1 // SPDX-License-Identifier: GPL-2.0
3 * Basic worker thread pool for io_uring
5 * Copyright (C) 2019 Jens Axboe
8 #include <linux/kernel.h>
9 #include <linux/init.h>
10 #include <linux/errno.h>
11 #include <linux/sched/signal.h>
12 #include <linux/percpu.h>
13 #include <linux/slab.h>
14 #include <linux/rculist_nulls.h>
15 #include <linux/cpu.h>
16 #include <linux/tracehook.h>
17 #include <linux/audit.h>
18 #include <uapi/linux/io_uring.h>
22 #define WORKER_IDLE_TIMEOUT (5 * HZ)
25 IO_WORKER_F_UP = 1, /* up and active */
26 IO_WORKER_F_RUNNING = 2, /* account as running */
27 IO_WORKER_F_FREE = 4, /* worker on free list */
28 IO_WORKER_F_BOUND = 8, /* is doing bounded work */
32 IO_WQ_BIT_EXIT = 0, /* wq exiting */
36 IO_ACCT_STALLED_BIT = 0, /* stalled on hash */
40 * One for each thread in a wqe pool
45 struct hlist_nulls_node nulls_node;
46 struct list_head all_list;
47 struct task_struct *task;
50 struct io_wq_work *cur_work;
51 struct io_wq_work *next_work;
54 struct completion ref_done;
56 unsigned long create_state;
57 struct callback_head create_work;
62 struct work_struct work;
66 #if BITS_PER_LONG == 64
67 #define IO_WQ_HASH_ORDER 6
69 #define IO_WQ_HASH_ORDER 5
72 #define IO_WQ_NR_HASH_BUCKETS (1u << IO_WQ_HASH_ORDER)
79 struct io_wq_work_list work_list;
90 * Per-node worker thread pool
94 struct io_wqe_acct acct[2];
98 struct hlist_nulls_head free_list;
99 struct list_head all_list;
101 struct wait_queue_entry wait;
104 struct io_wq_work *hash_tail[IO_WQ_NR_HASH_BUCKETS];
106 cpumask_var_t cpu_mask;
115 free_work_fn *free_work;
116 io_wq_work_fn *do_work;
118 struct io_wq_hash *hash;
120 atomic_t worker_refs;
121 struct completion worker_done;
123 struct hlist_node cpuhp_node;
125 struct task_struct *task;
127 struct io_wqe *wqes[];
130 static enum cpuhp_state io_wq_online;
132 struct io_cb_cancel_data {
140 static bool create_io_worker(struct io_wq *wq, struct io_wqe *wqe, int index);
141 static void io_wqe_dec_running(struct io_worker *worker);
142 static bool io_acct_cancel_pending_work(struct io_wqe *wqe,
143 struct io_wqe_acct *acct,
144 struct io_cb_cancel_data *match);
145 static void create_worker_cb(struct callback_head *cb);
146 static void io_wq_cancel_tw_create(struct io_wq *wq);
148 static bool io_worker_get(struct io_worker *worker)
150 return refcount_inc_not_zero(&worker->ref);
153 static void io_worker_release(struct io_worker *worker)
155 if (refcount_dec_and_test(&worker->ref))
156 complete(&worker->ref_done);
159 static inline struct io_wqe_acct *io_get_acct(struct io_wqe *wqe, bool bound)
161 return &wqe->acct[bound ? IO_WQ_ACCT_BOUND : IO_WQ_ACCT_UNBOUND];
164 static inline struct io_wqe_acct *io_work_get_acct(struct io_wqe *wqe,
165 struct io_wq_work *work)
167 return io_get_acct(wqe, !(work->flags & IO_WQ_WORK_UNBOUND));
170 static inline struct io_wqe_acct *io_wqe_get_acct(struct io_worker *worker)
172 return io_get_acct(worker->wqe, worker->flags & IO_WORKER_F_BOUND);
175 static void io_worker_ref_put(struct io_wq *wq)
177 if (atomic_dec_and_test(&wq->worker_refs))
178 complete(&wq->worker_done);
181 static void io_worker_cancel_cb(struct io_worker *worker)
183 struct io_wqe_acct *acct = io_wqe_get_acct(worker);
184 struct io_wqe *wqe = worker->wqe;
185 struct io_wq *wq = wqe->wq;
187 atomic_dec(&acct->nr_running);
188 raw_spin_lock(&worker->wqe->lock);
190 raw_spin_unlock(&worker->wqe->lock);
191 io_worker_ref_put(wq);
192 clear_bit_unlock(0, &worker->create_state);
193 io_worker_release(worker);
196 static bool io_task_worker_match(struct callback_head *cb, void *data)
198 struct io_worker *worker;
200 if (cb->func != create_worker_cb)
202 worker = container_of(cb, struct io_worker, create_work);
203 return worker == data;
206 static void io_worker_exit(struct io_worker *worker)
208 struct io_wqe *wqe = worker->wqe;
209 struct io_wq *wq = wqe->wq;
212 struct callback_head *cb = task_work_cancel_match(wq->task,
213 io_task_worker_match, worker);
217 io_worker_cancel_cb(worker);
220 io_worker_release(worker);
221 wait_for_completion(&worker->ref_done);
223 raw_spin_lock(&wqe->lock);
224 if (worker->flags & IO_WORKER_F_FREE)
225 hlist_nulls_del_rcu(&worker->nulls_node);
226 list_del_rcu(&worker->all_list);
228 io_wqe_dec_running(worker);
230 current->flags &= ~PF_IO_WORKER;
232 raw_spin_unlock(&wqe->lock);
234 kfree_rcu(worker, rcu);
235 io_worker_ref_put(wqe->wq);
239 static inline bool io_acct_run_queue(struct io_wqe_acct *acct)
241 if (!wq_list_empty(&acct->work_list) &&
242 !test_bit(IO_ACCT_STALLED_BIT, &acct->flags))
248 * Check head of free list for an available worker. If one isn't available,
249 * caller must create one.
