io-wq.c 27.0 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
// 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/mmu_context.h>
#include <linux/sched/mm.h>
#include <linux/percpu.h>
#include <linux/slab.h>
#include <linux/kthread.h>
#include <linux/rculist_nulls.h>
19
#include <linux/uaccess.h>
20
#include <linux/fs_struct.h>
21
#include <linux/task_work.h>
22 23 24 25 26 27 28 29 30 31 32

#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 */
	IO_WORKER_F_EXITING	= 8,	/* worker exiting */
	IO_WORKER_F_FIXED	= 16,	/* static idle worker */
33
	IO_WORKER_F_BOUND	= 32,	/* is doing bounded work */
34 35 36 37 38
};

enum {
	IO_WQ_BIT_EXIT		= 0,	/* wq exiting */
	IO_WQ_BIT_CANCEL	= 1,	/* cancel work on list */
39
	IO_WQ_BIT_ERROR		= 2,	/* error on setup */
40 41 42 43 44 45 46 47 48 49 50 51 52
};

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;
53
	struct list_head all_list;
54 55
	struct task_struct *task;
	struct io_wqe *wqe;
56

57
	struct io_wq_work *cur_work;
58
	spinlock_t lock;
59 60 61

	struct rcu_head rcu;
	struct mm_struct *mm;
62 63
	const struct cred *cur_creds;
	const struct cred *saved_creds;
64
	struct files_struct *restore_files;
65
	struct fs_struct *restore_fs;
66 67 68 69 70 71 72 73
};

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

74 75
#define IO_WQ_NR_HASH_BUCKETS	(1u << IO_WQ_HASH_ORDER)

76 77 78 79 80 81 82 83 84 85 86
struct io_wqe_acct {
	unsigned nr_workers;
	unsigned max_workers;
	atomic_t nr_running;
};

enum {
	IO_WQ_ACCT_BOUND,
	IO_WQ_ACCT_UNBOUND,
};

87 88 89 90 91 92
/*
 * Per-node worker thread pool
 */
struct io_wqe {
	struct {
		spinlock_t lock;
J
Jens Axboe 已提交
93
		struct io_wq_work_list work_list;
94 95 96 97 98
		unsigned long hash_map;
		unsigned flags;
	} ____cacheline_aligned_in_smp;

	int node;
99
	struct io_wqe_acct acct[2];
100

101
	struct hlist_nulls_head free_list;
102
	struct list_head all_list;
103 104

	struct io_wq *wq;
105
	struct io_wq_work *hash_tail[IO_WQ_NR_HASH_BUCKETS];
106 107 108 109 110 111 112 113 114
};

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

115
	free_work_fn *free_work;
116

117
	struct task_struct *manager;
118
	struct user_struct *user;
119 120
	refcount_t refs;
	struct completion done;
J
Jens Axboe 已提交
121 122

	refcount_t use_refs;
123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142
};

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)
{
143 144
	bool dropped_lock = false;

145 146 147
	if (worker->saved_creds) {
		revert_creds(worker->saved_creds);
		worker->cur_creds = worker->saved_creds = NULL;
148 149
	}

150 151 152 153 154 155 156 157 158 159
	if (current->files != worker->restore_files) {
		__acquire(&wqe->lock);
		spin_unlock_irq(&wqe->lock);
		dropped_lock = true;

		task_lock(current);
		current->files = worker->restore_files;
		task_unlock(current);
	}

160 161 162
	if (current->fs != worker->restore_fs)
		current->fs = worker->restore_fs;

163 164 165 166 167
	/*
	 * 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) {
168 169 170 171 172
		if (!dropped_lock) {
			__acquire(&wqe->lock);
			spin_unlock_irq(&wqe->lock);
			dropped_lock = true;
		}
173 174 175 176 177 178 179
		__set_current_state(TASK_RUNNING);
		set_fs(KERNEL_DS);
		unuse_mm(worker->mm);
		mmput(worker->mm);
		worker->mm = NULL;
	}

180
	return dropped_lock;
181 182
}

183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200
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];
}

