]>
git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - block/bio.c
2 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
13 * You should have received a copy of the GNU General Public Licens
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
19 #include <linux/swap.h>
20 #include <linux/bio.h>
21 #include <linux/blkdev.h>
22 #include <linux/uio.h>
23 #include <linux/iocontext.h>
24 #include <linux/slab.h>
25 #include <linux/init.h>
26 #include <linux/kernel.h>
27 #include <linux/export.h>
28 #include <linux/mempool.h>
29 #include <linux/workqueue.h>
30 #include <linux/cgroup.h>
32 #include <trace/events/block.h>
35 * Test patch to inline a certain number of bi_io_vec's inside the bio
36 * itself, to shrink a bio data allocation from two mempool calls to one
38 #define BIO_INLINE_VECS 4
41 * if you change this list, also change bvec_alloc or things will
42 * break badly! cannot be bigger than what you can fit into an
45 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
46 static struct biovec_slab bvec_slabs
[BIOVEC_NR_POOLS
] __read_mostly
= {
47 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES
),
52 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
53 * IO code that does not need private memory pools.
55 struct bio_set
*fs_bio_set
;
56 EXPORT_SYMBOL(fs_bio_set
);
59 * Our slab pool management
62 struct kmem_cache
*slab
;
63 unsigned int slab_ref
;
64 unsigned int slab_size
;
67 static DEFINE_MUTEX(bio_slab_lock
);
68 static struct bio_slab
*bio_slabs
;
69 static unsigned int bio_slab_nr
, bio_slab_max
;
71 static struct kmem_cache
*bio_find_or_create_slab(unsigned int extra_size
)
73 unsigned int sz
= sizeof(struct bio
) + extra_size
;
74 struct kmem_cache
*slab
= NULL
;
75 struct bio_slab
*bslab
, *new_bio_slabs
;
76 unsigned int new_bio_slab_max
;
77 unsigned int i
, entry
= -1;
79 mutex_lock(&bio_slab_lock
);
82 while (i
< bio_slab_nr
) {
83 bslab
= &bio_slabs
[i
];
85 if (!bslab
->slab
&& entry
== -1)
87 else if (bslab
->slab_size
== sz
) {
98 if (bio_slab_nr
== bio_slab_max
&& entry
== -1) {
99 new_bio_slab_max
= bio_slab_max
<< 1;
100 new_bio_slabs
= krealloc(bio_slabs
,
101 new_bio_slab_max
* sizeof(struct bio_slab
),
105 bio_slab_max
= new_bio_slab_max
;
106 bio_slabs
= new_bio_slabs
;
109 entry
= bio_slab_nr
++;
111 bslab
= &bio_slabs
[entry
];
113 snprintf(bslab
->name
, sizeof(bslab
->name
), "bio-%d", entry
);
114 slab
= kmem_cache_create(bslab
->name
, sz
, ARCH_KMALLOC_MINALIGN
,
115 SLAB_HWCACHE_ALIGN
, NULL
);
121 bslab
->slab_size
= sz
;
123 mutex_unlock(&bio_slab_lock
);
127 static void bio_put_slab(struct bio_set
*bs
)
129 struct bio_slab
*bslab
= NULL
;
132 mutex_lock(&bio_slab_lock
);
134 for (i
= 0; i
< bio_slab_nr
; i
++) {
135 if (bs
->bio_slab
== bio_slabs
[i
].slab
) {
136 bslab
= &bio_slabs
[i
];
141 if (WARN(!bslab
, KERN_ERR
"bio: unable to find slab!\n"))
144 WARN_ON(!bslab
->slab_ref
);
146 if (--bslab
->slab_ref
)
149 kmem_cache_destroy(bslab
->slab
);
153 mutex_unlock(&bio_slab_lock
);
156 unsigned int bvec_nr_vecs(unsigned short idx
)
158 return bvec_slabs
[idx
].nr_vecs
;
161 void bvec_free(mempool_t
*pool
, struct bio_vec
*bv
, unsigned int idx
)
163 BIO_BUG_ON(idx
>= BIOVEC_NR_POOLS
);
165 if (idx
== BIOVEC_MAX_IDX
)
166 mempool_free(bv
, pool
);
168 struct biovec_slab
*bvs
= bvec_slabs
+ idx
;
170 kmem_cache_free(bvs
->slab
, bv
);
174 struct bio_vec
*bvec_alloc(gfp_t gfp_mask
, int nr
, unsigned long *idx
,
180 * see comment near bvec_array define!
198 case 129 ... BIO_MAX_PAGES
:
206 * idx now points to the pool we want to allocate from. only the
207 * 1-vec entry pool is mempool backed.
209 if (*idx
== BIOVEC_MAX_IDX
) {
211 bvl
= mempool_alloc(pool
, gfp_mask
);
213 struct biovec_slab
*bvs
= bvec_slabs
+ *idx
;
214 gfp_t __gfp_mask
= gfp_mask
& ~(__GFP_WAIT
| __GFP_IO
);
217 * Make this allocation restricted and don't dump info on
218 * allocation failures, since we'll fallback to the mempool
219 * in case of failure.
221 __gfp_mask
|= __GFP_NOMEMALLOC
| __GFP_NORETRY
| __GFP_NOWARN
;
224 * Try a slab allocation. If this fails and __GFP_WAIT
225 * is set, retry with the 1-entry mempool
227 bvl
= kmem_cache_alloc(bvs
->slab
, __gfp_mask
);
228 if (unlikely(!bvl
&& (gfp_mask
& __GFP_WAIT
))) {
229 *idx
= BIOVEC_MAX_IDX
;
237 static void __bio_free(struct bio
*bio
)
239 bio_disassociate_task(bio
);
241 if (bio_integrity(bio
))
242 bio_integrity_free(bio
);
245 static void bio_free(struct bio
*bio
)
247 struct bio_set
*bs
= bio
->bi_pool
;
253 if (bio_flagged(bio
, BIO_OWNS_VEC
))
254 bvec_free(bs
->bvec_pool
, bio
->bi_io_vec
, BIO_POOL_IDX(bio
));
257 * If we have front padding, adjust the bio pointer before freeing
262 mempool_free(p
, bs
->bio_pool
);
264 /* Bio was allocated by bio_kmalloc() */
269 void bio_init(struct bio
*bio
)
271 memset(bio
, 0, sizeof(*bio
));
272 bio
->bi_flags
= 1 << BIO_UPTODATE
;
273 atomic_set(&bio
->bi_remaining
, 1);
274 atomic_set(&bio
->bi_cnt
, 1);
276 EXPORT_SYMBOL(bio_init
);
279 * bio_reset - reinitialize a bio
283 * After calling bio_reset(), @bio will be in the same state as a freshly
284 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
285 * preserved are the ones that are initialized by bio_alloc_bioset(). See
286 * comment in struct bio.
