]>
git.proxmox.com Git - mirror_ubuntu-zesty-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
[BIOVEC_NR_POOLS
] __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
)
163 BIO_BUG_ON(idx
>= BIOVEC_NR_POOLS
);
165 if (idx
== BIOVEC_MAX_IDX
)
166 mempool_free(bv
, pool
);
168 struct biovec_slab
*bvs
= bvec_slabs
+ idx
;
170 kmem_cache_free(bvs
->slab
, bv
);
174 struct bio_vec
*bvec_alloc(gfp_t gfp_mask
, int nr
, unsigned long *idx
,
180 * see comment near bvec_array define!
198 case 129 ... BIO_MAX_PAGES
:
206 * idx now points to the pool we want to allocate from. only the
207 * 1-vec entry pool is mempool backed.
209 if (*idx
== BIOVEC_MAX_IDX
) {
211 bvl
= mempool_alloc(pool
, gfp_mask
);
213 struct biovec_slab
*bvs
= bvec_slabs
+ *idx
;
214 gfp_t __gfp_mask
= gfp_mask
& ~(__GFP_WAIT
| __GFP_IO
);
217 * Make this allocation restricted and don't dump info on
218 * allocation failures, since we'll fallback to the mempool
219 * in case of failure.
221 __gfp_mask
|= __GFP_NOMEMALLOC
| __GFP_NORETRY
| __GFP_NOWARN
;
224 * Try a slab allocation. If this fails and __GFP_WAIT
225 * is set, retry with the 1-entry mempool
227 bvl
= kmem_cache_alloc(bvs
->slab
, __gfp_mask
);
228 if (unlikely(!bvl
&& (gfp_mask
& __GFP_WAIT
))) {
229 *idx
= BIOVEC_MAX_IDX
;
237 static void __bio_free(struct bio
*bio
)
239 bio_disassociate_task(bio
);
241 if (bio_integrity(bio
))
242 bio_integrity_free(bio
);
245 static void bio_free(struct bio
*bio
)
247 struct bio_set
*bs
= bio
->bi_pool
;
253 if (bio_flagged(bio
, BIO_OWNS_VEC
))
254 bvec_free(bs
->bvec_pool
, bio
->bi_io_vec
, BIO_POOL_IDX(bio
));
257 * If we have front padding, adjust the bio pointer before freeing
262 mempool_free(p
, bs
->bio_pool
);
264 /* Bio was allocated by bio_kmalloc() */
269 void bio_init(struct bio
*bio
)
271 memset(bio
, 0, sizeof(*bio
));
272 bio
->bi_flags
= 1 << BIO_UPTODATE
;
273 atomic_set(&bio
->__bi_remaining
, 1);
274 atomic_set(&bio
->__bi_cnt
, 1);
276 EXPORT_SYMBOL(bio_init
);
279 * bio_reset - reinitialize a bio
283 * After calling bio_reset(), @bio will be in the same state as a freshly
284 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
285 * preserved are the ones that are initialized by bio_alloc_bioset(). See
286 * comment in struct bio.
288 void bio_reset(struct bio
*bio
)
290 unsigned long flags
= bio
->bi_flags
& (~0UL << BIO_RESET_BITS
);
294 memset(bio
, 0, BIO_RESET_BYTES
);
295 bio
->bi_flags
= flags
| (1 << BIO_UPTODATE
);
296 atomic_set(&bio
->__bi_remaining
, 1);
298 EXPORT_SYMBOL(bio_reset
);
300 static void bio_chain_endio(struct bio
*bio
, int error
)
302 bio_endio(bio
->bi_private
, error
);
307 * bio_chain - chain bio completions
308 * @bio: the target bio
309 * @parent: the @bio's parent bio
311 * The caller won't have a bi_end_io called when @bio completes - instead,
312 * @parent's bi_end_io won't be called until both @parent and @bio have
313 * completed; the chained bio will also be freed when it completes.
315 * The caller must not set bi_private or bi_end_io in @bio.
317 void bio_chain(struct bio
*bio
, struct bio
*parent
)
319 BUG_ON(bio
->bi_private
|| bio
->bi_end_io
);
321 bio
->bi_private
= parent
;
322 bio
->bi_end_io
= bio_chain_endio
;
323 bio_inc_remaining(parent
);
325 EXPORT_SYMBOL(bio_chain
);
327 static void bio_alloc_rescue(struct work_struct
*work
)
329 struct bio_set
*bs
= container_of(work
, struct bio_set
, rescue_work
);
333 spin_lock(&bs
->rescue_lock
);
334 bio
= bio_list_pop(&bs
->rescue_list
);
335 spin_unlock(&bs
->rescue_lock
);
340 generic_make_request(bio
);
344 static void punt_bios_to_rescuer(struct bio_set
*bs
)
346 struct bio_list punt
, nopunt
;
350 * In order to guarantee forward progress we must punt only bios that
351 * were allocated from this bio_set; otherwise, if there was a bio on
352 * there for a stacking driver higher up in the stack, processing it
353 * could require allocating bios from this bio_set, and doing that from
354 * our own rescuer would be bad.
356 * Since bio lists are singly linked, pop them all instead of trying to
357 * remove from the middle of the list:
360 bio_list_init(&punt
);
361 bio_list_init(&nopunt
);
363 while ((bio
= bio_list_pop(current
->bio_list
)))
364 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
366 *current
->bio_list
= nopunt
;
368 spin_lock(&bs
->rescue_lock
);
369 bio_list_merge(&bs
->rescue_list
, &punt
);
370 spin_unlock(&bs
->rescue_lock
);
372 queue_work(bs
->rescue_workqueue
, &bs
->rescue_work
);
376 * bio_alloc_bioset - allocate a bio for I/O
377 * @gfp_mask: the GFP_ mask given to the slab allocator
378 * @nr_iovecs: number of iovecs to pre-allocate
379 * @bs: the bio_set to allocate from.
382 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
383 * backed by the @bs's mempool.
385 * When @bs is not NULL, if %__GFP_WAIT is set then bio_alloc will always be
386 * able to allocate a bio. This is due to the mempool guarantees. To make this
387 * work, callers must never allocate more than 1 bio at a time from this pool.
388 * Callers that need to allocate more than 1 bio must always submit the
389 * previously allocated bio for IO before attempting to allocate a new one.
