]>
git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - block/bio.c
2 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
13 * You should have received a copy of the GNU General Public Licens
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
19 #include <linux/swap.h>
20 #include <linux/bio.h>
21 #include <linux/blkdev.h>
22 #include <linux/uio.h>
23 #include <linux/iocontext.h>
24 #include <linux/slab.h>
25 #include <linux/init.h>
26 #include <linux/kernel.h>
27 #include <linux/export.h>
28 #include <linux/mempool.h>
29 #include <linux/workqueue.h>
30 #include <linux/cgroup.h>
32 #include <trace/events/block.h>
35 * Test patch to inline a certain number of bi_io_vec's inside the bio
36 * itself, to shrink a bio data allocation from two mempool calls to one
38 #define BIO_INLINE_VECS 4
41 * if you change this list, also change bvec_alloc or things will
42 * break badly! cannot be bigger than what you can fit into an
45 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
46 static struct biovec_slab bvec_slabs
[BVEC_POOL_NR
] __read_mostly
= {
47 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES
),
52 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
53 * IO code that does not need private memory pools.
55 struct bio_set
*fs_bio_set
;
56 EXPORT_SYMBOL(fs_bio_set
);
59 * Our slab pool management
62 struct kmem_cache
*slab
;
63 unsigned int slab_ref
;
64 unsigned int slab_size
;
67 static DEFINE_MUTEX(bio_slab_lock
);
68 static struct bio_slab
*bio_slabs
;
69 static unsigned int bio_slab_nr
, bio_slab_max
;
71 static struct kmem_cache
*bio_find_or_create_slab(unsigned int extra_size
)
73 unsigned int sz
= sizeof(struct bio
) + extra_size
;
74 struct kmem_cache
*slab
= NULL
;
75 struct bio_slab
*bslab
, *new_bio_slabs
;
76 unsigned int new_bio_slab_max
;
77 unsigned int i
, entry
= -1;
79 mutex_lock(&bio_slab_lock
);
82 while (i
< bio_slab_nr
) {
83 bslab
= &bio_slabs
[i
];
85 if (!bslab
->slab
&& entry
== -1)
87 else if (bslab
->slab_size
== sz
) {
98 if (bio_slab_nr
== bio_slab_max
&& entry
== -1) {
99 new_bio_slab_max
= bio_slab_max
<< 1;
100 new_bio_slabs
= krealloc(bio_slabs
,
101 new_bio_slab_max
* sizeof(struct bio_slab
),
105 bio_slab_max
= new_bio_slab_max
;
106 bio_slabs
= new_bio_slabs
;
109 entry
= bio_slab_nr
++;
111 bslab
= &bio_slabs
[entry
];
113 snprintf(bslab
->name
, sizeof(bslab
->name
), "bio-%d", entry
);
114 slab
= kmem_cache_create(bslab
->name
, sz
, ARCH_KMALLOC_MINALIGN
,
115 SLAB_HWCACHE_ALIGN
, NULL
);
121 bslab
->slab_size
= sz
;
123 mutex_unlock(&bio_slab_lock
);
127 static void bio_put_slab(struct bio_set
*bs
)
129 struct bio_slab
*bslab
= NULL
;
132 mutex_lock(&bio_slab_lock
);
134 for (i
= 0; i
< bio_slab_nr
; i
++) {
135 if (bs
->bio_slab
== bio_slabs
[i
].slab
) {
136 bslab
= &bio_slabs
[i
];
141 if (WARN(!bslab
, KERN_ERR
"bio: unable to find slab!\n"))
144 WARN_ON(!bslab
->slab_ref
);
146 if (--bslab
->slab_ref
)
149 kmem_cache_destroy(bslab
->slab
);
153 mutex_unlock(&bio_slab_lock
);
156 unsigned int bvec_nr_vecs(unsigned short idx
)
158 return bvec_slabs
[idx
].nr_vecs
;
161 void bvec_free(mempool_t
*pool
, struct bio_vec
*bv
, unsigned int idx
)
167 BIO_BUG_ON(idx
>= BVEC_POOL_NR
);
169 if (idx
== BVEC_POOL_MAX
) {
170 mempool_free(bv
, pool
);
172 struct biovec_slab
*bvs
= bvec_slabs
+ idx
;
174 kmem_cache_free(bvs
->slab
, bv
);
178 struct bio_vec
*bvec_alloc(gfp_t gfp_mask
, int nr
, unsigned long *idx
,
184 * see comment near bvec_array define!
202 case 129 ... BIO_MAX_PAGES
:
210 * idx now points to the pool we want to allocate from. only the
211 * 1-vec entry pool is mempool backed.
213 if (*idx
== BVEC_POOL_MAX
) {
215 bvl
= mempool_alloc(pool
, gfp_mask
);
217 struct biovec_slab
*bvs
= bvec_slabs
+ *idx
;
218 gfp_t __gfp_mask
= gfp_mask
& ~(__GFP_DIRECT_RECLAIM
| __GFP_IO
);
221 * Make this allocation restricted and don't dump info on
222 * allocation failures, since we'll fallback to the mempool
223 * in case of failure.
225 __gfp_mask
|= __GFP_NOMEMALLOC
| __GFP_NORETRY
| __GFP_NOWARN
;
228 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
229 * is set, retry with the 1-entry mempool
231 bvl
= kmem_cache_alloc(bvs
->slab
, __gfp_mask
);
232 if (unlikely(!bvl
&& (gfp_mask
& __GFP_DIRECT_RECLAIM
))) {
233 *idx
= BVEC_POOL_MAX
;
242 static void __bio_free(struct bio
*bio
)
244 bio_disassociate_task(bio
);
246 if (bio_integrity(bio
))
247 bio_integrity_free(bio
);
250 static void bio_free(struct bio
*bio
)
252 struct bio_set
*bs
= bio
->bi_pool
;
258 bvec_free(bs
->bvec_pool
, bio
->bi_io_vec
, BVEC_POOL_IDX(bio
));
261 * If we have front padding, adjust the bio pointer before freeing
266 mempool_free(p
, bs
->bio_pool
);
268 /* Bio was allocated by bio_kmalloc() */
273 void bio_init(struct bio
*bio
, struct bio_vec
*table
,
274 unsigned short max_vecs
)
276 memset(bio
, 0, sizeof(*bio
));
277 atomic_set(&bio
->__bi_remaining
, 1);
278 atomic_set(&bio
->__bi_cnt
, 1);
280 bio
->bi_io_vec
= table
;
281 bio
->bi_max_vecs
= max_vecs
;
283 EXPORT_SYMBOL(bio_init
);
286 * bio_reset - reinitialize a bio
290 * After calling bio_reset(), @bio will be in the same state as a freshly
291 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
292 * preserved are the ones that are initialized by bio_alloc_bioset(). See
293 * comment in struct bio.