288 void bio_reset(struct bio
*bio
)
290 unsigned long flags
= bio
->bi_flags
& (~0UL << BIO_RESET_BITS
);
294 memset(bio
, 0, BIO_RESET_BYTES
);
295 bio
->bi_flags
= flags
|(1 << BIO_UPTODATE
);
296 atomic_set(&bio
->bi_remaining
, 1);
298 EXPORT_SYMBOL(bio_reset
);
300 static void bio_chain_endio(struct bio
*bio
, int error
)
302 bio_endio(bio
->bi_private
, error
);
307 * bio_chain - chain bio completions
308 * @bio: the target bio
309 * @parent: the @bio's parent bio
311 * The caller won't have a bi_end_io called when @bio completes - instead,
312 * @parent's bi_end_io won't be called until both @parent and @bio have
313 * completed; the chained bio will also be freed when it completes.
315 * The caller must not set bi_private or bi_end_io in @bio.
317 void bio_chain(struct bio
*bio
, struct bio
*parent
)
319 BUG_ON(bio
->bi_private
|| bio
->bi_end_io
);
321 bio
->bi_private
= parent
;
322 bio
->bi_end_io
= bio_chain_endio
;
323 atomic_inc(&parent
->bi_remaining
);
325 EXPORT_SYMBOL(bio_chain
);
327 static void bio_alloc_rescue(struct work_struct
*work
)
329 struct bio_set
*bs
= container_of(work
, struct bio_set
, rescue_work
);
333 spin_lock(&bs
->rescue_lock
);
334 bio
= bio_list_pop(&bs
->rescue_list
);
335 spin_unlock(&bs
->rescue_lock
);
340 generic_make_request(bio
);
344 static void punt_bios_to_rescuer(struct bio_set
*bs
)
346 struct bio_list punt
, nopunt
;
350 * In order to guarantee forward progress we must punt only bios that
351 * were allocated from this bio_set; otherwise, if there was a bio on
352 * there for a stacking driver higher up in the stack, processing it
353 * could require allocating bios from this bio_set, and doing that from
354 * our own rescuer would be bad.
356 * Since bio lists are singly linked, pop them all instead of trying to
357 * remove from the middle of the list:
360 bio_list_init(&punt
);
361 bio_list_init(&nopunt
);
363 while ((bio
= bio_list_pop(current
->bio_list
)))
364 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
366 *current
->bio_list
= nopunt
;
368 spin_lock(&bs
->rescue_lock
);
369 bio_list_merge(&bs
->rescue_list
, &punt
);
370 spin_unlock(&bs
->rescue_lock
);
372 queue_work(bs
->rescue_workqueue
, &bs
->rescue_work
);
376 * bio_alloc_bioset - allocate a bio for I/O
377 * @gfp_mask: the GFP_ mask given to the slab allocator
378 * @nr_iovecs: number of iovecs to pre-allocate
379 * @bs: the bio_set to allocate from.
382 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
383 * backed by the @bs's mempool.
385 * When @bs is not NULL, if %__GFP_WAIT is set then bio_alloc will always be
386 * able to allocate a bio. This is due to the mempool guarantees. To make this
387 * work, callers must never allocate more than 1 bio at a time from this pool.
388 * Callers that need to allocate more than 1 bio must always submit the
389 * previously allocated bio for IO before attempting to allocate a new one.
390 * Failure to do so can cause deadlocks under memory pressure.
392 * Note that when running under generic_make_request() (i.e. any block
393 * driver), bios are not submitted until after you return - see the code in
394 * generic_make_request() that converts recursion into iteration, to prevent
397 * This would normally mean allocating multiple bios under
398 * generic_make_request() would be susceptible to deadlocks, but we have
399 * deadlock avoidance code that resubmits any blocked bios from a rescuer
402 * However, we do not guarantee forward progress for allocations from other
403 * mempools. Doing multiple allocations from the same mempool under
404 * generic_make_request() should be avoided - instead, use bio_set's front_pad
405 * for per bio allocations.
408 * Pointer to new bio on success, NULL on failure.
410 struct bio
*bio_alloc_bioset(gfp_t gfp_mask
, int nr_iovecs
, struct bio_set
*bs
)
412 gfp_t saved_gfp
= gfp_mask
;
414 unsigned inline_vecs
;
415 unsigned long idx
= BIO_POOL_NONE
;
416 struct bio_vec
*bvl
= NULL
;
421 if (nr_iovecs
> UIO_MAXIOV
)
424 p
= kmalloc(sizeof(struct bio
) +
425 nr_iovecs
* sizeof(struct bio_vec
),
428 inline_vecs
= nr_iovecs
;
430 /* should not use nobvec bioset for nr_iovecs > 0 */
431 if (WARN_ON_ONCE(!bs
->bvec_pool
&& nr_iovecs
> 0))
434 * generic_make_request() converts recursion to iteration; this
435 * means if we're running beneath it, any bios we allocate and
436 * submit will not be submitted (and thus freed) until after we
439 * This exposes us to a potential deadlock if we allocate
440 * multiple bios from the same bio_set() while running
441 * underneath generic_make_request(). If we were to allocate
442 * multiple bios (say a stacking block driver that was splitting
443 * bios), we would deadlock if we exhausted the mempool's
446 * We solve this, and guarantee forward progress, with a rescuer
447 * workqueue per bio_set. If we go to allocate and there are
448 * bios on current->bio_list, we first try the allocation
449 * without __GFP_WAIT; if that fails, we punt those bios we
450 * would be blocking to the rescuer workqueue before we retry
451 * with the original gfp_flags.
