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>
31 #include <linux/blk-cgroup.h>
33 #include <trace/events/block.h>
35 #include "blk-rq-qos.h"
38 * Test patch to inline a certain number of bi_io_vec's inside the bio
39 * itself, to shrink a bio data allocation from two mempool calls to one
41 #define BIO_INLINE_VECS 4
44 * if you change this list, also change bvec_alloc or things will
45 * break badly! cannot be bigger than what you can fit into an
48 #define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n }
49 static struct biovec_slab bvec_slabs
[BVEC_POOL_NR
] __read_mostly
= {
50 BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES
, max
),
55 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
56 * IO code that does not need private memory pools.
58 struct bio_set fs_bio_set
;
59 EXPORT_SYMBOL(fs_bio_set
);
62 * Our slab pool management
65 struct kmem_cache
*slab
;
66 unsigned int slab_ref
;
67 unsigned int slab_size
;
70 static DEFINE_MUTEX(bio_slab_lock
);
71 static struct bio_slab
*bio_slabs
;
72 static unsigned int bio_slab_nr
, bio_slab_max
;
74 static struct kmem_cache
*bio_find_or_create_slab(unsigned int extra_size
)
76 unsigned int sz
= sizeof(struct bio
) + extra_size
;
77 struct kmem_cache
*slab
= NULL
;
78 struct bio_slab
*bslab
, *new_bio_slabs
;
79 unsigned int new_bio_slab_max
;
80 unsigned int i
, entry
= -1;
82 mutex_lock(&bio_slab_lock
);
85 while (i
< bio_slab_nr
) {
86 bslab
= &bio_slabs
[i
];
88 if (!bslab
->slab
&& entry
== -1)
90 else if (bslab
->slab_size
== sz
) {
101 if (bio_slab_nr
== bio_slab_max
&& entry
== -1) {
102 new_bio_slab_max
= bio_slab_max
<< 1;
103 new_bio_slabs
= krealloc(bio_slabs
,
104 new_bio_slab_max
* sizeof(struct bio_slab
),
108 bio_slab_max
= new_bio_slab_max
;
109 bio_slabs
= new_bio_slabs
;
112 entry
= bio_slab_nr
++;
114 bslab
= &bio_slabs
[entry
];
116 snprintf(bslab
->name
, sizeof(bslab
->name
), "bio-%d", entry
);
117 slab
= kmem_cache_create(bslab
->name
, sz
, ARCH_KMALLOC_MINALIGN
,
118 SLAB_HWCACHE_ALIGN
, NULL
);
124 bslab
->slab_size
= sz
;
126 mutex_unlock(&bio_slab_lock
);
130 static void bio_put_slab(struct bio_set
*bs
)
132 struct bio_slab
*bslab
= NULL
;
135 mutex_lock(&bio_slab_lock
);
137 for (i
= 0; i
< bio_slab_nr
; i
++) {
138 if (bs
->bio_slab
== bio_slabs
[i
].slab
) {
139 bslab
= &bio_slabs
[i
];
144 if (WARN(!bslab
, KERN_ERR
"bio: unable to find slab!\n"))
147 WARN_ON(!bslab
->slab_ref
);
149 if (--bslab
->slab_ref
)
152 kmem_cache_destroy(bslab
->slab
);
156 mutex_unlock(&bio_slab_lock
);
159 unsigned int bvec_nr_vecs(unsigned short idx
)
161 return bvec_slabs
[--idx
].nr_vecs
;
164 void bvec_free(mempool_t
*pool
, struct bio_vec
*bv
, unsigned int idx
)
170 BIO_BUG_ON(idx
>= BVEC_POOL_NR
);
172 if (idx
== BVEC_POOL_MAX
) {
173 mempool_free(bv
, pool
);
175 struct biovec_slab
*bvs
= bvec_slabs
+ idx
;
177 kmem_cache_free(bvs
->slab
, bv
);
181 struct bio_vec
*bvec_alloc(gfp_t gfp_mask
, int nr
, unsigned long *idx
,
187 * see comment near bvec_array define!
205 case 129 ... BIO_MAX_PAGES
:
213 * idx now points to the pool we want to allocate from. only the
214 * 1-vec entry pool is mempool backed.
216 if (*idx
== BVEC_POOL_MAX
) {
218 bvl
= mempool_alloc(pool
, gfp_mask
);
220 struct biovec_slab
*bvs
= bvec_slabs
+ *idx
;
221 gfp_t __gfp_mask
= gfp_mask
& ~(__GFP_DIRECT_RECLAIM
| __GFP_IO
);
224 * Make this allocation restricted and don't dump info on
225 * allocation failures, since we'll fallback to the mempool
226 * in case of failure.
228 __gfp_mask
|= __GFP_NOMEMALLOC
| __GFP_NORETRY
| __GFP_NOWARN
;
231 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
232 * is set, retry with the 1-entry mempool
234 bvl
= kmem_cache_alloc(bvs
->slab
, __gfp_mask
);
235 if (unlikely(!bvl
&& (gfp_mask
& __GFP_DIRECT_RECLAIM
))) {
236 *idx
= BVEC_POOL_MAX
;
245 void bio_uninit(struct bio
*bio
)
247 bio_disassociate_task(bio
);
249 EXPORT_SYMBOL(bio_uninit
);
251 static void bio_free(struct bio
*bio
)
253 struct bio_set
*bs
= bio
->bi_pool
;
259 bvec_free(&bs
->bvec_pool
, bio
->bi_io_vec
, BVEC_POOL_IDX(bio
));
262 * If we have front padding, adjust the bio pointer before freeing
267 mempool_free(p
, &bs
->bio_pool
);
269 /* Bio was allocated by bio_kmalloc() */
275 * Users of this function have their own bio allocation. Subsequently,
276 * they must remember to pair any call to bio_init() with bio_uninit()
277 * when IO has completed, or when the bio is released.
279 void bio_init(struct bio
*bio
, struct bio_vec
*table
,
280 unsigned short max_vecs
)
282 memset(bio
, 0, sizeof(*bio
));
283 atomic_set(&bio
->__bi_remaining
, 1);
284 atomic_set(&bio
->__bi_cnt
, 1);
286 bio
->bi_io_vec
= table
;
287 bio
->bi_max_vecs
= max_vecs
;
289 EXPORT_SYMBOL(bio_init
);
292 * bio_reset - reinitialize a bio
296 * After calling bio_reset(), @bio will be in the same state as a freshly
297 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
298 * preserved are the ones that are initialized by bio_alloc_bioset(). See
299 * comment in struct bio.
301 void bio_reset(struct bio
*bio
)
303 unsigned long flags
= bio
->bi_flags
& (~0UL << BIO_RESET_BITS
);
307 memset(bio
, 0, BIO_RESET_BYTES
);
308 bio
->bi_flags
= flags
;
309 atomic_set(&bio
->__bi_remaining
, 1);
311 EXPORT_SYMBOL(bio_reset
);
313 static struct bio
*__bio_chain_endio(struct bio
*bio
)
315 struct bio
*parent
= bio
->bi_private
;
317 if (!parent
->bi_status
)
318 parent
->bi_status
= bio
->bi_status
;
323 static void bio_chain_endio(struct bio
*bio
)
325 bio_endio(__bio_chain_endio(bio
));
329 * bio_chain - chain bio completions
330 * @bio: the target bio
331 * @parent: the @bio's parent bio
333 * The caller won't have a bi_end_io called when @bio completes - instead,
334 * @parent's bi_end_io won't be called until both @parent and @bio have
335 * completed; the chained bio will also be freed when it completes.
