18839a0e0STejun Heo /* 24fed947cSTejun Heo * Functions to sequence FLUSH and FUA writes. 3ae1b1539STejun Heo * 4ae1b1539STejun Heo * Copyright (C) 2011 Max Planck Institute for Gravitational Physics 5ae1b1539STejun Heo * Copyright (C) 2011 Tejun Heo <tj@kernel.org> 6ae1b1539STejun Heo * 7ae1b1539STejun Heo * This file is released under the GPLv2. 8ae1b1539STejun Heo * 9ae1b1539STejun Heo * REQ_{FLUSH|FUA} requests are decomposed to sequences consisted of three 10ae1b1539STejun Heo * optional steps - PREFLUSH, DATA and POSTFLUSH - according to the request 11ae1b1539STejun Heo * properties and hardware capability. 12ae1b1539STejun Heo * 13ae1b1539STejun Heo * If a request doesn't have data, only REQ_FLUSH makes sense, which 14ae1b1539STejun Heo * indicates a simple flush request. If there is data, REQ_FLUSH indicates 15ae1b1539STejun Heo * that the device cache should be flushed before the data is executed, and 16ae1b1539STejun Heo * REQ_FUA means that the data must be on non-volatile media on request 17ae1b1539STejun Heo * completion. 18ae1b1539STejun Heo * 19ae1b1539STejun Heo * If the device doesn't have writeback cache, FLUSH and FUA don't make any 20ae1b1539STejun Heo * difference. The requests are either completed immediately if there's no 21ae1b1539STejun Heo * data or executed as normal requests otherwise. 22ae1b1539STejun Heo * 23ae1b1539STejun Heo * If the device has writeback cache and supports FUA, REQ_FLUSH is 24ae1b1539STejun Heo * translated to PREFLUSH but REQ_FUA is passed down directly with DATA. 25ae1b1539STejun Heo * 26ae1b1539STejun Heo * If the device has writeback cache and doesn't support FUA, REQ_FLUSH is 27ae1b1539STejun Heo * translated to PREFLUSH and REQ_FUA to POSTFLUSH. 28ae1b1539STejun Heo * 29ae1b1539STejun Heo * The actual execution of flush is double buffered. Whenever a request 30ae1b1539STejun Heo * needs to execute PRE or POSTFLUSH, it queues at 31ae1b1539STejun Heo * q->flush_queue[q->flush_pending_idx]. Once certain criteria are met, a 32ae1b1539STejun Heo * flush is issued and the pending_idx is toggled. When the flush 33ae1b1539STejun Heo * completes, all the requests which were pending are proceeded to the next 34ae1b1539STejun Heo * step. This allows arbitrary merging of different types of FLUSH/FUA 35ae1b1539STejun Heo * requests. 36ae1b1539STejun Heo * 37ae1b1539STejun Heo * Currently, the following conditions are used to determine when to issue 38ae1b1539STejun Heo * flush. 39ae1b1539STejun Heo * 40ae1b1539STejun Heo * C1. At any given time, only one flush shall be in progress. This makes 41ae1b1539STejun Heo * double buffering sufficient. 42ae1b1539STejun Heo * 43ae1b1539STejun Heo * C2. Flush is deferred if any request is executing DATA of its sequence. 44ae1b1539STejun Heo * This avoids issuing separate POSTFLUSHes for requests which shared 45ae1b1539STejun Heo * PREFLUSH. 46ae1b1539STejun Heo * 47ae1b1539STejun Heo * C3. The second condition is ignored if there is a request which has 48ae1b1539STejun Heo * waited longer than FLUSH_PENDING_TIMEOUT. This is to avoid 49ae1b1539STejun Heo * starvation in the unlikely case where there are continuous stream of 50ae1b1539STejun Heo * FUA (without FLUSH) requests. 