390 * Failure to do so can cause deadlocks under memory pressure.
392 * Note that when running under generic_make_request() (i.e. any block
393 * driver), bios are not submitted until after you return - see the code in
394 * generic_make_request() that converts recursion into iteration, to prevent
397 * This would normally mean allocating multiple bios under
398 * generic_make_request() would be susceptible to deadlocks, but we have
399 * deadlock avoidance code that resubmits any blocked bios from a rescuer
402 * However, we do not guarantee forward progress for allocations from other
403 * mempools. Doing multiple allocations from the same mempool under
404 * generic_make_request() should be avoided - instead, use bio_set's front_pad
405 * for per bio allocations.
408 * Pointer to new bio on success, NULL on failure.
410 struct bio
*bio_alloc_bioset(gfp_t gfp_mask
, int nr_iovecs
, struct bio_set
*bs
)
412 gfp_t saved_gfp
= gfp_mask
;
414 unsigned inline_vecs
;
415 unsigned long idx
= BIO_POOL_NONE
;
416 struct bio_vec
*bvl
= NULL
;
421 if (nr_iovecs
> UIO_MAXIOV
)
424 p
= kmalloc(sizeof(struct bio
) +
425 nr_iovecs
* sizeof(struct bio_vec
),
428 inline_vecs
= nr_iovecs
;
430 /* should not use nobvec bioset for nr_iovecs > 0 */
431 if (WARN_ON_ONCE(!bs
->bvec_pool
&& nr_iovecs
> 0))
434 * generic_make_request() converts recursion to iteration; this
435 * means if we're running beneath it, any bios we allocate and
436 * submit will not be submitted (and thus freed) until after we
439 * This exposes us to a potential deadlock if we allocate
440 * multiple bios from the same bio_set() while running
441 * underneath generic_make_request(). If we were to allocate
442 * multiple bios (say a stacking block driver that was splitting
443 * bios), we would deadlock if we exhausted the mempool's
446 * We solve this, and guarantee forward progress, with a rescuer
447 * workqueue per bio_set. If we go to allocate and there are
448 * bios on current->bio_list, we first try the allocation
449 * without __GFP_WAIT; if that fails, we punt those bios we
450 * would be blocking to the rescuer workqueue before we retry
451 * with the original gfp_flags.
454 if (current
->bio_list
&& !bio_list_empty(current
->bio_list
))
455 gfp_mask
&= ~__GFP_WAIT
;
457 p
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
458 if (!p
&& gfp_mask
!= saved_gfp
) {
459 punt_bios_to_rescuer(bs
);
460 gfp_mask
= saved_gfp
;
461 p
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
464 front_pad
= bs
->front_pad
;
465 inline_vecs
= BIO_INLINE_VECS
;
474 if (nr_iovecs
> inline_vecs
) {
475 bvl
= bvec_alloc(gfp_mask
, nr_iovecs
, &idx
, bs
->bvec_pool
);
476 if (!bvl
&& gfp_mask
!= saved_gfp
) {
477 punt_bios_to_rescuer(bs
);
478 gfp_mask
= saved_gfp
;
479 bvl
= bvec_alloc(gfp_mask
, nr_iovecs
, &idx
, bs
->bvec_pool
);
485 bio
->bi_flags
|= 1 << BIO_OWNS_VEC
;
486 } else if (nr_iovecs
) {
487 bvl
= bio
->bi_inline_vecs
;
491 bio
->bi_flags
|= idx
<< BIO_POOL_OFFSET
;
492 bio
->bi_max_vecs
= nr_iovecs
;
493 bio
->bi_io_vec
= bvl
;
497 mempool_free(p
, bs
->bio_pool
);
500 EXPORT_SYMBOL(bio_alloc_bioset
);
502 void zero_fill_bio(struct bio
*bio
)
506 struct bvec_iter iter
;
508 bio_for_each_segment(bv
, bio
, iter
) {
509 char *data
= bvec_kmap_irq(&bv
, &flags
);
510 memset(data
, 0, bv
.bv_len
);
511 flush_dcache_page(bv
.bv_page
);
512 bvec_kunmap_irq(data
, &flags
);
515 EXPORT_SYMBOL(zero_fill_bio
);
518 * bio_put - release a reference to a bio
519 * @bio: bio to release reference to
522 * Put a reference to a &struct bio, either one you have gotten with
523 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
525 void bio_put(struct bio
*bio
)
527 if (!bio_flagged(bio
, BIO_REFFED
))
530 BIO_BUG_ON(!atomic_read(&bio
->__bi_cnt
));
535 if (atomic_dec_and_test(&bio
->__bi_cnt
))
539 EXPORT_SYMBOL(bio_put
);
541 inline int bio_phys_segments(struct request_queue
*q
, struct bio
*bio
)
543 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
544 blk_recount_segments(q
, bio
);
546 return bio
->bi_phys_segments
;
548 EXPORT_SYMBOL(bio_phys_segments
);
551 * __bio_clone_fast - clone a bio that shares the original bio's biovec
552 * @bio: destination bio
553 * @bio_src: bio to clone
555 * Clone a &bio. Caller will own the returned bio, but not
556 * the actual data it points to. Reference count of returned
559 * Caller must ensure that @bio_src is not freed before @bio.
561 void __bio_clone_fast(struct bio
*bio
, struct bio
*bio_src
)
563 BUG_ON(bio
->bi_pool
&& BIO_POOL_IDX(bio
) != BIO_POOL_NONE
);
566 * most users will be overriding ->bi_bdev with a new target,
567 * so we don't set nor calculate new physical/hw segment counts here
569 bio
->bi_bdev
= bio_src
->bi_bdev
;
570 bio
->bi_flags
|= 1 << BIO_CLONED
;
571 bio
->bi_rw
= bio_src
->bi_rw
;
572 bio
->bi_iter
= bio_src
->bi_iter
;
573 bio
->bi_io_vec
= bio_src
->bi_io_vec
;
575 EXPORT_SYMBOL(__bio_clone_fast
);
578 * bio_clone_fast - clone a bio that shares the original bio's biovec
580 * @gfp_mask: allocation priority
581 * @bs: bio_set to allocate from
583 * Like __bio_clone_fast, only also allocates the returned bio
585 struct bio
*bio_clone_fast(struct bio
*bio
, gfp_t gfp_mask
, struct bio_set
*bs
)
589 b
= bio_alloc_bioset(gfp_mask
, 0, bs
);
593 __bio_clone_fast(b
, bio
);
595 if (bio_integrity(bio
)) {
598 ret
= bio_integrity_clone(b
, bio
, gfp_mask
);
608 EXPORT_SYMBOL(bio_clone_fast
);
611 * bio_clone_bioset - clone a bio
612 * @bio_src: bio to clone
613 * @gfp_mask: allocation priority
614 * @bs: bio_set to allocate from
616 * Clone bio. Caller will own the returned bio, but not the actual data it
617 * points to. Reference count of returned bio will be one.