337 * The caller must not set bi_private or bi_end_io in @bio.
339 void bio_chain(struct bio
*bio
, struct bio
*parent
)
341 BUG_ON(bio
->bi_private
|| bio
->bi_end_io
);
343 bio
->bi_private
= parent
;
344 bio
->bi_end_io
= bio_chain_endio
;
345 bio_inc_remaining(parent
);
347 EXPORT_SYMBOL(bio_chain
);
349 static void bio_alloc_rescue(struct work_struct
*work
)
351 struct bio_set
*bs
= container_of(work
, struct bio_set
, rescue_work
);
355 spin_lock(&bs
->rescue_lock
);
356 bio
= bio_list_pop(&bs
->rescue_list
);
357 spin_unlock(&bs
->rescue_lock
);
362 generic_make_request(bio
);
366 static void punt_bios_to_rescuer(struct bio_set
*bs
)
368 struct bio_list punt
, nopunt
;
371 if (WARN_ON_ONCE(!bs
->rescue_workqueue
))
374 * In order to guarantee forward progress we must punt only bios that
375 * were allocated from this bio_set; otherwise, if there was a bio on
376 * there for a stacking driver higher up in the stack, processing it
377 * could require allocating bios from this bio_set, and doing that from
378 * our own rescuer would be bad.
380 * Since bio lists are singly linked, pop them all instead of trying to
381 * remove from the middle of the list:
384 bio_list_init(&punt
);
385 bio_list_init(&nopunt
);
387 while ((bio
= bio_list_pop(¤t
->bio_list
[0])))
388 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
389 current
->bio_list
[0] = nopunt
;
391 bio_list_init(&nopunt
);
392 while ((bio
= bio_list_pop(¤t
->bio_list
[1])))
393 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
394 current
->bio_list
[1] = nopunt
;
396 spin_lock(&bs
->rescue_lock
);
397 bio_list_merge(&bs
->rescue_list
, &punt
);
398 spin_unlock(&bs
->rescue_lock
);
400 queue_work(bs
->rescue_workqueue
, &bs
->rescue_work
);
404 * bio_alloc_bioset - allocate a bio for I/O
405 * @gfp_mask: the GFP_* mask given to the slab allocator
406 * @nr_iovecs: number of iovecs to pre-allocate
407 * @bs: the bio_set to allocate from.
410 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
411 * backed by the @bs's mempool.
413 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
414 * always be able to allocate a bio. This is due to the mempool guarantees.
415 * To make this work, callers must never allocate more than 1 bio at a time
416 * from this pool. Callers that need to allocate more than 1 bio must always
417 * submit the previously allocated bio for IO before attempting to allocate
418 * a new one. Failure to do so can cause deadlocks under memory pressure.
420 * Note that when running under generic_make_request() (i.e. any block
421 * driver), bios are not submitted until after you return - see the code in
422 * generic_make_request() that converts recursion into iteration, to prevent
425 * This would normally mean allocating multiple bios under
426 * generic_make_request() would be susceptible to deadlocks, but we have
427 * deadlock avoidance code that resubmits any blocked bios from a rescuer
430 * However, we do not guarantee forward progress for allocations from other
431 * mempools. Doing multiple allocations from the same mempool under
432 * generic_make_request() should be avoided - instead, use bio_set's front_pad
433 * for per bio allocations.
436 * Pointer to new bio on success, NULL on failure.
438 struct bio
*bio_alloc_bioset(gfp_t gfp_mask
, unsigned int nr_iovecs
,
441 gfp_t saved_gfp
= gfp_mask
;
443 unsigned inline_vecs
;
444 struct bio_vec
*bvl
= NULL
;
449 if (nr_iovecs
> UIO_MAXIOV
)
452 p
= kmalloc(sizeof(struct bio
) +
453 nr_iovecs
* sizeof(struct bio_vec
),
456 inline_vecs
= nr_iovecs
;
458 /* should not use nobvec bioset for nr_iovecs > 0 */
459 if (WARN_ON_ONCE(!mempool_initialized(&bs
->bvec_pool
) &&
463 * generic_make_request() converts recursion to iteration; this
464 * means if we're running beneath it, any bios we allocate and
465 * submit will not be submitted (and thus freed) until after we
468 * This exposes us to a potential deadlock if we allocate
469 * multiple bios from the same bio_set() while running
470 * underneath generic_make_request(). If we were to allocate
471 * multiple bios (say a stacking block driver that was splitting
472 * bios), we would deadlock if we exhausted the mempool's
475 * We solve this, and guarantee forward progress, with a rescuer
476 * workqueue per bio_set. If we go to allocate and there are
477 * bios on current->bio_list, we first try the allocation
478 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
479 * bios we would be blocking to the rescuer workqueue before
480 * we retry with the original gfp_flags.
483 if (current
->bio_list
&&
484 (!bio_list_empty(¤t
->bio_list
[0]) ||
485 !bio_list_empty(¤t
->bio_list
[1])) &&
486 bs
->rescue_workqueue
)
487 gfp_mask
&= ~__GFP_DIRECT_RECLAIM
;
489 p
= mempool_alloc(&bs
->bio_pool
, gfp_mask
);
490 if (!p
&& gfp_mask
!= saved_gfp
) {
491 punt_bios_to_rescuer(bs
);
492 gfp_mask
= saved_gfp
;
493 p
= mempool_alloc(&bs
->bio_pool
, gfp_mask
);
496 front_pad
= bs
->front_pad
;
497 inline_vecs
= BIO_INLINE_VECS
;
504 bio_init(bio
, NULL
, 0);
506 if (nr_iovecs
> inline_vecs
) {
507 unsigned long idx
= 0;
509 bvl
= bvec_alloc(gfp_mask
, nr_iovecs
, &idx
, &bs
->bvec_pool
);
510 if (!bvl
&& gfp_mask
!= saved_gfp
) {
511 punt_bios_to_rescuer(bs
);
512 gfp_mask
= saved_gfp
;
513 bvl
= bvec_alloc(gfp_mask
, nr_iovecs
, &idx
, &bs
->bvec_pool
);
519 bio
->bi_flags
|= idx
<< BVEC_POOL_OFFSET
;
520 } else if (nr_iovecs
) {
521 bvl
= bio
->bi_inline_vecs
;
525 bio
->bi_max_vecs
= nr_iovecs
;
526 bio
->bi_io_vec
= bvl
;
530 mempool_free(p
, &bs
->bio_pool
);
533 EXPORT_SYMBOL(bio_alloc_bioset
);
535 void zero_fill_bio_iter(struct bio
*bio
, struct bvec_iter start
)
539 struct bvec_iter iter
;
541 __bio_for_each_segment(bv
, bio
, iter
, start
) {
542 char *data
= bvec_kmap_irq(&bv
, &flags
);
543 memset(data
, 0, bv
.bv_len
);
544 flush_dcache_page(bv
.bv_page
);
545 bvec_kunmap_irq(data
, &flags
);
548 EXPORT_SYMBOL(zero_fill_bio_iter
);
551 * bio_put - release a reference to a bio
552 * @bio: bio to release reference to
555 * Put a reference to a &struct bio, either one you have gotten with
556 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
558 void bio_put(struct bio
*bio
)
560 if (!bio_flagged(bio
, BIO_REFFED
))
563 BIO_BUG_ON(!atomic_read(&bio
->__bi_cnt
));
568 if (atomic_dec_and_test(&bio
->__bi_cnt
))
572 EXPORT_SYMBOL(bio_put
);
574 inline int bio_phys_segments(struct request_queue
*q
, struct bio
*bio
)
576 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
577 blk_recount_segments(q
, bio
);
579 return bio
->bi_phys_segments
;
581 EXPORT_SYMBOL(bio_phys_segments
);
584 * __bio_clone_fast - clone a bio that shares the original bio's biovec
585 * @bio: destination bio
586 * @bio_src: bio to clone
588 * Clone a &bio. Caller will own the returned bio, but not
589 * the actual data it points to. Reference count of returned
592 * Caller must ensure that @bio_src is not freed before @bio.