295 void bio_reset(struct bio
*bio
)
297 unsigned long flags
= bio
->bi_flags
& (~0UL << BIO_RESET_BITS
);
301 memset(bio
, 0, BIO_RESET_BYTES
);
302 bio
->bi_flags
= flags
;
303 atomic_set(&bio
->__bi_remaining
, 1);
305 EXPORT_SYMBOL(bio_reset
);
307 static struct bio
*__bio_chain_endio(struct bio
*bio
)
309 struct bio
*parent
= bio
->bi_private
;
311 if (!parent
->bi_error
)
312 parent
->bi_error
= bio
->bi_error
;
317 static void bio_chain_endio(struct bio
*bio
)
319 bio_endio(__bio_chain_endio(bio
));
323 * bio_chain - chain bio completions
324 * @bio: the target bio
325 * @parent: the @bio's parent bio
327 * The caller won't have a bi_end_io called when @bio completes - instead,
328 * @parent's bi_end_io won't be called until both @parent and @bio have
329 * completed; the chained bio will also be freed when it completes.
331 * The caller must not set bi_private or bi_end_io in @bio.
333 void bio_chain(struct bio
*bio
, struct bio
*parent
)
335 BUG_ON(bio
->bi_private
|| bio
->bi_end_io
);
337 bio
->bi_private
= parent
;
338 bio
->bi_end_io
= bio_chain_endio
;
339 bio_inc_remaining(parent
);
341 EXPORT_SYMBOL(bio_chain
);
343 static void bio_alloc_rescue(struct work_struct
*work
)
345 struct bio_set
*bs
= container_of(work
, struct bio_set
, rescue_work
);
349 spin_lock(&bs
->rescue_lock
);
350 bio
= bio_list_pop(&bs
->rescue_list
);
351 spin_unlock(&bs
->rescue_lock
);
356 generic_make_request(bio
);
360 static void punt_bios_to_rescuer(struct bio_set
*bs
)
362 struct bio_list punt
, nopunt
;
366 * In order to guarantee forward progress we must punt only bios that
367 * were allocated from this bio_set; otherwise, if there was a bio on
368 * there for a stacking driver higher up in the stack, processing it
369 * could require allocating bios from this bio_set, and doing that from
370 * our own rescuer would be bad.
372 * Since bio lists are singly linked, pop them all instead of trying to
373 * remove from the middle of the list:
376 bio_list_init(&punt
);
377 bio_list_init(&nopunt
);
379 while ((bio
= bio_list_pop(¤t
->bio_list
[0])))
380 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
381 current
->bio_list
[0] = nopunt
;
383 bio_list_init(&nopunt
);
384 while ((bio
= bio_list_pop(¤t
->bio_list
[1])))
385 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
386 current
->bio_list
[1] = nopunt
;
388 spin_lock(&bs
->rescue_lock
);
389 bio_list_merge(&bs
->rescue_list
, &punt
);
390 spin_unlock(&bs
->rescue_lock
);
392 queue_work(bs
->rescue_workqueue
, &bs
->rescue_work
);
396 * bio_alloc_bioset - allocate a bio for I/O
397 * @gfp_mask: the GFP_ mask given to the slab allocator
398 * @nr_iovecs: number of iovecs to pre-allocate
399 * @bs: the bio_set to allocate from.
402 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
403 * backed by the @bs's mempool.
405 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
406 * always be able to allocate a bio. This is due to the mempool guarantees.
407 * To make this work, callers must never allocate more than 1 bio at a time
408 * from this pool. Callers that need to allocate more than 1 bio must always
409 * submit the previously allocated bio for IO before attempting to allocate
410 * a new one. Failure to do so can cause deadlocks under memory pressure.
412 * Note that when running under generic_make_request() (i.e. any block
413 * driver), bios are not submitted until after you return - see the code in
414 * generic_make_request() that converts recursion into iteration, to prevent
417 * This would normally mean allocating multiple bios under
418 * generic_make_request() would be susceptible to deadlocks, but we have
419 * deadlock avoidance code that resubmits any blocked bios from a rescuer
422 * However, we do not guarantee forward progress for allocations from other
423 * mempools. Doing multiple allocations from the same mempool under
424 * generic_make_request() should be avoided - instead, use bio_set's front_pad
425 * for per bio allocations.
428 * Pointer to new bio on success, NULL on failure.
430 struct bio
*bio_alloc_bioset(gfp_t gfp_mask
, unsigned int nr_iovecs
,
433 gfp_t saved_gfp
= gfp_mask
;
435 unsigned inline_vecs
;
436 struct bio_vec
*bvl
= NULL
;
441 if (nr_iovecs
> UIO_MAXIOV
)
444 p
= kmalloc(sizeof(struct bio
) +
445 nr_iovecs
* sizeof(struct bio_vec
),
448 inline_vecs
= nr_iovecs
;
450 /* should not use nobvec bioset for nr_iovecs > 0 */
451 if (WARN_ON_ONCE(!bs
->bvec_pool
&& nr_iovecs
> 0))
454 * generic_make_request() converts recursion to iteration; this
455 * means if we're running beneath it, any bios we allocate and
456 * submit will not be submitted (and thus freed) until after we
459 * This exposes us to a potential deadlock if we allocate
460 * multiple bios from the same bio_set() while running
461 * underneath generic_make_request(). If we were to allocate
462 * multiple bios (say a stacking block driver that was splitting
463 * bios), we would deadlock if we exhausted the mempool's
466 * We solve this, and guarantee forward progress, with a rescuer
467 * workqueue per bio_set. If we go to allocate and there are
468 * bios on current->bio_list, we first try the allocation
469 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
470 * bios we would be blocking to the rescuer workqueue before
471 * we retry with the original gfp_flags.