201 202 203
static void io_worker_exit(struct io_worker *worker)
{
	struct io_wqe *wqe = worker->wqe;
204 205
	struct io_wqe_acct *acct = io_wqe_get_acct(wqe, worker);
	unsigned nr_workers;
206 207 208 209 210 211 212 213 214 215 216 217 218

	/*
	 * 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)
219 220 221
		atomic_dec(&acct->nr_running);
	if (!(worker->flags & IO_WORKER_F_BOUND))
		atomic_dec(&wqe->wq->user->processes);
222 223 224 225 226
	worker->flags = 0;
	preempt_enable();

	spin_lock_irq(&wqe->lock);
	hlist_nulls_del_rcu(&worker->nulls_node);
227
	list_del_rcu(&worker->all_list);
228 229 230 231
	if (__io_worker_unuse(wqe, worker)) {
		__release(&wqe->lock);
		spin_lock_irq(&wqe->lock);
	}
232 233 234
	acct->nr_workers--;
	nr_workers = wqe->acct[IO_WQ_ACCT_BOUND].nr_workers +
			wqe->acct[IO_WQ_ACCT_UNBOUND].nr_workers;
235 236 237
	spin_unlock_irq(&wqe->lock);

	/* all workers gone, wq exit can proceed */
238
	if (!nr_workers && refcount_dec_and_test(&wqe->wq->refs))
239 240
		complete(&wqe->wq->done);

241
	kfree_rcu(worker, rcu);
242 243
}

244 245 246
static inline bool io_wqe_run_queue(struct io_wqe *wqe)
	__must_hold(wqe->lock)
{
J
Jens Axboe 已提交
247 248
	if (!wq_list_empty(&wqe->work_list) &&
	    !(wqe->flags & IO_WQE_FLAG_STALLED))
249 250 251 252 253 254 255 256 257 258 259 260 261 262
		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;

263
	n = rcu_dereference(hlist_nulls_first_rcu(&wqe->free_list));
264 265 266 267 268
	if (is_a_nulls(n))
		return false;

	worker = hlist_nulls_entry(n, struct io_worker, nulls_node);
	if (io_worker_get(worker)) {
269
		wake_up_process(worker->task);
270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314
		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);
}

315 316 317 318 319 320 321
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);
322
	worker->restore_files = current->files;
323
	worker->restore_fs = current->fs;
324
	io_wqe_inc_running(wqe, worker);
325 326 327 328 329 330 331 332 333 334
}

/*
 * 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)
{
335 336
	bool worker_bound, work_bound;

337 338 339 340
	if (worker->flags & IO_WORKER_F_FREE) {
		worker->flags &= ~IO_WORKER_F_FREE;
		hlist_nulls_del_init_rcu(&worker->nulls_node);
	}
341 342 343 344 345

	/*
	 * If worker is moving from bound to unbound (or vice versa), then
	 * ensure we update the running accounting.
	 */
346 347 348
	worker_bound = (worker->flags & IO_WORKER_F_BOUND) != 0;
	work_bound = (work->flags & IO_WQ_WORK_UNBOUND) == 0;
	if (worker_bound != work_bound) {
349 350 351 352 353 354 355 356 357 358 359 360 361 362
		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);
	 }
363 364 365 366 367 368 369 370 371 372 373 374 375 376
}

/*
 * 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;
377
		hlist_nulls_add_head_rcu(&worker->nulls_node, &wqe->free_list);
378 379 380 381 382
	}

	return __io_worker_unuse(wqe, worker);
}

P
Pavel Begunkov 已提交
383 384 385 386 387 388
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)
389 390
	__must_hold(wqe->lock)
{
J
Jens Axboe 已提交
391
	struct io_wq_work_node *node, *prev;
392
	struct io_wq_work *work, *tail;
P
Pavel Begunkov 已提交
393
	unsigned int hash;
394

J
Jens Axboe 已提交
395 396 397
	wq_list_for_each(node, prev, &wqe->work_list) {
		work = container_of(node, struct io_wq_work, list);

