io-wq.c 28.3 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17
// SPDX-License-Identifier: GPL-2.0
/*
 * Basic worker thread pool for io_uring
 *
 * Copyright (C) 2019 Jens Axboe
 *
 */
#include <linux/kernel.h>
#include <linux/init.h>
#include <linux/errno.h>
#include <linux/sched/signal.h>
#include <linux/mm.h>
#include <linux/sched/mm.h>
#include <linux/percpu.h>
#include <linux/slab.h>
#include <linux/kthread.h>
#include <linux/rculist_nulls.h>
18
#include <linux/fs_struct.h>
19
#include <linux/task_work.h>
20
#include <linux/blk-cgroup.h>
21 22 23 24 25 26 27 28 29

#include "io-wq.h"

#define WORKER_IDLE_TIMEOUT	(5 * HZ)

enum {
	IO_WORKER_F_UP		= 1,	/* up and active */
	IO_WORKER_F_RUNNING	= 2,	/* account as running */
	IO_WORKER_F_FREE	= 4,	/* worker on free list */
30 31
	IO_WORKER_F_FIXED	= 8,	/* static idle worker */
	IO_WORKER_F_BOUND	= 16,	/* is doing bounded work */
32 33 34 35 36
};

enum {
	IO_WQ_BIT_EXIT		= 0,	/* wq exiting */
	IO_WQ_BIT_CANCEL	= 1,	/* cancel work on list */
37
	IO_WQ_BIT_ERROR		= 2,	/* error on setup */
38 39 40 41 42 43 44 45 46 47 48 49 50
};

enum {
	IO_WQE_FLAG_STALLED	= 1,	/* stalled on hash */
};

/*
 * One for each thread in a wqe pool
 */
struct io_worker {
	refcount_t ref;
	unsigned flags;
	struct hlist_nulls_node nulls_node;
51
	struct list_head all_list;
52 53
	struct task_struct *task;
	struct io_wqe *wqe;
54

55
	struct io_wq_work *cur_work;
56
	spinlock_t lock;
57 58 59

	struct rcu_head rcu;
	struct mm_struct *mm;
60 61 62
#ifdef CONFIG_BLK_CGROUP
	struct cgroup_subsys_state *blkcg_css;
#endif
63 64
	const struct cred *cur_creds;
	const struct cred *saved_creds;
65
	struct files_struct *restore_files;
66
	struct nsproxy *restore_nsproxy;
67
	struct fs_struct *restore_fs;
68 69 70 71 72 73 74 75
};

#if BITS_PER_LONG == 64
#define IO_WQ_HASH_ORDER	6
#else
#define IO_WQ_HASH_ORDER	5
#endif

76 77
#define IO_WQ_NR_HASH_BUCKETS	(1u << IO_WQ_HASH_ORDER)

78 79 80 81 82 83 84 85 86 87 88
struct io_wqe_acct {
	unsigned nr_workers;
	unsigned max_workers;
	atomic_t nr_running;
};

enum {
	IO_WQ_ACCT_BOUND,
	IO_WQ_ACCT_UNBOUND,
};

89 90 91 92 93
/*
 * Per-node worker thread pool
 */
struct io_wqe {
	struct {
94
		raw_spinlock_t lock;
J
Jens Axboe 已提交
95
		struct io_wq_work_list work_list;
96 97 98 99 100
		unsigned long hash_map;
		unsigned flags;
	} ____cacheline_aligned_in_smp;

	int node;
101
	struct io_wqe_acct acct[2];
102

103
	struct hlist_nulls_head free_list;
104
	struct list_head all_list;
105 106

	struct io_wq *wq;
107
	struct io_wq_work *hash_tail[IO_WQ_NR_HASH_BUCKETS];
108 109 110 111 112 113 114 115 116
};

/*
 * Per io_wq state
  */
struct io_wq {
	struct io_wqe **wqes;
	unsigned long state;

117
	free_work_fn *free_work;
118
	io_wq_work_fn *do_work;
119

120
	struct task_struct *manager;
121
	struct user_struct *user;
122 123
	refcount_t refs;
	struct completion done;
J
Jens Axboe 已提交
124 125

	refcount_t use_refs;
126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145
};

static bool io_worker_get(struct io_worker *worker)
{
	return refcount_inc_not_zero(&worker->ref);
}

static void io_worker_release(struct io_worker *worker)
{
	if (refcount_dec_and_test(&worker->ref))
		wake_up_process(worker->task);
}

/*
 * Note: drops the wqe->lock if returning true! The caller must re-acquire
 * the lock in that case. Some callers need to restart handling if this
 * happens, so we can't just re-acquire the lock on behalf of the caller.
 */
static bool __io_worker_unuse(struct io_wqe *wqe, struct io_worker *worker)
{
146 147
	bool dropped_lock = false;

148 149 150
	if (worker->saved_creds) {
		revert_creds(worker->saved_creds);
		worker->cur_creds = worker->saved_creds = NULL;
151 152
	}

153 154
	if (current->files != worker->restore_files) {
		__acquire(&wqe->lock);
155
		raw_spin_unlock_irq(&wqe->lock);
156 157 158 159
		dropped_lock = true;

		task_lock(current);
		current->files = worker->restore_files;
160
		current->nsproxy = worker->restore_nsproxy;
161 162 163
		task_unlock(current);
	}

164 165 166
	if (current->fs != worker->restore_fs)
		current->fs = worker->restore_fs;

