blob: e621c9f5729ffc01396bc446e9c15a8dc03e9e5b [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
Jens Axboeb5dba592019-12-11 14:02:38 -0700503struct io_close {
504 struct file *file;
Jens Axboeb5dba592019-12-11 14:02:38 -0700505 int fd;
506};
507
Jens Axboead8a48a2019-11-15 08:49:11 -0700508struct io_timeout_data {
509 struct io_kiocb *req;
510 struct hrtimer timer;
511 struct timespec64 ts;
512 enum hrtimer_mode mode;
513};
514
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700515struct io_accept {
516 struct file *file;
517 struct sockaddr __user *addr;
518 int __user *addr_len;
519 int flags;
Jens Axboe09952e32020-03-19 20:16:56 -0600520 unsigned long nofile;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700521};
522
523struct io_sync {
524 struct file *file;
525 loff_t len;
526 loff_t off;
527 int flags;
Jens Axboed63d1b52019-12-10 10:38:56 -0700528 int mode;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700529};
530
Jens Axboefbf23842019-12-17 18:45:56 -0700531struct io_cancel {
532 struct file *file;
533 u64 addr;
534};
535
Jens Axboeb29472e2019-12-17 18:50:29 -0700536struct io_timeout {
537 struct file *file;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +0300538 u32 off;
539 u32 target_seq;
Pavel Begunkov135fcde2020-07-13 23:37:12 +0300540 struct list_head list;
Pavel Begunkov90cd7e42020-10-27 23:25:36 +0000541 /* head of the link, used by linked timeouts only */
542 struct io_kiocb *head;
Jens Axboeb29472e2019-12-17 18:50:29 -0700543};
544
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100545struct io_timeout_rem {
546 struct file *file;
547 u64 addr;
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000548
549 /* timeout update */
550 struct timespec64 ts;
551 u32 flags;
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100552};
553
Jens Axboe9adbd452019-12-20 08:45:55 -0700554struct io_rw {
555 /* NOTE: kiocb has the file as the first member, so don't do it here */
556 struct kiocb kiocb;
557 u64 addr;
558 u64 len;
559};
560
Jens Axboe3fbb51c2019-12-20 08:51:52 -0700561struct io_connect {
562 struct file *file;
563 struct sockaddr __user *addr;
564 int addr_len;
565};
566
Jens Axboee47293f2019-12-20 08:58:21 -0700567struct io_sr_msg {
568 struct file *file;
Jens Axboefddafac2020-01-04 20:19:44 -0700569 union {
Pavel Begunkov4af34172021-04-11 01:46:30 +0100570 struct compat_msghdr __user *umsg_compat;
571 struct user_msghdr __user *umsg;
572 void __user *buf;
Jens Axboefddafac2020-01-04 20:19:44 -0700573 };
Jens Axboee47293f2019-12-20 08:58:21 -0700574 int msg_flags;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700575 int bgid;
Jens Axboefddafac2020-01-04 20:19:44 -0700576 size_t len;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700577 struct io_buffer *kbuf;
Jens Axboee47293f2019-12-20 08:58:21 -0700578};
579
Jens Axboe15b71ab2019-12-11 11:20:36 -0700580struct io_open {
581 struct file *file;
582 int dfd;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700583 struct filename *filename;
Jens Axboec12cedf2020-01-08 17:41:21 -0700584 struct open_how how;
Jens Axboe4022e7a2020-03-19 19:23:18 -0600585 unsigned long nofile;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700586};
587
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000588struct io_rsrc_update {
Jens Axboe05f3fb32019-12-09 11:22:50 -0700589 struct file *file;
590 u64 arg;
591 u32 nr_args;
592 u32 offset;
593};
594
Jens Axboe4840e412019-12-25 22:03:45 -0700595struct io_fadvise {
596 struct file *file;
597 u64 offset;
598 u32 len;
599 u32 advice;
600};
601
Jens Axboec1ca7572019-12-25 22:18:28 -0700602struct io_madvise {
603 struct file *file;
604 u64 addr;
605 u32 len;
606 u32 advice;
607};
608
Jens Axboe3e4827b2020-01-08 15:18:09 -0700609struct io_epoll {
610 struct file *file;
611 int epfd;
612 int op;
613 int fd;
614 struct epoll_event event;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700615};
616
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300617struct io_splice {
618 struct file *file_out;
619 struct file *file_in;
620 loff_t off_out;
621 loff_t off_in;
622 u64 len;
623 unsigned int flags;
624};
625
Jens Axboeddf0322d2020-02-23 16:41:33 -0700626struct io_provide_buf {
627 struct file *file;
628 __u64 addr;
Pavel Begunkov38134ad2021-04-15 13:07:39 +0100629 __u32 len;
Jens Axboeddf0322d2020-02-23 16:41:33 -0700630 __u32 bgid;
631 __u16 nbufs;
632 __u16 bid;
633};
634
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700635struct io_statx {
636 struct file *file;
637 int dfd;
638 unsigned int mask;
639 unsigned int flags;
Bijan Mottahedehe62753e2020-05-22 21:31:18 -0700640 const char __user *filename;
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700641 struct statx __user *buffer;
642};
643
Jens Axboe36f4fa62020-09-05 11:14:22 -0600644struct io_shutdown {
645 struct file *file;
646 int how;
647};
648
Jens Axboe80a261f2020-09-28 14:23:58 -0600649struct io_rename {
650 struct file *file;
651 int old_dfd;
652 int new_dfd;
653 struct filename *oldpath;
654 struct filename *newpath;
655 int flags;
656};
657
Jens Axboe14a11432020-09-28 14:27:37 -0600658struct io_unlink {
659 struct file *file;
660 int dfd;
661 int flags;
662 struct filename *filename;
663};
664
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300665struct io_completion {
666 struct file *file;
667 struct list_head list;
Pavel Begunkov8c3f9cd2021-02-28 22:35:15 +0000668 u32 cflags;
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300669};
670
Jens Axboef499a022019-12-02 16:28:46 -0700671struct io_async_connect {
672 struct sockaddr_storage address;
673};
674
Jens Axboe03b12302019-12-02 18:50:25 -0700675struct io_async_msghdr {
676 struct iovec fast_iov[UIO_FASTIOV];
Pavel Begunkov257e84a2021-02-05 00:58:00 +0000677 /* points to an allocated iov, if NULL we use fast_iov instead */
678 struct iovec *free_iov;
Jens Axboe03b12302019-12-02 18:50:25 -0700679 struct sockaddr __user *uaddr;
680 struct msghdr msg;
Jens Axboeb5379162020-02-09 11:29:15 -0700681 struct sockaddr_storage addr;
Jens Axboe03b12302019-12-02 18:50:25 -0700682};
683
Jens Axboef67676d2019-12-02 11:03:47 -0700684struct io_async_rw {
685 struct iovec fast_iov[UIO_FASTIOV];
Jens Axboeff6165b2020-08-13 09:47:43 -0600686 const struct iovec *free_iovec;
687 struct iov_iter iter;
Jens Axboe227c0c92020-08-13 11:51:40 -0600688 size_t bytes_done;
Jens Axboebcf5a062020-05-22 09:24:42 -0600689 struct wait_page_queue wpq;
Jens Axboef67676d2019-12-02 11:03:47 -0700690};
691
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300692enum {
693 REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT,
694 REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT,
695 REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT,
696 REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT,
697 REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700698 REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300699
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300700 REQ_F_FAIL_LINK_BIT,
701 REQ_F_INFLIGHT_BIT,
702 REQ_F_CUR_POS_BIT,
703 REQ_F_NOWAIT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300704 REQ_F_LINK_TIMEOUT_BIT,
Pavel Begunkov99bc4c32020-02-07 22:04:45 +0300705 REQ_F_NEED_CLEANUP_BIT,
Jens Axboed7718a92020-02-14 22:23:12 -0700706 REQ_F_POLLED_BIT,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700707 REQ_F_BUFFER_SELECTED_BIT,
Pavel Begunkov900fad42020-10-19 16:39:16 +0100708 REQ_F_LTIMEOUT_ACTIVE_BIT,
Pavel Begunkove342c802021-01-19 13:32:47 +0000709 REQ_F_COMPLETE_INLINE_BIT,
Jens Axboe230d50d2021-04-01 20:41:15 -0600710 REQ_F_REISSUE_BIT,
Pavel Begunkov8c130822021-03-22 01:58:32 +0000711 REQ_F_DONT_REISSUE_BIT,
Jens Axboe7b29f922021-03-12 08:30:14 -0700712 /* keep async read/write and isreg together and in order */
713 REQ_F_ASYNC_READ_BIT,
714 REQ_F_ASYNC_WRITE_BIT,
715 REQ_F_ISREG_BIT,
Jens Axboe84557872020-03-03 15:28:17 -0700716
717 /* not a real bit, just to check we're not overflowing the space */
718 __REQ_F_LAST_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300719};
720
721enum {
722 /* ctx owns file */
723 REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT),
724 /* drain existing IO first */
725 REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT),
726 /* linked sqes */
727 REQ_F_LINK = BIT(REQ_F_LINK_BIT),
728 /* doesn't sever on completion < 0 */
729 REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT),
730 /* IOSQE_ASYNC */
731 REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT),
Jens Axboebcda7ba2020-02-23 16:42:51 -0700732 /* IOSQE_BUFFER_SELECT */
733 REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT),
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300734
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300735 /* fail rest of links */
736 REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT),
Pavel Begunkovb05a1bc2021-03-04 13:59:24 +0000737 /* on inflight list, should be cancelled and waited on exit reliably */
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300738 REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT),
739 /* read/write uses file position */
740 REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT),
741 /* must not punt to workers */
742 REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT),
Pavel Begunkov900fad42020-10-19 16:39:16 +0100743 /* has or had linked timeout */
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300744 REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT),
Pavel Begunkov99bc4c32020-02-07 22:04:45 +0300745 /* needs cleanup */
746 REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT),
Jens Axboed7718a92020-02-14 22:23:12 -0700747 /* already went through poll handler */
748 REQ_F_POLLED = BIT(REQ_F_POLLED_BIT),
Jens Axboebcda7ba2020-02-23 16:42:51 -0700749 /* buffer already selected */
750 REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT),
Pavel Begunkov900fad42020-10-19 16:39:16 +0100751 /* linked timeout is active, i.e. prepared by link's head */
752 REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT),
Pavel Begunkove342c802021-01-19 13:32:47 +0000753 /* completion is deferred through io_comp_state */
754 REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT),
Jens Axboe230d50d2021-04-01 20:41:15 -0600755 /* caller should reissue async */
756 REQ_F_REISSUE = BIT(REQ_F_REISSUE_BIT),
Pavel Begunkov8c130822021-03-22 01:58:32 +0000757 /* don't attempt request reissue, see io_rw_reissue() */
758 REQ_F_DONT_REISSUE = BIT(REQ_F_DONT_REISSUE_BIT),
Jens Axboe7b29f922021-03-12 08:30:14 -0700759 /* supports async reads */
760 REQ_F_ASYNC_READ = BIT(REQ_F_ASYNC_READ_BIT),
761 /* supports async writes */
762 REQ_F_ASYNC_WRITE = BIT(REQ_F_ASYNC_WRITE_BIT),
763 /* regular file */
764 REQ_F_ISREG = BIT(REQ_F_ISREG_BIT),
Jens Axboed7718a92020-02-14 22:23:12 -0700765};
766
767struct async_poll {
768 struct io_poll_iocb poll;
Jens Axboe807abcb2020-07-17 17:09:27 -0600769 struct io_poll_iocb *double_poll;
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300770};
771
Jens Axboe7cbf1722021-02-10 00:03:20 +0000772struct io_task_work {
773 struct io_wq_work_node node;
774 task_work_func_t func;
775};
776
Jens Axboe09bb8392019-03-13 12:39:28 -0600777/*
778 * NOTE! Each of the iocb union members has the file pointer
779 * as the first entry in their struct definition. So you can
780 * access the file pointer through any of the sub-structs,
781 * or directly as just 'ki_filp' in this struct.
782 */
Jens Axboe2b188cc2019-01-07 10:46:33 -0700783struct io_kiocb {
Jens Axboe221c5eb2019-01-17 09:41:58 -0700784 union {
Jens Axboe09bb8392019-03-13 12:39:28 -0600785 struct file *file;
Jens Axboe9adbd452019-12-20 08:45:55 -0700786 struct io_rw rw;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700787 struct io_poll_iocb poll;
Pavel Begunkov9d805892021-04-13 02:58:40 +0100788 struct io_poll_update poll_update;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700789 struct io_accept accept;
790 struct io_sync sync;
Jens Axboefbf23842019-12-17 18:45:56 -0700791 struct io_cancel cancel;
Jens Axboeb29472e2019-12-17 18:50:29 -0700792 struct io_timeout timeout;
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100793 struct io_timeout_rem timeout_rem;
Jens Axboe3fbb51c2019-12-20 08:51:52 -0700794 struct io_connect connect;
Jens Axboee47293f2019-12-20 08:58:21 -0700795 struct io_sr_msg sr_msg;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700796 struct io_open open;
Jens Axboeb5dba592019-12-11 14:02:38 -0700797 struct io_close close;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000798 struct io_rsrc_update rsrc_update;
Jens Axboe4840e412019-12-25 22:03:45 -0700799 struct io_fadvise fadvise;
Jens Axboec1ca7572019-12-25 22:18:28 -0700800 struct io_madvise madvise;
Jens Axboe3e4827b2020-01-08 15:18:09 -0700801 struct io_epoll epoll;
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300802 struct io_splice splice;
Jens Axboeddf0322d2020-02-23 16:41:33 -0700803 struct io_provide_buf pbuf;
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700804 struct io_statx statx;
Jens Axboe36f4fa62020-09-05 11:14:22 -0600805 struct io_shutdown shutdown;
Jens Axboe80a261f2020-09-28 14:23:58 -0600806 struct io_rename rename;
Jens Axboe14a11432020-09-28 14:27:37 -0600807 struct io_unlink unlink;
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300808 /* use only after cleaning per-op data, see io_clean_op() */
809 struct io_completion compl;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700810 };
Jens Axboe2b188cc2019-01-07 10:46:33 -0700811
Jens Axboee8c2bc12020-08-15 18:44:09 -0700812 /* opcode allocated if it needs to store data for async defer */
813 void *async_data;
Jens Axboed625c6e2019-12-17 19:53:05 -0700814 u8 opcode;
Xiaoguang Wang65a65432020-06-11 23:39:36 +0800815 /* polled IO has completed */
816 u8 iopoll_completed;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700817
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -0700818 u16 buf_index;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +0300819 u32 result;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -0700820
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300821 struct io_ring_ctx *ctx;
822 unsigned int flags;
Jens Axboeabc54d62021-02-24 13:32:30 -0700823 atomic_t refs;
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300824 struct task_struct *task;
825 u64 user_data;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700826
Pavel Begunkovf2f87372020-10-27 23:25:37 +0000827 struct io_kiocb *link;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000828 struct percpu_ref *fixed_rsrc_refs;
Jens Axboed7718a92020-02-14 22:23:12 -0700829
Pavel Begunkovb303fe22021-04-11 01:46:26 +0100830 /* used with ctx->iopoll_list with reads/writes */
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300831 struct list_head inflight_entry;
Jens Axboe7cbf1722021-02-10 00:03:20 +0000832 union {
833 struct io_task_work io_task_work;
834 struct callback_head task_work;
835 };
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300836 /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */
837 struct hlist_node hash_node;
838 struct async_poll *apoll;
839 struct io_wq_work work;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700840};
841
Pavel Begunkov13bf43f2021-03-06 11:02:12 +0000842struct io_tctx_node {
843 struct list_head ctx_node;
844 struct task_struct *task;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +0000845 struct io_ring_ctx *ctx;
846};
847
Pavel Begunkov27dc8332020-07-13 23:37:14 +0300848struct io_defer_entry {
849 struct list_head list;
850 struct io_kiocb *req;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +0300851 u32 seq;
Pavel Begunkov27dc8332020-07-13 23:37:14 +0300852};
853
Jens Axboed3656342019-12-18 09:50:26 -0700854struct io_op_def {
Jens Axboed3656342019-12-18 09:50:26 -0700855 /* needs req->file assigned */
856 unsigned needs_file : 1;
Jens Axboed3656342019-12-18 09:50:26 -0700857 /* hash wq insertion if file is a regular file */
858 unsigned hash_reg_file : 1;
859 /* unbound wq insertion if file is a non-regular file */
860 unsigned unbound_nonreg_file : 1;
Jens Axboe66f4af92020-01-16 15:36:52 -0700861 /* opcode is not supported by this kernel */
862 unsigned not_supported : 1;
Jens Axboe8a727582020-02-20 09:59:44 -0700863 /* set if opcode supports polled "wait" */
864 unsigned pollin : 1;
865 unsigned pollout : 1;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700866 /* op supports buffer selection */
867 unsigned buffer_select : 1;
Pavel Begunkov26f05052021-02-28 22:35:18 +0000868 /* do prep async if is going to be punted */
869 unsigned needs_async_setup : 1;
Jens Axboe27926b62020-10-28 09:33:23 -0600870 /* should block plug */
871 unsigned plug : 1;
Jens Axboee8c2bc12020-08-15 18:44:09 -0700872 /* size of async data needed, if any */
873 unsigned short async_size;
Jens Axboed3656342019-12-18 09:50:26 -0700874};
875
Jens Axboe09186822020-10-13 15:01:40 -0600876static const struct io_op_def io_op_defs[] = {
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300877 [IORING_OP_NOP] = {},
878 [IORING_OP_READV] = {
Jens Axboed3656342019-12-18 09:50:26 -0700879 .needs_file = 1,
880 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700881 .pollin = 1,
Jens Axboe4d954c22020-02-27 07:31:19 -0700882 .buffer_select = 1,
Pavel Begunkov26f05052021-02-28 22:35:18 +0000883 .needs_async_setup = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600884 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700885 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700886 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300887 [IORING_OP_WRITEV] = {
Jens Axboed3656342019-12-18 09:50:26 -0700888 .needs_file = 1,
889 .hash_reg_file = 1,
890 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700891 .pollout = 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_FSYNC] = {
Jens Axboed3656342019-12-18 09:50:26 -0700897 .needs_file = 1,
898 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300899 [IORING_OP_READ_FIXED] = {
Jens Axboed3656342019-12-18 09:50:26 -0700900 .needs_file = 1,
901 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700902 .pollin = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600903 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700904 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700905 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300906 [IORING_OP_WRITE_FIXED] = {
Jens Axboed3656342019-12-18 09:50:26 -0700907 .needs_file = 1,
908 .hash_reg_file = 1,
909 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700910 .pollout = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600911 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700912 .async_size = sizeof(struct io_async_rw),
Jens Axboed3656342019-12-18 09:50:26 -0700913 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300914 [IORING_OP_POLL_ADD] = {
Jens Axboed3656342019-12-18 09:50:26 -0700915 .needs_file = 1,
916 .unbound_nonreg_file = 1,
917 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300918 [IORING_OP_POLL_REMOVE] = {},
919 [IORING_OP_SYNC_FILE_RANGE] = {
Jens Axboed3656342019-12-18 09:50:26 -0700920 .needs_file = 1,
921 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300922 [IORING_OP_SENDMSG] = {
Jens Axboed3656342019-12-18 09:50:26 -0700923 .needs_file = 1,
924 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700925 .pollout = 1,
Pavel Begunkov26f05052021-02-28 22:35:18 +0000926 .needs_async_setup = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700927 .async_size = sizeof(struct io_async_msghdr),
Jens Axboed3656342019-12-18 09:50:26 -0700928 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300929 [IORING_OP_RECVMSG] = {
Jens Axboed3656342019-12-18 09:50:26 -0700930 .needs_file = 1,
931 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700932 .pollin = 1,
Jens Axboe52de1fe2020-02-27 10:15:42 -0700933 .buffer_select = 1,
Pavel Begunkov26f05052021-02-28 22:35:18 +0000934 .needs_async_setup = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700935 .async_size = sizeof(struct io_async_msghdr),
Jens Axboed3656342019-12-18 09:50:26 -0700936 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300937 [IORING_OP_TIMEOUT] = {
Jens Axboee8c2bc12020-08-15 18:44:09 -0700938 .async_size = sizeof(struct io_timeout_data),
Jens Axboed3656342019-12-18 09:50:26 -0700939 },
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000940 [IORING_OP_TIMEOUT_REMOVE] = {
941 /* used by timeout updates' prep() */
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000942 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300943 [IORING_OP_ACCEPT] = {
Jens Axboed3656342019-12-18 09:50:26 -0700944 .needs_file = 1,
945 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700946 .pollin = 1,
Jens Axboed3656342019-12-18 09:50:26 -0700947 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300948 [IORING_OP_ASYNC_CANCEL] = {},
949 [IORING_OP_LINK_TIMEOUT] = {
Jens Axboee8c2bc12020-08-15 18:44:09 -0700950 .async_size = sizeof(struct io_timeout_data),
Jens Axboed3656342019-12-18 09:50:26 -0700951 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300952 [IORING_OP_CONNECT] = {
Jens Axboed3656342019-12-18 09:50:26 -0700953 .needs_file = 1,
954 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700955 .pollout = 1,
Pavel Begunkov26f05052021-02-28 22:35:18 +0000956 .needs_async_setup = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700957 .async_size = sizeof(struct io_async_connect),
Jens Axboed3656342019-12-18 09:50:26 -0700958 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300959 [IORING_OP_FALLOCATE] = {
Jens Axboed3656342019-12-18 09:50:26 -0700960 .needs_file = 1,
961 },
Jens Axboe44526be2021-02-15 13:32:18 -0700962 [IORING_OP_OPENAT] = {},
963 [IORING_OP_CLOSE] = {},
964 [IORING_OP_FILES_UPDATE] = {},
965 [IORING_OP_STATX] = {},
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300966 [IORING_OP_READ] = {
Jens Axboe3a6820f2019-12-22 15:19:35 -0700967 .needs_file = 1,
968 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700969 .pollin = 1,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700970 .buffer_select = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600971 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700972 .async_size = sizeof(struct io_async_rw),
Jens Axboe3a6820f2019-12-22 15:19:35 -0700973 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300974 [IORING_OP_WRITE] = {
Jens Axboe3a6820f2019-12-22 15:19:35 -0700975 .needs_file = 1,
976 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700977 .pollout = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600978 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700979 .async_size = sizeof(struct io_async_rw),
Jens Axboe3a6820f2019-12-22 15:19:35 -0700980 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300981 [IORING_OP_FADVISE] = {
Jens Axboe4840e412019-12-25 22:03:45 -0700982 .needs_file = 1,
983 },
Jens Axboe44526be2021-02-15 13:32:18 -0700984 [IORING_OP_MADVISE] = {},
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300985 [IORING_OP_SEND] = {
Jens Axboefddafac2020-01-04 20:19:44 -0700986 .needs_file = 1,
987 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700988 .pollout = 1,
Jens Axboefddafac2020-01-04 20:19:44 -0700989 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300990 [IORING_OP_RECV] = {
Jens Axboefddafac2020-01-04 20:19:44 -0700991 .needs_file = 1,
992 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700993 .pollin = 1,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700994 .buffer_select = 1,
Jens Axboefddafac2020-01-04 20:19:44 -0700995 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300996 [IORING_OP_OPENAT2] = {
Jens Axboecebdb982020-01-08 17:59:24 -0700997 },
Jens Axboe3e4827b2020-01-08 15:18:09 -0700998 [IORING_OP_EPOLL_CTL] = {
999 .unbound_nonreg_file = 1,
Jens Axboe3e4827b2020-01-08 15:18:09 -07001000 },
Pavel Begunkov7d67af22020-02-24 11:32:45 +03001001 [IORING_OP_SPLICE] = {
1002 .needs_file = 1,
1003 .hash_reg_file = 1,
1004 .unbound_nonreg_file = 1,
Jens Axboeddf0322d2020-02-23 16:41:33 -07001005 },
1006 [IORING_OP_PROVIDE_BUFFERS] = {},
Jens Axboe067524e2020-03-02 16:32:28 -07001007 [IORING_OP_REMOVE_BUFFERS] = {},
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03001008 [IORING_OP_TEE] = {
1009 .needs_file = 1,
1010 .hash_reg_file = 1,
1011 .unbound_nonreg_file = 1,
1012 },
Jens Axboe36f4fa62020-09-05 11:14:22 -06001013 [IORING_OP_SHUTDOWN] = {
1014 .needs_file = 1,
1015 },
Jens Axboe44526be2021-02-15 13:32:18 -07001016 [IORING_OP_RENAMEAT] = {},
1017 [IORING_OP_UNLINKAT] = {},
Jens Axboed3656342019-12-18 09:50:26 -07001018};
1019
Pavel Begunkov7a612352021-03-09 00:37:59 +00001020static bool io_disarm_next(struct io_kiocb *req);
Pavel Begunkovd56d9382021-03-06 11:02:13 +00001021static void io_uring_del_task_file(unsigned long index);
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00001022static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
1023 struct task_struct *task,
1024 struct files_struct *files);
Pavel Begunkov734551d2021-04-18 14:52:09 +01001025static void io_uring_cancel_sqpoll(struct io_sq_data *sqd);
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01001026static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx);
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00001027
Pavel Begunkovff6421642021-04-11 01:46:32 +01001028static bool io_cqring_fill_event(struct io_kiocb *req, long res, unsigned cflags);
Jackie Liuec9c02a2019-11-08 23:50:36 +08001029static void io_put_req(struct io_kiocb *req);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001030static void io_put_req_deferred(struct io_kiocb *req, int nr);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001031static void io_dismantle_req(struct io_kiocb *req);
1032static void io_put_task(struct task_struct *task, int nr);
Jens Axboe94ae5e72019-11-14 19:39:52 -07001033static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req);
1034static void io_queue_linked_timeout(struct io_kiocb *req);
Jens Axboe05f3fb32019-12-09 11:22:50 -07001035static int __io_sqe_files_update(struct io_ring_ctx *ctx,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001036 struct io_uring_rsrc_update *ip,
Jens Axboe05f3fb32019-12-09 11:22:50 -07001037 unsigned nr_args);
Pavel Begunkov68fb8972021-03-19 17:22:41 +00001038static void io_clean_op(struct io_kiocb *req);
Pavel Begunkov8371adf2020-10-10 18:34:08 +01001039static struct file *io_file_get(struct io_submit_state *state,
1040 struct io_kiocb *req, int fd, bool fixed);
Pavel Begunkovc5eef2b942021-02-10 00:03:22 +00001041static void __io_queue_sqe(struct io_kiocb *req);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001042static void io_rsrc_put_work(struct work_struct *work);
Jens Axboede0617e2019-04-06 21:51:27 -06001043
Pavel Begunkov907d1df2021-01-26 23:35:10 +00001044static void io_req_task_queue(struct io_kiocb *req);
Jens Axboe65453d12021-02-10 00:03:21 +00001045static void io_submit_flush_completions(struct io_comp_state *cs,
1046 struct io_ring_ctx *ctx);
Jens Axboe50826202021-02-23 09:02:26 -07001047static bool io_poll_remove_waitqs(struct io_kiocb *req);
Pavel Begunkov179ae0d2021-02-28 22:35:20 +00001048static int io_req_prep_async(struct io_kiocb *req);
Jens Axboe9a56a232019-01-09 09:06:50 -07001049
Jens Axboe2b188cc2019-01-07 10:46:33 -07001050static struct kmem_cache *req_cachep;
1051
Jens Axboe09186822020-10-13 15:01:40 -06001052static const struct file_operations io_uring_fops;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001053
1054struct sock *io_uring_get_socket(struct file *file)
1055{
1056#if defined(CONFIG_UNIX)
1057 if (file->f_op == &io_uring_fops) {
1058 struct io_ring_ctx *ctx = file->private_data;
1059
1060 return ctx->ring_sock->sk;
1061 }
1062#endif
1063 return NULL;
1064}
1065EXPORT_SYMBOL(io_uring_get_socket);
1066
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001067#define io_for_each_link(pos, head) \
1068 for (pos = (head); pos; pos = pos->link)
1069
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01001070static inline void io_req_set_rsrc_node(struct io_kiocb *req)
Jens Axboec40f6372020-06-25 15:39:59 -06001071{
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001072 struct io_ring_ctx *ctx = req->ctx;
1073
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001074 if (!req->fixed_rsrc_refs) {
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01001075 req->fixed_rsrc_refs = &ctx->rsrc_node->refs;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001076 percpu_ref_get(req->fixed_rsrc_refs);
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001077 }
1078}
1079
Pavel Begunkovf70865d2021-04-11 01:46:40 +01001080static void io_refs_resurrect(struct percpu_ref *ref, struct completion *compl)
1081{
1082 bool got = percpu_ref_tryget(ref);
1083
1084 /* already at zero, wait for ->release() */
1085 if (!got)
1086 wait_for_completion(compl);
1087 percpu_ref_resurrect(ref);
1088 if (got)
1089 percpu_ref_put(ref);
1090}
1091
Pavel Begunkov08d23632020-11-06 13:00:22 +00001092static bool io_match_task(struct io_kiocb *head,
1093 struct task_struct *task,
1094 struct files_struct *files)
1095{
1096 struct io_kiocb *req;
1097
Pavel Begunkov68207682021-03-22 01:58:25 +00001098 if (task && head->task != task)
Pavel Begunkov08d23632020-11-06 13:00:22 +00001099 return false;
1100 if (!files)
1101 return true;
1102
1103 io_for_each_link(req, head) {
Pavel Begunkovb05a1bc2021-03-04 13:59:24 +00001104 if (req->flags & REQ_F_INFLIGHT)
Jens Axboe02a13672021-01-23 15:49:31 -07001105 return true;
Pavel Begunkov08d23632020-11-06 13:00:22 +00001106 }
1107 return false;
1108}
1109
Jens Axboec40f6372020-06-25 15:39:59 -06001110static inline void req_set_fail_links(struct io_kiocb *req)
1111{
Pavel Begunkove4335ed2021-04-11 01:46:39 +01001112 if (req->flags & REQ_F_LINK)
Jens Axboec40f6372020-06-25 15:39:59 -06001113 req->flags |= REQ_F_FAIL_LINK;
1114}
Jens Axboe4a38aed22020-05-14 17:21:15 -06001115
Jens Axboe2b188cc2019-01-07 10:46:33 -07001116static void io_ring_ctx_ref_free(struct percpu_ref *ref)
1117{
1118 struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs);
1119
Jens Axboe0f158b42020-05-14 17:18:39 -06001120 complete(&ctx->ref_comp);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001121}
1122
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03001123static inline bool io_is_timeout_noseq(struct io_kiocb *req)
1124{
1125 return !req->timeout.off;
1126}
1127
Jens Axboe2b188cc2019-01-07 10:46:33 -07001128static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
1129{
1130 struct io_ring_ctx *ctx;
Jens Axboe78076bb2019-12-04 19:56:40 -07001131 int hash_bits;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001132
1133 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1134 if (!ctx)
1135 return NULL;
1136
Jens Axboe78076bb2019-12-04 19:56:40 -07001137 /*
1138 * Use 5 bits less than the max cq entries, that should give us around
1139 * 32 entries per hash list if totally full and uniformly spread.
1140 */
1141 hash_bits = ilog2(p->cq_entries);
1142 hash_bits -= 5;
1143 if (hash_bits <= 0)
1144 hash_bits = 1;
1145 ctx->cancel_hash_bits = hash_bits;
1146 ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head),
1147 GFP_KERNEL);
1148 if (!ctx->cancel_hash)
1149 goto err;
1150 __hash_init(ctx->cancel_hash, 1U << hash_bits);
1151
Roman Gushchin21482892019-05-07 10:01:48 -07001152 if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free,
Jens Axboe206aefd2019-11-07 18:27:42 -07001153 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL))
1154 goto err;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001155
1156 ctx->flags = p->flags;
Jens Axboe90554202020-09-03 12:12:41 -06001157 init_waitqueue_head(&ctx->sqo_sq_wait);
Jens Axboe69fb2132020-09-14 11:16:23 -06001158 INIT_LIST_HEAD(&ctx->sqd_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001159 init_waitqueue_head(&ctx->cq_wait);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001160 INIT_LIST_HEAD(&ctx->cq_overflow_list);
Jens Axboe0f158b42020-05-14 17:18:39 -06001161 init_completion(&ctx->ref_comp);
Jens Axboe9e15c3a2021-03-13 12:29:43 -07001162 xa_init_flags(&ctx->io_buffers, XA_FLAGS_ALLOC1);
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00001163 xa_init_flags(&ctx->personalities, XA_FLAGS_ALLOC1);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001164 mutex_init(&ctx->uring_lock);
1165 init_waitqueue_head(&ctx->wait);
1166 spin_lock_init(&ctx->completion_lock);
Pavel Begunkov540e32a2020-07-13 23:37:09 +03001167 INIT_LIST_HEAD(&ctx->iopoll_list);
Jens Axboede0617e2019-04-06 21:51:27 -06001168 INIT_LIST_HEAD(&ctx->defer_list);
Jens Axboe5262f562019-09-17 12:26:57 -06001169 INIT_LIST_HEAD(&ctx->timeout_list);
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00001170 spin_lock_init(&ctx->rsrc_ref_lock);
1171 INIT_LIST_HEAD(&ctx->rsrc_ref_list);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001172 INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work);
1173 init_llist_head(&ctx->rsrc_put_llist);
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00001174 INIT_LIST_HEAD(&ctx->tctx_list);
Jens Axboe1b4c3512021-02-10 00:03:19 +00001175 INIT_LIST_HEAD(&ctx->submit_state.comp.free_list);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001176 INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001177 return ctx;
Jens Axboe206aefd2019-11-07 18:27:42 -07001178err:
Jens Axboe78076bb2019-12-04 19:56:40 -07001179 kfree(ctx->cancel_hash);
Jens Axboe206aefd2019-11-07 18:27:42 -07001180 kfree(ctx);
1181 return NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001182}
1183
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001184static bool req_need_defer(struct io_kiocb *req, u32 seq)
Jens Axboede0617e2019-04-06 21:51:27 -06001185{
Jens Axboe2bc99302020-07-09 09:43:27 -06001186 if (unlikely(req->flags & REQ_F_IO_DRAIN)) {
1187 struct io_ring_ctx *ctx = req->ctx;
Jackie Liua197f662019-11-08 08:09:12 -07001188
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001189 return seq != ctx->cached_cq_tail
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001190 + READ_ONCE(ctx->cached_cq_overflow);
Jens Axboe2bc99302020-07-09 09:43:27 -06001191 }
Jens Axboe7adf4ea2019-10-10 21:42:58 -06001192
Bob Liu9d858b22019-11-13 18:06:25 +08001193 return false;
Jens Axboe7adf4ea2019-10-10 21:42:58 -06001194}
1195
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00001196static void io_req_track_inflight(struct io_kiocb *req)
1197{
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00001198 if (!(req->flags & REQ_F_INFLIGHT)) {
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00001199 req->flags |= REQ_F_INFLIGHT;
Pavel Begunkovb303fe22021-04-11 01:46:26 +01001200 atomic_inc(&current->io_uring->inflight_tracked);
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00001201 }
1202}
1203
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001204static void io_prep_async_work(struct io_kiocb *req)
Jens Axboe561fb042019-10-24 07:25:42 -06001205{
Jens Axboed3656342019-12-18 09:50:26 -07001206 const struct io_op_def *def = &io_op_defs[req->opcode];
Pavel Begunkov23329512020-10-10 18:34:06 +01001207 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe54a91f32019-09-10 09:15:04 -06001208
Jens Axboe003e8dc2021-03-06 09:22:27 -07001209 if (!req->work.creds)
1210 req->work.creds = get_current_cred();
1211
Pavel Begunkove1d675d2021-03-22 01:58:29 +00001212 req->work.list.next = NULL;
1213 req->work.flags = 0;
Pavel Begunkovfeaadc42020-10-22 16:47:16 +01001214 if (req->flags & REQ_F_FORCE_ASYNC)
1215 req->work.flags |= IO_WQ_WORK_CONCURRENT;
1216
Jens Axboed3656342019-12-18 09:50:26 -07001217 if (req->flags & REQ_F_ISREG) {
Pavel Begunkov23329512020-10-10 18:34:06 +01001218 if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL))
Pavel Begunkov8766dd52020-03-14 00:31:04 +03001219 io_wq_hash_work(&req->work, file_inode(req->file));
Jens Axboe4b982bd2021-04-01 08:38:34 -06001220 } else if (!req->file || !S_ISBLK(file_inode(req->file)->i_mode)) {
Jens Axboed3656342019-12-18 09:50:26 -07001221 if (def->unbound_nonreg_file)
Jens Axboe3529d8c2019-12-19 18:24:38 -07001222 req->work.flags |= IO_WQ_WORK_UNBOUND;
Jens Axboe54a91f32019-09-10 09:15:04 -06001223 }
Pavel Begunkove1d675d2021-03-22 01:58:29 +00001224
1225 switch (req->opcode) {
1226 case IORING_OP_SPLICE:
1227 case IORING_OP_TEE:
Pavel Begunkove1d675d2021-03-22 01:58:29 +00001228 if (!S_ISREG(file_inode(req->splice.file_in)->i_mode))
1229 req->work.flags |= IO_WQ_WORK_UNBOUND;
1230 break;
1231 }
Jens Axboe561fb042019-10-24 07:25:42 -06001232}
1233
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001234static void io_prep_async_link(struct io_kiocb *req)
1235{
1236 struct io_kiocb *cur;
1237
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001238 io_for_each_link(cur, req)
1239 io_prep_async_work(cur);
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001240}
1241
Pavel Begunkovebf93662021-03-01 18:20:47 +00001242static void io_queue_async_work(struct io_kiocb *req)
Jens Axboe561fb042019-10-24 07:25:42 -06001243{
Jackie Liua197f662019-11-08 08:09:12 -07001244 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001245 struct io_kiocb *link = io_prep_linked_timeout(req);
Jens Axboe5aa75ed2021-02-16 12:56:50 -07001246 struct io_uring_task *tctx = req->task->io_uring;
Jens Axboe561fb042019-10-24 07:25:42 -06001247
Jens Axboe3bfe6102021-02-16 14:15:30 -07001248 BUG_ON(!tctx);
1249 BUG_ON(!tctx->io_wq);
Jens Axboe561fb042019-10-24 07:25:42 -06001250
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001251 /* init ->work of the whole link before punting */
1252 io_prep_async_link(req);
Pavel Begunkovd07f1e8a2021-03-22 01:45:58 +00001253 trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req,
1254 &req->work, req->flags);
Pavel Begunkovebf93662021-03-01 18:20:47 +00001255 io_wq_enqueue(tctx->io_wq, &req->work);
Jens Axboe7271ef32020-08-10 09:55:22 -06001256 if (link)
1257 io_queue_linked_timeout(link);
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001258}
1259
Pavel Begunkov1ee41602021-03-25 18:32:42 +00001260static void io_kill_timeout(struct io_kiocb *req, int status)
Pavel Begunkov8c855882021-04-13 02:58:41 +01001261 __must_hold(&req->ctx->completion_lock)
Jens Axboe5262f562019-09-17 12:26:57 -06001262{
Jens Axboee8c2bc12020-08-15 18:44:09 -07001263 struct io_timeout_data *io = req->async_data;
Jens Axboe5262f562019-09-17 12:26:57 -06001264
Pavel Begunkovfd9c7bc2021-04-13 02:58:42 +01001265 if (hrtimer_try_to_cancel(&io->timer) != -1) {
Pavel Begunkov01cec8c2020-07-30 18:43:50 +03001266 atomic_set(&req->ctx->cq_timeouts,
1267 atomic_read(&req->ctx->cq_timeouts) + 1);
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001268 list_del_init(&req->timeout.list);
Pavel Begunkovff6421642021-04-11 01:46:32 +01001269 io_cqring_fill_event(req, status, 0);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001270 io_put_req_deferred(req, 1);
Jens Axboe5262f562019-09-17 12:26:57 -06001271 }
1272}
1273
Pavel Begunkov04518942020-05-26 20:34:05 +03001274static void __io_queue_deferred(struct io_ring_ctx *ctx)
1275{
1276 do {
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001277 struct io_defer_entry *de = list_first_entry(&ctx->defer_list,
1278 struct io_defer_entry, list);
Pavel Begunkov04518942020-05-26 20:34:05 +03001279
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001280 if (req_need_defer(de->req, de->seq))
Pavel Begunkov04518942020-05-26 20:34:05 +03001281 break;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001282 list_del_init(&de->list);
Pavel Begunkov907d1df2021-01-26 23:35:10 +00001283 io_req_task_queue(de->req);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001284 kfree(de);
Pavel Begunkov04518942020-05-26 20:34:05 +03001285 } while (!list_empty(&ctx->defer_list));
1286}
1287
Pavel Begunkov360428f2020-05-30 14:54:17 +03001288static void io_flush_timeouts(struct io_ring_ctx *ctx)
1289{
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001290 u32 seq;
1291
1292 if (list_empty(&ctx->timeout_list))
1293 return;
1294
1295 seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
1296
1297 do {
1298 u32 events_needed, events_got;
Pavel Begunkov360428f2020-05-30 14:54:17 +03001299 struct io_kiocb *req = list_first_entry(&ctx->timeout_list,
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001300 struct io_kiocb, timeout.list);
Pavel Begunkov360428f2020-05-30 14:54:17 +03001301
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03001302 if (io_is_timeout_noseq(req))
Pavel Begunkov360428f2020-05-30 14:54:17 +03001303 break;
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001304
1305 /*
1306 * Since seq can easily wrap around over time, subtract
1307 * the last seq at which timeouts were flushed before comparing.
1308 * Assuming not more than 2^31-1 events have happened since,
1309 * these subtractions won't have wrapped, so we can check if
1310 * target is in [last_seq, current_seq] by comparing the two.
1311 */
1312 events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush;
1313 events_got = seq - ctx->cq_last_tm_flush;
1314 if (events_got < events_needed)
Pavel Begunkov360428f2020-05-30 14:54:17 +03001315 break;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03001316
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001317 list_del_init(&req->timeout.list);
Pavel Begunkov1ee41602021-03-25 18:32:42 +00001318 io_kill_timeout(req, 0);
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001319 } while (!list_empty(&ctx->timeout_list));
1320
1321 ctx->cq_last_tm_flush = seq;
Pavel Begunkov360428f2020-05-30 14:54:17 +03001322}
1323
Jens Axboede0617e2019-04-06 21:51:27 -06001324static void io_commit_cqring(struct io_ring_ctx *ctx)
1325{
Pavel Begunkov360428f2020-05-30 14:54:17 +03001326 io_flush_timeouts(ctx);
Pavel Begunkovec30e042021-01-19 13:32:38 +00001327
1328 /* order cqe stores with ring update */
1329 smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail);
Jens Axboede0617e2019-04-06 21:51:27 -06001330
Pavel Begunkov04518942020-05-26 20:34:05 +03001331 if (unlikely(!list_empty(&ctx->defer_list)))
1332 __io_queue_deferred(ctx);
Jens Axboede0617e2019-04-06 21:51:27 -06001333}
1334
Jens Axboe90554202020-09-03 12:12:41 -06001335static inline bool io_sqring_full(struct io_ring_ctx *ctx)
1336{
1337 struct io_rings *r = ctx->rings;
1338
1339 return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries;
1340}
1341
Pavel Begunkov888aae22021-01-19 13:32:39 +00001342static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx)
1343{
1344 return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head);
1345}
1346
Pavel Begunkov8d133262021-04-11 01:46:33 +01001347static inline struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001348{
Hristo Venev75b28af2019-08-26 17:23:46 +00001349 struct io_rings *rings = ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001350 unsigned tail;
1351
Stefan Bühler115e12e2019-04-24 23:54:18 +02001352 /*
1353 * writes to the cq entry need to come after reading head; the
1354 * control dependency is enough as we're using WRITE_ONCE to
1355 * fill the cq entry
1356 */
Pavel Begunkov888aae22021-01-19 13:32:39 +00001357 if (__io_cqring_events(ctx) == rings->cq_ring_entries)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001358 return NULL;
1359
Pavel Begunkov888aae22021-01-19 13:32:39 +00001360 tail = ctx->cached_cq_tail++;
Hristo Venev75b28af2019-08-26 17:23:46 +00001361 return &rings->cqes[tail & ctx->cq_mask];
Jens Axboe2b188cc2019-01-07 10:46:33 -07001362}
1363
Jens Axboef2842ab2020-01-08 11:04:00 -07001364static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx)
1365{
Pavel Begunkov44c769d2021-04-11 01:46:31 +01001366 if (likely(!ctx->cq_ev_fd))
Jens Axboef0b493e2020-02-01 21:30:11 -07001367 return false;
Stefano Garzarella7e55a192020-05-15 18:38:05 +02001368 if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED)
1369 return false;
Pavel Begunkov44c769d2021-04-11 01:46:31 +01001370 return !ctx->eventfd_async || io_wq_current_is_worker();
Jens Axboef2842ab2020-01-08 11:04:00 -07001371}
1372
Jens Axboeb41e9852020-02-17 09:52:41 -07001373static void io_cqring_ev_posted(struct io_ring_ctx *ctx)
Jens Axboe8c838782019-03-12 15:48:16 -06001374{
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001375 /* see waitqueue_active() comment */
1376 smp_mb();
1377
Jens Axboe8c838782019-03-12 15:48:16 -06001378 if (waitqueue_active(&ctx->wait))
1379 wake_up(&ctx->wait);
Jens Axboe534ca6d2020-09-02 13:52:19 -06001380 if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait))
1381 wake_up(&ctx->sq_data->wait);
Jens Axboeb41e9852020-02-17 09:52:41 -07001382 if (io_should_trigger_evfd(ctx))
Jens Axboe9b402842019-04-11 11:45:41 -06001383 eventfd_signal(ctx->cq_ev_fd, 1);
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001384 if (waitqueue_active(&ctx->cq_wait)) {
Pavel Begunkov4aa84f22021-01-07 03:15:42 +00001385 wake_up_interruptible(&ctx->cq_wait);
1386 kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
1387 }
Jens Axboe8c838782019-03-12 15:48:16 -06001388}
1389
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001390static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx)
1391{
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001392 /* see waitqueue_active() comment */
1393 smp_mb();
1394
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001395 if (ctx->flags & IORING_SETUP_SQPOLL) {
1396 if (waitqueue_active(&ctx->wait))
1397 wake_up(&ctx->wait);
1398 }
1399 if (io_should_trigger_evfd(ctx))
1400 eventfd_signal(ctx->cq_ev_fd, 1);
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001401 if (waitqueue_active(&ctx->cq_wait)) {
Pavel Begunkov4aa84f22021-01-07 03:15:42 +00001402 wake_up_interruptible(&ctx->cq_wait);
1403 kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
1404 }
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001405}
1406
Jens Axboec4a2ed72019-11-21 21:01:26 -07001407/* Returns true if there are no backlogged entries after the flush */
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00001408static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001409{
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001410 struct io_rings *rings = ctx->rings;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001411 unsigned long flags;
Jens Axboeb18032b2021-01-24 16:58:56 -07001412 bool all_flushed, posted;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001413
Pavel Begunkove23de152020-12-17 00:24:37 +00001414 if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries)
1415 return false;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001416
Jens Axboeb18032b2021-01-24 16:58:56 -07001417 posted = false;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001418 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00001419 while (!list_empty(&ctx->cq_overflow_list)) {
1420 struct io_uring_cqe *cqe = io_get_cqring(ctx);
1421 struct io_overflow_cqe *ocqe;
Jens Axboee6c8aa92020-09-28 13:10:13 -06001422
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001423 if (!cqe && !force)
1424 break;
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00001425 ocqe = list_first_entry(&ctx->cq_overflow_list,
1426 struct io_overflow_cqe, list);
1427 if (cqe)
1428 memcpy(cqe, &ocqe->cqe, sizeof(*cqe));
1429 else
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001430 WRITE_ONCE(ctx->rings->cq_overflow,
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00001431 ++ctx->cached_cq_overflow);
Jens Axboeb18032b2021-01-24 16:58:56 -07001432 posted = true;
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00001433 list_del(&ocqe->list);
1434 kfree(ocqe);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001435 }
1436
Pavel Begunkov09e88402020-12-17 00:24:38 +00001437 all_flushed = list_empty(&ctx->cq_overflow_list);
1438 if (all_flushed) {
1439 clear_bit(0, &ctx->sq_check_overflow);
1440 clear_bit(0, &ctx->cq_check_overflow);
1441 ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW;
1442 }
Pavel Begunkov46930142020-07-30 18:43:49 +03001443
Jens Axboeb18032b2021-01-24 16:58:56 -07001444 if (posted)
1445 io_commit_cqring(ctx);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001446 spin_unlock_irqrestore(&ctx->completion_lock, flags);
Jens Axboeb18032b2021-01-24 16:58:56 -07001447 if (posted)
1448 io_cqring_ev_posted(ctx);
Pavel Begunkov09e88402020-12-17 00:24:38 +00001449 return all_flushed;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001450}
1451
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00001452static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force)
Pavel Begunkov6c503152021-01-04 20:36:36 +00001453{
Jens Axboeca0a2652021-03-04 17:15:48 -07001454 bool ret = true;
1455
Pavel Begunkov6c503152021-01-04 20:36:36 +00001456 if (test_bit(0, &ctx->cq_check_overflow)) {
1457 /* iopoll syncs against uring_lock, not completion_lock */
1458 if (ctx->flags & IORING_SETUP_IOPOLL)
1459 mutex_lock(&ctx->uring_lock);
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00001460 ret = __io_cqring_overflow_flush(ctx, force);
Pavel Begunkov6c503152021-01-04 20:36:36 +00001461 if (ctx->flags & IORING_SETUP_IOPOLL)
1462 mutex_unlock(&ctx->uring_lock);
1463 }
Jens Axboeca0a2652021-03-04 17:15:48 -07001464
1465 return ret;
Pavel Begunkov6c503152021-01-04 20:36:36 +00001466}
1467
Jens Axboeabc54d62021-02-24 13:32:30 -07001468/*
1469 * Shamelessly stolen from the mm implementation of page reference checking,
1470 * see commit f958d7b528b1 for details.
1471 */
1472#define req_ref_zero_or_close_to_overflow(req) \
1473 ((unsigned int) atomic_read(&(req->refs)) + 127u <= 127u)
1474
Jens Axboede9b4cc2021-02-24 13:28:27 -07001475static inline bool req_ref_inc_not_zero(struct io_kiocb *req)
1476{
Jens Axboeabc54d62021-02-24 13:32:30 -07001477 return atomic_inc_not_zero(&req->refs);
Jens Axboede9b4cc2021-02-24 13:28:27 -07001478}
1479
1480static inline bool req_ref_sub_and_test(struct io_kiocb *req, int refs)
1481{
Jens Axboeabc54d62021-02-24 13:32:30 -07001482 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
1483 return atomic_sub_and_test(refs, &req->refs);
Jens Axboede9b4cc2021-02-24 13:28:27 -07001484}
1485
1486static inline bool req_ref_put_and_test(struct io_kiocb *req)
1487{
Jens Axboeabc54d62021-02-24 13:32:30 -07001488 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
1489 return atomic_dec_and_test(&req->refs);
Jens Axboede9b4cc2021-02-24 13:28:27 -07001490}
1491
1492static inline void req_ref_put(struct io_kiocb *req)
1493{
Jens Axboeabc54d62021-02-24 13:32:30 -07001494 WARN_ON_ONCE(req_ref_put_and_test(req));
Jens Axboede9b4cc2021-02-24 13:28:27 -07001495}
1496
1497static inline void req_ref_get(struct io_kiocb *req)
1498{
Jens Axboeabc54d62021-02-24 13:32:30 -07001499 WARN_ON_ONCE(req_ref_zero_or_close_to_overflow(req));
1500 atomic_inc(&req->refs);
Jens Axboede9b4cc2021-02-24 13:28:27 -07001501}
1502
Pavel Begunkov8d133262021-04-11 01:46:33 +01001503static bool io_cqring_event_overflow(struct io_kiocb *req, long res,
1504 unsigned int cflags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001505{
Jens Axboe78e19bb2019-11-06 15:21:34 -07001506 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovcce4b8b2021-04-13 02:58:44 +01001507 struct io_overflow_cqe *ocqe;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001508
Pavel Begunkovcce4b8b2021-04-13 02:58:44 +01001509 ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT);
1510 if (!ocqe) {
1511 /*
1512 * If we're in ring overflow flush mode, or in task cancel mode,
1513 * or cannot allocate an overflow entry, then we need to drop it
1514 * on the floor.
1515 */
1516 WRITE_ONCE(ctx->rings->cq_overflow, ++ctx->cached_cq_overflow);
1517 return false;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001518 }
Pavel Begunkovcce4b8b2021-04-13 02:58:44 +01001519 if (list_empty(&ctx->cq_overflow_list)) {
1520 set_bit(0, &ctx->sq_check_overflow);
1521 set_bit(0, &ctx->cq_check_overflow);
1522 ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW;
1523 }
1524 ocqe->cqe.user_data = req->user_data;
1525 ocqe->cqe.res = res;
1526 ocqe->cqe.flags = cflags;
1527 list_add_tail(&ocqe->list, &ctx->cq_overflow_list);
1528 return true;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001529}
1530
Pavel Begunkov8d133262021-04-11 01:46:33 +01001531static inline bool __io_cqring_fill_event(struct io_kiocb *req, long res,
1532 unsigned int cflags)
1533{
1534 struct io_ring_ctx *ctx = req->ctx;
1535 struct io_uring_cqe *cqe;
1536
1537 trace_io_uring_complete(ctx, req->user_data, res, cflags);
1538
1539 /*
1540 * If we can't get a cq entry, userspace overflowed the
1541 * submission (by quite a lot). Increment the overflow count in
1542 * the ring.
1543 */
1544 cqe = io_get_cqring(ctx);
1545 if (likely(cqe)) {
1546 WRITE_ONCE(cqe->user_data, req->user_data);
1547 WRITE_ONCE(cqe->res, res);
1548 WRITE_ONCE(cqe->flags, cflags);
1549 return true;
1550 }
1551 return io_cqring_event_overflow(req, res, cflags);
1552}
1553
1554/* not as hot to bloat with inlining */
1555static noinline bool io_cqring_fill_event(struct io_kiocb *req, long res,
1556 unsigned int cflags)
1557{
1558 return __io_cqring_fill_event(req, res, cflags);
1559}
1560
Pavel Begunkov7a612352021-03-09 00:37:59 +00001561static void io_req_complete_post(struct io_kiocb *req, long res,
1562 unsigned int cflags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001563{
Jens Axboe78e19bb2019-11-06 15:21:34 -07001564 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001565 unsigned long flags;
1566
1567 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkov8d133262021-04-11 01:46:33 +01001568 __io_cqring_fill_event(req, res, cflags);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001569 /*
1570 * If we're the last reference to this request, add to our locked
1571 * free_list cache.
1572 */
Jens Axboede9b4cc2021-02-24 13:28:27 -07001573 if (req_ref_put_and_test(req)) {
Jens Axboec7dae4b2021-02-09 19:53:37 -07001574 struct io_comp_state *cs = &ctx->submit_state.comp;
1575
Pavel Begunkov7a612352021-03-09 00:37:59 +00001576 if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) {
1577 if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK))
1578 io_disarm_next(req);
1579 if (req->link) {
1580 io_req_task_queue(req->link);
1581 req->link = NULL;
1582 }
1583 }
Jens Axboec7dae4b2021-02-09 19:53:37 -07001584 io_dismantle_req(req);
1585 io_put_task(req->task, 1);
1586 list_add(&req->compl.list, &cs->locked_free_list);
1587 cs->locked_free_nr++;
Pavel Begunkov180f8292021-03-14 20:57:09 +00001588 } else {
1589 if (!percpu_ref_tryget(&ctx->refs))
1590 req = NULL;
1591 }
Pavel Begunkov7a612352021-03-09 00:37:59 +00001592 io_commit_cqring(ctx);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001593 spin_unlock_irqrestore(&ctx->completion_lock, flags);
Pavel Begunkov7a612352021-03-09 00:37:59 +00001594
Pavel Begunkov180f8292021-03-14 20:57:09 +00001595 if (req) {
1596 io_cqring_ev_posted(ctx);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001597 percpu_ref_put(&ctx->refs);
Pavel Begunkov180f8292021-03-14 20:57:09 +00001598 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07001599}
1600
Jens Axboe4e3d9ff2021-04-15 17:44:34 -06001601static inline bool io_req_needs_clean(struct io_kiocb *req)
1602{
Jens Axboe75652a302021-04-15 09:52:40 -06001603 return req->flags & (REQ_F_BUFFER_SELECTED | REQ_F_NEED_CLEANUP |
Pavel Begunkov3a0a6902021-04-20 12:03:31 +01001604 REQ_F_POLLED | REQ_F_INFLIGHT);
Jens Axboe4e3d9ff2021-04-15 17:44:34 -06001605}
1606
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001607static void io_req_complete_state(struct io_kiocb *req, long res,
Pavel Begunkov889fca72021-02-10 00:03:09 +00001608 unsigned int cflags)
Jens Axboebcda7ba2020-02-23 16:42:51 -07001609{
Jens Axboe4e3d9ff2021-04-15 17:44:34 -06001610 if (io_req_needs_clean(req))
Pavel Begunkov68fb8972021-03-19 17:22:41 +00001611 io_clean_op(req);
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001612 req->result = res;
1613 req->compl.cflags = cflags;
Pavel Begunkove342c802021-01-19 13:32:47 +00001614 req->flags |= REQ_F_COMPLETE_INLINE;
Jens Axboee1e16092020-06-22 09:17:17 -06001615}
Jens Axboe2b188cc2019-01-07 10:46:33 -07001616
Pavel Begunkov889fca72021-02-10 00:03:09 +00001617static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags,
1618 long res, unsigned cflags)
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001619{
Pavel Begunkov889fca72021-02-10 00:03:09 +00001620 if (issue_flags & IO_URING_F_COMPLETE_DEFER)
1621 io_req_complete_state(req, res, cflags);
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001622 else
Jens Axboec7dae4b2021-02-09 19:53:37 -07001623 io_req_complete_post(req, res, cflags);
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001624}
Jens Axboebcda7ba2020-02-23 16:42:51 -07001625
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001626static inline void io_req_complete(struct io_kiocb *req, long res)
Jens Axboee1e16092020-06-22 09:17:17 -06001627{
Pavel Begunkov889fca72021-02-10 00:03:09 +00001628 __io_req_complete(req, 0, res, 0);
Jens Axboebcda7ba2020-02-23 16:42:51 -07001629}
1630
Pavel Begunkovf41db2732021-02-28 22:35:12 +00001631static void io_req_complete_failed(struct io_kiocb *req, long res)
1632{
1633 req_set_fail_links(req);
1634 io_put_req(req);
1635 io_req_complete_post(req, res, 0);
1636}
1637
Pavel Begunkovdac7a092021-03-19 17:22:39 +00001638static void io_flush_cached_locked_reqs(struct io_ring_ctx *ctx,
1639 struct io_comp_state *cs)
1640{
1641 spin_lock_irq(&ctx->completion_lock);
1642 list_splice_init(&cs->locked_free_list, &cs->free_list);
1643 cs->locked_free_nr = 0;
1644 spin_unlock_irq(&ctx->completion_lock);
1645}
1646
Pavel Begunkovdd78f492021-03-19 17:22:35 +00001647/* Returns true IFF there are requests in the cache */
Jens Axboec7dae4b2021-02-09 19:53:37 -07001648static bool io_flush_cached_reqs(struct io_ring_ctx *ctx)
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001649{
Jens Axboec7dae4b2021-02-09 19:53:37 -07001650 struct io_submit_state *state = &ctx->submit_state;
1651 struct io_comp_state *cs = &state->comp;
Pavel Begunkovdd78f492021-03-19 17:22:35 +00001652 int nr;
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001653
Jens Axboec7dae4b2021-02-09 19:53:37 -07001654 /*
1655 * If we have more than a batch's worth of requests in our IRQ side
1656 * locked cache, grab the lock and move them over to our submission
1657 * side cache.
