Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 1 | /* |
Tejun Heo | 4fed947 | 2010-09-03 11:56:17 +0200 | [diff] [blame] | 2 | * Functions to sequence FLUSH and FUA writes. |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 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 | * |
| 13 | * If a request doesn't have data, only REQ_FLUSH makes sense, which |
| 14 | * indicates a simple flush request. If there is data, REQ_FLUSH indicates |
| 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 | * |
| 23 | * If the device has writeback cache and supports FUA, REQ_FLUSH is |
| 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_FLUSH is |
| 27 | * translated to PREFLUSH and REQ_FUA to POSTFLUSH. |
| 28 | * |
| 29 | * The actual execution of flush is double buffered. Whenever a request |
| 30 | * needs to execute PRE or POSTFLUSH, it queues at |
| 31 | * q->flush_queue[q->flush_pending_idx]. Once certain criteria are met, a |
| 32 | * 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 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 |
| 59 | * sequence is complete. This is implemented by testing REQ_FLUSH_SEQ in |
| 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. |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 65 | */ |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 66 | |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 67 | #include <linux/kernel.h> |
| 68 | #include <linux/module.h> |
| 69 | #include <linux/bio.h> |
| 70 | #include <linux/blkdev.h> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 71 | #include <linux/gfp.h> |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 72 | |
| 73 | #include "blk.h" |
| 74 | |
Tejun Heo | 4fed947 | 2010-09-03 11:56:17 +0200 | [diff] [blame] | 75 | /* FLUSH/FUA sequences */ |
| 76 | enum { |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 77 | REQ_FSEQ_PREFLUSH = (1 << 0), /* pre-flushing in progress */ |
| 78 | REQ_FSEQ_DATA = (1 << 1), /* data write in progress */ |
| 79 | REQ_FSEQ_POSTFLUSH = (1 << 2), /* post-flushing in progress */ |
| 80 | REQ_FSEQ_DONE = (1 << 3), |
Tejun Heo | 4fed947 | 2010-09-03 11:56:17 +0200 | [diff] [blame] | 81 | |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 82 | REQ_FSEQ_ACTIONS = REQ_FSEQ_PREFLUSH | REQ_FSEQ_DATA | |
| 83 | REQ_FSEQ_POSTFLUSH, |
Tejun Heo | 47f70d5 | 2010-09-03 11:56:17 +0200 | [diff] [blame] | 84 | |
| 85 | /* |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 86 | * If flush has been pending longer than the following timeout, |
| 87 | * it's issued even if flush_data requests are still in flight. |
Tejun Heo | 47f70d5 | 2010-09-03 11:56:17 +0200 | [diff] [blame] | 88 | */ |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 89 | FLUSH_PENDING_TIMEOUT = 5 * HZ, |
| 90 | }; |
| 91 | |
| 92 | static bool blk_kick_flush(struct request_queue *q); |
| 93 | |
| 94 | static unsigned int blk_flush_policy(unsigned int fflags, struct request *rq) |
| 95 | { |
| 96 | unsigned int policy = 0; |
| 97 | |
| 98 | if (fflags & REQ_FLUSH) { |
| 99 | if (rq->cmd_flags & REQ_FLUSH) |
| 100 | policy |= REQ_FSEQ_PREFLUSH; |
| 101 | if (blk_rq_sectors(rq)) |
| 102 | policy |= REQ_FSEQ_DATA; |
| 103 | if (!(fflags & REQ_FUA) && (rq->cmd_flags & REQ_FUA)) |
| 104 | policy |= REQ_FSEQ_POSTFLUSH; |
| 105 | } |
| 106 | return policy; |
Tejun Heo | 47f70d5 | 2010-09-03 11:56:17 +0200 | [diff] [blame] | 107 | } |
