]>
git.proxmox.com Git - mirror_ubuntu-hirsute-kernel.git/blob - block/bio.c
2 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
13 * You should have received a copy of the GNU General Public Licens
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
19 #include <linux/swap.h>
20 #include <linux/bio.h>
21 #include <linux/blkdev.h>
22 #include <linux/uio.h>
23 #include <linux/iocontext.h>
24 #include <linux/slab.h>
25 #include <linux/init.h>
26 #include <linux/kernel.h>
27 #include <linux/export.h>
28 #include <linux/mempool.h>
29 #include <linux/workqueue.h>
30 #include <linux/cgroup.h>
32 #include <trace/events/block.h>
36 * Test patch to inline a certain number of bi_io_vec's inside the bio
37 * itself, to shrink a bio data allocation from two mempool calls to one
39 #define BIO_INLINE_VECS 4
42 * if you change this list, also change bvec_alloc or things will
43 * break badly! cannot be bigger than what you can fit into an
46 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
47 static struct biovec_slab bvec_slabs
[BVEC_POOL_NR
] __read_mostly
= {
48 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES
),
53 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
54 * IO code that does not need private memory pools.
56 struct bio_set
*fs_bio_set
;
57 EXPORT_SYMBOL(fs_bio_set
);
60 * Our slab pool management
63 struct kmem_cache
*slab
;
64 unsigned int slab_ref
;
65 unsigned int slab_size
;
68 static DEFINE_MUTEX(bio_slab_lock
);
69 static struct bio_slab
*bio_slabs
;
70 static unsigned int bio_slab_nr
, bio_slab_max
;
72 static struct kmem_cache
*bio_find_or_create_slab(unsigned int extra_size
)
74 unsigned int sz
= sizeof(struct bio
) + extra_size
;
75 struct kmem_cache
*slab
= NULL
;
76 struct bio_slab
*bslab
, *new_bio_slabs
;
77 unsigned int new_bio_slab_max
;
78 unsigned int i
, entry
= -1;
80 mutex_lock(&bio_slab_lock
);
83 while (i
< bio_slab_nr
) {
84 bslab
= &bio_slabs
[i
];
86 if (!bslab
->slab
&& entry
== -1)
88 else if (bslab
->slab_size
== sz
) {
99 if (bio_slab_nr
== bio_slab_max
&& entry
== -1) {
100 new_bio_slab_max
= bio_slab_max
<< 1;
101 new_bio_slabs
= krealloc(bio_slabs
,
102 new_bio_slab_max
* sizeof(struct bio_slab
),
106 bio_slab_max
= new_bio_slab_max
;
107 bio_slabs
= new_bio_slabs
;
110 entry
= bio_slab_nr
++;
112 bslab
= &bio_slabs
[entry
];
114 snprintf(bslab
->name
, sizeof(bslab
->name
), "bio-%d", entry
);
115 slab
= kmem_cache_create(bslab
->name
, sz
, ARCH_KMALLOC_MINALIGN
,
116 SLAB_HWCACHE_ALIGN
, NULL
);
122 bslab
->slab_size
= sz
;
124 mutex_unlock(&bio_slab_lock
);
128 static void bio_put_slab(struct bio_set
*bs
)
130 struct bio_slab
*bslab
= NULL
;
133 mutex_lock(&bio_slab_lock
);
135 for (i
= 0; i
< bio_slab_nr
; i
++) {
136 if (bs
->bio_slab
== bio_slabs
[i
].slab
) {
137 bslab
= &bio_slabs
[i
];
142 if (WARN(!bslab
, KERN_ERR
"bio: unable to find slab!\n"))
145 WARN_ON(!bslab
->slab_ref
);
147 if (--bslab
->slab_ref
)
150 kmem_cache_destroy(bslab
->slab
);
154 mutex_unlock(&bio_slab_lock
);
157 unsigned int bvec_nr_vecs(unsigned short idx
)
159 return bvec_slabs
[idx
].nr_vecs
;
162 void bvec_free(mempool_t
*pool
, struct bio_vec
*bv
, unsigned int idx
)
168 BIO_BUG_ON(idx
>= BVEC_POOL_NR
);
170 if (idx
== BVEC_POOL_MAX
) {
171 mempool_free(bv
, pool
);
173 struct biovec_slab
*bvs
= bvec_slabs
+ idx
;
175 kmem_cache_free(bvs
->slab
, bv
);
179 struct bio_vec
*bvec_alloc(gfp_t gfp_mask
, int nr
, unsigned long *idx
,
185 * see comment near bvec_array define!
203 case 129 ... BIO_MAX_PAGES
:
211 * idx now points to the pool we want to allocate from. only the
212 * 1-vec entry pool is mempool backed.
214 if (*idx
== BVEC_POOL_MAX
) {
216 bvl
= mempool_alloc(pool
, gfp_mask
);
218 struct biovec_slab
*bvs
= bvec_slabs
+ *idx
;
219 gfp_t __gfp_mask
= gfp_mask
& ~(__GFP_DIRECT_RECLAIM
| __GFP_IO
);
222 * Make this allocation restricted and don't dump info on
223 * allocation failures, since we'll fallback to the mempool
224 * in case of failure.
226 __gfp_mask
|= __GFP_NOMEMALLOC
| __GFP_NORETRY
| __GFP_NOWARN
;
229 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
230 * is set, retry with the 1-entry mempool
232 bvl
= kmem_cache_alloc(bvs
->slab
, __gfp_mask
);
233 if (unlikely(!bvl
&& (gfp_mask
& __GFP_DIRECT_RECLAIM
))) {
234 *idx
= BVEC_POOL_MAX
;
243 static void __bio_free(struct bio
*bio
)
245 bio_disassociate_task(bio
);
247 if (bio_integrity(bio
))
248 bio_integrity_free(bio
);
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() */
274 void bio_init(struct bio
*bio
, struct bio_vec
*table
,
275 unsigned short max_vecs
)
277 memset(bio
, 0, sizeof(*bio
));
278 atomic_set(&bio
->__bi_remaining
, 1);
279 atomic_set(&bio
->__bi_cnt
, 1);
281 bio
->bi_io_vec
= table
;
282 bio
->bi_max_vecs
= max_vecs
;
284 EXPORT_SYMBOL(bio_init
);
287 * bio_reset - reinitialize a bio
291 * After calling bio_reset(), @bio will be in the same state as a freshly
292 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
293 * preserved are the ones that are initialized by bio_alloc_bioset(). See
294 * comment in struct bio.
