blob: 5c0b1a7dba805311612d381b01de5b2bdda1be77 [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
Pavel Begunkov45d189c2021-02-10 00:03:07 +0000190enum io_uring_cmd_flags {
191 IO_URING_F_NONBLOCK = 1,
Pavel Begunkov889fca72021-02-10 00:03:09 +0000192 IO_URING_F_COMPLETE_DEFER = 2,
Pavel Begunkov45d189c2021-02-10 00:03:07 +0000193};
194
Jens Axboeedafcce2019-01-09 09:16:05 -0700195struct io_mapped_ubuf {
196 u64 ubuf;
197 size_t len;
198 struct bio_vec *bvec;
199 unsigned int nr_bvecs;
Jens Axboede293932020-09-17 16:19:16 -0600200 unsigned long acct_pages;
Jens Axboeedafcce2019-01-09 09:16:05 -0700201};
202
Bijan Mottahedeh50238532021-01-15 17:37:45 +0000203struct io_ring_ctx;
204
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000205struct io_rsrc_put {
206 struct list_head list;
Bijan Mottahedeh50238532021-01-15 17:37:45 +0000207 union {
208 void *rsrc;
209 struct file *file;
210 };
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000211};
212
213struct fixed_rsrc_table {
Jens Axboe65e19f52019-10-26 07:20:21 -0600214 struct file **files;
Jens Axboe31b51512019-01-18 22:56:34 -0700215};
216
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000217struct fixed_rsrc_ref_node {
Xiaoguang Wang05589552020-03-31 14:05:18 +0800218 struct percpu_ref refs;
219 struct list_head node;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000220 struct list_head rsrc_list;
221 struct fixed_rsrc_data *rsrc_data;
Bijan Mottahedeh50238532021-01-15 17:37:45 +0000222 void (*rsrc_put)(struct io_ring_ctx *ctx,
223 struct io_rsrc_put *prsrc);
Jens Axboe4a38aed22020-05-14 17:21:15 -0600224 struct llist_node llist;
Pavel Begunkove2978222020-11-18 14:56:26 +0000225 bool done;
Xiaoguang Wang05589552020-03-31 14:05:18 +0800226};
227
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000228struct fixed_rsrc_data {
229 struct fixed_rsrc_table *table;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700230 struct io_ring_ctx *ctx;
231
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000232 struct fixed_rsrc_ref_node *node;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700233 struct percpu_ref refs;
Jens Axboe05f3fb32019-12-09 11:22:50 -0700234 struct completion done;
235};
236
Jens Axboe5a2e7452020-02-23 16:23:11 -0700237struct io_buffer {
238 struct list_head list;
239 __u64 addr;
240 __s32 len;
241 __u16 bid;
242};
243
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200244struct io_restriction {
245 DECLARE_BITMAP(register_op, IORING_REGISTER_LAST);
246 DECLARE_BITMAP(sqe_op, IORING_OP_LAST);
247 u8 sqe_flags_allowed;
248 u8 sqe_flags_required;
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +0200249 bool registered;
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200250};
251
Jens Axboe534ca6d2020-09-02 13:52:19 -0600252struct io_sq_data {
253 refcount_t refs;
Jens Axboe69fb2132020-09-14 11:16:23 -0600254 struct mutex lock;
255
256 /* ctx's that are using this sqd */
257 struct list_head ctx_list;
258 struct list_head ctx_new_list;
259 struct mutex ctx_lock;
260
Jens Axboe534ca6d2020-09-02 13:52:19 -0600261 struct task_struct *thread;
262 struct wait_queue_head wait;
Xiaoguang Wang08369242020-11-03 14:15:59 +0800263
264 unsigned sq_thread_idle;
Jens Axboe534ca6d2020-09-02 13:52:19 -0600265};
266
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000267#define IO_IOPOLL_BATCH 8
Pavel Begunkov6dd0be12021-02-10 00:03:13 +0000268#define IO_COMPL_BATCH 32
Pavel Begunkov6ff119a2021-02-10 00:03:18 +0000269#define IO_REQ_CACHE_SIZE 32
Pavel Begunkovbf019da2021-02-10 00:03:17 +0000270#define IO_REQ_ALLOC_BATCH 8
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000271
272struct io_comp_state {
Pavel Begunkov6dd0be12021-02-10 00:03:13 +0000273 struct io_kiocb *reqs[IO_COMPL_BATCH];
Jens Axboe1b4c3512021-02-10 00:03:19 +0000274 unsigned int nr;
Jens Axboec7dae4b2021-02-09 19:53:37 -0700275 unsigned int locked_free_nr;
276 /* inline/task_work completion list, under ->uring_lock */
Jens Axboe1b4c3512021-02-10 00:03:19 +0000277 struct list_head free_list;
Jens Axboec7dae4b2021-02-09 19:53:37 -0700278 /* IRQ completion list, under ->completion_lock */
279 struct list_head locked_free_list;
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000280};
281
282struct io_submit_state {
283 struct blk_plug plug;
284
285 /*
286 * io_kiocb alloc cache
287 */
Pavel Begunkovbf019da2021-02-10 00:03:17 +0000288 void *reqs[IO_REQ_CACHE_SIZE];
Pavel Begunkov258b29a2021-02-10 00:03:10 +0000289 unsigned int free_reqs;
290
291 bool plug_started;
292
293 /*
294 * Batch completion logic
295 */
296 struct io_comp_state comp;
297
298 /*
299 * File reference cache
300 */
301 struct file *file;
302 unsigned int fd;
303 unsigned int file_refs;
304 unsigned int ios_left;
305};
306
Jens Axboe2b188cc2019-01-07 10:46:33 -0700307struct io_ring_ctx {
308 struct {
309 struct percpu_ref refs;
310 } ____cacheline_aligned_in_smp;
311
312 struct {
313 unsigned int flags;
Randy Dunlape1d85332020-02-05 20:57:10 -0800314 unsigned int compat: 1;
Bijan Mottahedehaad5d8d2020-06-16 16:36:08 -0700315 unsigned int limit_mem: 1;
Randy Dunlape1d85332020-02-05 20:57:10 -0800316 unsigned int cq_overflow_flushed: 1;
317 unsigned int drain_next: 1;
318 unsigned int eventfd_async: 1;
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200319 unsigned int restricted: 1;
Pavel Begunkovd9d05212021-01-08 20:57:25 +0000320 unsigned int sqo_dead: 1;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700321
Hristo Venev75b28af2019-08-26 17:23:46 +0000322 /*
323 * Ring buffer of indices into array of io_uring_sqe, which is
324 * mmapped by the application using the IORING_OFF_SQES offset.
325 *
326 * This indirection could e.g. be used to assign fixed
327 * io_uring_sqe entries to operations and only submit them to
328 * the queue when needed.
329 *
330 * The kernel modifies neither the indices array nor the entries
331 * array.
332 */
333 u32 *sq_array;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700334 unsigned cached_sq_head;
335 unsigned sq_entries;
336 unsigned sq_mask;
Jens Axboe6c271ce2019-01-10 11:22:30 -0700337 unsigned sq_thread_idle;
Jens Axboe498ccd92019-10-25 10:04:25 -0600338 unsigned cached_sq_dropped;
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +0100339 unsigned cached_cq_overflow;
Jens Axboead3eb2c2019-12-18 17:12:20 -0700340 unsigned long sq_check_overflow;
Jens Axboede0617e2019-04-06 21:51:27 -0600341
342 struct list_head defer_list;
Jens Axboe5262f562019-09-17 12:26:57 -0600343 struct list_head timeout_list;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -0700344 struct list_head cq_overflow_list;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700345
Jens Axboead3eb2c2019-12-18 17:12:20 -0700346 struct io_uring_sqe *sq_sqes;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700347 } ____cacheline_aligned_in_smp;
348
Jens Axboe3c1a2ea2021-02-11 10:48:03 -0700349 struct {
350 struct mutex uring_lock;
351 wait_queue_head_t wait;
352 } ____cacheline_aligned_in_smp;
353
354 struct io_submit_state submit_state;
355
Hristo Venev75b28af2019-08-26 17:23:46 +0000356 struct io_rings *rings;
357
Jens Axboe2b188cc2019-01-07 10:46:33 -0700358 /* IO offload */
Jens Axboe561fb042019-10-24 07:25:42 -0600359 struct io_wq *io_wq;
Jens Axboe2aede0e2020-09-14 10:45:53 -0600360
361 /*
362 * For SQPOLL usage - we hold a reference to the parent task, so we
363 * have access to the ->files
364 */
365 struct task_struct *sqo_task;
366
367 /* Only used for accounting purposes */
368 struct mm_struct *mm_account;
369
Dennis Zhou91d8f512020-09-16 13:41:05 -0700370#ifdef CONFIG_BLK_CGROUP
371 struct cgroup_subsys_state *sqo_blkcg_css;
372#endif
373
Jens Axboe534ca6d2020-09-02 13:52:19 -0600374 struct io_sq_data *sq_data; /* if using sq thread polling */
375
Jens Axboe90554202020-09-03 12:12:41 -0600376 struct wait_queue_head sqo_sq_wait;
Jens Axboe69fb2132020-09-14 11:16:23 -0600377 struct list_head sqd_list;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700378
Jens Axboe6b063142019-01-10 22:13:58 -0700379 /*
380 * If used, fixed file set. Writers must ensure that ->refs is dead,
381 * readers must ensure that ->refs is alive as long as the file* is
382 * used. Only updated through io_uring_register(2).
383 */
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000384 struct fixed_rsrc_data *file_data;
Jens Axboe6b063142019-01-10 22:13:58 -0700385 unsigned nr_user_files;
386
Jens Axboeedafcce2019-01-09 09:16:05 -0700387 /* if used, fixed mapped user buffers */
388 unsigned nr_user_bufs;
389 struct io_mapped_ubuf *user_bufs;
390
Jens Axboe2b188cc2019-01-07 10:46:33 -0700391 struct user_struct *user;
392
Jens Axboe0b8c0ec2019-12-02 08:50:00 -0700393 const struct cred *creds;
Jens Axboe181e4482019-11-25 08:52:30 -0700394
Jens Axboe4ea33a92020-10-15 13:46:44 -0600395#ifdef CONFIG_AUDIT
396 kuid_t loginuid;
397 unsigned int sessionid;
398#endif
399
Jens Axboe0f158b42020-05-14 17:18:39 -0600400 struct completion ref_comp;
401 struct completion sq_thread_comp;
Jens Axboe206aefd2019-11-07 18:27:42 -0700402
403#if defined(CONFIG_UNIX)
404 struct socket *ring_sock;
405#endif
406
Jens Axboe5a2e7452020-02-23 16:23:11 -0700407 struct idr io_buffer_idr;
408
Jens Axboe071698e2020-01-28 10:04:42 -0700409 struct idr personality_idr;
410
Jens Axboe206aefd2019-11-07 18:27:42 -0700411 struct {
412 unsigned cached_cq_tail;
413 unsigned cq_entries;
414 unsigned cq_mask;
415 atomic_t cq_timeouts;
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -0500416 unsigned cq_last_tm_flush;
Jens Axboead3eb2c2019-12-18 17:12:20 -0700417 unsigned long cq_check_overflow;
Jens Axboe206aefd2019-11-07 18:27:42 -0700418 struct wait_queue_head cq_wait;
419 struct fasync_struct *cq_fasync;
420 struct eventfd_ctx *cq_ev_fd;
421 } ____cacheline_aligned_in_smp;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700422
423 struct {
Jens Axboe2b188cc2019-01-07 10:46:33 -0700424 spinlock_t completion_lock;
Jens Axboee94f1412019-12-19 12:06:02 -0700425
Jens Axboedef596e2019-01-09 08:59:42 -0700426 /*
Pavel Begunkov540e32a2020-07-13 23:37:09 +0300427 * ->iopoll_list is protected by the ctx->uring_lock for
Jens Axboedef596e2019-01-09 08:59:42 -0700428 * io_uring instances that don't use IORING_SETUP_SQPOLL.
429 * For SQPOLL, only the single threaded io_sq_thread() will
430 * manipulate the list, hence no extra locking is needed there.
431 */
Pavel Begunkov540e32a2020-07-13 23:37:09 +0300432 struct list_head iopoll_list;
Jens Axboe78076bb2019-12-04 19:56:40 -0700433 struct hlist_head *cancel_hash;
434 unsigned cancel_hash_bits;
Jens Axboee94f1412019-12-19 12:06:02 -0700435 bool poll_multi_file;
Jens Axboefcb323c2019-10-24 12:39:47 -0600436
437 spinlock_t inflight_lock;
438 struct list_head inflight_list;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700439 } ____cacheline_aligned_in_smp;
Jens Axboe85faa7b2020-04-09 18:14:00 -0600440
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000441 struct delayed_work rsrc_put_work;
442 struct llist_head rsrc_put_llist;
Bijan Mottahedehd67d2262021-01-15 17:37:46 +0000443 struct list_head rsrc_ref_list;
444 spinlock_t rsrc_ref_lock;
Jens Axboe4a38aed22020-05-14 17:21:15 -0600445
Stefano Garzarella21b55db2020-08-27 16:58:30 +0200446 struct io_restriction restrictions;
Jens Axboe3c1a2ea2021-02-11 10:48:03 -0700447
448 /* Keep this last, we don't need it for the fast path */
449 struct work_struct exit_work;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700450};
451
Jens Axboe09bb8392019-03-13 12:39:28 -0600452/*
453 * First field must be the file pointer in all the
454 * iocb unions! See also 'struct kiocb' in <linux/fs.h>
455 */
Jens Axboe221c5eb2019-01-17 09:41:58 -0700456struct io_poll_iocb {
457 struct file *file;
Pavel Begunkov018043b2020-10-27 23:17:18 +0000458 struct wait_queue_head *head;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700459 __poll_t events;
Jens Axboe8c838782019-03-12 15:48:16 -0600460 bool done;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700461 bool canceled;
Jens Axboe392edb42019-12-09 17:52:20 -0700462 struct wait_queue_entry wait;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700463};
464
Pavel Begunkov018043b2020-10-27 23:17:18 +0000465struct io_poll_remove {
466 struct file *file;
467 u64 addr;
468};
469
Jens Axboeb5dba592019-12-11 14:02:38 -0700470struct io_close {
471 struct file *file;
Jens Axboeb5dba592019-12-11 14:02:38 -0700472 int fd;
473};
474
Jens Axboead8a48a2019-11-15 08:49:11 -0700475struct io_timeout_data {
476 struct io_kiocb *req;
477 struct hrtimer timer;
478 struct timespec64 ts;
479 enum hrtimer_mode mode;
480};
481
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700482struct io_accept {
483 struct file *file;
484 struct sockaddr __user *addr;
485 int __user *addr_len;
486 int flags;
Jens Axboe09952e32020-03-19 20:16:56 -0600487 unsigned long nofile;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700488};
489
490struct io_sync {
491 struct file *file;
492 loff_t len;
493 loff_t off;
494 int flags;
Jens Axboed63d1b52019-12-10 10:38:56 -0700495 int mode;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700496};
497
Jens Axboefbf23842019-12-17 18:45:56 -0700498struct io_cancel {
499 struct file *file;
500 u64 addr;
501};
502
Jens Axboeb29472e2019-12-17 18:50:29 -0700503struct io_timeout {
504 struct file *file;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +0300505 u32 off;
506 u32 target_seq;
Pavel Begunkov135fcde2020-07-13 23:37:12 +0300507 struct list_head list;
Pavel Begunkov90cd7e42020-10-27 23:25:36 +0000508 /* head of the link, used by linked timeouts only */
509 struct io_kiocb *head;
Jens Axboeb29472e2019-12-17 18:50:29 -0700510};
511
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100512struct io_timeout_rem {
513 struct file *file;
514 u64 addr;
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000515
516 /* timeout update */
517 struct timespec64 ts;
518 u32 flags;
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100519};
520
Jens Axboe9adbd452019-12-20 08:45:55 -0700521struct io_rw {
522 /* NOTE: kiocb has the file as the first member, so don't do it here */
523 struct kiocb kiocb;
524 u64 addr;
525 u64 len;
526};
527
Jens Axboe3fbb51c2019-12-20 08:51:52 -0700528struct io_connect {
529 struct file *file;
530 struct sockaddr __user *addr;
531 int addr_len;
532};
533
Jens Axboee47293f2019-12-20 08:58:21 -0700534struct io_sr_msg {
535 struct file *file;
Jens Axboefddafac2020-01-04 20:19:44 -0700536 union {
Pavel Begunkov270a5942020-07-12 20:41:04 +0300537 struct user_msghdr __user *umsg;
Jens Axboefddafac2020-01-04 20:19:44 -0700538 void __user *buf;
539 };
Jens Axboee47293f2019-12-20 08:58:21 -0700540 int msg_flags;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700541 int bgid;
Jens Axboefddafac2020-01-04 20:19:44 -0700542 size_t len;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700543 struct io_buffer *kbuf;
Jens Axboee47293f2019-12-20 08:58:21 -0700544};
545
Jens Axboe15b71ab2019-12-11 11:20:36 -0700546struct io_open {
547 struct file *file;
548 int dfd;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700549 struct filename *filename;
Jens Axboec12cedf2020-01-08 17:41:21 -0700550 struct open_how how;
Jens Axboe4022e7a2020-03-19 19:23:18 -0600551 unsigned long nofile;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700552};
553
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000554struct io_rsrc_update {
Jens Axboe05f3fb32019-12-09 11:22:50 -0700555 struct file *file;
556 u64 arg;
557 u32 nr_args;
558 u32 offset;
559};
560
Jens Axboe4840e412019-12-25 22:03:45 -0700561struct io_fadvise {
562 struct file *file;
563 u64 offset;
564 u32 len;
565 u32 advice;
566};
567
Jens Axboec1ca7572019-12-25 22:18:28 -0700568struct io_madvise {
569 struct file *file;
570 u64 addr;
571 u32 len;
572 u32 advice;
573};
574
Jens Axboe3e4827b2020-01-08 15:18:09 -0700575struct io_epoll {
576 struct file *file;
577 int epfd;
578 int op;
579 int fd;
580 struct epoll_event event;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700581};
582
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300583struct io_splice {
584 struct file *file_out;
585 struct file *file_in;
586 loff_t off_out;
587 loff_t off_in;
588 u64 len;
589 unsigned int flags;
590};
591
Jens Axboeddf0322d2020-02-23 16:41:33 -0700592struct io_provide_buf {
593 struct file *file;
594 __u64 addr;
595 __s32 len;
596 __u32 bgid;
597 __u16 nbufs;
598 __u16 bid;
599};
600
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700601struct io_statx {
602 struct file *file;
603 int dfd;
604 unsigned int mask;
605 unsigned int flags;
Bijan Mottahedehe62753e2020-05-22 21:31:18 -0700606 const char __user *filename;
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700607 struct statx __user *buffer;
608};
609
Jens Axboe36f4fa62020-09-05 11:14:22 -0600610struct io_shutdown {
611 struct file *file;
612 int how;
613};
614
Jens Axboe80a261f2020-09-28 14:23:58 -0600615struct io_rename {
616 struct file *file;
617 int old_dfd;
618 int new_dfd;
619 struct filename *oldpath;
620 struct filename *newpath;
621 int flags;
622};
623
Jens Axboe14a11432020-09-28 14:27:37 -0600624struct io_unlink {
625 struct file *file;
626 int dfd;
627 int flags;
628 struct filename *filename;
629};
630
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300631struct io_completion {
632 struct file *file;
633 struct list_head list;
Pavel Begunkov0f7e4662020-07-13 23:37:16 +0300634 int cflags;
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300635};
636
Jens Axboef499a022019-12-02 16:28:46 -0700637struct io_async_connect {
638 struct sockaddr_storage address;
639};
640
Jens Axboe03b12302019-12-02 18:50:25 -0700641struct io_async_msghdr {
642 struct iovec fast_iov[UIO_FASTIOV];
Pavel Begunkov257e84a2021-02-05 00:58:00 +0000643 /* points to an allocated iov, if NULL we use fast_iov instead */
644 struct iovec *free_iov;
Jens Axboe03b12302019-12-02 18:50:25 -0700645 struct sockaddr __user *uaddr;
646 struct msghdr msg;
Jens Axboeb5379162020-02-09 11:29:15 -0700647 struct sockaddr_storage addr;
Jens Axboe03b12302019-12-02 18:50:25 -0700648};
649
Jens Axboef67676d2019-12-02 11:03:47 -0700650struct io_async_rw {
651 struct iovec fast_iov[UIO_FASTIOV];
Jens Axboeff6165b2020-08-13 09:47:43 -0600652 const struct iovec *free_iovec;
653 struct iov_iter iter;
Jens Axboe227c0c92020-08-13 11:51:40 -0600654 size_t bytes_done;
Jens Axboebcf5a062020-05-22 09:24:42 -0600655 struct wait_page_queue wpq;
Jens Axboef67676d2019-12-02 11:03:47 -0700656};
657
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300658enum {
659 REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT,
660 REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT,
661 REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT,
662 REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT,
663 REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700664 REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300665
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300666 REQ_F_FAIL_LINK_BIT,
667 REQ_F_INFLIGHT_BIT,
668 REQ_F_CUR_POS_BIT,
669 REQ_F_NOWAIT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300670 REQ_F_LINK_TIMEOUT_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300671 REQ_F_ISREG_BIT,
Pavel Begunkov99bc4c32020-02-07 22:04:45 +0300672 REQ_F_NEED_CLEANUP_BIT,
Jens Axboed7718a92020-02-14 22:23:12 -0700673 REQ_F_POLLED_BIT,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700674 REQ_F_BUFFER_SELECTED_BIT,
Jens Axboe5b0bbee2020-04-27 10:41:22 -0600675 REQ_F_NO_FILE_TABLE_BIT,
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +0800676 REQ_F_WORK_INITIALIZED_BIT,
Pavel Begunkov900fad42020-10-19 16:39:16 +0100677 REQ_F_LTIMEOUT_ACTIVE_BIT,
Pavel Begunkove342c802021-01-19 13:32:47 +0000678 REQ_F_COMPLETE_INLINE_BIT,
Jens Axboe84557872020-03-03 15:28:17 -0700679
680 /* not a real bit, just to check we're not overflowing the space */
681 __REQ_F_LAST_BIT,
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300682};
683
684enum {
685 /* ctx owns file */
686 REQ_F_FIXED_FILE = BIT(REQ_F_FIXED_FILE_BIT),
687 /* drain existing IO first */
688 REQ_F_IO_DRAIN = BIT(REQ_F_IO_DRAIN_BIT),
689 /* linked sqes */
690 REQ_F_LINK = BIT(REQ_F_LINK_BIT),
691 /* doesn't sever on completion < 0 */
692 REQ_F_HARDLINK = BIT(REQ_F_HARDLINK_BIT),
693 /* IOSQE_ASYNC */
694 REQ_F_FORCE_ASYNC = BIT(REQ_F_FORCE_ASYNC_BIT),
Jens Axboebcda7ba2020-02-23 16:42:51 -0700695 /* IOSQE_BUFFER_SELECT */
696 REQ_F_BUFFER_SELECT = BIT(REQ_F_BUFFER_SELECT_BIT),
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300697
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300698 /* fail rest of links */
699 REQ_F_FAIL_LINK = BIT(REQ_F_FAIL_LINK_BIT),
700 /* on inflight list */
701 REQ_F_INFLIGHT = BIT(REQ_F_INFLIGHT_BIT),
702 /* read/write uses file position */
703 REQ_F_CUR_POS = BIT(REQ_F_CUR_POS_BIT),
704 /* must not punt to workers */
705 REQ_F_NOWAIT = BIT(REQ_F_NOWAIT_BIT),
Pavel Begunkov900fad42020-10-19 16:39:16 +0100706 /* has or had linked timeout */
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300707 REQ_F_LINK_TIMEOUT = BIT(REQ_F_LINK_TIMEOUT_BIT),
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300708 /* regular file */
709 REQ_F_ISREG = BIT(REQ_F_ISREG_BIT),
Pavel Begunkov99bc4c32020-02-07 22:04:45 +0300710 /* needs cleanup */
711 REQ_F_NEED_CLEANUP = BIT(REQ_F_NEED_CLEANUP_BIT),
Jens Axboed7718a92020-02-14 22:23:12 -0700712 /* already went through poll handler */
713 REQ_F_POLLED = BIT(REQ_F_POLLED_BIT),
Jens Axboebcda7ba2020-02-23 16:42:51 -0700714 /* buffer already selected */
715 REQ_F_BUFFER_SELECTED = BIT(REQ_F_BUFFER_SELECTED_BIT),
Jens Axboe5b0bbee2020-04-27 10:41:22 -0600716 /* doesn't need file table for this request */
717 REQ_F_NO_FILE_TABLE = BIT(REQ_F_NO_FILE_TABLE_BIT),
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +0800718 /* io_wq_work is initialized */
719 REQ_F_WORK_INITIALIZED = BIT(REQ_F_WORK_INITIALIZED_BIT),
Pavel Begunkov900fad42020-10-19 16:39:16 +0100720 /* linked timeout is active, i.e. prepared by link's head */
721 REQ_F_LTIMEOUT_ACTIVE = BIT(REQ_F_LTIMEOUT_ACTIVE_BIT),
Pavel Begunkove342c802021-01-19 13:32:47 +0000722 /* completion is deferred through io_comp_state */
723 REQ_F_COMPLETE_INLINE = BIT(REQ_F_COMPLETE_INLINE_BIT),
Jens Axboed7718a92020-02-14 22:23:12 -0700724};
725
726struct async_poll {
727 struct io_poll_iocb poll;
Jens Axboe807abcb2020-07-17 17:09:27 -0600728 struct io_poll_iocb *double_poll;
Pavel Begunkov6b47ee62020-01-18 20:22:41 +0300729};
730
Jens Axboe7cbf1722021-02-10 00:03:20 +0000731struct io_task_work {
732 struct io_wq_work_node node;
733 task_work_func_t func;
734};
735
Jens Axboe09bb8392019-03-13 12:39:28 -0600736/*
737 * NOTE! Each of the iocb union members has the file pointer
738 * as the first entry in their struct definition. So you can
739 * access the file pointer through any of the sub-structs,
740 * or directly as just 'ki_filp' in this struct.
741 */
Jens Axboe2b188cc2019-01-07 10:46:33 -0700742struct io_kiocb {
Jens Axboe221c5eb2019-01-17 09:41:58 -0700743 union {
Jens Axboe09bb8392019-03-13 12:39:28 -0600744 struct file *file;
Jens Axboe9adbd452019-12-20 08:45:55 -0700745 struct io_rw rw;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700746 struct io_poll_iocb poll;
Pavel Begunkov018043b2020-10-27 23:17:18 +0000747 struct io_poll_remove poll_remove;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -0700748 struct io_accept accept;
749 struct io_sync sync;
Jens Axboefbf23842019-12-17 18:45:56 -0700750 struct io_cancel cancel;
Jens Axboeb29472e2019-12-17 18:50:29 -0700751 struct io_timeout timeout;
Pavel Begunkov0bdf7a22020-10-10 18:34:10 +0100752 struct io_timeout_rem timeout_rem;
Jens Axboe3fbb51c2019-12-20 08:51:52 -0700753 struct io_connect connect;
Jens Axboee47293f2019-12-20 08:58:21 -0700754 struct io_sr_msg sr_msg;
Jens Axboe15b71ab2019-12-11 11:20:36 -0700755 struct io_open open;
Jens Axboeb5dba592019-12-11 14:02:38 -0700756 struct io_close close;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000757 struct io_rsrc_update rsrc_update;
Jens Axboe4840e412019-12-25 22:03:45 -0700758 struct io_fadvise fadvise;
Jens Axboec1ca7572019-12-25 22:18:28 -0700759 struct io_madvise madvise;
Jens Axboe3e4827b2020-01-08 15:18:09 -0700760 struct io_epoll epoll;
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300761 struct io_splice splice;
Jens Axboeddf0322d2020-02-23 16:41:33 -0700762 struct io_provide_buf pbuf;
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -0700763 struct io_statx statx;
Jens Axboe36f4fa62020-09-05 11:14:22 -0600764 struct io_shutdown shutdown;
Jens Axboe80a261f2020-09-28 14:23:58 -0600765 struct io_rename rename;
Jens Axboe14a11432020-09-28 14:27:37 -0600766 struct io_unlink unlink;
Pavel Begunkov3ca405e2020-07-13 23:37:08 +0300767 /* use only after cleaning per-op data, see io_clean_op() */
768 struct io_completion compl;
Jens Axboe221c5eb2019-01-17 09:41:58 -0700769 };
Jens Axboe2b188cc2019-01-07 10:46:33 -0700770
Jens Axboee8c2bc12020-08-15 18:44:09 -0700771 /* opcode allocated if it needs to store data for async defer */
772 void *async_data;
Jens Axboed625c6e2019-12-17 19:53:05 -0700773 u8 opcode;
Xiaoguang Wang65a65432020-06-11 23:39:36 +0800774 /* polled IO has completed */
775 u8 iopoll_completed;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700776
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -0700777 u16 buf_index;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +0300778 u32 result;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -0700779
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300780 struct io_ring_ctx *ctx;
781 unsigned int flags;
782 refcount_t refs;
783 struct task_struct *task;
784 u64 user_data;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700785
Pavel Begunkovf2f87372020-10-27 23:25:37 +0000786 struct io_kiocb *link;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +0000787 struct percpu_ref *fixed_rsrc_refs;
Jens Axboed7718a92020-02-14 22:23:12 -0700788
Pavel Begunkovd21ffe72020-07-13 23:37:10 +0300789 /*
790 * 1. used with ctx->iopoll_list with reads/writes
791 * 2. to track reqs with ->files (see io_op_def::file_table)
792 */
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300793 struct list_head inflight_entry;
Jens Axboe7cbf1722021-02-10 00:03:20 +0000794 union {
795 struct io_task_work io_task_work;
796 struct callback_head task_work;
797 };
Pavel Begunkov010e8e62020-07-30 18:43:45 +0300798 /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */
799 struct hlist_node hash_node;
800 struct async_poll *apoll;
801 struct io_wq_work work;
Jens Axboe2b188cc2019-01-07 10:46:33 -0700802};
803
Pavel Begunkov27dc8332020-07-13 23:37:14 +0300804struct io_defer_entry {
805 struct list_head list;
806 struct io_kiocb *req;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +0300807 u32 seq;
Pavel Begunkov27dc8332020-07-13 23:37:14 +0300808};
809
Jens Axboed3656342019-12-18 09:50:26 -0700810struct io_op_def {
Jens Axboed3656342019-12-18 09:50:26 -0700811 /* needs req->file assigned */
812 unsigned needs_file : 1;
Jens Axboed3656342019-12-18 09:50:26 -0700813 /* hash wq insertion if file is a regular file */
814 unsigned hash_reg_file : 1;
815 /* unbound wq insertion if file is a non-regular file */
816 unsigned unbound_nonreg_file : 1;
Jens Axboe66f4af92020-01-16 15:36:52 -0700817 /* opcode is not supported by this kernel */
818 unsigned not_supported : 1;
Jens Axboe8a727582020-02-20 09:59:44 -0700819 /* set if opcode supports polled "wait" */
820 unsigned pollin : 1;
821 unsigned pollout : 1;
Jens Axboebcda7ba2020-02-23 16:42:51 -0700822 /* op supports buffer selection */
823 unsigned buffer_select : 1;
Jens Axboee8c2bc12020-08-15 18:44:09 -0700824 /* must always have async data allocated */
825 unsigned needs_async_data : 1;
Jens Axboe27926b62020-10-28 09:33:23 -0600826 /* should block plug */
827 unsigned plug : 1;
Jens Axboee8c2bc12020-08-15 18:44:09 -0700828 /* size of async data needed, if any */
829 unsigned short async_size;
Jens Axboe0f203762020-10-14 09:23:55 -0600830 unsigned work_flags;
Jens Axboed3656342019-12-18 09:50:26 -0700831};
832
Jens Axboe09186822020-10-13 15:01:40 -0600833static const struct io_op_def io_op_defs[] = {
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300834 [IORING_OP_NOP] = {},
835 [IORING_OP_READV] = {
Jens Axboed3656342019-12-18 09:50:26 -0700836 .needs_file = 1,
837 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700838 .pollin = 1,
Jens Axboe4d954c22020-02-27 07:31:19 -0700839 .buffer_select = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700840 .needs_async_data = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600841 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700842 .async_size = sizeof(struct io_async_rw),
Jens Axboe0f203762020-10-14 09:23:55 -0600843 .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG,
Jens Axboed3656342019-12-18 09:50:26 -0700844 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300845 [IORING_OP_WRITEV] = {
Jens Axboed3656342019-12-18 09:50:26 -0700846 .needs_file = 1,
847 .hash_reg_file = 1,
848 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700849 .pollout = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700850 .needs_async_data = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600851 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700852 .async_size = sizeof(struct io_async_rw),
Jens Axboe69228332020-10-20 14:28:41 -0600853 .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG |
854 IO_WQ_WORK_FSIZE,
Jens Axboed3656342019-12-18 09:50:26 -0700855 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300856 [IORING_OP_FSYNC] = {
Jens Axboed3656342019-12-18 09:50:26 -0700857 .needs_file = 1,
Jens Axboe0f203762020-10-14 09:23:55 -0600858 .work_flags = IO_WQ_WORK_BLKCG,
Jens Axboed3656342019-12-18 09:50:26 -0700859 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300860 [IORING_OP_READ_FIXED] = {
Jens Axboed3656342019-12-18 09:50:26 -0700861 .needs_file = 1,
862 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700863 .pollin = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600864 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700865 .async_size = sizeof(struct io_async_rw),
Jens Axboe4017eb92020-10-22 14:14:12 -0600866 .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_MM,
Jens Axboed3656342019-12-18 09:50:26 -0700867 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300868 [IORING_OP_WRITE_FIXED] = {
Jens Axboed3656342019-12-18 09:50:26 -0700869 .needs_file = 1,
870 .hash_reg_file = 1,
871 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700872 .pollout = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600873 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700874 .async_size = sizeof(struct io_async_rw),
Jens Axboe4017eb92020-10-22 14:14:12 -0600875 .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_FSIZE |
876 IO_WQ_WORK_MM,
Jens Axboed3656342019-12-18 09:50:26 -0700877 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300878 [IORING_OP_POLL_ADD] = {
Jens Axboed3656342019-12-18 09:50:26 -0700879 .needs_file = 1,
880 .unbound_nonreg_file = 1,
881 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300882 [IORING_OP_POLL_REMOVE] = {},
883 [IORING_OP_SYNC_FILE_RANGE] = {
Jens Axboed3656342019-12-18 09:50:26 -0700884 .needs_file = 1,
Jens Axboe0f203762020-10-14 09:23:55 -0600885 .work_flags = IO_WQ_WORK_BLKCG,
Jens Axboed3656342019-12-18 09:50:26 -0700886 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300887 [IORING_OP_SENDMSG] = {
Jens Axboed3656342019-12-18 09:50:26 -0700888 .needs_file = 1,
889 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700890 .pollout = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700891 .needs_async_data = 1,
892 .async_size = sizeof(struct io_async_msghdr),
Pavel Begunkov10cad2c2020-11-07 13:20:39 +0000893 .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG,
Jens Axboed3656342019-12-18 09:50:26 -0700894 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300895 [IORING_OP_RECVMSG] = {
Jens Axboed3656342019-12-18 09:50:26 -0700896 .needs_file = 1,
897 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700898 .pollin = 1,
Jens Axboe52de1fe2020-02-27 10:15:42 -0700899 .buffer_select = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700900 .needs_async_data = 1,
901 .async_size = sizeof(struct io_async_msghdr),
Pavel Begunkov10cad2c2020-11-07 13:20:39 +0000902 .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG,
Jens Axboed3656342019-12-18 09:50:26 -0700903 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300904 [IORING_OP_TIMEOUT] = {
Jens Axboee8c2bc12020-08-15 18:44:09 -0700905 .needs_async_data = 1,
906 .async_size = sizeof(struct io_timeout_data),
Jens Axboe0f203762020-10-14 09:23:55 -0600907 .work_flags = IO_WQ_WORK_MM,
Jens Axboed3656342019-12-18 09:50:26 -0700908 },
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +0000909 [IORING_OP_TIMEOUT_REMOVE] = {
910 /* used by timeout updates' prep() */
911 .work_flags = IO_WQ_WORK_MM,
912 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300913 [IORING_OP_ACCEPT] = {
Jens Axboed3656342019-12-18 09:50:26 -0700914 .needs_file = 1,
915 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700916 .pollin = 1,
Jens Axboe0f203762020-10-14 09:23:55 -0600917 .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_FILES,
Jens Axboed3656342019-12-18 09:50:26 -0700918 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300919 [IORING_OP_ASYNC_CANCEL] = {},
920 [IORING_OP_LINK_TIMEOUT] = {
Jens Axboee8c2bc12020-08-15 18:44:09 -0700921 .needs_async_data = 1,
922 .async_size = sizeof(struct io_timeout_data),
Jens Axboe0f203762020-10-14 09:23:55 -0600923 .work_flags = IO_WQ_WORK_MM,
Jens Axboed3656342019-12-18 09:50:26 -0700924 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300925 [IORING_OP_CONNECT] = {
Jens Axboed3656342019-12-18 09:50:26 -0700926 .needs_file = 1,
927 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700928 .pollout = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700929 .needs_async_data = 1,
930 .async_size = sizeof(struct io_async_connect),
Jens Axboe0f203762020-10-14 09:23:55 -0600931 .work_flags = IO_WQ_WORK_MM,
Jens Axboed3656342019-12-18 09:50:26 -0700932 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300933 [IORING_OP_FALLOCATE] = {
Jens Axboed3656342019-12-18 09:50:26 -0700934 .needs_file = 1,
Jens Axboe69228332020-10-20 14:28:41 -0600935 .work_flags = IO_WQ_WORK_BLKCG | IO_WQ_WORK_FSIZE,
Jens Axboed3656342019-12-18 09:50:26 -0700936 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300937 [IORING_OP_OPENAT] = {
Jens Axboe0f203762020-10-14 09:23:55 -0600938 .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_BLKCG |
Jens Axboe14587a462020-09-05 11:36:08 -0600939 IO_WQ_WORK_FS | IO_WQ_WORK_MM,
Jens Axboed3656342019-12-18 09:50:26 -0700940 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300941 [IORING_OP_CLOSE] = {
Jens Axboe0f203762020-10-14 09:23:55 -0600942 .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_BLKCG,
Jens Axboed3656342019-12-18 09:50:26 -0700943 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300944 [IORING_OP_FILES_UPDATE] = {
Jens Axboe0f203762020-10-14 09:23:55 -0600945 .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_MM,
Jens Axboed3656342019-12-18 09:50:26 -0700946 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300947 [IORING_OP_STATX] = {
Jens Axboe0f203762020-10-14 09:23:55 -0600948 .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_MM |
949 IO_WQ_WORK_FS | IO_WQ_WORK_BLKCG,
Jens Axboed3656342019-12-18 09:50:26 -0700950 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300951 [IORING_OP_READ] = {
Jens Axboe3a6820f2019-12-22 15:19:35 -0700952 .needs_file = 1,
953 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700954 .pollin = 1,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700955 .buffer_select = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600956 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700957 .async_size = sizeof(struct io_async_rw),
Jens Axboe0f203762020-10-14 09:23:55 -0600958 .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG,
Jens Axboe3a6820f2019-12-22 15:19:35 -0700959 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300960 [IORING_OP_WRITE] = {
Jens Axboe3a6820f2019-12-22 15:19:35 -0700961 .needs_file = 1,
962 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700963 .pollout = 1,
Jens Axboe27926b62020-10-28 09:33:23 -0600964 .plug = 1,
Jens Axboee8c2bc12020-08-15 18:44:09 -0700965 .async_size = sizeof(struct io_async_rw),
Jens Axboe69228332020-10-20 14:28:41 -0600966 .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG |
967 IO_WQ_WORK_FSIZE,
Jens Axboe3a6820f2019-12-22 15:19:35 -0700968 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300969 [IORING_OP_FADVISE] = {
Jens Axboe4840e412019-12-25 22:03:45 -0700970 .needs_file = 1,
Jens Axboe0f203762020-10-14 09:23:55 -0600971 .work_flags = IO_WQ_WORK_BLKCG,
Jens Axboe4840e412019-12-25 22:03:45 -0700972 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300973 [IORING_OP_MADVISE] = {
Jens Axboe0f203762020-10-14 09:23:55 -0600974 .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG,
Jens Axboec1ca7572019-12-25 22:18:28 -0700975 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300976 [IORING_OP_SEND] = {
Jens Axboefddafac2020-01-04 20:19:44 -0700977 .needs_file = 1,
978 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700979 .pollout = 1,
Jens Axboe0f203762020-10-14 09:23:55 -0600980 .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG,
Jens Axboefddafac2020-01-04 20:19:44 -0700981 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300982 [IORING_OP_RECV] = {
Jens Axboefddafac2020-01-04 20:19:44 -0700983 .needs_file = 1,
984 .unbound_nonreg_file = 1,
Jens Axboe8a727582020-02-20 09:59:44 -0700985 .pollin = 1,
Jens Axboebcda7ba2020-02-23 16:42:51 -0700986 .buffer_select = 1,
Jens Axboe0f203762020-10-14 09:23:55 -0600987 .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG,
Jens Axboefddafac2020-01-04 20:19:44 -0700988 },
Pavel Begunkov0463b6c2020-01-18 21:35:38 +0300989 [IORING_OP_OPENAT2] = {
Jens Axboe0f203762020-10-14 09:23:55 -0600990 .work_flags = IO_WQ_WORK_FILES | IO_WQ_WORK_FS |
Jens Axboe14587a462020-09-05 11:36:08 -0600991 IO_WQ_WORK_BLKCG | IO_WQ_WORK_MM,
Jens Axboecebdb982020-01-08 17:59:24 -0700992 },
Jens Axboe3e4827b2020-01-08 15:18:09 -0700993 [IORING_OP_EPOLL_CTL] = {
994 .unbound_nonreg_file = 1,
Jens Axboe0f203762020-10-14 09:23:55 -0600995 .work_flags = IO_WQ_WORK_FILES,
Jens Axboe3e4827b2020-01-08 15:18:09 -0700996 },
Pavel Begunkov7d67af22020-02-24 11:32:45 +0300997 [IORING_OP_SPLICE] = {
998 .needs_file = 1,
999 .hash_reg_file = 1,
1000 .unbound_nonreg_file = 1,
Jens Axboe0f203762020-10-14 09:23:55 -06001001 .work_flags = IO_WQ_WORK_BLKCG,
Jens Axboeddf0322d2020-02-23 16:41:33 -07001002 },
1003 [IORING_OP_PROVIDE_BUFFERS] = {},
Jens Axboe067524e2020-03-02 16:32:28 -07001004 [IORING_OP_REMOVE_BUFFERS] = {},
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03001005 [IORING_OP_TEE] = {
1006 .needs_file = 1,
1007 .hash_reg_file = 1,
1008 .unbound_nonreg_file = 1,
1009 },
Jens Axboe36f4fa62020-09-05 11:14:22 -06001010 [IORING_OP_SHUTDOWN] = {
1011 .needs_file = 1,
1012 },
Jens Axboe80a261f2020-09-28 14:23:58 -06001013 [IORING_OP_RENAMEAT] = {
1014 .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_FILES |
1015 IO_WQ_WORK_FS | IO_WQ_WORK_BLKCG,
1016 },
Jens Axboe14a11432020-09-28 14:27:37 -06001017 [IORING_OP_UNLINKAT] = {
1018 .work_flags = IO_WQ_WORK_MM | IO_WQ_WORK_FILES |
1019 IO_WQ_WORK_FS | IO_WQ_WORK_BLKCG,
1020 },
Jens Axboed3656342019-12-18 09:50:26 -07001021};
1022
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00001023static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
1024 struct task_struct *task,
1025 struct files_struct *files);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001026static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node);
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00001027static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node(
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00001028 struct io_ring_ctx *ctx);
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00001029static void init_fixed_file_ref_node(struct io_ring_ctx *ctx,
1030 struct fixed_rsrc_ref_node *ref_node);
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00001031
Pavel Begunkov23faba32021-02-11 18:28:22 +00001032static bool io_rw_reissue(struct io_kiocb *req);
Jens Axboe78e19bb2019-11-06 15:21:34 -07001033static void io_cqring_fill_event(struct io_kiocb *req, long res);
Jackie Liuec9c02a2019-11-08 23:50:36 +08001034static void io_put_req(struct io_kiocb *req);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001035static void io_put_req_deferred(struct io_kiocb *req, int nr);
Jens Axboec40f6372020-06-25 15:39:59 -06001036static void io_double_put_req(struct io_kiocb *req);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001037static void io_dismantle_req(struct io_kiocb *req);
1038static void io_put_task(struct task_struct *task, int nr);
1039static void io_queue_next(struct io_kiocb *req);
Jens Axboe94ae5e72019-11-14 19:39:52 -07001040static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req);
Jens Axboe7271ef32020-08-10 09:55:22 -06001041static void __io_queue_linked_timeout(struct io_kiocb *req);
Jens Axboe94ae5e72019-11-14 19:39:52 -07001042static void io_queue_linked_timeout(struct io_kiocb *req);
Jens Axboe05f3fb32019-12-09 11:22:50 -07001043static int __io_sqe_files_update(struct io_ring_ctx *ctx,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001044 struct io_uring_rsrc_update *ip,
Jens Axboe05f3fb32019-12-09 11:22:50 -07001045 unsigned nr_args);
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001046static void __io_clean_op(struct io_kiocb *req);
Pavel Begunkov8371adf2020-10-10 18:34:08 +01001047static struct file *io_file_get(struct io_submit_state *state,
1048 struct io_kiocb *req, int fd, bool fixed);
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00001049static void __io_queue_sqe(struct io_kiocb *req);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001050static void io_rsrc_put_work(struct work_struct *work);
Jens Axboede0617e2019-04-06 21:51:27 -06001051
Pavel Begunkov847595d2021-02-04 13:52:06 +00001052static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec,
1053 struct iov_iter *iter, bool needs_lock);
Jens Axboeff6165b2020-08-13 09:47:43 -06001054static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec,
1055 const struct iovec *fast_iov,
Jens Axboe227c0c92020-08-13 11:51:40 -06001056 struct iov_iter *iter, bool force);
Pavel Begunkov907d1df2021-01-26 23:35:10 +00001057static void io_req_task_queue(struct io_kiocb *req);
Jens Axboe65453d12021-02-10 00:03:21 +00001058static void io_submit_flush_completions(struct io_comp_state *cs,
1059 struct io_ring_ctx *ctx);
Jens Axboe9a56a232019-01-09 09:06:50 -07001060
Jens Axboe2b188cc2019-01-07 10:46:33 -07001061static struct kmem_cache *req_cachep;
1062
Jens Axboe09186822020-10-13 15:01:40 -06001063static const struct file_operations io_uring_fops;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001064
1065struct sock *io_uring_get_socket(struct file *file)
1066{
1067#if defined(CONFIG_UNIX)
1068 if (file->f_op == &io_uring_fops) {
1069 struct io_ring_ctx *ctx = file->private_data;
1070
1071 return ctx->ring_sock->sk;
1072 }
1073#endif
1074 return NULL;
1075}
1076EXPORT_SYMBOL(io_uring_get_socket);
1077
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001078#define io_for_each_link(pos, head) \
1079 for (pos = (head); pos; pos = pos->link)
1080
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001081static inline void io_clean_op(struct io_kiocb *req)
1082{
Pavel Begunkov9d5c8192021-01-24 15:08:14 +00001083 if (req->flags & (REQ_F_NEED_CLEANUP | REQ_F_BUFFER_SELECTED))
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03001084 __io_clean_op(req);
1085}
1086
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001087static inline void io_set_resource_node(struct io_kiocb *req)
Jens Axboec40f6372020-06-25 15:39:59 -06001088{
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001089 struct io_ring_ctx *ctx = req->ctx;
1090
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001091 if (!req->fixed_rsrc_refs) {
1092 req->fixed_rsrc_refs = &ctx->file_data->node->refs;
1093 percpu_ref_get(req->fixed_rsrc_refs);
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00001094 }
1095}
1096
Pavel Begunkov08d23632020-11-06 13:00:22 +00001097static bool io_match_task(struct io_kiocb *head,
1098 struct task_struct *task,
1099 struct files_struct *files)
1100{
1101 struct io_kiocb *req;
1102
Jens Axboe84965ff2021-01-23 15:51:11 -07001103 if (task && head->task != task) {
1104 /* in terms of cancelation, always match if req task is dead */
1105 if (head->task->flags & PF_EXITING)
1106 return true;
Pavel Begunkov08d23632020-11-06 13:00:22 +00001107 return false;
Jens Axboe84965ff2021-01-23 15:51:11 -07001108 }
Pavel Begunkov08d23632020-11-06 13:00:22 +00001109 if (!files)
1110 return true;
1111
1112 io_for_each_link(req, head) {
Jens Axboe02a13672021-01-23 15:49:31 -07001113 if (!(req->flags & REQ_F_WORK_INITIALIZED))
1114 continue;
1115 if (req->file && req->file->f_op == &io_uring_fops)
1116 return true;
1117 if ((req->work.flags & IO_WQ_WORK_FILES) &&
Pavel Begunkov08d23632020-11-06 13:00:22 +00001118 req->work.identity->files == files)
1119 return true;
1120 }
1121 return false;
1122}
1123
Jens Axboe28cea78a2020-09-14 10:51:17 -06001124static void io_sq_thread_drop_mm_files(void)
Jens Axboec40f6372020-06-25 15:39:59 -06001125{
Jens Axboe28cea78a2020-09-14 10:51:17 -06001126 struct files_struct *files = current->files;
Jens Axboec40f6372020-06-25 15:39:59 -06001127 struct mm_struct *mm = current->mm;
1128
1129 if (mm) {
1130 kthread_unuse_mm(mm);
1131 mmput(mm);
Jens Axboe4b70cf92020-11-02 10:39:05 -07001132 current->mm = NULL;
Jens Axboec40f6372020-06-25 15:39:59 -06001133 }
Jens Axboe28cea78a2020-09-14 10:51:17 -06001134 if (files) {
1135 struct nsproxy *nsproxy = current->nsproxy;
1136
1137 task_lock(current);
1138 current->files = NULL;
1139 current->nsproxy = NULL;
1140 task_unlock(current);
1141 put_files_struct(files);
1142 put_nsproxy(nsproxy);
1143 }
1144}
1145
Pavel Begunkov1a38ffc2020-11-08 12:55:55 +00001146static int __io_sq_thread_acquire_files(struct io_ring_ctx *ctx)
Jens Axboe28cea78a2020-09-14 10:51:17 -06001147{
1148 if (!current->files) {
1149 struct files_struct *files;
1150 struct nsproxy *nsproxy;
1151
1152 task_lock(ctx->sqo_task);
1153 files = ctx->sqo_task->files;
1154 if (!files) {
1155 task_unlock(ctx->sqo_task);
Pavel Begunkov1a38ffc2020-11-08 12:55:55 +00001156 return -EOWNERDEAD;
Jens Axboe28cea78a2020-09-14 10:51:17 -06001157 }
1158 atomic_inc(&files->count);
1159 get_nsproxy(ctx->sqo_task->nsproxy);
1160 nsproxy = ctx->sqo_task->nsproxy;
1161 task_unlock(ctx->sqo_task);
1162
1163 task_lock(current);
1164 current->files = files;
1165 current->nsproxy = nsproxy;
1166 task_unlock(current);
1167 }
Pavel Begunkov1a38ffc2020-11-08 12:55:55 +00001168 return 0;
Jens Axboec40f6372020-06-25 15:39:59 -06001169}
1170
1171static int __io_sq_thread_acquire_mm(struct io_ring_ctx *ctx)
1172{
Jens Axboe4b70cf92020-11-02 10:39:05 -07001173 struct mm_struct *mm;
1174
1175 if (current->mm)
1176 return 0;
1177
Jens Axboe4b70cf92020-11-02 10:39:05 -07001178 task_lock(ctx->sqo_task);
1179 mm = ctx->sqo_task->mm;
1180 if (unlikely(!mm || !mmget_not_zero(mm)))
1181 mm = NULL;
1182 task_unlock(ctx->sqo_task);
1183
1184 if (mm) {
1185 kthread_use_mm(mm);
1186 return 0;
Jens Axboec40f6372020-06-25 15:39:59 -06001187 }
1188
Jens Axboe4b70cf92020-11-02 10:39:05 -07001189 return -EFAULT;
Jens Axboec40f6372020-06-25 15:39:59 -06001190}
1191
Pavel Begunkov4e326352021-02-12 03:23:52 +00001192static int __io_sq_thread_acquire_mm_files(struct io_ring_ctx *ctx,
1193 struct io_kiocb *req)
Jens Axboec40f6372020-06-25 15:39:59 -06001194{
Jens Axboe28cea78a2020-09-14 10:51:17 -06001195 const struct io_op_def *def = &io_op_defs[req->opcode];
Pavel Begunkov1a38ffc2020-11-08 12:55:55 +00001196 int ret;
Jens Axboe28cea78a2020-09-14 10:51:17 -06001197
1198 if (def->work_flags & IO_WQ_WORK_MM) {
Pavel Begunkov1a38ffc2020-11-08 12:55:55 +00001199 ret = __io_sq_thread_acquire_mm(ctx);
Jens Axboe28cea78a2020-09-14 10:51:17 -06001200 if (unlikely(ret))
1201 return ret;
1202 }
1203
Pavel Begunkov1a38ffc2020-11-08 12:55:55 +00001204 if (def->needs_file || (def->work_flags & IO_WQ_WORK_FILES)) {
1205 ret = __io_sq_thread_acquire_files(ctx);
1206 if (unlikely(ret))
1207 return ret;
1208 }
Jens Axboe28cea78a2020-09-14 10:51:17 -06001209
1210 return 0;
Jens Axboec40f6372020-06-25 15:39:59 -06001211}
1212
Pavel Begunkov4e326352021-02-12 03:23:52 +00001213static inline int io_sq_thread_acquire_mm_files(struct io_ring_ctx *ctx,
1214 struct io_kiocb *req)
1215{
1216 if (unlikely(current->flags & PF_EXITING))
1217 return -EFAULT;
1218 if (!(ctx->flags & IORING_SETUP_SQPOLL))
1219 return 0;
1220 return __io_sq_thread_acquire_mm_files(ctx, req);
1221}
1222
Dennis Zhou91d8f512020-09-16 13:41:05 -07001223static void io_sq_thread_associate_blkcg(struct io_ring_ctx *ctx,
1224 struct cgroup_subsys_state **cur_css)
1225
1226{
1227#ifdef CONFIG_BLK_CGROUP
1228 /* puts the old one when swapping */
1229 if (*cur_css != ctx->sqo_blkcg_css) {
1230 kthread_associate_blkcg(ctx->sqo_blkcg_css);
1231 *cur_css = ctx->sqo_blkcg_css;
1232 }
1233#endif
1234}
1235
1236static void io_sq_thread_unassociate_blkcg(void)
1237{
1238#ifdef CONFIG_BLK_CGROUP
1239 kthread_associate_blkcg(NULL);
1240#endif
1241}
1242
Jens Axboec40f6372020-06-25 15:39:59 -06001243static inline void req_set_fail_links(struct io_kiocb *req)
1244{
1245 if ((req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) == REQ_F_LINK)
1246 req->flags |= REQ_F_FAIL_LINK;
1247}
Jens Axboe4a38aed22020-05-14 17:21:15 -06001248
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08001249/*
Jens Axboe1e6fa522020-10-15 08:46:24 -06001250 * None of these are dereferenced, they are simply used to check if any of
1251 * them have changed. If we're under current and check they are still the
1252 * same, we're fine to grab references to them for actual out-of-line use.
1253 */
1254static void io_init_identity(struct io_identity *id)
1255{
1256 id->files = current->files;
1257 id->mm = current->mm;
1258#ifdef CONFIG_BLK_CGROUP
1259 rcu_read_lock();
1260 id->blkcg_css = blkcg_css();
1261 rcu_read_unlock();
1262#endif
1263 id->creds = current_cred();
1264 id->nsproxy = current->nsproxy;
1265 id->fs = current->fs;
1266 id->fsize = rlimit(RLIMIT_FSIZE);
Jens Axboe4ea33a92020-10-15 13:46:44 -06001267#ifdef CONFIG_AUDIT
1268 id->loginuid = current->loginuid;
1269 id->sessionid = current->sessionid;
1270#endif
Jens Axboe1e6fa522020-10-15 08:46:24 -06001271 refcount_set(&id->count, 1);
1272}
1273
Pavel Begunkovec99ca62020-10-18 10:17:38 +01001274static inline void __io_req_init_async(struct io_kiocb *req)
1275{
1276 memset(&req->work, 0, sizeof(req->work));
1277 req->flags |= REQ_F_WORK_INITIALIZED;
1278}
1279
Jens Axboe1e6fa522020-10-15 08:46:24 -06001280/*
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08001281 * Note: must call io_req_init_async() for the first time you
1282 * touch any members of io_wq_work.
1283 */
1284static inline void io_req_init_async(struct io_kiocb *req)
1285{
Jens Axboe500a3732020-10-15 17:38:03 -06001286 struct io_uring_task *tctx = current->io_uring;
1287
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08001288 if (req->flags & REQ_F_WORK_INITIALIZED)
1289 return;
1290
Pavel Begunkovec99ca62020-10-18 10:17:38 +01001291 __io_req_init_async(req);
Jens Axboe500a3732020-10-15 17:38:03 -06001292
1293 /* Grab a ref if this isn't our static identity */
1294 req->work.identity = tctx->identity;
1295 if (tctx->identity != &tctx->__identity)
1296 refcount_inc(&req->work.identity->count);
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08001297}
1298
Jens Axboe2b188cc2019-01-07 10:46:33 -07001299static void io_ring_ctx_ref_free(struct percpu_ref *ref)
1300{
1301 struct io_ring_ctx *ctx = container_of(ref, struct io_ring_ctx, refs);
1302
Jens Axboe0f158b42020-05-14 17:18:39 -06001303 complete(&ctx->ref_comp);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001304}
1305
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03001306static inline bool io_is_timeout_noseq(struct io_kiocb *req)
1307{
1308 return !req->timeout.off;
1309}
1310
Jens Axboe2b188cc2019-01-07 10:46:33 -07001311static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
1312{
1313 struct io_ring_ctx *ctx;
Jens Axboe78076bb2019-12-04 19:56:40 -07001314 int hash_bits;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001315
1316 ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
1317 if (!ctx)
1318 return NULL;
1319
Jens Axboe78076bb2019-12-04 19:56:40 -07001320 /*
1321 * Use 5 bits less than the max cq entries, that should give us around
1322 * 32 entries per hash list if totally full and uniformly spread.
1323 */
1324 hash_bits = ilog2(p->cq_entries);
1325 hash_bits -= 5;
1326 if (hash_bits <= 0)
1327 hash_bits = 1;
1328 ctx->cancel_hash_bits = hash_bits;
1329 ctx->cancel_hash = kmalloc((1U << hash_bits) * sizeof(struct hlist_head),
1330 GFP_KERNEL);
1331 if (!ctx->cancel_hash)
1332 goto err;
1333 __hash_init(ctx->cancel_hash, 1U << hash_bits);
1334
Roman Gushchin21482892019-05-07 10:01:48 -07001335 if (percpu_ref_init(&ctx->refs, io_ring_ctx_ref_free,
Jens Axboe206aefd2019-11-07 18:27:42 -07001336 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL))
1337 goto err;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001338
1339 ctx->flags = p->flags;
Jens Axboe90554202020-09-03 12:12:41 -06001340 init_waitqueue_head(&ctx->sqo_sq_wait);
Jens Axboe69fb2132020-09-14 11:16:23 -06001341 INIT_LIST_HEAD(&ctx->sqd_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001342 init_waitqueue_head(&ctx->cq_wait);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001343 INIT_LIST_HEAD(&ctx->cq_overflow_list);
Jens Axboe0f158b42020-05-14 17:18:39 -06001344 init_completion(&ctx->ref_comp);
1345 init_completion(&ctx->sq_thread_comp);
Jens Axboe5a2e7452020-02-23 16:23:11 -07001346 idr_init(&ctx->io_buffer_idr);
Jens Axboe071698e2020-01-28 10:04:42 -07001347 idr_init(&ctx->personality_idr);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001348 mutex_init(&ctx->uring_lock);
1349 init_waitqueue_head(&ctx->wait);
1350 spin_lock_init(&ctx->completion_lock);
Pavel Begunkov540e32a2020-07-13 23:37:09 +03001351 INIT_LIST_HEAD(&ctx->iopoll_list);
Jens Axboede0617e2019-04-06 21:51:27 -06001352 INIT_LIST_HEAD(&ctx->defer_list);
Jens Axboe5262f562019-09-17 12:26:57 -06001353 INIT_LIST_HEAD(&ctx->timeout_list);
Jens Axboefcb323c2019-10-24 12:39:47 -06001354 spin_lock_init(&ctx->inflight_lock);
1355 INIT_LIST_HEAD(&ctx->inflight_list);
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00001356 spin_lock_init(&ctx->rsrc_ref_lock);
1357 INIT_LIST_HEAD(&ctx->rsrc_ref_list);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00001358 INIT_DELAYED_WORK(&ctx->rsrc_put_work, io_rsrc_put_work);
1359 init_llist_head(&ctx->rsrc_put_llist);
Jens Axboe1b4c3512021-02-10 00:03:19 +00001360 INIT_LIST_HEAD(&ctx->submit_state.comp.free_list);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001361 INIT_LIST_HEAD(&ctx->submit_state.comp.locked_free_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001362 return ctx;
Jens Axboe206aefd2019-11-07 18:27:42 -07001363err:
Jens Axboe78076bb2019-12-04 19:56:40 -07001364 kfree(ctx->cancel_hash);
Jens Axboe206aefd2019-11-07 18:27:42 -07001365 kfree(ctx);
1366 return NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001367}
1368
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001369static bool req_need_defer(struct io_kiocb *req, u32 seq)
Jens Axboede0617e2019-04-06 21:51:27 -06001370{
Jens Axboe2bc99302020-07-09 09:43:27 -06001371 if (unlikely(req->flags & REQ_F_IO_DRAIN)) {
1372 struct io_ring_ctx *ctx = req->ctx;
Jackie Liua197f662019-11-08 08:09:12 -07001373
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001374 return seq != ctx->cached_cq_tail
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001375 + READ_ONCE(ctx->cached_cq_overflow);
Jens Axboe2bc99302020-07-09 09:43:27 -06001376 }
Jens Axboe7adf4ea2019-10-10 21:42:58 -06001377
Bob Liu9d858b22019-11-13 18:06:25 +08001378 return false;
Jens Axboe7adf4ea2019-10-10 21:42:58 -06001379}
1380
Jens Axboe5c3462c2020-10-15 09:02:33 -06001381static void io_put_identity(struct io_uring_task *tctx, struct io_kiocb *req)
Jens Axboe1e6fa522020-10-15 08:46:24 -06001382{
Jens Axboe500a3732020-10-15 17:38:03 -06001383 if (req->work.identity == &tctx->__identity)
Jens Axboe1e6fa522020-10-15 08:46:24 -06001384 return;
1385 if (refcount_dec_and_test(&req->work.identity->count))
1386 kfree(req->work.identity);
1387}
1388
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01001389static void io_req_clean_work(struct io_kiocb *req)
Jens Axboecccf0ee2020-01-27 16:34:48 -07001390{
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08001391 if (!(req->flags & REQ_F_WORK_INITIALIZED))
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01001392 return;
Jens Axboe51a4cc12020-08-10 10:55:56 -06001393
Pavel Begunkove86d0042021-02-01 18:59:54 +00001394 if (req->work.flags & IO_WQ_WORK_MM)
Jens Axboe98447d62020-10-14 10:48:51 -06001395 mmdrop(req->work.identity->mm);
Dennis Zhou91d8f512020-09-16 13:41:05 -07001396#ifdef CONFIG_BLK_CGROUP
Pavel Begunkove86d0042021-02-01 18:59:54 +00001397 if (req->work.flags & IO_WQ_WORK_BLKCG)
Jens Axboe98447d62020-10-14 10:48:51 -06001398 css_put(req->work.identity->blkcg_css);
Jens Axboedfead8a2020-10-14 10:12:37 -06001399#endif
Pavel Begunkove86d0042021-02-01 18:59:54 +00001400 if (req->work.flags & IO_WQ_WORK_CREDS)
Jens Axboe98447d62020-10-14 10:48:51 -06001401 put_cred(req->work.identity->creds);
Jens Axboedfead8a2020-10-14 10:12:37 -06001402 if (req->work.flags & IO_WQ_WORK_FS) {
Jens Axboe98447d62020-10-14 10:48:51 -06001403 struct fs_struct *fs = req->work.identity->fs;
Jens Axboeff002b32020-02-07 16:05:21 -07001404
Jens Axboe98447d62020-10-14 10:48:51 -06001405 spin_lock(&req->work.identity->fs->lock);
Jens Axboeff002b32020-02-07 16:05:21 -07001406 if (--fs->users)
1407 fs = NULL;
Jens Axboe98447d62020-10-14 10:48:51 -06001408 spin_unlock(&req->work.identity->fs->lock);
Jens Axboeff002b32020-02-07 16:05:21 -07001409 if (fs)
1410 free_fs_struct(fs);
1411 }
Pavel Begunkov34e08fe2021-02-01 18:59:53 +00001412 if (req->work.flags & IO_WQ_WORK_FILES) {
1413 put_files_struct(req->work.identity->files);
1414 put_nsproxy(req->work.identity->nsproxy);
Pavel Begunkov34e08fe2021-02-01 18:59:53 +00001415 }
1416 if (req->flags & REQ_F_INFLIGHT) {
1417 struct io_ring_ctx *ctx = req->ctx;
1418 struct io_uring_task *tctx = req->task->io_uring;
1419 unsigned long flags;
1420
1421 spin_lock_irqsave(&ctx->inflight_lock, flags);
1422 list_del(&req->inflight_entry);
1423 spin_unlock_irqrestore(&ctx->inflight_lock, flags);
1424 req->flags &= ~REQ_F_INFLIGHT;
1425 if (atomic_read(&tctx->in_idle))
1426 wake_up(&tctx->wait);
1427 }
Jens Axboe51a4cc12020-08-10 10:55:56 -06001428
Pavel Begunkove86d0042021-02-01 18:59:54 +00001429 req->flags &= ~REQ_F_WORK_INITIALIZED;
1430 req->work.flags &= ~(IO_WQ_WORK_MM | IO_WQ_WORK_BLKCG | IO_WQ_WORK_FS |
1431 IO_WQ_WORK_CREDS | IO_WQ_WORK_FILES);
Jens Axboe5c3462c2020-10-15 09:02:33 -06001432 io_put_identity(req->task->io_uring, req);
Jens Axboe1e6fa522020-10-15 08:46:24 -06001433}
1434
1435/*
1436 * Create a private copy of io_identity, since some fields don't match
1437 * the current context.
