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_blkg(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 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
;
583 * __bio_clone_fast - clone a bio that shares the original bio's biovec
584 * @bio: destination bio
585 * @bio_src: bio to clone
587 * Clone a &bio. Caller will own the returned bio, but not
588 * the actual data it points to. Reference count of returned
591 * Caller must ensure that @bio_src is not freed before @bio.
593 void __bio_clone_fast(struct bio
*bio
, struct bio
*bio_src
)
595 BUG_ON(bio
->bi_pool
&& BVEC_POOL_IDX(bio
));
598 * most users will be overriding ->bi_disk with a new target,
599 * so we don't set nor calculate new physical/hw segment counts here
601 bio
->bi_disk
= bio_src
->bi_disk
;
602 bio
->bi_partno
= bio_src
->bi_partno
;
603 bio_set_flag(bio
, BIO_CLONED
);
604 if (bio_flagged(bio_src
, BIO_THROTTLED
))
605 bio_set_flag(bio
, BIO_THROTTLED
);
606 bio
->bi_opf
= bio_src
->bi_opf
;
607 bio
->bi_ioprio
= bio_src
->bi_ioprio
;
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
);
613 blkcg_bio_issue_init(bio
);
615 EXPORT_SYMBOL(__bio_clone_fast
);
618 * bio_clone_fast - clone a bio that shares the original bio's biovec
620 * @gfp_mask: allocation priority
621 * @bs: bio_set to allocate from
623 * Like __bio_clone_fast, only also allocates the returned bio
625 struct bio
*bio_clone_fast(struct bio
*bio
, gfp_t gfp_mask
, struct bio_set
*bs
)
629 b
= bio_alloc_bioset(gfp_mask
, 0, bs
);
633 __bio_clone_fast(b
, bio
);
635 if (bio_integrity(bio
)) {
638 ret
= bio_integrity_clone(b
, bio
, gfp_mask
);
648 EXPORT_SYMBOL(bio_clone_fast
);
651 * bio_add_pc_page - attempt to add page to bio
652 * @q: the target queue
653 * @bio: destination bio
655 * @len: vec entry length
656 * @offset: vec entry offset
658 * Attempt to add a page to the bio_vec maplist. This can fail for a
659 * number of reasons, such as the bio being full or target block device
660 * limitations. The target block device must allow bio's up to PAGE_SIZE,
661 * so it is always possible to add a single page to an empty bio.
663 * This should only be used by REQ_PC bios.
665 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
, struct page
666 *page
, unsigned int len
, unsigned int offset
)
668 int retried_segments
= 0;
669 struct bio_vec
*bvec
;
672 * cloned bio must not modify vec list
674 if (unlikely(bio_flagged(bio
, BIO_CLONED
)))
677 if (((bio
->bi_iter
.bi_size
+ len
) >> 9) > queue_max_hw_sectors(q
))
681 * For filesystems with a blocksize smaller than the pagesize
682 * we will often be called with the same page as last time and
683 * a consecutive offset. Optimize this special case.
685 if (bio
->bi_vcnt
> 0) {
686 struct bio_vec
*prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
688 if (page
== prev
->bv_page
&&
689 offset
== prev
->bv_offset
+ prev
->bv_len
) {
691 bio
->bi_iter
.bi_size
+= len
;
696 * If the queue doesn't support SG gaps and adding this
697 * offset would create a gap, disallow it.
699 if (bvec_gap_to_prev(q
, prev
, offset
))
707 * setup the new entry, we might clear it again later if we
708 * cannot add the page
710 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
711 bvec
->bv_page
= page
;
713 bvec
->bv_offset
= offset
;
715 bio
->bi_phys_segments
++;
716 bio
->bi_iter
.bi_size
+= len
;
719 * Perform a recount if the number of segments is greater
720 * than queue_max_segments(q).
723 while (bio
->bi_phys_segments
> queue_max_segments(q
)) {
725 if (retried_segments
)
728 retried_segments
= 1;
729 blk_recount_segments(q
, bio
);
732 /* If we may be able to merge these biovecs, force a recount */
733 if (bio
->bi_vcnt
> 1 && biovec_phys_mergeable(q
, bvec
- 1, bvec
))
734 bio_clear_flag(bio
, BIO_SEG_VALID
);
740 bvec
->bv_page
= NULL
;
744 bio
->bi_iter
.bi_size
-= len
;
745 blk_recount_segments(q
, bio
);
748 EXPORT_SYMBOL(bio_add_pc_page
);
751 * __bio_try_merge_page - try appending data to an existing bvec.
752 * @bio: destination bio
754 * @len: length of the data to add
755 * @off: offset of the data in @page
756 * @same_page: if %true only merge if the new data is in the same physical
757 * page as the last segment of the bio.
759 * Try to add the data at @page + @off to the last bvec of @bio. This is a
760 * a useful optimisation for file systems with a block size smaller than the
763 * Return %true on success or %false on failure.
765 bool __bio_try_merge_page(struct bio
*bio
, struct page
*page
,
766 unsigned int len
, unsigned int off
, bool same_page
)
768 if (WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
)))
771 if (bio
->bi_vcnt
> 0) {
772 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
773 phys_addr_t vec_end_addr
= page_to_phys(bv
->bv_page
) +
774 bv
->bv_offset
+ bv
->bv_len
- 1;
775 phys_addr_t page_addr
= page_to_phys(page
);
777 if (vec_end_addr
+ 1 != page_addr
+ off
)
779 if (same_page
&& (vec_end_addr
& PAGE_MASK
) != page_addr
)
783 bio
->bi_iter
.bi_size
+= len
;
788 EXPORT_SYMBOL_GPL(__bio_try_merge_page
);
791 * __bio_add_page - add page to a bio in a new segment
792 * @bio: destination bio
794 * @len: length of the data to add
795 * @off: offset of the data in @page
797 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
798 * that @bio has space for another bvec.
