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/iocontext.h>
23 #include <linux/slab.h>
24 #include <linux/init.h>
25 #include <linux/kernel.h>
26 #include <linux/module.h>
27 #include <linux/mempool.h>
28 #include <linux/workqueue.h>
29 #include <linux/cgroup.h>
30 #include <scsi/sg.h> /* for struct sg_iovec */
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
40 static mempool_t
*bio_split_pool __read_mostly
;
43 * if you change this list, also change bvec_alloc or things will
44 * break badly! cannot be bigger than what you can fit into an
47 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
48 static struct biovec_slab bvec_slabs
[BIOVEC_NR_POOLS
] __read_mostly
= {
49 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES
),
54 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
55 * IO code that does not need private memory pools.
57 struct bio_set
*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
;
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) {
100 bio_slabs
= krealloc(bio_slabs
,
101 bio_slab_max
* sizeof(struct bio_slab
),
107 entry
= bio_slab_nr
++;
109 bslab
= &bio_slabs
[entry
];
111 snprintf(bslab
->name
, sizeof(bslab
->name
), "bio-%d", entry
);
112 slab
= kmem_cache_create(bslab
->name
, sz
, 0, SLAB_HWCACHE_ALIGN
, NULL
);
116 printk(KERN_INFO
"bio: create slab <%s> at %d\n", bslab
->name
, entry
);
119 bslab
->slab_size
= sz
;
121 mutex_unlock(&bio_slab_lock
);
125 static void bio_put_slab(struct bio_set
*bs
)
127 struct bio_slab
*bslab
= NULL
;
130 mutex_lock(&bio_slab_lock
);
132 for (i
= 0; i
< bio_slab_nr
; i
++) {
133 if (bs
->bio_slab
== bio_slabs
[i
].slab
) {
134 bslab
= &bio_slabs
[i
];
139 if (WARN(!bslab
, KERN_ERR
"bio: unable to find slab!\n"))
142 WARN_ON(!bslab
->slab_ref
);
144 if (--bslab
->slab_ref
)
147 kmem_cache_destroy(bslab
->slab
);
151 mutex_unlock(&bio_slab_lock
);
154 unsigned int bvec_nr_vecs(unsigned short idx
)
156 return bvec_slabs
[idx
].nr_vecs
;
159 void bvec_free_bs(struct bio_set
*bs
, struct bio_vec
*bv
, unsigned int idx
)
161 BIO_BUG_ON(idx
>= BIOVEC_NR_POOLS
);
163 if (idx
== BIOVEC_MAX_IDX
)
164 mempool_free(bv
, bs
->bvec_pool
);
166 struct biovec_slab
*bvs
= bvec_slabs
+ idx
;
168 kmem_cache_free(bvs
->slab
, bv
);
172 struct bio_vec
*bvec_alloc_bs(gfp_t gfp_mask
, int nr
, unsigned long *idx
,
178 * see comment near bvec_array define!
196 case 129 ... BIO_MAX_PAGES
:
204 * idx now points to the pool we want to allocate from. only the
205 * 1-vec entry pool is mempool backed.
207 if (*idx
== BIOVEC_MAX_IDX
) {
209 bvl
= mempool_alloc(bs
->bvec_pool
, gfp_mask
);
211 struct biovec_slab
*bvs
= bvec_slabs
+ *idx
;
212 gfp_t __gfp_mask
= gfp_mask
& ~(__GFP_WAIT
| __GFP_IO
);
215 * Make this allocation restricted and don't dump info on
216 * allocation failures, since we'll fallback to the mempool
217 * in case of failure.
219 __gfp_mask
|= __GFP_NOMEMALLOC
| __GFP_NORETRY
| __GFP_NOWARN
;
222 * Try a slab allocation. If this fails and __GFP_WAIT
223 * is set, retry with the 1-entry mempool
225 bvl
= kmem_cache_alloc(bvs
->slab
, __gfp_mask
);
226 if (unlikely(!bvl
&& (gfp_mask
& __GFP_WAIT
))) {
227 *idx
= BIOVEC_MAX_IDX
;
235 void bio_free(struct bio
*bio
, struct bio_set
*bs
)
239 if (bio_has_allocated_vec(bio
))
240 bvec_free_bs(bs
, bio
->bi_io_vec
, BIO_POOL_IDX(bio
));
242 if (bio_integrity(bio
))
243 bio_integrity_free(bio
, bs
);
246 * If we have front padding, adjust the bio pointer before freeing
252 mempool_free(p
, bs
->bio_pool
);
254 EXPORT_SYMBOL(bio_free
);
256 void bio_init(struct bio
*bio
)
258 memset(bio
, 0, sizeof(*bio
));
259 bio
->bi_flags
= 1 << BIO_UPTODATE
;
260 atomic_set(&bio
->bi_cnt
, 1);
262 EXPORT_SYMBOL(bio_init
);
265 * bio_alloc_bioset - allocate a bio for I/O
266 * @gfp_mask: the GFP_ mask given to the slab allocator
267 * @nr_iovecs: number of iovecs to pre-allocate
268 * @bs: the bio_set to allocate from.
271 * bio_alloc_bioset will try its own mempool to satisfy the allocation.
272 * If %__GFP_WAIT is set then we will block on the internal pool waiting
273 * for a &struct bio to become free.
275 * Note that the caller must set ->bi_destructor on successful return
276 * of a bio, to do the appropriate freeing of the bio once the reference
277 * count drops to zero.