| 108 | |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 109 | static unsigned int blk_flush_cur_seq(struct request *rq) |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 110 | { |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 111 | return 1 << ffz(rq->flush.seq); |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 112 | } |
| 113 | |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 114 | static void blk_flush_restore_request(struct request *rq) |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 115 | { |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 116 | /* |
| 117 | * After flush data completion, @rq->bio is %NULL but we need to |
| 118 | * complete the bio again. @rq->biotail is guaranteed to equal the |
| 119 | * original @rq->bio. Restore it. |
| 120 | */ |
| 121 | rq->bio = rq->biotail; |
| 122 | |
| 123 | /* make @rq a normal request */ |
| 124 | rq->cmd_flags &= ~REQ_FLUSH_SEQ; |
| 125 | rq->end_io = NULL; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 126 | } |
| 127 | |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 128 | /** |
| 129 | * blk_flush_complete_seq - complete flush sequence |
| 130 | * @rq: FLUSH/FUA request being sequenced |
| 131 | * @seq: sequences to complete (mask of %REQ_FSEQ_*, can be zero) |
| 132 | * @error: whether an error occurred |
| 133 | * |
| 134 | * @rq just completed @seq part of its flush sequence, record the |
| 135 | * completion and trigger the next step. |
| 136 | * |
| 137 | * CONTEXT: |
| 138 | * spin_lock_irq(q->queue_lock) |
| 139 | * |
| 140 | * RETURNS: |
| 141 | * %true if requests were added to the dispatch queue, %false otherwise. |
| 142 | */ |
| 143 | static bool blk_flush_complete_seq(struct request *rq, unsigned int seq, |
| 144 | int error) |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 145 | { |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 146 | struct request_queue *q = rq->q; |
| 147 | struct list_head *pending = &q->flush_queue[q->flush_pending_idx]; |
| 148 | bool queued = false; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 149 | |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 150 | BUG_ON(rq->flush.seq & seq); |
| 151 | rq->flush.seq |= seq; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 152 | |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 153 | if (likely(!error)) |
| 154 | seq = blk_flush_cur_seq(rq); |
| 155 | else |
| 156 | seq = REQ_FSEQ_DONE; |
Tejun Heo | 28e7d18 | 2010-09-03 11:56:16 +0200 | [diff] [blame] | 157 | |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 158 | switch (seq) { |
| 159 | case REQ_FSEQ_PREFLUSH: |
| 160 | case REQ_FSEQ_POSTFLUSH: |
| 161 | /* queue for flush */ |
| 162 | if (list_empty(pending)) |
| 163 | q->flush_pending_since = jiffies; |
| 164 | list_move_tail(&rq->flush.list, pending); |
Tejun Heo | 28e7d18 | 2010-09-03 11:56:16 +0200 | [diff] [blame] | 165 | break; |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 166 | |
| 167 | case REQ_FSEQ_DATA: |
| 168 | list_move_tail(&rq->flush.list, &q->flush_data_in_flight); |
| 169 | list_add(&rq->queuelist, &q->queue_head); |
| 170 | queued = true; |
| 171 | break; |
| 172 | |
| 173 | case REQ_FSEQ_DONE: |
Tejun Heo | 09d60c7 | 2010-09-03 11:56:17 +0200 | [diff] [blame] | 174 | /* |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 175 | * @rq was previously adjusted by blk_flush_issue() for |