51ae1b1539STejun Heo * 52ae1b1539STejun Heo * For devices which support FUA, it isn't clear whether C2 (and thus C3) 53ae1b1539STejun Heo * is beneficial. 54ae1b1539STejun Heo * 55ae1b1539STejun Heo * Note that a sequenced FLUSH/FUA request with DATA is completed twice. 56ae1b1539STejun Heo * Once while executing DATA and again after the whole sequence is 57ae1b1539STejun Heo * complete. The first completion updates the contained bio but doesn't 58ae1b1539STejun Heo * finish it so that the bio submitter is notified only after the whole 59ae1b1539STejun Heo * sequence is complete. This is implemented by testing REQ_FLUSH_SEQ in 60ae1b1539STejun Heo * req_bio_endio(). 61ae1b1539STejun Heo * 62ae1b1539STejun Heo * The above peculiarity requires that each FLUSH/FUA request has only one 63ae1b1539STejun Heo * bio attached to it, which is guaranteed as they aren't allowed to be 64ae1b1539STejun Heo * merged in the usual way. 658839a0e0STejun Heo */ 66ae1b1539STejun Heo 678839a0e0STejun Heo #include <linux/kernel.h> 688839a0e0STejun Heo #include <linux/module.h> 698839a0e0STejun Heo #include <linux/bio.h> 708839a0e0STejun Heo #include <linux/blkdev.h> 718839a0e0STejun Heo #include <linux/gfp.h> 72320ae51fSJens Axboe #include <linux/blk-mq.h> 738839a0e0STejun Heo 748839a0e0STejun Heo #include "blk.h" 75320ae51fSJens Axboe #include "blk-mq.h" 768839a0e0STejun Heo 774fed947cSTejun Heo /* FLUSH/FUA sequences */ 784fed947cSTejun Heo enum { 79ae1b1539STejun Heo REQ_FSEQ_PREFLUSH = (1 << 0), /* pre-flushing in progress */ 80ae1b1539STejun Heo REQ_FSEQ_DATA = (1 << 1), /* data write in progress */ 81ae1b1539STejun Heo REQ_FSEQ_POSTFLUSH = (1 << 2), /* post-flushing in progress */ 82ae1b1539STejun Heo REQ_FSEQ_DONE = (1 << 3), 83ae1b1539STejun Heo 84ae1b1539STejun Heo REQ_FSEQ_ACTIONS = REQ_FSEQ_PREFLUSH | REQ_FSEQ_DATA | 85ae1b1539STejun Heo REQ_FSEQ_POSTFLUSH, 86ae1b1539STejun Heo 87ae1b1539STejun Heo /* 88ae1b1539STejun Heo * If flush has been pending longer than the following timeout, 89ae1b1539STejun Heo * it's issued even if flush_data requests are still in flight. 90ae1b1539STejun Heo */ 91ae1b1539STejun Heo FLUSH_PENDING_TIMEOUT = 5 * HZ, 924fed947cSTejun Heo }; 934fed947cSTejun Heo 94ae1b1539STejun Heo static bool blk_kick_flush(struct request_queue *q); 958839a0e0STejun Heo 96ae1b1539STejun Heo static unsigned int blk_flush_policy(unsigned int fflags, struct request *rq) 978839a0e0STejun Heo { 98ae1b1539STejun Heo unsigned int policy = 0; 99ae1b1539STejun Heo 100fa1bf42fSJeff Moyer if (blk_rq_sectors(rq)) 101fa1bf42fSJeff Moyer policy |= REQ_FSEQ_DATA; 102fa1bf42fSJeff Moyer 103ae1b1539STejun Heo if (fflags & REQ_FLUSH) { 104ae1b1539STejun Heo if (rq->cmd_flags & REQ_FLUSH) 105ae1b1539STejun Heo policy |= REQ_FSEQ_PREFLUSH; 106ae1b1539STejun Heo if (!(fflags & REQ_FUA) && (rq->cmd_flags & REQ_FUA)) 107ae1b1539STejun Heo policy |= REQ_FSEQ_POSTFLUSH; 108ae1b1539STejun Heo } 109ae1b1539STejun Heo return policy; 1108839a0e0STejun Heo } 1118839a0e0STejun Heo 112ae1b1539STejun Heo static unsigned int blk_flush_cur_seq(struct request *rq) 1138839a0e0STejun Heo { 114ae1b1539STejun Heo return 1 << ffz(rq->flush.seq); 1158839a0e0STejun Heo } 1168839a0e0STejun Heo 117ae1b1539STejun Heo static void blk_flush_restore_request(struct request *rq) 11847f70d5aSTejun Heo { 11947f70d5aSTejun Heo /* 120ae1b1539STejun