594 void __bio_clone_fast(struct bio
*bio
, struct bio
*bio_src
)
596 BUG_ON(bio
->bi_pool
&& BVEC_POOL_IDX(bio
));
599 * most users will be overriding ->bi_disk with a new target,
600 * so we don't set nor calculate new physical/hw segment counts here
602 bio
->bi_disk
= bio_src
->bi_disk
;
603 bio
->bi_partno
= bio_src
->bi_partno
;
604 bio_set_flag(bio
, BIO_CLONED
);
605 if (bio_flagged(bio_src
, BIO_THROTTLED
))
606 bio_set_flag(bio
, BIO_THROTTLED
);
607 bio
->bi_opf
= bio_src
->bi_opf
;
608 bio
->bi_write_hint
= bio_src
->bi_write_hint
;
609 bio
->bi_iter
= bio_src
->bi_iter
;
610 bio
->bi_io_vec
= bio_src
->bi_io_vec
;
612 bio_clone_blkg_association(bio
, bio_src
);
614 blkcg_bio_issue_init(bio
);
616 EXPORT_SYMBOL(__bio_clone_fast
);
619 * bio_clone_fast - clone a bio that shares the original bio's biovec
621 * @gfp_mask: allocation priority
622 * @bs: bio_set to allocate from
624 * Like __bio_clone_fast, only also allocates the returned bio
626 struct bio
*bio_clone_fast(struct bio
*bio
, gfp_t gfp_mask
, struct bio_set
*bs
)
630 b
= bio_alloc_bioset(gfp_mask
, 0, bs
);
634 __bio_clone_fast(b
, bio
);
636 if (bio_integrity(bio
)) {
639 ret
= bio_integrity_clone(b
, bio
, gfp_mask
);
649 EXPORT_SYMBOL(bio_clone_fast
);
652 * bio_add_pc_page - attempt to add page to bio
653 * @q: the target queue
654 * @bio: destination bio
656 * @len: vec entry length
657 * @offset: vec entry offset
659 * Attempt to add a page to the bio_vec maplist. This can fail for a
660 * number of reasons, such as the bio being full or target block device
661 * limitations. The target block device must allow bio's up to PAGE_SIZE,
662 * so it is always possible to add a single page to an empty bio.
664 * This should only be used by REQ_PC bios.
666 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
, struct page
667 *page
, unsigned int len
, unsigned int offset
)
669 int retried_segments
= 0;
670 struct bio_vec
*bvec
;
673 * cloned bio must not modify vec list
675 if (unlikely(bio_flagged(bio
, BIO_CLONED
)))
678 if (((bio
->bi_iter
.bi_size
+ len
) >> 9) > queue_max_hw_sectors(q
))
682 * For filesystems with a blocksize smaller than the pagesize
683 * we will often be called with the same page as last time and
684 * a consecutive offset. Optimize this special case.
686 if (bio
->bi_vcnt
> 0) {
687 struct bio_vec
*prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
689 if (page
== prev
->bv_page
&&
690 offset
== prev
->bv_offset
+ prev
->bv_len
) {
692 bio
->bi_iter
.bi_size
+= len
;
697 * If the queue doesn't support SG gaps and adding this
698 * offset would create a gap, disallow it.
700 if (bvec_gap_to_prev(q
, prev
, offset
))
708 * setup the new entry, we might clear it again later if we
709 * cannot add the page
711 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
712 bvec
->bv_page
= page
;
714 bvec
->bv_offset
= offset
;
716 bio
->bi_phys_segments
++;
717 bio
->bi_iter
.bi_size
+= len
;
720 * Perform a recount if the number of segments is greater
721 * than queue_max_segments(q).
724 while (bio
->bi_phys_segments
> queue_max_segments(q
)) {
726 if (retried_segments
)
729 retried_segments
= 1;
730 blk_recount_segments(q
, bio
);
733 /* If we may be able to merge these biovecs, force a recount */
734 if (bio
->bi_vcnt
> 1 && biovec_phys_mergeable(q
, bvec
- 1, bvec
))
735 bio_clear_flag(bio
, BIO_SEG_VALID
);
741 bvec
->bv_page
= NULL
;
745 bio
->bi_iter
.bi_size
-= len
;
746 blk_recount_segments(q
, bio
);
749 EXPORT_SYMBOL(bio_add_pc_page
);
752 * __bio_try_merge_page - try appending data to an existing bvec.
753 * @bio: destination bio
755 * @len: length of the data to add
756 * @off: offset of the data in @page
758 * Try to add the data at @page + @off to the last bvec of @bio. This is a
759 * a useful optimisation for file systems with a block size smaller than the
762 * Return %true on success or %false on failure.
764 bool __bio_try_merge_page(struct bio
*bio
, struct page
*page
,
765 unsigned int len
, unsigned int off
)
767 if (WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
)))
770 if (bio
->bi_vcnt
> 0) {
771 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
773 if (page
== bv
->bv_page
&& off
== bv
->bv_offset
+ bv
->bv_len
) {
775 bio
->bi_iter
.bi_size
+= len
;
781 EXPORT_SYMBOL_GPL(__bio_try_merge_page
);
784 * __bio_add_page - add page to a bio in a new segment
785 * @bio: destination bio
787 * @len: length of the data to add
788 * @off: offset of the data in @page
790 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
791 * that @bio has space for another bvec.
793 void __bio_add_page(struct bio
*bio
, struct page
*page
,
794 unsigned int len
, unsigned int off
)
796 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
798 WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
));
799 WARN_ON_ONCE(bio_full(bio
));
805 bio
->bi_iter
.bi_size
+= len
;
808 EXPORT_SYMBOL_GPL(__bio_add_page
);
811 * bio_add_page - attempt to add page to bio
812 * @bio: destination bio
814 * @len: vec entry length
815 * @offset: vec entry offset
817 * Attempt to add a page to the bio_vec maplist. This will only fail
818 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
820 int bio_add_page(struct bio
*bio
, struct page
*page
,
821 unsigned int len
, unsigned int offset
)
823 if (!__bio_try_merge_page(bio
, page
, len
, offset
)) {
826 __bio_add_page(bio
, page
, len
, offset
);
830 EXPORT_SYMBOL(bio_add_page
);
832 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
835 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
836 * @bio: bio to add pages to
837 * @iter: iov iterator describing the region to be mapped
839 * Pins pages from *iter and appends them to @bio's bvec array. The
840 * pages will have to be released using put_page() when done.