| 176 | * flush sequencing and may already have gone through the |
| 177 | * flush data request completion path. Restore @rq for |
| 178 | * normal completion and end it. |
Tejun Heo | 09d60c7 | 2010-09-03 11:56:17 +0200 | [diff] [blame] | 179 | */ |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 180 | BUG_ON(!list_empty(&rq->queuelist)); |
| 181 | list_del_init(&rq->flush.list); |
| 182 | blk_flush_restore_request(rq); |
| 183 | __blk_end_request_all(rq, error); |
Tejun Heo | 28e7d18 | 2010-09-03 11:56:16 +0200 | [diff] [blame] | 184 | break; |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 185 | |
Tejun Heo | 28e7d18 | 2010-09-03 11:56:16 +0200 | [diff] [blame] | 186 | default: |
| 187 | BUG(); |
| 188 | } |
Christoph Hellwig | cde4c40 | 2010-09-03 11:56:17 +0200 | [diff] [blame] | 189 | |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 190 | return blk_kick_flush(q) | queued; |
Tejun Heo | 28e7d18 | 2010-09-03 11:56:16 +0200 | [diff] [blame] | 191 | } |
| 192 | |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 193 | static void flush_end_io(struct request *flush_rq, int error) |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 194 | { |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 195 | struct request_queue *q = flush_rq->q; |
| 196 | struct list_head *running = &q->flush_queue[q->flush_running_idx]; |
| 197 | bool was_empty = elv_queue_empty(q); |
| 198 | bool queued = false; |
| 199 | struct request *rq, *n; |
Tejun Heo | 8f11b3e | 2008-11-28 13:32:05 +0900 | [diff] [blame] | 200 | |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 201 | BUG_ON(q->flush_pending_idx == q->flush_running_idx); |
| 202 | |
| 203 | /* account completion of the flush request */ |
| 204 | q->flush_running_idx ^= 1; |
| 205 | elv_completed_request(q, flush_rq); |
| 206 | |
| 207 | /* and push the waiting requests to the next stage */ |
| 208 | list_for_each_entry_safe(rq, n, running, flush.list) { |
| 209 | unsigned int seq = blk_flush_cur_seq(rq); |
| 210 | |
| 211 | BUG_ON(seq != REQ_FSEQ_PREFLUSH && seq != REQ_FSEQ_POSTFLUSH); |
| 212 | queued |= blk_flush_complete_seq(rq, seq, error); |
Tejun Heo | 28e7d18 | 2010-09-03 11:56:16 +0200 | [diff] [blame] | 213 | } |
| 214 | |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 215 | /* |
| 216 | * Moving a request silently to empty queue_head may stall the |
Tejun Heo | 255bb49 | 2011-03-02 08:48:06 -0500 | [diff] [blame] | 217 | * queue. Kick the queue in those cases. This function is called |
| 218 | * from request completion path and calling directly into |
| 219 | * request_fn may confuse the driver. Always use kblockd. |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 220 | */ |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 221 | if (queued && was_empty) |
Tejun Heo | 255bb49 | 2011-03-02 08:48:06 -0500 | [diff] [blame] | 222 | __blk_run_queue(q, true); |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 223 | } |
| 224 | |
| 225 | /** |
| 226 | * blk_kick_flush - consider issuing flush request |
| 227 | * @q: request_queue being kicked |
| 228 | * |
| 229 | * Flush related states of @q have changed, consider issuing flush request. |
| 230 | * Please read the comment at the top of this file for more info. |
| 231 | * |
| 232 | * CONTEXT: |
| 233 | * spin_lock_irq(q->queue_lock) |
| 234 | * |
| 235 | * RETURNS: |
