2 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
13 * You should have received a copy of the GNU General Public Licens
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
19 #include <linux/swap.h>
20 #include <linux/bio.h>
21 #include <linux/blkdev.h>
22 #include <linux/uio.h>
23 #include <linux/iocontext.h>
24 #include <linux/slab.h>
25 #include <linux/init.h>
26 #include <linux/kernel.h>
27 #include <linux/export.h>
28 #include <linux/mempool.h>
29 #include <linux/workqueue.h>
30 #include <linux/cgroup.h>
31 #include <scsi/sg.h> /* for struct sg_iovec */
33 #include <trace/events/block.h>
36 * Test patch to inline a certain number of bi_io_vec's inside the bio
37 * itself, to shrink a bio data allocation from two mempool calls to one
39 #define BIO_INLINE_VECS 4
41 static mempool_t
*bio_split_pool __read_mostly
;
44 * if you change this list, also change bvec_alloc or things will
45 * break badly! cannot be bigger than what you can fit into an
48 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
49 static struct biovec_slab bvec_slabs
[BIOVEC_NR_POOLS
] __read_mostly
= {
50 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES
),
55 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
56 * IO code that does not need private memory pools.
58 struct bio_set
*fs_bio_set
;
59 EXPORT_SYMBOL(fs_bio_set
);
62 * Our slab pool management
65 struct kmem_cache
*slab
;
66 unsigned int slab_ref
;
67 unsigned int slab_size
;
70 static DEFINE_MUTEX(bio_slab_lock
);
71 static struct bio_slab
*bio_slabs
;
72 static unsigned int bio_slab_nr
, bio_slab_max
;
74 static struct kmem_cache
*bio_find_or_create_slab(unsigned int extra_size
)
76 unsigned int sz
= sizeof(struct bio
) + extra_size
;
77 struct kmem_cache
*slab
= NULL
;
78 struct bio_slab
*bslab
, *new_bio_slabs
;
79 unsigned int new_bio_slab_max
;
80 unsigned int i
, entry
= -1;
82 mutex_lock(&bio_slab_lock
);
85 while (i
< bio_slab_nr
) {
86 bslab
= &bio_slabs
[i
];
88 if (!bslab
->slab
&& entry
== -1)
90 else if (bslab
->slab_size
== sz
) {
101 if (bio_slab_nr
== bio_slab_max
&& entry
== -1) {
102 new_bio_slab_max
= bio_slab_max
<< 1;
103 new_bio_slabs
= krealloc(bio_slabs
,
104 new_bio_slab_max
* sizeof(struct bio_slab
),
108 bio_slab_max
= new_bio_slab_max
;
109 bio_slabs
= new_bio_slabs
;
112 entry
= bio_slab_nr
++;
114 bslab
= &bio_slabs
[entry
];
116 snprintf(bslab
->name
, sizeof(bslab
->name
), "bio-%d", entry
);
117 slab
= kmem_cache_create(bslab
->name
, sz
, 0, SLAB_HWCACHE_ALIGN
, NULL
);
121 printk(KERN_INFO
"bio: create slab <%s> at %d\n", bslab
->name
, entry
);
124 bslab
->slab_size
= sz
;
126 mutex_unlock(&bio_slab_lock
);
130 static void bio_put_slab(struct bio_set
*bs
)
132 struct bio_slab
*bslab
= NULL
;
135 mutex_lock(&bio_slab_lock
);
137 for (i
= 0; i
< bio_slab_nr
; i
++) {
138 if (bs
->bio_slab
== bio_slabs
[i
].slab
) {
139 bslab
= &bio_slabs
[i
];
144 if (WARN(!bslab
, KERN_ERR
"bio: unable to find slab!\n"))
147 WARN_ON(!bslab
->slab_ref
);
149 if (--bslab
->slab_ref
)
152 kmem_cache_destroy(bslab
->slab
);
156 mutex_unlock(&bio_slab_lock
);
159 unsigned int bvec_nr_vecs(unsigned short idx
)
161 return bvec_slabs
[idx
].nr_vecs
;
164 void bvec_free_bs(struct bio_set
*bs
, struct bio_vec
*bv
, unsigned int idx
)
166 BIO_BUG_ON(idx
>= BIOVEC_NR_POOLS
);
168 if (idx
== BIOVEC_MAX_IDX
)
169 mempool_free(bv
, bs
->bvec_pool
);
171 struct biovec_slab
*bvs
= bvec_slabs
+ idx
;
173 kmem_cache_free(bvs
->slab
, bv
);
177 struct bio_vec
*bvec_alloc_bs(gfp_t gfp_mask
, int nr
, unsigned long *idx
,
183 * see comment near bvec_array define!
201 case 129 ... BIO_MAX_PAGES
:
209 * idx now points to the pool we want to allocate from. only the
210 * 1-vec entry pool is mempool backed.
212 if (*idx
== BIOVEC_MAX_IDX
) {
214 bvl
= mempool_alloc(bs
->bvec_pool
, gfp_mask
);
216 struct biovec_slab
*bvs
= bvec_slabs
+ *idx
;
217 gfp_t __gfp_mask
= gfp_mask
& ~(__GFP_WAIT
| __GFP_IO
);
220 * Make this allocation restricted and don't dump info on
221 * allocation failures, since we'll fallback to the mempool
222 * in case of failure.
