blob: 3632b5a4f13f0250cb69c09306024366cb0d6d46 [file] [log] [blame]
Jens Axboe2b188cc2019-01-07 10:46:33 -07001// SPDX-License-Identifier: GPL-2.0
2/*
3 * Shared application/kernel submission and completion ring pairs, for
4 * supporting fast/efficient IO.
5 *
6 * A note on the read/write ordering memory barriers that are matched between
Stefan Bühler1e84b972019-04-24 23:54:16 +02007 * the application and kernel side.
8 *
9 * After the application reads the CQ ring tail, it must use an
10 * appropriate smp_rmb() to pair with the smp_wmb() the kernel uses
11 * before writing the tail (using smp_load_acquire to read the tail will
12 * do). It also needs a smp_mb() before updating CQ head (ordering the
13 * entry load(s) with the head store), pairing with an implicit barrier
14 * through a control-dependency in io_get_cqring (smp_store_release to
15 * store head will do). Failure to do so could lead to reading invalid
16 * CQ entries.
17 *
18 * Likewise, the application must use an appropriate smp_wmb() before
19 * writing the SQ tail (ordering SQ entry stores with the tail store),
20 * which pairs with smp_load_acquire in io_get_sqring (smp_store_release
21 * to store the tail will do). And it needs a barrier ordering the SQ
22 * head load before writing new SQ entries (smp_load_acquire to read
23 * head will do).
24 *
25 * When using the SQ poll thread (IORING_SETUP_SQPOLL), the application
26 * needs to check the SQ flags for IORING_SQ_NEED_WAKEUP *after*
27 * updating the SQ tail; a full memory barrier smp_mb() is needed
28 * between.
Jens Axboe2b188cc2019-01-07 10:46:33 -070029 *
30 * Also see the examples in the liburing library:
31 *
32 * git://git.kernel.dk/liburing
33 *
34 * io_uring also uses READ/WRITE_ONCE() for _any_ store or load that happens
35 * from data shared between the kernel and application. This is done both
36 * for ordering purposes, but also to ensure that once a value is loaded from
37 * data that the application could potentially modify, it remains stable.
38 *
39 * Copyright (C) 2018-2019 Jens Axboe
Christoph Hellwigc992fe22019-01-11 09:43:02 -070040 * Copyright (c) 2018-2019 Christoph Hellwig
Jens Axboe2b188cc2019-01-07 10:46:33 -070041 */
42#include <linux/kernel.h>
43#include <linux/init.h>
44#include <linux/errno.h>
45#include <linux/syscalls.h>
46#include <linux/compat.h>
Jens Axboe52de1fe2020-02-27 10:15:42 -070047#include <net/compat.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070048#include <linux/refcount.h>
49#include <linux/uio.h>
Pavel Begunkov6b47ee62020-01-18 20:22:41 +030050#include <linux/bits.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070051
52#include <linux/sched/signal.h>
53#include <linux/fs.h>
54#include <linux/file.h>
55#include <linux/fdtable.h>
56#include <linux/mm.h>
57#include <linux/mman.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070058#include <linux/percpu.h>
59#include <linux/slab.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070060#include <linux/blkdev.h>
Jens Axboeedafcce2019-01-09 09:16:05 -070061#include <linux/bvec.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070062#include <linux/net.h>
63#include <net/sock.h>
64#include <net/af_unix.h>
Jens Axboe6b063142019-01-10 22:13:58 -070065#include <net/scm.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070066#include <linux/anon_inodes.h>
67#include <linux/sched/mm.h>
68#include <linux/uaccess.h>
69#include <linux/nospec.h>
Jens Axboeedafcce2019-01-09 09:16:05 -070070#include <linux/sizes.h>
71#include <linux/hugetlb.h>
Jens Axboeaa4c3962019-11-29 10:14:00 -070072#include <linux/highmem.h>
Jens Axboe15b71ab2019-12-11 11:20:36 -070073#include <linux/namei.h>
74#include <linux/fsnotify.h>
Jens Axboe4840e412019-12-25 22:03:45 -070075#include <linux/fadvise.h>
Jens Axboe3e4827b2020-01-08 15:18:09 -070076#include <linux/eventpoll.h>
Pavel Begunkov7d67af22020-02-24 11:32:45 +030077#include <linux/splice.h>
Jens Axboeb41e9852020-02-17 09:52:41 -070078#include <linux/task_work.h>
Jens Axboebcf5a062020-05-22 09:24:42 -060079#include <linux/pagemap.h>
Jens Axboe0f212202020-09-13 13:09:39 -060080#include <linux/io_uring.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070081
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +020082#define CREATE_TRACE_POINTS
83#include <trace/events/io_uring.h>
84
Jens Axboe2b188cc2019-01-07 10:46:33 -070085#include <uapi/linux/io_uring.h>
86
87#include "internal.h"
Jens Axboe561fb042019-10-24 07:25:42 -060088#include "io-wq.h"
Jens Axboe2b188cc2019-01-07 10:46:33 -070089
Daniel Xu5277dea2019-09-14 14:23:45 -070090#define IORING_MAX_ENTRIES 32768
Jens Axboe33a107f2019-10-04 12:10:03 -060091#define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES)
Jens Axboe65e19f52019-10-26 07:20:21 -060092
93/*
94 * Shift of 9 is 512 entries, or exactly one page on 64-bit archs
95 */
96#define IORING_FILE_TABLE_SHIFT 9
97#define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT)
98#define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1)
99#define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE)
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200100#define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \
101 IORING_REGISTER_LAST + IORING_OP_LAST)
Jens Axboe2b188cc2019-01-07 10:46:33 -0700102
Pavel Begunkovb16fed662021-02-18 18:29:40 +0000103#define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \
104 IOSQE_IO_HARDLINK | IOSQE_ASYNC | \
105 IOSQE_BUFFER_SELECT)
106
Jens Axboe2b188cc2019-01-07 10:46:33 -0700107struct io_uring {
108 u32 head ____cacheline_aligned_in_smp;
109 u32 tail ____cacheline_aligned_in_smp;
110};
111
Stefan Bühler1e84b972019-04-24 23:54:16 +0200112/*
Hristo Venev75b28af2019-08-26 17:23:46 +0000113 * This data is shared with the application through the mmap at offsets
114 * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING.
Stefan Bühler1e84b972019-04-24 23:54:16 +0200115 *
116 * The offsets to the member fields are published through struct
117 * io_sqring_offsets when calling io_uring_setup.
118 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000119struct io_rings {
Stefan Bühler1e84b972019-04-24 23:54:16 +0200120 /*
121 * Head and tail offsets into the ring; the offsets need to be
122 * masked to get valid indices.
123 *
Hristo Venev75b28af2019-08-26 17:23:46 +0000124 * The kernel controls head of the sq ring and the tail of the cq ring,
125 * and the application controls tail of the sq ring and the head of the
126 * cq ring.
Stefan Bühler1e84b972019-04-24 23:54:16 +0200127 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000128 struct io_uring sq, cq;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200129 /*
Hristo Venev75b28af2019-08-26 17:23:46 +0000130 * Bitmasks to apply to head and tail offsets (constant, equals
Stefan Bühler1e84b972019-04-24 23:54:16 +0200131 * ring_entries - 1)
132 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000133 u32 sq_ring_mask, cq_ring_mask;
134 /* Ring sizes (constant, power of 2) */
135 u32 sq_ring_entries, cq_ring_entries;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200136 /*
137 * Number of invalid entries dropped by the kernel due to
138 * invalid index stored in array
139 *
140 * Written by the kernel, shouldn't be modified by the
141 * application (i.e. get number of "new events" by comparing to
142 * cached value).
143 *
144 * After a new SQ head value was read by the application this
145 * counter includes all submissions that were dropped reaching
146 * the new SQ head (and possibly more).
147 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000148 u32 sq_dropped;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200149 /*
Stefano Garzarella0d9b5b32020-05-15 18:38:04 +0200150 * Runtime SQ flags
Stefan Bühler1e84b972019-04-24 23:54:16 +0200151 *
152 * Written by the kernel, shouldn't be modified by the
153 * application.
154 *
155 * The application needs a full memory barrier before checking
156 * for IORING_SQ_NEED_WAKEUP after updating the sq tail.
157 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000158 u32 sq_flags;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200159 /*
Stefano Garzarella0d9b5b32020-05-15 18:38:04 +0200160 * Runtime CQ flags
161 *
162 * Written by the application, shouldn't be modified by the
163 * kernel.
164 */
165 u32 cq_flags;
166 /*
Stefan Bühler1e84b972019-04-24 23:54:16 +0200167 * Number of completion events lost because the queue was full;
168 * this should be avoided by the application by making sure
LimingWu0b4295b2019-12-05 20:18:18 +0800169 * there are not more requests pending than there is space in
Stefan Bühler1e84b972019-04-24 23:54:16 +0200170 * the completion queue.
171 *
172 * Written by the kernel, shouldn't be modified by the
173 * application (i.e. get number of "new events" by comparing to
174 * cached value).
175 *
176 * As completion events come in out of order this counter is not
177 * ordered with any other data.
178 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000179 u32 cq_overflow;
Stefan Bühler1e84b972019-04-24 23:54:16 +0200180 /*
181 * Ring buffer of completion events.
182 *
183 * The kernel writes completion events fresh every time they are
184 * produced, so the application is allowed to modify pending
185 * entries.
186 */
Hristo Venev75b28af2019-08-26 17:23:46 +0000187 struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700188};
189
Pavel Begunkov45d189c2021-02-10 00:03:07 +0000190enum io_uring_cmd_flags {
191 IO_URING_F_NONBLOCK = 1,
Pavel Begunkov889fca72021-02-10 00:03:09 +0000192 IO_URING_F_COMPLETE_DEFER = 2,
Pavel Begunkov45d189c2021-02-10 00:03:07 +0000193};
194
Jens Axboeedafcce2019-01-09 09:16:05 -0700195struct io_mapped_ubuf {
196 u64 ubuf;
Pavel Begunkov4751f532021-04-01 15:43:55 +0100197 u64 ubuf_end;
Jens Axboeedafcce2019-01-09 09:16:05 -0700198 struct bio_vec *bvec;
199 unsigned int nr_bvecs;
Jens Axboede293932020-09-17 16:19:16 -0600200 unsigned long acct_pages;
Jens Axboeedafcce2019-01-09 09:16:05 -0700201};
202
Bijan Mottahedeh50238532021-01-15 17:37:45 +0000203struct io_ring_ctx;
204
Pavel Begunkov6c2450a2021-02-23 12:40:22 +0000205struct io_overflow_cqe {
206 struct io_uring_cqe cqe;
207 struct list_head list;
208};
209
Pavel Begunkova04b0ac2021-04-01 15:44:04 +0100210struct io_fixed_file {
211 /* file * with additional FFS_* flags */
212 unsigned long file_ptr;
213};
214
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000215struct io_rsrc_put {
216 struct list_head list;
Bijan Mottahedeh50238532021-01-15 17:37:45 +0000217 union {
218 void *rsrc;
219 struct file *file;
220 };
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000221};
222
Pavel Begunkovaeca2412021-04-11 01:46:37 +0100223struct io_file_table {
224 /* two level table */
225 struct io_fixed_file **files;
Jens Axboe31b51512019-01-18 22:56:34 -0700226};
227
Pavel Begunkovb895c9a2021-04-01 15:43:40 +0100228struct io_rsrc_node {
Xiaoguang Wang05589552020-03-31 14:05:18 +0800229 struct percpu_ref refs;
230 struct list_head node;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000231 struct list_head rsrc_list;
Pavel Begunkovb895c9a2021-04-01 15:43:40 +0100232 struct io_rsrc_data *rsrc_data;
Jens Axboe4a38aed22020-05-14 17:21:15 -0600233 struct llist_node llist;
Pavel Begunkove2978222020-11-18 14:56:26 +0000234 bool done;
Xiaoguang Wang05589552020-03-31 14:05:18 +0800235};
236
Pavel Begunkov40ae0ff2021-04-01 15:43:44 +0100237typedef void (rsrc_put_fn)(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc);
238
Pavel Begunkovb895c9a2021-04-01 15:43:40 +0100239struct io_rsrc_data {
Jens Axboe05f3fb32019-12-09 11:22:50 -0700240 struct io_ring_ctx *ctx;
241
Pavel Begunkov40ae0ff2021-04-01 15:43:44 +0100242 rsrc_put_fn *do_put;
Pavel Begunkov3e942492021-04-11 01:46:34 +0100243 atomic_t refs;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700244 struct completion done;
Hao Xu8bad28d2021-02-19 17:19:36 +0800245 bool quiesce;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700246};
247
Jens Axboe5a2e7452020-02-23 16:23:11 -0700248struct io_buffer {
249 struct list_head list;
250 __u64 addr;
251 __s32 len;
252 __u16 bid;
253};
254
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200255struct io_restriction {
256 DECLARE_BITMAP(register_op, IORING_REGISTER_LAST);
257 DECLARE_BITMAP(sqe_op, IORING_OP_LAST);
258 u8 sqe_flags_allowed;
259 u8 sqe_flags_required;
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +0200260 bool registered;
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200261};
262
Jens Axboe37d1e2e2021-02-17 21:03:43 -0700263enum {
264 IO_SQ_THREAD_SHOULD_STOP = 0,
265 IO_SQ_THREAD_SHOULD_PARK,
266};
267
Jens Axboe534ca6d2020-09-02 13:52:19 -0600268struct io_sq_data {
269 refcount_t refs;
Pavel Begunkov9e138a42021-03-14 20:57:12 +0000270 atomic_t park_pending;
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +0000271 struct mutex lock;
Jens Axboe69fb2132020-09-14 11:16:23 -0600272
273 /* ctx's that are using this sqd */
274 struct list_head ctx_list;
Jens Axboe69fb2132020-09-14 11:16:23 -0600275
Jens Axboe534ca6d2020-09-02 13:52:19 -0600276 struct task_struct *thread;
277 struct wait_queue_head wait;
Xiaoguang Wang08369242020-11-03 14:15:59 +0800278
279 unsigned sq_thread_idle;
Jens Axboe37d1e2e2021-02-17 21:03:43 -0700280 int sq_cpu;
281 pid_t task_pid;
Jens Axboe5c2469e2021-03-11 10:17:56 -0700282 pid_t task_tgid;
Jens Axboe37d1e2e2021-02-17 21:03:43 -0700283
284 unsigned long state;
Jens Axboe37d1e2e2021-02-17 21:03:43 -0700285 struct completion exited;
Pavel Begunkovb7f5a0b2021-03-15 14:23:08 +0000286 struct callback_head *park_task_work;
Jens Axboe534ca6d2020-09-02 13:52:19 -0600287};
288
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000289#define IO_IOPOLL_BATCH 8
Pavel Begunkov6dd0be12021-02-10 00:03:13 +0000290#define IO_COMPL_BATCH 32
Pavel Begunkov6ff119a2021-02-10 00:03:18 +0000291#define IO_REQ_CACHE_SIZE 32
Pavel Begunkovbf019da2021-02-10 00:03:17 +0000292#define IO_REQ_ALLOC_BATCH 8
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000293
294struct io_comp_state {
Pavel Begunkov6dd0be12021-02-10 00:03:13 +0000295 struct io_kiocb *reqs[IO_COMPL_BATCH];
Jens Axboe1b4c3512021-02-10 00:03:19 +0000296 unsigned int nr;
Jens Axboec7dae4b2021-02-09 19:53:37 -0700297 unsigned int locked_free_nr;
298 /* inline/task_work completion list, under ->uring_lock */
Jens Axboe1b4c3512021-02-10 00:03:19 +0000299 struct list_head free_list;
Jens Axboec7dae4b2021-02-09 19:53:37 -0700300 /* IRQ completion list, under ->completion_lock */
301 struct list_head locked_free_list;
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000302};
303
Pavel Begunkova1ab7b32021-02-18 18:29:42 +0000304struct io_submit_link {
305 struct io_kiocb *head;
306 struct io_kiocb *last;
307};
308
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000309struct io_submit_state {
310 struct blk_plug plug;
Pavel Begunkova1ab7b32021-02-18 18:29:42 +0000311 struct io_submit_link link;
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000312
313 /*
314 * io_kiocb alloc cache
315 */
Pavel Begunkovbf019da2021-02-10 00:03:17 +0000316 void *reqs[IO_REQ_CACHE_SIZE];
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000317 unsigned int free_reqs;
318
319 bool plug_started;
320
321 /*
322 * Batch completion logic
323 */
324 struct io_comp_state comp;
325
326 /*
327 * File reference cache
328 */
329 struct file *file;
330 unsigned int fd;
331 unsigned int file_refs;
332 unsigned int ios_left;
333};
334
Jens Axboe2b188cc2019-01-07 10:46:33 -0700335struct io_ring_ctx {
336 struct {
337 struct percpu_ref refs;
338 } ____cacheline_aligned_in_smp;
339
340 struct {
341 unsigned int flags;
Randy Dunlape1d85332020-02-05 20:57:10 -0800342 unsigned int compat: 1;
Randy Dunlape1d85332020-02-05 20:57:10 -0800343 unsigned int drain_next: 1;
344 unsigned int eventfd_async: 1;
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200345 unsigned int restricted: 1;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700346
Hristo Venev75b28af2019-08-26 17:23:46 +0000347 /*
348 * Ring buffer of indices into array of io_uring_sqe, which is
349 * mmapped by the application using the IORING_OFF_SQES offset.
350 *
351 * This indirection could e.g. be used to assign fixed
352 * io_uring_sqe entries to operations and only submit them to
353 * the queue when needed.
354 *
355 * The kernel modifies neither the indices array nor the entries
356 * array.
357 */
358 u32 *sq_array;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700359 unsigned cached_sq_head;
360 unsigned sq_entries;
361 unsigned sq_mask;
Jens Axboe6c271ce2019-01-10 11:22:30 -0700362 unsigned sq_thread_idle;
Jens Axboe498ccd92019-10-25 10:04:25 -0600363 unsigned cached_sq_dropped;
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +0100364 unsigned cached_cq_overflow;
Jens Axboead3eb2c2019-12-18 17:12:20 -0700365 unsigned long sq_check_overflow;
Jens Axboede0617e2019-04-06 21:51:27 -0600366
Jens Axboee9418942021-02-19 12:33:30 -0700367 /* hashed buffered write serialization */
368 struct io_wq_hash *hash_map;
369
Jens Axboede0617e2019-04-06 21:51:27 -0600370 struct list_head defer_list;
Jens Axboe5262f562019-09-17 12:26:57 -0600371 struct list_head timeout_list;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -0700372 struct list_head cq_overflow_list;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700373
Jens Axboead3eb2c2019-12-18 17:12:20 -0700374 struct io_uring_sqe *sq_sqes;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700375 } ____cacheline_aligned_in_smp;
376
Jens Axboe3c1a2ea2021-02-11 10:48:03 -0700377 struct {
378 struct mutex uring_lock;
379 wait_queue_head_t wait;
380 } ____cacheline_aligned_in_smp;
381
382 struct io_submit_state submit_state;
383
Hristo Venev75b28af2019-08-26 17:23:46 +0000384 struct io_rings *rings;
385
Jens Axboe2aede0e2020-09-14 10:45:53 -0600386 /* Only used for accounting purposes */
387 struct mm_struct *mm_account;
388
Stefan Metzmacher7c30f36a2021-03-07 11:54:28 +0100389 const struct cred *sq_creds; /* cred used for __io_sq_thread() */
Jens Axboe534ca6d2020-09-02 13:52:19 -0600390 struct io_sq_data *sq_data; /* if using sq thread polling */
391
Jens Axboe90554202020-09-03 12:12:41 -0600392 struct wait_queue_head sqo_sq_wait;
Jens Axboe69fb2132020-09-14 11:16:23 -0600393 struct list_head sqd_list;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700394
Jens Axboe6b063142019-01-10 22:13:58 -0700395 /*
396 * If used, fixed file set. Writers must ensure that ->refs is dead,
397 * readers must ensure that ->refs is alive as long as the file* is
398 * used. Only updated through io_uring_register(2).
399 */
Pavel Begunkovb895c9a2021-04-01 15:43:40 +0100400 struct io_rsrc_data *file_data;
Pavel Begunkovaeca2412021-04-11 01:46:37 +0100401 struct io_file_table file_table;
Jens Axboe6b063142019-01-10 22:13:58 -0700402 unsigned nr_user_files;
403
Jens Axboeedafcce2019-01-09 09:16:05 -0700404 /* if used, fixed mapped user buffers */
405 unsigned nr_user_bufs;
406 struct io_mapped_ubuf *user_bufs;
407
Jens Axboe2b188cc2019-01-07 10:46:33 -0700408 struct user_struct *user;
409
Jens Axboe0f158b42020-05-14 17:18:39 -0600410 struct completion ref_comp;
Jens Axboe206aefd2019-11-07 18:27:42 -0700411
412#if defined(CONFIG_UNIX)
413 struct socket *ring_sock;
414#endif
415
Jens Axboe9e15c3a2021-03-13 12:29:43 -0700416 struct xarray io_buffers;
Jens Axboe5a2e7452020-02-23 16:23:11 -0700417
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +0000418 struct xarray personalities;
419 u32 pers_next;
Jens Axboe071698e2020-01-28 10:04:42 -0700420
Jens Axboe206aefd2019-11-07 18:27:42 -0700421 struct {
422 unsigned cached_cq_tail;
423 unsigned cq_entries;
424 unsigned cq_mask;
425 atomic_t cq_timeouts;
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -0500426 unsigned cq_last_tm_flush;
Jens Axboead3eb2c2019-12-18 17:12:20 -0700427 unsigned long cq_check_overflow;
Jens Axboe206aefd2019-11-07 18:27:42 -0700428 struct wait_queue_head cq_wait;
429 struct fasync_struct *cq_fasync;
430 struct eventfd_ctx *cq_ev_fd;
431 } ____cacheline_aligned_in_smp;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700432
433 struct {
Jens Axboe2b188cc2019-01-07 10:46:33 -0700434 spinlock_t completion_lock;
Jens Axboee94f1412019-12-19 12:06:02 -0700435
Jens Axboedef596e2019-01-09 08:59:42 -0700436 /*
Pavel Begunkov540e32a2020-07-13 23:37:09 +0300437 * ->iopoll_list is protected by the ctx->uring_lock for
Jens Axboedef596e2019-01-09 08:59:42 -0700438 * io_uring instances that don't use IORING_SETUP_SQPOLL.
439 * For SQPOLL, only the single threaded io_sq_thread() will
440 * manipulate the list, hence no extra locking is needed there.
441 */
Pavel Begunkov540e32a2020-07-13 23:37:09 +0300442 struct list_head iopoll_list;
Jens Axboe78076bb2019-12-04 19:56:40 -0700443 struct hlist_head *cancel_hash;
444 unsigned cancel_hash_bits;
Jens Axboee94f1412019-12-19 12:06:02 -0700445 bool poll_multi_file;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700446 } ____cacheline_aligned_in_smp;
Jens Axboe85faa7b2020-04-09 18:14:00 -0600447
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000448 struct delayed_work rsrc_put_work;
449 struct llist_head rsrc_put_llist;
Bijan Mottahedehd67d2262021-01-15 17:37:46 +0000450 struct list_head rsrc_ref_list;
451 spinlock_t rsrc_ref_lock;
Pavel Begunkova7f0ed52021-04-01 15:43:46 +0100452 struct io_rsrc_node *rsrc_node;
Pavel Begunkovb895c9a2021-04-01 15:43:40 +0100453 struct io_rsrc_node *rsrc_backup_node;
Jens Axboe4a38aed22020-05-14 17:21:15 -0600454
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200455 struct io_restriction restrictions;
Jens Axboe3c1a2ea2021-02-11 10:48:03 -0700456
Jens Axboe7c25c0d2021-02-16 07:17:00 -0700457 /* exit task_work */
458 struct callback_head *exit_task_work;
459
Jens Axboe3c1a2ea2021-02-11 10:48:03 -0700460 /* Keep this last, we don't need it for the fast path */
461 struct work_struct exit_work;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +0000462 struct list_head tctx_list;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700463};
464
Stefan Metzmacher53e043b2021-03-15 12:56:56 +0100465struct io_uring_task {
466 /* submission side */
467 struct xarray xa;
468 struct wait_queue_head wait;
Stefan Metzmacheree53fb22021-03-15 12:56:57 +0100469 const struct io_ring_ctx *last;
470 struct io_wq *io_wq;
Stefan Metzmacher53e043b2021-03-15 12:56:56 +0100471 struct percpu_counter inflight;
Pavel Begunkovb303fe22021-04-11 01:46:26 +0100472 atomic_t inflight_tracked;
Stefan Metzmacher53e043b2021-03-15 12:56:56 +0100473 atomic_t in_idle;
Stefan Metzmacher53e043b2021-03-15 12:56:56 +0100474
475 spinlock_t task_lock;
476 struct io_wq_work_list task_list;
477 unsigned long task_state;
478 struct callback_head task_work;
479};
480
Jens Axboe09bb8392019-03-13 12:39:28 -0600481/*
482 * First field must be the file pointer in all the
483 * iocb unions! See also 'struct kiocb' in <linux/fs.h>
484 */
Jens Axboe221c5eb2019-01-17 09:41:58 -0700485struct io_poll_iocb {
486 struct file *file;
Pavel Begunkov018043b2020-10-27 23:17:18 +0000487 struct wait_queue_head *head;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700488 __poll_t events;
Jens Axboe8c838782019-03-12 15:48:16 -0600489 bool done;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700490 bool canceled;
Pavel Begunkov9d805892021-04-13 02:58:40 +0100491 struct wait_queue_entry wait;
492};
493
494struct io_poll_update {
495 struct file *file;
496 u64 old_user_data;
497 u64 new_user_data;
498 __poll_t events;
Jens Axboeb69de282021-03-17 08:37:41 -0600499 bool update_events;
500 bool update_user_data;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700501};
502
Pavel Begunkov018043b2020-10-27 23:17:18 +0000503struct io_poll_remove {
504 struct file *file;
505 u64 addr;
506};
507
Jens Axboeb5dba592019-12-11 14:02:38 -0700508struct io_close {
509 struct file *file;
Jens Axboeb5dba592019-12-11 14:02:38 -0700510 int fd;
511};
512
Jens Axboead8a48a2019-11-15 08:49:11 -0700513struct io_timeout_data {
514 struct io_kiocb *req;
515 struct hrtimer timer;
516 struct timespec64 ts;
517 enum hrtimer_mode mode;
518};
519
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700520struct io_accept {
521 struct file *file;
522 struct sockaddr __user *addr;
523 int __user *addr_len;
524 int flags;
Jens Axboe09952e32020-03-19 20:16:56 -0600525 unsigned long nofile;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700526};
527
528struct io_sync {
529 struct file *file;
530 loff_t len;
531 loff_t off;
532 int flags;
Jens Axboed63d1b52019-12-10 10:38:56 -0700533 int mode;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700534};
535
Jens Axboefbf23842019-12-17 18:45:56 -0700536struct io_cancel {
537 struct file *file;
538 u64 addr;
539};
540
Jens Axboeb29472e2019-12-17 18:50:29 -0700541struct io_timeout {
542 struct file *file;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +0300543 u32 off;
544 u32 target_seq;
Pavel Begunkov135fcde2020-07-13 23:37:12 +0300545 struct list_head list;
Pavel Begunkov90cd7e42020-10-27 23:25:36 +0000546 /* head of the link, used by linked timeouts only */
547 struct io_kiocb *head;
Jens Axboeb29472e2019-12-17 18:50:29 -0700548};
549
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100550struct io_timeout_rem {
551 struct file *file;
552 u64 addr;
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000553
554 /* timeout update */
555 struct timespec64 ts;
556 u32 flags;
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100557};
558
Jens Axboe9adbd452019-12-20 08:45:55 -0700559struct io_rw {
560 /* NOTE: kiocb has the file as the first member, so don't do it here */
561 struct kiocb kiocb;
562 u64 addr;
563 u64 len;
564};
565
Jens Axboe3fbb51c2019-12-20 08:51:52 -0700566struct io_connect {
567 struct file *file;
568 struct sockaddr __user *addr;
569 int addr_len;
570};
571
Jens Axboee47293f2019-12-20 08:58:21 -0700572struct io_sr_msg {
573 struct file *file;
Jens Axboefddafac2020-01-04 20:19:44 -0700574 union {
Pavel Begunkov4af34172021-04-11 01:46:30 +0100575 struct compat_msghdr __user *umsg_compat;
576 struct user_msghdr __user *umsg;
577 void __user *buf;
Jens Axboefddafac2020-01-04 20:19:44 -0700578 };
Jens Axboee47293f2019-12-20 08:58:21 -0700579 int msg_flags;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700580 int bgid;
Jens Axboefddafac2020-01-04 20:19:44 -0700581 size_t len;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700582 struct io_buffer *kbuf;
Jens Axboee47293f2019-12-20 08:58:21 -0700583};
584
Jens Axboe15b71ab2019-12-11 11:20:36 -0700585struct io_open {
586 struct file *file;
587 int dfd;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700588 struct filename *filename;
Jens Axboec12cedf2020-01-08 17:41:21 -0700589 struct open_how how;
Jens Axboe4022e7a2020-03-19 19:23:18 -0600590 unsigned long nofile;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700591};
592
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000593struct io_rsrc_update {
Jens Axboe05f3fb32019-12-09 11:22:50 -0700594 struct file *file;
595 u64 arg;
596 u32 nr_args;
597 u32 offset;
598};
599
Jens Axboe4840e412019-12-25 22:03:45 -0700600struct io_fadvise {
601 struct file *file;
602 u64 offset;
603 u32 len;
604 u32 advice;
605};
606
Jens Axboec1ca7572019-12-25 22:18:28 -0700607struct io_madvise {
608 struct file *file;
609 u64 addr;
610 u32 len;
611 u32 advice;
612};
613
Jens Axboe3e4827b2020-01-08 15:18:09 -0700614struct io_epoll {
615 struct file *file;
616 int epfd;
617 int op;
618 int fd;
619 struct epoll_event event;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700620};
621
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300622struct io_splice {
623 struct file *file_out;
624 struct file *file_in;
625 loff_t off_out;
626 loff_t off_in;
627 u64 len;
628 unsigned int flags;
629};
630
Jens Axboeddf0322d2020-02-23 16:41:33 -0700631struct io_provide_buf {
632 struct file *file;
633 __u64 addr;
634 __s32 len;
635 __u32 bgid;
636 __u16 nbufs;
637 __u16 bid;
638};
639
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700640struct io_statx {
641 struct file *file;
642 int dfd;
643 unsigned int mask;
644 unsigned int flags;
Bijan Mottahedehe62753e2020-05-22 21:31:18 -0700645 const char __user *filename;
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700646 struct statx __user *buffer;
647};
648
Jens Axboe36f4fa62020-09-05 11:14:22 -0600649struct io_shutdown {
650 struct file *file;
651 int how;
652};
653
Jens Axboe80a261f2020-09-28 14:23:58 -0600654struct io_rename {
655 struct file *file;
656 int old_dfd;
657 int new_dfd;
658 struct filename *oldpath;
659 struct filename *newpath;
660 int flags;
661};
662
Jens Axboe14a11432020-09-28 14:27:37 -0600663struct io_unlink {
664 struct file *file;
665 int dfd;
666 int flags;
667 struct filename *filename;
668};
669
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300670struct io_completion {
671 struct file *file;
672 struct list_head list;
Pavel Begunkov8c3f9cd2021-02-28 22:35:15 +0000673 u32 cflags;
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300674};
675
Jens Axboef499a022019-12-02 16:28:46 -0700676struct io_async_connect {
677 struct sockaddr_storage address;
678};
679
Jens Axboe03b12302019-12-02 18:50:25 -0700680struct io_async_msghdr {
681 struct iovec fast_iov[UIO_FASTIOV];
Pavel Begunkov257e84a2021-02-05 00:58:00 +0000682 /* points to an allocated iov, if NULL we use fast_iov instead */
683 struct iovec *free_iov;
Jens Axboe03b12302019-12-02 18:50:25 -0700684 struct sockaddr __user *uaddr;
685 struct msghdr msg;
Jens Axboeb5379162020-02-09 11:29:15 -0700686 struct sockaddr_storage addr;
Jens Axboe03b12302019-12-02 18:50:25 -0700687};
688
Jens Axboef67676d2019-12-02 11:03:47 -0700689struct io_async_rw {
690 struct iovec fast_iov[UIO_FASTIOV];
Jens Axboeff6165b2020-08-13 09:47:43 -0600691 const struct iovec *free_iovec;
692 struct iov_iter iter;
Jens Axboe227c0c92020-08-13 11:51:40 -0600693 size_t bytes_done;
Jens Axboebcf5a062020-05-22 09:24:42 -0600694 struct wait_page_queue wpq;
Jens Axboef67676d2019-12-02 11:03:47 -0700695};
696
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300697enum {
698 REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT,
699 REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT,
700 REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT,
701 REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT,
702 REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700703 REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300704
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300705 REQ_F_FAIL_LINK_BIT,
706 REQ_F_INFLIGHT_BIT,
707 REQ_F_CUR_POS_BIT,
708 REQ_F_NOWAIT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300709 REQ_F_LINK_TIMEOUT_BIT,
Pavel Begunkov99bc4c32020-02-07 22:04:45 +0300710 REQ_F_NEED_CLEANUP_BIT,
Jens Axboed7718a92020-02-14 22:23:12 -0700711 REQ_F_POLLED_BIT,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700712 REQ_F_BUFFER_SELECTED_BIT,
Pavel Begunkov900fad42020-10-19 16:39:16 +0100713 REQ_F_LTIMEOUT_ACTIVE_BIT,
Pavel Begunkove342c802021-01-19 13:32:47 +0000714 REQ_F_COMPLETE_INLINE_BIT,
Jens Axboe230d50d2021-04-01 20:41:15 -0600715 REQ_F_REISSUE_BIT,
Pavel Begunkov8c130822021-03-22 01:58:32 +0000716 REQ_F_DONT_REISSUE_BIT,
Pavel Begunkov9d805892021-04-13 02:58:40 +0100717 REQ_F_POLL_UPDATE_BIT,
Jens Axboe7b29f922021-03-12 08:30:14 -0700718 /* keep async read/write and isreg together and in order */
719 REQ_F_ASYNC_READ_BIT,
720 REQ_F_ASYNC_WRITE_BIT,
721 REQ_F_ISREG_BIT,
Jens Axboe84557872020-03-03 15:28:17 -0700722
723 /* not a real bit, just to check we're not overflowing the space */
724 __REQ_F_LAST_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300725};
726
727enum {
728 /* ctx owns file */
729 REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT),
730 /* drain existing IO first */
731 REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT),
732 /* linked sqes */
733 REQ_F_LINK = BIT(REQ_F_LINK_BIT),
734 /* doesn't sever on completion < 0 */
735 REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT),
736 /* IOSQE_ASYNC */
737 REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT),
Jens Axboebcda7ba2020-02-23 16:42:51 -0700738 /* IOSQE_BUFFER_SELECT */
739 REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT),
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300740
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300741 /* fail rest of links */
742 REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT),
Pavel Begunkovb05a1bc2021-03-04 13:59:24 +0000743 /* on inflight list, should be cancelled and waited on exit reliably */
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300744 REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT),
745 /* read/write uses file position */
746 REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT),
747 /* must not punt to workers */
748 REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT),
Pavel Begunkov900fad42020-10-19 16:39:16 +0100749 /* has or had linked timeout */
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300750 REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT),
Pavel Begunkov99bc4c32020-02-07 22:04:45 +0300751 /* needs cleanup */
752 REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT),
Jens Axboed7718a92020-02-14 22:23:12 -0700753 /* already went through poll handler */
754 REQ_F_POLLED = BIT(REQ_F_POLLED_BIT),
Jens Axboebcda7ba2020-02-23 16:42:51 -0700755 /* buffer already selected */
756 REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT),
Pavel Begunkov900fad42020-10-19 16:39:16 +0100757 /* linked timeout is active, i.e. prepared by link's head */
758 REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT),
Pavel Begunkove342c802021-01-19 13:32:47 +0000759 /* completion is deferred through io_comp_state */
760 REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT),
Jens Axboe230d50d2021-04-01 20:41:15 -0600761 /* caller should reissue async */
762 REQ_F_REISSUE = BIT(REQ_F_REISSUE_BIT),
Pavel Begunkov8c130822021-03-22 01:58:32 +0000763 /* don't attempt request reissue, see io_rw_reissue() */
764 REQ_F_DONT_REISSUE = BIT(REQ_F_DONT_REISSUE_BIT),
Pavel Begunkov9d805892021-04-13 02:58:40 +0100765 /* switches between poll and poll update */
766 REQ_F_POLL_UPDATE = BIT(REQ_F_POLL_UPDATE_BIT),
Jens Axboe7b29f922021-03-12 08:30:14 -0700767 /* supports async reads */
768 REQ_F_ASYNC_READ = BIT(REQ_F_ASYNC_READ_BIT),
769 /* supports async writes */
770 REQ_F_ASYNC_WRITE = BIT(REQ_F_ASYNC_WRITE_BIT),
771 /* regular file */
772 REQ_F_ISREG = BIT(REQ_F_ISREG_BIT),
Jens Axboed7718a92020-02-14 22:23:12 -0700773};
774
775struct async_poll {
776 struct io_poll_iocb poll;
Jens Axboe807abcb2020-07-17 17:09:27 -0600777 struct io_poll_iocb *double_poll;
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300778};
779
Jens Axboe7cbf1722021-02-10 00:03:20 +0000780struct io_task_work {
781 struct io_wq_work_node node;
782 task_work_func_t func;
783};
784
Jens Axboe09bb8392019-03-13 12:39:28 -0600785/*
786 * NOTE! Each of the iocb union members has the file pointer
787 * as the first entry in their struct definition. So you can
788 * access the file pointer through any of the sub-structs,
789 * or directly as just 'ki_filp' in this struct.
790 */
Jens Axboe2b188cc2019-01-07 10:46:33 -0700791struct io_kiocb {
Jens Axboe221c5eb2019-01-17 09:41:58 -0700792 union {
Jens Axboe09bb8392019-03-13 12:39:28 -0600793 struct file *file;
Jens Axboe9adbd452019-12-20 08:45:55 -0700794 struct io_rw rw;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700795 struct io_poll_iocb poll;
Pavel Begunkov9d805892021-04-13 02:58:40 +0100796 struct io_poll_update poll_update;
Pavel Begunkov018043b2020-10-27 23:17:18 +0000797 struct io_poll_remove poll_remove;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700798 struct io_accept accept;
799 struct io_sync sync;
Jens Axboefbf23842019-12-17 18:45:56 -0700800 struct io_cancel cancel;
Jens Axboeb29472e2019-12-17 18:50:29 -0700801 struct io_timeout timeout;
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100802 struct io_timeout_rem timeout_rem;
Jens Axboe3fbb51c2019-12-20 08:51:52 -0700803 struct io_connect connect;
Jens Axboee47293f2019-12-20 08:58:21 -0700804 struct io_sr_msg sr_msg;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700805 struct io_open open;
Jens Axboeb5dba592019-12-11 14:02:38 -0700806 struct io_close close;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000807 struct io_rsrc_update rsrc_update;
Jens Axboe4840e412019-12-25 22:03:45 -0700808 struct io_fadvise fadvise;
Jens Axboec1ca7572019-12-25 22:18:28 -0700809 struct io_madvise madvise;
Jens Axboe3e4827b2020-01-08 15:18:09 -0700810 struct io_epoll epoll;
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300811 struct io_splice splice;
Jens Axboeddf0322d2020-02-23 16:41:33 -0700812 struct io_provide_buf pbuf;
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700813 struct io_statx statx;
Jens Axboe36f4fa62020-09-05 11:14:22 -0600814 struct io_shutdown shutdown;
Jens Axboe80a261f2020-09-28 14:23:58 -0600815 struct io_rename rename;
Jens Axboe14a11432020-09-28 14:27:37 -0600816 struct io_unlink unlink;
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300817 /* use only after cleaning per-op data, see io_clean_op() */
818 struct io_completion compl;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700819 };
Jens Axboe2b188cc2019-01-07 10:46:33 -0700820
Jens Axboee8c2bc12020-08-15 18:44:09 -0700821 /* opcode allocated if it needs to store data for async defer */
822 void *async_data;
Jens Axboed625c6e2019-12-17 19:53:05 -0700823 u8 opcode;
Xiaoguang Wang65a65432020-06-11 23:39:36 +0800824 /* polled IO has completed */
825 u8 iopoll_completed;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700826
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -0700827 u16 buf_index;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +0300828 u32 result;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -0700829
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300830 struct io_ring_ctx *ctx;
831 unsigned int flags;
Jens Axboeabc54d62021-02-24 13:32:30 -0700832 atomic_t refs;
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300833 struct task_struct *task;
834 u64 user_data;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700835
Pavel Begunkovf2f87372020-10-27 23:25:37 +0000836 struct io_kiocb *link;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000837 struct percpu_ref *fixed_rsrc_refs;
Jens Axboed7718a92020-02-14 22:23:12 -0700838
Pavel Begunkovb303fe22021-04-11 01:46:26 +0100839 /* used with ctx->iopoll_list with reads/writes */
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300840 struct list_head inflight_entry;
Jens Axboe7cbf1722021-02-10 00:03:20 +0000841 union {
842 struct io_task_work io_task_work;
843 struct callback_head task_work;
844 };
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300845 /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */
846 struct hlist_node hash_node;
847 struct async_poll *apoll;
848 struct io_wq_work work;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700849};
850
Pavel Begunkov13bf43f2021-03-06 11:02:12 +0000851struct io_tctx_node {
852 struct list_head ctx_node;
853 struct task_struct *task;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +0000854 struct io_ring_ctx *ctx;
855};
856
Pavel Begunkov27dc8332020-07-13 23:37:14 +0300857struct io_defer_entry {
858 struct list_head list;
859 struct io_kiocb *req;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +0300860 u32 seq;
Pavel Begunkov27dc8332020-07-13 23:37:14 +0300861};
862
Jens Axboed3656342019-12-18 09:50:26 -0700863struct io_op_def {
Jens Axboed3656342019-12-18 09:50:26 -0700864 /* needs req->file assigned */
865 unsigned needs_file : 1;
Jens Axboed3656342019-12-18 09:50:26 -0700866 /* hash wq insertion if file is a regular file */
867 unsigned hash_reg_file : 1;
868 /* unbound wq insertion if file is a non-regular file */
869 unsigned unbound_nonreg_file : 1;
Jens Axboe66f4af92020-01-16 15:36:52 -0700870 /* opcode is not supported by this kernel */
871 unsigned not_supported : 1;
Jens Axboe8a727582020-02-20 09:59:44 -0700872 /* set if opcode supports polled "wait" */
873 unsigned pollin : 1;
874 unsigned pollout : 1;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700875 /* op supports buffer selection */
876 unsigned buffer_select : 1;
Pavel Begunkov26f05052021-02-28 22:35:18 +0000877 /* do prep async if is going to be punted */
878 unsigned needs_async_setup : 1;
Jens Axboe27926b62020-10-28 09:33:23 -0600879 /* should block plug */
880 unsigned plug : 1;
Jens Axboee8c2bc12020-08-15 18:44:09 -0700881 /* size of async data needed, if any */
882 unsigned short async_size;
Jens Axboed3656342019-12-18 09:50:26 -0700883};
884
Jens Axboe09186822020-10-13 15:01:40 -0600885static const struct io_op_def io_op_defs[] = {
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300886 [IORING_OP_NOP] = {},
887 [IORING_OP_READV] = {
Jens Axboed3656342019-12-18 09:50:26 -0700888 .needs_file = 1,
889 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700890 .pollin = 1,
Jens Axboe4d954c22020-02-27 07:31:19 -0700891 .buffer_select = 1,
Pavel Begunkov26f05052021-02-28 22:35:18 +0000892 .needs_async_setup = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600893 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700894 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700895 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300896 [IORING_OP_WRITEV] = {
Jens Axboed3656342019-12-18 09:50:26 -0700897 .needs_file = 1,
898 .hash_reg_file = 1,
899 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700900 .pollout = 1,
Pavel Begunkov26f05052021-02-28 22:35:18 +0000901 .needs_async_setup = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600902 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700903 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700904 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300905 [IORING_OP_FSYNC] = {
Jens Axboed3656342019-12-18 09:50:26 -0700906 .needs_file = 1,
907 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300908 [IORING_OP_READ_FIXED] = {
Jens Axboed3656342019-12-18 09:50:26 -0700909 .needs_file = 1,
910 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700911 .pollin = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600912 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700913 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700914 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300915 [IORING_OP_WRITE_FIXED] = {
Jens Axboed3656342019-12-18 09:50:26 -0700916 .needs_file = 1,
917 .hash_reg_file = 1,
918 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700919 .pollout = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600920 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700921 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700922 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300923 [IORING_OP_POLL_ADD] = {
Jens Axboed3656342019-12-18 09:50:26 -0700924 .needs_file = 1,
925 .unbound_nonreg_file = 1,
926 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300927 [IORING_OP_POLL_REMOVE] = {},
928 [IORING_OP_SYNC_FILE_RANGE] = {
Jens Axboed3656342019-12-18 09:50:26 -0700929 .needs_file = 1,
930 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300931 [IORING_OP_SENDMSG] = {
Jens Axboed3656342019-12-18 09:50:26 -0700932 .needs_file = 1,
933 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700934 .pollout = 1,
Pavel Begunkov26f05052021-02-28 22:35:18 +0000935 .needs_async_setup = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700936 .async_size = sizeof(struct io_async_msghdr),
Jens Axboed3656342019-12-18 09:50:26 -0700937 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300938 [IORING_OP_RECVMSG] = {
Jens Axboed3656342019-12-18 09:50:26 -0700939 .needs_file = 1,
940 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700941 .pollin = 1,
Jens Axboe52de1fe2020-02-27 10:15:42 -0700942 .buffer_select = 1,
Pavel Begunkov26f05052021-02-28 22:35:18 +0000943 .needs_async_setup = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700944 .async_size = sizeof(struct io_async_msghdr),
Jens Axboed3656342019-12-18 09:50:26 -0700945 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300946 [IORING_OP_TIMEOUT] = {
Jens Axboee8c2bc12020-08-15 18:44:09 -0700947 .async_size = sizeof(struct io_timeout_data),
Jens Axboed3656342019-12-18 09:50:26 -0700948 },
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000949 [IORING_OP_TIMEOUT_REMOVE] = {
950 /* used by timeout updates' prep() */
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000951 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300952 [IORING_OP_ACCEPT] = {
Jens Axboed3656342019-12-18 09:50:26 -0700953 .needs_file = 1,
954 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700955 .pollin = 1,
Jens Axboed3656342019-12-18 09:50:26 -0700956 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300957 [IORING_OP_ASYNC_CANCEL] = {},
958 [IORING_OP_LINK_TIMEOUT] = {
Jens Axboee8c2bc12020-08-15 18:44:09 -0700959 .async_size = sizeof(struct io_timeout_data),
Jens Axboed3656342019-12-18 09:50:26 -0700960 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300961 [IORING_OP_CONNECT] = {
Jens Axboed3656342019-12-18 09:50:26 -0700962 .needs_file = 1,
963 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700964 .pollout = 1,
Pavel Begunkov26f05052021-02-28 22:35:18 +0000965 .needs_async_setup = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700966 .async_size = sizeof(struct io_async_connect),
Jens Axboed3656342019-12-18 09:50:26 -0700967 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300968 [IORING_OP_FALLOCATE] = {
Jens Axboed3656342019-12-18 09:50:26 -0700969 .needs_file = 1,
970 },
Jens Axboe44526be2021-02-15 13:32:18 -0700971 [IORING_OP_OPENAT] = {},
972 [IORING_OP_CLOSE] = {},
973 [IORING_OP_FILES_UPDATE] = {},
974 [IORING_OP_STATX] = {},
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300975 [IORING_OP_READ] = {
Jens Axboe3a6820f2019-12-22 15:19:35 -0700976 .needs_file = 1,
977 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700978 .pollin = 1,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700979 .buffer_select = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600980 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700981 .async_size = sizeof(struct io_async_rw),
Jens Axboe3a6820f2019-12-22 15:19:35 -0700982 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300983 [IORING_OP_WRITE] = {
Jens Axboe3a6820f2019-12-22 15:19:35 -0700984 .needs_file = 1,
985 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700986 .pollout = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600987 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700988 .async_size = sizeof(struct io_async_rw),
Jens Axboe3a6820f2019-12-22 15:19:35 -0700989 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300990 [IORING_OP_FADVISE] = {
Jens Axboe4840e412019-12-25 22:03:45 -0700991 .needs_file = 1,
992 },
Jens Axboe44526be2021-02-15 13:32:18 -0700993 [IORING_OP_MADVISE] = {},
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300994 [IORING_OP_SEND] = {
Jens Axboefddafac2020-01-04 20:19:44 -0700995 .needs_file = 1,
996 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700997 .pollout = 1,
Jens Axboefddafac2020-01-04 20:19:44 -0700998 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300999 [IORING_OP_RECV] = {
Jens Axboefddafac2020-01-04 20:19:44 -07001000 .needs_file = 1,
1001 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -07001002 .pollin = 1,
Jens Axboebcda7ba2020-02-23 16:42:51 -07001003 .buffer_select = 1,
Jens Axboefddafac2020-01-04 20:19:44 -07001004 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +03001005 [IORING_OP_OPENAT2] = {
Jens Axboecebdb982020-01-08 17:59:24 -07001006 },
Jens Axboe3e4827b2020-01-08 15:18:09 -07001007 [IORING_OP_EPOLL_CTL] = {
1008 .unbound_nonreg_file = 1,
Jens Axboe3e4827b2020-01-08 15:18:09 -07001009 },
Pavel Begunkov7d67af22020-02-24 11:32:45 +03001010 [IORING_OP_SPLICE] = {
1011 .needs_file = 1,
1012 .hash_reg_file = 1,
1013 .unbound_nonreg_file = 1,
Jens Axboeddf0322d2020-02-23 16:41:33 -07001014 },
1015 [IORING_OP_PROVIDE_BUFFERS] = {},
Jens Axboe067524e2020-03-02 16:32:28 -07001016 [IORING_OP_REMOVE_BUFFERS] = {},
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03001017 [IORING_OP_TEE] = {
1018 .needs_file = 1,
1019 .hash_reg_file = 1,
1020 .unbound_nonreg_file = 1,
1021 },
Jens Axboe36f4fa62020-09-05 11:14:22 -06001022 [IORING_OP_SHUTDOWN] = {
1023 .needs_file = 1,
1024 },
Jens Axboe44526be2021-02-15 13:32:18 -07001025 [IORING_OP_RENAMEAT] = {},
1026 [IORING_OP_UNLINKAT] = {},
Jens Axboed3656342019-12-18 09:50:26 -07001027};
1028
Pavel Begunkov7a612352021-03-09 00:37:59 +00001029static bool io_disarm_next(struct io_kiocb *req);
Pavel Begunkovd56d9382021-03-06 11:02:13 +00001030static void io_uring_del_task_file(unsigned long index);
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00001031static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
1032 struct task_struct *task,
1033 struct files_struct *files);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07001034static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx);
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01001035static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx);
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00001036
Pavel Begunkovff6421642021-04-11 01:46:32 +01001037static bool io_cqring_fill_event(struct io_kiocb *req, long res, unsigned cflags);
Jackie Liuec9c02a2019-11-08 23:50:36 +08001038static void io_put_req(struct io_kiocb *req);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001039static void io_put_req_deferred(struct io_kiocb *req, int nr);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001040static void io_dismantle_req(struct io_kiocb *req);
1041static void io_put_task(struct task_struct *task, int nr);
Jens Axboe94ae5e72019-11-14 19:39:52 -07001042static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req);
1043static void io_queue_linked_timeout(struct io_kiocb *req);
Jens Axboe05f3fb32019-12-09 11:22:50 -07001044static int __io_sqe_files_update(struct io_ring_ctx *ctx,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001045 struct io_uring_rsrc_update *ip,
Jens Axboe05f3fb32019-12-09 11:22:50 -07001046 unsigned nr_args);
Pavel Begunkov68fb8972021-03-19 17:22:41 +00001047static void io_clean_op(struct io_kiocb *req);
Pavel Begunkov8371adf2020-10-10 18:34:08 +01001048static struct file *io_file_get(struct io_submit_state *state,
1049 struct io_kiocb *req, int fd, bool fixed);
Pavel Begunkovc5eef2b942021-02-10 00:03:22 +00001050static void __io_queue_sqe(struct io_kiocb *req);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001051static void io_rsrc_put_work(struct work_struct *work);
Jens Axboede0617e2019-04-06 21:51:27 -06001052
Pavel Begunkov907d1df2021-01-26 23:35:10 +00001053static void io_req_task_queue(struct io_kiocb *req);
Jens Axboe65453d12021-02-10 00:03:21 +00001054static void io_submit_flush_completions(struct io_comp_state *cs,
1055 struct io_ring_ctx *ctx);
Jens Axboe50826202021-02-23 09:02:26 -07001056static bool io_poll_remove_waitqs(struct io_kiocb *req);
Pavel Begunkov179ae0d2021-02-28 22:35:20 +00001057static int io_req_prep_async(struct io_kiocb *req);
Jens Axboe9a56a232019-01-09 09:06:50 -07001058
Jens Axboe2b188cc2019-01-07 10:46:33 -07001059static struct kmem_cache *req_cachep;
1060
Jens Axboe09186822020-10-13 15:01:40 -06001061static const struct file_operations io_uring_fops;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001062
1063struct sock *io_uring_get_socket(struct file *file)
1064{
1065#if defined(CONFIG_UNIX)
1066 if (file->f_op == &io_uring_fops) {
1067 struct io_ring_ctx *ctx = file->private_data;
1068
1069 return ctx->ring_sock->sk;
1070 }
1071#endif
1072 return NULL;
1073}
1074EXPORT_SYMBOL(io_uring_get_socket);
1075
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001076#define io_for_each_link(pos, head) \
1077 for (pos = (head); pos; pos = pos->link)
1078
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01001079static inline void io_req_set_rsrc_node(struct io_kiocb *req)
Jens Axboec40f6372020-06-25 15:39:59 -06001080{
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001081 struct io_ring_ctx *ctx = req->ctx;
1082
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001083 if (!req->fixed_rsrc_refs) {
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01001084 req->fixed_rsrc_refs = &ctx->rsrc_node->refs;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001085 percpu_ref_get(req->fixed_rsrc_refs);
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001086 }
1087}
1088
Pavel Begunkovf70865d2021-04-11 01:46:40 +01001089static void io_refs_resurrect(struct percpu_ref *ref, struct completion *compl)
1090{
1091 bool got = percpu_ref_tryget(ref);
1092
1093 /* already at zero, wait for ->release() */
1094 if (!got)
1095 wait_for_completion(compl);
1096 percpu_ref_resurrect(ref);
1097 if (got)
1098 percpu_ref_put(ref);
1099}
1100
Pavel Begunkov08d23632020-11-06 13:00:22 +00001101static bool io_match_task(struct io_kiocb *head,
1102 struct task_struct *task,
1103 struct files_struct *files)
1104{
1105 struct io_kiocb *req;
1106
Pavel Begunkov68207682021-03-22 01:58:25 +00001107 if (task && head->task != task)
Pavel Begunkov08d23632020-11-06 13:00:22 +00001108 return false;
1109 if (!files)
1110 return true;
1111
1112 io_for_each_link(req, head) {
Pavel Begunkovb05a1bc2021-03-04 13:59:24 +00001113 if (req->flags & REQ_F_INFLIGHT)
Jens Axboe02a13672021-01-23 15:49:31 -07001114 return true;
Pavel Begunkov08d23632020-11-06 13:00:22 +00001115 }
1116 return false;
1117}
1118
Jens Axboec40f6372020-06-25 15:39:59 -06001119static inline void req_set_fail_links(struct io_kiocb *req)
1120{
Pavel Begunkove4335ed2021-04-11 01:46:39 +01001121 if (req->flags & REQ_F_LINK)
Jens Axboec40f6372020-06-25 15:39:59 -06001122 req->flags |= REQ_F_FAIL_LINK;
1123}
Jens Axboe4a38aed22020-05-14 17:21:15 -06001124
Jens Axboe2b188cc2019-01-07 10:46:33 -07001125static void io_ring_ctx_ref_free(struct percpu_ref *ref)
1126{
1127 struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs);
1128
Jens Axboe0f158b42020-05-14 17:18:39 -06001129 complete(&ctx->ref_comp);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001130}
1131
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03001132static inline bool io_is_timeout_noseq(struct io_kiocb *req)
1133{
1134 return !req->timeout.off;
1135}
1136
Jens Axboe2b188cc2019-01-07 10:46:33 -07001137static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
1138{
1139 struct io_ring_ctx *ctx;
Jens Axboe78076bb2019-12-04 19:56:40 -07001140 int hash_bits;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001141
1142 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1143 if (!ctx)
1144 return NULL;
1145
Jens Axboe78076bb2019-12-04 19:56:40 -07001146 /*
1147 * Use 5 bits less than the max cq entries, that should give us around
1148 * 32 entries per hash list if totally full and uniformly spread.