| 236 | * %true if flush was issued, %false otherwise. |
| 237 | */ |
| 238 | static bool blk_kick_flush(struct request_queue *q) |
| 239 | { |
| 240 | struct list_head *pending = &q->flush_queue[q->flush_pending_idx]; |
| 241 | struct request *first_rq = |
| 242 | list_first_entry(pending, struct request, flush.list); |
| 243 | |
| 244 | /* C1 described at the top of this file */ |
| 245 | if (q->flush_pending_idx != q->flush_running_idx || list_empty(pending)) |
| 246 | return false; |
| 247 | |
| 248 | /* C2 and C3 */ |
| 249 | if (!list_empty(&q->flush_data_in_flight) && |
| 250 | time_before(jiffies, |
| 251 | q->flush_pending_since + FLUSH_PENDING_TIMEOUT)) |
| 252 | return false; |
Tejun Heo | 28e7d18 | 2010-09-03 11:56:16 +0200 | [diff] [blame] | 253 | |
| 254 | /* |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 255 | * Issue flush and toggle pending_idx. This makes pending_idx |
| 256 | * different from running_idx, which means flush is in flight. |
Tejun Heo | 28e7d18 | 2010-09-03 11:56:16 +0200 | [diff] [blame] | 257 | */ |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 258 | blk_rq_init(q, &q->flush_rq); |
| 259 | q->flush_rq.cmd_type = REQ_TYPE_FS; |
| 260 | q->flush_rq.cmd_flags = WRITE_FLUSH | REQ_FLUSH_SEQ; |
| 261 | q->flush_rq.rq_disk = first_rq->rq_disk; |
| 262 | q->flush_rq.end_io = flush_end_io; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 263 | |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 264 | q->flush_pending_idx ^= 1; |
Tejun Heo | e83a46b | 2011-03-04 19:09:02 +0100 | [diff] [blame^] | 265 | elv_insert(q, &q->flush_rq, ELEVATOR_INSERT_REQUEUE); |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 266 | return true; |
| 267 | } |
| 268 | |
| 269 | static void flush_data_end_io(struct request *rq, int error) |
| 270 | { |
| 271 | struct request_queue *q = rq->q; |
| 272 | bool was_empty = elv_queue_empty(q); |
| 273 | |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 274 | /* |
Tejun Heo | e83a46b | 2011-03-04 19:09:02 +0100 | [diff] [blame^] | 275 | * After populating an empty queue, kick it to avoid stall. Read |
| 276 | * the comment in flush_end_io(). |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 277 | */ |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 278 | if (blk_flush_complete_seq(rq, REQ_FSEQ_DATA, error) && was_empty) |
Tejun Heo | e83a46b | 2011-03-04 19:09:02 +0100 | [diff] [blame^] | 279 | __blk_run_queue(q, true); |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 280 | } |
| 281 | |
| 282 | /** |
| 283 | * blk_insert_flush - insert a new FLUSH/FUA request |
| 284 | * @rq: request to insert |
| 285 | * |
| 286 | * To be called from elv_insert() for %ELEVATOR_INSERT_FLUSH insertions. |
| 287 | * @rq is being submitted. Analyze what needs to be done and put it on the |
| 288 | * right queue. |
| 289 | * |
| 290 | * CONTEXT: |
| 291 | * spin_lock_irq(q->queue_lock) |
| 292 | */ |
| 293 | void blk_insert_flush(struct request *rq) |
| 294 | { |
| 295 | struct request_queue *q = rq->q; |
| 296 | unsigned int fflags = q->flush_flags; /* may change, cache */ |
| 297 | unsigned int policy = blk_flush_policy(fflags, rq); |
| 298 | |
| 299 | BUG_ON(rq->end_io); |
| 300 | BUG_ON(!rq->bio || rq->bio != rq->biotail); |
| 301 | |
| 302 | /* |
| 303 | * @policy now records what operations need to be done. Adjust |
| 304 | * REQ_FLUSH and FUA for the driver. |
| 305 | */ |