800 void __bio_add_page(struct bio
*bio
, struct page
*page
,
801 unsigned int len
, unsigned int off
)
803 struct bio_vec
*bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
805 WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
));
806 WARN_ON_ONCE(bio_full(bio
));
812 bio
->bi_iter
.bi_size
+= len
;
815 EXPORT_SYMBOL_GPL(__bio_add_page
);
818 * bio_add_page - attempt to add page to bio
819 * @bio: destination bio
821 * @len: vec entry length
822 * @offset: vec entry offset
824 * Attempt to add a page to the bio_vec maplist. This will only fail
825 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
827 int bio_add_page(struct bio
*bio
, struct page
*page
,
828 unsigned int len
, unsigned int offset
)
830 if (!__bio_try_merge_page(bio
, page
, len
, offset
, false)) {
833 __bio_add_page(bio
, page
, len
, offset
);
837 EXPORT_SYMBOL(bio_add_page
);
839 static int __bio_iov_bvec_add_pages(struct bio
*bio
, struct iov_iter
*iter
)
841 const struct bio_vec
*bv
= iter
->bvec
;
845 if (WARN_ON_ONCE(iter
->iov_offset
> bv
->bv_len
))
848 len
= min_t(size_t, bv
->bv_len
- iter
->iov_offset
, iter
->count
);
849 size
= bio_add_page(bio
, bv
->bv_page
, len
,
850 bv
->bv_offset
+ iter
->iov_offset
);
852 if (!bio_flagged(bio
, BIO_NO_PAGE_REF
)) {
856 mp_bvec_for_each_page(page
, bv
, i
)
860 iov_iter_advance(iter
, size
);
867 #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
870 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
871 * @bio: bio to add pages to
872 * @iter: iov iterator describing the region to be mapped
874 * Pins pages from *iter and appends them to @bio's bvec array. The
875 * pages will have to be released using put_page() when done.
876 * For multi-segment *iter, this function only adds pages from the
877 * the next non-empty segment of the iov iterator.
879 static int __bio_iov_iter_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
881 unsigned short nr_pages
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
882 unsigned short entries_left
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
883 struct bio_vec
*bv
= bio
->bi_io_vec
+ bio
->bi_vcnt
;
884 struct page
**pages
= (struct page
**)bv
;
890 * Move page array up in the allocated memory for the bio vecs as far as
891 * possible so that we can start filling biovecs from the beginning
892 * without overwriting the temporary page array.
894 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC
< 2);
895 pages
+= entries_left
* (PAGE_PTRS_PER_BVEC
- 1);
897 size
= iov_iter_get_pages(iter
, pages
, LONG_MAX
, nr_pages
, &offset
);
898 if (unlikely(size
<= 0))
899 return size
? size
: -EFAULT
;
901 for (left
= size
, i
= 0; left
> 0; left
-= len
, i
++) {
902 struct page
*page
= pages
[i
];
904 len
= min_t(size_t, PAGE_SIZE
- offset
, left
);
905 if (WARN_ON_ONCE(bio_add_page(bio
, page
, len
, offset
) != len
))
910 iov_iter_advance(iter
, size
);
915 * bio_iov_iter_get_pages - add user or kernel pages to a bio
916 * @bio: bio to add pages to
917 * @iter: iov iterator describing the region to be added
919 * This takes either an iterator pointing to user memory, or one pointing to
920 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
921 * map them into the kernel. On IO completion, the caller should put those
922 * pages. If we're adding kernel pages, and the caller told us it's safe to
923 * do so, we just have to add the pages to the bio directly. We don't grab an
924 * extra reference to those pages (the user should already have that), and we
925 * don't put the page on IO completion. The caller needs to check if the bio is
926 * flagged BIO_NO_PAGE_REF on IO completion. If it isn't, then pages should be
929 * The function tries, but does not guarantee, to pin as many pages as
930 * fit into the bio, or are requested in *iter, whatever is smaller. If
931 * MM encounters an error pinning the requested pages, it stops. Error
932 * is returned only if 0 pages could be pinned.
934 int bio_iov_iter_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
936 const bool is_bvec
= iov_iter_is_bvec(iter
);
937 unsigned short orig_vcnt
= bio
->bi_vcnt
;
940 * If this is a BVEC iter, then the pages are kernel pages. Don't
941 * release them on IO completion, if the caller asked us to.
943 if (is_bvec
&& iov_iter_bvec_no_ref(iter
))
944 bio_set_flag(bio
, BIO_NO_PAGE_REF
);
950 ret
= __bio_iov_bvec_add_pages(bio
, iter
);
952 ret
= __bio_iov_iter_get_pages(bio
, iter
);
955 return bio
->bi_vcnt
> orig_vcnt
? 0 : ret
;
957 } while (iov_iter_count(iter
) && !bio_full(bio
));
962 static void submit_bio_wait_endio(struct bio
*bio
)
964 complete(bio
->bi_private
);
968 * submit_bio_wait - submit a bio, and wait until it completes
969 * @bio: The &struct bio which describes the I/O
971 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
972 * bio_endio() on failure.
