blk-flush.c 14.5 KB
Newer Older
1
// SPDX-License-Identifier: GPL-2.0
2
/*
3
 * Functions to sequence PREFLUSH and FUA writes.
4
5
6
7
 *
 * Copyright (C) 2011		Max Planck Institute for Gravitational Physics
 * Copyright (C) 2011		Tejun Heo <tj@kernel.org>
 *
8
 * REQ_{PREFLUSH|FUA} requests are decomposed to sequences consisted of three
9
10
11
 * optional steps - PREFLUSH, DATA and POSTFLUSH - according to the request
 * properties and hardware capability.
 *
12
13
 * 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
14
15
16
17
 * 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.
 *
18
19
20
 * 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.
21
 *
22
 * If the device has writeback cache and supports FUA, REQ_PREFLUSH is
23
24
 * translated to PREFLUSH but REQ_FUA is passed down directly with DATA.
 *
25
26
 * If the device has writeback cache and doesn't support FUA, REQ_PREFLUSH
 * is translated to PREFLUSH and REQ_FUA to POSTFLUSH.
27
28
29
 *
 * The actual execution of flush is double buffered.  Whenever a request
 * needs to execute PRE or POSTFLUSH, it queues at
30
 * fq->flush_queue[fq->flush_pending_idx].  Once certain criteria are met, a
31
 * REQ_OP_FLUSH is issued and the pending_idx is toggled.  When the flush
32
 * completes, all the requests which were pending are proceeded to the next
33
 * step.  This allows arbitrary merging of different types of PREFLUSH/FUA
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
 * 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
49
 *     FUA (without PREFLUSH) requests.
50
51
52
53
 *
 * For devices which support FUA, it isn't clear whether C2 (and thus C3)
 * is beneficial.
 *
54
 * Note that a sequenced PREFLUSH/FUA request with DATA is completed twice.
55
56
57
 * 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
58
 * sequence is complete.  This is implemented by testing RQF_FLUSH_SEQ in
59
60
 * req_bio_endio().
 *
61
 * The above peculiarity requires that each PREFLUSH/FUA request has only one
62
63
 * bio attached to it, which is guaranteed as they aren't allowed to be
 * merged in the usual way.
64
 */
65

66
67
68
69
#include <linux/kernel.h>
#include <linux/module.h>
#include <linux/bio.h>
#include <linux/blkdev.h>
70
#include <linux/gfp.h>
71
#include <linux/blk-mq.h>
72
73

#include "blk.h"
74
#include "blk-mq.h"
75
#include "blk-mq-tag.h"
76
#include "blk-mq-sched.h"
77

78
/* PREFLUSH/FUA sequences */
79
enum {
80
81
82
83
84
85
86
87
88
89
90
91
92
	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,
93
94
};

95
static void blk_kick_flush(struct request_queue *q,
96
			   struct blk_flush_queue *fq, unsigned int flags);
97

Jens Axboe's avatar
Jens Axboe committed
98
static unsigned int blk_flush_policy(unsigned long fflags, struct request *rq)
99
{
100
	unsigned int policy = 0;
101

102
103
104
	if (blk_rq_sectors(rq))
		policy |= REQ_FSEQ_DATA;

Jens Axboe's avatar
Jens Axboe committed
105
	if (fflags & (1UL << QUEUE_FLAG_WC)) {
106
		if (rq->cmd_flags & REQ_PREFLUSH)
107
			policy |= REQ_FSEQ_PREFLUSH;
Jens Axboe's avatar
Jens Axboe committed
108
109
		if (!(fflags & (1UL << QUEUE_FLAG_FUA)) &&
		    (rq->cmd_flags & REQ_FUA))
110
			policy |= REQ_FSEQ_POSTFLUSH;
111
	}
112
	return policy;
113
114
}

115
static unsigned int blk_flush_cur_seq(struct request *rq)
116
{
117
118
	return 1 << ffz(rq->flush.seq);
}
119

