1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Contains the core associated with submission side polling of the SQ
4 * ring, offloading submissions from the application to a kernel thread.
5 */
6 #include <linux/kernel.h>
7 #include <linux/errno.h>
8 #include <linux/file.h>
9 #include <linux/mm.h>
10 #include <linux/slab.h>
11 #include <linux/audit.h>
12 #include <linux/security.h>
13 #include <linux/io_uring.h>
14
15 #include <uapi/linux/io_uring.h>
16
17 #include "io_uring.h"
18 #include "napi.h"
19 #include "sqpoll.h"
20
21 #define IORING_SQPOLL_CAP_ENTRIES_VALUE 8
22 #define IORING_TW_CAP_ENTRIES_VALUE 8
23
24 enum {
25 IO_SQ_THREAD_SHOULD_STOP = 0,
26 IO_SQ_THREAD_SHOULD_PARK,
27 };
28
29 void io_sq_thread_unpark(struct io_sq_data *sqd)
30 __releases(&sqd->lock)
31 {
32 WARN_ON_ONCE(sqd->thread == current);
33
34 /*
35 * Do the dance but not conditional clear_bit() because it'd race with
36 * other threads incrementing park_pending and setting the bit.
37 */
38 clear_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
39 if (atomic_dec_return(&sqd->park_pending))
40 set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
41 mutex_unlock(&sqd->lock);
42 }
43
44 void io_sq_thread_park(struct io_sq_data *sqd)
45 __acquires(&sqd->lock)
46 {
47 WARN_ON_ONCE(sqd->thread == current);
48
49 atomic_inc(&sqd->park_pending);
50 set_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state);
51 mutex_lock(&sqd->lock);
52 if (sqd->thread)
53 wake_up_process(sqd->thread);
54 }
55
56 void io_sq_thread_stop(struct io_sq_data *sqd)
57 {
58 WARN_ON_ONCE(sqd->thread == current);
59 WARN_ON_ONCE(test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state));
60
61 set_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state);
62 mutex_lock(&sqd->lock);
63 if (sqd->thread)
64 wake_up_process(sqd->thread);
65 mutex_unlock(&sqd->lock);
66 wait_for_completion(&sqd->exited);
67 }
68
69 void io_put_sq_data(struct io_sq_data *sqd)
70 {
71 if (refcount_dec_and_test(&sqd->refs)) {
72 WARN_ON_ONCE(atomic_read(&sqd->park_pending));
73
74 io_sq_thread_stop(sqd);
75 kfree(sqd);
76 }
77 }
78
79 static __cold void io_sqd_update_thread_idle(struct io_sq_data *sqd)
80 {
81 struct io_ring_ctx *ctx;
82 unsigned sq_thread_idle = 0;
83
84 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
85 sq_thread_idle = max(sq_thread_idle, ctx->sq_thread_idle);
86 sqd->sq_thread_idle = sq_thread_idle;
87 }
88
89 void io_sq_thread_finish(struct io_ring_ctx *ctx)
90 {
91 struct io_sq_data *sqd = ctx->sq_data;
92
93 if (sqd) {
94 io_sq_thread_park(sqd);
95 list_del_init(&ctx->sqd_list);
96 io_sqd_update_thread_idle(sqd);
97 io_sq_thread_unpark(sqd);
98
99 io_put_sq_data(sqd);
100 ctx->sq_data = NULL;
101 }
102 }
103
104 static struct io_sq_data *io_attach_sq_data(struct io_uring_params *p)
105 {
106 struct io_ring_ctx *ctx_attach;
107 struct io_sq_data *sqd;
108 struct fd f;
109
110 f = fdget(p->wq_fd);
111 if (!f.file)
112 return ERR_PTR(-ENXIO);
113 if (!io_is_uring_fops(f.file)) {
114 fdput(f);
115 return ERR_PTR(-EINVAL);
116 }
117
118 ctx_attach = f.file->private_data;