974 * WARNING: Unlike to how submit_bio() is usually used, this function does not
975 * result in bio reference to be consumed. The caller must drop the reference
978 int submit_bio_wait(struct bio
*bio
)
980 DECLARE_COMPLETION_ONSTACK_MAP(done
, bio
->bi_disk
->lockdep_map
);
982 bio
->bi_private
= &done
;
983 bio
->bi_end_io
= submit_bio_wait_endio
;
984 bio
->bi_opf
|= REQ_SYNC
;
986 wait_for_completion_io(&done
);
988 return blk_status_to_errno(bio
->bi_status
);
990 EXPORT_SYMBOL(submit_bio_wait
);
993 * bio_advance - increment/complete a bio by some number of bytes
994 * @bio: bio to advance
995 * @bytes: number of bytes to complete
997 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
998 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
999 * be updated on the last bvec as well.
1001 * @bio will then represent the remaining, uncompleted portion of the io.
1003 void bio_advance(struct bio
*bio
, unsigned bytes
)
1005 if (bio_integrity(bio
))
1006 bio_integrity_advance(bio
, bytes
);
1008 bio_advance_iter(bio
, &bio
->bi_iter
, bytes
);
1010 EXPORT_SYMBOL(bio_advance
);
1012 void bio_copy_data_iter(struct bio
*dst
, struct bvec_iter
*dst_iter
,
1013 struct bio
*src
, struct bvec_iter
*src_iter
)
1015 struct bio_vec src_bv
, dst_bv
;
1016 void *src_p
, *dst_p
;
1019 while (src_iter
->bi_size
&& dst_iter
->bi_size
) {
1020 src_bv
= bio_iter_iovec(src
, *src_iter
);
1021 dst_bv
= bio_iter_iovec(dst
, *dst_iter
);
1023 bytes
= min(src_bv
.bv_len
, dst_bv
.bv_len
);
1025 src_p
= kmap_atomic(src_bv
.bv_page
);
1026 dst_p
= kmap_atomic(dst_bv
.bv_page
);
1028 memcpy(dst_p
+ dst_bv
.bv_offset
,
1029 src_p
+ src_bv
.bv_offset
,
1032 kunmap_atomic(dst_p
);
1033 kunmap_atomic(src_p
);
1035 flush_dcache_page(dst_bv
.bv_page
);
1037 bio_advance_iter(src
, src_iter
, bytes
);
1038 bio_advance_iter(dst
, dst_iter
, bytes
);
1041 EXPORT_SYMBOL(bio_copy_data_iter
);
1044 * bio_copy_data - copy contents of data buffers from one bio to another
1046 * @dst: destination bio
1048 * Stops when it reaches the end of either @src or @dst - that is, copies
1049 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1051 void bio_copy_data(struct bio
*dst
, struct bio
*src
)
1053 struct bvec_iter src_iter
= src
->bi_iter
;
1054 struct bvec_iter dst_iter
= dst
->bi_iter
;
1056 bio_copy_data_iter(dst
, &dst_iter
, src
, &src_iter
);
1058 EXPORT_SYMBOL(bio_copy_data
);
1061 * bio_list_copy_data - copy contents of data buffers from one chain of bios to
1063 * @src: source bio list
1064 * @dst: destination bio list
1066 * Stops when it reaches the end of either the @src list or @dst list - that is,
1067 * copies min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of
1070 void bio_list_copy_data(struct bio
*dst
, struct bio
*src
)
1072 struct bvec_iter src_iter
= src
->bi_iter
;
1073 struct bvec_iter dst_iter
= dst
->bi_iter
;
1076 if (!src_iter
.bi_size
) {
1081 src_iter
= src
->bi_iter
;
1084 if (!dst_iter
.bi_size
) {
1089 dst_iter
= dst
->bi_iter
;
1092 bio_copy_data_iter(dst
, &dst_iter
, src
, &src_iter
);
1095 EXPORT_SYMBOL(bio_list_copy_data
);
1097 struct bio_map_data
{
1099 struct iov_iter iter
;
1103 static struct bio_map_data
*bio_alloc_map_data(struct iov_iter
*data
,
1106 struct bio_map_data
*bmd
;
1107 if (data
->nr_segs
> UIO_MAXIOV
)
1110 bmd
= kmalloc(sizeof(struct bio_map_data
) +
1111 sizeof(struct iovec
) * data
->nr_segs
, gfp_mask
);
1114 memcpy(bmd
->iov
, data
->iov
, sizeof(struct iovec
) * data
->nr_segs
);
1116 bmd
->iter
.iov
= bmd
->iov
;
1121 * bio_copy_from_iter - copy all pages from iov_iter to bio
1122 * @bio: The &struct bio which describes the I/O as destination
1123 * @iter: iov_iter as source
1125 * Copy all pages from iov_iter to bio.
1126 * Returns 0 on success, or error on failure.
1128 static int bio_copy_from_iter(struct bio
*bio
, struct iov_iter
*iter
)
1131 struct bio_vec
*bvec
;
1132 struct bvec_iter_all iter_all
;
1134 bio_for_each_segment_all(bvec
, bio
, i
, iter_all
) {
1137 ret
= copy_page_from_iter(bvec
->bv_page
,
1142 if (!iov_iter_count(iter
))
1145 if (ret
< bvec
->bv_len
)
1153 * bio_copy_to_iter - copy all pages from bio to iov_iter
1154 * @bio: The &struct bio which describes the I/O as source
1155 * @iter: iov_iter as destination
1157 * Copy all pages from bio to iov_iter.
1158 * Returns 0 on success, or error on failure.