398
		/* not hashed, can run anytime */
399
		if (!io_wq_is_hashed(work)) {
400
			wq_list_del(&wqe->work_list, node, prev);
401 402 403 404
			return work;
		}

		/* hashed, can run if not already running */
P
Pavel Begunkov 已提交
405 406 407
		hash = io_get_work_hash(work);
		if (!(wqe->hash_map & BIT(hash))) {
			wqe->hash_map |= BIT(hash);
408 409 410 411
			/* 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);
412 413 414 415 416 417 418
			return work;
		}
	}

	return NULL;
}

419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455
static void io_wq_switch_mm(struct io_worker *worker, struct io_wq_work *work)
{
	if (worker->mm) {
		unuse_mm(worker->mm);
		mmput(worker->mm);
		worker->mm = NULL;
	}
	if (!work->mm) {
		set_fs(KERNEL_DS);
		return;
	}
	if (mmget_not_zero(work->mm)) {
		use_mm(work->mm);
		if (!worker->mm)
			set_fs(USER_DS);
		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;
}

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;
}

456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474
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;
		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);
}

static void io_assign_current_work(struct io_worker *worker,
				   struct io_wq_work *work)
{
475 476 477 478 479 480
	if (work) {
		/* flush pending signals before assigning new work */
		if (signal_pending(current))
			flush_signals(current);
		cond_resched();
	}
481 482 483 484 485 486

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

P
Pavel Begunkov 已提交
487 488
static void io_wqe_enqueue(struct io_wqe *wqe, struct io_wq_work *work);

489 490 491 492 493 494 495
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 {
496
		struct io_wq_work *work;
P
Pavel Begunkov 已提交
497
		unsigned int hash;
498
get_next:
499 500 501 502 503 504 505
		/*
		 * 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
506
		work = io_get_next_work(wqe);
507 508
		if (work)
			__io_worker_busy(wqe, worker, work);
J
Jens Axboe 已提交
509
		else if (!wq_list_empty(&wqe->work_list))
510 511 512 513 514
			wqe->flags |= IO_WQE_FLAG_STALLED;

		spin_unlock_irq(&wqe->lock);
		if (!work)
			break;
515
		io_assign_current_work(worker, work);
516

517 518
		/* handle a whole dependent link */
		do {
519
			struct io_wq_work *old_work, *next_hashed, *linked;
520

521
			next_hashed = wq_next_work(work);
522
			io_impersonate_work(worker, work);
523 524 525 526 527 528 529
			/*
			 * 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;

P
Pavel Begunkov 已提交
530
			hash = io_get_work_hash(work);
531 532 533 534 535 536 537 538 539 540
			linked = old_work = work;
			linked->func(&linked);
			linked = (old_work == linked) ? NULL : linked;

			work = next_hashed;
			if (!work && linked && !io_wq_is_hashed(linked)) {
				work = linked;
				linked = NULL;
			}
			io_assign_current_work(worker, work);
541
			wq->free_work(old_work);
542

543 544 545 546
			if (linked)
				io_wqe_enqueue(wqe, linked);

			if (hash != -1U && !next_hashed) {
547 548 549 550 551
				spin_lock_irq(&wqe->lock);
				wqe->hash_map &= ~BIT_ULL(hash);
				wqe->flags &= ~IO_WQE_FLAG_STALLED;
				/* dependent work is not hashed */
				hash = -1U;
552 553 554 555
				/* skip unnecessary unlock-lock wqe->lock */
				if (!work)
					goto get_next;
				spin_unlock_irq(&wqe->lock);
556
			}
557
		} while (work);
558

559
		spin_lock_irq(&wqe->lock);
560 561 562 563 564 565 566 567 568 569 570 571
	} 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)) {
572
		set_current_state(TASK_INTERRUPTIBLE);
573
loop:
574 575 576 577
		spin_lock_irq(&wqe->lock);
		if (io_wqe_run_queue(wqe)) {
			__set_current_state(TASK_RUNNING);
			io_worker_handle_work(worker);
578
			goto loop;
579 580 581 582
		}
		/* drops the lock on success, retry */
		if (__io_worker_idle(wqe, worker)) {
			__release(&wqe->lock);
583
			goto loop;
584 585 586 587 588 589 590 591 592 593 594 595 596 597
		}
		spin_unlock_irq(&wqe->lock);
		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)) {
		spin_lock_irq(&wqe->lock);
J
Jens Axboe 已提交
598
		if (!wq_list_empty(&wqe->work_list))
599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620
			io_worker_handle_work(worker);
		else
			spin_unlock_irq(&wqe->lock);
	}

