]>
Commit | Line | Data |
---|---|---|
86db1e29 | 1 | /* |
4fed947c | 2 | * Functions to sequence FLUSH and FUA writes. |
ae1b1539 TH |
3 | * |
4 | * Copyright (C) 2011 Max Planck Institute for Gravitational Physics | |
5 | * Copyright (C) 2011 Tejun Heo <tj@kernel.org> | |
6 | * | |
7 | * This file is released under the GPLv2. | |
8 | * | |
9 | * REQ_{FLUSH|FUA} requests are decomposed to sequences consisted of three | |
10 | * optional steps - PREFLUSH, DATA and POSTFLUSH - according to the request | |
11 | * properties and hardware capability. | |
12 | * | |
28a8f0d3 MC |
13 | * If a request doesn't have data, only REQ_PREFLUSH makes sense, which |
14 | * indicates a simple flush request. If there is data, REQ_PREFLUSH indicates | |
ae1b1539 TH |
15 | * that the device cache should be flushed before the data is executed, and |
16 | * REQ_FUA means that the data must be on non-volatile media on request | |
17 | * completion. | |
18 | * | |
19 | * If the device doesn't have writeback cache, FLUSH and FUA don't make any | |
20 | * difference. The requests are either completed immediately if there's no | |
21 | * data or executed as normal requests otherwise. | |
22 | * | |
28a8f0d3 | 23 | * If the device has writeback cache and supports FUA, REQ_PREFLUSH is |
ae1b1539 TH |
24 | * translated to PREFLUSH but REQ_FUA is passed down directly with DATA. |
25 | * | |
28a8f0d3 MC |
26 | * If the device has writeback cache and doesn't support FUA, REQ_PREFLUSH |
27 | * is translated to PREFLUSH and REQ_FUA to POSTFLUSH. | |
ae1b1539 TH |
28 | * |
29 | * The actual execution of flush is double buffered. Whenever a request | |
30 | * needs to execute PRE or POSTFLUSH, it queues at | |
7c94e1c1 | 31 | * fq->flush_queue[fq->flush_pending_idx]. Once certain criteria are met, a |
3a5e02ce | 32 | * REQ_OP_FLUSH is issued and the pending_idx is toggled. When the flush |
ae1b1539 TH |
33 | * completes, all the requests which were pending are proceeded to the next |
34 | * step. This allows arbitrary merging of different types of FLUSH/FUA | |
35 | * requests. | |
36 | * | |
37 | * Currently, the following conditions are used to determine when to issue | |
38 | * flush. | |
39 | * | |
40 | * C1. At any given time, only one flush shall be in progress. This makes | |
41 | * double buffering sufficient. | |
42 | * | |
43 | * C2. Flush is deferred if any request is executing DATA of its sequence. | |
44 | * This avoids issuing separate POSTFLUSHes for requests which shared | |
45 | * PREFLUSH. | |
46 | * | |
47 | * C3. The second condition is ignored if there is a request which has | |
48 | * waited longer than FLUSH_PENDING_TIMEOUT. This is to avoid | |
49 | * starvation in the unlikely case where there are continuous stream of | |
50 | * FUA (without FLUSH) requests. | |
51 | * | |
52 | * For devices which support FUA, it isn't clear whether C2 (and thus C3) | |
53 | * is beneficial. | |
54 | * | |
55 | * Note that a sequenced FLUSH/FUA request with DATA is completed twice. | |
56 | * Once while executing DATA and again after the whole sequence is | |
