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/slab.h>
23 #include <linux/init.h>
24 #include <linux/kernel.h>
25 #include <linux/module.h>
26 #include <linux/mempool.h>
27 #include <linux/workqueue.h>
28 #include <linux/blktrace_api.h>
29 #include <trace/block.h>
30 #include <scsi/sg.h> /* for struct sg_iovec */
32 DEFINE_TRACE(block_split
);
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 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 struct bio_slab
*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("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 * If 'bs' is given, lookup the pool and do the mempool alloc.
179 * If not, this is a bio_kmalloc() allocation and just do a
180 * kzalloc() for the exact number of vecs right away.
183 bvl
= kmalloc(nr
* sizeof(struct bio_vec
), gfp_mask
);
186 * see comment near bvec_array define!
204 case 129 ... BIO_MAX_PAGES
:
212 * idx now points to the pool we want to allocate from. only the
213 * 1-vec entry pool is mempool backed.
215 if (*idx
== BIOVEC_MAX_IDX
) {
217 bvl
= mempool_alloc(bs
->bvec_pool
, gfp_mask
);
219 struct biovec_slab
*bvs
= bvec_slabs
+ *idx
;
220 gfp_t __gfp_mask
= gfp_mask
& ~(__GFP_WAIT
| __GFP_IO
);
223 * Make this allocation restricted and don't dump info on
224 * allocation failures, since we'll fallback to the mempool
225 * in case of failure.
227 __gfp_mask
|= __GFP_NOMEMALLOC
| __GFP_NORETRY
| __GFP_NOWARN
;
230 * Try a slab allocation. If this fails and __GFP_WAIT
231 * is set, retry with the 1-entry mempool
233 bvl
= kmem_cache_alloc(bvs
->slab
, __gfp_mask
);
234 if (unlikely(!bvl
&& (gfp_mask
& __GFP_WAIT
))) {
235 *idx
= BIOVEC_MAX_IDX
;
243 void bio_free(struct bio
*bio
, struct bio_set
*bs
)
247 if (bio_has_allocated_vec(bio
))
248 bvec_free_bs(bs
, bio
->bi_io_vec
, BIO_POOL_IDX(bio
));
250 if (bio_integrity(bio
))
251 bio_integrity_free(bio
, bs
);
254 * If we have front padding, adjust the bio pointer before freeing
260 mempool_free(p
, bs
->bio_pool
);
264 * default destructor for a bio allocated with bio_alloc_bioset()
266 static void bio_fs_destructor(struct bio
*bio
)
268 bio_free(bio
, fs_bio_set
);
271 static void bio_kmalloc_destructor(struct bio
*bio
)
273 if (bio_has_allocated_vec(bio
))
274 kfree(bio
->bi_io_vec
);
278 void bio_init(struct bio
*bio
)
280 memset(bio
, 0, sizeof(*bio
));
281 bio
->bi_flags
= 1 << BIO_UPTODATE
;
282 bio
->bi_comp_cpu
= -1;
283 atomic_set(&bio
->bi_cnt
, 1);
287 * bio_alloc_bioset - allocate a bio for I/O
288 * @gfp_mask: the GFP_ mask given to the slab allocator
289 * @nr_iovecs: number of iovecs to pre-allocate
290 * @bs: the bio_set to allocate from. If %NULL, just use kmalloc
293 * bio_alloc_bioset will first try its own mempool to satisfy the allocation.
294 * If %__GFP_WAIT is set then we will block on the internal pool waiting
295 * for a &struct bio to become free. If a %NULL @bs is passed in, we will
296 * fall back to just using @kmalloc to allocate the required memory.
298 * Note that the caller must set ->bi_destructor on succesful return
299 * of a bio, to do the appropriate freeing of the bio once the reference
300 * count drops to zero.
302 struct bio
*bio_alloc_bioset(gfp_t gfp_mask
, int nr_iovecs
, struct bio_set
*bs
)
304 struct bio
*bio
= NULL
;
305 void *uninitialized_var(p
);
308 p
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
311 bio
= p
+ bs
->front_pad
;
313 bio
= kmalloc(sizeof(*bio
), gfp_mask
);
316 struct bio_vec
*bvl
= NULL
;
319 if (likely(nr_iovecs
)) {
320 unsigned long uninitialized_var(idx
);
322 if (nr_iovecs
<= BIO_INLINE_VECS
) {
324 bvl
= bio
->bi_inline_vecs
;
325 nr_iovecs
= BIO_INLINE_VECS
;
327 bvl
= bvec_alloc_bs(gfp_mask
, nr_iovecs
, &idx
,
329 nr_iovecs
= bvec_nr_vecs(idx
);
331 if (unlikely(!bvl
)) {
333 mempool_free(p
, bs
->bio_pool
);
339 bio
->bi_flags
|= idx
<< BIO_POOL_OFFSET
;
340 bio
->bi_max_vecs
= nr_iovecs
;
342 bio
->bi_io_vec
= bvl
;
348 struct bio
*bio_alloc(gfp_t gfp_mask
, int nr_iovecs
)
350 struct bio
*bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, fs_bio_set
);
353 bio
->bi_destructor
= bio_fs_destructor
;
359 * Like bio_alloc(), but doesn't use a mempool backing. This means that
360 * it CAN fail, but while bio_alloc() can only be used for allocations
361 * that have a short (finite) life span, bio_kmalloc() should be used
362 * for more permanent bio allocations (like allocating some bio's for
363 * initalization or setup purposes).