1160 static int bio_copy_to_iter(struct bio
*bio
, struct iov_iter iter
)
1163 struct bio_vec
*bvec
;
1164 struct bvec_iter_all iter_all
;
1166 bio_for_each_segment_all(bvec
, bio
, i
, iter_all
) {
1169 ret
= copy_page_to_iter(bvec
->bv_page
,
1174 if (!iov_iter_count(&iter
))
1177 if (ret
< bvec
->bv_len
)
1184 void bio_free_pages(struct bio
*bio
)
1186 struct bio_vec
*bvec
;
1188 struct bvec_iter_all iter_all
;
1190 bio_for_each_segment_all(bvec
, bio
, i
, iter_all
)
1191 __free_page(bvec
->bv_page
);
1193 EXPORT_SYMBOL(bio_free_pages
);
1196 * bio_uncopy_user - finish previously mapped bio
1197 * @bio: bio being terminated
1199 * Free pages allocated from bio_copy_user_iov() and write back data
1200 * to user space in case of a read.
1202 int bio_uncopy_user(struct bio
*bio
)
1204 struct bio_map_data
*bmd
= bio
->bi_private
;
1207 if (!bio_flagged(bio
, BIO_NULL_MAPPED
)) {
1209 * if we're in a workqueue, the request is orphaned, so
1210 * don't copy into a random user address space, just free
1211 * and return -EINTR so user space doesn't expect any data.
1215 else if (bio_data_dir(bio
) == READ
)
1216 ret
= bio_copy_to_iter(bio
, bmd
->iter
);
1217 if (bmd
->is_our_pages
)
1218 bio_free_pages(bio
);
1226 * bio_copy_user_iov - copy user data to bio
1227 * @q: destination block queue
1228 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1229 * @iter: iovec iterator
1230 * @gfp_mask: memory allocation flags
1232 * Prepares and returns a bio for indirect user io, bouncing data
1233 * to/from kernel pages as necessary. Must be paired with
1234 * call bio_uncopy_user() on io completion.
1236 struct bio
*bio_copy_user_iov(struct request_queue
*q
,
1237 struct rq_map_data
*map_data
,
1238 struct iov_iter
*iter
,
1241 struct bio_map_data
*bmd
;
1246 unsigned int len
= iter
->count
;
1247 unsigned int offset
= map_data
? offset_in_page(map_data
->offset
) : 0;
1249 bmd
= bio_alloc_map_data(iter
, gfp_mask
);
1251 return ERR_PTR(-ENOMEM
);
1254 * We need to do a deep copy of the iov_iter including the iovecs.
1255 * The caller provided iov might point to an on-stack or otherwise
1258 bmd
->is_our_pages
= map_data
? 0 : 1;
1260 nr_pages
= DIV_ROUND_UP(offset
+ len
, PAGE_SIZE
);
1261 if (nr_pages
> BIO_MAX_PAGES
)
1262 nr_pages
= BIO_MAX_PAGES
;
1265 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1272 nr_pages
= 1 << map_data
->page_order
;
1273 i
= map_data
->offset
/ PAGE_SIZE
;
1276 unsigned int bytes
= PAGE_SIZE
;
1284 if (i
== map_data
->nr_entries
* nr_pages
) {
1289 page
= map_data
->pages
[i
/ nr_pages
];
1290 page
+= (i
% nr_pages
);
1294 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
1301 if (bio_add_pc_page(q
, bio
, page
, bytes
, offset
) < bytes
) {
1315 map_data
->offset
+= bio
->bi_iter
.bi_size
;
1320 if ((iov_iter_rw(iter
) == WRITE
&& (!map_data
|| !map_data
->null_mapped
)) ||
1321 (map_data
&& map_data
->from_user
)) {
1322 ret
= bio_copy_from_iter(bio
, iter
);
1326 if (bmd
->is_our_pages
)
1328 iov_iter_advance(iter
, bio
->bi_iter
.bi_size
);
1331 bio
->bi_private
= bmd
;
1332 if (map_data
&& map_data
->null_mapped
)
1333 bio_set_flag(bio
, BIO_NULL_MAPPED
);
1337 bio_free_pages(bio
);
1341 return ERR_PTR(ret
);
1345 * bio_map_user_iov - map user iovec into bio
1346 * @q: the struct request_queue for the bio
1347 * @iter: iovec iterator
1348 * @gfp_mask: memory allocation flags
1350 * Map the user space address into a bio suitable for io to a block
1351 * device. Returns an error pointer in case of error.
1353 struct bio
*bio_map_user_iov(struct request_queue
*q
,
1354 struct iov_iter
*iter
,
1360 struct bio_vec
*bvec
;
1361 struct bvec_iter_all iter_all
;
1363 if (!iov_iter_count(iter
))
1364 return ERR_PTR(-EINVAL
);
1366 bio
= bio_kmalloc(gfp_mask
, iov_iter_npages(iter
, BIO_MAX_PAGES
));
1368 return ERR_PTR(-ENOMEM
);
1370 while (iov_iter_count(iter
)) {
1371 struct page
**pages
;
1373 size_t offs
, added
= 0;
1376 bytes
= iov_iter_get_pages_alloc(iter
, &pages
, LONG_MAX
, &offs
);
1377 if (unlikely(bytes
<= 0)) {
1378 ret
= bytes
? bytes
: -EFAULT
;
1382 npages
= DIV_ROUND_UP(offs
+ bytes
, PAGE_SIZE
);
1384 if (unlikely(offs
& queue_dma_alignment(q
))) {
1388 for (j
= 0; j
< npages
; j
++) {
1389 struct page
*page
= pages
[j
];
1390 unsigned int n
= PAGE_SIZE
- offs
;
1391 unsigned short prev_bi_vcnt
= bio
->bi_vcnt
;
1396 if (!bio_add_pc_page(q
, bio
, page
, n
, offs
))
1400 * check if vector was merged with previous
1401 * drop page reference if needed
1403 if (bio
->bi_vcnt
== prev_bi_vcnt
)
1410 iov_iter_advance(iter
, added
);
1413 * release the pages we didn't map into the bio, if any
1416 put_page(pages
[j
++]);
1418 /* couldn't stuff something into bio? */
1423 bio_set_flag(bio
, BIO_USER_MAPPED
);
1426 * subtle -- if bio_map_user_iov() ended up bouncing a bio,
1427 * it would normally disappear when its bi_end_io is run.
