]>
git.proxmox.com Git - mirror_ubuntu-eoan-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 * Increment chain count for the bio. Make sure the CHAIN flag update
308 * is visible before the raised count.
310 static inline void bio_inc_remaining(struct bio
*bio
)
312 bio
->bi_flags
|= (1 << BIO_CHAIN
);
313 smp_mb__before_atomic();
314 atomic_inc(&bio
->__bi_remaining
);
318 * bio_chain - chain bio completions
319 * @bio: the target bio
320 * @parent: the @bio's parent bio
322 * The caller won't have a bi_end_io called when @bio completes - instead,
323 * @parent's bi_end_io won't be called until both @parent and @bio have
324 * completed; the chained bio will also be freed when it completes.
326 * The caller must not set bi_private or bi_end_io in @bio.
328 void bio_chain(struct bio
*bio
, struct bio
*parent
)
330 BUG_ON(bio
->bi_private
|| bio
->bi_end_io
);
332 bio
->bi_private
= parent
;
333 bio
->bi_end_io
= bio_chain_endio
;
334 bio_inc_remaining(parent
);
336 EXPORT_SYMBOL(bio_chain
);
338 static void bio_alloc_rescue(struct work_struct
*work
)
340 struct bio_set
*bs
= container_of(work
, struct bio_set
, rescue_work
);
344 spin_lock(&bs
->rescue_lock
);
345 bio
= bio_list_pop(&bs
->rescue_list
);
346 spin_unlock(&bs
->rescue_lock
);
351 generic_make_request(bio
);
355 static void punt_bios_to_rescuer(struct bio_set
*bs
)
357 struct bio_list punt
, nopunt
;
361 * In order to guarantee forward progress we must punt only bios that
362 * were allocated from this bio_set; otherwise, if there was a bio on
363 * there for a stacking driver higher up in the stack, processing it
364 * could require allocating bios from this bio_set, and doing that from
365 * our own rescuer would be bad.
367 * Since bio lists are singly linked, pop them all instead of trying to
368 * remove from the middle of the list:
371 bio_list_init(&punt
);
372 bio_list_init(&nopunt
);
374 while ((bio
= bio_list_pop(current
->bio_list
)))
375 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
377 *current
->bio_list
= nopunt
;
379 spin_lock(&bs
->rescue_lock
);
380 bio_list_merge(&bs
->rescue_list
, &punt
);
381 spin_unlock(&bs
->rescue_lock
);
383 queue_work(bs
->rescue_workqueue
, &bs
->rescue_work
);
387 * bio_alloc_bioset - allocate a bio for I/O
388 * @gfp_mask: the GFP_ mask given to the slab allocator
389 * @nr_iovecs: number of iovecs to pre-allocate
390 * @bs: the bio_set to allocate from.
393 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
394 * backed by the @bs's mempool.
396 * When @bs is not NULL, if %__GFP_WAIT is set then bio_alloc will always be
397 * able to allocate a bio. This is due to the mempool guarantees. To make this
398 * work, callers must never allocate more than 1 bio at a time from this pool.
399 * Callers that need to allocate more than 1 bio must always submit the
400 * previously allocated bio for IO before attempting to allocate a new one.
401 * Failure to do so can cause deadlocks under memory pressure.
403 * Note that when running under generic_make_request() (i.e. any block
404 * driver), bios are not submitted until after you return - see the code in
405 * generic_make_request() that converts recursion into iteration, to prevent
408 * This would normally mean allocating multiple bios under
409 * generic_make_request() would be susceptible to deadlocks, but we have
410 * deadlock avoidance code that resubmits any blocked bios from a rescuer
413 * However, we do not guarantee forward progress for allocations from other
414 * mempools. Doing multiple allocations from the same mempool under
415 * generic_make_request() should be avoided - instead, use bio_set's front_pad
416 * for per bio allocations.
419 * Pointer to new bio on success, NULL on failure.
421 struct bio
*bio_alloc_bioset(gfp_t gfp_mask
, int nr_iovecs
, struct bio_set
*bs
)
423 gfp_t saved_gfp
= gfp_mask
;
425 unsigned inline_vecs
;
426 unsigned long idx
= BIO_POOL_NONE
;
427 struct bio_vec
*bvl
= NULL
;
432 if (nr_iovecs
> UIO_MAXIOV
)
435 p
= kmalloc(sizeof(struct bio
) +
436 nr_iovecs
* sizeof(struct bio_vec
),
439 inline_vecs
= nr_iovecs
;
441 /* should not use nobvec bioset for nr_iovecs > 0 */
442 if (WARN_ON_ONCE(!bs
->bvec_pool
&& nr_iovecs
> 0))
445 * generic_make_request() converts recursion to iteration; this
446 * means if we're running beneath it, any bios we allocate and
447 * submit will not be submitted (and thus freed) until after we
450 * This exposes us to a potential deadlock if we allocate
451 * multiple bios from the same bio_set() while running
452 * underneath generic_make_request(). If we were to allocate
453 * multiple bios (say a stacking block driver that was splitting
454 * bios), we would deadlock if we exhausted the mempool's
457 * We solve this, and guarantee forward progress, with a rescuer
458 * workqueue per bio_set. If we go to allocate and there are
459 * bios on current->bio_list, we first try the allocation
460 * without __GFP_WAIT; if that fails, we punt those bios we
461 * would be blocking to the rescuer workqueue before we retry
462 * with the original gfp_flags.