	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;
621
	io_wqe_inc_running(wqe, worker);
622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641
}

/*
 * 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;

	spin_lock_irq(&wqe->lock);
642
	io_wqe_dec_running(wqe, worker);
643 644 645
	spin_unlock_irq(&wqe->lock);
}

646
static bool create_io_worker(struct io_wq *wq, struct io_wqe *wqe, int index)
647
{
648
	struct io_wqe_acct *acct =&wqe->acct[index];
649 650
	struct io_worker *worker;

651
	worker = kzalloc_node(sizeof(*worker), GFP_KERNEL, wqe->node);
652
	if (!worker)
653
		return false;
654 655 656 657

	refcount_set(&worker->ref, 1);
	worker->nulls_node.pprev = NULL;
	worker->wqe = wqe;
658
	spin_lock_init(&worker->lock);
659 660

	worker->task = kthread_create_on_node(io_wqe_worker, worker, wqe->node,
661
				"io_wqe_worker-%d/%d", index, wqe->node);
662 663
	if (IS_ERR(worker->task)) {
		kfree(worker);
664
		return false;
665 666 667
	}

	spin_lock_irq(&wqe->lock);
668
	hlist_nulls_add_head_rcu(&worker->nulls_node, &wqe->free_list);
669
	list_add_tail_rcu(&worker->all_list, &wqe->all_list);
670
	worker->flags |= IO_WORKER_F_FREE;
671 672 673
	if (index == IO_WQ_ACCT_BOUND)
		worker->flags |= IO_WORKER_F_BOUND;
	if (!acct->nr_workers && (worker->flags & IO_WORKER_F_BOUND))
674
		worker->flags |= IO_WORKER_F_FIXED;
675
	acct->nr_workers++;
676 677
	spin_unlock_irq(&wqe->lock);

678 679 680
	if (index == IO_WQ_ACCT_UNBOUND)
		atomic_inc(&wq->user->processes);

681
	wake_up_process(worker->task);
682
	return true;
683 684
}

685
static inline bool io_wqe_need_worker(struct io_wqe *wqe, int index)
686 687
	__must_hold(wqe->lock)
{
688
	struct io_wqe_acct *acct = &wqe->acct[index];
689

690
	/* if we have available workers or no work, no need */
691
	if (!hlist_nulls_empty(&wqe->free_list) || !io_wqe_run_queue(wqe))
692 693
		return false;
	return acct->nr_workers < acct->max_workers;
694 695 696 697 698 699 700 701
}

/*
 * 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
702 703
	int workers_to_create = num_possible_nodes();
	int node;
704

705
	/* create fixed workers */
J
Jann Horn 已提交
706 707
	refcount_set(&wq->refs, workers_to_create);
	for_each_node(node) {
708 709
		if (!node_online(node))
			continue;
J
Jann Horn 已提交
710 711 712
		if (!create_io_worker(wq, wq->wqes[node], IO_WQ_ACCT_BOUND))
			goto err;
		workers_to_create--;
713
	}
714

715 716 717
	while (workers_to_create--)
		refcount_dec(&wq->refs);

718 719 720
	complete(&wq->done);

	while (!kthread_should_stop()) {
721 722 723
		if (current->task_works)
			task_work_run();

J
Jann Horn 已提交
724 725
		for_each_node(node) {
			struct io_wqe *wqe = wq->wqes[node];
726
			bool fork_worker[2] = { false, false };
727

728 729 730
			if (!node_online(node))
				continue;