454 if (current
->bio_list
&& !bio_list_empty(current
->bio_list
))
455 gfp_mask
&= ~__GFP_WAIT
;
457 p
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
458 if (!p
&& gfp_mask
!= saved_gfp
) {
459 punt_bios_to_rescuer(bs
);
460 gfp_mask
= saved_gfp
;
461 p
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
464 front_pad
= bs
->front_pad
;
465 inline_vecs
= BIO_INLINE_VECS
;
474 if (nr_iovecs
> inline_vecs
) {
475 bvl
= bvec_alloc(gfp_mask
, nr_iovecs
, &idx
, bs
->bvec_pool
);
476 if (!bvl
&& gfp_mask
!= saved_gfp
) {
477 punt_bios_to_rescuer(bs
);
478 gfp_mask
= saved_gfp
;
479 bvl
= bvec_alloc(gfp_mask
, nr_iovecs
, &idx
, bs
->bvec_pool
);
485 bio
->bi_flags
|= 1 << BIO_OWNS_VEC
;
486 } else if (nr_iovecs
) {
487 bvl
= bio
->bi_inline_vecs
;
491 bio
->bi_flags
|= idx
<< BIO_POOL_OFFSET
;
492 bio
->bi_max_vecs
= nr_iovecs
;
493 bio
->bi_io_vec
= bvl
;
497 mempool_free(p
, bs
->bio_pool
);
500 EXPORT_SYMBOL(bio_alloc_bioset
);
502 void zero_fill_bio(struct bio
*bio
)
506 struct bvec_iter iter
;
508 bio_for_each_segment(bv
, bio
, iter
) {
509 char *data
= bvec_kmap_irq(&bv
, &flags
);
510 memset(data
, 0, bv
.bv_len
);
511 flush_dcache_page(bv
.bv_page
);
512 bvec_kunmap_irq(data
, &flags
);
515 EXPORT_SYMBOL(zero_fill_bio
);
518 * bio_put - release a reference to a bio
519 * @bio: bio to release reference to
522 * Put a reference to a &struct bio, either one you have gotten with
523 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
525 void bio_put(struct bio
*bio
)
527 BIO_BUG_ON(!atomic_read(&bio
->bi_cnt
));
532 if (atomic_dec_and_test(&bio
->bi_cnt
))
535 EXPORT_SYMBOL(bio_put
);
537 inline int bio_phys_segments(struct request_queue
*q
, struct bio
*bio
)
539 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
540 blk_recount_segments(q
, bio
);
542 return bio
->bi_phys_segments
;
544 EXPORT_SYMBOL(bio_phys_segments
);
547 * __bio_clone_fast - clone a bio that shares the original bio's biovec
548 * @bio: destination bio
549 * @bio_src: bio to clone
551 * Clone a &bio. Caller will own the returned bio, but not
552 * the actual data it points to. Reference count of returned
555 * Caller must ensure that @bio_src is not freed before @bio.
557 void __bio_clone_fast(struct bio
*bio
, struct bio
*bio_src
)
559 BUG_ON(bio
->bi_pool
&& BIO_POOL_IDX(bio
) != BIO_POOL_NONE
);
562 * most users will be overriding ->bi_bdev with a new target,
563 * so we don't set nor calculate new physical/hw segment counts here
565 bio
->bi_bdev
= bio_src
->bi_bdev
;
566 bio
->bi_flags
|= 1 << BIO_CLONED
;
567 bio
->bi_rw
= bio_src
->bi_rw
;
568 bio
->bi_iter
= bio_src
->bi_iter
;
569 bio
->bi_io_vec
= bio_src
->bi_io_vec
;
571 EXPORT_SYMBOL(__bio_clone_fast
);
574 * bio_clone_fast - clone a bio that shares the original bio's biovec
576 * @gfp_mask: allocation priority
577 * @bs: bio_set to allocate from
579 * Like __bio_clone_fast, only also allocates the returned bio
581 struct bio
*bio_clone_fast(struct bio
*bio
, gfp_t gfp_mask
, struct bio_set
*bs
)
585 b
= bio_alloc_bioset(gfp_mask
, 0, bs
);
589 __bio_clone_fast(b
, bio
);
591 if (bio_integrity(bio
)) {
594 ret
= bio_integrity_clone(b
, bio
, gfp_mask
);
604 EXPORT_SYMBOL(bio_clone_fast
);
607 * bio_clone_bioset - clone a bio
608 * @bio_src: bio to clone
609 * @gfp_mask: allocation priority
610 * @bs: bio_set to allocate from
612 * Clone bio. Caller will own the returned bio, but not the actual data it
613 * points to. Reference count of returned bio will be one.
615 struct bio
*bio_clone_bioset(struct bio
*bio_src
, gfp_t gfp_mask
,
618 struct bvec_iter iter
;
623 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
624 * bio_src->bi_io_vec to bio->bi_io_vec.
626 * We can't do that anymore, because:
628 * - The point of cloning the biovec is to produce a bio with a biovec
629 * the caller can modify: bi_idx and bi_bvec_done should be 0.
631 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
632 * we tried to clone the whole thing bio_alloc_bioset() would fail.
633 * But the clone should succeed as long as the number of biovecs we
634 * actually need to allocate is fewer than BIO_MAX_PAGES.
636 * - Lastly, bi_vcnt should not be looked at or relied upon by code
637 * that does not own the bio - reason being drivers don't use it for
638 * iterating over the biovec anymore, so expecting it to be kept up
639 * to date (i.e. for clones that share the parent biovec) is just
640 * asking for trouble and would force extra work on
641 * __bio_clone_fast() anyways.
644 bio
= bio_alloc_bioset(gfp_mask
, bio_segments(bio_src
), bs
);
648 bio
->bi_bdev
= bio_src
->bi_bdev
;
649 bio
->bi_rw
= bio_src
->bi_rw
;
650 bio
->bi_iter
.bi_sector
= bio_src
->bi_iter
.bi_sector
;
651 bio
->bi_iter
.bi_size
= bio_src
->bi_iter
.bi_size
;
653 if (bio
->bi_rw
& REQ_DISCARD
)
654 goto integrity_clone
;
656 if (bio
->bi_rw
& REQ_WRITE_SAME
) {
657 bio
->bi_io_vec
[bio
->bi_vcnt
++] = bio_src
->bi_io_vec
[0];
658 goto integrity_clone
;
661 bio_for_each_segment(bv
, bio_src
, iter
)
662 bio
->bi_io_vec
[bio
->bi_vcnt
++] = bv
;
665 if (bio_integrity(bio_src
)) {
668 ret
= bio_integrity_clone(bio
, bio_src
, gfp_mask
);
677 EXPORT_SYMBOL(bio_clone_bioset
);
680 * bio_get_nr_vecs - return approx number of vecs
683 * Return the approximate number of pages we can send to this target.
684 * There's no guarantee that you will be able to fit this number of pages
685 * into a bio, it does not account for dynamic restrictions that vary
688 int bio_get_nr_vecs(struct block_device
*bdev
)
690 struct request_queue
*q
= bdev_get_queue(bdev
);
693 nr_pages
= min_t(unsigned,
694 queue_max_segments(q
),
695 queue_max_sectors(q
) / (PAGE_SIZE
>> 9) + 1);
697 return min_t(unsigned, nr_pages
, BIO_MAX_PAGES
);
700 EXPORT_SYMBOL(bio_get_nr_vecs
);
702 static int __bio_add_page(struct request_queue
*q
, struct bio
*bio
, struct page
703 *page
, unsigned int len
, unsigned int offset
,
704 unsigned int max_sectors
)
706 int retried_segments
= 0;
707 struct bio_vec
*bvec
;
710 * cloned bio must not modify vec list
712 if (unlikely(bio_flagged(bio
, BIO_CLONED
)))
715 if (((bio
->bi_iter
.bi_size
+ len
) >> 9) > max_sectors
)
719 * For filesystems with a blocksize smaller than the pagesize
720 * we will often be called with the same page as last time and
721 * a consecutive offset. Optimize this special case.