1149 */
1150 hash_bits = ilog2(p->cq_entries);
1151 hash_bits -= 5;
1152 if (hash_bits <= 0)
1153 hash_bits = 1;
1154 ctx->cancel_hash_bits = hash_bits;
1155 ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head),
1156 GFP_KERNEL);
1157 if (!ctx->cancel_hash)
1158 goto err;
1159 __hash_init(ctx->cancel_hash, 1U << hash_bits);
1160
Roman Gushchin21482892019-05-07 10:01:48 -07001161 if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free,
Jens Axboe206aefd2019-11-07 18:27:42 -07001162 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL))
1163 goto err;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001164
1165 ctx->flags = p->flags;
Jens Axboe90554202020-09-03 12:12:41 -06001166 init_waitqueue_head(&ctx->sqo_sq_wait);
Jens Axboe69fb2132020-09-14 11:16:23 -06001167 INIT_LIST_HEAD(&ctx->sqd_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001168 init_waitqueue_head(&ctx->cq_wait);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001169 INIT_LIST_HEAD(&ctx->cq_overflow_list);
Jens Axboe0f158b42020-05-14 17:18:39 -06001170 init_completion(&ctx->ref_comp);
Jens Axboe9e15c3a2021-03-13 12:29:43 -07001171 xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1);
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00001172 xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001173 mutex_init(&ctx->uring_lock);
1174 init_waitqueue_head(&ctx->wait);
1175 spin_lock_init(&ctx->completion_lock);
Pavel Begunkov540e32a2020-07-13 23:37:09 +03001176 INIT_LIST_HEAD(&ctx->iopoll_list);
Jens Axboede0617e2019-04-06 21:51:27 -06001177 INIT_LIST_HEAD(&ctx->defer_list);
Jens Axboe5262f562019-09-17 12:26:57 -06001178 INIT_LIST_HEAD(&ctx->timeout_list);
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00001179 spin_lock_init(&ctx->rsrc_ref_lock);
1180 INIT_LIST_HEAD(&ctx->rsrc_ref_list);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001181 INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work);
1182 init_llist_head(&ctx->rsrc_put_llist);
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00001183 INIT_LIST_HEAD(&ctx->tctx_list);
Jens Axboe1b4c3512021-02-10 00:03:19 +00001184 INIT_LIST_HEAD(&ctx->submit_state.comp.free_list);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001185 INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001186 return ctx;
Jens Axboe206aefd2019-11-07 18:27:42 -07001187err:
Jens Axboe78076bb2019-12-04 19:56:40 -07001188 kfree(ctx->cancel_hash);
Jens Axboe206aefd2019-11-07 18:27:42 -07001189 kfree(ctx);
1190 return NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001191}
1192
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001193static bool req_need_defer(struct io_kiocb *req, u32 seq)
Jens Axboede0617e2019-04-06 21:51:27 -06001194{
Jens Axboe2bc99302020-07-09 09:43:27 -06001195 if (unlikely(req->flags & REQ_F_IO_DRAIN)) {
1196 struct io_ring_ctx *ctx = req->ctx;
Jackie Liua197f662019-11-08 08:09:12 -07001197
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001198 return seq != ctx->cached_cq_tail
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001199 + READ_ONCE(ctx->cached_cq_overflow);
Jens Axboe2bc99302020-07-09 09:43:27 -06001200 }
Jens Axboe7adf4ea2019-10-10 21:42:58 -06001201
Bob Liu9d858b22019-11-13 18:06:25 +08001202 return false;
Jens Axboe7adf4ea2019-10-10 21:42:58 -06001203}
1204
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00001205static void io_req_track_inflight(struct io_kiocb *req)
1206{
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00001207 if (!(req->flags & REQ_F_INFLIGHT)) {
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00001208 req->flags |= REQ_F_INFLIGHT;
Pavel Begunkovb303fe22021-04-11 01:46:26 +01001209 atomic_inc(&current->io_uring->inflight_tracked);
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00001210 }
1211}
1212
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001213static void io_prep_async_work(struct io_kiocb *req)
Jens Axboe561fb042019-10-24 07:25:42 -06001214{
Jens Axboed3656342019-12-18 09:50:26 -07001215 const struct io_op_def *def = &io_op_defs[req->opcode];
Pavel Begunkov23329512020-10-10 18:34:06 +01001216 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe54a91f32019-09-10 09:15:04 -06001217
Jens Axboe003e8dc2021-03-06 09:22:27 -07001218 if (!req->work.creds)
1219 req->work.creds = get_current_cred();
1220
Pavel Begunkove1d675d2021-03-22 01:58:29 +00001221 req->work.list.next = NULL;
1222 req->work.flags = 0;
Pavel Begunkovfeaadc42020-10-22 16:47:16 +01001223 if (req->flags & REQ_F_FORCE_ASYNC)
1224 req->work.flags |= IO_WQ_WORK_CONCURRENT;
1225
Jens Axboed3656342019-12-18 09:50:26 -07001226 if (req->flags & REQ_F_ISREG) {
Pavel Begunkov23329512020-10-10 18:34:06 +01001227 if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL))
Pavel Begunkov8766dd52020-03-14 00:31:04 +03001228 io_wq_hash_work(&req->work, file_inode(req->file));
Jens Axboe4b982bd2021-04-01 08:38:34 -06001229 } else if (!req->file || !S_ISBLK(file_inode(req->file)->i_mode)) {
Jens Axboed3656342019-12-18 09:50:26 -07001230 if (def->unbound_nonreg_file)
Jens Axboe3529d8c2019-12-19 18:24:38 -07001231 req->work.flags |= IO_WQ_WORK_UNBOUND;
Jens Axboe54a91f32019-09-10 09:15:04 -06001232 }
Pavel Begunkove1d675d2021-03-22 01:58:29 +00001233
1234 switch (req->opcode) {
1235 case IORING_OP_SPLICE:
1236 case IORING_OP_TEE:
Pavel Begunkove1d675d2021-03-22 01:58:29 +00001237 if (!S_ISREG(file_inode(req->splice.file_in)->i_mode))
1238 req->work.flags |= IO_WQ_WORK_UNBOUND;
1239 break;
1240 }
Jens Axboe561fb042019-10-24 07:25:42 -06001241}
1242
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001243static void io_prep_async_link(struct io_kiocb *req)
1244{
1245 struct io_kiocb *cur;
1246
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001247 io_for_each_link(cur, req)
1248 io_prep_async_work(cur);
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001249}
1250
Pavel Begunkovebf93662021-03-01 18:20:47 +00001251static void io_queue_async_work(struct io_kiocb *req)
Jens Axboe561fb042019-10-24 07:25:42 -06001252{
Jackie Liua197f662019-11-08 08:09:12 -07001253 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001254 struct io_kiocb *link = io_prep_linked_timeout(req);
Jens Axboe5aa75ed2021-02-16 12:56:50 -07001255 struct io_uring_task *tctx = req->task->io_uring;
Jens Axboe561fb042019-10-24 07:25:42 -06001256
Jens Axboe3bfe6102021-02-16 14:15:30 -07001257 BUG_ON(!tctx);
1258 BUG_ON(!tctx->io_wq);
Jens Axboe561fb042019-10-24 07:25:42 -06001259
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001260 /* init ->work of the whole link before punting */
1261 io_prep_async_link(req);
Pavel Begunkovd07f1e8a2021-03-22 01:45:58 +00001262 trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req,
1263 &req->work, req->flags);
Pavel Begunkovebf93662021-03-01 18:20:47 +00001264 io_wq_enqueue(tctx->io_wq, &req->work);
Jens Axboe7271ef32020-08-10 09:55:22 -06001265 if (link)
1266 io_queue_linked_timeout(link);
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001267}
1268
Pavel Begunkov1ee41602021-03-25 18:32:42 +00001269static void io_kill_timeout(struct io_kiocb *req, int status)
Pavel Begunkov8c855882021-04-13 02:58:41 +01001270 __must_hold(&req->ctx->completion_lock)
Jens Axboe5262f562019-09-17 12:26:57 -06001271{
Jens Axboee8c2bc12020-08-15 18:44:09 -07001272 struct io_timeout_data *io = req->async_data;
Jens Axboe5262f562019-09-17 12:26:57 -06001273
Pavel Begunkovfd9c7bc2021-04-13 02:58:42 +01001274 if (hrtimer_try_to_cancel(&io->timer) != -1) {
Pavel Begunkov01cec8c2020-07-30 18:43:50 +03001275 atomic_set(&req->ctx->cq_timeouts,
1276 atomic_read(&req->ctx->cq_timeouts) + 1);
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001277 list_del_init(&req->timeout.list);
Pavel Begunkovff6421642021-04-11 01:46:32 +01001278 io_cqring_fill_event(req, status, 0);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001279 io_put_req_deferred(req, 1);
Jens Axboe5262f562019-09-17 12:26:57 -06001280 }
1281}
1282
Pavel Begunkov04518942020-05-26 20:34:05 +03001283static void __io_queue_deferred(struct io_ring_ctx *ctx)
1284{
1285 do {
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001286 struct io_defer_entry *de = list_first_entry(&ctx->defer_list,
1287 struct io_defer_entry, list);
Pavel Begunkov04518942020-05-26 20:34:05 +03001288
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001289 if (req_need_defer(de->req, de->seq))
Pavel Begunkov04518942020-05-26 20:34:05 +03001290 break;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001291 list_del_init(&de->list);
Pavel Begunkov907d1df2021-01-26 23:35:10 +00001292 io_req_task_queue(de->req);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001293 kfree(de);
Pavel Begunkov04518942020-05-26 20:34:05 +03001294 } while (!list_empty(&ctx->defer_list));
1295}
1296
Pavel Begunkov360428f2020-05-30 14:54:17 +03001297static void io_flush_timeouts(struct io_ring_ctx *ctx)
1298{
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001299 u32 seq;
1300
1301 if (list_empty(&ctx->timeout_list))
1302 return;
1303
1304 seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
1305
1306 do {
1307 u32 events_needed, events_got;
Pavel Begunkov360428f2020-05-30 14:54:17 +03001308 struct io_kiocb *req = list_first_entry(&ctx->timeout_list,
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001309 struct io_kiocb, timeout.list);
Pavel Begunkov360428f2020-05-30 14:54:17 +03001310
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03001311 if (io_is_timeout_noseq(req))
Pavel Begunkov360428f2020-05-30 14:54:17 +03001312 break;
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001313
1314 /*
1315 * Since seq can easily wrap around over time, subtract
1316 * the last seq at which timeouts were flushed before comparing.
1317 * Assuming not more than 2^31-1 events have happened since,
1318 * these subtractions won't have wrapped, so we can check if
1319 * target is in [last_seq, current_seq] by comparing the two.
1320 */
1321 events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush;
1322 events_got = seq - ctx->cq_last_tm_flush;
1323 if (events_got < events_needed)
Pavel Begunkov360428f2020-05-30 14:54:17 +03001324 break;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03001325
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001326 list_del_init(&req->timeout.list);
Pavel Begunkov1ee41602021-03-25 18:32:42 +00001327 io_kill_timeout(req, 0);
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001328 } while (!list_empty(&ctx->timeout_list));
1329
1330 ctx->cq_last_tm_flush = seq;
Pavel Begunkov360428f2020-05-30 14:54:17 +03001331}
1332
Jens Axboede0617e2019-04-06 21:51:27 -06001333static void io_commit_cqring(struct io_ring_ctx *ctx)
1334{
Pavel Begunkov360428f2020-05-30 14:54:17 +03001335 io_flush_timeouts(ctx);
Pavel Begunkovec30e042021-01-19 13:32:38 +00001336
1337 /* order cqe stores with ring update */
1338 smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail);
Jens Axboede0617e2019-04-06 21:51:27 -06001339
Pavel Begunkov04518942020-05-26 20:34:05 +03001340 if (unlikely(!list_empty(&ctx->defer_list)))
1341 __io_queue_deferred(ctx);
Jens Axboede0617e2019-04-06 21:51:27 -06001342}
1343
Jens Axboe90554202020-09-03 12:12:41 -06001344static inline bool io_sqring_full(struct io_ring_ctx *ctx)
1345{
1346 struct io_rings *r = ctx->rings;
1347
1348 return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries;
1349}
1350
Pavel Begunkov888aae22021-01-19 13:32:39 +00001351static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx)
1352{
1353 return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head);
1354}
1355
Pavel Begunkov8d133262021-04-11 01:46:33 +01001356static inline struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001357{
Hristo Venev75b28af2019-08-26 17:23:46 +00001358 struct io_rings *rings = ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001359 unsigned tail;
1360
Stefan Bühler115e12e2019-04-24 23:54:18 +02001361 /*
1362 * writes to the cq entry need to come after reading head; the
1363 * control dependency is enough as we're using WRITE_ONCE to
1364 * fill the cq entry
1365 */
Pavel Begunkov888aae22021-01-19 13:32:39 +00001366 if (__io_cqring_events(ctx) == rings->cq_ring_entries)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001367 return NULL;
1368
Pavel Begunkov888aae22021-01-19 13:32:39 +00001369 tail = ctx->cached_cq_tail++;
Hristo Venev75b28af2019-08-26 17:23:46 +00001370 return &rings->cqes[tail & ctx->cq_mask];
Jens Axboe2b188cc2019-01-07 10:46:33 -07001371}
1372
Jens Axboef2842ab2020-01-08 11:04:00 -07001373static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx)
1374{
Pavel Begunkov44c769d2021-04-11 01:46:31 +01001375 if (likely(!ctx->cq_ev_fd))
Jens Axboef0b493e2020-02-01 21:30:11 -07001376 return false;
Stefano Garzarella7e55a192020-05-15 18:38:05 +02001377 if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED)
1378 return false;
Pavel Begunkov44c769d2021-04-11 01:46:31 +01001379 return !ctx->eventfd_async || io_wq_current_is_worker();
Jens Axboef2842ab2020-01-08 11:04:00 -07001380}
1381
Jens Axboeb41e9852020-02-17 09:52:41 -07001382static void io_cqring_ev_posted(struct io_ring_ctx *ctx)
Jens Axboe8c838782019-03-12 15:48:16 -06001383{
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001384 /* see waitqueue_active() comment */
1385 smp_mb();
1386
Jens Axboe8c838782019-03-12 15:48:16 -06001387 if (waitqueue_active(&ctx->wait))
1388 wake_up(&ctx->wait);
Jens Axboe534ca6d2020-09-02 13:52:19 -06001389 if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait))
1390 wake_up(&ctx->sq_data->wait);
Jens Axboeb41e9852020-02-17 09:52:41 -07001391 if (io_should_trigger_evfd(ctx))
Jens Axboe9b402842019-04-11 11:45:41 -06001392 eventfd_signal(ctx->cq_ev_fd, 1);
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001393 if (waitqueue_active(&ctx->cq_wait)) {
Pavel Begunkov4aa84f22021-01-07 03:15:42 +00001394 wake_up_interruptible(&ctx->cq_wait);
1395 kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
1396 }
Jens Axboe8c838782019-03-12 15:48:16 -06001397}
1398
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001399static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx)
1400{
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001401 /* see waitqueue_active() comment */
1402 smp_mb();
1403
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001404 if (ctx->flags & IORING_SETUP_SQPOLL) {
1405 if (waitqueue_active(&ctx->wait))
1406 wake_up(&ctx->wait);
1407 }
1408 if (io_should_trigger_evfd(ctx))
1409 eventfd_signal(ctx->cq_ev_fd, 1);
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001410 if (waitqueue_active(&ctx->cq_wait)) {
Pavel Begunkov4aa84f22021-01-07 03:15:42 +00001411 wake_up_interruptible(&ctx->cq_wait);
1412 kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
1413 }
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001414}
1415
Jens Axboec4a2ed72019-11-21 21:01:26 -07001416/* Returns true if there are no backlogged entries after the flush */
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00001417static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001418{
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001419 struct io_rings *rings = ctx->rings;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001420 unsigned long flags;
Jens Axboeb18032b2021-01-24 16:58:56 -07001421 bool all_flushed, posted;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001422
Pavel Begunkove23de152020-12-17 00:24:37 +00001423 if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries)
1424 return false;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001425
Jens Axboeb18032b2021-01-24 16:58:56 -07001426 posted = false;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001427 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00001428 while (!list_empty(&ctx->cq_overflow_list)) {
1429 struct io_uring_cqe *cqe = io_get_cqring(ctx);
1430 struct io_overflow_cqe *ocqe;
Jens Axboee6c8aa92020-09-28 13:10:13 -06001431
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001432 if (!cqe && !force)
1433 break;
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00001434 ocqe = list_first_entry(&ctx->cq_overflow_list,
1435 struct io_overflow_cqe, list);
1436 if (cqe)
1437 memcpy(cqe, &ocqe->cqe, sizeof(*cqe));
1438 else
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001439 WRITE_ONCE(ctx->rings->cq_overflow,
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00001440 ++ctx->cached_cq_overflow);
Jens Axboeb18032b2021-01-24 16:58:56 -07001441 posted = true;
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00001442 list_del(&ocqe->list);
1443 kfree(ocqe);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001444 }
1445
Pavel Begunkov09e88402020-12-17 00:24:38 +00001446 all_flushed = list_empty(&ctx->cq_overflow_list);
1447 if (all_flushed) {
1448 clear_bit(0, &ctx->sq_check_overflow);
1449 clear_bit(0, &ctx->cq_check_overflow);
1450 ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW;
1451 }
Pavel Begunkov46930142020-07-30 18:43:49 +03001452
Jens Axboeb18032b2021-01-24 16:58:56 -07001453 if (posted)
1454 io_commit_cqring(ctx);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001455 spin_unlock_irqrestore(&ctx->completion_lock, flags);
Jens Axboeb18032b2021-01-24 16:58:56 -07001456 if (posted)
1457 io_cqring_ev_posted(ctx);
Pavel Begunkov09e88402020-12-17 00:24:38 +00001458 return all_flushed;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001459}
1460
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00001461static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force)
Pavel Begunkov6c503152021-01-04 20:36:36 +00001462{
Jens Axboeca0a2652021-03-04 17:15:48 -07001463 bool ret = true;
1464
Pavel Begunkov6c503152021-01-04 20:36:36 +00001465 if (test_bit(0, &ctx->cq_check_overflow)) {
1466 /* iopoll syncs against uring_lock, not completion_lock */
1467 if (ctx->flags & IORING_SETUP_IOPOLL)
1468 mutex_lock(&ctx->uring_lock);
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00001469 ret = __io_cqring_overflow_flush(ctx, force);
Pavel Begunkov6c503152021-01-04 20:36:36 +00001470 if (ctx->flags & IORING_SETUP_IOPOLL)
1471 mutex_unlock(&ctx->uring_lock);
1472 }
Jens Axboeca0a2652021-03-04 17:15:48 -07001473
1474 return ret;
Pavel Begunkov6c503152021-01-04 20:36:36 +00001475}
1476
Jens Axboeabc54d62021-02-24 13:32:30 -07001477/*
1478 * Shamelessly stolen from the mm implementation of page reference checking,
1479 * see commit f958d7b528b1 for details.
1480 */
1481#define req_ref_zero_or_close_to_overflow(req) \
1482 ((unsigned int) atomic_read(&(req->refs)) + 127u <= 127u)
1483
Jens Axboede9b4cc2021-02-24 13:28:27 -07001484static inline bool req_ref_inc_not_zero(struct io_kiocb *req)
1485{
Jens Axboeabc54d62021-02-24 13:32:30 -07001486 return atomic_inc_not_zero(&req->refs);
Jens Axboede9b4cc2021-02-24 13:28:27 -07001487}
1488
1489static inline bool req_ref_sub_and_test(struct io_kiocb *req, int refs)
1490{
Jens Axboeabc54d62021-02-24 13:32:30 -07001491 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
1492 return atomic_sub_and_test(refs, &req->refs);
Jens Axboede9b4cc2021-02-24 13:28:27 -07001493}
1494
1495static inline bool req_ref_put_and_test(struct io_kiocb *req)
1496{
Jens Axboeabc54d62021-02-24 13:32:30 -07001497 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
1498 return atomic_dec_and_test(&req->refs);
Jens Axboede9b4cc2021-02-24 13:28:27 -07001499}
1500
1501static inline void req_ref_put(struct io_kiocb *req)
1502{
Jens Axboeabc54d62021-02-24 13:32:30 -07001503 WARN_ON_ONCE(req_ref_put_and_test(req));
Jens Axboede9b4cc2021-02-24 13:28:27 -07001504}
1505
1506static inline void req_ref_get(struct io_kiocb *req)
1507{
Jens Axboeabc54d62021-02-24 13:32:30 -07001508 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
1509 atomic_inc(&req->refs);
Jens Axboede9b4cc2021-02-24 13:28:27 -07001510}
1511
Pavel Begunkov8d133262021-04-11 01:46:33 +01001512static bool io_cqring_event_overflow(struct io_kiocb *req, long res,
1513 unsigned int cflags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001514{
Jens Axboe78e19bb2019-11-06 15:21:34 -07001515 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovcce4b8b2021-04-13 02:58:44 +01001516 struct io_overflow_cqe *ocqe;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001517
Pavel Begunkovcce4b8b2021-04-13 02:58:44 +01001518 ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT);
1519 if (!ocqe) {
1520 /*
1521 * If we're in ring overflow flush mode, or in task cancel mode,
1522 * or cannot allocate an overflow entry, then we need to drop it
1523 * on the floor.
1524 */
1525 WRITE_ONCE(ctx->rings->cq_overflow, ++ctx->cached_cq_overflow);
1526 return false;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001527 }
Pavel Begunkovcce4b8b2021-04-13 02:58:44 +01001528 if (list_empty(&ctx->cq_overflow_list)) {
1529 set_bit(0, &ctx->sq_check_overflow);
1530 set_bit(0, &ctx->cq_check_overflow);
1531 ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW;
1532 }
1533 ocqe->cqe.user_data = req->user_data;
1534 ocqe->cqe.res = res;
1535 ocqe->cqe.flags = cflags;
1536 list_add_tail(&ocqe->list, &ctx->cq_overflow_list);
1537 return true;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001538}
1539
Pavel Begunkov8d133262021-04-11 01:46:33 +01001540static inline bool __io_cqring_fill_event(struct io_kiocb *req, long res,
1541 unsigned int cflags)
1542{
1543 struct io_ring_ctx *ctx = req->ctx;
1544 struct io_uring_cqe *cqe;
1545
1546 trace_io_uring_complete(ctx, req->user_data, res, cflags);
1547
1548 /*
1549 * If we can't get a cq entry, userspace overflowed the
1550 * submission (by quite a lot). Increment the overflow count in
1551 * the ring.
1552 */
1553 cqe = io_get_cqring(ctx);
1554 if (likely(cqe)) {
1555 WRITE_ONCE(cqe->user_data, req->user_data);
1556 WRITE_ONCE(cqe->res, res);
1557 WRITE_ONCE(cqe->flags, cflags);
1558 return true;
1559 }
1560 return io_cqring_event_overflow(req, res, cflags);
1561}
1562
1563/* not as hot to bloat with inlining */
1564static noinline bool io_cqring_fill_event(struct io_kiocb *req, long res,
1565 unsigned int cflags)
1566{
1567 return __io_cqring_fill_event(req, res, cflags);
1568}
1569
Pavel Begunkov7a612352021-03-09 00:37:59 +00001570static void io_req_complete_post(struct io_kiocb *req, long res,
1571 unsigned int cflags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001572{
Jens Axboe78e19bb2019-11-06 15:21:34 -07001573 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001574 unsigned long flags;
1575
1576 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkov8d133262021-04-11 01:46:33 +01001577 __io_cqring_fill_event(req, res, cflags);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001578 /*
1579 * If we're the last reference to this request, add to our locked
1580 * free_list cache.
1581 */
Jens Axboede9b4cc2021-02-24 13:28:27 -07001582 if (req_ref_put_and_test(req)) {
Jens Axboec7dae4b2021-02-09 19:53:37 -07001583 struct io_comp_state *cs = &ctx->submit_state.comp;
1584
Pavel Begunkov7a612352021-03-09 00:37:59 +00001585 if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) {
1586 if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK))
1587 io_disarm_next(req);
1588 if (req->link) {
1589 io_req_task_queue(req->link);
1590 req->link = NULL;
1591 }
1592 }
Jens Axboec7dae4b2021-02-09 19:53:37 -07001593 io_dismantle_req(req);
1594 io_put_task(req->task, 1);
1595 list_add(&req->compl.list, &cs->locked_free_list);
1596 cs->locked_free_nr++;
Pavel Begunkov180f8292021-03-14 20:57:09 +00001597 } else {
1598 if (!percpu_ref_tryget(&ctx->refs))
1599 req = NULL;
1600 }
Pavel Begunkov7a612352021-03-09 00:37:59 +00001601 io_commit_cqring(ctx);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001602 spin_unlock_irqrestore(&ctx->completion_lock, flags);
Pavel Begunkov7a612352021-03-09 00:37:59 +00001603
Pavel Begunkov180f8292021-03-14 20:57:09 +00001604 if (req) {
1605 io_cqring_ev_posted(ctx);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001606 percpu_ref_put(&ctx->refs);
Pavel Begunkov180f8292021-03-14 20:57:09 +00001607 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07001608}
1609
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001610static void io_req_complete_state(struct io_kiocb *req, long res,
Pavel Begunkov889fca72021-02-10 00:03:09 +00001611 unsigned int cflags)
Jens Axboebcda7ba2020-02-23 16:42:51 -07001612{
Pavel Begunkov68fb8972021-03-19 17:22:41 +00001613 if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED))
1614 io_clean_op(req);
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001615 req->result = res;
1616 req->compl.cflags = cflags;
Pavel Begunkove342c802021-01-19 13:32:47 +00001617 req->flags |= REQ_F_COMPLETE_INLINE;
Jens Axboee1e16092020-06-22 09:17:17 -06001618}
Jens Axboe2b188cc2019-01-07 10:46:33 -07001619
Pavel Begunkov889fca72021-02-10 00:03:09 +00001620static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags,
1621 long res, unsigned cflags)
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001622{
Pavel Begunkov889fca72021-02-10 00:03:09 +00001623 if (issue_flags & IO_URING_F_COMPLETE_DEFER)
1624 io_req_complete_state(req, res, cflags);
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001625 else
Jens Axboec7dae4b2021-02-09 19:53:37 -07001626 io_req_complete_post(req, res, cflags);
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001627}
Jens Axboebcda7ba2020-02-23 16:42:51 -07001628
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001629static inline void io_req_complete(struct io_kiocb *req, long res)
Jens Axboee1e16092020-06-22 09:17:17 -06001630{
Pavel Begunkov889fca72021-02-10 00:03:09 +00001631 __io_req_complete(req, 0, res, 0);
Jens Axboebcda7ba2020-02-23 16:42:51 -07001632}
1633
Pavel Begunkovf41db2732021-02-28 22:35:12 +00001634static void io_req_complete_failed(struct io_kiocb *req, long res)
1635{
1636 req_set_fail_links(req);
1637 io_put_req(req);
1638 io_req_complete_post(req, res, 0);
1639}
1640
Pavel Begunkovdac7a092021-03-19 17:22:39 +00001641static void io_flush_cached_locked_reqs(struct io_ring_ctx *ctx,
1642 struct io_comp_state *cs)
1643{
1644 spin_lock_irq(&ctx->completion_lock);
1645 list_splice_init(&cs->locked_free_list, &cs->free_list);
1646 cs->locked_free_nr = 0;
1647 spin_unlock_irq(&ctx->completion_lock);
1648}
1649
Pavel Begunkovdd78f492021-03-19 17:22:35 +00001650/* Returns true IFF there are requests in the cache */
Jens Axboec7dae4b2021-02-09 19:53:37 -07001651static bool io_flush_cached_reqs(struct io_ring_ctx *ctx)
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001652{
Jens Axboec7dae4b2021-02-09 19:53:37 -07001653 struct io_submit_state *state = &ctx->submit_state;
1654 struct io_comp_state *cs = &state->comp;
Pavel Begunkovdd78f492021-03-19 17:22:35 +00001655 int nr;
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001656
Jens Axboec7dae4b2021-02-09 19:53:37 -07001657 /*
1658 * If we have more than a batch's worth of requests in our IRQ side
1659 * locked cache, grab the lock and move them over to our submission
1660 * side cache.
1661 */
Pavel Begunkovdac7a092021-03-19 17:22:39 +00001662 if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH)
1663 io_flush_cached_locked_reqs(ctx, cs);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001664
Pavel Begunkovdd78f492021-03-19 17:22:35 +00001665 nr = state->free_reqs;
Jens Axboec7dae4b2021-02-09 19:53:37 -07001666 while (!list_empty(&cs->free_list)) {
Pavel Begunkovdd78f492021-03-19 17:22:35 +00001667 struct io_kiocb *req = list_first_entry(&cs->free_list,
1668 struct io_kiocb, compl.list);
1669
Jens Axboe2b188cc2019-01-07 10:46:33 -07001670 list_del(&req->compl.list);
Pavel Begunkovdd78f492021-03-19 17:22:35 +00001671 state->reqs[nr++] = req;
1672 if (nr == ARRAY_SIZE(state->reqs))
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001673 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001674 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07001675
Pavel Begunkovdd78f492021-03-19 17:22:35 +00001676 state->free_reqs = nr;
1677 return nr != 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001678}
1679
Pavel Begunkov258b29a2021-02-10 00:03:10 +00001680static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001681{
Pavel Begunkov258b29a2021-02-10 00:03:10 +00001682 struct io_submit_state *state = &ctx->submit_state;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001683
Pavel Begunkovbf019da2021-02-10 00:03:17 +00001684 BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs));
Jens Axboe2b188cc2019-01-07 10:46:33 -07001685
Pavel Begunkovf6b6c7d2020-06-21 13:09:53 +03001686 if (!state->free_reqs) {
Pavel Begunkov291b2822020-09-30 22:57:01 +03001687 gfp_t gfp = GFP_KERNEL | __GFP_NOWARN;
Jens Axboe2579f912019-01-09 09:10:43 -07001688 int ret;
1689
Jens Axboec7dae4b2021-02-09 19:53:37 -07001690 if (io_flush_cached_reqs(ctx))
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001691 goto got_req;
1692
Pavel Begunkovbf019da2021-02-10 00:03:17 +00001693 ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH,
1694 state->reqs);
Jens Axboefd6fab22019-03-14 16:30:06 -06001695
1696 /*
1697 * Bulk alloc is all-or-nothing. If we fail to get a batch,
1698 * retry single alloc to be on the safe side.
1699 */
1700 if (unlikely(ret <= 0)) {
1701 state->reqs[0] = kmem_cache_alloc(req_cachep, gfp);
1702 if (!state->reqs[0])
Pavel Begunkov3893f392021-02-10 00:03:15 +00001703 return NULL;
Jens Axboefd6fab22019-03-14 16:30:06 -06001704 ret = 1;
1705 }
Pavel Begunkov291b2822020-09-30 22:57:01 +03001706 state->free_reqs = ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001707 }
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001708got_req:
Pavel Begunkov291b2822020-09-30 22:57:01 +03001709 state->free_reqs--;
1710 return state->reqs[state->free_reqs];
Jens Axboe2b188cc2019-01-07 10:46:33 -07001711}
1712
Pavel Begunkove1d767f2021-03-19 17:22:43 +00001713static inline void io_put_file(struct file *file)
Pavel Begunkov8da11c12020-02-24 11:32:44 +03001714{
Pavel Begunkove1d767f2021-03-19 17:22:43 +00001715 if (file)
Pavel Begunkov8da11c12020-02-24 11:32:44 +03001716 fput(file);
1717}
1718
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01001719static void io_dismantle_req(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001720{
Pavel Begunkov094bae42021-03-19 17:22:42 +00001721 unsigned int flags = req->flags;
1722
Pavel Begunkove1d767f2021-03-19 17:22:43 +00001723 if (!(flags & REQ_F_FIXED_FILE))
1724 io_put_file(req->file);
Pavel Begunkov094bae42021-03-19 17:22:42 +00001725 if (flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED |
1726 REQ_F_INFLIGHT)) {
1727 io_clean_op(req);
1728
1729 if (req->flags & REQ_F_INFLIGHT) {
Pavel Begunkovb303fe22021-04-11 01:46:26 +01001730 struct io_uring_task *tctx = req->task->io_uring;
Pavel Begunkov094bae42021-03-19 17:22:42 +00001731
Pavel Begunkovb303fe22021-04-11 01:46:26 +01001732 atomic_dec(&tctx->inflight_tracked);
Pavel Begunkov094bae42021-03-19 17:22:42 +00001733 req->flags &= ~REQ_F_INFLIGHT;
1734 }
1735 }
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001736 if (req->fixed_rsrc_refs)
1737 percpu_ref_put(req->fixed_rsrc_refs);
Pavel Begunkov094bae42021-03-19 17:22:42 +00001738 if (req->async_data)
1739 kfree(req->async_data);
Jens Axboe003e8dc2021-03-06 09:22:27 -07001740 if (req->work.creds) {
1741 put_cred(req->work.creds);
1742 req->work.creds = NULL;
1743 }
Pavel Begunkove6543a82020-06-28 12:52:30 +03001744}
Pavel Begunkov2b85edf2019-12-28 14:13:03 +03001745
Pavel Begunkovb23fcf42021-03-01 18:20:48 +00001746/* must to be called somewhat shortly after putting a request */
Pavel Begunkov7c660732021-01-25 11:42:21 +00001747static inline void io_put_task(struct task_struct *task, int nr)
1748{
1749 struct io_uring_task *tctx = task->io_uring;
1750
1751 percpu_counter_sub(&tctx->inflight, nr);
1752 if (unlikely(atomic_read(&tctx->in_idle)))
1753 wake_up(&tctx->wait);
1754 put_task_struct_many(task, nr);
1755}
1756
Pavel Begunkov216578e2020-10-13 09:44:00 +01001757static void __io_free_req(struct io_kiocb *req)
Pavel Begunkove6543a82020-06-28 12:52:30 +03001758{
Jens Axboe51a4cc12020-08-10 10:55:56 -06001759 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovecfc5172020-06-29 13:13:03 +03001760
Pavel Begunkov216578e2020-10-13 09:44:00 +01001761 io_dismantle_req(req);
Pavel Begunkov7c660732021-01-25 11:42:21 +00001762 io_put_task(req->task, 1);
Pavel Begunkove6543a82020-06-28 12:52:30 +03001763
Pavel Begunkov3893f392021-02-10 00:03:15 +00001764 kmem_cache_free(req_cachep, req);
Pavel Begunkovecfc5172020-06-29 13:13:03 +03001765 percpu_ref_put(&ctx->refs);
Jens Axboee65ef562019-03-12 10:16:44 -06001766}
1767
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001768static inline void io_remove_next_linked(struct io_kiocb *req)
1769{
1770 struct io_kiocb *nxt = req->link;
1771
1772 req->link = nxt->link;
1773 nxt->link = NULL;
1774}
1775
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001776static bool io_kill_linked_timeout(struct io_kiocb *req)
1777 __must_hold(&req->ctx->completion_lock)
Jens Axboe9e645e112019-05-10 16:07:28 -06001778{
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001779 struct io_kiocb *link = req->link;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001780
Pavel Begunkov900fad42020-10-19 16:39:16 +01001781 /*
1782 * Can happen if a linked timeout fired and link had been like
1783 * req -> link t-out -> link t-out [-> ...]
1784 */
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001785 if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) {
1786 struct io_timeout_data *io = link->async_data;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001787
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001788 io_remove_next_linked(req);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00001789 link->timeout.head = NULL;
Pavel Begunkovfd9c7bc2021-04-13 02:58:42 +01001790 if (hrtimer_try_to_cancel(&io->timer) != -1) {
Pavel Begunkovff6421642021-04-11 01:46:32 +01001791 io_cqring_fill_event(link, -ECANCELED, 0);
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001792 io_put_req_deferred(link, 1);
Pavel Begunkovd4729fb2021-03-22 01:58:24 +00001793 return true;
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001794 }
1795 }
Pavel Begunkovd4729fb2021-03-22 01:58:24 +00001796 return false;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001797}
1798
Pavel Begunkovd148ca42020-10-18 10:17:39 +01001799static void io_fail_links(struct io_kiocb *req)
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001800 __must_hold(&req->ctx->completion_lock)
Jens Axboe9e645e112019-05-10 16:07:28 -06001801{
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001802 struct io_kiocb *nxt, *link = req->link;
Jens Axboe9e645e112019-05-10 16:07:28 -06001803
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001804 req->link = NULL;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001805 while (link) {
1806 nxt = link->link;
1807 link->link = NULL;
1808
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02001809 trace_io_uring_fail_link(req, link);
Pavel Begunkovff6421642021-04-11 01:46:32 +01001810 io_cqring_fill_event(link, -ECANCELED, 0);
Jens Axboe1575f212021-02-27 15:20:49 -07001811 io_put_req_deferred(link, 2);
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001812 link = nxt;
Jens Axboe9e645e112019-05-10 16:07:28 -06001813 }
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001814}
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001815
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001816static bool io_disarm_next(struct io_kiocb *req)
1817 __must_hold(&req->ctx->completion_lock)
1818{
1819 bool posted = false;
1820
1821 if (likely(req->flags & REQ_F_LINK_TIMEOUT))
1822 posted = io_kill_linked_timeout(req);
Pavel Begunkove4335ed2021-04-11 01:46:39 +01001823 if (unlikely((req->flags & REQ_F_FAIL_LINK) &&
1824 !(req->flags & REQ_F_HARDLINK))) {
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001825 posted |= (req->link != NULL);
1826 io_fail_links(req);
1827 }
1828 return posted;
Jens Axboe9e645e112019-05-10 16:07:28 -06001829}
1830
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001831static struct io_kiocb *__io_req_find_next(struct io_kiocb *req)
Jens Axboe9e645e112019-05-10 16:07:28 -06001832{
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001833 struct io_kiocb *nxt;
Jens Axboe2665abf2019-11-05 12:40:47 -07001834
Jens Axboe9e645e112019-05-10 16:07:28 -06001835 /*
1836 * If LINK is set, we have dependent requests in this chain. If we
1837 * didn't fail this request, queue the first one up, moving any other
1838 * dependencies to the next request. In case of failure, fail the rest
1839 * of the chain.
1840 */
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001841 if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) {
1842 struct io_ring_ctx *ctx = req->ctx;
1843 unsigned long flags;
1844 bool posted;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001845
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001846 spin_lock_irqsave(&ctx->completion_lock, flags);
1847 posted = io_disarm_next(req);
1848 if (posted)
1849 io_commit_cqring(req->ctx);
1850 spin_unlock_irqrestore(&ctx->completion_lock, flags);
1851 if (posted)
1852 io_cqring_ev_posted(ctx);
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001853 }
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001854 nxt = req->link;
1855 req->link = NULL;
1856 return nxt;
Jens Axboe4d7dd462019-11-20 13:03:52 -07001857}
Jens Axboe2665abf2019-11-05 12:40:47 -07001858
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001859static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req)
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001860{
Pavel Begunkovcdbff982021-02-12 18:41:16 +00001861 if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK))))
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001862 return NULL;
1863 return __io_req_find_next(req);
1864}
1865
Pavel Begunkov2c323952021-02-28 22:04:53 +00001866static void ctx_flush_and_put(struct io_ring_ctx *ctx)
1867{
1868 if (!ctx)
1869 return;
1870 if (ctx->submit_state.comp.nr) {
1871 mutex_lock(&ctx->uring_lock);
1872 io_submit_flush_completions(&ctx->submit_state.comp, ctx);
1873 mutex_unlock(&ctx->uring_lock);
1874 }
1875 percpu_ref_put(&ctx->refs);
1876}
1877
Jens Axboe7cbf1722021-02-10 00:03:20 +00001878static bool __tctx_task_work(struct io_uring_task *tctx)
1879{
Jens Axboe65453d12021-02-10 00:03:21 +00001880 struct io_ring_ctx *ctx = NULL;
Jens Axboe7cbf1722021-02-10 00:03:20 +00001881 struct io_wq_work_list list;
1882 struct io_wq_work_node *node;
1883
1884 if (wq_list_empty(&tctx->task_list))
1885 return false;
1886
Jens Axboe0b81e802021-02-16 10:33:53 -07001887 spin_lock_irq(&tctx->task_lock);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001888 list = tctx->task_list;
1889 INIT_WQ_LIST(&tctx->task_list);
Jens Axboe0b81e802021-02-16 10:33:53 -07001890 spin_unlock_irq(&tctx->task_lock);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001891
1892 node = list.first;
1893 while (node) {
1894 struct io_wq_work_node *next = node->next;
1895 struct io_kiocb *req;
1896
1897 req = container_of(node, struct io_kiocb, io_task_work.node);
Pavel Begunkov2c323952021-02-28 22:04:53 +00001898 if (req->ctx != ctx) {
1899 ctx_flush_and_put(ctx);
1900 ctx = req->ctx;
1901 percpu_ref_get(&ctx->refs);
1902 }
1903
Jens Axboe7cbf1722021-02-10 00:03:20 +00001904 req->task_work.func(&req->task_work);
1905 node = next;
Jens Axboe65453d12021-02-10 00:03:21 +00001906 }
1907
Pavel Begunkov2c323952021-02-28 22:04:53 +00001908 ctx_flush_and_put(ctx);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001909 return list.first != NULL;
1910}
1911
1912static void tctx_task_work(struct callback_head *cb)
1913{
1914 struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work);
1915
Jens Axboe1d5f3602021-02-26 14:54:16 -07001916 clear_bit(0, &tctx->task_state);
1917
Jens Axboe7cbf1722021-02-10 00:03:20 +00001918 while (__tctx_task_work(tctx))
1919 cond_resched();
Jens Axboe7cbf1722021-02-10 00:03:20 +00001920}
1921
Pavel Begunkovc15b79d2021-03-19 17:22:44 +00001922static int io_req_task_work_add(struct io_kiocb *req)
Jens Axboe7cbf1722021-02-10 00:03:20 +00001923{
Pavel Begunkovc15b79d2021-03-19 17:22:44 +00001924 struct task_struct *tsk = req->task;
Jens Axboe7cbf1722021-02-10 00:03:20 +00001925 struct io_uring_task *tctx = tsk->io_uring;
Pavel Begunkovc15b79d2021-03-19 17:22:44 +00001926 enum task_work_notify_mode notify;
Jens Axboe7cbf1722021-02-10 00:03:20 +00001927 struct io_wq_work_node *node, *prev;
Jens Axboe0b81e802021-02-16 10:33:53 -07001928 unsigned long flags;
Pavel Begunkovc15b79d2021-03-19 17:22:44 +00001929 int ret = 0;
1930
1931 if (unlikely(tsk->flags & PF_EXITING))
1932 return -ESRCH;
Jens Axboe7cbf1722021-02-10 00:03:20 +00001933
1934 WARN_ON_ONCE(!tctx);
1935
Jens Axboe0b81e802021-02-16 10:33:53 -07001936 spin_lock_irqsave(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001937 wq_list_add_tail(&req->io_task_work.node, &tctx->task_list);
Jens Axboe0b81e802021-02-16 10:33:53 -07001938 spin_unlock_irqrestore(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001939
1940 /* task_work already pending, we're done */
1941 if (test_bit(0, &tctx->task_state) ||
1942 test_and_set_bit(0, &tctx->task_state))
1943 return 0;
1944
Pavel Begunkovc15b79d2021-03-19 17:22:44 +00001945 /*
1946 * SQPOLL kernel thread doesn't need notification, just a wakeup. For
1947 * all other cases, use TWA_SIGNAL unconditionally to ensure we're
1948 * processing task_work. There's no reliable way to tell if TWA_RESUME
1949 * will do the job.
1950 */
1951 notify = (req->ctx->flags & IORING_SETUP_SQPOLL) ? TWA_NONE : TWA_SIGNAL;
1952
1953 if (!task_work_add(tsk, &tctx->task_work, notify)) {
1954 wake_up_process(tsk);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001955 return 0;
Pavel Begunkovc15b79d2021-03-19 17:22:44 +00001956 }
Jens Axboe7cbf1722021-02-10 00:03:20 +00001957
1958 /*
1959 * Slow path - we failed, find and delete work. if the work is not
1960 * in the list, it got run and we're fine.
1961 */
Jens Axboe0b81e802021-02-16 10:33:53 -07001962 spin_lock_irqsave(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001963 wq_list_for_each(node, prev, &tctx->task_list) {
1964 if (&req->io_task_work.node == node) {
1965 wq_list_del(&tctx->task_list, node, prev);
1966 ret = 1;
1967 break;
1968 }
1969 }
Jens Axboe0b81e802021-02-16 10:33:53 -07001970 spin_unlock_irqrestore(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001971 clear_bit(0, &tctx->task_state);
1972 return ret;
1973}
1974
Pavel Begunkov9b465712021-03-15 14:23:07 +00001975static bool io_run_task_work_head(struct callback_head **work_head)
1976{
1977 struct callback_head *work, *next;
1978 bool executed = false;
1979
1980 do {
1981 work = xchg(work_head, NULL);
1982 if (!work)
1983 break;
1984
1985 do {
1986 next = work->next;
1987 work->func(work);
1988 work = next;
1989 cond_resched();
1990 } while (work);
1991 executed = true;
1992 } while (1);
1993
1994 return executed;
1995}
1996
1997static void io_task_work_add_head(struct callback_head **work_head,
1998 struct callback_head *task_work)
1999{
2000 struct callback_head *head;
2001
2002 do {
2003 head = READ_ONCE(*work_head);
2004 task_work->next = head;
2005 } while (cmpxchg(work_head, head, task_work) != head);
2006}
2007
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002008static void io_req_task_work_add_fallback(struct io_kiocb *req,
Jens Axboe7cbf1722021-02-10 00:03:20 +00002009 task_work_func_t cb)
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002010{
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002011 init_task_work(&req->task_work, cb);
Pavel Begunkov9b465712021-03-15 14:23:07 +00002012 io_task_work_add_head(&req->ctx->exit_task_work, &req->task_work);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002013}
2014
Jens Axboec40f6372020-06-25 15:39:59 -06002015static void io_req_task_cancel(struct callback_head *cb)
2016{
2017 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
Jens Axboe87ceb6a2020-09-14 08:20:12 -06002018 struct io_ring_ctx *ctx = req->ctx;
Jens Axboec40f6372020-06-25 15:39:59 -06002019
Pavel Begunkove83acd72021-02-28 22:35:09 +00002020 /* ctx is guaranteed to stay alive while we hold uring_lock */
Pavel Begunkov792bb6e2021-02-18 22:32:51 +00002021 mutex_lock(&ctx->uring_lock);
Pavel Begunkov25935532021-03-19 17:22:40 +00002022 io_req_complete_failed(req, req->result);
Pavel Begunkov792bb6e2021-02-18 22:32:51 +00002023 mutex_unlock(&ctx->uring_lock);
Jens Axboec40f6372020-06-25 15:39:59 -06002024}
2025
2026static void __io_req_task_submit(struct io_kiocb *req)
2027{
2028 struct io_ring_ctx *ctx = req->ctx;
2029
Pavel Begunkov04fc6c82021-02-12 03:23:54 +00002030 /* ctx stays valid until unlock, even if we drop all ours ctx->refs */
Pavel Begunkov81b6d052021-01-04 20:36:35 +00002031 mutex_lock(&ctx->uring_lock);
Pavel Begunkov70aacfe2021-03-01 13:02:15 +00002032 if (!(current->flags & PF_EXITING) && !current->in_execve)
Pavel Begunkovc5eef2b942021-02-10 00:03:22 +00002033 __io_queue_sqe(req);
Pavel Begunkov81b6d052021-01-04 20:36:35 +00002034 else
Pavel Begunkov25935532021-03-19 17:22:40 +00002035 io_req_complete_failed(req, -EFAULT);
Pavel Begunkov81b6d052021-01-04 20:36:35 +00002036 mutex_unlock(&ctx->uring_lock);
Jens Axboe9e645e112019-05-10 16:07:28 -06002037}
2038
Jens Axboec40f6372020-06-25 15:39:59 -06002039static void io_req_task_submit(struct callback_head *cb)
2040{
2041 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
2042
2043 __io_req_task_submit(req);
2044}
2045
Pavel Begunkova3df76982021-02-18 22:32:52 +00002046static void io_req_task_queue_fail(struct io_kiocb *req, int ret)
2047{
Pavel Begunkova3df76982021-02-18 22:32:52 +00002048 req->result = ret;
2049 req->task_work.func = io_req_task_cancel;
2050
2051 if (unlikely(io_req_task_work_add(req)))
2052 io_req_task_work_add_fallback(req, io_req_task_cancel);
2053}
2054
Pavel Begunkov2c4b8eb2021-02-28 22:35:10 +00002055static void io_req_task_queue(struct io_kiocb *req)
2056{
2057 req->task_work.func = io_req_task_submit;
2058
2059 if (unlikely(io_req_task_work_add(req)))
2060 io_req_task_queue_fail(req, -ECANCELED);
2061}
2062
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002063static inline void io_queue_next(struct io_kiocb *req)
Jackie Liuc69f8db2019-11-09 11:00:08 +08002064{
Pavel Begunkov9b5f7bd92020-06-29 13:13:00 +03002065 struct io_kiocb *nxt = io_req_find_next(req);
Pavel Begunkov944e58b2019-11-21 23:21:01 +03002066
Pavel Begunkov906a8c32020-06-27 14:04:55 +03002067 if (nxt)
2068 io_req_task_queue(nxt);
Jackie Liuc69f8db2019-11-09 11:00:08 +08002069}
2070
Jens Axboe9e645e112019-05-10 16:07:28 -06002071static void io_free_req(struct io_kiocb *req)
2072{
Pavel Begunkovc3524382020-06-28 12:52:32 +03002073 io_queue_next(req);
Jens Axboe9e645e112019-05-10 16:07:28 -06002074 __io_free_req(req);
Jens Axboee65ef562019-03-12 10:16:44 -06002075}
2076
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002077struct req_batch {
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002078 struct task_struct *task;
2079 int task_refs;
Jens Axboe1b4c3512021-02-10 00:03:19 +00002080 int ctx_refs;
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002081};
2082
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002083static inline void io_init_req_batch(struct req_batch *rb)
Pavel Begunkov7a743e22020-03-03 21:33:13 +03002084{
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002085 rb->task_refs = 0;
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002086 rb->ctx_refs = 0;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002087 rb->task = NULL;
2088}
Pavel Begunkov8766dd52020-03-14 00:31:04 +03002089
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002090static void io_req_free_batch_finish(struct io_ring_ctx *ctx,
2091 struct req_batch *rb)
2092{
Pavel Begunkov6e833d52021-02-11 18:28:20 +00002093 if (rb->task)
Pavel Begunkov7c660732021-01-25 11:42:21 +00002094 io_put_task(rb->task, rb->task_refs);
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002095 if (rb->ctx_refs)
2096 percpu_ref_put_many(&ctx->refs, rb->ctx_refs);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002097}
2098
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002099static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req,
2100 struct io_submit_state *state)
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002101{
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002102 io_queue_next(req);
Pavel Begunkov96670652021-03-19 17:22:32 +00002103 io_dismantle_req(req);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002104
Jens Axboee3bc8e92020-09-24 08:45:57 -06002105 if (req->task != rb->task) {
Pavel Begunkov7c660732021-01-25 11:42:21 +00002106 if (rb->task)
2107 io_put_task(rb->task, rb->task_refs);
Jens Axboee3bc8e92020-09-24 08:45:57 -06002108 rb->task = req->task;
2109 rb->task_refs = 0;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002110 }
Jens Axboee3bc8e92020-09-24 08:45:57 -06002111 rb->task_refs++;
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002112 rb->ctx_refs++;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002113
Pavel Begunkovbd759042021-02-12 03:23:50 +00002114 if (state->free_reqs != ARRAY_SIZE(state->reqs))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002115 state->reqs[state->free_reqs++] = req;
Pavel Begunkovbd759042021-02-12 03:23:50 +00002116 else
2117 list_add(&req->compl.list, &state->comp.free_list);
Pavel Begunkov7a743e22020-03-03 21:33:13 +03002118}
2119
Pavel Begunkov905c1722021-02-10 00:03:14 +00002120static void io_submit_flush_completions(struct io_comp_state *cs,
2121 struct io_ring_ctx *ctx)
2122{
2123 int i, nr = cs->nr;
2124 struct io_kiocb *req;
2125 struct req_batch rb;
2126
2127 io_init_req_batch(&rb);
2128 spin_lock_irq(&ctx->completion_lock);
2129 for (i = 0; i < nr; i++) {
2130 req = cs->reqs[i];
Pavel Begunkov8d133262021-04-11 01:46:33 +01002131 __io_cqring_fill_event(req, req->result, req->compl.cflags);
Pavel Begunkov905c1722021-02-10 00:03:14 +00002132 }
2133 io_commit_cqring(ctx);
2134 spin_unlock_irq(&ctx->completion_lock);
2135
2136 io_cqring_ev_posted(ctx);
2137 for (i = 0; i < nr; i++) {
2138 req = cs->reqs[i];
2139
2140 /* submission and completion refs */
Jens Axboede9b4cc2021-02-24 13:28:27 -07002141 if (req_ref_sub_and_test(req, 2))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002142 io_req_free_batch(&rb, req, &ctx->submit_state);
Pavel Begunkov905c1722021-02-10 00:03:14 +00002143 }
2144
2145 io_req_free_batch_finish(ctx, &rb);
2146 cs->nr = 0;
Jens Axboee65ef562019-03-12 10:16:44 -06002147}
2148
Jens Axboeba816ad2019-09-28 11:36:45 -06002149/*
2150 * Drop reference to request, return next in chain (if there is one) if this
2151 * was the last reference to this request.
2152 */
Pavel Begunkov0d850352021-03-19 17:22:37 +00002153static inline struct io_kiocb *io_put_req_find_next(struct io_kiocb *req)
Jens Axboee65ef562019-03-12 10:16:44 -06002154{
Pavel Begunkov9b5f7bd92020-06-29 13:13:00 +03002155 struct io_kiocb *nxt = NULL;
2156
Jens Axboede9b4cc2021-02-24 13:28:27 -07002157 if (req_ref_put_and_test(req)) {
Pavel Begunkov9b5f7bd92020-06-29 13:13:00 +03002158 nxt = io_req_find_next(req);
Jens Axboe4d7dd462019-11-20 13:03:52 -07002159 __io_free_req(req);
Jens Axboe2a44f462020-02-25 13:25:41 -07002160 }
Pavel Begunkov9b5f7bd92020-06-29 13:13:00 +03002161 return nxt;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002162}
2163
Pavel Begunkov0d850352021-03-19 17:22:37 +00002164static inline void io_put_req(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002165{
Jens Axboede9b4cc2021-02-24 13:28:27 -07002166 if (req_ref_put_and_test(req))
Jens Axboedef596e2019-01-09 08:59:42 -07002167 io_free_req(req);
2168}
2169
Pavel Begunkov216578e2020-10-13 09:44:00 +01002170static void io_put_req_deferred_cb(struct callback_head *cb)
2171{
2172 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
2173
2174 io_free_req(req);
2175}
2176
2177static void io_free_req_deferred(struct io_kiocb *req)
2178{
Jens Axboe7cbf1722021-02-10 00:03:20 +00002179 req->task_work.func = io_put_req_deferred_cb;
Pavel Begunkova05432f2021-03-19 17:22:38 +00002180 if (unlikely(io_req_task_work_add(req)))
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002181 io_req_task_work_add_fallback(req, io_put_req_deferred_cb);
Pavel Begunkov216578e2020-10-13 09:44:00 +01002182}
2183
2184static inline void io_put_req_deferred(struct io_kiocb *req, int refs)
2185{
Jens Axboede9b4cc2021-02-24 13:28:27 -07002186 if (req_ref_sub_and_test(req, refs))
Pavel Begunkov216578e2020-10-13 09:44:00 +01002187 io_free_req_deferred(req);
2188}
2189
Pavel Begunkov6c503152021-01-04 20:36:36 +00002190static unsigned io_cqring_events(struct io_ring_ctx *ctx)
Jens Axboea3a0e432019-08-20 11:03:11 -06002191{
2192 /* See comment at the top of this file */
2193 smp_rmb();
Pavel Begunkove23de152020-12-17 00:24:37 +00002194 return __io_cqring_events(ctx);
Jens Axboea3a0e432019-08-20 11:03:11 -06002195}
2196
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03002197static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx)
2198{
2199 struct io_rings *rings = ctx->rings;
2200
2201 /* make sure SQ entry isn't read before tail */
2202 return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head;
2203}
2204
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002205static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf)
Jens Axboee94f1412019-12-19 12:06:02 -07002206{
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002207 unsigned int cflags;
Jens Axboee94f1412019-12-19 12:06:02 -07002208
Jens Axboebcda7ba2020-02-23 16:42:51 -07002209 cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT;
2210 cflags |= IORING_CQE_F_BUFFER;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03002211 req->flags &= ~REQ_F_BUFFER_SELECTED;
Jens Axboebcda7ba2020-02-23 16:42:51 -07002212 kfree(kbuf);
2213 return cflags;
2214}
2215
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002216static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req)
2217{
2218 struct io_buffer *kbuf;
2219
2220 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
2221 return io_put_kbuf(req, kbuf);
2222}
2223
Jens Axboe4c6e2772020-07-01 11:29:10 -06002224static inline bool io_run_task_work(void)
2225{
Jens Axboe6200b0a2020-09-13 14:38:30 -06002226 /*
2227 * Not safe to run on exiting task, and the task_work handling will
2228 * not add work to such a task.
2229 */
2230 if (unlikely(current->flags & PF_EXITING))
2231 return false;
Jens Axboe4c6e2772020-07-01 11:29:10 -06002232 if (current->task_works) {
2233 __set_current_state(TASK_RUNNING);
2234 task_work_run();
2235 return true;
2236 }
2237
2238 return false;
2239}
2240
Jens Axboedef596e2019-01-09 08:59:42 -07002241/*
2242 * Find and free completed poll iocbs
2243 */
2244static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events,
2245 struct list_head *done)
2246{
Jens Axboe8237e042019-12-28 10:48:22 -07002247 struct req_batch rb;
Jens Axboedef596e2019-01-09 08:59:42 -07002248 struct io_kiocb *req;
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002249
2250 /* order with ->result store in io_complete_rw_iopoll() */
2251 smp_rmb();
Jens Axboedef596e2019-01-09 08:59:42 -07002252
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002253 io_init_req_batch(&rb);
Jens Axboedef596e2019-01-09 08:59:42 -07002254 while (!list_empty(done)) {
Jens Axboebcda7ba2020-02-23 16:42:51 -07002255 int cflags = 0;
2256
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002257 req = list_first_entry(done, struct io_kiocb, inflight_entry);
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002258 list_del(&req->inflight_entry);
Jens Axboedef596e2019-01-09 08:59:42 -07002259
Pavel Begunkov8c130822021-03-22 01:58:32 +00002260 if (READ_ONCE(req->result) == -EAGAIN &&
2261 !(req->flags & REQ_F_DONT_REISSUE)) {
Pavel Begunkovf1613402021-02-11 18:28:21 +00002262 req->iopoll_completed = 0;
Pavel Begunkov8c130822021-03-22 01:58:32 +00002263 req_ref_get(req);
2264 io_queue_async_work(req);
2265 continue;
Pavel Begunkovf1613402021-02-11 18:28:21 +00002266 }
2267
Jens Axboebcda7ba2020-02-23 16:42:51 -07002268 if (req->flags & REQ_F_BUFFER_SELECTED)
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002269 cflags = io_put_rw_kbuf(req);
Jens Axboebcda7ba2020-02-23 16:42:51 -07002270
Pavel Begunkov8d133262021-04-11 01:46:33 +01002271 __io_cqring_fill_event(req, req->result, cflags);
Jens Axboedef596e2019-01-09 08:59:42 -07002272 (*nr_events)++;
2273
Jens Axboede9b4cc2021-02-24 13:28:27 -07002274 if (req_ref_put_and_test(req))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002275 io_req_free_batch(&rb, req, &ctx->submit_state);
Jens Axboedef596e2019-01-09 08:59:42 -07002276 }
Jens Axboedef596e2019-01-09 08:59:42 -07002277
Jens Axboe09bb8392019-03-13 12:39:28 -06002278 io_commit_cqring(ctx);
Pavel Begunkov80c18e42021-01-07 03:15:41 +00002279 io_cqring_ev_posted_iopoll(ctx);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002280 io_req_free_batch_finish(ctx, &rb);
Bijan Mottahedeh581f9812020-04-03 13:51:33 -07002281}
2282
Jens Axboedef596e2019-01-09 08:59:42 -07002283static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events,
2284 long min)
2285{
2286 struct io_kiocb *req, *tmp;
2287 LIST_HEAD(done);
2288 bool spin;
2289 int ret;
2290
2291 /*
2292 * Only spin for completions if we don't have multiple devices hanging
2293 * off our complete list, and we're under the requested amount.
2294 */
2295 spin = !ctx->poll_multi_file && *nr_events < min;
2296
2297 ret = 0;
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002298 list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) {
Jens Axboe9adbd452019-12-20 08:45:55 -07002299 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboedef596e2019-01-09 08:59:42 -07002300
2301 /*
Bijan Mottahedeh581f9812020-04-03 13:51:33 -07002302 * Move completed and retryable entries to our local lists.
2303 * If we find a request that requires polling, break out
2304 * and complete those lists first, if we have entries there.
Jens Axboedef596e2019-01-09 08:59:42 -07002305 */
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002306 if (READ_ONCE(req->iopoll_completed)) {
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002307 list_move_tail(&req->inflight_entry, &done);
Jens Axboedef596e2019-01-09 08:59:42 -07002308 continue;
2309 }
2310 if (!list_empty(&done))
2311 break;
2312
2313 ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin);
2314 if (ret < 0)
2315 break;
2316
Pavel Begunkov3aadc232020-07-06 17:59:29 +03002317 /* iopoll may have completed current req */
2318 if (READ_ONCE(req->iopoll_completed))
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002319 list_move_tail(&req->inflight_entry, &done);
Pavel Begunkov3aadc232020-07-06 17:59:29 +03002320
Jens Axboedef596e2019-01-09 08:59:42 -07002321 if (ret && spin)
2322 spin = false;
2323 ret = 0;
2324 }
2325
2326 if (!list_empty(&done))
2327 io_iopoll_complete(ctx, nr_events, &done);
2328
2329 return ret;
2330}
2331
2332/*
Brian Gianforcarod195a662019-12-13 03:09:50 -08002333 * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a
Jens Axboedef596e2019-01-09 08:59:42 -07002334 * non-spinning poll check - we'll still enter the driver poll loop, but only
2335 * as a non-spinning completion check.
2336 */
2337static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events,
2338 long min)
2339{
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002340 while (!list_empty(&ctx->iopoll_list) && !need_resched()) {
Jens Axboedef596e2019-01-09 08:59:42 -07002341 int ret;
2342
2343 ret = io_do_iopoll(ctx, nr_events, min);
2344 if (ret < 0)
2345 return ret;
Pavel Begunkoveba0a4d2020-07-06 17:59:30 +03002346 if (*nr_events >= min)
Jens Axboedef596e2019-01-09 08:59:42 -07002347 return 0;
2348 }
2349
2350 return 1;
2351}
2352
2353/*
2354 * We can't just wait for polled events to come to us, we have to actively
2355 * find and complete them.
2356 */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002357static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx)
Jens Axboedef596e2019-01-09 08:59:42 -07002358{
2359 if (!(ctx->flags & IORING_SETUP_IOPOLL))
2360 return;
2361
2362 mutex_lock(&ctx->uring_lock);
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002363 while (!list_empty(&ctx->iopoll_list)) {
Jens Axboedef596e2019-01-09 08:59:42 -07002364 unsigned int nr_events = 0;
2365
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002366 io_do_iopoll(ctx, &nr_events, 0);
Jens Axboe08f54392019-08-21 22:19:11 -06002367
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002368 /* let it sleep and repeat later if can't complete a request */
2369 if (nr_events == 0)
2370 break;
Jens Axboe08f54392019-08-21 22:19:11 -06002371 /*
2372 * Ensure we allow local-to-the-cpu processing to take place,
2373 * in this case we need to ensure that we reap all events.
Pavel Begunkov3fcee5a2020-07-06 17:59:31 +03002374 * Also let task_work, etc. to progress by releasing the mutex
Jens Axboe08f54392019-08-21 22:19:11 -06002375 */
Pavel Begunkov3fcee5a2020-07-06 17:59:31 +03002376 if (need_resched()) {
2377 mutex_unlock(&ctx->uring_lock);
2378 cond_resched();
2379 mutex_lock(&ctx->uring_lock);
2380 }
Jens Axboedef596e2019-01-09 08:59:42 -07002381 }
2382 mutex_unlock(&ctx->uring_lock);
2383}
2384
Pavel Begunkov7668b922020-07-07 16:36:21 +03002385static int io_iopoll_check(struct io_ring_ctx *ctx, long min)
Jens Axboedef596e2019-01-09 08:59:42 -07002386{
Pavel Begunkov7668b922020-07-07 16:36:21 +03002387 unsigned int nr_events = 0;
Jens Axboe2b2ed972019-10-25 10:06:15 -06002388 int iters = 0, ret = 0;
Jens Axboedef596e2019-01-09 08:59:42 -07002389
Xiaoguang Wangc7849be2020-02-22 14:46:05 +08002390 /*
2391 * We disallow the app entering submit/complete with polling, but we
2392 * still need to lock the ring to prevent racing with polled issue
2393 * that got punted to a workqueue.
