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/export.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
, *new_bio_slabs
;
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 new_bio_slabs
= krealloc(bio_slabs
,
101 bio_slab_max
* sizeof(struct bio_slab
),
105 bio_slabs
= new_bio_slabs
;
108 entry
= bio_slab_nr
++;
110 bslab
= &bio_slabs
[entry
];
112 snprintf(bslab
->name
, sizeof(bslab
->name
), "bio-%d", entry
);
113 slab
= kmem_cache_create(bslab
->name
, sz
, 0, SLAB_HWCACHE_ALIGN
, NULL
);
117 printk(KERN_INFO
"bio: create slab <%s> at %d\n", bslab
->name
, entry
);
120 bslab
->slab_size
= sz
;
122 mutex_unlock(&bio_slab_lock
);
126 static void bio_put_slab(struct bio_set
*bs
)
128 struct bio_slab
*bslab
= NULL
;
131 mutex_lock(&bio_slab_lock
);
133 for (i
= 0; i
< bio_slab_nr
; i
++) {
134 if (bs
->bio_slab
== bio_slabs
[i
].slab
) {
135 bslab
= &bio_slabs
[i
];
140 if (WARN(!bslab
, KERN_ERR
"bio: unable to find slab!\n"))
143 WARN_ON(!bslab
->slab_ref
);
145 if (--bslab
->slab_ref
)
148 kmem_cache_destroy(bslab
->slab
);
152 mutex_unlock(&bio_slab_lock
);
155 unsigned int bvec_nr_vecs(unsigned short idx
)
157 return bvec_slabs
[idx
].nr_vecs
;
160 void bvec_free_bs(struct bio_set
*bs
, struct bio_vec
*bv
, unsigned int idx
)
162 BIO_BUG_ON(idx
>= BIOVEC_NR_POOLS
);
164 if (idx
== BIOVEC_MAX_IDX
)
165 mempool_free(bv
, bs
->bvec_pool
);
167 struct biovec_slab
*bvs
= bvec_slabs
+ idx
;
169 kmem_cache_free(bvs
->slab
, bv
);
173 struct bio_vec
*bvec_alloc_bs(gfp_t gfp_mask
, int nr
, unsigned long *idx
,
179 * see comment near bvec_array define!
197 case 129 ... BIO_MAX_PAGES
:
205 * idx now points to the pool we want to allocate from. only the
206 * 1-vec entry pool is mempool backed.
208 if (*idx
== BIOVEC_MAX_IDX
) {
210 bvl
= mempool_alloc(bs
->bvec_pool
, gfp_mask
);
212 struct biovec_slab
*bvs
= bvec_slabs
+ *idx
;
213 gfp_t __gfp_mask
= gfp_mask
& ~(__GFP_WAIT
| __GFP_IO
);
216 * Make this allocation restricted and don't dump info on
217 * allocation failures, since we'll fallback to the mempool
218 * in case of failure.
220 __gfp_mask
|= __GFP_NOMEMALLOC
| __GFP_NORETRY
| __GFP_NOWARN
;
223 * Try a slab allocation. If this fails and __GFP_WAIT
224 * is set, retry with the 1-entry mempool
226 bvl
= kmem_cache_alloc(bvs
->slab
, __gfp_mask
);
227 if (unlikely(!bvl
&& (gfp_mask
& __GFP_WAIT
))) {
228 *idx
= BIOVEC_MAX_IDX
;
236 void bio_free(struct bio
*bio
, struct bio_set
*bs
)
240 if (bio_has_allocated_vec(bio
))
241 bvec_free_bs(bs
, bio
->bi_io_vec
, BIO_POOL_IDX(bio
));
243 if (bio_integrity(bio
))
244 bio_integrity_free(bio
);
247 * If we have front padding, adjust the bio pointer before freeing
253 mempool_free(p
, bs
->bio_pool
);
255 EXPORT_SYMBOL(bio_free
);
257 void bio_init(struct bio
*bio
)
259 memset(bio
, 0, sizeof(*bio
));
260 bio
->bi_flags
= 1 << BIO_UPTODATE
;
261 atomic_set(&bio
->bi_cnt
, 1);
263 EXPORT_SYMBOL(bio_init
);
266 * bio_reset - reinitialize a bio
270 * After calling bio_reset(), @bio will be in the same state as a freshly
271 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
272 * preserved are the ones that are initialized by bio_alloc_bioset(). See
273 * comment in struct bio.
