blob: 8018c7076b25ce66a8215500ffa3531d43f4204b [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 Axboe6c271ce2019-01-10 11:22:30 -070060#include <linux/kthread.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070061#include <linux/blkdev.h>
Jens Axboeedafcce2019-01-09 09:16:05 -070062#include <linux/bvec.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070063#include <linux/net.h>
64#include <net/sock.h>
65#include <net/af_unix.h>
Jens Axboe6b063142019-01-10 22:13:58 -070066#include <net/scm.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070067#include <linux/anon_inodes.h>
68#include <linux/sched/mm.h>
69#include <linux/uaccess.h>
70#include <linux/nospec.h>
Jens Axboeedafcce2019-01-09 09:16:05 -070071#include <linux/sizes.h>
72#include <linux/hugetlb.h>
Jens Axboeaa4c3962019-11-29 10:14:00 -070073#include <linux/highmem.h>
Jens Axboe15b71ab2019-12-11 11:20:36 -070074#include <linux/namei.h>
75#include <linux/fsnotify.h>
Jens Axboe4840e412019-12-25 22:03:45 -070076#include <linux/fadvise.h>
Jens Axboe3e4827b2020-01-08 15:18:09 -070077#include <linux/eventpoll.h>
Jens Axboeff002b32020-02-07 16:05:21 -070078#include <linux/fs_struct.h>
Pavel Begunkov7d67af22020-02-24 11:32:45 +030079#include <linux/splice.h>
Jens Axboeb41e9852020-02-17 09:52:41 -070080#include <linux/task_work.h>
Jens Axboebcf5a062020-05-22 09:24:42 -060081#include <linux/pagemap.h>
Jens Axboe0f212202020-09-13 13:09:39 -060082#include <linux/io_uring.h>
Dennis Zhou91d8f512020-09-16 13:41:05 -070083#include <linux/blk-cgroup.h>
Jens Axboe4ea33a92020-10-15 13:46:44 -060084#include <linux/audit.h>
Jens Axboe2b188cc2019-01-07 10:46:33 -070085
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +020086#define CREATE_TRACE_POINTS
87#include <trace/events/io_uring.h>
88
Jens Axboe2b188cc2019-01-07 10:46:33 -070089#include <uapi/linux/io_uring.h>
90
91#include "internal.h"
Jens Axboe561fb042019-10-24 07:25:42 -060092#include "io-wq.h"
Jens Axboe2b188cc2019-01-07 10:46:33 -070093
Daniel Xu5277dea2019-09-14 14:23:45 -070094#define IORING_MAX_ENTRIES 32768
Jens Axboe33a107f2019-10-04 12:10:03 -060095#define IORING_MAX_CQ_ENTRIES (2 * IORING_MAX_ENTRIES)
Jens Axboe65e19f52019-10-26 07:20:21 -060096
97/*
98 * Shift of 9 is 512 entries, or exactly one page on 64-bit archs
99 */
100#define IORING_FILE_TABLE_SHIFT 9
101#define IORING_MAX_FILES_TABLE (1U << IORING_FILE_TABLE_SHIFT)
102#define IORING_FILE_TABLE_MASK (IORING_MAX_FILES_TABLE - 1)
103#define IORING_MAX_FIXED_FILES (64 * IORING_MAX_FILES_TABLE)
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200104#define IORING_MAX_RESTRICTIONS (IORING_RESTRICTION_LAST + \
105 IORING_REGISTER_LAST + IORING_OP_LAST)
Jens Axboe2b188cc2019-01-07 10:46:33 -0700106
107struct 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
Jens Axboeedafcce2019-01-09 09:16:05 -0700190struct io_mapped_ubuf {
191 u64 ubuf;
192 size_t len;
193 struct bio_vec *bvec;
194 unsigned int nr_bvecs;
Jens Axboede293932020-09-17 16:19:16 -0600195 unsigned long acct_pages;
Jens Axboeedafcce2019-01-09 09:16:05 -0700196};
197
Jens Axboe65e19f52019-10-26 07:20:21 -0600198struct fixed_file_table {
199 struct file **files;
Jens Axboe31b51512019-01-18 22:56:34 -0700200};
201
Xiaoguang Wang05589552020-03-31 14:05:18 +0800202struct fixed_file_ref_node {
203 struct percpu_ref refs;
204 struct list_head node;
205 struct list_head file_list;
206 struct fixed_file_data *file_data;
Jens Axboe4a38aed22020-05-14 17:21:15 -0600207 struct llist_node llist;
Xiaoguang Wang05589552020-03-31 14:05:18 +0800208};
209
Jens Axboe05f3fb32019-12-09 11:22:50 -0700210struct fixed_file_data {
211 struct fixed_file_table *table;
212 struct io_ring_ctx *ctx;
213
Pavel Begunkovb2e96852020-10-10 18:34:16 +0100214 struct fixed_file_ref_node *node;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700215 struct percpu_ref refs;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700216 struct completion done;
Xiaoguang Wang05589552020-03-31 14:05:18 +0800217 struct list_head ref_list;
218 spinlock_t lock;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700219};
220
Jens Axboe5a2e7452020-02-23 16:23:11 -0700221struct io_buffer {
222 struct list_head list;
223 __u64 addr;
224 __s32 len;
225 __u16 bid;
226};
227
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200228struct io_restriction {
229 DECLARE_BITMAP(register_op, IORING_REGISTER_LAST);
230 DECLARE_BITMAP(sqe_op, IORING_OP_LAST);
231 u8 sqe_flags_allowed;
232 u8 sqe_flags_required;
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +0200233 bool registered;
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200234};
235
Jens Axboe534ca6d2020-09-02 13:52:19 -0600236struct io_sq_data {
237 refcount_t refs;
Jens Axboe69fb2132020-09-14 11:16:23 -0600238 struct mutex lock;
239
240 /* ctx's that are using this sqd */
241 struct list_head ctx_list;
242 struct list_head ctx_new_list;
243 struct mutex ctx_lock;
244
Jens Axboe534ca6d2020-09-02 13:52:19 -0600245 struct task_struct *thread;
246 struct wait_queue_head wait;
247};
248
Jens Axboe2b188cc2019-01-07 10:46:33 -0700249struct io_ring_ctx {
250 struct {
251 struct percpu_ref refs;
252 } ____cacheline_aligned_in_smp;
253
254 struct {
255 unsigned int flags;
Randy Dunlape1d85332020-02-05 20:57:10 -0800256 unsigned int compat: 1;
Bijan Mottahedehaad5d8d2020-06-16 16:36:08 -0700257 unsigned int limit_mem: 1;
Randy Dunlape1d85332020-02-05 20:57:10 -0800258 unsigned int cq_overflow_flushed: 1;
259 unsigned int drain_next: 1;
260 unsigned int eventfd_async: 1;
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200261 unsigned int restricted: 1;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700262
Hristo Venev75b28af2019-08-26 17:23:46 +0000263 /*
264 * Ring buffer of indices into array of io_uring_sqe, which is
265 * mmapped by the application using the IORING_OFF_SQES offset.
266 *
267 * This indirection could e.g. be used to assign fixed
268 * io_uring_sqe entries to operations and only submit them to
269 * the queue when needed.
270 *
271 * The kernel modifies neither the indices array nor the entries
272 * array.
273 */
274 u32 *sq_array;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700275 unsigned cached_sq_head;
276 unsigned sq_entries;
277 unsigned sq_mask;
Jens Axboe6c271ce2019-01-10 11:22:30 -0700278 unsigned sq_thread_idle;
Jens Axboe498ccd92019-10-25 10:04:25 -0600279 unsigned cached_sq_dropped;
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +0100280 unsigned cached_cq_overflow;
Jens Axboead3eb2c2019-12-18 17:12:20 -0700281 unsigned long sq_check_overflow;
Jens Axboede0617e2019-04-06 21:51:27 -0600282
283 struct list_head defer_list;
Jens Axboe5262f562019-09-17 12:26:57 -0600284 struct list_head timeout_list;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -0700285 struct list_head cq_overflow_list;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700286
Jens Axboefcb323c2019-10-24 12:39:47 -0600287 wait_queue_head_t inflight_wait;
Jens Axboead3eb2c2019-12-18 17:12:20 -0700288 struct io_uring_sqe *sq_sqes;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700289 } ____cacheline_aligned_in_smp;
290
Hristo Venev75b28af2019-08-26 17:23:46 +0000291 struct io_rings *rings;
292
Jens Axboe2b188cc2019-01-07 10:46:33 -0700293 /* IO offload */
Jens Axboe561fb042019-10-24 07:25:42 -0600294 struct io_wq *io_wq;
Jens Axboe2aede0e2020-09-14 10:45:53 -0600295
296 /*
297 * For SQPOLL usage - we hold a reference to the parent task, so we
298 * have access to the ->files
299 */
300 struct task_struct *sqo_task;
301
302 /* Only used for accounting purposes */
303 struct mm_struct *mm_account;
304
Dennis Zhou91d8f512020-09-16 13:41:05 -0700305#ifdef CONFIG_BLK_CGROUP
306 struct cgroup_subsys_state *sqo_blkcg_css;
307#endif
308
Jens Axboe534ca6d2020-09-02 13:52:19 -0600309 struct io_sq_data *sq_data; /* if using sq thread polling */
310
Jens Axboe90554202020-09-03 12:12:41 -0600311 struct wait_queue_head sqo_sq_wait;
Jens Axboe6a779382020-09-02 12:21:41 -0600312 struct wait_queue_entry sqo_wait_entry;
Jens Axboe69fb2132020-09-14 11:16:23 -0600313 struct list_head sqd_list;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700314
Jens Axboe6b063142019-01-10 22:13:58 -0700315 /*
316 * If used, fixed file set. Writers must ensure that ->refs is dead,
317 * readers must ensure that ->refs is alive as long as the file* is
318 * used. Only updated through io_uring_register(2).
319 */
Jens Axboe05f3fb32019-12-09 11:22:50 -0700320 struct fixed_file_data *file_data;
Jens Axboe6b063142019-01-10 22:13:58 -0700321 unsigned nr_user_files;
322
Jens Axboeedafcce2019-01-09 09:16:05 -0700323 /* if used, fixed mapped user buffers */
324 unsigned nr_user_bufs;
325 struct io_mapped_ubuf *user_bufs;
326
Jens Axboe2b188cc2019-01-07 10:46:33 -0700327 struct user_struct *user;
328
Jens Axboe0b8c0ec2019-12-02 08:50:00 -0700329 const struct cred *creds;
Jens Axboe181e4482019-11-25 08:52:30 -0700330
Jens Axboe4ea33a92020-10-15 13:46:44 -0600331#ifdef CONFIG_AUDIT
332 kuid_t loginuid;
333 unsigned int sessionid;
334#endif
335
Jens Axboe0f158b42020-05-14 17:18:39 -0600336 struct completion ref_comp;
337 struct completion sq_thread_comp;
Jens Axboe206aefd2019-11-07 18:27:42 -0700338
Jens Axboe0ddf92e2019-11-08 08:52:53 -0700339 /* if all else fails... */
340 struct io_kiocb *fallback_req;
341
Jens Axboe206aefd2019-11-07 18:27:42 -0700342#if defined(CONFIG_UNIX)
343 struct socket *ring_sock;
344#endif
345
Jens Axboe5a2e7452020-02-23 16:23:11 -0700346 struct idr io_buffer_idr;
347
Jens Axboe071698e2020-01-28 10:04:42 -0700348 struct idr personality_idr;
349
Jens Axboe206aefd2019-11-07 18:27:42 -0700350 struct {
351 unsigned cached_cq_tail;
352 unsigned cq_entries;
353 unsigned cq_mask;
354 atomic_t cq_timeouts;
Jens Axboead3eb2c2019-12-18 17:12:20 -0700355 unsigned long cq_check_overflow;
Jens Axboe206aefd2019-11-07 18:27:42 -0700356 struct wait_queue_head cq_wait;
357 struct fasync_struct *cq_fasync;
358 struct eventfd_ctx *cq_ev_fd;
359 } ____cacheline_aligned_in_smp;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700360
361 struct {
362 struct mutex uring_lock;
363 wait_queue_head_t wait;
364 } ____cacheline_aligned_in_smp;
365
366 struct {
367 spinlock_t completion_lock;
Jens Axboee94f1412019-12-19 12:06:02 -0700368
Jens Axboedef596e2019-01-09 08:59:42 -0700369 /*
Pavel Begunkov540e32a2020-07-13 23:37:09 +0300370 * ->iopoll_list is protected by the ctx->uring_lock for
Jens Axboedef596e2019-01-09 08:59:42 -0700371 * io_uring instances that don't use IORING_SETUP_SQPOLL.
372 * For SQPOLL, only the single threaded io_sq_thread() will
373 * manipulate the list, hence no extra locking is needed there.
374 */
Pavel Begunkov540e32a2020-07-13 23:37:09 +0300375 struct list_head iopoll_list;
Jens Axboe78076bb2019-12-04 19:56:40 -0700376 struct hlist_head *cancel_hash;
377 unsigned cancel_hash_bits;
Jens Axboee94f1412019-12-19 12:06:02 -0700378 bool poll_multi_file;
Jens Axboefcb323c2019-10-24 12:39:47 -0600379
380 spinlock_t inflight_lock;
381 struct list_head inflight_list;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700382 } ____cacheline_aligned_in_smp;
Jens Axboe85faa7b2020-04-09 18:14:00 -0600383
Jens Axboe4a38aed22020-05-14 17:21:15 -0600384 struct delayed_work file_put_work;
385 struct llist_head file_put_llist;
386
Jens Axboe85faa7b2020-04-09 18:14:00 -0600387 struct work_struct exit_work;
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200388 struct io_restriction restrictions;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700389};
390
Jens Axboe09bb8392019-03-13 12:39:28 -0600391/*
392 * First field must be the file pointer in all the
393 * iocb unions! See also 'struct kiocb' in <linux/fs.h>
394 */
Jens Axboe221c5eb2019-01-17 09:41:58 -0700395struct io_poll_iocb {
396 struct file *file;
Jens Axboe0969e782019-12-17 18:40:57 -0700397 union {
398 struct wait_queue_head *head;
399 u64 addr;
400 };
Jens Axboe221c5eb2019-01-17 09:41:58 -0700401 __poll_t events;
Jens Axboe8c838782019-03-12 15:48:16 -0600402 bool done;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700403 bool canceled;
Jens Axboe392edb42019-12-09 17:52:20 -0700404 struct wait_queue_entry wait;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700405};
406
Jens Axboeb5dba592019-12-11 14:02:38 -0700407struct io_close {
408 struct file *file;
409 struct file *put_file;
410 int fd;
411};
412
Jens Axboead8a48a2019-11-15 08:49:11 -0700413struct io_timeout_data {
414 struct io_kiocb *req;
415 struct hrtimer timer;
416 struct timespec64 ts;
417 enum hrtimer_mode mode;
418};
419
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700420struct io_accept {
421 struct file *file;
422 struct sockaddr __user *addr;
423 int __user *addr_len;
424 int flags;
Jens Axboe09952e32020-03-19 20:16:56 -0600425 unsigned long nofile;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700426};
427
428struct io_sync {
429 struct file *file;
430 loff_t len;
431 loff_t off;
432 int flags;
Jens Axboed63d1b52019-12-10 10:38:56 -0700433 int mode;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700434};
435
Jens Axboefbf23842019-12-17 18:45:56 -0700436struct io_cancel {
437 struct file *file;
438 u64 addr;
439};
440
Jens Axboeb29472e2019-12-17 18:50:29 -0700441struct io_timeout {
442 struct file *file;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +0300443 u32 off;
444 u32 target_seq;
Pavel Begunkov135fcde2020-07-13 23:37:12 +0300445 struct list_head list;
Jens Axboeb29472e2019-12-17 18:50:29 -0700446};
447
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100448struct io_timeout_rem {
449 struct file *file;
450 u64 addr;
451};
452
Jens Axboe9adbd452019-12-20 08:45:55 -0700453struct io_rw {
454 /* NOTE: kiocb has the file as the first member, so don't do it here */
455 struct kiocb kiocb;
456 u64 addr;
457 u64 len;
458};
459
Jens Axboe3fbb51c2019-12-20 08:51:52 -0700460struct io_connect {
461 struct file *file;
462 struct sockaddr __user *addr;
463 int addr_len;
464};
465
Jens Axboee47293f2019-12-20 08:58:21 -0700466struct io_sr_msg {
467 struct file *file;
Jens Axboefddafac2020-01-04 20:19:44 -0700468 union {
Pavel Begunkov270a5942020-07-12 20:41:04 +0300469 struct user_msghdr __user *umsg;
Jens Axboefddafac2020-01-04 20:19:44 -0700470 void __user *buf;
471 };
Jens Axboee47293f2019-12-20 08:58:21 -0700472 int msg_flags;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700473 int bgid;
Jens Axboefddafac2020-01-04 20:19:44 -0700474 size_t len;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700475 struct io_buffer *kbuf;
Jens Axboee47293f2019-12-20 08:58:21 -0700476};
477
Jens Axboe15b71ab2019-12-11 11:20:36 -0700478struct io_open {
479 struct file *file;
480 int dfd;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700481 struct filename *filename;
Jens Axboec12cedf2020-01-08 17:41:21 -0700482 struct open_how how;
Jens Axboe4022e7a2020-03-19 19:23:18 -0600483 unsigned long nofile;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700484};
485
Jens Axboe05f3fb32019-12-09 11:22:50 -0700486struct io_files_update {
487 struct file *file;
488 u64 arg;
489 u32 nr_args;
490 u32 offset;
491};
492
Jens Axboe4840e412019-12-25 22:03:45 -0700493struct io_fadvise {
494 struct file *file;
495 u64 offset;
496 u32 len;
497 u32 advice;
498};
499
Jens Axboec1ca7572019-12-25 22:18:28 -0700500struct io_madvise {
501 struct file *file;
502 u64 addr;
503 u32 len;
504 u32 advice;
505};
506
Jens Axboe3e4827b2020-01-08 15:18:09 -0700507struct io_epoll {
508 struct file *file;
509 int epfd;
510 int op;
511 int fd;
512 struct epoll_event event;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700513};
514
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300515struct io_splice {
516 struct file *file_out;
517 struct file *file_in;
518 loff_t off_out;
519 loff_t off_in;
520 u64 len;
521 unsigned int flags;
522};
523
Jens Axboeddf0322d2020-02-23 16:41:33 -0700524struct io_provide_buf {
525 struct file *file;
526 __u64 addr;
527 __s32 len;
528 __u32 bgid;
529 __u16 nbufs;
530 __u16 bid;
531};
532
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700533struct io_statx {
534 struct file *file;
535 int dfd;
536 unsigned int mask;
537 unsigned int flags;
Bijan Mottahedehe62753e2020-05-22 21:31:18 -0700538 const char __user *filename;
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700539 struct statx __user *buffer;
540};
541
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300542struct io_completion {
543 struct file *file;
544 struct list_head list;
Pavel Begunkov0f7e4662020-07-13 23:37:16 +0300545 int cflags;
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300546};
547
Jens Axboef499a022019-12-02 16:28:46 -0700548struct io_async_connect {
549 struct sockaddr_storage address;
550};
551
Jens Axboe03b12302019-12-02 18:50:25 -0700552struct io_async_msghdr {
553 struct iovec fast_iov[UIO_FASTIOV];
554 struct iovec *iov;
555 struct sockaddr __user *uaddr;
556 struct msghdr msg;
Jens Axboeb5379162020-02-09 11:29:15 -0700557 struct sockaddr_storage addr;
Jens Axboe03b12302019-12-02 18:50:25 -0700558};
559
Jens Axboef67676d2019-12-02 11:03:47 -0700560struct io_async_rw {
561 struct iovec fast_iov[UIO_FASTIOV];
Jens Axboeff6165b2020-08-13 09:47:43 -0600562 const struct iovec *free_iovec;
563 struct iov_iter iter;
Jens Axboe227c0c92020-08-13 11:51:40 -0600564 size_t bytes_done;
Jens Axboebcf5a062020-05-22 09:24:42 -0600565 struct wait_page_queue wpq;
Jens Axboef67676d2019-12-02 11:03:47 -0700566};
567
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300568enum {
569 REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT,
570 REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT,
571 REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT,
572 REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT,
573 REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700574 REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300575
Pavel Begunkovdea3b492020-04-12 02:05:04 +0300576 REQ_F_LINK_HEAD_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300577 REQ_F_FAIL_LINK_BIT,
578 REQ_F_INFLIGHT_BIT,
579 REQ_F_CUR_POS_BIT,
580 REQ_F_NOWAIT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300581 REQ_F_LINK_TIMEOUT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300582 REQ_F_ISREG_BIT,
Pavel Begunkov99bc4c32020-02-07 22:04:45 +0300583 REQ_F_NEED_CLEANUP_BIT,
Jens Axboed7718a92020-02-14 22:23:12 -0700584 REQ_F_POLLED_BIT,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700585 REQ_F_BUFFER_SELECTED_BIT,
Jens Axboe5b0bbee2020-04-27 10:41:22 -0600586 REQ_F_NO_FILE_TABLE_BIT,
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +0800587 REQ_F_WORK_INITIALIZED_BIT,
Pavel Begunkov900fad42020-10-19 16:39:16 +0100588 REQ_F_LTIMEOUT_ACTIVE_BIT,
Jens Axboe84557872020-03-03 15:28:17 -0700589
590 /* not a real bit, just to check we're not overflowing the space */
591 __REQ_F_LAST_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300592};
593
594enum {
595 /* ctx owns file */
596 REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT),
597 /* drain existing IO first */
598 REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT),
599 /* linked sqes */
600 REQ_F_LINK = BIT(REQ_F_LINK_BIT),
601 /* doesn't sever on completion < 0 */
602 REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT),
603 /* IOSQE_ASYNC */
604 REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT),
Jens Axboebcda7ba2020-02-23 16:42:51 -0700605 /* IOSQE_BUFFER_SELECT */
606 REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT),
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300607
Pavel Begunkovdea3b492020-04-12 02:05:04 +0300608 /* head of a link */
609 REQ_F_LINK_HEAD = BIT(REQ_F_LINK_HEAD_BIT),
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300610 /* fail rest of links */
611 REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT),
612 /* on inflight list */
613 REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT),
614 /* read/write uses file position */
615 REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT),
616 /* must not punt to workers */
617 REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT),
Pavel Begunkov900fad42020-10-19 16:39:16 +0100618 /* has or had linked timeout */
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300619 REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT),
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300620 /* regular file */
621 REQ_F_ISREG = BIT(REQ_F_ISREG_BIT),
Pavel Begunkov99bc4c32020-02-07 22:04:45 +0300622 /* needs cleanup */
623 REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT),
Jens Axboed7718a92020-02-14 22:23:12 -0700624 /* already went through poll handler */
625 REQ_F_POLLED = BIT(REQ_F_POLLED_BIT),
Jens Axboebcda7ba2020-02-23 16:42:51 -0700626 /* buffer already selected */
627 REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT),
Jens Axboe5b0bbee2020-04-27 10:41:22 -0600628 /* doesn't need file table for this request */
629 REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT),
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +0800630 /* io_wq_work is initialized */
631 REQ_F_WORK_INITIALIZED = BIT(REQ_F_WORK_INITIALIZED_BIT),
Pavel Begunkov900fad42020-10-19 16:39:16 +0100632 /* linked timeout is active, i.e. prepared by link's head */
633 REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT),
Jens Axboed7718a92020-02-14 22:23:12 -0700634};
635
636struct async_poll {
637 struct io_poll_iocb poll;
Jens Axboe807abcb2020-07-17 17:09:27 -0600638 struct io_poll_iocb *double_poll;
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300639};
640
Jens Axboe09bb8392019-03-13 12:39:28 -0600641/*
642 * NOTE! Each of the iocb union members has the file pointer
643 * as the first entry in their struct definition. So you can
644 * access the file pointer through any of the sub-structs,
645 * or directly as just 'ki_filp' in this struct.
646 */
Jens Axboe2b188cc2019-01-07 10:46:33 -0700647struct io_kiocb {
Jens Axboe221c5eb2019-01-17 09:41:58 -0700648 union {
Jens Axboe09bb8392019-03-13 12:39:28 -0600649 struct file *file;
Jens Axboe9adbd452019-12-20 08:45:55 -0700650 struct io_rw rw;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700651 struct io_poll_iocb poll;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700652 struct io_accept accept;
653 struct io_sync sync;
Jens Axboefbf23842019-12-17 18:45:56 -0700654 struct io_cancel cancel;
Jens Axboeb29472e2019-12-17 18:50:29 -0700655 struct io_timeout timeout;
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100656 struct io_timeout_rem timeout_rem;
Jens Axboe3fbb51c2019-12-20 08:51:52 -0700657 struct io_connect connect;
Jens Axboee47293f2019-12-20 08:58:21 -0700658 struct io_sr_msg sr_msg;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700659 struct io_open open;
Jens Axboeb5dba592019-12-11 14:02:38 -0700660 struct io_close close;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700661 struct io_files_update files_update;
Jens Axboe4840e412019-12-25 22:03:45 -0700662 struct io_fadvise fadvise;
Jens Axboec1ca7572019-12-25 22:18:28 -0700663 struct io_madvise madvise;
Jens Axboe3e4827b2020-01-08 15:18:09 -0700664 struct io_epoll epoll;
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300665 struct io_splice splice;
Jens Axboeddf0322d2020-02-23 16:41:33 -0700666 struct io_provide_buf pbuf;
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700667 struct io_statx statx;
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300668 /* use only after cleaning per-op data, see io_clean_op() */
669 struct io_completion compl;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700670 };
Jens Axboe2b188cc2019-01-07 10:46:33 -0700671
Jens Axboee8c2bc12020-08-15 18:44:09 -0700672 /* opcode allocated if it needs to store data for async defer */
673 void *async_data;
Jens Axboed625c6e2019-12-17 19:53:05 -0700674 u8 opcode;
Xiaoguang Wang65a65432020-06-11 23:39:36 +0800675 /* polled IO has completed */
676 u8 iopoll_completed;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700677
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -0700678 u16 buf_index;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +0300679 u32 result;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -0700680
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300681 struct io_ring_ctx *ctx;
682 unsigned int flags;
683 refcount_t refs;
684 struct task_struct *task;
685 u64 user_data;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700686
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300687 struct list_head link_list;
Jens Axboed7718a92020-02-14 22:23:12 -0700688
Pavel Begunkovd21ffe72020-07-13 23:37:10 +0300689 /*
690 * 1. used with ctx->iopoll_list with reads/writes
691 * 2. to track reqs with ->files (see io_op_def::file_table)
692 */
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300693 struct list_head inflight_entry;
Jens Axboefcb323c2019-10-24 12:39:47 -0600694
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300695 struct percpu_ref *fixed_file_refs;
696 struct callback_head task_work;
697 /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */
698 struct hlist_node hash_node;
699 struct async_poll *apoll;
700 struct io_wq_work work;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700701};
702
Pavel Begunkov27dc8332020-07-13 23:37:14 +0300703struct io_defer_entry {
704 struct list_head list;
705 struct io_kiocb *req;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +0300706 u32 seq;
Pavel Begunkov27dc8332020-07-13 23:37:14 +0300707};
708
Jens Axboedef596e2019-01-09 08:59:42 -0700709#define IO_IOPOLL_BATCH 8
Jens Axboe2b188cc2019-01-07 10:46:33 -0700710
Jens Axboe013538b2020-06-22 09:29:15 -0600711struct io_comp_state {
712 unsigned int nr;
713 struct list_head list;
714 struct io_ring_ctx *ctx;
715};
716
Jens Axboe9a56a232019-01-09 09:06:50 -0700717struct io_submit_state {
718 struct blk_plug plug;
719
720 /*
Jens Axboe2579f912019-01-09 09:10:43 -0700721 * io_kiocb alloc cache
722 */
723 void *reqs[IO_IOPOLL_BATCH];
Pavel Begunkov6c8a3132020-02-01 03:58:00 +0300724 unsigned int free_reqs;
Jens Axboe2579f912019-01-09 09:10:43 -0700725
726 /*
Jens Axboe013538b2020-06-22 09:29:15 -0600727 * Batch completion logic
728 */
729 struct io_comp_state comp;
730
731 /*
Jens Axboe9a56a232019-01-09 09:06:50 -0700732 * File reference cache
733 */
734 struct file *file;
735 unsigned int fd;
736 unsigned int has_refs;
Jens Axboe9a56a232019-01-09 09:06:50 -0700737 unsigned int ios_left;
738};
739
Jens Axboed3656342019-12-18 09:50:26 -0700740struct io_op_def {
Jens Axboed3656342019-12-18 09:50:26 -0700741 /* needs req->file assigned */
742 unsigned needs_file : 1;
Jens Axboefd2206e2020-06-02 16:40:47 -0600743 /* don't fail if file grab fails */
744 unsigned needs_file_no_error : 1;
Jens Axboed3656342019-12-18 09:50:26 -0700745 /* hash wq insertion if file is a regular file */
746 unsigned hash_reg_file : 1;
747 /* unbound wq insertion if file is a non-regular file */
748 unsigned unbound_nonreg_file : 1;
Jens Axboe66f4af92020-01-16 15:36:52 -0700749 /* opcode is not supported by this kernel */
750 unsigned not_supported : 1;
Jens Axboe8a727582020-02-20 09:59:44 -0700751 /* set if opcode supports polled "wait" */
752 unsigned pollin : 1;
753 unsigned pollout : 1;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700754 /* op supports buffer selection */
755 unsigned buffer_select : 1;
Jens Axboee8c2bc12020-08-15 18:44:09 -0700756 /* must always have async data allocated */
757 unsigned needs_async_data : 1;
758 /* size of async data needed, if any */
759 unsigned short async_size;
Jens Axboe0f203762020-10-14 09:23:55 -0600760 unsigned work_flags;
Jens Axboed3656342019-12-18 09:50:26 -0700761};
762
Jens Axboe09186822020-10-13 15:01:40 -0600763static const struct io_op_def io_op_defs[] = {
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300764 [IORING_OP_NOP] = {},
765 [IORING_OP_READV] = {
Jens Axboed3656342019-12-18 09:50:26 -0700766 .needs_file = 1,
767 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700768 .pollin = 1,
Jens Axboe4d954c22020-02-27 07:31:19 -0700769 .buffer_select = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700770 .needs_async_data = 1,
771 .async_size = sizeof(struct io_async_rw),
Jens Axboe0f203762020-10-14 09:23:55 -0600772 .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG,
Jens Axboed3656342019-12-18 09:50:26 -0700773 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300774 [IORING_OP_WRITEV] = {
Jens Axboed3656342019-12-18 09:50:26 -0700775 .needs_file = 1,
776 .hash_reg_file = 1,
777 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700778 .pollout = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700779 .needs_async_data = 1,
780 .async_size = sizeof(struct io_async_rw),
Jens Axboe69228332020-10-20 14:28:41 -0600781 .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG |
782 IO_WQ_WORK_FSIZE,
Jens Axboed3656342019-12-18 09:50:26 -0700783 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300784 [IORING_OP_FSYNC] = {
Jens Axboed3656342019-12-18 09:50:26 -0700785 .needs_file = 1,
Jens Axboe0f203762020-10-14 09:23:55 -0600786 .work_flags = IO_WQ_WORK_BLKCG,
Jens Axboed3656342019-12-18 09:50:26 -0700787 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300788 [IORING_OP_READ_FIXED] = {
Jens Axboed3656342019-12-18 09:50:26 -0700789 .needs_file = 1,
790 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700791 .pollin = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700792 .async_size = sizeof(struct io_async_rw),
Jens Axboe4017eb92020-10-22 14:14:12 -0600793 .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_MM,
Jens Axboed3656342019-12-18 09:50:26 -0700794 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300795 [IORING_OP_WRITE_FIXED] = {
Jens Axboed3656342019-12-18 09:50:26 -0700796 .needs_file = 1,
797 .hash_reg_file = 1,
798 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700799 .pollout = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700800 .async_size = sizeof(struct io_async_rw),
Jens Axboe4017eb92020-10-22 14:14:12 -0600801 .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_FSIZE |
802 IO_WQ_WORK_MM,
Jens Axboed3656342019-12-18 09:50:26 -0700803 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300804 [IORING_OP_POLL_ADD] = {
Jens Axboed3656342019-12-18 09:50:26 -0700805 .needs_file = 1,
806 .unbound_nonreg_file = 1,
807 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300808 [IORING_OP_POLL_REMOVE] = {},
809 [IORING_OP_SYNC_FILE_RANGE] = {
Jens Axboed3656342019-12-18 09:50:26 -0700810 .needs_file = 1,
Jens Axboe0f203762020-10-14 09:23:55 -0600811 .work_flags = IO_WQ_WORK_BLKCG,
Jens Axboed3656342019-12-18 09:50:26 -0700812 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300813 [IORING_OP_SENDMSG] = {
Jens Axboed3656342019-12-18 09:50:26 -0700814 .needs_file = 1,
815 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700816 .pollout = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700817 .needs_async_data = 1,
818 .async_size = sizeof(struct io_async_msghdr),
Jens Axboe0f203762020-10-14 09:23:55 -0600819 .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG |
820 IO_WQ_WORK_FS,
Jens Axboed3656342019-12-18 09:50:26 -0700821 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300822 [IORING_OP_RECVMSG] = {
Jens Axboed3656342019-12-18 09:50:26 -0700823 .needs_file = 1,
824 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700825 .pollin = 1,
Jens Axboe52de1fe2020-02-27 10:15:42 -0700826 .buffer_select = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700827 .needs_async_data = 1,
828 .async_size = sizeof(struct io_async_msghdr),
Jens Axboe0f203762020-10-14 09:23:55 -0600829 .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG |
830 IO_WQ_WORK_FS,
Jens Axboed3656342019-12-18 09:50:26 -0700831 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300832 [IORING_OP_TIMEOUT] = {
Jens Axboee8c2bc12020-08-15 18:44:09 -0700833 .needs_async_data = 1,
834 .async_size = sizeof(struct io_timeout_data),
Jens Axboe0f203762020-10-14 09:23:55 -0600835 .work_flags = IO_WQ_WORK_MM,
Jens Axboed3656342019-12-18 09:50:26 -0700836 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300837 [IORING_OP_TIMEOUT_REMOVE] = {},
838 [IORING_OP_ACCEPT] = {
Jens Axboed3656342019-12-18 09:50:26 -0700839 .needs_file = 1,
840 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700841 .pollin = 1,
Jens Axboe0f203762020-10-14 09:23:55 -0600842 .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_FILES,
Jens Axboed3656342019-12-18 09:50:26 -0700843 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300844 [IORING_OP_ASYNC_CANCEL] = {},
845 [IORING_OP_LINK_TIMEOUT] = {
Jens Axboee8c2bc12020-08-15 18:44:09 -0700846 .needs_async_data = 1,
847 .async_size = sizeof(struct io_timeout_data),
Jens Axboe0f203762020-10-14 09:23:55 -0600848 .work_flags = IO_WQ_WORK_MM,
Jens Axboed3656342019-12-18 09:50:26 -0700849 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300850 [IORING_OP_CONNECT] = {
Jens Axboed3656342019-12-18 09:50:26 -0700851 .needs_file = 1,
852 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700853 .pollout = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700854 .needs_async_data = 1,
855 .async_size = sizeof(struct io_async_connect),
Jens Axboe0f203762020-10-14 09:23:55 -0600856 .work_flags = IO_WQ_WORK_MM,
Jens Axboed3656342019-12-18 09:50:26 -0700857 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300858 [IORING_OP_FALLOCATE] = {
Jens Axboed3656342019-12-18 09:50:26 -0700859 .needs_file = 1,
Jens Axboe69228332020-10-20 14:28:41 -0600860 .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_FSIZE,
Jens Axboed3656342019-12-18 09:50:26 -0700861 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300862 [IORING_OP_OPENAT] = {
Jens Axboe0f203762020-10-14 09:23:55 -0600863 .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_BLKCG |
864 IO_WQ_WORK_FS,
Jens Axboed3656342019-12-18 09:50:26 -0700865 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300866 [IORING_OP_CLOSE] = {
Jens Axboefd2206e2020-06-02 16:40:47 -0600867 .needs_file = 1,
868 .needs_file_no_error = 1,
Jens Axboe0f203762020-10-14 09:23:55 -0600869 .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_BLKCG,
Jens Axboed3656342019-12-18 09:50:26 -0700870 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300871 [IORING_OP_FILES_UPDATE] = {
Jens Axboe0f203762020-10-14 09:23:55 -0600872 .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_MM,
Jens Axboed3656342019-12-18 09:50:26 -0700873 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300874 [IORING_OP_STATX] = {
Jens Axboe0f203762020-10-14 09:23:55 -0600875 .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_MM |
876 IO_WQ_WORK_FS | IO_WQ_WORK_BLKCG,
Jens Axboed3656342019-12-18 09:50:26 -0700877 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300878 [IORING_OP_READ] = {
Jens Axboe3a6820f2019-12-22 15:19:35 -0700879 .needs_file = 1,
880 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700881 .pollin = 1,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700882 .buffer_select = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700883 .async_size = sizeof(struct io_async_rw),
Jens Axboe0f203762020-10-14 09:23:55 -0600884 .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG,
Jens Axboe3a6820f2019-12-22 15:19:35 -0700885 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300886 [IORING_OP_WRITE] = {
Jens Axboe3a6820f2019-12-22 15:19:35 -0700887 .needs_file = 1,
888 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700889 .pollout = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700890 .async_size = sizeof(struct io_async_rw),
Jens Axboe69228332020-10-20 14:28:41 -0600891 .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG |
892 IO_WQ_WORK_FSIZE,
Jens Axboe3a6820f2019-12-22 15:19:35 -0700893 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300894 [IORING_OP_FADVISE] = {
Jens Axboe4840e412019-12-25 22:03:45 -0700895 .needs_file = 1,
Jens Axboe0f203762020-10-14 09:23:55 -0600896 .work_flags = IO_WQ_WORK_BLKCG,
Jens Axboe4840e412019-12-25 22:03:45 -0700897 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300898 [IORING_OP_MADVISE] = {
Jens Axboe0f203762020-10-14 09:23:55 -0600899 .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG,
Jens Axboec1ca7572019-12-25 22:18:28 -0700900 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300901 [IORING_OP_SEND] = {
Jens Axboefddafac2020-01-04 20:19:44 -0700902 .needs_file = 1,
903 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700904 .pollout = 1,
Jens Axboe0f203762020-10-14 09:23:55 -0600905 .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG,
Jens Axboefddafac2020-01-04 20:19:44 -0700906 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300907 [IORING_OP_RECV] = {
Jens Axboefddafac2020-01-04 20:19:44 -0700908 .needs_file = 1,
909 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700910 .pollin = 1,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700911 .buffer_select = 1,
Jens Axboe0f203762020-10-14 09:23:55 -0600912 .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG,
Jens Axboefddafac2020-01-04 20:19:44 -0700913 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300914 [IORING_OP_OPENAT2] = {
Jens Axboe0f203762020-10-14 09:23:55 -0600915 .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_FS |
916 IO_WQ_WORK_BLKCG,
Jens Axboecebdb982020-01-08 17:59:24 -0700917 },
Jens Axboe3e4827b2020-01-08 15:18:09 -0700918 [IORING_OP_EPOLL_CTL] = {
919 .unbound_nonreg_file = 1,
Jens Axboe0f203762020-10-14 09:23:55 -0600920 .work_flags = IO_WQ_WORK_FILES,
Jens Axboe3e4827b2020-01-08 15:18:09 -0700921 },
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300922 [IORING_OP_SPLICE] = {
923 .needs_file = 1,
924 .hash_reg_file = 1,
925 .unbound_nonreg_file = 1,
Jens Axboe0f203762020-10-14 09:23:55 -0600926 .work_flags = IO_WQ_WORK_BLKCG,
Jens Axboeddf0322d2020-02-23 16:41:33 -0700927 },
928 [IORING_OP_PROVIDE_BUFFERS] = {},
Jens Axboe067524e2020-03-02 16:32:28 -0700929 [IORING_OP_REMOVE_BUFFERS] = {},
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +0300930 [IORING_OP_TEE] = {
931 .needs_file = 1,
932 .hash_reg_file = 1,
933 .unbound_nonreg_file = 1,
934 },
Jens Axboed3656342019-12-18 09:50:26 -0700935};
936
Bijan Mottahedeh2e0464d2020-06-16 16:36:10 -0700937enum io_mem_account {
938 ACCT_LOCKED,
939 ACCT_PINNED,
940};
941
Pavel Begunkov81b68a52020-07-30 18:43:46 +0300942static void __io_complete_rw(struct io_kiocb *req, long res, long res2,
943 struct io_comp_state *cs);
Jens Axboe78e19bb2019-11-06 15:21:34 -0700944static void io_cqring_fill_event(struct io_kiocb *req, long res);
Jackie Liuec9c02a2019-11-08 23:50:36 +0800945static void io_put_req(struct io_kiocb *req);
Pavel Begunkov216578e2020-10-13 09:44:00 +0100946static void io_put_req_deferred(struct io_kiocb *req, int nr);
Jens Axboec40f6372020-06-25 15:39:59 -0600947static void io_double_put_req(struct io_kiocb *req);
Jens Axboe94ae5e72019-11-14 19:39:52 -0700948static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req);
Jens Axboe7271ef32020-08-10 09:55:22 -0600949static void __io_queue_linked_timeout(struct io_kiocb *req);
Jens Axboe94ae5e72019-11-14 19:39:52 -0700950static void io_queue_linked_timeout(struct io_kiocb *req);
Jens Axboe05f3fb32019-12-09 11:22:50 -0700951static int __io_sqe_files_update(struct io_ring_ctx *ctx,
952 struct io_uring_files_update *ip,
953 unsigned nr_args);
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300954static void __io_clean_op(struct io_kiocb *req);
Pavel Begunkov8371adf2020-10-10 18:34:08 +0100955static struct file *io_file_get(struct io_submit_state *state,
956 struct io_kiocb *req, int fd, bool fixed);
Pavel Begunkovc1379e22020-09-30 22:57:56 +0300957static void __io_queue_sqe(struct io_kiocb *req, struct io_comp_state *cs);
Jens Axboe4349f302020-07-09 15:07:01 -0600958static void io_file_put_work(struct work_struct *work);
Jens Axboede0617e2019-04-06 21:51:27 -0600959
Jens Axboeb63534c2020-06-04 11:28:00 -0600960static ssize_t io_import_iovec(int rw, struct io_kiocb *req,
961 struct iovec **iovec, struct iov_iter *iter,
962 bool needs_lock);
Jens Axboeff6165b2020-08-13 09:47:43 -0600963static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec,
964 const struct iovec *fast_iov,
Jens Axboe227c0c92020-08-13 11:51:40 -0600965 struct iov_iter *iter, bool force);
Jens Axboe2b188cc2019-01-07 10:46:33 -0700966
967static struct kmem_cache *req_cachep;
968
Jens Axboe09186822020-10-13 15:01:40 -0600969static const struct file_operations io_uring_fops;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700970
971struct sock *io_uring_get_socket(struct file *file)
972{
973#if defined(CONFIG_UNIX)
974 if (file->f_op == &io_uring_fops) {
975 struct io_ring_ctx *ctx = file->private_data;
976
977 return ctx->ring_sock->sk;
978 }
979#endif
980 return NULL;
981}
982EXPORT_SYMBOL(io_uring_get_socket);
983
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300984static inline void io_clean_op(struct io_kiocb *req)
985{
Pavel Begunkovbb175342020-08-20 11:33:35 +0300986 if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED |
987 REQ_F_INFLIGHT))
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300988 __io_clean_op(req);
989}
990
Jens Axboe4349f302020-07-09 15:07:01 -0600991static void io_sq_thread_drop_mm(void)
Jens Axboec40f6372020-06-25 15:39:59 -0600992{
993 struct mm_struct *mm = current->mm;
994
995 if (mm) {
996 kthread_unuse_mm(mm);
997 mmput(mm);
Jens Axboe4b70cf92020-11-02 10:39:05 -0700998 current->mm = NULL;
Jens Axboec40f6372020-06-25 15:39:59 -0600999 }
1000}
1001
1002static int __io_sq_thread_acquire_mm(struct io_ring_ctx *ctx)
1003{
Jens Axboe4b70cf92020-11-02 10:39:05 -07001004 struct mm_struct *mm;
1005
1006 if (current->mm)
1007 return 0;
1008
1009 /* Should never happen */
1010 if (unlikely(!(ctx->flags & IORING_SETUP_SQPOLL)))
1011 return -EFAULT;
1012
1013 task_lock(ctx->sqo_task);
1014 mm = ctx->sqo_task->mm;
1015 if (unlikely(!mm || !mmget_not_zero(mm)))
1016 mm = NULL;
1017 task_unlock(ctx->sqo_task);
1018
1019 if (mm) {
1020 kthread_use_mm(mm);
1021 return 0;
Jens Axboec40f6372020-06-25 15:39:59 -06001022 }
1023
Jens Axboe4b70cf92020-11-02 10:39:05 -07001024 return -EFAULT;
Jens Axboec40f6372020-06-25 15:39:59 -06001025}
1026
1027static int io_sq_thread_acquire_mm(struct io_ring_ctx *ctx,
1028 struct io_kiocb *req)
1029{
Jens Axboe0f203762020-10-14 09:23:55 -06001030 if (!(io_op_defs[req->opcode].work_flags & IO_WQ_WORK_MM))
Jens Axboec40f6372020-06-25 15:39:59 -06001031 return 0;
1032 return __io_sq_thread_acquire_mm(ctx);
1033}
1034
Dennis Zhou91d8f512020-09-16 13:41:05 -07001035static void io_sq_thread_associate_blkcg(struct io_ring_ctx *ctx,
1036 struct cgroup_subsys_state **cur_css)
1037
1038{
1039#ifdef CONFIG_BLK_CGROUP
1040 /* puts the old one when swapping */
1041 if (*cur_css != ctx->sqo_blkcg_css) {
1042 kthread_associate_blkcg(ctx->sqo_blkcg_css);
1043 *cur_css = ctx->sqo_blkcg_css;
1044 }
1045#endif
1046}
1047
1048static void io_sq_thread_unassociate_blkcg(void)
1049{
1050#ifdef CONFIG_BLK_CGROUP
1051 kthread_associate_blkcg(NULL);
1052#endif
1053}
1054
Jens Axboec40f6372020-06-25 15:39:59 -06001055static inline void req_set_fail_links(struct io_kiocb *req)
1056{
1057 if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK)
1058 req->flags |= REQ_F_FAIL_LINK;
1059}
Jens Axboe4a38aed22020-05-14 17:21:15 -06001060
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08001061/*
Jens Axboe1e6fa522020-10-15 08:46:24 -06001062 * None of these are dereferenced, they are simply used to check if any of
1063 * them have changed. If we're under current and check they are still the
1064 * same, we're fine to grab references to them for actual out-of-line use.
1065 */
1066static void io_init_identity(struct io_identity *id)
1067{
1068 id->files = current->files;
1069 id->mm = current->mm;
1070#ifdef CONFIG_BLK_CGROUP
1071 rcu_read_lock();
1072 id->blkcg_css = blkcg_css();
1073 rcu_read_unlock();
1074#endif
1075 id->creds = current_cred();
1076 id->nsproxy = current->nsproxy;
1077 id->fs = current->fs;
1078 id->fsize = rlimit(RLIMIT_FSIZE);
Jens Axboe4ea33a92020-10-15 13:46:44 -06001079#ifdef CONFIG_AUDIT
1080 id->loginuid = current->loginuid;
1081 id->sessionid = current->sessionid;
1082#endif
Jens Axboe1e6fa522020-10-15 08:46:24 -06001083 refcount_set(&id->count, 1);
1084}
1085
Pavel Begunkovec99ca62020-10-18 10:17:38 +01001086static inline void __io_req_init_async(struct io_kiocb *req)
1087{
1088 memset(&req->work, 0, sizeof(req->work));
1089 req->flags |= REQ_F_WORK_INITIALIZED;
1090}
1091
Jens Axboe1e6fa522020-10-15 08:46:24 -06001092/*
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08001093 * Note: must call io_req_init_async() for the first time you
1094 * touch any members of io_wq_work.
1095 */
1096static inline void io_req_init_async(struct io_kiocb *req)
1097{
Jens Axboe500a3732020-10-15 17:38:03 -06001098 struct io_uring_task *tctx = current->io_uring;
1099
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08001100 if (req->flags & REQ_F_WORK_INITIALIZED)
1101 return;
1102
Pavel Begunkovec99ca62020-10-18 10:17:38 +01001103 __io_req_init_async(req);
Jens Axboe500a3732020-10-15 17:38:03 -06001104
1105 /* Grab a ref if this isn't our static identity */
1106 req->work.identity = tctx->identity;
1107 if (tctx->identity != &tctx->__identity)
1108 refcount_inc(&req->work.identity->count);
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08001109}
1110
Pavel Begunkov0cdaf762020-05-17 14:13:40 +03001111static inline bool io_async_submit(struct io_ring_ctx *ctx)
1112{
1113 return ctx->flags & IORING_SETUP_SQPOLL;
1114}
1115
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 Axboe0ddf92e2019-11-08 08:52:53 -07001137 ctx->fallback_req = kmem_cache_alloc(req_cachep, GFP_KERNEL);
1138 if (!ctx->fallback_req)
1139 goto err;
1140
Jens Axboe78076bb2019-12-04 19:56:40 -07001141 /*
1142 * Use 5 bits less than the max cq entries, that should give us around
1143 * 32 entries per hash list if totally full and uniformly spread.
1144 */
1145 hash_bits = ilog2(p->cq_entries);
1146 hash_bits -= 5;
1147 if (hash_bits <= 0)
1148 hash_bits = 1;
1149 ctx->cancel_hash_bits = hash_bits;
1150 ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head),
1151 GFP_KERNEL);
1152 if (!ctx->cancel_hash)
1153 goto err;
1154 __hash_init(ctx->cancel_hash, 1U << hash_bits);
1155
Roman Gushchin21482892019-05-07 10:01:48 -07001156 if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free,
Jens Axboe206aefd2019-11-07 18:27:42 -07001157 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL))
1158 goto err;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001159
1160 ctx->flags = p->flags;
Jens Axboe90554202020-09-03 12:12:41 -06001161 init_waitqueue_head(&ctx->sqo_sq_wait);
Jens Axboe69fb2132020-09-14 11:16:23 -06001162 INIT_LIST_HEAD(&ctx->sqd_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001163 init_waitqueue_head(&ctx->cq_wait);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001164 INIT_LIST_HEAD(&ctx->cq_overflow_list);
Jens Axboe0f158b42020-05-14 17:18:39 -06001165 init_completion(&ctx->ref_comp);
1166 init_completion(&ctx->sq_thread_comp);
Jens Axboe5a2e7452020-02-23 16:23:11 -07001167 idr_init(&ctx->io_buffer_idr);
Jens Axboe071698e2020-01-28 10:04:42 -07001168 idr_init(&ctx->personality_idr);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001169 mutex_init(&ctx->uring_lock);
1170 init_waitqueue_head(&ctx->wait);
1171 spin_lock_init(&ctx->completion_lock);
Pavel Begunkov540e32a2020-07-13 23:37:09 +03001172 INIT_LIST_HEAD(&ctx->iopoll_list);
Jens Axboede0617e2019-04-06 21:51:27 -06001173 INIT_LIST_HEAD(&ctx->defer_list);
Jens Axboe5262f562019-09-17 12:26:57 -06001174 INIT_LIST_HEAD(&ctx->timeout_list);
Jens Axboefcb323c2019-10-24 12:39:47 -06001175 init_waitqueue_head(&ctx->inflight_wait);
1176 spin_lock_init(&ctx->inflight_lock);
1177 INIT_LIST_HEAD(&ctx->inflight_list);
Jens Axboe4a38aed22020-05-14 17:21:15 -06001178 INIT_DELAYED_WORK(&ctx->file_put_work, io_file_put_work);
1179 init_llist_head(&ctx->file_put_llist);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001180 return ctx;
Jens Axboe206aefd2019-11-07 18:27:42 -07001181err:
Jens Axboe0ddf92e2019-11-08 08:52:53 -07001182 if (ctx->fallback_req)
1183 kmem_cache_free(req_cachep, ctx->fallback_req);
Jens Axboe78076bb2019-12-04 19:56:40 -07001184 kfree(ctx->cancel_hash);
Jens Axboe206aefd2019-11-07 18:27:42 -07001185 kfree(ctx);
1186 return NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001187}
1188
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001189static bool req_need_defer(struct io_kiocb *req, u32 seq)
Jens Axboede0617e2019-04-06 21:51:27 -06001190{
Jens Axboe2bc99302020-07-09 09:43:27 -06001191 if (unlikely(req->flags & REQ_F_IO_DRAIN)) {
1192 struct io_ring_ctx *ctx = req->ctx;
Jackie Liua197f662019-11-08 08:09:12 -07001193
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001194 return seq != ctx->cached_cq_tail
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001195 + READ_ONCE(ctx->cached_cq_overflow);
Jens Axboe2bc99302020-07-09 09:43:27 -06001196 }
Jens Axboe7adf4ea2019-10-10 21:42:58 -06001197
Bob Liu9d858b22019-11-13 18:06:25 +08001198 return false;
Jens Axboe7adf4ea2019-10-10 21:42:58 -06001199}
1200
Jens Axboede0617e2019-04-06 21:51:27 -06001201static void __io_commit_cqring(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001202{
Hristo Venev75b28af2019-08-26 17:23:46 +00001203 struct io_rings *rings = ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001204
Pavel Begunkov07910152020-01-17 03:52:46 +03001205 /* order cqe stores with ring update */
1206 smp_store_release(&rings->cq.tail, ctx->cached_cq_tail);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001207
Pavel Begunkov07910152020-01-17 03:52:46 +03001208 if (wq_has_sleeper(&ctx->cq_wait)) {
1209 wake_up_interruptible(&ctx->cq_wait);
1210 kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001211 }
1212}
1213
Jens Axboe5c3462c2020-10-15 09:02:33 -06001214static void io_put_identity(struct io_uring_task *tctx, struct io_kiocb *req)
Jens Axboe1e6fa522020-10-15 08:46:24 -06001215{
Jens Axboe500a3732020-10-15 17:38:03 -06001216 if (req->work.identity == &tctx->__identity)
Jens Axboe1e6fa522020-10-15 08:46:24 -06001217 return;
1218 if (refcount_dec_and_test(&req->work.identity->count))
1219 kfree(req->work.identity);
1220}
1221
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01001222static void io_req_clean_work(struct io_kiocb *req)
Jens Axboecccf0ee2020-01-27 16:34:48 -07001223{
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08001224 if (!(req->flags & REQ_F_WORK_INITIALIZED))
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01001225 return;
Jens Axboe51a4cc12020-08-10 10:55:56 -06001226
1227 req->flags &= ~REQ_F_WORK_INITIALIZED;
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08001228
Jens Axboedfead8a2020-10-14 10:12:37 -06001229 if (req->work.flags & IO_WQ_WORK_MM) {
Jens Axboe98447d62020-10-14 10:48:51 -06001230 mmdrop(req->work.identity->mm);
Jens Axboedfead8a2020-10-14 10:12:37 -06001231 req->work.flags &= ~IO_WQ_WORK_MM;
Jens Axboecccf0ee2020-01-27 16:34:48 -07001232 }
Dennis Zhou91d8f512020-09-16 13:41:05 -07001233#ifdef CONFIG_BLK_CGROUP
Jens Axboedfead8a2020-10-14 10:12:37 -06001234 if (req->work.flags & IO_WQ_WORK_BLKCG) {
Jens Axboe98447d62020-10-14 10:48:51 -06001235 css_put(req->work.identity->blkcg_css);
Jens Axboedfead8a2020-10-14 10:12:37 -06001236 req->work.flags &= ~IO_WQ_WORK_BLKCG;
Jens Axboecccf0ee2020-01-27 16:34:48 -07001237 }
Jens Axboedfead8a2020-10-14 10:12:37 -06001238#endif
1239 if (req->work.flags & IO_WQ_WORK_CREDS) {
Jens Axboe98447d62020-10-14 10:48:51 -06001240 put_cred(req->work.identity->creds);
Jens Axboedfead8a2020-10-14 10:12:37 -06001241 req->work.flags &= ~IO_WQ_WORK_CREDS;
1242 }
1243 if (req->work.flags & IO_WQ_WORK_FS) {
Jens Axboe98447d62020-10-14 10:48:51 -06001244 struct fs_struct *fs = req->work.identity->fs;
Jens Axboeff002b32020-02-07 16:05:21 -07001245
Jens Axboe98447d62020-10-14 10:48:51 -06001246 spin_lock(&req->work.identity->fs->lock);
Jens Axboeff002b32020-02-07 16:05:21 -07001247 if (--fs->users)
1248 fs = NULL;
Jens Axboe98447d62020-10-14 10:48:51 -06001249 spin_unlock(&req->work.identity->fs->lock);
Jens Axboeff002b32020-02-07 16:05:21 -07001250 if (fs)
1251 free_fs_struct(fs);
Jens Axboedfead8a2020-10-14 10:12:37 -06001252 req->work.flags &= ~IO_WQ_WORK_FS;
Jens Axboeff002b32020-02-07 16:05:21 -07001253 }
Jens Axboe51a4cc12020-08-10 10:55:56 -06001254
Jens Axboe5c3462c2020-10-15 09:02:33 -06001255 io_put_identity(req->task->io_uring, req);
Jens Axboe1e6fa522020-10-15 08:46:24 -06001256}
1257
1258/*
1259 * Create a private copy of io_identity, since some fields don't match
1260 * the current context.
1261 */
1262static bool io_identity_cow(struct io_kiocb *req)
1263{
Jens Axboe5c3462c2020-10-15 09:02:33 -06001264 struct io_uring_task *tctx = current->io_uring;
Jens Axboe1e6fa522020-10-15 08:46:24 -06001265 const struct cred *creds = NULL;
1266 struct io_identity *id;
1267
1268 if (req->work.flags & IO_WQ_WORK_CREDS)
1269 creds = req->work.identity->creds;
1270
1271 id = kmemdup(req->work.identity, sizeof(*id), GFP_KERNEL);
1272 if (unlikely(!id)) {
1273 req->work.flags |= IO_WQ_WORK_CANCEL;
1274 return false;
1275 }
1276
1277 /*
1278 * We can safely just re-init the creds we copied Either the field
1279 * matches the current one, or we haven't grabbed it yet. The only
1280 * exception is ->creds, through registered personalities, so handle
1281 * that one separately.
1282 */
1283 io_init_identity(id);
1284 if (creds)
1285 req->work.identity->creds = creds;
1286
1287 /* add one for this request */
1288 refcount_inc(&id->count);
1289
Jens Axboecb8a8ae2020-11-03 12:19:07 -07001290 /* drop tctx and req identity references, if needed */
1291 if (tctx->identity != &tctx->__identity &&
1292 refcount_dec_and_test(&tctx->identity->count))
1293 kfree(tctx->identity);
1294 if (req->work.identity != &tctx->__identity &&
1295 refcount_dec_and_test(&req->work.identity->count))
Jens Axboe1e6fa522020-10-15 08:46:24 -06001296 kfree(req->work.identity);
1297
1298 req->work.identity = id;
Jens Axboe500a3732020-10-15 17:38:03 -06001299 tctx->identity = id;
Jens Axboe1e6fa522020-10-15 08:46:24 -06001300 return true;
1301}
1302
1303static bool io_grab_identity(struct io_kiocb *req)
1304{
1305 const struct io_op_def *def = &io_op_defs[req->opcode];
Jens Axboe5c3462c2020-10-15 09:02:33 -06001306 struct io_identity *id = req->work.identity;
Jens Axboe1e6fa522020-10-15 08:46:24 -06001307 struct io_ring_ctx *ctx = req->ctx;
1308
Jens Axboe69228332020-10-20 14:28:41 -06001309 if (def->work_flags & IO_WQ_WORK_FSIZE) {
1310 if (id->fsize != rlimit(RLIMIT_FSIZE))
1311 return false;
1312 req->work.flags |= IO_WQ_WORK_FSIZE;
1313 }
Jens Axboe1e6fa522020-10-15 08:46:24 -06001314
1315 if (!(req->work.flags & IO_WQ_WORK_FILES) &&
1316 (def->work_flags & IO_WQ_WORK_FILES) &&
1317 !(req->flags & REQ_F_NO_FILE_TABLE)) {
1318 if (id->files != current->files ||
1319 id->nsproxy != current->nsproxy)
1320 return false;
1321 atomic_inc(&id->files->count);
1322 get_nsproxy(id->nsproxy);
1323 req->flags |= REQ_F_INFLIGHT;
1324
1325 spin_lock_irq(&ctx->inflight_lock);
1326 list_add(&req->inflight_entry, &ctx->inflight_list);
1327 spin_unlock_irq(&ctx->inflight_lock);
1328 req->work.flags |= IO_WQ_WORK_FILES;
1329 }
1330#ifdef CONFIG_BLK_CGROUP
1331 if (!(req->work.flags & IO_WQ_WORK_BLKCG) &&
1332 (def->work_flags & IO_WQ_WORK_BLKCG)) {
1333 rcu_read_lock();
1334 if (id->blkcg_css != blkcg_css()) {
1335 rcu_read_unlock();
1336 return false;
1337 }
1338 /*
1339 * This should be rare, either the cgroup is dying or the task
1340 * is moving cgroups. Just punt to root for the handful of ios.