2394 */
2395 mutex_lock(&ctx->uring_lock);
Jens Axboedef596e2019-01-09 08:59:42 -07002396 do {
Jens Axboe500f9fb2019-08-19 12:15:59 -06002397 /*
Jens Axboea3a0e432019-08-20 11:03:11 -06002398 * Don't enter poll loop if we already have events pending.
2399 * If we do, we can potentially be spinning for commands that
2400 * already triggered a CQE (eg in error).
2401 */
Pavel Begunkov6c503152021-01-04 20:36:36 +00002402 if (test_bit(0, &ctx->cq_check_overflow))
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00002403 __io_cqring_overflow_flush(ctx, false);
Pavel Begunkov6c503152021-01-04 20:36:36 +00002404 if (io_cqring_events(ctx))
Jens Axboea3a0e432019-08-20 11:03:11 -06002405 break;
2406
2407 /*
Jens Axboe500f9fb2019-08-19 12:15:59 -06002408 * If a submit got punted to a workqueue, we can have the
2409 * application entering polling for a command before it gets
2410 * issued. That app will hold the uring_lock for the duration
2411 * of the poll right here, so we need to take a breather every
2412 * now and then to ensure that the issue has a chance to add
2413 * the poll to the issued list. Otherwise we can spin here
2414 * forever, while the workqueue is stuck trying to acquire the
2415 * very same mutex.
2416 */
2417 if (!(++iters & 7)) {
2418 mutex_unlock(&ctx->uring_lock);
Jens Axboe4c6e2772020-07-01 11:29:10 -06002419 io_run_task_work();
Jens Axboe500f9fb2019-08-19 12:15:59 -06002420 mutex_lock(&ctx->uring_lock);
2421 }
2422
Pavel Begunkov7668b922020-07-07 16:36:21 +03002423 ret = io_iopoll_getevents(ctx, &nr_events, min);
Jens Axboedef596e2019-01-09 08:59:42 -07002424 if (ret <= 0)
2425 break;
2426 ret = 0;
Pavel Begunkov7668b922020-07-07 16:36:21 +03002427 } while (min && !nr_events && !need_resched());
Jens Axboedef596e2019-01-09 08:59:42 -07002428
Jens Axboe500f9fb2019-08-19 12:15:59 -06002429 mutex_unlock(&ctx->uring_lock);
Jens Axboedef596e2019-01-09 08:59:42 -07002430 return ret;
2431}
2432
Jens Axboe491381ce2019-10-17 09:20:46 -06002433static void kiocb_end_write(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002434{
Jens Axboe491381ce2019-10-17 09:20:46 -06002435 /*
2436 * Tell lockdep we inherited freeze protection from submission
2437 * thread.
2438 */
2439 if (req->flags & REQ_F_ISREG) {
Pavel Begunkov1c986792021-03-22 01:58:31 +00002440 struct super_block *sb = file_inode(req->file)->i_sb;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002441
Pavel Begunkov1c986792021-03-22 01:58:31 +00002442 __sb_writers_acquired(sb, SB_FREEZE_WRITE);
2443 sb_end_write(sb);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002444 }
2445}
2446
Jens Axboeb63534c2020-06-04 11:28:00 -06002447#ifdef CONFIG_BLOCK
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002448static bool io_resubmit_prep(struct io_kiocb *req)
Jens Axboeb63534c2020-06-04 11:28:00 -06002449{
Pavel Begunkovab454432021-03-22 01:58:33 +00002450 struct io_async_rw *rw = req->async_data;
2451
2452 if (!rw)
2453 return !io_req_prep_async(req);
2454 /* may have left rw->iter inconsistent on -EIOCBQUEUED */
2455 iov_iter_revert(&rw->iter, req->result - iov_iter_count(&rw->iter));
2456 return true;
Jens Axboeb63534c2020-06-04 11:28:00 -06002457}
Jens Axboeb63534c2020-06-04 11:28:00 -06002458
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002459static bool io_rw_should_reissue(struct io_kiocb *req)
Jens Axboeb63534c2020-06-04 11:28:00 -06002460{
Jens Axboe355afae2020-09-02 09:30:31 -06002461 umode_t mode = file_inode(req->file)->i_mode;
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002462 struct io_ring_ctx *ctx = req->ctx;
Jens Axboeb63534c2020-06-04 11:28:00 -06002463
Jens Axboe355afae2020-09-02 09:30:31 -06002464 if (!S_ISBLK(mode) && !S_ISREG(mode))
2465 return false;
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002466 if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() &&
2467 !(ctx->flags & IORING_SETUP_IOPOLL)))
Jens Axboeb63534c2020-06-04 11:28:00 -06002468 return false;
Jens Axboe7c977a52021-02-23 19:17:35 -07002469 /*
2470 * If ref is dying, we might be running poll reap from the exit work.
2471 * Don't attempt to reissue from that path, just let it fail with
2472 * -EAGAIN.
2473 */
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002474 if (percpu_ref_is_dying(&ctx->refs))
2475 return false;
2476 return true;
2477}
Jens Axboee82ad482021-04-02 19:45:34 -06002478#else
Jens Axboea1ff1e32021-04-12 06:40:02 -06002479static bool io_resubmit_prep(struct io_kiocb *req)
2480{
2481 return false;
2482}
Jens Axboee82ad482021-04-02 19:45:34 -06002483static bool io_rw_should_reissue(struct io_kiocb *req)
2484{
2485 return false;
2486}
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002487#endif
2488
Jens Axboea1d7c392020-06-22 11:09:46 -06002489static void __io_complete_rw(struct io_kiocb *req, long res, long res2,
Pavel Begunkov889fca72021-02-10 00:03:09 +00002490 unsigned int issue_flags)
Jens Axboea1d7c392020-06-22 11:09:46 -06002491{
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002492 int cflags = 0;
2493
Pavel Begunkovb65c1282021-03-22 01:45:59 +00002494 if (req->rw.kiocb.ki_flags & IOCB_WRITE)
2495 kiocb_end_write(req);
Pavel Begunkov9532b992021-03-22 01:58:34 +00002496 if (res != req->result) {
2497 if ((res == -EAGAIN || res == -EOPNOTSUPP) &&
2498 io_rw_should_reissue(req)) {
2499 req->flags |= REQ_F_REISSUE;
2500 return;
2501 }
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002502 req_set_fail_links(req);
Pavel Begunkov9532b992021-03-22 01:58:34 +00002503 }
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002504 if (req->flags & REQ_F_BUFFER_SELECTED)
2505 cflags = io_put_rw_kbuf(req);
2506 __io_req_complete(req, issue_flags, res, cflags);
Jens Axboeba816ad2019-09-28 11:36:45 -06002507}
2508
2509static void io_complete_rw(struct kiocb *kiocb, long res, long res2)
2510{
Jens Axboe9adbd452019-12-20 08:45:55 -07002511 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboeba816ad2019-09-28 11:36:45 -06002512
Pavel Begunkov889fca72021-02-10 00:03:09 +00002513 __io_complete_rw(req, res, res2, 0);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002514}
2515
Jens Axboedef596e2019-01-09 08:59:42 -07002516static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2)
2517{
Jens Axboe9adbd452019-12-20 08:45:55 -07002518 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboedef596e2019-01-09 08:59:42 -07002519
Jens Axboe491381ce2019-10-17 09:20:46 -06002520 if (kiocb->ki_flags & IOCB_WRITE)
2521 kiocb_end_write(req);
Pavel Begunkov9532b992021-03-22 01:58:34 +00002522 if (unlikely(res != req->result)) {
Jens Axboea1ff1e32021-04-12 06:40:02 -06002523 if (!(res == -EAGAIN && io_rw_should_reissue(req) &&
2524 io_resubmit_prep(req))) {
Pavel Begunkov9532b992021-03-22 01:58:34 +00002525 req_set_fail_links(req);
2526 req->flags |= REQ_F_DONT_REISSUE;
2527 }
Pavel Begunkov8c130822021-03-22 01:58:32 +00002528 }
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002529
2530 WRITE_ONCE(req->result, res);
Jens Axboeb9b0e0d2021-02-23 08:18:36 -07002531 /* order with io_iopoll_complete() checking ->result */
Pavel Begunkovcd664b02020-06-25 12:37:10 +03002532 smp_wmb();
2533 WRITE_ONCE(req->iopoll_completed, 1);
Jens Axboedef596e2019-01-09 08:59:42 -07002534}
2535
2536/*
2537 * After the iocb has been issued, it's safe to be found on the poll list.
2538 * Adding the kiocb to the list AFTER submission ensures that we don't
2539 * find it from a io_iopoll_getevents() thread before the issuer is done
2540 * accessing the kiocb cookie.
2541 */
Xiaoguang Wang2e9dbe92020-11-13 00:44:08 +08002542static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async)
Jens Axboedef596e2019-01-09 08:59:42 -07002543{
2544 struct io_ring_ctx *ctx = req->ctx;
2545
2546 /*
2547 * Track whether we have multiple files in our lists. This will impact
2548 * how we do polling eventually, not spinning if we're on potentially
2549 * different devices.
2550 */
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002551 if (list_empty(&ctx->iopoll_list)) {
Jens Axboedef596e2019-01-09 08:59:42 -07002552 ctx->poll_multi_file = false;
2553 } else if (!ctx->poll_multi_file) {
2554 struct io_kiocb *list_req;
2555
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002556 list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb,
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002557 inflight_entry);
Jens Axboe9adbd452019-12-20 08:45:55 -07002558 if (list_req->file != req->file)
Jens Axboedef596e2019-01-09 08:59:42 -07002559 ctx->poll_multi_file = true;
2560 }
2561
2562 /*
2563 * For fast devices, IO may have already completed. If it has, add
2564 * it to the front so we find it first.
2565 */
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002566 if (READ_ONCE(req->iopoll_completed))
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002567 list_add(&req->inflight_entry, &ctx->iopoll_list);
Jens Axboedef596e2019-01-09 08:59:42 -07002568 else
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002569 list_add_tail(&req->inflight_entry, &ctx->iopoll_list);
Xiaoguang Wangbdcd3ea2020-02-25 22:12:08 +08002570
Xiaoguang Wang2e9dbe92020-11-13 00:44:08 +08002571 /*
2572 * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread
2573 * task context or in io worker task context. If current task context is
2574 * sq thread, we don't need to check whether should wake up sq thread.
2575 */
2576 if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) &&
Jens Axboe534ca6d2020-09-02 13:52:19 -06002577 wq_has_sleeper(&ctx->sq_data->wait))
2578 wake_up(&ctx->sq_data->wait);
Jens Axboedef596e2019-01-09 08:59:42 -07002579}
2580
Pavel Begunkov9f13c352020-05-17 14:13:41 +03002581static inline void io_state_file_put(struct io_submit_state *state)
2582{
Pavel Begunkov02b23a92021-01-19 13:32:41 +00002583 if (state->file_refs) {
2584 fput_many(state->file, state->file_refs);
2585 state->file_refs = 0;
2586 }
Jens Axboe9a56a232019-01-09 09:06:50 -07002587}
2588
2589/*
2590 * Get as many references to a file as we have IOs left in this submission,
2591 * assuming most submissions are for one file, or at least that each file
2592 * has more than one submission.
2593 */
Pavel Begunkov8da11c12020-02-24 11:32:44 +03002594static struct file *__io_file_get(struct io_submit_state *state, int fd)
Jens Axboe9a56a232019-01-09 09:06:50 -07002595{
2596 if (!state)
2597 return fget(fd);
2598
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002599 if (state->file_refs) {
Jens Axboe9a56a232019-01-09 09:06:50 -07002600 if (state->fd == fd) {
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002601 state->file_refs--;
Jens Axboe9a56a232019-01-09 09:06:50 -07002602 return state->file;
2603 }
Pavel Begunkov02b23a92021-01-19 13:32:41 +00002604 io_state_file_put(state);
Jens Axboe9a56a232019-01-09 09:06:50 -07002605 }
2606 state->file = fget_many(fd, state->ios_left);
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002607 if (unlikely(!state->file))
Jens Axboe9a56a232019-01-09 09:06:50 -07002608 return NULL;
2609
2610 state->fd = fd;
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002611 state->file_refs = state->ios_left - 1;
Jens Axboe9a56a232019-01-09 09:06:50 -07002612 return state->file;
2613}
2614
Jens Axboe4503b762020-06-01 10:00:27 -06002615static bool io_bdev_nowait(struct block_device *bdev)
2616{
Jeffle Xu9ba0d0c2020-10-19 16:59:42 +08002617 return !bdev || blk_queue_nowait(bdev_get_queue(bdev));
Jens Axboe4503b762020-06-01 10:00:27 -06002618}
2619
Jens Axboe2b188cc2019-01-07 10:46:33 -07002620/*
2621 * If we tracked the file through the SCM inflight mechanism, we could support
2622 * any file. For now, just ensure that anything potentially problematic is done
2623 * inline.
2624 */
Jens Axboe7b29f922021-03-12 08:30:14 -07002625static bool __io_file_supports_async(struct file *file, int rw)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002626{
2627 umode_t mode = file_inode(file)->i_mode;
2628
Jens Axboe4503b762020-06-01 10:00:27 -06002629 if (S_ISBLK(mode)) {
Christoph Hellwig4e7b5672020-11-23 13:38:40 +01002630 if (IS_ENABLED(CONFIG_BLOCK) &&
2631 io_bdev_nowait(I_BDEV(file->f_mapping->host)))
Jens Axboe4503b762020-06-01 10:00:27 -06002632 return true;
2633 return false;
2634 }
2635 if (S_ISCHR(mode) || S_ISSOCK(mode))
Jens Axboe2b188cc2019-01-07 10:46:33 -07002636 return true;
Jens Axboe4503b762020-06-01 10:00:27 -06002637 if (S_ISREG(mode)) {
Christoph Hellwig4e7b5672020-11-23 13:38:40 +01002638 if (IS_ENABLED(CONFIG_BLOCK) &&
2639 io_bdev_nowait(file->f_inode->i_sb->s_bdev) &&
Jens Axboe4503b762020-06-01 10:00:27 -06002640 file->f_op != &io_uring_fops)
2641 return true;
2642 return false;
2643 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002644
Jens Axboec5b85622020-06-09 19:23:05 -06002645 /* any ->read/write should understand O_NONBLOCK */
2646 if (file->f_flags & O_NONBLOCK)
2647 return true;
2648
Jens Axboeaf197f52020-04-28 13:15:06 -06002649 if (!(file->f_mode & FMODE_NOWAIT))
2650 return false;
2651
2652 if (rw == READ)
2653 return file->f_op->read_iter != NULL;
2654
2655 return file->f_op->write_iter != NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002656}
2657
Jens Axboe7b29f922021-03-12 08:30:14 -07002658static bool io_file_supports_async(struct io_kiocb *req, int rw)
2659{
2660 if (rw == READ && (req->flags & REQ_F_ASYNC_READ))
2661 return true;
2662 else if (rw == WRITE && (req->flags & REQ_F_ASYNC_WRITE))
2663 return true;
2664
2665 return __io_file_supports_async(req->file, rw);
2666}
2667
Pavel Begunkova88fc402020-09-30 22:57:53 +03002668static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002669{
Jens Axboedef596e2019-01-09 08:59:42 -07002670 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe9adbd452019-12-20 08:45:55 -07002671 struct kiocb *kiocb = &req->rw.kiocb;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002672 struct file *file = req->file;
Jens Axboe09bb8392019-03-13 12:39:28 -06002673 unsigned ioprio;
2674 int ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002675
Jens Axboe7b29f922021-03-12 08:30:14 -07002676 if (!(req->flags & REQ_F_ISREG) && S_ISREG(file_inode(file)->i_mode))
Jens Axboe491381ce2019-10-17 09:20:46 -06002677 req->flags |= REQ_F_ISREG;
2678
Jens Axboe2b188cc2019-01-07 10:46:33 -07002679 kiocb->ki_pos = READ_ONCE(sqe->off);
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002680 if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) {
Jens Axboeba042912019-12-25 16:33:42 -07002681 req->flags |= REQ_F_CUR_POS;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002682 kiocb->ki_pos = file->f_pos;
Jens Axboeba042912019-12-25 16:33:42 -07002683 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002684 kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp));
Pavel Begunkov3e577dc2020-02-01 03:58:42 +03002685 kiocb->ki_flags = iocb_flags(kiocb->ki_filp);
2686 ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags));
2687 if (unlikely(ret))
2688 return ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002689
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002690 /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */
2691 if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK))
2692 req->flags |= REQ_F_NOWAIT;
2693
Jens Axboe2b188cc2019-01-07 10:46:33 -07002694 ioprio = READ_ONCE(sqe->ioprio);
2695 if (ioprio) {
2696 ret = ioprio_check_cap(ioprio);
2697 if (ret)
Jens Axboe09bb8392019-03-13 12:39:28 -06002698 return ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002699
2700 kiocb->ki_ioprio = ioprio;
2701 } else
2702 kiocb->ki_ioprio = get_current_ioprio();
2703
Jens Axboedef596e2019-01-09 08:59:42 -07002704 if (ctx->flags & IORING_SETUP_IOPOLL) {
Jens Axboedef596e2019-01-09 08:59:42 -07002705 if (!(kiocb->ki_flags & IOCB_DIRECT) ||
2706 !kiocb->ki_filp->f_op->iopoll)
Jens Axboe09bb8392019-03-13 12:39:28 -06002707 return -EOPNOTSUPP;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002708
Jens Axboedef596e2019-01-09 08:59:42 -07002709 kiocb->ki_flags |= IOCB_HIPRI;
2710 kiocb->ki_complete = io_complete_rw_iopoll;
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002711 req->iopoll_completed = 0;
Jens Axboedef596e2019-01-09 08:59:42 -07002712 } else {
Jens Axboe09bb8392019-03-13 12:39:28 -06002713 if (kiocb->ki_flags & IOCB_HIPRI)
2714 return -EINVAL;
Jens Axboedef596e2019-01-09 08:59:42 -07002715 kiocb->ki_complete = io_complete_rw;
2716 }
Jens Axboe9adbd452019-12-20 08:45:55 -07002717
Jens Axboe3529d8c2019-12-19 18:24:38 -07002718 req->rw.addr = READ_ONCE(sqe->addr);
2719 req->rw.len = READ_ONCE(sqe->len);
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002720 req->buf_index = READ_ONCE(sqe->buf_index);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002721 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002722}
2723
2724static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret)
2725{
2726 switch (ret) {
2727 case -EIOCBQUEUED:
2728 break;
2729 case -ERESTARTSYS:
2730 case -ERESTARTNOINTR:
2731 case -ERESTARTNOHAND:
2732 case -ERESTART_RESTARTBLOCK:
2733 /*
2734 * We can't just restart the syscall, since previously
2735 * submitted sqes may already be in progress. Just fail this
2736 * IO with EINTR.
2737 */
2738 ret = -EINTR;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05002739 fallthrough;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002740 default:
2741 kiocb->ki_complete(kiocb, ret, 0);
2742 }
2743}
2744
Jens Axboea1d7c392020-06-22 11:09:46 -06002745static void kiocb_done(struct kiocb *kiocb, ssize_t ret,
Pavel Begunkov889fca72021-02-10 00:03:09 +00002746 unsigned int issue_flags)
Jens Axboeba816ad2019-09-28 11:36:45 -06002747{
Jens Axboeba042912019-12-25 16:33:42 -07002748 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboee8c2bc12020-08-15 18:44:09 -07002749 struct io_async_rw *io = req->async_data;
Pavel Begunkov97284632021-04-08 19:28:03 +01002750 bool check_reissue = kiocb->ki_complete == io_complete_rw;
Jens Axboeba042912019-12-25 16:33:42 -07002751
Jens Axboe227c0c92020-08-13 11:51:40 -06002752 /* add previously done IO, if any */
Jens Axboee8c2bc12020-08-15 18:44:09 -07002753 if (io && io->bytes_done > 0) {
Jens Axboe227c0c92020-08-13 11:51:40 -06002754 if (ret < 0)
Jens Axboee8c2bc12020-08-15 18:44:09 -07002755 ret = io->bytes_done;
Jens Axboe227c0c92020-08-13 11:51:40 -06002756 else
Jens Axboee8c2bc12020-08-15 18:44:09 -07002757 ret += io->bytes_done;
Jens Axboe227c0c92020-08-13 11:51:40 -06002758 }
2759
Jens Axboeba042912019-12-25 16:33:42 -07002760 if (req->flags & REQ_F_CUR_POS)
2761 req->file->f_pos = kiocb->ki_pos;
Pavel Begunkovbcaec082020-02-24 11:30:18 +03002762 if (ret >= 0 && kiocb->ki_complete == io_complete_rw)
Pavel Begunkov889fca72021-02-10 00:03:09 +00002763 __io_complete_rw(req, ret, 0, issue_flags);
Jens Axboeba816ad2019-09-28 11:36:45 -06002764 else
2765 io_rw_done(kiocb, ret);
Pavel Begunkov97284632021-04-08 19:28:03 +01002766
2767 if (check_reissue && req->flags & REQ_F_REISSUE) {
2768 req->flags &= ~REQ_F_REISSUE;
Pavel Begunkov8c130822021-03-22 01:58:32 +00002769 if (!io_resubmit_prep(req)) {
2770 req_ref_get(req);
2771 io_queue_async_work(req);
2772 } else {
Pavel Begunkov97284632021-04-08 19:28:03 +01002773 int cflags = 0;
2774
2775 req_set_fail_links(req);
2776 if (req->flags & REQ_F_BUFFER_SELECTED)
2777 cflags = io_put_rw_kbuf(req);
2778 __io_req_complete(req, issue_flags, ret, cflags);
2779 }
2780 }
Jens Axboeba816ad2019-09-28 11:36:45 -06002781}
2782
Pavel Begunkov847595d2021-02-04 13:52:06 +00002783static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter)
Jens Axboeedafcce2019-01-09 09:16:05 -07002784{
Jens Axboe9adbd452019-12-20 08:45:55 -07002785 struct io_ring_ctx *ctx = req->ctx;
2786 size_t len = req->rw.len;
Jens Axboeedafcce2019-01-09 09:16:05 -07002787 struct io_mapped_ubuf *imu;
Pavel Begunkov4be1c612020-09-06 00:45:48 +03002788 u16 index, buf_index = req->buf_index;
Pavel Begunkov75769e32021-04-01 15:43:54 +01002789 u64 buf_end, buf_addr = req->rw.addr;
Jens Axboeedafcce2019-01-09 09:16:05 -07002790 size_t offset;
Jens Axboeedafcce2019-01-09 09:16:05 -07002791
Jens Axboeedafcce2019-01-09 09:16:05 -07002792 if (unlikely(buf_index >= ctx->nr_user_bufs))
2793 return -EFAULT;
Jens Axboeedafcce2019-01-09 09:16:05 -07002794 index = array_index_nospec(buf_index, ctx->nr_user_bufs);
2795 imu = &ctx->user_bufs[index];
Jens Axboe9adbd452019-12-20 08:45:55 -07002796 buf_addr = req->rw.addr;
Jens Axboeedafcce2019-01-09 09:16:05 -07002797
Pavel Begunkov75769e32021-04-01 15:43:54 +01002798 if (unlikely(check_add_overflow(buf_addr, (u64)len, &buf_end)))
Jens Axboeedafcce2019-01-09 09:16:05 -07002799 return -EFAULT;
2800 /* not inside the mapped region */
Pavel Begunkov4751f532021-04-01 15:43:55 +01002801 if (unlikely(buf_addr < imu->ubuf || buf_end > imu->ubuf_end))
Jens Axboeedafcce2019-01-09 09:16:05 -07002802 return -EFAULT;
2803
2804 /*
2805 * May not be a start of buffer, set size appropriately
2806 * and advance us to the beginning.
2807 */
2808 offset = buf_addr - imu->ubuf;
2809 iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len);
Jens Axboebd11b3a2019-07-20 08:37:31 -06002810
2811 if (offset) {
2812 /*
2813 * Don't use iov_iter_advance() here, as it's really slow for
2814 * using the latter parts of a big fixed buffer - it iterates
2815 * over each segment manually. We can cheat a bit here, because
2816 * we know that:
2817 *
2818 * 1) it's a BVEC iter, we set it up
2819 * 2) all bvecs are PAGE_SIZE in size, except potentially the
2820 * first and last bvec
2821 *
2822 * So just find our index, and adjust the iterator afterwards.
2823 * If the offset is within the first bvec (or the whole first
2824 * bvec, just use iov_iter_advance(). This makes it easier
2825 * since we can just skip the first segment, which may not
2826 * be PAGE_SIZE aligned.
2827 */
2828 const struct bio_vec *bvec = imu->bvec;
2829
2830 if (offset <= bvec->bv_len) {
2831 iov_iter_advance(iter, offset);
2832 } else {
2833 unsigned long seg_skip;
2834
2835 /* skip first vec */
2836 offset -= bvec->bv_len;
2837 seg_skip = 1 + (offset >> PAGE_SHIFT);
2838
2839 iter->bvec = bvec + seg_skip;
2840 iter->nr_segs -= seg_skip;
Aleix Roca Nonell99c79f62019-08-15 14:03:22 +02002841 iter->count -= bvec->bv_len + offset;
Jens Axboebd11b3a2019-07-20 08:37:31 -06002842 iter->iov_offset = offset & ~PAGE_MASK;
Jens Axboebd11b3a2019-07-20 08:37:31 -06002843 }
2844 }
2845
Pavel Begunkov847595d2021-02-04 13:52:06 +00002846 return 0;
Jens Axboeedafcce2019-01-09 09:16:05 -07002847}
2848
Jens Axboebcda7ba2020-02-23 16:42:51 -07002849static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock)
2850{
2851 if (needs_lock)
2852 mutex_unlock(&ctx->uring_lock);
2853}
2854
2855static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock)
2856{
2857 /*
2858 * "Normal" inline submissions always hold the uring_lock, since we
2859 * grab it from the system call. Same is true for the SQPOLL offload.
2860 * The only exception is when we've detached the request and issue it
2861 * from an async worker thread, grab the lock for that case.
2862 */
2863 if (needs_lock)
2864 mutex_lock(&ctx->uring_lock);
2865}
2866
2867static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len,
2868 int bgid, struct io_buffer *kbuf,
2869 bool needs_lock)
2870{
2871 struct io_buffer *head;
2872
2873 if (req->flags & REQ_F_BUFFER_SELECTED)
2874 return kbuf;
2875
2876 io_ring_submit_lock(req->ctx, needs_lock);
2877
2878 lockdep_assert_held(&req->ctx->uring_lock);
2879
Jens Axboe9e15c3a2021-03-13 12:29:43 -07002880 head = xa_load(&req->ctx->io_buffers, bgid);
Jens Axboebcda7ba2020-02-23 16:42:51 -07002881 if (head) {
2882 if (!list_empty(&head->list)) {
2883 kbuf = list_last_entry(&head->list, struct io_buffer,
2884 list);
2885 list_del(&kbuf->list);
2886 } else {
2887 kbuf = head;
Jens Axboe9e15c3a2021-03-13 12:29:43 -07002888 xa_erase(&req->ctx->io_buffers, bgid);
Jens Axboebcda7ba2020-02-23 16:42:51 -07002889 }
2890 if (*len > kbuf->len)
2891 *len = kbuf->len;
2892 } else {
2893 kbuf = ERR_PTR(-ENOBUFS);
2894 }
2895
2896 io_ring_submit_unlock(req->ctx, needs_lock);
2897
2898 return kbuf;
2899}
2900
Jens Axboe4d954c22020-02-27 07:31:19 -07002901static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len,
2902 bool needs_lock)
2903{
2904 struct io_buffer *kbuf;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002905 u16 bgid;
Jens Axboe4d954c22020-02-27 07:31:19 -07002906
2907 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002908 bgid = req->buf_index;
Jens Axboe4d954c22020-02-27 07:31:19 -07002909 kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock);
2910 if (IS_ERR(kbuf))
2911 return kbuf;
2912 req->rw.addr = (u64) (unsigned long) kbuf;
2913 req->flags |= REQ_F_BUFFER_SELECTED;
2914 return u64_to_user_ptr(kbuf->addr);
2915}
2916
2917#ifdef CONFIG_COMPAT
2918static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov,
2919 bool needs_lock)
2920{
2921 struct compat_iovec __user *uiov;
2922 compat_ssize_t clen;
2923 void __user *buf;
2924 ssize_t len;
2925
2926 uiov = u64_to_user_ptr(req->rw.addr);
2927 if (!access_ok(uiov, sizeof(*uiov)))
2928 return -EFAULT;
2929 if (__get_user(clen, &uiov->iov_len))
2930 return -EFAULT;
2931 if (clen < 0)
2932 return -EINVAL;
2933
2934 len = clen;
2935 buf = io_rw_buffer_select(req, &len, needs_lock);
2936 if (IS_ERR(buf))
2937 return PTR_ERR(buf);
2938 iov[0].iov_base = buf;
2939 iov[0].iov_len = (compat_size_t) len;
2940 return 0;
2941}
2942#endif
2943
2944static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov,
2945 bool needs_lock)
2946{
2947 struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr);
2948 void __user *buf;
2949 ssize_t len;
2950
2951 if (copy_from_user(iov, uiov, sizeof(*uiov)))
2952 return -EFAULT;
2953
2954 len = iov[0].iov_len;
2955 if (len < 0)
2956 return -EINVAL;
2957 buf = io_rw_buffer_select(req, &len, needs_lock);
2958 if (IS_ERR(buf))
2959 return PTR_ERR(buf);
2960 iov[0].iov_base = buf;
2961 iov[0].iov_len = len;
2962 return 0;
2963}
2964
2965static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov,
2966 bool needs_lock)
2967{
Jens Axboedddb3e22020-06-04 11:27:01 -06002968 if (req->flags & REQ_F_BUFFER_SELECTED) {
2969 struct io_buffer *kbuf;
2970
2971 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
2972 iov[0].iov_base = u64_to_user_ptr(kbuf->addr);
2973 iov[0].iov_len = kbuf->len;
Jens Axboe4d954c22020-02-27 07:31:19 -07002974 return 0;
Jens Axboedddb3e22020-06-04 11:27:01 -06002975 }
Pavel Begunkovdd201662020-12-19 03:15:43 +00002976 if (req->rw.len != 1)
Jens Axboe4d954c22020-02-27 07:31:19 -07002977 return -EINVAL;
2978
2979#ifdef CONFIG_COMPAT
2980 if (req->ctx->compat)
2981 return io_compat_import(req, iov, needs_lock);
2982#endif
2983
2984 return __io_iov_buffer_select(req, iov, needs_lock);
2985}
2986
Pavel Begunkov847595d2021-02-04 13:52:06 +00002987static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec,
2988 struct iov_iter *iter, bool needs_lock)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002989{
Jens Axboe9adbd452019-12-20 08:45:55 -07002990 void __user *buf = u64_to_user_ptr(req->rw.addr);
2991 size_t sqe_len = req->rw.len;
Pavel Begunkov847595d2021-02-04 13:52:06 +00002992 u8 opcode = req->opcode;
Jens Axboe4d954c22020-02-27 07:31:19 -07002993 ssize_t ret;
Jens Axboeedafcce2019-01-09 09:16:05 -07002994
Pavel Begunkov7d009162019-11-25 23:14:40 +03002995 if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) {
Jens Axboeedafcce2019-01-09 09:16:05 -07002996 *iovec = NULL;
Jens Axboe9adbd452019-12-20 08:45:55 -07002997 return io_import_fixed(req, rw, iter);
Jens Axboeedafcce2019-01-09 09:16:05 -07002998 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002999
Jens Axboebcda7ba2020-02-23 16:42:51 -07003000 /* buffer index only valid with fixed read/write, or buffer select */
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07003001 if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT))
Jens Axboe9adbd452019-12-20 08:45:55 -07003002 return -EINVAL;
3003
Jens Axboe3a6820f2019-12-22 15:19:35 -07003004 if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) {
Jens Axboebcda7ba2020-02-23 16:42:51 -07003005 if (req->flags & REQ_F_BUFFER_SELECT) {
Jens Axboe4d954c22020-02-27 07:31:19 -07003006 buf = io_rw_buffer_select(req, &sqe_len, needs_lock);
Pavel Begunkov867a23e2020-08-20 11:34:39 +03003007 if (IS_ERR(buf))
Jens Axboe4d954c22020-02-27 07:31:19 -07003008 return PTR_ERR(buf);
Jens Axboe3f9d6442020-03-11 12:27:04 -06003009 req->rw.len = sqe_len;
Jens Axboebcda7ba2020-02-23 16:42:51 -07003010 }
3011
Jens Axboe3a6820f2019-12-22 15:19:35 -07003012 ret = import_single_range(rw, buf, sqe_len, *iovec, iter);
3013 *iovec = NULL;
David Laight10fc72e2020-11-07 13:16:25 +00003014 return ret;
Jens Axboe3a6820f2019-12-22 15:19:35 -07003015 }
3016
Jens Axboe4d954c22020-02-27 07:31:19 -07003017 if (req->flags & REQ_F_BUFFER_SELECT) {
3018 ret = io_iov_buffer_select(req, *iovec, needs_lock);
Pavel Begunkov847595d2021-02-04 13:52:06 +00003019 if (!ret)
3020 iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len);
Jens Axboe4d954c22020-02-27 07:31:19 -07003021 *iovec = NULL;
3022 return ret;
3023 }
3024
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02003025 return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter,
3026 req->ctx->compat);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003027}
3028
Jens Axboe0fef9482020-08-26 10:36:20 -06003029static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb)
3030{
Pavel Begunkov5b09e372020-09-30 22:57:15 +03003031 return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos;
Jens Axboe0fef9482020-08-26 10:36:20 -06003032}
3033
Jens Axboe32960612019-09-23 11:05:34 -06003034/*
3035 * For files that don't have ->read_iter() and ->write_iter(), handle them
3036 * by looping over ->read() or ->write() manually.
3037 */
Jens Axboe4017eb92020-10-22 14:14:12 -06003038static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter)
Jens Axboe32960612019-09-23 11:05:34 -06003039{
Jens Axboe4017eb92020-10-22 14:14:12 -06003040 struct kiocb *kiocb = &req->rw.kiocb;
3041 struct file *file = req->file;
Jens Axboe32960612019-09-23 11:05:34 -06003042 ssize_t ret = 0;
3043
3044 /*
3045 * Don't support polled IO through this interface, and we can't
3046 * support non-blocking either. For the latter, this just causes
3047 * the kiocb to be handled from an async context.
3048 */
3049 if (kiocb->ki_flags & IOCB_HIPRI)
3050 return -EOPNOTSUPP;
3051 if (kiocb->ki_flags & IOCB_NOWAIT)
3052 return -EAGAIN;
3053
3054 while (iov_iter_count(iter)) {
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003055 struct iovec iovec;
Jens Axboe32960612019-09-23 11:05:34 -06003056 ssize_t nr;
3057
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003058 if (!iov_iter_is_bvec(iter)) {
3059 iovec = iov_iter_iovec(iter);
3060 } else {
Jens Axboe4017eb92020-10-22 14:14:12 -06003061 iovec.iov_base = u64_to_user_ptr(req->rw.addr);
3062 iovec.iov_len = req->rw.len;
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003063 }
3064
Jens Axboe32960612019-09-23 11:05:34 -06003065 if (rw == READ) {
3066 nr = file->f_op->read(file, iovec.iov_base,
Jens Axboe0fef9482020-08-26 10:36:20 -06003067 iovec.iov_len, io_kiocb_ppos(kiocb));
Jens Axboe32960612019-09-23 11:05:34 -06003068 } else {
3069 nr = file->f_op->write(file, iovec.iov_base,
Jens Axboe0fef9482020-08-26 10:36:20 -06003070 iovec.iov_len, io_kiocb_ppos(kiocb));
Jens Axboe32960612019-09-23 11:05:34 -06003071 }
3072
3073 if (nr < 0) {
3074 if (!ret)
3075 ret = nr;
3076 break;
3077 }
3078 ret += nr;
3079 if (nr != iovec.iov_len)
3080 break;
Jens Axboe4017eb92020-10-22 14:14:12 -06003081 req->rw.len -= nr;
3082 req->rw.addr += nr;
Jens Axboe32960612019-09-23 11:05:34 -06003083 iov_iter_advance(iter, nr);
3084 }
3085
3086 return ret;
3087}
3088
Jens Axboeff6165b2020-08-13 09:47:43 -06003089static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec,
3090 const struct iovec *fast_iov, struct iov_iter *iter)
Jens Axboef67676d2019-12-02 11:03:47 -07003091{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003092 struct io_async_rw *rw = req->async_data;
Pavel Begunkovb64e3442020-07-13 22:59:18 +03003093
Jens Axboeff6165b2020-08-13 09:47:43 -06003094 memcpy(&rw->iter, iter, sizeof(*iter));
Pavel Begunkovafb87652020-09-06 00:45:46 +03003095 rw->free_iovec = iovec;
Jens Axboe227c0c92020-08-13 11:51:40 -06003096 rw->bytes_done = 0;
Jens Axboeff6165b2020-08-13 09:47:43 -06003097 /* can only be fixed buffers, no need to do anything */
Pavel Begunkov9c3a2052020-11-23 23:20:27 +00003098 if (iov_iter_is_bvec(iter))
Jens Axboeff6165b2020-08-13 09:47:43 -06003099 return;
Pavel Begunkovb64e3442020-07-13 22:59:18 +03003100 if (!iovec) {
Jens Axboeff6165b2020-08-13 09:47:43 -06003101 unsigned iov_off = 0;
3102
3103 rw->iter.iov = rw->fast_iov;
3104 if (iter->iov != fast_iov) {
3105 iov_off = iter->iov - fast_iov;
3106 rw->iter.iov += iov_off;
3107 }
3108 if (rw->fast_iov != fast_iov)
3109 memcpy(rw->fast_iov + iov_off, fast_iov + iov_off,
Xiaoguang Wang45097da2020-04-08 22:29:58 +08003110 sizeof(struct iovec) * iter->nr_segs);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03003111 } else {
3112 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboef67676d2019-12-02 11:03:47 -07003113 }
3114}
3115
Pavel Begunkov6cb78682021-02-28 22:35:17 +00003116static inline int io_alloc_async_data(struct io_kiocb *req)
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003117{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003118 WARN_ON_ONCE(!io_op_defs[req->opcode].async_size);
3119 req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL);
3120 return req->async_data == NULL;
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003121}
3122
Jens Axboeff6165b2020-08-13 09:47:43 -06003123static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec,
3124 const struct iovec *fast_iov,
Jens Axboe227c0c92020-08-13 11:51:40 -06003125 struct iov_iter *iter, bool force)
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003126{
Pavel Begunkov26f05052021-02-28 22:35:18 +00003127 if (!force && !io_op_defs[req->opcode].needs_async_setup)
Jens Axboe74566df2020-01-13 19:23:24 -07003128 return 0;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003129 if (!req->async_data) {
Pavel Begunkov6cb78682021-02-28 22:35:17 +00003130 if (io_alloc_async_data(req)) {
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003131 kfree(iovec);
Jens Axboe5d204bc2020-01-31 12:06:52 -07003132 return -ENOMEM;
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003133 }
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003134
Jens Axboeff6165b2020-08-13 09:47:43 -06003135 io_req_map_rw(req, iovec, fast_iov, iter);
Jens Axboe5d204bc2020-01-31 12:06:52 -07003136 }
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003137 return 0;
Jens Axboef67676d2019-12-02 11:03:47 -07003138}
3139
Pavel Begunkov73debe62020-09-30 22:57:54 +03003140static inline int io_rw_prep_async(struct io_kiocb *req, int rw)
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003141{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003142 struct io_async_rw *iorw = req->async_data;
Pavel Begunkovf4bff102020-09-06 00:45:45 +03003143 struct iovec *iov = iorw->fast_iov;
Pavel Begunkov847595d2021-02-04 13:52:06 +00003144 int ret;
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003145
Pavel Begunkov2846c482020-11-07 13:16:27 +00003146 ret = io_import_iovec(rw, req, &iov, &iorw->iter, false);
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003147 if (unlikely(ret < 0))
3148 return ret;
3149
Pavel Begunkovab0b1962020-09-06 00:45:47 +03003150 iorw->bytes_done = 0;
3151 iorw->free_iovec = iov;
3152 if (iov)
3153 req->flags |= REQ_F_NEED_CLEANUP;
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003154 return 0;
3155}
3156
Pavel Begunkov73debe62020-09-30 22:57:54 +03003157static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07003158{
Jens Axboe3529d8c2019-12-19 18:24:38 -07003159 if (unlikely(!(req->file->f_mode & FMODE_READ)))
3160 return -EBADF;
Pavel Begunkov93642ef2021-02-18 18:29:44 +00003161 return io_prep_rw(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07003162}
3163
Jens Axboec1dd91d2020-08-03 16:43:59 -06003164/*
3165 * This is our waitqueue callback handler, registered through lock_page_async()
3166 * when we initially tried to do the IO with the iocb armed our waitqueue.
3167 * This gets called when the page is unlocked, and we generally expect that to
3168 * happen when the page IO is completed and the page is now uptodate. This will
3169 * queue a task_work based retry of the operation, attempting to copy the data
3170 * again. If the latter fails because the page was NOT uptodate, then we will
3171 * do a thread based blocking retry of the operation. That's the unexpected
3172 * slow path.
3173 */
Jens Axboebcf5a062020-05-22 09:24:42 -06003174static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode,
3175 int sync, void *arg)
3176{
3177 struct wait_page_queue *wpq;
3178 struct io_kiocb *req = wait->private;
Jens Axboebcf5a062020-05-22 09:24:42 -06003179 struct wait_page_key *key = arg;
Jens Axboebcf5a062020-05-22 09:24:42 -06003180
3181 wpq = container_of(wait, struct wait_page_queue, wait);
3182
Linus Torvaldscdc8fcb2020-08-03 13:01:22 -07003183 if (!wake_page_match(wpq, key))
3184 return 0;
3185
Hao Xuc8d317a2020-09-29 20:00:45 +08003186 req->rw.kiocb.ki_flags &= ~IOCB_WAITQ;
Jens Axboebcf5a062020-05-22 09:24:42 -06003187 list_del_init(&wait->entry);
3188
Jens Axboebcf5a062020-05-22 09:24:42 -06003189 /* submit ref gets dropped, acquire a new one */
Jens Axboede9b4cc2021-02-24 13:28:27 -07003190 req_ref_get(req);
Pavel Begunkov921b9052021-02-12 03:23:53 +00003191 io_req_task_queue(req);
Jens Axboebcf5a062020-05-22 09:24:42 -06003192 return 1;
3193}
3194
Jens Axboec1dd91d2020-08-03 16:43:59 -06003195/*
3196 * This controls whether a given IO request should be armed for async page
3197 * based retry. If we return false here, the request is handed to the async
3198 * worker threads for retry. If we're doing buffered reads on a regular file,
3199 * we prepare a private wait_page_queue entry and retry the operation. This
3200 * will either succeed because the page is now uptodate and unlocked, or it
3201 * will register a callback when the page is unlocked at IO completion. Through
3202 * that callback, io_uring uses task_work to setup a retry of the operation.
3203 * That retry will attempt the buffered read again. The retry will generally
3204 * succeed, or in rare cases where it fails, we then fall back to using the
3205 * async worker threads for a blocking retry.
3206 */
Jens Axboe227c0c92020-08-13 11:51:40 -06003207static bool io_rw_should_retry(struct io_kiocb *req)
Jens Axboebcf5a062020-05-22 09:24:42 -06003208{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003209 struct io_async_rw *rw = req->async_data;
3210 struct wait_page_queue *wait = &rw->wpq;
Jens Axboebcf5a062020-05-22 09:24:42 -06003211 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboebcf5a062020-05-22 09:24:42 -06003212
3213 /* never retry for NOWAIT, we just complete with -EAGAIN */
3214 if (req->flags & REQ_F_NOWAIT)
3215 return false;
3216
Jens Axboe227c0c92020-08-13 11:51:40 -06003217 /* Only for buffered IO */
Jens Axboe3b2a4432020-08-16 10:58:43 -07003218 if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI))
Jens Axboebcf5a062020-05-22 09:24:42 -06003219 return false;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003220
Jens Axboebcf5a062020-05-22 09:24:42 -06003221 /*
3222 * just use poll if we can, and don't attempt if the fs doesn't
3223 * support callback based unlocks
3224 */
3225 if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC))
3226 return false;
3227
Jens Axboe3b2a4432020-08-16 10:58:43 -07003228 wait->wait.func = io_async_buf_func;
3229 wait->wait.private = req;
3230 wait->wait.flags = 0;
3231 INIT_LIST_HEAD(&wait->wait.entry);
3232 kiocb->ki_flags |= IOCB_WAITQ;
Hao Xuc8d317a2020-09-29 20:00:45 +08003233 kiocb->ki_flags &= ~IOCB_NOWAIT;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003234 kiocb->ki_waitq = wait;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003235 return true;
Jens Axboebcf5a062020-05-22 09:24:42 -06003236}
3237
3238static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter)
3239{
3240 if (req->file->f_op->read_iter)
3241 return call_read_iter(req->file, &req->rw.kiocb, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003242 else if (req->file->f_op->read)
Jens Axboe4017eb92020-10-22 14:14:12 -06003243 return loop_rw_iter(READ, req, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003244 else
3245 return -EINVAL;
Jens Axboebcf5a062020-05-22 09:24:42 -06003246}
3247
Pavel Begunkov889fca72021-02-10 00:03:09 +00003248static int io_read(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003249{
3250 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
Jens Axboe9adbd452019-12-20 08:45:55 -07003251 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboeff6165b2020-08-13 09:47:43 -06003252 struct iov_iter __iter, *iter = &__iter;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003253 struct io_async_rw *rw = req->async_data;
Jens Axboe227c0c92020-08-13 11:51:40 -06003254 ssize_t io_size, ret, ret2;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003255 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003256
Pavel Begunkov2846c482020-11-07 13:16:27 +00003257 if (rw) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07003258 iter = &rw->iter;
Pavel Begunkov2846c482020-11-07 13:16:27 +00003259 iovec = NULL;
3260 } else {
3261 ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock);
3262 if (ret < 0)
3263 return ret;
3264 }
Pavel Begunkov632546c2020-11-07 13:16:26 +00003265 io_size = iov_iter_count(iter);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003266 req->result = io_size;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003267
Jens Axboefd6c2e42019-12-18 12:19:41 -07003268 /* Ensure we clear previously set non-block flag */
3269 if (!force_nonblock)
Jens Axboe29de5f62020-02-20 09:56:08 -07003270 kiocb->ki_flags &= ~IOCB_NOWAIT;
Pavel Begunkova88fc402020-09-30 22:57:53 +03003271 else
3272 kiocb->ki_flags |= IOCB_NOWAIT;
3273
Pavel Begunkov24c74672020-06-21 13:09:51 +03003274 /* If the file doesn't support async, just async punt */
Jens Axboe7b29f922021-03-12 08:30:14 -07003275 if (force_nonblock && !io_file_supports_async(req, READ)) {
Pavel Begunkov6713e7a2021-02-04 13:51:59 +00003276 ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003277 return ret ?: -EAGAIN;
Pavel Begunkov6713e7a2021-02-04 13:51:59 +00003278 }
Jens Axboe9e645e112019-05-10 16:07:28 -06003279
Pavel Begunkov632546c2020-11-07 13:16:26 +00003280 ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size);
Pavel Begunkov5ea5dd42021-02-04 13:52:03 +00003281 if (unlikely(ret)) {
3282 kfree(iovec);
3283 return ret;
3284 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07003285
Jens Axboe227c0c92020-08-13 11:51:40 -06003286 ret = io_iter_do_read(req, iter);
Jens Axboe32960612019-09-23 11:05:34 -06003287
Jens Axboe230d50d2021-04-01 20:41:15 -06003288 if (ret == -EAGAIN || (req->flags & REQ_F_REISSUE)) {
Pavel Begunkov6ad7f2332021-04-08 01:54:39 +01003289 req->flags &= ~REQ_F_REISSUE;
Jens Axboeeefdf302020-08-27 16:40:19 -06003290 /* IOPOLL retry should happen for io-wq threads */
3291 if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboef91daf52020-08-15 15:58:42 -07003292 goto done;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003293 /* no retry on NONBLOCK nor RWF_NOWAIT */
3294 if (req->flags & REQ_F_NOWAIT)
Jens Axboe355afae2020-09-02 09:30:31 -06003295 goto done;
Jens Axboe84216312020-08-24 11:45:26 -06003296 /* some cases will consume bytes even on error returns */
Pavel Begunkov632546c2020-11-07 13:16:26 +00003297 iov_iter_revert(iter, io_size - iov_iter_count(iter));
Jens Axboef38c7e32020-09-25 15:23:43 -06003298 ret = 0;
Jens Axboe230d50d2021-04-01 20:41:15 -06003299 } else if (ret == -EIOCBQUEUED) {
3300 goto out_free;
Pavel Begunkov7335e3b2021-02-04 13:52:02 +00003301 } else if (ret <= 0 || ret == io_size || !force_nonblock ||
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003302 (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) {
Pavel Begunkov7335e3b2021-02-04 13:52:02 +00003303 /* read all, failed, already did sync or don't want to retry */
Jens Axboe00d23d52020-08-25 12:59:22 -06003304 goto done;
Jens Axboe227c0c92020-08-13 11:51:40 -06003305 }
3306
Jens Axboe227c0c92020-08-13 11:51:40 -06003307 ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003308 if (ret2)
3309 return ret2;
3310
Pavel Begunkovfe1cdd52021-02-17 21:02:36 +00003311 iovec = NULL;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003312 rw = req->async_data;
Jens Axboe227c0c92020-08-13 11:51:40 -06003313 /* now use our persistent iterator, if we aren't already */
Jens Axboee8c2bc12020-08-15 18:44:09 -07003314 iter = &rw->iter;
Jens Axboe227c0c92020-08-13 11:51:40 -06003315
Pavel Begunkovb23df912021-02-04 13:52:04 +00003316 do {
3317 io_size -= ret;
3318 rw->bytes_done += ret;
3319 /* if we can retry, do so with the callbacks armed */
3320 if (!io_rw_should_retry(req)) {
3321 kiocb->ki_flags &= ~IOCB_WAITQ;
3322 return -EAGAIN;
3323 }
3324
3325 /*
3326 * Now retry read with the IOCB_WAITQ parts set in the iocb. If
3327 * we get -EIOCBQUEUED, then we'll get a notification when the
3328 * desired page gets unlocked. We can also get a partial read
3329 * here, and if we do, then just retry at the new offset.
3330 */
3331 ret = io_iter_do_read(req, iter);
3332 if (ret == -EIOCBQUEUED)
3333 return 0;
Jens Axboe227c0c92020-08-13 11:51:40 -06003334 /* we got some bytes, but not all. retry. */
Jens Axboeb5b0ecb2021-03-04 21:02:58 -07003335 kiocb->ki_flags &= ~IOCB_WAITQ;
Pavel Begunkovb23df912021-02-04 13:52:04 +00003336 } while (ret > 0 && ret < io_size);
Jens Axboe227c0c92020-08-13 11:51:40 -06003337done:
Pavel Begunkov889fca72021-02-10 00:03:09 +00003338 kiocb_done(kiocb, ret, issue_flags);
Pavel Begunkovfe1cdd52021-02-17 21:02:36 +00003339out_free:
3340 /* it's faster to check here then delegate to kfree */
3341 if (iovec)
3342 kfree(iovec);
Pavel Begunkov5ea5dd42021-02-04 13:52:03 +00003343 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003344}
3345
Pavel Begunkov73debe62020-09-30 22:57:54 +03003346static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07003347{
Jens Axboe3529d8c2019-12-19 18:24:38 -07003348 if (unlikely(!(req->file->f_mode & FMODE_WRITE)))
3349 return -EBADF;
Pavel Begunkov93642ef2021-02-18 18:29:44 +00003350 return io_prep_rw(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07003351}
3352
Pavel Begunkov889fca72021-02-10 00:03:09 +00003353static int io_write(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003354{
3355 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
Jens Axboe9adbd452019-12-20 08:45:55 -07003356 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboeff6165b2020-08-13 09:47:43 -06003357 struct iov_iter __iter, *iter = &__iter;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003358 struct io_async_rw *rw = req->async_data;
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003359 ssize_t ret, ret2, io_size;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003360 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003361
Pavel Begunkov2846c482020-11-07 13:16:27 +00003362 if (rw) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07003363 iter = &rw->iter;
Pavel Begunkov2846c482020-11-07 13:16:27 +00003364 iovec = NULL;
3365 } else {
3366 ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock);
3367 if (ret < 0)
3368 return ret;
3369 }
Pavel Begunkov632546c2020-11-07 13:16:26 +00003370 io_size = iov_iter_count(iter);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003371 req->result = io_size;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003372
Jens Axboefd6c2e42019-12-18 12:19:41 -07003373 /* Ensure we clear previously set non-block flag */
3374 if (!force_nonblock)
Pavel Begunkova88fc402020-09-30 22:57:53 +03003375 kiocb->ki_flags &= ~IOCB_NOWAIT;
3376 else
3377 kiocb->ki_flags |= IOCB_NOWAIT;
Jens Axboefd6c2e42019-12-18 12:19:41 -07003378
Pavel Begunkov24c74672020-06-21 13:09:51 +03003379 /* If the file doesn't support async, just async punt */
Jens Axboe7b29f922021-03-12 08:30:14 -07003380 if (force_nonblock && !io_file_supports_async(req, WRITE))
Jens Axboef67676d2019-12-02 11:03:47 -07003381 goto copy_iov;
Jens Axboef67676d2019-12-02 11:03:47 -07003382
Jens Axboe10d59342019-12-09 20:16:22 -07003383 /* file path doesn't support NOWAIT for non-direct_IO */
3384 if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) &&
3385 (req->flags & REQ_F_ISREG))
Jens Axboef67676d2019-12-02 11:03:47 -07003386 goto copy_iov;
Jens Axboe9e645e112019-05-10 16:07:28 -06003387
Pavel Begunkov632546c2020-11-07 13:16:26 +00003388 ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003389 if (unlikely(ret))
3390 goto out_free;
Roman Penyaev9bf79332019-03-25 20:09:24 +01003391
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003392 /*
3393 * Open-code file_start_write here to grab freeze protection,
3394 * which will be released by another thread in
3395 * io_complete_rw(). Fool lockdep by telling it the lock got
3396 * released so that it doesn't complain about the held lock when
3397 * we return to userspace.
3398 */
3399 if (req->flags & REQ_F_ISREG) {
Darrick J. Wong8a3c84b2020-11-10 16:50:21 -08003400 sb_start_write(file_inode(req->file)->i_sb);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003401 __sb_writers_release(file_inode(req->file)->i_sb,
3402 SB_FREEZE_WRITE);
3403 }
3404 kiocb->ki_flags |= IOCB_WRITE;
Roman Penyaev9bf79332019-03-25 20:09:24 +01003405
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003406 if (req->file->f_op->write_iter)
Jens Axboeff6165b2020-08-13 09:47:43 -06003407 ret2 = call_write_iter(req->file, kiocb, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003408 else if (req->file->f_op->write)
Jens Axboe4017eb92020-10-22 14:14:12 -06003409 ret2 = loop_rw_iter(WRITE, req, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003410 else
3411 ret2 = -EINVAL;
Jens Axboe4ed734b2020-03-20 11:23:41 -06003412
Pavel Begunkov6ad7f2332021-04-08 01:54:39 +01003413 if (req->flags & REQ_F_REISSUE) {
3414 req->flags &= ~REQ_F_REISSUE;
Jens Axboe230d50d2021-04-01 20:41:15 -06003415 ret2 = -EAGAIN;
Pavel Begunkov6ad7f2332021-04-08 01:54:39 +01003416 }
Jens Axboe230d50d2021-04-01 20:41:15 -06003417
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003418 /*
3419 * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just
3420 * retry them without IOCB_NOWAIT.
3421 */
3422 if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT))
3423 ret2 = -EAGAIN;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003424 /* no retry on NONBLOCK nor RWF_NOWAIT */
3425 if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT))
Jens Axboe355afae2020-09-02 09:30:31 -06003426 goto done;
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003427 if (!force_nonblock || ret2 != -EAGAIN) {
Jens Axboeeefdf302020-08-27 16:40:19 -06003428 /* IOPOLL retry should happen for io-wq threads */
3429 if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN)
3430 goto copy_iov;
Jens Axboe355afae2020-09-02 09:30:31 -06003431done:
Pavel Begunkov889fca72021-02-10 00:03:09 +00003432 kiocb_done(kiocb, ret2, issue_flags);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003433 } else {
Jens Axboef67676d2019-12-02 11:03:47 -07003434copy_iov:
Jens Axboe84216312020-08-24 11:45:26 -06003435 /* some cases will consume bytes even on error returns */
Pavel Begunkov632546c2020-11-07 13:16:26 +00003436 iov_iter_revert(iter, io_size - iov_iter_count(iter));
Jens Axboe227c0c92020-08-13 11:51:40 -06003437 ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003438 return ret ?: -EAGAIN;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003439 }
Jens Axboe31b51512019-01-18 22:56:34 -07003440out_free:
Pavel Begunkovf261c162020-08-20 11:34:10 +03003441 /* it's reportedly faster than delegating the null check to kfree() */
Pavel Begunkov252917c2020-07-13 22:59:20 +03003442 if (iovec)
Xiaoguang Wang6f2cc162020-06-18 15:01:56 +08003443 kfree(iovec);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003444 return ret;
3445}
3446
Jens Axboe80a261f2020-09-28 14:23:58 -06003447static int io_renameat_prep(struct io_kiocb *req,
3448 const struct io_uring_sqe *sqe)
3449{
3450 struct io_rename *ren = &req->rename;
3451 const char __user *oldf, *newf;
3452
3453 if (unlikely(req->flags & REQ_F_FIXED_FILE))
3454 return -EBADF;
3455
3456 ren->old_dfd = READ_ONCE(sqe->fd);
3457 oldf = u64_to_user_ptr(READ_ONCE(sqe->addr));
3458 newf = u64_to_user_ptr(READ_ONCE(sqe->addr2));
3459 ren->new_dfd = READ_ONCE(sqe->len);
3460 ren->flags = READ_ONCE(sqe->rename_flags);
3461
3462 ren->oldpath = getname(oldf);
3463 if (IS_ERR(ren->oldpath))
3464 return PTR_ERR(ren->oldpath);
3465
3466 ren->newpath = getname(newf);
3467 if (IS_ERR(ren->newpath)) {
3468 putname(ren->oldpath);
3469 return PTR_ERR(ren->newpath);
3470 }
3471
3472 req->flags |= REQ_F_NEED_CLEANUP;
3473 return 0;
3474}
3475
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003476static int io_renameat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe80a261f2020-09-28 14:23:58 -06003477{
3478 struct io_rename *ren = &req->rename;
3479 int ret;
3480
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003481 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe80a261f2020-09-28 14:23:58 -06003482 return -EAGAIN;
3483
3484 ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd,
3485 ren->newpath, ren->flags);
3486
3487 req->flags &= ~REQ_F_NEED_CLEANUP;
3488 if (ret < 0)
3489 req_set_fail_links(req);
3490 io_req_complete(req, ret);
3491 return 0;
3492}
3493
Jens Axboe14a11432020-09-28 14:27:37 -06003494static int io_unlinkat_prep(struct io_kiocb *req,
3495 const struct io_uring_sqe *sqe)
3496{
3497 struct io_unlink *un = &req->unlink;
3498 const char __user *fname;
3499
3500 if (unlikely(req->flags & REQ_F_FIXED_FILE))
3501 return -EBADF;
3502
3503 un->dfd = READ_ONCE(sqe->fd);
3504
3505 un->flags = READ_ONCE(sqe->unlink_flags);
3506 if (un->flags & ~AT_REMOVEDIR)
3507 return -EINVAL;
3508
3509 fname = u64_to_user_ptr(READ_ONCE(sqe->addr));
3510 un->filename = getname(fname);
3511 if (IS_ERR(un->filename))
3512 return PTR_ERR(un->filename);
3513
3514 req->flags |= REQ_F_NEED_CLEANUP;
3515 return 0;
3516}
3517
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003518static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe14a11432020-09-28 14:27:37 -06003519{
3520 struct io_unlink *un = &req->unlink;
3521 int ret;
3522
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003523 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe14a11432020-09-28 14:27:37 -06003524 return -EAGAIN;
3525
3526 if (un->flags & AT_REMOVEDIR)
3527 ret = do_rmdir(un->dfd, un->filename);
3528 else
3529 ret = do_unlinkat(un->dfd, un->filename);
3530
3531 req->flags &= ~REQ_F_NEED_CLEANUP;
3532 if (ret < 0)
3533 req_set_fail_links(req);
3534 io_req_complete(req, ret);
3535 return 0;
3536}
3537
Jens Axboe36f4fa62020-09-05 11:14:22 -06003538static int io_shutdown_prep(struct io_kiocb *req,
3539 const struct io_uring_sqe *sqe)
3540{
3541#if defined(CONFIG_NET)
3542 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3543 return -EINVAL;
3544 if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags ||
3545 sqe->buf_index)
3546 return -EINVAL;
3547
3548 req->shutdown.how = READ_ONCE(sqe->len);
3549 return 0;
3550#else
3551 return -EOPNOTSUPP;
3552#endif
3553}
3554
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003555static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe36f4fa62020-09-05 11:14:22 -06003556{
3557#if defined(CONFIG_NET)
3558 struct socket *sock;
3559 int ret;
3560
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003561 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe36f4fa62020-09-05 11:14:22 -06003562 return -EAGAIN;
3563
Linus Torvalds48aba792020-12-16 12:44:05 -08003564 sock = sock_from_file(req->file);
Jens Axboe36f4fa62020-09-05 11:14:22 -06003565 if (unlikely(!sock))
Linus Torvalds48aba792020-12-16 12:44:05 -08003566 return -ENOTSOCK;
Jens Axboe36f4fa62020-09-05 11:14:22 -06003567
3568 ret = __sys_shutdown_sock(sock, req->shutdown.how);
Jens Axboea1464682020-12-14 20:57:27 -07003569 if (ret < 0)
3570 req_set_fail_links(req);
Jens Axboe36f4fa62020-09-05 11:14:22 -06003571 io_req_complete(req, ret);
3572 return 0;
3573#else
3574 return -EOPNOTSUPP;
3575#endif
3576}
3577
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003578static int __io_splice_prep(struct io_kiocb *req,
3579 const struct io_uring_sqe *sqe)
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003580{
3581 struct io_splice* sp = &req->splice;
3582 unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003583
Pavel Begunkov3232dd02020-06-03 18:03:22 +03003584 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3585 return -EINVAL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003586
3587 sp->file_in = NULL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003588 sp->len = READ_ONCE(sqe->len);
3589 sp->flags = READ_ONCE(sqe->splice_flags);
3590
3591 if (unlikely(sp->flags & ~valid_flags))
3592 return -EINVAL;
3593
Pavel Begunkov8371adf2020-10-10 18:34:08 +01003594 sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in),
3595 (sp->flags & SPLICE_F_FD_IN_FIXED));
3596 if (!sp->file_in)
3597 return -EBADF;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003598 req->flags |= REQ_F_NEED_CLEANUP;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003599 return 0;
3600}
3601
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003602static int io_tee_prep(struct io_kiocb *req,
3603 const struct io_uring_sqe *sqe)
3604{
3605 if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off))
3606 return -EINVAL;
3607 return __io_splice_prep(req, sqe);
3608}
3609
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003610static int io_tee(struct io_kiocb *req, unsigned int issue_flags)
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003611{
3612 struct io_splice *sp = &req->splice;
3613 struct file *in = sp->file_in;
3614 struct file *out = sp->file_out;
3615 unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED;
3616 long ret = 0;
3617
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003618 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003619 return -EAGAIN;
3620 if (sp->len)
3621 ret = do_tee(in, out, sp->len, flags);
3622
Pavel Begunkove1d767f2021-03-19 17:22:43 +00003623 if (!(sp->flags & SPLICE_F_FD_IN_FIXED))
3624 io_put_file(in);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003625 req->flags &= ~REQ_F_NEED_CLEANUP;
3626
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003627 if (ret != sp->len)
3628 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003629 io_req_complete(req, ret);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003630 return 0;
3631}
3632
3633static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3634{
3635 struct io_splice* sp = &req->splice;
3636
3637 sp->off_in = READ_ONCE(sqe->splice_off_in);
3638 sp->off_out = READ_ONCE(sqe->off);
3639 return __io_splice_prep(req, sqe);
3640}
3641
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003642static int io_splice(struct io_kiocb *req, unsigned int issue_flags)
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003643{
3644 struct io_splice *sp = &req->splice;
3645 struct file *in = sp->file_in;
3646 struct file *out = sp->file_out;
3647 unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED;
3648 loff_t *poff_in, *poff_out;
Pavel Begunkovc9687422020-05-04 23:00:54 +03003649 long ret = 0;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003650
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003651 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov2fb3e822020-05-01 17:09:38 +03003652 return -EAGAIN;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003653
3654 poff_in = (sp->off_in == -1) ? NULL : &sp->off_in;
3655 poff_out = (sp->off_out == -1) ? NULL : &sp->off_out;
Pavel Begunkovc9687422020-05-04 23:00:54 +03003656
Jens Axboe948a7742020-05-17 14:21:38 -06003657 if (sp->len)
Pavel Begunkovc9687422020-05-04 23:00:54 +03003658 ret = do_splice(in, poff_in, out, poff_out, sp->len, flags);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003659
Pavel Begunkove1d767f2021-03-19 17:22:43 +00003660 if (!(sp->flags & SPLICE_F_FD_IN_FIXED))
3661 io_put_file(in);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003662 req->flags &= ~REQ_F_NEED_CLEANUP;
3663
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003664 if (ret != sp->len)
3665 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003666 io_req_complete(req, ret);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003667 return 0;
3668}
3669
Jens Axboe2b188cc2019-01-07 10:46:33 -07003670/*
3671 * IORING_OP_NOP just posts a completion event, nothing else.