251 static bool io_wqe_activate_free_worker(struct io_wqe *wqe,
252 struct io_wqe_acct *acct)
255 struct hlist_nulls_node *n;
256 struct io_worker *worker;
259 * Iterate free_list and see if we can find an idle worker to
260 * activate. If a given worker is on the free_list but in the process
261 * of exiting, keep trying.
263 hlist_nulls_for_each_entry_rcu(worker, n, &wqe->free_list, nulls_node) {
264 if (!io_worker_get(worker))
266 if (io_wqe_get_acct(worker) != acct) {
267 io_worker_release(worker);
270 if (wake_up_process(worker->task)) {
271 io_worker_release(worker);
274 io_worker_release(worker);
281 * We need a worker. If we find a free one, we're good. If not, and we're
282 * below the max number of workers, create one.
284 static bool io_wqe_create_worker(struct io_wqe *wqe, struct io_wqe_acct *acct)
287 * Most likely an attempt to queue unbounded work on an io_wq that
288 * wasn't setup with any unbounded workers.
290 if (unlikely(!acct->max_workers))
291 pr_warn_once("io-wq is not configured for unbound workers");
293 raw_spin_lock(&wqe->lock);
294 if (acct->nr_workers >= acct->max_workers) {
295 raw_spin_unlock(&wqe->lock);
299 raw_spin_unlock(&wqe->lock);
300 atomic_inc(&acct->nr_running);
301 atomic_inc(&wqe->wq->worker_refs);
302 return create_io_worker(wqe->wq, wqe, acct->index);
305 static void io_wqe_inc_running(struct io_worker *worker)
307 struct io_wqe_acct *acct = io_wqe_get_acct(worker);
309 atomic_inc(&acct->nr_running);
312 static void create_worker_cb(struct callback_head *cb)
314 struct io_worker *worker;
317 struct io_wqe_acct *acct;
318 bool do_create = false;
320 worker = container_of(cb, struct io_worker, create_work);
323 acct = &wqe->acct[worker->create_index];
324 raw_spin_lock(&wqe->lock);
325 if (acct->nr_workers < acct->max_workers) {
329 raw_spin_unlock(&wqe->lock);
331 create_io_worker(wq, wqe, worker->create_index);
333 atomic_dec(&acct->nr_running);
334 io_worker_ref_put(wq);
336 clear_bit_unlock(0, &worker->create_state);
337 io_worker_release(worker);
340 static bool io_queue_worker_create(struct io_worker *worker,
341 struct io_wqe_acct *acct,
342 task_work_func_t func)
344 struct io_wqe *wqe = worker->wqe;
345 struct io_wq *wq = wqe->wq;
347 /* raced with exit, just ignore create call */
348 if (test_bit(IO_WQ_BIT_EXIT, &wq->state))
350 if (!io_worker_get(worker))
353 * create_state manages ownership of create_work/index. We should
354 * only need one entry per worker, as the worker going to sleep
355 * will trigger the condition, and waking will clear it once it
356 * runs the task_work.
358 if (test_bit(0, &worker->create_state) ||
359 test_and_set_bit_lock(0, &worker->create_state))
362 atomic_inc(&wq->worker_refs);
363 init_task_work(&worker->create_work, func);
364 worker->create_index = acct->index;
365 if (!task_work_add(wq->task, &worker->create_work, TWA_SIGNAL)) {
367 * EXIT may have been set after checking it above, check after
368 * adding the task_work and remove any creation item if it is
369 * now set. wq exit does that too, but we can have added this
370 * work item after we canceled in io_wq_exit_workers().