1658 */
Pavel Begunkovdac7a092021-03-19 17:22:39 +00001659 if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH)
1660 io_flush_cached_locked_reqs(ctx, cs);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001661
Pavel Begunkovdd78f492021-03-19 17:22:35 +00001662 nr = state->free_reqs;
Jens Axboec7dae4b2021-02-09 19:53:37 -07001663 while (!list_empty(&cs->free_list)) {
Pavel Begunkovdd78f492021-03-19 17:22:35 +00001664 struct io_kiocb *req = list_first_entry(&cs->free_list,
1665 struct io_kiocb, compl.list);
1666
Jens Axboe2b188cc2019-01-07 10:46:33 -07001667 list_del(&req->compl.list);
Pavel Begunkovdd78f492021-03-19 17:22:35 +00001668 state->reqs[nr++] = req;
1669 if (nr == ARRAY_SIZE(state->reqs))
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001670 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001671 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07001672
Pavel Begunkovdd78f492021-03-19 17:22:35 +00001673 state->free_reqs = nr;
1674 return nr != 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001675}
1676
Pavel Begunkov258b29a2021-02-10 00:03:10 +00001677static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001678{
Pavel Begunkov258b29a2021-02-10 00:03:10 +00001679 struct io_submit_state *state = &ctx->submit_state;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001680
Pavel Begunkovbf019da2021-02-10 00:03:17 +00001681 BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs));
Jens Axboe2b188cc2019-01-07 10:46:33 -07001682
Pavel Begunkovf6b6c7d2020-06-21 13:09:53 +03001683 if (!state->free_reqs) {
Pavel Begunkov291b2822020-09-30 22:57:01 +03001684 gfp_t gfp = GFP_KERNEL | __GFP_NOWARN;
Jens Axboe2579f912019-01-09 09:10:43 -07001685 int ret;
1686
Jens Axboec7dae4b2021-02-09 19:53:37 -07001687 if (io_flush_cached_reqs(ctx))
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001688 goto got_req;
1689
Pavel Begunkovbf019da2021-02-10 00:03:17 +00001690 ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH,
1691 state->reqs);
Jens Axboefd6fab22019-03-14 16:30:06 -06001692
1693 /*
1694 * Bulk alloc is all-or-nothing. If we fail to get a batch,
1695 * retry single alloc to be on the safe side.
1696 */
1697 if (unlikely(ret <= 0)) {
1698 state->reqs[0] = kmem_cache_alloc(req_cachep, gfp);
1699 if (!state->reqs[0])
Pavel Begunkov3893f392021-02-10 00:03:15 +00001700 return NULL;
Jens Axboefd6fab22019-03-14 16:30:06 -06001701 ret = 1;
1702 }
Pavel Begunkov291b2822020-09-30 22:57:01 +03001703 state->free_reqs = ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001704 }
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001705got_req:
Pavel Begunkov291b2822020-09-30 22:57:01 +03001706 state->free_reqs--;
1707 return state->reqs[state->free_reqs];
Jens Axboe2b188cc2019-01-07 10:46:33 -07001708}
1709
Pavel Begunkove1d767f2021-03-19 17:22:43 +00001710static inline void io_put_file(struct file *file)
Pavel Begunkov8da11c12020-02-24 11:32:44 +03001711{
Pavel Begunkove1d767f2021-03-19 17:22:43 +00001712 if (file)
Pavel Begunkov8da11c12020-02-24 11:32:44 +03001713 fput(file);
1714}
1715
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01001716static void io_dismantle_req(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001717{
Pavel Begunkov094bae42021-03-19 17:22:42 +00001718 unsigned int flags = req->flags;
1719
Pavel Begunkov3a0a6902021-04-20 12:03:31 +01001720 if (io_req_needs_clean(req))
1721 io_clean_op(req);
Pavel Begunkove1d767f2021-03-19 17:22:43 +00001722 if (!(flags & REQ_F_FIXED_FILE))
1723 io_put_file(req->file);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001724 if (req->fixed_rsrc_refs)
1725 percpu_ref_put(req->fixed_rsrc_refs);
Pavel Begunkov094bae42021-03-19 17:22:42 +00001726 if (req->async_data)
1727 kfree(req->async_data);
Jens Axboe003e8dc2021-03-06 09:22:27 -07001728 if (req->work.creds) {
1729 put_cred(req->work.creds);
1730 req->work.creds = NULL;
1731 }
Pavel Begunkove6543a82020-06-28 12:52:30 +03001732}
Pavel Begunkov2b85edf2019-12-28 14:13:03 +03001733
Pavel Begunkovb23fcf42021-03-01 18:20:48 +00001734/* must to be called somewhat shortly after putting a request */
Pavel Begunkov7c660732021-01-25 11:42:21 +00001735static inline void io_put_task(struct task_struct *task, int nr)
1736{
1737 struct io_uring_task *tctx = task->io_uring;
1738
1739 percpu_counter_sub(&tctx->inflight, nr);
1740 if (unlikely(atomic_read(&tctx->in_idle)))
1741 wake_up(&tctx->wait);
1742 put_task_struct_many(task, nr);
1743}
1744
Pavel Begunkov216578e2020-10-13 09:44:00 +01001745static void __io_free_req(struct io_kiocb *req)
Pavel Begunkove6543a82020-06-28 12:52:30 +03001746{
Jens Axboe51a4cc12020-08-10 10:55:56 -06001747 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovecfc5172020-06-29 13:13:03 +03001748
Pavel Begunkov216578e2020-10-13 09:44:00 +01001749 io_dismantle_req(req);
Pavel Begunkov7c660732021-01-25 11:42:21 +00001750 io_put_task(req->task, 1);
Pavel Begunkove6543a82020-06-28 12:52:30 +03001751
Pavel Begunkov3893f392021-02-10 00:03:15 +00001752 kmem_cache_free(req_cachep, req);
Pavel Begunkovecfc5172020-06-29 13:13:03 +03001753 percpu_ref_put(&ctx->refs);
Jens Axboee65ef562019-03-12 10:16:44 -06001754}
1755
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001756static inline void io_remove_next_linked(struct io_kiocb *req)
1757{
1758 struct io_kiocb *nxt = req->link;
1759
1760 req->link = nxt->link;
1761 nxt->link = NULL;
1762}
1763
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001764static bool io_kill_linked_timeout(struct io_kiocb *req)
1765 __must_hold(&req->ctx->completion_lock)
Jens Axboe9e645e112019-05-10 16:07:28 -06001766{
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001767 struct io_kiocb *link = req->link;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001768
Pavel Begunkov900fad42020-10-19 16:39:16 +01001769 /*
1770 * Can happen if a linked timeout fired and link had been like
1771 * req -> link t-out -> link t-out [-> ...]
1772 */
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001773 if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) {
1774 struct io_timeout_data *io = link->async_data;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001775
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001776 io_remove_next_linked(req);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00001777 link->timeout.head = NULL;
Pavel Begunkovfd9c7bc2021-04-13 02:58:42 +01001778 if (hrtimer_try_to_cancel(&io->timer) != -1) {
Pavel Begunkovff6421642021-04-11 01:46:32 +01001779 io_cqring_fill_event(link, -ECANCELED, 0);
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001780 io_put_req_deferred(link, 1);
Pavel Begunkovd4729fb2021-03-22 01:58:24 +00001781 return true;
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001782 }
1783 }
Pavel Begunkovd4729fb2021-03-22 01:58:24 +00001784 return false;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001785}
1786
Pavel Begunkovd148ca42020-10-18 10:17:39 +01001787static void io_fail_links(struct io_kiocb *req)
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001788 __must_hold(&req->ctx->completion_lock)
Jens Axboe9e645e112019-05-10 16:07:28 -06001789{
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001790 struct io_kiocb *nxt, *link = req->link;
Jens Axboe9e645e112019-05-10 16:07:28 -06001791
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001792 req->link = NULL;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001793 while (link) {
1794 nxt = link->link;
1795 link->link = NULL;
1796
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02001797 trace_io_uring_fail_link(req, link);
Pavel Begunkovff6421642021-04-11 01:46:32 +01001798 io_cqring_fill_event(link, -ECANCELED, 0);
Jens Axboe1575f212021-02-27 15:20:49 -07001799 io_put_req_deferred(link, 2);
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001800 link = nxt;
Jens Axboe9e645e112019-05-10 16:07:28 -06001801 }
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001802}
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001803
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001804static bool io_disarm_next(struct io_kiocb *req)
1805 __must_hold(&req->ctx->completion_lock)
1806{
1807 bool posted = false;
1808
1809 if (likely(req->flags & REQ_F_LINK_TIMEOUT))
1810 posted = io_kill_linked_timeout(req);
Pavel Begunkove4335ed2021-04-11 01:46:39 +01001811 if (unlikely((req->flags & REQ_F_FAIL_LINK) &&
1812 !(req->flags & REQ_F_HARDLINK))) {
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001813 posted |= (req->link != NULL);
1814 io_fail_links(req);
1815 }
1816 return posted;
Jens Axboe9e645e112019-05-10 16:07:28 -06001817}
1818
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001819static struct io_kiocb *__io_req_find_next(struct io_kiocb *req)
Jens Axboe9e645e112019-05-10 16:07:28 -06001820{
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001821 struct io_kiocb *nxt;
Jens Axboe2665abf2019-11-05 12:40:47 -07001822
Jens Axboe9e645e112019-05-10 16:07:28 -06001823 /*
1824 * If LINK is set, we have dependent requests in this chain. If we
1825 * didn't fail this request, queue the first one up, moving any other
1826 * dependencies to the next request. In case of failure, fail the rest
1827 * of the chain.
1828 */
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001829 if (req->flags & (REQ_F_LINK_TIMEOUT | REQ_F_FAIL_LINK)) {
1830 struct io_ring_ctx *ctx = req->ctx;
1831 unsigned long flags;
1832 bool posted;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001833
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001834 spin_lock_irqsave(&ctx->completion_lock, flags);
1835 posted = io_disarm_next(req);
1836 if (posted)
1837 io_commit_cqring(req->ctx);
1838 spin_unlock_irqrestore(&ctx->completion_lock, flags);
1839 if (posted)
1840 io_cqring_ev_posted(ctx);
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001841 }
Pavel Begunkov33cc89a2021-03-09 00:37:58 +00001842 nxt = req->link;
1843 req->link = NULL;
1844 return nxt;
Jens Axboe4d7dd462019-11-20 13:03:52 -07001845}
Jens Axboe2665abf2019-11-05 12:40:47 -07001846
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001847static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req)
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001848{
Pavel Begunkovcdbff982021-02-12 18:41:16 +00001849 if (likely(!(req->flags & (REQ_F_LINK|REQ_F_HARDLINK))))
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001850 return NULL;
1851 return __io_req_find_next(req);
1852}
1853
Pavel Begunkov2c323952021-02-28 22:04:53 +00001854static void ctx_flush_and_put(struct io_ring_ctx *ctx)
1855{
1856 if (!ctx)
1857 return;
1858 if (ctx->submit_state.comp.nr) {
1859 mutex_lock(&ctx->uring_lock);
1860 io_submit_flush_completions(&ctx->submit_state.comp, ctx);
1861 mutex_unlock(&ctx->uring_lock);
1862 }
1863 percpu_ref_put(&ctx->refs);
1864}
1865
Jens Axboe7cbf1722021-02-10 00:03:20 +00001866static bool __tctx_task_work(struct io_uring_task *tctx)
1867{
Jens Axboe65453d12021-02-10 00:03:21 +00001868 struct io_ring_ctx *ctx = NULL;
Jens Axboe7cbf1722021-02-10 00:03:20 +00001869 struct io_wq_work_list list;
1870 struct io_wq_work_node *node;
1871
1872 if (wq_list_empty(&tctx->task_list))
1873 return false;
1874
Jens Axboe0b81e802021-02-16 10:33:53 -07001875 spin_lock_irq(&tctx->task_lock);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001876 list = tctx->task_list;
1877 INIT_WQ_LIST(&tctx->task_list);
Jens Axboe0b81e802021-02-16 10:33:53 -07001878 spin_unlock_irq(&tctx->task_lock);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001879
1880 node = list.first;
1881 while (node) {
1882 struct io_wq_work_node *next = node->next;
1883 struct io_kiocb *req;
1884
1885 req = container_of(node, struct io_kiocb, io_task_work.node);
Pavel Begunkov2c323952021-02-28 22:04:53 +00001886 if (req->ctx != ctx) {
1887 ctx_flush_and_put(ctx);
1888 ctx = req->ctx;
1889 percpu_ref_get(&ctx->refs);
1890 }
1891
Jens Axboe7cbf1722021-02-10 00:03:20 +00001892 req->task_work.func(&req->task_work);
1893 node = next;
Jens Axboe65453d12021-02-10 00:03:21 +00001894 }
1895
Pavel Begunkov2c323952021-02-28 22:04:53 +00001896 ctx_flush_and_put(ctx);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001897 return list.first != NULL;
1898}
1899
1900static void tctx_task_work(struct callback_head *cb)
1901{
1902 struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work);
1903
Jens Axboe1d5f3602021-02-26 14:54:16 -07001904 clear_bit(0, &tctx->task_state);
1905
Jens Axboe7cbf1722021-02-10 00:03:20 +00001906 while (__tctx_task_work(tctx))
1907 cond_resched();
Jens Axboe7cbf1722021-02-10 00:03:20 +00001908}
1909
Pavel Begunkovc15b79d2021-03-19 17:22:44 +00001910static int io_req_task_work_add(struct io_kiocb *req)
Jens Axboe7cbf1722021-02-10 00:03:20 +00001911{
Pavel Begunkovc15b79d2021-03-19 17:22:44 +00001912 struct task_struct *tsk = req->task;
Jens Axboe7cbf1722021-02-10 00:03:20 +00001913 struct io_uring_task *tctx = tsk->io_uring;
Pavel Begunkovc15b79d2021-03-19 17:22:44 +00001914 enum task_work_notify_mode notify;
Jens Axboe7cbf1722021-02-10 00:03:20 +00001915 struct io_wq_work_node *node, *prev;
Jens Axboe0b81e802021-02-16 10:33:53 -07001916 unsigned long flags;
Pavel Begunkovc15b79d2021-03-19 17:22:44 +00001917 int ret = 0;
1918
1919 if (unlikely(tsk->flags & PF_EXITING))
1920 return -ESRCH;
Jens Axboe7cbf1722021-02-10 00:03:20 +00001921
1922 WARN_ON_ONCE(!tctx);
1923
Jens Axboe0b81e802021-02-16 10:33:53 -07001924 spin_lock_irqsave(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001925 wq_list_add_tail(&req->io_task_work.node, &tctx->task_list);
Jens Axboe0b81e802021-02-16 10:33:53 -07001926 spin_unlock_irqrestore(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001927
1928 /* task_work already pending, we're done */
1929 if (test_bit(0, &tctx->task_state) ||
1930 test_and_set_bit(0, &tctx->task_state))
1931 return 0;
1932
Pavel Begunkovc15b79d2021-03-19 17:22:44 +00001933 /*
1934 * SQPOLL kernel thread doesn't need notification, just a wakeup. For
1935 * all other cases, use TWA_SIGNAL unconditionally to ensure we're
1936 * processing task_work. There's no reliable way to tell if TWA_RESUME
1937 * will do the job.
1938 */
1939 notify = (req->ctx->flags & IORING_SETUP_SQPOLL) ? TWA_NONE : TWA_SIGNAL;
1940
1941 if (!task_work_add(tsk, &tctx->task_work, notify)) {
1942 wake_up_process(tsk);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001943 return 0;
Pavel Begunkovc15b79d2021-03-19 17:22:44 +00001944 }
Jens Axboe7cbf1722021-02-10 00:03:20 +00001945
1946 /*
1947 * Slow path - we failed, find and delete work. if the work is not
1948 * in the list, it got run and we're fine.
1949 */
Jens Axboe0b81e802021-02-16 10:33:53 -07001950 spin_lock_irqsave(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001951 wq_list_for_each(node, prev, &tctx->task_list) {
1952 if (&req->io_task_work.node == node) {
1953 wq_list_del(&tctx->task_list, node, prev);
1954 ret = 1;
1955 break;
1956 }
1957 }
Jens Axboe0b81e802021-02-16 10:33:53 -07001958 spin_unlock_irqrestore(&tctx->task_lock, flags);
Jens Axboe7cbf1722021-02-10 00:03:20 +00001959 clear_bit(0, &tctx->task_state);
1960 return ret;
1961}
1962
Pavel Begunkov9b465712021-03-15 14:23:07 +00001963static bool io_run_task_work_head(struct callback_head **work_head)
1964{
1965 struct callback_head *work, *next;
1966 bool executed = false;
1967
1968 do {
1969 work = xchg(work_head, NULL);
1970 if (!work)
1971 break;
1972
1973 do {
1974 next = work->next;
1975 work->func(work);
1976 work = next;
1977 cond_resched();
1978 } while (work);
1979 executed = true;
1980 } while (1);
1981
1982 return executed;
1983}
1984
1985static void io_task_work_add_head(struct callback_head **work_head,
1986 struct callback_head *task_work)
1987{
1988 struct callback_head *head;
1989
1990 do {
1991 head = READ_ONCE(*work_head);
1992 task_work->next = head;
1993 } while (cmpxchg(work_head, head, task_work) != head);
1994}
1995
Pavel Begunkoveab30c42021-01-19 13:32:42 +00001996static void io_req_task_work_add_fallback(struct io_kiocb *req,
Jens Axboe7cbf1722021-02-10 00:03:20 +00001997 task_work_func_t cb)
Pavel Begunkoveab30c42021-01-19 13:32:42 +00001998{
Pavel Begunkoveab30c42021-01-19 13:32:42 +00001999 init_task_work(&req->task_work, cb);
Pavel Begunkov9b465712021-03-15 14:23:07 +00002000 io_task_work_add_head(&req->ctx->exit_task_work, &req->task_work);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002001}
2002
Jens Axboec40f6372020-06-25 15:39:59 -06002003static void io_req_task_cancel(struct callback_head *cb)
2004{
2005 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
Jens Axboe87ceb6a2020-09-14 08:20:12 -06002006 struct io_ring_ctx *ctx = req->ctx;
Jens Axboec40f6372020-06-25 15:39:59 -06002007
Pavel Begunkove83acd72021-02-28 22:35:09 +00002008 /* ctx is guaranteed to stay alive while we hold uring_lock */
Pavel Begunkov792bb6e2021-02-18 22:32:51 +00002009 mutex_lock(&ctx->uring_lock);
Pavel Begunkov25935532021-03-19 17:22:40 +00002010 io_req_complete_failed(req, req->result);
Pavel Begunkov792bb6e2021-02-18 22:32:51 +00002011 mutex_unlock(&ctx->uring_lock);
Jens Axboec40f6372020-06-25 15:39:59 -06002012}
2013
2014static void __io_req_task_submit(struct io_kiocb *req)
2015{
2016 struct io_ring_ctx *ctx = req->ctx;
2017
Pavel Begunkov04fc6c82021-02-12 03:23:54 +00002018 /* ctx stays valid until unlock, even if we drop all ours ctx->refs */
Pavel Begunkov81b6d052021-01-04 20:36:35 +00002019 mutex_lock(&ctx->uring_lock);
Pavel Begunkov70aacfe2021-03-01 13:02:15 +00002020 if (!(current->flags & PF_EXITING) && !current->in_execve)
Pavel Begunkovc5eef2b942021-02-10 00:03:22 +00002021 __io_queue_sqe(req);
Pavel Begunkov81b6d052021-01-04 20:36:35 +00002022 else
Pavel Begunkov25935532021-03-19 17:22:40 +00002023 io_req_complete_failed(req, -EFAULT);
Pavel Begunkov81b6d052021-01-04 20:36:35 +00002024 mutex_unlock(&ctx->uring_lock);
Jens Axboe9e645e112019-05-10 16:07:28 -06002025}
2026
Jens Axboec40f6372020-06-25 15:39:59 -06002027static void io_req_task_submit(struct callback_head *cb)
2028{
2029 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
2030
2031 __io_req_task_submit(req);
2032}
2033
Pavel Begunkova3df76982021-02-18 22:32:52 +00002034static void io_req_task_queue_fail(struct io_kiocb *req, int ret)
2035{
Pavel Begunkova3df76982021-02-18 22:32:52 +00002036 req->result = ret;
2037 req->task_work.func = io_req_task_cancel;
2038
2039 if (unlikely(io_req_task_work_add(req)))
2040 io_req_task_work_add_fallback(req, io_req_task_cancel);
2041}
2042
Pavel Begunkov2c4b8eb2021-02-28 22:35:10 +00002043static void io_req_task_queue(struct io_kiocb *req)
2044{
2045 req->task_work.func = io_req_task_submit;
2046
2047 if (unlikely(io_req_task_work_add(req)))
2048 io_req_task_queue_fail(req, -ECANCELED);
2049}
2050
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002051static inline void io_queue_next(struct io_kiocb *req)
Jackie Liuc69f8db2019-11-09 11:00:08 +08002052{
Pavel Begunkov9b5f7bd92020-06-29 13:13:00 +03002053 struct io_kiocb *nxt = io_req_find_next(req);
Pavel Begunkov944e58b2019-11-21 23:21:01 +03002054
Pavel Begunkov906a8c32020-06-27 14:04:55 +03002055 if (nxt)
2056 io_req_task_queue(nxt);
Jackie Liuc69f8db2019-11-09 11:00:08 +08002057}
2058
Jens Axboe9e645e112019-05-10 16:07:28 -06002059static void io_free_req(struct io_kiocb *req)
2060{
Pavel Begunkovc3524382020-06-28 12:52:32 +03002061 io_queue_next(req);
Jens Axboe9e645e112019-05-10 16:07:28 -06002062 __io_free_req(req);
Jens Axboee65ef562019-03-12 10:16:44 -06002063}
2064
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002065struct req_batch {
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002066 struct task_struct *task;
2067 int task_refs;
Jens Axboe1b4c3512021-02-10 00:03:19 +00002068 int ctx_refs;
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002069};
2070
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002071static inline void io_init_req_batch(struct req_batch *rb)
Pavel Begunkov7a743e22020-03-03 21:33:13 +03002072{
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002073 rb->task_refs = 0;
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002074 rb->ctx_refs = 0;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002075 rb->task = NULL;
2076}
Pavel Begunkov8766dd52020-03-14 00:31:04 +03002077
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002078static void io_req_free_batch_finish(struct io_ring_ctx *ctx,
2079 struct req_batch *rb)
2080{
Pavel Begunkov6e833d52021-02-11 18:28:20 +00002081 if (rb->task)
Pavel Begunkov7c660732021-01-25 11:42:21 +00002082 io_put_task(rb->task, rb->task_refs);
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002083 if (rb->ctx_refs)
2084 percpu_ref_put_many(&ctx->refs, rb->ctx_refs);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002085}
2086
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002087static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req,
2088 struct io_submit_state *state)
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002089{
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002090 io_queue_next(req);
Pavel Begunkov96670652021-03-19 17:22:32 +00002091 io_dismantle_req(req);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002092
Jens Axboee3bc8e92020-09-24 08:45:57 -06002093 if (req->task != rb->task) {
Pavel Begunkov7c660732021-01-25 11:42:21 +00002094 if (rb->task)
2095 io_put_task(rb->task, rb->task_refs);
Jens Axboee3bc8e92020-09-24 08:45:57 -06002096 rb->task = req->task;
2097 rb->task_refs = 0;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002098 }
Jens Axboee3bc8e92020-09-24 08:45:57 -06002099 rb->task_refs++;
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002100 rb->ctx_refs++;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002101
Pavel Begunkovbd759042021-02-12 03:23:50 +00002102 if (state->free_reqs != ARRAY_SIZE(state->reqs))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002103 state->reqs[state->free_reqs++] = req;
Pavel Begunkovbd759042021-02-12 03:23:50 +00002104 else
2105 list_add(&req->compl.list, &state->comp.free_list);
Pavel Begunkov7a743e22020-03-03 21:33:13 +03002106}
2107
Pavel Begunkov905c1722021-02-10 00:03:14 +00002108static void io_submit_flush_completions(struct io_comp_state *cs,
2109 struct io_ring_ctx *ctx)
2110{
2111 int i, nr = cs->nr;
2112 struct io_kiocb *req;
2113 struct req_batch rb;
2114
2115 io_init_req_batch(&rb);
2116 spin_lock_irq(&ctx->completion_lock);
2117 for (i = 0; i < nr; i++) {
2118 req = cs->reqs[i];
Pavel Begunkov8d133262021-04-11 01:46:33 +01002119 __io_cqring_fill_event(req, req->result, req->compl.cflags);
Pavel Begunkov905c1722021-02-10 00:03:14 +00002120 }
2121 io_commit_cqring(ctx);
2122 spin_unlock_irq(&ctx->completion_lock);
2123
2124 io_cqring_ev_posted(ctx);
2125 for (i = 0; i < nr; i++) {
2126 req = cs->reqs[i];
2127
2128 /* submission and completion refs */
Jens Axboede9b4cc2021-02-24 13:28:27 -07002129 if (req_ref_sub_and_test(req, 2))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002130 io_req_free_batch(&rb, req, &ctx->submit_state);
Pavel Begunkov905c1722021-02-10 00:03:14 +00002131 }
2132
2133 io_req_free_batch_finish(ctx, &rb);
2134 cs->nr = 0;
Jens Axboee65ef562019-03-12 10:16:44 -06002135}
2136
Jens Axboeba816ad2019-09-28 11:36:45 -06002137/*
2138 * Drop reference to request, return next in chain (if there is one) if this
2139 * was the last reference to this request.
2140 */
Pavel Begunkov0d850352021-03-19 17:22:37 +00002141static inline struct io_kiocb *io_put_req_find_next(struct io_kiocb *req)
Jens Axboee65ef562019-03-12 10:16:44 -06002142{
Pavel Begunkov9b5f7bd92020-06-29 13:13:00 +03002143 struct io_kiocb *nxt = NULL;
2144
Jens Axboede9b4cc2021-02-24 13:28:27 -07002145 if (req_ref_put_and_test(req)) {
Pavel Begunkov9b5f7bd92020-06-29 13:13:00 +03002146 nxt = io_req_find_next(req);
Jens Axboe4d7dd462019-11-20 13:03:52 -07002147 __io_free_req(req);
Jens Axboe2a44f462020-02-25 13:25:41 -07002148 }
Pavel Begunkov9b5f7bd92020-06-29 13:13:00 +03002149 return nxt;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002150}
2151
Pavel Begunkov0d850352021-03-19 17:22:37 +00002152static inline void io_put_req(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002153{
Jens Axboede9b4cc2021-02-24 13:28:27 -07002154 if (req_ref_put_and_test(req))
Jens Axboedef596e2019-01-09 08:59:42 -07002155 io_free_req(req);
2156}
2157
Pavel Begunkov216578e2020-10-13 09:44:00 +01002158static void io_put_req_deferred_cb(struct callback_head *cb)
2159{
2160 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
2161
2162 io_free_req(req);
2163}
2164
2165static void io_free_req_deferred(struct io_kiocb *req)
2166{
Jens Axboe7cbf1722021-02-10 00:03:20 +00002167 req->task_work.func = io_put_req_deferred_cb;
Pavel Begunkova05432f2021-03-19 17:22:38 +00002168 if (unlikely(io_req_task_work_add(req)))
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002169 io_req_task_work_add_fallback(req, io_put_req_deferred_cb);
Pavel Begunkov216578e2020-10-13 09:44:00 +01002170}
2171
2172static inline void io_put_req_deferred(struct io_kiocb *req, int refs)
2173{
Jens Axboede9b4cc2021-02-24 13:28:27 -07002174 if (req_ref_sub_and_test(req, refs))
Pavel Begunkov216578e2020-10-13 09:44:00 +01002175 io_free_req_deferred(req);
2176}
2177
Pavel Begunkov6c503152021-01-04 20:36:36 +00002178static unsigned io_cqring_events(struct io_ring_ctx *ctx)
Jens Axboea3a0e432019-08-20 11:03:11 -06002179{
2180 /* See comment at the top of this file */
2181 smp_rmb();
Pavel Begunkove23de152020-12-17 00:24:37 +00002182 return __io_cqring_events(ctx);
Jens Axboea3a0e432019-08-20 11:03:11 -06002183}
2184
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03002185static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx)
2186{
2187 struct io_rings *rings = ctx->rings;
2188
2189 /* make sure SQ entry isn't read before tail */
2190 return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head;
2191}
2192
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002193static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf)
Jens Axboee94f1412019-12-19 12:06:02 -07002194{
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002195 unsigned int cflags;
Jens Axboee94f1412019-12-19 12:06:02 -07002196
Jens Axboebcda7ba2020-02-23 16:42:51 -07002197 cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT;
2198 cflags |= IORING_CQE_F_BUFFER;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03002199 req->flags &= ~REQ_F_BUFFER_SELECTED;
Jens Axboebcda7ba2020-02-23 16:42:51 -07002200 kfree(kbuf);
2201 return cflags;
2202}
2203
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002204static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req)
2205{
2206 struct io_buffer *kbuf;
2207
2208 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
2209 return io_put_kbuf(req, kbuf);
2210}
2211
Jens Axboe4c6e2772020-07-01 11:29:10 -06002212static inline bool io_run_task_work(void)
2213{
Jens Axboe6200b0a2020-09-13 14:38:30 -06002214 /*
2215 * Not safe to run on exiting task, and the task_work handling will
2216 * not add work to such a task.
2217 */
2218 if (unlikely(current->flags & PF_EXITING))
2219 return false;
Jens Axboe4c6e2772020-07-01 11:29:10 -06002220 if (current->task_works) {
2221 __set_current_state(TASK_RUNNING);
2222 task_work_run();
2223 return true;
2224 }
2225
2226 return false;
2227}
2228
Jens Axboedef596e2019-01-09 08:59:42 -07002229/*
2230 * Find and free completed poll iocbs
2231 */
2232static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events,
2233 struct list_head *done)
2234{
Jens Axboe8237e042019-12-28 10:48:22 -07002235 struct req_batch rb;
Jens Axboedef596e2019-01-09 08:59:42 -07002236 struct io_kiocb *req;
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002237
2238 /* order with ->result store in io_complete_rw_iopoll() */
2239 smp_rmb();
Jens Axboedef596e2019-01-09 08:59:42 -07002240
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002241 io_init_req_batch(&rb);
Jens Axboedef596e2019-01-09 08:59:42 -07002242 while (!list_empty(done)) {
Jens Axboebcda7ba2020-02-23 16:42:51 -07002243 int cflags = 0;
2244
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002245 req = list_first_entry(done, struct io_kiocb, inflight_entry);
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002246 list_del(&req->inflight_entry);
Jens Axboedef596e2019-01-09 08:59:42 -07002247
Pavel Begunkov8c130822021-03-22 01:58:32 +00002248 if (READ_ONCE(req->result) == -EAGAIN &&
2249 !(req->flags & REQ_F_DONT_REISSUE)) {
Pavel Begunkovf1613402021-02-11 18:28:21 +00002250 req->iopoll_completed = 0;
Pavel Begunkov8c130822021-03-22 01:58:32 +00002251 req_ref_get(req);
2252 io_queue_async_work(req);
2253 continue;
Pavel Begunkovf1613402021-02-11 18:28:21 +00002254 }
2255
Jens Axboebcda7ba2020-02-23 16:42:51 -07002256 if (req->flags & REQ_F_BUFFER_SELECTED)
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002257 cflags = io_put_rw_kbuf(req);
Jens Axboebcda7ba2020-02-23 16:42:51 -07002258
Pavel Begunkov8d133262021-04-11 01:46:33 +01002259 __io_cqring_fill_event(req, req->result, cflags);
Jens Axboedef596e2019-01-09 08:59:42 -07002260 (*nr_events)++;
2261
Jens Axboede9b4cc2021-02-24 13:28:27 -07002262 if (req_ref_put_and_test(req))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002263 io_req_free_batch(&rb, req, &ctx->submit_state);
Jens Axboedef596e2019-01-09 08:59:42 -07002264 }
Jens Axboedef596e2019-01-09 08:59:42 -07002265
Jens Axboe09bb8392019-03-13 12:39:28 -06002266 io_commit_cqring(ctx);
Pavel Begunkov80c18e42021-01-07 03:15:41 +00002267 io_cqring_ev_posted_iopoll(ctx);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002268 io_req_free_batch_finish(ctx, &rb);
Bijan Mottahedeh581f9812020-04-03 13:51:33 -07002269}
2270
Jens Axboedef596e2019-01-09 08:59:42 -07002271static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events,
2272 long min)
2273{
2274 struct io_kiocb *req, *tmp;
2275 LIST_HEAD(done);
2276 bool spin;
2277 int ret;
2278
2279 /*
2280 * Only spin for completions if we don't have multiple devices hanging
2281 * off our complete list, and we're under the requested amount.
2282 */
2283 spin = !ctx->poll_multi_file && *nr_events < min;
2284
2285 ret = 0;
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002286 list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) {
Jens Axboe9adbd452019-12-20 08:45:55 -07002287 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboedef596e2019-01-09 08:59:42 -07002288
2289 /*
Bijan Mottahedeh581f9812020-04-03 13:51:33 -07002290 * Move completed and retryable entries to our local lists.
2291 * If we find a request that requires polling, break out
2292 * and complete those lists first, if we have entries there.
Jens Axboedef596e2019-01-09 08:59:42 -07002293 */
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002294 if (READ_ONCE(req->iopoll_completed)) {
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002295 list_move_tail(&req->inflight_entry, &done);
Jens Axboedef596e2019-01-09 08:59:42 -07002296 continue;
2297 }
2298 if (!list_empty(&done))
2299 break;
2300
2301 ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin);
2302 if (ret < 0)
2303 break;
2304
Pavel Begunkov3aadc232020-07-06 17:59:29 +03002305 /* iopoll may have completed current req */
2306 if (READ_ONCE(req->iopoll_completed))
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002307 list_move_tail(&req->inflight_entry, &done);
Pavel Begunkov3aadc232020-07-06 17:59:29 +03002308
Jens Axboedef596e2019-01-09 08:59:42 -07002309 if (ret && spin)
2310 spin = false;
2311 ret = 0;
2312 }
2313
2314 if (!list_empty(&done))
2315 io_iopoll_complete(ctx, nr_events, &done);
2316
2317 return ret;
2318}
2319
2320/*
Jens Axboedef596e2019-01-09 08:59:42 -07002321 * We can't just wait for polled events to come to us, we have to actively
2322 * find and complete them.
2323 */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002324static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx)
Jens Axboedef596e2019-01-09 08:59:42 -07002325{
2326 if (!(ctx->flags & IORING_SETUP_IOPOLL))
2327 return;
2328
2329 mutex_lock(&ctx->uring_lock);
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002330 while (!list_empty(&ctx->iopoll_list)) {
Jens Axboedef596e2019-01-09 08:59:42 -07002331 unsigned int nr_events = 0;
2332
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002333 io_do_iopoll(ctx, &nr_events, 0);
Jens Axboe08f54392019-08-21 22:19:11 -06002334
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002335 /* let it sleep and repeat later if can't complete a request */
2336 if (nr_events == 0)
2337 break;
Jens Axboe08f54392019-08-21 22:19:11 -06002338 /*
2339 * Ensure we allow local-to-the-cpu processing to take place,
2340 * in this case we need to ensure that we reap all events.
Pavel Begunkov3fcee5a2020-07-06 17:59:31 +03002341 * Also let task_work, etc. to progress by releasing the mutex
Jens Axboe08f54392019-08-21 22:19:11 -06002342 */
Pavel Begunkov3fcee5a2020-07-06 17:59:31 +03002343 if (need_resched()) {
2344 mutex_unlock(&ctx->uring_lock);
2345 cond_resched();
2346 mutex_lock(&ctx->uring_lock);
2347 }
Jens Axboedef596e2019-01-09 08:59:42 -07002348 }
2349 mutex_unlock(&ctx->uring_lock);
2350}
2351
Pavel Begunkov7668b922020-07-07 16:36:21 +03002352static int io_iopoll_check(struct io_ring_ctx *ctx, long min)
Jens Axboedef596e2019-01-09 08:59:42 -07002353{
Pavel Begunkov7668b922020-07-07 16:36:21 +03002354 unsigned int nr_events = 0;
Pavel Begunkove9979b32021-04-13 02:58:45 +01002355 int ret = 0;
Jens Axboedef596e2019-01-09 08:59:42 -07002356
Xiaoguang Wangc7849be2020-02-22 14:46:05 +08002357 /*
2358 * We disallow the app entering submit/complete with polling, but we
2359 * still need to lock the ring to prevent racing with polled issue
2360 * that got punted to a workqueue.
2361 */
2362 mutex_lock(&ctx->uring_lock);
Pavel Begunkovf39c8a52021-04-13 02:58:46 +01002363 /*
2364 * Don't enter poll loop if we already have events pending.
2365 * If we do, we can potentially be spinning for commands that
2366 * already triggered a CQE (eg in error).
2367 */
2368 if (test_bit(0, &ctx->cq_check_overflow))
2369 __io_cqring_overflow_flush(ctx, false);
2370 if (io_cqring_events(ctx))
2371 goto out;
Jens Axboedef596e2019-01-09 08:59:42 -07002372 do {
Jens Axboe500f9fb2019-08-19 12:15:59 -06002373 /*
2374 * If a submit got punted to a workqueue, we can have the
2375 * application entering polling for a command before it gets
2376 * issued. That app will hold the uring_lock for the duration
2377 * of the poll right here, so we need to take a breather every
2378 * now and then to ensure that the issue has a chance to add
2379 * the poll to the issued list. Otherwise we can spin here
2380 * forever, while the workqueue is stuck trying to acquire the
2381 * very same mutex.
2382 */
Pavel Begunkove9979b32021-04-13 02:58:45 +01002383 if (list_empty(&ctx->iopoll_list)) {
Jens Axboe500f9fb2019-08-19 12:15:59 -06002384 mutex_unlock(&ctx->uring_lock);
Jens Axboe4c6e2772020-07-01 11:29:10 -06002385 io_run_task_work();
Jens Axboe500f9fb2019-08-19 12:15:59 -06002386 mutex_lock(&ctx->uring_lock);
Pavel Begunkove9979b32021-04-13 02:58:45 +01002387
2388 if (list_empty(&ctx->iopoll_list))
2389 break;
Jens Axboe500f9fb2019-08-19 12:15:59 -06002390 }
Pavel Begunkovf39c8a52021-04-13 02:58:46 +01002391 ret = io_do_iopoll(ctx, &nr_events, min);
2392 } while (!ret && nr_events < min && !need_resched());
2393out:
Jens Axboe500f9fb2019-08-19 12:15:59 -06002394 mutex_unlock(&ctx->uring_lock);
Jens Axboedef596e2019-01-09 08:59:42 -07002395 return ret;
2396}
2397
Jens Axboe491381ce2019-10-17 09:20:46 -06002398static void kiocb_end_write(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002399{
Jens Axboe491381ce2019-10-17 09:20:46 -06002400 /*
2401 * Tell lockdep we inherited freeze protection from submission
2402 * thread.
2403 */
2404 if (req->flags & REQ_F_ISREG) {
Pavel Begunkov1c986792021-03-22 01:58:31 +00002405 struct super_block *sb = file_inode(req->file)->i_sb;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002406
Pavel Begunkov1c986792021-03-22 01:58:31 +00002407 __sb_writers_acquired(sb, SB_FREEZE_WRITE);
2408 sb_end_write(sb);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002409 }
2410}
2411
Jens Axboeb63534c2020-06-04 11:28:00 -06002412#ifdef CONFIG_BLOCK
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002413static bool io_resubmit_prep(struct io_kiocb *req)
Jens Axboeb63534c2020-06-04 11:28:00 -06002414{
Pavel Begunkovab454432021-03-22 01:58:33 +00002415 struct io_async_rw *rw = req->async_data;
2416
2417 if (!rw)
2418 return !io_req_prep_async(req);
2419 /* may have left rw->iter inconsistent on -EIOCBQUEUED */
2420 iov_iter_revert(&rw->iter, req->result - iov_iter_count(&rw->iter));
2421 return true;
Jens Axboeb63534c2020-06-04 11:28:00 -06002422}
Jens Axboeb63534c2020-06-04 11:28:00 -06002423
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002424static bool io_rw_should_reissue(struct io_kiocb *req)
Jens Axboeb63534c2020-06-04 11:28:00 -06002425{
Jens Axboe355afae2020-09-02 09:30:31 -06002426 umode_t mode = file_inode(req->file)->i_mode;
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002427 struct io_ring_ctx *ctx = req->ctx;
Jens Axboeb63534c2020-06-04 11:28:00 -06002428
Jens Axboe355afae2020-09-02 09:30:31 -06002429 if (!S_ISBLK(mode) && !S_ISREG(mode))
2430 return false;
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002431 if ((req->flags & REQ_F_NOWAIT) || (io_wq_current_is_worker() &&
2432 !(ctx->flags & IORING_SETUP_IOPOLL)))
Jens Axboeb63534c2020-06-04 11:28:00 -06002433 return false;
Jens Axboe7c977a52021-02-23 19:17:35 -07002434 /*
2435 * If ref is dying, we might be running poll reap from the exit work.
2436 * Don't attempt to reissue from that path, just let it fail with
2437 * -EAGAIN.
2438 */
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002439 if (percpu_ref_is_dying(&ctx->refs))
2440 return false;
2441 return true;
2442}
Jens Axboee82ad482021-04-02 19:45:34 -06002443#else
Jens Axboea1ff1e32021-04-12 06:40:02 -06002444static bool io_resubmit_prep(struct io_kiocb *req)
2445{
2446 return false;
2447}
Jens Axboee82ad482021-04-02 19:45:34 -06002448static bool io_rw_should_reissue(struct io_kiocb *req)
2449{
2450 return false;
2451}
Jens Axboe3e6a0d32021-03-01 13:56:00 -07002452#endif
2453
Jens Axboea1d7c392020-06-22 11:09:46 -06002454static void __io_complete_rw(struct io_kiocb *req, long res, long res2,
Pavel Begunkov889fca72021-02-10 00:03:09 +00002455 unsigned int issue_flags)
Jens Axboea1d7c392020-06-22 11:09:46 -06002456{
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002457 int cflags = 0;
2458
Pavel Begunkovb65c1282021-03-22 01:45:59 +00002459 if (req->rw.kiocb.ki_flags & IOCB_WRITE)
2460 kiocb_end_write(req);
Pavel Begunkov9532b992021-03-22 01:58:34 +00002461 if (res != req->result) {
2462 if ((res == -EAGAIN || res == -EOPNOTSUPP) &&
2463 io_rw_should_reissue(req)) {
2464 req->flags |= REQ_F_REISSUE;
2465 return;
2466 }
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002467 req_set_fail_links(req);
Pavel Begunkov9532b992021-03-22 01:58:34 +00002468 }
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002469 if (req->flags & REQ_F_BUFFER_SELECTED)
2470 cflags = io_put_rw_kbuf(req);
2471 __io_req_complete(req, issue_flags, res, cflags);
Jens Axboeba816ad2019-09-28 11:36:45 -06002472}
2473
2474static void io_complete_rw(struct kiocb *kiocb, long res, long res2)
2475{
Jens Axboe9adbd452019-12-20 08:45:55 -07002476 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboeba816ad2019-09-28 11:36:45 -06002477
Pavel Begunkov889fca72021-02-10 00:03:09 +00002478 __io_complete_rw(req, res, res2, 0);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002479}
2480
Jens Axboedef596e2019-01-09 08:59:42 -07002481static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2)
2482{
Jens Axboe9adbd452019-12-20 08:45:55 -07002483 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboedef596e2019-01-09 08:59:42 -07002484
Jens Axboe491381ce2019-10-17 09:20:46 -06002485 if (kiocb->ki_flags & IOCB_WRITE)
2486 kiocb_end_write(req);
Pavel Begunkov9532b992021-03-22 01:58:34 +00002487 if (unlikely(res != req->result)) {
Jens Axboea1ff1e32021-04-12 06:40:02 -06002488 if (!(res == -EAGAIN && io_rw_should_reissue(req) &&
2489 io_resubmit_prep(req))) {
Pavel Begunkov9532b992021-03-22 01:58:34 +00002490 req_set_fail_links(req);
2491 req->flags |= REQ_F_DONT_REISSUE;
2492 }
Pavel Begunkov8c130822021-03-22 01:58:32 +00002493 }
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002494
2495 WRITE_ONCE(req->result, res);
Jens Axboeb9b0e0d2021-02-23 08:18:36 -07002496 /* order with io_iopoll_complete() checking ->result */
Pavel Begunkovcd664b02020-06-25 12:37:10 +03002497 smp_wmb();
2498 WRITE_ONCE(req->iopoll_completed, 1);
Jens Axboedef596e2019-01-09 08:59:42 -07002499}
2500
2501/*
2502 * After the iocb has been issued, it's safe to be found on the poll list.
2503 * Adding the kiocb to the list AFTER submission ensures that we don't
Pavel Begunkovf39c8a52021-04-13 02:58:46 +01002504 * find it from a io_do_iopoll() thread before the issuer is done
Jens Axboedef596e2019-01-09 08:59:42 -07002505 * accessing the kiocb cookie.
2506 */
Xiaoguang Wang2e9dbe92020-11-13 00:44:08 +08002507static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async)
Jens Axboedef596e2019-01-09 08:59:42 -07002508{
2509 struct io_ring_ctx *ctx = req->ctx;
2510
2511 /*
2512 * Track whether we have multiple files in our lists. This will impact
2513 * how we do polling eventually, not spinning if we're on potentially
2514 * different devices.
2515 */
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002516 if (list_empty(&ctx->iopoll_list)) {
Jens Axboedef596e2019-01-09 08:59:42 -07002517 ctx->poll_multi_file = false;
2518 } else if (!ctx->poll_multi_file) {
2519 struct io_kiocb *list_req;
2520
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002521 list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb,
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002522 inflight_entry);
Jens Axboe9adbd452019-12-20 08:45:55 -07002523 if (list_req->file != req->file)
Jens Axboedef596e2019-01-09 08:59:42 -07002524 ctx->poll_multi_file = true;
2525 }
2526
2527 /*
2528 * For fast devices, IO may have already completed. If it has, add
2529 * it to the front so we find it first.
2530 */
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002531 if (READ_ONCE(req->iopoll_completed))
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002532 list_add(&req->inflight_entry, &ctx->iopoll_list);
Jens Axboedef596e2019-01-09 08:59:42 -07002533 else
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002534 list_add_tail(&req->inflight_entry, &ctx->iopoll_list);
Xiaoguang Wangbdcd3ea2020-02-25 22:12:08 +08002535
Xiaoguang Wang2e9dbe92020-11-13 00:44:08 +08002536 /*
2537 * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread
2538 * task context or in io worker task context. If current task context is
2539 * sq thread, we don't need to check whether should wake up sq thread.
2540 */
2541 if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) &&
Jens Axboe534ca6d2020-09-02 13:52:19 -06002542 wq_has_sleeper(&ctx->sq_data->wait))
2543 wake_up(&ctx->sq_data->wait);
Jens Axboedef596e2019-01-09 08:59:42 -07002544}
2545
Pavel Begunkov9f13c352020-05-17 14:13:41 +03002546static inline void io_state_file_put(struct io_submit_state *state)
2547{
Pavel Begunkov02b23a92021-01-19 13:32:41 +00002548 if (state->file_refs) {
2549 fput_many(state->file, state->file_refs);
2550 state->file_refs = 0;
2551 }
Jens Axboe9a56a232019-01-09 09:06:50 -07002552}
2553
2554/*
2555 * Get as many references to a file as we have IOs left in this submission,
2556 * assuming most submissions are for one file, or at least that each file
2557 * has more than one submission.
2558 */
Pavel Begunkov8da11c12020-02-24 11:32:44 +03002559static struct file *__io_file_get(struct io_submit_state *state, int fd)
Jens Axboe9a56a232019-01-09 09:06:50 -07002560{
2561 if (!state)
2562 return fget(fd);
2563
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002564 if (state->file_refs) {
Jens Axboe9a56a232019-01-09 09:06:50 -07002565 if (state->fd == fd) {
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002566 state->file_refs--;
Jens Axboe9a56a232019-01-09 09:06:50 -07002567 return state->file;
2568 }
Pavel Begunkov02b23a92021-01-19 13:32:41 +00002569 io_state_file_put(state);
Jens Axboe9a56a232019-01-09 09:06:50 -07002570 }
2571 state->file = fget_many(fd, state->ios_left);
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002572 if (unlikely(!state->file))
Jens Axboe9a56a232019-01-09 09:06:50 -07002573 return NULL;
2574
2575 state->fd = fd;
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002576 state->file_refs = state->ios_left - 1;
Jens Axboe9a56a232019-01-09 09:06:50 -07002577 return state->file;
2578}
2579
Jens Axboe4503b762020-06-01 10:00:27 -06002580static bool io_bdev_nowait(struct block_device *bdev)
2581{
Jeffle Xu9ba0d0c2020-10-19 16:59:42 +08002582 return !bdev || blk_queue_nowait(bdev_get_queue(bdev));
Jens Axboe4503b762020-06-01 10:00:27 -06002583}
2584
Jens Axboe2b188cc2019-01-07 10:46:33 -07002585/*
2586 * If we tracked the file through the SCM inflight mechanism, we could support
2587 * any file. For now, just ensure that anything potentially problematic is done
2588 * inline.
2589 */
Jens Axboe7b29f922021-03-12 08:30:14 -07002590static bool __io_file_supports_async(struct file *file, int rw)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002591{
2592 umode_t mode = file_inode(file)->i_mode;
2593
Jens Axboe4503b762020-06-01 10:00:27 -06002594 if (S_ISBLK(mode)) {
Christoph Hellwig4e7b5672020-11-23 13:38:40 +01002595 if (IS_ENABLED(CONFIG_BLOCK) &&
2596 io_bdev_nowait(I_BDEV(file->f_mapping->host)))
Jens Axboe4503b762020-06-01 10:00:27 -06002597 return true;
2598 return false;
2599 }
2600 if (S_ISCHR(mode) || S_ISSOCK(mode))
Jens Axboe2b188cc2019-01-07 10:46:33 -07002601 return true;
Jens Axboe4503b762020-06-01 10:00:27 -06002602 if (S_ISREG(mode)) {
Christoph Hellwig4e7b5672020-11-23 13:38:40 +01002603 if (IS_ENABLED(CONFIG_BLOCK) &&
2604 io_bdev_nowait(file->f_inode->i_sb->s_bdev) &&
Jens Axboe4503b762020-06-01 10:00:27 -06002605 file->f_op != &io_uring_fops)
2606 return true;
2607 return false;
2608 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002609
Jens Axboec5b85622020-06-09 19:23:05 -06002610 /* any ->read/write should understand O_NONBLOCK */
2611 if (file->f_flags & O_NONBLOCK)
2612 return true;
2613
Jens Axboeaf197f52020-04-28 13:15:06 -06002614 if (!(file->f_mode & FMODE_NOWAIT))
2615 return false;
2616
2617 if (rw == READ)
2618 return file->f_op->read_iter != NULL;
2619
2620 return file->f_op->write_iter != NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002621}
2622
Jens Axboe7b29f922021-03-12 08:30:14 -07002623static bool io_file_supports_async(struct io_kiocb *req, int rw)
2624{
2625 if (rw == READ && (req->flags & REQ_F_ASYNC_READ))
2626 return true;
2627 else if (rw == WRITE && (req->flags & REQ_F_ASYNC_WRITE))
2628 return true;
2629
2630 return __io_file_supports_async(req->file, rw);
2631}
2632
Pavel Begunkova88fc402020-09-30 22:57:53 +03002633static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002634{
Jens Axboedef596e2019-01-09 08:59:42 -07002635 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe9adbd452019-12-20 08:45:55 -07002636 struct kiocb *kiocb = &req->rw.kiocb;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002637 struct file *file = req->file;
Jens Axboe09bb8392019-03-13 12:39:28 -06002638 unsigned ioprio;
2639 int ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002640
Jens Axboe7b29f922021-03-12 08:30:14 -07002641 if (!(req->flags & REQ_F_ISREG) && S_ISREG(file_inode(file)->i_mode))
Jens Axboe491381ce2019-10-17 09:20:46 -06002642 req->flags |= REQ_F_ISREG;
2643
Jens Axboe2b188cc2019-01-07 10:46:33 -07002644 kiocb->ki_pos = READ_ONCE(sqe->off);
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002645 if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) {
Jens Axboeba042912019-12-25 16:33:42 -07002646 req->flags |= REQ_F_CUR_POS;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002647 kiocb->ki_pos = file->f_pos;
Jens Axboeba042912019-12-25 16:33:42 -07002648 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002649 kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp));
Pavel Begunkov3e577dc2020-02-01 03:58:42 +03002650 kiocb->ki_flags = iocb_flags(kiocb->ki_filp);
2651 ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags));
2652 if (unlikely(ret))
2653 return ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002654
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002655 /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */
2656 if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK))
2657 req->flags |= REQ_F_NOWAIT;
2658
Jens Axboe2b188cc2019-01-07 10:46:33 -07002659 ioprio = READ_ONCE(sqe->ioprio);
2660 if (ioprio) {
2661 ret = ioprio_check_cap(ioprio);
2662 if (ret)
Jens Axboe09bb8392019-03-13 12:39:28 -06002663 return ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002664
2665 kiocb->ki_ioprio = ioprio;
2666 } else
2667 kiocb->ki_ioprio = get_current_ioprio();
2668
Jens Axboedef596e2019-01-09 08:59:42 -07002669 if (ctx->flags & IORING_SETUP_IOPOLL) {
Jens Axboedef596e2019-01-09 08:59:42 -07002670 if (!(kiocb->ki_flags & IOCB_DIRECT) ||
2671 !kiocb->ki_filp->f_op->iopoll)
Jens Axboe09bb8392019-03-13 12:39:28 -06002672 return -EOPNOTSUPP;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002673
Jens Axboedef596e2019-01-09 08:59:42 -07002674 kiocb->ki_flags |= IOCB_HIPRI;
2675 kiocb->ki_complete = io_complete_rw_iopoll;
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002676 req->iopoll_completed = 0;
Jens Axboedef596e2019-01-09 08:59:42 -07002677 } else {
Jens Axboe09bb8392019-03-13 12:39:28 -06002678 if (kiocb->ki_flags & IOCB_HIPRI)
2679 return -EINVAL;
Jens Axboedef596e2019-01-09 08:59:42 -07002680 kiocb->ki_complete = io_complete_rw;
2681 }
Jens Axboe9adbd452019-12-20 08:45:55 -07002682
Jens Axboe3529d8c2019-12-19 18:24:38 -07002683 req->rw.addr = READ_ONCE(sqe->addr);
2684 req->rw.len = READ_ONCE(sqe->len);
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002685 req->buf_index = READ_ONCE(sqe->buf_index);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002686 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002687}
2688
2689static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret)
2690{
2691 switch (ret) {
2692 case -EIOCBQUEUED:
2693 break;
2694 case -ERESTARTSYS:
2695 case -ERESTARTNOINTR:
2696 case -ERESTARTNOHAND:
2697 case -ERESTART_RESTARTBLOCK:
2698 /*
2699 * We can't just restart the syscall, since previously
2700 * submitted sqes may already be in progress. Just fail this
2701 * IO with EINTR.