296 void bio_reset(struct bio
*bio
)
298 unsigned long flags
= bio
->bi_flags
& (~0UL << BIO_RESET_BITS
);
302 memset(bio
, 0, BIO_RESET_BYTES
);
303 bio
->bi_flags
= flags
;
304 atomic_set(&bio
->__bi_remaining
, 1);
306 EXPORT_SYMBOL(bio_reset
);
308 static struct bio
*__bio_chain_endio(struct bio
*bio
)
310 struct bio
*parent
= bio
->bi_private
;
312 if (!parent
->bi_error
)
313 parent
->bi_error
= bio
->bi_error
;
318 static void bio_chain_endio(struct bio
*bio
)
320 bio_endio(__bio_chain_endio(bio
));
324 * bio_chain - chain bio completions
325 * @bio: the target bio
326 * @parent: the @bio's parent bio
328 * The caller won't have a bi_end_io called when @bio completes - instead,
329 * @parent's bi_end_io won't be called until both @parent and @bio have
330 * completed; the chained bio will also be freed when it completes.
332 * The caller must not set bi_private or bi_end_io in @bio.
334 void bio_chain(struct bio
*bio
, struct bio
*parent
)
336 BUG_ON(bio
->bi_private
|| bio
->bi_end_io
);
338 bio
->bi_private
= parent
;
339 bio
->bi_end_io
= bio_chain_endio
;
340 bio_inc_remaining(parent
);
342 EXPORT_SYMBOL(bio_chain
);
344 static void bio_alloc_rescue(struct work_struct
*work
)
346 struct bio_set
*bs
= container_of(work
, struct bio_set
, rescue_work
);
350 spin_lock(&bs
->rescue_lock
);
351 bio
= bio_list_pop(&bs
->rescue_list
);
352 spin_unlock(&bs
->rescue_lock
);
357 generic_make_request(bio
);
361 static void punt_bios_to_rescuer(struct bio_set
*bs
)
363 struct bio_list punt
, nopunt
;
367 * In order to guarantee forward progress we must punt only bios that
368 * were allocated from this bio_set; otherwise, if there was a bio on
369 * there for a stacking driver higher up in the stack, processing it
370 * could require allocating bios from this bio_set, and doing that from
371 * our own rescuer would be bad.
373 * Since bio lists are singly linked, pop them all instead of trying to
374 * remove from the middle of the list:
377 bio_list_init(&punt
);
378 bio_list_init(&nopunt
);
380 while ((bio
= bio_list_pop(¤t
->bio_list
[0])))
381 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
382 current
->bio_list
[0] = nopunt
;
384 bio_list_init(&nopunt
);
385 while ((bio
= bio_list_pop(¤t
->bio_list
[1])))
386 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
387 current
->bio_list
[1] = nopunt
;
389 spin_lock(&bs
->rescue_lock
);
390 bio_list_merge(&bs
->rescue_list
, &punt
);
391 spin_unlock(&bs
->rescue_lock
);
393 queue_work(bs
->rescue_workqueue
, &bs
->rescue_work
);
397 * bio_alloc_bioset - allocate a bio for I/O
398 * @gfp_mask: the GFP_ mask given to the slab allocator
399 * @nr_iovecs: number of iovecs to pre-allocate
400 * @bs: the bio_set to allocate from.
403 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
404 * backed by the @bs's mempool.
406 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
407 * always be able to allocate a bio. This is due to the mempool guarantees.
408 * To make this work, callers must never allocate more than 1 bio at a time
409 * from this pool. Callers that need to allocate more than 1 bio must always
410 * submit the previously allocated bio for IO before attempting to allocate
411 * a new one. Failure to do so can cause deadlocks under memory pressure.
413 * Note that when running under generic_make_request() (i.e. any block
414 * driver), bios are not submitted until after you return - see the code in
415 * generic_make_request() that converts recursion into iteration, to prevent
418 * This would normally mean allocating multiple bios under
419 * generic_make_request() would be susceptible to deadlocks, but we have
420 * deadlock avoidance code that resubmits any blocked bios from a rescuer
423 * However, we do not guarantee forward progress for allocations from other
424 * mempools. Doing multiple allocations from the same mempool under
425 * generic_make_request() should be avoided - instead, use bio_set's front_pad
426 * for per bio allocations.
429 * Pointer to new bio on success, NULL on failure.
431 struct bio
*bio_alloc_bioset(gfp_t gfp_mask
, unsigned int nr_iovecs
,
434 gfp_t saved_gfp
= gfp_mask
;
436 unsigned inline_vecs
;
437 struct bio_vec
*bvl
= NULL
;
442 if (nr_iovecs
> UIO_MAXIOV
)
445 p
= kmalloc(sizeof(struct bio
) +
446 nr_iovecs
* sizeof(struct bio_vec
),
449 inline_vecs
= nr_iovecs
;
451 /* should not use nobvec bioset for nr_iovecs > 0 */
452 if (WARN_ON_ONCE(!bs
->bvec_pool
&& nr_iovecs
> 0))
455 * generic_make_request() converts recursion to iteration; this
456 * means if we're running beneath it, any bios we allocate and
457 * submit will not be submitted (and thus freed) until after we
460 * This exposes us to a potential deadlock if we allocate
461 * multiple bios from the same bio_set() while running
462 * underneath generic_make_request(). If we were to allocate
463 * multiple bios (say a stacking block driver that was splitting
464 * bios), we would deadlock if we exhausted the mempool's
467 * We solve this, and guarantee forward progress, with a rescuer
468 * workqueue per bio_set. If we go to allocate and there are
469 * bios on current->bio_list, we first try the allocation
470 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
471 * bios we would be blocking to the rescuer workqueue before
472 * we retry with the original gfp_flags.