619 struct bio
*bio_clone_bioset(struct bio
*bio_src
, gfp_t gfp_mask
,
622 struct bvec_iter iter
;
627 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
628 * bio_src->bi_io_vec to bio->bi_io_vec.
630 * We can't do that anymore, because:
632 * - The point of cloning the biovec is to produce a bio with a biovec
633 * the caller can modify: bi_idx and bi_bvec_done should be 0.
635 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
636 * we tried to clone the whole thing bio_alloc_bioset() would fail.
637 * But the clone should succeed as long as the number of biovecs we
638 * actually need to allocate is fewer than BIO_MAX_PAGES.
640 * - Lastly, bi_vcnt should not be looked at or relied upon by code
641 * that does not own the bio - reason being drivers don't use it for
642 * iterating over the biovec anymore, so expecting it to be kept up
643 * to date (i.e. for clones that share the parent biovec) is just
644 * asking for trouble and would force extra work on
645 * __bio_clone_fast() anyways.
648 bio
= bio_alloc_bioset(gfp_mask
, bio_segments(bio_src
), bs
);
652 bio
->bi_bdev
= bio_src
->bi_bdev
;
653 bio
->bi_rw
= bio_src
->bi_rw
;
654 bio
->bi_iter
.bi_sector
= bio_src
->bi_iter
.bi_sector
;
655 bio
->bi_iter
.bi_size
= bio_src
->bi_iter
.bi_size
;
657 if (bio
->bi_rw
& REQ_DISCARD
)
658 goto integrity_clone
;
660 if (bio
->bi_rw
& REQ_WRITE_SAME
) {
661 bio
->bi_io_vec
[bio
->bi_vcnt
++] = bio_src
->bi_io_vec
[0];
662 goto integrity_clone
;
665 bio_for_each_segment(bv
, bio_src
, iter
)
666 bio
->bi_io_vec
[bio
->bi_vcnt
++] = bv
;
669 if (bio_integrity(bio_src
)) {
672 ret
= bio_integrity_clone(bio
, bio_src
, gfp_mask
);
681 EXPORT_SYMBOL(bio_clone_bioset
);
684 * bio_get_nr_vecs - return approx number of vecs
687 * Return the approximate number of pages we can send to this target.
688 * There's no guarantee that you will be able to fit this number of pages
689 * into a bio, it does not account for dynamic restrictions that vary
692 int bio_get_nr_vecs(struct block_device
*bdev
)
694 struct request_queue
*q
= bdev_get_queue(bdev
);
697 nr_pages
= min_t(unsigned,
698 queue_max_segments(q
),
699 queue_max_sectors(q
) / (PAGE_SIZE
>> 9) + 1);
701 return min_t(unsigned, nr_pages
, BIO_MAX_PAGES
);
704 EXPORT_SYMBOL(bio_get_nr_vecs
);
706 static int __bio_add_page(struct request_queue
*q
, struct bio
*bio
, struct page
707 *page
, unsigned int len
, unsigned int offset
,
708 unsigned int max_sectors
)
710 int retried_segments
= 0;
711 struct bio_vec
*bvec
;
714 * cloned bio must not modify vec list
716 if (unlikely(bio_flagged(bio
, BIO_CLONED
)))
719 if (((bio
->bi_iter
.bi_size
+ len
) >> 9) > max_sectors
)
723 * For filesystems with a blocksize smaller than the pagesize
724 * we will often be called with the same page as last time and
725 * a consecutive offset. Optimize this special case.
727 if (bio
->bi_vcnt
> 0) {
728 struct bio_vec
*prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
730 if (page
== prev
->bv_page
&&
731 offset
== prev
->bv_offset
+ prev
->bv_len
) {
732 unsigned int prev_bv_len
= prev
->bv_len
;
735 if (q
->merge_bvec_fn
) {
736 struct bvec_merge_data bvm
= {
737 /* prev_bvec is already charged in
738 bi_size, discharge it in order to
739 simulate merging updated prev_bvec
741 .bi_bdev
= bio
->bi_bdev
,
742 .bi_sector
= bio
->bi_iter
.bi_sector
,
743 .bi_size
= bio
->bi_iter
.bi_size
-
748 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < prev
->bv_len
) {
754 bio
->bi_iter
.bi_size
+= len
;
759 * If the queue doesn't support SG gaps and adding this
760 * offset would create a gap, disallow it.
762 if (q
->queue_flags
& (1 << QUEUE_FLAG_SG_GAPS
) &&
763 bvec_gap_to_prev(prev
, offset
))
767 if (bio
->bi_vcnt
>= bio
->bi_max_vecs
)
771 * setup the new entry, we might clear it again later if we
772 * cannot add the page
774 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
775 bvec
->bv_page
= page
;
777 bvec
->bv_offset
= offset
;
779 bio
->bi_phys_segments
++;
780 bio
->bi_iter
.bi_size
+= len
;
783 * Perform a recount if the number of segments is greater
784 * than queue_max_segments(q).