841 * For multi-segment *iter, this function only adds pages from the
842 * the next non-empty segment of the iov iterator.
844 static int __bio_iov_iter_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
846 unsigned short nr_pages
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
847 unsigned short entries_left
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
848 struct bio_vec
*bv
= bio
->bi_io_vec
+ bio
->bi_vcnt
;
849 struct page
**pages
= (struct page
**)bv
;
855 * Move page array up in the allocated memory for the bio vecs as far as
856 * possible so that we can start filling biovecs from the beginning
857 * without overwriting the temporary page array.
859 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC
< 2);
860 pages
+= entries_left
* (PAGE_PTRS_PER_BVEC
- 1);
862 size
= iov_iter_get_pages(iter
, pages
, LONG_MAX
, nr_pages
, &offset
);
863 if (unlikely(size
<= 0))
864 return size
? size
: -EFAULT
;
866 for (left
= size
, i
= 0; left
> 0; left
-= len
, i
++) {
867 struct page
*page
= pages
[i
];
869 len
= min_t(size_t, PAGE_SIZE
- offset
, left
);
870 if (WARN_ON_ONCE(bio_add_page(bio
, page
, len
, offset
) != len
))
875 iov_iter_advance(iter
, size
);
880 * bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
881 * @bio: bio to add pages to
882 * @iter: iov iterator describing the region to be mapped
884 * Pins pages from *iter and appends them to @bio's bvec array. The
885 * pages will have to be released using put_page() when done.
886 * The function tries, but does not guarantee, to pin as many pages as
887 * fit into the bio, or are requested in *iter, whatever is smaller.
888 * If MM encounters an error pinning the requested pages, it stops.
889 * Error is returned only if 0 pages could be pinned.
891 int bio_iov_iter_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
893 unsigned short orig_vcnt
= bio
->bi_vcnt
;
896 int ret
= __bio_iov_iter_get_pages(bio
, iter
);
899 return bio
->bi_vcnt
> orig_vcnt
? 0 : ret
;
901 } while (iov_iter_count(iter
) && !bio_full(bio
));
905 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages
);
907 static void submit_bio_wait_endio(struct bio
*bio
)
909 complete(bio
->bi_private
);
913 * submit_bio_wait - submit a bio, and wait until it completes
914 * @bio: The &struct bio which describes the I/O
916 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
917 * bio_endio() on failure.
919 * WARNING: Unlike to how submit_bio() is usually used, this function does not
920 * result in bio reference to be consumed. The caller must drop the reference
923 int submit_bio_wait(struct bio
*bio
)
925 DECLARE_COMPLETION_ONSTACK_MAP(done
, bio
->bi_disk
->lockdep_map
);
927 bio
->bi_private
= &done
;
928 bio
->bi_end_io
= submit_bio_wait_endio
;
929 bio
->bi_opf
|= REQ_SYNC
;
931 wait_for_completion_io(&done
);
933 return blk_status_to_errno(bio
->bi_status
);
935 EXPORT_SYMBOL(submit_bio_wait
);
938 * bio_advance - increment/complete a bio by some number of bytes
939 * @bio: bio to advance
940 * @bytes: number of bytes to complete
942 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
943 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
944 * be updated on the last bvec as well.
946 * @bio will then represent the remaining, uncompleted portion of the io.
948 void bio_advance(struct bio
*bio
, unsigned bytes
)
950 if (bio_integrity(bio
))
951 bio_integrity_advance(bio
, bytes
);
953 bio_advance_iter(bio
, &bio
->bi_iter
, bytes
);
955 EXPORT_SYMBOL(bio_advance
);
957 void bio_copy_data_iter(struct bio
*dst
, struct bvec_iter
*dst_iter
,
958 struct bio
*src
, struct bvec_iter
*src_iter
)
960 struct bio_vec src_bv
, dst_bv
;
964 while (src_iter
->bi_size
&& dst_iter
->bi_size
) {
965 src_bv
= bio_iter_iovec(src
, *src_iter
);
966 dst_bv
= bio_iter_iovec(dst
, *dst_iter
);
968 bytes
= min(src_bv
.bv_len
, dst_bv
.bv_len
);
970 src_p
= kmap_atomic(src_bv
.bv_page
);
971 dst_p
= kmap_atomic(dst_bv
.bv_page
);
973 memcpy(dst_p
+ dst_bv
.bv_offset
,
974 src_p
+ src_bv
.bv_offset
,
977 kunmap_atomic(dst_p
);
978 kunmap_atomic(src_p
);
980 flush_dcache_page(dst_bv
.bv_page
);
982 bio_advance_iter(src
, src_iter
, bytes
);
983 bio_advance_iter(dst
, dst_iter
, bytes
);
986 EXPORT_SYMBOL(bio_copy_data_iter
);
989 * bio_copy_data - copy contents of data buffers from one bio to another
991 * @dst: destination bio
993 * Stops when it reaches the end of either @src or @dst - that is, copies
994 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
996 void bio_copy_data(struct bio
*dst
, struct bio
*src
)
998 struct bvec_iter src_iter
= src
->bi_iter
;
999 struct bvec_iter dst_iter
= dst
->bi_iter
;
1001 bio_copy_data_iter(dst
, &dst_iter
, src
, &src_iter
);
1003 EXPORT_SYMBOL(bio_copy_data
);
1006 * bio_list_copy_data - copy contents of data buffers from one chain of bios to
1008 * @src: source bio list
1009 * @dst: destination bio list
1011 * Stops when it reaches the end of either the @src list or @dst list - that is,
1012 * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of
1015 void bio_list_copy_data(struct bio
*dst
, struct bio
*src
)
1017 struct bvec_iter src_iter
= src
->bi_iter
;
1018 struct bvec_iter dst_iter
= dst
->bi_iter
;
1021 if (!src_iter
.bi_size
) {
1026 src_iter
= src
->bi_iter
;
1029 if (!dst_iter
.bi_size
) {
1034 dst_iter
= dst
->bi_iter
;
1037 bio_copy_data_iter(dst
, &dst_iter
, src
, &src_iter
);
1040 EXPORT_SYMBOL(bio_list_copy_data
);
1042 struct bio_map_data
{
1044 struct iov_iter iter
;
1048 static struct bio_map_data
*bio_alloc_map_data(struct iov_iter
*data
,
1051 struct bio_map_data
*bmd
;
1052 if (data
->nr_segs
> UIO_MAXIOV
)
1055 bmd
= kmalloc(sizeof(struct bio_map_data
) +
1056 sizeof(struct iovec
) * data
->nr_segs
, gfp_mask
);
1059 memcpy(bmd
->iov
, data
->iov
, sizeof(struct iovec
) * data
->nr_segs
);
1061 bmd
->iter
.iov
= bmd
->iov
;
1066 * bio_copy_from_iter - copy all pages from iov_iter to bio
1067 * @bio: The &struct bio which describes the I/O as destination
1068 * @iter: iov_iter as source
1070 * Copy all pages from iov_iter to bio.