Heo * After flush data completion, @rq->bio is %NULL but we need to 121ae1b1539STejun Heo * complete the bio again. @rq->biotail is guaranteed to equal the 122ae1b1539STejun Heo * original @rq->bio. Restore it. 12347f70d5aSTejun Heo */ 124ae1b1539STejun Heo rq->bio = rq->biotail; 125ae1b1539STejun Heo 126ae1b1539STejun Heo /* make @rq a normal request */ 127ae1b1539STejun Heo rq->cmd_flags &= ~REQ_FLUSH_SEQ; 1284853abaaSJeff Moyer rq->end_io = rq->flush.saved_end_io; 129320ae51fSJens Axboe 130320ae51fSJens Axboe blk_clear_rq_complete(rq); 131320ae51fSJens Axboe } 132320ae51fSJens Axboe 13318741986SChristoph Hellwig static void mq_flush_run(struct work_struct *work) 134320ae51fSJens Axboe { 135320ae51fSJens Axboe struct request *rq; 136320ae51fSJens Axboe 13718741986SChristoph Hellwig rq = container_of(work, struct request, mq_flush_work); 138320ae51fSJens Axboe 139320ae51fSJens Axboe memset(&rq->csd, 0, sizeof(rq->csd)); 140feb71daeSChristoph Hellwig blk_mq_insert_request(rq, false, true, false); 141320ae51fSJens Axboe } 142320ae51fSJens Axboe 14310beafc1SMike Snitzer static bool blk_flush_queue_rq(struct request *rq, bool add_front) 144320ae51fSJens Axboe { 14518741986SChristoph Hellwig if (rq->q->mq_ops) { 14618741986SChristoph Hellwig INIT_WORK(&rq->mq_flush_work, mq_flush_run); 14759c3d45eSJens Axboe kblockd_schedule_work(&rq->mq_flush_work); 14818741986SChristoph Hellwig return false; 14918741986SChristoph Hellwig } else { 15010beafc1SMike Snitzer if (add_front) 15110beafc1SMike Snitzer list_add(&rq->queuelist, &rq->q->queue_head); 15210beafc1SMike Snitzer else 15318741986SChristoph Hellwig list_add_tail(&rq->queuelist, &rq->q->queue_head); 15418741986SChristoph Hellwig return true; 15518741986SChristoph Hellwig } 15647f70d5aSTejun Heo } 15747f70d5aSTejun Heo 158ae1b1539STejun Heo /** 159ae1b1539STejun Heo * blk_flush_complete_seq - complete flush sequence 160ae1b1539STejun Heo * @rq: FLUSH/FUA request being sequenced 161ae1b1539STejun Heo * @seq: sequences to complete (mask of %REQ_FSEQ_*, can be zero) 162ae1b1539STejun Heo * @error: whether an error occurred 163ae1b1539STejun Heo * 164ae1b1539STejun Heo * @rq just completed @seq part of its flush sequence, record the 165ae1b1539STejun Heo * completion and trigger the next step. 166ae1b1539STejun Heo * 167ae1b1539STejun Heo * CONTEXT: 168320ae51fSJens Axboe * spin_lock_irq(q->queue_lock or q->mq_flush_lock) 169ae1b1539STejun Heo * 170ae1b1539STejun Heo * RETURNS: 171ae1b1539STejun Heo * %true if requests were added to the dispatch queue, %false otherwise. 172ae1b1539STejun Heo */ 173ae1b1539STejun Heo static bool blk_flush_complete_seq(struct request *rq, unsigned int seq, 174ae1b1539STejun Heo int error) 1758839a0e0STejun Heo { 176ae1b1539STejun Heo struct request_queue *q = rq->q; 177ae1b1539STejun Heo struct list_head *pending = &q->flush_queue[q->flush_pending_idx]; 178320ae51fSJens Axboe bool queued = false, kicked; 1798839a0e0STejun Heo 180ae1b1539STejun Heo BUG_ON(rq->flush.seq & seq); 181ae1b1539STejun Heo rq->flush.seq |= seq; 1828839a0e0STejun Heo 183ae1b1539STejun Heo if (likely(!error)) 184ae1b1539STejun Heo seq = blk_flush_cur_seq(rq); 185ae1b1539STejun Heo else 186ae1b1539STejun Heo seq = REQ_FSEQ_DONE; 1878839a0e0STejun Heo 188ae1b1539STejun Heo