365 struct bio
*bio_kmalloc(gfp_t gfp_mask
, int nr_iovecs
)
367 struct bio
*bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, NULL
);
370 bio
->bi_destructor
= bio_kmalloc_destructor
;
375 void zero_fill_bio(struct bio
*bio
)
381 bio_for_each_segment(bv
, bio
, i
) {
382 char *data
= bvec_kmap_irq(bv
, &flags
);
383 memset(data
, 0, bv
->bv_len
);
384 flush_dcache_page(bv
->bv_page
);
385 bvec_kunmap_irq(data
, &flags
);
388 EXPORT_SYMBOL(zero_fill_bio
);
391 * bio_put - release a reference to a bio
392 * @bio: bio to release reference to
395 * Put a reference to a &struct bio, either one you have gotten with
396 * bio_alloc or bio_get. The last put of a bio will free it.
398 void bio_put(struct bio
*bio
)
400 BIO_BUG_ON(!atomic_read(&bio
->bi_cnt
));
405 if (atomic_dec_and_test(&bio
->bi_cnt
)) {
407 bio
->bi_destructor(bio
);
411 inline int bio_phys_segments(struct request_queue
*q
, struct bio
*bio
)
413 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
414 blk_recount_segments(q
, bio
);
416 return bio
->bi_phys_segments
;
420 * __bio_clone - clone a bio
421 * @bio: destination bio
422 * @bio_src: bio to clone
424 * Clone a &bio. Caller will own the returned bio, but not
425 * the actual data it points to. Reference count of returned
428 void __bio_clone(struct bio
*bio
, struct bio
*bio_src
)
430 memcpy(bio
->bi_io_vec
, bio_src
->bi_io_vec
,
431 bio_src
->bi_max_vecs
* sizeof(struct bio_vec
));
434 * most users will be overriding ->bi_bdev with a new target,
435 * so we don't set nor calculate new physical/hw segment counts here
437 bio
->bi_sector
= bio_src
->bi_sector
;
438 bio
->bi_bdev
= bio_src
->bi_bdev
;
439 bio
->bi_flags
|= 1 << BIO_CLONED
;
440 bio
->bi_rw
= bio_src
->bi_rw
;
441 bio
->bi_vcnt
= bio_src
->bi_vcnt
;
442 bio
->bi_size
= bio_src
->bi_size
;
443 bio
->bi_idx
= bio_src
->bi_idx
;
447 * bio_clone - clone a bio
449 * @gfp_mask: allocation priority
451 * Like __bio_clone, only also allocates the returned bio
453 struct bio
*bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
455 struct bio
*b
= bio_alloc_bioset(gfp_mask
, bio
->bi_max_vecs
, fs_bio_set
);
460 b
->bi_destructor
= bio_fs_destructor
;
463 if (bio_integrity(bio
)) {
466 ret
= bio_integrity_clone(b
, bio
, gfp_mask
, fs_bio_set
);
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
= ((q
->max_sectors
<< 9) + PAGE_SIZE
- 1) >> PAGE_SHIFT
;
492 if (nr_pages
> q
->max_phys_segments
)
493 nr_pages
= q
->max_phys_segments
;
494 if (nr_pages
> q
->max_hw_segments
)
495 nr_pages
= q
->max_hw_segments
;
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
) {
528 if (q
->merge_bvec_fn
) {
529 struct bvec_merge_data bvm
= {
530 .bi_bdev
= bio
->bi_bdev
,
531 .bi_sector
= bio
->bi_sector
,
532 .bi_size
= bio
->bi_size
,
536 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < len
) {
546 if (bio
->bi_vcnt
>= bio
->bi_max_vecs
)
550 * we might lose a segment or two here, but rather that than
551 * make this too complex.