1071 * Returns 0 on success, or error on failure.
1073 static int bio_copy_from_iter(struct bio
*bio
, struct iov_iter
*iter
)
1076 struct bio_vec
*bvec
;
1078 bio_for_each_segment_all(bvec
, bio
, i
) {
1081 ret
= copy_page_from_iter(bvec
->bv_page
,
1086 if (!iov_iter_count(iter
))
1089 if (ret
< bvec
->bv_len
)
1097 * bio_copy_to_iter - copy all pages from bio to iov_iter
1098 * @bio: The &struct bio which describes the I/O as source
1099 * @iter: iov_iter as destination
1101 * Copy all pages from bio to iov_iter.
1102 * Returns 0 on success, or error on failure.
1104 static int bio_copy_to_iter(struct bio
*bio
, struct iov_iter iter
)
1107 struct bio_vec
*bvec
;
1109 bio_for_each_segment_all(bvec
, bio
, i
) {
1112 ret
= copy_page_to_iter(bvec
->bv_page
,
1117 if (!iov_iter_count(&iter
))
1120 if (ret
< bvec
->bv_len
)
1127 void bio_free_pages(struct bio
*bio
)
1129 struct bio_vec
*bvec
;
1132 bio_for_each_segment_all(bvec
, bio
, i
)
1133 __free_page(bvec
->bv_page
);
1135 EXPORT_SYMBOL(bio_free_pages
);
1138 * bio_uncopy_user - finish previously mapped bio
1139 * @bio: bio being terminated
1141 * Free pages allocated from bio_copy_user_iov() and write back data
1142 * to user space in case of a read.
1144 int bio_uncopy_user(struct bio
*bio
)
1146 struct bio_map_data
*bmd
= bio
->bi_private
;
1149 if (!bio_flagged(bio
, BIO_NULL_MAPPED
)) {
1151 * if we're in a workqueue, the request is orphaned, so
1152 * don't copy into a random user address space, just free
1153 * and return -EINTR so user space doesn't expect any data.
1157 else if (bio_data_dir(bio
) == READ
)
1158 ret
= bio_copy_to_iter(bio
, bmd
->iter
);
1159 if (bmd
->is_our_pages
)
1160 bio_free_pages(bio
);
1168 * bio_copy_user_iov - copy user data to bio
1169 * @q: destination block queue
1170 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1171 * @iter: iovec iterator
1172 * @gfp_mask: memory allocation flags
1174 * Prepares and returns a bio for indirect user io, bouncing data
1175 * to/from kernel pages as necessary. Must be paired with
1176 * call bio_uncopy_user() on io completion.
1178 struct bio
*bio_copy_user_iov(struct request_queue
*q
,
1179 struct rq_map_data
*map_data
,
1180 struct iov_iter
*iter
,
1183 struct bio_map_data
*bmd
;
1188 unsigned int len
= iter
->count
;
1189 unsigned int offset
= map_data
? offset_in_page(map_data
->offset
) : 0;
1191 bmd
= bio_alloc_map_data(iter
, gfp_mask
);
1193 return ERR_PTR(-ENOMEM
);
1196 * We need to do a deep copy of the iov_iter including the iovecs.
1197 * The caller provided iov might point to an on-stack or otherwise
1200 bmd
->is_our_pages
= map_data
? 0 : 1;
1202 nr_pages
= DIV_ROUND_UP(offset
+ len
, PAGE_SIZE
);
1203 if (nr_pages
> BIO_MAX_PAGES
)
1204 nr_pages
= BIO_MAX_PAGES
;
1207 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1214 nr_pages
= 1 << map_data
->page_order
;
1215 i
= map_data
->offset
/ PAGE_SIZE
;
1218 unsigned int bytes
= PAGE_SIZE
;
1226 if (i
== map_data
->nr_entries
* nr_pages
) {
1231 page
= map_data
->pages
[i
/ nr_pages
];
1232 page
+= (i
% nr_pages
);
1236 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
1243 if (bio_add_pc_page(q
, bio
, page
, bytes
, offset
) < bytes
)
1254 map_data
->offset
+= bio
->bi_iter
.bi_size
;
1259 if ((iov_iter_rw(iter
) == WRITE
&& (!map_data
|| !map_data
->null_mapped
)) ||
1260 (map_data
&& map_data
->from_user
)) {
1261 ret
= bio_copy_from_iter(bio
, iter
);
1265 iov_iter_advance(iter
, bio
->bi_iter
.bi_size
);
1268 bio
->bi_private
= bmd
;
1269 if (map_data
&& map_data
->null_mapped
)
1270 bio_set_flag(bio
, BIO_NULL_MAPPED
);
1274 bio_free_pages(bio
);
1278 return ERR_PTR(ret
);
1282 * bio_map_user_iov - map user iovec into bio
1283 * @q: the struct request_queue for the bio
1284 * @iter: iovec iterator
1285 * @gfp_mask: memory allocation flags
1287 * Map the user space address into a bio suitable for io to a block
1288 * device. Returns an error pointer in case of error.
1290 struct bio
*bio_map_user_iov(struct request_queue
*q
,
1291 struct iov_iter
*iter
,
1297 struct bio_vec
*bvec
;
1299 if (!iov_iter_count(iter
))
1300 return ERR_PTR(-EINVAL
);
1302 bio
= bio_kmalloc(gfp_mask
, iov_iter_npages(iter
, BIO_MAX_PAGES
));
1304 return ERR_PTR(-ENOMEM
);
1306 while (iov_iter_count(iter
)) {
1307 struct page
**pages
;
1309 size_t offs
, added
= 0;
1312 bytes
= iov_iter_get_pages_alloc(iter
, &pages
, LONG_MAX
, &offs
);
1313 if (unlikely(bytes
<= 0)) {
1314 ret
= bytes
? bytes
: -EFAULT
;
1318 npages
= DIV_ROUND_UP(offs
+ bytes
, PAGE_SIZE
);
1320 if (unlikely(offs
& queue_dma_alignment(q
))) {
1324 for (j
= 0; j
< npages
; j
++) {
1325 struct page
*page
= pages
[j
];
1326 unsigned int n
= PAGE_SIZE
- offs
;
1327 unsigned short prev_bi_vcnt
= bio
->bi_vcnt
;
1332 if (!bio_add_pc_page(q
, bio
, page
, n
, offs
))
1336 * check if vector was merged with previous
1337 * drop page reference if needed
1339 if (bio
->bi_vcnt
== prev_bi_vcnt
)
1346 iov_iter_advance(iter
, added
);
1349 * release the pages we didn't map into the bio, if any
1352 put_page(pages
[j
++]);
1354 /* couldn't stuff something into bio? */
1359 bio_set_flag(bio
, BIO_USER_MAPPED
);
1362 * subtle -- if bio_map_user_iov() ended up bouncing a bio,
1363 * it would normally disappear when its bi_end_io is run.