167 168 169 170 171
	/*
	 * If we have an active mm, we need to drop the wq lock before unusing
	 * it. If we do, return true and let the caller retry the idle loop.
	 */
	if (worker->mm) {
172 173
		if (!dropped_lock) {
			__acquire(&wqe->lock);
174
			raw_spin_unlock_irq(&wqe->lock);
175 176
			dropped_lock = true;
		}
177
		__set_current_state(TASK_RUNNING);
178
		kthread_unuse_mm(worker->mm);
179 180 181 182
		mmput(worker->mm);
		worker->mm = NULL;
	}

183 184 185 186 187 188 189
#ifdef CONFIG_BLK_CGROUP
	if (worker->blkcg_css) {
		kthread_associate_blkcg(NULL);
		worker->blkcg_css = NULL;
	}
#endif

190
	return dropped_lock;
191 192
}

193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210
static inline struct io_wqe_acct *io_work_get_acct(struct io_wqe *wqe,
						   struct io_wq_work *work)
{
	if (work->flags & IO_WQ_WORK_UNBOUND)
		return &wqe->acct[IO_WQ_ACCT_UNBOUND];

	return &wqe->acct[IO_WQ_ACCT_BOUND];
}

static inline struct io_wqe_acct *io_wqe_get_acct(struct io_wqe *wqe,
						  struct io_worker *worker)
{
	if (worker->flags & IO_WORKER_F_BOUND)
		return &wqe->acct[IO_WQ_ACCT_BOUND];

	return &wqe->acct[IO_WQ_ACCT_UNBOUND];
}

211 212 213
static void io_worker_exit(struct io_worker *worker)
{
	struct io_wqe *wqe = worker->wqe;
214
	struct io_wqe_acct *acct = io_wqe_get_acct(wqe, worker);
215 216 217 218 219 220 221 222 223 224 225 226 227

	/*
	 * If we're not at zero, someone else is holding a brief reference
	 * to the worker. Wait for that to go away.
	 */
	set_current_state(TASK_INTERRUPTIBLE);
	if (!refcount_dec_and_test(&worker->ref))
		schedule();
	__set_current_state(TASK_RUNNING);

	preempt_disable();
	current->flags &= ~PF_IO_WORKER;
	if (worker->flags & IO_WORKER_F_RUNNING)
228 229 230
		atomic_dec(&acct->nr_running);
	if (!(worker->flags & IO_WORKER_F_BOUND))
		atomic_dec(&wqe->wq->user->processes);
231 232 233
	worker->flags = 0;
	preempt_enable();

234
	raw_spin_lock_irq(&wqe->lock);
235
	hlist_nulls_del_rcu(&worker->nulls_node);
236
	list_del_rcu(&worker->all_list);
237 238
	if (__io_worker_unuse(wqe, worker)) {
		__release(&wqe->lock);
239
		raw_spin_lock_irq(&wqe->lock);
240
	}
241
	acct->nr_workers--;
242
	raw_spin_unlock_irq(&wqe->lock);
243

244
	kfree_rcu(worker, rcu);
245 246
	if (refcount_dec_and_test(&wqe->wq->refs))
		complete(&wqe->wq->done);
247 248
}

249 250 251
static inline bool io_wqe_run_queue(struct io_wqe *wqe)
	__must_hold(wqe->lock)
{
J
Jens Axboe 已提交
252 253
	if (!wq_list_empty(&wqe->work_list) &&
	    !(wqe->flags & IO_WQE_FLAG_STALLED))
254 255 256 257 258 259 260 261 262 263 264 265 266 267
		return true;
	return false;
}

/*
 * Check head of free list for an available worker. If one isn't available,
 * caller must wake up the wq manager to create one.
 */
static bool io_wqe_activate_free_worker(struct io_wqe *wqe)
	__must_hold(RCU)
{
	struct hlist_nulls_node *n;
	struct io_worker *worker;

268
	n = rcu_dereference(hlist_nulls_first_rcu(&wqe->free_list));
269 270 271 272 273
	if (is_a_nulls(n))
		return false;

	worker = hlist_nulls_entry(n, struct io_worker, nulls_node);
	if (io_worker_get(worker)) {
J
Jens Axboe 已提交
274
		wake_up_process(worker->task);
275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319
		io_worker_release(worker);
		return true;
	}

	return false;
}

/*
 * We need a worker. If we find a free one, we're good. If not, and we're
 * below the max number of workers, wake up the manager to create one.
 */
static void io_wqe_wake_worker(struct io_wqe *wqe, struct io_wqe_acct *acct)
{
	bool ret;

	/*
	 * Most likely an attempt to queue unbounded work on an io_wq that
	 * wasn't setup with any unbounded workers.
	 */
	WARN_ON_ONCE(!acct->max_workers);

	rcu_read_lock();
	ret = io_wqe_activate_free_worker(wqe);
	rcu_read_unlock();

	if (!ret && acct->nr_workers < acct->max_workers)
		wake_up_process(wqe->wq->manager);
}

static void io_wqe_inc_running(struct io_wqe *wqe, struct io_worker *worker)
{
	struct io_wqe_acct *acct = io_wqe_get_acct(wqe, worker);

	atomic_inc(&acct->nr_running);
}

static void io_wqe_dec_running(struct io_wqe *wqe, struct io_worker *worker)
	__must_hold(wqe->lock)
{
	struct io_wqe_acct *acct = io_wqe_get_acct(wqe, worker);

	if (atomic_dec_and_test(&acct->nr_running) && io_wqe_run_queue(wqe))
		io_wqe_wake_worker(wqe, acct);
}