475 if (current
->bio_list
&&
476 (!bio_list_empty(¤t
->bio_list
[0]) ||
477 !bio_list_empty(¤t
->bio_list
[1])))
478 gfp_mask
&= ~__GFP_DIRECT_RECLAIM
;
480 p
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
481 if (!p
&& gfp_mask
!= saved_gfp
) {
482 punt_bios_to_rescuer(bs
);
483 gfp_mask
= saved_gfp
;
484 p
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
487 front_pad
= bs
->front_pad
;
488 inline_vecs
= BIO_INLINE_VECS
;
495 bio_init(bio
, NULL
, 0);
497 if (nr_iovecs
> inline_vecs
) {
498 unsigned long idx
= 0;
500 bvl
= bvec_alloc(gfp_mask
, nr_iovecs
, &idx
, bs
->bvec_pool
);
501 if (!bvl
&& gfp_mask
!= saved_gfp
) {
502 punt_bios_to_rescuer(bs
);
503 gfp_mask
= saved_gfp
;
504 bvl
= bvec_alloc(gfp_mask
, nr_iovecs
, &idx
, bs
->bvec_pool
);
510 bio
->bi_flags
|= idx
<< BVEC_POOL_OFFSET
;
511 } else if (nr_iovecs
) {
512 bvl
= bio
->bi_inline_vecs
;
516 bio
->bi_max_vecs
= nr_iovecs
;
517 bio
->bi_io_vec
= bvl
;
521 mempool_free(p
, bs
->bio_pool
);
524 EXPORT_SYMBOL(bio_alloc_bioset
);
526 void zero_fill_bio(struct bio
*bio
)
530 struct bvec_iter iter
;
532 bio_for_each_segment(bv
, bio
, iter
) {
533 char *data
= bvec_kmap_irq(&bv
, &flags
);
534 memset(data
, 0, bv
.bv_len
);
535 flush_dcache_page(bv
.bv_page
);
536 bvec_kunmap_irq(data
, &flags
);
539 EXPORT_SYMBOL(zero_fill_bio
);
542 * bio_put - release a reference to a bio
543 * @bio: bio to release reference to
546 * Put a reference to a &struct bio, either one you have gotten with
547 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
549 void bio_put(struct bio
*bio
)
551 if (!bio_flagged(bio
, BIO_REFFED
))
554 BIO_BUG_ON(!atomic_read(&bio
->__bi_cnt
));
559 if (atomic_dec_and_test(&bio
->__bi_cnt
))
563 EXPORT_SYMBOL(bio_put
);
565 inline int bio_phys_segments(struct request_queue
*q
, struct bio
*bio
)
567 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
568 blk_recount_segments(q
, bio
);
570 return bio
->bi_phys_segments
;
572 EXPORT_SYMBOL(bio_phys_segments
);
575 * __bio_clone_fast - clone a bio that shares the original bio's biovec
576 * @bio: destination bio
577 * @bio_src: bio to clone
579 * Clone a &bio. Caller will own the returned bio, but not
580 * the actual data it points to. Reference count of returned
583 * Caller must ensure that @bio_src is not freed before @bio.
585 void __bio_clone_fast(struct bio
*bio
, struct bio
*bio_src
)
587 BUG_ON(bio
->bi_pool
&& BVEC_POOL_IDX(bio
));
590 * most users will be overriding ->bi_bdev with a new target,
591 * so we don't set nor calculate new physical/hw segment counts here
593 bio
->bi_bdev
= bio_src
->bi_bdev
;
594 bio_set_flag(bio
, BIO_CLONED
);
595 bio
->bi_opf
= bio_src
->bi_opf
;
596 bio
->bi_iter
= bio_src
->bi_iter
;
597 bio
->bi_io_vec
= bio_src
->bi_io_vec
;
599 bio_clone_blkcg_association(bio
, bio_src
);
601 EXPORT_SYMBOL(__bio_clone_fast
);
604 * bio_clone_fast - clone a bio that shares the original bio's biovec
606 * @gfp_mask: allocation priority
607 * @bs: bio_set to allocate from
609 * Like __bio_clone_fast, only also allocates the returned bio
611 struct bio
*bio_clone_fast(struct bio
*bio
, gfp_t gfp_mask
, struct bio_set
*bs
)
615 b
= bio_alloc_bioset(gfp_mask
, 0, bs
);
619 __bio_clone_fast(b
, bio
);
621 if (bio_integrity(bio
)) {
624 ret
= bio_integrity_clone(b
, bio
, gfp_mask
);
634 EXPORT_SYMBOL(bio_clone_fast
);
637 * bio_clone_bioset - clone a bio
638 * @bio_src: bio to clone
639 * @gfp_mask: allocation priority
640 * @bs: bio_set to allocate from
642 * Clone bio. Caller will own the returned bio, but not the actual data it
643 * points to. Reference count of returned bio will be one.
645 struct bio
*bio_clone_bioset(struct bio
*bio_src
, gfp_t gfp_mask
,
648 struct bvec_iter iter
;
653 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
654 * bio_src->bi_io_vec to bio->bi_io_vec.
656 * We can't do that anymore, because:
658 * - The point of cloning the biovec is to produce a bio with a biovec
659 * the caller can modify: bi_idx and bi_bvec_done should be 0.
661 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
662 * we tried to clone the whole thing bio_alloc_bioset() would fail.
663 * But the clone should succeed as long as the number of biovecs we
664 * actually need to allocate is fewer than BIO_MAX_PAGES.
666 * - Lastly, bi_vcnt should not be looked at or relied upon by code
667 * that does not own the bio - reason being drivers don't use it for
668 * iterating over the biovec anymore, so expecting it to be kept up
669 * to date (i.e. for clones that share the parent biovec) is just
670 * asking for trouble and would force extra work on
671 * __bio_clone_fast() anyways.
674 bio
= bio_alloc_bioset(gfp_mask
, bio_segments(bio_src
), bs
);
677 bio
->bi_bdev
= bio_src
->bi_bdev
;
678 bio
->bi_opf
= bio_src
->bi_opf
;
679 bio
->bi_iter
.bi_sector
= bio_src
->bi_iter
.bi_sector
;
680 bio
->bi_iter
.bi_size
= bio_src
->bi_iter
.bi_size
;
682 switch (bio_op(bio
)) {
684 case REQ_OP_SECURE_ERASE
:
685 case REQ_OP_WRITE_ZEROES
:
687 case REQ_OP_WRITE_SAME
:
688 bio
->bi_io_vec
[bio
->bi_vcnt
++] = bio_src
->bi_io_vec
[0];
691 bio_for_each_segment(bv
, bio_src
, iter
)
692 bio
->bi_io_vec
[bio
->bi_vcnt
++] = bv
;
696 if (bio_integrity(bio_src
)) {
699 ret
= bio_integrity_clone(bio
, bio_src
, gfp_mask
);
706 bio_clone_blkcg_association(bio
, bio_src
);
710 EXPORT_SYMBOL(bio_clone_bioset
);
713 * bio_add_pc_page - attempt to add page to bio
714 * @q: the target queue
715 * @bio: destination bio
717 * @len: vec entry length
718 * @offset: vec entry offset
720 * Attempt to add a page to the bio_vec maplist. This can fail for a
721 * number of reasons, such as the bio being full or target block device
722 * limitations. The target block device must allow bio's up to PAGE_SIZE,
723 * so it is always possible to add a single page to an empty bio.
725 * This should only be used by REQ_PC bios.
727 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
, struct page
728 *page
, unsigned int len
, unsigned int offset
)
730 int retried_segments
= 0;
731 struct bio_vec
*bvec
;
734 * cloned bio must not modify vec list
736 if (unlikely(bio_flagged(bio
, BIO_CLONED
)))
739 if (((bio
->bi_iter
.bi_size
+ len
) >> 9) > queue_max_hw_sectors(q
))
743 * For filesystems with a blocksize smaller than the pagesize
744 * we will often be called with the same page as last time and
745 * a consecutive offset. Optimize this special case.
747 if (bio
->bi_vcnt
> 0) {
748 struct bio_vec
*prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
750 if (page
== prev
->bv_page
&&
751 offset
== prev
->bv_offset
+ prev
->bv_len
) {
753 bio
->bi_iter
.bi_size
+= len
;
758 * If the queue doesn't support SG gaps and adding this
759 * offset would create a gap, disallow it.