731
			spin_lock_irq(&wqe->lock);
732 733 734 735
			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;
736
			spin_unlock_irq(&wqe->lock);
737 738 739 740
			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);
741 742 743 744 745
		}
		set_current_state(TASK_INTERRUPTIBLE);
		schedule_timeout(HZ);
	}

746 747 748
	if (current->task_works)
		task_work_run();

749
	return 0;
750 751 752
err:
	set_bit(IO_WQ_BIT_ERROR, &wq->state);
	set_bit(IO_WQ_BIT_EXIT, &wq->state);
J
Jann Horn 已提交
753
	if (refcount_sub_and_test(workers_to_create, &wq->refs))
754 755
		complete(&wq->done);
	return 0;
756 757
}

758 759 760 761 762 763 764 765 766 767 768
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();
769
	free_worker = !hlist_nulls_empty(&wqe->free_list);
770 771 772 773 774 775 776 777 778 779 780
	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;
}

781
static void io_run_cancel(struct io_wq_work *work, struct io_wqe *wqe)
782
{
783 784
	struct io_wq *wq = wqe->wq;

785 786 787 788 789 790
	do {
		struct io_wq_work *old_work = work;

		work->flags |= IO_WQ_WORK_CANCEL;
		work->func(&work);
		work = (work == old_work) ? NULL : work;
791
		wq->free_work(old_work);
792 793 794
	} while (work);
}

795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814
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);
}

815 816
static void io_wqe_enqueue(struct io_wqe *wqe, struct io_wq_work *work)
{
817
	struct io_wqe_acct *acct = io_work_get_acct(wqe, work);
818
	int work_flags;
819 820
	unsigned long flags;

821 822 823 824 825 826 827
	/*
	 * 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))) {
828
		io_run_cancel(work, wqe);
829 830 831
		return;
	}

832
	work_flags = work->flags;
833
	spin_lock_irqsave(&wqe->lock, flags);
834
	io_wqe_insert_work(wqe, work);
835 836 837
	wqe->flags &= ~IO_WQE_FLAG_STALLED;
	spin_unlock_irqrestore(&wqe->lock, flags);

838 839
	if ((work_flags & IO_WQ_WORK_CONCURRENT) ||
	    !atomic_read(&acct->nr_running))
840
		io_wqe_wake_worker(wqe, acct);
841 842 843 844 845 846 847 848 849 850
}

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);
}

/*
851 852
 * Work items that hash to the same value will not be done in parallel.
 * Used to limit concurrent writes, generally hashed by inode.
853
 */
854
void io_wq_hash_work(struct io_wq_work *work, void *val)
855
{
856
	unsigned int bit;
857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878

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

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;

879
	list_for_each_entry_rcu(worker, &wqe->all_list, all_list) {
880
		if (io_worker_get(worker)) {
881 882 883
			/* no task if node is/was offline */
			if (worker->task)
				ret = func(worker, data);
884 885 886 887 888
			io_worker_release(worker);
			if (ret)
				break;
		}
	}
889

890 891 892 893 894
	return ret;
}

void io_wq_cancel_all(struct io_wq *wq)
{
J
Jann Horn 已提交
895
	int node;
896 897 898 899

	set_bit(IO_WQ_BIT_CANCEL, &wq->state);

	rcu_read_lock();
J
Jann Horn 已提交
900 901
	for_each_node(node) {
		struct io_wqe *wqe = wq->wqes[node];
902

903
		io_wq_for_each_worker(wqe, io_wqe_worker_send_sig, NULL);
904 905 906 907
	}
	rcu_read_unlock();
}

908
struct io_cb_cancel_data {
909 910
	work_cancel_fn *fn;
	void *data;
911 912
};

913
static bool io_wq_worker_cancel(struct io_worker *worker, void *data)
914
{
915
	struct io_cb_cancel_data *match = data;
916
	unsigned long flags;
917 918 919 920
	bool ret = false;