2702 */
2703 ret = -EINTR;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05002704 fallthrough;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002705 default:
2706 kiocb->ki_complete(kiocb, ret, 0);
2707 }
2708}
2709
Jens Axboea1d7c392020-06-22 11:09:46 -06002710static void kiocb_done(struct kiocb *kiocb, ssize_t ret,
Pavel Begunkov889fca72021-02-10 00:03:09 +00002711 unsigned int issue_flags)
Jens Axboeba816ad2019-09-28 11:36:45 -06002712{
Jens Axboeba042912019-12-25 16:33:42 -07002713 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboee8c2bc12020-08-15 18:44:09 -07002714 struct io_async_rw *io = req->async_data;
Pavel Begunkov97284632021-04-08 19:28:03 +01002715 bool check_reissue = kiocb->ki_complete == io_complete_rw;
Jens Axboeba042912019-12-25 16:33:42 -07002716
Jens Axboe227c0c92020-08-13 11:51:40 -06002717 /* add previously done IO, if any */
Jens Axboee8c2bc12020-08-15 18:44:09 -07002718 if (io && io->bytes_done > 0) {
Jens Axboe227c0c92020-08-13 11:51:40 -06002719 if (ret < 0)
Jens Axboee8c2bc12020-08-15 18:44:09 -07002720 ret = io->bytes_done;
Jens Axboe227c0c92020-08-13 11:51:40 -06002721 else
Jens Axboee8c2bc12020-08-15 18:44:09 -07002722 ret += io->bytes_done;
Jens Axboe227c0c92020-08-13 11:51:40 -06002723 }
2724
Jens Axboeba042912019-12-25 16:33:42 -07002725 if (req->flags & REQ_F_CUR_POS)
2726 req->file->f_pos = kiocb->ki_pos;
Pavel Begunkovbcaec082020-02-24 11:30:18 +03002727 if (ret >= 0 && kiocb->ki_complete == io_complete_rw)
Pavel Begunkov889fca72021-02-10 00:03:09 +00002728 __io_complete_rw(req, ret, 0, issue_flags);
Jens Axboeba816ad2019-09-28 11:36:45 -06002729 else
2730 io_rw_done(kiocb, ret);
Pavel Begunkov97284632021-04-08 19:28:03 +01002731
2732 if (check_reissue && req->flags & REQ_F_REISSUE) {
2733 req->flags &= ~REQ_F_REISSUE;
Jens Axboea7be7c22021-04-15 11:31:14 -06002734 if (io_resubmit_prep(req)) {
Pavel Begunkov8c130822021-03-22 01:58:32 +00002735 req_ref_get(req);
2736 io_queue_async_work(req);
2737 } else {
Pavel Begunkov97284632021-04-08 19:28:03 +01002738 int cflags = 0;
2739
2740 req_set_fail_links(req);
2741 if (req->flags & REQ_F_BUFFER_SELECTED)
2742 cflags = io_put_rw_kbuf(req);
2743 __io_req_complete(req, issue_flags, ret, cflags);
2744 }
2745 }
Jens Axboeba816ad2019-09-28 11:36:45 -06002746}
2747
Pavel Begunkov847595d2021-02-04 13:52:06 +00002748static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter)
Jens Axboeedafcce2019-01-09 09:16:05 -07002749{
Jens Axboe9adbd452019-12-20 08:45:55 -07002750 struct io_ring_ctx *ctx = req->ctx;
2751 size_t len = req->rw.len;
Jens Axboeedafcce2019-01-09 09:16:05 -07002752 struct io_mapped_ubuf *imu;
Pavel Begunkov4be1c612020-09-06 00:45:48 +03002753 u16 index, buf_index = req->buf_index;
Pavel Begunkov75769e32021-04-01 15:43:54 +01002754 u64 buf_end, buf_addr = req->rw.addr;
Jens Axboeedafcce2019-01-09 09:16:05 -07002755 size_t offset;
Jens Axboeedafcce2019-01-09 09:16:05 -07002756
Jens Axboeedafcce2019-01-09 09:16:05 -07002757 if (unlikely(buf_index >= ctx->nr_user_bufs))
2758 return -EFAULT;
Jens Axboeedafcce2019-01-09 09:16:05 -07002759 index = array_index_nospec(buf_index, ctx->nr_user_bufs);
2760 imu = &ctx->user_bufs[index];
Jens Axboe9adbd452019-12-20 08:45:55 -07002761 buf_addr = req->rw.addr;
Jens Axboeedafcce2019-01-09 09:16:05 -07002762
Pavel Begunkov75769e32021-04-01 15:43:54 +01002763 if (unlikely(check_add_overflow(buf_addr, (u64)len, &buf_end)))
Jens Axboeedafcce2019-01-09 09:16:05 -07002764 return -EFAULT;
2765 /* not inside the mapped region */
Pavel Begunkov4751f532021-04-01 15:43:55 +01002766 if (unlikely(buf_addr < imu->ubuf || buf_end > imu->ubuf_end))
Jens Axboeedafcce2019-01-09 09:16:05 -07002767 return -EFAULT;
2768
2769 /*
2770 * May not be a start of buffer, set size appropriately
2771 * and advance us to the beginning.
2772 */
2773 offset = buf_addr - imu->ubuf;
2774 iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len);
Jens Axboebd11b3a2019-07-20 08:37:31 -06002775
2776 if (offset) {
2777 /*
2778 * Don't use iov_iter_advance() here, as it's really slow for
2779 * using the latter parts of a big fixed buffer - it iterates
2780 * over each segment manually. We can cheat a bit here, because
2781 * we know that:
2782 *
2783 * 1) it's a BVEC iter, we set it up
2784 * 2) all bvecs are PAGE_SIZE in size, except potentially the
2785 * first and last bvec
2786 *
2787 * So just find our index, and adjust the iterator afterwards.
2788 * If the offset is within the first bvec (or the whole first
2789 * bvec, just use iov_iter_advance(). This makes it easier
2790 * since we can just skip the first segment, which may not
2791 * be PAGE_SIZE aligned.
2792 */
2793 const struct bio_vec *bvec = imu->bvec;
2794
2795 if (offset <= bvec->bv_len) {
2796 iov_iter_advance(iter, offset);
2797 } else {
2798 unsigned long seg_skip;
2799
2800 /* skip first vec */
2801 offset -= bvec->bv_len;
2802 seg_skip = 1 + (offset >> PAGE_SHIFT);
2803
2804 iter->bvec = bvec + seg_skip;
2805 iter->nr_segs -= seg_skip;
Aleix Roca Nonell99c79f62019-08-15 14:03:22 +02002806 iter->count -= bvec->bv_len + offset;
Jens Axboebd11b3a2019-07-20 08:37:31 -06002807 iter->iov_offset = offset & ~PAGE_MASK;
Jens Axboebd11b3a2019-07-20 08:37:31 -06002808 }
2809 }
2810
Pavel Begunkov847595d2021-02-04 13:52:06 +00002811 return 0;
Jens Axboeedafcce2019-01-09 09:16:05 -07002812}
2813
Jens Axboebcda7ba2020-02-23 16:42:51 -07002814static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock)
2815{
2816 if (needs_lock)
2817 mutex_unlock(&ctx->uring_lock);
2818}
2819
2820static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock)
2821{
2822 /*
2823 * "Normal" inline submissions always hold the uring_lock, since we
2824 * grab it from the system call. Same is true for the SQPOLL offload.
2825 * The only exception is when we've detached the request and issue it
2826 * from an async worker thread, grab the lock for that case.
2827 */
2828 if (needs_lock)
2829 mutex_lock(&ctx->uring_lock);
2830}
2831
2832static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len,
2833 int bgid, struct io_buffer *kbuf,
2834 bool needs_lock)
2835{
2836 struct io_buffer *head;
2837
2838 if (req->flags & REQ_F_BUFFER_SELECTED)
2839 return kbuf;
2840
2841 io_ring_submit_lock(req->ctx, needs_lock);
2842
2843 lockdep_assert_held(&req->ctx->uring_lock);
2844
Jens Axboe9e15c3a2021-03-13 12:29:43 -07002845 head = xa_load(&req->ctx->io_buffers, bgid);
Jens Axboebcda7ba2020-02-23 16:42:51 -07002846 if (head) {
2847 if (!list_empty(&head->list)) {
2848 kbuf = list_last_entry(&head->list, struct io_buffer,
2849 list);
2850 list_del(&kbuf->list);
2851 } else {
2852 kbuf = head;
Jens Axboe9e15c3a2021-03-13 12:29:43 -07002853 xa_erase(&req->ctx->io_buffers, bgid);
Jens Axboebcda7ba2020-02-23 16:42:51 -07002854 }
2855 if (*len > kbuf->len)
2856 *len = kbuf->len;
2857 } else {
2858 kbuf = ERR_PTR(-ENOBUFS);
2859 }
2860
2861 io_ring_submit_unlock(req->ctx, needs_lock);
2862
2863 return kbuf;
2864}
2865
Jens Axboe4d954c22020-02-27 07:31:19 -07002866static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len,
2867 bool needs_lock)
2868{
2869 struct io_buffer *kbuf;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002870 u16 bgid;
Jens Axboe4d954c22020-02-27 07:31:19 -07002871
2872 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002873 bgid = req->buf_index;
Jens Axboe4d954c22020-02-27 07:31:19 -07002874 kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock);
2875 if (IS_ERR(kbuf))
2876 return kbuf;
2877 req->rw.addr = (u64) (unsigned long) kbuf;
2878 req->flags |= REQ_F_BUFFER_SELECTED;
2879 return u64_to_user_ptr(kbuf->addr);
2880}
2881
2882#ifdef CONFIG_COMPAT
2883static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov,
2884 bool needs_lock)
2885{
2886 struct compat_iovec __user *uiov;
2887 compat_ssize_t clen;
2888 void __user *buf;
2889 ssize_t len;
2890
2891 uiov = u64_to_user_ptr(req->rw.addr);
2892 if (!access_ok(uiov, sizeof(*uiov)))
2893 return -EFAULT;
2894 if (__get_user(clen, &uiov->iov_len))
2895 return -EFAULT;
2896 if (clen < 0)
2897 return -EINVAL;
2898
2899 len = clen;
2900 buf = io_rw_buffer_select(req, &len, needs_lock);
2901 if (IS_ERR(buf))
2902 return PTR_ERR(buf);
2903 iov[0].iov_base = buf;
2904 iov[0].iov_len = (compat_size_t) len;
2905 return 0;
2906}
2907#endif
2908
2909static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov,
2910 bool needs_lock)
2911{
2912 struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr);
2913 void __user *buf;
2914 ssize_t len;
2915
2916 if (copy_from_user(iov, uiov, sizeof(*uiov)))
2917 return -EFAULT;
2918
2919 len = iov[0].iov_len;
2920 if (len < 0)
2921 return -EINVAL;
2922 buf = io_rw_buffer_select(req, &len, needs_lock);
2923 if (IS_ERR(buf))
2924 return PTR_ERR(buf);
2925 iov[0].iov_base = buf;
2926 iov[0].iov_len = len;
2927 return 0;
2928}
2929
2930static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov,
2931 bool needs_lock)
2932{
Jens Axboedddb3e22020-06-04 11:27:01 -06002933 if (req->flags & REQ_F_BUFFER_SELECTED) {
2934 struct io_buffer *kbuf;
2935
2936 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
2937 iov[0].iov_base = u64_to_user_ptr(kbuf->addr);
2938 iov[0].iov_len = kbuf->len;
Jens Axboe4d954c22020-02-27 07:31:19 -07002939 return 0;
Jens Axboedddb3e22020-06-04 11:27:01 -06002940 }
Pavel Begunkovdd201662020-12-19 03:15:43 +00002941 if (req->rw.len != 1)
Jens Axboe4d954c22020-02-27 07:31:19 -07002942 return -EINVAL;
2943
2944#ifdef CONFIG_COMPAT
2945 if (req->ctx->compat)
2946 return io_compat_import(req, iov, needs_lock);
2947#endif
2948
2949 return __io_iov_buffer_select(req, iov, needs_lock);
2950}
2951
Pavel Begunkov847595d2021-02-04 13:52:06 +00002952static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec,
2953 struct iov_iter *iter, bool needs_lock)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002954{
Jens Axboe9adbd452019-12-20 08:45:55 -07002955 void __user *buf = u64_to_user_ptr(req->rw.addr);
2956 size_t sqe_len = req->rw.len;
Pavel Begunkov847595d2021-02-04 13:52:06 +00002957 u8 opcode = req->opcode;
Jens Axboe4d954c22020-02-27 07:31:19 -07002958 ssize_t ret;
Jens Axboeedafcce2019-01-09 09:16:05 -07002959
Pavel Begunkov7d009162019-11-25 23:14:40 +03002960 if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) {
Jens Axboeedafcce2019-01-09 09:16:05 -07002961 *iovec = NULL;
Jens Axboe9adbd452019-12-20 08:45:55 -07002962 return io_import_fixed(req, rw, iter);
Jens Axboeedafcce2019-01-09 09:16:05 -07002963 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002964
Jens Axboebcda7ba2020-02-23 16:42:51 -07002965 /* buffer index only valid with fixed read/write, or buffer select */
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002966 if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT))
Jens Axboe9adbd452019-12-20 08:45:55 -07002967 return -EINVAL;
2968
Jens Axboe3a6820f2019-12-22 15:19:35 -07002969 if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) {
Jens Axboebcda7ba2020-02-23 16:42:51 -07002970 if (req->flags & REQ_F_BUFFER_SELECT) {
Jens Axboe4d954c22020-02-27 07:31:19 -07002971 buf = io_rw_buffer_select(req, &sqe_len, needs_lock);
Pavel Begunkov867a23e2020-08-20 11:34:39 +03002972 if (IS_ERR(buf))
Jens Axboe4d954c22020-02-27 07:31:19 -07002973 return PTR_ERR(buf);
Jens Axboe3f9d6442020-03-11 12:27:04 -06002974 req->rw.len = sqe_len;
Jens Axboebcda7ba2020-02-23 16:42:51 -07002975 }
2976
Jens Axboe3a6820f2019-12-22 15:19:35 -07002977 ret = import_single_range(rw, buf, sqe_len, *iovec, iter);
2978 *iovec = NULL;
David Laight10fc72e2020-11-07 13:16:25 +00002979 return ret;
Jens Axboe3a6820f2019-12-22 15:19:35 -07002980 }
2981
Jens Axboe4d954c22020-02-27 07:31:19 -07002982 if (req->flags & REQ_F_BUFFER_SELECT) {
2983 ret = io_iov_buffer_select(req, *iovec, needs_lock);
Pavel Begunkov847595d2021-02-04 13:52:06 +00002984 if (!ret)
2985 iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len);
Jens Axboe4d954c22020-02-27 07:31:19 -07002986 *iovec = NULL;
2987 return ret;
2988 }
2989
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02002990 return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter,
2991 req->ctx->compat);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002992}
2993
Jens Axboe0fef9482020-08-26 10:36:20 -06002994static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb)
2995{
Pavel Begunkov5b09e372020-09-30 22:57:15 +03002996 return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos;
Jens Axboe0fef9482020-08-26 10:36:20 -06002997}
2998
Jens Axboe32960612019-09-23 11:05:34 -06002999/*
3000 * For files that don't have ->read_iter() and ->write_iter(), handle them
3001 * by looping over ->read() or ->write() manually.
3002 */
Jens Axboe4017eb92020-10-22 14:14:12 -06003003static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter)
Jens Axboe32960612019-09-23 11:05:34 -06003004{
Jens Axboe4017eb92020-10-22 14:14:12 -06003005 struct kiocb *kiocb = &req->rw.kiocb;
3006 struct file *file = req->file;
Jens Axboe32960612019-09-23 11:05:34 -06003007 ssize_t ret = 0;
3008
3009 /*
3010 * Don't support polled IO through this interface, and we can't
3011 * support non-blocking either. For the latter, this just causes
3012 * the kiocb to be handled from an async context.
3013 */
3014 if (kiocb->ki_flags & IOCB_HIPRI)
3015 return -EOPNOTSUPP;
3016 if (kiocb->ki_flags & IOCB_NOWAIT)
3017 return -EAGAIN;
3018
3019 while (iov_iter_count(iter)) {
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003020 struct iovec iovec;
Jens Axboe32960612019-09-23 11:05:34 -06003021 ssize_t nr;
3022
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003023 if (!iov_iter_is_bvec(iter)) {
3024 iovec = iov_iter_iovec(iter);
3025 } else {
Jens Axboe4017eb92020-10-22 14:14:12 -06003026 iovec.iov_base = u64_to_user_ptr(req->rw.addr);
3027 iovec.iov_len = req->rw.len;
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003028 }
3029
Jens Axboe32960612019-09-23 11:05:34 -06003030 if (rw == READ) {
3031 nr = file->f_op->read(file, iovec.iov_base,
Jens Axboe0fef9482020-08-26 10:36:20 -06003032 iovec.iov_len, io_kiocb_ppos(kiocb));
Jens Axboe32960612019-09-23 11:05:34 -06003033 } else {
3034 nr = file->f_op->write(file, iovec.iov_base,
Jens Axboe0fef9482020-08-26 10:36:20 -06003035 iovec.iov_len, io_kiocb_ppos(kiocb));
Jens Axboe32960612019-09-23 11:05:34 -06003036 }
3037
3038 if (nr < 0) {
3039 if (!ret)
3040 ret = nr;
3041 break;
3042 }
3043 ret += nr;
3044 if (nr != iovec.iov_len)
3045 break;
Jens Axboe4017eb92020-10-22 14:14:12 -06003046 req->rw.len -= nr;
3047 req->rw.addr += nr;
Jens Axboe32960612019-09-23 11:05:34 -06003048 iov_iter_advance(iter, nr);
3049 }
3050
3051 return ret;
3052}
3053
Jens Axboeff6165b2020-08-13 09:47:43 -06003054static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec,
3055 const struct iovec *fast_iov, struct iov_iter *iter)
Jens Axboef67676d2019-12-02 11:03:47 -07003056{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003057 struct io_async_rw *rw = req->async_data;
Pavel Begunkovb64e3442020-07-13 22:59:18 +03003058
Jens Axboeff6165b2020-08-13 09:47:43 -06003059 memcpy(&rw->iter, iter, sizeof(*iter));
Pavel Begunkovafb87652020-09-06 00:45:46 +03003060 rw->free_iovec = iovec;
Jens Axboe227c0c92020-08-13 11:51:40 -06003061 rw->bytes_done = 0;
Jens Axboeff6165b2020-08-13 09:47:43 -06003062 /* can only be fixed buffers, no need to do anything */
Pavel Begunkov9c3a2052020-11-23 23:20:27 +00003063 if (iov_iter_is_bvec(iter))
Jens Axboeff6165b2020-08-13 09:47:43 -06003064 return;
Pavel Begunkovb64e3442020-07-13 22:59:18 +03003065 if (!iovec) {
Jens Axboeff6165b2020-08-13 09:47:43 -06003066 unsigned iov_off = 0;
3067
3068 rw->iter.iov = rw->fast_iov;
3069 if (iter->iov != fast_iov) {
3070 iov_off = iter->iov - fast_iov;
3071 rw->iter.iov += iov_off;
3072 }
3073 if (rw->fast_iov != fast_iov)
3074 memcpy(rw->fast_iov + iov_off, fast_iov + iov_off,
Xiaoguang Wang45097da2020-04-08 22:29:58 +08003075 sizeof(struct iovec) * iter->nr_segs);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03003076 } else {
3077 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboef67676d2019-12-02 11:03:47 -07003078 }
3079}
3080
Pavel Begunkov6cb78682021-02-28 22:35:17 +00003081static inline int io_alloc_async_data(struct io_kiocb *req)
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003082{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003083 WARN_ON_ONCE(!io_op_defs[req->opcode].async_size);
3084 req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL);
3085 return req->async_data == NULL;
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003086}
3087
Jens Axboeff6165b2020-08-13 09:47:43 -06003088static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec,
3089 const struct iovec *fast_iov,
Jens Axboe227c0c92020-08-13 11:51:40 -06003090 struct iov_iter *iter, bool force)
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003091{
Pavel Begunkov26f05052021-02-28 22:35:18 +00003092 if (!force && !io_op_defs[req->opcode].needs_async_setup)
Jens Axboe74566df2020-01-13 19:23:24 -07003093 return 0;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003094 if (!req->async_data) {
Pavel Begunkov6cb78682021-02-28 22:35:17 +00003095 if (io_alloc_async_data(req)) {
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003096 kfree(iovec);
Jens Axboe5d204bc2020-01-31 12:06:52 -07003097 return -ENOMEM;
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003098 }
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003099
Jens Axboeff6165b2020-08-13 09:47:43 -06003100 io_req_map_rw(req, iovec, fast_iov, iter);
Jens Axboe5d204bc2020-01-31 12:06:52 -07003101 }
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003102 return 0;
Jens Axboef67676d2019-12-02 11:03:47 -07003103}
3104
Pavel Begunkov73debe62020-09-30 22:57:54 +03003105static inline int io_rw_prep_async(struct io_kiocb *req, int rw)
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003106{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003107 struct io_async_rw *iorw = req->async_data;
Pavel Begunkovf4bff102020-09-06 00:45:45 +03003108 struct iovec *iov = iorw->fast_iov;
Pavel Begunkov847595d2021-02-04 13:52:06 +00003109 int ret;
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003110
Pavel Begunkov2846c482020-11-07 13:16:27 +00003111 ret = io_import_iovec(rw, req, &iov, &iorw->iter, false);
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003112 if (unlikely(ret < 0))
3113 return ret;
3114
Pavel Begunkovab0b1962020-09-06 00:45:47 +03003115 iorw->bytes_done = 0;
3116 iorw->free_iovec = iov;
3117 if (iov)
3118 req->flags |= REQ_F_NEED_CLEANUP;
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003119 return 0;
3120}
3121
Pavel Begunkov73debe62020-09-30 22:57:54 +03003122static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07003123{
Jens Axboe3529d8c2019-12-19 18:24:38 -07003124 if (unlikely(!(req->file->f_mode & FMODE_READ)))
3125 return -EBADF;
Pavel Begunkov93642ef2021-02-18 18:29:44 +00003126 return io_prep_rw(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07003127}
3128
Jens Axboec1dd91d2020-08-03 16:43:59 -06003129/*
3130 * This is our waitqueue callback handler, registered through lock_page_async()
3131 * when we initially tried to do the IO with the iocb armed our waitqueue.
3132 * This gets called when the page is unlocked, and we generally expect that to
3133 * happen when the page IO is completed and the page is now uptodate. This will
3134 * queue a task_work based retry of the operation, attempting to copy the data
3135 * again. If the latter fails because the page was NOT uptodate, then we will
3136 * do a thread based blocking retry of the operation. That's the unexpected
3137 * slow path.
3138 */
Jens Axboebcf5a062020-05-22 09:24:42 -06003139static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode,
3140 int sync, void *arg)
3141{
3142 struct wait_page_queue *wpq;
3143 struct io_kiocb *req = wait->private;
Jens Axboebcf5a062020-05-22 09:24:42 -06003144 struct wait_page_key *key = arg;
Jens Axboebcf5a062020-05-22 09:24:42 -06003145
3146 wpq = container_of(wait, struct wait_page_queue, wait);
3147
Linus Torvaldscdc8fcb2020-08-03 13:01:22 -07003148 if (!wake_page_match(wpq, key))
3149 return 0;
3150
Hao Xuc8d317a2020-09-29 20:00:45 +08003151 req->rw.kiocb.ki_flags &= ~IOCB_WAITQ;
Jens Axboebcf5a062020-05-22 09:24:42 -06003152 list_del_init(&wait->entry);
3153
Jens Axboebcf5a062020-05-22 09:24:42 -06003154 /* submit ref gets dropped, acquire a new one */
Jens Axboede9b4cc2021-02-24 13:28:27 -07003155 req_ref_get(req);
Pavel Begunkov921b9052021-02-12 03:23:53 +00003156 io_req_task_queue(req);
Jens Axboebcf5a062020-05-22 09:24:42 -06003157 return 1;
3158}
3159
Jens Axboec1dd91d2020-08-03 16:43:59 -06003160/*
3161 * This controls whether a given IO request should be armed for async page
3162 * based retry. If we return false here, the request is handed to the async
3163 * worker threads for retry. If we're doing buffered reads on a regular file,
3164 * we prepare a private wait_page_queue entry and retry the operation. This
3165 * will either succeed because the page is now uptodate and unlocked, or it
3166 * will register a callback when the page is unlocked at IO completion. Through
3167 * that callback, io_uring uses task_work to setup a retry of the operation.
3168 * That retry will attempt the buffered read again. The retry will generally
3169 * succeed, or in rare cases where it fails, we then fall back to using the
3170 * async worker threads for a blocking retry.
3171 */
Jens Axboe227c0c92020-08-13 11:51:40 -06003172static bool io_rw_should_retry(struct io_kiocb *req)
Jens Axboebcf5a062020-05-22 09:24:42 -06003173{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003174 struct io_async_rw *rw = req->async_data;
3175 struct wait_page_queue *wait = &rw->wpq;
Jens Axboebcf5a062020-05-22 09:24:42 -06003176 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboebcf5a062020-05-22 09:24:42 -06003177
3178 /* never retry for NOWAIT, we just complete with -EAGAIN */
3179 if (req->flags & REQ_F_NOWAIT)
3180 return false;
3181
Jens Axboe227c0c92020-08-13 11:51:40 -06003182 /* Only for buffered IO */
Jens Axboe3b2a4432020-08-16 10:58:43 -07003183 if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI))
Jens Axboebcf5a062020-05-22 09:24:42 -06003184 return false;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003185
Jens Axboebcf5a062020-05-22 09:24:42 -06003186 /*
3187 * just use poll if we can, and don't attempt if the fs doesn't
3188 * support callback based unlocks
3189 */
3190 if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC))
3191 return false;
3192
Jens Axboe3b2a4432020-08-16 10:58:43 -07003193 wait->wait.func = io_async_buf_func;
3194 wait->wait.private = req;
3195 wait->wait.flags = 0;
3196 INIT_LIST_HEAD(&wait->wait.entry);
3197 kiocb->ki_flags |= IOCB_WAITQ;
Hao Xuc8d317a2020-09-29 20:00:45 +08003198 kiocb->ki_flags &= ~IOCB_NOWAIT;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003199 kiocb->ki_waitq = wait;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003200 return true;
Jens Axboebcf5a062020-05-22 09:24:42 -06003201}
3202
3203static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter)
3204{
3205 if (req->file->f_op->read_iter)
3206 return call_read_iter(req->file, &req->rw.kiocb, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003207 else if (req->file->f_op->read)
Jens Axboe4017eb92020-10-22 14:14:12 -06003208 return loop_rw_iter(READ, req, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003209 else
3210 return -EINVAL;
Jens Axboebcf5a062020-05-22 09:24:42 -06003211}
3212
Pavel Begunkov889fca72021-02-10 00:03:09 +00003213static int io_read(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003214{
3215 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
Jens Axboe9adbd452019-12-20 08:45:55 -07003216 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboeff6165b2020-08-13 09:47:43 -06003217 struct iov_iter __iter, *iter = &__iter;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003218 struct io_async_rw *rw = req->async_data;
Jens Axboe227c0c92020-08-13 11:51:40 -06003219 ssize_t io_size, ret, ret2;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003220 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003221
Pavel Begunkov2846c482020-11-07 13:16:27 +00003222 if (rw) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07003223 iter = &rw->iter;
Pavel Begunkov2846c482020-11-07 13:16:27 +00003224 iovec = NULL;
3225 } else {
3226 ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock);
3227 if (ret < 0)
3228 return ret;
3229 }
Pavel Begunkov632546c2020-11-07 13:16:26 +00003230 io_size = iov_iter_count(iter);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003231 req->result = io_size;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003232
Jens Axboefd6c2e42019-12-18 12:19:41 -07003233 /* Ensure we clear previously set non-block flag */
3234 if (!force_nonblock)
Jens Axboe29de5f62020-02-20 09:56:08 -07003235 kiocb->ki_flags &= ~IOCB_NOWAIT;
Pavel Begunkova88fc402020-09-30 22:57:53 +03003236 else
3237 kiocb->ki_flags |= IOCB_NOWAIT;
3238
Pavel Begunkov24c74672020-06-21 13:09:51 +03003239 /* If the file doesn't support async, just async punt */
Jens Axboe7b29f922021-03-12 08:30:14 -07003240 if (force_nonblock && !io_file_supports_async(req, READ)) {
Pavel Begunkov6713e7a2021-02-04 13:51:59 +00003241 ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003242 return ret ?: -EAGAIN;
Pavel Begunkov6713e7a2021-02-04 13:51:59 +00003243 }
Jens Axboe9e645e112019-05-10 16:07:28 -06003244
Pavel Begunkov632546c2020-11-07 13:16:26 +00003245 ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size);
Pavel Begunkov5ea5dd42021-02-04 13:52:03 +00003246 if (unlikely(ret)) {
3247 kfree(iovec);
3248 return ret;
3249 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07003250
Jens Axboe227c0c92020-08-13 11:51:40 -06003251 ret = io_iter_do_read(req, iter);
Jens Axboe32960612019-09-23 11:05:34 -06003252
Jens Axboe230d50d2021-04-01 20:41:15 -06003253 if (ret == -EAGAIN || (req->flags & REQ_F_REISSUE)) {
Pavel Begunkov6ad7f2332021-04-08 01:54:39 +01003254 req->flags &= ~REQ_F_REISSUE;
Jens Axboeeefdf302020-08-27 16:40:19 -06003255 /* IOPOLL retry should happen for io-wq threads */
3256 if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboef91daf52020-08-15 15:58:42 -07003257 goto done;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003258 /* no retry on NONBLOCK nor RWF_NOWAIT */
3259 if (req->flags & REQ_F_NOWAIT)
Jens Axboe355afae2020-09-02 09:30:31 -06003260 goto done;
Jens Axboe84216312020-08-24 11:45:26 -06003261 /* some cases will consume bytes even on error returns */
Pavel Begunkov632546c2020-11-07 13:16:26 +00003262 iov_iter_revert(iter, io_size - iov_iter_count(iter));
Jens Axboef38c7e32020-09-25 15:23:43 -06003263 ret = 0;
Jens Axboe230d50d2021-04-01 20:41:15 -06003264 } else if (ret == -EIOCBQUEUED) {
3265 goto out_free;
Pavel Begunkov7335e3b2021-02-04 13:52:02 +00003266 } else if (ret <= 0 || ret == io_size || !force_nonblock ||
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003267 (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) {
Pavel Begunkov7335e3b2021-02-04 13:52:02 +00003268 /* read all, failed, already did sync or don't want to retry */
Jens Axboe00d23d52020-08-25 12:59:22 -06003269 goto done;
Jens Axboe227c0c92020-08-13 11:51:40 -06003270 }
3271
Jens Axboe227c0c92020-08-13 11:51:40 -06003272 ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003273 if (ret2)
3274 return ret2;
3275
Pavel Begunkovfe1cdd52021-02-17 21:02:36 +00003276 iovec = NULL;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003277 rw = req->async_data;
Jens Axboe227c0c92020-08-13 11:51:40 -06003278 /* now use our persistent iterator, if we aren't already */
Jens Axboee8c2bc12020-08-15 18:44:09 -07003279 iter = &rw->iter;
Jens Axboe227c0c92020-08-13 11:51:40 -06003280
Pavel Begunkovb23df912021-02-04 13:52:04 +00003281 do {
3282 io_size -= ret;
3283 rw->bytes_done += ret;
3284 /* if we can retry, do so with the callbacks armed */
3285 if (!io_rw_should_retry(req)) {
3286 kiocb->ki_flags &= ~IOCB_WAITQ;
3287 return -EAGAIN;
3288 }
3289
3290 /*
3291 * Now retry read with the IOCB_WAITQ parts set in the iocb. If
3292 * we get -EIOCBQUEUED, then we'll get a notification when the
3293 * desired page gets unlocked. We can also get a partial read
3294 * here, and if we do, then just retry at the new offset.
3295 */
3296 ret = io_iter_do_read(req, iter);
3297 if (ret == -EIOCBQUEUED)
3298 return 0;
Jens Axboe227c0c92020-08-13 11:51:40 -06003299 /* we got some bytes, but not all. retry. */
Jens Axboeb5b0ecb2021-03-04 21:02:58 -07003300 kiocb->ki_flags &= ~IOCB_WAITQ;
Pavel Begunkovb23df912021-02-04 13:52:04 +00003301 } while (ret > 0 && ret < io_size);
Jens Axboe227c0c92020-08-13 11:51:40 -06003302done:
Pavel Begunkov889fca72021-02-10 00:03:09 +00003303 kiocb_done(kiocb, ret, issue_flags);
Pavel Begunkovfe1cdd52021-02-17 21:02:36 +00003304out_free:
3305 /* it's faster to check here then delegate to kfree */
3306 if (iovec)
3307 kfree(iovec);
Pavel Begunkov5ea5dd42021-02-04 13:52:03 +00003308 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003309}
3310
Pavel Begunkov73debe62020-09-30 22:57:54 +03003311static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07003312{
Jens Axboe3529d8c2019-12-19 18:24:38 -07003313 if (unlikely(!(req->file->f_mode & FMODE_WRITE)))
3314 return -EBADF;
Pavel Begunkov93642ef2021-02-18 18:29:44 +00003315 return io_prep_rw(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07003316}
3317
Pavel Begunkov889fca72021-02-10 00:03:09 +00003318static int io_write(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003319{
3320 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
Jens Axboe9adbd452019-12-20 08:45:55 -07003321 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboeff6165b2020-08-13 09:47:43 -06003322 struct iov_iter __iter, *iter = &__iter;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003323 struct io_async_rw *rw = req->async_data;
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003324 ssize_t ret, ret2, io_size;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003325 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003326
Pavel Begunkov2846c482020-11-07 13:16:27 +00003327 if (rw) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07003328 iter = &rw->iter;
Pavel Begunkov2846c482020-11-07 13:16:27 +00003329 iovec = NULL;
3330 } else {
3331 ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock);
3332 if (ret < 0)
3333 return ret;
3334 }
Pavel Begunkov632546c2020-11-07 13:16:26 +00003335 io_size = iov_iter_count(iter);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003336 req->result = io_size;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003337
Jens Axboefd6c2e42019-12-18 12:19:41 -07003338 /* Ensure we clear previously set non-block flag */
3339 if (!force_nonblock)
Pavel Begunkova88fc402020-09-30 22:57:53 +03003340 kiocb->ki_flags &= ~IOCB_NOWAIT;
3341 else
3342 kiocb->ki_flags |= IOCB_NOWAIT;
Jens Axboefd6c2e42019-12-18 12:19:41 -07003343
Pavel Begunkov24c74672020-06-21 13:09:51 +03003344 /* If the file doesn't support async, just async punt */
Jens Axboe7b29f922021-03-12 08:30:14 -07003345 if (force_nonblock && !io_file_supports_async(req, WRITE))
Jens Axboef67676d2019-12-02 11:03:47 -07003346 goto copy_iov;
Jens Axboef67676d2019-12-02 11:03:47 -07003347
Jens Axboe10d59342019-12-09 20:16:22 -07003348 /* file path doesn't support NOWAIT for non-direct_IO */
3349 if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) &&
3350 (req->flags & REQ_F_ISREG))
Jens Axboef67676d2019-12-02 11:03:47 -07003351 goto copy_iov;
Jens Axboe9e645e112019-05-10 16:07:28 -06003352
Pavel Begunkov632546c2020-11-07 13:16:26 +00003353 ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003354 if (unlikely(ret))
3355 goto out_free;
Roman Penyaev9bf79332019-03-25 20:09:24 +01003356
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003357 /*
3358 * Open-code file_start_write here to grab freeze protection,
3359 * which will be released by another thread in
3360 * io_complete_rw(). Fool lockdep by telling it the lock got
3361 * released so that it doesn't complain about the held lock when
3362 * we return to userspace.
3363 */
3364 if (req->flags & REQ_F_ISREG) {
Darrick J. Wong8a3c84b2020-11-10 16:50:21 -08003365 sb_start_write(file_inode(req->file)->i_sb);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003366 __sb_writers_release(file_inode(req->file)->i_sb,
3367 SB_FREEZE_WRITE);
3368 }
3369 kiocb->ki_flags |= IOCB_WRITE;
Roman Penyaev9bf79332019-03-25 20:09:24 +01003370
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003371 if (req->file->f_op->write_iter)
Jens Axboeff6165b2020-08-13 09:47:43 -06003372 ret2 = call_write_iter(req->file, kiocb, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003373 else if (req->file->f_op->write)
Jens Axboe4017eb92020-10-22 14:14:12 -06003374 ret2 = loop_rw_iter(WRITE, req, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003375 else
3376 ret2 = -EINVAL;
Jens Axboe4ed734b2020-03-20 11:23:41 -06003377
Pavel Begunkov6ad7f2332021-04-08 01:54:39 +01003378 if (req->flags & REQ_F_REISSUE) {
3379 req->flags &= ~REQ_F_REISSUE;
Jens Axboe230d50d2021-04-01 20:41:15 -06003380 ret2 = -EAGAIN;
Pavel Begunkov6ad7f2332021-04-08 01:54:39 +01003381 }
Jens Axboe230d50d2021-04-01 20:41:15 -06003382
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003383 /*
3384 * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just
3385 * retry them without IOCB_NOWAIT.
3386 */
3387 if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT))
3388 ret2 = -EAGAIN;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003389 /* no retry on NONBLOCK nor RWF_NOWAIT */
3390 if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT))
Jens Axboe355afae2020-09-02 09:30:31 -06003391 goto done;
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003392 if (!force_nonblock || ret2 != -EAGAIN) {
Jens Axboeeefdf302020-08-27 16:40:19 -06003393 /* IOPOLL retry should happen for io-wq threads */
3394 if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN)
3395 goto copy_iov;
Jens Axboe355afae2020-09-02 09:30:31 -06003396done:
Pavel Begunkov889fca72021-02-10 00:03:09 +00003397 kiocb_done(kiocb, ret2, issue_flags);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003398 } else {
Jens Axboef67676d2019-12-02 11:03:47 -07003399copy_iov:
Jens Axboe84216312020-08-24 11:45:26 -06003400 /* some cases will consume bytes even on error returns */
Pavel Begunkov632546c2020-11-07 13:16:26 +00003401 iov_iter_revert(iter, io_size - iov_iter_count(iter));
Jens Axboe227c0c92020-08-13 11:51:40 -06003402 ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003403 return ret ?: -EAGAIN;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003404 }
Jens Axboe31b51512019-01-18 22:56:34 -07003405out_free:
Pavel Begunkovf261c162020-08-20 11:34:10 +03003406 /* it's reportedly faster than delegating the null check to kfree() */
Pavel Begunkov252917c2020-07-13 22:59:20 +03003407 if (iovec)
Xiaoguang Wang6f2cc162020-06-18 15:01:56 +08003408 kfree(iovec);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003409 return ret;
3410}
3411
Jens Axboe80a261f2020-09-28 14:23:58 -06003412static int io_renameat_prep(struct io_kiocb *req,
3413 const struct io_uring_sqe *sqe)
3414{
3415 struct io_rename *ren = &req->rename;
3416 const char __user *oldf, *newf;
3417
3418 if (unlikely(req->flags & REQ_F_FIXED_FILE))
3419 return -EBADF;
3420
3421 ren->old_dfd = READ_ONCE(sqe->fd);
3422 oldf = u64_to_user_ptr(READ_ONCE(sqe->addr));
3423 newf = u64_to_user_ptr(READ_ONCE(sqe->addr2));
3424 ren->new_dfd = READ_ONCE(sqe->len);
3425 ren->flags = READ_ONCE(sqe->rename_flags);
3426
3427 ren->oldpath = getname(oldf);
3428 if (IS_ERR(ren->oldpath))
3429 return PTR_ERR(ren->oldpath);
3430
3431 ren->newpath = getname(newf);
3432 if (IS_ERR(ren->newpath)) {
3433 putname(ren->oldpath);
3434 return PTR_ERR(ren->newpath);
3435 }
3436
3437 req->flags |= REQ_F_NEED_CLEANUP;
3438 return 0;
3439}
3440
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003441static int io_renameat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe80a261f2020-09-28 14:23:58 -06003442{
3443 struct io_rename *ren = &req->rename;
3444 int ret;
3445
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003446 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe80a261f2020-09-28 14:23:58 -06003447 return -EAGAIN;
3448
3449 ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd,
3450 ren->newpath, ren->flags);
3451
3452 req->flags &= ~REQ_F_NEED_CLEANUP;
3453 if (ret < 0)
3454 req_set_fail_links(req);
3455 io_req_complete(req, ret);
3456 return 0;
3457}
3458
Jens Axboe14a11432020-09-28 14:27:37 -06003459static int io_unlinkat_prep(struct io_kiocb *req,
3460 const struct io_uring_sqe *sqe)
3461{
3462 struct io_unlink *un = &req->unlink;
3463 const char __user *fname;
3464
3465 if (unlikely(req->flags & REQ_F_FIXED_FILE))
3466 return -EBADF;
3467
3468 un->dfd = READ_ONCE(sqe->fd);
3469
3470 un->flags = READ_ONCE(sqe->unlink_flags);
3471 if (un->flags & ~AT_REMOVEDIR)
3472 return -EINVAL;
3473
3474 fname = u64_to_user_ptr(READ_ONCE(sqe->addr));
3475 un->filename = getname(fname);
3476 if (IS_ERR(un->filename))
3477 return PTR_ERR(un->filename);
3478
3479 req->flags |= REQ_F_NEED_CLEANUP;
3480 return 0;
3481}
3482
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003483static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe14a11432020-09-28 14:27:37 -06003484{
3485 struct io_unlink *un = &req->unlink;
3486 int ret;
3487
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003488 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe14a11432020-09-28 14:27:37 -06003489 return -EAGAIN;
3490
3491 if (un->flags & AT_REMOVEDIR)
3492 ret = do_rmdir(un->dfd, un->filename);
3493 else
3494 ret = do_unlinkat(un->dfd, un->filename);
3495
3496 req->flags &= ~REQ_F_NEED_CLEANUP;
3497 if (ret < 0)
3498 req_set_fail_links(req);
3499 io_req_complete(req, ret);
3500 return 0;
3501}
3502
Jens Axboe36f4fa62020-09-05 11:14:22 -06003503static int io_shutdown_prep(struct io_kiocb *req,
3504 const struct io_uring_sqe *sqe)
3505{
3506#if defined(CONFIG_NET)
3507 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3508 return -EINVAL;
3509 if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags ||
3510 sqe->buf_index)
3511 return -EINVAL;
3512
3513 req->shutdown.how = READ_ONCE(sqe->len);
3514 return 0;
3515#else
3516 return -EOPNOTSUPP;
3517#endif
3518}
3519
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003520static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe36f4fa62020-09-05 11:14:22 -06003521{
3522#if defined(CONFIG_NET)
3523 struct socket *sock;
3524 int ret;
3525
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003526 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe36f4fa62020-09-05 11:14:22 -06003527 return -EAGAIN;
3528
Linus Torvalds48aba792020-12-16 12:44:05 -08003529 sock = sock_from_file(req->file);
Jens Axboe36f4fa62020-09-05 11:14:22 -06003530 if (unlikely(!sock))
Linus Torvalds48aba792020-12-16 12:44:05 -08003531 return -ENOTSOCK;
Jens Axboe36f4fa62020-09-05 11:14:22 -06003532
3533 ret = __sys_shutdown_sock(sock, req->shutdown.how);
Jens Axboea1464682020-12-14 20:57:27 -07003534 if (ret < 0)
3535 req_set_fail_links(req);
Jens Axboe36f4fa62020-09-05 11:14:22 -06003536 io_req_complete(req, ret);
3537 return 0;
3538#else
3539 return -EOPNOTSUPP;
3540#endif
3541}
3542
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003543static int __io_splice_prep(struct io_kiocb *req,
3544 const struct io_uring_sqe *sqe)
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003545{
3546 struct io_splice* sp = &req->splice;
3547 unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003548
Pavel Begunkov3232dd02020-06-03 18:03:22 +03003549 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3550 return -EINVAL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003551
3552 sp->file_in = NULL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003553 sp->len = READ_ONCE(sqe->len);
3554 sp->flags = READ_ONCE(sqe->splice_flags);
3555
3556 if (unlikely(sp->flags & ~valid_flags))
3557 return -EINVAL;
3558
Pavel Begunkov8371adf2020-10-10 18:34:08 +01003559 sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in),
3560 (sp->flags & SPLICE_F_FD_IN_FIXED));
3561 if (!sp->file_in)
3562 return -EBADF;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003563 req->flags |= REQ_F_NEED_CLEANUP;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003564 return 0;
3565}
3566
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003567static int io_tee_prep(struct io_kiocb *req,
3568 const struct io_uring_sqe *sqe)
3569{
3570 if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off))
3571 return -EINVAL;
3572 return __io_splice_prep(req, sqe);
3573}
3574
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003575static int io_tee(struct io_kiocb *req, unsigned int issue_flags)
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003576{
3577 struct io_splice *sp = &req->splice;
3578 struct file *in = sp->file_in;
3579 struct file *out = sp->file_out;
3580 unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED;
3581 long ret = 0;
3582
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003583 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003584 return -EAGAIN;
3585 if (sp->len)
3586 ret = do_tee(in, out, sp->len, flags);
3587
Pavel Begunkove1d767f2021-03-19 17:22:43 +00003588 if (!(sp->flags & SPLICE_F_FD_IN_FIXED))
3589 io_put_file(in);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003590 req->flags &= ~REQ_F_NEED_CLEANUP;
3591
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003592 if (ret != sp->len)
3593 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003594 io_req_complete(req, ret);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003595 return 0;
3596}
3597
3598static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3599{
3600 struct io_splice* sp = &req->splice;
3601
3602 sp->off_in = READ_ONCE(sqe->splice_off_in);
3603 sp->off_out = READ_ONCE(sqe->off);
3604 return __io_splice_prep(req, sqe);
3605}
3606
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003607static int io_splice(struct io_kiocb *req, unsigned int issue_flags)
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003608{
3609 struct io_splice *sp = &req->splice;
3610 struct file *in = sp->file_in;
3611 struct file *out = sp->file_out;
3612 unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED;
3613 loff_t *poff_in, *poff_out;
Pavel Begunkovc9687422020-05-04 23:00:54 +03003614 long ret = 0;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003615
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003616 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov2fb3e822020-05-01 17:09:38 +03003617 return -EAGAIN;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003618
3619 poff_in = (sp->off_in == -1) ? NULL : &sp->off_in;
3620 poff_out = (sp->off_out == -1) ? NULL : &sp->off_out;
Pavel Begunkovc9687422020-05-04 23:00:54 +03003621
Jens Axboe948a7742020-05-17 14:21:38 -06003622 if (sp->len)
Pavel Begunkovc9687422020-05-04 23:00:54 +03003623 ret = do_splice(in, poff_in, out, poff_out, sp->len, flags);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003624
Pavel Begunkove1d767f2021-03-19 17:22:43 +00003625 if (!(sp->flags & SPLICE_F_FD_IN_FIXED))
3626 io_put_file(in);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003627 req->flags &= ~REQ_F_NEED_CLEANUP;
3628
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003629 if (ret != sp->len)
3630 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003631 io_req_complete(req, ret);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003632 return 0;
3633}
3634
Jens Axboe2b188cc2019-01-07 10:46:33 -07003635/*
3636 * IORING_OP_NOP just posts a completion event, nothing else.
3637 */
Pavel Begunkov889fca72021-02-10 00:03:09 +00003638static int io_nop(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003639{
3640 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003641
Jens Axboedef596e2019-01-09 08:59:42 -07003642 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
3643 return -EINVAL;
3644
Pavel Begunkov889fca72021-02-10 00:03:09 +00003645 __io_req_complete(req, issue_flags, 0, 0);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003646 return 0;
3647}
3648
Pavel Begunkov1155c762021-02-18 18:29:38 +00003649static int io_fsync_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003650{
Jens Axboe6b063142019-01-10 22:13:58 -07003651 struct io_ring_ctx *ctx = req->ctx;
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003652
Jens Axboe09bb8392019-03-13 12:39:28 -06003653 if (!req->file)
3654 return -EBADF;
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003655
Jens Axboe6b063142019-01-10 22:13:58 -07003656 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboedef596e2019-01-09 08:59:42 -07003657 return -EINVAL;
Jens Axboeedafcce2019-01-09 09:16:05 -07003658 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003659 return -EINVAL;
3660
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003661 req->sync.flags = READ_ONCE(sqe->fsync_flags);
3662 if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC))
3663 return -EINVAL;
3664
3665 req->sync.off = READ_ONCE(sqe->off);
3666 req->sync.len = READ_ONCE(sqe->len);
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003667 return 0;
3668}
3669
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003670static int io_fsync(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe78912932020-01-14 22:09:06 -07003671{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003672 loff_t end = req->sync.off + req->sync.len;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003673 int ret;
3674
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003675 /* fsync always requires a blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003676 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003677 return -EAGAIN;
3678
Jens Axboe9adbd452019-12-20 08:45:55 -07003679 ret = vfs_fsync_range(req->file, req->sync.off,
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003680 end > 0 ? end : LLONG_MAX,
3681 req->sync.flags & IORING_FSYNC_DATASYNC);
3682 if (ret < 0)
3683 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003684 io_req_complete(req, ret);
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003685 return 0;
3686}
3687
Jens Axboed63d1b52019-12-10 10:38:56 -07003688static int io_fallocate_prep(struct io_kiocb *req,
3689 const struct io_uring_sqe *sqe)
3690{
3691 if (sqe->ioprio || sqe->buf_index || sqe->rw_flags)
3692 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03003693 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3694 return -EINVAL;
Jens Axboed63d1b52019-12-10 10:38:56 -07003695
3696 req->sync.off = READ_ONCE(sqe->off);
3697 req->sync.len = READ_ONCE(sqe->addr);
3698 req->sync.mode = READ_ONCE(sqe->len);
3699 return 0;
3700}
3701
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003702static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboed63d1b52019-12-10 10:38:56 -07003703{
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003704 int ret;
Jens Axboed63d1b52019-12-10 10:38:56 -07003705
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003706 /* fallocate always requiring blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003707 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003708 return -EAGAIN;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003709 ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off,
3710 req->sync.len);
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003711 if (ret < 0)
3712 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003713 io_req_complete(req, ret);
Jens Axboed63d1b52019-12-10 10:38:56 -07003714 return 0;
3715}
3716
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003717static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe15b71ab2019-12-11 11:20:36 -07003718{
Jens Axboef8748882020-01-08 17:47:02 -07003719 const char __user *fname;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003720 int ret;
3721
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003722 if (unlikely(sqe->ioprio || sqe->buf_index))
Jens Axboe15b71ab2019-12-11 11:20:36 -07003723 return -EINVAL;
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003724 if (unlikely(req->flags & REQ_F_FIXED_FILE))
Jens Axboecf3040c2020-02-06 21:31:40 -07003725 return -EBADF;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003726
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003727 /* open.how should be already initialised */
3728 if (!(req->open.how.flags & O_PATH) && force_o_largefile())
Jens Axboe08a1d26eb2020-04-08 09:20:54 -06003729 req->open.how.flags |= O_LARGEFILE;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003730
Pavel Begunkov25e72d12020-06-03 18:03:23 +03003731 req->open.dfd = READ_ONCE(sqe->fd);
3732 fname = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboef8748882020-01-08 17:47:02 -07003733 req->open.filename = getname(fname);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003734 if (IS_ERR(req->open.filename)) {
3735 ret = PTR_ERR(req->open.filename);
3736 req->open.filename = NULL;
3737 return ret;
3738 }
Jens Axboe4022e7a2020-03-19 19:23:18 -06003739 req->open.nofile = rlimit(RLIMIT_NOFILE);
Pavel Begunkov8fef80b2020-02-07 23:59:53 +03003740 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003741 return 0;
3742}
3743
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003744static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3745{
3746 u64 flags, mode;
3747
Jens Axboe14587a462020-09-05 11:36:08 -06003748 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe4eb8dde2020-09-18 19:36:24 -06003749 return -EINVAL;
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003750 mode = READ_ONCE(sqe->len);
3751 flags = READ_ONCE(sqe->open_flags);
3752 req->open.how = build_open_how(flags, mode);
3753 return __io_openat_prep(req, sqe);
3754}
3755
Jens Axboecebdb982020-01-08 17:59:24 -07003756static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3757{
3758 struct open_how __user *how;
Jens Axboecebdb982020-01-08 17:59:24 -07003759 size_t len;
3760 int ret;
3761
Jens Axboe14587a462020-09-05 11:36:08 -06003762 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe4eb8dde2020-09-18 19:36:24 -06003763 return -EINVAL;
Jens Axboecebdb982020-01-08 17:59:24 -07003764 how = u64_to_user_ptr(READ_ONCE(sqe->addr2));
3765 len = READ_ONCE(sqe->len);
Jens Axboecebdb982020-01-08 17:59:24 -07003766 if (len < OPEN_HOW_SIZE_VER0)
3767 return -EINVAL;
3768
3769 ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how,
3770 len);
3771 if (ret)
3772 return ret;
3773
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003774 return __io_openat_prep(req, sqe);
Jens Axboecebdb982020-01-08 17:59:24 -07003775}
3776
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003777static int io_openat2(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe15b71ab2019-12-11 11:20:36 -07003778{
3779 struct open_flags op;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003780 struct file *file;
Jens Axboe3a81fd02020-12-10 12:25:36 -07003781 bool nonblock_set;
3782 bool resolve_nonblock;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003783 int ret;
3784
Jens Axboecebdb982020-01-08 17:59:24 -07003785 ret = build_open_flags(&req->open.how, &op);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003786 if (ret)
3787 goto err;
Jens Axboe3a81fd02020-12-10 12:25:36 -07003788 nonblock_set = op.open_flag & O_NONBLOCK;
3789 resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003790 if (issue_flags & IO_URING_F_NONBLOCK) {
Jens Axboe3a81fd02020-12-10 12:25:36 -07003791 /*
3792 * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open,
3793 * it'll always -EAGAIN
3794 */
3795 if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE))
3796 return -EAGAIN;
3797 op.lookup_flags |= LOOKUP_CACHED;
3798 op.open_flag |= O_NONBLOCK;
3799 }
Jens Axboe15b71ab2019-12-11 11:20:36 -07003800
Jens Axboe4022e7a2020-03-19 19:23:18 -06003801 ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003802 if (ret < 0)
3803 goto err;
3804
3805 file = do_filp_open(req->open.dfd, req->open.filename, &op);
Jens Axboe3a81fd02020-12-10 12:25:36 -07003806 /* only retry if RESOLVE_CACHED wasn't already set by application */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003807 if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) &&
3808 file == ERR_PTR(-EAGAIN)) {
Jens Axboe3a81fd02020-12-10 12:25:36 -07003809 /*
3810 * We could hang on to this 'fd', but seems like marginal
3811 * gain for something that is now known to be a slower path.
3812 * So just put it, and we'll get a new one when we retry.
