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_blkcg_association(bio
, bio_src
);
614 EXPORT_SYMBOL(__bio_clone_fast
);
617 * bio_clone_fast - clone a bio that shares the original bio's biovec
619 * @gfp_mask: allocation priority
620 * @bs: bio_set to allocate from
622 * Like __bio_clone_fast, only also allocates the returned bio
624 struct bio
*bio_clone_fast(struct bio
*bio
, gfp_t gfp_mask
, struct bio_set
*bs
)
628 b
= bio_alloc_bioset(gfp_mask
, 0, bs
);
632 __bio_clone_fast(b
, bio
);
634 if (bio_integrity(bio
)) {
637 ret
= bio_integrity_clone(b
, bio
, gfp_mask
);
647 EXPORT_SYMBOL(bio_clone_fast
);
650 * bio_add_pc_page - attempt to add page to bio
651 * @q: the target queue
652 * @bio: destination bio
654 * @len: vec entry length
655 * @offset: vec entry offset
657 * Attempt to add a page to the bio_vec maplist. This can fail for a
658 * number of reasons, such as the bio being full or target block device
659 * limitations. The target block device must allow bio's up to PAGE_SIZE,
660 * so it is always possible to add a single page to an empty bio.
662 * This should only be used by REQ_PC bios.
664 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
, struct page
665 *page
, unsigned int len
, unsigned int offset
)
667 int retried_segments
= 0;
668 struct bio_vec
*bvec
;
671 * cloned bio must not modify vec list
673 if (unlikely(bio_flagged(bio
, BIO_CLONED
)))
676 if (((bio
->bi_iter
.bi_size
+ len
) >> 9) > queue_max_hw_sectors(q
))
680 * For filesystems with a blocksize smaller than the pagesize
681 * we will often be called with the same page as last time and
682 * a consecutive offset. Optimize this special case.
684 if (bio
->bi_vcnt
> 0) {
685 struct bio_vec
*prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
687 if (page
== prev
->bv_page
&&
688 offset
== prev
->bv_offset
+ prev
->bv_len
) {
690 bio
->bi_iter
.bi_size
+= len
;
695 * If the queue doesn't support SG gaps and adding this
696 * offset would create a gap, disallow it.
698 if (bvec_gap_to_prev(q
, prev
, offset
))
706 * setup the new entry, we might clear it again later if we
707 * cannot add the page
709 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
710 bvec
->bv_page
= page
;
712 bvec
->bv_offset
= offset
;
714 bio
->bi_phys_segments
++;
715 bio
->bi_iter
.bi_size
+= len
;
718 * Perform a recount if the number of segments is greater
719 * than queue_max_segments(q).
722 while (bio
->bi_phys_segments
> queue_max_segments(q
)) {
724 if (retried_segments
)
727 retried_segments
= 1;
728 blk_recount_segments(q
, bio
);
731 /* If we may be able to merge these biovecs, force a recount */
732 if (bio
->bi_vcnt
> 1 && (BIOVEC_PHYS_MERGEABLE(bvec
-1, bvec
)))
733 bio_clear_flag(bio
, BIO_SEG_VALID
);
739 bvec
->bv_page
= NULL
;
743 bio
->bi_iter
.bi_size
-= len
;
744 blk_recount_segments(q
, bio
);
747 EXPORT_SYMBOL(bio_add_pc_page
);
750 * __bio_try_merge_page - try appending data to an existing bvec.
751 * @bio: destination bio
753 * @len: length of the data to add
754 * @off: offset of the data in @page
756 * Try to add the data at @page + @off to the last bvec of @bio. This is a
757 * a useful optimisation for file systems with a block size smaller than the
760 * Return %true on success or %false on failure.
762 bool __bio_try_merge_page(struct bio
*bio
, struct page
*page
,
763 unsigned int len
, unsigned int off
)
765 if (WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
)))
768 if (bio
->bi_vcnt
> 0) {
769 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
771 if (page
== bv
->bv_page
&& off
== bv
->bv_offset
+ bv
->bv_len
) {
773 bio
->bi_iter
.bi_size
+= len
;
779 EXPORT_SYMBOL_GPL(__bio_try_merge_page
);
782 * __bio_add_page - add page to a bio in a new segment
783 * @bio: destination bio
785 * @len: length of the data to add
786 * @off: offset of the data in @page
788 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
789 * that @bio has space for another bvec.
791 void __bio_add_page(struct bio
*bio
, struct page
*page
,
792 unsigned int len
, unsigned int off
)
794 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
796 WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
));
797 WARN_ON_ONCE(bio_full(bio
));
803 bio
->bi_iter
.bi_size
+= len
;
806 EXPORT_SYMBOL_GPL(__bio_add_page
);
809 * bio_add_page - attempt to add page to bio
810 * @bio: destination bio
812 * @len: vec entry length
813 * @offset: vec entry offset
815 * Attempt to add a page to the bio_vec maplist. This will only fail
816 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
818 int bio_add_page(struct bio
*bio
, struct page
*page
,
819 unsigned int len
, unsigned int offset
)
821 if (!__bio_try_merge_page(bio
, page
, len
, offset
)) {
824 __bio_add_page(bio
, page
, len
, offset
);
828 EXPORT_SYMBOL(bio_add_page
);
830 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
833 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
834 * @bio: bio to add pages to
835 * @iter: iov iterator describing the region to be mapped
837 * Pins pages from *iter and appends them to @bio's bvec array. The
838 * pages will have to be released using put_page() when done.