224 __gfp_mask
|= __GFP_NOMEMALLOC
| __GFP_NORETRY
| __GFP_NOWARN
;
227 * Try a slab allocation. If this fails and __GFP_WAIT
228 * is set, retry with the 1-entry mempool
230 bvl
= kmem_cache_alloc(bvs
->slab
, __gfp_mask
);
231 if (unlikely(!bvl
&& (gfp_mask
& __GFP_WAIT
))) {
232 *idx
= BIOVEC_MAX_IDX
;
240 static void __bio_free(struct bio
*bio
)
242 bio_disassociate_task(bio
);
244 if (bio_integrity(bio
))
245 bio_integrity_free(bio
);
248 static void bio_free(struct bio
*bio
)
250 struct bio_set
*bs
= bio
->bi_pool
;
256 if (bio_has_allocated_vec(bio
))
257 bvec_free_bs(bs
, bio
->bi_io_vec
, BIO_POOL_IDX(bio
));
260 * If we have front padding, adjust the bio pointer before freeing
265 mempool_free(p
, bs
->bio_pool
);
267 /* Bio was allocated by bio_kmalloc() */
272 void bio_init(struct bio
*bio
)
274 memset(bio
, 0, sizeof(*bio
));
275 bio
->bi_flags
= 1 << BIO_UPTODATE
;
276 atomic_set(&bio
->bi_cnt
, 1);
278 EXPORT_SYMBOL(bio_init
);
281 * bio_reset - reinitialize a bio
285 * After calling bio_reset(), @bio will be in the same state as a freshly
286 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
287 * preserved are the ones that are initialized by bio_alloc_bioset(). See
288 * comment in struct bio.
290 void bio_reset(struct bio
*bio
)
292 unsigned long flags
= bio
->bi_flags
& (~0UL << BIO_RESET_BITS
);
296 memset(bio
, 0, BIO_RESET_BYTES
);
297 bio
->bi_flags
= flags
|(1 << BIO_UPTODATE
);
299 EXPORT_SYMBOL(bio_reset
);
302 * bio_alloc_bioset - allocate a bio for I/O
303 * @gfp_mask: the GFP_ mask given to the slab allocator
304 * @nr_iovecs: number of iovecs to pre-allocate
305 * @bs: the bio_set to allocate from.
308 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
309 * backed by the @bs's mempool.
311 * When @bs is not NULL, if %__GFP_WAIT is set then bio_alloc will always be
312 * able to allocate a bio. This is due to the mempool guarantees. To make this
313 * work, callers must never allocate more than 1 bio at a time from this pool.
314 * Callers that need to allocate more than 1 bio must always submit the
315 * previously allocated bio for IO before attempting to allocate a new one.
316 * Failure to do so can cause deadlocks under memory pressure.
319 * Pointer to new bio on success, NULL on failure.
321 struct bio
*bio_alloc_bioset(gfp_t gfp_mask
, int nr_iovecs
, struct bio_set
*bs
)
324 unsigned inline_vecs
;
325 unsigned long idx
= BIO_POOL_NONE
;
326 struct bio_vec
*bvl
= NULL
;
331 if (nr_iovecs
> UIO_MAXIOV
)
334 p
= kmalloc(sizeof(struct bio
) +
335 nr_iovecs
* sizeof(struct bio_vec
),
338 inline_vecs
= nr_iovecs
;
340 p
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
341 front_pad
= bs
->front_pad
;
342 inline_vecs
= BIO_INLINE_VECS
;
351 if (nr_iovecs
> inline_vecs
) {
352 bvl
= bvec_alloc_bs(gfp_mask
, nr_iovecs
, &idx
, bs
);
355 } else if (nr_iovecs
) {
356 bvl
= bio
->bi_inline_vecs
;
360 bio
->bi_flags
|= idx
<< BIO_POOL_OFFSET
;
361 bio
->bi_max_vecs
= nr_iovecs
;
362 bio
->bi_io_vec
= bvl
;
366 mempool_free(p
, bs
->bio_pool
);
369 EXPORT_SYMBOL(bio_alloc_bioset
);
371 void zero_fill_bio(struct bio
*bio
)
377 bio_for_each_segment(bv
, bio
, i
) {
378 char *data
= bvec_kmap_irq(bv
, &flags
);
379 memset(data
, 0, bv
->bv_len
);
380 flush_dcache_page(bv
->bv_page
);
381 bvec_kunmap_irq(data
, &flags
);
384 EXPORT_SYMBOL(zero_fill_bio
);
387 * bio_put - release a reference to a bio
388 * @bio: bio to release reference to
391 * Put a reference to a &struct bio, either one you have gotten with
392 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
394 void bio_put(struct bio
*bio
)
396 BIO_BUG_ON(!atomic_read(&bio
->bi_cnt
));
401 if (atomic_dec_and_test(&bio
->bi_cnt
))
404 EXPORT_SYMBOL(bio_put
);
406 inline int bio_phys_segments(struct request_queue
*q
, struct bio
*bio
)
408 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
409 blk_recount_segments(q
, bio
);
411 return bio
->bi_phys_segments
;
413 EXPORT_SYMBOL(bio_phys_segments
);
416 * __bio_clone - clone a bio
417 * @bio: destination bio
418 * @bio_src: bio to clone
420 * Clone a &bio. Caller will own the returned bio, but not
421 * the actual data it points to. Reference count of returned
424 void __bio_clone(struct bio
*bio
, struct bio
*bio_src
)
426 memcpy(bio
->bi_io_vec
, bio_src
->bi_io_vec
,
427 bio_src
->bi_max_vecs
* sizeof(struct bio_vec
));
430 * most users will be overriding ->bi_bdev with a new target,
431 * so we don't set nor calculate new physical/hw segment counts here
433 bio
->bi_sector
= bio_src
->bi_sector
;
434 bio
->bi_bdev
= bio_src
->bi_bdev
;
435 bio
->bi_flags
|= 1 << BIO_CLONED
;
436 bio
->bi_rw
= bio_src
->bi_rw
;
437 bio
->bi_vcnt
= bio_src
->bi_vcnt
;
438 bio
->bi_size
= bio_src
->bi_size
;
439 bio
->bi_idx
= bio_src
->bi_idx
;
441 EXPORT_SYMBOL(__bio_clone
);
444 * bio_clone_bioset - clone a bio
446 * @gfp_mask: allocation priority
447 * @bs: bio_set to allocate from
449 * Like __bio_clone, only also allocates the returned bio
451 struct bio
*bio_clone_bioset(struct bio
*bio
, gfp_t gfp_mask
,
456 b
= bio_alloc_bioset(gfp_mask
, bio
->bi_max_vecs
, bs
);
462 if (bio_integrity(bio
)) {
465 ret
= bio_integrity_clone(b
, bio
, gfp_mask
);
475 EXPORT_SYMBOL(bio_clone_bioset
);
478 * bio_get_nr_vecs - return approx number of vecs
481 * Return the approximate number of pages we can send to this target.