465 if (current
->bio_list
&& !bio_list_empty(current
->bio_list
))
466 gfp_mask
&= ~__GFP_WAIT
;
468 p
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
469 if (!p
&& gfp_mask
!= saved_gfp
) {
470 punt_bios_to_rescuer(bs
);
471 gfp_mask
= saved_gfp
;
472 p
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
475 front_pad
= bs
->front_pad
;
476 inline_vecs
= BIO_INLINE_VECS
;
485 if (nr_iovecs
> inline_vecs
) {
486 bvl
= bvec_alloc(gfp_mask
, nr_iovecs
, &idx
, bs
->bvec_pool
);
487 if (!bvl
&& gfp_mask
!= saved_gfp
) {
488 punt_bios_to_rescuer(bs
);
489 gfp_mask
= saved_gfp
;
490 bvl
= bvec_alloc(gfp_mask
, nr_iovecs
, &idx
, bs
->bvec_pool
);
496 bio
->bi_flags
|= 1 << BIO_OWNS_VEC
;
497 } else if (nr_iovecs
) {
498 bvl
= bio
->bi_inline_vecs
;
502 bio
->bi_flags
|= idx
<< BIO_POOL_OFFSET
;
503 bio
->bi_max_vecs
= nr_iovecs
;
504 bio
->bi_io_vec
= bvl
;
508 mempool_free(p
, bs
->bio_pool
);
511 EXPORT_SYMBOL(bio_alloc_bioset
);
513 void zero_fill_bio(struct bio
*bio
)
517 struct bvec_iter iter
;
519 bio_for_each_segment(bv
, bio
, iter
) {
520 char *data
= bvec_kmap_irq(&bv
, &flags
);
521 memset(data
, 0, bv
.bv_len
);
522 flush_dcache_page(bv
.bv_page
);
523 bvec_kunmap_irq(data
, &flags
);
526 EXPORT_SYMBOL(zero_fill_bio
);
529 * bio_put - release a reference to a bio
530 * @bio: bio to release reference to
533 * Put a reference to a &struct bio, either one you have gotten with
534 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
536 void bio_put(struct bio
*bio
)
538 if (!bio_flagged(bio
, BIO_REFFED
))
541 BIO_BUG_ON(!atomic_read(&bio
->__bi_cnt
));
546 if (atomic_dec_and_test(&bio
->__bi_cnt
))
550 EXPORT_SYMBOL(bio_put
);
552 inline int bio_phys_segments(struct request_queue
*q
, struct bio
*bio
)
554 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
555 blk_recount_segments(q
, bio
);
557 return bio
->bi_phys_segments
;
559 EXPORT_SYMBOL(bio_phys_segments
);
562 * __bio_clone_fast - clone a bio that shares the original bio's biovec
563 * @bio: destination bio
564 * @bio_src: bio to clone
566 * Clone a &bio. Caller will own the returned bio, but not
567 * the actual data it points to. Reference count of returned
570 * Caller must ensure that @bio_src is not freed before @bio.
572 void __bio_clone_fast(struct bio
*bio
, struct bio
*bio_src
)
574 BUG_ON(bio
->bi_pool
&& BIO_POOL_IDX(bio
) != BIO_POOL_NONE
);
577 * most users will be overriding ->bi_bdev with a new target,
578 * so we don't set nor calculate new physical/hw segment counts here
580 bio
->bi_bdev
= bio_src
->bi_bdev
;
581 bio
->bi_flags
|= 1 << BIO_CLONED
;
582 bio
->bi_rw
= bio_src
->bi_rw
;
583 bio
->bi_iter
= bio_src
->bi_iter
;
584 bio
->bi_io_vec
= bio_src
->bi_io_vec
;
586 EXPORT_SYMBOL(__bio_clone_fast
);
589 * bio_clone_fast - clone a bio that shares the original bio's biovec
591 * @gfp_mask: allocation priority
592 * @bs: bio_set to allocate from
594 * Like __bio_clone_fast, only also allocates the returned bio
596 struct bio
*bio_clone_fast(struct bio
*bio
, gfp_t gfp_mask
, struct bio_set
*bs
)
600 b
= bio_alloc_bioset(gfp_mask
, 0, bs
);
604 __bio_clone_fast(b
, bio
);
606 if (bio_integrity(bio
)) {
609 ret
= bio_integrity_clone(b
, bio
, gfp_mask
);
619 EXPORT_SYMBOL(bio_clone_fast
);
622 * bio_clone_bioset - clone a bio
623 * @bio_src: bio to clone
624 * @gfp_mask: allocation priority
625 * @bs: bio_set to allocate from
627 * Clone bio. Caller will own the returned bio, but not the actual data it
628 * points to. Reference count of returned bio will be one.
630 struct bio
*bio_clone_bioset(struct bio
*bio_src
, gfp_t gfp_mask
,
633 struct bvec_iter iter
;
638 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
639 * bio_src->bi_io_vec to bio->bi_io_vec.
641 * We can't do that anymore, because:
643 * - The point of cloning the biovec is to produce a bio with a biovec
644 * the caller can modify: bi_idx and bi_bvec_done should be 0.
646 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
647 * we tried to clone the whole thing bio_alloc_bioset() would fail.
648 * But the clone should succeed as long as the number of biovecs we
649 * actually need to allocate is fewer than BIO_MAX_PAGES.
651 * - Lastly, bi_vcnt should not be looked at or relied upon by code
652 * that does not own the bio - reason being drivers don't use it for
653 * iterating over the biovec anymore, so expecting it to be kept up
654 * to date (i.e. for clones that share the parent biovec) is just
655 * asking for trouble and would force extra work on
656 * __bio_clone_fast() anyways.
659 bio
= bio_alloc_bioset(gfp_mask
, bio_segments(bio_src
), bs
);
663 bio
->bi_bdev
= bio_src
->bi_bdev
;
664 bio
->bi_rw
= bio_src
->bi_rw
;
665 bio
->bi_iter
.bi_sector
= bio_src
->bi_iter
.bi_sector
;
666 bio
->bi_iter
.bi_size
= bio_src
->bi_iter
.bi_size
;
668 if (bio
->bi_rw
& REQ_DISCARD
)
669 goto integrity_clone
;
671 if (bio
->bi_rw
& REQ_WRITE_SAME
) {
672 bio
->bi_io_vec
[bio
->bi_vcnt
++] = bio_src
->bi_io_vec
[0];
673 goto integrity_clone
;
676 bio_for_each_segment(bv
, bio_src
, iter
)
677 bio
->bi_io_vec
[bio
->bi_vcnt
++] = bv
;
680 if (bio_integrity(bio_src
)) {
683 ret
= bio_integrity_clone(bio
, bio_src
, gfp_mask
);
692 EXPORT_SYMBOL(bio_clone_bioset
);
695 * bio_get_nr_vecs - return approx number of vecs
698 * Return the approximate number of pages we can send to this target.