57 | * complete. The first completion updates the contained bio but doesn't | |
58 | * finish it so that the bio submitter is notified only after the whole | |
e8064021 | 59 | * sequence is complete. This is implemented by testing RQF_FLUSH_SEQ in |
ae1b1539 TH |
60 | * req_bio_endio(). |
61 | * | |
62 | * The above peculiarity requires that each FLUSH/FUA request has only one | |
63 | * bio attached to it, which is guaranteed as they aren't allowed to be | |
64 | * merged in the usual way. | |
86db1e29 | 65 | */ |
ae1b1539 | 66 | |
86db1e29 JA |
67 | #include <linux/kernel.h> |
68 | #include <linux/module.h> | |
69 | #include <linux/bio.h> | |
70 | #include <linux/blkdev.h> | |
5a0e3ad6 | 71 | #include <linux/gfp.h> |
320ae51f | 72 | #include <linux/blk-mq.h> |
86db1e29 JA |
73 | |
74 | #include "blk.h" | |
320ae51f | 75 | #include "blk-mq.h" |
0048b483 | 76 | #include "blk-mq-tag.h" |
bd166ef1 | 77 | #include "blk-mq-sched.h" |
86db1e29 | 78 | |
4fed947c TH |
79 | /* FLUSH/FUA sequences */ |
80 | enum { | |
ae1b1539 TH |
81 | REQ_FSEQ_PREFLUSH = (1 << 0), /* pre-flushing in progress */ |
82 | REQ_FSEQ_DATA = (1 << 1), /* data write in progress */ | |
83 | REQ_FSEQ_POSTFLUSH = (1 << 2), /* post-flushing in progress */ | |
84 | REQ_FSEQ_DONE = (1 << 3), | |
85 | ||
86 | REQ_FSEQ_ACTIONS = REQ_FSEQ_PREFLUSH | REQ_FSEQ_DATA | | |
87 | REQ_FSEQ_POSTFLUSH, | |
88 | ||
89 | /* | |
90 | * If flush has been pending longer than the following timeout, | |
91 | * it's issued even if flush_data requests are still in flight. | |
92 | */ | |
93 | FLUSH_PENDING_TIMEOUT = 5 * HZ, | |
4fed947c TH |
94 | }; |
95 | ||
0bae352d ML |
96 | static bool blk_kick_flush(struct request_queue *q, |
97 | struct blk_flush_queue *fq); | |
28e7d184 | 98 | |
c888a8f9 | 99 | static unsigned int blk_flush_policy(unsigned long fflags, struct request *rq) |
86db1e29 | 100 | { |
ae1b1539 | 101 | unsigned int policy = 0; |
86db1e29 | 102 | |
fa1bf42f JM |
103 | if (blk_rq_sectors(rq)) |
104 | policy |= REQ_FSEQ_DATA; | |
105 | ||
c888a8f9 | 106 | if (fflags & (1UL << QUEUE_FLAG_WC)) { |
28a8f0d3 | 107 | if (rq->cmd_flags & REQ_PREFLUSH) |
ae1b1539 | 108 | policy |= REQ_FSEQ_PREFLUSH; |
c888a8f9 JA |
109 | if (!(fflags & (1UL << QUEUE_FLAG_FUA)) && |
110 | (rq->cmd_flags & REQ_FUA)) | |
ae1b1539 | 111 | policy |= REQ_FSEQ_POSTFLUSH; |
28e7d184 | 112 | } |
ae1b1539 | 113 | return policy; |
86db1e29 JA |
114 | } |
115 | ||
ae1b1539 | 116 | static unsigned int blk_flush_cur_seq(struct request *rq) |
47f70d5a | 117 | { |
ae1b1539 TH |
118 | return 1 << ffz(rq->flush.seq); |
119 | } | |
47f70d5a | 120 | |
ae1b1539 TH |
121 | static void blk_flush_restore_request(struct request *rq) |
122 | { | |
47f70d5a | 123 | /* |
ae1b1539 TH |
124 | * After flush data completion, @rq->bio is %NULL but we need to |
125 | * complete the bio again. @rq->biotail is guaranteed to equal the | |
126 | * original @rq->bio. Restore it. | |
47f70d5a | 127 | */ |
ae1b1539 TH |
128 | rq->bio = rq->biotail; |
129 | ||
130 | /* make @rq a normal request */ | |