482 * There's no guarantee that you will be able to fit this number of pages
483 * into a bio, it does not account for dynamic restrictions that vary
486 int bio_get_nr_vecs(struct block_device
*bdev
)
488 struct request_queue
*q
= bdev_get_queue(bdev
);
491 nr_pages
= min_t(unsigned,
492 queue_max_segments(q
),
493 queue_max_sectors(q
) / (PAGE_SIZE
>> 9) + 1);
495 return min_t(unsigned, nr_pages
, BIO_MAX_PAGES
);
498 EXPORT_SYMBOL(bio_get_nr_vecs
);
500 static int __bio_add_page(struct request_queue
*q
, struct bio
*bio
, struct page
501 *page
, unsigned int len
, unsigned int offset
,
502 unsigned short max_sectors
)
504 int retried_segments
= 0;
505 struct bio_vec
*bvec
;
508 * cloned bio must not modify vec list
510 if (unlikely(bio_flagged(bio
, BIO_CLONED
)))
513 if (((bio
->bi_size
+ len
) >> 9) > max_sectors
)
517 * For filesystems with a blocksize smaller than the pagesize
518 * we will often be called with the same page as last time and
519 * a consecutive offset. Optimize this special case.
521 if (bio
->bi_vcnt
> 0) {
522 struct bio_vec
*prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
524 if (page
== prev
->bv_page
&&
525 offset
== prev
->bv_offset
+ prev
->bv_len
) {
526 unsigned int prev_bv_len
= prev
->bv_len
;
529 if (q
->merge_bvec_fn
) {
530 struct bvec_merge_data bvm
= {
531 /* prev_bvec is already charged in
532 bi_size, discharge it in order to
533 simulate merging updated prev_bvec
535 .bi_bdev
= bio
->bi_bdev
,
536 .bi_sector
= bio
->bi_sector
,
537 .bi_size
= bio
->bi_size
- prev_bv_len
,
541 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < prev
->bv_len
) {
551 if (bio
->bi_vcnt
>= bio
->bi_max_vecs
)
555 * we might lose a segment or two here, but rather that than
556 * make this too complex.
559 while (bio
->bi_phys_segments
>= queue_max_segments(q
)) {
561 if (retried_segments
)
564 retried_segments
= 1;
565 blk_recount_segments(q
, bio
);
569 * setup the new entry, we might clear it again later if we
570 * cannot add the page
572 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
573 bvec
->bv_page
= page
;
575 bvec
->bv_offset
= offset
;
578 * if queue has other restrictions (eg varying max sector size
579 * depending on offset), it can specify a merge_bvec_fn in the
580 * queue to get further control
582 if (q
->merge_bvec_fn
) {
583 struct bvec_merge_data bvm
= {
584 .bi_bdev
= bio
->bi_bdev
,
585 .bi_sector
= bio
->bi_sector
,
586 .bi_size
= bio
->bi_size
,
591 * merge_bvec_fn() returns number of bytes it can accept
594 if (q
->merge_bvec_fn(q
, &bvm
, bvec
) < bvec
->bv_len
) {
595 bvec
->bv_page
= NULL
;
602 /* If we may be able to merge these biovecs, force a recount */
603 if (bio
->bi_vcnt
&& (BIOVEC_PHYS_MERGEABLE(bvec
-1, bvec
)))
604 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
607 bio
->bi_phys_segments
++;
614 * bio_add_pc_page - attempt to add page to bio
615 * @q: the target queue
616 * @bio: destination bio
618 * @len: vec entry length
619 * @offset: vec entry offset
621 * Attempt to add a page to the bio_vec maplist. This can fail for a
622 * number of reasons, such as the bio being full or target block device
623 * limitations. The target block device must allow bio's up to PAGE_SIZE,
624 * so it is always possible to add a single page to an empty bio.
626 * This should only be used by REQ_PC bios.
628 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
, struct page
*page
,
629 unsigned int len
, unsigned int offset
)
631 return __bio_add_page(q
, bio
, page
, len
, offset
,
632 queue_max_hw_sectors(q
));
634 EXPORT_SYMBOL(bio_add_pc_page
);
637 * bio_add_page - attempt to add page to bio
638 * @bio: destination bio
640 * @len: vec entry length
641 * @offset: vec entry offset
643 * Attempt to add a page to the bio_vec maplist. This can fail for a
644 * number of reasons, such as the bio being full or target block device
645 * limitations. The target block device must allow bio's up to PAGE_SIZE,
646 * so it is always possible to add a single page to an empty bio.