1428 * however, we need it for the unmap, so grab an extra
1435 bio_for_each_segment_all(bvec
, bio
, j
, iter_all
) {
1436 put_page(bvec
->bv_page
);
1439 return ERR_PTR(ret
);
1442 static void __bio_unmap_user(struct bio
*bio
)
1444 struct bio_vec
*bvec
;
1446 struct bvec_iter_all iter_all
;
1449 * make sure we dirty pages we wrote to
1451 bio_for_each_segment_all(bvec
, bio
, i
, iter_all
) {
1452 if (bio_data_dir(bio
) == READ
)
1453 set_page_dirty_lock(bvec
->bv_page
);
1455 put_page(bvec
->bv_page
);
1462 * bio_unmap_user - unmap a bio
1463 * @bio: the bio being unmapped
1465 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from
1468 * bio_unmap_user() may sleep.
1470 void bio_unmap_user(struct bio
*bio
)
1472 __bio_unmap_user(bio
);
1476 static void bio_map_kern_endio(struct bio
*bio
)
1482 * bio_map_kern - map kernel address into bio
1483 * @q: the struct request_queue for the bio
1484 * @data: pointer to buffer to map
1485 * @len: length in bytes
1486 * @gfp_mask: allocation flags for bio allocation
1488 * Map the kernel address into a bio suitable for io to a block
1489 * device. Returns an error pointer in case of error.
1491 struct bio
*bio_map_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1494 unsigned long kaddr
= (unsigned long)data
;
1495 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1496 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1497 const int nr_pages
= end
- start
;
1501 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1503 return ERR_PTR(-ENOMEM
);
1505 offset
= offset_in_page(kaddr
);
1506 for (i
= 0; i
< nr_pages
; i
++) {
1507 unsigned int bytes
= PAGE_SIZE
- offset
;
1515 if (bio_add_pc_page(q
, bio
, virt_to_page(data
), bytes
,
1517 /* we don't support partial mappings */
1519 return ERR_PTR(-EINVAL
);
1527 bio
->bi_end_io
= bio_map_kern_endio
;
1530 EXPORT_SYMBOL(bio_map_kern
);
1532 static void bio_copy_kern_endio(struct bio
*bio
)
1534 bio_free_pages(bio
);
1538 static void bio_copy_kern_endio_read(struct bio
*bio
)
1540 char *p
= bio
->bi_private
;
1541 struct bio_vec
*bvec
;
1543 struct bvec_iter_all iter_all
;
1545 bio_for_each_segment_all(bvec
, bio
, i
, iter_all
) {
1546 memcpy(p
, page_address(bvec
->bv_page
), bvec
->bv_len
);
1550 bio_copy_kern_endio(bio
);
1554 * bio_copy_kern - copy kernel address into bio
1555 * @q: the struct request_queue for the bio
1556 * @data: pointer to buffer to copy
1557 * @len: length in bytes
1558 * @gfp_mask: allocation flags for bio and page allocation
1559 * @reading: data direction is READ
1561 * copy the kernel address into a bio suitable for io to a block
1562 * device. Returns an error pointer in case of error.
1564 struct bio
*bio_copy_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1565 gfp_t gfp_mask
, int reading
)
1567 unsigned long kaddr
= (unsigned long)data
;
1568 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1569 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1578 return ERR_PTR(-EINVAL
);
1580 nr_pages
= end
- start
;
1581 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1583 return ERR_PTR(-ENOMEM
);
1587 unsigned int bytes
= PAGE_SIZE
;
1592 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
1597 memcpy(page_address(page
), p
, bytes
);
1599 if (bio_add_pc_page(q
, bio
, page
, bytes
, 0) < bytes
)
1607 bio
->bi_end_io
= bio_copy_kern_endio_read
;
1608 bio
->bi_private
= data
;
1610 bio
->bi_end_io
= bio_copy_kern_endio
;
1616 bio_free_pages(bio
);
1618 return ERR_PTR(-ENOMEM
);
1622 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1623 * for performing direct-IO in BIOs.
1625 * The problem is that we cannot run set_page_dirty() from interrupt context
1626 * because the required locks are not interrupt-safe. So what we can do is to
1627 * mark the pages dirty _before_ performing IO. And in interrupt context,
1628 * check that the pages are still dirty. If so, fine. If not, redirty them
1629 * in process context.
1631 * We special-case compound pages here: normally this means reads into hugetlb
1632 * pages. The logic in here doesn't really work right for compound pages
1633 * because the VM does not uniformly chase down the head page in all cases.
1634 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1635 * handle them at all. So we skip compound pages here at an early stage.
1637 * Note that this code is very hard to test under normal circumstances because
1638 * direct-io pins the pages with get_user_pages(). This makes
1639 * is_page_cache_freeable return false, and the VM will not clean the pages.
1640 * But other code (eg, flusher threads) could clean the pages if they are mapped
1643 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1644 * deferred bio dirtying paths.