839 * For multi-segment *iter, this function only adds pages from the
840 * the next non-empty segment of the iov iterator.
842 static int __bio_iov_iter_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
844 unsigned short nr_pages
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
845 unsigned short entries_left
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
846 struct bio_vec
*bv
= bio
->bi_io_vec
+ bio
->bi_vcnt
;
847 struct page
**pages
= (struct page
**)bv
;
853 * Move page array up in the allocated memory for the bio vecs as far as
854 * possible so that we can start filling biovecs from the beginning
855 * without overwriting the temporary page array.
857 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC
< 2);
858 pages
+= entries_left
* (PAGE_PTRS_PER_BVEC
- 1);
860 size
= iov_iter_get_pages(iter
, pages
, LONG_MAX
, nr_pages
, &offset
);
861 if (unlikely(size
<= 0))
862 return size
? size
: -EFAULT
;
864 for (left
= size
, i
= 0; left
> 0; left
-= len
, i
++) {
865 struct page
*page
= pages
[i
];
867 len
= min_t(size_t, PAGE_SIZE
- offset
, left
);
868 if (WARN_ON_ONCE(bio_add_page(bio
, page
, len
, offset
) != len
))
873 iov_iter_advance(iter
, size
);
878 * bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
879 * @bio: bio to add pages to
880 * @iter: iov iterator describing the region to be mapped
882 * Pins pages from *iter and appends them to @bio's bvec array. The
883 * pages will have to be released using put_page() when done.
884 * The function tries, but does not guarantee, to pin as many pages as
885 * fit into the bio, or are requested in *iter, whatever is smaller.
886 * If MM encounters an error pinning the requested pages, it stops.
887 * Error is returned only if 0 pages could be pinned.
889 int bio_iov_iter_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
891 unsigned short orig_vcnt
= bio
->bi_vcnt
;
894 int ret
= __bio_iov_iter_get_pages(bio
, iter
);
897 return bio
->bi_vcnt
> orig_vcnt
? 0 : ret
;
899 } while (iov_iter_count(iter
) && !bio_full(bio
));
903 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages
);
905 static void submit_bio_wait_endio(struct bio
*bio
)
907 complete(bio
->bi_private
);
911 * submit_bio_wait - submit a bio, and wait until it completes
912 * @bio: The &struct bio which describes the I/O
914 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
915 * bio_endio() on failure.
917 * WARNING: Unlike to how submit_bio() is usually used, this function does not
918 * result in bio reference to be consumed. The caller must drop the reference
921 int submit_bio_wait(struct bio
*bio
)
923 DECLARE_COMPLETION_ONSTACK_MAP(done
, bio
->bi_disk
->lockdep_map
);
925 bio
->bi_private
= &done
;
926 bio
->bi_end_io
= submit_bio_wait_endio
;
927 bio
->bi_opf
|= REQ_SYNC
;
929 wait_for_completion_io(&done
);
931 return blk_status_to_errno(bio
->bi_status
);
933 EXPORT_SYMBOL(submit_bio_wait
);
936 * bio_advance - increment/complete a bio by some number of bytes
937 * @bio: bio to advance
938 * @bytes: number of bytes to complete
940 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
941 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
942 * be updated on the last bvec as well.
944 * @bio will then represent the remaining, uncompleted portion of the io.
