2 * Functions related to barrier IO handling
4 #include <linux/kernel.h>
5 #include <linux/module.h>
7 #include <linux/blkdev.h>
12 * blk_queue_ordered - does this queue support ordered writes
13 * @q: the request queue
14 * @ordered: one of QUEUE_ORDERED_*
15 * @prepare_flush_fn: rq setup helper for cache flush ordered writes
18 * For journalled file systems, doing ordered writes on a commit
19 * block instead of explicitly doing wait_on_buffer (which is bad
20 * for performance) can be a big win. Block drivers supporting this
21 * feature should call this function and indicate so.
24 int blk_queue_ordered(struct request_queue *q, unsigned ordered,
25 prepare_flush_fn *prepare_flush_fn)
27 if (!prepare_flush_fn && (ordered & (QUEUE_ORDERED_DO_PREFLUSH |
28 QUEUE_ORDERED_DO_POSTFLUSH))) {
29 printk(KERN_ERR "%s: prepare_flush_fn required\n", __func__);
33 if (ordered != QUEUE_ORDERED_NONE &&
34 ordered != QUEUE_ORDERED_DRAIN &&
35 ordered != QUEUE_ORDERED_DRAIN_FLUSH &&
36 ordered != QUEUE_ORDERED_DRAIN_FUA &&
37 ordered != QUEUE_ORDERED_TAG &&
38 ordered != QUEUE_ORDERED_TAG_FLUSH &&
39 ordered != QUEUE_ORDERED_TAG_FUA) {
40 printk(KERN_ERR "blk_queue_ordered: bad value %d\n", ordered);
45 q->next_ordered = ordered;
46 q->prepare_flush_fn = prepare_flush_fn;
50 EXPORT_SYMBOL(blk_queue_ordered);
53 * Cache flushing for ordered writes handling
55 unsigned blk_ordered_cur_seq(struct request_queue *q)
59 return 1 << ffz(q->ordseq);
62 unsigned blk_ordered_req_seq(struct request *rq)
64 struct request_queue *q = rq->q;
66 BUG_ON(q->ordseq == 0);
68 if (rq == &q->pre_flush_rq)
69 return QUEUE_ORDSEQ_PREFLUSH;
71 return QUEUE_ORDSEQ_BAR;
72 if (rq == &q->post_flush_rq)
73 return QUEUE_ORDSEQ_POSTFLUSH;
76 * !fs requests don't need to follow barrier ordering. Always
77 * put them at the front. This fixes the following deadlock.
79 * http://thread.gmane.org/gmane.linux.kernel/537473
81 if (!blk_fs_request(rq))
82 return QUEUE_ORDSEQ_DRAIN;
84 if ((rq->cmd_flags & REQ_ORDERED_COLOR) ==
85 (q->orig_bar_rq->cmd_flags & REQ_ORDERED_COLOR))
86 return QUEUE_ORDSEQ_DRAIN;
88 return QUEUE_ORDSEQ_DONE;
91 bool blk_ordered_complete_seq(struct request_queue *q, unsigned seq, int error)
95 if (error && !q->orderr)
98 BUG_ON(q->ordseq & seq);
101 if (blk_ordered_cur_seq(q) != QUEUE_ORDSEQ_DONE)
105 * Okay, sequence complete.
110 if (__blk_end_request(rq, q->orderr, blk_rq_bytes(rq)))
116 static void pre_flush_end_io(struct request *rq, int error)
118 elv_completed_request(rq->q, rq);
119 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_PREFLUSH, error);
122 static void bar_end_io(struct request *rq, int error)
124 elv_completed_request(rq->q, rq);
125 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_BAR, error);
128 static void post_flush_end_io(struct request *rq, int error)
130 elv_completed_request(rq->q, rq);
131 blk_ordered_complete_seq(rq->q, QUEUE_ORDSEQ_POSTFLUSH, error);
134 static void queue_flush(struct request_queue *q, unsigned which)
137 rq_end_io_fn *end_io;
139 if (which == QUEUE_ORDERED_DO_PREFLUSH) {
140 rq = &q->pre_flush_rq;
141 end_io = pre_flush_end_io;
143 rq = &q->post_flush_rq;
144 end_io = post_flush_end_io;
148 rq->cmd_flags = REQ_HARDBARRIER;
149 rq->rq_disk = q->bar_rq.rq_disk;
151 q->prepare_flush_fn(q, rq);
153 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
156 static inline bool start_ordered(struct request_queue *q, struct request **rqp)
158 struct request *rq = *rqp;
162 q->ordered = q->next_ordered;
163 q->ordseq |= QUEUE_ORDSEQ_STARTED;
166 * For an empty barrier, there's no actual BAR request, which
167 * in turn makes POSTFLUSH unnecessary. Mask them off.
169 if (!rq->hard_nr_sectors) {
170 q->ordered &= ~(QUEUE_ORDERED_DO_BAR |
171 QUEUE_ORDERED_DO_POSTFLUSH);
173 * Empty barrier on a write-through device w/ ordered
174 * tag has no command to issue and without any command
175 * to issue, ordering by tag can't be used. Drain
178 if ((q->ordered & QUEUE_ORDERED_BY_TAG) &&
179 !(q->ordered & QUEUE_ORDERED_DO_PREFLUSH)) {
180 q->ordered &= ~QUEUE_ORDERED_BY_TAG;
181 q->ordered |= QUEUE_ORDERED_BY_DRAIN;
185 /* stash away the original request */
186 elv_dequeue_request(q, rq);
191 * Queue ordered sequence. As we stack them at the head, we
192 * need to queue in reverse order. Note that we rely on that
193 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
194 * request gets inbetween ordered sequence.