120
121
static void blk_flush_restore_request(struct request *rq)
{
122
	/*
123
124
125
	 * 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.
126
	 */
127
128
129
	rq->bio = rq->biotail;

	/* make @rq a normal request */
130
	rq->rq_flags &= ~RQF_FLUSH_SEQ;
131
	rq->end_io = rq->flush.saved_end_io;
132
133
}

134
static void blk_flush_queue_rq(struct request *rq, bool add_front)
135
{
136
	blk_mq_add_to_requeue_list(rq, add_front, true);
137
138
}

139
140
/**
 * blk_flush_complete_seq - complete flush sequence
141
 * @rq: PREFLUSH/FUA request being sequenced
142
 * @fq: flush queue
143
144
145
146
147
148
149
 * @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:
150
 * spin_lock_irq(fq->mq_flush_lock)
151
152
153
154
 *
 * RETURNS:
 * %true if requests were added to the dispatch queue, %false otherwise.
 */
155
static void blk_flush_complete_seq(struct request *rq,
156
				   struct blk_flush_queue *fq,
157
				   unsigned int seq, blk_status_t error)
158
{
159
	struct request_queue *q = rq->q;
160
	struct list_head *pending = &fq->flush_queue[fq->flush_pending_idx];
161
	unsigned int cmd_flags;
162
163
164

	BUG_ON(rq->flush.seq & seq);
	rq->flush.seq |= seq;
165
	cmd_flags = rq->cmd_flags;
166
167
168
169
170
171
172
173
174
175
176

	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))
177
			fq->flush_pending_since = jiffies;
178
179
180
181
		list_move_tail(&rq->flush.list, pending);
		break;

	case REQ_FSEQ_DATA:
182
		list_move_tail(&rq->flush.list, &fq->flush_data_in_flight);
183
		blk_flush_queue_rq(rq, true);
184
185
186
187
188
189
190
191
192
193
194
195
		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);
196
		blk_mq_end_request(rq, error);
197
198
199
200
201
202
		break;

	default:
		BUG();
	}

203
	blk_kick_flush(q, fq, cmd_flags);
204
205
}

206
static void flush_end_io(struct request *flush_rq, blk_status_t error)
207
{
208
	struct request_queue *q = flush_rq->q;
209
	struct list_head *running;
210
	struct request *rq, *n;
211
	unsigned long flags = 0;
212
	struct blk_flush_queue *fq = blk_get_flush_queue(q, flush_rq->mq_ctx);
213
	struct blk_mq_hw_ctx *hctx;
214

215
216
	/* release the tag's ownership to the req cloned from */
	spin_lock_irqsave(&fq->mq_flush_lock, flags);
217
218
219
220
221
222
223
224
225
226

	if (!refcount_dec_and_test(&flush_rq->ref)) {
		fq->rq_status = error;
		spin_unlock_irqrestore(&fq->mq_flush_lock, flags);
		return;
	}

	if (fq->rq_status != BLK_STS_OK)
		error = fq->rq_status;

227
	hctx = flush_rq->mq_hctx;
228
229
230
231
	if (!q->elevator) {
		blk_mq_tag_set_rq(hctx, flush_rq->tag, fq->orig_rq);
		flush_rq->tag = -1;
	} else {
232
		blk_mq_put_driver_tag(flush_rq);
233
		flush_rq->internal_tag = -1;
234
	}
235

236
237
	running = &fq->flush_queue[fq->flush_running_idx];
	BUG_ON(fq->flush_pending_idx == fq->flush_running_idx);
238
239

	/* account completion of the flush request */
240
	fq->flush_running_idx ^= 1;
241

242
243
244
245
246
	/* 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);
247
		blk_flush_complete_seq(rq, fq, seq, error);
248
249
	}

250
	fq->flush_queue_delayed = 0;
251
	spin_unlock_irqrestore(&fq->mq_flush_lock, flags);
252
253
}