1341 */
1342 if (css_tryget_online(id->blkcg_css))
1343 req->work.flags |= IO_WQ_WORK_BLKCG;
1344 rcu_read_unlock();
1345 }
1346#endif
1347 if (!(req->work.flags & IO_WQ_WORK_CREDS)) {
1348 if (id->creds != current_cred())
1349 return false;
1350 get_cred(id->creds);
1351 req->work.flags |= IO_WQ_WORK_CREDS;
1352 }
Jens Axboe4ea33a92020-10-15 13:46:44 -06001353#ifdef CONFIG_AUDIT
1354 if (!uid_eq(current->loginuid, id->loginuid) ||
1355 current->sessionid != id->sessionid)
1356 return false;
1357#endif
Jens Axboe1e6fa522020-10-15 08:46:24 -06001358 if (!(req->work.flags & IO_WQ_WORK_FS) &&
1359 (def->work_flags & IO_WQ_WORK_FS)) {
1360 if (current->fs != id->fs)
1361 return false;
1362 spin_lock(&id->fs->lock);
1363 if (!id->fs->in_exec) {
1364 id->fs->users++;
1365 req->work.flags |= IO_WQ_WORK_FS;
1366 } else {
1367 req->work.flags |= IO_WQ_WORK_CANCEL;
1368 }
1369 spin_unlock(&current->fs->lock);
1370 }
1371
1372 return true;
Jens Axboe561fb042019-10-24 07:25:42 -06001373}
1374
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001375static void io_prep_async_work(struct io_kiocb *req)
Jens Axboe561fb042019-10-24 07:25:42 -06001376{
Jens Axboed3656342019-12-18 09:50:26 -07001377 const struct io_op_def *def = &io_op_defs[req->opcode];
Pavel Begunkov23329512020-10-10 18:34:06 +01001378 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5c3462c2020-10-15 09:02:33 -06001379 struct io_identity *id;
Jens Axboe54a91f32019-09-10 09:15:04 -06001380
Pavel Begunkov16d59802020-07-12 16:16:47 +03001381 io_req_init_async(req);
Jens Axboe5c3462c2020-10-15 09:02:33 -06001382 id = req->work.identity;
Pavel Begunkov16d59802020-07-12 16:16:47 +03001383
Pavel Begunkovfeaadc42020-10-22 16:47:16 +01001384 if (req->flags & REQ_F_FORCE_ASYNC)
1385 req->work.flags |= IO_WQ_WORK_CONCURRENT;
1386
Jens Axboed3656342019-12-18 09:50:26 -07001387 if (req->flags & REQ_F_ISREG) {
Pavel Begunkov23329512020-10-10 18:34:06 +01001388 if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL))
Pavel Begunkov8766dd52020-03-14 00:31:04 +03001389 io_wq_hash_work(&req->work, file_inode(req->file));
Jens Axboed3656342019-12-18 09:50:26 -07001390 } else {
1391 if (def->unbound_nonreg_file)
Jens Axboe3529d8c2019-12-19 18:24:38 -07001392 req->work.flags |= IO_WQ_WORK_UNBOUND;
Jens Axboe54a91f32019-09-10 09:15:04 -06001393 }
Pavel Begunkov23329512020-10-10 18:34:06 +01001394
Jens Axboe1e6fa522020-10-15 08:46:24 -06001395 /* ->mm can never change on us */
Jens Axboedfead8a2020-10-14 10:12:37 -06001396 if (!(req->work.flags & IO_WQ_WORK_MM) &&
1397 (def->work_flags & IO_WQ_WORK_MM)) {
Jens Axboe1e6fa522020-10-15 08:46:24 -06001398 mmgrab(id->mm);
Jens Axboedfead8a2020-10-14 10:12:37 -06001399 req->work.flags |= IO_WQ_WORK_MM;
Pavel Begunkov23329512020-10-10 18:34:06 +01001400 }
Jens Axboe1e6fa522020-10-15 08:46:24 -06001401
1402 /* if we fail grabbing identity, we must COW, regrab, and retry */
1403 if (io_grab_identity(req))
1404 return;
1405
1406 if (!io_identity_cow(req))
1407 return;
1408
1409 /* can't fail at this point */
1410 if (!io_grab_identity(req))
1411 WARN_ON(1);
Jens Axboe561fb042019-10-24 07:25:42 -06001412}
1413
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001414static void io_prep_async_link(struct io_kiocb *req)
1415{
1416 struct io_kiocb *cur;
1417
1418 io_prep_async_work(req);
1419 if (req->flags & REQ_F_LINK_HEAD)
1420 list_for_each_entry(cur, &req->link_list, link_list)
1421 io_prep_async_work(cur);
1422}
1423
Jens Axboe7271ef32020-08-10 09:55:22 -06001424static struct io_kiocb *__io_queue_async_work(struct io_kiocb *req)
Jens Axboe561fb042019-10-24 07:25:42 -06001425{
Jackie Liua197f662019-11-08 08:09:12 -07001426 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001427 struct io_kiocb *link = io_prep_linked_timeout(req);
Jens Axboe561fb042019-10-24 07:25:42 -06001428
Pavel Begunkov8766dd52020-03-14 00:31:04 +03001429 trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req,
1430 &req->work, req->flags);
1431 io_wq_enqueue(ctx->io_wq, &req->work);
Jens Axboe7271ef32020-08-10 09:55:22 -06001432 return link;
Jens Axboe18d9be12019-09-10 09:13:05 -06001433}
1434
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001435static void io_queue_async_work(struct io_kiocb *req)
1436{
Jens Axboe7271ef32020-08-10 09:55:22 -06001437 struct io_kiocb *link;
1438
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001439 /* init ->work of the whole link before punting */
1440 io_prep_async_link(req);
Jens Axboe7271ef32020-08-10 09:55:22 -06001441 link = __io_queue_async_work(req);
1442
1443 if (link)
1444 io_queue_linked_timeout(link);
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001445}
1446
Jens Axboe5262f562019-09-17 12:26:57 -06001447static void io_kill_timeout(struct io_kiocb *req)
1448{
Jens Axboee8c2bc12020-08-15 18:44:09 -07001449 struct io_timeout_data *io = req->async_data;
Jens Axboe5262f562019-09-17 12:26:57 -06001450 int ret;
1451
Jens Axboee8c2bc12020-08-15 18:44:09 -07001452 ret = hrtimer_try_to_cancel(&io->timer);
Jens Axboe5262f562019-09-17 12:26:57 -06001453 if (ret != -1) {
Pavel Begunkov01cec8c2020-07-30 18:43:50 +03001454 atomic_set(&req->ctx->cq_timeouts,
1455 atomic_read(&req->ctx->cq_timeouts) + 1);
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001456 list_del_init(&req->timeout.list);
Jens Axboe78e19bb2019-11-06 15:21:34 -07001457 io_cqring_fill_event(req, 0);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001458 io_put_req_deferred(req, 1);
Jens Axboe5262f562019-09-17 12:26:57 -06001459 }
1460}
1461
Jens Axboef3606e32020-09-22 08:18:24 -06001462static bool io_task_match(struct io_kiocb *req, struct task_struct *tsk)
1463{
1464 struct io_ring_ctx *ctx = req->ctx;
1465
1466 if (!tsk || req->task == tsk)
1467 return true;
Jens Axboe534ca6d2020-09-02 13:52:19 -06001468 if (ctx->flags & IORING_SETUP_SQPOLL) {
1469 if (ctx->sq_data && req->task == ctx->sq_data->thread)
1470 return true;
1471 }
Jens Axboef3606e32020-09-22 08:18:24 -06001472 return false;
1473}
1474
Jens Axboe76e1b642020-09-26 15:05:03 -06001475/*
1476 * Returns true if we found and killed one or more timeouts
1477 */
1478static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk)
Jens Axboe5262f562019-09-17 12:26:57 -06001479{
1480 struct io_kiocb *req, *tmp;
Jens Axboe76e1b642020-09-26 15:05:03 -06001481 int canceled = 0;
Jens Axboe5262f562019-09-17 12:26:57 -06001482
1483 spin_lock_irq(&ctx->completion_lock);
Jens Axboef3606e32020-09-22 08:18:24 -06001484 list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) {
Jens Axboe76e1b642020-09-26 15:05:03 -06001485 if (io_task_match(req, tsk)) {
Jens Axboef3606e32020-09-22 08:18:24 -06001486 io_kill_timeout(req);
Jens Axboe76e1b642020-09-26 15:05:03 -06001487 canceled++;
1488 }
Jens Axboef3606e32020-09-22 08:18:24 -06001489 }
Jens Axboe5262f562019-09-17 12:26:57 -06001490 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe76e1b642020-09-26 15:05:03 -06001491 return canceled != 0;
Jens Axboe5262f562019-09-17 12:26:57 -06001492}
1493
Pavel Begunkov04518942020-05-26 20:34:05 +03001494static void __io_queue_deferred(struct io_ring_ctx *ctx)
1495{
1496 do {
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001497 struct io_defer_entry *de = list_first_entry(&ctx->defer_list,
1498 struct io_defer_entry, list);
Jens Axboe7271ef32020-08-10 09:55:22 -06001499 struct io_kiocb *link;
Pavel Begunkov04518942020-05-26 20:34:05 +03001500
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001501 if (req_need_defer(de->req, de->seq))
Pavel Begunkov04518942020-05-26 20:34:05 +03001502 break;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001503 list_del_init(&de->list);
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001504 /* punt-init is done before queueing for defer */
Jens Axboe7271ef32020-08-10 09:55:22 -06001505 link = __io_queue_async_work(de->req);
1506 if (link) {
1507 __io_queue_linked_timeout(link);
1508 /* drop submission reference */
Pavel Begunkov216578e2020-10-13 09:44:00 +01001509 io_put_req_deferred(link, 1);
Jens Axboe7271ef32020-08-10 09:55:22 -06001510 }
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001511 kfree(de);
Pavel Begunkov04518942020-05-26 20:34:05 +03001512 } while (!list_empty(&ctx->defer_list));
1513}
1514
Pavel Begunkov360428f2020-05-30 14:54:17 +03001515static void io_flush_timeouts(struct io_ring_ctx *ctx)
1516{
1517 while (!list_empty(&ctx->timeout_list)) {
1518 struct io_kiocb *req = list_first_entry(&ctx->timeout_list,
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001519 struct io_kiocb, timeout.list);
Pavel Begunkov360428f2020-05-30 14:54:17 +03001520
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03001521 if (io_is_timeout_noseq(req))
Pavel Begunkov360428f2020-05-30 14:54:17 +03001522 break;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03001523 if (req->timeout.target_seq != ctx->cached_cq_tail
1524 - atomic_read(&ctx->cq_timeouts))
Pavel Begunkov360428f2020-05-30 14:54:17 +03001525 break;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03001526
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001527 list_del_init(&req->timeout.list);
Pavel Begunkov360428f2020-05-30 14:54:17 +03001528 io_kill_timeout(req);
1529 }
1530}
1531
Jens Axboede0617e2019-04-06 21:51:27 -06001532static void io_commit_cqring(struct io_ring_ctx *ctx)
1533{
Pavel Begunkov360428f2020-05-30 14:54:17 +03001534 io_flush_timeouts(ctx);
Jens Axboede0617e2019-04-06 21:51:27 -06001535 __io_commit_cqring(ctx);
1536
Pavel Begunkov04518942020-05-26 20:34:05 +03001537 if (unlikely(!list_empty(&ctx->defer_list)))
1538 __io_queue_deferred(ctx);
Jens Axboede0617e2019-04-06 21:51:27 -06001539}
1540
Jens Axboe90554202020-09-03 12:12:41 -06001541static inline bool io_sqring_full(struct io_ring_ctx *ctx)
1542{
1543 struct io_rings *r = ctx->rings;
1544
1545 return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries;
1546}
1547
Jens Axboe2b188cc2019-01-07 10:46:33 -07001548static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx)
1549{
Hristo Venev75b28af2019-08-26 17:23:46 +00001550 struct io_rings *rings = ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001551 unsigned tail;
1552
1553 tail = ctx->cached_cq_tail;
Stefan Bühler115e12e2019-04-24 23:54:18 +02001554 /*
1555 * writes to the cq entry need to come after reading head; the
1556 * control dependency is enough as we're using WRITE_ONCE to
1557 * fill the cq entry
1558 */
Hristo Venev75b28af2019-08-26 17:23:46 +00001559 if (tail - READ_ONCE(rings->cq.head) == rings->cq_ring_entries)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001560 return NULL;
1561
1562 ctx->cached_cq_tail++;
Hristo Venev75b28af2019-08-26 17:23:46 +00001563 return &rings->cqes[tail & ctx->cq_mask];
Jens Axboe2b188cc2019-01-07 10:46:33 -07001564}
1565
Jens Axboef2842ab2020-01-08 11:04:00 -07001566static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx)
1567{
Jens Axboef0b493e2020-02-01 21:30:11 -07001568 if (!ctx->cq_ev_fd)
1569 return false;
Stefano Garzarella7e55a192020-05-15 18:38:05 +02001570 if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED)
1571 return false;
Jens Axboef2842ab2020-01-08 11:04:00 -07001572 if (!ctx->eventfd_async)
1573 return true;
Jens Axboeb41e9852020-02-17 09:52:41 -07001574 return io_wq_current_is_worker();
Jens Axboef2842ab2020-01-08 11:04:00 -07001575}
1576
Jens Axboeb41e9852020-02-17 09:52:41 -07001577static void io_cqring_ev_posted(struct io_ring_ctx *ctx)
Jens Axboe8c838782019-03-12 15:48:16 -06001578{
1579 if (waitqueue_active(&ctx->wait))
1580 wake_up(&ctx->wait);
Jens Axboe534ca6d2020-09-02 13:52:19 -06001581 if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait))
1582 wake_up(&ctx->sq_data->wait);
Jens Axboeb41e9852020-02-17 09:52:41 -07001583 if (io_should_trigger_evfd(ctx))
Jens Axboe9b402842019-04-11 11:45:41 -06001584 eventfd_signal(ctx->cq_ev_fd, 1);
Jens Axboe8c838782019-03-12 15:48:16 -06001585}
1586
Pavel Begunkov46930142020-07-30 18:43:49 +03001587static void io_cqring_mark_overflow(struct io_ring_ctx *ctx)
1588{
1589 if (list_empty(&ctx->cq_overflow_list)) {
1590 clear_bit(0, &ctx->sq_check_overflow);
1591 clear_bit(0, &ctx->cq_check_overflow);
1592 ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW;
1593 }
1594}
1595
Pavel Begunkov99b32802020-11-04 13:39:31 +00001596static inline bool __io_match_files(struct io_kiocb *req,
1597 struct files_struct *files)
Jens Axboee6c8aa92020-09-28 13:10:13 -06001598{
Pavel Begunkov99b32802020-11-04 13:39:31 +00001599 return ((req->flags & REQ_F_WORK_INITIALIZED) &&
1600 (req->work.flags & IO_WQ_WORK_FILES)) &&
1601 req->work.identity->files == files;
1602}
1603
1604static bool io_match_files(struct io_kiocb *req,
1605 struct files_struct *files)
1606{
1607 struct io_kiocb *link;
1608
Jens Axboee6c8aa92020-09-28 13:10:13 -06001609 if (!files)
1610 return true;
Pavel Begunkov99b32802020-11-04 13:39:31 +00001611 if (__io_match_files(req, files))
1612 return true;
1613 if (req->flags & REQ_F_LINK_HEAD) {
1614 list_for_each_entry(link, &req->link_list, link_list) {
1615 if (__io_match_files(link, files))
1616 return true;
1617 }
1618 }
Jens Axboee6c8aa92020-09-28 13:10:13 -06001619 return false;
1620}
1621
Jens Axboec4a2ed72019-11-21 21:01:26 -07001622/* Returns true if there are no backlogged entries after the flush */
Jens Axboee6c8aa92020-09-28 13:10:13 -06001623static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
1624 struct task_struct *tsk,
1625 struct files_struct *files)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001626{
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001627 struct io_rings *rings = ctx->rings;
Jens Axboee6c8aa92020-09-28 13:10:13 -06001628 struct io_kiocb *req, *tmp;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001629 struct io_uring_cqe *cqe;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001630 unsigned long flags;
1631 LIST_HEAD(list);
1632
1633 if (!force) {
1634 if (list_empty_careful(&ctx->cq_overflow_list))
Jens Axboec4a2ed72019-11-21 21:01:26 -07001635 return true;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001636 if ((ctx->cached_cq_tail - READ_ONCE(rings->cq.head) ==
1637 rings->cq_ring_entries))
Jens Axboec4a2ed72019-11-21 21:01:26 -07001638 return false;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001639 }
1640
1641 spin_lock_irqsave(&ctx->completion_lock, flags);
1642
1643 /* if force is set, the ring is going away. always drop after that */
1644 if (force)
Jens Axboe69b3e542020-01-08 11:01:46 -07001645 ctx->cq_overflow_flushed = 1;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001646
Jens Axboec4a2ed72019-11-21 21:01:26 -07001647 cqe = NULL;
Jens Axboee6c8aa92020-09-28 13:10:13 -06001648 list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) {
1649 if (tsk && req->task != tsk)
1650 continue;
1651 if (!io_match_files(req, files))
1652 continue;
1653
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001654 cqe = io_get_cqring(ctx);
1655 if (!cqe && !force)
1656 break;
1657
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001658 list_move(&req->compl.list, &list);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001659 if (cqe) {
1660 WRITE_ONCE(cqe->user_data, req->user_data);
1661 WRITE_ONCE(cqe->res, req->result);
Pavel Begunkov0f7e4662020-07-13 23:37:16 +03001662 WRITE_ONCE(cqe->flags, req->compl.cflags);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001663 } else {
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001664 ctx->cached_cq_overflow++;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001665 WRITE_ONCE(ctx->rings->cq_overflow,
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001666 ctx->cached_cq_overflow);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001667 }
1668 }
1669
1670 io_commit_cqring(ctx);
Pavel Begunkov46930142020-07-30 18:43:49 +03001671 io_cqring_mark_overflow(ctx);
1672
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001673 spin_unlock_irqrestore(&ctx->completion_lock, flags);
1674 io_cqring_ev_posted(ctx);
1675
1676 while (!list_empty(&list)) {
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001677 req = list_first_entry(&list, struct io_kiocb, compl.list);
1678 list_del(&req->compl.list);
Jackie Liuec9c02a2019-11-08 23:50:36 +08001679 io_put_req(req);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001680 }
Jens Axboec4a2ed72019-11-21 21:01:26 -07001681
1682 return cqe != NULL;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001683}
1684
Jens Axboebcda7ba2020-02-23 16:42:51 -07001685static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001686{
Jens Axboe78e19bb2019-11-06 15:21:34 -07001687 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001688 struct io_uring_cqe *cqe;
1689
Jens Axboe78e19bb2019-11-06 15:21:34 -07001690 trace_io_uring_complete(ctx, req->user_data, res);
Jens Axboe51c3ff62019-11-03 06:52:50 -07001691
Jens Axboe2b188cc2019-01-07 10:46:33 -07001692 /*
1693 * If we can't get a cq entry, userspace overflowed the
1694 * submission (by quite a lot). Increment the overflow count in
1695 * the ring.
1696 */
1697 cqe = io_get_cqring(ctx);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001698 if (likely(cqe)) {
Jens Axboe78e19bb2019-11-06 15:21:34 -07001699 WRITE_ONCE(cqe->user_data, req->user_data);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001700 WRITE_ONCE(cqe->res, res);
Jens Axboebcda7ba2020-02-23 16:42:51 -07001701 WRITE_ONCE(cqe->flags, cflags);
Jens Axboefdaf0832020-10-30 09:37:30 -06001702 } else if (ctx->cq_overflow_flushed ||
1703 atomic_read(&req->task->io_uring->in_idle)) {
Jens Axboe0f212202020-09-13 13:09:39 -06001704 /*
1705 * If we're in ring overflow flush mode, or in task cancel mode,
1706 * then we cannot store the request for later flushing, we need
1707 * to drop it on the floor.
1708 */
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001709 ctx->cached_cq_overflow++;
1710 WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001711 } else {
Jens Axboead3eb2c2019-12-18 17:12:20 -07001712 if (list_empty(&ctx->cq_overflow_list)) {
1713 set_bit(0, &ctx->sq_check_overflow);
1714 set_bit(0, &ctx->cq_check_overflow);
Xiaoguang Wang6d5f9042020-07-09 09:15:29 +08001715 ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW;
Jens Axboead3eb2c2019-12-18 17:12:20 -07001716 }
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001717 io_clean_op(req);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001718 req->result = res;
Pavel Begunkov0f7e4662020-07-13 23:37:16 +03001719 req->compl.cflags = cflags;
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001720 refcount_inc(&req->refs);
1721 list_add_tail(&req->compl.list, &ctx->cq_overflow_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001722 }
1723}
1724
Jens Axboebcda7ba2020-02-23 16:42:51 -07001725static void io_cqring_fill_event(struct io_kiocb *req, long res)
1726{
1727 __io_cqring_fill_event(req, res, 0);
1728}
1729
Jens Axboee1e16092020-06-22 09:17:17 -06001730static void io_cqring_add_event(struct io_kiocb *req, long res, long cflags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001731{
Jens Axboe78e19bb2019-11-06 15:21:34 -07001732 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001733 unsigned long flags;
1734
1735 spin_lock_irqsave(&ctx->completion_lock, flags);
Jens Axboebcda7ba2020-02-23 16:42:51 -07001736 __io_cqring_fill_event(req, res, cflags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001737 io_commit_cqring(ctx);
1738 spin_unlock_irqrestore(&ctx->completion_lock, flags);
1739
Jens Axboe8c838782019-03-12 15:48:16 -06001740 io_cqring_ev_posted(ctx);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001741}
1742
Jens Axboe229a7b62020-06-22 10:13:11 -06001743static void io_submit_flush_completions(struct io_comp_state *cs)
Jens Axboebcda7ba2020-02-23 16:42:51 -07001744{
Jens Axboe229a7b62020-06-22 10:13:11 -06001745 struct io_ring_ctx *ctx = cs->ctx;
1746
1747 spin_lock_irq(&ctx->completion_lock);
1748 while (!list_empty(&cs->list)) {
1749 struct io_kiocb *req;
1750
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001751 req = list_first_entry(&cs->list, struct io_kiocb, compl.list);
1752 list_del(&req->compl.list);
Pavel Begunkov0f7e4662020-07-13 23:37:16 +03001753 __io_cqring_fill_event(req, req->result, req->compl.cflags);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001754
1755 /*
1756 * io_free_req() doesn't care about completion_lock unless one
1757 * of these flags is set. REQ_F_WORK_INITIALIZED is in the list
1758 * because of a potential deadlock with req->work.fs->lock
1759 */
1760 if (req->flags & (REQ_F_FAIL_LINK|REQ_F_LINK_TIMEOUT
1761 |REQ_F_WORK_INITIALIZED)) {
Jens Axboe229a7b62020-06-22 10:13:11 -06001762 spin_unlock_irq(&ctx->completion_lock);
1763 io_put_req(req);
1764 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001765 } else {
1766 io_put_req(req);
Jens Axboe229a7b62020-06-22 10:13:11 -06001767 }
1768 }
1769 io_commit_cqring(ctx);
1770 spin_unlock_irq(&ctx->completion_lock);
1771
1772 io_cqring_ev_posted(ctx);
1773 cs->nr = 0;
1774}
1775
1776static void __io_req_complete(struct io_kiocb *req, long res, unsigned cflags,
1777 struct io_comp_state *cs)
1778{
1779 if (!cs) {
1780 io_cqring_add_event(req, res, cflags);
1781 io_put_req(req);
1782 } else {
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001783 io_clean_op(req);
Jens Axboe229a7b62020-06-22 10:13:11 -06001784 req->result = res;
Pavel Begunkov0f7e4662020-07-13 23:37:16 +03001785 req->compl.cflags = cflags;
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001786 list_add_tail(&req->compl.list, &cs->list);
Jens Axboe229a7b62020-06-22 10:13:11 -06001787 if (++cs->nr >= 32)
1788 io_submit_flush_completions(cs);
1789 }
Jens Axboee1e16092020-06-22 09:17:17 -06001790}
1791
1792static void io_req_complete(struct io_kiocb *req, long res)
1793{
Jens Axboe229a7b62020-06-22 10:13:11 -06001794 __io_req_complete(req, res, 0, NULL);
Jens Axboebcda7ba2020-02-23 16:42:51 -07001795}
1796
Jens Axboe0ddf92e2019-11-08 08:52:53 -07001797static inline bool io_is_fallback_req(struct io_kiocb *req)
1798{
1799 return req == (struct io_kiocb *)
1800 ((unsigned long) req->ctx->fallback_req & ~1UL);
1801}
1802
1803static struct io_kiocb *io_get_fallback_req(struct io_ring_ctx *ctx)
1804{
1805 struct io_kiocb *req;
1806
1807 req = ctx->fallback_req;
Bijan Mottahedehdd461af2020-04-29 17:47:50 -07001808 if (!test_and_set_bit_lock(0, (unsigned long *) &ctx->fallback_req))
Jens Axboe0ddf92e2019-11-08 08:52:53 -07001809 return req;
1810
1811 return NULL;
1812}
1813
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03001814static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx,
1815 struct io_submit_state *state)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001816{
Pavel Begunkovf6b6c7d2020-06-21 13:09:53 +03001817 if (!state->free_reqs) {
Pavel Begunkov291b2822020-09-30 22:57:01 +03001818 gfp_t gfp = GFP_KERNEL | __GFP_NOWARN;
Jens Axboe2579f912019-01-09 09:10:43 -07001819 size_t sz;
1820 int ret;
1821
1822 sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs));
Jens Axboefd6fab22019-03-14 16:30:06 -06001823 ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs);
1824
1825 /*
1826 * Bulk alloc is all-or-nothing. If we fail to get a batch,
1827 * retry single alloc to be on the safe side.
1828 */
1829 if (unlikely(ret <= 0)) {
1830 state->reqs[0] = kmem_cache_alloc(req_cachep, gfp);
1831 if (!state->reqs[0])
Jens Axboe0ddf92e2019-11-08 08:52:53 -07001832 goto fallback;
Jens Axboefd6fab22019-03-14 16:30:06 -06001833 ret = 1;
1834 }
Pavel Begunkov291b2822020-09-30 22:57:01 +03001835 state->free_reqs = ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001836 }
1837
Pavel Begunkov291b2822020-09-30 22:57:01 +03001838 state->free_reqs--;
1839 return state->reqs[state->free_reqs];
Jens Axboe0ddf92e2019-11-08 08:52:53 -07001840fallback:
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03001841 return io_get_fallback_req(ctx);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001842}
1843
Pavel Begunkov8da11c12020-02-24 11:32:44 +03001844static inline void io_put_file(struct io_kiocb *req, struct file *file,
1845 bool fixed)
1846{
1847 if (fixed)
Xiaoguang Wang05589552020-03-31 14:05:18 +08001848 percpu_ref_put(req->fixed_file_refs);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03001849 else
1850 fput(file);
1851}
1852
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01001853static void io_dismantle_req(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001854{
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001855 io_clean_op(req);
Pavel Begunkov929a3af2020-02-19 00:19:09 +03001856
Jens Axboee8c2bc12020-08-15 18:44:09 -07001857 if (req->async_data)
1858 kfree(req->async_data);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03001859 if (req->file)
1860 io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE));
Jens Axboefcb323c2019-10-24 12:39:47 -06001861
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01001862 io_req_clean_work(req);
Pavel Begunkove6543a82020-06-28 12:52:30 +03001863}
Pavel Begunkov2b85edf2019-12-28 14:13:03 +03001864
Pavel Begunkov216578e2020-10-13 09:44:00 +01001865static void __io_free_req(struct io_kiocb *req)
Pavel Begunkove6543a82020-06-28 12:52:30 +03001866{
Jens Axboe0f212202020-09-13 13:09:39 -06001867 struct io_uring_task *tctx = req->task->io_uring;
Jens Axboe51a4cc12020-08-10 10:55:56 -06001868 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovecfc5172020-06-29 13:13:03 +03001869
Pavel Begunkov216578e2020-10-13 09:44:00 +01001870 io_dismantle_req(req);
Pavel Begunkove6543a82020-06-28 12:52:30 +03001871
Jens Axboed8a6df12020-10-15 16:24:45 -06001872 percpu_counter_dec(&tctx->inflight);
Jens Axboefdaf0832020-10-30 09:37:30 -06001873 if (atomic_read(&tctx->in_idle))
Jens Axboe0f212202020-09-13 13:09:39 -06001874 wake_up(&tctx->wait);
Jens Axboee3bc8e92020-09-24 08:45:57 -06001875 put_task_struct(req->task);
1876
Pavel Begunkovb1e50e52020-04-08 08:58:44 +03001877 if (likely(!io_is_fallback_req(req)))
1878 kmem_cache_free(req_cachep, req);
1879 else
Pavel Begunkovecfc5172020-06-29 13:13:03 +03001880 clear_bit_unlock(0, (unsigned long *) &ctx->fallback_req);
1881 percpu_ref_put(&ctx->refs);
Jens Axboee65ef562019-03-12 10:16:44 -06001882}
1883
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001884static void io_kill_linked_timeout(struct io_kiocb *req)
Jens Axboe9e645e112019-05-10 16:07:28 -06001885{
Jackie Liua197f662019-11-08 08:09:12 -07001886 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001887 struct io_kiocb *link;
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001888 bool cancelled = false;
1889 unsigned long flags;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001890
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001891 spin_lock_irqsave(&ctx->completion_lock, flags);
1892 link = list_first_entry_or_null(&req->link_list, struct io_kiocb,
1893 link_list);
Pavel Begunkov900fad42020-10-19 16:39:16 +01001894 /*
1895 * Can happen if a linked timeout fired and link had been like
1896 * req -> link t-out -> link t-out [-> ...]
1897 */
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001898 if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) {
1899 struct io_timeout_data *io = link->async_data;
1900 int ret;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001901
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001902 list_del_init(&link->link_list);
1903 ret = hrtimer_try_to_cancel(&io->timer);
1904 if (ret != -1) {
1905 io_cqring_fill_event(link, -ECANCELED);
1906 io_commit_cqring(ctx);
1907 cancelled = true;
1908 }
1909 }
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001910 req->flags &= ~REQ_F_LINK_TIMEOUT;
Pavel Begunkov216578e2020-10-13 09:44:00 +01001911 spin_unlock_irqrestore(&ctx->completion_lock, flags);
Jens Axboeab0b6452020-06-30 08:43:15 -06001912
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001913 if (cancelled) {
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001914 io_cqring_ev_posted(ctx);
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01001915 io_put_req(link);
1916 }
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001917}
1918
Pavel Begunkov9b5f7bd92020-06-29 13:13:00 +03001919static struct io_kiocb *io_req_link_next(struct io_kiocb *req)
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001920{
1921 struct io_kiocb *nxt;
Jens Axboe4d7dd462019-11-20 13:03:52 -07001922
Jens Axboe9e645e112019-05-10 16:07:28 -06001923 /*
1924 * The list should never be empty when we are called here. But could
1925 * potentially happen if the chain is messed up, check to be on the
1926 * safe side.
1927 */
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001928 if (unlikely(list_empty(&req->link_list)))
Pavel Begunkov9b5f7bd92020-06-29 13:13:00 +03001929 return NULL;
Jens Axboe94ae5e72019-11-14 19:39:52 -07001930
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001931 nxt = list_first_entry(&req->link_list, struct io_kiocb, link_list);
1932 list_del_init(&req->link_list);
1933 if (!list_empty(&nxt->link_list))
1934 nxt->flags |= REQ_F_LINK_HEAD;
Pavel Begunkov9b5f7bd92020-06-29 13:13:00 +03001935 return nxt;
Jens Axboe9e645e112019-05-10 16:07:28 -06001936}
1937
1938/*
Pavel Begunkovdea3b492020-04-12 02:05:04 +03001939 * Called if REQ_F_LINK_HEAD is set, and we fail the head request
Jens Axboe9e645e112019-05-10 16:07:28 -06001940 */
Pavel Begunkovd148ca42020-10-18 10:17:39 +01001941static void io_fail_links(struct io_kiocb *req)
Jens Axboe9e645e112019-05-10 16:07:28 -06001942{
Jens Axboe2665abf2019-11-05 12:40:47 -07001943 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovd148ca42020-10-18 10:17:39 +01001944 unsigned long flags;
Jens Axboe9e645e112019-05-10 16:07:28 -06001945
Pavel Begunkovd148ca42020-10-18 10:17:39 +01001946 spin_lock_irqsave(&ctx->completion_lock, flags);
Jens Axboe9e645e112019-05-10 16:07:28 -06001947 while (!list_empty(&req->link_list)) {
Pavel Begunkov44932332019-12-05 16:16:35 +03001948 struct io_kiocb *link = list_first_entry(&req->link_list,
1949 struct io_kiocb, link_list);
Jens Axboe9e645e112019-05-10 16:07:28 -06001950
Pavel Begunkov44932332019-12-05 16:16:35 +03001951 list_del_init(&link->link_list);
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02001952 trace_io_uring_fail_link(req, link);
Jens Axboe2665abf2019-11-05 12:40:47 -07001953
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001954 io_cqring_fill_event(link, -ECANCELED);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001955
1956 /*
1957 * It's ok to free under spinlock as they're not linked anymore,
1958 * but avoid REQ_F_WORK_INITIALIZED because it may deadlock on
1959 * work.fs->lock.
1960 */
1961 if (link->flags & REQ_F_WORK_INITIALIZED)
1962 io_put_req_deferred(link, 2);
1963 else
1964 io_double_put_req(link);
Jens Axboe9e645e112019-05-10 16:07:28 -06001965 }
Jens Axboe2665abf2019-11-05 12:40:47 -07001966
1967 io_commit_cqring(ctx);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001968 spin_unlock_irqrestore(&ctx->completion_lock, flags);
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001969
Jens Axboe2665abf2019-11-05 12:40:47 -07001970 io_cqring_ev_posted(ctx);
Jens Axboe9e645e112019-05-10 16:07:28 -06001971}
1972
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001973static struct io_kiocb *__io_req_find_next(struct io_kiocb *req)
Jens Axboe9e645e112019-05-10 16:07:28 -06001974{
Pavel Begunkov9b0d9112020-06-28 12:52:34 +03001975 req->flags &= ~REQ_F_LINK_HEAD;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03001976 if (req->flags & REQ_F_LINK_TIMEOUT)
1977 io_kill_linked_timeout(req);
Jens Axboe2665abf2019-11-05 12:40:47 -07001978
Jens Axboe9e645e112019-05-10 16:07:28 -06001979 /*
1980 * If LINK is set, we have dependent requests in this chain. If we
1981 * didn't fail this request, queue the first one up, moving any other
1982 * dependencies to the next request. In case of failure, fail the rest
1983 * of the chain.
1984 */
Pavel Begunkov9b5f7bd92020-06-29 13:13:00 +03001985 if (likely(!(req->flags & REQ_F_FAIL_LINK)))
1986 return io_req_link_next(req);
1987 io_fail_links(req);
1988 return NULL;
Jens Axboe4d7dd462019-11-20 13:03:52 -07001989}
Jens Axboe2665abf2019-11-05 12:40:47 -07001990
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03001991static struct io_kiocb *io_req_find_next(struct io_kiocb *req)
1992{
1993 if (likely(!(req->flags & REQ_F_LINK_HEAD)))
1994 return NULL;
1995 return __io_req_find_next(req);
1996}
1997
Jens Axboe87c43112020-09-30 21:00:14 -06001998static int io_req_task_work_add(struct io_kiocb *req, bool twa_signal_ok)
Jens Axboec2c4c832020-07-01 15:37:11 -06001999{
2000 struct task_struct *tsk = req->task;
2001 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe91989c72020-10-16 09:02:26 -06002002 enum task_work_notify_mode notify;
2003 int ret;
Jens Axboec2c4c832020-07-01 15:37:11 -06002004
Jens Axboe6200b0a2020-09-13 14:38:30 -06002005 if (tsk->flags & PF_EXITING)
2006 return -ESRCH;
2007
Jens Axboec2c4c832020-07-01 15:37:11 -06002008 /*
Jens Axboe0ba9c9e2020-08-06 19:41:50 -06002009 * SQPOLL kernel thread doesn't need notification, just a wakeup. For
2010 * all other cases, use TWA_SIGNAL unconditionally to ensure we're
2011 * processing task_work. There's no reliable way to tell if TWA_RESUME
2012 * will do the job.
Jens Axboec2c4c832020-07-01 15:37:11 -06002013 */
Jens Axboe91989c72020-10-16 09:02:26 -06002014 notify = TWA_NONE;
Jens Axboefd7d6de2020-08-23 11:00:37 -06002015 if (!(ctx->flags & IORING_SETUP_SQPOLL) && twa_signal_ok)
Jens Axboec2c4c832020-07-01 15:37:11 -06002016 notify = TWA_SIGNAL;
2017
Jens Axboe87c43112020-09-30 21:00:14 -06002018 ret = task_work_add(tsk, &req->task_work, notify);
Jens Axboec2c4c832020-07-01 15:37:11 -06002019 if (!ret)
2020 wake_up_process(tsk);
Jens Axboe0ba9c9e2020-08-06 19:41:50 -06002021
Jens Axboec2c4c832020-07-01 15:37:11 -06002022 return ret;
2023}
2024
Jens Axboec40f6372020-06-25 15:39:59 -06002025static void __io_req_task_cancel(struct io_kiocb *req, int error)
2026{
2027 struct io_ring_ctx *ctx = req->ctx;
2028
2029 spin_lock_irq(&ctx->completion_lock);
2030 io_cqring_fill_event(req, error);
2031 io_commit_cqring(ctx);
2032 spin_unlock_irq(&ctx->completion_lock);
2033
2034 io_cqring_ev_posted(ctx);
2035 req_set_fail_links(req);
2036 io_double_put_req(req);
2037}
2038
2039static void io_req_task_cancel(struct callback_head *cb)
2040{
2041 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
Jens Axboe87ceb6a2020-09-14 08:20:12 -06002042 struct io_ring_ctx *ctx = req->ctx;
Jens Axboec40f6372020-06-25 15:39:59 -06002043
2044 __io_req_task_cancel(req, -ECANCELED);
Jens Axboe87ceb6a2020-09-14 08:20:12 -06002045 percpu_ref_put(&ctx->refs);
Jens Axboec40f6372020-06-25 15:39:59 -06002046}
2047
2048static void __io_req_task_submit(struct io_kiocb *req)
2049{
2050 struct io_ring_ctx *ctx = req->ctx;
2051
Jens Axboec40f6372020-06-25 15:39:59 -06002052 if (!__io_sq_thread_acquire_mm(ctx)) {
2053 mutex_lock(&ctx->uring_lock);
Pavel Begunkovc1379e22020-09-30 22:57:56 +03002054 __io_queue_sqe(req, NULL);
Jens Axboec40f6372020-06-25 15:39:59 -06002055 mutex_unlock(&ctx->uring_lock);
Jens Axboe2665abf2019-11-05 12:40:47 -07002056 } else {
Jens Axboec40f6372020-06-25 15:39:59 -06002057 __io_req_task_cancel(req, -EFAULT);
Jens Axboe2665abf2019-11-05 12:40:47 -07002058 }
Jens Axboe9e645e112019-05-10 16:07:28 -06002059}
2060
Jens Axboec40f6372020-06-25 15:39:59 -06002061static void io_req_task_submit(struct callback_head *cb)
2062{
2063 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
Jens Axboe6d816e02020-08-11 08:04:14 -06002064 struct io_ring_ctx *ctx = req->ctx;
Jens Axboec40f6372020-06-25 15:39:59 -06002065
2066 __io_req_task_submit(req);
Jens Axboe6d816e02020-08-11 08:04:14 -06002067 percpu_ref_put(&ctx->refs);
Jens Axboec40f6372020-06-25 15:39:59 -06002068}
2069
2070static void io_req_task_queue(struct io_kiocb *req)
2071{
Jens Axboec40f6372020-06-25 15:39:59 -06002072 int ret;
2073
2074 init_task_work(&req->task_work, io_req_task_submit);
Jens Axboe6d816e02020-08-11 08:04:14 -06002075 percpu_ref_get(&req->ctx->refs);
Jens Axboec40f6372020-06-25 15:39:59 -06002076
Jens Axboe87c43112020-09-30 21:00:14 -06002077 ret = io_req_task_work_add(req, true);
Jens Axboec40f6372020-06-25 15:39:59 -06002078 if (unlikely(ret)) {
Jens Axboec2c4c832020-07-01 15:37:11 -06002079 struct task_struct *tsk;
2080
Jens Axboec40f6372020-06-25 15:39:59 -06002081 init_task_work(&req->task_work, io_req_task_cancel);
2082 tsk = io_wq_get_task(req->ctx->io_wq);
Jens Axboe91989c72020-10-16 09:02:26 -06002083 task_work_add(tsk, &req->task_work, TWA_NONE);
Jens Axboec2c4c832020-07-01 15:37:11 -06002084 wake_up_process(tsk);
Jens Axboec40f6372020-06-25 15:39:59 -06002085 }
Jens Axboec40f6372020-06-25 15:39:59 -06002086}
2087
Pavel Begunkovc3524382020-06-28 12:52:32 +03002088static void io_queue_next(struct io_kiocb *req)
Jackie Liuc69f8db2019-11-09 11:00:08 +08002089{
Pavel Begunkov9b5f7bd92020-06-29 13:13:00 +03002090 struct io_kiocb *nxt = io_req_find_next(req);
Pavel Begunkov944e58b2019-11-21 23:21:01 +03002091
Pavel Begunkov906a8c32020-06-27 14:04:55 +03002092 if (nxt)
2093 io_req_task_queue(nxt);
Jackie Liuc69f8db2019-11-09 11:00:08 +08002094}
2095
Jens Axboe9e645e112019-05-10 16:07:28 -06002096static void io_free_req(struct io_kiocb *req)
2097{
Pavel Begunkovc3524382020-06-28 12:52:32 +03002098 io_queue_next(req);
Jens Axboe9e645e112019-05-10 16:07:28 -06002099 __io_free_req(req);
Jens Axboee65ef562019-03-12 10:16:44 -06002100}
2101
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002102struct req_batch {
2103 void *reqs[IO_IOPOLL_BATCH];
2104 int to_free;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002105
2106 struct task_struct *task;
2107 int task_refs;
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002108};
2109
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002110static inline void io_init_req_batch(struct req_batch *rb)
Pavel Begunkov7a743e22020-03-03 21:33:13 +03002111{
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002112 rb->to_free = 0;
2113 rb->task_refs = 0;
2114 rb->task = NULL;
2115}
Pavel Begunkov8766dd52020-03-14 00:31:04 +03002116
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002117static void __io_req_free_batch_flush(struct io_ring_ctx *ctx,
2118 struct req_batch *rb)
2119{
2120 kmem_cache_free_bulk(req_cachep, rb->to_free, rb->reqs);
2121 percpu_ref_put_many(&ctx->refs, rb->to_free);
2122 rb->to_free = 0;
2123}
Pavel Begunkov7a743e22020-03-03 21:33:13 +03002124
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002125static void io_req_free_batch_finish(struct io_ring_ctx *ctx,
2126 struct req_batch *rb)
2127{
2128 if (rb->to_free)
2129 __io_req_free_batch_flush(ctx, rb);
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002130 if (rb->task) {
Jens Axboed8a6df12020-10-15 16:24:45 -06002131 struct io_uring_task *tctx = rb->task->io_uring;
2132
2133 percpu_counter_sub(&tctx->inflight, rb->task_refs);
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002134 put_task_struct_many(rb->task, rb->task_refs);
2135 rb->task = NULL;
2136 }
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002137}
2138
2139static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req)
2140{
2141 if (unlikely(io_is_fallback_req(req))) {
2142 io_free_req(req);
2143 return;
2144 }
2145 if (req->flags & REQ_F_LINK_HEAD)
2146 io_queue_next(req);
2147
Jens Axboee3bc8e92020-09-24 08:45:57 -06002148 if (req->task != rb->task) {
Jens Axboe0f212202020-09-13 13:09:39 -06002149 if (rb->task) {
Jens Axboed8a6df12020-10-15 16:24:45 -06002150 struct io_uring_task *tctx = rb->task->io_uring;
2151
2152 percpu_counter_sub(&tctx->inflight, rb->task_refs);
Jens Axboee3bc8e92020-09-24 08:45:57 -06002153 put_task_struct_many(rb->task, rb->task_refs);
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002154 }
Jens Axboee3bc8e92020-09-24 08:45:57 -06002155 rb->task = req->task;
2156 rb->task_refs = 0;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002157 }
Jens Axboee3bc8e92020-09-24 08:45:57 -06002158 rb->task_refs++;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002159
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01002160 io_dismantle_req(req);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002161 rb->reqs[rb->to_free++] = req;
2162 if (unlikely(rb->to_free == ARRAY_SIZE(rb->reqs)))
2163 __io_req_free_batch_flush(req->ctx, rb);
Pavel Begunkov7a743e22020-03-03 21:33:13 +03002164}
2165
Jens Axboeba816ad2019-09-28 11:36:45 -06002166/*
2167 * Drop reference to request, return next in chain (if there is one) if this
2168 * was the last reference to this request.
2169 */
Pavel Begunkov9b5f7bd92020-06-29 13:13:00 +03002170static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req)
Jens Axboee65ef562019-03-12 10:16:44 -06002171{
Pavel Begunkov9b5f7bd92020-06-29 13:13:00 +03002172 struct io_kiocb *nxt = NULL;
2173
Jens Axboe2a44f462020-02-25 13:25:41 -07002174 if (refcount_dec_and_test(&req->refs)) {
Pavel Begunkov9b5f7bd92020-06-29 13:13:00 +03002175 nxt = io_req_find_next(req);
Jens Axboe4d7dd462019-11-20 13:03:52 -07002176 __io_free_req(req);
Jens Axboe2a44f462020-02-25 13:25:41 -07002177 }
Pavel Begunkov9b5f7bd92020-06-29 13:13:00 +03002178 return nxt;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002179}
2180
Jens Axboe2b188cc2019-01-07 10:46:33 -07002181static void io_put_req(struct io_kiocb *req)
2182{
Jens Axboedef596e2019-01-09 08:59:42 -07002183 if (refcount_dec_and_test(&req->refs))
2184 io_free_req(req);
2185}
2186
Pavel Begunkov216578e2020-10-13 09:44:00 +01002187static void io_put_req_deferred_cb(struct callback_head *cb)
2188{
2189 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
2190
2191 io_free_req(req);
2192}
2193
2194static void io_free_req_deferred(struct io_kiocb *req)
2195{
2196 int ret;
2197
2198 init_task_work(&req->task_work, io_put_req_deferred_cb);
2199 ret = io_req_task_work_add(req, true);
2200 if (unlikely(ret)) {
2201 struct task_struct *tsk;
2202
2203 tsk = io_wq_get_task(req->ctx->io_wq);
Jens Axboe91989c72020-10-16 09:02:26 -06002204 task_work_add(tsk, &req->task_work, TWA_NONE);
Pavel Begunkov216578e2020-10-13 09:44:00 +01002205 wake_up_process(tsk);
2206 }
2207}
2208
2209static inline void io_put_req_deferred(struct io_kiocb *req, int refs)
2210{
2211 if (refcount_sub_and_test(refs, &req->refs))
2212 io_free_req_deferred(req);
2213}
2214
Pavel Begunkovf4db7182020-06-25 18:20:54 +03002215static struct io_wq_work *io_steal_work(struct io_kiocb *req)
Pavel Begunkov7a743e22020-03-03 21:33:13 +03002216{
Pavel Begunkov6df1db62020-07-03 22:15:06 +03002217 struct io_kiocb *nxt;
Pavel Begunkov7a743e22020-03-03 21:33:13 +03002218
Pavel Begunkovf4db7182020-06-25 18:20:54 +03002219 /*
2220 * A ref is owned by io-wq in which context we're. So, if that's the
2221 * last one, it's safe to steal next work. False negatives are Ok,
2222 * it just will be re-punted async in io_put_work()
2223 */
2224 if (refcount_read(&req->refs) != 1)
2225 return NULL;
2226
Pavel Begunkov9b5f7bd92020-06-29 13:13:00 +03002227 nxt = io_req_find_next(req);
Pavel Begunkov6df1db62020-07-03 22:15:06 +03002228 return nxt ? &nxt->work : NULL;
Pavel Begunkov7a743e22020-03-03 21:33:13 +03002229}
2230
Jens Axboe978db572019-11-14 22:39:04 -07002231static void io_double_put_req(struct io_kiocb *req)
2232{
2233 /* drop both submit and complete references */
2234 if (refcount_sub_and_test(2, &req->refs))
2235 io_free_req(req);
2236}
2237
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07002238static unsigned io_cqring_events(struct io_ring_ctx *ctx, bool noflush)
Jens Axboea3a0e432019-08-20 11:03:11 -06002239{
Jens Axboe84f97dc2019-11-06 11:27:53 -07002240 struct io_rings *rings = ctx->rings;
2241
Jens Axboead3eb2c2019-12-18 17:12:20 -07002242 if (test_bit(0, &ctx->cq_check_overflow)) {
2243 /*
2244 * noflush == true is from the waitqueue handler, just ensure
2245 * we wake up the task, and the next invocation will flush the
2246 * entries. We cannot safely to it from here.
2247 */
2248 if (noflush && !list_empty(&ctx->cq_overflow_list))
2249 return -1U;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07002250
Jens Axboee6c8aa92020-09-28 13:10:13 -06002251 io_cqring_overflow_flush(ctx, false, NULL, NULL);
Jens Axboead3eb2c2019-12-18 17:12:20 -07002252 }
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07002253
Jens Axboea3a0e432019-08-20 11:03:11 -06002254 /* See comment at the top of this file */
2255 smp_rmb();
Jens Axboead3eb2c2019-12-18 17:12:20 -07002256 return ctx->cached_cq_tail - READ_ONCE(rings->cq.head);
Jens Axboea3a0e432019-08-20 11:03:11 -06002257}
2258
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03002259static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx)
2260{
2261 struct io_rings *rings = ctx->rings;
2262
2263 /* make sure SQ entry isn't read before tail */
2264 return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head;
2265}
2266
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002267static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf)
Jens Axboee94f1412019-12-19 12:06:02 -07002268{
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002269 unsigned int cflags;
Jens Axboee94f1412019-12-19 12:06:02 -07002270
Jens Axboebcda7ba2020-02-23 16:42:51 -07002271 cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT;
2272 cflags |= IORING_CQE_F_BUFFER;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03002273 req->flags &= ~REQ_F_BUFFER_SELECTED;
Jens Axboebcda7ba2020-02-23 16:42:51 -07002274 kfree(kbuf);
2275 return cflags;
2276}
2277
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002278static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req)
2279{
2280 struct io_buffer *kbuf;
2281
2282 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
2283 return io_put_kbuf(req, kbuf);
2284}
2285
Jens Axboe4c6e2772020-07-01 11:29:10 -06002286static inline bool io_run_task_work(void)
2287{
Jens Axboe6200b0a2020-09-13 14:38:30 -06002288 /*
2289 * Not safe to run on exiting task, and the task_work handling will
2290 * not add work to such a task.
2291 */
2292 if (unlikely(current->flags & PF_EXITING))
2293 return false;
Jens Axboe4c6e2772020-07-01 11:29:10 -06002294 if (current->task_works) {
2295 __set_current_state(TASK_RUNNING);
2296 task_work_run();
2297 return true;
2298 }
2299
2300 return false;
2301}
2302
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002303static void io_iopoll_queue(struct list_head *again)
2304{
2305 struct io_kiocb *req;
2306
2307 do {
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002308 req = list_first_entry(again, struct io_kiocb, inflight_entry);
2309 list_del(&req->inflight_entry);
Pavel Begunkov81b68a52020-07-30 18:43:46 +03002310 __io_complete_rw(req, -EAGAIN, 0, NULL);
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002311 } while (!list_empty(again));
2312}
2313
Jens Axboedef596e2019-01-09 08:59:42 -07002314/*
2315 * Find and free completed poll iocbs
2316 */
2317static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events,
2318 struct list_head *done)
2319{
Jens Axboe8237e042019-12-28 10:48:22 -07002320 struct req_batch rb;
Jens Axboedef596e2019-01-09 08:59:42 -07002321 struct io_kiocb *req;
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002322 LIST_HEAD(again);
2323
2324 /* order with ->result store in io_complete_rw_iopoll() */
2325 smp_rmb();
Jens Axboedef596e2019-01-09 08:59:42 -07002326
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002327 io_init_req_batch(&rb);
Jens Axboedef596e2019-01-09 08:59:42 -07002328 while (!list_empty(done)) {
Jens Axboebcda7ba2020-02-23 16:42:51 -07002329 int cflags = 0;
2330
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002331 req = list_first_entry(done, struct io_kiocb, inflight_entry);
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002332 if (READ_ONCE(req->result) == -EAGAIN) {
Jens Axboe56450c22020-08-26 18:58:26 -06002333 req->result = 0;
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002334 req->iopoll_completed = 0;
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002335 list_move_tail(&req->inflight_entry, &again);
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002336 continue;
2337 }
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002338 list_del(&req->inflight_entry);
Jens Axboedef596e2019-01-09 08:59:42 -07002339
Jens Axboebcda7ba2020-02-23 16:42:51 -07002340 if (req->flags & REQ_F_BUFFER_SELECTED)
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002341 cflags = io_put_rw_kbuf(req);
Jens Axboebcda7ba2020-02-23 16:42:51 -07002342
2343 __io_cqring_fill_event(req, req->result, cflags);
Jens Axboedef596e2019-01-09 08:59:42 -07002344 (*nr_events)++;
2345
Pavel Begunkovc3524382020-06-28 12:52:32 +03002346 if (refcount_dec_and_test(&req->refs))
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002347 io_req_free_batch(&rb, req);
Jens Axboedef596e2019-01-09 08:59:42 -07002348 }
Jens Axboedef596e2019-01-09 08:59:42 -07002349
Jens Axboe09bb8392019-03-13 12:39:28 -06002350 io_commit_cqring(ctx);
Xiaoguang Wang32b22442020-03-11 09:26:09 +08002351 if (ctx->flags & IORING_SETUP_SQPOLL)
2352 io_cqring_ev_posted(ctx);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002353 io_req_free_batch_finish(ctx, &rb);
Jens Axboedef596e2019-01-09 08:59:42 -07002354
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002355 if (!list_empty(&again))
2356 io_iopoll_queue(&again);
Bijan Mottahedeh581f9812020-04-03 13:51:33 -07002357}
2358
Jens Axboedef596e2019-01-09 08:59:42 -07002359static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events,
2360 long min)
2361{
2362 struct io_kiocb *req, *tmp;
2363 LIST_HEAD(done);
2364 bool spin;
2365 int ret;
2366
2367 /*
2368 * Only spin for completions if we don't have multiple devices hanging
2369 * off our complete list, and we're under the requested amount.
2370 */
2371 spin = !ctx->poll_multi_file && *nr_events < min;
2372
2373 ret = 0;
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002374 list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) {
Jens Axboe9adbd452019-12-20 08:45:55 -07002375 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboedef596e2019-01-09 08:59:42 -07002376
2377 /*
Bijan Mottahedeh581f9812020-04-03 13:51:33 -07002378 * Move completed and retryable entries to our local lists.
2379 * If we find a request that requires polling, break out
2380 * and complete those lists first, if we have entries there.
Jens Axboedef596e2019-01-09 08:59:42 -07002381 */
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002382 if (READ_ONCE(req->iopoll_completed)) {
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002383 list_move_tail(&req->inflight_entry, &done);
Jens Axboedef596e2019-01-09 08:59:42 -07002384 continue;
2385 }
2386 if (!list_empty(&done))
2387 break;
2388
2389 ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin);
2390 if (ret < 0)
2391 break;
2392
Pavel Begunkov3aadc232020-07-06 17:59:29 +03002393 /* iopoll may have completed current req */
2394 if (READ_ONCE(req->iopoll_completed))
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002395 list_move_tail(&req->inflight_entry, &done);
Pavel Begunkov3aadc232020-07-06 17:59:29 +03002396
Jens Axboedef596e2019-01-09 08:59:42 -07002397 if (ret && spin)
2398 spin = false;
2399 ret = 0;
2400 }
2401
2402 if (!list_empty(&done))
2403 io_iopoll_complete(ctx, nr_events, &done);
2404
2405 return ret;
2406}
2407
2408/*
Brian Gianforcarod195a662019-12-13 03:09:50 -08002409 * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a
Jens Axboedef596e2019-01-09 08:59:42 -07002410 * non-spinning poll check - we'll still enter the driver poll loop, but only
2411 * as a non-spinning completion check.
2412 */
2413static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events,
2414 long min)
2415{
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002416 while (!list_empty(&ctx->iopoll_list) && !need_resched()) {
Jens Axboedef596e2019-01-09 08:59:42 -07002417 int ret;
2418
2419 ret = io_do_iopoll(ctx, nr_events, min);
2420 if (ret < 0)
2421 return ret;
Pavel Begunkoveba0a4d2020-07-06 17:59:30 +03002422 if (*nr_events >= min)
Jens Axboedef596e2019-01-09 08:59:42 -07002423 return 0;
2424 }
2425
2426 return 1;
2427}
2428
2429/*
2430 * We can't just wait for polled events to come to us, we have to actively
2431 * find and complete them.
2432 */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002433static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx)
Jens Axboedef596e2019-01-09 08:59:42 -07002434{
2435 if (!(ctx->flags & IORING_SETUP_IOPOLL))
2436 return;
2437
2438 mutex_lock(&ctx->uring_lock);
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002439 while (!list_empty(&ctx->iopoll_list)) {
Jens Axboedef596e2019-01-09 08:59:42 -07002440 unsigned int nr_events = 0;
2441
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002442 io_do_iopoll(ctx, &nr_events, 0);
Jens Axboe08f54392019-08-21 22:19:11 -06002443
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002444 /* let it sleep and repeat later if can't complete a request */
2445 if (nr_events == 0)
2446 break;
Jens Axboe08f54392019-08-21 22:19:11 -06002447 /*
2448 * Ensure we allow local-to-the-cpu processing to take place,
2449 * in this case we need to ensure that we reap all events.