372 if (test_bit(IO_WQ_BIT_EXIT, &wq->state))
373 io_wq_cancel_tw_create(wq);
374 io_worker_ref_put(wq);
377 io_worker_ref_put(wq);
378 clear_bit_unlock(0, &worker->create_state);
380 io_worker_release(worker);
382 atomic_dec(&acct->nr_running);
383 io_worker_ref_put(wq);
387 static void io_wqe_dec_running(struct io_worker *worker)
388 __must_hold(wqe->lock)
390 struct io_wqe_acct *acct = io_wqe_get_acct(worker);
391 struct io_wqe *wqe = worker->wqe;
393 if (!(worker->flags & IO_WORKER_F_UP))
396 if (atomic_dec_and_test(&acct->nr_running) && io_acct_run_queue(acct)) {
397 atomic_inc(&acct->nr_running);
398 atomic_inc(&wqe->wq->worker_refs);
399 raw_spin_unlock(&wqe->lock);
400 io_queue_worker_create(worker, acct, create_worker_cb);
401 raw_spin_lock(&wqe->lock);
406 * Worker will start processing some work. Move it to the busy list, if
407 * it's currently on the freelist
409 static void __io_worker_busy(struct io_wqe *wqe, struct io_worker *worker)
410 __must_hold(wqe->lock)
412 if (worker->flags & IO_WORKER_F_FREE) {
413 worker->flags &= ~IO_WORKER_F_FREE;
414 hlist_nulls_del_init_rcu(&worker->nulls_node);
419 * No work, worker going to sleep. Move to freelist, and unuse mm if we
420 * have one attached. Dropping the mm may potentially sleep, so we drop
421 * the lock in that case and return success. Since the caller has to
422 * retry the loop in that case (we changed task state), we don't regrab
423 * the lock if we return success.
425 static void __io_worker_idle(struct io_wqe *wqe, struct io_worker *worker)
426 __must_hold(wqe->lock)
428 if (!(worker->flags & IO_WORKER_F_FREE)) {
429 worker->flags |= IO_WORKER_F_FREE;
430 hlist_nulls_add_head_rcu(&worker->nulls_node, &wqe->free_list);
434 static inline unsigned int io_get_work_hash(struct io_wq_work *work)
436 return work->flags >> IO_WQ_HASH_SHIFT;
439 static bool io_wait_on_hash(struct io_wqe *wqe, unsigned int hash)
441 struct io_wq *wq = wqe->wq;
444 spin_lock_irq(&wq->hash->wait.lock);
445 if (list_empty(&wqe->wait.entry)) {
446 __add_wait_queue(&wq->hash->wait, &wqe->wait);
447 if (!test_bit(hash, &wq->hash->map)) {
448 __set_current_state(TASK_RUNNING);
449 list_del_init(&wqe->wait.entry);
453 spin_unlock_irq(&wq->hash->wait.lock);
457 static struct io_wq_work *io_get_next_work(struct io_wqe_acct *acct,
458 struct io_worker *worker)
459 __must_hold(wqe->lock)
461 struct io_wq_work_node *node, *prev;
462 struct io_wq_work *work, *tail;
463 unsigned int stall_hash = -1U;
464 struct io_wqe *wqe = worker->wqe;
466 wq_list_for_each(node, prev, &acct->work_list) {
469 work = container_of(node, struct io_wq_work, list);
471 /* not hashed, can run anytime */
472 if (!io_wq_is_hashed(work)) {
473 wq_list_del(&acct->work_list, node, prev);
477 hash = io_get_work_hash(work);
478 /* all items with this hash lie in [work, tail] */
479 tail = wqe->hash_tail[hash];
481 /* hashed, can run if not already running */
482 if (!test_and_set_bit(hash, &wqe->wq->hash->map)) {
483 wqe->hash_tail[hash] = NULL;
484 wq_list_cut(&acct->work_list, &tail->list, prev);
487 if (stall_hash == -1U)
489 /* fast forward to a next hash, for-each will fix up @prev */
493 if (stall_hash != -1U) {
497 * Set this before dropping the lock to avoid racing with new
498 * work being added and clearing the stalled bit.
500 set_bit(IO_ACCT_STALLED_BIT, &acct->flags);
501 raw_spin_unlock(&wqe->lock);
502 unstalled = io_wait_on_hash(wqe, stall_hash);
503 raw_spin_lock(&wqe->lock);
505 clear_bit(IO_ACCT_STALLED_BIT, &acct->flags);
506 if (wq_has_sleeper(&wqe->wq->hash->wait))
507 wake_up(&wqe->wq->hash->wait);
514 static bool io_flush_signals(void)
516 if (unlikely(test_thread_flag(TIF_NOTIFY_SIGNAL))) {
517 __set_current_state(TASK_RUNNING);
518 tracehook_notify_signal();
524 static void io_assign_current_work(struct io_worker *worker,
525 struct io_wq_work *work)
532 raw_spin_lock(&worker->lock);
533 worker->cur_work = work;
534 worker->next_work = NULL;
535 raw_spin_unlock(&worker->lock);
538 static void io_wqe_enqueue(struct io_wqe *wqe, struct io_wq_work *work);
540 static void io_worker_handle_work(struct io_worker *worker)
541 __releases(wqe->lock)
543 struct io_wqe_acct *acct = io_wqe_get_acct(worker);
544 struct io_wqe *wqe = worker->wqe;
545 struct io_wq *wq = wqe->wq;
546 bool do_kill = test_bit(IO_WQ_BIT_EXIT, &wq->state);
549 struct io_wq_work *work;
552 * If we got some work, mark us as busy. If we didn't, but
553 * the list isn't empty, it means we stalled on hashed work.