787 while (bio
->bi_phys_segments
> queue_max_segments(q
)) {
789 if (retried_segments
)
792 retried_segments
= 1;
793 blk_recount_segments(q
, bio
);
797 * if queue has other restrictions (eg varying max sector size
798 * depending on offset), it can specify a merge_bvec_fn in the
799 * queue to get further control
801 if (q
->merge_bvec_fn
) {
802 struct bvec_merge_data bvm
= {
803 .bi_bdev
= bio
->bi_bdev
,
804 .bi_sector
= bio
->bi_iter
.bi_sector
,
805 .bi_size
= bio
->bi_iter
.bi_size
- len
,
810 * merge_bvec_fn() returns number of bytes it can accept
813 if (q
->merge_bvec_fn(q
, &bvm
, bvec
) < bvec
->bv_len
)
817 /* If we may be able to merge these biovecs, force a recount */
818 if (bio
->bi_vcnt
> 1 && (BIOVEC_PHYS_MERGEABLE(bvec
-1, bvec
)))
819 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
825 bvec
->bv_page
= NULL
;
829 bio
->bi_iter
.bi_size
-= len
;
830 blk_recount_segments(q
, bio
);
835 * bio_add_pc_page - attempt to add page to bio
836 * @q: the target queue
837 * @bio: destination bio
839 * @len: vec entry length
840 * @offset: vec entry offset
842 * Attempt to add a page to the bio_vec maplist. This can fail for a
843 * number of reasons, such as the bio being full or target block device
844 * limitations. The target block device must allow bio's up to PAGE_SIZE,
845 * so it is always possible to add a single page to an empty bio.
847 * This should only be used by REQ_PC bios.
849 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
, struct page
*page
,
850 unsigned int len
, unsigned int offset
)
852 return __bio_add_page(q
, bio
, page
, len
, offset
,
853 queue_max_hw_sectors(q
));
855 EXPORT_SYMBOL(bio_add_pc_page
);
858 * bio_add_page - attempt to add page to bio
859 * @bio: destination bio
861 * @len: vec entry length
862 * @offset: vec entry offset
864 * Attempt to add a page to the bio_vec maplist. This can fail for a
865 * number of reasons, such as the bio being full or target block device
866 * limitations. The target block device must allow bio's up to PAGE_SIZE,
867 * so it is always possible to add a single page to an empty bio.
869 int bio_add_page(struct bio
*bio
, struct page
*page
, unsigned int len
,
872 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
873 unsigned int max_sectors
;
875 max_sectors
= blk_max_size_offset(q
, bio
->bi_iter
.bi_sector
);
876 if ((max_sectors
< (len
>> 9)) && !bio
->bi_iter
.bi_size
)
877 max_sectors
= len
>> 9;
879 return __bio_add_page(q
, bio
, page
, len
, offset
, max_sectors
);
881 EXPORT_SYMBOL(bio_add_page
);
883 struct submit_bio_ret
{
884 struct completion event
;
888 static void submit_bio_wait_endio(struct bio
*bio
, int error
)
890 struct submit_bio_ret
*ret
= bio
->bi_private
;
893 complete(&ret
->event
);
897 * submit_bio_wait - submit a bio, and wait until it completes
898 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
899 * @bio: The &struct bio which describes the I/O
901 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
902 * bio_endio() on failure.
904 int submit_bio_wait(int rw
, struct bio
*bio
)
906 struct submit_bio_ret ret
;
909 init_completion(&ret
.event
);
910 bio
->bi_private
= &ret
;
911 bio
->bi_end_io
= submit_bio_wait_endio
;
913 wait_for_completion(&ret
.event
);
917 EXPORT_SYMBOL(submit_bio_wait
);
920 * bio_advance - increment/complete a bio by some number of bytes
921 * @bio: bio to advance
922 * @bytes: number of bytes to complete
924 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
925 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
926 * be updated on the last bvec as well.
928 * @bio will then represent the remaining, uncompleted portion of the io.
930 void bio_advance(struct bio
*bio
, unsigned bytes
)
932 if (bio_integrity(bio
))
933 bio_integrity_advance(bio
, bytes
);
935 bio_advance_iter(bio
, &bio
->bi_iter
, bytes
);
937 EXPORT_SYMBOL(bio_advance
);
940 * bio_alloc_pages - allocates a single page for each bvec in a bio
941 * @bio: bio to allocate pages for
942 * @gfp_mask: flags for allocation
944 * Allocates pages up to @bio->bi_vcnt.
946 * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
949 int bio_alloc_pages(struct bio
*bio
, gfp_t gfp_mask
)
954 bio_for_each_segment_all(bv
, bio
, i
) {
955 bv
->bv_page
= alloc_page(gfp_mask
);
957 while (--bv
>= bio
->bi_io_vec
)
958 __free_page(bv
->bv_page
);
965 EXPORT_SYMBOL(bio_alloc_pages
);
968 * bio_copy_data - copy contents of data buffers from one chain of bios to
970 * @src: source bio list
971 * @dst: destination bio list
973 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
974 * @src and @dst as linked lists of bios.
976 * Stops when it reaches the end of either @src or @dst - that is, copies
977 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
979 void bio_copy_data(struct bio
*dst
, struct bio
*src
)
981 struct bvec_iter src_iter
, dst_iter
;
982 struct bio_vec src_bv
, dst_bv
;
986 src_iter
= src
->bi_iter
;
987 dst_iter
= dst
->bi_iter
;
990 if (!src_iter
.bi_size
) {
995 src_iter
= src
->bi_iter
;
998 if (!dst_iter
.bi_size
) {
1003 dst_iter
= dst
->bi_iter
;
1006 src_bv
= bio_iter_iovec(src
, src_iter
);
1007 dst_bv
= bio_iter_iovec(dst
, dst_iter
);
1009 bytes
= min(src_bv
.bv_len
, dst_bv
.bv_len
);
1011 src_p
= kmap_atomic(src_bv
.bv_page
);
1012 dst_p
= kmap_atomic(dst_bv
.bv_page
);
1014 memcpy(dst_p
+ dst_bv
.bv_offset
,
1015 src_p
+ src_bv
.bv_offset
,
1018 kunmap_atomic(dst_p
);
1019 kunmap_atomic(src_p
);
1021 bio_advance_iter(src
, &src_iter
, bytes
);
1022 bio_advance_iter(dst
, &dst_iter
, bytes
);
1025 EXPORT_SYMBOL(bio_copy_data
);
1027 struct bio_map_data
{
1029 struct iov_iter iter
;
1033 static struct bio_map_data
*bio_alloc_map_data(unsigned int iov_count
,
1036 if (iov_count
> UIO_MAXIOV
)
1039 return kmalloc(sizeof(struct bio_map_data
) +
1040 sizeof(struct iovec
) * iov_count
, gfp_mask
);
1044 * bio_copy_from_iter - copy all pages from iov_iter to bio
1045 * @bio: The &struct bio which describes the I/O as destination
1046 * @iter: iov_iter as source
1048 * Copy all pages from iov_iter to bio.