	/*
	 * Hold the lock to avoid ->cur_work going out of scope, caller
921
	 * may dereference the passed in work.
922
	 */
923
	spin_lock_irqsave(&worker->lock, flags);
924
	if (worker->cur_work &&
925
	    !(worker->cur_work->flags & IO_WQ_WORK_NO_CANCEL) &&
926
	    match->fn(worker->cur_work, match->data)) {
927
		send_sig(SIGINT, worker->task, 1);
928
		ret = true;
929
	}
930
	spin_unlock_irqrestore(&worker->lock, flags);
931

932
	return ret;
933 934 935
}

static enum io_wq_cancel io_wqe_cancel_work(struct io_wqe *wqe,
936
					    struct io_cb_cancel_data *match)
937
{
J
Jens Axboe 已提交
938
	struct io_wq_work_node *node, *prev;
939
	struct io_wq_work *work;
940
	unsigned long flags;
941 942 943 944 945 946 947
	bool found = false;

	/*
	 * 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.
	 */
948
	spin_lock_irqsave(&wqe->lock, flags);
J
Jens Axboe 已提交
949 950 951
	wq_list_for_each(node, prev, &wqe->work_list) {
		work = container_of(node, struct io_wq_work, list);

952
		if (match->fn(work, match->data)) {
953
			wq_list_del(&wqe->work_list, node, prev);
954 955 956 957
			found = true;
			break;
		}
	}
958
	spin_unlock_irqrestore(&wqe->lock, flags);
959 960

	if (found) {
961
		io_run_cancel(work, wqe);
962 963 964 965 966 967
		return IO_WQ_CANCEL_OK;
	}

	/*
	 * 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
968
	 * as an indication that we attempt to signal cancellation. The
969 970 971
	 * completion will run normally in this case.
	 */
	rcu_read_lock();
972
	found = io_wq_for_each_worker(wqe, io_wq_worker_cancel, match);
973 974 975 976
	rcu_read_unlock();
	return found ? IO_WQ_CANCEL_RUNNING : IO_WQ_CANCEL_NOTFOUND;
}

977 978
enum io_wq_cancel io_wq_cancel_cb(struct io_wq *wq, work_cancel_fn *cancel,
				  void *data)
979
{
980 981 982
	struct io_cb_cancel_data match = {
		.fn	= cancel,
		.data	= data,
983
	};
984
	enum io_wq_cancel ret = IO_WQ_CANCEL_NOTFOUND;
J
Jann Horn 已提交
985
	int node;
986

J
Jann Horn 已提交
987 988
	for_each_node(node) {
		struct io_wqe *wqe = wq->wqes[node];
989

990
		ret = io_wqe_cancel_work(wqe, &match);
991 992 993 994 995 996 997
		if (ret != IO_WQ_CANCEL_NOTFOUND)
			break;
	}

	return ret;
}

998 999 1000 1001 1002 1003 1004 1005 1006 1007
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)
{
	return io_wq_cancel_cb(wq, io_wq_io_cb_cancel_data, (void *)cwork);
}

1008 1009 1010 1011
static bool io_wq_pid_match(struct io_wq_work *work, void *data)
{
	pid_t pid = (pid_t) (unsigned long) data;

1012
	return work->task_pid == pid;
1013 1014 1015 1016
}

enum io_wq_cancel io_wq_cancel_pid(struct io_wq *wq, pid_t pid)
{
1017
	void *data = (void *) (unsigned long) pid;
1018

1019
	return io_wq_cancel_cb(wq, io_wq_pid_match, data);
1020 1021
}

1022
struct io_wq *io_wq_create(unsigned bounded, struct io_wq_data *data)
1023
{
J
Jann Horn 已提交
1024
	int ret = -ENOMEM, node;
1025 1026
	struct io_wq *wq;

1027 1028 1029
	if (WARN_ON_ONCE(!data->free_work))
		return ERR_PTR(-EINVAL);

1030
	wq = kzalloc(sizeof(*wq), GFP_KERNEL);
1031 1032 1033
	if (!wq)
		return ERR_PTR(-ENOMEM);

J
Jann Horn 已提交
1034
	wq->wqes = kcalloc(nr_node_ids, sizeof(struct io_wqe *), GFP_KERNEL);
1035 1036 1037 1038 1039
	if (!wq->wqes) {
		kfree(wq);
		return ERR_PTR(-ENOMEM);
	}