1438 */
1439static bool io_identity_cow(struct io_kiocb *req)
1440{
Jens Axboe5c3462c2020-10-15 09:02:33 -06001441 struct io_uring_task *tctx = current->io_uring;
Jens Axboe1e6fa522020-10-15 08:46:24 -06001442 const struct cred *creds = NULL;
1443 struct io_identity *id;
1444
1445 if (req->work.flags & IO_WQ_WORK_CREDS)
1446 creds = req->work.identity->creds;
1447
1448 id = kmemdup(req->work.identity, sizeof(*id), GFP_KERNEL);
1449 if (unlikely(!id)) {
1450 req->work.flags |= IO_WQ_WORK_CANCEL;
1451 return false;
1452 }
1453
1454 /*
1455 * We can safely just re-init the creds we copied Either the field
1456 * matches the current one, or we haven't grabbed it yet. The only
1457 * exception is ->creds, through registered personalities, so handle
1458 * that one separately.
1459 */
1460 io_init_identity(id);
1461 if (creds)
Pavel Begunkove8c954d2020-12-06 22:22:46 +00001462 id->creds = creds;
Jens Axboe1e6fa522020-10-15 08:46:24 -06001463
1464 /* add one for this request */
1465 refcount_inc(&id->count);
1466
Jens Axboecb8a8ae2020-11-03 12:19:07 -07001467 /* drop tctx and req identity references, if needed */
1468 if (tctx->identity != &tctx->__identity &&
1469 refcount_dec_and_test(&tctx->identity->count))
1470 kfree(tctx->identity);
1471 if (req->work.identity != &tctx->__identity &&
1472 refcount_dec_and_test(&req->work.identity->count))
Jens Axboe1e6fa522020-10-15 08:46:24 -06001473 kfree(req->work.identity);
1474
1475 req->work.identity = id;
Jens Axboe500a3732020-10-15 17:38:03 -06001476 tctx->identity = id;
Jens Axboe1e6fa522020-10-15 08:46:24 -06001477 return true;
1478}
1479
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00001480static void io_req_track_inflight(struct io_kiocb *req)
1481{
1482 struct io_ring_ctx *ctx = req->ctx;
1483
1484 if (!(req->flags & REQ_F_INFLIGHT)) {
1485 io_req_init_async(req);
1486 req->flags |= REQ_F_INFLIGHT;
1487
1488 spin_lock_irq(&ctx->inflight_lock);
1489 list_add(&req->inflight_entry, &ctx->inflight_list);
1490 spin_unlock_irq(&ctx->inflight_lock);
1491 }
1492}
1493
Jens Axboe1e6fa522020-10-15 08:46:24 -06001494static bool io_grab_identity(struct io_kiocb *req)
1495{
1496 const struct io_op_def *def = &io_op_defs[req->opcode];
Jens Axboe5c3462c2020-10-15 09:02:33 -06001497 struct io_identity *id = req->work.identity;
Jens Axboe1e6fa522020-10-15 08:46:24 -06001498
Jens Axboe69228332020-10-20 14:28:41 -06001499 if (def->work_flags & IO_WQ_WORK_FSIZE) {
1500 if (id->fsize != rlimit(RLIMIT_FSIZE))
1501 return false;
1502 req->work.flags |= IO_WQ_WORK_FSIZE;
1503 }
Jens Axboe1e6fa522020-10-15 08:46:24 -06001504#ifdef CONFIG_BLK_CGROUP
1505 if (!(req->work.flags & IO_WQ_WORK_BLKCG) &&
1506 (def->work_flags & IO_WQ_WORK_BLKCG)) {
1507 rcu_read_lock();
1508 if (id->blkcg_css != blkcg_css()) {
1509 rcu_read_unlock();
1510 return false;
1511 }
1512 /*
1513 * This should be rare, either the cgroup is dying or the task
1514 * is moving cgroups. Just punt to root for the handful of ios.
1515 */
1516 if (css_tryget_online(id->blkcg_css))
1517 req->work.flags |= IO_WQ_WORK_BLKCG;
1518 rcu_read_unlock();
1519 }
1520#endif
1521 if (!(req->work.flags & IO_WQ_WORK_CREDS)) {
1522 if (id->creds != current_cred())
1523 return false;
1524 get_cred(id->creds);
1525 req->work.flags |= IO_WQ_WORK_CREDS;
1526 }
Jens Axboe4ea33a92020-10-15 13:46:44 -06001527#ifdef CONFIG_AUDIT
1528 if (!uid_eq(current->loginuid, id->loginuid) ||
1529 current->sessionid != id->sessionid)
1530 return false;
1531#endif
Jens Axboe1e6fa522020-10-15 08:46:24 -06001532 if (!(req->work.flags & IO_WQ_WORK_FS) &&
1533 (def->work_flags & IO_WQ_WORK_FS)) {
1534 if (current->fs != id->fs)
1535 return false;
1536 spin_lock(&id->fs->lock);
1537 if (!id->fs->in_exec) {
1538 id->fs->users++;
1539 req->work.flags |= IO_WQ_WORK_FS;
1540 } else {
1541 req->work.flags |= IO_WQ_WORK_CANCEL;
1542 }
1543 spin_unlock(&current->fs->lock);
1544 }
Pavel Begunkovaf604702020-11-25 18:41:28 +00001545 if (!(req->work.flags & IO_WQ_WORK_FILES) &&
1546 (def->work_flags & IO_WQ_WORK_FILES) &&
1547 !(req->flags & REQ_F_NO_FILE_TABLE)) {
1548 if (id->files != current->files ||
1549 id->nsproxy != current->nsproxy)
1550 return false;
1551 atomic_inc(&id->files->count);
1552 get_nsproxy(id->nsproxy);
Pavel Begunkovaf604702020-11-25 18:41:28 +00001553 req->work.flags |= IO_WQ_WORK_FILES;
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00001554 io_req_track_inflight(req);
Pavel Begunkovaf604702020-11-25 18:41:28 +00001555 }
Jens Axboe77788772020-12-29 10:50:46 -07001556 if (!(req->work.flags & IO_WQ_WORK_MM) &&
1557 (def->work_flags & IO_WQ_WORK_MM)) {
1558 if (id->mm != current->mm)
1559 return false;
1560 mmgrab(id->mm);
1561 req->work.flags |= IO_WQ_WORK_MM;
1562 }
Jens Axboe1e6fa522020-10-15 08:46:24 -06001563
1564 return true;
Jens Axboe561fb042019-10-24 07:25:42 -06001565}
1566
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001567static void io_prep_async_work(struct io_kiocb *req)
Jens Axboe561fb042019-10-24 07:25:42 -06001568{
Jens Axboed3656342019-12-18 09:50:26 -07001569 const struct io_op_def *def = &io_op_defs[req->opcode];
Pavel Begunkov23329512020-10-10 18:34:06 +01001570 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe54a91f32019-09-10 09:15:04 -06001571
Pavel Begunkov16d59802020-07-12 16:16:47 +03001572 io_req_init_async(req);
1573
Pavel Begunkovfeaadc42020-10-22 16:47:16 +01001574 if (req->flags & REQ_F_FORCE_ASYNC)
1575 req->work.flags |= IO_WQ_WORK_CONCURRENT;
1576
Jens Axboed3656342019-12-18 09:50:26 -07001577 if (req->flags & REQ_F_ISREG) {
Pavel Begunkov23329512020-10-10 18:34:06 +01001578 if (def->hash_reg_file || (ctx->flags & IORING_SETUP_IOPOLL))
Pavel Begunkov8766dd52020-03-14 00:31:04 +03001579 io_wq_hash_work(&req->work, file_inode(req->file));
Jens Axboed3656342019-12-18 09:50:26 -07001580 } else {
1581 if (def->unbound_nonreg_file)
Jens Axboe3529d8c2019-12-19 18:24:38 -07001582 req->work.flags |= IO_WQ_WORK_UNBOUND;
Jens Axboe54a91f32019-09-10 09:15:04 -06001583 }
Pavel Begunkov23329512020-10-10 18:34:06 +01001584
Jens Axboe1e6fa522020-10-15 08:46:24 -06001585 /* if we fail grabbing identity, we must COW, regrab, and retry */
1586 if (io_grab_identity(req))
1587 return;
1588
1589 if (!io_identity_cow(req))
1590 return;
1591
1592 /* can't fail at this point */
1593 if (!io_grab_identity(req))
1594 WARN_ON(1);
Jens Axboe561fb042019-10-24 07:25:42 -06001595}
1596
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001597static void io_prep_async_link(struct io_kiocb *req)
1598{
1599 struct io_kiocb *cur;
1600
Pavel Begunkovf2f87372020-10-27 23:25:37 +00001601 io_for_each_link(cur, req)
1602 io_prep_async_work(cur);
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001603}
1604
Jens Axboe7271ef32020-08-10 09:55:22 -06001605static struct io_kiocb *__io_queue_async_work(struct io_kiocb *req)
Jens Axboe561fb042019-10-24 07:25:42 -06001606{
Jackie Liua197f662019-11-08 08:09:12 -07001607 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001608 struct io_kiocb *link = io_prep_linked_timeout(req);
Jens Axboe561fb042019-10-24 07:25:42 -06001609
Pavel Begunkov8766dd52020-03-14 00:31:04 +03001610 trace_io_uring_queue_async_work(ctx, io_wq_is_hashed(&req->work), req,
1611 &req->work, req->flags);
1612 io_wq_enqueue(ctx->io_wq, &req->work);
Jens Axboe7271ef32020-08-10 09:55:22 -06001613 return link;
Jens Axboe18d9be12019-09-10 09:13:05 -06001614}
1615
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001616static void io_queue_async_work(struct io_kiocb *req)
1617{
Jens Axboe7271ef32020-08-10 09:55:22 -06001618 struct io_kiocb *link;
1619
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001620 /* init ->work of the whole link before punting */
1621 io_prep_async_link(req);
Jens Axboe7271ef32020-08-10 09:55:22 -06001622 link = __io_queue_async_work(req);
1623
1624 if (link)
1625 io_queue_linked_timeout(link);
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03001626}
1627
Jens Axboe5262f562019-09-17 12:26:57 -06001628static void io_kill_timeout(struct io_kiocb *req)
1629{
Jens Axboee8c2bc12020-08-15 18:44:09 -07001630 struct io_timeout_data *io = req->async_data;
Jens Axboe5262f562019-09-17 12:26:57 -06001631 int ret;
1632
Jens Axboee8c2bc12020-08-15 18:44:09 -07001633 ret = hrtimer_try_to_cancel(&io->timer);
Jens Axboe5262f562019-09-17 12:26:57 -06001634 if (ret != -1) {
Pavel Begunkov01cec8c2020-07-30 18:43:50 +03001635 atomic_set(&req->ctx->cq_timeouts,
1636 atomic_read(&req->ctx->cq_timeouts) + 1);
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001637 list_del_init(&req->timeout.list);
Jens Axboe78e19bb2019-11-06 15:21:34 -07001638 io_cqring_fill_event(req, 0);
Pavel Begunkov216578e2020-10-13 09:44:00 +01001639 io_put_req_deferred(req, 1);
Jens Axboe5262f562019-09-17 12:26:57 -06001640 }
1641}
1642
Jens Axboe76e1b642020-09-26 15:05:03 -06001643/*
1644 * Returns true if we found and killed one or more timeouts
1645 */
Pavel Begunkov6b819282020-11-06 13:00:25 +00001646static bool io_kill_timeouts(struct io_ring_ctx *ctx, struct task_struct *tsk,
1647 struct files_struct *files)
Jens Axboe5262f562019-09-17 12:26:57 -06001648{
1649 struct io_kiocb *req, *tmp;
Jens Axboe76e1b642020-09-26 15:05:03 -06001650 int canceled = 0;
Jens Axboe5262f562019-09-17 12:26:57 -06001651
1652 spin_lock_irq(&ctx->completion_lock);
Jens Axboef3606e32020-09-22 08:18:24 -06001653 list_for_each_entry_safe(req, tmp, &ctx->timeout_list, timeout.list) {
Pavel Begunkov6b819282020-11-06 13:00:25 +00001654 if (io_match_task(req, tsk, files)) {
Jens Axboef3606e32020-09-22 08:18:24 -06001655 io_kill_timeout(req);
Jens Axboe76e1b642020-09-26 15:05:03 -06001656 canceled++;
1657 }
Jens Axboef3606e32020-09-22 08:18:24 -06001658 }
Jens Axboe5262f562019-09-17 12:26:57 -06001659 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe76e1b642020-09-26 15:05:03 -06001660 return canceled != 0;
Jens Axboe5262f562019-09-17 12:26:57 -06001661}
1662
Pavel Begunkov04518942020-05-26 20:34:05 +03001663static void __io_queue_deferred(struct io_ring_ctx *ctx)
1664{
1665 do {
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001666 struct io_defer_entry *de = list_first_entry(&ctx->defer_list,
1667 struct io_defer_entry, list);
Pavel Begunkov04518942020-05-26 20:34:05 +03001668
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03001669 if (req_need_defer(de->req, de->seq))
Pavel Begunkov04518942020-05-26 20:34:05 +03001670 break;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001671 list_del_init(&de->list);
Pavel Begunkov907d1df2021-01-26 23:35:10 +00001672 io_req_task_queue(de->req);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03001673 kfree(de);
Pavel Begunkov04518942020-05-26 20:34:05 +03001674 } while (!list_empty(&ctx->defer_list));
1675}
1676
Pavel Begunkov360428f2020-05-30 14:54:17 +03001677static void io_flush_timeouts(struct io_ring_ctx *ctx)
1678{
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001679 u32 seq;
1680
1681 if (list_empty(&ctx->timeout_list))
1682 return;
1683
1684 seq = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
1685
1686 do {
1687 u32 events_needed, events_got;
Pavel Begunkov360428f2020-05-30 14:54:17 +03001688 struct io_kiocb *req = list_first_entry(&ctx->timeout_list,
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001689 struct io_kiocb, timeout.list);
Pavel Begunkov360428f2020-05-30 14:54:17 +03001690
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03001691 if (io_is_timeout_noseq(req))
Pavel Begunkov360428f2020-05-30 14:54:17 +03001692 break;
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001693
1694 /*
1695 * Since seq can easily wrap around over time, subtract
1696 * the last seq at which timeouts were flushed before comparing.
1697 * Assuming not more than 2^31-1 events have happened since,
1698 * these subtractions won't have wrapped, so we can check if
1699 * target is in [last_seq, current_seq] by comparing the two.
1700 */
1701 events_needed = req->timeout.target_seq - ctx->cq_last_tm_flush;
1702 events_got = seq - ctx->cq_last_tm_flush;
1703 if (events_got < events_needed)
Pavel Begunkov360428f2020-05-30 14:54:17 +03001704 break;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03001705
Pavel Begunkov135fcde2020-07-13 23:37:12 +03001706 list_del_init(&req->timeout.list);
Pavel Begunkov360428f2020-05-30 14:54:17 +03001707 io_kill_timeout(req);
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05001708 } while (!list_empty(&ctx->timeout_list));
1709
1710 ctx->cq_last_tm_flush = seq;
Pavel Begunkov360428f2020-05-30 14:54:17 +03001711}
1712
Jens Axboede0617e2019-04-06 21:51:27 -06001713static void io_commit_cqring(struct io_ring_ctx *ctx)
1714{
Pavel Begunkov360428f2020-05-30 14:54:17 +03001715 io_flush_timeouts(ctx);
Pavel Begunkovec30e042021-01-19 13:32:38 +00001716
1717 /* order cqe stores with ring update */
1718 smp_store_release(&ctx->rings->cq.tail, ctx->cached_cq_tail);
Jens Axboede0617e2019-04-06 21:51:27 -06001719
Pavel Begunkov04518942020-05-26 20:34:05 +03001720 if (unlikely(!list_empty(&ctx->defer_list)))
1721 __io_queue_deferred(ctx);
Jens Axboede0617e2019-04-06 21:51:27 -06001722}
1723
Jens Axboe90554202020-09-03 12:12:41 -06001724static inline bool io_sqring_full(struct io_ring_ctx *ctx)
1725{
1726 struct io_rings *r = ctx->rings;
1727
1728 return READ_ONCE(r->sq.tail) - ctx->cached_sq_head == r->sq_ring_entries;
1729}
1730
Pavel Begunkov888aae22021-01-19 13:32:39 +00001731static inline unsigned int __io_cqring_events(struct io_ring_ctx *ctx)
1732{
1733 return ctx->cached_cq_tail - READ_ONCE(ctx->rings->cq.head);
1734}
1735
Jens Axboe2b188cc2019-01-07 10:46:33 -07001736static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx)
1737{
Hristo Venev75b28af2019-08-26 17:23:46 +00001738 struct io_rings *rings = ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001739 unsigned tail;
1740
Stefan Bühler115e12e2019-04-24 23:54:18 +02001741 /*
1742 * writes to the cq entry need to come after reading head; the
1743 * control dependency is enough as we're using WRITE_ONCE to
1744 * fill the cq entry
1745 */
Pavel Begunkov888aae22021-01-19 13:32:39 +00001746 if (__io_cqring_events(ctx) == rings->cq_ring_entries)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001747 return NULL;
1748
Pavel Begunkov888aae22021-01-19 13:32:39 +00001749 tail = ctx->cached_cq_tail++;
Hristo Venev75b28af2019-08-26 17:23:46 +00001750 return &rings->cqes[tail & ctx->cq_mask];
Jens Axboe2b188cc2019-01-07 10:46:33 -07001751}
1752
Jens Axboef2842ab2020-01-08 11:04:00 -07001753static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx)
1754{
Jens Axboef0b493e2020-02-01 21:30:11 -07001755 if (!ctx->cq_ev_fd)
1756 return false;
Stefano Garzarella7e55a192020-05-15 18:38:05 +02001757 if (READ_ONCE(ctx->rings->cq_flags) & IORING_CQ_EVENTFD_DISABLED)
1758 return false;
Jens Axboef2842ab2020-01-08 11:04:00 -07001759 if (!ctx->eventfd_async)
1760 return true;
Jens Axboeb41e9852020-02-17 09:52:41 -07001761 return io_wq_current_is_worker();
Jens Axboef2842ab2020-01-08 11:04:00 -07001762}
1763
Jens Axboeb41e9852020-02-17 09:52:41 -07001764static void io_cqring_ev_posted(struct io_ring_ctx *ctx)
Jens Axboe8c838782019-03-12 15:48:16 -06001765{
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001766 /* see waitqueue_active() comment */
1767 smp_mb();
1768
Jens Axboe8c838782019-03-12 15:48:16 -06001769 if (waitqueue_active(&ctx->wait))
1770 wake_up(&ctx->wait);
Jens Axboe534ca6d2020-09-02 13:52:19 -06001771 if (ctx->sq_data && waitqueue_active(&ctx->sq_data->wait))
1772 wake_up(&ctx->sq_data->wait);
Jens Axboeb41e9852020-02-17 09:52:41 -07001773 if (io_should_trigger_evfd(ctx))
Jens Axboe9b402842019-04-11 11:45:41 -06001774 eventfd_signal(ctx->cq_ev_fd, 1);
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001775 if (waitqueue_active(&ctx->cq_wait)) {
Pavel Begunkov4aa84f22021-01-07 03:15:42 +00001776 wake_up_interruptible(&ctx->cq_wait);
1777 kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
1778 }
Jens Axboe8c838782019-03-12 15:48:16 -06001779}
1780
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001781static void io_cqring_ev_posted_iopoll(struct io_ring_ctx *ctx)
1782{
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001783 /* see waitqueue_active() comment */
1784 smp_mb();
1785
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001786 if (ctx->flags & IORING_SETUP_SQPOLL) {
1787 if (waitqueue_active(&ctx->wait))
1788 wake_up(&ctx->wait);
1789 }
1790 if (io_should_trigger_evfd(ctx))
1791 eventfd_signal(ctx->cq_ev_fd, 1);
Pavel Begunkovb1445e52021-01-07 03:15:43 +00001792 if (waitqueue_active(&ctx->cq_wait)) {
Pavel Begunkov4aa84f22021-01-07 03:15:42 +00001793 wake_up_interruptible(&ctx->cq_wait);
1794 kill_fasync(&ctx->cq_fasync, SIGIO, POLL_IN);
1795 }
Pavel Begunkov80c18e42021-01-07 03:15:41 +00001796}
1797
Jens Axboec4a2ed72019-11-21 21:01:26 -07001798/* Returns true if there are no backlogged entries after the flush */
Pavel Begunkov6c503152021-01-04 20:36:36 +00001799static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
1800 struct task_struct *tsk,
1801 struct files_struct *files)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001802{
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001803 struct io_rings *rings = ctx->rings;
Jens Axboee6c8aa92020-09-28 13:10:13 -06001804 struct io_kiocb *req, *tmp;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001805 struct io_uring_cqe *cqe;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001806 unsigned long flags;
Jens Axboeb18032b2021-01-24 16:58:56 -07001807 bool all_flushed, posted;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001808 LIST_HEAD(list);
1809
Pavel Begunkove23de152020-12-17 00:24:37 +00001810 if (!force && __io_cqring_events(ctx) == rings->cq_ring_entries)
1811 return false;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001812
Jens Axboeb18032b2021-01-24 16:58:56 -07001813 posted = false;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001814 spin_lock_irqsave(&ctx->completion_lock, flags);
Jens Axboee6c8aa92020-09-28 13:10:13 -06001815 list_for_each_entry_safe(req, tmp, &ctx->cq_overflow_list, compl.list) {
Pavel Begunkov08d23632020-11-06 13:00:22 +00001816 if (!io_match_task(req, tsk, files))
Jens Axboee6c8aa92020-09-28 13:10:13 -06001817 continue;
1818
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001819 cqe = io_get_cqring(ctx);
1820 if (!cqe && !force)
1821 break;
1822
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001823 list_move(&req->compl.list, &list);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001824 if (cqe) {
1825 WRITE_ONCE(cqe->user_data, req->user_data);
1826 WRITE_ONCE(cqe->res, req->result);
Pavel Begunkov0f7e4662020-07-13 23:37:16 +03001827 WRITE_ONCE(cqe->flags, req->compl.cflags);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001828 } else {
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001829 ctx->cached_cq_overflow++;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001830 WRITE_ONCE(ctx->rings->cq_overflow,
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001831 ctx->cached_cq_overflow);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001832 }
Jens Axboeb18032b2021-01-24 16:58:56 -07001833 posted = true;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001834 }
1835
Pavel Begunkov09e88402020-12-17 00:24:38 +00001836 all_flushed = list_empty(&ctx->cq_overflow_list);
1837 if (all_flushed) {
1838 clear_bit(0, &ctx->sq_check_overflow);
1839 clear_bit(0, &ctx->cq_check_overflow);
1840 ctx->rings->sq_flags &= ~IORING_SQ_CQ_OVERFLOW;
1841 }
Pavel Begunkov46930142020-07-30 18:43:49 +03001842
Jens Axboeb18032b2021-01-24 16:58:56 -07001843 if (posted)
1844 io_commit_cqring(ctx);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001845 spin_unlock_irqrestore(&ctx->completion_lock, flags);
Jens Axboeb18032b2021-01-24 16:58:56 -07001846 if (posted)
1847 io_cqring_ev_posted(ctx);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001848
1849 while (!list_empty(&list)) {
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001850 req = list_first_entry(&list, struct io_kiocb, compl.list);
1851 list_del(&req->compl.list);
Jackie Liuec9c02a2019-11-08 23:50:36 +08001852 io_put_req(req);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001853 }
Jens Axboec4a2ed72019-11-21 21:01:26 -07001854
Pavel Begunkov09e88402020-12-17 00:24:38 +00001855 return all_flushed;
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001856}
1857
Pavel Begunkov6c503152021-01-04 20:36:36 +00001858static void io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
1859 struct task_struct *tsk,
1860 struct files_struct *files)
1861{
1862 if (test_bit(0, &ctx->cq_check_overflow)) {
1863 /* iopoll syncs against uring_lock, not completion_lock */
1864 if (ctx->flags & IORING_SETUP_IOPOLL)
1865 mutex_lock(&ctx->uring_lock);
1866 __io_cqring_overflow_flush(ctx, force, tsk, files);
1867 if (ctx->flags & IORING_SETUP_IOPOLL)
1868 mutex_unlock(&ctx->uring_lock);
1869 }
1870}
1871
Jens Axboebcda7ba2020-02-23 16:42:51 -07001872static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001873{
Jens Axboe78e19bb2019-11-06 15:21:34 -07001874 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001875 struct io_uring_cqe *cqe;
1876
Jens Axboe78e19bb2019-11-06 15:21:34 -07001877 trace_io_uring_complete(ctx, req->user_data, res);
Jens Axboe51c3ff62019-11-03 06:52:50 -07001878
Jens Axboe2b188cc2019-01-07 10:46:33 -07001879 /*
1880 * If we can't get a cq entry, userspace overflowed the
1881 * submission (by quite a lot). Increment the overflow count in
1882 * the ring.
1883 */
1884 cqe = io_get_cqring(ctx);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001885 if (likely(cqe)) {
Jens Axboe78e19bb2019-11-06 15:21:34 -07001886 WRITE_ONCE(cqe->user_data, req->user_data);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001887 WRITE_ONCE(cqe->res, res);
Jens Axboebcda7ba2020-02-23 16:42:51 -07001888 WRITE_ONCE(cqe->flags, cflags);
Jens Axboefdaf0832020-10-30 09:37:30 -06001889 } else if (ctx->cq_overflow_flushed ||
1890 atomic_read(&req->task->io_uring->in_idle)) {
Jens Axboe0f212202020-09-13 13:09:39 -06001891 /*
1892 * If we're in ring overflow flush mode, or in task cancel mode,
1893 * then we cannot store the request for later flushing, we need
1894 * to drop it on the floor.
1895 */
Pavel Begunkov2c3bac6d2020-10-18 10:17:40 +01001896 ctx->cached_cq_overflow++;
1897 WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001898 } else {
Jens Axboead3eb2c2019-12-18 17:12:20 -07001899 if (list_empty(&ctx->cq_overflow_list)) {
1900 set_bit(0, &ctx->sq_check_overflow);
1901 set_bit(0, &ctx->cq_check_overflow);
Xiaoguang Wang6d5f9042020-07-09 09:15:29 +08001902 ctx->rings->sq_flags |= IORING_SQ_CQ_OVERFLOW;
Jens Axboead3eb2c2019-12-18 17:12:20 -07001903 }
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001904 io_clean_op(req);
Jens Axboe1d7bb1d2019-11-06 11:31:17 -07001905 req->result = res;
Pavel Begunkov0f7e4662020-07-13 23:37:16 +03001906 req->compl.cflags = cflags;
Pavel Begunkov40d8ddd2020-07-13 23:37:11 +03001907 refcount_inc(&req->refs);
1908 list_add_tail(&req->compl.list, &ctx->cq_overflow_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001909 }
1910}
1911
Jens Axboebcda7ba2020-02-23 16:42:51 -07001912static void io_cqring_fill_event(struct io_kiocb *req, long res)
1913{
1914 __io_cqring_fill_event(req, res, 0);
1915}
1916
Jens Axboec7dae4b2021-02-09 19:53:37 -07001917static inline void io_req_complete_post(struct io_kiocb *req, long res,
1918 unsigned int cflags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07001919{
Jens Axboe78e19bb2019-11-06 15:21:34 -07001920 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001921 unsigned long flags;
1922
1923 spin_lock_irqsave(&ctx->completion_lock, flags);
Jens Axboebcda7ba2020-02-23 16:42:51 -07001924 __io_cqring_fill_event(req, res, cflags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07001925 io_commit_cqring(ctx);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001926 /*
1927 * If we're the last reference to this request, add to our locked
1928 * free_list cache.
1929 */
1930 if (refcount_dec_and_test(&req->refs)) {
1931 struct io_comp_state *cs = &ctx->submit_state.comp;
1932
1933 io_dismantle_req(req);
1934 io_put_task(req->task, 1);
1935 list_add(&req->compl.list, &cs->locked_free_list);
1936 cs->locked_free_nr++;
1937 } else
1938 req = NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07001939 spin_unlock_irqrestore(&ctx->completion_lock, flags);
1940
Jens Axboe8c838782019-03-12 15:48:16 -06001941 io_cqring_ev_posted(ctx);
Jens Axboec7dae4b2021-02-09 19:53:37 -07001942 if (req) {
1943 io_queue_next(req);
1944 percpu_ref_put(&ctx->refs);
1945 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07001946}
1947
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001948static void io_req_complete_state(struct io_kiocb *req, long res,
Pavel Begunkov889fca72021-02-10 00:03:09 +00001949 unsigned int cflags)
Jens Axboe229a7b62020-06-22 10:13:11 -06001950{
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001951 io_clean_op(req);
1952 req->result = res;
1953 req->compl.cflags = cflags;
Pavel Begunkove342c802021-01-19 13:32:47 +00001954 req->flags |= REQ_F_COMPLETE_INLINE;
Jens Axboee1e16092020-06-22 09:17:17 -06001955}
1956
Pavel Begunkov889fca72021-02-10 00:03:09 +00001957static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags,
1958 long res, unsigned cflags)
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001959{
Pavel Begunkov889fca72021-02-10 00:03:09 +00001960 if (issue_flags & IO_URING_F_COMPLETE_DEFER)
1961 io_req_complete_state(req, res, cflags);
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001962 else
Jens Axboec7dae4b2021-02-09 19:53:37 -07001963 io_req_complete_post(req, res, cflags);
Pavel Begunkova38d68d2021-01-19 13:32:45 +00001964}
1965
1966static inline void io_req_complete(struct io_kiocb *req, long res)
Jens Axboee1e16092020-06-22 09:17:17 -06001967{
Pavel Begunkov889fca72021-02-10 00:03:09 +00001968 __io_req_complete(req, 0, res, 0);
Jens Axboebcda7ba2020-02-23 16:42:51 -07001969}
1970
Jens Axboec7dae4b2021-02-09 19:53:37 -07001971static bool io_flush_cached_reqs(struct io_ring_ctx *ctx)
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001972{
Jens Axboec7dae4b2021-02-09 19:53:37 -07001973 struct io_submit_state *state = &ctx->submit_state;
1974 struct io_comp_state *cs = &state->comp;
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001975 struct io_kiocb *req = NULL;
1976
Jens Axboec7dae4b2021-02-09 19:53:37 -07001977 /*
1978 * If we have more than a batch's worth of requests in our IRQ side
1979 * locked cache, grab the lock and move them over to our submission
1980 * side cache.
1981 */
1982 if (READ_ONCE(cs->locked_free_nr) > IO_COMPL_BATCH) {
1983 spin_lock_irq(&ctx->completion_lock);
1984 list_splice_init(&cs->locked_free_list, &cs->free_list);
1985 cs->locked_free_nr = 0;
1986 spin_unlock_irq(&ctx->completion_lock);
1987 }
1988
1989 while (!list_empty(&cs->free_list)) {
1990 req = list_first_entry(&cs->free_list, struct io_kiocb,
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00001991 compl.list);
1992 list_del(&req->compl.list);
1993 state->reqs[state->free_reqs++] = req;
1994 if (state->free_reqs == ARRAY_SIZE(state->reqs))
1995 break;
1996 }
1997
1998 return req != NULL;
1999}
2000
Pavel Begunkov258b29a2021-02-10 00:03:10 +00002001static struct io_kiocb *io_alloc_req(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002002{
Pavel Begunkov258b29a2021-02-10 00:03:10 +00002003 struct io_submit_state *state = &ctx->submit_state;
2004
Pavel Begunkovbf019da2021-02-10 00:03:17 +00002005 BUILD_BUG_ON(IO_REQ_ALLOC_BATCH > ARRAY_SIZE(state->reqs));
2006
Pavel Begunkovf6b6c7d2020-06-21 13:09:53 +03002007 if (!state->free_reqs) {
Pavel Begunkov291b2822020-09-30 22:57:01 +03002008 gfp_t gfp = GFP_KERNEL | __GFP_NOWARN;
Jens Axboe2579f912019-01-09 09:10:43 -07002009 int ret;
2010
Jens Axboec7dae4b2021-02-09 19:53:37 -07002011 if (io_flush_cached_reqs(ctx))
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00002012 goto got_req;
2013
Pavel Begunkovbf019da2021-02-10 00:03:17 +00002014 ret = kmem_cache_alloc_bulk(req_cachep, gfp, IO_REQ_ALLOC_BATCH,
2015 state->reqs);
Jens Axboefd6fab22019-03-14 16:30:06 -06002016
2017 /*
2018 * Bulk alloc is all-or-nothing. If we fail to get a batch,
2019 * retry single alloc to be on the safe side.
2020 */
2021 if (unlikely(ret <= 0)) {
2022 state->reqs[0] = kmem_cache_alloc(req_cachep, gfp);
2023 if (!state->reqs[0])
Pavel Begunkov3893f392021-02-10 00:03:15 +00002024 return NULL;
Jens Axboefd6fab22019-03-14 16:30:06 -06002025 ret = 1;
2026 }
Pavel Begunkov291b2822020-09-30 22:57:01 +03002027 state->free_reqs = ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002028 }
Pavel Begunkove5d1bc02021-02-10 00:03:23 +00002029got_req:
Pavel Begunkov291b2822020-09-30 22:57:01 +03002030 state->free_reqs--;
2031 return state->reqs[state->free_reqs];
Jens Axboe2b188cc2019-01-07 10:46:33 -07002032}
2033
Pavel Begunkov8da11c12020-02-24 11:32:44 +03002034static inline void io_put_file(struct io_kiocb *req, struct file *file,
2035 bool fixed)
2036{
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00002037 if (!fixed)
Pavel Begunkov8da11c12020-02-24 11:32:44 +03002038 fput(file);
2039}
2040
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01002041static void io_dismantle_req(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002042{
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03002043 io_clean_op(req);
Pavel Begunkov929a3af2020-02-19 00:19:09 +03002044
Jens Axboee8c2bc12020-08-15 18:44:09 -07002045 if (req->async_data)
2046 kfree(req->async_data);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03002047 if (req->file)
2048 io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE));
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00002049 if (req->fixed_rsrc_refs)
2050 percpu_ref_put(req->fixed_rsrc_refs);
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01002051 io_req_clean_work(req);
Pavel Begunkove6543a82020-06-28 12:52:30 +03002052}
Pavel Begunkov2b85edf2019-12-28 14:13:03 +03002053
Pavel Begunkov7c660732021-01-25 11:42:21 +00002054static inline void io_put_task(struct task_struct *task, int nr)
2055{
2056 struct io_uring_task *tctx = task->io_uring;
2057
2058 percpu_counter_sub(&tctx->inflight, nr);
2059 if (unlikely(atomic_read(&tctx->in_idle)))
2060 wake_up(&tctx->wait);
2061 put_task_struct_many(task, nr);
2062}
2063
Pavel Begunkov216578e2020-10-13 09:44:00 +01002064static void __io_free_req(struct io_kiocb *req)
Pavel Begunkove6543a82020-06-28 12:52:30 +03002065{
Jens Axboe51a4cc12020-08-10 10:55:56 -06002066 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovecfc5172020-06-29 13:13:03 +03002067
Pavel Begunkov216578e2020-10-13 09:44:00 +01002068 io_dismantle_req(req);
Pavel Begunkov7c660732021-01-25 11:42:21 +00002069 io_put_task(req->task, 1);
Jens Axboee3bc8e92020-09-24 08:45:57 -06002070
Pavel Begunkov3893f392021-02-10 00:03:15 +00002071 kmem_cache_free(req_cachep, req);
Pavel Begunkovecfc5172020-06-29 13:13:03 +03002072 percpu_ref_put(&ctx->refs);
Jens Axboee65ef562019-03-12 10:16:44 -06002073}
2074
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002075static inline void io_remove_next_linked(struct io_kiocb *req)
2076{
2077 struct io_kiocb *nxt = req->link;
2078
2079 req->link = nxt->link;
2080 nxt->link = NULL;
2081}
2082
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01002083static void io_kill_linked_timeout(struct io_kiocb *req)
Jens Axboe9e645e112019-05-10 16:07:28 -06002084{
Jackie Liua197f662019-11-08 08:09:12 -07002085 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03002086 struct io_kiocb *link;
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01002087 bool cancelled = false;
2088 unsigned long flags;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03002089
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01002090 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002091 link = req->link;
2092
Pavel Begunkov900fad42020-10-19 16:39:16 +01002093 /*
2094 * Can happen if a linked timeout fired and link had been like
2095 * req -> link t-out -> link t-out [-> ...]
2096 */
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01002097 if (link && (link->flags & REQ_F_LTIMEOUT_ACTIVE)) {
2098 struct io_timeout_data *io = link->async_data;
2099 int ret;
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03002100
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002101 io_remove_next_linked(req);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00002102 link->timeout.head = NULL;
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01002103 ret = hrtimer_try_to_cancel(&io->timer);
2104 if (ret != -1) {
2105 io_cqring_fill_event(link, -ECANCELED);
2106 io_commit_cqring(ctx);
2107 cancelled = true;
2108 }
2109 }
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03002110 req->flags &= ~REQ_F_LINK_TIMEOUT;
Pavel Begunkov216578e2020-10-13 09:44:00 +01002111 spin_unlock_irqrestore(&ctx->completion_lock, flags);
Jens Axboeab0b6452020-06-30 08:43:15 -06002112
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01002113 if (cancelled) {
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03002114 io_cqring_ev_posted(ctx);
Pavel Begunkovc9abd7a2020-10-22 16:43:11 +01002115 io_put_req(link);
2116 }
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03002117}
2118
Jens Axboe4d7dd462019-11-20 13:03:52 -07002119
Pavel Begunkovd148ca42020-10-18 10:17:39 +01002120static void io_fail_links(struct io_kiocb *req)
Jens Axboe9e645e112019-05-10 16:07:28 -06002121{
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002122 struct io_kiocb *link, *nxt;
Jens Axboe2665abf2019-11-05 12:40:47 -07002123 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovd148ca42020-10-18 10:17:39 +01002124 unsigned long flags;
Jens Axboe9e645e112019-05-10 16:07:28 -06002125
Pavel Begunkovd148ca42020-10-18 10:17:39 +01002126 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002127 link = req->link;
2128 req->link = NULL;
Jens Axboe9e645e112019-05-10 16:07:28 -06002129
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002130 while (link) {
2131 nxt = link->link;
2132 link->link = NULL;
2133
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02002134 trace_io_uring_fail_link(req, link);
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03002135 io_cqring_fill_event(link, -ECANCELED);
Pavel Begunkov216578e2020-10-13 09:44:00 +01002136
2137 /*
2138 * It's ok to free under spinlock as they're not linked anymore,
2139 * but avoid REQ_F_WORK_INITIALIZED because it may deadlock on
2140 * work.fs->lock.
2141 */
2142 if (link->flags & REQ_F_WORK_INITIALIZED)
2143 io_put_req_deferred(link, 2);
2144 else
2145 io_double_put_req(link);
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002146 link = nxt;
Jens Axboe9e645e112019-05-10 16:07:28 -06002147 }
Jens Axboe2665abf2019-11-05 12:40:47 -07002148 io_commit_cqring(ctx);
Pavel Begunkov216578e2020-10-13 09:44:00 +01002149 spin_unlock_irqrestore(&ctx->completion_lock, flags);
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03002150
Jens Axboe2665abf2019-11-05 12:40:47 -07002151 io_cqring_ev_posted(ctx);
Jens Axboe9e645e112019-05-10 16:07:28 -06002152}
2153
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03002154static struct io_kiocb *__io_req_find_next(struct io_kiocb *req)
Jens Axboe9e645e112019-05-10 16:07:28 -06002155{
Pavel Begunkov7c86ffe2020-06-29 13:12:59 +03002156 if (req->flags & REQ_F_LINK_TIMEOUT)
2157 io_kill_linked_timeout(req);
Jens Axboe2665abf2019-11-05 12:40:47 -07002158
Jens Axboe9e645e112019-05-10 16:07:28 -06002159 /*
2160 * If LINK is set, we have dependent requests in this chain. If we
2161 * didn't fail this request, queue the first one up, moving any other
2162 * dependencies to the next request. In case of failure, fail the rest
2163 * of the chain.
2164 */
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002165 if (likely(!(req->flags & REQ_F_FAIL_LINK))) {
2166 struct io_kiocb *nxt = req->link;
2167
2168 req->link = NULL;
2169 return nxt;
2170 }
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002171 io_fail_links(req);
2172 return NULL;
Jens Axboe4d7dd462019-11-20 13:03:52 -07002173}
Jens Axboe2665abf2019-11-05 12:40:47 -07002174
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002175static inline struct io_kiocb *io_req_find_next(struct io_kiocb *req)
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03002176{
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002177 if (likely(!(req->link) && !(req->flags & REQ_F_LINK_TIMEOUT)))
Pavel Begunkov3fa5e0f2020-06-30 15:20:43 +03002178 return NULL;
2179 return __io_req_find_next(req);
2180}
2181
Jens Axboe7cbf1722021-02-10 00:03:20 +00002182static bool __tctx_task_work(struct io_uring_task *tctx)
2183{
Jens Axboe65453d12021-02-10 00:03:21 +00002184 struct io_ring_ctx *ctx = NULL;
Jens Axboe7cbf1722021-02-10 00:03:20 +00002185 struct io_wq_work_list list;
2186 struct io_wq_work_node *node;
2187
2188 if (wq_list_empty(&tctx->task_list))
2189 return false;
2190
2191 spin_lock(&tctx->task_lock);
2192 list = tctx->task_list;
2193 INIT_WQ_LIST(&tctx->task_list);
2194 spin_unlock(&tctx->task_lock);
2195
2196 node = list.first;
2197 while (node) {
2198 struct io_wq_work_node *next = node->next;
Jens Axboe65453d12021-02-10 00:03:21 +00002199 struct io_ring_ctx *this_ctx;
Jens Axboe7cbf1722021-02-10 00:03:20 +00002200 struct io_kiocb *req;
2201
2202 req = container_of(node, struct io_kiocb, io_task_work.node);
Jens Axboe65453d12021-02-10 00:03:21 +00002203 this_ctx = req->ctx;
Jens Axboe7cbf1722021-02-10 00:03:20 +00002204 req->task_work.func(&req->task_work);
2205 node = next;
Jens Axboe65453d12021-02-10 00:03:21 +00002206
2207 if (!ctx) {
2208 ctx = this_ctx;
2209 } else if (ctx != this_ctx) {
2210 mutex_lock(&ctx->uring_lock);
2211 io_submit_flush_completions(&ctx->submit_state.comp, ctx);
2212 mutex_unlock(&ctx->uring_lock);
2213 ctx = this_ctx;
2214 }
2215 }
2216
2217 if (ctx && ctx->submit_state.comp.nr) {
2218 mutex_lock(&ctx->uring_lock);
2219 io_submit_flush_completions(&ctx->submit_state.comp, ctx);
2220 mutex_unlock(&ctx->uring_lock);
Jens Axboe7cbf1722021-02-10 00:03:20 +00002221 }
2222
2223 return list.first != NULL;
2224}
2225
2226static void tctx_task_work(struct callback_head *cb)
2227{
2228 struct io_uring_task *tctx = container_of(cb, struct io_uring_task, task_work);
2229
2230 while (__tctx_task_work(tctx))
2231 cond_resched();
2232
2233 clear_bit(0, &tctx->task_state);
2234}
2235
2236static int io_task_work_add(struct task_struct *tsk, struct io_kiocb *req,
2237 enum task_work_notify_mode notify)
2238{
2239 struct io_uring_task *tctx = tsk->io_uring;
2240 struct io_wq_work_node *node, *prev;
2241 int ret;
2242
2243 WARN_ON_ONCE(!tctx);
2244
2245 spin_lock(&tctx->task_lock);
2246 wq_list_add_tail(&req->io_task_work.node, &tctx->task_list);
2247 spin_unlock(&tctx->task_lock);
2248
2249 /* task_work already pending, we're done */
2250 if (test_bit(0, &tctx->task_state) ||
2251 test_and_set_bit(0, &tctx->task_state))
2252 return 0;
2253
2254 if (!task_work_add(tsk, &tctx->task_work, notify))
2255 return 0;
2256
2257 /*
2258 * Slow path - we failed, find and delete work. if the work is not
2259 * in the list, it got run and we're fine.
2260 */
2261 ret = 0;
2262 spin_lock(&tctx->task_lock);
2263 wq_list_for_each(node, prev, &tctx->task_list) {
2264 if (&req->io_task_work.node == node) {
2265 wq_list_del(&tctx->task_list, node, prev);
2266 ret = 1;
2267 break;
2268 }
2269 }
2270 spin_unlock(&tctx->task_lock);
2271 clear_bit(0, &tctx->task_state);
2272 return ret;
2273}
2274
Jens Axboe355fb9e2020-10-22 20:19:35 -06002275static int io_req_task_work_add(struct io_kiocb *req)
Jens Axboec2c4c832020-07-01 15:37:11 -06002276{
2277 struct task_struct *tsk = req->task;
2278 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe91989c72020-10-16 09:02:26 -06002279 enum task_work_notify_mode notify;
2280 int ret;
Jens Axboec2c4c832020-07-01 15:37:11 -06002281
Jens Axboe6200b0a2020-09-13 14:38:30 -06002282 if (tsk->flags & PF_EXITING)
2283 return -ESRCH;
2284
Jens Axboec2c4c832020-07-01 15:37:11 -06002285 /*
Jens Axboe0ba9c9e2020-08-06 19:41:50 -06002286 * SQPOLL kernel thread doesn't need notification, just a wakeup. For
2287 * all other cases, use TWA_SIGNAL unconditionally to ensure we're
2288 * processing task_work. There's no reliable way to tell if TWA_RESUME
2289 * will do the job.
Jens Axboec2c4c832020-07-01 15:37:11 -06002290 */
Jens Axboe91989c72020-10-16 09:02:26 -06002291 notify = TWA_NONE;
Jens Axboe355fb9e2020-10-22 20:19:35 -06002292 if (!(ctx->flags & IORING_SETUP_SQPOLL))
Jens Axboec2c4c832020-07-01 15:37:11 -06002293 notify = TWA_SIGNAL;
2294
Jens Axboe7cbf1722021-02-10 00:03:20 +00002295 ret = io_task_work_add(tsk, req, notify);
Jens Axboec2c4c832020-07-01 15:37:11 -06002296 if (!ret)
2297 wake_up_process(tsk);
Jens Axboe0ba9c9e2020-08-06 19:41:50 -06002298
Jens Axboec2c4c832020-07-01 15:37:11 -06002299 return ret;
2300}
2301
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002302static void io_req_task_work_add_fallback(struct io_kiocb *req,
Jens Axboe7cbf1722021-02-10 00:03:20 +00002303 task_work_func_t cb)
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002304{
2305 struct task_struct *tsk = io_wq_get_task(req->ctx->io_wq);
2306
2307 init_task_work(&req->task_work, cb);
2308 task_work_add(tsk, &req->task_work, TWA_NONE);
2309 wake_up_process(tsk);
2310}
2311
Jens Axboec40f6372020-06-25 15:39:59 -06002312static void __io_req_task_cancel(struct io_kiocb *req, int error)
2313{
2314 struct io_ring_ctx *ctx = req->ctx;
2315
2316 spin_lock_irq(&ctx->completion_lock);
2317 io_cqring_fill_event(req, error);
2318 io_commit_cqring(ctx);
2319 spin_unlock_irq(&ctx->completion_lock);
2320
2321 io_cqring_ev_posted(ctx);
2322 req_set_fail_links(req);
2323 io_double_put_req(req);
2324}
2325
2326static void io_req_task_cancel(struct callback_head *cb)
2327{
2328 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
Jens Axboe87ceb6a2020-09-14 08:20:12 -06002329 struct io_ring_ctx *ctx = req->ctx;
Jens Axboec40f6372020-06-25 15:39:59 -06002330
2331 __io_req_task_cancel(req, -ECANCELED);
Jens Axboe87ceb6a2020-09-14 08:20:12 -06002332 percpu_ref_put(&ctx->refs);
Jens Axboec40f6372020-06-25 15:39:59 -06002333}
2334
2335static void __io_req_task_submit(struct io_kiocb *req)
2336{
2337 struct io_ring_ctx *ctx = req->ctx;
2338
Pavel Begunkov81b6d052021-01-04 20:36:35 +00002339 mutex_lock(&ctx->uring_lock);
Pavel Begunkov4e326352021-02-12 03:23:52 +00002340 if (!ctx->sqo_dead && !io_sq_thread_acquire_mm_files(ctx, req))
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00002341 __io_queue_sqe(req);
Pavel Begunkov81b6d052021-01-04 20:36:35 +00002342 else
Jens Axboec40f6372020-06-25 15:39:59 -06002343 __io_req_task_cancel(req, -EFAULT);
Pavel Begunkov81b6d052021-01-04 20:36:35 +00002344 mutex_unlock(&ctx->uring_lock);
Jens Axboe9e645e112019-05-10 16:07:28 -06002345}
2346
Jens Axboec40f6372020-06-25 15:39:59 -06002347static void io_req_task_submit(struct callback_head *cb)
2348{
2349 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
Jens Axboe6d816e02020-08-11 08:04:14 -06002350 struct io_ring_ctx *ctx = req->ctx;
Jens Axboec40f6372020-06-25 15:39:59 -06002351
2352 __io_req_task_submit(req);
Jens Axboe6d816e02020-08-11 08:04:14 -06002353 percpu_ref_put(&ctx->refs);
Jens Axboec40f6372020-06-25 15:39:59 -06002354}
2355
2356static void io_req_task_queue(struct io_kiocb *req)
2357{
Jens Axboec40f6372020-06-25 15:39:59 -06002358 int ret;
2359
Jens Axboe7cbf1722021-02-10 00:03:20 +00002360 req->task_work.func = io_req_task_submit;
Jens Axboe6d816e02020-08-11 08:04:14 -06002361 percpu_ref_get(&req->ctx->refs);
Jens Axboec40f6372020-06-25 15:39:59 -06002362
Jens Axboe355fb9e2020-10-22 20:19:35 -06002363 ret = io_req_task_work_add(req);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002364 if (unlikely(ret))
2365 io_req_task_work_add_fallback(req, io_req_task_cancel);
Jens Axboec40f6372020-06-25 15:39:59 -06002366}
2367
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002368static inline void io_queue_next(struct io_kiocb *req)
Jackie Liuc69f8db2019-11-09 11:00:08 +08002369{
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002370 struct io_kiocb *nxt = io_req_find_next(req);
Pavel Begunkov944e58b2019-11-21 23:21:01 +03002371
Pavel Begunkov906a8c32020-06-27 14:04:55 +03002372 if (nxt)
2373 io_req_task_queue(nxt);
Jackie Liuc69f8db2019-11-09 11:00:08 +08002374}
2375
Jens Axboe9e645e112019-05-10 16:07:28 -06002376static void io_free_req(struct io_kiocb *req)
2377{
Pavel Begunkovc3524382020-06-28 12:52:32 +03002378 io_queue_next(req);
Jens Axboe9e645e112019-05-10 16:07:28 -06002379 __io_free_req(req);
Jens Axboee65ef562019-03-12 10:16:44 -06002380}
2381
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002382struct req_batch {
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002383 struct task_struct *task;
2384 int task_refs;
Jens Axboe1b4c3512021-02-10 00:03:19 +00002385 int ctx_refs;
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002386};
2387
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002388static inline void io_init_req_batch(struct req_batch *rb)
Pavel Begunkov7a743e22020-03-03 21:33:13 +03002389{
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002390 rb->task_refs = 0;
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002391 rb->ctx_refs = 0;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002392 rb->task = NULL;
2393}
Pavel Begunkov8766dd52020-03-14 00:31:04 +03002394
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002395static void io_req_free_batch_finish(struct io_ring_ctx *ctx,
2396 struct req_batch *rb)
2397{
Pavel Begunkov6e833d52021-02-11 18:28:20 +00002398 if (rb->task)
Pavel Begunkov7c660732021-01-25 11:42:21 +00002399 io_put_task(rb->task, rb->task_refs);
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002400 if (rb->ctx_refs)
2401 percpu_ref_put_many(&ctx->refs, rb->ctx_refs);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002402}
2403
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002404static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req,
2405 struct io_submit_state *state)
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002406{
Pavel Begunkovf2f87372020-10-27 23:25:37 +00002407 io_queue_next(req);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002408
Jens Axboee3bc8e92020-09-24 08:45:57 -06002409 if (req->task != rb->task) {
Pavel Begunkov7c660732021-01-25 11:42:21 +00002410 if (rb->task)
2411 io_put_task(rb->task, rb->task_refs);
Jens Axboee3bc8e92020-09-24 08:45:57 -06002412 rb->task = req->task;
2413 rb->task_refs = 0;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002414 }
Jens Axboee3bc8e92020-09-24 08:45:57 -06002415 rb->task_refs++;
Pavel Begunkov9ae72462021-02-10 00:03:16 +00002416 rb->ctx_refs++;
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002417
Pavel Begunkov4edf20f2020-10-13 09:43:59 +01002418 io_dismantle_req(req);
Pavel Begunkovbd759042021-02-12 03:23:50 +00002419 if (state->free_reqs != ARRAY_SIZE(state->reqs))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002420 state->reqs[state->free_reqs++] = req;
Pavel Begunkovbd759042021-02-12 03:23:50 +00002421 else
2422 list_add(&req->compl.list, &state->comp.free_list);
Pavel Begunkov7a743e22020-03-03 21:33:13 +03002423}
2424
Pavel Begunkov905c1722021-02-10 00:03:14 +00002425static void io_submit_flush_completions(struct io_comp_state *cs,
2426 struct io_ring_ctx *ctx)
2427{
2428 int i, nr = cs->nr;
2429 struct io_kiocb *req;
2430 struct req_batch rb;
2431
2432 io_init_req_batch(&rb);
2433 spin_lock_irq(&ctx->completion_lock);
2434 for (i = 0; i < nr; i++) {
2435 req = cs->reqs[i];
2436 __io_cqring_fill_event(req, req->result, req->compl.cflags);
2437 }
2438 io_commit_cqring(ctx);
2439 spin_unlock_irq(&ctx->completion_lock);
2440
2441 io_cqring_ev_posted(ctx);
2442 for (i = 0; i < nr; i++) {
2443 req = cs->reqs[i];
2444
2445 /* submission and completion refs */
2446 if (refcount_sub_and_test(2, &req->refs))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002447 io_req_free_batch(&rb, req, &ctx->submit_state);
Pavel Begunkov905c1722021-02-10 00:03:14 +00002448 }
2449
2450 io_req_free_batch_finish(ctx, &rb);
2451 cs->nr = 0;
2452}
2453
Jens Axboeba816ad2019-09-28 11:36:45 -06002454/*
2455 * Drop reference to request, return next in chain (if there is one) if this
2456 * was the last reference to this request.
2457 */
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002458static struct io_kiocb *io_put_req_find_next(struct io_kiocb *req)
Jens Axboee65ef562019-03-12 10:16:44 -06002459{
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002460 struct io_kiocb *nxt = NULL;
2461
Jens Axboe2a44f462020-02-25 13:25:41 -07002462 if (refcount_dec_and_test(&req->refs)) {
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002463 nxt = io_req_find_next(req);
Jens Axboe4d7dd462019-11-20 13:03:52 -07002464 __io_free_req(req);
Jens Axboe2a44f462020-02-25 13:25:41 -07002465 }
Pavel Begunkov9b5f7bd2020-06-29 13:13:00 +03002466 return nxt;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002467}
2468
Jens Axboe2b188cc2019-01-07 10:46:33 -07002469static void io_put_req(struct io_kiocb *req)
2470{
Jens Axboedef596e2019-01-09 08:59:42 -07002471 if (refcount_dec_and_test(&req->refs))
2472 io_free_req(req);
2473}
2474
Pavel Begunkov216578e2020-10-13 09:44:00 +01002475static void io_put_req_deferred_cb(struct callback_head *cb)
2476{
2477 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
2478
2479 io_free_req(req);
2480}
2481
2482static void io_free_req_deferred(struct io_kiocb *req)
2483{
2484 int ret;
2485
Jens Axboe7cbf1722021-02-10 00:03:20 +00002486 req->task_work.func = io_put_req_deferred_cb;
Jens Axboe355fb9e2020-10-22 20:19:35 -06002487 ret = io_req_task_work_add(req);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00002488 if (unlikely(ret))
2489 io_req_task_work_add_fallback(req, io_put_req_deferred_cb);
Pavel Begunkov216578e2020-10-13 09:44:00 +01002490}
2491
2492static inline void io_put_req_deferred(struct io_kiocb *req, int refs)
2493{
2494 if (refcount_sub_and_test(refs, &req->refs))
2495 io_free_req_deferred(req);
2496}
2497
Jens Axboe978db572019-11-14 22:39:04 -07002498static void io_double_put_req(struct io_kiocb *req)
2499{
2500 /* drop both submit and complete references */
2501 if (refcount_sub_and_test(2, &req->refs))
2502 io_free_req(req);
2503}
2504
Pavel Begunkov6c503152021-01-04 20:36:36 +00002505static unsigned io_cqring_events(struct io_ring_ctx *ctx)
Jens Axboea3a0e432019-08-20 11:03:11 -06002506{
2507 /* See comment at the top of this file */
2508 smp_rmb();
Pavel Begunkove23de152020-12-17 00:24:37 +00002509 return __io_cqring_events(ctx);
Jens Axboea3a0e432019-08-20 11:03:11 -06002510}
2511
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03002512static inline unsigned int io_sqring_entries(struct io_ring_ctx *ctx)
2513{
2514 struct io_rings *rings = ctx->rings;
2515
2516 /* make sure SQ entry isn't read before tail */
2517 return smp_load_acquire(&rings->sq.tail) - ctx->cached_sq_head;
2518}
2519
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002520static unsigned int io_put_kbuf(struct io_kiocb *req, struct io_buffer *kbuf)
Jens Axboee94f1412019-12-19 12:06:02 -07002521{
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002522 unsigned int cflags;
Jens Axboee94f1412019-12-19 12:06:02 -07002523
Jens Axboebcda7ba2020-02-23 16:42:51 -07002524 cflags = kbuf->bid << IORING_CQE_BUFFER_SHIFT;
2525 cflags |= IORING_CQE_F_BUFFER;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03002526 req->flags &= ~REQ_F_BUFFER_SELECTED;
Jens Axboebcda7ba2020-02-23 16:42:51 -07002527 kfree(kbuf);
2528 return cflags;
2529}
2530
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002531static inline unsigned int io_put_rw_kbuf(struct io_kiocb *req)
2532{
2533 struct io_buffer *kbuf;
2534
2535 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
2536 return io_put_kbuf(req, kbuf);
2537}
2538
Jens Axboe4c6e2772020-07-01 11:29:10 -06002539static inline bool io_run_task_work(void)
2540{
Jens Axboe6200b0a2020-09-13 14:38:30 -06002541 /*
2542 * Not safe to run on exiting task, and the task_work handling will
2543 * not add work to such a task.
2544 */
2545 if (unlikely(current->flags & PF_EXITING))
2546 return false;
Jens Axboe4c6e2772020-07-01 11:29:10 -06002547 if (current->task_works) {
2548 __set_current_state(TASK_RUNNING);
2549 task_work_run();
2550 return true;
2551 }
2552
2553 return false;
2554}
2555
Jens Axboedef596e2019-01-09 08:59:42 -07002556/*
2557 * Find and free completed poll iocbs
2558 */
2559static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events,
2560 struct list_head *done)
2561{
Jens Axboe8237e042019-12-28 10:48:22 -07002562 struct req_batch rb;
Jens Axboedef596e2019-01-09 08:59:42 -07002563 struct io_kiocb *req;
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002564
2565 /* order with ->result store in io_complete_rw_iopoll() */
2566 smp_rmb();
Jens Axboedef596e2019-01-09 08:59:42 -07002567
Pavel Begunkov5af1d132020-07-18 11:32:52 +03002568 io_init_req_batch(&rb);
Jens Axboedef596e2019-01-09 08:59:42 -07002569 while (!list_empty(done)) {
Jens Axboebcda7ba2020-02-23 16:42:51 -07002570 int cflags = 0;
2571
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002572 req = list_first_entry(done, struct io_kiocb, inflight_entry);
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002573 list_del(&req->inflight_entry);
Jens Axboedef596e2019-01-09 08:59:42 -07002574
Pavel Begunkovf1613402021-02-11 18:28:21 +00002575 if (READ_ONCE(req->result) == -EAGAIN) {
2576 req->iopoll_completed = 0;
Pavel Begunkov23faba32021-02-11 18:28:22 +00002577 if (io_rw_reissue(req))
Pavel Begunkovf1613402021-02-11 18:28:21 +00002578 continue;
2579 }
2580
Jens Axboebcda7ba2020-02-23 16:42:51 -07002581 if (req->flags & REQ_F_BUFFER_SELECTED)
Pavel Begunkov8ff069b2020-07-16 23:28:04 +03002582 cflags = io_put_rw_kbuf(req);
Jens Axboebcda7ba2020-02-23 16:42:51 -07002583
2584 __io_cqring_fill_event(req, req->result, cflags);
Jens Axboedef596e2019-01-09 08:59:42 -07002585 (*nr_events)++;
2586
Pavel Begunkovc3524382020-06-28 12:52:32 +03002587 if (refcount_dec_and_test(&req->refs))
Pavel Begunkov6ff119a2021-02-10 00:03:18 +00002588 io_req_free_batch(&rb, req, &ctx->submit_state);
Jens Axboedef596e2019-01-09 08:59:42 -07002589 }
Jens Axboedef596e2019-01-09 08:59:42 -07002590
Jens Axboe09bb8392019-03-13 12:39:28 -06002591 io_commit_cqring(ctx);
Pavel Begunkov80c18e42021-01-07 03:15:41 +00002592 io_cqring_ev_posted_iopoll(ctx);
Pavel Begunkov2d6500d2020-06-28 12:52:33 +03002593 io_req_free_batch_finish(ctx, &rb);
Bijan Mottahedeh581f9812020-04-03 13:51:33 -07002594}
2595
Jens Axboedef596e2019-01-09 08:59:42 -07002596static int io_do_iopoll(struct io_ring_ctx *ctx, unsigned int *nr_events,
2597 long min)
2598{
2599 struct io_kiocb *req, *tmp;
2600 LIST_HEAD(done);
2601 bool spin;
2602 int ret;
2603
2604 /*
2605 * Only spin for completions if we don't have multiple devices hanging
2606 * off our complete list, and we're under the requested amount.
2607 */
2608 spin = !ctx->poll_multi_file && *nr_events < min;
2609
2610 ret = 0;
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002611 list_for_each_entry_safe(req, tmp, &ctx->iopoll_list, inflight_entry) {
Jens Axboe9adbd452019-12-20 08:45:55 -07002612 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboedef596e2019-01-09 08:59:42 -07002613
2614 /*
Bijan Mottahedeh581f9812020-04-03 13:51:33 -07002615 * Move completed and retryable entries to our local lists.
2616 * If we find a request that requires polling, break out
2617 * and complete those lists first, if we have entries there.
Jens Axboedef596e2019-01-09 08:59:42 -07002618 */
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002619 if (READ_ONCE(req->iopoll_completed)) {
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002620 list_move_tail(&req->inflight_entry, &done);
Jens Axboedef596e2019-01-09 08:59:42 -07002621 continue;
2622 }
2623 if (!list_empty(&done))
2624 break;
2625
2626 ret = kiocb->ki_filp->f_op->iopoll(kiocb, spin);
2627 if (ret < 0)
2628 break;
2629
Pavel Begunkov3aadc232020-07-06 17:59:29 +03002630 /* iopoll may have completed current req */
2631 if (READ_ONCE(req->iopoll_completed))
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002632 list_move_tail(&req->inflight_entry, &done);
Pavel Begunkov3aadc232020-07-06 17:59:29 +03002633
Jens Axboedef596e2019-01-09 08:59:42 -07002634 if (ret && spin)
2635 spin = false;
2636 ret = 0;
2637 }
2638
2639 if (!list_empty(&done))
2640 io_iopoll_complete(ctx, nr_events, &done);
2641
2642 return ret;
2643}
2644
2645/*
Brian Gianforcarod195a662019-12-13 03:09:50 -08002646 * Poll for a minimum of 'min' events. Note that if min == 0 we consider that a
Jens Axboedef596e2019-01-09 08:59:42 -07002647 * non-spinning poll check - we'll still enter the driver poll loop, but only
2648 * as a non-spinning completion check.
2649 */
2650static int io_iopoll_getevents(struct io_ring_ctx *ctx, unsigned int *nr_events,
2651 long min)
2652{
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002653 while (!list_empty(&ctx->iopoll_list) && !need_resched()) {
Jens Axboedef596e2019-01-09 08:59:42 -07002654 int ret;
2655
2656 ret = io_do_iopoll(ctx, nr_events, min);
2657 if (ret < 0)
2658 return ret;
Pavel Begunkoveba0a4d2020-07-06 17:59:30 +03002659 if (*nr_events >= min)
Jens Axboedef596e2019-01-09 08:59:42 -07002660 return 0;
2661 }
2662
2663 return 1;
2664}
2665
2666/*
2667 * We can't just wait for polled events to come to us, we have to actively
2668 * find and complete them.
2669 */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002670static void io_iopoll_try_reap_events(struct io_ring_ctx *ctx)
Jens Axboedef596e2019-01-09 08:59:42 -07002671{
2672 if (!(ctx->flags & IORING_SETUP_IOPOLL))
2673 return;
2674
2675 mutex_lock(&ctx->uring_lock);
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002676 while (!list_empty(&ctx->iopoll_list)) {
Jens Axboedef596e2019-01-09 08:59:42 -07002677 unsigned int nr_events = 0;
2678
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002679 io_do_iopoll(ctx, &nr_events, 0);
Jens Axboe08f54392019-08-21 22:19:11 -06002680
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03002681 /* let it sleep and repeat later if can't complete a request */
2682 if (nr_events == 0)
2683 break;
Jens Axboe08f54392019-08-21 22:19:11 -06002684 /*
2685 * Ensure we allow local-to-the-cpu processing to take place,
2686 * in this case we need to ensure that we reap all events.
