blk-flush.c 17 KB
Newer Older
1
/*
2
 * Functions to sequence PREFLUSH and FUA writes.
3 4 5 6 7 8
 *
 * Copyright (C) 2011		Max Planck Institute for Gravitational Physics
 * Copyright (C) 2011		Tejun Heo <tj@kernel.org>
 *
 * This file is released under the GPLv2.
 *
9
 * REQ_{PREFLUSH|FUA} requests are decomposed to sequences consisted of three
10 11 12
 * optional steps - PREFLUSH, DATA and POSTFLUSH - according to the request
 * properties and hardware capability.
 *
13 14
 * If a request doesn't have data, only REQ_PREFLUSH makes sense, which
 * indicates a simple flush request.  If there is data, REQ_PREFLUSH indicates
15 16 17 18
 * that the device cache should be flushed before the data is executed, and
 * REQ_FUA means that the data must be on non-volatile media on request
 * completion.
 *
19 20 21
 * If the device doesn't have writeback cache, PREFLUSH and FUA don't make any
 * difference.  The requests are either completed immediately if there's no data
 * or executed as normal requests otherwise.
22
 *
23
 * If the device has writeback cache and supports FUA, REQ_PREFLUSH is
24 25
 * translated to PREFLUSH but REQ_FUA is passed down directly with DATA.
 *
26 27
 * If the device has writeback cache and doesn't support FUA, REQ_PREFLUSH
 * is translated to PREFLUSH and REQ_FUA to POSTFLUSH.
28 29 30
 *
 * The actual execution of flush is double buffered.  Whenever a request
 * needs to execute PRE or POSTFLUSH, it queues at
31
 * fq->flush_queue[fq->flush_pending_idx].  Once certain criteria are met, a
32
 * REQ_OP_FLUSH is issued and the pending_idx is toggled.  When the flush
33
 * completes, all the requests which were pending are proceeded to the next
34
 * step.  This allows arbitrary merging of different types of PREFLUSH/FUA
35 36 37 38 39 40 41 42 43 44 45 46 47 48 49
 * requests.
 *
 * Currently, the following conditions are used to determine when to issue
 * flush.
 *
 * C1. At any given time, only one flush shall be in progress.  This makes
 *     double buffering sufficient.
 *
 * C2. Flush is deferred if any request is executing DATA of its sequence.
 *     This avoids issuing separate POSTFLUSHes for requests which shared
 *     PREFLUSH.
 *
 * C3. The second condition is ignored if there is a request which has
 *     waited longer than FLUSH_PENDING_TIMEOUT.  This is to avoid
 *     starvation in the unlikely case where there are continuous stream of
50
 *     FUA (without PREFLUSH) requests.
51 52 53 54
 *
 * For devices which support FUA, it isn't clear whether C2 (and thus C3)
 * is beneficial.
 *
55
 * Note that a sequenced PREFLUSH/FUA request with DATA is completed twice.
56 57 58
 * Once while executing DATA and again after the whole sequence is
 * complete.  The first completion updates the contained bio but doesn't
 * finish it so that the bio submitter is notified only after the whole
59
 * sequence is complete.  This is implemented by testing RQF_FLUSH_SEQ in
60 61
 * req_bio_endio().
 *
62
 * The above peculiarity requires that each PREFLUSH/FUA request has only one
63 64
 * bio attached to it, which is guaranteed as they aren't allowed to be
 * merged in the usual way.
65
 */
66

67 68 69 70
#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/bio.h>
#include <linux/blkdev.h>
71
#include <linux/gfp.h>
72
#include <linux/blk-mq.h>
73 74

#include "blk.h"
75
#include "blk-mq.h"
76
#include "blk-mq-tag.h"
77
#include "blk-mq-sched.h"
78

79
/* PREFLUSH/FUA sequences */
80
enum {
81 82 83 84 85 86 87 88 89 90 91 92 93
	REQ_FSEQ_PREFLUSH	= (1 << 0), /* pre-flushing in progress */
	REQ_FSEQ_DATA		= (1 << 1), /* data write in progress */
	REQ_FSEQ_POSTFLUSH	= (1 << 2), /* post-flushing in progress */
	REQ_FSEQ_DONE		= (1 << 3),

