2 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
13 * You should have received a copy of the GNU General Public Licens
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
19 #include <linux/swap.h>
20 #include <linux/bio.h>
21 #include <linux/blkdev.h>
22 #include <linux/uio.h>
23 #include <linux/iocontext.h>
24 #include <linux/slab.h>
25 #include <linux/init.h>
26 #include <linux/kernel.h>
27 #include <linux/export.h>
28 #include <linux/mempool.h>
29 #include <linux/workqueue.h>
30 #include <linux/cgroup.h>
32 #include <trace/events/block.h>
36 * Test patch to inline a certain number of bi_io_vec's inside the bio
37 * itself, to shrink a bio data allocation from two mempool calls to one
39 #define BIO_INLINE_VECS 4
42 * if you change this list, also change bvec_alloc or things will
43 * break badly! cannot be bigger than what you can fit into an
46 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
47 static struct biovec_slab bvec_slabs
[BVEC_POOL_NR
] __read_mostly
= {
48 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES
),
53 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
54 * IO code that does not need private memory pools.
56 struct bio_set
*fs_bio_set
;
57 EXPORT_SYMBOL(fs_bio_set
);
60 * Our slab pool management
63 struct kmem_cache
*slab
;
64 unsigned int slab_ref
;
65 unsigned int slab_size
;
68 static DEFINE_MUTEX(bio_slab_lock
);
69 static struct bio_slab
*bio_slabs
;
70 static unsigned int bio_slab_nr
, bio_slab_max
;
72 static struct kmem_cache
*bio_find_or_create_slab(unsigned int extra_size
)
74 unsigned int sz
= sizeof(struct bio
) + extra_size
;
75 struct kmem_cache
*slab
= NULL
;
76 struct bio_slab
*bslab
, *new_bio_slabs
;
77 unsigned int new_bio_slab_max
;
78 unsigned int i
, entry
= -1;
80 mutex_lock(&bio_slab_lock
);
83 while (i
< bio_slab_nr
) {
84 bslab
= &bio_slabs
[i
];
86 if (!bslab
->slab
&& entry
== -1)
88 else if (bslab
->slab_size
== sz
) {
99 if (bio_slab_nr
== bio_slab_max
&& entry
== -1) {
100 new_bio_slab_max
= bio_slab_max
<< 1;
101 new_bio_slabs
= krealloc(bio_slabs
,
102 new_bio_slab_max
* sizeof(struct bio_slab
),
106 bio_slab_max
= new_bio_slab_max
;
107 bio_slabs
= new_bio_slabs
;
110 entry
= bio_slab_nr
++;
112 bslab
= &bio_slabs
[entry
];
114 snprintf(bslab
->name
, sizeof(bslab
->name
), "bio-%d", entry
);
115 slab
= kmem_cache_create(bslab
->name
, sz
, ARCH_KMALLOC_MINALIGN
,
116 SLAB_HWCACHE_ALIGN
, NULL
);
122 bslab
->slab_size
= sz
;
124 mutex_unlock(&bio_slab_lock
);
128 static void bio_put_slab(struct bio_set
*bs
)
130 struct bio_slab
*bslab
= NULL
;
133 mutex_lock(&bio_slab_lock
);
135 for (i
= 0; i
< bio_slab_nr
; i
++) {
136 if (bs
->bio_slab
== bio_slabs
[i
].slab
) {
137 bslab
= &bio_slabs
[i
];
142 if (WARN(!bslab
, KERN_ERR
"bio: unable to find slab!\n"))
145 WARN_ON(!bslab
->slab_ref
);
147 if (--bslab
->slab_ref
)
150 kmem_cache_destroy(bslab
->slab
);
154 mutex_unlock(&bio_slab_lock
);
157 unsigned int bvec_nr_vecs(unsigned short idx
)
159 return bvec_slabs
[idx
].nr_vecs
;
162 void bvec_free(mempool_t
*pool
, struct bio_vec
*bv
, unsigned int idx
)
168 BIO_BUG_ON(idx
>= BVEC_POOL_NR
);
170 if (idx
== BVEC_POOL_MAX
) {
171 mempool_free(bv
, pool
);
173 struct biovec_slab
*bvs
= bvec_slabs
+ idx
;
175 kmem_cache_free(bvs
->slab
, bv
);
179 struct bio_vec
*bvec_alloc(gfp_t gfp_mask
, int nr
, unsigned long *idx
,
185 * see comment near bvec_array define!
203 case 129 ... BIO_MAX_PAGES
:
211 * idx now points to the pool we want to allocate from. only the
212 * 1-vec entry pool is mempool backed.
214 if (*idx
== BVEC_POOL_MAX
) {
216 bvl
= mempool_alloc(pool
, gfp_mask
);
218 struct biovec_slab
*bvs
= bvec_slabs
+ *idx
;
219 gfp_t __gfp_mask
= gfp_mask
& ~(__GFP_DIRECT_RECLAIM
| __GFP_IO
);
222 * Make this allocation restricted and don't dump info on
223 * allocation failures, since we'll fallback to the mempool
224 * in case of failure.
226 __gfp_mask
|= __GFP_NOMEMALLOC
| __GFP_NORETRY
| __GFP_NOWARN
;
229 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
230 * is set, retry with the 1-entry mempool
232 bvl
= kmem_cache_alloc(bvs
->slab
, __gfp_mask
);
233 if (unlikely(!bvl
&& (gfp_mask
& __GFP_DIRECT_RECLAIM
))) {
234 *idx
= BVEC_POOL_MAX
;
243 void bio_uninit(struct bio
*bio
)
245 bio_disassociate_task(bio
);
247 EXPORT_SYMBOL(bio_uninit
);
249 static void bio_free(struct bio
*bio
)
251 struct bio_set
*bs
= bio
->bi_pool
;
257 bvec_free(bs
->bvec_pool
, bio
->bi_io_vec
, BVEC_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() */
273 * Users of this function have their own bio allocation. Subsequently,
274 * they must remember to pair any call to bio_init() with bio_uninit()
275 * when IO has completed, or when the bio is released.
277 void bio_init(struct bio
*bio
, struct bio_vec
*table
,
278 unsigned short max_vecs
)
280 memset(bio
, 0, sizeof(*bio
));
281 atomic_set(&bio
->__bi_remaining
, 1);
282 atomic_set(&bio
->__bi_cnt
, 1);
284 bio
->bi_io_vec
= table
;
285 bio
->bi_max_vecs
= max_vecs
;
287 EXPORT_SYMBOL(bio_init
);
290 * bio_reset - reinitialize a bio
294 * After calling bio_reset(), @bio will be in the same state as a freshly
295 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
296 * preserved are the ones that are initialized by bio_alloc_bioset(). See
297 * comment in struct bio.