275 void bio_reset(struct bio
*bio
)
277 unsigned long flags
= bio
->bi_flags
& (~0UL << BIO_RESET_BITS
);
279 if (bio_integrity(bio
))
280 bio_integrity_free(bio
);
282 bio_disassociate_task(bio
);
284 memset(bio
, 0, BIO_RESET_BYTES
);
285 bio
->bi_flags
= flags
|(1 << BIO_UPTODATE
);
287 EXPORT_SYMBOL(bio_reset
);
290 * bio_alloc_bioset - allocate a bio for I/O
291 * @gfp_mask: the GFP_ mask given to the slab allocator
292 * @nr_iovecs: number of iovecs to pre-allocate
293 * @bs: the bio_set to allocate from.
296 * bio_alloc_bioset will try its own mempool to satisfy the allocation.
297 * If %__GFP_WAIT is set then we will block on the internal pool waiting
298 * for a &struct bio to become free.
300 struct bio
*bio_alloc_bioset(gfp_t gfp_mask
, int nr_iovecs
, struct bio_set
*bs
)
302 unsigned long idx
= BIO_POOL_NONE
;
303 struct bio_vec
*bvl
= NULL
;
307 p
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
310 bio
= p
+ bs
->front_pad
;
315 if (unlikely(!nr_iovecs
))
318 if (nr_iovecs
<= BIO_INLINE_VECS
) {
319 bvl
= bio
->bi_inline_vecs
;
320 nr_iovecs
= BIO_INLINE_VECS
;
322 bvl
= bvec_alloc_bs(gfp_mask
, nr_iovecs
, &idx
, bs
);
326 nr_iovecs
= bvec_nr_vecs(idx
);
329 bio
->bi_flags
|= idx
<< BIO_POOL_OFFSET
;
330 bio
->bi_max_vecs
= nr_iovecs
;
331 bio
->bi_io_vec
= bvl
;
335 mempool_free(p
, bs
->bio_pool
);
338 EXPORT_SYMBOL(bio_alloc_bioset
);
341 * bio_alloc - allocate a new bio, memory pool backed
342 * @gfp_mask: allocation mask to use
343 * @nr_iovecs: number of iovecs
345 * bio_alloc will allocate a bio and associated bio_vec array that can hold
346 * at least @nr_iovecs entries. Allocations will be done from the
347 * fs_bio_set. Also see @bio_alloc_bioset and @bio_kmalloc.
349 * If %__GFP_WAIT is set, then bio_alloc will always be able to allocate
350 * a bio. This is due to the mempool guarantees. To make this work, callers
351 * must never allocate more than 1 bio at a time from this pool. Callers
352 * that need to allocate more than 1 bio must always submit the previously
353 * allocated bio for IO before attempting to allocate a new one. Failure to
354 * do so can cause livelocks under memory pressure.
357 * Pointer to new bio on success, NULL on failure.
359 struct bio
*bio_alloc(gfp_t gfp_mask
, unsigned int nr_iovecs
)
361 return bio_alloc_bioset(gfp_mask
, nr_iovecs
, fs_bio_set
);
363 EXPORT_SYMBOL(bio_alloc
);
365 static void bio_kmalloc_destructor(struct bio
*bio
)
367 if (bio_integrity(bio
))
368 bio_integrity_free(bio
);
373 * bio_kmalloc - allocate a bio for I/O using kmalloc()
374 * @gfp_mask: the GFP_ mask given to the slab allocator
375 * @nr_iovecs: number of iovecs to pre-allocate
378 * Allocate a new bio with @nr_iovecs bvecs. If @gfp_mask contains
379 * %__GFP_WAIT, the allocation is guaranteed to succeed.