Pavel Begunkov3fcee5a2020-07-06 17:59:31 +03002450 * Also let task_work, etc. to progress by releasing the mutex
Jens Axboe08f54392019-08-21 22:19:11 -06002451 */
Pavel Begunkov3fcee5a2020-07-06 17:59:31 +03002452 if (need_resched()) {
2453 mutex_unlock(&ctx->uring_lock);
2454 cond_resched();
2455 mutex_lock(&ctx->uring_lock);
2456 }
Jens Axboedef596e2019-01-09 08:59:42 -07002457 }
2458 mutex_unlock(&ctx->uring_lock);
2459}
2460
Pavel Begunkov7668b922020-07-07 16:36:21 +03002461static int io_iopoll_check(struct io_ring_ctx *ctx, long min)
Jens Axboedef596e2019-01-09 08:59:42 -07002462{
Pavel Begunkov7668b922020-07-07 16:36:21 +03002463 unsigned int nr_events = 0;
Jens Axboe2b2ed972019-10-25 10:06:15 -06002464 int iters = 0, ret = 0;
Jens Axboedef596e2019-01-09 08:59:42 -07002465
Xiaoguang Wangc7849be2020-02-22 14:46:05 +08002466 /*
2467 * We disallow the app entering submit/complete with polling, but we
2468 * still need to lock the ring to prevent racing with polled issue
2469 * that got punted to a workqueue.
2470 */
2471 mutex_lock(&ctx->uring_lock);
Jens Axboedef596e2019-01-09 08:59:42 -07002472 do {
Jens Axboe500f9fb2019-08-19 12:15:59 -06002473 /*
Jens Axboea3a0e432019-08-20 11:03:11 -06002474 * Don't enter poll loop if we already have events pending.
2475 * If we do, we can potentially be spinning for commands that
2476 * already triggered a CQE (eg in error).
2477 */
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07002478 if (io_cqring_events(ctx, false))
Jens Axboea3a0e432019-08-20 11:03:11 -06002479 break;
2480
2481 /*
Jens Axboe500f9fb2019-08-19 12:15:59 -06002482 * If a submit got punted to a workqueue, we can have the
2483 * application entering polling for a command before it gets
2484 * issued. That app will hold the uring_lock for the duration
2485 * of the poll right here, so we need to take a breather every
2486 * now and then to ensure that the issue has a chance to add
2487 * the poll to the issued list. Otherwise we can spin here
2488 * forever, while the workqueue is stuck trying to acquire the
2489 * very same mutex.
2490 */
2491 if (!(++iters & 7)) {
2492 mutex_unlock(&ctx->uring_lock);
Jens Axboe4c6e2772020-07-01 11:29:10 -06002493 io_run_task_work();
Jens Axboe500f9fb2019-08-19 12:15:59 -06002494 mutex_lock(&ctx->uring_lock);
2495 }
2496
Pavel Begunkov7668b922020-07-07 16:36:21 +03002497 ret = io_iopoll_getevents(ctx, &nr_events, min);
Jens Axboedef596e2019-01-09 08:59:42 -07002498 if (ret <= 0)
2499 break;
2500 ret = 0;
Pavel Begunkov7668b922020-07-07 16:36:21 +03002501 } while (min && !nr_events && !need_resched());
Jens Axboedef596e2019-01-09 08:59:42 -07002502
Jens Axboe500f9fb2019-08-19 12:15:59 -06002503 mutex_unlock(&ctx->uring_lock);
Jens Axboedef596e2019-01-09 08:59:42 -07002504 return ret;
2505}
2506
Jens Axboe491381ce2019-10-17 09:20:46 -06002507static void kiocb_end_write(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002508{
Jens Axboe491381ce2019-10-17 09:20:46 -06002509 /*
2510 * Tell lockdep we inherited freeze protection from submission
2511 * thread.
2512 */
2513 if (req->flags & REQ_F_ISREG) {
2514 struct inode *inode = file_inode(req->file);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002515
Jens Axboe491381ce2019-10-17 09:20:46 -06002516 __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002517 }
Jens Axboe491381ce2019-10-17 09:20:46 -06002518 file_end_write(req->file);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002519}
2520
Jens Axboea1d7c392020-06-22 11:09:46 -06002521static void io_complete_rw_common(struct kiocb *kiocb, long res,
2522 struct io_comp_state *cs)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002523{
Jens Axboe9adbd452019-12-20 08:45:55 -07002524 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboebcda7ba2020-02-23 16:42:51 -07002525 int cflags = 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002526
Jens Axboe491381ce2019-10-17 09:20:46 -06002527 if (kiocb->ki_flags & IOCB_WRITE)
2528 kiocb_end_write(req);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002529
Jens Axboe4e88d6e2019-12-07 20:59:47 -07002530 if (res != req->result)
2531 req_set_fail_links(req);
Jens Axboebcda7ba2020-02-23 16:42:51 -07002532 if (req->flags & REQ_F_BUFFER_SELECTED)
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002533 cflags = io_put_rw_kbuf(req);
Jens Axboea1d7c392020-06-22 11:09:46 -06002534 __io_req_complete(req, res, cflags, cs);
Jens Axboeba816ad2019-09-28 11:36:45 -06002535}
2536
Jens Axboeb63534c2020-06-04 11:28:00 -06002537#ifdef CONFIG_BLOCK
2538static bool io_resubmit_prep(struct io_kiocb *req, int error)
2539{
2540 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
2541 ssize_t ret = -ECANCELED;
2542 struct iov_iter iter;
2543 int rw;
2544
2545 if (error) {
2546 ret = error;
2547 goto end_req;
2548 }
2549
2550 switch (req->opcode) {
2551 case IORING_OP_READV:
2552 case IORING_OP_READ_FIXED:
2553 case IORING_OP_READ:
2554 rw = READ;
2555 break;
2556 case IORING_OP_WRITEV:
2557 case IORING_OP_WRITE_FIXED:
2558 case IORING_OP_WRITE:
2559 rw = WRITE;
2560 break;
2561 default:
2562 printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n",
2563 req->opcode);
2564 goto end_req;
2565 }
2566
Jens Axboee8c2bc12020-08-15 18:44:09 -07002567 if (!req->async_data) {
Jens Axboe8f3d7492020-09-14 09:28:14 -06002568 ret = io_import_iovec(rw, req, &iovec, &iter, false);
2569 if (ret < 0)
2570 goto end_req;
2571 ret = io_setup_async_rw(req, iovec, inline_vecs, &iter, false);
2572 if (!ret)
2573 return true;
2574 kfree(iovec);
2575 } else {
Jens Axboeb63534c2020-06-04 11:28:00 -06002576 return true;
Jens Axboe8f3d7492020-09-14 09:28:14 -06002577 }
Jens Axboeb63534c2020-06-04 11:28:00 -06002578end_req:
Jens Axboeb63534c2020-06-04 11:28:00 -06002579 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06002580 io_req_complete(req, ret);
Jens Axboeb63534c2020-06-04 11:28:00 -06002581 return false;
2582}
Jens Axboeb63534c2020-06-04 11:28:00 -06002583#endif
2584
2585static bool io_rw_reissue(struct io_kiocb *req, long res)
2586{
2587#ifdef CONFIG_BLOCK
Jens Axboe355afae2020-09-02 09:30:31 -06002588 umode_t mode = file_inode(req->file)->i_mode;
Jens Axboeb63534c2020-06-04 11:28:00 -06002589 int ret;
2590
Jens Axboe355afae2020-09-02 09:30:31 -06002591 if (!S_ISBLK(mode) && !S_ISREG(mode))
2592 return false;
Jens Axboeb63534c2020-06-04 11:28:00 -06002593 if ((res != -EAGAIN && res != -EOPNOTSUPP) || io_wq_current_is_worker())
2594 return false;
2595
Jens Axboefdee9462020-08-27 16:46:24 -06002596 ret = io_sq_thread_acquire_mm(req->ctx, req);
Jens Axboe6d816e02020-08-11 08:04:14 -06002597
Jens Axboefdee9462020-08-27 16:46:24 -06002598 if (io_resubmit_prep(req, ret)) {
2599 refcount_inc(&req->refs);
2600 io_queue_async_work(req);
Jens Axboeb63534c2020-06-04 11:28:00 -06002601 return true;
Jens Axboefdee9462020-08-27 16:46:24 -06002602 }
2603
Jens Axboeb63534c2020-06-04 11:28:00 -06002604#endif
2605 return false;
2606}
2607
Jens Axboea1d7c392020-06-22 11:09:46 -06002608static void __io_complete_rw(struct io_kiocb *req, long res, long res2,
2609 struct io_comp_state *cs)
2610{
2611 if (!io_rw_reissue(req, res))
2612 io_complete_rw_common(&req->rw.kiocb, res, cs);
Jens Axboeba816ad2019-09-28 11:36:45 -06002613}
2614
2615static void io_complete_rw(struct kiocb *kiocb, long res, long res2)
2616{
Jens Axboe9adbd452019-12-20 08:45:55 -07002617 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboeba816ad2019-09-28 11:36:45 -06002618
Jens Axboea1d7c392020-06-22 11:09:46 -06002619 __io_complete_rw(req, res, res2, NULL);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002620}
2621
Jens Axboedef596e2019-01-09 08:59:42 -07002622static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2)
2623{
Jens Axboe9adbd452019-12-20 08:45:55 -07002624 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboedef596e2019-01-09 08:59:42 -07002625
Jens Axboe491381ce2019-10-17 09:20:46 -06002626 if (kiocb->ki_flags & IOCB_WRITE)
2627 kiocb_end_write(req);
Jens Axboedef596e2019-01-09 08:59:42 -07002628
Xiaoguang Wang2d7d6792020-06-16 02:06:37 +08002629 if (res != -EAGAIN && res != req->result)
Jens Axboe4e88d6e2019-12-07 20:59:47 -07002630 req_set_fail_links(req);
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002631
2632 WRITE_ONCE(req->result, res);
2633 /* order with io_poll_complete() checking ->result */
Pavel Begunkovcd664b02020-06-25 12:37:10 +03002634 smp_wmb();
2635 WRITE_ONCE(req->iopoll_completed, 1);
Jens Axboedef596e2019-01-09 08:59:42 -07002636}
2637
2638/*
2639 * After the iocb has been issued, it's safe to be found on the poll list.
2640 * Adding the kiocb to the list AFTER submission ensures that we don't
2641 * find it from a io_iopoll_getevents() thread before the issuer is done
2642 * accessing the kiocb cookie.
2643 */
2644static void io_iopoll_req_issued(struct io_kiocb *req)
2645{
2646 struct io_ring_ctx *ctx = req->ctx;
2647
2648 /*
2649 * Track whether we have multiple files in our lists. This will impact
2650 * how we do polling eventually, not spinning if we're on potentially
2651 * different devices.
2652 */
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002653 if (list_empty(&ctx->iopoll_list)) {
Jens Axboedef596e2019-01-09 08:59:42 -07002654 ctx->poll_multi_file = false;
2655 } else if (!ctx->poll_multi_file) {
2656 struct io_kiocb *list_req;
2657
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002658 list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb,
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002659 inflight_entry);
Jens Axboe9adbd452019-12-20 08:45:55 -07002660 if (list_req->file != req->file)
Jens Axboedef596e2019-01-09 08:59:42 -07002661 ctx->poll_multi_file = true;
2662 }
2663
2664 /*
2665 * For fast devices, IO may have already completed. If it has, add
2666 * it to the front so we find it first.
2667 */
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002668 if (READ_ONCE(req->iopoll_completed))
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002669 list_add(&req->inflight_entry, &ctx->iopoll_list);
Jens Axboedef596e2019-01-09 08:59:42 -07002670 else
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002671 list_add_tail(&req->inflight_entry, &ctx->iopoll_list);
Xiaoguang Wangbdcd3ea2020-02-25 22:12:08 +08002672
2673 if ((ctx->flags & IORING_SETUP_SQPOLL) &&
Jens Axboe534ca6d2020-09-02 13:52:19 -06002674 wq_has_sleeper(&ctx->sq_data->wait))
2675 wake_up(&ctx->sq_data->wait);
Jens Axboedef596e2019-01-09 08:59:42 -07002676}
2677
Pavel Begunkov9f13c352020-05-17 14:13:41 +03002678static void __io_state_file_put(struct io_submit_state *state)
Jens Axboe9a56a232019-01-09 09:06:50 -07002679{
Pavel Begunkov06ef3602020-07-16 23:28:33 +03002680 if (state->has_refs)
2681 fput_many(state->file, state->has_refs);
Pavel Begunkov9f13c352020-05-17 14:13:41 +03002682 state->file = NULL;
2683}
2684
2685static inline void io_state_file_put(struct io_submit_state *state)
2686{
2687 if (state->file)
2688 __io_state_file_put(state);
Jens Axboe9a56a232019-01-09 09:06:50 -07002689}
2690
2691/*
2692 * Get as many references to a file as we have IOs left in this submission,
2693 * assuming most submissions are for one file, or at least that each file
2694 * has more than one submission.
2695 */
Pavel Begunkov8da11c12020-02-24 11:32:44 +03002696static struct file *__io_file_get(struct io_submit_state *state, int fd)
Jens Axboe9a56a232019-01-09 09:06:50 -07002697{
2698 if (!state)
2699 return fget(fd);
2700
2701 if (state->file) {
2702 if (state->fd == fd) {
Pavel Begunkov06ef3602020-07-16 23:28:33 +03002703 state->has_refs--;
Jens Axboe9a56a232019-01-09 09:06:50 -07002704 return state->file;
2705 }
Pavel Begunkov9f13c352020-05-17 14:13:41 +03002706 __io_state_file_put(state);
Jens Axboe9a56a232019-01-09 09:06:50 -07002707 }
2708 state->file = fget_many(fd, state->ios_left);
2709 if (!state->file)
2710 return NULL;
2711
2712 state->fd = fd;
Pavel Begunkov71b547c2020-10-10 18:34:09 +01002713 state->has_refs = state->ios_left - 1;
Jens Axboe9a56a232019-01-09 09:06:50 -07002714 return state->file;
2715}
2716
Jens Axboe4503b762020-06-01 10:00:27 -06002717static bool io_bdev_nowait(struct block_device *bdev)
2718{
2719#ifdef CONFIG_BLOCK
Jeffle Xu9ba0d0c2020-10-19 16:59:42 +08002720 return !bdev || blk_queue_nowait(bdev_get_queue(bdev));
Jens Axboe4503b762020-06-01 10:00:27 -06002721#else
2722 return true;
2723#endif
2724}
2725
Jens Axboe2b188cc2019-01-07 10:46:33 -07002726/*
2727 * If we tracked the file through the SCM inflight mechanism, we could support
2728 * any file. For now, just ensure that anything potentially problematic is done
2729 * inline.
2730 */
Jens Axboeaf197f52020-04-28 13:15:06 -06002731static bool io_file_supports_async(struct file *file, int rw)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002732{
2733 umode_t mode = file_inode(file)->i_mode;
2734
Jens Axboe4503b762020-06-01 10:00:27 -06002735 if (S_ISBLK(mode)) {
2736 if (io_bdev_nowait(file->f_inode->i_bdev))
2737 return true;
2738 return false;
2739 }
2740 if (S_ISCHR(mode) || S_ISSOCK(mode))
Jens Axboe2b188cc2019-01-07 10:46:33 -07002741 return true;
Jens Axboe4503b762020-06-01 10:00:27 -06002742 if (S_ISREG(mode)) {
2743 if (io_bdev_nowait(file->f_inode->i_sb->s_bdev) &&
2744 file->f_op != &io_uring_fops)
2745 return true;
2746 return false;
2747 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002748
Jens Axboec5b85622020-06-09 19:23:05 -06002749 /* any ->read/write should understand O_NONBLOCK */
2750 if (file->f_flags & O_NONBLOCK)
2751 return true;
2752
Jens Axboeaf197f52020-04-28 13:15:06 -06002753 if (!(file->f_mode & FMODE_NOWAIT))
2754 return false;
2755
2756 if (rw == READ)
2757 return file->f_op->read_iter != NULL;
2758
2759 return file->f_op->write_iter != NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002760}
2761
Pavel Begunkova88fc402020-09-30 22:57:53 +03002762static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002763{
Jens Axboedef596e2019-01-09 08:59:42 -07002764 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe9adbd452019-12-20 08:45:55 -07002765 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboe09bb8392019-03-13 12:39:28 -06002766 unsigned ioprio;
2767 int ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002768
Jens Axboe491381ce2019-10-17 09:20:46 -06002769 if (S_ISREG(file_inode(req->file)->i_mode))
2770 req->flags |= REQ_F_ISREG;
2771
Jens Axboe2b188cc2019-01-07 10:46:33 -07002772 kiocb->ki_pos = READ_ONCE(sqe->off);
Jens Axboeba042912019-12-25 16:33:42 -07002773 if (kiocb->ki_pos == -1 && !(req->file->f_mode & FMODE_STREAM)) {
2774 req->flags |= REQ_F_CUR_POS;
2775 kiocb->ki_pos = req->file->f_pos;
2776 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002777 kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp));
Pavel Begunkov3e577dc2020-02-01 03:58:42 +03002778 kiocb->ki_flags = iocb_flags(kiocb->ki_filp);
2779 ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags));
2780 if (unlikely(ret))
2781 return ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002782
2783 ioprio = READ_ONCE(sqe->ioprio);
2784 if (ioprio) {
2785 ret = ioprio_check_cap(ioprio);
2786 if (ret)
Jens Axboe09bb8392019-03-13 12:39:28 -06002787 return ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002788
2789 kiocb->ki_ioprio = ioprio;
2790 } else
2791 kiocb->ki_ioprio = get_current_ioprio();
2792
Stefan Bühler8449eed2019-04-27 20:34:19 +02002793 /* don't allow async punt if RWF_NOWAIT was requested */
Jens Axboec5b85622020-06-09 19:23:05 -06002794 if (kiocb->ki_flags & IOCB_NOWAIT)
Stefan Bühler8449eed2019-04-27 20:34:19 +02002795 req->flags |= REQ_F_NOWAIT;
2796
Jens Axboedef596e2019-01-09 08:59:42 -07002797 if (ctx->flags & IORING_SETUP_IOPOLL) {
Jens Axboedef596e2019-01-09 08:59:42 -07002798 if (!(kiocb->ki_flags & IOCB_DIRECT) ||
2799 !kiocb->ki_filp->f_op->iopoll)
Jens Axboe09bb8392019-03-13 12:39:28 -06002800 return -EOPNOTSUPP;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002801
Jens Axboedef596e2019-01-09 08:59:42 -07002802 kiocb->ki_flags |= IOCB_HIPRI;
2803 kiocb->ki_complete = io_complete_rw_iopoll;
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002804 req->iopoll_completed = 0;
Jens Axboedef596e2019-01-09 08:59:42 -07002805 } else {
Jens Axboe09bb8392019-03-13 12:39:28 -06002806 if (kiocb->ki_flags & IOCB_HIPRI)
2807 return -EINVAL;
Jens Axboedef596e2019-01-09 08:59:42 -07002808 kiocb->ki_complete = io_complete_rw;
2809 }
Jens Axboe9adbd452019-12-20 08:45:55 -07002810
Jens Axboe3529d8c2019-12-19 18:24:38 -07002811 req->rw.addr = READ_ONCE(sqe->addr);
2812 req->rw.len = READ_ONCE(sqe->len);
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002813 req->buf_index = READ_ONCE(sqe->buf_index);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002814 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002815}
2816
2817static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret)
2818{
2819 switch (ret) {
2820 case -EIOCBQUEUED:
2821 break;
2822 case -ERESTARTSYS:
2823 case -ERESTARTNOINTR:
2824 case -ERESTARTNOHAND:
2825 case -ERESTART_RESTARTBLOCK:
2826 /*
2827 * We can't just restart the syscall, since previously
2828 * submitted sqes may already be in progress. Just fail this
2829 * IO with EINTR.
2830 */
2831 ret = -EINTR;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05002832 fallthrough;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002833 default:
2834 kiocb->ki_complete(kiocb, ret, 0);
2835 }
2836}
2837
Jens Axboea1d7c392020-06-22 11:09:46 -06002838static void kiocb_done(struct kiocb *kiocb, ssize_t ret,
2839 struct io_comp_state *cs)
Jens Axboeba816ad2019-09-28 11:36:45 -06002840{
Jens Axboeba042912019-12-25 16:33:42 -07002841 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboee8c2bc12020-08-15 18:44:09 -07002842 struct io_async_rw *io = req->async_data;
Jens Axboeba042912019-12-25 16:33:42 -07002843
Jens Axboe227c0c92020-08-13 11:51:40 -06002844 /* add previously done IO, if any */
Jens Axboee8c2bc12020-08-15 18:44:09 -07002845 if (io && io->bytes_done > 0) {
Jens Axboe227c0c92020-08-13 11:51:40 -06002846 if (ret < 0)
Jens Axboee8c2bc12020-08-15 18:44:09 -07002847 ret = io->bytes_done;
Jens Axboe227c0c92020-08-13 11:51:40 -06002848 else
Jens Axboee8c2bc12020-08-15 18:44:09 -07002849 ret += io->bytes_done;
Jens Axboe227c0c92020-08-13 11:51:40 -06002850 }
2851
Jens Axboeba042912019-12-25 16:33:42 -07002852 if (req->flags & REQ_F_CUR_POS)
2853 req->file->f_pos = kiocb->ki_pos;
Pavel Begunkovbcaec082020-02-24 11:30:18 +03002854 if (ret >= 0 && kiocb->ki_complete == io_complete_rw)
Jens Axboea1d7c392020-06-22 11:09:46 -06002855 __io_complete_rw(req, ret, 0, cs);
Jens Axboeba816ad2019-09-28 11:36:45 -06002856 else
2857 io_rw_done(kiocb, ret);
2858}
2859
Jens Axboe9adbd452019-12-20 08:45:55 -07002860static ssize_t io_import_fixed(struct io_kiocb *req, int rw,
Pavel Begunkov7d009162019-11-25 23:14:40 +03002861 struct iov_iter *iter)
Jens Axboeedafcce2019-01-09 09:16:05 -07002862{
Jens Axboe9adbd452019-12-20 08:45:55 -07002863 struct io_ring_ctx *ctx = req->ctx;
2864 size_t len = req->rw.len;
Jens Axboeedafcce2019-01-09 09:16:05 -07002865 struct io_mapped_ubuf *imu;
Pavel Begunkov4be1c612020-09-06 00:45:48 +03002866 u16 index, buf_index = req->buf_index;
Jens Axboeedafcce2019-01-09 09:16:05 -07002867 size_t offset;
2868 u64 buf_addr;
2869
Jens Axboeedafcce2019-01-09 09:16:05 -07002870 if (unlikely(buf_index >= ctx->nr_user_bufs))
2871 return -EFAULT;
Jens Axboeedafcce2019-01-09 09:16:05 -07002872 index = array_index_nospec(buf_index, ctx->nr_user_bufs);
2873 imu = &ctx->user_bufs[index];
Jens Axboe9adbd452019-12-20 08:45:55 -07002874 buf_addr = req->rw.addr;
Jens Axboeedafcce2019-01-09 09:16:05 -07002875
2876 /* overflow */
2877 if (buf_addr + len < buf_addr)
2878 return -EFAULT;
2879 /* not inside the mapped region */
2880 if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len)
2881 return -EFAULT;
2882
2883 /*
2884 * May not be a start of buffer, set size appropriately
2885 * and advance us to the beginning.
2886 */
2887 offset = buf_addr - imu->ubuf;
2888 iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len);
Jens Axboebd11b3a2019-07-20 08:37:31 -06002889
2890 if (offset) {
2891 /*
2892 * Don't use iov_iter_advance() here, as it's really slow for
2893 * using the latter parts of a big fixed buffer - it iterates
2894 * over each segment manually. We can cheat a bit here, because
2895 * we know that:
2896 *
2897 * 1) it's a BVEC iter, we set it up
2898 * 2) all bvecs are PAGE_SIZE in size, except potentially the
2899 * first and last bvec
2900 *
2901 * So just find our index, and adjust the iterator afterwards.
2902 * If the offset is within the first bvec (or the whole first
2903 * bvec, just use iov_iter_advance(). This makes it easier
2904 * since we can just skip the first segment, which may not
2905 * be PAGE_SIZE aligned.
2906 */
2907 const struct bio_vec *bvec = imu->bvec;
2908
2909 if (offset <= bvec->bv_len) {
2910 iov_iter_advance(iter, offset);
2911 } else {
2912 unsigned long seg_skip;
2913
2914 /* skip first vec */
2915 offset -= bvec->bv_len;
2916 seg_skip = 1 + (offset >> PAGE_SHIFT);
2917
2918 iter->bvec = bvec + seg_skip;
2919 iter->nr_segs -= seg_skip;
Aleix Roca Nonell99c79f62019-08-15 14:03:22 +02002920 iter->count -= bvec->bv_len + offset;
Jens Axboebd11b3a2019-07-20 08:37:31 -06002921 iter->iov_offset = offset & ~PAGE_MASK;
Jens Axboebd11b3a2019-07-20 08:37:31 -06002922 }
2923 }
2924
Jens Axboe5e559562019-11-13 16:12:46 -07002925 return len;
Jens Axboeedafcce2019-01-09 09:16:05 -07002926}
2927
Jens Axboebcda7ba2020-02-23 16:42:51 -07002928static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock)
2929{
2930 if (needs_lock)
2931 mutex_unlock(&ctx->uring_lock);
2932}
2933
2934static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock)
2935{
2936 /*
2937 * "Normal" inline submissions always hold the uring_lock, since we
2938 * grab it from the system call. Same is true for the SQPOLL offload.
2939 * The only exception is when we've detached the request and issue it
2940 * from an async worker thread, grab the lock for that case.
2941 */
2942 if (needs_lock)
2943 mutex_lock(&ctx->uring_lock);
2944}
2945
2946static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len,
2947 int bgid, struct io_buffer *kbuf,
2948 bool needs_lock)
2949{
2950 struct io_buffer *head;
2951
2952 if (req->flags & REQ_F_BUFFER_SELECTED)
2953 return kbuf;
2954
2955 io_ring_submit_lock(req->ctx, needs_lock);
2956
2957 lockdep_assert_held(&req->ctx->uring_lock);
2958
2959 head = idr_find(&req->ctx->io_buffer_idr, bgid);
2960 if (head) {
2961 if (!list_empty(&head->list)) {
2962 kbuf = list_last_entry(&head->list, struct io_buffer,
2963 list);
2964 list_del(&kbuf->list);
2965 } else {
2966 kbuf = head;
2967 idr_remove(&req->ctx->io_buffer_idr, bgid);
2968 }
2969 if (*len > kbuf->len)
2970 *len = kbuf->len;
2971 } else {
2972 kbuf = ERR_PTR(-ENOBUFS);
2973 }
2974
2975 io_ring_submit_unlock(req->ctx, needs_lock);
2976
2977 return kbuf;
2978}
2979
Jens Axboe4d954c22020-02-27 07:31:19 -07002980static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len,
2981 bool needs_lock)
2982{
2983 struct io_buffer *kbuf;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002984 u16 bgid;
Jens Axboe4d954c22020-02-27 07:31:19 -07002985
2986 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07002987 bgid = req->buf_index;
Jens Axboe4d954c22020-02-27 07:31:19 -07002988 kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock);
2989 if (IS_ERR(kbuf))
2990 return kbuf;
2991 req->rw.addr = (u64) (unsigned long) kbuf;
2992 req->flags |= REQ_F_BUFFER_SELECTED;
2993 return u64_to_user_ptr(kbuf->addr);
2994}
2995
2996#ifdef CONFIG_COMPAT
2997static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov,
2998 bool needs_lock)
2999{
3000 struct compat_iovec __user *uiov;
3001 compat_ssize_t clen;
3002 void __user *buf;
3003 ssize_t len;
3004
3005 uiov = u64_to_user_ptr(req->rw.addr);
3006 if (!access_ok(uiov, sizeof(*uiov)))
3007 return -EFAULT;
3008 if (__get_user(clen, &uiov->iov_len))
3009 return -EFAULT;
3010 if (clen < 0)
3011 return -EINVAL;
3012
3013 len = clen;
3014 buf = io_rw_buffer_select(req, &len, needs_lock);
3015 if (IS_ERR(buf))
3016 return PTR_ERR(buf);
3017 iov[0].iov_base = buf;
3018 iov[0].iov_len = (compat_size_t) len;
3019 return 0;
3020}
3021#endif
3022
3023static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov,
3024 bool needs_lock)
3025{
3026 struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr);
3027 void __user *buf;
3028 ssize_t len;
3029
3030 if (copy_from_user(iov, uiov, sizeof(*uiov)))
3031 return -EFAULT;
3032
3033 len = iov[0].iov_len;
3034 if (len < 0)
3035 return -EINVAL;
3036 buf = io_rw_buffer_select(req, &len, needs_lock);
3037 if (IS_ERR(buf))
3038 return PTR_ERR(buf);
3039 iov[0].iov_base = buf;
3040 iov[0].iov_len = len;
3041 return 0;
3042}
3043
3044static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov,
3045 bool needs_lock)
3046{
Jens Axboedddb3e22020-06-04 11:27:01 -06003047 if (req->flags & REQ_F_BUFFER_SELECTED) {
3048 struct io_buffer *kbuf;
3049
3050 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
3051 iov[0].iov_base = u64_to_user_ptr(kbuf->addr);
3052 iov[0].iov_len = kbuf->len;
Jens Axboe4d954c22020-02-27 07:31:19 -07003053 return 0;
Jens Axboedddb3e22020-06-04 11:27:01 -06003054 }
Jens Axboe4d954c22020-02-27 07:31:19 -07003055 if (!req->rw.len)
3056 return 0;
3057 else if (req->rw.len > 1)
3058 return -EINVAL;
3059
3060#ifdef CONFIG_COMPAT
3061 if (req->ctx->compat)
3062 return io_compat_import(req, iov, needs_lock);
3063#endif
3064
3065 return __io_iov_buffer_select(req, iov, needs_lock);
3066}
3067
Jens Axboe8452fd02020-08-18 13:58:33 -07003068static ssize_t __io_import_iovec(int rw, struct io_kiocb *req,
3069 struct iovec **iovec, struct iov_iter *iter,
3070 bool needs_lock)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003071{
Jens Axboe9adbd452019-12-20 08:45:55 -07003072 void __user *buf = u64_to_user_ptr(req->rw.addr);
3073 size_t sqe_len = req->rw.len;
Jens Axboe4d954c22020-02-27 07:31:19 -07003074 ssize_t ret;
Jens Axboeedafcce2019-01-09 09:16:05 -07003075 u8 opcode;
3076
Jens Axboed625c6e2019-12-17 19:53:05 -07003077 opcode = req->opcode;
Pavel Begunkov7d009162019-11-25 23:14:40 +03003078 if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) {
Jens Axboeedafcce2019-01-09 09:16:05 -07003079 *iovec = NULL;
Jens Axboe9adbd452019-12-20 08:45:55 -07003080 return io_import_fixed(req, rw, iter);
Jens Axboeedafcce2019-01-09 09:16:05 -07003081 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07003082
Jens Axboebcda7ba2020-02-23 16:42:51 -07003083 /* buffer index only valid with fixed read/write, or buffer select */
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07003084 if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT))
Jens Axboe9adbd452019-12-20 08:45:55 -07003085 return -EINVAL;
3086
Jens Axboe3a6820f2019-12-22 15:19:35 -07003087 if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) {
Jens Axboebcda7ba2020-02-23 16:42:51 -07003088 if (req->flags & REQ_F_BUFFER_SELECT) {
Jens Axboe4d954c22020-02-27 07:31:19 -07003089 buf = io_rw_buffer_select(req, &sqe_len, needs_lock);
Pavel Begunkov867a23e2020-08-20 11:34:39 +03003090 if (IS_ERR(buf))
Jens Axboe4d954c22020-02-27 07:31:19 -07003091 return PTR_ERR(buf);
Jens Axboe3f9d6442020-03-11 12:27:04 -06003092 req->rw.len = sqe_len;
Jens Axboebcda7ba2020-02-23 16:42:51 -07003093 }
3094
Jens Axboe3a6820f2019-12-22 15:19:35 -07003095 ret = import_single_range(rw, buf, sqe_len, *iovec, iter);
3096 *iovec = NULL;
Jens Axboe3a901592020-02-25 17:48:55 -07003097 return ret < 0 ? ret : sqe_len;
Jens Axboe3a6820f2019-12-22 15:19:35 -07003098 }
3099
Jens Axboe4d954c22020-02-27 07:31:19 -07003100 if (req->flags & REQ_F_BUFFER_SELECT) {
3101 ret = io_iov_buffer_select(req, *iovec, needs_lock);
Jens Axboe3f9d6442020-03-11 12:27:04 -06003102 if (!ret) {
3103 ret = (*iovec)->iov_len;
3104 iov_iter_init(iter, rw, *iovec, 1, ret);
3105 }
Jens Axboe4d954c22020-02-27 07:31:19 -07003106 *iovec = NULL;
3107 return ret;
3108 }
3109
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02003110 return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter,
3111 req->ctx->compat);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003112}
3113
Jens Axboe8452fd02020-08-18 13:58:33 -07003114static ssize_t io_import_iovec(int rw, struct io_kiocb *req,
3115 struct iovec **iovec, struct iov_iter *iter,
3116 bool needs_lock)
3117{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003118 struct io_async_rw *iorw = req->async_data;
3119
3120 if (!iorw)
Jens Axboe8452fd02020-08-18 13:58:33 -07003121 return __io_import_iovec(rw, req, iovec, iter, needs_lock);
3122 *iovec = NULL;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003123 return iov_iter_count(&iorw->iter);
Jens Axboe8452fd02020-08-18 13:58:33 -07003124}
3125
Jens Axboe0fef9482020-08-26 10:36:20 -06003126static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb)
3127{
Pavel Begunkov5b09e372020-09-30 22:57:15 +03003128 return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos;
Jens Axboe0fef9482020-08-26 10:36:20 -06003129}
3130
Jens Axboe32960612019-09-23 11:05:34 -06003131/*
3132 * For files that don't have ->read_iter() and ->write_iter(), handle them
3133 * by looping over ->read() or ->write() manually.
3134 */
Jens Axboe4017eb92020-10-22 14:14:12 -06003135static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter)
Jens Axboe32960612019-09-23 11:05:34 -06003136{
Jens Axboe4017eb92020-10-22 14:14:12 -06003137 struct kiocb *kiocb = &req->rw.kiocb;
3138 struct file *file = req->file;
Jens Axboe32960612019-09-23 11:05:34 -06003139 ssize_t ret = 0;
3140
3141 /*
3142 * Don't support polled IO through this interface, and we can't
3143 * support non-blocking either. For the latter, this just causes
3144 * the kiocb to be handled from an async context.
3145 */
3146 if (kiocb->ki_flags & IOCB_HIPRI)
3147 return -EOPNOTSUPP;
3148 if (kiocb->ki_flags & IOCB_NOWAIT)
3149 return -EAGAIN;
3150
3151 while (iov_iter_count(iter)) {
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003152 struct iovec iovec;
Jens Axboe32960612019-09-23 11:05:34 -06003153 ssize_t nr;
3154
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003155 if (!iov_iter_is_bvec(iter)) {
3156 iovec = iov_iter_iovec(iter);
3157 } else {
Jens Axboe4017eb92020-10-22 14:14:12 -06003158 iovec.iov_base = u64_to_user_ptr(req->rw.addr);
3159 iovec.iov_len = req->rw.len;
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003160 }
3161
Jens Axboe32960612019-09-23 11:05:34 -06003162 if (rw == READ) {
3163 nr = file->f_op->read(file, iovec.iov_base,
Jens Axboe0fef9482020-08-26 10:36:20 -06003164 iovec.iov_len, io_kiocb_ppos(kiocb));
Jens Axboe32960612019-09-23 11:05:34 -06003165 } else {
3166 nr = file->f_op->write(file, iovec.iov_base,
Jens Axboe0fef9482020-08-26 10:36:20 -06003167 iovec.iov_len, io_kiocb_ppos(kiocb));
Jens Axboe32960612019-09-23 11:05:34 -06003168 }
3169
3170 if (nr < 0) {
3171 if (!ret)
3172 ret = nr;
3173 break;
3174 }
3175 ret += nr;
3176 if (nr != iovec.iov_len)
3177 break;
Jens Axboe4017eb92020-10-22 14:14:12 -06003178 req->rw.len -= nr;
3179 req->rw.addr += nr;
Jens Axboe32960612019-09-23 11:05:34 -06003180 iov_iter_advance(iter, nr);
3181 }
3182
3183 return ret;
3184}
3185
Jens Axboeff6165b2020-08-13 09:47:43 -06003186static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec,
3187 const struct iovec *fast_iov, struct iov_iter *iter)
Jens Axboef67676d2019-12-02 11:03:47 -07003188{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003189 struct io_async_rw *rw = req->async_data;
Pavel Begunkovb64e3442020-07-13 22:59:18 +03003190
Jens Axboeff6165b2020-08-13 09:47:43 -06003191 memcpy(&rw->iter, iter, sizeof(*iter));
Pavel Begunkovafb87652020-09-06 00:45:46 +03003192 rw->free_iovec = iovec;
Jens Axboe227c0c92020-08-13 11:51:40 -06003193 rw->bytes_done = 0;
Jens Axboeff6165b2020-08-13 09:47:43 -06003194 /* can only be fixed buffers, no need to do anything */
3195 if (iter->type == ITER_BVEC)
3196 return;
Pavel Begunkovb64e3442020-07-13 22:59:18 +03003197 if (!iovec) {
Jens Axboeff6165b2020-08-13 09:47:43 -06003198 unsigned iov_off = 0;
3199
3200 rw->iter.iov = rw->fast_iov;
3201 if (iter->iov != fast_iov) {
3202 iov_off = iter->iov - fast_iov;
3203 rw->iter.iov += iov_off;
3204 }
3205 if (rw->fast_iov != fast_iov)
3206 memcpy(rw->fast_iov + iov_off, fast_iov + iov_off,
Xiaoguang Wang45097da2020-04-08 22:29:58 +08003207 sizeof(struct iovec) * iter->nr_segs);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03003208 } else {
3209 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboef67676d2019-12-02 11:03:47 -07003210 }
3211}
3212
Jens Axboee8c2bc12020-08-15 18:44:09 -07003213static inline int __io_alloc_async_data(struct io_kiocb *req)
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003214{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003215 WARN_ON_ONCE(!io_op_defs[req->opcode].async_size);
3216 req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL);
3217 return req->async_data == NULL;
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003218}
3219
Jens Axboee8c2bc12020-08-15 18:44:09 -07003220static int io_alloc_async_data(struct io_kiocb *req)
Jens Axboef67676d2019-12-02 11:03:47 -07003221{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003222 if (!io_op_defs[req->opcode].needs_async_data)
Jens Axboed3656342019-12-18 09:50:26 -07003223 return 0;
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003224
Jens Axboee8c2bc12020-08-15 18:44:09 -07003225 return __io_alloc_async_data(req);
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003226}
3227
Jens Axboeff6165b2020-08-13 09:47:43 -06003228static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec,
3229 const struct iovec *fast_iov,
Jens Axboe227c0c92020-08-13 11:51:40 -06003230 struct iov_iter *iter, bool force)
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003231{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003232 if (!force && !io_op_defs[req->opcode].needs_async_data)
Jens Axboe74566df2020-01-13 19:23:24 -07003233 return 0;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003234 if (!req->async_data) {
3235 if (__io_alloc_async_data(req))
Jens Axboe5d204bc2020-01-31 12:06:52 -07003236 return -ENOMEM;
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003237
Jens Axboeff6165b2020-08-13 09:47:43 -06003238 io_req_map_rw(req, iovec, fast_iov, iter);
Jens Axboe5d204bc2020-01-31 12:06:52 -07003239 }
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003240 return 0;
Jens Axboef67676d2019-12-02 11:03:47 -07003241}
3242
Pavel Begunkov73debe62020-09-30 22:57:54 +03003243static inline int io_rw_prep_async(struct io_kiocb *req, int rw)
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003244{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003245 struct io_async_rw *iorw = req->async_data;
Pavel Begunkovf4bff102020-09-06 00:45:45 +03003246 struct iovec *iov = iorw->fast_iov;
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003247 ssize_t ret;
3248
Pavel Begunkov73debe62020-09-30 22:57:54 +03003249 ret = __io_import_iovec(rw, req, &iov, &iorw->iter, false);
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003250 if (unlikely(ret < 0))
3251 return ret;
3252
Pavel Begunkovab0b1962020-09-06 00:45:47 +03003253 iorw->bytes_done = 0;
3254 iorw->free_iovec = iov;
3255 if (iov)
3256 req->flags |= REQ_F_NEED_CLEANUP;
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003257 return 0;
3258}
3259
Pavel Begunkov73debe62020-09-30 22:57:54 +03003260static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07003261{
3262 ssize_t ret;
3263
Pavel Begunkova88fc402020-09-30 22:57:53 +03003264 ret = io_prep_rw(req, sqe);
Jens Axboe3529d8c2019-12-19 18:24:38 -07003265 if (ret)
3266 return ret;
Jens Axboef67676d2019-12-02 11:03:47 -07003267
Jens Axboe3529d8c2019-12-19 18:24:38 -07003268 if (unlikely(!(req->file->f_mode & FMODE_READ)))
3269 return -EBADF;
Jens Axboef67676d2019-12-02 11:03:47 -07003270
Pavel Begunkov5f798be2020-02-08 13:28:02 +03003271 /* either don't need iovec imported or already have it */
Pavel Begunkov2d199892020-09-30 22:57:35 +03003272 if (!req->async_data)
Jens Axboe3529d8c2019-12-19 18:24:38 -07003273 return 0;
Pavel Begunkov73debe62020-09-30 22:57:54 +03003274 return io_rw_prep_async(req, READ);
Jens Axboef67676d2019-12-02 11:03:47 -07003275}
3276
Jens Axboec1dd91d2020-08-03 16:43:59 -06003277/*
3278 * This is our waitqueue callback handler, registered through lock_page_async()
3279 * when we initially tried to do the IO with the iocb armed our waitqueue.
3280 * This gets called when the page is unlocked, and we generally expect that to
3281 * happen when the page IO is completed and the page is now uptodate. This will
3282 * queue a task_work based retry of the operation, attempting to copy the data
3283 * again. If the latter fails because the page was NOT uptodate, then we will
3284 * do a thread based blocking retry of the operation. That's the unexpected
3285 * slow path.
3286 */
Jens Axboebcf5a062020-05-22 09:24:42 -06003287static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode,
3288 int sync, void *arg)
3289{
3290 struct wait_page_queue *wpq;
3291 struct io_kiocb *req = wait->private;
Jens Axboebcf5a062020-05-22 09:24:42 -06003292 struct wait_page_key *key = arg;
Jens Axboebcf5a062020-05-22 09:24:42 -06003293 int ret;
3294
3295 wpq = container_of(wait, struct wait_page_queue, wait);
3296
Linus Torvaldscdc8fcb2020-08-03 13:01:22 -07003297 if (!wake_page_match(wpq, key))
3298 return 0;
3299
Hao Xuc8d317a2020-09-29 20:00:45 +08003300 req->rw.kiocb.ki_flags &= ~IOCB_WAITQ;
Jens Axboebcf5a062020-05-22 09:24:42 -06003301 list_del_init(&wait->entry);
3302
Pavel Begunkove7375122020-07-12 20:42:04 +03003303 init_task_work(&req->task_work, io_req_task_submit);
Jens Axboe6d816e02020-08-11 08:04:14 -06003304 percpu_ref_get(&req->ctx->refs);
3305
Jens Axboebcf5a062020-05-22 09:24:42 -06003306 /* submit ref gets dropped, acquire a new one */
3307 refcount_inc(&req->refs);
Jens Axboe87c43112020-09-30 21:00:14 -06003308 ret = io_req_task_work_add(req, true);
Jens Axboebcf5a062020-05-22 09:24:42 -06003309 if (unlikely(ret)) {
Jens Axboec2c4c832020-07-01 15:37:11 -06003310 struct task_struct *tsk;
3311
Jens Axboebcf5a062020-05-22 09:24:42 -06003312 /* queue just for cancelation */
Pavel Begunkove7375122020-07-12 20:42:04 +03003313 init_task_work(&req->task_work, io_req_task_cancel);
Jens Axboebcf5a062020-05-22 09:24:42 -06003314 tsk = io_wq_get_task(req->ctx->io_wq);
Jens Axboe91989c72020-10-16 09:02:26 -06003315 task_work_add(tsk, &req->task_work, TWA_NONE);
Jens Axboec2c4c832020-07-01 15:37:11 -06003316 wake_up_process(tsk);
Jens Axboebcf5a062020-05-22 09:24:42 -06003317 }
Jens Axboebcf5a062020-05-22 09:24:42 -06003318 return 1;
3319}
3320
Jens Axboec1dd91d2020-08-03 16:43:59 -06003321/*
3322 * This controls whether a given IO request should be armed for async page
3323 * based retry. If we return false here, the request is handed to the async
3324 * worker threads for retry. If we're doing buffered reads on a regular file,
3325 * we prepare a private wait_page_queue entry and retry the operation. This
3326 * will either succeed because the page is now uptodate and unlocked, or it
3327 * will register a callback when the page is unlocked at IO completion. Through
3328 * that callback, io_uring uses task_work to setup a retry of the operation.
3329 * That retry will attempt the buffered read again. The retry will generally
3330 * succeed, or in rare cases where it fails, we then fall back to using the
3331 * async worker threads for a blocking retry.
3332 */
Jens Axboe227c0c92020-08-13 11:51:40 -06003333static bool io_rw_should_retry(struct io_kiocb *req)
Jens Axboebcf5a062020-05-22 09:24:42 -06003334{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003335 struct io_async_rw *rw = req->async_data;
3336 struct wait_page_queue *wait = &rw->wpq;
Jens Axboebcf5a062020-05-22 09:24:42 -06003337 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboebcf5a062020-05-22 09:24:42 -06003338
3339 /* never retry for NOWAIT, we just complete with -EAGAIN */
3340 if (req->flags & REQ_F_NOWAIT)
3341 return false;
3342
Jens Axboe227c0c92020-08-13 11:51:40 -06003343 /* Only for buffered IO */
Jens Axboe3b2a4432020-08-16 10:58:43 -07003344 if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI))
Jens Axboebcf5a062020-05-22 09:24:42 -06003345 return false;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003346
Jens Axboebcf5a062020-05-22 09:24:42 -06003347 /*
3348 * just use poll if we can, and don't attempt if the fs doesn't
3349 * support callback based unlocks
3350 */
3351 if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC))
3352 return false;
3353
Jens Axboe3b2a4432020-08-16 10:58:43 -07003354 wait->wait.func = io_async_buf_func;
3355 wait->wait.private = req;
3356 wait->wait.flags = 0;
3357 INIT_LIST_HEAD(&wait->wait.entry);
3358 kiocb->ki_flags |= IOCB_WAITQ;
Hao Xuc8d317a2020-09-29 20:00:45 +08003359 kiocb->ki_flags &= ~IOCB_NOWAIT;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003360 kiocb->ki_waitq = wait;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003361 return true;
Jens Axboebcf5a062020-05-22 09:24:42 -06003362}
3363
3364static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter)
3365{
3366 if (req->file->f_op->read_iter)
3367 return call_read_iter(req->file, &req->rw.kiocb, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003368 else if (req->file->f_op->read)
Jens Axboe4017eb92020-10-22 14:14:12 -06003369 return loop_rw_iter(READ, req, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003370 else
3371 return -EINVAL;
Jens Axboebcf5a062020-05-22 09:24:42 -06003372}
3373
Jens Axboea1d7c392020-06-22 11:09:46 -06003374static int io_read(struct io_kiocb *req, bool force_nonblock,
3375 struct io_comp_state *cs)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003376{
3377 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
Jens Axboe9adbd452019-12-20 08:45:55 -07003378 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboeff6165b2020-08-13 09:47:43 -06003379 struct iov_iter __iter, *iter = &__iter;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003380 struct io_async_rw *rw = req->async_data;
Jens Axboe227c0c92020-08-13 11:51:40 -06003381 ssize_t io_size, ret, ret2;
Jens Axboe31b51512019-01-18 22:56:34 -07003382 size_t iov_count;
Jens Axboef5cac8b2020-09-14 09:30:38 -06003383 bool no_async;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003384
Jens Axboee8c2bc12020-08-15 18:44:09 -07003385 if (rw)
3386 iter = &rw->iter;
Jens Axboeff6165b2020-08-13 09:47:43 -06003387
3388 ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock);
Jens Axboe06b76d42019-12-19 14:44:26 -07003389 if (ret < 0)
3390 return ret;
Jens Axboeeefdf302020-08-27 16:40:19 -06003391 iov_count = iov_iter_count(iter);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003392 io_size = ret;
3393 req->result = io_size;
Jens Axboe227c0c92020-08-13 11:51:40 -06003394 ret = 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003395
Jens Axboefd6c2e42019-12-18 12:19:41 -07003396 /* Ensure we clear previously set non-block flag */
3397 if (!force_nonblock)
Jens Axboe29de5f62020-02-20 09:56:08 -07003398 kiocb->ki_flags &= ~IOCB_NOWAIT;
Pavel Begunkova88fc402020-09-30 22:57:53 +03003399 else
3400 kiocb->ki_flags |= IOCB_NOWAIT;
3401
Jens Axboefd6c2e42019-12-18 12:19:41 -07003402
Pavel Begunkov24c74672020-06-21 13:09:51 +03003403 /* If the file doesn't support async, just async punt */
Jens Axboef5cac8b2020-09-14 09:30:38 -06003404 no_async = force_nonblock && !io_file_supports_async(req->file, READ);
3405 if (no_async)
Jens Axboef67676d2019-12-02 11:03:47 -07003406 goto copy_iov;
Jens Axboe9e645e112019-05-10 16:07:28 -06003407
Jens Axboe0fef9482020-08-26 10:36:20 -06003408 ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), iov_count);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003409 if (unlikely(ret))
3410 goto out_free;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003411
Jens Axboe227c0c92020-08-13 11:51:40 -06003412 ret = io_iter_do_read(req, iter);
Jens Axboe32960612019-09-23 11:05:34 -06003413
Jens Axboe227c0c92020-08-13 11:51:40 -06003414 if (!ret) {
3415 goto done;
3416 } else if (ret == -EIOCBQUEUED) {
3417 ret = 0;
3418 goto out_free;
3419 } else if (ret == -EAGAIN) {
Jens Axboeeefdf302020-08-27 16:40:19 -06003420 /* IOPOLL retry should happen for io-wq threads */
3421 if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboef91daf52020-08-15 15:58:42 -07003422 goto done;
Jens Axboe355afae2020-09-02 09:30:31 -06003423 /* no retry on NONBLOCK marked file */
3424 if (req->file->f_flags & O_NONBLOCK)
3425 goto done;
Jens Axboe84216312020-08-24 11:45:26 -06003426 /* some cases will consume bytes even on error returns */
3427 iov_iter_revert(iter, iov_count - iov_iter_count(iter));
Jens Axboef38c7e32020-09-25 15:23:43 -06003428 ret = 0;
3429 goto copy_iov;
Jens Axboe227c0c92020-08-13 11:51:40 -06003430 } else if (ret < 0) {
Jens Axboe00d23d52020-08-25 12:59:22 -06003431 /* make sure -ERESTARTSYS -> -EINTR is done */
3432 goto done;
Jens Axboe227c0c92020-08-13 11:51:40 -06003433 }
3434
3435 /* read it all, or we did blocking attempt. no retry. */
Jens Axboef91daf52020-08-15 15:58:42 -07003436 if (!iov_iter_count(iter) || !force_nonblock ||
3437 (req->file->f_flags & O_NONBLOCK))
Jens Axboe227c0c92020-08-13 11:51:40 -06003438 goto done;
3439
3440 io_size -= ret;
3441copy_iov:
3442 ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true);
3443 if (ret2) {
3444 ret = ret2;
3445 goto out_free;
3446 }
Jens Axboef5cac8b2020-09-14 09:30:38 -06003447 if (no_async)
3448 return -EAGAIN;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003449 rw = req->async_data;
Jens Axboe227c0c92020-08-13 11:51:40 -06003450 /* it's copied and will be cleaned with ->io */
3451 iovec = NULL;
3452 /* now use our persistent iterator, if we aren't already */
Jens Axboee8c2bc12020-08-15 18:44:09 -07003453 iter = &rw->iter;
Jens Axboe227c0c92020-08-13 11:51:40 -06003454retry:
Jens Axboee8c2bc12020-08-15 18:44:09 -07003455 rw->bytes_done += ret;
Jens Axboe227c0c92020-08-13 11:51:40 -06003456 /* if we can retry, do so with the callbacks armed */
3457 if (!io_rw_should_retry(req)) {
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003458 kiocb->ki_flags &= ~IOCB_WAITQ;
3459 return -EAGAIN;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003460 }
Jens Axboe227c0c92020-08-13 11:51:40 -06003461
3462 /*
3463 * Now retry read with the IOCB_WAITQ parts set in the iocb. If we
3464 * get -EIOCBQUEUED, then we'll get a notification when the desired
3465 * page gets unlocked. We can also get a partial read here, and if we
3466 * do, then just retry at the new offset.
3467 */
3468 ret = io_iter_do_read(req, iter);
3469 if (ret == -EIOCBQUEUED) {
3470 ret = 0;
3471 goto out_free;
3472 } else if (ret > 0 && ret < io_size) {
3473 /* we got some bytes, but not all. retry. */
3474 goto retry;
3475 }
3476done:
3477 kiocb_done(kiocb, ret, cs);
3478 ret = 0;
Jens Axboef67676d2019-12-02 11:03:47 -07003479out_free:
Pavel Begunkovf261c162020-08-20 11:34:10 +03003480 /* it's reportedly faster than delegating the null check to kfree() */
Pavel Begunkov252917c2020-07-13 22:59:20 +03003481 if (iovec)
Xiaoguang Wang6f2cc162020-06-18 15:01:56 +08003482 kfree(iovec);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003483 return ret;
3484}
3485
Pavel Begunkov73debe62020-09-30 22:57:54 +03003486static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07003487{
3488 ssize_t ret;
3489
Pavel Begunkova88fc402020-09-30 22:57:53 +03003490 ret = io_prep_rw(req, sqe);
Jens Axboe3529d8c2019-12-19 18:24:38 -07003491 if (ret)
3492 return ret;
Jens Axboef67676d2019-12-02 11:03:47 -07003493
Jens Axboe3529d8c2019-12-19 18:24:38 -07003494 if (unlikely(!(req->file->f_mode & FMODE_WRITE)))
3495 return -EBADF;
Jens Axboef67676d2019-12-02 11:03:47 -07003496
Pavel Begunkov5f798be2020-02-08 13:28:02 +03003497 /* either don't need iovec imported or already have it */
Pavel Begunkov2d199892020-09-30 22:57:35 +03003498 if (!req->async_data)
Jens Axboe3529d8c2019-12-19 18:24:38 -07003499 return 0;
Pavel Begunkov73debe62020-09-30 22:57:54 +03003500 return io_rw_prep_async(req, WRITE);
Jens Axboef67676d2019-12-02 11:03:47 -07003501}
3502
Jens Axboea1d7c392020-06-22 11:09:46 -06003503static int io_write(struct io_kiocb *req, bool force_nonblock,
3504 struct io_comp_state *cs)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003505{
3506 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
Jens Axboe9adbd452019-12-20 08:45:55 -07003507 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboeff6165b2020-08-13 09:47:43 -06003508 struct iov_iter __iter, *iter = &__iter;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003509 struct io_async_rw *rw = req->async_data;
Jens Axboe31b51512019-01-18 22:56:34 -07003510 size_t iov_count;
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003511 ssize_t ret, ret2, io_size;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003512
Jens Axboee8c2bc12020-08-15 18:44:09 -07003513 if (rw)
3514 iter = &rw->iter;
Jens Axboeff6165b2020-08-13 09:47:43 -06003515
3516 ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock);
Jens Axboe06b76d42019-12-19 14:44:26 -07003517 if (ret < 0)
3518 return ret;
Jens Axboeeefdf302020-08-27 16:40:19 -06003519 iov_count = iov_iter_count(iter);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003520 io_size = ret;
3521 req->result = io_size;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003522
Jens Axboefd6c2e42019-12-18 12:19:41 -07003523 /* Ensure we clear previously set non-block flag */
3524 if (!force_nonblock)
Pavel Begunkova88fc402020-09-30 22:57:53 +03003525 kiocb->ki_flags &= ~IOCB_NOWAIT;
3526 else
3527 kiocb->ki_flags |= IOCB_NOWAIT;
Jens Axboefd6c2e42019-12-18 12:19:41 -07003528
Pavel Begunkov24c74672020-06-21 13:09:51 +03003529 /* If the file doesn't support async, just async punt */
Jens Axboeaf197f52020-04-28 13:15:06 -06003530 if (force_nonblock && !io_file_supports_async(req->file, WRITE))
Jens Axboef67676d2019-12-02 11:03:47 -07003531 goto copy_iov;
Jens Axboef67676d2019-12-02 11:03:47 -07003532
Jens Axboe10d59342019-12-09 20:16:22 -07003533 /* file path doesn't support NOWAIT for non-direct_IO */
3534 if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) &&
3535 (req->flags & REQ_F_ISREG))
Jens Axboef67676d2019-12-02 11:03:47 -07003536 goto copy_iov;
Jens Axboe9e645e112019-05-10 16:07:28 -06003537
Jens Axboe0fef9482020-08-26 10:36:20 -06003538 ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), iov_count);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003539 if (unlikely(ret))
3540 goto out_free;
Roman Penyaev9bf79332019-03-25 20:09:24 +01003541
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003542 /*
3543 * Open-code file_start_write here to grab freeze protection,
3544 * which will be released by another thread in
3545 * io_complete_rw(). Fool lockdep by telling it the lock got
3546 * released so that it doesn't complain about the held lock when
3547 * we return to userspace.
3548 */
3549 if (req->flags & REQ_F_ISREG) {
3550 __sb_start_write(file_inode(req->file)->i_sb,
3551 SB_FREEZE_WRITE, true);
3552 __sb_writers_release(file_inode(req->file)->i_sb,
3553 SB_FREEZE_WRITE);
3554 }
3555 kiocb->ki_flags |= IOCB_WRITE;
Roman Penyaev9bf79332019-03-25 20:09:24 +01003556
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003557 if (req->file->f_op->write_iter)
Jens Axboeff6165b2020-08-13 09:47:43 -06003558 ret2 = call_write_iter(req->file, kiocb, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003559 else if (req->file->f_op->write)
Jens Axboe4017eb92020-10-22 14:14:12 -06003560 ret2 = loop_rw_iter(WRITE, req, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003561 else
3562 ret2 = -EINVAL;
Jens Axboe4ed734b2020-03-20 11:23:41 -06003563
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003564 /*
3565 * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just
3566 * retry them without IOCB_NOWAIT.
3567 */
3568 if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT))
3569 ret2 = -EAGAIN;
Jens Axboe355afae2020-09-02 09:30:31 -06003570 /* no retry on NONBLOCK marked file */
3571 if (ret2 == -EAGAIN && (req->file->f_flags & O_NONBLOCK))
3572 goto done;
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003573 if (!force_nonblock || ret2 != -EAGAIN) {
Jens Axboeeefdf302020-08-27 16:40:19 -06003574 /* IOPOLL retry should happen for io-wq threads */
3575 if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN)
3576 goto copy_iov;
Jens Axboe355afae2020-09-02 09:30:31 -06003577done:
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003578 kiocb_done(kiocb, ret2, cs);
3579 } else {
Jens Axboef67676d2019-12-02 11:03:47 -07003580copy_iov:
Jens Axboe84216312020-08-24 11:45:26 -06003581 /* some cases will consume bytes even on error returns */
3582 iov_iter_revert(iter, iov_count - iov_iter_count(iter));
Jens Axboe227c0c92020-08-13 11:51:40 -06003583 ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false);
Jens Axboeff6165b2020-08-13 09:47:43 -06003584 if (!ret)
3585 return -EAGAIN;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003586 }
Jens Axboe31b51512019-01-18 22:56:34 -07003587out_free:
Pavel Begunkovf261c162020-08-20 11:34:10 +03003588 /* it's reportedly faster than delegating the null check to kfree() */
Pavel Begunkov252917c2020-07-13 22:59:20 +03003589 if (iovec)
Xiaoguang Wang6f2cc162020-06-18 15:01:56 +08003590 kfree(iovec);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003591 return ret;
3592}
3593
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003594static int __io_splice_prep(struct io_kiocb *req,
3595 const struct io_uring_sqe *sqe)
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003596{
3597 struct io_splice* sp = &req->splice;
3598 unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003599
Pavel Begunkov3232dd02020-06-03 18:03:22 +03003600 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3601 return -EINVAL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003602
3603 sp->file_in = NULL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003604 sp->len = READ_ONCE(sqe->len);
3605 sp->flags = READ_ONCE(sqe->splice_flags);
3606
3607 if (unlikely(sp->flags & ~valid_flags))
3608 return -EINVAL;
3609
Pavel Begunkov8371adf2020-10-10 18:34:08 +01003610 sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in),
3611 (sp->flags & SPLICE_F_FD_IN_FIXED));
3612 if (!sp->file_in)
3613 return -EBADF;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003614 req->flags |= REQ_F_NEED_CLEANUP;
3615
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08003616 if (!S_ISREG(file_inode(sp->file_in)->i_mode)) {
3617 /*
3618 * Splice operation will be punted aync, and here need to
3619 * modify io_wq_work.flags, so initialize io_wq_work firstly.