648 int bio_add_page(struct bio
*bio
, struct page
*page
, unsigned int len
,
651 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
652 return __bio_add_page(q
, bio
, page
, len
, offset
, queue_max_sectors(q
));
654 EXPORT_SYMBOL(bio_add_page
);
656 struct bio_map_data
{
657 struct bio_vec
*iovecs
;
658 struct sg_iovec
*sgvecs
;
663 static void bio_set_map_data(struct bio_map_data
*bmd
, struct bio
*bio
,
664 struct sg_iovec
*iov
, int iov_count
,
667 memcpy(bmd
->iovecs
, bio
->bi_io_vec
, sizeof(struct bio_vec
) * bio
->bi_vcnt
);
668 memcpy(bmd
->sgvecs
, iov
, sizeof(struct sg_iovec
) * iov_count
);
669 bmd
->nr_sgvecs
= iov_count
;
670 bmd
->is_our_pages
= is_our_pages
;
671 bio
->bi_private
= bmd
;
674 static void bio_free_map_data(struct bio_map_data
*bmd
)
681 static struct bio_map_data
*bio_alloc_map_data(int nr_segs
,
682 unsigned int iov_count
,
685 struct bio_map_data
*bmd
;
687 if (iov_count
> UIO_MAXIOV
)
690 bmd
= kmalloc(sizeof(*bmd
), gfp_mask
);
694 bmd
->iovecs
= kmalloc(sizeof(struct bio_vec
) * nr_segs
, gfp_mask
);
700 bmd
->sgvecs
= kmalloc(sizeof(struct sg_iovec
) * iov_count
, gfp_mask
);
709 static int __bio_copy_iov(struct bio
*bio
, struct bio_vec
*iovecs
,
710 struct sg_iovec
*iov
, int iov_count
,
711 int to_user
, int from_user
, int do_free_page
)
714 struct bio_vec
*bvec
;
716 unsigned int iov_off
= 0;
718 __bio_for_each_segment(bvec
, bio
, i
, 0) {
719 char *bv_addr
= page_address(bvec
->bv_page
);
720 unsigned int bv_len
= iovecs
[i
].bv_len
;
722 while (bv_len
&& iov_idx
< iov_count
) {
724 char __user
*iov_addr
;
726 bytes
= min_t(unsigned int,
727 iov
[iov_idx
].iov_len
- iov_off
, bv_len
);
728 iov_addr
= iov
[iov_idx
].iov_base
+ iov_off
;
732 ret
= copy_to_user(iov_addr
, bv_addr
,
736 ret
= copy_from_user(bv_addr
, iov_addr
,
748 if (iov
[iov_idx
].iov_len
== iov_off
) {
755 __free_page(bvec
->bv_page
);
762 * bio_uncopy_user - finish previously mapped bio
763 * @bio: bio being terminated
765 * Free pages allocated from bio_copy_user() and write back data
766 * to user space in case of a read.
768 int bio_uncopy_user(struct bio
*bio
)
770 struct bio_map_data
*bmd
= bio
->bi_private
;
773 if (!bio_flagged(bio
, BIO_NULL_MAPPED
))
774 ret
= __bio_copy_iov(bio
, bmd
->iovecs
, bmd
->sgvecs
,
775 bmd
->nr_sgvecs
, bio_data_dir(bio
) == READ
,
776 0, bmd
->is_our_pages
);
777 bio_free_map_data(bmd
);
781 EXPORT_SYMBOL(bio_uncopy_user
);
784 * bio_copy_user_iov - copy user data to bio
785 * @q: destination block queue
786 * @map_data: pointer to the rq_map_data holding pages (if necessary)
788 * @iov_count: number of elements in the iovec
789 * @write_to_vm: bool indicating writing to pages or not
790 * @gfp_mask: memory allocation flags
792 * Prepares and returns a bio for indirect user io, bouncing data
793 * to/from kernel pages as necessary. Must be paired with
794 * call bio_uncopy_user() on io completion.
796 struct bio
*bio_copy_user_iov(struct request_queue
*q
,
797 struct rq_map_data
*map_data
,
798 struct sg_iovec
*iov
, int iov_count
,
799 int write_to_vm
, gfp_t gfp_mask
)
801 struct bio_map_data
*bmd
;
802 struct bio_vec
*bvec
;
807 unsigned int len
= 0;
808 unsigned int offset
= map_data
? map_data
->offset
& ~PAGE_MASK
: 0;
810 for (i
= 0; i
< iov_count
; i
++) {
815 uaddr
= (unsigned long)iov
[i
].iov_base
;
816 end
= (uaddr
+ iov
[i
].iov_len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
817 start
= uaddr
>> PAGE_SHIFT
;
823 return ERR_PTR(-EINVAL
);
825 nr_pages
+= end
- start
;
826 len
+= iov
[i
].iov_len
;
832 bmd
= bio_alloc_map_data(nr_pages
, iov_count
, gfp_mask
);
834 return ERR_PTR(-ENOMEM
);
837 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
842 bio
->bi_rw
|= REQ_WRITE
;
847 nr_pages
= 1 << map_data
->page_order
;
848 i
= map_data
->offset
/ PAGE_SIZE
;
851 unsigned int bytes
= PAGE_SIZE
;
859 if (i
== map_data
->nr_entries
* nr_pages
) {
864 page
= map_data
->pages
[i
/ nr_pages
];
865 page
+= (i
% nr_pages
);
869 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
876 if (bio_add_pc_page(q
, bio
, page
, bytes
, offset
) < bytes
)
889 if ((!write_to_vm
&& (!map_data
|| !map_data
->null_mapped
)) ||
890 (map_data
&& map_data
->from_user
)) {
891 ret
= __bio_copy_iov(bio
, bio
->bi_io_vec
, iov
, iov_count
, 0, 1, 0);
896 bio_set_map_data(bmd
, bio
, iov
, iov_count
, map_data
? 0 : 1);
900 bio_for_each_segment(bvec
, bio
, i
)
901 __free_page(bvec
->bv_page
);
905 bio_free_map_data(bmd
);
910 * bio_copy_user - copy user data to bio
911 * @q: destination block queue
912 * @map_data: pointer to the rq_map_data holding pages (if necessary)
913 * @uaddr: start of user address
914 * @len: length in bytes
915 * @write_to_vm: bool indicating writing to pages or not
916 * @gfp_mask: memory allocation flags
918 * Prepares and returns a bio for indirect user io, bouncing data
919 * to/from kernel pages as necessary. Must be paired with
920 * call bio_uncopy_user() on io completion.