switch (seq) { 189ae1b1539STejun Heo case REQ_FSEQ_PREFLUSH: 190ae1b1539STejun Heo case REQ_FSEQ_POSTFLUSH: 191ae1b1539STejun Heo /* queue for flush */ 192ae1b1539STejun Heo if (list_empty(pending)) 193ae1b1539STejun Heo q->flush_pending_since = jiffies; 194ae1b1539STejun Heo list_move_tail(&rq->flush.list, pending); 1958839a0e0STejun Heo break; 196ae1b1539STejun Heo 197ae1b1539STejun Heo case REQ_FSEQ_DATA: 198ae1b1539STejun Heo list_move_tail(&rq->flush.list, &q->flush_data_in_flight); 19910beafc1SMike Snitzer queued = blk_flush_queue_rq(rq, true); 200ae1b1539STejun Heo break; 201ae1b1539STejun Heo 202ae1b1539STejun Heo case REQ_FSEQ_DONE: 20309d60c70STejun Heo /* 204ae1b1539STejun Heo * @rq was previously adjusted by blk_flush_issue() for 205ae1b1539STejun Heo * flush sequencing and may already have gone through the 206ae1b1539STejun Heo * flush data request completion path. Restore @rq for 207ae1b1539STejun Heo * normal completion and end it. 20809d60c70STejun Heo */ 209ae1b1539STejun Heo BUG_ON(!list_empty(&rq->queuelist)); 210ae1b1539STejun Heo list_del_init(&rq->flush.list); 211ae1b1539STejun Heo blk_flush_restore_request(rq); 212320ae51fSJens Axboe if (q->mq_ops) 213320ae51fSJens Axboe blk_mq_end_io(rq, error); 214320ae51fSJens Axboe else 215ae1b1539STejun Heo __blk_end_request_all(rq, error); 2168839a0e0STejun Heo break; 217ae1b1539STejun Heo 2188839a0e0STejun Heo default: 2198839a0e0STejun Heo BUG(); 2208839a0e0STejun Heo } 221cde4c406SChristoph Hellwig 222320ae51fSJens Axboe kicked = blk_kick_flush(q); 223320ae51fSJens Axboe return kicked | queued; 2248839a0e0STejun Heo } 2258839a0e0STejun Heo 226ae1b1539STejun Heo static void flush_end_io(struct request *flush_rq, int error) 2278839a0e0STejun Heo { 228ae1b1539STejun Heo struct request_queue *q = flush_rq->q; 229320ae51fSJens Axboe struct list_head *running; 230ae1b1539STejun Heo bool queued = false; 231ae1b1539STejun Heo struct request *rq, *n; 232320ae51fSJens Axboe unsigned long flags = 0; 2338839a0e0STejun Heo 23418741986SChristoph Hellwig if (q->mq_ops) 235320ae51fSJens Axboe spin_lock_irqsave(&q->mq_flush_lock, flags); 23618741986SChristoph Hellwig 237320ae51fSJens Axboe running = &q->flush_queue[q->flush_running_idx]; 238ae1b1539STejun Heo BUG_ON(q->flush_pending_idx == q->flush_running_idx); 239ae1b1539STejun Heo 240ae1b1539STejun Heo /* account completion of the flush request */ 241ae1b1539STejun Heo q->flush_running_idx ^= 1; 242320ae51fSJens Axboe 243320ae51fSJens Axboe if (!q->mq_ops) 244ae1b1539STejun Heo elv_completed_request(q, flush_rq); 245ae1b1539STejun Heo 246ae1b1539STejun Heo /* and push the waiting requests to the next stage */ 247ae1b1539STejun Heo list_for_each_entry_safe(rq, n, running, flush.list) { 248ae1b1539STejun Heo unsigned int seq = blk_flush_cur_seq(rq); 249ae1b1539STejun Heo 250ae1b1539STejun Heo BUG_ON(seq != REQ_FSEQ_PREFLUSH && seq != REQ_FSEQ_POSTFLUSH); 251ae1b1539STejun Heo queued |= blk_flush_complete_seq(rq, seq, error); 252ae1b1539STejun Heo } 253ae1b1539STejun Heo 2548839a0e0STejun Heo /* 2553ac0cc45Sshaohua.li@intel.com * Kick the queue to avoid stall for two cases: 2563ac0cc45Sshaohua.li@intel.com * 1. Moving a request silently to empty queue_head may stall the 2573ac0cc45Sshaohua.li@intel.com * queue. 