320 321 322 323 324 325 326
static void io_worker_start(struct io_wqe *wqe, struct io_worker *worker)
{
	allow_kernel_signal(SIGINT);

	current->flags |= PF_IO_WORKER;

	worker->flags |= (IO_WORKER_F_UP | IO_WORKER_F_RUNNING);
327
	worker->restore_files = current->files;
328
	worker->restore_nsproxy = current->nsproxy;
329
	worker->restore_fs = current->fs;
330
	io_wqe_inc_running(wqe, worker);
331 332 333 334 335 336 337 338 339 340
}

/*
 * Worker will start processing some work. Move it to the busy list, if
 * it's currently on the freelist
 */
static void __io_worker_busy(struct io_wqe *wqe, struct io_worker *worker,
			     struct io_wq_work *work)
	__must_hold(wqe->lock)
{
341 342
	bool worker_bound, work_bound;

343 344 345 346
	if (worker->flags & IO_WORKER_F_FREE) {
		worker->flags &= ~IO_WORKER_F_FREE;
		hlist_nulls_del_init_rcu(&worker->nulls_node);
	}
347 348 349 350 351

	/*
	 * If worker is moving from bound to unbound (or vice versa), then
	 * ensure we update the running accounting.
	 */
352 353 354
	worker_bound = (worker->flags & IO_WORKER_F_BOUND) != 0;
	work_bound = (work->flags & IO_WQ_WORK_UNBOUND) == 0;
	if (worker_bound != work_bound) {
355 356 357 358 359 360 361 362 363 364 365 366 367 368
		io_wqe_dec_running(wqe, worker);
		if (work_bound) {
			worker->flags |= IO_WORKER_F_BOUND;
			wqe->acct[IO_WQ_ACCT_UNBOUND].nr_workers--;
			wqe->acct[IO_WQ_ACCT_BOUND].nr_workers++;
			atomic_dec(&wqe->wq->user->processes);
		} else {
			worker->flags &= ~IO_WORKER_F_BOUND;
			wqe->acct[IO_WQ_ACCT_UNBOUND].nr_workers++;
			wqe->acct[IO_WQ_ACCT_BOUND].nr_workers--;
			atomic_inc(&wqe->wq->user->processes);
		}
		io_wqe_inc_running(wqe, worker);
	 }
369 370 371 372 373 374 375 376 377 378 379 380 381 382
}

/*
 * No work, worker going to sleep. Move to freelist, and unuse mm if we
 * have one attached. Dropping the mm may potentially sleep, so we drop
 * the lock in that case and return success. Since the caller has to
 * retry the loop in that case (we changed task state), we don't regrab
 * the lock if we return success.
 */
static bool __io_worker_idle(struct io_wqe *wqe, struct io_worker *worker)
	__must_hold(wqe->lock)
{
	if (!(worker->flags & IO_WORKER_F_FREE)) {
		worker->flags |= IO_WORKER_F_FREE;
383
		hlist_nulls_add_head_rcu(&worker->nulls_node, &wqe->free_list);
384 385 386 387 388
	}

	return __io_worker_unuse(wqe, worker);
}

P
Pavel Begunkov 已提交
389 390 391 392 393 394
static inline unsigned int io_get_work_hash(struct io_wq_work *work)
{
	return work->flags >> IO_WQ_HASH_SHIFT;
}

static struct io_wq_work *io_get_next_work(struct io_wqe *wqe)
395 396
	__must_hold(wqe->lock)
{
J
Jens Axboe 已提交
397
	struct io_wq_work_node *node, *prev;
398
	struct io_wq_work *work, *tail;
P
Pavel Begunkov 已提交
399
	unsigned int hash;
400

J
Jens Axboe 已提交
401 402 403
	wq_list_for_each(node, prev, &wqe->work_list) {
		work = container_of(node, struct io_wq_work, list);

404
		/* not hashed, can run anytime */
405
		if (!io_wq_is_hashed(work)) {
406
			wq_list_del(&wqe->work_list, node, prev);
407 408 409 410
			return work;
		}

		/* hashed, can run if not already running */
P
Pavel Begunkov 已提交
411 412 413
		hash = io_get_work_hash(work);
		if (!(wqe->hash_map & BIT(hash))) {
			wqe->hash_map |= BIT(hash);
414 415 416 417
			/* all items with this hash lie in [work, tail] */
			tail = wqe->hash_tail[hash];
			wqe->hash_tail[hash] = NULL;
			wq_list_cut(&wqe->work_list, &tail->list, prev);
418 419 420 421 422 423 424
			return work;
		}
	}

	return NULL;
}

425 426 427
static void io_wq_switch_mm(struct io_worker *worker, struct io_wq_work *work)
{
	if (worker->mm) {
428
		kthread_unuse_mm(worker->mm);
429 430 431
		mmput(worker->mm);
		worker->mm = NULL;
	}
432
	if (!work->mm)
433
		return;
434

435
	if (mmget_not_zero(work->mm)) {
436
		kthread_use_mm(work->mm);
437 438 439 440 441 442 443 444 445 446
		worker->mm = work->mm;
		/* hang on to this mm */
		work->mm = NULL;
		return;
	}

	/* failed grabbing mm, ensure work gets cancelled */
	work->flags |= IO_WQ_WORK_CANCEL;
}

447 448 449 450 451 452 453 454 455 456 457
static inline void io_wq_switch_blkcg(struct io_worker *worker,
				      struct io_wq_work *work)
{
#ifdef CONFIG_BLK_CGROUP
	if (work->blkcg_css != worker->blkcg_css) {
		kthread_associate_blkcg(work->blkcg_css);
		worker->blkcg_css = work->blkcg_css;
	}
#endif
}