922 struct bio
*bio_copy_user(struct request_queue
*q
, struct rq_map_data
*map_data
,
923 unsigned long uaddr
, unsigned int len
,
924 int write_to_vm
, gfp_t gfp_mask
)
928 iov
.iov_base
= (void __user
*)uaddr
;
931 return bio_copy_user_iov(q
, map_data
, &iov
, 1, write_to_vm
, gfp_mask
);
933 EXPORT_SYMBOL(bio_copy_user
);
935 static struct bio
*__bio_map_user_iov(struct request_queue
*q
,
936 struct block_device
*bdev
,
937 struct sg_iovec
*iov
, int iov_count
,
938 int write_to_vm
, gfp_t gfp_mask
)
947 for (i
= 0; i
< iov_count
; i
++) {
948 unsigned long uaddr
= (unsigned long)iov
[i
].iov_base
;
949 unsigned long len
= iov
[i
].iov_len
;
950 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
951 unsigned long start
= uaddr
>> PAGE_SHIFT
;
957 return ERR_PTR(-EINVAL
);
959 nr_pages
+= end
- start
;
961 * buffer must be aligned to at least hardsector size for now
963 if (uaddr
& queue_dma_alignment(q
))
964 return ERR_PTR(-EINVAL
);
968 return ERR_PTR(-EINVAL
);
970 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
972 return ERR_PTR(-ENOMEM
);
975 pages
= kcalloc(nr_pages
, sizeof(struct page
*), gfp_mask
);
979 for (i
= 0; i
< iov_count
; i
++) {
980 unsigned long uaddr
= (unsigned long)iov
[i
].iov_base
;
981 unsigned long len
= iov
[i
].iov_len
;
982 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
983 unsigned long start
= uaddr
>> PAGE_SHIFT
;
984 const int local_nr_pages
= end
- start
;
985 const int page_limit
= cur_page
+ local_nr_pages
;
987 ret
= get_user_pages_fast(uaddr
, local_nr_pages
,
988 write_to_vm
, &pages
[cur_page
]);
989 if (ret
< local_nr_pages
) {
994 offset
= uaddr
& ~PAGE_MASK
;
995 for (j
= cur_page
; j
< page_limit
; j
++) {
996 unsigned int bytes
= PAGE_SIZE
- offset
;
1007 if (bio_add_pc_page(q
, bio
, pages
[j
], bytes
, offset
) <
1017 * release the pages we didn't map into the bio, if any
1019 while (j
< page_limit
)
1020 page_cache_release(pages
[j
++]);
1026 * set data direction, and check if mapped pages need bouncing
1029 bio
->bi_rw
|= REQ_WRITE
;
1031 bio
->bi_bdev
= bdev
;
1032 bio
->bi_flags
|= (1 << BIO_USER_MAPPED
);
1036 for (i
= 0; i
< nr_pages
; i
++) {
1039 page_cache_release(pages
[i
]);
1044 return ERR_PTR(ret
);
1048 * bio_map_user - map user address into bio
1049 * @q: the struct request_queue for the bio
1050 * @bdev: destination block device
1051 * @uaddr: start of user address
1052 * @len: length in bytes
1053 * @write_to_vm: bool indicating writing to pages or not
1054 * @gfp_mask: memory allocation flags
1056 * Map the user space address into a bio suitable for io to a block
1057 * device. Returns an error pointer in case of error.
1059 struct bio
*bio_map_user(struct request_queue
*q
, struct block_device
*bdev
,
1060 unsigned long uaddr
, unsigned int len
, int write_to_vm
,
1063 struct sg_iovec iov
;
1065 iov
.iov_base
= (void __user
*)uaddr
;
1068 return bio_map_user_iov(q
, bdev
, &iov
, 1, write_to_vm
, gfp_mask
);
1070 EXPORT_SYMBOL(bio_map_user
);
1073 * bio_map_user_iov - map user sg_iovec table into bio
1074 * @q: the struct request_queue for the bio
1075 * @bdev: destination block device
1077 * @iov_count: number of elements in the iovec
1078 * @write_to_vm: bool indicating writing to pages or not
1079 * @gfp_mask: memory allocation flags
1081 * Map the user space address into a bio suitable for io to a block
1082 * device. Returns an error pointer in case of error.