2583ac0cc45Sshaohua.li@intel.com * 2. When flush request is running in non-queueable queue, the 2593ac0cc45Sshaohua.li@intel.com * queue is hold. Restart the queue after flush request is finished 2603ac0cc45Sshaohua.li@intel.com * to avoid stall. 2613ac0cc45Sshaohua.li@intel.com * This function is called from request completion path and calling 2623ac0cc45Sshaohua.li@intel.com * directly into request_fn may confuse the driver. Always use 2633ac0cc45Sshaohua.li@intel.com * kblockd. 2648839a0e0STejun Heo */ 265320ae51fSJens Axboe if (queued || q->flush_queue_delayed) { 26618741986SChristoph Hellwig WARN_ON(q->mq_ops); 26724ecfbe2SChristoph Hellwig blk_run_queue_async(q); 268320ae51fSJens Axboe } 2693ac0cc45Sshaohua.li@intel.com q->flush_queue_delayed = 0; 270320ae51fSJens Axboe if (q->mq_ops) 271320ae51fSJens Axboe spin_unlock_irqrestore(&q->mq_flush_lock, flags); 272320ae51fSJens Axboe } 273320ae51fSJens Axboe 274ae1b1539STejun Heo /** 275ae1b1539STejun Heo * blk_kick_flush - consider issuing flush request 276ae1b1539STejun Heo * @q: request_queue being kicked 2774fed947cSTejun Heo * 278ae1b1539STejun Heo * Flush related states of @q have changed, consider issuing flush request. 279ae1b1539STejun Heo * Please read the comment at the top of this file for more info. 280ae1b1539STejun Heo * 281ae1b1539STejun Heo * CONTEXT: 282320ae51fSJens Axboe * spin_lock_irq(q->queue_lock or q->mq_flush_lock) 283ae1b1539STejun Heo * 284ae1b1539STejun Heo * RETURNS: 285ae1b1539STejun Heo * %true if flush was issued, %false otherwise. 2868839a0e0STejun Heo */ 287ae1b1539STejun Heo static bool blk_kick_flush(struct request_queue *q) 288ae1b1539STejun Heo { 289ae1b1539STejun Heo struct list_head *pending = &q->flush_queue[q->flush_pending_idx]; 290ae1b1539STejun Heo struct request *first_rq = 291ae1b1539STejun Heo list_first_entry(pending, struct request, flush.list); 292ae1b1539STejun Heo 293ae1b1539STejun Heo /* C1 described at the top of this file */ 294ae1b1539STejun Heo if (q->flush_pending_idx != q->flush_running_idx || list_empty(pending)) 295ae1b1539STejun Heo return false; 296ae1b1539STejun Heo 297ae1b1539STejun Heo /* C2 and C3 */ 298ae1b1539STejun Heo if (!list_empty(&q->flush_data_in_flight) && 299ae1b1539STejun Heo time_before(jiffies, 300ae1b1539STejun Heo q->flush_pending_since + FLUSH_PENDING_TIMEOUT)) 301ae1b1539STejun Heo return false; 302ae1b1539STejun Heo 303ae1b1539STejun Heo /* 304ae1b1539STejun Heo * Issue flush and toggle pending_idx. This makes pending_idx 305ae1b1539STejun Heo * different from running_idx, which means flush is in flight. 306ae1b1539STejun Heo */ 307320ae51fSJens Axboe q->flush_pending_idx ^= 1; 30818741986SChristoph Hellwig 3099d74e257SChristoph Hellwig blk_rq_init(q, q->flush_rq); 3108727af4bSChristoph Hellwig if (q->mq_ops) 3118727af4bSChristoph Hellwig blk_mq_clone_flush_request(q->flush_rq, first_rq); 312320ae51fSJens Axboe 31318741986SChristoph Hellwig q->flush_rq->cmd_type = REQ_TYPE_FS; 31418741986SChristoph Hellwig q->flush_rq->cmd_flags = WRITE_FLUSH | REQ_FLUSH_SEQ; 31518741986SChristoph Hellwig q->flush_rq->rq_disk = first_rq->rq_disk; 31618741986SChristoph Hellwig q->flush_rq->end_io = flush_end_io; 317ae1b1539STejun Heo 31810beafc1SMike Snitzer return blk_flush_queue_rq(q->flush_rq, false); 319ae1b1539STejun Heo } 320ae1b1539STejun Heo 321ae1b1539STejun Heo static void flush_data_end_io(struct request *rq, int error) 322ae1b1539STejun Heo { 323ae1b1539STejun Heo struct request_queue *q = rq->q; 324ae1b1539STejun Heo 3258839a0e0STejun Heo /* 326e83a46bbSTejun Heo * After populating an empty queue, kick it to avoid stall. Read 327e83a46bbSTejun Heo * the comment in flush_end_io(). 