946 void bio_advance(struct bio
*bio
, unsigned bytes
)
948 if (bio_integrity(bio
))
949 bio_integrity_advance(bio
, bytes
);
951 bio_advance_iter(bio
, &bio
->bi_iter
, bytes
);
953 EXPORT_SYMBOL(bio_advance
);
955 void bio_copy_data_iter(struct bio
*dst
, struct bvec_iter
*dst_iter
,
956 struct bio
*src
, struct bvec_iter
*src_iter
)
958 struct bio_vec src_bv
, dst_bv
;
962 while (src_iter
->bi_size
&& dst_iter
->bi_size
) {
963 src_bv
= bio_iter_iovec(src
, *src_iter
);
964 dst_bv
= bio_iter_iovec(dst
, *dst_iter
);
966 bytes
= min(src_bv
.bv_len
, dst_bv
.bv_len
);
968 src_p
= kmap_atomic(src_bv
.bv_page
);
969 dst_p
= kmap_atomic(dst_bv
.bv_page
);
971 memcpy(dst_p
+ dst_bv
.bv_offset
,
972 src_p
+ src_bv
.bv_offset
,
975 kunmap_atomic(dst_p
);
976 kunmap_atomic(src_p
);
978 flush_dcache_page(dst_bv
.bv_page
);
980 bio_advance_iter(src
, src_iter
, bytes
);
981 bio_advance_iter(dst
, dst_iter
, bytes
);
984 EXPORT_SYMBOL(bio_copy_data_iter
);
987 * bio_copy_data - copy contents of data buffers from one bio to another
989 * @dst: destination bio
991 * Stops when it reaches the end of either @src or @dst - that is, copies
992 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
994 void bio_copy_data(struct bio
*dst
, struct bio
*src
)
996 struct bvec_iter src_iter
= src
->bi_iter
;
997 struct bvec_iter dst_iter
= dst
->bi_iter
;
999 bio_copy_data_iter(dst
, &dst_iter
, src
, &src_iter
);
1001 EXPORT_SYMBOL(bio_copy_data
);
1004 * bio_list_copy_data - copy contents of data buffers from one chain of bios to
1006 * @src: source bio list
1007 * @dst: destination bio list
1009 * Stops when it reaches the end of either the @src list or @dst list - that is,
1010 * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of
1013 void bio_list_copy_data(struct bio
*dst
, struct bio
*src
)
1015 struct bvec_iter src_iter
= src
->bi_iter
;
1016 struct bvec_iter dst_iter
= dst
->bi_iter
;
1019 if (!src_iter
.bi_size
) {
1024 src_iter
= src
->bi_iter
;
1027 if (!dst_iter
.bi_size
) {
1032 dst_iter
= dst
->bi_iter
;
1035 bio_copy_data_iter(dst
, &dst_iter
, src
, &src_iter
);
1038 EXPORT_SYMBOL(bio_list_copy_data
);
1040 struct bio_map_data
{
1042 struct iov_iter iter
;
1046 static struct bio_map_data
*bio_alloc_map_data(struct iov_iter
*data
,
1049 struct bio_map_data
*bmd
;
1050 if (data
->nr_segs
> UIO_MAXIOV
)
1053 bmd
= kmalloc(sizeof(struct bio_map_data
) +
1054 sizeof(struct iovec
) * data
->nr_segs
, gfp_mask
);
1057 memcpy(bmd
->iov
, data
->iov
, sizeof(struct iovec
) * data
->nr_segs
);
1059 bmd
->iter
.iov
= bmd
->iov
;
1064 * bio_copy_from_iter - copy all pages from iov_iter to bio
1065 * @bio: The &struct bio which describes the I/O as destination
1066 * @iter: iov_iter as source
1068 * Copy all pages from iov_iter to bio.
1069 * Returns 0 on success, or error on failure.
1071 static int bio_copy_from_iter(struct bio
*bio
, struct iov_iter
*iter
)
1074 struct bio_vec
*bvec
;
1076 bio_for_each_segment_all(bvec
, bio
, i
) {
1079 ret
= copy_page_from_iter(bvec
->bv_page
,
1084 if (!iov_iter_count(iter
))
1087 if (ret
< bvec
->bv_len
)
1095 * bio_copy_to_iter - copy all pages from bio to iov_iter
1096 * @bio: The &struct bio which describes the I/O as source
1097 * @iter: iov_iter as destination
1099 * Copy all pages from bio to iov_iter.
1100 * Returns 0 on success, or error on failure.
1102 static int bio_copy_to_iter(struct bio
*bio
, struct iov_iter iter
)
1105 struct bio_vec
*bvec
;
1107 bio_for_each_segment_all(bvec
, bio
, i
) {
1110 ret
= copy_page_to_iter(bvec
->bv_page
,
1115 if (!iov_iter_count(&iter
))
1118 if (ret
< bvec
->bv_len
)
1125 void bio_free_pages(struct bio
*bio
)
1127 struct bio_vec
*bvec
;
1130 bio_for_each_segment_all(bvec
, bio
, i
)
1131 __free_page(bvec
->bv_page
);
1133 EXPORT_SYMBOL(bio_free_pages
);
1136 * bio_uncopy_user - finish previously mapped bio
1137 * @bio: bio being terminated
1139 * Free pages allocated from bio_copy_user_iov() and write back data
1140 * to user space in case of a read.
1142 int bio_uncopy_user(struct bio
*bio
)
1144 struct bio_map_data
*bmd
= bio
->bi_private
;
1147 if (!bio_flagged(bio
, BIO_NULL_MAPPED
)) {
1149 * if we're in a workqueue, the request is orphaned, so
1150 * don't copy into a random user address space, just free
1151 * and return -EINTR so user space doesn't expect any data.