299 void bio_reset(struct bio
*bio
)
301 unsigned long flags
= bio
->bi_flags
& (~0UL << BIO_RESET_BITS
);
305 memset(bio
, 0, BIO_RESET_BYTES
);
306 bio
->bi_flags
= flags
;
307 atomic_set(&bio
->__bi_remaining
, 1);
309 EXPORT_SYMBOL(bio_reset
);
311 static struct bio
*__bio_chain_endio(struct bio
*bio
)
313 struct bio
*parent
= bio
->bi_private
;
315 if (!parent
->bi_status
)
316 parent
->bi_status
= bio
->bi_status
;
321 static void bio_chain_endio(struct bio
*bio
)
323 bio_endio(__bio_chain_endio(bio
));
327 * bio_chain - chain bio completions
328 * @bio: the target bio
329 * @parent: the @bio's parent bio
331 * The caller won't have a bi_end_io called when @bio completes - instead,
332 * @parent's bi_end_io won't be called until both @parent and @bio have
333 * completed; the chained bio will also be freed when it completes.
335 * The caller must not set bi_private or bi_end_io in @bio.
337 void bio_chain(struct bio
*bio
, struct bio
*parent
)
339 BUG_ON(bio
->bi_private
|| bio
->bi_end_io
);
341 bio
->bi_private
= parent
;
342 bio
->bi_end_io
= bio_chain_endio
;
343 bio_inc_remaining(parent
);
345 EXPORT_SYMBOL(bio_chain
);
347 static void bio_alloc_rescue(struct work_struct
*work
)
349 struct bio_set
*bs
= container_of(work
, struct bio_set
, rescue_work
);
353 spin_lock(&bs
->rescue_lock
);
354 bio
= bio_list_pop(&bs
->rescue_list
);
355 spin_unlock(&bs
->rescue_lock
);
360 generic_make_request(bio
);
364 static void punt_bios_to_rescuer(struct bio_set
*bs
)
366 struct bio_list punt
, nopunt
;
369 if (WARN_ON_ONCE(!bs
->rescue_workqueue
))
372 * In order to guarantee forward progress we must punt only bios that
373 * were allocated from this bio_set; otherwise, if there was a bio on
374 * there for a stacking driver higher up in the stack, processing it
375 * could require allocating bios from this bio_set, and doing that from
376 * our own rescuer would be bad.
378 * Since bio lists are singly linked, pop them all instead of trying to
379 * remove from the middle of the list:
382 bio_list_init(&punt
);
383 bio_list_init(&nopunt
);
385 while ((bio
= bio_list_pop(¤t
->bio_list
[0])))
386 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
387 current
->bio_list
[0] = nopunt
;
389 bio_list_init(&nopunt
);
390 while ((bio
= bio_list_pop(¤t
->bio_list
[1])))
391 bio_list_add(bio
->bi_pool
== bs
? &punt
: &nopunt
, bio
);
392 current
->bio_list
[1] = nopunt
;
394 spin_lock(&bs
->rescue_lock
);
395 bio_list_merge(&bs
->rescue_list
, &punt
);
396 spin_unlock(&bs
->rescue_lock
);
398 queue_work(bs
->rescue_workqueue
, &bs
->rescue_work
);
402 * bio_alloc_bioset - allocate a bio for I/O
403 * @gfp_mask: the GFP_ mask given to the slab allocator
404 * @nr_iovecs: number of iovecs to pre-allocate
405 * @bs: the bio_set to allocate from.
408 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
409 * backed by the @bs's mempool.
411 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
412 * always be able to allocate a bio. This is due to the mempool guarantees.
413 * To make this work, callers must never allocate more than 1 bio at a time
414 * from this pool. Callers that need to allocate more than 1 bio must always
415 * submit the previously allocated bio for IO before attempting to allocate
416 * a new one. Failure to do so can cause deadlocks under memory pressure.
418 * Note that when running under generic_make_request() (i.e. any block
419 * driver), bios are not submitted until after you return - see the code in
420 * generic_make_request() that converts recursion into iteration, to prevent
423 * This would normally mean allocating multiple bios under
424 * generic_make_request() would be susceptible to deadlocks, but we have
425 * deadlock avoidance code that resubmits any blocked bios from a rescuer
428 * However, we do not guarantee forward progress for allocations from other
429 * mempools. Doing multiple allocations from the same mempool under
430 * generic_make_request() should be avoided - instead, use bio_set's front_pad
431 * for per bio allocations.
434 * Pointer to new bio on success, NULL on failure.
436 struct bio
*bio_alloc_bioset(gfp_t gfp_mask
, unsigned int nr_iovecs
,
439 gfp_t saved_gfp
= gfp_mask
;
441 unsigned inline_vecs
;
442 struct bio_vec
*bvl
= NULL
;
447 if (nr_iovecs
> UIO_MAXIOV
)
450 p
= kmalloc(sizeof(struct bio
) +
451 nr_iovecs
* sizeof(struct bio_vec
),
454 inline_vecs
= nr_iovecs
;
456 /* should not use nobvec bioset for nr_iovecs > 0 */
457 if (WARN_ON_ONCE(!bs
->bvec_pool
&& nr_iovecs
> 0))
460 * generic_make_request() converts recursion to iteration; this
461 * means if we're running beneath it, any bios we allocate and
462 * submit will not be submitted (and thus freed) until after we
465 * This exposes us to a potential deadlock if we allocate
466 * multiple bios from the same bio_set() while running
467 * underneath generic_make_request(). If we were to allocate
468 * multiple bios (say a stacking block driver that was splitting
469 * bios), we would deadlock if we exhausted the mempool's
472 * We solve this, and guarantee forward progress, with a rescuer
473 * workqueue per bio_set. If we go to allocate and there are
474 * bios on current->bio_list, we first try the allocation
475 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
476 * bios we would be blocking to the rescuer workqueue before
477 * we retry with the original gfp_flags.