554 * Mark us stalled so we don't keep looking for work when we
555 * can't make progress, any work completion or insertion will
556 * clear the stalled flag.
558 work = io_get_next_work(acct, worker);
560 __io_worker_busy(wqe, worker);
563 * Make sure cancelation can find this, even before
564 * it becomes the active work. That avoids a window
565 * where the work has been removed from our general
566 * work list, but isn't yet discoverable as the
567 * current work item for this worker.
569 raw_spin_lock(&worker->lock);
570 worker->next_work = work;
571 raw_spin_unlock(&worker->lock);
573 raw_spin_unlock(&wqe->lock);
576 io_assign_current_work(worker, work);
577 __set_current_state(TASK_RUNNING);
579 /* handle a whole dependent link */
581 struct io_wq_work *next_hashed, *linked;
582 unsigned int hash = io_get_work_hash(work);
584 next_hashed = wq_next_work(work);
586 if (unlikely(do_kill) && (work->flags & IO_WQ_WORK_UNBOUND))
587 work->flags |= IO_WQ_WORK_CANCEL;
589 io_assign_current_work(worker, NULL);
591 linked = wq->free_work(work);
593 if (!work && linked && !io_wq_is_hashed(linked)) {
597 io_assign_current_work(worker, work);
599 io_wqe_enqueue(wqe, linked);
601 if (hash != -1U && !next_hashed) {
602 /* serialize hash clear with wake_up() */
603 spin_lock_irq(&wq->hash->wait.lock);
604 clear_bit(hash, &wq->hash->map);
605 clear_bit(IO_ACCT_STALLED_BIT, &acct->flags);
606 spin_unlock_irq(&wq->hash->wait.lock);
607 if (wq_has_sleeper(&wq->hash->wait))
608 wake_up(&wq->hash->wait);
612 raw_spin_lock(&wqe->lock);
616 static int io_wqe_worker(void *data)
618 struct io_worker *worker = data;
619 struct io_wqe_acct *acct = io_wqe_get_acct(worker);
620 struct io_wqe *wqe = worker->wqe;
621 struct io_wq *wq = wqe->wq;
622 bool last_timeout = false;
623 char buf[TASK_COMM_LEN];
625 worker->flags |= (IO_WORKER_F_UP | IO_WORKER_F_RUNNING);
627 snprintf(buf, sizeof(buf), "iou-wrk-%d", wq->task->pid);
628 set_task_comm(current, buf);
630 audit_alloc_kernel(current);
632 while (!test_bit(IO_WQ_BIT_EXIT, &wq->state)) {
635 set_current_state(TASK_INTERRUPTIBLE);
637 raw_spin_lock(&wqe->lock);
638 if (io_acct_run_queue(acct)) {
639 io_worker_handle_work(worker);
642 /* timed out, exit unless we're the last worker */
643 if (last_timeout && acct->nr_workers > 1) {
645 raw_spin_unlock(&wqe->lock);
646 __set_current_state(TASK_RUNNING);
649 last_timeout = false;
650 __io_worker_idle(wqe, worker);
651 raw_spin_unlock(&wqe->lock);
652 if (io_flush_signals())
654 ret = schedule_timeout(WORKER_IDLE_TIMEOUT);
655 if (signal_pending(current)) {
658 if (!get_signal(&ksig))
665 if (test_bit(IO_WQ_BIT_EXIT, &wq->state)) {
666 raw_spin_lock(&wqe->lock);
667 io_worker_handle_work(worker);
671 io_worker_exit(worker);
676 * Called when a worker is scheduled in. Mark us as currently running.
678 void io_wq_worker_running(struct task_struct *tsk)
680 struct io_worker *worker = tsk->worker_private;
684 if (!(worker->flags & IO_WORKER_F_UP))
686 if (worker->flags & IO_WORKER_F_RUNNING)
688 worker->flags |= IO_WORKER_F_RUNNING;
689 io_wqe_inc_running(worker);
693 * Called when worker is going to sleep. If there are no workers currently
694 * running and we have work pending, wake up a free one or create a new one.