1155 else if (bio_data_dir(bio
) == READ
)
1156 ret
= bio_copy_to_iter(bio
, bmd
->iter
);
1157 if (bmd
->is_our_pages
)
1158 bio_free_pages(bio
);
1166 * bio_copy_user_iov - copy user data to bio
1167 * @q: destination block queue
1168 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1169 * @iter: iovec iterator
1170 * @gfp_mask: memory allocation flags
1172 * Prepares and returns a bio for indirect user io, bouncing data
1173 * to/from kernel pages as necessary. Must be paired with
1174 * call bio_uncopy_user() on io completion.
1176 struct bio
*bio_copy_user_iov(struct request_queue
*q
,
1177 struct rq_map_data
*map_data
,
1178 struct iov_iter
*iter
,
1181 struct bio_map_data
*bmd
;
1186 unsigned int len
= iter
->count
;
1187 unsigned int offset
= map_data
? offset_in_page(map_data
->offset
) : 0;
1189 bmd
= bio_alloc_map_data(iter
, gfp_mask
);
1191 return ERR_PTR(-ENOMEM
);
1194 * We need to do a deep copy of the iov_iter including the iovecs.
1195 * The caller provided iov might point to an on-stack or otherwise
1198 bmd
->is_our_pages
= map_data
? 0 : 1;
1200 nr_pages
= DIV_ROUND_UP(offset
+ len
, PAGE_SIZE
);
1201 if (nr_pages
> BIO_MAX_PAGES
)
1202 nr_pages
= BIO_MAX_PAGES
;
1205 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1212 nr_pages
= 1 << map_data
->page_order
;
1213 i
= map_data
->offset
/ PAGE_SIZE
;
1216 unsigned int bytes
= PAGE_SIZE
;
1224 if (i
== map_data
->nr_entries
* nr_pages
) {
1229 page
= map_data
->pages
[i
/ nr_pages
];
1230 page
+= (i
% nr_pages
);
1234 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
1241 if (bio_add_pc_page(q
, bio
, page
, bytes
, offset
) < bytes
)
1252 map_data
->offset
+= bio
->bi_iter
.bi_size
;
1257 if (((iter
->type
& WRITE
) && (!map_data
|| !map_data
->null_mapped
)) ||
1258 (map_data
&& map_data
->from_user
)) {
1259 ret
= bio_copy_from_iter(bio
, iter
);
1263 iov_iter_advance(iter
, bio
->bi_iter
.bi_size
);
1266 bio
->bi_private
= bmd
;
1267 if (map_data
&& map_data
->null_mapped
)
1268 bio_set_flag(bio
, BIO_NULL_MAPPED
);
1272 bio_free_pages(bio
);
1276 return ERR_PTR(ret
);
1280 * bio_map_user_iov - map user iovec into bio
1281 * @q: the struct request_queue for the bio
1282 * @iter: iovec iterator
1283 * @gfp_mask: memory allocation flags
1285 * Map the user space address into a bio suitable for io to a block
1286 * device. Returns an error pointer in case of error.
1288 struct bio
*bio_map_user_iov(struct request_queue
*q
,
1289 struct iov_iter
*iter
,
1295 struct bio_vec
*bvec
;
1297 if (!iov_iter_count(iter
))
1298 return ERR_PTR(-EINVAL
);
1300 bio
= bio_kmalloc(gfp_mask
, iov_iter_npages(iter
, BIO_MAX_PAGES
));
1302 return ERR_PTR(-ENOMEM
);
1304 while (iov_iter_count(iter
)) {
1305 struct page
**pages
;
1307 size_t offs
, added
= 0;
1310 bytes
= iov_iter_get_pages_alloc(iter
, &pages
, LONG_MAX
, &offs
);
1311 if (unlikely(bytes
<= 0)) {
1312 ret
= bytes
? bytes
: -EFAULT
;
1316 npages
= DIV_ROUND_UP(offs
+ bytes
, PAGE_SIZE
);
1318 if (unlikely(offs
& queue_dma_alignment(q
))) {
1322 for (j
= 0; j
< npages
; j
++) {
1323 struct page
*page
= pages
[j
];
1324 unsigned int n
= PAGE_SIZE
- offs
;
1325 unsigned short prev_bi_vcnt
= bio
->bi_vcnt
;
1330 if (!bio_add_pc_page(q
, bio
, page
, n
, offs
))
1334 * check if vector was merged with previous
1335 * drop page reference if needed
1337 if (bio
->bi_vcnt
== prev_bi_vcnt
)
1344 iov_iter_advance(iter
, added
);
1347 * release the pages we didn't map into the bio, if any
1350 put_page(pages
[j
++]);
1352 /* couldn't stuff something into bio? */
1357 bio_set_flag(bio
, BIO_USER_MAPPED
);
1360 * subtle -- if bio_map_user_iov() ended up bouncing a bio,
1361 * it would normally disappear when its bi_end_io is run.