	REQ_FSEQ_ACTIONS	= REQ_FSEQ_PREFLUSH | REQ_FSEQ_DATA |
				  REQ_FSEQ_POSTFLUSH,

	/*
	 * If flush has been pending longer than the following timeout,
	 * it's issued even if flush_data requests are still in flight.
	 */
	FLUSH_PENDING_TIMEOUT	= 5 * HZ,
94 95
};

96 97
static bool blk_kick_flush(struct request_queue *q,
			   struct blk_flush_queue *fq);
98

Jens Axboe's avatar
Jens Axboe committed
99
static unsigned int blk_flush_policy(unsigned long fflags, struct request *rq)
100
{
101
	unsigned int policy = 0;
102

103 104 105
	if (blk_rq_sectors(rq))
		policy |= REQ_FSEQ_DATA;

Jens Axboe's avatar
Jens Axboe committed
106
	if (fflags & (1UL << QUEUE_FLAG_WC)) {
107
		if (rq->cmd_flags & REQ_PREFLUSH)
108
			policy |= REQ_FSEQ_PREFLUSH;
Jens Axboe's avatar
Jens Axboe committed
109 110
		if (!(fflags & (1UL << QUEUE_FLAG_FUA)) &&
		    (rq->cmd_flags & REQ_FUA))
111
			policy |= REQ_FSEQ_POSTFLUSH;
112
	}
113
	return policy;
114 115
}

116
static unsigned int blk_flush_cur_seq(struct request *rq)
117
{
118 119
	return 1 << ffz(rq->flush.seq);
}
120

121 122
static void blk_flush_restore_request(struct request *rq)
{
123
	/*
124 125 126
	 * After flush data completion, @rq->bio is %NULL but we need to
	 * complete the bio again.  @rq->biotail is guaranteed to equal the
	 * original @rq->bio.  Restore it.
127
	 */
128 129 130
	rq->bio = rq->biotail;

	/* make @rq a normal request */
131
	rq->rq_flags &= ~RQF_FLUSH_SEQ;
132
	rq->end_io = rq->flush.saved_end_io;
133 134
}

135
static bool blk_flush_queue_rq(struct request *rq, bool add_front)
136
{
137
	if (rq->q->mq_ops) {
138
		blk_mq_add_to_requeue_list(rq, add_front, true);
139 140
		return false;
	} else {
141 142 143 144
		if (add_front)
			list_add(&rq->queuelist, &rq->q->queue_head);
		else
			list_add_tail(&rq->queuelist, &rq->q->queue_head);
145 146
		return true;
	}
147 148
}

149 150
/**
 * blk_flush_complete_seq - complete flush sequence
151
 * @rq: PREFLUSH/FUA request being sequenced
152
 * @fq: flush queue
153 154 155 156 157 158 159
 * @seq: sequences to complete (mask of %REQ_FSEQ_*, can be zero)
 * @error: whether an error occurred
 *
 * @rq just completed @seq part of its flush sequence, record the
 * completion and trigger the next step.
 *
 * CONTEXT:
160
 * spin_lock_irq(q->queue_lock or fq->mq_flush_lock)
161 162 163 164
 *
 * RETURNS:
 * %true if requests were added to the dispatch queue, %false otherwise.
 */
165 166
static bool blk_flush_complete_seq(struct request *rq,
				   struct blk_flush_queue *fq,
167
				   unsigned int seq, blk_status_t error)
168
{
169
	struct request_queue *q = rq->q;
170
	struct list_head *pending = &fq->flush_queue[fq->flush_pending_idx];
171
	bool queued = false, kicked;
172 173 174 175 176 177 178 179 180 181 182 183 184 185