474 if (current
->bio_list
&&
475 (!bio_list_empty(¤t
->bio_list
[0]) ||
476 !bio_list_empty(¤t
->bio_list
[1])))
477 gfp_mask
&= ~__GFP_DIRECT_RECLAIM
;
479 p
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
480 if (!p
&& gfp_mask
!= saved_gfp
) {
481 punt_bios_to_rescuer(bs
);
482 gfp_mask
= saved_gfp
;
483 p
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
486 front_pad
= bs
->front_pad
;
487 inline_vecs
= BIO_INLINE_VECS
;
494 bio_init(bio
, NULL
, 0);
496 if (nr_iovecs
> inline_vecs
) {
497 unsigned long idx
= 0;
499 bvl
= bvec_alloc(gfp_mask
, nr_iovecs
, &idx
, bs
->bvec_pool
);
500 if (!bvl
&& gfp_mask
!= saved_gfp
) {
501 punt_bios_to_rescuer(bs
);
502 gfp_mask
= saved_gfp
;
503 bvl
= bvec_alloc(gfp_mask
, nr_iovecs
, &idx
, bs
->bvec_pool
);
509 bio
->bi_flags
|= idx
<< BVEC_POOL_OFFSET
;
510 } else if (nr_iovecs
) {
511 bvl
= bio
->bi_inline_vecs
;
515 bio
->bi_max_vecs
= nr_iovecs
;
516 bio
->bi_io_vec
= bvl
;
520 mempool_free(p
, bs
->bio_pool
);
523 EXPORT_SYMBOL(bio_alloc_bioset
);
525 void zero_fill_bio(struct bio
*bio
)
529 struct bvec_iter iter
;
531 bio_for_each_segment(bv
, bio
, iter
) {
532 char *data
= bvec_kmap_irq(&bv
, &flags
);
533 memset(data
, 0, bv
.bv_len
);
534 flush_dcache_page(bv
.bv_page
);
535 bvec_kunmap_irq(data
, &flags
);
538 EXPORT_SYMBOL(zero_fill_bio
);
541 * bio_put - release a reference to a bio
542 * @bio: bio to release reference to
545 * Put a reference to a &struct bio, either one you have gotten with
546 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
548 void bio_put(struct bio
*bio
)
550 if (!bio_flagged(bio
, BIO_REFFED
))
553 BIO_BUG_ON(!atomic_read(&bio
->__bi_cnt
));
558 if (atomic_dec_and_test(&bio
->__bi_cnt
))
562 EXPORT_SYMBOL(bio_put
);
564 inline int bio_phys_segments(struct request_queue
*q
, struct bio
*bio
)
566 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
567 blk_recount_segments(q
, bio
);
569 return bio
->bi_phys_segments
;
571 EXPORT_SYMBOL(bio_phys_segments
);
574 * __bio_clone_fast - clone a bio that shares the original bio's biovec
575 * @bio: destination bio
576 * @bio_src: bio to clone
578 * Clone a &bio. Caller will own the returned bio, but not
579 * the actual data it points to. Reference count of returned
582 * Caller must ensure that @bio_src is not freed before @bio.
584 void __bio_clone_fast(struct bio
*bio
, struct bio
*bio_src
)
586 BUG_ON(bio
->bi_pool
&& BVEC_POOL_IDX(bio
));
589 * most users will be overriding ->bi_bdev with a new target,
590 * so we don't set nor calculate new physical/hw segment counts here
592 bio
->bi_bdev
= bio_src
->bi_bdev
;
593 bio_set_flag(bio
, BIO_CLONED
);
594 bio
->bi_opf
= bio_src
->bi_opf
;
595 bio
->bi_iter
= bio_src
->bi_iter
;
596 bio
->bi_io_vec
= bio_src
->bi_io_vec
;
598 bio_clone_blkcg_association(bio
, bio_src
);
600 EXPORT_SYMBOL(__bio_clone_fast
);
603 * bio_clone_fast - clone a bio that shares the original bio's biovec
605 * @gfp_mask: allocation priority
606 * @bs: bio_set to allocate from
608 * Like __bio_clone_fast, only also allocates the returned bio
610 struct bio
*bio_clone_fast(struct bio
*bio
, gfp_t gfp_mask
, struct bio_set
*bs
)
614 b
= bio_alloc_bioset(gfp_mask
, 0, bs
);
618 __bio_clone_fast(b
, bio
);
620 if (bio_integrity(bio
)) {
623 ret
= bio_integrity_clone(b
, bio
, gfp_mask
);
633 EXPORT_SYMBOL(bio_clone_fast
);
635 static struct bio
*__bio_clone_bioset(struct bio
*bio_src
, gfp_t gfp_mask
,
636 struct bio_set
*bs
, int offset
,
639 struct bvec_iter iter
;
642 struct bvec_iter iter_src
= bio_src
->bi_iter
;
644 /* for supporting partial clone */
645 if (offset
|| size
!= bio_src
->bi_iter
.bi_size
) {
646 bio_advance_iter(bio_src
, &iter_src
, offset
);
647 iter_src
.bi_size
= size
;
651 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
652 * bio_src->bi_io_vec to bio->bi_io_vec.
654 * We can't do that anymore, because:
656 * - The point of cloning the biovec is to produce a bio with a biovec
657 * the caller can modify: bi_idx and bi_bvec_done should be 0.
659 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
660 * we tried to clone the whole thing bio_alloc_bioset() would fail.
661 * But the clone should succeed as long as the number of biovecs we
662 * actually need to allocate is fewer than BIO_MAX_PAGES.
664 * - Lastly, bi_vcnt should not be looked at or relied upon by code
665 * that does not own the bio - reason being drivers don't use it for
666 * iterating over the biovec anymore, so expecting it to be kept up
667 * to date (i.e. for clones that share the parent biovec) is just
668 * asking for trouble and would force extra work on
669 * __bio_clone_fast() anyways.
672 bio
= bio_alloc_bioset(gfp_mask
, __bio_segments(bio_src
,
676 bio
->bi_bdev
= bio_src
->bi_bdev
;
677 bio
->bi_opf
= bio_src
->bi_opf
;
678 bio
->bi_iter
.bi_sector
= bio_src
->bi_iter
.bi_sector
;
679 bio
->bi_iter
.bi_size
= bio_src
->bi_iter
.bi_size
;
681 switch (bio_op(bio
)) {
683 case REQ_OP_SECURE_ERASE
:
684 case REQ_OP_WRITE_ZEROES
:
686 case REQ_OP_WRITE_SAME
:
687 bio
->bi_io_vec
[bio
->bi_vcnt
++] = bio_src
->bi_io_vec
[0];
690 __bio_for_each_segment(bv
, bio_src
, iter
, iter_src
)
691 bio
->bi_io_vec
[bio
->bi_vcnt
++] = bv
;
695 if (bio_integrity(bio_src
)) {
698 ret
= bio_integrity_clone(bio
, bio_src
, gfp_mask
);
705 bio_clone_blkcg_association(bio
, bio_src
);
711 * bio_clone_bioset - clone a bio
712 * @bio_src: bio to clone
713 * @gfp_mask: allocation priority
714 * @bs: bio_set to allocate from
716 * Clone bio. Caller will own the returned bio, but not the actual data it
717 * points to. Reference count of returned bio will be one.
719 struct bio
*bio_clone_bioset(struct bio
*bio_src
, gfp_t gfp_mask
,
722 return __bio_clone_bioset(bio_src
, gfp_mask
, bs
, 0,
723 bio_src
->bi_iter
.bi_size
);
725 EXPORT_SYMBOL(bio_clone_bioset
);
728 * bio_clone_bioset_partial - clone a partial bio
729 * @bio_src: bio to clone
730 * @gfp_mask: allocation priority
731 * @bs: bio_set to allocate from
732 * @offset: cloned starting from the offset
733 * @size: size for the cloned bio
735 * Clone bio. Caller will own the returned bio, but not the actual data it
736 * points to. Reference count of returned bio will be one.