699 * There's no guarantee that you will be able to fit this number of pages
700 * into a bio, it does not account for dynamic restrictions that vary
703 int bio_get_nr_vecs(struct block_device
*bdev
)
705 struct request_queue
*q
= bdev_get_queue(bdev
);
708 nr_pages
= min_t(unsigned,
709 queue_max_segments(q
),
710 queue_max_sectors(q
) / (PAGE_SIZE
>> 9) + 1);
712 return min_t(unsigned, nr_pages
, BIO_MAX_PAGES
);
715 EXPORT_SYMBOL(bio_get_nr_vecs
);
717 static int __bio_add_page(struct request_queue
*q
, struct bio
*bio
, struct page
718 *page
, unsigned int len
, unsigned int offset
,
719 unsigned int max_sectors
)
721 int retried_segments
= 0;
722 struct bio_vec
*bvec
;
725 * cloned bio must not modify vec list
727 if (unlikely(bio_flagged(bio
, BIO_CLONED
)))
730 if (((bio
->bi_iter
.bi_size
+ len
) >> 9) > max_sectors
)
734 * For filesystems with a blocksize smaller than the pagesize
735 * we will often be called with the same page as last time and
736 * a consecutive offset. Optimize this special case.
738 if (bio
->bi_vcnt
> 0) {
739 struct bio_vec
*prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
741 if (page
== prev
->bv_page
&&
742 offset
== prev
->bv_offset
+ prev
->bv_len
) {
743 unsigned int prev_bv_len
= prev
->bv_len
;
746 if (q
->merge_bvec_fn
) {
747 struct bvec_merge_data bvm
= {
748 /* prev_bvec is already charged in
749 bi_size, discharge it in order to
750 simulate merging updated prev_bvec
752 .bi_bdev
= bio
->bi_bdev
,
753 .bi_sector
= bio
->bi_iter
.bi_sector
,
754 .bi_size
= bio
->bi_iter
.bi_size
-
759 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < prev
->bv_len
) {
765 bio
->bi_iter
.bi_size
+= len
;
770 * If the queue doesn't support SG gaps and adding this
771 * offset would create a gap, disallow it.
773 if (q
->queue_flags
& (1 << QUEUE_FLAG_SG_GAPS
) &&
774 bvec_gap_to_prev(prev
, offset
))
778 if (bio
->bi_vcnt
>= bio
->bi_max_vecs
)
782 * setup the new entry, we might clear it again later if we
783 * cannot add the page
785 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
786 bvec
->bv_page
= page
;
788 bvec
->bv_offset
= offset
;
790 bio
->bi_phys_segments
++;
791 bio
->bi_iter
.bi_size
+= len
;
794 * Perform a recount if the number of segments is greater
795 * than queue_max_segments(q).
798 while (bio
->bi_phys_segments
> queue_max_segments(q
)) {
800 if (retried_segments
)
803 retried_segments
= 1;
804 blk_recount_segments(q
, bio
);
808 * if queue has other restrictions (eg varying max sector size
809 * depending on offset), it can specify a merge_bvec_fn in the
810 * queue to get further control
812 if (q
->merge_bvec_fn
) {
813 struct bvec_merge_data bvm
= {
814 .bi_bdev
= bio
->bi_bdev
,
815 .bi_sector
= bio
->bi_iter
.bi_sector
,
816 .bi_size
= bio
->bi_iter
.bi_size
- len
,
821 * merge_bvec_fn() returns number of bytes it can accept
824 if (q
->merge_bvec_fn(q
, &bvm
, bvec
) < bvec
->bv_len
)
828 /* If we may be able to merge these biovecs, force a recount */
829 if (bio
->bi_vcnt
> 1 && (BIOVEC_PHYS_MERGEABLE(bvec
-1, bvec
)))
830 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
836 bvec
->bv_page
= NULL
;
840 bio
->bi_iter
.bi_size
-= len
;
841 blk_recount_segments(q
, bio
);
846 * bio_add_pc_page - attempt to add page to bio
847 * @q: the target queue
848 * @bio: destination bio
850 * @len: vec entry length
851 * @offset: vec entry offset
853 * Attempt to add a page to the bio_vec maplist. This can fail for a
854 * number of reasons, such as the bio being full or target block device
855 * limitations. The target block device must allow bio's up to PAGE_SIZE,
856 * so it is always possible to add a single page to an empty bio.
858 * This should only be used by REQ_PC bios.
860 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
, struct page
*page
,
861 unsigned int len
, unsigned int offset
)
863 return __bio_add_page(q
, bio
, page
, len
, offset
,
864 queue_max_hw_sectors(q
));
866 EXPORT_SYMBOL(bio_add_pc_page
);
869 * bio_add_page - attempt to add page to bio
870 * @bio: destination bio
872 * @len: vec entry length
873 * @offset: vec entry offset
875 * Attempt to add a page to the bio_vec maplist. This can fail for a
876 * number of reasons, such as the bio being full or target block device
877 * limitations. The target block device must allow bio's up to PAGE_SIZE,
878 * so it is always possible to add a single page to an empty bio.
880 int bio_add_page(struct bio
*bio
, struct page
*page
, unsigned int len
,
883 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
884 unsigned int max_sectors
;
886 max_sectors
= blk_max_size_offset(q
, bio
->bi_iter
.bi_sector
);
887 if ((max_sectors
< (len
>> 9)) && !bio
->bi_iter
.bi_size
)
888 max_sectors
= len
>> 9;
890 return __bio_add_page(q
, bio
, page
, len
, offset
, max_sectors
);
892 EXPORT_SYMBOL(bio_add_page
);
894 struct submit_bio_ret
{
895 struct completion event
;
899 static void submit_bio_wait_endio(struct bio
*bio
, int error
)
901 struct submit_bio_ret
*ret
= bio
->bi_private
;
904 complete(&ret
->event
);
908 * submit_bio_wait - submit a bio, and wait until it completes
909 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
910 * @bio: The &struct bio which describes the I/O
912 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
913 * bio_endio() on failure.