	BUG_ON(rq->flush.seq & seq);
	rq->flush.seq |= seq;

	if (likely(!error))
		seq = blk_flush_cur_seq(rq);
	else
		seq = REQ_FSEQ_DONE;

	switch (seq) {
	case REQ_FSEQ_PREFLUSH:
	case REQ_FSEQ_POSTFLUSH:
		/* queue for flush */
		if (list_empty(pending))
186
			fq->flush_pending_since = jiffies;
187 188 189 190
		list_move_tail(&rq->flush.list, pending);
		break;

	case REQ_FSEQ_DATA:
191
		list_move_tail(&rq->flush.list, &fq->flush_data_in_flight);
192
		queued = blk_flush_queue_rq(rq, true);
193 194 195 196 197 198 199 200 201 202 203 204
		break;

	case REQ_FSEQ_DONE:
		/*
		 * @rq was previously adjusted by blk_flush_issue() for
		 * flush sequencing and may already have gone through the
		 * flush data request completion path.  Restore @rq for
		 * normal completion and end it.
		 */
		BUG_ON(!list_empty(&rq->queuelist));
		list_del_init(&rq->flush.list);
		blk_flush_restore_request(rq);
205
		if (q->mq_ops)
206
			blk_mq_end_request(rq, error);
207 208
		else
			__blk_end_request_all(rq, error);
209 210 211 212 213 214
		break;

	default:
		BUG();
	}

215
	kicked = blk_kick_flush(q, fq);
216
	return kicked | queued;
217 218
}

219
static void flush_end_io(struct request *flush_rq, blk_status_t error)
220
{
221
	struct request_queue *q = flush_rq->q;
222
	struct list_head *running;
223 224
	bool queued = false;
	struct request *rq, *n;
225
	unsigned long flags = 0;
226
	struct blk_flush_queue *fq = blk_get_flush_queue(q, flush_rq->mq_ctx);
227

228
	if (q->mq_ops) {
229 230 231
		struct blk_mq_hw_ctx *hctx;

		/* release the tag's ownership to the req cloned from */
232
		spin_lock_irqsave(&fq->mq_flush_lock, flags);
233
		hctx = blk_mq_map_queue(q, flush_rq->mq_ctx->cpu);
234 235 236 237 238 239 240
		if (!q->elevator) {
			blk_mq_tag_set_rq(hctx, flush_rq->tag, fq->orig_rq);
			flush_rq->tag = -1;
		} else {
			blk_mq_put_driver_tag_hctx(hctx, flush_rq);
			flush_rq->internal_tag = -1;
		}
241
	}
242

243 244
	running = &fq->flush_queue[fq->flush_running_idx];
	BUG_ON(fq->flush_pending_idx == fq->flush_running_idx);
245 246

	/* account completion of the flush request */
247
	fq->flush_running_idx ^= 1;
248 249 250

	if (!q->mq_ops)
		elv_completed_request(q, flush_rq);
251 252 253 254 255 256

	/* and push the waiting requests to the next stage */
	list_for_each_entry_safe(rq, n, running, flush.list) {
		unsigned int seq = blk_flush_cur_seq(rq);

		BUG_ON(seq != REQ_FSEQ_PREFLUSH && seq != REQ_FSEQ_POSTFLUSH);
257
		queued |= blk_flush_complete_seq(rq, fq, seq, error);
258 259
	}

260
	/*
261 262 263 264 265 266 267 268 269
	 * Kick the queue to avoid stall for two cases:
	 * 1. Moving a request silently to empty queue_head may stall the
	 * queue.
	 * 2. When flush request is running in non-queueable queue, the
	 * queue is hold. Restart the queue after flush request is finished
	 * to avoid stall.
	 * This function is called from request completion path and calling
	 * directly into request_fn may confuse the driver.  Always use
	 * kblockd.
270
	 */
271
	if (queued || fq->flush_queue_delayed) {
272 273
		WARN_ON(q->mq_ops);
		blk_run_queue_async(q);
274
	}
275
	fq->flush_queue_delayed = 0;
276
	if (q->mq_ops)
277
		spin_unlock_irqrestore(&fq->mq_flush_lock, flags);
278 279
}