3813 */
3814 put_unused_fd(ret);
3815 return -EAGAIN;
3816 }
3817
Jens Axboe15b71ab2019-12-11 11:20:36 -07003818 if (IS_ERR(file)) {
3819 put_unused_fd(ret);
3820 ret = PTR_ERR(file);
3821 } else {
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003822 if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set)
Jens Axboe3a81fd02020-12-10 12:25:36 -07003823 file->f_flags &= ~O_NONBLOCK;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003824 fsnotify_open(file);
3825 fd_install(ret, file);
3826 }
3827err:
3828 putname(req->open.filename);
Pavel Begunkov8fef80b2020-02-07 23:59:53 +03003829 req->flags &= ~REQ_F_NEED_CLEANUP;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003830 if (ret < 0)
3831 req_set_fail_links(req);
Pavel Begunkov0bdf3392021-04-11 01:46:29 +01003832 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003833 return 0;
3834}
3835
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003836static int io_openat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboecebdb982020-01-08 17:59:24 -07003837{
Pavel Begunkove45cff52021-02-28 22:35:14 +00003838 return io_openat2(req, issue_flags);
Jens Axboecebdb982020-01-08 17:59:24 -07003839}
3840
Jens Axboe067524e2020-03-02 16:32:28 -07003841static int io_remove_buffers_prep(struct io_kiocb *req,
3842 const struct io_uring_sqe *sqe)
3843{
3844 struct io_provide_buf *p = &req->pbuf;
3845 u64 tmp;
3846
3847 if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off)
3848 return -EINVAL;
3849
3850 tmp = READ_ONCE(sqe->fd);
3851 if (!tmp || tmp > USHRT_MAX)
3852 return -EINVAL;
3853
3854 memset(p, 0, sizeof(*p));
3855 p->nbufs = tmp;
3856 p->bgid = READ_ONCE(sqe->buf_group);
3857 return 0;
3858}
3859
3860static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf,
3861 int bgid, unsigned nbufs)
3862{
3863 unsigned i = 0;
3864
3865 /* shouldn't happen */
3866 if (!nbufs)
3867 return 0;
3868
3869 /* the head kbuf is the list itself */
3870 while (!list_empty(&buf->list)) {
3871 struct io_buffer *nxt;
3872
3873 nxt = list_first_entry(&buf->list, struct io_buffer, list);
3874 list_del(&nxt->list);
3875 kfree(nxt);
3876 if (++i == nbufs)
3877 return i;
3878 }
3879 i++;
3880 kfree(buf);
Jens Axboe9e15c3a2021-03-13 12:29:43 -07003881 xa_erase(&ctx->io_buffers, bgid);
Jens Axboe067524e2020-03-02 16:32:28 -07003882
3883 return i;
3884}
3885
Pavel Begunkov889fca72021-02-10 00:03:09 +00003886static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe067524e2020-03-02 16:32:28 -07003887{
3888 struct io_provide_buf *p = &req->pbuf;
3889 struct io_ring_ctx *ctx = req->ctx;
3890 struct io_buffer *head;
3891 int ret = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003892 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe067524e2020-03-02 16:32:28 -07003893
3894 io_ring_submit_lock(ctx, !force_nonblock);
3895
3896 lockdep_assert_held(&ctx->uring_lock);
3897
3898 ret = -ENOENT;
Jens Axboe9e15c3a2021-03-13 12:29:43 -07003899 head = xa_load(&ctx->io_buffers, p->bgid);
Jens Axboe067524e2020-03-02 16:32:28 -07003900 if (head)
3901 ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs);
Jens Axboe067524e2020-03-02 16:32:28 -07003902 if (ret < 0)
3903 req_set_fail_links(req);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00003904
Pavel Begunkov9fb8cb42021-02-28 22:35:13 +00003905 /* complete before unlock, IOPOLL may need the lock */
3906 __io_req_complete(req, issue_flags, ret, 0);
3907 io_ring_submit_unlock(ctx, !force_nonblock);
Jens Axboe067524e2020-03-02 16:32:28 -07003908 return 0;
3909}
3910
Jens Axboeddf0322d2020-02-23 16:41:33 -07003911static int io_provide_buffers_prep(struct io_kiocb *req,
3912 const struct io_uring_sqe *sqe)
3913{
Pavel Begunkov38134ad2021-04-15 13:07:39 +01003914 unsigned long size, tmp_check;
Jens Axboeddf0322d2020-02-23 16:41:33 -07003915 struct io_provide_buf *p = &req->pbuf;
3916 u64 tmp;
3917
3918 if (sqe->ioprio || sqe->rw_flags)
3919 return -EINVAL;
3920
3921 tmp = READ_ONCE(sqe->fd);
3922 if (!tmp || tmp > USHRT_MAX)
3923 return -E2BIG;
3924 p->nbufs = tmp;
3925 p->addr = READ_ONCE(sqe->addr);
3926 p->len = READ_ONCE(sqe->len);
3927
Pavel Begunkov38134ad2021-04-15 13:07:39 +01003928 if (check_mul_overflow((unsigned long)p->len, (unsigned long)p->nbufs,
3929 &size))
3930 return -EOVERFLOW;
3931 if (check_add_overflow((unsigned long)p->addr, size, &tmp_check))
3932 return -EOVERFLOW;
3933
Pavel Begunkovd81269f2021-03-19 10:21:19 +00003934 size = (unsigned long)p->len * p->nbufs;
3935 if (!access_ok(u64_to_user_ptr(p->addr), size))
Jens Axboeddf0322d2020-02-23 16:41:33 -07003936 return -EFAULT;
3937
3938 p->bgid = READ_ONCE(sqe->buf_group);
3939 tmp = READ_ONCE(sqe->off);
3940 if (tmp > USHRT_MAX)
3941 return -E2BIG;
3942 p->bid = tmp;
3943 return 0;
3944}
3945
3946static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head)
3947{
3948 struct io_buffer *buf;
3949 u64 addr = pbuf->addr;
3950 int i, bid = pbuf->bid;
3951
3952 for (i = 0; i < pbuf->nbufs; i++) {
3953 buf = kmalloc(sizeof(*buf), GFP_KERNEL);
3954 if (!buf)
3955 break;
3956
3957 buf->addr = addr;
3958 buf->len = pbuf->len;
3959 buf->bid = bid;
3960 addr += pbuf->len;
3961 bid++;
3962 if (!*head) {
3963 INIT_LIST_HEAD(&buf->list);
3964 *head = buf;
3965 } else {
3966 list_add_tail(&buf->list, &(*head)->list);
3967 }
3968 }
3969
3970 return i ? i : -ENOMEM;
3971}
3972
Pavel Begunkov889fca72021-02-10 00:03:09 +00003973static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeddf0322d2020-02-23 16:41:33 -07003974{
3975 struct io_provide_buf *p = &req->pbuf;
3976 struct io_ring_ctx *ctx = req->ctx;
3977 struct io_buffer *head, *list;
3978 int ret = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003979 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboeddf0322d2020-02-23 16:41:33 -07003980
3981 io_ring_submit_lock(ctx, !force_nonblock);
3982
3983 lockdep_assert_held(&ctx->uring_lock);
3984
Jens Axboe9e15c3a2021-03-13 12:29:43 -07003985 list = head = xa_load(&ctx->io_buffers, p->bgid);
Jens Axboeddf0322d2020-02-23 16:41:33 -07003986
3987 ret = io_add_buffers(p, &head);
Jens Axboe9e15c3a2021-03-13 12:29:43 -07003988 if (ret >= 0 && !list) {
3989 ret = xa_insert(&ctx->io_buffers, p->bgid, head, GFP_KERNEL);
3990 if (ret < 0)
Jens Axboe067524e2020-03-02 16:32:28 -07003991 __io_remove_buffers(ctx, head, p->bgid, -1U);
Jens Axboeddf0322d2020-02-23 16:41:33 -07003992 }
Jens Axboeddf0322d2020-02-23 16:41:33 -07003993 if (ret < 0)
3994 req_set_fail_links(req);
Pavel Begunkov9fb8cb42021-02-28 22:35:13 +00003995 /* complete before unlock, IOPOLL may need the lock */
3996 __io_req_complete(req, issue_flags, ret, 0);
3997 io_ring_submit_unlock(ctx, !force_nonblock);
Jens Axboeddf0322d2020-02-23 16:41:33 -07003998 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003999}
4000
Jens Axboe3e4827b2020-01-08 15:18:09 -07004001static int io_epoll_ctl_prep(struct io_kiocb *req,
4002 const struct io_uring_sqe *sqe)
4003{
4004#if defined(CONFIG_EPOLL)
4005 if (sqe->ioprio || sqe->buf_index)
4006 return -EINVAL;
Jens Axboe6ca56f82020-09-18 16:51:19 -06004007 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL)))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004008 return -EINVAL;
Jens Axboe3e4827b2020-01-08 15:18:09 -07004009
4010 req->epoll.epfd = READ_ONCE(sqe->fd);
4011 req->epoll.op = READ_ONCE(sqe->len);
4012 req->epoll.fd = READ_ONCE(sqe->off);
4013
4014 if (ep_op_has_event(req->epoll.op)) {
4015 struct epoll_event __user *ev;
4016
4017 ev = u64_to_user_ptr(READ_ONCE(sqe->addr));
4018 if (copy_from_user(&req->epoll.event, ev, sizeof(*ev)))
4019 return -EFAULT;
4020 }
4021
4022 return 0;
4023#else
4024 return -EOPNOTSUPP;
4025#endif
4026}
4027
Pavel Begunkov889fca72021-02-10 00:03:09 +00004028static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe3e4827b2020-01-08 15:18:09 -07004029{
4030#if defined(CONFIG_EPOLL)
4031 struct io_epoll *ie = &req->epoll;
4032 int ret;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004033 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe3e4827b2020-01-08 15:18:09 -07004034
4035 ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock);
4036 if (force_nonblock && ret == -EAGAIN)
4037 return -EAGAIN;
4038
4039 if (ret < 0)
4040 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004041 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe3e4827b2020-01-08 15:18:09 -07004042 return 0;
4043#else
4044 return -EOPNOTSUPP;
4045#endif
4046}
4047
Jens Axboec1ca7572019-12-25 22:18:28 -07004048static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4049{
4050#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
4051 if (sqe->ioprio || sqe->buf_index || sqe->off)
4052 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004053 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4054 return -EINVAL;
Jens Axboec1ca7572019-12-25 22:18:28 -07004055
4056 req->madvise.addr = READ_ONCE(sqe->addr);
4057 req->madvise.len = READ_ONCE(sqe->len);
4058 req->madvise.advice = READ_ONCE(sqe->fadvise_advice);
4059 return 0;
4060#else
4061 return -EOPNOTSUPP;
4062#endif
4063}
4064
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004065static int io_madvise(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboec1ca7572019-12-25 22:18:28 -07004066{
4067#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
4068 struct io_madvise *ma = &req->madvise;
4069 int ret;
4070
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004071 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboec1ca7572019-12-25 22:18:28 -07004072 return -EAGAIN;
4073
Minchan Kim0726b012020-10-17 16:14:50 -07004074 ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice);
Jens Axboec1ca7572019-12-25 22:18:28 -07004075 if (ret < 0)
4076 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004077 io_req_complete(req, ret);
Jens Axboec1ca7572019-12-25 22:18:28 -07004078 return 0;
4079#else
4080 return -EOPNOTSUPP;
4081#endif
4082}
4083
Jens Axboe4840e412019-12-25 22:03:45 -07004084static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4085{
4086 if (sqe->ioprio || sqe->buf_index || sqe->addr)
4087 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004088 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4089 return -EINVAL;
Jens Axboe4840e412019-12-25 22:03:45 -07004090
4091 req->fadvise.offset = READ_ONCE(sqe->off);
4092 req->fadvise.len = READ_ONCE(sqe->len);
4093 req->fadvise.advice = READ_ONCE(sqe->fadvise_advice);
4094 return 0;
4095}
4096
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004097static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe4840e412019-12-25 22:03:45 -07004098{
4099 struct io_fadvise *fa = &req->fadvise;
4100 int ret;
4101
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004102 if (issue_flags & IO_URING_F_NONBLOCK) {
Jens Axboe3e694262020-02-01 09:22:49 -07004103 switch (fa->advice) {
4104 case POSIX_FADV_NORMAL:
4105 case POSIX_FADV_RANDOM:
4106 case POSIX_FADV_SEQUENTIAL:
4107 break;
4108 default:
4109 return -EAGAIN;
4110 }
4111 }
Jens Axboe4840e412019-12-25 22:03:45 -07004112
4113 ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice);
4114 if (ret < 0)
4115 req_set_fail_links(req);
Pavel Begunkov0bdf3392021-04-11 01:46:29 +01004116 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe4840e412019-12-25 22:03:45 -07004117 return 0;
4118}
4119
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004120static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4121{
Jens Axboe6ca56f82020-09-18 16:51:19 -06004122 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL)))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004123 return -EINVAL;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004124 if (sqe->ioprio || sqe->buf_index)
4125 return -EINVAL;
Pavel Begunkov9c280f92020-04-08 08:58:46 +03004126 if (req->flags & REQ_F_FIXED_FILE)
Jens Axboecf3040c2020-02-06 21:31:40 -07004127 return -EBADF;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004128
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004129 req->statx.dfd = READ_ONCE(sqe->fd);
4130 req->statx.mask = READ_ONCE(sqe->len);
Bijan Mottahedehe62753e2020-05-22 21:31:18 -07004131 req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr));
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004132 req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2));
4133 req->statx.flags = READ_ONCE(sqe->statx_flags);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004134
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004135 return 0;
4136}
4137
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004138static int io_statx(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004139{
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004140 struct io_statx *ctx = &req->statx;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004141 int ret;
4142
Pavel Begunkov59d70012021-03-22 01:58:30 +00004143 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004144 return -EAGAIN;
4145
Bijan Mottahedehe62753e2020-05-22 21:31:18 -07004146 ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask,
4147 ctx->buffer);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004148
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004149 if (ret < 0)
4150 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004151 io_req_complete(req, ret);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004152 return 0;
4153}
4154
Jens Axboeb5dba592019-12-11 14:02:38 -07004155static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4156{
Jens Axboe14587a462020-09-05 11:36:08 -06004157 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004158 return -EINVAL;
Jens Axboeb5dba592019-12-11 14:02:38 -07004159 if (sqe->ioprio || sqe->off || sqe->addr || sqe->len ||
4160 sqe->rw_flags || sqe->buf_index)
4161 return -EINVAL;
Pavel Begunkov9c280f92020-04-08 08:58:46 +03004162 if (req->flags & REQ_F_FIXED_FILE)
Jens Axboecf3040c2020-02-06 21:31:40 -07004163 return -EBADF;
Jens Axboeb5dba592019-12-11 14:02:38 -07004164
4165 req->close.fd = READ_ONCE(sqe->fd);
Jens Axboeb5dba592019-12-11 14:02:38 -07004166 return 0;
4167}
4168
Pavel Begunkov889fca72021-02-10 00:03:09 +00004169static int io_close(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeb5dba592019-12-11 14:02:38 -07004170{
Jens Axboe9eac1902021-01-19 15:50:37 -07004171 struct files_struct *files = current->files;
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004172 struct io_close *close = &req->close;
Jens Axboe9eac1902021-01-19 15:50:37 -07004173 struct fdtable *fdt;
Pavel Begunkova1fde922021-04-11 01:46:28 +01004174 struct file *file = NULL;
4175 int ret = -EBADF;
Jens Axboeb5dba592019-12-11 14:02:38 -07004176
Jens Axboe9eac1902021-01-19 15:50:37 -07004177 spin_lock(&files->file_lock);
4178 fdt = files_fdtable(files);
4179 if (close->fd >= fdt->max_fds) {
4180 spin_unlock(&files->file_lock);
4181 goto err;
4182 }
4183 file = fdt->fd[close->fd];
Pavel Begunkova1fde922021-04-11 01:46:28 +01004184 if (!file || file->f_op == &io_uring_fops) {
Jens Axboe9eac1902021-01-19 15:50:37 -07004185 spin_unlock(&files->file_lock);
4186 file = NULL;
4187 goto err;
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004188 }
Jens Axboeb5dba592019-12-11 14:02:38 -07004189
4190 /* if the file has a flush method, be safe and punt to async */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004191 if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) {
Jens Axboe9eac1902021-01-19 15:50:37 -07004192 spin_unlock(&files->file_lock);
Pavel Begunkov0bf0eef2020-05-26 20:34:06 +03004193 return -EAGAIN;
Pavel Begunkova2100672020-03-02 23:45:16 +03004194 }
Jens Axboeb5dba592019-12-11 14:02:38 -07004195
Jens Axboe9eac1902021-01-19 15:50:37 -07004196 ret = __close_fd_get_file(close->fd, &file);
4197 spin_unlock(&files->file_lock);
4198 if (ret < 0) {
4199 if (ret == -ENOENT)
4200 ret = -EBADF;
4201 goto err;
4202 }
4203
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004204 /* No ->flush() or already async, safely close from here */
Jens Axboe9eac1902021-01-19 15:50:37 -07004205 ret = filp_close(file, current->files);
4206err:
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004207 if (ret < 0)
4208 req_set_fail_links(req);
Jens Axboe9eac1902021-01-19 15:50:37 -07004209 if (file)
4210 fput(file);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004211 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe1a417f42020-01-31 17:16:48 -07004212 return 0;
Jens Axboeb5dba592019-12-11 14:02:38 -07004213}
4214
Pavel Begunkov1155c762021-02-18 18:29:38 +00004215static int io_sfr_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004216{
4217 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004218
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004219 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
4220 return -EINVAL;
4221 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
4222 return -EINVAL;
4223
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004224 req->sync.off = READ_ONCE(sqe->off);
4225 req->sync.len = READ_ONCE(sqe->len);
4226 req->sync.flags = READ_ONCE(sqe->sync_range_flags);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004227 return 0;
4228}
4229
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004230static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004231{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004232 int ret;
4233
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004234 /* sync_file_range always requires a blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004235 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004236 return -EAGAIN;
4237
Jens Axboe9adbd452019-12-20 08:45:55 -07004238 ret = sync_file_range(req->file, req->sync.off, req->sync.len,
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004239 req->sync.flags);
4240 if (ret < 0)
4241 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004242 io_req_complete(req, ret);
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004243 return 0;
4244}
4245
YueHaibing469956e2020-03-04 15:53:52 +08004246#if defined(CONFIG_NET)
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004247static int io_setup_async_msg(struct io_kiocb *req,
4248 struct io_async_msghdr *kmsg)
4249{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004250 struct io_async_msghdr *async_msg = req->async_data;
4251
4252 if (async_msg)
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004253 return -EAGAIN;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004254 if (io_alloc_async_data(req)) {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004255 kfree(kmsg->free_iov);
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004256 return -ENOMEM;
4257 }
Jens Axboee8c2bc12020-08-15 18:44:09 -07004258 async_msg = req->async_data;
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004259 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004260 memcpy(async_msg, kmsg, sizeof(*kmsg));
Pavel Begunkov2a780802021-02-05 00:57:58 +00004261 async_msg->msg.msg_name = &async_msg->addr;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004262 /* if were using fast_iov, set it to the new one */
4263 if (!async_msg->free_iov)
4264 async_msg->msg.msg_iter.iov = async_msg->fast_iov;
4265
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004266 return -EAGAIN;
4267}
4268
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004269static int io_sendmsg_copy_hdr(struct io_kiocb *req,
4270 struct io_async_msghdr *iomsg)
4271{
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004272 iomsg->msg.msg_name = &iomsg->addr;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004273 iomsg->free_iov = iomsg->fast_iov;
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004274 return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004275 req->sr_msg.msg_flags, &iomsg->free_iov);
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004276}
4277
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004278static int io_sendmsg_prep_async(struct io_kiocb *req)
4279{
4280 int ret;
4281
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004282 ret = io_sendmsg_copy_hdr(req, req->async_data);
4283 if (!ret)
4284 req->flags |= REQ_F_NEED_CLEANUP;
4285 return ret;
4286}
4287
Jens Axboe3529d8c2019-12-19 18:24:38 -07004288static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboeaa1fa282019-04-19 13:38:09 -06004289{
Jens Axboee47293f2019-12-20 08:58:21 -07004290 struct io_sr_msg *sr = &req->sr_msg;
Jens Axboe03b12302019-12-02 18:50:25 -07004291
Pavel Begunkovd2b6f482020-06-03 18:03:25 +03004292 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4293 return -EINVAL;
4294
Pavel Begunkov270a5942020-07-12 20:41:04 +03004295 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboefddafac2020-01-04 20:19:44 -07004296 sr->len = READ_ONCE(sqe->len);
Pavel Begunkov04411802021-04-01 15:44:00 +01004297 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
4298 if (sr->msg_flags & MSG_DONTWAIT)
4299 req->flags |= REQ_F_NOWAIT;
Jens Axboe3529d8c2019-12-19 18:24:38 -07004300
Jens Axboed8768362020-02-27 14:17:49 -07004301#ifdef CONFIG_COMPAT
4302 if (req->ctx->compat)
4303 sr->msg_flags |= MSG_CMSG_COMPAT;
4304#endif
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004305 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004306}
4307
Pavel Begunkov889fca72021-02-10 00:03:09 +00004308static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004309{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004310 struct io_async_msghdr iomsg, *kmsg;
Jens Axboe03b12302019-12-02 18:50:25 -07004311 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004312 unsigned flags;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004313 int min_ret = 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004314 int ret;
4315
Florent Revestdba4a922020-12-04 12:36:04 +01004316 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004317 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004318 return -ENOTSOCK;
Jens Axboe03b12302019-12-02 18:50:25 -07004319
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004320 kmsg = req->async_data;
4321 if (!kmsg) {
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004322 ret = io_sendmsg_copy_hdr(req, &iomsg);
Jens Axboefddafac2020-01-04 20:19:44 -07004323 if (ret)
4324 return ret;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004325 kmsg = &iomsg;
Jens Axboefddafac2020-01-04 20:19:44 -07004326 }
4327
Pavel Begunkov04411802021-04-01 15:44:00 +01004328 flags = req->sr_msg.msg_flags;
4329 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004330 flags |= MSG_DONTWAIT;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004331 if (flags & MSG_WAITALL)
4332 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
4333
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004334 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004335 if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004336 return io_setup_async_msg(req, kmsg);
4337 if (ret == -ERESTARTSYS)
4338 ret = -EINTR;
4339
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004340 /* fast path, check for non-NULL to avoid function call */
4341 if (kmsg->free_iov)
4342 kfree(kmsg->free_iov);
Jens Axboe03b12302019-12-02 18:50:25 -07004343 req->flags &= ~REQ_F_NEED_CLEANUP;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004344 if (ret < min_ret)
Jens Axboefddafac2020-01-04 20:19:44 -07004345 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004346 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboefddafac2020-01-04 20:19:44 -07004347 return 0;
Jens Axboefddafac2020-01-04 20:19:44 -07004348}
4349
Pavel Begunkov889fca72021-02-10 00:03:09 +00004350static int io_send(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004351{
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004352 struct io_sr_msg *sr = &req->sr_msg;
4353 struct msghdr msg;
4354 struct iovec iov;
Jens Axboe03b12302019-12-02 18:50:25 -07004355 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004356 unsigned flags;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004357 int min_ret = 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004358 int ret;
4359
Florent Revestdba4a922020-12-04 12:36:04 +01004360 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004361 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004362 return -ENOTSOCK;
Jens Axboe03b12302019-12-02 18:50:25 -07004363
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004364 ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter);
4365 if (unlikely(ret))
Zheng Bin14db8412020-09-09 20:12:37 +08004366 return ret;
Jens Axboe03b12302019-12-02 18:50:25 -07004367
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004368 msg.msg_name = NULL;
4369 msg.msg_control = NULL;
4370 msg.msg_controllen = 0;
4371 msg.msg_namelen = 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004372
Pavel Begunkov04411802021-04-01 15:44:00 +01004373 flags = req->sr_msg.msg_flags;
4374 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004375 flags |= MSG_DONTWAIT;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004376 if (flags & MSG_WAITALL)
4377 min_ret = iov_iter_count(&msg.msg_iter);
4378
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004379 msg.msg_flags = flags;
4380 ret = sock_sendmsg(sock, &msg);
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004381 if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004382 return -EAGAIN;
4383 if (ret == -ERESTARTSYS)
4384 ret = -EINTR;
Jens Axboe03b12302019-12-02 18:50:25 -07004385
Stefan Metzmacher00312752021-03-20 20:33:36 +01004386 if (ret < min_ret)
Jens Axboe03b12302019-12-02 18:50:25 -07004387 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004388 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe03b12302019-12-02 18:50:25 -07004389 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004390}
4391
Pavel Begunkov1400e692020-07-12 20:41:05 +03004392static int __io_recvmsg_copy_hdr(struct io_kiocb *req,
4393 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004394{
4395 struct io_sr_msg *sr = &req->sr_msg;
4396 struct iovec __user *uiov;
4397 size_t iov_len;
4398 int ret;
4399
Pavel Begunkov1400e692020-07-12 20:41:05 +03004400 ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg,
4401 &iomsg->uaddr, &uiov, &iov_len);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004402 if (ret)
4403 return ret;
4404
4405 if (req->flags & REQ_F_BUFFER_SELECT) {
4406 if (iov_len > 1)
4407 return -EINVAL;
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004408 if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov)))
Jens Axboe52de1fe2020-02-27 10:15:42 -07004409 return -EFAULT;
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004410 sr->len = iomsg->fast_iov[0].iov_len;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004411 iomsg->free_iov = NULL;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004412 } else {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004413 iomsg->free_iov = iomsg->fast_iov;
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004414 ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004415 &iomsg->free_iov, &iomsg->msg.msg_iter,
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004416 false);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004417 if (ret > 0)
4418 ret = 0;
4419 }
4420
4421 return ret;
4422}
4423
4424#ifdef CONFIG_COMPAT
4425static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req,
Pavel Begunkov1400e692020-07-12 20:41:05 +03004426 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004427{
Jens Axboe52de1fe2020-02-27 10:15:42 -07004428 struct io_sr_msg *sr = &req->sr_msg;
4429 struct compat_iovec __user *uiov;
4430 compat_uptr_t ptr;
4431 compat_size_t len;
4432 int ret;
4433
Pavel Begunkov4af34172021-04-11 01:46:30 +01004434 ret = __get_compat_msghdr(&iomsg->msg, sr->umsg_compat, &iomsg->uaddr,
4435 &ptr, &len);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004436 if (ret)
4437 return ret;
4438
4439 uiov = compat_ptr(ptr);
4440 if (req->flags & REQ_F_BUFFER_SELECT) {
4441 compat_ssize_t clen;
4442
4443 if (len > 1)
4444 return -EINVAL;
4445 if (!access_ok(uiov, sizeof(*uiov)))
4446 return -EFAULT;
4447 if (__get_user(clen, &uiov->iov_len))
4448 return -EFAULT;
4449 if (clen < 0)
4450 return -EINVAL;
Pavel Begunkov2d280bc2020-11-29 18:33:32 +00004451 sr->len = clen;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004452 iomsg->free_iov = NULL;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004453 } else {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004454 iomsg->free_iov = iomsg->fast_iov;
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004455 ret = __import_iovec(READ, (struct iovec __user *)uiov, len,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004456 UIO_FASTIOV, &iomsg->free_iov,
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004457 &iomsg->msg.msg_iter, true);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004458 if (ret < 0)
4459 return ret;
4460 }
4461
4462 return 0;
4463}
Jens Axboe03b12302019-12-02 18:50:25 -07004464#endif
Jens Axboe52de1fe2020-02-27 10:15:42 -07004465
Pavel Begunkov1400e692020-07-12 20:41:05 +03004466static int io_recvmsg_copy_hdr(struct io_kiocb *req,
4467 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004468{
Pavel Begunkov1400e692020-07-12 20:41:05 +03004469 iomsg->msg.msg_name = &iomsg->addr;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004470
4471#ifdef CONFIG_COMPAT
4472 if (req->ctx->compat)
Pavel Begunkov1400e692020-07-12 20:41:05 +03004473 return __io_compat_recvmsg_copy_hdr(req, iomsg);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004474#endif
4475
Pavel Begunkov1400e692020-07-12 20:41:05 +03004476 return __io_recvmsg_copy_hdr(req, iomsg);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004477}
4478
Jens Axboebcda7ba2020-02-23 16:42:51 -07004479static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req,
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004480 bool needs_lock)
Jens Axboebcda7ba2020-02-23 16:42:51 -07004481{
4482 struct io_sr_msg *sr = &req->sr_msg;
4483 struct io_buffer *kbuf;
4484
Jens Axboebcda7ba2020-02-23 16:42:51 -07004485 kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock);
4486 if (IS_ERR(kbuf))
4487 return kbuf;
4488
4489 sr->kbuf = kbuf;
4490 req->flags |= REQ_F_BUFFER_SELECTED;
Jens Axboebcda7ba2020-02-23 16:42:51 -07004491 return kbuf;
Jens Axboe03b12302019-12-02 18:50:25 -07004492}
4493
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004494static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req)
4495{
4496 return io_put_kbuf(req, req->sr_msg.kbuf);
4497}
4498
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004499static int io_recvmsg_prep_async(struct io_kiocb *req)
Jens Axboe03b12302019-12-02 18:50:25 -07004500{
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004501 int ret;
Jens Axboe06b76d42019-12-19 14:44:26 -07004502
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004503 ret = io_recvmsg_copy_hdr(req, req->async_data);
4504 if (!ret)
4505 req->flags |= REQ_F_NEED_CLEANUP;
4506 return ret;
4507}
4508
4509static int io_recvmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4510{
4511 struct io_sr_msg *sr = &req->sr_msg;
4512
Pavel Begunkovd2b6f482020-06-03 18:03:25 +03004513 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4514 return -EINVAL;
4515
Pavel Begunkov270a5942020-07-12 20:41:04 +03004516 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboe0b7b21e2020-01-31 08:34:59 -07004517 sr->len = READ_ONCE(sqe->len);
Jens Axboebcda7ba2020-02-23 16:42:51 -07004518 sr->bgid = READ_ONCE(sqe->buf_group);
Pavel Begunkov04411802021-04-01 15:44:00 +01004519 sr->msg_flags = READ_ONCE(sqe->msg_flags) | MSG_NOSIGNAL;
4520 if (sr->msg_flags & MSG_DONTWAIT)
4521 req->flags |= REQ_F_NOWAIT;
Jens Axboe3529d8c2019-12-19 18:24:38 -07004522
Jens Axboed8768362020-02-27 14:17:49 -07004523#ifdef CONFIG_COMPAT
4524 if (req->ctx->compat)
4525 sr->msg_flags |= MSG_CMSG_COMPAT;
4526#endif
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004527 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004528}
4529
Pavel Begunkov889fca72021-02-10 00:03:09 +00004530static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004531{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004532 struct io_async_msghdr iomsg, *kmsg;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004533 struct socket *sock;
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004534 struct io_buffer *kbuf;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004535 unsigned flags;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004536 int min_ret = 0;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004537 int ret, cflags = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004538 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004539
Florent Revestdba4a922020-12-04 12:36:04 +01004540 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004541 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004542 return -ENOTSOCK;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004543
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004544 kmsg = req->async_data;
4545 if (!kmsg) {
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004546 ret = io_recvmsg_copy_hdr(req, &iomsg);
4547 if (ret)
Pavel Begunkov681fda82020-07-15 22:20:45 +03004548 return ret;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004549 kmsg = &iomsg;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004550 }
4551
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004552 if (req->flags & REQ_F_BUFFER_SELECT) {
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004553 kbuf = io_recv_buffer_select(req, !force_nonblock);
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004554 if (IS_ERR(kbuf))
4555 return PTR_ERR(kbuf);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004556 kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr);
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004557 kmsg->fast_iov[0].iov_len = req->sr_msg.len;
4558 iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov,
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004559 1, req->sr_msg.len);
4560 }
4561
Pavel Begunkov04411802021-04-01 15:44:00 +01004562 flags = req->sr_msg.msg_flags;
4563 if (force_nonblock)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004564 flags |= MSG_DONTWAIT;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004565 if (flags & MSG_WAITALL)
4566 min_ret = iov_iter_count(&kmsg->msg.msg_iter);
4567
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004568 ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg,
4569 kmsg->uaddr, flags);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03004570 if (force_nonblock && ret == -EAGAIN)
4571 return io_setup_async_msg(req, kmsg);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004572 if (ret == -ERESTARTSYS)
4573 ret = -EINTR;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03004574
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004575 if (req->flags & REQ_F_BUFFER_SELECTED)
4576 cflags = io_put_recv_kbuf(req);
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004577 /* fast path, check for non-NULL to avoid function call */
4578 if (kmsg->free_iov)
4579 kfree(kmsg->free_iov);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004580 req->flags &= ~REQ_F_NEED_CLEANUP;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004581 if (ret < min_ret || ((flags & MSG_WAITALL) && (kmsg->msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))))
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004582 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004583 __io_req_complete(req, issue_flags, ret, cflags);
Jens Axboe0fa03c62019-04-19 13:34:07 -06004584 return 0;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004585}
4586
Pavel Begunkov889fca72021-02-10 00:03:09 +00004587static int io_recv(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboefddafac2020-01-04 20:19:44 -07004588{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004589 struct io_buffer *kbuf;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004590 struct io_sr_msg *sr = &req->sr_msg;
4591 struct msghdr msg;
4592 void __user *buf = sr->buf;
Jens Axboefddafac2020-01-04 20:19:44 -07004593 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004594 struct iovec iov;
4595 unsigned flags;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004596 int min_ret = 0;
Jens Axboebcda7ba2020-02-23 16:42:51 -07004597 int ret, cflags = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004598 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboefddafac2020-01-04 20:19:44 -07004599
Florent Revestdba4a922020-12-04 12:36:04 +01004600 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004601 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004602 return -ENOTSOCK;
Jens Axboefddafac2020-01-04 20:19:44 -07004603
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004604 if (req->flags & REQ_F_BUFFER_SELECT) {
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004605 kbuf = io_recv_buffer_select(req, !force_nonblock);
Jens Axboebcda7ba2020-02-23 16:42:51 -07004606 if (IS_ERR(kbuf))
4607 return PTR_ERR(kbuf);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004608 buf = u64_to_user_ptr(kbuf->addr);
Jens Axboefddafac2020-01-04 20:19:44 -07004609 }
4610
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004611 ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter);
Pavel Begunkov14c32ee2020-07-16 23:28:01 +03004612 if (unlikely(ret))
4613 goto out_free;
Jens Axboefddafac2020-01-04 20:19:44 -07004614
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004615 msg.msg_name = NULL;
4616 msg.msg_control = NULL;
4617 msg.msg_controllen = 0;
4618 msg.msg_namelen = 0;
4619 msg.msg_iocb = NULL;
4620 msg.msg_flags = 0;
4621
Pavel Begunkov04411802021-04-01 15:44:00 +01004622 flags = req->sr_msg.msg_flags;
4623 if (force_nonblock)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004624 flags |= MSG_DONTWAIT;
Stefan Metzmacher00312752021-03-20 20:33:36 +01004625 if (flags & MSG_WAITALL)
4626 min_ret = iov_iter_count(&msg.msg_iter);
4627
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004628 ret = sock_recvmsg(sock, &msg, flags);
4629 if (force_nonblock && ret == -EAGAIN)
4630 return -EAGAIN;
4631 if (ret == -ERESTARTSYS)
4632 ret = -EINTR;
Pavel Begunkov14c32ee2020-07-16 23:28:01 +03004633out_free:
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004634 if (req->flags & REQ_F_BUFFER_SELECTED)
4635 cflags = io_put_recv_kbuf(req);
Stefan Metzmacher00312752021-03-20 20:33:36 +01004636 if (ret < min_ret || ((flags & MSG_WAITALL) && (msg.msg_flags & (MSG_TRUNC | MSG_CTRUNC))))
Jens Axboefddafac2020-01-04 20:19:44 -07004637 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004638 __io_req_complete(req, issue_flags, ret, cflags);
Jens Axboefddafac2020-01-04 20:19:44 -07004639 return 0;
Jens Axboefddafac2020-01-04 20:19:44 -07004640}
4641
Jens Axboe3529d8c2019-12-19 18:24:38 -07004642static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004643{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004644 struct io_accept *accept = &req->accept;
4645
Jens Axboe14587a462020-09-05 11:36:08 -06004646 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe17f2fe32019-10-17 14:42:58 -06004647 return -EINVAL;
Hrvoje Zeba8042d6c2019-11-25 14:40:22 -05004648 if (sqe->ioprio || sqe->len || sqe->buf_index)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004649 return -EINVAL;
4650
Jens Axboed55e5f52019-12-11 16:12:15 -07004651 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
4652 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004653 accept->flags = READ_ONCE(sqe->accept_flags);
Jens Axboe09952e32020-03-19 20:16:56 -06004654 accept->nofile = rlimit(RLIMIT_NOFILE);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004655 return 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004656}
Jens Axboe17f2fe32019-10-17 14:42:58 -06004657
Pavel Begunkov889fca72021-02-10 00:03:09 +00004658static int io_accept(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004659{
4660 struct io_accept *accept = &req->accept;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004661 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004662 unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004663 int ret;
4664
Jiufei Xuee697dee2020-06-10 13:41:59 +08004665 if (req->file->f_flags & O_NONBLOCK)
4666 req->flags |= REQ_F_NOWAIT;
4667
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004668 ret = __sys_accept4_file(req->file, file_flags, accept->addr,
Jens Axboe09952e32020-03-19 20:16:56 -06004669 accept->addr_len, accept->flags,
4670 accept->nofile);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004671 if (ret == -EAGAIN && force_nonblock)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004672 return -EAGAIN;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004673 if (ret < 0) {
4674 if (ret == -ERESTARTSYS)
4675 ret = -EINTR;
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004676 req_set_fail_links(req);
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004677 }
Pavel Begunkov889fca72021-02-10 00:03:09 +00004678 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe17f2fe32019-10-17 14:42:58 -06004679 return 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004680}
4681
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004682static int io_connect_prep_async(struct io_kiocb *req)
4683{
4684 struct io_async_connect *io = req->async_data;
4685 struct io_connect *conn = &req->connect;
4686
4687 return move_addr_to_kernel(conn->addr, conn->addr_len, &io->address);
4688}
4689
Jens Axboe3529d8c2019-12-19 18:24:38 -07004690static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef499a022019-12-02 16:28:46 -07004691{
Jens Axboe3529d8c2019-12-19 18:24:38 -07004692 struct io_connect *conn = &req->connect;
Jens Axboef499a022019-12-02 16:28:46 -07004693
Jens Axboe14587a462020-09-05 11:36:08 -06004694 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004695 return -EINVAL;
4696 if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags)
4697 return -EINVAL;
4698
Jens Axboe3529d8c2019-12-19 18:24:38 -07004699 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
4700 conn->addr_len = READ_ONCE(sqe->addr2);
Pavel Begunkov93642ef2021-02-18 18:29:44 +00004701 return 0;
Jens Axboef499a022019-12-02 16:28:46 -07004702}
4703
Pavel Begunkov889fca72021-02-10 00:03:09 +00004704static int io_connect(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboef8e85cf2019-11-23 14:24:24 -07004705{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004706 struct io_async_connect __io, *io;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004707 unsigned file_flags;
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004708 int ret;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004709 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004710
Jens Axboee8c2bc12020-08-15 18:44:09 -07004711 if (req->async_data) {
4712 io = req->async_data;
Jens Axboef499a022019-12-02 16:28:46 -07004713 } else {
Jens Axboe3529d8c2019-12-19 18:24:38 -07004714 ret = move_addr_to_kernel(req->connect.addr,
4715 req->connect.addr_len,
Jens Axboee8c2bc12020-08-15 18:44:09 -07004716 &__io.address);
Jens Axboef499a022019-12-02 16:28:46 -07004717 if (ret)
4718 goto out;
4719 io = &__io;
4720 }
4721
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004722 file_flags = force_nonblock ? O_NONBLOCK : 0;
4723
Jens Axboee8c2bc12020-08-15 18:44:09 -07004724 ret = __sys_connect_file(req->file, &io->address,
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004725 req->connect.addr_len, file_flags);
Jens Axboe87f80d62019-12-03 11:23:54 -07004726 if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07004727 if (req->async_data)
Jens Axboeb7bb4f72019-12-15 22:13:43 -07004728 return -EAGAIN;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004729 if (io_alloc_async_data(req)) {
Jens Axboef499a022019-12-02 16:28:46 -07004730 ret = -ENOMEM;
4731 goto out;
4732 }
Jens Axboee8c2bc12020-08-15 18:44:09 -07004733 memcpy(req->async_data, &__io, sizeof(__io));
Jens Axboef8e85cf2019-11-23 14:24:24 -07004734 return -EAGAIN;
Jens Axboef499a022019-12-02 16:28:46 -07004735 }
Jens Axboef8e85cf2019-11-23 14:24:24 -07004736 if (ret == -ERESTARTSYS)
4737 ret = -EINTR;
Jens Axboef499a022019-12-02 16:28:46 -07004738out:
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004739 if (ret < 0)
4740 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004741 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboef8e85cf2019-11-23 14:24:24 -07004742 return 0;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004743}
YueHaibing469956e2020-03-04 15:53:52 +08004744#else /* !CONFIG_NET */
Jens Axboe99a10082021-02-19 09:35:19 -07004745#define IO_NETOP_FN(op) \
4746static int io_##op(struct io_kiocb *req, unsigned int issue_flags) \
4747{ \
4748 return -EOPNOTSUPP; \
Jens Axboef8e85cf2019-11-23 14:24:24 -07004749}
4750
Jens Axboe99a10082021-02-19 09:35:19 -07004751#define IO_NETOP_PREP(op) \
4752IO_NETOP_FN(op) \
4753static int io_##op##_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) \
4754{ \
4755 return -EOPNOTSUPP; \
4756} \
4757
4758#define IO_NETOP_PREP_ASYNC(op) \
4759IO_NETOP_PREP(op) \
4760static int io_##op##_prep_async(struct io_kiocb *req) \
4761{ \
4762 return -EOPNOTSUPP; \
YueHaibing469956e2020-03-04 15:53:52 +08004763}
4764
Jens Axboe99a10082021-02-19 09:35:19 -07004765IO_NETOP_PREP_ASYNC(sendmsg);
4766IO_NETOP_PREP_ASYNC(recvmsg);
4767IO_NETOP_PREP_ASYNC(connect);
4768IO_NETOP_PREP(accept);
4769IO_NETOP_FN(send);
4770IO_NETOP_FN(recv);
YueHaibing469956e2020-03-04 15:53:52 +08004771#endif /* CONFIG_NET */
Jens Axboe17f2fe32019-10-17 14:42:58 -06004772
Jens Axboed7718a92020-02-14 22:23:12 -07004773struct io_poll_table {
4774 struct poll_table_struct pt;
4775 struct io_kiocb *req;
4776 int error;
4777};
4778
Jens Axboed7718a92020-02-14 22:23:12 -07004779static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll,
4780 __poll_t mask, task_work_func_t func)
4781{
Jens Axboeaa96bf82020-04-03 11:26:26 -06004782 int ret;
Jens Axboed7718a92020-02-14 22:23:12 -07004783
4784 /* for instances that support it check for an event match first: */
4785 if (mask && !(mask & poll->events))
4786 return 0;
4787
4788 trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask);
4789
4790 list_del_init(&poll->wait.entry);
4791
Jens Axboed7718a92020-02-14 22:23:12 -07004792 req->result = mask;
Jens Axboe7cbf1722021-02-10 00:03:20 +00004793 req->task_work.func = func;
Jens Axboe6d816e02020-08-11 08:04:14 -06004794
Jens Axboed7718a92020-02-14 22:23:12 -07004795 /*
Jens Axboee3aabf92020-05-18 11:04:17 -06004796 * If this fails, then the task is exiting. When a task exits, the
4797 * work gets canceled, so just cancel this request as well instead
4798 * of executing it. We can't safely execute it anyway, as we may not
4799 * have the needed state needed for it anyway.
Jens Axboed7718a92020-02-14 22:23:12 -07004800 */
Jens Axboe355fb9e2020-10-22 20:19:35 -06004801 ret = io_req_task_work_add(req);
Jens Axboeaa96bf82020-04-03 11:26:26 -06004802 if (unlikely(ret)) {
Jens Axboee3aabf92020-05-18 11:04:17 -06004803 WRITE_ONCE(poll->canceled, true);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00004804 io_req_task_work_add_fallback(req, func);
Jens Axboeaa96bf82020-04-03 11:26:26 -06004805 }
Jens Axboed7718a92020-02-14 22:23:12 -07004806 return 1;
4807}
4808
Jens Axboe74ce6ce2020-04-13 11:09:12 -06004809static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll)
4810 __acquires(&req->ctx->completion_lock)
4811{
4812 struct io_ring_ctx *ctx = req->ctx;
4813
4814 if (!req->result && !READ_ONCE(poll->canceled)) {
4815 struct poll_table_struct pt = { ._key = poll->events };
4816
4817 req->result = vfs_poll(req->file, &pt) & poll->events;
4818 }
4819
4820 spin_lock_irq(&ctx->completion_lock);
4821 if (!req->result && !READ_ONCE(poll->canceled)) {
4822 add_wait_queue(poll->head, &poll->wait);
4823 return true;
4824 }
4825
4826 return false;
4827}
4828
Jens Axboed4e7cd32020-08-15 11:44:50 -07004829static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req)
Jens Axboe18bceab2020-05-15 11:56:54 -06004830{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004831 /* pure poll stashes this in ->async_data, poll driven retry elsewhere */
Jens Axboed4e7cd32020-08-15 11:44:50 -07004832 if (req->opcode == IORING_OP_POLL_ADD)
Jens Axboee8c2bc12020-08-15 18:44:09 -07004833 return req->async_data;
Jens Axboed4e7cd32020-08-15 11:44:50 -07004834 return req->apoll->double_poll;
4835}
4836
4837static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req)
4838{
4839 if (req->opcode == IORING_OP_POLL_ADD)
4840 return &req->poll;
4841 return &req->apoll->poll;
4842}
4843
4844static void io_poll_remove_double(struct io_kiocb *req)
Pavel Begunkove07785b2021-04-01 15:43:57 +01004845 __must_hold(&req->ctx->completion_lock)
Jens Axboed4e7cd32020-08-15 11:44:50 -07004846{
4847 struct io_poll_iocb *poll = io_poll_get_double(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004848
4849 lockdep_assert_held(&req->ctx->completion_lock);
4850
4851 if (poll && poll->head) {
4852 struct wait_queue_head *head = poll->head;
4853
4854 spin_lock(&head->lock);
4855 list_del_init(&poll->wait.entry);
4856 if (poll->wait.private)
Jens Axboede9b4cc2021-02-24 13:28:27 -07004857 req_ref_put(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004858 poll->head = NULL;
4859 spin_unlock(&head->lock);
4860 }
4861}
4862
Pavel Begunkove27414b2021-04-09 09:13:20 +01004863static bool io_poll_complete(struct io_kiocb *req, __poll_t mask)
Pavel Begunkove07785b2021-04-01 15:43:57 +01004864 __must_hold(&req->ctx->completion_lock)
Jens Axboe18bceab2020-05-15 11:56:54 -06004865{
4866 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe88e41cf2021-02-22 22:08:01 -07004867 unsigned flags = IORING_CQE_F_MORE;
Pavel Begunkove27414b2021-04-09 09:13:20 +01004868 int error;
Jens Axboe18bceab2020-05-15 11:56:54 -06004869
Pavel Begunkove27414b2021-04-09 09:13:20 +01004870 if (READ_ONCE(req->poll.canceled)) {
Jens Axboe45ab03b2021-02-23 08:19:33 -07004871 error = -ECANCELED;
Jens Axboe88e41cf2021-02-22 22:08:01 -07004872 req->poll.events |= EPOLLONESHOT;
Pavel Begunkove27414b2021-04-09 09:13:20 +01004873 } else {
Jens Axboe50826202021-02-23 09:02:26 -07004874 error = mangle_poll(mask);
Pavel Begunkove27414b2021-04-09 09:13:20 +01004875 }
Jens Axboeb69de282021-03-17 08:37:41 -06004876 if (req->poll.events & EPOLLONESHOT)
4877 flags = 0;
Pavel Begunkovff6421642021-04-11 01:46:32 +01004878 if (!io_cqring_fill_event(req, error, flags)) {
Jens Axboe50826202021-02-23 09:02:26 -07004879 io_poll_remove_waitqs(req);
Jens Axboe88e41cf2021-02-22 22:08:01 -07004880 req->poll.done = true;
4881 flags = 0;
4882 }
Jens Axboe18bceab2020-05-15 11:56:54 -06004883 io_commit_cqring(ctx);
Jens Axboe88e41cf2021-02-22 22:08:01 -07004884 return !(flags & IORING_CQE_F_MORE);
Jens Axboe18bceab2020-05-15 11:56:54 -06004885}
4886
Jens Axboe18bceab2020-05-15 11:56:54 -06004887static void io_poll_task_func(struct callback_head *cb)
4888{
4889 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
Jens Axboe6d816e02020-08-11 08:04:14 -06004890 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004891 struct io_kiocb *nxt;
Jens Axboe18bceab2020-05-15 11:56:54 -06004892
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004893 if (io_poll_rewait(req, &req->poll)) {
4894 spin_unlock_irq(&ctx->completion_lock);
4895 } else {
Pavel Begunkovf40b9642021-04-09 09:13:19 +01004896 bool done;
Jens Axboe88e41cf2021-02-22 22:08:01 -07004897
Pavel Begunkove27414b2021-04-09 09:13:20 +01004898 done = io_poll_complete(req, req->result);
Jens Axboe88e41cf2021-02-22 22:08:01 -07004899 if (done) {
4900 hash_del(&req->hash_node);
Pavel Begunkovf40b9642021-04-09 09:13:19 +01004901 } else {
Jens Axboe88e41cf2021-02-22 22:08:01 -07004902 req->result = 0;
4903 add_wait_queue(req->poll.head, &req->poll.wait);
4904 }
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004905 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkovf40b9642021-04-09 09:13:19 +01004906 io_cqring_ev_posted(ctx);
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004907
Jens Axboe88e41cf2021-02-22 22:08:01 -07004908 if (done) {
4909 nxt = io_put_req_find_next(req);
4910 if (nxt)
4911 __io_req_task_submit(nxt);
4912 }
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004913 }
Jens Axboe18bceab2020-05-15 11:56:54 -06004914}
4915
4916static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode,
4917 int sync, void *key)
4918{
4919 struct io_kiocb *req = wait->private;
Jens Axboed4e7cd32020-08-15 11:44:50 -07004920 struct io_poll_iocb *poll = io_poll_get_single(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004921 __poll_t mask = key_to_poll(key);
4922
4923 /* for instances that support it check for an event match first: */
4924 if (mask && !(mask & poll->events))
4925 return 0;
Jens Axboe88e41cf2021-02-22 22:08:01 -07004926 if (!(poll->events & EPOLLONESHOT))
4927 return poll->wait.func(&poll->wait, mode, sync, key);
Jens Axboe18bceab2020-05-15 11:56:54 -06004928
Jens Axboe8706e042020-09-28 08:38:54 -06004929 list_del_init(&wait->entry);
4930
Jens Axboe807abcb2020-07-17 17:09:27 -06004931 if (poll && poll->head) {
Jens Axboe18bceab2020-05-15 11:56:54 -06004932 bool done;
4933
Jens Axboe807abcb2020-07-17 17:09:27 -06004934 spin_lock(&poll->head->lock);
4935 done = list_empty(&poll->wait.entry);
Jens Axboe18bceab2020-05-15 11:56:54 -06004936 if (!done)
Jens Axboe807abcb2020-07-17 17:09:27 -06004937 list_del_init(&poll->wait.entry);
Jens Axboed4e7cd32020-08-15 11:44:50 -07004938 /* make sure double remove sees this as being gone */
4939 wait->private = NULL;
Jens Axboe807abcb2020-07-17 17:09:27 -06004940 spin_unlock(&poll->head->lock);
Jens Axboec8b5e262020-10-25 13:53:26 -06004941 if (!done) {
4942 /* use wait func handler, so it matches the rq type */
4943 poll->wait.func(&poll->wait, mode, sync, key);
4944 }
Jens Axboe18bceab2020-05-15 11:56:54 -06004945 }
Jens Axboede9b4cc2021-02-24 13:28:27 -07004946 req_ref_put(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004947 return 1;
4948}
4949
4950static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events,
4951 wait_queue_func_t wake_func)
4952{
4953 poll->head = NULL;
4954 poll->done = false;
4955 poll->canceled = false;
Jens Axboe464dca62021-03-19 14:06:24 -06004956#define IO_POLL_UNMASK (EPOLLERR|EPOLLHUP|EPOLLNVAL|EPOLLRDHUP)
4957 /* mask in events that we always want/need */
4958 poll->events = events | IO_POLL_UNMASK;
Jens Axboe18bceab2020-05-15 11:56:54 -06004959 INIT_LIST_HEAD(&poll->wait.entry);
4960 init_waitqueue_func_entry(&poll->wait, wake_func);
4961}
4962
4963static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt,
Jens Axboe807abcb2020-07-17 17:09:27 -06004964 struct wait_queue_head *head,
4965 struct io_poll_iocb **poll_ptr)
Jens Axboe18bceab2020-05-15 11:56:54 -06004966{
4967 struct io_kiocb *req = pt->req;
4968
4969 /*
4970 * If poll->head is already set, it's because the file being polled
4971 * uses multiple waitqueues for poll handling (eg one for read, one
4972 * for write). Setup a separate io_poll_iocb if this happens.
4973 */
4974 if (unlikely(poll->head)) {
Pavel Begunkov58852d42020-10-16 20:55:56 +01004975 struct io_poll_iocb *poll_one = poll;
4976
Jens Axboe18bceab2020-05-15 11:56:54 -06004977 /* already have a 2nd entry, fail a third attempt */
Jens Axboe807abcb2020-07-17 17:09:27 -06004978 if (*poll_ptr) {
Jens Axboe18bceab2020-05-15 11:56:54 -06004979 pt->error = -EINVAL;
4980 return;
4981 }
Jens Axboeea6a693d2021-04-15 09:47:13 -06004982 /*
4983 * Can't handle multishot for double wait for now, turn it
4984 * into one-shot mode.