480 if (current
->bio_list
&&
481 (!bio_list_empty(¤t
->bio_list
[0]) ||
482 !bio_list_empty(¤t
->bio_list
[1])) &&
483 bs
->rescue_workqueue
)
484 gfp_mask
&= ~__GFP_DIRECT_RECLAIM
;
486 p
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
487 if (!p
&& gfp_mask
!= saved_gfp
) {
488 punt_bios_to_rescuer(bs
);
489 gfp_mask
= saved_gfp
;
490 p
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
493 front_pad
= bs
->front_pad
;
494 inline_vecs
= BIO_INLINE_VECS
;
501 bio_init(bio
, NULL
, 0);
503 if (nr_iovecs
> inline_vecs
) {
504 unsigned long idx
= 0;
506 bvl
= bvec_alloc(gfp_mask
, nr_iovecs
, &idx
, bs
->bvec_pool
);
507 if (!bvl
&& gfp_mask
!= saved_gfp
) {
508 punt_bios_to_rescuer(bs
);
509 gfp_mask
= saved_gfp
;
510 bvl
= bvec_alloc(gfp_mask
, nr_iovecs
, &idx
, bs
->bvec_pool
);
516 bio
->bi_flags
|= idx
<< BVEC_POOL_OFFSET
;
517 } else if (nr_iovecs
) {
518 bvl
= bio
->bi_inline_vecs
;
522 bio
->bi_max_vecs
= nr_iovecs
;
523 bio
->bi_io_vec
= bvl
;
527 mempool_free(p
, bs
->bio_pool
);
530 EXPORT_SYMBOL(bio_alloc_bioset
);
532 void zero_fill_bio(struct bio
*bio
)
536 struct bvec_iter iter
;
538 bio_for_each_segment(bv
, bio
, iter
) {
539 char *data
= bvec_kmap_irq(&bv
, &flags
);
540 memset(data
, 0, bv
.bv_len
);
541 flush_dcache_page(bv
.bv_page
);
542 bvec_kunmap_irq(data
, &flags
);
545 EXPORT_SYMBOL(zero_fill_bio
);
548 * bio_put - release a reference to a bio
549 * @bio: bio to release reference to
552 * Put a reference to a &struct bio, either one you have gotten with
553 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
555 void bio_put(struct bio
*bio
)
557 if (!bio_flagged(bio
, BIO_REFFED
))
560 BIO_BUG_ON(!atomic_read(&bio
->__bi_cnt
));
565 if (atomic_dec_and_test(&bio
->__bi_cnt
))
569 EXPORT_SYMBOL(bio_put
);
571 inline int bio_phys_segments(struct request_queue
*q
, struct bio
*bio
)
573 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
574 blk_recount_segments(q
, bio
);
576 return bio
->bi_phys_segments
;
578 EXPORT_SYMBOL(bio_phys_segments
);
581 * __bio_clone_fast - clone a bio that shares the original bio's biovec
582 * @bio: destination bio
583 * @bio_src: bio to clone
585 * Clone a &bio. Caller will own the returned bio, but not
586 * the actual data it points to. Reference count of returned
589 * Caller must ensure that @bio_src is not freed before @bio.
591 void __bio_clone_fast(struct bio
*bio
, struct bio
*bio_src
)
593 BUG_ON(bio
->bi_pool
&& BVEC_POOL_IDX(bio
));
596 * most users will be overriding ->bi_disk with a new target,
597 * so we don't set nor calculate new physical/hw segment counts here
599 bio
->bi_disk
= bio_src
->bi_disk
;
600 bio
->bi_partno
= bio_src
->bi_partno
;
601 bio_set_flag(bio
, BIO_CLONED
);
602 if (bio_flagged(bio_src
, BIO_THROTTLED
))
603 bio_set_flag(bio
, BIO_THROTTLED
);
604 bio
->bi_opf
= bio_src
->bi_opf
;
605 bio
->bi_write_hint
= bio_src
->bi_write_hint
;
606 bio
->bi_iter
= bio_src
->bi_iter
;
607 bio
->bi_io_vec
= bio_src
->bi_io_vec
;
609 bio_clone_blkcg_association(bio
, bio_src
);
611 EXPORT_SYMBOL(__bio_clone_fast
);
614 * bio_clone_fast - clone a bio that shares the original bio's biovec
616 * @gfp_mask: allocation priority
617 * @bs: bio_set to allocate from
619 * Like __bio_clone_fast, only also allocates the returned bio
621 struct bio
*bio_clone_fast(struct bio
*bio
, gfp_t gfp_mask
, struct bio_set
*bs
)
625 b
= bio_alloc_bioset(gfp_mask
, 0, bs
);
629 __bio_clone_fast(b
, bio
);
631 if (bio_integrity(bio
)) {
634 ret
= bio_integrity_clone(b
, bio
, gfp_mask
);
644 EXPORT_SYMBOL(bio_clone_fast
);
647 * bio_clone_bioset - clone a bio
648 * @bio_src: bio to clone
649 * @gfp_mask: allocation priority
650 * @bs: bio_set to allocate from
652 * Clone bio. Caller will own the returned bio, but not the actual data it
653 * points to. Reference count of returned bio will be one.
655 struct bio
*bio_clone_bioset(struct bio
*bio_src
, gfp_t gfp_mask
,
658 struct bvec_iter iter
;
663 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
664 * bio_src->bi_io_vec to bio->bi_io_vec.
666 * We can't do that anymore, because:
668 * - The point of cloning the biovec is to produce a bio with a biovec
669 * the caller can modify: bi_idx and bi_bvec_done should be 0.
671 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
672 * we tried to clone the whole thing bio_alloc_bioset() would fail.
673 * But the clone should succeed as long as the number of biovecs we
674 * actually need to allocate is fewer than BIO_MAX_PAGES.
676 * - Lastly, bi_vcnt should not be looked at or relied upon by code
677 * that does not own the bio - reason being drivers don't use it for
678 * iterating over the biovec anymore, so expecting it to be kept up
679 * to date (i.e. for clones that share the parent biovec) is just
680 * asking for trouble and would force extra work on
681 * __bio_clone_fast() anyways.
684 bio
= bio_alloc_bioset(gfp_mask
, bio_segments(bio_src
), bs
);
687 bio
->bi_disk
= bio_src
->bi_disk
;
688 bio
->bi_opf
= bio_src
->bi_opf
;
689 bio
->bi_write_hint
= bio_src
->bi_write_hint
;
690 bio
->bi_iter
.bi_sector
= bio_src
->bi_iter
.bi_sector
;
691 bio
->bi_iter
.bi_size
= bio_src
->bi_iter
.bi_size
;
693 switch (bio_op(bio
)) {
695 case REQ_OP_SECURE_ERASE
:
696 case REQ_OP_WRITE_ZEROES
:
698 case REQ_OP_WRITE_SAME
:
699 bio
->bi_io_vec
[bio
->bi_vcnt
++] = bio_src
->bi_io_vec
[0];
702 bio_for_each_segment(bv
, bio_src
, iter
)
703 bio
->bi_io_vec
[bio
->bi_vcnt
++] = bv
;
707 if (bio_integrity(bio_src
)) {
710 ret
= bio_integrity_clone(bio
, bio_src
, gfp_mask
);
717 bio_clone_blkcg_association(bio
, bio_src
);
721 EXPORT_SYMBOL(bio_clone_bioset
);
724 * bio_add_pc_page - attempt to add page to bio
725 * @q: the target queue
726 * @bio: destination bio
728 * @len: vec entry length
729 * @offset: vec entry offset
731 * Attempt to add a page to the bio_vec maplist. This can fail for a
732 * number of reasons, such as the bio being full or target block device
733 * limitations. The target block device must allow bio's up to PAGE_SIZE,
734 * so it is always possible to add a single page to an empty bio.
736 * This should only be used by REQ_PC bios.
738 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
, struct page
739 *page
, unsigned int len
, unsigned int offset
)
741 int retried_segments
= 0;
742 struct bio_vec
*bvec
;
745 * cloned bio must not modify vec list
747 if (unlikely(bio_flagged(bio
, BIO_CLONED
)))
750 if (((bio
->bi_iter
.bi_size
+ len
) >> 9) > queue_max_hw_sectors(q
))
754 * For filesystems with a blocksize smaller than the pagesize
755 * we will often be called with the same page as last time and
756 * a consecutive offset. Optimize this special case.