1084 struct bio
*bio_map_user_iov(struct request_queue
*q
, struct block_device
*bdev
,
1085 struct sg_iovec
*iov
, int iov_count
,
1086 int write_to_vm
, gfp_t gfp_mask
)
1090 bio
= __bio_map_user_iov(q
, bdev
, iov
, iov_count
, write_to_vm
,
1096 * subtle -- if __bio_map_user() ended up bouncing a bio,
1097 * it would normally disappear when its bi_end_io is run.
1098 * however, we need it for the unmap, so grab an extra
1106 static void __bio_unmap_user(struct bio
*bio
)
1108 struct bio_vec
*bvec
;
1112 * make sure we dirty pages we wrote to
1114 __bio_for_each_segment(bvec
, bio
, i
, 0) {
1115 if (bio_data_dir(bio
) == READ
)
1116 set_page_dirty_lock(bvec
->bv_page
);
1118 page_cache_release(bvec
->bv_page
);
1125 * bio_unmap_user - unmap a bio
1126 * @bio: the bio being unmapped
1128 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1129 * a process context.
1131 * bio_unmap_user() may sleep.
1133 void bio_unmap_user(struct bio
*bio
)
1135 __bio_unmap_user(bio
);
1138 EXPORT_SYMBOL(bio_unmap_user
);
1140 static void bio_map_kern_endio(struct bio
*bio
, int err
)
1145 static struct bio
*__bio_map_kern(struct request_queue
*q
, void *data
,
1146 unsigned int len
, gfp_t gfp_mask
)
1148 unsigned long kaddr
= (unsigned long)data
;
1149 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1150 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1151 const int nr_pages
= end
- start
;
1155 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1157 return ERR_PTR(-ENOMEM
);
1159 offset
= offset_in_page(kaddr
);
1160 for (i
= 0; i
< nr_pages
; i
++) {
1161 unsigned int bytes
= PAGE_SIZE
- offset
;
1169 if (bio_add_pc_page(q
, bio
, virt_to_page(data
), bytes
,
1178 bio
->bi_end_io
= bio_map_kern_endio
;
1183 * bio_map_kern - map kernel address into bio
1184 * @q: the struct request_queue for the bio
1185 * @data: pointer to buffer to map
1186 * @len: length in bytes
1187 * @gfp_mask: allocation flags for bio allocation
1189 * Map the kernel address into a bio suitable for io to a block
1190 * device. Returns an error pointer in case of error.
1192 struct bio
*bio_map_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1197 bio
= __bio_map_kern(q
, data
, len
, gfp_mask
);
1201 if (bio
->bi_size
== len
)
1205 * Don't support partial mappings.
1208 return ERR_PTR(-EINVAL
);
1210 EXPORT_SYMBOL(bio_map_kern
);
1212 static void bio_copy_kern_endio(struct bio
*bio
, int err
)
1214 struct bio_vec
*bvec
;
1215 const int read
= bio_data_dir(bio
) == READ
;
1216 struct bio_map_data
*bmd
= bio
->bi_private
;
1218 char *p
= bmd
->sgvecs
[0].iov_base
;
1220 __bio_for_each_segment(bvec
, bio
, i
, 0) {
1221 char *addr
= page_address(bvec
->bv_page
);
1222 int len
= bmd
->iovecs
[i
].bv_len
;
1225 memcpy(p
, addr
, len
);
1227 __free_page(bvec
->bv_page
);
1231 bio_free_map_data(bmd
);
1236 * bio_copy_kern - copy kernel address into bio
1237 * @q: the struct request_queue for the bio
1238 * @data: pointer to buffer to copy
1239 * @len: length in bytes
1240 * @gfp_mask: allocation flags for bio and page allocation
1241 * @reading: data direction is READ
1243 * copy the kernel address into a bio suitable for io to a block
1244 * device. Returns an error pointer in case of error.
1246 struct bio
*bio_copy_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1247 gfp_t gfp_mask
, int reading
)
1250 struct bio_vec
*bvec
;
1253 bio
= bio_copy_user(q
, NULL
, (unsigned long)data
, len
, 1, gfp_mask
);
1260 bio_for_each_segment(bvec
, bio
, i
) {
1261 char *addr
= page_address(bvec
->bv_page
);
1263 memcpy(addr
, p
, bvec
->bv_len
);
1268 bio
->bi_end_io
= bio_copy_kern_endio
;
1272 EXPORT_SYMBOL(bio_copy_kern
);
1275 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1276 * for performing direct-IO in BIOs.
1278 * The problem is that we cannot run set_page_dirty() from interrupt context
1279 * because the required locks are not interrupt-safe. So what we can do is to
1280 * mark the pages dirty _before_ performing IO. And in interrupt context,
1281 * check that the pages are still dirty. If so, fine. If not, redirty them
1282 * in process context.
1284 * We special-case compound pages here: normally this means reads into hugetlb
1285 * pages. The logic in here doesn't really work right for compound pages
1286 * because the VM does not uniformly chase down the head page in all cases.
1287 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1288 * handle them at all. So we skip compound pages here at an early stage.
1290 * Note that this code is very hard to test under normal circumstances because
1291 * direct-io pins the pages with get_user_pages(). This makes
1292 * is_page_cache_freeable return false, and the VM will not clean the pages.
1293 * But other code (eg, flusher threads) could clean the pages if they are mapped
1296 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1297 * deferred bio dirtying paths.