1049 * Returns 0 on success, or error on failure.
1051 static int bio_copy_from_iter(struct bio
*bio
, struct iov_iter iter
)
1054 struct bio_vec
*bvec
;
1056 bio_for_each_segment_all(bvec
, bio
, i
) {
1059 ret
= copy_page_from_iter(bvec
->bv_page
,
1064 if (!iov_iter_count(&iter
))
1067 if (ret
< bvec
->bv_len
)
1075 * bio_copy_to_iter - copy all pages from bio to iov_iter
1076 * @bio: The &struct bio which describes the I/O as source
1077 * @iter: iov_iter as destination
1079 * Copy all pages from bio to iov_iter.
1080 * Returns 0 on success, or error on failure.
1082 static int bio_copy_to_iter(struct bio
*bio
, struct iov_iter iter
)
1085 struct bio_vec
*bvec
;
1087 bio_for_each_segment_all(bvec
, bio
, i
) {
1090 ret
= copy_page_to_iter(bvec
->bv_page
,
1095 if (!iov_iter_count(&iter
))
1098 if (ret
< bvec
->bv_len
)
1105 static void bio_free_pages(struct bio
*bio
)
1107 struct bio_vec
*bvec
;
1110 bio_for_each_segment_all(bvec
, bio
, i
)
1111 __free_page(bvec
->bv_page
);
1115 * bio_uncopy_user - finish previously mapped bio
1116 * @bio: bio being terminated
1118 * Free pages allocated from bio_copy_user_iov() and write back data
1119 * to user space in case of a read.
1121 int bio_uncopy_user(struct bio
*bio
)
1123 struct bio_map_data
*bmd
= bio
->bi_private
;
1126 if (!bio_flagged(bio
, BIO_NULL_MAPPED
)) {
1128 * if we're in a workqueue, the request is orphaned, so
1129 * don't copy into a random user address space, just free.
1131 if (current
->mm
&& bio_data_dir(bio
) == READ
)
1132 ret
= bio_copy_to_iter(bio
, bmd
->iter
);
1133 if (bmd
->is_our_pages
)
1134 bio_free_pages(bio
);
1140 EXPORT_SYMBOL(bio_uncopy_user
);
1143 * bio_copy_user_iov - copy user data to bio
1144 * @q: destination block queue
1145 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1146 * @iter: iovec iterator
1147 * @gfp_mask: memory allocation flags
1149 * Prepares and returns a bio for indirect user io, bouncing data
1150 * to/from kernel pages as necessary. Must be paired with
1151 * call bio_uncopy_user() on io completion.
1153 struct bio
*bio_copy_user_iov(struct request_queue
*q
,
1154 struct rq_map_data
*map_data
,
1155 const struct iov_iter
*iter
,
1158 struct bio_map_data
*bmd
;
1163 unsigned int len
= iter
->count
;
1164 unsigned int offset
= map_data
? map_data
->offset
& ~PAGE_MASK
: 0;
1166 for (i
= 0; i
< iter
->nr_segs
; i
++) {
1167 unsigned long uaddr
;
1169 unsigned long start
;
1171 uaddr
= (unsigned long) iter
->iov
[i
].iov_base
;
1172 end
= (uaddr
+ iter
->iov
[i
].iov_len
+ PAGE_SIZE
- 1)
1174 start
= uaddr
>> PAGE_SHIFT
;
1180 return ERR_PTR(-EINVAL
);
1182 nr_pages
+= end
- start
;
1188 bmd
= bio_alloc_map_data(iter
->nr_segs
, gfp_mask
);
1190 return ERR_PTR(-ENOMEM
);
1193 * We need to do a deep copy of the iov_iter including the iovecs.
1194 * The caller provided iov might point to an on-stack or otherwise
1197 bmd
->is_our_pages
= map_data
? 0 : 1;
1198 memcpy(bmd
->iov
, iter
->iov
, sizeof(struct iovec
) * iter
->nr_segs
);
1199 iov_iter_init(&bmd
->iter
, iter
->type
, bmd
->iov
,
1200 iter
->nr_segs
, iter
->count
);
1203 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1207 if (iter
->type
& WRITE
)
1208 bio
->bi_rw
|= REQ_WRITE
;
1213 nr_pages
= 1 << map_data
->page_order
;
1214 i
= map_data
->offset
/ PAGE_SIZE
;
1217 unsigned int bytes
= PAGE_SIZE
;
1225 if (i
== map_data
->nr_entries
* nr_pages
) {
1230 page
= map_data
->pages
[i
/ nr_pages
];
1231 page
+= (i
% nr_pages
);
1235 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
1242 if (bio_add_pc_page(q
, bio
, page
, bytes
, offset
) < bytes
)
1255 if (((iter
->type
& WRITE
) && (!map_data
|| !map_data
->null_mapped
)) ||
1256 (map_data
&& map_data
->from_user
)) {
1257 ret
= bio_copy_from_iter(bio
, *iter
);
1262 bio
->bi_private
= bmd
;
1266 bio_free_pages(bio
);
1270 return ERR_PTR(ret
);
1274 * bio_map_user_iov - map user iovec into bio
1275 * @q: the struct request_queue for the bio
1276 * @iter: iovec iterator
1277 * @gfp_mask: memory allocation flags
1279 * Map the user space address into a bio suitable for io to a block
1280 * device. Returns an error pointer in case of error.
