2 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
13 * You should have received a copy of the GNU General Public Licens
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
19 #include <linux/swap.h>
20 #include <linux/bio.h>
21 #include <linux/blkdev.h>
22 #include <linux/uio.h>
23 #include <linux/iocontext.h>
24 #include <linux/slab.h>
25 #include <linux/init.h>
26 #include <linux/kernel.h>
27 #include <linux/export.h>
28 #include <linux/mempool.h>
29 #include <linux/workqueue.h>
30 #include <linux/cgroup.h>
31 #include <scsi/sg.h> /* for struct sg_iovec */
33 #include <trace/events/block.h>
36 * Test patch to inline a certain number of bi_io_vec's inside the bio
37 * itself, to shrink a bio data allocation from two mempool calls to one
39 #define BIO_INLINE_VECS 4
41 static mempool_t
*bio_split_pool __read_mostly
;
44 * if you change this list, also change bvec_alloc or things will
45 * break badly! cannot be bigger than what you can fit into an
48 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
49 static struct biovec_slab bvec_slabs
[BIOVEC_NR_POOLS
] __read_mostly
= {
50 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES
),
55 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
56 * IO code that does not need private memory pools.
58 struct bio_set
*fs_bio_set
;
59 EXPORT_SYMBOL(fs_bio_set
);
62 * Our slab pool management
65 struct kmem_cache
*slab
;
66 unsigned int slab_ref
;
67 unsigned int slab_size
;
70 static DEFINE_MUTEX(bio_slab_lock
);
71 static struct bio_slab
*bio_slabs
;
72 static unsigned int bio_slab_nr
, bio_slab_max
;
74 static struct kmem_cache
*bio_find_or_create_slab(unsigned int extra_size
)
76 unsigned int sz
= sizeof(struct bio
) + extra_size
;
77 struct kmem_cache
*slab
= NULL
;
78 struct bio_slab
*bslab
, *new_bio_slabs
;
79 unsigned int new_bio_slab_max
;
80 unsigned int i
, entry
= -1;
82 mutex_lock(&bio_slab_lock
);
85 while (i
< bio_slab_nr
) {
86 bslab
= &bio_slabs
[i
];
88 if (!bslab
->slab
&& entry
== -1)
90 else if (bslab
->slab_size
== sz
) {
101 if (bio_slab_nr
== bio_slab_max
&& entry
== -1) {
102 new_bio_slab_max
= bio_slab_max
<< 1;
103 new_bio_slabs
= krealloc(bio_slabs
,
104 new_bio_slab_max
* sizeof(struct bio_slab
),
108 bio_slab_max
= new_bio_slab_max
;
109 bio_slabs
= new_bio_slabs
;
112 entry
= bio_slab_nr
++;
114 bslab
= &bio_slabs
[entry
];
116 snprintf(bslab
->name
, sizeof(bslab
->name
), "bio-%d", entry
);
117 slab
= kmem_cache_create(bslab
->name
, sz
, 0, SLAB_HWCACHE_ALIGN
, NULL
);
121 printk(KERN_INFO
"bio: create slab <%s> at %d\n", bslab
->name
, entry
);
124 bslab
->slab_size
= sz
;
126 mutex_unlock(&bio_slab_lock
);
130 static void bio_put_slab(struct bio_set
*bs
)
132 struct bio_slab
*bslab
= NULL
;
135 mutex_lock(&bio_slab_lock
);
137 for (i
= 0; i
< bio_slab_nr
; i
++) {
138 if (bs
->bio_slab
== bio_slabs
[i
].slab
) {
139 bslab
= &bio_slabs
[i
];
144 if (WARN(!bslab
, KERN_ERR
"bio: unable to find slab!\n"))
147 WARN_ON(!bslab
->slab_ref
);
149 if (--bslab
->slab_ref
)
152 kmem_cache_destroy(bslab
->slab
);
156 mutex_unlock(&bio_slab_lock
);
159 unsigned int bvec_nr_vecs(unsigned short idx
)
161 return bvec_slabs
[idx
].nr_vecs
;
164 void bvec_free(mempool_t
*pool
, struct bio_vec
*bv
, unsigned int idx
)
166 BIO_BUG_ON(idx
>= BIOVEC_NR_POOLS
);
168 if (idx
== BIOVEC_MAX_IDX
)
169 mempool_free(bv
, pool
);
171 struct biovec_slab
*bvs
= bvec_slabs
+ idx
;
173 kmem_cache_free(bvs
->slab
, bv
);
177 struct bio_vec
*bvec_alloc(gfp_t gfp_mask
, int nr
, unsigned long *idx
,
183 * see comment near bvec_array define!
201 case 129 ... BIO_MAX_PAGES
:
209 * idx now points to the pool we want to allocate from. only the
210 * 1-vec entry pool is mempool backed.
212 if (*idx
== BIOVEC_MAX_IDX
) {
214 bvl
= mempool_alloc(pool
, gfp_mask
);
216 struct biovec_slab
*bvs
= bvec_slabs
+ *idx
;
217 gfp_t __gfp_mask
= gfp_mask
& ~(__GFP_WAIT
| __GFP_IO
);
220 * Make this allocation restricted and don't dump info on
221 * allocation failures, since we'll fallback to the mempool
222 * in case of failure.
224 __gfp_mask
|= __GFP_NOMEMALLOC
| __GFP_NORETRY
| __GFP_NOWARN
;
227 * Try a slab allocation. If this fails and __GFP_WAIT
228 * is set, retry with the 1-entry mempool
230 bvl
= kmem_cache_alloc(bvs
->slab
, __gfp_mask
);
231 if (unlikely(!bvl
&& (gfp_mask
& __GFP_WAIT
))) {
232 *idx
= BIOVEC_MAX_IDX
;
240 static void __bio_free(struct bio
*bio
)
242 bio_disassociate_task(bio
);
244 if (bio_integrity(bio
))
245 bio_integrity_free(bio
);
248 static void bio_free(struct bio
*bio
)
250 struct bio_set
*bs
= bio
->bi_pool
;
256 if (bio_flagged(bio
, BIO_OWNS_VEC
))
257 bvec_free(bs
->bvec_pool
, bio
->bi_io_vec
, BIO_POOL_IDX(bio
));
260 * If we have front padding, adjust the bio pointer before freeing
265 mempool_free(p
, bs
->bio_pool
);
267 /* Bio was allocated by bio_kmalloc() */
272 void bio_init(struct bio
*bio
)
274 memset(bio
, 0, sizeof(*bio
));
275 bio
->bi_flags
= 1 << BIO_UPTODATE
;
276 atomic_set(&bio
->bi_cnt
, 1);
278 EXPORT_SYMBOL(bio_init
);
281 * bio_reset - reinitialize a bio
285 * After calling bio_reset(), @bio will be in the same state as a freshly
286 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
287 * preserved are the ones that are initialized by bio_alloc_bioset(). See
288 * comment in struct bio.