4985 */
4986 if (!(req->poll.events & EPOLLONESHOT))
4987 req->poll.events |= EPOLLONESHOT;
Jens Axboe1c3b3e62021-02-28 16:07:30 -07004988 /* double add on the same waitqueue head, ignore */
4989 if (poll->head == head)
4990 return;
Jens Axboe18bceab2020-05-15 11:56:54 -06004991 poll = kmalloc(sizeof(*poll), GFP_ATOMIC);
4992 if (!poll) {
4993 pt->error = -ENOMEM;
4994 return;
4995 }
Pavel Begunkov58852d42020-10-16 20:55:56 +01004996 io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake);
Jens Axboede9b4cc2021-02-24 13:28:27 -07004997 req_ref_get(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004998 poll->wait.private = req;
Jens Axboe807abcb2020-07-17 17:09:27 -06004999 *poll_ptr = poll;
Jens Axboe18bceab2020-05-15 11:56:54 -06005000 }
5001
5002 pt->error = 0;
5003 poll->head = head;
Jiufei Xuea31eb4a2020-06-17 17:53:56 +08005004
5005 if (poll->events & EPOLLEXCLUSIVE)
5006 add_wait_queue_exclusive(head, &poll->wait);
5007 else
5008 add_wait_queue(head, &poll->wait);
Jens Axboe18bceab2020-05-15 11:56:54 -06005009}
5010
5011static void io_async_queue_proc(struct file *file, struct wait_queue_head *head,
5012 struct poll_table_struct *p)
5013{
5014 struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
Jens Axboe807abcb2020-07-17 17:09:27 -06005015 struct async_poll *apoll = pt->req->apoll;
Jens Axboe18bceab2020-05-15 11:56:54 -06005016
Jens Axboe807abcb2020-07-17 17:09:27 -06005017 __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll);
Jens Axboe18bceab2020-05-15 11:56:54 -06005018}
5019
Jens Axboed7718a92020-02-14 22:23:12 -07005020static void io_async_task_func(struct callback_head *cb)
5021{
5022 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
5023 struct async_poll *apoll = req->apoll;
5024 struct io_ring_ctx *ctx = req->ctx;
5025
5026 trace_io_uring_task_run(req->ctx, req->opcode, req->user_data);
5027
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005028 if (io_poll_rewait(req, &apoll->poll)) {
Jens Axboed7718a92020-02-14 22:23:12 -07005029 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005030 return;
Jens Axboed7718a92020-02-14 22:23:12 -07005031 }
5032
Pavel Begunkov0ea13b42021-04-09 09:13:21 +01005033 hash_del(&req->hash_node);
Jens Axboed4e7cd32020-08-15 11:44:50 -07005034 io_poll_remove_double(req);
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005035 spin_unlock_irq(&ctx->completion_lock);
5036
Pavel Begunkov0be0b0e2020-06-30 15:20:42 +03005037 if (!READ_ONCE(apoll->poll.canceled))
5038 __io_req_task_submit(req);
5039 else
Pavel Begunkov25935532021-03-19 17:22:40 +00005040 io_req_complete_failed(req, -ECANCELED);
Jens Axboed7718a92020-02-14 22:23:12 -07005041}
5042
5043static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
5044 void *key)
5045{
5046 struct io_kiocb *req = wait->private;
5047 struct io_poll_iocb *poll = &req->apoll->poll;
5048
5049 trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data,
5050 key_to_poll(key));
5051
5052 return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func);
5053}
5054
5055static void io_poll_req_insert(struct io_kiocb *req)
5056{
5057 struct io_ring_ctx *ctx = req->ctx;
5058 struct hlist_head *list;
5059
5060 list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)];
5061 hlist_add_head(&req->hash_node, list);
5062}
5063
5064static __poll_t __io_arm_poll_handler(struct io_kiocb *req,
5065 struct io_poll_iocb *poll,
5066 struct io_poll_table *ipt, __poll_t mask,
5067 wait_queue_func_t wake_func)
5068 __acquires(&ctx->completion_lock)
5069{
5070 struct io_ring_ctx *ctx = req->ctx;
5071 bool cancel = false;
5072
Pavel Begunkov4d52f332020-10-18 10:17:43 +01005073 INIT_HLIST_NODE(&req->hash_node);
Jens Axboe18bceab2020-05-15 11:56:54 -06005074 io_init_poll_iocb(poll, mask, wake_func);
Pavel Begunkovb90cd192020-06-21 13:09:52 +03005075 poll->file = req->file;
Jens Axboe18bceab2020-05-15 11:56:54 -06005076 poll->wait.private = req;
Jens Axboed7718a92020-02-14 22:23:12 -07005077
5078 ipt->pt._key = mask;
5079 ipt->req = req;
5080 ipt->error = -EINVAL;
5081
Jens Axboed7718a92020-02-14 22:23:12 -07005082 mask = vfs_poll(req->file, &ipt->pt) & poll->events;
5083
5084 spin_lock_irq(&ctx->completion_lock);
5085 if (likely(poll->head)) {
5086 spin_lock(&poll->head->lock);
5087 if (unlikely(list_empty(&poll->wait.entry))) {
5088 if (ipt->error)
5089 cancel = true;
5090 ipt->error = 0;
5091 mask = 0;
5092 }
Jens Axboe88e41cf2021-02-22 22:08:01 -07005093 if ((mask && (poll->events & EPOLLONESHOT)) || ipt->error)
Jens Axboed7718a92020-02-14 22:23:12 -07005094 list_del_init(&poll->wait.entry);
5095 else if (cancel)
5096 WRITE_ONCE(poll->canceled, true);
5097 else if (!poll->done) /* actually waiting for an event */
5098 io_poll_req_insert(req);
5099 spin_unlock(&poll->head->lock);
5100 }
5101
5102 return mask;
5103}
5104
5105static bool io_arm_poll_handler(struct io_kiocb *req)
5106{
5107 const struct io_op_def *def = &io_op_defs[req->opcode];
5108 struct io_ring_ctx *ctx = req->ctx;
5109 struct async_poll *apoll;
5110 struct io_poll_table ipt;
5111 __poll_t mask, ret;
Jens Axboe9dab14b2020-08-25 12:27:50 -06005112 int rw;
Jens Axboed7718a92020-02-14 22:23:12 -07005113
5114 if (!req->file || !file_can_poll(req->file))
5115 return false;
Pavel Begunkov24c74672020-06-21 13:09:51 +03005116 if (req->flags & REQ_F_POLLED)
Jens Axboed7718a92020-02-14 22:23:12 -07005117 return false;
Jens Axboe9dab14b2020-08-25 12:27:50 -06005118 if (def->pollin)
5119 rw = READ;
5120 else if (def->pollout)
5121 rw = WRITE;
5122 else
5123 return false;
5124 /* if we can't nonblock try, then no point in arming a poll handler */
Jens Axboe7b29f922021-03-12 08:30:14 -07005125 if (!io_file_supports_async(req, rw))
Jens Axboed7718a92020-02-14 22:23:12 -07005126 return false;
5127
5128 apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC);
5129 if (unlikely(!apoll))
5130 return false;
Jens Axboe807abcb2020-07-17 17:09:27 -06005131 apoll->double_poll = NULL;
Jens Axboed7718a92020-02-14 22:23:12 -07005132
5133 req->flags |= REQ_F_POLLED;
Jens Axboed7718a92020-02-14 22:23:12 -07005134 req->apoll = apoll;
Jens Axboed7718a92020-02-14 22:23:12 -07005135
Jens Axboe88e41cf2021-02-22 22:08:01 -07005136 mask = EPOLLONESHOT;
Jens Axboed7718a92020-02-14 22:23:12 -07005137 if (def->pollin)
Nathan Chancellor8755d972020-03-02 16:01:19 -07005138 mask |= POLLIN | POLLRDNORM;
Jens Axboed7718a92020-02-14 22:23:12 -07005139 if (def->pollout)
5140 mask |= POLLOUT | POLLWRNORM;
Luke Hsiao901341b2020-08-21 21:41:05 -07005141
5142 /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */
5143 if ((req->opcode == IORING_OP_RECVMSG) &&
5144 (req->sr_msg.msg_flags & MSG_ERRQUEUE))
5145 mask &= ~POLLIN;
5146
Jens Axboed7718a92020-02-14 22:23:12 -07005147 mask |= POLLERR | POLLPRI;
5148
5149 ipt.pt._qproc = io_async_queue_proc;
5150
5151 ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask,
5152 io_async_wake);
Jens Axboea36da652020-08-11 09:50:19 -06005153 if (ret || ipt.error) {
Jens Axboed4e7cd32020-08-15 11:44:50 -07005154 io_poll_remove_double(req);
Jens Axboed7718a92020-02-14 22:23:12 -07005155 spin_unlock_irq(&ctx->completion_lock);
Jens Axboed7718a92020-02-14 22:23:12 -07005156 return false;
5157 }
5158 spin_unlock_irq(&ctx->completion_lock);
5159 trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask,
5160 apoll->poll.events);
5161 return true;
5162}
5163
5164static bool __io_poll_remove_one(struct io_kiocb *req,
Jens Axboeb2e720a2021-03-31 09:03:03 -06005165 struct io_poll_iocb *poll, bool do_cancel)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005166 __must_hold(&req->ctx->completion_lock)
Jens Axboed7718a92020-02-14 22:23:12 -07005167{
Jens Axboeb41e9852020-02-17 09:52:41 -07005168 bool do_complete = false;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005169
Jens Axboe50826202021-02-23 09:02:26 -07005170 if (!poll->head)
5171 return false;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005172 spin_lock(&poll->head->lock);
Jens Axboeb2e720a2021-03-31 09:03:03 -06005173 if (do_cancel)
5174 WRITE_ONCE(poll->canceled, true);
Jens Axboe392edb42019-12-09 17:52:20 -07005175 if (!list_empty(&poll->wait.entry)) {
5176 list_del_init(&poll->wait.entry);
Jens Axboeb41e9852020-02-17 09:52:41 -07005177 do_complete = true;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005178 }
5179 spin_unlock(&poll->head->lock);
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005180 hash_del(&req->hash_node);
Jens Axboed7718a92020-02-14 22:23:12 -07005181 return do_complete;
5182}
5183
Jens Axboeb2c3f7e2021-02-23 08:58:04 -07005184static bool io_poll_remove_waitqs(struct io_kiocb *req)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005185 __must_hold(&req->ctx->completion_lock)
Jens Axboed7718a92020-02-14 22:23:12 -07005186{
5187 bool do_complete;
5188
Jens Axboed4e7cd32020-08-15 11:44:50 -07005189 io_poll_remove_double(req);
Pavel Begunkove31001a2021-04-13 02:58:43 +01005190 do_complete = __io_poll_remove_one(req, io_poll_get_single(req), true);
Jens Axboed4e7cd32020-08-15 11:44:50 -07005191
Pavel Begunkove31001a2021-04-13 02:58:43 +01005192 if (req->opcode != IORING_OP_POLL_ADD && do_complete) {
Jens Axboed7718a92020-02-14 22:23:12 -07005193 /* non-poll requests have submit ref still */
Pavel Begunkove31001a2021-04-13 02:58:43 +01005194 req_ref_put(req);
Xiaoguang Wangb1f573b2020-04-12 14:50:54 +08005195 }
Jens Axboeb2c3f7e2021-02-23 08:58:04 -07005196 return do_complete;
5197}
5198
5199static bool io_poll_remove_one(struct io_kiocb *req)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005200 __must_hold(&req->ctx->completion_lock)
Jens Axboeb2c3f7e2021-02-23 08:58:04 -07005201{
5202 bool do_complete;
5203
5204 do_complete = io_poll_remove_waitqs(req);
Jens Axboeb41e9852020-02-17 09:52:41 -07005205 if (do_complete) {
Pavel Begunkovff6421642021-04-11 01:46:32 +01005206 io_cqring_fill_event(req, -ECANCELED, 0);
Jens Axboeb41e9852020-02-17 09:52:41 -07005207 io_commit_cqring(req->ctx);
Jens Axboef254ac02020-08-12 17:33:30 -06005208 req_set_fail_links(req);
Pavel Begunkov216578e2020-10-13 09:44:00 +01005209 io_put_req_deferred(req, 1);
Jens Axboeb41e9852020-02-17 09:52:41 -07005210 }
5211
5212 return do_complete;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005213}
5214
Jens Axboe76e1b642020-09-26 15:05:03 -06005215/*
5216 * Returns true if we found and killed one or more poll requests
5217 */
Pavel Begunkov6b819282020-11-06 13:00:25 +00005218static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk,
5219 struct files_struct *files)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005220{
Jens Axboe78076bb2019-12-04 19:56:40 -07005221 struct hlist_node *tmp;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005222 struct io_kiocb *req;
Jens Axboe8e2e1fa2020-04-13 17:05:14 -06005223 int posted = 0, i;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005224
5225 spin_lock_irq(&ctx->completion_lock);
Jens Axboe78076bb2019-12-04 19:56:40 -07005226 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
5227 struct hlist_head *list;
5228
5229 list = &ctx->cancel_hash[i];
Jens Axboef3606e32020-09-22 08:18:24 -06005230 hlist_for_each_entry_safe(req, tmp, list, hash_node) {
Pavel Begunkov6b819282020-11-06 13:00:25 +00005231 if (io_match_task(req, tsk, files))
Jens Axboef3606e32020-09-22 08:18:24 -06005232 posted += io_poll_remove_one(req);
5233 }
Jens Axboe221c5eb2019-01-17 09:41:58 -07005234 }
5235 spin_unlock_irq(&ctx->completion_lock);
Jens Axboeb41e9852020-02-17 09:52:41 -07005236
Jens Axboe8e2e1fa2020-04-13 17:05:14 -06005237 if (posted)
5238 io_cqring_ev_posted(ctx);
Jens Axboe76e1b642020-09-26 15:05:03 -06005239
5240 return posted != 0;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005241}
5242
Pavel Begunkov9ba5fac2021-04-14 13:38:35 +01005243static struct io_kiocb *io_poll_find(struct io_ring_ctx *ctx, __u64 sqe_addr,
5244 bool poll_only)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005245 __must_hold(&ctx->completion_lock)
Jens Axboe47f46762019-11-09 17:43:02 -07005246{
Jens Axboe78076bb2019-12-04 19:56:40 -07005247 struct hlist_head *list;
Jens Axboe47f46762019-11-09 17:43:02 -07005248 struct io_kiocb *req;
5249
Jens Axboe78076bb2019-12-04 19:56:40 -07005250 list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)];
5251 hlist_for_each_entry(req, list, hash_node) {
Jens Axboeb41e9852020-02-17 09:52:41 -07005252 if (sqe_addr != req->user_data)
5253 continue;
Pavel Begunkov9ba5fac2021-04-14 13:38:35 +01005254 if (poll_only && req->opcode != IORING_OP_POLL_ADD)
5255 continue;
Jens Axboeb2cb8052021-03-17 08:17:19 -06005256 return req;
Jens Axboe47f46762019-11-09 17:43:02 -07005257 }
Jens Axboeb2cb8052021-03-17 08:17:19 -06005258 return NULL;
5259}
5260
Pavel Begunkov9ba5fac2021-04-14 13:38:35 +01005261static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr,
5262 bool poll_only)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005263 __must_hold(&ctx->completion_lock)
Jens Axboeb2cb8052021-03-17 08:17:19 -06005264{
5265 struct io_kiocb *req;
5266
Pavel Begunkov9ba5fac2021-04-14 13:38:35 +01005267 req = io_poll_find(ctx, sqe_addr, poll_only);
Jens Axboeb2cb8052021-03-17 08:17:19 -06005268 if (!req)
5269 return -ENOENT;
5270 if (io_poll_remove_one(req))
5271 return 0;
5272
5273 return -EALREADY;
Jens Axboe47f46762019-11-09 17:43:02 -07005274}
5275
Pavel Begunkov9096af32021-04-14 13:38:36 +01005276static __poll_t io_poll_parse_events(const struct io_uring_sqe *sqe,
5277 unsigned int flags)
5278{
5279 u32 events;
5280
5281 events = READ_ONCE(sqe->poll32_events);
5282#ifdef __BIG_ENDIAN
5283 events = swahw32(events);
5284#endif
5285 if (!(flags & IORING_POLL_ADD_MULTI))
5286 events |= EPOLLONESHOT;
5287 return demangle_poll(events) | (events & (EPOLLEXCLUSIVE|EPOLLONESHOT));
5288}
5289
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005290static int io_poll_update_prep(struct io_kiocb *req,
Jens Axboe3529d8c2019-12-19 18:24:38 -07005291 const struct io_uring_sqe *sqe)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005292{
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005293 struct io_poll_update *upd = &req->poll_update;
5294 u32 flags;
5295
Jens Axboe221c5eb2019-01-17 09:41:58 -07005296 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5297 return -EINVAL;
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005298 if (sqe->ioprio || sqe->buf_index)
5299 return -EINVAL;
5300 flags = READ_ONCE(sqe->len);
5301 if (flags & ~(IORING_POLL_UPDATE_EVENTS | IORING_POLL_UPDATE_USER_DATA |
5302 IORING_POLL_ADD_MULTI))
5303 return -EINVAL;
5304 /* meaningless without update */
5305 if (flags == IORING_POLL_ADD_MULTI)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005306 return -EINVAL;
5307
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005308 upd->old_user_data = READ_ONCE(sqe->addr);
5309 upd->update_events = flags & IORING_POLL_UPDATE_EVENTS;
5310 upd->update_user_data = flags & IORING_POLL_UPDATE_USER_DATA;
Jens Axboe0969e782019-12-17 18:40:57 -07005311
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005312 upd->new_user_data = READ_ONCE(sqe->off);
5313 if (!upd->update_user_data && upd->new_user_data)
5314 return -EINVAL;
5315 if (upd->update_events)
5316 upd->events = io_poll_parse_events(sqe, flags);
5317 else if (sqe->poll32_events)
5318 return -EINVAL;
Jens Axboe0969e782019-12-17 18:40:57 -07005319
Jens Axboe221c5eb2019-01-17 09:41:58 -07005320 return 0;
5321}
5322
Jens Axboe221c5eb2019-01-17 09:41:58 -07005323static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
5324 void *key)
5325{
Jens Axboec2f2eb72020-02-10 09:07:05 -07005326 struct io_kiocb *req = wait->private;
5327 struct io_poll_iocb *poll = &req->poll;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005328
Jens Axboed7718a92020-02-14 22:23:12 -07005329 return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005330}
5331
Jens Axboe221c5eb2019-01-17 09:41:58 -07005332static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head,
5333 struct poll_table_struct *p)
5334{
5335 struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
5336
Jens Axboee8c2bc12020-08-15 18:44:09 -07005337 __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data);
Jens Axboeeac406c62019-11-14 12:09:58 -07005338}
5339
Jens Axboe3529d8c2019-12-19 18:24:38 -07005340static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005341{
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005342 struct io_poll_iocb *poll = &req->poll;
5343 u32 flags;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005344
5345 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5346 return -EINVAL;
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005347 if (sqe->ioprio || sqe->buf_index || sqe->off || sqe->addr)
Jens Axboe88e41cf2021-02-22 22:08:01 -07005348 return -EINVAL;
5349 flags = READ_ONCE(sqe->len);
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005350 if (flags & ~IORING_POLL_ADD_MULTI)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005351 return -EINVAL;
Pavel Begunkov9096af32021-04-14 13:38:36 +01005352
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005353 poll->events = io_poll_parse_events(sqe, flags);
Jens Axboe0969e782019-12-17 18:40:57 -07005354 return 0;
5355}
5356
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005357static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe0969e782019-12-17 18:40:57 -07005358{
5359 struct io_poll_iocb *poll = &req->poll;
5360 struct io_ring_ctx *ctx = req->ctx;
5361 struct io_poll_table ipt;
Jens Axboe0969e782019-12-17 18:40:57 -07005362 __poll_t mask;
Jens Axboe0969e782019-12-17 18:40:57 -07005363
Jens Axboed7718a92020-02-14 22:23:12 -07005364 ipt.pt._qproc = io_poll_queue_proc;
Jens Axboe36703242019-07-25 10:20:18 -06005365
Jens Axboed7718a92020-02-14 22:23:12 -07005366 mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events,
5367 io_poll_wake);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005368
Jens Axboe8c838782019-03-12 15:48:16 -06005369 if (mask) { /* no async, we'd stolen it */
Jens Axboe8c838782019-03-12 15:48:16 -06005370 ipt.error = 0;
Pavel Begunkove27414b2021-04-09 09:13:20 +01005371 io_poll_complete(req, mask);
Jens Axboe8c838782019-03-12 15:48:16 -06005372 }
Jens Axboe221c5eb2019-01-17 09:41:58 -07005373 spin_unlock_irq(&ctx->completion_lock);
5374
Jens Axboe8c838782019-03-12 15:48:16 -06005375 if (mask) {
5376 io_cqring_ev_posted(ctx);
Jens Axboe88e41cf2021-02-22 22:08:01 -07005377 if (poll->events & EPOLLONESHOT)
5378 io_put_req(req);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005379 }
Jens Axboe8c838782019-03-12 15:48:16 -06005380 return ipt.error;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005381}
5382
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005383static int io_poll_update(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeb69de282021-03-17 08:37:41 -06005384{
5385 struct io_ring_ctx *ctx = req->ctx;
5386 struct io_kiocb *preq;
Jens Axboecb3b200e2021-04-06 09:49:31 -06005387 bool completing;
Jens Axboeb69de282021-03-17 08:37:41 -06005388 int ret;
5389
5390 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov9ba5fac2021-04-14 13:38:35 +01005391 preq = io_poll_find(ctx, req->poll_update.old_user_data, true);
Jens Axboeb69de282021-03-17 08:37:41 -06005392 if (!preq) {
5393 ret = -ENOENT;
5394 goto err;
Jens Axboeb69de282021-03-17 08:37:41 -06005395 }
Jens Axboecb3b200e2021-04-06 09:49:31 -06005396
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005397 if (!req->poll_update.update_events && !req->poll_update.update_user_data) {
5398 completing = true;
5399 ret = io_poll_remove_one(preq) ? 0 : -EALREADY;
5400 goto err;
5401 }
5402
Jens Axboecb3b200e2021-04-06 09:49:31 -06005403 /*
5404 * Don't allow racy completion with singleshot, as we cannot safely
5405 * update those. For multishot, if we're racing with completion, just
5406 * let completion re-add it.
5407 */
5408 completing = !__io_poll_remove_one(preq, &preq->poll, false);
5409 if (completing && (preq->poll.events & EPOLLONESHOT)) {
5410 ret = -EALREADY;
5411 goto err;
Jens Axboeb69de282021-03-17 08:37:41 -06005412 }
5413 /* we now have a detached poll request. reissue. */
5414 ret = 0;
5415err:
Jens Axboeb69de282021-03-17 08:37:41 -06005416 if (ret < 0) {
Jens Axboecb3b200e2021-04-06 09:49:31 -06005417 spin_unlock_irq(&ctx->completion_lock);
Jens Axboeb69de282021-03-17 08:37:41 -06005418 req_set_fail_links(req);
5419 io_req_complete(req, ret);
5420 return 0;
5421 }
5422 /* only mask one event flags, keep behavior flags */
Pavel Begunkov9d805892021-04-13 02:58:40 +01005423 if (req->poll_update.update_events) {
Jens Axboeb69de282021-03-17 08:37:41 -06005424 preq->poll.events &= ~0xffff;
Pavel Begunkov9d805892021-04-13 02:58:40 +01005425 preq->poll.events |= req->poll_update.events & 0xffff;
Jens Axboeb69de282021-03-17 08:37:41 -06005426 preq->poll.events |= IO_POLL_UNMASK;
5427 }
Pavel Begunkov9d805892021-04-13 02:58:40 +01005428 if (req->poll_update.update_user_data)
5429 preq->user_data = req->poll_update.new_user_data;
Jens Axboecb3b200e2021-04-06 09:49:31 -06005430 spin_unlock_irq(&ctx->completion_lock);
5431
Jens Axboeb69de282021-03-17 08:37:41 -06005432 /* complete update request, we're done with it */
5433 io_req_complete(req, ret);
5434
Jens Axboecb3b200e2021-04-06 09:49:31 -06005435 if (!completing) {
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005436 ret = io_poll_add(preq, issue_flags);
Jens Axboecb3b200e2021-04-06 09:49:31 -06005437 if (ret < 0) {
5438 req_set_fail_links(preq);
5439 io_req_complete(preq, ret);
5440 }
Jens Axboeb69de282021-03-17 08:37:41 -06005441 }
5442 return 0;
5443}
5444
Jens Axboe5262f562019-09-17 12:26:57 -06005445static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer)
5446{
Jens Axboead8a48a2019-11-15 08:49:11 -07005447 struct io_timeout_data *data = container_of(timer,
5448 struct io_timeout_data, timer);
5449 struct io_kiocb *req = data->req;
5450 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5262f562019-09-17 12:26:57 -06005451 unsigned long flags;
5452
Jens Axboe5262f562019-09-17 12:26:57 -06005453 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkova71976f2020-10-10 18:34:11 +01005454 list_del_init(&req->timeout.list);
Pavel Begunkov01cec8c2020-07-30 18:43:50 +03005455 atomic_set(&req->ctx->cq_timeouts,
5456 atomic_read(&req->ctx->cq_timeouts) + 1);
5457
Pavel Begunkovff6421642021-04-11 01:46:32 +01005458 io_cqring_fill_event(req, -ETIME, 0);
Jens Axboe5262f562019-09-17 12:26:57 -06005459 io_commit_cqring(ctx);
5460 spin_unlock_irqrestore(&ctx->completion_lock, flags);
5461
5462 io_cqring_ev_posted(ctx);
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005463 req_set_fail_links(req);
Jens Axboe5262f562019-09-17 12:26:57 -06005464 io_put_req(req);
5465 return HRTIMER_NORESTART;
5466}
5467
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005468static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx,
5469 __u64 user_data)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005470 __must_hold(&ctx->completion_lock)
Jens Axboe47f46762019-11-09 17:43:02 -07005471{
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005472 struct io_timeout_data *io;
Jens Axboef254ac02020-08-12 17:33:30 -06005473 struct io_kiocb *req;
Pavel Begunkovfd9c7bc2021-04-13 02:58:42 +01005474 bool found = false;
Jens Axboef254ac02020-08-12 17:33:30 -06005475
5476 list_for_each_entry(req, &ctx->timeout_list, timeout.list) {
Pavel Begunkovfd9c7bc2021-04-13 02:58:42 +01005477 found = user_data == req->user_data;
5478 if (found)
Jens Axboef254ac02020-08-12 17:33:30 -06005479 break;
Jens Axboef254ac02020-08-12 17:33:30 -06005480 }
Pavel Begunkovfd9c7bc2021-04-13 02:58:42 +01005481 if (!found)
5482 return ERR_PTR(-ENOENT);
Jens Axboef254ac02020-08-12 17:33:30 -06005483
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005484 io = req->async_data;
Pavel Begunkovfd9c7bc2021-04-13 02:58:42 +01005485 if (hrtimer_try_to_cancel(&io->timer) == -1)
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005486 return ERR_PTR(-EALREADY);
5487 list_del_init(&req->timeout.list);
5488 return req;
5489}
5490
5491static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005492 __must_hold(&ctx->completion_lock)
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005493{
5494 struct io_kiocb *req = io_timeout_extract(ctx, user_data);
5495
5496 if (IS_ERR(req))
5497 return PTR_ERR(req);
5498
5499 req_set_fail_links(req);
Pavel Begunkovff6421642021-04-11 01:46:32 +01005500 io_cqring_fill_event(req, -ECANCELED, 0);
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005501 io_put_req_deferred(req, 1);
5502 return 0;
Jens Axboef254ac02020-08-12 17:33:30 -06005503}
5504
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005505static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data,
5506 struct timespec64 *ts, enum hrtimer_mode mode)
Pavel Begunkove07785b2021-04-01 15:43:57 +01005507 __must_hold(&ctx->completion_lock)
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005508{
5509 struct io_kiocb *req = io_timeout_extract(ctx, user_data);
5510 struct io_timeout_data *data;
5511
5512 if (IS_ERR(req))
5513 return PTR_ERR(req);
5514
5515 req->timeout.off = 0; /* noseq */
5516 data = req->async_data;
5517 list_add_tail(&req->timeout.list, &ctx->timeout_list);
5518 hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode);
5519 data->timer.function = io_timeout_fn;
5520 hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode);
5521 return 0;
Jens Axboe47f46762019-11-09 17:43:02 -07005522}
5523
Jens Axboe3529d8c2019-12-19 18:24:38 -07005524static int io_timeout_remove_prep(struct io_kiocb *req,
5525 const struct io_uring_sqe *sqe)
Jens Axboeb29472e2019-12-17 18:50:29 -07005526{
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005527 struct io_timeout_rem *tr = &req->timeout_rem;
5528
Jens Axboeb29472e2019-12-17 18:50:29 -07005529 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5530 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005531 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5532 return -EINVAL;
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005533 if (sqe->ioprio || sqe->buf_index || sqe->len)
Jens Axboeb29472e2019-12-17 18:50:29 -07005534 return -EINVAL;
5535
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005536 tr->addr = READ_ONCE(sqe->addr);
5537 tr->flags = READ_ONCE(sqe->timeout_flags);
5538 if (tr->flags & IORING_TIMEOUT_UPDATE) {
5539 if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS))
5540 return -EINVAL;
5541 if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2)))
5542 return -EFAULT;
5543 } else if (tr->flags) {
5544 /* timeout removal doesn't support flags */
5545 return -EINVAL;
5546 }
5547
Jens Axboeb29472e2019-12-17 18:50:29 -07005548 return 0;
5549}
5550
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005551static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags)
5552{
5553 return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS
5554 : HRTIMER_MODE_REL;
5555}
5556
Jens Axboe11365042019-10-16 09:08:32 -06005557/*
5558 * Remove or update an existing timeout command
5559 */
Pavel Begunkov61e98202021-02-10 00:03:08 +00005560static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe11365042019-10-16 09:08:32 -06005561{
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005562 struct io_timeout_rem *tr = &req->timeout_rem;
Jens Axboe11365042019-10-16 09:08:32 -06005563 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe47f46762019-11-09 17:43:02 -07005564 int ret;
Jens Axboe11365042019-10-16 09:08:32 -06005565
Jens Axboe11365042019-10-16 09:08:32 -06005566 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005567 if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE))
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005568 ret = io_timeout_cancel(ctx, tr->addr);
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005569 else
5570 ret = io_timeout_update(ctx, tr->addr, &tr->ts,
5571 io_translate_timeout_mode(tr->flags));
Jens Axboe11365042019-10-16 09:08:32 -06005572
Pavel Begunkovff6421642021-04-11 01:46:32 +01005573 io_cqring_fill_event(req, ret, 0);
Jens Axboe11365042019-10-16 09:08:32 -06005574 io_commit_cqring(ctx);
5575 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe5262f562019-09-17 12:26:57 -06005576 io_cqring_ev_posted(ctx);
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005577 if (ret < 0)
5578 req_set_fail_links(req);
Jackie Liuec9c02a2019-11-08 23:50:36 +08005579 io_put_req(req);
Jens Axboe11365042019-10-16 09:08:32 -06005580 return 0;
Jens Axboe5262f562019-09-17 12:26:57 -06005581}
5582
Jens Axboe3529d8c2019-12-19 18:24:38 -07005583static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe,
Jens Axboe2d283902019-12-04 11:08:05 -07005584 bool is_timeout_link)
Jens Axboe5262f562019-09-17 12:26:57 -06005585{
Jens Axboead8a48a2019-11-15 08:49:11 -07005586 struct io_timeout_data *data;
Jens Axboea41525a2019-10-15 16:48:15 -06005587 unsigned flags;
Pavel Begunkov56080b02020-05-26 20:34:04 +03005588 u32 off = READ_ONCE(sqe->off);
Jens Axboe5262f562019-09-17 12:26:57 -06005589
Jens Axboead8a48a2019-11-15 08:49:11 -07005590 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe5262f562019-09-17 12:26:57 -06005591 return -EINVAL;
Jens Axboead8a48a2019-11-15 08:49:11 -07005592 if (sqe->ioprio || sqe->buf_index || sqe->len != 1)
Jens Axboea41525a2019-10-15 16:48:15 -06005593 return -EINVAL;
Pavel Begunkov56080b02020-05-26 20:34:04 +03005594 if (off && is_timeout_link)
Jens Axboe2d283902019-12-04 11:08:05 -07005595 return -EINVAL;
Jens Axboea41525a2019-10-15 16:48:15 -06005596 flags = READ_ONCE(sqe->timeout_flags);
5597 if (flags & ~IORING_TIMEOUT_ABS)
Jens Axboe5262f562019-09-17 12:26:57 -06005598 return -EINVAL;
Arnd Bergmannbdf20072019-10-01 09:53:29 -06005599
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005600 req->timeout.off = off;
Jens Axboe26a61672019-12-20 09:02:01 -07005601
Jens Axboee8c2bc12020-08-15 18:44:09 -07005602 if (!req->async_data && io_alloc_async_data(req))
Jens Axboe26a61672019-12-20 09:02:01 -07005603 return -ENOMEM;
5604
Jens Axboee8c2bc12020-08-15 18:44:09 -07005605 data = req->async_data;
Jens Axboead8a48a2019-11-15 08:49:11 -07005606 data->req = req;
Jens Axboead8a48a2019-11-15 08:49:11 -07005607
5608 if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr)))
Jens Axboe5262f562019-09-17 12:26:57 -06005609 return -EFAULT;
5610
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005611 data->mode = io_translate_timeout_mode(flags);
Jens Axboead8a48a2019-11-15 08:49:11 -07005612 hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode);
Pavel Begunkov2482b582021-03-25 18:32:44 +00005613 if (is_timeout_link)
5614 io_req_track_inflight(req);
Jens Axboead8a48a2019-11-15 08:49:11 -07005615 return 0;
5616}
5617
Pavel Begunkov61e98202021-02-10 00:03:08 +00005618static int io_timeout(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboead8a48a2019-11-15 08:49:11 -07005619{
Jens Axboead8a48a2019-11-15 08:49:11 -07005620 struct io_ring_ctx *ctx = req->ctx;
Jens Axboee8c2bc12020-08-15 18:44:09 -07005621 struct io_timeout_data *data = req->async_data;
Jens Axboead8a48a2019-11-15 08:49:11 -07005622 struct list_head *entry;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005623 u32 tail, off = req->timeout.off;
Jens Axboead8a48a2019-11-15 08:49:11 -07005624
Pavel Begunkov733f5c92020-05-26 20:34:03 +03005625 spin_lock_irq(&ctx->completion_lock);
Jens Axboe93bd25b2019-11-11 23:34:31 -07005626
Jens Axboe5262f562019-09-17 12:26:57 -06005627 /*
5628 * sqe->off holds how many events that need to occur for this
Jens Axboe93bd25b2019-11-11 23:34:31 -07005629 * timeout event to be satisfied. If it isn't set, then this is
5630 * a pure timeout request, sequence isn't used.
Jens Axboe5262f562019-09-17 12:26:57 -06005631 */
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03005632 if (io_is_timeout_noseq(req)) {
Jens Axboe93bd25b2019-11-11 23:34:31 -07005633 entry = ctx->timeout_list.prev;
5634 goto add;
5635 }
Jens Axboe5262f562019-09-17 12:26:57 -06005636
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005637 tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
5638 req->timeout.target_seq = tail + off;
Jens Axboe5262f562019-09-17 12:26:57 -06005639
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05005640 /* Update the last seq here in case io_flush_timeouts() hasn't.
5641 * This is safe because ->completion_lock is held, and submissions
5642 * and completions are never mixed in the same ->completion_lock section.
5643 */
5644 ctx->cq_last_tm_flush = tail;
5645
Jens Axboe5262f562019-09-17 12:26:57 -06005646 /*
5647 * Insertion sort, ensuring the first entry in the list is always
5648 * the one we need first.
5649 */
Jens Axboe5262f562019-09-17 12:26:57 -06005650 list_for_each_prev(entry, &ctx->timeout_list) {
Pavel Begunkov135fcde2020-07-13 23:37:12 +03005651 struct io_kiocb *nxt = list_entry(entry, struct io_kiocb,
5652 timeout.list);
Jens Axboe5262f562019-09-17 12:26:57 -06005653
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03005654 if (io_is_timeout_noseq(nxt))
Jens Axboe93bd25b2019-11-11 23:34:31 -07005655 continue;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005656 /* nxt.seq is behind @tail, otherwise would've been completed */
5657 if (off >= nxt->timeout.target_seq - tail)
Jens Axboe5262f562019-09-17 12:26:57 -06005658 break;
5659 }
Jens Axboe93bd25b2019-11-11 23:34:31 -07005660add:
Pavel Begunkov135fcde2020-07-13 23:37:12 +03005661 list_add(&req->timeout.list, entry);
Jens Axboead8a48a2019-11-15 08:49:11 -07005662 data->timer.function = io_timeout_fn;
5663 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode);
Jens Axboe842f9612019-10-29 12:34:10 -06005664 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe5262f562019-09-17 12:26:57 -06005665 return 0;
5666}
5667
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005668struct io_cancel_data {
5669 struct io_ring_ctx *ctx;
5670 u64 user_data;
5671};
5672
Jens Axboe62755e32019-10-28 21:49:21 -06005673static bool io_cancel_cb(struct io_wq_work *work, void *data)
Jens Axboede0617e2019-04-06 21:51:27 -06005674{
Jens Axboe62755e32019-10-28 21:49:21 -06005675 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005676 struct io_cancel_data *cd = data;
Jens Axboede0617e2019-04-06 21:51:27 -06005677
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005678 return req->ctx == cd->ctx && req->user_data == cd->user_data;
Jens Axboe62755e32019-10-28 21:49:21 -06005679}
5680
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005681static int io_async_cancel_one(struct io_uring_task *tctx, u64 user_data,
5682 struct io_ring_ctx *ctx)
Jens Axboe62755e32019-10-28 21:49:21 -06005683{
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005684 struct io_cancel_data data = { .ctx = ctx, .user_data = user_data, };
Jens Axboe62755e32019-10-28 21:49:21 -06005685 enum io_wq_cancel cancel_ret;
Jens Axboe62755e32019-10-28 21:49:21 -06005686 int ret = 0;
5687
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005688 if (!tctx || !tctx->io_wq)
Jens Axboe5aa75ed2021-02-16 12:56:50 -07005689 return -ENOENT;
5690
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005691 cancel_ret = io_wq_cancel_cb(tctx->io_wq, io_cancel_cb, &data, false);
Jens Axboe62755e32019-10-28 21:49:21 -06005692 switch (cancel_ret) {
5693 case IO_WQ_CANCEL_OK:
5694 ret = 0;
5695 break;
5696 case IO_WQ_CANCEL_RUNNING:
5697 ret = -EALREADY;
5698 break;
5699 case IO_WQ_CANCEL_NOTFOUND:
5700 ret = -ENOENT;
5701 break;
5702 }
5703
Jens Axboee977d6d2019-11-05 12:39:45 -07005704 return ret;
5705}
5706
Jens Axboe47f46762019-11-09 17:43:02 -07005707static void io_async_find_and_cancel(struct io_ring_ctx *ctx,
5708 struct io_kiocb *req, __u64 sqe_addr,
Pavel Begunkov014db002020-03-03 21:33:12 +03005709 int success_ret)
Jens Axboe47f46762019-11-09 17:43:02 -07005710{
5711 unsigned long flags;
5712 int ret;
5713
Pavel Begunkovf458dd842021-03-08 12:14:14 +00005714 ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx);
Jens Axboe47f46762019-11-09 17:43:02 -07005715 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkovdf9727a2021-04-01 15:43:59 +01005716 if (ret != -ENOENT)
5717 goto done;
Jens Axboe47f46762019-11-09 17:43:02 -07005718 ret = io_timeout_cancel(ctx, sqe_addr);
5719 if (ret != -ENOENT)
5720 goto done;
Pavel Begunkov9ba5fac2021-04-14 13:38:35 +01005721 ret = io_poll_cancel(ctx, sqe_addr, false);
Jens Axboe47f46762019-11-09 17:43:02 -07005722done:
Jens Axboeb0dd8a42019-11-18 12:14:54 -07005723 if (!ret)
5724 ret = success_ret;
Pavel Begunkovff6421642021-04-11 01:46:32 +01005725 io_cqring_fill_event(req, ret, 0);
Jens Axboe47f46762019-11-09 17:43:02 -07005726 io_commit_cqring(ctx);
5727 spin_unlock_irqrestore(&ctx->completion_lock, flags);
5728 io_cqring_ev_posted(ctx);
5729
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005730 if (ret < 0)
5731 req_set_fail_links(req);
Jens Axboe47f46762019-11-09 17:43:02 -07005732}
5733
Jens Axboe3529d8c2019-12-19 18:24:38 -07005734static int io_async_cancel_prep(struct io_kiocb *req,
5735 const struct io_uring_sqe *sqe)
Jens Axboee977d6d2019-11-05 12:39:45 -07005736{
Jens Axboefbf23842019-12-17 18:45:56 -07005737 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboee977d6d2019-11-05 12:39:45 -07005738 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005739 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5740 return -EINVAL;
5741 if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags)
Jens Axboee977d6d2019-11-05 12:39:45 -07005742 return -EINVAL;
5743
Jens Axboefbf23842019-12-17 18:45:56 -07005744 req->cancel.addr = READ_ONCE(sqe->addr);
5745 return 0;
5746}
5747
Pavel Begunkov61e98202021-02-10 00:03:08 +00005748static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboefbf23842019-12-17 18:45:56 -07005749{
5750 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkov58f99372021-03-12 16:25:55 +00005751 u64 sqe_addr = req->cancel.addr;
5752 struct io_tctx_node *node;
5753 int ret;
Jens Axboefbf23842019-12-17 18:45:56 -07005754
Pavel Begunkov58f99372021-03-12 16:25:55 +00005755 /* tasks should wait for their io-wq threads, so safe w/o sync */
5756 ret = io_async_cancel_one(req->task->io_uring, sqe_addr, ctx);
5757 spin_lock_irq(&ctx->completion_lock);
5758 if (ret != -ENOENT)
5759 goto done;
5760 ret = io_timeout_cancel(ctx, sqe_addr);
5761 if (ret != -ENOENT)
5762 goto done;
Pavel Begunkov9ba5fac2021-04-14 13:38:35 +01005763 ret = io_poll_cancel(ctx, sqe_addr, false);
Pavel Begunkov58f99372021-03-12 16:25:55 +00005764 if (ret != -ENOENT)
5765 goto done;
5766 spin_unlock_irq(&ctx->completion_lock);
5767
5768 /* slow path, try all io-wq's */
5769 io_ring_submit_lock(ctx, !(issue_flags & IO_URING_F_NONBLOCK));
5770 ret = -ENOENT;
5771 list_for_each_entry(node, &ctx->tctx_list, ctx_node) {
5772 struct io_uring_task *tctx = node->task->io_uring;
5773
Pavel Begunkov58f99372021-03-12 16:25:55 +00005774 ret = io_async_cancel_one(tctx, req->cancel.addr, ctx);
5775 if (ret != -ENOENT)
5776 break;
5777 }
5778 io_ring_submit_unlock(ctx, !(issue_flags & IO_URING_F_NONBLOCK));
5779
5780 spin_lock_irq(&ctx->completion_lock);
5781done:
Pavel Begunkovff6421642021-04-11 01:46:32 +01005782 io_cqring_fill_event(req, ret, 0);
Pavel Begunkov58f99372021-03-12 16:25:55 +00005783 io_commit_cqring(ctx);
5784 spin_unlock_irq(&ctx->completion_lock);
5785 io_cqring_ev_posted(ctx);
5786
5787 if (ret < 0)
5788 req_set_fail_links(req);
5789 io_put_req(req);
Jens Axboe62755e32019-10-28 21:49:21 -06005790 return 0;
5791}
5792
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005793static int io_rsrc_update_prep(struct io_kiocb *req,
Jens Axboe05f3fb32019-12-09 11:22:50 -07005794 const struct io_uring_sqe *sqe)
5795{
Jens Axboe6ca56f82020-09-18 16:51:19 -06005796 if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL))
5797 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005798 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5799 return -EINVAL;
5800 if (sqe->ioprio || sqe->rw_flags)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005801 return -EINVAL;
5802
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005803 req->rsrc_update.offset = READ_ONCE(sqe->off);
5804 req->rsrc_update.nr_args = READ_ONCE(sqe->len);
5805 if (!req->rsrc_update.nr_args)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005806 return -EINVAL;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005807 req->rsrc_update.arg = READ_ONCE(sqe->addr);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005808 return 0;
5809}
5810
Pavel Begunkov889fca72021-02-10 00:03:09 +00005811static int io_files_update(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005812{
5813 struct io_ring_ctx *ctx = req->ctx;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005814 struct io_uring_rsrc_update up;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005815 int ret;
5816
Pavel Begunkov45d189c2021-02-10 00:03:07 +00005817 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005818 return -EAGAIN;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005819
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005820 up.offset = req->rsrc_update.offset;
5821 up.data = req->rsrc_update.arg;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005822
5823 mutex_lock(&ctx->uring_lock);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005824 ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005825 mutex_unlock(&ctx->uring_lock);
5826
5827 if (ret < 0)
5828 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00005829 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005830 return 0;
5831}
5832
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005833static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07005834{
Jens Axboed625c6e2019-12-17 19:53:05 -07005835 switch (req->opcode) {
Jens Axboee7815732019-12-17 19:45:06 -07005836 case IORING_OP_NOP:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005837 return 0;
Jens Axboef67676d2019-12-02 11:03:47 -07005838 case IORING_OP_READV:
5839 case IORING_OP_READ_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005840 case IORING_OP_READ:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005841 return io_read_prep(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07005842 case IORING_OP_WRITEV:
5843 case IORING_OP_WRITE_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005844 case IORING_OP_WRITE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005845 return io_write_prep(req, sqe);
Jens Axboe0969e782019-12-17 18:40:57 -07005846 case IORING_OP_POLL_ADD:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005847 return io_poll_add_prep(req, sqe);
Jens Axboe0969e782019-12-17 18:40:57 -07005848 case IORING_OP_POLL_REMOVE:
Pavel Begunkovc5de0032021-04-14 13:38:37 +01005849 return io_poll_update_prep(req, sqe);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005850 case IORING_OP_FSYNC:
Pavel Begunkov1155c762021-02-18 18:29:38 +00005851 return io_fsync_prep(req, sqe);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005852 case IORING_OP_SYNC_FILE_RANGE:
Pavel Begunkov1155c762021-02-18 18:29:38 +00005853 return io_sfr_prep(req, sqe);
Jens Axboe03b12302019-12-02 18:50:25 -07005854 case IORING_OP_SENDMSG:
Jens Axboefddafac2020-01-04 20:19:44 -07005855 case IORING_OP_SEND:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005856 return io_sendmsg_prep(req, sqe);
Jens Axboe03b12302019-12-02 18:50:25 -07005857 case IORING_OP_RECVMSG:
Jens Axboefddafac2020-01-04 20:19:44 -07005858 case IORING_OP_RECV:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005859 return io_recvmsg_prep(req, sqe);
Jens Axboef499a022019-12-02 16:28:46 -07005860 case IORING_OP_CONNECT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005861 return io_connect_prep(req, sqe);
Jens Axboe2d283902019-12-04 11:08:05 -07005862 case IORING_OP_TIMEOUT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005863 return io_timeout_prep(req, sqe, false);
Jens Axboeb29472e2019-12-17 18:50:29 -07005864 case IORING_OP_TIMEOUT_REMOVE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005865 return io_timeout_remove_prep(req, sqe);
Jens Axboefbf23842019-12-17 18:45:56 -07005866 case IORING_OP_ASYNC_CANCEL:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005867 return io_async_cancel_prep(req, sqe);
Jens Axboe2d283902019-12-04 11:08:05 -07005868 case IORING_OP_LINK_TIMEOUT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005869 return io_timeout_prep(req, sqe, true);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005870 case IORING_OP_ACCEPT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005871 return io_accept_prep(req, sqe);
Jens Axboed63d1b52019-12-10 10:38:56 -07005872 case IORING_OP_FALLOCATE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005873 return io_fallocate_prep(req, sqe);
Jens Axboe15b71ab2019-12-11 11:20:36 -07005874 case IORING_OP_OPENAT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005875 return io_openat_prep(req, sqe);
Jens Axboeb5dba592019-12-11 14:02:38 -07005876 case IORING_OP_CLOSE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005877 return io_close_prep(req, sqe);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005878 case IORING_OP_FILES_UPDATE:
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00005879 return io_rsrc_update_prep(req, sqe);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07005880 case IORING_OP_STATX:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005881 return io_statx_prep(req, sqe);
Jens Axboe4840e412019-12-25 22:03:45 -07005882 case IORING_OP_FADVISE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005883 return io_fadvise_prep(req, sqe);
Jens Axboec1ca7572019-12-25 22:18:28 -07005884 case IORING_OP_MADVISE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005885 return io_madvise_prep(req, sqe);
Jens Axboecebdb982020-01-08 17:59:24 -07005886 case IORING_OP_OPENAT2:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005887 return io_openat2_prep(req, sqe);
Jens Axboe3e4827b2020-01-08 15:18:09 -07005888 case IORING_OP_EPOLL_CTL:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005889 return io_epoll_ctl_prep(req, sqe);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03005890 case IORING_OP_SPLICE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005891 return io_splice_prep(req, sqe);
Jens Axboeddf0322d2020-02-23 16:41:33 -07005892 case IORING_OP_PROVIDE_BUFFERS:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005893 return io_provide_buffers_prep(req, sqe);
Jens Axboe067524e2020-03-02 16:32:28 -07005894 case IORING_OP_REMOVE_BUFFERS:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005895 return io_remove_buffers_prep(req, sqe);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03005896 case IORING_OP_TEE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005897 return io_tee_prep(req, sqe);
Jens Axboe36f4fa62020-09-05 11:14:22 -06005898 case IORING_OP_SHUTDOWN:
5899 return io_shutdown_prep(req, sqe);
Jens Axboe80a261f2020-09-28 14:23:58 -06005900 case IORING_OP_RENAMEAT:
5901 return io_renameat_prep(req, sqe);
Jens Axboe14a11432020-09-28 14:27:37 -06005902 case IORING_OP_UNLINKAT:
5903 return io_unlinkat_prep(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07005904 }
5905
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005906 printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n",
5907 req->opcode);
5908 return-EINVAL;
5909}
5910
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005911static int io_req_prep_async(struct io_kiocb *req)
Jens Axboedef596e2019-01-09 08:59:42 -07005912{
Pavel Begunkovb7e298d2021-02-28 22:35:19 +00005913 if (!io_op_defs[req->opcode].needs_async_setup)
5914 return 0;
5915 if (WARN_ON_ONCE(req->async_data))
5916 return -EFAULT;
5917 if (io_alloc_async_data(req))
5918 return -EAGAIN;
5919
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005920 switch (req->opcode) {
5921 case IORING_OP_READV:
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005922 return io_rw_prep_async(req, READ);
5923 case IORING_OP_WRITEV:
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005924 return io_rw_prep_async(req, WRITE);
5925 case IORING_OP_SENDMSG:
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005926 return io_sendmsg_prep_async(req);
5927 case IORING_OP_RECVMSG:
Pavel Begunkov93642ef2021-02-18 18:29:44 +00005928 return io_recvmsg_prep_async(req);
5929 case IORING_OP_CONNECT:
5930 return io_connect_prep_async(req);
5931 }
Pavel Begunkovb7e298d2021-02-28 22:35:19 +00005932 printk_once(KERN_WARNING "io_uring: prep_async() bad opcode %d\n",
5933 req->opcode);
5934 return -EFAULT;
Jens Axboedef596e2019-01-09 08:59:42 -07005935}
5936
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005937static u32 io_get_sequence(struct io_kiocb *req)
5938{
5939 struct io_kiocb *pos;
5940 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00005941 u32 total_submitted, nr_reqs = 0;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005942
Pavel Begunkovf2f87372020-10-27 23:25:37 +00005943 io_for_each_link(pos, req)
5944 nr_reqs++;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005945
5946 total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped;
5947 return total_submitted - nr_reqs;
5948}
5949
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005950static int io_req_defer(struct io_kiocb *req)
Jens Axboedef596e2019-01-09 08:59:42 -07005951{
5952 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005953 struct io_defer_entry *de;
Jens Axboedef596e2019-01-09 08:59:42 -07005954 int ret;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005955 u32 seq;
Jens Axboedef596e2019-01-09 08:59:42 -07005956
5957 /* Still need defer if there is pending req in defer list. */
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005958 if (likely(list_empty_careful(&ctx->defer_list) &&
5959 !(req->flags & REQ_F_IO_DRAIN)))
5960 return 0;
5961
5962 seq = io_get_sequence(req);
5963 /* Still a chance to pass the sequence check */
5964 if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list))
Jens Axboedef596e2019-01-09 08:59:42 -07005965 return 0;
5966
Pavel Begunkovb7e298d2021-02-28 22:35:19 +00005967 ret = io_req_prep_async(req);
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00005968 if (ret)
5969 return ret;
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03005970 io_prep_async_link(req);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005971 de = kmalloc(sizeof(*de), GFP_KERNEL);
5972 if (!de)
5973 return -ENOMEM;
Jens Axboe31b51512019-01-18 22:56:34 -07005974
5975 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005976 if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) {
Jens Axboe31b51512019-01-18 22:56:34 -07005977 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005978 kfree(de);
Pavel Begunkovae348172020-07-23 20:25:20 +03005979 io_queue_async_work(req);
5980 return -EIOCBQUEUED;
Jens Axboe31b51512019-01-18 22:56:34 -07005981 }
5982
5983 trace_io_uring_defer(ctx, req, req->user_data);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005984 de->req = req;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005985 de->seq = seq;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005986 list_add_tail(&de->list, &ctx->defer_list);
Jens Axboe31b51512019-01-18 22:56:34 -07005987 spin_unlock_irq(&ctx->completion_lock);
5988 return -EIOCBQUEUED;
5989}
5990
Pavel Begunkov68fb8972021-03-19 17:22:41 +00005991static void io_clean_op(struct io_kiocb *req)
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03005992{
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005993 if (req->flags & REQ_F_BUFFER_SELECTED) {
5994 switch (req->opcode) {
5995 case IORING_OP_READV:
5996 case IORING_OP_READ_FIXED:
5997 case IORING_OP_READ:
Jens Axboebcda7ba2020-02-23 16:42:51 -07005998 kfree((void *)(unsigned long)req->rw.addr);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005999 break;
6000 case IORING_OP_RECVMSG:
6001 case IORING_OP_RECV:
Jens Axboe52de1fe2020-02-27 10:15:42 -07006002 kfree(req->sr_msg.kbuf);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006003 break;
6004 }
6005 req->flags &= ~REQ_F_BUFFER_SELECTED;
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03006006 }
6007
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006008 if (req->flags & REQ_F_NEED_CLEANUP) {
6009 switch (req->opcode) {
6010 case IORING_OP_READV:
6011 case IORING_OP_READ_FIXED:
6012 case IORING_OP_READ:
6013 case IORING_OP_WRITEV:
6014 case IORING_OP_WRITE_FIXED:
Jens Axboee8c2bc12020-08-15 18:44:09 -07006015 case IORING_OP_WRITE: {
6016 struct io_async_rw *io = req->async_data;
6017 if (io->free_iovec)
6018 kfree(io->free_iovec);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006019 break;
Jens Axboee8c2bc12020-08-15 18:44:09 -07006020 }
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006021 case IORING_OP_RECVMSG:
Jens Axboee8c2bc12020-08-15 18:44:09 -07006022 case IORING_OP_SENDMSG: {
6023 struct io_async_msghdr *io = req->async_data;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00006024
6025 kfree(io->free_iov);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006026 break;
Jens Axboee8c2bc12020-08-15 18:44:09 -07006027 }
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006028 case IORING_OP_SPLICE:
6029 case IORING_OP_TEE:
Pavel Begunkove1d767f2021-03-19 17:22:43 +00006030 if (!(req->splice.flags & SPLICE_F_FD_IN_FIXED))
6031 io_put_file(req->splice.file_in);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006032 break;
Jens Axboef3cd48502020-09-24 14:55:54 -06006033 case IORING_OP_OPENAT:
6034 case IORING_OP_OPENAT2:
6035 if (req->open.filename)
6036 putname(req->open.filename);
6037 break;
Jens Axboe80a261f2020-09-28 14:23:58 -06006038 case IORING_OP_RENAMEAT:
6039 putname(req->rename.oldpath);
6040 putname(req->rename.newpath);
6041 break;
Jens Axboe14a11432020-09-28 14:27:37 -06006042 case IORING_OP_UNLINKAT:
6043 putname(req->unlink.filename);
6044 break;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006045 }
6046 req->flags &= ~REQ_F_NEED_CLEANUP;
6047 }
Jens Axboe75652a302021-04-15 09:52:40 -06006048 if ((req->flags & REQ_F_POLLED) && req->apoll) {
6049 kfree(req->apoll->double_poll);
6050 kfree(req->apoll);
6051 req->apoll = NULL;
6052 }
Pavel Begunkov3a0a6902021-04-20 12:03:31 +01006053 if (req->flags & REQ_F_INFLIGHT) {
6054 struct io_uring_task *tctx = req->task->io_uring;
6055
6056 atomic_dec(&tctx->inflight_tracked);
6057 req->flags &= ~REQ_F_INFLIGHT;
6058 }
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03006059}
6060
Pavel Begunkov889fca72021-02-10 00:03:09 +00006061static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeedafcce2019-01-09 09:16:05 -07006062{
Jens Axboeedafcce2019-01-09 09:16:05 -07006063 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5730b272021-02-27 15:57:30 -07006064 const struct cred *creds = NULL;
Jens Axboed625c6e2019-12-17 19:53:05 -07006065 int ret;
Jens Axboeedafcce2019-01-09 09:16:05 -07006066
Jens Axboe003e8dc2021-03-06 09:22:27 -07006067 if (req->work.creds && req->work.creds != current_cred())
6068 creds = override_creds(req->work.creds);
Jens Axboe5730b272021-02-27 15:57:30 -07006069
Jens Axboed625c6e2019-12-17 19:53:05 -07006070 switch (req->opcode) {
Jens Axboe2b188cc2019-01-07 10:46:33 -07006071 case IORING_OP_NOP:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006072 ret = io_nop(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006073 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006074 case IORING_OP_READV:
Jens Axboe3529d8c2019-12-19 18:24:38 -07006075 case IORING_OP_READ_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07006076 case IORING_OP_READ:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006077 ret = io_read(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006078 break;
6079 case IORING_OP_WRITEV:
Jens Axboe2b188cc2019-01-07 10:46:33 -07006080 case IORING_OP_WRITE_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07006081 case IORING_OP_WRITE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006082 ret = io_write(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006083 break;
6084 case IORING_OP_FSYNC:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006085 ret = io_fsync(req, issue_flags);
Jackie Liuba5290c2019-10-09 09:19:59 +08006086 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006087 case IORING_OP_POLL_ADD:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006088 ret = io_poll_add(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006089 break;
6090 case IORING_OP_POLL_REMOVE:
Pavel Begunkovc5de0032021-04-14 13:38:37 +01006091 ret = io_poll_update(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006092 break;
Jens Axboeb76da702019-11-20 13:05:32 -07006093 case IORING_OP_SYNC_FILE_RANGE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006094 ret = io_sync_file_range(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07006095 break;
6096 case IORING_OP_SENDMSG:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006097 ret = io_sendmsg(req, issue_flags);
Pavel Begunkov062d04d2020-10-10 18:34:12 +01006098 break;
Jens Axboefddafac2020-01-04 20:19:44 -07006099 case IORING_OP_SEND:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006100 ret = io_send(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07006101 break;
6102 case IORING_OP_RECVMSG:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006103 ret = io_recvmsg(req, issue_flags);
Pavel Begunkov062d04d2020-10-10 18:34:12 +01006104 break;
Jens Axboefddafac2020-01-04 20:19:44 -07006105 case IORING_OP_RECV:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006106 ret = io_recv(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07006107 break;
Jens Axboe561fb042019-10-24 07:25:42 -06006108 case IORING_OP_TIMEOUT:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006109 ret = io_timeout(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006110 break;
6111 case IORING_OP_TIMEOUT_REMOVE:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006112 ret = io_timeout_remove(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006113 break;
6114 case IORING_OP_ACCEPT:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006115 ret = io_accept(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006116 break;
6117 case IORING_OP_CONNECT:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006118 ret = io_connect(req, issue_flags);
Jens Axboe31b51512019-01-18 22:56:34 -07006119 break;
6120 case IORING_OP_ASYNC_CANCEL:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006121 ret = io_async_cancel(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006122 break;
Jens Axboed63d1b52019-12-10 10:38:56 -07006123 case IORING_OP_FALLOCATE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006124 ret = io_fallocate(req, issue_flags);
Jens Axboed63d1b52019-12-10 10:38:56 -07006125 break;
Jens Axboe15b71ab2019-12-11 11:20:36 -07006126 case IORING_OP_OPENAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006127 ret = io_openat(req, issue_flags);
Jens Axboe15b71ab2019-12-11 11:20:36 -07006128 break;
Jens Axboeb5dba592019-12-11 14:02:38 -07006129 case IORING_OP_CLOSE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006130 ret = io_close(req, issue_flags);
Jens Axboeb5dba592019-12-11 14:02:38 -07006131 break;
Jens Axboe05f3fb32019-12-09 11:22:50 -07006132 case IORING_OP_FILES_UPDATE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006133 ret = io_files_update(req, issue_flags);
Jens Axboe05f3fb32019-12-09 11:22:50 -07006134 break;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07006135 case IORING_OP_STATX:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006136 ret = io_statx(req, issue_flags);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07006137 break;
Jens Axboe4840e412019-12-25 22:03:45 -07006138 case IORING_OP_FADVISE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006139 ret = io_fadvise(req, issue_flags);
Jens Axboe4840e412019-12-25 22:03:45 -07006140 break;
Jens Axboec1ca7572019-12-25 22:18:28 -07006141 case IORING_OP_MADVISE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006142 ret = io_madvise(req, issue_flags);
Jens Axboec1ca7572019-12-25 22:18:28 -07006143 break;
Jens Axboecebdb982020-01-08 17:59:24 -07006144 case IORING_OP_OPENAT2:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006145 ret = io_openat2(req, issue_flags);
Jens Axboecebdb982020-01-08 17:59:24 -07006146 break;
Jens Axboe3e4827b2020-01-08 15:18:09 -07006147 case IORING_OP_EPOLL_CTL:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006148 ret = io_epoll_ctl(req, issue_flags);
Jens Axboe3e4827b2020-01-08 15:18:09 -07006149 break;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03006150 case IORING_OP_SPLICE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006151 ret = io_splice(req, issue_flags);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03006152 break;
Jens Axboeddf0322d2020-02-23 16:41:33 -07006153 case IORING_OP_PROVIDE_BUFFERS:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006154 ret = io_provide_buffers(req, issue_flags);
Jens Axboeddf0322d2020-02-23 16:41:33 -07006155 break;
Jens Axboe067524e2020-03-02 16:32:28 -07006156 case IORING_OP_REMOVE_BUFFERS:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006157 ret = io_remove_buffers(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006158 break;
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03006159 case IORING_OP_TEE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006160 ret = io_tee(req, issue_flags);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03006161 break;
Jens Axboe36f4fa62020-09-05 11:14:22 -06006162 case IORING_OP_SHUTDOWN:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006163 ret = io_shutdown(req, issue_flags);
Jens Axboe36f4fa62020-09-05 11:14:22 -06006164 break;
Jens Axboe80a261f2020-09-28 14:23:58 -06006165 case IORING_OP_RENAMEAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006166 ret = io_renameat(req, issue_flags);
Jens Axboe80a261f2020-09-28 14:23:58 -06006167 break;
Jens Axboe14a11432020-09-28 14:27:37 -06006168 case IORING_OP_UNLINKAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006169 ret = io_unlinkat(req, issue_flags);
Jens Axboe14a11432020-09-28 14:27:37 -06006170 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006171 default:
6172 ret = -EINVAL;
6173 break;
6174 }
Jens Axboe31b51512019-01-18 22:56:34 -07006175
Jens Axboe5730b272021-02-27 15:57:30 -07006176 if (creds)
6177 revert_creds(creds);
6178
Jens Axboe2b188cc2019-01-07 10:46:33 -07006179 if (ret)
6180 return ret;
6181
Jens Axboeb5325762020-05-19 21:20:27 -06006182 /* If the op doesn't have a file, we're not polling for it */
6183 if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) {
Jens Axboe11ba8202020-01-15 21:51:17 -07006184 const bool in_async = io_wq_current_is_worker();
6185
Jens Axboe11ba8202020-01-15 21:51:17 -07006186 /* workqueue context doesn't hold uring_lock, grab it now */
6187 if (in_async)
6188 mutex_lock(&ctx->uring_lock);
6189
Xiaoguang Wang2e9dbe92020-11-13 00:44:08 +08006190 io_iopoll_req_issued(req, in_async);
Jens Axboe11ba8202020-01-15 21:51:17 -07006191
6192 if (in_async)
6193 mutex_unlock(&ctx->uring_lock);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006194 }
6195
6196 return 0;
6197}
6198
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00006199static void io_wq_submit_work(struct io_wq_work *work)
Pavel Begunkovd4c81f32020-06-08 21:08:19 +03006200{
Jens Axboe2b188cc2019-01-07 10:46:33 -07006201 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Pavel Begunkov6df1db62020-07-03 22:15:06 +03006202 struct io_kiocb *timeout;
Jens Axboe561fb042019-10-24 07:25:42 -06006203 int ret = 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006204
Pavel Begunkov6df1db62020-07-03 22:15:06 +03006205 timeout = io_prep_linked_timeout(req);
6206 if (timeout)
6207 io_queue_linked_timeout(timeout);
Pavel Begunkovd4c81f32020-06-08 21:08:19 +03006208
Jens Axboe4014d942021-01-19 15:53:54 -07006209 if (work->flags & IO_WQ_WORK_CANCEL)
Jens Axboe561fb042019-10-24 07:25:42 -06006210 ret = -ECANCELED;
Jens Axboe31b51512019-01-18 22:56:34 -07006211
Jens Axboe561fb042019-10-24 07:25:42 -06006212 if (!ret) {
Jens Axboe561fb042019-10-24 07:25:42 -06006213 do {
Pavel Begunkov889fca72021-02-10 00:03:09 +00006214 ret = io_issue_sqe(req, 0);
Jens Axboe561fb042019-10-24 07:25:42 -06006215 /*
6216 * We can get EAGAIN for polled IO even though we're
6217 * forcing a sync submission from here, since we can't
6218 * wait for request slots on the block side.