3620 */
3621 io_req_init_async(req);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003622 req->work.flags |= IO_WQ_WORK_UNBOUND;
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08003623 }
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003624
3625 return 0;
3626}
3627
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003628static int io_tee_prep(struct io_kiocb *req,
3629 const struct io_uring_sqe *sqe)
3630{
3631 if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off))
3632 return -EINVAL;
3633 return __io_splice_prep(req, sqe);
3634}
3635
3636static int io_tee(struct io_kiocb *req, bool force_nonblock)
3637{
3638 struct io_splice *sp = &req->splice;
3639 struct file *in = sp->file_in;
3640 struct file *out = sp->file_out;
3641 unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED;
3642 long ret = 0;
3643
3644 if (force_nonblock)
3645 return -EAGAIN;
3646 if (sp->len)
3647 ret = do_tee(in, out, sp->len, flags);
3648
3649 io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED));
3650 req->flags &= ~REQ_F_NEED_CLEANUP;
3651
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003652 if (ret != sp->len)
3653 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003654 io_req_complete(req, ret);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003655 return 0;
3656}
3657
3658static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3659{
3660 struct io_splice* sp = &req->splice;
3661
3662 sp->off_in = READ_ONCE(sqe->splice_off_in);
3663 sp->off_out = READ_ONCE(sqe->off);
3664 return __io_splice_prep(req, sqe);
3665}
3666
Pavel Begunkov014db002020-03-03 21:33:12 +03003667static int io_splice(struct io_kiocb *req, bool force_nonblock)
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003668{
3669 struct io_splice *sp = &req->splice;
3670 struct file *in = sp->file_in;
3671 struct file *out = sp->file_out;
3672 unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED;
3673 loff_t *poff_in, *poff_out;
Pavel Begunkovc9687422020-05-04 23:00:54 +03003674 long ret = 0;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003675
Pavel Begunkov2fb3e822020-05-01 17:09:38 +03003676 if (force_nonblock)
3677 return -EAGAIN;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003678
3679 poff_in = (sp->off_in == -1) ? NULL : &sp->off_in;
3680 poff_out = (sp->off_out == -1) ? NULL : &sp->off_out;
Pavel Begunkovc9687422020-05-04 23:00:54 +03003681
Jens Axboe948a7742020-05-17 14:21:38 -06003682 if (sp->len)
Pavel Begunkovc9687422020-05-04 23:00:54 +03003683 ret = do_splice(in, poff_in, out, poff_out, sp->len, flags);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003684
3685 io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED));
3686 req->flags &= ~REQ_F_NEED_CLEANUP;
3687
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003688 if (ret != sp->len)
3689 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003690 io_req_complete(req, ret);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003691 return 0;
3692}
3693
Jens Axboe2b188cc2019-01-07 10:46:33 -07003694/*
3695 * IORING_OP_NOP just posts a completion event, nothing else.
3696 */
Jens Axboe229a7b62020-06-22 10:13:11 -06003697static int io_nop(struct io_kiocb *req, struct io_comp_state *cs)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003698{
3699 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003700
Jens Axboedef596e2019-01-09 08:59:42 -07003701 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
3702 return -EINVAL;
3703
Jens Axboe229a7b62020-06-22 10:13:11 -06003704 __io_req_complete(req, 0, 0, cs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003705 return 0;
3706}
3707
Jens Axboe3529d8c2019-12-19 18:24:38 -07003708static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003709{
Jens Axboe6b063142019-01-10 22:13:58 -07003710 struct io_ring_ctx *ctx = req->ctx;
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003711
Jens Axboe09bb8392019-03-13 12:39:28 -06003712 if (!req->file)
3713 return -EBADF;
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003714
Jens Axboe6b063142019-01-10 22:13:58 -07003715 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboedef596e2019-01-09 08:59:42 -07003716 return -EINVAL;
Jens Axboeedafcce2019-01-09 09:16:05 -07003717 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003718 return -EINVAL;
3719
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003720 req->sync.flags = READ_ONCE(sqe->fsync_flags);
3721 if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC))
3722 return -EINVAL;
3723
3724 req->sync.off = READ_ONCE(sqe->off);
3725 req->sync.len = READ_ONCE(sqe->len);
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003726 return 0;
3727}
3728
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003729static int io_fsync(struct io_kiocb *req, bool force_nonblock)
Jens Axboe78912932020-01-14 22:09:06 -07003730{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003731 loff_t end = req->sync.off + req->sync.len;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003732 int ret;
3733
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003734 /* fsync always requires a blocking context */
3735 if (force_nonblock)
3736 return -EAGAIN;
3737
Jens Axboe9adbd452019-12-20 08:45:55 -07003738 ret = vfs_fsync_range(req->file, req->sync.off,
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07003739 end > 0 ? end : LLONG_MAX,
3740 req->sync.flags & IORING_FSYNC_DATASYNC);
3741 if (ret < 0)
3742 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003743 io_req_complete(req, ret);
Christoph Hellwigc992fe22019-01-11 09:43:02 -07003744 return 0;
3745}
3746
Jens Axboed63d1b52019-12-10 10:38:56 -07003747static int io_fallocate_prep(struct io_kiocb *req,
3748 const struct io_uring_sqe *sqe)
3749{
3750 if (sqe->ioprio || sqe->buf_index || sqe->rw_flags)
3751 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03003752 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3753 return -EINVAL;
Jens Axboed63d1b52019-12-10 10:38:56 -07003754
3755 req->sync.off = READ_ONCE(sqe->off);
3756 req->sync.len = READ_ONCE(sqe->addr);
3757 req->sync.mode = READ_ONCE(sqe->len);
3758 return 0;
3759}
3760
Pavel Begunkov014db002020-03-03 21:33:12 +03003761static int io_fallocate(struct io_kiocb *req, bool force_nonblock)
Jens Axboed63d1b52019-12-10 10:38:56 -07003762{
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003763 int ret;
Jens Axboed63d1b52019-12-10 10:38:56 -07003764
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003765 /* fallocate always requiring blocking context */
3766 if (force_nonblock)
3767 return -EAGAIN;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003768 ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off,
3769 req->sync.len);
Pavel Begunkovac45abc2020-06-08 21:08:18 +03003770 if (ret < 0)
3771 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003772 io_req_complete(req, ret);
Jens Axboed63d1b52019-12-10 10:38:56 -07003773 return 0;
3774}
3775
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003776static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe15b71ab2019-12-11 11:20:36 -07003777{
Jens Axboef8748882020-01-08 17:47:02 -07003778 const char __user *fname;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003779 int ret;
3780
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003781 if (unlikely(sqe->ioprio || sqe->buf_index))
Jens Axboe15b71ab2019-12-11 11:20:36 -07003782 return -EINVAL;
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003783 if (unlikely(req->flags & REQ_F_FIXED_FILE))
Jens Axboecf3040c2020-02-06 21:31:40 -07003784 return -EBADF;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003785
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003786 /* open.how should be already initialised */
3787 if (!(req->open.how.flags & O_PATH) && force_o_largefile())
Jens Axboe08a1d26eb2020-04-08 09:20:54 -06003788 req->open.how.flags |= O_LARGEFILE;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003789
Pavel Begunkov25e72d12020-06-03 18:03:23 +03003790 req->open.dfd = READ_ONCE(sqe->fd);
3791 fname = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboef8748882020-01-08 17:47:02 -07003792 req->open.filename = getname(fname);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003793 if (IS_ERR(req->open.filename)) {
3794 ret = PTR_ERR(req->open.filename);
3795 req->open.filename = NULL;
3796 return ret;
3797 }
Jens Axboe4022e7a2020-03-19 19:23:18 -06003798 req->open.nofile = rlimit(RLIMIT_NOFILE);
Pavel Begunkov8fef80b2020-02-07 23:59:53 +03003799 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003800 return 0;
3801}
3802
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003803static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3804{
3805 u64 flags, mode;
3806
Jens Axboe4eb8dde2020-09-18 19:36:24 -06003807 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL)))
3808 return -EINVAL;
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003809 mode = READ_ONCE(sqe->len);
3810 flags = READ_ONCE(sqe->open_flags);
3811 req->open.how = build_open_how(flags, mode);
3812 return __io_openat_prep(req, sqe);
3813}
3814
Jens Axboecebdb982020-01-08 17:59:24 -07003815static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3816{
3817 struct open_how __user *how;
Jens Axboecebdb982020-01-08 17:59:24 -07003818 size_t len;
3819 int ret;
3820
Jens Axboe4eb8dde2020-09-18 19:36:24 -06003821 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL)))
3822 return -EINVAL;
Jens Axboecebdb982020-01-08 17:59:24 -07003823 how = u64_to_user_ptr(READ_ONCE(sqe->addr2));
3824 len = READ_ONCE(sqe->len);
Jens Axboecebdb982020-01-08 17:59:24 -07003825 if (len < OPEN_HOW_SIZE_VER0)
3826 return -EINVAL;
3827
3828 ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how,
3829 len);
3830 if (ret)
3831 return ret;
3832
Pavel Begunkovec65fea2020-06-03 18:03:24 +03003833 return __io_openat_prep(req, sqe);
Jens Axboecebdb982020-01-08 17:59:24 -07003834}
3835
Pavel Begunkov014db002020-03-03 21:33:12 +03003836static int io_openat2(struct io_kiocb *req, bool force_nonblock)
Jens Axboe15b71ab2019-12-11 11:20:36 -07003837{
3838 struct open_flags op;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003839 struct file *file;
3840 int ret;
3841
Jens Axboef86cd202020-01-29 13:46:44 -07003842 if (force_nonblock)
Jens Axboe15b71ab2019-12-11 11:20:36 -07003843 return -EAGAIN;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003844
Jens Axboecebdb982020-01-08 17:59:24 -07003845 ret = build_open_flags(&req->open.how, &op);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003846 if (ret)
3847 goto err;
3848
Jens Axboe4022e7a2020-03-19 19:23:18 -06003849 ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003850 if (ret < 0)
3851 goto err;
3852
3853 file = do_filp_open(req->open.dfd, req->open.filename, &op);
3854 if (IS_ERR(file)) {
3855 put_unused_fd(ret);
3856 ret = PTR_ERR(file);
3857 } else {
3858 fsnotify_open(file);
3859 fd_install(ret, file);
3860 }
3861err:
3862 putname(req->open.filename);
Pavel Begunkov8fef80b2020-02-07 23:59:53 +03003863 req->flags &= ~REQ_F_NEED_CLEANUP;
Jens Axboe15b71ab2019-12-11 11:20:36 -07003864 if (ret < 0)
3865 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003866 io_req_complete(req, ret);
Jens Axboe15b71ab2019-12-11 11:20:36 -07003867 return 0;
3868}
3869
Pavel Begunkov014db002020-03-03 21:33:12 +03003870static int io_openat(struct io_kiocb *req, bool force_nonblock)
Jens Axboecebdb982020-01-08 17:59:24 -07003871{
Pavel Begunkov014db002020-03-03 21:33:12 +03003872 return io_openat2(req, force_nonblock);
Jens Axboecebdb982020-01-08 17:59:24 -07003873}
3874
Jens Axboe067524e2020-03-02 16:32:28 -07003875static int io_remove_buffers_prep(struct io_kiocb *req,
3876 const struct io_uring_sqe *sqe)
3877{
3878 struct io_provide_buf *p = &req->pbuf;
3879 u64 tmp;
3880
3881 if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off)
3882 return -EINVAL;
3883
3884 tmp = READ_ONCE(sqe->fd);
3885 if (!tmp || tmp > USHRT_MAX)
3886 return -EINVAL;
3887
3888 memset(p, 0, sizeof(*p));
3889 p->nbufs = tmp;
3890 p->bgid = READ_ONCE(sqe->buf_group);
3891 return 0;
3892}
3893
3894static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf,
3895 int bgid, unsigned nbufs)
3896{
3897 unsigned i = 0;
3898
3899 /* shouldn't happen */
3900 if (!nbufs)
3901 return 0;
3902
3903 /* the head kbuf is the list itself */
3904 while (!list_empty(&buf->list)) {
3905 struct io_buffer *nxt;
3906
3907 nxt = list_first_entry(&buf->list, struct io_buffer, list);
3908 list_del(&nxt->list);
3909 kfree(nxt);
3910 if (++i == nbufs)
3911 return i;
3912 }
3913 i++;
3914 kfree(buf);
3915 idr_remove(&ctx->io_buffer_idr, bgid);
3916
3917 return i;
3918}
3919
Jens Axboe229a7b62020-06-22 10:13:11 -06003920static int io_remove_buffers(struct io_kiocb *req, bool force_nonblock,
3921 struct io_comp_state *cs)
Jens Axboe067524e2020-03-02 16:32:28 -07003922{
3923 struct io_provide_buf *p = &req->pbuf;
3924 struct io_ring_ctx *ctx = req->ctx;
3925 struct io_buffer *head;
3926 int ret = 0;
3927
3928 io_ring_submit_lock(ctx, !force_nonblock);
3929
3930 lockdep_assert_held(&ctx->uring_lock);
3931
3932 ret = -ENOENT;
3933 head = idr_find(&ctx->io_buffer_idr, p->bgid);
3934 if (head)
3935 ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs);
3936
3937 io_ring_submit_lock(ctx, !force_nonblock);
3938 if (ret < 0)
3939 req_set_fail_links(req);
Jens Axboe229a7b62020-06-22 10:13:11 -06003940 __io_req_complete(req, ret, 0, cs);
Jens Axboe067524e2020-03-02 16:32:28 -07003941 return 0;
3942}
3943
Jens Axboeddf0322d2020-02-23 16:41:33 -07003944static int io_provide_buffers_prep(struct io_kiocb *req,
3945 const struct io_uring_sqe *sqe)
3946{
3947 struct io_provide_buf *p = &req->pbuf;
3948 u64 tmp;
3949
3950 if (sqe->ioprio || sqe->rw_flags)
3951 return -EINVAL;
3952
3953 tmp = READ_ONCE(sqe->fd);
3954 if (!tmp || tmp > USHRT_MAX)
3955 return -E2BIG;
3956 p->nbufs = tmp;
3957 p->addr = READ_ONCE(sqe->addr);
3958 p->len = READ_ONCE(sqe->len);
3959
Bijan Mottahedehefe68c12020-06-04 18:01:52 -07003960 if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs)))
Jens Axboeddf0322d2020-02-23 16:41:33 -07003961 return -EFAULT;
3962
3963 p->bgid = READ_ONCE(sqe->buf_group);
3964 tmp = READ_ONCE(sqe->off);
3965 if (tmp > USHRT_MAX)
3966 return -E2BIG;
3967 p->bid = tmp;
3968 return 0;
3969}
3970
3971static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head)
3972{
3973 struct io_buffer *buf;
3974 u64 addr = pbuf->addr;
3975 int i, bid = pbuf->bid;
3976
3977 for (i = 0; i < pbuf->nbufs; i++) {
3978 buf = kmalloc(sizeof(*buf), GFP_KERNEL);
3979 if (!buf)
3980 break;
3981
3982 buf->addr = addr;
3983 buf->len = pbuf->len;
3984 buf->bid = bid;
3985 addr += pbuf->len;
3986 bid++;
3987 if (!*head) {
3988 INIT_LIST_HEAD(&buf->list);
3989 *head = buf;
3990 } else {
3991 list_add_tail(&buf->list, &(*head)->list);
3992 }
3993 }
3994
3995 return i ? i : -ENOMEM;
3996}
3997
Jens Axboe229a7b62020-06-22 10:13:11 -06003998static int io_provide_buffers(struct io_kiocb *req, bool force_nonblock,
3999 struct io_comp_state *cs)
Jens Axboeddf0322d2020-02-23 16:41:33 -07004000{
4001 struct io_provide_buf *p = &req->pbuf;
4002 struct io_ring_ctx *ctx = req->ctx;
4003 struct io_buffer *head, *list;
4004 int ret = 0;
4005
4006 io_ring_submit_lock(ctx, !force_nonblock);
4007
4008 lockdep_assert_held(&ctx->uring_lock);
4009
4010 list = head = idr_find(&ctx->io_buffer_idr, p->bgid);
4011
4012 ret = io_add_buffers(p, &head);
4013 if (ret < 0)
4014 goto out;
4015
4016 if (!list) {
4017 ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1,
4018 GFP_KERNEL);
4019 if (ret < 0) {
Jens Axboe067524e2020-03-02 16:32:28 -07004020 __io_remove_buffers(ctx, head, p->bgid, -1U);
Jens Axboeddf0322d2020-02-23 16:41:33 -07004021 goto out;
4022 }
4023 }
4024out:
4025 io_ring_submit_unlock(ctx, !force_nonblock);
4026 if (ret < 0)
4027 req_set_fail_links(req);
Jens Axboe229a7b62020-06-22 10:13:11 -06004028 __io_req_complete(req, ret, 0, cs);
Jens Axboeddf0322d2020-02-23 16:41:33 -07004029 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07004030}
4031
Jens Axboe3e4827b2020-01-08 15:18:09 -07004032static int io_epoll_ctl_prep(struct io_kiocb *req,
4033 const struct io_uring_sqe *sqe)
4034{
4035#if defined(CONFIG_EPOLL)
4036 if (sqe->ioprio || sqe->buf_index)
4037 return -EINVAL;
Jens Axboe6ca56f82020-09-18 16:51:19 -06004038 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL)))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004039 return -EINVAL;
Jens Axboe3e4827b2020-01-08 15:18:09 -07004040
4041 req->epoll.epfd = READ_ONCE(sqe->fd);
4042 req->epoll.op = READ_ONCE(sqe->len);
4043 req->epoll.fd = READ_ONCE(sqe->off);
4044
4045 if (ep_op_has_event(req->epoll.op)) {
4046 struct epoll_event __user *ev;
4047
4048 ev = u64_to_user_ptr(READ_ONCE(sqe->addr));
4049 if (copy_from_user(&req->epoll.event, ev, sizeof(*ev)))
4050 return -EFAULT;
4051 }
4052
4053 return 0;
4054#else
4055 return -EOPNOTSUPP;
4056#endif
4057}
4058
Jens Axboe229a7b62020-06-22 10:13:11 -06004059static int io_epoll_ctl(struct io_kiocb *req, bool force_nonblock,
4060 struct io_comp_state *cs)
Jens Axboe3e4827b2020-01-08 15:18:09 -07004061{
4062#if defined(CONFIG_EPOLL)
4063 struct io_epoll *ie = &req->epoll;
4064 int ret;
4065
4066 ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock);
4067 if (force_nonblock && ret == -EAGAIN)
4068 return -EAGAIN;
4069
4070 if (ret < 0)
4071 req_set_fail_links(req);
Jens Axboe229a7b62020-06-22 10:13:11 -06004072 __io_req_complete(req, ret, 0, cs);
Jens Axboe3e4827b2020-01-08 15:18:09 -07004073 return 0;
4074#else
4075 return -EOPNOTSUPP;
4076#endif
4077}
4078
Jens Axboec1ca7572019-12-25 22:18:28 -07004079static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4080{
4081#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
4082 if (sqe->ioprio || sqe->buf_index || sqe->off)
4083 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004084 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4085 return -EINVAL;
Jens Axboec1ca7572019-12-25 22:18:28 -07004086
4087 req->madvise.addr = READ_ONCE(sqe->addr);
4088 req->madvise.len = READ_ONCE(sqe->len);
4089 req->madvise.advice = READ_ONCE(sqe->fadvise_advice);
4090 return 0;
4091#else
4092 return -EOPNOTSUPP;
4093#endif
4094}
4095
Pavel Begunkov014db002020-03-03 21:33:12 +03004096static int io_madvise(struct io_kiocb *req, bool force_nonblock)
Jens Axboec1ca7572019-12-25 22:18:28 -07004097{
4098#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
4099 struct io_madvise *ma = &req->madvise;
4100 int ret;
4101
4102 if (force_nonblock)
4103 return -EAGAIN;
4104
Minchan Kim0726b012020-10-17 16:14:50 -07004105 ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice);
Jens Axboec1ca7572019-12-25 22:18:28 -07004106 if (ret < 0)
4107 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004108 io_req_complete(req, ret);
Jens Axboec1ca7572019-12-25 22:18:28 -07004109 return 0;
4110#else
4111 return -EOPNOTSUPP;
4112#endif
4113}
4114
Jens Axboe4840e412019-12-25 22:03:45 -07004115static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4116{
4117 if (sqe->ioprio || sqe->buf_index || sqe->addr)
4118 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004119 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4120 return -EINVAL;
Jens Axboe4840e412019-12-25 22:03:45 -07004121
4122 req->fadvise.offset = READ_ONCE(sqe->off);
4123 req->fadvise.len = READ_ONCE(sqe->len);
4124 req->fadvise.advice = READ_ONCE(sqe->fadvise_advice);
4125 return 0;
4126}
4127
Pavel Begunkov014db002020-03-03 21:33:12 +03004128static int io_fadvise(struct io_kiocb *req, bool force_nonblock)
Jens Axboe4840e412019-12-25 22:03:45 -07004129{
4130 struct io_fadvise *fa = &req->fadvise;
4131 int ret;
4132
Jens Axboe3e694262020-02-01 09:22:49 -07004133 if (force_nonblock) {
4134 switch (fa->advice) {
4135 case POSIX_FADV_NORMAL:
4136 case POSIX_FADV_RANDOM:
4137 case POSIX_FADV_SEQUENTIAL:
4138 break;
4139 default:
4140 return -EAGAIN;
4141 }
4142 }
Jens Axboe4840e412019-12-25 22:03:45 -07004143
4144 ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice);
4145 if (ret < 0)
4146 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004147 io_req_complete(req, ret);
Jens Axboe4840e412019-12-25 22:03:45 -07004148 return 0;
4149}
4150
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004151static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4152{
Jens Axboe6ca56f82020-09-18 16:51:19 -06004153 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL)))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004154 return -EINVAL;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004155 if (sqe->ioprio || sqe->buf_index)
4156 return -EINVAL;
Pavel Begunkov9c280f92020-04-08 08:58:46 +03004157 if (req->flags & REQ_F_FIXED_FILE)
Jens Axboecf3040c2020-02-06 21:31:40 -07004158 return -EBADF;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004159
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004160 req->statx.dfd = READ_ONCE(sqe->fd);
4161 req->statx.mask = READ_ONCE(sqe->len);
Bijan Mottahedehe62753e2020-05-22 21:31:18 -07004162 req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr));
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004163 req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2));
4164 req->statx.flags = READ_ONCE(sqe->statx_flags);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004165
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004166 return 0;
4167}
4168
Pavel Begunkov014db002020-03-03 21:33:12 +03004169static int io_statx(struct io_kiocb *req, bool force_nonblock)
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004170{
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004171 struct io_statx *ctx = &req->statx;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004172 int ret;
4173
Jens Axboe5b0bbee2020-04-27 10:41:22 -06004174 if (force_nonblock) {
4175 /* only need file table for an actual valid fd */
4176 if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD)
4177 req->flags |= REQ_F_NO_FILE_TABLE;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004178 return -EAGAIN;
Jens Axboe5b0bbee2020-04-27 10:41:22 -06004179 }
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004180
Bijan Mottahedehe62753e2020-05-22 21:31:18 -07004181 ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask,
4182 ctx->buffer);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004183
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004184 if (ret < 0)
4185 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004186 io_req_complete(req, ret);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004187 return 0;
4188}
4189
Jens Axboeb5dba592019-12-11 14:02:38 -07004190static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4191{
4192 /*
4193 * If we queue this for async, it must not be cancellable. That would
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08004194 * leave the 'file' in an undeterminate state, and here need to modify
4195 * io_wq_work.flags, so initialize io_wq_work firstly.
Jens Axboeb5dba592019-12-11 14:02:38 -07004196 */
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08004197 io_req_init_async(req);
Jens Axboeb5dba592019-12-11 14:02:38 -07004198 req->work.flags |= IO_WQ_WORK_NO_CANCEL;
4199
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004200 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL)))
4201 return -EINVAL;
Jens Axboeb5dba592019-12-11 14:02:38 -07004202 if (sqe->ioprio || sqe->off || sqe->addr || sqe->len ||
4203 sqe->rw_flags || sqe->buf_index)
4204 return -EINVAL;
Pavel Begunkov9c280f92020-04-08 08:58:46 +03004205 if (req->flags & REQ_F_FIXED_FILE)
Jens Axboecf3040c2020-02-06 21:31:40 -07004206 return -EBADF;
Jens Axboeb5dba592019-12-11 14:02:38 -07004207
4208 req->close.fd = READ_ONCE(sqe->fd);
Jens Axboe0f212202020-09-13 13:09:39 -06004209 if ((req->file && req->file->f_op == &io_uring_fops))
Jens Axboefd2206e2020-06-02 16:40:47 -06004210 return -EBADF;
4211
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004212 req->close.put_file = NULL;
Jens Axboeb5dba592019-12-11 14:02:38 -07004213 return 0;
4214}
4215
Jens Axboe229a7b62020-06-22 10:13:11 -06004216static int io_close(struct io_kiocb *req, bool force_nonblock,
4217 struct io_comp_state *cs)
Jens Axboeb5dba592019-12-11 14:02:38 -07004218{
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004219 struct io_close *close = &req->close;
Jens Axboeb5dba592019-12-11 14:02:38 -07004220 int ret;
4221
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004222 /* might be already done during nonblock submission */
4223 if (!close->put_file) {
4224 ret = __close_fd_get_file(close->fd, &close->put_file);
4225 if (ret < 0)
4226 return (ret == -ENOENT) ? -EBADF : ret;
4227 }
Jens Axboeb5dba592019-12-11 14:02:38 -07004228
4229 /* if the file has a flush method, be safe and punt to async */
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004230 if (close->put_file->f_op->flush && force_nonblock) {
Pavel Begunkov24c74672020-06-21 13:09:51 +03004231 /* was never set, but play safe */
4232 req->flags &= ~REQ_F_NOWAIT;
Pavel Begunkov0bf0eef2020-05-26 20:34:06 +03004233 /* avoid grabbing files - we don't need the files */
Pavel Begunkov24c74672020-06-21 13:09:51 +03004234 req->flags |= REQ_F_NO_FILE_TABLE;
Pavel Begunkov0bf0eef2020-05-26 20:34:06 +03004235 return -EAGAIN;
Pavel Begunkova2100672020-03-02 23:45:16 +03004236 }
Jens Axboeb5dba592019-12-11 14:02:38 -07004237
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004238 /* No ->flush() or already async, safely close from here */
Jens Axboe98447d62020-10-14 10:48:51 -06004239 ret = filp_close(close->put_file, req->work.identity->files);
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004240 if (ret < 0)
4241 req_set_fail_links(req);
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004242 fput(close->put_file);
4243 close->put_file = NULL;
Jens Axboe229a7b62020-06-22 10:13:11 -06004244 __io_req_complete(req, ret, 0, cs);
Jens Axboe1a417f42020-01-31 17:16:48 -07004245 return 0;
Jens Axboeb5dba592019-12-11 14:02:38 -07004246}
4247
Jens Axboe3529d8c2019-12-19 18:24:38 -07004248static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004249{
4250 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004251
4252 if (!req->file)
4253 return -EBADF;
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004254
4255 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
4256 return -EINVAL;
4257 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
4258 return -EINVAL;
4259
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004260 req->sync.off = READ_ONCE(sqe->off);
4261 req->sync.len = READ_ONCE(sqe->len);
4262 req->sync.flags = READ_ONCE(sqe->sync_range_flags);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004263 return 0;
4264}
4265
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004266static int io_sync_file_range(struct io_kiocb *req, bool force_nonblock)
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004267{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004268 int ret;
4269
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004270 /* sync_file_range always requires a blocking context */
4271 if (force_nonblock)
4272 return -EAGAIN;
4273
Jens Axboe9adbd452019-12-20 08:45:55 -07004274 ret = sync_file_range(req->file, req->sync.off, req->sync.len,
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004275 req->sync.flags);
4276 if (ret < 0)
4277 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004278 io_req_complete(req, ret);
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004279 return 0;
4280}
4281
YueHaibing469956e2020-03-04 15:53:52 +08004282#if defined(CONFIG_NET)
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004283static int io_setup_async_msg(struct io_kiocb *req,
4284 struct io_async_msghdr *kmsg)
4285{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004286 struct io_async_msghdr *async_msg = req->async_data;
4287
4288 if (async_msg)
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004289 return -EAGAIN;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004290 if (io_alloc_async_data(req)) {
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004291 if (kmsg->iov != kmsg->fast_iov)
4292 kfree(kmsg->iov);
4293 return -ENOMEM;
4294 }
Jens Axboee8c2bc12020-08-15 18:44:09 -07004295 async_msg = req->async_data;
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004296 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004297 memcpy(async_msg, kmsg, sizeof(*kmsg));
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004298 return -EAGAIN;
4299}
4300
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004301static int io_sendmsg_copy_hdr(struct io_kiocb *req,
4302 struct io_async_msghdr *iomsg)
4303{
4304 iomsg->iov = iomsg->fast_iov;
4305 iomsg->msg.msg_name = &iomsg->addr;
4306 return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg,
4307 req->sr_msg.msg_flags, &iomsg->iov);
4308}
4309
Jens Axboe3529d8c2019-12-19 18:24:38 -07004310static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboeaa1fa282019-04-19 13:38:09 -06004311{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004312 struct io_async_msghdr *async_msg = req->async_data;
Jens Axboee47293f2019-12-20 08:58:21 -07004313 struct io_sr_msg *sr = &req->sr_msg;
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004314 int ret;
Jens Axboe03b12302019-12-02 18:50:25 -07004315
Pavel Begunkovd2b6f482020-06-03 18:03:25 +03004316 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4317 return -EINVAL;
4318
Jens Axboee47293f2019-12-20 08:58:21 -07004319 sr->msg_flags = READ_ONCE(sqe->msg_flags);
Pavel Begunkov270a5942020-07-12 20:41:04 +03004320 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboefddafac2020-01-04 20:19:44 -07004321 sr->len = READ_ONCE(sqe->len);
Jens Axboe3529d8c2019-12-19 18:24:38 -07004322
Jens Axboed8768362020-02-27 14:17:49 -07004323#ifdef CONFIG_COMPAT
4324 if (req->ctx->compat)
4325 sr->msg_flags |= MSG_CMSG_COMPAT;
4326#endif
4327
Jens Axboee8c2bc12020-08-15 18:44:09 -07004328 if (!async_msg || !io_op_defs[req->opcode].needs_async_data)
Jens Axboe3529d8c2019-12-19 18:24:38 -07004329 return 0;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004330 ret = io_sendmsg_copy_hdr(req, async_msg);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004331 if (!ret)
4332 req->flags |= REQ_F_NEED_CLEANUP;
4333 return ret;
Jens Axboe03b12302019-12-02 18:50:25 -07004334}
4335
Jens Axboe229a7b62020-06-22 10:13:11 -06004336static int io_sendmsg(struct io_kiocb *req, bool force_nonblock,
4337 struct io_comp_state *cs)
Jens Axboe03b12302019-12-02 18:50:25 -07004338{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004339 struct io_async_msghdr iomsg, *kmsg;
Jens Axboe03b12302019-12-02 18:50:25 -07004340 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004341 unsigned flags;
Jens Axboe03b12302019-12-02 18:50:25 -07004342 int ret;
4343
Jens Axboe03b12302019-12-02 18:50:25 -07004344 sock = sock_from_file(req->file, &ret);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004345 if (unlikely(!sock))
4346 return ret;
Jens Axboe03b12302019-12-02 18:50:25 -07004347
Jens Axboee8c2bc12020-08-15 18:44:09 -07004348 if (req->async_data) {
4349 kmsg = req->async_data;
4350 kmsg->msg.msg_name = &kmsg->addr;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004351 /* if iov is set, it's allocated already */
4352 if (!kmsg->iov)
4353 kmsg->iov = kmsg->fast_iov;
4354 kmsg->msg.msg_iter.iov = kmsg->iov;
4355 } else {
4356 ret = io_sendmsg_copy_hdr(req, &iomsg);
Jens Axboefddafac2020-01-04 20:19:44 -07004357 if (ret)
4358 return ret;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004359 kmsg = &iomsg;
Jens Axboefddafac2020-01-04 20:19:44 -07004360 }
4361
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004362 flags = req->sr_msg.msg_flags;
4363 if (flags & MSG_DONTWAIT)
4364 req->flags |= REQ_F_NOWAIT;
4365 else if (force_nonblock)
4366 flags |= MSG_DONTWAIT;
4367
4368 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
4369 if (force_nonblock && ret == -EAGAIN)
4370 return io_setup_async_msg(req, kmsg);
4371 if (ret == -ERESTARTSYS)
4372 ret = -EINTR;
4373
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004374 if (kmsg->iov != kmsg->fast_iov)
Jens Axboe03b12302019-12-02 18:50:25 -07004375 kfree(kmsg->iov);
4376 req->flags &= ~REQ_F_NEED_CLEANUP;
Jens Axboefddafac2020-01-04 20:19:44 -07004377 if (ret < 0)
4378 req_set_fail_links(req);
Jens Axboe229a7b62020-06-22 10:13:11 -06004379 __io_req_complete(req, ret, 0, cs);
Jens Axboefddafac2020-01-04 20:19:44 -07004380 return 0;
Jens Axboefddafac2020-01-04 20:19:44 -07004381}
4382
Jens Axboe229a7b62020-06-22 10:13:11 -06004383static int io_send(struct io_kiocb *req, bool force_nonblock,
4384 struct io_comp_state *cs)
Jens Axboe03b12302019-12-02 18:50:25 -07004385{
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004386 struct io_sr_msg *sr = &req->sr_msg;
4387 struct msghdr msg;
4388 struct iovec iov;
Jens Axboe03b12302019-12-02 18:50:25 -07004389 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004390 unsigned flags;
Jens Axboe03b12302019-12-02 18:50:25 -07004391 int ret;
4392
4393 sock = sock_from_file(req->file, &ret);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004394 if (unlikely(!sock))
4395 return ret;
Jens Axboe03b12302019-12-02 18:50:25 -07004396
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004397 ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter);
4398 if (unlikely(ret))
Zheng Bin14db8412020-09-09 20:12:37 +08004399 return ret;
Jens Axboe03b12302019-12-02 18:50:25 -07004400
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004401 msg.msg_name = NULL;
4402 msg.msg_control = NULL;
4403 msg.msg_controllen = 0;
4404 msg.msg_namelen = 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004405
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004406 flags = req->sr_msg.msg_flags;
4407 if (flags & MSG_DONTWAIT)
4408 req->flags |= REQ_F_NOWAIT;
4409 else if (force_nonblock)
4410 flags |= MSG_DONTWAIT;
Jens Axboe03b12302019-12-02 18:50:25 -07004411
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004412 msg.msg_flags = flags;
4413 ret = sock_sendmsg(sock, &msg);
4414 if (force_nonblock && ret == -EAGAIN)
4415 return -EAGAIN;
4416 if (ret == -ERESTARTSYS)
4417 ret = -EINTR;
Jens Axboe03b12302019-12-02 18:50:25 -07004418
Jens Axboe03b12302019-12-02 18:50:25 -07004419 if (ret < 0)
4420 req_set_fail_links(req);
Jens Axboe229a7b62020-06-22 10:13:11 -06004421 __io_req_complete(req, ret, 0, cs);
Jens Axboe03b12302019-12-02 18:50:25 -07004422 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004423}
4424
Pavel Begunkov1400e692020-07-12 20:41:05 +03004425static int __io_recvmsg_copy_hdr(struct io_kiocb *req,
4426 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004427{
4428 struct io_sr_msg *sr = &req->sr_msg;
4429 struct iovec __user *uiov;
4430 size_t iov_len;
4431 int ret;
4432
Pavel Begunkov1400e692020-07-12 20:41:05 +03004433 ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg,
4434 &iomsg->uaddr, &uiov, &iov_len);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004435 if (ret)
4436 return ret;
4437
4438 if (req->flags & REQ_F_BUFFER_SELECT) {
4439 if (iov_len > 1)
4440 return -EINVAL;
Pavel Begunkov1400e692020-07-12 20:41:05 +03004441 if (copy_from_user(iomsg->iov, uiov, sizeof(*uiov)))
Jens Axboe52de1fe2020-02-27 10:15:42 -07004442 return -EFAULT;
Pavel Begunkov1400e692020-07-12 20:41:05 +03004443 sr->len = iomsg->iov[0].iov_len;
4444 iov_iter_init(&iomsg->msg.msg_iter, READ, iomsg->iov, 1,
Jens Axboe52de1fe2020-02-27 10:15:42 -07004445 sr->len);
Pavel Begunkov1400e692020-07-12 20:41:05 +03004446 iomsg->iov = NULL;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004447 } else {
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004448 ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV,
4449 &iomsg->iov, &iomsg->msg.msg_iter,
4450 false);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004451 if (ret > 0)
4452 ret = 0;
4453 }
4454
4455 return ret;
4456}
4457
4458#ifdef CONFIG_COMPAT
4459static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req,
Pavel Begunkov1400e692020-07-12 20:41:05 +03004460 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004461{
4462 struct compat_msghdr __user *msg_compat;
4463 struct io_sr_msg *sr = &req->sr_msg;
4464 struct compat_iovec __user *uiov;
4465 compat_uptr_t ptr;
4466 compat_size_t len;
4467 int ret;
4468
Pavel Begunkov270a5942020-07-12 20:41:04 +03004469 msg_compat = (struct compat_msghdr __user *) sr->umsg;
Pavel Begunkov1400e692020-07-12 20:41:05 +03004470 ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr,
Jens Axboe52de1fe2020-02-27 10:15:42 -07004471 &ptr, &len);
4472 if (ret)
4473 return ret;
4474
4475 uiov = compat_ptr(ptr);
4476 if (req->flags & REQ_F_BUFFER_SELECT) {
4477 compat_ssize_t clen;
4478
4479 if (len > 1)
4480 return -EINVAL;
4481 if (!access_ok(uiov, sizeof(*uiov)))
4482 return -EFAULT;
4483 if (__get_user(clen, &uiov->iov_len))
4484 return -EFAULT;
4485 if (clen < 0)
4486 return -EINVAL;
Pavel Begunkov1400e692020-07-12 20:41:05 +03004487 sr->len = iomsg->iov[0].iov_len;
4488 iomsg->iov = NULL;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004489 } else {
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004490 ret = __import_iovec(READ, (struct iovec __user *)uiov, len,
4491 UIO_FASTIOV, &iomsg->iov,
4492 &iomsg->msg.msg_iter, true);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004493 if (ret < 0)
4494 return ret;
4495 }
4496
4497 return 0;
4498}
Jens Axboe03b12302019-12-02 18:50:25 -07004499#endif
Jens Axboe52de1fe2020-02-27 10:15:42 -07004500
Pavel Begunkov1400e692020-07-12 20:41:05 +03004501static int io_recvmsg_copy_hdr(struct io_kiocb *req,
4502 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004503{
Pavel Begunkov1400e692020-07-12 20:41:05 +03004504 iomsg->msg.msg_name = &iomsg->addr;
4505 iomsg->iov = iomsg->fast_iov;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004506
4507#ifdef CONFIG_COMPAT
4508 if (req->ctx->compat)
Pavel Begunkov1400e692020-07-12 20:41:05 +03004509 return __io_compat_recvmsg_copy_hdr(req, iomsg);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004510#endif
4511
Pavel Begunkov1400e692020-07-12 20:41:05 +03004512 return __io_recvmsg_copy_hdr(req, iomsg);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004513}
4514
Jens Axboebcda7ba2020-02-23 16:42:51 -07004515static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req,
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004516 bool needs_lock)
Jens Axboebcda7ba2020-02-23 16:42:51 -07004517{
4518 struct io_sr_msg *sr = &req->sr_msg;
4519 struct io_buffer *kbuf;
4520
Jens Axboebcda7ba2020-02-23 16:42:51 -07004521 kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock);
4522 if (IS_ERR(kbuf))
4523 return kbuf;
4524
4525 sr->kbuf = kbuf;
4526 req->flags |= REQ_F_BUFFER_SELECTED;
Jens Axboebcda7ba2020-02-23 16:42:51 -07004527 return kbuf;
Jens Axboe03b12302019-12-02 18:50:25 -07004528}
4529
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004530static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req)
4531{
4532 return io_put_kbuf(req, req->sr_msg.kbuf);
4533}
4534
Jens Axboe3529d8c2019-12-19 18:24:38 -07004535static int io_recvmsg_prep(struct io_kiocb *req,
4536 const struct io_uring_sqe *sqe)
Jens Axboe03b12302019-12-02 18:50:25 -07004537{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004538 struct io_async_msghdr *async_msg = req->async_data;
Jens Axboee47293f2019-12-20 08:58:21 -07004539 struct io_sr_msg *sr = &req->sr_msg;
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004540 int ret;
Jens Axboe06b76d42019-12-19 14:44:26 -07004541
Pavel Begunkovd2b6f482020-06-03 18:03:25 +03004542 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4543 return -EINVAL;
4544
Jens Axboe3529d8c2019-12-19 18:24:38 -07004545 sr->msg_flags = READ_ONCE(sqe->msg_flags);
Pavel Begunkov270a5942020-07-12 20:41:04 +03004546 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboe0b7b21e2020-01-31 08:34:59 -07004547 sr->len = READ_ONCE(sqe->len);
Jens Axboebcda7ba2020-02-23 16:42:51 -07004548 sr->bgid = READ_ONCE(sqe->buf_group);
Jens Axboe3529d8c2019-12-19 18:24:38 -07004549
Jens Axboed8768362020-02-27 14:17:49 -07004550#ifdef CONFIG_COMPAT
4551 if (req->ctx->compat)
4552 sr->msg_flags |= MSG_CMSG_COMPAT;
4553#endif
4554
Jens Axboee8c2bc12020-08-15 18:44:09 -07004555 if (!async_msg || !io_op_defs[req->opcode].needs_async_data)
Jens Axboe06b76d42019-12-19 14:44:26 -07004556 return 0;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004557 ret = io_recvmsg_copy_hdr(req, async_msg);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004558 if (!ret)
4559 req->flags |= REQ_F_NEED_CLEANUP;
4560 return ret;
Jens Axboe03b12302019-12-02 18:50:25 -07004561}
4562
Jens Axboe229a7b62020-06-22 10:13:11 -06004563static int io_recvmsg(struct io_kiocb *req, bool force_nonblock,
4564 struct io_comp_state *cs)
Jens Axboe03b12302019-12-02 18:50:25 -07004565{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004566 struct io_async_msghdr iomsg, *kmsg;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004567 struct socket *sock;
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004568 struct io_buffer *kbuf;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004569 unsigned flags;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004570 int ret, cflags = 0;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004571
Jens Axboe0fa03c62019-04-19 13:34:07 -06004572 sock = sock_from_file(req->file, &ret);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004573 if (unlikely(!sock))
4574 return ret;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004575
Jens Axboee8c2bc12020-08-15 18:44:09 -07004576 if (req->async_data) {
4577 kmsg = req->async_data;
4578 kmsg->msg.msg_name = &kmsg->addr;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004579 /* if iov is set, it's allocated already */
4580 if (!kmsg->iov)
4581 kmsg->iov = kmsg->fast_iov;
4582 kmsg->msg.msg_iter.iov = kmsg->iov;
4583 } else {
4584 ret = io_recvmsg_copy_hdr(req, &iomsg);
4585 if (ret)
Pavel Begunkov681fda82020-07-15 22:20:45 +03004586 return ret;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004587 kmsg = &iomsg;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004588 }
4589
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004590 if (req->flags & REQ_F_BUFFER_SELECT) {
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004591 kbuf = io_recv_buffer_select(req, !force_nonblock);
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004592 if (IS_ERR(kbuf))
4593 return PTR_ERR(kbuf);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004594 kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr);
4595 iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->iov,
4596 1, req->sr_msg.len);
4597 }
4598
4599 flags = req->sr_msg.msg_flags;
4600 if (flags & MSG_DONTWAIT)
4601 req->flags |= REQ_F_NOWAIT;
4602 else if (force_nonblock)
4603 flags |= MSG_DONTWAIT;
4604
4605 ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg,
4606 kmsg->uaddr, flags);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03004607 if (force_nonblock && ret == -EAGAIN)
4608 return io_setup_async_msg(req, kmsg);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004609 if (ret == -ERESTARTSYS)
4610 ret = -EINTR;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03004611
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004612 if (req->flags & REQ_F_BUFFER_SELECTED)
4613 cflags = io_put_recv_kbuf(req);
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004614 if (kmsg->iov != kmsg->fast_iov)
Jens Axboe0b416c3e2019-12-15 10:57:46 -07004615 kfree(kmsg->iov);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004616 req->flags &= ~REQ_F_NEED_CLEANUP;
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004617 if (ret < 0)
4618 req_set_fail_links(req);
Jens Axboe229a7b62020-06-22 10:13:11 -06004619 __io_req_complete(req, ret, cflags, cs);
Jens Axboe0fa03c62019-04-19 13:34:07 -06004620 return 0;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004621}
4622
Jens Axboe229a7b62020-06-22 10:13:11 -06004623static int io_recv(struct io_kiocb *req, bool force_nonblock,
4624 struct io_comp_state *cs)
Jens Axboefddafac2020-01-04 20:19:44 -07004625{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004626 struct io_buffer *kbuf;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004627 struct io_sr_msg *sr = &req->sr_msg;
4628 struct msghdr msg;
4629 void __user *buf = sr->buf;
Jens Axboefddafac2020-01-04 20:19:44 -07004630 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004631 struct iovec iov;
4632 unsigned flags;
Jens Axboebcda7ba2020-02-23 16:42:51 -07004633 int ret, cflags = 0;
Jens Axboefddafac2020-01-04 20:19:44 -07004634
Jens Axboefddafac2020-01-04 20:19:44 -07004635 sock = sock_from_file(req->file, &ret);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004636 if (unlikely(!sock))
4637 return ret;
Jens Axboefddafac2020-01-04 20:19:44 -07004638
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004639 if (req->flags & REQ_F_BUFFER_SELECT) {
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004640 kbuf = io_recv_buffer_select(req, !force_nonblock);
Jens Axboebcda7ba2020-02-23 16:42:51 -07004641 if (IS_ERR(kbuf))
4642 return PTR_ERR(kbuf);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004643 buf = u64_to_user_ptr(kbuf->addr);
Jens Axboefddafac2020-01-04 20:19:44 -07004644 }
4645
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004646 ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter);
Pavel Begunkov14c32ee2020-07-16 23:28:01 +03004647 if (unlikely(ret))
4648 goto out_free;
Jens Axboefddafac2020-01-04 20:19:44 -07004649
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004650 msg.msg_name = NULL;
4651 msg.msg_control = NULL;
4652 msg.msg_controllen = 0;
4653 msg.msg_namelen = 0;
4654 msg.msg_iocb = NULL;
4655 msg.msg_flags = 0;
4656
4657 flags = req->sr_msg.msg_flags;
4658 if (flags & MSG_DONTWAIT)
4659 req->flags |= REQ_F_NOWAIT;
4660 else if (force_nonblock)
4661 flags |= MSG_DONTWAIT;
4662
4663 ret = sock_recvmsg(sock, &msg, flags);
4664 if (force_nonblock && ret == -EAGAIN)
4665 return -EAGAIN;
4666 if (ret == -ERESTARTSYS)
4667 ret = -EINTR;
Pavel Begunkov14c32ee2020-07-16 23:28:01 +03004668out_free:
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004669 if (req->flags & REQ_F_BUFFER_SELECTED)
4670 cflags = io_put_recv_kbuf(req);
Jens Axboefddafac2020-01-04 20:19:44 -07004671 if (ret < 0)
4672 req_set_fail_links(req);
Jens Axboe229a7b62020-06-22 10:13:11 -06004673 __io_req_complete(req, ret, cflags, cs);
Jens Axboefddafac2020-01-04 20:19:44 -07004674 return 0;
Jens Axboefddafac2020-01-04 20:19:44 -07004675}
4676
Jens Axboe3529d8c2019-12-19 18:24:38 -07004677static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004678{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004679 struct io_accept *accept = &req->accept;
4680
Jens Axboe17f2fe32019-10-17 14:42:58 -06004681 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL)))
4682 return -EINVAL;
Hrvoje Zeba8042d6c2019-11-25 14:40:22 -05004683 if (sqe->ioprio || sqe->len || sqe->buf_index)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004684 return -EINVAL;
4685
Jens Axboed55e5f52019-12-11 16:12:15 -07004686 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
4687 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004688 accept->flags = READ_ONCE(sqe->accept_flags);
Jens Axboe09952e32020-03-19 20:16:56 -06004689 accept->nofile = rlimit(RLIMIT_NOFILE);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004690 return 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004691}
Jens Axboe17f2fe32019-10-17 14:42:58 -06004692
Jens Axboe229a7b62020-06-22 10:13:11 -06004693static int io_accept(struct io_kiocb *req, bool force_nonblock,
4694 struct io_comp_state *cs)
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004695{
4696 struct io_accept *accept = &req->accept;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004697 unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004698 int ret;
4699
Jiufei Xuee697dee2020-06-10 13:41:59 +08004700 if (req->file->f_flags & O_NONBLOCK)
4701 req->flags |= REQ_F_NOWAIT;
4702
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004703 ret = __sys_accept4_file(req->file, file_flags, accept->addr,
Jens Axboe09952e32020-03-19 20:16:56 -06004704 accept->addr_len, accept->flags,
4705 accept->nofile);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004706 if (ret == -EAGAIN && force_nonblock)
Jens Axboe17f2fe32019-10-17 14:42:58 -06004707 return -EAGAIN;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004708 if (ret < 0) {
4709 if (ret == -ERESTARTSYS)
4710 ret = -EINTR;
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004711 req_set_fail_links(req);
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004712 }
Jens Axboe229a7b62020-06-22 10:13:11 -06004713 __io_req_complete(req, ret, 0, cs);
Jens Axboe17f2fe32019-10-17 14:42:58 -06004714 return 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004715}
4716
Jens Axboe3529d8c2019-12-19 18:24:38 -07004717static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef499a022019-12-02 16:28:46 -07004718{
Jens Axboe3529d8c2019-12-19 18:24:38 -07004719 struct io_connect *conn = &req->connect;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004720 struct io_async_connect *io = req->async_data;
Jens Axboef499a022019-12-02 16:28:46 -07004721
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004722 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL|IORING_SETUP_SQPOLL)))
4723 return -EINVAL;
4724 if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags)
4725 return -EINVAL;
4726
Jens Axboe3529d8c2019-12-19 18:24:38 -07004727 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
4728 conn->addr_len = READ_ONCE(sqe->addr2);
4729
4730 if (!io)
4731 return 0;
4732
4733 return move_addr_to_kernel(conn->addr, conn->addr_len,
Jens Axboee8c2bc12020-08-15 18:44:09 -07004734 &io->address);
Jens Axboef499a022019-12-02 16:28:46 -07004735}
4736
Jens Axboe229a7b62020-06-22 10:13:11 -06004737static int io_connect(struct io_kiocb *req, bool force_nonblock,
4738 struct io_comp_state *cs)
Jens Axboef8e85cf2019-11-23 14:24:24 -07004739{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004740 struct io_async_connect __io, *io;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004741 unsigned file_flags;
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004742 int ret;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004743
Jens Axboee8c2bc12020-08-15 18:44:09 -07004744 if (req->async_data) {
4745 io = req->async_data;
Jens Axboef499a022019-12-02 16:28:46 -07004746 } else {
Jens Axboe3529d8c2019-12-19 18:24:38 -07004747 ret = move_addr_to_kernel(req->connect.addr,
4748 req->connect.addr_len,
Jens Axboee8c2bc12020-08-15 18:44:09 -07004749 &__io.address);
Jens Axboef499a022019-12-02 16:28:46 -07004750 if (ret)
4751 goto out;
4752 io = &__io;
4753 }
4754
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004755 file_flags = force_nonblock ? O_NONBLOCK : 0;
4756
Jens Axboee8c2bc12020-08-15 18:44:09 -07004757 ret = __sys_connect_file(req->file, &io->address,
Jens Axboe3fbb51c2019-12-20 08:51:52 -07004758 req->connect.addr_len, file_flags);
Jens Axboe87f80d62019-12-03 11:23:54 -07004759 if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07004760 if (req->async_data)
Jens Axboeb7bb4f72019-12-15 22:13:43 -07004761 return -EAGAIN;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004762 if (io_alloc_async_data(req)) {
Jens Axboef499a022019-12-02 16:28:46 -07004763 ret = -ENOMEM;
4764 goto out;
4765 }
Jens Axboee8c2bc12020-08-15 18:44:09 -07004766 io = req->async_data;
4767 memcpy(req->async_data, &__io, sizeof(__io));
Jens Axboef8e85cf2019-11-23 14:24:24 -07004768 return -EAGAIN;
Jens Axboef499a022019-12-02 16:28:46 -07004769 }
Jens Axboef8e85cf2019-11-23 14:24:24 -07004770 if (ret == -ERESTARTSYS)
4771 ret = -EINTR;
Jens Axboef499a022019-12-02 16:28:46 -07004772out:
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004773 if (ret < 0)
4774 req_set_fail_links(req);
Jens Axboe229a7b62020-06-22 10:13:11 -06004775 __io_req_complete(req, ret, 0, cs);
Jens Axboef8e85cf2019-11-23 14:24:24 -07004776 return 0;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004777}
YueHaibing469956e2020-03-04 15:53:52 +08004778#else /* !CONFIG_NET */
4779static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4780{
Jens Axboef8e85cf2019-11-23 14:24:24 -07004781 return -EOPNOTSUPP;
Jens Axboef8e85cf2019-11-23 14:24:24 -07004782}
4783
Randy Dunlap1e16c2f2020-06-26 16:32:50 -07004784static int io_sendmsg(struct io_kiocb *req, bool force_nonblock,
4785 struct io_comp_state *cs)
Jens Axboe221c5eb2019-01-17 09:41:58 -07004786{
YueHaibing469956e2020-03-04 15:53:52 +08004787 return -EOPNOTSUPP;
4788}
4789
Randy Dunlap1e16c2f2020-06-26 16:32:50 -07004790static int io_send(struct io_kiocb *req, bool force_nonblock,
4791 struct io_comp_state *cs)
YueHaibing469956e2020-03-04 15:53:52 +08004792{
4793 return -EOPNOTSUPP;
4794}
4795
4796static int io_recvmsg_prep(struct io_kiocb *req,
4797 const struct io_uring_sqe *sqe)
4798{
4799 return -EOPNOTSUPP;
4800}
4801
Randy Dunlap1e16c2f2020-06-26 16:32:50 -07004802static int io_recvmsg(struct io_kiocb *req, bool force_nonblock,
4803 struct io_comp_state *cs)
YueHaibing469956e2020-03-04 15:53:52 +08004804{
4805 return -EOPNOTSUPP;
4806}
4807
Randy Dunlap1e16c2f2020-06-26 16:32:50 -07004808static int io_recv(struct io_kiocb *req, bool force_nonblock,
4809 struct io_comp_state *cs)
YueHaibing469956e2020-03-04 15:53:52 +08004810{
4811 return -EOPNOTSUPP;
4812}
4813
4814static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4815{
4816 return -EOPNOTSUPP;
4817}
4818
Randy Dunlap1e16c2f2020-06-26 16:32:50 -07004819static int io_accept(struct io_kiocb *req, bool force_nonblock,
4820 struct io_comp_state *cs)
YueHaibing469956e2020-03-04 15:53:52 +08004821{
4822 return -EOPNOTSUPP;
4823}
4824
4825static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4826{
4827 return -EOPNOTSUPP;
4828}
4829
Randy Dunlap1e16c2f2020-06-26 16:32:50 -07004830static int io_connect(struct io_kiocb *req, bool force_nonblock,
4831 struct io_comp_state *cs)
YueHaibing469956e2020-03-04 15:53:52 +08004832{
4833 return -EOPNOTSUPP;
4834}
4835#endif /* CONFIG_NET */
Jens Axboe2b188cc2019-01-07 10:46:33 -07004836
Jens Axboed7718a92020-02-14 22:23:12 -07004837struct io_poll_table {
4838 struct poll_table_struct pt;
4839 struct io_kiocb *req;
4840 int error;
4841};
4842
Jens Axboed7718a92020-02-14 22:23:12 -07004843static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll,
4844 __poll_t mask, task_work_func_t func)
4845{
Jens Axboefd7d6de2020-08-23 11:00:37 -06004846 bool twa_signal_ok;
Jens Axboeaa96bf82020-04-03 11:26:26 -06004847 int ret;
Jens Axboed7718a92020-02-14 22:23:12 -07004848
4849 /* for instances that support it check for an event match first: */
4850 if (mask && !(mask & poll->events))
4851 return 0;
4852
4853 trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask);
4854
4855 list_del_init(&poll->wait.entry);
4856
Jens Axboed7718a92020-02-14 22:23:12 -07004857 req->result = mask;
4858 init_task_work(&req->task_work, func);
Jens Axboe6d816e02020-08-11 08:04:14 -06004859 percpu_ref_get(&req->ctx->refs);
4860
Jens Axboed7718a92020-02-14 22:23:12 -07004861 /*
Jens Axboefd7d6de2020-08-23 11:00:37 -06004862 * If we using the signalfd wait_queue_head for this wakeup, then
4863 * it's not safe to use TWA_SIGNAL as we could be recursing on the
4864 * tsk->sighand->siglock on doing the wakeup. Should not be needed
4865 * either, as the normal wakeup will suffice.
4866 */
4867 twa_signal_ok = (poll->head != &req->task->sighand->signalfd_wqh);
4868
4869 /*
Jens Axboee3aabf92020-05-18 11:04:17 -06004870 * If this fails, then the task is exiting. When a task exits, the
4871 * work gets canceled, so just cancel this request as well instead
4872 * of executing it. We can't safely execute it anyway, as we may not
4873 * have the needed state needed for it anyway.