696 void io_wq_worker_sleeping(struct task_struct *tsk)
698 struct io_worker *worker = tsk->worker_private;
702 if (!(worker->flags & IO_WORKER_F_UP))
704 if (!(worker->flags & IO_WORKER_F_RUNNING))
707 worker->flags &= ~IO_WORKER_F_RUNNING;
709 raw_spin_lock(&worker->wqe->lock);
710 io_wqe_dec_running(worker);
711 raw_spin_unlock(&worker->wqe->lock);
714 static void io_init_new_worker(struct io_wqe *wqe, struct io_worker *worker,
715 struct task_struct *tsk)
717 tsk->worker_private = worker;
719 set_cpus_allowed_ptr(tsk, wqe->cpu_mask);
720 tsk->flags |= PF_NO_SETAFFINITY;
722 raw_spin_lock(&wqe->lock);
723 hlist_nulls_add_head_rcu(&worker->nulls_node, &wqe->free_list);
724 list_add_tail_rcu(&worker->all_list, &wqe->all_list);
725 worker->flags |= IO_WORKER_F_FREE;
726 raw_spin_unlock(&wqe->lock);
727 wake_up_new_task(tsk);
730 static bool io_wq_work_match_all(struct io_wq_work *work, void *data)
735 static inline bool io_should_retry_thread(long err)
738 * Prevent perpetual task_work retry, if the task (or its group) is
741 if (fatal_signal_pending(current))
747 case -ERESTARTNOINTR:
748 case -ERESTARTNOHAND:
755 static void create_worker_cont(struct callback_head *cb)
757 struct io_worker *worker;
758 struct task_struct *tsk;
761 worker = container_of(cb, struct io_worker, create_work);
762 clear_bit_unlock(0, &worker->create_state);
764 tsk = create_io_thread(io_wqe_worker, worker, wqe->node);
766 io_init_new_worker(wqe, worker, tsk);
767 io_worker_release(worker);
769 } else if (!io_should_retry_thread(PTR_ERR(tsk))) {
770 struct io_wqe_acct *acct = io_wqe_get_acct(worker);
772 atomic_dec(&acct->nr_running);
773 raw_spin_lock(&wqe->lock);
775 if (!acct->nr_workers) {
776 struct io_cb_cancel_data match = {
777 .fn = io_wq_work_match_all,
781 while (io_acct_cancel_pending_work(wqe, acct, &match))
782 raw_spin_lock(&wqe->lock);
784 raw_spin_unlock(&wqe->lock);
785 io_worker_ref_put(wqe->wq);
790 /* re-create attempts grab a new worker ref, drop the existing one */
791 io_worker_release(worker);
792 schedule_work(&worker->work);
795 static void io_workqueue_create(struct work_struct *work)
797 struct io_worker *worker = container_of(work, struct io_worker, work);
798 struct io_wqe_acct *acct = io_wqe_get_acct(worker);
800 if (!io_queue_worker_create(worker, acct, create_worker_cont))
804 static bool create_io_worker(struct io_wq *wq, struct io_wqe *wqe, int index)
806 struct io_wqe_acct *acct = &wqe->acct[index];
807 struct io_worker *worker;
808 struct task_struct *tsk;
810 __set_current_state(TASK_RUNNING);
812 worker = kzalloc_node(sizeof(*worker), GFP_KERNEL, wqe->node);
815 atomic_dec(&acct->nr_running);
816 raw_spin_lock(&wqe->lock);
818 raw_spin_unlock(&wqe->lock);
819 io_worker_ref_put(wq);
823 refcount_set(&worker->ref, 1);
825 raw_spin_lock_init(&worker->lock);
826 init_completion(&worker->ref_done);
828 if (index == IO_WQ_ACCT_BOUND)
829 worker->flags |= IO_WORKER_F_BOUND;
831 tsk = create_io_thread(io_wqe_worker, worker, wqe->node);
833 io_init_new_worker(wqe, worker, tsk);
834 } else if (!io_should_retry_thread(PTR_ERR(tsk))) {
838 INIT_WORK(&worker->work, io_workqueue_create);
839 schedule_work(&worker->work);
846 * Iterate the passed in list and call the specific function for each
847 * worker that isn't exiting
849 static bool io_wq_for_each_worker(struct io_wqe *wqe,
850 bool (*func)(struct io_worker *, void *),
853 struct io_worker *worker;
856 list_for_each_entry_rcu(worker, &wqe->all_list, all_list) {
857 if (io_worker_get(worker)) {
858 /* no task if node is/was offline */
860 ret = func(worker, data);
861 io_worker_release(worker);
870 static bool io_wq_worker_wake(struct io_worker *worker, void *data)
872 set_notify_signal(worker->task);
873 wake_up_process(worker->task);
877 static void io_run_cancel(struct io_wq_work *work, struct io_wqe *wqe)
879 struct io_wq *wq = wqe->wq;
882 work->flags |= IO_WQ_WORK_CANCEL;
884 work = wq->free_work(work);
888 static void io_wqe_insert_work(struct io_wqe *wqe, struct io_wq_work *work)
890 struct io_wqe_acct *acct = io_work_get_acct(wqe, work);
892 struct io_wq_work *tail;
894 if (!io_wq_is_hashed(work)) {
896 wq_list_add_tail(&work->list, &acct->work_list);
900 hash = io_get_work_hash(work);
901 tail = wqe->hash_tail[hash];
902 wqe->hash_tail[hash] = work;
906 wq_list_add_after(&work->list, &tail->list, &acct->work_list);
909 static bool io_wq_work_match_item(struct io_wq_work *work, void *data)
914 static void io_wqe_enqueue(struct io_wqe *wqe, struct io_wq_work *work)
916 struct io_wqe_acct *acct = io_work_get_acct(wqe, work);
917 unsigned work_flags = work->flags;
921 * If io-wq is exiting for this task, or if the request has explicitly
922 * been marked as one that should not get executed, cancel it here.