279 struct bio
*bio_alloc_bioset(gfp_t gfp_mask
, int nr_iovecs
, struct bio_set
*bs
)
281 unsigned long idx
= BIO_POOL_NONE
;
282 struct bio_vec
*bvl
= NULL
;
286 p
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
289 bio
= p
+ bs
->front_pad
;
293 if (unlikely(!nr_iovecs
))
296 if (nr_iovecs
<= BIO_INLINE_VECS
) {
297 bvl
= bio
->bi_inline_vecs
;
298 nr_iovecs
= BIO_INLINE_VECS
;
300 bvl
= bvec_alloc_bs(gfp_mask
, nr_iovecs
, &idx
, bs
);
304 nr_iovecs
= bvec_nr_vecs(idx
);
307 bio
->bi_flags
|= idx
<< BIO_POOL_OFFSET
;
308 bio
->bi_max_vecs
= nr_iovecs
;
309 bio
->bi_io_vec
= bvl
;
313 mempool_free(p
, bs
->bio_pool
);
316 EXPORT_SYMBOL(bio_alloc_bioset
);
318 static void bio_fs_destructor(struct bio
*bio
)
320 bio_free(bio
, fs_bio_set
);
324 * bio_alloc - allocate a new bio, memory pool backed
325 * @gfp_mask: allocation mask to use
326 * @nr_iovecs: number of iovecs
328 * bio_alloc will allocate a bio and associated bio_vec array that can hold
329 * at least @nr_iovecs entries. Allocations will be done from the
330 * fs_bio_set. Also see @bio_alloc_bioset and @bio_kmalloc.
332 * If %__GFP_WAIT is set, then bio_alloc will always be able to allocate
333 * a bio. This is due to the mempool guarantees. To make this work, callers
334 * must never allocate more than 1 bio at a time from this pool. Callers
335 * that need to allocate more than 1 bio must always submit the previously
336 * allocated bio for IO before attempting to allocate a new one. Failure to
337 * do so can cause livelocks under memory pressure.
340 * Pointer to new bio on success, NULL on failure.
342 struct bio
*bio_alloc(gfp_t gfp_mask
, unsigned int nr_iovecs
)
344 struct bio
*bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, fs_bio_set
);
347 bio
->bi_destructor
= bio_fs_destructor
;
351 EXPORT_SYMBOL(bio_alloc
);
353 static void bio_kmalloc_destructor(struct bio
*bio
)
355 if (bio_integrity(bio
))
356 bio_integrity_free(bio
, fs_bio_set
);
361 * bio_kmalloc - allocate a bio for I/O using kmalloc()
362 * @gfp_mask: the GFP_ mask given to the slab allocator
363 * @nr_iovecs: number of iovecs to pre-allocate
366 * Allocate a new bio with @nr_iovecs bvecs. If @gfp_mask contains
367 * %__GFP_WAIT, the allocation is guaranteed to succeed.
370 struct bio
*bio_kmalloc(gfp_t gfp_mask
, unsigned int nr_iovecs
)
374 if (nr_iovecs
> UIO_MAXIOV
)
377 bio
= kmalloc(sizeof(struct bio
) + nr_iovecs
* sizeof(struct bio_vec
),
383 bio
->bi_flags
|= BIO_POOL_NONE
<< BIO_POOL_OFFSET
;
384 bio
->bi_max_vecs
= nr_iovecs
;
385 bio
->bi_io_vec
= bio
->bi_inline_vecs
;
386 bio
->bi_destructor
= bio_kmalloc_destructor
;
390 EXPORT_SYMBOL(bio_kmalloc
);
392 void zero_fill_bio(struct bio
*bio
)
398 bio_for_each_segment(bv
, bio
, i
) {
399 char *data
= bvec_kmap_irq(bv
, &flags
);
400 memset(data
, 0, bv
->bv_len
);
401 flush_dcache_page(bv
->bv_page
);
402 bvec_kunmap_irq(data
, &flags
);
405 EXPORT_SYMBOL(zero_fill_bio
);
408 * bio_put - release a reference to a bio
409 * @bio: bio to release reference to
412 * Put a reference to a &struct bio, either one you have gotten with
413 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
415 void bio_put(struct bio
*bio
)
417 BIO_BUG_ON(!atomic_read(&bio
->bi_cnt
));
422 if (atomic_dec_and_test(&bio
->bi_cnt
)) {
423 bio_disassociate_task(bio
);
425 bio
->bi_destructor(bio
);
428 EXPORT_SYMBOL(bio_put
);
430 inline int bio_phys_segments(struct request_queue
*q
, struct bio
*bio
)
432 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
433 blk_recount_segments(q
, bio
);
435 return bio
->bi_phys_segments
;
437 EXPORT_SYMBOL(bio_phys_segments
);
440 * __bio_clone - clone a bio
441 * @bio: destination bio
442 * @bio_src: bio to clone
444 * Clone a &bio. Caller will own the returned bio, but not
445 * the actual data it points to. Reference count of returned
448 void __bio_clone(struct bio
*bio
, struct bio
*bio_src
)
450 memcpy(bio
->bi_io_vec
, bio_src
->bi_io_vec
,
451 bio_src
->bi_max_vecs
* sizeof(struct bio_vec
));
454 * most users will be overriding ->bi_bdev with a new target,
455 * so we don't set nor calculate new physical/hw segment counts here
457 bio
->bi_sector
= bio_src
->bi_sector
;
458 bio
->bi_bdev
= bio_src
->bi_bdev
;
459 bio
->bi_flags
|= 1 << BIO_CLONED
;
460 bio
->bi_rw
= bio_src
->bi_rw
;
461 bio
->bi_vcnt
= bio_src
->bi_vcnt
;
462 bio
->bi_size
= bio_src
->bi_size
;
463 bio
->bi_idx
= bio_src
->bi_idx
;
465 EXPORT_SYMBOL(__bio_clone
);
468 * bio_clone - clone a bio
470 * @gfp_mask: allocation priority
472 * Like __bio_clone, only also allocates the returned bio
474 struct bio
*bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
476 struct bio
*b
= bio_alloc_bioset(gfp_mask
, bio
->bi_max_vecs
, fs_bio_set
);
481 b
->bi_destructor
= bio_fs_destructor
;
484 if (bio_integrity(bio
)) {
487 ret
= bio_integrity_clone(b
, bio
, gfp_mask
, fs_bio_set
);
497 EXPORT_SYMBOL(bio_clone
);
500 * bio_get_nr_vecs - return approx number of vecs
503 * Return the approximate number of pages we can send to this target.