Pavel Begunkov3fcee5a2020-07-06 17:59:31 +03002687 * Also let task_work, etc. to progress by releasing the mutex
Jens Axboe08f54392019-08-21 22:19:11 -06002688 */
Pavel Begunkov3fcee5a2020-07-06 17:59:31 +03002689 if (need_resched()) {
2690 mutex_unlock(&ctx->uring_lock);
2691 cond_resched();
2692 mutex_lock(&ctx->uring_lock);
2693 }
Jens Axboedef596e2019-01-09 08:59:42 -07002694 }
2695 mutex_unlock(&ctx->uring_lock);
2696}
2697
Pavel Begunkov7668b922020-07-07 16:36:21 +03002698static int io_iopoll_check(struct io_ring_ctx *ctx, long min)
Jens Axboedef596e2019-01-09 08:59:42 -07002699{
Pavel Begunkov7668b922020-07-07 16:36:21 +03002700 unsigned int nr_events = 0;
Jens Axboe2b2ed972019-10-25 10:06:15 -06002701 int iters = 0, ret = 0;
Jens Axboedef596e2019-01-09 08:59:42 -07002702
Xiaoguang Wangc7849be2020-02-22 14:46:05 +08002703 /*
2704 * We disallow the app entering submit/complete with polling, but we
2705 * still need to lock the ring to prevent racing with polled issue
2706 * that got punted to a workqueue.
2707 */
2708 mutex_lock(&ctx->uring_lock);
Jens Axboedef596e2019-01-09 08:59:42 -07002709 do {
Jens Axboe500f9fb2019-08-19 12:15:59 -06002710 /*
Jens Axboea3a0e432019-08-20 11:03:11 -06002711 * Don't enter poll loop if we already have events pending.
2712 * If we do, we can potentially be spinning for commands that
2713 * already triggered a CQE (eg in error).
2714 */
Pavel Begunkov6c503152021-01-04 20:36:36 +00002715 if (test_bit(0, &ctx->cq_check_overflow))
2716 __io_cqring_overflow_flush(ctx, false, NULL, NULL);
2717 if (io_cqring_events(ctx))
Jens Axboea3a0e432019-08-20 11:03:11 -06002718 break;
2719
2720 /*
Jens Axboe500f9fb2019-08-19 12:15:59 -06002721 * If a submit got punted to a workqueue, we can have the
2722 * application entering polling for a command before it gets
2723 * issued. That app will hold the uring_lock for the duration
2724 * of the poll right here, so we need to take a breather every
2725 * now and then to ensure that the issue has a chance to add
2726 * the poll to the issued list. Otherwise we can spin here
2727 * forever, while the workqueue is stuck trying to acquire the
2728 * very same mutex.
2729 */
2730 if (!(++iters & 7)) {
2731 mutex_unlock(&ctx->uring_lock);
Jens Axboe4c6e2772020-07-01 11:29:10 -06002732 io_run_task_work();
Jens Axboe500f9fb2019-08-19 12:15:59 -06002733 mutex_lock(&ctx->uring_lock);
2734 }
2735
Pavel Begunkov7668b922020-07-07 16:36:21 +03002736 ret = io_iopoll_getevents(ctx, &nr_events, min);
Jens Axboedef596e2019-01-09 08:59:42 -07002737 if (ret <= 0)
2738 break;
2739 ret = 0;
Pavel Begunkov7668b922020-07-07 16:36:21 +03002740 } while (min && !nr_events && !need_resched());
Jens Axboedef596e2019-01-09 08:59:42 -07002741
Jens Axboe500f9fb2019-08-19 12:15:59 -06002742 mutex_unlock(&ctx->uring_lock);
Jens Axboedef596e2019-01-09 08:59:42 -07002743 return ret;
2744}
2745
Jens Axboe491381ce2019-10-17 09:20:46 -06002746static void kiocb_end_write(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002747{
Jens Axboe491381ce2019-10-17 09:20:46 -06002748 /*
2749 * Tell lockdep we inherited freeze protection from submission
2750 * thread.
2751 */
2752 if (req->flags & REQ_F_ISREG) {
2753 struct inode *inode = file_inode(req->file);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002754
Jens Axboe491381ce2019-10-17 09:20:46 -06002755 __sb_writers_acquired(inode->i_sb, SB_FREEZE_WRITE);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002756 }
Jens Axboe491381ce2019-10-17 09:20:46 -06002757 file_end_write(req->file);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002758}
2759
Jens Axboeb63534c2020-06-04 11:28:00 -06002760#ifdef CONFIG_BLOCK
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002761static bool io_resubmit_prep(struct io_kiocb *req)
Jens Axboeb63534c2020-06-04 11:28:00 -06002762{
2763 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
Colin Ian King4a245472021-02-10 20:00:07 +00002764 int rw, ret;
Jens Axboeb63534c2020-06-04 11:28:00 -06002765 struct iov_iter iter;
Jens Axboeb63534c2020-06-04 11:28:00 -06002766
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002767 /* already prepared */
2768 if (req->async_data)
2769 return true;
Jens Axboeb63534c2020-06-04 11:28:00 -06002770
2771 switch (req->opcode) {
2772 case IORING_OP_READV:
2773 case IORING_OP_READ_FIXED:
2774 case IORING_OP_READ:
2775 rw = READ;
2776 break;
2777 case IORING_OP_WRITEV:
2778 case IORING_OP_WRITE_FIXED:
2779 case IORING_OP_WRITE:
2780 rw = WRITE;
2781 break;
2782 default:
2783 printk_once(KERN_WARNING "io_uring: bad opcode in resubmit %d\n",
2784 req->opcode);
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002785 return false;
Jens Axboeb63534c2020-06-04 11:28:00 -06002786 }
2787
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002788 ret = io_import_iovec(rw, req, &iovec, &iter, false);
2789 if (ret < 0)
2790 return false;
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00002791 return !io_setup_async_rw(req, iovec, inline_vecs, &iter, false);
Jens Axboeb63534c2020-06-04 11:28:00 -06002792}
Jens Axboeb63534c2020-06-04 11:28:00 -06002793#endif
2794
Pavel Begunkov23faba32021-02-11 18:28:22 +00002795static bool io_rw_reissue(struct io_kiocb *req)
Jens Axboeb63534c2020-06-04 11:28:00 -06002796{
2797#ifdef CONFIG_BLOCK
Pavel Begunkov23faba32021-02-11 18:28:22 +00002798 umode_t mode = file_inode(req->file)->i_mode;
Jens Axboeb63534c2020-06-04 11:28:00 -06002799 int ret;
2800
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002801 if (!S_ISBLK(mode) && !S_ISREG(mode))
2802 return false;
2803 if ((req->flags & REQ_F_NOWAIT) || io_wq_current_is_worker())
Jens Axboeb63534c2020-06-04 11:28:00 -06002804 return false;
2805
Pavel Begunkov55e6ac12021-01-08 20:57:22 +00002806 lockdep_assert_held(&req->ctx->uring_lock);
2807
Jens Axboe28cea78a2020-09-14 10:51:17 -06002808 ret = io_sq_thread_acquire_mm_files(req->ctx, req);
Jens Axboe6d816e02020-08-11 08:04:14 -06002809
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002810 if (!ret && io_resubmit_prep(req)) {
Jens Axboefdee9462020-08-27 16:46:24 -06002811 refcount_inc(&req->refs);
2812 io_queue_async_work(req);
Jens Axboeb63534c2020-06-04 11:28:00 -06002813 return true;
Jens Axboefdee9462020-08-27 16:46:24 -06002814 }
Pavel Begunkovdc2a6e92021-01-19 13:32:35 +00002815 req_set_fail_links(req);
Jens Axboeb63534c2020-06-04 11:28:00 -06002816#endif
2817 return false;
2818}
2819
Jens Axboea1d7c392020-06-22 11:09:46 -06002820static void __io_complete_rw(struct io_kiocb *req, long res, long res2,
Pavel Begunkov889fca72021-02-10 00:03:09 +00002821 unsigned int issue_flags)
Jens Axboea1d7c392020-06-22 11:09:46 -06002822{
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002823 int cflags = 0;
2824
Pavel Begunkov23faba32021-02-11 18:28:22 +00002825 if ((res == -EAGAIN || res == -EOPNOTSUPP) && io_rw_reissue(req))
2826 return;
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002827 if (res != req->result)
2828 req_set_fail_links(req);
Pavel Begunkov23faba32021-02-11 18:28:22 +00002829
Pavel Begunkov2f8e45f2021-02-11 18:28:23 +00002830 if (req->rw.kiocb.ki_flags & IOCB_WRITE)
2831 kiocb_end_write(req);
2832 if (req->flags & REQ_F_BUFFER_SELECTED)
2833 cflags = io_put_rw_kbuf(req);
2834 __io_req_complete(req, issue_flags, res, cflags);
Jens Axboeba816ad2019-09-28 11:36:45 -06002835}
2836
2837static void io_complete_rw(struct kiocb *kiocb, long res, long res2)
2838{
Jens Axboe9adbd452019-12-20 08:45:55 -07002839 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboeba816ad2019-09-28 11:36:45 -06002840
Pavel Begunkov889fca72021-02-10 00:03:09 +00002841 __io_complete_rw(req, res, res2, 0);
Jens Axboe2b188cc2019-01-07 10:46:33 -07002842}
2843
Jens Axboedef596e2019-01-09 08:59:42 -07002844static void io_complete_rw_iopoll(struct kiocb *kiocb, long res, long res2)
2845{
Jens Axboe9adbd452019-12-20 08:45:55 -07002846 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboedef596e2019-01-09 08:59:42 -07002847
Jens Axboe491381ce2019-10-17 09:20:46 -06002848 if (kiocb->ki_flags & IOCB_WRITE)
2849 kiocb_end_write(req);
Jens Axboedef596e2019-01-09 08:59:42 -07002850
Xiaoguang Wang2d7d6792020-06-16 02:06:37 +08002851 if (res != -EAGAIN && res != req->result)
Jens Axboe4e88d6e2019-12-07 20:59:47 -07002852 req_set_fail_links(req);
Xiaoguang Wangbbde0172020-06-16 02:06:38 +08002853
2854 WRITE_ONCE(req->result, res);
2855 /* order with io_poll_complete() checking ->result */
Pavel Begunkovcd664b02020-06-25 12:37:10 +03002856 smp_wmb();
2857 WRITE_ONCE(req->iopoll_completed, 1);
Jens Axboedef596e2019-01-09 08:59:42 -07002858}
2859
2860/*
2861 * After the iocb has been issued, it's safe to be found on the poll list.
2862 * Adding the kiocb to the list AFTER submission ensures that we don't
2863 * find it from a io_iopoll_getevents() thread before the issuer is done
2864 * accessing the kiocb cookie.
2865 */
Xiaoguang Wang2e9dbe92020-11-13 00:44:08 +08002866static void io_iopoll_req_issued(struct io_kiocb *req, bool in_async)
Jens Axboedef596e2019-01-09 08:59:42 -07002867{
2868 struct io_ring_ctx *ctx = req->ctx;
2869
2870 /*
2871 * Track whether we have multiple files in our lists. This will impact
2872 * how we do polling eventually, not spinning if we're on potentially
2873 * different devices.
2874 */
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002875 if (list_empty(&ctx->iopoll_list)) {
Jens Axboedef596e2019-01-09 08:59:42 -07002876 ctx->poll_multi_file = false;
2877 } else if (!ctx->poll_multi_file) {
2878 struct io_kiocb *list_req;
2879
Pavel Begunkov540e32a2020-07-13 23:37:09 +03002880 list_req = list_first_entry(&ctx->iopoll_list, struct io_kiocb,
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002881 inflight_entry);
Jens Axboe9adbd452019-12-20 08:45:55 -07002882 if (list_req->file != req->file)
Jens Axboedef596e2019-01-09 08:59:42 -07002883 ctx->poll_multi_file = true;
2884 }
2885
2886 /*
2887 * For fast devices, IO may have already completed. If it has, add
2888 * it to the front so we find it first.
2889 */
Xiaoguang Wang65a65432020-06-11 23:39:36 +08002890 if (READ_ONCE(req->iopoll_completed))
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002891 list_add(&req->inflight_entry, &ctx->iopoll_list);
Jens Axboedef596e2019-01-09 08:59:42 -07002892 else
Pavel Begunkovd21ffe72020-07-13 23:37:10 +03002893 list_add_tail(&req->inflight_entry, &ctx->iopoll_list);
Xiaoguang Wangbdcd3ea2020-02-25 22:12:08 +08002894
Xiaoguang Wang2e9dbe92020-11-13 00:44:08 +08002895 /*
2896 * If IORING_SETUP_SQPOLL is enabled, sqes are either handled in sq thread
2897 * task context or in io worker task context. If current task context is
2898 * sq thread, we don't need to check whether should wake up sq thread.
2899 */
2900 if (in_async && (ctx->flags & IORING_SETUP_SQPOLL) &&
Jens Axboe534ca6d2020-09-02 13:52:19 -06002901 wq_has_sleeper(&ctx->sq_data->wait))
2902 wake_up(&ctx->sq_data->wait);
Jens Axboedef596e2019-01-09 08:59:42 -07002903}
2904
Pavel Begunkov9f13c352020-05-17 14:13:41 +03002905static inline void io_state_file_put(struct io_submit_state *state)
2906{
Pavel Begunkov02b23a92021-01-19 13:32:41 +00002907 if (state->file_refs) {
2908 fput_many(state->file, state->file_refs);
2909 state->file_refs = 0;
2910 }
Jens Axboe9a56a232019-01-09 09:06:50 -07002911}
2912
2913/*
2914 * Get as many references to a file as we have IOs left in this submission,
2915 * assuming most submissions are for one file, or at least that each file
2916 * has more than one submission.
2917 */
Pavel Begunkov8da11c12020-02-24 11:32:44 +03002918static struct file *__io_file_get(struct io_submit_state *state, int fd)
Jens Axboe9a56a232019-01-09 09:06:50 -07002919{
2920 if (!state)
2921 return fget(fd);
2922
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002923 if (state->file_refs) {
Jens Axboe9a56a232019-01-09 09:06:50 -07002924 if (state->fd == fd) {
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002925 state->file_refs--;
Jens Axboe9a56a232019-01-09 09:06:50 -07002926 return state->file;
2927 }
Pavel Begunkov02b23a92021-01-19 13:32:41 +00002928 io_state_file_put(state);
Jens Axboe9a56a232019-01-09 09:06:50 -07002929 }
2930 state->file = fget_many(fd, state->ios_left);
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002931 if (unlikely(!state->file))
Jens Axboe9a56a232019-01-09 09:06:50 -07002932 return NULL;
2933
2934 state->fd = fd;
Pavel Begunkov6e1271e2020-11-20 15:50:50 +00002935 state->file_refs = state->ios_left - 1;
Jens Axboe9a56a232019-01-09 09:06:50 -07002936 return state->file;
2937}
2938
Jens Axboe4503b762020-06-01 10:00:27 -06002939static bool io_bdev_nowait(struct block_device *bdev)
2940{
Jeffle Xu9ba0d0c2020-10-19 16:59:42 +08002941 return !bdev || blk_queue_nowait(bdev_get_queue(bdev));
Jens Axboe4503b762020-06-01 10:00:27 -06002942}
2943
Jens Axboe2b188cc2019-01-07 10:46:33 -07002944/*
2945 * If we tracked the file through the SCM inflight mechanism, we could support
2946 * any file. For now, just ensure that anything potentially problematic is done
2947 * inline.
2948 */
Jens Axboeaf197f52020-04-28 13:15:06 -06002949static bool io_file_supports_async(struct file *file, int rw)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002950{
2951 umode_t mode = file_inode(file)->i_mode;
2952
Jens Axboe4503b762020-06-01 10:00:27 -06002953 if (S_ISBLK(mode)) {
Christoph Hellwig4e7b5672020-11-23 13:38:40 +01002954 if (IS_ENABLED(CONFIG_BLOCK) &&
2955 io_bdev_nowait(I_BDEV(file->f_mapping->host)))
Jens Axboe4503b762020-06-01 10:00:27 -06002956 return true;
2957 return false;
2958 }
2959 if (S_ISCHR(mode) || S_ISSOCK(mode))
Jens Axboe2b188cc2019-01-07 10:46:33 -07002960 return true;
Jens Axboe4503b762020-06-01 10:00:27 -06002961 if (S_ISREG(mode)) {
Christoph Hellwig4e7b5672020-11-23 13:38:40 +01002962 if (IS_ENABLED(CONFIG_BLOCK) &&
2963 io_bdev_nowait(file->f_inode->i_sb->s_bdev) &&
Jens Axboe4503b762020-06-01 10:00:27 -06002964 file->f_op != &io_uring_fops)
2965 return true;
2966 return false;
2967 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002968
Jens Axboec5b85622020-06-09 19:23:05 -06002969 /* any ->read/write should understand O_NONBLOCK */
2970 if (file->f_flags & O_NONBLOCK)
2971 return true;
2972
Jens Axboeaf197f52020-04-28 13:15:06 -06002973 if (!(file->f_mode & FMODE_NOWAIT))
2974 return false;
2975
2976 if (rw == READ)
2977 return file->f_op->read_iter != NULL;
2978
2979 return file->f_op->write_iter != NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002980}
2981
Pavel Begunkova88fc402020-09-30 22:57:53 +03002982static int io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe2b188cc2019-01-07 10:46:33 -07002983{
Jens Axboedef596e2019-01-09 08:59:42 -07002984 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe9adbd452019-12-20 08:45:55 -07002985 struct kiocb *kiocb = &req->rw.kiocb;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002986 struct file *file = req->file;
Jens Axboe09bb8392019-03-13 12:39:28 -06002987 unsigned ioprio;
2988 int ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07002989
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002990 if (S_ISREG(file_inode(file)->i_mode))
Jens Axboe491381ce2019-10-17 09:20:46 -06002991 req->flags |= REQ_F_ISREG;
2992
Jens Axboe2b188cc2019-01-07 10:46:33 -07002993 kiocb->ki_pos = READ_ONCE(sqe->off);
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002994 if (kiocb->ki_pos == -1 && !(file->f_mode & FMODE_STREAM)) {
Jens Axboeba042912019-12-25 16:33:42 -07002995 req->flags |= REQ_F_CUR_POS;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00002996 kiocb->ki_pos = file->f_pos;
Jens Axboeba042912019-12-25 16:33:42 -07002997 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07002998 kiocb->ki_hint = ki_hint_validate(file_write_hint(kiocb->ki_filp));
Pavel Begunkov3e577dc2020-02-01 03:58:42 +03002999 kiocb->ki_flags = iocb_flags(kiocb->ki_filp);
3000 ret = kiocb_set_rw_flags(kiocb, READ_ONCE(sqe->rw_flags));
3001 if (unlikely(ret))
3002 return ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003003
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003004 /* don't allow async punt for O_NONBLOCK or RWF_NOWAIT */
3005 if ((kiocb->ki_flags & IOCB_NOWAIT) || (file->f_flags & O_NONBLOCK))
3006 req->flags |= REQ_F_NOWAIT;
3007
Jens Axboe2b188cc2019-01-07 10:46:33 -07003008 ioprio = READ_ONCE(sqe->ioprio);
3009 if (ioprio) {
3010 ret = ioprio_check_cap(ioprio);
3011 if (ret)
Jens Axboe09bb8392019-03-13 12:39:28 -06003012 return ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003013
3014 kiocb->ki_ioprio = ioprio;
3015 } else
3016 kiocb->ki_ioprio = get_current_ioprio();
3017
Jens Axboedef596e2019-01-09 08:59:42 -07003018 if (ctx->flags & IORING_SETUP_IOPOLL) {
Jens Axboedef596e2019-01-09 08:59:42 -07003019 if (!(kiocb->ki_flags & IOCB_DIRECT) ||
3020 !kiocb->ki_filp->f_op->iopoll)
Jens Axboe09bb8392019-03-13 12:39:28 -06003021 return -EOPNOTSUPP;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003022
Jens Axboedef596e2019-01-09 08:59:42 -07003023 kiocb->ki_flags |= IOCB_HIPRI;
3024 kiocb->ki_complete = io_complete_rw_iopoll;
Xiaoguang Wang65a65432020-06-11 23:39:36 +08003025 req->iopoll_completed = 0;
Jens Axboedef596e2019-01-09 08:59:42 -07003026 } else {
Jens Axboe09bb8392019-03-13 12:39:28 -06003027 if (kiocb->ki_flags & IOCB_HIPRI)
3028 return -EINVAL;
Jens Axboedef596e2019-01-09 08:59:42 -07003029 kiocb->ki_complete = io_complete_rw;
3030 }
Jens Axboe9adbd452019-12-20 08:45:55 -07003031
Jens Axboe3529d8c2019-12-19 18:24:38 -07003032 req->rw.addr = READ_ONCE(sqe->addr);
3033 req->rw.len = READ_ONCE(sqe->len);
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07003034 req->buf_index = READ_ONCE(sqe->buf_index);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003035 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003036}
3037
3038static inline void io_rw_done(struct kiocb *kiocb, ssize_t ret)
3039{
3040 switch (ret) {
3041 case -EIOCBQUEUED:
3042 break;
3043 case -ERESTARTSYS:
3044 case -ERESTARTNOINTR:
3045 case -ERESTARTNOHAND:
3046 case -ERESTART_RESTARTBLOCK:
3047 /*
3048 * We can't just restart the syscall, since previously
3049 * submitted sqes may already be in progress. Just fail this
3050 * IO with EINTR.
3051 */
3052 ret = -EINTR;
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05003053 fallthrough;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003054 default:
3055 kiocb->ki_complete(kiocb, ret, 0);
3056 }
3057}
3058
Jens Axboea1d7c392020-06-22 11:09:46 -06003059static void kiocb_done(struct kiocb *kiocb, ssize_t ret,
Pavel Begunkov889fca72021-02-10 00:03:09 +00003060 unsigned int issue_flags)
Jens Axboeba816ad2019-09-28 11:36:45 -06003061{
Jens Axboeba042912019-12-25 16:33:42 -07003062 struct io_kiocb *req = container_of(kiocb, struct io_kiocb, rw.kiocb);
Jens Axboee8c2bc12020-08-15 18:44:09 -07003063 struct io_async_rw *io = req->async_data;
Jens Axboeba042912019-12-25 16:33:42 -07003064
Jens Axboe227c0c92020-08-13 11:51:40 -06003065 /* add previously done IO, if any */
Jens Axboee8c2bc12020-08-15 18:44:09 -07003066 if (io && io->bytes_done > 0) {
Jens Axboe227c0c92020-08-13 11:51:40 -06003067 if (ret < 0)
Jens Axboee8c2bc12020-08-15 18:44:09 -07003068 ret = io->bytes_done;
Jens Axboe227c0c92020-08-13 11:51:40 -06003069 else
Jens Axboee8c2bc12020-08-15 18:44:09 -07003070 ret += io->bytes_done;
Jens Axboe227c0c92020-08-13 11:51:40 -06003071 }
3072
Jens Axboeba042912019-12-25 16:33:42 -07003073 if (req->flags & REQ_F_CUR_POS)
3074 req->file->f_pos = kiocb->ki_pos;
Pavel Begunkovbcaec082020-02-24 11:30:18 +03003075 if (ret >= 0 && kiocb->ki_complete == io_complete_rw)
Pavel Begunkov889fca72021-02-10 00:03:09 +00003076 __io_complete_rw(req, ret, 0, issue_flags);
Jens Axboeba816ad2019-09-28 11:36:45 -06003077 else
3078 io_rw_done(kiocb, ret);
3079}
3080
Pavel Begunkov847595d2021-02-04 13:52:06 +00003081static int io_import_fixed(struct io_kiocb *req, int rw, struct iov_iter *iter)
Jens Axboeedafcce2019-01-09 09:16:05 -07003082{
Jens Axboe9adbd452019-12-20 08:45:55 -07003083 struct io_ring_ctx *ctx = req->ctx;
3084 size_t len = req->rw.len;
Jens Axboeedafcce2019-01-09 09:16:05 -07003085 struct io_mapped_ubuf *imu;
Pavel Begunkov4be1c612020-09-06 00:45:48 +03003086 u16 index, buf_index = req->buf_index;
Jens Axboeedafcce2019-01-09 09:16:05 -07003087 size_t offset;
3088 u64 buf_addr;
3089
Jens Axboeedafcce2019-01-09 09:16:05 -07003090 if (unlikely(buf_index >= ctx->nr_user_bufs))
3091 return -EFAULT;
Jens Axboeedafcce2019-01-09 09:16:05 -07003092 index = array_index_nospec(buf_index, ctx->nr_user_bufs);
3093 imu = &ctx->user_bufs[index];
Jens Axboe9adbd452019-12-20 08:45:55 -07003094 buf_addr = req->rw.addr;
Jens Axboeedafcce2019-01-09 09:16:05 -07003095
3096 /* overflow */
3097 if (buf_addr + len < buf_addr)
3098 return -EFAULT;
3099 /* not inside the mapped region */
3100 if (buf_addr < imu->ubuf || buf_addr + len > imu->ubuf + imu->len)
3101 return -EFAULT;
3102
3103 /*
3104 * May not be a start of buffer, set size appropriately
3105 * and advance us to the beginning.
3106 */
3107 offset = buf_addr - imu->ubuf;
3108 iov_iter_bvec(iter, rw, imu->bvec, imu->nr_bvecs, offset + len);
Jens Axboebd11b3a2019-07-20 08:37:31 -06003109
3110 if (offset) {
3111 /*
3112 * Don't use iov_iter_advance() here, as it's really slow for
3113 * using the latter parts of a big fixed buffer - it iterates
3114 * over each segment manually. We can cheat a bit here, because
3115 * we know that:
3116 *
3117 * 1) it's a BVEC iter, we set it up
3118 * 2) all bvecs are PAGE_SIZE in size, except potentially the
3119 * first and last bvec
3120 *
3121 * So just find our index, and adjust the iterator afterwards.
3122 * If the offset is within the first bvec (or the whole first
3123 * bvec, just use iov_iter_advance(). This makes it easier
3124 * since we can just skip the first segment, which may not
3125 * be PAGE_SIZE aligned.
3126 */
3127 const struct bio_vec *bvec = imu->bvec;
3128
3129 if (offset <= bvec->bv_len) {
3130 iov_iter_advance(iter, offset);
3131 } else {
3132 unsigned long seg_skip;
3133
3134 /* skip first vec */
3135 offset -= bvec->bv_len;
3136 seg_skip = 1 + (offset >> PAGE_SHIFT);
3137
3138 iter->bvec = bvec + seg_skip;
3139 iter->nr_segs -= seg_skip;
Aleix Roca Nonell99c79f62019-08-15 14:03:22 +02003140 iter->count -= bvec->bv_len + offset;
Jens Axboebd11b3a2019-07-20 08:37:31 -06003141 iter->iov_offset = offset & ~PAGE_MASK;
Jens Axboebd11b3a2019-07-20 08:37:31 -06003142 }
3143 }
3144
Pavel Begunkov847595d2021-02-04 13:52:06 +00003145 return 0;
Jens Axboeedafcce2019-01-09 09:16:05 -07003146}
3147
Jens Axboebcda7ba2020-02-23 16:42:51 -07003148static void io_ring_submit_unlock(struct io_ring_ctx *ctx, bool needs_lock)
3149{
3150 if (needs_lock)
3151 mutex_unlock(&ctx->uring_lock);
3152}
3153
3154static void io_ring_submit_lock(struct io_ring_ctx *ctx, bool needs_lock)
3155{
3156 /*
3157 * "Normal" inline submissions always hold the uring_lock, since we
3158 * grab it from the system call. Same is true for the SQPOLL offload.
3159 * The only exception is when we've detached the request and issue it
3160 * from an async worker thread, grab the lock for that case.
3161 */
3162 if (needs_lock)
3163 mutex_lock(&ctx->uring_lock);
3164}
3165
3166static struct io_buffer *io_buffer_select(struct io_kiocb *req, size_t *len,
3167 int bgid, struct io_buffer *kbuf,
3168 bool needs_lock)
3169{
3170 struct io_buffer *head;
3171
3172 if (req->flags & REQ_F_BUFFER_SELECTED)
3173 return kbuf;
3174
3175 io_ring_submit_lock(req->ctx, needs_lock);
3176
3177 lockdep_assert_held(&req->ctx->uring_lock);
3178
3179 head = idr_find(&req->ctx->io_buffer_idr, bgid);
3180 if (head) {
3181 if (!list_empty(&head->list)) {
3182 kbuf = list_last_entry(&head->list, struct io_buffer,
3183 list);
3184 list_del(&kbuf->list);
3185 } else {
3186 kbuf = head;
3187 idr_remove(&req->ctx->io_buffer_idr, bgid);
3188 }
3189 if (*len > kbuf->len)
3190 *len = kbuf->len;
3191 } else {
3192 kbuf = ERR_PTR(-ENOBUFS);
3193 }
3194
3195 io_ring_submit_unlock(req->ctx, needs_lock);
3196
3197 return kbuf;
3198}
3199
Jens Axboe4d954c22020-02-27 07:31:19 -07003200static void __user *io_rw_buffer_select(struct io_kiocb *req, size_t *len,
3201 bool needs_lock)
3202{
3203 struct io_buffer *kbuf;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07003204 u16 bgid;
Jens Axboe4d954c22020-02-27 07:31:19 -07003205
3206 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07003207 bgid = req->buf_index;
Jens Axboe4d954c22020-02-27 07:31:19 -07003208 kbuf = io_buffer_select(req, len, bgid, kbuf, needs_lock);
3209 if (IS_ERR(kbuf))
3210 return kbuf;
3211 req->rw.addr = (u64) (unsigned long) kbuf;
3212 req->flags |= REQ_F_BUFFER_SELECTED;
3213 return u64_to_user_ptr(kbuf->addr);
3214}
3215
3216#ifdef CONFIG_COMPAT
3217static ssize_t io_compat_import(struct io_kiocb *req, struct iovec *iov,
3218 bool needs_lock)
3219{
3220 struct compat_iovec __user *uiov;
3221 compat_ssize_t clen;
3222 void __user *buf;
3223 ssize_t len;
3224
3225 uiov = u64_to_user_ptr(req->rw.addr);
3226 if (!access_ok(uiov, sizeof(*uiov)))
3227 return -EFAULT;
3228 if (__get_user(clen, &uiov->iov_len))
3229 return -EFAULT;
3230 if (clen < 0)
3231 return -EINVAL;
3232
3233 len = clen;
3234 buf = io_rw_buffer_select(req, &len, needs_lock);
3235 if (IS_ERR(buf))
3236 return PTR_ERR(buf);
3237 iov[0].iov_base = buf;
3238 iov[0].iov_len = (compat_size_t) len;
3239 return 0;
3240}
3241#endif
3242
3243static ssize_t __io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov,
3244 bool needs_lock)
3245{
3246 struct iovec __user *uiov = u64_to_user_ptr(req->rw.addr);
3247 void __user *buf;
3248 ssize_t len;
3249
3250 if (copy_from_user(iov, uiov, sizeof(*uiov)))
3251 return -EFAULT;
3252
3253 len = iov[0].iov_len;
3254 if (len < 0)
3255 return -EINVAL;
3256 buf = io_rw_buffer_select(req, &len, needs_lock);
3257 if (IS_ERR(buf))
3258 return PTR_ERR(buf);
3259 iov[0].iov_base = buf;
3260 iov[0].iov_len = len;
3261 return 0;
3262}
3263
3264static ssize_t io_iov_buffer_select(struct io_kiocb *req, struct iovec *iov,
3265 bool needs_lock)
3266{
Jens Axboedddb3e22020-06-04 11:27:01 -06003267 if (req->flags & REQ_F_BUFFER_SELECTED) {
3268 struct io_buffer *kbuf;
3269
3270 kbuf = (struct io_buffer *) (unsigned long) req->rw.addr;
3271 iov[0].iov_base = u64_to_user_ptr(kbuf->addr);
3272 iov[0].iov_len = kbuf->len;
Jens Axboe4d954c22020-02-27 07:31:19 -07003273 return 0;
Jens Axboedddb3e22020-06-04 11:27:01 -06003274 }
Pavel Begunkovdd201662020-12-19 03:15:43 +00003275 if (req->rw.len != 1)
Jens Axboe4d954c22020-02-27 07:31:19 -07003276 return -EINVAL;
3277
3278#ifdef CONFIG_COMPAT
3279 if (req->ctx->compat)
3280 return io_compat_import(req, iov, needs_lock);
3281#endif
3282
3283 return __io_iov_buffer_select(req, iov, needs_lock);
3284}
3285
Pavel Begunkov847595d2021-02-04 13:52:06 +00003286static int io_import_iovec(int rw, struct io_kiocb *req, struct iovec **iovec,
3287 struct iov_iter *iter, bool needs_lock)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003288{
Jens Axboe9adbd452019-12-20 08:45:55 -07003289 void __user *buf = u64_to_user_ptr(req->rw.addr);
3290 size_t sqe_len = req->rw.len;
Pavel Begunkov847595d2021-02-04 13:52:06 +00003291 u8 opcode = req->opcode;
Jens Axboe4d954c22020-02-27 07:31:19 -07003292 ssize_t ret;
Jens Axboeedafcce2019-01-09 09:16:05 -07003293
Pavel Begunkov7d009162019-11-25 23:14:40 +03003294 if (opcode == IORING_OP_READ_FIXED || opcode == IORING_OP_WRITE_FIXED) {
Jens Axboeedafcce2019-01-09 09:16:05 -07003295 *iovec = NULL;
Jens Axboe9adbd452019-12-20 08:45:55 -07003296 return io_import_fixed(req, rw, iter);
Jens Axboeedafcce2019-01-09 09:16:05 -07003297 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07003298
Jens Axboebcda7ba2020-02-23 16:42:51 -07003299 /* buffer index only valid with fixed read/write, or buffer select */
Bijan Mottahedeh4f4eeba2020-05-19 14:52:49 -07003300 if (req->buf_index && !(req->flags & REQ_F_BUFFER_SELECT))
Jens Axboe9adbd452019-12-20 08:45:55 -07003301 return -EINVAL;
3302
Jens Axboe3a6820f2019-12-22 15:19:35 -07003303 if (opcode == IORING_OP_READ || opcode == IORING_OP_WRITE) {
Jens Axboebcda7ba2020-02-23 16:42:51 -07003304 if (req->flags & REQ_F_BUFFER_SELECT) {
Jens Axboe4d954c22020-02-27 07:31:19 -07003305 buf = io_rw_buffer_select(req, &sqe_len, needs_lock);
Pavel Begunkov867a23e2020-08-20 11:34:39 +03003306 if (IS_ERR(buf))
Jens Axboe4d954c22020-02-27 07:31:19 -07003307 return PTR_ERR(buf);
Jens Axboe3f9d6442020-03-11 12:27:04 -06003308 req->rw.len = sqe_len;
Jens Axboebcda7ba2020-02-23 16:42:51 -07003309 }
3310
Jens Axboe3a6820f2019-12-22 15:19:35 -07003311 ret = import_single_range(rw, buf, sqe_len, *iovec, iter);
3312 *iovec = NULL;
David Laight10fc72e2020-11-07 13:16:25 +00003313 return ret;
Jens Axboe3a6820f2019-12-22 15:19:35 -07003314 }
3315
Jens Axboe4d954c22020-02-27 07:31:19 -07003316 if (req->flags & REQ_F_BUFFER_SELECT) {
3317 ret = io_iov_buffer_select(req, *iovec, needs_lock);
Pavel Begunkov847595d2021-02-04 13:52:06 +00003318 if (!ret)
3319 iov_iter_init(iter, rw, *iovec, 1, (*iovec)->iov_len);
Jens Axboe4d954c22020-02-27 07:31:19 -07003320 *iovec = NULL;
3321 return ret;
3322 }
3323
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02003324 return __import_iovec(rw, buf, sqe_len, UIO_FASTIOV, iovec, iter,
3325 req->ctx->compat);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003326}
3327
Jens Axboe0fef9482020-08-26 10:36:20 -06003328static inline loff_t *io_kiocb_ppos(struct kiocb *kiocb)
3329{
Pavel Begunkov5b09e372020-09-30 22:57:15 +03003330 return (kiocb->ki_filp->f_mode & FMODE_STREAM) ? NULL : &kiocb->ki_pos;
Jens Axboe0fef9482020-08-26 10:36:20 -06003331}
3332
Jens Axboe32960612019-09-23 11:05:34 -06003333/*
3334 * For files that don't have ->read_iter() and ->write_iter(), handle them
3335 * by looping over ->read() or ->write() manually.
3336 */
Jens Axboe4017eb92020-10-22 14:14:12 -06003337static ssize_t loop_rw_iter(int rw, struct io_kiocb *req, struct iov_iter *iter)
Jens Axboe32960612019-09-23 11:05:34 -06003338{
Jens Axboe4017eb92020-10-22 14:14:12 -06003339 struct kiocb *kiocb = &req->rw.kiocb;
3340 struct file *file = req->file;
Jens Axboe32960612019-09-23 11:05:34 -06003341 ssize_t ret = 0;
3342
3343 /*
3344 * Don't support polled IO through this interface, and we can't
3345 * support non-blocking either. For the latter, this just causes
3346 * the kiocb to be handled from an async context.
3347 */
3348 if (kiocb->ki_flags & IOCB_HIPRI)
3349 return -EOPNOTSUPP;
3350 if (kiocb->ki_flags & IOCB_NOWAIT)
3351 return -EAGAIN;
3352
3353 while (iov_iter_count(iter)) {
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003354 struct iovec iovec;
Jens Axboe32960612019-09-23 11:05:34 -06003355 ssize_t nr;
3356
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003357 if (!iov_iter_is_bvec(iter)) {
3358 iovec = iov_iter_iovec(iter);
3359 } else {
Jens Axboe4017eb92020-10-22 14:14:12 -06003360 iovec.iov_base = u64_to_user_ptr(req->rw.addr);
3361 iovec.iov_len = req->rw.len;
Pavel Begunkov311ae9e2019-11-24 11:58:24 +03003362 }
3363
Jens Axboe32960612019-09-23 11:05:34 -06003364 if (rw == READ) {
3365 nr = file->f_op->read(file, iovec.iov_base,
Jens Axboe0fef9482020-08-26 10:36:20 -06003366 iovec.iov_len, io_kiocb_ppos(kiocb));
Jens Axboe32960612019-09-23 11:05:34 -06003367 } else {
3368 nr = file->f_op->write(file, iovec.iov_base,
Jens Axboe0fef9482020-08-26 10:36:20 -06003369 iovec.iov_len, io_kiocb_ppos(kiocb));
Jens Axboe32960612019-09-23 11:05:34 -06003370 }
3371
3372 if (nr < 0) {
3373 if (!ret)
3374 ret = nr;
3375 break;
3376 }
3377 ret += nr;
3378 if (nr != iovec.iov_len)
3379 break;
Jens Axboe4017eb92020-10-22 14:14:12 -06003380 req->rw.len -= nr;
3381 req->rw.addr += nr;
Jens Axboe32960612019-09-23 11:05:34 -06003382 iov_iter_advance(iter, nr);
3383 }
3384
3385 return ret;
3386}
3387
Jens Axboeff6165b2020-08-13 09:47:43 -06003388static void io_req_map_rw(struct io_kiocb *req, const struct iovec *iovec,
3389 const struct iovec *fast_iov, struct iov_iter *iter)
Jens Axboef67676d2019-12-02 11:03:47 -07003390{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003391 struct io_async_rw *rw = req->async_data;
Pavel Begunkovb64e3442020-07-13 22:59:18 +03003392
Jens Axboeff6165b2020-08-13 09:47:43 -06003393 memcpy(&rw->iter, iter, sizeof(*iter));
Pavel Begunkovafb87652020-09-06 00:45:46 +03003394 rw->free_iovec = iovec;
Jens Axboe227c0c92020-08-13 11:51:40 -06003395 rw->bytes_done = 0;
Jens Axboeff6165b2020-08-13 09:47:43 -06003396 /* can only be fixed buffers, no need to do anything */
Pavel Begunkov9c3a2052020-11-23 23:20:27 +00003397 if (iov_iter_is_bvec(iter))
Jens Axboeff6165b2020-08-13 09:47:43 -06003398 return;
Pavel Begunkovb64e3442020-07-13 22:59:18 +03003399 if (!iovec) {
Jens Axboeff6165b2020-08-13 09:47:43 -06003400 unsigned iov_off = 0;
3401
3402 rw->iter.iov = rw->fast_iov;
3403 if (iter->iov != fast_iov) {
3404 iov_off = iter->iov - fast_iov;
3405 rw->iter.iov += iov_off;
3406 }
3407 if (rw->fast_iov != fast_iov)
3408 memcpy(rw->fast_iov + iov_off, fast_iov + iov_off,
Xiaoguang Wang45097da2020-04-08 22:29:58 +08003409 sizeof(struct iovec) * iter->nr_segs);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03003410 } else {
3411 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboef67676d2019-12-02 11:03:47 -07003412 }
3413}
3414
Jens Axboee8c2bc12020-08-15 18:44:09 -07003415static inline int __io_alloc_async_data(struct io_kiocb *req)
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003416{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003417 WARN_ON_ONCE(!io_op_defs[req->opcode].async_size);
3418 req->async_data = kmalloc(io_op_defs[req->opcode].async_size, GFP_KERNEL);
3419 return req->async_data == NULL;
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003420}
3421
Jens Axboee8c2bc12020-08-15 18:44:09 -07003422static int io_alloc_async_data(struct io_kiocb *req)
Jens Axboef67676d2019-12-02 11:03:47 -07003423{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003424 if (!io_op_defs[req->opcode].needs_async_data)
Jens Axboed3656342019-12-18 09:50:26 -07003425 return 0;
Xiaoguang Wang3d9932a2020-03-27 15:36:52 +08003426
Jens Axboee8c2bc12020-08-15 18:44:09 -07003427 return __io_alloc_async_data(req);
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003428}
3429
Jens Axboeff6165b2020-08-13 09:47:43 -06003430static int io_setup_async_rw(struct io_kiocb *req, const struct iovec *iovec,
3431 const struct iovec *fast_iov,
Jens Axboe227c0c92020-08-13 11:51:40 -06003432 struct iov_iter *iter, bool force)
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003433{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003434 if (!force && !io_op_defs[req->opcode].needs_async_data)
Jens Axboe74566df2020-01-13 19:23:24 -07003435 return 0;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003436 if (!req->async_data) {
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003437 if (__io_alloc_async_data(req)) {
3438 kfree(iovec);
Jens Axboe5d204bc2020-01-31 12:06:52 -07003439 return -ENOMEM;
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003440 }
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003441
Jens Axboeff6165b2020-08-13 09:47:43 -06003442 io_req_map_rw(req, iovec, fast_iov, iter);
Jens Axboe5d204bc2020-01-31 12:06:52 -07003443 }
Jens Axboeb7bb4f72019-12-15 22:13:43 -07003444 return 0;
Jens Axboef67676d2019-12-02 11:03:47 -07003445}
3446
Pavel Begunkov73debe62020-09-30 22:57:54 +03003447static inline int io_rw_prep_async(struct io_kiocb *req, int rw)
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003448{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003449 struct io_async_rw *iorw = req->async_data;
Pavel Begunkovf4bff102020-09-06 00:45:45 +03003450 struct iovec *iov = iorw->fast_iov;
Pavel Begunkov847595d2021-02-04 13:52:06 +00003451 int ret;
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003452
Pavel Begunkov2846c482020-11-07 13:16:27 +00003453 ret = io_import_iovec(rw, req, &iov, &iorw->iter, false);
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003454 if (unlikely(ret < 0))
3455 return ret;
3456
Pavel Begunkovab0b1962020-09-06 00:45:47 +03003457 iorw->bytes_done = 0;
3458 iorw->free_iovec = iov;
3459 if (iov)
3460 req->flags |= REQ_F_NEED_CLEANUP;
Pavel Begunkovc3e330a2020-07-13 22:59:19 +03003461 return 0;
3462}
3463
Pavel Begunkov73debe62020-09-30 22:57:54 +03003464static int io_read_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07003465{
3466 ssize_t ret;
3467
Pavel Begunkova88fc402020-09-30 22:57:53 +03003468 ret = io_prep_rw(req, sqe);
Jens Axboe3529d8c2019-12-19 18:24:38 -07003469 if (ret)
3470 return ret;
Jens Axboef67676d2019-12-02 11:03:47 -07003471
Jens Axboe3529d8c2019-12-19 18:24:38 -07003472 if (unlikely(!(req->file->f_mode & FMODE_READ)))
3473 return -EBADF;
Jens Axboef67676d2019-12-02 11:03:47 -07003474
Pavel Begunkov5f798be2020-02-08 13:28:02 +03003475 /* either don't need iovec imported or already have it */
Pavel Begunkov2d199892020-09-30 22:57:35 +03003476 if (!req->async_data)
Jens Axboe3529d8c2019-12-19 18:24:38 -07003477 return 0;
Pavel Begunkov73debe62020-09-30 22:57:54 +03003478 return io_rw_prep_async(req, READ);
Jens Axboef67676d2019-12-02 11:03:47 -07003479}
3480
Jens Axboec1dd91d2020-08-03 16:43:59 -06003481/*
3482 * This is our waitqueue callback handler, registered through lock_page_async()
3483 * when we initially tried to do the IO with the iocb armed our waitqueue.
3484 * This gets called when the page is unlocked, and we generally expect that to
3485 * happen when the page IO is completed and the page is now uptodate. This will
3486 * queue a task_work based retry of the operation, attempting to copy the data
3487 * again. If the latter fails because the page was NOT uptodate, then we will
3488 * do a thread based blocking retry of the operation. That's the unexpected
3489 * slow path.
3490 */
Jens Axboebcf5a062020-05-22 09:24:42 -06003491static int io_async_buf_func(struct wait_queue_entry *wait, unsigned mode,
3492 int sync, void *arg)
3493{
3494 struct wait_page_queue *wpq;
3495 struct io_kiocb *req = wait->private;
Jens Axboebcf5a062020-05-22 09:24:42 -06003496 struct wait_page_key *key = arg;
Jens Axboebcf5a062020-05-22 09:24:42 -06003497
3498 wpq = container_of(wait, struct wait_page_queue, wait);
3499
Linus Torvaldscdc8fcb2020-08-03 13:01:22 -07003500 if (!wake_page_match(wpq, key))
3501 return 0;
3502
Hao Xuc8d317a2020-09-29 20:00:45 +08003503 req->rw.kiocb.ki_flags &= ~IOCB_WAITQ;
Jens Axboebcf5a062020-05-22 09:24:42 -06003504 list_del_init(&wait->entry);
3505
Jens Axboebcf5a062020-05-22 09:24:42 -06003506 /* submit ref gets dropped, acquire a new one */
3507 refcount_inc(&req->refs);
Pavel Begunkov921b9052021-02-12 03:23:53 +00003508 io_req_task_queue(req);
Jens Axboebcf5a062020-05-22 09:24:42 -06003509 return 1;
3510}
3511
Jens Axboec1dd91d2020-08-03 16:43:59 -06003512/*
3513 * This controls whether a given IO request should be armed for async page
3514 * based retry. If we return false here, the request is handed to the async
3515 * worker threads for retry. If we're doing buffered reads on a regular file,
3516 * we prepare a private wait_page_queue entry and retry the operation. This
3517 * will either succeed because the page is now uptodate and unlocked, or it
3518 * will register a callback when the page is unlocked at IO completion. Through
3519 * that callback, io_uring uses task_work to setup a retry of the operation.
3520 * That retry will attempt the buffered read again. The retry will generally
3521 * succeed, or in rare cases where it fails, we then fall back to using the
3522 * async worker threads for a blocking retry.
3523 */
Jens Axboe227c0c92020-08-13 11:51:40 -06003524static bool io_rw_should_retry(struct io_kiocb *req)
Jens Axboebcf5a062020-05-22 09:24:42 -06003525{
Jens Axboee8c2bc12020-08-15 18:44:09 -07003526 struct io_async_rw *rw = req->async_data;
3527 struct wait_page_queue *wait = &rw->wpq;
Jens Axboebcf5a062020-05-22 09:24:42 -06003528 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboebcf5a062020-05-22 09:24:42 -06003529
3530 /* never retry for NOWAIT, we just complete with -EAGAIN */
3531 if (req->flags & REQ_F_NOWAIT)
3532 return false;
3533
Jens Axboe227c0c92020-08-13 11:51:40 -06003534 /* Only for buffered IO */
Jens Axboe3b2a4432020-08-16 10:58:43 -07003535 if (kiocb->ki_flags & (IOCB_DIRECT | IOCB_HIPRI))
Jens Axboebcf5a062020-05-22 09:24:42 -06003536 return false;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003537
Jens Axboebcf5a062020-05-22 09:24:42 -06003538 /*
3539 * just use poll if we can, and don't attempt if the fs doesn't
3540 * support callback based unlocks
3541 */
3542 if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC))
3543 return false;
3544
Jens Axboe3b2a4432020-08-16 10:58:43 -07003545 wait->wait.func = io_async_buf_func;
3546 wait->wait.private = req;
3547 wait->wait.flags = 0;
3548 INIT_LIST_HEAD(&wait->wait.entry);
3549 kiocb->ki_flags |= IOCB_WAITQ;
Hao Xuc8d317a2020-09-29 20:00:45 +08003550 kiocb->ki_flags &= ~IOCB_NOWAIT;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003551 kiocb->ki_waitq = wait;
Jens Axboe3b2a4432020-08-16 10:58:43 -07003552 return true;
Jens Axboebcf5a062020-05-22 09:24:42 -06003553}
3554
3555static int io_iter_do_read(struct io_kiocb *req, struct iov_iter *iter)
3556{
3557 if (req->file->f_op->read_iter)
3558 return call_read_iter(req->file, &req->rw.kiocb, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003559 else if (req->file->f_op->read)
Jens Axboe4017eb92020-10-22 14:14:12 -06003560 return loop_rw_iter(READ, req, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003561 else
3562 return -EINVAL;
Jens Axboebcf5a062020-05-22 09:24:42 -06003563}
3564
Pavel Begunkov889fca72021-02-10 00:03:09 +00003565static int io_read(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003566{
3567 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
Jens Axboe9adbd452019-12-20 08:45:55 -07003568 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboeff6165b2020-08-13 09:47:43 -06003569 struct iov_iter __iter, *iter = &__iter;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003570 struct io_async_rw *rw = req->async_data;
Jens Axboe227c0c92020-08-13 11:51:40 -06003571 ssize_t io_size, ret, ret2;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003572 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003573
Pavel Begunkov2846c482020-11-07 13:16:27 +00003574 if (rw) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07003575 iter = &rw->iter;
Pavel Begunkov2846c482020-11-07 13:16:27 +00003576 iovec = NULL;
3577 } else {
3578 ret = io_import_iovec(READ, req, &iovec, iter, !force_nonblock);
3579 if (ret < 0)
3580 return ret;
3581 }
Pavel Begunkov632546c2020-11-07 13:16:26 +00003582 io_size = iov_iter_count(iter);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003583 req->result = io_size;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003584
Jens Axboefd6c2e42019-12-18 12:19:41 -07003585 /* Ensure we clear previously set non-block flag */
3586 if (!force_nonblock)
Jens Axboe29de5f62020-02-20 09:56:08 -07003587 kiocb->ki_flags &= ~IOCB_NOWAIT;
Pavel Begunkova88fc402020-09-30 22:57:53 +03003588 else
3589 kiocb->ki_flags |= IOCB_NOWAIT;
3590
Pavel Begunkov24c74672020-06-21 13:09:51 +03003591 /* If the file doesn't support async, just async punt */
Pavel Begunkov6713e7a2021-02-04 13:51:59 +00003592 if (force_nonblock && !io_file_supports_async(req->file, READ)) {
3593 ret = io_setup_async_rw(req, iovec, inline_vecs, iter, true);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003594 return ret ?: -EAGAIN;
Pavel Begunkov6713e7a2021-02-04 13:51:59 +00003595 }
Jens Axboe9e645e112019-05-10 16:07:28 -06003596
Pavel Begunkov632546c2020-11-07 13:16:26 +00003597 ret = rw_verify_area(READ, req->file, io_kiocb_ppos(kiocb), io_size);
Pavel Begunkov5ea5dd42021-02-04 13:52:03 +00003598 if (unlikely(ret)) {
3599 kfree(iovec);
3600 return ret;
3601 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07003602
Jens Axboe227c0c92020-08-13 11:51:40 -06003603 ret = io_iter_do_read(req, iter);
Jens Axboe32960612019-09-23 11:05:34 -06003604
Pavel Begunkov57cd6572021-02-01 18:59:56 +00003605 if (ret == -EIOCBQUEUED) {
Pavel Begunkov5ea5dd42021-02-04 13:52:03 +00003606 /* it's faster to check here then delegate to kfree */
3607 if (iovec)
3608 kfree(iovec);
3609 return 0;
Jens Axboe227c0c92020-08-13 11:51:40 -06003610 } else if (ret == -EAGAIN) {
Jens Axboeeefdf302020-08-27 16:40:19 -06003611 /* IOPOLL retry should happen for io-wq threads */
3612 if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboef91daf52020-08-15 15:58:42 -07003613 goto done;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003614 /* no retry on NONBLOCK nor RWF_NOWAIT */
3615 if (req->flags & REQ_F_NOWAIT)
Jens Axboe355afae2020-09-02 09:30:31 -06003616 goto done;
Jens Axboe84216312020-08-24 11:45:26 -06003617 /* some cases will consume bytes even on error returns */
Pavel Begunkov632546c2020-11-07 13:16:26 +00003618 iov_iter_revert(iter, io_size - iov_iter_count(iter));
Jens Axboef38c7e32020-09-25 15:23:43 -06003619 ret = 0;
Pavel Begunkov7335e3b2021-02-04 13:52:02 +00003620 } else if (ret <= 0 || ret == io_size || !force_nonblock ||
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003621 (req->flags & REQ_F_NOWAIT) || !(req->flags & REQ_F_ISREG)) {
Pavel Begunkov7335e3b2021-02-04 13:52:02 +00003622 /* read all, failed, already did sync or don't want to retry */
Jens Axboe00d23d52020-08-25 12:59:22 -06003623 goto done;
Jens Axboe227c0c92020-08-13 11:51:40 -06003624 }
3625
Jens Axboe227c0c92020-08-13 11:51:40 -06003626 ret2 = io_setup_async_rw(req, iovec, inline_vecs, iter, true);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003627 if (ret2)
3628 return ret2;
3629
Jens Axboee8c2bc12020-08-15 18:44:09 -07003630 rw = req->async_data;
Jens Axboe227c0c92020-08-13 11:51:40 -06003631 /* now use our persistent iterator, if we aren't already */
Jens Axboee8c2bc12020-08-15 18:44:09 -07003632 iter = &rw->iter;
Jens Axboe227c0c92020-08-13 11:51:40 -06003633
Pavel Begunkovb23df912021-02-04 13:52:04 +00003634 do {
3635 io_size -= ret;
3636 rw->bytes_done += ret;
3637 /* if we can retry, do so with the callbacks armed */
3638 if (!io_rw_should_retry(req)) {
3639 kiocb->ki_flags &= ~IOCB_WAITQ;
3640 return -EAGAIN;
3641 }
3642
3643 /*
3644 * Now retry read with the IOCB_WAITQ parts set in the iocb. If
3645 * we get -EIOCBQUEUED, then we'll get a notification when the
3646 * desired page gets unlocked. We can also get a partial read
3647 * here, and if we do, then just retry at the new offset.
3648 */
3649 ret = io_iter_do_read(req, iter);
3650 if (ret == -EIOCBQUEUED)
3651 return 0;
3652 /* we got some bytes, but not all. retry. */
3653 } while (ret > 0 && ret < io_size);
Jens Axboe227c0c92020-08-13 11:51:40 -06003654done:
Pavel Begunkov889fca72021-02-10 00:03:09 +00003655 kiocb_done(kiocb, ret, issue_flags);
Pavel Begunkov5ea5dd42021-02-04 13:52:03 +00003656 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003657}
3658
Pavel Begunkov73debe62020-09-30 22:57:54 +03003659static int io_write_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07003660{
3661 ssize_t ret;
3662
Pavel Begunkova88fc402020-09-30 22:57:53 +03003663 ret = io_prep_rw(req, sqe);
Jens Axboe3529d8c2019-12-19 18:24:38 -07003664 if (ret)
3665 return ret;
Jens Axboef67676d2019-12-02 11:03:47 -07003666
Jens Axboe3529d8c2019-12-19 18:24:38 -07003667 if (unlikely(!(req->file->f_mode & FMODE_WRITE)))
3668 return -EBADF;
Jens Axboef67676d2019-12-02 11:03:47 -07003669
Pavel Begunkov5f798be2020-02-08 13:28:02 +03003670 /* either don't need iovec imported or already have it */
Pavel Begunkov2d199892020-09-30 22:57:35 +03003671 if (!req->async_data)
Jens Axboe3529d8c2019-12-19 18:24:38 -07003672 return 0;
Pavel Begunkov73debe62020-09-30 22:57:54 +03003673 return io_rw_prep_async(req, WRITE);
Jens Axboef67676d2019-12-02 11:03:47 -07003674}
3675
Pavel Begunkov889fca72021-02-10 00:03:09 +00003676static int io_write(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07003677{
3678 struct iovec inline_vecs[UIO_FASTIOV], *iovec = inline_vecs;
Jens Axboe9adbd452019-12-20 08:45:55 -07003679 struct kiocb *kiocb = &req->rw.kiocb;
Jens Axboeff6165b2020-08-13 09:47:43 -06003680 struct iov_iter __iter, *iter = &__iter;
Jens Axboee8c2bc12020-08-15 18:44:09 -07003681 struct io_async_rw *rw = req->async_data;
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003682 ssize_t ret, ret2, io_size;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003683 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003684
Pavel Begunkov2846c482020-11-07 13:16:27 +00003685 if (rw) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07003686 iter = &rw->iter;
Pavel Begunkov2846c482020-11-07 13:16:27 +00003687 iovec = NULL;
3688 } else {
3689 ret = io_import_iovec(WRITE, req, &iovec, iter, !force_nonblock);
3690 if (ret < 0)
3691 return ret;
3692 }
Pavel Begunkov632546c2020-11-07 13:16:26 +00003693 io_size = iov_iter_count(iter);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003694 req->result = io_size;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003695
Jens Axboefd6c2e42019-12-18 12:19:41 -07003696 /* Ensure we clear previously set non-block flag */
3697 if (!force_nonblock)
Pavel Begunkova88fc402020-09-30 22:57:53 +03003698 kiocb->ki_flags &= ~IOCB_NOWAIT;
3699 else
3700 kiocb->ki_flags |= IOCB_NOWAIT;
Jens Axboefd6c2e42019-12-18 12:19:41 -07003701
Pavel Begunkov24c74672020-06-21 13:09:51 +03003702 /* If the file doesn't support async, just async punt */
Jens Axboeaf197f52020-04-28 13:15:06 -06003703 if (force_nonblock && !io_file_supports_async(req->file, WRITE))
Jens Axboef67676d2019-12-02 11:03:47 -07003704 goto copy_iov;
Jens Axboef67676d2019-12-02 11:03:47 -07003705
Jens Axboe10d59342019-12-09 20:16:22 -07003706 /* file path doesn't support NOWAIT for non-direct_IO */
3707 if (force_nonblock && !(kiocb->ki_flags & IOCB_DIRECT) &&
3708 (req->flags & REQ_F_ISREG))
Jens Axboef67676d2019-12-02 11:03:47 -07003709 goto copy_iov;
Jens Axboe9e645e112019-05-10 16:07:28 -06003710
Pavel Begunkov632546c2020-11-07 13:16:26 +00003711 ret = rw_verify_area(WRITE, req->file, io_kiocb_ppos(kiocb), io_size);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003712 if (unlikely(ret))
3713 goto out_free;
Roman Penyaev9bf79332019-03-25 20:09:24 +01003714
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003715 /*
3716 * Open-code file_start_write here to grab freeze protection,
3717 * which will be released by another thread in
3718 * io_complete_rw(). Fool lockdep by telling it the lock got
3719 * released so that it doesn't complain about the held lock when
3720 * we return to userspace.
3721 */
3722 if (req->flags & REQ_F_ISREG) {
Darrick J. Wong8a3c84b2020-11-10 16:50:21 -08003723 sb_start_write(file_inode(req->file)->i_sb);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003724 __sb_writers_release(file_inode(req->file)->i_sb,
3725 SB_FREEZE_WRITE);
3726 }
3727 kiocb->ki_flags |= IOCB_WRITE;
Roman Penyaev9bf79332019-03-25 20:09:24 +01003728
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003729 if (req->file->f_op->write_iter)
Jens Axboeff6165b2020-08-13 09:47:43 -06003730 ret2 = call_write_iter(req->file, kiocb, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003731 else if (req->file->f_op->write)
Jens Axboe4017eb92020-10-22 14:14:12 -06003732 ret2 = loop_rw_iter(WRITE, req, iter);
Guoyu Huang2dd21112020-08-05 03:53:50 -07003733 else
3734 ret2 = -EINVAL;
Jens Axboe4ed734b2020-03-20 11:23:41 -06003735
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003736 /*
3737 * Raw bdev writes will return -EOPNOTSUPP for IOCB_NOWAIT. Just
3738 * retry them without IOCB_NOWAIT.
3739 */
3740 if (ret2 == -EOPNOTSUPP && (kiocb->ki_flags & IOCB_NOWAIT))
3741 ret2 = -EAGAIN;
Pavel Begunkov75c668c2021-02-04 13:52:05 +00003742 /* no retry on NONBLOCK nor RWF_NOWAIT */
3743 if (ret2 == -EAGAIN && (req->flags & REQ_F_NOWAIT))
Jens Axboe355afae2020-09-02 09:30:31 -06003744 goto done;
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003745 if (!force_nonblock || ret2 != -EAGAIN) {
Jens Axboeeefdf302020-08-27 16:40:19 -06003746 /* IOPOLL retry should happen for io-wq threads */
3747 if ((req->ctx->flags & IORING_SETUP_IOPOLL) && ret2 == -EAGAIN)
3748 goto copy_iov;
Jens Axboe355afae2020-09-02 09:30:31 -06003749done:
Pavel Begunkov889fca72021-02-10 00:03:09 +00003750 kiocb_done(kiocb, ret2, issue_flags);
Pavel Begunkovfa15baf2020-08-01 13:50:02 +03003751 } else {
Jens Axboef67676d2019-12-02 11:03:47 -07003752copy_iov:
Jens Axboe84216312020-08-24 11:45:26 -06003753 /* some cases will consume bytes even on error returns */
Pavel Begunkov632546c2020-11-07 13:16:26 +00003754 iov_iter_revert(iter, io_size - iov_iter_count(iter));
Jens Axboe227c0c92020-08-13 11:51:40 -06003755 ret = io_setup_async_rw(req, iovec, inline_vecs, iter, false);
Pavel Begunkov6bf985d2021-02-04 13:52:01 +00003756 return ret ?: -EAGAIN;
Jens Axboe2b188cc2019-01-07 10:46:33 -07003757 }
Jens Axboe31b51512019-01-18 22:56:34 -07003758out_free:
Pavel Begunkovf261c162020-08-20 11:34:10 +03003759 /* it's reportedly faster than delegating the null check to kfree() */
Pavel Begunkov252917c2020-07-13 22:59:20 +03003760 if (iovec)
Xiaoguang Wang6f2cc162020-06-18 15:01:56 +08003761 kfree(iovec);
Jens Axboe2b188cc2019-01-07 10:46:33 -07003762 return ret;
3763}
3764
Jens Axboe80a261f2020-09-28 14:23:58 -06003765static int io_renameat_prep(struct io_kiocb *req,
3766 const struct io_uring_sqe *sqe)
3767{
3768 struct io_rename *ren = &req->rename;
3769 const char __user *oldf, *newf;
3770
3771 if (unlikely(req->flags & REQ_F_FIXED_FILE))
3772 return -EBADF;
3773
3774 ren->old_dfd = READ_ONCE(sqe->fd);
3775 oldf = u64_to_user_ptr(READ_ONCE(sqe->addr));
3776 newf = u64_to_user_ptr(READ_ONCE(sqe->addr2));
3777 ren->new_dfd = READ_ONCE(sqe->len);
3778 ren->flags = READ_ONCE(sqe->rename_flags);
3779
3780 ren->oldpath = getname(oldf);
3781 if (IS_ERR(ren->oldpath))
3782 return PTR_ERR(ren->oldpath);
3783
3784 ren->newpath = getname(newf);
3785 if (IS_ERR(ren->newpath)) {
3786 putname(ren->oldpath);
3787 return PTR_ERR(ren->newpath);
3788 }
3789
3790 req->flags |= REQ_F_NEED_CLEANUP;
3791 return 0;
3792}
3793
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003794static int io_renameat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe80a261f2020-09-28 14:23:58 -06003795{
3796 struct io_rename *ren = &req->rename;
3797 int ret;
3798
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003799 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe80a261f2020-09-28 14:23:58 -06003800 return -EAGAIN;
3801
3802 ret = do_renameat2(ren->old_dfd, ren->oldpath, ren->new_dfd,
3803 ren->newpath, ren->flags);
3804
3805 req->flags &= ~REQ_F_NEED_CLEANUP;
3806 if (ret < 0)
3807 req_set_fail_links(req);
3808 io_req_complete(req, ret);
3809 return 0;
3810}
3811
Jens Axboe14a11432020-09-28 14:27:37 -06003812static int io_unlinkat_prep(struct io_kiocb *req,
3813 const struct io_uring_sqe *sqe)
3814{
3815 struct io_unlink *un = &req->unlink;
3816 const char __user *fname;
3817
3818 if (unlikely(req->flags & REQ_F_FIXED_FILE))
3819 return -EBADF;
3820
3821 un->dfd = READ_ONCE(sqe->fd);
3822
3823 un->flags = READ_ONCE(sqe->unlink_flags);
3824 if (un->flags & ~AT_REMOVEDIR)
3825 return -EINVAL;
3826
3827 fname = u64_to_user_ptr(READ_ONCE(sqe->addr));
3828 un->filename = getname(fname);
3829 if (IS_ERR(un->filename))
3830 return PTR_ERR(un->filename);
3831
3832 req->flags |= REQ_F_NEED_CLEANUP;
3833 return 0;
3834}
3835
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003836static int io_unlinkat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe14a11432020-09-28 14:27:37 -06003837{
3838 struct io_unlink *un = &req->unlink;
3839 int ret;
3840
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003841 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe14a11432020-09-28 14:27:37 -06003842 return -EAGAIN;
3843
3844 if (un->flags & AT_REMOVEDIR)
3845 ret = do_rmdir(un->dfd, un->filename);
3846 else
3847 ret = do_unlinkat(un->dfd, un->filename);
3848
3849 req->flags &= ~REQ_F_NEED_CLEANUP;
3850 if (ret < 0)
3851 req_set_fail_links(req);
3852 io_req_complete(req, ret);
3853 return 0;
3854}
3855
Jens Axboe36f4fa62020-09-05 11:14:22 -06003856static int io_shutdown_prep(struct io_kiocb *req,
3857 const struct io_uring_sqe *sqe)
3858{
3859#if defined(CONFIG_NET)
3860 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3861 return -EINVAL;
3862 if (sqe->ioprio || sqe->off || sqe->addr || sqe->rw_flags ||
3863 sqe->buf_index)
3864 return -EINVAL;
3865
3866 req->shutdown.how = READ_ONCE(sqe->len);
3867 return 0;
3868#else
3869 return -EOPNOTSUPP;
3870#endif
3871}
3872
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003873static int io_shutdown(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe36f4fa62020-09-05 11:14:22 -06003874{
3875#if defined(CONFIG_NET)
3876 struct socket *sock;
3877 int ret;
3878
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003879 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe36f4fa62020-09-05 11:14:22 -06003880 return -EAGAIN;
3881
Linus Torvalds48aba792020-12-16 12:44:05 -08003882 sock = sock_from_file(req->file);
Jens Axboe36f4fa62020-09-05 11:14:22 -06003883 if (unlikely(!sock))
Linus Torvalds48aba792020-12-16 12:44:05 -08003884 return -ENOTSOCK;
Jens Axboe36f4fa62020-09-05 11:14:22 -06003885
3886 ret = __sys_shutdown_sock(sock, req->shutdown.how);
Jens Axboea1464682020-12-14 20:57:27 -07003887 if (ret < 0)
3888 req_set_fail_links(req);
Jens Axboe36f4fa62020-09-05 11:14:22 -06003889 io_req_complete(req, ret);
3890 return 0;
3891#else
3892 return -EOPNOTSUPP;
3893#endif
3894}
3895
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003896static int __io_splice_prep(struct io_kiocb *req,
3897 const struct io_uring_sqe *sqe)
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003898{
3899 struct io_splice* sp = &req->splice;
3900 unsigned int valid_flags = SPLICE_F_FD_IN_FIXED | SPLICE_F_ALL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003901
Pavel Begunkov3232dd02020-06-03 18:03:22 +03003902 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
3903 return -EINVAL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003904
3905 sp->file_in = NULL;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003906 sp->len = READ_ONCE(sqe->len);
3907 sp->flags = READ_ONCE(sqe->splice_flags);
3908
3909 if (unlikely(sp->flags & ~valid_flags))
3910 return -EINVAL;
3911
Pavel Begunkov8371adf2020-10-10 18:34:08 +01003912 sp->file_in = io_file_get(NULL, req, READ_ONCE(sqe->splice_fd_in),
3913 (sp->flags & SPLICE_F_FD_IN_FIXED));
3914 if (!sp->file_in)
3915 return -EBADF;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003916 req->flags |= REQ_F_NEED_CLEANUP;
3917
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08003918 if (!S_ISREG(file_inode(sp->file_in)->i_mode)) {
3919 /*
3920 * Splice operation will be punted aync, and here need to
3921 * modify io_wq_work.flags, so initialize io_wq_work firstly.