723 if (bio
->bi_vcnt
> 0) {
724 struct bio_vec
*prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
726 if (page
== prev
->bv_page
&&
727 offset
== prev
->bv_offset
+ prev
->bv_len
) {
728 unsigned int prev_bv_len
= prev
->bv_len
;
731 if (q
->merge_bvec_fn
) {
732 struct bvec_merge_data bvm
= {
733 /* prev_bvec is already charged in
734 bi_size, discharge it in order to
735 simulate merging updated prev_bvec
737 .bi_bdev
= bio
->bi_bdev
,
738 .bi_sector
= bio
->bi_iter
.bi_sector
,
739 .bi_size
= bio
->bi_iter
.bi_size
-
744 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < prev
->bv_len
) {
750 bio
->bi_iter
.bi_size
+= len
;
755 * If the queue doesn't support SG gaps and adding this
756 * offset would create a gap, disallow it.
758 if (q
->queue_flags
& (1 << QUEUE_FLAG_SG_GAPS
) &&
759 bvec_gap_to_prev(prev
, offset
))
763 if (bio
->bi_vcnt
>= bio
->bi_max_vecs
)
767 * setup the new entry, we might clear it again later if we
768 * cannot add the page
770 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
771 bvec
->bv_page
= page
;
773 bvec
->bv_offset
= offset
;
775 bio
->bi_phys_segments
++;
776 bio
->bi_iter
.bi_size
+= len
;
779 * Perform a recount if the number of segments is greater
780 * than queue_max_segments(q).
783 while (bio
->bi_phys_segments
> queue_max_segments(q
)) {
785 if (retried_segments
)
788 retried_segments
= 1;
789 blk_recount_segments(q
, bio
);
793 * if queue has other restrictions (eg varying max sector size
794 * depending on offset), it can specify a merge_bvec_fn in the
795 * queue to get further control
797 if (q
->merge_bvec_fn
) {
798 struct bvec_merge_data bvm
= {
799 .bi_bdev
= bio
->bi_bdev
,
800 .bi_sector
= bio
->bi_iter
.bi_sector
,
801 .bi_size
= bio
->bi_iter
.bi_size
- len
,
806 * merge_bvec_fn() returns number of bytes it can accept
809 if (q
->merge_bvec_fn(q
, &bvm
, bvec
) < bvec
->bv_len
)
813 /* If we may be able to merge these biovecs, force a recount */
814 if (bio
->bi_vcnt
> 1 && (BIOVEC_PHYS_MERGEABLE(bvec
-1, bvec
)))
815 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
821 bvec
->bv_page
= NULL
;
825 bio
->bi_iter
.bi_size
-= len
;
826 blk_recount_segments(q
, bio
);
831 * bio_add_pc_page - attempt to add page to bio
832 * @q: the target queue
833 * @bio: destination bio
835 * @len: vec entry length
836 * @offset: vec entry offset
838 * Attempt to add a page to the bio_vec maplist. This can fail for a
839 * number of reasons, such as the bio being full or target block device
840 * limitations. The target block device must allow bio's up to PAGE_SIZE,
841 * so it is always possible to add a single page to an empty bio.
843 * This should only be used by REQ_PC bios.
845 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
, struct page
*page
,
846 unsigned int len
, unsigned int offset
)
848 return __bio_add_page(q
, bio
, page
, len
, offset
,
849 queue_max_hw_sectors(q
));
851 EXPORT_SYMBOL(bio_add_pc_page
);
854 * bio_add_page - attempt to add page to bio
855 * @bio: destination bio
857 * @len: vec entry length
858 * @offset: vec entry offset
860 * Attempt to add a page to the bio_vec maplist. This can fail for a
861 * number of reasons, such as the bio being full or target block device
862 * limitations. The target block device must allow bio's up to PAGE_SIZE,
863 * so it is always possible to add a single page to an empty bio.
865 int bio_add_page(struct bio
*bio
, struct page
*page
, unsigned int len
,
868 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
869 unsigned int max_sectors
;
871 max_sectors
= blk_max_size_offset(q
, bio
->bi_iter
.bi_sector
);
872 if ((max_sectors
< (len
>> 9)) && !bio
->bi_iter
.bi_size
)
873 max_sectors
= len
>> 9;
875 return __bio_add_page(q
, bio
, page
, len
, offset
, max_sectors
);
877 EXPORT_SYMBOL(bio_add_page
);
879 struct submit_bio_ret
{
880 struct completion event
;
884 static void submit_bio_wait_endio(struct bio
*bio
, int error
)
886 struct submit_bio_ret
*ret
= bio
->bi_private
;
889 complete(&ret
->event
);
893 * submit_bio_wait - submit a bio, and wait until it completes
894 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
895 * @bio: The &struct bio which describes the I/O
897 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
898 * bio_endio() on failure.
900 int submit_bio_wait(int rw
, struct bio
*bio
)
902 struct submit_bio_ret ret
;
905 init_completion(&ret
.event
);
906 bio
->bi_private
= &ret
;
907 bio
->bi_end_io
= submit_bio_wait_endio
;
909 wait_for_completion(&ret
.event
);
913 EXPORT_SYMBOL(submit_bio_wait
);
916 * bio_advance - increment/complete a bio by some number of bytes
917 * @bio: bio to advance
918 * @bytes: number of bytes to complete
920 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
921 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
922 * be updated on the last bvec as well.
924 * @bio will then represent the remaining, uncompleted portion of the io.
926 void bio_advance(struct bio
*bio
, unsigned bytes
)
928 if (bio_integrity(bio
))
929 bio_integrity_advance(bio
, bytes
);
931 bio_advance_iter(bio
, &bio
->bi_iter
, bytes
);
933 EXPORT_SYMBOL(bio_advance
);
936 * bio_alloc_pages - allocates a single page for each bvec in a bio
937 * @bio: bio to allocate pages for
938 * @gfp_mask: flags for allocation
940 * Allocates pages up to @bio->bi_vcnt.
942 * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
945 int bio_alloc_pages(struct bio
*bio
, gfp_t gfp_mask
)
950 bio_for_each_segment_all(bv
, bio
, i
) {
951 bv
->bv_page
= alloc_page(gfp_mask
);
953 while (--bv
>= bio
->bi_io_vec
)
954 __free_page(bv
->bv_page
);
961 EXPORT_SYMBOL(bio_alloc_pages
);
964 * bio_copy_data - copy contents of data buffers from one chain of bios to
966 * @src: source bio list
967 * @dst: destination bio list
969 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
970 * @src and @dst as linked lists of bios.