504 * There's no guarantee that you will be able to fit this number of pages
505 * into a bio, it does not account for dynamic restrictions that vary
508 int bio_get_nr_vecs(struct block_device
*bdev
)
510 struct request_queue
*q
= bdev_get_queue(bdev
);
511 return min_t(unsigned,
512 queue_max_segments(q
),
513 queue_max_sectors(q
) / (PAGE_SIZE
>> 9) + 1);
515 EXPORT_SYMBOL(bio_get_nr_vecs
);
517 static int __bio_add_page(struct request_queue
*q
, struct bio
*bio
, struct page
518 *page
, unsigned int len
, unsigned int offset
,
519 unsigned short max_sectors
)
521 int retried_segments
= 0;
522 struct bio_vec
*bvec
;
525 * cloned bio must not modify vec list
527 if (unlikely(bio_flagged(bio
, BIO_CLONED
)))
530 if (((bio
->bi_size
+ len
) >> 9) > max_sectors
)
534 * For filesystems with a blocksize smaller than the pagesize
535 * we will often be called with the same page as last time and
536 * a consecutive offset. Optimize this special case.
538 if (bio
->bi_vcnt
> 0) {
539 struct bio_vec
*prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
541 if (page
== prev
->bv_page
&&
542 offset
== prev
->bv_offset
+ prev
->bv_len
) {
543 unsigned int prev_bv_len
= prev
->bv_len
;
546 if (q
->merge_bvec_fn
) {
547 struct bvec_merge_data bvm
= {
548 /* prev_bvec is already charged in
549 bi_size, discharge it in order to
550 simulate merging updated prev_bvec
552 .bi_bdev
= bio
->bi_bdev
,
553 .bi_sector
= bio
->bi_sector
,
554 .bi_size
= bio
->bi_size
- prev_bv_len
,
558 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < prev
->bv_len
) {
568 if (bio
->bi_vcnt
>= bio
->bi_max_vecs
)
572 * we might lose a segment or two here, but rather that than
573 * make this too complex.
576 while (bio
->bi_phys_segments
>= queue_max_segments(q
)) {
578 if (retried_segments
)
581 retried_segments
= 1;
582 blk_recount_segments(q
, bio
);
586 * setup the new entry, we might clear it again later if we
587 * cannot add the page
589 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
590 bvec
->bv_page
= page
;
592 bvec
->bv_offset
= offset
;
595 * if queue has other restrictions (eg varying max sector size
596 * depending on offset), it can specify a merge_bvec_fn in the
597 * queue to get further control
599 if (q
->merge_bvec_fn
) {
600 struct bvec_merge_data bvm
= {
601 .bi_bdev
= bio
->bi_bdev
,
602 .bi_sector
= bio
->bi_sector
,
603 .bi_size
= bio
->bi_size
,
608 * merge_bvec_fn() returns number of bytes it can accept
611 if (q
->merge_bvec_fn(q
, &bvm
, bvec
) < bvec
->bv_len
) {
612 bvec
->bv_page
= NULL
;
619 /* If we may be able to merge these biovecs, force a recount */
620 if (bio
->bi_vcnt
&& (BIOVEC_PHYS_MERGEABLE(bvec
-1, bvec
)))
621 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
624 bio
->bi_phys_segments
++;
631 * bio_add_pc_page - attempt to add page to bio
632 * @q: the target queue
633 * @bio: destination bio
635 * @len: vec entry length
636 * @offset: vec entry offset
638 * Attempt to add a page to the bio_vec maplist. This can fail for a
639 * number of reasons, such as the bio being full or target block device
640 * limitations. The target block device must allow bio's up to PAGE_SIZE,
641 * so it is always possible to add a single page to an empty bio.
643 * This should only be used by REQ_PC bios.
645 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
, struct page
*page
,
646 unsigned int len
, unsigned int offset
)
648 return __bio_add_page(q
, bio
, page
, len
, offset
,
649 queue_max_hw_sectors(q
));
651 EXPORT_SYMBOL(bio_add_pc_page
);
654 * bio_add_page - attempt to add page to bio
655 * @bio: destination bio
657 * @len: vec entry length
658 * @offset: vec entry offset
660 * Attempt to add a page to the bio_vec maplist. This can fail for a
661 * number of reasons, such as the bio being full or target block device
662 * limitations. The target block device must allow bio's up to PAGE_SIZE,
663 * so it is always possible to add a single page to an empty bio.