290 void bio_reset(struct bio
*bio
)
292 unsigned long flags
= bio
->bi_flags
& (~0UL << BIO_RESET_BITS
);
296 memset(bio
, 0, BIO_RESET_BYTES
);
297 bio
->bi_flags
= flags
|(1 << BIO_UPTODATE
);
299 EXPORT_SYMBOL(bio_reset
);
301 static void bio_alloc_rescue(struct work_struct
*work
)
303 struct bio_set
*bs
= container_of(work
, struct bio_set
, rescue_work
);
307 spin_lock(&bs
->rescue_lock
);
308 bio
= bio_list_pop(&bs
->rescue_list
);
309 spin_unlock(&bs
->rescue_lock
);
314 generic_make_request(bio
);
318 static void punt_bios_to_rescuer(struct bio_set
*bs
)
320 struct bio_list punt
, nopunt
;
324 * In order to guarantee forward progress we must punt only bios that
325 * were allocated from this bio_set; otherwise, if there was a bio on
326 * there for a stacking driver higher up in the stack, processing it
327 * could require allocating bios from this bio_set, and doing that from
328 * our own rescuer would be bad.
330 * Since bio lists are singly linked, pop them all instead of trying to
331 * remove from the middle of the list:
334 bio_list_init(&punt
);
335 bio_list_init(&nopunt
);
337 while ((bio
= bio_list_pop(current
->bio_list
)))
338 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
340 *current
->bio_list
= nopunt
;
342 spin_lock(&bs
->rescue_lock
);
343 bio_list_merge(&bs
->rescue_list
, &punt
);
344 spin_unlock(&bs
->rescue_lock
);
346 queue_work(bs
->rescue_workqueue
, &bs
->rescue_work
);
350 * bio_alloc_bioset - allocate a bio for I/O
351 * @gfp_mask: the GFP_ mask given to the slab allocator
352 * @nr_iovecs: number of iovecs to pre-allocate
353 * @bs: the bio_set to allocate from.
356 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
357 * backed by the @bs's mempool.
359 * When @bs is not NULL, if %__GFP_WAIT is set then bio_alloc will always be
360 * able to allocate a bio. This is due to the mempool guarantees. To make this
361 * work, callers must never allocate more than 1 bio at a time from this pool.
362 * Callers that need to allocate more than 1 bio must always submit the
363 * previously allocated bio for IO before attempting to allocate a new one.
364 * Failure to do so can cause deadlocks under memory pressure.
366 * Note that when running under generic_make_request() (i.e. any block
367 * driver), bios are not submitted until after you return - see the code in
368 * generic_make_request() that converts recursion into iteration, to prevent
371 * This would normally mean allocating multiple bios under
372 * generic_make_request() would be susceptible to deadlocks, but we have
373 * deadlock avoidance code that resubmits any blocked bios from a rescuer
376 * However, we do not guarantee forward progress for allocations from other
377 * mempools. Doing multiple allocations from the same mempool under
378 * generic_make_request() should be avoided - instead, use bio_set's front_pad
379 * for per bio allocations.
382 * Pointer to new bio on success, NULL on failure.
384 struct bio
*bio_alloc_bioset(gfp_t gfp_mask
, int nr_iovecs
, struct bio_set
*bs
)
386 gfp_t saved_gfp
= gfp_mask
;
388 unsigned inline_vecs
;
389 unsigned long idx
= BIO_POOL_NONE
;
390 struct bio_vec
*bvl
= NULL
;
395 if (nr_iovecs
> UIO_MAXIOV
)
398 p
= kmalloc(sizeof(struct bio
) +
399 nr_iovecs
* sizeof(struct bio_vec
),
402 inline_vecs
= nr_iovecs
;
405 * generic_make_request() converts recursion to iteration; this
406 * means if we're running beneath it, any bios we allocate and
407 * submit will not be submitted (and thus freed) until after we
410 * This exposes us to a potential deadlock if we allocate
411 * multiple bios from the same bio_set() while running
412 * underneath generic_make_request(). If we were to allocate
413 * multiple bios (say a stacking block driver that was splitting
414 * bios), we would deadlock if we exhausted the mempool's
417 * We solve this, and guarantee forward progress, with a rescuer
418 * workqueue per bio_set. If we go to allocate and there are
419 * bios on current->bio_list, we first try the allocation
420 * without __GFP_WAIT; if that fails, we punt those bios we
421 * would be blocking to the rescuer workqueue before we retry
422 * with the original gfp_flags.
425 if (current
->bio_list
&& !bio_list_empty(current
->bio_list
))
426 gfp_mask
&= ~__GFP_WAIT
;
428 p
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
429 if (!p
&& gfp_mask
!= saved_gfp
) {
430 punt_bios_to_rescuer(bs
);
431 gfp_mask
= saved_gfp
;
432 p
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
435 front_pad
= bs
->front_pad
;
436 inline_vecs
= BIO_INLINE_VECS
;
445 if (nr_iovecs
> inline_vecs
) {
446 bvl
= bvec_alloc(gfp_mask
, nr_iovecs
, &idx
, bs
->bvec_pool
);
447 if (!bvl
&& gfp_mask
!= saved_gfp
) {
448 punt_bios_to_rescuer(bs
);
449 gfp_mask
= saved_gfp
;
450 bvl
= bvec_alloc(gfp_mask
, nr_iovecs
, &idx
, bs
->bvec_pool
);
456 bio
->bi_flags
|= 1 << BIO_OWNS_VEC
;
457 } else if (nr_iovecs
) {
458 bvl
= bio
->bi_inline_vecs
;
462 bio
->bi_flags
|= idx
<< BIO_POOL_OFFSET
;
463 bio
->bi_max_vecs
= nr_iovecs
;
464 bio
->bi_io_vec
= bvl
;
468 mempool_free(p
, bs
->bio_pool
);
471 EXPORT_SYMBOL(bio_alloc_bioset
);
473 void zero_fill_bio(struct bio
*bio
)
477 struct bvec_iter iter
;
479 bio_for_each_segment(bv
, bio
, iter
) {
480 char *data
= bvec_kmap_irq(&bv
, &flags
);
481 memset(data
, 0, bv
.bv_len
);
482 flush_dcache_page(bv
.bv_page
);
483 bvec_kunmap_irq(data
, &flags
);
486 EXPORT_SYMBOL(zero_fill_bio
);
489 * bio_put - release a reference to a bio
490 * @bio: bio to release reference to
493 * Put a reference to a &struct bio, either one you have gotten with
494 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
496 void bio_put(struct bio
*bio
)
498 BIO_BUG_ON(!atomic_read(&bio
->bi_cnt
));
503 if (atomic_dec_and_test(&bio
->bi_cnt
))
506 EXPORT_SYMBOL(bio_put
);
508 inline int bio_phys_segments(struct request_queue
*q
, struct bio
*bio
)
510 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
511 blk_recount_segments(q
, bio
);
513 return bio
->bi_phys_segments
;
515 EXPORT_SYMBOL(bio_phys_segments
);
518 * __bio_clone_fast - clone a bio that shares the original bio's biovec
519 * @bio: destination bio
520 * @bio_src: bio to clone
522 * Clone a &bio. Caller will own the returned bio, but not
523 * the actual data it points to. Reference count of returned
526 * Caller must ensure that @bio_src is not freed before @bio.