1362 * however, we need it for the unmap, so grab an extra
1369 bio_for_each_segment_all(bvec
, bio
, j
) {
1370 put_page(bvec
->bv_page
);
1373 return ERR_PTR(ret
);
1376 static void __bio_unmap_user(struct bio
*bio
)
1378 struct bio_vec
*bvec
;
1382 * make sure we dirty pages we wrote to
1384 bio_for_each_segment_all(bvec
, bio
, i
) {
1385 if (bio_data_dir(bio
) == READ
)
1386 set_page_dirty_lock(bvec
->bv_page
);
1388 put_page(bvec
->bv_page
);
1395 * bio_unmap_user - unmap a bio
1396 * @bio: the bio being unmapped
1398 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from
1401 * bio_unmap_user() may sleep.
1403 void bio_unmap_user(struct bio
*bio
)
1405 __bio_unmap_user(bio
);
1409 static void bio_map_kern_endio(struct bio
*bio
)
1415 * bio_map_kern - map kernel address into bio
1416 * @q: the struct request_queue for the bio
1417 * @data: pointer to buffer to map
1418 * @len: length in bytes
1419 * @gfp_mask: allocation flags for bio allocation
1421 * Map the kernel address into a bio suitable for io to a block
1422 * device. Returns an error pointer in case of error.
1424 struct bio
*bio_map_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1427 unsigned long kaddr
= (unsigned long)data
;
1428 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1429 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1430 const int nr_pages
= end
- start
;
1434 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1436 return ERR_PTR(-ENOMEM
);
1438 offset
= offset_in_page(kaddr
);
1439 for (i
= 0; i
< nr_pages
; i
++) {
1440 unsigned int bytes
= PAGE_SIZE
- offset
;
1448 if (bio_add_pc_page(q
, bio
, virt_to_page(data
), bytes
,
1450 /* we don't support partial mappings */
1452 return ERR_PTR(-EINVAL
);
1460 bio
->bi_end_io
= bio_map_kern_endio
;
1463 EXPORT_SYMBOL(bio_map_kern
);
1465 static void bio_copy_kern_endio(struct bio
*bio
)
1467 bio_free_pages(bio
);
1471 static void bio_copy_kern_endio_read(struct bio
*bio
)
1473 char *p
= bio
->bi_private
;
1474 struct bio_vec
*bvec
;
1477 bio_for_each_segment_all(bvec
, bio
, i
) {
1478 memcpy(p
, page_address(bvec
->bv_page
), bvec
->bv_len
);
1482 bio_copy_kern_endio(bio
);
1486 * bio_copy_kern - copy kernel address into bio
1487 * @q: the struct request_queue for the bio
1488 * @data: pointer to buffer to copy
1489 * @len: length in bytes
1490 * @gfp_mask: allocation flags for bio and page allocation
1491 * @reading: data direction is READ
1493 * copy the kernel address into a bio suitable for io to a block
1494 * device. Returns an error pointer in case of error.
1496 struct bio
*bio_copy_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1497 gfp_t gfp_mask
, int reading
)
1499 unsigned long kaddr
= (unsigned long)data
;
1500 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1501 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1510 return ERR_PTR(-EINVAL
);
1512 nr_pages
= end
- start
;
1513 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1515 return ERR_PTR(-ENOMEM
);
1519 unsigned int bytes
= PAGE_SIZE
;
1524 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
1529 memcpy(page_address(page
), p
, bytes
);
1531 if (bio_add_pc_page(q
, bio
, page
, bytes
, 0) < bytes
)
1539 bio
->bi_end_io
= bio_copy_kern_endio_read
;
1540 bio
->bi_private
= data
;
1542 bio
->bi_end_io
= bio_copy_kern_endio
;
1548 bio_free_pages(bio
);
1550 return ERR_PTR(-ENOMEM
);
1554 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1555 * for performing direct-IO in BIOs.
1557 * The problem is that we cannot run set_page_dirty() from interrupt context
1558 * because the required locks are not interrupt-safe. So what we can do is to
1559 * mark the pages dirty _before_ performing IO. And in interrupt context,
1560 * check that the pages are still dirty. If so, fine. If not, redirty them
1561 * in process context.
1563 * We special-case compound pages here: normally this means reads into hugetlb
1564 * pages. The logic in here doesn't really work right for compound pages
1565 * because the VM does not uniformly chase down the head page in all cases.
1566 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1567 * handle them at all. So we skip compound pages here at an early stage.
1569 * Note that this code is very hard to test under normal circumstances because
1570 * direct-io pins the pages with get_user_pages(). This makes
1571 * is_page_cache_freeable return false, and the VM will not clean the pages.
1572 * But other code (eg, flusher threads) could clean the pages if they are mapped
1575 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1576 * deferred bio dirtying paths.
1580 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1582 void bio_set_pages_dirty(struct bio
*bio
)
1584 struct bio_vec
*bvec
;
1587 bio_for_each_segment_all(bvec
, bio
, i
) {
1588 if (!PageCompound(bvec
->bv_page
))
1589 set_page_dirty_lock(bvec
->bv_page
);
1592 EXPORT_SYMBOL_GPL(bio_set_pages_dirty
);
1594 static void bio_release_pages(struct bio
*bio
)
1596 struct bio_vec
*bvec
;
1599 bio_for_each_segment_all(bvec
, bio
, i
)
1600 put_page(bvec
->bv_page
);
1604 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1605 * If they are, then fine. If, however, some pages are clean then they must
1606 * have been written out during the direct-IO read. So we take another ref on
1607 * the BIO and re-dirty the pages in process context.