458 459 460 461 462 463 464 465 466 467 468 469
static void io_wq_switch_creds(struct io_worker *worker,
			       struct io_wq_work *work)
{
	const struct cred *old_creds = override_creds(work->creds);

	worker->cur_creds = work->creds;
	if (worker->saved_creds)
		put_cred(old_creds); /* creds set by previous switch */
	else
		worker->saved_creds = old_creds;
}

470 471 472 473 474 475
static void io_impersonate_work(struct io_worker *worker,
				struct io_wq_work *work)
{
	if (work->files && current->files != work->files) {
		task_lock(current);
		current->files = work->files;
476
		current->nsproxy = work->nsproxy;
477 478 479 480 481 482 483 484
		task_unlock(current);
	}
	if (work->fs && current->fs != work->fs)
		current->fs = work->fs;
	if (work->mm != worker->mm)
		io_wq_switch_mm(worker, work);
	if (worker->cur_creds != work->creds)
		io_wq_switch_creds(worker, work);
485
	current->signal->rlim[RLIMIT_FSIZE].rlim_cur = work->fsize;
486
	io_wq_switch_blkcg(worker, work);
487 488 489 490 491
}

static void io_assign_current_work(struct io_worker *worker,
				   struct io_wq_work *work)
{
492 493 494 495 496 497
	if (work) {
		/* flush pending signals before assigning new work */
		if (signal_pending(current))
			flush_signals(current);
		cond_resched();
	}
498 499 500 501 502 503

	spin_lock_irq(&worker->lock);
	worker->cur_work = work;
	spin_unlock_irq(&worker->lock);
}

P
Pavel Begunkov 已提交
504 505
static void io_wqe_enqueue(struct io_wqe *wqe, struct io_wq_work *work);

506 507 508 509 510 511 512
static void io_worker_handle_work(struct io_worker *worker)
	__releases(wqe->lock)
{
	struct io_wqe *wqe = worker->wqe;
	struct io_wq *wq = wqe->wq;

	do {
513
		struct io_wq_work *work;
514
get_next:
515 516 517 518 519 520 521
		/*
		 * If we got some work, mark us as busy. If we didn't, but
		 * the list isn't empty, it means we stalled on hashed work.
		 * Mark us stalled so we don't keep looking for work when we
		 * can't make progress, any work completion or insertion will
		 * clear the stalled flag.
		 */
P
Pavel Begunkov 已提交
522
		work = io_get_next_work(wqe);
523 524
		if (work)
			__io_worker_busy(wqe, worker, work);
J
Jens Axboe 已提交
525
		else if (!wq_list_empty(&wqe->work_list))
526 527
			wqe->flags |= IO_WQE_FLAG_STALLED;

528
		raw_spin_unlock_irq(&wqe->lock);
529 530
		if (!work)
			break;
531
		io_assign_current_work(worker, work);
532

533 534
		/* handle a whole dependent link */
		do {
535
			struct io_wq_work *old_work, *next_hashed, *linked;
P
Pavel Begunkov 已提交
536
			unsigned int hash = io_get_work_hash(work);
537

538
			next_hashed = wq_next_work(work);
539
			io_impersonate_work(worker, work);
540 541 542 543 544 545 546
			/*
			 * OK to set IO_WQ_WORK_CANCEL even for uncancellable
			 * work, the worker function will do the right thing.
			 */
			if (test_bit(IO_WQ_BIT_CANCEL, &wq->state))
				work->flags |= IO_WQ_WORK_CANCEL;

547 548
			old_work = work;
			linked = wq->do_work(work);
549 550 551 552 553 554 555

			work = next_hashed;
			if (!work && linked && !io_wq_is_hashed(linked)) {
				work = linked;
				linked = NULL;
			}
			io_assign_current_work(worker, work);
556
			wq->free_work(old_work);
557

558 559 560 561
			if (linked)
				io_wqe_enqueue(wqe, linked);

			if (hash != -1U && !next_hashed) {
562
				raw_spin_lock_irq(&wqe->lock);
563 564
				wqe->hash_map &= ~BIT_ULL(hash);
				wqe->flags &= ~IO_WQE_FLAG_STALLED;
565 566 567
				/* skip unnecessary unlock-lock wqe->lock */
				if (!work)
					goto get_next;
568
				raw_spin_unlock_irq(&wqe->lock);
569
			}
570
		} while (work);
571

572
		raw_spin_lock_irq(&wqe->lock);
573 574 575 576 577 578 579 580 581 582 583 584
	} while (1);
}

static int io_wqe_worker(void *data)
{
	struct io_worker *worker = data;
	struct io_wqe *wqe = worker->wqe;
	struct io_wq *wq = wqe->wq;

	io_worker_start(wqe, worker);

	while (!test_bit(IO_WQ_BIT_EXIT, &wq->state)) {
J
Jens Axboe 已提交
585
		set_current_state(TASK_INTERRUPTIBLE);
586
loop:
587
		raw_spin_lock_irq(&wqe->lock);
588 589 590
		if (io_wqe_run_queue(wqe)) {
			__set_current_state(TASK_RUNNING);
			io_worker_handle_work(worker);
591
			goto loop;
592 593 594 595
		}
		/* drops the lock on success, retry */
		if (__io_worker_idle(wqe, worker)) {
			__release(&wqe->lock);
596
			goto loop;
597
		}
598
		raw_spin_unlock_irq(&wqe->lock);
599 600 601 602 603 604 605 606 607 608 609
		if (signal_pending(current))
			flush_signals(current);
		if (schedule_timeout(WORKER_IDLE_TIMEOUT))
			continue;
		/* timed out, exit unless we're the fixed worker */
		if (test_bit(IO_WQ_BIT_EXIT, &wq->state) ||
		    !(worker->flags & IO_WORKER_F_FIXED))
			break;
	}