758 if (bio
->bi_vcnt
> 0) {
759 struct bio_vec
*prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
761 if (page
== prev
->bv_page
&&
762 offset
== prev
->bv_offset
+ prev
->bv_len
) {
764 bio
->bi_iter
.bi_size
+= len
;
769 * If the queue doesn't support SG gaps and adding this
770 * offset would create a gap, disallow it.
772 if (bvec_gap_to_prev(q
, prev
, offset
))
776 if (bio
->bi_vcnt
>= bio
->bi_max_vecs
)
780 * setup the new entry, we might clear it again later if we
781 * cannot add the page
783 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
784 bvec
->bv_page
= page
;
786 bvec
->bv_offset
= offset
;
788 bio
->bi_phys_segments
++;
789 bio
->bi_iter
.bi_size
+= len
;
792 * Perform a recount if the number of segments is greater
793 * than queue_max_segments(q).
796 while (bio
->bi_phys_segments
> queue_max_segments(q
)) {
798 if (retried_segments
)
801 retried_segments
= 1;
802 blk_recount_segments(q
, bio
);
805 /* If we may be able to merge these biovecs, force a recount */
806 if (bio
->bi_vcnt
> 1 && (BIOVEC_PHYS_MERGEABLE(bvec
-1, bvec
)))
807 bio_clear_flag(bio
, BIO_SEG_VALID
);
813 bvec
->bv_page
= NULL
;
817 bio
->bi_iter
.bi_size
-= len
;
818 blk_recount_segments(q
, bio
);
821 EXPORT_SYMBOL(bio_add_pc_page
);
824 * bio_add_page - attempt to add page to bio
825 * @bio: destination bio
827 * @len: vec entry length
828 * @offset: vec entry offset
830 * Attempt to add a page to the bio_vec maplist. This will only fail
831 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
833 int bio_add_page(struct bio
*bio
, struct page
*page
,
834 unsigned int len
, unsigned int offset
)
839 * cloned bio must not modify vec list
841 if (WARN_ON_ONCE(bio_flagged(bio
, BIO_CLONED
)))
845 * For filesystems with a blocksize smaller than the pagesize
846 * we will often be called with the same page as last time and
847 * a consecutive offset. Optimize this special case.
849 if (bio
->bi_vcnt
> 0) {
850 bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
852 if (page
== bv
->bv_page
&&
853 offset
== bv
->bv_offset
+ bv
->bv_len
) {
859 if (bio
->bi_vcnt
>= bio
->bi_max_vecs
)
862 bv
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
865 bv
->bv_offset
= offset
;
869 bio
->bi_iter
.bi_size
+= len
;
872 EXPORT_SYMBOL(bio_add_page
);
875 * bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
876 * @bio: bio to add pages to
877 * @iter: iov iterator describing the region to be mapped
879 * Pins as many pages from *iter and appends them to @bio's bvec array. The
880 * pages will have to be released using put_page() when done.
882 int bio_iov_iter_get_pages(struct bio
*bio
, struct iov_iter
*iter
)
884 unsigned short nr_pages
= bio
->bi_max_vecs
- bio
->bi_vcnt
;
885 struct bio_vec
*bv
= bio
->bi_io_vec
+ bio
->bi_vcnt
;
886 struct page
**pages
= (struct page
**)bv
;
890 size
= iov_iter_get_pages(iter
, pages
, LONG_MAX
, nr_pages
, &offset
);
891 if (unlikely(size
<= 0))
892 return size
? size
: -EFAULT
;
893 nr_pages
= (size
+ offset
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
896 * Deep magic below: We need to walk the pinned pages backwards
897 * because we are abusing the space allocated for the bio_vecs
898 * for the page array. Because the bio_vecs are larger than the
899 * page pointers by definition this will always work. But it also
900 * means we can't use bio_add_page, so any changes to it's semantics
901 * need to be reflected here as well.
903 bio
->bi_iter
.bi_size
+= size
;
904 bio
->bi_vcnt
+= nr_pages
;
906 diff
= (nr_pages
* PAGE_SIZE
- offset
) - size
;
908 bv
[nr_pages
].bv_page
= pages
[nr_pages
];
909 bv
[nr_pages
].bv_len
= PAGE_SIZE
;
910 bv
[nr_pages
].bv_offset
= 0;
913 bv
[0].bv_offset
+= offset
;
914 bv
[0].bv_len
-= offset
;
916 bv
[bio
->bi_vcnt
- 1].bv_len
-= diff
;
918 iov_iter_advance(iter
, size
);
921 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages
);
923 struct submit_bio_ret
{
924 struct completion event
;
928 static void submit_bio_wait_endio(struct bio
*bio
)
930 struct submit_bio_ret
*ret
= bio
->bi_private
;
932 ret
->error
= blk_status_to_errno(bio
->bi_status
);
933 complete(&ret
->event
);
937 * submit_bio_wait - submit a bio, and wait until it completes
938 * @bio: The &struct bio which describes the I/O
940 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
941 * bio_endio() on failure.
943 * WARNING: Unlike to how submit_bio() is usually used, this function does not
944 * result in bio reference to be consumed. The caller must drop the reference
947 int submit_bio_wait(struct bio
*bio
)
949 struct submit_bio_ret ret
;
951 init_completion(&ret
.event
);
952 bio
->bi_private
= &ret
;
953 bio
->bi_end_io
= submit_bio_wait_endio
;
954 bio
->bi_opf
|= REQ_SYNC
;
956 wait_for_completion_io(&ret
.event
);
960 EXPORT_SYMBOL(submit_bio_wait
);
963 * bio_advance - increment/complete a bio by some number of bytes
964 * @bio: bio to advance
965 * @bytes: number of bytes to complete
967 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
968 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
969 * be updated on the last bvec as well.
971 * @bio will then represent the remaining, uncompleted portion of the io.
973 void bio_advance(struct bio
*bio
, unsigned bytes
)
975 if (bio_integrity(bio
))
976 bio_integrity_advance(bio
, bytes
);
978 bio_advance_iter(bio
, &bio
->bi_iter
, bytes
);
980 EXPORT_SYMBOL(bio_advance
);
983 * bio_alloc_pages - allocates a single page for each bvec in a bio
984 * @bio: bio to allocate pages for
985 * @gfp_mask: flags for allocation
987 * Allocates pages up to @bio->bi_vcnt.
989 * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
992 int bio_alloc_pages(struct bio
*bio
, gfp_t gfp_mask
)
997 bio_for_each_segment_all(bv
, bio
, i
) {
998 bv
->bv_page
= alloc_page(gfp_mask
);
1000 while (--bv
>= bio
->bi_io_vec
)
1001 __free_page(bv
->bv_page
);
1008 EXPORT_SYMBOL(bio_alloc_pages
);
1011 * bio_copy_data - copy contents of data buffers from one chain of bios to
1013 * @src: source bio list
1014 * @dst: destination bio list
1016 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
1017 * @src and @dst as linked lists of bios.