6219 */
6220 if (ret != -EAGAIN)
6221 break;
6222 cond_resched();
6223 } while (1);
6224 }
Jens Axboe31b51512019-01-18 22:56:34 -07006225
Pavel Begunkova3df76982021-02-18 22:32:52 +00006226 /* avoid locking problems by failing it from a clean context */
Jens Axboe561fb042019-10-24 07:25:42 -06006227 if (ret) {
Pavel Begunkova3df76982021-02-18 22:32:52 +00006228 /* io-wq is going to take one down */
Jens Axboede9b4cc2021-02-24 13:28:27 -07006229 req_ref_get(req);
Pavel Begunkova3df76982021-02-18 22:32:52 +00006230 io_req_task_queue_fail(req, ret);
Jens Axboeedafcce2019-01-09 09:16:05 -07006231 }
Jens Axboe31b51512019-01-18 22:56:34 -07006232}
Jens Axboe2b188cc2019-01-07 10:46:33 -07006233
Jens Axboe7b29f922021-03-12 08:30:14 -07006234#define FFS_ASYNC_READ 0x1UL
6235#define FFS_ASYNC_WRITE 0x2UL
6236#ifdef CONFIG_64BIT
6237#define FFS_ISREG 0x4UL
6238#else
6239#define FFS_ISREG 0x0UL
6240#endif
6241#define FFS_MASK ~(FFS_ASYNC_READ|FFS_ASYNC_WRITE|FFS_ISREG)
6242
Pavel Begunkovaeca2412021-04-11 01:46:37 +01006243static inline struct io_fixed_file *io_fixed_file_slot(struct io_file_table *table,
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01006244 unsigned i)
Jens Axboe09bb8392019-03-13 12:39:28 -06006245{
Pavel Begunkovaeca2412021-04-11 01:46:37 +01006246 struct io_fixed_file *table_l2;
Jens Axboe65e19f52019-10-26 07:20:21 -06006247
Pavel Begunkovaeca2412021-04-11 01:46:37 +01006248 table_l2 = table->files[i >> IORING_FILE_TABLE_SHIFT];
6249 return &table_l2[i & IORING_FILE_TABLE_MASK];
Pavel Begunkovdafecf12021-02-28 22:35:11 +00006250}
6251
6252static inline struct file *io_file_from_index(struct io_ring_ctx *ctx,
6253 int index)
6254{
Pavel Begunkovaeca2412021-04-11 01:46:37 +01006255 struct io_fixed_file *slot = io_fixed_file_slot(&ctx->file_table, index);
Jens Axboe7b29f922021-03-12 08:30:14 -07006256
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01006257 return (struct file *) (slot->file_ptr & FFS_MASK);
Jens Axboe65e19f52019-10-26 07:20:21 -06006258}
6259
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01006260static void io_fixed_file_set(struct io_fixed_file *file_slot, struct file *file)
Pavel Begunkov9a321c92021-04-01 15:44:01 +01006261{
6262 unsigned long file_ptr = (unsigned long) file;
6263
6264 if (__io_file_supports_async(file, READ))
6265 file_ptr |= FFS_ASYNC_READ;
6266 if (__io_file_supports_async(file, WRITE))
6267 file_ptr |= FFS_ASYNC_WRITE;
6268 if (S_ISREG(file_inode(file)->i_mode))
6269 file_ptr |= FFS_ISREG;
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01006270 file_slot->file_ptr = file_ptr;
Pavel Begunkov9a321c92021-04-01 15:44:01 +01006271}
6272
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006273static struct file *io_file_get(struct io_submit_state *state,
6274 struct io_kiocb *req, int fd, bool fixed)
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006275{
6276 struct io_ring_ctx *ctx = req->ctx;
6277 struct file *file;
6278
6279 if (fixed) {
Jens Axboe7b29f922021-03-12 08:30:14 -07006280 unsigned long file_ptr;
6281
Pavel Begunkov479f5172020-10-10 18:34:07 +01006282 if (unlikely((unsigned int)fd >= ctx->nr_user_files))
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006283 return NULL;
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006284 fd = array_index_nospec(fd, ctx->nr_user_files);
Pavel Begunkovaeca2412021-04-11 01:46:37 +01006285 file_ptr = io_fixed_file_slot(&ctx->file_table, fd)->file_ptr;
Jens Axboe7b29f922021-03-12 08:30:14 -07006286 file = (struct file *) (file_ptr & FFS_MASK);
6287 file_ptr &= ~FFS_MASK;
6288 /* mask in overlapping REQ_F and FFS bits */
6289 req->flags |= (file_ptr << REQ_F_ASYNC_READ_BIT);
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01006290 io_req_set_rsrc_node(req);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006291 } else {
6292 trace_io_uring_file_get(ctx, fd);
6293 file = __io_file_get(state, fd);
Jens Axboed44f5542021-03-12 08:27:05 -07006294
6295 /* we don't allow fixed io_uring files */
6296 if (file && unlikely(file->f_op == &io_uring_fops))
6297 io_req_track_inflight(req);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006298 }
6299
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006300 return file;
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006301}
6302
Jens Axboe2665abf2019-11-05 12:40:47 -07006303static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer)
6304{
Jens Axboead8a48a2019-11-15 08:49:11 -07006305 struct io_timeout_data *data = container_of(timer,
6306 struct io_timeout_data, timer);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006307 struct io_kiocb *prev, *req = data->req;
Jens Axboe2665abf2019-11-05 12:40:47 -07006308 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2665abf2019-11-05 12:40:47 -07006309 unsigned long flags;
Jens Axboe2665abf2019-11-05 12:40:47 -07006310
6311 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006312 prev = req->timeout.head;
6313 req->timeout.head = NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006314
6315 /*
6316 * We don't expect the list to be empty, that will only happen if we
6317 * race with the completion of the linked work.
6318 */
Jens Axboede9b4cc2021-02-24 13:28:27 -07006319 if (prev && req_ref_inc_not_zero(prev))
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006320 io_remove_next_linked(prev);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006321 else
6322 prev = NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006323 spin_unlock_irqrestore(&ctx->completion_lock, flags);
6324
6325 if (prev) {
Pavel Begunkov014db002020-03-03 21:33:12 +03006326 io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME);
Pavel Begunkov9ae1f8d2021-02-01 18:59:51 +00006327 io_put_req_deferred(prev, 1);
Jens Axboe47f46762019-11-09 17:43:02 -07006328 } else {
Pavel Begunkov9ae1f8d2021-02-01 18:59:51 +00006329 io_req_complete_post(req, -ETIME, 0);
Jens Axboe2665abf2019-11-05 12:40:47 -07006330 }
Pavel Begunkovdf9727a2021-04-01 15:43:59 +01006331 io_put_req_deferred(req, 1);
Jens Axboe2665abf2019-11-05 12:40:47 -07006332 return HRTIMER_NORESTART;
6333}
6334
Pavel Begunkovde968c12021-03-19 17:22:33 +00006335static void io_queue_linked_timeout(struct io_kiocb *req)
Jens Axboe2665abf2019-11-05 12:40:47 -07006336{
Pavel Begunkovde968c12021-03-19 17:22:33 +00006337 struct io_ring_ctx *ctx = req->ctx;
6338
6339 spin_lock_irq(&ctx->completion_lock);
Jens Axboe76a46e02019-11-10 23:34:16 -07006340 /*
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006341 * If the back reference is NULL, then our linked request finished
6342 * before we got a chance to setup the timer
Jens Axboe76a46e02019-11-10 23:34:16 -07006343 */
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006344 if (req->timeout.head) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07006345 struct io_timeout_data *data = req->async_data;
Jens Axboe94ae5e72019-11-14 19:39:52 -07006346
Jens Axboead8a48a2019-11-15 08:49:11 -07006347 data->timer.function = io_link_timeout_fn;
6348 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts),
6349 data->mode);
Jens Axboe2665abf2019-11-05 12:40:47 -07006350 }
Jens Axboe76a46e02019-11-10 23:34:16 -07006351 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe2665abf2019-11-05 12:40:47 -07006352 /* drop submission reference */
Jens Axboe76a46e02019-11-10 23:34:16 -07006353 io_put_req(req);
Jens Axboe2665abf2019-11-05 12:40:47 -07006354}
6355
Jens Axboead8a48a2019-11-15 08:49:11 -07006356static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req)
Jens Axboe2665abf2019-11-05 12:40:47 -07006357{
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006358 struct io_kiocb *nxt = req->link;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006359
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006360 if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) ||
6361 nxt->opcode != IORING_OP_LINK_TIMEOUT)
Jens Axboed7718a92020-02-14 22:23:12 -07006362 return NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006363
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006364 nxt->timeout.head = req;
Pavel Begunkov900fad42020-10-19 16:39:16 +01006365 nxt->flags |= REQ_F_LTIMEOUT_ACTIVE;
Jens Axboe76a46e02019-11-10 23:34:16 -07006366 req->flags |= REQ_F_LINK_TIMEOUT;
Jens Axboe76a46e02019-11-10 23:34:16 -07006367 return nxt;
Jens Axboe2665abf2019-11-05 12:40:47 -07006368}
6369
Pavel Begunkovc5eef2b942021-02-10 00:03:22 +00006370static void __io_queue_sqe(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07006371{
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006372 struct io_kiocb *linked_timeout = io_prep_linked_timeout(req);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006373 int ret;
6374
Pavel Begunkovc5eef2b942021-02-10 00:03:22 +00006375 ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER);
Jens Axboe491381ce2019-10-17 09:20:46 -06006376
6377 /*
6378 * We async punt it if the file wasn't marked NOWAIT, or if the file
6379 * doesn't support non-blocking read/write attempts
6380 */
Pavel Begunkov18400382021-03-19 17:22:34 +00006381 if (likely(!ret)) {
Pavel Begunkov0d63c142020-10-22 16:47:18 +01006382 /* drop submission reference */
Pavel Begunkove342c802021-01-19 13:32:47 +00006383 if (req->flags & REQ_F_COMPLETE_INLINE) {
Pavel Begunkovc5eef2b942021-02-10 00:03:22 +00006384 struct io_ring_ctx *ctx = req->ctx;
6385 struct io_comp_state *cs = &ctx->submit_state.comp;
Jens Axboee65ef562019-03-12 10:16:44 -06006386
Pavel Begunkov6dd0be12021-02-10 00:03:13 +00006387 cs->reqs[cs->nr++] = req;
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006388 if (cs->nr == ARRAY_SIZE(cs->reqs))
Pavel Begunkovc5eef2b942021-02-10 00:03:22 +00006389 io_submit_flush_completions(cs, ctx);
Pavel Begunkov9affd662021-01-19 13:32:46 +00006390 } else {
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006391 io_put_req(req);
Pavel Begunkov0d63c142020-10-22 16:47:18 +01006392 }
Pavel Begunkov18400382021-03-19 17:22:34 +00006393 } else if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) {
6394 if (!io_arm_poll_handler(req)) {
6395 /*
6396 * Queued up for async execution, worker will release
6397 * submit reference when the iocb is actually submitted.
6398 */
6399 io_queue_async_work(req);
6400 }
Pavel Begunkov0d63c142020-10-22 16:47:18 +01006401 } else {
Pavel Begunkovf41db2732021-02-28 22:35:12 +00006402 io_req_complete_failed(req, ret);
Jens Axboe9e645e112019-05-10 16:07:28 -06006403 }
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006404 if (linked_timeout)
6405 io_queue_linked_timeout(linked_timeout);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006406}
6407
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006408static void io_queue_sqe(struct io_kiocb *req)
Jackie Liu4fe2c962019-09-09 20:50:40 +08006409{
6410 int ret;
6411
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006412 ret = io_req_defer(req);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006413 if (ret) {
6414 if (ret != -EIOCBQUEUED) {
Pavel Begunkov11185912020-01-22 23:09:35 +03006415fail_req:
Pavel Begunkovf41db2732021-02-28 22:35:12 +00006416 io_req_complete_failed(req, ret);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006417 }
Pavel Begunkov25508782019-12-30 21:24:47 +03006418 } else if (req->flags & REQ_F_FORCE_ASYNC) {
Pavel Begunkovb7e298d2021-02-28 22:35:19 +00006419 ret = io_req_prep_async(req);
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006420 if (unlikely(ret))
6421 goto fail_req;
Jens Axboece35a472019-12-17 08:04:44 -07006422 io_queue_async_work(req);
6423 } else {
Pavel Begunkovc5eef2b942021-02-10 00:03:22 +00006424 __io_queue_sqe(req);
Jens Axboece35a472019-12-17 08:04:44 -07006425 }
Jackie Liu4fe2c962019-09-09 20:50:40 +08006426}
6427
Stefano Garzarella21b55db2020-08-27 16:58:30 +02006428/*
6429 * Check SQE restrictions (opcode and flags).
6430 *
6431 * Returns 'true' if SQE is allowed, 'false' otherwise.
6432 */
6433static inline bool io_check_restriction(struct io_ring_ctx *ctx,
6434 struct io_kiocb *req,
6435 unsigned int sqe_flags)
6436{
6437 if (!ctx->restricted)
6438 return true;
6439
6440 if (!test_bit(req->opcode, ctx->restrictions.sqe_op))
6441 return false;
6442
6443 if ((sqe_flags & ctx->restrictions.sqe_flags_required) !=
6444 ctx->restrictions.sqe_flags_required)
6445 return false;
6446
6447 if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed |
6448 ctx->restrictions.sqe_flags_required))
6449 return false;
6450
6451 return true;
6452}
6453
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006454static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req,
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006455 const struct io_uring_sqe *sqe)
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006456{
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006457 struct io_submit_state *state;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006458 unsigned int sqe_flags;
Jens Axboe003e8dc2021-03-06 09:22:27 -07006459 int personality, ret = 0;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006460
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006461 req->opcode = READ_ONCE(sqe->opcode);
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006462 /* same numerical values with corresponding REQ_F_*, safe to copy */
6463 req->flags = sqe_flags = READ_ONCE(sqe->flags);
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006464 req->user_data = READ_ONCE(sqe->user_data);
Jens Axboee8c2bc12020-08-15 18:44:09 -07006465 req->async_data = NULL;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006466 req->file = NULL;
6467 req->ctx = ctx;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006468 req->link = NULL;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006469 req->fixed_rsrc_refs = NULL;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006470 /* one is dropped after submission, the other at completion */
Jens Axboeabc54d62021-02-24 13:32:30 -07006471 atomic_set(&req->refs, 2);
Pavel Begunkov4dd28242020-06-15 10:33:13 +03006472 req->task = current;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006473 req->result = 0;
Jens Axboe93e68e02021-03-09 07:02:21 -07006474 req->work.creds = NULL;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006475
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006476 /* enforce forwards compatibility on users */
Pavel Begunkovebf4a5d2021-02-20 01:39:53 +00006477 if (unlikely(sqe_flags & ~SQE_VALID_FLAGS)) {
6478 req->flags = 0;
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006479 return -EINVAL;
Pavel Begunkovebf4a5d2021-02-20 01:39:53 +00006480 }
Pavel Begunkov5be9ad12021-02-12 18:41:17 +00006481
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006482 if (unlikely(req->opcode >= IORING_OP_LAST))
6483 return -EINVAL;
6484
Stefano Garzarella21b55db2020-08-27 16:58:30 +02006485 if (unlikely(!io_check_restriction(ctx, req, sqe_flags)))
6486 return -EACCES;
6487
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006488 if ((sqe_flags & IOSQE_BUFFER_SELECT) &&
6489 !io_op_defs[req->opcode].buffer_select)
6490 return -EOPNOTSUPP;
6491
Jens Axboe003e8dc2021-03-06 09:22:27 -07006492 personality = READ_ONCE(sqe->personality);
6493 if (personality) {
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00006494 req->work.creds = xa_load(&ctx->personalities, personality);
Jens Axboe003e8dc2021-03-06 09:22:27 -07006495 if (!req->work.creds)
6496 return -EINVAL;
6497 get_cred(req->work.creds);
Jens Axboe003e8dc2021-03-06 09:22:27 -07006498 }
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006499 state = &ctx->submit_state;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006500
Jens Axboe27926b62020-10-28 09:33:23 -06006501 /*
6502 * Plug now if we have more than 1 IO left after this, and the target
6503 * is potentially a read/write to block based storage.
6504 */
6505 if (!state->plug_started && state->ios_left > 1 &&
6506 io_op_defs[req->opcode].plug) {
6507 blk_start_plug(&state->plug);
6508 state->plug_started = true;
6509 }
Jens Axboe63ff8222020-05-07 14:56:15 -06006510
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006511 if (io_op_defs[req->opcode].needs_file) {
6512 bool fixed = req->flags & REQ_F_FIXED_FILE;
Jens Axboe63ff8222020-05-07 14:56:15 -06006513
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006514 req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed);
Pavel Begunkovba13e232021-02-01 18:59:52 +00006515 if (unlikely(!req->file))
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006516 ret = -EBADF;
6517 }
6518
Pavel Begunkov71b547c2020-10-10 18:34:09 +01006519 state->ios_left--;
6520 return ret;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006521}
6522
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006523static int io_submit_sqe(struct io_ring_ctx *ctx, struct io_kiocb *req,
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006524 const struct io_uring_sqe *sqe)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006525{
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006526 struct io_submit_link *link = &ctx->submit_state.link;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006527 int ret;
6528
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006529 ret = io_init_req(ctx, req, sqe);
6530 if (unlikely(ret)) {
6531fail_req:
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006532 if (link->head) {
6533 /* fail even hard links since we don't submit */
Pavel Begunkovcf109602021-02-18 18:29:43 +00006534 link->head->flags |= REQ_F_FAIL_LINK;
Pavel Begunkovf41db2732021-02-28 22:35:12 +00006535 io_req_complete_failed(link->head, -ECANCELED);
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006536 link->head = NULL;
6537 }
Pavel Begunkovf41db2732021-02-28 22:35:12 +00006538 io_req_complete_failed(req, ret);
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006539 return ret;
6540 }
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006541 ret = io_req_prep(req, sqe);
6542 if (unlikely(ret))
6543 goto fail_req;
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006544
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006545 /* don't need @sqe from now on */
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006546 trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data,
6547 true, ctx->flags & IORING_SETUP_SQPOLL);
6548
Jens Axboe6c271ce2019-01-10 11:22:30 -07006549 /*
6550 * If we already have a head request, queue this one for async
6551 * submittal once the head completes. If we don't have a head but
6552 * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be
6553 * submitted sync once the chain is complete. If none of those
6554 * conditions are true (normal request), then just queue it.
6555 */
6556 if (link->head) {
6557 struct io_kiocb *head = link->head;
6558
6559 /*
6560 * Taking sequential execution of a link, draining both sides
6561 * of the link also fullfils IOSQE_IO_DRAIN semantics for all
6562 * requests in the link. So, it drains the head and the
6563 * next after the link request. The last one is done via
6564 * drain_next flag to persist the effect across calls.
6565 */
6566 if (req->flags & REQ_F_IO_DRAIN) {
6567 head->flags |= REQ_F_IO_DRAIN;
6568 ctx->drain_next = 1;
6569 }
Pavel Begunkovb7e298d2021-02-28 22:35:19 +00006570 ret = io_req_prep_async(req);
Pavel Begunkovcf109602021-02-18 18:29:43 +00006571 if (unlikely(ret))
Pavel Begunkova6b8cadc2021-02-18 18:29:41 +00006572 goto fail_req;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006573 trace_io_uring_link(ctx, req, head);
6574 link->last->link = req;
6575 link->last = req;
6576
6577 /* last request of a link, enqueue the link */
6578 if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) {
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006579 io_queue_sqe(head);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006580 link->head = NULL;
6581 }
Jackie Liu4fe2c962019-09-09 20:50:40 +08006582 } else {
6583 if (unlikely(ctx->drain_next)) {
6584 req->flags |= REQ_F_IO_DRAIN;
6585 ctx->drain_next = 0;
6586 }
6587 if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) {
Jackie Liu4fe2c962019-09-09 20:50:40 +08006588 link->head = req;
6589 link->last = req;
6590 } else {
Pavel Begunkovbe7053b2021-02-18 18:29:45 +00006591 io_queue_sqe(req);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006592 }
6593 }
6594
6595 return 0;
6596}
6597
6598/*
6599 * Batched submission is done, ensure local IO is flushed out.
6600 */
6601static void io_submit_state_end(struct io_submit_state *state,
6602 struct io_ring_ctx *ctx)
Pavel Begunkov1b4a51b2019-11-21 11:54:28 +03006603{
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006604 if (state->link.head)
Pavel Begunkovde59bc12021-02-18 18:29:47 +00006605 io_queue_sqe(state->link.head);
Jens Axboe3529d8c2019-12-19 18:24:38 -07006606 if (state->comp.nr)
Jens Axboe9e645e112019-05-10 16:07:28 -06006607 io_submit_flush_completions(&state->comp, ctx);
Jackie Liua197f662019-11-08 08:09:12 -07006608 if (state->plug_started)
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006609 blk_finish_plug(&state->plug);
Jens Axboe75c6a032020-01-28 10:15:23 -07006610 io_state_file_put(state);
Jens Axboe9e645e112019-05-10 16:07:28 -06006611}
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006612
Jens Axboe9e645e112019-05-10 16:07:28 -06006613/*
6614 * Start submission side cache.
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006615 */
Jens Axboe9e645e112019-05-10 16:07:28 -06006616static void io_submit_state_start(struct io_submit_state *state,
Pavel Begunkov196be952019-11-07 01:41:06 +03006617 unsigned int max_ios)
Jens Axboe9e645e112019-05-10 16:07:28 -06006618{
6619 state->plug_started = false;
Jens Axboebcda7ba2020-02-23 16:42:51 -07006620 state->ios_left = max_ios;
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006621 /* set only head, no need to init link_last in advance */
6622 state->link.head = NULL;
Jens Axboe75c6a032020-01-28 10:15:23 -07006623}
6624
Jens Axboe193155c2020-02-22 23:22:19 -07006625static void io_commit_sqring(struct io_ring_ctx *ctx)
6626{
Jens Axboe75c6a032020-01-28 10:15:23 -07006627 struct io_rings *rings = ctx->rings;
6628
6629 /*
Jens Axboe193155c2020-02-22 23:22:19 -07006630 * Ensure any loads from the SQEs are done at this point,
Jens Axboe75c6a032020-01-28 10:15:23 -07006631 * since once we write the new head, the application could
6632 * write new data to them.
Pavel Begunkov6b47ee62020-01-18 20:22:41 +03006633 */
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006634 smp_store_release(&rings->sq.head, ctx->cached_sq_head);
Jens Axboebcda7ba2020-02-23 16:42:51 -07006635}
6636
Jens Axboe9e645e112019-05-10 16:07:28 -06006637/*
Jens Axboe3529d8c2019-12-19 18:24:38 -07006638 * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory
Jens Axboe9e645e112019-05-10 16:07:28 -06006639 * that is mapped by userspace. This means that care needs to be taken to
6640 * ensure that reads are stable, as we cannot rely on userspace always
Jens Axboe78e19bb2019-11-06 15:21:34 -07006641 * being a good citizen. If members of the sqe are validated and then later
6642 * used, it's important that those reads are done through READ_ONCE() to
Pavel Begunkov2e6e1fd2019-12-05 16:15:45 +03006643 * prevent a re-load down the line.
Jens Axboe9e645e112019-05-10 16:07:28 -06006644 */
6645static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx)
Jens Axboe9e645e112019-05-10 16:07:28 -06006646{
6647 u32 *sq_array = ctx->sq_array;
6648 unsigned head;
6649
6650 /*
6651 * The cached sq head (or cq tail) serves two purposes:
6652 *
6653 * 1) allows us to batch the cost of updating the user visible
Pavel Begunkov9d763772019-12-17 02:22:07 +03006654 * head updates.
Jens Axboe9e645e112019-05-10 16:07:28 -06006655 * 2) allows the kernel side to track the head on its own, even
Pavel Begunkov8cdf2192020-01-25 00:40:24 +03006656 * though the application is the one updating it.
6657 */
6658 head = READ_ONCE(sq_array[ctx->cached_sq_head++ & ctx->sq_mask]);
6659 if (likely(head < ctx->sq_entries))
6660 return &ctx->sq_sqes[head];
6661
6662 /* drop invalid entries */
Pavel Begunkov711be032020-01-17 03:57:59 +03006663 ctx->cached_sq_dropped++;
6664 WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped);
6665 return NULL;
6666}
Jens Axboeb7bb4f72019-12-15 22:13:43 -07006667
Jens Axboe0f212202020-09-13 13:09:39 -06006668static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006669{
Pavel Begunkov46c4e162021-02-18 18:29:37 +00006670 int submitted = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006671
Pavel Begunkovee7d46d92019-12-30 21:24:45 +03006672 /* make sure SQ entry isn't read before tail */
6673 nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx));
Pavel Begunkov9ef4f122019-12-30 21:24:44 +03006674
Pavel Begunkov2b85edf2019-12-28 14:13:03 +03006675 if (!percpu_ref_tryget_many(&ctx->refs, nr))
6676 return -EAGAIN;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006677
Jens Axboed8a6df12020-10-15 16:24:45 -06006678 percpu_counter_add(&current->io_uring->inflight, nr);
Jens Axboefaf7b512020-10-07 12:48:53 -06006679 refcount_add(nr, &current->usage);
Pavel Begunkovba88ff12021-02-10 00:03:11 +00006680 io_submit_state_start(&ctx->submit_state, nr);
Pavel Begunkovb14cca02020-01-17 04:45:59 +03006681
Pavel Begunkov46c4e162021-02-18 18:29:37 +00006682 while (submitted < nr) {
Jens Axboe3529d8c2019-12-19 18:24:38 -07006683 const struct io_uring_sqe *sqe;
Pavel Begunkov196be952019-11-07 01:41:06 +03006684 struct io_kiocb *req;
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03006685
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006686 req = io_alloc_req(ctx);
Pavel Begunkov196be952019-11-07 01:41:06 +03006687 if (unlikely(!req)) {
6688 if (!submitted)
6689 submitted = -EAGAIN;
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03006690 break;
Jens Axboe9e645e112019-05-10 16:07:28 -06006691 }
Pavel Begunkov4fccfcb2021-02-12 11:55:17 +00006692 sqe = io_get_sqe(ctx);
6693 if (unlikely(!sqe)) {
6694 kmem_cache_free(req_cachep, req);
6695 break;
6696 }
Jens Axboed3656342019-12-18 09:50:26 -07006697 /* will complete beyond this point, count as submitted */
6698 submitted++;
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006699 if (io_submit_sqe(ctx, req, sqe))
Jens Axboed3656342019-12-18 09:50:26 -07006700 break;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006701 }
6702
Pavel Begunkov9466f432020-01-25 22:34:01 +03006703 if (unlikely(submitted != nr)) {
6704 int ref_used = (submitted == -EAGAIN) ? 0 : submitted;
Jens Axboed8a6df12020-10-15 16:24:45 -06006705 struct io_uring_task *tctx = current->io_uring;
6706 int unused = nr - ref_used;
Pavel Begunkov9466f432020-01-25 22:34:01 +03006707
Jens Axboed8a6df12020-10-15 16:24:45 -06006708 percpu_ref_put_many(&ctx->refs, unused);
6709 percpu_counter_sub(&tctx->inflight, unused);
6710 put_task_struct_many(current, unused);
Pavel Begunkov9466f432020-01-25 22:34:01 +03006711 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07006712
Pavel Begunkova1ab7b32021-02-18 18:29:42 +00006713 io_submit_state_end(&ctx->submit_state, ctx);
Pavel Begunkovae9428c2019-11-06 00:22:14 +03006714 /* Commit SQ ring head once we've consumed and submitted all SQEs */
6715 io_commit_sqring(ctx);
6716
Jens Axboe6c271ce2019-01-10 11:22:30 -07006717 return submitted;
6718}
6719
Xiaoguang Wang23b36282020-07-23 20:57:24 +08006720static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx)
6721{
6722 /* Tell userspace we may need a wakeup call */
6723 spin_lock_irq(&ctx->completion_lock);
6724 ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP;
6725 spin_unlock_irq(&ctx->completion_lock);
6726}
6727
6728static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx)
6729{
6730 spin_lock_irq(&ctx->completion_lock);
6731 ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP;
6732 spin_unlock_irq(&ctx->completion_lock);
6733}
6734
Xiaoguang Wang08369242020-11-03 14:15:59 +08006735static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006736{
Jens Axboec8d1ba52020-09-14 11:07:26 -06006737 unsigned int to_submit;
Xiaoguang Wangbdcd3ea2020-02-25 22:12:08 +08006738 int ret = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006739
Jens Axboec8d1ba52020-09-14 11:07:26 -06006740 to_submit = io_sqring_entries(ctx);
Jens Axboee95eee22020-09-08 09:11:32 -06006741 /* if we're handling multiple rings, cap submit size for fairness */
6742 if (cap_entries && to_submit > 8)
6743 to_submit = 8;
6744
Xiaoguang Wang906a3c62020-11-12 14:56:00 +08006745 if (!list_empty(&ctx->iopoll_list) || to_submit) {
6746 unsigned nr_events = 0;
6747
Xiaoguang Wang08369242020-11-03 14:15:59 +08006748 mutex_lock(&ctx->uring_lock);
Xiaoguang Wang906a3c62020-11-12 14:56:00 +08006749 if (!list_empty(&ctx->iopoll_list))
6750 io_do_iopoll(ctx, &nr_events, 0);
6751
Pavel Begunkov3b763ba2021-04-18 14:52:08 +01006752 /*
6753 * Don't submit if refs are dying, good for io_uring_register(),
6754 * but also it is relied upon by io_ring_exit_work()
6755 */
Pavel Begunkov0298ef92021-03-08 13:20:57 +00006756 if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) &&
6757 !(ctx->flags & IORING_SETUP_R_DISABLED))
Xiaoguang Wang08369242020-11-03 14:15:59 +08006758 ret = io_submit_sqes(ctx, to_submit);
6759 mutex_unlock(&ctx->uring_lock);
6760 }
Jens Axboe90554202020-09-03 12:12:41 -06006761
6762 if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait))
6763 wake_up(&ctx->sqo_sq_wait);
6764
Xiaoguang Wang08369242020-11-03 14:15:59 +08006765 return ret;
6766}
6767
6768static void io_sqd_update_thread_idle(struct io_sq_data *sqd)
6769{
6770 struct io_ring_ctx *ctx;
6771 unsigned sq_thread_idle = 0;
6772
Pavel Begunkovc9dca272021-03-10 13:13:55 +00006773 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6774 sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006775 sqd->sq_thread_idle = sq_thread_idle;
Jens Axboec8d1ba52020-09-14 11:07:26 -06006776}
6777
Jens Axboe6c271ce2019-01-10 11:22:30 -07006778static int io_sq_thread(void *data)
6779{
Jens Axboe69fb2132020-09-14 11:16:23 -06006780 struct io_sq_data *sqd = data;
6781 struct io_ring_ctx *ctx;
Xiaoguang Wanga0d9205f2020-11-12 14:55:59 +08006782 unsigned long timeout = 0;
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006783 char buf[TASK_COMM_LEN];
Xiaoguang Wang08369242020-11-03 14:15:59 +08006784 DEFINE_WAIT(wait);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006785
Pavel Begunkov696ee882021-04-01 09:55:04 +01006786 snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006787 set_task_comm(current, buf);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006788 current->pf_io_worker = NULL;
Jens Axboe28cea78a2020-09-14 10:51:17 -06006789
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006790 if (sqd->sq_cpu != -1)
6791 set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu));
6792 else
6793 set_cpus_allowed_ptr(current, cpu_online_mask);
6794 current->flags |= PF_NO_SETAFFINITY;
6795
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00006796 mutex_lock(&sqd->lock);
Jens Axboe05962f92021-03-06 13:58:48 -07006797 while (!test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state)) {
Xiaoguang Wang08369242020-11-03 14:15:59 +08006798 int ret;
6799 bool cap_entries, sqt_spin, needs_sched;
Jens Axboec1edbf52019-11-10 16:56:04 -07006800
Jens Axboe82734c52021-03-29 06:52:44 -06006801 if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) ||
6802 signal_pending(current)) {
6803 bool did_sig = false;
6804
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00006805 mutex_unlock(&sqd->lock);
Jens Axboe82734c52021-03-29 06:52:44 -06006806 if (signal_pending(current)) {
6807 struct ksignal ksig;
6808
6809 did_sig = get_signal(&ksig);
6810 }
Jens Axboe05962f92021-03-06 13:58:48 -07006811 cond_resched();
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00006812 mutex_lock(&sqd->lock);
Jens Axboe82734c52021-03-29 06:52:44 -06006813 if (did_sig)
6814 break;
Pavel Begunkov521d6a72021-03-11 23:29:38 +00006815 io_run_task_work();
Pavel Begunkovb7f5a0b2021-03-15 14:23:08 +00006816 io_run_task_work_head(&sqd->park_task_work);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006817 timeout = jiffies + sqd->sq_thread_idle;
Pavel Begunkov7d41e852021-03-10 13:13:54 +00006818 continue;
Xiaoguang Wang08369242020-11-03 14:15:59 +08006819 }
Xiaoguang Wang08369242020-11-03 14:15:59 +08006820 sqt_spin = false;
Jens Axboee95eee22020-09-08 09:11:32 -06006821 cap_entries = !list_is_singular(&sqd->ctx_list);
Jens Axboe69fb2132020-09-14 11:16:23 -06006822 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
Stefan Metzmacher7c30f36a2021-03-07 11:54:28 +01006823 const struct cred *creds = NULL;
6824
6825 if (ctx->sq_creds != current_cred())
6826 creds = override_creds(ctx->sq_creds);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006827 ret = __io_sq_thread(ctx, cap_entries);
Stefan Metzmacher7c30f36a2021-03-07 11:54:28 +01006828 if (creds)
6829 revert_creds(creds);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006830 if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list)))
6831 sqt_spin = true;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006832 }
6833
Xiaoguang Wang08369242020-11-03 14:15:59 +08006834 if (sqt_spin || !time_after(jiffies, timeout)) {
Jens Axboec8d1ba52020-09-14 11:07:26 -06006835 io_run_task_work();
6836 cond_resched();
Xiaoguang Wang08369242020-11-03 14:15:59 +08006837 if (sqt_spin)
6838 timeout = jiffies + sqd->sq_thread_idle;
6839 continue;
6840 }
6841
Xiaoguang Wang08369242020-11-03 14:15:59 +08006842 needs_sched = true;
6843 prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE);
6844 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
6845 if ((ctx->flags & IORING_SETUP_IOPOLL) &&
6846 !list_empty_careful(&ctx->iopoll_list)) {
6847 needs_sched = false;
6848 break;
6849 }
6850 if (io_sqring_entries(ctx)) {
6851 needs_sched = false;
6852 break;
6853 }
6854 }
6855
Jens Axboe05962f92021-03-06 13:58:48 -07006856 if (needs_sched && !test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state)) {
Jens Axboe69fb2132020-09-14 11:16:23 -06006857 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6858 io_ring_set_wakeup_flag(ctx);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006859
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00006860 mutex_unlock(&sqd->lock);
Jens Axboe69fb2132020-09-14 11:16:23 -06006861 schedule();
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00006862 mutex_lock(&sqd->lock);
Jens Axboe69fb2132020-09-14 11:16:23 -06006863 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6864 io_ring_clear_wakeup_flag(ctx);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006865 }
Xiaoguang Wang08369242020-11-03 14:15:59 +08006866
6867 finish_wait(&sqd->wait, &wait);
Pavel Begunkovb7f5a0b2021-03-15 14:23:08 +00006868 io_run_task_work_head(&sqd->park_task_work);
Xiaoguang Wang08369242020-11-03 14:15:59 +08006869 timeout = jiffies + sqd->sq_thread_idle;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006870 }
6871
Pavel Begunkov734551d2021-04-18 14:52:09 +01006872 io_uring_cancel_sqpoll(sqd);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006873 sqd->thread = NULL;
Jens Axboe05962f92021-03-06 13:58:48 -07006874 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
Jens Axboe5f3f26f2021-02-25 10:17:46 -07006875 io_ring_set_wakeup_flag(ctx);
Pavel Begunkov521d6a72021-03-11 23:29:38 +00006876 io_run_task_work();
Pavel Begunkovb7f5a0b2021-03-15 14:23:08 +00006877 io_run_task_work_head(&sqd->park_task_work);
Pavel Begunkov734551d2021-04-18 14:52:09 +01006878 mutex_unlock(&sqd->lock);
6879
Jens Axboe37d1e2e2021-02-17 21:03:43 -07006880 complete(&sqd->exited);
6881 do_exit(0);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006882}
6883
Jens Axboebda52162019-09-24 13:47:15 -06006884struct io_wait_queue {
6885 struct wait_queue_entry wq;
6886 struct io_ring_ctx *ctx;
6887 unsigned to_wait;
6888 unsigned nr_timeouts;
6889};
6890
Pavel Begunkov6c503152021-01-04 20:36:36 +00006891static inline bool io_should_wake(struct io_wait_queue *iowq)
Jens Axboebda52162019-09-24 13:47:15 -06006892{
6893 struct io_ring_ctx *ctx = iowq->ctx;
6894
6895 /*
Brian Gianforcarod195a662019-12-13 03:09:50 -08006896 * Wake up if we have enough events, or if a timeout occurred since we
Jens Axboebda52162019-09-24 13:47:15 -06006897 * started waiting. For timeouts, we always want to return to userspace,
6898 * regardless of event count.
6899 */
Pavel Begunkov6c503152021-01-04 20:36:36 +00006900 return io_cqring_events(ctx) >= iowq->to_wait ||
Jens Axboebda52162019-09-24 13:47:15 -06006901 atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts;
6902}
6903
6904static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode,
6905 int wake_flags, void *key)
6906{
6907 struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue,
6908 wq);
6909
Pavel Begunkov6c503152021-01-04 20:36:36 +00006910 /*
6911 * Cannot safely flush overflowed CQEs from here, ensure we wake up
6912 * the task, and the next invocation will do it.
6913 */
6914 if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow))
6915 return autoremove_wake_function(curr, mode, wake_flags, key);
6916 return -1;
Jens Axboebda52162019-09-24 13:47:15 -06006917}
6918
Jens Axboeaf9c1a42020-09-24 13:32:18 -06006919static int io_run_task_work_sig(void)
6920{
6921 if (io_run_task_work())
6922 return 1;
6923 if (!signal_pending(current))
6924 return 0;
Jens Axboe0b8cfa92021-03-21 14:16:08 -06006925 if (test_thread_flag(TIF_NOTIFY_SIGNAL))
Jens Axboe792ee0f62020-10-22 20:17:18 -06006926 return -ERESTARTSYS;
Jens Axboeaf9c1a42020-09-24 13:32:18 -06006927 return -EINTR;
6928}
6929
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00006930/* when returns >0, the caller should retry */
6931static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx,
6932 struct io_wait_queue *iowq,
6933 signed long *timeout)
6934{
6935 int ret;
6936
6937 /* make sure we run task_work before checking for signals */
6938 ret = io_run_task_work_sig();
6939 if (ret || io_should_wake(iowq))
6940 return ret;
6941 /* let the caller flush overflows, retry */
6942 if (test_bit(0, &ctx->cq_check_overflow))
6943 return 1;
6944
6945 *timeout = schedule_timeout(*timeout);
6946 return !*timeout ? -ETIME : 1;
6947}
6948
Jens Axboe2b188cc2019-01-07 10:46:33 -07006949/*
6950 * Wait until events become available, if we don't already have some. The
6951 * application must reap them itself, as they reside on the shared cq ring.
6952 */
6953static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
Hao Xuc73ebb62020-11-03 10:54:37 +08006954 const sigset_t __user *sig, size_t sigsz,
6955 struct __kernel_timespec __user *uts)
Jens Axboe2b188cc2019-01-07 10:46:33 -07006956{
Jens Axboebda52162019-09-24 13:47:15 -06006957 struct io_wait_queue iowq = {
6958 .wq = {
6959 .private = current,
6960 .func = io_wake_function,
6961 .entry = LIST_HEAD_INIT(iowq.wq.entry),
6962 },
6963 .ctx = ctx,
6964 .to_wait = min_events,
6965 };
Hristo Venev75b28af2019-08-26 17:23:46 +00006966 struct io_rings *rings = ctx->rings;
Pavel Begunkovc1d5a222021-02-04 13:51:57 +00006967 signed long timeout = MAX_SCHEDULE_TIMEOUT;
6968 int ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006969
Jens Axboeb41e9852020-02-17 09:52:41 -07006970 do {
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00006971 io_cqring_overflow_flush(ctx, false);
Pavel Begunkov6c503152021-01-04 20:36:36 +00006972 if (io_cqring_events(ctx) >= min_events)
Jens Axboeb41e9852020-02-17 09:52:41 -07006973 return 0;
Jens Axboe4c6e2772020-07-01 11:29:10 -06006974 if (!io_run_task_work())
Jens Axboeb41e9852020-02-17 09:52:41 -07006975 break;
Jens Axboeb41e9852020-02-17 09:52:41 -07006976 } while (1);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006977
6978 if (sig) {
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01006979#ifdef CONFIG_COMPAT
6980 if (in_compat_syscall())
6981 ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig,
Oleg Nesterovb7724342019-07-16 16:29:53 -07006982 sigsz);
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01006983 else
6984#endif
Oleg Nesterovb7724342019-07-16 16:29:53 -07006985 ret = set_user_sigmask(sig, sigsz);
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01006986
Jens Axboe2b188cc2019-01-07 10:46:33 -07006987 if (ret)
6988 return ret;
6989 }
6990
Hao Xuc73ebb62020-11-03 10:54:37 +08006991 if (uts) {
Pavel Begunkovc1d5a222021-02-04 13:51:57 +00006992 struct timespec64 ts;
6993
Hao Xuc73ebb62020-11-03 10:54:37 +08006994 if (get_timespec64(&ts, uts))
6995 return -EFAULT;
6996 timeout = timespec64_to_jiffies(&ts);
6997 }
6998
Jens Axboebda52162019-09-24 13:47:15 -06006999 iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts);
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02007000 trace_io_uring_cqring_wait(ctx, min_events);
Jens Axboebda52162019-09-24 13:47:15 -06007001 do {
Jens Axboeca0a2652021-03-04 17:15:48 -07007002 /* if we can't even flush overflow, don't wait for more */
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00007003 if (!io_cqring_overflow_flush(ctx, false)) {
Jens Axboeca0a2652021-03-04 17:15:48 -07007004 ret = -EBUSY;
7005 break;
7006 }
Jens Axboebda52162019-09-24 13:47:15 -06007007 prepare_to_wait_exclusive(&ctx->wait, &iowq.wq,
7008 TASK_INTERRUPTIBLE);
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00007009 ret = io_cqring_wait_schedule(ctx, &iowq, &timeout);
7010 finish_wait(&ctx->wait, &iowq.wq);
Jens Axboeca0a2652021-03-04 17:15:48 -07007011 cond_resched();
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00007012 } while (ret > 0);
Jens Axboebda52162019-09-24 13:47:15 -06007013
Jens Axboeb7db41c2020-07-04 08:55:50 -06007014 restore_saved_sigmask_unless(ret == -EINTR);
Jens Axboe2b188cc2019-01-07 10:46:33 -07007015
Hristo Venev75b28af2019-08-26 17:23:46 +00007016 return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07007017}
7018
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007019static void io_free_file_tables(struct io_file_table *table, unsigned nr_files)
Pavel Begunkov846a4ef2021-04-01 15:44:03 +01007020{
7021 unsigned i, nr_tables = DIV_ROUND_UP(nr_files, IORING_MAX_FILES_TABLE);
7022
7023 for (i = 0; i < nr_tables; i++)
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007024 kfree(table->files[i]);
7025 kfree(table->files);
7026 table->files = NULL;
Pavel Begunkov846a4ef2021-04-01 15:44:03 +01007027}
7028
Jens Axboe6b063142019-01-10 22:13:58 -07007029static void __io_sqe_files_unregister(struct io_ring_ctx *ctx)
7030{
7031#if defined(CONFIG_UNIX)
7032 if (ctx->ring_sock) {
7033 struct sock *sock = ctx->ring_sock->sk;
7034 struct sk_buff *skb;
7035
7036 while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL)
7037 kfree_skb(skb);
7038 }
7039#else
7040 int i;
7041
Jens Axboe65e19f52019-10-26 07:20:21 -06007042 for (i = 0; i < ctx->nr_user_files; i++) {
7043 struct file *file;
7044
7045 file = io_file_from_index(ctx, i);
7046 if (file)
7047 fput(file);
7048 }
Jens Axboe6b063142019-01-10 22:13:58 -07007049#endif
Pavel Begunkov08480402021-04-13 02:58:38 +01007050 io_free_file_tables(&ctx->file_table, ctx->nr_user_files);
7051 kfree(ctx->file_data);
7052 ctx->file_data = NULL;
7053 ctx->nr_user_files = 0;
Jens Axboe6b063142019-01-10 22:13:58 -07007054}
7055
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007056static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx)
Pavel Begunkov1642b442020-12-30 21:34:14 +00007057{
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007058 spin_lock_bh(&ctx->rsrc_ref_lock);
Pavel Begunkov1642b442020-12-30 21:34:14 +00007059}
7060
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007061static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx)
Jens Axboe6b063142019-01-10 22:13:58 -07007062{
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007063 spin_unlock_bh(&ctx->rsrc_ref_lock);
7064}
7065
Pavel Begunkov28a9fe22021-04-01 15:43:47 +01007066static void io_rsrc_node_destroy(struct io_rsrc_node *ref_node)
7067{
7068 percpu_ref_exit(&ref_node->refs);
7069 kfree(ref_node);
7070}
7071
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007072static void io_rsrc_node_switch(struct io_ring_ctx *ctx,
7073 struct io_rsrc_data *data_to_kill)
Jens Axboe6b063142019-01-10 22:13:58 -07007074{
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007075 WARN_ON_ONCE(!ctx->rsrc_backup_node);
7076 WARN_ON_ONCE(data_to_kill && !ctx->rsrc_node);
Pavel Begunkov82fbcfa2021-04-01 15:43:43 +01007077
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007078 if (data_to_kill) {
7079 struct io_rsrc_node *rsrc_node = ctx->rsrc_node;
Pavel Begunkov82fbcfa2021-04-01 15:43:43 +01007080
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007081 rsrc_node->rsrc_data = data_to_kill;
7082 io_rsrc_ref_lock(ctx);
7083 list_add_tail(&rsrc_node->node, &ctx->rsrc_ref_list);
7084 io_rsrc_ref_unlock(ctx);
Pavel Begunkov82fbcfa2021-04-01 15:43:43 +01007085
Pavel Begunkov3e942492021-04-11 01:46:34 +01007086 atomic_inc(&data_to_kill->refs);
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007087 percpu_ref_kill(&rsrc_node->refs);
7088 ctx->rsrc_node = NULL;
7089 }
7090
7091 if (!ctx->rsrc_node) {
7092 ctx->rsrc_node = ctx->rsrc_backup_node;
7093 ctx->rsrc_backup_node = NULL;
7094 }
Jens Axboe6b063142019-01-10 22:13:58 -07007095}
7096
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007097static int io_rsrc_node_switch_start(struct io_ring_ctx *ctx)
Pavel Begunkov8dd03af2021-03-19 17:22:36 +00007098{
7099 if (ctx->rsrc_backup_node)
7100 return 0;
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007101 ctx->rsrc_backup_node = io_rsrc_node_alloc(ctx);
Pavel Begunkov8dd03af2021-03-19 17:22:36 +00007102 return ctx->rsrc_backup_node ? 0 : -ENOMEM;
7103}
7104
Pavel Begunkov40ae0ff2021-04-01 15:43:44 +01007105static int io_rsrc_ref_quiesce(struct io_rsrc_data *data, struct io_ring_ctx *ctx)
Hao Xu8bad28d2021-02-19 17:19:36 +08007106{
7107 int ret;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007108
Pavel Begunkov215c3902021-04-01 15:43:48 +01007109 /* As we may drop ->uring_lock, other task may have started quiesce */
Hao Xu8bad28d2021-02-19 17:19:36 +08007110 if (data->quiesce)
7111 return -ENXIO;
7112
7113 data->quiesce = true;
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00007114 do {
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007115 ret = io_rsrc_node_switch_start(ctx);
Pavel Begunkov8dd03af2021-03-19 17:22:36 +00007116 if (ret)
Pavel Begunkovf2303b12021-02-20 18:03:49 +00007117 break;
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007118 io_rsrc_node_switch(ctx, data);
7119
Pavel Begunkov3e942492021-04-11 01:46:34 +01007120 /* kill initial ref, already quiesced if zero */
7121 if (atomic_dec_and_test(&data->refs))
7122 break;
Hao Xu8bad28d2021-02-19 17:19:36 +08007123 flush_delayed_work(&ctx->rsrc_put_work);
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00007124 ret = wait_for_completion_interruptible(&data->done);
7125 if (!ret)
7126 break;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007127
Pavel Begunkov3e942492021-04-11 01:46:34 +01007128 atomic_inc(&data->refs);
7129 /* wait for all works potentially completing data->done */
7130 flush_delayed_work(&ctx->rsrc_put_work);
Jens Axboecb5e1b82021-02-25 07:37:35 -07007131 reinit_completion(&data->done);
Pavel Begunkov8dd03af2021-03-19 17:22:36 +00007132
Hao Xu8bad28d2021-02-19 17:19:36 +08007133 mutex_unlock(&ctx->uring_lock);
7134 ret = io_run_task_work_sig();
7135 mutex_lock(&ctx->uring_lock);
Pavel Begunkovf2303b12021-02-20 18:03:49 +00007136 } while (ret >= 0);
Hao Xu8bad28d2021-02-19 17:19:36 +08007137 data->quiesce = false;
7138
Hao Xu8bad28d2021-02-19 17:19:36 +08007139 return ret;
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007140}
7141
Pavel Begunkov40ae0ff2021-04-01 15:43:44 +01007142static struct io_rsrc_data *io_rsrc_data_alloc(struct io_ring_ctx *ctx,
7143 rsrc_put_fn *do_put)
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007144{
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007145 struct io_rsrc_data *data;
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007146
7147 data = kzalloc(sizeof(*data), GFP_KERNEL);
7148 if (!data)
7149 return NULL;
7150
Pavel Begunkov3e942492021-04-11 01:46:34 +01007151 atomic_set(&data->refs, 1);
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007152 data->ctx = ctx;
Pavel Begunkov40ae0ff2021-04-01 15:43:44 +01007153 data->do_put = do_put;
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007154 init_completion(&data->done);
7155 return data;
7156}
7157
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007158static int io_sqe_files_unregister(struct io_ring_ctx *ctx)
7159{
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007160 int ret;
7161
Pavel Begunkov08480402021-04-13 02:58:38 +01007162 if (!ctx->file_data)
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007163 return -ENXIO;
Pavel Begunkov08480402021-04-13 02:58:38 +01007164 ret = io_rsrc_ref_quiesce(ctx->file_data, ctx);
7165 if (!ret)
7166 __io_sqe_files_unregister(ctx);
7167 return ret;
Jens Axboe6b063142019-01-10 22:13:58 -07007168}
7169
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007170static void io_sq_thread_unpark(struct io_sq_data *sqd)
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007171 __releases(&sqd->lock)
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007172{
Pavel Begunkov521d6a72021-03-11 23:29:38 +00007173 WARN_ON_ONCE(sqd->thread == current);
7174
Pavel Begunkov9e138a42021-03-14 20:57:12 +00007175 /*
7176 * Do the dance but not conditional clear_bit() because it'd race with
7177 * other threads incrementing park_pending and setting the bit.
7178 */
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007179 clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
Pavel Begunkov9e138a42021-03-14 20:57:12 +00007180 if (atomic_dec_return(&sqd->park_pending))
7181 set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007182 mutex_unlock(&sqd->lock);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007183}
7184
Jens Axboe86e0d672021-03-05 08:44:39 -07007185static void io_sq_thread_park(struct io_sq_data *sqd)
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007186 __acquires(&sqd->lock)
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007187{
Pavel Begunkov521d6a72021-03-11 23:29:38 +00007188 WARN_ON_ONCE(sqd->thread == current);
7189
Pavel Begunkov9e138a42021-03-14 20:57:12 +00007190 atomic_inc(&sqd->park_pending);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007191 set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007192 mutex_lock(&sqd->lock);
Jens Axboe05962f92021-03-06 13:58:48 -07007193 if (sqd->thread)
Jens Axboe86e0d672021-03-05 08:44:39 -07007194 wake_up_process(sqd->thread);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007195}
7196
7197static void io_sq_thread_stop(struct io_sq_data *sqd)
7198{
Pavel Begunkov521d6a72021-03-11 23:29:38 +00007199 WARN_ON_ONCE(sqd->thread == current);
Pavel Begunkov88885f62021-04-11 01:46:38 +01007200 WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state));
Pavel Begunkov521d6a72021-03-11 23:29:38 +00007201
Jens Axboe05962f92021-03-06 13:58:48 -07007202 set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state);
Pavel Begunkov88885f62021-04-11 01:46:38 +01007203 mutex_lock(&sqd->lock);
Jens Axboee8f98f242021-03-09 16:32:13 -07007204 if (sqd->thread)
7205 wake_up_process(sqd->thread);
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007206 mutex_unlock(&sqd->lock);
Jens Axboe05962f92021-03-06 13:58:48 -07007207 wait_for_completion(&sqd->exited);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007208}
7209
Jens Axboe534ca6d2020-09-02 13:52:19 -06007210static void io_put_sq_data(struct io_sq_data *sqd)
Jens Axboe6c271ce2019-01-10 11:22:30 -07007211{
Jens Axboe534ca6d2020-09-02 13:52:19 -06007212 if (refcount_dec_and_test(&sqd->refs)) {
Pavel Begunkov9e138a42021-03-14 20:57:12 +00007213 WARN_ON_ONCE(atomic_read(&sqd->park_pending));
7214
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007215 io_sq_thread_stop(sqd);
7216 kfree(sqd);
7217 }
7218}
7219
7220static void io_sq_thread_finish(struct io_ring_ctx *ctx)
7221{
7222 struct io_sq_data *sqd = ctx->sq_data;
7223
7224 if (sqd) {
Jens Axboe05962f92021-03-06 13:58:48 -07007225 io_sq_thread_park(sqd);
Pavel Begunkov521d6a72021-03-11 23:29:38 +00007226 list_del_init(&ctx->sqd_list);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007227 io_sqd_update_thread_idle(sqd);
Jens Axboe05962f92021-03-06 13:58:48 -07007228 io_sq_thread_unpark(sqd);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007229
7230 io_put_sq_data(sqd);
7231 ctx->sq_data = NULL;
Jens Axboe534ca6d2020-09-02 13:52:19 -06007232 }
7233}
7234
Jens Axboeaa061652020-09-02 14:50:27 -06007235static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p)
7236{
7237 struct io_ring_ctx *ctx_attach;
7238 struct io_sq_data *sqd;
7239 struct fd f;
7240
7241 f = fdget(p->wq_fd);
7242 if (!f.file)
7243 return ERR_PTR(-ENXIO);
7244 if (f.file->f_op != &io_uring_fops) {
7245 fdput(f);
7246 return ERR_PTR(-EINVAL);
7247 }
7248
7249 ctx_attach = f.file->private_data;
7250 sqd = ctx_attach->sq_data;
7251 if (!sqd) {
7252 fdput(f);
7253 return ERR_PTR(-EINVAL);
7254 }
Jens Axboe5c2469e2021-03-11 10:17:56 -07007255 if (sqd->task_tgid != current->tgid) {
7256 fdput(f);
7257 return ERR_PTR(-EPERM);
7258 }
Jens Axboeaa061652020-09-02 14:50:27 -06007259
7260 refcount_inc(&sqd->refs);
7261 fdput(f);
7262 return sqd;
7263}
7264
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007265static struct io_sq_data *io_get_sq_data(struct io_uring_params *p,
7266 bool *attached)
Jens Axboe534ca6d2020-09-02 13:52:19 -06007267{
7268 struct io_sq_data *sqd;
7269
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007270 *attached = false;
Jens Axboe5c2469e2021-03-11 10:17:56 -07007271 if (p->flags & IORING_SETUP_ATTACH_WQ) {
7272 sqd = io_attach_sq_data(p);
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007273 if (!IS_ERR(sqd)) {
7274 *attached = true;
Jens Axboe5c2469e2021-03-11 10:17:56 -07007275 return sqd;
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007276 }
Jens Axboe5c2469e2021-03-11 10:17:56 -07007277 /* fall through for EPERM case, setup new sqd/task */
7278 if (PTR_ERR(sqd) != -EPERM)
7279 return sqd;
7280 }
Jens Axboeaa061652020-09-02 14:50:27 -06007281
Jens Axboe534ca6d2020-09-02 13:52:19 -06007282 sqd = kzalloc(sizeof(*sqd), GFP_KERNEL);
7283 if (!sqd)
7284 return ERR_PTR(-ENOMEM);
7285
Pavel Begunkov9e138a42021-03-14 20:57:12 +00007286 atomic_set(&sqd->park_pending, 0);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007287 refcount_set(&sqd->refs, 1);
Jens Axboe69fb2132020-09-14 11:16:23 -06007288 INIT_LIST_HEAD(&sqd->ctx_list);
Pavel Begunkov09a6f4e2021-03-14 20:57:10 +00007289 mutex_init(&sqd->lock);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007290 init_waitqueue_head(&sqd->wait);
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007291 init_completion(&sqd->exited);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007292 return sqd;
7293}
7294
Jens Axboe6b063142019-01-10 22:13:58 -07007295#if defined(CONFIG_UNIX)
Jens Axboe6b063142019-01-10 22:13:58 -07007296/*
7297 * Ensure the UNIX gc is aware of our file set, so we are certain that
7298 * the io_uring can be safely unregistered on process exit, even if we have
7299 * loops in the file referencing.
7300 */
7301static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset)
7302{
7303 struct sock *sk = ctx->ring_sock->sk;
7304 struct scm_fp_list *fpl;
7305 struct sk_buff *skb;
Jens Axboe08a451732019-10-03 08:11:03 -06007306 int i, nr_files;
Jens Axboe6b063142019-01-10 22:13:58 -07007307
Jens Axboe6b063142019-01-10 22:13:58 -07007308 fpl = kzalloc(sizeof(*fpl), GFP_KERNEL);
7309 if (!fpl)
7310 return -ENOMEM;
7311
7312 skb = alloc_skb(0, GFP_KERNEL);
7313 if (!skb) {
7314 kfree(fpl);
7315 return -ENOMEM;
7316 }
7317
7318 skb->sk = sk;
Jens Axboe6b063142019-01-10 22:13:58 -07007319
Jens Axboe08a451732019-10-03 08:11:03 -06007320 nr_files = 0;
Jens Axboe62e398b2021-02-21 16:19:37 -07007321 fpl->user = get_uid(current_user());
Jens Axboe6b063142019-01-10 22:13:58 -07007322 for (i = 0; i < nr; i++) {
Jens Axboe65e19f52019-10-26 07:20:21 -06007323 struct file *file = io_file_from_index(ctx, i + offset);
7324
7325 if (!file)
Jens Axboe08a451732019-10-03 08:11:03 -06007326 continue;
Jens Axboe65e19f52019-10-26 07:20:21 -06007327 fpl->fp[nr_files] = get_file(file);
Jens Axboe08a451732019-10-03 08:11:03 -06007328 unix_inflight(fpl->user, fpl->fp[nr_files]);
7329 nr_files++;
Jens Axboe6b063142019-01-10 22:13:58 -07007330 }
7331
Jens Axboe08a451732019-10-03 08:11:03 -06007332 if (nr_files) {
7333 fpl->max = SCM_MAX_FD;
7334 fpl->count = nr_files;
7335 UNIXCB(skb).fp = fpl;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007336 skb->destructor = unix_destruct_scm;
Jens Axboe08a451732019-10-03 08:11:03 -06007337 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
7338 skb_queue_head(&sk->sk_receive_queue, skb);
Jens Axboe6b063142019-01-10 22:13:58 -07007339
Jens Axboe08a451732019-10-03 08:11:03 -06007340 for (i = 0; i < nr_files; i++)
7341 fput(fpl->fp[i]);
7342 } else {
7343 kfree_skb(skb);
7344 kfree(fpl);
7345 }
Jens Axboe6b063142019-01-10 22:13:58 -07007346
7347 return 0;
7348}
7349
7350/*
7351 * If UNIX sockets are enabled, fd passing can cause a reference cycle which
7352 * causes regular reference counting to break down. We rely on the UNIX
7353 * garbage collection to take care of this problem for us.