915 int submit_bio_wait(int rw
, struct bio
*bio
)
917 struct submit_bio_ret ret
;
920 init_completion(&ret
.event
);
921 bio
->bi_private
= &ret
;
922 bio
->bi_end_io
= submit_bio_wait_endio
;
924 wait_for_completion(&ret
.event
);
928 EXPORT_SYMBOL(submit_bio_wait
);
931 * bio_advance - increment/complete a bio by some number of bytes
932 * @bio: bio to advance
933 * @bytes: number of bytes to complete
935 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
936 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
937 * be updated on the last bvec as well.
939 * @bio will then represent the remaining, uncompleted portion of the io.
941 void bio_advance(struct bio
*bio
, unsigned bytes
)
943 if (bio_integrity(bio
))
944 bio_integrity_advance(bio
, bytes
);
946 bio_advance_iter(bio
, &bio
->bi_iter
, bytes
);
948 EXPORT_SYMBOL(bio_advance
);
951 * bio_alloc_pages - allocates a single page for each bvec in a bio
952 * @bio: bio to allocate pages for
953 * @gfp_mask: flags for allocation
955 * Allocates pages up to @bio->bi_vcnt.
957 * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
960 int bio_alloc_pages(struct bio
*bio
, gfp_t gfp_mask
)
965 bio_for_each_segment_all(bv
, bio
, i
) {
966 bv
->bv_page
= alloc_page(gfp_mask
);
968 while (--bv
>= bio
->bi_io_vec
)
969 __free_page(bv
->bv_page
);
976 EXPORT_SYMBOL(bio_alloc_pages
);
979 * bio_copy_data - copy contents of data buffers from one chain of bios to
981 * @src: source bio list
982 * @dst: destination bio list
984 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
985 * @src and @dst as linked lists of bios.
987 * Stops when it reaches the end of either @src or @dst - that is, copies
988 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
990 void bio_copy_data(struct bio
*dst
, struct bio
*src
)
992 struct bvec_iter src_iter
, dst_iter
;
993 struct bio_vec src_bv
, dst_bv
;
997 src_iter
= src
->bi_iter
;
998 dst_iter
= dst
->bi_iter
;
1001 if (!src_iter
.bi_size
) {
1006 src_iter
= src
->bi_iter
;
1009 if (!dst_iter
.bi_size
) {
1014 dst_iter
= dst
->bi_iter
;
1017 src_bv
= bio_iter_iovec(src
, src_iter
);
1018 dst_bv
= bio_iter_iovec(dst
, dst_iter
);
1020 bytes
= min(src_bv
.bv_len
, dst_bv
.bv_len
);
1022 src_p
= kmap_atomic(src_bv
.bv_page
);
1023 dst_p
= kmap_atomic(dst_bv
.bv_page
);
1025 memcpy(dst_p
+ dst_bv
.bv_offset
,
1026 src_p
+ src_bv
.bv_offset
,
1029 kunmap_atomic(dst_p
);
1030 kunmap_atomic(src_p
);
1032 bio_advance_iter(src
, &src_iter
, bytes
);
1033 bio_advance_iter(dst
, &dst_iter
, bytes
);
1036 EXPORT_SYMBOL(bio_copy_data
);
1038 struct bio_map_data
{
1040 struct iov_iter iter
;
1044 static struct bio_map_data
*bio_alloc_map_data(unsigned int iov_count
,
1047 if (iov_count
> UIO_MAXIOV
)
1050 return kmalloc(sizeof(struct bio_map_data
) +
1051 sizeof(struct iovec
) * iov_count
, gfp_mask
);
1055 * bio_copy_from_iter - copy all pages from iov_iter to bio
1056 * @bio: The &struct bio which describes the I/O as destination
1057 * @iter: iov_iter as source
1059 * Copy all pages from iov_iter to bio.
1060 * Returns 0 on success, or error on failure.
1062 static int bio_copy_from_iter(struct bio
*bio
, struct iov_iter iter
)
1065 struct bio_vec
*bvec
;
1067 bio_for_each_segment_all(bvec
, bio
, i
) {
1070 ret
= copy_page_from_iter(bvec
->bv_page
,
1075 if (!iov_iter_count(&iter
))
1078 if (ret
< bvec
->bv_len
)
1086 * bio_copy_to_iter - copy all pages from bio to iov_iter
1087 * @bio: The &struct bio which describes the I/O as source
1088 * @iter: iov_iter as destination
1090 * Copy all pages from bio to iov_iter.
1091 * Returns 0 on success, or error on failure.
1093 static int bio_copy_to_iter(struct bio
*bio
, struct iov_iter iter
)
1096 struct bio_vec
*bvec
;
1098 bio_for_each_segment_all(bvec
, bio
, i
) {
1101 ret
= copy_page_to_iter(bvec
->bv_page
,
1106 if (!iov_iter_count(&iter
))
1109 if (ret
< bvec
->bv_len
)
1116 static void bio_free_pages(struct bio
*bio
)
1118 struct bio_vec
*bvec
;
1121 bio_for_each_segment_all(bvec
, bio
, i
)
1122 __free_page(bvec
->bv_page
);
1126 * bio_uncopy_user - finish previously mapped bio
1127 * @bio: bio being terminated
1129 * Free pages allocated from bio_copy_user_iov() and write back data
1130 * to user space in case of a read.
1132 int bio_uncopy_user(struct bio
*bio
)
1134 struct bio_map_data
*bmd
= bio
->bi_private
;
1137 if (!bio_flagged(bio
, BIO_NULL_MAPPED
)) {
1139 * if we're in a workqueue, the request is orphaned, so
1140 * don't copy into a random user address space, just free.