972 * Stops when it reaches the end of either @src or @dst - that is, copies
973 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
975 void bio_copy_data(struct bio
*dst
, struct bio
*src
)
977 struct bvec_iter src_iter
, dst_iter
;
978 struct bio_vec src_bv
, dst_bv
;
982 src_iter
= src
->bi_iter
;
983 dst_iter
= dst
->bi_iter
;
986 if (!src_iter
.bi_size
) {
991 src_iter
= src
->bi_iter
;
994 if (!dst_iter
.bi_size
) {
999 dst_iter
= dst
->bi_iter
;
1002 src_bv
= bio_iter_iovec(src
, src_iter
);
1003 dst_bv
= bio_iter_iovec(dst
, dst_iter
);
1005 bytes
= min(src_bv
.bv_len
, dst_bv
.bv_len
);
1007 src_p
= kmap_atomic(src_bv
.bv_page
);
1008 dst_p
= kmap_atomic(dst_bv
.bv_page
);
1010 memcpy(dst_p
+ dst_bv
.bv_offset
,
1011 src_p
+ src_bv
.bv_offset
,
1014 kunmap_atomic(dst_p
);
1015 kunmap_atomic(src_p
);
1017 bio_advance_iter(src
, &src_iter
, bytes
);
1018 bio_advance_iter(dst
, &dst_iter
, bytes
);
1021 EXPORT_SYMBOL(bio_copy_data
);
1023 struct bio_map_data
{
1025 struct iov_iter iter
;
1029 static struct bio_map_data
*bio_alloc_map_data(unsigned int iov_count
,
1032 if (iov_count
> UIO_MAXIOV
)
1035 return kmalloc(sizeof(struct bio_map_data
) +
1036 sizeof(struct iovec
) * iov_count
, gfp_mask
);
1039 static int __bio_copy_iov(struct bio
*bio
, const struct iov_iter
*iter
,
1040 int to_user
, int from_user
)
1043 struct bio_vec
*bvec
;
1044 struct iov_iter iov_iter
= *iter
;
1046 bio_for_each_segment_all(bvec
, bio
, i
) {
1047 char *bv_addr
= page_address(bvec
->bv_page
);
1048 unsigned int bv_len
= bvec
->bv_len
;
1050 while (bv_len
&& iov_iter
.count
) {
1051 struct iovec iov
= iov_iter_iovec(&iov_iter
);
1052 unsigned int bytes
= min_t(unsigned int, bv_len
,
1057 ret
= copy_to_user(iov
.iov_base
,
1061 ret
= copy_from_user(bv_addr
,
1071 iov_iter_advance(&iov_iter
, bytes
);
1078 static void bio_free_pages(struct bio
*bio
)
1080 struct bio_vec
*bvec
;
1083 bio_for_each_segment_all(bvec
, bio
, i
)
1084 __free_page(bvec
->bv_page
);
1088 * bio_uncopy_user - finish previously mapped bio
1089 * @bio: bio being terminated
1091 * Free pages allocated from bio_copy_user_iov() and write back data
1092 * to user space in case of a read.
1094 int bio_uncopy_user(struct bio
*bio
)
1096 struct bio_map_data
*bmd
= bio
->bi_private
;
1099 if (!bio_flagged(bio
, BIO_NULL_MAPPED
)) {
1101 * if we're in a workqueue, the request is orphaned, so
1102 * don't copy into a random user address space, just free.
1105 ret
= __bio_copy_iov(bio
, &bmd
->iter
,
1106 bio_data_dir(bio
) == READ
, 0);
1107 if (bmd
->is_our_pages
)
1108 bio_free_pages(bio
);
1114 EXPORT_SYMBOL(bio_uncopy_user
);
1117 * bio_copy_user_iov - copy user data to bio
1118 * @q: destination block queue
1119 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1120 * @iter: iovec iterator
1121 * @gfp_mask: memory allocation flags
1123 * Prepares and returns a bio for indirect user io, bouncing data
1124 * to/from kernel pages as necessary. Must be paired with
1125 * call bio_uncopy_user() on io completion.
1127 struct bio
*bio_copy_user_iov(struct request_queue
*q
,
1128 struct rq_map_data
*map_data
,
1129 const struct iov_iter
*iter
,
1132 struct bio_map_data
*bmd
;
1137 unsigned int len
= iter
->count
;
1138 unsigned int offset
= map_data
? map_data
->offset
& ~PAGE_MASK
: 0;
1140 for (i
= 0; i
< iter
->nr_segs
; i
++) {
1141 unsigned long uaddr
;
1143 unsigned long start
;
1145 uaddr
= (unsigned long) iter
->iov
[i
].iov_base
;
1146 end
= (uaddr
+ iter
->iov
[i
].iov_len
+ PAGE_SIZE
- 1)
1148 start
= uaddr
>> PAGE_SHIFT
;
1154 return ERR_PTR(-EINVAL
);
1156 nr_pages
+= end
- start
;
1162 bmd
= bio_alloc_map_data(iter
->nr_segs
, gfp_mask
);
1164 return ERR_PTR(-ENOMEM
);
1167 * We need to do a deep copy of the iov_iter including the iovecs.