1282 struct bio
*bio_map_user_iov(struct request_queue
*q
,
1283 const struct iov_iter
*iter
,
1288 struct page
**pages
;
1295 iov_for_each(iov
, i
, *iter
) {
1296 unsigned long uaddr
= (unsigned long) iov
.iov_base
;
1297 unsigned long len
= iov
.iov_len
;
1298 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1299 unsigned long start
= uaddr
>> PAGE_SHIFT
;
1305 return ERR_PTR(-EINVAL
);
1307 nr_pages
+= end
- start
;
1309 * buffer must be aligned to at least hardsector size for now
1311 if (uaddr
& queue_dma_alignment(q
))
1312 return ERR_PTR(-EINVAL
);
1316 return ERR_PTR(-EINVAL
);
1318 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1320 return ERR_PTR(-ENOMEM
);
1323 pages
= kcalloc(nr_pages
, sizeof(struct page
*), gfp_mask
);
1327 iov_for_each(iov
, i
, *iter
) {
1328 unsigned long uaddr
= (unsigned long) iov
.iov_base
;
1329 unsigned long len
= iov
.iov_len
;
1330 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1331 unsigned long start
= uaddr
>> PAGE_SHIFT
;
1332 const int local_nr_pages
= end
- start
;
1333 const int page_limit
= cur_page
+ local_nr_pages
;
1335 ret
= get_user_pages_fast(uaddr
, local_nr_pages
,
1336 (iter
->type
& WRITE
) != WRITE
,
1338 if (ret
< local_nr_pages
) {
1343 offset
= uaddr
& ~PAGE_MASK
;
1344 for (j
= cur_page
; j
< page_limit
; j
++) {
1345 unsigned int bytes
= PAGE_SIZE
- offset
;
1356 if (bio_add_pc_page(q
, bio
, pages
[j
], bytes
, offset
) <
1366 * release the pages we didn't map into the bio, if any
1368 while (j
< page_limit
)
1369 page_cache_release(pages
[j
++]);
1375 * set data direction, and check if mapped pages need bouncing
1377 if (iter
->type
& WRITE
)
1378 bio
->bi_rw
|= REQ_WRITE
;
1380 bio
->bi_flags
|= (1 << BIO_USER_MAPPED
);
1383 * subtle -- if __bio_map_user() ended up bouncing a bio,
1384 * it would normally disappear when its bi_end_io is run.
1385 * however, we need it for the unmap, so grab an extra
1392 for (j
= 0; j
< nr_pages
; j
++) {
1395 page_cache_release(pages
[j
]);
1400 return ERR_PTR(ret
);
1403 static void __bio_unmap_user(struct bio
*bio
)
1405 struct bio_vec
*bvec
;
1409 * make sure we dirty pages we wrote to
1411 bio_for_each_segment_all(bvec
, bio
, i
) {
1412 if (bio_data_dir(bio
) == READ
)
1413 set_page_dirty_lock(bvec
->bv_page
);
1415 page_cache_release(bvec
->bv_page
);
1422 * bio_unmap_user - unmap a bio
1423 * @bio: the bio being unmapped
1425 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1426 * a process context.
1428 * bio_unmap_user() may sleep.
1430 void bio_unmap_user(struct bio
*bio
)
1432 __bio_unmap_user(bio
);
1435 EXPORT_SYMBOL(bio_unmap_user
);
1437 static void bio_map_kern_endio(struct bio
*bio
, int err
)
1443 * bio_map_kern - map kernel address into bio
1444 * @q: the struct request_queue for the bio
1445 * @data: pointer to buffer to map
1446 * @len: length in bytes
1447 * @gfp_mask: allocation flags for bio allocation
1449 * Map the kernel address into a bio suitable for io to a block
1450 * device. Returns an error pointer in case of error.
1452 struct bio
*bio_map_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1455 unsigned long kaddr
= (unsigned long)data
;
1456 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1457 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1458 const int nr_pages
= end
- start
;
1462 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1464 return ERR_PTR(-ENOMEM
);
1466 offset
= offset_in_page(kaddr
);
1467 for (i
= 0; i
< nr_pages
; i
++) {
1468 unsigned int bytes
= PAGE_SIZE
- offset
;
1476 if (bio_add_pc_page(q
, bio
, virt_to_page(data
), bytes
,
1478 /* we don't support partial mappings */
1480 return ERR_PTR(-EINVAL
);
1488 bio
->bi_end_io
= bio_map_kern_endio
;
1491 EXPORT_SYMBOL(bio_map_kern
);
1493 static void bio_copy_kern_endio(struct bio
*bio
, int err
)
1495 bio_free_pages(bio
);
1499 static void bio_copy_kern_endio_read(struct bio
*bio
, int err
)
1501 char *p
= bio
->bi_private
;
1502 struct bio_vec
*bvec
;
1505 bio_for_each_segment_all(bvec
, bio
, i
) {
1506 memcpy(p
, page_address(bvec
->bv_page
), bvec
->bv_len
);
1510 bio_copy_kern_endio(bio
, err
);
1514 * bio_copy_kern - copy kernel address into bio
1515 * @q: the struct request_queue for the bio
1516 * @data: pointer to buffer to copy
1517 * @len: length in bytes
1518 * @gfp_mask: allocation flags for bio and page allocation
1519 * @reading: data direction is READ
1521 * copy the kernel address into a bio suitable for io to a block
1522 * device. Returns an error pointer in case of error.
1524 struct bio
*bio_copy_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1525 gfp_t gfp_mask
, int reading
)
1527 unsigned long kaddr
= (unsigned long)data
;
1528 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1529 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1538 return ERR_PTR(-EINVAL
);
1540 nr_pages
= end
- start
;
1541 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1543 return ERR_PTR(-ENOMEM
);
1547 unsigned int bytes
= PAGE_SIZE
;
1552 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
1557 memcpy(page_address(page
), p
, bytes
);
1559 if (bio_add_pc_page(q
, bio
, page
, bytes
, 0) < bytes
)
1567 bio
->bi_end_io
= bio_copy_kern_endio_read
;
1568 bio
->bi_private
= data
;
1570 bio
->bi_end_io
= bio_copy_kern_endio
;
1571 bio
->bi_rw
|= REQ_WRITE
;
1577 bio_free_pages(bio
);
1579 return ERR_PTR(-ENOMEM
);
1581 EXPORT_SYMBOL(bio_copy_kern
);
1584 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1585 * for performing direct-IO in BIOs.
1587 * The problem is that we cannot run set_page_dirty() from interrupt context
1588 * because the required locks are not interrupt-safe. So what we can do is to
1589 * mark the pages dirty _before_ performing IO. And in interrupt context,
1590 * check that the pages are still dirty. If so, fine. If not, redirty them
1591 * in process context.
1593 * We special-case compound pages here: normally this means reads into hugetlb
1594 * pages. The logic in here doesn't really work right for compound pages
1595 * because the VM does not uniformly chase down the head page in all cases.
1596 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1597 * handle them at all. So we skip compound pages here at an early stage.
1599 * Note that this code is very hard to test under normal circumstances because
1600 * direct-io pins the pages with get_user_pages(). This makes
1601 * is_page_cache_freeable return false, and the VM will not clean the pages.