3288839a0e0STejun Heo */ 32973c10101SJens Axboe if (blk_flush_complete_seq(rq, REQ_FSEQ_DATA, error)) 33024ecfbe2SChristoph Hellwig blk_run_queue_async(q); 331ae1b1539STejun Heo } 332ae1b1539STejun Heo 333320ae51fSJens Axboe static void mq_flush_data_end_io(struct request *rq, int error) 334320ae51fSJens Axboe { 335320ae51fSJens Axboe struct request_queue *q = rq->q; 336320ae51fSJens Axboe struct blk_mq_hw_ctx *hctx; 337320ae51fSJens Axboe struct blk_mq_ctx *ctx; 338320ae51fSJens Axboe unsigned long flags; 339320ae51fSJens Axboe 340320ae51fSJens Axboe ctx = rq->mq_ctx; 341320ae51fSJens Axboe hctx = q->mq_ops->map_queue(q, ctx->cpu); 342320ae51fSJens Axboe 343320ae51fSJens Axboe /* 344320ae51fSJens Axboe * After populating an empty queue, kick it to avoid stall. Read 345320ae51fSJens Axboe * the comment in flush_end_io(). 346320ae51fSJens Axboe */ 347320ae51fSJens Axboe spin_lock_irqsave(&q->mq_flush_lock, flags); 348320ae51fSJens Axboe if (blk_flush_complete_seq(rq, REQ_FSEQ_DATA, error)) 349320ae51fSJens Axboe blk_mq_run_hw_queue(hctx, true); 350320ae51fSJens Axboe spin_unlock_irqrestore(&q->mq_flush_lock, flags); 351320ae51fSJens Axboe } 352320ae51fSJens Axboe 353ae1b1539STejun Heo /** 354ae1b1539STejun Heo * blk_insert_flush - insert a new FLUSH/FUA request 355ae1b1539STejun Heo * @rq: request to insert 356ae1b1539STejun Heo * 357b710a480SJens Axboe * To be called from __elv_add_request() for %ELEVATOR_INSERT_FLUSH insertions. 358320ae51fSJens Axboe * or __blk_mq_run_hw_queue() to dispatch request. 359ae1b1539STejun Heo * @rq is being submitted. Analyze what needs to be done and put it on the 360ae1b1539STejun Heo * right queue. 361ae1b1539STejun Heo * 362ae1b1539STejun Heo * CONTEXT: 363320ae51fSJens Axboe * spin_lock_irq(q->queue_lock) in !mq case 364ae1b1539STejun Heo */ 365ae1b1539STejun Heo void blk_insert_flush(struct request *rq) 366ae1b1539STejun Heo { 367ae1b1539STejun Heo struct request_queue *q = rq->q; 368ae1b1539STejun Heo unsigned int fflags = q->flush_flags; /* may change, cache */ 369ae1b1539STejun Heo unsigned int policy = blk_flush_policy(fflags, rq); 370ae1b1539STejun Heo 371ae1b1539STejun Heo /* 372ae1b1539STejun Heo * @policy now records what operations need to be done. Adjust 373ae1b1539STejun Heo * REQ_FLUSH and FUA for the driver. 374ae1b1539STejun Heo */ 3754fed947cSTejun Heo rq->cmd_flags &= ~REQ_FLUSH; 376ae1b1539STejun Heo if (!(fflags & REQ_FUA)) 3774fed947cSTejun Heo rq->cmd_flags &= ~REQ_FUA; 378ae1b1539STejun Heo 379ae1b1539STejun Heo /* 3804853abaaSJeff Moyer * An empty flush handed down from a stacking driver may 3814853abaaSJeff Moyer * translate into nothing if the underlying device does not 3824853abaaSJeff Moyer * advertise a write-back cache. In this case, simply 3834853abaaSJeff Moyer * complete the request. 