1364 * however, we need it for the unmap, so grab an extra
1371 bio_for_each_segment_all(bvec
, bio
, j
) {
1372 put_page(bvec
->bv_page
);
1375 return ERR_PTR(ret
);
1378 static void __bio_unmap_user(struct bio
*bio
)
1380 struct bio_vec
*bvec
;
1384 * make sure we dirty pages we wrote to
1386 bio_for_each_segment_all(bvec
, bio
, i
) {
1387 if (bio_data_dir(bio
) == READ
)
1388 set_page_dirty_lock(bvec
->bv_page
);
1390 put_page(bvec
->bv_page
);
1397 * bio_unmap_user - unmap a bio
1398 * @bio: the bio being unmapped
1400 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from
1403 * bio_unmap_user() may sleep.
1405 void bio_unmap_user(struct bio
*bio
)
1407 __bio_unmap_user(bio
);
1411 static void bio_map_kern_endio(struct bio
*bio
)
1417 * bio_map_kern - map kernel address into bio
1418 * @q: the struct request_queue for the bio
1419 * @data: pointer to buffer to map
1420 * @len: length in bytes
1421 * @gfp_mask: allocation flags for bio allocation
1423 * Map the kernel address into a bio suitable for io to a block
1424 * device. Returns an error pointer in case of error.
1426 struct bio
*bio_map_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1429 unsigned long kaddr
= (unsigned long)data
;
1430 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1431 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1432 const int nr_pages
= end
- start
;
1436 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1438 return ERR_PTR(-ENOMEM
);
1440 offset
= offset_in_page(kaddr
);
1441 for (i
= 0; i
< nr_pages
; i
++) {
1442 unsigned int bytes
= PAGE_SIZE
- offset
;
1450 if (bio_add_pc_page(q
, bio
, virt_to_page(data
), bytes
,
1452 /* we don't support partial mappings */
1454 return ERR_PTR(-EINVAL
);
1462 bio
->bi_end_io
= bio_map_kern_endio
;
1465 EXPORT_SYMBOL(bio_map_kern
);
1467 static void bio_copy_kern_endio(struct bio
*bio
)
1469 bio_free_pages(bio
);
1473 static void bio_copy_kern_endio_read(struct bio
*bio
)
1475 char *p
= bio
->bi_private
;
1476 struct bio_vec
*bvec
;
1479 bio_for_each_segment_all(bvec
, bio
, i
) {
1480 memcpy(p
, page_address(bvec
->bv_page
), bvec
->bv_len
);
1484 bio_copy_kern_endio(bio
);
1488 * bio_copy_kern - copy kernel address into bio
1489 * @q: the struct request_queue for the bio
1490 * @data: pointer to buffer to copy
1491 * @len: length in bytes
1492 * @gfp_mask: allocation flags for bio and page allocation
1493 * @reading: data direction is READ
1495 * copy the kernel address into a bio suitable for io to a block
1496 * device. Returns an error pointer in case of error.
1498 struct bio
*bio_copy_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1499 gfp_t gfp_mask
, int reading
)
1501 unsigned long kaddr
= (unsigned long)data
;
1502 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1503 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1512 return ERR_PTR(-EINVAL
);
1514 nr_pages
= end
- start
;
1515 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1517 return ERR_PTR(-ENOMEM
);
1521 unsigned int bytes
= PAGE_SIZE
;
1526 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
1531 memcpy(page_address(page
), p
, bytes
);
1533 if (bio_add_pc_page(q
, bio
, page
, bytes
, 0) < bytes
)
1541 bio
->bi_end_io
= bio_copy_kern_endio_read
;
1542 bio
->bi_private
= data
;
1544 bio
->bi_end_io
= bio_copy_kern_endio
;
1550 bio_free_pages(bio
);
1552 return ERR_PTR(-ENOMEM
);
1556 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1557 * for performing direct-IO in BIOs.
1559 * The problem is that we cannot run set_page_dirty() from interrupt context
1560 * because the required locks are not interrupt-safe. So what we can do is to
1561 * mark the pages dirty _before_ performing IO. And in interrupt context,
1562 * check that the pages are still dirty. If so, fine. If not, redirty them
1563 * in process context.
1565 * We special-case compound pages here: normally this means reads into hugetlb
1566 * pages. The logic in here doesn't really work right for compound pages
1567 * because the VM does not uniformly chase down the head page in all cases.
1568 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1569 * handle them at all. So we skip compound pages here at an early stage.
1571 * Note that this code is very hard to test under normal circumstances because
1572 * direct-io pins the pages with get_user_pages(). This makes
1573 * is_page_cache_freeable return false, and the VM will not clean the pages.
1574 * But other code (eg, flusher threads) could clean the pages if they are mapped
1577 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1578 * deferred bio dirtying paths.
1582 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1584 void bio_set_pages_dirty(struct bio
*bio
)
1586 struct bio_vec
*bvec
;
1589 bio_for_each_segment_all(bvec
, bio
, i
) {
1590 if (!PageCompound(bvec
->bv_page
))
1591 set_page_dirty_lock(bvec
->bv_page
);
1594 EXPORT_SYMBOL_GPL(bio_set_pages_dirty
);
1596 static void bio_release_pages(struct bio
*bio
)
1598 struct bio_vec
*bvec
;
1601 bio_for_each_segment_all(bvec
, bio
, i
)
1602 put_page(bvec
->bv_page
);
1606 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1607 * If they are, then fine. If, however, some pages are clean then they must
1608 * have been written out during the direct-IO read. So we take another ref on
1609 * the BIO and re-dirty the pages in process context.
1611 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1612 * here on. It will run one put_page() against each page and will run one
1613 * bio_put() against the BIO.
1616 static void bio_dirty_fn(struct work_struct
*work
);
1618 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
1619 static DEFINE_SPINLOCK(bio_dirty_lock
);
1620 static struct bio
*bio_dirty_list
;
1623 * This runs in process context
1625 static void bio_dirty_fn(struct work_struct
*work
)
1627 struct bio
*bio
, *next
;
1629 spin_lock_irq(&bio_dirty_lock
);
1630 next
= bio_dirty_list
;
1631 bio_dirty_list
= NULL
;
1632 spin_unlock_irq(&bio_dirty_lock
);
1634 while ((bio
= next
) != NULL
) {
1635 next
= bio
->bi_private
;
1637 bio_set_pages_dirty(bio
);
1638 bio_release_pages(bio
);
1643 void bio_check_pages_dirty(struct bio
*bio
)
1645 struct bio_vec
*bvec
;
1646 unsigned long flags
;
1649 bio_for_each_segment_all(bvec
, bio
, i
) {
1650 if (!PageDirty(bvec
->bv_page
) && !PageCompound(bvec
->bv_page
))
1654 bio_release_pages(bio
);
1658 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1659 bio
->bi_private
= bio_dirty_list
;
1660 bio_dirty_list
= bio
;
1661 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1662 schedule_work(&bio_dirty_work
);
1664 EXPORT_SYMBOL_GPL(bio_check_pages_dirty
);
1666 void generic_start_io_acct(struct request_queue
*q
, int op
,
1667 unsigned long sectors
, struct hd_struct
*part
)
1669 const int sgrp
= op_stat_group(op
);
1670 int cpu
= part_stat_lock();
1672 part_round_stats(q
, cpu
, part
);
1673 part_stat_inc(cpu
, part
, ios
[sgrp
]);
1674 part_stat_add(cpu
, part
, sectors
[sgrp
], sectors
);
1675 part_inc_in_flight(q
, part
, op_is_write(op
));
1679 EXPORT_SYMBOL(generic_start_io_acct
);
1681 void generic_end_io_acct(struct request_queue
*q
, int req_op
,
1682 struct hd_struct
*part
, unsigned long start_time
)
1684 unsigned long duration
= jiffies
- start_time
;
1685 const int sgrp
= op_stat_group(req_op
);
1686 int cpu
= part_stat_lock();
1688 part_stat_add(cpu
, part
, nsecs
[sgrp
], jiffies_to_nsecs(duration
));
1689 part_round_stats(q
, cpu
, part
);
1690 part_dec_in_flight(q
, part
, op_is_write(req_op
));
1694 EXPORT_SYMBOL(generic_end_io_acct
);
1696 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1697 void bio_flush_dcache_pages(struct bio
*bi
)
1699 struct bio_vec bvec
;
1700 struct bvec_iter iter
;
1702 bio_for_each_segment(bvec
, bi
, iter
)
1703 flush_dcache_page(bvec
.bv_page
);
1705 EXPORT_SYMBOL(bio_flush_dcache_pages
);
1708 static inline bool bio_remaining_done(struct bio
*bio
)
1711 * If we're not chaining, then ->__bi_remaining is always 1 and
1712 * we always end io on the first invocation.