1142 if (current
->mm
&& bio_data_dir(bio
) == READ
)
1143 ret
= bio_copy_to_iter(bio
, bmd
->iter
);
1144 if (bmd
->is_our_pages
)
1145 bio_free_pages(bio
);
1151 EXPORT_SYMBOL(bio_uncopy_user
);
1154 * bio_copy_user_iov - copy user data to bio
1155 * @q: destination block queue
1156 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1157 * @iter: iovec iterator
1158 * @gfp_mask: memory allocation flags
1160 * Prepares and returns a bio for indirect user io, bouncing data
1161 * to/from kernel pages as necessary. Must be paired with
1162 * call bio_uncopy_user() on io completion.
1164 struct bio
*bio_copy_user_iov(struct request_queue
*q
,
1165 struct rq_map_data
*map_data
,
1166 const struct iov_iter
*iter
,
1169 struct bio_map_data
*bmd
;
1174 unsigned int len
= iter
->count
;
1175 unsigned int offset
= map_data
? map_data
->offset
& ~PAGE_MASK
: 0;
1177 for (i
= 0; i
< iter
->nr_segs
; i
++) {
1178 unsigned long uaddr
;
1180 unsigned long start
;
1182 uaddr
= (unsigned long) iter
->iov
[i
].iov_base
;
1183 end
= (uaddr
+ iter
->iov
[i
].iov_len
+ PAGE_SIZE
- 1)
1185 start
= uaddr
>> PAGE_SHIFT
;
1191 return ERR_PTR(-EINVAL
);
1193 nr_pages
+= end
- start
;
1199 bmd
= bio_alloc_map_data(iter
->nr_segs
, gfp_mask
);
1201 return ERR_PTR(-ENOMEM
);
1204 * We need to do a deep copy of the iov_iter including the iovecs.
1205 * The caller provided iov might point to an on-stack or otherwise
1208 bmd
->is_our_pages
= map_data
? 0 : 1;
1209 memcpy(bmd
->iov
, iter
->iov
, sizeof(struct iovec
) * iter
->nr_segs
);
1210 iov_iter_init(&bmd
->iter
, iter
->type
, bmd
->iov
,
1211 iter
->nr_segs
, iter
->count
);
1214 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1218 if (iter
->type
& WRITE
)
1219 bio
->bi_rw
|= REQ_WRITE
;
1224 nr_pages
= 1 << map_data
->page_order
;
1225 i
= map_data
->offset
/ PAGE_SIZE
;
1228 unsigned int bytes
= PAGE_SIZE
;
1236 if (i
== map_data
->nr_entries
* nr_pages
) {
1241 page
= map_data
->pages
[i
/ nr_pages
];
1242 page
+= (i
% nr_pages
);
1246 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
1253 if (bio_add_pc_page(q
, bio
, page
, bytes
, offset
) < bytes
)
1266 if (((iter
->type
& WRITE
) && (!map_data
|| !map_data
->null_mapped
)) ||
1267 (map_data
&& map_data
->from_user
)) {
1268 ret
= bio_copy_from_iter(bio
, *iter
);
1273 bio
->bi_private
= bmd
;
1277 bio_free_pages(bio
);
1281 return ERR_PTR(ret
);
1285 * bio_map_user_iov - map user iovec into bio
1286 * @q: the struct request_queue for the bio
1287 * @iter: iovec iterator
1288 * @gfp_mask: memory allocation flags
1290 * Map the user space address into a bio suitable for io to a block
1291 * device. Returns an error pointer in case of error.
1293 struct bio
*bio_map_user_iov(struct request_queue
*q
,
1294 const struct iov_iter
*iter
,
1299 struct page
**pages
;
1306 iov_for_each(iov
, i
, *iter
) {
1307 unsigned long uaddr
= (unsigned long) iov
.iov_base
;
1308 unsigned long len
= iov
.iov_len
;
1309 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1310 unsigned long start
= uaddr
>> PAGE_SHIFT
;
1316 return ERR_PTR(-EINVAL
);
1318 nr_pages
+= end
- start
;
1320 * buffer must be aligned to at least hardsector size for now
1322 if (uaddr
& queue_dma_alignment(q
))
1323 return ERR_PTR(-EINVAL
);
1327 return ERR_PTR(-EINVAL
);
1329 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1331 return ERR_PTR(-ENOMEM
);
1334 pages
= kcalloc(nr_pages
, sizeof(struct page
*), gfp_mask
);
1338 iov_for_each(iov
, i
, *iter
) {
1339 unsigned long uaddr
= (unsigned long) iov
.iov_base
;
1340 unsigned long len
= iov
.iov_len
;
1341 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1342 unsigned long start
= uaddr
>> PAGE_SHIFT
;
1343 const int local_nr_pages
= end
- start
;
1344 const int page_limit
= cur_page
+ local_nr_pages
;
1346 ret
= get_user_pages_fast(uaddr
, local_nr_pages
,
1347 (iter
->type
& WRITE
) != WRITE
,
1349 if (ret
< local_nr_pages
) {
1354 offset
= uaddr
& ~PAGE_MASK
;
1355 for (j
= cur_page
; j
< page_limit
; j
++) {
1356 unsigned int bytes
= PAGE_SIZE
- offset
;
1367 if (bio_add_pc_page(q
, bio
, pages
[j
], bytes
, offset
) <
1377 * release the pages we didn't map into the bio, if any
1379 while (j
< page_limit
)
1380 page_cache_release(pages
[j
++]);
1386 * set data direction, and check if mapped pages need bouncing
1388 if (iter
->type
& WRITE
)
1389 bio
->bi_rw
|= REQ_WRITE
;
1391 bio
->bi_flags
|= (1 << BIO_USER_MAPPED
);
1394 * subtle -- if __bio_map_user() ended up bouncing a bio,
1395 * it would normally disappear when its bi_end_io is run.