554 while (bio
->bi_phys_segments
>= q
->max_phys_segments
555 || bio
->bi_phys_segments
>= q
->max_hw_segments
) {
557 if (retried_segments
)
560 retried_segments
= 1;
561 blk_recount_segments(q
, bio
);
565 * setup the new entry, we might clear it again later if we
566 * cannot add the page
568 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
569 bvec
->bv_page
= page
;
571 bvec
->bv_offset
= offset
;
574 * if queue has other restrictions (eg varying max sector size
575 * depending on offset), it can specify a merge_bvec_fn in the
576 * queue to get further control
578 if (q
->merge_bvec_fn
) {
579 struct bvec_merge_data bvm
= {
580 .bi_bdev
= bio
->bi_bdev
,
581 .bi_sector
= bio
->bi_sector
,
582 .bi_size
= bio
->bi_size
,
587 * merge_bvec_fn() returns number of bytes it can accept
590 if (q
->merge_bvec_fn(q
, &bvm
, bvec
) < len
) {
591 bvec
->bv_page
= NULL
;
598 /* If we may be able to merge these biovecs, force a recount */
599 if (bio
->bi_vcnt
&& (BIOVEC_PHYS_MERGEABLE(bvec
-1, bvec
)))
600 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
603 bio
->bi_phys_segments
++;
610 * bio_add_pc_page - attempt to add page to bio
611 * @q: the target queue
612 * @bio: destination bio
614 * @len: vec entry length
615 * @offset: vec entry offset
617 * Attempt to add a page to the bio_vec maplist. This can fail for a
618 * number of reasons, such as the bio being full or target block
619 * device limitations. The target block device must allow bio's
620 * smaller than PAGE_SIZE, so it is always possible to add a single
621 * page to an empty bio. This should only be used by REQ_PC bios.
623 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
, struct page
*page
,
624 unsigned int len
, unsigned int offset
)
626 return __bio_add_page(q
, bio
, page
, len
, offset
, q
->max_hw_sectors
);
630 * bio_add_page - attempt to add page to bio
631 * @bio: destination bio
633 * @len: vec entry length
634 * @offset: vec entry offset
636 * Attempt to add a page to the bio_vec maplist. This can fail for a
637 * number of reasons, such as the bio being full or target block
638 * device limitations. The target block device must allow bio's
639 * smaller than PAGE_SIZE, so it is always possible to add a single
640 * page to an empty bio.
642 int bio_add_page(struct bio
*bio
, struct page
*page
, unsigned int len
,
645 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
646 return __bio_add_page(q
, bio
, page
, len
, offset
, q
->max_sectors
);
649 struct bio_map_data
{
650 struct bio_vec
*iovecs
;
651 struct sg_iovec
*sgvecs
;
656 static void bio_set_map_data(struct bio_map_data
*bmd
, struct bio
*bio
,
657 struct sg_iovec
*iov
, int iov_count
,
660 memcpy(bmd
->iovecs
, bio
->bi_io_vec
, sizeof(struct bio_vec
) * bio
->bi_vcnt
);
661 memcpy(bmd
->sgvecs
, iov
, sizeof(struct sg_iovec
) * iov_count
);
662 bmd
->nr_sgvecs
= iov_count
;
663 bmd
->is_our_pages
= is_our_pages
;
664 bio
->bi_private
= bmd
;
667 static void bio_free_map_data(struct bio_map_data
*bmd
)
674 static struct bio_map_data
*bio_alloc_map_data(int nr_segs
, int iov_count
,
677 struct bio_map_data
*bmd
= kmalloc(sizeof(*bmd
), gfp_mask
);
682 bmd
->iovecs
= kmalloc(sizeof(struct bio_vec
) * nr_segs
, gfp_mask
);
688 bmd
->sgvecs
= kmalloc(sizeof(struct sg_iovec
) * iov_count
, gfp_mask
);
697 static int __bio_copy_iov(struct bio
*bio
, struct bio_vec
*iovecs
,
698 struct sg_iovec
*iov
, int iov_count
, int uncopy
,
702 struct bio_vec
*bvec
;
704 unsigned int iov_off
= 0;
705 int read
= bio_data_dir(bio
) == READ
;
707 __bio_for_each_segment(bvec
, bio
, i
, 0) {
708 char *bv_addr
= page_address(bvec
->bv_page
);
709 unsigned int bv_len
= iovecs
[i
].bv_len
;
711 while (bv_len
&& iov_idx
< iov_count
) {
715 bytes
= min_t(unsigned int,
716 iov
[iov_idx
].iov_len
- iov_off
, bv_len
);
717 iov_addr
= iov
[iov_idx
].iov_base
+ iov_off
;
720 if (!read
&& !uncopy
)
721 ret
= copy_from_user(bv_addr
, iov_addr
,
724 ret
= copy_to_user(iov_addr
, bv_addr
,
736 if (iov
[iov_idx
].iov_len
== iov_off
) {
743 __free_page(bvec
->bv_page
);
750 * bio_uncopy_user - finish previously mapped bio
751 * @bio: bio being terminated
753 * Free pages allocated from bio_copy_user() and write back data
754 * to user space in case of a read.