924 if (test_bit(IO_WQ_BIT_EXIT, &wqe->wq->state) ||
925 (work->flags & IO_WQ_WORK_CANCEL)) {
926 io_run_cancel(work, wqe);
930 raw_spin_lock(&wqe->lock);
931 io_wqe_insert_work(wqe, work);
932 clear_bit(IO_ACCT_STALLED_BIT, &acct->flags);
935 do_create = !io_wqe_activate_free_worker(wqe, acct);
938 raw_spin_unlock(&wqe->lock);
940 if (do_create && ((work_flags & IO_WQ_WORK_CONCURRENT) ||
941 !atomic_read(&acct->nr_running))) {
944 did_create = io_wqe_create_worker(wqe, acct);
945 if (likely(did_create))
948 raw_spin_lock(&wqe->lock);
949 /* fatal condition, failed to create the first worker */
950 if (!acct->nr_workers) {
951 struct io_cb_cancel_data match = {
952 .fn = io_wq_work_match_item,
957 if (io_acct_cancel_pending_work(wqe, acct, &match))
958 raw_spin_lock(&wqe->lock);
960 raw_spin_unlock(&wqe->lock);
964 void io_wq_enqueue(struct io_wq *wq, struct io_wq_work *work)
966 struct io_wqe *wqe = wq->wqes[numa_node_id()];
968 io_wqe_enqueue(wqe, work);
972 * Work items that hash to the same value will not be done in parallel.
973 * Used to limit concurrent writes, generally hashed by inode.
975 void io_wq_hash_work(struct io_wq_work *work, void *val)
979 bit = hash_ptr(val, IO_WQ_HASH_ORDER);
980 work->flags |= (IO_WQ_WORK_HASHED | (bit << IO_WQ_HASH_SHIFT));
983 static bool __io_wq_worker_cancel(struct io_worker *worker,
984 struct io_cb_cancel_data *match,
985 struct io_wq_work *work)
987 if (work && match->fn(work, match->data)) {
988 work->flags |= IO_WQ_WORK_CANCEL;
989 set_notify_signal(worker->task);
996 static bool io_wq_worker_cancel(struct io_worker *worker, void *data)
998 struct io_cb_cancel_data *match = data;
1001 * Hold the lock to avoid ->cur_work going out of scope, caller
1002 * may dereference the passed in work.
1004 raw_spin_lock(&worker->lock);
1005 if (__io_wq_worker_cancel(worker, match, worker->cur_work) ||
1006 __io_wq_worker_cancel(worker, match, worker->next_work))
1007 match->nr_running++;
1008 raw_spin_unlock(&worker->lock);
1010 return match->nr_running && !match->cancel_all;
1013 static inline void io_wqe_remove_pending(struct io_wqe *wqe,
1014 struct io_wq_work *work,
1015 struct io_wq_work_node *prev)
1017 struct io_wqe_acct *acct = io_work_get_acct(wqe, work);
1018 unsigned int hash = io_get_work_hash(work);
1019 struct io_wq_work *prev_work = NULL;
1021 if (io_wq_is_hashed(work) && work == wqe->hash_tail[hash]) {
1023 prev_work = container_of(prev, struct io_wq_work, list);
1024 if (prev_work && io_get_work_hash(prev_work) == hash)
1025 wqe->hash_tail[hash] = prev_work;
1027 wqe->hash_tail[hash] = NULL;
1029 wq_list_del(&acct->work_list, &work->list, prev);
1032 static bool io_acct_cancel_pending_work(struct io_wqe *wqe,
1033 struct io_wqe_acct *acct,
1034 struct io_cb_cancel_data *match)
1035 __releases(wqe->lock)
1037 struct io_wq_work_node *node, *prev;
1038 struct io_wq_work *work;
1040 wq_list_for_each(node, prev, &acct->work_list) {
1041 work = container_of(node, struct io_wq_work, list);
1042 if (!match->fn(work, match->data))
1044 io_wqe_remove_pending(wqe, work, prev);
1045 raw_spin_unlock(&wqe->lock);
1046 io_run_cancel(work, wqe);
1047 match->nr_pending++;
1048 /* not safe to continue after unlock */
1055 static void io_wqe_cancel_pending_work(struct io_wqe *wqe,
1056 struct io_cb_cancel_data *match)
1060 for (i = 0; i < IO_WQ_ACCT_NR; i++) {
1061 struct io_wqe_acct *acct = io_get_acct(wqe, i == 0);
1063 if (io_acct_cancel_pending_work(wqe, acct, match)) {
1064 raw_spin_lock(&wqe->lock);
1065 if (match->cancel_all)
1072 static void io_wqe_cancel_running_work(struct io_wqe *wqe,
1073 struct io_cb_cancel_data *match)
1076 io_wq_for_each_worker(wqe, io_wq_worker_cancel, match);
1080 enum io_wq_cancel io_wq_cancel_cb(struct io_wq *wq, work_cancel_fn *cancel,
1081 void *data, bool cancel_all)
1083 struct io_cb_cancel_data match = {
1086 .cancel_all = cancel_all,
1091 * First check pending list, if we're lucky we can just remove it
1092 * from there. CANCEL_OK means that the work is returned as-new,
1093 * no completion will be posted for it.