e8064021 | 131 | rq->rq_flags &= ~RQF_FLUSH_SEQ; |
4853abaa | 132 | rq->end_io = rq->flush.saved_end_io; |
320ae51f JA |
133 | } |
134 | ||
10beafc1 | 135 | static bool blk_flush_queue_rq(struct request *rq, bool add_front) |
320ae51f | 136 | { |
18741986 | 137 | if (rq->q->mq_ops) { |
2b053aca | 138 | blk_mq_add_to_requeue_list(rq, add_front, true); |
18741986 CH |
139 | return false; |
140 | } else { | |
10beafc1 MS |
141 | if (add_front) |
142 | list_add(&rq->queuelist, &rq->q->queue_head); | |
143 | else | |
144 | list_add_tail(&rq->queuelist, &rq->q->queue_head); | |
18741986 CH |
145 | return true; |
146 | } | |
47f70d5a TH |
147 | } |
148 | ||
ae1b1539 TH |
149 | /** |
150 | * blk_flush_complete_seq - complete flush sequence | |
151 | * @rq: FLUSH/FUA request being sequenced | |
0bae352d | 152 | * @fq: flush queue |
ae1b1539 TH |
153 | * @seq: sequences to complete (mask of %REQ_FSEQ_*, can be zero) |
154 | * @error: whether an error occurred | |
155 | * | |
156 | * @rq just completed @seq part of its flush sequence, record the | |
157 | * completion and trigger the next step. | |
158 | * | |
159 | * CONTEXT: | |
7c94e1c1 | 160 | * spin_lock_irq(q->queue_lock or fq->mq_flush_lock) |
ae1b1539 TH |
161 | * |
162 | * RETURNS: | |
163 | * %true if requests were added to the dispatch queue, %false otherwise. | |
164 | */ | |
0bae352d ML |
165 | static bool blk_flush_complete_seq(struct request *rq, |
166 | struct blk_flush_queue *fq, | |
2a842aca | 167 | unsigned int seq, blk_status_t error) |
86db1e29 | 168 | { |
ae1b1539 | 169 | struct request_queue *q = rq->q; |
7c94e1c1 | 170 | struct list_head *pending = &fq->flush_queue[fq->flush_pending_idx]; |
320ae51f | 171 | bool queued = false, kicked; |
ae1b1539 TH |
172 | |
173 | BUG_ON(rq->flush.seq & seq); | |
174 | rq->flush.seq |= seq; | |
175 | ||
176 | if (likely(!error)) | |
177 | seq = blk_flush_cur_seq(rq); | |
178 | else | |
179 | seq = REQ_FSEQ_DONE; | |
180 | ||
181 | switch (seq) { | |
182 | case REQ_FSEQ_PREFLUSH: | |
183 | case REQ_FSEQ_POSTFLUSH: | |
184 | /* queue for flush */ | |
185 | if (list_empty(pending)) | |
7c94e1c1 | 186 | fq->flush_pending_since = jiffies; |
ae1b1539 TH |
187 | list_move_tail(&rq->flush.list, pending); |
188 | break; | |
189 | ||
190 | case REQ_FSEQ_DATA: | |
7c94e1c1 | 191 | list_move_tail(&rq->flush.list, &fq->flush_data_in_flight); |
10beafc1 | 192 | queued = blk_flush_queue_rq(rq, true); |
ae1b1539 TH |
193 | break; |
194 | ||
195 | case REQ_FSEQ_DONE: | |
196 | /* | |
197 | * @rq was previously adjusted by blk_flush_issue() for | |
198 | * flush sequencing and may already have gone through the | |
199 | * flush data request completion path. Restore @rq for | |
200 | * normal completion and end it. | |
201 | */ | |
202 | BUG_ON(!list_empty(&rq->queuelist)); | |
203 | list_del_init(&rq->flush.list); | |
204 | blk_flush_restore_request(rq); | |
320ae51f | 205 | if (q->mq_ops) |
c8a446ad | 206 | blk_mq_end_request(rq, error); |
320ae51f JA |
207 | else |
208 | __blk_end_request_all(rq, error); | |
ae1b1539 TH |
209 | break; |
210 | ||