1396 * however, we need it for the unmap, so grab an extra
1403 for (j
= 0; j
< nr_pages
; j
++) {
1406 page_cache_release(pages
[j
]);
1411 return ERR_PTR(ret
);
1414 static void __bio_unmap_user(struct bio
*bio
)
1416 struct bio_vec
*bvec
;
1420 * make sure we dirty pages we wrote to
1422 bio_for_each_segment_all(bvec
, bio
, i
) {
1423 if (bio_data_dir(bio
) == READ
)
1424 set_page_dirty_lock(bvec
->bv_page
);
1426 page_cache_release(bvec
->bv_page
);
1433 * bio_unmap_user - unmap a bio
1434 * @bio: the bio being unmapped
1436 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1437 * a process context.
1439 * bio_unmap_user() may sleep.
1441 void bio_unmap_user(struct bio
*bio
)
1443 __bio_unmap_user(bio
);
1446 EXPORT_SYMBOL(bio_unmap_user
);
1448 static void bio_map_kern_endio(struct bio
*bio
, int err
)
1454 * bio_map_kern - map kernel address into bio
1455 * @q: the struct request_queue for the bio
1456 * @data: pointer to buffer to map
1457 * @len: length in bytes
1458 * @gfp_mask: allocation flags for bio allocation
1460 * Map the kernel address into a bio suitable for io to a block
1461 * device. Returns an error pointer in case of error.
1463 struct bio
*bio_map_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1466 unsigned long kaddr
= (unsigned long)data
;
1467 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1468 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1469 const int nr_pages
= end
- start
;
1473 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1475 return ERR_PTR(-ENOMEM
);
1477 offset
= offset_in_page(kaddr
);
1478 for (i
= 0; i
< nr_pages
; i
++) {
1479 unsigned int bytes
= PAGE_SIZE
- offset
;
1487 if (bio_add_pc_page(q
, bio
, virt_to_page(data
), bytes
,
1489 /* we don't support partial mappings */
1491 return ERR_PTR(-EINVAL
);
1499 bio
->bi_end_io
= bio_map_kern_endio
;
1502 EXPORT_SYMBOL(bio_map_kern
);
1504 static void bio_copy_kern_endio(struct bio
*bio
, int err
)
1506 bio_free_pages(bio
);
1510 static void bio_copy_kern_endio_read(struct bio
*bio
, int err
)
1512 char *p
= bio
->bi_private
;
1513 struct bio_vec
*bvec
;
1516 bio_for_each_segment_all(bvec
, bio
, i
) {
1517 memcpy(p
, page_address(bvec
->bv_page
), bvec
->bv_len
);
1521 bio_copy_kern_endio(bio
, err
);
1525 * bio_copy_kern - copy kernel address into bio
1526 * @q: the struct request_queue for the bio
1527 * @data: pointer to buffer to copy
1528 * @len: length in bytes
1529 * @gfp_mask: allocation flags for bio and page allocation
1530 * @reading: data direction is READ
1532 * copy the kernel address into a bio suitable for io to a block
1533 * device. Returns an error pointer in case of error.
1535 struct bio
*bio_copy_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1536 gfp_t gfp_mask
, int reading
)
1538 unsigned long kaddr
= (unsigned long)data
;
1539 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1540 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1549 return ERR_PTR(-EINVAL
);
1551 nr_pages
= end
- start
;
1552 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1554 return ERR_PTR(-ENOMEM
);
1558 unsigned int bytes
= PAGE_SIZE
;
1563 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
1568 memcpy(page_address(page
), p
, bytes
);
1570 if (bio_add_pc_page(q
, bio
, page
, bytes
, 0) < bytes
)
1578 bio
->bi_end_io
= bio_copy_kern_endio_read
;
1579 bio
->bi_private
= data
;
1581 bio
->bi_end_io
= bio_copy_kern_endio
;
1582 bio
->bi_rw
|= REQ_WRITE
;
1588 bio_free_pages(bio
);
1590 return ERR_PTR(-ENOMEM
);
1592 EXPORT_SYMBOL(bio_copy_kern
);
1595 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1596 * for performing direct-IO in BIOs.
1598 * The problem is that we cannot run set_page_dirty() from interrupt context
1599 * because the required locks are not interrupt-safe. So what we can do is to
1600 * mark the pages dirty _before_ performing IO. And in interrupt context,
1601 * check that the pages are still dirty. If so, fine. If not, redirty them
1602 * in process context.
1604 * We special-case compound pages here: normally this means reads into hugetlb
1605 * pages. The logic in here doesn't really work right for compound pages
1606 * because the VM does not uniformly chase down the head page in all cases.
1607 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1608 * handle them at all. So we skip compound pages here at an early stage.
1610 * Note that this code is very hard to test under normal circumstances because
1611 * direct-io pins the pages with get_user_pages(). This makes
1612 * is_page_cache_freeable return false, and the VM will not clean the pages.
1613 * But other code (eg, flusher threads) could clean the pages if they are mapped
1616 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1617 * deferred bio dirtying paths.
1621 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1623 void bio_set_pages_dirty(struct bio
*bio
)
1625 struct bio_vec
*bvec
;
1628 bio_for_each_segment_all(bvec
, bio
, i
) {
1629 struct page
*page
= bvec
->bv_page
;
1631 if (page
&& !PageCompound(page
))
1632 set_page_dirty_lock(page
);
1636 static void bio_release_pages(struct bio
*bio
)
1638 struct bio_vec
*bvec
;
1641 bio_for_each_segment_all(bvec
, bio
, i
) {
1642 struct page
*page
= bvec
->bv_page
;
1650 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1651 * If they are, then fine. If, however, some pages are clean then they must
1652 * have been written out during the direct-IO read. So we take another ref on
1653 * the BIO and the offending pages and re-dirty the pages in process context.
1655 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1656 * here on. It will run one page_cache_release() against each page and will
1657 * run one bio_put() against the BIO.