761 if (bvec_gap_to_prev(q
, prev
, offset
))
765 if (bio
->bi_vcnt
>= bio
->bi_max_vecs
)
769 * setup the new entry, we might clear it again later if we
770 * cannot add the page
772 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
773 bvec
->bv_page
= page
;
775 bvec
->bv_offset
= offset
;
777 bio
->bi_phys_segments
++;
778 bio
->bi_iter
.bi_size
+= len
;
781 * Perform a recount if the number of segments is greater
782 * than queue_max_segments(q).
785 while (bio
->bi_phys_segments
> queue_max_segments(q
)) {
787 if (retried_segments
)
790 retried_segments
= 1;
791 blk_recount_segments(q
, bio
);
794 /* If we may be able to merge these biovecs, force a recount */
795 if (bio
->bi_vcnt
> 1 && (BIOVEC_PHYS_MERGEABLE(bvec
-1, bvec
)))
796 bio_clear_flag(bio
, BIO_SEG_VALID
);
802 bvec
->bv_page
= NULL
;
806 bio
->bi_iter
.bi_size
-= len
;
807 blk_recount_segments(q
, bio
);
810 EXPORT_SYMBOL(bio_add_pc_page
);
813 * bio_add_page - attempt to add page to bio
814 * @bio: destination bio
816 * @len: vec entry length
817 * @offset: vec entry offset
819 * Attempt to add a page to the bio_vec maplist. This will only fail
820 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
822 int bio_add_page(struct bio
*bio
, struct page
*page
,
823 unsigned int len
, unsigned int offset
)
828 * cloned bio must not modify vec list
830 if (WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
)))
834 * For filesystems with a blocksize smaller than the pagesize
835 * we will often be called with the same page as last time and
836 * a consecutive offset. Optimize this special case.
838 if (bio
->bi_vcnt
> 0) {
839 bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
841 if (page
== bv
->bv_page
&&
842 offset
== bv
->bv_offset
+ bv
->bv_len
) {
848 if (bio
->bi_vcnt
>= bio
->bi_max_vecs
)
851 bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
854 bv
->bv_offset
= offset
;
858 bio
->bi_iter
.bi_size
+= len
;
861 EXPORT_SYMBOL(bio_add_page
);
864 * bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
865 * @bio: bio to add pages to
866 * @iter: iov iterator describing the region to be mapped
868 * Pins as many pages from *iter and appends them to @bio's bvec array. The
869 * pages will have to be released using put_page() when done.
871 int bio_iov_iter_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
873 unsigned short nr_pages
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
874 struct bio_vec
*bv
= bio
->bi_io_vec
+ bio
->bi_vcnt
;
875 struct page
**pages
= (struct page
**)bv
;
879 size
= iov_iter_get_pages(iter
, pages
, LONG_MAX
, nr_pages
, &offset
);
880 if (unlikely(size
<= 0))
881 return size
? size
: -EFAULT
;
882 nr_pages
= (size
+ offset
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
885 * Deep magic below: We need to walk the pinned pages backwards
886 * because we are abusing the space allocated for the bio_vecs
887 * for the page array. Because the bio_vecs are larger than the
888 * page pointers by definition this will always work. But it also
889 * means we can't use bio_add_page, so any changes to it's semantics
890 * need to be reflected here as well.
892 bio
->bi_iter
.bi_size
+= size
;
893 bio
->bi_vcnt
+= nr_pages
;
895 diff
= (nr_pages
* PAGE_SIZE
- offset
) - size
;
897 bv
[nr_pages
].bv_page
= pages
[nr_pages
];
898 bv
[nr_pages
].bv_len
= PAGE_SIZE
;
899 bv
[nr_pages
].bv_offset
= 0;
902 bv
[0].bv_offset
+= offset
;
903 bv
[0].bv_len
-= offset
;
905 bv
[bio
->bi_vcnt
- 1].bv_len
-= diff
;
907 iov_iter_advance(iter
, size
);
910 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages
);
912 struct submit_bio_ret
{
913 struct completion event
;
917 static void submit_bio_wait_endio(struct bio
*bio
)
919 struct submit_bio_ret
*ret
= bio
->bi_private
;
921 ret
->error
= bio
->bi_error
;
922 complete(&ret
->event
);
926 * submit_bio_wait - submit a bio, and wait until it completes
927 * @bio: The &struct bio which describes the I/O
929 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
930 * bio_endio() on failure.
932 int submit_bio_wait(struct bio
*bio
)
934 struct submit_bio_ret ret
;
936 init_completion(&ret
.event
);
937 bio
->bi_private
= &ret
;
938 bio
->bi_end_io
= submit_bio_wait_endio
;
939 bio
->bi_opf
|= REQ_SYNC
;
941 wait_for_completion_io(&ret
.event
);
945 EXPORT_SYMBOL(submit_bio_wait
);
948 * bio_advance - increment/complete a bio by some number of bytes
949 * @bio: bio to advance
950 * @bytes: number of bytes to complete
952 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
953 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
954 * be updated on the last bvec as well.
956 * @bio will then represent the remaining, uncompleted portion of the io.
958 void bio_advance(struct bio
*bio
, unsigned bytes
)
960 if (bio_integrity(bio
))
961 bio_integrity_advance(bio
, bytes
);
963 bio_advance_iter(bio
, &bio
->bi_iter
, bytes
);
965 EXPORT_SYMBOL(bio_advance
);
968 * bio_alloc_pages - allocates a single page for each bvec in a bio
969 * @bio: bio to allocate pages for
970 * @gfp_mask: flags for allocation
972 * Allocates pages up to @bio->bi_vcnt.
974 * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
977 int bio_alloc_pages(struct bio
*bio
, gfp_t gfp_mask
)
982 bio_for_each_segment_all(bv
, bio
, i
) {
983 bv
->bv_page
= alloc_page(gfp_mask
);
985 while (--bv
>= bio
->bi_io_vec
)
986 __free_page(bv
->bv_page
);
993 EXPORT_SYMBOL(bio_alloc_pages
);
996 * bio_copy_data - copy contents of data buffers from one chain of bios to
998 * @src: source bio list
999 * @dst: destination bio list
1001 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
1002 * @src and @dst as linked lists of bios.