528 void __bio_clone_fast(struct bio
*bio
, struct bio
*bio_src
)
530 BUG_ON(bio
->bi_pool
&& BIO_POOL_IDX(bio
) != BIO_POOL_NONE
);
533 * most users will be overriding ->bi_bdev with a new target,
534 * so we don't set nor calculate new physical/hw segment counts here
536 bio
->bi_bdev
= bio_src
->bi_bdev
;
537 bio
->bi_flags
|= 1 << BIO_CLONED
;
538 bio
->bi_rw
= bio_src
->bi_rw
;
539 bio
->bi_iter
= bio_src
->bi_iter
;
540 bio
->bi_io_vec
= bio_src
->bi_io_vec
;
542 EXPORT_SYMBOL(__bio_clone_fast
);
545 * bio_clone_fast - clone a bio that shares the original bio's biovec
547 * @gfp_mask: allocation priority
548 * @bs: bio_set to allocate from
550 * Like __bio_clone_fast, only also allocates the returned bio
552 struct bio
*bio_clone_fast(struct bio
*bio
, gfp_t gfp_mask
, struct bio_set
*bs
)
556 b
= bio_alloc_bioset(gfp_mask
, 0, bs
);
560 __bio_clone_fast(b
, bio
);
562 if (bio_integrity(bio
)) {
565 ret
= bio_integrity_clone(b
, bio
, gfp_mask
);
575 EXPORT_SYMBOL(bio_clone_fast
);
578 * bio_clone_bioset - clone a bio
579 * @bio_src: bio to clone
580 * @gfp_mask: allocation priority
581 * @bs: bio_set to allocate from
583 * Clone bio. Caller will own the returned bio, but not the actual data it
584 * points to. Reference count of returned bio will be one.
586 struct bio
*bio_clone_bioset(struct bio
*bio_src
, gfp_t gfp_mask
,
589 unsigned nr_iovecs
= 0;
590 struct bvec_iter iter
;
595 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
596 * bio_src->bi_io_vec to bio->bi_io_vec.
598 * We can't do that anymore, because:
600 * - The point of cloning the biovec is to produce a bio with a biovec
601 * the caller can modify: bi_idx and bi_bvec_done should be 0.
603 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
604 * we tried to clone the whole thing bio_alloc_bioset() would fail.
605 * But the clone should succeed as long as the number of biovecs we
606 * actually need to allocate is fewer than BIO_MAX_PAGES.
608 * - Lastly, bi_vcnt should not be looked at or relied upon by code
609 * that does not own the bio - reason being drivers don't use it for
610 * iterating over the biovec anymore, so expecting it to be kept up
611 * to date (i.e. for clones that share the parent biovec) is just
612 * asking for trouble and would force extra work on
613 * __bio_clone_fast() anyways.
616 bio_for_each_segment(bv
, bio_src
, iter
)
619 bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, bs
);
623 bio
->bi_bdev
= bio_src
->bi_bdev
;
624 bio
->bi_rw
= bio_src
->bi_rw
;
625 bio
->bi_iter
.bi_sector
= bio_src
->bi_iter
.bi_sector
;
626 bio
->bi_iter
.bi_size
= bio_src
->bi_iter
.bi_size
;
628 bio_for_each_segment(bv
, bio_src
, iter
)
629 bio
->bi_io_vec
[bio
->bi_vcnt
++] = bv
;
631 if (bio_integrity(bio_src
)) {
634 ret
= bio_integrity_clone(bio
, bio_src
, gfp_mask
);
643 EXPORT_SYMBOL(bio_clone_bioset
);
646 * bio_get_nr_vecs - return approx number of vecs
649 * Return the approximate number of pages we can send to this target.
650 * There's no guarantee that you will be able to fit this number of pages
651 * into a bio, it does not account for dynamic restrictions that vary
654 int bio_get_nr_vecs(struct block_device
*bdev
)
656 struct request_queue
*q
= bdev_get_queue(bdev
);
659 nr_pages
= min_t(unsigned,
660 queue_max_segments(q
),
661 queue_max_sectors(q
) / (PAGE_SIZE
>> 9) + 1);
663 return min_t(unsigned, nr_pages
, BIO_MAX_PAGES
);
666 EXPORT_SYMBOL(bio_get_nr_vecs
);
668 static int __bio_add_page(struct request_queue
*q
, struct bio
*bio
, struct page
669 *page
, unsigned int len
, unsigned int offset
,
670 unsigned int max_sectors
)
672 int retried_segments
= 0;
673 struct bio_vec
*bvec
;
676 * cloned bio must not modify vec list
678 if (unlikely(bio_flagged(bio
, BIO_CLONED
)))
681 if (((bio
->bi_iter
.bi_size
+ len
) >> 9) > max_sectors
)
685 * For filesystems with a blocksize smaller than the pagesize
686 * we will often be called with the same page as last time and
687 * a consecutive offset. Optimize this special case.
689 if (bio
->bi_vcnt
> 0) {
690 struct bio_vec
*prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
692 if (page
== prev
->bv_page
&&
693 offset
== prev
->bv_offset
+ prev
->bv_len
) {
694 unsigned int prev_bv_len
= prev
->bv_len
;
697 if (q
->merge_bvec_fn
) {
698 struct bvec_merge_data bvm
= {
699 /* prev_bvec is already charged in
700 bi_size, discharge it in order to
701 simulate merging updated prev_bvec
703 .bi_bdev
= bio
->bi_bdev
,
704 .bi_sector
= bio
->bi_iter
.bi_sector
,
705 .bi_size
= bio
->bi_iter
.bi_size
-
710 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < prev
->bv_len
) {
720 if (bio
->bi_vcnt
>= bio
->bi_max_vecs
)
724 * we might lose a segment or two here, but rather that than
725 * make this too complex.