1714 if (!bio_flagged(bio
, BIO_CHAIN
))
1717 BUG_ON(atomic_read(&bio
->__bi_remaining
) <= 0);
1719 if (atomic_dec_and_test(&bio
->__bi_remaining
)) {
1720 bio_clear_flag(bio
, BIO_CHAIN
);
1728 * bio_endio - end I/O on a bio
1732 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1733 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1734 * bio unless they own it and thus know that it has an end_io function.
1736 * bio_endio() can be called several times on a bio that has been chained
1737 * using bio_chain(). The ->bi_end_io() function will only be called the
1738 * last time. At this point the BLK_TA_COMPLETE tracing event will be
1739 * generated if BIO_TRACE_COMPLETION is set.
1741 void bio_endio(struct bio
*bio
)
1744 if (!bio_remaining_done(bio
))
1746 if (!bio_integrity_endio(bio
))
1750 rq_qos_done_bio(bio
->bi_disk
->queue
, bio
);
1753 * Need to have a real endio function for chained bios, otherwise
1754 * various corner cases will break (like stacking block devices that
1755 * save/restore bi_end_io) - however, we want to avoid unbounded
1756 * recursion and blowing the stack. Tail call optimization would
1757 * handle this, but compiling with frame pointers also disables
1758 * gcc's sibling call optimization.
1760 if (bio
->bi_end_io
== bio_chain_endio
) {
1761 bio
= __bio_chain_endio(bio
);
1765 if (bio
->bi_disk
&& bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
1766 trace_block_bio_complete(bio
->bi_disk
->queue
, bio
,
1767 blk_status_to_errno(bio
->bi_status
));
1768 bio_clear_flag(bio
, BIO_TRACE_COMPLETION
);
1771 blk_throtl_bio_endio(bio
);
1772 /* release cgroup info */
1775 bio
->bi_end_io(bio
);
1777 EXPORT_SYMBOL(bio_endio
);
1780 * bio_split - split a bio
1781 * @bio: bio to split
1782 * @sectors: number of sectors to split from the front of @bio
1784 * @bs: bio set to allocate from
1786 * Allocates and returns a new bio which represents @sectors from the start of
1787 * @bio, and updates @bio to represent the remaining sectors.
1789 * Unless this is a discard request the newly allocated bio will point
1790 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1791 * @bio is not freed before the split.
1793 struct bio
*bio_split(struct bio
*bio
, int sectors
,
1794 gfp_t gfp
, struct bio_set
*bs
)
1798 BUG_ON(sectors
<= 0);
1799 BUG_ON(sectors
>= bio_sectors(bio
));
1801 split
= bio_clone_fast(bio
, gfp
, bs
);
1805 split
->bi_iter
.bi_size
= sectors
<< 9;
1807 if (bio_integrity(split
))
1808 bio_integrity_trim(split
);
1810 bio_advance(bio
, split
->bi_iter
.bi_size
);
1812 if (bio_flagged(bio
, BIO_TRACE_COMPLETION
))
1813 bio_set_flag(split
, BIO_TRACE_COMPLETION
);
1817 EXPORT_SYMBOL(bio_split
);
1820 * bio_trim - trim a bio
1822 * @offset: number of sectors to trim from the front of @bio
1823 * @size: size we want to trim @bio to, in sectors
1825 void bio_trim(struct bio
*bio
, int offset
, int size
)
1827 /* 'bio' is a cloned bio which we need to trim to match
1828 * the given offset and size.
1832 if (offset
== 0 && size
== bio
->bi_iter
.bi_size
)
1835 bio_clear_flag(bio
, BIO_SEG_VALID
);
1837 bio_advance(bio
, offset
<< 9);
1839 bio
->bi_iter
.bi_size
= size
;
1841 if (bio_integrity(bio
))
1842 bio_integrity_trim(bio
);
1845 EXPORT_SYMBOL_GPL(bio_trim
);
1848 * create memory pools for biovec's in a bio_set.
1849 * use the global biovec slabs created for general use.
1851 int biovec_init_pool(mempool_t
*pool
, int pool_entries
)
1853 struct biovec_slab
*bp
= bvec_slabs
+ BVEC_POOL_MAX
;
1855 return mempool_init_slab_pool(pool
, pool_entries
, bp
->slab
);
1859 * bioset_exit - exit a bioset initialized with bioset_init()
1861 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1864 void bioset_exit(struct bio_set
*bs
)
1866 if (bs
->rescue_workqueue
)
1867 destroy_workqueue(bs
->rescue_workqueue
);
1868 bs
->rescue_workqueue
= NULL
;
1870 mempool_exit(&bs
->bio_pool
);
1871 mempool_exit(&bs
->bvec_pool
);
1873 bioset_integrity_free(bs
);
1876 bs
->bio_slab
= NULL
;
1878 EXPORT_SYMBOL(bioset_exit
);
1881 * bioset_init - Initialize a bio_set
1882 * @bs: pool to initialize
1883 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1884 * @front_pad: Number of bytes to allocate in front of the returned bio
1885 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1886 * and %BIOSET_NEED_RESCUER
1889 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1890 * to ask for a number of bytes to be allocated in front of the bio.
1891 * Front pad allocation is useful for embedding the bio inside
1892 * another structure, to avoid allocating extra data to go with the bio.
1893 * Note that the bio must be embedded at the END of that structure always,
1894 * or things will break badly.
1895 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1896 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast().
1897 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
1898 * dispatch queued requests when the mempool runs out of space.