	if (test_bit(IO_WQ_BIT_EXIT, &wq->state)) {
610
		raw_spin_lock_irq(&wqe->lock);
J
Jens Axboe 已提交
611
		if (!wq_list_empty(&wqe->work_list))
612 613
			io_worker_handle_work(worker);
		else
614
			raw_spin_unlock_irq(&wqe->lock);
615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633
	}

	io_worker_exit(worker);
	return 0;
}

/*
 * Called when a worker is scheduled in. Mark us as currently running.
 */
void io_wq_worker_running(struct task_struct *tsk)
{
	struct io_worker *worker = kthread_data(tsk);
	struct io_wqe *wqe = worker->wqe;

	if (!(worker->flags & IO_WORKER_F_UP))
		return;
	if (worker->flags & IO_WORKER_F_RUNNING)
		return;
	worker->flags |= IO_WORKER_F_RUNNING;
634
	io_wqe_inc_running(wqe, worker);
635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653
}

/*
 * Called when worker is going to sleep. If there are no workers currently
 * running and we have work pending, wake up a free one or have the manager
 * set one up.
 */
void io_wq_worker_sleeping(struct task_struct *tsk)
{
	struct io_worker *worker = kthread_data(tsk);
	struct io_wqe *wqe = worker->wqe;

	if (!(worker->flags & IO_WORKER_F_UP))
		return;
	if (!(worker->flags & IO_WORKER_F_RUNNING))
		return;

	worker->flags &= ~IO_WORKER_F_RUNNING;

654
	raw_spin_lock_irq(&wqe->lock);
655
	io_wqe_dec_running(wqe, worker);
656
	raw_spin_unlock_irq(&wqe->lock);
657 658
}

659
static bool create_io_worker(struct io_wq *wq, struct io_wqe *wqe, int index)
660
{
661
	struct io_wqe_acct *acct = &wqe->acct[index];
662 663
	struct io_worker *worker;

664
	worker = kzalloc_node(sizeof(*worker), GFP_KERNEL, wqe->node);
665
	if (!worker)
666
		return false;
667 668 669 670

	refcount_set(&worker->ref, 1);
	worker->nulls_node.pprev = NULL;
	worker->wqe = wqe;
671
	spin_lock_init(&worker->lock);
672 673

	worker->task = kthread_create_on_node(io_wqe_worker, worker, wqe->node,
674
				"io_wqe_worker-%d/%d", index, wqe->node);
675 676
	if (IS_ERR(worker->task)) {
		kfree(worker);
677
		return false;
678
	}
679
	kthread_bind_mask(worker->task, cpumask_of_node(wqe->node));
680

681
	raw_spin_lock_irq(&wqe->lock);
682
	hlist_nulls_add_head_rcu(&worker->nulls_node, &wqe->free_list);
683
	list_add_tail_rcu(&worker->all_list, &wqe->all_list);
684
	worker->flags |= IO_WORKER_F_FREE;
685 686 687
	if (index == IO_WQ_ACCT_BOUND)
		worker->flags |= IO_WORKER_F_BOUND;
	if (!acct->nr_workers && (worker->flags & IO_WORKER_F_BOUND))
688
		worker->flags |= IO_WORKER_F_FIXED;
689
	acct->nr_workers++;
690
	raw_spin_unlock_irq(&wqe->lock);
691

692 693 694
	if (index == IO_WQ_ACCT_UNBOUND)
		atomic_inc(&wq->user->processes);

695
	refcount_inc(&wq->refs);
696
	wake_up_process(worker->task);
697
	return true;
698 699
}

700
static inline bool io_wqe_need_worker(struct io_wqe *wqe, int index)
701 702
	__must_hold(wqe->lock)
{
703
	struct io_wqe_acct *acct = &wqe->acct[index];
704

705
	/* if we have available workers or no work, no need */
706
	if (!hlist_nulls_empty(&wqe->free_list) || !io_wqe_run_queue(wqe))
707 708
		return false;
	return acct->nr_workers < acct->max_workers;
709 710
}

711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747
static bool io_wqe_worker_send_sig(struct io_worker *worker, void *data)
{
	send_sig(SIGINT, worker->task, 1);
	return false;
}

/*
 * Iterate the passed in list and call the specific function for each
 * worker that isn't exiting
 */
static bool io_wq_for_each_worker(struct io_wqe *wqe,
				  bool (*func)(struct io_worker *, void *),
				  void *data)
{
	struct io_worker *worker;
	bool ret = false;

	list_for_each_entry_rcu(worker, &wqe->all_list, all_list) {
		if (io_worker_get(worker)) {
			/* no task if node is/was offline */
			if (worker->task)
				ret = func(worker, data);
			io_worker_release(worker);
			if (ret)
				break;
		}
	}

	return ret;
}

static bool io_wq_worker_wake(struct io_worker *worker, void *data)
{
	wake_up_process(worker->task);
	return false;
}

748 749 750 751 752 753
/*
 * Manager thread. Tasked with creating new workers, if we need them.
 */
static int io_wq_manager(void *data)
{
	struct io_wq *wq = data;
J
Jann Horn 已提交
754
	int node;
755

756
	/* create fixed workers */
757
	refcount_set(&wq->refs, 1);
J
Jann Horn 已提交
758
	for_each_node(node) {
759 760
		if (!node_online(node))
			continue;
761 762 763 764 765
		if (create_io_worker(wq, wq->wqes[node], IO_WQ_ACCT_BOUND))
			continue;
		set_bit(IO_WQ_BIT_ERROR, &wq->state);
		set_bit(IO_WQ_BIT_EXIT, &wq->state);
		goto out;
766
	}
767