1648 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1650 void bio_set_pages_dirty(struct bio
*bio
)
1652 struct bio_vec
*bvec
;
1654 struct bvec_iter_all iter_all
;
1656 bio_for_each_segment_all(bvec
, bio
, i
, iter_all
) {
1657 if (!PageCompound(bvec
->bv_page
))
1658 set_page_dirty_lock(bvec
->bv_page
);
1662 static void bio_release_pages(struct bio
*bio
)
1664 struct bio_vec
*bvec
;
1666 struct bvec_iter_all iter_all
;
1668 bio_for_each_segment_all(bvec
, bio
, i
, iter_all
)
1669 put_page(bvec
->bv_page
);
1673 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1674 * If they are, then fine. If, however, some pages are clean then they must
1675 * have been written out during the direct-IO read. So we take another ref on
1676 * the BIO and re-dirty the pages in process context.
1678 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1679 * here on. It will run one put_page() against each page and will run one
1680 * bio_put() against the BIO.
1683 static void bio_dirty_fn(struct work_struct
*work
);
1685 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
1686 static DEFINE_SPINLOCK(bio_dirty_lock
);
1687 static struct bio
*bio_dirty_list
;
1690 * This runs in process context
1692 static void bio_dirty_fn(struct work_struct
*work
)
1694 struct bio
*bio
, *next
;
1696 spin_lock_irq(&bio_dirty_lock
);
1697 next
= bio_dirty_list
;
1698 bio_dirty_list
= NULL
;
1699 spin_unlock_irq(&bio_dirty_lock
);
1701 while ((bio
= next
) != NULL
) {
1702 next
= bio
->bi_private
;
1704 bio_set_pages_dirty(bio
);
1705 if (!bio_flagged(bio
, BIO_NO_PAGE_REF
))
1706 bio_release_pages(bio
);
1711 void bio_check_pages_dirty(struct bio
*bio
)
1713 struct bio_vec
*bvec
;
1714 unsigned long flags
;
1716 struct bvec_iter_all iter_all
;
1718 bio_for_each_segment_all(bvec
, bio
, i
, iter_all
) {
1719 if (!PageDirty(bvec
->bv_page
) && !PageCompound(bvec
->bv_page
))
1723 if (!bio_flagged(bio
, BIO_NO_PAGE_REF
))
1724 bio_release_pages(bio
);
1728 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1729 bio
->bi_private
= bio_dirty_list
;
1730 bio_dirty_list
= bio
;
1731 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1732 schedule_work(&bio_dirty_work
);
1735 void update_io_ticks(struct hd_struct
*part
, unsigned long now
)
1737 unsigned long stamp
;
1739 stamp
= READ_ONCE(part
->stamp
);
1740 if (unlikely(stamp
!= now
)) {
1741 if (likely(cmpxchg(&part
->stamp
, stamp
, now
) == stamp
)) {
1742 __part_stat_add(part
, io_ticks
, 1);
1746 part
= &part_to_disk(part
)->part0
;
1751 void generic_start_io_acct(struct request_queue
*q
, int op
,
1752 unsigned long sectors
, struct hd_struct
*part
)
1754 const int sgrp
= op_stat_group(op
);
1758 update_io_ticks(part
, jiffies
);
1759 part_stat_inc(part
, ios
[sgrp
]);
1760 part_stat_add(part
, sectors
[sgrp
], sectors
);
1761 part_inc_in_flight(q
, part
, op_is_write(op
));
1765 EXPORT_SYMBOL(generic_start_io_acct
);
1767 void generic_end_io_acct(struct request_queue
*q
, int req_op
,
1768 struct hd_struct
*part
, unsigned long start_time
)
1770 unsigned long now
= jiffies
;
1771 unsigned long duration
= now
- start_time
;
1772 const int sgrp
= op_stat_group(req_op
);
1776 update_io_ticks(part
, now
);
1777 part_stat_add(part
, nsecs
[sgrp
], jiffies_to_nsecs(duration
));
1778 part_stat_add(part
, time_in_queue
, duration
);
1779 part_dec_in_flight(q
, part
, op_is_write(req_op
));
1783 EXPORT_SYMBOL(generic_end_io_acct
);
1785 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1786 void bio_flush_dcache_pages(struct bio
*bi
)
1788 struct bio_vec bvec
;
1789 struct bvec_iter iter
;
1791 bio_for_each_segment(bvec
, bi
, iter
)
1792 flush_dcache_page(bvec
.bv_page
);
1794 EXPORT_SYMBOL(bio_flush_dcache_pages
);
1797 static inline bool bio_remaining_done(struct bio
*bio
)
1800 * If we're not chaining, then ->__bi_remaining is always 1 and
1801 * we always end io on the first invocation.
1803 if (!bio_flagged(bio
, BIO_CHAIN
))
1806 BUG_ON(atomic_read(&bio
->__bi_remaining
) <= 0);
1808 if (atomic_dec_and_test(&bio
->__bi_remaining
)) {
1809 bio_clear_flag(bio
, BIO_CHAIN
);
1817 * bio_endio - end I/O on a bio
1821 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1822 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1823 * bio unless they own it and thus know that it has an end_io function.
1825 * bio_endio() can be called several times on a bio that has been chained
1826 * using bio_chain(). The ->bi_end_io() function will only be called the
1827 * last time. At this point the BLK_TA_COMPLETE tracing event will be
1828 * generated if BIO_TRACE_COMPLETION is set.