211 | default: | |
212 | BUG(); | |
213 | } | |
214 | ||
0bae352d | 215 | kicked = blk_kick_flush(q, fq); |
320ae51f | 216 | return kicked | queued; |
86db1e29 JA |
217 | } |
218 | ||
2a842aca | 219 | static void flush_end_io(struct request *flush_rq, blk_status_t error) |
86db1e29 | 220 | { |
ae1b1539 | 221 | struct request_queue *q = flush_rq->q; |
320ae51f | 222 | struct list_head *running; |
ae1b1539 TH |
223 | bool queued = false; |
224 | struct request *rq, *n; | |
320ae51f | 225 | unsigned long flags = 0; |
e97c293c | 226 | struct blk_flush_queue *fq = blk_get_flush_queue(q, flush_rq->mq_ctx); |
ae1b1539 | 227 | |
22302375 | 228 | if (q->mq_ops) { |
0048b483 ML |
229 | struct blk_mq_hw_ctx *hctx; |
230 | ||
231 | /* release the tag's ownership to the req cloned from */ | |
7c94e1c1 | 232 | spin_lock_irqsave(&fq->mq_flush_lock, flags); |
7d7e0f90 | 233 | hctx = blk_mq_map_queue(q, flush_rq->mq_ctx->cpu); |
0048b483 | 234 | blk_mq_tag_set_rq(hctx, flush_rq->tag, fq->orig_rq); |
7ddab5de | 235 | flush_rq->tag = -1; |
22302375 | 236 | } |
18741986 | 237 | |
7c94e1c1 ML |
238 | running = &fq->flush_queue[fq->flush_running_idx]; |
239 | BUG_ON(fq->flush_pending_idx == fq->flush_running_idx); | |
ae1b1539 TH |
240 | |
241 | /* account completion of the flush request */ | |
7c94e1c1 | 242 | fq->flush_running_idx ^= 1; |
320ae51f JA |
243 | |
244 | if (!q->mq_ops) | |
245 | elv_completed_request(q, flush_rq); | |
ae1b1539 TH |
246 | |
247 | /* and push the waiting requests to the next stage */ | |
248 | list_for_each_entry_safe(rq, n, running, flush.list) { | |
249 | unsigned int seq = blk_flush_cur_seq(rq); | |
250 | ||
251 | BUG_ON(seq != REQ_FSEQ_PREFLUSH && seq != REQ_FSEQ_POSTFLUSH); | |
0bae352d | 252 | queued |= blk_flush_complete_seq(rq, fq, seq, error); |
ae1b1539 TH |
253 | } |
254 | ||
47f70d5a | 255 | /* |
3ac0cc45 | 256 | * Kick the queue to avoid stall for two cases: |
257 | * 1. Moving a request silently to empty queue_head may stall the | |
258 | * queue. | |
259 | * 2. When flush request is running in non-queueable queue, the | |
260 | * queue is hold. Restart the queue after flush request is finished | |
261 | * to avoid stall. | |
262 | * This function is called from request completion path and calling | |
263 | * directly into request_fn may confuse the driver. Always use | |
264 | * kblockd. | |
47f70d5a | 265 | */ |
7c94e1c1 | 266 | if (queued || fq->flush_queue_delayed) { |
18741986 CH |
267 | WARN_ON(q->mq_ops); |
268 | blk_run_queue_async(q); | |
320ae51f | 269 | } |
7c94e1c1 | 270 | fq->flush_queue_delayed = 0; |
320ae51f | 271 | if (q->mq_ops) |
7c94e1c1 | 272 | spin_unlock_irqrestore(&fq->mq_flush_lock, flags); |
320ae51f JA |
273 | } |
274 | ||
ae1b1539 TH |
275 | /** |
276 | * blk_kick_flush - consider issuing flush request | |
277 | * @q: request_queue being kicked | |
0bae352d | 278 | * @fq: flush queue |
ae1b1539 TH |
279 | * |
280 | * Flush related states of @q have changed, consider issuing flush request. | |
281 | * Please read the comment at the top of this file for more info. | |
282 | * | |