768 769 770
	complete(&wq->done);

	while (!kthread_should_stop()) {
771 772 773
		if (current->task_works)
			task_work_run();

J
Jann Horn 已提交
774 775
		for_each_node(node) {
			struct io_wqe *wqe = wq->wqes[node];
776
			bool fork_worker[2] = { false, false };
777

778 779 780
			if (!node_online(node))
				continue;

781
			raw_spin_lock_irq(&wqe->lock);
782 783 784 785
			if (io_wqe_need_worker(wqe, IO_WQ_ACCT_BOUND))
				fork_worker[IO_WQ_ACCT_BOUND] = true;
			if (io_wqe_need_worker(wqe, IO_WQ_ACCT_UNBOUND))
				fork_worker[IO_WQ_ACCT_UNBOUND] = true;
786
			raw_spin_unlock_irq(&wqe->lock);
787 788 789 790
			if (fork_worker[IO_WQ_ACCT_BOUND])
				create_io_worker(wq, wqe, IO_WQ_ACCT_BOUND);
			if (fork_worker[IO_WQ_ACCT_UNBOUND])
				create_io_worker(wq, wqe, IO_WQ_ACCT_UNBOUND);
791 792 793 794 795
		}
		set_current_state(TASK_INTERRUPTIBLE);
		schedule_timeout(HZ);
	}

796 797 798
	if (current->task_works)
		task_work_run();

799 800
out:
	if (refcount_dec_and_test(&wq->refs)) {
801
		complete(&wq->done);
802 803 804 805 806 807 808 809 810
		return 0;
	}
	/* if ERROR is set and we get here, we have workers to wake */
	if (test_bit(IO_WQ_BIT_ERROR, &wq->state)) {
		rcu_read_lock();
		for_each_node(node)
			io_wq_for_each_worker(wq->wqes[node], io_wq_worker_wake, NULL);
		rcu_read_unlock();
	}
811
	return 0;
812 813
}

814 815 816 817 818 819 820 821 822 823 824
static bool io_wq_can_queue(struct io_wqe *wqe, struct io_wqe_acct *acct,
			    struct io_wq_work *work)
{
	bool free_worker;

	if (!(work->flags & IO_WQ_WORK_UNBOUND))
		return true;
	if (atomic_read(&acct->nr_running))
		return true;

	rcu_read_lock();
825
	free_worker = !hlist_nulls_empty(&wqe->free_list);
826 827 828 829 830 831 832 833 834 835 836
	rcu_read_unlock();
	if (free_worker)
		return true;

	if (atomic_read(&wqe->wq->user->processes) >= acct->max_workers &&
	    !(capable(CAP_SYS_RESOURCE) || capable(CAP_SYS_ADMIN)))
		return false;

	return true;
}

837
static void io_run_cancel(struct io_wq_work *work, struct io_wqe *wqe)
838
{
839 840
	struct io_wq *wq = wqe->wq;

841 842 843 844
	do {
		struct io_wq_work *old_work = work;

		work->flags |= IO_WQ_WORK_CANCEL;
845
		work = wq->do_work(work);
846
		wq->free_work(old_work);
847 848 849
	} while (work);
}

850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869
static void io_wqe_insert_work(struct io_wqe *wqe, struct io_wq_work *work)
{
	unsigned int hash;
	struct io_wq_work *tail;

	if (!io_wq_is_hashed(work)) {
append:
		wq_list_add_tail(&work->list, &wqe->work_list);
		return;
	}

	hash = io_get_work_hash(work);
	tail = wqe->hash_tail[hash];
	wqe->hash_tail[hash] = work;
	if (!tail)
		goto append;

	wq_list_add_after(&work->list, &tail->list, &wqe->work_list);
}

870 871
static void io_wqe_enqueue(struct io_wqe *wqe, struct io_wq_work *work)
{
872
	struct io_wqe_acct *acct = io_work_get_acct(wqe, work);
873
	int work_flags;
874 875
	unsigned long flags;

876 877 878 879 880 881 882
	/*
	 * Do early check to see if we need a new unbound worker, and if we do,
	 * if we're allowed to do so. This isn't 100% accurate as there's a
	 * gap between this check and incrementing the value, but that's OK.
	 * It's close enough to not be an issue, fork() has the same delay.
	 */
	if (unlikely(!io_wq_can_queue(wqe, acct, work))) {
883
		io_run_cancel(work, wqe);
884 885 886
		return;
	}

887
	work_flags = work->flags;
888
	raw_spin_lock_irqsave(&wqe->lock, flags);
889
	io_wqe_insert_work(wqe, work);
890
	wqe->flags &= ~IO_WQE_FLAG_STALLED;
891
	raw_spin_unlock_irqrestore(&wqe->lock, flags);
892

893 894
	if ((work_flags & IO_WQ_WORK_CONCURRENT) ||
	    !atomic_read(&acct->nr_running))
895
		io_wqe_wake_worker(wqe, acct);
896 897 898 899 900 901 902 903 904 905
}

void io_wq_enqueue(struct io_wq *wq, struct io_wq_work *work)
{
	struct io_wqe *wqe = wq->wqes[numa_node_id()];

	io_wqe_enqueue(wqe, work);
}

/*
906 907
 * Work items that hash to the same value will not be done in parallel.
 * Used to limit concurrent writes, generally hashed by inode.
908
 */
909
void io_wq_hash_work(struct io_wq_work *work, void *val)
910
{
911
	unsigned int bit;
912 913 914 915 916 917 918

	bit = hash_ptr(val, IO_WQ_HASH_ORDER);
	work->flags |= (IO_WQ_WORK_HASHED | (bit << IO_WQ_HASH_SHIFT));
}

void io_wq_cancel_all(struct io_wq *wq)
{
J
Jann Horn 已提交
919
	int node;
920 921 922 923

	set_bit(IO_WQ_BIT_CANCEL, &wq->state);

	rcu_read_lock();
J
Jann Horn 已提交
924 925
	for_each_node(node) {
		struct io_wqe *wqe = wq->wqes[node];
926