1602 * But other code (eg, flusher threads) could clean the pages if they are mapped
1605 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1606 * deferred bio dirtying paths.
1610 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1612 void bio_set_pages_dirty(struct bio
*bio
)
1614 struct bio_vec
*bvec
;
1617 bio_for_each_segment_all(bvec
, bio
, i
) {
1618 struct page
*page
= bvec
->bv_page
;
1620 if (page
&& !PageCompound(page
))
1621 set_page_dirty_lock(page
);
1625 static void bio_release_pages(struct bio
*bio
)
1627 struct bio_vec
*bvec
;
1630 bio_for_each_segment_all(bvec
, bio
, i
) {
1631 struct page
*page
= bvec
->bv_page
;
1639 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1640 * If they are, then fine. If, however, some pages are clean then they must
1641 * have been written out during the direct-IO read. So we take another ref on
1642 * the BIO and the offending pages and re-dirty the pages in process context.
1644 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1645 * here on. It will run one page_cache_release() against each page and will
1646 * run one bio_put() against the BIO.
1649 static void bio_dirty_fn(struct work_struct
*work
);
1651 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
1652 static DEFINE_SPINLOCK(bio_dirty_lock
);
1653 static struct bio
*bio_dirty_list
;
1656 * This runs in process context
1658 static void bio_dirty_fn(struct work_struct
*work
)
1660 unsigned long flags
;
1663 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1664 bio
= bio_dirty_list
;
1665 bio_dirty_list
= NULL
;
1666 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1669 struct bio
*next
= bio
->bi_private
;
1671 bio_set_pages_dirty(bio
);
1672 bio_release_pages(bio
);
1678 void bio_check_pages_dirty(struct bio
*bio
)
1680 struct bio_vec
*bvec
;
1681 int nr_clean_pages
= 0;
1684 bio_for_each_segment_all(bvec
, bio
, i
) {
1685 struct page
*page
= bvec
->bv_page
;
1687 if (PageDirty(page
) || PageCompound(page
)) {
1688 page_cache_release(page
);
1689 bvec
->bv_page
= NULL
;
1695 if (nr_clean_pages
) {
1696 unsigned long flags
;
1698 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1699 bio
->bi_private
= bio_dirty_list
;
1700 bio_dirty_list
= bio
;
1701 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1702 schedule_work(&bio_dirty_work
);
1708 void generic_start_io_acct(int rw
, unsigned long sectors
,
1709 struct hd_struct
*part
)
1711 int cpu
= part_stat_lock();
1713 part_round_stats(cpu
, part
);
1714 part_stat_inc(cpu
, part
, ios
[rw
]);
1715 part_stat_add(cpu
, part
, sectors
[rw
], sectors
);
1716 part_inc_in_flight(part
, rw
);
1720 EXPORT_SYMBOL(generic_start_io_acct
);
1722 void generic_end_io_acct(int rw
, struct hd_struct
*part
,
1723 unsigned long start_time
)
1725 unsigned long duration
= jiffies
- start_time
;
1726 int cpu
= part_stat_lock();
1728 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1729 part_round_stats(cpu
, part
);
1730 part_dec_in_flight(part
, rw
);
1734 EXPORT_SYMBOL(generic_end_io_acct
);
1736 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1737 void bio_flush_dcache_pages(struct bio
*bi
)
1739 struct bio_vec bvec
;
1740 struct bvec_iter iter
;
1742 bio_for_each_segment(bvec
, bi
, iter
)
1743 flush_dcache_page(bvec
.bv_page
);
1745 EXPORT_SYMBOL(bio_flush_dcache_pages
);
1748 static inline bool bio_remaining_done(struct bio
*bio
)
1751 * If we're not chaining, then ->__bi_remaining is always 1 and
1752 * we always end io on the first invocation.
1754 if (!bio_flagged(bio
, BIO_CHAIN
))
1757 BUG_ON(atomic_read(&bio
->__bi_remaining
) <= 0);
1759 if (atomic_dec_and_test(&bio
->__bi_remaining
))
1766 * bio_endio - end I/O on a bio
1768 * @error: error, if any
1771 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1772 * preferred way to end I/O on a bio, it takes care of clearing
1773 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1774 * established -Exxxx (-EIO, for instance) error values in case
1775 * something went wrong. No one should call bi_end_io() directly on a
1776 * bio unless they own it and thus know that it has an end_io
1779 void bio_endio(struct bio
*bio
, int error
)
1783 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1784 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
1787 if (unlikely(!bio_remaining_done(bio
)))
1791 * Need to have a real endio function for chained bios,
1792 * otherwise various corner cases will break (like stacking
1793 * block devices that save/restore bi_end_io) - however, we want
1794 * to avoid unbounded recursion and blowing the stack. Tail call
1795 * optimization would handle this, but compiling with frame
1796 * pointers also disables gcc's sibling call optimization.
1798 if (bio
->bi_end_io
== bio_chain_endio
) {
1799 struct bio
*parent
= bio
->bi_private
;
1804 bio
->bi_end_io(bio
, error
);
1809 EXPORT_SYMBOL(bio_endio
);
1812 * bio_endio_nodec - end I/O on a bio, without decrementing bi_remaining
1814 * @error: error, if any
1816 * For code that has saved and restored bi_end_io; thing hard before using this
1817 * function, probably you should've cloned the entire bio.
1819 void bio_endio_nodec(struct bio
*bio
, int error
)
1822 * If it's not flagged as a chain, we are not going to dec the count
1824 if (bio_flagged(bio
, BIO_CHAIN
))
1825 bio_inc_remaining(bio
);
1827 bio_endio(bio
, error
);
1829 EXPORT_SYMBOL(bio_endio_nodec
);
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 * The newly allocated bio will point to @bio's bi_io_vec; it is the caller's
1842 * responsibility to ensure that @bio is not freed before the split.
