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 /* stash away the original request */
174 elv_dequeue_request(q
, rq
);
179 * Queue ordered sequence. As we stack them at the head, we
180 * need to queue in reverse order. Note that we rely on that
181 * no fs request uses ELEVATOR_INSERT_FRONT and thus no fs
182 * request gets inbetween ordered sequence.
184 if (q
->ordered
& QUEUE_ORDERED_DO_POSTFLUSH
) {
185 queue_flush(q
, QUEUE_ORDERED_DO_POSTFLUSH
);
186 rq
= &q
->post_flush_rq
;
188 skip
|= QUEUE_ORDSEQ_POSTFLUSH
;
190 if (q
->ordered
& QUEUE_ORDERED_DO_BAR
) {
193 /* initialize proxy request and queue it */
195 if (bio_data_dir(q
->orig_bar_rq
->bio
) == WRITE
)
196 rq
->cmd_flags
|= REQ_RW
;
197 if (q
->ordered
& QUEUE_ORDERED_DO_FUA
)
198 rq
->cmd_flags
|= REQ_FUA
;
199 init_request_from_bio(rq
, q
->orig_bar_rq
->bio
);
200 rq
->end_io
= bar_end_io
;
202 elv_insert(q
, rq
, ELEVATOR_INSERT_FRONT
);
204 skip
|= QUEUE_ORDSEQ_BAR
;
206 if (q
->ordered
& QUEUE_ORDERED_DO_PREFLUSH
) {
207 queue_flush(q
, QUEUE_ORDERED_DO_PREFLUSH
);
208 rq
= &q
->pre_flush_rq
;
210 skip
|= QUEUE_ORDSEQ_PREFLUSH
;
212 if ((q
->ordered
& QUEUE_ORDERED_BY_DRAIN
) && q
->in_flight
)
215 skip
|= QUEUE_ORDSEQ_DRAIN
;
220 * Complete skipped sequences. If whole sequence is complete,
221 * return false to tell elevator that this request is gone.
223 return !blk_ordered_complete_seq(q
, skip
, 0);
226 bool blk_do_ordered(struct request_queue
*q
, struct request
**rqp
)
228 struct request
*rq
= *rqp
;
229 const int is_barrier
= blk_fs_request(rq
) && blk_barrier_rq(rq
);
235 if (q
->next_ordered
!= QUEUE_ORDERED_NONE
)
236 return start_ordered(q
, rqp
);
239 * Queue ordering not supported. Terminate
242 elv_dequeue_request(q
, rq
);
243 if (__blk_end_request(rq
, -EOPNOTSUPP
,
252 * Ordered sequence in progress
255 /* Special requests are not subject to ordering rules. */
256 if (!blk_fs_request(rq
) &&
257 rq
!= &q
->pre_flush_rq
&& rq
!= &q
->post_flush_rq
)
260 if (q
->ordered
& QUEUE_ORDERED_BY_TAG
) {
261 /* Ordered by tag. Blocking the next barrier is enough. */
262 if (is_barrier
&& rq
!= &q
->bar_rq
)
265 /* Ordered by draining. Wait for turn. */
266 WARN_ON(blk_ordered_req_seq(rq
) < blk_ordered_cur_seq(q
));
267 if (blk_ordered_req_seq(rq
) > blk_ordered_cur_seq(q
))
274 static void bio_end_empty_barrier(struct bio
*bio
, int err
)
277 if (err
== -EOPNOTSUPP
)
278 set_bit(BIO_EOPNOTSUPP
, &bio
->bi_flags
);
279 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
282 complete(bio
->bi_private
);
286 * blkdev_issue_flush - queue a flush
287 * @bdev: blockdev to issue flush for
288 * @error_sector: error sector
291 * Issue a flush for the block device in question. Caller can supply
292 * room for storing the error offset in case of a flush error, if they
293 * wish to. Caller must run wait_for_completion() on its own.
295 int blkdev_issue_flush(struct block_device
*bdev
, sector_t
*error_sector
)
297 DECLARE_COMPLETION_ONSTACK(wait
);
298 struct request_queue
*q
;
302 if (bdev
->bd_disk
== NULL
)
305 q
= bdev_get_queue(bdev
);
309 bio
= bio_alloc(GFP_KERNEL
, 0);
313 bio
->bi_end_io
= bio_end_empty_barrier
;
314 bio
->bi_private
= &wait
;
316 submit_bio(WRITE_BARRIER
, bio
);
318 wait_for_completion(&wait
);
321 * The driver must store the error location in ->bi_sector, if
322 * it supports it. For non-stacked drivers, this should be copied
326 *error_sector
= bio
->bi_sector
;
329 if (bio_flagged(bio
, BIO_EOPNOTSUPP
))
331 else if (!bio_flagged(bio
, BIO_UPTODATE
))
337 EXPORT_SYMBOL(blkdev_issue_flush
);
339 static void blkdev_discard_end_io(struct bio
*bio
, int err
)
342 if (err
== -EOPNOTSUPP
)
343 set_bit(BIO_EOPNOTSUPP
, &bio
->bi_flags
);
344 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
351 * blkdev_issue_discard - queue a discard
352 * @bdev: blockdev to issue discard for
353 * @sector: start sector
354 * @nr_sects: number of sectors to discard
355 * @gfp_mask: memory allocation flags (for bio_alloc)
358 * Issue a discard request for the sectors in question. Does not wait.
360 int blkdev_issue_discard(struct block_device
*bdev
,
361 sector_t sector
, sector_t nr_sects
, gfp_t gfp_mask
)
363 struct request_queue
*q
;
367 if (bdev
->bd_disk
== NULL
)
370 q
= bdev_get_queue(bdev
);
374 if (!q
->prepare_discard_fn
)
377 while (nr_sects
&& !ret
) {
378 bio
= bio_alloc(gfp_mask
, 0);
382 bio
->bi_end_io
= blkdev_discard_end_io
;
385 bio
->bi_sector
= sector
;
387 if (nr_sects
> q
->max_hw_sectors
) {
388 bio
->bi_size
= q
->max_hw_sectors
<< 9;
389 nr_sects
-= q
->max_hw_sectors
;
390 sector
+= q
->max_hw_sectors
;
392 bio
->bi_size
= nr_sects
<< 9;
396 submit_bio(DISCARD_BARRIER
, bio
);
398 /* Check if it failed immediately */
399 if (bio_flagged(bio
, BIO_EOPNOTSUPP
))
401 else if (!bio_flagged(bio
, BIO_UPTODATE
))
407 EXPORT_SYMBOL(blkdev_issue_discard
);