927
		io_wq_for_each_worker(wqe, io_wqe_worker_send_sig, NULL);
928 929 930 931
	}
	rcu_read_unlock();
}

932
struct io_cb_cancel_data {
933 934
	work_cancel_fn *fn;
	void *data;
935 936 937
	int nr_running;
	int nr_pending;
	bool cancel_all;
938 939
};

940
static bool io_wq_worker_cancel(struct io_worker *worker, void *data)
941
{
942
	struct io_cb_cancel_data *match = data;
943
	unsigned long flags;
944 945 946

	/*
	 * Hold the lock to avoid ->cur_work going out of scope, caller
947
	 * may dereference the passed in work.
948
	 */
949
	spin_lock_irqsave(&worker->lock, flags);
950
	if (worker->cur_work &&
951
	    !(worker->cur_work->flags & IO_WQ_WORK_NO_CANCEL) &&
952
	    match->fn(worker->cur_work, match->data)) {
953
		send_sig(SIGINT, worker->task, 1);
954
		match->nr_running++;
955
	}
956
	spin_unlock_irqrestore(&worker->lock, flags);
957

958
	return match->nr_running && !match->cancel_all;
959 960
}

961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978
static inline void io_wqe_remove_pending(struct io_wqe *wqe,
					 struct io_wq_work *work,
					 struct io_wq_work_node *prev)
{
	unsigned int hash = io_get_work_hash(work);
	struct io_wq_work *prev_work = NULL;

	if (io_wq_is_hashed(work) && work == wqe->hash_tail[hash]) {
		if (prev)
			prev_work = container_of(prev, struct io_wq_work, list);
		if (prev_work && io_get_work_hash(prev_work) == hash)
			wqe->hash_tail[hash] = prev_work;
		else
			wqe->hash_tail[hash] = NULL;
	}
	wq_list_del(&wqe->work_list, &work->list, prev);
}

979
static void io_wqe_cancel_pending_work(struct io_wqe *wqe,
980
				       struct io_cb_cancel_data *match)
981
{
J
Jens Axboe 已提交
982
	struct io_wq_work_node *node, *prev;
983
	struct io_wq_work *work;
984
	unsigned long flags;
985

986
retry:
987
	raw_spin_lock_irqsave(&wqe->lock, flags);
J
Jens Axboe 已提交
988 989
	wq_list_for_each(node, prev, &wqe->work_list) {
		work = container_of(node, struct io_wq_work, list);
990 991
		if (!match->fn(work, match->data))
			continue;
992
		io_wqe_remove_pending(wqe, work, prev);
993
		raw_spin_unlock_irqrestore(&wqe->lock, flags);
994 995 996 997 998 999 1000
		io_run_cancel(work, wqe);
		match->nr_pending++;
		if (!match->cancel_all)
			return;

		/* not safe to continue after unlock */
		goto retry;
1001
	}
1002
	raw_spin_unlock_irqrestore(&wqe->lock, flags);
1003 1004
}

1005
static void io_wqe_cancel_running_work(struct io_wqe *wqe,
1006 1007
				       struct io_cb_cancel_data *match)
{
1008
	rcu_read_lock();
1009
	io_wq_for_each_worker(wqe, io_wq_worker_cancel, match);
1010 1011 1012
	rcu_read_unlock();
}

1013
enum io_wq_cancel io_wq_cancel_cb(struct io_wq *wq, work_cancel_fn *cancel,
1014
				  void *data, bool cancel_all)
1015
{
1016
	struct io_cb_cancel_data match = {
1017 1018 1019
		.fn		= cancel,
		.data		= data,
		.cancel_all	= cancel_all,
1020
	};
J
Jann Horn 已提交
1021
	int node;
1022

1023 1024 1025 1026 1027
	/*
	 * First check pending list, if we're lucky we can just remove it
	 * from there. CANCEL_OK means that the work is returned as-new,
	 * no completion will be posted for it.
	 */
J
Jann Horn 已提交
1028 1029
	for_each_node(node) {
		struct io_wqe *wqe = wq->wqes[node];
1030

1031 1032
		io_wqe_cancel_pending_work(wqe, &match);
		if (match.nr_pending && !match.cancel_all)
1033
			return IO_WQ_CANCEL_OK;
1034 1035
	}

1036 1037 1038 1039 1040 1041 1042 1043 1044
	/*
	 * Now check if a free (going busy) or busy worker has the work
	 * currently running. If we find it there, we'll return CANCEL_RUNNING
	 * as an indication that we attempt to signal cancellation. The
	 * completion will run normally in this case.
	 */
	for_each_node(node) {
		struct io_wqe *wqe = wq->wqes[node];

1045 1046
		io_wqe_cancel_running_work(wqe, &match);
		if (match.nr_running && !match.cancel_all)
1047 1048 1049
			return IO_WQ_CANCEL_RUNNING;
	}