3844853abaaSJeff Moyer */ 3854853abaaSJeff Moyer if (!policy) { 386320ae51fSJens Axboe if (q->mq_ops) 387320ae51fSJens Axboe blk_mq_end_io(rq, 0); 388320ae51fSJens Axboe else 3894853abaaSJeff Moyer __blk_end_bidi_request(rq, 0, 0, 0); 3904853abaaSJeff Moyer return; 3914853abaaSJeff Moyer } 3924853abaaSJeff Moyer 393834f9f61SJeff Moyer BUG_ON(rq->bio != rq->biotail); /*assumes zero or single bio rq */ 3944853abaaSJeff Moyer 3954853abaaSJeff Moyer /* 396ae1b1539STejun Heo * If there's data but flush is not necessary, the request can be 397ae1b1539STejun Heo * processed directly without going through flush machinery. Queue 398ae1b1539STejun Heo * for normal execution. 399ae1b1539STejun Heo */ 400ae1b1539STejun Heo if ((policy & REQ_FSEQ_DATA) && 401ae1b1539STejun Heo !(policy & (REQ_FSEQ_PREFLUSH | REQ_FSEQ_POSTFLUSH))) { 402320ae51fSJens Axboe if (q->mq_ops) { 403feb71daeSChristoph Hellwig blk_mq_insert_request(rq, false, false, true); 404320ae51fSJens Axboe } else 40553d63e6bSJens Axboe list_add_tail(&rq->queuelist, &q->queue_head); 406ae1b1539STejun Heo return; 4078839a0e0STejun Heo } 4088839a0e0STejun Heo 4098839a0e0STejun Heo /* 410ae1b1539STejun Heo * @rq should go through flush machinery. Mark it part of flush 411ae1b1539STejun Heo * sequence and submit for further processing. 4128839a0e0STejun Heo */ 413ae1b1539STejun Heo memset(&rq->flush, 0, sizeof(rq->flush)); 414ae1b1539STejun Heo INIT_LIST_HEAD(&rq->flush.list); 415ae1b1539STejun Heo rq->cmd_flags |= REQ_FLUSH_SEQ; 4164853abaaSJeff Moyer rq->flush.saved_end_io = rq->end_io; /* Usually NULL */ 417320ae51fSJens Axboe if (q->mq_ops) { 418320ae51fSJens Axboe rq->end_io = mq_flush_data_end_io; 419320ae51fSJens Axboe 420320ae51fSJens Axboe spin_lock_irq(&q->mq_flush_lock); 421320ae51fSJens Axboe blk_flush_complete_seq(rq, REQ_FSEQ_ACTIONS & ~policy, 0); 422320ae51fSJens Axboe spin_unlock_irq(&q->mq_flush_lock); 423320ae51fSJens Axboe return; 424320ae51fSJens Axboe } 425ae1b1539STejun Heo rq->end_io = flush_data_end_io; 426ae1b1539STejun Heo 427ae1b1539STejun Heo blk_flush_complete_seq(rq, REQ_FSEQ_ACTIONS & ~policy, 0); 428ae1b1539STejun Heo } 429ae1b1539STejun Heo 430ae1b1539STejun Heo /** 431ae1b1539STejun Heo * blk_abort_flushes - @q is being aborted, abort flush requests 432ae1b1539STejun Heo * @q: request_queue being aborted 433ae1b1539STejun Heo * 434ae1b1539STejun Heo * To be called from elv_abort_queue(). @q is being aborted. Prepare all 435ae1b1539STejun Heo * FLUSH/FUA requests for abortion. 436ae1b1539STejun Heo * 437ae1b1539STejun Heo * CONTEXT: 438ae1b1539STejun Heo * spin_lock_irq(q->queue_lock) 439ae1b1539STejun Heo */ 440ae1b1539STejun Heo void blk_abort_flushes(struct request_queue *q) 441ae1b1539STejun Heo { 442ae1b1539STejun Heo struct request *rq, *n; 443ae1b1539STejun Heo int i; 444ae1b1539STejun Heo 445ae1b1539STejun Heo /* 446ae1b1539STejun Heo * Requests in flight for data are already owned by the dispatch 447ae1b1539STejun Heo * queue or the device driver. Just restore for normal completion. 448ae1b1539STejun Heo */ 449ae1b1539STejun Heo list_for_each_entry_safe(rq, n, &q->flush_data_in_flight, flush.list) { 450ae1b1539STejun Heo list_del_init(&rq->flush.list); 451ae1b1539STejun Heo blk_flush_restore_request(rq); 4528839a0e0STejun Heo } 4538839a0e0STejun Heo 4548839a0e0STejun Heo /* 455ae1b1539STejun Heo * We need to give away requests on flush queues. Restore for 456ae1b1539STejun Heo * normal completion and put them on the dispatch queue. 