665 int bio_add_page(struct bio
*bio
, struct page
*page
, unsigned int len
,
668 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
669 return __bio_add_page(q
, bio
, page
, len
, offset
, queue_max_sectors(q
));
671 EXPORT_SYMBOL(bio_add_page
);
673 struct bio_map_data
{
674 struct bio_vec
*iovecs
;
675 struct sg_iovec
*sgvecs
;
680 static void bio_set_map_data(struct bio_map_data
*bmd
, struct bio
*bio
,
681 struct sg_iovec
*iov
, int iov_count
,
684 memcpy(bmd
->iovecs
, bio
->bi_io_vec
, sizeof(struct bio_vec
) * bio
->bi_vcnt
);
685 memcpy(bmd
->sgvecs
, iov
, sizeof(struct sg_iovec
) * iov_count
);
686 bmd
->nr_sgvecs
= iov_count
;
687 bmd
->is_our_pages
= is_our_pages
;
688 bio
->bi_private
= bmd
;
691 static void bio_free_map_data(struct bio_map_data
*bmd
)
698 static struct bio_map_data
*bio_alloc_map_data(int nr_segs
,
699 unsigned int iov_count
,
702 struct bio_map_data
*bmd
;
704 if (iov_count
> UIO_MAXIOV
)
707 bmd
= kmalloc(sizeof(*bmd
), gfp_mask
);
711 bmd
->iovecs
= kmalloc(sizeof(struct bio_vec
) * nr_segs
, gfp_mask
);
717 bmd
->sgvecs
= kmalloc(sizeof(struct sg_iovec
) * iov_count
, gfp_mask
);
726 static int __bio_copy_iov(struct bio
*bio
, struct bio_vec
*iovecs
,
727 struct sg_iovec
*iov
, int iov_count
,
728 int to_user
, int from_user
, int do_free_page
)
731 struct bio_vec
*bvec
;
733 unsigned int iov_off
= 0;
735 __bio_for_each_segment(bvec
, bio
, i
, 0) {
736 char *bv_addr
= page_address(bvec
->bv_page
);
737 unsigned int bv_len
= iovecs
[i
].bv_len
;
739 while (bv_len
&& iov_idx
< iov_count
) {
741 char __user
*iov_addr
;
743 bytes
= min_t(unsigned int,
744 iov
[iov_idx
].iov_len
- iov_off
, bv_len
);
745 iov_addr
= iov
[iov_idx
].iov_base
+ iov_off
;
749 ret
= copy_to_user(iov_addr
, bv_addr
,
753 ret
= copy_from_user(bv_addr
, iov_addr
,
765 if (iov
[iov_idx
].iov_len
== iov_off
) {
772 __free_page(bvec
->bv_page
);
779 * bio_uncopy_user - finish previously mapped bio
780 * @bio: bio being terminated
782 * Free pages allocated from bio_copy_user() and write back data
783 * to user space in case of a read.
785 int bio_uncopy_user(struct bio
*bio
)
787 struct bio_map_data
*bmd
= bio
->bi_private
;
790 if (!bio_flagged(bio
, BIO_NULL_MAPPED
))
791 ret
= __bio_copy_iov(bio
, bmd
->iovecs
, bmd
->sgvecs
,
792 bmd
->nr_sgvecs
, bio_data_dir(bio
) == READ
,
793 0, bmd
->is_our_pages
);
794 bio_free_map_data(bmd
);
798 EXPORT_SYMBOL(bio_uncopy_user
);
801 * bio_copy_user_iov - copy user data to bio
802 * @q: destination block queue
803 * @map_data: pointer to the rq_map_data holding pages (if necessary)
805 * @iov_count: number of elements in the iovec
806 * @write_to_vm: bool indicating writing to pages or not
807 * @gfp_mask: memory allocation flags
809 * Prepares and returns a bio for indirect user io, bouncing data
810 * to/from kernel pages as necessary. Must be paired with
811 * call bio_uncopy_user() on io completion.
813 struct bio
*bio_copy_user_iov(struct request_queue
*q
,
814 struct rq_map_data
*map_data
,
815 struct sg_iovec
*iov
, int iov_count
,
816 int write_to_vm
, gfp_t gfp_mask
)
818 struct bio_map_data
*bmd
;
819 struct bio_vec
*bvec
;
824 unsigned int len
= 0;
825 unsigned int offset
= map_data
? map_data
->offset
& ~PAGE_MASK
: 0;
827 for (i
= 0; i
< iov_count
; i
++) {
832 uaddr
= (unsigned long)iov
[i
].iov_base
;
833 end
= (uaddr
+ iov
[i
].iov_len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
834 start
= uaddr
>> PAGE_SHIFT
;
840 return ERR_PTR(-EINVAL
);
842 nr_pages
+= end
- start
;
843 len
+= iov
[i
].iov_len
;
849 bmd
= bio_alloc_map_data(nr_pages
, iov_count
, gfp_mask
);
851 return ERR_PTR(-ENOMEM
);
854 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
859 bio
->bi_rw
|= REQ_WRITE
;
864 nr_pages
= 1 << map_data
->page_order
;
865 i
= map_data
->offset
/ PAGE_SIZE
;
868 unsigned int bytes
= PAGE_SIZE
;
876 if (i
== map_data
->nr_entries
* nr_pages
) {
881 page
= map_data
->pages
[i
/ nr_pages
];
882 page
+= (i
% nr_pages
);
886 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
893 if (bio_add_pc_page(q
, bio
, page
, bytes
, offset
) < bytes
)
906 if ((!write_to_vm
&& (!map_data
|| !map_data
->null_mapped
)) ||
907 (map_data
&& map_data
->from_user
)) {
908 ret
= __bio_copy_iov(bio
, bio
->bi_io_vec
, iov
, iov_count
, 0, 1, 0);
913 bio_set_map_data(bmd
, bio
, iov
, iov_count
, map_data
? 0 : 1);
917 bio_for_each_segment(bvec
, bio
, i
)
918 __free_page(bvec
->bv_page
);
922 bio_free_map_data(bmd
);
927 * bio_copy_user - copy user data to bio
928 * @q: destination block queue
929 * @map_data: pointer to the rq_map_data holding pages (if necessary)
930 * @uaddr: start of user address
931 * @len: length in bytes
932 * @write_to_vm: bool indicating writing to pages or not
933 * @gfp_mask: memory allocation flags
935 * Prepares and returns a bio for indirect user io, bouncing data
936 * to/from kernel pages as necessary. Must be paired with
937 * call bio_uncopy_user() on io completion.