738 struct bio
*bio_clone_bioset_partial(struct bio
*bio_src
, gfp_t gfp_mask
,
739 struct bio_set
*bs
, int offset
,
742 return __bio_clone_bioset(bio_src
, gfp_mask
, bs
, offset
, size
);
744 EXPORT_SYMBOL(bio_clone_bioset_partial
);
747 * bio_add_pc_page - attempt to add page to bio
748 * @q: the target queue
749 * @bio: destination bio
751 * @len: vec entry length
752 * @offset: vec entry offset
754 * Attempt to add a page to the bio_vec maplist. This can fail for a
755 * number of reasons, such as the bio being full or target block device
756 * limitations. The target block device must allow bio's up to PAGE_SIZE,
757 * so it is always possible to add a single page to an empty bio.
759 * This should only be used by REQ_PC bios.
761 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
, struct page
762 *page
, unsigned int len
, unsigned int offset
)
764 int retried_segments
= 0;
765 struct bio_vec
*bvec
;
768 * cloned bio must not modify vec list
770 if (unlikely(bio_flagged(bio
, BIO_CLONED
)))
773 if (((bio
->bi_iter
.bi_size
+ len
) >> 9) > queue_max_hw_sectors(q
))
777 * For filesystems with a blocksize smaller than the pagesize
778 * we will often be called with the same page as last time and
779 * a consecutive offset. Optimize this special case.
781 if (bio
->bi_vcnt
> 0) {
782 struct bio_vec
*prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
784 if (page
== prev
->bv_page
&&
785 offset
== prev
->bv_offset
+ prev
->bv_len
) {
787 bio
->bi_iter
.bi_size
+= len
;
792 * If the queue doesn't support SG gaps and adding this
793 * offset would create a gap, disallow it.
795 if (bvec_gap_to_prev(q
, prev
, offset
))
799 if (bio
->bi_vcnt
>= bio
->bi_max_vecs
)
803 * setup the new entry, we might clear it again later if we
804 * cannot add the page
806 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
807 bvec
->bv_page
= page
;
809 bvec
->bv_offset
= offset
;
811 bio
->bi_phys_segments
++;
812 bio
->bi_iter
.bi_size
+= len
;
815 * Perform a recount if the number of segments is greater
816 * than queue_max_segments(q).
819 while (bio
->bi_phys_segments
> queue_max_segments(q
)) {
821 if (retried_segments
)
824 retried_segments
= 1;
825 blk_recount_segments(q
, bio
);
828 /* If we may be able to merge these biovecs, force a recount */
829 if (bio
->bi_vcnt
> 1 && (BIOVEC_PHYS_MERGEABLE(bvec
-1, bvec
)))
830 bio_clear_flag(bio
, BIO_SEG_VALID
);
836 bvec
->bv_page
= NULL
;
840 bio
->bi_iter
.bi_size
-= len
;
841 blk_recount_segments(q
, bio
);
844 EXPORT_SYMBOL(bio_add_pc_page
);
847 * bio_add_page - attempt to add page to bio
848 * @bio: destination bio
850 * @len: vec entry length
851 * @offset: vec entry offset
853 * Attempt to add a page to the bio_vec maplist. This will only fail
854 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
856 int bio_add_page(struct bio
*bio
, struct page
*page
,
857 unsigned int len
, unsigned int offset
)
862 * cloned bio must not modify vec list
864 if (WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
)))
868 * For filesystems with a blocksize smaller than the pagesize
869 * we will often be called with the same page as last time and
870 * a consecutive offset. Optimize this special case.
872 if (bio
->bi_vcnt
> 0) {
873 bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
875 if (page
== bv
->bv_page
&&
876 offset
== bv
->bv_offset
+ bv
->bv_len
) {
882 if (bio
->bi_vcnt
>= bio
->bi_max_vecs
)
885 bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
888 bv
->bv_offset
= offset
;
892 bio
->bi_iter
.bi_size
+= len
;
895 EXPORT_SYMBOL(bio_add_page
);
898 * bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
899 * @bio: bio to add pages to
900 * @iter: iov iterator describing the region to be mapped
902 * Pins as many pages from *iter and appends them to @bio's bvec array. The
903 * pages will have to be released using put_page() when done.
905 int bio_iov_iter_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
907 unsigned short nr_pages
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
908 struct bio_vec
*bv
= bio
->bi_io_vec
+ bio
->bi_vcnt
;
909 struct page
**pages
= (struct page
**)bv
;
913 size
= iov_iter_get_pages(iter
, pages
, LONG_MAX
, nr_pages
, &offset
);
914 if (unlikely(size
<= 0))
915 return size
? size
: -EFAULT
;
916 nr_pages
= (size
+ offset
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
919 * Deep magic below: We need to walk the pinned pages backwards
920 * because we are abusing the space allocated for the bio_vecs
921 * for the page array. Because the bio_vecs are larger than the
922 * page pointers by definition this will always work. But it also
923 * means we can't use bio_add_page, so any changes to it's semantics
924 * need to be reflected here as well.
926 bio
->bi_iter
.bi_size
+= size
;
927 bio
->bi_vcnt
+= nr_pages
;
929 diff
= (nr_pages
* PAGE_SIZE
- offset
) - size
;
931 bv
[nr_pages
].bv_page
= pages
[nr_pages
];
932 bv
[nr_pages
].bv_len
= PAGE_SIZE
;
933 bv
[nr_pages
].bv_offset
= 0;
936 bv
[0].bv_offset
+= offset
;
937 bv
[0].bv_len
-= offset
;
939 bv
[bio
->bi_vcnt
- 1].bv_len
-= diff
;
941 iov_iter_advance(iter
, size
);
944 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages
);
946 struct submit_bio_ret
{
947 struct completion event
;
951 static void submit_bio_wait_endio(struct bio
*bio
)
953 struct submit_bio_ret
*ret
= bio
->bi_private
;
955 ret
->error
= bio
->bi_error
;
956 complete(&ret
->event
);
960 * submit_bio_wait - submit a bio, and wait until it completes
961 * @bio: The &struct bio which describes the I/O
963 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
964 * bio_endio() on failure.
966 int submit_bio_wait(struct bio
*bio
)
968 struct submit_bio_ret ret
;
970 init_completion(&ret
.event
);
971 bio
->bi_private
= &ret
;
972 bio
->bi_end_io
= submit_bio_wait_endio
;
973 bio
->bi_opf
|= REQ_SYNC
;
975 wait_for_completion_io(&ret
.event
);
979 EXPORT_SYMBOL(submit_bio_wait
);
982 * bio_advance - increment/complete a bio by some number of bytes
983 * @bio: bio to advance
984 * @bytes: number of bytes to complete
986 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
987 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
988 * be updated on the last bvec as well.
990 * @bio will then represent the remaining, uncompleted portion of the io.
992 void bio_advance(struct bio
*bio
, unsigned bytes
)
994 if (bio_integrity(bio
))
995 bio_integrity_advance(bio
, bytes
);
997 bio_advance_iter(bio
, &bio
->bi_iter
, bytes
);
999 EXPORT_SYMBOL(bio_advance
);
1002 * bio_alloc_pages - allocates a single page for each bvec in a bio
1003 * @bio: bio to allocate pages for
1004 * @gfp_mask: flags for allocation
1006 * Allocates pages up to @bio->bi_vcnt.