3922 */
3923 io_req_init_async(req);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003924 req->work.flags |= IO_WQ_WORK_UNBOUND;
Xiaoguang Wang7cdaf582020-06-10 19:41:19 +08003925 }
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003926
3927 return 0;
3928}
3929
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003930static int io_tee_prep(struct io_kiocb *req,
3931 const struct io_uring_sqe *sqe)
3932{
3933 if (READ_ONCE(sqe->splice_off_in) || READ_ONCE(sqe->off))
3934 return -EINVAL;
3935 return __io_splice_prep(req, sqe);
3936}
3937
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003938static int io_tee(struct io_kiocb *req, unsigned int issue_flags)
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003939{
3940 struct io_splice *sp = &req->splice;
3941 struct file *in = sp->file_in;
3942 struct file *out = sp->file_out;
3943 unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED;
3944 long ret = 0;
3945
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003946 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003947 return -EAGAIN;
3948 if (sp->len)
3949 ret = do_tee(in, out, sp->len, flags);
3950
3951 io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED));
3952 req->flags &= ~REQ_F_NEED_CLEANUP;
3953
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003954 if (ret != sp->len)
3955 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003956 io_req_complete(req, ret);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03003957 return 0;
3958}
3959
3960static int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
3961{
3962 struct io_splice* sp = &req->splice;
3963
3964 sp->off_in = READ_ONCE(sqe->splice_off_in);
3965 sp->off_out = READ_ONCE(sqe->off);
3966 return __io_splice_prep(req, sqe);
3967}
3968
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003969static int io_splice(struct io_kiocb *req, unsigned int issue_flags)
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003970{
3971 struct io_splice *sp = &req->splice;
3972 struct file *in = sp->file_in;
3973 struct file *out = sp->file_out;
3974 unsigned int flags = sp->flags & ~SPLICE_F_FD_IN_FIXED;
3975 loff_t *poff_in, *poff_out;
Pavel Begunkovc9687422020-05-04 23:00:54 +03003976 long ret = 0;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003977
Pavel Begunkov45d189c2021-02-10 00:03:07 +00003978 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov2fb3e822020-05-01 17:09:38 +03003979 return -EAGAIN;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003980
3981 poff_in = (sp->off_in == -1) ? NULL : &sp->off_in;
3982 poff_out = (sp->off_out == -1) ? NULL : &sp->off_out;
Pavel Begunkovc9687422020-05-04 23:00:54 +03003983
Jens Axboe948a7742020-05-17 14:21:38 -06003984 if (sp->len)
Pavel Begunkovc9687422020-05-04 23:00:54 +03003985 ret = do_splice(in, poff_in, out, poff_out, sp->len, flags);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003986
3987 io_put_file(req, in, (sp->flags & SPLICE_F_FD_IN_FIXED));
3988 req->flags &= ~REQ_F_NEED_CLEANUP;
3989
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003990 if (ret != sp->len)
3991 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06003992 io_req_complete(req, ret);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03003993 return 0;
3994}
3995
Jens Axboe2b188cc2019-01-07 10:46:33 -07003996/*
3997 * IORING_OP_NOP just posts a completion event, nothing else.
3998 */
Pavel Begunkov889fca72021-02-10 00:03:09 +00003999static int io_nop(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe2b188cc2019-01-07 10:46:33 -07004000{
4001 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2b188cc2019-01-07 10:46:33 -07004002
Jens Axboedef596e2019-01-09 08:59:42 -07004003 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
4004 return -EINVAL;
4005
Pavel Begunkov889fca72021-02-10 00:03:09 +00004006 __io_req_complete(req, issue_flags, 0, 0);
Jens Axboe2b188cc2019-01-07 10:46:33 -07004007 return 0;
4008}
4009
Jens Axboe3529d8c2019-12-19 18:24:38 -07004010static int io_prep_fsync(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Christoph Hellwigc992fe22019-01-11 09:43:02 -07004011{
Jens Axboe6b063142019-01-10 22:13:58 -07004012 struct io_ring_ctx *ctx = req->ctx;
Christoph Hellwigc992fe22019-01-11 09:43:02 -07004013
Jens Axboe09bb8392019-03-13 12:39:28 -06004014 if (!req->file)
4015 return -EBADF;
Christoph Hellwigc992fe22019-01-11 09:43:02 -07004016
Jens Axboe6b063142019-01-10 22:13:58 -07004017 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboedef596e2019-01-09 08:59:42 -07004018 return -EINVAL;
Jens Axboeedafcce2019-01-09 09:16:05 -07004019 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
Christoph Hellwigc992fe22019-01-11 09:43:02 -07004020 return -EINVAL;
4021
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004022 req->sync.flags = READ_ONCE(sqe->fsync_flags);
4023 if (unlikely(req->sync.flags & ~IORING_FSYNC_DATASYNC))
4024 return -EINVAL;
4025
4026 req->sync.off = READ_ONCE(sqe->off);
4027 req->sync.len = READ_ONCE(sqe->len);
Christoph Hellwigc992fe22019-01-11 09:43:02 -07004028 return 0;
4029}
4030
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004031static int io_fsync(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe78912932020-01-14 22:09:06 -07004032{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004033 loff_t end = req->sync.off + req->sync.len;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004034 int ret;
4035
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004036 /* fsync always requires a blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004037 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004038 return -EAGAIN;
4039
Jens Axboe9adbd452019-12-20 08:45:55 -07004040 ret = vfs_fsync_range(req->file, req->sync.off,
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004041 end > 0 ? end : LLONG_MAX,
4042 req->sync.flags & IORING_FSYNC_DATASYNC);
4043 if (ret < 0)
4044 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004045 io_req_complete(req, ret);
Christoph Hellwigc992fe22019-01-11 09:43:02 -07004046 return 0;
4047}
4048
Jens Axboed63d1b52019-12-10 10:38:56 -07004049static int io_fallocate_prep(struct io_kiocb *req,
4050 const struct io_uring_sqe *sqe)
4051{
4052 if (sqe->ioprio || sqe->buf_index || sqe->rw_flags)
4053 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004054 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4055 return -EINVAL;
Jens Axboed63d1b52019-12-10 10:38:56 -07004056
4057 req->sync.off = READ_ONCE(sqe->off);
4058 req->sync.len = READ_ONCE(sqe->addr);
4059 req->sync.mode = READ_ONCE(sqe->len);
4060 return 0;
4061}
4062
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004063static int io_fallocate(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboed63d1b52019-12-10 10:38:56 -07004064{
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004065 int ret;
Jens Axboed63d1b52019-12-10 10:38:56 -07004066
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004067 /* fallocate always requiring blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004068 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004069 return -EAGAIN;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004070 ret = vfs_fallocate(req->file, req->sync.mode, req->sync.off,
4071 req->sync.len);
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004072 if (ret < 0)
4073 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004074 io_req_complete(req, ret);
Jens Axboed63d1b52019-12-10 10:38:56 -07004075 return 0;
4076}
4077
Pavel Begunkovec65fea2020-06-03 18:03:24 +03004078static int __io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe15b71ab2019-12-11 11:20:36 -07004079{
Jens Axboef8748882020-01-08 17:47:02 -07004080 const char __user *fname;
Jens Axboe15b71ab2019-12-11 11:20:36 -07004081 int ret;
4082
Pavel Begunkovec65fea2020-06-03 18:03:24 +03004083 if (unlikely(sqe->ioprio || sqe->buf_index))
Jens Axboe15b71ab2019-12-11 11:20:36 -07004084 return -EINVAL;
Pavel Begunkovec65fea2020-06-03 18:03:24 +03004085 if (unlikely(req->flags & REQ_F_FIXED_FILE))
Jens Axboecf3040c2020-02-06 21:31:40 -07004086 return -EBADF;
Jens Axboe15b71ab2019-12-11 11:20:36 -07004087
Pavel Begunkovec65fea2020-06-03 18:03:24 +03004088 /* open.how should be already initialised */
4089 if (!(req->open.how.flags & O_PATH) && force_o_largefile())
Jens Axboe08a1d26eb2020-04-08 09:20:54 -06004090 req->open.how.flags |= O_LARGEFILE;
Jens Axboe15b71ab2019-12-11 11:20:36 -07004091
Pavel Begunkov25e72d12020-06-03 18:03:23 +03004092 req->open.dfd = READ_ONCE(sqe->fd);
4093 fname = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboef8748882020-01-08 17:47:02 -07004094 req->open.filename = getname(fname);
Jens Axboe15b71ab2019-12-11 11:20:36 -07004095 if (IS_ERR(req->open.filename)) {
4096 ret = PTR_ERR(req->open.filename);
4097 req->open.filename = NULL;
4098 return ret;
4099 }
Jens Axboe4022e7a2020-03-19 19:23:18 -06004100 req->open.nofile = rlimit(RLIMIT_NOFILE);
Pavel Begunkov8fef80b2020-02-07 23:59:53 +03004101 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboe15b71ab2019-12-11 11:20:36 -07004102 return 0;
4103}
4104
Pavel Begunkovec65fea2020-06-03 18:03:24 +03004105static int io_openat_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4106{
4107 u64 flags, mode;
4108
Jens Axboe14587a462020-09-05 11:36:08 -06004109 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe4eb8dde2020-09-18 19:36:24 -06004110 return -EINVAL;
Pavel Begunkovec65fea2020-06-03 18:03:24 +03004111 mode = READ_ONCE(sqe->len);
4112 flags = READ_ONCE(sqe->open_flags);
4113 req->open.how = build_open_how(flags, mode);
4114 return __io_openat_prep(req, sqe);
4115}
4116
Jens Axboecebdb982020-01-08 17:59:24 -07004117static int io_openat2_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4118{
4119 struct open_how __user *how;
Jens Axboecebdb982020-01-08 17:59:24 -07004120 size_t len;
4121 int ret;
4122
Jens Axboe14587a462020-09-05 11:36:08 -06004123 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe4eb8dde2020-09-18 19:36:24 -06004124 return -EINVAL;
Jens Axboecebdb982020-01-08 17:59:24 -07004125 how = u64_to_user_ptr(READ_ONCE(sqe->addr2));
4126 len = READ_ONCE(sqe->len);
Jens Axboecebdb982020-01-08 17:59:24 -07004127 if (len < OPEN_HOW_SIZE_VER0)
4128 return -EINVAL;
4129
4130 ret = copy_struct_from_user(&req->open.how, sizeof(req->open.how), how,
4131 len);
4132 if (ret)
4133 return ret;
4134
Pavel Begunkovec65fea2020-06-03 18:03:24 +03004135 return __io_openat_prep(req, sqe);
Jens Axboecebdb982020-01-08 17:59:24 -07004136}
4137
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004138static int io_openat2(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe15b71ab2019-12-11 11:20:36 -07004139{
4140 struct open_flags op;
Jens Axboe15b71ab2019-12-11 11:20:36 -07004141 struct file *file;
Jens Axboe3a81fd02020-12-10 12:25:36 -07004142 bool nonblock_set;
4143 bool resolve_nonblock;
Jens Axboe15b71ab2019-12-11 11:20:36 -07004144 int ret;
4145
Jens Axboecebdb982020-01-08 17:59:24 -07004146 ret = build_open_flags(&req->open.how, &op);
Jens Axboe15b71ab2019-12-11 11:20:36 -07004147 if (ret)
4148 goto err;
Jens Axboe3a81fd02020-12-10 12:25:36 -07004149 nonblock_set = op.open_flag & O_NONBLOCK;
4150 resolve_nonblock = req->open.how.resolve & RESOLVE_CACHED;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004151 if (issue_flags & IO_URING_F_NONBLOCK) {
Jens Axboe3a81fd02020-12-10 12:25:36 -07004152 /*
4153 * Don't bother trying for O_TRUNC, O_CREAT, or O_TMPFILE open,
4154 * it'll always -EAGAIN
4155 */
4156 if (req->open.how.flags & (O_TRUNC | O_CREAT | O_TMPFILE))
4157 return -EAGAIN;
4158 op.lookup_flags |= LOOKUP_CACHED;
4159 op.open_flag |= O_NONBLOCK;
4160 }
Jens Axboe15b71ab2019-12-11 11:20:36 -07004161
Jens Axboe4022e7a2020-03-19 19:23:18 -06004162 ret = __get_unused_fd_flags(req->open.how.flags, req->open.nofile);
Jens Axboe15b71ab2019-12-11 11:20:36 -07004163 if (ret < 0)
4164 goto err;
4165
4166 file = do_filp_open(req->open.dfd, req->open.filename, &op);
Jens Axboe3a81fd02020-12-10 12:25:36 -07004167 /* only retry if RESOLVE_CACHED wasn't already set by application */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004168 if ((!resolve_nonblock && (issue_flags & IO_URING_F_NONBLOCK)) &&
4169 file == ERR_PTR(-EAGAIN)) {
Jens Axboe3a81fd02020-12-10 12:25:36 -07004170 /*
4171 * We could hang on to this 'fd', but seems like marginal
4172 * gain for something that is now known to be a slower path.
4173 * So just put it, and we'll get a new one when we retry.
4174 */
4175 put_unused_fd(ret);
4176 return -EAGAIN;
4177 }
4178
Jens Axboe15b71ab2019-12-11 11:20:36 -07004179 if (IS_ERR(file)) {
4180 put_unused_fd(ret);
4181 ret = PTR_ERR(file);
4182 } else {
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004183 if ((issue_flags & IO_URING_F_NONBLOCK) && !nonblock_set)
Jens Axboe3a81fd02020-12-10 12:25:36 -07004184 file->f_flags &= ~O_NONBLOCK;
Jens Axboe15b71ab2019-12-11 11:20:36 -07004185 fsnotify_open(file);
4186 fd_install(ret, file);
4187 }
4188err:
4189 putname(req->open.filename);
Pavel Begunkov8fef80b2020-02-07 23:59:53 +03004190 req->flags &= ~REQ_F_NEED_CLEANUP;
Jens Axboe15b71ab2019-12-11 11:20:36 -07004191 if (ret < 0)
4192 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004193 io_req_complete(req, ret);
Jens Axboe15b71ab2019-12-11 11:20:36 -07004194 return 0;
4195}
4196
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004197static int io_openat(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboecebdb982020-01-08 17:59:24 -07004198{
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004199 return io_openat2(req, issue_flags & IO_URING_F_NONBLOCK);
Jens Axboecebdb982020-01-08 17:59:24 -07004200}
4201
Jens Axboe067524e2020-03-02 16:32:28 -07004202static int io_remove_buffers_prep(struct io_kiocb *req,
4203 const struct io_uring_sqe *sqe)
4204{
4205 struct io_provide_buf *p = &req->pbuf;
4206 u64 tmp;
4207
4208 if (sqe->ioprio || sqe->rw_flags || sqe->addr || sqe->len || sqe->off)
4209 return -EINVAL;
4210
4211 tmp = READ_ONCE(sqe->fd);
4212 if (!tmp || tmp > USHRT_MAX)
4213 return -EINVAL;
4214
4215 memset(p, 0, sizeof(*p));
4216 p->nbufs = tmp;
4217 p->bgid = READ_ONCE(sqe->buf_group);
4218 return 0;
4219}
4220
4221static int __io_remove_buffers(struct io_ring_ctx *ctx, struct io_buffer *buf,
4222 int bgid, unsigned nbufs)
4223{
4224 unsigned i = 0;
4225
4226 /* shouldn't happen */
4227 if (!nbufs)
4228 return 0;
4229
4230 /* the head kbuf is the list itself */
4231 while (!list_empty(&buf->list)) {
4232 struct io_buffer *nxt;
4233
4234 nxt = list_first_entry(&buf->list, struct io_buffer, list);
4235 list_del(&nxt->list);
4236 kfree(nxt);
4237 if (++i == nbufs)
4238 return i;
4239 }
4240 i++;
4241 kfree(buf);
4242 idr_remove(&ctx->io_buffer_idr, bgid);
4243
4244 return i;
4245}
4246
Pavel Begunkov889fca72021-02-10 00:03:09 +00004247static int io_remove_buffers(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe067524e2020-03-02 16:32:28 -07004248{
4249 struct io_provide_buf *p = &req->pbuf;
4250 struct io_ring_ctx *ctx = req->ctx;
4251 struct io_buffer *head;
4252 int ret = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004253 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe067524e2020-03-02 16:32:28 -07004254
4255 io_ring_submit_lock(ctx, !force_nonblock);
4256
4257 lockdep_assert_held(&ctx->uring_lock);
4258
4259 ret = -ENOENT;
4260 head = idr_find(&ctx->io_buffer_idr, p->bgid);
4261 if (head)
4262 ret = __io_remove_buffers(ctx, head, p->bgid, p->nbufs);
Jens Axboe067524e2020-03-02 16:32:28 -07004263 if (ret < 0)
4264 req_set_fail_links(req);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00004265
4266 /* need to hold the lock to complete IOPOLL requests */
4267 if (ctx->flags & IORING_SETUP_IOPOLL) {
Pavel Begunkov889fca72021-02-10 00:03:09 +00004268 __io_req_complete(req, issue_flags, ret, 0);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00004269 io_ring_submit_unlock(ctx, !force_nonblock);
4270 } else {
4271 io_ring_submit_unlock(ctx, !force_nonblock);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004272 __io_req_complete(req, issue_flags, ret, 0);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00004273 }
Jens Axboe067524e2020-03-02 16:32:28 -07004274 return 0;
4275}
4276
Jens Axboeddf0322d2020-02-23 16:41:33 -07004277static int io_provide_buffers_prep(struct io_kiocb *req,
4278 const struct io_uring_sqe *sqe)
4279{
4280 struct io_provide_buf *p = &req->pbuf;
4281 u64 tmp;
4282
4283 if (sqe->ioprio || sqe->rw_flags)
4284 return -EINVAL;
4285
4286 tmp = READ_ONCE(sqe->fd);
4287 if (!tmp || tmp > USHRT_MAX)
4288 return -E2BIG;
4289 p->nbufs = tmp;
4290 p->addr = READ_ONCE(sqe->addr);
4291 p->len = READ_ONCE(sqe->len);
4292
Bijan Mottahedehefe68c12020-06-04 18:01:52 -07004293 if (!access_ok(u64_to_user_ptr(p->addr), (p->len * p->nbufs)))
Jens Axboeddf0322d2020-02-23 16:41:33 -07004294 return -EFAULT;
4295
4296 p->bgid = READ_ONCE(sqe->buf_group);
4297 tmp = READ_ONCE(sqe->off);
4298 if (tmp > USHRT_MAX)
4299 return -E2BIG;
4300 p->bid = tmp;
4301 return 0;
4302}
4303
4304static int io_add_buffers(struct io_provide_buf *pbuf, struct io_buffer **head)
4305{
4306 struct io_buffer *buf;
4307 u64 addr = pbuf->addr;
4308 int i, bid = pbuf->bid;
4309
4310 for (i = 0; i < pbuf->nbufs; i++) {
4311 buf = kmalloc(sizeof(*buf), GFP_KERNEL);
4312 if (!buf)
4313 break;
4314
4315 buf->addr = addr;
4316 buf->len = pbuf->len;
4317 buf->bid = bid;
4318 addr += pbuf->len;
4319 bid++;
4320 if (!*head) {
4321 INIT_LIST_HEAD(&buf->list);
4322 *head = buf;
4323 } else {
4324 list_add_tail(&buf->list, &(*head)->list);
4325 }
4326 }
4327
4328 return i ? i : -ENOMEM;
4329}
4330
Pavel Begunkov889fca72021-02-10 00:03:09 +00004331static int io_provide_buffers(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeddf0322d2020-02-23 16:41:33 -07004332{
4333 struct io_provide_buf *p = &req->pbuf;
4334 struct io_ring_ctx *ctx = req->ctx;
4335 struct io_buffer *head, *list;
4336 int ret = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004337 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboeddf0322d2020-02-23 16:41:33 -07004338
4339 io_ring_submit_lock(ctx, !force_nonblock);
4340
4341 lockdep_assert_held(&ctx->uring_lock);
4342
4343 list = head = idr_find(&ctx->io_buffer_idr, p->bgid);
4344
4345 ret = io_add_buffers(p, &head);
4346 if (ret < 0)
4347 goto out;
4348
4349 if (!list) {
4350 ret = idr_alloc(&ctx->io_buffer_idr, head, p->bgid, p->bgid + 1,
4351 GFP_KERNEL);
4352 if (ret < 0) {
Jens Axboe067524e2020-03-02 16:32:28 -07004353 __io_remove_buffers(ctx, head, p->bgid, -1U);
Jens Axboeddf0322d2020-02-23 16:41:33 -07004354 goto out;
4355 }
4356 }
4357out:
Jens Axboeddf0322d2020-02-23 16:41:33 -07004358 if (ret < 0)
4359 req_set_fail_links(req);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00004360
4361 /* need to hold the lock to complete IOPOLL requests */
4362 if (ctx->flags & IORING_SETUP_IOPOLL) {
Pavel Begunkov889fca72021-02-10 00:03:09 +00004363 __io_req_complete(req, issue_flags, ret, 0);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00004364 io_ring_submit_unlock(ctx, !force_nonblock);
4365 } else {
4366 io_ring_submit_unlock(ctx, !force_nonblock);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004367 __io_req_complete(req, issue_flags, ret, 0);
Pavel Begunkov31bff9a2020-12-06 22:22:43 +00004368 }
Jens Axboeddf0322d2020-02-23 16:41:33 -07004369 return 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07004370}
4371
Jens Axboe3e4827b2020-01-08 15:18:09 -07004372static int io_epoll_ctl_prep(struct io_kiocb *req,
4373 const struct io_uring_sqe *sqe)
4374{
4375#if defined(CONFIG_EPOLL)
4376 if (sqe->ioprio || sqe->buf_index)
4377 return -EINVAL;
Jens Axboe6ca56f82020-09-18 16:51:19 -06004378 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL)))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004379 return -EINVAL;
Jens Axboe3e4827b2020-01-08 15:18:09 -07004380
4381 req->epoll.epfd = READ_ONCE(sqe->fd);
4382 req->epoll.op = READ_ONCE(sqe->len);
4383 req->epoll.fd = READ_ONCE(sqe->off);
4384
4385 if (ep_op_has_event(req->epoll.op)) {
4386 struct epoll_event __user *ev;
4387
4388 ev = u64_to_user_ptr(READ_ONCE(sqe->addr));
4389 if (copy_from_user(&req->epoll.event, ev, sizeof(*ev)))
4390 return -EFAULT;
4391 }
4392
4393 return 0;
4394#else
4395 return -EOPNOTSUPP;
4396#endif
4397}
4398
Pavel Begunkov889fca72021-02-10 00:03:09 +00004399static int io_epoll_ctl(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe3e4827b2020-01-08 15:18:09 -07004400{
4401#if defined(CONFIG_EPOLL)
4402 struct io_epoll *ie = &req->epoll;
4403 int ret;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004404 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe3e4827b2020-01-08 15:18:09 -07004405
4406 ret = do_epoll_ctl(ie->epfd, ie->op, ie->fd, &ie->event, force_nonblock);
4407 if (force_nonblock && ret == -EAGAIN)
4408 return -EAGAIN;
4409
4410 if (ret < 0)
4411 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004412 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe3e4827b2020-01-08 15:18:09 -07004413 return 0;
4414#else
4415 return -EOPNOTSUPP;
4416#endif
4417}
4418
Jens Axboec1ca7572019-12-25 22:18:28 -07004419static int io_madvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4420{
4421#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
4422 if (sqe->ioprio || sqe->buf_index || sqe->off)
4423 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004424 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4425 return -EINVAL;
Jens Axboec1ca7572019-12-25 22:18:28 -07004426
4427 req->madvise.addr = READ_ONCE(sqe->addr);
4428 req->madvise.len = READ_ONCE(sqe->len);
4429 req->madvise.advice = READ_ONCE(sqe->fadvise_advice);
4430 return 0;
4431#else
4432 return -EOPNOTSUPP;
4433#endif
4434}
4435
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004436static int io_madvise(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboec1ca7572019-12-25 22:18:28 -07004437{
4438#if defined(CONFIG_ADVISE_SYSCALLS) && defined(CONFIG_MMU)
4439 struct io_madvise *ma = &req->madvise;
4440 int ret;
4441
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004442 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboec1ca7572019-12-25 22:18:28 -07004443 return -EAGAIN;
4444
Minchan Kim0726b012020-10-17 16:14:50 -07004445 ret = do_madvise(current->mm, ma->addr, ma->len, ma->advice);
Jens Axboec1ca7572019-12-25 22:18:28 -07004446 if (ret < 0)
4447 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004448 io_req_complete(req, ret);
Jens Axboec1ca7572019-12-25 22:18:28 -07004449 return 0;
4450#else
4451 return -EOPNOTSUPP;
4452#endif
4453}
4454
Jens Axboe4840e412019-12-25 22:03:45 -07004455static int io_fadvise_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4456{
4457 if (sqe->ioprio || sqe->buf_index || sqe->addr)
4458 return -EINVAL;
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004459 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4460 return -EINVAL;
Jens Axboe4840e412019-12-25 22:03:45 -07004461
4462 req->fadvise.offset = READ_ONCE(sqe->off);
4463 req->fadvise.len = READ_ONCE(sqe->len);
4464 req->fadvise.advice = READ_ONCE(sqe->fadvise_advice);
4465 return 0;
4466}
4467
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004468static int io_fadvise(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe4840e412019-12-25 22:03:45 -07004469{
4470 struct io_fadvise *fa = &req->fadvise;
4471 int ret;
4472
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004473 if (issue_flags & IO_URING_F_NONBLOCK) {
Jens Axboe3e694262020-02-01 09:22:49 -07004474 switch (fa->advice) {
4475 case POSIX_FADV_NORMAL:
4476 case POSIX_FADV_RANDOM:
4477 case POSIX_FADV_SEQUENTIAL:
4478 break;
4479 default:
4480 return -EAGAIN;
4481 }
4482 }
Jens Axboe4840e412019-12-25 22:03:45 -07004483
4484 ret = vfs_fadvise(req->file, fa->offset, fa->len, fa->advice);
4485 if (ret < 0)
4486 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004487 io_req_complete(req, ret);
Jens Axboe4840e412019-12-25 22:03:45 -07004488 return 0;
4489}
4490
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004491static int io_statx_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4492{
Jens Axboe6ca56f82020-09-18 16:51:19 -06004493 if (unlikely(req->ctx->flags & (IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL)))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004494 return -EINVAL;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004495 if (sqe->ioprio || sqe->buf_index)
4496 return -EINVAL;
Pavel Begunkov9c280f92020-04-08 08:58:46 +03004497 if (req->flags & REQ_F_FIXED_FILE)
Jens Axboecf3040c2020-02-06 21:31:40 -07004498 return -EBADF;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004499
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004500 req->statx.dfd = READ_ONCE(sqe->fd);
4501 req->statx.mask = READ_ONCE(sqe->len);
Bijan Mottahedehe62753e2020-05-22 21:31:18 -07004502 req->statx.filename = u64_to_user_ptr(READ_ONCE(sqe->addr));
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004503 req->statx.buffer = u64_to_user_ptr(READ_ONCE(sqe->addr2));
4504 req->statx.flags = READ_ONCE(sqe->statx_flags);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004505
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004506 return 0;
4507}
4508
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004509static int io_statx(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004510{
Bijan Mottahedeh1d9e1282020-05-22 21:31:16 -07004511 struct io_statx *ctx = &req->statx;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004512 int ret;
4513
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004514 if (issue_flags & IO_URING_F_NONBLOCK) {
Jens Axboe5b0bbee2020-04-27 10:41:22 -06004515 /* only need file table for an actual valid fd */
4516 if (ctx->dfd == -1 || ctx->dfd == AT_FDCWD)
4517 req->flags |= REQ_F_NO_FILE_TABLE;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004518 return -EAGAIN;
Jens Axboe5b0bbee2020-04-27 10:41:22 -06004519 }
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004520
Bijan Mottahedehe62753e2020-05-22 21:31:18 -07004521 ret = do_statx(ctx->dfd, ctx->filename, ctx->flags, ctx->mask,
4522 ctx->buffer);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004523
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004524 if (ret < 0)
4525 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004526 io_req_complete(req, ret);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07004527 return 0;
4528}
4529
Jens Axboeb5dba592019-12-11 14:02:38 -07004530static int io_close_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
4531{
Jens Axboe14587a462020-09-05 11:36:08 -06004532 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Pavel Begunkov3232dd02020-06-03 18:03:22 +03004533 return -EINVAL;
Jens Axboeb5dba592019-12-11 14:02:38 -07004534 if (sqe->ioprio || sqe->off || sqe->addr || sqe->len ||
4535 sqe->rw_flags || sqe->buf_index)
4536 return -EINVAL;
Pavel Begunkov9c280f92020-04-08 08:58:46 +03004537 if (req->flags & REQ_F_FIXED_FILE)
Jens Axboecf3040c2020-02-06 21:31:40 -07004538 return -EBADF;
Jens Axboeb5dba592019-12-11 14:02:38 -07004539
4540 req->close.fd = READ_ONCE(sqe->fd);
Jens Axboeb5dba592019-12-11 14:02:38 -07004541 return 0;
4542}
4543
Pavel Begunkov889fca72021-02-10 00:03:09 +00004544static int io_close(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeb5dba592019-12-11 14:02:38 -07004545{
Jens Axboe9eac1902021-01-19 15:50:37 -07004546 struct files_struct *files = current->files;
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004547 struct io_close *close = &req->close;
Jens Axboe9eac1902021-01-19 15:50:37 -07004548 struct fdtable *fdt;
4549 struct file *file;
Jens Axboeb5dba592019-12-11 14:02:38 -07004550 int ret;
4551
Jens Axboe9eac1902021-01-19 15:50:37 -07004552 file = NULL;
4553 ret = -EBADF;
4554 spin_lock(&files->file_lock);
4555 fdt = files_fdtable(files);
4556 if (close->fd >= fdt->max_fds) {
4557 spin_unlock(&files->file_lock);
4558 goto err;
4559 }
4560 file = fdt->fd[close->fd];
4561 if (!file) {
4562 spin_unlock(&files->file_lock);
4563 goto err;
4564 }
4565
4566 if (file->f_op == &io_uring_fops) {
4567 spin_unlock(&files->file_lock);
4568 file = NULL;
4569 goto err;
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004570 }
Jens Axboeb5dba592019-12-11 14:02:38 -07004571
4572 /* if the file has a flush method, be safe and punt to async */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004573 if (file->f_op->flush && (issue_flags & IO_URING_F_NONBLOCK)) {
Jens Axboe9eac1902021-01-19 15:50:37 -07004574 spin_unlock(&files->file_lock);
Pavel Begunkov0bf0eef2020-05-26 20:34:06 +03004575 return -EAGAIN;
Pavel Begunkova2100672020-03-02 23:45:16 +03004576 }
Jens Axboeb5dba592019-12-11 14:02:38 -07004577
Jens Axboe9eac1902021-01-19 15:50:37 -07004578 ret = __close_fd_get_file(close->fd, &file);
4579 spin_unlock(&files->file_lock);
4580 if (ret < 0) {
4581 if (ret == -ENOENT)
4582 ret = -EBADF;
4583 goto err;
4584 }
4585
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004586 /* No ->flush() or already async, safely close from here */
Jens Axboe9eac1902021-01-19 15:50:37 -07004587 ret = filp_close(file, current->files);
4588err:
Pavel Begunkov3af73b22020-06-08 21:08:17 +03004589 if (ret < 0)
4590 req_set_fail_links(req);
Jens Axboe9eac1902021-01-19 15:50:37 -07004591 if (file)
4592 fput(file);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004593 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe1a417f42020-01-31 17:16:48 -07004594 return 0;
Jens Axboeb5dba592019-12-11 14:02:38 -07004595}
4596
Jens Axboe3529d8c2019-12-19 18:24:38 -07004597static int io_prep_sfr(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004598{
4599 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004600
4601 if (!req->file)
4602 return -EBADF;
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004603
4604 if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
4605 return -EINVAL;
4606 if (unlikely(sqe->addr || sqe->ioprio || sqe->buf_index))
4607 return -EINVAL;
4608
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004609 req->sync.off = READ_ONCE(sqe->off);
4610 req->sync.len = READ_ONCE(sqe->len);
4611 req->sync.flags = READ_ONCE(sqe->sync_range_flags);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004612 return 0;
4613}
4614
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004615static int io_sync_file_range(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004616{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004617 int ret;
4618
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004619 /* sync_file_range always requires a blocking context */
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004620 if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkovac45abc2020-06-08 21:08:18 +03004621 return -EAGAIN;
4622
Jens Axboe9adbd452019-12-20 08:45:55 -07004623 ret = sync_file_range(req->file, req->sync.off, req->sync.len,
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07004624 req->sync.flags);
4625 if (ret < 0)
4626 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06004627 io_req_complete(req, ret);
Jens Axboe5d17b4a2019-04-09 14:56:44 -06004628 return 0;
4629}
4630
YueHaibing469956e2020-03-04 15:53:52 +08004631#if defined(CONFIG_NET)
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004632static int io_setup_async_msg(struct io_kiocb *req,
4633 struct io_async_msghdr *kmsg)
4634{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004635 struct io_async_msghdr *async_msg = req->async_data;
4636
4637 if (async_msg)
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004638 return -EAGAIN;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004639 if (io_alloc_async_data(req)) {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004640 kfree(kmsg->free_iov);
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004641 return -ENOMEM;
4642 }
Jens Axboee8c2bc12020-08-15 18:44:09 -07004643 async_msg = req->async_data;
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004644 req->flags |= REQ_F_NEED_CLEANUP;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004645 memcpy(async_msg, kmsg, sizeof(*kmsg));
Pavel Begunkov2a780802021-02-05 00:57:58 +00004646 async_msg->msg.msg_name = &async_msg->addr;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004647 /* if were using fast_iov, set it to the new one */
4648 if (!async_msg->free_iov)
4649 async_msg->msg.msg_iter.iov = async_msg->fast_iov;
4650
Pavel Begunkov02d27d82020-02-28 10:36:36 +03004651 return -EAGAIN;
4652}
4653
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004654static int io_sendmsg_copy_hdr(struct io_kiocb *req,
4655 struct io_async_msghdr *iomsg)
4656{
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004657 iomsg->msg.msg_name = &iomsg->addr;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004658 iomsg->free_iov = iomsg->fast_iov;
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004659 return sendmsg_copy_msghdr(&iomsg->msg, req->sr_msg.umsg,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004660 req->sr_msg.msg_flags, &iomsg->free_iov);
Pavel Begunkov2ae523e2020-07-12 20:41:06 +03004661}
4662
Jens Axboe3529d8c2019-12-19 18:24:38 -07004663static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboeaa1fa282019-04-19 13:38:09 -06004664{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004665 struct io_async_msghdr *async_msg = req->async_data;
Jens Axboee47293f2019-12-20 08:58:21 -07004666 struct io_sr_msg *sr = &req->sr_msg;
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004667 int ret;
Jens Axboe03b12302019-12-02 18:50:25 -07004668
Pavel Begunkovd2b6f482020-06-03 18:03:25 +03004669 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4670 return -EINVAL;
4671
Jens Axboee47293f2019-12-20 08:58:21 -07004672 sr->msg_flags = READ_ONCE(sqe->msg_flags);
Pavel Begunkov270a5942020-07-12 20:41:04 +03004673 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboefddafac2020-01-04 20:19:44 -07004674 sr->len = READ_ONCE(sqe->len);
Jens Axboe3529d8c2019-12-19 18:24:38 -07004675
Jens Axboed8768362020-02-27 14:17:49 -07004676#ifdef CONFIG_COMPAT
4677 if (req->ctx->compat)
4678 sr->msg_flags |= MSG_CMSG_COMPAT;
4679#endif
4680
Jens Axboee8c2bc12020-08-15 18:44:09 -07004681 if (!async_msg || !io_op_defs[req->opcode].needs_async_data)
Jens Axboe3529d8c2019-12-19 18:24:38 -07004682 return 0;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004683 ret = io_sendmsg_copy_hdr(req, async_msg);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004684 if (!ret)
4685 req->flags |= REQ_F_NEED_CLEANUP;
4686 return ret;
Jens Axboe03b12302019-12-02 18:50:25 -07004687}
4688
Pavel Begunkov889fca72021-02-10 00:03:09 +00004689static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004690{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004691 struct io_async_msghdr iomsg, *kmsg;
Jens Axboe03b12302019-12-02 18:50:25 -07004692 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004693 unsigned flags;
Jens Axboe03b12302019-12-02 18:50:25 -07004694 int ret;
4695
Florent Revestdba4a922020-12-04 12:36:04 +01004696 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004697 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004698 return -ENOTSOCK;
Jens Axboe03b12302019-12-02 18:50:25 -07004699
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004700 kmsg = req->async_data;
4701 if (!kmsg) {
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004702 ret = io_sendmsg_copy_hdr(req, &iomsg);
Jens Axboefddafac2020-01-04 20:19:44 -07004703 if (ret)
4704 return ret;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004705 kmsg = &iomsg;
Jens Axboefddafac2020-01-04 20:19:44 -07004706 }
4707
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004708 flags = req->sr_msg.msg_flags;
4709 if (flags & MSG_DONTWAIT)
4710 req->flags |= REQ_F_NOWAIT;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004711 else if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004712 flags |= MSG_DONTWAIT;
4713
4714 ret = __sys_sendmsg_sock(sock, &kmsg->msg, flags);
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004715 if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004716 return io_setup_async_msg(req, kmsg);
4717 if (ret == -ERESTARTSYS)
4718 ret = -EINTR;
4719
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004720 /* fast path, check for non-NULL to avoid function call */
4721 if (kmsg->free_iov)
4722 kfree(kmsg->free_iov);
Jens Axboe03b12302019-12-02 18:50:25 -07004723 req->flags &= ~REQ_F_NEED_CLEANUP;
Jens Axboefddafac2020-01-04 20:19:44 -07004724 if (ret < 0)
4725 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004726 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboefddafac2020-01-04 20:19:44 -07004727 return 0;
Jens Axboefddafac2020-01-04 20:19:44 -07004728}
4729
Pavel Begunkov889fca72021-02-10 00:03:09 +00004730static int io_send(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004731{
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004732 struct io_sr_msg *sr = &req->sr_msg;
4733 struct msghdr msg;
4734 struct iovec iov;
Jens Axboe03b12302019-12-02 18:50:25 -07004735 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004736 unsigned flags;
Jens Axboe03b12302019-12-02 18:50:25 -07004737 int ret;
4738
Florent Revestdba4a922020-12-04 12:36:04 +01004739 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004740 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004741 return -ENOTSOCK;
Jens Axboe03b12302019-12-02 18:50:25 -07004742
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004743 ret = import_single_range(WRITE, sr->buf, sr->len, &iov, &msg.msg_iter);
4744 if (unlikely(ret))
Zheng Bin14db8412020-09-09 20:12:37 +08004745 return ret;
Jens Axboe03b12302019-12-02 18:50:25 -07004746
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004747 msg.msg_name = NULL;
4748 msg.msg_control = NULL;
4749 msg.msg_controllen = 0;
4750 msg.msg_namelen = 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004751
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004752 flags = req->sr_msg.msg_flags;
4753 if (flags & MSG_DONTWAIT)
4754 req->flags |= REQ_F_NOWAIT;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004755 else if (issue_flags & IO_URING_F_NONBLOCK)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004756 flags |= MSG_DONTWAIT;
Jens Axboe03b12302019-12-02 18:50:25 -07004757
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004758 msg.msg_flags = flags;
4759 ret = sock_sendmsg(sock, &msg);
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004760 if ((issue_flags & IO_URING_F_NONBLOCK) && ret == -EAGAIN)
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004761 return -EAGAIN;
4762 if (ret == -ERESTARTSYS)
4763 ret = -EINTR;
Jens Axboe03b12302019-12-02 18:50:25 -07004764
Jens Axboe03b12302019-12-02 18:50:25 -07004765 if (ret < 0)
4766 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004767 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe03b12302019-12-02 18:50:25 -07004768 return 0;
Jens Axboe03b12302019-12-02 18:50:25 -07004769}
4770
Pavel Begunkov1400e692020-07-12 20:41:05 +03004771static int __io_recvmsg_copy_hdr(struct io_kiocb *req,
4772 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004773{
4774 struct io_sr_msg *sr = &req->sr_msg;
4775 struct iovec __user *uiov;
4776 size_t iov_len;
4777 int ret;
4778
Pavel Begunkov1400e692020-07-12 20:41:05 +03004779 ret = __copy_msghdr_from_user(&iomsg->msg, sr->umsg,
4780 &iomsg->uaddr, &uiov, &iov_len);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004781 if (ret)
4782 return ret;
4783
4784 if (req->flags & REQ_F_BUFFER_SELECT) {
4785 if (iov_len > 1)
4786 return -EINVAL;
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004787 if (copy_from_user(iomsg->fast_iov, uiov, sizeof(*uiov)))
Jens Axboe52de1fe2020-02-27 10:15:42 -07004788 return -EFAULT;
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004789 sr->len = iomsg->fast_iov[0].iov_len;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004790 iomsg->free_iov = NULL;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004791 } else {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004792 iomsg->free_iov = iomsg->fast_iov;
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004793 ret = __import_iovec(READ, uiov, iov_len, UIO_FASTIOV,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004794 &iomsg->free_iov, &iomsg->msg.msg_iter,
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004795 false);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004796 if (ret > 0)
4797 ret = 0;
4798 }
4799
4800 return ret;
4801}
4802
4803#ifdef CONFIG_COMPAT
4804static int __io_compat_recvmsg_copy_hdr(struct io_kiocb *req,
Pavel Begunkov1400e692020-07-12 20:41:05 +03004805 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004806{
4807 struct compat_msghdr __user *msg_compat;
4808 struct io_sr_msg *sr = &req->sr_msg;
4809 struct compat_iovec __user *uiov;
4810 compat_uptr_t ptr;
4811 compat_size_t len;
4812 int ret;
4813
Pavel Begunkov270a5942020-07-12 20:41:04 +03004814 msg_compat = (struct compat_msghdr __user *) sr->umsg;
Pavel Begunkov1400e692020-07-12 20:41:05 +03004815 ret = __get_compat_msghdr(&iomsg->msg, msg_compat, &iomsg->uaddr,
Jens Axboe52de1fe2020-02-27 10:15:42 -07004816 &ptr, &len);
4817 if (ret)
4818 return ret;
4819
4820 uiov = compat_ptr(ptr);
4821 if (req->flags & REQ_F_BUFFER_SELECT) {
4822 compat_ssize_t clen;
4823
4824 if (len > 1)
4825 return -EINVAL;
4826 if (!access_ok(uiov, sizeof(*uiov)))
4827 return -EFAULT;
4828 if (__get_user(clen, &uiov->iov_len))
4829 return -EFAULT;
4830 if (clen < 0)
4831 return -EINVAL;
Pavel Begunkov2d280bc2020-11-29 18:33:32 +00004832 sr->len = clen;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004833 iomsg->free_iov = NULL;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004834 } else {
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004835 iomsg->free_iov = iomsg->fast_iov;
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004836 ret = __import_iovec(READ, (struct iovec __user *)uiov, len,
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004837 UIO_FASTIOV, &iomsg->free_iov,
Christoph Hellwig89cd35c2020-09-25 06:51:41 +02004838 &iomsg->msg.msg_iter, true);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004839 if (ret < 0)
4840 return ret;
4841 }
4842
4843 return 0;
4844}
Jens Axboe03b12302019-12-02 18:50:25 -07004845#endif
Jens Axboe52de1fe2020-02-27 10:15:42 -07004846
Pavel Begunkov1400e692020-07-12 20:41:05 +03004847static int io_recvmsg_copy_hdr(struct io_kiocb *req,
4848 struct io_async_msghdr *iomsg)
Jens Axboe52de1fe2020-02-27 10:15:42 -07004849{
Pavel Begunkov1400e692020-07-12 20:41:05 +03004850 iomsg->msg.msg_name = &iomsg->addr;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004851
4852#ifdef CONFIG_COMPAT
4853 if (req->ctx->compat)
Pavel Begunkov1400e692020-07-12 20:41:05 +03004854 return __io_compat_recvmsg_copy_hdr(req, iomsg);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004855#endif
4856
Pavel Begunkov1400e692020-07-12 20:41:05 +03004857 return __io_recvmsg_copy_hdr(req, iomsg);
Jens Axboe52de1fe2020-02-27 10:15:42 -07004858}
4859
Jens Axboebcda7ba2020-02-23 16:42:51 -07004860static struct io_buffer *io_recv_buffer_select(struct io_kiocb *req,
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004861 bool needs_lock)
Jens Axboebcda7ba2020-02-23 16:42:51 -07004862{
4863 struct io_sr_msg *sr = &req->sr_msg;
4864 struct io_buffer *kbuf;
4865
Jens Axboebcda7ba2020-02-23 16:42:51 -07004866 kbuf = io_buffer_select(req, &sr->len, sr->bgid, sr->kbuf, needs_lock);
4867 if (IS_ERR(kbuf))
4868 return kbuf;
4869
4870 sr->kbuf = kbuf;
4871 req->flags |= REQ_F_BUFFER_SELECTED;
Jens Axboebcda7ba2020-02-23 16:42:51 -07004872 return kbuf;
Jens Axboe03b12302019-12-02 18:50:25 -07004873}
4874
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004875static inline unsigned int io_put_recv_kbuf(struct io_kiocb *req)
4876{
4877 return io_put_kbuf(req, req->sr_msg.kbuf);
4878}
4879
Jens Axboe3529d8c2019-12-19 18:24:38 -07004880static int io_recvmsg_prep(struct io_kiocb *req,
4881 const struct io_uring_sqe *sqe)
Jens Axboe03b12302019-12-02 18:50:25 -07004882{
Jens Axboee8c2bc12020-08-15 18:44:09 -07004883 struct io_async_msghdr *async_msg = req->async_data;
Jens Axboee47293f2019-12-20 08:58:21 -07004884 struct io_sr_msg *sr = &req->sr_msg;
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004885 int ret;
Jens Axboe06b76d42019-12-19 14:44:26 -07004886
Pavel Begunkovd2b6f482020-06-03 18:03:25 +03004887 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
4888 return -EINVAL;
4889
Jens Axboe3529d8c2019-12-19 18:24:38 -07004890 sr->msg_flags = READ_ONCE(sqe->msg_flags);
Pavel Begunkov270a5942020-07-12 20:41:04 +03004891 sr->umsg = u64_to_user_ptr(READ_ONCE(sqe->addr));
Jens Axboe0b7b21e2020-01-31 08:34:59 -07004892 sr->len = READ_ONCE(sqe->len);
Jens Axboebcda7ba2020-02-23 16:42:51 -07004893 sr->bgid = READ_ONCE(sqe->buf_group);
Jens Axboe3529d8c2019-12-19 18:24:38 -07004894
Jens Axboed8768362020-02-27 14:17:49 -07004895#ifdef CONFIG_COMPAT
4896 if (req->ctx->compat)
4897 sr->msg_flags |= MSG_CMSG_COMPAT;
4898#endif
4899
Jens Axboee8c2bc12020-08-15 18:44:09 -07004900 if (!async_msg || !io_op_defs[req->opcode].needs_async_data)
Jens Axboe06b76d42019-12-19 14:44:26 -07004901 return 0;
Jens Axboee8c2bc12020-08-15 18:44:09 -07004902 ret = io_recvmsg_copy_hdr(req, async_msg);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004903 if (!ret)
4904 req->flags |= REQ_F_NEED_CLEANUP;
4905 return ret;
Jens Axboe03b12302019-12-02 18:50:25 -07004906}
4907
Pavel Begunkov889fca72021-02-10 00:03:09 +00004908static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe03b12302019-12-02 18:50:25 -07004909{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004910 struct io_async_msghdr iomsg, *kmsg;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004911 struct socket *sock;
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004912 struct io_buffer *kbuf;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004913 unsigned flags;
Jens Axboe52de1fe2020-02-27 10:15:42 -07004914 int ret, cflags = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004915 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004916
Florent Revestdba4a922020-12-04 12:36:04 +01004917 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004918 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004919 return -ENOTSOCK;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004920
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004921 kmsg = req->async_data;
4922 if (!kmsg) {
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004923 ret = io_recvmsg_copy_hdr(req, &iomsg);
4924 if (ret)
Pavel Begunkov681fda82020-07-15 22:20:45 +03004925 return ret;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004926 kmsg = &iomsg;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004927 }
4928
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004929 if (req->flags & REQ_F_BUFFER_SELECT) {
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004930 kbuf = io_recv_buffer_select(req, !force_nonblock);
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004931 if (IS_ERR(kbuf))
4932 return PTR_ERR(kbuf);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004933 kmsg->fast_iov[0].iov_base = u64_to_user_ptr(kbuf->addr);
Pavel Begunkov5476dfe2021-02-05 00:57:59 +00004934 kmsg->fast_iov[0].iov_len = req->sr_msg.len;
4935 iov_iter_init(&kmsg->msg.msg_iter, READ, kmsg->fast_iov,
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004936 1, req->sr_msg.len);
4937 }
4938
4939 flags = req->sr_msg.msg_flags;
4940 if (flags & MSG_DONTWAIT)
4941 req->flags |= REQ_F_NOWAIT;
4942 else if (force_nonblock)
4943 flags |= MSG_DONTWAIT;
4944
4945 ret = __sys_recvmsg_sock(sock, &kmsg->msg, req->sr_msg.umsg,
4946 kmsg->uaddr, flags);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03004947 if (force_nonblock && ret == -EAGAIN)
4948 return io_setup_async_msg(req, kmsg);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004949 if (ret == -ERESTARTSYS)
4950 ret = -EINTR;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03004951
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004952 if (req->flags & REQ_F_BUFFER_SELECTED)
4953 cflags = io_put_recv_kbuf(req);
Pavel Begunkov257e84a2021-02-05 00:58:00 +00004954 /* fast path, check for non-NULL to avoid function call */
4955 if (kmsg->free_iov)
4956 kfree(kmsg->free_iov);
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03004957 req->flags &= ~REQ_F_NEED_CLEANUP;
Jens Axboe4e88d6e2019-12-07 20:59:47 -07004958 if (ret < 0)
4959 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00004960 __io_req_complete(req, issue_flags, ret, cflags);
Jens Axboe0fa03c62019-04-19 13:34:07 -06004961 return 0;
Jens Axboe0fa03c62019-04-19 13:34:07 -06004962}
4963
Pavel Begunkov889fca72021-02-10 00:03:09 +00004964static int io_recv(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboefddafac2020-01-04 20:19:44 -07004965{
Pavel Begunkov6b754c82020-07-16 23:28:00 +03004966 struct io_buffer *kbuf;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004967 struct io_sr_msg *sr = &req->sr_msg;
4968 struct msghdr msg;
4969 void __user *buf = sr->buf;
Jens Axboefddafac2020-01-04 20:19:44 -07004970 struct socket *sock;
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004971 struct iovec iov;
4972 unsigned flags;
Jens Axboebcda7ba2020-02-23 16:42:51 -07004973 int ret, cflags = 0;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00004974 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboefddafac2020-01-04 20:19:44 -07004975
Florent Revestdba4a922020-12-04 12:36:04 +01004976 sock = sock_from_file(req->file);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004977 if (unlikely(!sock))
Florent Revestdba4a922020-12-04 12:36:04 +01004978 return -ENOTSOCK;
Jens Axboefddafac2020-01-04 20:19:44 -07004979
Pavel Begunkovbc02ef32020-07-16 23:28:03 +03004980 if (req->flags & REQ_F_BUFFER_SELECT) {
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03004981 kbuf = io_recv_buffer_select(req, !force_nonblock);
Jens Axboebcda7ba2020-02-23 16:42:51 -07004982 if (IS_ERR(kbuf))
4983 return PTR_ERR(kbuf);
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004984 buf = u64_to_user_ptr(kbuf->addr);
Jens Axboefddafac2020-01-04 20:19:44 -07004985 }
4986
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004987 ret = import_single_range(READ, buf, sr->len, &iov, &msg.msg_iter);
Pavel Begunkov14c32ee2020-07-16 23:28:01 +03004988 if (unlikely(ret))
4989 goto out_free;
Jens Axboefddafac2020-01-04 20:19:44 -07004990
Pavel Begunkov7a7cacb2020-07-16 23:27:59 +03004991 msg.msg_name = NULL;
4992 msg.msg_control = NULL;
4993 msg.msg_controllen = 0;
4994 msg.msg_namelen = 0;
4995 msg.msg_iocb = NULL;
4996 msg.msg_flags = 0;
4997
4998 flags = req->sr_msg.msg_flags;
4999 if (flags & MSG_DONTWAIT)
5000 req->flags |= REQ_F_NOWAIT;
5001 else if (force_nonblock)
5002 flags |= MSG_DONTWAIT;
5003
5004 ret = sock_recvmsg(sock, &msg, flags);
5005 if (force_nonblock && ret == -EAGAIN)
5006 return -EAGAIN;
5007 if (ret == -ERESTARTSYS)
5008 ret = -EINTR;
Pavel Begunkov14c32ee2020-07-16 23:28:01 +03005009out_free:
Pavel Begunkov7fbb1b52020-07-16 23:28:05 +03005010 if (req->flags & REQ_F_BUFFER_SELECTED)
5011 cflags = io_put_recv_kbuf(req);
Jens Axboefddafac2020-01-04 20:19:44 -07005012 if (ret < 0)
5013 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00005014 __io_req_complete(req, issue_flags, ret, cflags);
Jens Axboefddafac2020-01-04 20:19:44 -07005015 return 0;
Jens Axboefddafac2020-01-04 20:19:44 -07005016}
5017
Jens Axboe3529d8c2019-12-19 18:24:38 -07005018static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe17f2fe32019-10-17 14:42:58 -06005019{
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005020 struct io_accept *accept = &req->accept;
5021
Jens Axboe14587a462020-09-05 11:36:08 -06005022 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe17f2fe32019-10-17 14:42:58 -06005023 return -EINVAL;
Hrvoje Zeba8042d6c2019-11-25 14:40:22 -05005024 if (sqe->ioprio || sqe->len || sqe->buf_index)
Jens Axboe17f2fe32019-10-17 14:42:58 -06005025 return -EINVAL;
5026
Jens Axboed55e5f52019-12-11 16:12:15 -07005027 accept->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
5028 accept->addr_len = u64_to_user_ptr(READ_ONCE(sqe->addr2));
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005029 accept->flags = READ_ONCE(sqe->accept_flags);
Jens Axboe09952e32020-03-19 20:16:56 -06005030 accept->nofile = rlimit(RLIMIT_NOFILE);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005031 return 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005032}
Jens Axboe17f2fe32019-10-17 14:42:58 -06005033
Pavel Begunkov889fca72021-02-10 00:03:09 +00005034static int io_accept(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005035{
5036 struct io_accept *accept = &req->accept;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00005037 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03005038 unsigned int file_flags = force_nonblock ? O_NONBLOCK : 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005039 int ret;
5040
Jiufei Xuee697dee2020-06-10 13:41:59 +08005041 if (req->file->f_flags & O_NONBLOCK)
5042 req->flags |= REQ_F_NOWAIT;
5043
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005044 ret = __sys_accept4_file(req->file, file_flags, accept->addr,
Jens Axboe09952e32020-03-19 20:16:56 -06005045 accept->addr_len, accept->flags,
5046 accept->nofile);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005047 if (ret == -EAGAIN && force_nonblock)
Jens Axboe17f2fe32019-10-17 14:42:58 -06005048 return -EAGAIN;
Pavel Begunkovac45abc2020-06-08 21:08:18 +03005049 if (ret < 0) {
5050 if (ret == -ERESTARTSYS)
5051 ret = -EINTR;
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005052 req_set_fail_links(req);
Pavel Begunkovac45abc2020-06-08 21:08:18 +03005053 }
Pavel Begunkov889fca72021-02-10 00:03:09 +00005054 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe17f2fe32019-10-17 14:42:58 -06005055 return 0;
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07005056}
5057
Jens Axboe3529d8c2019-12-19 18:24:38 -07005058static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef499a022019-12-02 16:28:46 -07005059{
Jens Axboe3529d8c2019-12-19 18:24:38 -07005060 struct io_connect *conn = &req->connect;
Jens Axboee8c2bc12020-08-15 18:44:09 -07005061 struct io_async_connect *io = req->async_data;
Jens Axboef499a022019-12-02 16:28:46 -07005062
Jens Axboe14587a462020-09-05 11:36:08 -06005063 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe3fbb51c2019-12-20 08:51:52 -07005064 return -EINVAL;
5065 if (sqe->ioprio || sqe->len || sqe->buf_index || sqe->rw_flags)
5066 return -EINVAL;
5067
Jens Axboe3529d8c2019-12-19 18:24:38 -07005068 conn->addr = u64_to_user_ptr(READ_ONCE(sqe->addr));
5069 conn->addr_len = READ_ONCE(sqe->addr2);
5070
5071 if (!io)
5072 return 0;
5073
5074 return move_addr_to_kernel(conn->addr, conn->addr_len,
Jens Axboee8c2bc12020-08-15 18:44:09 -07005075 &io->address);
Jens Axboef499a022019-12-02 16:28:46 -07005076}
5077
Pavel Begunkov889fca72021-02-10 00:03:09 +00005078static int io_connect(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboef8e85cf2019-11-23 14:24:24 -07005079{
Jens Axboee8c2bc12020-08-15 18:44:09 -07005080 struct io_async_connect __io, *io;
Jens Axboef8e85cf2019-11-23 14:24:24 -07005081 unsigned file_flags;
Jens Axboe3fbb51c2019-12-20 08:51:52 -07005082 int ret;
Pavel Begunkov45d189c2021-02-10 00:03:07 +00005083 bool force_nonblock = issue_flags & IO_URING_F_NONBLOCK;
Jens Axboef8e85cf2019-11-23 14:24:24 -07005084
Jens Axboee8c2bc12020-08-15 18:44:09 -07005085 if (req->async_data) {
5086 io = req->async_data;
Jens Axboef499a022019-12-02 16:28:46 -07005087 } else {
Jens Axboe3529d8c2019-12-19 18:24:38 -07005088 ret = move_addr_to_kernel(req->connect.addr,
5089 req->connect.addr_len,
Jens Axboee8c2bc12020-08-15 18:44:09 -07005090 &__io.address);
Jens Axboef499a022019-12-02 16:28:46 -07005091 if (ret)
5092 goto out;
5093 io = &__io;
5094 }
5095
Jens Axboe3fbb51c2019-12-20 08:51:52 -07005096 file_flags = force_nonblock ? O_NONBLOCK : 0;
5097
Jens Axboee8c2bc12020-08-15 18:44:09 -07005098 ret = __sys_connect_file(req->file, &io->address,
Jens Axboe3fbb51c2019-12-20 08:51:52 -07005099 req->connect.addr_len, file_flags);
Jens Axboe87f80d62019-12-03 11:23:54 -07005100 if ((ret == -EAGAIN || ret == -EINPROGRESS) && force_nonblock) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07005101 if (req->async_data)
Jens Axboeb7bb4f72019-12-15 22:13:43 -07005102 return -EAGAIN;
Jens Axboee8c2bc12020-08-15 18:44:09 -07005103 if (io_alloc_async_data(req)) {
Jens Axboef499a022019-12-02 16:28:46 -07005104 ret = -ENOMEM;
5105 goto out;
5106 }
Jens Axboee8c2bc12020-08-15 18:44:09 -07005107 io = req->async_data;
5108 memcpy(req->async_data, &__io, sizeof(__io));
Jens Axboef8e85cf2019-11-23 14:24:24 -07005109 return -EAGAIN;
Jens Axboef499a022019-12-02 16:28:46 -07005110 }
Jens Axboef8e85cf2019-11-23 14:24:24 -07005111 if (ret == -ERESTARTSYS)
5112 ret = -EINTR;
Jens Axboef499a022019-12-02 16:28:46 -07005113out:
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005114 if (ret < 0)
5115 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00005116 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboef8e85cf2019-11-23 14:24:24 -07005117 return 0;
Jens Axboef8e85cf2019-11-23 14:24:24 -07005118}
YueHaibing469956e2020-03-04 15:53:52 +08005119#else /* !CONFIG_NET */
5120static int io_sendmsg_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
5121{
Jens Axboef8e85cf2019-11-23 14:24:24 -07005122 return -EOPNOTSUPP;
Jens Axboef8e85cf2019-11-23 14:24:24 -07005123}
5124
Pavel Begunkov889fca72021-02-10 00:03:09 +00005125static int io_sendmsg(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005126{
YueHaibing469956e2020-03-04 15:53:52 +08005127 return -EOPNOTSUPP;
5128}
5129
Pavel Begunkov889fca72021-02-10 00:03:09 +00005130static int io_send(struct io_kiocb *req, unsigned int issue_flags)
YueHaibing469956e2020-03-04 15:53:52 +08005131{
5132 return -EOPNOTSUPP;
5133}
5134
5135static int io_recvmsg_prep(struct io_kiocb *req,
5136 const struct io_uring_sqe *sqe)
5137{
5138 return -EOPNOTSUPP;
5139}
5140
Pavel Begunkov889fca72021-02-10 00:03:09 +00005141static int io_recvmsg(struct io_kiocb *req, unsigned int issue_flags)
YueHaibing469956e2020-03-04 15:53:52 +08005142{
5143 return -EOPNOTSUPP;
5144}
5145
Pavel Begunkov889fca72021-02-10 00:03:09 +00005146static int io_recv(struct io_kiocb *req, unsigned int issue_flags)
YueHaibing469956e2020-03-04 15:53:52 +08005147{
5148 return -EOPNOTSUPP;
5149}
5150
5151static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
5152{
5153 return -EOPNOTSUPP;
5154}
5155
Pavel Begunkov889fca72021-02-10 00:03:09 +00005156static int io_accept(struct io_kiocb *req, unsigned int issue_flags)
YueHaibing469956e2020-03-04 15:53:52 +08005157{
5158 return -EOPNOTSUPP;
5159}
5160
5161static int io_connect_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
5162{
5163 return -EOPNOTSUPP;
5164}
5165
Pavel Begunkov889fca72021-02-10 00:03:09 +00005166static int io_connect(struct io_kiocb *req, unsigned int issue_flags)
YueHaibing469956e2020-03-04 15:53:52 +08005167{
5168 return -EOPNOTSUPP;
5169}
5170#endif /* CONFIG_NET */
Jens Axboe2b188cc2019-01-07 10:46:33 -07005171
Jens Axboed7718a92020-02-14 22:23:12 -07005172struct io_poll_table {
5173 struct poll_table_struct pt;
5174 struct io_kiocb *req;
5175 int error;
5176};
5177
Jens Axboed7718a92020-02-14 22:23:12 -07005178static int __io_async_wake(struct io_kiocb *req, struct io_poll_iocb *poll,
5179 __poll_t mask, task_work_func_t func)
5180{
Jens Axboeaa96bf82020-04-03 11:26:26 -06005181 int ret;
Jens Axboed7718a92020-02-14 22:23:12 -07005182
5183 /* for instances that support it check for an event match first: */
5184 if (mask && !(mask & poll->events))
5185 return 0;
5186
5187 trace_io_uring_task_add(req->ctx, req->opcode, req->user_data, mask);
5188
5189 list_del_init(&poll->wait.entry);
5190
Jens Axboed7718a92020-02-14 22:23:12 -07005191 req->result = mask;
Jens Axboe7cbf1722021-02-10 00:03:20 +00005192 req->task_work.func = func;
Jens Axboe6d816e02020-08-11 08:04:14 -06005193 percpu_ref_get(&req->ctx->refs);
5194
Jens Axboed7718a92020-02-14 22:23:12 -07005195 /*
Jens Axboee3aabf92020-05-18 11:04:17 -06005196 * If this fails, then the task is exiting. When a task exits, the
5197 * work gets canceled, so just cancel this request as well instead
5198 * of executing it. We can't safely execute it anyway, as we may not
5199 * have the needed state needed for it anyway.