728 while (bio
->bi_phys_segments
>= queue_max_segments(q
)) {
730 if (retried_segments
)
733 retried_segments
= 1;
734 blk_recount_segments(q
, bio
);
738 * setup the new entry, we might clear it again later if we
739 * cannot add the page
741 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
742 bvec
->bv_page
= page
;
744 bvec
->bv_offset
= offset
;
747 * if queue has other restrictions (eg varying max sector size
748 * depending on offset), it can specify a merge_bvec_fn in the
749 * queue to get further control
751 if (q
->merge_bvec_fn
) {
752 struct bvec_merge_data bvm
= {
753 .bi_bdev
= bio
->bi_bdev
,
754 .bi_sector
= bio
->bi_iter
.bi_sector
,
755 .bi_size
= bio
->bi_iter
.bi_size
,
760 * merge_bvec_fn() returns number of bytes it can accept
763 if (q
->merge_bvec_fn(q
, &bvm
, bvec
) < bvec
->bv_len
) {
764 bvec
->bv_page
= NULL
;
771 /* If we may be able to merge these biovecs, force a recount */
772 if (bio
->bi_vcnt
&& (BIOVEC_PHYS_MERGEABLE(bvec
-1, bvec
)))
773 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
776 bio
->bi_phys_segments
++;
778 bio
->bi_iter
.bi_size
+= len
;
783 * bio_add_pc_page - attempt to add page to bio
784 * @q: the target queue
785 * @bio: destination bio
787 * @len: vec entry length
788 * @offset: vec entry offset
790 * Attempt to add a page to the bio_vec maplist. This can fail for a
791 * number of reasons, such as the bio being full or target block device
792 * limitations. The target block device must allow bio's up to PAGE_SIZE,
793 * so it is always possible to add a single page to an empty bio.
795 * This should only be used by REQ_PC bios.
797 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
, struct page
*page
,
798 unsigned int len
, unsigned int offset
)
800 return __bio_add_page(q
, bio
, page
, len
, offset
,
801 queue_max_hw_sectors(q
));
803 EXPORT_SYMBOL(bio_add_pc_page
);
806 * bio_add_page - attempt to add page to bio
807 * @bio: destination bio
809 * @len: vec entry length
810 * @offset: vec entry offset
812 * Attempt to add a page to the bio_vec maplist. This can fail for a
813 * number of reasons, such as the bio being full or target block device
814 * limitations. The target block device must allow bio's up to PAGE_SIZE,
815 * so it is always possible to add a single page to an empty bio.
817 int bio_add_page(struct bio
*bio
, struct page
*page
, unsigned int len
,
820 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
821 return __bio_add_page(q
, bio
, page
, len
, offset
, queue_max_sectors(q
));
823 EXPORT_SYMBOL(bio_add_page
);
825 struct submit_bio_ret
{
826 struct completion event
;
830 static void submit_bio_wait_endio(struct bio
*bio
, int error
)
832 struct submit_bio_ret
*ret
= bio
->bi_private
;
835 complete(&ret
->event
);
839 * submit_bio_wait - submit a bio, and wait until it completes
840 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
841 * @bio: The &struct bio which describes the I/O
843 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
844 * bio_endio() on failure.
846 int submit_bio_wait(int rw
, struct bio
*bio
)
848 struct submit_bio_ret ret
;
851 init_completion(&ret
.event
);
852 bio
->bi_private
= &ret
;
853 bio
->bi_end_io
= submit_bio_wait_endio
;
855 wait_for_completion(&ret
.event
);
859 EXPORT_SYMBOL(submit_bio_wait
);
862 * bio_advance - increment/complete a bio by some number of bytes
863 * @bio: bio to advance
864 * @bytes: number of bytes to complete
866 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
867 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
868 * be updated on the last bvec as well.
870 * @bio will then represent the remaining, uncompleted portion of the io.
872 void bio_advance(struct bio
*bio
, unsigned bytes
)
874 if (bio_integrity(bio
))
875 bio_integrity_advance(bio
, bytes
);
877 bio_advance_iter(bio
, &bio
->bi_iter
, bytes
);
879 EXPORT_SYMBOL(bio_advance
);
882 * bio_alloc_pages - allocates a single page for each bvec in a bio
883 * @bio: bio to allocate pages for
884 * @gfp_mask: flags for allocation
886 * Allocates pages up to @bio->bi_vcnt.
888 * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
891 int bio_alloc_pages(struct bio
*bio
, gfp_t gfp_mask
)
896 bio_for_each_segment_all(bv
, bio
, i
) {
897 bv
->bv_page
= alloc_page(gfp_mask
);
899 while (--bv
>= bio
->bi_io_vec
)
900 __free_page(bv
->bv_page
);
907 EXPORT_SYMBOL(bio_alloc_pages
);
910 * bio_copy_data - copy contents of data buffers from one chain of bios to
912 * @src: source bio list
913 * @dst: destination bio list
915 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
916 * @src and @dst as linked lists of bios.