Jens Axboed7718a92020-02-14 22:23:12 -07004874 */
Jens Axboe87c43112020-09-30 21:00:14 -06004875 ret = io_req_task_work_add(req, twa_signal_ok);
Jens Axboeaa96bf82020-04-03 11:26:26 -06004876 if (unlikely(ret)) {
Jens Axboec2c4c832020-07-01 15:37:11 -06004877 struct task_struct *tsk;
4878
Jens Axboee3aabf92020-05-18 11:04:17 -06004879 WRITE_ONCE(poll->canceled, true);
Jens Axboeaa96bf82020-04-03 11:26:26 -06004880 tsk = io_wq_get_task(req->ctx->io_wq);
Jens Axboe91989c72020-10-16 09:02:26 -06004881 task_work_add(tsk, &req->task_work, TWA_NONE);
Jens Axboece593a62020-06-30 12:39:05 -06004882 wake_up_process(tsk);
Jens Axboeaa96bf82020-04-03 11:26:26 -06004883 }
Jens Axboed7718a92020-02-14 22:23:12 -07004884 return 1;
4885}
4886
Jens Axboe74ce6ce2020-04-13 11:09:12 -06004887static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll)
4888 __acquires(&req->ctx->completion_lock)
4889{
4890 struct io_ring_ctx *ctx = req->ctx;
4891
4892 if (!req->result && !READ_ONCE(poll->canceled)) {
4893 struct poll_table_struct pt = { ._key = poll->events };
4894
4895 req->result = vfs_poll(req->file, &pt) & poll->events;
4896 }
4897
4898 spin_lock_irq(&ctx->completion_lock);
4899 if (!req->result && !READ_ONCE(poll->canceled)) {
4900 add_wait_queue(poll->head, &poll->wait);
4901 return true;
4902 }
4903
4904 return false;
4905}
4906
Jens Axboed4e7cd32020-08-15 11:44:50 -07004907static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req)
Jens Axboe18bceab2020-05-15 11:56:54 -06004908{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004909 /* pure poll stashes this in ->async_data, poll driven retry elsewhere */
Jens Axboed4e7cd32020-08-15 11:44:50 -07004910 if (req->opcode == IORING_OP_POLL_ADD)
Jens Axboee8c2bc12020-08-15 18:44:09 -07004911 return req->async_data;
Jens Axboed4e7cd32020-08-15 11:44:50 -07004912 return req->apoll->double_poll;
4913}
4914
4915static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req)
4916{
4917 if (req->opcode == IORING_OP_POLL_ADD)
4918 return &req->poll;
4919 return &req->apoll->poll;
4920}
4921
4922static void io_poll_remove_double(struct io_kiocb *req)
4923{
4924 struct io_poll_iocb *poll = io_poll_get_double(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004925
4926 lockdep_assert_held(&req->ctx->completion_lock);
4927
4928 if (poll && poll->head) {
4929 struct wait_queue_head *head = poll->head;
4930
4931 spin_lock(&head->lock);
4932 list_del_init(&poll->wait.entry);
4933 if (poll->wait.private)
4934 refcount_dec(&req->refs);
4935 poll->head = NULL;
4936 spin_unlock(&head->lock);
4937 }
4938}
4939
4940static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error)
4941{
4942 struct io_ring_ctx *ctx = req->ctx;
4943
Jens Axboed4e7cd32020-08-15 11:44:50 -07004944 io_poll_remove_double(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004945 req->poll.done = true;
4946 io_cqring_fill_event(req, error ? error : mangle_poll(mask));
4947 io_commit_cqring(ctx);
4948}
4949
Jens Axboe18bceab2020-05-15 11:56:54 -06004950static void io_poll_task_func(struct callback_head *cb)
4951{
4952 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
Jens Axboe6d816e02020-08-11 08:04:14 -06004953 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004954 struct io_kiocb *nxt;
Jens Axboe18bceab2020-05-15 11:56:54 -06004955
Pavel Begunkovdd221f462020-10-18 10:17:42 +01004956 if (io_poll_rewait(req, &req->poll)) {
4957 spin_unlock_irq(&ctx->completion_lock);
4958 } else {
4959 hash_del(&req->hash_node);
4960 io_poll_complete(req, req->result, 0);
4961 spin_unlock_irq(&ctx->completion_lock);
4962
4963 nxt = io_put_req_find_next(req);
4964 io_cqring_ev_posted(ctx);
4965 if (nxt)
4966 __io_req_task_submit(nxt);
4967 }
4968
Jens Axboe6d816e02020-08-11 08:04:14 -06004969 percpu_ref_put(&ctx->refs);
Jens Axboe18bceab2020-05-15 11:56:54 -06004970}
4971
4972static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode,
4973 int sync, void *key)
4974{
4975 struct io_kiocb *req = wait->private;
Jens Axboed4e7cd32020-08-15 11:44:50 -07004976 struct io_poll_iocb *poll = io_poll_get_single(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06004977 __poll_t mask = key_to_poll(key);
4978
4979 /* for instances that support it check for an event match first: */
4980 if (mask && !(mask & poll->events))
4981 return 0;
4982
Jens Axboe8706e042020-09-28 08:38:54 -06004983 list_del_init(&wait->entry);
4984
Jens Axboe807abcb2020-07-17 17:09:27 -06004985 if (poll && poll->head) {
Jens Axboe18bceab2020-05-15 11:56:54 -06004986 bool done;
4987
Jens Axboe807abcb2020-07-17 17:09:27 -06004988 spin_lock(&poll->head->lock);
4989 done = list_empty(&poll->wait.entry);
Jens Axboe18bceab2020-05-15 11:56:54 -06004990 if (!done)
Jens Axboe807abcb2020-07-17 17:09:27 -06004991 list_del_init(&poll->wait.entry);
Jens Axboed4e7cd32020-08-15 11:44:50 -07004992 /* make sure double remove sees this as being gone */
4993 wait->private = NULL;
Jens Axboe807abcb2020-07-17 17:09:27 -06004994 spin_unlock(&poll->head->lock);
Jens Axboec8b5e262020-10-25 13:53:26 -06004995 if (!done) {
4996 /* use wait func handler, so it matches the rq type */
4997 poll->wait.func(&poll->wait, mode, sync, key);
4998 }
Jens Axboe18bceab2020-05-15 11:56:54 -06004999 }
5000 refcount_dec(&req->refs);
5001 return 1;
5002}
5003
5004static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events,
5005 wait_queue_func_t wake_func)
5006{
5007 poll->head = NULL;
5008 poll->done = false;
5009 poll->canceled = false;
5010 poll->events = events;
5011 INIT_LIST_HEAD(&poll->wait.entry);
5012 init_waitqueue_func_entry(&poll->wait, wake_func);
5013}
5014
5015static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt,
Jens Axboe807abcb2020-07-17 17:09:27 -06005016 struct wait_queue_head *head,
5017 struct io_poll_iocb **poll_ptr)
Jens Axboe18bceab2020-05-15 11:56:54 -06005018{
5019 struct io_kiocb *req = pt->req;
5020
5021 /*
5022 * If poll->head is already set, it's because the file being polled
5023 * uses multiple waitqueues for poll handling (eg one for read, one
5024 * for write). Setup a separate io_poll_iocb if this happens.
5025 */
5026 if (unlikely(poll->head)) {
Pavel Begunkov58852d42020-10-16 20:55:56 +01005027 struct io_poll_iocb *poll_one = poll;
5028
Jens Axboe18bceab2020-05-15 11:56:54 -06005029 /* already have a 2nd entry, fail a third attempt */
Jens Axboe807abcb2020-07-17 17:09:27 -06005030 if (*poll_ptr) {
Jens Axboe18bceab2020-05-15 11:56:54 -06005031 pt->error = -EINVAL;
5032 return;
5033 }
5034 poll = kmalloc(sizeof(*poll), GFP_ATOMIC);
5035 if (!poll) {
5036 pt->error = -ENOMEM;
5037 return;
5038 }
Pavel Begunkov58852d42020-10-16 20:55:56 +01005039 io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake);
Jens Axboe18bceab2020-05-15 11:56:54 -06005040 refcount_inc(&req->refs);
5041 poll->wait.private = req;
Jens Axboe807abcb2020-07-17 17:09:27 -06005042 *poll_ptr = poll;
Jens Axboe18bceab2020-05-15 11:56:54 -06005043 }
5044
5045 pt->error = 0;
5046 poll->head = head;
Jiufei Xuea31eb4a2020-06-17 17:53:56 +08005047
5048 if (poll->events & EPOLLEXCLUSIVE)
5049 add_wait_queue_exclusive(head, &poll->wait);
5050 else
5051 add_wait_queue(head, &poll->wait);
Jens Axboe18bceab2020-05-15 11:56:54 -06005052}
5053
5054static void io_async_queue_proc(struct file *file, struct wait_queue_head *head,
5055 struct poll_table_struct *p)
5056{
5057 struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
Jens Axboe807abcb2020-07-17 17:09:27 -06005058 struct async_poll *apoll = pt->req->apoll;
Jens Axboe18bceab2020-05-15 11:56:54 -06005059
Jens Axboe807abcb2020-07-17 17:09:27 -06005060 __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll);
Jens Axboe18bceab2020-05-15 11:56:54 -06005061}
5062
Jens Axboed7718a92020-02-14 22:23:12 -07005063static void io_async_task_func(struct callback_head *cb)
5064{
5065 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
5066 struct async_poll *apoll = req->apoll;
5067 struct io_ring_ctx *ctx = req->ctx;
5068
5069 trace_io_uring_task_run(req->ctx, req->opcode, req->user_data);
5070
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005071 if (io_poll_rewait(req, &apoll->poll)) {
Jens Axboed7718a92020-02-14 22:23:12 -07005072 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe6d816e02020-08-11 08:04:14 -06005073 percpu_ref_put(&ctx->refs);
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005074 return;
Jens Axboed7718a92020-02-14 22:23:12 -07005075 }
5076
Jens Axboe31067252020-05-17 17:43:31 -06005077 /* If req is still hashed, it cannot have been canceled. Don't check. */
Pavel Begunkov0be0b0e2020-06-30 15:20:42 +03005078 if (hash_hashed(&req->hash_node))
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005079 hash_del(&req->hash_node);
Jens Axboe2bae0472020-04-13 11:16:34 -06005080
Jens Axboed4e7cd32020-08-15 11:44:50 -07005081 io_poll_remove_double(req);
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005082 spin_unlock_irq(&ctx->completion_lock);
5083
Pavel Begunkov0be0b0e2020-06-30 15:20:42 +03005084 if (!READ_ONCE(apoll->poll.canceled))
5085 __io_req_task_submit(req);
5086 else
5087 __io_req_task_cancel(req, -ECANCELED);
Dan Carpenteraa340842020-07-08 21:47:11 +03005088
Jens Axboe6d816e02020-08-11 08:04:14 -06005089 percpu_ref_put(&ctx->refs);
Jens Axboe807abcb2020-07-17 17:09:27 -06005090 kfree(apoll->double_poll);
Jens Axboe31067252020-05-17 17:43:31 -06005091 kfree(apoll);
Jens Axboed7718a92020-02-14 22:23:12 -07005092}
5093
5094static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
5095 void *key)
5096{
5097 struct io_kiocb *req = wait->private;
5098 struct io_poll_iocb *poll = &req->apoll->poll;
5099
5100 trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data,
5101 key_to_poll(key));
5102
5103 return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func);
5104}
5105
5106static void io_poll_req_insert(struct io_kiocb *req)
5107{
5108 struct io_ring_ctx *ctx = req->ctx;
5109 struct hlist_head *list;
5110
5111 list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)];
5112 hlist_add_head(&req->hash_node, list);
5113}
5114
5115static __poll_t __io_arm_poll_handler(struct io_kiocb *req,
5116 struct io_poll_iocb *poll,
5117 struct io_poll_table *ipt, __poll_t mask,
5118 wait_queue_func_t wake_func)
5119 __acquires(&ctx->completion_lock)
5120{
5121 struct io_ring_ctx *ctx = req->ctx;
5122 bool cancel = false;
5123
Pavel Begunkov4d52f332020-10-18 10:17:43 +01005124 INIT_HLIST_NODE(&req->hash_node);
Jens Axboe18bceab2020-05-15 11:56:54 -06005125 io_init_poll_iocb(poll, mask, wake_func);
Pavel Begunkovb90cd192020-06-21 13:09:52 +03005126 poll->file = req->file;
Jens Axboe18bceab2020-05-15 11:56:54 -06005127 poll->wait.private = req;
Jens Axboed7718a92020-02-14 22:23:12 -07005128
5129 ipt->pt._key = mask;
5130 ipt->req = req;
5131 ipt->error = -EINVAL;
5132
Jens Axboed7718a92020-02-14 22:23:12 -07005133 mask = vfs_poll(req->file, &ipt->pt) & poll->events;
5134
5135 spin_lock_irq(&ctx->completion_lock);
5136 if (likely(poll->head)) {
5137 spin_lock(&poll->head->lock);
5138 if (unlikely(list_empty(&poll->wait.entry))) {
5139 if (ipt->error)
5140 cancel = true;
5141 ipt->error = 0;
5142 mask = 0;
5143 }
5144 if (mask || ipt->error)
5145 list_del_init(&poll->wait.entry);
5146 else if (cancel)
5147 WRITE_ONCE(poll->canceled, true);
5148 else if (!poll->done) /* actually waiting for an event */
5149 io_poll_req_insert(req);
5150 spin_unlock(&poll->head->lock);
5151 }
5152
5153 return mask;
5154}
5155
5156static bool io_arm_poll_handler(struct io_kiocb *req)
5157{
5158 const struct io_op_def *def = &io_op_defs[req->opcode];
5159 struct io_ring_ctx *ctx = req->ctx;
5160 struct async_poll *apoll;
5161 struct io_poll_table ipt;
5162 __poll_t mask, ret;
Jens Axboe9dab14b2020-08-25 12:27:50 -06005163 int rw;
Jens Axboed7718a92020-02-14 22:23:12 -07005164
5165 if (!req->file || !file_can_poll(req->file))
5166 return false;
Pavel Begunkov24c74672020-06-21 13:09:51 +03005167 if (req->flags & REQ_F_POLLED)
Jens Axboed7718a92020-02-14 22:23:12 -07005168 return false;
Jens Axboe9dab14b2020-08-25 12:27:50 -06005169 if (def->pollin)
5170 rw = READ;
5171 else if (def->pollout)
5172 rw = WRITE;
5173 else
5174 return false;
5175 /* if we can't nonblock try, then no point in arming a poll handler */
5176 if (!io_file_supports_async(req->file, rw))
Jens Axboed7718a92020-02-14 22:23:12 -07005177 return false;
5178
5179 apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC);
5180 if (unlikely(!apoll))
5181 return false;
Jens Axboe807abcb2020-07-17 17:09:27 -06005182 apoll->double_poll = NULL;
Jens Axboed7718a92020-02-14 22:23:12 -07005183
5184 req->flags |= REQ_F_POLLED;
Jens Axboed7718a92020-02-14 22:23:12 -07005185 req->apoll = apoll;
Jens Axboed7718a92020-02-14 22:23:12 -07005186
Nathan Chancellor8755d972020-03-02 16:01:19 -07005187 mask = 0;
Jens Axboed7718a92020-02-14 22:23:12 -07005188 if (def->pollin)
Nathan Chancellor8755d972020-03-02 16:01:19 -07005189 mask |= POLLIN | POLLRDNORM;
Jens Axboed7718a92020-02-14 22:23:12 -07005190 if (def->pollout)
5191 mask |= POLLOUT | POLLWRNORM;
Luke Hsiao901341b2020-08-21 21:41:05 -07005192
5193 /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */
5194 if ((req->opcode == IORING_OP_RECVMSG) &&
5195 (req->sr_msg.msg_flags & MSG_ERRQUEUE))
5196 mask &= ~POLLIN;
5197
Jens Axboed7718a92020-02-14 22:23:12 -07005198 mask |= POLLERR | POLLPRI;
5199
5200 ipt.pt._qproc = io_async_queue_proc;
5201
5202 ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask,
5203 io_async_wake);
Jens Axboea36da652020-08-11 09:50:19 -06005204 if (ret || ipt.error) {
Jens Axboed4e7cd32020-08-15 11:44:50 -07005205 io_poll_remove_double(req);
Jens Axboed7718a92020-02-14 22:23:12 -07005206 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe807abcb2020-07-17 17:09:27 -06005207 kfree(apoll->double_poll);
Jens Axboed7718a92020-02-14 22:23:12 -07005208 kfree(apoll);
5209 return false;
5210 }
5211 spin_unlock_irq(&ctx->completion_lock);
5212 trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask,
5213 apoll->poll.events);
5214 return true;
5215}
5216
5217static bool __io_poll_remove_one(struct io_kiocb *req,
5218 struct io_poll_iocb *poll)
5219{
Jens Axboeb41e9852020-02-17 09:52:41 -07005220 bool do_complete = false;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005221
5222 spin_lock(&poll->head->lock);
5223 WRITE_ONCE(poll->canceled, true);
Jens Axboe392edb42019-12-09 17:52:20 -07005224 if (!list_empty(&poll->wait.entry)) {
5225 list_del_init(&poll->wait.entry);
Jens Axboeb41e9852020-02-17 09:52:41 -07005226 do_complete = true;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005227 }
5228 spin_unlock(&poll->head->lock);
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005229 hash_del(&req->hash_node);
Jens Axboed7718a92020-02-14 22:23:12 -07005230 return do_complete;
5231}
5232
5233static bool io_poll_remove_one(struct io_kiocb *req)
5234{
5235 bool do_complete;
5236
Jens Axboed4e7cd32020-08-15 11:44:50 -07005237 io_poll_remove_double(req);
5238
Jens Axboed7718a92020-02-14 22:23:12 -07005239 if (req->opcode == IORING_OP_POLL_ADD) {
5240 do_complete = __io_poll_remove_one(req, &req->poll);
5241 } else {
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005242 struct async_poll *apoll = req->apoll;
5243
Jens Axboed7718a92020-02-14 22:23:12 -07005244 /* non-poll requests have submit ref still */
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005245 do_complete = __io_poll_remove_one(req, &apoll->poll);
5246 if (do_complete) {
Jens Axboed7718a92020-02-14 22:23:12 -07005247 io_put_req(req);
Jens Axboe807abcb2020-07-17 17:09:27 -06005248 kfree(apoll->double_poll);
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005249 kfree(apoll);
5250 }
Xiaoguang Wangb1f573b2020-04-12 14:50:54 +08005251 }
5252
Jens Axboeb41e9852020-02-17 09:52:41 -07005253 if (do_complete) {
5254 io_cqring_fill_event(req, -ECANCELED);
5255 io_commit_cqring(req->ctx);
Jens Axboef254ac02020-08-12 17:33:30 -06005256 req_set_fail_links(req);
Pavel Begunkov216578e2020-10-13 09:44:00 +01005257 io_put_req_deferred(req, 1);
Jens Axboeb41e9852020-02-17 09:52:41 -07005258 }
5259
5260 return do_complete;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005261}
5262
Jens Axboe76e1b642020-09-26 15:05:03 -06005263/*
5264 * Returns true if we found and killed one or more poll requests
5265 */
5266static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005267{
Jens Axboe78076bb2019-12-04 19:56:40 -07005268 struct hlist_node *tmp;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005269 struct io_kiocb *req;
Jens Axboe8e2e1fa2020-04-13 17:05:14 -06005270 int posted = 0, i;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005271
5272 spin_lock_irq(&ctx->completion_lock);
Jens Axboe78076bb2019-12-04 19:56:40 -07005273 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
5274 struct hlist_head *list;
5275
5276 list = &ctx->cancel_hash[i];
Jens Axboef3606e32020-09-22 08:18:24 -06005277 hlist_for_each_entry_safe(req, tmp, list, hash_node) {
5278 if (io_task_match(req, tsk))
5279 posted += io_poll_remove_one(req);
5280 }
Jens Axboe221c5eb2019-01-17 09:41:58 -07005281 }
5282 spin_unlock_irq(&ctx->completion_lock);
Jens Axboeb41e9852020-02-17 09:52:41 -07005283
Jens Axboe8e2e1fa2020-04-13 17:05:14 -06005284 if (posted)
5285 io_cqring_ev_posted(ctx);
Jens Axboe76e1b642020-09-26 15:05:03 -06005286
5287 return posted != 0;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005288}
5289
Jens Axboe47f46762019-11-09 17:43:02 -07005290static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr)
5291{
Jens Axboe78076bb2019-12-04 19:56:40 -07005292 struct hlist_head *list;
Jens Axboe47f46762019-11-09 17:43:02 -07005293 struct io_kiocb *req;
5294
Jens Axboe78076bb2019-12-04 19:56:40 -07005295 list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)];
5296 hlist_for_each_entry(req, list, hash_node) {
Jens Axboeb41e9852020-02-17 09:52:41 -07005297 if (sqe_addr != req->user_data)
5298 continue;
5299 if (io_poll_remove_one(req))
Jens Axboeeac406c62019-11-14 12:09:58 -07005300 return 0;
Jens Axboeb41e9852020-02-17 09:52:41 -07005301 return -EALREADY;
Jens Axboe47f46762019-11-09 17:43:02 -07005302 }
5303
5304 return -ENOENT;
5305}
5306
Jens Axboe3529d8c2019-12-19 18:24:38 -07005307static int io_poll_remove_prep(struct io_kiocb *req,
5308 const struct io_uring_sqe *sqe)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005309{
Jens Axboe221c5eb2019-01-17 09:41:58 -07005310 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5311 return -EINVAL;
5312 if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index ||
5313 sqe->poll_events)
5314 return -EINVAL;
5315
Jens Axboe0969e782019-12-17 18:40:57 -07005316 req->poll.addr = READ_ONCE(sqe->addr);
Jens Axboe0969e782019-12-17 18:40:57 -07005317 return 0;
5318}
5319
5320/*
5321 * Find a running poll command that matches one specified in sqe->addr,
5322 * and remove it if found.
5323 */
5324static int io_poll_remove(struct io_kiocb *req)
5325{
5326 struct io_ring_ctx *ctx = req->ctx;
5327 u64 addr;
5328 int ret;
5329
Jens Axboe0969e782019-12-17 18:40:57 -07005330 addr = req->poll.addr;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005331 spin_lock_irq(&ctx->completion_lock);
Jens Axboe0969e782019-12-17 18:40:57 -07005332 ret = io_poll_cancel(ctx, addr);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005333 spin_unlock_irq(&ctx->completion_lock);
5334
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005335 if (ret < 0)
5336 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06005337 io_req_complete(req, ret);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005338 return 0;
5339}
5340
Jens Axboe221c5eb2019-01-17 09:41:58 -07005341static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
5342 void *key)
5343{
Jens Axboec2f2eb72020-02-10 09:07:05 -07005344 struct io_kiocb *req = wait->private;
5345 struct io_poll_iocb *poll = &req->poll;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005346
Jens Axboed7718a92020-02-14 22:23:12 -07005347 return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005348}
5349
Jens Axboe221c5eb2019-01-17 09:41:58 -07005350static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head,
5351 struct poll_table_struct *p)
5352{
5353 struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
5354
Jens Axboee8c2bc12020-08-15 18:44:09 -07005355 __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data);
Jens Axboeeac406c62019-11-14 12:09:58 -07005356}
5357
Jens Axboe3529d8c2019-12-19 18:24:38 -07005358static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005359{
5360 struct io_poll_iocb *poll = &req->poll;
Jiufei Xue5769a352020-06-17 17:53:55 +08005361 u32 events;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005362
5363 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5364 return -EINVAL;
5365 if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index)
5366 return -EINVAL;
5367
Jiufei Xue5769a352020-06-17 17:53:55 +08005368 events = READ_ONCE(sqe->poll32_events);
5369#ifdef __BIG_ENDIAN
5370 events = swahw32(events);
5371#endif
Jiufei Xuea31eb4a2020-06-17 17:53:56 +08005372 poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP |
5373 (events & EPOLLEXCLUSIVE);
Jens Axboe0969e782019-12-17 18:40:57 -07005374 return 0;
5375}
5376
Pavel Begunkov014db002020-03-03 21:33:12 +03005377static int io_poll_add(struct io_kiocb *req)
Jens Axboe0969e782019-12-17 18:40:57 -07005378{
5379 struct io_poll_iocb *poll = &req->poll;
5380 struct io_ring_ctx *ctx = req->ctx;
5381 struct io_poll_table ipt;
Jens Axboe0969e782019-12-17 18:40:57 -07005382 __poll_t mask;
Jens Axboe0969e782019-12-17 18:40:57 -07005383
Jens Axboed7718a92020-02-14 22:23:12 -07005384 ipt.pt._qproc = io_poll_queue_proc;
Jens Axboe36703242019-07-25 10:20:18 -06005385
Jens Axboed7718a92020-02-14 22:23:12 -07005386 mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events,
5387 io_poll_wake);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005388
Jens Axboe8c838782019-03-12 15:48:16 -06005389 if (mask) { /* no async, we'd stolen it */
Jens Axboe8c838782019-03-12 15:48:16 -06005390 ipt.error = 0;
Jens Axboeb0dd8a42019-11-18 12:14:54 -07005391 io_poll_complete(req, mask, 0);
Jens Axboe8c838782019-03-12 15:48:16 -06005392 }
Jens Axboe221c5eb2019-01-17 09:41:58 -07005393 spin_unlock_irq(&ctx->completion_lock);
5394
Jens Axboe8c838782019-03-12 15:48:16 -06005395 if (mask) {
5396 io_cqring_ev_posted(ctx);
Pavel Begunkov014db002020-03-03 21:33:12 +03005397 io_put_req(req);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005398 }
Jens Axboe8c838782019-03-12 15:48:16 -06005399 return ipt.error;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005400}
5401
Jens Axboe5262f562019-09-17 12:26:57 -06005402static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer)
5403{
Jens Axboead8a48a2019-11-15 08:49:11 -07005404 struct io_timeout_data *data = container_of(timer,
5405 struct io_timeout_data, timer);
5406 struct io_kiocb *req = data->req;
5407 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5262f562019-09-17 12:26:57 -06005408 unsigned long flags;
5409
Jens Axboe5262f562019-09-17 12:26:57 -06005410 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkova71976f2020-10-10 18:34:11 +01005411 list_del_init(&req->timeout.list);
Pavel Begunkov01cec8c2020-07-30 18:43:50 +03005412 atomic_set(&req->ctx->cq_timeouts,
5413 atomic_read(&req->ctx->cq_timeouts) + 1);
5414
Jens Axboe78e19bb2019-11-06 15:21:34 -07005415 io_cqring_fill_event(req, -ETIME);
Jens Axboe5262f562019-09-17 12:26:57 -06005416 io_commit_cqring(ctx);
5417 spin_unlock_irqrestore(&ctx->completion_lock, flags);
5418
5419 io_cqring_ev_posted(ctx);
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005420 req_set_fail_links(req);
Jens Axboe5262f562019-09-17 12:26:57 -06005421 io_put_req(req);
5422 return HRTIMER_NORESTART;
5423}
5424
Jens Axboef254ac02020-08-12 17:33:30 -06005425static int __io_timeout_cancel(struct io_kiocb *req)
Jens Axboe47f46762019-11-09 17:43:02 -07005426{
Jens Axboee8c2bc12020-08-15 18:44:09 -07005427 struct io_timeout_data *io = req->async_data;
Jens Axboef254ac02020-08-12 17:33:30 -06005428 int ret;
Jens Axboe47f46762019-11-09 17:43:02 -07005429
Jens Axboee8c2bc12020-08-15 18:44:09 -07005430 ret = hrtimer_try_to_cancel(&io->timer);
Jens Axboe47f46762019-11-09 17:43:02 -07005431 if (ret == -1)
5432 return -EALREADY;
Pavel Begunkova71976f2020-10-10 18:34:11 +01005433 list_del_init(&req->timeout.list);
Jens Axboe47f46762019-11-09 17:43:02 -07005434
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005435 req_set_fail_links(req);
Jens Axboe47f46762019-11-09 17:43:02 -07005436 io_cqring_fill_event(req, -ECANCELED);
Pavel Begunkov216578e2020-10-13 09:44:00 +01005437 io_put_req_deferred(req, 1);
Jens Axboe47f46762019-11-09 17:43:02 -07005438 return 0;
5439}
5440
Jens Axboef254ac02020-08-12 17:33:30 -06005441static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data)
5442{
5443 struct io_kiocb *req;
5444 int ret = -ENOENT;
5445
5446 list_for_each_entry(req, &ctx->timeout_list, timeout.list) {
5447 if (user_data == req->user_data) {
5448 ret = 0;
5449 break;
5450 }
5451 }
5452
5453 if (ret == -ENOENT)
5454 return ret;
5455
5456 return __io_timeout_cancel(req);
5457}
5458
Jens Axboe3529d8c2019-12-19 18:24:38 -07005459static int io_timeout_remove_prep(struct io_kiocb *req,
5460 const struct io_uring_sqe *sqe)
Jens Axboeb29472e2019-12-17 18:50:29 -07005461{
Jens Axboeb29472e2019-12-17 18:50:29 -07005462 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5463 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005464 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5465 return -EINVAL;
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +01005466 if (sqe->ioprio || sqe->buf_index || sqe->len || sqe->timeout_flags)
Jens Axboeb29472e2019-12-17 18:50:29 -07005467 return -EINVAL;
5468
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +01005469 req->timeout_rem.addr = READ_ONCE(sqe->addr);
Jens Axboeb29472e2019-12-17 18:50:29 -07005470 return 0;
5471}
5472
Jens Axboe11365042019-10-16 09:08:32 -06005473/*
5474 * Remove or update an existing timeout command
5475 */
Jens Axboefc4df992019-12-10 14:38:45 -07005476static int io_timeout_remove(struct io_kiocb *req)
Jens Axboe11365042019-10-16 09:08:32 -06005477{
5478 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe47f46762019-11-09 17:43:02 -07005479 int ret;
Jens Axboe11365042019-10-16 09:08:32 -06005480
Jens Axboe11365042019-10-16 09:08:32 -06005481 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +01005482 ret = io_timeout_cancel(ctx, req->timeout_rem.addr);
Jens Axboe11365042019-10-16 09:08:32 -06005483
Jens Axboe47f46762019-11-09 17:43:02 -07005484 io_cqring_fill_event(req, ret);
Jens Axboe11365042019-10-16 09:08:32 -06005485 io_commit_cqring(ctx);
5486 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe5262f562019-09-17 12:26:57 -06005487 io_cqring_ev_posted(ctx);
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005488 if (ret < 0)
5489 req_set_fail_links(req);
Jackie Liuec9c02a2019-11-08 23:50:36 +08005490 io_put_req(req);
Jens Axboe11365042019-10-16 09:08:32 -06005491 return 0;
Jens Axboe5262f562019-09-17 12:26:57 -06005492}
5493
Jens Axboe3529d8c2019-12-19 18:24:38 -07005494static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe,
Jens Axboe2d283902019-12-04 11:08:05 -07005495 bool is_timeout_link)
Jens Axboe5262f562019-09-17 12:26:57 -06005496{
Jens Axboead8a48a2019-11-15 08:49:11 -07005497 struct io_timeout_data *data;
Jens Axboea41525a2019-10-15 16:48:15 -06005498 unsigned flags;
Pavel Begunkov56080b02020-05-26 20:34:04 +03005499 u32 off = READ_ONCE(sqe->off);
Jens Axboe5262f562019-09-17 12:26:57 -06005500
Jens Axboead8a48a2019-11-15 08:49:11 -07005501 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe5262f562019-09-17 12:26:57 -06005502 return -EINVAL;
Jens Axboead8a48a2019-11-15 08:49:11 -07005503 if (sqe->ioprio || sqe->buf_index || sqe->len != 1)
Jens Axboea41525a2019-10-15 16:48:15 -06005504 return -EINVAL;
Pavel Begunkov56080b02020-05-26 20:34:04 +03005505 if (off && is_timeout_link)
Jens Axboe2d283902019-12-04 11:08:05 -07005506 return -EINVAL;
Jens Axboea41525a2019-10-15 16:48:15 -06005507 flags = READ_ONCE(sqe->timeout_flags);
5508 if (flags & ~IORING_TIMEOUT_ABS)
Jens Axboe5262f562019-09-17 12:26:57 -06005509 return -EINVAL;
Arnd Bergmannbdf20072019-10-01 09:53:29 -06005510
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005511 req->timeout.off = off;
Jens Axboe26a61672019-12-20 09:02:01 -07005512
Jens Axboee8c2bc12020-08-15 18:44:09 -07005513 if (!req->async_data && io_alloc_async_data(req))
Jens Axboe26a61672019-12-20 09:02:01 -07005514 return -ENOMEM;
5515
Jens Axboee8c2bc12020-08-15 18:44:09 -07005516 data = req->async_data;
Jens Axboead8a48a2019-11-15 08:49:11 -07005517 data->req = req;
Jens Axboead8a48a2019-11-15 08:49:11 -07005518
5519 if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr)))
Jens Axboe5262f562019-09-17 12:26:57 -06005520 return -EFAULT;
5521
Jens Axboe11365042019-10-16 09:08:32 -06005522 if (flags & IORING_TIMEOUT_ABS)
Jens Axboead8a48a2019-11-15 08:49:11 -07005523 data->mode = HRTIMER_MODE_ABS;
Jens Axboe11365042019-10-16 09:08:32 -06005524 else
Jens Axboead8a48a2019-11-15 08:49:11 -07005525 data->mode = HRTIMER_MODE_REL;
Jens Axboe11365042019-10-16 09:08:32 -06005526
Jens Axboead8a48a2019-11-15 08:49:11 -07005527 hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode);
5528 return 0;
5529}
5530
Jens Axboefc4df992019-12-10 14:38:45 -07005531static int io_timeout(struct io_kiocb *req)
Jens Axboead8a48a2019-11-15 08:49:11 -07005532{
Jens Axboead8a48a2019-11-15 08:49:11 -07005533 struct io_ring_ctx *ctx = req->ctx;
Jens Axboee8c2bc12020-08-15 18:44:09 -07005534 struct io_timeout_data *data = req->async_data;
Jens Axboead8a48a2019-11-15 08:49:11 -07005535 struct list_head *entry;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005536 u32 tail, off = req->timeout.off;
Jens Axboead8a48a2019-11-15 08:49:11 -07005537
Pavel Begunkov733f5c92020-05-26 20:34:03 +03005538 spin_lock_irq(&ctx->completion_lock);
Jens Axboe93bd25b2019-11-11 23:34:31 -07005539
Jens Axboe5262f562019-09-17 12:26:57 -06005540 /*
5541 * sqe->off holds how many events that need to occur for this
Jens Axboe93bd25b2019-11-11 23:34:31 -07005542 * timeout event to be satisfied. If it isn't set, then this is
5543 * a pure timeout request, sequence isn't used.
Jens Axboe5262f562019-09-17 12:26:57 -06005544 */
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03005545 if (io_is_timeout_noseq(req)) {
Jens Axboe93bd25b2019-11-11 23:34:31 -07005546 entry = ctx->timeout_list.prev;
5547 goto add;
5548 }
Jens Axboe5262f562019-09-17 12:26:57 -06005549
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005550 tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
5551 req->timeout.target_seq = tail + off;
Jens Axboe5262f562019-09-17 12:26:57 -06005552
5553 /*
5554 * Insertion sort, ensuring the first entry in the list is always
5555 * the one we need first.
5556 */
Jens Axboe5262f562019-09-17 12:26:57 -06005557 list_for_each_prev(entry, &ctx->timeout_list) {
Pavel Begunkov135fcde2020-07-13 23:37:12 +03005558 struct io_kiocb *nxt = list_entry(entry, struct io_kiocb,
5559 timeout.list);
Jens Axboe5262f562019-09-17 12:26:57 -06005560
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03005561 if (io_is_timeout_noseq(nxt))
Jens Axboe93bd25b2019-11-11 23:34:31 -07005562 continue;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005563 /* nxt.seq is behind @tail, otherwise would've been completed */
5564 if (off >= nxt->timeout.target_seq - tail)
Jens Axboe5262f562019-09-17 12:26:57 -06005565 break;
5566 }
Jens Axboe93bd25b2019-11-11 23:34:31 -07005567add:
Pavel Begunkov135fcde2020-07-13 23:37:12 +03005568 list_add(&req->timeout.list, entry);
Jens Axboead8a48a2019-11-15 08:49:11 -07005569 data->timer.function = io_timeout_fn;
5570 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode);
Jens Axboe842f9612019-10-29 12:34:10 -06005571 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe5262f562019-09-17 12:26:57 -06005572 return 0;
5573}
5574
Jens Axboe62755e32019-10-28 21:49:21 -06005575static bool io_cancel_cb(struct io_wq_work *work, void *data)
Jens Axboede0617e2019-04-06 21:51:27 -06005576{
Jens Axboe62755e32019-10-28 21:49:21 -06005577 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Jens Axboede0617e2019-04-06 21:51:27 -06005578
Jens Axboe62755e32019-10-28 21:49:21 -06005579 return req->user_data == (unsigned long) data;
5580}
5581
Jens Axboee977d6d2019-11-05 12:39:45 -07005582static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr)
Jens Axboe62755e32019-10-28 21:49:21 -06005583{
Jens Axboe62755e32019-10-28 21:49:21 -06005584 enum io_wq_cancel cancel_ret;
Jens Axboe62755e32019-10-28 21:49:21 -06005585 int ret = 0;
5586
Pavel Begunkov4f26bda2020-06-15 10:24:03 +03005587 cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr, false);
Jens Axboe62755e32019-10-28 21:49:21 -06005588 switch (cancel_ret) {
5589 case IO_WQ_CANCEL_OK:
5590 ret = 0;
5591 break;
5592 case IO_WQ_CANCEL_RUNNING:
5593 ret = -EALREADY;
5594 break;
5595 case IO_WQ_CANCEL_NOTFOUND:
5596 ret = -ENOENT;
5597 break;
5598 }
5599
Jens Axboee977d6d2019-11-05 12:39:45 -07005600 return ret;
5601}
5602
Jens Axboe47f46762019-11-09 17:43:02 -07005603static void io_async_find_and_cancel(struct io_ring_ctx *ctx,
5604 struct io_kiocb *req, __u64 sqe_addr,
Pavel Begunkov014db002020-03-03 21:33:12 +03005605 int success_ret)
Jens Axboe47f46762019-11-09 17:43:02 -07005606{
5607 unsigned long flags;
5608 int ret;
5609
5610 ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr);
5611 if (ret != -ENOENT) {
5612 spin_lock_irqsave(&ctx->completion_lock, flags);
5613 goto done;
5614 }
5615
5616 spin_lock_irqsave(&ctx->completion_lock, flags);
5617 ret = io_timeout_cancel(ctx, sqe_addr);
5618 if (ret != -ENOENT)
5619 goto done;
5620 ret = io_poll_cancel(ctx, sqe_addr);
5621done:
Jens Axboeb0dd8a42019-11-18 12:14:54 -07005622 if (!ret)
5623 ret = success_ret;
Jens Axboe47f46762019-11-09 17:43:02 -07005624 io_cqring_fill_event(req, ret);
5625 io_commit_cqring(ctx);
5626 spin_unlock_irqrestore(&ctx->completion_lock, flags);
5627 io_cqring_ev_posted(ctx);
5628
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005629 if (ret < 0)
5630 req_set_fail_links(req);
Pavel Begunkov014db002020-03-03 21:33:12 +03005631 io_put_req(req);
Jens Axboe47f46762019-11-09 17:43:02 -07005632}
5633
Jens Axboe3529d8c2019-12-19 18:24:38 -07005634static int io_async_cancel_prep(struct io_kiocb *req,
5635 const struct io_uring_sqe *sqe)
Jens Axboee977d6d2019-11-05 12:39:45 -07005636{
Jens Axboefbf23842019-12-17 18:45:56 -07005637 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboee977d6d2019-11-05 12:39:45 -07005638 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005639 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5640 return -EINVAL;
5641 if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags)
Jens Axboee977d6d2019-11-05 12:39:45 -07005642 return -EINVAL;
5643
Jens Axboefbf23842019-12-17 18:45:56 -07005644 req->cancel.addr = READ_ONCE(sqe->addr);
5645 return 0;
5646}
5647
Pavel Begunkov014db002020-03-03 21:33:12 +03005648static int io_async_cancel(struct io_kiocb *req)
Jens Axboefbf23842019-12-17 18:45:56 -07005649{
5650 struct io_ring_ctx *ctx = req->ctx;
Jens Axboefbf23842019-12-17 18:45:56 -07005651
Pavel Begunkov014db002020-03-03 21:33:12 +03005652 io_async_find_and_cancel(ctx, req, req->cancel.addr, 0);
Jens Axboe62755e32019-10-28 21:49:21 -06005653 return 0;
5654}
5655
Jens Axboe05f3fb32019-12-09 11:22:50 -07005656static int io_files_update_prep(struct io_kiocb *req,
5657 const struct io_uring_sqe *sqe)
5658{
Jens Axboe6ca56f82020-09-18 16:51:19 -06005659 if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL))
5660 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005661 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5662 return -EINVAL;
5663 if (sqe->ioprio || sqe->rw_flags)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005664 return -EINVAL;
5665
5666 req->files_update.offset = READ_ONCE(sqe->off);
5667 req->files_update.nr_args = READ_ONCE(sqe->len);
5668 if (!req->files_update.nr_args)
5669 return -EINVAL;
5670 req->files_update.arg = READ_ONCE(sqe->addr);
5671 return 0;
5672}
5673
Jens Axboe229a7b62020-06-22 10:13:11 -06005674static int io_files_update(struct io_kiocb *req, bool force_nonblock,
5675 struct io_comp_state *cs)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005676{
5677 struct io_ring_ctx *ctx = req->ctx;
5678 struct io_uring_files_update up;
5679 int ret;
5680
Jens Axboef86cd202020-01-29 13:46:44 -07005681 if (force_nonblock)
Jens Axboe05f3fb32019-12-09 11:22:50 -07005682 return -EAGAIN;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005683
5684 up.offset = req->files_update.offset;
5685 up.fds = req->files_update.arg;
5686
5687 mutex_lock(&ctx->uring_lock);
5688 ret = __io_sqe_files_update(ctx, &up, req->files_update.nr_args);
5689 mutex_unlock(&ctx->uring_lock);
5690
5691 if (ret < 0)
5692 req_set_fail_links(req);
Jens Axboe229a7b62020-06-22 10:13:11 -06005693 __io_req_complete(req, ret, 0, cs);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005694 return 0;
5695}
5696
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005697static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07005698{
Jens Axboed625c6e2019-12-17 19:53:05 -07005699 switch (req->opcode) {
Jens Axboee7815732019-12-17 19:45:06 -07005700 case IORING_OP_NOP:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005701 return 0;
Jens Axboef67676d2019-12-02 11:03:47 -07005702 case IORING_OP_READV:
5703 case IORING_OP_READ_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005704 case IORING_OP_READ:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005705 return io_read_prep(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07005706 case IORING_OP_WRITEV:
5707 case IORING_OP_WRITE_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005708 case IORING_OP_WRITE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005709 return io_write_prep(req, sqe);
Jens Axboe0969e782019-12-17 18:40:57 -07005710 case IORING_OP_POLL_ADD:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005711 return io_poll_add_prep(req, sqe);
Jens Axboe0969e782019-12-17 18:40:57 -07005712 case IORING_OP_POLL_REMOVE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005713 return io_poll_remove_prep(req, sqe);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005714 case IORING_OP_FSYNC:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005715 return io_prep_fsync(req, sqe);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005716 case IORING_OP_SYNC_FILE_RANGE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005717 return io_prep_sfr(req, sqe);
Jens Axboe03b12302019-12-02 18:50:25 -07005718 case IORING_OP_SENDMSG:
Jens Axboefddafac2020-01-04 20:19:44 -07005719 case IORING_OP_SEND:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005720 return io_sendmsg_prep(req, sqe);
Jens Axboe03b12302019-12-02 18:50:25 -07005721 case IORING_OP_RECVMSG:
Jens Axboefddafac2020-01-04 20:19:44 -07005722 case IORING_OP_RECV:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005723 return io_recvmsg_prep(req, sqe);
Jens Axboef499a022019-12-02 16:28:46 -07005724 case IORING_OP_CONNECT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005725 return io_connect_prep(req, sqe);
Jens Axboe2d283902019-12-04 11:08:05 -07005726 case IORING_OP_TIMEOUT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005727 return io_timeout_prep(req, sqe, false);
Jens Axboeb29472e2019-12-17 18:50:29 -07005728 case IORING_OP_TIMEOUT_REMOVE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005729 return io_timeout_remove_prep(req, sqe);
Jens Axboefbf23842019-12-17 18:45:56 -07005730 case IORING_OP_ASYNC_CANCEL:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005731 return io_async_cancel_prep(req, sqe);
Jens Axboe2d283902019-12-04 11:08:05 -07005732 case IORING_OP_LINK_TIMEOUT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005733 return io_timeout_prep(req, sqe, true);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005734 case IORING_OP_ACCEPT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005735 return io_accept_prep(req, sqe);
Jens Axboed63d1b52019-12-10 10:38:56 -07005736 case IORING_OP_FALLOCATE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005737 return io_fallocate_prep(req, sqe);
Jens Axboe15b71ab2019-12-11 11:20:36 -07005738 case IORING_OP_OPENAT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005739 return io_openat_prep(req, sqe);
Jens Axboeb5dba592019-12-11 14:02:38 -07005740 case IORING_OP_CLOSE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005741 return io_close_prep(req, sqe);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005742 case IORING_OP_FILES_UPDATE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005743 return io_files_update_prep(req, sqe);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07005744 case IORING_OP_STATX:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005745 return io_statx_prep(req, sqe);
Jens Axboe4840e412019-12-25 22:03:45 -07005746 case IORING_OP_FADVISE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005747 return io_fadvise_prep(req, sqe);
Jens Axboec1ca7572019-12-25 22:18:28 -07005748 case IORING_OP_MADVISE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005749 return io_madvise_prep(req, sqe);
Jens Axboecebdb982020-01-08 17:59:24 -07005750 case IORING_OP_OPENAT2:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005751 return io_openat2_prep(req, sqe);
Jens Axboe3e4827b2020-01-08 15:18:09 -07005752 case IORING_OP_EPOLL_CTL:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005753 return io_epoll_ctl_prep(req, sqe);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03005754 case IORING_OP_SPLICE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005755 return io_splice_prep(req, sqe);
Jens Axboeddf0322d2020-02-23 16:41:33 -07005756 case IORING_OP_PROVIDE_BUFFERS:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005757 return io_provide_buffers_prep(req, sqe);
Jens Axboe067524e2020-03-02 16:32:28 -07005758 case IORING_OP_REMOVE_BUFFERS:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005759 return io_remove_buffers_prep(req, sqe);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03005760 case IORING_OP_TEE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005761 return io_tee_prep(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07005762 }
5763
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005764 printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n",
5765 req->opcode);
5766 return-EINVAL;
5767}
5768
Jens Axboedef596e2019-01-09 08:59:42 -07005769static int io_req_defer_prep(struct io_kiocb *req,
5770 const struct io_uring_sqe *sqe)
Jens Axboedef596e2019-01-09 08:59:42 -07005771{
Jens Axboedef596e2019-01-09 08:59:42 -07005772 if (!sqe)
Jens Axboe2b188cc2019-01-07 10:46:33 -07005773 return 0;
Jens Axboee8c2bc12020-08-15 18:44:09 -07005774 if (io_alloc_async_data(req))
Jens Axboeb76da702019-11-20 13:05:32 -07005775 return -EAGAIN;
Pavel Begunkovbfe76552020-09-30 22:57:55 +03005776 return io_req_prep(req, sqe);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005777}
5778
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005779static u32 io_get_sequence(struct io_kiocb *req)
5780{
5781 struct io_kiocb *pos;
5782 struct io_ring_ctx *ctx = req->ctx;
5783 u32 total_submitted, nr_reqs = 1;
5784
5785 if (req->flags & REQ_F_LINK_HEAD)
5786 list_for_each_entry(pos, &req->link_list, link_list)
5787 nr_reqs++;
5788
5789 total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped;
5790 return total_submitted - nr_reqs;
5791}
5792
Jens Axboe3529d8c2019-12-19 18:24:38 -07005793static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe2b188cc2019-01-07 10:46:33 -07005794{
5795 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005796 struct io_defer_entry *de;
Jens Axboe2b188cc2019-01-07 10:46:33 -07005797 int ret;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005798 u32 seq;
Jens Axboe2b188cc2019-01-07 10:46:33 -07005799
5800 /* Still need defer if there is pending req in defer list. */
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005801 if (likely(list_empty_careful(&ctx->defer_list) &&
5802 !(req->flags & REQ_F_IO_DRAIN)))
5803 return 0;
5804
5805 seq = io_get_sequence(req);
5806 /* Still a chance to pass the sequence check */
5807 if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list))
Jens Axboe2b188cc2019-01-07 10:46:33 -07005808 return 0;
5809
Jens Axboee8c2bc12020-08-15 18:44:09 -07005810 if (!req->async_data) {
Pavel Begunkov650b5482020-05-17 14:02:11 +03005811 ret = io_req_defer_prep(req, sqe);
Pavel Begunkov327d6d92020-07-15 12:46:51 +03005812 if (ret)
Pavel Begunkov650b5482020-05-17 14:02:11 +03005813 return ret;
5814 }
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03005815 io_prep_async_link(req);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005816 de = kmalloc(sizeof(*de), GFP_KERNEL);
5817 if (!de)
5818 return -ENOMEM;
Jens Axboe31b51512019-01-18 22:56:34 -07005819
5820 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005821 if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) {
Jens Axboe31b51512019-01-18 22:56:34 -07005822 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005823 kfree(de);
Pavel Begunkovae348172020-07-23 20:25:20 +03005824 io_queue_async_work(req);
5825 return -EIOCBQUEUED;
Jens Axboe31b51512019-01-18 22:56:34 -07005826 }
5827
5828 trace_io_uring_defer(ctx, req, req->user_data);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005829 de->req = req;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03005830 de->seq = seq;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03005831 list_add_tail(&de->list, &ctx->defer_list);
Jens Axboe31b51512019-01-18 22:56:34 -07005832 spin_unlock_irq(&ctx->completion_lock);
5833 return -EIOCBQUEUED;
5834}
Jens Axboeedafcce2019-01-09 09:16:05 -07005835
Jens Axboef573d382020-09-22 10:19:24 -06005836static void io_req_drop_files(struct io_kiocb *req)
5837{
5838 struct io_ring_ctx *ctx = req->ctx;
5839 unsigned long flags;
5840
5841 spin_lock_irqsave(&ctx->inflight_lock, flags);
5842 list_del(&req->inflight_entry);
5843 if (waitqueue_active(&ctx->inflight_wait))
5844 wake_up(&ctx->inflight_wait);
5845 spin_unlock_irqrestore(&ctx->inflight_lock, flags);
5846 req->flags &= ~REQ_F_INFLIGHT;
Jens Axboe98447d62020-10-14 10:48:51 -06005847 put_files_struct(req->work.identity->files);
5848 put_nsproxy(req->work.identity->nsproxy);
Jens Axboedfead8a2020-10-14 10:12:37 -06005849 req->work.flags &= ~IO_WQ_WORK_FILES;
Jens Axboef573d382020-09-22 10:19:24 -06005850}
5851
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03005852static void __io_clean_op(struct io_kiocb *req)
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03005853{
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005854 if (req->flags & REQ_F_BUFFER_SELECTED) {
5855 switch (req->opcode) {
5856 case IORING_OP_READV:
5857 case IORING_OP_READ_FIXED:
5858 case IORING_OP_READ:
Jens Axboebcda7ba2020-02-23 16:42:51 -07005859 kfree((void *)(unsigned long)req->rw.addr);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005860 break;
5861 case IORING_OP_RECVMSG:
5862 case IORING_OP_RECV:
Jens Axboe52de1fe2020-02-27 10:15:42 -07005863 kfree(req->sr_msg.kbuf);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005864 break;
5865 }
5866 req->flags &= ~REQ_F_BUFFER_SELECTED;
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03005867 }
5868
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005869 if (req->flags & REQ_F_NEED_CLEANUP) {
5870 switch (req->opcode) {
5871 case IORING_OP_READV:
5872 case IORING_OP_READ_FIXED:
5873 case IORING_OP_READ:
5874 case IORING_OP_WRITEV:
5875 case IORING_OP_WRITE_FIXED:
Jens Axboee8c2bc12020-08-15 18:44:09 -07005876 case IORING_OP_WRITE: {
5877 struct io_async_rw *io = req->async_data;
5878 if (io->free_iovec)
5879 kfree(io->free_iovec);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005880 break;
Jens Axboee8c2bc12020-08-15 18:44:09 -07005881 }
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005882 case IORING_OP_RECVMSG:
Jens Axboee8c2bc12020-08-15 18:44:09 -07005883 case IORING_OP_SENDMSG: {
5884 struct io_async_msghdr *io = req->async_data;
5885 if (io->iov != io->fast_iov)
5886 kfree(io->iov);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005887 break;
Jens Axboee8c2bc12020-08-15 18:44:09 -07005888 }
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005889 case IORING_OP_SPLICE:
5890 case IORING_OP_TEE:
5891 io_put_file(req, req->splice.file_in,
5892 (req->splice.flags & SPLICE_F_FD_IN_FIXED));
5893 break;
Jens Axboef3cd48502020-09-24 14:55:54 -06005894 case IORING_OP_OPENAT:
5895 case IORING_OP_OPENAT2:
5896 if (req->open.filename)
5897 putname(req->open.filename);
5898 break;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03005899 }
5900 req->flags &= ~REQ_F_NEED_CLEANUP;
5901 }
Pavel Begunkovbb175342020-08-20 11:33:35 +03005902
Jens Axboef573d382020-09-22 10:19:24 -06005903 if (req->flags & REQ_F_INFLIGHT)
5904 io_req_drop_files(req);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03005905}
5906
Pavel Begunkovc1379e22020-09-30 22:57:56 +03005907static int io_issue_sqe(struct io_kiocb *req, bool force_nonblock,
5908 struct io_comp_state *cs)
Jens Axboeedafcce2019-01-09 09:16:05 -07005909{
Jens Axboeedafcce2019-01-09 09:16:05 -07005910 struct io_ring_ctx *ctx = req->ctx;
Jens Axboed625c6e2019-12-17 19:53:05 -07005911 int ret;
Jens Axboeedafcce2019-01-09 09:16:05 -07005912
Jens Axboed625c6e2019-12-17 19:53:05 -07005913 switch (req->opcode) {
Jens Axboe2b188cc2019-01-07 10:46:33 -07005914 case IORING_OP_NOP:
Jens Axboe229a7b62020-06-22 10:13:11 -06005915 ret = io_nop(req, cs);
Jens Axboe31b51512019-01-18 22:56:34 -07005916 break;
5917 case IORING_OP_READV:
Jens Axboe3529d8c2019-12-19 18:24:38 -07005918 case IORING_OP_READ_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005919 case IORING_OP_READ:
Jens Axboea1d7c392020-06-22 11:09:46 -06005920 ret = io_read(req, force_nonblock, cs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005921 break;
5922 case IORING_OP_WRITEV:
Jens Axboe2b188cc2019-01-07 10:46:33 -07005923 case IORING_OP_WRITE_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07005924 case IORING_OP_WRITE:
Jens Axboea1d7c392020-06-22 11:09:46 -06005925 ret = io_write(req, force_nonblock, cs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005926 break;
5927 case IORING_OP_FSYNC:
Pavel Begunkov014db002020-03-03 21:33:12 +03005928 ret = io_fsync(req, force_nonblock);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005929 break;
5930 case IORING_OP_POLL_ADD:
Pavel Begunkov014db002020-03-03 21:33:12 +03005931 ret = io_poll_add(req);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005932 break;
5933 case IORING_OP_POLL_REMOVE:
Jens Axboeb76da702019-11-20 13:05:32 -07005934 ret = io_poll_remove(req);
5935 break;
5936 case IORING_OP_SYNC_FILE_RANGE:
Pavel Begunkov014db002020-03-03 21:33:12 +03005937 ret = io_sync_file_range(req, force_nonblock);
Jens Axboeb76da702019-11-20 13:05:32 -07005938 break;
5939 case IORING_OP_SENDMSG:
Pavel Begunkov062d04d2020-10-10 18:34:12 +01005940 ret = io_sendmsg(req, force_nonblock, cs);
5941 break;
Jens Axboefddafac2020-01-04 20:19:44 -07005942 case IORING_OP_SEND:
Pavel Begunkov062d04d2020-10-10 18:34:12 +01005943 ret = io_send(req, force_nonblock, cs);
Jens Axboeb76da702019-11-20 13:05:32 -07005944 break;
5945 case IORING_OP_RECVMSG:
Pavel Begunkov062d04d2020-10-10 18:34:12 +01005946 ret = io_recvmsg(req, force_nonblock, cs);
5947 break;
Jens Axboefddafac2020-01-04 20:19:44 -07005948 case IORING_OP_RECV:
Pavel Begunkov062d04d2020-10-10 18:34:12 +01005949 ret = io_recv(req, force_nonblock, cs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005950 break;
5951 case IORING_OP_TIMEOUT:
5952 ret = io_timeout(req);
5953 break;
5954 case IORING_OP_TIMEOUT_REMOVE:
5955 ret = io_timeout_remove(req);
5956 break;
5957 case IORING_OP_ACCEPT:
Jens Axboe229a7b62020-06-22 10:13:11 -06005958 ret = io_accept(req, force_nonblock, cs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005959 break;
5960 case IORING_OP_CONNECT:
Jens Axboe229a7b62020-06-22 10:13:11 -06005961 ret = io_connect(req, force_nonblock, cs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005962 break;
5963 case IORING_OP_ASYNC_CANCEL:
Pavel Begunkov014db002020-03-03 21:33:12 +03005964 ret = io_async_cancel(req);
Jens Axboe2b188cc2019-01-07 10:46:33 -07005965 break;
Jens Axboed63d1b52019-12-10 10:38:56 -07005966 case IORING_OP_FALLOCATE:
Pavel Begunkov014db002020-03-03 21:33:12 +03005967 ret = io_fallocate(req, force_nonblock);
Jens Axboed63d1b52019-12-10 10:38:56 -07005968 break;
Jens Axboe15b71ab2019-12-11 11:20:36 -07005969 case IORING_OP_OPENAT:
Pavel Begunkov014db002020-03-03 21:33:12 +03005970 ret = io_openat(req, force_nonblock);
Jens Axboe15b71ab2019-12-11 11:20:36 -07005971 break;
Jens Axboeb5dba592019-12-11 14:02:38 -07005972 case IORING_OP_CLOSE:
Jens Axboe229a7b62020-06-22 10:13:11 -06005973 ret = io_close(req, force_nonblock, cs);
Jens Axboeb5dba592019-12-11 14:02:38 -07005974 break;
Jens Axboe05f3fb32019-12-09 11:22:50 -07005975 case IORING_OP_FILES_UPDATE:
Jens Axboe229a7b62020-06-22 10:13:11 -06005976 ret = io_files_update(req, force_nonblock, cs);
Jens Axboe05f3fb32019-12-09 11:22:50 -07005977 break;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07005978 case IORING_OP_STATX:
Pavel Begunkov014db002020-03-03 21:33:12 +03005979 ret = io_statx(req, force_nonblock);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07005980 break;
Jens Axboe4840e412019-12-25 22:03:45 -07005981 case IORING_OP_FADVISE:
Pavel Begunkov014db002020-03-03 21:33:12 +03005982 ret = io_fadvise(req, force_nonblock);
Jens Axboe4840e412019-12-25 22:03:45 -07005983 break;
Jens Axboec1ca7572019-12-25 22:18:28 -07005984 case IORING_OP_MADVISE:
Pavel Begunkov014db002020-03-03 21:33:12 +03005985 ret = io_madvise(req, force_nonblock);
Jens Axboec1ca7572019-12-25 22:18:28 -07005986 break;
Jens Axboecebdb982020-01-08 17:59:24 -07005987 case IORING_OP_OPENAT2:
Pavel Begunkov014db002020-03-03 21:33:12 +03005988 ret = io_openat2(req, force_nonblock);
Jens Axboecebdb982020-01-08 17:59:24 -07005989 break;
Jens Axboe3e4827b2020-01-08 15:18:09 -07005990 case IORING_OP_EPOLL_CTL:
Jens Axboe229a7b62020-06-22 10:13:11 -06005991 ret = io_epoll_ctl(req, force_nonblock, cs);
Jens Axboe3e4827b2020-01-08 15:18:09 -07005992 break;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03005993 case IORING_OP_SPLICE:
Pavel Begunkov014db002020-03-03 21:33:12 +03005994 ret = io_splice(req, force_nonblock);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03005995 break;
Jens Axboeddf0322d2020-02-23 16:41:33 -07005996 case IORING_OP_PROVIDE_BUFFERS:
Jens Axboe229a7b62020-06-22 10:13:11 -06005997 ret = io_provide_buffers(req, force_nonblock, cs);
Jens Axboeddf0322d2020-02-23 16:41:33 -07005998 break;
Jens Axboe067524e2020-03-02 16:32:28 -07005999 case IORING_OP_REMOVE_BUFFERS:
Jens Axboe229a7b62020-06-22 10:13:11 -06006000 ret = io_remove_buffers(req, force_nonblock, cs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006001 break;
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03006002 case IORING_OP_TEE:
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03006003 ret = io_tee(req, force_nonblock);
6004 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006005 default:
6006 ret = -EINVAL;
6007 break;
Jens Axboe31b51512019-01-18 22:56:34 -07006008 }
6009
6010 if (ret)
Jens Axboeedafcce2019-01-09 09:16:05 -07006011 return ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006012
Jens Axboeb5325762020-05-19 21:20:27 -06006013 /* If the op doesn't have a file, we're not polling for it */
6014 if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) {
Jens Axboe11ba8202020-01-15 21:51:17 -07006015 const bool in_async = io_wq_current_is_worker();
6016
Jens Axboe11ba8202020-01-15 21:51:17 -07006017 /* workqueue context doesn't hold uring_lock, grab it now */
6018 if (in_async)
6019 mutex_lock(&ctx->uring_lock);
6020
Jens Axboe2b188cc2019-01-07 10:46:33 -07006021 io_iopoll_req_issued(req);
Jens Axboe11ba8202020-01-15 21:51:17 -07006022
6023 if (in_async)
6024 mutex_unlock(&ctx->uring_lock);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006025 }
6026
6027 return 0;
6028}
6029
Pavel Begunkovf4db7182020-06-25 18:20:54 +03006030static struct io_wq_work *io_wq_submit_work(struct io_wq_work *work)
Pavel Begunkovd4c81f32020-06-08 21:08:19 +03006031{
Jens Axboe2b188cc2019-01-07 10:46:33 -07006032 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Pavel Begunkov6df1db62020-07-03 22:15:06 +03006033 struct io_kiocb *timeout;
Jens Axboe561fb042019-10-24 07:25:42 -06006034 int ret = 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006035
Pavel Begunkov6df1db62020-07-03 22:15:06 +03006036 timeout = io_prep_linked_timeout(req);
6037 if (timeout)
6038 io_queue_linked_timeout(timeout);
Pavel Begunkovd4c81f32020-06-08 21:08:19 +03006039
Jens Axboe0c9d5cc2019-12-11 19:29:43 -07006040 /* if NO_CANCEL is set, we must still run the work */
6041 if ((work->flags & (IO_WQ_WORK_CANCEL|IO_WQ_WORK_NO_CANCEL)) ==
6042 IO_WQ_WORK_CANCEL) {
Jens Axboe561fb042019-10-24 07:25:42 -06006043 ret = -ECANCELED;
Jens Axboe0c9d5cc2019-12-11 19:29:43 -07006044 }
Jens Axboe31b51512019-01-18 22:56:34 -07006045
Jens Axboe561fb042019-10-24 07:25:42 -06006046 if (!ret) {
Jens Axboe561fb042019-10-24 07:25:42 -06006047 do {
Pavel Begunkovc1379e22020-09-30 22:57:56 +03006048 ret = io_issue_sqe(req, false, NULL);
Jens Axboe561fb042019-10-24 07:25:42 -06006049 /*
6050 * We can get EAGAIN for polled IO even though we're
6051 * forcing a sync submission from here, since we can't
6052 * wait for request slots on the block side.