1609 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1610 * here on. It will run one put_page() against each page and will run one
1611 * bio_put() against the BIO.
1614 static void bio_dirty_fn(struct work_struct
*work
);
1616 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
1617 static DEFINE_SPINLOCK(bio_dirty_lock
);
1618 static struct bio
*bio_dirty_list
;
1621 * This runs in process context
1623 static void bio_dirty_fn(struct work_struct
*work
)
1625 struct bio
*bio
, *next
;
1627 spin_lock_irq(&bio_dirty_lock
);
1628 next
= bio_dirty_list
;
1629 bio_dirty_list
= NULL
;
1630 spin_unlock_irq(&bio_dirty_lock
);
1632 while ((bio
= next
) != NULL
) {
1633 next
= bio
->bi_private
;
1635 bio_set_pages_dirty(bio
);
1636 bio_release_pages(bio
);
1641 void bio_check_pages_dirty(struct bio
*bio
)
1643 struct bio_vec
*bvec
;
1644 unsigned long flags
;
1647 bio_for_each_segment_all(bvec
, bio
, i
) {
1648 if (!PageDirty(bvec
->bv_page
) && !PageCompound(bvec
->bv_page
))
1652 bio_release_pages(bio
);
1656 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1657 bio
->bi_private
= bio_dirty_list
;
1658 bio_dirty_list
= bio
;
1659 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1660 schedule_work(&bio_dirty_work
);
1662 EXPORT_SYMBOL_GPL(bio_check_pages_dirty
);
1664 void generic_start_io_acct(struct request_queue
*q
, int op
,
1665 unsigned long sectors
, struct hd_struct
*part
)
1667 const int sgrp
= op_stat_group(op
);
1668 int cpu
= part_stat_lock();
1670 part_round_stats(q
, cpu
, part
);
1671 part_stat_inc(cpu
, part
, ios
[sgrp
]);
1672 part_stat_add(cpu
, part
, sectors
[sgrp
], sectors
);
1673 part_inc_in_flight(q
, part
, op_is_write(op
));
1677 EXPORT_SYMBOL(generic_start_io_acct
);
1679 void generic_end_io_acct(struct request_queue
*q
, int req_op
,
1680 struct hd_struct
*part
, unsigned long start_time
)
1682 unsigned long duration
= jiffies
- start_time
;
1683 const int sgrp
= op_stat_group(req_op
);
1684 int cpu
= part_stat_lock();
1686 part_stat_add(cpu
, part
, ticks
[sgrp
], duration
);
1687 part_round_stats(q
, cpu
, part
);
1688 part_dec_in_flight(q
, part
, op_is_write(req_op
));
1692 EXPORT_SYMBOL(generic_end_io_acct
);
1694 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1695 void bio_flush_dcache_pages(struct bio
*bi
)
1697 struct bio_vec bvec
;
1698 struct bvec_iter iter
;
1700 bio_for_each_segment(bvec
, bi
, iter
)
1701 flush_dcache_page(bvec
.bv_page
);
1703 EXPORT_SYMBOL(bio_flush_dcache_pages
);
1706 static inline bool bio_remaining_done(struct bio
*bio
)
1709 * If we're not chaining, then ->__bi_remaining is always 1 and
1710 * we always end io on the first invocation.
1712 if (!bio_flagged(bio
, BIO_CHAIN
))
1715 BUG_ON(atomic_read(&bio
->__bi_remaining
) <= 0);
1717 if (atomic_dec_and_test(&bio
->__bi_remaining
)) {
1718 bio_clear_flag(bio
, BIO_CHAIN
);
1726 * bio_endio - end I/O on a bio
1730 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1731 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1732 * bio unless they own it and thus know that it has an end_io function.
1734 * bio_endio() can be called several times on a bio that has been chained
1735 * using bio_chain(). The ->bi_end_io() function will only be called the
1736 * last time. At this point the BLK_TA_COMPLETE tracing event will be
1737 * generated if BIO_TRACE_COMPLETION is set.
1739 void bio_endio(struct bio
*bio
)
1742 if (!bio_remaining_done(bio
))
1744 if (!bio_integrity_endio(bio
))
1748 rq_qos_done_bio(bio
->bi_disk
->queue
, bio
);
1751 * Need to have a real endio function for chained bios, otherwise
1752 * various corner cases will break (like stacking block devices that
1753 * save/restore bi_end_io) - however, we want to avoid unbounded
1754 * recursion and blowing the stack. Tail call optimization would
1755 * handle this, but compiling with frame pointers also disables
1756 * gcc's sibling call optimization.