280 281 282
/**
 * blk_kick_flush - consider issuing flush request
 * @q: request_queue being kicked
283
 * @fq: flush queue
284 285 286 287 288
 *
 * Flush related states of @q have changed, consider issuing flush request.
 * Please read the comment at the top of this file for more info.
 *
 * CONTEXT:
289
 * spin_lock_irq(q->queue_lock or fq->mq_flush_lock)
290 291 292 293
 *
 * RETURNS:
 * %true if flush was issued, %false otherwise.
 */
294
static bool blk_kick_flush(struct request_queue *q, struct blk_flush_queue *fq)
295
{
296
	struct list_head *pending = &fq->flush_queue[fq->flush_pending_idx];
297 298
	struct request *first_rq =
		list_first_entry(pending, struct request, flush.list);
299
	struct request *flush_rq = fq->flush_rq;
300 301

	/* C1 described at the top of this file */
302
	if (fq->flush_pending_idx != fq->flush_running_idx || list_empty(pending))
303 304
		return false;

305 306 307 308 309 310
	/* C2 and C3
	 *
	 * For blk-mq + scheduling, we can risk having all driver tags
	 * assigned to empty flushes, and we deadlock if we are expecting
	 * other requests to make progress. Don't defer for that case.
	 */
311
	if (!list_empty(&fq->flush_data_in_flight) &&
312
	    !(q->mq_ops && q->elevator) &&
313
	    time_before(jiffies,
314
			fq->flush_pending_since + FLUSH_PENDING_TIMEOUT))
315 316 317 318 319 320
		return false;

	/*
	 * Issue flush and toggle pending_idx.  This makes pending_idx
	 * different from running_idx, which means flush is in flight.
	 */
321
	fq->flush_pending_idx ^= 1;
322

323
	blk_rq_init(q, flush_rq);
324 325

	/*
326 327 328 329 330 331
	 * In case of none scheduler, borrow tag from the first request
	 * since they can't be in flight at the same time. And acquire
	 * the tag's ownership for flush req.
	 *
	 * In case of IO scheduler, flush rq need to borrow scheduler tag
	 * just for cheating put/get driver tag.
332 333
	 */
	if (q->mq_ops) {
334 335
		struct blk_mq_hw_ctx *hctx;

336
		flush_rq->mq_ctx = first_rq->mq_ctx;
337

338 339 340 341 342 343 344 345
		if (!q->elevator) {
			fq->orig_rq = first_rq;
			flush_rq->tag = first_rq->tag;
			hctx = blk_mq_map_queue(q, first_rq->mq_ctx->cpu);
			blk_mq_tag_set_rq(hctx, first_rq->tag, flush_rq);
		} else {
			flush_rq->internal_tag = first_rq->internal_tag;
		}
346
	}
347

348
	flush_rq->cmd_flags = REQ_OP_FLUSH | REQ_PREFLUSH;
349
	flush_rq->rq_flags |= RQF_FLUSH_SEQ;
350 351
	flush_rq->rq_disk = first_rq->rq_disk;
	flush_rq->end_io = flush_end_io;
352

353
	return blk_flush_queue_rq(flush_rq, false);
354 355
}

356
static void flush_data_end_io(struct request *rq, blk_status_t error)
357
{
358
	struct request_queue *q = rq->q;
359
	struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL);
360

361 362
	lockdep_assert_held(q->queue_lock);

363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388
	/*
	 * Updating q->in_flight[] here for making this tag usable
	 * early. Because in blk_queue_start_tag(),
	 * q->in_flight[BLK_RW_ASYNC] is used to limit async I/O and
	 * reserve tags for sync I/O.
	 *
	 * More importantly this way can avoid the following I/O
	 * deadlock:
	 *
	 * - suppose there are 40 fua requests comming to flush queue
	 *   and queue depth is 31
	 * - 30 rqs are scheduled then blk_queue_start_tag() can't alloc
	 *   tag for async I/O any more
	 * - all the 30 rqs are completed before FLUSH_PENDING_TIMEOUT
	 *   and flush_data_end_io() is called
	 * - the other rqs still can't go ahead if not updating
	 *   q->in_flight[BLK_RW_ASYNC] here, meantime these rqs
	 *   are held in flush data queue and make no progress of
	 *   handling post flush rq
	 * - only after the post flush rq is handled, all these rqs
	 *   can be completed
	 */

	elv_completed_request(q, rq);