6053 */
6054 if (ret != -EAGAIN)
6055 break;
6056 cond_resched();
6057 } while (1);
6058 }
Jens Axboe31b51512019-01-18 22:56:34 -07006059
Jens Axboe561fb042019-10-24 07:25:42 -06006060 if (ret) {
Jens Axboe4e88d6e2019-12-07 20:59:47 -07006061 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06006062 io_req_complete(req, ret);
Jens Axboeedafcce2019-01-09 09:16:05 -07006063 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07006064
Pavel Begunkovf4db7182020-06-25 18:20:54 +03006065 return io_steal_work(req);
Jens Axboe31b51512019-01-18 22:56:34 -07006066}
Jens Axboe2b188cc2019-01-07 10:46:33 -07006067
Jens Axboe65e19f52019-10-26 07:20:21 -06006068static inline struct file *io_file_from_index(struct io_ring_ctx *ctx,
6069 int index)
Jens Axboe09bb8392019-03-13 12:39:28 -06006070{
Jens Axboe65e19f52019-10-26 07:20:21 -06006071 struct fixed_file_table *table;
6072
Jens Axboe05f3fb32019-12-09 11:22:50 -07006073 table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT];
Xiaoming Ni84695082020-05-11 19:25:43 +08006074 return table->files[index & IORING_FILE_TABLE_MASK];
Jens Axboe65e19f52019-10-26 07:20:21 -06006075}
6076
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006077static struct file *io_file_get(struct io_submit_state *state,
6078 struct io_kiocb *req, int fd, bool fixed)
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006079{
6080 struct io_ring_ctx *ctx = req->ctx;
6081 struct file *file;
6082
6083 if (fixed) {
Pavel Begunkov479f5172020-10-10 18:34:07 +01006084 if (unlikely((unsigned int)fd >= ctx->nr_user_files))
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006085 return NULL;
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006086 fd = array_index_nospec(fd, ctx->nr_user_files);
6087 file = io_file_from_index(ctx, fd);
Jens Axboefd2206e2020-06-02 16:40:47 -06006088 if (file) {
Pavel Begunkovb2e96852020-10-10 18:34:16 +01006089 req->fixed_file_refs = &ctx->file_data->node->refs;
Jens Axboefd2206e2020-06-02 16:40:47 -06006090 percpu_ref_get(req->fixed_file_refs);
6091 }
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006092 } else {
6093 trace_io_uring_file_get(ctx, fd);
6094 file = __io_file_get(state, fd);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006095 }
6096
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006097 return file;
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006098}
6099
Jens Axboe3529d8c2019-12-19 18:24:38 -07006100static int io_req_set_file(struct io_submit_state *state, struct io_kiocb *req,
Jens Axboe63ff8222020-05-07 14:56:15 -06006101 int fd)
Jens Axboe09bb8392019-03-13 12:39:28 -06006102{
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006103 bool fixed;
Jens Axboe09bb8392019-03-13 12:39:28 -06006104
Jens Axboe63ff8222020-05-07 14:56:15 -06006105 fixed = (req->flags & REQ_F_FIXED_FILE) != 0;
Pavel Begunkov0cdaf762020-05-17 14:13:40 +03006106 if (unlikely(!fixed && io_async_submit(req->ctx)))
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006107 return -EBADF;
Jens Axboe09bb8392019-03-13 12:39:28 -06006108
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006109 req->file = io_file_get(state, req, fd, fixed);
6110 if (req->file || io_op_defs[req->opcode].needs_file_no_error)
Jens Axboef86cd202020-01-29 13:46:44 -07006111 return 0;
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006112 return -EBADF;
Pavel Begunkovf56040b2020-07-23 20:25:21 +03006113}
6114
Jens Axboe2665abf2019-11-05 12:40:47 -07006115static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer)
6116{
Jens Axboead8a48a2019-11-15 08:49:11 -07006117 struct io_timeout_data *data = container_of(timer,
6118 struct io_timeout_data, timer);
6119 struct io_kiocb *req = data->req;
Jens Axboe2665abf2019-11-05 12:40:47 -07006120 struct io_ring_ctx *ctx = req->ctx;
6121 struct io_kiocb *prev = NULL;
6122 unsigned long flags;
Jens Axboe2665abf2019-11-05 12:40:47 -07006123
6124 spin_lock_irqsave(&ctx->completion_lock, flags);
6125
6126 /*
6127 * We don't expect the list to be empty, that will only happen if we
6128 * race with the completion of the linked work.
6129 */
Pavel Begunkov44932332019-12-05 16:16:35 +03006130 if (!list_empty(&req->link_list)) {
6131 prev = list_entry(req->link_list.prev, struct io_kiocb,
6132 link_list);
Pavel Begunkov900fad42020-10-19 16:39:16 +01006133 if (refcount_inc_not_zero(&prev->refs))
Pavel Begunkov44932332019-12-05 16:16:35 +03006134 list_del_init(&req->link_list);
Pavel Begunkov900fad42020-10-19 16:39:16 +01006135 else
Jens Axboe76a46e02019-11-10 23:34:16 -07006136 prev = NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006137 }
6138
6139 spin_unlock_irqrestore(&ctx->completion_lock, flags);
6140
6141 if (prev) {
Jens Axboe4e88d6e2019-12-07 20:59:47 -07006142 req_set_fail_links(prev);
Pavel Begunkov014db002020-03-03 21:33:12 +03006143 io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME);
Jens Axboe76a46e02019-11-10 23:34:16 -07006144 io_put_req(prev);
Jens Axboe47f46762019-11-09 17:43:02 -07006145 } else {
Jens Axboee1e16092020-06-22 09:17:17 -06006146 io_req_complete(req, -ETIME);
Jens Axboe2665abf2019-11-05 12:40:47 -07006147 }
Jens Axboe2665abf2019-11-05 12:40:47 -07006148 return HRTIMER_NORESTART;
6149}
6150
Jens Axboe7271ef32020-08-10 09:55:22 -06006151static void __io_queue_linked_timeout(struct io_kiocb *req)
Jens Axboe2665abf2019-11-05 12:40:47 -07006152{
Jens Axboe76a46e02019-11-10 23:34:16 -07006153 /*
6154 * If the list is now empty, then our linked request finished before
6155 * we got a chance to setup the timer
6156 */
Pavel Begunkov44932332019-12-05 16:16:35 +03006157 if (!list_empty(&req->link_list)) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07006158 struct io_timeout_data *data = req->async_data;
Jens Axboe94ae5e72019-11-14 19:39:52 -07006159
Jens Axboead8a48a2019-11-15 08:49:11 -07006160 data->timer.function = io_link_timeout_fn;
6161 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts),
6162 data->mode);
Jens Axboe2665abf2019-11-05 12:40:47 -07006163 }
Jens Axboe7271ef32020-08-10 09:55:22 -06006164}
6165
6166static void io_queue_linked_timeout(struct io_kiocb *req)
6167{
6168 struct io_ring_ctx *ctx = req->ctx;
6169
6170 spin_lock_irq(&ctx->completion_lock);
6171 __io_queue_linked_timeout(req);
Jens Axboe76a46e02019-11-10 23:34:16 -07006172 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe2665abf2019-11-05 12:40:47 -07006173
Jens Axboe2665abf2019-11-05 12:40:47 -07006174 /* drop submission reference */
Jens Axboe76a46e02019-11-10 23:34:16 -07006175 io_put_req(req);
Jens Axboe2665abf2019-11-05 12:40:47 -07006176}
6177
Jens Axboead8a48a2019-11-15 08:49:11 -07006178static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req)
Jens Axboe2665abf2019-11-05 12:40:47 -07006179{
6180 struct io_kiocb *nxt;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006181
Pavel Begunkovdea3b492020-04-12 02:05:04 +03006182 if (!(req->flags & REQ_F_LINK_HEAD))
Jens Axboe2665abf2019-11-05 12:40:47 -07006183 return NULL;
Pavel Begunkov6df1db62020-07-03 22:15:06 +03006184 if (req->flags & REQ_F_LINK_TIMEOUT)
Jens Axboed7718a92020-02-14 22:23:12 -07006185 return NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006186
Pavel Begunkov44932332019-12-05 16:16:35 +03006187 nxt = list_first_entry_or_null(&req->link_list, struct io_kiocb,
6188 link_list);
Jens Axboed625c6e2019-12-17 19:53:05 -07006189 if (!nxt || nxt->opcode != IORING_OP_LINK_TIMEOUT)
Jens Axboe76a46e02019-11-10 23:34:16 -07006190 return NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006191
Pavel Begunkov900fad42020-10-19 16:39:16 +01006192 nxt->flags |= REQ_F_LTIMEOUT_ACTIVE;
Jens Axboe76a46e02019-11-10 23:34:16 -07006193 req->flags |= REQ_F_LINK_TIMEOUT;
Jens Axboe76a46e02019-11-10 23:34:16 -07006194 return nxt;
Jens Axboe2665abf2019-11-05 12:40:47 -07006195}
6196
Pavel Begunkovc1379e22020-09-30 22:57:56 +03006197static void __io_queue_sqe(struct io_kiocb *req, struct io_comp_state *cs)
Jens Axboe2b188cc2019-01-07 10:46:33 -07006198{
Jens Axboe4a0a7a12019-12-09 20:01:01 -07006199 struct io_kiocb *linked_timeout;
Jens Axboe193155c2020-02-22 23:22:19 -07006200 const struct cred *old_creds = NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006201 int ret;
6202
Jens Axboe4a0a7a12019-12-09 20:01:01 -07006203again:
6204 linked_timeout = io_prep_linked_timeout(req);
6205
Pavel Begunkov2e5aa6c2020-10-18 10:17:37 +01006206 if ((req->flags & REQ_F_WORK_INITIALIZED) &&
6207 (req->work.flags & IO_WQ_WORK_CREDS) &&
Jens Axboe98447d62020-10-14 10:48:51 -06006208 req->work.identity->creds != current_cred()) {
Jens Axboe193155c2020-02-22 23:22:19 -07006209 if (old_creds)
6210 revert_creds(old_creds);
Jens Axboe98447d62020-10-14 10:48:51 -06006211 if (old_creds == req->work.identity->creds)
Jens Axboe193155c2020-02-22 23:22:19 -07006212 old_creds = NULL; /* restored original creds */
6213 else
Jens Axboe98447d62020-10-14 10:48:51 -06006214 old_creds = override_creds(req->work.identity->creds);
Jens Axboe193155c2020-02-22 23:22:19 -07006215 }
6216
Pavel Begunkovc1379e22020-09-30 22:57:56 +03006217 ret = io_issue_sqe(req, true, cs);
Jens Axboe491381ce2019-10-17 09:20:46 -06006218
6219 /*
6220 * We async punt it if the file wasn't marked NOWAIT, or if the file
6221 * doesn't support non-blocking read/write attempts
6222 */
Pavel Begunkov24c74672020-06-21 13:09:51 +03006223 if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) {
Pavel Begunkovf063c542020-07-25 14:41:59 +03006224 if (!io_arm_poll_handler(req)) {
Pavel Begunkovf063c542020-07-25 14:41:59 +03006225 /*
6226 * Queued up for async execution, worker will release
6227 * submit reference when the iocb is actually submitted.
6228 */
6229 io_queue_async_work(req);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006230 }
Pavel Begunkovbbad27b2019-11-19 23:32:47 +03006231
Pavel Begunkovf063c542020-07-25 14:41:59 +03006232 if (linked_timeout)
6233 io_queue_linked_timeout(linked_timeout);
Pavel Begunkov0d63c142020-10-22 16:47:18 +01006234 } else if (likely(!ret)) {
6235 /* drop submission reference */
6236 req = io_put_req_find_next(req);
6237 if (linked_timeout)
6238 io_queue_linked_timeout(linked_timeout);
Jens Axboee65ef562019-03-12 10:16:44 -06006239
Pavel Begunkov0d63c142020-10-22 16:47:18 +01006240 if (req) {
6241 if (!(req->flags & REQ_F_FORCE_ASYNC))
6242 goto again;
6243 io_queue_async_work(req);
6244 }
6245 } else {
Pavel Begunkov652532a2020-07-03 22:15:07 +03006246 /* un-prep timeout, so it'll be killed as any other linked */
6247 req->flags &= ~REQ_F_LINK_TIMEOUT;
Jens Axboe4e88d6e2019-12-07 20:59:47 -07006248 req_set_fail_links(req);
Jens Axboee65ef562019-03-12 10:16:44 -06006249 io_put_req(req);
Pavel Begunkov652532a2020-07-03 22:15:07 +03006250 io_req_complete(req, ret);
Jens Axboe9e645e112019-05-10 16:07:28 -06006251 }
Pavel Begunkov652532a2020-07-03 22:15:07 +03006252
Jens Axboe193155c2020-02-22 23:22:19 -07006253 if (old_creds)
6254 revert_creds(old_creds);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006255}
6256
Jens Axboef13fad72020-06-22 09:34:30 -06006257static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe,
6258 struct io_comp_state *cs)
Jackie Liu4fe2c962019-09-09 20:50:40 +08006259{
6260 int ret;
6261
Jens Axboe3529d8c2019-12-19 18:24:38 -07006262 ret = io_req_defer(req, sqe);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006263 if (ret) {
6264 if (ret != -EIOCBQUEUED) {
Pavel Begunkov11185912020-01-22 23:09:35 +03006265fail_req:
Jens Axboe4e88d6e2019-12-07 20:59:47 -07006266 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06006267 io_put_req(req);
6268 io_req_complete(req, ret);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006269 }
Pavel Begunkov25508782019-12-30 21:24:47 +03006270 } else if (req->flags & REQ_F_FORCE_ASYNC) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07006271 if (!req->async_data) {
Pavel Begunkovbd2ab182020-05-17 14:02:12 +03006272 ret = io_req_defer_prep(req, sqe);
Pavel Begunkov327d6d92020-07-15 12:46:51 +03006273 if (unlikely(ret))
Pavel Begunkovbd2ab182020-05-17 14:02:12 +03006274 goto fail_req;
6275 }
Jens Axboece35a472019-12-17 08:04:44 -07006276 io_queue_async_work(req);
6277 } else {
Pavel Begunkovc1379e22020-09-30 22:57:56 +03006278 if (sqe) {
6279 ret = io_req_prep(req, sqe);
6280 if (unlikely(ret))
6281 goto fail_req;
6282 }
6283 __io_queue_sqe(req, cs);
Jens Axboece35a472019-12-17 08:04:44 -07006284 }
Jackie Liu4fe2c962019-09-09 20:50:40 +08006285}
6286
Jens Axboef13fad72020-06-22 09:34:30 -06006287static inline void io_queue_link_head(struct io_kiocb *req,
6288 struct io_comp_state *cs)
Jackie Liu4fe2c962019-09-09 20:50:40 +08006289{
Jens Axboe94ae5e72019-11-14 19:39:52 -07006290 if (unlikely(req->flags & REQ_F_FAIL_LINK)) {
Jens Axboee1e16092020-06-22 09:17:17 -06006291 io_put_req(req);
6292 io_req_complete(req, -ECANCELED);
Pavel Begunkov1b4a51b2019-11-21 11:54:28 +03006293 } else
Jens Axboef13fad72020-06-22 09:34:30 -06006294 io_queue_sqe(req, NULL, cs);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006295}
6296
Pavel Begunkov1d4240c2020-04-12 02:05:03 +03006297static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe,
Jens Axboef13fad72020-06-22 09:34:30 -06006298 struct io_kiocb **link, struct io_comp_state *cs)
Jens Axboe9e645e112019-05-10 16:07:28 -06006299{
Jackie Liua197f662019-11-08 08:09:12 -07006300 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006301 int ret;
Jens Axboe9e645e112019-05-10 16:07:28 -06006302
Jens Axboe9e645e112019-05-10 16:07:28 -06006303 /*
6304 * If we already have a head request, queue this one for async
6305 * submittal once the head completes. If we don't have a head but
6306 * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be
6307 * submitted sync once the chain is complete. If none of those
6308 * conditions are true (normal request), then just queue it.
6309 */
6310 if (*link) {
Pavel Begunkov9d763772019-12-17 02:22:07 +03006311 struct io_kiocb *head = *link;
Jens Axboe9e645e112019-05-10 16:07:28 -06006312
Pavel Begunkov8cdf2192020-01-25 00:40:24 +03006313 /*
6314 * Taking sequential execution of a link, draining both sides
6315 * of the link also fullfils IOSQE_IO_DRAIN semantics for all
6316 * requests in the link. So, it drains the head and the
6317 * next after the link request. The last one is done via
6318 * drain_next flag to persist the effect across calls.
6319 */
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006320 if (req->flags & REQ_F_IO_DRAIN) {
Pavel Begunkov711be032020-01-17 03:57:59 +03006321 head->flags |= REQ_F_IO_DRAIN;
6322 ctx->drain_next = 1;
6323 }
Jens Axboe3529d8c2019-12-19 18:24:38 -07006324 ret = io_req_defer_prep(req, sqe);
Pavel Begunkov327d6d92020-07-15 12:46:51 +03006325 if (unlikely(ret)) {
Jens Axboe4e88d6e2019-12-07 20:59:47 -07006326 /* fail even hard links since we don't submit */
Pavel Begunkov9d763772019-12-17 02:22:07 +03006327 head->flags |= REQ_F_FAIL_LINK;
Pavel Begunkov1d4240c2020-04-12 02:05:03 +03006328 return ret;
Jens Axboe2d283902019-12-04 11:08:05 -07006329 }
Pavel Begunkov9d763772019-12-17 02:22:07 +03006330 trace_io_uring_link(ctx, req, head);
6331 list_add_tail(&req->link_list, &head->link_list);
Jens Axboe9e645e112019-05-10 16:07:28 -06006332
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006333 /* last request of a link, enqueue the link */
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006334 if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) {
Jens Axboef13fad72020-06-22 09:34:30 -06006335 io_queue_link_head(head, cs);
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006336 *link = NULL;
6337 }
Jens Axboe9e645e112019-05-10 16:07:28 -06006338 } else {
Pavel Begunkov711be032020-01-17 03:57:59 +03006339 if (unlikely(ctx->drain_next)) {
6340 req->flags |= REQ_F_IO_DRAIN;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006341 ctx->drain_next = 0;
Pavel Begunkov711be032020-01-17 03:57:59 +03006342 }
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006343 if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) {
Pavel Begunkovdea3b492020-04-12 02:05:04 +03006344 req->flags |= REQ_F_LINK_HEAD;
Pavel Begunkov711be032020-01-17 03:57:59 +03006345 INIT_LIST_HEAD(&req->link_list);
Pavel Begunkovf1d96a82020-03-13 22:29:14 +03006346
Pavel Begunkov711be032020-01-17 03:57:59 +03006347 ret = io_req_defer_prep(req, sqe);
Pavel Begunkov327d6d92020-07-15 12:46:51 +03006348 if (unlikely(ret))
Pavel Begunkov711be032020-01-17 03:57:59 +03006349 req->flags |= REQ_F_FAIL_LINK;
6350 *link = req;
6351 } else {
Jens Axboef13fad72020-06-22 09:34:30 -06006352 io_queue_sqe(req, sqe, cs);
Pavel Begunkov711be032020-01-17 03:57:59 +03006353 }
Jens Axboe9e645e112019-05-10 16:07:28 -06006354 }
Pavel Begunkov2e6e1fd2019-12-05 16:15:45 +03006355
Pavel Begunkov1d4240c2020-04-12 02:05:03 +03006356 return 0;
Jens Axboe9e645e112019-05-10 16:07:28 -06006357}
6358
Jens Axboe9a56a232019-01-09 09:06:50 -07006359/*
6360 * Batched submission is done, ensure local IO is flushed out.
6361 */
6362static void io_submit_state_end(struct io_submit_state *state)
6363{
Jens Axboef13fad72020-06-22 09:34:30 -06006364 if (!list_empty(&state->comp.list))
6365 io_submit_flush_completions(&state->comp);
Jens Axboe9a56a232019-01-09 09:06:50 -07006366 blk_finish_plug(&state->plug);
Pavel Begunkov9f13c352020-05-17 14:13:41 +03006367 io_state_file_put(state);
Jens Axboe2579f912019-01-09 09:10:43 -07006368 if (state->free_reqs)
Pavel Begunkov6c8a3132020-02-01 03:58:00 +03006369 kmem_cache_free_bulk(req_cachep, state->free_reqs, state->reqs);
Jens Axboe9a56a232019-01-09 09:06:50 -07006370}
6371
6372/*
6373 * Start submission side cache.
6374 */
6375static void io_submit_state_start(struct io_submit_state *state,
Jens Axboe013538b2020-06-22 09:29:15 -06006376 struct io_ring_ctx *ctx, unsigned int max_ios)
Jens Axboe9a56a232019-01-09 09:06:50 -07006377{
6378 blk_start_plug(&state->plug);
Jens Axboe013538b2020-06-22 09:29:15 -06006379 state->comp.nr = 0;
6380 INIT_LIST_HEAD(&state->comp.list);
6381 state->comp.ctx = ctx;
Jens Axboe2579f912019-01-09 09:10:43 -07006382 state->free_reqs = 0;
Jens Axboe9a56a232019-01-09 09:06:50 -07006383 state->file = NULL;
6384 state->ios_left = max_ios;
6385}
6386
Jens Axboe2b188cc2019-01-07 10:46:33 -07006387static void io_commit_sqring(struct io_ring_ctx *ctx)
6388{
Hristo Venev75b28af2019-08-26 17:23:46 +00006389 struct io_rings *rings = ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006390
Pavel Begunkovcaf582c2019-12-30 21:24:46 +03006391 /*
6392 * Ensure any loads from the SQEs are done at this point,
6393 * since once we write the new head, the application could
6394 * write new data to them.
6395 */
6396 smp_store_release(&rings->sq.head, ctx->cached_sq_head);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006397}
6398
6399/*
Jens Axboe3529d8c2019-12-19 18:24:38 -07006400 * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory
Jens Axboe2b188cc2019-01-07 10:46:33 -07006401 * that is mapped by userspace. This means that care needs to be taken to
6402 * ensure that reads are stable, as we cannot rely on userspace always
6403 * being a good citizen. If members of the sqe are validated and then later
6404 * used, it's important that those reads are done through READ_ONCE() to
6405 * prevent a re-load down the line.
6406 */
Pavel Begunkov709b3022020-04-08 08:58:43 +03006407static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07006408{
Hristo Venev75b28af2019-08-26 17:23:46 +00006409 u32 *sq_array = ctx->sq_array;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006410 unsigned head;
6411
6412 /*
6413 * The cached sq head (or cq tail) serves two purposes:
6414 *
6415 * 1) allows us to batch the cost of updating the user visible
6416 * head updates.
6417 * 2) allows the kernel side to track the head on its own, even
6418 * though the application is the one updating it.
6419 */
Pavel Begunkovee7d46d92019-12-30 21:24:45 +03006420 head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]);
Pavel Begunkov709b3022020-04-08 08:58:43 +03006421 if (likely(head < ctx->sq_entries))
6422 return &ctx->sq_sqes[head];
Jens Axboe2b188cc2019-01-07 10:46:33 -07006423
6424 /* drop invalid entries */
Jens Axboe498ccd92019-10-25 10:04:25 -06006425 ctx->cached_sq_dropped++;
Pavel Begunkovee7d46d92019-12-30 21:24:45 +03006426 WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped);
Pavel Begunkov709b3022020-04-08 08:58:43 +03006427 return NULL;
6428}
6429
6430static inline void io_consume_sqe(struct io_ring_ctx *ctx)
6431{
6432 ctx->cached_sq_head++;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006433}
6434
Stefano Garzarella21b55db2020-08-27 16:58:30 +02006435/*
6436 * Check SQE restrictions (opcode and flags).
6437 *
6438 * Returns 'true' if SQE is allowed, 'false' otherwise.
6439 */
6440static inline bool io_check_restriction(struct io_ring_ctx *ctx,
6441 struct io_kiocb *req,
6442 unsigned int sqe_flags)
6443{
6444 if (!ctx->restricted)
6445 return true;
6446
6447 if (!test_bit(req->opcode, ctx->restrictions.sqe_op))
6448 return false;
6449
6450 if ((sqe_flags & ctx->restrictions.sqe_flags_required) !=
6451 ctx->restrictions.sqe_flags_required)
6452 return false;
6453
6454 if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed |
6455 ctx->restrictions.sqe_flags_required))
6456 return false;
6457
6458 return true;
6459}
6460
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006461#define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \
6462 IOSQE_IO_HARDLINK | IOSQE_ASYNC | \
6463 IOSQE_BUFFER_SELECT)
6464
6465static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req,
6466 const struct io_uring_sqe *sqe,
Pavel Begunkov0cdaf762020-05-17 14:13:40 +03006467 struct io_submit_state *state)
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006468{
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006469 unsigned int sqe_flags;
Pavel Begunkov71b547c2020-10-10 18:34:09 +01006470 int id, ret;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006471
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006472 req->opcode = READ_ONCE(sqe->opcode);
6473 req->user_data = READ_ONCE(sqe->user_data);
Jens Axboee8c2bc12020-08-15 18:44:09 -07006474 req->async_data = NULL;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006475 req->file = NULL;
6476 req->ctx = ctx;
6477 req->flags = 0;
6478 /* one is dropped after submission, the other at completion */
6479 refcount_set(&req->refs, 2);
Pavel Begunkov4dd28242020-06-15 10:33:13 +03006480 req->task = current;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006481 req->result = 0;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006482
6483 if (unlikely(req->opcode >= IORING_OP_LAST))
6484 return -EINVAL;
6485
Jens Axboe9d8426a2020-06-16 18:42:49 -06006486 if (unlikely(io_sq_thread_acquire_mm(ctx, req)))
6487 return -EFAULT;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006488
6489 sqe_flags = READ_ONCE(sqe->flags);
6490 /* enforce forwards compatibility on users */
6491 if (unlikely(sqe_flags & ~SQE_VALID_FLAGS))
6492 return -EINVAL;
6493
Stefano Garzarella21b55db2020-08-27 16:58:30 +02006494 if (unlikely(!io_check_restriction(ctx, req, sqe_flags)))
6495 return -EACCES;
6496
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006497 if ((sqe_flags & IOSQE_BUFFER_SELECT) &&
6498 !io_op_defs[req->opcode].buffer_select)
6499 return -EOPNOTSUPP;
6500
6501 id = READ_ONCE(sqe->personality);
6502 if (id) {
Jens Axboe1e6fa522020-10-15 08:46:24 -06006503 struct io_identity *iod;
6504
Jens Axboe1e6fa522020-10-15 08:46:24 -06006505 iod = idr_find(&ctx->personality_idr, id);
6506 if (unlikely(!iod))
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006507 return -EINVAL;
Jens Axboe1e6fa522020-10-15 08:46:24 -06006508 refcount_inc(&iod->count);
Pavel Begunkovec99ca62020-10-18 10:17:38 +01006509
6510 __io_req_init_async(req);
Jens Axboe1e6fa522020-10-15 08:46:24 -06006511 get_cred(iod->creds);
6512 req->work.identity = iod;
Jens Axboedfead8a2020-10-14 10:12:37 -06006513 req->work.flags |= IO_WQ_WORK_CREDS;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006514 }
6515
6516 /* same numerical values with corresponding REQ_F_*, safe to copy */
Pavel Begunkovc11368a52020-05-17 14:13:42 +03006517 req->flags |= sqe_flags;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006518
Jens Axboe63ff8222020-05-07 14:56:15 -06006519 if (!io_op_defs[req->opcode].needs_file)
6520 return 0;
6521
Pavel Begunkov71b547c2020-10-10 18:34:09 +01006522 ret = io_req_set_file(state, req, READ_ONCE(sqe->fd));
6523 state->ios_left--;
6524 return ret;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006525}
6526
Jens Axboe0f212202020-09-13 13:09:39 -06006527static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006528{
Jens Axboeac8691c2020-06-01 08:30:41 -06006529 struct io_submit_state state;
Jens Axboe9e645e112019-05-10 16:07:28 -06006530 struct io_kiocb *link = NULL;
Jens Axboe9e645e112019-05-10 16:07:28 -06006531 int i, submitted = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006532
Jens Axboec4a2ed72019-11-21 21:01:26 -07006533 /* if we have a backlog and couldn't flush it all, return BUSY */
Jens Axboead3eb2c2019-12-18 17:12:20 -07006534 if (test_bit(0, &ctx->sq_check_overflow)) {
6535 if (!list_empty(&ctx->cq_overflow_list) &&
Jens Axboee6c8aa92020-09-28 13:10:13 -06006536 !io_cqring_overflow_flush(ctx, false, NULL, NULL))
Jens Axboead3eb2c2019-12-18 17:12:20 -07006537 return -EBUSY;
6538 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07006539
Pavel Begunkovee7d46d92019-12-30 21:24:45 +03006540 /* make sure SQ entry isn't read before tail */
6541 nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx));
Pavel Begunkov9ef4f122019-12-30 21:24:44 +03006542
Pavel Begunkov2b85edf2019-12-28 14:13:03 +03006543 if (!percpu_ref_tryget_many(&ctx->refs, nr))
6544 return -EAGAIN;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006545
Jens Axboed8a6df12020-10-15 16:24:45 -06006546 percpu_counter_add(&current->io_uring->inflight, nr);
Jens Axboefaf7b512020-10-07 12:48:53 -06006547 refcount_add(nr, &current->usage);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006548
Jens Axboe6c271ce2019-01-10 11:22:30 -07006549 io_submit_state_start(&state, ctx, nr);
Pavel Begunkovb14cca02020-01-17 04:45:59 +03006550
Jens Axboe6c271ce2019-01-10 11:22:30 -07006551 for (i = 0; i < nr; i++) {
Jens Axboe3529d8c2019-12-19 18:24:38 -07006552 const struct io_uring_sqe *sqe;
Pavel Begunkov196be952019-11-07 01:41:06 +03006553 struct io_kiocb *req;
Pavel Begunkov1cb1edb2020-02-06 21:16:09 +03006554 int err;
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03006555
Pavel Begunkovb1e50e52020-04-08 08:58:44 +03006556 sqe = io_get_sqe(ctx);
6557 if (unlikely(!sqe)) {
6558 io_consume_sqe(ctx);
6559 break;
6560 }
Jens Axboeac8691c2020-06-01 08:30:41 -06006561 req = io_alloc_req(ctx, &state);
Pavel Begunkov196be952019-11-07 01:41:06 +03006562 if (unlikely(!req)) {
6563 if (!submitted)
6564 submitted = -EAGAIN;
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03006565 break;
Jens Axboe9e645e112019-05-10 16:07:28 -06006566 }
Pavel Begunkov709b3022020-04-08 08:58:43 +03006567 io_consume_sqe(ctx);
Jens Axboed3656342019-12-18 09:50:26 -07006568 /* will complete beyond this point, count as submitted */
6569 submitted++;
6570
Pavel Begunkov692d8362020-10-10 18:34:13 +01006571 err = io_init_req(ctx, req, sqe, &state);
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006572 if (unlikely(err)) {
Pavel Begunkov1cb1edb2020-02-06 21:16:09 +03006573fail_req:
Jens Axboee1e16092020-06-22 09:17:17 -06006574 io_put_req(req);
6575 io_req_complete(req, err);
Jens Axboed3656342019-12-18 09:50:26 -07006576 break;
6577 }
6578
Jens Axboe354420f2020-01-08 18:55:15 -07006579 trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data,
Pavel Begunkov0cdaf762020-05-17 14:13:40 +03006580 true, io_async_submit(ctx));
Jens Axboef13fad72020-06-22 09:34:30 -06006581 err = io_submit_sqe(req, sqe, &link, &state.comp);
Pavel Begunkov1d4240c2020-04-12 02:05:03 +03006582 if (err)
6583 goto fail_req;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006584 }
6585
Pavel Begunkov9466f432020-01-25 22:34:01 +03006586 if (unlikely(submitted != nr)) {
6587 int ref_used = (submitted == -EAGAIN) ? 0 : submitted;
Jens Axboed8a6df12020-10-15 16:24:45 -06006588 struct io_uring_task *tctx = current->io_uring;
6589 int unused = nr - ref_used;
Pavel Begunkov9466f432020-01-25 22:34:01 +03006590
Jens Axboed8a6df12020-10-15 16:24:45 -06006591 percpu_ref_put_many(&ctx->refs, unused);
6592 percpu_counter_sub(&tctx->inflight, unused);
6593 put_task_struct_many(current, unused);
Pavel Begunkov9466f432020-01-25 22:34:01 +03006594 }
Jens Axboe9e645e112019-05-10 16:07:28 -06006595 if (link)
Jens Axboef13fad72020-06-22 09:34:30 -06006596 io_queue_link_head(link, &state.comp);
Jens Axboeac8691c2020-06-01 08:30:41 -06006597 io_submit_state_end(&state);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006598
Pavel Begunkovae9428c2019-11-06 00:22:14 +03006599 /* Commit SQ ring head once we've consumed and submitted all SQEs */
6600 io_commit_sqring(ctx);
6601
Jens Axboe6c271ce2019-01-10 11:22:30 -07006602 return submitted;
6603}
6604
Xiaoguang Wang23b36282020-07-23 20:57:24 +08006605static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx)
6606{
6607 /* Tell userspace we may need a wakeup call */
6608 spin_lock_irq(&ctx->completion_lock);
6609 ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP;
6610 spin_unlock_irq(&ctx->completion_lock);
6611}
6612
6613static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx)
6614{
6615 spin_lock_irq(&ctx->completion_lock);
6616 ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP;
6617 spin_unlock_irq(&ctx->completion_lock);
6618}
6619
Jens Axboe3f0e64d2020-09-02 12:42:47 -06006620static int io_sq_wake_function(struct wait_queue_entry *wqe, unsigned mode,
6621 int sync, void *key)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006622{
Jens Axboe3f0e64d2020-09-02 12:42:47 -06006623 struct io_ring_ctx *ctx = container_of(wqe, struct io_ring_ctx, sqo_wait_entry);
6624 int ret;
6625
6626 ret = autoremove_wake_function(wqe, mode, sync, key);
6627 if (ret) {
6628 unsigned long flags;
6629
6630 spin_lock_irqsave(&ctx->completion_lock, flags);
6631 ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP;
6632 spin_unlock_irqrestore(&ctx->completion_lock, flags);
6633 }
6634 return ret;
6635}
6636
Jens Axboec8d1ba52020-09-14 11:07:26 -06006637enum sq_ret {
6638 SQT_IDLE = 1,
6639 SQT_SPIN = 2,
6640 SQT_DID_WORK = 4,
6641};
6642
6643static enum sq_ret __io_sq_thread(struct io_ring_ctx *ctx,
Jens Axboee95eee22020-09-08 09:11:32 -06006644 unsigned long start_jiffies, bool cap_entries)
Jens Axboec8d1ba52020-09-14 11:07:26 -06006645{
6646 unsigned long timeout = start_jiffies + ctx->sq_thread_idle;
Jens Axboe534ca6d2020-09-02 13:52:19 -06006647 struct io_sq_data *sqd = ctx->sq_data;
Jens Axboec8d1ba52020-09-14 11:07:26 -06006648 unsigned int to_submit;
Xiaoguang Wangbdcd3ea2020-02-25 22:12:08 +08006649 int ret = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006650
Jens Axboec8d1ba52020-09-14 11:07:26 -06006651again:
6652 if (!list_empty(&ctx->iopoll_list)) {
6653 unsigned nr_events = 0;
Jackie Liua4c0b3d2019-07-08 13:41:12 +08006654
Jens Axboec8d1ba52020-09-14 11:07:26 -06006655 mutex_lock(&ctx->uring_lock);
6656 if (!list_empty(&ctx->iopoll_list) && !need_resched())
6657 io_do_iopoll(ctx, &nr_events, 0);
6658 mutex_unlock(&ctx->uring_lock);
6659 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07006660
Jens Axboec8d1ba52020-09-14 11:07:26 -06006661 to_submit = io_sqring_entries(ctx);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006662
Jens Axboec8d1ba52020-09-14 11:07:26 -06006663 /*
6664 * If submit got -EBUSY, flag us as needing the application
6665 * to enter the kernel to reap and flush events.
6666 */
6667 if (!to_submit || ret == -EBUSY || need_resched()) {
6668 /*
6669 * Drop cur_mm before scheduling, we can't hold it for
6670 * long periods (or over schedule()). Do this before
6671 * adding ourselves to the waitqueue, as the unuse/drop
6672 * may sleep.
6673 */
6674 io_sq_thread_drop_mm();
Jens Axboe6c271ce2019-01-10 11:22:30 -07006675
Jens Axboec8d1ba52020-09-14 11:07:26 -06006676 /*
6677 * We're polling. If we're within the defined idle
6678 * period, then let us spin without work before going
6679 * to sleep. The exception is if we got EBUSY doing
6680 * more IO, we should wait for the application to
6681 * reap events and wake us up.
6682 */
6683 if (!list_empty(&ctx->iopoll_list) || need_resched() ||
6684 (!time_after(jiffies, timeout) && ret != -EBUSY &&
6685 !percpu_ref_is_dying(&ctx->refs)))
6686 return SQT_SPIN;
6687
Jens Axboe534ca6d2020-09-02 13:52:19 -06006688 prepare_to_wait(&sqd->wait, &ctx->sqo_wait_entry,
Jens Axboec8d1ba52020-09-14 11:07:26 -06006689 TASK_INTERRUPTIBLE);
6690
6691 /*
6692 * While doing polled IO, before going to sleep, we need
6693 * to check if there are new reqs added to iopoll_list,
6694 * it is because reqs may have been punted to io worker
6695 * and will be added to iopoll_list later, hence check
6696 * the iopoll_list again.
6697 */
6698 if ((ctx->flags & IORING_SETUP_IOPOLL) &&
6699 !list_empty_careful(&ctx->iopoll_list)) {
Jens Axboe534ca6d2020-09-02 13:52:19 -06006700 finish_wait(&sqd->wait, &ctx->sqo_wait_entry);
Jens Axboec8d1ba52020-09-14 11:07:26 -06006701 goto again;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006702 }
6703
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03006704 to_submit = io_sqring_entries(ctx);
Jens Axboec8d1ba52020-09-14 11:07:26 -06006705 if (!to_submit || ret == -EBUSY)
6706 return SQT_IDLE;
6707 }
6708
Jens Axboe534ca6d2020-09-02 13:52:19 -06006709 finish_wait(&sqd->wait, &ctx->sqo_wait_entry);
Jens Axboec8d1ba52020-09-14 11:07:26 -06006710 io_ring_clear_wakeup_flag(ctx);
6711
Jens Axboee95eee22020-09-08 09:11:32 -06006712 /* if we're handling multiple rings, cap submit size for fairness */
6713 if (cap_entries && to_submit > 8)
6714 to_submit = 8;
6715
Jens Axboec8d1ba52020-09-14 11:07:26 -06006716 mutex_lock(&ctx->uring_lock);
6717 if (likely(!percpu_ref_is_dying(&ctx->refs)))
6718 ret = io_submit_sqes(ctx, to_submit);
6719 mutex_unlock(&ctx->uring_lock);
Jens Axboe90554202020-09-03 12:12:41 -06006720
6721 if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait))
6722 wake_up(&ctx->sqo_sq_wait);
6723
Jens Axboec8d1ba52020-09-14 11:07:26 -06006724 return SQT_DID_WORK;
6725}
6726
Jens Axboe69fb2132020-09-14 11:16:23 -06006727static void io_sqd_init_new(struct io_sq_data *sqd)
6728{
6729 struct io_ring_ctx *ctx;
6730
6731 while (!list_empty(&sqd->ctx_new_list)) {
6732 ctx = list_first_entry(&sqd->ctx_new_list, struct io_ring_ctx, sqd_list);
6733 init_wait(&ctx->sqo_wait_entry);
6734 ctx->sqo_wait_entry.func = io_sq_wake_function;
6735 list_move_tail(&ctx->sqd_list, &sqd->ctx_list);
6736 complete(&ctx->sq_thread_comp);
6737 }
6738}
6739
Jens Axboe6c271ce2019-01-10 11:22:30 -07006740static int io_sq_thread(void *data)
6741{
Dennis Zhou91d8f512020-09-16 13:41:05 -07006742 struct cgroup_subsys_state *cur_css = NULL;
Jens Axboe69fb2132020-09-14 11:16:23 -06006743 const struct cred *old_cred = NULL;
6744 struct io_sq_data *sqd = data;
6745 struct io_ring_ctx *ctx;
Jens Axboec8d1ba52020-09-14 11:07:26 -06006746 unsigned long start_jiffies;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006747
Jens Axboec8d1ba52020-09-14 11:07:26 -06006748 start_jiffies = jiffies;
Jens Axboe69fb2132020-09-14 11:16:23 -06006749 while (!kthread_should_stop()) {
6750 enum sq_ret ret = 0;
Jens Axboee95eee22020-09-08 09:11:32 -06006751 bool cap_entries;
Jens Axboec1edbf52019-11-10 16:56:04 -07006752
6753 /*
Jens Axboe69fb2132020-09-14 11:16:23 -06006754 * Any changes to the sqd lists are synchronized through the
6755 * kthread parking. This synchronizes the thread vs users,
6756 * the users are synchronized on the sqd->ctx_lock.
Jens Axboec1edbf52019-11-10 16:56:04 -07006757 */
Jens Axboe69fb2132020-09-14 11:16:23 -06006758 if (kthread_should_park())
6759 kthread_parkme();
6760
6761 if (unlikely(!list_empty(&sqd->ctx_new_list)))
6762 io_sqd_init_new(sqd);
6763
Jens Axboee95eee22020-09-08 09:11:32 -06006764 cap_entries = !list_is_singular(&sqd->ctx_list);
6765
Jens Axboe69fb2132020-09-14 11:16:23 -06006766 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
6767 if (current->cred != ctx->creds) {
6768 if (old_cred)
6769 revert_creds(old_cred);
6770 old_cred = override_creds(ctx->creds);
6771 }
Dennis Zhou91d8f512020-09-16 13:41:05 -07006772 io_sq_thread_associate_blkcg(ctx, &cur_css);
Jens Axboe4ea33a92020-10-15 13:46:44 -06006773#ifdef CONFIG_AUDIT
6774 current->loginuid = ctx->loginuid;
6775 current->sessionid = ctx->sessionid;
6776#endif
Jens Axboe69fb2132020-09-14 11:16:23 -06006777
Jens Axboee95eee22020-09-08 09:11:32 -06006778 ret |= __io_sq_thread(ctx, start_jiffies, cap_entries);
Jens Axboe69fb2132020-09-14 11:16:23 -06006779
Jens Axboe4349f302020-07-09 15:07:01 -06006780 io_sq_thread_drop_mm();
Jens Axboe6c271ce2019-01-10 11:22:30 -07006781 }
6782
Jens Axboe69fb2132020-09-14 11:16:23 -06006783 if (ret & SQT_SPIN) {
Jens Axboec8d1ba52020-09-14 11:07:26 -06006784 io_run_task_work();
6785 cond_resched();
Jens Axboe69fb2132020-09-14 11:16:23 -06006786 } else if (ret == SQT_IDLE) {
6787 if (kthread_should_park())
6788 continue;
6789 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6790 io_ring_set_wakeup_flag(ctx);
6791 schedule();
6792 start_jiffies = jiffies;
6793 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
6794 io_ring_clear_wakeup_flag(ctx);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006795 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07006796 }
6797
Jens Axboe4c6e2772020-07-01 11:29:10 -06006798 io_run_task_work();
Jens Axboeb41e9852020-02-17 09:52:41 -07006799
Dennis Zhou91d8f512020-09-16 13:41:05 -07006800 if (cur_css)
6801 io_sq_thread_unassociate_blkcg();
Jens Axboe69fb2132020-09-14 11:16:23 -06006802 if (old_cred)
6803 revert_creds(old_cred);
Jens Axboe06058632019-04-13 09:26:03 -06006804
Roman Penyaev2bbcd6d2019-05-16 10:53:57 +02006805 kthread_parkme();
Jens Axboe06058632019-04-13 09:26:03 -06006806
Jens Axboe6c271ce2019-01-10 11:22:30 -07006807 return 0;
6808}
6809
Jens Axboebda52162019-09-24 13:47:15 -06006810struct io_wait_queue {
6811 struct wait_queue_entry wq;
6812 struct io_ring_ctx *ctx;
6813 unsigned to_wait;
6814 unsigned nr_timeouts;
6815};
6816
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07006817static inline bool io_should_wake(struct io_wait_queue *iowq, bool noflush)
Jens Axboebda52162019-09-24 13:47:15 -06006818{
6819 struct io_ring_ctx *ctx = iowq->ctx;
6820
6821 /*
Brian Gianforcarod195a662019-12-13 03:09:50 -08006822 * Wake up if we have enough events, or if a timeout occurred since we
Jens Axboebda52162019-09-24 13:47:15 -06006823 * started waiting. For timeouts, we always want to return to userspace,
6824 * regardless of event count.
6825 */
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07006826 return io_cqring_events(ctx, noflush) >= iowq->to_wait ||
Jens Axboebda52162019-09-24 13:47:15 -06006827 atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts;
6828}
6829
6830static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode,
6831 int wake_flags, void *key)
6832{
6833 struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue,
6834 wq);
6835
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07006836 /* use noflush == true, as we can't safely rely on locking context */
6837 if (!io_should_wake(iowq, true))
Jens Axboebda52162019-09-24 13:47:15 -06006838 return -1;
6839
6840 return autoremove_wake_function(curr, mode, wake_flags, key);
6841}
6842
Jens Axboeaf9c1a42020-09-24 13:32:18 -06006843static int io_run_task_work_sig(void)
6844{
6845 if (io_run_task_work())
6846 return 1;
6847 if (!signal_pending(current))
6848 return 0;
6849 if (current->jobctl & JOBCTL_TASK_WORK) {
6850 spin_lock_irq(&current->sighand->siglock);
6851 current->jobctl &= ~JOBCTL_TASK_WORK;
6852 recalc_sigpending();
6853 spin_unlock_irq(&current->sighand->siglock);
6854 return 1;
6855 }
6856 return -EINTR;
6857}
6858
Jens Axboe2b188cc2019-01-07 10:46:33 -07006859/*
6860 * Wait until events become available, if we don't already have some. The
6861 * application must reap them itself, as they reside on the shared cq ring.
6862 */
6863static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
6864 const sigset_t __user *sig, size_t sigsz)
6865{
Jens Axboebda52162019-09-24 13:47:15 -06006866 struct io_wait_queue iowq = {
6867 .wq = {
6868 .private = current,
6869 .func = io_wake_function,
6870 .entry = LIST_HEAD_INIT(iowq.wq.entry),
6871 },
6872 .ctx = ctx,
6873 .to_wait = min_events,
6874 };
Hristo Venev75b28af2019-08-26 17:23:46 +00006875 struct io_rings *rings = ctx->rings;
Jackie Liue9ffa5c2019-10-29 11:16:42 +08006876 int ret = 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006877
Jens Axboeb41e9852020-02-17 09:52:41 -07006878 do {
6879 if (io_cqring_events(ctx, false) >= min_events)
6880 return 0;
Jens Axboe4c6e2772020-07-01 11:29:10 -06006881 if (!io_run_task_work())
Jens Axboeb41e9852020-02-17 09:52:41 -07006882 break;
Jens Axboeb41e9852020-02-17 09:52:41 -07006883 } while (1);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006884
6885 if (sig) {
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01006886#ifdef CONFIG_COMPAT
6887 if (in_compat_syscall())
6888 ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig,
Oleg Nesterovb7724342019-07-16 16:29:53 -07006889 sigsz);
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01006890 else
6891#endif
Oleg Nesterovb7724342019-07-16 16:29:53 -07006892 ret = set_user_sigmask(sig, sigsz);
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01006893
Jens Axboe2b188cc2019-01-07 10:46:33 -07006894 if (ret)
6895 return ret;
6896 }
6897
Jens Axboebda52162019-09-24 13:47:15 -06006898 iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts);
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02006899 trace_io_uring_cqring_wait(ctx, min_events);
Jens Axboebda52162019-09-24 13:47:15 -06006900 do {
6901 prepare_to_wait_exclusive(&ctx->wait, &iowq.wq,
6902 TASK_INTERRUPTIBLE);
Jens Axboece593a62020-06-30 12:39:05 -06006903 /* make sure we run task_work before checking for signals */
Jens Axboeaf9c1a42020-09-24 13:32:18 -06006904 ret = io_run_task_work_sig();
6905 if (ret > 0)
Jens Axboe4c6e2772020-07-01 11:29:10 -06006906 continue;
Jens Axboeaf9c1a42020-09-24 13:32:18 -06006907 else if (ret < 0)
Jens Axboece593a62020-06-30 12:39:05 -06006908 break;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07006909 if (io_should_wake(&iowq, false))
Jens Axboebda52162019-09-24 13:47:15 -06006910 break;
6911 schedule();
Jens Axboebda52162019-09-24 13:47:15 -06006912 } while (1);
6913 finish_wait(&ctx->wait, &iowq.wq);
6914
Jens Axboeb7db41c2020-07-04 08:55:50 -06006915 restore_saved_sigmask_unless(ret == -EINTR);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006916
Hristo Venev75b28af2019-08-26 17:23:46 +00006917 return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006918}
6919
Jens Axboe6b063142019-01-10 22:13:58 -07006920static void __io_sqe_files_unregister(struct io_ring_ctx *ctx)
6921{
6922#if defined(CONFIG_UNIX)
6923 if (ctx->ring_sock) {
6924 struct sock *sock = ctx->ring_sock->sk;
6925 struct sk_buff *skb;
6926
6927 while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL)
6928 kfree_skb(skb);
6929 }
6930#else
6931 int i;
6932
Jens Axboe65e19f52019-10-26 07:20:21 -06006933 for (i = 0; i < ctx->nr_user_files; i++) {
6934 struct file *file;
6935
6936 file = io_file_from_index(ctx, i);
6937 if (file)
6938 fput(file);
6939 }
Jens Axboe6b063142019-01-10 22:13:58 -07006940#endif
6941}
6942
Jens Axboe05f3fb32019-12-09 11:22:50 -07006943static void io_file_ref_kill(struct percpu_ref *ref)
6944{
6945 struct fixed_file_data *data;
6946
6947 data = container_of(ref, struct fixed_file_data, refs);
6948 complete(&data->done);
6949}
6950
Jens Axboe6b063142019-01-10 22:13:58 -07006951static int io_sqe_files_unregister(struct io_ring_ctx *ctx)
6952{
Jens Axboe05f3fb32019-12-09 11:22:50 -07006953 struct fixed_file_data *data = ctx->file_data;
Xiaoguang Wang05589552020-03-31 14:05:18 +08006954 struct fixed_file_ref_node *ref_node = NULL;
Jens Axboe65e19f52019-10-26 07:20:21 -06006955 unsigned nr_tables, i;
6956
Jens Axboe05f3fb32019-12-09 11:22:50 -07006957 if (!data)
Jens Axboe6b063142019-01-10 22:13:58 -07006958 return -ENXIO;
6959
Jens Axboe6a4d07c2020-05-15 14:30:38 -06006960 spin_lock(&data->lock);
Xiaoguang Wang05589552020-03-31 14:05:18 +08006961 if (!list_empty(&data->ref_list))
6962 ref_node = list_first_entry(&data->ref_list,
6963 struct fixed_file_ref_node, node);
Jens Axboe6a4d07c2020-05-15 14:30:38 -06006964 spin_unlock(&data->lock);
Xiaoguang Wang05589552020-03-31 14:05:18 +08006965 if (ref_node)
6966 percpu_ref_kill(&ref_node->refs);
6967
6968 percpu_ref_kill(&data->refs);
6969
6970 /* wait for all refs nodes to complete */
Jens Axboe4a38aed22020-05-14 17:21:15 -06006971 flush_delayed_work(&ctx->file_put_work);
Jens Axboe2faf8522020-02-04 19:54:55 -07006972 wait_for_completion(&data->done);
Jens Axboe05f3fb32019-12-09 11:22:50 -07006973
Jens Axboe6b063142019-01-10 22:13:58 -07006974 __io_sqe_files_unregister(ctx);
Jens Axboe65e19f52019-10-26 07:20:21 -06006975 nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE);
6976 for (i = 0; i < nr_tables; i++)
Jens Axboe05f3fb32019-12-09 11:22:50 -07006977 kfree(data->table[i].files);
6978 kfree(data->table);
Xiaoguang Wang05589552020-03-31 14:05:18 +08006979 percpu_ref_exit(&data->refs);
6980 kfree(data);
Jens Axboe05f3fb32019-12-09 11:22:50 -07006981 ctx->file_data = NULL;
Jens Axboe6b063142019-01-10 22:13:58 -07006982 ctx->nr_user_files = 0;
6983 return 0;
6984}
6985
Jens Axboe534ca6d2020-09-02 13:52:19 -06006986static void io_put_sq_data(struct io_sq_data *sqd)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006987{
Jens Axboe534ca6d2020-09-02 13:52:19 -06006988 if (refcount_dec_and_test(&sqd->refs)) {
Roman Penyaev2bbcd6d2019-05-16 10:53:57 +02006989 /*
6990 * The park is a bit of a work-around, without it we get
6991 * warning spews on shutdown with SQPOLL set and affinity
6992 * set to a single CPU.
6993 */
Jens Axboe534ca6d2020-09-02 13:52:19 -06006994 if (sqd->thread) {
6995 kthread_park(sqd->thread);
6996 kthread_stop(sqd->thread);
6997 }
6998
6999 kfree(sqd);
7000 }
7001}
7002
Jens Axboeaa061652020-09-02 14:50:27 -06007003static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p)
7004{
7005 struct io_ring_ctx *ctx_attach;
7006 struct io_sq_data *sqd;
7007 struct fd f;
7008
7009 f = fdget(p->wq_fd);
7010 if (!f.file)
7011 return ERR_PTR(-ENXIO);
7012 if (f.file->f_op != &io_uring_fops) {
7013 fdput(f);
7014 return ERR_PTR(-EINVAL);
7015 }
7016
7017 ctx_attach = f.file->private_data;
7018 sqd = ctx_attach->sq_data;
7019 if (!sqd) {
7020 fdput(f);
7021 return ERR_PTR(-EINVAL);
7022 }
7023
7024 refcount_inc(&sqd->refs);
7025 fdput(f);
7026 return sqd;
7027}
7028
Jens Axboe534ca6d2020-09-02 13:52:19 -06007029static struct io_sq_data *io_get_sq_data(struct io_uring_params *p)
7030{
7031 struct io_sq_data *sqd;
7032
Jens Axboeaa061652020-09-02 14:50:27 -06007033 if (p->flags & IORING_SETUP_ATTACH_WQ)
7034 return io_attach_sq_data(p);
7035
Jens Axboe534ca6d2020-09-02 13:52:19 -06007036 sqd = kzalloc(sizeof(*sqd), GFP_KERNEL);
7037 if (!sqd)
7038 return ERR_PTR(-ENOMEM);
7039
7040 refcount_set(&sqd->refs, 1);
Jens Axboe69fb2132020-09-14 11:16:23 -06007041 INIT_LIST_HEAD(&sqd->ctx_list);
7042 INIT_LIST_HEAD(&sqd->ctx_new_list);
7043 mutex_init(&sqd->ctx_lock);
7044 mutex_init(&sqd->lock);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007045 init_waitqueue_head(&sqd->wait);
7046 return sqd;
7047}
7048
Jens Axboe69fb2132020-09-14 11:16:23 -06007049static void io_sq_thread_unpark(struct io_sq_data *sqd)
7050 __releases(&sqd->lock)
7051{
7052 if (!sqd->thread)
7053 return;
7054 kthread_unpark(sqd->thread);
7055 mutex_unlock(&sqd->lock);
7056}
7057
7058static void io_sq_thread_park(struct io_sq_data *sqd)
7059 __acquires(&sqd->lock)
7060{
7061 if (!sqd->thread)
7062 return;
7063 mutex_lock(&sqd->lock);
7064 kthread_park(sqd->thread);
7065}
7066
Jens Axboe534ca6d2020-09-02 13:52:19 -06007067static void io_sq_thread_stop(struct io_ring_ctx *ctx)
7068{
7069 struct io_sq_data *sqd = ctx->sq_data;
7070
7071 if (sqd) {
7072 if (sqd->thread) {
7073 /*
7074 * We may arrive here from the error branch in
7075 * io_sq_offload_create() where the kthread is created
7076 * without being waked up, thus wake it up now to make
7077 * sure the wait will complete.
7078 */
7079 wake_up_process(sqd->thread);
7080 wait_for_completion(&ctx->sq_thread_comp);
Jens Axboe69fb2132020-09-14 11:16:23 -06007081
7082 io_sq_thread_park(sqd);
7083 }
7084
7085 mutex_lock(&sqd->ctx_lock);
7086 list_del(&ctx->sqd_list);
7087 mutex_unlock(&sqd->ctx_lock);
7088
7089 if (sqd->thread) {
7090 finish_wait(&sqd->wait, &ctx->sqo_wait_entry);
7091 io_sq_thread_unpark(sqd);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007092 }
7093
7094 io_put_sq_data(sqd);
7095 ctx->sq_data = NULL;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007096 }
7097}
7098
Jens Axboe6b063142019-01-10 22:13:58 -07007099static void io_finish_async(struct io_ring_ctx *ctx)
7100{
Jens Axboe6c271ce2019-01-10 11:22:30 -07007101 io_sq_thread_stop(ctx);
7102
Jens Axboe561fb042019-10-24 07:25:42 -06007103 if (ctx->io_wq) {
7104 io_wq_destroy(ctx->io_wq);
7105 ctx->io_wq = NULL;
Jens Axboe6b063142019-01-10 22:13:58 -07007106 }
7107}
7108
7109#if defined(CONFIG_UNIX)
Jens Axboe6b063142019-01-10 22:13:58 -07007110/*
7111 * Ensure the UNIX gc is aware of our file set, so we are certain that
7112 * the io_uring can be safely unregistered on process exit, even if we have
7113 * loops in the file referencing.