254
255
256
/**
 * blk_kick_flush - consider issuing flush request
 * @q: request_queue being kicked
257
 * @fq: flush queue
258
 * @flags: cmd_flags of the original request
259
260
261
262
263
 *
 * 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:
264
 * spin_lock_irq(fq->mq_flush_lock)
265
266
 *
 */
267
static void blk_kick_flush(struct request_queue *q, struct blk_flush_queue *fq,
268
			   unsigned int flags)
269
{
270
	struct list_head *pending = &fq->flush_queue[fq->flush_pending_idx];
271
272
	struct request *first_rq =
		list_first_entry(pending, struct request, flush.list);
273
	struct request *flush_rq = fq->flush_rq;
274
275

	/* C1 described at the top of this file */
276
	if (fq->flush_pending_idx != fq->flush_running_idx || list_empty(pending))
277
		return;
278

279
280
281
282
283
284
	/* 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.
	 */
Jens Axboe's avatar
Jens Axboe committed
285
	if (!list_empty(&fq->flush_data_in_flight) && q->elevator &&
286
	    time_before(jiffies,
287
			fq->flush_pending_since + FLUSH_PENDING_TIMEOUT))
288
		return;
289
290
291
292
293

	/*
	 * Issue flush and toggle pending_idx.  This makes pending_idx
	 * different from running_idx, which means flush is in flight.
	 */
294
	fq->flush_pending_idx ^= 1;
295

296
	blk_rq_init(q, flush_rq);
297
298

	/*
299
300
301
302
303
304
	 * 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.
305
	 */
306
	flush_rq->mq_ctx = first_rq->mq_ctx;
307
	flush_rq->mq_hctx = first_rq->mq_hctx;
308
309
310
311

	if (!q->elevator) {
		fq->orig_rq = first_rq;
		flush_rq->tag = first_rq->tag;
312
		blk_mq_tag_set_rq(flush_rq->mq_hctx, first_rq->tag, flush_rq);
313
314
	} else {
		flush_rq->internal_tag = first_rq->internal_tag;
315
	}
316

317
	flush_rq->cmd_flags = REQ_OP_FLUSH | REQ_PREFLUSH;
318
	flush_rq->cmd_flags |= (flags & REQ_DRV) | (flags & REQ_FAILFAST_MASK);
319
	flush_rq->rq_flags |= RQF_FLUSH_SEQ;
320
321
	flush_rq->rq_disk = first_rq->rq_disk;
	flush_rq->end_io = flush_end_io;
322

323
	blk_flush_queue_rq(flush_rq, false);
324
325
}

326
static void mq_flush_data_end_io(struct request *rq, blk_status_t error)
327
328
{
	struct request_queue *q = rq->q;
329
	struct blk_mq_hw_ctx *hctx = rq->mq_hctx;
330
	struct blk_mq_ctx *ctx = rq->mq_ctx;
331
	unsigned long flags;
332
	struct blk_flush_queue *fq = blk_get_flush_queue(q, ctx);
333

334
335
	if (q->elevator) {
		WARN_ON(rq->tag < 0);
336
		blk_mq_put_driver_tag(rq);
337
338
	}

339
340
341
342
	/*
	 * After populating an empty queue, kick it to avoid stall.  Read
	 * the comment in flush_end_io().
	 */
343
	spin_lock_irqsave(&fq->mq_flush_lock, flags);
344
	blk_flush_complete_seq(rq, fq, REQ_FSEQ_DATA, error);
345
	spin_unlock_irqrestore(&fq->mq_flush_lock, flags);
346

347
	blk_mq_sched_restart(hctx);
348
349
}

350
/**
351
 * blk_insert_flush - insert a new PREFLUSH/FUA request
352
353
 * @rq: request to insert
 *
354
 * To be called from __elv_add_request() for %ELEVATOR_INSERT_FLUSH insertions.
355
 * or __blk_mq_run_hw_queue() to dispatch request.
356
357
358
359
 * @rq is being submitted.  Analyze what needs to be done and put it on the
 * right queue.
 */
void blk_insert_flush(struct request *rq)
360
{
361
	struct request_queue *q = rq->q;
Jens Axboe's avatar
Jens Axboe committed
362
	unsigned long fflags = q->queue_flags;	/* may change, cache */
363
	unsigned int policy = blk_flush_policy(fflags, rq);
364
	struct blk_flush_queue *fq = blk_get_flush_queue(q, rq->mq_ctx);
365