1830 void bio_endio(struct bio
*bio
)
1833 if (!bio_remaining_done(bio
))
1835 if (!bio_integrity_endio(bio
))
1839 rq_qos_done_bio(bio
->bi_disk
->queue
, bio
);
1842 * Need to have a real endio function for chained bios, otherwise
1843 * various corner cases will break (like stacking block devices that
1844 * save/restore bi_end_io) - however, we want to avoid unbounded
1845 * recursion and blowing the stack. Tail call optimization would
1846 * handle this, but compiling with frame pointers also disables
1847 * gcc's sibling call optimization.
1849 if (bio
->bi_end_io
== bio_chain_endio
) {
1850 bio
= __bio_chain_endio(bio
);
1854 if (bio
->bi_disk
&& bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
1855 trace_block_bio_complete(bio
->bi_disk
->queue
, bio
,
1856 blk_status_to_errno(bio
->bi_status
));
1857 bio_clear_flag(bio
, BIO_TRACE_COMPLETION
);
1860 blk_throtl_bio_endio(bio
);
1861 /* release cgroup info */
1864 bio
->bi_end_io(bio
);
1866 EXPORT_SYMBOL(bio_endio
);
1869 * bio_split - split a bio
1870 * @bio: bio to split
1871 * @sectors: number of sectors to split from the front of @bio
1873 * @bs: bio set to allocate from
1875 * Allocates and returns a new bio which represents @sectors from the start of
1876 * @bio, and updates @bio to represent the remaining sectors.
1878 * Unless this is a discard request the newly allocated bio will point
1879 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1880 * @bio is not freed before the split.
1882 struct bio
*bio_split(struct bio
*bio
, int sectors
,
1883 gfp_t gfp
, struct bio_set
*bs
)
1887 BUG_ON(sectors
<= 0);
1888 BUG_ON(sectors
>= bio_sectors(bio
));
1890 split
= bio_clone_fast(bio
, gfp
, bs
);
1894 split
->bi_iter
.bi_size
= sectors
<< 9;
1896 if (bio_integrity(split
))
1897 bio_integrity_trim(split
);
1899 bio_advance(bio
, split
->bi_iter
.bi_size
);
1901 if (bio_flagged(bio
, BIO_TRACE_COMPLETION
))
1902 bio_set_flag(split
, BIO_TRACE_COMPLETION
);
1906 EXPORT_SYMBOL(bio_split
);
1909 * bio_trim - trim a bio
1911 * @offset: number of sectors to trim from the front of @bio
1912 * @size: size we want to trim @bio to, in sectors
1914 void bio_trim(struct bio
*bio
, int offset
, int size
)
1916 /* 'bio' is a cloned bio which we need to trim to match
1917 * the given offset and size.
1921 if (offset
== 0 && size
== bio
->bi_iter
.bi_size
)
1924 bio_clear_flag(bio
, BIO_SEG_VALID
);
1926 bio_advance(bio
, offset
<< 9);
1928 bio
->bi_iter
.bi_size
= size
;
1930 if (bio_integrity(bio
))
1931 bio_integrity_trim(bio
);
1934 EXPORT_SYMBOL_GPL(bio_trim
);
1937 * create memory pools for biovec's in a bio_set.
1938 * use the global biovec slabs created for general use.
1940 int biovec_init_pool(mempool_t
*pool
, int pool_entries
)
1942 struct biovec_slab
*bp
= bvec_slabs
+ BVEC_POOL_MAX
;
1944 return mempool_init_slab_pool(pool
, pool_entries
, bp
->slab
);
1948 * bioset_exit - exit a bioset initialized with bioset_init()
1950 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1953 void bioset_exit(struct bio_set
*bs
)
1955 if (bs
->rescue_workqueue
)
1956 destroy_workqueue(bs
->rescue_workqueue
);
1957 bs
->rescue_workqueue
= NULL
;
1959 mempool_exit(&bs
->bio_pool
);
1960 mempool_exit(&bs
->bvec_pool
);
1962 bioset_integrity_free(bs
);
1965 bs
->bio_slab
= NULL
;
1967 EXPORT_SYMBOL(bioset_exit
);
1970 * bioset_init - Initialize a bio_set
1971 * @bs: pool to initialize
1972 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1973 * @front_pad: Number of bytes to allocate in front of the returned bio
1974 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1975 * and %BIOSET_NEED_RESCUER
1978 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1979 * to ask for a number of bytes to be allocated in front of the bio.
1980 * Front pad allocation is useful for embedding the bio inside
1981 * another structure, to avoid allocating extra data to go with the bio.
1982 * Note that the bio must be embedded at the END of that structure always,
1983 * or things will break badly.
1984 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1985 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast().
1986 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
1987 * dispatch queued requests when the mempool runs out of space.
1990 int bioset_init(struct bio_set
*bs
,
1991 unsigned int pool_size
,
1992 unsigned int front_pad
,
1995 unsigned int back_pad
= BIO_INLINE_VECS
* sizeof(struct bio_vec
);
1997 bs
->front_pad
= front_pad
;
1999 spin_lock_init(&bs
->rescue_lock
);
2000 bio_list_init(&bs
->rescue_list
);
2001 INIT_WORK(&bs
->rescue_work
, bio_alloc_rescue
);
2003 bs
->bio_slab
= bio_find_or_create_slab(front_pad
+ back_pad
);
2007 if (mempool_init_slab_pool(&bs
->bio_pool
, pool_size
, bs
->bio_slab
))
2010 if ((flags
& BIOSET_NEED_BVECS
) &&
2011 biovec_init_pool(&bs
->bvec_pool
, pool_size
))
2014 if (!(flags
& BIOSET_NEED_RESCUER
))
2017 bs
->rescue_workqueue
= alloc_workqueue("bioset", WQ_MEM_RECLAIM
, 0);
2018 if (!bs
->rescue_workqueue
)
2026 EXPORT_SYMBOL(bioset_init
);
2029 * Initialize and setup a new bio_set, based on the settings from
2032 int bioset_init_from_src(struct bio_set
*bs
, struct bio_set
*src
)
2037 if (src
->bvec_pool
.min_nr
)
2038 flags
|= BIOSET_NEED_BVECS
;
2039 if (src
->rescue_workqueue
)
2040 flags
|= BIOSET_NEED_RESCUER
;
2042 return bioset_init(bs
, src
->bio_pool
.min_nr
, src
->front_pad
, flags
);
2044 EXPORT_SYMBOL(bioset_init_from_src
);
2046 #ifdef CONFIG_BLK_CGROUP
2049 * bio_disassociate_blkg - puts back the blkg reference if associated
2052 * Helper to disassociate the blkg from @bio if a blkg is associated.