1301 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1303 void bio_set_pages_dirty(struct bio
*bio
)
1305 struct bio_vec
*bvec
= bio
->bi_io_vec
;
1308 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
1309 struct page
*page
= bvec
[i
].bv_page
;
1311 if (page
&& !PageCompound(page
))
1312 set_page_dirty_lock(page
);
1316 static void bio_release_pages(struct bio
*bio
)
1318 struct bio_vec
*bvec
= bio
->bi_io_vec
;
1321 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
1322 struct page
*page
= bvec
[i
].bv_page
;
1330 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1331 * If they are, then fine. If, however, some pages are clean then they must
1332 * have been written out during the direct-IO read. So we take another ref on
1333 * the BIO and the offending pages and re-dirty the pages in process context.
1335 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1336 * here on. It will run one page_cache_release() against each page and will
1337 * run one bio_put() against the BIO.
1340 static void bio_dirty_fn(struct work_struct
*work
);
1342 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
1343 static DEFINE_SPINLOCK(bio_dirty_lock
);
1344 static struct bio
*bio_dirty_list
;
1347 * This runs in process context
1349 static void bio_dirty_fn(struct work_struct
*work
)
1351 unsigned long flags
;
1354 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1355 bio
= bio_dirty_list
;
1356 bio_dirty_list
= NULL
;
1357 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1360 struct bio
*next
= bio
->bi_private
;
1362 bio_set_pages_dirty(bio
);
1363 bio_release_pages(bio
);
1369 void bio_check_pages_dirty(struct bio
*bio
)
1371 struct bio_vec
*bvec
= bio
->bi_io_vec
;
1372 int nr_clean_pages
= 0;
1375 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
1376 struct page
*page
= bvec
[i
].bv_page
;
1378 if (PageDirty(page
) || PageCompound(page
)) {
1379 page_cache_release(page
);
1380 bvec
[i
].bv_page
= NULL
;
1386 if (nr_clean_pages
) {
1387 unsigned long flags
;
1389 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1390 bio
->bi_private
= bio_dirty_list
;
1391 bio_dirty_list
= bio
;
1392 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1393 schedule_work(&bio_dirty_work
);
1399 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1400 void bio_flush_dcache_pages(struct bio
*bi
)
1403 struct bio_vec
*bvec
;
1405 bio_for_each_segment(bvec
, bi
, i
)
1406 flush_dcache_page(bvec
->bv_page
);
1408 EXPORT_SYMBOL(bio_flush_dcache_pages
);
1412 * bio_endio - end I/O on a bio
1414 * @error: error, if any
1417 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1418 * preferred way to end I/O on a bio, it takes care of clearing
1419 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1420 * established -Exxxx (-EIO, for instance) error values in case
1421 * something went wrong. No one should call bi_end_io() directly on a
1422 * bio unless they own it and thus know that it has an end_io
1425 void bio_endio(struct bio
*bio
, int error
)
1428 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1429 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
1433 bio
->bi_end_io(bio
, error
);
1435 EXPORT_SYMBOL(bio_endio
);
1437 void bio_pair_release(struct bio_pair
*bp
)
1439 if (atomic_dec_and_test(&bp
->cnt
)) {
1440 struct bio
*master
= bp
->bio1
.bi_private
;
1442 bio_endio(master
, bp
->error
);
1443 mempool_free(bp
, bp
->bio2
.bi_private
);
1446 EXPORT_SYMBOL(bio_pair_release
);
1448 static void bio_pair_end_1(struct bio
*bi
, int err
)
1450 struct bio_pair
*bp
= container_of(bi
, struct bio_pair
, bio1
);
1455 bio_pair_release(bp
);
1458 static void bio_pair_end_2(struct bio
*bi
, int err
)
1460 struct bio_pair
*bp
= container_of(bi
, struct bio_pair
, bio2
);
1465 bio_pair_release(bp
);
1469 * split a bio - only worry about a bio with a single page in its iovec
1471 struct bio_pair
*bio_split(struct bio
*bi
, int first_sectors
)
1473 struct bio_pair
*bp
= mempool_alloc(bio_split_pool
, GFP_NOIO
);
1478 trace_block_split(bdev_get_queue(bi
->bi_bdev
), bi
,
1479 bi
->bi_sector
+ first_sectors
);
1481 BUG_ON(bi
->bi_vcnt
!= 1 && bi
->bi_vcnt
!= 0);
1482 BUG_ON(bi
->bi_idx
!= 0);
1483 atomic_set(&bp
->cnt
, 3);
1487 bp
->bio2
.bi_sector
+= first_sectors
;
1488 bp
->bio2
.bi_size
-= first_sectors
<< 9;
1489 bp
->bio1
.bi_size
= first_sectors
<< 9;
1491 if (bi
->bi_vcnt
!= 0) {
1492 bp
->bv1
= bi
->bi_io_vec
[0];
1493 bp
->bv2
= bi
->bi_io_vec
[0];
1495 if (bio_is_rw(bi
)) {
1496 bp
->bv2
.bv_offset
+= first_sectors
<< 9;
1497 bp
->bv2
.bv_len
-= first_sectors
<< 9;
1498 bp
->bv1
.bv_len
= first_sectors
<< 9;
1501 bp
->bio1
.bi_io_vec
= &bp
->bv1
;
1502 bp
->bio2
.bi_io_vec
= &bp
->bv2
;
1504 bp
->bio1
.bi_max_vecs
= 1;
1505 bp
->bio2
.bi_max_vecs
= 1;
1508 bp
->bio1
.bi_end_io
= bio_pair_end_1
;
1509 bp
->bio2
.bi_end_io
= bio_pair_end_2
;
1511 bp
->bio1
.bi_private
= bi
;
1512 bp
->bio2
.bi_private
= bio_split_pool
;
1514 if (bio_integrity(bi
))
1515 bio_integrity_split(bi
, bp
, first_sectors
);
1519 EXPORT_SYMBOL(bio_split
);
1522 * bio_sector_offset - Find hardware sector offset in bio
1523 * @bio: bio to inspect
1524 * @index: bio_vec index
1525 * @offset: offset in bv_page
1527 * Return the number of hardware sectors between beginning of bio
1528 * and an end point indicated by a bio_vec index and an offset
1529 * within that vector's page.