939 struct bio
*bio_copy_user(struct request_queue
*q
, struct rq_map_data
*map_data
,
940 unsigned long uaddr
, unsigned int len
,
941 int write_to_vm
, gfp_t gfp_mask
)
945 iov
.iov_base
= (void __user
*)uaddr
;
948 return bio_copy_user_iov(q
, map_data
, &iov
, 1, write_to_vm
, gfp_mask
);
950 EXPORT_SYMBOL(bio_copy_user
);
952 static struct bio
*__bio_map_user_iov(struct request_queue
*q
,
953 struct block_device
*bdev
,
954 struct sg_iovec
*iov
, int iov_count
,
955 int write_to_vm
, gfp_t gfp_mask
)
964 for (i
= 0; i
< iov_count
; i
++) {
965 unsigned long uaddr
= (unsigned long)iov
[i
].iov_base
;
966 unsigned long len
= iov
[i
].iov_len
;
967 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
968 unsigned long start
= uaddr
>> PAGE_SHIFT
;
974 return ERR_PTR(-EINVAL
);
976 nr_pages
+= end
- start
;
978 * buffer must be aligned to at least hardsector size for now
980 if (uaddr
& queue_dma_alignment(q
))
981 return ERR_PTR(-EINVAL
);
985 return ERR_PTR(-EINVAL
);
987 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
989 return ERR_PTR(-ENOMEM
);
992 pages
= kcalloc(nr_pages
, sizeof(struct page
*), gfp_mask
);
996 for (i
= 0; i
< iov_count
; i
++) {
997 unsigned long uaddr
= (unsigned long)iov
[i
].iov_base
;
998 unsigned long len
= iov
[i
].iov_len
;
999 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1000 unsigned long start
= uaddr
>> PAGE_SHIFT
;
1001 const int local_nr_pages
= end
- start
;
1002 const int page_limit
= cur_page
+ local_nr_pages
;
1004 ret
= get_user_pages_fast(uaddr
, local_nr_pages
,
1005 write_to_vm
, &pages
[cur_page
]);
1006 if (ret
< local_nr_pages
) {
1011 offset
= uaddr
& ~PAGE_MASK
;
1012 for (j
= cur_page
; j
< page_limit
; j
++) {
1013 unsigned int bytes
= PAGE_SIZE
- offset
;
1024 if (bio_add_pc_page(q
, bio
, pages
[j
], bytes
, offset
) <
1034 * release the pages we didn't map into the bio, if any
1036 while (j
< page_limit
)
1037 page_cache_release(pages
[j
++]);
1043 * set data direction, and check if mapped pages need bouncing
1046 bio
->bi_rw
|= REQ_WRITE
;
1048 bio
->bi_bdev
= bdev
;
1049 bio
->bi_flags
|= (1 << BIO_USER_MAPPED
);
1053 for (i
= 0; i
< nr_pages
; i
++) {
1056 page_cache_release(pages
[i
]);
1061 return ERR_PTR(ret
);
1065 * bio_map_user - map user address into bio
1066 * @q: the struct request_queue for the bio
1067 * @bdev: destination block device
1068 * @uaddr: start of user address
1069 * @len: length in bytes
1070 * @write_to_vm: bool indicating writing to pages or not
1071 * @gfp_mask: memory allocation flags
1073 * Map the user space address into a bio suitable for io to a block
1074 * device. Returns an error pointer in case of error.
1076 struct bio
*bio_map_user(struct request_queue
*q
, struct block_device
*bdev
,
1077 unsigned long uaddr
, unsigned int len
, int write_to_vm
,
1080 struct sg_iovec iov
;
1082 iov
.iov_base
= (void __user
*)uaddr
;
1085 return bio_map_user_iov(q
, bdev
, &iov
, 1, write_to_vm
, gfp_mask
);
1087 EXPORT_SYMBOL(bio_map_user
);
1090 * bio_map_user_iov - map user sg_iovec table into bio
1091 * @q: the struct request_queue for the bio
1092 * @bdev: destination block device
1094 * @iov_count: number of elements in the iovec
1095 * @write_to_vm: bool indicating writing to pages or not
1096 * @gfp_mask: memory allocation flags
1098 * Map the user space address into a bio suitable for io to a block
1099 * device. Returns an error pointer in case of error.