7354 */
7355static int io_sqe_files_scm(struct io_ring_ctx *ctx)
7356{
7357 unsigned left, total;
7358 int ret = 0;
7359
7360 total = 0;
7361 left = ctx->nr_user_files;
7362 while (left) {
7363 unsigned this_files = min_t(unsigned, left, SCM_MAX_FD);
Jens Axboe6b063142019-01-10 22:13:58 -07007364
7365 ret = __io_sqe_files_scm(ctx, this_files, total);
7366 if (ret)
7367 break;
7368 left -= this_files;
7369 total += this_files;
7370 }
7371
7372 if (!ret)
7373 return 0;
7374
7375 while (total < ctx->nr_user_files) {
Jens Axboe65e19f52019-10-26 07:20:21 -06007376 struct file *file = io_file_from_index(ctx, total);
7377
7378 if (file)
7379 fput(file);
Jens Axboe6b063142019-01-10 22:13:58 -07007380 total++;
7381 }
7382
7383 return ret;
7384}
7385#else
7386static int io_sqe_files_scm(struct io_ring_ctx *ctx)
7387{
7388 return 0;
7389}
7390#endif
7391
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007392static bool io_alloc_file_tables(struct io_file_table *table, unsigned nr_files)
Jens Axboe65e19f52019-10-26 07:20:21 -06007393{
Pavel Begunkov846a4ef2021-04-01 15:44:03 +01007394 unsigned i, nr_tables = DIV_ROUND_UP(nr_files, IORING_MAX_FILES_TABLE);
7395
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007396 table->files = kcalloc(nr_tables, sizeof(*table->files), GFP_KERNEL);
7397 if (!table->files)
Pavel Begunkov846a4ef2021-04-01 15:44:03 +01007398 return false;
Jens Axboe65e19f52019-10-26 07:20:21 -06007399
7400 for (i = 0; i < nr_tables; i++) {
Pavel Begunkov846a4ef2021-04-01 15:44:03 +01007401 unsigned int this_files = min(nr_files, IORING_MAX_FILES_TABLE);
Jens Axboe65e19f52019-10-26 07:20:21 -06007402
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007403 table->files[i] = kcalloc(this_files, sizeof(*table->files[i]),
Jens Axboe65e19f52019-10-26 07:20:21 -06007404 GFP_KERNEL);
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007405 if (!table->files[i])
Jens Axboe65e19f52019-10-26 07:20:21 -06007406 break;
7407 nr_files -= this_files;
7408 }
7409
7410 if (i == nr_tables)
Pavel Begunkov846a4ef2021-04-01 15:44:03 +01007411 return true;
Jens Axboe65e19f52019-10-26 07:20:21 -06007412
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007413 io_free_file_tables(table, nr_tables * IORING_MAX_FILES_TABLE);
Pavel Begunkov846a4ef2021-04-01 15:44:03 +01007414 return false;
Jens Axboe65e19f52019-10-26 07:20:21 -06007415}
7416
Pavel Begunkov47e90392021-04-01 15:43:56 +01007417static void io_rsrc_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc)
Jens Axboec3a31e62019-10-03 13:59:56 -06007418{
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007419 struct file *file = prsrc->file;
Jens Axboec3a31e62019-10-03 13:59:56 -06007420#if defined(CONFIG_UNIX)
Jens Axboec3a31e62019-10-03 13:59:56 -06007421 struct sock *sock = ctx->ring_sock->sk;
7422 struct sk_buff_head list, *head = &sock->sk_receive_queue;
7423 struct sk_buff *skb;
7424 int i;
7425
7426 __skb_queue_head_init(&list);
7427
7428 /*
7429 * Find the skb that holds this file in its SCM_RIGHTS. When found,
7430 * remove this entry and rearrange the file array.
7431 */
7432 skb = skb_dequeue(head);
7433 while (skb) {
7434 struct scm_fp_list *fp;
7435
7436 fp = UNIXCB(skb).fp;
7437 for (i = 0; i < fp->count; i++) {
7438 int left;
7439
7440 if (fp->fp[i] != file)
7441 continue;
7442
7443 unix_notinflight(fp->user, fp->fp[i]);
7444 left = fp->count - 1 - i;
7445 if (left) {
7446 memmove(&fp->fp[i], &fp->fp[i + 1],
7447 left * sizeof(struct file *));
7448 }
7449 fp->count--;
7450 if (!fp->count) {
7451 kfree_skb(skb);
7452 skb = NULL;
7453 } else {
7454 __skb_queue_tail(&list, skb);
7455 }
7456 fput(file);
7457 file = NULL;
7458 break;
7459 }
7460
7461 if (!file)
7462 break;
7463
7464 __skb_queue_tail(&list, skb);
7465
7466 skb = skb_dequeue(head);
7467 }
7468
7469 if (skb_peek(&list)) {
7470 spin_lock_irq(&head->lock);
7471 while ((skb = __skb_dequeue(&list)) != NULL)
7472 __skb_queue_tail(head, skb);
7473 spin_unlock_irq(&head->lock);
7474 }
7475#else
Jens Axboe05f3fb32019-12-09 11:22:50 -07007476 fput(file);
Jens Axboec3a31e62019-10-03 13:59:56 -06007477#endif
7478}
7479
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007480static void __io_rsrc_put_work(struct io_rsrc_node *ref_node)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007481{
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007482 struct io_rsrc_data *rsrc_data = ref_node->rsrc_data;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007483 struct io_ring_ctx *ctx = rsrc_data->ctx;
7484 struct io_rsrc_put *prsrc, *tmp;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007485
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007486 list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) {
7487 list_del(&prsrc->list);
Pavel Begunkov40ae0ff2021-04-01 15:43:44 +01007488 rsrc_data->do_put(ctx, prsrc);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007489 kfree(prsrc);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007490 }
7491
Pavel Begunkov28a9fe22021-04-01 15:43:47 +01007492 io_rsrc_node_destroy(ref_node);
Pavel Begunkov3e942492021-04-11 01:46:34 +01007493 if (atomic_dec_and_test(&rsrc_data->refs))
7494 complete(&rsrc_data->done);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007495}
7496
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007497static void io_rsrc_put_work(struct work_struct *work)
Jens Axboe4a38aed22020-05-14 17:21:15 -06007498{
7499 struct io_ring_ctx *ctx;
7500 struct llist_node *node;
7501
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007502 ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work);
7503 node = llist_del_all(&ctx->rsrc_put_llist);
Jens Axboe4a38aed22020-05-14 17:21:15 -06007504
7505 while (node) {
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007506 struct io_rsrc_node *ref_node;
Jens Axboe4a38aed22020-05-14 17:21:15 -06007507 struct llist_node *next = node->next;
7508
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007509 ref_node = llist_entry(node, struct io_rsrc_node, llist);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007510 __io_rsrc_put_work(ref_node);
Jens Axboe4a38aed22020-05-14 17:21:15 -06007511 node = next;
7512 }
7513}
7514
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00007515static void io_rsrc_node_ref_zero(struct percpu_ref *ref)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007516{
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007517 struct io_rsrc_node *node = container_of(ref, struct io_rsrc_node, refs);
Pavel Begunkov3e942492021-04-11 01:46:34 +01007518 struct io_ring_ctx *ctx = node->rsrc_data->ctx;
Pavel Begunkove2978222020-11-18 14:56:26 +00007519 bool first_add = false;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007520
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007521 io_rsrc_ref_lock(ctx);
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007522 node->done = true;
Pavel Begunkove2978222020-11-18 14:56:26 +00007523
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00007524 while (!list_empty(&ctx->rsrc_ref_list)) {
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007525 node = list_first_entry(&ctx->rsrc_ref_list,
7526 struct io_rsrc_node, node);
Pavel Begunkove2978222020-11-18 14:56:26 +00007527 /* recycle ref nodes in order */
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007528 if (!node->done)
Pavel Begunkove2978222020-11-18 14:56:26 +00007529 break;
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007530 list_del(&node->node);
7531 first_add |= llist_add(&node->llist, &ctx->rsrc_put_llist);
Pavel Begunkove2978222020-11-18 14:56:26 +00007532 }
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007533 io_rsrc_ref_unlock(ctx);
Pavel Begunkove2978222020-11-18 14:56:26 +00007534
Pavel Begunkov3e942492021-04-11 01:46:34 +01007535 if (first_add)
7536 mod_delayed_work(system_wq, &ctx->rsrc_put_work, HZ);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007537}
7538
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007539static struct io_rsrc_node *io_rsrc_node_alloc(struct io_ring_ctx *ctx)
Xiaoguang Wang05589552020-03-31 14:05:18 +08007540{
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007541 struct io_rsrc_node *ref_node;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007542
7543 ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL);
7544 if (!ref_node)
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007545 return NULL;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007546
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00007547 if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero,
Xiaoguang Wang05589552020-03-31 14:05:18 +08007548 0, GFP_KERNEL)) {
7549 kfree(ref_node);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007550 return NULL;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007551 }
7552 INIT_LIST_HEAD(&ref_node->node);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007553 INIT_LIST_HEAD(&ref_node->rsrc_list);
Pavel Begunkove2978222020-11-18 14:56:26 +00007554 ref_node->done = false;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007555 return ref_node;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007556}
7557
Jens Axboe05f3fb32019-12-09 11:22:50 -07007558static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg,
7559 unsigned nr_args)
7560{
7561 __s32 __user *fds = (__s32 __user *) arg;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007562 struct file *file;
Pavel Begunkovf3baed32021-04-01 15:43:42 +01007563 int fd, ret;
Pavel Begunkov846a4ef2021-04-01 15:44:03 +01007564 unsigned i;
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007565 struct io_rsrc_data *file_data;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007566
7567 if (ctx->file_data)
7568 return -EBUSY;
7569 if (!nr_args)
7570 return -EINVAL;
7571 if (nr_args > IORING_MAX_FIXED_FILES)
7572 return -EMFILE;
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007573 ret = io_rsrc_node_switch_start(ctx);
Pavel Begunkovf3baed32021-04-01 15:43:42 +01007574 if (ret)
7575 return ret;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007576
Pavel Begunkov47e90392021-04-01 15:43:56 +01007577 file_data = io_rsrc_data_alloc(ctx, io_rsrc_file_put);
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007578 if (!file_data)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007579 return -ENOMEM;
Dan Carpenter13770a72021-02-01 15:23:42 +03007580 ctx->file_data = file_data;
Pavel Begunkovf3baed32021-04-01 15:43:42 +01007581 ret = -ENOMEM;
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007582 if (!io_alloc_file_tables(&ctx->file_table, nr_args))
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007583 goto out_free;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007584
Jens Axboe05f3fb32019-12-09 11:22:50 -07007585 for (i = 0; i < nr_args; i++, ctx->nr_user_files++) {
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007586 if (copy_from_user(&fd, &fds[i], sizeof(fd))) {
7587 ret = -EFAULT;
7588 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007589 }
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007590 /* allow sparse sets */
7591 if (fd == -1)
7592 continue;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007593
Jens Axboe05f3fb32019-12-09 11:22:50 -07007594 file = fget(fd);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007595 ret = -EBADF;
7596 if (!file)
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007597 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007598
7599 /*
7600 * Don't allow io_uring instances to be registered. If UNIX
7601 * isn't enabled, then this causes a reference cycle and this
7602 * instance can never get freed. If UNIX is enabled we'll
7603 * handle it just fine, but there's still no point in allowing
7604 * a ring fd as it doesn't support regular read/write anyway.
7605 */
7606 if (file->f_op == &io_uring_fops) {
7607 fput(file);
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007608 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007609 }
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007610 io_fixed_file_set(io_fixed_file_slot(&ctx->file_table, i), file);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007611 }
7612
Jens Axboe05f3fb32019-12-09 11:22:50 -07007613 ret = io_sqe_files_scm(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007614 if (ret) {
Pavel Begunkov08480402021-04-13 02:58:38 +01007615 __io_sqe_files_unregister(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007616 return ret;
7617 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07007618
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007619 io_rsrc_node_switch(ctx, NULL);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007620 return ret;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007621out_fput:
7622 for (i = 0; i < ctx->nr_user_files; i++) {
7623 file = io_file_from_index(ctx, i);
7624 if (file)
7625 fput(file);
7626 }
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007627 io_free_file_tables(&ctx->file_table, nr_args);
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007628 ctx->nr_user_files = 0;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007629out_free:
Pavel Begunkov3e942492021-04-11 01:46:34 +01007630 kfree(ctx->file_data);
Jens Axboe55cbc252020-10-14 07:35:57 -06007631 ctx->file_data = NULL;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007632 return ret;
7633}
7634
Jens Axboec3a31e62019-10-03 13:59:56 -06007635static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file,
7636 int index)
7637{
7638#if defined(CONFIG_UNIX)
7639 struct sock *sock = ctx->ring_sock->sk;
7640 struct sk_buff_head *head = &sock->sk_receive_queue;
7641 struct sk_buff *skb;
7642
7643 /*
7644 * See if we can merge this file into an existing skb SCM_RIGHTS
7645 * file set. If there's no room, fall back to allocating a new skb
7646 * and filling it in.
7647 */
7648 spin_lock_irq(&head->lock);
7649 skb = skb_peek(head);
7650 if (skb) {
7651 struct scm_fp_list *fpl = UNIXCB(skb).fp;
7652
7653 if (fpl->count < SCM_MAX_FD) {
7654 __skb_unlink(skb, head);
7655 spin_unlock_irq(&head->lock);
7656 fpl->fp[fpl->count] = get_file(file);
7657 unix_inflight(fpl->user, fpl->fp[fpl->count]);
7658 fpl->count++;
7659 spin_lock_irq(&head->lock);
7660 __skb_queue_head(head, skb);
7661 } else {
7662 skb = NULL;
7663 }
7664 }
7665 spin_unlock_irq(&head->lock);
7666
7667 if (skb) {
7668 fput(file);
7669 return 0;
7670 }
7671
7672 return __io_sqe_files_scm(ctx, 1, index);
7673#else
7674 return 0;
7675#endif
7676}
7677
Pavel Begunkove7c78372021-04-01 15:43:45 +01007678static int io_queue_rsrc_removal(struct io_rsrc_data *data,
7679 struct io_rsrc_node *node, void *rsrc)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007680{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007681 struct io_rsrc_put *prsrc;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007682
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007683 prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL);
7684 if (!prsrc)
Hillf Dantona5318d32020-03-23 17:47:15 +08007685 return -ENOMEM;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007686
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007687 prsrc->rsrc = rsrc;
Pavel Begunkove7c78372021-04-01 15:43:45 +01007688 list_add(&prsrc->list, &node->rsrc_list);
Hillf Dantona5318d32020-03-23 17:47:15 +08007689 return 0;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007690}
7691
7692static int __io_sqe_files_update(struct io_ring_ctx *ctx,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007693 struct io_uring_rsrc_update *up,
Jens Axboe05f3fb32019-12-09 11:22:50 -07007694 unsigned nr_args)
7695{
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01007696 struct io_rsrc_data *data = ctx->file_data;
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01007697 struct io_fixed_file *file_slot;
7698 struct file *file;
Jens Axboec3a31e62019-10-03 13:59:56 -06007699 __s32 __user *fds;
7700 int fd, i, err;
7701 __u32 done;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007702 bool needs_switch = false;
Jens Axboec3a31e62019-10-03 13:59:56 -06007703
Jens Axboe05f3fb32019-12-09 11:22:50 -07007704 if (check_add_overflow(up->offset, nr_args, &done))
Jens Axboec3a31e62019-10-03 13:59:56 -06007705 return -EOVERFLOW;
7706 if (done > ctx->nr_user_files)
7707 return -EINVAL;
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007708 err = io_rsrc_node_switch_start(ctx);
Pavel Begunkov8dd03af2021-03-19 17:22:36 +00007709 if (err)
7710 return err;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007711
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007712 fds = u64_to_user_ptr(up->data);
Pavel Begunkov67973b92021-01-26 13:51:09 +00007713 for (done = 0; done < nr_args; done++) {
Jens Axboec3a31e62019-10-03 13:59:56 -06007714 err = 0;
7715 if (copy_from_user(&fd, &fds[done], sizeof(fd))) {
7716 err = -EFAULT;
7717 break;
7718 }
noah4e0377a2021-01-26 15:23:28 -05007719 if (fd == IORING_REGISTER_FILES_SKIP)
7720 continue;
7721
Pavel Begunkov67973b92021-01-26 13:51:09 +00007722 i = array_index_nospec(up->offset + done, ctx->nr_user_files);
Pavel Begunkovaeca2412021-04-11 01:46:37 +01007723 file_slot = io_fixed_file_slot(&ctx->file_table, i);
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007724
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01007725 if (file_slot->file_ptr) {
7726 file = (struct file *)(file_slot->file_ptr & FFS_MASK);
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007727 err = io_queue_rsrc_removal(data, ctx->rsrc_node, file);
Hillf Dantona5318d32020-03-23 17:47:15 +08007728 if (err)
7729 break;
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01007730 file_slot->file_ptr = 0;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007731 needs_switch = true;
Jens Axboec3a31e62019-10-03 13:59:56 -06007732 }
7733 if (fd != -1) {
Jens Axboec3a31e62019-10-03 13:59:56 -06007734 file = fget(fd);
7735 if (!file) {
7736 err = -EBADF;
7737 break;
7738 }
7739 /*
7740 * Don't allow io_uring instances to be registered. If
7741 * UNIX isn't enabled, then this causes a reference
7742 * cycle and this instance can never get freed. If UNIX
7743 * is enabled we'll handle it just fine, but there's
7744 * still no point in allowing a ring fd as it doesn't
7745 * support regular read/write anyway.
7746 */
7747 if (file->f_op == &io_uring_fops) {
7748 fput(file);
7749 err = -EBADF;
7750 break;
7751 }
Pavel Begunkov9a321c92021-04-01 15:44:01 +01007752 io_fixed_file_set(file_slot, file);
Jens Axboec3a31e62019-10-03 13:59:56 -06007753 err = io_sqe_file_register(ctx, file, i);
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007754 if (err) {
Pavel Begunkova04b0ac2021-04-01 15:44:04 +01007755 file_slot->file_ptr = 0;
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007756 fput(file);
Jens Axboec3a31e62019-10-03 13:59:56 -06007757 break;
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007758 }
Jens Axboec3a31e62019-10-03 13:59:56 -06007759 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07007760 }
7761
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01007762 if (needs_switch)
7763 io_rsrc_node_switch(ctx, data);
Jens Axboec3a31e62019-10-03 13:59:56 -06007764 return done ? done : err;
7765}
Xiaoguang Wang05589552020-03-31 14:05:18 +08007766
Jens Axboe05f3fb32019-12-09 11:22:50 -07007767static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg,
7768 unsigned nr_args)
7769{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007770 struct io_uring_rsrc_update up;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007771
7772 if (!ctx->file_data)
7773 return -ENXIO;
7774 if (!nr_args)
7775 return -EINVAL;
7776 if (copy_from_user(&up, arg, sizeof(up)))
7777 return -EFAULT;
7778 if (up.resv)
7779 return -EINVAL;
7780
7781 return __io_sqe_files_update(ctx, &up, nr_args);
7782}
Jens Axboec3a31e62019-10-03 13:59:56 -06007783
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00007784static struct io_wq_work *io_free_work(struct io_wq_work *work)
Jens Axboe7d723062019-11-12 22:31:31 -07007785{
7786 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
7787
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00007788 req = io_put_req_find_next(req);
7789 return req ? &req->work : NULL;
Jens Axboe7d723062019-11-12 22:31:31 -07007790}
7791
Jens Axboe685fe7f2021-03-08 09:37:51 -07007792static struct io_wq *io_init_wq_offload(struct io_ring_ctx *ctx,
7793 struct task_struct *task)
Pavel Begunkov24369c22020-01-28 03:15:48 +03007794{
Jens Axboee9418942021-02-19 12:33:30 -07007795 struct io_wq_hash *hash;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007796 struct io_wq_data data;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007797 unsigned int concurrency;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007798
Jens Axboee9418942021-02-19 12:33:30 -07007799 hash = ctx->hash_map;
7800 if (!hash) {
7801 hash = kzalloc(sizeof(*hash), GFP_KERNEL);
7802 if (!hash)
7803 return ERR_PTR(-ENOMEM);
7804 refcount_set(&hash->refs, 1);
7805 init_waitqueue_head(&hash->wait);
7806 ctx->hash_map = hash;
7807 }
7808
7809 data.hash = hash;
Jens Axboe685fe7f2021-03-08 09:37:51 -07007810 data.task = task;
Pavel Begunkove9fd9392020-03-04 16:14:12 +03007811 data.free_work = io_free_work;
Pavel Begunkovf5fa38c2020-06-08 21:08:20 +03007812 data.do_work = io_wq_submit_work;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007813
Jens Axboed25e3a32021-02-16 11:41:41 -07007814 /* Do QD, or 4 * CPUS, whatever is smallest */
7815 concurrency = min(ctx->sq_entries, 4 * num_online_cpus());
Pavel Begunkov24369c22020-01-28 03:15:48 +03007816
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007817 return io_wq_create(concurrency, &data);
Pavel Begunkov24369c22020-01-28 03:15:48 +03007818}
7819
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007820static int io_uring_alloc_task_context(struct task_struct *task,
7821 struct io_ring_ctx *ctx)
Jens Axboe0f212202020-09-13 13:09:39 -06007822{
7823 struct io_uring_task *tctx;
Jens Axboed8a6df12020-10-15 16:24:45 -06007824 int ret;
Jens Axboe0f212202020-09-13 13:09:39 -06007825
7826 tctx = kmalloc(sizeof(*tctx), GFP_KERNEL);
7827 if (unlikely(!tctx))
7828 return -ENOMEM;
7829
Jens Axboed8a6df12020-10-15 16:24:45 -06007830 ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL);
7831 if (unlikely(ret)) {
7832 kfree(tctx);
7833 return ret;
7834 }
7835
Jens Axboe685fe7f2021-03-08 09:37:51 -07007836 tctx->io_wq = io_init_wq_offload(ctx, task);
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007837 if (IS_ERR(tctx->io_wq)) {
7838 ret = PTR_ERR(tctx->io_wq);
7839 percpu_counter_destroy(&tctx->inflight);
7840 kfree(tctx);
7841 return ret;
7842 }
7843
Jens Axboe0f212202020-09-13 13:09:39 -06007844 xa_init(&tctx->xa);
7845 init_waitqueue_head(&tctx->wait);
7846 tctx->last = NULL;
Jens Axboefdaf0832020-10-30 09:37:30 -06007847 atomic_set(&tctx->in_idle, 0);
Pavel Begunkovb303fe22021-04-11 01:46:26 +01007848 atomic_set(&tctx->inflight_tracked, 0);
Jens Axboe0f212202020-09-13 13:09:39 -06007849 task->io_uring = tctx;
Jens Axboe7cbf1722021-02-10 00:03:20 +00007850 spin_lock_init(&tctx->task_lock);
7851 INIT_WQ_LIST(&tctx->task_list);
7852 tctx->task_state = 0;
7853 init_task_work(&tctx->task_work, tctx_task_work);
Jens Axboe0f212202020-09-13 13:09:39 -06007854 return 0;
7855}
7856
7857void __io_uring_free(struct task_struct *tsk)
7858{
7859 struct io_uring_task *tctx = tsk->io_uring;
7860
7861 WARN_ON_ONCE(!xa_empty(&tctx->xa));
Pavel Begunkovef8eaa42021-02-27 11:16:45 +00007862 WARN_ON_ONCE(tctx->io_wq);
7863
Jens Axboed8a6df12020-10-15 16:24:45 -06007864 percpu_counter_destroy(&tctx->inflight);
Jens Axboe0f212202020-09-13 13:09:39 -06007865 kfree(tctx);
7866 tsk->io_uring = NULL;
7867}
7868
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02007869static int io_sq_offload_create(struct io_ring_ctx *ctx,
7870 struct io_uring_params *p)
Jens Axboe2b188cc2019-01-07 10:46:33 -07007871{
7872 int ret;
7873
Jens Axboed25e3a32021-02-16 11:41:41 -07007874 /* Retain compatibility with failing for an invalid attach attempt */
7875 if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) ==
7876 IORING_SETUP_ATTACH_WQ) {
7877 struct fd f;
7878
7879 f = fdget(p->wq_fd);
7880 if (!f.file)
7881 return -ENXIO;
7882 if (f.file->f_op != &io_uring_fops) {
7883 fdput(f);
7884 return -EINVAL;
7885 }
7886 fdput(f);
7887 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07007888 if (ctx->flags & IORING_SETUP_SQPOLL) {
Jens Axboe46fe18b2021-03-04 12:39:36 -07007889 struct task_struct *tsk;
Jens Axboe534ca6d2020-09-02 13:52:19 -06007890 struct io_sq_data *sqd;
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007891 bool attached;
Jens Axboe534ca6d2020-09-02 13:52:19 -06007892
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007893 sqd = io_get_sq_data(p, &attached);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007894 if (IS_ERR(sqd)) {
7895 ret = PTR_ERR(sqd);
7896 goto err;
7897 }
Jens Axboe69fb2132020-09-14 11:16:23 -06007898
Stefan Metzmacher7c30f36a2021-03-07 11:54:28 +01007899 ctx->sq_creds = get_current_cred();
Jens Axboe534ca6d2020-09-02 13:52:19 -06007900 ctx->sq_data = sqd;
Jens Axboe917257d2019-04-13 09:28:55 -06007901 ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle);
7902 if (!ctx->sq_thread_idle)
7903 ctx->sq_thread_idle = HZ;
7904
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007905 ret = 0;
Pavel Begunkov78d7f6b2021-03-10 13:13:53 +00007906 io_sq_thread_park(sqd);
Pavel Begunkovde75a3d2021-03-18 11:54:35 +00007907 list_add(&ctx->sqd_list, &sqd->ctx_list);
7908 io_sqd_update_thread_idle(sqd);
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007909 /* don't attach to a dying SQPOLL thread, would be racy */
Pavel Begunkovde75a3d2021-03-18 11:54:35 +00007910 if (attached && !sqd->thread)
Pavel Begunkov26984fb2021-03-11 23:29:37 +00007911 ret = -ENXIO;
Pavel Begunkov78d7f6b2021-03-10 13:13:53 +00007912 io_sq_thread_unpark(sqd);
7913
Pavel Begunkovde75a3d2021-03-18 11:54:35 +00007914 if (ret < 0)
7915 goto err;
7916 if (attached)
Jens Axboe5aa75ed2021-02-16 12:56:50 -07007917 return 0;
Jens Axboeaa061652020-09-02 14:50:27 -06007918
Jens Axboe6c271ce2019-01-10 11:22:30 -07007919 if (p->flags & IORING_SETUP_SQ_AFF) {
Jens Axboe44a9bd12019-05-14 20:00:30 -06007920 int cpu = p->sq_thread_cpu;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007921
Jens Axboe917257d2019-04-13 09:28:55 -06007922 ret = -EINVAL;
Jens Axboe44a9bd12019-05-14 20:00:30 -06007923 if (cpu >= nr_cpu_ids)
Jens Axboee8f98f242021-03-09 16:32:13 -07007924 goto err_sqpoll;
Shenghui Wang7889f442019-05-07 16:03:19 +08007925 if (!cpu_online(cpu))
Jens Axboee8f98f242021-03-09 16:32:13 -07007926 goto err_sqpoll;
Jens Axboe917257d2019-04-13 09:28:55 -06007927
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007928 sqd->sq_cpu = cpu;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007929 } else {
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007930 sqd->sq_cpu = -1;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007931 }
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007932
7933 sqd->task_pid = current->pid;
Jens Axboe5c2469e2021-03-11 10:17:56 -07007934 sqd->task_tgid = current->tgid;
Jens Axboe46fe18b2021-03-04 12:39:36 -07007935 tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE);
7936 if (IS_ERR(tsk)) {
7937 ret = PTR_ERR(tsk);
Jens Axboee8f98f242021-03-09 16:32:13 -07007938 goto err_sqpoll;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007939 }
Pavel Begunkov97a73a02021-03-08 17:30:54 +00007940
Jens Axboe46fe18b2021-03-04 12:39:36 -07007941 sqd->thread = tsk;
Pavel Begunkov97a73a02021-03-08 17:30:54 +00007942 ret = io_uring_alloc_task_context(tsk, ctx);
Jens Axboe46fe18b2021-03-04 12:39:36 -07007943 wake_up_new_task(tsk);
Jens Axboe0f212202020-09-13 13:09:39 -06007944 if (ret)
7945 goto err;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007946 } else if (p->flags & IORING_SETUP_SQ_AFF) {
7947 /* Can't have SQ_AFF without SQPOLL */
7948 ret = -EINVAL;
7949 goto err;
7950 }
7951
Jens Axboe2b188cc2019-01-07 10:46:33 -07007952 return 0;
7953err:
Jens Axboe37d1e2e2021-02-17 21:03:43 -07007954 io_sq_thread_finish(ctx);
Jens Axboe2b188cc2019-01-07 10:46:33 -07007955 return ret;
Jens Axboee8f98f242021-03-09 16:32:13 -07007956err_sqpoll:
7957 complete(&ctx->sq_data->exited);
7958 goto err;
Jens Axboe2b188cc2019-01-07 10:46:33 -07007959}
7960
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007961static inline void __io_unaccount_mem(struct user_struct *user,
7962 unsigned long nr_pages)
Jens Axboe2b188cc2019-01-07 10:46:33 -07007963{
7964 atomic_long_sub(nr_pages, &user->locked_vm);
7965}
7966
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007967static inline int __io_account_mem(struct user_struct *user,
7968 unsigned long nr_pages)
Jens Axboe2b188cc2019-01-07 10:46:33 -07007969{
7970 unsigned long page_limit, cur_pages, new_pages;
7971
7972 /* Don't allow more pages than we can safely lock */
7973 page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
7974
7975 do {
7976 cur_pages = atomic_long_read(&user->locked_vm);
7977 new_pages = cur_pages + nr_pages;
7978 if (new_pages > page_limit)
7979 return -ENOMEM;
7980 } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages,
7981 new_pages) != cur_pages);
7982
7983 return 0;
7984}
7985
Jens Axboe26bfa89e2021-02-09 20:14:12 -07007986static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007987{
Jens Axboe62e398b2021-02-21 16:19:37 -07007988 if (ctx->user)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007989 __io_unaccount_mem(ctx->user, nr_pages);
Bijan Mottahedeh30975822020-06-16 16:36:09 -07007990
Jens Axboe26bfa89e2021-02-09 20:14:12 -07007991 if (ctx->mm_account)
7992 atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm);
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007993}
7994
Jens Axboe26bfa89e2021-02-09 20:14:12 -07007995static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007996{
Bijan Mottahedeh30975822020-06-16 16:36:09 -07007997 int ret;
7998
Jens Axboe62e398b2021-02-21 16:19:37 -07007999 if (ctx->user) {
Bijan Mottahedeh30975822020-06-16 16:36:09 -07008000 ret = __io_account_mem(ctx->user, nr_pages);
8001 if (ret)
8002 return ret;
8003 }
8004
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008005 if (ctx->mm_account)
8006 atomic64_add(nr_pages, &ctx->mm_account->pinned_vm);
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008007
8008 return 0;
8009}
8010
Jens Axboe2b188cc2019-01-07 10:46:33 -07008011static void io_mem_free(void *ptr)
8012{
Mark Rutland52e04ef2019-04-30 17:30:21 +01008013 struct page *page;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008014
Mark Rutland52e04ef2019-04-30 17:30:21 +01008015 if (!ptr)
8016 return;
8017
8018 page = virt_to_head_page(ptr);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008019 if (put_page_testzero(page))
8020 free_compound_page(page);
8021}
8022
8023static void *io_mem_alloc(size_t size)
8024{
8025 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP |
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008026 __GFP_NORETRY | __GFP_ACCOUNT;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008027
8028 return (void *) __get_free_pages(gfp_flags, get_order(size));
8029}
8030
Hristo Venev75b28af2019-08-26 17:23:46 +00008031static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries,
8032 size_t *sq_offset)
8033{
8034 struct io_rings *rings;
8035 size_t off, sq_array_size;
8036
8037 off = struct_size(rings, cqes, cq_entries);
8038 if (off == SIZE_MAX)
8039 return SIZE_MAX;
8040
8041#ifdef CONFIG_SMP
8042 off = ALIGN(off, SMP_CACHE_BYTES);
8043 if (off == 0)
8044 return SIZE_MAX;
8045#endif
8046
Dmitry Vyukovb36200f2020-07-11 11:31:11 +02008047 if (sq_offset)
8048 *sq_offset = off;
8049
Hristo Venev75b28af2019-08-26 17:23:46 +00008050 sq_array_size = array_size(sizeof(u32), sq_entries);
8051 if (sq_array_size == SIZE_MAX)
8052 return SIZE_MAX;
8053
8054 if (check_add_overflow(off, sq_array_size, &off))
8055 return SIZE_MAX;
8056
Hristo Venev75b28af2019-08-26 17:23:46 +00008057 return off;
8058}
8059
Pavel Begunkov7f61a1e92021-04-11 01:46:35 +01008060static void io_buffer_unmap(struct io_ring_ctx *ctx, struct io_mapped_ubuf *imu)
8061{
8062 unsigned int i;
8063
8064 for (i = 0; i < imu->nr_bvecs; i++)
8065 unpin_user_page(imu->bvec[i].bv_page);
8066 if (imu->acct_pages)
8067 io_unaccount_mem(ctx, imu->acct_pages);
8068 kvfree(imu->bvec);
8069 imu->nr_bvecs = 0;
8070}
8071
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008072static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx)
Jens Axboeedafcce2019-01-09 09:16:05 -07008073{
Pavel Begunkov7f61a1e92021-04-11 01:46:35 +01008074 unsigned int i;
Jens Axboeedafcce2019-01-09 09:16:05 -07008075
8076 if (!ctx->user_bufs)
8077 return -ENXIO;
8078
Pavel Begunkov7f61a1e92021-04-11 01:46:35 +01008079 for (i = 0; i < ctx->nr_user_bufs; i++)
8080 io_buffer_unmap(ctx, &ctx->user_bufs[i]);
Jens Axboeedafcce2019-01-09 09:16:05 -07008081 kfree(ctx->user_bufs);
8082 ctx->user_bufs = NULL;
8083 ctx->nr_user_bufs = 0;
8084 return 0;
8085}
8086
8087static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst,
8088 void __user *arg, unsigned index)
8089{
8090 struct iovec __user *src;
8091
8092#ifdef CONFIG_COMPAT
8093 if (ctx->compat) {
8094 struct compat_iovec __user *ciovs;
8095 struct compat_iovec ciov;
8096
8097 ciovs = (struct compat_iovec __user *) arg;
8098 if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov)))
8099 return -EFAULT;
8100
Jens Axboed55e5f52019-12-11 16:12:15 -07008101 dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base);
Jens Axboeedafcce2019-01-09 09:16:05 -07008102 dst->iov_len = ciov.iov_len;
8103 return 0;
8104 }
8105#endif
8106 src = (struct iovec __user *) arg;
8107 if (copy_from_user(dst, &src[index], sizeof(*dst)))
8108 return -EFAULT;
8109 return 0;
8110}
8111
Jens Axboede293932020-09-17 16:19:16 -06008112/*
8113 * Not super efficient, but this is just a registration time. And we do cache
8114 * the last compound head, so generally we'll only do a full search if we don't
8115 * match that one.
8116 *
8117 * We check if the given compound head page has already been accounted, to
8118 * avoid double accounting it. This allows us to account the full size of the
8119 * page, not just the constituent pages of a huge page.
8120 */
8121static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages,
8122 int nr_pages, struct page *hpage)
8123{
8124 int i, j;
8125
8126 /* check current page array */
8127 for (i = 0; i < nr_pages; i++) {
8128 if (!PageCompound(pages[i]))
8129 continue;
8130 if (compound_head(pages[i]) == hpage)
8131 return true;
8132 }
8133
8134 /* check previously registered pages */
8135 for (i = 0; i < ctx->nr_user_bufs; i++) {
8136 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
8137
8138 for (j = 0; j < imu->nr_bvecs; j++) {
8139 if (!PageCompound(imu->bvec[j].bv_page))
8140 continue;
8141 if (compound_head(imu->bvec[j].bv_page) == hpage)
8142 return true;
8143 }
8144 }
8145
8146 return false;
8147}
8148
8149static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages,
8150 int nr_pages, struct io_mapped_ubuf *imu,
8151 struct page **last_hpage)
8152{
8153 int i, ret;
8154
8155 for (i = 0; i < nr_pages; i++) {
8156 if (!PageCompound(pages[i])) {
8157 imu->acct_pages++;
8158 } else {
8159 struct page *hpage;
8160
8161 hpage = compound_head(pages[i]);
8162 if (hpage == *last_hpage)
8163 continue;
8164 *last_hpage = hpage;
8165 if (headpage_already_acct(ctx, pages, i, hpage))
8166 continue;
8167 imu->acct_pages += page_size(hpage) >> PAGE_SHIFT;
8168 }
8169 }
8170
8171 if (!imu->acct_pages)
8172 return 0;
8173
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008174 ret = io_account_mem(ctx, imu->acct_pages);
Jens Axboede293932020-09-17 16:19:16 -06008175 if (ret)
8176 imu->acct_pages = 0;
8177 return ret;
8178}
8179
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008180static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov,
8181 struct io_mapped_ubuf *imu,
8182 struct page **last_hpage)
Jens Axboeedafcce2019-01-09 09:16:05 -07008183{
8184 struct vm_area_struct **vmas = NULL;
8185 struct page **pages = NULL;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008186 unsigned long off, start, end, ubuf;
8187 size_t size;
8188 int ret, pret, nr_pages, i;
Jens Axboeedafcce2019-01-09 09:16:05 -07008189
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008190 ubuf = (unsigned long) iov->iov_base;
8191 end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
8192 start = ubuf >> PAGE_SHIFT;
8193 nr_pages = end - start;
8194
8195 ret = -ENOMEM;
8196
8197 pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL);
8198 if (!pages)
8199 goto done;
8200
8201 vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *),
8202 GFP_KERNEL);
8203 if (!vmas)
8204 goto done;
8205
8206 imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec),
8207 GFP_KERNEL);
8208 if (!imu->bvec)
8209 goto done;
8210
8211 ret = 0;
8212 mmap_read_lock(current->mm);
8213 pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM,
8214 pages, vmas);
8215 if (pret == nr_pages) {
8216 /* don't support file backed memory */
8217 for (i = 0; i < nr_pages; i++) {
8218 struct vm_area_struct *vma = vmas[i];
8219
8220 if (vma->vm_file &&
8221 !is_file_hugepages(vma->vm_file)) {
8222 ret = -EOPNOTSUPP;
8223 break;
8224 }
8225 }
8226 } else {
8227 ret = pret < 0 ? pret : -EFAULT;
8228 }
8229 mmap_read_unlock(current->mm);
8230 if (ret) {
8231 /*
8232 * if we did partial map, or found file backed vmas,
8233 * release any pages we did get
8234 */
8235 if (pret > 0)
8236 unpin_user_pages(pages, pret);
8237 kvfree(imu->bvec);
8238 goto done;
8239 }
8240
8241 ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage);
8242 if (ret) {
8243 unpin_user_pages(pages, pret);
8244 kvfree(imu->bvec);
8245 goto done;
8246 }
8247
8248 off = ubuf & ~PAGE_MASK;
8249 size = iov->iov_len;
8250 for (i = 0; i < nr_pages; i++) {
8251 size_t vec_len;
8252
8253 vec_len = min_t(size_t, size, PAGE_SIZE - off);
8254 imu->bvec[i].bv_page = pages[i];
8255 imu->bvec[i].bv_len = vec_len;
8256 imu->bvec[i].bv_offset = off;
8257 off = 0;
8258 size -= vec_len;
8259 }
8260 /* store original address for later verification */
8261 imu->ubuf = ubuf;
Pavel Begunkov4751f532021-04-01 15:43:55 +01008262 imu->ubuf_end = ubuf + iov->iov_len;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008263 imu->nr_bvecs = nr_pages;
8264 ret = 0;
8265done:
8266 kvfree(pages);
8267 kvfree(vmas);
8268 return ret;
8269}
8270
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008271static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008272{
Pavel Begunkov87094462021-04-11 01:46:36 +01008273 ctx->user_bufs = kcalloc(nr_args, sizeof(*ctx->user_bufs), GFP_KERNEL);
8274 return ctx->user_bufs ? 0 : -ENOMEM;
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008275}
8276
8277static int io_buffer_validate(struct iovec *iov)
8278{
Pavel Begunkov50e96982021-03-24 22:59:01 +00008279 unsigned long tmp, acct_len = iov->iov_len + (PAGE_SIZE - 1);
8280
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008281 /*
8282 * Don't impose further limits on the size and buffer
8283 * constraints here, we'll -EINVAL later when IO is
8284 * submitted if they are wrong.
8285 */
8286 if (!iov->iov_base || !iov->iov_len)
8287 return -EFAULT;
8288
8289 /* arbitrary limit, but we need something */
8290 if (iov->iov_len > SZ_1G)
8291 return -EFAULT;
8292
Pavel Begunkov50e96982021-03-24 22:59:01 +00008293 if (check_add_overflow((unsigned long)iov->iov_base, acct_len, &tmp))
8294 return -EOVERFLOW;
8295
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008296 return 0;
8297}
8298
8299static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg,
8300 unsigned int nr_args)
8301{
8302 int i, ret;
8303 struct iovec iov;
8304 struct page *last_hpage = NULL;
8305
Pavel Begunkov87094462021-04-11 01:46:36 +01008306 if (ctx->user_bufs)
8307 return -EBUSY;
8308 if (!nr_args || nr_args > UIO_MAXIOV)
8309 return -EINVAL;
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008310 ret = io_buffers_map_alloc(ctx, nr_args);
8311 if (ret)
8312 return ret;
8313
Pavel Begunkov87094462021-04-11 01:46:36 +01008314 for (i = 0; i < nr_args; i++, ctx->nr_user_bufs++) {
Jens Axboeedafcce2019-01-09 09:16:05 -07008315 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
Jens Axboeedafcce2019-01-09 09:16:05 -07008316
8317 ret = io_copy_iov(ctx, &iov, arg, i);
8318 if (ret)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008319 break;
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008320 ret = io_buffer_validate(&iov);
8321 if (ret)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008322 break;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008323 ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage);
8324 if (ret)
8325 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07008326 }
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008327
8328 if (ret)
8329 io_sqe_buffers_unregister(ctx);
8330
Jens Axboeedafcce2019-01-09 09:16:05 -07008331 return ret;
8332}
8333
Jens Axboe9b402842019-04-11 11:45:41 -06008334static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg)
8335{
8336 __s32 __user *fds = arg;
8337 int fd;
8338
8339 if (ctx->cq_ev_fd)
8340 return -EBUSY;
8341
8342 if (copy_from_user(&fd, fds, sizeof(*fds)))
8343 return -EFAULT;
8344
8345 ctx->cq_ev_fd = eventfd_ctx_fdget(fd);
8346 if (IS_ERR(ctx->cq_ev_fd)) {
8347 int ret = PTR_ERR(ctx->cq_ev_fd);
8348 ctx->cq_ev_fd = NULL;
8349 return ret;
8350 }
8351
8352 return 0;
8353}
8354
8355static int io_eventfd_unregister(struct io_ring_ctx *ctx)
8356{
8357 if (ctx->cq_ev_fd) {
8358 eventfd_ctx_put(ctx->cq_ev_fd);
8359 ctx->cq_ev_fd = NULL;
8360 return 0;
8361 }
8362
8363 return -ENXIO;
8364}
8365
Jens Axboe5a2e7452020-02-23 16:23:11 -07008366static void io_destroy_buffers(struct io_ring_ctx *ctx)
8367{
Jens Axboe9e15c3a2021-03-13 12:29:43 -07008368 struct io_buffer *buf;
8369 unsigned long index;
8370
8371 xa_for_each(&ctx->io_buffers, index, buf)
8372 __io_remove_buffers(ctx, buf, index, -1U);
Jens Axboe5a2e7452020-02-23 16:23:11 -07008373}
8374
Jens Axboe68e68ee2021-02-13 09:00:02 -07008375static void io_req_cache_free(struct list_head *list, struct task_struct *tsk)
Jens Axboe1b4c3512021-02-10 00:03:19 +00008376{
Jens Axboe68e68ee2021-02-13 09:00:02 -07008377 struct io_kiocb *req, *nxt;
Jens Axboe1b4c3512021-02-10 00:03:19 +00008378
Jens Axboe68e68ee2021-02-13 09:00:02 -07008379 list_for_each_entry_safe(req, nxt, list, compl.list) {
8380 if (tsk && req->task != tsk)
8381 continue;
Jens Axboe1b4c3512021-02-10 00:03:19 +00008382 list_del(&req->compl.list);
8383 kmem_cache_free(req_cachep, req);
8384 }
8385}
8386
Jens Axboe4010fec2021-02-27 15:04:18 -07008387static void io_req_caches_free(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07008388{
Pavel Begunkovbf019da2021-02-10 00:03:17 +00008389 struct io_submit_state *submit_state = &ctx->submit_state;
Pavel Begunkove5547d22021-02-23 22:17:20 +00008390 struct io_comp_state *cs = &ctx->submit_state.comp;
Pavel Begunkovbf019da2021-02-10 00:03:17 +00008391
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008392 mutex_lock(&ctx->uring_lock);
8393
Pavel Begunkov8e5c66c2021-02-22 11:45:55 +00008394 if (submit_state->free_reqs) {
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008395 kmem_cache_free_bulk(req_cachep, submit_state->free_reqs,
8396 submit_state->reqs);
Pavel Begunkov8e5c66c2021-02-22 11:45:55 +00008397 submit_state->free_reqs = 0;
8398 }
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008399
Pavel Begunkovdac7a092021-03-19 17:22:39 +00008400 io_flush_cached_locked_reqs(ctx, cs);
Pavel Begunkove5547d22021-02-23 22:17:20 +00008401 io_req_cache_free(&cs->free_list, NULL);
Jens Axboe9a4fdbd2021-02-13 09:09:44 -07008402 mutex_unlock(&ctx->uring_lock);
8403}
8404
Jens Axboe2b188cc2019-01-07 10:46:33 -07008405static void io_ring_ctx_free(struct io_ring_ctx *ctx)
8406{
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008407 io_sq_thread_finish(ctx);
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008408 io_sqe_buffers_unregister(ctx);
Jens Axboe2aede0e2020-09-14 10:45:53 -06008409
Jens Axboe37d1e2e2021-02-17 21:03:43 -07008410 if (ctx->mm_account) {
Jens Axboe2aede0e2020-09-14 10:45:53 -06008411 mmdrop(ctx->mm_account);
8412 ctx->mm_account = NULL;
Bijan Mottahedeh30975822020-06-16 16:36:09 -07008413 }
Jens Axboedef596e2019-01-09 08:59:42 -07008414
Hao Xu8bad28d2021-02-19 17:19:36 +08008415 mutex_lock(&ctx->uring_lock);
Pavel Begunkov08480402021-04-13 02:58:38 +01008416 if (ctx->file_data) {
8417 if (!atomic_dec_and_test(&ctx->file_data->refs))
8418 wait_for_completion(&ctx->file_data->done);
8419 __io_sqe_files_unregister(ctx);
8420 }
Pavel Begunkovc4ea0602021-04-01 15:43:58 +01008421 if (ctx->rings)
8422 __io_cqring_overflow_flush(ctx, true);
Hao Xu8bad28d2021-02-19 17:19:36 +08008423 mutex_unlock(&ctx->uring_lock);
Jens Axboe9b402842019-04-11 11:45:41 -06008424 io_eventfd_unregister(ctx);
Jens Axboe5a2e7452020-02-23 16:23:11 -07008425 io_destroy_buffers(ctx);
Pavel Begunkov07db2982021-04-20 12:03:32 +01008426 if (ctx->sq_creds)
8427 put_cred(ctx->sq_creds);
Jens Axboedef596e2019-01-09 08:59:42 -07008428
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01008429 /* there are no registered resources left, nobody uses it */
8430 if (ctx->rsrc_node)
8431 io_rsrc_node_destroy(ctx->rsrc_node);
Pavel Begunkov8dd03af2021-03-19 17:22:36 +00008432 if (ctx->rsrc_backup_node)
Pavel Begunkovb895c9a2021-04-01 15:43:40 +01008433 io_rsrc_node_destroy(ctx->rsrc_backup_node);
Pavel Begunkova7f0ed52021-04-01 15:43:46 +01008434 flush_delayed_work(&ctx->rsrc_put_work);
8435
8436 WARN_ON_ONCE(!list_empty(&ctx->rsrc_ref_list));
8437 WARN_ON_ONCE(!llist_empty(&ctx->rsrc_put_llist));
Pavel Begunkov8dd03af2021-03-19 17:22:36 +00008438
Jens Axboe2b188cc2019-01-07 10:46:33 -07008439#if defined(CONFIG_UNIX)
Eric Biggers355e8d262019-06-12 14:58:43 -07008440 if (ctx->ring_sock) {
8441 ctx->ring_sock->file = NULL; /* so that iput() is called */
Jens Axboe2b188cc2019-01-07 10:46:33 -07008442 sock_release(ctx->ring_sock);
Eric Biggers355e8d262019-06-12 14:58:43 -07008443 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07008444#endif
8445
Hristo Venev75b28af2019-08-26 17:23:46 +00008446 io_mem_free(ctx->rings);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008447 io_mem_free(ctx->sq_sqes);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008448
8449 percpu_ref_exit(&ctx->refs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008450 free_uid(ctx->user);
Jens Axboe4010fec2021-02-27 15:04:18 -07008451 io_req_caches_free(ctx);
Jens Axboee9418942021-02-19 12:33:30 -07008452 if (ctx->hash_map)
8453 io_wq_put_hash(ctx->hash_map);
Jens Axboe78076bb2019-12-04 19:56:40 -07008454 kfree(ctx->cancel_hash);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008455 kfree(ctx);
8456}
8457
8458static __poll_t io_uring_poll(struct file *file, poll_table *wait)
8459{
8460 struct io_ring_ctx *ctx = file->private_data;
8461 __poll_t mask = 0;
8462
8463 poll_wait(file, &ctx->cq_wait, wait);
Stefan Bühler4f7067c2019-04-24 23:54:17 +02008464 /*
8465 * synchronizes with barrier from wq_has_sleeper call in
8466 * io_commit_cqring
8467 */
Jens Axboe2b188cc2019-01-07 10:46:33 -07008468 smp_rmb();
Jens Axboe90554202020-09-03 12:12:41 -06008469 if (!io_sqring_full(ctx))
Jens Axboe2b188cc2019-01-07 10:46:33 -07008470 mask |= EPOLLOUT | EPOLLWRNORM;
Hao Xued670c32021-02-05 16:34:21 +08008471
8472 /*
8473 * Don't flush cqring overflow list here, just do a simple check.
8474 * Otherwise there could possible be ABBA deadlock:
8475 * CPU0 CPU1
8476 * ---- ----
8477 * lock(&ctx->uring_lock);
8478 * lock(&ep->mtx);
8479 * lock(&ctx->uring_lock);
8480 * lock(&ep->mtx);
8481 *
8482 * Users may get EPOLLIN meanwhile seeing nothing in cqring, this
8483 * pushs them to do the flush.
8484 */
8485 if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow))
Jens Axboe2b188cc2019-01-07 10:46:33 -07008486 mask |= EPOLLIN | EPOLLRDNORM;
8487
8488 return mask;
8489}
8490
8491static int io_uring_fasync(int fd, struct file *file, int on)
8492{
8493 struct io_ring_ctx *ctx = file->private_data;
8494
8495 return fasync_helper(fd, file, on, &ctx->cq_fasync);
8496}
8497
Yejune Deng0bead8c2020-12-24 11:02:20 +08008498static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id)
Jens Axboe071698e2020-01-28 10:04:42 -07008499{
Jens Axboe4379bf82021-02-15 13:40:22 -07008500 const struct cred *creds;
Jens Axboe071698e2020-01-28 10:04:42 -07008501
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00008502 creds = xa_erase(&ctx->personalities, id);
Jens Axboe4379bf82021-02-15 13:40:22 -07008503 if (creds) {
8504 put_cred(creds);
Yejune Deng0bead8c2020-12-24 11:02:20 +08008505 return 0;
Jens Axboe1e6fa522020-10-15 08:46:24 -06008506 }
Yejune Deng0bead8c2020-12-24 11:02:20 +08008507
8508 return -EINVAL;
8509}
8510
Pavel Begunkov9b465712021-03-15 14:23:07 +00008511static inline bool io_run_ctx_fallback(struct io_ring_ctx *ctx)
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008512{
Pavel Begunkov9b465712021-03-15 14:23:07 +00008513 return io_run_task_work_head(&ctx->exit_task_work);
Jens Axboe7c25c0d2021-02-16 07:17:00 -07008514}
8515
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008516struct io_tctx_exit {
8517 struct callback_head task_work;
8518 struct completion completion;
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00008519 struct io_ring_ctx *ctx;
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008520};
8521
8522static void io_tctx_exit_cb(struct callback_head *cb)
8523{
8524 struct io_uring_task *tctx = current->io_uring;
8525 struct io_tctx_exit *work;
8526
8527 work = container_of(cb, struct io_tctx_exit, task_work);
8528 /*
8529 * When @in_idle, we're in cancellation and it's racy to remove the
8530 * node. It'll be removed by the end of cancellation, just ignore it.
8531 */
8532 if (!atomic_read(&tctx->in_idle))
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00008533 io_uring_del_task_file((unsigned long)work->ctx);
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008534 complete(&work->completion);
8535}
8536
Jens Axboe85faa7b2020-04-09 18:14:00 -06008537static void io_ring_exit_work(struct work_struct *work)
8538{
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008539 struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx, exit_work);
Pavel Begunkovb5bb3a22021-03-06 11:02:16 +00008540 unsigned long timeout = jiffies + HZ * 60 * 5;
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008541 struct io_tctx_exit exit;
8542 struct io_tctx_node *node;
8543 int ret;
Jens Axboe85faa7b2020-04-09 18:14:00 -06008544
Jens Axboe56952e92020-06-17 15:00:04 -06008545 /*
8546 * If we're doing polled IO and end up having requests being
8547 * submitted async (out-of-line), then completions can come in while
8548 * we're waiting for refs to drop. We need to reap these manually,
8549 * as nobody else will be looking for them.
8550 */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008551 do {
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008552 io_uring_try_cancel_requests(ctx, NULL, NULL);
Pavel Begunkovb5bb3a22021-03-06 11:02:16 +00008553
8554 WARN_ON_ONCE(time_after(jiffies, timeout));
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008555 } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20));
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008556
Pavel Begunkov7f006512021-04-14 13:38:34 +01008557 init_completion(&exit.completion);
8558 init_task_work(&exit.task_work, io_tctx_exit_cb);
8559 exit.ctx = ctx;
Pavel Begunkov89b50662021-04-01 15:43:50 +01008560 /*
8561 * Some may use context even when all refs and requests have been put,
8562 * and they are free to do so while still holding uring_lock or
8563 * completion_lock, see __io_req_task_submit(). Apart from other work,
8564 * this lock/unlock section also waits them to finish.
8565 */
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008566 mutex_lock(&ctx->uring_lock);
8567 while (!list_empty(&ctx->tctx_list)) {
Pavel Begunkovb5bb3a22021-03-06 11:02:16 +00008568 WARN_ON_ONCE(time_after(jiffies, timeout));
8569
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008570 node = list_first_entry(&ctx->tctx_list, struct io_tctx_node,
8571 ctx_node);
Pavel Begunkov7f006512021-04-14 13:38:34 +01008572 /* don't spin on a single task if cancellation failed */
8573 list_rotate_left(&ctx->tctx_list);
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008574 ret = task_work_add(node->task, &exit.task_work, TWA_SIGNAL);
8575 if (WARN_ON_ONCE(ret))
8576 continue;
8577 wake_up_process(node->task);
8578
8579 mutex_unlock(&ctx->uring_lock);
8580 wait_for_completion(&exit.completion);
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008581 mutex_lock(&ctx->uring_lock);
8582 }
8583 mutex_unlock(&ctx->uring_lock);
Pavel Begunkov89b50662021-04-01 15:43:50 +01008584 spin_lock_irq(&ctx->completion_lock);
8585 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkovd56d9382021-03-06 11:02:13 +00008586
Jens Axboe85faa7b2020-04-09 18:14:00 -06008587 io_ring_ctx_free(ctx);
8588}
8589
Pavel Begunkov80c4cbd2021-03-25 18:32:43 +00008590/* Returns true if we found and killed one or more timeouts */
8591static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk,
8592 struct files_struct *files)
8593{
8594 struct io_kiocb *req, *tmp;
8595 int canceled = 0;
8596
8597 spin_lock_irq(&ctx->completion_lock);
8598 list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) {
8599 if (io_match_task(req, tsk, files)) {
8600 io_kill_timeout(req, -ECANCELED);
8601 canceled++;
8602 }
8603 }
Pavel Begunkov51520422021-03-29 11:39:29 +01008604 if (canceled != 0)
8605 io_commit_cqring(ctx);
Pavel Begunkov80c4cbd2021-03-25 18:32:43 +00008606 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkov80c4cbd2021-03-25 18:32:43 +00008607 if (canceled != 0)
8608 io_cqring_ev_posted(ctx);
8609 return canceled != 0;
8610}
8611
Jens Axboe2b188cc2019-01-07 10:46:33 -07008612static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx)
8613{
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00008614 unsigned long index;
8615 struct creds *creds;
8616
Jens Axboe2b188cc2019-01-07 10:46:33 -07008617 mutex_lock(&ctx->uring_lock);
8618 percpu_ref_kill(&ctx->refs);
Pavel Begunkov634578f2020-12-06 22:22:44 +00008619 if (ctx->rings)
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00008620 __io_cqring_overflow_flush(ctx, true);
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00008621 xa_for_each(&ctx->personalities, index, creds)
8622 io_unregister_personality(ctx, index);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008623 mutex_unlock(&ctx->uring_lock);
8624
Pavel Begunkov6b819282020-11-06 13:00:25 +00008625 io_kill_timeouts(ctx, NULL, NULL);
8626 io_poll_remove_all(ctx, NULL, NULL);
Jens Axboe561fb042019-10-24 07:25:42 -06008627
Jens Axboe15dff282019-11-13 09:09:23 -07008628 /* if we failed setting up the ctx, we might not have any rings */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008629 io_iopoll_try_reap_events(ctx);
Jens Axboe309fc032020-07-10 09:13:34 -06008630
Jens Axboe85faa7b2020-04-09 18:14:00 -06008631 INIT_WORK(&ctx->exit_work, io_ring_exit_work);
Jens Axboefc666772020-08-19 11:10:51 -06008632 /*
8633 * Use system_unbound_wq to avoid spawning tons of event kworkers
8634 * if we're exiting a ton of rings at the same time. It just adds
8635 * noise and overhead, there's no discernable change in runtime
8636 * over using system_wq.