3672 */
Pavel Begunkov889fca72021-02-10 00:03:09 +00003673static int io_nop(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003674{
3675 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003676
Jens Axboedef596e2019-01-09 08:59:42 -07003677 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
3678 return -EINVAL;
3679
Pavel Begunkov889fca72021-02-10 00:03:09 +00003680 __io_req_complete(req, issue_flags, 0, 0);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003681 return 0;
3682}
3683
Pavel Begunkov1155c762021-02-18 18:29:38 +00003684static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003685{
Jens Axboe6b063142019-01-10 22:13:58 -07003686 struct io_ring_ctx *ctx = req->ctx;
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003687
Jens Axboe09bb8392019-03-13 12:39:28 -06003688 if (!req->file)
3689 return -EBADF;
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003690
Jens Axboe6b063142019-01-10 22:13:58 -07003691 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboedef596e2019-01-09 08:59:42 -07003692 return -EINVAL;
Jens Axboeedafcce2019-01-09 09:16:05 -07003693 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003694 return -EINVAL;
3695
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003696 req->sync.flags = READ_ONCE(sqe->fsync_flags);
3697 if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC))
3698 return -EINVAL;
3699
3700 req->sync.off = READ_ONCE(sqe->off);
3701 req->sync.len = READ_ONCE(sqe->len);
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003702 return 0;
3703}
3704
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003705static int io_fsync(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe78912932020-01-14 22:09:06 -07003706{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003707 loff_t end = req->sync.off + req->sync.len;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003708 int ret;
3709
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003710 /* fsync always requires a blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003711 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003712 return -EAGAIN;
3713
Jens Axboe9adbd452019-12-20 08:45:55 -07003714 ret = vfs_fsync_range(req->file, req->sync.off,
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003715 end > 0 ? end : LLONG_MAX,
3716 req->sync.flags & IORING_FSYNC_DATASYNC);
3717 if (ret < 0)
3718 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003719 io_req_complete(req, ret);
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003720 return 0;
3721}
3722
Jens Axboed63d1b52019-12-10 10:38:56 -07003723static int io_fallocate_prep(struct io_kiocb *req,
3724 const struct io_uring_sqe *sqe)
3725{
3726 if (sqe->ioprio || sqe->buf_index || sqe->rw_flags)
3727 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03003728 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3729 return -EINVAL;
Jens Axboed63d1b52019-12-10 10:38:56 -07003730
3731 req->sync.off = READ_ONCE(sqe->off);
3732 req->sync.len = READ_ONCE(sqe->addr);
3733 req->sync.mode = READ_ONCE(sqe->len);
3734 return 0;
3735}
3736
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003737static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboed63d1b52019-12-10 10:38:56 -07003738{
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003739 int ret;
Jens Axboed63d1b52019-12-10 10:38:56 -07003740
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003741 /* fallocate always requiring blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003742 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003743 return -EAGAIN;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003744 ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off,
3745 req->sync.len);
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003746 if (ret < 0)
3747 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003748 io_req_complete(req, ret);
Jens Axboed63d1b52019-12-10 10:38:56 -07003749 return 0;
3750}
3751
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003752static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe15b71ab2019-12-11 11:20:36 -07003753{
Jens Axboef8748882020-01-08 17:47:02 -07003754 const char __user *fname;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003755 int ret;
3756
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003757 if (unlikely(sqe->ioprio || sqe->buf_index))
Jens Axboe15b71ab2019-12-11 11:20:36 -07003758 return -EINVAL;
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003759 if (unlikely(req->flags & REQ_F_FIXED_FILE))
Jens Axboecf3040c2020-02-06 21:31:40 -07003760 return -EBADF;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003761
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003762 /* open.how should be already initialised */
3763 if (!(req->open.how.flags & O_PATH) && force_o_largefile())
Jens Axboe08a1d26eb2020-04-08 09:20:54 -06003764 req->open.how.flags |= O_LARGEFILE;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003765
Pavel Begunkov25e72d12020-06-03 18:03:23 +03003766 req->open.dfd = READ_ONCE(sqe->fd);
3767 fname = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboef8748882020-01-08 17:47:02 -07003768 req->open.filename = getname(fname);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003769 if (IS_ERR(req->open.filename)) {
3770 ret = PTR_ERR(req->open.filename);
3771 req->open.filename = NULL;
3772 return ret;
3773 }
Jens Axboe4022e7a2020-03-19 19:23:18 -06003774 req->open.nofile = rlimit(RLIMIT_NOFILE);
Pavel Begunkov8fef80b2020-02-07 23:59:53 +03003775 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003776 return 0;
3777}
3778
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003779static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3780{
3781 u64 flags, mode;
3782
Jens Axboe14587a462020-09-05 11:36:08 -06003783 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe4eb8dde2020-09-18 19:36:24 -06003784 return -EINVAL;
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003785 mode = READ_ONCE(sqe->len);
3786 flags = READ_ONCE(sqe->open_flags);
3787 req->open.how = build_open_how(flags, mode);
3788 return __io_openat_prep(req, sqe);
3789}
3790
Jens Axboecebdb982020-01-08 17:59:24 -07003791static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3792{
3793 struct open_how __user *how;
Jens Axboecebdb982020-01-08 17:59:24 -07003794 size_t len;
3795 int ret;
3796
Jens Axboe14587a462020-09-05 11:36:08 -06003797 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe4eb8dde2020-09-18 19:36:24 -06003798 return -EINVAL;
Jens Axboecebdb982020-01-08 17:59:24 -07003799 how = u64_to_user_ptr(READ_ONCE(sqe->addr2));
3800 len = READ_ONCE(sqe->len);
Jens Axboecebdb982020-01-08 17:59:24 -07003801 if (len < OPEN_HOW_SIZE_VER0)
3802 return -EINVAL;
3803
3804 ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how,
3805 len);
3806 if (ret)
3807 return ret;
3808
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003809 return __io_openat_prep(req, sqe);
Jens Axboecebdb982020-01-08 17:59:24 -07003810}
3811
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003812static int io_openat2(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe15b71ab2019-12-11 11:20:36 -07003813{
3814 struct open_flags op;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003815 struct file *file;
Jens Axboe3a81fd02020-12-10 12:25:36 -07003816 bool nonblock_set;
3817 bool resolve_nonblock;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003818 int ret;
3819
Jens Axboecebdb982020-01-08 17:59:24 -07003820 ret = build_open_flags(&req->open.how, &op);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003821 if (ret)
3822 goto err;
Jens Axboe3a81fd02020-12-10 12:25:36 -07003823 nonblock_set = op.open_flag & O_NONBLOCK;
3824 resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003825 if (issue_flags & IO_URING_F_NONBLOCK) {
Jens Axboe3a81fd02020-12-10 12:25:36 -07003826 /*
3827 * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open,
3828 * it'll always -EAGAIN
3829 */
3830 if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE))
3831 return -EAGAIN;
3832 op.lookup_flags |= LOOKUP_CACHED;
3833 op.open_flag |= O_NONBLOCK;
3834 }
Jens Axboe15b71ab2019-12-11 11:20:36 -07003835
Jens Axboe4022e7a2020-03-19 19:23:18 -06003836 ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003837 if (ret < 0)
3838 goto err;
3839
3840 file = do_filp_open(req->open.dfd, req->open.filename, &op);
Jens Axboe3a81fd02020-12-10 12:25:36 -07003841 /* only retry if RESOLVE_CACHED wasn't already set by application */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003842 if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) &&
3843 file == ERR_PTR(-EAGAIN)) {
Jens Axboe3a81fd02020-12-10 12:25:36 -07003844 /*
3845 * We could hang on to this 'fd', but seems like marginal
3846 * gain for something that is now known to be a slower path.
3847 * So just put it, and we'll get a new one when we retry.
3848 */
3849 put_unused_fd(ret);
3850 return -EAGAIN;
3851 }
3852
Jens Axboe15b71ab2019-12-11 11:20:36 -07003853 if (IS_ERR(file)) {
3854 put_unused_fd(ret);
3855 ret = PTR_ERR(file);
3856 } else {
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003857 if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set)
Jens Axboe3a81fd02020-12-10 12:25:36 -07003858 file->f_flags &= ~O_NONBLOCK;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003859 fsnotify_open(file);
3860 fd_install(ret, file);
3861 }
3862err:
3863 putname(req->open.filename);
Pavel Begunkov8fef80b2020-02-07 23:59:53 +03003864 req->flags &= ~REQ_F_NEED_CLEANUP;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003865 if (ret < 0)
3866 req_set_fail_links(req);
Pavel Begunkov0bdf3392021-04-11 01:46:29 +01003867 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003868 return 0;
3869}
3870
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003871static int io_openat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboecebdb982020-01-08 17:59:24 -07003872{
Pavel Begunkove45cff52021-02-28 22:35:14 +00003873 return io_openat2(req, issue_flags);
Jens Axboecebdb982020-01-08 17:59:24 -07003874}
3875
Jens Axboe067524e2020-03-02 16:32:28 -07003876static int io_remove_buffers_prep(struct io_kiocb *req,
3877 const struct io_uring_sqe *sqe)
3878{
3879 struct io_provide_buf *p = &req->pbuf;
3880 u64 tmp;
3881
3882 if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off)
3883 return -EINVAL;
3884
3885 tmp = READ_ONCE(sqe->fd);
3886 if (!tmp || tmp > USHRT_MAX)
3887 return -EINVAL;
3888
3889 memset(p, 0, sizeof(*p));
3890 p->nbufs = tmp;
3891 p->bgid = READ_ONCE(sqe->buf_group);
3892 return 0;
3893}
3894
3895static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf,
3896 int bgid, unsigned nbufs)
3897{
3898 unsigned i = 0;
3899
3900 /* shouldn't happen */
3901 if (!nbufs)
3902 return 0;
3903
3904 /* the head kbuf is the list itself */
3905 while (!list_empty(&buf->list)) {
3906 struct io_buffer *nxt;
3907
3908 nxt = list_first_entry(&buf->list, struct io_buffer, list);
3909 list_del(&nxt->list);
3910 kfree(nxt);
3911 if (++i == nbufs)
3912 return i;
3913 }
3914 i++;
3915 kfree(buf);
Jens Axboe9e15c3a2021-03-13 12:29:43 -07003916 xa_erase(&ctx->io_buffers, bgid);
Jens Axboe067524e2020-03-02 16:32:28 -07003917
3918 return i;
3919}
3920
Pavel Begunkov889fca72021-02-10 00:03:09 +00003921static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe067524e2020-03-02 16:32:28 -07003922{
3923 struct io_provide_buf *p = &req->pbuf;
3924 struct io_ring_ctx *ctx = req->ctx;
3925 struct io_buffer *head;
3926 int ret = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003927 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe067524e2020-03-02 16:32:28 -07003928
3929 io_ring_submit_lock(ctx, !force_nonblock);
3930
3931 lockdep_assert_held(&ctx->uring_lock);
3932
3933 ret = -ENOENT;
Jens Axboe9e15c3a2021-03-13 12:29:43 -07003934 head = xa_load(&ctx->io_buffers, p->bgid);
Jens Axboe067524e2020-03-02 16:32:28 -07003935 if (head)
3936 ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs);
Jens Axboe067524e2020-03-02 16:32:28 -07003937 if (ret < 0)
3938 req_set_fail_links(req);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00003939
Pavel Begunkov9fb8cb42021-02-28 22:35:13 +00003940 /* complete before unlock, IOPOLL may need the lock */
3941 __io_req_complete(req, issue_flags, ret, 0);
3942 io_ring_submit_unlock(ctx, !force_nonblock);
Jens Axboe067524e2020-03-02 16:32:28 -07003943 return 0;
3944}
3945
Jens Axboeddf0322d2020-02-23 16:41:33 -07003946static int io_provide_buffers_prep(struct io_kiocb *req,
3947 const struct io_uring_sqe *sqe)
3948{
Pavel Begunkovd81269f2021-03-19 10:21:19 +00003949 unsigned long size;
Jens Axboeddf0322d2020-02-23 16:41:33 -07003950 struct io_provide_buf *p = &req->pbuf;
3951 u64 tmp;
3952
3953 if (sqe->ioprio || sqe->rw_flags)
3954 return -EINVAL;
3955
3956 tmp = READ_ONCE(sqe->fd);
3957 if (!tmp || tmp > USHRT_MAX)
3958 return -E2BIG;
3959 p->nbufs = tmp;
3960 p->addr = READ_ONCE(sqe->addr);
3961 p->len = READ_ONCE(sqe->len);
3962
Pavel Begunkovd81269f2021-03-19 10:21:19 +00003963 size = (unsigned long)p->len * p->nbufs;
3964 if (!access_ok(u64_to_user_ptr(p->addr), size))
Jens Axboeddf0322d2020-02-23 16:41:33 -07003965 return -EFAULT;
3966
3967 p->bgid = READ_ONCE(sqe->buf_group);
3968 tmp = READ_ONCE(sqe->off);
3969 if (tmp > USHRT_MAX)
3970 return -E2BIG;
3971 p->bid = tmp;
3972 return 0;
3973}
3974
3975static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head)
3976{
3977 struct io_buffer *buf;
3978 u64 addr = pbuf->addr;
3979 int i, bid = pbuf->bid;
3980
3981 for (i = 0; i < pbuf->nbufs; i++) {
3982 buf = kmalloc(sizeof(*buf), GFP_KERNEL);
3983 if (!buf)
3984 break;
3985
3986 buf->addr = addr;
3987 buf->len = pbuf->len;
3988 buf->bid = bid;
3989 addr += pbuf->len;
3990 bid++;
3991 if (!*head) {
3992 INIT_LIST_HEAD(&buf->list);
3993 *head = buf;
3994 } else {
3995 list_add_tail(&buf->list, &(*head)->list);
3996 }
3997 }
3998
3999 return i ? i : -ENOMEM;
4000}
4001
Pavel Begunkov889fca72021-02-10 00:03:09 +00004002static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeddf0322d2020-02-23 16:41:33 -07004003{
4004 struct io_provide_buf *p = &req->pbuf;
4005 struct io_ring_ctx *ctx = req->ctx;
4006 struct io_buffer *head, *list;
4007 int ret = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004008 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboeddf0322d2020-02-23 16:41:33 -07004009
4010 io_ring_submit_lock(ctx, !force_nonblock);
4011
4012 lockdep_assert_held(&ctx->uring_lock);
4013
Jens Axboe9e15c3a2021-03-13 12:29:43 -07004014 list = head = xa_load(&ctx->io_buffers, p->bgid);
Jens Axboeddf0322d2020-02-23 16:41:33 -07004015
4016 ret = io_add_buffers(p, &head);
Jens Axboe9e15c3a2021-03-13 12:29:43 -07004017 if (ret >= 0 && !list) {
4018 ret = xa_insert(&ctx->io_buffers, p->bgid, head, GFP_KERNEL);
4019 if (ret < 0)
Jens Axboe067524e2020-03-02 16:32:28 -07004020 __io_remove_buffers(ctx, head, p->bgid, -1U);
Jens Axboeddf0322d2020-02-23 16:41:33 -07004021 }
Jens Axboeddf0322d2020-02-23 16:41:33 -07004022 if (ret < 0)
4023 req_set_fail_links(req);
Pavel Begunkov9fb8cb42021-02-28 22:35:13 +00004024 /* complete before unlock, IOPOLL may need the lock */
4025 __io_req_complete(req, issue_flags, ret, 0);
4026 io_ring_submit_unlock(ctx, !force_nonblock);
Jens Axboeddf0322d2020-02-23 16:41:33 -07004027 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07004028}
4029
Jens Axboe3e4827b2020-01-08 15:18:09 -07004030static int io_epoll_ctl_prep(struct io_kiocb *req,
4031 const struct io_uring_sqe *sqe)
4032{
4033#if defined(CONFIG_EPOLL)
4034 if (sqe->ioprio || sqe->buf_index)
4035 return -EINVAL;
Jens Axboe6ca56f82020-09-18 16:51:19 -06004036 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL)))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004037 return -EINVAL;
Jens Axboe3e4827b2020-01-08 15:18:09 -07004038
4039 req->epoll.epfd = READ_ONCE(sqe->fd);
4040 req->epoll.op = READ_ONCE(sqe->len);
4041 req->epoll.fd = READ_ONCE(sqe->off);
4042
4043 if (ep_op_has_event(req->epoll.op)) {
4044 struct epoll_event __user *ev;
4045
4046 ev = u64_to_user_ptr(READ_ONCE(sqe->addr));
4047 if (copy_from_user(&req->epoll.event, ev, sizeof(*ev)))
4048 return -EFAULT;
4049 }
4050
4051 return 0;
4052#else
4053 return -EOPNOTSUPP;
4054#endif
4055}
4056
Pavel Begunkov889fca72021-02-10 00:03:09 +00004057static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe3e4827b2020-01-08 15:18:09 -07004058{
4059#if defined(CONFIG_EPOLL)
4060 struct io_epoll *ie = &req->epoll;
4061 int ret;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004062 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe3e4827b2020-01-08 15:18:09 -07004063
4064 ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock);
4065 if (force_nonblock && ret == -EAGAIN)
4066 return -EAGAIN;
4067
4068 if (ret < 0)
4069 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004070 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe3e4827b2020-01-08 15:18:09 -07004071 return 0;
4072#else
4073 return -EOPNOTSUPP;
4074#endif
4075}
4076
Jens Axboec1ca7572019-12-25 22:18:28 -07004077static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4078{
4079#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
4080 if (sqe->ioprio || sqe->buf_index || sqe->off)
4081 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004082 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4083 return -EINVAL;
Jens Axboec1ca7572019-12-25 22:18:28 -07004084
4085 req->madvise.addr = READ_ONCE(sqe->addr);
4086 req->madvise.len = READ_ONCE(sqe->len);
4087 req->madvise.advice = READ_ONCE(sqe->fadvise_advice);
4088 return 0;
4089#else
4090 return -EOPNOTSUPP;
4091#endif
4092}
4093
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004094static int io_madvise(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboec1ca7572019-12-25 22:18:28 -07004095{
4096#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
4097 struct io_madvise *ma = &req->madvise;
4098 int ret;
4099
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004100 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboec1ca7572019-12-25 22:18:28 -07004101 return -EAGAIN;
4102
Minchan Kim0726b012020-10-17 16:14:50 -07004103 ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice);
Jens Axboec1ca7572019-12-25 22:18:28 -07004104 if (ret < 0)
4105 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004106 io_req_complete(req, ret);
Jens Axboec1ca7572019-12-25 22:18:28 -07004107 return 0;
4108#else
4109 return -EOPNOTSUPP;
4110#endif
4111}
4112
Jens Axboe4840e412019-12-25 22:03:45 -07004113static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4114{
4115 if (sqe->ioprio || sqe->buf_index || sqe->addr)
4116 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004117 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4118 return -EINVAL;
Jens Axboe4840e412019-12-25 22:03:45 -07004119
4120 req->fadvise.offset = READ_ONCE(sqe->off);
4121 req->fadvise.len = READ_ONCE(sqe->len);
4122 req->fadvise.advice = READ_ONCE(sqe->fadvise_advice);
4123 return 0;
4124}
4125
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004126static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe4840e412019-12-25 22:03:45 -07004127{
4128 struct io_fadvise *fa = &req->fadvise;
4129 int ret;
4130
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004131 if (issue_flags & IO_URING_F_NONBLOCK) {
Jens Axboe3e694262020-02-01 09:22:49 -07004132 switch (fa->advice) {
4133 case POSIX_FADV_NORMAL:
4134 case POSIX_FADV_RANDOM:
4135 case POSIX_FADV_SEQUENTIAL:
4136 break;
4137 default:
4138 return -EAGAIN;
4139 }
4140 }
Jens Axboe4840e412019-12-25 22:03:45 -07004141
4142 ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice);
4143 if (ret < 0)
4144 req_set_fail_links(req);
Pavel Begunkov0bdf3392021-04-11 01:46:29 +01004145 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe4840e412019-12-25 22:03:45 -07004146 return 0;
4147}
4148
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004149static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4150{
Jens Axboe6ca56f82020-09-18 16:51:19 -06004151 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL)))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004152 return -EINVAL;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004153 if (sqe->ioprio || sqe->buf_index)
4154 return -EINVAL;
Pavel Begunkov9c280f92020-04-08 08:58:46 +03004155 if (req->flags & REQ_F_FIXED_FILE)
Jens Axboecf3040c2020-02-06 21:31:40 -07004156 return -EBADF;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004157
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004158 req->statx.dfd = READ_ONCE(sqe->fd);
4159 req->statx.mask = READ_ONCE(sqe->len);
Bijan Mottahedehe62753e2020-05-22 21:31:18 -07004160 req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr));
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004161 req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2));
4162 req->statx.flags = READ_ONCE(sqe->statx_flags);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004163
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004164 return 0;
4165}
4166
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004167static int io_statx(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004168{
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004169 struct io_statx *ctx = &req->statx;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004170 int ret;
4171
Pavel Begunkov59d70012021-03-22 01:58:30 +00004172 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004173 return -EAGAIN;
4174
Bijan Mottahedehe62753e2020-05-22 21:31:18 -07004175 ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask,
4176 ctx->buffer);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004177
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004178 if (ret < 0)
4179 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004180 io_req_complete(req, ret);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004181 return 0;
4182}
4183
Jens Axboeb5dba592019-12-11 14:02:38 -07004184static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4185{
Jens Axboe14587a462020-09-05 11:36:08 -06004186 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004187 return -EINVAL;
Jens Axboeb5dba592019-12-11 14:02:38 -07004188 if (sqe->ioprio || sqe->off || sqe->addr || sqe->len ||
4189 sqe->rw_flags || sqe->buf_index)
4190 return -EINVAL;
Pavel Begunkov9c280f92020-04-08 08:58:46 +03004191 if (req->flags & REQ_F_FIXED_FILE)
Jens Axboecf3040c2020-02-06 21:31:40 -07004192 return -EBADF;
Jens Axboeb5dba592019-12-11 14:02:38 -07004193
4194 req->close.fd = READ_ONCE(sqe->fd);
Jens Axboeb5dba592019-12-11 14:02:38 -07004195 return 0;
4196}
4197
Pavel Begunkov889fca72021-02-10 00:03:09 +00004198static int io_close(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeb5dba592019-12-11 14:02:38 -07004199{
Jens Axboe9eac1902021-01-19 15:50:37 -07004200 struct files_struct *files = current->files;
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004201 struct io_close *close = &req->close;
Jens Axboe9eac1902021-01-19 15:50:37 -07004202 struct fdtable *fdt;
Pavel Begunkova1fde922021-04-11 01:46:28 +01004203 struct file *file = NULL;
4204 int ret = -EBADF;
Jens Axboeb5dba592019-12-11 14:02:38 -07004205
Jens Axboe9eac1902021-01-19 15:50:37 -07004206 spin_lock(&files->file_lock);
4207 fdt = files_fdtable(files);
4208 if (close->fd >= fdt->max_fds) {
4209 spin_unlock(&files->file_lock);
4210 goto err;
4211 }
4212 file = fdt->fd[close->fd];
Pavel Begunkova1fde922021-04-11 01:46:28 +01004213 if (!file || file->f_op == &io_uring_fops) {
Jens Axboe9eac1902021-01-19 15:50:37 -07004214 spin_unlock(&files->file_lock);
4215 file = NULL;
4216 goto err;
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004217 }
Jens Axboeb5dba592019-12-11 14:02:38 -07004218
4219 /* if the file has a flush method, be safe and punt to async */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004220 if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) {
Jens Axboe9eac1902021-01-19 15:50:37 -07004221 spin_unlock(&files->file_lock);
Pavel Begunkov0bf0eef2020-05-26 20:34:06 +03004222 return -EAGAIN;
Pavel Begunkova2100672020-03-02 23:45:16 +03004223 }
Jens Axboeb5dba592019-12-11 14:02:38 -07004224
Jens Axboe9eac1902021-01-19 15:50:37 -07004225 ret = __close_fd_get_file(close->fd, &file);
4226 spin_unlock(&files->file_lock);
4227 if (ret < 0) {
4228 if (ret == -ENOENT)
4229 ret = -EBADF;
4230 goto err;
4231 }
4232
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004233 /* No ->flush() or already async, safely close from here */
Jens Axboe9eac1902021-01-19 15:50:37 -07004234 ret = filp_close(file, current->files);
4235err:
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004236 if (ret < 0)
4237 req_set_fail_links(req);
Jens Axboe9eac1902021-01-19 15:50:37 -07004238 if (file)
4239 fput(file);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004240 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe1a417f42020-01-31 17:16:48 -07004241 return 0;
Jens Axboeb5dba592019-12-11 14:02:38 -07004242}
4243
Pavel Begunkov1155c762021-02-18 18:29:38 +00004244static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004245{
4246 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004247
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004248 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
4249 return -EINVAL;
4250 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
4251 return -EINVAL;
4252
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004253 req->sync.off = READ_ONCE(sqe->off);
4254 req->sync.len = READ_ONCE(sqe->len);
4255 req->sync.flags = READ_ONCE(sqe->sync_range_flags);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004256 return 0;
4257}
4258
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004259static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004260{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004261 int ret;
4262
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004263 /* sync_file_range always requires a blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004264 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004265 return -EAGAIN;
4266
Jens Axboe9adbd452019-12-20 08:45:55 -07004267 ret = sync_file_range(req->file, req->sync.off, req->sync.len,
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004268 req->sync.flags);
4269 if (ret < 0)
4270 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004271 io_req_complete(req, ret);
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004272 return 0;
4273}
4274
YueHaibing469956e2020-03-04 15:53:52 +08004275#if defined(CONFIG_NET)
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004276static int io_setup_async_msg(struct io_kiocb *req,
4277 struct io_async_msghdr *kmsg)
4278{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004279 struct io_async_msghdr *async_msg = req->async_data;
4280
4281 if (async_msg)
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004282 return -EAGAIN;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004283 if (io_alloc_async_data(req)) {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004284 kfree(kmsg->free_iov);
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004285 return -ENOMEM;
4286 }
Jens Axboee8c2bc12020-08-15 18:44:09 -07004287 async_msg = req->async_data;
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004288 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004289 memcpy(async_msg, kmsg, sizeof(*kmsg));
Pavel Begunkov2a780802021-02-05 00:57:58 +00004290 async_msg->msg.msg_name = &async_msg->addr;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004291 /* if were using fast_iov, set it to the new one */
4292 if (!async_msg->free_iov)
4293 async_msg->msg.msg_iter.iov = async_msg->fast_iov;
4294
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004295 return -EAGAIN;
4296}
4297
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004298static int io_sendmsg_copy_hdr(struct io_kiocb *req,
4299 struct io_async_msghdr *iomsg)
4300{
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004301 iomsg->msg.msg_name = &iomsg->addr;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004302 iomsg->free_iov = iomsg->fast_iov;
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004303 return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004304 req->sr_msg.msg_flags, &iomsg->free_iov);
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004305}
4306
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004307static int io_sendmsg_prep_async(struct io_kiocb *req)
4308{
4309 int ret;
4310
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004311 ret = io_sendmsg_copy_hdr(req, req->async_data);
4312 if (!ret)
4313 req->flags |= REQ_F_NEED_CLEANUP;
4314 return ret;
4315}
4316
Jens Axboe3529d8c2019-12-19 18:24:38 -07004317static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboeaa1fa282019-04-19 13:38:09 -06004318{
Jens Axboee47293f2019-12-20 08:58:21 -07004319 struct io_sr_msg *sr = &req->sr_msg;
Jens Axboe03b12302019-12-02 18:50:25 -07004320
Pavel Begunkovd2b6f482020-06-03 18:03:25 +03004321 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4322 return -EINVAL;
4323
Pavel Begunkov270a5942020-07-12 20:41:04 +03004324 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboefddafac2020-01-04 20:19:44 -07004325 sr->len = READ_ONCE(sqe->len);
Pavel Begunkov04411802021-04-01 15:44:00 +01004326 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
4327 if (sr->msg_flags & MSG_DONTWAIT)
4328 req->flags |= REQ_F_NOWAIT;
Jens Axboe3529d8c2019-12-19 18:24:38 -07004329
Jens Axboed8768362020-02-27 14:17:49 -07004330#ifdef CONFIG_COMPAT
4331 if (req->ctx->compat)
4332 sr->msg_flags |= MSG_CMSG_COMPAT;
4333#endif
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004334 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004335}
4336
Pavel Begunkov889fca72021-02-10 00:03:09 +00004337static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004338{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004339 struct io_async_msghdr iomsg, *kmsg;
Jens Axboe03b12302019-12-02 18:50:25 -07004340 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004341 unsigned flags;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004342 int min_ret = 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004343 int ret;
4344
Florent Revestdba4a922020-12-04 12:36:04 +01004345 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004346 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004347 return -ENOTSOCK;
Jens Axboe03b12302019-12-02 18:50:25 -07004348
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004349 kmsg = req->async_data;
4350 if (!kmsg) {
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004351 ret = io_sendmsg_copy_hdr(req, &iomsg);
Jens Axboefddafac2020-01-04 20:19:44 -07004352 if (ret)
4353 return ret;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004354 kmsg = &iomsg;
Jens Axboefddafac2020-01-04 20:19:44 -07004355 }
4356
Pavel Begunkov04411802021-04-01 15:44:00 +01004357 flags = req->sr_msg.msg_flags;
4358 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004359 flags |= MSG_DONTWAIT;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004360 if (flags & MSG_WAITALL)
4361 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
4362
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004363 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004364 if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004365 return io_setup_async_msg(req, kmsg);
4366 if (ret == -ERESTARTSYS)
4367 ret = -EINTR;
4368
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004369 /* fast path, check for non-NULL to avoid function call */
4370 if (kmsg->free_iov)
4371 kfree(kmsg->free_iov);
Jens Axboe03b12302019-12-02 18:50:25 -07004372 req->flags &= ~REQ_F_NEED_CLEANUP;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004373 if (ret < min_ret)
Jens Axboefddafac2020-01-04 20:19:44 -07004374 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004375 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboefddafac2020-01-04 20:19:44 -07004376 return 0;
Jens Axboefddafac2020-01-04 20:19:44 -07004377}
4378
Pavel Begunkov889fca72021-02-10 00:03:09 +00004379static int io_send(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004380{
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004381 struct io_sr_msg *sr = &req->sr_msg;
4382 struct msghdr msg;
4383 struct iovec iov;
Jens Axboe03b12302019-12-02 18:50:25 -07004384 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004385 unsigned flags;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004386 int min_ret = 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004387 int ret;
4388
Florent Revestdba4a922020-12-04 12:36:04 +01004389 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004390 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004391 return -ENOTSOCK;
Jens Axboe03b12302019-12-02 18:50:25 -07004392
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004393 ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter);
4394 if (unlikely(ret))
Zheng Bin14db8412020-09-09 20:12:37 +08004395 return ret;
Jens Axboe03b12302019-12-02 18:50:25 -07004396
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004397 msg.msg_name = NULL;
4398 msg.msg_control = NULL;
4399 msg.msg_controllen = 0;
4400 msg.msg_namelen = 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004401
Pavel Begunkov04411802021-04-01 15:44:00 +01004402 flags = req->sr_msg.msg_flags;
4403 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004404 flags |= MSG_DONTWAIT;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004405 if (flags & MSG_WAITALL)
4406 min_ret = iov_iter_count(&msg.msg_iter);
4407
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004408 msg.msg_flags = flags;
4409 ret = sock_sendmsg(sock, &msg);
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004410 if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004411 return -EAGAIN;
4412 if (ret == -ERESTARTSYS)
4413 ret = -EINTR;
Jens Axboe03b12302019-12-02 18:50:25 -07004414
Stefan Metzmacher00312752021-03-20 20:33:36 +01004415 if (ret < min_ret)
Jens Axboe03b12302019-12-02 18:50:25 -07004416 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004417 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe03b12302019-12-02 18:50:25 -07004418 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004419}
4420
Pavel Begunkov1400e692020-07-12 20:41:05 +03004421static int __io_recvmsg_copy_hdr(struct io_kiocb *req,
4422 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004423{
4424 struct io_sr_msg *sr = &req->sr_msg;
4425 struct iovec __user *uiov;
4426 size_t iov_len;
4427 int ret;
4428
Pavel Begunkov1400e692020-07-12 20:41:05 +03004429 ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg,
4430 &iomsg->uaddr, &uiov, &iov_len);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004431 if (ret)
4432 return ret;
4433
4434 if (req->flags & REQ_F_BUFFER_SELECT) {
4435 if (iov_len > 1)
4436 return -EINVAL;
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004437 if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov)))
Jens Axboe52de1fe2020-02-27 10:15:42 -07004438 return -EFAULT;
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004439 sr->len = iomsg->fast_iov[0].iov_len;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004440 iomsg->free_iov = NULL;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004441 } else {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004442 iomsg->free_iov = iomsg->fast_iov;
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004443 ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004444 &iomsg->free_iov, &iomsg->msg.msg_iter,
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004445 false);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004446 if (ret > 0)
4447 ret = 0;
4448 }
4449
4450 return ret;
4451}
4452
4453#ifdef CONFIG_COMPAT
4454static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req,
Pavel Begunkov1400e692020-07-12 20:41:05 +03004455 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004456{
Jens Axboe52de1fe2020-02-27 10:15:42 -07004457 struct io_sr_msg *sr = &req->sr_msg;
4458 struct compat_iovec __user *uiov;
4459 compat_uptr_t ptr;
4460 compat_size_t len;
4461 int ret;
4462
Pavel Begunkov4af34172021-04-11 01:46:30 +01004463 ret = __get_compat_msghdr(&iomsg->msg, sr->umsg_compat, &iomsg->uaddr,
4464 &ptr, &len);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004465 if (ret)
4466 return ret;
4467
4468 uiov = compat_ptr(ptr);
4469 if (req->flags & REQ_F_BUFFER_SELECT) {
4470 compat_ssize_t clen;
4471
4472 if (len > 1)
4473 return -EINVAL;
4474 if (!access_ok(uiov, sizeof(*uiov)))
4475 return -EFAULT;
4476 if (__get_user(clen, &uiov->iov_len))
4477 return -EFAULT;
4478 if (clen < 0)
4479 return -EINVAL;
Pavel Begunkov2d280bc2020-11-29 18:33:32 +00004480 sr->len = clen;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004481 iomsg->free_iov = NULL;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004482 } else {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004483 iomsg->free_iov = iomsg->fast_iov;
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004484 ret = __import_iovec(READ, (struct iovec __user *)uiov, len,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004485 UIO_FASTIOV, &iomsg->free_iov,
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004486 &iomsg->msg.msg_iter, true);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004487 if (ret < 0)
4488 return ret;
4489 }
4490
4491 return 0;
4492}
Jens Axboe03b12302019-12-02 18:50:25 -07004493#endif
Jens Axboe52de1fe2020-02-27 10:15:42 -07004494
Pavel Begunkov1400e692020-07-12 20:41:05 +03004495static int io_recvmsg_copy_hdr(struct io_kiocb *req,
4496 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004497{
Pavel Begunkov1400e692020-07-12 20:41:05 +03004498 iomsg->msg.msg_name = &iomsg->addr;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004499
4500#ifdef CONFIG_COMPAT
4501 if (req->ctx->compat)
Pavel Begunkov1400e692020-07-12 20:41:05 +03004502 return __io_compat_recvmsg_copy_hdr(req, iomsg);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004503#endif
4504
Pavel Begunkov1400e692020-07-12 20:41:05 +03004505 return __io_recvmsg_copy_hdr(req, iomsg);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004506}
4507
Jens Axboebcda7ba2020-02-23 16:42:51 -07004508static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req,
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004509 bool needs_lock)
Jens Axboebcda7ba2020-02-23 16:42:51 -07004510{
4511 struct io_sr_msg *sr = &req->sr_msg;
4512 struct io_buffer *kbuf;
4513
Jens Axboebcda7ba2020-02-23 16:42:51 -07004514 kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock);
4515 if (IS_ERR(kbuf))
4516 return kbuf;
4517
4518 sr->kbuf = kbuf;
4519 req->flags |= REQ_F_BUFFER_SELECTED;
Jens Axboebcda7ba2020-02-23 16:42:51 -07004520 return kbuf;
Jens Axboe03b12302019-12-02 18:50:25 -07004521}
4522
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004523static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req)
4524{
4525 return io_put_kbuf(req, req->sr_msg.kbuf);
4526}
4527
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004528static int io_recvmsg_prep_async(struct io_kiocb *req)
Jens Axboe03b12302019-12-02 18:50:25 -07004529{
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004530 int ret;
Jens Axboe06b76d42019-12-19 14:44:26 -07004531
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004532 ret = io_recvmsg_copy_hdr(req, req->async_data);
4533 if (!ret)
4534 req->flags |= REQ_F_NEED_CLEANUP;
4535 return ret;
4536}
4537
4538static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4539{
4540 struct io_sr_msg *sr = &req->sr_msg;
4541
Pavel Begunkovd2b6f482020-06-03 18:03:25 +03004542 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4543 return -EINVAL;
4544
Pavel Begunkov270a5942020-07-12 20:41:04 +03004545 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboe0b7b21e2020-01-31 08:34:59 -07004546 sr->len = READ_ONCE(sqe->len);
Jens Axboebcda7ba2020-02-23 16:42:51 -07004547 sr->bgid = READ_ONCE(sqe->buf_group);
Pavel Begunkov04411802021-04-01 15:44:00 +01004548 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
4549 if (sr->msg_flags & MSG_DONTWAIT)
4550 req->flags |= REQ_F_NOWAIT;
Jens Axboe3529d8c2019-12-19 18:24:38 -07004551
Jens Axboed8768362020-02-27 14:17:49 -07004552#ifdef CONFIG_COMPAT
4553 if (req->ctx->compat)
4554 sr->msg_flags |= MSG_CMSG_COMPAT;
4555#endif
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004556 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004557}
4558
Pavel Begunkov889fca72021-02-10 00:03:09 +00004559static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004560{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004561 struct io_async_msghdr iomsg, *kmsg;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004562 struct socket *sock;
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004563 struct io_buffer *kbuf;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004564 unsigned flags;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004565 int min_ret = 0;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004566 int ret, cflags = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004567 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004568
Florent Revestdba4a922020-12-04 12:36:04 +01004569 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004570 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004571 return -ENOTSOCK;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004572
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004573 kmsg = req->async_data;
4574 if (!kmsg) {
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004575 ret = io_recvmsg_copy_hdr(req, &iomsg);
4576 if (ret)
Pavel Begunkov681fda82020-07-15 22:20:45 +03004577 return ret;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004578 kmsg = &iomsg;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004579 }
4580
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004581 if (req->flags & REQ_F_BUFFER_SELECT) {
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004582 kbuf = io_recv_buffer_select(req, !force_nonblock);
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004583 if (IS_ERR(kbuf))
4584 return PTR_ERR(kbuf);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004585 kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr);
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004586 kmsg->fast_iov[0].iov_len = req->sr_msg.len;
4587 iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov,
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004588 1, req->sr_msg.len);
4589 }
4590
Pavel Begunkov04411802021-04-01 15:44:00 +01004591 flags = req->sr_msg.msg_flags;
4592 if (force_nonblock)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004593 flags |= MSG_DONTWAIT;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004594 if (flags & MSG_WAITALL)
4595 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
4596
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004597 ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg,
4598 kmsg->uaddr, flags);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03004599 if (force_nonblock && ret == -EAGAIN)
4600 return io_setup_async_msg(req, kmsg);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004601 if (ret == -ERESTARTSYS)
4602 ret = -EINTR;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03004603
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004604 if (req->flags & REQ_F_BUFFER_SELECTED)
4605 cflags = io_put_recv_kbuf(req);
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004606 /* fast path, check for non-NULL to avoid function call */
4607 if (kmsg->free_iov)
4608 kfree(kmsg->free_iov);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004609 req->flags &= ~REQ_F_NEED_CLEANUP;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004610 if (ret < min_ret || ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))))
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004611 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004612 __io_req_complete(req, issue_flags, ret, cflags);
Jens Axboe0fa03c62019-04-19 13:34:07 -06004613 return 0;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004614}
4615
Pavel Begunkov889fca72021-02-10 00:03:09 +00004616static int io_recv(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboefddafac2020-01-04 20:19:44 -07004617{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004618 struct io_buffer *kbuf;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004619 struct io_sr_msg *sr = &req->sr_msg;
4620 struct msghdr msg;
4621 void __user *buf = sr->buf;
Jens Axboefddafac2020-01-04 20:19:44 -07004622 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004623 struct iovec iov;
4624 unsigned flags;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004625 int min_ret = 0;
Jens Axboebcda7ba2020-02-23 16:42:51 -07004626 int ret, cflags = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004627 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboefddafac2020-01-04 20:19:44 -07004628
Florent Revestdba4a922020-12-04 12:36:04 +01004629 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004630 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004631 return -ENOTSOCK;
Jens Axboefddafac2020-01-04 20:19:44 -07004632
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004633 if (req->flags & REQ_F_BUFFER_SELECT) {
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004634 kbuf = io_recv_buffer_select(req, !force_nonblock);
Jens Axboebcda7ba2020-02-23 16:42:51 -07004635 if (IS_ERR(kbuf))
4636 return PTR_ERR(kbuf);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004637 buf = u64_to_user_ptr(kbuf->addr);
Jens Axboefddafac2020-01-04 20:19:44 -07004638 }
4639
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004640 ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter);
Pavel Begunkov14c32ee2020-07-16 23:28:01 +03004641 if (unlikely(ret))
4642 goto out_free;
Jens Axboefddafac2020-01-04 20:19:44 -07004643
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004644 msg.msg_name = NULL;
4645 msg.msg_control = NULL;
4646 msg.msg_controllen = 0;
4647 msg.msg_namelen = 0;
4648 msg.msg_iocb = NULL;
4649 msg.msg_flags = 0;
4650
Pavel Begunkov04411802021-04-01 15:44:00 +01004651 flags = req->sr_msg.msg_flags;
4652 if (force_nonblock)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004653 flags |= MSG_DONTWAIT;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004654 if (flags & MSG_WAITALL)
4655 min_ret = iov_iter_count(&msg.msg_iter);
4656
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004657 ret = sock_recvmsg(sock, &msg, flags);
4658 if (force_nonblock && ret == -EAGAIN)
4659 return -EAGAIN;
4660 if (ret == -ERESTARTSYS)
4661 ret = -EINTR;
Pavel Begunkov14c32ee2020-07-16 23:28:01 +03004662out_free:
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004663 if (req->flags & REQ_F_BUFFER_SELECTED)
4664 cflags = io_put_recv_kbuf(req);
Stefan Metzmacher00312752021-03-20 20:33:36 +01004665 if (ret < min_ret || ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))))
Jens Axboefddafac2020-01-04 20:19:44 -07004666 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004667 __io_req_complete(req, issue_flags, ret, cflags);
Jens Axboefddafac2020-01-04 20:19:44 -07004668 return 0;
Jens Axboefddafac2020-01-04 20:19:44 -07004669}
4670
Jens Axboe3529d8c2019-12-19 18:24:38 -07004671static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004672{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004673 struct io_accept *accept = &req->accept;
4674
Jens Axboe14587a462020-09-05 11:36:08 -06004675 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe17f2fe32019-10-17 14:42:58 -06004676 return -EINVAL;
Hrvoje Zeba8042d6c2019-11-25 14:40:22 -05004677 if (sqe->ioprio || sqe->len || sqe->buf_index)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004678 return -EINVAL;
4679
Jens Axboed55e5f52019-12-11 16:12:15 -07004680 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
4681 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004682 accept->flags = READ_ONCE(sqe->accept_flags);
Jens Axboe09952e32020-03-19 20:16:56 -06004683 accept->nofile = rlimit(RLIMIT_NOFILE);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004684 return 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004685}
Jens Axboe17f2fe32019-10-17 14:42:58 -06004686
Pavel Begunkov889fca72021-02-10 00:03:09 +00004687static int io_accept(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004688{
4689 struct io_accept *accept = &req->accept;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004690 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004691 unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004692 int ret;
4693
Jiufei Xuee697dee2020-06-10 13:41:59 +08004694 if (req->file->f_flags & O_NONBLOCK)
4695 req->flags |= REQ_F_NOWAIT;
4696
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004697 ret = __sys_accept4_file(req->file, file_flags, accept->addr,
Jens Axboe09952e32020-03-19 20:16:56 -06004698 accept->addr_len, accept->flags,
4699 accept->nofile);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004700 if (ret == -EAGAIN && force_nonblock)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004701 return -EAGAIN;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004702 if (ret < 0) {
4703 if (ret == -ERESTARTSYS)
4704 ret = -EINTR;
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004705 req_set_fail_links(req);
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004706 }
Pavel Begunkov889fca72021-02-10 00:03:09 +00004707 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe17f2fe32019-10-17 14:42:58 -06004708 return 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004709}
4710
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004711static int io_connect_prep_async(struct io_kiocb *req)
4712{
4713 struct io_async_connect *io = req->async_data;
4714 struct io_connect *conn = &req->connect;
4715
4716 return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address);
4717}
4718
Jens Axboe3529d8c2019-12-19 18:24:38 -07004719static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef499a022019-12-02 16:28:46 -07004720{
Jens Axboe3529d8c2019-12-19 18:24:38 -07004721 struct io_connect *conn = &req->connect;
Jens Axboef499a022019-12-02 16:28:46 -07004722
Jens Axboe14587a462020-09-05 11:36:08 -06004723 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004724 return -EINVAL;
4725 if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags)
4726 return -EINVAL;
4727
Jens Axboe3529d8c2019-12-19 18:24:38 -07004728 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
4729 conn->addr_len = READ_ONCE(sqe->addr2);
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004730 return 0;
Jens Axboef499a022019-12-02 16:28:46 -07004731}
4732
Pavel Begunkov889fca72021-02-10 00:03:09 +00004733static int io_connect(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboef8e85cf2019-11-23 14:24:24 -07004734{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004735 struct io_async_connect __io, *io;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004736 unsigned file_flags;
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004737 int ret;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004738 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004739
Jens Axboee8c2bc12020-08-15 18:44:09 -07004740 if (req->async_data) {
4741 io = req->async_data;
Jens Axboef499a022019-12-02 16:28:46 -07004742 } else {
Jens Axboe3529d8c2019-12-19 18:24:38 -07004743 ret = move_addr_to_kernel(req->connect.addr,
4744 req->connect.addr_len,
Jens Axboee8c2bc12020-08-15 18:44:09 -07004745 &__io.address);
Jens Axboef499a022019-12-02 16:28:46 -07004746 if (ret)
4747 goto out;
4748 io = &__io;
4749 }
4750
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004751 file_flags = force_nonblock ? O_NONBLOCK : 0;
4752
Jens Axboee8c2bc12020-08-15 18:44:09 -07004753 ret = __sys_connect_file(req->file, &io->address,
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004754 req->connect.addr_len, file_flags);
Jens Axboe87f80d62019-12-03 11:23:54 -07004755 if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07004756 if (req->async_data)
Jens Axboeb7bb4f72019-12-15 22:13:43 -07004757 return -EAGAIN;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004758 if (io_alloc_async_data(req)) {
Jens Axboef499a022019-12-02 16:28:46 -07004759 ret = -ENOMEM;
4760 goto out;
4761 }
Jens Axboee8c2bc12020-08-15 18:44:09 -07004762 memcpy(req->async_data, &__io, sizeof(__io));
Jens Axboef8e85cf2019-11-23 14:24:24 -07004763 return -EAGAIN;
Jens Axboef499a022019-12-02 16:28:46 -07004764 }
Jens Axboef8e85cf2019-11-23 14:24:24 -07004765 if (ret == -ERESTARTSYS)
4766 ret = -EINTR;
Jens Axboef499a022019-12-02 16:28:46 -07004767out:
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004768 if (ret < 0)
4769 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004770 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboef8e85cf2019-11-23 14:24:24 -07004771 return 0;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004772}
YueHaibing469956e2020-03-04 15:53:52 +08004773#else /* !CONFIG_NET */
Jens Axboe99a10082021-02-19 09:35:19 -07004774#define IO_NETOP_FN(op) \
4775static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \
4776{ \
4777 return -EOPNOTSUPP; \
Jens Axboef8e85cf2019-11-23 14:24:24 -07004778}
4779
Jens Axboe99a10082021-02-19 09:35:19 -07004780#define IO_NETOP_PREP(op) \
4781IO_NETOP_FN(op) \
4782static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \
4783{ \
4784 return -EOPNOTSUPP; \
4785} \
4786
4787#define IO_NETOP_PREP_ASYNC(op) \
4788IO_NETOP_PREP(op) \
4789static int io_##op##_prep_async(struct io_kiocb *req) \
4790{ \
4791 return -EOPNOTSUPP; \
YueHaibing469956e2020-03-04 15:53:52 +08004792}
4793
Jens Axboe99a10082021-02-19 09:35:19 -07004794IO_NETOP_PREP_ASYNC(sendmsg);
4795IO_NETOP_PREP_ASYNC(recvmsg);
4796IO_NETOP_PREP_ASYNC(connect);
4797IO_NETOP_PREP(accept);
4798IO_NETOP_FN(send);
4799IO_NETOP_FN(recv);
YueHaibing469956e2020-03-04 15:53:52 +08004800#endif /* CONFIG_NET */
Jens Axboe17f2fe32019-10-17 14:42:58 -06004801
Jens Axboed7718a92020-02-14 22:23:12 -07004802struct io_poll_table {
4803 struct poll_table_struct pt;
4804 struct io_kiocb *req;
4805 int error;
4806};
4807
Jens Axboed7718a92020-02-14 22:23:12 -07004808static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll,
4809 __poll_t mask, task_work_func_t func)
4810{
Jens Axboeaa96bf82020-04-03 11:26:26 -06004811 int ret;
Jens Axboed7718a92020-02-14 22:23:12 -07004812
4813 /* for instances that support it check for an event match first: */
4814 if (mask && !(mask & poll->events))
4815 return 0;
4816
4817 trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask);
4818
4819 list_del_init(&poll->wait.entry);
4820
Jens Axboed7718a92020-02-14 22:23:12 -07004821 req->result = mask;
Jens Axboe7cbf1722021-02-10 00:03:20 +00004822 req->task_work.func = func;
Jens Axboe6d816e02020-08-11 08:04:14 -06004823
Jens Axboed7718a92020-02-14 22:23:12 -07004824 /*
Jens Axboee3aabf92020-05-18 11:04:17 -06004825 * If this fails, then the task is exiting. When a task exits, the
4826 * work gets canceled, so just cancel this request as well instead
4827 * of executing it. We can't safely execute it anyway, as we may not
4828 * have the needed state needed for it anyway.
Jens Axboed7718a92020-02-14 22:23:12 -07004829 */
Jens Axboe355fb9e2020-10-22 20:19:35 -06004830 ret = io_req_task_work_add(req);
Jens Axboeaa96bf82020-04-03 11:26:26 -06004831 if (unlikely(ret)) {
Jens Axboee3aabf92020-05-18 11:04:17 -06004832 WRITE_ONCE(poll->canceled, true);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00004833 io_req_task_work_add_fallback(req, func);
Jens Axboeaa96bf82020-04-03 11:26:26 -06004834 }
Jens Axboed7718a92020-02-14 22:23:12 -07004835 return 1;
4836}
4837
Jens Axboe74ce6ce2020-04-13 11:09:12 -06004838static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll)
4839 __acquires(&req->ctx->completion_lock)
4840{
4841 struct io_ring_ctx *ctx = req->ctx;
4842
4843 if (!req->result && !READ_ONCE(poll->canceled)) {
4844 struct poll_table_struct pt = { ._key = poll->events };
4845
4846 req->result = vfs_poll(req->file, &pt) & poll->events;
4847 }
4848
4849 spin_lock_irq(&ctx->completion_lock);
4850 if (!req->result && !READ_ONCE(poll->canceled)) {
4851 add_wait_queue(poll->head, &poll->wait);
4852 return true;
4853 }
4854
4855 return false;
4856}
4857
Jens Axboed4e7cd32020-08-15 11:44:50 -07004858static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req)
Jens Axboe18bceab2020-05-15 11:56:54 -06004859{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004860 /* pure poll stashes this in ->async_data, poll driven retry elsewhere */
Jens Axboed4e7cd32020-08-15 11:44:50 -07004861 if (req->opcode == IORING_OP_POLL_ADD)
Jens Axboee8c2bc12020-08-15 18:44:09 -07004862 return req->async_data;
Jens Axboed4e7cd32020-08-15 11:44:50 -07004863 return req->apoll->double_poll;
4864}
4865
4866static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req)
4867{
4868 if (req->opcode == IORING_OP_POLL_ADD)
4869 return &req->poll;
4870 return &req->apoll->poll;
4871}
4872
4873static void io_poll_remove_double(struct io_kiocb *req)
Pavel Begunkove07785b2021-04-01 15:43:57 +01004874 __must_hold(&req->ctx->completion_lock)
Jens Axboed4e7cd32020-08-15 11:44:50 -07004875{
4876 struct io_poll_iocb *poll = io_poll_get_double(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004877
4878 lockdep_assert_held(&req->ctx->completion_lock);
4879
4880 if (poll && poll->head) {
4881 struct wait_queue_head *head = poll->head;
4882
4883 spin_lock(&head->lock);
4884 list_del_init(&poll->wait.entry);
4885 if (poll->wait.private)
Jens Axboede9b4cc2021-02-24 13:28:27 -07004886 req_ref_put(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004887 poll->head = NULL;
4888 spin_unlock(&head->lock);
4889 }
4890}
4891
Pavel Begunkove27414b2021-04-09 09:13:20 +01004892static bool io_poll_complete(struct io_kiocb *req, __poll_t mask)
Pavel Begunkove07785b2021-04-01 15:43:57 +01004893 __must_hold(&req->ctx->completion_lock)
Jens Axboe18bceab2020-05-15 11:56:54 -06004894{
4895 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe88e41cf2021-02-22 22:08:01 -07004896 unsigned flags = IORING_CQE_F_MORE;
Pavel Begunkove27414b2021-04-09 09:13:20 +01004897 int error;
Jens Axboe18bceab2020-05-15 11:56:54 -06004898
Pavel Begunkove27414b2021-04-09 09:13:20 +01004899 if (READ_ONCE(req->poll.canceled)) {
Jens Axboe45ab03b2021-02-23 08:19:33 -07004900 error = -ECANCELED;
Jens Axboe88e41cf2021-02-22 22:08:01 -07004901 req->poll.events |= EPOLLONESHOT;
Pavel Begunkove27414b2021-04-09 09:13:20 +01004902 } else {
Jens Axboe50826202021-02-23 09:02:26 -07004903 error = mangle_poll(mask);
Pavel Begunkove27414b2021-04-09 09:13:20 +01004904 }
Jens Axboeb69de282021-03-17 08:37:41 -06004905 if (req->poll.events & EPOLLONESHOT)
4906 flags = 0;
Pavel Begunkovff6421642021-04-11 01:46:32 +01004907 if (!io_cqring_fill_event(req, error, flags)) {
Jens Axboe50826202021-02-23 09:02:26 -07004908 io_poll_remove_waitqs(req);
Jens Axboe88e41cf2021-02-22 22:08:01 -07004909 req->poll.done = true;
4910 flags = 0;
4911 }
Jens Axboe18bceab2020-05-15 11:56:54 -06004912 io_commit_cqring(ctx);
Jens Axboe88e41cf2021-02-22 22:08:01 -07004913 return !(flags & IORING_CQE_F_MORE);
Jens Axboe18bceab2020-05-15 11:56:54 -06004914}
4915
Jens Axboe18bceab2020-05-15 11:56:54 -06004916static void io_poll_task_func(struct callback_head *cb)
4917{
4918 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
Jens Axboe6d816e02020-08-11 08:04:14 -06004919 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004920 struct io_kiocb *nxt;
Jens Axboe18bceab2020-05-15 11:56:54 -06004921
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004922 if (io_poll_rewait(req, &req->poll)) {
4923 spin_unlock_irq(&ctx->completion_lock);
4924 } else {
Pavel Begunkovf40b9642021-04-09 09:13:19 +01004925 bool done;
Jens Axboe88e41cf2021-02-22 22:08:01 -07004926
Pavel Begunkove27414b2021-04-09 09:13:20 +01004927 done = io_poll_complete(req, req->result);
Jens Axboe88e41cf2021-02-22 22:08:01 -07004928 if (done) {
4929 hash_del(&req->hash_node);
Pavel Begunkovf40b9642021-04-09 09:13:19 +01004930 } else {
Jens Axboe88e41cf2021-02-22 22:08:01 -07004931 req->result = 0;
4932 add_wait_queue(req->poll.head, &req->poll.wait);
4933 }
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004934 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkovf40b9642021-04-09 09:13:19 +01004935 io_cqring_ev_posted(ctx);
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004936
Jens Axboe88e41cf2021-02-22 22:08:01 -07004937 if (done) {
4938 nxt = io_put_req_find_next(req);
4939 if (nxt)
4940 __io_req_task_submit(nxt);
4941 }
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004942 }
Jens Axboe18bceab2020-05-15 11:56:54 -06004943}
4944
4945static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode,
4946 int sync, void *key)
4947{
4948 struct io_kiocb *req = wait->private;
Jens Axboed4e7cd32020-08-15 11:44:50 -07004949 struct io_poll_iocb *poll = io_poll_get_single(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004950 __poll_t mask = key_to_poll(key);
4951
4952 /* for instances that support it check for an event match first: */
4953 if (mask && !(mask & poll->events))
4954 return 0;
Jens Axboe88e41cf2021-02-22 22:08:01 -07004955 if (!(poll->events & EPOLLONESHOT))
4956 return poll->wait.func(&poll->wait, mode, sync, key);
Jens Axboe18bceab2020-05-15 11:56:54 -06004957
Jens Axboe8706e042020-09-28 08:38:54 -06004958 list_del_init(&wait->entry);
4959
Jens Axboe807abcb2020-07-17 17:09:27 -06004960 if (poll && poll->head) {
Jens Axboe18bceab2020-05-15 11:56:54 -06004961 bool done;
4962
Jens Axboe807abcb2020-07-17 17:09:27 -06004963 spin_lock(&poll->head->lock);
4964 done = list_empty(&poll->wait.entry);
Jens Axboe18bceab2020-05-15 11:56:54 -06004965 if (!done)
Jens Axboe807abcb2020-07-17 17:09:27 -06004966 list_del_init(&poll->wait.entry);
Jens Axboed4e7cd32020-08-15 11:44:50 -07004967 /* make sure double remove sees this as being gone */
4968 wait->private = NULL;
Jens Axboe807abcb2020-07-17 17:09:27 -06004969 spin_unlock(&poll->head->lock);
Jens Axboec8b5e262020-10-25 13:53:26 -06004970 if (!done) {
4971 /* use wait func handler, so it matches the rq type */
4972 poll->wait.func(&poll->wait, mode, sync, key);
4973 }
Jens Axboe18bceab2020-05-15 11:56:54 -06004974 }
Jens Axboede9b4cc2021-02-24 13:28:27 -07004975 req_ref_put(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004976 return 1;
4977}
4978
4979static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events,
4980 wait_queue_func_t wake_func)
4981{
4982 poll->head = NULL;
4983 poll->done = false;
4984 poll->canceled = false;
Jens Axboe464dca62021-03-19 14:06:24 -06004985#define IO_POLL_UNMASK (EPOLLERR|EPOLLHUP|EPOLLNVAL|EPOLLRDHUP)
4986 /* mask in events that we always want/need */
4987 poll->events = events | IO_POLL_UNMASK;
Jens Axboe18bceab2020-05-15 11:56:54 -06004988 INIT_LIST_HEAD(&poll->wait.entry);
4989 init_waitqueue_func_entry(&poll->wait, wake_func);
4990}
4991
4992static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt,
Jens Axboe807abcb2020-07-17 17:09:27 -06004993 struct wait_queue_head *head,
4994 struct io_poll_iocb **poll_ptr)
Jens Axboe18bceab2020-05-15 11:56:54 -06004995{
4996 struct io_kiocb *req = pt->req;
4997
4998 /*
4999 * If poll->head is already set, it's because the file being polled
5000 * uses multiple waitqueues for poll handling (eg one for read, one
5001 * for write). Setup a separate io_poll_iocb if this happens.