Jens Axboed7718a92020-02-14 22:23:12 -07005200 */
Jens Axboe355fb9e2020-10-22 20:19:35 -06005201 ret = io_req_task_work_add(req);
Jens Axboeaa96bf82020-04-03 11:26:26 -06005202 if (unlikely(ret)) {
Jens Axboee3aabf92020-05-18 11:04:17 -06005203 WRITE_ONCE(poll->canceled, true);
Pavel Begunkoveab30c42021-01-19 13:32:42 +00005204 io_req_task_work_add_fallback(req, func);
Jens Axboeaa96bf82020-04-03 11:26:26 -06005205 }
Jens Axboed7718a92020-02-14 22:23:12 -07005206 return 1;
5207}
5208
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005209static bool io_poll_rewait(struct io_kiocb *req, struct io_poll_iocb *poll)
5210 __acquires(&req->ctx->completion_lock)
5211{
5212 struct io_ring_ctx *ctx = req->ctx;
5213
5214 if (!req->result && !READ_ONCE(poll->canceled)) {
5215 struct poll_table_struct pt = { ._key = poll->events };
5216
5217 req->result = vfs_poll(req->file, &pt) & poll->events;
5218 }
5219
5220 spin_lock_irq(&ctx->completion_lock);
5221 if (!req->result && !READ_ONCE(poll->canceled)) {
5222 add_wait_queue(poll->head, &poll->wait);
5223 return true;
5224 }
5225
5226 return false;
5227}
5228
Jens Axboed4e7cd32020-08-15 11:44:50 -07005229static struct io_poll_iocb *io_poll_get_double(struct io_kiocb *req)
Jens Axboe18bceab2020-05-15 11:56:54 -06005230{
Jens Axboee8c2bc12020-08-15 18:44:09 -07005231 /* pure poll stashes this in ->async_data, poll driven retry elsewhere */
Jens Axboed4e7cd32020-08-15 11:44:50 -07005232 if (req->opcode == IORING_OP_POLL_ADD)
Jens Axboee8c2bc12020-08-15 18:44:09 -07005233 return req->async_data;
Jens Axboed4e7cd32020-08-15 11:44:50 -07005234 return req->apoll->double_poll;
5235}
5236
5237static struct io_poll_iocb *io_poll_get_single(struct io_kiocb *req)
5238{
5239 if (req->opcode == IORING_OP_POLL_ADD)
5240 return &req->poll;
5241 return &req->apoll->poll;
5242}
5243
5244static void io_poll_remove_double(struct io_kiocb *req)
5245{
5246 struct io_poll_iocb *poll = io_poll_get_double(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06005247
5248 lockdep_assert_held(&req->ctx->completion_lock);
5249
5250 if (poll && poll->head) {
5251 struct wait_queue_head *head = poll->head;
5252
5253 spin_lock(&head->lock);
5254 list_del_init(&poll->wait.entry);
5255 if (poll->wait.private)
5256 refcount_dec(&req->refs);
5257 poll->head = NULL;
5258 spin_unlock(&head->lock);
5259 }
5260}
5261
5262static void io_poll_complete(struct io_kiocb *req, __poll_t mask, int error)
5263{
5264 struct io_ring_ctx *ctx = req->ctx;
5265
Jens Axboed4e7cd32020-08-15 11:44:50 -07005266 io_poll_remove_double(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06005267 req->poll.done = true;
5268 io_cqring_fill_event(req, error ? error : mangle_poll(mask));
5269 io_commit_cqring(ctx);
5270}
5271
Jens Axboe18bceab2020-05-15 11:56:54 -06005272static void io_poll_task_func(struct callback_head *cb)
5273{
5274 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
Jens Axboe6d816e02020-08-11 08:04:14 -06005275 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovdd221f462020-10-18 10:17:42 +01005276 struct io_kiocb *nxt;
Jens Axboe18bceab2020-05-15 11:56:54 -06005277
Pavel Begunkovdd221f462020-10-18 10:17:42 +01005278 if (io_poll_rewait(req, &req->poll)) {
5279 spin_unlock_irq(&ctx->completion_lock);
5280 } else {
5281 hash_del(&req->hash_node);
5282 io_poll_complete(req, req->result, 0);
5283 spin_unlock_irq(&ctx->completion_lock);
5284
5285 nxt = io_put_req_find_next(req);
5286 io_cqring_ev_posted(ctx);
5287 if (nxt)
5288 __io_req_task_submit(nxt);
5289 }
5290
Jens Axboe6d816e02020-08-11 08:04:14 -06005291 percpu_ref_put(&ctx->refs);
Jens Axboe18bceab2020-05-15 11:56:54 -06005292}
5293
5294static int io_poll_double_wake(struct wait_queue_entry *wait, unsigned mode,
5295 int sync, void *key)
5296{
5297 struct io_kiocb *req = wait->private;
Jens Axboed4e7cd32020-08-15 11:44:50 -07005298 struct io_poll_iocb *poll = io_poll_get_single(req);
Jens Axboe18bceab2020-05-15 11:56:54 -06005299 __poll_t mask = key_to_poll(key);
5300
5301 /* for instances that support it check for an event match first: */
5302 if (mask && !(mask & poll->events))
5303 return 0;
5304
Jens Axboe8706e042020-09-28 08:38:54 -06005305 list_del_init(&wait->entry);
5306
Jens Axboe807abcb2020-07-17 17:09:27 -06005307 if (poll && poll->head) {
Jens Axboe18bceab2020-05-15 11:56:54 -06005308 bool done;
5309
Jens Axboe807abcb2020-07-17 17:09:27 -06005310 spin_lock(&poll->head->lock);
5311 done = list_empty(&poll->wait.entry);
Jens Axboe18bceab2020-05-15 11:56:54 -06005312 if (!done)
Jens Axboe807abcb2020-07-17 17:09:27 -06005313 list_del_init(&poll->wait.entry);
Jens Axboed4e7cd32020-08-15 11:44:50 -07005314 /* make sure double remove sees this as being gone */
5315 wait->private = NULL;
Jens Axboe807abcb2020-07-17 17:09:27 -06005316 spin_unlock(&poll->head->lock);
Jens Axboec8b5e262020-10-25 13:53:26 -06005317 if (!done) {
5318 /* use wait func handler, so it matches the rq type */
5319 poll->wait.func(&poll->wait, mode, sync, key);
5320 }
Jens Axboe18bceab2020-05-15 11:56:54 -06005321 }
5322 refcount_dec(&req->refs);
5323 return 1;
5324}
5325
5326static void io_init_poll_iocb(struct io_poll_iocb *poll, __poll_t events,
5327 wait_queue_func_t wake_func)
5328{
5329 poll->head = NULL;
5330 poll->done = false;
5331 poll->canceled = false;
5332 poll->events = events;
5333 INIT_LIST_HEAD(&poll->wait.entry);
5334 init_waitqueue_func_entry(&poll->wait, wake_func);
5335}
5336
5337static void __io_queue_proc(struct io_poll_iocb *poll, struct io_poll_table *pt,
Jens Axboe807abcb2020-07-17 17:09:27 -06005338 struct wait_queue_head *head,
5339 struct io_poll_iocb **poll_ptr)
Jens Axboe18bceab2020-05-15 11:56:54 -06005340{
5341 struct io_kiocb *req = pt->req;
5342
5343 /*
5344 * If poll->head is already set, it's because the file being polled
5345 * uses multiple waitqueues for poll handling (eg one for read, one
5346 * for write). Setup a separate io_poll_iocb if this happens.
5347 */
5348 if (unlikely(poll->head)) {
Pavel Begunkov58852d42020-10-16 20:55:56 +01005349 struct io_poll_iocb *poll_one = poll;
5350
Jens Axboe18bceab2020-05-15 11:56:54 -06005351 /* already have a 2nd entry, fail a third attempt */
Jens Axboe807abcb2020-07-17 17:09:27 -06005352 if (*poll_ptr) {
Jens Axboe18bceab2020-05-15 11:56:54 -06005353 pt->error = -EINVAL;
5354 return;
5355 }
5356 poll = kmalloc(sizeof(*poll), GFP_ATOMIC);
5357 if (!poll) {
5358 pt->error = -ENOMEM;
5359 return;
5360 }
Pavel Begunkov58852d42020-10-16 20:55:56 +01005361 io_init_poll_iocb(poll, poll_one->events, io_poll_double_wake);
Jens Axboe18bceab2020-05-15 11:56:54 -06005362 refcount_inc(&req->refs);
5363 poll->wait.private = req;
Jens Axboe807abcb2020-07-17 17:09:27 -06005364 *poll_ptr = poll;
Jens Axboe18bceab2020-05-15 11:56:54 -06005365 }
5366
5367 pt->error = 0;
5368 poll->head = head;
Jiufei Xuea31eb4a2020-06-17 17:53:56 +08005369
5370 if (poll->events & EPOLLEXCLUSIVE)
5371 add_wait_queue_exclusive(head, &poll->wait);
5372 else
5373 add_wait_queue(head, &poll->wait);
Jens Axboe18bceab2020-05-15 11:56:54 -06005374}
5375
5376static void io_async_queue_proc(struct file *file, struct wait_queue_head *head,
5377 struct poll_table_struct *p)
5378{
5379 struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
Jens Axboe807abcb2020-07-17 17:09:27 -06005380 struct async_poll *apoll = pt->req->apoll;
Jens Axboe18bceab2020-05-15 11:56:54 -06005381
Jens Axboe807abcb2020-07-17 17:09:27 -06005382 __io_queue_proc(&apoll->poll, pt, head, &apoll->double_poll);
Jens Axboe18bceab2020-05-15 11:56:54 -06005383}
5384
Jens Axboed7718a92020-02-14 22:23:12 -07005385static void io_async_task_func(struct callback_head *cb)
5386{
5387 struct io_kiocb *req = container_of(cb, struct io_kiocb, task_work);
5388 struct async_poll *apoll = req->apoll;
5389 struct io_ring_ctx *ctx = req->ctx;
5390
5391 trace_io_uring_task_run(req->ctx, req->opcode, req->user_data);
5392
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005393 if (io_poll_rewait(req, &apoll->poll)) {
Jens Axboed7718a92020-02-14 22:23:12 -07005394 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe6d816e02020-08-11 08:04:14 -06005395 percpu_ref_put(&ctx->refs);
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005396 return;
Jens Axboed7718a92020-02-14 22:23:12 -07005397 }
5398
Jens Axboe31067252020-05-17 17:43:31 -06005399 /* If req is still hashed, it cannot have been canceled. Don't check. */
Pavel Begunkov0be0b0e2020-06-30 15:20:42 +03005400 if (hash_hashed(&req->hash_node))
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005401 hash_del(&req->hash_node);
Jens Axboe2bae0472020-04-13 11:16:34 -06005402
Jens Axboed4e7cd32020-08-15 11:44:50 -07005403 io_poll_remove_double(req);
Jens Axboe74ce6ce2020-04-13 11:09:12 -06005404 spin_unlock_irq(&ctx->completion_lock);
5405
Pavel Begunkov0be0b0e2020-06-30 15:20:42 +03005406 if (!READ_ONCE(apoll->poll.canceled))
5407 __io_req_task_submit(req);
5408 else
5409 __io_req_task_cancel(req, -ECANCELED);
Dan Carpenteraa340842020-07-08 21:47:11 +03005410
Jens Axboe6d816e02020-08-11 08:04:14 -06005411 percpu_ref_put(&ctx->refs);
Jens Axboe807abcb2020-07-17 17:09:27 -06005412 kfree(apoll->double_poll);
Jens Axboe31067252020-05-17 17:43:31 -06005413 kfree(apoll);
Jens Axboed7718a92020-02-14 22:23:12 -07005414}
5415
5416static int io_async_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
5417 void *key)
5418{
5419 struct io_kiocb *req = wait->private;
5420 struct io_poll_iocb *poll = &req->apoll->poll;
5421
5422 trace_io_uring_poll_wake(req->ctx, req->opcode, req->user_data,
5423 key_to_poll(key));
5424
5425 return __io_async_wake(req, poll, key_to_poll(key), io_async_task_func);
5426}
5427
5428static void io_poll_req_insert(struct io_kiocb *req)
5429{
5430 struct io_ring_ctx *ctx = req->ctx;
5431 struct hlist_head *list;
5432
5433 list = &ctx->cancel_hash[hash_long(req->user_data, ctx->cancel_hash_bits)];
5434 hlist_add_head(&req->hash_node, list);
5435}
5436
5437static __poll_t __io_arm_poll_handler(struct io_kiocb *req,
5438 struct io_poll_iocb *poll,
5439 struct io_poll_table *ipt, __poll_t mask,
5440 wait_queue_func_t wake_func)
5441 __acquires(&ctx->completion_lock)
5442{
5443 struct io_ring_ctx *ctx = req->ctx;
5444 bool cancel = false;
5445
Pavel Begunkov4d52f332020-10-18 10:17:43 +01005446 INIT_HLIST_NODE(&req->hash_node);
Jens Axboe18bceab2020-05-15 11:56:54 -06005447 io_init_poll_iocb(poll, mask, wake_func);
Pavel Begunkovb90cd192020-06-21 13:09:52 +03005448 poll->file = req->file;
Jens Axboe18bceab2020-05-15 11:56:54 -06005449 poll->wait.private = req;
Jens Axboed7718a92020-02-14 22:23:12 -07005450
5451 ipt->pt._key = mask;
5452 ipt->req = req;
5453 ipt->error = -EINVAL;
5454
Jens Axboed7718a92020-02-14 22:23:12 -07005455 mask = vfs_poll(req->file, &ipt->pt) & poll->events;
5456
5457 spin_lock_irq(&ctx->completion_lock);
5458 if (likely(poll->head)) {
5459 spin_lock(&poll->head->lock);
5460 if (unlikely(list_empty(&poll->wait.entry))) {
5461 if (ipt->error)
5462 cancel = true;
5463 ipt->error = 0;
5464 mask = 0;
5465 }
5466 if (mask || ipt->error)
5467 list_del_init(&poll->wait.entry);
5468 else if (cancel)
5469 WRITE_ONCE(poll->canceled, true);
5470 else if (!poll->done) /* actually waiting for an event */
5471 io_poll_req_insert(req);
5472 spin_unlock(&poll->head->lock);
5473 }
5474
5475 return mask;
5476}
5477
5478static bool io_arm_poll_handler(struct io_kiocb *req)
5479{
5480 const struct io_op_def *def = &io_op_defs[req->opcode];
5481 struct io_ring_ctx *ctx = req->ctx;
5482 struct async_poll *apoll;
5483 struct io_poll_table ipt;
5484 __poll_t mask, ret;
Jens Axboe9dab14b2020-08-25 12:27:50 -06005485 int rw;
Jens Axboed7718a92020-02-14 22:23:12 -07005486
5487 if (!req->file || !file_can_poll(req->file))
5488 return false;
Pavel Begunkov24c74672020-06-21 13:09:51 +03005489 if (req->flags & REQ_F_POLLED)
Jens Axboed7718a92020-02-14 22:23:12 -07005490 return false;
Jens Axboe9dab14b2020-08-25 12:27:50 -06005491 if (def->pollin)
5492 rw = READ;
5493 else if (def->pollout)
5494 rw = WRITE;
5495 else
5496 return false;
5497 /* if we can't nonblock try, then no point in arming a poll handler */
5498 if (!io_file_supports_async(req->file, rw))
Jens Axboed7718a92020-02-14 22:23:12 -07005499 return false;
5500
5501 apoll = kmalloc(sizeof(*apoll), GFP_ATOMIC);
5502 if (unlikely(!apoll))
5503 return false;
Jens Axboe807abcb2020-07-17 17:09:27 -06005504 apoll->double_poll = NULL;
Jens Axboed7718a92020-02-14 22:23:12 -07005505
5506 req->flags |= REQ_F_POLLED;
Jens Axboed7718a92020-02-14 22:23:12 -07005507 req->apoll = apoll;
Jens Axboed7718a92020-02-14 22:23:12 -07005508
Nathan Chancellor8755d972020-03-02 16:01:19 -07005509 mask = 0;
Jens Axboed7718a92020-02-14 22:23:12 -07005510 if (def->pollin)
Nathan Chancellor8755d972020-03-02 16:01:19 -07005511 mask |= POLLIN | POLLRDNORM;
Jens Axboed7718a92020-02-14 22:23:12 -07005512 if (def->pollout)
5513 mask |= POLLOUT | POLLWRNORM;
Luke Hsiao901341b2020-08-21 21:41:05 -07005514
5515 /* If reading from MSG_ERRQUEUE using recvmsg, ignore POLLIN */
5516 if ((req->opcode == IORING_OP_RECVMSG) &&
5517 (req->sr_msg.msg_flags & MSG_ERRQUEUE))
5518 mask &= ~POLLIN;
5519
Jens Axboed7718a92020-02-14 22:23:12 -07005520 mask |= POLLERR | POLLPRI;
5521
5522 ipt.pt._qproc = io_async_queue_proc;
5523
5524 ret = __io_arm_poll_handler(req, &apoll->poll, &ipt, mask,
5525 io_async_wake);
Jens Axboea36da652020-08-11 09:50:19 -06005526 if (ret || ipt.error) {
Jens Axboed4e7cd32020-08-15 11:44:50 -07005527 io_poll_remove_double(req);
Jens Axboed7718a92020-02-14 22:23:12 -07005528 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe807abcb2020-07-17 17:09:27 -06005529 kfree(apoll->double_poll);
Jens Axboed7718a92020-02-14 22:23:12 -07005530 kfree(apoll);
5531 return false;
5532 }
5533 spin_unlock_irq(&ctx->completion_lock);
5534 trace_io_uring_poll_arm(ctx, req->opcode, req->user_data, mask,
5535 apoll->poll.events);
5536 return true;
5537}
5538
5539static bool __io_poll_remove_one(struct io_kiocb *req,
5540 struct io_poll_iocb *poll)
5541{
Jens Axboeb41e9852020-02-17 09:52:41 -07005542 bool do_complete = false;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005543
5544 spin_lock(&poll->head->lock);
5545 WRITE_ONCE(poll->canceled, true);
Jens Axboe392edb42019-12-09 17:52:20 -07005546 if (!list_empty(&poll->wait.entry)) {
5547 list_del_init(&poll->wait.entry);
Jens Axboeb41e9852020-02-17 09:52:41 -07005548 do_complete = true;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005549 }
5550 spin_unlock(&poll->head->lock);
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005551 hash_del(&req->hash_node);
Jens Axboed7718a92020-02-14 22:23:12 -07005552 return do_complete;
5553}
5554
5555static bool io_poll_remove_one(struct io_kiocb *req)
5556{
5557 bool do_complete;
5558
Jens Axboed4e7cd32020-08-15 11:44:50 -07005559 io_poll_remove_double(req);
5560
Jens Axboed7718a92020-02-14 22:23:12 -07005561 if (req->opcode == IORING_OP_POLL_ADD) {
5562 do_complete = __io_poll_remove_one(req, &req->poll);
5563 } else {
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005564 struct async_poll *apoll = req->apoll;
5565
Jens Axboed7718a92020-02-14 22:23:12 -07005566 /* non-poll requests have submit ref still */
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005567 do_complete = __io_poll_remove_one(req, &apoll->poll);
5568 if (do_complete) {
Jens Axboed7718a92020-02-14 22:23:12 -07005569 io_put_req(req);
Jens Axboe807abcb2020-07-17 17:09:27 -06005570 kfree(apoll->double_poll);
Jens Axboe3bfa5bc2020-05-17 13:54:12 -06005571 kfree(apoll);
5572 }
Xiaoguang Wangb1f573b2020-04-12 14:50:54 +08005573 }
5574
Jens Axboeb41e9852020-02-17 09:52:41 -07005575 if (do_complete) {
5576 io_cqring_fill_event(req, -ECANCELED);
5577 io_commit_cqring(req->ctx);
Jens Axboef254ac02020-08-12 17:33:30 -06005578 req_set_fail_links(req);
Pavel Begunkov216578e2020-10-13 09:44:00 +01005579 io_put_req_deferred(req, 1);
Jens Axboeb41e9852020-02-17 09:52:41 -07005580 }
5581
5582 return do_complete;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005583}
5584
Jens Axboe76e1b642020-09-26 15:05:03 -06005585/*
5586 * Returns true if we found and killed one or more poll requests
5587 */
Pavel Begunkov6b819282020-11-06 13:00:25 +00005588static bool io_poll_remove_all(struct io_ring_ctx *ctx, struct task_struct *tsk,
5589 struct files_struct *files)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005590{
Jens Axboe78076bb2019-12-04 19:56:40 -07005591 struct hlist_node *tmp;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005592 struct io_kiocb *req;
Jens Axboe8e2e1fa2020-04-13 17:05:14 -06005593 int posted = 0, i;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005594
5595 spin_lock_irq(&ctx->completion_lock);
Jens Axboe78076bb2019-12-04 19:56:40 -07005596 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
5597 struct hlist_head *list;
5598
5599 list = &ctx->cancel_hash[i];
Jens Axboef3606e32020-09-22 08:18:24 -06005600 hlist_for_each_entry_safe(req, tmp, list, hash_node) {
Pavel Begunkov6b819282020-11-06 13:00:25 +00005601 if (io_match_task(req, tsk, files))
Jens Axboef3606e32020-09-22 08:18:24 -06005602 posted += io_poll_remove_one(req);
5603 }
Jens Axboe221c5eb2019-01-17 09:41:58 -07005604 }
5605 spin_unlock_irq(&ctx->completion_lock);
Jens Axboeb41e9852020-02-17 09:52:41 -07005606
Jens Axboe8e2e1fa2020-04-13 17:05:14 -06005607 if (posted)
5608 io_cqring_ev_posted(ctx);
Jens Axboe76e1b642020-09-26 15:05:03 -06005609
5610 return posted != 0;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005611}
5612
Jens Axboe47f46762019-11-09 17:43:02 -07005613static int io_poll_cancel(struct io_ring_ctx *ctx, __u64 sqe_addr)
5614{
Jens Axboe78076bb2019-12-04 19:56:40 -07005615 struct hlist_head *list;
Jens Axboe47f46762019-11-09 17:43:02 -07005616 struct io_kiocb *req;
5617
Jens Axboe78076bb2019-12-04 19:56:40 -07005618 list = &ctx->cancel_hash[hash_long(sqe_addr, ctx->cancel_hash_bits)];
5619 hlist_for_each_entry(req, list, hash_node) {
Jens Axboeb41e9852020-02-17 09:52:41 -07005620 if (sqe_addr != req->user_data)
5621 continue;
5622 if (io_poll_remove_one(req))
Jens Axboeeac406c2019-11-14 12:09:58 -07005623 return 0;
Jens Axboeb41e9852020-02-17 09:52:41 -07005624 return -EALREADY;
Jens Axboe47f46762019-11-09 17:43:02 -07005625 }
5626
5627 return -ENOENT;
5628}
5629
Jens Axboe3529d8c2019-12-19 18:24:38 -07005630static int io_poll_remove_prep(struct io_kiocb *req,
5631 const struct io_uring_sqe *sqe)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005632{
Jens Axboe221c5eb2019-01-17 09:41:58 -07005633 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5634 return -EINVAL;
5635 if (sqe->ioprio || sqe->off || sqe->len || sqe->buf_index ||
5636 sqe->poll_events)
5637 return -EINVAL;
5638
Pavel Begunkov018043b2020-10-27 23:17:18 +00005639 req->poll_remove.addr = READ_ONCE(sqe->addr);
Jens Axboe0969e782019-12-17 18:40:57 -07005640 return 0;
5641}
5642
5643/*
5644 * Find a running poll command that matches one specified in sqe->addr,
5645 * and remove it if found.
5646 */
Pavel Begunkov61e98202021-02-10 00:03:08 +00005647static int io_poll_remove(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe0969e782019-12-17 18:40:57 -07005648{
5649 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe0969e782019-12-17 18:40:57 -07005650 int ret;
5651
Jens Axboe221c5eb2019-01-17 09:41:58 -07005652 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov018043b2020-10-27 23:17:18 +00005653 ret = io_poll_cancel(ctx, req->poll_remove.addr);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005654 spin_unlock_irq(&ctx->completion_lock);
5655
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005656 if (ret < 0)
5657 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06005658 io_req_complete(req, ret);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005659 return 0;
5660}
5661
Jens Axboe221c5eb2019-01-17 09:41:58 -07005662static int io_poll_wake(struct wait_queue_entry *wait, unsigned mode, int sync,
5663 void *key)
5664{
Jens Axboec2f2eb72020-02-10 09:07:05 -07005665 struct io_kiocb *req = wait->private;
5666 struct io_poll_iocb *poll = &req->poll;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005667
Jens Axboed7718a92020-02-14 22:23:12 -07005668 return __io_async_wake(req, poll, key_to_poll(key), io_poll_task_func);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005669}
5670
Jens Axboe221c5eb2019-01-17 09:41:58 -07005671static void io_poll_queue_proc(struct file *file, struct wait_queue_head *head,
5672 struct poll_table_struct *p)
5673{
5674 struct io_poll_table *pt = container_of(p, struct io_poll_table, pt);
5675
Jens Axboee8c2bc12020-08-15 18:44:09 -07005676 __io_queue_proc(&pt->req->poll, pt, head, (struct io_poll_iocb **) &pt->req->async_data);
Jens Axboeeac406c2019-11-14 12:09:58 -07005677}
5678
Jens Axboe3529d8c2019-12-19 18:24:38 -07005679static int io_poll_add_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe221c5eb2019-01-17 09:41:58 -07005680{
5681 struct io_poll_iocb *poll = &req->poll;
Jiufei Xue5769a352020-06-17 17:53:55 +08005682 u32 events;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005683
5684 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5685 return -EINVAL;
5686 if (sqe->addr || sqe->ioprio || sqe->off || sqe->len || sqe->buf_index)
5687 return -EINVAL;
5688
Jiufei Xue5769a352020-06-17 17:53:55 +08005689 events = READ_ONCE(sqe->poll32_events);
5690#ifdef __BIG_ENDIAN
5691 events = swahw32(events);
5692#endif
Jiufei Xuea31eb4a2020-06-17 17:53:56 +08005693 poll->events = demangle_poll(events) | EPOLLERR | EPOLLHUP |
5694 (events & EPOLLEXCLUSIVE);
Jens Axboe0969e782019-12-17 18:40:57 -07005695 return 0;
5696}
5697
Pavel Begunkov61e98202021-02-10 00:03:08 +00005698static int io_poll_add(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe0969e782019-12-17 18:40:57 -07005699{
5700 struct io_poll_iocb *poll = &req->poll;
5701 struct io_ring_ctx *ctx = req->ctx;
5702 struct io_poll_table ipt;
Jens Axboe0969e782019-12-17 18:40:57 -07005703 __poll_t mask;
Jens Axboe0969e782019-12-17 18:40:57 -07005704
Jens Axboed7718a92020-02-14 22:23:12 -07005705 ipt.pt._qproc = io_poll_queue_proc;
Jens Axboe36703242019-07-25 10:20:18 -06005706
Jens Axboed7718a92020-02-14 22:23:12 -07005707 mask = __io_arm_poll_handler(req, &req->poll, &ipt, poll->events,
5708 io_poll_wake);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005709
Jens Axboe8c838782019-03-12 15:48:16 -06005710 if (mask) { /* no async, we'd stolen it */
Jens Axboe8c838782019-03-12 15:48:16 -06005711 ipt.error = 0;
Jens Axboeb0dd8a42019-11-18 12:14:54 -07005712 io_poll_complete(req, mask, 0);
Jens Axboe8c838782019-03-12 15:48:16 -06005713 }
Jens Axboe221c5eb2019-01-17 09:41:58 -07005714 spin_unlock_irq(&ctx->completion_lock);
5715
Jens Axboe8c838782019-03-12 15:48:16 -06005716 if (mask) {
5717 io_cqring_ev_posted(ctx);
Pavel Begunkov014db002020-03-03 21:33:12 +03005718 io_put_req(req);
Jens Axboe221c5eb2019-01-17 09:41:58 -07005719 }
Jens Axboe8c838782019-03-12 15:48:16 -06005720 return ipt.error;
Jens Axboe221c5eb2019-01-17 09:41:58 -07005721}
5722
Jens Axboe5262f562019-09-17 12:26:57 -06005723static enum hrtimer_restart io_timeout_fn(struct hrtimer *timer)
5724{
Jens Axboead8a48a2019-11-15 08:49:11 -07005725 struct io_timeout_data *data = container_of(timer,
5726 struct io_timeout_data, timer);
5727 struct io_kiocb *req = data->req;
5728 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe5262f562019-09-17 12:26:57 -06005729 unsigned long flags;
5730
Jens Axboe5262f562019-09-17 12:26:57 -06005731 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkova71976f2020-10-10 18:34:11 +01005732 list_del_init(&req->timeout.list);
Pavel Begunkov01cec8c2020-07-30 18:43:50 +03005733 atomic_set(&req->ctx->cq_timeouts,
5734 atomic_read(&req->ctx->cq_timeouts) + 1);
5735
Jens Axboe78e19bb2019-11-06 15:21:34 -07005736 io_cqring_fill_event(req, -ETIME);
Jens Axboe5262f562019-09-17 12:26:57 -06005737 io_commit_cqring(ctx);
5738 spin_unlock_irqrestore(&ctx->completion_lock, flags);
5739
5740 io_cqring_ev_posted(ctx);
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005741 req_set_fail_links(req);
Jens Axboe5262f562019-09-17 12:26:57 -06005742 io_put_req(req);
5743 return HRTIMER_NORESTART;
5744}
5745
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005746static struct io_kiocb *io_timeout_extract(struct io_ring_ctx *ctx,
5747 __u64 user_data)
Jens Axboe47f46762019-11-09 17:43:02 -07005748{
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005749 struct io_timeout_data *io;
Jens Axboef254ac02020-08-12 17:33:30 -06005750 struct io_kiocb *req;
5751 int ret = -ENOENT;
5752
5753 list_for_each_entry(req, &ctx->timeout_list, timeout.list) {
5754 if (user_data == req->user_data) {
5755 ret = 0;
5756 break;
5757 }
5758 }
5759
5760 if (ret == -ENOENT)
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005761 return ERR_PTR(ret);
Jens Axboef254ac02020-08-12 17:33:30 -06005762
Pavel Begunkovfbd15842020-11-30 19:11:15 +00005763 io = req->async_data;
5764 ret = hrtimer_try_to_cancel(&io->timer);
5765 if (ret == -1)
5766 return ERR_PTR(-EALREADY);
5767 list_del_init(&req->timeout.list);
5768 return req;
5769}
5770
5771static int io_timeout_cancel(struct io_ring_ctx *ctx, __u64 user_data)
5772{
5773 struct io_kiocb *req = io_timeout_extract(ctx, user_data);
5774
5775 if (IS_ERR(req))
5776 return PTR_ERR(req);
5777
5778 req_set_fail_links(req);
5779 io_cqring_fill_event(req, -ECANCELED);
5780 io_put_req_deferred(req, 1);
5781 return 0;
Jens Axboef254ac02020-08-12 17:33:30 -06005782}
5783
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005784static int io_timeout_update(struct io_ring_ctx *ctx, __u64 user_data,
5785 struct timespec64 *ts, enum hrtimer_mode mode)
5786{
5787 struct io_kiocb *req = io_timeout_extract(ctx, user_data);
5788 struct io_timeout_data *data;
5789
5790 if (IS_ERR(req))
5791 return PTR_ERR(req);
5792
5793 req->timeout.off = 0; /* noseq */
5794 data = req->async_data;
5795 list_add_tail(&req->timeout.list, &ctx->timeout_list);
5796 hrtimer_init(&data->timer, CLOCK_MONOTONIC, mode);
5797 data->timer.function = io_timeout_fn;
5798 hrtimer_start(&data->timer, timespec64_to_ktime(*ts), mode);
5799 return 0;
Jens Axboe47f46762019-11-09 17:43:02 -07005800}
5801
Jens Axboe3529d8c2019-12-19 18:24:38 -07005802static int io_timeout_remove_prep(struct io_kiocb *req,
5803 const struct io_uring_sqe *sqe)
Jens Axboeb29472e2019-12-17 18:50:29 -07005804{
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005805 struct io_timeout_rem *tr = &req->timeout_rem;
5806
Jens Axboeb29472e2019-12-17 18:50:29 -07005807 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
5808 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06005809 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
5810 return -EINVAL;
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005811 if (sqe->ioprio || sqe->buf_index || sqe->len)
Jens Axboeb29472e2019-12-17 18:50:29 -07005812 return -EINVAL;
5813
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005814 tr->addr = READ_ONCE(sqe->addr);
5815 tr->flags = READ_ONCE(sqe->timeout_flags);
5816 if (tr->flags & IORING_TIMEOUT_UPDATE) {
5817 if (tr->flags & ~(IORING_TIMEOUT_UPDATE|IORING_TIMEOUT_ABS))
5818 return -EINVAL;
5819 if (get_timespec64(&tr->ts, u64_to_user_ptr(sqe->addr2)))
5820 return -EFAULT;
5821 } else if (tr->flags) {
5822 /* timeout removal doesn't support flags */
5823 return -EINVAL;
5824 }
5825
Jens Axboeb29472e2019-12-17 18:50:29 -07005826 return 0;
5827}
5828
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005829static inline enum hrtimer_mode io_translate_timeout_mode(unsigned int flags)
5830{
5831 return (flags & IORING_TIMEOUT_ABS) ? HRTIMER_MODE_ABS
5832 : HRTIMER_MODE_REL;
5833}
5834
Jens Axboe11365042019-10-16 09:08:32 -06005835/*
5836 * Remove or update an existing timeout command
5837 */
Pavel Begunkov61e98202021-02-10 00:03:08 +00005838static int io_timeout_remove(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe11365042019-10-16 09:08:32 -06005839{
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005840 struct io_timeout_rem *tr = &req->timeout_rem;
Jens Axboe11365042019-10-16 09:08:32 -06005841 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe47f46762019-11-09 17:43:02 -07005842 int ret;
Jens Axboe11365042019-10-16 09:08:32 -06005843
Jens Axboe11365042019-10-16 09:08:32 -06005844 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005845 if (!(req->timeout_rem.flags & IORING_TIMEOUT_UPDATE))
Pavel Begunkov9c8e11b2020-11-30 19:11:16 +00005846 ret = io_timeout_cancel(ctx, tr->addr);
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005847 else
5848 ret = io_timeout_update(ctx, tr->addr, &tr->ts,
5849 io_translate_timeout_mode(tr->flags));
Jens Axboe11365042019-10-16 09:08:32 -06005850
Jens Axboe47f46762019-11-09 17:43:02 -07005851 io_cqring_fill_event(req, ret);
Jens Axboe11365042019-10-16 09:08:32 -06005852 io_commit_cqring(ctx);
5853 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe5262f562019-09-17 12:26:57 -06005854 io_cqring_ev_posted(ctx);
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005855 if (ret < 0)
5856 req_set_fail_links(req);
Jackie Liuec9c02a2019-11-08 23:50:36 +08005857 io_put_req(req);
Jens Axboe11365042019-10-16 09:08:32 -06005858 return 0;
Jens Axboe5262f562019-09-17 12:26:57 -06005859}
5860
Jens Axboe3529d8c2019-12-19 18:24:38 -07005861static int io_timeout_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe,
Jens Axboe2d283902019-12-04 11:08:05 -07005862 bool is_timeout_link)
Jens Axboe5262f562019-09-17 12:26:57 -06005863{
Jens Axboead8a48a2019-11-15 08:49:11 -07005864 struct io_timeout_data *data;
Jens Axboea41525a2019-10-15 16:48:15 -06005865 unsigned flags;
Pavel Begunkov56080b02020-05-26 20:34:04 +03005866 u32 off = READ_ONCE(sqe->off);
Jens Axboe5262f562019-09-17 12:26:57 -06005867
Jens Axboead8a48a2019-11-15 08:49:11 -07005868 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboe5262f562019-09-17 12:26:57 -06005869 return -EINVAL;
Jens Axboead8a48a2019-11-15 08:49:11 -07005870 if (sqe->ioprio || sqe->buf_index || sqe->len != 1)
Jens Axboea41525a2019-10-15 16:48:15 -06005871 return -EINVAL;
Pavel Begunkov56080b02020-05-26 20:34:04 +03005872 if (off && is_timeout_link)
Jens Axboe2d283902019-12-04 11:08:05 -07005873 return -EINVAL;
Jens Axboea41525a2019-10-15 16:48:15 -06005874 flags = READ_ONCE(sqe->timeout_flags);
5875 if (flags & ~IORING_TIMEOUT_ABS)
Jens Axboe5262f562019-09-17 12:26:57 -06005876 return -EINVAL;
Arnd Bergmannbdf20072019-10-01 09:53:29 -06005877
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005878 req->timeout.off = off;
Jens Axboe26a61672019-12-20 09:02:01 -07005879
Jens Axboee8c2bc12020-08-15 18:44:09 -07005880 if (!req->async_data && io_alloc_async_data(req))
Jens Axboe26a61672019-12-20 09:02:01 -07005881 return -ENOMEM;
5882
Jens Axboee8c2bc12020-08-15 18:44:09 -07005883 data = req->async_data;
Jens Axboead8a48a2019-11-15 08:49:11 -07005884 data->req = req;
Jens Axboead8a48a2019-11-15 08:49:11 -07005885
5886 if (get_timespec64(&data->ts, u64_to_user_ptr(sqe->addr)))
Jens Axboe5262f562019-09-17 12:26:57 -06005887 return -EFAULT;
5888
Pavel Begunkov8662dae2021-01-19 13:32:44 +00005889 data->mode = io_translate_timeout_mode(flags);
Jens Axboead8a48a2019-11-15 08:49:11 -07005890 hrtimer_init(&data->timer, CLOCK_MONOTONIC, data->mode);
5891 return 0;
5892}
5893
Pavel Begunkov61e98202021-02-10 00:03:08 +00005894static int io_timeout(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboead8a48a2019-11-15 08:49:11 -07005895{
Jens Axboead8a48a2019-11-15 08:49:11 -07005896 struct io_ring_ctx *ctx = req->ctx;
Jens Axboee8c2bc12020-08-15 18:44:09 -07005897 struct io_timeout_data *data = req->async_data;
Jens Axboead8a48a2019-11-15 08:49:11 -07005898 struct list_head *entry;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005899 u32 tail, off = req->timeout.off;
Jens Axboead8a48a2019-11-15 08:49:11 -07005900
Pavel Begunkov733f5c92020-05-26 20:34:03 +03005901 spin_lock_irq(&ctx->completion_lock);
Jens Axboe93bd25b2019-11-11 23:34:31 -07005902
Jens Axboe5262f562019-09-17 12:26:57 -06005903 /*
5904 * sqe->off holds how many events that need to occur for this
Jens Axboe93bd25b2019-11-11 23:34:31 -07005905 * timeout event to be satisfied. If it isn't set, then this is
5906 * a pure timeout request, sequence isn't used.
Jens Axboe5262f562019-09-17 12:26:57 -06005907 */
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03005908 if (io_is_timeout_noseq(req)) {
Jens Axboe93bd25b2019-11-11 23:34:31 -07005909 entry = ctx->timeout_list.prev;
5910 goto add;
5911 }
Jens Axboe5262f562019-09-17 12:26:57 -06005912
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005913 tail = ctx->cached_cq_tail - atomic_read(&ctx->cq_timeouts);
5914 req->timeout.target_seq = tail + off;
Jens Axboe5262f562019-09-17 12:26:57 -06005915
Marcelo Diop-Gonzalezf0105052021-01-15 11:54:40 -05005916 /* Update the last seq here in case io_flush_timeouts() hasn't.
5917 * This is safe because ->completion_lock is held, and submissions
5918 * and completions are never mixed in the same ->completion_lock section.
5919 */
5920 ctx->cq_last_tm_flush = tail;
5921
Jens Axboe5262f562019-09-17 12:26:57 -06005922 /*
5923 * Insertion sort, ensuring the first entry in the list is always
5924 * the one we need first.
5925 */
Jens Axboe5262f562019-09-17 12:26:57 -06005926 list_for_each_prev(entry, &ctx->timeout_list) {
Pavel Begunkov135fcde2020-07-13 23:37:12 +03005927 struct io_kiocb *nxt = list_entry(entry, struct io_kiocb,
5928 timeout.list);
Jens Axboe5262f562019-09-17 12:26:57 -06005929
Pavel Begunkov8eb7e2d2020-06-29 13:13:02 +03005930 if (io_is_timeout_noseq(nxt))
Jens Axboe93bd25b2019-11-11 23:34:31 -07005931 continue;
Pavel Begunkovbfe68a22020-05-30 14:54:18 +03005932 /* nxt.seq is behind @tail, otherwise would've been completed */
5933 if (off >= nxt->timeout.target_seq - tail)
Jens Axboe5262f562019-09-17 12:26:57 -06005934 break;
5935 }
Jens Axboe93bd25b2019-11-11 23:34:31 -07005936add:
Pavel Begunkov135fcde2020-07-13 23:37:12 +03005937 list_add(&req->timeout.list, entry);
Jens Axboead8a48a2019-11-15 08:49:11 -07005938 data->timer.function = io_timeout_fn;
5939 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts), data->mode);
Jens Axboe842f9612019-10-29 12:34:10 -06005940 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe5262f562019-09-17 12:26:57 -06005941 return 0;
5942}
5943
Jens Axboe62755e32019-10-28 21:49:21 -06005944static bool io_cancel_cb(struct io_wq_work *work, void *data)
Jens Axboede0617e2019-04-06 21:51:27 -06005945{
Jens Axboe62755e32019-10-28 21:49:21 -06005946 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Jens Axboede0617e2019-04-06 21:51:27 -06005947
Jens Axboe62755e32019-10-28 21:49:21 -06005948 return req->user_data == (unsigned long) data;
5949}
5950
Jens Axboee977d6d2019-11-05 12:39:45 -07005951static int io_async_cancel_one(struct io_ring_ctx *ctx, void *sqe_addr)
Jens Axboe62755e32019-10-28 21:49:21 -06005952{
Jens Axboe62755e32019-10-28 21:49:21 -06005953 enum io_wq_cancel cancel_ret;
Jens Axboe62755e32019-10-28 21:49:21 -06005954 int ret = 0;
5955
Pavel Begunkov4f26bda2020-06-15 10:24:03 +03005956 cancel_ret = io_wq_cancel_cb(ctx->io_wq, io_cancel_cb, sqe_addr, false);
Jens Axboe62755e32019-10-28 21:49:21 -06005957 switch (cancel_ret) {
5958 case IO_WQ_CANCEL_OK:
5959 ret = 0;
5960 break;
5961 case IO_WQ_CANCEL_RUNNING:
5962 ret = -EALREADY;
5963 break;
5964 case IO_WQ_CANCEL_NOTFOUND:
5965 ret = -ENOENT;
5966 break;
5967 }
5968
Jens Axboee977d6d2019-11-05 12:39:45 -07005969 return ret;
5970}
5971
Jens Axboe47f46762019-11-09 17:43:02 -07005972static void io_async_find_and_cancel(struct io_ring_ctx *ctx,
5973 struct io_kiocb *req, __u64 sqe_addr,
Pavel Begunkov014db002020-03-03 21:33:12 +03005974 int success_ret)
Jens Axboe47f46762019-11-09 17:43:02 -07005975{
5976 unsigned long flags;
5977 int ret;
5978
5979 ret = io_async_cancel_one(ctx, (void *) (unsigned long) sqe_addr);
5980 if (ret != -ENOENT) {
5981 spin_lock_irqsave(&ctx->completion_lock, flags);
5982 goto done;
5983 }
5984
5985 spin_lock_irqsave(&ctx->completion_lock, flags);
5986 ret = io_timeout_cancel(ctx, sqe_addr);
5987 if (ret != -ENOENT)
5988 goto done;
5989 ret = io_poll_cancel(ctx, sqe_addr);
5990done:
Jens Axboeb0dd8a42019-11-18 12:14:54 -07005991 if (!ret)
5992 ret = success_ret;
Jens Axboe47f46762019-11-09 17:43:02 -07005993 io_cqring_fill_event(req, ret);
5994 io_commit_cqring(ctx);
5995 spin_unlock_irqrestore(&ctx->completion_lock, flags);
5996 io_cqring_ev_posted(ctx);
5997
Jens Axboe4e88d6e2019-12-07 20:59:47 -07005998 if (ret < 0)
5999 req_set_fail_links(req);
Pavel Begunkov014db002020-03-03 21:33:12 +03006000 io_put_req(req);
Jens Axboe47f46762019-11-09 17:43:02 -07006001}
6002
Jens Axboe3529d8c2019-12-19 18:24:38 -07006003static int io_async_cancel_prep(struct io_kiocb *req,
6004 const struct io_uring_sqe *sqe)
Jens Axboee977d6d2019-11-05 12:39:45 -07006005{
Jens Axboefbf23842019-12-17 18:45:56 -07006006 if (unlikely(req->ctx->flags & IORING_SETUP_IOPOLL))
Jens Axboee977d6d2019-11-05 12:39:45 -07006007 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06006008 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
6009 return -EINVAL;
6010 if (sqe->ioprio || sqe->off || sqe->len || sqe->cancel_flags)
Jens Axboee977d6d2019-11-05 12:39:45 -07006011 return -EINVAL;
6012
Jens Axboefbf23842019-12-17 18:45:56 -07006013 req->cancel.addr = READ_ONCE(sqe->addr);
6014 return 0;
6015}
6016
Pavel Begunkov61e98202021-02-10 00:03:08 +00006017static int io_async_cancel(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboefbf23842019-12-17 18:45:56 -07006018{
6019 struct io_ring_ctx *ctx = req->ctx;
Jens Axboefbf23842019-12-17 18:45:56 -07006020
Pavel Begunkov014db002020-03-03 21:33:12 +03006021 io_async_find_and_cancel(ctx, req, req->cancel.addr, 0);
Jens Axboe62755e32019-10-28 21:49:21 -06006022 return 0;
6023}
6024
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006025static int io_rsrc_update_prep(struct io_kiocb *req,
Jens Axboe05f3fb32019-12-09 11:22:50 -07006026 const struct io_uring_sqe *sqe)
6027{
Jens Axboe6ca56f82020-09-18 16:51:19 -06006028 if (unlikely(req->ctx->flags & IORING_SETUP_SQPOLL))
6029 return -EINVAL;
Daniele Albano61710e42020-07-18 14:15:16 -06006030 if (unlikely(req->flags & (REQ_F_FIXED_FILE | REQ_F_BUFFER_SELECT)))
6031 return -EINVAL;
6032 if (sqe->ioprio || sqe->rw_flags)
Jens Axboe05f3fb32019-12-09 11:22:50 -07006033 return -EINVAL;
6034
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006035 req->rsrc_update.offset = READ_ONCE(sqe->off);
6036 req->rsrc_update.nr_args = READ_ONCE(sqe->len);
6037 if (!req->rsrc_update.nr_args)
Jens Axboe05f3fb32019-12-09 11:22:50 -07006038 return -EINVAL;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006039 req->rsrc_update.arg = READ_ONCE(sqe->addr);
Jens Axboe05f3fb32019-12-09 11:22:50 -07006040 return 0;
6041}
6042
Pavel Begunkov889fca72021-02-10 00:03:09 +00006043static int io_files_update(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboe05f3fb32019-12-09 11:22:50 -07006044{
6045 struct io_ring_ctx *ctx = req->ctx;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006046 struct io_uring_rsrc_update up;
Jens Axboe05f3fb32019-12-09 11:22:50 -07006047 int ret;
6048
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006049 if (issue_flags & IO_URING_F_NONBLOCK)
Jens Axboe05f3fb32019-12-09 11:22:50 -07006050 return -EAGAIN;
Jens Axboe05f3fb32019-12-09 11:22:50 -07006051
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006052 up.offset = req->rsrc_update.offset;
6053 up.data = req->rsrc_update.arg;
Jens Axboe05f3fb32019-12-09 11:22:50 -07006054
6055 mutex_lock(&ctx->uring_lock);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006056 ret = __io_sqe_files_update(ctx, &up, req->rsrc_update.nr_args);
Jens Axboe05f3fb32019-12-09 11:22:50 -07006057 mutex_unlock(&ctx->uring_lock);
6058
6059 if (ret < 0)
6060 req_set_fail_links(req);
Pavel Begunkov889fca72021-02-10 00:03:09 +00006061 __io_req_complete(req, issue_flags, ret, 0);
Jens Axboe05f3fb32019-12-09 11:22:50 -07006062 return 0;
6063}
6064
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006065static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboef67676d2019-12-02 11:03:47 -07006066{
Jens Axboed625c6e2019-12-17 19:53:05 -07006067 switch (req->opcode) {
Jens Axboee7815732019-12-17 19:45:06 -07006068 case IORING_OP_NOP:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006069 return 0;
Jens Axboef67676d2019-12-02 11:03:47 -07006070 case IORING_OP_READV:
6071 case IORING_OP_READ_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07006072 case IORING_OP_READ:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006073 return io_read_prep(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07006074 case IORING_OP_WRITEV:
6075 case IORING_OP_WRITE_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07006076 case IORING_OP_WRITE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006077 return io_write_prep(req, sqe);
Jens Axboe0969e782019-12-17 18:40:57 -07006078 case IORING_OP_POLL_ADD:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006079 return io_poll_add_prep(req, sqe);
Jens Axboe0969e782019-12-17 18:40:57 -07006080 case IORING_OP_POLL_REMOVE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006081 return io_poll_remove_prep(req, sqe);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07006082 case IORING_OP_FSYNC:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006083 return io_prep_fsync(req, sqe);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07006084 case IORING_OP_SYNC_FILE_RANGE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006085 return io_prep_sfr(req, sqe);
Jens Axboe03b12302019-12-02 18:50:25 -07006086 case IORING_OP_SENDMSG:
Jens Axboefddafac2020-01-04 20:19:44 -07006087 case IORING_OP_SEND:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006088 return io_sendmsg_prep(req, sqe);
Jens Axboe03b12302019-12-02 18:50:25 -07006089 case IORING_OP_RECVMSG:
Jens Axboefddafac2020-01-04 20:19:44 -07006090 case IORING_OP_RECV:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006091 return io_recvmsg_prep(req, sqe);
Jens Axboef499a022019-12-02 16:28:46 -07006092 case IORING_OP_CONNECT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006093 return io_connect_prep(req, sqe);
Jens Axboe2d283902019-12-04 11:08:05 -07006094 case IORING_OP_TIMEOUT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006095 return io_timeout_prep(req, sqe, false);
Jens Axboeb29472e2019-12-17 18:50:29 -07006096 case IORING_OP_TIMEOUT_REMOVE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006097 return io_timeout_remove_prep(req, sqe);
Jens Axboefbf23842019-12-17 18:45:56 -07006098 case IORING_OP_ASYNC_CANCEL:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006099 return io_async_cancel_prep(req, sqe);
Jens Axboe2d283902019-12-04 11:08:05 -07006100 case IORING_OP_LINK_TIMEOUT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006101 return io_timeout_prep(req, sqe, true);
Jens Axboe8ed8d3c2019-12-16 11:55:28 -07006102 case IORING_OP_ACCEPT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006103 return io_accept_prep(req, sqe);
Jens Axboed63d1b52019-12-10 10:38:56 -07006104 case IORING_OP_FALLOCATE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006105 return io_fallocate_prep(req, sqe);
Jens Axboe15b71ab2019-12-11 11:20:36 -07006106 case IORING_OP_OPENAT:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006107 return io_openat_prep(req, sqe);
Jens Axboeb5dba592019-12-11 14:02:38 -07006108 case IORING_OP_CLOSE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006109 return io_close_prep(req, sqe);
Jens Axboe05f3fb32019-12-09 11:22:50 -07006110 case IORING_OP_FILES_UPDATE:
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006111 return io_rsrc_update_prep(req, sqe);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07006112 case IORING_OP_STATX:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006113 return io_statx_prep(req, sqe);
Jens Axboe4840e412019-12-25 22:03:45 -07006114 case IORING_OP_FADVISE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006115 return io_fadvise_prep(req, sqe);
Jens Axboec1ca7572019-12-25 22:18:28 -07006116 case IORING_OP_MADVISE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006117 return io_madvise_prep(req, sqe);
Jens Axboecebdb982020-01-08 17:59:24 -07006118 case IORING_OP_OPENAT2:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006119 return io_openat2_prep(req, sqe);
Jens Axboe3e4827b2020-01-08 15:18:09 -07006120 case IORING_OP_EPOLL_CTL:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006121 return io_epoll_ctl_prep(req, sqe);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03006122 case IORING_OP_SPLICE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006123 return io_splice_prep(req, sqe);
Jens Axboeddf0322d2020-02-23 16:41:33 -07006124 case IORING_OP_PROVIDE_BUFFERS:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006125 return io_provide_buffers_prep(req, sqe);
Jens Axboe067524e2020-03-02 16:32:28 -07006126 case IORING_OP_REMOVE_BUFFERS:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006127 return io_remove_buffers_prep(req, sqe);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03006128 case IORING_OP_TEE:
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006129 return io_tee_prep(req, sqe);
Jens Axboe36f4fa62020-09-05 11:14:22 -06006130 case IORING_OP_SHUTDOWN:
6131 return io_shutdown_prep(req, sqe);
Jens Axboe80a261f2020-09-28 14:23:58 -06006132 case IORING_OP_RENAMEAT:
6133 return io_renameat_prep(req, sqe);
Jens Axboe14a11432020-09-28 14:27:37 -06006134 case IORING_OP_UNLINKAT:
6135 return io_unlinkat_prep(req, sqe);
Jens Axboef67676d2019-12-02 11:03:47 -07006136 }
6137
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006138 printk_once(KERN_WARNING "io_uring: unhandled opcode %d\n",
6139 req->opcode);
6140 return-EINVAL;
6141}
6142
Jens Axboedef596e2019-01-09 08:59:42 -07006143static int io_req_defer_prep(struct io_kiocb *req,
6144 const struct io_uring_sqe *sqe)
Jens Axboedef596e2019-01-09 08:59:42 -07006145{
Jens Axboedef596e2019-01-09 08:59:42 -07006146 if (!sqe)
Jens Axboe2b188cc2019-01-07 10:46:33 -07006147 return 0;
Jens Axboee8c2bc12020-08-15 18:44:09 -07006148 if (io_alloc_async_data(req))
Jens Axboeb76da702019-11-20 13:05:32 -07006149 return -EAGAIN;
Pavel Begunkovbfe76552020-09-30 22:57:55 +03006150 return io_req_prep(req, sqe);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006151}
6152
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03006153static u32 io_get_sequence(struct io_kiocb *req)
6154{
6155 struct io_kiocb *pos;
6156 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006157 u32 total_submitted, nr_reqs = 0;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03006158
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006159 io_for_each_link(pos, req)
6160 nr_reqs++;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03006161
6162 total_submitted = ctx->cached_sq_head - ctx->cached_sq_dropped;
6163 return total_submitted - nr_reqs;
6164}
6165
Jens Axboe3529d8c2019-12-19 18:24:38 -07006166static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jens Axboe2b188cc2019-01-07 10:46:33 -07006167{
6168 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03006169 struct io_defer_entry *de;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006170 int ret;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03006171 u32 seq;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006172
6173 /* Still need defer if there is pending req in defer list. */
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03006174 if (likely(list_empty_careful(&ctx->defer_list) &&
6175 !(req->flags & REQ_F_IO_DRAIN)))
6176 return 0;
6177
6178 seq = io_get_sequence(req);
6179 /* Still a chance to pass the sequence check */
6180 if (!req_need_defer(req, seq) && list_empty_careful(&ctx->defer_list))
Jens Axboe2b188cc2019-01-07 10:46:33 -07006181 return 0;
6182
Jens Axboee8c2bc12020-08-15 18:44:09 -07006183 if (!req->async_data) {
Pavel Begunkov650b5482020-05-17 14:02:11 +03006184 ret = io_req_defer_prep(req, sqe);
Pavel Begunkov327d6d92020-07-15 12:46:51 +03006185 if (ret)
Pavel Begunkov650b5482020-05-17 14:02:11 +03006186 return ret;
6187 }
Pavel Begunkovcbdcb432020-06-29 19:18:43 +03006188 io_prep_async_link(req);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03006189 de = kmalloc(sizeof(*de), GFP_KERNEL);
6190 if (!de)
6191 return -ENOMEM;
Jens Axboe31b51512019-01-18 22:56:34 -07006192
6193 spin_lock_irq(&ctx->completion_lock);
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03006194 if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) {
Jens Axboe31b51512019-01-18 22:56:34 -07006195 spin_unlock_irq(&ctx->completion_lock);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03006196 kfree(de);
Pavel Begunkovae348172020-07-23 20:25:20 +03006197 io_queue_async_work(req);
6198 return -EIOCBQUEUED;
Jens Axboe31b51512019-01-18 22:56:34 -07006199 }
6200
6201 trace_io_uring_defer(ctx, req, req->user_data);
Pavel Begunkov27dc8332020-07-13 23:37:14 +03006202 de->req = req;
Pavel Begunkov9cf7c102020-07-13 23:37:15 +03006203 de->seq = seq;
Pavel Begunkov27dc8332020-07-13 23:37:14 +03006204 list_add_tail(&de->list, &ctx->defer_list);
Jens Axboe31b51512019-01-18 22:56:34 -07006205 spin_unlock_irq(&ctx->completion_lock);
6206 return -EIOCBQUEUED;
6207}
Jens Axboeedafcce2019-01-09 09:16:05 -07006208
Pavel Begunkov3ca405e2020-07-13 23:37:08 +03006209static void __io_clean_op(struct io_kiocb *req)
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03006210{
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006211 if (req->flags & REQ_F_BUFFER_SELECTED) {
6212 switch (req->opcode) {
6213 case IORING_OP_READV:
6214 case IORING_OP_READ_FIXED:
6215 case IORING_OP_READ:
Jens Axboebcda7ba2020-02-23 16:42:51 -07006216 kfree((void *)(unsigned long)req->rw.addr);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006217 break;
6218 case IORING_OP_RECVMSG:
6219 case IORING_OP_RECV:
Jens Axboe52de1fe2020-02-27 10:15:42 -07006220 kfree(req->sr_msg.kbuf);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006221 break;
6222 }
6223 req->flags &= ~REQ_F_BUFFER_SELECTED;
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03006224 }
6225
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006226 if (req->flags & REQ_F_NEED_CLEANUP) {
6227 switch (req->opcode) {
6228 case IORING_OP_READV:
6229 case IORING_OP_READ_FIXED:
6230 case IORING_OP_READ:
6231 case IORING_OP_WRITEV:
6232 case IORING_OP_WRITE_FIXED:
Jens Axboee8c2bc12020-08-15 18:44:09 -07006233 case IORING_OP_WRITE: {
6234 struct io_async_rw *io = req->async_data;
6235 if (io->free_iovec)
6236 kfree(io->free_iovec);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006237 break;
Jens Axboee8c2bc12020-08-15 18:44:09 -07006238 }
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006239 case IORING_OP_RECVMSG:
Jens Axboee8c2bc12020-08-15 18:44:09 -07006240 case IORING_OP_SENDMSG: {
6241 struct io_async_msghdr *io = req->async_data;
Pavel Begunkov257e84a2021-02-05 00:58:00 +00006242
6243 kfree(io->free_iov);
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006244 break;
Jens Axboee8c2bc12020-08-15 18:44:09 -07006245 }
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006246 case IORING_OP_SPLICE:
6247 case IORING_OP_TEE:
6248 io_put_file(req, req->splice.file_in,
6249 (req->splice.flags & SPLICE_F_FD_IN_FIXED));
6250 break;
Jens Axboef3cd48502020-09-24 14:55:54 -06006251 case IORING_OP_OPENAT:
6252 case IORING_OP_OPENAT2:
6253 if (req->open.filename)
6254 putname(req->open.filename);
6255 break;
Jens Axboe80a261f2020-09-28 14:23:58 -06006256 case IORING_OP_RENAMEAT:
6257 putname(req->rename.oldpath);
6258 putname(req->rename.newpath);
6259 break;
Jens Axboe14a11432020-09-28 14:27:37 -06006260 case IORING_OP_UNLINKAT:
6261 putname(req->unlink.filename);
6262 break;
Pavel Begunkov0e1b6fe32020-07-16 23:28:02 +03006263 }
6264 req->flags &= ~REQ_F_NEED_CLEANUP;
6265 }
Pavel Begunkov99bc4c32020-02-07 22:04:45 +03006266}
6267
Pavel Begunkov889fca72021-02-10 00:03:09 +00006268static int io_issue_sqe(struct io_kiocb *req, unsigned int issue_flags)
Jens Axboeedafcce2019-01-09 09:16:05 -07006269{
Jens Axboeedafcce2019-01-09 09:16:05 -07006270 struct io_ring_ctx *ctx = req->ctx;
Jens Axboed625c6e2019-12-17 19:53:05 -07006271 int ret;
Jens Axboeedafcce2019-01-09 09:16:05 -07006272
Jens Axboed625c6e2019-12-17 19:53:05 -07006273 switch (req->opcode) {
Jens Axboe2b188cc2019-01-07 10:46:33 -07006274 case IORING_OP_NOP:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006275 ret = io_nop(req, issue_flags);
Jens Axboe31b51512019-01-18 22:56:34 -07006276 break;
6277 case IORING_OP_READV:
Jens Axboe3529d8c2019-12-19 18:24:38 -07006278 case IORING_OP_READ_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07006279 case IORING_OP_READ:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006280 ret = io_read(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006281 break;
6282 case IORING_OP_WRITEV:
Jens Axboe2b188cc2019-01-07 10:46:33 -07006283 case IORING_OP_WRITE_FIXED:
Jens Axboe3a6820f2019-12-22 15:19:35 -07006284 case IORING_OP_WRITE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006285 ret = io_write(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006286 break;
6287 case IORING_OP_FSYNC:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006288 ret = io_fsync(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006289 break;
6290 case IORING_OP_POLL_ADD:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006291 ret = io_poll_add(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006292 break;
6293 case IORING_OP_POLL_REMOVE:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006294 ret = io_poll_remove(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07006295 break;
6296 case IORING_OP_SYNC_FILE_RANGE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006297 ret = io_sync_file_range(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07006298 break;
6299 case IORING_OP_SENDMSG:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006300 ret = io_sendmsg(req, issue_flags);
Pavel Begunkov062d04d2020-10-10 18:34:12 +01006301 break;
Jens Axboefddafac2020-01-04 20:19:44 -07006302 case IORING_OP_SEND:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006303 ret = io_send(req, issue_flags);
Jens Axboeb76da702019-11-20 13:05:32 -07006304 break;
6305 case IORING_OP_RECVMSG:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006306 ret = io_recvmsg(req, issue_flags);
Pavel Begunkov062d04d2020-10-10 18:34:12 +01006307 break;
Jens Axboefddafac2020-01-04 20:19:44 -07006308 case IORING_OP_RECV:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006309 ret = io_recv(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006310 break;
6311 case IORING_OP_TIMEOUT:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006312 ret = io_timeout(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006313 break;
6314 case IORING_OP_TIMEOUT_REMOVE:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006315 ret = io_timeout_remove(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006316 break;
6317 case IORING_OP_ACCEPT:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006318 ret = io_accept(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006319 break;
6320 case IORING_OP_CONNECT:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006321 ret = io_connect(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006322 break;
6323 case IORING_OP_ASYNC_CANCEL:
Pavel Begunkov61e98202021-02-10 00:03:08 +00006324 ret = io_async_cancel(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006325 break;
Jens Axboed63d1b52019-12-10 10:38:56 -07006326 case IORING_OP_FALLOCATE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006327 ret = io_fallocate(req, issue_flags);
Jens Axboed63d1b52019-12-10 10:38:56 -07006328 break;
Jens Axboe15b71ab2019-12-11 11:20:36 -07006329 case IORING_OP_OPENAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006330 ret = io_openat(req, issue_flags);
Jens Axboe15b71ab2019-12-11 11:20:36 -07006331 break;
Jens Axboeb5dba592019-12-11 14:02:38 -07006332 case IORING_OP_CLOSE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006333 ret = io_close(req, issue_flags);
Jens Axboeb5dba592019-12-11 14:02:38 -07006334 break;
Jens Axboe05f3fb32019-12-09 11:22:50 -07006335 case IORING_OP_FILES_UPDATE:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006336 ret = io_files_update(req, issue_flags);
Jens Axboe05f3fb32019-12-09 11:22:50 -07006337 break;
Jens Axboeeddc7ef2019-12-13 21:18:10 -07006338 case IORING_OP_STATX:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006339 ret = io_statx(req, issue_flags);
Jens Axboeeddc7ef2019-12-13 21:18:10 -07006340 break;
Jens Axboe4840e412019-12-25 22:03:45 -07006341 case IORING_OP_FADVISE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006342 ret = io_fadvise(req, issue_flags);
Jens Axboe4840e412019-12-25 22:03:45 -07006343 break;
Jens Axboec1ca7572019-12-25 22:18:28 -07006344 case IORING_OP_MADVISE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006345 ret = io_madvise(req, issue_flags);
Jens Axboec1ca7572019-12-25 22:18:28 -07006346 break;
Jens Axboecebdb982020-01-08 17:59:24 -07006347 case IORING_OP_OPENAT2:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006348 ret = io_openat2(req, issue_flags);
Jens Axboecebdb982020-01-08 17:59:24 -07006349 break;
Jens Axboe3e4827b2020-01-08 15:18:09 -07006350 case IORING_OP_EPOLL_CTL:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006351 ret = io_epoll_ctl(req, issue_flags);
Jens Axboe3e4827b2020-01-08 15:18:09 -07006352 break;
Pavel Begunkov7d67af22020-02-24 11:32:45 +03006353 case IORING_OP_SPLICE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006354 ret = io_splice(req, issue_flags);
Pavel Begunkov7d67af22020-02-24 11:32:45 +03006355 break;
Jens Axboeddf0322d2020-02-23 16:41:33 -07006356 case IORING_OP_PROVIDE_BUFFERS:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006357 ret = io_provide_buffers(req, issue_flags);
Jens Axboeddf0322d2020-02-23 16:41:33 -07006358 break;
Jens Axboe067524e2020-03-02 16:32:28 -07006359 case IORING_OP_REMOVE_BUFFERS:
Pavel Begunkov889fca72021-02-10 00:03:09 +00006360 ret = io_remove_buffers(req, issue_flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006361 break;
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03006362 case IORING_OP_TEE:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006363 ret = io_tee(req, issue_flags);
Pavel Begunkovf2a8d5c2020-05-17 14:18:06 +03006364 break;
Jens Axboe36f4fa62020-09-05 11:14:22 -06006365 case IORING_OP_SHUTDOWN:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006366 ret = io_shutdown(req, issue_flags);
Jens Axboe36f4fa62020-09-05 11:14:22 -06006367 break;
Jens Axboe80a261f2020-09-28 14:23:58 -06006368 case IORING_OP_RENAMEAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006369 ret = io_renameat(req, issue_flags);
Jens Axboe80a261f2020-09-28 14:23:58 -06006370 break;
Jens Axboe14a11432020-09-28 14:27:37 -06006371 case IORING_OP_UNLINKAT:
Pavel Begunkov45d189c2021-02-10 00:03:07 +00006372 ret = io_unlinkat(req, issue_flags);
Jens Axboe14a11432020-09-28 14:27:37 -06006373 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006374 default:
6375 ret = -EINVAL;
6376 break;
Jens Axboe31b51512019-01-18 22:56:34 -07006377 }
6378
6379 if (ret)
Jens Axboeedafcce2019-01-09 09:16:05 -07006380 return ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006381
Jens Axboeb5325762020-05-19 21:20:27 -06006382 /* If the op doesn't have a file, we're not polling for it */
6383 if ((ctx->flags & IORING_SETUP_IOPOLL) && req->file) {
Jens Axboe11ba8202020-01-15 21:51:17 -07006384 const bool in_async = io_wq_current_is_worker();
6385
Jens Axboe11ba8202020-01-15 21:51:17 -07006386 /* workqueue context doesn't hold uring_lock, grab it now */
6387 if (in_async)
6388 mutex_lock(&ctx->uring_lock);
6389
Xiaoguang Wang2e9dbe92020-11-13 00:44:08 +08006390 io_iopoll_req_issued(req, in_async);
Jens Axboe11ba8202020-01-15 21:51:17 -07006391
6392 if (in_async)
6393 mutex_unlock(&ctx->uring_lock);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006394 }
6395
6396 return 0;
6397}
6398
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00006399static void io_wq_submit_work(struct io_wq_work *work)
Pavel Begunkovd4c81f32020-06-08 21:08:19 +03006400{
Jens Axboe2b188cc2019-01-07 10:46:33 -07006401 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Pavel Begunkov6df1db62020-07-03 22:15:06 +03006402 struct io_kiocb *timeout;
Jens Axboe561fb042019-10-24 07:25:42 -06006403 int ret = 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006404
Pavel Begunkov6df1db62020-07-03 22:15:06 +03006405 timeout = io_prep_linked_timeout(req);
6406 if (timeout)
6407 io_queue_linked_timeout(timeout);
Pavel Begunkovd4c81f32020-06-08 21:08:19 +03006408
Jens Axboe4014d942021-01-19 15:53:54 -07006409 if (work->flags & IO_WQ_WORK_CANCEL)
Jens Axboe561fb042019-10-24 07:25:42 -06006410 ret = -ECANCELED;
Jens Axboe31b51512019-01-18 22:56:34 -07006411
Jens Axboe561fb042019-10-24 07:25:42 -06006412 if (!ret) {
Jens Axboe561fb042019-10-24 07:25:42 -06006413 do {
Pavel Begunkov889fca72021-02-10 00:03:09 +00006414 ret = io_issue_sqe(req, 0);
Jens Axboe561fb042019-10-24 07:25:42 -06006415 /*
6416 * We can get EAGAIN for polled IO even though we're
6417 * forcing a sync submission from here, since we can't
6418 * wait for request slots on the block side.