1019 * Stops when it reaches the end of either @src or @dst - that is, copies
1020 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1022 void bio_copy_data(struct bio
*dst
, struct bio
*src
)
1024 struct bvec_iter src_iter
, dst_iter
;
1025 struct bio_vec src_bv
, dst_bv
;
1026 void *src_p
, *dst_p
;
1029 src_iter
= src
->bi_iter
;
1030 dst_iter
= dst
->bi_iter
;
1033 if (!src_iter
.bi_size
) {
1038 src_iter
= src
->bi_iter
;
1041 if (!dst_iter
.bi_size
) {
1046 dst_iter
= dst
->bi_iter
;
1049 src_bv
= bio_iter_iovec(src
, src_iter
);
1050 dst_bv
= bio_iter_iovec(dst
, dst_iter
);
1052 bytes
= min(src_bv
.bv_len
, dst_bv
.bv_len
);
1054 src_p
= kmap_atomic(src_bv
.bv_page
);
1055 dst_p
= kmap_atomic(dst_bv
.bv_page
);
1057 memcpy(dst_p
+ dst_bv
.bv_offset
,
1058 src_p
+ src_bv
.bv_offset
,
1061 kunmap_atomic(dst_p
);
1062 kunmap_atomic(src_p
);
1064 bio_advance_iter(src
, &src_iter
, bytes
);
1065 bio_advance_iter(dst
, &dst_iter
, bytes
);
1068 EXPORT_SYMBOL(bio_copy_data
);
1070 struct bio_map_data
{
1072 struct iov_iter iter
;
1076 static struct bio_map_data
*bio_alloc_map_data(unsigned int iov_count
,
1079 if (iov_count
> UIO_MAXIOV
)
1082 return kmalloc(sizeof(struct bio_map_data
) +
1083 sizeof(struct iovec
) * iov_count
, gfp_mask
);
1087 * bio_copy_from_iter - copy all pages from iov_iter to bio
1088 * @bio: The &struct bio which describes the I/O as destination
1089 * @iter: iov_iter as source
1091 * Copy all pages from iov_iter to bio.
1092 * Returns 0 on success, or error on failure.
1094 static int bio_copy_from_iter(struct bio
*bio
, struct iov_iter iter
)
1097 struct bio_vec
*bvec
;
1099 bio_for_each_segment_all(bvec
, bio
, i
) {
1102 ret
= copy_page_from_iter(bvec
->bv_page
,
1107 if (!iov_iter_count(&iter
))
1110 if (ret
< bvec
->bv_len
)
1118 * bio_copy_to_iter - copy all pages from bio to iov_iter
1119 * @bio: The &struct bio which describes the I/O as source
1120 * @iter: iov_iter as destination
1122 * Copy all pages from bio to iov_iter.
1123 * Returns 0 on success, or error on failure.
1125 static int bio_copy_to_iter(struct bio
*bio
, struct iov_iter iter
)
1128 struct bio_vec
*bvec
;
1130 bio_for_each_segment_all(bvec
, bio
, i
) {
1133 ret
= copy_page_to_iter(bvec
->bv_page
,
1138 if (!iov_iter_count(&iter
))
1141 if (ret
< bvec
->bv_len
)
1148 void bio_free_pages(struct bio
*bio
)
1150 struct bio_vec
*bvec
;
1153 bio_for_each_segment_all(bvec
, bio
, i
)
1154 __free_page(bvec
->bv_page
);
1156 EXPORT_SYMBOL(bio_free_pages
);
1159 * bio_uncopy_user - finish previously mapped bio
1160 * @bio: bio being terminated
1162 * Free pages allocated from bio_copy_user_iov() and write back data
1163 * to user space in case of a read.
1165 int bio_uncopy_user(struct bio
*bio
)
1167 struct bio_map_data
*bmd
= bio
->bi_private
;
1170 if (!bio_flagged(bio
, BIO_NULL_MAPPED
)) {
1172 * if we're in a workqueue, the request is orphaned, so
1173 * don't copy into a random user address space, just free
1174 * and return -EINTR so user space doesn't expect any data.
1178 else if (bio_data_dir(bio
) == READ
)
1179 ret
= bio_copy_to_iter(bio
, bmd
->iter
);
1180 if (bmd
->is_our_pages
)
1181 bio_free_pages(bio
);
1189 * bio_copy_user_iov - copy user data to bio
1190 * @q: destination block queue
1191 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1192 * @iter: iovec iterator
1193 * @gfp_mask: memory allocation flags
1195 * Prepares and returns a bio for indirect user io, bouncing data
1196 * to/from kernel pages as necessary. Must be paired with
1197 * call bio_uncopy_user() on io completion.
1199 struct bio
*bio_copy_user_iov(struct request_queue
*q
,
1200 struct rq_map_data
*map_data
,
1201 const struct iov_iter
*iter
,
1204 struct bio_map_data
*bmd
;
1209 unsigned int len
= iter
->count
;
1210 unsigned int offset
= map_data
? offset_in_page(map_data
->offset
) : 0;
1212 for (i
= 0; i
< iter
->nr_segs
; i
++) {
1213 unsigned long uaddr
;
1215 unsigned long start
;
1217 uaddr
= (unsigned long) iter
->iov
[i
].iov_base
;
1218 end
= (uaddr
+ iter
->iov
[i
].iov_len
+ PAGE_SIZE
- 1)
1220 start
= uaddr
>> PAGE_SHIFT
;
1226 return ERR_PTR(-EINVAL
);
1228 nr_pages
+= end
- start
;
1234 bmd
= bio_alloc_map_data(iter
->nr_segs
, gfp_mask
);
1236 return ERR_PTR(-ENOMEM
);
1239 * We need to do a deep copy of the iov_iter including the iovecs.
1240 * The caller provided iov might point to an on-stack or otherwise
1243 bmd
->is_our_pages
= map_data
? 0 : 1;
1244 memcpy(bmd
->iov
, iter
->iov
, sizeof(struct iovec
) * iter
->nr_segs
);
1246 bmd
->iter
.iov
= bmd
->iov
;
1249 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1256 nr_pages
= 1 << map_data
->page_order
;
1257 i
= map_data
->offset
/ PAGE_SIZE
;
1260 unsigned int bytes
= PAGE_SIZE
;
1268 if (i
== map_data
->nr_entries
* nr_pages
) {
1273 page
= map_data
->pages
[i
/ nr_pages
];
1274 page
+= (i
% nr_pages
);
1278 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
1285 if (bio_add_pc_page(q
, bio
, page
, bytes
, offset
) < bytes
)
1298 if (((iter
->type
& WRITE
) && (!map_data
|| !map_data
->null_mapped
)) ||
1299 (map_data
&& map_data
->from_user
)) {
1300 ret
= bio_copy_from_iter(bio
, *iter
);
1305 bio
->bi_private
= bmd
;
1309 bio_free_pages(bio
);
1313 return ERR_PTR(ret
);
1317 * bio_map_user_iov - map user iovec into bio
1318 * @q: the struct request_queue for the bio
1319 * @iter: iovec iterator
1320 * @gfp_mask: memory allocation flags
1322 * Map the user space address into a bio suitable for io to a block
1323 * device. Returns an error pointer in case of error.