1660 static void bio_dirty_fn(struct work_struct
*work
);
1662 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
1663 static DEFINE_SPINLOCK(bio_dirty_lock
);
1664 static struct bio
*bio_dirty_list
;
1667 * This runs in process context
1669 static void bio_dirty_fn(struct work_struct
*work
)
1671 unsigned long flags
;
1674 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1675 bio
= bio_dirty_list
;
1676 bio_dirty_list
= NULL
;
1677 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1680 struct bio
*next
= bio
->bi_private
;
1682 bio_set_pages_dirty(bio
);
1683 bio_release_pages(bio
);
1689 void bio_check_pages_dirty(struct bio
*bio
)
1691 struct bio_vec
*bvec
;
1692 int nr_clean_pages
= 0;
1695 bio_for_each_segment_all(bvec
, bio
, i
) {
1696 struct page
*page
= bvec
->bv_page
;
1698 if (PageDirty(page
) || PageCompound(page
)) {
1699 page_cache_release(page
);
1700 bvec
->bv_page
= NULL
;
1706 if (nr_clean_pages
) {
1707 unsigned long flags
;
1709 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1710 bio
->bi_private
= bio_dirty_list
;
1711 bio_dirty_list
= bio
;
1712 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1713 schedule_work(&bio_dirty_work
);
1719 void generic_start_io_acct(int rw
, unsigned long sectors
,
1720 struct hd_struct
*part
)
1722 int cpu
= part_stat_lock();
1724 part_round_stats(cpu
, part
);
1725 part_stat_inc(cpu
, part
, ios
[rw
]);
1726 part_stat_add(cpu
, part
, sectors
[rw
], sectors
);
1727 part_inc_in_flight(part
, rw
);
1731 EXPORT_SYMBOL(generic_start_io_acct
);
1733 void generic_end_io_acct(int rw
, struct hd_struct
*part
,
1734 unsigned long start_time
)
1736 unsigned long duration
= jiffies
- start_time
;
1737 int cpu
= part_stat_lock();
1739 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1740 part_round_stats(cpu
, part
);
1741 part_dec_in_flight(part
, rw
);
1745 EXPORT_SYMBOL(generic_end_io_acct
);
1747 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1748 void bio_flush_dcache_pages(struct bio
*bi
)
1750 struct bio_vec bvec
;
1751 struct bvec_iter iter
;
1753 bio_for_each_segment(bvec
, bi
, iter
)
1754 flush_dcache_page(bvec
.bv_page
);
1756 EXPORT_SYMBOL(bio_flush_dcache_pages
);
1759 static inline bool bio_remaining_done(struct bio
*bio
)
1762 * If we're not chaining, then ->__bi_remaining is always 1 and
1763 * we always end io on the first invocation.
1765 if (!bio_flagged(bio
, BIO_CHAIN
))
1768 BUG_ON(atomic_read(&bio
->__bi_remaining
) <= 0);
1770 if (atomic_dec_and_test(&bio
->__bi_remaining
)) {
1771 clear_bit(BIO_CHAIN
, &bio
->bi_flags
);
1779 * bio_endio - end I/O on a bio
1781 * @error: error, if any
1784 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1785 * preferred way to end I/O on a bio, it takes care of clearing
1786 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1787 * established -Exxxx (-EIO, for instance) error values in case
1788 * something went wrong. No one should call bi_end_io() directly on a
1789 * bio unless they own it and thus know that it has an end_io
1792 void bio_endio(struct bio
*bio
, int error
)
1796 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1797 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
1800 if (unlikely(!bio_remaining_done(bio
)))
1804 * Need to have a real endio function for chained bios,
1805 * otherwise various corner cases will break (like stacking
1806 * block devices that save/restore bi_end_io) - however, we want
1807 * to avoid unbounded recursion and blowing the stack. Tail call
1808 * optimization would handle this, but compiling with frame
1809 * pointers also disables gcc's sibling call optimization.
1811 if (bio
->bi_end_io
== bio_chain_endio
) {
1812 struct bio
*parent
= bio
->bi_private
;
1817 bio
->bi_end_io(bio
, error
);
1822 EXPORT_SYMBOL(bio_endio
);
1825 * bio_split - split a bio
1826 * @bio: bio to split
1827 * @sectors: number of sectors to split from the front of @bio
1829 * @bs: bio set to allocate from
1831 * Allocates and returns a new bio which represents @sectors from the start of
1832 * @bio, and updates @bio to represent the remaining sectors.
1834 * Unless this is a discard request the newly allocated bio will point
1835 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1836 * @bio is not freed before the split.
1838 struct bio
*bio_split(struct bio
*bio
, int sectors
,
1839 gfp_t gfp
, struct bio_set
*bs
)
1841 struct bio
*split
= NULL
;
1843 BUG_ON(sectors
<= 0);
1844 BUG_ON(sectors
>= bio_sectors(bio
));
1847 * Discards need a mutable bio_vec to accommodate the payload
1848 * required by the DSM TRIM and UNMAP commands.
1850 if (bio
->bi_rw
& REQ_DISCARD
)
1851 split
= bio_clone_bioset(bio
, gfp
, bs
);
1853 split
= bio_clone_fast(bio
, gfp
, bs
);
1858 split
->bi_iter
.bi_size
= sectors
<< 9;
1860 if (bio_integrity(split
))
1861 bio_integrity_trim(split
, 0, sectors
);
1863 bio_advance(bio
, split
->bi_iter
.bi_size
);
1867 EXPORT_SYMBOL(bio_split
);
1870 * bio_trim - trim a bio
1872 * @offset: number of sectors to trim from the front of @bio
1873 * @size: size we want to trim @bio to, in sectors
1875 void bio_trim(struct bio
*bio
, int offset
, int size
)
1877 /* 'bio' is a cloned bio which we need to trim to match
1878 * the given offset and size.
1882 if (offset
== 0 && size
== bio
->bi_iter
.bi_size
)
1885 clear_bit(BIO_SEG_VALID
, &bio
->bi_flags
);
1887 bio_advance(bio
, offset
<< 9);
1889 bio
->bi_iter
.bi_size
= size
;
1891 EXPORT_SYMBOL_GPL(bio_trim
);
1894 * create memory pools for biovec's in a bio_set.