6419 */
6420 if (ret != -EAGAIN)
6421 break;
6422 cond_resched();
6423 } while (1);
6424 }
Jens Axboe31b51512019-01-18 22:56:34 -07006425
Jens Axboe561fb042019-10-24 07:25:42 -06006426 if (ret) {
Xiaoguang Wangc07e6712020-12-14 23:49:41 +08006427 struct io_ring_ctx *lock_ctx = NULL;
Xiaoguang Wangdad1b122020-12-06 22:22:42 +00006428
Xiaoguang Wangc07e6712020-12-14 23:49:41 +08006429 if (req->ctx->flags & IORING_SETUP_IOPOLL)
6430 lock_ctx = req->ctx;
6431
6432 /*
6433 * io_iopoll_complete() does not hold completion_lock to
6434 * complete polled io, so here for polled io, we can not call
6435 * io_req_complete() directly, otherwise there maybe concurrent
6436 * access to cqring, defer_list, etc, which is not safe. Given
6437 * that io_iopoll_complete() is always called under uring_lock,
6438 * so here for polled io, we also get uring_lock to complete
6439 * it.
6440 */
6441 if (lock_ctx)
6442 mutex_lock(&lock_ctx->uring_lock);
6443
6444 req_set_fail_links(req);
6445 io_req_complete(req, ret);
6446
6447 if (lock_ctx)
6448 mutex_unlock(&lock_ctx->uring_lock);
Jens Axboeedafcce2019-01-09 09:16:05 -07006449 }
Jens Axboe31b51512019-01-18 22:56:34 -07006450}
Jens Axboe2b188cc2019-01-07 10:46:33 -07006451
Jens Axboe65e19f52019-10-26 07:20:21 -06006452static inline struct file *io_file_from_index(struct io_ring_ctx *ctx,
6453 int index)
Jens Axboe09bb8392019-03-13 12:39:28 -06006454{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006455 struct fixed_rsrc_table *table;
Jens Axboe65e19f52019-10-26 07:20:21 -06006456
Jens Axboe05f3fb32019-12-09 11:22:50 -07006457 table = &ctx->file_data->table[index >> IORING_FILE_TABLE_SHIFT];
Xiaoming Ni84695082020-05-11 19:25:43 +08006458 return table->files[index & IORING_FILE_TABLE_MASK];
Jens Axboe65e19f52019-10-26 07:20:21 -06006459}
6460
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006461static struct file *io_file_get(struct io_submit_state *state,
6462 struct io_kiocb *req, int fd, bool fixed)
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006463{
6464 struct io_ring_ctx *ctx = req->ctx;
6465 struct file *file;
6466
6467 if (fixed) {
Pavel Begunkov479f5172020-10-10 18:34:07 +01006468 if (unlikely((unsigned int)fd >= ctx->nr_user_files))
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006469 return NULL;
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006470 fd = array_index_nospec(fd, ctx->nr_user_files);
6471 file = io_file_from_index(ctx, fd);
Pavel Begunkov36f72fe2020-11-18 19:57:26 +00006472 io_set_resource_node(req);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006473 } else {
6474 trace_io_uring_file_get(ctx, fd);
6475 file = __io_file_get(state, fd);
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006476 }
6477
Pavel Begunkovce3d5aa2021-02-01 18:59:55 +00006478 if (file && unlikely(file->f_op == &io_uring_fops))
6479 io_req_track_inflight(req);
Pavel Begunkov8371adf2020-10-10 18:34:08 +01006480 return file;
Pavel Begunkov8da11c12020-02-24 11:32:44 +03006481}
6482
Jens Axboe2665abf2019-11-05 12:40:47 -07006483static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer)
6484{
Jens Axboead8a48a2019-11-15 08:49:11 -07006485 struct io_timeout_data *data = container_of(timer,
6486 struct io_timeout_data, timer);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006487 struct io_kiocb *prev, *req = data->req;
Jens Axboe2665abf2019-11-05 12:40:47 -07006488 struct io_ring_ctx *ctx = req->ctx;
Jens Axboe2665abf2019-11-05 12:40:47 -07006489 unsigned long flags;
Jens Axboe2665abf2019-11-05 12:40:47 -07006490
6491 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006492 prev = req->timeout.head;
6493 req->timeout.head = NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006494
6495 /*
6496 * We don't expect the list to be empty, that will only happen if we
6497 * race with the completion of the linked work.
6498 */
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006499 if (prev && refcount_inc_not_zero(&prev->refs))
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006500 io_remove_next_linked(prev);
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006501 else
6502 prev = NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006503 spin_unlock_irqrestore(&ctx->completion_lock, flags);
6504
6505 if (prev) {
Jens Axboe4e88d6e2019-12-07 20:59:47 -07006506 req_set_fail_links(prev);
Pavel Begunkov014db002020-03-03 21:33:12 +03006507 io_async_find_and_cancel(ctx, req, prev->user_data, -ETIME);
Pavel Begunkov9ae1f8d2021-02-01 18:59:51 +00006508 io_put_req_deferred(prev, 1);
Jens Axboe47f46762019-11-09 17:43:02 -07006509 } else {
Pavel Begunkov9ae1f8d2021-02-01 18:59:51 +00006510 io_req_complete_post(req, -ETIME, 0);
6511 io_put_req_deferred(req, 1);
Jens Axboe2665abf2019-11-05 12:40:47 -07006512 }
Jens Axboe2665abf2019-11-05 12:40:47 -07006513 return HRTIMER_NORESTART;
6514}
6515
Jens Axboe7271ef32020-08-10 09:55:22 -06006516static void __io_queue_linked_timeout(struct io_kiocb *req)
Jens Axboe2665abf2019-11-05 12:40:47 -07006517{
Jens Axboe76a46e02019-11-10 23:34:16 -07006518 /*
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006519 * If the back reference is NULL, then our linked request finished
6520 * before we got a chance to setup the timer
Jens Axboe76a46e02019-11-10 23:34:16 -07006521 */
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006522 if (req->timeout.head) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07006523 struct io_timeout_data *data = req->async_data;
Jens Axboe94ae5e72019-11-14 19:39:52 -07006524
Jens Axboead8a48a2019-11-15 08:49:11 -07006525 data->timer.function = io_link_timeout_fn;
6526 hrtimer_start(&data->timer, timespec64_to_ktime(data->ts),
6527 data->mode);
Jens Axboe2665abf2019-11-05 12:40:47 -07006528 }
Jens Axboe7271ef32020-08-10 09:55:22 -06006529}
6530
6531static void io_queue_linked_timeout(struct io_kiocb *req)
6532{
6533 struct io_ring_ctx *ctx = req->ctx;
6534
6535 spin_lock_irq(&ctx->completion_lock);
6536 __io_queue_linked_timeout(req);
Jens Axboe76a46e02019-11-10 23:34:16 -07006537 spin_unlock_irq(&ctx->completion_lock);
Jens Axboe2665abf2019-11-05 12:40:47 -07006538
Jens Axboe2665abf2019-11-05 12:40:47 -07006539 /* drop submission reference */
Jens Axboe76a46e02019-11-10 23:34:16 -07006540 io_put_req(req);
Jens Axboe2665abf2019-11-05 12:40:47 -07006541}
6542
Jens Axboead8a48a2019-11-15 08:49:11 -07006543static struct io_kiocb *io_prep_linked_timeout(struct io_kiocb *req)
Jens Axboe2665abf2019-11-05 12:40:47 -07006544{
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006545 struct io_kiocb *nxt = req->link;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006546
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006547 if (!nxt || (req->flags & REQ_F_LINK_TIMEOUT) ||
6548 nxt->opcode != IORING_OP_LINK_TIMEOUT)
Jens Axboed7718a92020-02-14 22:23:12 -07006549 return NULL;
Jens Axboe2665abf2019-11-05 12:40:47 -07006550
Pavel Begunkov90cd7e42020-10-27 23:25:36 +00006551 nxt->timeout.head = req;
Pavel Begunkov900fad42020-10-19 16:39:16 +01006552 nxt->flags |= REQ_F_LTIMEOUT_ACTIVE;
Jens Axboe76a46e02019-11-10 23:34:16 -07006553 req->flags |= REQ_F_LINK_TIMEOUT;
Jens Axboe76a46e02019-11-10 23:34:16 -07006554 return nxt;
Jens Axboe2665abf2019-11-05 12:40:47 -07006555}
6556
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006557static void __io_queue_sqe(struct io_kiocb *req)
Jens Axboe2b188cc2019-01-07 10:46:33 -07006558{
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006559 struct io_kiocb *linked_timeout = io_prep_linked_timeout(req);
Jens Axboe193155c2020-02-22 23:22:19 -07006560 const struct cred *old_creds = NULL;
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006561 int ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006562
Pavel Begunkov2e5aa6c2020-10-18 10:17:37 +01006563 if ((req->flags & REQ_F_WORK_INITIALIZED) &&
6564 (req->work.flags & IO_WQ_WORK_CREDS) &&
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006565 req->work.identity->creds != current_cred())
6566 old_creds = override_creds(req->work.identity->creds);
Jens Axboe193155c2020-02-22 23:22:19 -07006567
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006568 ret = io_issue_sqe(req, IO_URING_F_NONBLOCK|IO_URING_F_COMPLETE_DEFER);
Jens Axboe491381ce2019-10-17 09:20:46 -06006569
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006570 if (old_creds)
6571 revert_creds(old_creds);
6572
Jens Axboe491381ce2019-10-17 09:20:46 -06006573 /*
6574 * We async punt it if the file wasn't marked NOWAIT, or if the file
6575 * doesn't support non-blocking read/write attempts
6576 */
Pavel Begunkov24c74672020-06-21 13:09:51 +03006577 if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) {
Pavel Begunkovf063c542020-07-25 14:41:59 +03006578 if (!io_arm_poll_handler(req)) {
Pavel Begunkovf063c542020-07-25 14:41:59 +03006579 /*
6580 * Queued up for async execution, worker will release
6581 * submit reference when the iocb is actually submitted.
6582 */
6583 io_queue_async_work(req);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006584 }
Pavel Begunkov0d63c142020-10-22 16:47:18 +01006585 } else if (likely(!ret)) {
6586 /* drop submission reference */
Pavel Begunkove342c802021-01-19 13:32:47 +00006587 if (req->flags & REQ_F_COMPLETE_INLINE) {
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006588 struct io_ring_ctx *ctx = req->ctx;
6589 struct io_comp_state *cs = &ctx->submit_state.comp;
6590
Pavel Begunkov6dd0be12021-02-10 00:03:13 +00006591 cs->reqs[cs->nr++] = req;
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006592 if (cs->nr == ARRAY_SIZE(cs->reqs))
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006593 io_submit_flush_completions(cs, ctx);
Pavel Begunkov9affd662021-01-19 13:32:46 +00006594 } else {
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006595 io_put_req(req);
Pavel Begunkov0d63c142020-10-22 16:47:18 +01006596 }
6597 } else {
Jens Axboe4e88d6e2019-12-07 20:59:47 -07006598 req_set_fail_links(req);
Jens Axboee65ef562019-03-12 10:16:44 -06006599 io_put_req(req);
Pavel Begunkov652532a2020-07-03 22:15:07 +03006600 io_req_complete(req, ret);
Jens Axboe9e645e112019-05-10 16:07:28 -06006601 }
Pavel Begunkovd3d72982021-02-12 03:23:51 +00006602 if (linked_timeout)
6603 io_queue_linked_timeout(linked_timeout);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006604}
6605
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006606static void io_queue_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe)
Jackie Liu4fe2c962019-09-09 20:50:40 +08006607{
6608 int ret;
6609
Jens Axboe3529d8c2019-12-19 18:24:38 -07006610 ret = io_req_defer(req, sqe);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006611 if (ret) {
6612 if (ret != -EIOCBQUEUED) {
Pavel Begunkov11185912020-01-22 23:09:35 +03006613fail_req:
Jens Axboe4e88d6e2019-12-07 20:59:47 -07006614 req_set_fail_links(req);
Jens Axboee1e16092020-06-22 09:17:17 -06006615 io_put_req(req);
6616 io_req_complete(req, ret);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006617 }
Pavel Begunkov25508782019-12-30 21:24:47 +03006618 } else if (req->flags & REQ_F_FORCE_ASYNC) {
Jens Axboee8c2bc12020-08-15 18:44:09 -07006619 if (!req->async_data) {
Pavel Begunkovbd2ab182020-05-17 14:02:12 +03006620 ret = io_req_defer_prep(req, sqe);
Pavel Begunkov327d6d92020-07-15 12:46:51 +03006621 if (unlikely(ret))
Pavel Begunkovbd2ab182020-05-17 14:02:12 +03006622 goto fail_req;
6623 }
Jens Axboece35a472019-12-17 08:04:44 -07006624 io_queue_async_work(req);
6625 } else {
Pavel Begunkovc1379e22020-09-30 22:57:56 +03006626 if (sqe) {
6627 ret = io_req_prep(req, sqe);
6628 if (unlikely(ret))
6629 goto fail_req;
6630 }
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006631 __io_queue_sqe(req);
Jens Axboece35a472019-12-17 08:04:44 -07006632 }
Jackie Liu4fe2c962019-09-09 20:50:40 +08006633}
6634
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006635static inline void io_queue_link_head(struct io_kiocb *req)
Jackie Liu4fe2c962019-09-09 20:50:40 +08006636{
Jens Axboe94ae5e72019-11-14 19:39:52 -07006637 if (unlikely(req->flags & REQ_F_FAIL_LINK)) {
Jens Axboee1e16092020-06-22 09:17:17 -06006638 io_put_req(req);
6639 io_req_complete(req, -ECANCELED);
Pavel Begunkov1b4a51b2019-11-21 11:54:28 +03006640 } else
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006641 io_queue_sqe(req, NULL);
Jackie Liu4fe2c962019-09-09 20:50:40 +08006642}
6643
Pavel Begunkov863e0562020-10-27 23:25:35 +00006644struct io_submit_link {
6645 struct io_kiocb *head;
6646 struct io_kiocb *last;
6647};
6648
Pavel Begunkov1d4240c2020-04-12 02:05:03 +03006649static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe,
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006650 struct io_submit_link *link)
Jens Axboe9e645e112019-05-10 16:07:28 -06006651{
Jackie Liua197f662019-11-08 08:09:12 -07006652 struct io_ring_ctx *ctx = req->ctx;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006653 int ret;
Jens Axboe9e645e112019-05-10 16:07:28 -06006654
Jens Axboe9e645e112019-05-10 16:07:28 -06006655 /*
6656 * If we already have a head request, queue this one for async
6657 * submittal once the head completes. If we don't have a head but
6658 * IOSQE_IO_LINK is set in the sqe, start a new head. This one will be
6659 * submitted sync once the chain is complete. If none of those
6660 * conditions are true (normal request), then just queue it.
6661 */
Pavel Begunkov863e0562020-10-27 23:25:35 +00006662 if (link->head) {
6663 struct io_kiocb *head = link->head;
Jens Axboe9e645e112019-05-10 16:07:28 -06006664
Pavel Begunkov8cdf2192020-01-25 00:40:24 +03006665 /*
6666 * Taking sequential execution of a link, draining both sides
6667 * of the link also fullfils IOSQE_IO_DRAIN semantics for all
6668 * requests in the link. So, it drains the head and the
6669 * next after the link request. The last one is done via
6670 * drain_next flag to persist the effect across calls.
6671 */
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006672 if (req->flags & REQ_F_IO_DRAIN) {
Pavel Begunkov711be032020-01-17 03:57:59 +03006673 head->flags |= REQ_F_IO_DRAIN;
6674 ctx->drain_next = 1;
6675 }
Jens Axboe3529d8c2019-12-19 18:24:38 -07006676 ret = io_req_defer_prep(req, sqe);
Pavel Begunkov327d6d92020-07-15 12:46:51 +03006677 if (unlikely(ret)) {
Jens Axboe4e88d6e2019-12-07 20:59:47 -07006678 /* fail even hard links since we don't submit */
Pavel Begunkov9d763772019-12-17 02:22:07 +03006679 head->flags |= REQ_F_FAIL_LINK;
Pavel Begunkov1d4240c2020-04-12 02:05:03 +03006680 return ret;
Jens Axboe2d283902019-12-04 11:08:05 -07006681 }
Pavel Begunkov9d763772019-12-17 02:22:07 +03006682 trace_io_uring_link(ctx, req, head);
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006683 link->last->link = req;
Pavel Begunkov863e0562020-10-27 23:25:35 +00006684 link->last = req;
Jens Axboe9e645e112019-05-10 16:07:28 -06006685
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006686 /* last request of a link, enqueue the link */
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006687 if (!(req->flags & (REQ_F_LINK | REQ_F_HARDLINK))) {
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006688 io_queue_link_head(head);
Pavel Begunkov863e0562020-10-27 23:25:35 +00006689 link->head = NULL;
Pavel Begunkov32fe5252019-12-17 22:26:58 +03006690 }
Jens Axboe9e645e112019-05-10 16:07:28 -06006691 } else {
Pavel Begunkov711be032020-01-17 03:57:59 +03006692 if (unlikely(ctx->drain_next)) {
6693 req->flags |= REQ_F_IO_DRAIN;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006694 ctx->drain_next = 0;
Pavel Begunkov711be032020-01-17 03:57:59 +03006695 }
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006696 if (req->flags & (REQ_F_LINK | REQ_F_HARDLINK)) {
Pavel Begunkov711be032020-01-17 03:57:59 +03006697 ret = io_req_defer_prep(req, sqe);
Pavel Begunkov327d6d92020-07-15 12:46:51 +03006698 if (unlikely(ret))
Pavel Begunkov711be032020-01-17 03:57:59 +03006699 req->flags |= REQ_F_FAIL_LINK;
Pavel Begunkov863e0562020-10-27 23:25:35 +00006700 link->head = req;
6701 link->last = req;
Pavel Begunkov711be032020-01-17 03:57:59 +03006702 } else {
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006703 io_queue_sqe(req, sqe);
Pavel Begunkov711be032020-01-17 03:57:59 +03006704 }
Jens Axboe9e645e112019-05-10 16:07:28 -06006705 }
Pavel Begunkov2e6e1fd2019-12-05 16:15:45 +03006706
Pavel Begunkov1d4240c2020-04-12 02:05:03 +03006707 return 0;
Jens Axboe9e645e112019-05-10 16:07:28 -06006708}
6709
Jens Axboe9a56a232019-01-09 09:06:50 -07006710/*
6711 * Batched submission is done, ensure local IO is flushed out.
6712 */
Pavel Begunkovba88ff12021-02-10 00:03:11 +00006713static void io_submit_state_end(struct io_submit_state *state,
6714 struct io_ring_ctx *ctx)
Jens Axboe9a56a232019-01-09 09:06:50 -07006715{
Pavel Begunkov6dd0be12021-02-10 00:03:13 +00006716 if (state->comp.nr)
Pavel Begunkovba88ff12021-02-10 00:03:11 +00006717 io_submit_flush_completions(&state->comp, ctx);
Jens Axboe27926b62020-10-28 09:33:23 -06006718 if (state->plug_started)
6719 blk_finish_plug(&state->plug);
Pavel Begunkov9f13c352020-05-17 14:13:41 +03006720 io_state_file_put(state);
Jens Axboe9a56a232019-01-09 09:06:50 -07006721}
6722
6723/*
6724 * Start submission side cache.
6725 */
6726static void io_submit_state_start(struct io_submit_state *state,
Pavel Begunkovba88ff12021-02-10 00:03:11 +00006727 unsigned int max_ios)
Jens Axboe9a56a232019-01-09 09:06:50 -07006728{
Jens Axboe27926b62020-10-28 09:33:23 -06006729 state->plug_started = false;
Jens Axboe9a56a232019-01-09 09:06:50 -07006730 state->ios_left = max_ios;
6731}
6732
Jens Axboe2b188cc2019-01-07 10:46:33 -07006733static void io_commit_sqring(struct io_ring_ctx *ctx)
6734{
Hristo Venev75b28af2019-08-26 17:23:46 +00006735 struct io_rings *rings = ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006736
Pavel Begunkovcaf582c2019-12-30 21:24:46 +03006737 /*
6738 * Ensure any loads from the SQEs are done at this point,
6739 * since once we write the new head, the application could
6740 * write new data to them.
6741 */
6742 smp_store_release(&rings->sq.head, ctx->cached_sq_head);
Jens Axboe2b188cc2019-01-07 10:46:33 -07006743}
6744
6745/*
Jens Axboe3529d8c2019-12-19 18:24:38 -07006746 * Fetch an sqe, if one is available. Note that sqe_ptr will point to memory
Jens Axboe2b188cc2019-01-07 10:46:33 -07006747 * that is mapped by userspace. This means that care needs to be taken to
6748 * ensure that reads are stable, as we cannot rely on userspace always
6749 * being a good citizen. If members of the sqe are validated and then later
6750 * used, it's important that those reads are done through READ_ONCE() to
6751 * prevent a re-load down the line.
6752 */
Pavel Begunkov709b3022020-04-08 08:58:43 +03006753static const struct io_uring_sqe *io_get_sqe(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07006754{
Hristo Venev75b28af2019-08-26 17:23:46 +00006755 u32 *sq_array = ctx->sq_array;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006756 unsigned head;
6757
6758 /*
6759 * The cached sq head (or cq tail) serves two purposes:
6760 *
6761 * 1) allows us to batch the cost of updating the user visible
6762 * head updates.
6763 * 2) allows the kernel side to track the head on its own, even
6764 * though the application is the one updating it.
6765 */
Pavel Begunkovee7d46d2019-12-30 21:24:45 +03006766 head = READ_ONCE(sq_array[ctx->cached_sq_head & ctx->sq_mask]);
Pavel Begunkov709b3022020-04-08 08:58:43 +03006767 if (likely(head < ctx->sq_entries))
6768 return &ctx->sq_sqes[head];
Jens Axboe2b188cc2019-01-07 10:46:33 -07006769
6770 /* drop invalid entries */
Jens Axboe498ccd92019-10-25 10:04:25 -06006771 ctx->cached_sq_dropped++;
Pavel Begunkovee7d46d2019-12-30 21:24:45 +03006772 WRITE_ONCE(ctx->rings->sq_dropped, ctx->cached_sq_dropped);
Pavel Begunkov709b3022020-04-08 08:58:43 +03006773 return NULL;
6774}
6775
6776static inline void io_consume_sqe(struct io_ring_ctx *ctx)
6777{
6778 ctx->cached_sq_head++;
Jens Axboe2b188cc2019-01-07 10:46:33 -07006779}
6780
Stefano Garzarella21b55db2020-08-27 16:58:30 +02006781/*
6782 * Check SQE restrictions (opcode and flags).
6783 *
6784 * Returns 'true' if SQE is allowed, 'false' otherwise.
6785 */
6786static inline bool io_check_restriction(struct io_ring_ctx *ctx,
6787 struct io_kiocb *req,
6788 unsigned int sqe_flags)
6789{
6790 if (!ctx->restricted)
6791 return true;
6792
6793 if (!test_bit(req->opcode, ctx->restrictions.sqe_op))
6794 return false;
6795
6796 if ((sqe_flags & ctx->restrictions.sqe_flags_required) !=
6797 ctx->restrictions.sqe_flags_required)
6798 return false;
6799
6800 if (sqe_flags & ~(ctx->restrictions.sqe_flags_allowed |
6801 ctx->restrictions.sqe_flags_required))
6802 return false;
6803
6804 return true;
6805}
6806
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006807#define SQE_VALID_FLAGS (IOSQE_FIXED_FILE|IOSQE_IO_DRAIN|IOSQE_IO_LINK| \
6808 IOSQE_IO_HARDLINK | IOSQE_ASYNC | \
6809 IOSQE_BUFFER_SELECT)
6810
6811static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req,
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006812 const struct io_uring_sqe *sqe)
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006813{
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006814 struct io_submit_state *state;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006815 unsigned int sqe_flags;
Pavel Begunkov71b547c2020-10-10 18:34:09 +01006816 int id, ret;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006817
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006818 req->opcode = READ_ONCE(sqe->opcode);
6819 req->user_data = READ_ONCE(sqe->user_data);
Jens Axboee8c2bc12020-08-15 18:44:09 -07006820 req->async_data = NULL;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006821 req->file = NULL;
6822 req->ctx = ctx;
6823 req->flags = 0;
Pavel Begunkovf2f87372020-10-27 23:25:37 +00006824 req->link = NULL;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00006825 req->fixed_rsrc_refs = NULL;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006826 /* one is dropped after submission, the other at completion */
6827 refcount_set(&req->refs, 2);
Pavel Begunkov4dd28242020-06-15 10:33:13 +03006828 req->task = current;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006829 req->result = 0;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006830
6831 if (unlikely(req->opcode >= IORING_OP_LAST))
6832 return -EINVAL;
6833
Jens Axboe28cea78a2020-09-14 10:51:17 -06006834 if (unlikely(io_sq_thread_acquire_mm_files(ctx, req)))
Jens Axboe9d8426a2020-06-16 18:42:49 -06006835 return -EFAULT;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006836
6837 sqe_flags = READ_ONCE(sqe->flags);
6838 /* enforce forwards compatibility on users */
6839 if (unlikely(sqe_flags & ~SQE_VALID_FLAGS))
6840 return -EINVAL;
6841
Stefano Garzarella21b55db2020-08-27 16:58:30 +02006842 if (unlikely(!io_check_restriction(ctx, req, sqe_flags)))
6843 return -EACCES;
6844
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006845 if ((sqe_flags & IOSQE_BUFFER_SELECT) &&
6846 !io_op_defs[req->opcode].buffer_select)
6847 return -EOPNOTSUPP;
6848
6849 id = READ_ONCE(sqe->personality);
6850 if (id) {
Jens Axboe1e6fa522020-10-15 08:46:24 -06006851 struct io_identity *iod;
6852
Jens Axboe1e6fa522020-10-15 08:46:24 -06006853 iod = idr_find(&ctx->personality_idr, id);
6854 if (unlikely(!iod))
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006855 return -EINVAL;
Jens Axboe1e6fa522020-10-15 08:46:24 -06006856 refcount_inc(&iod->count);
Pavel Begunkovec99ca62020-10-18 10:17:38 +01006857
6858 __io_req_init_async(req);
Jens Axboe1e6fa522020-10-15 08:46:24 -06006859 get_cred(iod->creds);
6860 req->work.identity = iod;
Jens Axboedfead8a2020-10-14 10:12:37 -06006861 req->work.flags |= IO_WQ_WORK_CREDS;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006862 }
6863
6864 /* same numerical values with corresponding REQ_F_*, safe to copy */
Pavel Begunkovc11368a52020-05-17 14:13:42 +03006865 req->flags |= sqe_flags;
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006866 state = &ctx->submit_state;
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006867
Jens Axboe27926b62020-10-28 09:33:23 -06006868 /*
6869 * Plug now if we have more than 1 IO left after this, and the target
6870 * is potentially a read/write to block based storage.
6871 */
6872 if (!state->plug_started && state->ios_left > 1 &&
6873 io_op_defs[req->opcode].plug) {
6874 blk_start_plug(&state->plug);
6875 state->plug_started = true;
6876 }
Jens Axboe63ff8222020-05-07 14:56:15 -06006877
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006878 ret = 0;
6879 if (io_op_defs[req->opcode].needs_file) {
6880 bool fixed = req->flags & REQ_F_FIXED_FILE;
Jens Axboe63ff8222020-05-07 14:56:15 -06006881
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006882 req->file = io_file_get(state, req, READ_ONCE(sqe->fd), fixed);
Pavel Begunkovba13e232021-02-01 18:59:52 +00006883 if (unlikely(!req->file))
Pavel Begunkovbd5bbda2020-11-20 15:50:51 +00006884 ret = -EBADF;
6885 }
6886
Pavel Begunkov71b547c2020-10-10 18:34:09 +01006887 state->ios_left--;
6888 return ret;
Pavel Begunkov0553b8b2020-04-08 08:58:45 +03006889}
6890
Jens Axboe0f212202020-09-13 13:09:39 -06006891static int io_submit_sqes(struct io_ring_ctx *ctx, unsigned int nr)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006892{
Pavel Begunkov863e0562020-10-27 23:25:35 +00006893 struct io_submit_link link;
Jens Axboe9e645e112019-05-10 16:07:28 -06006894 int i, submitted = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006895
Jens Axboec4a2ed72019-11-21 21:01:26 -07006896 /* if we have a backlog and couldn't flush it all, return BUSY */
Jens Axboead3eb2c2019-12-18 17:12:20 -07006897 if (test_bit(0, &ctx->sq_check_overflow)) {
Pavel Begunkov6c503152021-01-04 20:36:36 +00006898 if (!__io_cqring_overflow_flush(ctx, false, NULL, NULL))
Jens Axboead3eb2c2019-12-18 17:12:20 -07006899 return -EBUSY;
6900 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07006901
Pavel Begunkovee7d46d2019-12-30 21:24:45 +03006902 /* make sure SQ entry isn't read before tail */
6903 nr = min3(nr, ctx->sq_entries, io_sqring_entries(ctx));
Pavel Begunkov9ef4f122019-12-30 21:24:44 +03006904
Pavel Begunkov2b85edf2019-12-28 14:13:03 +03006905 if (!percpu_ref_tryget_many(&ctx->refs, nr))
6906 return -EAGAIN;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006907
Jens Axboed8a6df12020-10-15 16:24:45 -06006908 percpu_counter_add(&current->io_uring->inflight, nr);
Jens Axboefaf7b512020-10-07 12:48:53 -06006909 refcount_add(nr, &current->usage);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006910
Pavel Begunkovba88ff12021-02-10 00:03:11 +00006911 io_submit_state_start(&ctx->submit_state, nr);
Pavel Begunkov863e0562020-10-27 23:25:35 +00006912 link.head = NULL;
Pavel Begunkovb14cca02020-01-17 04:45:59 +03006913
Jens Axboe6c271ce2019-01-10 11:22:30 -07006914 for (i = 0; i < nr; i++) {
Jens Axboe3529d8c2019-12-19 18:24:38 -07006915 const struct io_uring_sqe *sqe;
Pavel Begunkov196be952019-11-07 01:41:06 +03006916 struct io_kiocb *req;
Pavel Begunkov1cb1edb2020-02-06 21:16:09 +03006917 int err;
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03006918
Pavel Begunkovb1e50e52020-04-08 08:58:44 +03006919 sqe = io_get_sqe(ctx);
6920 if (unlikely(!sqe)) {
6921 io_consume_sqe(ctx);
6922 break;
6923 }
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006924 req = io_alloc_req(ctx);
Pavel Begunkov196be952019-11-07 01:41:06 +03006925 if (unlikely(!req)) {
6926 if (!submitted)
6927 submitted = -EAGAIN;
Pavel Begunkovfb5ccc92019-10-25 12:31:30 +03006928 break;
Jens Axboe9e645e112019-05-10 16:07:28 -06006929 }
Pavel Begunkov709b3022020-04-08 08:58:43 +03006930 io_consume_sqe(ctx);
Jens Axboed3656342019-12-18 09:50:26 -07006931 /* will complete beyond this point, count as submitted */
6932 submitted++;
6933
Pavel Begunkov258b29a2021-02-10 00:03:10 +00006934 err = io_init_req(ctx, req, sqe);
Pavel Begunkovef4ff582020-04-12 02:05:05 +03006935 if (unlikely(err)) {
Pavel Begunkov1cb1edb2020-02-06 21:16:09 +03006936fail_req:
Jens Axboee1e16092020-06-22 09:17:17 -06006937 io_put_req(req);
6938 io_req_complete(req, err);
Jens Axboed3656342019-12-18 09:50:26 -07006939 break;
6940 }
6941
Jens Axboe354420f2020-01-08 18:55:15 -07006942 trace_io_uring_submit_sqe(ctx, req->opcode, req->user_data,
Pavel Begunkov2d7e9352021-01-19 13:32:37 +00006943 true, ctx->flags & IORING_SETUP_SQPOLL);
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006944 err = io_submit_sqe(req, sqe, &link);
Pavel Begunkov1d4240c2020-04-12 02:05:03 +03006945 if (err)
6946 goto fail_req;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006947 }
6948
Pavel Begunkov9466f432020-01-25 22:34:01 +03006949 if (unlikely(submitted != nr)) {
6950 int ref_used = (submitted == -EAGAIN) ? 0 : submitted;
Jens Axboed8a6df12020-10-15 16:24:45 -06006951 struct io_uring_task *tctx = current->io_uring;
6952 int unused = nr - ref_used;
Pavel Begunkov9466f432020-01-25 22:34:01 +03006953
Jens Axboed8a6df12020-10-15 16:24:45 -06006954 percpu_ref_put_many(&ctx->refs, unused);
6955 percpu_counter_sub(&tctx->inflight, unused);
6956 put_task_struct_many(current, unused);
Pavel Begunkov9466f432020-01-25 22:34:01 +03006957 }
Pavel Begunkov863e0562020-10-27 23:25:35 +00006958 if (link.head)
Pavel Begunkovc5eef2b2021-02-10 00:03:22 +00006959 io_queue_link_head(link.head);
Pavel Begunkovba88ff12021-02-10 00:03:11 +00006960 io_submit_state_end(&ctx->submit_state, ctx);
Jens Axboe6c271ce2019-01-10 11:22:30 -07006961
Pavel Begunkovae9428c2019-11-06 00:22:14 +03006962 /* Commit SQ ring head once we've consumed and submitted all SQEs */
6963 io_commit_sqring(ctx);
6964
Jens Axboe6c271ce2019-01-10 11:22:30 -07006965 return submitted;
6966}
6967
Xiaoguang Wang23b36282020-07-23 20:57:24 +08006968static inline void io_ring_set_wakeup_flag(struct io_ring_ctx *ctx)
6969{
6970 /* Tell userspace we may need a wakeup call */
6971 spin_lock_irq(&ctx->completion_lock);
6972 ctx->rings->sq_flags |= IORING_SQ_NEED_WAKEUP;
6973 spin_unlock_irq(&ctx->completion_lock);
6974}
6975
6976static inline void io_ring_clear_wakeup_flag(struct io_ring_ctx *ctx)
6977{
6978 spin_lock_irq(&ctx->completion_lock);
6979 ctx->rings->sq_flags &= ~IORING_SQ_NEED_WAKEUP;
6980 spin_unlock_irq(&ctx->completion_lock);
6981}
6982
Xiaoguang Wang08369242020-11-03 14:15:59 +08006983static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries)
Jens Axboe6c271ce2019-01-10 11:22:30 -07006984{
Jens Axboec8d1ba52020-09-14 11:07:26 -06006985 unsigned int to_submit;
Xiaoguang Wangbdcd3ea2020-02-25 22:12:08 +08006986 int ret = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07006987
Jens Axboec8d1ba52020-09-14 11:07:26 -06006988 to_submit = io_sqring_entries(ctx);
Jens Axboee95eee22020-09-08 09:11:32 -06006989 /* if we're handling multiple rings, cap submit size for fairness */
6990 if (cap_entries && to_submit > 8)
6991 to_submit = 8;
6992
Xiaoguang Wang906a3c62020-11-12 14:56:00 +08006993 if (!list_empty(&ctx->iopoll_list) || to_submit) {
6994 unsigned nr_events = 0;
6995
Xiaoguang Wang08369242020-11-03 14:15:59 +08006996 mutex_lock(&ctx->uring_lock);
Xiaoguang Wang906a3c62020-11-12 14:56:00 +08006997 if (!list_empty(&ctx->iopoll_list))
6998 io_do_iopoll(ctx, &nr_events, 0);
6999
Pavel Begunkovd9d05212021-01-08 20:57:25 +00007000 if (to_submit && !ctx->sqo_dead &&
7001 likely(!percpu_ref_is_dying(&ctx->refs)))
Xiaoguang Wang08369242020-11-03 14:15:59 +08007002 ret = io_submit_sqes(ctx, to_submit);
7003 mutex_unlock(&ctx->uring_lock);
7004 }
Jens Axboe90554202020-09-03 12:12:41 -06007005
7006 if (!io_sqring_full(ctx) && wq_has_sleeper(&ctx->sqo_sq_wait))
7007 wake_up(&ctx->sqo_sq_wait);
7008
Xiaoguang Wang08369242020-11-03 14:15:59 +08007009 return ret;
7010}
7011
7012static void io_sqd_update_thread_idle(struct io_sq_data *sqd)
7013{
7014 struct io_ring_ctx *ctx;
7015 unsigned sq_thread_idle = 0;
7016
7017 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
7018 if (sq_thread_idle < ctx->sq_thread_idle)
7019 sq_thread_idle = ctx->sq_thread_idle;
7020 }
7021
7022 sqd->sq_thread_idle = sq_thread_idle;
Jens Axboec8d1ba52020-09-14 11:07:26 -06007023}
7024
Jens Axboe69fb2132020-09-14 11:16:23 -06007025static void io_sqd_init_new(struct io_sq_data *sqd)
7026{
7027 struct io_ring_ctx *ctx;
7028
7029 while (!list_empty(&sqd->ctx_new_list)) {
7030 ctx = list_first_entry(&sqd->ctx_new_list, struct io_ring_ctx, sqd_list);
Jens Axboe69fb2132020-09-14 11:16:23 -06007031 list_move_tail(&ctx->sqd_list, &sqd->ctx_list);
7032 complete(&ctx->sq_thread_comp);
7033 }
Xiaoguang Wang08369242020-11-03 14:15:59 +08007034
7035 io_sqd_update_thread_idle(sqd);
Jens Axboe69fb2132020-09-14 11:16:23 -06007036}
7037
Jens Axboe6c271ce2019-01-10 11:22:30 -07007038static int io_sq_thread(void *data)
7039{
Dennis Zhou91d8f512020-09-16 13:41:05 -07007040 struct cgroup_subsys_state *cur_css = NULL;
Jens Axboe28cea78a2020-09-14 10:51:17 -06007041 struct files_struct *old_files = current->files;
7042 struct nsproxy *old_nsproxy = current->nsproxy;
Jens Axboe69fb2132020-09-14 11:16:23 -06007043 const struct cred *old_cred = NULL;
7044 struct io_sq_data *sqd = data;
7045 struct io_ring_ctx *ctx;
Xiaoguang Wanga0d92052020-11-12 14:55:59 +08007046 unsigned long timeout = 0;
Xiaoguang Wang08369242020-11-03 14:15:59 +08007047 DEFINE_WAIT(wait);
Jens Axboe6c271ce2019-01-10 11:22:30 -07007048
Jens Axboe28cea78a2020-09-14 10:51:17 -06007049 task_lock(current);
7050 current->files = NULL;
7051 current->nsproxy = NULL;
7052 task_unlock(current);
7053
Jens Axboe69fb2132020-09-14 11:16:23 -06007054 while (!kthread_should_stop()) {
Xiaoguang Wang08369242020-11-03 14:15:59 +08007055 int ret;
7056 bool cap_entries, sqt_spin, needs_sched;
Jens Axboec1edbf52019-11-10 16:56:04 -07007057
7058 /*
Jens Axboe69fb2132020-09-14 11:16:23 -06007059 * Any changes to the sqd lists are synchronized through the
7060 * kthread parking. This synchronizes the thread vs users,
7061 * the users are synchronized on the sqd->ctx_lock.
Jens Axboec1edbf52019-11-10 16:56:04 -07007062 */
Xiaoguang Wang65b2b212020-11-19 17:44:46 +08007063 if (kthread_should_park()) {
Jens Axboe69fb2132020-09-14 11:16:23 -06007064 kthread_parkme();
Xiaoguang Wang65b2b212020-11-19 17:44:46 +08007065 /*
7066 * When sq thread is unparked, in case the previous park operation
7067 * comes from io_put_sq_data(), which means that sq thread is going
7068 * to be stopped, so here needs to have a check.
7069 */
7070 if (kthread_should_stop())
7071 break;
7072 }
Jens Axboe69fb2132020-09-14 11:16:23 -06007073
Xiaoguang Wang08369242020-11-03 14:15:59 +08007074 if (unlikely(!list_empty(&sqd->ctx_new_list))) {
Jens Axboe69fb2132020-09-14 11:16:23 -06007075 io_sqd_init_new(sqd);
Xiaoguang Wang08369242020-11-03 14:15:59 +08007076 timeout = jiffies + sqd->sq_thread_idle;
7077 }
Jens Axboe69fb2132020-09-14 11:16:23 -06007078
Xiaoguang Wang08369242020-11-03 14:15:59 +08007079 sqt_spin = false;
Jens Axboee95eee22020-09-08 09:11:32 -06007080 cap_entries = !list_is_singular(&sqd->ctx_list);
Jens Axboe69fb2132020-09-14 11:16:23 -06007081 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
7082 if (current->cred != ctx->creds) {
7083 if (old_cred)
7084 revert_creds(old_cred);
7085 old_cred = override_creds(ctx->creds);
7086 }
Dennis Zhou91d8f512020-09-16 13:41:05 -07007087 io_sq_thread_associate_blkcg(ctx, &cur_css);
Jens Axboe4ea33a92020-10-15 13:46:44 -06007088#ifdef CONFIG_AUDIT
7089 current->loginuid = ctx->loginuid;
7090 current->sessionid = ctx->sessionid;
7091#endif
Jens Axboe69fb2132020-09-14 11:16:23 -06007092
Xiaoguang Wang08369242020-11-03 14:15:59 +08007093 ret = __io_sq_thread(ctx, cap_entries);
7094 if (!sqt_spin && (ret > 0 || !list_empty(&ctx->iopoll_list)))
7095 sqt_spin = true;
Jens Axboe69fb2132020-09-14 11:16:23 -06007096
Jens Axboe28cea78a2020-09-14 10:51:17 -06007097 io_sq_thread_drop_mm_files();
Jens Axboe6c271ce2019-01-10 11:22:30 -07007098 }
7099
Xiaoguang Wang08369242020-11-03 14:15:59 +08007100 if (sqt_spin || !time_after(jiffies, timeout)) {
Jens Axboec8d1ba52020-09-14 11:07:26 -06007101 io_run_task_work();
Pavel Begunkovd434ab62021-01-11 04:00:30 +00007102 io_sq_thread_drop_mm_files();
Jens Axboec8d1ba52020-09-14 11:07:26 -06007103 cond_resched();
Xiaoguang Wang08369242020-11-03 14:15:59 +08007104 if (sqt_spin)
7105 timeout = jiffies + sqd->sq_thread_idle;
7106 continue;
7107 }
7108
Xiaoguang Wang08369242020-11-03 14:15:59 +08007109 needs_sched = true;
7110 prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE);
7111 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
7112 if ((ctx->flags & IORING_SETUP_IOPOLL) &&
7113 !list_empty_careful(&ctx->iopoll_list)) {
7114 needs_sched = false;
7115 break;
7116 }
7117 if (io_sqring_entries(ctx)) {
7118 needs_sched = false;
7119 break;
7120 }
7121 }
7122
Hao Xu8b28fdf2021-01-31 22:39:04 +08007123 if (needs_sched && !kthread_should_park()) {
Jens Axboe69fb2132020-09-14 11:16:23 -06007124 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
7125 io_ring_set_wakeup_flag(ctx);
Xiaoguang Wang08369242020-11-03 14:15:59 +08007126
Jens Axboe69fb2132020-09-14 11:16:23 -06007127 schedule();
Jens Axboe69fb2132020-09-14 11:16:23 -06007128 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
7129 io_ring_clear_wakeup_flag(ctx);
Jens Axboe6c271ce2019-01-10 11:22:30 -07007130 }
Xiaoguang Wang08369242020-11-03 14:15:59 +08007131
7132 finish_wait(&sqd->wait, &wait);
7133 timeout = jiffies + sqd->sq_thread_idle;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007134 }
7135
Jens Axboe4c6e2772020-07-01 11:29:10 -06007136 io_run_task_work();
Pavel Begunkovd434ab62021-01-11 04:00:30 +00007137 io_sq_thread_drop_mm_files();
Jens Axboeb41e9852020-02-17 09:52:41 -07007138
Dennis Zhou91d8f512020-09-16 13:41:05 -07007139 if (cur_css)
7140 io_sq_thread_unassociate_blkcg();
Jens Axboe69fb2132020-09-14 11:16:23 -06007141 if (old_cred)
7142 revert_creds(old_cred);
Jens Axboe06058632019-04-13 09:26:03 -06007143
Jens Axboe28cea78a2020-09-14 10:51:17 -06007144 task_lock(current);
7145 current->files = old_files;
7146 current->nsproxy = old_nsproxy;
7147 task_unlock(current);
7148
Roman Penyaev2bbcd6d2019-05-16 10:53:57 +02007149 kthread_parkme();
Jens Axboe06058632019-04-13 09:26:03 -06007150
Jens Axboe6c271ce2019-01-10 11:22:30 -07007151 return 0;
7152}
7153
Jens Axboebda52162019-09-24 13:47:15 -06007154struct io_wait_queue {
7155 struct wait_queue_entry wq;
7156 struct io_ring_ctx *ctx;
7157 unsigned to_wait;
7158 unsigned nr_timeouts;
7159};
7160
Pavel Begunkov6c503152021-01-04 20:36:36 +00007161static inline bool io_should_wake(struct io_wait_queue *iowq)
Jens Axboebda52162019-09-24 13:47:15 -06007162{
7163 struct io_ring_ctx *ctx = iowq->ctx;
7164
7165 /*
Brian Gianforcarod195a662019-12-13 03:09:50 -08007166 * Wake up if we have enough events, or if a timeout occurred since we
Jens Axboebda52162019-09-24 13:47:15 -06007167 * started waiting. For timeouts, we always want to return to userspace,
7168 * regardless of event count.
7169 */
Pavel Begunkov6c503152021-01-04 20:36:36 +00007170 return io_cqring_events(ctx) >= iowq->to_wait ||
Jens Axboebda52162019-09-24 13:47:15 -06007171 atomic_read(&ctx->cq_timeouts) != iowq->nr_timeouts;
7172}
7173
7174static int io_wake_function(struct wait_queue_entry *curr, unsigned int mode,
7175 int wake_flags, void *key)
7176{
7177 struct io_wait_queue *iowq = container_of(curr, struct io_wait_queue,
7178 wq);
7179
Pavel Begunkov6c503152021-01-04 20:36:36 +00007180 /*
7181 * Cannot safely flush overflowed CQEs from here, ensure we wake up
7182 * the task, and the next invocation will do it.
7183 */
7184 if (io_should_wake(iowq) || test_bit(0, &iowq->ctx->cq_check_overflow))
7185 return autoremove_wake_function(curr, mode, wake_flags, key);
7186 return -1;
Jens Axboebda52162019-09-24 13:47:15 -06007187}
7188
Jens Axboeaf9c1a42020-09-24 13:32:18 -06007189static int io_run_task_work_sig(void)
7190{
7191 if (io_run_task_work())
7192 return 1;
7193 if (!signal_pending(current))
7194 return 0;
Jens Axboe792ee0f62020-10-22 20:17:18 -06007195 if (test_tsk_thread_flag(current, TIF_NOTIFY_SIGNAL))
7196 return -ERESTARTSYS;
Jens Axboeaf9c1a42020-09-24 13:32:18 -06007197 return -EINTR;
7198}
7199
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00007200/* when returns >0, the caller should retry */
7201static inline int io_cqring_wait_schedule(struct io_ring_ctx *ctx,
7202 struct io_wait_queue *iowq,
7203 signed long *timeout)
7204{
7205 int ret;
7206
7207 /* make sure we run task_work before checking for signals */
7208 ret = io_run_task_work_sig();
7209 if (ret || io_should_wake(iowq))
7210 return ret;
7211 /* let the caller flush overflows, retry */
7212 if (test_bit(0, &ctx->cq_check_overflow))
7213 return 1;
7214
7215 *timeout = schedule_timeout(*timeout);
7216 return !*timeout ? -ETIME : 1;
7217}
7218
Jens Axboe2b188cc2019-01-07 10:46:33 -07007219/*
7220 * Wait until events become available, if we don't already have some. The
7221 * application must reap them itself, as they reside on the shared cq ring.
7222 */
7223static int io_cqring_wait(struct io_ring_ctx *ctx, int min_events,
Hao Xuc73ebb62020-11-03 10:54:37 +08007224 const sigset_t __user *sig, size_t sigsz,
7225 struct __kernel_timespec __user *uts)
Jens Axboe2b188cc2019-01-07 10:46:33 -07007226{
Jens Axboebda52162019-09-24 13:47:15 -06007227 struct io_wait_queue iowq = {
7228 .wq = {
7229 .private = current,
7230 .func = io_wake_function,
7231 .entry = LIST_HEAD_INIT(iowq.wq.entry),
7232 },
7233 .ctx = ctx,
7234 .to_wait = min_events,
7235 };
Hristo Venev75b28af2019-08-26 17:23:46 +00007236 struct io_rings *rings = ctx->rings;
Pavel Begunkovc1d5a222021-02-04 13:51:57 +00007237 signed long timeout = MAX_SCHEDULE_TIMEOUT;
7238 int ret;
Jens Axboe2b188cc2019-01-07 10:46:33 -07007239
Jens Axboeb41e9852020-02-17 09:52:41 -07007240 do {
Pavel Begunkov6c503152021-01-04 20:36:36 +00007241 io_cqring_overflow_flush(ctx, false, NULL, NULL);
7242 if (io_cqring_events(ctx) >= min_events)
Jens Axboeb41e9852020-02-17 09:52:41 -07007243 return 0;
Jens Axboe4c6e2772020-07-01 11:29:10 -06007244 if (!io_run_task_work())
Jens Axboeb41e9852020-02-17 09:52:41 -07007245 break;
Jens Axboeb41e9852020-02-17 09:52:41 -07007246 } while (1);
Jens Axboe2b188cc2019-01-07 10:46:33 -07007247
7248 if (sig) {
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01007249#ifdef CONFIG_COMPAT
7250 if (in_compat_syscall())
7251 ret = set_compat_user_sigmask((const compat_sigset_t __user *)sig,
Oleg Nesterovb7724342019-07-16 16:29:53 -07007252 sigsz);
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01007253 else
7254#endif
Oleg Nesterovb7724342019-07-16 16:29:53 -07007255 ret = set_user_sigmask(sig, sigsz);
Arnd Bergmann9e75ad52019-03-25 15:34:53 +01007256
Jens Axboe2b188cc2019-01-07 10:46:33 -07007257 if (ret)
7258 return ret;
7259 }
7260
Hao Xuc73ebb62020-11-03 10:54:37 +08007261 if (uts) {
Pavel Begunkovc1d5a222021-02-04 13:51:57 +00007262 struct timespec64 ts;
7263
Hao Xuc73ebb62020-11-03 10:54:37 +08007264 if (get_timespec64(&ts, uts))
7265 return -EFAULT;
7266 timeout = timespec64_to_jiffies(&ts);
7267 }
7268
Jens Axboebda52162019-09-24 13:47:15 -06007269 iowq.nr_timeouts = atomic_read(&ctx->cq_timeouts);
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02007270 trace_io_uring_cqring_wait(ctx, min_events);
Jens Axboebda52162019-09-24 13:47:15 -06007271 do {
Pavel Begunkov6c503152021-01-04 20:36:36 +00007272 io_cqring_overflow_flush(ctx, false, NULL, NULL);
Jens Axboebda52162019-09-24 13:47:15 -06007273 prepare_to_wait_exclusive(&ctx->wait, &iowq.wq,
7274 TASK_INTERRUPTIBLE);
Pavel Begunkoveeb60b92021-02-04 13:51:58 +00007275 ret = io_cqring_wait_schedule(ctx, &iowq, &timeout);
7276 finish_wait(&ctx->wait, &iowq.wq);
7277 } while (ret > 0);
Jens Axboebda52162019-09-24 13:47:15 -06007278
Jens Axboeb7db41c2020-07-04 08:55:50 -06007279 restore_saved_sigmask_unless(ret == -EINTR);
Jens Axboe2b188cc2019-01-07 10:46:33 -07007280
Hristo Venev75b28af2019-08-26 17:23:46 +00007281 return READ_ONCE(rings->cq.head) == READ_ONCE(rings->cq.tail) ? ret : 0;
Jens Axboe2b188cc2019-01-07 10:46:33 -07007282}
7283
Jens Axboe6b063142019-01-10 22:13:58 -07007284static void __io_sqe_files_unregister(struct io_ring_ctx *ctx)
7285{
7286#if defined(CONFIG_UNIX)
7287 if (ctx->ring_sock) {
7288 struct sock *sock = ctx->ring_sock->sk;
7289 struct sk_buff *skb;
7290
7291 while ((skb = skb_dequeue(&sock->sk_receive_queue)) != NULL)
7292 kfree_skb(skb);
7293 }
7294#else
7295 int i;
7296
Jens Axboe65e19f52019-10-26 07:20:21 -06007297 for (i = 0; i < ctx->nr_user_files; i++) {
7298 struct file *file;
7299
7300 file = io_file_from_index(ctx, i);
7301 if (file)
7302 fput(file);
7303 }
Jens Axboe6b063142019-01-10 22:13:58 -07007304#endif
7305}
7306
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00007307static void io_rsrc_data_ref_zero(struct percpu_ref *ref)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007308{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007309 struct fixed_rsrc_data *data;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007310
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007311 data = container_of(ref, struct fixed_rsrc_data, refs);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007312 complete(&data->done);
7313}
7314
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007315static inline void io_rsrc_ref_lock(struct io_ring_ctx *ctx)
7316{
7317 spin_lock_bh(&ctx->rsrc_ref_lock);
7318}
7319
7320static inline void io_rsrc_ref_unlock(struct io_ring_ctx *ctx)
7321{
7322 spin_unlock_bh(&ctx->rsrc_ref_lock);
7323}
7324
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00007325static void io_sqe_rsrc_set_node(struct io_ring_ctx *ctx,
7326 struct fixed_rsrc_data *rsrc_data,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007327 struct fixed_rsrc_ref_node *ref_node)
Pavel Begunkov1642b442020-12-30 21:34:14 +00007328{
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007329 io_rsrc_ref_lock(ctx);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007330 rsrc_data->node = ref_node;
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00007331 list_add_tail(&ref_node->node, &ctx->rsrc_ref_list);
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007332 io_rsrc_ref_unlock(ctx);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007333 percpu_ref_get(&rsrc_data->refs);
Pavel Begunkov1642b442020-12-30 21:34:14 +00007334}
7335
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007336static int io_rsrc_ref_quiesce(struct fixed_rsrc_data *data,
7337 struct io_ring_ctx *ctx,
7338 struct fixed_rsrc_ref_node *backup_node)
Jens Axboe6b063142019-01-10 22:13:58 -07007339{
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007340 struct fixed_rsrc_ref_node *ref_node;
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00007341 int ret;
Jens Axboe65e19f52019-10-26 07:20:21 -06007342
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007343 io_rsrc_ref_lock(ctx);
Pavel Begunkov1e5d7702020-11-18 14:56:25 +00007344 ref_node = data->node;
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007345 io_rsrc_ref_unlock(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007346 if (ref_node)
7347 percpu_ref_kill(&ref_node->refs);
7348
7349 percpu_ref_kill(&data->refs);
7350
7351 /* wait for all refs nodes to complete */
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007352 flush_delayed_work(&ctx->rsrc_put_work);
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00007353 do {
7354 ret = wait_for_completion_interruptible(&data->done);
7355 if (!ret)
7356 break;
7357 ret = io_run_task_work_sig();
7358 if (ret < 0) {
7359 percpu_ref_resurrect(&data->refs);
7360 reinit_completion(&data->done);
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00007361 io_sqe_rsrc_set_node(ctx, data, backup_node);
Pavel Begunkov1ffc5422020-12-30 21:34:15 +00007362 return ret;
7363 }
7364 } while (1);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007365
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007366 destroy_fixed_rsrc_ref_node(backup_node);
7367 return 0;
7368}
7369
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007370static struct fixed_rsrc_data *alloc_fixed_rsrc_data(struct io_ring_ctx *ctx)
7371{
7372 struct fixed_rsrc_data *data;
7373
7374 data = kzalloc(sizeof(*data), GFP_KERNEL);
7375 if (!data)
7376 return NULL;
7377
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00007378 if (percpu_ref_init(&data->refs, io_rsrc_data_ref_zero,
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007379 PERCPU_REF_ALLOW_REINIT, GFP_KERNEL)) {
7380 kfree(data);
7381 return NULL;
7382 }
7383 data->ctx = ctx;
7384 init_completion(&data->done);
7385 return data;
7386}
7387
7388static void free_fixed_rsrc_data(struct fixed_rsrc_data *data)
7389{
7390 percpu_ref_exit(&data->refs);
7391 kfree(data->table);
7392 kfree(data);
7393}
7394
Bijan Mottahedehd7954b22021-01-15 17:37:50 +00007395static int io_sqe_files_unregister(struct io_ring_ctx *ctx)
7396{
7397 struct fixed_rsrc_data *data = ctx->file_data;
7398 struct fixed_rsrc_ref_node *backup_node;
7399 unsigned nr_tables, i;
7400 int ret;
7401
7402 if (!data)
7403 return -ENXIO;
7404 backup_node = alloc_fixed_rsrc_ref_node(ctx);
7405 if (!backup_node)
7406 return -ENOMEM;
7407 init_fixed_file_ref_node(ctx, backup_node);
7408
7409 ret = io_rsrc_ref_quiesce(data, ctx, backup_node);
7410 if (ret)
7411 return ret;
7412
Jens Axboe6b063142019-01-10 22:13:58 -07007413 __io_sqe_files_unregister(ctx);
Jens Axboe65e19f52019-10-26 07:20:21 -06007414 nr_tables = DIV_ROUND_UP(ctx->nr_user_files, IORING_MAX_FILES_TABLE);
7415 for (i = 0; i < nr_tables; i++)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007416 kfree(data->table[i].files);
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007417 free_fixed_rsrc_data(data);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007418 ctx->file_data = NULL;
Jens Axboe6b063142019-01-10 22:13:58 -07007419 ctx->nr_user_files = 0;
7420 return 0;
7421}
7422
Jens Axboe534ca6d2020-09-02 13:52:19 -06007423static void io_put_sq_data(struct io_sq_data *sqd)
Jens Axboe6c271ce2019-01-10 11:22:30 -07007424{
Jens Axboe534ca6d2020-09-02 13:52:19 -06007425 if (refcount_dec_and_test(&sqd->refs)) {
Roman Penyaev2bbcd6d2019-05-16 10:53:57 +02007426 /*
7427 * The park is a bit of a work-around, without it we get
7428 * warning spews on shutdown with SQPOLL set and affinity
7429 * set to a single CPU.
7430 */
Jens Axboe534ca6d2020-09-02 13:52:19 -06007431 if (sqd->thread) {
7432 kthread_park(sqd->thread);
7433 kthread_stop(sqd->thread);
7434 }
7435
7436 kfree(sqd);
7437 }
7438}
7439
Jens Axboeaa061652020-09-02 14:50:27 -06007440static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p)
7441{
7442 struct io_ring_ctx *ctx_attach;
7443 struct io_sq_data *sqd;
7444 struct fd f;
7445
7446 f = fdget(p->wq_fd);
7447 if (!f.file)
7448 return ERR_PTR(-ENXIO);
7449 if (f.file->f_op != &io_uring_fops) {
7450 fdput(f);
7451 return ERR_PTR(-EINVAL);
7452 }
7453
7454 ctx_attach = f.file->private_data;
7455 sqd = ctx_attach->sq_data;
7456 if (!sqd) {
7457 fdput(f);
7458 return ERR_PTR(-EINVAL);
7459 }
7460
7461 refcount_inc(&sqd->refs);
7462 fdput(f);
7463 return sqd;
7464}
7465
Jens Axboe534ca6d2020-09-02 13:52:19 -06007466static struct io_sq_data *io_get_sq_data(struct io_uring_params *p)
7467{
7468 struct io_sq_data *sqd;
7469
Jens Axboeaa061652020-09-02 14:50:27 -06007470 if (p->flags & IORING_SETUP_ATTACH_WQ)
7471 return io_attach_sq_data(p);
7472
Jens Axboe534ca6d2020-09-02 13:52:19 -06007473 sqd = kzalloc(sizeof(*sqd), GFP_KERNEL);
7474 if (!sqd)
7475 return ERR_PTR(-ENOMEM);
7476
7477 refcount_set(&sqd->refs, 1);
Jens Axboe69fb2132020-09-14 11:16:23 -06007478 INIT_LIST_HEAD(&sqd->ctx_list);
7479 INIT_LIST_HEAD(&sqd->ctx_new_list);
7480 mutex_init(&sqd->ctx_lock);
7481 mutex_init(&sqd->lock);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007482 init_waitqueue_head(&sqd->wait);
7483 return sqd;
7484}
7485
Jens Axboe69fb2132020-09-14 11:16:23 -06007486static void io_sq_thread_unpark(struct io_sq_data *sqd)
7487 __releases(&sqd->lock)
7488{
7489 if (!sqd->thread)
7490 return;
7491 kthread_unpark(sqd->thread);
7492 mutex_unlock(&sqd->lock);
7493}
7494
7495static void io_sq_thread_park(struct io_sq_data *sqd)
7496 __acquires(&sqd->lock)
7497{
7498 if (!sqd->thread)
7499 return;
7500 mutex_lock(&sqd->lock);
7501 kthread_park(sqd->thread);
7502}
7503
Jens Axboe534ca6d2020-09-02 13:52:19 -06007504static void io_sq_thread_stop(struct io_ring_ctx *ctx)
7505{
7506 struct io_sq_data *sqd = ctx->sq_data;
7507
7508 if (sqd) {
7509 if (sqd->thread) {
7510 /*
7511 * We may arrive here from the error branch in
7512 * io_sq_offload_create() where the kthread is created
7513 * without being waked up, thus wake it up now to make
7514 * sure the wait will complete.