756 int bio_uncopy_user(struct bio
*bio
)
758 struct bio_map_data
*bmd
= bio
->bi_private
;
761 if (!bio_flagged(bio
, BIO_NULL_MAPPED
))
762 ret
= __bio_copy_iov(bio
, bmd
->iovecs
, bmd
->sgvecs
,
763 bmd
->nr_sgvecs
, 1, bmd
->is_our_pages
);
764 bio_free_map_data(bmd
);
770 * bio_copy_user_iov - copy user data to bio
771 * @q: destination block queue
772 * @map_data: pointer to the rq_map_data holding pages (if necessary)
774 * @iov_count: number of elements in the iovec
775 * @write_to_vm: bool indicating writing to pages or not
776 * @gfp_mask: memory allocation flags
778 * Prepares and returns a bio for indirect user io, bouncing data
779 * to/from kernel pages as necessary. Must be paired with
780 * call bio_uncopy_user() on io completion.
782 struct bio
*bio_copy_user_iov(struct request_queue
*q
,
783 struct rq_map_data
*map_data
,
784 struct sg_iovec
*iov
, int iov_count
,
785 int write_to_vm
, gfp_t gfp_mask
)
787 struct bio_map_data
*bmd
;
788 struct bio_vec
*bvec
;
793 unsigned int len
= 0;
794 unsigned int offset
= map_data
? map_data
->offset
& ~PAGE_MASK
: 0;
796 for (i
= 0; i
< iov_count
; i
++) {
801 uaddr
= (unsigned long)iov
[i
].iov_base
;
802 end
= (uaddr
+ iov
[i
].iov_len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
803 start
= uaddr
>> PAGE_SHIFT
;
805 nr_pages
+= end
- start
;
806 len
+= iov
[i
].iov_len
;
809 bmd
= bio_alloc_map_data(nr_pages
, iov_count
, gfp_mask
);
811 return ERR_PTR(-ENOMEM
);
814 bio
= bio_alloc(gfp_mask
, nr_pages
);
818 bio
->bi_rw
|= (!write_to_vm
<< BIO_RW
);
823 nr_pages
= 1 << map_data
->page_order
;
824 i
= map_data
->offset
/ PAGE_SIZE
;
827 unsigned int bytes
= PAGE_SIZE
;
835 if (i
== map_data
->nr_entries
* nr_pages
) {
840 page
= map_data
->pages
[i
/ nr_pages
];
841 page
+= (i
% nr_pages
);
845 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
852 if (bio_add_pc_page(q
, bio
, page
, bytes
, offset
) < bytes
)
865 if (!write_to_vm
&& (!map_data
|| !map_data
->null_mapped
)) {
866 ret
= __bio_copy_iov(bio
, bio
->bi_io_vec
, iov
, iov_count
, 0, 0);
871 bio_set_map_data(bmd
, bio
, iov
, iov_count
, map_data
? 0 : 1);
875 bio_for_each_segment(bvec
, bio
, i
)
876 __free_page(bvec
->bv_page
);
880 bio_free_map_data(bmd
);
885 * bio_copy_user - copy user data to bio
886 * @q: destination block queue
887 * @map_data: pointer to the rq_map_data holding pages (if necessary)
888 * @uaddr: start of user address
889 * @len: length in bytes
890 * @write_to_vm: bool indicating writing to pages or not
891 * @gfp_mask: memory allocation flags
893 * Prepares and returns a bio for indirect user io, bouncing data
894 * to/from kernel pages as necessary. Must be paired with
895 * call bio_uncopy_user() on io completion.