1101 struct bio
*bio_map_user_iov(struct request_queue
*q
, struct block_device
*bdev
,
1102 struct sg_iovec
*iov
, int iov_count
,
1103 int write_to_vm
, gfp_t gfp_mask
)
1107 bio
= __bio_map_user_iov(q
, bdev
, iov
, iov_count
, write_to_vm
,
1113 * subtle -- if __bio_map_user() ended up bouncing a bio,
1114 * it would normally disappear when its bi_end_io is run.
1115 * however, we need it for the unmap, so grab an extra
1123 static void __bio_unmap_user(struct bio
*bio
)
1125 struct bio_vec
*bvec
;
1129 * make sure we dirty pages we wrote to
1131 __bio_for_each_segment(bvec
, bio
, i
, 0) {
1132 if (bio_data_dir(bio
) == READ
)
1133 set_page_dirty_lock(bvec
->bv_page
);
1135 page_cache_release(bvec
->bv_page
);
1142 * bio_unmap_user - unmap a bio
1143 * @bio: the bio being unmapped
1145 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1146 * a process context.
1148 * bio_unmap_user() may sleep.
1150 void bio_unmap_user(struct bio
*bio
)
1152 __bio_unmap_user(bio
);
1155 EXPORT_SYMBOL(bio_unmap_user
);
1157 static void bio_map_kern_endio(struct bio
*bio
, int err
)
1162 static struct bio
*__bio_map_kern(struct request_queue
*q
, void *data
,
1163 unsigned int len
, gfp_t gfp_mask
)
1165 unsigned long kaddr
= (unsigned long)data
;
1166 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1167 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1168 const int nr_pages
= end
- start
;
1172 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1174 return ERR_PTR(-ENOMEM
);
1176 offset
= offset_in_page(kaddr
);
1177 for (i
= 0; i
< nr_pages
; i
++) {
1178 unsigned int bytes
= PAGE_SIZE
- offset
;
1186 if (bio_add_pc_page(q
, bio
, virt_to_page(data
), bytes
,
1195 bio
->bi_end_io
= bio_map_kern_endio
;
1200 * bio_map_kern - map kernel address into bio
1201 * @q: the struct request_queue for the bio
1202 * @data: pointer to buffer to map
1203 * @len: length in bytes
1204 * @gfp_mask: allocation flags for bio allocation
1206 * Map the kernel address into a bio suitable for io to a block
1207 * device. Returns an error pointer in case of error.
1209 struct bio
*bio_map_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1214 bio
= __bio_map_kern(q
, data
, len
, gfp_mask
);
1218 if (bio
->bi_size
== len
)
1222 * Don't support partial mappings.
1225 return ERR_PTR(-EINVAL
);
1227 EXPORT_SYMBOL(bio_map_kern
);
1229 static void bio_copy_kern_endio(struct bio
*bio
, int err
)
1231 struct bio_vec
*bvec
;
1232 const int read
= bio_data_dir(bio
) == READ
;
1233 struct bio_map_data
*bmd
= bio
->bi_private
;
1235 char *p
= bmd
->sgvecs
[0].iov_base
;
1237 __bio_for_each_segment(bvec
, bio
, i
, 0) {
1238 char *addr
= page_address(bvec
->bv_page
);
1239 int len
= bmd
->iovecs
[i
].bv_len
;
1242 memcpy(p
, addr
, len
);
1244 __free_page(bvec
->bv_page
);
1248 bio_free_map_data(bmd
);
1253 * bio_copy_kern - copy kernel address into bio
1254 * @q: the struct request_queue for the bio
1255 * @data: pointer to buffer to copy
1256 * @len: length in bytes
1257 * @gfp_mask: allocation flags for bio and page allocation
1258 * @reading: data direction is READ
1260 * copy the kernel address into a bio suitable for io to a block
1261 * device. Returns an error pointer in case of error.
1263 struct bio
*bio_copy_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1264 gfp_t gfp_mask
, int reading
)
1267 struct bio_vec
*bvec
;
1270 bio
= bio_copy_user(q
, NULL
, (unsigned long)data
, len
, 1, gfp_mask
);
1277 bio_for_each_segment(bvec
, bio
, i
) {
1278 char *addr
= page_address(bvec
->bv_page
);
1280 memcpy(addr
, p
, bvec
->bv_len
);
1285 bio
->bi_end_io
= bio_copy_kern_endio
;
1289 EXPORT_SYMBOL(bio_copy_kern
);
1292 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1293 * for performing direct-IO in BIOs.
1295 * The problem is that we cannot run set_page_dirty() from interrupt context
1296 * because the required locks are not interrupt-safe. So what we can do is to
1297 * mark the pages dirty _before_ performing IO. And in interrupt context,
1298 * check that the pages are still dirty. If so, fine. If not, redirty them
1299 * in process context.
1301 * We special-case compound pages here: normally this means reads into hugetlb
1302 * pages. The logic in here doesn't really work right for compound pages
1303 * because the VM does not uniformly chase down the head page in all cases.
1304 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1305 * handle them at all. So we skip compound pages here at an early stage.
1307 * Note that this code is very hard to test under normal circumstances because
1308 * direct-io pins the pages with get_user_pages(). This makes
1309 * is_page_cache_freeable return false, and the VM will not clean the pages.
1310 * But other code (eg, pdflush) could clean the pages if they are mapped
1313 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1314 * deferred bio dirtying paths.