283 | * CONTEXT: | |
7c94e1c1 | 284 | * spin_lock_irq(q->queue_lock or fq->mq_flush_lock) |
ae1b1539 TH |
285 | * |
286 | * RETURNS: | |
287 | * %true if flush was issued, %false otherwise. | |
288 | */ | |
0bae352d | 289 | static bool blk_kick_flush(struct request_queue *q, struct blk_flush_queue *fq) |
86db1e29 | 290 | { |
7c94e1c1 | 291 | struct list_head *pending = &fq->flush_queue[fq->flush_pending_idx]; |
ae1b1539 TH |
292 | struct request *first_rq = |
293 | list_first_entry(pending, struct request, flush.list); | |
7c94e1c1 | 294 | struct request *flush_rq = fq->flush_rq; |
ae1b1539 TH |
295 | |
296 | /* C1 described at the top of this file */ | |
7c94e1c1 | 297 | if (fq->flush_pending_idx != fq->flush_running_idx || list_empty(pending)) |
ae1b1539 TH |
298 | return false; |
299 | ||
7520872c JA |
300 | /* C2 and C3 |
301 | * | |
302 | * For blk-mq + scheduling, we can risk having all driver tags | |
303 | * assigned to empty flushes, and we deadlock if we are expecting | |
304 | * other requests to make progress. Don't defer for that case. | |
305 | */ | |
7c94e1c1 | 306 | if (!list_empty(&fq->flush_data_in_flight) && |
7520872c | 307 | !(q->mq_ops && q->elevator) && |
ae1b1539 | 308 | time_before(jiffies, |
7c94e1c1 | 309 | fq->flush_pending_since + FLUSH_PENDING_TIMEOUT)) |
ae1b1539 TH |
310 | return false; |
311 | ||
312 | /* | |
313 | * Issue flush and toggle pending_idx. This makes pending_idx | |
314 | * different from running_idx, which means flush is in flight. | |
315 | */ | |
7c94e1c1 | 316 | fq->flush_pending_idx ^= 1; |
18741986 | 317 | |
7ddab5de | 318 | blk_rq_init(q, flush_rq); |
f70ced09 ML |
319 | |
320 | /* | |
321 | * Borrow tag from the first request since they can't | |
0048b483 ML |
322 | * be in flight at the same time. And acquire the tag's |
323 | * ownership for flush req. | |
f70ced09 ML |
324 | */ |
325 | if (q->mq_ops) { | |
0048b483 ML |
326 | struct blk_mq_hw_ctx *hctx; |
327 | ||
f70ced09 ML |
328 | flush_rq->mq_ctx = first_rq->mq_ctx; |
329 | flush_rq->tag = first_rq->tag; | |
0048b483 ML |
330 | fq->orig_rq = first_rq; |
331 | ||
7d7e0f90 | 332 | hctx = blk_mq_map_queue(q, first_rq->mq_ctx->cpu); |
0048b483 | 333 | blk_mq_tag_set_rq(hctx, first_rq->tag, flush_rq); |
f70ced09 | 334 | } |
320ae51f | 335 | |
70fd7614 | 336 | flush_rq->cmd_flags = REQ_OP_FLUSH | REQ_PREFLUSH; |
e8064021 | 337 | flush_rq->rq_flags |= RQF_FLUSH_SEQ; |
7ddab5de ML |
338 | flush_rq->rq_disk = first_rq->rq_disk; |
339 | flush_rq->end_io = flush_end_io; | |
ae1b1539 | 340 | |
7ddab5de | 341 | return blk_flush_queue_rq(flush_rq, false); |
86db1e29 JA |
342 | } |
343 | ||
2a842aca | 344 | static void flush_data_end_io(struct request *rq, blk_status_t error) |
86db1e29 | 345 | { |
ae1b1539 | 346 | struct request_queue *q = rq->q; |
e97c293c | 347 | struct blk_flush_queue *fq = blk_get_flush_queue(q, NULL); |
ae1b1539 | 348 | |
94d7dea4 ML |
349 | /* |
350 | * Updating q->in_flight[] here for making this tag usable | |
351 | * early. Because in blk_queue_start_tag(), | |