1004 * Stops when it reaches the end of either @src or @dst - that is, copies
1005 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1007 void bio_copy_data(struct bio
*dst
, struct bio
*src
)
1009 struct bvec_iter src_iter
, dst_iter
;
1010 struct bio_vec src_bv
, dst_bv
;
1011 void *src_p
, *dst_p
;
1014 src_iter
= src
->bi_iter
;
1015 dst_iter
= dst
->bi_iter
;
1018 if (!src_iter
.bi_size
) {
1023 src_iter
= src
->bi_iter
;
1026 if (!dst_iter
.bi_size
) {
1031 dst_iter
= dst
->bi_iter
;
1034 src_bv
= bio_iter_iovec(src
, src_iter
);
1035 dst_bv
= bio_iter_iovec(dst
, dst_iter
);
1037 bytes
= min(src_bv
.bv_len
, dst_bv
.bv_len
);
1039 src_p
= kmap_atomic(src_bv
.bv_page
);
1040 dst_p
= kmap_atomic(dst_bv
.bv_page
);
1042 memcpy(dst_p
+ dst_bv
.bv_offset
,
1043 src_p
+ src_bv
.bv_offset
,
1046 kunmap_atomic(dst_p
);
1047 kunmap_atomic(src_p
);
1049 bio_advance_iter(src
, &src_iter
, bytes
);
1050 bio_advance_iter(dst
, &dst_iter
, bytes
);
1053 EXPORT_SYMBOL(bio_copy_data
);
1055 struct bio_map_data
{
1057 struct iov_iter iter
;
1061 static struct bio_map_data
*bio_alloc_map_data(unsigned int iov_count
,
1064 if (iov_count
> UIO_MAXIOV
)
1067 return kmalloc(sizeof(struct bio_map_data
) +
1068 sizeof(struct iovec
) * iov_count
, gfp_mask
);
1072 * bio_copy_from_iter - copy all pages from iov_iter to bio
1073 * @bio: The &struct bio which describes the I/O as destination
1074 * @iter: iov_iter as source
1076 * Copy all pages from iov_iter to bio.
1077 * Returns 0 on success, or error on failure.
1079 static int bio_copy_from_iter(struct bio
*bio
, struct iov_iter iter
)
1082 struct bio_vec
*bvec
;
1084 bio_for_each_segment_all(bvec
, bio
, i
) {
1087 ret
= copy_page_from_iter(bvec
->bv_page
,
1092 if (!iov_iter_count(&iter
))
1095 if (ret
< bvec
->bv_len
)
1103 * bio_copy_to_iter - copy all pages from bio to iov_iter
1104 * @bio: The &struct bio which describes the I/O as source
1105 * @iter: iov_iter as destination
1107 * Copy all pages from bio to iov_iter.
1108 * Returns 0 on success, or error on failure.
1110 static int bio_copy_to_iter(struct bio
*bio
, struct iov_iter iter
)
1113 struct bio_vec
*bvec
;
1115 bio_for_each_segment_all(bvec
, bio
, i
) {
1118 ret
= copy_page_to_iter(bvec
->bv_page
,
1123 if (!iov_iter_count(&iter
))
1126 if (ret
< bvec
->bv_len
)
1133 void bio_free_pages(struct bio
*bio
)
1135 struct bio_vec
*bvec
;
1138 bio_for_each_segment_all(bvec
, bio
, i
)
1139 __free_page(bvec
->bv_page
);
1141 EXPORT_SYMBOL(bio_free_pages
);
1144 * bio_uncopy_user - finish previously mapped bio
1145 * @bio: bio being terminated
1147 * Free pages allocated from bio_copy_user_iov() and write back data
1148 * to user space in case of a read.
1150 int bio_uncopy_user(struct bio
*bio
)
1152 struct bio_map_data
*bmd
= bio
->bi_private
;
1155 if (!bio_flagged(bio
, BIO_NULL_MAPPED
)) {
1157 * if we're in a workqueue, the request is orphaned, so
1158 * don't copy into a random user address space, just free
1159 * and return -EINTR so user space doesn't expect any data.
1163 else if (bio_data_dir(bio
) == READ
)
1164 ret
= bio_copy_to_iter(bio
, bmd
->iter
);
1165 if (bmd
->is_our_pages
)
1166 bio_free_pages(bio
);
1174 * bio_copy_user_iov - copy user data to bio
1175 * @q: destination block queue
1176 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1177 * @iter: iovec iterator
1178 * @gfp_mask: memory allocation flags
1180 * Prepares and returns a bio for indirect user io, bouncing data
1181 * to/from kernel pages as necessary. Must be paired with
1182 * call bio_uncopy_user() on io completion.
1184 struct bio
*bio_copy_user_iov(struct request_queue
*q
,
1185 struct rq_map_data
*map_data
,
1186 const struct iov_iter
*iter
,
1189 struct bio_map_data
*bmd
;
1194 unsigned int len
= iter
->count
;
1195 unsigned int offset
= map_data
? offset_in_page(map_data
->offset
) : 0;
1197 for (i
= 0; i
< iter
->nr_segs
; i
++) {
1198 unsigned long uaddr
;
1200 unsigned long start
;
1202 uaddr
= (unsigned long) iter
->iov
[i
].iov_base
;
1203 end
= (uaddr
+ iter
->iov
[i
].iov_len
+ PAGE_SIZE
- 1)
1205 start
= uaddr
>> PAGE_SHIFT
;
1211 return ERR_PTR(-EINVAL
);
1213 nr_pages
+= end
- start
;
1219 bmd
= bio_alloc_map_data(iter
->nr_segs
, gfp_mask
);
1221 return ERR_PTR(-ENOMEM
);
1224 * We need to do a deep copy of the iov_iter including the iovecs.
1225 * The caller provided iov might point to an on-stack or otherwise
1228 bmd
->is_our_pages
= map_data
? 0 : 1;
1229 memcpy(bmd
->iov
, iter
->iov
, sizeof(struct iovec
) * iter
->nr_segs
);
1230 iov_iter_init(&bmd
->iter
, iter
->type
, bmd
->iov
,
1231 iter
->nr_segs
, iter
->count
);
1234 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1241 nr_pages
= 1 << map_data
->page_order
;
1242 i
= map_data
->offset
/ PAGE_SIZE
;
1245 unsigned int bytes
= PAGE_SIZE
;
1253 if (i
== map_data
->nr_entries
* nr_pages
) {
1258 page
= map_data
->pages
[i
/ nr_pages
];
1259 page
+= (i
% nr_pages
);
1263 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
1270 if (bio_add_pc_page(q
, bio
, page
, bytes
, offset
) < bytes
)
1283 if (((iter
->type
& WRITE
) && (!map_data
|| !map_data
->null_mapped
)) ||
1284 (map_data
&& map_data
->from_user
)) {
1285 ret
= bio_copy_from_iter(bio
, *iter
);
1290 bio
->bi_private
= bmd
;
1294 bio_free_pages(bio
);
1298 return ERR_PTR(ret
);
1302 * bio_map_user_iov - map user iovec into bio
1303 * @q: the struct request_queue for the bio
1304 * @iter: iovec iterator
1305 * @gfp_mask: memory allocation flags
1307 * Map the user space address into a bio suitable for io to a block
1308 * device. Returns an error pointer in case of error.