1040
	wq->free_work = data->free_work;
1041

1042
	/* caller must already hold a reference to this */
1043
	wq->user = data->user;
1044

J
Jann Horn 已提交
1045
	for_each_node(node) {
1046
		struct io_wqe *wqe;
1047
		int alloc_node = node;
1048

1049 1050 1051
		if (!node_online(alloc_node))
			alloc_node = NUMA_NO_NODE;
		wqe = kzalloc_node(sizeof(struct io_wqe), GFP_KERNEL, alloc_node);
1052
		if (!wqe)
J
Jann Horn 已提交
1053 1054
			goto err;
		wq->wqes[node] = wqe;
1055
		wqe->node = alloc_node;
1056 1057
		wqe->acct[IO_WQ_ACCT_BOUND].max_workers = bounded;
		atomic_set(&wqe->acct[IO_WQ_ACCT_BOUND].nr_running, 0);
1058
		if (wq->user) {
1059 1060 1061 1062
			wqe->acct[IO_WQ_ACCT_UNBOUND].max_workers =
					task_rlimit(current, RLIMIT_NPROC);
		}
		atomic_set(&wqe->acct[IO_WQ_ACCT_UNBOUND].nr_running, 0);
1063 1064
		wqe->wq = wq;
		spin_lock_init(&wqe->lock);
J
Jens Axboe 已提交
1065
		INIT_WQ_LIST(&wqe->work_list);
1066
		INIT_HLIST_NULLS_HEAD(&wqe->free_list, 0);
1067
		INIT_LIST_HEAD(&wqe->all_list);
1068 1069 1070 1071 1072 1073 1074
	}

	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);
1075 1076 1077 1078 1079
		wait_for_completion(&wq->done);
		if (test_bit(IO_WQ_BIT_ERROR, &wq->state)) {
			ret = -ENOMEM;
			goto err;
		}
J
Jens Axboe 已提交
1080
		refcount_set(&wq->use_refs, 1);
1081
		reinit_completion(&wq->done);
1082 1083 1084 1085 1086
		return wq;
	}

	ret = PTR_ERR(wq->manager);
	complete(&wq->done);
1087
err:
J
Jann Horn 已提交
1088 1089
	for_each_node(node)
		kfree(wq->wqes[node]);
1090 1091
	kfree(wq->wqes);
	kfree(wq);
1092 1093 1094
	return ERR_PTR(ret);
}

1095 1096
bool io_wq_get(struct io_wq *wq, struct io_wq_data *data)
{
1097
	if (data->free_work != wq->free_work)
1098 1099 1100 1101 1102
		return false;

	return refcount_inc_not_zero(&wq->use_refs);
}

1103 1104 1105 1106 1107 1108
static bool io_wq_worker_wake(struct io_worker *worker, void *data)
{
	wake_up_process(worker->task);
	return false;
}

J
Jens Axboe 已提交
1109
static void __io_wq_destroy(struct io_wq *wq)
1110
{
J
Jann Horn 已提交
1111
	int node;
1112

1113 1114
	set_bit(IO_WQ_BIT_EXIT, &wq->state);
	if (wq->manager)
1115 1116 1117
		kthread_stop(wq->manager);

	rcu_read_lock();
J
Jann Horn 已提交
1118 1119
	for_each_node(node)
		io_wq_for_each_worker(wq->wqes[node], io_wq_worker_wake, NULL);
1120 1121 1122 1123
	rcu_read_unlock();

	wait_for_completion(&wq->done);

J
Jann Horn 已提交
1124 1125
	for_each_node(node)
		kfree(wq->wqes[node]);
1126 1127 1128
	kfree(wq->wqes);
	kfree(wq);
}
J
Jens Axboe 已提交
1129 1130 1131 1132 1133 1134

void io_wq_destroy(struct io_wq *wq)
{
	if (refcount_dec_and_test(&wq->use_refs))
		__io_wq_destroy(wq);
}
1135 1136 1137 1138 1139

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