1901 int bioset_init(struct bio_set
*bs
,
1902 unsigned int pool_size
,
1903 unsigned int front_pad
,
1906 unsigned int back_pad
= BIO_INLINE_VECS
* sizeof(struct bio_vec
);
1908 bs
->front_pad
= front_pad
;
1910 spin_lock_init(&bs
->rescue_lock
);
1911 bio_list_init(&bs
->rescue_list
);
1912 INIT_WORK(&bs
->rescue_work
, bio_alloc_rescue
);
1914 bs
->bio_slab
= bio_find_or_create_slab(front_pad
+ back_pad
);
1918 if (mempool_init_slab_pool(&bs
->bio_pool
, pool_size
, bs
->bio_slab
))
1921 if ((flags
& BIOSET_NEED_BVECS
) &&
1922 biovec_init_pool(&bs
->bvec_pool
, pool_size
))
1925 if (!(flags
& BIOSET_NEED_RESCUER
))
1928 bs
->rescue_workqueue
= alloc_workqueue("bioset", WQ_MEM_RECLAIM
, 0);
1929 if (!bs
->rescue_workqueue
)
1937 EXPORT_SYMBOL(bioset_init
);
1940 * Initialize and setup a new bio_set, based on the settings from
1943 int bioset_init_from_src(struct bio_set
*bs
, struct bio_set
*src
)
1948 if (src
->bvec_pool
.min_nr
)
1949 flags
|= BIOSET_NEED_BVECS
;
1950 if (src
->rescue_workqueue
)
1951 flags
|= BIOSET_NEED_RESCUER
;
1953 return bioset_init(bs
, src
->bio_pool
.min_nr
, src
->front_pad
, flags
);
1955 EXPORT_SYMBOL(bioset_init_from_src
);
1957 #ifdef CONFIG_BLK_CGROUP
1960 * bio_associate_blkg - associate a bio with the a blkg
1962 * @blkg: the blkg to associate
1964 * This tries to associate @bio with the specified blkg. Association failure
1965 * is handled by walking up the blkg tree. Therefore, the blkg associated can
1966 * be anything between @blkg and the root_blkg. This situation only happens
1967 * when a cgroup is dying and then the remaining bios will spill to the closest
1970 * A reference will be taken on the @blkg and will be released when @bio is
1973 int bio_associate_blkg(struct bio
*bio
, struct blkcg_gq
*blkg
)
1975 if (unlikely(bio
->bi_blkg
))
1977 bio
->bi_blkg
= blkg_tryget_closest(blkg
);
1982 * __bio_associate_blkg_from_css - internal blkg association function
1984 * This in the core association function that all association paths rely on.
1985 * A blkg reference is taken which is released upon freeing of the bio.
1987 static int __bio_associate_blkg_from_css(struct bio
*bio
,
1988 struct cgroup_subsys_state
*css
)
1990 struct request_queue
*q
= bio
->bi_disk
->queue
;
1991 struct blkcg_gq
*blkg
;
1996 if (!css
|| !css
->parent
)
1997 blkg
= q
->root_blkg
;
1999 blkg
= blkg_lookup_create(css_to_blkcg(css
), q
);
2001 ret
= bio_associate_blkg(bio
, blkg
);
2008 * bio_associate_blkg_from_css - associate a bio with a specified css
2012 * Associate @bio with the blkg found by combining the css's blkg and the
2013 * request_queue of the @bio. This falls back to the queue's root_blkg if
2014 * the association fails with the css.
2016 int bio_associate_blkg_from_css(struct bio
*bio
,
2017 struct cgroup_subsys_state
*css
)
2019 if (unlikely(bio
->bi_blkg
))
2021 return __bio_associate_blkg_from_css(bio
, css
);
2023 EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css
);
2027 * bio_associate_blkg_from_page - associate a bio with the page's blkg
2029 * @page: the page to lookup the blkcg from
2031 * Associate @bio with the blkg from @page's owning memcg and the respective
2032 * request_queue. If cgroup_e_css returns NULL, fall back to the queue's
2035 * Note: this must be called after bio has an associated device.
2037 int bio_associate_blkg_from_page(struct bio
*bio
, struct page
*page
)
2039 struct cgroup_subsys_state
*css
;
2042 if (unlikely(bio
->bi_blkg
))
2044 if (!page
->mem_cgroup
)
2049 css
= cgroup_e_css(page
->mem_cgroup
->css
.cgroup
, &io_cgrp_subsys
);
2051 ret
= __bio_associate_blkg_from_css(bio
, css
);
2056 #endif /* CONFIG_MEMCG */
2059 * bio_associate_create_blkg - associate a bio with a blkg from q
2060 * @q: request_queue where bio is going
2063 * Associate @bio with the blkg found from the bio's css and the request_queue.
2064 * If one is not found, bio_lookup_blkg creates the blkg. This falls back to
2065 * the queue's root_blkg if association fails.
2067 int bio_associate_create_blkg(struct request_queue
*q
, struct bio
*bio
)
2069 struct cgroup_subsys_state
*css
;
2072 /* someone has already associated this bio with a blkg */
2080 ret
= __bio_associate_blkg_from_css(bio
, css
);
2087 * bio_reassociate_blkg - reassociate a bio with a blkg from q
2088 * @q: request_queue where bio is going
2091 * When submitting a bio, multiple recursive calls to make_request() may occur.
2092 * This causes the initial associate done in blkcg_bio_issue_check() to be
2093 * incorrect and reference the prior request_queue. This performs reassociation
2094 * when this situation happens.
2096 int bio_reassociate_blkg(struct request_queue
*q
, struct bio
*bio
)
2099 blkg_put(bio
->bi_blkg
);
2100 bio
->bi_blkg
= NULL
;
2103 return bio_associate_create_blkg(q
, bio
);
2107 * bio_disassociate_task - undo bio_associate_current()
2110 void bio_disassociate_task(struct bio
*bio
)
2113 put_io_context(bio
->bi_ioc
);
2117 blkg_put(bio
->bi_blkg
);
2118 bio
->bi_blkg
= NULL
;
2123 * bio_clone_blkg_association - clone blkg association from src to dst bio
2124 * @dst: destination bio
2127 void bio_clone_blkg_association(struct bio
*dst
, struct bio
*src
)
2130 bio_associate_blkg(dst
, src
->bi_blkg
);
2132 EXPORT_SYMBOL_GPL(bio_clone_blkg_association
);
2133 #endif /* CONFIG_BLK_CGROUP */
2135 static void __init
biovec_init_slabs(void)
2139 for (i
= 0; i
< BVEC_POOL_NR
; i
++) {
2141 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
2143 if (bvs
->nr_vecs
<= BIO_INLINE_VECS
) {
2148 size
= bvs
->nr_vecs
* sizeof(struct bio_vec
);
2149 bvs
->slab
= kmem_cache_create(bvs
->name
, size
, 0,
2150 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
2154 static int __init
init_bio(void)
2158 bio_slabs
= kcalloc(bio_slab_max
, sizeof(struct bio_slab
),
2161 panic("bio: can't allocate bios\n");
2163 bio_integrity_init();
2164 biovec_init_slabs();
2166 if (bioset_init(&fs_bio_set
, BIO_POOL_SIZE
, 0, BIOSET_NEED_BVECS
))
2167 panic("bio: can't allocate bios\n");
2169 if (bioset_integrity_create(&fs_bio_set
, BIO_POOL_SIZE
))
2170 panic("bio: can't create integrity pool\n");
2174 subsys_initcall(init_bio
);