1758 if (bio
->bi_end_io
== bio_chain_endio
) {
1759 bio
= __bio_chain_endio(bio
);
1763 if (bio
->bi_disk
&& bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
1764 trace_block_bio_complete(bio
->bi_disk
->queue
, bio
,
1765 blk_status_to_errno(bio
->bi_status
));
1766 bio_clear_flag(bio
, BIO_TRACE_COMPLETION
);
1769 blk_throtl_bio_endio(bio
);
1770 /* release cgroup info */
1773 bio
->bi_end_io(bio
);
1775 EXPORT_SYMBOL(bio_endio
);
1778 * bio_split - split a bio
1779 * @bio: bio to split
1780 * @sectors: number of sectors to split from the front of @bio
1782 * @bs: bio set to allocate from
1784 * Allocates and returns a new bio which represents @sectors from the start of
1785 * @bio, and updates @bio to represent the remaining sectors.
1787 * Unless this is a discard request the newly allocated bio will point
1788 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1789 * @bio is not freed before the split.
1791 struct bio
*bio_split(struct bio
*bio
, int sectors
,
1792 gfp_t gfp
, struct bio_set
*bs
)
1796 BUG_ON(sectors
<= 0);
1797 BUG_ON(sectors
>= bio_sectors(bio
));
1799 split
= bio_clone_fast(bio
, gfp
, bs
);
1803 split
->bi_iter
.bi_size
= sectors
<< 9;
1805 if (bio_integrity(split
))
1806 bio_integrity_trim(split
);
1808 bio_advance(bio
, split
->bi_iter
.bi_size
);
1810 if (bio_flagged(bio
, BIO_TRACE_COMPLETION
))
1811 bio_set_flag(split
, BIO_TRACE_COMPLETION
);
1815 EXPORT_SYMBOL(bio_split
);
1818 * bio_trim - trim a bio
1820 * @offset: number of sectors to trim from the front of @bio
1821 * @size: size we want to trim @bio to, in sectors
1823 void bio_trim(struct bio
*bio
, int offset
, int size
)
1825 /* 'bio' is a cloned bio which we need to trim to match
1826 * the given offset and size.
1830 if (offset
== 0 && size
== bio
->bi_iter
.bi_size
)
1833 bio_clear_flag(bio
, BIO_SEG_VALID
);
1835 bio_advance(bio
, offset
<< 9);
1837 bio
->bi_iter
.bi_size
= size
;
1839 if (bio_integrity(bio
))
1840 bio_integrity_trim(bio
);
1843 EXPORT_SYMBOL_GPL(bio_trim
);
1846 * create memory pools for biovec's in a bio_set.
1847 * use the global biovec slabs created for general use.
1849 int biovec_init_pool(mempool_t
*pool
, int pool_entries
)
1851 struct biovec_slab
*bp
= bvec_slabs
+ BVEC_POOL_MAX
;
1853 return mempool_init_slab_pool(pool
, pool_entries
, bp
->slab
);
1857 * bioset_exit - exit a bioset initialized with bioset_init()
1859 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1862 void bioset_exit(struct bio_set
*bs
)
1864 if (bs
->rescue_workqueue
)
1865 destroy_workqueue(bs
->rescue_workqueue
);
1866 bs
->rescue_workqueue
= NULL
;
1868 mempool_exit(&bs
->bio_pool
);
1869 mempool_exit(&bs
->bvec_pool
);
1871 bioset_integrity_free(bs
);
1874 bs
->bio_slab
= NULL
;
1876 EXPORT_SYMBOL(bioset_exit
);
1879 * bioset_init - Initialize a bio_set
1880 * @bs: pool to initialize
1881 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1882 * @front_pad: Number of bytes to allocate in front of the returned bio
1883 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1884 * and %BIOSET_NEED_RESCUER
1887 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1888 * to ask for a number of bytes to be allocated in front of the bio.
1889 * Front pad allocation is useful for embedding the bio inside
1890 * another structure, to avoid allocating extra data to go with the bio.
1891 * Note that the bio must be embedded at the END of that structure always,
1892 * or things will break badly.
1893 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1894 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast().
1895 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
1896 * dispatch queued requests when the mempool runs out of space.