366
367
	/*
	 * @policy now records what operations need to be done.  Adjust
368
	 * REQ_PREFLUSH and FUA for the driver.
369
	 */
370
	rq->cmd_flags &= ~REQ_PREFLUSH;
Jens Axboe's avatar
Jens Axboe committed
371
	if (!(fflags & (1UL << QUEUE_FLAG_FUA)))
372
373
		rq->cmd_flags &= ~REQ_FUA;

374
375
376
377
378
379
380
	/*
	 * 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;

381
382
383
384
385
386
387
	/*
	 * 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) {
388
		blk_mq_end_request(rq, 0);
389
390
391
		return;
	}

392
	BUG_ON(rq->bio != rq->biotail); /*assumes zero or single bio rq */
393

394
395
396
397
398
399
400
	/*
	 * 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))) {
401
		blk_mq_request_bypass_insert(rq, false);
402
		return;
403
	}
404

405
406
407
408
409
410
	/*
	 * @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);
411
	rq->rq_flags |= RQF_FLUSH_SEQ;
412
	rq->flush.saved_end_io = rq->end_io; /* Usually NULL */
413

414
	rq->end_io = mq_flush_data_end_io;
415

416
	spin_lock_irq(&fq->mq_flush_lock);
417
	blk_flush_complete_seq(rq, fq, REQ_FSEQ_ACTIONS & ~policy, 0);
418
	spin_unlock_irq(&fq->mq_flush_lock);
419
420
421
422
423
}

/**
 * blkdev_issue_flush - queue a flush
 * @bdev:	blockdev to issue flush for
424
 * @gfp_mask:	memory allocation flags (for bio_alloc)
425
426
427
428
429
 * @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
430
 *    wish to.
431
 */
432
int blkdev_issue_flush(struct block_device *bdev, gfp_t gfp_mask,
433
		sector_t *error_sector)
434
435
436
{
	struct request_queue *q;
	struct bio *bio;
437
	int ret = 0;
438
439
440
441
442
443
444
445

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

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

446
447
448
449
	/*
	 * 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
450
	 * the flush.
451
452
453
454
	 */
	if (!q->make_request_fn)
		return -ENXIO;

455
	bio = bio_alloc(gfp_mask, 0);
456
	bio_set_dev(bio, bdev);
457
	bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH;
458

459
	ret = submit_bio_wait(bio);
460
461
462
463
464
465
466

	/*
	 * 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)
467
		*error_sector = bio->bi_iter.bi_sector;
468
469
470
471
472

	bio_put(bio);
	return ret;
}
EXPORT_SYMBOL(blkdev_issue_flush);
473

474
struct blk_flush_queue *blk_alloc_flush_queue(struct request_queue *q,
475
		int node, int cmd_size, gfp_t flags)
476
{
477
478
	struct blk_flush_queue *fq;
	int rq_sz = sizeof(struct request);
479

480
	fq = kzalloc_node(sizeof(*fq), flags, node);
481
482
	if (!fq)
		goto fail;
483

484
	spin_lock_init(&fq->mq_flush_lock);
485

486
	rq_sz = round_up(rq_sz + cmd_size, cache_line_size());
487
	fq->flush_rq = kzalloc_node(rq_sz, flags, node);
488
489
490
491
492
493
494
495
496
497
498
499
500
	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;
501
}
502

503
void blk_free_flush_queue(struct blk_flush_queue *fq)
504
{
505
506
507
	/* bio based request queue hasn't flush queue */
	if (!fq)
		return;
508

509
510
511
	kfree(fq->flush_rq);
	kfree(fq);
}