897 struct bio
*bio_copy_user(struct request_queue
*q
, struct rq_map_data
*map_data
,
898 unsigned long uaddr
, unsigned int len
,
899 int write_to_vm
, gfp_t gfp_mask
)
903 iov
.iov_base
= (void __user
*)uaddr
;
906 return bio_copy_user_iov(q
, map_data
, &iov
, 1, write_to_vm
, gfp_mask
);
909 static struct bio
*__bio_map_user_iov(struct request_queue
*q
,
910 struct block_device
*bdev
,
911 struct sg_iovec
*iov
, int iov_count
,
912 int write_to_vm
, gfp_t gfp_mask
)
921 for (i
= 0; i
< iov_count
; i
++) {
922 unsigned long uaddr
= (unsigned long)iov
[i
].iov_base
;
923 unsigned long len
= iov
[i
].iov_len
;
924 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
925 unsigned long start
= uaddr
>> PAGE_SHIFT
;
927 nr_pages
+= end
- start
;
929 * buffer must be aligned to at least hardsector size for now
931 if (uaddr
& queue_dma_alignment(q
))
932 return ERR_PTR(-EINVAL
);
936 return ERR_PTR(-EINVAL
);
938 bio
= bio_alloc(gfp_mask
, nr_pages
);
940 return ERR_PTR(-ENOMEM
);
943 pages
= kcalloc(nr_pages
, sizeof(struct page
*), 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
;
952 const int local_nr_pages
= end
- start
;
953 const int page_limit
= cur_page
+ local_nr_pages
;
955 ret
= get_user_pages_fast(uaddr
, local_nr_pages
,
956 write_to_vm
, &pages
[cur_page
]);
957 if (ret
< local_nr_pages
) {
962 offset
= uaddr
& ~PAGE_MASK
;
963 for (j
= cur_page
; j
< page_limit
; j
++) {
964 unsigned int bytes
= PAGE_SIZE
- offset
;
975 if (bio_add_pc_page(q
, bio
, pages
[j
], bytes
, offset
) <
985 * release the pages we didn't map into the bio, if any
987 while (j
< page_limit
)
988 page_cache_release(pages
[j
++]);
994 * set data direction, and check if mapped pages need bouncing
997 bio
->bi_rw
|= (1 << BIO_RW
);
1000 bio
->bi_flags
|= (1 << BIO_USER_MAPPED
);
1004 for (i
= 0; i
< nr_pages
; i
++) {
1007 page_cache_release(pages
[i
]);
1012 return ERR_PTR(ret
);
1016 * bio_map_user - map user address into bio
1017 * @q: the struct request_queue for the bio
1018 * @bdev: destination block device
1019 * @uaddr: start of user address
1020 * @len: length in bytes
1021 * @write_to_vm: bool indicating writing to pages or not
1022 * @gfp_mask: memory allocation flags
1024 * Map the user space address into a bio suitable for io to a block
1025 * device. Returns an error pointer in case of error.
1027 struct bio
*bio_map_user(struct request_queue
*q
, struct block_device
*bdev
,
1028 unsigned long uaddr
, unsigned int len
, int write_to_vm
,
1031 struct sg_iovec iov
;
1033 iov
.iov_base
= (void __user
*)uaddr
;
1036 return bio_map_user_iov(q
, bdev
, &iov
, 1, write_to_vm
, gfp_mask
);
1040 * bio_map_user_iov - map user sg_iovec table into bio
1041 * @q: the struct request_queue for the bio
1042 * @bdev: destination block device
1044 * @iov_count: number of elements in the iovec
1045 * @write_to_vm: bool indicating writing to pages or not
1046 * @gfp_mask: memory allocation flags
1048 * Map the user space address into a bio suitable for io to a block
1049 * device. Returns an error pointer in case of error.
1051 struct bio
*bio_map_user_iov(struct request_queue
*q
, struct block_device
*bdev
,
1052 struct sg_iovec
*iov
, int iov_count
,
1053 int write_to_vm
, gfp_t gfp_mask
)
1057 bio
= __bio_map_user_iov(q
, bdev
, iov
, iov_count
, write_to_vm
,
1063 * subtle -- if __bio_map_user() ended up bouncing a bio,
1064 * it would normally disappear when its bi_end_io is run.
1065 * however, we need it for the unmap, so grab an extra
1073 static void __bio_unmap_user(struct bio
*bio
)
1075 struct bio_vec
*bvec
;
1079 * make sure we dirty pages we wrote to
1081 __bio_for_each_segment(bvec
, bio
, i
, 0) {
1082 if (bio_data_dir(bio
) == READ
)
1083 set_page_dirty_lock(bvec
->bv_page
);
1085 page_cache_release(bvec
->bv_page
);
1092 * bio_unmap_user - unmap a bio
1093 * @bio: the bio being unmapped
1095 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1096 * a process context.
1098 * bio_unmap_user() may sleep.
1100 void bio_unmap_user(struct bio
*bio
)
1102 __bio_unmap_user(bio
);
1106 static void bio_map_kern_endio(struct bio
*bio
, int err
)
1112 static struct bio
*__bio_map_kern(struct request_queue
*q
, void *data
,
1113 unsigned int len
, gfp_t gfp_mask
)
1115 unsigned long kaddr
= (unsigned long)data
;
1116 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1117 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1118 const int nr_pages
= end
- start
;
1122 bio
= bio_alloc(gfp_mask
, nr_pages
);
1124 return ERR_PTR(-ENOMEM
);
1126 offset
= offset_in_page(kaddr
);
1127 for (i
= 0; i
< nr_pages
; i
++) {
1128 unsigned int bytes
= PAGE_SIZE
- offset
;
1136 if (bio_add_pc_page(q
, bio
, virt_to_page(data
), bytes
,
1145 bio
->bi_end_io
= bio_map_kern_endio
;
1150 * bio_map_kern - map kernel address into bio
1151 * @q: the struct request_queue for the bio
1152 * @data: pointer to buffer to map
1153 * @len: length in bytes
1154 * @gfp_mask: allocation flags for bio allocation
1156 * Map the kernel address into a bio suitable for io to a block
1157 * device. Returns an error pointer in case of error.