918 * Stops when it reaches the end of either @src or @dst - that is, copies
919 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
921 void bio_copy_data(struct bio
*dst
, struct bio
*src
)
923 struct bvec_iter src_iter
, dst_iter
;
924 struct bio_vec src_bv
, dst_bv
;
928 src_iter
= src
->bi_iter
;
929 dst_iter
= dst
->bi_iter
;
932 if (!src_iter
.bi_size
) {
937 src_iter
= src
->bi_iter
;
940 if (!dst_iter
.bi_size
) {
945 dst_iter
= dst
->bi_iter
;
948 src_bv
= bio_iter_iovec(src
, src_iter
);
949 dst_bv
= bio_iter_iovec(dst
, dst_iter
);
951 bytes
= min(src_bv
.bv_len
, dst_bv
.bv_len
);
953 src_p
= kmap_atomic(src_bv
.bv_page
);
954 dst_p
= kmap_atomic(dst_bv
.bv_page
);
956 memcpy(dst_p
+ dst_bv
.bv_offset
,
957 src_p
+ src_bv
.bv_offset
,
960 kunmap_atomic(dst_p
);
961 kunmap_atomic(src_p
);
963 bio_advance_iter(src
, &src_iter
, bytes
);
964 bio_advance_iter(dst
, &dst_iter
, bytes
);
967 EXPORT_SYMBOL(bio_copy_data
);
969 struct bio_map_data
{
972 struct sg_iovec sgvecs
[];
975 static void bio_set_map_data(struct bio_map_data
*bmd
, struct bio
*bio
,
976 struct sg_iovec
*iov
, int iov_count
,
979 memcpy(bmd
->sgvecs
, iov
, sizeof(struct sg_iovec
) * iov_count
);
980 bmd
->nr_sgvecs
= iov_count
;
981 bmd
->is_our_pages
= is_our_pages
;
982 bio
->bi_private
= bmd
;
985 static struct bio_map_data
*bio_alloc_map_data(int nr_segs
,
986 unsigned int iov_count
,
989 if (iov_count
> UIO_MAXIOV
)
992 return kmalloc(sizeof(struct bio_map_data
) +
993 sizeof(struct sg_iovec
) * iov_count
, gfp_mask
);
996 static int __bio_copy_iov(struct bio
*bio
, struct sg_iovec
*iov
, int iov_count
,
997 int to_user
, int from_user
, int do_free_page
)
1000 struct bio_vec
*bvec
;
1002 unsigned int iov_off
= 0;
1004 bio_for_each_segment_all(bvec
, bio
, i
) {
1005 char *bv_addr
= page_address(bvec
->bv_page
);
1006 unsigned int bv_len
= bvec
->bv_len
;
1008 while (bv_len
&& iov_idx
< iov_count
) {
1010 char __user
*iov_addr
;
1012 bytes
= min_t(unsigned int,
1013 iov
[iov_idx
].iov_len
- iov_off
, bv_len
);
1014 iov_addr
= iov
[iov_idx
].iov_base
+ iov_off
;
1018 ret
= copy_to_user(iov_addr
, bv_addr
,
1022 ret
= copy_from_user(bv_addr
, iov_addr
,
1034 if (iov
[iov_idx
].iov_len
== iov_off
) {
1041 __free_page(bvec
->bv_page
);
1048 * bio_uncopy_user - finish previously mapped bio
1049 * @bio: bio being terminated
1051 * Free pages allocated from bio_copy_user() and write back data
1052 * to user space in case of a read.
1054 int bio_uncopy_user(struct bio
*bio
)
1056 struct bio_map_data
*bmd
= bio
->bi_private
;
1057 struct bio_vec
*bvec
;
1060 if (!bio_flagged(bio
, BIO_NULL_MAPPED
)) {
1062 * if we're in a workqueue, the request is orphaned, so
1063 * don't copy into a random user address space, just free.
1066 ret
= __bio_copy_iov(bio
, bmd
->sgvecs
, bmd
->nr_sgvecs
,
1067 bio_data_dir(bio
) == READ
,
1068 0, bmd
->is_our_pages
);
1069 else if (bmd
->is_our_pages
)
1070 bio_for_each_segment_all(bvec
, bio
, i
)
1071 __free_page(bvec
->bv_page
);
1077 EXPORT_SYMBOL(bio_uncopy_user
);
1080 * bio_copy_user_iov - copy user data to bio
1081 * @q: destination block queue
1082 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1084 * @iov_count: number of elements in the iovec
1085 * @write_to_vm: bool indicating writing to pages or not
1086 * @gfp_mask: memory allocation flags
1088 * Prepares and returns a bio for indirect user io, bouncing data
1089 * to/from kernel pages as necessary. Must be paired with
1090 * call bio_uncopy_user() on io completion.
1092 struct bio
*bio_copy_user_iov(struct request_queue
*q
,
1093 struct rq_map_data
*map_data
,
1094 struct sg_iovec
*iov
, int iov_count
,
1095 int write_to_vm
, gfp_t gfp_mask
)
1097 struct bio_map_data
*bmd
;
1098 struct bio_vec
*bvec
;
1103 unsigned int len
= 0;
1104 unsigned int offset
= map_data
? map_data
->offset
& ~PAGE_MASK
: 0;
1106 for (i
= 0; i
< iov_count
; i
++) {
1107 unsigned long uaddr
;
1109 unsigned long start
;
1111 uaddr
= (unsigned long)iov
[i
].iov_base
;
1112 end
= (uaddr
+ iov
[i
].iov_len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1113 start
= uaddr
>> PAGE_SHIFT
;
1119 return ERR_PTR(-EINVAL
);
1121 nr_pages
+= end
- start
;
1122 len
+= iov
[i
].iov_len
;
1128 bmd
= bio_alloc_map_data(nr_pages
, iov_count
, gfp_mask
);
1130 return ERR_PTR(-ENOMEM
);
1133 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1138 bio
->bi_rw
|= REQ_WRITE
;
1143 nr_pages
= 1 << map_data
->page_order
;
1144 i
= map_data
->offset
/ PAGE_SIZE
;
1147 unsigned int bytes
= PAGE_SIZE
;
1155 if (i
== map_data
->nr_entries
* nr_pages
) {
1160 page
= map_data
->pages
[i
/ nr_pages
];
1161 page
+= (i
% nr_pages
);
1165 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
1172 if (bio_add_pc_page(q
, bio
, page
, bytes
, offset
) < bytes
)
1185 if ((!write_to_vm
&& (!map_data
|| !map_data
->null_mapped
)) ||
1186 (map_data
&& map_data
->from_user
)) {
1187 ret
= __bio_copy_iov(bio
, iov
, iov_count
, 0, 1, 0);
1192 bio_set_map_data(bmd
, bio
, iov
, iov_count
, map_data
? 0 : 1);
1196 bio_for_each_segment_all(bvec
, bio
, i
)
1197 __free_page(bvec
->bv_page
);
1202 return ERR_PTR(ret
);
1206 * bio_copy_user - copy user data to bio
1207 * @q: destination block queue
1208 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1209 * @uaddr: start of user address
1210 * @len: length in bytes
1211 * @write_to_vm: bool indicating writing to pages or not
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(struct request_queue
*q
, struct rq_map_data
*map_data
,
1219 unsigned long uaddr
, unsigned int len
,
1220 int write_to_vm
, gfp_t gfp_mask
)
1222 struct sg_iovec iov
;
1224 iov
.iov_base
= (void __user
*)uaddr
;
1227 return bio_copy_user_iov(q
, map_data
, &iov
, 1, write_to_vm
, gfp_mask
);
1229 EXPORT_SYMBOL(bio_copy_user
);
1231 static struct bio
*__bio_map_user_iov(struct request_queue
*q
,
1232 struct block_device
*bdev
,
1233 struct sg_iovec
*iov
, int iov_count
,
1234 int write_to_vm
, gfp_t gfp_mask
)
1238 struct page
**pages
;
1243 for (i
= 0; i
< iov_count
; i
++) {