1310 struct bio
*bio_map_user_iov(struct request_queue
*q
,
1311 const struct iov_iter
*iter
,
1316 struct page
**pages
;
1323 iov_for_each(iov
, i
, *iter
) {
1324 unsigned long uaddr
= (unsigned long) iov
.iov_base
;
1325 unsigned long len
= iov
.iov_len
;
1326 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1327 unsigned long start
= uaddr
>> PAGE_SHIFT
;
1333 return ERR_PTR(-EINVAL
);
1335 nr_pages
+= end
- start
;
1337 * buffer must be aligned to at least logical block size for now
1339 if (uaddr
& queue_dma_alignment(q
))
1340 return ERR_PTR(-EINVAL
);
1344 return ERR_PTR(-EINVAL
);
1346 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1348 return ERR_PTR(-ENOMEM
);
1351 pages
= kcalloc(nr_pages
, sizeof(struct page
*), gfp_mask
);
1355 iov_for_each(iov
, i
, *iter
) {
1356 unsigned long uaddr
= (unsigned long) iov
.iov_base
;
1357 unsigned long len
= iov
.iov_len
;
1358 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1359 unsigned long start
= uaddr
>> PAGE_SHIFT
;
1360 const int local_nr_pages
= end
- start
;
1361 const int page_limit
= cur_page
+ local_nr_pages
;
1363 ret
= get_user_pages_fast(uaddr
, local_nr_pages
,
1364 (iter
->type
& WRITE
) != WRITE
,
1366 if (ret
< local_nr_pages
) {
1371 offset
= offset_in_page(uaddr
);
1372 for (j
= cur_page
; j
< page_limit
; j
++) {
1373 unsigned int bytes
= PAGE_SIZE
- offset
;
1384 if (bio_add_pc_page(q
, bio
, pages
[j
], bytes
, offset
) <
1394 * release the pages we didn't map into the bio, if any
1396 while (j
< page_limit
)
1397 put_page(pages
[j
++]);
1402 bio_set_flag(bio
, BIO_USER_MAPPED
);
1405 * subtle -- if bio_map_user_iov() ended up bouncing a bio,
1406 * it would normally disappear when its bi_end_io is run.
1407 * however, we need it for the unmap, so grab an extra
1414 for (j
= 0; j
< nr_pages
; j
++) {
1422 return ERR_PTR(ret
);
1425 static void __bio_unmap_user(struct bio
*bio
)
1427 struct bio_vec
*bvec
;
1431 * make sure we dirty pages we wrote to
1433 bio_for_each_segment_all(bvec
, bio
, i
) {
1434 if (bio_data_dir(bio
) == READ
)
1435 set_page_dirty_lock(bvec
->bv_page
);
1437 put_page(bvec
->bv_page
);
1444 * bio_unmap_user - unmap a bio
1445 * @bio: the bio being unmapped
1447 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from
1450 * bio_unmap_user() may sleep.
1452 void bio_unmap_user(struct bio
*bio
)
1454 __bio_unmap_user(bio
);
1458 static void bio_map_kern_endio(struct bio
*bio
)
1464 * bio_map_kern - map kernel address into bio
1465 * @q: the struct request_queue for the bio
1466 * @data: pointer to buffer to map
1467 * @len: length in bytes
1468 * @gfp_mask: allocation flags for bio allocation
1470 * Map the kernel address into a bio suitable for io to a block
1471 * device. Returns an error pointer in case of error.
1473 struct bio
*bio_map_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1476 unsigned long kaddr
= (unsigned long)data
;
1477 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1478 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1479 const int nr_pages
= end
- start
;
1483 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1485 return ERR_PTR(-ENOMEM
);
1487 offset
= offset_in_page(kaddr
);
1488 for (i
= 0; i
< nr_pages
; i
++) {
1489 unsigned int bytes
= PAGE_SIZE
- offset
;
1497 if (bio_add_pc_page(q
, bio
, virt_to_page(data
), bytes
,
1499 /* we don't support partial mappings */
1501 return ERR_PTR(-EINVAL
);
1509 bio
->bi_end_io
= bio_map_kern_endio
;
1512 EXPORT_SYMBOL(bio_map_kern
);
1514 static void bio_copy_kern_endio(struct bio
*bio
)
1516 bio_free_pages(bio
);
1520 static void bio_copy_kern_endio_read(struct bio
*bio
)
1522 char *p
= bio
->bi_private
;
1523 struct bio_vec
*bvec
;
1526 bio_for_each_segment_all(bvec
, bio
, i
) {
1527 memcpy(p
, page_address(bvec
->bv_page
), bvec
->bv_len
);
1531 bio_copy_kern_endio(bio
);
1535 * bio_copy_kern - copy kernel address into bio
1536 * @q: the struct request_queue for the bio
1537 * @data: pointer to buffer to copy
1538 * @len: length in bytes
1539 * @gfp_mask: allocation flags for bio and page allocation
1540 * @reading: data direction is READ
1542 * copy the kernel address into a bio suitable for io to a block
1543 * device. Returns an error pointer in case of error.
1545 struct bio
*bio_copy_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1546 gfp_t gfp_mask
, int reading
)
1548 unsigned long kaddr
= (unsigned long)data
;
1549 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1550 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1559 return ERR_PTR(-EINVAL
);
1561 nr_pages
= end
- start
;
1562 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1564 return ERR_PTR(-ENOMEM
);
1568 unsigned int bytes
= PAGE_SIZE
;
1573 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
1578 memcpy(page_address(page
), p
, bytes
);
1580 if (bio_add_pc_page(q
, bio
, page
, bytes
, 0) < bytes
)
1588 bio
->bi_end_io
= bio_copy_kern_endio_read
;
1589 bio
->bi_private
= data
;
1591 bio
->bi_end_io
= bio_copy_kern_endio
;
1597 bio_free_pages(bio
);
1599 return ERR_PTR(-ENOMEM
);
1603 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1604 * for performing direct-IO in BIOs.
1606 * The problem is that we cannot run set_page_dirty() from interrupt context
1607 * because the required locks are not interrupt-safe. So what we can do is to
1608 * mark the pages dirty _before_ performing IO. And in interrupt context,
1609 * check that the pages are still dirty. If so, fine. If not, redirty them
1610 * in process context.
1612 * We special-case compound pages here: normally this means reads into hugetlb
1613 * pages. The logic in here doesn't really work right for compound pages
1614 * because the VM does not uniformly chase down the head page in all cases.
1615 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1616 * handle them at all. So we skip compound pages here at an early stage.
1618 * Note that this code is very hard to test under normal circumstances because
1619 * direct-io pins the pages with get_user_pages(). This makes
1620 * is_page_cache_freeable return false, and the VM will not clean the pages.
1621 * But other code (eg, flusher threads) could clean the pages if they are mapped
1624 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1625 * deferred bio dirtying paths.