2054 void bio_disassociate_blkg(struct bio
*bio
)
2057 blkg_put(bio
->bi_blkg
);
2058 bio
->bi_blkg
= NULL
;
2061 EXPORT_SYMBOL_GPL(bio_disassociate_blkg
);
2064 * __bio_associate_blkg - associate a bio with the a blkg
2066 * @blkg: the blkg to associate
2068 * This tries to associate @bio with the specified @blkg. Association failure
2069 * is handled by walking up the blkg tree. Therefore, the blkg associated can
2070 * be anything between @blkg and the root_blkg. This situation only happens
2071 * when a cgroup is dying and then the remaining bios will spill to the closest
2074 * A reference will be taken on the @blkg and will be released when @bio is
2077 static void __bio_associate_blkg(struct bio
*bio
, struct blkcg_gq
*blkg
)
2079 bio_disassociate_blkg(bio
);
2081 bio
->bi_blkg
= blkg_tryget_closest(blkg
);
2085 * bio_associate_blkg_from_css - associate a bio with a specified css
2089 * Associate @bio with the blkg found by combining the css's blkg and the
2090 * request_queue of the @bio. This falls back to the queue's root_blkg if
2091 * the association fails with the css.
2093 void bio_associate_blkg_from_css(struct bio
*bio
,
2094 struct cgroup_subsys_state
*css
)
2096 struct request_queue
*q
= bio
->bi_disk
->queue
;
2097 struct blkcg_gq
*blkg
;
2101 if (!css
|| !css
->parent
)
2102 blkg
= q
->root_blkg
;
2104 blkg
= blkg_lookup_create(css_to_blkcg(css
), q
);
2106 __bio_associate_blkg(bio
, blkg
);
2110 EXPORT_SYMBOL_GPL(bio_associate_blkg_from_css
);
2114 * bio_associate_blkg_from_page - associate a bio with the page's blkg
2116 * @page: the page to lookup the blkcg from
2118 * Associate @bio with the blkg from @page's owning memcg and the respective
2119 * request_queue. If cgroup_e_css returns %NULL, fall back to the queue's
2122 void bio_associate_blkg_from_page(struct bio
*bio
, struct page
*page
)
2124 struct cgroup_subsys_state
*css
;
2126 if (!page
->mem_cgroup
)
2131 css
= cgroup_e_css(page
->mem_cgroup
->css
.cgroup
, &io_cgrp_subsys
);
2132 bio_associate_blkg_from_css(bio
, css
);
2136 #endif /* CONFIG_MEMCG */
2139 * bio_associate_blkg - associate a bio with a blkg
2142 * Associate @bio with the blkg found from the bio's css and request_queue.
2143 * If one is not found, bio_lookup_blkg() creates the blkg. If a blkg is
2144 * already associated, the css is reused and association redone as the
2145 * request_queue may have changed.
2147 void bio_associate_blkg(struct bio
*bio
)
2149 struct cgroup_subsys_state
*css
;
2154 css
= &bio_blkcg(bio
)->css
;
2158 bio_associate_blkg_from_css(bio
, css
);
2162 EXPORT_SYMBOL_GPL(bio_associate_blkg
);
2165 * bio_clone_blkg_association - clone blkg association from src to dst bio
2166 * @dst: destination bio
2169 void bio_clone_blkg_association(struct bio
*dst
, struct bio
*src
)
2174 __bio_associate_blkg(dst
, src
->bi_blkg
);
2178 EXPORT_SYMBOL_GPL(bio_clone_blkg_association
);
2179 #endif /* CONFIG_BLK_CGROUP */
2181 static void __init
biovec_init_slabs(void)
2185 for (i
= 0; i
< BVEC_POOL_NR
; i
++) {
2187 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
2189 if (bvs
->nr_vecs
<= BIO_INLINE_VECS
) {
2194 size
= bvs
->nr_vecs
* sizeof(struct bio_vec
);
2195 bvs
->slab
= kmem_cache_create(bvs
->name
, size
, 0,
2196 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
2200 static int __init
init_bio(void)
2204 bio_slabs
= kcalloc(bio_slab_max
, sizeof(struct bio_slab
),
2207 panic("bio: can't allocate bios\n");
2209 bio_integrity_init();
2210 biovec_init_slabs();
2212 if (bioset_init(&fs_bio_set
, BIO_POOL_SIZE
, 0, BIOSET_NEED_BVECS
))
2213 panic("bio: can't allocate bios\n");
2215 if (bioset_integrity_create(&fs_bio_set
, BIO_POOL_SIZE
))
2216 panic("bio: can't create integrity pool\n");
2220 subsys_initcall(init_bio
);