1325 struct bio
*bio_map_user_iov(struct request_queue
*q
,
1326 const struct iov_iter
*iter
,
1331 struct page
**pages
;
1337 struct bio_vec
*bvec
;
1339 iov_for_each(iov
, i
, *iter
) {
1340 unsigned long uaddr
= (unsigned long) iov
.iov_base
;
1341 unsigned long len
= iov
.iov_len
;
1342 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1343 unsigned long start
= uaddr
>> PAGE_SHIFT
;
1349 return ERR_PTR(-EINVAL
);
1351 nr_pages
+= end
- start
;
1353 * buffer must be aligned to at least logical block size for now
1355 if (uaddr
& queue_dma_alignment(q
))
1356 return ERR_PTR(-EINVAL
);
1360 return ERR_PTR(-EINVAL
);
1362 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1364 return ERR_PTR(-ENOMEM
);
1367 pages
= kcalloc(nr_pages
, sizeof(struct page
*), gfp_mask
);
1371 iov_for_each(iov
, i
, *iter
) {
1372 unsigned long uaddr
= (unsigned long) iov
.iov_base
;
1373 unsigned long len
= iov
.iov_len
;
1374 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1375 unsigned long start
= uaddr
>> PAGE_SHIFT
;
1376 const int local_nr_pages
= end
- start
;
1377 const int page_limit
= cur_page
+ local_nr_pages
;
1379 ret
= get_user_pages_fast(uaddr
, local_nr_pages
,
1380 (iter
->type
& WRITE
) != WRITE
,
1382 if (unlikely(ret
< local_nr_pages
)) {
1383 for (j
= cur_page
; j
< page_limit
; j
++) {
1392 offset
= offset_in_page(uaddr
);
1393 for (j
= cur_page
; j
< page_limit
; j
++) {
1394 unsigned int bytes
= PAGE_SIZE
- offset
;
1395 unsigned short prev_bi_vcnt
= bio
->bi_vcnt
;
1406 if (bio_add_pc_page(q
, bio
, pages
[j
], bytes
, offset
) <
1411 * check if vector was merged with previous
1412 * drop page reference if needed
1414 if (bio
->bi_vcnt
== prev_bi_vcnt
)
1423 * release the pages we didn't map into the bio, if any
1425 while (j
< page_limit
)
1426 put_page(pages
[j
++]);
1431 bio_set_flag(bio
, BIO_USER_MAPPED
);
1434 * subtle -- if bio_map_user_iov() ended up bouncing a bio,
1435 * it would normally disappear when its bi_end_io is run.
1436 * however, we need it for the unmap, so grab an extra
1443 bio_for_each_segment_all(bvec
, bio
, j
) {
1444 put_page(bvec
->bv_page
);
1449 return ERR_PTR(ret
);
1452 static void __bio_unmap_user(struct bio
*bio
)
1454 struct bio_vec
*bvec
;
1458 * make sure we dirty pages we wrote to
1460 bio_for_each_segment_all(bvec
, bio
, i
) {
1461 if (bio_data_dir(bio
) == READ
)
1462 set_page_dirty_lock(bvec
->bv_page
);
1464 put_page(bvec
->bv_page
);
1471 * bio_unmap_user - unmap a bio
1472 * @bio: the bio being unmapped
1474 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from
1477 * bio_unmap_user() may sleep.
1479 void bio_unmap_user(struct bio
*bio
)
1481 __bio_unmap_user(bio
);
1485 static void bio_map_kern_endio(struct bio
*bio
)
1491 * bio_map_kern - map kernel address into bio
1492 * @q: the struct request_queue for the bio
1493 * @data: pointer to buffer to map
1494 * @len: length in bytes
1495 * @gfp_mask: allocation flags for bio allocation
1497 * Map the kernel address into a bio suitable for io to a block
1498 * device. Returns an error pointer in case of error.
1500 struct bio
*bio_map_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1503 unsigned long kaddr
= (unsigned long)data
;
1504 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1505 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1506 const int nr_pages
= end
- start
;
1510 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1512 return ERR_PTR(-ENOMEM
);
1514 offset
= offset_in_page(kaddr
);
1515 for (i
= 0; i
< nr_pages
; i
++) {
1516 unsigned int bytes
= PAGE_SIZE
- offset
;
1524 if (bio_add_pc_page(q
, bio
, virt_to_page(data
), bytes
,
1526 /* we don't support partial mappings */
1528 return ERR_PTR(-EINVAL
);
1536 bio
->bi_end_io
= bio_map_kern_endio
;
1539 EXPORT_SYMBOL(bio_map_kern
);
1541 static void bio_copy_kern_endio(struct bio
*bio
)
1543 bio_free_pages(bio
);
1547 static void bio_copy_kern_endio_read(struct bio
*bio
)
1549 char *p
= bio
->bi_private
;
1550 struct bio_vec
*bvec
;
1553 bio_for_each_segment_all(bvec
, bio
, i
) {
1554 memcpy(p
, page_address(bvec
->bv_page
), bvec
->bv_len
);
1558 bio_copy_kern_endio(bio
);
1562 * bio_copy_kern - copy kernel address into bio
1563 * @q: the struct request_queue for the bio
1564 * @data: pointer to buffer to copy
1565 * @len: length in bytes
1566 * @gfp_mask: allocation flags for bio and page allocation
1567 * @reading: data direction is READ
1569 * copy the kernel address into a bio suitable for io to a block
1570 * device. Returns an error pointer in case of error.
1572 struct bio
*bio_copy_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1573 gfp_t gfp_mask
, int reading
)
1575 unsigned long kaddr
= (unsigned long)data
;
1576 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1577 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1586 return ERR_PTR(-EINVAL
);
1588 nr_pages
= end
- start
;
1589 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1591 return ERR_PTR(-ENOMEM
);
1595 unsigned int bytes
= PAGE_SIZE
;
1600 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
1605 memcpy(page_address(page
), p
, bytes
);
1607 if (bio_add_pc_page(q
, bio
, page
, bytes
, 0) < bytes
)
1615 bio
->bi_end_io
= bio_copy_kern_endio_read
;
1616 bio
->bi_private
= data
;
1618 bio
->bi_end_io
= bio_copy_kern_endio
;
1624 bio_free_pages(bio
);
1626 return ERR_PTR(-ENOMEM
);
1630 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1631 * for performing direct-IO in BIOs.
1633 * The problem is that we cannot run set_page_dirty() from interrupt context
1634 * because the required locks are not interrupt-safe. So what we can do is to
1635 * mark the pages dirty _before_ performing IO. And in interrupt context,
1636 * check that the pages are still dirty. If so, fine. If not, redirty them
1637 * in process context.
1639 * We special-case compound pages here: normally this means reads into hugetlb
1640 * pages. The logic in here doesn't really work right for compound pages
1641 * because the VM does not uniformly chase down the head page in all cases.