352 | * q->in_flight[BLK_RW_ASYNC] is used to limit async I/O and | |
353 | * reserve tags for sync I/O. | |
354 | * | |
355 | * More importantly this way can avoid the following I/O | |
356 | * deadlock: | |
357 | * | |
358 | * - suppose there are 40 fua requests comming to flush queue | |
359 | * and queue depth is 31 | |
360 | * - 30 rqs are scheduled then blk_queue_start_tag() can't alloc | |
361 | * tag for async I/O any more | |
362 | * - all the 30 rqs are completed before FLUSH_PENDING_TIMEOUT | |
363 | * and flush_data_end_io() is called | |
364 | * - the other rqs still can't go ahead if not updating | |
365 | * q->in_flight[BLK_RW_ASYNC] here, meantime these rqs | |
366 | * are held in flush data queue and make no progress of | |
367 | * handling post flush rq | |
368 | * - only after the post flush rq is handled, all these rqs | |
369 | * can be completed | |
370 | */ | |
371 | ||
372 | elv_completed_request(q, rq); | |
373 | ||
374 | /* for avoiding double accounting */ | |
36869cb9 | 375 | rq->rq_flags &= ~RQF_STARTED; |
94d7dea4 | 376 | |
e83a46bb TH |
377 | /* |
378 | * After populating an empty queue, kick it to avoid stall. Read | |
379 | * the comment in flush_end_io(). | |
380 | */ | |
0bae352d | 381 | if (blk_flush_complete_seq(rq, fq, REQ_FSEQ_DATA, error)) |
24ecfbe2 | 382 | blk_run_queue_async(q); |
86db1e29 JA |
383 | } |
384 | ||
2a842aca | 385 | static void mq_flush_data_end_io(struct request *rq, blk_status_t error) |
320ae51f JA |
386 | { |
387 | struct request_queue *q = rq->q; | |
388 | struct blk_mq_hw_ctx *hctx; | |
e97c293c | 389 | struct blk_mq_ctx *ctx = rq->mq_ctx; |
320ae51f | 390 | unsigned long flags; |
e97c293c | 391 | struct blk_flush_queue *fq = blk_get_flush_queue(q, ctx); |
320ae51f | 392 | |
7d7e0f90 | 393 | hctx = blk_mq_map_queue(q, ctx->cpu); |
320ae51f JA |
394 | |
395 | /* | |
396 | * After populating an empty queue, kick it to avoid stall. Read | |
397 | * the comment in flush_end_io(). | |
398 | */ | |
7c94e1c1 | 399 | spin_lock_irqsave(&fq->mq_flush_lock, flags); |
bd166ef1 | 400 | blk_flush_complete_seq(rq, fq, REQ_FSEQ_DATA, error); |
7c94e1c1 | 401 | spin_unlock_irqrestore(&fq->mq_flush_lock, flags); |
bd166ef1 JA |
402 | |
403 | blk_mq_run_hw_queue(hctx, true); | |
320ae51f JA |
404 | } |
405 | ||
ae1b1539 TH |
406 | /** |
407 | * blk_insert_flush - insert a new FLUSH/FUA request | |
408 | * @rq: request to insert | |
409 | * | |
b710a480 | 410 | * To be called from __elv_add_request() for %ELEVATOR_INSERT_FLUSH insertions. |
320ae51f | 411 | * or __blk_mq_run_hw_queue() to dispatch request. |
ae1b1539 TH |
412 | * @rq is being submitted. Analyze what needs to be done and put it on the |
413 | * right queue. | |
414 | * | |
415 | * CONTEXT: | |
320ae51f | 416 | * spin_lock_irq(q->queue_lock) in !mq case |
ae1b1539 TH |
417 | */ |
418 | void blk_insert_flush(struct request *rq) | |
86db1e29 | 419 | { |
ae1b1539 | 420 | struct request_queue *q = rq->q; |
c888a8f9 | 421 | unsigned long fflags = q->queue_flags; /* may change, cache */ |
ae1b1539 | 422 | unsigned int policy = blk_flush_policy(fflags, rq); |