1899 int bioset_init(struct bio_set
*bs
,
1900 unsigned int pool_size
,
1901 unsigned int front_pad
,
1904 unsigned int back_pad
= BIO_INLINE_VECS
* sizeof(struct bio_vec
);
1906 bs
->front_pad
= front_pad
;
1908 spin_lock_init(&bs
->rescue_lock
);
1909 bio_list_init(&bs
->rescue_list
);
1910 INIT_WORK(&bs
->rescue_work
, bio_alloc_rescue
);
1912 bs
->bio_slab
= bio_find_or_create_slab(front_pad
+ back_pad
);
1916 if (mempool_init_slab_pool(&bs
->bio_pool
, pool_size
, bs
->bio_slab
))
1919 if ((flags
& BIOSET_NEED_BVECS
) &&
1920 biovec_init_pool(&bs
->bvec_pool
, pool_size
))
1923 if (!(flags
& BIOSET_NEED_RESCUER
))
1926 bs
->rescue_workqueue
= alloc_workqueue("bioset", WQ_MEM_RECLAIM
, 0);
1927 if (!bs
->rescue_workqueue
)
1935 EXPORT_SYMBOL(bioset_init
);
1938 * Initialize and setup a new bio_set, based on the settings from
1941 int bioset_init_from_src(struct bio_set
*bs
, struct bio_set
*src
)
1946 if (src
->bvec_pool
.min_nr
)
1947 flags
|= BIOSET_NEED_BVECS
;
1948 if (src
->rescue_workqueue
)
1949 flags
|= BIOSET_NEED_RESCUER
;
1951 return bioset_init(bs
, src
->bio_pool
.min_nr
, src
->front_pad
, flags
);
1953 EXPORT_SYMBOL(bioset_init_from_src
);
1955 #ifdef CONFIG_BLK_CGROUP
1959 * bio_associate_blkcg_from_page - associate a bio with the page's blkcg
1961 * @page: the page to lookup the blkcg from
1963 * Associate @bio with the blkcg from @page's owning memcg. This works like
1964 * every other associate function wrt references.
1966 int bio_associate_blkcg_from_page(struct bio
*bio
, struct page
*page
)
1968 struct cgroup_subsys_state
*blkcg_css
;
1970 if (unlikely(bio
->bi_css
))
1972 if (!page
->mem_cgroup
)
1974 blkcg_css
= cgroup_get_e_css(page
->mem_cgroup
->css
.cgroup
,
1976 bio
->bi_css
= blkcg_css
;
1979 #endif /* CONFIG_MEMCG */
1982 * bio_associate_blkcg - associate a bio with the specified blkcg
1984 * @blkcg_css: css of the blkcg to associate
1986 * Associate @bio with the blkcg specified by @blkcg_css. Block layer will
1987 * treat @bio as if it were issued by a task which belongs to the blkcg.
1989 * This function takes an extra reference of @blkcg_css which will be put
1990 * when @bio is released. The caller must own @bio and is responsible for
1991 * synchronizing calls to this function.
1993 int bio_associate_blkcg(struct bio
*bio
, struct cgroup_subsys_state
*blkcg_css
)
1995 if (unlikely(bio
->bi_css
))
1998 bio
->bi_css
= blkcg_css
;
2001 EXPORT_SYMBOL_GPL(bio_associate_blkcg
);
2004 * bio_associate_blkg - associate a bio with the specified blkg
2006 * @blkg: the blkg to associate
2008 * Associate @bio with the blkg specified by @blkg. This is the queue specific
2009 * blkcg information associated with the @bio, a reference will be taken on the
2010 * @blkg and will be freed when the bio is freed.
2012 int bio_associate_blkg(struct bio
*bio
, struct blkcg_gq
*blkg
)
2014 if (unlikely(bio
->bi_blkg
))
2016 if (!blkg_try_get(blkg
))
2018 bio
->bi_blkg
= blkg
;
2023 * bio_disassociate_task - undo bio_associate_current()
2026 void bio_disassociate_task(struct bio
*bio
)
2029 put_io_context(bio
->bi_ioc
);
2033 css_put(bio
->bi_css
);
2037 blkg_put(bio
->bi_blkg
);
2038 bio
->bi_blkg
= NULL
;
2043 * bio_clone_blkcg_association - clone blkcg association from src to dst bio
2044 * @dst: destination bio
2047 void bio_clone_blkcg_association(struct bio
*dst
, struct bio
*src
)
2050 WARN_ON(bio_associate_blkcg(dst
, src
->bi_css
));
2052 EXPORT_SYMBOL_GPL(bio_clone_blkcg_association
);
2053 #endif /* CONFIG_BLK_CGROUP */
2055 static void __init
biovec_init_slabs(void)
2059 for (i
= 0; i
< BVEC_POOL_NR
; i
++) {
2061 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
2063 if (bvs
->nr_vecs
<= BIO_INLINE_VECS
) {
2068 size
= bvs
->nr_vecs
* sizeof(struct bio_vec
);
2069 bvs
->slab
= kmem_cache_create(bvs
->name
, size
, 0,
2070 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
2074 static int __init
init_bio(void)
2078 bio_slabs
= kcalloc(bio_slab_max
, sizeof(struct bio_slab
),
2081 panic("bio: can't allocate bios\n");
2083 bio_integrity_init();
2084 biovec_init_slabs();
2086 if (bioset_init(&fs_bio_set
, BIO_POOL_SIZE
, 0, BIOSET_NEED_BVECS
))
2087 panic("bio: can't allocate bios\n");
2089 if (bioset_integrity_create(&fs_bio_set
, BIO_POOL_SIZE
))
2090 panic("bio: can't create integrity pool\n");
2094 subsys_initcall(init_bio
);