1159 struct bio
*bio_map_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1164 bio
= __bio_map_kern(q
, data
, len
, gfp_mask
);
1168 if (bio
->bi_size
== len
)
1172 * Don't support partial mappings.
1175 return ERR_PTR(-EINVAL
);
1178 static void bio_copy_kern_endio(struct bio
*bio
, int err
)
1180 struct bio_vec
*bvec
;
1181 const int read
= bio_data_dir(bio
) == READ
;
1182 struct bio_map_data
*bmd
= bio
->bi_private
;
1184 char *p
= bmd
->sgvecs
[0].iov_base
;
1186 __bio_for_each_segment(bvec
, bio
, i
, 0) {
1187 char *addr
= page_address(bvec
->bv_page
);
1188 int len
= bmd
->iovecs
[i
].bv_len
;
1191 memcpy(p
, addr
, len
);
1193 __free_page(bvec
->bv_page
);
1197 bio_free_map_data(bmd
);
1202 * bio_copy_kern - copy kernel address into bio
1203 * @q: the struct request_queue for the bio
1204 * @data: pointer to buffer to copy
1205 * @len: length in bytes
1206 * @gfp_mask: allocation flags for bio and page allocation
1207 * @reading: data direction is READ
1209 * copy the kernel address into a bio suitable for io to a block
1210 * device. Returns an error pointer in case of error.
1212 struct bio
*bio_copy_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1213 gfp_t gfp_mask
, int reading
)
1216 struct bio_vec
*bvec
;
1219 bio
= bio_copy_user(q
, NULL
, (unsigned long)data
, len
, 1, gfp_mask
);
1226 bio_for_each_segment(bvec
, bio
, i
) {
1227 char *addr
= page_address(bvec
->bv_page
);
1229 memcpy(addr
, p
, bvec
->bv_len
);
1234 bio
->bi_end_io
= bio_copy_kern_endio
;
1240 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1241 * for performing direct-IO in BIOs.
1243 * The problem is that we cannot run set_page_dirty() from interrupt context
1244 * because the required locks are not interrupt-safe. So what we can do is to
1245 * mark the pages dirty _before_ performing IO. And in interrupt context,
1246 * check that the pages are still dirty. If so, fine. If not, redirty them
1247 * in process context.
1249 * We special-case compound pages here: normally this means reads into hugetlb
1250 * pages. The logic in here doesn't really work right for compound pages
1251 * because the VM does not uniformly chase down the head page in all cases.
1252 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1253 * handle them at all. So we skip compound pages here at an early stage.
1255 * Note that this code is very hard to test under normal circumstances because
1256 * direct-io pins the pages with get_user_pages(). This makes
1257 * is_page_cache_freeable return false, and the VM will not clean the pages.
1258 * But other code (eg, pdflush) could clean the pages if they are mapped
1261 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1262 * deferred bio dirtying paths.
1266 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1268 void bio_set_pages_dirty(struct bio
*bio
)
1270 struct bio_vec
*bvec
= bio
->bi_io_vec
;
1273 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
1274 struct page
*page
= bvec
[i
].bv_page
;
1276 if (page
&& !PageCompound(page
))
1277 set_page_dirty_lock(page
);
1281 static void bio_release_pages(struct bio
*bio
)
1283 struct bio_vec
*bvec
= bio
->bi_io_vec
;
1286 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
1287 struct page
*page
= bvec
[i
].bv_page
;
1295 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1296 * If they are, then fine. If, however, some pages are clean then they must
1297 * have been written out during the direct-IO read. So we take another ref on
1298 * the BIO and the offending pages and re-dirty the pages in process context.
1300 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1301 * here on. It will run one page_cache_release() against each page and will
1302 * run one bio_put() against the BIO.