1642 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1643 * handle them at all. So we skip compound pages here at an early stage.
1645 * Note that this code is very hard to test under normal circumstances because
1646 * direct-io pins the pages with get_user_pages(). This makes
1647 * is_page_cache_freeable return false, and the VM will not clean the pages.
1648 * But other code (eg, flusher threads) could clean the pages if they are mapped
1651 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1652 * deferred bio dirtying paths.
1656 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1658 void bio_set_pages_dirty(struct bio
*bio
)
1660 struct bio_vec
*bvec
;
1663 bio_for_each_segment_all(bvec
, bio
, i
) {
1664 struct page
*page
= bvec
->bv_page
;
1666 if (page
&& !PageCompound(page
))
1667 set_page_dirty_lock(page
);
1671 static void bio_release_pages(struct bio
*bio
)
1673 struct bio_vec
*bvec
;
1676 bio_for_each_segment_all(bvec
, bio
, i
) {
1677 struct page
*page
= bvec
->bv_page
;
1685 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1686 * If they are, then fine. If, however, some pages are clean then they must
1687 * have been written out during the direct-IO read. So we take another ref on
1688 * the BIO and the offending pages and re-dirty the pages in process context.
1690 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1691 * here on. It will run one put_page() against each page and will run one
1692 * bio_put() against the BIO.
1695 static void bio_dirty_fn(struct work_struct
*work
);
1697 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
1698 static DEFINE_SPINLOCK(bio_dirty_lock
);
1699 static struct bio
*bio_dirty_list
;
1702 * This runs in process context
1704 static void bio_dirty_fn(struct work_struct
*work
)
1706 unsigned long flags
;
1709 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1710 bio
= bio_dirty_list
;
1711 bio_dirty_list
= NULL
;
1712 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1715 struct bio
*next
= bio
->bi_private
;
1717 bio_set_pages_dirty(bio
);
1718 bio_release_pages(bio
);
1724 void bio_check_pages_dirty(struct bio
*bio
)
1726 struct bio_vec
*bvec
;
1727 int nr_clean_pages
= 0;
1730 bio_for_each_segment_all(bvec
, bio
, i
) {
1731 struct page
*page
= bvec
->bv_page
;
1733 if (PageDirty(page
) || PageCompound(page
)) {
1735 bvec
->bv_page
= NULL
;
1741 if (nr_clean_pages
) {
1742 unsigned long flags
;
1744 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1745 bio
->bi_private
= bio_dirty_list
;
1746 bio_dirty_list
= bio
;
1747 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1748 schedule_work(&bio_dirty_work
);
1754 void generic_start_io_acct(struct request_queue
*q
, int rw
,
1755 unsigned long sectors
, struct hd_struct
*part
)
1757 int cpu
= part_stat_lock();
1759 part_round_stats(q
, cpu
, part
);
1760 part_stat_inc(cpu
, part
, ios
[rw
]);
1761 part_stat_add(cpu
, part
, sectors
[rw
], sectors
);
1762 part_inc_in_flight(q
, part
, rw
);
1766 EXPORT_SYMBOL(generic_start_io_acct
);
1768 void generic_end_io_acct(struct request_queue
*q
, int rw
,
1769 struct hd_struct
*part
, unsigned long start_time
)
1771 unsigned long duration
= jiffies
- start_time
;
1772 int cpu
= part_stat_lock();
1774 part_stat_add(cpu
, part
, ticks
[rw
], duration
);
1775 part_round_stats(q
, cpu
, part
);
1776 part_dec_in_flight(q
, part
, rw
);
1780 EXPORT_SYMBOL(generic_end_io_acct
);
1782 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1783 void bio_flush_dcache_pages(struct bio
*bi
)
1785 struct bio_vec bvec
;
1786 struct bvec_iter iter
;
1788 bio_for_each_segment(bvec
, bi
, iter
)
1789 flush_dcache_page(bvec
.bv_page
);
1791 EXPORT_SYMBOL(bio_flush_dcache_pages
);
1794 static inline bool bio_remaining_done(struct bio
*bio
)
1797 * If we're not chaining, then ->__bi_remaining is always 1 and
1798 * we always end io on the first invocation.
1800 if (!bio_flagged(bio
, BIO_CHAIN
))
1803 BUG_ON(atomic_read(&bio
->__bi_remaining
) <= 0);
1805 if (atomic_dec_and_test(&bio
->__bi_remaining
)) {
1806 bio_clear_flag(bio
, BIO_CHAIN
);
1814 * bio_endio - end I/O on a bio
1818 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1819 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1820 * bio unless they own it and thus know that it has an end_io function.
1822 * bio_endio() can be called several times on a bio that has been chained
1823 * using bio_chain(). The ->bi_end_io() function will only be called the
1824 * last time. At this point the BLK_TA_COMPLETE tracing event will be
1825 * generated if BIO_TRACE_COMPLETION is set.
1827 void bio_endio(struct bio
*bio
)
1830 if (!bio_remaining_done(bio
))
1832 if (!bio_integrity_endio(bio
))
1836 * Need to have a real endio function for chained bios, otherwise
1837 * various corner cases will break (like stacking block devices that
1838 * save/restore bi_end_io) - however, we want to avoid unbounded
1839 * recursion and blowing the stack. Tail call optimization would
1840 * handle this, but compiling with frame pointers also disables
1841 * gcc's sibling call optimization.
1843 if (bio
->bi_end_io
== bio_chain_endio
) {
1844 bio
= __bio_chain_endio(bio
);
1848 if (bio
->bi_disk
&& bio_flagged(bio
, BIO_TRACE_COMPLETION
)) {
1849 trace_block_bio_complete(bio
->bi_disk
->queue
, bio
,
1850 blk_status_to_errno(bio
->bi_status
));
1851 bio_clear_flag(bio
, BIO_TRACE_COMPLETION
);
1854 blk_throtl_bio_endio(bio
);
1855 /* release cgroup info */
1858 bio
->bi_end_io(bio
);
1860 EXPORT_SYMBOL(bio_endio
);
1863 * bio_split - split a bio
1864 * @bio: bio to split
1865 * @sectors: number of sectors to split from the front of @bio
1867 * @bs: bio set to allocate from
1869 * Allocates and returns a new bio which represents @sectors from the start of
1870 * @bio, and updates @bio to represent the remaining sectors.
1872 * Unless this is a discard request the newly allocated bio will point
1873 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1874 * @bio is not freed before the split.