1895 * use the global biovec slabs created for general use.
1897 mempool_t
*biovec_create_pool(int pool_entries
)
1899 struct biovec_slab
*bp
= bvec_slabs
+ BIOVEC_MAX_IDX
;
1901 return mempool_create_slab_pool(pool_entries
, bp
->slab
);
1904 void bioset_free(struct bio_set
*bs
)
1906 if (bs
->rescue_workqueue
)
1907 destroy_workqueue(bs
->rescue_workqueue
);
1910 mempool_destroy(bs
->bio_pool
);
1913 mempool_destroy(bs
->bvec_pool
);
1915 bioset_integrity_free(bs
);
1920 EXPORT_SYMBOL(bioset_free
);
1922 static struct bio_set
*__bioset_create(unsigned int pool_size
,
1923 unsigned int front_pad
,
1924 bool create_bvec_pool
)
1926 unsigned int back_pad
= BIO_INLINE_VECS
* sizeof(struct bio_vec
);
1929 bs
= kzalloc(sizeof(*bs
), GFP_KERNEL
);
1933 bs
->front_pad
= front_pad
;
1935 spin_lock_init(&bs
->rescue_lock
);
1936 bio_list_init(&bs
->rescue_list
);
1937 INIT_WORK(&bs
->rescue_work
, bio_alloc_rescue
);
1939 bs
->bio_slab
= bio_find_or_create_slab(front_pad
+ back_pad
);
1940 if (!bs
->bio_slab
) {
1945 bs
->bio_pool
= mempool_create_slab_pool(pool_size
, bs
->bio_slab
);
1949 if (create_bvec_pool
) {
1950 bs
->bvec_pool
= biovec_create_pool(pool_size
);
1955 bs
->rescue_workqueue
= alloc_workqueue("bioset", WQ_MEM_RECLAIM
, 0);
1956 if (!bs
->rescue_workqueue
)
1966 * bioset_create - Create a bio_set
1967 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1968 * @front_pad: Number of bytes to allocate in front of the returned bio
1971 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1972 * to ask for a number of bytes to be allocated in front of the bio.
1973 * Front pad allocation is useful for embedding the bio inside
1974 * another structure, to avoid allocating extra data to go with the bio.
1975 * Note that the bio must be embedded at the END of that structure always,
1976 * or things will break badly.
1978 struct bio_set
*bioset_create(unsigned int pool_size
, unsigned int front_pad
)
1980 return __bioset_create(pool_size
, front_pad
, true);
1982 EXPORT_SYMBOL(bioset_create
);
1985 * bioset_create_nobvec - Create a bio_set without bio_vec mempool
1986 * @pool_size: Number of bio to cache in the mempool
1987 * @front_pad: Number of bytes to allocate in front of the returned bio
1990 * Same functionality as bioset_create() except that mempool is not
1991 * created for bio_vecs. Saving some memory for bio_clone_fast() users.
1993 struct bio_set
*bioset_create_nobvec(unsigned int pool_size
, unsigned int front_pad
)
1995 return __bioset_create(pool_size
, front_pad
, false);
1997 EXPORT_SYMBOL(bioset_create_nobvec
);
1999 #ifdef CONFIG_BLK_CGROUP
2002 * bio_associate_blkcg - associate a bio with the specified blkcg
2004 * @blkcg_css: css of the blkcg to associate
2006 * Associate @bio with the blkcg specified by @blkcg_css. Block layer will
2007 * treat @bio as if it were issued by a task which belongs to the blkcg.
2009 * This function takes an extra reference of @blkcg_css which will be put
2010 * when @bio is released. The caller must own @bio and is responsible for
2011 * synchronizing calls to this function.
2013 int bio_associate_blkcg(struct bio
*bio
, struct cgroup_subsys_state
*blkcg_css
)
2015 if (unlikely(bio
->bi_css
))
2018 bio
->bi_css
= blkcg_css
;
2021 EXPORT_SYMBOL_GPL(bio_associate_blkcg
);
2024 * bio_associate_current - associate a bio with %current
2027 * Associate @bio with %current if it hasn't been associated yet. Block
2028 * layer will treat @bio as if it were issued by %current no matter which
2029 * task actually issues it.
2031 * This function takes an extra reference of @task's io_context and blkcg
2032 * which will be put when @bio is released. The caller must own @bio,
2033 * ensure %current->io_context exists, and is responsible for synchronizing
2034 * calls to this function.
2036 int bio_associate_current(struct bio
*bio
)
2038 struct io_context
*ioc
;
2043 ioc
= current
->io_context
;
2047 get_io_context_active(ioc
);
2049 bio
->bi_css
= task_get_css(current
, blkio_cgrp_id
);
2052 EXPORT_SYMBOL_GPL(bio_associate_current
);
2055 * bio_disassociate_task - undo bio_associate_current()
2058 void bio_disassociate_task(struct bio
*bio
)
2061 put_io_context(bio
->bi_ioc
);
2065 css_put(bio
->bi_css
);
2070 #endif /* CONFIG_BLK_CGROUP */
2072 static void __init
biovec_init_slabs(void)
2076 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
2078 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
2080 if (bvs
->nr_vecs
<= BIO_INLINE_VECS
) {
2085 size
= bvs
->nr_vecs
* sizeof(struct bio_vec
);
2086 bvs
->slab
= kmem_cache_create(bvs
->name
, size
, 0,
2087 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
2091 static int __init
init_bio(void)
2095 bio_slabs
= kzalloc(bio_slab_max
* sizeof(struct bio_slab
), GFP_KERNEL
);
2097 panic("bio: can't allocate bios\n");
2099 bio_integrity_init();
2100 biovec_init_slabs();
2102 fs_bio_set
= bioset_create(BIO_POOL_SIZE
, 0);
2104 panic("bio: can't allocate bios\n");
2106 if (bioset_integrity_create(fs_bio_set
, BIO_POOL_SIZE
))
2107 panic("bio: can't create integrity pool\n");
2111 subsys_initcall(init_bio
);