1008 * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
1011 int bio_alloc_pages(struct bio
*bio
, gfp_t gfp_mask
)
1016 bio_for_each_segment_all(bv
, bio
, i
) {
1017 bv
->bv_page
= alloc_page(gfp_mask
);
1019 while (--bv
>= bio
->bi_io_vec
)
1020 __free_page(bv
->bv_page
);
1027 EXPORT_SYMBOL(bio_alloc_pages
);
1030 * bio_copy_data - copy contents of data buffers from one chain of bios to
1032 * @src: source bio list
1033 * @dst: destination bio list
1035 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
1036 * @src and @dst as linked lists of bios.
1038 * Stops when it reaches the end of either @src or @dst - that is, copies
1039 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1041 void bio_copy_data(struct bio
*dst
, struct bio
*src
)
1043 struct bvec_iter src_iter
, dst_iter
;
1044 struct bio_vec src_bv
, dst_bv
;
1045 void *src_p
, *dst_p
;
1048 src_iter
= src
->bi_iter
;
1049 dst_iter
= dst
->bi_iter
;
1052 if (!src_iter
.bi_size
) {
1057 src_iter
= src
->bi_iter
;
1060 if (!dst_iter
.bi_size
) {
1065 dst_iter
= dst
->bi_iter
;
1068 src_bv
= bio_iter_iovec(src
, src_iter
);
1069 dst_bv
= bio_iter_iovec(dst
, dst_iter
);
1071 bytes
= min(src_bv
.bv_len
, dst_bv
.bv_len
);
1073 src_p
= kmap_atomic(src_bv
.bv_page
);
1074 dst_p
= kmap_atomic(dst_bv
.bv_page
);
1076 memcpy(dst_p
+ dst_bv
.bv_offset
,
1077 src_p
+ src_bv
.bv_offset
,
1080 kunmap_atomic(dst_p
);
1081 kunmap_atomic(src_p
);
1083 bio_advance_iter(src
, &src_iter
, bytes
);
1084 bio_advance_iter(dst
, &dst_iter
, bytes
);
1087 EXPORT_SYMBOL(bio_copy_data
);
1089 struct bio_map_data
{
1091 struct iov_iter iter
;
1095 static struct bio_map_data
*bio_alloc_map_data(unsigned int iov_count
,
1098 if (iov_count
> UIO_MAXIOV
)
1101 return kmalloc(sizeof(struct bio_map_data
) +
1102 sizeof(struct iovec
) * iov_count
, gfp_mask
);
1106 * bio_copy_from_iter - copy all pages from iov_iter to bio
1107 * @bio: The &struct bio which describes the I/O as destination
1108 * @iter: iov_iter as source
1110 * Copy all pages from iov_iter to bio.
1111 * Returns 0 on success, or error on failure.
1113 static int bio_copy_from_iter(struct bio
*bio
, struct iov_iter iter
)
1116 struct bio_vec
*bvec
;
1118 bio_for_each_segment_all(bvec
, bio
, i
) {
1121 ret
= copy_page_from_iter(bvec
->bv_page
,
1126 if (!iov_iter_count(&iter
))
1129 if (ret
< bvec
->bv_len
)
1137 * bio_copy_to_iter - copy all pages from bio to iov_iter
1138 * @bio: The &struct bio which describes the I/O as source
1139 * @iter: iov_iter as destination
1141 * Copy all pages from bio to iov_iter.
1142 * Returns 0 on success, or error on failure.
1144 static int bio_copy_to_iter(struct bio
*bio
, struct iov_iter iter
)
1147 struct bio_vec
*bvec
;
1149 bio_for_each_segment_all(bvec
, bio
, i
) {
1152 ret
= copy_page_to_iter(bvec
->bv_page
,
1157 if (!iov_iter_count(&iter
))
1160 if (ret
< bvec
->bv_len
)
1167 void bio_free_pages(struct bio
*bio
)
1169 struct bio_vec
*bvec
;
1172 bio_for_each_segment_all(bvec
, bio
, i
)
1173 __free_page(bvec
->bv_page
);
1175 EXPORT_SYMBOL(bio_free_pages
);
1178 * bio_uncopy_user - finish previously mapped bio
1179 * @bio: bio being terminated
1181 * Free pages allocated from bio_copy_user_iov() and write back data
1182 * to user space in case of a read.
1184 int bio_uncopy_user(struct bio
*bio
)
1186 struct bio_map_data
*bmd
= bio
->bi_private
;
1189 if (!bio_flagged(bio
, BIO_NULL_MAPPED
)) {
1191 * if we're in a workqueue, the request is orphaned, so
1192 * don't copy into a random user address space, just free
1193 * and return -EINTR so user space doesn't expect any data.
1197 else if (bio_data_dir(bio
) == READ
)
1198 ret
= bio_copy_to_iter(bio
, bmd
->iter
);
1199 if (bmd
->is_our_pages
)
1200 bio_free_pages(bio
);
1208 * bio_copy_user_iov - copy user data to bio
1209 * @q: destination block queue
1210 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1211 * @iter: iovec iterator
1212 * @gfp_mask: memory allocation flags
1214 * Prepares and returns a bio for indirect user io, bouncing data
1215 * to/from kernel pages as necessary. Must be paired with
1216 * call bio_uncopy_user() on io completion.
1218 struct bio
*bio_copy_user_iov(struct request_queue
*q
,
1219 struct rq_map_data
*map_data
,
1220 const struct iov_iter
*iter
,
1223 struct bio_map_data
*bmd
;
1228 unsigned int len
= iter
->count
;
1229 unsigned int offset
= map_data
? offset_in_page(map_data
->offset
) : 0;
1231 for (i
= 0; i
< iter
->nr_segs
; i
++) {
1232 unsigned long uaddr
;
1234 unsigned long start
;
1236 uaddr
= (unsigned long) iter
->iov
[i
].iov_base
;
1237 end
= (uaddr
+ iter
->iov
[i
].iov_len
+ PAGE_SIZE
- 1)
1239 start
= uaddr
>> PAGE_SHIFT
;
1245 return ERR_PTR(-EINVAL
);
1247 nr_pages
+= end
- start
;
1253 bmd
= bio_alloc_map_data(iter
->nr_segs
, gfp_mask
);
1255 return ERR_PTR(-ENOMEM
);
1258 * We need to do a deep copy of the iov_iter including the iovecs.