1168 * The caller provided iov might point to an on-stack or otherwise
1171 bmd
->is_our_pages
= map_data
? 0 : 1;
1172 memcpy(bmd
->iov
, iter
->iov
, sizeof(struct iovec
) * iter
->nr_segs
);
1173 iov_iter_init(&bmd
->iter
, iter
->type
, bmd
->iov
,
1174 iter
->nr_segs
, iter
->count
);
1177 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1181 if (iter
->type
& WRITE
)
1182 bio
->bi_rw
|= REQ_WRITE
;
1187 nr_pages
= 1 << map_data
->page_order
;
1188 i
= map_data
->offset
/ PAGE_SIZE
;
1191 unsigned int bytes
= PAGE_SIZE
;
1199 if (i
== map_data
->nr_entries
* nr_pages
) {
1204 page
= map_data
->pages
[i
/ nr_pages
];
1205 page
+= (i
% nr_pages
);
1209 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
1216 if (bio_add_pc_page(q
, bio
, page
, bytes
, offset
) < bytes
)
1229 if (((iter
->type
& WRITE
) && (!map_data
|| !map_data
->null_mapped
)) ||
1230 (map_data
&& map_data
->from_user
)) {
1231 ret
= __bio_copy_iov(bio
, iter
, 0, 1);
1236 bio
->bi_private
= bmd
;
1240 bio_free_pages(bio
);
1244 return ERR_PTR(ret
);
1247 static struct bio
*__bio_map_user_iov(struct request_queue
*q
,
1248 struct block_device
*bdev
,
1249 const struct iov_iter
*iter
,
1254 struct page
**pages
;
1261 iov_for_each(iov
, i
, *iter
) {
1262 unsigned long uaddr
= (unsigned long) iov
.iov_base
;
1263 unsigned long len
= iov
.iov_len
;
1264 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1265 unsigned long start
= uaddr
>> PAGE_SHIFT
;
1271 return ERR_PTR(-EINVAL
);
1273 nr_pages
+= end
- start
;
1275 * buffer must be aligned to at least hardsector size for now
1277 if (uaddr
& queue_dma_alignment(q
))
1278 return ERR_PTR(-EINVAL
);
1282 return ERR_PTR(-EINVAL
);
1284 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1286 return ERR_PTR(-ENOMEM
);
1289 pages
= kcalloc(nr_pages
, sizeof(struct page
*), gfp_mask
);
1293 iov_for_each(iov
, i
, *iter
) {
1294 unsigned long uaddr
= (unsigned long) iov
.iov_base
;
1295 unsigned long len
= iov
.iov_len
;
1296 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1297 unsigned long start
= uaddr
>> PAGE_SHIFT
;
1298 const int local_nr_pages
= end
- start
;
1299 const int page_limit
= cur_page
+ local_nr_pages
;
1301 ret
= get_user_pages_fast(uaddr
, local_nr_pages
,
1302 (iter
->type
& WRITE
) != WRITE
,
1304 if (ret
< local_nr_pages
) {
1309 offset
= uaddr
& ~PAGE_MASK
;
1310 for (j
= cur_page
; j
< page_limit
; j
++) {
1311 unsigned int bytes
= PAGE_SIZE
- offset
;
1322 if (bio_add_pc_page(q
, bio
, pages
[j
], bytes
, offset
) <
1332 * release the pages we didn't map into the bio, if any
1334 while (j
< page_limit
)
1335 page_cache_release(pages
[j
++]);
1341 * set data direction, and check if mapped pages need bouncing
1343 if (iter
->type
& WRITE
)
1344 bio
->bi_rw
|= REQ_WRITE
;
1346 bio
->bi_bdev
= bdev
;
1347 bio
->bi_flags
|= (1 << BIO_USER_MAPPED
);
1351 for (j
= 0; j
< nr_pages
; j
++) {
1354 page_cache_release(pages
[j
]);
1359 return ERR_PTR(ret
);
1363 * bio_map_user_iov - map user iovec into bio
1364 * @q: the struct request_queue for the bio
1365 * @bdev: destination block device
1366 * @iter: iovec iterator
1367 * @gfp_mask: memory allocation flags
1369 * Map the user space address into a bio suitable for io to a block
1370 * device. Returns an error pointer in case of error.
1372 struct bio
*bio_map_user_iov(struct request_queue
*q
, struct block_device
*bdev
,
1373 const struct iov_iter
*iter
,
1378 bio
= __bio_map_user_iov(q
, bdev
, iter
, gfp_mask
);
1383 * subtle -- if __bio_map_user() ended up bouncing a bio,
1384 * it would normally disappear when its bi_end_io is run.
1385 * however, we need it for the unmap, so grab an extra
1393 static void __bio_unmap_user(struct bio
*bio
)
1395 struct bio_vec
*bvec
;
1399 * make sure we dirty pages we wrote to
1401 bio_for_each_segment_all(bvec
, bio
, i
) {
1402 if (bio_data_dir(bio
) == READ
)
1403 set_page_dirty_lock(bvec
->bv_page
);
1405 page_cache_release(bvec
->bv_page
);
1412 * bio_unmap_user - unmap a bio
1413 * @bio: the bio being unmapped
1415 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1416 * a process context.
1418 * bio_unmap_user() may sleep.
1420 void bio_unmap_user(struct bio
*bio
)
1422 __bio_unmap_user(bio
);
1425 EXPORT_SYMBOL(bio_unmap_user
);
1427 static void bio_map_kern_endio(struct bio
*bio
, int err
)
1433 * bio_map_kern - map kernel address into bio
1434 * @q: the struct request_queue for the bio
1435 * @data: pointer to buffer to map
1436 * @len: length in bytes
1437 * @gfp_mask: allocation flags for bio allocation
1439 * Map the kernel address into a bio suitable for io to a block
1440 * device. Returns an error pointer in case of error.
1442 struct bio
*bio_map_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1445 unsigned long kaddr
= (unsigned long)data
;
1446 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1447 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1448 const int nr_pages
= end
- start
;
1452 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1454 return ERR_PTR(-ENOMEM
);
1456 offset
= offset_in_page(kaddr
);
1457 for (i
= 0; i
< nr_pages
; i
++) {
1458 unsigned int bytes
= PAGE_SIZE
- offset
;
1466 if (bio_add_pc_page(q
, bio
, virt_to_page(data
), bytes
,
1468 /* we don't support partial mappings */
1470 return ERR_PTR(-EINVAL
);
1478 bio
->bi_end_io
= bio_map_kern_endio
;
1481 EXPORT_SYMBOL(bio_map_kern
);
1483 static void bio_copy_kern_endio(struct bio
*bio
, int err
)
1485 bio_free_pages(bio
);
1489 static void bio_copy_kern_endio_read(struct bio
*bio
, int err
)
1491 char *p
= bio
->bi_private
;
1492 struct bio_vec
*bvec
;
1495 bio_for_each_segment_all(bvec
, bio
, i
) {
1496 memcpy(p
, page_address(bvec
->bv_page
), bvec
->bv_len
);
1500 bio_copy_kern_endio(bio
, err
);
1504 * bio_copy_kern - copy kernel address into bio
1505 * @q: the struct request_queue for the bio
1506 * @data: pointer to buffer to copy
1507 * @len: length in bytes
1508 * @gfp_mask: allocation flags for bio and page allocation
1509 * @reading: data direction is READ
1511 * copy the kernel address into a bio suitable for io to a block
1512 * device. Returns an error pointer in case of error.
1514 struct bio
*bio_copy_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1515 gfp_t gfp_mask
, int reading
)
1517 unsigned long kaddr
= (unsigned long)data
;
1518 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1519 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1528 return ERR_PTR(-EINVAL
);
1530 nr_pages
= end
- start
;
1531 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1533 return ERR_PTR(-ENOMEM
);
1537 unsigned int bytes
= PAGE_SIZE
;
1542 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
1547 memcpy(page_address(page
), p
, bytes
);
1549 if (bio_add_pc_page(q
, bio
, page
, bytes
, 0) < bytes
)
1557 bio
->bi_end_io
= bio_copy_kern_endio_read
;
1558 bio
->bi_private
= data
;
1560 bio
->bi_end_io
= bio_copy_kern_endio
;
1561 bio
->bi_rw
|= REQ_WRITE
;
1567 bio_free_pages(bio
);
1569 return ERR_PTR(-ENOMEM
);
1571 EXPORT_SYMBOL(bio_copy_kern
);
1574 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1575 * for performing direct-IO in BIOs.