7114 */
7115static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset)
7116{
7117 struct sock *sk = ctx->ring_sock->sk;
7118 struct scm_fp_list *fpl;
7119 struct sk_buff *skb;
Jens Axboe08a451732019-10-03 08:11:03 -06007120 int i, nr_files;
Jens Axboe6b063142019-01-10 22:13:58 -07007121
Jens Axboe6b063142019-01-10 22:13:58 -07007122 fpl = kzalloc(sizeof(*fpl), GFP_KERNEL);
7123 if (!fpl)
7124 return -ENOMEM;
7125
7126 skb = alloc_skb(0, GFP_KERNEL);
7127 if (!skb) {
7128 kfree(fpl);
7129 return -ENOMEM;
7130 }
7131
7132 skb->sk = sk;
Jens Axboe6b063142019-01-10 22:13:58 -07007133
Jens Axboe08a451732019-10-03 08:11:03 -06007134 nr_files = 0;
Jens Axboe6b063142019-01-10 22:13:58 -07007135 fpl->user = get_uid(ctx->user);
7136 for (i = 0; i < nr; i++) {
Jens Axboe65e19f52019-10-26 07:20:21 -06007137 struct file *file = io_file_from_index(ctx, i + offset);
7138
7139 if (!file)
Jens Axboe08a451732019-10-03 08:11:03 -06007140 continue;
Jens Axboe65e19f52019-10-26 07:20:21 -06007141 fpl->fp[nr_files] = get_file(file);
Jens Axboe08a451732019-10-03 08:11:03 -06007142 unix_inflight(fpl->user, fpl->fp[nr_files]);
7143 nr_files++;
Jens Axboe6b063142019-01-10 22:13:58 -07007144 }
7145
Jens Axboe08a451732019-10-03 08:11:03 -06007146 if (nr_files) {
7147 fpl->max = SCM_MAX_FD;
7148 fpl->count = nr_files;
7149 UNIXCB(skb).fp = fpl;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007150 skb->destructor = unix_destruct_scm;
Jens Axboe08a451732019-10-03 08:11:03 -06007151 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
7152 skb_queue_head(&sk->sk_receive_queue, skb);
Jens Axboe6b063142019-01-10 22:13:58 -07007153
Jens Axboe08a451732019-10-03 08:11:03 -06007154 for (i = 0; i < nr_files; i++)
7155 fput(fpl->fp[i]);
7156 } else {
7157 kfree_skb(skb);
7158 kfree(fpl);
7159 }
Jens Axboe6b063142019-01-10 22:13:58 -07007160
7161 return 0;
7162}
7163
7164/*
7165 * If UNIX sockets are enabled, fd passing can cause a reference cycle which
7166 * causes regular reference counting to break down. We rely on the UNIX
7167 * garbage collection to take care of this problem for us.
7168 */
7169static int io_sqe_files_scm(struct io_ring_ctx *ctx)
7170{
7171 unsigned left, total;
7172 int ret = 0;
7173
7174 total = 0;
7175 left = ctx->nr_user_files;
7176 while (left) {
7177 unsigned this_files = min_t(unsigned, left, SCM_MAX_FD);
Jens Axboe6b063142019-01-10 22:13:58 -07007178
7179 ret = __io_sqe_files_scm(ctx, this_files, total);
7180 if (ret)
7181 break;
7182 left -= this_files;
7183 total += this_files;
7184 }
7185
7186 if (!ret)
7187 return 0;
7188
7189 while (total < ctx->nr_user_files) {
Jens Axboe65e19f52019-10-26 07:20:21 -06007190 struct file *file = io_file_from_index(ctx, total);
7191
7192 if (file)
7193 fput(file);
Jens Axboe6b063142019-01-10 22:13:58 -07007194 total++;
7195 }
7196
7197 return ret;
7198}
7199#else
7200static int io_sqe_files_scm(struct io_ring_ctx *ctx)
7201{
7202 return 0;
7203}
7204#endif
7205
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007206static int io_sqe_alloc_file_tables(struct fixed_file_data *file_data,
7207 unsigned nr_tables, unsigned nr_files)
Jens Axboe65e19f52019-10-26 07:20:21 -06007208{
7209 int i;
7210
7211 for (i = 0; i < nr_tables; i++) {
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007212 struct fixed_file_table *table = &file_data->table[i];
Jens Axboe65e19f52019-10-26 07:20:21 -06007213 unsigned this_files;
7214
7215 this_files = min(nr_files, IORING_MAX_FILES_TABLE);
7216 table->files = kcalloc(this_files, sizeof(struct file *),
7217 GFP_KERNEL);
7218 if (!table->files)
7219 break;
7220 nr_files -= this_files;
7221 }
7222
7223 if (i == nr_tables)
7224 return 0;
7225
7226 for (i = 0; i < nr_tables; i++) {
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007227 struct fixed_file_table *table = &file_data->table[i];
Jens Axboe65e19f52019-10-26 07:20:21 -06007228 kfree(table->files);
7229 }
7230 return 1;
7231}
7232
Jens Axboe05f3fb32019-12-09 11:22:50 -07007233static void io_ring_file_put(struct io_ring_ctx *ctx, struct file *file)
Jens Axboec3a31e62019-10-03 13:59:56 -06007234{
7235#if defined(CONFIG_UNIX)
Jens Axboec3a31e62019-10-03 13:59:56 -06007236 struct sock *sock = ctx->ring_sock->sk;
7237 struct sk_buff_head list, *head = &sock->sk_receive_queue;
7238 struct sk_buff *skb;
7239 int i;
7240
7241 __skb_queue_head_init(&list);
7242
7243 /*
7244 * Find the skb that holds this file in its SCM_RIGHTS. When found,
7245 * remove this entry and rearrange the file array.
7246 */
7247 skb = skb_dequeue(head);
7248 while (skb) {
7249 struct scm_fp_list *fp;
7250
7251 fp = UNIXCB(skb).fp;
7252 for (i = 0; i < fp->count; i++) {
7253 int left;
7254
7255 if (fp->fp[i] != file)
7256 continue;
7257
7258 unix_notinflight(fp->user, fp->fp[i]);
7259 left = fp->count - 1 - i;
7260 if (left) {
7261 memmove(&fp->fp[i], &fp->fp[i + 1],
7262 left * sizeof(struct file *));
7263 }
7264 fp->count--;
7265 if (!fp->count) {
7266 kfree_skb(skb);
7267 skb = NULL;
7268 } else {
7269 __skb_queue_tail(&list, skb);
7270 }
7271 fput(file);
7272 file = NULL;
7273 break;
7274 }
7275
7276 if (!file)
7277 break;
7278
7279 __skb_queue_tail(&list, skb);
7280
7281 skb = skb_dequeue(head);
7282 }
7283
7284 if (skb_peek(&list)) {
7285 spin_lock_irq(&head->lock);
7286 while ((skb = __skb_dequeue(&list)) != NULL)
7287 __skb_queue_tail(head, skb);
7288 spin_unlock_irq(&head->lock);
7289 }
7290#else
Jens Axboe05f3fb32019-12-09 11:22:50 -07007291 fput(file);
Jens Axboec3a31e62019-10-03 13:59:56 -06007292#endif
7293}
7294
Jens Axboe05f3fb32019-12-09 11:22:50 -07007295struct io_file_put {
Xiaoguang Wang05589552020-03-31 14:05:18 +08007296 struct list_head list;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007297 struct file *file;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007298};
7299
Jens Axboe4a38aed22020-05-14 17:21:15 -06007300static void __io_file_put_work(struct fixed_file_ref_node *ref_node)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007301{
Jens Axboe4a38aed22020-05-14 17:21:15 -06007302 struct fixed_file_data *file_data = ref_node->file_data;
7303 struct io_ring_ctx *ctx = file_data->ctx;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007304 struct io_file_put *pfile, *tmp;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007305
7306 list_for_each_entry_safe(pfile, tmp, &ref_node->file_list, list) {
Jens Axboe6a4d07c2020-05-15 14:30:38 -06007307 list_del(&pfile->list);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007308 io_ring_file_put(ctx, pfile->file);
7309 kfree(pfile);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007310 }
7311
Jens Axboe6a4d07c2020-05-15 14:30:38 -06007312 spin_lock(&file_data->lock);
7313 list_del(&ref_node->node);
7314 spin_unlock(&file_data->lock);
Jens Axboe2faf8522020-02-04 19:54:55 -07007315
Xiaoguang Wang05589552020-03-31 14:05:18 +08007316 percpu_ref_exit(&ref_node->refs);
7317 kfree(ref_node);
7318 percpu_ref_put(&file_data->refs);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007319}
7320
Jens Axboe4a38aed22020-05-14 17:21:15 -06007321static void io_file_put_work(struct work_struct *work)
7322{
7323 struct io_ring_ctx *ctx;
7324 struct llist_node *node;
7325
7326 ctx = container_of(work, struct io_ring_ctx, file_put_work.work);
7327 node = llist_del_all(&ctx->file_put_llist);
7328
7329 while (node) {
7330 struct fixed_file_ref_node *ref_node;
7331 struct llist_node *next = node->next;
7332
7333 ref_node = llist_entry(node, struct fixed_file_ref_node, llist);
7334 __io_file_put_work(ref_node);
7335 node = next;
7336 }
7337}
7338
Jens Axboe05f3fb32019-12-09 11:22:50 -07007339static void io_file_data_ref_zero(struct percpu_ref *ref)
7340{
Xiaoguang Wang05589552020-03-31 14:05:18 +08007341 struct fixed_file_ref_node *ref_node;
Jens Axboe4a38aed22020-05-14 17:21:15 -06007342 struct io_ring_ctx *ctx;
7343 bool first_add;
7344 int delay = HZ;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007345
Xiaoguang Wang05589552020-03-31 14:05:18 +08007346 ref_node = container_of(ref, struct fixed_file_ref_node, refs);
Jens Axboe4a38aed22020-05-14 17:21:15 -06007347 ctx = ref_node->file_data->ctx;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007348
Jens Axboe4a38aed22020-05-14 17:21:15 -06007349 if (percpu_ref_is_dying(&ctx->file_data->refs))
7350 delay = 0;
7351
7352 first_add = llist_add(&ref_node->llist, &ctx->file_put_llist);
7353 if (!delay)
7354 mod_delayed_work(system_wq, &ctx->file_put_work, 0);
7355 else if (first_add)
7356 queue_delayed_work(system_wq, &ctx->file_put_work, delay);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007357}
7358
7359static struct fixed_file_ref_node *alloc_fixed_file_ref_node(
7360 struct io_ring_ctx *ctx)
7361{
7362 struct fixed_file_ref_node *ref_node;
7363
7364 ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL);
7365 if (!ref_node)
7366 return ERR_PTR(-ENOMEM);
7367
7368 if (percpu_ref_init(&ref_node->refs, io_file_data_ref_zero,
7369 0, GFP_KERNEL)) {
7370 kfree(ref_node);
7371 return ERR_PTR(-ENOMEM);
7372 }
7373 INIT_LIST_HEAD(&ref_node->node);
7374 INIT_LIST_HEAD(&ref_node->file_list);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007375 ref_node->file_data = ctx->file_data;
7376 return ref_node;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007377}
7378
7379static void destroy_fixed_file_ref_node(struct fixed_file_ref_node *ref_node)
7380{
7381 percpu_ref_exit(&ref_node->refs);
7382 kfree(ref_node);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007383}
7384
7385static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg,
7386 unsigned nr_args)
7387{
7388 __s32 __user *fds = (__s32 __user *) arg;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007389 unsigned nr_tables, i;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007390 struct file *file;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007391 int fd, ret = -ENOMEM;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007392 struct fixed_file_ref_node *ref_node;
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007393 struct fixed_file_data *file_data;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007394
7395 if (ctx->file_data)
7396 return -EBUSY;
7397 if (!nr_args)
7398 return -EINVAL;
7399 if (nr_args > IORING_MAX_FIXED_FILES)
7400 return -EMFILE;
7401
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007402 file_data = kzalloc(sizeof(*ctx->file_data), GFP_KERNEL);
7403 if (!file_data)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007404 return -ENOMEM;
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007405 file_data->ctx = ctx;
7406 init_completion(&file_data->done);
7407 INIT_LIST_HEAD(&file_data->ref_list);
7408 spin_lock_init(&file_data->lock);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007409
7410 nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE);
Colin Ian King035fbaf2020-10-12 15:03:41 +01007411 file_data->table = kcalloc(nr_tables, sizeof(*file_data->table),
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007412 GFP_KERNEL);
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007413 if (!file_data->table)
7414 goto out_free;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007415
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007416 if (percpu_ref_init(&file_data->refs, io_file_ref_kill,
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007417 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL))
7418 goto out_free;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007419
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007420 if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args))
7421 goto out_ref;
Jens Axboe55cbc252020-10-14 07:35:57 -06007422 ctx->file_data = file_data;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007423
7424 for (i = 0; i < nr_args; i++, ctx->nr_user_files++) {
7425 struct fixed_file_table *table;
7426 unsigned index;
7427
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007428 if (copy_from_user(&fd, &fds[i], sizeof(fd))) {
7429 ret = -EFAULT;
7430 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007431 }
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007432 /* allow sparse sets */
7433 if (fd == -1)
7434 continue;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007435
Jens Axboe05f3fb32019-12-09 11:22:50 -07007436 file = fget(fd);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007437 ret = -EBADF;
7438 if (!file)
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007439 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007440
7441 /*
7442 * Don't allow io_uring instances to be registered. If UNIX
7443 * isn't enabled, then this causes a reference cycle and this
7444 * instance can never get freed. If UNIX is enabled we'll
7445 * handle it just fine, but there's still no point in allowing
7446 * a ring fd as it doesn't support regular read/write anyway.
7447 */
7448 if (file->f_op == &io_uring_fops) {
7449 fput(file);
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007450 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007451 }
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007452 table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT];
7453 index = i & IORING_FILE_TABLE_MASK;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007454 table->files[index] = file;
7455 }
7456
Jens Axboe05f3fb32019-12-09 11:22:50 -07007457 ret = io_sqe_files_scm(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007458 if (ret) {
Jens Axboe05f3fb32019-12-09 11:22:50 -07007459 io_sqe_files_unregister(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007460 return ret;
7461 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07007462
Xiaoguang Wang05589552020-03-31 14:05:18 +08007463 ref_node = alloc_fixed_file_ref_node(ctx);
7464 if (IS_ERR(ref_node)) {
7465 io_sqe_files_unregister(ctx);
7466 return PTR_ERR(ref_node);
7467 }
7468
Pavel Begunkovb2e96852020-10-10 18:34:16 +01007469 file_data->node = ref_node;
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007470 spin_lock(&file_data->lock);
7471 list_add(&ref_node->node, &file_data->ref_list);
7472 spin_unlock(&file_data->lock);
7473 percpu_ref_get(&file_data->refs);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007474 return ret;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007475out_fput:
7476 for (i = 0; i < ctx->nr_user_files; i++) {
7477 file = io_file_from_index(ctx, i);
7478 if (file)
7479 fput(file);
7480 }
7481 for (i = 0; i < nr_tables; i++)
7482 kfree(file_data->table[i].files);
7483 ctx->nr_user_files = 0;
7484out_ref:
7485 percpu_ref_exit(&file_data->refs);
7486out_free:
7487 kfree(file_data->table);
7488 kfree(file_data);
Jens Axboe55cbc252020-10-14 07:35:57 -06007489 ctx->file_data = NULL;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007490 return ret;
7491}
7492
Jens Axboec3a31e62019-10-03 13:59:56 -06007493static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file,
7494 int index)
7495{
7496#if defined(CONFIG_UNIX)
7497 struct sock *sock = ctx->ring_sock->sk;
7498 struct sk_buff_head *head = &sock->sk_receive_queue;
7499 struct sk_buff *skb;
7500
7501 /*
7502 * See if we can merge this file into an existing skb SCM_RIGHTS
7503 * file set. If there's no room, fall back to allocating a new skb
7504 * and filling it in.
7505 */
7506 spin_lock_irq(&head->lock);
7507 skb = skb_peek(head);
7508 if (skb) {
7509 struct scm_fp_list *fpl = UNIXCB(skb).fp;
7510
7511 if (fpl->count < SCM_MAX_FD) {
7512 __skb_unlink(skb, head);
7513 spin_unlock_irq(&head->lock);
7514 fpl->fp[fpl->count] = get_file(file);
7515 unix_inflight(fpl->user, fpl->fp[fpl->count]);
7516 fpl->count++;
7517 spin_lock_irq(&head->lock);
7518 __skb_queue_head(head, skb);
7519 } else {
7520 skb = NULL;
7521 }
7522 }
7523 spin_unlock_irq(&head->lock);
7524
7525 if (skb) {
7526 fput(file);
7527 return 0;
7528 }
7529
7530 return __io_sqe_files_scm(ctx, 1, index);
7531#else
7532 return 0;
7533#endif
7534}
7535
Hillf Dantona5318d32020-03-23 17:47:15 +08007536static int io_queue_file_removal(struct fixed_file_data *data,
Xiaoguang Wang05589552020-03-31 14:05:18 +08007537 struct file *file)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007538{
Hillf Dantona5318d32020-03-23 17:47:15 +08007539 struct io_file_put *pfile;
Pavel Begunkovb2e96852020-10-10 18:34:16 +01007540 struct fixed_file_ref_node *ref_node = data->node;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007541
Jens Axboe05f3fb32019-12-09 11:22:50 -07007542 pfile = kzalloc(sizeof(*pfile), GFP_KERNEL);
Hillf Dantona5318d32020-03-23 17:47:15 +08007543 if (!pfile)
7544 return -ENOMEM;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007545
7546 pfile->file = file;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007547 list_add(&pfile->list, &ref_node->file_list);
7548
Hillf Dantona5318d32020-03-23 17:47:15 +08007549 return 0;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007550}
7551
7552static int __io_sqe_files_update(struct io_ring_ctx *ctx,
7553 struct io_uring_files_update *up,
7554 unsigned nr_args)
7555{
7556 struct fixed_file_data *data = ctx->file_data;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007557 struct fixed_file_ref_node *ref_node;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007558 struct file *file;
Jens Axboec3a31e62019-10-03 13:59:56 -06007559 __s32 __user *fds;
7560 int fd, i, err;
7561 __u32 done;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007562 bool needs_switch = false;
Jens Axboec3a31e62019-10-03 13:59:56 -06007563
Jens Axboe05f3fb32019-12-09 11:22:50 -07007564 if (check_add_overflow(up->offset, nr_args, &done))
Jens Axboec3a31e62019-10-03 13:59:56 -06007565 return -EOVERFLOW;
7566 if (done > ctx->nr_user_files)
7567 return -EINVAL;
7568
Xiaoguang Wang05589552020-03-31 14:05:18 +08007569 ref_node = alloc_fixed_file_ref_node(ctx);
7570 if (IS_ERR(ref_node))
7571 return PTR_ERR(ref_node);
7572
Jens Axboec3a31e62019-10-03 13:59:56 -06007573 done = 0;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007574 fds = u64_to_user_ptr(up->fds);
Jens Axboec3a31e62019-10-03 13:59:56 -06007575 while (nr_args) {
Jens Axboe65e19f52019-10-26 07:20:21 -06007576 struct fixed_file_table *table;
7577 unsigned index;
7578
Jens Axboec3a31e62019-10-03 13:59:56 -06007579 err = 0;
7580 if (copy_from_user(&fd, &fds[done], sizeof(fd))) {
7581 err = -EFAULT;
7582 break;
7583 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07007584 i = array_index_nospec(up->offset, ctx->nr_user_files);
7585 table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT];
Jens Axboe65e19f52019-10-26 07:20:21 -06007586 index = i & IORING_FILE_TABLE_MASK;
7587 if (table->files[index]) {
Jiufei Xue98dfd502020-09-01 13:35:02 +08007588 file = table->files[index];
Hillf Dantona5318d32020-03-23 17:47:15 +08007589 err = io_queue_file_removal(data, file);
7590 if (err)
7591 break;
Jens Axboe65e19f52019-10-26 07:20:21 -06007592 table->files[index] = NULL;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007593 needs_switch = true;
Jens Axboec3a31e62019-10-03 13:59:56 -06007594 }
7595 if (fd != -1) {
Jens Axboec3a31e62019-10-03 13:59:56 -06007596 file = fget(fd);
7597 if (!file) {
7598 err = -EBADF;
7599 break;
7600 }
7601 /*
7602 * Don't allow io_uring instances to be registered. If
7603 * UNIX isn't enabled, then this causes a reference
7604 * cycle and this instance can never get freed. If UNIX
7605 * is enabled we'll handle it just fine, but there's
7606 * still no point in allowing a ring fd as it doesn't
7607 * support regular read/write anyway.
7608 */
7609 if (file->f_op == &io_uring_fops) {
7610 fput(file);
7611 err = -EBADF;
7612 break;
7613 }
Jens Axboe65e19f52019-10-26 07:20:21 -06007614 table->files[index] = file;
Jens Axboec3a31e62019-10-03 13:59:56 -06007615 err = io_sqe_file_register(ctx, file, i);
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007616 if (err) {
Jiufei Xue95d1c8e2020-09-02 17:59:39 +08007617 table->files[index] = NULL;
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007618 fput(file);
Jens Axboec3a31e62019-10-03 13:59:56 -06007619 break;
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00007620 }
Jens Axboec3a31e62019-10-03 13:59:56 -06007621 }
7622 nr_args--;
7623 done++;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007624 up->offset++;
7625 }
7626
Xiaoguang Wang05589552020-03-31 14:05:18 +08007627 if (needs_switch) {
Pavel Begunkovb2e96852020-10-10 18:34:16 +01007628 percpu_ref_kill(&data->node->refs);
Jens Axboe6a4d07c2020-05-15 14:30:38 -06007629 spin_lock(&data->lock);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007630 list_add(&ref_node->node, &data->ref_list);
Pavel Begunkovb2e96852020-10-10 18:34:16 +01007631 data->node = ref_node;
Jens Axboe6a4d07c2020-05-15 14:30:38 -06007632 spin_unlock(&data->lock);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007633 percpu_ref_get(&ctx->file_data->refs);
7634 } else
7635 destroy_fixed_file_ref_node(ref_node);
Jens Axboec3a31e62019-10-03 13:59:56 -06007636
7637 return done ? done : err;
7638}
Xiaoguang Wang05589552020-03-31 14:05:18 +08007639
Jens Axboe05f3fb32019-12-09 11:22:50 -07007640static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg,
7641 unsigned nr_args)
7642{
7643 struct io_uring_files_update up;
7644
7645 if (!ctx->file_data)
7646 return -ENXIO;
7647 if (!nr_args)
7648 return -EINVAL;
7649 if (copy_from_user(&up, arg, sizeof(up)))
7650 return -EFAULT;
7651 if (up.resv)
7652 return -EINVAL;
7653
7654 return __io_sqe_files_update(ctx, &up, nr_args);
7655}
Jens Axboec3a31e62019-10-03 13:59:56 -06007656
Pavel Begunkove9fd9392020-03-04 16:14:12 +03007657static void io_free_work(struct io_wq_work *work)
Jens Axboe7d723062019-11-12 22:31:31 -07007658{
7659 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
7660
Pavel Begunkove9fd9392020-03-04 16:14:12 +03007661 /* Consider that io_steal_work() relies on this ref */
Jens Axboe7d723062019-11-12 22:31:31 -07007662 io_put_req(req);
7663}
7664
Pavel Begunkov24369c22020-01-28 03:15:48 +03007665static int io_init_wq_offload(struct io_ring_ctx *ctx,
7666 struct io_uring_params *p)
7667{
7668 struct io_wq_data data;
7669 struct fd f;
7670 struct io_ring_ctx *ctx_attach;
7671 unsigned int concurrency;
7672 int ret = 0;
7673
7674 data.user = ctx->user;
Pavel Begunkove9fd9392020-03-04 16:14:12 +03007675 data.free_work = io_free_work;
Pavel Begunkovf5fa38c2020-06-08 21:08:20 +03007676 data.do_work = io_wq_submit_work;
Pavel Begunkov24369c22020-01-28 03:15:48 +03007677
7678 if (!(p->flags & IORING_SETUP_ATTACH_WQ)) {
7679 /* Do QD, or 4 * CPUS, whatever is smallest */
7680 concurrency = min(ctx->sq_entries, 4 * num_online_cpus());
7681
7682 ctx->io_wq = io_wq_create(concurrency, &data);
7683 if (IS_ERR(ctx->io_wq)) {
7684 ret = PTR_ERR(ctx->io_wq);
7685 ctx->io_wq = NULL;
7686 }
7687 return ret;
7688 }
7689
7690 f = fdget(p->wq_fd);
7691 if (!f.file)
7692 return -EBADF;
7693
7694 if (f.file->f_op != &io_uring_fops) {
7695 ret = -EINVAL;
7696 goto out_fput;
7697 }
7698
7699 ctx_attach = f.file->private_data;
7700 /* @io_wq is protected by holding the fd */
7701 if (!io_wq_get(ctx_attach->io_wq, &data)) {
7702 ret = -EINVAL;
7703 goto out_fput;
7704 }
7705
7706 ctx->io_wq = ctx_attach->io_wq;
7707out_fput:
7708 fdput(f);
7709 return ret;
7710}
7711
Jens Axboe0f212202020-09-13 13:09:39 -06007712static int io_uring_alloc_task_context(struct task_struct *task)
7713{
7714 struct io_uring_task *tctx;
Jens Axboed8a6df12020-10-15 16:24:45 -06007715 int ret;
Jens Axboe0f212202020-09-13 13:09:39 -06007716
7717 tctx = kmalloc(sizeof(*tctx), GFP_KERNEL);
7718 if (unlikely(!tctx))
7719 return -ENOMEM;
7720
Jens Axboed8a6df12020-10-15 16:24:45 -06007721 ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL);
7722 if (unlikely(ret)) {
7723 kfree(tctx);
7724 return ret;
7725 }
7726
Jens Axboe0f212202020-09-13 13:09:39 -06007727 xa_init(&tctx->xa);
7728 init_waitqueue_head(&tctx->wait);
7729 tctx->last = NULL;
Jens Axboefdaf0832020-10-30 09:37:30 -06007730 atomic_set(&tctx->in_idle, 0);
7731 tctx->sqpoll = false;
Jens Axboe500a3732020-10-15 17:38:03 -06007732 io_init_identity(&tctx->__identity);
7733 tctx->identity = &tctx->__identity;
Jens Axboe0f212202020-09-13 13:09:39 -06007734 task->io_uring = tctx;
7735 return 0;
7736}
7737
7738void __io_uring_free(struct task_struct *tsk)
7739{
7740 struct io_uring_task *tctx = tsk->io_uring;
7741
7742 WARN_ON_ONCE(!xa_empty(&tctx->xa));
Jens Axboe500a3732020-10-15 17:38:03 -06007743 WARN_ON_ONCE(refcount_read(&tctx->identity->count) != 1);
7744 if (tctx->identity != &tctx->__identity)
7745 kfree(tctx->identity);
Jens Axboed8a6df12020-10-15 16:24:45 -06007746 percpu_counter_destroy(&tctx->inflight);
Jens Axboe0f212202020-09-13 13:09:39 -06007747 kfree(tctx);
7748 tsk->io_uring = NULL;
7749}
7750
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02007751static int io_sq_offload_create(struct io_ring_ctx *ctx,
7752 struct io_uring_params *p)
Jens Axboe2b188cc2019-01-07 10:46:33 -07007753{
7754 int ret;
7755
Jens Axboe6c271ce2019-01-10 11:22:30 -07007756 if (ctx->flags & IORING_SETUP_SQPOLL) {
Jens Axboe534ca6d2020-09-02 13:52:19 -06007757 struct io_sq_data *sqd;
7758
Jens Axboe3ec482d2019-04-08 10:51:01 -06007759 ret = -EPERM;
7760 if (!capable(CAP_SYS_ADMIN))
7761 goto err;
7762
Jens Axboe534ca6d2020-09-02 13:52:19 -06007763 sqd = io_get_sq_data(p);
7764 if (IS_ERR(sqd)) {
7765 ret = PTR_ERR(sqd);
7766 goto err;
7767 }
Jens Axboe69fb2132020-09-14 11:16:23 -06007768
Jens Axboe534ca6d2020-09-02 13:52:19 -06007769 ctx->sq_data = sqd;
Jens Axboe69fb2132020-09-14 11:16:23 -06007770 io_sq_thread_park(sqd);
7771 mutex_lock(&sqd->ctx_lock);
7772 list_add(&ctx->sqd_list, &sqd->ctx_new_list);
7773 mutex_unlock(&sqd->ctx_lock);
7774 io_sq_thread_unpark(sqd);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007775
Jens Axboe917257d2019-04-13 09:28:55 -06007776 ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle);
7777 if (!ctx->sq_thread_idle)
7778 ctx->sq_thread_idle = HZ;
7779
Jens Axboeaa061652020-09-02 14:50:27 -06007780 if (sqd->thread)
7781 goto done;
7782
Jens Axboe6c271ce2019-01-10 11:22:30 -07007783 if (p->flags & IORING_SETUP_SQ_AFF) {
Jens Axboe44a9bd12019-05-14 20:00:30 -06007784 int cpu = p->sq_thread_cpu;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007785
Jens Axboe917257d2019-04-13 09:28:55 -06007786 ret = -EINVAL;
Jens Axboe44a9bd12019-05-14 20:00:30 -06007787 if (cpu >= nr_cpu_ids)
7788 goto err;
Shenghui Wang7889f442019-05-07 16:03:19 +08007789 if (!cpu_online(cpu))
Jens Axboe917257d2019-04-13 09:28:55 -06007790 goto err;
7791
Jens Axboe69fb2132020-09-14 11:16:23 -06007792 sqd->thread = kthread_create_on_cpu(io_sq_thread, sqd,
Jens Axboe534ca6d2020-09-02 13:52:19 -06007793 cpu, "io_uring-sq");
Jens Axboe6c271ce2019-01-10 11:22:30 -07007794 } else {
Jens Axboe69fb2132020-09-14 11:16:23 -06007795 sqd->thread = kthread_create(io_sq_thread, sqd,
Jens Axboe6c271ce2019-01-10 11:22:30 -07007796 "io_uring-sq");
7797 }
Jens Axboe534ca6d2020-09-02 13:52:19 -06007798 if (IS_ERR(sqd->thread)) {
7799 ret = PTR_ERR(sqd->thread);
7800 sqd->thread = NULL;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007801 goto err;
7802 }
Jens Axboe534ca6d2020-09-02 13:52:19 -06007803 ret = io_uring_alloc_task_context(sqd->thread);
Jens Axboe0f212202020-09-13 13:09:39 -06007804 if (ret)
7805 goto err;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007806 } else if (p->flags & IORING_SETUP_SQ_AFF) {
7807 /* Can't have SQ_AFF without SQPOLL */
7808 ret = -EINVAL;
7809 goto err;
7810 }
7811
Jens Axboeaa061652020-09-02 14:50:27 -06007812done:
Pavel Begunkov24369c22020-01-28 03:15:48 +03007813 ret = io_init_wq_offload(ctx, p);
7814 if (ret)
Jens Axboe2b188cc2019-01-07 10:46:33 -07007815 goto err;
Jens Axboe2b188cc2019-01-07 10:46:33 -07007816
7817 return 0;
7818err:
Jens Axboe54a91f32019-09-10 09:15:04 -06007819 io_finish_async(ctx);
Jens Axboe2b188cc2019-01-07 10:46:33 -07007820 return ret;
7821}
7822
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02007823static void io_sq_offload_start(struct io_ring_ctx *ctx)
7824{
Jens Axboe534ca6d2020-09-02 13:52:19 -06007825 struct io_sq_data *sqd = ctx->sq_data;
7826
7827 if ((ctx->flags & IORING_SETUP_SQPOLL) && sqd->thread)
7828 wake_up_process(sqd->thread);
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02007829}
7830
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007831static inline void __io_unaccount_mem(struct user_struct *user,
7832 unsigned long nr_pages)
Jens Axboe2b188cc2019-01-07 10:46:33 -07007833{
7834 atomic_long_sub(nr_pages, &user->locked_vm);
7835}
7836
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007837static inline int __io_account_mem(struct user_struct *user,
7838 unsigned long nr_pages)
Jens Axboe2b188cc2019-01-07 10:46:33 -07007839{
7840 unsigned long page_limit, cur_pages, new_pages;
7841
7842 /* Don't allow more pages than we can safely lock */
7843 page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
7844
7845 do {
7846 cur_pages = atomic_long_read(&user->locked_vm);
7847 new_pages = cur_pages + nr_pages;
7848 if (new_pages > page_limit)
7849 return -ENOMEM;
7850 } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages,
7851 new_pages) != cur_pages);
7852
7853 return 0;
7854}
7855
Bijan Mottahedeh2e0464d2020-06-16 16:36:10 -07007856static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages,
7857 enum io_mem_account acct)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007858{
Bijan Mottahedehaad5d8d2020-06-16 16:36:08 -07007859 if (ctx->limit_mem)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007860 __io_unaccount_mem(ctx->user, nr_pages);
Bijan Mottahedeh30975822020-06-16 16:36:09 -07007861
Jens Axboe2aede0e2020-09-14 10:45:53 -06007862 if (ctx->mm_account) {
Bijan Mottahedeh2e0464d2020-06-16 16:36:10 -07007863 if (acct == ACCT_LOCKED)
Jens Axboe2aede0e2020-09-14 10:45:53 -06007864 ctx->mm_account->locked_vm -= nr_pages;
Bijan Mottahedeh2e0464d2020-06-16 16:36:10 -07007865 else if (acct == ACCT_PINNED)
Jens Axboe2aede0e2020-09-14 10:45:53 -06007866 atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm);
Bijan Mottahedeh2e0464d2020-06-16 16:36:10 -07007867 }
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007868}
7869
Bijan Mottahedeh2e0464d2020-06-16 16:36:10 -07007870static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages,
7871 enum io_mem_account acct)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007872{
Bijan Mottahedeh30975822020-06-16 16:36:09 -07007873 int ret;
7874
7875 if (ctx->limit_mem) {
7876 ret = __io_account_mem(ctx->user, nr_pages);
7877 if (ret)
7878 return ret;
7879 }
7880
Jens Axboe2aede0e2020-09-14 10:45:53 -06007881 if (ctx->mm_account) {
Bijan Mottahedeh2e0464d2020-06-16 16:36:10 -07007882 if (acct == ACCT_LOCKED)
Jens Axboe2aede0e2020-09-14 10:45:53 -06007883 ctx->mm_account->locked_vm += nr_pages;
Bijan Mottahedeh2e0464d2020-06-16 16:36:10 -07007884 else if (acct == ACCT_PINNED)
Jens Axboe2aede0e2020-09-14 10:45:53 -06007885 atomic64_add(nr_pages, &ctx->mm_account->pinned_vm);
Bijan Mottahedeh2e0464d2020-06-16 16:36:10 -07007886 }
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07007887
7888 return 0;
7889}
7890
Jens Axboe2b188cc2019-01-07 10:46:33 -07007891static void io_mem_free(void *ptr)
7892{
Mark Rutland52e04ef2019-04-30 17:30:21 +01007893 struct page *page;
Jens Axboe2b188cc2019-01-07 10:46:33 -07007894
Mark Rutland52e04ef2019-04-30 17:30:21 +01007895 if (!ptr)
7896 return;
7897
7898 page = virt_to_head_page(ptr);
Jens Axboe2b188cc2019-01-07 10:46:33 -07007899 if (put_page_testzero(page))
7900 free_compound_page(page);
7901}
7902
7903static void *io_mem_alloc(size_t size)
7904{
7905 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP |
7906 __GFP_NORETRY;
7907
7908 return (void *) __get_free_pages(gfp_flags, get_order(size));
7909}
7910
Hristo Venev75b28af2019-08-26 17:23:46 +00007911static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries,
7912 size_t *sq_offset)
7913{
7914 struct io_rings *rings;
7915 size_t off, sq_array_size;
7916
7917 off = struct_size(rings, cqes, cq_entries);
7918 if (off == SIZE_MAX)
7919 return SIZE_MAX;
7920
7921#ifdef CONFIG_SMP
7922 off = ALIGN(off, SMP_CACHE_BYTES);
7923 if (off == 0)
7924 return SIZE_MAX;
7925#endif
7926
Dmitry Vyukovb36200f2020-07-11 11:31:11 +02007927 if (sq_offset)
7928 *sq_offset = off;
7929
Hristo Venev75b28af2019-08-26 17:23:46 +00007930 sq_array_size = array_size(sizeof(u32), sq_entries);
7931 if (sq_array_size == SIZE_MAX)
7932 return SIZE_MAX;
7933
7934 if (check_add_overflow(off, sq_array_size, &off))
7935 return SIZE_MAX;
7936
Hristo Venev75b28af2019-08-26 17:23:46 +00007937 return off;
7938}
7939
Jens Axboe2b188cc2019-01-07 10:46:33 -07007940static unsigned long ring_pages(unsigned sq_entries, unsigned cq_entries)
7941{
Hristo Venev75b28af2019-08-26 17:23:46 +00007942 size_t pages;
Jens Axboe2b188cc2019-01-07 10:46:33 -07007943
Hristo Venev75b28af2019-08-26 17:23:46 +00007944 pages = (size_t)1 << get_order(
7945 rings_size(sq_entries, cq_entries, NULL));
7946 pages += (size_t)1 << get_order(
7947 array_size(sizeof(struct io_uring_sqe), sq_entries));
Jens Axboe2b188cc2019-01-07 10:46:33 -07007948
Hristo Venev75b28af2019-08-26 17:23:46 +00007949 return pages;
Jens Axboe2b188cc2019-01-07 10:46:33 -07007950}
7951
Jens Axboeedafcce2019-01-09 09:16:05 -07007952static int io_sqe_buffer_unregister(struct io_ring_ctx *ctx)
7953{
7954 int i, j;
7955
7956 if (!ctx->user_bufs)
7957 return -ENXIO;
7958
7959 for (i = 0; i < ctx->nr_user_bufs; i++) {
7960 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
7961
7962 for (j = 0; j < imu->nr_bvecs; j++)
John Hubbardf1f6a7d2020-01-30 22:13:35 -08007963 unpin_user_page(imu->bvec[j].bv_page);
Jens Axboeedafcce2019-01-09 09:16:05 -07007964
Jens Axboede293932020-09-17 16:19:16 -06007965 if (imu->acct_pages)
7966 io_unaccount_mem(ctx, imu->acct_pages, ACCT_PINNED);
Mark Rutlandd4ef6472019-05-01 16:59:16 +01007967 kvfree(imu->bvec);
Jens Axboeedafcce2019-01-09 09:16:05 -07007968 imu->nr_bvecs = 0;
7969 }
7970
7971 kfree(ctx->user_bufs);
7972 ctx->user_bufs = NULL;
7973 ctx->nr_user_bufs = 0;
7974 return 0;
7975}
7976
7977static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst,
7978 void __user *arg, unsigned index)
7979{
7980 struct iovec __user *src;
7981
7982#ifdef CONFIG_COMPAT
7983 if (ctx->compat) {
7984 struct compat_iovec __user *ciovs;
7985 struct compat_iovec ciov;
7986
7987 ciovs = (struct compat_iovec __user *) arg;
7988 if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov)))
7989 return -EFAULT;
7990
Jens Axboed55e5f52019-12-11 16:12:15 -07007991 dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base);
Jens Axboeedafcce2019-01-09 09:16:05 -07007992 dst->iov_len = ciov.iov_len;
7993 return 0;
7994 }
7995#endif
7996 src = (struct iovec __user *) arg;
7997 if (copy_from_user(dst, &src[index], sizeof(*dst)))
7998 return -EFAULT;
7999 return 0;
8000}
8001
Jens Axboede293932020-09-17 16:19:16 -06008002/*
8003 * Not super efficient, but this is just a registration time. And we do cache
8004 * the last compound head, so generally we'll only do a full search if we don't
8005 * match that one.
8006 *
8007 * We check if the given compound head page has already been accounted, to
8008 * avoid double accounting it. This allows us to account the full size of the
8009 * page, not just the constituent pages of a huge page.
8010 */
8011static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages,
8012 int nr_pages, struct page *hpage)
8013{
8014 int i, j;
8015
8016 /* check current page array */
8017 for (i = 0; i < nr_pages; i++) {
8018 if (!PageCompound(pages[i]))
8019 continue;
8020 if (compound_head(pages[i]) == hpage)
8021 return true;
8022 }
8023
8024 /* check previously registered pages */
8025 for (i = 0; i < ctx->nr_user_bufs; i++) {
8026 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
8027
8028 for (j = 0; j < imu->nr_bvecs; j++) {
8029 if (!PageCompound(imu->bvec[j].bv_page))
8030 continue;
8031 if (compound_head(imu->bvec[j].bv_page) == hpage)
8032 return true;
8033 }
8034 }
8035
8036 return false;
8037}
8038
8039static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages,
8040 int nr_pages, struct io_mapped_ubuf *imu,
8041 struct page **last_hpage)
8042{
8043 int i, ret;
8044
8045 for (i = 0; i < nr_pages; i++) {
8046 if (!PageCompound(pages[i])) {
8047 imu->acct_pages++;
8048 } else {
8049 struct page *hpage;
8050
8051 hpage = compound_head(pages[i]);
8052 if (hpage == *last_hpage)
8053 continue;
8054 *last_hpage = hpage;
8055 if (headpage_already_acct(ctx, pages, i, hpage))
8056 continue;
8057 imu->acct_pages += page_size(hpage) >> PAGE_SHIFT;
8058 }
8059 }
8060
8061 if (!imu->acct_pages)
8062 return 0;
8063
8064 ret = io_account_mem(ctx, imu->acct_pages, ACCT_PINNED);
8065 if (ret)
8066 imu->acct_pages = 0;
8067 return ret;
8068}
8069
Jens Axboeedafcce2019-01-09 09:16:05 -07008070static int io_sqe_buffer_register(struct io_ring_ctx *ctx, void __user *arg,
8071 unsigned nr_args)
8072{
8073 struct vm_area_struct **vmas = NULL;
8074 struct page **pages = NULL;
Jens Axboede293932020-09-17 16:19:16 -06008075 struct page *last_hpage = NULL;
Jens Axboeedafcce2019-01-09 09:16:05 -07008076 int i, j, got_pages = 0;
8077 int ret = -EINVAL;
8078
8079 if (ctx->user_bufs)
8080 return -EBUSY;
8081 if (!nr_args || nr_args > UIO_MAXIOV)
8082 return -EINVAL;
8083
8084 ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf),
8085 GFP_KERNEL);
8086 if (!ctx->user_bufs)
8087 return -ENOMEM;
8088
8089 for (i = 0; i < nr_args; i++) {
8090 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
8091 unsigned long off, start, end, ubuf;
8092 int pret, nr_pages;
8093 struct iovec iov;
8094 size_t size;
8095
8096 ret = io_copy_iov(ctx, &iov, arg, i);
8097 if (ret)
Pavel Begunkova2786822019-05-26 12:35:47 +03008098 goto err;
Jens Axboeedafcce2019-01-09 09:16:05 -07008099
8100 /*
8101 * Don't impose further limits on the size and buffer
8102 * constraints here, we'll -EINVAL later when IO is
8103 * submitted if they are wrong.
8104 */
8105 ret = -EFAULT;
8106 if (!iov.iov_base || !iov.iov_len)
8107 goto err;
8108
8109 /* arbitrary limit, but we need something */
8110 if (iov.iov_len > SZ_1G)
8111 goto err;
8112
8113 ubuf = (unsigned long) iov.iov_base;
8114 end = (ubuf + iov.iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
8115 start = ubuf >> PAGE_SHIFT;
8116 nr_pages = end - start;
8117
Jens Axboeedafcce2019-01-09 09:16:05 -07008118 ret = 0;
8119 if (!pages || nr_pages > got_pages) {
Denis Efremova8c73c12020-06-05 12:32:03 +03008120 kvfree(vmas);
8121 kvfree(pages);
Mark Rutlandd4ef6472019-05-01 16:59:16 +01008122 pages = kvmalloc_array(nr_pages, sizeof(struct page *),
Jens Axboeedafcce2019-01-09 09:16:05 -07008123 GFP_KERNEL);
Mark Rutlandd4ef6472019-05-01 16:59:16 +01008124 vmas = kvmalloc_array(nr_pages,
Jens Axboeedafcce2019-01-09 09:16:05 -07008125 sizeof(struct vm_area_struct *),
8126 GFP_KERNEL);
8127 if (!pages || !vmas) {
8128 ret = -ENOMEM;
Jens Axboeedafcce2019-01-09 09:16:05 -07008129 goto err;
8130 }
8131 got_pages = nr_pages;
8132 }
8133
Mark Rutlandd4ef6472019-05-01 16:59:16 +01008134 imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec),
Jens Axboeedafcce2019-01-09 09:16:05 -07008135 GFP_KERNEL);
8136 ret = -ENOMEM;
Jens Axboede293932020-09-17 16:19:16 -06008137 if (!imu->bvec)
Jens Axboeedafcce2019-01-09 09:16:05 -07008138 goto err;
Jens Axboeedafcce2019-01-09 09:16:05 -07008139
8140 ret = 0;
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07008141 mmap_read_lock(current->mm);
John Hubbard2113b052020-01-30 22:13:13 -08008142 pret = pin_user_pages(ubuf, nr_pages,
Ira Weiny932f4a62019-05-13 17:17:03 -07008143 FOLL_WRITE | FOLL_LONGTERM,
8144 pages, vmas);
Jens Axboeedafcce2019-01-09 09:16:05 -07008145 if (pret == nr_pages) {
8146 /* don't support file backed memory */
8147 for (j = 0; j < nr_pages; j++) {
8148 struct vm_area_struct *vma = vmas[j];
8149
8150 if (vma->vm_file &&
8151 !is_file_hugepages(vma->vm_file)) {
8152 ret = -EOPNOTSUPP;
8153 break;
8154 }
8155 }
8156 } else {
8157 ret = pret < 0 ? pret : -EFAULT;
8158 }
Michel Lespinassed8ed45c2020-06-08 21:33:25 -07008159 mmap_read_unlock(current->mm);
Jens Axboeedafcce2019-01-09 09:16:05 -07008160 if (ret) {
8161 /*
8162 * if we did partial map, or found file backed vmas,
8163 * release any pages we did get
8164 */
John Hubbard27c4d3a2019-08-04 19:32:06 -07008165 if (pret > 0)
John Hubbardf1f6a7d2020-01-30 22:13:35 -08008166 unpin_user_pages(pages, pret);
Jens Axboede293932020-09-17 16:19:16 -06008167 kvfree(imu->bvec);
8168 goto err;
8169 }
8170
8171 ret = io_buffer_account_pin(ctx, pages, pret, imu, &last_hpage);
8172 if (ret) {
8173 unpin_user_pages(pages, pret);
Mark Rutlandd4ef6472019-05-01 16:59:16 +01008174 kvfree(imu->bvec);
Jens Axboeedafcce2019-01-09 09:16:05 -07008175 goto err;
8176 }
8177
8178 off = ubuf & ~PAGE_MASK;
8179 size = iov.iov_len;
8180 for (j = 0; j < nr_pages; j++) {
8181 size_t vec_len;
8182
8183 vec_len = min_t(size_t, size, PAGE_SIZE - off);
8184 imu->bvec[j].bv_page = pages[j];
8185 imu->bvec[j].bv_len = vec_len;
8186 imu->bvec[j].bv_offset = off;
8187 off = 0;
8188 size -= vec_len;
8189 }
8190 /* store original address for later verification */
8191 imu->ubuf = ubuf;
8192 imu->len = iov.iov_len;
8193 imu->nr_bvecs = nr_pages;
8194
8195 ctx->nr_user_bufs++;
8196 }
Mark Rutlandd4ef6472019-05-01 16:59:16 +01008197 kvfree(pages);
8198 kvfree(vmas);
Jens Axboeedafcce2019-01-09 09:16:05 -07008199 return 0;
8200err:
Mark Rutlandd4ef6472019-05-01 16:59:16 +01008201 kvfree(pages);
8202 kvfree(vmas);
Jens Axboeedafcce2019-01-09 09:16:05 -07008203 io_sqe_buffer_unregister(ctx);
8204 return ret;
8205}
8206
Jens Axboe9b402842019-04-11 11:45:41 -06008207static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg)
8208{
8209 __s32 __user *fds = arg;
8210 int fd;
8211
8212 if (ctx->cq_ev_fd)
8213 return -EBUSY;
8214
8215 if (copy_from_user(&fd, fds, sizeof(*fds)))
8216 return -EFAULT;
8217
8218 ctx->cq_ev_fd = eventfd_ctx_fdget(fd);
8219 if (IS_ERR(ctx->cq_ev_fd)) {
8220 int ret = PTR_ERR(ctx->cq_ev_fd);
8221 ctx->cq_ev_fd = NULL;
8222 return ret;
8223 }
8224
8225 return 0;
8226}
8227
8228static int io_eventfd_unregister(struct io_ring_ctx *ctx)
8229{
8230 if (ctx->cq_ev_fd) {
8231 eventfd_ctx_put(ctx->cq_ev_fd);
8232 ctx->cq_ev_fd = NULL;
8233 return 0;
8234 }
8235
8236 return -ENXIO;
8237}
8238
Jens Axboe5a2e7452020-02-23 16:23:11 -07008239static int __io_destroy_buffers(int id, void *p, void *data)
8240{
8241 struct io_ring_ctx *ctx = data;
8242 struct io_buffer *buf = p;
8243
Jens Axboe067524e2020-03-02 16:32:28 -07008244 __io_remove_buffers(ctx, buf, id, -1U);
Jens Axboe5a2e7452020-02-23 16:23:11 -07008245 return 0;
8246}
8247
8248static void io_destroy_buffers(struct io_ring_ctx *ctx)
8249{
8250 idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx);
8251 idr_destroy(&ctx->io_buffer_idr);
8252}
8253
Jens Axboe2b188cc2019-01-07 10:46:33 -07008254static void io_ring_ctx_free(struct io_ring_ctx *ctx)
8255{
Jens Axboe6b063142019-01-10 22:13:58 -07008256 io_finish_async(ctx);
Jens Axboeedafcce2019-01-09 09:16:05 -07008257 io_sqe_buffer_unregister(ctx);
Jens Axboe2aede0e2020-09-14 10:45:53 -06008258
8259 if (ctx->sqo_task) {
8260 put_task_struct(ctx->sqo_task);
8261 ctx->sqo_task = NULL;
8262 mmdrop(ctx->mm_account);
8263 ctx->mm_account = NULL;
Bijan Mottahedeh30975822020-06-16 16:36:09 -07008264 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07008265
Dennis Zhou91d8f512020-09-16 13:41:05 -07008266#ifdef CONFIG_BLK_CGROUP
8267 if (ctx->sqo_blkcg_css)
8268 css_put(ctx->sqo_blkcg_css);
8269#endif
8270
Jens Axboe6b063142019-01-10 22:13:58 -07008271 io_sqe_files_unregister(ctx);
Jens Axboe9b402842019-04-11 11:45:41 -06008272 io_eventfd_unregister(ctx);
Jens Axboe5a2e7452020-02-23 16:23:11 -07008273 io_destroy_buffers(ctx);
Jens Axboe41726c92020-02-23 13:11:42 -07008274 idr_destroy(&ctx->personality_idr);
Jens Axboedef596e2019-01-09 08:59:42 -07008275
Jens Axboe2b188cc2019-01-07 10:46:33 -07008276#if defined(CONFIG_UNIX)
Eric Biggers355e8d262019-06-12 14:58:43 -07008277 if (ctx->ring_sock) {
8278 ctx->ring_sock->file = NULL; /* so that iput() is called */
Jens Axboe2b188cc2019-01-07 10:46:33 -07008279 sock_release(ctx->ring_sock);
Eric Biggers355e8d262019-06-12 14:58:43 -07008280 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07008281#endif
8282
Hristo Venev75b28af2019-08-26 17:23:46 +00008283 io_mem_free(ctx->rings);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008284 io_mem_free(ctx->sq_sqes);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008285
8286 percpu_ref_exit(&ctx->refs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008287 free_uid(ctx->user);
Jens Axboe181e4482019-11-25 08:52:30 -07008288 put_cred(ctx->creds);
Jens Axboe78076bb2019-12-04 19:56:40 -07008289 kfree(ctx->cancel_hash);
Jens Axboe0ddf92e2019-11-08 08:52:53 -07008290 kmem_cache_free(req_cachep, ctx->fallback_req);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008291 kfree(ctx);
8292}
8293
8294static __poll_t io_uring_poll(struct file *file, poll_table *wait)
8295{
8296 struct io_ring_ctx *ctx = file->private_data;
8297 __poll_t mask = 0;
8298
8299 poll_wait(file, &ctx->cq_wait, wait);
Stefan Bühler4f7067c2019-04-24 23:54:17 +02008300 /*
8301 * synchronizes with barrier from wq_has_sleeper call in
8302 * io_commit_cqring
8303 */
Jens Axboe2b188cc2019-01-07 10:46:33 -07008304 smp_rmb();
Jens Axboe90554202020-09-03 12:12:41 -06008305 if (!io_sqring_full(ctx))
Jens Axboe2b188cc2019-01-07 10:46:33 -07008306 mask |= EPOLLOUT | EPOLLWRNORM;
Stefano Garzarella63e5d812020-02-07 13:18:28 +01008307 if (io_cqring_events(ctx, false))
Jens Axboe2b188cc2019-01-07 10:46:33 -07008308 mask |= EPOLLIN | EPOLLRDNORM;
8309
8310 return mask;
8311}
8312
8313static int io_uring_fasync(int fd, struct file *file, int on)
8314{
8315 struct io_ring_ctx *ctx = file->private_data;
8316
8317 return fasync_helper(fd, file, on, &ctx->cq_fasync);
8318}
8319
Jens Axboe071698e2020-01-28 10:04:42 -07008320static int io_remove_personalities(int id, void *p, void *data)
8321{
8322 struct io_ring_ctx *ctx = data;
Jens Axboe1e6fa522020-10-15 08:46:24 -06008323 struct io_identity *iod;
Jens Axboe071698e2020-01-28 10:04:42 -07008324
Jens Axboe1e6fa522020-10-15 08:46:24 -06008325 iod = idr_remove(&ctx->personality_idr, id);
8326 if (iod) {
8327 put_cred(iod->creds);
8328 if (refcount_dec_and_test(&iod->count))
8329 kfree(iod);
8330 }
Jens Axboe071698e2020-01-28 10:04:42 -07008331 return 0;
8332}
8333
Jens Axboe85faa7b2020-04-09 18:14:00 -06008334static void io_ring_exit_work(struct work_struct *work)
8335{
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008336 struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx,
8337 exit_work);
Jens Axboe85faa7b2020-04-09 18:14:00 -06008338
Jens Axboe56952e92020-06-17 15:00:04 -06008339 /*
8340 * If we're doing polled IO and end up having requests being
8341 * submitted async (out-of-line), then completions can come in while
8342 * we're waiting for refs to drop. We need to reap these manually,
8343 * as nobody else will be looking for them.
8344 */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008345 do {
Jens Axboe56952e92020-06-17 15:00:04 -06008346 if (ctx->rings)
Jens Axboee6c8aa92020-09-28 13:10:13 -06008347 io_cqring_overflow_flush(ctx, true, NULL, NULL);
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008348 io_iopoll_try_reap_events(ctx);
8349 } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20));
Jens Axboe85faa7b2020-04-09 18:14:00 -06008350 io_ring_ctx_free(ctx);
8351}
8352
Jens Axboe2b188cc2019-01-07 10:46:33 -07008353static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx)
8354{
8355 mutex_lock(&ctx->uring_lock);
8356 percpu_ref_kill(&ctx->refs);
8357 mutex_unlock(&ctx->uring_lock);
8358
Jens Axboef3606e32020-09-22 08:18:24 -06008359 io_kill_timeouts(ctx, NULL);
8360 io_poll_remove_all(ctx, NULL);
Jens Axboe561fb042019-10-24 07:25:42 -06008361
8362 if (ctx->io_wq)
8363 io_wq_cancel_all(ctx->io_wq);
8364
Jens Axboe15dff282019-11-13 09:09:23 -07008365 /* if we failed setting up the ctx, we might not have any rings */
8366 if (ctx->rings)
Jens Axboee6c8aa92020-09-28 13:10:13 -06008367 io_cqring_overflow_flush(ctx, true, NULL, NULL);
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008368 io_iopoll_try_reap_events(ctx);
Jens Axboe071698e2020-01-28 10:04:42 -07008369 idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx);
Jens Axboe309fc032020-07-10 09:13:34 -06008370
8371 /*
8372 * Do this upfront, so we won't have a grace period where the ring
8373 * is closed but resources aren't reaped yet. This can cause
8374 * spurious failure in setting up a new ring.
8375 */
Jens Axboe760618f2020-07-24 12:53:31 -06008376 io_unaccount_mem(ctx, ring_pages(ctx->sq_entries, ctx->cq_entries),
8377 ACCT_LOCKED);
Jens Axboe309fc032020-07-10 09:13:34 -06008378
Jens Axboe85faa7b2020-04-09 18:14:00 -06008379 INIT_WORK(&ctx->exit_work, io_ring_exit_work);
Jens Axboefc666772020-08-19 11:10:51 -06008380 /*
8381 * Use system_unbound_wq to avoid spawning tons of event kworkers
8382 * if we're exiting a ton of rings at the same time. It just adds
8383 * noise and overhead, there's no discernable change in runtime
8384 * over using system_wq.
8385 */
8386 queue_work(system_unbound_wq, &ctx->exit_work);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008387}
8388
8389static int io_uring_release(struct inode *inode, struct file *file)
8390{
8391 struct io_ring_ctx *ctx = file->private_data;
8392
8393 file->private_data = NULL;
8394 io_ring_ctx_wait_and_kill(ctx);
8395 return 0;
8396}
8397
Pavel Begunkov67c4d9e2020-06-15 10:24:05 +03008398static bool io_wq_files_match(struct io_wq_work *work, void *data)
8399{
8400 struct files_struct *files = data;
8401
Jens Axboedfead8a2020-10-14 10:12:37 -06008402 return !files || ((work->flags & IO_WQ_WORK_FILES) &&
Jens Axboe98447d62020-10-14 10:48:51 -06008403 work->identity->files == files);
Pavel Begunkov67c4d9e2020-06-15 10:24:05 +03008404}
8405
Jens Axboef254ac02020-08-12 17:33:30 -06008406/*
8407 * Returns true if 'preq' is the link parent of 'req'
8408 */
8409static bool io_match_link(struct io_kiocb *preq, struct io_kiocb *req)
8410{
8411 struct io_kiocb *link;
8412
8413 if (!(preq->flags & REQ_F_LINK_HEAD))
8414 return false;
8415
8416 list_for_each_entry(link, &preq->link_list, link_list) {
8417 if (link == req)
8418 return true;
8419 }
8420
8421 return false;
8422}
8423
8424/*
8425 * We're looking to cancel 'req' because it's holding on to our files, but
8426 * 'req' could be a link to another request. See if it is, and cancel that
8427 * parent request if so.