119 sqd = ctx_attach->sq_data;
120 if (!sqd) {
121 fdput(f);
122 return ERR_PTR(-EINVAL);
123 }
124 if (sqd->task_tgid != current->tgid) {
125 fdput(f);
126 return ERR_PTR(-EPERM);
127 }
128
129 refcount_inc(&sqd->refs);
130 fdput(f);
131 return sqd;
132 }
133
134 static struct io_sq_data *io_get_sq_data(struct io_uring_params *p,
135 bool *attached)
136 {
137 struct io_sq_data *sqd;
138
139 *attached = false;
140 if (p->flags & IORING_SETUP_ATTACH_WQ) {
141 sqd = io_attach_sq_data(p);
142 if (!IS_ERR(sqd)) {
143 *attached = true;
144 return sqd;
145 }
146 /* fall through for EPERM case, setup new sqd/task */
147 if (PTR_ERR(sqd) != -EPERM)
148 return sqd;
149 }
150
151 sqd = kzalloc(sizeof(*sqd), GFP_KERNEL);
152 if (!sqd)
153 return ERR_PTR(-ENOMEM);
154
155 atomic_set(&sqd->park_pending, 0);
156 refcount_set(&sqd->refs, 1);
157 INIT_LIST_HEAD(&sqd->ctx_list);
158 mutex_init(&sqd->lock);
159 init_waitqueue_head(&sqd->wait);
160 init_completion(&sqd->exited);
161 return sqd;
162 }
163
164 static inline bool io_sqd_events_pending(struct io_sq_data *sqd)
165 {
166 return READ_ONCE(sqd->state);
167 }
168
169 static int __io_sq_thread(struct io_ring_ctx *ctx, bool cap_entries)
170 {
171 unsigned int to_submit;
172 int ret = 0;
173
174 to_submit = io_sqring_entries(ctx);
175 /* if we're handling multiple rings, cap submit size for fairness */
176 if (cap_entries && to_submit > IORING_SQPOLL_CAP_ENTRIES_VALUE)
177 to_submit = IORING_SQPOLL_CAP_ENTRIES_VALUE;
178
179 if (!wq_list_empty(&ctx->iopoll_list) || to_submit) {
180 const struct cred *creds = NULL;
181
182 if (ctx->sq_creds != current_cred())
183 creds = override_creds(ctx->sq_creds);
184
185 mutex_lock(&ctx->uring_lock);
186 if (!wq_list_empty(&ctx->iopoll_list))
187 io_do_iopoll(ctx, true);
188
189 /*
190 * Don't submit if refs are dying, good for io_uring_register(),
191 * but also it is relied upon by io_ring_exit_work()
192 */
193 if (to_submit && likely(!percpu_ref_is_dying(&ctx->refs)) &&
194 !(ctx->flags & IORING_SETUP_R_DISABLED))
195 ret = io_submit_sqes(ctx, to_submit);
196 mutex_unlock(&ctx->uring_lock);
197
198 if (io_napi(ctx))
199 ret += io_napi_sqpoll_busy_poll(ctx);
200
201 if (to_submit && wq_has_sleeper(&ctx->sqo_sq_wait))
202 wake_up(&ctx->sqo_sq_wait);
203 if (creds)
204 revert_creds(creds);
205 }
206
207 return ret;
208 }
209
210 static bool io_sqd_handle_event(struct io_sq_data *sqd)
211 {
212 bool did_sig = false;
213 struct ksignal ksig;
214
215 if (test_bit(IO_SQ_THREAD_SHOULD_PARK, &sqd->state) ||
216 signal_pending(current)) {
217 mutex_unlock(&sqd->lock);
218 if (signal_pending(current))
219 did_sig = get_signal(&ksig);
220 cond_resched();
221 mutex_lock(&sqd->lock);
222 sqd->sq_cpu = raw_smp_processor_id();
223 }
224 return did_sig || test_bit(IO_SQ_THREAD_SHOULD_STOP, &sqd->state);
225 }
226
227 /*
228 * Run task_work, processing the retry_list first. The retry_list holds
229 * entries that we passed on in the previous run, if we had more task_work
230 * than we were asked to process. Newly queued task_work isn't run until the
231 * retry list has been fully processed.