1318 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1320 void bio_set_pages_dirty(struct bio
*bio
)
1322 struct bio_vec
*bvec
= bio
->bi_io_vec
;
1325 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
1326 struct page
*page
= bvec
[i
].bv_page
;
1328 if (page
&& !PageCompound(page
))
1329 set_page_dirty_lock(page
);
1333 static void bio_release_pages(struct bio
*bio
)
1335 struct bio_vec
*bvec
= bio
->bi_io_vec
;
1338 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
1339 struct page
*page
= bvec
[i
].bv_page
;
1347 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1348 * If they are, then fine. If, however, some pages are clean then they must
1349 * have been written out during the direct-IO read. So we take another ref on
1350 * the BIO and the offending pages and re-dirty the pages in process context.
1352 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1353 * here on. It will run one page_cache_release() against each page and will
1354 * run one bio_put() against the BIO.
1357 static void bio_dirty_fn(struct work_struct
*work
);
1359 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
1360 static DEFINE_SPINLOCK(bio_dirty_lock
);
1361 static struct bio
*bio_dirty_list
;
1364 * This runs in process context
1366 static void bio_dirty_fn(struct work_struct
*work
)
1368 unsigned long flags
;
1371 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1372 bio
= bio_dirty_list
;
1373 bio_dirty_list
= NULL
;
1374 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1377 struct bio
*next
= bio
->bi_private
;
1379 bio_set_pages_dirty(bio
);
1380 bio_release_pages(bio
);
1386 void bio_check_pages_dirty(struct bio
*bio
)
1388 struct bio_vec
*bvec
= bio
->bi_io_vec
;
1389 int nr_clean_pages
= 0;
1392 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
1393 struct page
*page
= bvec
[i
].bv_page
;
1395 if (PageDirty(page
) || PageCompound(page
)) {
1396 page_cache_release(page
);
1397 bvec
[i
].bv_page
= NULL
;
1403 if (nr_clean_pages
) {
1404 unsigned long flags
;
1406 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1407 bio
->bi_private
= bio_dirty_list
;
1408 bio_dirty_list
= bio
;
1409 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1410 schedule_work(&bio_dirty_work
);
1416 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1417 void bio_flush_dcache_pages(struct bio
*bi
)
1420 struct bio_vec
*bvec
;
1422 bio_for_each_segment(bvec
, bi
, i
)
1423 flush_dcache_page(bvec
->bv_page
);
1425 EXPORT_SYMBOL(bio_flush_dcache_pages
);
1429 * bio_endio - end I/O on a bio
1431 * @error: error, if any
1434 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1435 * preferred way to end I/O on a bio, it takes care of clearing
1436 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1437 * established -Exxxx (-EIO, for instance) error values in case
1438 * something went wrong. No one should call bi_end_io() directly on a
1439 * bio unless they own it and thus know that it has an end_io
1442 void bio_endio(struct bio
*bio
, int error
)
1445 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1446 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
1450 bio
->bi_end_io(bio
, error
);
1452 EXPORT_SYMBOL(bio_endio
);
1454 void bio_pair_release(struct bio_pair
*bp
)
1456 if (atomic_dec_and_test(&bp
->cnt
)) {
1457 struct bio
*master
= bp
->bio1
.bi_private
;
1459 bio_endio(master
, bp
->error
);
1460 mempool_free(bp
, bp
->bio2
.bi_private
);
1463 EXPORT_SYMBOL(bio_pair_release
);
1465 static void bio_pair_end_1(struct bio
*bi
, int err
)
1467 struct bio_pair
*bp
= container_of(bi
, struct bio_pair
, bio1
);
1472 bio_pair_release(bp
);
1475 static void bio_pair_end_2(struct bio
*bi
, int err
)
1477 struct bio_pair
*bp
= container_of(bi
, struct bio_pair
, bio2
);
1482 bio_pair_release(bp
);
1486 * split a bio - only worry about a bio with a single page in its iovec
1488 struct bio_pair
*bio_split(struct bio
*bi
, int first_sectors
)
1490 struct bio_pair
*bp
= mempool_alloc(bio_split_pool
, GFP_NOIO
);
1495 trace_block_split(bdev_get_queue(bi
->bi_bdev
), bi
,
1496 bi
->bi_sector
+ first_sectors
);
1498 BUG_ON(bi
->bi_vcnt
!= 1);
1499 BUG_ON(bi
->bi_idx
!= 0);
1500 atomic_set(&bp
->cnt
, 3);
1504 bp
->bio2
.bi_sector
+= first_sectors
;
1505 bp
->bio2
.bi_size
-= first_sectors
<< 9;
1506 bp
->bio1
.bi_size
= first_sectors
<< 9;
1508 bp
->bv1
= bi
->bi_io_vec
[0];
1509 bp
->bv2
= bi
->bi_io_vec
[0];
1510 bp
->bv2
.bv_offset
+= first_sectors
<< 9;
1511 bp
->bv2
.bv_len
-= first_sectors
<< 9;
1512 bp
->bv1
.bv_len
= first_sectors
<< 9;
1514 bp
->bio1
.bi_io_vec
= &bp
->bv1
;
1515 bp
->bio2
.bi_io_vec
= &bp
->bv2
;
1517 bp
->bio1
.bi_max_vecs
= 1;
1518 bp
->bio2
.bi_max_vecs
= 1;
1520 bp
->bio1
.bi_end_io
= bio_pair_end_1
;
1521 bp
->bio2
.bi_end_io
= bio_pair_end_2
;
1523 bp
->bio1
.bi_private
= bi
;
1524 bp
->bio2
.bi_private
= bio_split_pool
;
1526 if (bio_integrity(bi
))
1527 bio_integrity_split(bi
, bp
, first_sectors
);
1531 EXPORT_SYMBOL(bio_split
);
1534 * bio_sector_offset - Find hardware sector offset in bio
1535 * @bio: bio to inspect
1536 * @index: bio_vec index
1537 * @offset: offset in bv_page
1539 * Return the number of hardware sectors between beginning of bio
1540 * and an end point indicated by a bio_vec index and an offset
1541 * within that vector's page.