1244 unsigned long uaddr
= (unsigned long)iov
[i
].iov_base
;
1245 unsigned long len
= iov
[i
].iov_len
;
1246 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1247 unsigned long start
= uaddr
>> PAGE_SHIFT
;
1253 return ERR_PTR(-EINVAL
);
1255 nr_pages
+= end
- start
;
1257 * buffer must be aligned to at least hardsector size for now
1259 if (uaddr
& queue_dma_alignment(q
))
1260 return ERR_PTR(-EINVAL
);
1264 return ERR_PTR(-EINVAL
);
1266 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1268 return ERR_PTR(-ENOMEM
);
1271 pages
= kcalloc(nr_pages
, sizeof(struct page
*), gfp_mask
);
1275 for (i
= 0; i
< iov_count
; i
++) {
1276 unsigned long uaddr
= (unsigned long)iov
[i
].iov_base
;
1277 unsigned long len
= iov
[i
].iov_len
;
1278 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1279 unsigned long start
= uaddr
>> PAGE_SHIFT
;
1280 const int local_nr_pages
= end
- start
;
1281 const int page_limit
= cur_page
+ local_nr_pages
;
1283 ret
= get_user_pages_fast(uaddr
, local_nr_pages
,
1284 write_to_vm
, &pages
[cur_page
]);
1285 if (ret
< local_nr_pages
) {
1290 offset
= uaddr
& ~PAGE_MASK
;
1291 for (j
= cur_page
; j
< page_limit
; j
++) {
1292 unsigned int bytes
= PAGE_SIZE
- offset
;
1303 if (bio_add_pc_page(q
, bio
, pages
[j
], bytes
, offset
) <
1313 * release the pages we didn't map into the bio, if any
1315 while (j
< page_limit
)
1316 page_cache_release(pages
[j
++]);
1322 * set data direction, and check if mapped pages need bouncing
1325 bio
->bi_rw
|= REQ_WRITE
;
1327 bio
->bi_bdev
= bdev
;
1328 bio
->bi_flags
|= (1 << BIO_USER_MAPPED
);
1332 for (i
= 0; i
< nr_pages
; i
++) {
1335 page_cache_release(pages
[i
]);
1340 return ERR_PTR(ret
);
1344 * bio_map_user - map user address into bio
1345 * @q: the struct request_queue for the bio
1346 * @bdev: destination block device
1347 * @uaddr: start of user address
1348 * @len: length in bytes
1349 * @write_to_vm: bool indicating writing to pages or not
1350 * @gfp_mask: memory allocation flags
1352 * Map the user space address into a bio suitable for io to a block
1353 * device. Returns an error pointer in case of error.
1355 struct bio
*bio_map_user(struct request_queue
*q
, struct block_device
*bdev
,
1356 unsigned long uaddr
, unsigned int len
, int write_to_vm
,
1359 struct sg_iovec iov
;
1361 iov
.iov_base
= (void __user
*)uaddr
;
1364 return bio_map_user_iov(q
, bdev
, &iov
, 1, write_to_vm
, gfp_mask
);
1366 EXPORT_SYMBOL(bio_map_user
);
1369 * bio_map_user_iov - map user sg_iovec table into bio
1370 * @q: the struct request_queue for the bio
1371 * @bdev: destination block device
1373 * @iov_count: number of elements in the iovec
1374 * @write_to_vm: bool indicating writing to pages or not
1375 * @gfp_mask: memory allocation flags
1377 * Map the user space address into a bio suitable for io to a block
1378 * device. Returns an error pointer in case of error.
1380 struct bio
*bio_map_user_iov(struct request_queue
*q
, struct block_device
*bdev
,
1381 struct sg_iovec
*iov
, int iov_count
,
1382 int write_to_vm
, gfp_t gfp_mask
)
1386 bio
= __bio_map_user_iov(q
, bdev
, iov
, iov_count
, write_to_vm
,
1392 * subtle -- if __bio_map_user() ended up bouncing a bio,
1393 * it would normally disappear when its bi_end_io is run.
1394 * however, we need it for the unmap, so grab an extra
1402 static void __bio_unmap_user(struct bio
*bio
)
1404 struct bio_vec
*bvec
;
1408 * make sure we dirty pages we wrote to
1410 bio_for_each_segment_all(bvec
, bio
, i
) {
1411 if (bio_data_dir(bio
) == READ
)
1412 set_page_dirty_lock(bvec
->bv_page
);
1414 page_cache_release(bvec
->bv_page
);
1421 * bio_unmap_user - unmap a bio
1422 * @bio: the bio being unmapped
1424 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1425 * a process context.
1427 * bio_unmap_user() may sleep.
1429 void bio_unmap_user(struct bio
*bio
)
1431 __bio_unmap_user(bio
);
1434 EXPORT_SYMBOL(bio_unmap_user
);
1436 static void bio_map_kern_endio(struct bio
*bio
, int err
)
1441 static struct bio
*__bio_map_kern(struct request_queue
*q
, void *data
,
1442 unsigned int len
, gfp_t gfp_mask
)
1444 unsigned long kaddr
= (unsigned long)data
;
1445 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1446 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1447 const int nr_pages
= end
- start
;
1451 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1453 return ERR_PTR(-ENOMEM
);
1455 offset
= offset_in_page(kaddr
);
1456 for (i
= 0; i
< nr_pages
; i
++) {
1457 unsigned int bytes
= PAGE_SIZE
- offset
;
1465 if (bio_add_pc_page(q
, bio
, virt_to_page(data
), bytes
,
1474 bio
->bi_end_io
= bio_map_kern_endio
;
1479 * bio_map_kern - map kernel address into bio
1480 * @q: the struct request_queue for the bio
1481 * @data: pointer to buffer to map
1482 * @len: length in bytes
1483 * @gfp_mask: allocation flags for bio allocation
1485 * Map the kernel address into a bio suitable for io to a block
1486 * device. Returns an error pointer in case of error.
1488 struct bio
*bio_map_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1493 bio
= __bio_map_kern(q
, data
, len
, gfp_mask
);
1497 if (bio
->bi_iter
.bi_size
== len
)
1501 * Don't support partial mappings.
1504 return ERR_PTR(-EINVAL
);
1506 EXPORT_SYMBOL(bio_map_kern
);
1508 static void bio_copy_kern_endio(struct bio
*bio
, int err
)
1510 struct bio_vec
*bvec
;
1511 const int read
= bio_data_dir(bio
) == READ
;
1512 struct bio_map_data
*bmd
= bio
->bi_private
;
1514 char *p
= bmd
->sgvecs
[0].iov_base
;
1516 bio_for_each_segment_all(bvec
, bio
, i
) {
1517 char *addr
= page_address(bvec
->bv_page
);
1520 memcpy(p
, addr
, bvec
->bv_len
);
1522 __free_page(bvec
->bv_page
);
1531 * bio_copy_kern - copy kernel address into bio
1532 * @q: the struct request_queue for the bio
1533 * @data: pointer to buffer to copy
1534 * @len: length in bytes
1535 * @gfp_mask: allocation flags for bio and page allocation
1536 * @reading: data direction is READ
1538 * copy the kernel address into a bio suitable for io to a block
1539 * device. Returns an error pointer in case of error.