1876 struct bio
*bio_split(struct bio
*bio
, int sectors
,
1877 gfp_t gfp
, struct bio_set
*bs
)
1879 struct bio
*split
= NULL
;
1881 BUG_ON(sectors
<= 0);
1882 BUG_ON(sectors
>= bio_sectors(bio
));
1884 split
= bio_clone_fast(bio
, gfp
, bs
);
1888 split
->bi_iter
.bi_size
= sectors
<< 9;
1890 if (bio_integrity(split
))
1891 bio_integrity_trim(split
);
1893 bio_advance(bio
, split
->bi_iter
.bi_size
);
1895 if (bio_flagged(bio
, BIO_TRACE_COMPLETION
))
1896 bio_set_flag(bio
, BIO_TRACE_COMPLETION
);
1900 EXPORT_SYMBOL(bio_split
);
1903 * bio_trim - trim a bio
1905 * @offset: number of sectors to trim from the front of @bio
1906 * @size: size we want to trim @bio to, in sectors
1908 void bio_trim(struct bio
*bio
, int offset
, int size
)
1910 /* 'bio' is a cloned bio which we need to trim to match
1911 * the given offset and size.
1915 if (offset
== 0 && size
== bio
->bi_iter
.bi_size
)
1918 bio_clear_flag(bio
, BIO_SEG_VALID
);
1920 bio_advance(bio
, offset
<< 9);
1922 bio
->bi_iter
.bi_size
= size
;
1924 if (bio_integrity(bio
))
1925 bio_integrity_trim(bio
);
1928 EXPORT_SYMBOL_GPL(bio_trim
);
1931 * create memory pools for biovec's in a bio_set.
1932 * use the global biovec slabs created for general use.
1934 mempool_t
*biovec_create_pool(int pool_entries
)
1936 struct biovec_slab
*bp
= bvec_slabs
+ BVEC_POOL_MAX
;
1938 return mempool_create_slab_pool(pool_entries
, bp
->slab
);
1941 void bioset_free(struct bio_set
*bs
)
1943 if (bs
->rescue_workqueue
)
1944 destroy_workqueue(bs
->rescue_workqueue
);
1947 mempool_destroy(bs
->bio_pool
);
1950 mempool_destroy(bs
->bvec_pool
);
1952 bioset_integrity_free(bs
);
1957 EXPORT_SYMBOL(bioset_free
);
1960 * bioset_create - Create a bio_set
1961 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1962 * @front_pad: Number of bytes to allocate in front of the returned bio
1963 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1964 * and %BIOSET_NEED_RESCUER
1967 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1968 * to ask for a number of bytes to be allocated in front of the bio.
1969 * Front pad allocation is useful for embedding the bio inside
1970 * another structure, to avoid allocating extra data to go with the bio.
1971 * Note that the bio must be embedded at the END of that structure always,
1972 * or things will break badly.
1973 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1974 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast().
1975 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
1976 * dispatch queued requests when the mempool runs out of space.
1979 struct bio_set
*bioset_create(unsigned int pool_size
,
1980 unsigned int front_pad
,
1983 unsigned int back_pad
= BIO_INLINE_VECS
* sizeof(struct bio_vec
);
1986 bs
= kzalloc(sizeof(*bs
), GFP_KERNEL
);
1990 bs
->front_pad
= front_pad
;
1992 spin_lock_init(&bs
->rescue_lock
);
1993 bio_list_init(&bs
->rescue_list
);
1994 INIT_WORK(&bs
->rescue_work
, bio_alloc_rescue
);
1996 bs
->bio_slab
= bio_find_or_create_slab(front_pad
+ back_pad
);
1997 if (!bs
->bio_slab
) {
2002 bs
->bio_pool
= mempool_create_slab_pool(pool_size
, bs
->bio_slab
);
2006 if (flags
& BIOSET_NEED_BVECS
) {
2007 bs
->bvec_pool
= biovec_create_pool(pool_size
);
2012 if (!(flags
& BIOSET_NEED_RESCUER
))
2015 bs
->rescue_workqueue
= alloc_workqueue("bioset", WQ_MEM_RECLAIM
, 0);
2016 if (!bs
->rescue_workqueue
)
2024 EXPORT_SYMBOL(bioset_create
);
2026 #ifdef CONFIG_BLK_CGROUP
2029 * bio_associate_blkcg - associate a bio with the specified blkcg
2031 * @blkcg_css: css of the blkcg to associate
2033 * Associate @bio with the blkcg specified by @blkcg_css. Block layer will
2034 * treat @bio as if it were issued by a task which belongs to the blkcg.
2036 * This function takes an extra reference of @blkcg_css which will be put
2037 * when @bio is released. The caller must own @bio and is responsible for
2038 * synchronizing calls to this function.
2040 int bio_associate_blkcg(struct bio
*bio
, struct cgroup_subsys_state
*blkcg_css
)
2042 if (unlikely(bio
->bi_css
))
2045 bio
->bi_css
= blkcg_css
;
2048 EXPORT_SYMBOL_GPL(bio_associate_blkcg
);
2051 * bio_associate_current - associate a bio with %current
2054 * Associate @bio with %current if it hasn't been associated yet. Block
2055 * layer will treat @bio as if it were issued by %current no matter which
2056 * task actually issues it.
2058 * This function takes an extra reference of @task's io_context and blkcg
2059 * which will be put when @bio is released. The caller must own @bio,
2060 * ensure %current->io_context exists, and is responsible for synchronizing
2061 * calls to this function.
2063 int bio_associate_current(struct bio
*bio
)
2065 struct io_context
*ioc
;
2070 ioc
= current
->io_context
;
2074 get_io_context_active(ioc
);
2076 bio
->bi_css
= task_get_css(current
, io_cgrp_id
);
2079 EXPORT_SYMBOL_GPL(bio_associate_current
);
2082 * bio_disassociate_task - undo bio_associate_current()
2085 void bio_disassociate_task(struct bio
*bio
)
2088 put_io_context(bio
->bi_ioc
);
2092 css_put(bio
->bi_css
);
2098 * bio_clone_blkcg_association - clone blkcg association from src to dst bio
2099 * @dst: destination bio
2102 void bio_clone_blkcg_association(struct bio
*dst
, struct bio
*src
)
2105 WARN_ON(bio_associate_blkcg(dst
, src
->bi_css
));
2107 EXPORT_SYMBOL_GPL(bio_clone_blkcg_association
);
2108 #endif /* CONFIG_BLK_CGROUP */
2110 static void __init
biovec_init_slabs(void)
2114 for (i
= 0; i
< BVEC_POOL_NR
; i
++) {
2116 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
2118 if (bvs
->nr_vecs
<= BIO_INLINE_VECS
) {
2123 size
= bvs
->nr_vecs
* sizeof(struct bio_vec
);
2124 bvs
->slab
= kmem_cache_create(bvs
->name
, size
, 0,
2125 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
2129 static int __init
init_bio(void)
2133 bio_slabs
= kzalloc(bio_slab_max
* sizeof(struct bio_slab
), GFP_KERNEL
);
2135 panic("bio: can't allocate bios\n");
2137 bio_integrity_init();
2138 biovec_init_slabs();
2140 fs_bio_set
= bioset_create(BIO_POOL_SIZE
, 0, BIOSET_NEED_BVECS
);
2142 panic("bio: can't allocate bios\n");
2144 if (bioset_integrity_create(fs_bio_set
, BIO_POOL_SIZE
))
2145 panic("bio: can't create integrity pool\n");
2149 subsys_initcall(init_bio
);