1531 sector_t
bio_sector_offset(struct bio
*bio
, unsigned short index
,
1532 unsigned int offset
)
1534 unsigned int sector_sz
;
1539 sector_sz
= queue_logical_block_size(bio
->bi_bdev
->bd_disk
->queue
);
1542 if (index
>= bio
->bi_idx
)
1543 index
= bio
->bi_vcnt
- 1;
1545 __bio_for_each_segment(bv
, bio
, i
, 0) {
1547 if (offset
> bv
->bv_offset
)
1548 sectors
+= (offset
- bv
->bv_offset
) / sector_sz
;
1552 sectors
+= bv
->bv_len
/ sector_sz
;
1557 EXPORT_SYMBOL(bio_sector_offset
);
1560 * create memory pools for biovec's in a bio_set.
1561 * use the global biovec slabs created for general use.
1563 static int biovec_create_pools(struct bio_set
*bs
, int pool_entries
)
1565 struct biovec_slab
*bp
= bvec_slabs
+ BIOVEC_MAX_IDX
;
1567 bs
->bvec_pool
= mempool_create_slab_pool(pool_entries
, bp
->slab
);
1574 static void biovec_free_pools(struct bio_set
*bs
)
1576 mempool_destroy(bs
->bvec_pool
);
1579 void bioset_free(struct bio_set
*bs
)
1582 mempool_destroy(bs
->bio_pool
);
1584 bioset_integrity_free(bs
);
1585 biovec_free_pools(bs
);
1590 EXPORT_SYMBOL(bioset_free
);
1593 * bioset_create - Create a bio_set
1594 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1595 * @front_pad: Number of bytes to allocate in front of the returned bio
1598 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1599 * to ask for a number of bytes to be allocated in front of the bio.
1600 * Front pad allocation is useful for embedding the bio inside
1601 * another structure, to avoid allocating extra data to go with the bio.
1602 * Note that the bio must be embedded at the END of that structure always,
1603 * or things will break badly.
1605 struct bio_set
*bioset_create(unsigned int pool_size
, unsigned int front_pad
)
1607 unsigned int back_pad
= BIO_INLINE_VECS
* sizeof(struct bio_vec
);
1610 bs
= kzalloc(sizeof(*bs
), GFP_KERNEL
);
1614 bs
->front_pad
= front_pad
;
1616 bs
->bio_slab
= bio_find_or_create_slab(front_pad
+ back_pad
);
1617 if (!bs
->bio_slab
) {
1622 bs
->bio_pool
= mempool_create_slab_pool(pool_size
, bs
->bio_slab
);
1626 if (!biovec_create_pools(bs
, pool_size
))
1633 EXPORT_SYMBOL(bioset_create
);
1635 #ifdef CONFIG_BLK_CGROUP
1637 * bio_associate_current - associate a bio with %current
1640 * Associate @bio with %current if it hasn't been associated yet. Block
1641 * layer will treat @bio as if it were issued by %current no matter which
1642 * task actually issues it.
1644 * This function takes an extra reference of @task's io_context and blkcg
1645 * which will be put when @bio is released. The caller must own @bio,
1646 * ensure %current->io_context exists, and is responsible for synchronizing
1647 * calls to this function.
1649 int bio_associate_current(struct bio
*bio
)
1651 struct io_context
*ioc
;
1652 struct cgroup_subsys_state
*css
;
1657 ioc
= current
->io_context
;
1661 /* acquire active ref on @ioc and associate */
1662 get_io_context_active(ioc
);
1665 /* associate blkcg if exists */
1667 css
= task_subsys_state(current
, blkio_subsys_id
);
1668 if (css
&& css_tryget(css
))
1676 * bio_disassociate_task - undo bio_associate_current()
1679 void bio_disassociate_task(struct bio
*bio
)
1682 put_io_context(bio
->bi_ioc
);
1686 css_put(bio
->bi_css
);
1691 #endif /* CONFIG_BLK_CGROUP */
1693 static void __init
biovec_init_slabs(void)
1697 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
1699 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
1701 if (bvs
->nr_vecs
<= BIO_INLINE_VECS
) {
1706 size
= bvs
->nr_vecs
* sizeof(struct bio_vec
);
1707 bvs
->slab
= kmem_cache_create(bvs
->name
, size
, 0,
1708 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
1712 static int __init
init_bio(void)
1716 bio_slabs
= kzalloc(bio_slab_max
* sizeof(struct bio_slab
), GFP_KERNEL
);
1718 panic("bio: can't allocate bios\n");
1720 bio_integrity_init();
1721 biovec_init_slabs();
1723 fs_bio_set
= bioset_create(BIO_POOL_SIZE
, 0);
1725 panic("bio: can't allocate bios\n");
1727 if (bioset_integrity_create(fs_bio_set
, BIO_POOL_SIZE
))
1728 panic("bio: can't create integrity pool\n");
1730 bio_split_pool
= mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES
,
1731 sizeof(struct bio_pair
));
1732 if (!bio_split_pool
)
1733 panic("bio: can't create split pool\n");
1737 subsys_initcall(init_bio
);