1629 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1631 void bio_set_pages_dirty(struct bio
*bio
)
1633 struct bio_vec
*bvec
;
1636 bio_for_each_segment_all(bvec
, bio
, i
) {
1637 struct page
*page
= bvec
->bv_page
;
1639 if (page
&& !PageCompound(page
))
1640 set_page_dirty_lock(page
);
1644 static void bio_release_pages(struct bio
*bio
)
1646 struct bio_vec
*bvec
;
1649 bio_for_each_segment_all(bvec
, bio
, i
) {
1650 struct page
*page
= bvec
->bv_page
;
1658 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1659 * If they are, then fine. If, however, some pages are clean then they must
1660 * have been written out during the direct-IO read. So we take another ref on
1661 * the BIO and the offending pages and re-dirty the pages in process context.
1663 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1664 * here on. It will run one put_page() against each page and will run one
1665 * bio_put() against the BIO.
1668 static void bio_dirty_fn(struct work_struct
*work
);
1670 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
1671 static DEFINE_SPINLOCK(bio_dirty_lock
);
1672 static struct bio
*bio_dirty_list
;
1675 * This runs in process context
1677 static void bio_dirty_fn(struct work_struct
*work
)
1679 unsigned long flags
;
1682 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1683 bio
= bio_dirty_list
;
1684 bio_dirty_list
= NULL
;
1685 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1688 struct bio
*next
= bio
->bi_private
;
1690 bio_set_pages_dirty(bio
);
1691 bio_release_pages(bio
);
1697 void bio_check_pages_dirty(struct bio
*bio
)
1699 struct bio_vec
*bvec
;
1700 int nr_clean_pages
= 0;
1703 bio_for_each_segment_all(bvec
, bio
, i
) {
1704 struct page
*page
= bvec
->bv_page
;
1706 if (PageDirty(page
) || PageCompound(page
)) {
1708 bvec
->bv_page
= NULL
;
1714 if (nr_clean_pages
) {
1715 unsigned long flags
;
1717 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1718 bio
->bi_private
= bio_dirty_list
;
1719 bio_dirty_list
= bio
;
1720 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1721 schedule_work(&bio_dirty_work
);
1727 void generic_start_io_acct(int rw
, unsigned long sectors
,
1728 struct hd_struct
*part
)
1730 int cpu
= part_stat_lock();
1732 part_round_stats(cpu
, part
);
1733 part_stat_inc(cpu
, part
, ios
[rw
]);
1734 part_stat_add(cpu
, part
, sectors
[rw
], sectors
);
1735 part_inc_in_flight(part
, rw
);
1739 EXPORT_SYMBOL(generic_start_io_acct
);
1741 void generic_end_io_acct(int rw
, struct hd_struct
*part
,
1742 unsigned long start_time
)
1744 unsigned long duration
= jiffies
- start_time
;
1745 int cpu
= part_stat_lock();
1747 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1748 part_round_stats(cpu
, part
);
1749 part_dec_in_flight(part
, rw
);
1753 EXPORT_SYMBOL(generic_end_io_acct
);
1755 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1756 void bio_flush_dcache_pages(struct bio
*bi
)
1758 struct bio_vec bvec
;
1759 struct bvec_iter iter
;
1761 bio_for_each_segment(bvec
, bi
, iter
)
1762 flush_dcache_page(bvec
.bv_page
);
1764 EXPORT_SYMBOL(bio_flush_dcache_pages
);
1767 static inline bool bio_remaining_done(struct bio
*bio
)
1770 * If we're not chaining, then ->__bi_remaining is always 1 and
1771 * we always end io on the first invocation.
1773 if (!bio_flagged(bio
, BIO_CHAIN
))
1776 BUG_ON(atomic_read(&bio
->__bi_remaining
) <= 0);
1778 if (atomic_dec_and_test(&bio
->__bi_remaining
)) {
1779 bio_clear_flag(bio
, BIO_CHAIN
);
1787 * bio_endio - end I/O on a bio
1791 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1792 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1793 * bio unless they own it and thus know that it has an end_io function.
1795 * bio_endio() can be called several times on a bio that has been chained
1796 * using bio_chain(). The ->bi_end_io() function will only be called the
1797 * last time. At this point the BLK_TA_COMPLETE tracing event will be
1798 * generated if BIO_TRACE_COMPLETION is set.
1800 void bio_endio(struct bio
*bio
)
1803 if (!bio_remaining_done(bio
))
1807 * Need to have a real endio function for chained bios, otherwise
1808 * various corner cases will break (like stacking block devices that
1809 * save/restore bi_end_io) - however, we want to avoid unbounded
1810 * recursion and blowing the stack. Tail call optimization would
1811 * handle this, but compiling with frame pointers also disables
1812 * gcc's sibling call optimization.
1814 if (bio
->bi_end_io
== bio_chain_endio
) {
1815 bio
= __bio_chain_endio(bio
);
1819 if (bio
->bi_bdev
&& bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
1820 trace_block_bio_complete(bdev_get_queue(bio
->bi_bdev
),
1821 bio
, bio
->bi_error
);
1822 bio_clear_flag(bio
, BIO_TRACE_COMPLETION
);
1825 blk_throtl_bio_endio(bio
);
1827 bio
->bi_end_io(bio
);
1829 EXPORT_SYMBOL(bio_endio
);
1832 * bio_split - split a bio
1833 * @bio: bio to split
1834 * @sectors: number of sectors to split from the front of @bio
1836 * @bs: bio set to allocate from
1838 * Allocates and returns a new bio which represents @sectors from the start of
1839 * @bio, and updates @bio to represent the remaining sectors.
1841 * Unless this is a discard request the newly allocated bio will point
1842 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1843 * @bio is not freed before the split.
1845 struct bio
*bio_split(struct bio
*bio
, int sectors
,
1846 gfp_t gfp
, struct bio_set
*bs
)
1848 struct bio
*split
= NULL
;
1850 BUG_ON(sectors
<= 0);
1851 BUG_ON(sectors
>= bio_sectors(bio
));
1853 split
= bio_clone_fast(bio
, gfp
, bs
);
1857 split
->bi_iter
.bi_size
= sectors
<< 9;
1859 if (bio_integrity(split
))
1860 bio_integrity_trim(split
, 0, sectors
);
1862 bio_advance(bio
, split
->bi_iter
.bi_size
);
1864 if (bio_flagged(bio
, BIO_TRACE_COMPLETION
))
1865 bio_set_flag(bio
, BIO_TRACE_COMPLETION
);
1869 EXPORT_SYMBOL(bio_split
);
1872 * bio_trim - trim a bio
1874 * @offset: number of sectors to trim from the front of @bio
1875 * @size: size we want to trim @bio to, in sectors
1877 void bio_trim(struct bio
*bio
, int offset
, int size
)
1879 /* 'bio' is a cloned bio which we need to trim to match
1880 * the given offset and size.