8428 */
8429static bool io_poll_remove_link(struct io_ring_ctx *ctx, struct io_kiocb *req)
8430{
8431 struct hlist_node *tmp;
8432 struct io_kiocb *preq;
8433 bool found = false;
8434 int i;
8435
8436 spin_lock_irq(&ctx->completion_lock);
8437 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
8438 struct hlist_head *list;
8439
8440 list = &ctx->cancel_hash[i];
8441 hlist_for_each_entry_safe(preq, tmp, list, hash_node) {
8442 found = io_match_link(preq, req);
8443 if (found) {
8444 io_poll_remove_one(preq);
8445 break;
8446 }
8447 }
8448 }
8449 spin_unlock_irq(&ctx->completion_lock);
8450 return found;
8451}
8452
8453static bool io_timeout_remove_link(struct io_ring_ctx *ctx,
8454 struct io_kiocb *req)
8455{
8456 struct io_kiocb *preq;
8457 bool found = false;
8458
8459 spin_lock_irq(&ctx->completion_lock);
8460 list_for_each_entry(preq, &ctx->timeout_list, timeout.list) {
8461 found = io_match_link(preq, req);
8462 if (found) {
8463 __io_timeout_cancel(preq);
8464 break;
8465 }
8466 }
8467 spin_unlock_irq(&ctx->completion_lock);
8468 return found;
8469}
8470
Jens Axboeb711d4e2020-08-16 08:23:05 -07008471static bool io_cancel_link_cb(struct io_wq_work *work, void *data)
8472{
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008473 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
8474 bool ret;
8475
8476 if (req->flags & REQ_F_LINK_TIMEOUT) {
8477 unsigned long flags;
8478 struct io_ring_ctx *ctx = req->ctx;
8479
8480 /* protect against races with linked timeouts */
8481 spin_lock_irqsave(&ctx->completion_lock, flags);
8482 ret = io_match_link(req, data);
8483 spin_unlock_irqrestore(&ctx->completion_lock, flags);
8484 } else {
8485 ret = io_match_link(req, data);
8486 }
8487 return ret;
Jens Axboeb711d4e2020-08-16 08:23:05 -07008488}
8489
8490static void io_attempt_cancel(struct io_ring_ctx *ctx, struct io_kiocb *req)
8491{
8492 enum io_wq_cancel cret;
8493
8494 /* cancel this particular work, if it's running */
8495 cret = io_wq_cancel_work(ctx->io_wq, &req->work);
8496 if (cret != IO_WQ_CANCEL_NOTFOUND)
8497 return;
8498
8499 /* find links that hold this pending, cancel those */
8500 cret = io_wq_cancel_cb(ctx->io_wq, io_cancel_link_cb, req, true);
8501 if (cret != IO_WQ_CANCEL_NOTFOUND)
8502 return;
8503
8504 /* if we have a poll link holding this pending, cancel that */
8505 if (io_poll_remove_link(ctx, req))
8506 return;
8507
8508 /* final option, timeout link is holding this req pending */
8509 io_timeout_remove_link(ctx, req);
8510}
8511
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008512static void io_cancel_defer_files(struct io_ring_ctx *ctx,
Pavel Begunkovef9865a2020-11-05 14:06:19 +00008513 struct task_struct *task,
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008514 struct files_struct *files)
8515{
8516 struct io_defer_entry *de = NULL;
8517 LIST_HEAD(list);
8518
8519 spin_lock_irq(&ctx->completion_lock);
8520 list_for_each_entry_reverse(de, &ctx->defer_list, list) {
Pavel Begunkovef9865a2020-11-05 14:06:19 +00008521 if (io_task_match(de->req, task) &&
8522 io_match_files(de->req, files)) {
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008523 list_cut_position(&list, &ctx->defer_list, &de->list);
8524 break;
8525 }
8526 }
8527 spin_unlock_irq(&ctx->completion_lock);
8528
8529 while (!list_empty(&list)) {
8530 de = list_first_entry(&list, struct io_defer_entry, list);
8531 list_del_init(&de->list);
8532 req_set_fail_links(de->req);
8533 io_put_req(de->req);
8534 io_req_complete(de->req, -ECANCELED);
8535 kfree(de);
8536 }
8537}
8538
Jens Axboe76e1b642020-09-26 15:05:03 -06008539/*
8540 * Returns true if we found and killed one or more files pinning requests
8541 */
8542static bool io_uring_cancel_files(struct io_ring_ctx *ctx,
Jens Axboefcb323c2019-10-24 12:39:47 -06008543 struct files_struct *files)
8544{
Pavel Begunkov67c4d9e2020-06-15 10:24:05 +03008545 if (list_empty_careful(&ctx->inflight_list))
Jens Axboe76e1b642020-09-26 15:05:03 -06008546 return false;
Pavel Begunkov67c4d9e2020-06-15 10:24:05 +03008547
8548 /* cancel all at once, should be faster than doing it one by one*/
8549 io_wq_cancel_cb(ctx->io_wq, io_wq_files_match, files, true);
8550
Jens Axboefcb323c2019-10-24 12:39:47 -06008551 while (!list_empty_careful(&ctx->inflight_list)) {
Xiaoguang Wangd8f1b972020-04-26 15:54:43 +08008552 struct io_kiocb *cancel_req = NULL, *req;
8553 DEFINE_WAIT(wait);
Jens Axboefcb323c2019-10-24 12:39:47 -06008554
8555 spin_lock_irq(&ctx->inflight_lock);
8556 list_for_each_entry(req, &ctx->inflight_list, inflight_entry) {
Jens Axboedfead8a2020-10-14 10:12:37 -06008557 if (files && (req->work.flags & IO_WQ_WORK_FILES) &&
Jens Axboe98447d62020-10-14 10:48:51 -06008558 req->work.identity->files != files)
Jens Axboe768134d2019-11-10 20:30:53 -07008559 continue;
8560 /* req is being completed, ignore */
8561 if (!refcount_inc_not_zero(&req->refs))
8562 continue;
8563 cancel_req = req;
8564 break;
Jens Axboefcb323c2019-10-24 12:39:47 -06008565 }
Jens Axboe768134d2019-11-10 20:30:53 -07008566 if (cancel_req)
Jens Axboefcb323c2019-10-24 12:39:47 -06008567 prepare_to_wait(&ctx->inflight_wait, &wait,
Jens Axboe768134d2019-11-10 20:30:53 -07008568 TASK_UNINTERRUPTIBLE);
Jens Axboefcb323c2019-10-24 12:39:47 -06008569 spin_unlock_irq(&ctx->inflight_lock);
8570
Jens Axboe768134d2019-11-10 20:30:53 -07008571 /* We need to keep going until we don't find a matching req */
8572 if (!cancel_req)
Jens Axboefcb323c2019-10-24 12:39:47 -06008573 break;
Pavel Begunkovbb175342020-08-20 11:33:35 +03008574 /* cancel this request, or head link requests */
8575 io_attempt_cancel(ctx, cancel_req);
8576 io_put_req(cancel_req);
Jens Axboe6200b0a2020-09-13 14:38:30 -06008577 /* cancellations _may_ trigger task work */
8578 io_run_task_work();
Jens Axboefcb323c2019-10-24 12:39:47 -06008579 schedule();
Xiaoguang Wangd8f1b972020-04-26 15:54:43 +08008580 finish_wait(&ctx->inflight_wait, &wait);
Jens Axboefcb323c2019-10-24 12:39:47 -06008581 }
Jens Axboe76e1b642020-09-26 15:05:03 -06008582
8583 return true;
Jens Axboefcb323c2019-10-24 12:39:47 -06008584}
8585
Pavel Begunkov801dd572020-06-15 10:33:14 +03008586static bool io_cancel_task_cb(struct io_wq_work *work, void *data)
Pavel Begunkov44e728b2020-06-15 10:24:04 +03008587{
Pavel Begunkov801dd572020-06-15 10:33:14 +03008588 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
8589 struct task_struct *task = data;
Pavel Begunkov44e728b2020-06-15 10:24:04 +03008590
Jens Axboef3606e32020-09-22 08:18:24 -06008591 return io_task_match(req, task);
Pavel Begunkov44e728b2020-06-15 10:24:04 +03008592}
8593
Jens Axboe0f212202020-09-13 13:09:39 -06008594static bool __io_uring_cancel_task_requests(struct io_ring_ctx *ctx,
8595 struct task_struct *task,
8596 struct files_struct *files)
8597{
8598 bool ret;
8599
8600 ret = io_uring_cancel_files(ctx, files);
8601 if (!files) {
8602 enum io_wq_cancel cret;
8603
8604 cret = io_wq_cancel_cb(ctx->io_wq, io_cancel_task_cb, task, true);
8605 if (cret != IO_WQ_CANCEL_NOTFOUND)
8606 ret = true;
8607
8608 /* SQPOLL thread does its own polling */
8609 if (!(ctx->flags & IORING_SETUP_SQPOLL)) {
8610 while (!list_empty_careful(&ctx->iopoll_list)) {
8611 io_iopoll_try_reap_events(ctx);
8612 ret = true;
8613 }
8614 }
8615
8616 ret |= io_poll_remove_all(ctx, task);
8617 ret |= io_kill_timeouts(ctx, task);
8618 }
8619
8620 return ret;
8621}
8622
8623/*
8624 * We need to iteratively cancel requests, in case a request has dependent
8625 * hard links. These persist even for failure of cancelations, hence keep
8626 * looping until none are found.
8627 */
8628static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx,
8629 struct files_struct *files)
8630{
8631 struct task_struct *task = current;
8632
Jens Axboefdaf0832020-10-30 09:37:30 -06008633 if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) {
Jens Axboe534ca6d2020-09-02 13:52:19 -06008634 task = ctx->sq_data->thread;
Jens Axboefdaf0832020-10-30 09:37:30 -06008635 atomic_inc(&task->io_uring->in_idle);
8636 io_sq_thread_park(ctx->sq_data);
8637 }
Jens Axboe0f212202020-09-13 13:09:39 -06008638
Pavel Begunkovef9865a2020-11-05 14:06:19 +00008639 if (files)
8640 io_cancel_defer_files(ctx, NULL, files);
8641 else
8642 io_cancel_defer_files(ctx, task, NULL);
8643
Jens Axboe0f212202020-09-13 13:09:39 -06008644 io_cqring_overflow_flush(ctx, true, task, files);
8645
8646 while (__io_uring_cancel_task_requests(ctx, task, files)) {
8647 io_run_task_work();
8648 cond_resched();
8649 }
Jens Axboefdaf0832020-10-30 09:37:30 -06008650
8651 if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) {
8652 atomic_dec(&task->io_uring->in_idle);
8653 /*
8654 * If the files that are going away are the ones in the thread
8655 * identity, clear them out.
8656 */
8657 if (task->io_uring->identity->files == files)
8658 task->io_uring->identity->files = NULL;
8659 io_sq_thread_unpark(ctx->sq_data);
8660 }
Jens Axboe0f212202020-09-13 13:09:39 -06008661}
8662
8663/*
8664 * Note that this task has used io_uring. We use it for cancelation purposes.
8665 */
Jens Axboefdaf0832020-10-30 09:37:30 -06008666static int io_uring_add_task_file(struct io_ring_ctx *ctx, struct file *file)
Jens Axboe0f212202020-09-13 13:09:39 -06008667{
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008668 struct io_uring_task *tctx = current->io_uring;
8669
8670 if (unlikely(!tctx)) {
Jens Axboe0f212202020-09-13 13:09:39 -06008671 int ret;
8672
8673 ret = io_uring_alloc_task_context(current);
8674 if (unlikely(ret))
8675 return ret;
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008676 tctx = current->io_uring;
Jens Axboe0f212202020-09-13 13:09:39 -06008677 }
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008678 if (tctx->last != file) {
8679 void *old = xa_load(&tctx->xa, (unsigned long)file);
Jens Axboe0f212202020-09-13 13:09:39 -06008680
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008681 if (!old) {
Jens Axboe0f212202020-09-13 13:09:39 -06008682 get_file(file);
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008683 xa_store(&tctx->xa, (unsigned long)file, file, GFP_KERNEL);
Jens Axboe0f212202020-09-13 13:09:39 -06008684 }
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01008685 tctx->last = file;
Jens Axboe0f212202020-09-13 13:09:39 -06008686 }
8687
Jens Axboefdaf0832020-10-30 09:37:30 -06008688 /*
8689 * This is race safe in that the task itself is doing this, hence it
8690 * cannot be going through the exit/cancel paths at the same time.
8691 * This cannot be modified while exit/cancel is running.
8692 */
8693 if (!tctx->sqpoll && (ctx->flags & IORING_SETUP_SQPOLL))
8694 tctx->sqpoll = true;
8695
Jens Axboe0f212202020-09-13 13:09:39 -06008696 return 0;
8697}
8698
8699/*
8700 * Remove this io_uring_file -> task mapping.
8701 */
8702static void io_uring_del_task_file(struct file *file)
8703{
8704 struct io_uring_task *tctx = current->io_uring;
Jens Axboe0f212202020-09-13 13:09:39 -06008705
8706 if (tctx->last == file)
8707 tctx->last = NULL;
Matthew Wilcox (Oracle)5e2ed8c2020-10-09 13:49:53 +01008708 file = xa_erase(&tctx->xa, (unsigned long)file);
Jens Axboe0f212202020-09-13 13:09:39 -06008709 if (file)
8710 fput(file);
8711}
8712
Jens Axboe0f212202020-09-13 13:09:39 -06008713/*
8714 * Drop task note for this file if we're the only ones that hold it after
8715 * pending fput()
8716 */
Pavel Begunkovc8fb20b2020-10-22 16:38:27 +01008717static void io_uring_attempt_task_drop(struct file *file)
Jens Axboe0f212202020-09-13 13:09:39 -06008718{
8719 if (!current->io_uring)
8720 return;
8721 /*
8722 * fput() is pending, will be 2 if the only other ref is our potential
8723 * task file note. If the task is exiting, drop regardless of count.
8724 */
Pavel Begunkovc8fb20b2020-10-22 16:38:27 +01008725 if (fatal_signal_pending(current) || (current->flags & PF_EXITING) ||
8726 atomic_long_read(&file->f_count) == 2)
8727 io_uring_del_task_file(file);
Jens Axboe0f212202020-09-13 13:09:39 -06008728}
8729
8730void __io_uring_files_cancel(struct files_struct *files)
8731{
8732 struct io_uring_task *tctx = current->io_uring;
Matthew Wilcox (Oracle)ce765372020-10-09 13:49:51 +01008733 struct file *file;
8734 unsigned long index;
Jens Axboe0f212202020-09-13 13:09:39 -06008735
8736 /* make sure overflow events are dropped */
Jens Axboefdaf0832020-10-30 09:37:30 -06008737 atomic_inc(&tctx->in_idle);
Jens Axboe0f212202020-09-13 13:09:39 -06008738
Matthew Wilcox (Oracle)ce765372020-10-09 13:49:51 +01008739 xa_for_each(&tctx->xa, index, file) {
8740 struct io_ring_ctx *ctx = file->private_data;
Jens Axboe0f212202020-09-13 13:09:39 -06008741
8742 io_uring_cancel_task_requests(ctx, files);
8743 if (files)
8744 io_uring_del_task_file(file);
Matthew Wilcox (Oracle)ce765372020-10-09 13:49:51 +01008745 }
Jens Axboefdaf0832020-10-30 09:37:30 -06008746
8747 atomic_dec(&tctx->in_idle);
8748}
8749
8750static s64 tctx_inflight(struct io_uring_task *tctx)
8751{
8752 unsigned long index;
8753 struct file *file;
8754 s64 inflight;
8755
8756 inflight = percpu_counter_sum(&tctx->inflight);
8757 if (!tctx->sqpoll)
8758 return inflight;
8759
8760 /*
8761 * If we have SQPOLL rings, then we need to iterate and find them, and
8762 * add the pending count for those.
8763 */
8764 xa_for_each(&tctx->xa, index, file) {
8765 struct io_ring_ctx *ctx = file->private_data;
8766
8767 if (ctx->flags & IORING_SETUP_SQPOLL) {
8768 struct io_uring_task *__tctx = ctx->sqo_task->io_uring;
8769
8770 inflight += percpu_counter_sum(&__tctx->inflight);
8771 }
8772 }
8773
8774 return inflight;
Jens Axboe0f212202020-09-13 13:09:39 -06008775}
8776
Jens Axboe0f212202020-09-13 13:09:39 -06008777/*
8778 * Find any io_uring fd that this task has registered or done IO on, and cancel
8779 * requests.
8780 */
8781void __io_uring_task_cancel(void)
8782{
8783 struct io_uring_task *tctx = current->io_uring;
8784 DEFINE_WAIT(wait);
Jens Axboed8a6df12020-10-15 16:24:45 -06008785 s64 inflight;
Jens Axboe0f212202020-09-13 13:09:39 -06008786
8787 /* make sure overflow events are dropped */
Jens Axboefdaf0832020-10-30 09:37:30 -06008788 atomic_inc(&tctx->in_idle);
Jens Axboe0f212202020-09-13 13:09:39 -06008789
Jens Axboed8a6df12020-10-15 16:24:45 -06008790 do {
Jens Axboe0f212202020-09-13 13:09:39 -06008791 /* read completions before cancelations */
Jens Axboefdaf0832020-10-30 09:37:30 -06008792 inflight = tctx_inflight(tctx);
Jens Axboed8a6df12020-10-15 16:24:45 -06008793 if (!inflight)
8794 break;
Jens Axboe0f212202020-09-13 13:09:39 -06008795 __io_uring_files_cancel(NULL);
8796
8797 prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE);
8798
8799 /*
8800 * If we've seen completions, retry. This avoids a race where
8801 * a completion comes in before we did prepare_to_wait().
8802 */
Jens Axboefdaf0832020-10-30 09:37:30 -06008803 if (inflight != tctx_inflight(tctx))
Jens Axboe0f212202020-09-13 13:09:39 -06008804 continue;
Jens Axboe0f212202020-09-13 13:09:39 -06008805 schedule();
Jens Axboed8a6df12020-10-15 16:24:45 -06008806 } while (1);
Jens Axboe0f212202020-09-13 13:09:39 -06008807
8808 finish_wait(&tctx->wait, &wait);
Jens Axboefdaf0832020-10-30 09:37:30 -06008809 atomic_dec(&tctx->in_idle);
Jens Axboefcb323c2019-10-24 12:39:47 -06008810}
8811
8812static int io_uring_flush(struct file *file, void *data)
8813{
Pavel Begunkovc8fb20b2020-10-22 16:38:27 +01008814 io_uring_attempt_task_drop(file);
Jens Axboefcb323c2019-10-24 12:39:47 -06008815 return 0;
8816}
8817
Roman Penyaev6c5c2402019-11-28 12:53:22 +01008818static void *io_uring_validate_mmap_request(struct file *file,
8819 loff_t pgoff, size_t sz)
Jens Axboe2b188cc2019-01-07 10:46:33 -07008820{
Jens Axboe2b188cc2019-01-07 10:46:33 -07008821 struct io_ring_ctx *ctx = file->private_data;
Roman Penyaev6c5c2402019-11-28 12:53:22 +01008822 loff_t offset = pgoff << PAGE_SHIFT;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008823 struct page *page;
8824 void *ptr;
8825
8826 switch (offset) {
8827 case IORING_OFF_SQ_RING:
Hristo Venev75b28af2019-08-26 17:23:46 +00008828 case IORING_OFF_CQ_RING:
8829 ptr = ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008830 break;
8831 case IORING_OFF_SQES:
8832 ptr = ctx->sq_sqes;
8833 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008834 default:
Roman Penyaev6c5c2402019-11-28 12:53:22 +01008835 return ERR_PTR(-EINVAL);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008836 }
8837
8838 page = virt_to_head_page(ptr);
Matthew Wilcox (Oracle)a50b8542019-09-23 15:34:25 -07008839 if (sz > page_size(page))
Roman Penyaev6c5c2402019-11-28 12:53:22 +01008840 return ERR_PTR(-EINVAL);
8841
8842 return ptr;
8843}
8844
8845#ifdef CONFIG_MMU
8846
8847static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
8848{
8849 size_t sz = vma->vm_end - vma->vm_start;
8850 unsigned long pfn;
8851 void *ptr;
8852
8853 ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz);
8854 if (IS_ERR(ptr))
8855 return PTR_ERR(ptr);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008856
8857 pfn = virt_to_phys(ptr) >> PAGE_SHIFT;
8858 return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot);
8859}
8860
Roman Penyaev6c5c2402019-11-28 12:53:22 +01008861#else /* !CONFIG_MMU */
8862
8863static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
8864{
8865 return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL;
8866}
8867
8868static unsigned int io_uring_nommu_mmap_capabilities(struct file *file)
8869{
8870 return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE;
8871}
8872
8873static unsigned long io_uring_nommu_get_unmapped_area(struct file *file,
8874 unsigned long addr, unsigned long len,
8875 unsigned long pgoff, unsigned long flags)
8876{
8877 void *ptr;
8878
8879 ptr = io_uring_validate_mmap_request(file, pgoff, len);
8880 if (IS_ERR(ptr))
8881 return PTR_ERR(ptr);
8882
8883 return (unsigned long) ptr;
8884}
8885
8886#endif /* !CONFIG_MMU */
8887
Jens Axboe90554202020-09-03 12:12:41 -06008888static void io_sqpoll_wait_sq(struct io_ring_ctx *ctx)
8889{
8890 DEFINE_WAIT(wait);
8891
8892 do {
8893 if (!io_sqring_full(ctx))
8894 break;
8895
8896 prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE);
8897
8898 if (!io_sqring_full(ctx))
8899 break;
8900
8901 schedule();
8902 } while (!signal_pending(current));
8903
8904 finish_wait(&ctx->sqo_sq_wait, &wait);
8905}
8906
Jens Axboe2b188cc2019-01-07 10:46:33 -07008907SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
8908 u32, min_complete, u32, flags, const sigset_t __user *, sig,
8909 size_t, sigsz)
8910{
8911 struct io_ring_ctx *ctx;
8912 long ret = -EBADF;
8913 int submitted = 0;
8914 struct fd f;
8915
Jens Axboe4c6e2772020-07-01 11:29:10 -06008916 io_run_task_work();
Jens Axboeb41e9852020-02-17 09:52:41 -07008917
Jens Axboe90554202020-09-03 12:12:41 -06008918 if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP |
8919 IORING_ENTER_SQ_WAIT))
Jens Axboe2b188cc2019-01-07 10:46:33 -07008920 return -EINVAL;
8921
8922 f = fdget(fd);
8923 if (!f.file)
8924 return -EBADF;
8925
8926 ret = -EOPNOTSUPP;
8927 if (f.file->f_op != &io_uring_fops)
8928 goto out_fput;
8929
8930 ret = -ENXIO;
8931 ctx = f.file->private_data;
8932 if (!percpu_ref_tryget(&ctx->refs))
8933 goto out_fput;
8934
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02008935 ret = -EBADFD;
8936 if (ctx->flags & IORING_SETUP_R_DISABLED)
8937 goto out;
8938
Jens Axboe6c271ce2019-01-10 11:22:30 -07008939 /*
8940 * For SQ polling, the thread will do all submissions and completions.
8941 * Just return the requested submit count, and wake the thread if
8942 * we were asked to.
8943 */
Jens Axboeb2a9ead2019-09-12 14:19:16 -06008944 ret = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07008945 if (ctx->flags & IORING_SETUP_SQPOLL) {
Jens Axboec1edbf52019-11-10 16:56:04 -07008946 if (!list_empty_careful(&ctx->cq_overflow_list))
Jens Axboee6c8aa92020-09-28 13:10:13 -06008947 io_cqring_overflow_flush(ctx, false, NULL, NULL);
Jens Axboe6c271ce2019-01-10 11:22:30 -07008948 if (flags & IORING_ENTER_SQ_WAKEUP)
Jens Axboe534ca6d2020-09-02 13:52:19 -06008949 wake_up(&ctx->sq_data->wait);
Jens Axboe90554202020-09-03 12:12:41 -06008950 if (flags & IORING_ENTER_SQ_WAIT)
8951 io_sqpoll_wait_sq(ctx);
Jens Axboe6c271ce2019-01-10 11:22:30 -07008952 submitted = to_submit;
Jens Axboeb2a9ead2019-09-12 14:19:16 -06008953 } else if (to_submit) {
Jens Axboefdaf0832020-10-30 09:37:30 -06008954 ret = io_uring_add_task_file(ctx, f.file);
Jens Axboe0f212202020-09-13 13:09:39 -06008955 if (unlikely(ret))
8956 goto out;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008957 mutex_lock(&ctx->uring_lock);
Jens Axboe0f212202020-09-13 13:09:39 -06008958 submitted = io_submit_sqes(ctx, to_submit);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008959 mutex_unlock(&ctx->uring_lock);
Pavel Begunkov7c504e652019-12-18 19:53:45 +03008960
8961 if (submitted != to_submit)
8962 goto out;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008963 }
8964 if (flags & IORING_ENTER_GETEVENTS) {
8965 min_complete = min(min_complete, ctx->cq_entries);
8966
Xiaoguang Wang32b22442020-03-11 09:26:09 +08008967 /*
8968 * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user
8969 * space applications don't need to do io completion events
8970 * polling again, they can rely on io_sq_thread to do polling
8971 * work, which can reduce cpu usage and uring_lock contention.
8972 */
8973 if (ctx->flags & IORING_SETUP_IOPOLL &&
8974 !(ctx->flags & IORING_SETUP_SQPOLL)) {
Pavel Begunkov7668b922020-07-07 16:36:21 +03008975 ret = io_iopoll_check(ctx, min_complete);
Jens Axboedef596e2019-01-09 08:59:42 -07008976 } else {
8977 ret = io_cqring_wait(ctx, min_complete, sig, sigsz);
8978 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07008979 }
8980
Pavel Begunkov7c504e652019-12-18 19:53:45 +03008981out:
Pavel Begunkov6805b32e2019-10-08 02:18:42 +03008982 percpu_ref_put(&ctx->refs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008983out_fput:
8984 fdput(f);
8985 return submitted ? submitted : ret;
8986}
8987
Tobias Klauserbebdb652020-02-26 18:38:32 +01008988#ifdef CONFIG_PROC_FS
Jens Axboe87ce9552020-01-30 08:25:34 -07008989static int io_uring_show_cred(int id, void *p, void *data)
8990{
Jens Axboe6b47ab82020-11-05 09:50:16 -07008991 struct io_identity *iod = p;
8992 const struct cred *cred = iod->creds;
Jens Axboe87ce9552020-01-30 08:25:34 -07008993 struct seq_file *m = data;
8994 struct user_namespace *uns = seq_user_ns(m);
8995 struct group_info *gi;
8996 kernel_cap_t cap;
8997 unsigned __capi;
8998 int g;
8999
9000 seq_printf(m, "%5d\n", id);
9001 seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid));
9002 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid));
9003 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid));
9004 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid));
9005 seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid));
9006 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid));
9007 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid));
9008 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid));
9009 seq_puts(m, "\n\tGroups:\t");
9010 gi = cred->group_info;
9011 for (g = 0; g < gi->ngroups; g++) {
9012 seq_put_decimal_ull(m, g ? " " : "",
9013 from_kgid_munged(uns, gi->gid[g]));
9014 }
9015 seq_puts(m, "\n\tCapEff:\t");
9016 cap = cred->cap_effective;
9017 CAP_FOR_EACH_U32(__capi)
9018 seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8);
9019 seq_putc(m, '\n');
9020 return 0;
9021}
9022
9023static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m)
9024{
Joseph Qidbbe9c62020-09-29 09:01:22 -06009025 struct io_sq_data *sq = NULL;
Jens Axboefad8e0d2020-09-28 08:57:48 -06009026 bool has_lock;
Jens Axboe87ce9552020-01-30 08:25:34 -07009027 int i;
9028
Jens Axboefad8e0d2020-09-28 08:57:48 -06009029 /*
9030 * Avoid ABBA deadlock between the seq lock and the io_uring mutex,
9031 * since fdinfo case grabs it in the opposite direction of normal use
9032 * cases. If we fail to get the lock, we just don't iterate any
9033 * structures that could be going away outside the io_uring mutex.
9034 */
9035 has_lock = mutex_trylock(&ctx->uring_lock);
9036
Joseph Qidbbe9c62020-09-29 09:01:22 -06009037 if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL))
9038 sq = ctx->sq_data;
9039
9040 seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1);
9041 seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1);
Jens Axboe87ce9552020-01-30 08:25:34 -07009042 seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009043 for (i = 0; has_lock && i < ctx->nr_user_files; i++) {
Jens Axboe87ce9552020-01-30 08:25:34 -07009044 struct fixed_file_table *table;
9045 struct file *f;
9046
9047 table = &ctx->file_data->table[i >> IORING_FILE_TABLE_SHIFT];
9048 f = table->files[i & IORING_FILE_TABLE_MASK];
9049 if (f)
9050 seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname);
9051 else
9052 seq_printf(m, "%5u: <none>\n", i);
9053 }
9054 seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009055 for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) {
Jens Axboe87ce9552020-01-30 08:25:34 -07009056 struct io_mapped_ubuf *buf = &ctx->user_bufs[i];
9057
9058 seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf,
9059 (unsigned int) buf->len);
9060 }
Jens Axboefad8e0d2020-09-28 08:57:48 -06009061 if (has_lock && !idr_is_empty(&ctx->personality_idr)) {
Jens Axboe87ce9552020-01-30 08:25:34 -07009062 seq_printf(m, "Personalities:\n");
9063 idr_for_each(&ctx->personality_idr, io_uring_show_cred, m);
9064 }
Jens Axboed7718a92020-02-14 22:23:12 -07009065 seq_printf(m, "PollList:\n");
9066 spin_lock_irq(&ctx->completion_lock);
9067 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
9068 struct hlist_head *list = &ctx->cancel_hash[i];
9069 struct io_kiocb *req;
9070
9071 hlist_for_each_entry(req, list, hash_node)
9072 seq_printf(m, " op=%d, task_works=%d\n", req->opcode,
9073 req->task->task_works != NULL);
9074 }
9075 spin_unlock_irq(&ctx->completion_lock);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009076 if (has_lock)
9077 mutex_unlock(&ctx->uring_lock);
Jens Axboe87ce9552020-01-30 08:25:34 -07009078}
9079
9080static void io_uring_show_fdinfo(struct seq_file *m, struct file *f)
9081{
9082 struct io_ring_ctx *ctx = f->private_data;
9083
9084 if (percpu_ref_tryget(&ctx->refs)) {
9085 __io_uring_show_fdinfo(ctx, m);
9086 percpu_ref_put(&ctx->refs);
9087 }
9088}
Tobias Klauserbebdb652020-02-26 18:38:32 +01009089#endif
Jens Axboe87ce9552020-01-30 08:25:34 -07009090
Jens Axboe2b188cc2019-01-07 10:46:33 -07009091static const struct file_operations io_uring_fops = {
9092 .release = io_uring_release,
Jens Axboefcb323c2019-10-24 12:39:47 -06009093 .flush = io_uring_flush,
Jens Axboe2b188cc2019-01-07 10:46:33 -07009094 .mmap = io_uring_mmap,
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009095#ifndef CONFIG_MMU
9096 .get_unmapped_area = io_uring_nommu_get_unmapped_area,
9097 .mmap_capabilities = io_uring_nommu_mmap_capabilities,
9098#endif
Jens Axboe2b188cc2019-01-07 10:46:33 -07009099 .poll = io_uring_poll,
9100 .fasync = io_uring_fasync,
Tobias Klauserbebdb652020-02-26 18:38:32 +01009101#ifdef CONFIG_PROC_FS
Jens Axboe87ce9552020-01-30 08:25:34 -07009102 .show_fdinfo = io_uring_show_fdinfo,
Tobias Klauserbebdb652020-02-26 18:38:32 +01009103#endif
Jens Axboe2b188cc2019-01-07 10:46:33 -07009104};
9105
9106static int io_allocate_scq_urings(struct io_ring_ctx *ctx,
9107 struct io_uring_params *p)
9108{
Hristo Venev75b28af2019-08-26 17:23:46 +00009109 struct io_rings *rings;
9110 size_t size, sq_array_offset;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009111
Jens Axboebd740482020-08-05 12:58:23 -06009112 /* make sure these are sane, as we already accounted them */
9113 ctx->sq_entries = p->sq_entries;
9114 ctx->cq_entries = p->cq_entries;
9115
Hristo Venev75b28af2019-08-26 17:23:46 +00009116 size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset);
9117 if (size == SIZE_MAX)
9118 return -EOVERFLOW;
9119
9120 rings = io_mem_alloc(size);
9121 if (!rings)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009122 return -ENOMEM;
9123
Hristo Venev75b28af2019-08-26 17:23:46 +00009124 ctx->rings = rings;
9125 ctx->sq_array = (u32 *)((char *)rings + sq_array_offset);
9126 rings->sq_ring_mask = p->sq_entries - 1;
9127 rings->cq_ring_mask = p->cq_entries - 1;
9128 rings->sq_ring_entries = p->sq_entries;
9129 rings->cq_ring_entries = p->cq_entries;
9130 ctx->sq_mask = rings->sq_ring_mask;
9131 ctx->cq_mask = rings->cq_ring_mask;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009132
9133 size = array_size(sizeof(struct io_uring_sqe), p->sq_entries);
Jens Axboeeb065d32019-11-20 09:26:29 -07009134 if (size == SIZE_MAX) {
9135 io_mem_free(ctx->rings);
9136 ctx->rings = NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009137 return -EOVERFLOW;
Jens Axboeeb065d32019-11-20 09:26:29 -07009138 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009139
9140 ctx->sq_sqes = io_mem_alloc(size);
Jens Axboeeb065d32019-11-20 09:26:29 -07009141 if (!ctx->sq_sqes) {
9142 io_mem_free(ctx->rings);
9143 ctx->rings = NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009144 return -ENOMEM;
Jens Axboeeb065d32019-11-20 09:26:29 -07009145 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009146
Jens Axboe2b188cc2019-01-07 10:46:33 -07009147 return 0;
9148}
9149
9150/*
9151 * Allocate an anonymous fd, this is what constitutes the application
9152 * visible backing of an io_uring instance. The application mmaps this
9153 * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled,
9154 * we have to tie this fd to a socket for file garbage collection purposes.
9155 */
9156static int io_uring_get_fd(struct io_ring_ctx *ctx)
9157{
9158 struct file *file;
9159 int ret;
9160
9161#if defined(CONFIG_UNIX)
9162 ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP,
9163 &ctx->ring_sock);
9164 if (ret)
9165 return ret;
9166#endif
9167
9168 ret = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
9169 if (ret < 0)
9170 goto err;
9171
9172 file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx,
9173 O_RDWR | O_CLOEXEC);
9174 if (IS_ERR(file)) {
Jens Axboe0f212202020-09-13 13:09:39 -06009175err_fd:
Jens Axboe2b188cc2019-01-07 10:46:33 -07009176 put_unused_fd(ret);
9177 ret = PTR_ERR(file);
9178 goto err;
9179 }
9180
9181#if defined(CONFIG_UNIX)
9182 ctx->ring_sock->file = file;
9183#endif
Jens Axboefdaf0832020-10-30 09:37:30 -06009184 if (unlikely(io_uring_add_task_file(ctx, file))) {
Jens Axboe0f212202020-09-13 13:09:39 -06009185 file = ERR_PTR(-ENOMEM);
9186 goto err_fd;
9187 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009188 fd_install(ret, file);
9189 return ret;
9190err:
9191#if defined(CONFIG_UNIX)
9192 sock_release(ctx->ring_sock);
9193 ctx->ring_sock = NULL;
9194#endif
9195 return ret;
9196}
9197
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009198static int io_uring_create(unsigned entries, struct io_uring_params *p,
9199 struct io_uring_params __user *params)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009200{
9201 struct user_struct *user = NULL;
9202 struct io_ring_ctx *ctx;
Bijan Mottahedehaad5d8d2020-06-16 16:36:08 -07009203 bool limit_mem;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009204 int ret;
9205
Jens Axboe8110c1a2019-12-28 15:39:54 -07009206 if (!entries)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009207 return -EINVAL;
Jens Axboe8110c1a2019-12-28 15:39:54 -07009208 if (entries > IORING_MAX_ENTRIES) {
9209 if (!(p->flags & IORING_SETUP_CLAMP))
9210 return -EINVAL;
9211 entries = IORING_MAX_ENTRIES;
9212 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009213
9214 /*
9215 * Use twice as many entries for the CQ ring. It's possible for the
9216 * application to drive a higher depth than the size of the SQ ring,
9217 * since the sqes are only used at submission time. This allows for
Jens Axboe33a107f2019-10-04 12:10:03 -06009218 * some flexibility in overcommitting a bit. If the application has
9219 * set IORING_SETUP_CQSIZE, it will have passed in the desired number
9220 * of CQ ring entries manually.
Jens Axboe2b188cc2019-01-07 10:46:33 -07009221 */
9222 p->sq_entries = roundup_pow_of_two(entries);
Jens Axboe33a107f2019-10-04 12:10:03 -06009223 if (p->flags & IORING_SETUP_CQSIZE) {
9224 /*
9225 * If IORING_SETUP_CQSIZE is set, we do the same roundup
9226 * to a power-of-two, if it isn't already. We do NOT impose
9227 * any cq vs sq ring sizing.
9228 */
Jens Axboe8110c1a2019-12-28 15:39:54 -07009229 if (p->cq_entries < p->sq_entries)
Jens Axboe33a107f2019-10-04 12:10:03 -06009230 return -EINVAL;
Jens Axboe8110c1a2019-12-28 15:39:54 -07009231 if (p->cq_entries > IORING_MAX_CQ_ENTRIES) {
9232 if (!(p->flags & IORING_SETUP_CLAMP))
9233 return -EINVAL;
9234 p->cq_entries = IORING_MAX_CQ_ENTRIES;
9235 }
Jens Axboe33a107f2019-10-04 12:10:03 -06009236 p->cq_entries = roundup_pow_of_two(p->cq_entries);
9237 } else {
9238 p->cq_entries = 2 * p->sq_entries;
9239 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009240
9241 user = get_uid(current_user());
Bijan Mottahedehaad5d8d2020-06-16 16:36:08 -07009242 limit_mem = !capable(CAP_IPC_LOCK);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009243
Bijan Mottahedehaad5d8d2020-06-16 16:36:08 -07009244 if (limit_mem) {
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07009245 ret = __io_account_mem(user,
Jens Axboe2b188cc2019-01-07 10:46:33 -07009246 ring_pages(p->sq_entries, p->cq_entries));
9247 if (ret) {
9248 free_uid(user);
9249 return ret;
9250 }
9251 }
9252
9253 ctx = io_ring_ctx_alloc(p);
9254 if (!ctx) {
Bijan Mottahedehaad5d8d2020-06-16 16:36:08 -07009255 if (limit_mem)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07009256 __io_unaccount_mem(user, ring_pages(p->sq_entries,
Jens Axboe2b188cc2019-01-07 10:46:33 -07009257 p->cq_entries));
9258 free_uid(user);
9259 return -ENOMEM;
9260 }
9261 ctx->compat = in_compat_syscall();
Jens Axboe2b188cc2019-01-07 10:46:33 -07009262 ctx->user = user;
Jens Axboe0b8c0ec2019-12-02 08:50:00 -07009263 ctx->creds = get_current_cred();
Jens Axboe4ea33a92020-10-15 13:46:44 -06009264#ifdef CONFIG_AUDIT
9265 ctx->loginuid = current->loginuid;
9266 ctx->sessionid = current->sessionid;
9267#endif
Jens Axboe2aede0e2020-09-14 10:45:53 -06009268 ctx->sqo_task = get_task_struct(current);
9269
9270 /*
9271 * This is just grabbed for accounting purposes. When a process exits,
9272 * the mm is exited and dropped before the files, hence we need to hang
9273 * on to this mm purely for the purposes of being able to unaccount
9274 * memory (locked/pinned vm). It's not used for anything else.
9275 */
Jens Axboe6b7898e2020-08-25 07:58:00 -06009276 mmgrab(current->mm);
Jens Axboe2aede0e2020-09-14 10:45:53 -06009277 ctx->mm_account = current->mm;
Jens Axboe6b7898e2020-08-25 07:58:00 -06009278
Dennis Zhou91d8f512020-09-16 13:41:05 -07009279#ifdef CONFIG_BLK_CGROUP
9280 /*
9281 * The sq thread will belong to the original cgroup it was inited in.
9282 * If the cgroup goes offline (e.g. disabling the io controller), then
9283 * issued bios will be associated with the closest cgroup later in the
9284 * block layer.
9285 */
9286 rcu_read_lock();
9287 ctx->sqo_blkcg_css = blkcg_css();
9288 ret = css_tryget_online(ctx->sqo_blkcg_css);
9289 rcu_read_unlock();
9290 if (!ret) {
9291 /* don't init against a dying cgroup, have the user try again */
9292 ctx->sqo_blkcg_css = NULL;
9293 ret = -ENODEV;
9294 goto err;
9295 }
9296#endif
Jens Axboe6c271ce2019-01-10 11:22:30 -07009297
Jens Axboe2b188cc2019-01-07 10:46:33 -07009298 /*
9299 * Account memory _before_ installing the file descriptor. Once
9300 * the descriptor is installed, it can get closed at any time. Also
Jens Axboe2b188cc2019-01-07 10:46:33 -07009301 * do this before hitting the general error path, as ring freeing
Hristo Venev75b28af2019-08-26 17:23:46 +00009302 * will un-account as well.
9303 */
9304 io_account_mem(ctx, ring_pages(p->sq_entries, p->cq_entries),
9305 ACCT_LOCKED);
9306 ctx->limit_mem = limit_mem;
9307
9308 ret = io_allocate_scq_urings(ctx, p);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009309 if (ret)
9310 goto err;
Hristo Venev75b28af2019-08-26 17:23:46 +00009311
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009312 ret = io_sq_offload_create(ctx, p);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009313 if (ret)
9314 goto err;
9315
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009316 if (!(p->flags & IORING_SETUP_R_DISABLED))
9317 io_sq_offload_start(ctx);
9318
Jens Axboe2b188cc2019-01-07 10:46:33 -07009319 memset(&p->sq_off, 0, sizeof(p->sq_off));
9320 p->sq_off.head = offsetof(struct io_rings, sq.head);
9321 p->sq_off.tail = offsetof(struct io_rings, sq.tail);
9322 p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask);
9323 p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries);
9324 p->sq_off.flags = offsetof(struct io_rings, sq_flags);
9325 p->sq_off.dropped = offsetof(struct io_rings, sq_dropped);
9326 p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings;
9327
9328 memset(&p->cq_off, 0, sizeof(p->cq_off));
Hristo Venev75b28af2019-08-26 17:23:46 +00009329 p->cq_off.head = offsetof(struct io_rings, cq.head);
9330 p->cq_off.tail = offsetof(struct io_rings, cq.tail);
9331 p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask);
9332 p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries);
9333 p->cq_off.overflow = offsetof(struct io_rings, cq_overflow);
9334 p->cq_off.cqes = offsetof(struct io_rings, cqes);
Stefano Garzarella0d9b5b32020-05-15 18:38:04 +02009335 p->cq_off.flags = offsetof(struct io_rings, cq_flags);
Jens Axboeac90f242019-09-06 10:26:21 -06009336
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009337 p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP |
9338 IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS |
Jiufei Xue5769a352020-06-17 17:53:55 +08009339 IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL |
9340 IORING_FEAT_POLL_32BITS;
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009341
9342 if (copy_to_user(params, p, sizeof(*p))) {
9343 ret = -EFAULT;
9344 goto err;
9345 }
Jens Axboed1719f72020-07-30 13:43:53 -06009346
9347 /*
Jens Axboe044c1ab2019-10-28 09:15:33 -06009348 * Install ring fd as the very last thing, so we don't risk someone
9349 * having closed it before we finish setup
9350 */
9351 ret = io_uring_get_fd(ctx);
9352 if (ret < 0)
9353 goto err;
9354
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02009355 trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009356 return ret;
9357err:
9358 io_ring_ctx_wait_and_kill(ctx);
9359 return ret;
9360}
9361
9362/*
9363 * Sets up an aio uring context, and returns the fd. Applications asks for a
9364 * ring size, we return the actual sq/cq ring sizes (among other things) in the
9365 * params structure passed in.
9366 */
9367static long io_uring_setup(u32 entries, struct io_uring_params __user *params)
9368{
9369 struct io_uring_params p;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009370 int i;
9371
9372 if (copy_from_user(&p, params, sizeof(p)))
9373 return -EFAULT;
9374 for (i = 0; i < ARRAY_SIZE(p.resv); i++) {
9375 if (p.resv[i])
9376 return -EINVAL;
9377 }
9378
Jens Axboe6c271ce2019-01-10 11:22:30 -07009379 if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL |
Jens Axboe8110c1a2019-12-28 15:39:54 -07009380 IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE |
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009381 IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ |
9382 IORING_SETUP_R_DISABLED))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009383 return -EINVAL;
9384
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009385 return io_uring_create(entries, &p, params);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009386}
9387
9388SYSCALL_DEFINE2(io_uring_setup, u32, entries,
9389 struct io_uring_params __user *, params)
9390{
9391 return io_uring_setup(entries, params);
9392}
9393
Jens Axboe66f4af92020-01-16 15:36:52 -07009394static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args)
9395{
9396 struct io_uring_probe *p;
9397 size_t size;
9398 int i, ret;
9399
9400 size = struct_size(p, ops, nr_args);
9401 if (size == SIZE_MAX)
9402 return -EOVERFLOW;
9403 p = kzalloc(size, GFP_KERNEL);
9404 if (!p)
9405 return -ENOMEM;
9406
9407 ret = -EFAULT;
9408 if (copy_from_user(p, arg, size))
9409 goto out;
9410 ret = -EINVAL;
9411 if (memchr_inv(p, 0, size))
9412 goto out;
9413
9414 p->last_op = IORING_OP_LAST - 1;
9415 if (nr_args > IORING_OP_LAST)
9416 nr_args = IORING_OP_LAST;
9417
9418 for (i = 0; i < nr_args; i++) {
9419 p->ops[i].op = i;
9420 if (!io_op_defs[i].not_supported)
9421 p->ops[i].flags = IO_URING_OP_SUPPORTED;
9422 }
9423 p->ops_len = i;
9424
9425 ret = 0;
9426 if (copy_to_user(arg, p, size))
9427 ret = -EFAULT;
9428out:
9429 kfree(p);
9430 return ret;
9431}
9432
Jens Axboe071698e2020-01-28 10:04:42 -07009433static int io_register_personality(struct io_ring_ctx *ctx)
9434{
Jens Axboe1e6fa522020-10-15 08:46:24 -06009435 struct io_identity *id;
9436 int ret;
Jens Axboe071698e2020-01-28 10:04:42 -07009437
Jens Axboe1e6fa522020-10-15 08:46:24 -06009438 id = kmalloc(sizeof(*id), GFP_KERNEL);
9439 if (unlikely(!id))
9440 return -ENOMEM;
9441
9442 io_init_identity(id);
9443 id->creds = get_current_cred();
9444
9445 ret = idr_alloc_cyclic(&ctx->personality_idr, id, 1, USHRT_MAX, GFP_KERNEL);
9446 if (ret < 0) {
9447 put_cred(id->creds);
9448 kfree(id);
9449 }
9450 return ret;
Jens Axboe071698e2020-01-28 10:04:42 -07009451}
9452
9453static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id)
9454{
Jens Axboe1e6fa522020-10-15 08:46:24 -06009455 struct io_identity *iod;
Jens Axboe071698e2020-01-28 10:04:42 -07009456
Jens Axboe1e6fa522020-10-15 08:46:24 -06009457 iod = idr_remove(&ctx->personality_idr, id);
9458 if (iod) {
9459 put_cred(iod->creds);
9460 if (refcount_dec_and_test(&iod->count))
9461 kfree(iod);
Jens Axboe071698e2020-01-28 10:04:42 -07009462 return 0;
9463 }
9464
9465 return -EINVAL;
9466}
9467
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009468static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg,
9469 unsigned int nr_args)
9470{
9471 struct io_uring_restriction *res;
9472 size_t size;
9473 int i, ret;
9474
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009475 /* Restrictions allowed only if rings started disabled */
9476 if (!(ctx->flags & IORING_SETUP_R_DISABLED))
9477 return -EBADFD;
9478
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009479 /* We allow only a single restrictions registration */
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009480 if (ctx->restrictions.registered)
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009481 return -EBUSY;
9482
9483 if (!arg || nr_args > IORING_MAX_RESTRICTIONS)
9484 return -EINVAL;
9485
9486 size = array_size(nr_args, sizeof(*res));
9487 if (size == SIZE_MAX)
9488 return -EOVERFLOW;
9489
9490 res = memdup_user(arg, size);
9491 if (IS_ERR(res))
9492 return PTR_ERR(res);
9493
9494 ret = 0;
9495
9496 for (i = 0; i < nr_args; i++) {
9497 switch (res[i].opcode) {
9498 case IORING_RESTRICTION_REGISTER_OP:
9499 if (res[i].register_op >= IORING_REGISTER_LAST) {
9500 ret = -EINVAL;
9501 goto out;
9502 }
9503
9504 __set_bit(res[i].register_op,
9505 ctx->restrictions.register_op);
9506 break;
9507 case IORING_RESTRICTION_SQE_OP:
9508 if (res[i].sqe_op >= IORING_OP_LAST) {
9509 ret = -EINVAL;
9510 goto out;
9511 }
9512
9513 __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op);
9514 break;
9515 case IORING_RESTRICTION_SQE_FLAGS_ALLOWED:
9516 ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags;
9517 break;
9518 case IORING_RESTRICTION_SQE_FLAGS_REQUIRED:
9519 ctx->restrictions.sqe_flags_required = res[i].sqe_flags;
9520 break;
9521 default:
9522 ret = -EINVAL;
9523 goto out;
9524 }
9525 }
9526
9527out:
9528 /* Reset all restrictions if an error happened */
9529 if (ret != 0)
9530 memset(&ctx->restrictions, 0, sizeof(ctx->restrictions));
9531 else
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009532 ctx->restrictions.registered = true;
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009533
9534 kfree(res);
9535 return ret;
9536}
9537
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009538static int io_register_enable_rings(struct io_ring_ctx *ctx)
9539{
9540 if (!(ctx->flags & IORING_SETUP_R_DISABLED))
9541 return -EBADFD;
9542
9543 if (ctx->restrictions.registered)
9544 ctx->restricted = 1;
9545
9546 ctx->flags &= ~IORING_SETUP_R_DISABLED;
9547
9548 io_sq_offload_start(ctx);
9549
9550 return 0;
9551}
9552
Jens Axboe071698e2020-01-28 10:04:42 -07009553static bool io_register_op_must_quiesce(int op)
9554{
9555 switch (op) {
9556 case IORING_UNREGISTER_FILES:
9557 case IORING_REGISTER_FILES_UPDATE:
9558 case IORING_REGISTER_PROBE:
9559 case IORING_REGISTER_PERSONALITY:
9560 case IORING_UNREGISTER_PERSONALITY:
9561 return false;
9562 default:
9563 return true;
9564 }
9565}
9566
Jens Axboeedafcce2019-01-09 09:16:05 -07009567static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode,
9568 void __user *arg, unsigned nr_args)
Jens Axboeb19062a2019-04-15 10:49:38 -06009569 __releases(ctx->uring_lock)
9570 __acquires(ctx->uring_lock)
Jens Axboeedafcce2019-01-09 09:16:05 -07009571{
9572 int ret;
9573
Jens Axboe35fa71a2019-04-22 10:23:23 -06009574 /*
9575 * We're inside the ring mutex, if the ref is already dying, then
9576 * someone else killed the ctx or is already going through
9577 * io_uring_register().
9578 */
9579 if (percpu_ref_is_dying(&ctx->refs))
9580 return -ENXIO;
9581
Jens Axboe071698e2020-01-28 10:04:42 -07009582 if (io_register_op_must_quiesce(opcode)) {
Jens Axboe05f3fb32019-12-09 11:22:50 -07009583 percpu_ref_kill(&ctx->refs);
Jens Axboeb19062a2019-04-15 10:49:38 -06009584
Jens Axboe05f3fb32019-12-09 11:22:50 -07009585 /*
9586 * Drop uring mutex before waiting for references to exit. If
9587 * another thread is currently inside io_uring_enter() it might
9588 * need to grab the uring_lock to make progress. If we hold it
9589 * here across the drain wait, then we can deadlock. It's safe
9590 * to drop the mutex here, since no new references will come in
9591 * after we've killed the percpu ref.
9592 */
9593 mutex_unlock(&ctx->uring_lock);
Jens Axboeaf9c1a42020-09-24 13:32:18 -06009594 do {
9595 ret = wait_for_completion_interruptible(&ctx->ref_comp);
9596 if (!ret)
9597 break;
Jens Axboeed6930c2020-10-08 19:09:46 -06009598 ret = io_run_task_work_sig();
9599 if (ret < 0)
9600 break;
Jens Axboeaf9c1a42020-09-24 13:32:18 -06009601 } while (1);
9602
Jens Axboe05f3fb32019-12-09 11:22:50 -07009603 mutex_lock(&ctx->uring_lock);
Jens Axboeaf9c1a42020-09-24 13:32:18 -06009604
Jens Axboec1503682020-01-08 08:26:07 -07009605 if (ret) {
9606 percpu_ref_resurrect(&ctx->refs);
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009607 goto out_quiesce;
9608 }
9609 }
9610
9611 if (ctx->restricted) {
9612 if (opcode >= IORING_REGISTER_LAST) {
9613 ret = -EINVAL;
9614 goto out;
9615 }
9616
9617 if (!test_bit(opcode, ctx->restrictions.register_op)) {
9618 ret = -EACCES;
Jens Axboec1503682020-01-08 08:26:07 -07009619 goto out;
9620 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07009621 }
Jens Axboeedafcce2019-01-09 09:16:05 -07009622
9623 switch (opcode) {
9624 case IORING_REGISTER_BUFFERS:
9625 ret = io_sqe_buffer_register(ctx, arg, nr_args);
9626 break;
9627 case IORING_UNREGISTER_BUFFERS:
9628 ret = -EINVAL;
9629 if (arg || nr_args)
9630 break;
9631 ret = io_sqe_buffer_unregister(ctx);
9632 break;
Jens Axboe6b063142019-01-10 22:13:58 -07009633 case IORING_REGISTER_FILES:
9634 ret = io_sqe_files_register(ctx, arg, nr_args);
9635 break;
9636 case IORING_UNREGISTER_FILES:
9637 ret = -EINVAL;
9638 if (arg || nr_args)
9639 break;
9640 ret = io_sqe_files_unregister(ctx);
9641 break;
Jens Axboec3a31e62019-10-03 13:59:56 -06009642 case IORING_REGISTER_FILES_UPDATE:
9643 ret = io_sqe_files_update(ctx, arg, nr_args);
9644 break;
Jens Axboe9b402842019-04-11 11:45:41 -06009645 case IORING_REGISTER_EVENTFD:
Jens Axboef2842ab2020-01-08 11:04:00 -07009646 case IORING_REGISTER_EVENTFD_ASYNC:
Jens Axboe9b402842019-04-11 11:45:41 -06009647 ret = -EINVAL;
9648 if (nr_args != 1)
9649 break;
9650 ret = io_eventfd_register(ctx, arg);
Jens Axboef2842ab2020-01-08 11:04:00 -07009651 if (ret)
9652 break;
9653 if (opcode == IORING_REGISTER_EVENTFD_ASYNC)
9654 ctx->eventfd_async = 1;
9655 else
9656 ctx->eventfd_async = 0;
Jens Axboe9b402842019-04-11 11:45:41 -06009657 break;
9658 case IORING_UNREGISTER_EVENTFD:
9659 ret = -EINVAL;
9660 if (arg || nr_args)
9661 break;
9662 ret = io_eventfd_unregister(ctx);
9663 break;
Jens Axboe66f4af92020-01-16 15:36:52 -07009664 case IORING_REGISTER_PROBE:
9665 ret = -EINVAL;
9666 if (!arg || nr_args > 256)
9667 break;
9668 ret = io_probe(ctx, arg, nr_args);
9669 break;
Jens Axboe071698e2020-01-28 10:04:42 -07009670 case IORING_REGISTER_PERSONALITY:
9671 ret = -EINVAL;
9672 if (arg || nr_args)
9673 break;
9674 ret = io_register_personality(ctx);
9675 break;
9676 case IORING_UNREGISTER_PERSONALITY:
9677 ret = -EINVAL;
9678 if (arg)
9679 break;
9680 ret = io_unregister_personality(ctx, nr_args);
9681 break;
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009682 case IORING_REGISTER_ENABLE_RINGS:
9683 ret = -EINVAL;
9684 if (arg || nr_args)
9685 break;
9686 ret = io_register_enable_rings(ctx);
9687 break;
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009688 case IORING_REGISTER_RESTRICTIONS:
9689 ret = io_register_restrictions(ctx, arg, nr_args);
9690 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07009691 default:
9692 ret = -EINVAL;
9693 break;
9694 }
9695
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009696out:
Jens Axboe071698e2020-01-28 10:04:42 -07009697 if (io_register_op_must_quiesce(opcode)) {
Jens Axboe05f3fb32019-12-09 11:22:50 -07009698 /* bring the ctx back to life */
Jens Axboe05f3fb32019-12-09 11:22:50 -07009699 percpu_ref_reinit(&ctx->refs);
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009700out_quiesce:
Jens Axboe0f158b42020-05-14 17:18:39 -06009701 reinit_completion(&ctx->ref_comp);
Jens Axboe05f3fb32019-12-09 11:22:50 -07009702 }
Jens Axboeedafcce2019-01-09 09:16:05 -07009703 return ret;
9704}
9705
9706SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode,
9707 void __user *, arg, unsigned int, nr_args)
9708{
9709 struct io_ring_ctx *ctx;
9710 long ret = -EBADF;
9711 struct fd f;
9712
9713 f = fdget(fd);
9714 if (!f.file)
9715 return -EBADF;
9716
9717 ret = -EOPNOTSUPP;
9718 if (f.file->f_op != &io_uring_fops)
9719 goto out_fput;
9720
9721 ctx = f.file->private_data;
9722
9723 mutex_lock(&ctx->uring_lock);
9724 ret = __io_uring_register(ctx, opcode, arg, nr_args);
9725 mutex_unlock(&ctx->uring_lock);
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02009726 trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs,
9727 ctx->cq_ev_fd != NULL, ret);
Jens Axboeedafcce2019-01-09 09:16:05 -07009728out_fput:
9729 fdput(f);
9730 return ret;
9731}
9732
Jens Axboe2b188cc2019-01-07 10:46:33 -07009733static int __init io_uring_init(void)
9734{
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009735#define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \
9736 BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \
9737 BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \
9738} while (0)
9739
9740#define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \
9741 __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename)
9742 BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64);
9743 BUILD_BUG_SQE_ELEM(0, __u8, opcode);
9744 BUILD_BUG_SQE_ELEM(1, __u8, flags);
9745 BUILD_BUG_SQE_ELEM(2, __u16, ioprio);
9746 BUILD_BUG_SQE_ELEM(4, __s32, fd);
9747 BUILD_BUG_SQE_ELEM(8, __u64, off);
9748 BUILD_BUG_SQE_ELEM(8, __u64, addr2);
9749 BUILD_BUG_SQE_ELEM(16, __u64, addr);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03009750 BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009751 BUILD_BUG_SQE_ELEM(24, __u32, len);
9752 BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags);
9753 BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags);
9754 BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags);
9755 BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags);
Jiufei Xue5769a352020-06-17 17:53:55 +08009756 BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events);
9757 BUILD_BUG_SQE_ELEM(28, __u32, poll32_events);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009758 BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags);
9759 BUILD_BUG_SQE_ELEM(28, __u32, msg_flags);
9760 BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags);
9761 BUILD_BUG_SQE_ELEM(28, __u32, accept_flags);
9762 BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags);
9763 BUILD_BUG_SQE_ELEM(28, __u32, open_flags);
9764 BUILD_BUG_SQE_ELEM(28, __u32, statx_flags);
9765 BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03009766 BUILD_BUG_SQE_ELEM(28, __u32, splice_flags);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009767 BUILD_BUG_SQE_ELEM(32, __u64, user_data);
9768 BUILD_BUG_SQE_ELEM(40, __u16, buf_index);
9769 BUILD_BUG_SQE_ELEM(42, __u16, personality);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03009770 BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +01009771
Jens Axboed3656342019-12-18 09:50:26 -07009772 BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST);
Jens Axboe84557872020-03-03 15:28:17 -07009773 BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int));
Jens Axboe2b188cc2019-01-07 10:46:33 -07009774 req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC);
9775 return 0;
9776};
9777__initcall(io_uring_init);