1541 struct bio
*bio_copy_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1542 gfp_t gfp_mask
, int reading
)
1545 struct bio_vec
*bvec
;
1548 bio
= bio_copy_user(q
, NULL
, (unsigned long)data
, len
, 1, gfp_mask
);
1555 bio_for_each_segment_all(bvec
, bio
, i
) {
1556 char *addr
= page_address(bvec
->bv_page
);
1558 memcpy(addr
, p
, bvec
->bv_len
);
1563 bio
->bi_end_io
= bio_copy_kern_endio
;
1567 EXPORT_SYMBOL(bio_copy_kern
);
1570 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1571 * for performing direct-IO in BIOs.
1573 * The problem is that we cannot run set_page_dirty() from interrupt context
1574 * because the required locks are not interrupt-safe. So what we can do is to
1575 * mark the pages dirty _before_ performing IO. And in interrupt context,
1576 * check that the pages are still dirty. If so, fine. If not, redirty them
1577 * in process context.
1579 * We special-case compound pages here: normally this means reads into hugetlb
1580 * pages. The logic in here doesn't really work right for compound pages
1581 * because the VM does not uniformly chase down the head page in all cases.
1582 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1583 * handle them at all. So we skip compound pages here at an early stage.
1585 * Note that this code is very hard to test under normal circumstances because
1586 * direct-io pins the pages with get_user_pages(). This makes
1587 * is_page_cache_freeable return false, and the VM will not clean the pages.
1588 * But other code (eg, flusher threads) could clean the pages if they are mapped
1591 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1592 * deferred bio dirtying paths.
1596 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1598 void bio_set_pages_dirty(struct bio
*bio
)
1600 struct bio_vec
*bvec
;
1603 bio_for_each_segment_all(bvec
, bio
, i
) {
1604 struct page
*page
= bvec
->bv_page
;
1606 if (page
&& !PageCompound(page
))
1607 set_page_dirty_lock(page
);
1611 static void bio_release_pages(struct bio
*bio
)
1613 struct bio_vec
*bvec
;
1616 bio_for_each_segment_all(bvec
, bio
, i
) {
1617 struct page
*page
= bvec
->bv_page
;
1625 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1626 * If they are, then fine. If, however, some pages are clean then they must
1627 * have been written out during the direct-IO read. So we take another ref on
1628 * the BIO and the offending pages and re-dirty the pages in process context.
1630 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1631 * here on. It will run one page_cache_release() against each page and will
1632 * run one bio_put() against the BIO.
1635 static void bio_dirty_fn(struct work_struct
*work
);
1637 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
1638 static DEFINE_SPINLOCK(bio_dirty_lock
);
1639 static struct bio
*bio_dirty_list
;
1642 * This runs in process context
1644 static void bio_dirty_fn(struct work_struct
*work
)
1646 unsigned long flags
;
1649 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1650 bio
= bio_dirty_list
;
1651 bio_dirty_list
= NULL
;
1652 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1655 struct bio
*next
= bio
->bi_private
;
1657 bio_set_pages_dirty(bio
);
1658 bio_release_pages(bio
);
1664 void bio_check_pages_dirty(struct bio
*bio
)
1666 struct bio_vec
*bvec
;
1667 int nr_clean_pages
= 0;
1670 bio_for_each_segment_all(bvec
, bio
, i
) {
1671 struct page
*page
= bvec
->bv_page
;
1673 if (PageDirty(page
) || PageCompound(page
)) {
1674 page_cache_release(page
);
1675 bvec
->bv_page
= NULL
;
1681 if (nr_clean_pages
) {
1682 unsigned long flags
;
1684 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1685 bio
->bi_private
= bio_dirty_list
;
1686 bio_dirty_list
= bio
;
1687 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1688 schedule_work(&bio_dirty_work
);
1694 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1695 void bio_flush_dcache_pages(struct bio
*bi
)
1697 struct bio_vec bvec
;
1698 struct bvec_iter iter
;
1700 bio_for_each_segment(bvec
, bi
, iter
)
1701 flush_dcache_page(bvec
.bv_page
);
1703 EXPORT_SYMBOL(bio_flush_dcache_pages
);
1707 * bio_endio - end I/O on a bio
1709 * @error: error, if any
1712 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1713 * preferred way to end I/O on a bio, it takes care of clearing
1714 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1715 * established -Exxxx (-EIO, for instance) error values in case
1716 * something went wrong. No one should call bi_end_io() directly on a
1717 * bio unless they own it and thus know that it has an end_io
1720 void bio_endio(struct bio
*bio
, int error
)
1723 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1724 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
1728 bio
->bi_end_io(bio
, error
);
1730 EXPORT_SYMBOL(bio_endio
);
1732 void bio_pair_release(struct bio_pair
*bp
)
1734 if (atomic_dec_and_test(&bp
->cnt
)) {
1735 struct bio
*master
= bp
->bio1
.bi_private
;
1737 bio_endio(master
, bp
->error
);
1738 mempool_free(bp
, bp
->bio2
.bi_private
);
1741 EXPORT_SYMBOL(bio_pair_release
);
1743 static void bio_pair_end_1(struct bio
*bi
, int err
)
1745 struct bio_pair
*bp
= container_of(bi
, struct bio_pair
, bio1
);
1750 bio_pair_release(bp
);
1753 static void bio_pair_end_2(struct bio
*bi
, int err
)
1755 struct bio_pair
*bp
= container_of(bi
, struct bio_pair
, bio2
);
1760 bio_pair_release(bp
);
1764 * split a bio - only worry about a bio with a single page in its iovec
1766 struct bio_pair
*bio_split(struct bio
*bi
, int first_sectors
)
1768 struct bio_pair
*bp
= mempool_alloc(bio_split_pool
, GFP_NOIO
);
1773 trace_block_split(bdev_get_queue(bi
->bi_bdev
), bi
,
1774 bi
->bi_iter
.bi_sector
+ first_sectors
);
1776 BUG_ON(bio_multiple_segments(bi
));
1777 atomic_set(&bp
->cnt
, 3);
1781 bp
->bio2
.bi_iter
.bi_sector
+= first_sectors
;
1782 bp
->bio2
.bi_iter
.bi_size
-= first_sectors
<< 9;
1783 bp
->bio1
.bi_iter
.bi_size
= first_sectors
<< 9;
1785 if (bi
->bi_vcnt
!= 0) {
1786 bp
->bv1
= bio_iovec(bi
);
1787 bp
->bv2
= bio_iovec(bi
);
1789 if (bio_is_rw(bi
)) {
1790 bp
->bv2
.bv_offset
+= first_sectors
<< 9;
1791 bp
->bv2
.bv_len
-= first_sectors
<< 9;
1792 bp
->bv1
.bv_len
= first_sectors
<< 9;
1795 bp
->bio1
.bi_io_vec
= &bp
->bv1
;
1796 bp
->bio2
.bi_io_vec
= &bp
->bv2
;
1798 bp
->bio1
.bi_max_vecs
= 1;
1799 bp
->bio2
.bi_max_vecs
= 1;
1802 bp
->bio1
.bi_end_io
= bio_pair_end_1
;
1803 bp
->bio2
.bi_end_io
= bio_pair_end_2
;
1805 bp
->bio1
.bi_private
= bi
;
1806 bp
->bio2
.bi_private
= bio_split_pool
;
1808 if (bio_integrity(bi
))
1809 bio_integrity_split(bi
, bp
, first_sectors
);
1813 EXPORT_SYMBOL(bio_split
);
1816 * bio_trim - trim a bio
1818 * @offset: number of sectors to trim from the front of @bio
1819 * @size: size we want to trim @bio to, in sectors
1821 void bio_trim(struct bio
*bio
, int offset
, int size
)
1823 /* 'bio' is a cloned bio which we need to trim to match
1824 * the given offset and size.