1095 * Then check if a free (going busy) or busy worker has the work
1096 * currently running. If we find it there, we'll return CANCEL_RUNNING
1097 * as an indication that we attempt to signal cancellation. The
1098 * completion will run normally in this case.
1100 * Do both of these while holding the wqe->lock, to ensure that
1101 * we'll find a work item regardless of state.
1103 for_each_node(node) {
1104 struct io_wqe *wqe = wq->wqes[node];
1106 raw_spin_lock(&wqe->lock);
1107 io_wqe_cancel_pending_work(wqe, &match);
1108 if (match.nr_pending && !match.cancel_all) {
1109 raw_spin_unlock(&wqe->lock);
1110 return IO_WQ_CANCEL_OK;
1113 io_wqe_cancel_running_work(wqe, &match);
1114 raw_spin_unlock(&wqe->lock);
1115 if (match.nr_running && !match.cancel_all)
1116 return IO_WQ_CANCEL_RUNNING;
1119 if (match.nr_running)
1120 return IO_WQ_CANCEL_RUNNING;
1121 if (match.nr_pending)
1122 return IO_WQ_CANCEL_OK;
1123 return IO_WQ_CANCEL_NOTFOUND;
1126 static int io_wqe_hash_wake(struct wait_queue_entry *wait, unsigned mode,
1127 int sync, void *key)
1129 struct io_wqe *wqe = container_of(wait, struct io_wqe, wait);
1132 list_del_init(&wait->entry);
1135 for (i = 0; i < IO_WQ_ACCT_NR; i++) {
1136 struct io_wqe_acct *acct = &wqe->acct[i];
1138 if (test_and_clear_bit(IO_ACCT_STALLED_BIT, &acct->flags))
1139 io_wqe_activate_free_worker(wqe, acct);
1145 struct io_wq *io_wq_create(unsigned bounded, struct io_wq_data *data)
1150 if (WARN_ON_ONCE(!data->free_work || !data->do_work))
1151 return ERR_PTR(-EINVAL);
1152 if (WARN_ON_ONCE(!bounded))
1153 return ERR_PTR(-EINVAL);
1155 wq = kzalloc(struct_size(wq, wqes, nr_node_ids), GFP_KERNEL);
1157 return ERR_PTR(-ENOMEM);
1158 ret = cpuhp_state_add_instance_nocalls(io_wq_online, &wq->cpuhp_node);
1162 refcount_inc(&data->hash->refs);
1163 wq->hash = data->hash;
1164 wq->free_work = data->free_work;
1165 wq->do_work = data->do_work;
1168 for_each_node(node) {
1170 int alloc_node = node;
1172 if (!node_online(alloc_node))
1173 alloc_node = NUMA_NO_NODE;
1174 wqe = kzalloc_node(sizeof(struct io_wqe), GFP_KERNEL, alloc_node);
1177 if (!alloc_cpumask_var(&wqe->cpu_mask, GFP_KERNEL))
1179 cpumask_copy(wqe->cpu_mask, cpumask_of_node(node));
1180 wq->wqes[node] = wqe;
1181 wqe->node = alloc_node;
1182 wqe->acct[IO_WQ_ACCT_BOUND].max_workers = bounded;
1183 wqe->acct[IO_WQ_ACCT_UNBOUND].max_workers =
1184 task_rlimit(current, RLIMIT_NPROC);
1185 INIT_LIST_HEAD(&wqe->wait.entry);
1186 wqe->wait.func = io_wqe_hash_wake;
1187 for (i = 0; i < IO_WQ_ACCT_NR; i++) {
1188 struct io_wqe_acct *acct = &wqe->acct[i];
1191 atomic_set(&acct->nr_running, 0);
1192 INIT_WQ_LIST(&acct->work_list);
1195 raw_spin_lock_init(&wqe->lock);
1196 INIT_HLIST_NULLS_HEAD(&wqe->free_list, 0);
1197 INIT_LIST_HEAD(&wqe->all_list);
1200 wq->task = get_task_struct(data->task);
1201 atomic_set(&wq->worker_refs, 1);
1202 init_completion(&wq->worker_done);
1205 io_wq_put_hash(data->hash);
1206 cpuhp_state_remove_instance_nocalls(io_wq_online, &wq->cpuhp_node);
1207 for_each_node(node) {
1208 if (!wq->wqes[node])
1210 free_cpumask_var(wq->wqes[node]->cpu_mask);
1211 kfree(wq->wqes[node]);
1215 return ERR_PTR(ret);
1218 static bool io_task_work_match(struct callback_head *cb, void *data)
1220 struct io_worker *worker;
1222 if (cb->func != create_worker_cb && cb->func != create_worker_cont)
1224 worker = container_of(cb, struct io_worker, create_work);
1225 return worker->wqe->wq == data;
1228 void io_wq_exit_start(struct io_wq *wq)
1230 set_bit(IO_WQ_BIT_EXIT, &wq->state);
1233 static void io_wq_cancel_tw_create(struct io_wq *wq)