Tejun Heo | 4fed947 | 2010-09-03 11:56:17 +0200 | [diff] [blame] | 306 | rq->cmd_flags &= ~REQ_FLUSH; |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 307 | if (!(fflags & REQ_FUA)) |
Tejun Heo | 4fed947 | 2010-09-03 11:56:17 +0200 | [diff] [blame] | 308 | rq->cmd_flags &= ~REQ_FUA; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 309 | |
Tejun Heo | ae1b153 | 2011-01-25 12:43:54 +0100 | [diff] [blame] | 310 | /* |
| 311 | * If there's data but flush is not necessary, the request can be |
| 312 | * processed directly without going through flush machinery. Queue |
| 313 | * for normal execution. |
| 314 | */ |
| 315 | if ((policy & REQ_FSEQ_DATA) && |
| 316 | !(policy & (REQ_FSEQ_PREFLUSH | REQ_FSEQ_POSTFLUSH))) { |
| 317 | list_add(&rq->queuelist, &q->queue_head); |
| 318 | return; |
| 319 | } |
| 320 | |
| 321 | /* |
| 322 | * @rq should go through flush machinery. Mark it part of flush |
| 323 | * sequence and submit for further processing. |
| 324 | */ |
| 325 | memset(&rq->flush, 0, sizeof(rq->flush)); |
| 326 | INIT_LIST_HEAD(&rq->flush.list); |
| 327 | rq->cmd_flags |= REQ_FLUSH_SEQ; |
| 328 | rq->end_io = flush_data_end_io; |
| 329 | |
| 330 | blk_flush_complete_seq(rq, REQ_FSEQ_ACTIONS & ~policy, 0); |
| 331 | } |
| 332 | |
| 333 | /** |
| 334 | * blk_abort_flushes - @q is being aborted, abort flush requests |
| 335 | * @q: request_queue being aborted |
| 336 | * |
| 337 | * To be called from elv_abort_queue(). @q is being aborted. Prepare all |
| 338 | * FLUSH/FUA requests for abortion. |
| 339 | * |
| 340 | * CONTEXT: |
| 341 | * spin_lock_irq(q->queue_lock) |
| 342 | */ |
| 343 | void blk_abort_flushes(struct request_queue *q) |
| 344 | { |
| 345 | struct request *rq, *n; |
| 346 | int i; |
| 347 | |
| 348 | /* |
| 349 | * Requests in flight for data are already owned by the dispatch |
| 350 | * queue or the device driver. Just restore for normal completion. |
| 351 | */ |
| 352 | list_for_each_entry_safe(rq, n, &q->flush_data_in_flight, flush.list) { |
| 353 | list_del_init(&rq->flush.list); |
| 354 | blk_flush_restore_request(rq); |
| 355 | } |
| 356 | |
| 357 | /* |
| 358 | * We need to give away requests on flush queues. Restore for |
| 359 | * normal completion and put them on the dispatch queue. |
| 360 | */ |
| 361 | for (i = 0; i < ARRAY_SIZE(q->flush_queue); i++) { |
| 362 | list_for_each_entry_safe(rq, n, &q->flush_queue[i], |
| 363 | flush.list) { |
| 364 | list_del_init(&rq->flush.list); |
| 365 | blk_flush_restore_request(rq); |
| 366 | list_add_tail(&rq->queuelist, &q->queue_head); |
| 367 | } |
| 368 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 369 | } |
| 370 | |
Tejun Heo | d391a2d | 2010-09-03 11:56:17 +0200 | [diff] [blame] | 371 | static void bio_end_flush(struct bio *bio, int err) |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 372 | { |
Tejun Heo | d391a2d | 2010-09-03 11:56:17 +0200 | [diff] [blame] | 373 | if (err) |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 374 | clear_bit(BIO_UPTODATE, &bio->bi_flags); |
Dmitry Monakhov | f17e232 | 2010-04-28 17:55:07 +0400 | [diff] [blame] | 375 | if (bio->bi_private) |
| 376 | complete(bio->bi_private); |
| 377 | bio_put(bio); |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 378 | } |
| 379 | |
| 380 | /** |
| 381 | * blkdev_issue_flush - queue a flush |
| 382 | * @bdev: blockdev to issue flush for |