5002 */
5003 if (unlikely(poll->head)) {
Pavel Begunkov58852d42020-10-16 20:55:56 +01005004 struct io_poll_iocb *poll_one = poll;
5005
Jens Axboe18bceab2020-05-15 11:56:54 -06005006 /* already have a 2nd entry, fail a third attempt */
Jens Axboe807abcb2020-07-17 17:09:27 -06005007 if (*poll_ptr) {
Jens Axboe18bceab2020-05-15 11:56:54 -06005008 pt->error = -EINVAL;
5009 return;
5010 }
Jens Axboe1c3b3e62021-02-28 16:07:30 -07005011 /* double add on the same waitqueue head, ignore */
5012 if (poll->head == head)
5013 return;
Jens Axboe18bceab2020-05-15 11:56:54 -06005014 poll = kmalloc(sizeof(*poll), GFP_ATOMIC);
5015 if (!poll) {
5016 pt->error = -ENOMEM;
5017 return;
5018 }
Pavel Begunkov58852d42020-10-16 20:55:56 +01005019 io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake);
Jens Axboede9b4cc2021-02-24 13:28:27 -07005020 req_ref_get(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06005021 poll->wait.private = req;
Jens Axboe807abcb2020-07-17 17:09:27 -06005022 *poll_ptr = poll;
Jens Axboe18bceab2020-05-15 11:56:54 -06005023 }
5024
5025 pt->error = 0;
5026 poll->head = head;
Jiufei Xuea31eb4a2020-06-17 17:53:56 +08005027
5028 if (poll->events & EPOLLEXCLUSIVE)
5029 add_wait_queue_exclusive(head, &poll->wait);
5030 else
5031 add_wait_queue(head, &poll->wait);
Jens Axboe18bceab2020-05-15 11:56:54 -06005032}
5033
5034static void io_async_queue_proc(struct file *file, struct wait_queue_head *head,
5035 struct poll_table_struct *p)
5036{
5037 struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
Jens Axboe807abcb2020-07-17 17:09:27 -06005038 struct async_poll *apoll = pt->req->apoll;
Jens Axboe18bceab2020-05-15 11:56:54 -06005039
Jens Axboe807abcb2020-07-17 17:09:27 -06005040 __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll);
Jens Axboe18bceab2020-05-15 11:56:54 -06005041}
5042
Jens Axboed7718a92020-02-14 22:23:12 -07005043static void io_async_task_func(struct callback_head *cb)
5044{
5045 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
5046 struct async_poll *apoll = req->apoll;
5047 struct io_ring_ctx *ctx = req->ctx;
5048
5049 trace_io_uring_task_run(req->ctx, req->opcode, req->user_data);
5050
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005051 if (io_poll_rewait(req, &apoll->poll)) {
Jens Axboed7718a92020-02-14 22:23:12 -07005052 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005053 return;
Jens Axboed7718a92020-02-14 22:23:12 -07005054 }
5055
Pavel Begunkov0ea13b42021-04-09 09:13:21 +01005056 hash_del(&req->hash_node);
Jens Axboed4e7cd32020-08-15 11:44:50 -07005057 io_poll_remove_double(req);
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005058 spin_unlock_irq(&ctx->completion_lock);
5059
Pavel Begunkov0be0b0e2020-06-30 15:20:42 +03005060 if (!READ_ONCE(apoll->poll.canceled))
5061 __io_req_task_submit(req);
5062 else
Pavel Begunkov25935532021-03-19 17:22:40 +00005063 io_req_complete_failed(req, -ECANCELED);
Dan Carpenteraa340842020-07-08 21:47:11 +03005064
Jens Axboe807abcb2020-07-17 17:09:27 -06005065 kfree(apoll->double_poll);
Jens Axboe31067252020-05-17 17:43:31 -06005066 kfree(apoll);
Jens Axboed7718a92020-02-14 22:23:12 -07005067}
5068
5069static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
5070 void *key)
5071{
5072 struct io_kiocb *req = wait->private;
5073 struct io_poll_iocb *poll = &req->apoll->poll;
5074
5075 trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data,
5076 key_to_poll(key));
5077
5078 return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func);
5079}
5080
5081static void io_poll_req_insert(struct io_kiocb *req)
5082{
5083 struct io_ring_ctx *ctx = req->ctx;
5084 struct hlist_head *list;
5085
5086 list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)];
5087 hlist_add_head(&req->hash_node, list);
5088}
5089
5090static __poll_t __io_arm_poll_handler(struct io_kiocb *req,
5091 struct io_poll_iocb *poll,
5092 struct io_poll_table *ipt, __poll_t mask,
5093 wait_queue_func_t wake_func)
5094 __acquires(&ctx->completion_lock)
5095{
5096 struct io_ring_ctx *ctx = req->ctx;
5097 bool cancel = false;
5098
Pavel Begunkov4d52f332020-10-18 10:17:43 +01005099 INIT_HLIST_NODE(&req->hash_node);
Jens Axboe18bceab2020-05-15 11:56:54 -06005100 io_init_poll_iocb(poll, mask, wake_func);
Pavel Begunkovb90cd192020-06-21 13:09:52 +03005101 poll->file = req->file;
Jens Axboe18bceab2020-05-15 11:56:54 -06005102 poll->wait.private = req;
Jens Axboed7718a92020-02-14 22:23:12 -07005103
5104 ipt->pt._key = mask;
5105 ipt->req = req;
5106 ipt->error = -EINVAL;
5107
Jens Axboed7718a92020-02-14 22:23:12 -07005108 mask = vfs_poll(req->file, &ipt->pt) & poll->events;
5109
5110 spin_lock_irq(&ctx->completion_lock);
5111 if (likely(poll->head)) {
5112 spin_lock(&poll->head->lock);
5113 if (unlikely(list_empty(&poll->wait.entry))) {
5114 if (ipt->error)
5115 cancel = true;
5116 ipt->error = 0;
5117 mask = 0;
5118 }
Jens Axboe88e41cf2021-02-22 22:08:01 -07005119 if ((mask && (poll->events & EPOLLONESHOT)) || ipt->error)
Jens Axboed7718a92020-02-14 22:23:12 -07005120 list_del_init(&poll->wait.entry);
5121 else if (cancel)
5122 WRITE_ONCE(poll->canceled, true);
5123 else if (!poll->done) /* actually waiting for an event */
5124 io_poll_req_insert(req);
5125 spin_unlock(&poll->head->lock);
5126 }
5127
5128 return mask;
5129}
5130
5131static bool io_arm_poll_handler(struct io_kiocb *req)
5132{
5133 const struct io_op_def *def = &io_op_defs[req->opcode];
5134 struct io_ring_ctx *ctx = req->ctx;
5135 struct async_poll *apoll;
5136 struct io_poll_table ipt;
5137 __poll_t mask, ret;
Jens Axboe9dab14b2020-08-25 12:27:50 -06005138 int rw;
Jens Axboed7718a92020-02-14 22:23:12 -07005139
5140 if (!req->file || !file_can_poll(req->file))
5141 return false;
Pavel Begunkov24c74672020-06-21 13:09:51 +03005142 if (req->flags & REQ_F_POLLED)
Jens Axboed7718a92020-02-14 22:23:12 -07005143 return false;
Jens Axboe9dab14b2020-08-25 12:27:50 -06005144 if (def->pollin)
5145 rw = READ;
5146 else if (def->pollout)
5147 rw = WRITE;
5148 else
5149 return false;
5150 /* if we can't nonblock try, then no point in arming a poll handler */
Jens Axboe7b29f922021-03-12 08:30:14 -07005151 if (!io_file_supports_async(req, rw))
Jens Axboed7718a92020-02-14 22:23:12 -07005152 return false;
5153
5154 apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC);
5155 if (unlikely(!apoll))
5156 return false;
Jens Axboe807abcb2020-07-17 17:09:27 -06005157 apoll->double_poll = NULL;
Jens Axboed7718a92020-02-14 22:23:12 -07005158
5159 req->flags |= REQ_F_POLLED;
Jens Axboed7718a92020-02-14 22:23:12 -07005160 req->apoll = apoll;
Jens Axboed7718a92020-02-14 22:23:12 -07005161
Jens Axboe88e41cf2021-02-22 22:08:01 -07005162 mask = EPOLLONESHOT;
Jens Axboed7718a92020-02-14 22:23:12 -07005163 if (def->pollin)
Nathan Chancellor8755d972020-03-02 16:01:19 -07005164 mask |= POLLIN | POLLRDNORM;
Jens Axboed7718a92020-02-14 22:23:12 -07005165 if (def->pollout)
5166 mask |= POLLOUT | POLLWRNORM;
Luke Hsiao901341b2020-08-21 21:41:05 -07005167
5168 /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */
5169 if ((req->opcode == IORING_OP_RECVMSG) &&
5170 (req->sr_msg.msg_flags & MSG_ERRQUEUE))
5171 mask &= ~POLLIN;
5172
Jens Axboed7718a92020-02-14 22:23:12 -07005173 mask |= POLLERR | POLLPRI;
5174
5175 ipt.pt._qproc = io_async_queue_proc;
5176
5177 ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask,
5178 io_async_wake);
Jens Axboea36da652020-08-11 09:50:19 -06005179 if (ret || ipt.error) {
Jens Axboed4e7cd32020-08-15 11:44:50 -07005180 io_poll_remove_double(req);
Jens Axboed7718a92020-02-14 22:23:12 -07005181 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe807abcb2020-07-17 17:09:27 -06005182 kfree(apoll->double_poll);
Jens Axboed7718a92020-02-14 22:23:12 -07005183 kfree(apoll);
5184 return false;
5185 }
5186 spin_unlock_irq(&ctx->completion_lock);
5187 trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask,
5188 apoll->poll.events);
5189 return true;
5190}
5191
5192static bool __io_poll_remove_one(struct io_kiocb *req,
Jens Axboeb2e720a2021-03-31 09:03:03 -06005193 struct io_poll_iocb *poll, bool do_cancel)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005194 __must_hold(&req->ctx->completion_lock)
Jens Axboed7718a92020-02-14 22:23:12 -07005195{
Jens Axboeb41e9852020-02-17 09:52:41 -07005196 bool do_complete = false;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005197
Jens Axboe50826202021-02-23 09:02:26 -07005198 if (!poll->head)
5199 return false;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005200 spin_lock(&poll->head->lock);
Jens Axboeb2e720a2021-03-31 09:03:03 -06005201 if (do_cancel)
5202 WRITE_ONCE(poll->canceled, true);
Jens Axboe392edb42019-12-09 17:52:20 -07005203 if (!list_empty(&poll->wait.entry)) {
5204 list_del_init(&poll->wait.entry);
Jens Axboeb41e9852020-02-17 09:52:41 -07005205 do_complete = true;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005206 }
5207 spin_unlock(&poll->head->lock);
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005208 hash_del(&req->hash_node);
Jens Axboed7718a92020-02-14 22:23:12 -07005209 return do_complete;
5210}
5211
Jens Axboeb2c3f7e2021-02-23 08:58:04 -07005212static bool io_poll_remove_waitqs(struct io_kiocb *req)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005213 __must_hold(&req->ctx->completion_lock)
Jens Axboed7718a92020-02-14 22:23:12 -07005214{
5215 bool do_complete;
5216
Jens Axboed4e7cd32020-08-15 11:44:50 -07005217 io_poll_remove_double(req);
Pavel Begunkove31001a2021-04-13 02:58:43 +01005218 do_complete = __io_poll_remove_one(req, io_poll_get_single(req), true);
Jens Axboed4e7cd32020-08-15 11:44:50 -07005219
Pavel Begunkove31001a2021-04-13 02:58:43 +01005220 if (req->opcode != IORING_OP_POLL_ADD && do_complete) {
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005221 struct async_poll *apoll = req->apoll;
5222
Jens Axboed7718a92020-02-14 22:23:12 -07005223 /* non-poll requests have submit ref still */
Pavel Begunkove31001a2021-04-13 02:58:43 +01005224 req_ref_put(req);
5225 kfree(apoll->double_poll);
5226 kfree(apoll);
Xiaoguang Wangb1f573b2020-04-12 14:50:54 +08005227 }
Jens Axboeb2c3f7e2021-02-23 08:58:04 -07005228 return do_complete;
5229}
5230
5231static bool io_poll_remove_one(struct io_kiocb *req)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005232 __must_hold(&req->ctx->completion_lock)
Jens Axboeb2c3f7e2021-02-23 08:58:04 -07005233{
5234 bool do_complete;
5235
5236 do_complete = io_poll_remove_waitqs(req);
Jens Axboeb41e9852020-02-17 09:52:41 -07005237 if (do_complete) {
Pavel Begunkovff6421642021-04-11 01:46:32 +01005238 io_cqring_fill_event(req, -ECANCELED, 0);
Jens Axboeb41e9852020-02-17 09:52:41 -07005239 io_commit_cqring(req->ctx);
Jens Axboef254ac02020-08-12 17:33:30 -06005240 req_set_fail_links(req);
Pavel Begunkov216578e2020-10-13 09:44:00 +01005241 io_put_req_deferred(req, 1);
Jens Axboeb41e9852020-02-17 09:52:41 -07005242 }
5243
5244 return do_complete;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005245}
5246
Jens Axboe76e1b642020-09-26 15:05:03 -06005247/*
5248 * Returns true if we found and killed one or more poll requests
5249 */
Pavel Begunkov6b819282020-11-06 13:00:25 +00005250static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk,
5251 struct files_struct *files)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005252{
Jens Axboe78076bb2019-12-04 19:56:40 -07005253 struct hlist_node *tmp;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005254 struct io_kiocb *req;
Jens Axboe8e2e1fa2020-04-13 17:05:14 -06005255 int posted = 0, i;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005256
5257 spin_lock_irq(&ctx->completion_lock);
Jens Axboe78076bb2019-12-04 19:56:40 -07005258 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
5259 struct hlist_head *list;
5260
5261 list = &ctx->cancel_hash[i];
Jens Axboef3606e32020-09-22 08:18:24 -06005262 hlist_for_each_entry_safe(req, tmp, list, hash_node) {
Pavel Begunkov6b819282020-11-06 13:00:25 +00005263 if (io_match_task(req, tsk, files))
Jens Axboef3606e32020-09-22 08:18:24 -06005264 posted += io_poll_remove_one(req);
5265 }
Jens Axboe221c5eb2019-01-17 09:41:58 -07005266 }
5267 spin_unlock_irq(&ctx->completion_lock);
Jens Axboeb41e9852020-02-17 09:52:41 -07005268
Jens Axboe8e2e1fa2020-04-13 17:05:14 -06005269 if (posted)
5270 io_cqring_ev_posted(ctx);
Jens Axboe76e1b642020-09-26 15:05:03 -06005271
5272 return posted != 0;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005273}
5274
Jens Axboeb2cb8052021-03-17 08:17:19 -06005275static struct io_kiocb *io_poll_find(struct io_ring_ctx *ctx, __u64 sqe_addr)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005276 __must_hold(&ctx->completion_lock)
Jens Axboe47f46762019-11-09 17:43:02 -07005277{
Jens Axboe78076bb2019-12-04 19:56:40 -07005278 struct hlist_head *list;
Jens Axboe47f46762019-11-09 17:43:02 -07005279 struct io_kiocb *req;
5280
Jens Axboe78076bb2019-12-04 19:56:40 -07005281 list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)];
5282 hlist_for_each_entry(req, list, hash_node) {
Jens Axboeb41e9852020-02-17 09:52:41 -07005283 if (sqe_addr != req->user_data)
5284 continue;
Jens Axboeb2cb8052021-03-17 08:17:19 -06005285 return req;
Jens Axboe47f46762019-11-09 17:43:02 -07005286 }
5287
Jens Axboeb2cb8052021-03-17 08:17:19 -06005288 return NULL;
5289}
5290
5291static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005292 __must_hold(&ctx->completion_lock)
Jens Axboeb2cb8052021-03-17 08:17:19 -06005293{
5294 struct io_kiocb *req;
5295
5296 req = io_poll_find(ctx, sqe_addr);
5297 if (!req)
5298 return -ENOENT;
5299 if (io_poll_remove_one(req))
5300 return 0;
5301
5302 return -EALREADY;
Jens Axboe47f46762019-11-09 17:43:02 -07005303}
5304
Jens Axboe3529d8c2019-12-19 18:24:38 -07005305static int io_poll_remove_prep(struct io_kiocb *req,
5306 const struct io_uring_sqe *sqe)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005307{
Jens Axboe221c5eb2019-01-17 09:41:58 -07005308 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5309 return -EINVAL;
5310 if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index ||
5311 sqe->poll_events)
5312 return -EINVAL;
5313
Pavel Begunkov018043b2020-10-27 23:17:18 +00005314 req->poll_remove.addr = READ_ONCE(sqe->addr);
Jens Axboe0969e782019-12-17 18:40:57 -07005315 return 0;
5316}
5317
5318/*
5319 * Find a running poll command that matches one specified in sqe->addr,
5320 * and remove it if found.
5321 */
Pavel Begunkov61e98202021-02-10 00:03:08 +00005322static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe0969e782019-12-17 18:40:57 -07005323{
5324 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe0969e782019-12-17 18:40:57 -07005325 int ret;
5326
Jens Axboe221c5eb2019-01-17 09:41:58 -07005327 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov018043b2020-10-27 23:17:18 +00005328 ret = io_poll_cancel(ctx, req->poll_remove.addr);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005329 spin_unlock_irq(&ctx->completion_lock);
5330
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005331 if (ret < 0)
5332 req_set_fail_links(req);
Pavel Begunkov0bdf3392021-04-11 01:46:29 +01005333 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005334 return 0;
5335}
5336
Jens Axboe221c5eb2019-01-17 09:41:58 -07005337static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
5338 void *key)
5339{
Jens Axboec2f2eb72020-02-10 09:07:05 -07005340 struct io_kiocb *req = wait->private;
5341 struct io_poll_iocb *poll = &req->poll;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005342
Jens Axboed7718a92020-02-14 22:23:12 -07005343 return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005344}
5345
Jens Axboe221c5eb2019-01-17 09:41:58 -07005346static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head,
5347 struct poll_table_struct *p)
5348{
5349 struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
5350
Jens Axboee8c2bc12020-08-15 18:44:09 -07005351 __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data);
Jens Axboeeac406c62019-11-14 12:09:58 -07005352}
5353
Jens Axboe3529d8c2019-12-19 18:24:38 -07005354static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005355{
Jens Axboe88e41cf2021-02-22 22:08:01 -07005356 u32 events, flags;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005357
5358 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5359 return -EINVAL;
Jens Axboeb69de282021-03-17 08:37:41 -06005360 if (sqe->ioprio || sqe->buf_index)
Jens Axboe88e41cf2021-02-22 22:08:01 -07005361 return -EINVAL;
5362 flags = READ_ONCE(sqe->len);
Jens Axboeb69de282021-03-17 08:37:41 -06005363 if (flags & ~(IORING_POLL_ADD_MULTI | IORING_POLL_UPDATE_EVENTS |
5364 IORING_POLL_UPDATE_USER_DATA))
Jens Axboe221c5eb2019-01-17 09:41:58 -07005365 return -EINVAL;
Jiufei Xue5769a352020-06-17 17:53:55 +08005366 events = READ_ONCE(sqe->poll32_events);
5367#ifdef __BIG_ENDIAN
5368 events = swahw32(events);
5369#endif
Jens Axboeb69de282021-03-17 08:37:41 -06005370 if (!(flags & IORING_POLL_ADD_MULTI))
Jens Axboe88e41cf2021-02-22 22:08:01 -07005371 events |= EPOLLONESHOT;
Pavel Begunkov9d805892021-04-13 02:58:40 +01005372 events = demangle_poll(events) |
5373 (events & (EPOLLEXCLUSIVE|EPOLLONESHOT));
Pavel Begunkov66d2d002021-04-13 02:58:39 +01005374
5375 if (flags & (IORING_POLL_UPDATE_EVENTS|IORING_POLL_UPDATE_USER_DATA)) {
Pavel Begunkov9d805892021-04-13 02:58:40 +01005376 struct io_poll_update *poll_upd = &req->poll_update;
5377
5378 req->flags |= REQ_F_POLL_UPDATE;
5379 poll_upd->events = events;
5380 poll_upd->old_user_data = READ_ONCE(sqe->addr);
5381 poll_upd->update_events = flags & IORING_POLL_UPDATE_EVENTS;
5382 poll_upd->update_user_data = flags & IORING_POLL_UPDATE_USER_DATA;
5383 if (poll_upd->update_user_data)
5384 poll_upd->new_user_data = READ_ONCE(sqe->off);
Pavel Begunkov66d2d002021-04-13 02:58:39 +01005385 } else {
Pavel Begunkov9d805892021-04-13 02:58:40 +01005386 struct io_poll_iocb *poll = &req->poll;
5387
5388 poll->events = events;
Pavel Begunkov66d2d002021-04-13 02:58:39 +01005389 if (sqe->off || sqe->addr)
5390 return -EINVAL;
Jens Axboeb69de282021-03-17 08:37:41 -06005391 }
Jens Axboe0969e782019-12-17 18:40:57 -07005392 return 0;
5393}
5394
Jens Axboeb69de282021-03-17 08:37:41 -06005395static int __io_poll_add(struct io_kiocb *req)
Jens Axboe0969e782019-12-17 18:40:57 -07005396{
5397 struct io_poll_iocb *poll = &req->poll;
5398 struct io_ring_ctx *ctx = req->ctx;
5399 struct io_poll_table ipt;
Jens Axboe0969e782019-12-17 18:40:57 -07005400 __poll_t mask;
Jens Axboe0969e782019-12-17 18:40:57 -07005401
Jens Axboed7718a92020-02-14 22:23:12 -07005402 ipt.pt._qproc = io_poll_queue_proc;
Jens Axboe36703242019-07-25 10:20:18 -06005403
Jens Axboed7718a92020-02-14 22:23:12 -07005404 mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events,
5405 io_poll_wake);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005406
Jens Axboe8c838782019-03-12 15:48:16 -06005407 if (mask) { /* no async, we'd stolen it */
Jens Axboe8c838782019-03-12 15:48:16 -06005408 ipt.error = 0;
Pavel Begunkove27414b2021-04-09 09:13:20 +01005409 io_poll_complete(req, mask);
Jens Axboe8c838782019-03-12 15:48:16 -06005410 }
Jens Axboe221c5eb2019-01-17 09:41:58 -07005411 spin_unlock_irq(&ctx->completion_lock);
5412
Jens Axboe8c838782019-03-12 15:48:16 -06005413 if (mask) {
5414 io_cqring_ev_posted(ctx);
Jens Axboe88e41cf2021-02-22 22:08:01 -07005415 if (poll->events & EPOLLONESHOT)
5416 io_put_req(req);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005417 }
Jens Axboe8c838782019-03-12 15:48:16 -06005418 return ipt.error;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005419}
5420
Jens Axboeb69de282021-03-17 08:37:41 -06005421static int io_poll_update(struct io_kiocb *req)
5422{
5423 struct io_ring_ctx *ctx = req->ctx;
5424 struct io_kiocb *preq;
Jens Axboecb3b200e2021-04-06 09:49:31 -06005425 bool completing;
Jens Axboeb69de282021-03-17 08:37:41 -06005426 int ret;
5427
5428 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov9d805892021-04-13 02:58:40 +01005429 preq = io_poll_find(ctx, req->poll_update.old_user_data);
Jens Axboeb69de282021-03-17 08:37:41 -06005430 if (!preq) {
5431 ret = -ENOENT;
5432 goto err;
5433 } else if (preq->opcode != IORING_OP_POLL_ADD) {
5434 /* don't allow internal poll updates */
5435 ret = -EACCES;
5436 goto err;
5437 }
Jens Axboecb3b200e2021-04-06 09:49:31 -06005438
5439 /*
5440 * Don't allow racy completion with singleshot, as we cannot safely
5441 * update those. For multishot, if we're racing with completion, just
5442 * let completion re-add it.
5443 */
5444 completing = !__io_poll_remove_one(preq, &preq->poll, false);
5445 if (completing && (preq->poll.events & EPOLLONESHOT)) {
5446 ret = -EALREADY;
5447 goto err;
Jens Axboeb69de282021-03-17 08:37:41 -06005448 }
5449 /* we now have a detached poll request. reissue. */
5450 ret = 0;
5451err:
Jens Axboeb69de282021-03-17 08:37:41 -06005452 if (ret < 0) {
Jens Axboecb3b200e2021-04-06 09:49:31 -06005453 spin_unlock_irq(&ctx->completion_lock);
Jens Axboeb69de282021-03-17 08:37:41 -06005454 req_set_fail_links(req);
5455 io_req_complete(req, ret);
5456 return 0;
5457 }
5458 /* only mask one event flags, keep behavior flags */
Pavel Begunkov9d805892021-04-13 02:58:40 +01005459 if (req->poll_update.update_events) {
Jens Axboeb69de282021-03-17 08:37:41 -06005460 preq->poll.events &= ~0xffff;
Pavel Begunkov9d805892021-04-13 02:58:40 +01005461 preq->poll.events |= req->poll_update.events & 0xffff;
Jens Axboeb69de282021-03-17 08:37:41 -06005462 preq->poll.events |= IO_POLL_UNMASK;
5463 }
Pavel Begunkov9d805892021-04-13 02:58:40 +01005464 if (req->poll_update.update_user_data)
5465 preq->user_data = req->poll_update.new_user_data;
Jens Axboeb69de282021-03-17 08:37:41 -06005466
Jens Axboecb3b200e2021-04-06 09:49:31 -06005467 spin_unlock_irq(&ctx->completion_lock);
5468
Jens Axboeb69de282021-03-17 08:37:41 -06005469 /* complete update request, we're done with it */
5470 io_req_complete(req, ret);
5471
Jens Axboecb3b200e2021-04-06 09:49:31 -06005472 if (!completing) {
5473 ret = __io_poll_add(preq);
5474 if (ret < 0) {
5475 req_set_fail_links(preq);
5476 io_req_complete(preq, ret);
5477 }
Jens Axboeb69de282021-03-17 08:37:41 -06005478 }
5479 return 0;
5480}
5481
5482static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags)
5483{
Pavel Begunkov9d805892021-04-13 02:58:40 +01005484 if (!(req->flags & REQ_F_POLL_UPDATE))
Jens Axboeb69de282021-03-17 08:37:41 -06005485 return __io_poll_add(req);
5486 return io_poll_update(req);
5487}
5488
Jens Axboe5262f562019-09-17 12:26:57 -06005489static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer)
5490{
Jens Axboead8a48a2019-11-15 08:49:11 -07005491 struct io_timeout_data *data = container_of(timer,
5492 struct io_timeout_data, timer);
5493 struct io_kiocb *req = data->req;
5494 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5262f562019-09-17 12:26:57 -06005495 unsigned long flags;
5496
Jens Axboe5262f562019-09-17 12:26:57 -06005497 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkova71976f2020-10-10 18:34:11 +01005498 list_del_init(&req->timeout.list);
Pavel Begunkov01cec8c2020-07-30 18:43:50 +03005499 atomic_set(&req->ctx->cq_timeouts,
5500 atomic_read(&req->ctx->cq_timeouts) + 1);
5501
Pavel Begunkovff6421642021-04-11 01:46:32 +01005502 io_cqring_fill_event(req, -ETIME, 0);
Jens Axboe5262f562019-09-17 12:26:57 -06005503 io_commit_cqring(ctx);
5504 spin_unlock_irqrestore(&ctx->completion_lock, flags);
5505
5506 io_cqring_ev_posted(ctx);
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005507 req_set_fail_links(req);
Jens Axboe5262f562019-09-17 12:26:57 -06005508 io_put_req(req);
5509 return HRTIMER_NORESTART;
5510}
5511
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005512static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx,
5513 __u64 user_data)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005514 __must_hold(&ctx->completion_lock)
Jens Axboe47f46762019-11-09 17:43:02 -07005515{
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005516 struct io_timeout_data *io;
Jens Axboef254ac02020-08-12 17:33:30 -06005517 struct io_kiocb *req;
Pavel Begunkovfd9c7bc2021-04-13 02:58:42 +01005518 bool found = false;
Jens Axboef254ac02020-08-12 17:33:30 -06005519
5520 list_for_each_entry(req, &ctx->timeout_list, timeout.list) {
Pavel Begunkovfd9c7bc2021-04-13 02:58:42 +01005521 found = user_data == req->user_data;
5522 if (found)
Jens Axboef254ac02020-08-12 17:33:30 -06005523 break;
Jens Axboef254ac02020-08-12 17:33:30 -06005524 }
Pavel Begunkovfd9c7bc2021-04-13 02:58:42 +01005525 if (!found)
5526 return ERR_PTR(-ENOENT);
Jens Axboef254ac02020-08-12 17:33:30 -06005527
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005528 io = req->async_data;
Pavel Begunkovfd9c7bc2021-04-13 02:58:42 +01005529 if (hrtimer_try_to_cancel(&io->timer) == -1)
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005530 return ERR_PTR(-EALREADY);
5531 list_del_init(&req->timeout.list);
5532 return req;
5533}
5534
5535static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005536 __must_hold(&ctx->completion_lock)
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005537{
5538 struct io_kiocb *req = io_timeout_extract(ctx, user_data);
5539
5540 if (IS_ERR(req))
5541 return PTR_ERR(req);
5542
5543 req_set_fail_links(req);
Pavel Begunkovff6421642021-04-11 01:46:32 +01005544 io_cqring_fill_event(req, -ECANCELED, 0);
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005545 io_put_req_deferred(req, 1);
5546 return 0;
Jens Axboef254ac02020-08-12 17:33:30 -06005547}
5548
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005549static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data,
5550 struct timespec64 *ts, enum hrtimer_mode mode)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005551 __must_hold(&ctx->completion_lock)
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005552{
5553 struct io_kiocb *req = io_timeout_extract(ctx, user_data);
5554 struct io_timeout_data *data;
5555
5556 if (IS_ERR(req))
5557 return PTR_ERR(req);
5558
5559 req->timeout.off = 0; /* noseq */
5560 data = req->async_data;
5561 list_add_tail(&req->timeout.list, &ctx->timeout_list);
5562 hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode);
5563 data->timer.function = io_timeout_fn;
5564 hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode);
5565 return 0;
Jens Axboe47f46762019-11-09 17:43:02 -07005566}
5567
Jens Axboe3529d8c2019-12-19 18:24:38 -07005568static int io_timeout_remove_prep(struct io_kiocb *req,
5569 const struct io_uring_sqe *sqe)
Jens Axboeb29472e2019-12-17 18:50:29 -07005570{
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005571 struct io_timeout_rem *tr = &req->timeout_rem;
5572
Jens Axboeb29472e2019-12-17 18:50:29 -07005573 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5574 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005575 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5576 return -EINVAL;
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005577 if (sqe->ioprio || sqe->buf_index || sqe->len)
Jens Axboeb29472e2019-12-17 18:50:29 -07005578 return -EINVAL;
5579
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005580 tr->addr = READ_ONCE(sqe->addr);
5581 tr->flags = READ_ONCE(sqe->timeout_flags);
5582 if (tr->flags & IORING_TIMEOUT_UPDATE) {
5583 if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS))
5584 return -EINVAL;
5585 if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2)))
5586 return -EFAULT;
5587 } else if (tr->flags) {
5588 /* timeout removal doesn't support flags */
5589 return -EINVAL;
5590 }
5591
Jens Axboeb29472e2019-12-17 18:50:29 -07005592 return 0;
5593}
5594
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005595static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags)
5596{
5597 return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS
5598 : HRTIMER_MODE_REL;
5599}
5600
Jens Axboe11365042019-10-16 09:08:32 -06005601/*
5602 * Remove or update an existing timeout command
5603 */
Pavel Begunkov61e98202021-02-10 00:03:08 +00005604static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe11365042019-10-16 09:08:32 -06005605{
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005606 struct io_timeout_rem *tr = &req->timeout_rem;
Jens Axboe11365042019-10-16 09:08:32 -06005607 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe47f46762019-11-09 17:43:02 -07005608 int ret;
Jens Axboe11365042019-10-16 09:08:32 -06005609
Jens Axboe11365042019-10-16 09:08:32 -06005610 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005611 if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE))
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005612 ret = io_timeout_cancel(ctx, tr->addr);
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005613 else
5614 ret = io_timeout_update(ctx, tr->addr, &tr->ts,
5615 io_translate_timeout_mode(tr->flags));
Jens Axboe11365042019-10-16 09:08:32 -06005616
Pavel Begunkovff6421642021-04-11 01:46:32 +01005617 io_cqring_fill_event(req, ret, 0);
Jens Axboe11365042019-10-16 09:08:32 -06005618 io_commit_cqring(ctx);
5619 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe5262f562019-09-17 12:26:57 -06005620 io_cqring_ev_posted(ctx);
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005621 if (ret < 0)
5622 req_set_fail_links(req);
Jackie Liuec9c02a2019-11-08 23:50:36 +08005623 io_put_req(req);
Jens Axboe11365042019-10-16 09:08:32 -06005624 return 0;
Jens Axboe5262f562019-09-17 12:26:57 -06005625}
5626
Jens Axboe3529d8c2019-12-19 18:24:38 -07005627static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe,
Jens Axboe2d283902019-12-04 11:08:05 -07005628 bool is_timeout_link)
Jens Axboe5262f562019-09-17 12:26:57 -06005629{
Jens Axboead8a48a2019-11-15 08:49:11 -07005630 struct io_timeout_data *data;
Jens Axboea41525a2019-10-15 16:48:15 -06005631 unsigned flags;
Pavel Begunkov56080b02020-05-26 20:34:04 +03005632 u32 off = READ_ONCE(sqe->off);
Jens Axboe5262f562019-09-17 12:26:57 -06005633
Jens Axboead8a48a2019-11-15 08:49:11 -07005634 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe5262f562019-09-17 12:26:57 -06005635 return -EINVAL;
Jens Axboead8a48a2019-11-15 08:49:11 -07005636 if (sqe->ioprio || sqe->buf_index || sqe->len != 1)
Jens Axboea41525a2019-10-15 16:48:15 -06005637 return -EINVAL;
Pavel Begunkov56080b02020-05-26 20:34:04 +03005638 if (off && is_timeout_link)
Jens Axboe2d283902019-12-04 11:08:05 -07005639 return -EINVAL;
Jens Axboea41525a2019-10-15 16:48:15 -06005640 flags = READ_ONCE(sqe->timeout_flags);
5641 if (flags & ~IORING_TIMEOUT_ABS)
Jens Axboe5262f562019-09-17 12:26:57 -06005642 return -EINVAL;
Arnd Bergmannbdf20072019-10-01 09:53:29 -06005643
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005644 req->timeout.off = off;
Jens Axboe26a61672019-12-20 09:02:01 -07005645
Jens Axboee8c2bc12020-08-15 18:44:09 -07005646 if (!req->async_data && io_alloc_async_data(req))
Jens Axboe26a61672019-12-20 09:02:01 -07005647 return -ENOMEM;
5648
Jens Axboee8c2bc12020-08-15 18:44:09 -07005649 data = req->async_data;
Jens Axboead8a48a2019-11-15 08:49:11 -07005650 data->req = req;
Jens Axboead8a48a2019-11-15 08:49:11 -07005651
5652 if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr)))
Jens Axboe5262f562019-09-17 12:26:57 -06005653 return -EFAULT;
5654
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005655 data->mode = io_translate_timeout_mode(flags);
Jens Axboead8a48a2019-11-15 08:49:11 -07005656 hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode);
Pavel Begunkov2482b582021-03-25 18:32:44 +00005657 if (is_timeout_link)
5658 io_req_track_inflight(req);
Jens Axboead8a48a2019-11-15 08:49:11 -07005659 return 0;
5660}
5661
Pavel Begunkov61e98202021-02-10 00:03:08 +00005662static int io_timeout(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboead8a48a2019-11-15 08:49:11 -07005663{
Jens Axboead8a48a2019-11-15 08:49:11 -07005664 struct io_ring_ctx *ctx = req->ctx;
Jens Axboee8c2bc12020-08-15 18:44:09 -07005665 struct io_timeout_data *data = req->async_data;
Jens Axboead8a48a2019-11-15 08:49:11 -07005666 struct list_head *entry;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005667 u32 tail, off = req->timeout.off;
Jens Axboead8a48a2019-11-15 08:49:11 -07005668
Pavel Begunkov733f5c92020-05-26 20:34:03 +03005669 spin_lock_irq(&ctx->completion_lock);
Jens Axboe93bd25b2019-11-11 23:34:31 -07005670
Jens Axboe5262f562019-09-17 12:26:57 -06005671 /*
5672 * sqe->off holds how many events that need to occur for this
Jens Axboe93bd25b2019-11-11 23:34:31 -07005673 * timeout event to be satisfied. If it isn't set, then this is
5674 * a pure timeout request, sequence isn't used.
Jens Axboe5262f562019-09-17 12:26:57 -06005675 */
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03005676 if (io_is_timeout_noseq(req)) {
Jens Axboe93bd25b2019-11-11 23:34:31 -07005677 entry = ctx->timeout_list.prev;
5678 goto add;
5679 }
Jens Axboe5262f562019-09-17 12:26:57 -06005680
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005681 tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
5682 req->timeout.target_seq = tail + off;
Jens Axboe5262f562019-09-17 12:26:57 -06005683
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05005684 /* Update the last seq here in case io_flush_timeouts() hasn't.
5685 * This is safe because ->completion_lock is held, and submissions
5686 * and completions are never mixed in the same ->completion_lock section.
5687 */
5688 ctx->cq_last_tm_flush = tail;
5689
Jens Axboe5262f562019-09-17 12:26:57 -06005690 /*
5691 * Insertion sort, ensuring the first entry in the list is always
5692 * the one we need first.
5693 */
Jens Axboe5262f562019-09-17 12:26:57 -06005694 list_for_each_prev(entry, &ctx->timeout_list) {
Pavel Begunkov135fcde2020-07-13 23:37:12 +03005695 struct io_kiocb *nxt = list_entry(entry, struct io_kiocb,
5696 timeout.list);
Jens Axboe5262f562019-09-17 12:26:57 -06005697
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03005698 if (io_is_timeout_noseq(nxt))
Jens Axboe93bd25b2019-11-11 23:34:31 -07005699 continue;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005700 /* nxt.seq is behind @tail, otherwise would've been completed */
5701 if (off >= nxt->timeout.target_seq - tail)
Jens Axboe5262f562019-09-17 12:26:57 -06005702 break;
5703 }
Jens Axboe93bd25b2019-11-11 23:34:31 -07005704add:
Pavel Begunkov135fcde2020-07-13 23:37:12 +03005705 list_add(&req->timeout.list, entry);
Jens Axboead8a48a2019-11-15 08:49:11 -07005706 data->timer.function = io_timeout_fn;
5707 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode);
Jens Axboe842f9612019-10-29 12:34:10 -06005708 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe5262f562019-09-17 12:26:57 -06005709 return 0;
5710}
5711
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005712struct io_cancel_data {
5713 struct io_ring_ctx *ctx;
5714 u64 user_data;
5715};
5716
Jens Axboe62755e32019-10-28 21:49:21 -06005717static bool io_cancel_cb(struct io_wq_work *work, void *data)
Jens Axboede0617e2019-04-06 21:51:27 -06005718{
Jens Axboe62755e32019-10-28 21:49:21 -06005719 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005720 struct io_cancel_data *cd = data;
Jens Axboede0617e2019-04-06 21:51:27 -06005721
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005722 return req->ctx == cd->ctx && req->user_data == cd->user_data;
Jens Axboe62755e32019-10-28 21:49:21 -06005723}
5724
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005725static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data,
5726 struct io_ring_ctx *ctx)
Jens Axboe62755e32019-10-28 21:49:21 -06005727{
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005728 struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, };
Jens Axboe62755e32019-10-28 21:49:21 -06005729 enum io_wq_cancel cancel_ret;
Jens Axboe62755e32019-10-28 21:49:21 -06005730 int ret = 0;
5731
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005732 if (!tctx || !tctx->io_wq)
Jens Axboe5aa75ed2021-02-16 12:56:50 -07005733 return -ENOENT;
5734
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005735 cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false);
Jens Axboe62755e32019-10-28 21:49:21 -06005736 switch (cancel_ret) {
5737 case IO_WQ_CANCEL_OK:
5738 ret = 0;
5739 break;
5740 case IO_WQ_CANCEL_RUNNING:
5741 ret = -EALREADY;
5742 break;
5743 case IO_WQ_CANCEL_NOTFOUND:
5744 ret = -ENOENT;
5745 break;
5746 }
5747
Jens Axboee977d6d2019-11-05 12:39:45 -07005748 return ret;
5749}
5750
Jens Axboe47f46762019-11-09 17:43:02 -07005751static void io_async_find_and_cancel(struct io_ring_ctx *ctx,
5752 struct io_kiocb *req, __u64 sqe_addr,
Pavel Begunkov014db002020-03-03 21:33:12 +03005753 int success_ret)
Jens Axboe47f46762019-11-09 17:43:02 -07005754{
5755 unsigned long flags;
5756 int ret;
5757
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005758 ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx);
Jens Axboe47f46762019-11-09 17:43:02 -07005759 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkovdf9727a2021-04-01 15:43:59 +01005760 if (ret != -ENOENT)
5761 goto done;
Jens Axboe47f46762019-11-09 17:43:02 -07005762 ret = io_timeout_cancel(ctx, sqe_addr);
5763 if (ret != -ENOENT)
5764 goto done;
5765 ret = io_poll_cancel(ctx, sqe_addr);
5766done:
Jens Axboeb0dd8a42019-11-18 12:14:54 -07005767 if (!ret)
5768 ret = success_ret;
Pavel Begunkovff6421642021-04-11 01:46:32 +01005769 io_cqring_fill_event(req, ret, 0);
Jens Axboe47f46762019-11-09 17:43:02 -07005770 io_commit_cqring(ctx);
5771 spin_unlock_irqrestore(&ctx->completion_lock, flags);
5772 io_cqring_ev_posted(ctx);
5773
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005774 if (ret < 0)
5775 req_set_fail_links(req);
Jens Axboe47f46762019-11-09 17:43:02 -07005776}
5777
Jens Axboe3529d8c2019-12-19 18:24:38 -07005778static int io_async_cancel_prep(struct io_kiocb *req,
5779 const struct io_uring_sqe *sqe)
Jens Axboee977d6d2019-11-05 12:39:45 -07005780{
Jens Axboefbf23842019-12-17 18:45:56 -07005781 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboee977d6d2019-11-05 12:39:45 -07005782 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005783 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5784 return -EINVAL;
5785 if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags)
Jens Axboee977d6d2019-11-05 12:39:45 -07005786 return -EINVAL;
5787
Jens Axboefbf23842019-12-17 18:45:56 -07005788 req->cancel.addr = READ_ONCE(sqe->addr);
5789 return 0;
5790}
5791
Pavel Begunkov61e98202021-02-10 00:03:08 +00005792static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboefbf23842019-12-17 18:45:56 -07005793{
5794 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkov58f99372021-03-12 16:25:55 +00005795 u64 sqe_addr = req->cancel.addr;
5796 struct io_tctx_node *node;
5797 int ret;
Jens Axboefbf23842019-12-17 18:45:56 -07005798
Pavel Begunkov58f99372021-03-12 16:25:55 +00005799 /* tasks should wait for their io-wq threads, so safe w/o sync */
5800 ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx);
5801 spin_lock_irq(&ctx->completion_lock);
5802 if (ret != -ENOENT)
5803 goto done;
5804 ret = io_timeout_cancel(ctx, sqe_addr);
5805 if (ret != -ENOENT)
5806 goto done;
5807 ret = io_poll_cancel(ctx, sqe_addr);
5808 if (ret != -ENOENT)
5809 goto done;
5810 spin_unlock_irq(&ctx->completion_lock);
5811
5812 /* slow path, try all io-wq's */
5813 io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK));
5814 ret = -ENOENT;
5815 list_for_each_entry(node, &ctx->tctx_list, ctx_node) {
5816 struct io_uring_task *tctx = node->task->io_uring;
5817
Pavel Begunkov58f99372021-03-12 16:25:55 +00005818 ret = io_async_cancel_one(tctx, req->cancel.addr, ctx);
5819 if (ret != -ENOENT)
5820 break;
5821 }
5822 io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK));
5823
5824 spin_lock_irq(&ctx->completion_lock);
5825done:
Pavel Begunkovff6421642021-04-11 01:46:32 +01005826 io_cqring_fill_event(req, ret, 0);
Pavel Begunkov58f99372021-03-12 16:25:55 +00005827 io_commit_cqring(ctx);
5828 spin_unlock_irq(&ctx->completion_lock);
5829 io_cqring_ev_posted(ctx);
5830
5831 if (ret < 0)
5832 req_set_fail_links(req);
5833 io_put_req(req);
Jens Axboe62755e32019-10-28 21:49:21 -06005834 return 0;
5835}
5836
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005837static int io_rsrc_update_prep(struct io_kiocb *req,
Jens Axboe05f3fb32019-12-09 11:22:50 -07005838 const struct io_uring_sqe *sqe)
5839{
Jens Axboe6ca56f82020-09-18 16:51:19 -06005840 if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL))
5841 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005842 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5843 return -EINVAL;
5844 if (sqe->ioprio || sqe->rw_flags)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005845 return -EINVAL;
5846
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005847 req->rsrc_update.offset = READ_ONCE(sqe->off);
5848 req->rsrc_update.nr_args = READ_ONCE(sqe->len);
5849 if (!req->rsrc_update.nr_args)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005850 return -EINVAL;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005851 req->rsrc_update.arg = READ_ONCE(sqe->addr);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005852 return 0;
5853}
5854
Pavel Begunkov889fca72021-02-10 00:03:09 +00005855static int io_files_update(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005856{
5857 struct io_ring_ctx *ctx = req->ctx;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005858 struct io_uring_rsrc_update up;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005859 int ret;
5860
Pavel Begunkov45d189c2021-02-10 00:03:07 +00005861 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005862 return -EAGAIN;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005863
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005864 up.offset = req->rsrc_update.offset;
5865 up.data = req->rsrc_update.arg;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005866
5867 mutex_lock(&ctx->uring_lock);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005868 ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005869 mutex_unlock(&ctx->uring_lock);
5870
5871 if (ret < 0)
5872 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00005873 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005874 return 0;
5875}
5876
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005877static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07005878{
Jens Axboed625c6e2019-12-17 19:53:05 -07005879 switch (req->opcode) {
Jens Axboee7815732019-12-17 19:45:06 -07005880 case IORING_OP_NOP:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005881 return 0;
Jens Axboef67676d2019-12-02 11:03:47 -07005882 case IORING_OP_READV:
5883 case IORING_OP_READ_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005884 case IORING_OP_READ:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005885 return io_read_prep(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07005886 case IORING_OP_WRITEV:
5887 case IORING_OP_WRITE_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005888 case IORING_OP_WRITE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005889 return io_write_prep(req, sqe);
Jens Axboe0969e782019-12-17 18:40:57 -07005890 case IORING_OP_POLL_ADD:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005891 return io_poll_add_prep(req, sqe);
Jens Axboe0969e782019-12-17 18:40:57 -07005892 case IORING_OP_POLL_REMOVE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005893 return io_poll_remove_prep(req, sqe);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005894 case IORING_OP_FSYNC:
Pavel Begunkov1155c762021-02-18 18:29:38 +00005895 return io_fsync_prep(req, sqe);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005896 case IORING_OP_SYNC_FILE_RANGE:
Pavel Begunkov1155c762021-02-18 18:29:38 +00005897 return io_sfr_prep(req, sqe);
Jens Axboe03b12302019-12-02 18:50:25 -07005898 case IORING_OP_SENDMSG:
Jens Axboefddafac2020-01-04 20:19:44 -07005899 case IORING_OP_SEND:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005900 return io_sendmsg_prep(req, sqe);
Jens Axboe03b12302019-12-02 18:50:25 -07005901 case IORING_OP_RECVMSG:
Jens Axboefddafac2020-01-04 20:19:44 -07005902 case IORING_OP_RECV:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005903 return io_recvmsg_prep(req, sqe);
Jens Axboef499a022019-12-02 16:28:46 -07005904 case IORING_OP_CONNECT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005905 return io_connect_prep(req, sqe);
Jens Axboe2d283902019-12-04 11:08:05 -07005906 case IORING_OP_TIMEOUT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005907 return io_timeout_prep(req, sqe, false);
Jens Axboeb29472e2019-12-17 18:50:29 -07005908 case IORING_OP_TIMEOUT_REMOVE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005909 return io_timeout_remove_prep(req, sqe);
Jens Axboefbf23842019-12-17 18:45:56 -07005910 case IORING_OP_ASYNC_CANCEL:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005911 return io_async_cancel_prep(req, sqe);
Jens Axboe2d283902019-12-04 11:08:05 -07005912 case IORING_OP_LINK_TIMEOUT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005913 return io_timeout_prep(req, sqe, true);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005914 case IORING_OP_ACCEPT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005915 return io_accept_prep(req, sqe);
Jens Axboed63d1b52019-12-10 10:38:56 -07005916 case IORING_OP_FALLOCATE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005917 return io_fallocate_prep(req, sqe);
Jens Axboe15b71ab2019-12-11 11:20:36 -07005918 case IORING_OP_OPENAT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005919 return io_openat_prep(req, sqe);
Jens Axboeb5dba592019-12-11 14:02:38 -07005920 case IORING_OP_CLOSE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005921 return io_close_prep(req, sqe);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005922 case IORING_OP_FILES_UPDATE:
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005923 return io_rsrc_update_prep(req, sqe);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07005924 case IORING_OP_STATX:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005925 return io_statx_prep(req, sqe);
Jens Axboe4840e412019-12-25 22:03:45 -07005926 case IORING_OP_FADVISE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005927 return io_fadvise_prep(req, sqe);
Jens Axboec1ca7572019-12-25 22:18:28 -07005928 case IORING_OP_MADVISE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005929 return io_madvise_prep(req, sqe);
Jens Axboecebdb982020-01-08 17:59:24 -07005930 case IORING_OP_OPENAT2:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005931 return io_openat2_prep(req, sqe);
Jens Axboe3e4827b2020-01-08 15:18:09 -07005932 case IORING_OP_EPOLL_CTL:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005933 return io_epoll_ctl_prep(req, sqe);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03005934 case IORING_OP_SPLICE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005935 return io_splice_prep(req, sqe);
Jens Axboeddf0322d2020-02-23 16:41:33 -07005936 case IORING_OP_PROVIDE_BUFFERS:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005937 return io_provide_buffers_prep(req, sqe);
Jens Axboe067524e2020-03-02 16:32:28 -07005938 case IORING_OP_REMOVE_BUFFERS:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005939 return io_remove_buffers_prep(req, sqe);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03005940 case IORING_OP_TEE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005941 return io_tee_prep(req, sqe);
Jens Axboe36f4fa62020-09-05 11:14:22 -06005942 case IORING_OP_SHUTDOWN:
5943 return io_shutdown_prep(req, sqe);
Jens Axboe80a261f2020-09-28 14:23:58 -06005944 case IORING_OP_RENAMEAT:
5945 return io_renameat_prep(req, sqe);
Jens Axboe14a11432020-09-28 14:27:37 -06005946 case IORING_OP_UNLINKAT:
5947 return io_unlinkat_prep(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07005948 }
5949
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005950 printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n",
5951 req->opcode);
5952 return-EINVAL;
5953}
5954
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005955static int io_req_prep_async(struct io_kiocb *req)
Jens Axboedef596e2019-01-09 08:59:42 -07005956{
Pavel Begunkovb7e298d2021-02-28 22:35:19 +00005957 if (!io_op_defs[req->opcode].needs_async_setup)
5958 return 0;
5959 if (WARN_ON_ONCE(req->async_data))
5960 return -EFAULT;
5961 if (io_alloc_async_data(req))
5962 return -EAGAIN;
5963
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005964 switch (req->opcode) {
5965 case IORING_OP_READV:
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005966 return io_rw_prep_async(req, READ);
5967 case IORING_OP_WRITEV:
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005968 return io_rw_prep_async(req, WRITE);
5969 case IORING_OP_SENDMSG:
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005970 return io_sendmsg_prep_async(req);
5971 case IORING_OP_RECVMSG:
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005972 return io_recvmsg_prep_async(req);
5973 case IORING_OP_CONNECT:
5974 return io_connect_prep_async(req);
5975 }
Pavel Begunkovb7e298d2021-02-28 22:35:19 +00005976 printk_once(KERN_WARNING "io_uring: prep_async() bad opcode %d\n",
5977 req->opcode);
5978 return -EFAULT;
Jens Axboedef596e2019-01-09 08:59:42 -07005979}
5980
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005981static u32 io_get_sequence(struct io_kiocb *req)
5982{
5983 struct io_kiocb *pos;
5984 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00005985 u32 total_submitted, nr_reqs = 0;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005986
Pavel Begunkovf2f87372020-10-27 23:25:37 +00005987 io_for_each_link(pos, req)
5988 nr_reqs++;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005989
5990 total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped;
5991 return total_submitted - nr_reqs;
5992}
5993
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005994static int io_req_defer(struct io_kiocb *req)
Jens Axboedef596e2019-01-09 08:59:42 -07005995{
5996 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005997 struct io_defer_entry *de;
Jens Axboedef596e2019-01-09 08:59:42 -07005998 int ret;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005999 u32 seq;
Jens Axboedef596e2019-01-09 08:59:42 -07006000
6001 /* Still need defer if there is pending req in defer list. */
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03006002 if (likely(list_empty_careful(&ctx->defer_list) &&
6003 !(req->flags & REQ_F_IO_DRAIN)))
6004 return 0;
6005
6006 seq = io_get_sequence(req);
6007 /* Still a chance to pass the sequence check */
6008 if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list))
Jens Axboedef596e2019-01-09 08:59:42 -07006009 return 0;
6010
Pavel Begunkovb7e298d2021-02-28 22:35:19 +00006011 ret = io_req_prep_async(req);
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006012 if (ret)
6013 return ret;
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03006014 io_prep_async_link(req);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03006015 de = kmalloc(sizeof(*de), GFP_KERNEL);
6016 if (!de)
6017 return -ENOMEM;
Jens Axboe31b51512019-01-18 22:56:34 -07006018
6019 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03006020 if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) {
Jens Axboe31b51512019-01-18 22:56:34 -07006021 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03006022 kfree(de);
Pavel Begunkovae348172020-07-23 20:25:20 +03006023 io_queue_async_work(req);
6024 return -EIOCBQUEUED;
Jens Axboe31b51512019-01-18 22:56:34 -07006025 }
6026
6027 trace_io_uring_defer(ctx, req, req->user_data);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03006028 de->req = req;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03006029 de->seq = seq;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03006030 list_add_tail(&de->list, &ctx->defer_list);
Jens Axboe31b51512019-01-18 22:56:34 -07006031 spin_unlock_irq(&ctx->completion_lock);
6032 return -EIOCBQUEUED;
6033}
6034
Pavel Begunkov68fb8972021-03-19 17:22:41 +00006035static void io_clean_op(struct io_kiocb *req)
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03006036{
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006037 if (req->flags & REQ_F_BUFFER_SELECTED) {
6038 switch (req->opcode) {
6039 case IORING_OP_READV:
6040 case IORING_OP_READ_FIXED:
6041 case IORING_OP_READ:
Jens Axboebcda7ba2020-02-23 16:42:51 -07006042 kfree((void *)(unsigned long)req->rw.addr);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006043 break;
6044 case IORING_OP_RECVMSG:
6045 case IORING_OP_RECV:
Jens Axboe52de1fe2020-02-27 10:15:42 -07006046 kfree(req->sr_msg.kbuf);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006047 break;
6048 }
6049 req->flags &= ~REQ_F_BUFFER_SELECTED;
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03006050 }
6051
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006052 if (req->flags & REQ_F_NEED_CLEANUP) {
6053 switch (req->opcode) {
6054 case IORING_OP_READV:
6055 case IORING_OP_READ_FIXED:
6056 case IORING_OP_READ:
6057 case IORING_OP_WRITEV:
6058 case IORING_OP_WRITE_FIXED:
Jens Axboee8c2bc12020-08-15 18:44:09 -07006059 case IORING_OP_WRITE: {
6060 struct io_async_rw *io = req->async_data;
6061 if (io->free_iovec)
6062 kfree(io->free_iovec);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006063 break;
Jens Axboee8c2bc12020-08-15 18:44:09 -07006064 }
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006065 case IORING_OP_RECVMSG:
Jens Axboee8c2bc12020-08-15 18:44:09 -07006066 case IORING_OP_SENDMSG: {
6067 struct io_async_msghdr *io = req->async_data;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00006068
6069 kfree(io->free_iov);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006070 break;
Jens Axboee8c2bc12020-08-15 18:44:09 -07006071 }
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006072 case IORING_OP_SPLICE:
6073 case IORING_OP_TEE:
Pavel Begunkove1d767f2021-03-19 17:22:43 +00006074 if (!(req->splice.flags & SPLICE_F_FD_IN_FIXED))
6075 io_put_file(req->splice.file_in);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006076 break;
Jens Axboef3cd48502020-09-24 14:55:54 -06006077 case IORING_OP_OPENAT:
6078 case IORING_OP_OPENAT2:
6079 if (req->open.filename)
6080 putname(req->open.filename);
6081 break;
Jens Axboe80a261f2020-09-28 14:23:58 -06006082 case IORING_OP_RENAMEAT:
6083 putname(req->rename.oldpath);
6084 putname(req->rename.newpath);
6085 break;
Jens Axboe14a11432020-09-28 14:27:37 -06006086 case IORING_OP_UNLINKAT:
6087 putname(req->unlink.filename);
6088 break;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006089 }
6090 req->flags &= ~REQ_F_NEED_CLEANUP;
6091 }
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03006092}
6093
Pavel Begunkov889fca72021-02-10 00:03:09 +00006094static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeedafcce2019-01-09 09:16:05 -07006095{
Jens Axboeedafcce2019-01-09 09:16:05 -07006096 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5730b272021-02-27 15:57:30 -07006097 const struct cred *creds = NULL;
Jens Axboed625c6e2019-12-17 19:53:05 -07006098 int ret;
Jens Axboeedafcce2019-01-09 09:16:05 -07006099
Jens Axboe003e8dc2021-03-06 09:22:27 -07006100 if (req->work.creds && req->work.creds != current_cred())
6101 creds = override_creds(req->work.creds);
Jens Axboe5730b272021-02-27 15:57:30 -07006102
Jens Axboed625c6e2019-12-17 19:53:05 -07006103 switch (req->opcode) {
Jens Axboe2b188cc2019-01-07 10:46:33 -07006104 case IORING_OP_NOP:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006105 ret = io_nop(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006106 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006107 case IORING_OP_READV:
Jens Axboe3529d8c2019-12-19 18:24:38 -07006108 case IORING_OP_READ_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07006109 case IORING_OP_READ:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006110 ret = io_read(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006111 break;
6112 case IORING_OP_WRITEV:
Jens Axboe2b188cc2019-01-07 10:46:33 -07006113 case IORING_OP_WRITE_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07006114 case IORING_OP_WRITE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006115 ret = io_write(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006116 break;
6117 case IORING_OP_FSYNC:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006118 ret = io_fsync(req, issue_flags);
Jackie Liuba5290c2019-10-09 09:19:59 +08006119 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006120 case IORING_OP_POLL_ADD:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006121 ret = io_poll_add(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006122 break;
6123 case IORING_OP_POLL_REMOVE:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006124 ret = io_poll_remove(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006125 break;
Jens Axboeb76da702019-11-20 13:05:32 -07006126 case IORING_OP_SYNC_FILE_RANGE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006127 ret = io_sync_file_range(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07006128 break;
6129 case IORING_OP_SENDMSG:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006130 ret = io_sendmsg(req, issue_flags);
Pavel Begunkov062d04d2020-10-10 18:34:12 +01006131 break;
Jens Axboefddafac2020-01-04 20:19:44 -07006132 case IORING_OP_SEND:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006133 ret = io_send(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07006134 break;
6135 case IORING_OP_RECVMSG:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006136 ret = io_recvmsg(req, issue_flags);
Pavel Begunkov062d04d2020-10-10 18:34:12 +01006137 break;
Jens Axboefddafac2020-01-04 20:19:44 -07006138 case IORING_OP_RECV:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006139 ret = io_recv(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07006140 break;
Jens Axboe561fb042019-10-24 07:25:42 -06006141 case IORING_OP_TIMEOUT:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006142 ret = io_timeout(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006143 break;
6144 case IORING_OP_TIMEOUT_REMOVE:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006145 ret = io_timeout_remove(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006146 break;
6147 case IORING_OP_ACCEPT:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006148 ret = io_accept(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006149 break;
6150 case IORING_OP_CONNECT:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006151 ret = io_connect(req, issue_flags);
Jens Axboe31b51512019-01-18 22:56:34 -07006152 break;
6153 case IORING_OP_ASYNC_CANCEL:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006154 ret = io_async_cancel(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006155 break;
Jens Axboed63d1b52019-12-10 10:38:56 -07006156 case IORING_OP_FALLOCATE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006157 ret = io_fallocate(req, issue_flags);
Jens Axboed63d1b52019-12-10 10:38:56 -07006158 break;
Jens Axboe15b71ab2019-12-11 11:20:36 -07006159 case IORING_OP_OPENAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006160 ret = io_openat(req, issue_flags);
Jens Axboe15b71ab2019-12-11 11:20:36 -07006161 break;
Jens Axboeb5dba592019-12-11 14:02:38 -07006162 case IORING_OP_CLOSE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006163 ret = io_close(req, issue_flags);
Jens Axboeb5dba592019-12-11 14:02:38 -07006164 break;
Jens Axboe05f3fb32019-12-09 11:22:50 -07006165 case IORING_OP_FILES_UPDATE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006166 ret = io_files_update(req, issue_flags);
Jens Axboe05f3fb32019-12-09 11:22:50 -07006167 break;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07006168 case IORING_OP_STATX:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006169 ret = io_statx(req, issue_flags);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07006170 break;
Jens Axboe4840e412019-12-25 22:03:45 -07006171 case IORING_OP_FADVISE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006172 ret = io_fadvise(req, issue_flags);
Jens Axboe4840e412019-12-25 22:03:45 -07006173 break;
Jens Axboec1ca7572019-12-25 22:18:28 -07006174 case IORING_OP_MADVISE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006175 ret = io_madvise(req, issue_flags);
Jens Axboec1ca7572019-12-25 22:18:28 -07006176 break;
Jens Axboecebdb982020-01-08 17:59:24 -07006177 case IORING_OP_OPENAT2:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006178 ret = io_openat2(req, issue_flags);
Jens Axboecebdb982020-01-08 17:59:24 -07006179 break;
Jens Axboe3e4827b2020-01-08 15:18:09 -07006180 case IORING_OP_EPOLL_CTL:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006181 ret = io_epoll_ctl(req, issue_flags);
Jens Axboe3e4827b2020-01-08 15:18:09 -07006182 break;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03006183 case IORING_OP_SPLICE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006184 ret = io_splice(req, issue_flags);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03006185 break;
Jens Axboeddf0322d2020-02-23 16:41:33 -07006186 case IORING_OP_PROVIDE_BUFFERS:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006187 ret = io_provide_buffers(req, issue_flags);
Jens Axboeddf0322d2020-02-23 16:41:33 -07006188 break;
Jens Axboe067524e2020-03-02 16:32:28 -07006189 case IORING_OP_REMOVE_BUFFERS:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006190 ret = io_remove_buffers(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006191 break;
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03006192 case IORING_OP_TEE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006193 ret = io_tee(req, issue_flags);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03006194 break;
Jens Axboe36f4fa62020-09-05 11:14:22 -06006195 case IORING_OP_SHUTDOWN:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006196 ret = io_shutdown(req, issue_flags);
Jens Axboe36f4fa62020-09-05 11:14:22 -06006197 break;
Jens Axboe80a261f2020-09-28 14:23:58 -06006198 case IORING_OP_RENAMEAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006199 ret = io_renameat(req, issue_flags);
Jens Axboe80a261f2020-09-28 14:23:58 -06006200 break;
Jens Axboe14a11432020-09-28 14:27:37 -06006201 case IORING_OP_UNLINKAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006202 ret = io_unlinkat(req, issue_flags);
Jens Axboe14a11432020-09-28 14:27:37 -06006203 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006204 default:
6205 ret = -EINVAL;
6206 break;
6207 }
Jens Axboe31b51512019-01-18 22:56:34 -07006208
Jens Axboe5730b272021-02-27 15:57:30 -07006209 if (creds)
6210 revert_creds(creds);
6211
Jens Axboe2b188cc2019-01-07 10:46:33 -07006212 if (ret)
6213 return ret;
6214
Jens Axboeb5325762020-05-19 21:20:27 -06006215 /* If the op doesn't have a file, we're not polling for it */
6216 if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) {
Jens Axboe11ba8202020-01-15 21:51:17 -07006217 const bool in_async = io_wq_current_is_worker();
6218
Jens Axboe11ba8202020-01-15 21:51:17 -07006219 /* workqueue context doesn't hold uring_lock, grab it now */
6220 if (in_async)
6221 mutex_lock(&ctx->uring_lock);
6222
Xiaoguang Wang2e9dbe92020-11-13 00:44:08 +08006223 io_iopoll_req_issued(req, in_async);
Jens Axboe11ba8202020-01-15 21:51:17 -07006224
6225 if (in_async)
6226 mutex_unlock(&ctx->uring_lock);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006227 }
6228
6229 return 0;
6230}
6231
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00006232static void io_wq_submit_work(struct io_wq_work *work)
Pavel Begunkovd4c81f32020-06-08 21:08:19 +03006233{
Jens Axboe2b188cc2019-01-07 10:46:33 -07006234 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Pavel Begunkov6df1db62020-07-03 22:15:06 +03006235 struct io_kiocb *timeout;
Jens Axboe561fb042019-10-24 07:25:42 -06006236 int ret = 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006237
Pavel Begunkov6df1db62020-07-03 22:15:06 +03006238 timeout = io_prep_linked_timeout(req);
6239 if (timeout)
6240 io_queue_linked_timeout(timeout);
Pavel Begunkovd4c81f32020-06-08 21:08:19 +03006241
Jens Axboe4014d942021-01-19 15:53:54 -07006242 if (work->flags & IO_WQ_WORK_CANCEL)
Jens Axboe561fb042019-10-24 07:25:42 -06006243 ret = -ECANCELED;
Jens Axboe31b51512019-01-18 22:56:34 -07006244
Jens Axboe561fb042019-10-24 07:25:42 -06006245 if (!ret) {
Jens Axboe561fb042019-10-24 07:25:42 -06006246 do {
Pavel Begunkov889fca72021-02-10 00:03:09 +00006247 ret = io_issue_sqe(req, 0);
Jens Axboe561fb042019-10-24 07:25:42 -06006248 /*
6249 * We can get EAGAIN for polled IO even though we're
6250 * forcing a sync submission from here, since we can't
6251 * wait for request slots on the block side.