1577 * The problem is that we cannot run set_page_dirty() from interrupt context
1578 * because the required locks are not interrupt-safe. So what we can do is to
1579 * mark the pages dirty _before_ performing IO. And in interrupt context,
1580 * check that the pages are still dirty. If so, fine. If not, redirty them
1581 * in process context.
1583 * We special-case compound pages here: normally this means reads into hugetlb
1584 * pages. The logic in here doesn't really work right for compound pages
1585 * because the VM does not uniformly chase down the head page in all cases.
1586 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1587 * handle them at all. So we skip compound pages here at an early stage.
1589 * Note that this code is very hard to test under normal circumstances because
1590 * direct-io pins the pages with get_user_pages(). This makes
1591 * is_page_cache_freeable return false, and the VM will not clean the pages.
1592 * But other code (eg, flusher threads) could clean the pages if they are mapped
1595 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1596 * deferred bio dirtying paths.
1600 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1602 void bio_set_pages_dirty(struct bio
*bio
)
1604 struct bio_vec
*bvec
;
1607 bio_for_each_segment_all(bvec
, bio
, i
) {
1608 struct page
*page
= bvec
->bv_page
;
1610 if (page
&& !PageCompound(page
))
1611 set_page_dirty_lock(page
);
1615 static void bio_release_pages(struct bio
*bio
)
1617 struct bio_vec
*bvec
;
1620 bio_for_each_segment_all(bvec
, bio
, i
) {
1621 struct page
*page
= bvec
->bv_page
;
1629 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1630 * If they are, then fine. If, however, some pages are clean then they must
1631 * have been written out during the direct-IO read. So we take another ref on
1632 * the BIO and the offending pages and re-dirty the pages in process context.
1634 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1635 * here on. It will run one page_cache_release() against each page and will
1636 * run one bio_put() against the BIO.
1639 static void bio_dirty_fn(struct work_struct
*work
);
1641 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
1642 static DEFINE_SPINLOCK(bio_dirty_lock
);
1643 static struct bio
*bio_dirty_list
;
1646 * This runs in process context
1648 static void bio_dirty_fn(struct work_struct
*work
)
1650 unsigned long flags
;
1653 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1654 bio
= bio_dirty_list
;
1655 bio_dirty_list
= NULL
;
1656 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1659 struct bio
*next
= bio
->bi_private
;
1661 bio_set_pages_dirty(bio
);
1662 bio_release_pages(bio
);
1668 void bio_check_pages_dirty(struct bio
*bio
)
1670 struct bio_vec
*bvec
;
1671 int nr_clean_pages
= 0;
1674 bio_for_each_segment_all(bvec
, bio
, i
) {
1675 struct page
*page
= bvec
->bv_page
;
1677 if (PageDirty(page
) || PageCompound(page
)) {
1678 page_cache_release(page
);
1679 bvec
->bv_page
= NULL
;
1685 if (nr_clean_pages
) {
1686 unsigned long flags
;
1688 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1689 bio
->bi_private
= bio_dirty_list
;
1690 bio_dirty_list
= bio
;
1691 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1692 schedule_work(&bio_dirty_work
);
1698 void generic_start_io_acct(int rw
, unsigned long sectors
,
1699 struct hd_struct
*part
)
1701 int cpu
= part_stat_lock();
1703 part_round_stats(cpu
, part
);
1704 part_stat_inc(cpu
, part
, ios
[rw
]);
1705 part_stat_add(cpu
, part
, sectors
[rw
], sectors
);
1706 part_inc_in_flight(part
, rw
);
1710 EXPORT_SYMBOL(generic_start_io_acct
);
1712 void generic_end_io_acct(int rw
, struct hd_struct
*part
,
1713 unsigned long start_time
)
1715 unsigned long duration
= jiffies
- start_time
;
1716 int cpu
= part_stat_lock();
1718 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1719 part_round_stats(cpu
, part
);
1720 part_dec_in_flight(part
, rw
);
1724 EXPORT_SYMBOL(generic_end_io_acct
);
1726 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1727 void bio_flush_dcache_pages(struct bio
*bi
)
1729 struct bio_vec bvec
;
1730 struct bvec_iter iter
;
1732 bio_for_each_segment(bvec
, bi
, iter
)
1733 flush_dcache_page(bvec
.bv_page
);
1735 EXPORT_SYMBOL(bio_flush_dcache_pages
);
1739 * bio_endio - end I/O on a bio
1741 * @error: error, if any
1744 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1745 * preferred way to end I/O on a bio, it takes care of clearing
1746 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1747 * established -Exxxx (-EIO, for instance) error values in case
1748 * something went wrong. No one should call bi_end_io() directly on a
1749 * bio unless they own it and thus know that it has an end_io
1752 void bio_endio(struct bio
*bio
, int error
)
1755 BUG_ON(atomic_read(&bio
->bi_remaining
) <= 0);
1758 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1759 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
1762 if (!atomic_dec_and_test(&bio
->bi_remaining
))
1766 * Need to have a real endio function for chained bios,
1767 * otherwise various corner cases will break (like stacking
1768 * block devices that save/restore bi_end_io) - however, we want
1769 * to avoid unbounded recursion and blowing the stack. Tail call
1770 * optimization would handle this, but compiling with frame
1771 * pointers also disables gcc's sibling call optimization.
1773 if (bio
->bi_end_io
== bio_chain_endio
) {
1774 struct bio
*parent
= bio
->bi_private
;
1779 bio
->bi_end_io(bio
, error
);
1784 EXPORT_SYMBOL(bio_endio
);
1787 * bio_endio_nodec - end I/O on a bio, without decrementing bi_remaining
1789 * @error: error, if any
1791 * For code that has saved and restored bi_end_io; thing hard before using this
1792 * function, probably you should've cloned the entire bio.
1794 void bio_endio_nodec(struct bio
*bio
, int error
)
1796 atomic_inc(&bio
->bi_remaining
);
1797 bio_endio(bio
, error
);
1799 EXPORT_SYMBOL(bio_endio_nodec
);
1802 * bio_split - split a bio
1803 * @bio: bio to split
1804 * @sectors: number of sectors to split from the front of @bio
1806 * @bs: bio set to allocate from
1808 * Allocates and returns a new bio which represents @sectors from the start of
1809 * @bio, and updates @bio to represent the remaining sectors.
1811 * The newly allocated bio will point to @bio's bi_io_vec; it is the caller's
1812 * responsibility to ensure that @bio is not freed before the split.