1844 struct bio
*bio_split(struct bio
*bio
, int sectors
,
1845 gfp_t gfp
, struct bio_set
*bs
)
1847 struct bio
*split
= NULL
;
1849 BUG_ON(sectors
<= 0);
1850 BUG_ON(sectors
>= bio_sectors(bio
));
1852 split
= bio_clone_fast(bio
, gfp
, bs
);
1856 split
->bi_iter
.bi_size
= sectors
<< 9;
1858 if (bio_integrity(split
))
1859 bio_integrity_trim(split
, 0, sectors
);
1861 bio_advance(bio
, split
->bi_iter
.bi_size
);
1865 EXPORT_SYMBOL(bio_split
);
1868 * bio_trim - trim a bio
1870 * @offset: number of sectors to trim from the front of @bio
1871 * @size: size we want to trim @bio to, in sectors
1873 void bio_trim(struct bio
*bio
, int offset
, int size
)
1875 /* 'bio' is a cloned bio which we need to trim to match
1876 * the given offset and size.
1880 if (offset
== 0 && size
== bio
->bi_iter
.bi_size
)
1883 clear_bit(BIO_SEG_VALID
, &bio
->bi_flags
);
1885 bio_advance(bio
, offset
<< 9);
1887 bio
->bi_iter
.bi_size
= size
;
1889 EXPORT_SYMBOL_GPL(bio_trim
);
1892 * create memory pools for biovec's in a bio_set.
1893 * use the global biovec slabs created for general use.
1895 mempool_t
*biovec_create_pool(int pool_entries
)
1897 struct biovec_slab
*bp
= bvec_slabs
+ BIOVEC_MAX_IDX
;
1899 return mempool_create_slab_pool(pool_entries
, bp
->slab
);
1902 void bioset_free(struct bio_set
*bs
)
1904 if (bs
->rescue_workqueue
)
1905 destroy_workqueue(bs
->rescue_workqueue
);
1908 mempool_destroy(bs
->bio_pool
);
1911 mempool_destroy(bs
->bvec_pool
);
1913 bioset_integrity_free(bs
);
1918 EXPORT_SYMBOL(bioset_free
);
1920 static struct bio_set
*__bioset_create(unsigned int pool_size
,
1921 unsigned int front_pad
,
1922 bool create_bvec_pool
)
1924 unsigned int back_pad
= BIO_INLINE_VECS
* sizeof(struct bio_vec
);
1927 bs
= kzalloc(sizeof(*bs
), GFP_KERNEL
);
1931 bs
->front_pad
= front_pad
;
1933 spin_lock_init(&bs
->rescue_lock
);
1934 bio_list_init(&bs
->rescue_list
);
1935 INIT_WORK(&bs
->rescue_work
, bio_alloc_rescue
);
1937 bs
->bio_slab
= bio_find_or_create_slab(front_pad
+ back_pad
);
1938 if (!bs
->bio_slab
) {
1943 bs
->bio_pool
= mempool_create_slab_pool(pool_size
, bs
->bio_slab
);
1947 if (create_bvec_pool
) {
1948 bs
->bvec_pool
= biovec_create_pool(pool_size
);
1953 bs
->rescue_workqueue
= alloc_workqueue("bioset", WQ_MEM_RECLAIM
, 0);
1954 if (!bs
->rescue_workqueue
)
1964 * bioset_create - Create a bio_set
1965 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1966 * @front_pad: Number of bytes to allocate in front of the returned bio
1969 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1970 * to ask for a number of bytes to be allocated in front of the bio.
1971 * Front pad allocation is useful for embedding the bio inside
1972 * another structure, to avoid allocating extra data to go with the bio.
1973 * Note that the bio must be embedded at the END of that structure always,
1974 * or things will break badly.
1976 struct bio_set
*bioset_create(unsigned int pool_size
, unsigned int front_pad
)
1978 return __bioset_create(pool_size
, front_pad
, true);
1980 EXPORT_SYMBOL(bioset_create
);
1983 * bioset_create_nobvec - Create a bio_set without bio_vec mempool
1984 * @pool_size: Number of bio to cache in the mempool
1985 * @front_pad: Number of bytes to allocate in front of the returned bio
1988 * Same functionality as bioset_create() except that mempool is not
1989 * created for bio_vecs. Saving some memory for bio_clone_fast() users.
1991 struct bio_set
*bioset_create_nobvec(unsigned int pool_size
, unsigned int front_pad
)
1993 return __bioset_create(pool_size
, front_pad
, false);
1995 EXPORT_SYMBOL(bioset_create_nobvec
);
1997 #ifdef CONFIG_BLK_CGROUP
1999 * bio_associate_current - associate a bio with %current
2002 * Associate @bio with %current if it hasn't been associated yet. Block
2003 * layer will treat @bio as if it were issued by %current no matter which
2004 * task actually issues it.
2006 * This function takes an extra reference of @task's io_context and blkcg
2007 * which will be put when @bio is released. The caller must own @bio,
2008 * ensure %current->io_context exists, and is responsible for synchronizing
2009 * calls to this function.
2011 int bio_associate_current(struct bio
*bio
)
2013 struct io_context
*ioc
;
2014 struct cgroup_subsys_state
*css
;
2019 ioc
= current
->io_context
;
2023 /* acquire active ref on @ioc and associate */
2024 get_io_context_active(ioc
);
2027 /* associate blkcg if exists */
2029 css
= task_css(current
, blkio_cgrp_id
);
2030 if (css
&& css_tryget_online(css
))
2038 * bio_disassociate_task - undo bio_associate_current()
2041 void bio_disassociate_task(struct bio
*bio
)
2044 put_io_context(bio
->bi_ioc
);
2048 css_put(bio
->bi_css
);
2053 #endif /* CONFIG_BLK_CGROUP */
2055 static void __init
biovec_init_slabs(void)
2059 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
2061 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
2063 if (bvs
->nr_vecs
<= BIO_INLINE_VECS
) {
2068 size
= bvs
->nr_vecs
* sizeof(struct bio_vec
);
2069 bvs
->slab
= kmem_cache_create(bvs
->name
, size
, 0,
2070 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
2074 static int __init
init_bio(void)
2078 bio_slabs
= kzalloc(bio_slab_max
* sizeof(struct bio_slab
), GFP_KERNEL
);
2080 panic("bio: can't allocate bios\n");
2082 bio_integrity_init();
2083 biovec_init_slabs();
2085 fs_bio_set
= bioset_create(BIO_POOL_SIZE
, 0);
2087 panic("bio: can't allocate bios\n");
2089 if (bioset_integrity_create(fs_bio_set
, BIO_POOL_SIZE
))
2090 panic("bio: can't create integrity pool\n");
2094 subsys_initcall(init_bio
);