6252 */
6253 if (ret != -EAGAIN)
6254 break;
6255 cond_resched();
6256 } while (1);
6257 }
Jens Axboe31b51512019-01-18 22:56:34 -07006258
Pavel Begunkova3df76982021-02-18 22:32:52 +00006259 /* avoid locking problems by failing it from a clean context */
Jens Axboe561fb042019-10-24 07:25:42 -06006260 if (ret) {
Pavel Begunkova3df76982021-02-18 22:32:52 +00006261 /* io-wq is going to take one down */
Jens Axboede9b4cc2021-02-24 13:28:27 -07006262 req_ref_get(req);
Pavel Begunkova3df76982021-02-18 22:32:52 +00006263 io_req_task_queue_fail(req, ret);
Jens Axboeedafcce2019-01-09 09:16:05 -07006264 }
Jens Axboe31b51512019-01-18 22:56:34 -07006265}
Jens Axboe2b188cc2019-01-07 10:46:33 -07006266
Jens Axboe7b29f922021-03-12 08:30:14 -07006267#define FFS_ASYNC_READ 0x1UL
6268#define FFS_ASYNC_WRITE 0x2UL
6269#ifdef CONFIG_64BIT
6270#define FFS_ISREG 0x4UL
6271#else
6272#define FFS_ISREG 0x0UL
6273#endif
6274#define FFS_MASK ~(FFS_ASYNC_READ|FFS_ASYNC_WRITE|FFS_ISREG)
6275
Pavel Begunkovaeca2412021-04-11 01:46:37 +01006276static inline struct io_fixed_file *io_fixed_file_slot(struct io_file_table *table,
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01006277 unsigned i)
Jens Axboe09bb8392019-03-13 12:39:28 -06006278{
Pavel Begunkovaeca2412021-04-11 01:46:37 +01006279 struct io_fixed_file *table_l2;
Jens Axboe65e19f52019-10-26 07:20:21 -06006280
Pavel Begunkovaeca2412021-04-11 01:46:37 +01006281 table_l2 = table->files[i >> IORING_FILE_TABLE_SHIFT];
6282 return &table_l2[i & IORING_FILE_TABLE_MASK];
Pavel Begunkovdafecf12021-02-28 22:35:11 +00006283}
6284
6285static inline struct file *io_file_from_index(struct io_ring_ctx *ctx,
6286 int index)
6287{
Pavel Begunkovaeca2412021-04-11 01:46:37 +01006288 struct io_fixed_file *slot = io_fixed_file_slot(&ctx->file_table, index);
Jens Axboe7b29f922021-03-12 08:30:14 -07006289
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01006290 return (struct file *) (slot->file_ptr & FFS_MASK);
Jens Axboe65e19f52019-10-26 07:20:21 -06006291}
6292
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01006293static void io_fixed_file_set(struct io_fixed_file *file_slot, struct file *file)
Pavel Begunkov9a321c92021-04-01 15:44:01 +01006294{
6295 unsigned long file_ptr = (unsigned long) file;
6296
6297 if (__io_file_supports_async(file, READ))
6298 file_ptr |= FFS_ASYNC_READ;
6299 if (__io_file_supports_async(file, WRITE))
6300 file_ptr |= FFS_ASYNC_WRITE;
6301 if (S_ISREG(file_inode(file)->i_mode))
6302 file_ptr |= FFS_ISREG;
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01006303 file_slot->file_ptr = file_ptr;
Pavel Begunkov9a321c92021-04-01 15:44:01 +01006304}
6305
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006306static struct file *io_file_get(struct io_submit_state *state,
6307 struct io_kiocb *req, int fd, bool fixed)
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006308{
6309 struct io_ring_ctx *ctx = req->ctx;
6310 struct file *file;
6311
6312 if (fixed) {
Jens Axboe7b29f922021-03-12 08:30:14 -07006313 unsigned long file_ptr;
6314
Pavel Begunkov479f5172020-10-10 18:34:07 +01006315 if (unlikely((unsigned int)fd >= ctx->nr_user_files))
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006316 return NULL;
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006317 fd = array_index_nospec(fd, ctx->nr_user_files);
Pavel Begunkovaeca2412021-04-11 01:46:37 +01006318 file_ptr = io_fixed_file_slot(&ctx->file_table, fd)->file_ptr;
Jens Axboe7b29f922021-03-12 08:30:14 -07006319 file = (struct file *) (file_ptr & FFS_MASK);
6320 file_ptr &= ~FFS_MASK;
6321 /* mask in overlapping REQ_F and FFS bits */
6322 req->flags |= (file_ptr << REQ_F_ASYNC_READ_BIT);
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01006323 io_req_set_rsrc_node(req);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006324 } else {
6325 trace_io_uring_file_get(ctx, fd);
6326 file = __io_file_get(state, fd);
Jens Axboed44f5542021-03-12 08:27:05 -07006327
6328 /* we don't allow fixed io_uring files */
6329 if (file && unlikely(file->f_op == &io_uring_fops))
6330 io_req_track_inflight(req);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006331 }
6332
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006333 return file;
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006334}
6335
Jens Axboe2665abf2019-11-05 12:40:47 -07006336static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer)
6337{
Jens Axboead8a48a2019-11-15 08:49:11 -07006338 struct io_timeout_data *data = container_of(timer,
6339 struct io_timeout_data, timer);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006340 struct io_kiocb *prev, *req = data->req;
Jens Axboe2665abf2019-11-05 12:40:47 -07006341 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2665abf2019-11-05 12:40:47 -07006342 unsigned long flags;
Jens Axboe2665abf2019-11-05 12:40:47 -07006343
6344 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006345 prev = req->timeout.head;
6346 req->timeout.head = NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006347
6348 /*
6349 * We don't expect the list to be empty, that will only happen if we
6350 * race with the completion of the linked work.
6351 */
Jens Axboede9b4cc2021-02-24 13:28:27 -07006352 if (prev && req_ref_inc_not_zero(prev))
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006353 io_remove_next_linked(prev);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006354 else
6355 prev = NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006356 spin_unlock_irqrestore(&ctx->completion_lock, flags);
6357
6358 if (prev) {
Pavel Begunkov014db002020-03-03 21:33:12 +03006359 io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME);
Pavel Begunkov9ae1f8d2021-02-01 18:59:51 +00006360 io_put_req_deferred(prev, 1);
Jens Axboe47f46762019-11-09 17:43:02 -07006361 } else {
Pavel Begunkov9ae1f8d2021-02-01 18:59:51 +00006362 io_req_complete_post(req, -ETIME, 0);
Jens Axboe2665abf2019-11-05 12:40:47 -07006363 }
Pavel Begunkovdf9727a2021-04-01 15:43:59 +01006364 io_put_req_deferred(req, 1);
Jens Axboe2665abf2019-11-05 12:40:47 -07006365 return HRTIMER_NORESTART;
6366}
6367
Pavel Begunkovde968c12021-03-19 17:22:33 +00006368static void io_queue_linked_timeout(struct io_kiocb *req)
Jens Axboe2665abf2019-11-05 12:40:47 -07006369{
Pavel Begunkovde968c12021-03-19 17:22:33 +00006370 struct io_ring_ctx *ctx = req->ctx;
6371
6372 spin_lock_irq(&ctx->completion_lock);
Jens Axboe76a46e02019-11-10 23:34:16 -07006373 /*
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006374 * If the back reference is NULL, then our linked request finished
6375 * before we got a chance to setup the timer
Jens Axboe76a46e02019-11-10 23:34:16 -07006376 */
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006377 if (req->timeout.head) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07006378 struct io_timeout_data *data = req->async_data;
Jens Axboe94ae5e72019-11-14 19:39:52 -07006379
Jens Axboead8a48a2019-11-15 08:49:11 -07006380 data->timer.function = io_link_timeout_fn;
6381 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts),
6382 data->mode);
Jens Axboe2665abf2019-11-05 12:40:47 -07006383 }
Jens Axboe76a46e02019-11-10 23:34:16 -07006384 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe2665abf2019-11-05 12:40:47 -07006385 /* drop submission reference */
Jens Axboe76a46e02019-11-10 23:34:16 -07006386 io_put_req(req);
Jens Axboe2665abf2019-11-05 12:40:47 -07006387}
6388
Jens Axboead8a48a2019-11-15 08:49:11 -07006389static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req)
Jens Axboe2665abf2019-11-05 12:40:47 -07006390{
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006391 struct io_kiocb *nxt = req->link;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006392
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006393 if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) ||
6394 nxt->opcode != IORING_OP_LINK_TIMEOUT)
Jens Axboed7718a92020-02-14 22:23:12 -07006395 return NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006396
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006397 nxt->timeout.head = req;
Pavel Begunkov900fad42020-10-19 16:39:16 +01006398 nxt->flags |= REQ_F_LTIMEOUT_ACTIVE;
Jens Axboe76a46e02019-11-10 23:34:16 -07006399 req->flags |= REQ_F_LINK_TIMEOUT;
Jens Axboe76a46e02019-11-10 23:34:16 -07006400 return nxt;
Jens Axboe2665abf2019-11-05 12:40:47 -07006401}
6402
Pavel Begunkovc5eef2b942021-02-10 00:03:22 +00006403static void __io_queue_sqe(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07006404{
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006405 struct io_kiocb *linked_timeout = io_prep_linked_timeout(req);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006406 int ret;
6407
Pavel Begunkovc5eef2b942021-02-10 00:03:22 +00006408 ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER);
Jens Axboe491381ce2019-10-17 09:20:46 -06006409
6410 /*
6411 * We async punt it if the file wasn't marked NOWAIT, or if the file
6412 * doesn't support non-blocking read/write attempts
6413 */
Pavel Begunkov18400382021-03-19 17:22:34 +00006414 if (likely(!ret)) {
Pavel Begunkov0d63c142020-10-22 16:47:18 +01006415 /* drop submission reference */
Pavel Begunkove342c802021-01-19 13:32:47 +00006416 if (req->flags & REQ_F_COMPLETE_INLINE) {
Pavel Begunkovc5eef2b942021-02-10 00:03:22 +00006417 struct io_ring_ctx *ctx = req->ctx;
6418 struct io_comp_state *cs = &ctx->submit_state.comp;
Jens Axboee65ef562019-03-12 10:16:44 -06006419
Pavel Begunkov6dd0be12021-02-10 00:03:13 +00006420 cs->reqs[cs->nr++] = req;
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006421 if (cs->nr == ARRAY_SIZE(cs->reqs))
Pavel Begunkovc5eef2b942021-02-10 00:03:22 +00006422 io_submit_flush_completions(cs, ctx);
Pavel Begunkov9affd662021-01-19 13:32:46 +00006423 } else {
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006424 io_put_req(req);
Pavel Begunkov0d63c142020-10-22 16:47:18 +01006425 }
Pavel Begunkov18400382021-03-19 17:22:34 +00006426 } else if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) {
6427 if (!io_arm_poll_handler(req)) {
6428 /*
6429 * Queued up for async execution, worker will release
6430 * submit reference when the iocb is actually submitted.
6431 */
6432 io_queue_async_work(req);
6433 }
Pavel Begunkov0d63c142020-10-22 16:47:18 +01006434 } else {
Pavel Begunkovf41db2732021-02-28 22:35:12 +00006435 io_req_complete_failed(req, ret);
Jens Axboe9e645e112019-05-10 16:07:28 -06006436 }
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006437 if (linked_timeout)
6438 io_queue_linked_timeout(linked_timeout);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006439}
6440
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006441static void io_queue_sqe(struct io_kiocb *req)
Jackie Liu4fe2c962019-09-09 20:50:40 +08006442{
6443 int ret;
6444
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006445 ret = io_req_defer(req);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006446 if (ret) {
6447 if (ret != -EIOCBQUEUED) {
Pavel Begunkov11185912020-01-22 23:09:35 +03006448fail_req:
Pavel Begunkovf41db2732021-02-28 22:35:12 +00006449 io_req_complete_failed(req, ret);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006450 }
Pavel Begunkov25508782019-12-30 21:24:47 +03006451 } else if (req->flags & REQ_F_FORCE_ASYNC) {
Pavel Begunkovb7e298d2021-02-28 22:35:19 +00006452 ret = io_req_prep_async(req);
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006453 if (unlikely(ret))
6454 goto fail_req;
Jens Axboece35a472019-12-17 08:04:44 -07006455 io_queue_async_work(req);
6456 } else {
Pavel Begunkovc5eef2b942021-02-10 00:03:22 +00006457 __io_queue_sqe(req);
Jens Axboece35a472019-12-17 08:04:44 -07006458 }
Jackie Liu4fe2c962019-09-09 20:50:40 +08006459}
6460
Stefano Garzarella21b55db2020-08-27 16:58:30 +02006461/*
6462 * Check SQE restrictions (opcode and flags).
6463 *
6464 * Returns 'true' if SQE is allowed, 'false' otherwise.
6465 */
6466static inline bool io_check_restriction(struct io_ring_ctx *ctx,
6467 struct io_kiocb *req,
6468 unsigned int sqe_flags)
6469{
6470 if (!ctx->restricted)
6471 return true;
6472
6473 if (!test_bit(req->opcode, ctx->restrictions.sqe_op))
6474 return false;
6475
6476 if ((sqe_flags & ctx->restrictions.sqe_flags_required) !=
6477 ctx->restrictions.sqe_flags_required)
6478 return false;
6479
6480 if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed |
6481 ctx->restrictions.sqe_flags_required))
6482 return false;
6483
6484 return true;
6485}
6486
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006487static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req,
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006488 const struct io_uring_sqe *sqe)
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006489{
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006490 struct io_submit_state *state;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006491 unsigned int sqe_flags;
Jens Axboe003e8dc2021-03-06 09:22:27 -07006492 int personality, ret = 0;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006493
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006494 req->opcode = READ_ONCE(sqe->opcode);
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006495 /* same numerical values with corresponding REQ_F_*, safe to copy */
6496 req->flags = sqe_flags = READ_ONCE(sqe->flags);
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006497 req->user_data = READ_ONCE(sqe->user_data);
Jens Axboee8c2bc12020-08-15 18:44:09 -07006498 req->async_data = NULL;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006499 req->file = NULL;
6500 req->ctx = ctx;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006501 req->link = NULL;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006502 req->fixed_rsrc_refs = NULL;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006503 /* one is dropped after submission, the other at completion */
Jens Axboeabc54d62021-02-24 13:32:30 -07006504 atomic_set(&req->refs, 2);
Pavel Begunkov4dd28242020-06-15 10:33:13 +03006505 req->task = current;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006506 req->result = 0;
Jens Axboe93e68e02021-03-09 07:02:21 -07006507 req->work.creds = NULL;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006508
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006509 /* enforce forwards compatibility on users */
Pavel Begunkovebf4a5d2021-02-20 01:39:53 +00006510 if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) {
6511 req->flags = 0;
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006512 return -EINVAL;
Pavel Begunkovebf4a5d2021-02-20 01:39:53 +00006513 }
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006514
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006515 if (unlikely(req->opcode >= IORING_OP_LAST))
6516 return -EINVAL;
6517
Stefano Garzarella21b55db2020-08-27 16:58:30 +02006518 if (unlikely(!io_check_restriction(ctx, req, sqe_flags)))
6519 return -EACCES;
6520
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006521 if ((sqe_flags & IOSQE_BUFFER_SELECT) &&
6522 !io_op_defs[req->opcode].buffer_select)
6523 return -EOPNOTSUPP;
6524
Jens Axboe003e8dc2021-03-06 09:22:27 -07006525 personality = READ_ONCE(sqe->personality);
6526 if (personality) {
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00006527 req->work.creds = xa_load(&ctx->personalities, personality);
Jens Axboe003e8dc2021-03-06 09:22:27 -07006528 if (!req->work.creds)
6529 return -EINVAL;
6530 get_cred(req->work.creds);
Jens Axboe003e8dc2021-03-06 09:22:27 -07006531 }
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006532 state = &ctx->submit_state;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006533
Jens Axboe27926b62020-10-28 09:33:23 -06006534 /*
6535 * Plug now if we have more than 1 IO left after this, and the target
6536 * is potentially a read/write to block based storage.
6537 */
6538 if (!state->plug_started && state->ios_left > 1 &&
6539 io_op_defs[req->opcode].plug) {
6540 blk_start_plug(&state->plug);
6541 state->plug_started = true;
6542 }
Jens Axboe63ff8222020-05-07 14:56:15 -06006543
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006544 if (io_op_defs[req->opcode].needs_file) {
6545 bool fixed = req->flags & REQ_F_FIXED_FILE;
Jens Axboe63ff8222020-05-07 14:56:15 -06006546
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006547 req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed);
Pavel Begunkovba13e232021-02-01 18:59:52 +00006548 if (unlikely(!req->file))
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006549 ret = -EBADF;
6550 }
6551
Pavel Begunkov71b547c2020-10-10 18:34:09 +01006552 state->ios_left--;
6553 return ret;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006554}
6555
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006556static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req,
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006557 const struct io_uring_sqe *sqe)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006558{
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006559 struct io_submit_link *link = &ctx->submit_state.link;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006560 int ret;
6561
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006562 ret = io_init_req(ctx, req, sqe);
6563 if (unlikely(ret)) {
6564fail_req:
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006565 if (link->head) {
6566 /* fail even hard links since we don't submit */
Pavel Begunkovcf109602021-02-18 18:29:43 +00006567 link->head->flags |= REQ_F_FAIL_LINK;
Pavel Begunkovf41db2732021-02-28 22:35:12 +00006568 io_req_complete_failed(link->head, -ECANCELED);
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006569 link->head = NULL;
6570 }
Pavel Begunkovf41db2732021-02-28 22:35:12 +00006571 io_req_complete_failed(req, ret);
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006572 return ret;
6573 }
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006574 ret = io_req_prep(req, sqe);
6575 if (unlikely(ret))
6576 goto fail_req;
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006577
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006578 /* don't need @sqe from now on */
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006579 trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data,
6580 true, ctx->flags & IORING_SETUP_SQPOLL);
6581
Jens Axboe6c271ce2019-01-10 11:22:30 -07006582 /*
6583 * If we already have a head request, queue this one for async
6584 * submittal once the head completes. If we don't have a head but
6585 * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be
6586 * submitted sync once the chain is complete. If none of those
6587 * conditions are true (normal request), then just queue it.
6588 */
6589 if (link->head) {
6590 struct io_kiocb *head = link->head;
6591
6592 /*
6593 * Taking sequential execution of a link, draining both sides
6594 * of the link also fullfils IOSQE_IO_DRAIN semantics for all
6595 * requests in the link. So, it drains the head and the
6596 * next after the link request. The last one is done via
6597 * drain_next flag to persist the effect across calls.
6598 */
6599 if (req->flags & REQ_F_IO_DRAIN) {
6600 head->flags |= REQ_F_IO_DRAIN;
6601 ctx->drain_next = 1;
6602 }
Pavel Begunkovb7e298d2021-02-28 22:35:19 +00006603 ret = io_req_prep_async(req);
Pavel Begunkovcf109602021-02-18 18:29:43 +00006604 if (unlikely(ret))
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006605 goto fail_req;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006606 trace_io_uring_link(ctx, req, head);
6607 link->last->link = req;
6608 link->last = req;
6609
6610 /* last request of a link, enqueue the link */
6611 if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) {
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006612 io_queue_sqe(head);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006613 link->head = NULL;
6614 }
Jackie Liu4fe2c962019-09-09 20:50:40 +08006615 } else {
6616 if (unlikely(ctx->drain_next)) {
6617 req->flags |= REQ_F_IO_DRAIN;
6618 ctx->drain_next = 0;
6619 }
6620 if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) {
Jackie Liu4fe2c962019-09-09 20:50:40 +08006621 link->head = req;
6622 link->last = req;
6623 } else {
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006624 io_queue_sqe(req);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006625 }
6626 }
6627
6628 return 0;
6629}
6630
6631/*
6632 * Batched submission is done, ensure local IO is flushed out.
6633 */
6634static void io_submit_state_end(struct io_submit_state *state,
6635 struct io_ring_ctx *ctx)
Pavel Begunkov1b4a51b2019-11-21 11:54:28 +03006636{
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006637 if (state->link.head)
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006638 io_queue_sqe(state->link.head);
Jens Axboe3529d8c2019-12-19 18:24:38 -07006639 if (state->comp.nr)
Jens Axboe9e645e112019-05-10 16:07:28 -06006640 io_submit_flush_completions(&state->comp, ctx);
Jackie Liua197f662019-11-08 08:09:12 -07006641 if (state->plug_started)
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006642 blk_finish_plug(&state->plug);
Jens Axboe75c6a032020-01-28 10:15:23 -07006643 io_state_file_put(state);
Jens Axboe9e645e112019-05-10 16:07:28 -06006644}
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006645
Jens Axboe9e645e112019-05-10 16:07:28 -06006646/*
6647 * Start submission side cache.
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006648 */
Jens Axboe9e645e112019-05-10 16:07:28 -06006649static void io_submit_state_start(struct io_submit_state *state,
Pavel Begunkov196be952019-11-07 01:41:06 +03006650 unsigned int max_ios)
Jens Axboe9e645e112019-05-10 16:07:28 -06006651{
6652 state->plug_started = false;
Jens Axboebcda7ba2020-02-23 16:42:51 -07006653 state->ios_left = max_ios;
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006654 /* set only head, no need to init link_last in advance */
6655 state->link.head = NULL;
Jens Axboe75c6a032020-01-28 10:15:23 -07006656}
6657
Jens Axboe193155c2020-02-22 23:22:19 -07006658static void io_commit_sqring(struct io_ring_ctx *ctx)
6659{
Jens Axboe75c6a032020-01-28 10:15:23 -07006660 struct io_rings *rings = ctx->rings;
6661
6662 /*
Jens Axboe193155c2020-02-22 23:22:19 -07006663 * Ensure any loads from the SQEs are done at this point,
Jens Axboe75c6a032020-01-28 10:15:23 -07006664 * since once we write the new head, the application could
6665 * write new data to them.
Pavel Begunkov6b47ee62020-01-18 20:22:41 +03006666 */
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006667 smp_store_release(&rings->sq.head, ctx->cached_sq_head);
Jens Axboebcda7ba2020-02-23 16:42:51 -07006668}
6669
Jens Axboe9e645e112019-05-10 16:07:28 -06006670/*
Jens Axboe3529d8c2019-12-19 18:24:38 -07006671 * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory
Jens Axboe9e645e112019-05-10 16:07:28 -06006672 * that is mapped by userspace. This means that care needs to be taken to
6673 * ensure that reads are stable, as we cannot rely on userspace always
Jens Axboe78e19bb2019-11-06 15:21:34 -07006674 * being a good citizen. If members of the sqe are validated and then later
6675 * used, it's important that those reads are done through READ_ONCE() to
Pavel Begunkov2e6e1fd2019-12-05 16:15:45 +03006676 * prevent a re-load down the line.
Jens Axboe9e645e112019-05-10 16:07:28 -06006677 */
6678static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx)
Jens Axboe9e645e112019-05-10 16:07:28 -06006679{
6680 u32 *sq_array = ctx->sq_array;
6681 unsigned head;
6682
6683 /*
6684 * The cached sq head (or cq tail) serves two purposes:
6685 *
6686 * 1) allows us to batch the cost of updating the user visible
Pavel Begunkov9d763772019-12-17 02:22:07 +03006687 * head updates.
Jens Axboe9e645e112019-05-10 16:07:28 -06006688 * 2) allows the kernel side to track the head on its own, even
Pavel Begunkov8cdf2192020-01-25 00:40:24 +03006689 * though the application is the one updating it.
6690 */
6691 head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]);
6692 if (likely(head < ctx->sq_entries))
6693 return &ctx->sq_sqes[head];
6694
6695 /* drop invalid entries */
Pavel Begunkov711be032020-01-17 03:57:59 +03006696 ctx->cached_sq_dropped++;
6697 WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped);
6698 return NULL;
6699}
Jens Axboeb7bb4f72019-12-15 22:13:43 -07006700
Jens Axboe0f212202020-09-13 13:09:39 -06006701static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006702{
Pavel Begunkov46c4e162021-02-18 18:29:37 +00006703 int submitted = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006704
Jens Axboec4a2ed72019-11-21 21:01:26 -07006705 /* if we have a backlog and couldn't flush it all, return BUSY */
Jens Axboead3eb2c2019-12-18 17:12:20 -07006706 if (test_bit(0, &ctx->sq_check_overflow)) {
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00006707 if (!__io_cqring_overflow_flush(ctx, false))
Jens Axboead3eb2c2019-12-18 17:12:20 -07006708 return -EBUSY;
6709 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07006710
Pavel Begunkovee7d46d92019-12-30 21:24:45 +03006711 /* make sure SQ entry isn't read before tail */
6712 nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx));
Pavel Begunkov9ef4f122019-12-30 21:24:44 +03006713
Pavel Begunkov2b85edf2019-12-28 14:13:03 +03006714 if (!percpu_ref_tryget_many(&ctx->refs, nr))
6715 return -EAGAIN;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006716
Jens Axboed8a6df12020-10-15 16:24:45 -06006717 percpu_counter_add(&current->io_uring->inflight, nr);
Jens Axboefaf7b512020-10-07 12:48:53 -06006718 refcount_add(nr, &current->usage);
Pavel Begunkovba88ff12021-02-10 00:03:11 +00006719 io_submit_state_start(&ctx->submit_state, nr);
Pavel Begunkovb14cca02020-01-17 04:45:59 +03006720
Pavel Begunkov46c4e162021-02-18 18:29:37 +00006721 while (submitted < nr) {
Jens Axboe3529d8c2019-12-19 18:24:38 -07006722 const struct io_uring_sqe *sqe;
Pavel Begunkov196be952019-11-07 01:41:06 +03006723 struct io_kiocb *req;
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03006724
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006725 req = io_alloc_req(ctx);
Pavel Begunkov196be952019-11-07 01:41:06 +03006726 if (unlikely(!req)) {
6727 if (!submitted)
6728 submitted = -EAGAIN;
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03006729 break;
Jens Axboe9e645e112019-05-10 16:07:28 -06006730 }
Pavel Begunkov4fccfcb2021-02-12 11:55:17 +00006731 sqe = io_get_sqe(ctx);
6732 if (unlikely(!sqe)) {
6733 kmem_cache_free(req_cachep, req);
6734 break;
6735 }
Jens Axboed3656342019-12-18 09:50:26 -07006736 /* will complete beyond this point, count as submitted */
6737 submitted++;
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006738 if (io_submit_sqe(ctx, req, sqe))
Jens Axboed3656342019-12-18 09:50:26 -07006739 break;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006740 }
6741
Pavel Begunkov9466f432020-01-25 22:34:01 +03006742 if (unlikely(submitted != nr)) {
6743 int ref_used = (submitted == -EAGAIN) ? 0 : submitted;
Jens Axboed8a6df12020-10-15 16:24:45 -06006744 struct io_uring_task *tctx = current->io_uring;
6745 int unused = nr - ref_used;
Pavel Begunkov9466f432020-01-25 22:34:01 +03006746
Jens Axboed8a6df12020-10-15 16:24:45 -06006747 percpu_ref_put_many(&ctx->refs, unused);
6748 percpu_counter_sub(&tctx->inflight, unused);
6749 put_task_struct_many(current, unused);
Pavel Begunkov9466f432020-01-25 22:34:01 +03006750 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07006751
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006752 io_submit_state_end(&ctx->submit_state, ctx);
Pavel Begunkovae9428c2019-11-06 00:22:14 +03006753 /* Commit SQ ring head once we've consumed and submitted all SQEs */
6754 io_commit_sqring(ctx);
6755
Jens Axboe6c271ce2019-01-10 11:22:30 -07006756 return submitted;
6757}
6758
Xiaoguang Wang23b36282020-07-23 20:57:24 +08006759static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx)
6760{
6761 /* Tell userspace we may need a wakeup call */
6762 spin_lock_irq(&ctx->completion_lock);
6763 ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP;
6764 spin_unlock_irq(&ctx->completion_lock);
6765}
6766
6767static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx)
6768{
6769 spin_lock_irq(&ctx->completion_lock);
6770 ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP;
6771 spin_unlock_irq(&ctx->completion_lock);
6772}
6773
Xiaoguang Wang08369242020-11-03 14:15:59 +08006774static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006775{
Jens Axboec8d1ba52020-09-14 11:07:26 -06006776 unsigned int to_submit;
Xiaoguang Wangbdcd3ea2020-02-25 22:12:08 +08006777 int ret = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006778
Jens Axboec8d1ba52020-09-14 11:07:26 -06006779 to_submit = io_sqring_entries(ctx);
Jens Axboee95eee22020-09-08 09:11:32 -06006780 /* if we're handling multiple rings, cap submit size for fairness */
6781 if (cap_entries && to_submit > 8)
6782 to_submit = 8;
6783
Xiaoguang Wang906a3c62020-11-12 14:56:00 +08006784 if (!list_empty(&ctx->iopoll_list) || to_submit) {
6785 unsigned nr_events = 0;
6786
Xiaoguang Wang08369242020-11-03 14:15:59 +08006787 mutex_lock(&ctx->uring_lock);
Xiaoguang Wang906a3c62020-11-12 14:56:00 +08006788 if (!list_empty(&ctx->iopoll_list))
6789 io_do_iopoll(ctx, &nr_events, 0);
6790
Pavel Begunkov0298ef92021-03-08 13:20:57 +00006791 if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) &&
6792 !(ctx->flags & IORING_SETUP_R_DISABLED))
Xiaoguang Wang08369242020-11-03 14:15:59 +08006793 ret = io_submit_sqes(ctx, to_submit);
6794 mutex_unlock(&ctx->uring_lock);
6795 }
Jens Axboe90554202020-09-03 12:12:41 -06006796
6797 if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait))
6798 wake_up(&ctx->sqo_sq_wait);
6799
Xiaoguang Wang08369242020-11-03 14:15:59 +08006800 return ret;
6801}
6802
6803static void io_sqd_update_thread_idle(struct io_sq_data *sqd)
6804{
6805 struct io_ring_ctx *ctx;
6806 unsigned sq_thread_idle = 0;
6807
Pavel Begunkovc9dca272021-03-10 13:13:55 +00006808 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6809 sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006810 sqd->sq_thread_idle = sq_thread_idle;
Jens Axboec8d1ba52020-09-14 11:07:26 -06006811}
6812
Jens Axboe6c271ce2019-01-10 11:22:30 -07006813static int io_sq_thread(void *data)
6814{
Jens Axboe69fb2132020-09-14 11:16:23 -06006815 struct io_sq_data *sqd = data;
6816 struct io_ring_ctx *ctx;
Xiaoguang Wanga0d9205f2020-11-12 14:55:59 +08006817 unsigned long timeout = 0;
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006818 char buf[TASK_COMM_LEN];
Xiaoguang Wang08369242020-11-03 14:15:59 +08006819 DEFINE_WAIT(wait);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006820
Pavel Begunkov696ee882021-04-01 09:55:04 +01006821 snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006822 set_task_comm(current, buf);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006823 current->pf_io_worker = NULL;
Jens Axboe28cea78a2020-09-14 10:51:17 -06006824
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006825 if (sqd->sq_cpu != -1)
6826 set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu));
6827 else
6828 set_cpus_allowed_ptr(current, cpu_online_mask);
6829 current->flags |= PF_NO_SETAFFINITY;
6830
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00006831 mutex_lock(&sqd->lock);
Jens Axboe05962f92021-03-06 13:58:48 -07006832 while (!test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)) {
Xiaoguang Wang08369242020-11-03 14:15:59 +08006833 int ret;
6834 bool cap_entries, sqt_spin, needs_sched;
Jens Axboec1edbf52019-11-10 16:56:04 -07006835
Jens Axboe82734c52021-03-29 06:52:44 -06006836 if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) ||
6837 signal_pending(current)) {
6838 bool did_sig = false;
6839
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00006840 mutex_unlock(&sqd->lock);
Jens Axboe82734c52021-03-29 06:52:44 -06006841 if (signal_pending(current)) {
6842 struct ksignal ksig;
6843
6844 did_sig = get_signal(&ksig);
6845 }
Jens Axboe05962f92021-03-06 13:58:48 -07006846 cond_resched();
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00006847 mutex_lock(&sqd->lock);
Jens Axboe82734c52021-03-29 06:52:44 -06006848 if (did_sig)
6849 break;
Pavel Begunkov521d6a72021-03-11 23:29:38 +00006850 io_run_task_work();
Pavel Begunkovb7f5a0b2021-03-15 14:23:08 +00006851 io_run_task_work_head(&sqd->park_task_work);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006852 timeout = jiffies + sqd->sq_thread_idle;
Pavel Begunkov7d41e852021-03-10 13:13:54 +00006853 continue;
Xiaoguang Wang08369242020-11-03 14:15:59 +08006854 }
Xiaoguang Wang08369242020-11-03 14:15:59 +08006855 sqt_spin = false;
Jens Axboee95eee22020-09-08 09:11:32 -06006856 cap_entries = !list_is_singular(&sqd->ctx_list);
Jens Axboe69fb2132020-09-14 11:16:23 -06006857 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
Stefan Metzmacher7c30f36a2021-03-07 11:54:28 +01006858 const struct cred *creds = NULL;
6859
6860 if (ctx->sq_creds != current_cred())
6861 creds = override_creds(ctx->sq_creds);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006862 ret = __io_sq_thread(ctx, cap_entries);
Stefan Metzmacher7c30f36a2021-03-07 11:54:28 +01006863 if (creds)
6864 revert_creds(creds);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006865 if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list)))
6866 sqt_spin = true;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006867 }
6868
Xiaoguang Wang08369242020-11-03 14:15:59 +08006869 if (sqt_spin || !time_after(jiffies, timeout)) {
Jens Axboec8d1ba52020-09-14 11:07:26 -06006870 io_run_task_work();
6871 cond_resched();
Xiaoguang Wang08369242020-11-03 14:15:59 +08006872 if (sqt_spin)
6873 timeout = jiffies + sqd->sq_thread_idle;
6874 continue;
6875 }
6876
Xiaoguang Wang08369242020-11-03 14:15:59 +08006877 needs_sched = true;
6878 prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE);
6879 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
6880 if ((ctx->flags & IORING_SETUP_IOPOLL) &&
6881 !list_empty_careful(&ctx->iopoll_list)) {
6882 needs_sched = false;
6883 break;
6884 }
6885 if (io_sqring_entries(ctx)) {
6886 needs_sched = false;
6887 break;
6888 }
6889 }
6890
Jens Axboe05962f92021-03-06 13:58:48 -07006891 if (needs_sched && !test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) {
Jens Axboe69fb2132020-09-14 11:16:23 -06006892 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6893 io_ring_set_wakeup_flag(ctx);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006894
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00006895 mutex_unlock(&sqd->lock);
Jens Axboe69fb2132020-09-14 11:16:23 -06006896 schedule();
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00006897 mutex_lock(&sqd->lock);
Jens Axboe69fb2132020-09-14 11:16:23 -06006898 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6899 io_ring_clear_wakeup_flag(ctx);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006900 }
Xiaoguang Wang08369242020-11-03 14:15:59 +08006901
6902 finish_wait(&sqd->wait, &wait);
Pavel Begunkovb7f5a0b2021-03-15 14:23:08 +00006903 io_run_task_work_head(&sqd->park_task_work);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006904 timeout = jiffies + sqd->sq_thread_idle;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006905 }
6906
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006907 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6908 io_uring_cancel_sqpoll(ctx);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006909 sqd->thread = NULL;
Jens Axboe05962f92021-03-06 13:58:48 -07006910 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
Jens Axboe5f3f26f2021-02-25 10:17:46 -07006911 io_ring_set_wakeup_flag(ctx);
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00006912 mutex_unlock(&sqd->lock);
Pavel Begunkov521d6a72021-03-11 23:29:38 +00006913
6914 io_run_task_work();
Pavel Begunkovb7f5a0b2021-03-15 14:23:08 +00006915 io_run_task_work_head(&sqd->park_task_work);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006916 complete(&sqd->exited);
6917 do_exit(0);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006918}
6919
Jens Axboebda52162019-09-24 13:47:15 -06006920struct io_wait_queue {
6921 struct wait_queue_entry wq;
6922 struct io_ring_ctx *ctx;
6923 unsigned to_wait;
6924 unsigned nr_timeouts;
6925};
6926
Pavel Begunkov6c503152021-01-04 20:36:36 +00006927static inline bool io_should_wake(struct io_wait_queue *iowq)
Jens Axboebda52162019-09-24 13:47:15 -06006928{
6929 struct io_ring_ctx *ctx = iowq->ctx;
6930
6931 /*
Brian Gianforcarod195a662019-12-13 03:09:50 -08006932 * Wake up if we have enough events, or if a timeout occurred since we
Jens Axboebda52162019-09-24 13:47:15 -06006933 * started waiting. For timeouts, we always want to return to userspace,
6934 * regardless of event count.
6935 */
Pavel Begunkov6c503152021-01-04 20:36:36 +00006936 return io_cqring_events(ctx) >= iowq->to_wait ||
Jens Axboebda52162019-09-24 13:47:15 -06006937 atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts;
6938}
6939
6940static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode,
6941 int wake_flags, void *key)
6942{
6943 struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue,
6944 wq);
6945
Pavel Begunkov6c503152021-01-04 20:36:36 +00006946 /*
6947 * Cannot safely flush overflowed CQEs from here, ensure we wake up
6948 * the task, and the next invocation will do it.
6949 */
6950 if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow))
6951 return autoremove_wake_function(curr, mode, wake_flags, key);
6952 return -1;
Jens Axboebda52162019-09-24 13:47:15 -06006953}
6954
Jens Axboeaf9c1a42020-09-24 13:32:18 -06006955static int io_run_task_work_sig(void)
6956{
6957 if (io_run_task_work())
6958 return 1;
6959 if (!signal_pending(current))
6960 return 0;
Jens Axboe0b8cfa92021-03-21 14:16:08 -06006961 if (test_thread_flag(TIF_NOTIFY_SIGNAL))
Jens Axboe792ee0f62020-10-22 20:17:18 -06006962 return -ERESTARTSYS;
Jens Axboeaf9c1a42020-09-24 13:32:18 -06006963 return -EINTR;
6964}
6965
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00006966/* when returns >0, the caller should retry */
6967static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx,
6968 struct io_wait_queue *iowq,
6969 signed long *timeout)
6970{
6971 int ret;
6972
6973 /* make sure we run task_work before checking for signals */
6974 ret = io_run_task_work_sig();
6975 if (ret || io_should_wake(iowq))
6976 return ret;
6977 /* let the caller flush overflows, retry */
6978 if (test_bit(0, &ctx->cq_check_overflow))
6979 return 1;
6980
6981 *timeout = schedule_timeout(*timeout);
6982 return !*timeout ? -ETIME : 1;
6983}
6984
Jens Axboe2b188cc2019-01-07 10:46:33 -07006985/*
6986 * Wait until events become available, if we don't already have some. The
6987 * application must reap them itself, as they reside on the shared cq ring.
6988 */
6989static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
Hao Xuc73ebb62020-11-03 10:54:37 +08006990 const sigset_t __user *sig, size_t sigsz,
6991 struct __kernel_timespec __user *uts)
Jens Axboe2b188cc2019-01-07 10:46:33 -07006992{
Jens Axboebda52162019-09-24 13:47:15 -06006993 struct io_wait_queue iowq = {
6994 .wq = {
6995 .private = current,
6996 .func = io_wake_function,
6997 .entry = LIST_HEAD_INIT(iowq.wq.entry),
6998 },
6999 .ctx = ctx,
7000 .to_wait = min_events,
7001 };
Hristo Venev75b28af2019-08-26 17:23:46 +00007002 struct io_rings *rings = ctx->rings;
Pavel Begunkovc1d5a222021-02-04 13:51:57 +00007003 signed long timeout = MAX_SCHEDULE_TIMEOUT;
7004 int ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07007005
Jens Axboeb41e9852020-02-17 09:52:41 -07007006 do {
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00007007 io_cqring_overflow_flush(ctx, false);
Pavel Begunkov6c503152021-01-04 20:36:36 +00007008 if (io_cqring_events(ctx) >= min_events)
Jens Axboeb41e9852020-02-17 09:52:41 -07007009 return 0;
Jens Axboe4c6e2772020-07-01 11:29:10 -06007010 if (!io_run_task_work())
Jens Axboeb41e9852020-02-17 09:52:41 -07007011 break;
Jens Axboeb41e9852020-02-17 09:52:41 -07007012 } while (1);
Jens Axboe2b188cc2019-01-07 10:46:33 -07007013
7014 if (sig) {
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01007015#ifdef CONFIG_COMPAT
7016 if (in_compat_syscall())
7017 ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig,
Oleg Nesterovb7724342019-07-16 16:29:53 -07007018 sigsz);
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01007019 else
7020#endif
Oleg Nesterovb7724342019-07-16 16:29:53 -07007021 ret = set_user_sigmask(sig, sigsz);
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01007022
Jens Axboe2b188cc2019-01-07 10:46:33 -07007023 if (ret)
7024 return ret;
7025 }
7026
Hao Xuc73ebb62020-11-03 10:54:37 +08007027 if (uts) {
Pavel Begunkovc1d5a222021-02-04 13:51:57 +00007028 struct timespec64 ts;
7029
Hao Xuc73ebb62020-11-03 10:54:37 +08007030 if (get_timespec64(&ts, uts))
7031 return -EFAULT;
7032 timeout = timespec64_to_jiffies(&ts);
7033 }
7034
Jens Axboebda52162019-09-24 13:47:15 -06007035 iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts);
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02007036 trace_io_uring_cqring_wait(ctx, min_events);
Jens Axboebda52162019-09-24 13:47:15 -06007037 do {
Jens Axboeca0a2652021-03-04 17:15:48 -07007038 /* if we can't even flush overflow, don't wait for more */
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00007039 if (!io_cqring_overflow_flush(ctx, false)) {
Jens Axboeca0a2652021-03-04 17:15:48 -07007040 ret = -EBUSY;
7041 break;
7042 }
Jens Axboebda52162019-09-24 13:47:15 -06007043 prepare_to_wait_exclusive(&ctx->wait, &iowq.wq,
7044 TASK_INTERRUPTIBLE);
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00007045 ret = io_cqring_wait_schedule(ctx, &iowq, &timeout);
7046 finish_wait(&ctx->wait, &iowq.wq);
Jens Axboeca0a2652021-03-04 17:15:48 -07007047 cond_resched();
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00007048 } while (ret > 0);
Jens Axboebda52162019-09-24 13:47:15 -06007049
Jens Axboeb7db41c2020-07-04 08:55:50 -06007050 restore_saved_sigmask_unless(ret == -EINTR);
Jens Axboe2b188cc2019-01-07 10:46:33 -07007051
Hristo Venev75b28af2019-08-26 17:23:46 +00007052 return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07007053}
7054
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007055static void io_free_file_tables(struct io_file_table *table, unsigned nr_files)
Pavel Begunkov846a4ef2021-04-01 15:44:03 +01007056{
7057 unsigned i, nr_tables = DIV_ROUND_UP(nr_files, IORING_MAX_FILES_TABLE);
7058
7059 for (i = 0; i < nr_tables; i++)
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007060 kfree(table->files[i]);
7061 kfree(table->files);
7062 table->files = NULL;
Pavel Begunkov846a4ef2021-04-01 15:44:03 +01007063}
7064
Jens Axboe6b063142019-01-10 22:13:58 -07007065static void __io_sqe_files_unregister(struct io_ring_ctx *ctx)
7066{
7067#if defined(CONFIG_UNIX)
7068 if (ctx->ring_sock) {
7069 struct sock *sock = ctx->ring_sock->sk;
7070 struct sk_buff *skb;
7071
7072 while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL)
7073 kfree_skb(skb);
7074 }
7075#else
7076 int i;
7077
Jens Axboe65e19f52019-10-26 07:20:21 -06007078 for (i = 0; i < ctx->nr_user_files; i++) {
7079 struct file *file;
7080
7081 file = io_file_from_index(ctx, i);
7082 if (file)
7083 fput(file);
7084 }
Jens Axboe6b063142019-01-10 22:13:58 -07007085#endif
Pavel Begunkov08480402021-04-13 02:58:38 +01007086 io_free_file_tables(&ctx->file_table, ctx->nr_user_files);
7087 kfree(ctx->file_data);
7088 ctx->file_data = NULL;
7089 ctx->nr_user_files = 0;
Jens Axboe6b063142019-01-10 22:13:58 -07007090}
7091
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007092static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx)
Pavel Begunkov1642b442020-12-30 21:34:14 +00007093{
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007094 spin_lock_bh(&ctx->rsrc_ref_lock);
Pavel Begunkov1642b442020-12-30 21:34:14 +00007095}
7096
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007097static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx)
Jens Axboe6b063142019-01-10 22:13:58 -07007098{
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007099 spin_unlock_bh(&ctx->rsrc_ref_lock);
7100}
7101
Pavel Begunkov28a9fe22021-04-01 15:43:47 +01007102static void io_rsrc_node_destroy(struct io_rsrc_node *ref_node)
7103{
7104 percpu_ref_exit(&ref_node->refs);
7105 kfree(ref_node);
7106}
7107
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007108static void io_rsrc_node_switch(struct io_ring_ctx *ctx,
7109 struct io_rsrc_data *data_to_kill)
Jens Axboe6b063142019-01-10 22:13:58 -07007110{
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007111 WARN_ON_ONCE(!ctx->rsrc_backup_node);
7112 WARN_ON_ONCE(data_to_kill && !ctx->rsrc_node);
Pavel Begunkov82fbcfa2021-04-01 15:43:43 +01007113
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007114 if (data_to_kill) {
7115 struct io_rsrc_node *rsrc_node = ctx->rsrc_node;
Pavel Begunkov82fbcfa2021-04-01 15:43:43 +01007116
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007117 rsrc_node->rsrc_data = data_to_kill;
7118 io_rsrc_ref_lock(ctx);
7119 list_add_tail(&rsrc_node->node, &ctx->rsrc_ref_list);
7120 io_rsrc_ref_unlock(ctx);
Pavel Begunkov82fbcfa2021-04-01 15:43:43 +01007121
Pavel Begunkov3e942492021-04-11 01:46:34 +01007122 atomic_inc(&data_to_kill->refs);
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007123 percpu_ref_kill(&rsrc_node->refs);
7124 ctx->rsrc_node = NULL;
7125 }
7126
7127 if (!ctx->rsrc_node) {
7128 ctx->rsrc_node = ctx->rsrc_backup_node;
7129 ctx->rsrc_backup_node = NULL;
7130 }
Jens Axboe6b063142019-01-10 22:13:58 -07007131}
7132
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007133static int io_rsrc_node_switch_start(struct io_ring_ctx *ctx)
Pavel Begunkov8dd03af2021-03-19 17:22:36 +00007134{
7135 if (ctx->rsrc_backup_node)
7136 return 0;
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007137 ctx->rsrc_backup_node = io_rsrc_node_alloc(ctx);
Pavel Begunkov8dd03af2021-03-19 17:22:36 +00007138 return ctx->rsrc_backup_node ? 0 : -ENOMEM;
7139}
7140
Pavel Begunkov40ae0ff2021-04-01 15:43:44 +01007141static int io_rsrc_ref_quiesce(struct io_rsrc_data *data, struct io_ring_ctx *ctx)
Hao Xu8bad28d2021-02-19 17:19:36 +08007142{
7143 int ret;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007144
Pavel Begunkov215c3902021-04-01 15:43:48 +01007145 /* As we may drop ->uring_lock, other task may have started quiesce */
Hao Xu8bad28d2021-02-19 17:19:36 +08007146 if (data->quiesce)
7147 return -ENXIO;
7148
7149 data->quiesce = true;
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00007150 do {
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007151 ret = io_rsrc_node_switch_start(ctx);
Pavel Begunkov8dd03af2021-03-19 17:22:36 +00007152 if (ret)
Pavel Begunkovf2303b12021-02-20 18:03:49 +00007153 break;
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007154 io_rsrc_node_switch(ctx, data);
7155
Pavel Begunkov3e942492021-04-11 01:46:34 +01007156 /* kill initial ref, already quiesced if zero */
7157 if (atomic_dec_and_test(&data->refs))
7158 break;
Hao Xu8bad28d2021-02-19 17:19:36 +08007159 flush_delayed_work(&ctx->rsrc_put_work);
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00007160 ret = wait_for_completion_interruptible(&data->done);
7161 if (!ret)
7162 break;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007163
Pavel Begunkov3e942492021-04-11 01:46:34 +01007164 atomic_inc(&data->refs);
7165 /* wait for all works potentially completing data->done */
7166 flush_delayed_work(&ctx->rsrc_put_work);
Jens Axboecb5e1b82021-02-25 07:37:35 -07007167 reinit_completion(&data->done);
Pavel Begunkov8dd03af2021-03-19 17:22:36 +00007168
Hao Xu8bad28d2021-02-19 17:19:36 +08007169 mutex_unlock(&ctx->uring_lock);
7170 ret = io_run_task_work_sig();
7171 mutex_lock(&ctx->uring_lock);
Pavel Begunkovf2303b12021-02-20 18:03:49 +00007172 } while (ret >= 0);
Hao Xu8bad28d2021-02-19 17:19:36 +08007173 data->quiesce = false;
7174
Hao Xu8bad28d2021-02-19 17:19:36 +08007175 return ret;
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007176}
7177
Pavel Begunkov40ae0ff2021-04-01 15:43:44 +01007178static struct io_rsrc_data *io_rsrc_data_alloc(struct io_ring_ctx *ctx,
7179 rsrc_put_fn *do_put)
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007180{
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007181 struct io_rsrc_data *data;
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007182
7183 data = kzalloc(sizeof(*data), GFP_KERNEL);
7184 if (!data)
7185 return NULL;
7186
Pavel Begunkov3e942492021-04-11 01:46:34 +01007187 atomic_set(&data->refs, 1);
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007188 data->ctx = ctx;
Pavel Begunkov40ae0ff2021-04-01 15:43:44 +01007189 data->do_put = do_put;
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007190 init_completion(&data->done);
7191 return data;
7192}
7193
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007194static int io_sqe_files_unregister(struct io_ring_ctx *ctx)
7195{
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007196 int ret;
7197
Pavel Begunkov08480402021-04-13 02:58:38 +01007198 if (!ctx->file_data)
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007199 return -ENXIO;
Pavel Begunkov08480402021-04-13 02:58:38 +01007200 ret = io_rsrc_ref_quiesce(ctx->file_data, ctx);
7201 if (!ret)
7202 __io_sqe_files_unregister(ctx);
7203 return ret;
Jens Axboe6b063142019-01-10 22:13:58 -07007204}
7205
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007206static void io_sq_thread_unpark(struct io_sq_data *sqd)
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007207 __releases(&sqd->lock)
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007208{
Pavel Begunkov521d6a72021-03-11 23:29:38 +00007209 WARN_ON_ONCE(sqd->thread == current);
7210
Pavel Begunkov9e138a42021-03-14 20:57:12 +00007211 /*
7212 * Do the dance but not conditional clear_bit() because it'd race with
7213 * other threads incrementing park_pending and setting the bit.