1828 if (offset
== 0 && size
== bio
->bi_iter
.bi_size
)
1831 clear_bit(BIO_SEG_VALID
, &bio
->bi_flags
);
1833 bio_advance(bio
, offset
<< 9);
1835 bio
->bi_iter
.bi_size
= size
;
1837 EXPORT_SYMBOL_GPL(bio_trim
);
1840 * create memory pools for biovec's in a bio_set.
1841 * use the global biovec slabs created for general use.
1843 mempool_t
*biovec_create_pool(struct bio_set
*bs
, int pool_entries
)
1845 struct biovec_slab
*bp
= bvec_slabs
+ BIOVEC_MAX_IDX
;
1847 return mempool_create_slab_pool(pool_entries
, bp
->slab
);
1850 void bioset_free(struct bio_set
*bs
)
1852 if (bs
->rescue_workqueue
)
1853 destroy_workqueue(bs
->rescue_workqueue
);
1856 mempool_destroy(bs
->bio_pool
);
1859 mempool_destroy(bs
->bvec_pool
);
1861 bioset_integrity_free(bs
);
1866 EXPORT_SYMBOL(bioset_free
);
1869 * bioset_create - Create a bio_set
1870 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1871 * @front_pad: Number of bytes to allocate in front of the returned bio
1874 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1875 * to ask for a number of bytes to be allocated in front of the bio.
1876 * Front pad allocation is useful for embedding the bio inside
1877 * another structure, to avoid allocating extra data to go with the bio.
1878 * Note that the bio must be embedded at the END of that structure always,
1879 * or things will break badly.
1881 struct bio_set
*bioset_create(unsigned int pool_size
, unsigned int front_pad
)
1883 unsigned int back_pad
= BIO_INLINE_VECS
* sizeof(struct bio_vec
);
1886 bs
= kzalloc(sizeof(*bs
), GFP_KERNEL
);
1890 bs
->front_pad
= front_pad
;
1892 spin_lock_init(&bs
->rescue_lock
);
1893 bio_list_init(&bs
->rescue_list
);
1894 INIT_WORK(&bs
->rescue_work
, bio_alloc_rescue
);
1896 bs
->bio_slab
= bio_find_or_create_slab(front_pad
+ back_pad
);
1897 if (!bs
->bio_slab
) {
1902 bs
->bio_pool
= mempool_create_slab_pool(pool_size
, bs
->bio_slab
);
1906 bs
->bvec_pool
= biovec_create_pool(bs
, pool_size
);
1910 bs
->rescue_workqueue
= alloc_workqueue("bioset", WQ_MEM_RECLAIM
, 0);
1911 if (!bs
->rescue_workqueue
)
1919 EXPORT_SYMBOL(bioset_create
);
1921 #ifdef CONFIG_BLK_CGROUP
1923 * bio_associate_current - associate a bio with %current
1926 * Associate @bio with %current if it hasn't been associated yet. Block
1927 * layer will treat @bio as if it were issued by %current no matter which
1928 * task actually issues it.
1930 * This function takes an extra reference of @task's io_context and blkcg
1931 * which will be put when @bio is released. The caller must own @bio,
1932 * ensure %current->io_context exists, and is responsible for synchronizing
1933 * calls to this function.
1935 int bio_associate_current(struct bio
*bio
)
1937 struct io_context
*ioc
;
1938 struct cgroup_subsys_state
*css
;
1943 ioc
= current
->io_context
;
1947 /* acquire active ref on @ioc and associate */
1948 get_io_context_active(ioc
);
1951 /* associate blkcg if exists */
1953 css
= task_css(current
, blkio_subsys_id
);
1954 if (css
&& css_tryget(css
))
1962 * bio_disassociate_task - undo bio_associate_current()
1965 void bio_disassociate_task(struct bio
*bio
)
1968 put_io_context(bio
->bi_ioc
);
1972 css_put(bio
->bi_css
);
1977 #endif /* CONFIG_BLK_CGROUP */
1979 static void __init
biovec_init_slabs(void)
1983 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
1985 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
1987 if (bvs
->nr_vecs
<= BIO_INLINE_VECS
) {
1992 size
= bvs
->nr_vecs
* sizeof(struct bio_vec
);
1993 bvs
->slab
= kmem_cache_create(bvs
->name
, size
, 0,
1994 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
1998 static int __init
init_bio(void)
2002 bio_slabs
= kzalloc(bio_slab_max
* sizeof(struct bio_slab
), GFP_KERNEL
);
2004 panic("bio: can't allocate bios\n");
2006 bio_integrity_init();
2007 biovec_init_slabs();
2009 fs_bio_set
= bioset_create(BIO_POOL_SIZE
, 0);
2011 panic("bio: can't allocate bios\n");
2013 if (bioset_integrity_create(fs_bio_set
, BIO_POOL_SIZE
))
2014 panic("bio: can't create integrity pool\n");
2016 bio_split_pool
= mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES
,
2017 sizeof(struct bio_pair
));
2018 if (!bio_split_pool
)
2019 panic("bio: can't create split pool\n");
2023 subsys_initcall(init_bio
);