	/* for avoiding double accounting */
389
	rq->rq_flags &= ~RQF_STARTED;
390

391 392 393 394
	/*
	 * After populating an empty queue, kick it to avoid stall.  Read
	 * the comment in flush_end_io().
	 */
395
	if (blk_flush_complete_seq(rq, fq, REQ_FSEQ_DATA, error))
396
		blk_run_queue_async(q);
397 398
}

399
static void mq_flush_data_end_io(struct request *rq, blk_status_t error)
400 401 402
{
	struct request_queue *q = rq->q;
	struct blk_mq_hw_ctx *hctx;
403
	struct blk_mq_ctx *ctx = rq->mq_ctx;
404
	unsigned long flags;
405
	struct blk_flush_queue *fq = blk_get_flush_queue(q, ctx);
406

407
	hctx = blk_mq_map_queue(q, ctx->cpu);
408

409 410 411 412 413
	if (q->elevator) {
		WARN_ON(rq->tag < 0);
		blk_mq_put_driver_tag_hctx(hctx, rq);
	}

414 415 416 417
	/*
	 * After populating an empty queue, kick it to avoid stall.  Read
	 * the comment in flush_end_io().
	 */
418
	spin_lock_irqsave(&fq->mq_flush_lock, flags);
419
	blk_flush_complete_seq(rq, fq, REQ_FSEQ_DATA, error);
420
	spin_unlock_irqrestore(&fq->mq_flush_lock, flags);
421 422

	blk_mq_run_hw_queue(hctx, true);
423 424
}

425
/**
426
 * blk_insert_flush - insert a new PREFLUSH/FUA request
427 428
 * @rq: request to insert
 *
429
 * To be called from __elv_add_request() for %ELEVATOR_INSERT_FLUSH insertions.
430
 * or __blk_mq_run_hw_queue() to dispatch request.
431 432 433 434
 * @rq is being submitted.  Analyze what needs to be done and put it on the
 * right queue.
 */
void blk_insert_flush(struct request *rq)
435
{
436
	struct request_queue *q = rq->q;
Jens Axboe's avatar
Jens Axboe committed
437
	unsigned long fflags = q->queue_flags;	/* may change, cache */
438
	unsigned int policy = blk_flush_policy(fflags, rq);
439
	struct blk_flush_queue *fq = blk_get_flush_queue(q, rq->mq_ctx);
440

441 442 443
	if (!q->mq_ops)
		lockdep_assert_held(q->queue_lock);

444 445
	/*
	 * @policy now records what operations need to be done.  Adjust
446
	 * REQ_PREFLUSH and FUA for the driver.
447
	 */
448
	rq->cmd_flags &= ~REQ_PREFLUSH;
Jens Axboe's avatar
Jens Axboe committed
449
	if (!(fflags & (1UL << QUEUE_FLAG_FUA)))
450 451
		rq->cmd_flags &= ~REQ_FUA;

452 453 454 455 456 457 458
	/*
	 * REQ_PREFLUSH|REQ_FUA implies REQ_SYNC, so if we clear any
	 * of those flags, we have to set REQ_SYNC to avoid skewing
	 * the request accounting.
	 */
	rq->cmd_flags |= REQ_SYNC;

459 460 461 462 463 464 465
	/*
	 * An empty flush handed down from a stacking driver may
	 * translate into nothing if the underlying device does not
	 * advertise a write-back cache.  In this case, simply
	 * complete the request.
	 */
	if (!policy) {
466
		if (q->mq_ops)
467
			blk_mq_end_request(rq, 0);
468
		else
469
			__blk_end_request(rq, 0, 0);
470 471 472
		return;
	}

473
	BUG_ON(rq->bio != rq->biotail); /*assumes zero or single bio rq */
474

475 476 477 478 479 480 481
	/*
	 * If there's data but flush is not necessary, the request can be
	 * processed directly without going through flush machinery.  Queue
	 * for normal execution.
	 */
	if ((policy & REQ_FSEQ_DATA) &&
	    !(policy & (REQ_FSEQ_PREFLUSH | REQ_FSEQ_POSTFLUSH))) {
482
		if (q->mq_ops)
483
			blk_mq_request_bypass_insert(rq, false);
484
		else
485
			list_add_tail(&rq->queuelist, &q->queue_head);
486
		return;
487
	}
488