Dmitry Monakhov | fbd9b09 | 2010-04-28 17:55:06 +0400 | [diff] [blame] | 383 | * @gfp_mask: memory allocation flags (for bio_alloc) |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 384 | * @error_sector: error sector |
| 385 | * |
| 386 | * Description: |
| 387 | * Issue a flush for the block device in question. Caller can supply |
| 388 | * room for storing the error offset in case of a flush error, if they |
Dmitry Monakhov | f17e232 | 2010-04-28 17:55:07 +0400 | [diff] [blame] | 389 | * wish to. If WAIT flag is not passed then caller may check only what |
| 390 | * request was pushed in some internal queue for later handling. |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 391 | */ |
Dmitry Monakhov | fbd9b09 | 2010-04-28 17:55:06 +0400 | [diff] [blame] | 392 | int blkdev_issue_flush(struct block_device *bdev, gfp_t gfp_mask, |
Christoph Hellwig | dd3932e | 2010-09-16 20:51:46 +0200 | [diff] [blame] | 393 | sector_t *error_sector) |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 394 | { |
| 395 | DECLARE_COMPLETION_ONSTACK(wait); |
| 396 | struct request_queue *q; |
| 397 | struct bio *bio; |
Dmitry Monakhov | fbd9b09 | 2010-04-28 17:55:06 +0400 | [diff] [blame] | 398 | int ret = 0; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 399 | |
| 400 | if (bdev->bd_disk == NULL) |
| 401 | return -ENXIO; |
| 402 | |
| 403 | q = bdev_get_queue(bdev); |
| 404 | if (!q) |
| 405 | return -ENXIO; |
| 406 | |
Dave Chinner | f10d9f6 | 2010-07-13 17:50:50 +1000 | [diff] [blame] | 407 | /* |
| 408 | * some block devices may not have their queue correctly set up here |
| 409 | * (e.g. loop device without a backing file) and so issuing a flush |
| 410 | * here will panic. Ensure there is a request function before issuing |
Tejun Heo | d391a2d | 2010-09-03 11:56:17 +0200 | [diff] [blame] | 411 | * the flush. |
Dave Chinner | f10d9f6 | 2010-07-13 17:50:50 +1000 | [diff] [blame] | 412 | */ |
| 413 | if (!q->make_request_fn) |
| 414 | return -ENXIO; |
| 415 | |
Dmitry Monakhov | fbd9b09 | 2010-04-28 17:55:06 +0400 | [diff] [blame] | 416 | bio = bio_alloc(gfp_mask, 0); |
Tejun Heo | d391a2d | 2010-09-03 11:56:17 +0200 | [diff] [blame] | 417 | bio->bi_end_io = bio_end_flush; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 418 | bio->bi_bdev = bdev; |
Christoph Hellwig | dd3932e | 2010-09-16 20:51:46 +0200 | [diff] [blame] | 419 | bio->bi_private = &wait; |
Dmitry Monakhov | f17e232 | 2010-04-28 17:55:07 +0400 | [diff] [blame] | 420 | |
| 421 | bio_get(bio); |
Tejun Heo | d391a2d | 2010-09-03 11:56:17 +0200 | [diff] [blame] | 422 | submit_bio(WRITE_FLUSH, bio); |
Christoph Hellwig | dd3932e | 2010-09-16 20:51:46 +0200 | [diff] [blame] | 423 | wait_for_completion(&wait); |
| 424 | |
| 425 | /* |
| 426 | * The driver must store the error location in ->bi_sector, if |
| 427 | * it supports it. For non-stacked drivers, this should be |
| 428 | * copied from blk_rq_pos(rq). |
| 429 | */ |
| 430 | if (error_sector) |
| 431 | *error_sector = bio->bi_sector; |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 432 | |
Tejun Heo | d391a2d | 2010-09-03 11:56:17 +0200 | [diff] [blame] | 433 | if (!bio_flagged(bio, BIO_UPTODATE)) |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 434 | ret = -EIO; |
| 435 | |
| 436 | bio_put(bio); |
| 437 | return ret; |
| 438 | } |
Jens Axboe | 86db1e2 | 2008-01-29 14:53:40 +0100 | [diff] [blame] | 439 | EXPORT_SYMBOL(blkdev_issue_flush); |