7214 */
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007215 clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
Pavel Begunkov9e138a42021-03-14 20:57:12 +00007216 if (atomic_dec_return(&sqd->park_pending))
7217 set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007218 mutex_unlock(&sqd->lock);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007219}
7220
Jens Axboe86e0d672021-03-05 08:44:39 -07007221static void io_sq_thread_park(struct io_sq_data *sqd)
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007222 __acquires(&sqd->lock)
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007223{
Pavel Begunkov521d6a72021-03-11 23:29:38 +00007224 WARN_ON_ONCE(sqd->thread == current);
7225
Pavel Begunkov9e138a42021-03-14 20:57:12 +00007226 atomic_inc(&sqd->park_pending);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007227 set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007228 mutex_lock(&sqd->lock);
Jens Axboe05962f92021-03-06 13:58:48 -07007229 if (sqd->thread)
Jens Axboe86e0d672021-03-05 08:44:39 -07007230 wake_up_process(sqd->thread);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007231}
7232
7233static void io_sq_thread_stop(struct io_sq_data *sqd)
7234{
Pavel Begunkov521d6a72021-03-11 23:29:38 +00007235 WARN_ON_ONCE(sqd->thread == current);
Pavel Begunkov88885f62021-04-11 01:46:38 +01007236 WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state));
Pavel Begunkov521d6a72021-03-11 23:29:38 +00007237
Jens Axboe05962f92021-03-06 13:58:48 -07007238 set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state);
Pavel Begunkov88885f62021-04-11 01:46:38 +01007239 mutex_lock(&sqd->lock);
Jens Axboee8f98f242021-03-09 16:32:13 -07007240 if (sqd->thread)
7241 wake_up_process(sqd->thread);
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007242 mutex_unlock(&sqd->lock);
Jens Axboe05962f92021-03-06 13:58:48 -07007243 wait_for_completion(&sqd->exited);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007244}
7245
Jens Axboe534ca6d2020-09-02 13:52:19 -06007246static void io_put_sq_data(struct io_sq_data *sqd)
Jens Axboe6c271ce2019-01-10 11:22:30 -07007247{
Jens Axboe534ca6d2020-09-02 13:52:19 -06007248 if (refcount_dec_and_test(&sqd->refs)) {
Pavel Begunkov9e138a42021-03-14 20:57:12 +00007249 WARN_ON_ONCE(atomic_read(&sqd->park_pending));
7250
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007251 io_sq_thread_stop(sqd);
7252 kfree(sqd);
7253 }
7254}
7255
7256static void io_sq_thread_finish(struct io_ring_ctx *ctx)
7257{
7258 struct io_sq_data *sqd = ctx->sq_data;
7259
7260 if (sqd) {
Jens Axboe05962f92021-03-06 13:58:48 -07007261 io_sq_thread_park(sqd);
Pavel Begunkov521d6a72021-03-11 23:29:38 +00007262 list_del_init(&ctx->sqd_list);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007263 io_sqd_update_thread_idle(sqd);
Jens Axboe05962f92021-03-06 13:58:48 -07007264 io_sq_thread_unpark(sqd);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007265
7266 io_put_sq_data(sqd);
7267 ctx->sq_data = NULL;
Stefan Metzmacher7c30f36a2021-03-07 11:54:28 +01007268 if (ctx->sq_creds)
7269 put_cred(ctx->sq_creds);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007270 }
7271}
7272
Jens Axboeaa061652020-09-02 14:50:27 -06007273static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p)
7274{
7275 struct io_ring_ctx *ctx_attach;
7276 struct io_sq_data *sqd;
7277 struct fd f;
7278
7279 f = fdget(p->wq_fd);
7280 if (!f.file)
7281 return ERR_PTR(-ENXIO);
7282 if (f.file->f_op != &io_uring_fops) {
7283 fdput(f);
7284 return ERR_PTR(-EINVAL);
7285 }
7286
7287 ctx_attach = f.file->private_data;
7288 sqd = ctx_attach->sq_data;
7289 if (!sqd) {
7290 fdput(f);
7291 return ERR_PTR(-EINVAL);
7292 }
Jens Axboe5c2469e2021-03-11 10:17:56 -07007293 if (sqd->task_tgid != current->tgid) {
7294 fdput(f);
7295 return ERR_PTR(-EPERM);
7296 }
Jens Axboeaa061652020-09-02 14:50:27 -06007297
7298 refcount_inc(&sqd->refs);
7299 fdput(f);
7300 return sqd;
7301}
7302
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007303static struct io_sq_data *io_get_sq_data(struct io_uring_params *p,
7304 bool *attached)
Jens Axboe534ca6d2020-09-02 13:52:19 -06007305{
7306 struct io_sq_data *sqd;
7307
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007308 *attached = false;
Jens Axboe5c2469e2021-03-11 10:17:56 -07007309 if (p->flags & IORING_SETUP_ATTACH_WQ) {
7310 sqd = io_attach_sq_data(p);
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007311 if (!IS_ERR(sqd)) {
7312 *attached = true;
Jens Axboe5c2469e2021-03-11 10:17:56 -07007313 return sqd;
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007314 }
Jens Axboe5c2469e2021-03-11 10:17:56 -07007315 /* fall through for EPERM case, setup new sqd/task */
7316 if (PTR_ERR(sqd) != -EPERM)
7317 return sqd;
7318 }
Jens Axboeaa061652020-09-02 14:50:27 -06007319
Jens Axboe534ca6d2020-09-02 13:52:19 -06007320 sqd = kzalloc(sizeof(*sqd), GFP_KERNEL);
7321 if (!sqd)
7322 return ERR_PTR(-ENOMEM);
7323
Pavel Begunkov9e138a42021-03-14 20:57:12 +00007324 atomic_set(&sqd->park_pending, 0);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007325 refcount_set(&sqd->refs, 1);
Jens Axboe69fb2132020-09-14 11:16:23 -06007326 INIT_LIST_HEAD(&sqd->ctx_list);
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007327 mutex_init(&sqd->lock);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007328 init_waitqueue_head(&sqd->wait);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007329 init_completion(&sqd->exited);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007330 return sqd;
7331}
7332
Jens Axboe6b063142019-01-10 22:13:58 -07007333#if defined(CONFIG_UNIX)
Jens Axboe6b063142019-01-10 22:13:58 -07007334/*
7335 * Ensure the UNIX gc is aware of our file set, so we are certain that
7336 * the io_uring can be safely unregistered on process exit, even if we have
7337 * loops in the file referencing.
7338 */
7339static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset)
7340{
7341 struct sock *sk = ctx->ring_sock->sk;
7342 struct scm_fp_list *fpl;
7343 struct sk_buff *skb;
Jens Axboe08a451732019-10-03 08:11:03 -06007344 int i, nr_files;
Jens Axboe6b063142019-01-10 22:13:58 -07007345
Jens Axboe6b063142019-01-10 22:13:58 -07007346 fpl = kzalloc(sizeof(*fpl), GFP_KERNEL);
7347 if (!fpl)
7348 return -ENOMEM;
7349
7350 skb = alloc_skb(0, GFP_KERNEL);
7351 if (!skb) {
7352 kfree(fpl);
7353 return -ENOMEM;
7354 }
7355
7356 skb->sk = sk;
Jens Axboe6b063142019-01-10 22:13:58 -07007357
Jens Axboe08a451732019-10-03 08:11:03 -06007358 nr_files = 0;
Jens Axboe62e398b2021-02-21 16:19:37 -07007359 fpl->user = get_uid(current_user());
Jens Axboe6b063142019-01-10 22:13:58 -07007360 for (i = 0; i < nr; i++) {
Jens Axboe65e19f52019-10-26 07:20:21 -06007361 struct file *file = io_file_from_index(ctx, i + offset);
7362
7363 if (!file)
Jens Axboe08a451732019-10-03 08:11:03 -06007364 continue;
Jens Axboe65e19f52019-10-26 07:20:21 -06007365 fpl->fp[nr_files] = get_file(file);
Jens Axboe08a451732019-10-03 08:11:03 -06007366 unix_inflight(fpl->user, fpl->fp[nr_files]);
7367 nr_files++;
Jens Axboe6b063142019-01-10 22:13:58 -07007368 }
7369
Jens Axboe08a451732019-10-03 08:11:03 -06007370 if (nr_files) {
7371 fpl->max = SCM_MAX_FD;
7372 fpl->count = nr_files;
7373 UNIXCB(skb).fp = fpl;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007374 skb->destructor = unix_destruct_scm;
Jens Axboe08a451732019-10-03 08:11:03 -06007375 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
7376 skb_queue_head(&sk->sk_receive_queue, skb);
Jens Axboe6b063142019-01-10 22:13:58 -07007377
Jens Axboe08a451732019-10-03 08:11:03 -06007378 for (i = 0; i < nr_files; i++)
7379 fput(fpl->fp[i]);
7380 } else {
7381 kfree_skb(skb);
7382 kfree(fpl);
7383 }
Jens Axboe6b063142019-01-10 22:13:58 -07007384
7385 return 0;
7386}
7387
7388/*
7389 * If UNIX sockets are enabled, fd passing can cause a reference cycle which
7390 * causes regular reference counting to break down. We rely on the UNIX
7391 * garbage collection to take care of this problem for us.
7392 */
7393static int io_sqe_files_scm(struct io_ring_ctx *ctx)
7394{
7395 unsigned left, total;
7396 int ret = 0;
7397
7398 total = 0;
7399 left = ctx->nr_user_files;
7400 while (left) {
7401 unsigned this_files = min_t(unsigned, left, SCM_MAX_FD);
Jens Axboe6b063142019-01-10 22:13:58 -07007402
7403 ret = __io_sqe_files_scm(ctx, this_files, total);
7404 if (ret)
7405 break;
7406 left -= this_files;
7407 total += this_files;
7408 }
7409
7410 if (!ret)
7411 return 0;
7412
7413 while (total < ctx->nr_user_files) {
Jens Axboe65e19f52019-10-26 07:20:21 -06007414 struct file *file = io_file_from_index(ctx, total);
7415
7416 if (file)
7417 fput(file);
Jens Axboe6b063142019-01-10 22:13:58 -07007418 total++;
7419 }
7420
7421 return ret;
7422}
7423#else
7424static int io_sqe_files_scm(struct io_ring_ctx *ctx)
7425{
7426 return 0;
7427}
7428#endif
7429
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007430static bool io_alloc_file_tables(struct io_file_table *table, unsigned nr_files)
Jens Axboe65e19f52019-10-26 07:20:21 -06007431{
Pavel Begunkov846a4ef2021-04-01 15:44:03 +01007432 unsigned i, nr_tables = DIV_ROUND_UP(nr_files, IORING_MAX_FILES_TABLE);
7433
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007434 table->files = kcalloc(nr_tables, sizeof(*table->files), GFP_KERNEL);
7435 if (!table->files)
Pavel Begunkov846a4ef2021-04-01 15:44:03 +01007436 return false;
Jens Axboe65e19f52019-10-26 07:20:21 -06007437
7438 for (i = 0; i < nr_tables; i++) {
Pavel Begunkov846a4ef2021-04-01 15:44:03 +01007439 unsigned int this_files = min(nr_files, IORING_MAX_FILES_TABLE);
Jens Axboe65e19f52019-10-26 07:20:21 -06007440
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007441 table->files[i] = kcalloc(this_files, sizeof(*table->files[i]),
Jens Axboe65e19f52019-10-26 07:20:21 -06007442 GFP_KERNEL);
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007443 if (!table->files[i])
Jens Axboe65e19f52019-10-26 07:20:21 -06007444 break;
7445 nr_files -= this_files;
7446 }
7447
7448 if (i == nr_tables)
Pavel Begunkov846a4ef2021-04-01 15:44:03 +01007449 return true;
Jens Axboe65e19f52019-10-26 07:20:21 -06007450
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007451 io_free_file_tables(table, nr_tables * IORING_MAX_FILES_TABLE);
Pavel Begunkov846a4ef2021-04-01 15:44:03 +01007452 return false;
Jens Axboe65e19f52019-10-26 07:20:21 -06007453}
7454
Pavel Begunkov47e90392021-04-01 15:43:56 +01007455static void io_rsrc_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc)
Jens Axboec3a31e62019-10-03 13:59:56 -06007456{
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007457 struct file *file = prsrc->file;
Jens Axboec3a31e62019-10-03 13:59:56 -06007458#if defined(CONFIG_UNIX)
Jens Axboec3a31e62019-10-03 13:59:56 -06007459 struct sock *sock = ctx->ring_sock->sk;
7460 struct sk_buff_head list, *head = &sock->sk_receive_queue;
7461 struct sk_buff *skb;
7462 int i;
7463
7464 __skb_queue_head_init(&list);
7465
7466 /*
7467 * Find the skb that holds this file in its SCM_RIGHTS. When found,
7468 * remove this entry and rearrange the file array.
7469 */
7470 skb = skb_dequeue(head);
7471 while (skb) {
7472 struct scm_fp_list *fp;
7473
7474 fp = UNIXCB(skb).fp;
7475 for (i = 0; i < fp->count; i++) {
7476 int left;
7477
7478 if (fp->fp[i] != file)
7479 continue;
7480
7481 unix_notinflight(fp->user, fp->fp[i]);
7482 left = fp->count - 1 - i;
7483 if (left) {
7484 memmove(&fp->fp[i], &fp->fp[i + 1],
7485 left * sizeof(struct file *));
7486 }
7487 fp->count--;
7488 if (!fp->count) {
7489 kfree_skb(skb);
7490 skb = NULL;
7491 } else {
7492 __skb_queue_tail(&list, skb);
7493 }
7494 fput(file);
7495 file = NULL;
7496 break;
7497 }
7498
7499 if (!file)
7500 break;
7501
7502 __skb_queue_tail(&list, skb);
7503
7504 skb = skb_dequeue(head);
7505 }
7506
7507 if (skb_peek(&list)) {
7508 spin_lock_irq(&head->lock);
7509 while ((skb = __skb_dequeue(&list)) != NULL)
7510 __skb_queue_tail(head, skb);
7511 spin_unlock_irq(&head->lock);
7512 }
7513#else
Jens Axboe05f3fb32019-12-09 11:22:50 -07007514 fput(file);
Jens Axboec3a31e62019-10-03 13:59:56 -06007515#endif
7516}
7517
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007518static void __io_rsrc_put_work(struct io_rsrc_node *ref_node)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007519{
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007520 struct io_rsrc_data *rsrc_data = ref_node->rsrc_data;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007521 struct io_ring_ctx *ctx = rsrc_data->ctx;
7522 struct io_rsrc_put *prsrc, *tmp;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007523
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007524 list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) {
7525 list_del(&prsrc->list);
Pavel Begunkov40ae0ff2021-04-01 15:43:44 +01007526 rsrc_data->do_put(ctx, prsrc);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007527 kfree(prsrc);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007528 }
7529
Pavel Begunkov28a9fe22021-04-01 15:43:47 +01007530 io_rsrc_node_destroy(ref_node);
Pavel Begunkov3e942492021-04-11 01:46:34 +01007531 if (atomic_dec_and_test(&rsrc_data->refs))
7532 complete(&rsrc_data->done);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007533}
7534
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007535static void io_rsrc_put_work(struct work_struct *work)
Jens Axboe4a38aed22020-05-14 17:21:15 -06007536{
7537 struct io_ring_ctx *ctx;
7538 struct llist_node *node;
7539
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007540 ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work);
7541 node = llist_del_all(&ctx->rsrc_put_llist);
Jens Axboe4a38aed22020-05-14 17:21:15 -06007542
7543 while (node) {
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007544 struct io_rsrc_node *ref_node;
Jens Axboe4a38aed22020-05-14 17:21:15 -06007545 struct llist_node *next = node->next;
7546
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007547 ref_node = llist_entry(node, struct io_rsrc_node, llist);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007548 __io_rsrc_put_work(ref_node);
Jens Axboe4a38aed22020-05-14 17:21:15 -06007549 node = next;
7550 }
7551}
7552
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00007553static void io_rsrc_node_ref_zero(struct percpu_ref *ref)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007554{
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007555 struct io_rsrc_node *node = container_of(ref, struct io_rsrc_node, refs);
Pavel Begunkov3e942492021-04-11 01:46:34 +01007556 struct io_ring_ctx *ctx = node->rsrc_data->ctx;
Pavel Begunkove2978222020-11-18 14:56:26 +00007557 bool first_add = false;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007558
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007559 io_rsrc_ref_lock(ctx);
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007560 node->done = true;
Pavel Begunkove2978222020-11-18 14:56:26 +00007561
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00007562 while (!list_empty(&ctx->rsrc_ref_list)) {
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007563 node = list_first_entry(&ctx->rsrc_ref_list,
7564 struct io_rsrc_node, node);
Pavel Begunkove2978222020-11-18 14:56:26 +00007565 /* recycle ref nodes in order */
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007566 if (!node->done)
Pavel Begunkove2978222020-11-18 14:56:26 +00007567 break;
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007568 list_del(&node->node);
7569 first_add |= llist_add(&node->llist, &ctx->rsrc_put_llist);
Pavel Begunkove2978222020-11-18 14:56:26 +00007570 }
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007571 io_rsrc_ref_unlock(ctx);
Pavel Begunkove2978222020-11-18 14:56:26 +00007572
Pavel Begunkov3e942492021-04-11 01:46:34 +01007573 if (first_add)
7574 mod_delayed_work(system_wq, &ctx->rsrc_put_work, HZ);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007575}
7576
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007577static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx)
Xiaoguang Wang05589552020-03-31 14:05:18 +08007578{
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007579 struct io_rsrc_node *ref_node;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007580
7581 ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL);
7582 if (!ref_node)
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007583 return NULL;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007584
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00007585 if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero,
Xiaoguang Wang05589552020-03-31 14:05:18 +08007586 0, GFP_KERNEL)) {
7587 kfree(ref_node);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007588 return NULL;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007589 }
7590 INIT_LIST_HEAD(&ref_node->node);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007591 INIT_LIST_HEAD(&ref_node->rsrc_list);
Pavel Begunkove2978222020-11-18 14:56:26 +00007592 ref_node->done = false;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007593 return ref_node;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007594}
7595
Jens Axboe05f3fb32019-12-09 11:22:50 -07007596static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg,
7597 unsigned nr_args)
7598{
7599 __s32 __user *fds = (__s32 __user *) arg;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007600 struct file *file;
Pavel Begunkovf3baed32021-04-01 15:43:42 +01007601 int fd, ret;
Pavel Begunkov846a4ef2021-04-01 15:44:03 +01007602 unsigned i;
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007603 struct io_rsrc_data *file_data;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007604
7605 if (ctx->file_data)
7606 return -EBUSY;
7607 if (!nr_args)
7608 return -EINVAL;
7609 if (nr_args > IORING_MAX_FIXED_FILES)
7610 return -EMFILE;
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007611 ret = io_rsrc_node_switch_start(ctx);
Pavel Begunkovf3baed32021-04-01 15:43:42 +01007612 if (ret)
7613 return ret;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007614
Pavel Begunkov47e90392021-04-01 15:43:56 +01007615 file_data = io_rsrc_data_alloc(ctx, io_rsrc_file_put);
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007616 if (!file_data)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007617 return -ENOMEM;
Dan Carpenter13770a72021-02-01 15:23:42 +03007618 ctx->file_data = file_data;
Pavel Begunkovf3baed32021-04-01 15:43:42 +01007619 ret = -ENOMEM;
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007620 if (!io_alloc_file_tables(&ctx->file_table, nr_args))
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007621 goto out_free;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007622
Jens Axboe05f3fb32019-12-09 11:22:50 -07007623 for (i = 0; i < nr_args; i++, ctx->nr_user_files++) {
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007624 if (copy_from_user(&fd, &fds[i], sizeof(fd))) {
7625 ret = -EFAULT;
7626 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007627 }
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007628 /* allow sparse sets */
7629 if (fd == -1)
7630 continue;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007631
Jens Axboe05f3fb32019-12-09 11:22:50 -07007632 file = fget(fd);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007633 ret = -EBADF;
7634 if (!file)
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007635 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007636
7637 /*
7638 * Don't allow io_uring instances to be registered. If UNIX
7639 * isn't enabled, then this causes a reference cycle and this
7640 * instance can never get freed. If UNIX is enabled we'll
7641 * handle it just fine, but there's still no point in allowing
7642 * a ring fd as it doesn't support regular read/write anyway.
7643 */
7644 if (file->f_op == &io_uring_fops) {
7645 fput(file);
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007646 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007647 }
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007648 io_fixed_file_set(io_fixed_file_slot(&ctx->file_table, i), file);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007649 }
7650
Jens Axboe05f3fb32019-12-09 11:22:50 -07007651 ret = io_sqe_files_scm(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007652 if (ret) {
Pavel Begunkov08480402021-04-13 02:58:38 +01007653 __io_sqe_files_unregister(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007654 return ret;
7655 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07007656
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007657 io_rsrc_node_switch(ctx, NULL);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007658 return ret;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007659out_fput:
7660 for (i = 0; i < ctx->nr_user_files; i++) {
7661 file = io_file_from_index(ctx, i);
7662 if (file)
7663 fput(file);
7664 }
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007665 io_free_file_tables(&ctx->file_table, nr_args);
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007666 ctx->nr_user_files = 0;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007667out_free:
Pavel Begunkov3e942492021-04-11 01:46:34 +01007668 kfree(ctx->file_data);
Jens Axboe55cbc252020-10-14 07:35:57 -06007669 ctx->file_data = NULL;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007670 return ret;
7671}
7672
Jens Axboec3a31e62019-10-03 13:59:56 -06007673static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file,
7674 int index)
7675{
7676#if defined(CONFIG_UNIX)
7677 struct sock *sock = ctx->ring_sock->sk;
7678 struct sk_buff_head *head = &sock->sk_receive_queue;
7679 struct sk_buff *skb;
7680
7681 /*
7682 * See if we can merge this file into an existing skb SCM_RIGHTS
7683 * file set. If there's no room, fall back to allocating a new skb
7684 * and filling it in.
7685 */
7686 spin_lock_irq(&head->lock);
7687 skb = skb_peek(head);
7688 if (skb) {
7689 struct scm_fp_list *fpl = UNIXCB(skb).fp;
7690
7691 if (fpl->count < SCM_MAX_FD) {
7692 __skb_unlink(skb, head);
7693 spin_unlock_irq(&head->lock);
7694 fpl->fp[fpl->count] = get_file(file);
7695 unix_inflight(fpl->user, fpl->fp[fpl->count]);
7696 fpl->count++;
7697 spin_lock_irq(&head->lock);
7698 __skb_queue_head(head, skb);
7699 } else {
7700 skb = NULL;
7701 }
7702 }
7703 spin_unlock_irq(&head->lock);
7704
7705 if (skb) {
7706 fput(file);
7707 return 0;
7708 }
7709
7710 return __io_sqe_files_scm(ctx, 1, index);
7711#else
7712 return 0;
7713#endif
7714}
7715
Pavel Begunkove7c78372021-04-01 15:43:45 +01007716static int io_queue_rsrc_removal(struct io_rsrc_data *data,
7717 struct io_rsrc_node *node, void *rsrc)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007718{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007719 struct io_rsrc_put *prsrc;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007720
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007721 prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL);
7722 if (!prsrc)
Hillf Dantona5318d32020-03-23 17:47:15 +08007723 return -ENOMEM;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007724
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007725 prsrc->rsrc = rsrc;
Pavel Begunkove7c78372021-04-01 15:43:45 +01007726 list_add(&prsrc->list, &node->rsrc_list);
Hillf Dantona5318d32020-03-23 17:47:15 +08007727 return 0;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007728}
7729
7730static int __io_sqe_files_update(struct io_ring_ctx *ctx,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007731 struct io_uring_rsrc_update *up,
Jens Axboe05f3fb32019-12-09 11:22:50 -07007732 unsigned nr_args)
7733{
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007734 struct io_rsrc_data *data = ctx->file_data;
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01007735 struct io_fixed_file *file_slot;
7736 struct file *file;
Jens Axboec3a31e62019-10-03 13:59:56 -06007737 __s32 __user *fds;
7738 int fd, i, err;
7739 __u32 done;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007740 bool needs_switch = false;
Jens Axboec3a31e62019-10-03 13:59:56 -06007741
Jens Axboe05f3fb32019-12-09 11:22:50 -07007742 if (check_add_overflow(up->offset, nr_args, &done))
Jens Axboec3a31e62019-10-03 13:59:56 -06007743 return -EOVERFLOW;
7744 if (done > ctx->nr_user_files)
7745 return -EINVAL;
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007746 err = io_rsrc_node_switch_start(ctx);
Pavel Begunkov8dd03af2021-03-19 17:22:36 +00007747 if (err)
7748 return err;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007749
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007750 fds = u64_to_user_ptr(up->data);
Pavel Begunkov67973b92021-01-26 13:51:09 +00007751 for (done = 0; done < nr_args; done++) {
Jens Axboec3a31e62019-10-03 13:59:56 -06007752 err = 0;
7753 if (copy_from_user(&fd, &fds[done], sizeof(fd))) {
7754 err = -EFAULT;
7755 break;
7756 }
noah4e0377a2021-01-26 15:23:28 -05007757 if (fd == IORING_REGISTER_FILES_SKIP)
7758 continue;
7759
Pavel Begunkov67973b92021-01-26 13:51:09 +00007760 i = array_index_nospec(up->offset + done, ctx->nr_user_files);
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007761 file_slot = io_fixed_file_slot(&ctx->file_table, i);
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007762
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01007763 if (file_slot->file_ptr) {
7764 file = (struct file *)(file_slot->file_ptr & FFS_MASK);
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007765 err = io_queue_rsrc_removal(data, ctx->rsrc_node, file);
Hillf Dantona5318d32020-03-23 17:47:15 +08007766 if (err)
7767 break;
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01007768 file_slot->file_ptr = 0;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007769 needs_switch = true;
Jens Axboec3a31e62019-10-03 13:59:56 -06007770 }
7771 if (fd != -1) {
Jens Axboec3a31e62019-10-03 13:59:56 -06007772 file = fget(fd);
7773 if (!file) {
7774 err = -EBADF;
7775 break;
7776 }
7777 /*
7778 * Don't allow io_uring instances to be registered. If
7779 * UNIX isn't enabled, then this causes a reference
7780 * cycle and this instance can never get freed. If UNIX
7781 * is enabled we'll handle it just fine, but there's
7782 * still no point in allowing a ring fd as it doesn't
7783 * support regular read/write anyway.
7784 */
7785 if (file->f_op == &io_uring_fops) {
7786 fput(file);
7787 err = -EBADF;
7788 break;
7789 }
Pavel Begunkov9a321c92021-04-01 15:44:01 +01007790 io_fixed_file_set(file_slot, file);
Jens Axboec3a31e62019-10-03 13:59:56 -06007791 err = io_sqe_file_register(ctx, file, i);
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007792 if (err) {
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01007793 file_slot->file_ptr = 0;
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007794 fput(file);
Jens Axboec3a31e62019-10-03 13:59:56 -06007795 break;
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007796 }
Jens Axboec3a31e62019-10-03 13:59:56 -06007797 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07007798 }
7799
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007800 if (needs_switch)
7801 io_rsrc_node_switch(ctx, data);
Jens Axboec3a31e62019-10-03 13:59:56 -06007802 return done ? done : err;
7803}
Xiaoguang Wang05589552020-03-31 14:05:18 +08007804
Jens Axboe05f3fb32019-12-09 11:22:50 -07007805static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg,
7806 unsigned nr_args)
7807{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007808 struct io_uring_rsrc_update up;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007809
7810 if (!ctx->file_data)
7811 return -ENXIO;
7812 if (!nr_args)
7813 return -EINVAL;
7814 if (copy_from_user(&up, arg, sizeof(up)))
7815 return -EFAULT;
7816 if (up.resv)
7817 return -EINVAL;
7818
7819 return __io_sqe_files_update(ctx, &up, nr_args);
7820}
Jens Axboec3a31e62019-10-03 13:59:56 -06007821
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00007822static struct io_wq_work *io_free_work(struct io_wq_work *work)
Jens Axboe7d723062019-11-12 22:31:31 -07007823{
7824 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
7825
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00007826 req = io_put_req_find_next(req);
7827 return req ? &req->work : NULL;
Jens Axboe7d723062019-11-12 22:31:31 -07007828}
7829
Jens Axboe685fe7f2021-03-08 09:37:51 -07007830static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx,
7831 struct task_struct *task)
Pavel Begunkov24369c22020-01-28 03:15:48 +03007832{
Jens Axboee9418942021-02-19 12:33:30 -07007833 struct io_wq_hash *hash;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007834 struct io_wq_data data;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007835 unsigned int concurrency;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007836
Jens Axboee9418942021-02-19 12:33:30 -07007837 hash = ctx->hash_map;
7838 if (!hash) {
7839 hash = kzalloc(sizeof(*hash), GFP_KERNEL);
7840 if (!hash)
7841 return ERR_PTR(-ENOMEM);
7842 refcount_set(&hash->refs, 1);
7843 init_waitqueue_head(&hash->wait);
7844 ctx->hash_map = hash;
7845 }
7846
7847 data.hash = hash;
Jens Axboe685fe7f2021-03-08 09:37:51 -07007848 data.task = task;
Pavel Begunkove9fd9392020-03-04 16:14:12 +03007849 data.free_work = io_free_work;
Pavel Begunkovf5fa38c2020-06-08 21:08:20 +03007850 data.do_work = io_wq_submit_work;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007851
Jens Axboed25e3a32021-02-16 11:41:41 -07007852 /* Do QD, or 4 * CPUS, whatever is smallest */
7853 concurrency = min(ctx->sq_entries, 4 * num_online_cpus());
Pavel Begunkov24369c22020-01-28 03:15:48 +03007854
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007855 return io_wq_create(concurrency, &data);
Pavel Begunkov24369c22020-01-28 03:15:48 +03007856}
7857
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007858static int io_uring_alloc_task_context(struct task_struct *task,
7859 struct io_ring_ctx *ctx)
Jens Axboe0f212202020-09-13 13:09:39 -06007860{
7861 struct io_uring_task *tctx;
Jens Axboed8a6df12020-10-15 16:24:45 -06007862 int ret;
Jens Axboe0f212202020-09-13 13:09:39 -06007863
7864 tctx = kmalloc(sizeof(*tctx), GFP_KERNEL);
7865 if (unlikely(!tctx))
7866 return -ENOMEM;
7867
Jens Axboed8a6df12020-10-15 16:24:45 -06007868 ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL);
7869 if (unlikely(ret)) {
7870 kfree(tctx);
7871 return ret;
7872 }
7873
Jens Axboe685fe7f2021-03-08 09:37:51 -07007874 tctx->io_wq = io_init_wq_offload(ctx, task);
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007875 if (IS_ERR(tctx->io_wq)) {
7876 ret = PTR_ERR(tctx->io_wq);
7877 percpu_counter_destroy(&tctx->inflight);
7878 kfree(tctx);
7879 return ret;
7880 }
7881
Jens Axboe0f212202020-09-13 13:09:39 -06007882 xa_init(&tctx->xa);
7883 init_waitqueue_head(&tctx->wait);
7884 tctx->last = NULL;
Jens Axboefdaf0832020-10-30 09:37:30 -06007885 atomic_set(&tctx->in_idle, 0);
Pavel Begunkovb303fe22021-04-11 01:46:26 +01007886 atomic_set(&tctx->inflight_tracked, 0);
Jens Axboe0f212202020-09-13 13:09:39 -06007887 task->io_uring = tctx;
Jens Axboe7cbf1722021-02-10 00:03:20 +00007888 spin_lock_init(&tctx->task_lock);
7889 INIT_WQ_LIST(&tctx->task_list);
7890 tctx->task_state = 0;
7891 init_task_work(&tctx->task_work, tctx_task_work);
Jens Axboe0f212202020-09-13 13:09:39 -06007892 return 0;
7893}
7894
7895void __io_uring_free(struct task_struct *tsk)
7896{
7897 struct io_uring_task *tctx = tsk->io_uring;
7898
7899 WARN_ON_ONCE(!xa_empty(&tctx->xa));
Pavel Begunkovef8eaa42021-02-27 11:16:45 +00007900 WARN_ON_ONCE(tctx->io_wq);
7901
Jens Axboed8a6df12020-10-15 16:24:45 -06007902 percpu_counter_destroy(&tctx->inflight);
Jens Axboe0f212202020-09-13 13:09:39 -06007903 kfree(tctx);
7904 tsk->io_uring = NULL;
7905}
7906
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02007907static int io_sq_offload_create(struct io_ring_ctx *ctx,
7908 struct io_uring_params *p)
Jens Axboe2b188cc2019-01-07 10:46:33 -07007909{
7910 int ret;
7911
Jens Axboed25e3a32021-02-16 11:41:41 -07007912 /* Retain compatibility with failing for an invalid attach attempt */
7913 if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) ==
7914 IORING_SETUP_ATTACH_WQ) {
7915 struct fd f;
7916
7917 f = fdget(p->wq_fd);
7918 if (!f.file)
7919 return -ENXIO;
7920 if (f.file->f_op != &io_uring_fops) {
7921 fdput(f);
7922 return -EINVAL;
7923 }
7924 fdput(f);
7925 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07007926 if (ctx->flags & IORING_SETUP_SQPOLL) {
Jens Axboe46fe18b2021-03-04 12:39:36 -07007927 struct task_struct *tsk;
Jens Axboe534ca6d2020-09-02 13:52:19 -06007928 struct io_sq_data *sqd;
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007929 bool attached;
Jens Axboe534ca6d2020-09-02 13:52:19 -06007930
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007931 sqd = io_get_sq_data(p, &attached);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007932 if (IS_ERR(sqd)) {
7933 ret = PTR_ERR(sqd);
7934 goto err;
7935 }
Jens Axboe69fb2132020-09-14 11:16:23 -06007936
Stefan Metzmacher7c30f36a2021-03-07 11:54:28 +01007937 ctx->sq_creds = get_current_cred();
Jens Axboe534ca6d2020-09-02 13:52:19 -06007938 ctx->sq_data = sqd;
Jens Axboe917257d2019-04-13 09:28:55 -06007939 ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle);
7940 if (!ctx->sq_thread_idle)
7941 ctx->sq_thread_idle = HZ;
7942
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007943 ret = 0;
Pavel Begunkov78d7f6b2021-03-10 13:13:53 +00007944 io_sq_thread_park(sqd);
Pavel Begunkovde75a3d2021-03-18 11:54:35 +00007945 list_add(&ctx->sqd_list, &sqd->ctx_list);
7946 io_sqd_update_thread_idle(sqd);
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007947 /* don't attach to a dying SQPOLL thread, would be racy */
Pavel Begunkovde75a3d2021-03-18 11:54:35 +00007948 if (attached && !sqd->thread)
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007949 ret = -ENXIO;
Pavel Begunkov78d7f6b2021-03-10 13:13:53 +00007950 io_sq_thread_unpark(sqd);
7951
Pavel Begunkovde75a3d2021-03-18 11:54:35 +00007952 if (ret < 0)
7953 goto err;
7954 if (attached)
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007955 return 0;
Jens Axboeaa061652020-09-02 14:50:27 -06007956
Jens Axboe6c271ce2019-01-10 11:22:30 -07007957 if (p->flags & IORING_SETUP_SQ_AFF) {
Jens Axboe44a9bd12019-05-14 20:00:30 -06007958 int cpu = p->sq_thread_cpu;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007959
Jens Axboe917257d2019-04-13 09:28:55 -06007960 ret = -EINVAL;
Jens Axboe44a9bd12019-05-14 20:00:30 -06007961 if (cpu >= nr_cpu_ids)
Jens Axboee8f98f242021-03-09 16:32:13 -07007962 goto err_sqpoll;
Shenghui Wang7889f442019-05-07 16:03:19 +08007963 if (!cpu_online(cpu))
Jens Axboee8f98f242021-03-09 16:32:13 -07007964 goto err_sqpoll;
Jens Axboe917257d2019-04-13 09:28:55 -06007965
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007966 sqd->sq_cpu = cpu;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007967 } else {
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007968 sqd->sq_cpu = -1;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007969 }
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007970
7971 sqd->task_pid = current->pid;
Jens Axboe5c2469e2021-03-11 10:17:56 -07007972 sqd->task_tgid = current->tgid;
Jens Axboe46fe18b2021-03-04 12:39:36 -07007973 tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE);
7974 if (IS_ERR(tsk)) {
7975 ret = PTR_ERR(tsk);
Jens Axboee8f98f242021-03-09 16:32:13 -07007976 goto err_sqpoll;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007977 }
Pavel Begunkov97a73a02021-03-08 17:30:54 +00007978
Jens Axboe46fe18b2021-03-04 12:39:36 -07007979 sqd->thread = tsk;
Pavel Begunkov97a73a02021-03-08 17:30:54 +00007980 ret = io_uring_alloc_task_context(tsk, ctx);
Jens Axboe46fe18b2021-03-04 12:39:36 -07007981 wake_up_new_task(tsk);
Jens Axboe0f212202020-09-13 13:09:39 -06007982 if (ret)
7983 goto err;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007984 } else if (p->flags & IORING_SETUP_SQ_AFF) {
7985 /* Can't have SQ_AFF without SQPOLL */
7986 ret = -EINVAL;
7987 goto err;
7988 }
7989
Jens Axboe2b188cc2019-01-07 10:46:33 -07007990 return 0;
7991err:
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007992 io_sq_thread_finish(ctx);
Jens Axboe2b188cc2019-01-07 10:46:33 -07007993 return ret;
Jens Axboee8f98f242021-03-09 16:32:13 -07007994err_sqpoll:
7995 complete(&ctx->sq_data->exited);
7996 goto err;
Jens Axboe2b188cc2019-01-07 10:46:33 -07007997}
7998
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007999static inline void __io_unaccount_mem(struct user_struct *user,
8000 unsigned long nr_pages)
Jens Axboe2b188cc2019-01-07 10:46:33 -07008001{
8002 atomic_long_sub(nr_pages, &user->locked_vm);
8003}
8004
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008005static inline int __io_account_mem(struct user_struct *user,
8006 unsigned long nr_pages)
Jens Axboe2b188cc2019-01-07 10:46:33 -07008007{
8008 unsigned long page_limit, cur_pages, new_pages;
8009
8010 /* Don't allow more pages than we can safely lock */
8011 page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
8012
8013 do {
8014 cur_pages = atomic_long_read(&user->locked_vm);
8015 new_pages = cur_pages + nr_pages;
8016 if (new_pages > page_limit)
8017 return -ENOMEM;
8018 } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages,
8019 new_pages) != cur_pages);
8020
8021 return 0;
8022}
8023
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008024static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008025{
Jens Axboe62e398b2021-02-21 16:19:37 -07008026 if (ctx->user)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008027 __io_unaccount_mem(ctx->user, nr_pages);
Bijan Mottahedeh30975822020-06-16 16:36:09 -07008028
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008029 if (ctx->mm_account)
8030 atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm);
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008031}
8032
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008033static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008034{
Bijan Mottahedeh30975822020-06-16 16:36:09 -07008035 int ret;
8036
Jens Axboe62e398b2021-02-21 16:19:37 -07008037 if (ctx->user) {
Bijan Mottahedeh30975822020-06-16 16:36:09 -07008038 ret = __io_account_mem(ctx->user, nr_pages);
8039 if (ret)
8040 return ret;
8041 }
8042
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008043 if (ctx->mm_account)
8044 atomic64_add(nr_pages, &ctx->mm_account->pinned_vm);
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008045
8046 return 0;
8047}
8048
Jens Axboe2b188cc2019-01-07 10:46:33 -07008049static void io_mem_free(void *ptr)
8050{
Mark Rutland52e04ef2019-04-30 17:30:21 +01008051 struct page *page;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008052
Mark Rutland52e04ef2019-04-30 17:30:21 +01008053 if (!ptr)
8054 return;
8055
8056 page = virt_to_head_page(ptr);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008057 if (put_page_testzero(page))
8058 free_compound_page(page);
8059}
8060
8061static void *io_mem_alloc(size_t size)
8062{
8063 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP |
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008064 __GFP_NORETRY | __GFP_ACCOUNT;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008065
8066 return (void *) __get_free_pages(gfp_flags, get_order(size));
8067}
8068
Hristo Venev75b28af2019-08-26 17:23:46 +00008069static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries,
8070 size_t *sq_offset)
8071{
8072 struct io_rings *rings;
8073 size_t off, sq_array_size;
8074
8075 off = struct_size(rings, cqes, cq_entries);
8076 if (off == SIZE_MAX)
8077 return SIZE_MAX;
8078
8079#ifdef CONFIG_SMP
8080 off = ALIGN(off, SMP_CACHE_BYTES);
8081 if (off == 0)
8082 return SIZE_MAX;
8083#endif
8084
Dmitry Vyukovb36200f2020-07-11 11:31:11 +02008085 if (sq_offset)
8086 *sq_offset = off;
8087
Hristo Venev75b28af2019-08-26 17:23:46 +00008088 sq_array_size = array_size(sizeof(u32), sq_entries);
8089 if (sq_array_size == SIZE_MAX)
8090 return SIZE_MAX;
8091
8092 if (check_add_overflow(off, sq_array_size, &off))
8093 return SIZE_MAX;
8094
Hristo Venev75b28af2019-08-26 17:23:46 +00008095 return off;
8096}
8097
Pavel Begunkov7f61a1e92021-04-11 01:46:35 +01008098static void io_buffer_unmap(struct io_ring_ctx *ctx, struct io_mapped_ubuf *imu)
8099{
8100 unsigned int i;
8101
8102 for (i = 0; i < imu->nr_bvecs; i++)
8103 unpin_user_page(imu->bvec[i].bv_page);
8104 if (imu->acct_pages)
8105 io_unaccount_mem(ctx, imu->acct_pages);
8106 kvfree(imu->bvec);
8107 imu->nr_bvecs = 0;
8108}
8109
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008110static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx)
Jens Axboeedafcce2019-01-09 09:16:05 -07008111{
Pavel Begunkov7f61a1e92021-04-11 01:46:35 +01008112 unsigned int i;
Jens Axboeedafcce2019-01-09 09:16:05 -07008113
8114 if (!ctx->user_bufs)
8115 return -ENXIO;
8116
Pavel Begunkov7f61a1e92021-04-11 01:46:35 +01008117 for (i = 0; i < ctx->nr_user_bufs; i++)
8118 io_buffer_unmap(ctx, &ctx->user_bufs[i]);
Jens Axboeedafcce2019-01-09 09:16:05 -07008119 kfree(ctx->user_bufs);
8120 ctx->user_bufs = NULL;
8121 ctx->nr_user_bufs = 0;
8122 return 0;
8123}
8124
8125static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst,
8126 void __user *arg, unsigned index)
8127{
8128 struct iovec __user *src;
8129
8130#ifdef CONFIG_COMPAT
8131 if (ctx->compat) {
8132 struct compat_iovec __user *ciovs;
8133 struct compat_iovec ciov;
8134
8135 ciovs = (struct compat_iovec __user *) arg;
8136 if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov)))
8137 return -EFAULT;
8138
Jens Axboed55e5f52019-12-11 16:12:15 -07008139 dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base);
Jens Axboeedafcce2019-01-09 09:16:05 -07008140 dst->iov_len = ciov.iov_len;
8141 return 0;
8142 }
8143#endif
8144 src = (struct iovec __user *) arg;
8145 if (copy_from_user(dst, &src[index], sizeof(*dst)))
8146 return -EFAULT;
8147 return 0;
8148}
8149
Jens Axboede293932020-09-17 16:19:16 -06008150/*
8151 * Not super efficient, but this is just a registration time. And we do cache
8152 * the last compound head, so generally we'll only do a full search if we don't
8153 * match that one.
8154 *
8155 * We check if the given compound head page has already been accounted, to
8156 * avoid double accounting it. This allows us to account the full size of the
8157 * page, not just the constituent pages of a huge page.
8158 */
8159static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages,
8160 int nr_pages, struct page *hpage)
8161{
8162 int i, j;
8163
8164 /* check current page array */
8165 for (i = 0; i < nr_pages; i++) {
8166 if (!PageCompound(pages[i]))
8167 continue;
8168 if (compound_head(pages[i]) == hpage)
8169 return true;
8170 }
8171
8172 /* check previously registered pages */
8173 for (i = 0; i < ctx->nr_user_bufs; i++) {
8174 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
8175
8176 for (j = 0; j < imu->nr_bvecs; j++) {
8177 if (!PageCompound(imu->bvec[j].bv_page))
8178 continue;
8179 if (compound_head(imu->bvec[j].bv_page) == hpage)
8180 return true;
8181 }
8182 }
8183
8184 return false;
8185}
8186
8187static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages,
8188 int nr_pages, struct io_mapped_ubuf *imu,
8189 struct page **last_hpage)
8190{
8191 int i, ret;
8192
8193 for (i = 0; i < nr_pages; i++) {
8194 if (!PageCompound(pages[i])) {
8195 imu->acct_pages++;
8196 } else {
8197 struct page *hpage;
8198
8199 hpage = compound_head(pages[i]);
8200 if (hpage == *last_hpage)
8201 continue;
8202 *last_hpage = hpage;
8203 if (headpage_already_acct(ctx, pages, i, hpage))
8204 continue;
8205 imu->acct_pages += page_size(hpage) >> PAGE_SHIFT;
8206 }
8207 }
8208
8209 if (!imu->acct_pages)
8210 return 0;
8211
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008212 ret = io_account_mem(ctx, imu->acct_pages);
Jens Axboede293932020-09-17 16:19:16 -06008213 if (ret)
8214 imu->acct_pages = 0;
8215 return ret;
8216}
8217
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008218static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov,
8219 struct io_mapped_ubuf *imu,
8220 struct page **last_hpage)
Jens Axboeedafcce2019-01-09 09:16:05 -07008221{
8222 struct vm_area_struct **vmas = NULL;
8223 struct page **pages = NULL;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008224 unsigned long off, start, end, ubuf;
8225 size_t size;
8226 int ret, pret, nr_pages, i;
Jens Axboeedafcce2019-01-09 09:16:05 -07008227
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008228 ubuf = (unsigned long) iov->iov_base;
8229 end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
8230 start = ubuf >> PAGE_SHIFT;
8231 nr_pages = end - start;
8232
8233 ret = -ENOMEM;
8234
8235 pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL);
8236 if (!pages)
8237 goto done;
8238
8239 vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *),
8240 GFP_KERNEL);
8241 if (!vmas)
8242 goto done;
8243
8244 imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec),
8245 GFP_KERNEL);
8246 if (!imu->bvec)
8247 goto done;
8248
8249 ret = 0;
8250 mmap_read_lock(current->mm);
8251 pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM,
8252 pages, vmas);
8253 if (pret == nr_pages) {
8254 /* don't support file backed memory */
8255 for (i = 0; i < nr_pages; i++) {
8256 struct vm_area_struct *vma = vmas[i];
8257
8258 if (vma->vm_file &&
8259 !is_file_hugepages(vma->vm_file)) {
8260 ret = -EOPNOTSUPP;
8261 break;
8262 }
8263 }
8264 } else {
8265 ret = pret < 0 ? pret : -EFAULT;
8266 }
8267 mmap_read_unlock(current->mm);
8268 if (ret) {
8269 /*
8270 * if we did partial map, or found file backed vmas,
8271 * release any pages we did get
8272 */
8273 if (pret > 0)
8274 unpin_user_pages(pages, pret);
8275 kvfree(imu->bvec);
8276 goto done;
8277 }
8278
8279 ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage);
8280 if (ret) {
8281 unpin_user_pages(pages, pret);
8282 kvfree(imu->bvec);
8283 goto done;
8284 }
8285
8286 off = ubuf & ~PAGE_MASK;
8287 size = iov->iov_len;
8288 for (i = 0; i < nr_pages; i++) {
8289 size_t vec_len;
8290
8291 vec_len = min_t(size_t, size, PAGE_SIZE - off);
8292 imu->bvec[i].bv_page = pages[i];
8293 imu->bvec[i].bv_len = vec_len;
8294 imu->bvec[i].bv_offset = off;
8295 off = 0;
8296 size -= vec_len;
8297 }
8298 /* store original address for later verification */
8299 imu->ubuf = ubuf;
Pavel Begunkov4751f532021-04-01 15:43:55 +01008300 imu->ubuf_end = ubuf + iov->iov_len;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008301 imu->nr_bvecs = nr_pages;
8302 ret = 0;
8303done:
8304 kvfree(pages);
8305 kvfree(vmas);
8306 return ret;
8307}
8308
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008309static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008310{
Pavel Begunkov87094462021-04-11 01:46:36 +01008311 ctx->user_bufs = kcalloc(nr_args, sizeof(*ctx->user_bufs), GFP_KERNEL);
8312 return ctx->user_bufs ? 0 : -ENOMEM;
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008313}
8314
8315static int io_buffer_validate(struct iovec *iov)
8316{
Pavel Begunkov50e96982021-03-24 22:59:01 +00008317 unsigned long tmp, acct_len = iov->iov_len + (PAGE_SIZE - 1);
8318
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008319 /*
8320 * Don't impose further limits on the size and buffer
8321 * constraints here, we'll -EINVAL later when IO is
8322 * submitted if they are wrong.
8323 */
8324 if (!iov->iov_base || !iov->iov_len)
8325 return -EFAULT;
8326
8327 /* arbitrary limit, but we need something */
8328 if (iov->iov_len > SZ_1G)
8329 return -EFAULT;
8330
Pavel Begunkov50e96982021-03-24 22:59:01 +00008331 if (check_add_overflow((unsigned long)iov->iov_base, acct_len, &tmp))
8332 return -EOVERFLOW;
8333
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008334 return 0;
8335}
8336
8337static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg,
8338 unsigned int nr_args)
8339{
8340 int i, ret;
8341 struct iovec iov;
8342 struct page *last_hpage = NULL;
8343
Pavel Begunkov87094462021-04-11 01:46:36 +01008344 if (ctx->user_bufs)
8345 return -EBUSY;
8346 if (!nr_args || nr_args > UIO_MAXIOV)
8347 return -EINVAL;
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008348 ret = io_buffers_map_alloc(ctx, nr_args);
8349 if (ret)
8350 return ret;
8351
Pavel Begunkov87094462021-04-11 01:46:36 +01008352 for (i = 0; i < nr_args; i++, ctx->nr_user_bufs++) {
Jens Axboeedafcce2019-01-09 09:16:05 -07008353 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
Jens Axboeedafcce2019-01-09 09:16:05 -07008354
8355 ret = io_copy_iov(ctx, &iov, arg, i);
8356 if (ret)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008357 break;
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008358 ret = io_buffer_validate(&iov);
8359 if (ret)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008360 break;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008361 ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage);
8362 if (ret)
8363 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07008364 }
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008365
8366 if (ret)
8367 io_sqe_buffers_unregister(ctx);
8368
Jens Axboeedafcce2019-01-09 09:16:05 -07008369 return ret;
8370}
8371
Jens Axboe9b402842019-04-11 11:45:41 -06008372static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg)
8373{
8374 __s32 __user *fds = arg;
8375 int fd;
8376
8377 if (ctx->cq_ev_fd)
8378 return -EBUSY;
8379
8380 if (copy_from_user(&fd, fds, sizeof(*fds)))
8381 return -EFAULT;
8382
8383 ctx->cq_ev_fd = eventfd_ctx_fdget(fd);
8384 if (IS_ERR(ctx->cq_ev_fd)) {
8385 int ret = PTR_ERR(ctx->cq_ev_fd);
8386 ctx->cq_ev_fd = NULL;
8387 return ret;
8388 }
8389
8390 return 0;
8391}
8392
8393static int io_eventfd_unregister(struct io_ring_ctx *ctx)
8394{
8395 if (ctx->cq_ev_fd) {
8396 eventfd_ctx_put(ctx->cq_ev_fd);
8397 ctx->cq_ev_fd = NULL;
8398 return 0;
8399 }
8400
8401 return -ENXIO;
8402}
8403
Jens Axboe5a2e7452020-02-23 16:23:11 -07008404static void io_destroy_buffers(struct io_ring_ctx *ctx)
8405{
Jens Axboe9e15c3a2021-03-13 12:29:43 -07008406 struct io_buffer *buf;
8407 unsigned long index;
8408
8409 xa_for_each(&ctx->io_buffers, index, buf)
8410 __io_remove_buffers(ctx, buf, index, -1U);
Jens Axboe5a2e7452020-02-23 16:23:11 -07008411}
8412
Jens Axboe68e68ee2021-02-13 09:00:02 -07008413static void io_req_cache_free(struct list_head *list, struct task_struct *tsk)
Jens Axboe1b4c3512021-02-10 00:03:19 +00008414{
Jens Axboe68e68ee2021-02-13 09:00:02 -07008415 struct io_kiocb *req, *nxt;
Jens Axboe1b4c3512021-02-10 00:03:19 +00008416
Jens Axboe68e68ee2021-02-13 09:00:02 -07008417 list_for_each_entry_safe(req, nxt, list, compl.list) {
8418 if (tsk && req->task != tsk)
8419 continue;
Jens Axboe1b4c3512021-02-10 00:03:19 +00008420 list_del(&req->compl.list);
8421 kmem_cache_free(req_cachep, req);
8422 }
8423}
8424
Jens Axboe4010fec2021-02-27 15:04:18 -07008425static void io_req_caches_free(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07008426{
Pavel Begunkovbf019da2021-02-10 00:03:17 +00008427 struct io_submit_state *submit_state = &ctx->submit_state;
Pavel Begunkove5547d22021-02-23 22:17:20 +00008428 struct io_comp_state *cs = &ctx->submit_state.comp;
Pavel Begunkovbf019da2021-02-10 00:03:17 +00008429
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008430 mutex_lock(&ctx->uring_lock);
8431
Pavel Begunkov8e5c66c2021-02-22 11:45:55 +00008432 if (submit_state->free_reqs) {
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008433 kmem_cache_free_bulk(req_cachep, submit_state->free_reqs,
8434 submit_state->reqs);
Pavel Begunkov8e5c66c2021-02-22 11:45:55 +00008435 submit_state->free_reqs = 0;
8436 }
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008437
Pavel Begunkovdac7a092021-03-19 17:22:39 +00008438 io_flush_cached_locked_reqs(ctx, cs);
Pavel Begunkove5547d22021-02-23 22:17:20 +00008439 io_req_cache_free(&cs->free_list, NULL);
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008440 mutex_unlock(&ctx->uring_lock);
8441}
8442
Jens Axboe2b188cc2019-01-07 10:46:33 -07008443static void io_ring_ctx_free(struct io_ring_ctx *ctx)
8444{
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008445 io_sq_thread_finish(ctx);
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008446 io_sqe_buffers_unregister(ctx);
Jens Axboe2aede0e2020-09-14 10:45:53 -06008447
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008448 if (ctx->mm_account) {
Jens Axboe2aede0e2020-09-14 10:45:53 -06008449 mmdrop(ctx->mm_account);
8450 ctx->mm_account = NULL;
Bijan Mottahedeh30975822020-06-16 16:36:09 -07008451 }
Jens Axboedef596e2019-01-09 08:59:42 -07008452
Hao Xu8bad28d2021-02-19 17:19:36 +08008453 mutex_lock(&ctx->uring_lock);
Pavel Begunkov08480402021-04-13 02:58:38 +01008454 if (ctx->file_data) {
8455 if (!atomic_dec_and_test(&ctx->file_data->refs))
8456 wait_for_completion(&ctx->file_data->done);
8457 __io_sqe_files_unregister(ctx);
8458 }
Pavel Begunkovc4ea0602021-04-01 15:43:58 +01008459 if (ctx->rings)
8460 __io_cqring_overflow_flush(ctx, true);
Hao Xu8bad28d2021-02-19 17:19:36 +08008461 mutex_unlock(&ctx->uring_lock);
Jens Axboe9b402842019-04-11 11:45:41 -06008462 io_eventfd_unregister(ctx);
Jens Axboe5a2e7452020-02-23 16:23:11 -07008463 io_destroy_buffers(ctx);
Jens Axboedef596e2019-01-09 08:59:42 -07008464
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01008465 /* there are no registered resources left, nobody uses it */
8466 if (ctx->rsrc_node)
8467 io_rsrc_node_destroy(ctx->rsrc_node);
Pavel Begunkov8dd03af2021-03-19 17:22:36 +00008468 if (ctx->rsrc_backup_node)
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01008469 io_rsrc_node_destroy(ctx->rsrc_backup_node);
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01008470 flush_delayed_work(&ctx->rsrc_put_work);
8471
8472 WARN_ON_ONCE(!list_empty(&ctx->rsrc_ref_list));
8473 WARN_ON_ONCE(!llist_empty(&ctx->rsrc_put_llist));
Pavel Begunkov8dd03af2021-03-19 17:22:36 +00008474
Jens Axboe2b188cc2019-01-07 10:46:33 -07008475#if defined(CONFIG_UNIX)
Eric Biggers355e8d262019-06-12 14:58:43 -07008476 if (ctx->ring_sock) {
8477 ctx->ring_sock->file = NULL; /* so that iput() is called */
Jens Axboe2b188cc2019-01-07 10:46:33 -07008478 sock_release(ctx->ring_sock);
Eric Biggers355e8d262019-06-12 14:58:43 -07008479 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07008480#endif
8481
Hristo Venev75b28af2019-08-26 17:23:46 +00008482 io_mem_free(ctx->rings);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008483 io_mem_free(ctx->sq_sqes);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008484
8485 percpu_ref_exit(&ctx->refs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008486 free_uid(ctx->user);
Jens Axboe4010fec2021-02-27 15:04:18 -07008487 io_req_caches_free(ctx);
Jens Axboee9418942021-02-19 12:33:30 -07008488 if (ctx->hash_map)
8489 io_wq_put_hash(ctx->hash_map);
Jens Axboe78076bb2019-12-04 19:56:40 -07008490 kfree(ctx->cancel_hash);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008491 kfree(ctx);
8492}
8493
8494static __poll_t io_uring_poll(struct file *file, poll_table *wait)
8495{
8496 struct io_ring_ctx *ctx = file->private_data;
8497 __poll_t mask = 0;
8498
8499 poll_wait(file, &ctx->cq_wait, wait);
Stefan Bühler4f7067c2019-04-24 23:54:17 +02008500 /*
8501 * synchronizes with barrier from wq_has_sleeper call in
8502 * io_commit_cqring
8503 */
Jens Axboe2b188cc2019-01-07 10:46:33 -07008504 smp_rmb();
Jens Axboe90554202020-09-03 12:12:41 -06008505 if (!io_sqring_full(ctx))
Jens Axboe2b188cc2019-01-07 10:46:33 -07008506 mask |= EPOLLOUT | EPOLLWRNORM;
Hao Xued670c32021-02-05 16:34:21 +08008507
8508 /*
8509 * Don't flush cqring overflow list here, just do a simple check.
8510 * Otherwise there could possible be ABBA deadlock:
8511 * CPU0 CPU1
8512 * ---- ----
8513 * lock(&ctx->uring_lock);
8514 * lock(&ep->mtx);
8515 * lock(&ctx->uring_lock);
8516 * lock(&ep->mtx);
8517 *
8518 * Users may get EPOLLIN meanwhile seeing nothing in cqring, this
8519 * pushs them to do the flush.
8520 */
8521 if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow))
Jens Axboe2b188cc2019-01-07 10:46:33 -07008522 mask |= EPOLLIN | EPOLLRDNORM;
8523
8524 return mask;
8525}
8526
8527static int io_uring_fasync(int fd, struct file *file, int on)
8528{
8529 struct io_ring_ctx *ctx = file->private_data;
8530
8531 return fasync_helper(fd, file, on, &ctx->cq_fasync);
8532}
8533
Yejune Deng0bead8c2020-12-24 11:02:20 +08008534static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id)
Jens Axboe071698e2020-01-28 10:04:42 -07008535{
Jens Axboe4379bf82021-02-15 13:40:22 -07008536 const struct cred *creds;
Jens Axboe071698e2020-01-28 10:04:42 -07008537
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00008538 creds = xa_erase(&ctx->personalities, id);
Jens Axboe4379bf82021-02-15 13:40:22 -07008539 if (creds) {
8540 put_cred(creds);
Yejune Deng0bead8c2020-12-24 11:02:20 +08008541 return 0;
Jens Axboe1e6fa522020-10-15 08:46:24 -06008542 }
Yejune Deng0bead8c2020-12-24 11:02:20 +08008543
8544 return -EINVAL;
8545}
8546
Pavel Begunkov9b465712021-03-15 14:23:07 +00008547static inline bool io_run_ctx_fallback(struct io_ring_ctx *ctx)
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008548{
Pavel Begunkov9b465712021-03-15 14:23:07 +00008549 return io_run_task_work_head(&ctx->exit_task_work);
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008550}
8551
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008552struct io_tctx_exit {
8553 struct callback_head task_work;
8554 struct completion completion;
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00008555 struct io_ring_ctx *ctx;
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008556};
8557
8558static void io_tctx_exit_cb(struct callback_head *cb)
8559{
8560 struct io_uring_task *tctx = current->io_uring;
8561 struct io_tctx_exit *work;
8562
8563 work = container_of(cb, struct io_tctx_exit, task_work);
8564 /*
8565 * When @in_idle, we're in cancellation and it's racy to remove the
8566 * node. It'll be removed by the end of cancellation, just ignore it.
8567 */
8568 if (!atomic_read(&tctx->in_idle))
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00008569 io_uring_del_task_file((unsigned long)work->ctx);
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008570 complete(&work->completion);
8571}
8572
Jens Axboe85faa7b2020-04-09 18:14:00 -06008573static void io_ring_exit_work(struct work_struct *work)
8574{
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008575 struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work);
Pavel Begunkovb5bb3a22021-03-06 11:02:16 +00008576 unsigned long timeout = jiffies + HZ * 60 * 5;
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008577 struct io_tctx_exit exit;
8578 struct io_tctx_node *node;
8579 int ret;
Jens Axboe85faa7b2020-04-09 18:14:00 -06008580
Pavel Begunkova185f1d2021-03-23 10:52:38 +00008581 /* prevent SQPOLL from submitting new requests */
8582 if (ctx->sq_data) {
8583 io_sq_thread_park(ctx->sq_data);
8584 list_del_init(&ctx->sqd_list);
8585 io_sqd_update_thread_idle(ctx->sq_data);
8586 io_sq_thread_unpark(ctx->sq_data);
8587 }
8588
Jens Axboe56952e92020-06-17 15:00:04 -06008589 /*
8590 * If we're doing polled IO and end up having requests being
8591 * submitted async (out-of-line), then completions can come in while
8592 * we're waiting for refs to drop. We need to reap these manually,
8593 * as nobody else will be looking for them.