7515 */
7516 wake_up_process(sqd->thread);
7517 wait_for_completion(&ctx->sq_thread_comp);
Jens Axboe69fb2132020-09-14 11:16:23 -06007518
7519 io_sq_thread_park(sqd);
7520 }
7521
7522 mutex_lock(&sqd->ctx_lock);
7523 list_del(&ctx->sqd_list);
Xiaoguang Wang08369242020-11-03 14:15:59 +08007524 io_sqd_update_thread_idle(sqd);
Jens Axboe69fb2132020-09-14 11:16:23 -06007525 mutex_unlock(&sqd->ctx_lock);
7526
Xiaoguang Wang08369242020-11-03 14:15:59 +08007527 if (sqd->thread)
Jens Axboe69fb2132020-09-14 11:16:23 -06007528 io_sq_thread_unpark(sqd);
Jens Axboe534ca6d2020-09-02 13:52:19 -06007529
7530 io_put_sq_data(sqd);
7531 ctx->sq_data = NULL;
Jens Axboe6c271ce2019-01-10 11:22:30 -07007532 }
7533}
7534
Jens Axboe6b063142019-01-10 22:13:58 -07007535static void io_finish_async(struct io_ring_ctx *ctx)
7536{
Jens Axboe6c271ce2019-01-10 11:22:30 -07007537 io_sq_thread_stop(ctx);
7538
Jens Axboe561fb042019-10-24 07:25:42 -06007539 if (ctx->io_wq) {
7540 io_wq_destroy(ctx->io_wq);
7541 ctx->io_wq = NULL;
Jens Axboe6b063142019-01-10 22:13:58 -07007542 }
7543}
7544
7545#if defined(CONFIG_UNIX)
Jens Axboe6b063142019-01-10 22:13:58 -07007546/*
7547 * Ensure the UNIX gc is aware of our file set, so we are certain that
7548 * the io_uring can be safely unregistered on process exit, even if we have
7549 * loops in the file referencing.
7550 */
7551static int __io_sqe_files_scm(struct io_ring_ctx *ctx, int nr, int offset)
7552{
7553 struct sock *sk = ctx->ring_sock->sk;
7554 struct scm_fp_list *fpl;
7555 struct sk_buff *skb;
Jens Axboe08a45172019-10-03 08:11:03 -06007556 int i, nr_files;
Jens Axboe6b063142019-01-10 22:13:58 -07007557
Jens Axboe6b063142019-01-10 22:13:58 -07007558 fpl = kzalloc(sizeof(*fpl), GFP_KERNEL);
7559 if (!fpl)
7560 return -ENOMEM;
7561
7562 skb = alloc_skb(0, GFP_KERNEL);
7563 if (!skb) {
7564 kfree(fpl);
7565 return -ENOMEM;
7566 }
7567
7568 skb->sk = sk;
Jens Axboe6b063142019-01-10 22:13:58 -07007569
Jens Axboe08a45172019-10-03 08:11:03 -06007570 nr_files = 0;
Jens Axboe6b063142019-01-10 22:13:58 -07007571 fpl->user = get_uid(ctx->user);
7572 for (i = 0; i < nr; i++) {
Jens Axboe65e19f52019-10-26 07:20:21 -06007573 struct file *file = io_file_from_index(ctx, i + offset);
7574
7575 if (!file)
Jens Axboe08a45172019-10-03 08:11:03 -06007576 continue;
Jens Axboe65e19f52019-10-26 07:20:21 -06007577 fpl->fp[nr_files] = get_file(file);
Jens Axboe08a45172019-10-03 08:11:03 -06007578 unix_inflight(fpl->user, fpl->fp[nr_files]);
7579 nr_files++;
Jens Axboe6b063142019-01-10 22:13:58 -07007580 }
7581
Jens Axboe08a45172019-10-03 08:11:03 -06007582 if (nr_files) {
7583 fpl->max = SCM_MAX_FD;
7584 fpl->count = nr_files;
7585 UNIXCB(skb).fp = fpl;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007586 skb->destructor = unix_destruct_scm;
Jens Axboe08a45172019-10-03 08:11:03 -06007587 refcount_add(skb->truesize, &sk->sk_wmem_alloc);
7588 skb_queue_head(&sk->sk_receive_queue, skb);
Jens Axboe6b063142019-01-10 22:13:58 -07007589
Jens Axboe08a45172019-10-03 08:11:03 -06007590 for (i = 0; i < nr_files; i++)
7591 fput(fpl->fp[i]);
7592 } else {
7593 kfree_skb(skb);
7594 kfree(fpl);
7595 }
Jens Axboe6b063142019-01-10 22:13:58 -07007596
7597 return 0;
7598}
7599
7600/*
7601 * If UNIX sockets are enabled, fd passing can cause a reference cycle which
7602 * causes regular reference counting to break down. We rely on the UNIX
7603 * garbage collection to take care of this problem for us.
7604 */
7605static int io_sqe_files_scm(struct io_ring_ctx *ctx)
7606{
7607 unsigned left, total;
7608 int ret = 0;
7609
7610 total = 0;
7611 left = ctx->nr_user_files;
7612 while (left) {
7613 unsigned this_files = min_t(unsigned, left, SCM_MAX_FD);
Jens Axboe6b063142019-01-10 22:13:58 -07007614
7615 ret = __io_sqe_files_scm(ctx, this_files, total);
7616 if (ret)
7617 break;
7618 left -= this_files;
7619 total += this_files;
7620 }
7621
7622 if (!ret)
7623 return 0;
7624
7625 while (total < ctx->nr_user_files) {
Jens Axboe65e19f52019-10-26 07:20:21 -06007626 struct file *file = io_file_from_index(ctx, total);
7627
7628 if (file)
7629 fput(file);
Jens Axboe6b063142019-01-10 22:13:58 -07007630 total++;
7631 }
7632
7633 return ret;
7634}
7635#else
7636static int io_sqe_files_scm(struct io_ring_ctx *ctx)
7637{
7638 return 0;
7639}
7640#endif
7641
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007642static int io_sqe_alloc_file_tables(struct fixed_rsrc_data *file_data,
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007643 unsigned nr_tables, unsigned nr_files)
Jens Axboe65e19f52019-10-26 07:20:21 -06007644{
7645 int i;
7646
7647 for (i = 0; i < nr_tables; i++) {
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007648 struct fixed_rsrc_table *table = &file_data->table[i];
Jens Axboe65e19f52019-10-26 07:20:21 -06007649 unsigned this_files;
7650
7651 this_files = min(nr_files, IORING_MAX_FILES_TABLE);
7652 table->files = kcalloc(this_files, sizeof(struct file *),
7653 GFP_KERNEL);
7654 if (!table->files)
7655 break;
7656 nr_files -= this_files;
7657 }
7658
7659 if (i == nr_tables)
7660 return 0;
7661
7662 for (i = 0; i < nr_tables; i++) {
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007663 struct fixed_rsrc_table *table = &file_data->table[i];
Jens Axboe65e19f52019-10-26 07:20:21 -06007664 kfree(table->files);
7665 }
7666 return 1;
7667}
7668
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007669static void io_ring_file_put(struct io_ring_ctx *ctx, struct io_rsrc_put *prsrc)
Jens Axboec3a31e62019-10-03 13:59:56 -06007670{
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007671 struct file *file = prsrc->file;
Jens Axboec3a31e62019-10-03 13:59:56 -06007672#if defined(CONFIG_UNIX)
Jens Axboec3a31e62019-10-03 13:59:56 -06007673 struct sock *sock = ctx->ring_sock->sk;
7674 struct sk_buff_head list, *head = &sock->sk_receive_queue;
7675 struct sk_buff *skb;
7676 int i;
7677
7678 __skb_queue_head_init(&list);
7679
7680 /*
7681 * Find the skb that holds this file in its SCM_RIGHTS. When found,
7682 * remove this entry and rearrange the file array.
7683 */
7684 skb = skb_dequeue(head);
7685 while (skb) {
7686 struct scm_fp_list *fp;
7687
7688 fp = UNIXCB(skb).fp;
7689 for (i = 0; i < fp->count; i++) {
7690 int left;
7691
7692 if (fp->fp[i] != file)
7693 continue;
7694
7695 unix_notinflight(fp->user, fp->fp[i]);
7696 left = fp->count - 1 - i;
7697 if (left) {
7698 memmove(&fp->fp[i], &fp->fp[i + 1],
7699 left * sizeof(struct file *));
7700 }
7701 fp->count--;
7702 if (!fp->count) {
7703 kfree_skb(skb);
7704 skb = NULL;
7705 } else {
7706 __skb_queue_tail(&list, skb);
7707 }
7708 fput(file);
7709 file = NULL;
7710 break;
7711 }
7712
7713 if (!file)
7714 break;
7715
7716 __skb_queue_tail(&list, skb);
7717
7718 skb = skb_dequeue(head);
7719 }
7720
7721 if (skb_peek(&list)) {
7722 spin_lock_irq(&head->lock);
7723 while ((skb = __skb_dequeue(&list)) != NULL)
7724 __skb_queue_tail(head, skb);
7725 spin_unlock_irq(&head->lock);
7726 }
7727#else
Jens Axboe05f3fb32019-12-09 11:22:50 -07007728 fput(file);
Jens Axboec3a31e62019-10-03 13:59:56 -06007729#endif
7730}
7731
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007732static void __io_rsrc_put_work(struct fixed_rsrc_ref_node *ref_node)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007733{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007734 struct fixed_rsrc_data *rsrc_data = ref_node->rsrc_data;
7735 struct io_ring_ctx *ctx = rsrc_data->ctx;
7736 struct io_rsrc_put *prsrc, *tmp;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007737
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007738 list_for_each_entry_safe(prsrc, tmp, &ref_node->rsrc_list, list) {
7739 list_del(&prsrc->list);
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007740 ref_node->rsrc_put(ctx, prsrc);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007741 kfree(prsrc);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007742 }
7743
Xiaoguang Wang05589552020-03-31 14:05:18 +08007744 percpu_ref_exit(&ref_node->refs);
7745 kfree(ref_node);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007746 percpu_ref_put(&rsrc_data->refs);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007747}
7748
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007749static void io_rsrc_put_work(struct work_struct *work)
Jens Axboe4a38aed22020-05-14 17:21:15 -06007750{
7751 struct io_ring_ctx *ctx;
7752 struct llist_node *node;
7753
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007754 ctx = container_of(work, struct io_ring_ctx, rsrc_put_work.work);
7755 node = llist_del_all(&ctx->rsrc_put_llist);
Jens Axboe4a38aed22020-05-14 17:21:15 -06007756
7757 while (node) {
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007758 struct fixed_rsrc_ref_node *ref_node;
Jens Axboe4a38aed22020-05-14 17:21:15 -06007759 struct llist_node *next = node->next;
7760
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007761 ref_node = llist_entry(node, struct fixed_rsrc_ref_node, llist);
7762 __io_rsrc_put_work(ref_node);
Jens Axboe4a38aed22020-05-14 17:21:15 -06007763 node = next;
7764 }
7765}
7766
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007767static struct file **io_fixed_file_slot(struct fixed_rsrc_data *file_data,
7768 unsigned i)
7769{
7770 struct fixed_rsrc_table *table;
7771
7772 table = &file_data->table[i >> IORING_FILE_TABLE_SHIFT];
7773 return &table->files[i & IORING_FILE_TABLE_MASK];
7774}
7775
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00007776static void io_rsrc_node_ref_zero(struct percpu_ref *ref)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007777{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007778 struct fixed_rsrc_ref_node *ref_node;
7779 struct fixed_rsrc_data *data;
Jens Axboe4a38aed22020-05-14 17:21:15 -06007780 struct io_ring_ctx *ctx;
Pavel Begunkove2978222020-11-18 14:56:26 +00007781 bool first_add = false;
Jens Axboe4a38aed22020-05-14 17:21:15 -06007782 int delay = HZ;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007783
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007784 ref_node = container_of(ref, struct fixed_rsrc_ref_node, refs);
7785 data = ref_node->rsrc_data;
Pavel Begunkove2978222020-11-18 14:56:26 +00007786 ctx = data->ctx;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007787
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007788 io_rsrc_ref_lock(ctx);
Pavel Begunkove2978222020-11-18 14:56:26 +00007789 ref_node->done = true;
7790
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00007791 while (!list_empty(&ctx->rsrc_ref_list)) {
7792 ref_node = list_first_entry(&ctx->rsrc_ref_list,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007793 struct fixed_rsrc_ref_node, node);
Pavel Begunkove2978222020-11-18 14:56:26 +00007794 /* recycle ref nodes in order */
7795 if (!ref_node->done)
7796 break;
7797 list_del(&ref_node->node);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007798 first_add |= llist_add(&ref_node->llist, &ctx->rsrc_put_llist);
Pavel Begunkove2978222020-11-18 14:56:26 +00007799 }
Bijan Mottahedeh2a63b2d2021-01-15 17:37:47 +00007800 io_rsrc_ref_unlock(ctx);
Pavel Begunkove2978222020-11-18 14:56:26 +00007801
7802 if (percpu_ref_is_dying(&data->refs))
Jens Axboe4a38aed22020-05-14 17:21:15 -06007803 delay = 0;
7804
Jens Axboe4a38aed22020-05-14 17:21:15 -06007805 if (!delay)
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007806 mod_delayed_work(system_wq, &ctx->rsrc_put_work, 0);
Jens Axboe4a38aed22020-05-14 17:21:15 -06007807 else if (first_add)
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007808 queue_delayed_work(system_wq, &ctx->rsrc_put_work, delay);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007809}
7810
Bijan Mottahedeh68025352021-01-15 17:37:48 +00007811static struct fixed_rsrc_ref_node *alloc_fixed_rsrc_ref_node(
Xiaoguang Wang05589552020-03-31 14:05:18 +08007812 struct io_ring_ctx *ctx)
7813{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007814 struct fixed_rsrc_ref_node *ref_node;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007815
7816 ref_node = kzalloc(sizeof(*ref_node), GFP_KERNEL);
7817 if (!ref_node)
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007818 return NULL;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007819
Bijan Mottahedeh00835dc2021-01-15 17:37:52 +00007820 if (percpu_ref_init(&ref_node->refs, io_rsrc_node_ref_zero,
Xiaoguang Wang05589552020-03-31 14:05:18 +08007821 0, GFP_KERNEL)) {
7822 kfree(ref_node);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007823 return NULL;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007824 }
7825 INIT_LIST_HEAD(&ref_node->node);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007826 INIT_LIST_HEAD(&ref_node->rsrc_list);
Bijan Mottahedeh68025352021-01-15 17:37:48 +00007827 ref_node->done = false;
7828 return ref_node;
7829}
7830
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007831static void init_fixed_file_ref_node(struct io_ring_ctx *ctx,
7832 struct fixed_rsrc_ref_node *ref_node)
Bijan Mottahedeh68025352021-01-15 17:37:48 +00007833{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007834 ref_node->rsrc_data = ctx->file_data;
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007835 ref_node->rsrc_put = io_ring_file_put;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007836}
7837
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007838static void destroy_fixed_rsrc_ref_node(struct fixed_rsrc_ref_node *ref_node)
Xiaoguang Wang05589552020-03-31 14:05:18 +08007839{
7840 percpu_ref_exit(&ref_node->refs);
7841 kfree(ref_node);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007842}
7843
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007844
Jens Axboe05f3fb32019-12-09 11:22:50 -07007845static int io_sqe_files_register(struct io_ring_ctx *ctx, void __user *arg,
7846 unsigned nr_args)
7847{
7848 __s32 __user *fds = (__s32 __user *) arg;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007849 unsigned nr_tables, i;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007850 struct file *file;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007851 int fd, ret = -ENOMEM;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007852 struct fixed_rsrc_ref_node *ref_node;
7853 struct fixed_rsrc_data *file_data;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007854
7855 if (ctx->file_data)
7856 return -EBUSY;
7857 if (!nr_args)
7858 return -EINVAL;
7859 if (nr_args > IORING_MAX_FIXED_FILES)
7860 return -EMFILE;
7861
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007862 file_data = alloc_fixed_rsrc_data(ctx);
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007863 if (!file_data)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007864 return -ENOMEM;
Dan Carpenter13770a72021-02-01 15:23:42 +03007865 ctx->file_data = file_data;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007866
7867 nr_tables = DIV_ROUND_UP(nr_args, IORING_MAX_FILES_TABLE);
Colin Ian King035fbaf2020-10-12 15:03:41 +01007868 file_data->table = kcalloc(nr_tables, sizeof(*file_data->table),
Pavel Begunkov5398ae62020-10-10 18:34:14 +01007869 GFP_KERNEL);
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007870 if (!file_data->table)
7871 goto out_free;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007872
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007873 if (io_sqe_alloc_file_tables(file_data, nr_tables, nr_args))
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007874 goto out_free;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007875
7876 for (i = 0; i < nr_args; i++, ctx->nr_user_files++) {
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007877 if (copy_from_user(&fd, &fds[i], sizeof(fd))) {
7878 ret = -EFAULT;
7879 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007880 }
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007881 /* allow sparse sets */
7882 if (fd == -1)
7883 continue;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007884
Jens Axboe05f3fb32019-12-09 11:22:50 -07007885 file = fget(fd);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007886 ret = -EBADF;
7887 if (!file)
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007888 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007889
7890 /*
7891 * Don't allow io_uring instances to be registered. If UNIX
7892 * isn't enabled, then this causes a reference cycle and this
7893 * instance can never get freed. If UNIX is enabled we'll
7894 * handle it just fine, but there's still no point in allowing
7895 * a ring fd as it doesn't support regular read/write anyway.
7896 */
7897 if (file->f_op == &io_uring_fops) {
7898 fput(file);
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007899 goto out_fput;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007900 }
Pavel Begunkovea64ec022021-02-04 13:52:07 +00007901 *io_fixed_file_slot(file_data, i) = file;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007902 }
7903
Jens Axboe05f3fb32019-12-09 11:22:50 -07007904 ret = io_sqe_files_scm(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007905 if (ret) {
Jens Axboe05f3fb32019-12-09 11:22:50 -07007906 io_sqe_files_unregister(ctx);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007907 return ret;
7908 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07007909
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007910 ref_node = alloc_fixed_rsrc_ref_node(ctx);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007911 if (!ref_node) {
Xiaoguang Wang05589552020-03-31 14:05:18 +08007912 io_sqe_files_unregister(ctx);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00007913 return -ENOMEM;
Xiaoguang Wang05589552020-03-31 14:05:18 +08007914 }
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00007915 init_fixed_file_ref_node(ctx, ref_node);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007916
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00007917 io_sqe_rsrc_set_node(ctx, file_data, ref_node);
Jens Axboe05f3fb32019-12-09 11:22:50 -07007918 return ret;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007919out_fput:
7920 for (i = 0; i < ctx->nr_user_files; i++) {
7921 file = io_file_from_index(ctx, i);
7922 if (file)
7923 fput(file);
7924 }
7925 for (i = 0; i < nr_tables; i++)
7926 kfree(file_data->table[i].files);
7927 ctx->nr_user_files = 0;
Pavel Begunkov600cf3f2020-10-10 18:34:15 +01007928out_free:
Bijan Mottahedeh1ad555c2021-01-15 17:37:51 +00007929 free_fixed_rsrc_data(ctx->file_data);
Jens Axboe55cbc252020-10-14 07:35:57 -06007930 ctx->file_data = NULL;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007931 return ret;
7932}
7933
Jens Axboec3a31e62019-10-03 13:59:56 -06007934static int io_sqe_file_register(struct io_ring_ctx *ctx, struct file *file,
7935 int index)
7936{
7937#if defined(CONFIG_UNIX)
7938 struct sock *sock = ctx->ring_sock->sk;
7939 struct sk_buff_head *head = &sock->sk_receive_queue;
7940 struct sk_buff *skb;
7941
7942 /*
7943 * See if we can merge this file into an existing skb SCM_RIGHTS
7944 * file set. If there's no room, fall back to allocating a new skb
7945 * and filling it in.
7946 */
7947 spin_lock_irq(&head->lock);
7948 skb = skb_peek(head);
7949 if (skb) {
7950 struct scm_fp_list *fpl = UNIXCB(skb).fp;
7951
7952 if (fpl->count < SCM_MAX_FD) {
7953 __skb_unlink(skb, head);
7954 spin_unlock_irq(&head->lock);
7955 fpl->fp[fpl->count] = get_file(file);
7956 unix_inflight(fpl->user, fpl->fp[fpl->count]);
7957 fpl->count++;
7958 spin_lock_irq(&head->lock);
7959 __skb_queue_head(head, skb);
7960 } else {
7961 skb = NULL;
7962 }
7963 }
7964 spin_unlock_irq(&head->lock);
7965
7966 if (skb) {
7967 fput(file);
7968 return 0;
7969 }
7970
7971 return __io_sqe_files_scm(ctx, 1, index);
7972#else
7973 return 0;
7974#endif
7975}
7976
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007977static int io_queue_rsrc_removal(struct fixed_rsrc_data *data, void *rsrc)
Jens Axboe05f3fb32019-12-09 11:22:50 -07007978{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007979 struct io_rsrc_put *prsrc;
7980 struct fixed_rsrc_ref_node *ref_node = data->node;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007981
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007982 prsrc = kzalloc(sizeof(*prsrc), GFP_KERNEL);
7983 if (!prsrc)
Hillf Dantona5318d32020-03-23 17:47:15 +08007984 return -ENOMEM;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007985
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007986 prsrc->rsrc = rsrc;
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007987 list_add(&prsrc->list, &ref_node->rsrc_list);
Xiaoguang Wang05589552020-03-31 14:05:18 +08007988
Hillf Dantona5318d32020-03-23 17:47:15 +08007989 return 0;
Jens Axboe05f3fb32019-12-09 11:22:50 -07007990}
7991
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007992static inline int io_queue_file_removal(struct fixed_rsrc_data *data,
7993 struct file *file)
7994{
Bijan Mottahedeh50238532021-01-15 17:37:45 +00007995 return io_queue_rsrc_removal(data, (void *)file);
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007996}
7997
Jens Axboe05f3fb32019-12-09 11:22:50 -07007998static int __io_sqe_files_update(struct io_ring_ctx *ctx,
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00007999 struct io_uring_rsrc_update *up,
Jens Axboe05f3fb32019-12-09 11:22:50 -07008000 unsigned nr_args)
8001{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00008002 struct fixed_rsrc_data *data = ctx->file_data;
8003 struct fixed_rsrc_ref_node *ref_node;
Pavel Begunkovea64ec022021-02-04 13:52:07 +00008004 struct file *file, **file_slot;
Jens Axboec3a31e62019-10-03 13:59:56 -06008005 __s32 __user *fds;
8006 int fd, i, err;
8007 __u32 done;
Xiaoguang Wang05589552020-03-31 14:05:18 +08008008 bool needs_switch = false;
Jens Axboec3a31e62019-10-03 13:59:56 -06008009
Jens Axboe05f3fb32019-12-09 11:22:50 -07008010 if (check_add_overflow(up->offset, nr_args, &done))
Jens Axboec3a31e62019-10-03 13:59:56 -06008011 return -EOVERFLOW;
8012 if (done > ctx->nr_user_files)
8013 return -EINVAL;
8014
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00008015 ref_node = alloc_fixed_rsrc_ref_node(ctx);
Matthew Wilcox (Oracle)3e2224c2021-01-06 16:09:26 +00008016 if (!ref_node)
8017 return -ENOMEM;
Pavel Begunkovbc9744c2021-01-15 17:37:49 +00008018 init_fixed_file_ref_node(ctx, ref_node);
Xiaoguang Wang05589552020-03-31 14:05:18 +08008019
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00008020 fds = u64_to_user_ptr(up->data);
Pavel Begunkov67973b92021-01-26 13:51:09 +00008021 for (done = 0; done < nr_args; done++) {
Jens Axboec3a31e62019-10-03 13:59:56 -06008022 err = 0;
8023 if (copy_from_user(&fd, &fds[done], sizeof(fd))) {
8024 err = -EFAULT;
8025 break;
8026 }
noah4e0377a2021-01-26 15:23:28 -05008027 if (fd == IORING_REGISTER_FILES_SKIP)
8028 continue;
8029
Pavel Begunkov67973b92021-01-26 13:51:09 +00008030 i = array_index_nospec(up->offset + done, ctx->nr_user_files);
Pavel Begunkovea64ec022021-02-04 13:52:07 +00008031 file_slot = io_fixed_file_slot(ctx->file_data, i);
8032
8033 if (*file_slot) {
8034 err = io_queue_file_removal(data, *file_slot);
Hillf Dantona5318d32020-03-23 17:47:15 +08008035 if (err)
8036 break;
Pavel Begunkovea64ec022021-02-04 13:52:07 +00008037 *file_slot = NULL;
Xiaoguang Wang05589552020-03-31 14:05:18 +08008038 needs_switch = true;
Jens Axboec3a31e62019-10-03 13:59:56 -06008039 }
8040 if (fd != -1) {
Jens Axboec3a31e62019-10-03 13:59:56 -06008041 file = fget(fd);
8042 if (!file) {
8043 err = -EBADF;
8044 break;
8045 }
8046 /*
8047 * Don't allow io_uring instances to be registered. If
8048 * UNIX isn't enabled, then this causes a reference
8049 * cycle and this instance can never get freed. If UNIX
8050 * is enabled we'll handle it just fine, but there's
8051 * still no point in allowing a ring fd as it doesn't
8052 * support regular read/write anyway.
8053 */
8054 if (file->f_op == &io_uring_fops) {
8055 fput(file);
8056 err = -EBADF;
8057 break;
8058 }
Jens Axboee68a3ff2021-02-11 07:45:08 -07008059 *file_slot = file;
Jens Axboec3a31e62019-10-03 13:59:56 -06008060 err = io_sqe_file_register(ctx, file, i);
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00008061 if (err) {
Jens Axboee68a3ff2021-02-11 07:45:08 -07008062 *file_slot = NULL;
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00008063 fput(file);
Jens Axboec3a31e62019-10-03 13:59:56 -06008064 break;
Yang Yingliangf3bd9da2020-07-09 10:11:41 +00008065 }
Jens Axboec3a31e62019-10-03 13:59:56 -06008066 }
Jens Axboe05f3fb32019-12-09 11:22:50 -07008067 }
8068
Xiaoguang Wang05589552020-03-31 14:05:18 +08008069 if (needs_switch) {
Pavel Begunkovb2e96852020-10-10 18:34:16 +01008070 percpu_ref_kill(&data->node->refs);
Bijan Mottahedehd67d2262021-01-15 17:37:46 +00008071 io_sqe_rsrc_set_node(ctx, data, ref_node);
Xiaoguang Wang05589552020-03-31 14:05:18 +08008072 } else
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00008073 destroy_fixed_rsrc_ref_node(ref_node);
Jens Axboec3a31e62019-10-03 13:59:56 -06008074
8075 return done ? done : err;
8076}
Xiaoguang Wang05589552020-03-31 14:05:18 +08008077
Jens Axboe05f3fb32019-12-09 11:22:50 -07008078static int io_sqe_files_update(struct io_ring_ctx *ctx, void __user *arg,
8079 unsigned nr_args)
8080{
Bijan Mottahedeh269bbe52021-01-15 17:37:44 +00008081 struct io_uring_rsrc_update up;
Jens Axboe05f3fb32019-12-09 11:22:50 -07008082
8083 if (!ctx->file_data)
8084 return -ENXIO;
8085 if (!nr_args)
8086 return -EINVAL;
8087 if (copy_from_user(&up, arg, sizeof(up)))
8088 return -EFAULT;
8089 if (up.resv)
8090 return -EINVAL;
8091
8092 return __io_sqe_files_update(ctx, &up, nr_args);
8093}
Jens Axboec3a31e62019-10-03 13:59:56 -06008094
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00008095static struct io_wq_work *io_free_work(struct io_wq_work *work)
Jens Axboe7d723062019-11-12 22:31:31 -07008096{
8097 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
8098
Pavel Begunkov5280f7e2021-02-04 13:52:08 +00008099 req = io_put_req_find_next(req);
8100 return req ? &req->work : NULL;
Jens Axboe7d723062019-11-12 22:31:31 -07008101}
8102
Pavel Begunkov24369c22020-01-28 03:15:48 +03008103static int io_init_wq_offload(struct io_ring_ctx *ctx,
8104 struct io_uring_params *p)
8105{
8106 struct io_wq_data data;
8107 struct fd f;
8108 struct io_ring_ctx *ctx_attach;
8109 unsigned int concurrency;
8110 int ret = 0;
8111
8112 data.user = ctx->user;
Pavel Begunkove9fd9392020-03-04 16:14:12 +03008113 data.free_work = io_free_work;
Pavel Begunkovf5fa38c2020-06-08 21:08:20 +03008114 data.do_work = io_wq_submit_work;
Pavel Begunkov24369c22020-01-28 03:15:48 +03008115
8116 if (!(p->flags & IORING_SETUP_ATTACH_WQ)) {
8117 /* Do QD, or 4 * CPUS, whatever is smallest */
8118 concurrency = min(ctx->sq_entries, 4 * num_online_cpus());
8119
8120 ctx->io_wq = io_wq_create(concurrency, &data);
8121 if (IS_ERR(ctx->io_wq)) {
8122 ret = PTR_ERR(ctx->io_wq);
8123 ctx->io_wq = NULL;
8124 }
8125 return ret;
8126 }
8127
8128 f = fdget(p->wq_fd);
8129 if (!f.file)
8130 return -EBADF;
8131
8132 if (f.file->f_op != &io_uring_fops) {
8133 ret = -EINVAL;
8134 goto out_fput;
8135 }
8136
8137 ctx_attach = f.file->private_data;
8138 /* @io_wq is protected by holding the fd */
8139 if (!io_wq_get(ctx_attach->io_wq, &data)) {
8140 ret = -EINVAL;
8141 goto out_fput;
8142 }
8143
8144 ctx->io_wq = ctx_attach->io_wq;
8145out_fput:
8146 fdput(f);
8147 return ret;
8148}
8149
Jens Axboe0f212202020-09-13 13:09:39 -06008150static int io_uring_alloc_task_context(struct task_struct *task)
8151{
8152 struct io_uring_task *tctx;
Jens Axboed8a6df12020-10-15 16:24:45 -06008153 int ret;
Jens Axboe0f212202020-09-13 13:09:39 -06008154
8155 tctx = kmalloc(sizeof(*tctx), GFP_KERNEL);
8156 if (unlikely(!tctx))
8157 return -ENOMEM;
8158
Jens Axboed8a6df12020-10-15 16:24:45 -06008159 ret = percpu_counter_init(&tctx->inflight, 0, GFP_KERNEL);
8160 if (unlikely(ret)) {
8161 kfree(tctx);
8162 return ret;
8163 }
8164
Jens Axboe0f212202020-09-13 13:09:39 -06008165 xa_init(&tctx->xa);
8166 init_waitqueue_head(&tctx->wait);
8167 tctx->last = NULL;
Jens Axboefdaf0832020-10-30 09:37:30 -06008168 atomic_set(&tctx->in_idle, 0);
8169 tctx->sqpoll = false;
Jens Axboe500a3732020-10-15 17:38:03 -06008170 io_init_identity(&tctx->__identity);
8171 tctx->identity = &tctx->__identity;
Jens Axboe0f212202020-09-13 13:09:39 -06008172 task->io_uring = tctx;
Jens Axboe7cbf1722021-02-10 00:03:20 +00008173 spin_lock_init(&tctx->task_lock);
8174 INIT_WQ_LIST(&tctx->task_list);
8175 tctx->task_state = 0;
8176 init_task_work(&tctx->task_work, tctx_task_work);
Jens Axboe0f212202020-09-13 13:09:39 -06008177 return 0;
8178}
8179
8180void __io_uring_free(struct task_struct *tsk)
8181{
8182 struct io_uring_task *tctx = tsk->io_uring;
8183
8184 WARN_ON_ONCE(!xa_empty(&tctx->xa));
Jens Axboe500a3732020-10-15 17:38:03 -06008185 WARN_ON_ONCE(refcount_read(&tctx->identity->count) != 1);
8186 if (tctx->identity != &tctx->__identity)
8187 kfree(tctx->identity);
Jens Axboed8a6df12020-10-15 16:24:45 -06008188 percpu_counter_destroy(&tctx->inflight);
Jens Axboe0f212202020-09-13 13:09:39 -06008189 kfree(tctx);
8190 tsk->io_uring = NULL;
8191}
8192
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02008193static int io_sq_offload_create(struct io_ring_ctx *ctx,
8194 struct io_uring_params *p)
Jens Axboe2b188cc2019-01-07 10:46:33 -07008195{
8196 int ret;
8197
Jens Axboe6c271ce2019-01-10 11:22:30 -07008198 if (ctx->flags & IORING_SETUP_SQPOLL) {
Jens Axboe534ca6d2020-09-02 13:52:19 -06008199 struct io_sq_data *sqd;
8200
Jens Axboe3ec482d2019-04-08 10:51:01 -06008201 ret = -EPERM;
Jens Axboece59fc62020-09-02 13:28:09 -06008202 if (!capable(CAP_SYS_ADMIN) && !capable(CAP_SYS_NICE))
Jens Axboe3ec482d2019-04-08 10:51:01 -06008203 goto err;
8204
Jens Axboe534ca6d2020-09-02 13:52:19 -06008205 sqd = io_get_sq_data(p);
8206 if (IS_ERR(sqd)) {
8207 ret = PTR_ERR(sqd);
8208 goto err;
8209 }
Jens Axboe69fb2132020-09-14 11:16:23 -06008210
Jens Axboe534ca6d2020-09-02 13:52:19 -06008211 ctx->sq_data = sqd;
Jens Axboe69fb2132020-09-14 11:16:23 -06008212 io_sq_thread_park(sqd);
8213 mutex_lock(&sqd->ctx_lock);
8214 list_add(&ctx->sqd_list, &sqd->ctx_new_list);
8215 mutex_unlock(&sqd->ctx_lock);
8216 io_sq_thread_unpark(sqd);
Jens Axboe534ca6d2020-09-02 13:52:19 -06008217
Jens Axboe917257d2019-04-13 09:28:55 -06008218 ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle);
8219 if (!ctx->sq_thread_idle)
8220 ctx->sq_thread_idle = HZ;
8221
Jens Axboeaa061652020-09-02 14:50:27 -06008222 if (sqd->thread)
8223 goto done;
8224
Jens Axboe6c271ce2019-01-10 11:22:30 -07008225 if (p->flags & IORING_SETUP_SQ_AFF) {
Jens Axboe44a9bd12019-05-14 20:00:30 -06008226 int cpu = p->sq_thread_cpu;
Jens Axboe6c271ce2019-01-10 11:22:30 -07008227
Jens Axboe917257d2019-04-13 09:28:55 -06008228 ret = -EINVAL;
Jens Axboe44a9bd12019-05-14 20:00:30 -06008229 if (cpu >= nr_cpu_ids)
8230 goto err;
Shenghui Wang7889f442019-05-07 16:03:19 +08008231 if (!cpu_online(cpu))
Jens Axboe917257d2019-04-13 09:28:55 -06008232 goto err;
8233
Jens Axboe69fb2132020-09-14 11:16:23 -06008234 sqd->thread = kthread_create_on_cpu(io_sq_thread, sqd,
Jens Axboe534ca6d2020-09-02 13:52:19 -06008235 cpu, "io_uring-sq");
Jens Axboe6c271ce2019-01-10 11:22:30 -07008236 } else {
Jens Axboe69fb2132020-09-14 11:16:23 -06008237 sqd->thread = kthread_create(io_sq_thread, sqd,
Jens Axboe6c271ce2019-01-10 11:22:30 -07008238 "io_uring-sq");
8239 }
Jens Axboe534ca6d2020-09-02 13:52:19 -06008240 if (IS_ERR(sqd->thread)) {
8241 ret = PTR_ERR(sqd->thread);
8242 sqd->thread = NULL;
Jens Axboe6c271ce2019-01-10 11:22:30 -07008243 goto err;
8244 }
Jens Axboe534ca6d2020-09-02 13:52:19 -06008245 ret = io_uring_alloc_task_context(sqd->thread);
Jens Axboe0f212202020-09-13 13:09:39 -06008246 if (ret)
8247 goto err;
Jens Axboe6c271ce2019-01-10 11:22:30 -07008248 } else if (p->flags & IORING_SETUP_SQ_AFF) {
8249 /* Can't have SQ_AFF without SQPOLL */
8250 ret = -EINVAL;
8251 goto err;
8252 }
8253
Jens Axboeaa061652020-09-02 14:50:27 -06008254done:
Pavel Begunkov24369c22020-01-28 03:15:48 +03008255 ret = io_init_wq_offload(ctx, p);
8256 if (ret)
Jens Axboe2b188cc2019-01-07 10:46:33 -07008257 goto err;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008258
8259 return 0;
8260err:
Jens Axboe54a91f32019-09-10 09:15:04 -06008261 io_finish_async(ctx);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008262 return ret;
8263}
8264
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02008265static void io_sq_offload_start(struct io_ring_ctx *ctx)
8266{
Jens Axboe534ca6d2020-09-02 13:52:19 -06008267 struct io_sq_data *sqd = ctx->sq_data;
8268
8269 if ((ctx->flags & IORING_SETUP_SQPOLL) && sqd->thread)
8270 wake_up_process(sqd->thread);
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02008271}
8272
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008273static inline void __io_unaccount_mem(struct user_struct *user,
8274 unsigned long nr_pages)
Jens Axboe2b188cc2019-01-07 10:46:33 -07008275{
8276 atomic_long_sub(nr_pages, &user->locked_vm);
8277}
8278
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008279static inline int __io_account_mem(struct user_struct *user,
8280 unsigned long nr_pages)
Jens Axboe2b188cc2019-01-07 10:46:33 -07008281{
8282 unsigned long page_limit, cur_pages, new_pages;
8283
8284 /* Don't allow more pages than we can safely lock */
8285 page_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
8286
8287 do {
8288 cur_pages = atomic_long_read(&user->locked_vm);
8289 new_pages = cur_pages + nr_pages;
8290 if (new_pages > page_limit)
8291 return -ENOMEM;
8292 } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages,
8293 new_pages) != cur_pages);
8294
8295 return 0;
8296}
8297
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008298static void io_unaccount_mem(struct io_ring_ctx *ctx, unsigned long nr_pages)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008299{
Bijan Mottahedehaad5d8d2020-06-16 16:36:08 -07008300 if (ctx->limit_mem)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008301 __io_unaccount_mem(ctx->user, nr_pages);
Bijan Mottahedeh30975822020-06-16 16:36:09 -07008302
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008303 if (ctx->mm_account)
8304 atomic64_sub(nr_pages, &ctx->mm_account->pinned_vm);
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008305}
8306
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008307static int io_account_mem(struct io_ring_ctx *ctx, unsigned long nr_pages)
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008308{
Bijan Mottahedeh30975822020-06-16 16:36:09 -07008309 int ret;
8310
8311 if (ctx->limit_mem) {
8312 ret = __io_account_mem(ctx->user, nr_pages);
8313 if (ret)
8314 return ret;
8315 }
8316
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008317 if (ctx->mm_account)
8318 atomic64_add(nr_pages, &ctx->mm_account->pinned_vm);
Bijan Mottahedeha087e2b2020-06-16 16:36:07 -07008319
8320 return 0;
8321}
8322
Jens Axboe2b188cc2019-01-07 10:46:33 -07008323static void io_mem_free(void *ptr)
8324{
Mark Rutland52e04ef2019-04-30 17:30:21 +01008325 struct page *page;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008326
Mark Rutland52e04ef2019-04-30 17:30:21 +01008327 if (!ptr)
8328 return;
8329
8330 page = virt_to_head_page(ptr);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008331 if (put_page_testzero(page))
8332 free_compound_page(page);
8333}
8334
8335static void *io_mem_alloc(size_t size)
8336{
8337 gfp_t gfp_flags = GFP_KERNEL | __GFP_ZERO | __GFP_NOWARN | __GFP_COMP |
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008338 __GFP_NORETRY | __GFP_ACCOUNT;
Jens Axboe2b188cc2019-01-07 10:46:33 -07008339
8340 return (void *) __get_free_pages(gfp_flags, get_order(size));
8341}
8342
Hristo Venev75b28af2019-08-26 17:23:46 +00008343static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries,
8344 size_t *sq_offset)
8345{
8346 struct io_rings *rings;
8347 size_t off, sq_array_size;
8348
8349 off = struct_size(rings, cqes, cq_entries);
8350 if (off == SIZE_MAX)
8351 return SIZE_MAX;
8352
8353#ifdef CONFIG_SMP
8354 off = ALIGN(off, SMP_CACHE_BYTES);
8355 if (off == 0)
8356 return SIZE_MAX;
8357#endif
8358
Dmitry Vyukovb36200f2020-07-11 11:31:11 +02008359 if (sq_offset)
8360 *sq_offset = off;
8361
Hristo Venev75b28af2019-08-26 17:23:46 +00008362 sq_array_size = array_size(sizeof(u32), sq_entries);
8363 if (sq_array_size == SIZE_MAX)
8364 return SIZE_MAX;
8365
8366 if (check_add_overflow(off, sq_array_size, &off))
8367 return SIZE_MAX;
8368
Hristo Venev75b28af2019-08-26 17:23:46 +00008369 return off;
8370}
8371
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008372static int io_sqe_buffers_unregister(struct io_ring_ctx *ctx)
Jens Axboeedafcce2019-01-09 09:16:05 -07008373{
8374 int i, j;
8375
8376 if (!ctx->user_bufs)
8377 return -ENXIO;
8378
8379 for (i = 0; i < ctx->nr_user_bufs; i++) {
8380 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
8381
8382 for (j = 0; j < imu->nr_bvecs; j++)
John Hubbardf1f6a7d2020-01-30 22:13:35 -08008383 unpin_user_page(imu->bvec[j].bv_page);
Jens Axboeedafcce2019-01-09 09:16:05 -07008384
Jens Axboede293932020-09-17 16:19:16 -06008385 if (imu->acct_pages)
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008386 io_unaccount_mem(ctx, imu->acct_pages);
Mark Rutlandd4ef6472019-05-01 16:59:16 +01008387 kvfree(imu->bvec);
Jens Axboeedafcce2019-01-09 09:16:05 -07008388 imu->nr_bvecs = 0;
8389 }
8390
8391 kfree(ctx->user_bufs);
8392 ctx->user_bufs = NULL;
8393 ctx->nr_user_bufs = 0;
8394 return 0;
8395}
8396
8397static int io_copy_iov(struct io_ring_ctx *ctx, struct iovec *dst,
8398 void __user *arg, unsigned index)
8399{
8400 struct iovec __user *src;
8401
8402#ifdef CONFIG_COMPAT
8403 if (ctx->compat) {
8404 struct compat_iovec __user *ciovs;
8405 struct compat_iovec ciov;
8406
8407 ciovs = (struct compat_iovec __user *) arg;
8408 if (copy_from_user(&ciov, &ciovs[index], sizeof(ciov)))
8409 return -EFAULT;
8410
Jens Axboed55e5f52019-12-11 16:12:15 -07008411 dst->iov_base = u64_to_user_ptr((u64)ciov.iov_base);
Jens Axboeedafcce2019-01-09 09:16:05 -07008412 dst->iov_len = ciov.iov_len;
8413 return 0;
8414 }
8415#endif
8416 src = (struct iovec __user *) arg;
8417 if (copy_from_user(dst, &src[index], sizeof(*dst)))
8418 return -EFAULT;
8419 return 0;
8420}
8421
Jens Axboede293932020-09-17 16:19:16 -06008422/*
8423 * Not super efficient, but this is just a registration time. And we do cache
8424 * the last compound head, so generally we'll only do a full search if we don't
8425 * match that one.
8426 *
8427 * We check if the given compound head page has already been accounted, to
8428 * avoid double accounting it. This allows us to account the full size of the
8429 * page, not just the constituent pages of a huge page.
8430 */
8431static bool headpage_already_acct(struct io_ring_ctx *ctx, struct page **pages,
8432 int nr_pages, struct page *hpage)
8433{
8434 int i, j;
8435
8436 /* check current page array */
8437 for (i = 0; i < nr_pages; i++) {
8438 if (!PageCompound(pages[i]))
8439 continue;
8440 if (compound_head(pages[i]) == hpage)
8441 return true;
8442 }
8443
8444 /* check previously registered pages */
8445 for (i = 0; i < ctx->nr_user_bufs; i++) {
8446 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
8447
8448 for (j = 0; j < imu->nr_bvecs; j++) {
8449 if (!PageCompound(imu->bvec[j].bv_page))
8450 continue;
8451 if (compound_head(imu->bvec[j].bv_page) == hpage)
8452 return true;
8453 }
8454 }
8455
8456 return false;
8457}
8458
8459static int io_buffer_account_pin(struct io_ring_ctx *ctx, struct page **pages,
8460 int nr_pages, struct io_mapped_ubuf *imu,
8461 struct page **last_hpage)
8462{
8463 int i, ret;
8464
8465 for (i = 0; i < nr_pages; i++) {
8466 if (!PageCompound(pages[i])) {
8467 imu->acct_pages++;
8468 } else {
8469 struct page *hpage;
8470
8471 hpage = compound_head(pages[i]);
8472 if (hpage == *last_hpage)
8473 continue;
8474 *last_hpage = hpage;
8475 if (headpage_already_acct(ctx, pages, i, hpage))
8476 continue;
8477 imu->acct_pages += page_size(hpage) >> PAGE_SHIFT;
8478 }
8479 }
8480
8481 if (!imu->acct_pages)
8482 return 0;
8483
Jens Axboe26bfa89e2021-02-09 20:14:12 -07008484 ret = io_account_mem(ctx, imu->acct_pages);
Jens Axboede293932020-09-17 16:19:16 -06008485 if (ret)
8486 imu->acct_pages = 0;
8487 return ret;
8488}
8489
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008490static int io_sqe_buffer_register(struct io_ring_ctx *ctx, struct iovec *iov,
8491 struct io_mapped_ubuf *imu,
8492 struct page **last_hpage)
Jens Axboeedafcce2019-01-09 09:16:05 -07008493{
8494 struct vm_area_struct **vmas = NULL;
8495 struct page **pages = NULL;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008496 unsigned long off, start, end, ubuf;
8497 size_t size;
8498 int ret, pret, nr_pages, i;
8499
8500 ubuf = (unsigned long) iov->iov_base;
8501 end = (ubuf + iov->iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
8502 start = ubuf >> PAGE_SHIFT;
8503 nr_pages = end - start;
8504
8505 ret = -ENOMEM;
8506
8507 pages = kvmalloc_array(nr_pages, sizeof(struct page *), GFP_KERNEL);
8508 if (!pages)
8509 goto done;
8510
8511 vmas = kvmalloc_array(nr_pages, sizeof(struct vm_area_struct *),
8512 GFP_KERNEL);
8513 if (!vmas)
8514 goto done;
8515
8516 imu->bvec = kvmalloc_array(nr_pages, sizeof(struct bio_vec),
8517 GFP_KERNEL);
8518 if (!imu->bvec)
8519 goto done;
8520
8521 ret = 0;
8522 mmap_read_lock(current->mm);
8523 pret = pin_user_pages(ubuf, nr_pages, FOLL_WRITE | FOLL_LONGTERM,
8524 pages, vmas);
8525 if (pret == nr_pages) {
8526 /* don't support file backed memory */
8527 for (i = 0; i < nr_pages; i++) {
8528 struct vm_area_struct *vma = vmas[i];
8529
8530 if (vma->vm_file &&
8531 !is_file_hugepages(vma->vm_file)) {
8532 ret = -EOPNOTSUPP;
8533 break;
8534 }
8535 }
8536 } else {
8537 ret = pret < 0 ? pret : -EFAULT;
8538 }
8539 mmap_read_unlock(current->mm);
8540 if (ret) {
8541 /*
8542 * if we did partial map, or found file backed vmas,
8543 * release any pages we did get
8544 */
8545 if (pret > 0)
8546 unpin_user_pages(pages, pret);
8547 kvfree(imu->bvec);
8548 goto done;
8549 }
8550
8551 ret = io_buffer_account_pin(ctx, pages, pret, imu, last_hpage);
8552 if (ret) {
8553 unpin_user_pages(pages, pret);
8554 kvfree(imu->bvec);
8555 goto done;
8556 }
8557
8558 off = ubuf & ~PAGE_MASK;
8559 size = iov->iov_len;
8560 for (i = 0; i < nr_pages; i++) {
8561 size_t vec_len;
8562
8563 vec_len = min_t(size_t, size, PAGE_SIZE - off);
8564 imu->bvec[i].bv_page = pages[i];
8565 imu->bvec[i].bv_len = vec_len;
8566 imu->bvec[i].bv_offset = off;
8567 off = 0;
8568 size -= vec_len;
8569 }
8570 /* store original address for later verification */
8571 imu->ubuf = ubuf;
8572 imu->len = iov->iov_len;
8573 imu->nr_bvecs = nr_pages;
8574 ret = 0;
8575done:
8576 kvfree(pages);
8577 kvfree(vmas);
8578 return ret;
8579}
8580
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008581static int io_buffers_map_alloc(struct io_ring_ctx *ctx, unsigned int nr_args)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008582{
Jens Axboeedafcce2019-01-09 09:16:05 -07008583 if (ctx->user_bufs)
8584 return -EBUSY;
8585 if (!nr_args || nr_args > UIO_MAXIOV)
8586 return -EINVAL;
8587
8588 ctx->user_bufs = kcalloc(nr_args, sizeof(struct io_mapped_ubuf),
8589 GFP_KERNEL);
8590 if (!ctx->user_bufs)
8591 return -ENOMEM;
8592
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008593 return 0;
8594}
8595
8596static int io_buffer_validate(struct iovec *iov)
8597{
8598 /*
8599 * Don't impose further limits on the size and buffer
8600 * constraints here, we'll -EINVAL later when IO is
8601 * submitted if they are wrong.
8602 */
8603 if (!iov->iov_base || !iov->iov_len)
8604 return -EFAULT;
8605
8606 /* arbitrary limit, but we need something */
8607 if (iov->iov_len > SZ_1G)
8608 return -EFAULT;
8609
8610 return 0;
8611}
8612
8613static int io_sqe_buffers_register(struct io_ring_ctx *ctx, void __user *arg,
8614 unsigned int nr_args)
8615{
8616 int i, ret;
8617 struct iovec iov;
8618 struct page *last_hpage = NULL;
8619
8620 ret = io_buffers_map_alloc(ctx, nr_args);
8621 if (ret)
8622 return ret;
8623
Jens Axboeedafcce2019-01-09 09:16:05 -07008624 for (i = 0; i < nr_args; i++) {
8625 struct io_mapped_ubuf *imu = &ctx->user_bufs[i];
Jens Axboeedafcce2019-01-09 09:16:05 -07008626
8627 ret = io_copy_iov(ctx, &iov, arg, i);
8628 if (ret)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008629 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07008630
Bijan Mottahedeh2b358602021-01-06 12:39:11 -08008631 ret = io_buffer_validate(&iov);
8632 if (ret)
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008633 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07008634
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008635 ret = io_sqe_buffer_register(ctx, &iov, imu, &last_hpage);
8636 if (ret)
8637 break;
Jens Axboeedafcce2019-01-09 09:16:05 -07008638
8639 ctx->nr_user_bufs++;
8640 }
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008641
8642 if (ret)
8643 io_sqe_buffers_unregister(ctx);
8644
Jens Axboeedafcce2019-01-09 09:16:05 -07008645 return ret;
8646}
8647
Jens Axboe9b402842019-04-11 11:45:41 -06008648static int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg)
8649{
8650 __s32 __user *fds = arg;
8651 int fd;
8652
8653 if (ctx->cq_ev_fd)
8654 return -EBUSY;
8655
8656 if (copy_from_user(&fd, fds, sizeof(*fds)))
8657 return -EFAULT;
8658
8659 ctx->cq_ev_fd = eventfd_ctx_fdget(fd);
8660 if (IS_ERR(ctx->cq_ev_fd)) {
8661 int ret = PTR_ERR(ctx->cq_ev_fd);
8662 ctx->cq_ev_fd = NULL;
8663 return ret;
8664 }
8665
8666 return 0;
8667}
8668
8669static int io_eventfd_unregister(struct io_ring_ctx *ctx)
8670{
8671 if (ctx->cq_ev_fd) {
8672 eventfd_ctx_put(ctx->cq_ev_fd);
8673 ctx->cq_ev_fd = NULL;
8674 return 0;
8675 }
8676
8677 return -ENXIO;
8678}
8679
Jens Axboe5a2e7452020-02-23 16:23:11 -07008680static int __io_destroy_buffers(int id, void *p, void *data)
8681{
8682 struct io_ring_ctx *ctx = data;
8683 struct io_buffer *buf = p;
8684
Jens Axboe067524e2020-03-02 16:32:28 -07008685 __io_remove_buffers(ctx, buf, id, -1U);
Jens Axboe5a2e7452020-02-23 16:23:11 -07008686 return 0;
8687}
8688
8689static void io_destroy_buffers(struct io_ring_ctx *ctx)
8690{
8691 idr_for_each(&ctx->io_buffer_idr, __io_destroy_buffers, ctx);
8692 idr_destroy(&ctx->io_buffer_idr);
8693}
8694
Jens Axboec7dae4b2021-02-09 19:53:37 -07008695static void io_req_cache_free(struct list_head *list)
Jens Axboe1b4c3512021-02-10 00:03:19 +00008696{
Jens Axboec7dae4b2021-02-09 19:53:37 -07008697 while (!list_empty(list)) {
Jens Axboe1b4c3512021-02-10 00:03:19 +00008698 struct io_kiocb *req;
8699
Jens Axboec7dae4b2021-02-09 19:53:37 -07008700 req = list_first_entry(list, struct io_kiocb, compl.list);
Jens Axboe1b4c3512021-02-10 00:03:19 +00008701 list_del(&req->compl.list);
8702 kmem_cache_free(req_cachep, req);
8703 }
8704}
8705
Jens Axboe2b188cc2019-01-07 10:46:33 -07008706static void io_ring_ctx_free(struct io_ring_ctx *ctx)
8707{
Pavel Begunkovbf019da2021-02-10 00:03:17 +00008708 struct io_submit_state *submit_state = &ctx->submit_state;
8709
Jens Axboe6b063142019-01-10 22:13:58 -07008710 io_finish_async(ctx);
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -08008711 io_sqe_buffers_unregister(ctx);
Jens Axboe2aede0e2020-09-14 10:45:53 -06008712
8713 if (ctx->sqo_task) {
8714 put_task_struct(ctx->sqo_task);
8715 ctx->sqo_task = NULL;
8716 mmdrop(ctx->mm_account);
8717 ctx->mm_account = NULL;
Bijan Mottahedeh30975822020-06-16 16:36:09 -07008718 }
Jens Axboedef596e2019-01-09 08:59:42 -07008719
Pavel Begunkovbf019da2021-02-10 00:03:17 +00008720 if (submit_state->free_reqs)
8721 kmem_cache_free_bulk(req_cachep, submit_state->free_reqs,
8722 submit_state->reqs);
8723
Dennis Zhou91d8f512020-09-16 13:41:05 -07008724#ifdef CONFIG_BLK_CGROUP
8725 if (ctx->sqo_blkcg_css)
8726 css_put(ctx->sqo_blkcg_css);
8727#endif
8728
Jens Axboe6b063142019-01-10 22:13:58 -07008729 io_sqe_files_unregister(ctx);
Jens Axboe9b402842019-04-11 11:45:41 -06008730 io_eventfd_unregister(ctx);
Jens Axboe5a2e7452020-02-23 16:23:11 -07008731 io_destroy_buffers(ctx);
Jens Axboe41726c92020-02-23 13:11:42 -07008732 idr_destroy(&ctx->personality_idr);
Jens Axboedef596e2019-01-09 08:59:42 -07008733
Jens Axboe2b188cc2019-01-07 10:46:33 -07008734#if defined(CONFIG_UNIX)
Eric Biggers355e8d22019-06-12 14:58:43 -07008735 if (ctx->ring_sock) {
8736 ctx->ring_sock->file = NULL; /* so that iput() is called */
Jens Axboe2b188cc2019-01-07 10:46:33 -07008737 sock_release(ctx->ring_sock);
Eric Biggers355e8d22019-06-12 14:58:43 -07008738 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07008739#endif
8740
Hristo Venev75b28af2019-08-26 17:23:46 +00008741 io_mem_free(ctx->rings);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008742 io_mem_free(ctx->sq_sqes);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008743
8744 percpu_ref_exit(&ctx->refs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008745 free_uid(ctx->user);
Jens Axboe181e4482019-11-25 08:52:30 -07008746 put_cred(ctx->creds);
Jens Axboe78076bb2019-12-04 19:56:40 -07008747 kfree(ctx->cancel_hash);
Jens Axboec7dae4b2021-02-09 19:53:37 -07008748 io_req_cache_free(&ctx->submit_state.comp.free_list);
8749 io_req_cache_free(&ctx->submit_state.comp.locked_free_list);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008750 kfree(ctx);
8751}
8752
8753static __poll_t io_uring_poll(struct file *file, poll_table *wait)
8754{
8755 struct io_ring_ctx *ctx = file->private_data;
8756 __poll_t mask = 0;
8757
8758 poll_wait(file, &ctx->cq_wait, wait);
Stefan Bühler4f7067c2019-04-24 23:54:17 +02008759 /*
8760 * synchronizes with barrier from wq_has_sleeper call in
8761 * io_commit_cqring
8762 */
Jens Axboe2b188cc2019-01-07 10:46:33 -07008763 smp_rmb();
Jens Axboe90554202020-09-03 12:12:41 -06008764 if (!io_sqring_full(ctx))
Jens Axboe2b188cc2019-01-07 10:46:33 -07008765 mask |= EPOLLOUT | EPOLLWRNORM;
Hao Xued670c32021-02-05 16:34:21 +08008766
8767 /*
8768 * Don't flush cqring overflow list here, just do a simple check.
8769 * Otherwise there could possible be ABBA deadlock:
8770 * CPU0 CPU1
8771 * ---- ----
8772 * lock(&ctx->uring_lock);
8773 * lock(&ep->mtx);
8774 * lock(&ctx->uring_lock);
8775 * lock(&ep->mtx);
8776 *
8777 * Users may get EPOLLIN meanwhile seeing nothing in cqring, this
8778 * pushs them to do the flush.
8779 */
8780 if (io_cqring_events(ctx) || test_bit(0, &ctx->cq_check_overflow))
Jens Axboe2b188cc2019-01-07 10:46:33 -07008781 mask |= EPOLLIN | EPOLLRDNORM;
8782
8783 return mask;
8784}
8785
8786static int io_uring_fasync(int fd, struct file *file, int on)
8787{
8788 struct io_ring_ctx *ctx = file->private_data;
8789
8790 return fasync_helper(fd, file, on, &ctx->cq_fasync);
8791}
8792
Yejune Deng0bead8c2020-12-24 11:02:20 +08008793static int io_unregister_personality(struct io_ring_ctx *ctx, unsigned id)
Jens Axboe071698e2020-01-28 10:04:42 -07008794{
Jens Axboe1e6fa522020-10-15 08:46:24 -06008795 struct io_identity *iod;
Jens Axboe071698e2020-01-28 10:04:42 -07008796
Jens Axboe1e6fa522020-10-15 08:46:24 -06008797 iod = idr_remove(&ctx->personality_idr, id);
8798 if (iod) {
8799 put_cred(iod->creds);
8800 if (refcount_dec_and_test(&iod->count))
8801 kfree(iod);
Yejune Deng0bead8c2020-12-24 11:02:20 +08008802 return 0;
Jens Axboe1e6fa522020-10-15 08:46:24 -06008803 }
Yejune Deng0bead8c2020-12-24 11:02:20 +08008804
8805 return -EINVAL;
8806}
8807
8808static int io_remove_personalities(int id, void *p, void *data)
8809{
8810 struct io_ring_ctx *ctx = data;
8811
8812 io_unregister_personality(ctx, id);
Jens Axboe071698e2020-01-28 10:04:42 -07008813 return 0;
8814}
8815
Jens Axboe85faa7b2020-04-09 18:14:00 -06008816static void io_ring_exit_work(struct work_struct *work)
8817{
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008818 struct io_ring_ctx *ctx = container_of(work, struct io_ring_ctx,
8819 exit_work);
Jens Axboe85faa7b2020-04-09 18:14:00 -06008820
Jens Axboe56952e92020-06-17 15:00:04 -06008821 /*
8822 * If we're doing polled IO and end up having requests being
8823 * submitted async (out-of-line), then completions can come in while
8824 * we're waiting for refs to drop. We need to reap these manually,
8825 * as nobody else will be looking for them.
8826 */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008827 do {
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008828 io_uring_try_cancel_requests(ctx, NULL, NULL);
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008829 } while (!wait_for_completion_timeout(&ctx->ref_comp, HZ/20));
Jens Axboe85faa7b2020-04-09 18:14:00 -06008830 io_ring_ctx_free(ctx);
8831}
8832
Jens Axboe00c18642020-12-20 10:45:02 -07008833static bool io_cancel_ctx_cb(struct io_wq_work *work, void *data)
8834{
8835 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
8836
8837 return req->ctx == data;
8838}
8839
Jens Axboe2b188cc2019-01-07 10:46:33 -07008840static void io_ring_ctx_wait_and_kill(struct io_ring_ctx *ctx)
8841{
8842 mutex_lock(&ctx->uring_lock);
8843 percpu_ref_kill(&ctx->refs);
Pavel Begunkovd9d05212021-01-08 20:57:25 +00008844
8845 if (WARN_ON_ONCE((ctx->flags & IORING_SETUP_SQPOLL) && !ctx->sqo_dead))
8846 ctx->sqo_dead = 1;
8847
Pavel Begunkovcda286f2020-12-17 00:24:35 +00008848 /* if force is set, the ring is going away. always drop after that */
8849 ctx->cq_overflow_flushed = 1;
Pavel Begunkov634578f2020-12-06 22:22:44 +00008850 if (ctx->rings)
Pavel Begunkov6c503152021-01-04 20:36:36 +00008851 __io_cqring_overflow_flush(ctx, true, NULL, NULL);
Pavel Begunkov5c766a92021-01-19 13:32:36 +00008852 idr_for_each(&ctx->personality_idr, io_remove_personalities, ctx);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008853 mutex_unlock(&ctx->uring_lock);
8854
Pavel Begunkov6b819282020-11-06 13:00:25 +00008855 io_kill_timeouts(ctx, NULL, NULL);
8856 io_poll_remove_all(ctx, NULL, NULL);
Jens Axboe561fb042019-10-24 07:25:42 -06008857
8858 if (ctx->io_wq)
Jens Axboe00c18642020-12-20 10:45:02 -07008859 io_wq_cancel_cb(ctx->io_wq, io_cancel_ctx_cb, ctx, true);
Jens Axboe561fb042019-10-24 07:25:42 -06008860
Jens Axboe15dff282019-11-13 09:09:23 -07008861 /* if we failed setting up the ctx, we might not have any rings */
Pavel Begunkovb2edc0a2020-07-07 16:36:22 +03008862 io_iopoll_try_reap_events(ctx);
Jens Axboe309fc032020-07-10 09:13:34 -06008863
Jens Axboe85faa7b2020-04-09 18:14:00 -06008864 INIT_WORK(&ctx->exit_work, io_ring_exit_work);
Jens Axboefc666772020-08-19 11:10:51 -06008865 /*
8866 * Use system_unbound_wq to avoid spawning tons of event kworkers
8867 * if we're exiting a ton of rings at the same time. It just adds
8868 * noise and overhead, there's no discernable change in runtime
8869 * over using system_wq.