1305 static void bio_dirty_fn(struct work_struct
*work
);
1307 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
1308 static DEFINE_SPINLOCK(bio_dirty_lock
);
1309 static struct bio
*bio_dirty_list
;
1312 * This runs in process context
1314 static void bio_dirty_fn(struct work_struct
*work
)
1316 unsigned long flags
;
1319 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1320 bio
= bio_dirty_list
;
1321 bio_dirty_list
= NULL
;
1322 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1325 struct bio
*next
= bio
->bi_private
;
1327 bio_set_pages_dirty(bio
);
1328 bio_release_pages(bio
);
1334 void bio_check_pages_dirty(struct bio
*bio
)
1336 struct bio_vec
*bvec
= bio
->bi_io_vec
;
1337 int nr_clean_pages
= 0;
1340 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
1341 struct page
*page
= bvec
[i
].bv_page
;
1343 if (PageDirty(page
) || PageCompound(page
)) {
1344 page_cache_release(page
);
1345 bvec
[i
].bv_page
= NULL
;
1351 if (nr_clean_pages
) {
1352 unsigned long flags
;
1354 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1355 bio
->bi_private
= bio_dirty_list
;
1356 bio_dirty_list
= bio
;
1357 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1358 schedule_work(&bio_dirty_work
);
1365 * bio_endio - end I/O on a bio
1367 * @error: error, if any
1370 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1371 * preferred way to end I/O on a bio, it takes care of clearing
1372 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1373 * established -Exxxx (-EIO, for instance) error values in case
1374 * something went wrong. Noone should call bi_end_io() directly on a
1375 * bio unless they own it and thus know that it has an end_io
1378 void bio_endio(struct bio
*bio
, int error
)
1381 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1382 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
1386 bio
->bi_end_io(bio
, error
);
1389 void bio_pair_release(struct bio_pair
*bp
)
1391 if (atomic_dec_and_test(&bp
->cnt
)) {
1392 struct bio
*master
= bp
->bio1
.bi_private
;
1394 bio_endio(master
, bp
->error
);
1395 mempool_free(bp
, bp
->bio2
.bi_private
);
1399 static void bio_pair_end_1(struct bio
*bi
, int err
)
1401 struct bio_pair
*bp
= container_of(bi
, struct bio_pair
, bio1
);
1406 bio_pair_release(bp
);
1409 static void bio_pair_end_2(struct bio
*bi
, int err
)
1411 struct bio_pair
*bp
= container_of(bi
, struct bio_pair
, bio2
);
1416 bio_pair_release(bp
);
1420 * split a bio - only worry about a bio with a single page
1423 struct bio_pair
*bio_split(struct bio
*bi
, int first_sectors
)
1425 struct bio_pair
*bp
= mempool_alloc(bio_split_pool
, GFP_NOIO
);
1430 trace_block_split(bdev_get_queue(bi
->bi_bdev
), bi
,
1431 bi
->bi_sector
+ first_sectors
);
1433 BUG_ON(bi
->bi_vcnt
!= 1);
1434 BUG_ON(bi
->bi_idx
!= 0);
1435 atomic_set(&bp
->cnt
, 3);
1439 bp
->bio2
.bi_sector
+= first_sectors
;
1440 bp
->bio2
.bi_size
-= first_sectors
<< 9;
1441 bp
->bio1
.bi_size
= first_sectors
<< 9;
1443 bp
->bv1
= bi
->bi_io_vec
[0];
1444 bp
->bv2
= bi
->bi_io_vec
[0];
1445 bp
->bv2
.bv_offset
+= first_sectors
<< 9;
1446 bp
->bv2
.bv_len
-= first_sectors
<< 9;
1447 bp
->bv1
.bv_len
= first_sectors
<< 9;
1449 bp
->bio1
.bi_io_vec
= &bp
->bv1
;
1450 bp
->bio2
.bi_io_vec
= &bp
->bv2
;
1452 bp
->bio1
.bi_max_vecs
= 1;
1453 bp
->bio2
.bi_max_vecs
= 1;
1455 bp
->bio1
.bi_end_io
= bio_pair_end_1
;
1456 bp
->bio2
.bi_end_io
= bio_pair_end_2
;
1458 bp
->bio1
.bi_private
= bi
;
1459 bp
->bio2
.bi_private
= bio_split_pool
;
1461 if (bio_integrity(bi
))
1462 bio_integrity_split(bi
, bp
, first_sectors
);
1468 * bio_sector_offset - Find hardware sector offset in bio
1469 * @bio: bio to inspect
1470 * @index: bio_vec index
1471 * @offset: offset in bv_page
1473 * Return the number of hardware sectors between beginning of bio
1474 * and an end point indicated by a bio_vec index and an offset
1475 * within that vector's page.