489 490 491 492 493 494
	/*
	 * @rq should go through flush machinery.  Mark it part of flush
	 * sequence and submit for further processing.
	 */
	memset(&rq->flush, 0, sizeof(rq->flush));
	INIT_LIST_HEAD(&rq->flush.list);
495
	rq->rq_flags |= RQF_FLUSH_SEQ;
496
	rq->flush.saved_end_io = rq->end_io; /* Usually NULL */
497 498 499
	if (q->mq_ops) {
		rq->end_io = mq_flush_data_end_io;

500
		spin_lock_irq(&fq->mq_flush_lock);
501
		blk_flush_complete_seq(rq, fq, REQ_FSEQ_ACTIONS & ~policy, 0);
502
		spin_unlock_irq(&fq->mq_flush_lock);
503 504
		return;
	}
505 506
	rq->end_io = flush_data_end_io;

507
	blk_flush_complete_seq(rq, fq, REQ_FSEQ_ACTIONS & ~policy, 0);
508 509 510 511 512
}

/**
 * blkdev_issue_flush - queue a flush
 * @bdev:	blockdev to issue flush for
513
 * @gfp_mask:	memory allocation flags (for bio_alloc)
514 515 516 517 518
 * @error_sector:	error sector
 *
 * Description:
 *    Issue a flush for the block device in question. Caller can supply
 *    room for storing the error offset in case of a flush error, if they
519
 *    wish to.
520
 */
521
int blkdev_issue_flush(struct block_device *bdev, gfp_t gfp_mask,
522
		sector_t *error_sector)
523 524 525
{
	struct request_queue *q;
	struct bio *bio;
526
	int ret = 0;
527 528 529 530 531 532 533 534

	if (bdev->bd_disk == NULL)
		return -ENXIO;

	q = bdev_get_queue(bdev);
	if (!q)
		return -ENXIO;

535 536 537 538
	/*
	 * some block devices may not have their queue correctly set up here
	 * (e.g. loop device without a backing file) and so issuing a flush
	 * here will panic. Ensure there is a request function before issuing
539
	 * the flush.
540 541 542 543
	 */
	if (!q->make_request_fn)
		return -ENXIO;

544
	bio = bio_alloc(gfp_mask, 0);
545
	bio_set_dev(bio, bdev);
546
	bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
547

548
	ret = submit_bio_wait(bio);
549 550 551 552 553 554 555

	/*
	 * The driver must store the error location in ->bi_sector, if
	 * it supports it. For non-stacked drivers, this should be
	 * copied from blk_rq_pos(rq).
	 */
	if (error_sector)
556
		*error_sector = bio->bi_iter.bi_sector;
557 558 559 560 561

	bio_put(bio);
	return ret;
}
EXPORT_SYMBOL(blkdev_issue_flush);
562

563 564
struct blk_flush_queue *blk_alloc_flush_queue(struct request_queue *q,
		int node, int cmd_size)
565
{
566 567
	struct blk_flush_queue *fq;
	int rq_sz = sizeof(struct request);
568

569
	fq = kzalloc_node(sizeof(*fq), GFP_KERNEL, node);
570 571
	if (!fq)
		goto fail;
572

573
	if (q->mq_ops)
574 575
		spin_lock_init(&fq->mq_flush_lock);

576
	rq_sz = round_up(rq_sz + cmd_size, cache_line_size());
577
	fq->flush_rq = kzalloc_node(rq_sz, GFP_KERNEL, node);
578 579 580 581 582 583 584 585 586 587 588 589 590
	if (!fq->flush_rq)
		goto fail_rq;

	INIT_LIST_HEAD(&fq->flush_queue[0]);
	INIT_LIST_HEAD(&fq->flush_queue[1]);
	INIT_LIST_HEAD(&fq->flush_data_in_flight);

	return fq;

 fail_rq:
	kfree(fq);
 fail:
	return NULL;
591
}
592

593
void blk_free_flush_queue(struct blk_flush_queue *fq)
594
{
595 596 597
	/* bio based request queue hasn't flush queue */
	if (!fq)
		return;
598

599 600 601
	kfree(fq->flush_rq);
	kfree(fq);
}