1543 sector_t
bio_sector_offset(struct bio
*bio
, unsigned short index
,
1544 unsigned int offset
)
1546 unsigned int sector_sz
;
1551 sector_sz
= queue_logical_block_size(bio
->bi_bdev
->bd_disk
->queue
);
1554 if (index
>= bio
->bi_idx
)
1555 index
= bio
->bi_vcnt
- 1;
1557 __bio_for_each_segment(bv
, bio
, i
, 0) {
1559 if (offset
> bv
->bv_offset
)
1560 sectors
+= (offset
- bv
->bv_offset
) / sector_sz
;
1564 sectors
+= bv
->bv_len
/ sector_sz
;
1569 EXPORT_SYMBOL(bio_sector_offset
);
1572 * create memory pools for biovec's in a bio_set.
1573 * use the global biovec slabs created for general use.
1575 static int biovec_create_pools(struct bio_set
*bs
, int pool_entries
)
1577 struct biovec_slab
*bp
= bvec_slabs
+ BIOVEC_MAX_IDX
;
1579 bs
->bvec_pool
= mempool_create_slab_pool(pool_entries
, bp
->slab
);
1586 static void biovec_free_pools(struct bio_set
*bs
)
1588 mempool_destroy(bs
->bvec_pool
);
1591 void bioset_free(struct bio_set
*bs
)
1594 mempool_destroy(bs
->bio_pool
);
1596 bioset_integrity_free(bs
);
1597 biovec_free_pools(bs
);
1602 EXPORT_SYMBOL(bioset_free
);
1605 * bioset_create - Create a bio_set
1606 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1607 * @front_pad: Number of bytes to allocate in front of the returned bio
1610 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1611 * to ask for a number of bytes to be allocated in front of the bio.
1612 * Front pad allocation is useful for embedding the bio inside
1613 * another structure, to avoid allocating extra data to go with the bio.
1614 * Note that the bio must be embedded at the END of that structure always,
1615 * or things will break badly.
1617 struct bio_set
*bioset_create(unsigned int pool_size
, unsigned int front_pad
)
1619 unsigned int back_pad
= BIO_INLINE_VECS
* sizeof(struct bio_vec
);
1622 bs
= kzalloc(sizeof(*bs
), GFP_KERNEL
);
1626 bs
->front_pad
= front_pad
;
1628 bs
->bio_slab
= bio_find_or_create_slab(front_pad
+ back_pad
);
1629 if (!bs
->bio_slab
) {
1634 bs
->bio_pool
= mempool_create_slab_pool(pool_size
, bs
->bio_slab
);
1638 if (!biovec_create_pools(bs
, pool_size
))
1645 EXPORT_SYMBOL(bioset_create
);
1647 #ifdef CONFIG_BLK_CGROUP
1649 * bio_associate_current - associate a bio with %current
1652 * Associate @bio with %current if it hasn't been associated yet. Block
1653 * layer will treat @bio as if it were issued by %current no matter which
1654 * task actually issues it.
1656 * This function takes an extra reference of @task's io_context and blkcg
1657 * which will be put when @bio is released. The caller must own @bio,
1658 * ensure %current->io_context exists, and is responsible for synchronizing
1659 * calls to this function.
1661 int bio_associate_current(struct bio
*bio
)
1663 struct io_context
*ioc
;
1664 struct cgroup_subsys_state
*css
;
1669 ioc
= current
->io_context
;
1673 /* acquire active ref on @ioc and associate */
1674 get_io_context_active(ioc
);
1677 /* associate blkcg if exists */
1679 css
= task_subsys_state(current
, blkio_subsys_id
);
1680 if (css
&& css_tryget(css
))
1688 * bio_disassociate_task - undo bio_associate_current()
1691 void bio_disassociate_task(struct bio
*bio
)
1694 put_io_context(bio
->bi_ioc
);
1698 css_put(bio
->bi_css
);
1703 #endif /* CONFIG_BLK_CGROUP */
1705 static void __init
biovec_init_slabs(void)
1709 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
1711 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
1713 if (bvs
->nr_vecs
<= BIO_INLINE_VECS
) {
1718 size
= bvs
->nr_vecs
* sizeof(struct bio_vec
);
1719 bvs
->slab
= kmem_cache_create(bvs
->name
, size
, 0,
1720 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
1724 static int __init
init_bio(void)
1728 bio_slabs
= kzalloc(bio_slab_max
* sizeof(struct bio_slab
), GFP_KERNEL
);
1730 panic("bio: can't allocate bios\n");
1732 bio_integrity_init();
1733 biovec_init_slabs();
1735 fs_bio_set
= bioset_create(BIO_POOL_SIZE
, 0);
1737 panic("bio: can't allocate bios\n");
1739 if (bioset_integrity_create(fs_bio_set
, BIO_POOL_SIZE
))
1740 panic("bio: can't create integrity pool\n");
1742 bio_split_pool
= mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES
,
1743 sizeof(struct bio_pair
));
1744 if (!bio_split_pool
)
1745 panic("bio: can't create split pool\n");
1749 subsys_initcall(init_bio
);