8870 */
8871 queue_work(system_unbound_wq, &ctx->exit_work);
Jens Axboe2b188cc2019-01-07 10:46:33 -07008872}
8873
8874static int io_uring_release(struct inode *inode, struct file *file)
8875{
8876 struct io_ring_ctx *ctx = file->private_data;
8877
8878 file->private_data = NULL;
8879 io_ring_ctx_wait_and_kill(ctx);
8880 return 0;
8881}
8882
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008883struct io_task_cancel {
8884 struct task_struct *task;
8885 struct files_struct *files;
8886};
Pavel Begunkov67c4d9e2020-06-15 10:24:05 +03008887
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008888static bool io_cancel_task_cb(struct io_wq_work *work, void *data)
Jens Axboeb711d4e2020-08-16 08:23:05 -07008889{
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008890 struct io_kiocb *req = container_of(work, struct io_kiocb, work);
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008891 struct io_task_cancel *cancel = data;
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008892 bool ret;
8893
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008894 if (cancel->files && (req->flags & REQ_F_LINK_TIMEOUT)) {
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008895 unsigned long flags;
8896 struct io_ring_ctx *ctx = req->ctx;
8897
8898 /* protect against races with linked timeouts */
8899 spin_lock_irqsave(&ctx->completion_lock, flags);
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008900 ret = io_match_task(req, cancel->task, cancel->files);
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008901 spin_unlock_irqrestore(&ctx->completion_lock, flags);
8902 } else {
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008903 ret = io_match_task(req, cancel->task, cancel->files);
Pavel Begunkov9a472ef2020-11-05 22:31:37 +00008904 }
8905 return ret;
Jens Axboeb711d4e2020-08-16 08:23:05 -07008906}
8907
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008908static void io_cancel_defer_files(struct io_ring_ctx *ctx,
Pavel Begunkovef9865a2020-11-05 14:06:19 +00008909 struct task_struct *task,
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008910 struct files_struct *files)
8911{
8912 struct io_defer_entry *de = NULL;
8913 LIST_HEAD(list);
8914
8915 spin_lock_irq(&ctx->completion_lock);
8916 list_for_each_entry_reverse(de, &ctx->defer_list, list) {
Pavel Begunkov08d23632020-11-06 13:00:22 +00008917 if (io_match_task(de->req, task, files)) {
Pavel Begunkovb7ddce32020-09-06 00:45:14 +03008918 list_cut_position(&list, &ctx->defer_list, &de->list);
8919 break;
8920 }
8921 }
8922 spin_unlock_irq(&ctx->completion_lock);
8923
8924 while (!list_empty(&list)) {
8925 de = list_first_entry(&list, struct io_defer_entry, list);
8926 list_del_init(&de->list);
8927 req_set_fail_links(de->req);
8928 io_put_req(de->req);
8929 io_req_complete(de->req, -ECANCELED);
8930 kfree(de);
8931 }
8932}
8933
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008934static void io_uring_try_cancel_requests(struct io_ring_ctx *ctx,
8935 struct task_struct *task,
8936 struct files_struct *files)
8937{
8938 struct io_task_cancel cancel = { .task = task, .files = files, };
8939
8940 while (1) {
8941 enum io_wq_cancel cret;
8942 bool ret = false;
8943
8944 if (ctx->io_wq) {
8945 cret = io_wq_cancel_cb(ctx->io_wq, io_cancel_task_cb,
8946 &cancel, true);
8947 ret |= (cret != IO_WQ_CANCEL_NOTFOUND);
8948 }
8949
8950 /* SQPOLL thread does its own polling */
8951 if (!(ctx->flags & IORING_SETUP_SQPOLL) && !files) {
8952 while (!list_empty_careful(&ctx->iopoll_list)) {
8953 io_iopoll_try_reap_events(ctx);
8954 ret = true;
8955 }
8956 }
8957
8958 ret |= io_poll_remove_all(ctx, task, files);
8959 ret |= io_kill_timeouts(ctx, task, files);
8960 ret |= io_run_task_work();
8961 io_cqring_overflow_flush(ctx, true, task, files);
8962 if (!ret)
8963 break;
8964 cond_resched();
8965 }
8966}
8967
Pavel Begunkovca70f002021-01-26 15:28:27 +00008968static int io_uring_count_inflight(struct io_ring_ctx *ctx,
8969 struct task_struct *task,
8970 struct files_struct *files)
8971{
8972 struct io_kiocb *req;
8973 int cnt = 0;
8974
8975 spin_lock_irq(&ctx->inflight_lock);
8976 list_for_each_entry(req, &ctx->inflight_list, inflight_entry)
8977 cnt += io_match_task(req, task, files);
8978 spin_unlock_irq(&ctx->inflight_lock);
8979 return cnt;
8980}
8981
Pavel Begunkovb52fda02020-11-06 13:00:24 +00008982static void io_uring_cancel_files(struct io_ring_ctx *ctx,
Pavel Begunkovdf9923f2020-11-06 13:00:23 +00008983 struct task_struct *task,
Jens Axboefcb323c2019-10-24 12:39:47 -06008984 struct files_struct *files)
8985{
Jens Axboefcb323c2019-10-24 12:39:47 -06008986 while (!list_empty_careful(&ctx->inflight_list)) {
Xiaoguang Wangd8f1b972020-04-26 15:54:43 +08008987 DEFINE_WAIT(wait);
Pavel Begunkovca70f002021-01-26 15:28:27 +00008988 int inflight;
Jens Axboefcb323c2019-10-24 12:39:47 -06008989
Pavel Begunkovca70f002021-01-26 15:28:27 +00008990 inflight = io_uring_count_inflight(ctx, task, files);
8991 if (!inflight)
Jens Axboefcb323c2019-10-24 12:39:47 -06008992 break;
Pavel Begunkovf6edbab2020-11-06 13:00:26 +00008993
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00008994 io_uring_try_cancel_requests(ctx, task, files);
Pavel Begunkov34343782021-02-10 11:45:42 +00008995
8996 if (ctx->sq_data)
8997 io_sq_thread_unpark(ctx->sq_data);
Pavel Begunkovca70f002021-01-26 15:28:27 +00008998 prepare_to_wait(&task->io_uring->wait, &wait,
8999 TASK_UNINTERRUPTIBLE);
9000 if (inflight == io_uring_count_inflight(ctx, task, files))
9001 schedule();
Pavel Begunkovc98de082020-11-15 12:56:32 +00009002 finish_wait(&task->io_uring->wait, &wait);
Pavel Begunkov34343782021-02-10 11:45:42 +00009003 if (ctx->sq_data)
9004 io_sq_thread_park(ctx->sq_data);
Jens Axboefcb323c2019-10-24 12:39:47 -06009005 }
9006}
9007
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009008static void io_disable_sqo_submit(struct io_ring_ctx *ctx)
9009{
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009010 mutex_lock(&ctx->uring_lock);
9011 ctx->sqo_dead = 1;
9012 mutex_unlock(&ctx->uring_lock);
9013
9014 /* make sure callers enter the ring to get error */
Pavel Begunkovb4411612021-01-13 12:42:24 +00009015 if (ctx->rings)
9016 io_ring_set_wakeup_flag(ctx);
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009017}
9018
Jens Axboe0f212202020-09-13 13:09:39 -06009019/*
9020 * We need to iteratively cancel requests, in case a request has dependent
9021 * hard links. These persist even for failure of cancelations, hence keep
9022 * looping until none are found.
9023 */
9024static void io_uring_cancel_task_requests(struct io_ring_ctx *ctx,
9025 struct files_struct *files)
9026{
9027 struct task_struct *task = current;
9028
Jens Axboefdaf0832020-10-30 09:37:30 -06009029 if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) {
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009030 io_disable_sqo_submit(ctx);
Jens Axboe534ca6d2020-09-02 13:52:19 -06009031 task = ctx->sq_data->thread;
Jens Axboefdaf0832020-10-30 09:37:30 -06009032 atomic_inc(&task->io_uring->in_idle);
9033 io_sq_thread_park(ctx->sq_data);
9034 }
Jens Axboe0f212202020-09-13 13:09:39 -06009035
Pavel Begunkovdf9923f2020-11-06 13:00:23 +00009036 io_cancel_defer_files(ctx, task, files);
Jens Axboe0f212202020-09-13 13:09:39 -06009037
Pavel Begunkov3a7efd12021-01-28 23:23:42 +00009038 io_uring_cancel_files(ctx, task, files);
Pavel Begunkovb52fda02020-11-06 13:00:24 +00009039 if (!files)
Pavel Begunkov9936c7c2021-02-04 13:51:56 +00009040 io_uring_try_cancel_requests(ctx, task, NULL);
Jens Axboefdaf0832020-10-30 09:37:30 -06009041
9042 if ((ctx->flags & IORING_SETUP_SQPOLL) && ctx->sq_data) {
9043 atomic_dec(&task->io_uring->in_idle);
9044 /*
9045 * If the files that are going away are the ones in the thread
9046 * identity, clear them out.
9047 */
9048 if (task->io_uring->identity->files == files)
9049 task->io_uring->identity->files = NULL;
9050 io_sq_thread_unpark(ctx->sq_data);
9051 }
Jens Axboe0f212202020-09-13 13:09:39 -06009052}
9053
9054/*
9055 * Note that this task has used io_uring. We use it for cancelation purposes.
9056 */
Jens Axboefdaf0832020-10-30 09:37:30 -06009057static int io_uring_add_task_file(struct io_ring_ctx *ctx, struct file *file)
Jens Axboe0f212202020-09-13 13:09:39 -06009058{
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01009059 struct io_uring_task *tctx = current->io_uring;
Pavel Begunkova528b042020-12-21 18:34:04 +00009060 int ret;
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01009061
9062 if (unlikely(!tctx)) {
Jens Axboe0f212202020-09-13 13:09:39 -06009063 ret = io_uring_alloc_task_context(current);
9064 if (unlikely(ret))
9065 return ret;
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01009066 tctx = current->io_uring;
Jens Axboe0f212202020-09-13 13:09:39 -06009067 }
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01009068 if (tctx->last != file) {
9069 void *old = xa_load(&tctx->xa, (unsigned long)file);
Jens Axboe0f212202020-09-13 13:09:39 -06009070
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01009071 if (!old) {
Jens Axboe0f212202020-09-13 13:09:39 -06009072 get_file(file);
Pavel Begunkova528b042020-12-21 18:34:04 +00009073 ret = xa_err(xa_store(&tctx->xa, (unsigned long)file,
9074 file, GFP_KERNEL));
9075 if (ret) {
9076 fput(file);
9077 return ret;
9078 }
Pavel Begunkovecfc8492021-01-25 11:42:20 +00009079
9080 /* one and only SQPOLL file note, held by sqo_task */
9081 WARN_ON_ONCE((ctx->flags & IORING_SETUP_SQPOLL) &&
9082 current != ctx->sqo_task);
Jens Axboe0f212202020-09-13 13:09:39 -06009083 }
Matthew Wilcox (Oracle)236434c2020-10-09 13:49:52 +01009084 tctx->last = file;
Jens Axboe0f212202020-09-13 13:09:39 -06009085 }
9086
Jens Axboefdaf0832020-10-30 09:37:30 -06009087 /*
9088 * This is race safe in that the task itself is doing this, hence it
9089 * cannot be going through the exit/cancel paths at the same time.
9090 * This cannot be modified while exit/cancel is running.
9091 */
9092 if (!tctx->sqpoll && (ctx->flags & IORING_SETUP_SQPOLL))
9093 tctx->sqpoll = true;
9094
Jens Axboe0f212202020-09-13 13:09:39 -06009095 return 0;
9096}
9097
9098/*
9099 * Remove this io_uring_file -> task mapping.
9100 */
9101static void io_uring_del_task_file(struct file *file)
9102{
9103 struct io_uring_task *tctx = current->io_uring;
Jens Axboe0f212202020-09-13 13:09:39 -06009104
9105 if (tctx->last == file)
9106 tctx->last = NULL;
Matthew Wilcox (Oracle)5e2ed8c2020-10-09 13:49:53 +01009107 file = xa_erase(&tctx->xa, (unsigned long)file);
Jens Axboe0f212202020-09-13 13:09:39 -06009108 if (file)
9109 fput(file);
9110}
9111
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00009112static void io_uring_remove_task_files(struct io_uring_task *tctx)
9113{
9114 struct file *file;
9115 unsigned long index;
9116
9117 xa_for_each(&tctx->xa, index, file)
9118 io_uring_del_task_file(file);
9119}
9120
Jens Axboe0f212202020-09-13 13:09:39 -06009121void __io_uring_files_cancel(struct files_struct *files)
9122{
9123 struct io_uring_task *tctx = current->io_uring;
Matthew Wilcox (Oracle)ce765372020-10-09 13:49:51 +01009124 struct file *file;
9125 unsigned long index;
Jens Axboe0f212202020-09-13 13:09:39 -06009126
9127 /* make sure overflow events are dropped */
Jens Axboefdaf0832020-10-30 09:37:30 -06009128 atomic_inc(&tctx->in_idle);
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00009129 xa_for_each(&tctx->xa, index, file)
9130 io_uring_cancel_task_requests(file->private_data, files);
Jens Axboefdaf0832020-10-30 09:37:30 -06009131 atomic_dec(&tctx->in_idle);
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00009132
9133 if (files)
9134 io_uring_remove_task_files(tctx);
Jens Axboefdaf0832020-10-30 09:37:30 -06009135}
9136
9137static s64 tctx_inflight(struct io_uring_task *tctx)
9138{
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00009139 return percpu_counter_sum(&tctx->inflight);
9140}
9141
9142static void io_uring_cancel_sqpoll(struct io_ring_ctx *ctx)
9143{
9144 struct io_uring_task *tctx;
Jens Axboefdaf0832020-10-30 09:37:30 -06009145 s64 inflight;
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00009146 DEFINE_WAIT(wait);
Jens Axboefdaf0832020-10-30 09:37:30 -06009147
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00009148 if (!ctx->sq_data)
9149 return;
9150 tctx = ctx->sq_data->thread->io_uring;
9151 io_disable_sqo_submit(ctx);
Jens Axboefdaf0832020-10-30 09:37:30 -06009152
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00009153 atomic_inc(&tctx->in_idle);
9154 do {
9155 /* read completions before cancelations */
9156 inflight = tctx_inflight(tctx);
9157 if (!inflight)
9158 break;
9159 io_uring_cancel_task_requests(ctx, NULL);
Jens Axboefdaf0832020-10-30 09:37:30 -06009160
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00009161 prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE);
9162 /*
9163 * If we've seen completions, retry without waiting. This
9164 * avoids a race where a completion comes in before we did
9165 * prepare_to_wait().
9166 */
9167 if (inflight == tctx_inflight(tctx))
9168 schedule();
9169 finish_wait(&tctx->wait, &wait);
9170 } while (1);
9171 atomic_dec(&tctx->in_idle);
Jens Axboe0f212202020-09-13 13:09:39 -06009172}
9173
Jens Axboe0f212202020-09-13 13:09:39 -06009174/*
9175 * Find any io_uring fd that this task has registered or done IO on, and cancel
9176 * requests.
9177 */
9178void __io_uring_task_cancel(void)
9179{
9180 struct io_uring_task *tctx = current->io_uring;
9181 DEFINE_WAIT(wait);
Jens Axboed8a6df12020-10-15 16:24:45 -06009182 s64 inflight;
Jens Axboe0f212202020-09-13 13:09:39 -06009183
9184 /* make sure overflow events are dropped */
Jens Axboefdaf0832020-10-30 09:37:30 -06009185 atomic_inc(&tctx->in_idle);
Jens Axboe0f212202020-09-13 13:09:39 -06009186
Pavel Begunkov0b5cd6c2021-01-17 02:29:56 +00009187 /* trigger io_disable_sqo_submit() */
Pavel Begunkov0e9ddb32021-02-07 22:34:26 +00009188 if (tctx->sqpoll) {
9189 struct file *file;
9190 unsigned long index;
9191
9192 xa_for_each(&tctx->xa, index, file)
9193 io_uring_cancel_sqpoll(file->private_data);
9194 }
Pavel Begunkov0b5cd6c2021-01-17 02:29:56 +00009195
Jens Axboed8a6df12020-10-15 16:24:45 -06009196 do {
Jens Axboe0f212202020-09-13 13:09:39 -06009197 /* read completions before cancelations */
Jens Axboefdaf0832020-10-30 09:37:30 -06009198 inflight = tctx_inflight(tctx);
Jens Axboed8a6df12020-10-15 16:24:45 -06009199 if (!inflight)
9200 break;
Jens Axboe0f212202020-09-13 13:09:39 -06009201 __io_uring_files_cancel(NULL);
9202
9203 prepare_to_wait(&tctx->wait, &wait, TASK_UNINTERRUPTIBLE);
9204
9205 /*
Pavel Begunkova1bb3cd2021-01-26 15:28:26 +00009206 * If we've seen completions, retry without waiting. This
9207 * avoids a race where a completion comes in before we did
9208 * prepare_to_wait().
Jens Axboe0f212202020-09-13 13:09:39 -06009209 */
Pavel Begunkova1bb3cd2021-01-26 15:28:26 +00009210 if (inflight == tctx_inflight(tctx))
9211 schedule();
Pavel Begunkovf57555e2020-12-20 13:21:44 +00009212 finish_wait(&tctx->wait, &wait);
Jens Axboed8a6df12020-10-15 16:24:45 -06009213 } while (1);
Jens Axboe0f212202020-09-13 13:09:39 -06009214
Jens Axboefdaf0832020-10-30 09:37:30 -06009215 atomic_dec(&tctx->in_idle);
Pavel Begunkovde7f1d92021-01-04 20:43:29 +00009216
9217 io_uring_remove_task_files(tctx);
Pavel Begunkov44e728b2020-06-15 10:24:04 +03009218}
9219
Jens Axboefcb323c2019-10-24 12:39:47 -06009220static int io_uring_flush(struct file *file, void *data)
9221{
Pavel Begunkov6b5733e2021-01-08 20:57:24 +00009222 struct io_uring_task *tctx = current->io_uring;
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009223 struct io_ring_ctx *ctx = file->private_data;
Pavel Begunkov6b5733e2021-01-08 20:57:24 +00009224
Jens Axboe84965ff2021-01-23 15:51:11 -07009225 if (fatal_signal_pending(current) || (current->flags & PF_EXITING))
9226 io_uring_cancel_task_requests(ctx, NULL);
9227
Pavel Begunkov6b5733e2021-01-08 20:57:24 +00009228 if (!tctx)
Pavel Begunkov4f793dc2021-01-08 20:57:23 +00009229 return 0;
9230
Pavel Begunkov6b5733e2021-01-08 20:57:24 +00009231 /* we should have cancelled and erased it before PF_EXITING */
9232 WARN_ON_ONCE((current->flags & PF_EXITING) &&
9233 xa_load(&tctx->xa, (unsigned long)file));
9234
Pavel Begunkov4f793dc2021-01-08 20:57:23 +00009235 /*
9236 * fput() is pending, will be 2 if the only other ref is our potential
9237 * task file note. If the task is exiting, drop regardless of count.
9238 */
Pavel Begunkov6b5733e2021-01-08 20:57:24 +00009239 if (atomic_long_read(&file->f_count) != 2)
9240 return 0;
Pavel Begunkov4f793dc2021-01-08 20:57:23 +00009241
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009242 if (ctx->flags & IORING_SETUP_SQPOLL) {
9243 /* there is only one file note, which is owned by sqo_task */
Pavel Begunkov4325cb42021-01-16 05:32:30 +00009244 WARN_ON_ONCE(ctx->sqo_task != current &&
9245 xa_load(&tctx->xa, (unsigned long)file));
9246 /* sqo_dead check is for when this happens after cancellation */
9247 WARN_ON_ONCE(ctx->sqo_task == current && !ctx->sqo_dead &&
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009248 !xa_load(&tctx->xa, (unsigned long)file));
9249
9250 io_disable_sqo_submit(ctx);
9251 }
9252
9253 if (!(ctx->flags & IORING_SETUP_SQPOLL) || ctx->sqo_task == current)
9254 io_uring_del_task_file(file);
Jens Axboefcb323c2019-10-24 12:39:47 -06009255 return 0;
9256}
9257
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009258static void *io_uring_validate_mmap_request(struct file *file,
9259 loff_t pgoff, size_t sz)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009260{
Jens Axboe2b188cc2019-01-07 10:46:33 -07009261 struct io_ring_ctx *ctx = file->private_data;
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009262 loff_t offset = pgoff << PAGE_SHIFT;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009263 struct page *page;
9264 void *ptr;
9265
9266 switch (offset) {
9267 case IORING_OFF_SQ_RING:
Hristo Venev75b28af2019-08-26 17:23:46 +00009268 case IORING_OFF_CQ_RING:
9269 ptr = ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009270 break;
9271 case IORING_OFF_SQES:
9272 ptr = ctx->sq_sqes;
9273 break;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009274 default:
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009275 return ERR_PTR(-EINVAL);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009276 }
9277
9278 page = virt_to_head_page(ptr);
Matthew Wilcox (Oracle)a50b8542019-09-23 15:34:25 -07009279 if (sz > page_size(page))
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009280 return ERR_PTR(-EINVAL);
9281
9282 return ptr;
9283}
9284
9285#ifdef CONFIG_MMU
9286
9287static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
9288{
9289 size_t sz = vma->vm_end - vma->vm_start;
9290 unsigned long pfn;
9291 void *ptr;
9292
9293 ptr = io_uring_validate_mmap_request(file, vma->vm_pgoff, sz);
9294 if (IS_ERR(ptr))
9295 return PTR_ERR(ptr);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009296
9297 pfn = virt_to_phys(ptr) >> PAGE_SHIFT;
9298 return remap_pfn_range(vma, vma->vm_start, pfn, sz, vma->vm_page_prot);
9299}
9300
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009301#else /* !CONFIG_MMU */
9302
9303static int io_uring_mmap(struct file *file, struct vm_area_struct *vma)
9304{
9305 return vma->vm_flags & (VM_SHARED | VM_MAYSHARE) ? 0 : -EINVAL;
9306}
9307
9308static unsigned int io_uring_nommu_mmap_capabilities(struct file *file)
9309{
9310 return NOMMU_MAP_DIRECT | NOMMU_MAP_READ | NOMMU_MAP_WRITE;
9311}
9312
9313static unsigned long io_uring_nommu_get_unmapped_area(struct file *file,
9314 unsigned long addr, unsigned long len,
9315 unsigned long pgoff, unsigned long flags)
9316{
9317 void *ptr;
9318
9319 ptr = io_uring_validate_mmap_request(file, pgoff, len);
9320 if (IS_ERR(ptr))
9321 return PTR_ERR(ptr);
9322
9323 return (unsigned long) ptr;
9324}
9325
9326#endif /* !CONFIG_MMU */
9327
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009328static int io_sqpoll_wait_sq(struct io_ring_ctx *ctx)
Jens Axboe90554202020-09-03 12:12:41 -06009329{
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009330 int ret = 0;
Jens Axboe90554202020-09-03 12:12:41 -06009331 DEFINE_WAIT(wait);
9332
9333 do {
9334 if (!io_sqring_full(ctx))
9335 break;
9336
9337 prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE);
9338
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009339 if (unlikely(ctx->sqo_dead)) {
9340 ret = -EOWNERDEAD;
9341 goto out;
9342 }
9343
Jens Axboe90554202020-09-03 12:12:41 -06009344 if (!io_sqring_full(ctx))
9345 break;
9346
9347 schedule();
9348 } while (!signal_pending(current));
9349
9350 finish_wait(&ctx->sqo_sq_wait, &wait);
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009351out:
9352 return ret;
Jens Axboe90554202020-09-03 12:12:41 -06009353}
9354
Hao Xuc73ebb62020-11-03 10:54:37 +08009355static int io_get_ext_arg(unsigned flags, const void __user *argp, size_t *argsz,
9356 struct __kernel_timespec __user **ts,
9357 const sigset_t __user **sig)
9358{
9359 struct io_uring_getevents_arg arg;
9360
9361 /*
9362 * If EXT_ARG isn't set, then we have no timespec and the argp pointer
9363 * is just a pointer to the sigset_t.
9364 */
9365 if (!(flags & IORING_ENTER_EXT_ARG)) {
9366 *sig = (const sigset_t __user *) argp;
9367 *ts = NULL;
9368 return 0;
9369 }
9370
9371 /*
9372 * EXT_ARG is set - ensure we agree on the size of it and copy in our
9373 * timespec and sigset_t pointers if good.
9374 */
9375 if (*argsz != sizeof(arg))
9376 return -EINVAL;
9377 if (copy_from_user(&arg, argp, sizeof(arg)))
9378 return -EFAULT;
9379 *sig = u64_to_user_ptr(arg.sigmask);
9380 *argsz = arg.sigmask_sz;
9381 *ts = u64_to_user_ptr(arg.ts);
9382 return 0;
9383}
9384
Jens Axboe2b188cc2019-01-07 10:46:33 -07009385SYSCALL_DEFINE6(io_uring_enter, unsigned int, fd, u32, to_submit,
Hao Xuc73ebb62020-11-03 10:54:37 +08009386 u32, min_complete, u32, flags, const void __user *, argp,
9387 size_t, argsz)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009388{
9389 struct io_ring_ctx *ctx;
9390 long ret = -EBADF;
9391 int submitted = 0;
9392 struct fd f;
9393
Jens Axboe4c6e2772020-07-01 11:29:10 -06009394 io_run_task_work();
Jens Axboeb41e9852020-02-17 09:52:41 -07009395
Jens Axboe90554202020-09-03 12:12:41 -06009396 if (flags & ~(IORING_ENTER_GETEVENTS | IORING_ENTER_SQ_WAKEUP |
Hao Xuc73ebb62020-11-03 10:54:37 +08009397 IORING_ENTER_SQ_WAIT | IORING_ENTER_EXT_ARG))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009398 return -EINVAL;
9399
9400 f = fdget(fd);
9401 if (!f.file)
9402 return -EBADF;
9403
9404 ret = -EOPNOTSUPP;
9405 if (f.file->f_op != &io_uring_fops)
9406 goto out_fput;
9407
9408 ret = -ENXIO;
9409 ctx = f.file->private_data;
9410 if (!percpu_ref_tryget(&ctx->refs))
9411 goto out_fput;
9412
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009413 ret = -EBADFD;
9414 if (ctx->flags & IORING_SETUP_R_DISABLED)
9415 goto out;
9416
Jens Axboe6c271ce2019-01-10 11:22:30 -07009417 /*
9418 * For SQ polling, the thread will do all submissions and completions.
9419 * Just return the requested submit count, and wake the thread if
9420 * we were asked to.
9421 */
Jens Axboeb2a9ead2019-09-12 14:19:16 -06009422 ret = 0;
Jens Axboe6c271ce2019-01-10 11:22:30 -07009423 if (ctx->flags & IORING_SETUP_SQPOLL) {
Pavel Begunkov6c503152021-01-04 20:36:36 +00009424 io_cqring_overflow_flush(ctx, false, NULL, NULL);
Pavel Begunkov89448c42020-12-17 00:24:39 +00009425
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009426 ret = -EOWNERDEAD;
9427 if (unlikely(ctx->sqo_dead))
9428 goto out;
Jens Axboe6c271ce2019-01-10 11:22:30 -07009429 if (flags & IORING_ENTER_SQ_WAKEUP)
Jens Axboe534ca6d2020-09-02 13:52:19 -06009430 wake_up(&ctx->sq_data->wait);
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009431 if (flags & IORING_ENTER_SQ_WAIT) {
9432 ret = io_sqpoll_wait_sq(ctx);
9433 if (ret)
9434 goto out;
9435 }
Jens Axboe6c271ce2019-01-10 11:22:30 -07009436 submitted = to_submit;
Jens Axboeb2a9ead2019-09-12 14:19:16 -06009437 } else if (to_submit) {
Jens Axboefdaf0832020-10-30 09:37:30 -06009438 ret = io_uring_add_task_file(ctx, f.file);
Jens Axboe0f212202020-09-13 13:09:39 -06009439 if (unlikely(ret))
9440 goto out;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009441 mutex_lock(&ctx->uring_lock);
Jens Axboe0f212202020-09-13 13:09:39 -06009442 submitted = io_submit_sqes(ctx, to_submit);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009443 mutex_unlock(&ctx->uring_lock);
Pavel Begunkov7c504e652019-12-18 19:53:45 +03009444
9445 if (submitted != to_submit)
9446 goto out;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009447 }
9448 if (flags & IORING_ENTER_GETEVENTS) {
Hao Xuc73ebb62020-11-03 10:54:37 +08009449 const sigset_t __user *sig;
9450 struct __kernel_timespec __user *ts;
9451
9452 ret = io_get_ext_arg(flags, argp, &argsz, &ts, &sig);
9453 if (unlikely(ret))
9454 goto out;
9455
Jens Axboe2b188cc2019-01-07 10:46:33 -07009456 min_complete = min(min_complete, ctx->cq_entries);
9457
Xiaoguang Wang32b22442020-03-11 09:26:09 +08009458 /*
9459 * When SETUP_IOPOLL and SETUP_SQPOLL are both enabled, user
9460 * space applications don't need to do io completion events
9461 * polling again, they can rely on io_sq_thread to do polling
9462 * work, which can reduce cpu usage and uring_lock contention.
9463 */
9464 if (ctx->flags & IORING_SETUP_IOPOLL &&
9465 !(ctx->flags & IORING_SETUP_SQPOLL)) {
Pavel Begunkov7668b922020-07-07 16:36:21 +03009466 ret = io_iopoll_check(ctx, min_complete);
Jens Axboedef596e2019-01-09 08:59:42 -07009467 } else {
Hao Xuc73ebb62020-11-03 10:54:37 +08009468 ret = io_cqring_wait(ctx, min_complete, sig, argsz, ts);
Jens Axboedef596e2019-01-09 08:59:42 -07009469 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009470 }
9471
Pavel Begunkov7c504e652019-12-18 19:53:45 +03009472out:
Pavel Begunkov6805b322019-10-08 02:18:42 +03009473 percpu_ref_put(&ctx->refs);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009474out_fput:
9475 fdput(f);
9476 return submitted ? submitted : ret;
9477}
9478
Tobias Klauserbebdb652020-02-26 18:38:32 +01009479#ifdef CONFIG_PROC_FS
Jens Axboe87ce9552020-01-30 08:25:34 -07009480static int io_uring_show_cred(int id, void *p, void *data)
9481{
Jens Axboe6b47ab82020-11-05 09:50:16 -07009482 struct io_identity *iod = p;
9483 const struct cred *cred = iod->creds;
Jens Axboe87ce9552020-01-30 08:25:34 -07009484 struct seq_file *m = data;
9485 struct user_namespace *uns = seq_user_ns(m);
9486 struct group_info *gi;
9487 kernel_cap_t cap;
9488 unsigned __capi;
9489 int g;
9490
9491 seq_printf(m, "%5d\n", id);
9492 seq_put_decimal_ull(m, "\tUid:\t", from_kuid_munged(uns, cred->uid));
9493 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->euid));
9494 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->suid));
9495 seq_put_decimal_ull(m, "\t\t", from_kuid_munged(uns, cred->fsuid));
9496 seq_put_decimal_ull(m, "\n\tGid:\t", from_kgid_munged(uns, cred->gid));
9497 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->egid));
9498 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->sgid));
9499 seq_put_decimal_ull(m, "\t\t", from_kgid_munged(uns, cred->fsgid));
9500 seq_puts(m, "\n\tGroups:\t");
9501 gi = cred->group_info;
9502 for (g = 0; g < gi->ngroups; g++) {
9503 seq_put_decimal_ull(m, g ? " " : "",
9504 from_kgid_munged(uns, gi->gid[g]));
9505 }
9506 seq_puts(m, "\n\tCapEff:\t");
9507 cap = cred->cap_effective;
9508 CAP_FOR_EACH_U32(__capi)
9509 seq_put_hex_ll(m, NULL, cap.cap[CAP_LAST_U32 - __capi], 8);
9510 seq_putc(m, '\n');
9511 return 0;
9512}
9513
9514static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m)
9515{
Joseph Qidbbe9c62020-09-29 09:01:22 -06009516 struct io_sq_data *sq = NULL;
Jens Axboefad8e0d2020-09-28 08:57:48 -06009517 bool has_lock;
Jens Axboe87ce9552020-01-30 08:25:34 -07009518 int i;
9519
Jens Axboefad8e0d2020-09-28 08:57:48 -06009520 /*
9521 * Avoid ABBA deadlock between the seq lock and the io_uring mutex,
9522 * since fdinfo case grabs it in the opposite direction of normal use
9523 * cases. If we fail to get the lock, we just don't iterate any
9524 * structures that could be going away outside the io_uring mutex.
9525 */
9526 has_lock = mutex_trylock(&ctx->uring_lock);
9527
Joseph Qidbbe9c62020-09-29 09:01:22 -06009528 if (has_lock && (ctx->flags & IORING_SETUP_SQPOLL))
9529 sq = ctx->sq_data;
9530
9531 seq_printf(m, "SqThread:\t%d\n", sq ? task_pid_nr(sq->thread) : -1);
9532 seq_printf(m, "SqThreadCpu:\t%d\n", sq ? task_cpu(sq->thread) : -1);
Jens Axboe87ce9552020-01-30 08:25:34 -07009533 seq_printf(m, "UserFiles:\t%u\n", ctx->nr_user_files);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009534 for (i = 0; has_lock && i < ctx->nr_user_files; i++) {
Pavel Begunkovea64ec022021-02-04 13:52:07 +00009535 struct file *f = *io_fixed_file_slot(ctx->file_data, i);
Jens Axboe87ce9552020-01-30 08:25:34 -07009536
Jens Axboe87ce9552020-01-30 08:25:34 -07009537 if (f)
9538 seq_printf(m, "%5u: %s\n", i, file_dentry(f)->d_iname);
9539 else
9540 seq_printf(m, "%5u: <none>\n", i);
9541 }
9542 seq_printf(m, "UserBufs:\t%u\n", ctx->nr_user_bufs);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009543 for (i = 0; has_lock && i < ctx->nr_user_bufs; i++) {
Jens Axboe87ce9552020-01-30 08:25:34 -07009544 struct io_mapped_ubuf *buf = &ctx->user_bufs[i];
9545
9546 seq_printf(m, "%5u: 0x%llx/%u\n", i, buf->ubuf,
9547 (unsigned int) buf->len);
9548 }
Jens Axboefad8e0d2020-09-28 08:57:48 -06009549 if (has_lock && !idr_is_empty(&ctx->personality_idr)) {
Jens Axboe87ce9552020-01-30 08:25:34 -07009550 seq_printf(m, "Personalities:\n");
9551 idr_for_each(&ctx->personality_idr, io_uring_show_cred, m);
9552 }
Jens Axboed7718a92020-02-14 22:23:12 -07009553 seq_printf(m, "PollList:\n");
9554 spin_lock_irq(&ctx->completion_lock);
9555 for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
9556 struct hlist_head *list = &ctx->cancel_hash[i];
9557 struct io_kiocb *req;
9558
9559 hlist_for_each_entry(req, list, hash_node)
9560 seq_printf(m, " op=%d, task_works=%d\n", req->opcode,
9561 req->task->task_works != NULL);
9562 }
9563 spin_unlock_irq(&ctx->completion_lock);
Jens Axboefad8e0d2020-09-28 08:57:48 -06009564 if (has_lock)
9565 mutex_unlock(&ctx->uring_lock);
Jens Axboe87ce9552020-01-30 08:25:34 -07009566}
9567
9568static void io_uring_show_fdinfo(struct seq_file *m, struct file *f)
9569{
9570 struct io_ring_ctx *ctx = f->private_data;
9571
9572 if (percpu_ref_tryget(&ctx->refs)) {
9573 __io_uring_show_fdinfo(ctx, m);
9574 percpu_ref_put(&ctx->refs);
9575 }
9576}
Tobias Klauserbebdb652020-02-26 18:38:32 +01009577#endif
Jens Axboe87ce9552020-01-30 08:25:34 -07009578
Jens Axboe2b188cc2019-01-07 10:46:33 -07009579static const struct file_operations io_uring_fops = {
9580 .release = io_uring_release,
Jens Axboefcb323c2019-10-24 12:39:47 -06009581 .flush = io_uring_flush,
Jens Axboe2b188cc2019-01-07 10:46:33 -07009582 .mmap = io_uring_mmap,
Roman Penyaev6c5c2402019-11-28 12:53:22 +01009583#ifndef CONFIG_MMU
9584 .get_unmapped_area = io_uring_nommu_get_unmapped_area,
9585 .mmap_capabilities = io_uring_nommu_mmap_capabilities,
9586#endif
Jens Axboe2b188cc2019-01-07 10:46:33 -07009587 .poll = io_uring_poll,
9588 .fasync = io_uring_fasync,
Tobias Klauserbebdb652020-02-26 18:38:32 +01009589#ifdef CONFIG_PROC_FS
Jens Axboe87ce9552020-01-30 08:25:34 -07009590 .show_fdinfo = io_uring_show_fdinfo,
Tobias Klauserbebdb652020-02-26 18:38:32 +01009591#endif
Jens Axboe2b188cc2019-01-07 10:46:33 -07009592};
9593
9594static int io_allocate_scq_urings(struct io_ring_ctx *ctx,
9595 struct io_uring_params *p)
9596{
Hristo Venev75b28af2019-08-26 17:23:46 +00009597 struct io_rings *rings;
9598 size_t size, sq_array_offset;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009599
Jens Axboebd740482020-08-05 12:58:23 -06009600 /* make sure these are sane, as we already accounted them */
9601 ctx->sq_entries = p->sq_entries;
9602 ctx->cq_entries = p->cq_entries;
9603
Hristo Venev75b28af2019-08-26 17:23:46 +00009604 size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset);
9605 if (size == SIZE_MAX)
9606 return -EOVERFLOW;
9607
9608 rings = io_mem_alloc(size);
9609 if (!rings)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009610 return -ENOMEM;
9611
Hristo Venev75b28af2019-08-26 17:23:46 +00009612 ctx->rings = rings;
9613 ctx->sq_array = (u32 *)((char *)rings + sq_array_offset);
9614 rings->sq_ring_mask = p->sq_entries - 1;
9615 rings->cq_ring_mask = p->cq_entries - 1;
9616 rings->sq_ring_entries = p->sq_entries;
9617 rings->cq_ring_entries = p->cq_entries;
9618 ctx->sq_mask = rings->sq_ring_mask;
9619 ctx->cq_mask = rings->cq_ring_mask;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009620
9621 size = array_size(sizeof(struct io_uring_sqe), p->sq_entries);
Jens Axboeeb065d32019-11-20 09:26:29 -07009622 if (size == SIZE_MAX) {
9623 io_mem_free(ctx->rings);
9624 ctx->rings = NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009625 return -EOVERFLOW;
Jens Axboeeb065d32019-11-20 09:26:29 -07009626 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009627
9628 ctx->sq_sqes = io_mem_alloc(size);
Jens Axboeeb065d32019-11-20 09:26:29 -07009629 if (!ctx->sq_sqes) {
9630 io_mem_free(ctx->rings);
9631 ctx->rings = NULL;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009632 return -ENOMEM;
Jens Axboeeb065d32019-11-20 09:26:29 -07009633 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009634
Jens Axboe2b188cc2019-01-07 10:46:33 -07009635 return 0;
9636}
9637
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009638static int io_uring_install_fd(struct io_ring_ctx *ctx, struct file *file)
9639{
9640 int ret, fd;
9641
9642 fd = get_unused_fd_flags(O_RDWR | O_CLOEXEC);
9643 if (fd < 0)
9644 return fd;
9645
9646 ret = io_uring_add_task_file(ctx, file);
9647 if (ret) {
9648 put_unused_fd(fd);
9649 return ret;
9650 }
9651 fd_install(fd, file);
9652 return fd;
9653}
9654
Jens Axboe2b188cc2019-01-07 10:46:33 -07009655/*
9656 * Allocate an anonymous fd, this is what constitutes the application
9657 * visible backing of an io_uring instance. The application mmaps this
9658 * fd to gain access to the SQ/CQ ring details. If UNIX sockets are enabled,
9659 * we have to tie this fd to a socket for file garbage collection purposes.
9660 */
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009661static struct file *io_uring_get_file(struct io_ring_ctx *ctx)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009662{
9663 struct file *file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009664#if defined(CONFIG_UNIX)
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009665 int ret;
9666
Jens Axboe2b188cc2019-01-07 10:46:33 -07009667 ret = sock_create_kern(&init_net, PF_UNIX, SOCK_RAW, IPPROTO_IP,
9668 &ctx->ring_sock);
9669 if (ret)
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009670 return ERR_PTR(ret);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009671#endif
9672
Jens Axboe2b188cc2019-01-07 10:46:33 -07009673 file = anon_inode_getfile("[io_uring]", &io_uring_fops, ctx,
9674 O_RDWR | O_CLOEXEC);
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009675#if defined(CONFIG_UNIX)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009676 if (IS_ERR(file)) {
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009677 sock_release(ctx->ring_sock);
9678 ctx->ring_sock = NULL;
9679 } else {
9680 ctx->ring_sock->file = file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009681 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009682#endif
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009683 return file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009684}
9685
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009686static int io_uring_create(unsigned entries, struct io_uring_params *p,
9687 struct io_uring_params __user *params)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009688{
9689 struct user_struct *user = NULL;
9690 struct io_ring_ctx *ctx;
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009691 struct file *file;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009692 int ret;
9693
Jens Axboe8110c1a2019-12-28 15:39:54 -07009694 if (!entries)
Jens Axboe2b188cc2019-01-07 10:46:33 -07009695 return -EINVAL;
Jens Axboe8110c1a2019-12-28 15:39:54 -07009696 if (entries > IORING_MAX_ENTRIES) {
9697 if (!(p->flags & IORING_SETUP_CLAMP))
9698 return -EINVAL;
9699 entries = IORING_MAX_ENTRIES;
9700 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009701
9702 /*
9703 * Use twice as many entries for the CQ ring. It's possible for the
9704 * application to drive a higher depth than the size of the SQ ring,
9705 * since the sqes are only used at submission time. This allows for
Jens Axboe33a107f2019-10-04 12:10:03 -06009706 * some flexibility in overcommitting a bit. If the application has
9707 * set IORING_SETUP_CQSIZE, it will have passed in the desired number
9708 * of CQ ring entries manually.
Jens Axboe2b188cc2019-01-07 10:46:33 -07009709 */
9710 p->sq_entries = roundup_pow_of_two(entries);
Jens Axboe33a107f2019-10-04 12:10:03 -06009711 if (p->flags & IORING_SETUP_CQSIZE) {
9712 /*
9713 * If IORING_SETUP_CQSIZE is set, we do the same roundup
9714 * to a power-of-two, if it isn't already. We do NOT impose
9715 * any cq vs sq ring sizing.
9716 */
Joseph Qieb2667b32020-11-24 15:03:03 +08009717 if (!p->cq_entries)
Jens Axboe33a107f2019-10-04 12:10:03 -06009718 return -EINVAL;
Jens Axboe8110c1a2019-12-28 15:39:54 -07009719 if (p->cq_entries > IORING_MAX_CQ_ENTRIES) {
9720 if (!(p->flags & IORING_SETUP_CLAMP))
9721 return -EINVAL;
9722 p->cq_entries = IORING_MAX_CQ_ENTRIES;
9723 }
Joseph Qieb2667b32020-11-24 15:03:03 +08009724 p->cq_entries = roundup_pow_of_two(p->cq_entries);
9725 if (p->cq_entries < p->sq_entries)
9726 return -EINVAL;
Jens Axboe33a107f2019-10-04 12:10:03 -06009727 } else {
9728 p->cq_entries = 2 * p->sq_entries;
9729 }
Jens Axboe2b188cc2019-01-07 10:46:33 -07009730
9731 user = get_uid(current_user());
Jens Axboe2b188cc2019-01-07 10:46:33 -07009732
9733 ctx = io_ring_ctx_alloc(p);
9734 if (!ctx) {
Jens Axboe2b188cc2019-01-07 10:46:33 -07009735 free_uid(user);
9736 return -ENOMEM;
9737 }
9738 ctx->compat = in_compat_syscall();
Jens Axboe26bfa89e2021-02-09 20:14:12 -07009739 ctx->limit_mem = !capable(CAP_IPC_LOCK);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009740 ctx->user = user;
Jens Axboe0b8c0ec2019-12-02 08:50:00 -07009741 ctx->creds = get_current_cred();
Jens Axboe4ea33a92020-10-15 13:46:44 -06009742#ifdef CONFIG_AUDIT
9743 ctx->loginuid = current->loginuid;
9744 ctx->sessionid = current->sessionid;
9745#endif
Jens Axboe2aede0e2020-09-14 10:45:53 -06009746 ctx->sqo_task = get_task_struct(current);
9747
9748 /*
9749 * This is just grabbed for accounting purposes. When a process exits,
9750 * the mm is exited and dropped before the files, hence we need to hang
9751 * on to this mm purely for the purposes of being able to unaccount
9752 * memory (locked/pinned vm). It's not used for anything else.
9753 */
Jens Axboe6b7898e2020-08-25 07:58:00 -06009754 mmgrab(current->mm);
Jens Axboe2aede0e2020-09-14 10:45:53 -06009755 ctx->mm_account = current->mm;
Jens Axboe6b7898e2020-08-25 07:58:00 -06009756
Dennis Zhou91d8f512020-09-16 13:41:05 -07009757#ifdef CONFIG_BLK_CGROUP
9758 /*
9759 * The sq thread will belong to the original cgroup it was inited in.
9760 * If the cgroup goes offline (e.g. disabling the io controller), then
9761 * issued bios will be associated with the closest cgroup later in the
9762 * block layer.
9763 */
9764 rcu_read_lock();
9765 ctx->sqo_blkcg_css = blkcg_css();
9766 ret = css_tryget_online(ctx->sqo_blkcg_css);
9767 rcu_read_unlock();
9768 if (!ret) {
9769 /* don't init against a dying cgroup, have the user try again */
9770 ctx->sqo_blkcg_css = NULL;
9771 ret = -ENODEV;
9772 goto err;
9773 }
9774#endif
Jens Axboe2b188cc2019-01-07 10:46:33 -07009775 ret = io_allocate_scq_urings(ctx, p);
9776 if (ret)
9777 goto err;
9778
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009779 ret = io_sq_offload_create(ctx, p);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009780 if (ret)
9781 goto err;
9782
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009783 if (!(p->flags & IORING_SETUP_R_DISABLED))
9784 io_sq_offload_start(ctx);
9785
Jens Axboe2b188cc2019-01-07 10:46:33 -07009786 memset(&p->sq_off, 0, sizeof(p->sq_off));
Hristo Venev75b28af2019-08-26 17:23:46 +00009787 p->sq_off.head = offsetof(struct io_rings, sq.head);
9788 p->sq_off.tail = offsetof(struct io_rings, sq.tail);
9789 p->sq_off.ring_mask = offsetof(struct io_rings, sq_ring_mask);
9790 p->sq_off.ring_entries = offsetof(struct io_rings, sq_ring_entries);
9791 p->sq_off.flags = offsetof(struct io_rings, sq_flags);
9792 p->sq_off.dropped = offsetof(struct io_rings, sq_dropped);
9793 p->sq_off.array = (char *)ctx->sq_array - (char *)ctx->rings;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009794
9795 memset(&p->cq_off, 0, sizeof(p->cq_off));
Hristo Venev75b28af2019-08-26 17:23:46 +00009796 p->cq_off.head = offsetof(struct io_rings, cq.head);
9797 p->cq_off.tail = offsetof(struct io_rings, cq.tail);
9798 p->cq_off.ring_mask = offsetof(struct io_rings, cq_ring_mask);
9799 p->cq_off.ring_entries = offsetof(struct io_rings, cq_ring_entries);
9800 p->cq_off.overflow = offsetof(struct io_rings, cq_overflow);
9801 p->cq_off.cqes = offsetof(struct io_rings, cqes);
Stefano Garzarella0d9b5b32020-05-15 18:38:04 +02009802 p->cq_off.flags = offsetof(struct io_rings, cq_flags);
Jens Axboeac90f242019-09-06 10:26:21 -06009803
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009804 p->features = IORING_FEAT_SINGLE_MMAP | IORING_FEAT_NODROP |
9805 IORING_FEAT_SUBMIT_STABLE | IORING_FEAT_RW_CUR_POS |
Jiufei Xue5769a352020-06-17 17:53:55 +08009806 IORING_FEAT_CUR_PERSONALITY | IORING_FEAT_FAST_POLL |
Hao Xuc73ebb62020-11-03 10:54:37 +08009807 IORING_FEAT_POLL_32BITS | IORING_FEAT_SQPOLL_NONFIXED |
9808 IORING_FEAT_EXT_ARG;
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009809
9810 if (copy_to_user(params, p, sizeof(*p))) {
9811 ret = -EFAULT;
9812 goto err;
9813 }
Jens Axboed1719f72020-07-30 13:43:53 -06009814
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009815 file = io_uring_get_file(ctx);
9816 if (IS_ERR(file)) {
9817 ret = PTR_ERR(file);
9818 goto err;
9819 }
9820
Jens Axboed1719f72020-07-30 13:43:53 -06009821 /*
Jens Axboe044c1ab2019-10-28 09:15:33 -06009822 * Install ring fd as the very last thing, so we don't risk someone
9823 * having closed it before we finish setup
9824 */
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009825 ret = io_uring_install_fd(ctx, file);
9826 if (ret < 0) {
Pavel Begunkov06585c42021-01-13 12:42:25 +00009827 io_disable_sqo_submit(ctx);
Pavel Begunkov9faadcc2020-12-21 18:34:05 +00009828 /* fput will clean it up */
9829 fput(file);
9830 return ret;
9831 }
Jens Axboe044c1ab2019-10-28 09:15:33 -06009832
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +02009833 trace_io_uring_create(ret, ctx, p->sq_entries, p->cq_entries, p->flags);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009834 return ret;
9835err:
Pavel Begunkovd9d05212021-01-08 20:57:25 +00009836 io_disable_sqo_submit(ctx);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009837 io_ring_ctx_wait_and_kill(ctx);
9838 return ret;
9839}
9840
9841/*
9842 * Sets up an aio uring context, and returns the fd. Applications asks for a
9843 * ring size, we return the actual sq/cq ring sizes (among other things) in the
9844 * params structure passed in.
9845 */
9846static long io_uring_setup(u32 entries, struct io_uring_params __user *params)
9847{
9848 struct io_uring_params p;
Jens Axboe2b188cc2019-01-07 10:46:33 -07009849 int i;
9850
9851 if (copy_from_user(&p, params, sizeof(p)))
9852 return -EFAULT;
9853 for (i = 0; i < ARRAY_SIZE(p.resv); i++) {
9854 if (p.resv[i])
9855 return -EINVAL;
9856 }
9857
Jens Axboe6c271ce2019-01-10 11:22:30 -07009858 if (p.flags & ~(IORING_SETUP_IOPOLL | IORING_SETUP_SQPOLL |
Jens Axboe8110c1a2019-12-28 15:39:54 -07009859 IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE |
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009860 IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ |
9861 IORING_SETUP_R_DISABLED))
Jens Axboe2b188cc2019-01-07 10:46:33 -07009862 return -EINVAL;
9863
Xiaoguang Wang7f136572020-05-05 16:28:53 +08009864 return io_uring_create(entries, &p, params);
Jens Axboe2b188cc2019-01-07 10:46:33 -07009865}
9866
9867SYSCALL_DEFINE2(io_uring_setup, u32, entries,
9868 struct io_uring_params __user *, params)
9869{
9870 return io_uring_setup(entries, params);
9871}
9872
Jens Axboe66f4af92020-01-16 15:36:52 -07009873static int io_probe(struct io_ring_ctx *ctx, void __user *arg, unsigned nr_args)
9874{
9875 struct io_uring_probe *p;
9876 size_t size;
9877 int i, ret;
9878
9879 size = struct_size(p, ops, nr_args);
9880 if (size == SIZE_MAX)
9881 return -EOVERFLOW;
9882 p = kzalloc(size, GFP_KERNEL);
9883 if (!p)
9884 return -ENOMEM;
9885
9886 ret = -EFAULT;
9887 if (copy_from_user(p, arg, size))
9888 goto out;
9889 ret = -EINVAL;
9890 if (memchr_inv(p, 0, size))
9891 goto out;
9892
9893 p->last_op = IORING_OP_LAST - 1;
9894 if (nr_args > IORING_OP_LAST)
9895 nr_args = IORING_OP_LAST;
9896
9897 for (i = 0; i < nr_args; i++) {
9898 p->ops[i].op = i;
9899 if (!io_op_defs[i].not_supported)
9900 p->ops[i].flags = IO_URING_OP_SUPPORTED;
9901 }
9902 p->ops_len = i;
9903
9904 ret = 0;
9905 if (copy_to_user(arg, p, size))
9906 ret = -EFAULT;
9907out:
9908 kfree(p);
9909 return ret;
9910}
9911
Jens Axboe071698e2020-01-28 10:04:42 -07009912static int io_register_personality(struct io_ring_ctx *ctx)
9913{
Jens Axboe1e6fa522020-10-15 08:46:24 -06009914 struct io_identity *id;
9915 int ret;
Jens Axboe071698e2020-01-28 10:04:42 -07009916
Jens Axboe1e6fa522020-10-15 08:46:24 -06009917 id = kmalloc(sizeof(*id), GFP_KERNEL);
9918 if (unlikely(!id))
9919 return -ENOMEM;
9920
9921 io_init_identity(id);
9922 id->creds = get_current_cred();
9923
9924 ret = idr_alloc_cyclic(&ctx->personality_idr, id, 1, USHRT_MAX, GFP_KERNEL);
9925 if (ret < 0) {
9926 put_cred(id->creds);
9927 kfree(id);
9928 }
9929 return ret;
Jens Axboe071698e2020-01-28 10:04:42 -07009930}
9931
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009932static int io_register_restrictions(struct io_ring_ctx *ctx, void __user *arg,
9933 unsigned int nr_args)
9934{
9935 struct io_uring_restriction *res;
9936 size_t size;
9937 int i, ret;
9938
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009939 /* Restrictions allowed only if rings started disabled */
9940 if (!(ctx->flags & IORING_SETUP_R_DISABLED))
9941 return -EBADFD;
9942
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009943 /* We allow only a single restrictions registration */
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009944 if (ctx->restrictions.registered)
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009945 return -EBUSY;
9946
9947 if (!arg || nr_args > IORING_MAX_RESTRICTIONS)
9948 return -EINVAL;
9949
9950 size = array_size(nr_args, sizeof(*res));
9951 if (size == SIZE_MAX)
9952 return -EOVERFLOW;
9953
9954 res = memdup_user(arg, size);
9955 if (IS_ERR(res))
9956 return PTR_ERR(res);
9957
9958 ret = 0;
9959
9960 for (i = 0; i < nr_args; i++) {
9961 switch (res[i].opcode) {
9962 case IORING_RESTRICTION_REGISTER_OP:
9963 if (res[i].register_op >= IORING_REGISTER_LAST) {
9964 ret = -EINVAL;
9965 goto out;
9966 }
9967
9968 __set_bit(res[i].register_op,
9969 ctx->restrictions.register_op);
9970 break;
9971 case IORING_RESTRICTION_SQE_OP:
9972 if (res[i].sqe_op >= IORING_OP_LAST) {
9973 ret = -EINVAL;
9974 goto out;
9975 }
9976
9977 __set_bit(res[i].sqe_op, ctx->restrictions.sqe_op);
9978 break;
9979 case IORING_RESTRICTION_SQE_FLAGS_ALLOWED:
9980 ctx->restrictions.sqe_flags_allowed = res[i].sqe_flags;
9981 break;
9982 case IORING_RESTRICTION_SQE_FLAGS_REQUIRED:
9983 ctx->restrictions.sqe_flags_required = res[i].sqe_flags;
9984 break;
9985 default:
9986 ret = -EINVAL;
9987 goto out;
9988 }
9989 }
9990
9991out:
9992 /* Reset all restrictions if an error happened */
9993 if (ret != 0)
9994 memset(&ctx->restrictions, 0, sizeof(ctx->restrictions));
9995 else
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +02009996 ctx->restrictions.registered = true;
Stefano Garzarella21b55db2020-08-27 16:58:30 +02009997
9998 kfree(res);
9999 return ret;
10000}
10001
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +020010002static int io_register_enable_rings(struct io_ring_ctx *ctx)
10003{
10004 if (!(ctx->flags & IORING_SETUP_R_DISABLED))
10005 return -EBADFD;
10006
10007 if (ctx->restrictions.registered)
10008 ctx->restricted = 1;
10009
10010 ctx->flags &= ~IORING_SETUP_R_DISABLED;
10011
10012 io_sq_offload_start(ctx);
10013
10014 return 0;
10015}
10016
Jens Axboe071698e2020-01-28 10:04:42 -070010017static bool io_register_op_must_quiesce(int op)
10018{
10019 switch (op) {
10020 case IORING_UNREGISTER_FILES:
10021 case IORING_REGISTER_FILES_UPDATE:
10022 case IORING_REGISTER_PROBE:
10023 case IORING_REGISTER_PERSONALITY:
10024 case IORING_UNREGISTER_PERSONALITY:
10025 return false;
10026 default:
10027 return true;
10028 }
10029}
10030
Jens Axboeedafcce2019-01-09 09:16:05 -070010031static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode,
10032 void __user *arg, unsigned nr_args)
Jens Axboeb19062a2019-04-15 10:49:38 -060010033 __releases(ctx->uring_lock)
10034 __acquires(ctx->uring_lock)
Jens Axboeedafcce2019-01-09 09:16:05 -070010035{
10036 int ret;
10037
Jens Axboe35fa71a2019-04-22 10:23:23 -060010038 /*
10039 * We're inside the ring mutex, if the ref is already dying, then
10040 * someone else killed the ctx or is already going through
10041 * io_uring_register().
10042 */
10043 if (percpu_ref_is_dying(&ctx->refs))
10044 return -ENXIO;
10045
Jens Axboe071698e2020-01-28 10:04:42 -070010046 if (io_register_op_must_quiesce(opcode)) {
Jens Axboe05f3fb32019-12-09 11:22:50 -070010047 percpu_ref_kill(&ctx->refs);
Jens Axboeb19062a2019-04-15 10:49:38 -060010048
Jens Axboe05f3fb32019-12-09 11:22:50 -070010049 /*
10050 * Drop uring mutex before waiting for references to exit. If
10051 * another thread is currently inside io_uring_enter() it might
10052 * need to grab the uring_lock to make progress. If we hold it
10053 * here across the drain wait, then we can deadlock. It's safe
10054 * to drop the mutex here, since no new references will come in
10055 * after we've killed the percpu ref.
10056 */
10057 mutex_unlock(&ctx->uring_lock);
Jens Axboeaf9c1a42020-09-24 13:32:18 -060010058 do {
10059 ret = wait_for_completion_interruptible(&ctx->ref_comp);
10060 if (!ret)
10061 break;
Jens Axboeed6930c2020-10-08 19:09:46 -060010062 ret = io_run_task_work_sig();
10063 if (ret < 0)
10064 break;
Jens Axboeaf9c1a42020-09-24 13:32:18 -060010065 } while (1);
10066
Jens Axboe05f3fb32019-12-09 11:22:50 -070010067 mutex_lock(&ctx->uring_lock);
Jens Axboeaf9c1a42020-09-24 13:32:18 -060010068
Jens Axboec1503682020-01-08 08:26:07 -070010069 if (ret) {
10070 percpu_ref_resurrect(&ctx->refs);
Stefano Garzarella21b55db2020-08-27 16:58:30 +020010071 goto out_quiesce;
10072 }
10073 }
10074
10075 if (ctx->restricted) {
10076 if (opcode >= IORING_REGISTER_LAST) {
10077 ret = -EINVAL;
10078 goto out;
10079 }
10080
10081 if (!test_bit(opcode, ctx->restrictions.register_op)) {
10082 ret = -EACCES;
Jens Axboec1503682020-01-08 08:26:07 -070010083 goto out;
10084 }
Jens Axboe05f3fb32019-12-09 11:22:50 -070010085 }
Jens Axboeedafcce2019-01-09 09:16:05 -070010086
10087 switch (opcode) {
10088 case IORING_REGISTER_BUFFERS:
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -080010089 ret = io_sqe_buffers_register(ctx, arg, nr_args);
Jens Axboeedafcce2019-01-09 09:16:05 -070010090 break;
10091 case IORING_UNREGISTER_BUFFERS:
10092 ret = -EINVAL;
10093 if (arg || nr_args)
10094 break;
Bijan Mottahedeh0a96bbe2021-01-06 12:39:10 -080010095 ret = io_sqe_buffers_unregister(ctx);
Jens Axboeedafcce2019-01-09 09:16:05 -070010096 break;
Jens Axboe6b063142019-01-10 22:13:58 -070010097 case IORING_REGISTER_FILES:
10098 ret = io_sqe_files_register(ctx, arg, nr_args);
10099 break;
10100 case IORING_UNREGISTER_FILES:
10101 ret = -EINVAL;
10102 if (arg || nr_args)
10103 break;
10104 ret = io_sqe_files_unregister(ctx);
10105 break;
Jens Axboec3a31e62019-10-03 13:59:56 -060010106 case IORING_REGISTER_FILES_UPDATE:
10107 ret = io_sqe_files_update(ctx, arg, nr_args);
10108 break;
Jens Axboe9b402842019-04-11 11:45:41 -060010109 case IORING_REGISTER_EVENTFD:
Jens Axboef2842ab2020-01-08 11:04:00 -070010110 case IORING_REGISTER_EVENTFD_ASYNC:
Jens Axboe9b402842019-04-11 11:45:41 -060010111 ret = -EINVAL;
10112 if (nr_args != 1)
10113 break;
10114 ret = io_eventfd_register(ctx, arg);
Jens Axboef2842ab2020-01-08 11:04:00 -070010115 if (ret)
10116 break;
10117 if (opcode == IORING_REGISTER_EVENTFD_ASYNC)
10118 ctx->eventfd_async = 1;
10119 else
10120 ctx->eventfd_async = 0;
Jens Axboe9b402842019-04-11 11:45:41 -060010121 break;
10122 case IORING_UNREGISTER_EVENTFD:
10123 ret = -EINVAL;
10124 if (arg || nr_args)
10125 break;
10126 ret = io_eventfd_unregister(ctx);
10127 break;
Jens Axboe66f4af92020-01-16 15:36:52 -070010128 case IORING_REGISTER_PROBE:
10129 ret = -EINVAL;
10130 if (!arg || nr_args > 256)
10131 break;
10132 ret = io_probe(ctx, arg, nr_args);
10133 break;
Jens Axboe071698e2020-01-28 10:04:42 -070010134 case IORING_REGISTER_PERSONALITY:
10135 ret = -EINVAL;
10136 if (arg || nr_args)
10137 break;
10138 ret = io_register_personality(ctx);
10139 break;
10140 case IORING_UNREGISTER_PERSONALITY:
10141 ret = -EINVAL;
10142 if (arg)
10143 break;
10144 ret = io_unregister_personality(ctx, nr_args);
10145 break;
Stefano Garzarella7e84e1c2020-08-27 16:58:31 +020010146 case IORING_REGISTER_ENABLE_RINGS:
10147 ret = -EINVAL;
10148 if (arg || nr_args)
10149 break;
10150 ret = io_register_enable_rings(ctx);
10151 break;
Stefano Garzarella21b55db2020-08-27 16:58:30 +020010152 case IORING_REGISTER_RESTRICTIONS:
10153 ret = io_register_restrictions(ctx, arg, nr_args);
10154 break;
Jens Axboeedafcce2019-01-09 09:16:05 -070010155 default:
10156 ret = -EINVAL;
10157 break;
10158 }
10159
Stefano Garzarella21b55db2020-08-27 16:58:30 +020010160out:
Jens Axboe071698e2020-01-28 10:04:42 -070010161 if (io_register_op_must_quiesce(opcode)) {
Jens Axboe05f3fb32019-12-09 11:22:50 -070010162 /* bring the ctx back to life */
Jens Axboe05f3fb32019-12-09 11:22:50 -070010163 percpu_ref_reinit(&ctx->refs);
Stefano Garzarella21b55db2020-08-27 16:58:30 +020010164out_quiesce:
Jens Axboe0f158b42020-05-14 17:18:39 -060010165 reinit_completion(&ctx->ref_comp);
Jens Axboe05f3fb32019-12-09 11:22:50 -070010166 }
Jens Axboeedafcce2019-01-09 09:16:05 -070010167 return ret;
10168}
10169
10170SYSCALL_DEFINE4(io_uring_register, unsigned int, fd, unsigned int, opcode,
10171 void __user *, arg, unsigned int, nr_args)
10172{
10173 struct io_ring_ctx *ctx;
10174 long ret = -EBADF;
10175 struct fd f;
10176
10177 f = fdget(fd);
10178 if (!f.file)
10179 return -EBADF;
10180
10181 ret = -EOPNOTSUPP;
10182 if (f.file->f_op != &io_uring_fops)
10183 goto out_fput;
10184
10185 ctx = f.file->private_data;
10186
10187 mutex_lock(&ctx->uring_lock);
10188 ret = __io_uring_register(ctx, opcode, arg, nr_args);
10189 mutex_unlock(&ctx->uring_lock);
Dmitrii Dolgovc826bd72019-10-15 19:02:01 +020010190 trace_io_uring_register(ctx, opcode, ctx->nr_user_files, ctx->nr_user_bufs,
10191 ctx->cq_ev_fd != NULL, ret);
Jens Axboeedafcce2019-01-09 09:16:05 -070010192out_fput:
10193 fdput(f);
10194 return ret;
10195}
10196
Jens Axboe2b188cc2019-01-07 10:46:33 -070010197static int __init io_uring_init(void)
10198{
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +010010199#define __BUILD_BUG_VERIFY_ELEMENT(stype, eoffset, etype, ename) do { \
10200 BUILD_BUG_ON(offsetof(stype, ename) != eoffset); \
10201 BUILD_BUG_ON(sizeof(etype) != sizeof_field(stype, ename)); \
10202} while (0)
10203
10204#define BUILD_BUG_SQE_ELEM(eoffset, etype, ename) \
10205 __BUILD_BUG_VERIFY_ELEMENT(struct io_uring_sqe, eoffset, etype, ename)
10206 BUILD_BUG_ON(sizeof(struct io_uring_sqe) != 64);
10207 BUILD_BUG_SQE_ELEM(0, __u8, opcode);
10208 BUILD_BUG_SQE_ELEM(1, __u8, flags);
10209 BUILD_BUG_SQE_ELEM(2, __u16, ioprio);
10210 BUILD_BUG_SQE_ELEM(4, __s32, fd);
10211 BUILD_BUG_SQE_ELEM(8, __u64, off);
10212 BUILD_BUG_SQE_ELEM(8, __u64, addr2);
10213 BUILD_BUG_SQE_ELEM(16, __u64, addr);
Pavel Begunkov7d67af22020-02-24 11:32:45 +030010214 BUILD_BUG_SQE_ELEM(16, __u64, splice_off_in);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +010010215 BUILD_BUG_SQE_ELEM(24, __u32, len);
10216 BUILD_BUG_SQE_ELEM(28, __kernel_rwf_t, rw_flags);
10217 BUILD_BUG_SQE_ELEM(28, /* compat */ int, rw_flags);
10218 BUILD_BUG_SQE_ELEM(28, /* compat */ __u32, rw_flags);
10219 BUILD_BUG_SQE_ELEM(28, __u32, fsync_flags);
Jiufei Xue5769a352020-06-17 17:53:55 +080010220 BUILD_BUG_SQE_ELEM(28, /* compat */ __u16, poll_events);
10221 BUILD_BUG_SQE_ELEM(28, __u32, poll32_events);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +010010222 BUILD_BUG_SQE_ELEM(28, __u32, sync_range_flags);
10223 BUILD_BUG_SQE_ELEM(28, __u32, msg_flags);
10224 BUILD_BUG_SQE_ELEM(28, __u32, timeout_flags);
10225 BUILD_BUG_SQE_ELEM(28, __u32, accept_flags);
10226 BUILD_BUG_SQE_ELEM(28, __u32, cancel_flags);
10227 BUILD_BUG_SQE_ELEM(28, __u32, open_flags);
10228 BUILD_BUG_SQE_ELEM(28, __u32, statx_flags);
10229 BUILD_BUG_SQE_ELEM(28, __u32, fadvise_advice);
Pavel Begunkov7d67af22020-02-24 11:32:45 +030010230 BUILD_BUG_SQE_ELEM(28, __u32, splice_flags);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +010010231 BUILD_BUG_SQE_ELEM(32, __u64, user_data);
10232 BUILD_BUG_SQE_ELEM(40, __u16, buf_index);
10233 BUILD_BUG_SQE_ELEM(42, __u16, personality);
Pavel Begunkov7d67af22020-02-24 11:32:45 +030010234 BUILD_BUG_SQE_ELEM(44, __s32, splice_fd_in);
Stefan Metzmacherd7f62e82020-01-29 14:39:41 +010010235
Jens Axboed3656342019-12-18 09:50:26 -070010236 BUILD_BUG_ON(ARRAY_SIZE(io_op_defs) != IORING_OP_LAST);
Jens Axboe84557872020-03-03 15:28:17 -070010237 BUILD_BUG_ON(__REQ_F_LAST_BIT >= 8 * sizeof(int));
Jens Axboe91f245d2021-02-09 13:48:50 -070010238 req_cachep = KMEM_CACHE(io_kiocb, SLAB_HWCACHE_ALIGN | SLAB_PANIC |
10239 SLAB_ACCOUNT);
Jens Axboe2b188cc2019-01-07 10:46:33 -070010240 return 0;
10241};
10242__initcall(io_uring_init);