1235 struct callback_head *cb;
1237 while ((cb = task_work_cancel_match(wq->task, io_task_work_match, wq)) != NULL) {
1238 struct io_worker *worker;
1240 worker = container_of(cb, struct io_worker, create_work);
1241 io_worker_cancel_cb(worker);
1245 static void io_wq_exit_workers(struct io_wq *wq)
1252 io_wq_cancel_tw_create(wq);
1255 for_each_node(node) {
1256 struct io_wqe *wqe = wq->wqes[node];
1258 io_wq_for_each_worker(wqe, io_wq_worker_wake, NULL);
1261 io_worker_ref_put(wq);
1262 wait_for_completion(&wq->worker_done);
1264 for_each_node(node) {
1265 spin_lock_irq(&wq->hash->wait.lock);
1266 list_del_init(&wq->wqes[node]->wait.entry);
1267 spin_unlock_irq(&wq->hash->wait.lock);
1269 put_task_struct(wq->task);
1273 static void io_wq_destroy(struct io_wq *wq)
1277 cpuhp_state_remove_instance_nocalls(io_wq_online, &wq->cpuhp_node);
1279 for_each_node(node) {
1280 struct io_wqe *wqe = wq->wqes[node];
1281 struct io_cb_cancel_data match = {
1282 .fn = io_wq_work_match_all,
1285 raw_spin_lock(&wqe->lock);
1286 io_wqe_cancel_pending_work(wqe, &match);
1287 raw_spin_unlock(&wqe->lock);
1288 free_cpumask_var(wqe->cpu_mask);
1291 io_wq_put_hash(wq->hash);
1295 void io_wq_put_and_exit(struct io_wq *wq)
1297 WARN_ON_ONCE(!test_bit(IO_WQ_BIT_EXIT, &wq->state));
1299 io_wq_exit_workers(wq);
1303 struct online_data {
1308 static bool io_wq_worker_affinity(struct io_worker *worker, void *data)
1310 struct online_data *od = data;
1313 cpumask_set_cpu(od->cpu, worker->wqe->cpu_mask);
1315 cpumask_clear_cpu(od->cpu, worker->wqe->cpu_mask);
1319 static int __io_wq_cpu_online(struct io_wq *wq, unsigned int cpu, bool online)
1321 struct online_data od = {
1329 io_wq_for_each_worker(wq->wqes[i], io_wq_worker_affinity, &od);
1334 static int io_wq_cpu_online(unsigned int cpu, struct hlist_node *node)
1336 struct io_wq *wq = hlist_entry_safe(node, struct io_wq, cpuhp_node);
1338 return __io_wq_cpu_online(wq, cpu, true);
1341 static int io_wq_cpu_offline(unsigned int cpu, struct hlist_node *node)
1343 struct io_wq *wq = hlist_entry_safe(node, struct io_wq, cpuhp_node);
1345 return __io_wq_cpu_online(wq, cpu, false);
1348 int io_wq_cpu_affinity(struct io_wq *wq, cpumask_var_t mask)
1354 struct io_wqe *wqe = wq->wqes[i];
1357 cpumask_copy(wqe->cpu_mask, mask);
1359 cpumask_copy(wqe->cpu_mask, cpumask_of_node(i));
1366 * Set max number of unbounded workers, returns old value. If new_count is 0,
1367 * then just return the old value.
1369 int io_wq_max_workers(struct io_wq *wq, int *new_count)
1371 int prev[IO_WQ_ACCT_NR];
1372 bool first_node = true;
1375 BUILD_BUG_ON((int) IO_WQ_ACCT_BOUND != (int) IO_WQ_BOUND);
1376 BUILD_BUG_ON((int) IO_WQ_ACCT_UNBOUND != (int) IO_WQ_UNBOUND);
1377 BUILD_BUG_ON((int) IO_WQ_ACCT_NR != 2);
1379 for (i = 0; i < 2; i++) {
1380 if (new_count[i] > task_rlimit(current, RLIMIT_NPROC))
1381 new_count[i] = task_rlimit(current, RLIMIT_NPROC);
1384 for (i = 0; i < IO_WQ_ACCT_NR; i++)
1388 for_each_node(node) {
1389 struct io_wqe *wqe = wq->wqes[node];
1390 struct io_wqe_acct *acct;
1392 raw_spin_lock(&wqe->lock);
1393 for (i = 0; i < IO_WQ_ACCT_NR; i++) {
1394 acct = &wqe->acct[i];
1396 prev[i] = max_t(int, acct->max_workers, prev[i]);
1398 acct->max_workers = new_count[i];
1400 raw_spin_unlock(&wqe->lock);
1405 for (i = 0; i < IO_WQ_ACCT_NR; i++)
1406 new_count[i] = prev[i];
1411 static __init int io_wq_init(void)
1415 ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN, "io-wq/online",
1416 io_wq_cpu_online, io_wq_cpu_offline);
1422 subsys_initcall(io_wq_init);