4578839a0e0STejun Heo */ 458ae1b1539STejun Heo for (i = 0; i < ARRAY_SIZE(q->flush_queue); i++) { 459ae1b1539STejun Heo list_for_each_entry_safe(rq, n, &q->flush_queue[i], 460ae1b1539STejun Heo flush.list) { 461ae1b1539STejun Heo list_del_init(&rq->flush.list); 462ae1b1539STejun Heo blk_flush_restore_request(rq); 463ae1b1539STejun Heo list_add_tail(&rq->queuelist, &q->queue_head); 464ae1b1539STejun Heo } 465ae1b1539STejun Heo } 4668839a0e0STejun Heo } 4678839a0e0STejun Heo 4688839a0e0STejun Heo /** 4698839a0e0STejun Heo * blkdev_issue_flush - queue a flush 4708839a0e0STejun Heo * @bdev: blockdev to issue flush for 4718839a0e0STejun Heo * @gfp_mask: memory allocation flags (for bio_alloc) 4728839a0e0STejun Heo * @error_sector: error sector 4738839a0e0STejun Heo * 4748839a0e0STejun Heo * Description: 4758839a0e0STejun Heo * Issue a flush for the block device in question. Caller can supply 4768839a0e0STejun Heo * room for storing the error offset in case of a flush error, if they 4778839a0e0STejun Heo * wish to. If WAIT flag is not passed then caller may check only what 4788839a0e0STejun Heo * request was pushed in some internal queue for later handling. 4798839a0e0STejun Heo */ 4808839a0e0STejun Heo int blkdev_issue_flush(struct block_device *bdev, gfp_t gfp_mask, 481dd3932edSChristoph Hellwig sector_t *error_sector) 4828839a0e0STejun Heo { 4838839a0e0STejun Heo struct request_queue *q; 4848839a0e0STejun Heo struct bio *bio; 4858839a0e0STejun Heo int ret = 0; 4868839a0e0STejun Heo 4878839a0e0STejun Heo if (bdev->bd_disk == NULL) 4888839a0e0STejun Heo return -ENXIO; 4898839a0e0STejun Heo 4908839a0e0STejun Heo q = bdev_get_queue(bdev); 4918839a0e0STejun Heo if (!q) 4928839a0e0STejun Heo return -ENXIO; 4938839a0e0STejun Heo 4948839a0e0STejun Heo /* 4958839a0e0STejun Heo * some block devices may not have their queue correctly set up here 4968839a0e0STejun Heo * (e.g. loop device without a backing file) and so issuing a flush 4978839a0e0STejun Heo * here will panic. Ensure there is a request function before issuing 498d391a2ddSTejun Heo * the flush. 4998839a0e0STejun Heo */ 5008839a0e0STejun Heo if (!q->make_request_fn) 5018839a0e0STejun Heo return -ENXIO; 5028839a0e0STejun Heo 5038839a0e0STejun Heo bio = bio_alloc(gfp_mask, 0); 5048839a0e0STejun Heo bio->bi_bdev = bdev; 5058839a0e0STejun Heo 50633879d45SKent Overstreet ret = submit_bio_wait(WRITE_FLUSH, bio); 507dd3932edSChristoph Hellwig 5088839a0e0STejun Heo /* 5098839a0e0STejun Heo * The driver must store the error location in ->bi_sector, if 5108839a0e0STejun Heo * it supports it. For non-stacked drivers, this should be 5118839a0e0STejun Heo * copied from blk_rq_pos(rq). 5128839a0e0STejun Heo */ 5138839a0e0STejun Heo if (error_sector) 5144f024f37SKent Overstreet *error_sector = bio->bi_iter.bi_sector; 5158839a0e0STejun Heo 5168839a0e0STejun Heo bio_put(bio); 5178839a0e0STejun Heo return ret; 5188839a0e0STejun Heo } 5198839a0e0STejun Heo EXPORT_SYMBOL(blkdev_issue_flush); 520320ae51fSJens Axboe 521320ae51fSJens Axboe void blk_mq_init_flush(struct request_queue *q) 522320ae51fSJens Axboe { 523320ae51fSJens Axboe spin_lock_init(&q->mq_flush_lock); 524320ae51fSJens Axboe } 525