1050 1051 1052 1053
	if (match.nr_running)
		return IO_WQ_CANCEL_RUNNING;
	if (match.nr_pending)
		return IO_WQ_CANCEL_OK;
1054
	return IO_WQ_CANCEL_NOTFOUND;
1055 1056
}

1057 1058 1059 1060 1061 1062 1063
static bool io_wq_io_cb_cancel_data(struct io_wq_work *work, void *data)
{
	return work == data;
}

enum io_wq_cancel io_wq_cancel_work(struct io_wq *wq, struct io_wq_work *cwork)
{
1064
	return io_wq_cancel_cb(wq, io_wq_io_cb_cancel_data, (void *)cwork, false);
1065 1066
}

1067
struct io_wq *io_wq_create(unsigned bounded, struct io_wq_data *data)
1068
{
J
Jann Horn 已提交
1069
	int ret = -ENOMEM, node;
1070 1071
	struct io_wq *wq;

1072
	if (WARN_ON_ONCE(!data->free_work || !data->do_work))
1073 1074
		return ERR_PTR(-EINVAL);

1075
	wq = kzalloc(sizeof(*wq), GFP_KERNEL);
1076 1077 1078
	if (!wq)
		return ERR_PTR(-ENOMEM);

J
Jann Horn 已提交
1079
	wq->wqes = kcalloc(nr_node_ids, sizeof(struct io_wqe *), GFP_KERNEL);
1080 1081 1082 1083 1084
	if (!wq->wqes) {
		kfree(wq);
		return ERR_PTR(-ENOMEM);
	}

1085
	wq->free_work = data->free_work;
1086
	wq->do_work = data->do_work;
1087

1088
	/* caller must already hold a reference to this */
1089
	wq->user = data->user;
1090

J
Jann Horn 已提交
1091
	for_each_node(node) {
1092
		struct io_wqe *wqe;
1093
		int alloc_node = node;
1094

1095 1096 1097
		if (!node_online(alloc_node))
			alloc_node = NUMA_NO_NODE;
		wqe = kzalloc_node(sizeof(struct io_wqe), GFP_KERNEL, alloc_node);
1098
		if (!wqe)
J
Jann Horn 已提交
1099 1100
			goto err;
		wq->wqes[node] = wqe;
1101
		wqe->node = alloc_node;
1102 1103
		wqe->acct[IO_WQ_ACCT_BOUND].max_workers = bounded;
		atomic_set(&wqe->acct[IO_WQ_ACCT_BOUND].nr_running, 0);
1104
		if (wq->user) {
1105 1106 1107 1108
			wqe->acct[IO_WQ_ACCT_UNBOUND].max_workers =
					task_rlimit(current, RLIMIT_NPROC);
		}
		atomic_set(&wqe->acct[IO_WQ_ACCT_UNBOUND].nr_running, 0);
1109
		wqe->wq = wq;
1110
		raw_spin_lock_init(&wqe->lock);
J
Jens Axboe 已提交
1111
		INIT_WQ_LIST(&wqe->work_list);
1112
		INIT_HLIST_NULLS_HEAD(&wqe->free_list, 0);
1113
		INIT_LIST_HEAD(&wqe->all_list);
1114 1115 1116 1117 1118 1119 1120
	}

	init_completion(&wq->done);

	wq->manager = kthread_create(io_wq_manager, wq, "io_wq_manager");
	if (!IS_ERR(wq->manager)) {
		wake_up_process(wq->manager);
1121 1122 1123 1124 1125
		wait_for_completion(&wq->done);
		if (test_bit(IO_WQ_BIT_ERROR, &wq->state)) {
			ret = -ENOMEM;
			goto err;
		}
J
Jens Axboe 已提交
1126
		refcount_set(&wq->use_refs, 1);
1127
		reinit_completion(&wq->done);
1128 1129 1130 1131 1132
		return wq;
	}

	ret = PTR_ERR(wq->manager);
	complete(&wq->done);
1133
err:
J
Jann Horn 已提交
1134 1135
	for_each_node(node)
		kfree(wq->wqes[node]);
1136 1137
	kfree(wq->wqes);
	kfree(wq);
1138 1139 1140
	return ERR_PTR(ret);
}

1141 1142
bool io_wq_get(struct io_wq *wq, struct io_wq_data *data)
{
1143
	if (data->free_work != wq->free_work || data->do_work != wq->do_work)
1144 1145 1146 1147 1148
		return false;

	return refcount_inc_not_zero(&wq->use_refs);
}

J
Jens Axboe 已提交
1149
static void __io_wq_destroy(struct io_wq *wq)
1150
{
J
Jann Horn 已提交
1151
	int node;
1152

1153 1154
	set_bit(IO_WQ_BIT_EXIT, &wq->state);
	if (wq->manager)
1155 1156 1157
		kthread_stop(wq->manager);

	rcu_read_lock();
J
Jann Horn 已提交
1158 1159
	for_each_node(node)
		io_wq_for_each_worker(wq->wqes[node], io_wq_worker_wake, NULL);
1160 1161 1162 1163
	rcu_read_unlock();

	wait_for_completion(&wq->done);

J
Jann Horn 已提交
1164 1165
	for_each_node(node)
		kfree(wq->wqes[node]);
1166 1167 1168
	kfree(wq->wqes);
	kfree(wq);
}
J
Jens Axboe 已提交
1169 1170 1171 1172 1173 1174

void io_wq_destroy(struct io_wq *wq)
{
	if (refcount_dec_and_test(&wq->use_refs))
		__io_wq_destroy(wq);
}
1175 1176 1177 1178 1179

struct task_struct *io_wq_get_task(struct io_wq *wq)
{
	return wq->manager;
}