1259 * The caller provided iov might point to an on-stack or otherwise
1262 bmd
->is_our_pages
= map_data
? 0 : 1;
1263 memcpy(bmd
->iov
, iter
->iov
, sizeof(struct iovec
) * iter
->nr_segs
);
1264 iov_iter_init(&bmd
->iter
, iter
->type
, bmd
->iov
,
1265 iter
->nr_segs
, iter
->count
);
1268 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1275 nr_pages
= 1 << map_data
->page_order
;
1276 i
= map_data
->offset
/ PAGE_SIZE
;
1279 unsigned int bytes
= PAGE_SIZE
;
1287 if (i
== map_data
->nr_entries
* nr_pages
) {
1292 page
= map_data
->pages
[i
/ nr_pages
];
1293 page
+= (i
% nr_pages
);
1297 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
1304 if (bio_add_pc_page(q
, bio
, page
, bytes
, offset
) < bytes
)
1317 if (((iter
->type
& WRITE
) && (!map_data
|| !map_data
->null_mapped
)) ||
1318 (map_data
&& map_data
->from_user
)) {
1319 ret
= bio_copy_from_iter(bio
, *iter
);
1324 bio
->bi_private
= bmd
;
1328 bio_free_pages(bio
);
1332 return ERR_PTR(ret
);
1336 * bio_map_user_iov - map user iovec into bio
1337 * @q: the struct request_queue for the bio
1338 * @iter: iovec iterator
1339 * @gfp_mask: memory allocation flags
1341 * Map the user space address into a bio suitable for io to a block
1342 * device. Returns an error pointer in case of error.
1344 struct bio
*bio_map_user_iov(struct request_queue
*q
,
1345 const struct iov_iter
*iter
,
1350 struct page
**pages
;
1357 iov_for_each(iov
, i
, *iter
) {
1358 unsigned long uaddr
= (unsigned long) iov
.iov_base
;
1359 unsigned long len
= iov
.iov_len
;
1360 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1361 unsigned long start
= uaddr
>> PAGE_SHIFT
;
1367 return ERR_PTR(-EINVAL
);
1369 nr_pages
+= end
- start
;
1371 * buffer must be aligned to at least logical block size for now
1373 if (uaddr
& queue_dma_alignment(q
))
1374 return ERR_PTR(-EINVAL
);
1378 return ERR_PTR(-EINVAL
);
1380 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1382 return ERR_PTR(-ENOMEM
);
1385 pages
= kcalloc(nr_pages
, sizeof(struct page
*), gfp_mask
);
1389 iov_for_each(iov
, i
, *iter
) {
1390 unsigned long uaddr
= (unsigned long) iov
.iov_base
;
1391 unsigned long len
= iov
.iov_len
;
1392 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1393 unsigned long start
= uaddr
>> PAGE_SHIFT
;
1394 const int local_nr_pages
= end
- start
;
1395 const int page_limit
= cur_page
+ local_nr_pages
;
1397 ret
= get_user_pages_fast(uaddr
, local_nr_pages
,
1398 (iter
->type
& WRITE
) != WRITE
,
1400 if (ret
< local_nr_pages
) {
1405 offset
= offset_in_page(uaddr
);
1406 for (j
= cur_page
; j
< page_limit
; j
++) {
1407 unsigned int bytes
= PAGE_SIZE
- offset
;
1418 if (bio_add_pc_page(q
, bio
, pages
[j
], bytes
, offset
) <
1428 * release the pages we didn't map into the bio, if any
1430 while (j
< page_limit
)
1431 put_page(pages
[j
++]);
1436 bio_set_flag(bio
, BIO_USER_MAPPED
);
1439 * subtle -- if bio_map_user_iov() ended up bouncing a bio,
1440 * it would normally disappear when its bi_end_io is run.
1441 * however, we need it for the unmap, so grab an extra
1448 for (j
= 0; j
< nr_pages
; j
++) {
1456 return ERR_PTR(ret
);
1459 static void __bio_unmap_user(struct bio
*bio
)
1461 struct bio_vec
*bvec
;
1465 * make sure we dirty pages we wrote to
1467 bio_for_each_segment_all(bvec
, bio
, i
) {
1468 if (bio_data_dir(bio
) == READ
)
1469 set_page_dirty_lock(bvec
->bv_page
);
1471 put_page(bvec
->bv_page
);
1478 * bio_unmap_user - unmap a bio
1479 * @bio: the bio being unmapped
1481 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from
1484 * bio_unmap_user() may sleep.
1486 void bio_unmap_user(struct bio
*bio
)
1488 __bio_unmap_user(bio
);
1492 static void bio_map_kern_endio(struct bio
*bio
)
1498 * bio_map_kern - map kernel address into bio
1499 * @q: the struct request_queue for the bio
1500 * @data: pointer to buffer to map
1501 * @len: length in bytes
1502 * @gfp_mask: allocation flags for bio allocation
1504 * Map the kernel address into a bio suitable for io to a block
1505 * device. Returns an error pointer in case of error.
1507 struct bio
*bio_map_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1510 unsigned long kaddr
= (unsigned long)data
;
1511 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1512 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1513 const int nr_pages
= end
- start
;
1517 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1519 return ERR_PTR(-ENOMEM
);
1521 offset
= offset_in_page(kaddr
);
1522 for (i
= 0; i
< nr_pages
; i
++) {
1523 unsigned int bytes
= PAGE_SIZE
- offset
;
1531 if (bio_add_pc_page(q
, bio
, virt_to_page(data
), bytes
,
1533 /* we don't support partial mappings */
1535 return ERR_PTR(-EINVAL
);
1543 bio
->bi_end_io
= bio_map_kern_endio
;
1546 EXPORT_SYMBOL(bio_map_kern
);
1548 static void bio_copy_kern_endio(struct bio
*bio
)
1550 bio_free_pages(bio
);
1554 static void bio_copy_kern_endio_read(struct bio
*bio
)
1556 char *p
= bio
->bi_private
;
1557 struct bio_vec
*bvec
;
1560 bio_for_each_segment_all(bvec
, bio
, i
) {
1561 memcpy(p
, page_address(bvec
->bv_page
), bvec
->bv_len
);
1565 bio_copy_kern_endio(bio
);
1569 * bio_copy_kern - copy kernel address into bio
1570 * @q: the struct request_queue for the bio
1571 * @data: pointer to buffer to copy
1572 * @len: length in bytes
1573 * @gfp_mask: allocation flags for bio and page allocation
1574 * @reading: data direction is READ
1576 * copy the kernel address into a bio suitable for io to a block
1577 * device. Returns an error pointer in case of error.
1579 struct bio
*bio_copy_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1580 gfp_t gfp_mask
, int reading
)
1582 unsigned long kaddr
= (unsigned long)data
;
1583 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1584 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1593 return ERR_PTR(-EINVAL
);
1595 nr_pages
= end
- start
;
1596 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1598 return ERR_PTR(-ENOMEM
);
1602 unsigned int bytes
= PAGE_SIZE
;
1607 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
1612 memcpy(page_address(page
), p
, bytes
);
1614 if (bio_add_pc_page(q
, bio
, page
, bytes
, 0) < bytes
)
1622 bio
->bi_end_io
= bio_copy_kern_endio_read
;
1623 bio
->bi_private
= data
;
1625 bio
->bi_end_io
= bio_copy_kern_endio
;
1631 bio_free_pages(bio
);
1633 return ERR_PTR(-ENOMEM
);
1637 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1638 * for performing direct-IO in BIOs.