1814 struct bio
*bio_split(struct bio
*bio
, int sectors
,
1815 gfp_t gfp
, struct bio_set
*bs
)
1817 struct bio
*split
= NULL
;
1819 BUG_ON(sectors
<= 0);
1820 BUG_ON(sectors
>= bio_sectors(bio
));
1822 split
= bio_clone_fast(bio
, gfp
, bs
);
1826 split
->bi_iter
.bi_size
= sectors
<< 9;
1828 if (bio_integrity(split
))
1829 bio_integrity_trim(split
, 0, sectors
);
1831 bio_advance(bio
, split
->bi_iter
.bi_size
);
1835 EXPORT_SYMBOL(bio_split
);
1838 * bio_trim - trim a bio
1840 * @offset: number of sectors to trim from the front of @bio
1841 * @size: size we want to trim @bio to, in sectors
1843 void bio_trim(struct bio
*bio
, int offset
, int size
)
1845 /* 'bio' is a cloned bio which we need to trim to match
1846 * the given offset and size.
1850 if (offset
== 0 && size
== bio
->bi_iter
.bi_size
)
1853 clear_bit(BIO_SEG_VALID
, &bio
->bi_flags
);
1855 bio_advance(bio
, offset
<< 9);
1857 bio
->bi_iter
.bi_size
= size
;
1859 EXPORT_SYMBOL_GPL(bio_trim
);
1862 * create memory pools for biovec's in a bio_set.
1863 * use the global biovec slabs created for general use.
1865 mempool_t
*biovec_create_pool(int pool_entries
)
1867 struct biovec_slab
*bp
= bvec_slabs
+ BIOVEC_MAX_IDX
;
1869 return mempool_create_slab_pool(pool_entries
, bp
->slab
);
1872 void bioset_free(struct bio_set
*bs
)
1874 if (bs
->rescue_workqueue
)
1875 destroy_workqueue(bs
->rescue_workqueue
);
1878 mempool_destroy(bs
->bio_pool
);
1881 mempool_destroy(bs
->bvec_pool
);
1883 bioset_integrity_free(bs
);
1888 EXPORT_SYMBOL(bioset_free
);
1890 static struct bio_set
*__bioset_create(unsigned int pool_size
,
1891 unsigned int front_pad
,
1892 bool create_bvec_pool
)
1894 unsigned int back_pad
= BIO_INLINE_VECS
* sizeof(struct bio_vec
);
1897 bs
= kzalloc(sizeof(*bs
), GFP_KERNEL
);
1901 bs
->front_pad
= front_pad
;
1903 spin_lock_init(&bs
->rescue_lock
);
1904 bio_list_init(&bs
->rescue_list
);
1905 INIT_WORK(&bs
->rescue_work
, bio_alloc_rescue
);
1907 bs
->bio_slab
= bio_find_or_create_slab(front_pad
+ back_pad
);
1908 if (!bs
->bio_slab
) {
1913 bs
->bio_pool
= mempool_create_slab_pool(pool_size
, bs
->bio_slab
);
1917 if (create_bvec_pool
) {
1918 bs
->bvec_pool
= biovec_create_pool(pool_size
);
1923 bs
->rescue_workqueue
= alloc_workqueue("bioset", WQ_MEM_RECLAIM
, 0);
1924 if (!bs
->rescue_workqueue
)
1934 * bioset_create - Create a bio_set
1935 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1936 * @front_pad: Number of bytes to allocate in front of the returned bio
1939 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1940 * to ask for a number of bytes to be allocated in front of the bio.
1941 * Front pad allocation is useful for embedding the bio inside
1942 * another structure, to avoid allocating extra data to go with the bio.
1943 * Note that the bio must be embedded at the END of that structure always,
1944 * or things will break badly.
1946 struct bio_set
*bioset_create(unsigned int pool_size
, unsigned int front_pad
)
1948 return __bioset_create(pool_size
, front_pad
, true);
1950 EXPORT_SYMBOL(bioset_create
);
1953 * bioset_create_nobvec - Create a bio_set without bio_vec mempool
1954 * @pool_size: Number of bio to cache in the mempool
1955 * @front_pad: Number of bytes to allocate in front of the returned bio
1958 * Same functionality as bioset_create() except that mempool is not
1959 * created for bio_vecs. Saving some memory for bio_clone_fast() users.
1961 struct bio_set
*bioset_create_nobvec(unsigned int pool_size
, unsigned int front_pad
)
1963 return __bioset_create(pool_size
, front_pad
, false);
1965 EXPORT_SYMBOL(bioset_create_nobvec
);
1967 #ifdef CONFIG_BLK_CGROUP
1969 * bio_associate_current - associate a bio with %current
1972 * Associate @bio with %current if it hasn't been associated yet. Block
1973 * layer will treat @bio as if it were issued by %current no matter which
1974 * task actually issues it.
1976 * This function takes an extra reference of @task's io_context and blkcg
1977 * which will be put when @bio is released. The caller must own @bio,
1978 * ensure %current->io_context exists, and is responsible for synchronizing
1979 * calls to this function.
1981 int bio_associate_current(struct bio
*bio
)
1983 struct io_context
*ioc
;
1984 struct cgroup_subsys_state
*css
;
1989 ioc
= current
->io_context
;
1993 /* acquire active ref on @ioc and associate */
1994 get_io_context_active(ioc
);
1997 /* associate blkcg if exists */
1999 css
= task_css(current
, blkio_cgrp_id
);
2000 if (css
&& css_tryget_online(css
))
2008 * bio_disassociate_task - undo bio_associate_current()
2011 void bio_disassociate_task(struct bio
*bio
)
2014 put_io_context(bio
->bi_ioc
);
2018 css_put(bio
->bi_css
);
2023 #endif /* CONFIG_BLK_CGROUP */
2025 static void __init
biovec_init_slabs(void)
2029 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
2031 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
2033 if (bvs
->nr_vecs
<= BIO_INLINE_VECS
) {
2038 size
= bvs
->nr_vecs
* sizeof(struct bio_vec
);
2039 bvs
->slab
= kmem_cache_create(bvs
->name
, size
, 0,
2040 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
2044 static int __init
init_bio(void)
2048 bio_slabs
= kzalloc(bio_slab_max
* sizeof(struct bio_slab
), GFP_KERNEL
);
2050 panic("bio: can't allocate bios\n");
2052 bio_integrity_init();
2053 biovec_init_slabs();
2055 fs_bio_set
= bioset_create(BIO_POOL_SIZE
, 0);
2057 panic("bio: can't allocate bios\n");
2059 if (bioset_integrity_create(fs_bio_set
, BIO_POOL_SIZE
))
2060 panic("bio: can't create integrity pool\n");
2064 subsys_initcall(init_bio
);