232 */
233 static unsigned int io_sq_tw(struct llist_node **retry_list, int max_entries)
234 {
235 struct io_uring_task *tctx = current->io_uring;
236 unsigned int count = 0;
237
238 if (*retry_list) {
239 *retry_list = io_handle_tw_list(*retry_list, &count, max_entries);
240 if (count >= max_entries)
241 goto out;
242 max_entries -= count;
243 }
244 *retry_list = tctx_task_work_run(tctx, max_entries, &count);
245 out:
246 if (task_work_pending(current))
247 task_work_run();
248 return count;
249 }
250
251 static bool io_sq_tw_pending(struct llist_node *retry_list)
252 {
253 struct io_uring_task *tctx = current->io_uring;
254
255 return retry_list || !llist_empty(&tctx->task_list);
256 }
257
258 static void io_sq_update_worktime(struct io_sq_data *sqd, struct rusage *start)
259 {
260 struct rusage end;
261
262 getrusage(current, RUSAGE_SELF, &end);
263 end.ru_stime.tv_sec -= start->ru_stime.tv_sec;
264 end.ru_stime.tv_usec -= start->ru_stime.tv_usec;
265
266 sqd->work_time += end.ru_stime.tv_usec + end.ru_stime.tv_sec * 1000000;
267 }
268
269 static int io_sq_thread(void *data)
270 {
271 struct llist_node *retry_list = NULL;
272 struct io_sq_data *sqd = data;
273 struct io_ring_ctx *ctx;
274 struct rusage start;
275 unsigned long timeout = 0;
276 char buf[TASK_COMM_LEN];
277 DEFINE_WAIT(wait);
278
279 /* offload context creation failed, just exit */
280 if (!current->io_uring)
281 goto err_out;
282
283 snprintf(buf, sizeof(buf), "iou-sqp-%d", sqd->task_pid);
284 set_task_comm(current, buf);
285
286 /* reset to our pid after we've set task_comm, for fdinfo */
287 sqd->task_pid = current->pid;
288
289 if (sqd->sq_cpu != -1) {
290 set_cpus_allowed_ptr(current, cpumask_of(sqd->sq_cpu));
291 } else {
292 set_cpus_allowed_ptr(current, cpu_online_mask);
293 sqd->sq_cpu = raw_smp_processor_id();
294 }
295
296 /*
297 * Force audit context to get setup, in case we do prep side async
298 * operations that would trigger an audit call before any issue side
299 * audit has been done.
300 */
301 audit_uring_entry(IORING_OP_NOP);
302 audit_uring_exit(true, 0);
303
304 mutex_lock(&sqd->lock);
305 while (1) {
306 bool cap_entries, sqt_spin = false;
307
308 if (io_sqd_events_pending(sqd) || signal_pending(current)) {
309 if (io_sqd_handle_event(sqd))
310 break;
311 timeout = jiffies + sqd->sq_thread_idle;
312 }
313
314 cap_entries = !list_is_singular(&sqd->ctx_list);
315 getrusage(current, RUSAGE_SELF, &start);
316 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
317 int ret = __io_sq_thread(ctx, cap_entries);
318
319 if (!sqt_spin && (ret > 0 || !wq_list_empty(&ctx->iopoll_list)))
320 sqt_spin = true;
321 }
322 if (io_sq_tw(&retry_list, IORING_TW_CAP_ENTRIES_VALUE))
323 sqt_spin = true;
324
325 if (sqt_spin || !time_after(jiffies, timeout)) {
326 if (sqt_spin) {
327 io_sq_update_worktime(sqd, &start);
328 timeout = jiffies + sqd->sq_thread_idle;
329 }
330 if (unlikely(need_resched())) {
331 mutex_unlock(&sqd->lock);
332 cond_resched();
333 mutex_lock(&sqd->lock);
334 sqd->sq_cpu = raw_smp_processor_id();
335 }
336 continue;
337 }
338
339 prepare_to_wait(&sqd->wait, &wait, TASK_INTERRUPTIBLE);
340 if (!io_sqd_events_pending(sqd) && !io_sq_tw_pending(retry_list)) {
341 bool needs_sched = true;
342
343 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list) {
344 atomic_or(IORING_SQ_NEED_WAKEUP,
345 &ctx->rings->sq_flags);
346 if ((ctx->flags & IORING_SETUP_IOPOLL) &&
347 !wq_list_empty(&ctx->iopoll_list)) {
348 needs_sched = false;
349 break;
350 }
351
352 /*
353 * Ensure the store of the wakeup flag is not
354 * reordered with the load of the SQ tail
355 */
356 smp_mb__after_atomic();
357
358 if (io_sqring_entries(ctx)) {
359 needs_sched = false;