1640 * The problem is that we cannot run set_page_dirty() from interrupt context
1641 * because the required locks are not interrupt-safe. So what we can do is to
1642 * mark the pages dirty _before_ performing IO. And in interrupt context,
1643 * check that the pages are still dirty. If so, fine. If not, redirty them
1644 * in process context.
1646 * We special-case compound pages here: normally this means reads into hugetlb
1647 * pages. The logic in here doesn't really work right for compound pages
1648 * because the VM does not uniformly chase down the head page in all cases.
1649 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1650 * handle them at all. So we skip compound pages here at an early stage.
1652 * Note that this code is very hard to test under normal circumstances because
1653 * direct-io pins the pages with get_user_pages(). This makes
1654 * is_page_cache_freeable return false, and the VM will not clean the pages.
1655 * But other code (eg, flusher threads) could clean the pages if they are mapped
1658 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1659 * deferred bio dirtying paths.
1663 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1665 void bio_set_pages_dirty(struct bio
*bio
)
1667 struct bio_vec
*bvec
;
1670 bio_for_each_segment_all(bvec
, bio
, i
) {
1671 struct page
*page
= bvec
->bv_page
;
1673 if (page
&& !PageCompound(page
))
1674 set_page_dirty_lock(page
);
1678 static void bio_release_pages(struct bio
*bio
)
1680 struct bio_vec
*bvec
;
1683 bio_for_each_segment_all(bvec
, bio
, i
) {
1684 struct page
*page
= bvec
->bv_page
;
1692 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1693 * If they are, then fine. If, however, some pages are clean then they must
1694 * have been written out during the direct-IO read. So we take another ref on
1695 * the BIO and the offending pages and re-dirty the pages in process context.
1697 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1698 * here on. It will run one put_page() against each page and will run one
1699 * bio_put() against the BIO.
1702 static void bio_dirty_fn(struct work_struct
*work
);
1704 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
1705 static DEFINE_SPINLOCK(bio_dirty_lock
);
1706 static struct bio
*bio_dirty_list
;
1709 * This runs in process context
1711 static void bio_dirty_fn(struct work_struct
*work
)
1713 unsigned long flags
;
1716 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1717 bio
= bio_dirty_list
;
1718 bio_dirty_list
= NULL
;
1719 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1722 struct bio
*next
= bio
->bi_private
;
1724 bio_set_pages_dirty(bio
);
1725 bio_release_pages(bio
);
1731 void bio_check_pages_dirty(struct bio
*bio
)
1733 struct bio_vec
*bvec
;
1734 int nr_clean_pages
= 0;
1737 bio_for_each_segment_all(bvec
, bio
, i
) {
1738 struct page
*page
= bvec
->bv_page
;
1740 if (PageDirty(page
) || PageCompound(page
)) {
1742 bvec
->bv_page
= NULL
;
1748 if (nr_clean_pages
) {
1749 unsigned long flags
;
1751 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1752 bio
->bi_private
= bio_dirty_list
;
1753 bio_dirty_list
= bio
;
1754 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1755 schedule_work(&bio_dirty_work
);
1761 void generic_start_io_acct(int rw
, unsigned long sectors
,
1762 struct hd_struct
*part
)
1764 int cpu
= part_stat_lock();
1766 part_round_stats(cpu
, part
);
1767 part_stat_inc(cpu
, part
, ios
[rw
]);
1768 part_stat_add(cpu
, part
, sectors
[rw
], sectors
);
1769 part_inc_in_flight(part
, rw
);
1773 EXPORT_SYMBOL(generic_start_io_acct
);
1775 void generic_end_io_acct(int rw
, struct hd_struct
*part
,
1776 unsigned long start_time
)
1778 unsigned long duration
= jiffies
- start_time
;
1779 int cpu
= part_stat_lock();
1781 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1782 part_round_stats(cpu
, part
);
1783 part_dec_in_flight(part
, rw
);
1787 EXPORT_SYMBOL(generic_end_io_acct
);
1789 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1790 void bio_flush_dcache_pages(struct bio
*bi
)
1792 struct bio_vec bvec
;
1793 struct bvec_iter iter
;
1795 bio_for_each_segment(bvec
, bi
, iter
)
1796 flush_dcache_page(bvec
.bv_page
);
1798 EXPORT_SYMBOL(bio_flush_dcache_pages
);
1801 static inline bool bio_remaining_done(struct bio
*bio
)
1804 * If we're not chaining, then ->__bi_remaining is always 1 and
1805 * we always end io on the first invocation.
1807 if (!bio_flagged(bio
, BIO_CHAIN
))
1810 BUG_ON(atomic_read(&bio
->__bi_remaining
) <= 0);
1812 if (atomic_dec_and_test(&bio
->__bi_remaining
)) {
1813 bio_clear_flag(bio
, BIO_CHAIN
);
1821 * bio_endio - end I/O on a bio
1825 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1826 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1827 * bio unless they own it and thus know that it has an end_io function.
1829 void bio_endio(struct bio
*bio
)
1832 if (!bio_remaining_done(bio
))
1836 * Need to have a real endio function for chained bios, otherwise
1837 * various corner cases will break (like stacking block devices that
1838 * save/restore bi_end_io) - however, we want to avoid unbounded
1839 * recursion and blowing the stack. Tail call optimization would
1840 * handle this, but compiling with frame pointers also disables
1841 * gcc's sibling call optimization.
1843 if (bio
->bi_end_io
== bio_chain_endio
) {
1844 bio
= __bio_chain_endio(bio
);
1849 bio
->bi_end_io(bio
);
1851 EXPORT_SYMBOL(bio_endio
);
1854 * bio_split - split a bio
1855 * @bio: bio to split
1856 * @sectors: number of sectors to split from the front of @bio
1858 * @bs: bio set to allocate from
1860 * Allocates and returns a new bio which represents @sectors from the start of
1861 * @bio, and updates @bio to represent the remaining sectors.
1863 * Unless this is a discard request the newly allocated bio will point
1864 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1865 * @bio is not freed before the split.
1867 struct bio
*bio_split(struct bio
*bio
, int sectors
,
1868 gfp_t gfp
, struct bio_set
*bs
)
1870 struct bio
*split
= NULL
;
1872 BUG_ON(sectors
<= 0);
1873 BUG_ON(sectors
>= bio_sectors(bio
));
1875 split
= bio_clone_fast(bio
, gfp
, bs
);
1879 split
->bi_iter
.bi_size
= sectors
<< 9;
1881 if (bio_integrity(split
))
1882 bio_integrity_trim(split
, 0, sectors
);
1884 bio_advance(bio
, split
->bi_iter
.bi_size
);
1888 EXPORT_SYMBOL(bio_split
);
1891 * bio_trim - trim a bio
1893 * @offset: number of sectors to trim from the front of @bio
1894 * @size: size we want to trim @bio to, in sectors
1896 void bio_trim(struct bio
*bio
, int offset
, int size
)
1898 /* 'bio' is a cloned bio which we need to trim to match
1899 * the given offset and size.