8594 */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008595 do {
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008596 io_uring_try_cancel_requests(ctx, NULL, NULL);
Pavel Begunkovb5bb3a22021-03-06 11:02:16 +00008597
8598 WARN_ON_ONCE(time_after(jiffies, timeout));
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008599 } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20));
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008600
Pavel Begunkov89b50662021-04-01 15:43:50 +01008601 /*
8602 * Some may use context even when all refs and requests have been put,
8603 * and they are free to do so while still holding uring_lock or
8604 * completion_lock, see __io_req_task_submit(). Apart from other work,
8605 * this lock/unlock section also waits them to finish.
8606 */
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008607 mutex_lock(&ctx->uring_lock);
8608 while (!list_empty(&ctx->tctx_list)) {
Pavel Begunkovb5bb3a22021-03-06 11:02:16 +00008609 WARN_ON_ONCE(time_after(jiffies, timeout));
8610
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008611 node = list_first_entry(&ctx->tctx_list, struct io_tctx_node,
8612 ctx_node);
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00008613 exit.ctx = ctx;
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008614 init_completion(&exit.completion);
8615 init_task_work(&exit.task_work, io_tctx_exit_cb);
8616 ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL);
8617 if (WARN_ON_ONCE(ret))
8618 continue;
8619 wake_up_process(node->task);
8620
8621 mutex_unlock(&ctx->uring_lock);
8622 wait_for_completion(&exit.completion);
8623 cond_resched();
8624 mutex_lock(&ctx->uring_lock);
8625 }
8626 mutex_unlock(&ctx->uring_lock);
Pavel Begunkov89b50662021-04-01 15:43:50 +01008627 spin_lock_irq(&ctx->completion_lock);
8628 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008629
Jens Axboe85faa7b2020-04-09 18:14:00 -06008630 io_ring_ctx_free(ctx);
8631}
8632
Pavel Begunkov80c4cbd2021-03-25 18:32:43 +00008633/* Returns true if we found and killed one or more timeouts */
8634static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk,
8635 struct files_struct *files)
8636{
8637 struct io_kiocb *req, *tmp;
8638 int canceled = 0;
8639
8640 spin_lock_irq(&ctx->completion_lock);
8641 list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) {
8642 if (io_match_task(req, tsk, files)) {
8643 io_kill_timeout(req, -ECANCELED);
8644 canceled++;
8645 }
8646 }
Pavel Begunkov51520422021-03-29 11:39:29 +01008647 if (canceled != 0)
8648 io_commit_cqring(ctx);
Pavel Begunkov80c4cbd2021-03-25 18:32:43 +00008649 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkov80c4cbd2021-03-25 18:32:43 +00008650 if (canceled != 0)
8651 io_cqring_ev_posted(ctx);
8652 return canceled != 0;
8653}
8654
Jens Axboe2b188cc2019-01-07 10:46:33 -07008655static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx)
8656{
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00008657 unsigned long index;
8658 struct creds *creds;
8659
Jens Axboe2b188cc2019-01-07 10:46:33 -07008660 mutex_lock(&ctx->uring_lock);
8661 percpu_ref_kill(&ctx->refs);
Pavel Begunkov634578f2020-12-06 22:22:44 +00008662 if (ctx->rings)
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00008663 __io_cqring_overflow_flush(ctx, true);
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00008664 xa_for_each(&ctx->personalities, index, creds)
8665 io_unregister_personality(ctx, index);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008666 mutex_unlock(&ctx->uring_lock);
8667
Pavel Begunkov6b819282020-11-06 13:00:25 +00008668 io_kill_timeouts(ctx, NULL, NULL);
8669 io_poll_remove_all(ctx, NULL, NULL);
Jens Axboe561fb042019-10-24 07:25:42 -06008670
Jens Axboe15dff282019-11-13 09:09:23 -07008671 /* if we failed setting up the ctx, we might not have any rings */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008672 io_iopoll_try_reap_events(ctx);
Jens Axboe309fc032020-07-10 09:13:34 -06008673
Jens Axboe85faa7b2020-04-09 18:14:00 -06008674 INIT_WORK(&ctx->exit_work, io_ring_exit_work);
Jens Axboefc666772020-08-19 11:10:51 -06008675 /*
8676 * Use system_unbound_wq to avoid spawning tons of event kworkers
8677 * if we're exiting a ton of rings at the same time. It just adds
8678 * noise and overhead, there's no discernable change in runtime
8679 * over using system_wq.
8680 */
8681 queue_work(system_unbound_wq, &ctx->exit_work);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008682}
8683
8684static int io_uring_release(struct inode *inode, struct file *file)
8685{
8686 struct io_ring_ctx *ctx = file->private_data;
8687
8688 file->private_data = NULL;
8689 io_ring_ctx_wait_and_kill(ctx);
8690 return 0;
8691}
8692
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008693struct io_task_cancel {
8694 struct task_struct *task;
8695 struct files_struct *files;
8696};
Pavel Begunkov67c4d9e2020-06-15 10:24:05 +03008697
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008698static bool io_cancel_task_cb(struct io_wq_work *work, void *data)
Jens Axboeb711d4e2020-08-16 08:23:05 -07008699{
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008700 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008701 struct io_task_cancel *cancel = data;
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008702 bool ret;
8703
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008704 if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) {
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008705 unsigned long flags;
8706 struct io_ring_ctx *ctx = req->ctx;
8707
8708 /* protect against races with linked timeouts */
8709 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008710 ret = io_match_task(req, cancel->task, cancel->files);
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008711 spin_unlock_irqrestore(&ctx->completion_lock, flags);
8712 } else {
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008713 ret = io_match_task(req, cancel->task, cancel->files);
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008714 }
8715 return ret;
Jens Axboeb711d4e2020-08-16 08:23:05 -07008716}
8717
Pavel Begunkove1915f72021-03-11 23:29:35 +00008718static bool io_cancel_defer_files(struct io_ring_ctx *ctx,
Pavel Begunkovef9865a2020-11-05 14:06:19 +00008719 struct task_struct *task,
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008720 struct files_struct *files)
8721{
Pavel Begunkove1915f72021-03-11 23:29:35 +00008722 struct io_defer_entry *de;
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008723 LIST_HEAD(list);
8724
8725 spin_lock_irq(&ctx->completion_lock);
8726 list_for_each_entry_reverse(de, &ctx->defer_list, list) {
Pavel Begunkov08d23632020-11-06 13:00:22 +00008727 if (io_match_task(de->req, task, files)) {
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008728 list_cut_position(&list, &ctx->defer_list, &de->list);
8729 break;
8730 }
8731 }
8732 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkove1915f72021-03-11 23:29:35 +00008733 if (list_empty(&list))
8734 return false;
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008735
8736 while (!list_empty(&list)) {
8737 de = list_first_entry(&list, struct io_defer_entry, list);
8738 list_del_init(&de->list);
Pavel Begunkovf41db2732021-02-28 22:35:12 +00008739 io_req_complete_failed(de->req, -ECANCELED);
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008740 kfree(de);
8741 }
Pavel Begunkove1915f72021-03-11 23:29:35 +00008742 return true;
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008743}
8744
Pavel Begunkov1b007642021-03-06 11:02:17 +00008745static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data)
8746{
8747 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
8748
8749 return req->ctx == data;
8750}
8751
8752static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx)
8753{
8754 struct io_tctx_node *node;
8755 enum io_wq_cancel cret;
8756 bool ret = false;
8757
8758 mutex_lock(&ctx->uring_lock);
8759 list_for_each_entry(node, &ctx->tctx_list, ctx_node) {
8760 struct io_uring_task *tctx = node->task->io_uring;
8761
8762 /*
8763 * io_wq will stay alive while we hold uring_lock, because it's
8764 * killed after ctx nodes, which requires to take the lock.
8765 */
8766 if (!tctx || !tctx->io_wq)
8767 continue;
8768 cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true);
8769 ret |= (cret != IO_WQ_CANCEL_NOTFOUND);
8770 }
8771 mutex_unlock(&ctx->uring_lock);
8772
8773 return ret;
8774}
8775
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008776static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
8777 struct task_struct *task,
8778 struct files_struct *files)
8779{
8780 struct io_task_cancel cancel = { .task = task, .files = files, };
Pavel Begunkov1b007642021-03-06 11:02:17 +00008781 struct io_uring_task *tctx = task ? task->io_uring : NULL;
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008782
8783 while (1) {
8784 enum io_wq_cancel cret;
8785 bool ret = false;
8786
Pavel Begunkov1b007642021-03-06 11:02:17 +00008787 if (!task) {
8788 ret |= io_uring_try_cancel_iowq(ctx);
8789 } else if (tctx && tctx->io_wq) {
8790 /*
8791 * Cancels requests of all rings, not only @ctx, but
8792 * it's fine as the task is in exit/exec.
8793 */
Jens Axboe5aa75ed2021-02-16 12:56:50 -07008794 cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb,
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008795 &cancel, true);
8796 ret |= (cret != IO_WQ_CANCEL_NOTFOUND);
8797 }
8798
8799 /* SQPOLL thread does its own polling */
Jens Axboed052d1d2021-03-11 10:49:20 -07008800 if ((!(ctx->flags & IORING_SETUP_SQPOLL) && !files) ||
8801 (ctx->sq_data && ctx->sq_data->thread == current)) {
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008802 while (!list_empty_careful(&ctx->iopoll_list)) {
8803 io_iopoll_try_reap_events(ctx);
8804 ret = true;
8805 }
8806 }
8807
Pavel Begunkove1915f72021-03-11 23:29:35 +00008808 ret |= io_cancel_defer_files(ctx, task, files);
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008809 ret |= io_poll_remove_all(ctx, task, files);
8810 ret |= io_kill_timeouts(ctx, task, files);
8811 ret |= io_run_task_work();
Pavel Begunkovba50a032021-02-26 15:47:56 +00008812 ret |= io_run_ctx_fallback(ctx);
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008813 if (!ret)
8814 break;
8815 cond_resched();
8816 }
8817}
8818
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00008819static int __io_uring_add_task_file(struct io_ring_ctx *ctx)
Jens Axboe0f212202020-09-13 13:09:39 -06008820{
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008821 struct io_uring_task *tctx = current->io_uring;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008822 struct io_tctx_node *node;
Pavel Begunkova528b042020-12-21 18:34:04 +00008823 int ret;
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008824
8825 if (unlikely(!tctx)) {
Jens Axboe5aa75ed2021-02-16 12:56:50 -07008826 ret = io_uring_alloc_task_context(current, ctx);
Jens Axboe0f212202020-09-13 13:09:39 -06008827 if (unlikely(ret))
8828 return ret;
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008829 tctx = current->io_uring;
Jens Axboe0f212202020-09-13 13:09:39 -06008830 }
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00008831 if (!xa_load(&tctx->xa, (unsigned long)ctx)) {
8832 node = kmalloc(sizeof(*node), GFP_KERNEL);
8833 if (!node)
8834 return -ENOMEM;
8835 node->ctx = ctx;
8836 node->task = current;
Jens Axboe0f212202020-09-13 13:09:39 -06008837
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00008838 ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx,
8839 node, GFP_KERNEL));
8840 if (ret) {
8841 kfree(node);
8842 return ret;
Jens Axboe0f212202020-09-13 13:09:39 -06008843 }
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00008844
8845 mutex_lock(&ctx->uring_lock);
8846 list_add(&node->ctx_node, &ctx->tctx_list);
8847 mutex_unlock(&ctx->uring_lock);
Jens Axboe0f212202020-09-13 13:09:39 -06008848 }
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00008849 tctx->last = ctx;
Jens Axboe0f212202020-09-13 13:09:39 -06008850 return 0;
8851}
8852
8853/*
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00008854 * Note that this task has used io_uring. We use it for cancelation purposes.
8855 */
8856static inline int io_uring_add_task_file(struct io_ring_ctx *ctx)
8857{
8858 struct io_uring_task *tctx = current->io_uring;
8859
8860 if (likely(tctx && tctx->last == ctx))
8861 return 0;
8862 return __io_uring_add_task_file(ctx);
8863}
8864
8865/*
Jens Axboe0f212202020-09-13 13:09:39 -06008866 * Remove this io_uring_file -> task mapping.
8867 */
Pavel Begunkov29412672021-03-06 11:02:11 +00008868static void io_uring_del_task_file(unsigned long index)
Jens Axboe0f212202020-09-13 13:09:39 -06008869{
8870 struct io_uring_task *tctx = current->io_uring;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008871 struct io_tctx_node *node;
Pavel Begunkov29412672021-03-06 11:02:11 +00008872
Pavel Begunkoveebd2e32021-03-06 11:02:14 +00008873 if (!tctx)
8874 return;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008875 node = xa_erase(&tctx->xa, index);
8876 if (!node)
Pavel Begunkov29412672021-03-06 11:02:11 +00008877 return;
Jens Axboe0f212202020-09-13 13:09:39 -06008878
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008879 WARN_ON_ONCE(current != node->task);
8880 WARN_ON_ONCE(list_empty(&node->ctx_node));
8881
8882 mutex_lock(&node->ctx->uring_lock);
8883 list_del(&node->ctx_node);
8884 mutex_unlock(&node->ctx->uring_lock);
8885
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00008886 if (tctx->last == node->ctx)
Jens Axboe0f212202020-09-13 13:09:39 -06008887 tctx->last = NULL;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008888 kfree(node);
Jens Axboe0f212202020-09-13 13:09:39 -06008889}
8890
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00008891static void io_uring_clean_tctx(struct io_uring_task *tctx)
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008892{
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008893 struct io_tctx_node *node;
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008894 unsigned long index;
8895
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008896 xa_for_each(&tctx->xa, index, node)
Pavel Begunkov29412672021-03-06 11:02:11 +00008897 io_uring_del_task_file(index);
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00008898 if (tctx->io_wq) {
8899 io_wq_put_and_exit(tctx->io_wq);
8900 tctx->io_wq = NULL;
8901 }
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008902}
8903
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01008904static s64 tctx_inflight(struct io_uring_task *tctx, bool tracked)
Pavel Begunkov368b2082021-04-11 01:46:25 +01008905{
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01008906 if (tracked)
8907 return atomic_read(&tctx->inflight_tracked);
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008908 return percpu_counter_sum(&tctx->inflight);
8909}
8910
8911static void io_sqpoll_cancel_cb(struct callback_head *cb)
8912{
8913 struct io_tctx_exit *work = container_of(cb, struct io_tctx_exit, task_work);
8914 struct io_ring_ctx *ctx = work->ctx;
8915 struct io_sq_data *sqd = ctx->sq_data;
8916
8917 if (sqd->thread)
8918 io_uring_cancel_sqpoll(ctx);
8919 complete(&work->completion);
8920}
8921
8922static void io_sqpoll_cancel_sync(struct io_ring_ctx *ctx)
8923{
8924 struct io_sq_data *sqd = ctx->sq_data;
8925 struct io_tctx_exit work = { .ctx = ctx, };
8926 struct task_struct *task;
8927
8928 io_sq_thread_park(sqd);
8929 list_del_init(&ctx->sqd_list);
8930 io_sqd_update_thread_idle(sqd);
8931 task = sqd->thread;
8932 if (task) {
8933 init_completion(&work.completion);
8934 init_task_work(&work.task_work, io_sqpoll_cancel_cb);
Pavel Begunkovb7f5a0b2021-03-15 14:23:08 +00008935 io_task_work_add_head(&sqd->park_task_work, &work.task_work);
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008936 wake_up_process(task);
8937 }
8938 io_sq_thread_unpark(sqd);
8939
8940 if (task)
8941 wait_for_completion(&work.completion);
8942}
8943
Pavel Begunkov368b2082021-04-11 01:46:25 +01008944static void io_uring_try_cancel(struct files_struct *files)
Jens Axboe0f212202020-09-13 13:09:39 -06008945{
8946 struct io_uring_task *tctx = current->io_uring;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008947 struct io_tctx_node *node;
Matthew Wilcox (Oracle)ce765372020-10-09 13:49:51 +01008948 unsigned long index;
Jens Axboe0f212202020-09-13 13:09:39 -06008949
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008950 xa_for_each(&tctx->xa, index, node) {
8951 struct io_ring_ctx *ctx = node->ctx;
8952
8953 if (ctx->sq_data) {
8954 io_sqpoll_cancel_sync(ctx);
8955 continue;
8956 }
Pavel Begunkov368b2082021-04-11 01:46:25 +01008957 io_uring_try_cancel_requests(ctx, current, files);
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008958 }
Jens Axboefdaf0832020-10-30 09:37:30 -06008959}
8960
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008961/* should only be called by SQPOLL task */
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008962static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx)
8963{
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008964 struct io_sq_data *sqd = ctx->sq_data;
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008965 struct io_uring_task *tctx = current->io_uring;
Jens Axboefdaf0832020-10-30 09:37:30 -06008966 s64 inflight;
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008967 DEFINE_WAIT(wait);
Jens Axboefdaf0832020-10-30 09:37:30 -06008968
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008969 WARN_ON_ONCE(!sqd || ctx->sq_data->thread != current);
8970
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008971 atomic_inc(&tctx->in_idle);
8972 do {
8973 /* read completions before cancelations */
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01008974 inflight = tctx_inflight(tctx, false);
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008975 if (!inflight)
8976 break;
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008977 io_uring_try_cancel_requests(ctx, current, NULL);
Jens Axboefdaf0832020-10-30 09:37:30 -06008978
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008979 prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE);
8980 /*
8981 * If we've seen completions, retry without waiting. This
8982 * avoids a race where a completion comes in before we did
8983 * prepare_to_wait().
8984 */
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01008985 if (inflight == tctx_inflight(tctx, false))
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008986 schedule();
8987 finish_wait(&tctx->wait, &wait);
8988 } while (1);
8989 atomic_dec(&tctx->in_idle);
Jens Axboe0f212202020-09-13 13:09:39 -06008990}
8991
Jens Axboe0f212202020-09-13 13:09:39 -06008992/*
8993 * Find any io_uring fd that this task has registered or done IO on, and cancel
8994 * requests.
8995 */
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01008996void __io_uring_cancel(struct files_struct *files)
Jens Axboe0f212202020-09-13 13:09:39 -06008997{
8998 struct io_uring_task *tctx = current->io_uring;
8999 DEFINE_WAIT(wait);
Jens Axboed8a6df12020-10-15 16:24:45 -06009000 s64 inflight;
Jens Axboe0f212202020-09-13 13:09:39 -06009001
9002 /* make sure overflow events are dropped */
Jens Axboefdaf0832020-10-30 09:37:30 -06009003 atomic_inc(&tctx->in_idle);
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01009004 io_uring_try_cancel(files);
Pavel Begunkov5a978dc2021-03-27 09:59:30 +00009005
Jens Axboed8a6df12020-10-15 16:24:45 -06009006 do {
Jens Axboe0f212202020-09-13 13:09:39 -06009007 /* read completions before cancelations */
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01009008 inflight = tctx_inflight(tctx, !!files);
Jens Axboed8a6df12020-10-15 16:24:45 -06009009 if (!inflight)
9010 break;
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01009011 io_uring_try_cancel(files);
Jens Axboe0f212202020-09-13 13:09:39 -06009012 prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE);
9013
9014 /*
Pavel Begunkova1bb3cd2021-01-26 15:28:26 +00009015 * If we've seen completions, retry without waiting. This
9016 * avoids a race where a completion comes in before we did
9017 * prepare_to_wait().
Jens Axboe0f212202020-09-13 13:09:39 -06009018 */
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01009019 if (inflight == tctx_inflight(tctx, !!files))
Pavel Begunkova1bb3cd2021-01-26 15:28:26 +00009020 schedule();
Pavel Begunkovf57555e2020-12-20 13:21:44 +00009021 finish_wait(&tctx->wait, &wait);
Jens Axboed8a6df12020-10-15 16:24:45 -06009022 } while (1);
Jens Axboefdaf0832020-10-30 09:37:30 -06009023 atomic_dec(&tctx->in_idle);
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00009024
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00009025 io_uring_clean_tctx(tctx);
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01009026 if (!files) {
9027 /* for exec all current's requests should be gone, kill tctx */
9028 __io_uring_free(current);
9029 }
Pavel Begunkov44e728b2020-06-15 10:24:04 +03009030}
9031
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009032static void *io_uring_validate_mmap_request(struct file *file,
9033 loff_t pgoff, size_t sz)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009034{
Jens Axboe2b188cc2019-01-07 10:46:33 -07009035 struct io_ring_ctx *ctx = file->private_data;
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009036 loff_t offset = pgoff << PAGE_SHIFT;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009037 struct page *page;
9038 void *ptr;
9039
9040 switch (offset) {
9041 case IORING_OFF_SQ_RING:
Hristo Venev75b28af2019-08-26 17:23:46 +00009042 case IORING_OFF_CQ_RING:
9043 ptr = ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009044 break;
9045 case IORING_OFF_SQES:
9046 ptr = ctx->sq_sqes;
9047 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009048 default:
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009049 return ERR_PTR(-EINVAL);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009050 }
9051
9052 page = virt_to_head_page(ptr);
Matthew Wilcox (Oracle)a50b8542019-09-23 15:34:25 -07009053 if (sz > page_size(page))
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009054 return ERR_PTR(-EINVAL);
9055
9056 return ptr;
9057}
9058
9059#ifdef CONFIG_MMU
9060
9061static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
9062{
9063 size_t sz = vma->vm_end - vma->vm_start;
9064 unsigned long pfn;
9065 void *ptr;
9066
9067 ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz);
9068 if (IS_ERR(ptr))
9069 return PTR_ERR(ptr);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009070
9071 pfn = virt_to_phys(ptr) >> PAGE_SHIFT;
9072 return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot);
9073}
9074
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009075#else /* !CONFIG_MMU */
9076
9077static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
9078{
9079 return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL;
9080}
9081
9082static unsigned int io_uring_nommu_mmap_capabilities(struct file *file)
9083{
9084 return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE;
9085}
9086
9087static unsigned long io_uring_nommu_get_unmapped_area(struct file *file,
9088 unsigned long addr, unsigned long len,
9089 unsigned long pgoff, unsigned long flags)
9090{
9091 void *ptr;
9092
9093 ptr = io_uring_validate_mmap_request(file, pgoff, len);
9094 if (IS_ERR(ptr))
9095 return PTR_ERR(ptr);
9096
9097 return (unsigned long) ptr;
9098}
9099
9100#endif /* !CONFIG_MMU */
9101
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009102static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx)
Jens Axboe90554202020-09-03 12:12:41 -06009103{
9104 DEFINE_WAIT(wait);
9105
9106 do {
9107 if (!io_sqring_full(ctx))
9108 break;
Jens Axboe90554202020-09-03 12:12:41 -06009109 prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE);
9110
9111 if (!io_sqring_full(ctx))
9112 break;
Jens Axboe90554202020-09-03 12:12:41 -06009113 schedule();
9114 } while (!signal_pending(current));
9115
9116 finish_wait(&ctx->sqo_sq_wait, &wait);
Yang Li51993282021-03-09 14:30:41 +08009117 return 0;
Jens Axboe90554202020-09-03 12:12:41 -06009118}
9119
Hao Xuc73ebb62020-11-03 10:54:37 +08009120static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz,
9121 struct __kernel_timespec __user **ts,
9122 const sigset_t __user **sig)
9123{
9124 struct io_uring_getevents_arg arg;
9125
9126 /*
9127 * If EXT_ARG isn't set, then we have no timespec and the argp pointer
9128 * is just a pointer to the sigset_t.
9129 */
9130 if (!(flags & IORING_ENTER_EXT_ARG)) {
9131 *sig = (const sigset_t __user *) argp;
9132 *ts = NULL;
9133 return 0;
9134 }
9135
9136 /*
9137 * EXT_ARG is set - ensure we agree on the size of it and copy in our
9138 * timespec and sigset_t pointers if good.
9139 */
9140 if (*argsz != sizeof(arg))
9141 return -EINVAL;
9142 if (copy_from_user(&arg, argp, sizeof(arg)))
9143 return -EFAULT;
9144 *sig = u64_to_user_ptr(arg.sigmask);
9145 *argsz = arg.sigmask_sz;
9146 *ts = u64_to_user_ptr(arg.ts);
9147 return 0;
9148}
9149
Jens Axboe2b188cc2019-01-07 10:46:33 -07009150SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
Hao Xuc73ebb62020-11-03 10:54:37 +08009151 u32, min_complete, u32, flags, const void __user *, argp,
9152 size_t, argsz)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009153{
9154 struct io_ring_ctx *ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009155 int submitted = 0;
9156 struct fd f;
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009157 long ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009158
Jens Axboe4c6e2772020-07-01 11:29:10 -06009159 io_run_task_work();
Jens Axboeb41e9852020-02-17 09:52:41 -07009160
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009161 if (unlikely(flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP |
9162 IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG)))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009163 return -EINVAL;
9164
9165 f = fdget(fd);
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009166 if (unlikely(!f.file))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009167 return -EBADF;
9168
9169 ret = -EOPNOTSUPP;
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009170 if (unlikely(f.file->f_op != &io_uring_fops))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009171 goto out_fput;
9172
9173 ret = -ENXIO;
9174 ctx = f.file->private_data;
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009175 if (unlikely(!percpu_ref_tryget(&ctx->refs)))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009176 goto out_fput;
9177
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009178 ret = -EBADFD;
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009179 if (unlikely(ctx->flags & IORING_SETUP_R_DISABLED))
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009180 goto out;
9181
Jens Axboe6c271ce2019-01-10 11:22:30 -07009182 /*
9183 * For SQ polling, the thread will do all submissions and completions.
9184 * Just return the requested submit count, and wake the thread if
9185 * we were asked to.
9186 */
Jens Axboeb2a9ead2019-09-12 14:19:16 -06009187 ret = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07009188 if (ctx->flags & IORING_SETUP_SQPOLL) {
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00009189 io_cqring_overflow_flush(ctx, false);
Pavel Begunkov89448c42020-12-17 00:24:39 +00009190
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009191 ret = -EOWNERDEAD;
Stefan Metzmacher04147482021-03-07 11:54:29 +01009192 if (unlikely(ctx->sq_data->thread == NULL)) {
9193 goto out;
9194 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07009195 if (flags & IORING_ENTER_SQ_WAKEUP)
Jens Axboe534ca6d2020-09-02 13:52:19 -06009196 wake_up(&ctx->sq_data->wait);
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009197 if (flags & IORING_ENTER_SQ_WAIT) {
9198 ret = io_sqpoll_wait_sq(ctx);
9199 if (ret)
9200 goto out;
9201 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07009202 submitted = to_submit;
Jens Axboeb2a9ead2019-09-12 14:19:16 -06009203 } else if (to_submit) {
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00009204 ret = io_uring_add_task_file(ctx);
Jens Axboe0f212202020-09-13 13:09:39 -06009205 if (unlikely(ret))
9206 goto out;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009207 mutex_lock(&ctx->uring_lock);
Jens Axboe0f212202020-09-13 13:09:39 -06009208 submitted = io_submit_sqes(ctx, to_submit);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009209 mutex_unlock(&ctx->uring_lock);
Pavel Begunkov7c504e652019-12-18 19:53:45 +03009210
9211 if (submitted != to_submit)
9212 goto out;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009213 }
9214 if (flags & IORING_ENTER_GETEVENTS) {
Hao Xuc73ebb62020-11-03 10:54:37 +08009215 const sigset_t __user *sig;
9216 struct __kernel_timespec __user *ts;
9217
9218 ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig);
9219 if (unlikely(ret))
9220 goto out;
9221
Jens Axboe2b188cc2019-01-07 10:46:33 -07009222 min_complete = min(min_complete, ctx->cq_entries);
9223
Xiaoguang Wang32b22442020-03-11 09:26:09 +08009224 /*
9225 * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user
9226 * space applications don't need to do io completion events
9227 * polling again, they can rely on io_sq_thread to do polling
9228 * work, which can reduce cpu usage and uring_lock contention.
9229 */
9230 if (ctx->flags & IORING_SETUP_IOPOLL &&
9231 !(ctx->flags & IORING_SETUP_SQPOLL)) {
Pavel Begunkov7668b922020-07-07 16:36:21 +03009232 ret = io_iopoll_check(ctx, min_complete);
Jens Axboedef596e2019-01-09 08:59:42 -07009233 } else {
Hao Xuc73ebb62020-11-03 10:54:37 +08009234 ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts);
Jens Axboedef596e2019-01-09 08:59:42 -07009235 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009236 }
9237
Pavel Begunkov7c504e652019-12-18 19:53:45 +03009238out:
Pavel Begunkov6805b32e2019-10-08 02:18:42 +03009239 percpu_ref_put(&ctx->refs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009240out_fput:
9241 fdput(f);
9242 return submitted ? submitted : ret;
9243}
9244
Tobias Klauserbebdb652020-02-26 18:38:32 +01009245#ifdef CONFIG_PROC_FS
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009246static int io_uring_show_cred(struct seq_file *m, unsigned int id,
9247 const struct cred *cred)
Jens Axboe87ce9552020-01-30 08:25:34 -07009248{
Jens Axboe87ce9552020-01-30 08:25:34 -07009249 struct user_namespace *uns = seq_user_ns(m);
9250 struct group_info *gi;
9251 kernel_cap_t cap;
9252 unsigned __capi;
9253 int g;
9254
9255 seq_printf(m, "%5d\n", id);
9256 seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid));
9257 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid));
9258 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid));
9259 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid));
9260 seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid));
9261 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid));
9262 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid));
9263 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid));
9264 seq_puts(m, "\n\tGroups:\t");
9265 gi = cred->group_info;
9266 for (g = 0; g < gi->ngroups; g++) {
9267 seq_put_decimal_ull(m, g ? " " : "",
9268 from_kgid_munged(uns, gi->gid[g]));
9269 }
9270 seq_puts(m, "\n\tCapEff:\t");
9271 cap = cred->cap_effective;
9272 CAP_FOR_EACH_U32(__capi)
9273 seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8);
9274 seq_putc(m, '\n');
9275 return 0;
9276}
9277
9278static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m)
9279{
Joseph Qidbbe9c62020-09-29 09:01:22 -06009280 struct io_sq_data *sq = NULL;
Jens Axboefad8e0d2020-09-28 08:57:48 -06009281 bool has_lock;
Jens Axboe87ce9552020-01-30 08:25:34 -07009282 int i;
9283
Jens Axboefad8e0d2020-09-28 08:57:48 -06009284 /*
9285 * Avoid ABBA deadlock between the seq lock and the io_uring mutex,
9286 * since fdinfo case grabs it in the opposite direction of normal use
9287 * cases. If we fail to get the lock, we just don't iterate any
9288 * structures that could be going away outside the io_uring mutex.
9289 */
9290 has_lock = mutex_trylock(&ctx->uring_lock);
9291
Jens Axboe5f3f26f2021-02-25 10:17:46 -07009292 if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) {
Joseph Qidbbe9c62020-09-29 09:01:22 -06009293 sq = ctx->sq_data;
Jens Axboe5f3f26f2021-02-25 10:17:46 -07009294 if (!sq->thread)
9295 sq = NULL;
9296 }
Joseph Qidbbe9c62020-09-29 09:01:22 -06009297
9298 seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1);
9299 seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1);
Jens Axboe87ce9552020-01-30 08:25:34 -07009300 seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009301 for (i = 0; has_lock && i < ctx->nr_user_files; i++) {
Jens Axboe7b29f922021-03-12 08:30:14 -07009302 struct file *f = io_file_from_index(ctx, i);
Jens Axboe87ce9552020-01-30 08:25:34 -07009303
Jens Axboe87ce9552020-01-30 08:25:34 -07009304 if (f)
9305 seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname);
9306 else
9307 seq_printf(m, "%5u: <none>\n", i);
9308 }
9309 seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009310 for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) {
Jens Axboe87ce9552020-01-30 08:25:34 -07009311 struct io_mapped_ubuf *buf = &ctx->user_bufs[i];
Pavel Begunkov4751f532021-04-01 15:43:55 +01009312 unsigned int len = buf->ubuf_end - buf->ubuf;
Jens Axboe87ce9552020-01-30 08:25:34 -07009313
Pavel Begunkov4751f532021-04-01 15:43:55 +01009314 seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, len);
Jens Axboe87ce9552020-01-30 08:25:34 -07009315 }
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009316 if (has_lock && !xa_empty(&ctx->personalities)) {
9317 unsigned long index;
9318 const struct cred *cred;
9319
Jens Axboe87ce9552020-01-30 08:25:34 -07009320 seq_printf(m, "Personalities:\n");
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009321 xa_for_each(&ctx->personalities, index, cred)
9322 io_uring_show_cred(m, index, cred);
Jens Axboe87ce9552020-01-30 08:25:34 -07009323 }
Jens Axboed7718a92020-02-14 22:23:12 -07009324 seq_printf(m, "PollList:\n");
9325 spin_lock_irq(&ctx->completion_lock);
9326 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
9327 struct hlist_head *list = &ctx->cancel_hash[i];
9328 struct io_kiocb *req;
9329
9330 hlist_for_each_entry(req, list, hash_node)
9331 seq_printf(m, " op=%d, task_works=%d\n", req->opcode,
9332 req->task->task_works != NULL);
9333 }
9334 spin_unlock_irq(&ctx->completion_lock);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009335 if (has_lock)
9336 mutex_unlock(&ctx->uring_lock);
Jens Axboe87ce9552020-01-30 08:25:34 -07009337}
9338
9339static void io_uring_show_fdinfo(struct seq_file *m, struct file *f)
9340{
9341 struct io_ring_ctx *ctx = f->private_data;
9342
9343 if (percpu_ref_tryget(&ctx->refs)) {
9344 __io_uring_show_fdinfo(ctx, m);
9345 percpu_ref_put(&ctx->refs);
9346 }
9347}
Tobias Klauserbebdb652020-02-26 18:38:32 +01009348#endif
Jens Axboe87ce9552020-01-30 08:25:34 -07009349
Jens Axboe2b188cc2019-01-07 10:46:33 -07009350static const struct file_operations io_uring_fops = {
9351 .release = io_uring_release,
9352 .mmap = io_uring_mmap,
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009353#ifndef CONFIG_MMU
9354 .get_unmapped_area = io_uring_nommu_get_unmapped_area,
9355 .mmap_capabilities = io_uring_nommu_mmap_capabilities,
9356#endif
Jens Axboe2b188cc2019-01-07 10:46:33 -07009357 .poll = io_uring_poll,
9358 .fasync = io_uring_fasync,
Tobias Klauserbebdb652020-02-26 18:38:32 +01009359#ifdef CONFIG_PROC_FS
Jens Axboe87ce9552020-01-30 08:25:34 -07009360 .show_fdinfo = io_uring_show_fdinfo,
Tobias Klauserbebdb652020-02-26 18:38:32 +01009361#endif
Jens Axboe2b188cc2019-01-07 10:46:33 -07009362};
9363
9364static int io_allocate_scq_urings(struct io_ring_ctx *ctx,
9365 struct io_uring_params *p)
9366{
Hristo Venev75b28af2019-08-26 17:23:46 +00009367 struct io_rings *rings;
9368 size_t size, sq_array_offset;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009369
Jens Axboebd740482020-08-05 12:58:23 -06009370 /* make sure these are sane, as we already accounted them */
9371 ctx->sq_entries = p->sq_entries;
9372 ctx->cq_entries = p->cq_entries;
9373
Hristo Venev75b28af2019-08-26 17:23:46 +00009374 size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset);
9375 if (size == SIZE_MAX)
9376 return -EOVERFLOW;
9377
9378 rings = io_mem_alloc(size);
9379 if (!rings)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009380 return -ENOMEM;
9381
Hristo Venev75b28af2019-08-26 17:23:46 +00009382 ctx->rings = rings;
9383 ctx->sq_array = (u32 *)((char *)rings + sq_array_offset);
9384 rings->sq_ring_mask = p->sq_entries - 1;
9385 rings->cq_ring_mask = p->cq_entries - 1;
9386 rings->sq_ring_entries = p->sq_entries;
9387 rings->cq_ring_entries = p->cq_entries;
9388 ctx->sq_mask = rings->sq_ring_mask;
9389 ctx->cq_mask = rings->cq_ring_mask;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009390
9391 size = array_size(sizeof(struct io_uring_sqe), p->sq_entries);
Jens Axboeeb065d32019-11-20 09:26:29 -07009392 if (size == SIZE_MAX) {
9393 io_mem_free(ctx->rings);
9394 ctx->rings = NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009395 return -EOVERFLOW;
Jens Axboeeb065d32019-11-20 09:26:29 -07009396 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009397
9398 ctx->sq_sqes = io_mem_alloc(size);
Jens Axboeeb065d32019-11-20 09:26:29 -07009399 if (!ctx->sq_sqes) {
9400 io_mem_free(ctx->rings);
9401 ctx->rings = NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009402 return -ENOMEM;
Jens Axboeeb065d32019-11-20 09:26:29 -07009403 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009404
Jens Axboe2b188cc2019-01-07 10:46:33 -07009405 return 0;
9406}
9407
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009408static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file)
9409{
9410 int ret, fd;
9411
9412 fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
9413 if (fd < 0)
9414 return fd;
9415
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00009416 ret = io_uring_add_task_file(ctx);
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009417 if (ret) {
9418 put_unused_fd(fd);
9419 return ret;
9420 }
9421 fd_install(fd, file);
9422 return fd;
9423}
9424
Jens Axboe2b188cc2019-01-07 10:46:33 -07009425/*
9426 * Allocate an anonymous fd, this is what constitutes the application
9427 * visible backing of an io_uring instance. The application mmaps this
9428 * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled,
9429 * we have to tie this fd to a socket for file garbage collection purposes.
9430 */
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009431static struct file *io_uring_get_file(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009432{
9433 struct file *file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009434#if defined(CONFIG_UNIX)
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009435 int ret;
9436
Jens Axboe2b188cc2019-01-07 10:46:33 -07009437 ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP,
9438 &ctx->ring_sock);
9439 if (ret)
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009440 return ERR_PTR(ret);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009441#endif
9442
Jens Axboe2b188cc2019-01-07 10:46:33 -07009443 file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx,
9444 O_RDWR | O_CLOEXEC);
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009445#if defined(CONFIG_UNIX)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009446 if (IS_ERR(file)) {
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009447 sock_release(ctx->ring_sock);
9448 ctx->ring_sock = NULL;
9449 } else {
9450 ctx->ring_sock->file = file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009451 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009452#endif
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009453 return file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009454}
9455
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009456static int io_uring_create(unsigned entries, struct io_uring_params *p,
9457 struct io_uring_params __user *params)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009458{
Jens Axboe2b188cc2019-01-07 10:46:33 -07009459 struct io_ring_ctx *ctx;
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009460 struct file *file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009461 int ret;
9462
Jens Axboe8110c1a2019-12-28 15:39:54 -07009463 if (!entries)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009464 return -EINVAL;
Jens Axboe8110c1a2019-12-28 15:39:54 -07009465 if (entries > IORING_MAX_ENTRIES) {
9466 if (!(p->flags & IORING_SETUP_CLAMP))
9467 return -EINVAL;
9468 entries = IORING_MAX_ENTRIES;
9469 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009470
9471 /*
9472 * Use twice as many entries for the CQ ring. It's possible for the
9473 * application to drive a higher depth than the size of the SQ ring,
9474 * since the sqes are only used at submission time. This allows for
Jens Axboe33a107f2019-10-04 12:10:03 -06009475 * some flexibility in overcommitting a bit. If the application has
9476 * set IORING_SETUP_CQSIZE, it will have passed in the desired number
9477 * of CQ ring entries manually.
Jens Axboe2b188cc2019-01-07 10:46:33 -07009478 */
9479 p->sq_entries = roundup_pow_of_two(entries);
Jens Axboe33a107f2019-10-04 12:10:03 -06009480 if (p->flags & IORING_SETUP_CQSIZE) {
9481 /*
9482 * If IORING_SETUP_CQSIZE is set, we do the same roundup
9483 * to a power-of-two, if it isn't already. We do NOT impose
9484 * any cq vs sq ring sizing.
9485 */
Joseph Qieb2667b32020-11-24 15:03:03 +08009486 if (!p->cq_entries)
Jens Axboe33a107f2019-10-04 12:10:03 -06009487 return -EINVAL;
Jens Axboe8110c1a2019-12-28 15:39:54 -07009488 if (p->cq_entries > IORING_MAX_CQ_ENTRIES) {
9489 if (!(p->flags & IORING_SETUP_CLAMP))
9490 return -EINVAL;
9491 p->cq_entries = IORING_MAX_CQ_ENTRIES;
9492 }
Joseph Qieb2667b32020-11-24 15:03:03 +08009493 p->cq_entries = roundup_pow_of_two(p->cq_entries);
9494 if (p->cq_entries < p->sq_entries)
9495 return -EINVAL;
Jens Axboe33a107f2019-10-04 12:10:03 -06009496 } else {
9497 p->cq_entries = 2 * p->sq_entries;
9498 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009499
Jens Axboe2b188cc2019-01-07 10:46:33 -07009500 ctx = io_ring_ctx_alloc(p);
Jens Axboe62e398b2021-02-21 16:19:37 -07009501 if (!ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009502 return -ENOMEM;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009503 ctx->compat = in_compat_syscall();
Jens Axboe62e398b2021-02-21 16:19:37 -07009504 if (!capable(CAP_IPC_LOCK))
9505 ctx->user = get_uid(current_user());
Jens Axboe2aede0e2020-09-14 10:45:53 -06009506
9507 /*
9508 * This is just grabbed for accounting purposes. When a process exits,
9509 * the mm is exited and dropped before the files, hence we need to hang
9510 * on to this mm purely for the purposes of being able to unaccount
9511 * memory (locked/pinned vm). It's not used for anything else.
9512 */
Jens Axboe6b7898e2020-08-25 07:58:00 -06009513 mmgrab(current->mm);
Jens Axboe2aede0e2020-09-14 10:45:53 -06009514 ctx->mm_account = current->mm;
Jens Axboe6b7898e2020-08-25 07:58:00 -06009515
Jens Axboe2b188cc2019-01-07 10:46:33 -07009516 ret = io_allocate_scq_urings(ctx, p);
9517 if (ret)
9518 goto err;
9519
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009520 ret = io_sq_offload_create(ctx, p);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009521 if (ret)
9522 goto err;
9523
Jens Axboe2b188cc2019-01-07 10:46:33 -07009524 memset(&p->sq_off, 0, sizeof(p->sq_off));
Hristo Venev75b28af2019-08-26 17:23:46 +00009525 p->sq_off.head = offsetof(struct io_rings, sq.head);
9526 p->sq_off.tail = offsetof(struct io_rings, sq.tail);
9527 p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask);
9528 p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries);
9529 p->sq_off.flags = offsetof(struct io_rings, sq_flags);
9530 p->sq_off.dropped = offsetof(struct io_rings, sq_dropped);
9531 p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009532
9533 memset(&p->cq_off, 0, sizeof(p->cq_off));
Hristo Venev75b28af2019-08-26 17:23:46 +00009534 p->cq_off.head = offsetof(struct io_rings, cq.head);
9535 p->cq_off.tail = offsetof(struct io_rings, cq.tail);
9536 p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask);
9537 p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries);
9538 p->cq_off.overflow = offsetof(struct io_rings, cq_overflow);
9539 p->cq_off.cqes = offsetof(struct io_rings, cqes);
Stefano Garzarella0d9b5b32020-05-15 18:38:04 +02009540 p->cq_off.flags = offsetof(struct io_rings, cq_flags);
Jens Axboeac90f242019-09-06 10:26:21 -06009541
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009542 p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP |
9543 IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS |
Jiufei Xue5769a352020-06-17 17:53:55 +08009544 IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL |
Hao Xuc73ebb62020-11-03 10:54:37 +08009545 IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED |
Jens Axboe1c0aa1fa2021-02-20 11:55:28 -07009546 IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS;
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009547
9548 if (copy_to_user(params, p, sizeof(*p))) {
9549 ret = -EFAULT;
9550 goto err;
9551 }
Jens Axboed1719f72020-07-30 13:43:53 -06009552
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009553 file = io_uring_get_file(ctx);
9554 if (IS_ERR(file)) {
9555 ret = PTR_ERR(file);
9556 goto err;
9557 }
9558
Jens Axboed1719f72020-07-30 13:43:53 -06009559 /*
Jens Axboe044c1ab2019-10-28 09:15:33 -06009560 * Install ring fd as the very last thing, so we don't risk someone
9561 * having closed it before we finish setup
9562 */
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009563 ret = io_uring_install_fd(ctx, file);
9564 if (ret < 0) {
9565 /* fput will clean it up */
9566 fput(file);
9567 return ret;
9568 }
Jens Axboe044c1ab2019-10-28 09:15:33 -06009569
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02009570 trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009571 return ret;
9572err:
9573 io_ring_ctx_wait_and_kill(ctx);
9574 return ret;
9575}
9576
9577/*
9578 * Sets up an aio uring context, and returns the fd. Applications asks for a
9579 * ring size, we return the actual sq/cq ring sizes (among other things) in the
9580 * params structure passed in.
9581 */
9582static long io_uring_setup(u32 entries, struct io_uring_params __user *params)
9583{
9584 struct io_uring_params p;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009585 int i;
9586
9587 if (copy_from_user(&p, params, sizeof(p)))
9588 return -EFAULT;
9589 for (i = 0; i < ARRAY_SIZE(p.resv); i++) {
9590 if (p.resv[i])
9591 return -EINVAL;
9592 }
9593
Jens Axboe6c271ce2019-01-10 11:22:30 -07009594 if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL |
Jens Axboe8110c1a2019-12-28 15:39:54 -07009595 IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE |
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009596 IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ |
9597 IORING_SETUP_R_DISABLED))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009598 return -EINVAL;
9599
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009600 return io_uring_create(entries, &p, params);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009601}
9602
9603SYSCALL_DEFINE2(io_uring_setup, u32, entries,
9604 struct io_uring_params __user *, params)
9605{
9606 return io_uring_setup(entries, params);
9607}
9608
Jens Axboe66f4af92020-01-16 15:36:52 -07009609static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args)
9610{
9611 struct io_uring_probe *p;
9612 size_t size;
9613 int i, ret;
9614
9615 size = struct_size(p, ops, nr_args);
9616 if (size == SIZE_MAX)
9617 return -EOVERFLOW;
9618 p = kzalloc(size, GFP_KERNEL);
9619 if (!p)
9620 return -ENOMEM;
9621
9622 ret = -EFAULT;
9623 if (copy_from_user(p, arg, size))
9624 goto out;
9625 ret = -EINVAL;
9626 if (memchr_inv(p, 0, size))
9627 goto out;
9628
9629 p->last_op = IORING_OP_LAST - 1;
9630 if (nr_args > IORING_OP_LAST)
9631 nr_args = IORING_OP_LAST;
9632
9633 for (i = 0; i < nr_args; i++) {
9634 p->ops[i].op = i;
9635 if (!io_op_defs[i].not_supported)
9636 p->ops[i].flags = IO_URING_OP_SUPPORTED;
9637 }
9638 p->ops_len = i;
9639
9640 ret = 0;
9641 if (copy_to_user(arg, p, size))
9642 ret = -EFAULT;
9643out:
9644 kfree(p);
9645 return ret;
9646}
9647
Jens Axboe071698e2020-01-28 10:04:42 -07009648static int io_register_personality(struct io_ring_ctx *ctx)
9649{
Jens Axboe4379bf82021-02-15 13:40:22 -07009650 const struct cred *creds;
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009651 u32 id;
Jens Axboe1e6fa522020-10-15 08:46:24 -06009652 int ret;
Jens Axboe071698e2020-01-28 10:04:42 -07009653
Jens Axboe4379bf82021-02-15 13:40:22 -07009654 creds = get_current_cred();
Jens Axboe1e6fa522020-10-15 08:46:24 -06009655
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009656 ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds,
9657 XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL);
9658 if (!ret)
9659 return id;
9660 put_cred(creds);
Jens Axboe1e6fa522020-10-15 08:46:24 -06009661 return ret;
Jens Axboe071698e2020-01-28 10:04:42 -07009662}
9663
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009664static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg,
9665 unsigned int nr_args)
9666{
9667 struct io_uring_restriction *res;
9668 size_t size;
9669 int i, ret;
9670
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009671 /* Restrictions allowed only if rings started disabled */
9672 if (!(ctx->flags & IORING_SETUP_R_DISABLED))
9673 return -EBADFD;
9674
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009675 /* We allow only a single restrictions registration */
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009676 if (ctx->restrictions.registered)
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009677 return -EBUSY;
9678
9679 if (!arg || nr_args > IORING_MAX_RESTRICTIONS)
9680 return -EINVAL;
9681
9682 size = array_size(nr_args, sizeof(*res));
9683 if (size == SIZE_MAX)
9684 return -EOVERFLOW;
9685
9686 res = memdup_user(arg, size);
9687 if (IS_ERR(res))
9688 return PTR_ERR(res);
9689
9690 ret = 0;
9691
9692 for (i = 0; i < nr_args; i++) {
9693 switch (res[i].opcode) {
9694 case IORING_RESTRICTION_REGISTER_OP:
9695 if (res[i].register_op >= IORING_REGISTER_LAST) {
9696 ret = -EINVAL;
9697 goto out;
9698 }
9699
9700 __set_bit(res[i].register_op,
9701 ctx->restrictions.register_op);
9702 break;
9703 case IORING_RESTRICTION_SQE_OP:
9704 if (res[i].sqe_op >= IORING_OP_LAST) {
9705 ret = -EINVAL;
9706 goto out;
9707 }
9708
9709 __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op);
9710 break;
9711 case IORING_RESTRICTION_SQE_FLAGS_ALLOWED:
9712 ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags;
9713 break;
9714 case IORING_RESTRICTION_SQE_FLAGS_REQUIRED:
9715 ctx->restrictions.sqe_flags_required = res[i].sqe_flags;
9716 break;
9717 default:
9718 ret = -EINVAL;
9719 goto out;
9720 }
9721 }
9722
9723out:
9724 /* Reset all restrictions if an error happened */
9725 if (ret != 0)
9726 memset(&ctx->restrictions, 0, sizeof(ctx->restrictions));
9727 else
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009728 ctx->restrictions.registered = true;
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009729
9730 kfree(res);
9731 return ret;
9732}
9733
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009734static int io_register_enable_rings(struct io_ring_ctx *ctx)
9735{
9736 if (!(ctx->flags & IORING_SETUP_R_DISABLED))
9737 return -EBADFD;
9738
9739 if (ctx->restrictions.registered)
9740 ctx->restricted = 1;
9741
Pavel Begunkov0298ef92021-03-08 13:20:57 +00009742 ctx->flags &= ~IORING_SETUP_R_DISABLED;
9743 if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait))
9744 wake_up(&ctx->sq_data->wait);
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009745 return 0;
9746}
9747
Jens Axboe071698e2020-01-28 10:04:42 -07009748static bool io_register_op_must_quiesce(int op)
9749{
9750 switch (op) {
Pavel Begunkovf4f7d212021-04-01 15:44:02 +01009751 case IORING_REGISTER_FILES:
Jens Axboe071698e2020-01-28 10:04:42 -07009752 case IORING_UNREGISTER_FILES:
9753 case IORING_REGISTER_FILES_UPDATE:
9754 case IORING_REGISTER_PROBE:
9755 case IORING_REGISTER_PERSONALITY:
9756 case IORING_UNREGISTER_PERSONALITY:
9757 return false;
9758 default:
9759 return true;
9760 }
9761}
9762
Jens Axboeedafcce2019-01-09 09:16:05 -07009763static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode,
9764 void __user *arg, unsigned nr_args)
Jens Axboeb19062a2019-04-15 10:49:38 -06009765 __releases(ctx->uring_lock)
9766 __acquires(ctx->uring_lock)
Jens Axboeedafcce2019-01-09 09:16:05 -07009767{
9768 int ret;
9769
Jens Axboe35fa71a2019-04-22 10:23:23 -06009770 /*
9771 * We're inside the ring mutex, if the ref is already dying, then
9772 * someone else killed the ctx or is already going through
9773 * io_uring_register().
9774 */
9775 if (percpu_ref_is_dying(&ctx->refs))
9776 return -ENXIO;
9777
Jens Axboe071698e2020-01-28 10:04:42 -07009778 if (io_register_op_must_quiesce(opcode)) {
Jens Axboe05f3fb32019-12-09 11:22:50 -07009779 percpu_ref_kill(&ctx->refs);
Jens Axboeb19062a2019-04-15 10:49:38 -06009780
Jens Axboe05f3fb32019-12-09 11:22:50 -07009781 /*
9782 * Drop uring mutex before waiting for references to exit. If
9783 * another thread is currently inside io_uring_enter() it might
9784 * need to grab the uring_lock to make progress. If we hold it
9785 * here across the drain wait, then we can deadlock. It's safe
9786 * to drop the mutex here, since no new references will come in
9787 * after we've killed the percpu ref.
9788 */
9789 mutex_unlock(&ctx->uring_lock);
Jens Axboeaf9c1a42020-09-24 13:32:18 -06009790 do {
9791 ret = wait_for_completion_interruptible(&ctx->ref_comp);
9792 if (!ret)
9793 break;
Jens Axboeed6930c2020-10-08 19:09:46 -06009794 ret = io_run_task_work_sig();
9795 if (ret < 0)
9796 break;
Jens Axboeaf9c1a42020-09-24 13:32:18 -06009797 } while (1);
Jens Axboe05f3fb32019-12-09 11:22:50 -07009798 mutex_lock(&ctx->uring_lock);
Jens Axboeaf9c1a42020-09-24 13:32:18 -06009799
Jens Axboec1503682020-01-08 08:26:07 -07009800 if (ret) {
Pavel Begunkovf70865d2021-04-11 01:46:40 +01009801 io_refs_resurrect(&ctx->refs, &ctx->ref_comp);
9802 return ret;
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009803 }
9804 }
9805
9806 if (ctx->restricted) {
9807 if (opcode >= IORING_REGISTER_LAST) {
9808 ret = -EINVAL;
9809 goto out;
9810 }
9811
9812 if (!test_bit(opcode, ctx->restrictions.register_op)) {
9813 ret = -EACCES;
Jens Axboec1503682020-01-08 08:26:07 -07009814 goto out;
9815 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07009816 }
Jens Axboeedafcce2019-01-09 09:16:05 -07009817
9818 switch (opcode) {
9819 case IORING_REGISTER_BUFFERS:
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08009820 ret = io_sqe_buffers_register(ctx, arg, nr_args);
Jens Axboeedafcce2019-01-09 09:16:05 -07009821 break;
9822 case IORING_UNREGISTER_BUFFERS:
9823 ret = -EINVAL;
9824 if (arg || nr_args)
9825 break;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08009826 ret = io_sqe_buffers_unregister(ctx);
Jens Axboeedafcce2019-01-09 09:16:05 -07009827 break;
Jens Axboe6b063142019-01-10 22:13:58 -07009828 case IORING_REGISTER_FILES:
9829 ret = io_sqe_files_register(ctx, arg, nr_args);
9830 break;
9831 case IORING_UNREGISTER_FILES:
9832 ret = -EINVAL;
9833 if (arg || nr_args)
9834 break;
9835 ret = io_sqe_files_unregister(ctx);
9836 break;
Jens Axboec3a31e62019-10-03 13:59:56 -06009837 case IORING_REGISTER_FILES_UPDATE:
9838 ret = io_sqe_files_update(ctx, arg, nr_args);
9839 break;
Jens Axboe9b402842019-04-11 11:45:41 -06009840 case IORING_REGISTER_EVENTFD:
Jens Axboef2842ab2020-01-08 11:04:00 -07009841 case IORING_REGISTER_EVENTFD_ASYNC:
Jens Axboe9b402842019-04-11 11:45:41 -06009842 ret = -EINVAL;
9843 if (nr_args != 1)
9844 break;
9845 ret = io_eventfd_register(ctx, arg);
Jens Axboef2842ab2020-01-08 11:04:00 -07009846 if (ret)
9847 break;
9848 if (opcode == IORING_REGISTER_EVENTFD_ASYNC)
9849 ctx->eventfd_async = 1;
9850 else
9851 ctx->eventfd_async = 0;
Jens Axboe9b402842019-04-11 11:45:41 -06009852 break;
9853 case IORING_UNREGISTER_EVENTFD:
9854 ret = -EINVAL;
9855 if (arg || nr_args)
9856 break;
9857 ret = io_eventfd_unregister(ctx);
9858 break;
Jens Axboe66f4af92020-01-16 15:36:52 -07009859 case IORING_REGISTER_PROBE:
9860 ret = -EINVAL;
9861 if (!arg || nr_args > 256)
9862 break;
9863 ret = io_probe(ctx, arg, nr_args);
9864 break;
Jens Axboe071698e2020-01-28 10:04:42 -07009865 case IORING_REGISTER_PERSONALITY:
9866 ret = -EINVAL;
9867 if (arg || nr_args)
9868 break;
9869 ret = io_register_personality(ctx);
9870 break;
9871 case IORING_UNREGISTER_PERSONALITY:
9872 ret = -EINVAL;
9873 if (arg)
9874 break;
9875 ret = io_unregister_personality(ctx, nr_args);
9876 break;
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009877 case IORING_REGISTER_ENABLE_RINGS:
9878 ret = -EINVAL;
9879 if (arg || nr_args)
9880 break;
9881 ret = io_register_enable_rings(ctx);
9882 break;
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009883 case IORING_REGISTER_RESTRICTIONS:
9884 ret = io_register_restrictions(ctx, arg, nr_args);
9885 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07009886 default:
9887 ret = -EINVAL;
9888 break;
9889 }
9890
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009891out:
Jens Axboe071698e2020-01-28 10:04:42 -07009892 if (io_register_op_must_quiesce(opcode)) {
Jens Axboe05f3fb32019-12-09 11:22:50 -07009893 /* bring the ctx back to life */
Jens Axboe05f3fb32019-12-09 11:22:50 -07009894 percpu_ref_reinit(&ctx->refs);
Jens Axboe0f158b42020-05-14 17:18:39 -06009895 reinit_completion(&ctx->ref_comp);
Jens Axboe05f3fb32019-12-09 11:22:50 -07009896 }
Jens Axboeedafcce2019-01-09 09:16:05 -07009897 return ret;
9898}
9899
9900SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode,
9901 void __user *, arg, unsigned int, nr_args)
9902{
9903 struct io_ring_ctx *ctx;
9904 long ret = -EBADF;
9905 struct fd f;
9906
9907 f = fdget(fd);
9908 if (!f.file)
9909 return -EBADF;
9910
9911 ret = -EOPNOTSUPP;
9912 if (f.file->f_op != &io_uring_fops)
9913 goto out_fput;
9914
9915 ctx = f.file->private_data;
9916
Pavel Begunkovb6c23dd2021-02-20 15:17:18 +00009917 io_run_task_work();
9918
Jens Axboeedafcce2019-01-09 09:16:05 -07009919 mutex_lock(&ctx->uring_lock);
9920 ret = __io_uring_register(ctx, opcode, arg, nr_args);
9921 mutex_unlock(&ctx->uring_lock);
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02009922 trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs,
9923 ctx->cq_ev_fd != NULL, ret);
Jens Axboeedafcce2019-01-09 09:16:05 -07009924out_fput:
9925 fdput(f);
9926 return ret;
9927}
9928
Jens Axboe2b188cc2019-01-07 10:46:33 -07009929static int __init io_uring_init(void)
9930{
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009931#define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \
9932 BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \
9933 BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \
9934} while (0)
9935
9936#define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \
9937 __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename)
9938 BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64);
9939 BUILD_BUG_SQE_ELEM(0, __u8, opcode);
9940 BUILD_BUG_SQE_ELEM(1, __u8, flags);
9941 BUILD_BUG_SQE_ELEM(2, __u16, ioprio);
9942 BUILD_BUG_SQE_ELEM(4, __s32, fd);
9943 BUILD_BUG_SQE_ELEM(8, __u64, off);
9944 BUILD_BUG_SQE_ELEM(8, __u64, addr2);
9945 BUILD_BUG_SQE_ELEM(16, __u64, addr);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03009946 BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009947 BUILD_BUG_SQE_ELEM(24, __u32, len);
9948 BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags);
9949 BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags);
9950 BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags);
9951 BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags);
Jiufei Xue5769a352020-06-17 17:53:55 +08009952 BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events);
9953 BUILD_BUG_SQE_ELEM(28, __u32, poll32_events);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009954 BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags);
9955 BUILD_BUG_SQE_ELEM(28, __u32, msg_flags);
9956 BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags);
9957 BUILD_BUG_SQE_ELEM(28, __u32, accept_flags);
9958 BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags);
9959 BUILD_BUG_SQE_ELEM(28, __u32, open_flags);
9960 BUILD_BUG_SQE_ELEM(28, __u32, statx_flags);
9961 BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03009962 BUILD_BUG_SQE_ELEM(28, __u32, splice_flags);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009963 BUILD_BUG_SQE_ELEM(32, __u64, user_data);
9964 BUILD_BUG_SQE_ELEM(40, __u16, buf_index);
9965 BUILD_BUG_SQE_ELEM(42, __u16, personality);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03009966 BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009967
Jens Axboed3656342019-12-18 09:50:26 -07009968 BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST);
Jens Axboe84557872020-03-03 15:28:17 -07009969 BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int));
Jens Axboe91f245d2021-02-09 13:48:50 -07009970 req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC |
9971 SLAB_ACCOUNT);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009972 return 0;
9973};
9974__initcall(io_uring_init);