196 if (q->ordered & QUEUE_ORDERED_DO_POSTFLUSH) {
197 queue_flush(q, QUEUE_ORDERED_DO_POSTFLUSH);
198 rq = &q->post_flush_rq;
200 skip |= QUEUE_ORDSEQ_POSTFLUSH;
202 if (q->ordered & QUEUE_ORDERED_DO_BAR) {
205 /* initialize proxy request and queue it */
207 if (bio_data_dir(q->orig_bar_rq->bio) == WRITE)
208 rq->cmd_flags |= REQ_RW;
209 if (q->ordered & QUEUE_ORDERED_DO_FUA)
210 rq->cmd_flags |= REQ_FUA;
211 init_request_from_bio(rq, q->orig_bar_rq->bio);
212 rq->end_io = bar_end_io;
214 elv_insert(q, rq, ELEVATOR_INSERT_FRONT);
216 skip |= QUEUE_ORDSEQ_BAR;
218 if (q->ordered & QUEUE_ORDERED_DO_PREFLUSH) {
219 queue_flush(q, QUEUE_ORDERED_DO_PREFLUSH);
220 rq = &q->pre_flush_rq;
222 skip |= QUEUE_ORDSEQ_PREFLUSH;
224 if ((q->ordered & QUEUE_ORDERED_BY_DRAIN) && q->in_flight)
227 skip |= QUEUE_ORDSEQ_DRAIN;
232 * Complete skipped sequences. If whole sequence is complete,
233 * return false to tell elevator that this request is gone.
235 return !blk_ordered_complete_seq(q, skip, 0);
238 bool blk_do_ordered(struct request_queue *q, struct request **rqp)
240 struct request *rq = *rqp;
241 const int is_barrier = blk_fs_request(rq) && blk_barrier_rq(rq);
247 if (q->next_ordered != QUEUE_ORDERED_NONE)
248 return start_ordered(q, rqp);
251 * Queue ordering not supported. Terminate
254 elv_dequeue_request(q, rq);
255 if (__blk_end_request(rq, -EOPNOTSUPP,
264 * Ordered sequence in progress
267 /* Special requests are not subject to ordering rules. */
268 if (!blk_fs_request(rq) &&
269 rq != &q->pre_flush_rq && rq != &q->post_flush_rq)
272 if (q->ordered & QUEUE_ORDERED_BY_TAG) {
273 /* Ordered by tag. Blocking the next barrier is enough. */
274 if (is_barrier && rq != &q->bar_rq)
277 /* Ordered by draining. Wait for turn. */
278 WARN_ON(blk_ordered_req_seq(rq) < blk_ordered_cur_seq(q));
279 if (blk_ordered_req_seq(rq) > blk_ordered_cur_seq(q))
286 static void bio_end_empty_barrier(struct bio *bio, int err)
289 if (err == -EOPNOTSUPP)
290 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
291 clear_bit(BIO_UPTODATE, &bio->bi_flags);
294 complete(bio->bi_private);
298 * blkdev_issue_flush - queue a flush
299 * @bdev: blockdev to issue flush for
300 * @error_sector: error sector
303 * Issue a flush for the block device in question. Caller can supply
304 * room for storing the error offset in case of a flush error, if they
305 * wish to. Caller must run wait_for_completion() on its own.
307 int blkdev_issue_flush(struct block_device *bdev, sector_t *error_sector)
309 DECLARE_COMPLETION_ONSTACK(wait);
310 struct request_queue *q;
314 if (bdev->bd_disk == NULL)
317 q = bdev_get_queue(bdev);
321 bio = bio_alloc(GFP_KERNEL, 0);
325 bio->bi_end_io = bio_end_empty_barrier;
326 bio->bi_private = &wait;
328 submit_bio(WRITE_BARRIER, bio);
330 wait_for_completion(&wait);
333 * The driver must store the error location in ->bi_sector, if
334 * it supports it. For non-stacked drivers, this should be copied
338 *error_sector = bio->bi_sector;
341 if (bio_flagged(bio, BIO_EOPNOTSUPP))
343 else if (!bio_flagged(bio, BIO_UPTODATE))
349 EXPORT_SYMBOL(blkdev_issue_flush);
351 static void blkdev_discard_end_io(struct bio *bio, int err)
354 if (err == -EOPNOTSUPP)
355 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
356 clear_bit(BIO_UPTODATE, &bio->bi_flags);
363 * blkdev_issue_discard - queue a discard
364 * @bdev: blockdev to issue discard for
365 * @sector: start sector
366 * @nr_sects: number of sectors to discard
367 * @gfp_mask: memory allocation flags (for bio_alloc)
370 * Issue a discard request for the sectors in question. Does not wait.
372 int blkdev_issue_discard(struct block_device *bdev,
373 sector_t sector, sector_t nr_sects, gfp_t gfp_mask)
375 struct request_queue *q;
379 if (bdev->bd_disk == NULL)
382 q = bdev_get_queue(bdev);
386 if (!q->prepare_discard_fn)
389 while (nr_sects && !ret) {
390 bio = bio_alloc(gfp_mask, 0);
394 bio->bi_end_io = blkdev_discard_end_io;
397 bio->bi_sector = sector;
399 if (nr_sects > q->max_hw_sectors) {
400 bio->bi_size = q->max_hw_sectors << 9;
401 nr_sects -= q->max_hw_sectors;
402 sector += q->max_hw_sectors;
404 bio->bi_size = nr_sects << 9;
408 submit_bio(DISCARD_BARRIER, bio);
410 /* Check if it failed immediately */
411 if (bio_flagged(bio, BIO_EOPNOTSUPP))
413 else if (!bio_flagged(bio, BIO_UPTODATE))
419 EXPORT_SYMBOL(blkdev_issue_discard);