1903 if (offset
== 0 && size
== bio
->bi_iter
.bi_size
)
1906 bio_clear_flag(bio
, BIO_SEG_VALID
);
1908 bio_advance(bio
, offset
<< 9);
1910 bio
->bi_iter
.bi_size
= size
;
1912 EXPORT_SYMBOL_GPL(bio_trim
);
1915 * create memory pools for biovec's in a bio_set.
1916 * use the global biovec slabs created for general use.
1918 mempool_t
*biovec_create_pool(int pool_entries
)
1920 struct biovec_slab
*bp
= bvec_slabs
+ BVEC_POOL_MAX
;
1922 return mempool_create_slab_pool(pool_entries
, bp
->slab
);
1925 void bioset_free(struct bio_set
*bs
)
1927 if (bs
->rescue_workqueue
)
1928 destroy_workqueue(bs
->rescue_workqueue
);
1931 mempool_destroy(bs
->bio_pool
);
1934 mempool_destroy(bs
->bvec_pool
);
1936 bioset_integrity_free(bs
);
1941 EXPORT_SYMBOL(bioset_free
);
1943 static struct bio_set
*__bioset_create(unsigned int pool_size
,
1944 unsigned int front_pad
,
1945 bool create_bvec_pool
)
1947 unsigned int back_pad
= BIO_INLINE_VECS
* sizeof(struct bio_vec
);
1950 bs
= kzalloc(sizeof(*bs
), GFP_KERNEL
);
1954 bs
->front_pad
= front_pad
;
1956 spin_lock_init(&bs
->rescue_lock
);
1957 bio_list_init(&bs
->rescue_list
);
1958 INIT_WORK(&bs
->rescue_work
, bio_alloc_rescue
);
1960 bs
->bio_slab
= bio_find_or_create_slab(front_pad
+ back_pad
);
1961 if (!bs
->bio_slab
) {
1966 bs
->bio_pool
= mempool_create_slab_pool(pool_size
, bs
->bio_slab
);
1970 if (create_bvec_pool
) {
1971 bs
->bvec_pool
= biovec_create_pool(pool_size
);
1976 bs
->rescue_workqueue
= alloc_workqueue("bioset", WQ_MEM_RECLAIM
, 0);
1977 if (!bs
->rescue_workqueue
)
1987 * bioset_create - Create a bio_set
1988 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1989 * @front_pad: Number of bytes to allocate in front of the returned bio
1992 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1993 * to ask for a number of bytes to be allocated in front of the bio.
1994 * Front pad allocation is useful for embedding the bio inside
1995 * another structure, to avoid allocating extra data to go with the bio.
1996 * Note that the bio must be embedded at the END of that structure always,
1997 * or things will break badly.
1999 struct bio_set
*bioset_create(unsigned int pool_size
, unsigned int front_pad
)
2001 return __bioset_create(pool_size
, front_pad
, true);
2003 EXPORT_SYMBOL(bioset_create
);
2006 * bioset_create_nobvec - Create a bio_set without bio_vec mempool
2007 * @pool_size: Number of bio to cache in the mempool
2008 * @front_pad: Number of bytes to allocate in front of the returned bio
2011 * Same functionality as bioset_create() except that mempool is not
2012 * created for bio_vecs. Saving some memory for bio_clone_fast() users.
2014 struct bio_set
*bioset_create_nobvec(unsigned int pool_size
, unsigned int front_pad
)
2016 return __bioset_create(pool_size
, front_pad
, false);
2018 EXPORT_SYMBOL(bioset_create_nobvec
);
2020 #ifdef CONFIG_BLK_CGROUP
2023 * bio_associate_blkcg - associate a bio with the specified blkcg
2025 * @blkcg_css: css of the blkcg to associate
2027 * Associate @bio with the blkcg specified by @blkcg_css. Block layer will
2028 * treat @bio as if it were issued by a task which belongs to the blkcg.
2030 * This function takes an extra reference of @blkcg_css which will be put
2031 * when @bio is released. The caller must own @bio and is responsible for
2032 * synchronizing calls to this function.
2034 int bio_associate_blkcg(struct bio
*bio
, struct cgroup_subsys_state
*blkcg_css
)
2036 if (unlikely(bio
->bi_css
))
2039 bio
->bi_css
= blkcg_css
;
2042 EXPORT_SYMBOL_GPL(bio_associate_blkcg
);
2045 * bio_associate_current - associate a bio with %current
2048 * Associate @bio with %current if it hasn't been associated yet. Block
2049 * layer will treat @bio as if it were issued by %current no matter which
2050 * task actually issues it.
2052 * This function takes an extra reference of @task's io_context and blkcg
2053 * which will be put when @bio is released. The caller must own @bio,
2054 * ensure %current->io_context exists, and is responsible for synchronizing
2055 * calls to this function.
2057 int bio_associate_current(struct bio
*bio
)
2059 struct io_context
*ioc
;
2064 ioc
= current
->io_context
;
2068 get_io_context_active(ioc
);
2070 bio
->bi_css
= task_get_css(current
, io_cgrp_id
);
2073 EXPORT_SYMBOL_GPL(bio_associate_current
);
2076 * bio_disassociate_task - undo bio_associate_current()
2079 void bio_disassociate_task(struct bio
*bio
)
2082 put_io_context(bio
->bi_ioc
);
2086 css_put(bio
->bi_css
);
2092 * bio_clone_blkcg_association - clone blkcg association from src to dst bio
2093 * @dst: destination bio
2096 void bio_clone_blkcg_association(struct bio
*dst
, struct bio
*src
)
2099 WARN_ON(bio_associate_blkcg(dst
, src
->bi_css
));
2102 #endif /* CONFIG_BLK_CGROUP */
2104 static void __init
biovec_init_slabs(void)
2108 for (i
= 0; i
< BVEC_POOL_NR
; i
++) {
2110 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
2112 if (bvs
->nr_vecs
<= BIO_INLINE_VECS
) {
2117 size
= bvs
->nr_vecs
* sizeof(struct bio_vec
);
2118 bvs
->slab
= kmem_cache_create(bvs
->name
, size
, 0,
2119 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
2123 static int __init
init_bio(void)
2127 bio_slabs
= kzalloc(bio_slab_max
* sizeof(struct bio_slab
), GFP_KERNEL
);
2129 panic("bio: can't allocate bios\n");
2131 bio_integrity_init();
2132 biovec_init_slabs();
2134 fs_bio_set
= bioset_create(BIO_POOL_SIZE
, 0);
2136 panic("bio: can't allocate bios\n");
2138 if (bioset_integrity_create(fs_bio_set
, BIO_POOL_SIZE
))
2139 panic("bio: can't create integrity pool\n");
2143 subsys_initcall(init_bio
);