e97c293c | 423 | struct blk_flush_queue *fq = blk_get_flush_queue(q, rq->mq_ctx); |
86db1e29 | 424 | |
ae1b1539 TH |
425 | /* |
426 | * @policy now records what operations need to be done. Adjust | |
28a8f0d3 | 427 | * REQ_PREFLUSH and FUA for the driver. |
ae1b1539 | 428 | */ |
28a8f0d3 | 429 | rq->cmd_flags &= ~REQ_PREFLUSH; |
c888a8f9 | 430 | if (!(fflags & (1UL << QUEUE_FLAG_FUA))) |
ae1b1539 TH |
431 | rq->cmd_flags &= ~REQ_FUA; |
432 | ||
ae5b2ec8 JA |
433 | /* |
434 | * REQ_PREFLUSH|REQ_FUA implies REQ_SYNC, so if we clear any | |
435 | * of those flags, we have to set REQ_SYNC to avoid skewing | |
436 | * the request accounting. | |
437 | */ | |
438 | rq->cmd_flags |= REQ_SYNC; | |
439 | ||
4853abaa JM |
440 | /* |
441 | * An empty flush handed down from a stacking driver may | |
442 | * translate into nothing if the underlying device does not | |
443 | * advertise a write-back cache. In this case, simply | |
444 | * complete the request. | |
445 | */ | |
446 | if (!policy) { | |
320ae51f | 447 | if (q->mq_ops) |
c8a446ad | 448 | blk_mq_end_request(rq, 0); |
320ae51f | 449 | else |
d0fac025 | 450 | __blk_end_request(rq, 0, 0); |
4853abaa JM |
451 | return; |
452 | } | |
453 | ||
834f9f61 | 454 | BUG_ON(rq->bio != rq->biotail); /*assumes zero or single bio rq */ |
4853abaa | 455 | |
ae1b1539 TH |
456 | /* |
457 | * If there's data but flush is not necessary, the request can be | |
458 | * processed directly without going through flush machinery. Queue | |
459 | * for normal execution. | |
460 | */ | |
461 | if ((policy & REQ_FSEQ_DATA) && | |
462 | !(policy & (REQ_FSEQ_PREFLUSH | REQ_FSEQ_POSTFLUSH))) { | |
bd166ef1 | 463 | if (q->mq_ops) |
bd6737f1 | 464 | blk_mq_sched_insert_request(rq, false, true, false, false); |
bd166ef1 | 465 | else |
dcd8376c | 466 | list_add_tail(&rq->queuelist, &q->queue_head); |
ae1b1539 | 467 | return; |
28e7d184 | 468 | } |
cde4c406 | 469 | |
ae1b1539 TH |
470 | /* |
471 | * @rq should go through flush machinery. Mark it part of flush | |
472 | * sequence and submit for further processing. | |
473 | */ | |
474 | memset(&rq->flush, 0, sizeof(rq->flush)); | |
475 | INIT_LIST_HEAD(&rq->flush.list); | |
e8064021 | 476 | rq->rq_flags |= RQF_FLUSH_SEQ; |
4853abaa | 477 | rq->flush.saved_end_io = rq->end_io; /* Usually NULL */ |
320ae51f JA |
478 | if (q->mq_ops) { |
479 | rq->end_io = mq_flush_data_end_io; | |
480 | ||
7c94e1c1 | 481 | spin_lock_irq(&fq->mq_flush_lock); |
0bae352d | 482 | blk_flush_complete_seq(rq, fq, REQ_FSEQ_ACTIONS & ~policy, 0); |
7c94e1c1 | 483 | spin_unlock_irq(&fq->mq_flush_lock); |
320ae51f JA |
484 | return; |
485 | } | |
ae1b1539 TH |
486 | rq->end_io = flush_data_end_io; |
487 | ||
0bae352d | 488 | blk_flush_complete_seq(rq, fq, REQ_FSEQ_ACTIONS & ~policy, 0); |
86db1e29 JA |
489 | } |
490 | ||
86db1e29 JA |
491 | /** |
492 | * blkdev_issue_flush - queue a flush | |
493 | * @bdev: blockdev to issue flush for | |
fbd9b09a | 494 | * @gfp_mask: memory allocation flags (for bio_alloc) |
86db1e29 JA |
495 | * @error_sector: error sector |
496 | * | |