1477 sector_t
bio_sector_offset(struct bio
*bio
, unsigned short index
,
1478 unsigned int offset
)
1480 unsigned int sector_sz
= queue_hardsect_size(bio
->bi_bdev
->bd_disk
->queue
);
1487 if (index
>= bio
->bi_idx
)
1488 index
= bio
->bi_vcnt
- 1;
1490 __bio_for_each_segment(bv
, bio
, i
, 0) {
1492 if (offset
> bv
->bv_offset
)
1493 sectors
+= (offset
- bv
->bv_offset
) / sector_sz
;
1497 sectors
+= bv
->bv_len
/ sector_sz
;
1502 EXPORT_SYMBOL(bio_sector_offset
);
1505 * create memory pools for biovec's in a bio_set.
1506 * use the global biovec slabs created for general use.
1508 static int biovec_create_pools(struct bio_set
*bs
, int pool_entries
)
1510 struct biovec_slab
*bp
= bvec_slabs
+ BIOVEC_MAX_IDX
;
1512 bs
->bvec_pool
= mempool_create_slab_pool(pool_entries
, bp
->slab
);
1519 static void biovec_free_pools(struct bio_set
*bs
)
1521 mempool_destroy(bs
->bvec_pool
);
1524 void bioset_free(struct bio_set
*bs
)
1527 mempool_destroy(bs
->bio_pool
);
1529 bioset_integrity_free(bs
);
1530 biovec_free_pools(bs
);
1537 * bioset_create - Create a bio_set
1538 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1539 * @front_pad: Number of bytes to allocate in front of the returned bio
1542 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1543 * to ask for a number of bytes to be allocated in front of the bio.
1544 * Front pad allocation is useful for embedding the bio inside
1545 * another structure, to avoid allocating extra data to go with the bio.
1546 * Note that the bio must be embedded at the END of that structure always,
1547 * or things will break badly.
1549 struct bio_set
*bioset_create(unsigned int pool_size
, unsigned int front_pad
)
1551 unsigned int back_pad
= BIO_INLINE_VECS
* sizeof(struct bio_vec
);
1554 bs
= kzalloc(sizeof(*bs
), GFP_KERNEL
);
1558 bs
->front_pad
= front_pad
;
1560 bs
->bio_slab
= bio_find_or_create_slab(front_pad
+ back_pad
);
1561 if (!bs
->bio_slab
) {
1566 bs
->bio_pool
= mempool_create_slab_pool(pool_size
, bs
->bio_slab
);
1570 if (bioset_integrity_create(bs
, pool_size
))
1573 if (!biovec_create_pools(bs
, pool_size
))
1581 static void __init
biovec_init_slabs(void)
1585 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
1587 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
1589 size
= bvs
->nr_vecs
* sizeof(struct bio_vec
);
1590 bvs
->slab
= kmem_cache_create(bvs
->name
, size
, 0,
1591 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
1595 static int __init
init_bio(void)
1599 bio_slabs
= kzalloc(bio_slab_max
* sizeof(struct bio_slab
), GFP_KERNEL
);
1601 panic("bio: can't allocate bios\n");
1603 bio_integrity_init_slab();
1604 biovec_init_slabs();
1606 fs_bio_set
= bioset_create(BIO_POOL_SIZE
, 0);
1608 panic("bio: can't allocate bios\n");
1610 bio_split_pool
= mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES
,
1611 sizeof(struct bio_pair
));
1612 if (!bio_split_pool
)
1613 panic("bio: can't create split pool\n");
1618 subsys_initcall(init_bio
);
1620 EXPORT_SYMBOL(bio_alloc
);
1621 EXPORT_SYMBOL(bio_kmalloc
);
1622 EXPORT_SYMBOL(bio_put
);
1623 EXPORT_SYMBOL(bio_free
);
1624 EXPORT_SYMBOL(bio_endio
);
1625 EXPORT_SYMBOL(bio_init
);
1626 EXPORT_SYMBOL(__bio_clone
);
1627 EXPORT_SYMBOL(bio_clone
);
1628 EXPORT_SYMBOL(bio_phys_segments
);
1629 EXPORT_SYMBOL(bio_add_page
);
1630 EXPORT_SYMBOL(bio_add_pc_page
);
1631 EXPORT_SYMBOL(bio_get_nr_vecs
);
1632 EXPORT_SYMBOL(bio_map_user
);
1633 EXPORT_SYMBOL(bio_unmap_user
);
1634 EXPORT_SYMBOL(bio_map_kern
);
1635 EXPORT_SYMBOL(bio_copy_kern
);
1636 EXPORT_SYMBOL(bio_pair_release
);
1637 EXPORT_SYMBOL(bio_split
);
1638 EXPORT_SYMBOL(bio_copy_user
);
1639 EXPORT_SYMBOL(bio_uncopy_user
);
1640 EXPORT_SYMBOL(bioset_create
);
1641 EXPORT_SYMBOL(bioset_free
);
1642 EXPORT_SYMBOL(bio_alloc_bioset
);