360 break;
361 }
362 }
363
364 if (needs_sched) {
365 mutex_unlock(&sqd->lock);
366 schedule();
367 mutex_lock(&sqd->lock);
368 sqd->sq_cpu = raw_smp_processor_id();
369 }
370 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
371 atomic_andnot(IORING_SQ_NEED_WAKEUP,
372 &ctx->rings->sq_flags);
373 }
374
375 finish_wait(&sqd->wait, &wait);
376 timeout = jiffies + sqd->sq_thread_idle;
377 }
378
379 if (retry_list)
380 io_sq_tw(&retry_list, UINT_MAX);
381
382 io_uring_cancel_generic(true, sqd);
383 sqd->thread = NULL;
384 list_for_each_entry(ctx, &sqd->ctx_list, sqd_list)
385 atomic_or(IORING_SQ_NEED_WAKEUP, &ctx->rings->sq_flags);
386 io_run_task_work();
387 mutex_unlock(&sqd->lock);
388 err_out:
389 complete(&sqd->exited);
390 do_exit(0);
391 }
392
393 void io_sqpoll_wait_sq(struct io_ring_ctx *ctx)
394 {
395 DEFINE_WAIT(wait);
396
397 do {
398 if (!io_sqring_full(ctx))
399 break;
400 prepare_to_wait(&ctx->sqo_sq_wait, &wait, TASK_INTERRUPTIBLE);
401
402 if (!io_sqring_full(ctx))
403 break;
404 schedule();
405 } while (!signal_pending(current));
406
407 finish_wait(&ctx->sqo_sq_wait, &wait);
408 }
409
410 __cold int io_sq_offload_create(struct io_ring_ctx *ctx,
411 struct io_uring_params *p)
412 {
413 int ret;
414
415 /* Retain compatibility with failing for an invalid attach attempt */
416 if ((ctx->flags & (IORING_SETUP_ATTACH_WQ | IORING_SETUP_SQPOLL)) ==
417 IORING_SETUP_ATTACH_WQ) {
418 struct fd f;
419
420 f = fdget(p->wq_fd);
421 if (!f.file)
422 return -ENXIO;
423 if (!io_is_uring_fops(f.file)) {
424 fdput(f);
425 return -EINVAL;
426 }
427 fdput(f);
428 }
429 if (ctx->flags & IORING_SETUP_SQPOLL) {
430 struct task_struct *tsk;
431 struct io_sq_data *sqd;
432 bool attached;
433
434 ret = security_uring_sqpoll();
435 if (ret)
436 return ret;
437
438 sqd = io_get_sq_data(p, &attached);
439 if (IS_ERR(sqd)) {
440 ret = PTR_ERR(sqd);
441 goto err;
442 }
443
444 ctx->sq_creds = get_current_cred();
445 ctx->sq_data = sqd;
446 ctx->sq_thread_idle = msecs_to_jiffies(p->sq_thread_idle);
447 if (!ctx->sq_thread_idle)
448 ctx->sq_thread_idle = HZ;
449
450 io_sq_thread_park(sqd);
451 list_add(&ctx->sqd_list, &sqd->ctx_list);
452 io_sqd_update_thread_idle(sqd);
453 /* don't attach to a dying SQPOLL thread, would be racy */
454 ret = (attached && !sqd->thread) ? -ENXIO : 0;
455 io_sq_thread_unpark(sqd);
456
457 if (ret < 0)
458 goto err;
459 if (attached)
460 return 0;
461
462 if (p->flags & IORING_SETUP_SQ_AFF) {
463 int cpu = p->sq_thread_cpu;
464
465 ret = -EINVAL;
466 if (cpu >= nr_cpu_ids || !cpu_online(cpu))
467 goto err_sqpoll;
468 sqd->sq_cpu = cpu;
469 } else {
470 sqd->sq_cpu = -1;
471 }
472
473 sqd->task_pid = current->pid;
474 sqd->task_tgid = current->tgid;
475 tsk = create_io_thread(io_sq_thread, sqd, NUMA_NO_NODE);
476 if (IS_ERR(tsk)) {
477 ret = PTR_ERR(tsk);
478 goto err_sqpoll;
479 }
480
481 sqd->thread = tsk;
482 ret = io_uring_alloc_task_context(tsk, ctx);
483 wake_up_new_task(tsk);
484 if (ret)
485 goto err;
486 } else if (p->flags & IORING_SETUP_SQ_AFF) {
487 /* Can't have SQ_AFF without SQPOLL */
488 ret = -EINVAL;
489 goto err;
490 }
491
492 return 0;
493 err_sqpoll:
494 complete(&ctx->sq_data->exited);
495 err:
496 io_sq_thread_finish(ctx);
497 return ret;
498 }
499
500 __cold int io_sqpoll_wq_cpu_affinity(struct io_ring_ctx *ctx,
501 cpumask_var_t mask)
502 {
503 struct io_sq_data *sqd = ctx->sq_data;
504 int ret = -EINVAL;
505
506 if (sqd) {
507 io_sq_thread_park(sqd);
508 /* Don't set affinity for a dying thread */
509 if (sqd->thread)
510 ret = io_wq_cpu_affinity(sqd->thread->io_uring, mask);
511 io_sq_thread_unpark(sqd);
512 }
513
514 return ret;
515 }
516
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.