1884 if (offset
== 0 && size
== bio
->bi_iter
.bi_size
)
1887 bio_clear_flag(bio
, BIO_SEG_VALID
);
1889 bio_advance(bio
, offset
<< 9);
1891 bio
->bi_iter
.bi_size
= size
;
1893 EXPORT_SYMBOL_GPL(bio_trim
);
1896 * create memory pools for biovec's in a bio_set.
1897 * use the global biovec slabs created for general use.
1899 mempool_t
*biovec_create_pool(int pool_entries
)
1901 struct biovec_slab
*bp
= bvec_slabs
+ BVEC_POOL_MAX
;
1903 return mempool_create_slab_pool(pool_entries
, bp
->slab
);
1906 void bioset_free(struct bio_set
*bs
)
1908 if (bs
->rescue_workqueue
)
1909 destroy_workqueue(bs
->rescue_workqueue
);
1912 mempool_destroy(bs
->bio_pool
);
1915 mempool_destroy(bs
->bvec_pool
);
1917 bioset_integrity_free(bs
);
1922 EXPORT_SYMBOL(bioset_free
);
1924 static struct bio_set
*__bioset_create(unsigned int pool_size
,
1925 unsigned int front_pad
,
1926 bool create_bvec_pool
)
1928 unsigned int back_pad
= BIO_INLINE_VECS
* sizeof(struct bio_vec
);
1931 bs
= kzalloc(sizeof(*bs
), GFP_KERNEL
);
1935 bs
->front_pad
= front_pad
;
1937 spin_lock_init(&bs
->rescue_lock
);
1938 bio_list_init(&bs
->rescue_list
);
1939 INIT_WORK(&bs
->rescue_work
, bio_alloc_rescue
);
1941 bs
->bio_slab
= bio_find_or_create_slab(front_pad
+ back_pad
);
1942 if (!bs
->bio_slab
) {
1947 bs
->bio_pool
= mempool_create_slab_pool(pool_size
, bs
->bio_slab
);
1951 if (create_bvec_pool
) {
1952 bs
->bvec_pool
= biovec_create_pool(pool_size
);
1957 bs
->rescue_workqueue
= alloc_workqueue("bioset", WQ_MEM_RECLAIM
, 0);
1958 if (!bs
->rescue_workqueue
)
1968 * bioset_create - Create a bio_set
1969 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1970 * @front_pad: Number of bytes to allocate in front of the returned bio
1973 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1974 * to ask for a number of bytes to be allocated in front of the bio.
1975 * Front pad allocation is useful for embedding the bio inside
1976 * another structure, to avoid allocating extra data to go with the bio.
1977 * Note that the bio must be embedded at the END of that structure always,
1978 * or things will break badly.
1980 struct bio_set
*bioset_create(unsigned int pool_size
, unsigned int front_pad
)
1982 return __bioset_create(pool_size
, front_pad
, true);
1984 EXPORT_SYMBOL(bioset_create
);
1987 * bioset_create_nobvec - Create a bio_set without bio_vec mempool
1988 * @pool_size: Number of bio to cache in the mempool
1989 * @front_pad: Number of bytes to allocate in front of the returned bio
1992 * Same functionality as bioset_create() except that mempool is not
1993 * created for bio_vecs. Saving some memory for bio_clone_fast() users.
1995 struct bio_set
*bioset_create_nobvec(unsigned int pool_size
, unsigned int front_pad
)
1997 return __bioset_create(pool_size
, front_pad
, false);
1999 EXPORT_SYMBOL(bioset_create_nobvec
);
2001 #ifdef CONFIG_BLK_CGROUP
2004 * bio_associate_blkcg - associate a bio with the specified blkcg
2006 * @blkcg_css: css of the blkcg to associate
2008 * Associate @bio with the blkcg specified by @blkcg_css. Block layer will
2009 * treat @bio as if it were issued by a task which belongs to the blkcg.
2011 * This function takes an extra reference of @blkcg_css which will be put
2012 * when @bio is released. The caller must own @bio and is responsible for
2013 * synchronizing calls to this function.
2015 int bio_associate_blkcg(struct bio
*bio
, struct cgroup_subsys_state
*blkcg_css
)
2017 if (unlikely(bio
->bi_css
))
2020 bio
->bi_css
= blkcg_css
;
2023 EXPORT_SYMBOL_GPL(bio_associate_blkcg
);
2026 * bio_associate_current - associate a bio with %current
2029 * Associate @bio with %current if it hasn't been associated yet. Block
2030 * layer will treat @bio as if it were issued by %current no matter which
2031 * task actually issues it.
2033 * This function takes an extra reference of @task's io_context and blkcg
2034 * which will be put when @bio is released. The caller must own @bio,
2035 * ensure %current->io_context exists, and is responsible for synchronizing
2036 * calls to this function.
2038 int bio_associate_current(struct bio
*bio
)
2040 struct io_context
*ioc
;
2045 ioc
= current
->io_context
;
2049 get_io_context_active(ioc
);
2051 bio
->bi_css
= task_get_css(current
, io_cgrp_id
);
2054 EXPORT_SYMBOL_GPL(bio_associate_current
);
2057 * bio_disassociate_task - undo bio_associate_current()
2060 void bio_disassociate_task(struct bio
*bio
)
2063 put_io_context(bio
->bi_ioc
);
2067 css_put(bio
->bi_css
);
2073 * bio_clone_blkcg_association - clone blkcg association from src to dst bio
2074 * @dst: destination bio
2077 void bio_clone_blkcg_association(struct bio
*dst
, struct bio
*src
)
2080 WARN_ON(bio_associate_blkcg(dst
, src
->bi_css
));
2083 #endif /* CONFIG_BLK_CGROUP */
2085 static void __init
biovec_init_slabs(void)
2089 for (i
= 0; i
< BVEC_POOL_NR
; i
++) {
2091 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
2093 if (bvs
->nr_vecs
<= BIO_INLINE_VECS
) {
2098 size
= bvs
->nr_vecs
* sizeof(struct bio_vec
);
2099 bvs
->slab
= kmem_cache_create(bvs
->name
, size
, 0,
2100 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
2104 static int __init
init_bio(void)
2108 bio_slabs
= kzalloc(bio_slab_max
* sizeof(struct bio_slab
), GFP_KERNEL
);
2110 panic("bio: can't allocate bios\n");
2112 bio_integrity_init();
2113 biovec_init_slabs();
2115 fs_bio_set
= bioset_create(BIO_POOL_SIZE
, 0);
2117 panic("bio: can't allocate bios\n");
2119 if (bioset_integrity_create(fs_bio_set
, BIO_POOL_SIZE
))
2120 panic("bio: can't create integrity pool\n");
2124 subsys_initcall(init_bio
);