8637 */
8638 queue_work(system_unbound_wq, &ctx->exit_work);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008639}
8640
8641static int io_uring_release(struct inode *inode, struct file *file)
8642{
8643 struct io_ring_ctx *ctx = file->private_data;
8644
8645 file->private_data = NULL;
8646 io_ring_ctx_wait_and_kill(ctx);
8647 return 0;
8648}
8649
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008650struct io_task_cancel {
8651 struct task_struct *task;
8652 struct files_struct *files;
8653};
Pavel Begunkov67c4d9e2020-06-15 10:24:05 +03008654
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008655static bool io_cancel_task_cb(struct io_wq_work *work, void *data)
Jens Axboeb711d4e2020-08-16 08:23:05 -07008656{
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008657 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008658 struct io_task_cancel *cancel = data;
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008659 bool ret;
8660
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008661 if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) {
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008662 unsigned long flags;
8663 struct io_ring_ctx *ctx = req->ctx;
8664
8665 /* protect against races with linked timeouts */
8666 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008667 ret = io_match_task(req, cancel->task, cancel->files);
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008668 spin_unlock_irqrestore(&ctx->completion_lock, flags);
8669 } else {
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008670 ret = io_match_task(req, cancel->task, cancel->files);
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008671 }
8672 return ret;
Jens Axboeb711d4e2020-08-16 08:23:05 -07008673}
8674
Pavel Begunkove1915f72021-03-11 23:29:35 +00008675static bool io_cancel_defer_files(struct io_ring_ctx *ctx,
Pavel Begunkovef9865a2020-11-05 14:06:19 +00008676 struct task_struct *task,
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008677 struct files_struct *files)
8678{
Pavel Begunkove1915f72021-03-11 23:29:35 +00008679 struct io_defer_entry *de;
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008680 LIST_HEAD(list);
8681
8682 spin_lock_irq(&ctx->completion_lock);
8683 list_for_each_entry_reverse(de, &ctx->defer_list, list) {
Pavel Begunkov08d23632020-11-06 13:00:22 +00008684 if (io_match_task(de->req, task, files)) {
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008685 list_cut_position(&list, &ctx->defer_list, &de->list);
8686 break;
8687 }
8688 }
8689 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkove1915f72021-03-11 23:29:35 +00008690 if (list_empty(&list))
8691 return false;
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008692
8693 while (!list_empty(&list)) {
8694 de = list_first_entry(&list, struct io_defer_entry, list);
8695 list_del_init(&de->list);
Pavel Begunkovf41db2732021-02-28 22:35:12 +00008696 io_req_complete_failed(de->req, -ECANCELED);
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008697 kfree(de);
8698 }
Pavel Begunkove1915f72021-03-11 23:29:35 +00008699 return true;
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008700}
8701
Pavel Begunkov1b007642021-03-06 11:02:17 +00008702static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data)
8703{
8704 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
8705
8706 return req->ctx == data;
8707}
8708
8709static bool io_uring_try_cancel_iowq(struct io_ring_ctx *ctx)
8710{
8711 struct io_tctx_node *node;
8712 enum io_wq_cancel cret;
8713 bool ret = false;
8714
8715 mutex_lock(&ctx->uring_lock);
8716 list_for_each_entry(node, &ctx->tctx_list, ctx_node) {
8717 struct io_uring_task *tctx = node->task->io_uring;
8718
8719 /*
8720 * io_wq will stay alive while we hold uring_lock, because it's
8721 * killed after ctx nodes, which requires to take the lock.
8722 */
8723 if (!tctx || !tctx->io_wq)
8724 continue;
8725 cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_ctx_cb, ctx, true);
8726 ret |= (cret != IO_WQ_CANCEL_NOTFOUND);
8727 }
8728 mutex_unlock(&ctx->uring_lock);
8729
8730 return ret;
8731}
8732
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008733static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
8734 struct task_struct *task,
8735 struct files_struct *files)
8736{
8737 struct io_task_cancel cancel = { .task = task, .files = files, };
Pavel Begunkov1b007642021-03-06 11:02:17 +00008738 struct io_uring_task *tctx = task ? task->io_uring : NULL;
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008739
8740 while (1) {
8741 enum io_wq_cancel cret;
8742 bool ret = false;
8743
Pavel Begunkov1b007642021-03-06 11:02:17 +00008744 if (!task) {
8745 ret |= io_uring_try_cancel_iowq(ctx);
8746 } else if (tctx && tctx->io_wq) {
8747 /*
8748 * Cancels requests of all rings, not only @ctx, but
8749 * it's fine as the task is in exit/exec.
8750 */
Jens Axboe5aa75ed2021-02-16 12:56:50 -07008751 cret = io_wq_cancel_cb(tctx->io_wq, io_cancel_task_cb,
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008752 &cancel, true);
8753 ret |= (cret != IO_WQ_CANCEL_NOTFOUND);
8754 }
8755
8756 /* SQPOLL thread does its own polling */
Jens Axboed052d1d2021-03-11 10:49:20 -07008757 if ((!(ctx->flags & IORING_SETUP_SQPOLL) && !files) ||
8758 (ctx->sq_data && ctx->sq_data->thread == current)) {
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008759 while (!list_empty_careful(&ctx->iopoll_list)) {
8760 io_iopoll_try_reap_events(ctx);
8761 ret = true;
8762 }
8763 }
8764
Pavel Begunkove1915f72021-03-11 23:29:35 +00008765 ret |= io_cancel_defer_files(ctx, task, files);
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008766 ret |= io_poll_remove_all(ctx, task, files);
8767 ret |= io_kill_timeouts(ctx, task, files);
8768 ret |= io_run_task_work();
Pavel Begunkovba50a032021-02-26 15:47:56 +00008769 ret |= io_run_ctx_fallback(ctx);
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008770 if (!ret)
8771 break;
8772 cond_resched();
8773 }
8774}
8775
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00008776static int __io_uring_add_task_file(struct io_ring_ctx *ctx)
Jens Axboe0f212202020-09-13 13:09:39 -06008777{
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008778 struct io_uring_task *tctx = current->io_uring;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008779 struct io_tctx_node *node;
Pavel Begunkova528b042020-12-21 18:34:04 +00008780 int ret;
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008781
8782 if (unlikely(!tctx)) {
Jens Axboe5aa75ed2021-02-16 12:56:50 -07008783 ret = io_uring_alloc_task_context(current, ctx);
Jens Axboe0f212202020-09-13 13:09:39 -06008784 if (unlikely(ret))
8785 return ret;
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008786 tctx = current->io_uring;
Jens Axboe0f212202020-09-13 13:09:39 -06008787 }
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00008788 if (!xa_load(&tctx->xa, (unsigned long)ctx)) {
8789 node = kmalloc(sizeof(*node), GFP_KERNEL);
8790 if (!node)
8791 return -ENOMEM;
8792 node->ctx = ctx;
8793 node->task = current;
Jens Axboe0f212202020-09-13 13:09:39 -06008794
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00008795 ret = xa_err(xa_store(&tctx->xa, (unsigned long)ctx,
8796 node, GFP_KERNEL));
8797 if (ret) {
8798 kfree(node);
8799 return ret;
Jens Axboe0f212202020-09-13 13:09:39 -06008800 }
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00008801
8802 mutex_lock(&ctx->uring_lock);
8803 list_add(&node->ctx_node, &ctx->tctx_list);
8804 mutex_unlock(&ctx->uring_lock);
Jens Axboe0f212202020-09-13 13:09:39 -06008805 }
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00008806 tctx->last = ctx;
Jens Axboe0f212202020-09-13 13:09:39 -06008807 return 0;
8808}
8809
8810/*
Pavel Begunkovcf27f3b2021-03-19 17:22:31 +00008811 * Note that this task has used io_uring. We use it for cancelation purposes.
8812 */
8813static inline int io_uring_add_task_file(struct io_ring_ctx *ctx)
8814{
8815 struct io_uring_task *tctx = current->io_uring;
8816
8817 if (likely(tctx && tctx->last == ctx))
8818 return 0;
8819 return __io_uring_add_task_file(ctx);
8820}
8821
8822/*
Jens Axboe0f212202020-09-13 13:09:39 -06008823 * Remove this io_uring_file -> task mapping.
8824 */
Pavel Begunkov29412672021-03-06 11:02:11 +00008825static void io_uring_del_task_file(unsigned long index)
Jens Axboe0f212202020-09-13 13:09:39 -06008826{
8827 struct io_uring_task *tctx = current->io_uring;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008828 struct io_tctx_node *node;
Pavel Begunkov29412672021-03-06 11:02:11 +00008829
Pavel Begunkoveebd2e32021-03-06 11:02:14 +00008830 if (!tctx)
8831 return;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008832 node = xa_erase(&tctx->xa, index);
8833 if (!node)
Pavel Begunkov29412672021-03-06 11:02:11 +00008834 return;
Jens Axboe0f212202020-09-13 13:09:39 -06008835
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008836 WARN_ON_ONCE(current != node->task);
8837 WARN_ON_ONCE(list_empty(&node->ctx_node));
8838
8839 mutex_lock(&node->ctx->uring_lock);
8840 list_del(&node->ctx_node);
8841 mutex_unlock(&node->ctx->uring_lock);
8842
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00008843 if (tctx->last == node->ctx)
Jens Axboe0f212202020-09-13 13:09:39 -06008844 tctx->last = NULL;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008845 kfree(node);
Jens Axboe0f212202020-09-13 13:09:39 -06008846}
8847
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00008848static void io_uring_clean_tctx(struct io_uring_task *tctx)
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008849{
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008850 struct io_tctx_node *node;
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008851 unsigned long index;
8852
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008853 xa_for_each(&tctx->xa, index, node)
Pavel Begunkov29412672021-03-06 11:02:11 +00008854 io_uring_del_task_file(index);
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00008855 if (tctx->io_wq) {
8856 io_wq_put_and_exit(tctx->io_wq);
8857 tctx->io_wq = NULL;
8858 }
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008859}
8860
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01008861static s64 tctx_inflight(struct io_uring_task *tctx, bool tracked)
Pavel Begunkov368b2082021-04-11 01:46:25 +01008862{
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01008863 if (tracked)
8864 return atomic_read(&tctx->inflight_tracked);
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008865 return percpu_counter_sum(&tctx->inflight);
8866}
8867
8868static void io_sqpoll_cancel_cb(struct callback_head *cb)
8869{
8870 struct io_tctx_exit *work = container_of(cb, struct io_tctx_exit, task_work);
Pavel Begunkov734551d2021-04-18 14:52:09 +01008871 struct io_sq_data *sqd = work->ctx->sq_data;
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008872
8873 if (sqd->thread)
Pavel Begunkov734551d2021-04-18 14:52:09 +01008874 io_uring_cancel_sqpoll(sqd);
8875 list_del_init(&work->ctx->sqd_list);
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008876 complete(&work->completion);
8877}
8878
8879static void io_sqpoll_cancel_sync(struct io_ring_ctx *ctx)
8880{
8881 struct io_sq_data *sqd = ctx->sq_data;
8882 struct io_tctx_exit work = { .ctx = ctx, };
8883 struct task_struct *task;
8884
8885 io_sq_thread_park(sqd);
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008886 io_sqd_update_thread_idle(sqd);
8887 task = sqd->thread;
8888 if (task) {
8889 init_completion(&work.completion);
8890 init_task_work(&work.task_work, io_sqpoll_cancel_cb);
Pavel Begunkovb7f5a0b2021-03-15 14:23:08 +00008891 io_task_work_add_head(&sqd->park_task_work, &work.task_work);
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008892 wake_up_process(task);
Pavel Begunkov734551d2021-04-18 14:52:09 +01008893 } else {
8894 list_del_init(&ctx->sqd_list);
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008895 }
8896 io_sq_thread_unpark(sqd);
8897
8898 if (task)
8899 wait_for_completion(&work.completion);
8900}
8901
Pavel Begunkov368b2082021-04-11 01:46:25 +01008902static void io_uring_try_cancel(struct files_struct *files)
Jens Axboe0f212202020-09-13 13:09:39 -06008903{
8904 struct io_uring_task *tctx = current->io_uring;
Pavel Begunkov13bf43f2021-03-06 11:02:12 +00008905 struct io_tctx_node *node;
Matthew Wilcox (Oracle)ce765372020-10-09 13:49:51 +01008906 unsigned long index;
Jens Axboe0f212202020-09-13 13:09:39 -06008907
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008908 xa_for_each(&tctx->xa, index, node) {
8909 struct io_ring_ctx *ctx = node->ctx;
8910
8911 if (ctx->sq_data) {
8912 io_sqpoll_cancel_sync(ctx);
8913 continue;
8914 }
Pavel Begunkov368b2082021-04-11 01:46:25 +01008915 io_uring_try_cancel_requests(ctx, current, files);
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008916 }
Jens Axboefdaf0832020-10-30 09:37:30 -06008917}
8918
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008919/* should only be called by SQPOLL task */
Pavel Begunkov734551d2021-04-18 14:52:09 +01008920static void io_uring_cancel_sqpoll(struct io_sq_data *sqd)
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008921{
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008922 struct io_uring_task *tctx = current->io_uring;
Pavel Begunkov734551d2021-04-18 14:52:09 +01008923 struct io_ring_ctx *ctx;
Jens Axboefdaf0832020-10-30 09:37:30 -06008924 s64 inflight;
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008925 DEFINE_WAIT(wait);
Jens Axboefdaf0832020-10-30 09:37:30 -06008926
Pavel Begunkov734551d2021-04-18 14:52:09 +01008927 WARN_ON_ONCE(!sqd || sqd->thread != current);
Pavel Begunkov521d6a72021-03-11 23:29:38 +00008928
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008929 atomic_inc(&tctx->in_idle);
8930 do {
8931 /* read completions before cancelations */
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01008932 inflight = tctx_inflight(tctx, false);
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008933 if (!inflight)
8934 break;
Pavel Begunkov734551d2021-04-18 14:52:09 +01008935 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
8936 io_uring_try_cancel_requests(ctx, current, NULL);
Jens Axboefdaf0832020-10-30 09:37:30 -06008937
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008938 prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE);
8939 /*
8940 * If we've seen completions, retry without waiting. This
8941 * avoids a race where a completion comes in before we did
8942 * prepare_to_wait().
8943 */
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01008944 if (inflight == tctx_inflight(tctx, false))
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00008945 schedule();
8946 finish_wait(&tctx->wait, &wait);
8947 } while (1);
8948 atomic_dec(&tctx->in_idle);
Jens Axboe0f212202020-09-13 13:09:39 -06008949}
8950
Jens Axboe0f212202020-09-13 13:09:39 -06008951/*
8952 * Find any io_uring fd that this task has registered or done IO on, and cancel
8953 * requests.
8954 */
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01008955void __io_uring_cancel(struct files_struct *files)
Jens Axboe0f212202020-09-13 13:09:39 -06008956{
8957 struct io_uring_task *tctx = current->io_uring;
8958 DEFINE_WAIT(wait);
Jens Axboed8a6df12020-10-15 16:24:45 -06008959 s64 inflight;
Jens Axboe0f212202020-09-13 13:09:39 -06008960
8961 /* make sure overflow events are dropped */
Jens Axboefdaf0832020-10-30 09:37:30 -06008962 atomic_inc(&tctx->in_idle);
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01008963 io_uring_try_cancel(files);
Pavel Begunkov5a978dc2021-03-27 09:59:30 +00008964
Jens Axboed8a6df12020-10-15 16:24:45 -06008965 do {
Jens Axboe0f212202020-09-13 13:09:39 -06008966 /* read completions before cancelations */
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01008967 inflight = tctx_inflight(tctx, !!files);
Jens Axboed8a6df12020-10-15 16:24:45 -06008968 if (!inflight)
8969 break;
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01008970 io_uring_try_cancel(files);
Jens Axboe0f212202020-09-13 13:09:39 -06008971 prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE);
8972
8973 /*
Pavel Begunkova1bb3cd2021-01-26 15:28:26 +00008974 * If we've seen completions, retry without waiting. This
8975 * avoids a race where a completion comes in before we did
8976 * prepare_to_wait().
Jens Axboe0f212202020-09-13 13:09:39 -06008977 */
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01008978 if (inflight == tctx_inflight(tctx, !!files))
Pavel Begunkova1bb3cd2021-01-26 15:28:26 +00008979 schedule();
Pavel Begunkovf57555e2020-12-20 13:21:44 +00008980 finish_wait(&tctx->wait, &wait);
Jens Axboed8a6df12020-10-15 16:24:45 -06008981 } while (1);
Jens Axboefdaf0832020-10-30 09:37:30 -06008982 atomic_dec(&tctx->in_idle);
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00008983
Pavel Begunkov8452d4a2021-02-27 11:16:46 +00008984 io_uring_clean_tctx(tctx);
Pavel Begunkov3f48cf12021-04-11 01:46:27 +01008985 if (!files) {
8986 /* for exec all current's requests should be gone, kill tctx */
8987 __io_uring_free(current);
8988 }
Pavel Begunkov44e728b2020-06-15 10:24:04 +03008989}
8990
Roman Penyaev6c5c2402019-11-28 12:53:22 +01008991static void *io_uring_validate_mmap_request(struct file *file,
8992 loff_t pgoff, size_t sz)
Jens Axboe2b188cc2019-01-07 10:46:33 -07008993{
Jens Axboe2b188cc2019-01-07 10:46:33 -07008994 struct io_ring_ctx *ctx = file->private_data;
Roman Penyaev6c5c2402019-11-28 12:53:22 +01008995 loff_t offset = pgoff << PAGE_SHIFT;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008996 struct page *page;
8997 void *ptr;
8998
8999 switch (offset) {
9000 case IORING_OFF_SQ_RING:
Hristo Venev75b28af2019-08-26 17:23:46 +00009001 case IORING_OFF_CQ_RING:
9002 ptr = ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009003 break;
9004 case IORING_OFF_SQES:
9005 ptr = ctx->sq_sqes;
9006 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009007 default:
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009008 return ERR_PTR(-EINVAL);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009009 }
9010
9011 page = virt_to_head_page(ptr);
Matthew Wilcox (Oracle)a50b8542019-09-23 15:34:25 -07009012 if (sz > page_size(page))
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009013 return ERR_PTR(-EINVAL);
9014
9015 return ptr;
9016}
9017
9018#ifdef CONFIG_MMU
9019
9020static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
9021{
9022 size_t sz = vma->vm_end - vma->vm_start;
9023 unsigned long pfn;
9024 void *ptr;
9025
9026 ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz);
9027 if (IS_ERR(ptr))
9028 return PTR_ERR(ptr);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009029
9030 pfn = virt_to_phys(ptr) >> PAGE_SHIFT;
9031 return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot);
9032}
9033
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009034#else /* !CONFIG_MMU */
9035
9036static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
9037{
9038 return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL;
9039}
9040
9041static unsigned int io_uring_nommu_mmap_capabilities(struct file *file)
9042{
9043 return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE;
9044}
9045
9046static unsigned long io_uring_nommu_get_unmapped_area(struct file *file,
9047 unsigned long addr, unsigned long len,
9048 unsigned long pgoff, unsigned long flags)
9049{
9050 void *ptr;
9051
9052 ptr = io_uring_validate_mmap_request(file, pgoff, len);
9053 if (IS_ERR(ptr))
9054 return PTR_ERR(ptr);
9055
9056 return (unsigned long) ptr;
9057}
9058
9059#endif /* !CONFIG_MMU */
9060
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009061static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx)
Jens Axboe90554202020-09-03 12:12:41 -06009062{
9063 DEFINE_WAIT(wait);
9064
9065 do {
9066 if (!io_sqring_full(ctx))
9067 break;
Jens Axboe90554202020-09-03 12:12:41 -06009068 prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE);
9069
9070 if (!io_sqring_full(ctx))
9071 break;
Jens Axboe90554202020-09-03 12:12:41 -06009072 schedule();
9073 } while (!signal_pending(current));
9074
9075 finish_wait(&ctx->sqo_sq_wait, &wait);
Yang Li51993282021-03-09 14:30:41 +08009076 return 0;
Jens Axboe90554202020-09-03 12:12:41 -06009077}
9078
Hao Xuc73ebb62020-11-03 10:54:37 +08009079static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz,
9080 struct __kernel_timespec __user **ts,
9081 const sigset_t __user **sig)
9082{
9083 struct io_uring_getevents_arg arg;
9084
9085 /*
9086 * If EXT_ARG isn't set, then we have no timespec and the argp pointer
9087 * is just a pointer to the sigset_t.
9088 */
9089 if (!(flags & IORING_ENTER_EXT_ARG)) {
9090 *sig = (const sigset_t __user *) argp;
9091 *ts = NULL;
9092 return 0;
9093 }
9094
9095 /*
9096 * EXT_ARG is set - ensure we agree on the size of it and copy in our
9097 * timespec and sigset_t pointers if good.
9098 */
9099 if (*argsz != sizeof(arg))
9100 return -EINVAL;
9101 if (copy_from_user(&arg, argp, sizeof(arg)))
9102 return -EFAULT;
9103 *sig = u64_to_user_ptr(arg.sigmask);
9104 *argsz = arg.sigmask_sz;
9105 *ts = u64_to_user_ptr(arg.ts);
9106 return 0;
9107}
9108
Jens Axboe2b188cc2019-01-07 10:46:33 -07009109SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
Hao Xuc73ebb62020-11-03 10:54:37 +08009110 u32, min_complete, u32, flags, const void __user *, argp,
9111 size_t, argsz)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009112{
9113 struct io_ring_ctx *ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009114 int submitted = 0;
9115 struct fd f;
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009116 long ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009117
Jens Axboe4c6e2772020-07-01 11:29:10 -06009118 io_run_task_work();
Jens Axboeb41e9852020-02-17 09:52:41 -07009119
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009120 if (unlikely(flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP |
9121 IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG)))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009122 return -EINVAL;
9123
9124 f = fdget(fd);
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009125 if (unlikely(!f.file))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009126 return -EBADF;
9127
9128 ret = -EOPNOTSUPP;
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009129 if (unlikely(f.file->f_op != &io_uring_fops))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009130 goto out_fput;
9131
9132 ret = -ENXIO;
9133 ctx = f.file->private_data;
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009134 if (unlikely(!percpu_ref_tryget(&ctx->refs)))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009135 goto out_fput;
9136
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009137 ret = -EBADFD;
Pavel Begunkov33f993d2021-03-19 17:22:30 +00009138 if (unlikely(ctx->flags & IORING_SETUP_R_DISABLED))
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009139 goto out;
9140
Jens Axboe6c271ce2019-01-10 11:22:30 -07009141 /*
9142 * For SQ polling, the thread will do all submissions and completions.
9143 * Just return the requested submit count, and wake the thread if
9144 * we were asked to.
9145 */
Jens Axboeb2a9ead2019-09-12 14:19:16 -06009146 ret = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07009147 if (ctx->flags & IORING_SETUP_SQPOLL) {
Pavel Begunkov6c2450a2021-02-23 12:40:22 +00009148 io_cqring_overflow_flush(ctx, false);
Pavel Begunkov89448c42020-12-17 00:24:39 +00009149
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009150 ret = -EOWNERDEAD;
Stefan Metzmacher04147482021-03-07 11:54:29 +01009151 if (unlikely(ctx->sq_data->thread == NULL)) {
9152 goto out;
9153 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07009154 if (flags & IORING_ENTER_SQ_WAKEUP)
Jens Axboe534ca6d2020-09-02 13:52:19 -06009155 wake_up(&ctx->sq_data->wait);
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009156 if (flags & IORING_ENTER_SQ_WAIT) {
9157 ret = io_sqpoll_wait_sq(ctx);
9158 if (ret)
9159 goto out;
9160 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07009161 submitted = to_submit;
Jens Axboeb2a9ead2019-09-12 14:19:16 -06009162 } else if (to_submit) {
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00009163 ret = io_uring_add_task_file(ctx);
Jens Axboe0f212202020-09-13 13:09:39 -06009164 if (unlikely(ret))
9165 goto out;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009166 mutex_lock(&ctx->uring_lock);
Jens Axboe0f212202020-09-13 13:09:39 -06009167 submitted = io_submit_sqes(ctx, to_submit);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009168 mutex_unlock(&ctx->uring_lock);
Pavel Begunkov7c504e652019-12-18 19:53:45 +03009169
9170 if (submitted != to_submit)
9171 goto out;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009172 }
9173 if (flags & IORING_ENTER_GETEVENTS) {
Hao Xuc73ebb62020-11-03 10:54:37 +08009174 const sigset_t __user *sig;
9175 struct __kernel_timespec __user *ts;
9176
9177 ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig);
9178 if (unlikely(ret))
9179 goto out;
9180
Jens Axboe2b188cc2019-01-07 10:46:33 -07009181 min_complete = min(min_complete, ctx->cq_entries);
9182
Xiaoguang Wang32b22442020-03-11 09:26:09 +08009183 /*
9184 * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user
9185 * space applications don't need to do io completion events
9186 * polling again, they can rely on io_sq_thread to do polling
9187 * work, which can reduce cpu usage and uring_lock contention.
9188 */
9189 if (ctx->flags & IORING_SETUP_IOPOLL &&
9190 !(ctx->flags & IORING_SETUP_SQPOLL)) {
Pavel Begunkov7668b922020-07-07 16:36:21 +03009191 ret = io_iopoll_check(ctx, min_complete);
Jens Axboedef596e2019-01-09 08:59:42 -07009192 } else {
Hao Xuc73ebb62020-11-03 10:54:37 +08009193 ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts);
Jens Axboedef596e2019-01-09 08:59:42 -07009194 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009195 }
9196
Pavel Begunkov7c504e652019-12-18 19:53:45 +03009197out:
Pavel Begunkov6805b32e2019-10-08 02:18:42 +03009198 percpu_ref_put(&ctx->refs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009199out_fput:
9200 fdput(f);
9201 return submitted ? submitted : ret;
9202}
9203
Tobias Klauserbebdb652020-02-26 18:38:32 +01009204#ifdef CONFIG_PROC_FS
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009205static int io_uring_show_cred(struct seq_file *m, unsigned int id,
9206 const struct cred *cred)
Jens Axboe87ce9552020-01-30 08:25:34 -07009207{
Jens Axboe87ce9552020-01-30 08:25:34 -07009208 struct user_namespace *uns = seq_user_ns(m);
9209 struct group_info *gi;
9210 kernel_cap_t cap;
9211 unsigned __capi;
9212 int g;
9213
9214 seq_printf(m, "%5d\n", id);
9215 seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid));
9216 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid));
9217 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid));
9218 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid));
9219 seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid));
9220 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid));
9221 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid));
9222 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid));
9223 seq_puts(m, "\n\tGroups:\t");
9224 gi = cred->group_info;
9225 for (g = 0; g < gi->ngroups; g++) {
9226 seq_put_decimal_ull(m, g ? " " : "",
9227 from_kgid_munged(uns, gi->gid[g]));
9228 }
9229 seq_puts(m, "\n\tCapEff:\t");
9230 cap = cred->cap_effective;
9231 CAP_FOR_EACH_U32(__capi)
9232 seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8);
9233 seq_putc(m, '\n');
9234 return 0;
9235}
9236
9237static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m)
9238{
Joseph Qidbbe9c62020-09-29 09:01:22 -06009239 struct io_sq_data *sq = NULL;
Jens Axboefad8e0d2020-09-28 08:57:48 -06009240 bool has_lock;
Jens Axboe87ce9552020-01-30 08:25:34 -07009241 int i;
9242
Jens Axboefad8e0d2020-09-28 08:57:48 -06009243 /*
9244 * Avoid ABBA deadlock between the seq lock and the io_uring mutex,
9245 * since fdinfo case grabs it in the opposite direction of normal use
9246 * cases. If we fail to get the lock, we just don't iterate any
9247 * structures that could be going away outside the io_uring mutex.
9248 */
9249 has_lock = mutex_trylock(&ctx->uring_lock);
9250
Jens Axboe5f3f26f2021-02-25 10:17:46 -07009251 if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL)) {
Joseph Qidbbe9c62020-09-29 09:01:22 -06009252 sq = ctx->sq_data;
Jens Axboe5f3f26f2021-02-25 10:17:46 -07009253 if (!sq->thread)
9254 sq = NULL;
9255 }
Joseph Qidbbe9c62020-09-29 09:01:22 -06009256
9257 seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1);
9258 seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1);
Jens Axboe87ce9552020-01-30 08:25:34 -07009259 seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009260 for (i = 0; has_lock && i < ctx->nr_user_files; i++) {
Jens Axboe7b29f922021-03-12 08:30:14 -07009261 struct file *f = io_file_from_index(ctx, i);
Jens Axboe87ce9552020-01-30 08:25:34 -07009262
Jens Axboe87ce9552020-01-30 08:25:34 -07009263 if (f)
9264 seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname);
9265 else
9266 seq_printf(m, "%5u: <none>\n", i);
9267 }
9268 seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009269 for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) {
Jens Axboe87ce9552020-01-30 08:25:34 -07009270 struct io_mapped_ubuf *buf = &ctx->user_bufs[i];
Pavel Begunkov4751f532021-04-01 15:43:55 +01009271 unsigned int len = buf->ubuf_end - buf->ubuf;
Jens Axboe87ce9552020-01-30 08:25:34 -07009272
Pavel Begunkov4751f532021-04-01 15:43:55 +01009273 seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf, len);
Jens Axboe87ce9552020-01-30 08:25:34 -07009274 }
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009275 if (has_lock && !xa_empty(&ctx->personalities)) {
9276 unsigned long index;
9277 const struct cred *cred;
9278
Jens Axboe87ce9552020-01-30 08:25:34 -07009279 seq_printf(m, "Personalities:\n");
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009280 xa_for_each(&ctx->personalities, index, cred)
9281 io_uring_show_cred(m, index, cred);
Jens Axboe87ce9552020-01-30 08:25:34 -07009282 }
Jens Axboed7718a92020-02-14 22:23:12 -07009283 seq_printf(m, "PollList:\n");
9284 spin_lock_irq(&ctx->completion_lock);
9285 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
9286 struct hlist_head *list = &ctx->cancel_hash[i];
9287 struct io_kiocb *req;
9288
9289 hlist_for_each_entry(req, list, hash_node)
9290 seq_printf(m, " op=%d, task_works=%d\n", req->opcode,
9291 req->task->task_works != NULL);
9292 }
9293 spin_unlock_irq(&ctx->completion_lock);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009294 if (has_lock)
9295 mutex_unlock(&ctx->uring_lock);
Jens Axboe87ce9552020-01-30 08:25:34 -07009296}
9297
9298static void io_uring_show_fdinfo(struct seq_file *m, struct file *f)
9299{
9300 struct io_ring_ctx *ctx = f->private_data;
9301
9302 if (percpu_ref_tryget(&ctx->refs)) {
9303 __io_uring_show_fdinfo(ctx, m);
9304 percpu_ref_put(&ctx->refs);
9305 }
9306}
Tobias Klauserbebdb652020-02-26 18:38:32 +01009307#endif
Jens Axboe87ce9552020-01-30 08:25:34 -07009308
Jens Axboe2b188cc2019-01-07 10:46:33 -07009309static const struct file_operations io_uring_fops = {
9310 .release = io_uring_release,
9311 .mmap = io_uring_mmap,
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009312#ifndef CONFIG_MMU
9313 .get_unmapped_area = io_uring_nommu_get_unmapped_area,
9314 .mmap_capabilities = io_uring_nommu_mmap_capabilities,
9315#endif
Jens Axboe2b188cc2019-01-07 10:46:33 -07009316 .poll = io_uring_poll,
9317 .fasync = io_uring_fasync,
Tobias Klauserbebdb652020-02-26 18:38:32 +01009318#ifdef CONFIG_PROC_FS
Jens Axboe87ce9552020-01-30 08:25:34 -07009319 .show_fdinfo = io_uring_show_fdinfo,
Tobias Klauserbebdb652020-02-26 18:38:32 +01009320#endif
Jens Axboe2b188cc2019-01-07 10:46:33 -07009321};
9322
9323static int io_allocate_scq_urings(struct io_ring_ctx *ctx,
9324 struct io_uring_params *p)
9325{
Hristo Venev75b28af2019-08-26 17:23:46 +00009326 struct io_rings *rings;
9327 size_t size, sq_array_offset;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009328
Jens Axboebd740482020-08-05 12:58:23 -06009329 /* make sure these are sane, as we already accounted them */
9330 ctx->sq_entries = p->sq_entries;
9331 ctx->cq_entries = p->cq_entries;
9332
Hristo Venev75b28af2019-08-26 17:23:46 +00009333 size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset);
9334 if (size == SIZE_MAX)
9335 return -EOVERFLOW;
9336
9337 rings = io_mem_alloc(size);
9338 if (!rings)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009339 return -ENOMEM;
9340
Hristo Venev75b28af2019-08-26 17:23:46 +00009341 ctx->rings = rings;
9342 ctx->sq_array = (u32 *)((char *)rings + sq_array_offset);
9343 rings->sq_ring_mask = p->sq_entries - 1;
9344 rings->cq_ring_mask = p->cq_entries - 1;
9345 rings->sq_ring_entries = p->sq_entries;
9346 rings->cq_ring_entries = p->cq_entries;
9347 ctx->sq_mask = rings->sq_ring_mask;
9348 ctx->cq_mask = rings->cq_ring_mask;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009349
9350 size = array_size(sizeof(struct io_uring_sqe), p->sq_entries);
Jens Axboeeb065d32019-11-20 09:26:29 -07009351 if (size == SIZE_MAX) {
9352 io_mem_free(ctx->rings);
9353 ctx->rings = NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009354 return -EOVERFLOW;
Jens Axboeeb065d32019-11-20 09:26:29 -07009355 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009356
9357 ctx->sq_sqes = io_mem_alloc(size);
Jens Axboeeb065d32019-11-20 09:26:29 -07009358 if (!ctx->sq_sqes) {
9359 io_mem_free(ctx->rings);
9360 ctx->rings = NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009361 return -ENOMEM;
Jens Axboeeb065d32019-11-20 09:26:29 -07009362 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009363
Jens Axboe2b188cc2019-01-07 10:46:33 -07009364 return 0;
9365}
9366
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009367static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file)
9368{
9369 int ret, fd;
9370
9371 fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
9372 if (fd < 0)
9373 return fd;
9374
Pavel Begunkovbaf186c2021-03-06 11:02:15 +00009375 ret = io_uring_add_task_file(ctx);
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009376 if (ret) {
9377 put_unused_fd(fd);
9378 return ret;
9379 }
9380 fd_install(fd, file);
9381 return fd;
9382}
9383
Jens Axboe2b188cc2019-01-07 10:46:33 -07009384/*
9385 * Allocate an anonymous fd, this is what constitutes the application
9386 * visible backing of an io_uring instance. The application mmaps this
9387 * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled,
9388 * we have to tie this fd to a socket for file garbage collection purposes.
9389 */
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009390static struct file *io_uring_get_file(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009391{
9392 struct file *file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009393#if defined(CONFIG_UNIX)
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009394 int ret;
9395
Jens Axboe2b188cc2019-01-07 10:46:33 -07009396 ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP,
9397 &ctx->ring_sock);
9398 if (ret)
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009399 return ERR_PTR(ret);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009400#endif
9401
Jens Axboe2b188cc2019-01-07 10:46:33 -07009402 file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx,
9403 O_RDWR | O_CLOEXEC);
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009404#if defined(CONFIG_UNIX)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009405 if (IS_ERR(file)) {
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009406 sock_release(ctx->ring_sock);
9407 ctx->ring_sock = NULL;
9408 } else {
9409 ctx->ring_sock->file = file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009410 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009411#endif
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009412 return file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009413}
9414
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009415static int io_uring_create(unsigned entries, struct io_uring_params *p,
9416 struct io_uring_params __user *params)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009417{
Jens Axboe2b188cc2019-01-07 10:46:33 -07009418 struct io_ring_ctx *ctx;
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009419 struct file *file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009420 int ret;
9421
Jens Axboe8110c1a2019-12-28 15:39:54 -07009422 if (!entries)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009423 return -EINVAL;
Jens Axboe8110c1a2019-12-28 15:39:54 -07009424 if (entries > IORING_MAX_ENTRIES) {
9425 if (!(p->flags & IORING_SETUP_CLAMP))
9426 return -EINVAL;
9427 entries = IORING_MAX_ENTRIES;
9428 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009429
9430 /*
9431 * Use twice as many entries for the CQ ring. It's possible for the
9432 * application to drive a higher depth than the size of the SQ ring,
9433 * since the sqes are only used at submission time. This allows for
Jens Axboe33a107f2019-10-04 12:10:03 -06009434 * some flexibility in overcommitting a bit. If the application has
9435 * set IORING_SETUP_CQSIZE, it will have passed in the desired number
9436 * of CQ ring entries manually.
Jens Axboe2b188cc2019-01-07 10:46:33 -07009437 */
9438 p->sq_entries = roundup_pow_of_two(entries);
Jens Axboe33a107f2019-10-04 12:10:03 -06009439 if (p->flags & IORING_SETUP_CQSIZE) {
9440 /*
9441 * If IORING_SETUP_CQSIZE is set, we do the same roundup
9442 * to a power-of-two, if it isn't already. We do NOT impose
9443 * any cq vs sq ring sizing.
9444 */
Joseph Qieb2667b32020-11-24 15:03:03 +08009445 if (!p->cq_entries)
Jens Axboe33a107f2019-10-04 12:10:03 -06009446 return -EINVAL;
Jens Axboe8110c1a2019-12-28 15:39:54 -07009447 if (p->cq_entries > IORING_MAX_CQ_ENTRIES) {
9448 if (!(p->flags & IORING_SETUP_CLAMP))
9449 return -EINVAL;
9450 p->cq_entries = IORING_MAX_CQ_ENTRIES;
9451 }
Joseph Qieb2667b32020-11-24 15:03:03 +08009452 p->cq_entries = roundup_pow_of_two(p->cq_entries);
9453 if (p->cq_entries < p->sq_entries)
9454 return -EINVAL;
Jens Axboe33a107f2019-10-04 12:10:03 -06009455 } else {
9456 p->cq_entries = 2 * p->sq_entries;
9457 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009458
Jens Axboe2b188cc2019-01-07 10:46:33 -07009459 ctx = io_ring_ctx_alloc(p);
Jens Axboe62e398b2021-02-21 16:19:37 -07009460 if (!ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009461 return -ENOMEM;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009462 ctx->compat = in_compat_syscall();
Jens Axboe62e398b2021-02-21 16:19:37 -07009463 if (!capable(CAP_IPC_LOCK))
9464 ctx->user = get_uid(current_user());
Jens Axboe2aede0e2020-09-14 10:45:53 -06009465
9466 /*
9467 * This is just grabbed for accounting purposes. When a process exits,
9468 * the mm is exited and dropped before the files, hence we need to hang
9469 * on to this mm purely for the purposes of being able to unaccount
9470 * memory (locked/pinned vm). It's not used for anything else.
9471 */
Jens Axboe6b7898e2020-08-25 07:58:00 -06009472 mmgrab(current->mm);
Jens Axboe2aede0e2020-09-14 10:45:53 -06009473 ctx->mm_account = current->mm;
Jens Axboe6b7898e2020-08-25 07:58:00 -06009474
Jens Axboe2b188cc2019-01-07 10:46:33 -07009475 ret = io_allocate_scq_urings(ctx, p);
9476 if (ret)
9477 goto err;
9478
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009479 ret = io_sq_offload_create(ctx, p);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009480 if (ret)
9481 goto err;
9482
Jens Axboe2b188cc2019-01-07 10:46:33 -07009483 memset(&p->sq_off, 0, sizeof(p->sq_off));
Hristo Venev75b28af2019-08-26 17:23:46 +00009484 p->sq_off.head = offsetof(struct io_rings, sq.head);
9485 p->sq_off.tail = offsetof(struct io_rings, sq.tail);
9486 p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask);
9487 p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries);
9488 p->sq_off.flags = offsetof(struct io_rings, sq_flags);
9489 p->sq_off.dropped = offsetof(struct io_rings, sq_dropped);
9490 p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009491
9492 memset(&p->cq_off, 0, sizeof(p->cq_off));
Hristo Venev75b28af2019-08-26 17:23:46 +00009493 p->cq_off.head = offsetof(struct io_rings, cq.head);
9494 p->cq_off.tail = offsetof(struct io_rings, cq.tail);
9495 p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask);
9496 p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries);
9497 p->cq_off.overflow = offsetof(struct io_rings, cq_overflow);
9498 p->cq_off.cqes = offsetof(struct io_rings, cqes);
Stefano Garzarella0d9b5b32020-05-15 18:38:04 +02009499 p->cq_off.flags = offsetof(struct io_rings, cq_flags);
Jens Axboeac90f242019-09-06 10:26:21 -06009500
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009501 p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP |
9502 IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS |
Jiufei Xue5769a352020-06-17 17:53:55 +08009503 IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL |
Hao Xuc73ebb62020-11-03 10:54:37 +08009504 IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED |
Jens Axboe1c0aa1fa2021-02-20 11:55:28 -07009505 IORING_FEAT_EXT_ARG | IORING_FEAT_NATIVE_WORKERS;
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009506
9507 if (copy_to_user(params, p, sizeof(*p))) {
9508 ret = -EFAULT;
9509 goto err;
9510 }
Jens Axboed1719f72020-07-30 13:43:53 -06009511
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009512 file = io_uring_get_file(ctx);
9513 if (IS_ERR(file)) {
9514 ret = PTR_ERR(file);
9515 goto err;
9516 }
9517
Jens Axboed1719f72020-07-30 13:43:53 -06009518 /*
Jens Axboe044c1ab2019-10-28 09:15:33 -06009519 * Install ring fd as the very last thing, so we don't risk someone
9520 * having closed it before we finish setup
9521 */
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009522 ret = io_uring_install_fd(ctx, file);
9523 if (ret < 0) {
9524 /* fput will clean it up */
9525 fput(file);
9526 return ret;
9527 }
Jens Axboe044c1ab2019-10-28 09:15:33 -06009528
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02009529 trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009530 return ret;
9531err:
9532 io_ring_ctx_wait_and_kill(ctx);
9533 return ret;
9534}
9535
9536/*
9537 * Sets up an aio uring context, and returns the fd. Applications asks for a
9538 * ring size, we return the actual sq/cq ring sizes (among other things) in the
9539 * params structure passed in.
9540 */
9541static long io_uring_setup(u32 entries, struct io_uring_params __user *params)
9542{
9543 struct io_uring_params p;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009544 int i;
9545
9546 if (copy_from_user(&p, params, sizeof(p)))
9547 return -EFAULT;
9548 for (i = 0; i < ARRAY_SIZE(p.resv); i++) {
9549 if (p.resv[i])
9550 return -EINVAL;
9551 }
9552
Jens Axboe6c271ce2019-01-10 11:22:30 -07009553 if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL |
Jens Axboe8110c1a2019-12-28 15:39:54 -07009554 IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE |
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009555 IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ |
9556 IORING_SETUP_R_DISABLED))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009557 return -EINVAL;
9558
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009559 return io_uring_create(entries, &p, params);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009560}
9561
9562SYSCALL_DEFINE2(io_uring_setup, u32, entries,
9563 struct io_uring_params __user *, params)
9564{
9565 return io_uring_setup(entries, params);
9566}
9567
Jens Axboe66f4af92020-01-16 15:36:52 -07009568static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args)
9569{
9570 struct io_uring_probe *p;
9571 size_t size;
9572 int i, ret;
9573
9574 size = struct_size(p, ops, nr_args);
9575 if (size == SIZE_MAX)
9576 return -EOVERFLOW;
9577 p = kzalloc(size, GFP_KERNEL);
9578 if (!p)
9579 return -ENOMEM;
9580
9581 ret = -EFAULT;
9582 if (copy_from_user(p, arg, size))
9583 goto out;
9584 ret = -EINVAL;
9585 if (memchr_inv(p, 0, size))
9586 goto out;
9587
9588 p->last_op = IORING_OP_LAST - 1;
9589 if (nr_args > IORING_OP_LAST)
9590 nr_args = IORING_OP_LAST;
9591
9592 for (i = 0; i < nr_args; i++) {
9593 p->ops[i].op = i;
9594 if (!io_op_defs[i].not_supported)
9595 p->ops[i].flags = IO_URING_OP_SUPPORTED;
9596 }
9597 p->ops_len = i;
9598
9599 ret = 0;
9600 if (copy_to_user(arg, p, size))
9601 ret = -EFAULT;
9602out:
9603 kfree(p);
9604 return ret;
9605}
9606
Jens Axboe071698e2020-01-28 10:04:42 -07009607static int io_register_personality(struct io_ring_ctx *ctx)
9608{
Jens Axboe4379bf82021-02-15 13:40:22 -07009609 const struct cred *creds;
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009610 u32 id;
Jens Axboe1e6fa522020-10-15 08:46:24 -06009611 int ret;
Jens Axboe071698e2020-01-28 10:04:42 -07009612
Jens Axboe4379bf82021-02-15 13:40:22 -07009613 creds = get_current_cred();
Jens Axboe1e6fa522020-10-15 08:46:24 -06009614
Matthew Wilcox (Oracle)61cf9372021-03-08 14:16:16 +00009615 ret = xa_alloc_cyclic(&ctx->personalities, &id, (void *)creds,
9616 XA_LIMIT(0, USHRT_MAX), &ctx->pers_next, GFP_KERNEL);
9617 if (!ret)
9618 return id;
9619 put_cred(creds);
Jens Axboe1e6fa522020-10-15 08:46:24 -06009620 return ret;
Jens Axboe071698e2020-01-28 10:04:42 -07009621}
9622
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009623static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg,
9624 unsigned int nr_args)
9625{
9626 struct io_uring_restriction *res;
9627 size_t size;
9628 int i, ret;
9629
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009630 /* Restrictions allowed only if rings started disabled */
9631 if (!(ctx->flags & IORING_SETUP_R_DISABLED))
9632 return -EBADFD;
9633
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009634 /* We allow only a single restrictions registration */
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009635 if (ctx->restrictions.registered)
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009636 return -EBUSY;
9637
9638 if (!arg || nr_args > IORING_MAX_RESTRICTIONS)
9639 return -EINVAL;
9640
9641 size = array_size(nr_args, sizeof(*res));
9642 if (size == SIZE_MAX)
9643 return -EOVERFLOW;
9644
9645 res = memdup_user(arg, size);
9646 if (IS_ERR(res))
9647 return PTR_ERR(res);
9648
9649 ret = 0;
9650
9651 for (i = 0; i < nr_args; i++) {
9652 switch (res[i].opcode) {
9653 case IORING_RESTRICTION_REGISTER_OP:
9654 if (res[i].register_op >= IORING_REGISTER_LAST) {
9655 ret = -EINVAL;
9656 goto out;
9657 }
9658
9659 __set_bit(res[i].register_op,
9660 ctx->restrictions.register_op);
9661 break;
9662 case IORING_RESTRICTION_SQE_OP:
9663 if (res[i].sqe_op >= IORING_OP_LAST) {
9664 ret = -EINVAL;
9665 goto out;
9666 }
9667
9668 __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op);
9669 break;
9670 case IORING_RESTRICTION_SQE_FLAGS_ALLOWED:
9671 ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags;
9672 break;
9673 case IORING_RESTRICTION_SQE_FLAGS_REQUIRED:
9674 ctx->restrictions.sqe_flags_required = res[i].sqe_flags;
9675 break;
9676 default:
9677 ret = -EINVAL;
9678 goto out;
9679 }
9680 }
9681
9682out:
9683 /* Reset all restrictions if an error happened */
9684 if (ret != 0)
9685 memset(&ctx->restrictions, 0, sizeof(ctx->restrictions));
9686 else
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009687 ctx->restrictions.registered = true;
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009688
9689 kfree(res);
9690 return ret;
9691}
9692
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009693static int io_register_enable_rings(struct io_ring_ctx *ctx)
9694{
9695 if (!(ctx->flags & IORING_SETUP_R_DISABLED))
9696 return -EBADFD;
9697
9698 if (ctx->restrictions.registered)
9699 ctx->restricted = 1;
9700
Pavel Begunkov0298ef92021-03-08 13:20:57 +00009701 ctx->flags &= ~IORING_SETUP_R_DISABLED;
9702 if (ctx->sq_data && wq_has_sleeper(&ctx->sq_data->wait))
9703 wake_up(&ctx->sq_data->wait);
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009704 return 0;
9705}
9706
Jens Axboe071698e2020-01-28 10:04:42 -07009707static bool io_register_op_must_quiesce(int op)
9708{
9709 switch (op) {
Pavel Begunkovf4f7d212021-04-01 15:44:02 +01009710 case IORING_REGISTER_FILES:
Jens Axboe071698e2020-01-28 10:04:42 -07009711 case IORING_UNREGISTER_FILES:
9712 case IORING_REGISTER_FILES_UPDATE:
9713 case IORING_REGISTER_PROBE:
9714 case IORING_REGISTER_PERSONALITY:
9715 case IORING_UNREGISTER_PERSONALITY:
9716 return false;
9717 default:
9718 return true;
9719 }
9720}
9721
Jens Axboeedafcce2019-01-09 09:16:05 -07009722static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode,
9723 void __user *arg, unsigned nr_args)
Jens Axboeb19062a2019-04-15 10:49:38 -06009724 __releases(ctx->uring_lock)
9725 __acquires(ctx->uring_lock)
Jens Axboeedafcce2019-01-09 09:16:05 -07009726{
9727 int ret;
9728
Jens Axboe35fa71a2019-04-22 10:23:23 -06009729 /*
9730 * We're inside the ring mutex, if the ref is already dying, then
9731 * someone else killed the ctx or is already going through
9732 * io_uring_register().
9733 */
9734 if (percpu_ref_is_dying(&ctx->refs))
9735 return -ENXIO;
9736
Pavel Begunkov75c40212021-04-15 13:07:40 +01009737 if (ctx->restricted) {
9738 if (opcode >= IORING_REGISTER_LAST)
9739 return -EINVAL;
9740 opcode = array_index_nospec(opcode, IORING_REGISTER_LAST);
9741 if (!test_bit(opcode, ctx->restrictions.register_op))
9742 return -EACCES;
9743 }
9744
Jens Axboe071698e2020-01-28 10:04:42 -07009745 if (io_register_op_must_quiesce(opcode)) {
Jens Axboe05f3fb32019-12-09 11:22:50 -07009746 percpu_ref_kill(&ctx->refs);
Jens Axboeb19062a2019-04-15 10:49:38 -06009747
Jens Axboe05f3fb32019-12-09 11:22:50 -07009748 /*
9749 * Drop uring mutex before waiting for references to exit. If
9750 * another thread is currently inside io_uring_enter() it might
9751 * need to grab the uring_lock to make progress. If we hold it
9752 * here across the drain wait, then we can deadlock. It's safe
9753 * to drop the mutex here, since no new references will come in
9754 * after we've killed the percpu ref.
9755 */
9756 mutex_unlock(&ctx->uring_lock);
Jens Axboeaf9c1a42020-09-24 13:32:18 -06009757 do {
9758 ret = wait_for_completion_interruptible(&ctx->ref_comp);
9759 if (!ret)
9760 break;
Jens Axboeed6930c2020-10-08 19:09:46 -06009761 ret = io_run_task_work_sig();
9762 if (ret < 0)
9763 break;
Jens Axboeaf9c1a42020-09-24 13:32:18 -06009764 } while (1);
Jens Axboe05f3fb32019-12-09 11:22:50 -07009765 mutex_lock(&ctx->uring_lock);
Jens Axboeaf9c1a42020-09-24 13:32:18 -06009766
Jens Axboec1503682020-01-08 08:26:07 -07009767 if (ret) {
Pavel Begunkovf70865d2021-04-11 01:46:40 +01009768 io_refs_resurrect(&ctx->refs, &ctx->ref_comp);
9769 return ret;
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009770 }
9771 }
9772
Jens Axboeedafcce2019-01-09 09:16:05 -07009773 switch (opcode) {
9774 case IORING_REGISTER_BUFFERS:
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08009775 ret = io_sqe_buffers_register(ctx, arg, nr_args);
Jens Axboeedafcce2019-01-09 09:16:05 -07009776 break;
9777 case IORING_UNREGISTER_BUFFERS:
9778 ret = -EINVAL;
9779 if (arg || nr_args)
9780 break;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08009781 ret = io_sqe_buffers_unregister(ctx);
Jens Axboeedafcce2019-01-09 09:16:05 -07009782 break;
Jens Axboe6b063142019-01-10 22:13:58 -07009783 case IORING_REGISTER_FILES:
9784 ret = io_sqe_files_register(ctx, arg, nr_args);
9785 break;
9786 case IORING_UNREGISTER_FILES:
9787 ret = -EINVAL;
9788 if (arg || nr_args)
9789 break;
9790 ret = io_sqe_files_unregister(ctx);
9791 break;
Jens Axboec3a31e62019-10-03 13:59:56 -06009792 case IORING_REGISTER_FILES_UPDATE:
9793 ret = io_sqe_files_update(ctx, arg, nr_args);
9794 break;
Jens Axboe9b402842019-04-11 11:45:41 -06009795 case IORING_REGISTER_EVENTFD:
Jens Axboef2842ab2020-01-08 11:04:00 -07009796 case IORING_REGISTER_EVENTFD_ASYNC:
Jens Axboe9b402842019-04-11 11:45:41 -06009797 ret = -EINVAL;
9798 if (nr_args != 1)
9799 break;
9800 ret = io_eventfd_register(ctx, arg);
Jens Axboef2842ab2020-01-08 11:04:00 -07009801 if (ret)
9802 break;
9803 if (opcode == IORING_REGISTER_EVENTFD_ASYNC)
9804 ctx->eventfd_async = 1;
9805 else
9806 ctx->eventfd_async = 0;
Jens Axboe9b402842019-04-11 11:45:41 -06009807 break;
9808 case IORING_UNREGISTER_EVENTFD:
9809 ret = -EINVAL;
9810 if (arg || nr_args)
9811 break;
9812 ret = io_eventfd_unregister(ctx);
9813 break;
Jens Axboe66f4af92020-01-16 15:36:52 -07009814 case IORING_REGISTER_PROBE:
9815 ret = -EINVAL;
9816 if (!arg || nr_args > 256)
9817 break;
9818 ret = io_probe(ctx, arg, nr_args);
9819 break;
Jens Axboe071698e2020-01-28 10:04:42 -07009820 case IORING_REGISTER_PERSONALITY:
9821 ret = -EINVAL;
9822 if (arg || nr_args)
9823 break;
9824 ret = io_register_personality(ctx);
9825 break;
9826 case IORING_UNREGISTER_PERSONALITY:
9827 ret = -EINVAL;
9828 if (arg)
9829 break;
9830 ret = io_unregister_personality(ctx, nr_args);
9831 break;
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009832 case IORING_REGISTER_ENABLE_RINGS:
9833 ret = -EINVAL;
9834 if (arg || nr_args)
9835 break;
9836 ret = io_register_enable_rings(ctx);
9837 break;
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009838 case IORING_REGISTER_RESTRICTIONS:
9839 ret = io_register_restrictions(ctx, arg, nr_args);
9840 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07009841 default:
9842 ret = -EINVAL;
9843 break;
9844 }
9845
Jens Axboe071698e2020-01-28 10:04:42 -07009846 if (io_register_op_must_quiesce(opcode)) {
Jens Axboe05f3fb32019-12-09 11:22:50 -07009847 /* bring the ctx back to life */
Jens Axboe05f3fb32019-12-09 11:22:50 -07009848 percpu_ref_reinit(&ctx->refs);
Jens Axboe0f158b42020-05-14 17:18:39 -06009849 reinit_completion(&ctx->ref_comp);
Jens Axboe05f3fb32019-12-09 11:22:50 -07009850 }
Jens Axboeedafcce2019-01-09 09:16:05 -07009851 return ret;
9852}
9853
9854SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode,
9855 void __user *, arg, unsigned int, nr_args)
9856{
9857 struct io_ring_ctx *ctx;
9858 long ret = -EBADF;
9859 struct fd f;
9860
9861 f = fdget(fd);
9862 if (!f.file)
9863 return -EBADF;
9864
9865 ret = -EOPNOTSUPP;
9866 if (f.file->f_op != &io_uring_fops)
9867 goto out_fput;
9868
9869 ctx = f.file->private_data;
9870
Pavel Begunkovb6c23dd2021-02-20 15:17:18 +00009871 io_run_task_work();
9872
Jens Axboeedafcce2019-01-09 09:16:05 -07009873 mutex_lock(&ctx->uring_lock);
9874 ret = __io_uring_register(ctx, opcode, arg, nr_args);
9875 mutex_unlock(&ctx->uring_lock);
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02009876 trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs,
9877 ctx->cq_ev_fd != NULL, ret);
Jens Axboeedafcce2019-01-09 09:16:05 -07009878out_fput:
9879 fdput(f);
9880 return ret;
9881}
9882
Jens Axboe2b188cc2019-01-07 10:46:33 -07009883static int __init io_uring_init(void)
9884{
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009885#define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \
9886 BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \
9887 BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \
9888} while (0)
9889
9890#define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \
9891 __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename)
9892 BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64);
9893 BUILD_BUG_SQE_ELEM(0, __u8, opcode);
9894 BUILD_BUG_SQE_ELEM(1, __u8, flags);
9895 BUILD_BUG_SQE_ELEM(2, __u16, ioprio);
9896 BUILD_BUG_SQE_ELEM(4, __s32, fd);
9897 BUILD_BUG_SQE_ELEM(8, __u64, off);
9898 BUILD_BUG_SQE_ELEM(8, __u64, addr2);
9899 BUILD_BUG_SQE_ELEM(16, __u64, addr);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03009900 BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009901 BUILD_BUG_SQE_ELEM(24, __u32, len);
9902 BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags);
9903 BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags);
9904 BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags);
9905 BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags);
Jiufei Xue5769a352020-06-17 17:53:55 +08009906 BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events);
9907 BUILD_BUG_SQE_ELEM(28, __u32, poll32_events);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009908 BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags);
9909 BUILD_BUG_SQE_ELEM(28, __u32, msg_flags);
9910 BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags);
9911 BUILD_BUG_SQE_ELEM(28, __u32, accept_flags);
9912 BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags);
9913 BUILD_BUG_SQE_ELEM(28, __u32, open_flags);
9914 BUILD_BUG_SQE_ELEM(28, __u32, statx_flags);
9915 BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03009916 BUILD_BUG_SQE_ELEM(28, __u32, splice_flags);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009917 BUILD_BUG_SQE_ELEM(32, __u64, user_data);
9918 BUILD_BUG_SQE_ELEM(40, __u16, buf_index);
9919 BUILD_BUG_SQE_ELEM(42, __u16, personality);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03009920 BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009921
Jens Axboed3656342019-12-18 09:50:26 -07009922 BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST);
Jens Axboe84557872020-03-03 15:28:17 -07009923 BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int));
Jens Axboe91f245d2021-02-09 13:48:50 -07009924 req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC |
9925 SLAB_ACCOUNT);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009926 return 0;
9927};
9928__initcall(io_uring_init);