497 | * Description: | |
498 | * Issue a flush for the block device in question. Caller can supply | |
499 | * room for storing the error offset in case of a flush error, if they | |
1be7d207 | 500 | * wish to. |
86db1e29 | 501 | */ |
fbd9b09a | 502 | int blkdev_issue_flush(struct block_device *bdev, gfp_t gfp_mask, |
dd3932ed | 503 | sector_t *error_sector) |
86db1e29 | 504 | { |
86db1e29 JA |
505 | struct request_queue *q; |
506 | struct bio *bio; | |
fbd9b09a | 507 | int ret = 0; |
86db1e29 JA |
508 | |
509 | if (bdev->bd_disk == NULL) | |
510 | return -ENXIO; | |
511 | ||
512 | q = bdev_get_queue(bdev); | |
513 | if (!q) | |
514 | return -ENXIO; | |
515 | ||
f10d9f61 DC |
516 | /* |
517 | * some block devices may not have their queue correctly set up here | |
518 | * (e.g. loop device without a backing file) and so issuing a flush | |
519 | * here will panic. Ensure there is a request function before issuing | |
d391a2dd | 520 | * the flush. |
f10d9f61 DC |
521 | */ |
522 | if (!q->make_request_fn) | |
523 | return -ENXIO; | |
524 | ||
fbd9b09a | 525 | bio = bio_alloc(gfp_mask, 0); |
86db1e29 | 526 | bio->bi_bdev = bdev; |
70fd7614 | 527 | bio->bi_opf = REQ_OP_WRITE | REQ_PREFLUSH; |
86db1e29 | 528 | |
4e49ea4a | 529 | ret = submit_bio_wait(bio); |
dd3932ed CH |
530 | |
531 | /* | |
532 | * The driver must store the error location in ->bi_sector, if | |
533 | * it supports it. For non-stacked drivers, this should be | |
534 | * copied from blk_rq_pos(rq). | |
535 | */ | |
536 | if (error_sector) | |
4f024f37 | 537 | *error_sector = bio->bi_iter.bi_sector; |
86db1e29 | 538 | |
86db1e29 JA |
539 | bio_put(bio); |
540 | return ret; | |
541 | } | |
86db1e29 | 542 | EXPORT_SYMBOL(blkdev_issue_flush); |
320ae51f | 543 | |
f70ced09 ML |
544 | struct blk_flush_queue *blk_alloc_flush_queue(struct request_queue *q, |
545 | int node, int cmd_size) | |
320ae51f | 546 | { |
7c94e1c1 ML |
547 | struct blk_flush_queue *fq; |
548 | int rq_sz = sizeof(struct request); | |
1bcb1ead | 549 | |
f70ced09 | 550 | fq = kzalloc_node(sizeof(*fq), GFP_KERNEL, node); |
7c94e1c1 ML |
551 | if (!fq) |
552 | goto fail; | |
1bcb1ead | 553 | |
6d247d7f | 554 | if (q->mq_ops) |
7c94e1c1 | 555 | spin_lock_init(&fq->mq_flush_lock); |
7c94e1c1 | 556 | |
6d247d7f | 557 | rq_sz = round_up(rq_sz + cmd_size, cache_line_size()); |
f70ced09 | 558 | fq->flush_rq = kzalloc_node(rq_sz, GFP_KERNEL, node); |
7c94e1c1 ML |
559 | if (!fq->flush_rq) |
560 | goto fail_rq; | |
561 | ||
562 | INIT_LIST_HEAD(&fq->flush_queue[0]); | |
563 | INIT_LIST_HEAD(&fq->flush_queue[1]); | |
564 | INIT_LIST_HEAD(&fq->flush_data_in_flight); | |
565 | ||
566 | return fq; | |
567 | ||
568 | fail_rq: | |
569 | kfree(fq); | |
570 | fail: | |
571 | return NULL; | |
320ae51f | 572 | } |
f3552655 | 573 | |
ba483388 | 574 | void blk_free_flush_queue(struct blk_flush_queue *fq) |
f3552655 | 575 | { |
7c94e1c1 ML |
576 | /* bio based request queue hasn't flush queue */ |
577 | if (!fq) | |
578 | return; | |
3c09676c | 579 | |
7c94e1c1 ML |
580 | kfree(fq->flush_rq); |
581 | kfree(fq); | |
582 | } |