382 struct bio
*bio_kmalloc(gfp_t gfp_mask
, unsigned int nr_iovecs
)
386 if (nr_iovecs
> UIO_MAXIOV
)
389 bio
= kmalloc(sizeof(struct bio
) + nr_iovecs
* sizeof(struct bio_vec
),
395 bio
->bi_flags
|= BIO_POOL_NONE
<< BIO_POOL_OFFSET
;
396 bio
->bi_max_vecs
= nr_iovecs
;
397 bio
->bi_io_vec
= bio
->bi_inline_vecs
;
398 bio
->bi_destructor
= bio_kmalloc_destructor
;
402 EXPORT_SYMBOL(bio_kmalloc
);
404 void zero_fill_bio(struct bio
*bio
)
410 bio_for_each_segment(bv
, bio
, i
) {
411 char *data
= bvec_kmap_irq(bv
, &flags
);
412 memset(data
, 0, bv
->bv_len
);
413 flush_dcache_page(bv
->bv_page
);
414 bvec_kunmap_irq(data
, &flags
);
417 EXPORT_SYMBOL(zero_fill_bio
);
420 * bio_put - release a reference to a bio
421 * @bio: bio to release reference to
424 * Put a reference to a &struct bio, either one you have gotten with
425 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
427 void bio_put(struct bio
*bio
)
429 BIO_BUG_ON(!atomic_read(&bio
->bi_cnt
));
434 if (atomic_dec_and_test(&bio
->bi_cnt
)) {
435 bio_disassociate_task(bio
);
439 * This if statement is temporary - bi_pool is replacing
440 * bi_destructor, but bi_destructor will be taken out in another
444 bio_free(bio
, bio
->bi_pool
);
446 bio
->bi_destructor(bio
);
449 EXPORT_SYMBOL(bio_put
);
451 inline int bio_phys_segments(struct request_queue
*q
, struct bio
*bio
)
453 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
454 blk_recount_segments(q
, bio
);
456 return bio
->bi_phys_segments
;
458 EXPORT_SYMBOL(bio_phys_segments
);
461 * __bio_clone - clone a bio
462 * @bio: destination bio
463 * @bio_src: bio to clone
465 * Clone a &bio. Caller will own the returned bio, but not
466 * the actual data it points to. Reference count of returned
469 void __bio_clone(struct bio
*bio
, struct bio
*bio_src
)
471 memcpy(bio
->bi_io_vec
, bio_src
->bi_io_vec
,
472 bio_src
->bi_max_vecs
* sizeof(struct bio_vec
));
475 * most users will be overriding ->bi_bdev with a new target,
476 * so we don't set nor calculate new physical/hw segment counts here
478 bio
->bi_sector
= bio_src
->bi_sector
;
479 bio
->bi_bdev
= bio_src
->bi_bdev
;
480 bio
->bi_flags
|= 1 << BIO_CLONED
;
481 bio
->bi_rw
= bio_src
->bi_rw
;
482 bio
->bi_vcnt
= bio_src
->bi_vcnt
;
483 bio
->bi_size
= bio_src
->bi_size
;
484 bio
->bi_idx
= bio_src
->bi_idx
;
486 EXPORT_SYMBOL(__bio_clone
);
489 * bio_clone - clone a bio
491 * @gfp_mask: allocation priority
493 * Like __bio_clone, only also allocates the returned bio
495 struct bio
*bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
497 struct bio
*b
= bio_alloc(gfp_mask
, bio
->bi_max_vecs
);
504 if (bio_integrity(bio
)) {
507 ret
= bio_integrity_clone(b
, bio
, gfp_mask
);
517 EXPORT_SYMBOL(bio_clone
);
520 * bio_get_nr_vecs - return approx number of vecs
523 * Return the approximate number of pages we can send to this target.
524 * There's no guarantee that you will be able to fit this number of pages
525 * into a bio, it does not account for dynamic restrictions that vary
528 int bio_get_nr_vecs(struct block_device
*bdev
)
530 struct request_queue
*q
= bdev_get_queue(bdev
);
533 nr_pages
= min_t(unsigned,
534 queue_max_segments(q
),
535 queue_max_sectors(q
) / (PAGE_SIZE
>> 9) + 1);
537 return min_t(unsigned, nr_pages
, BIO_MAX_PAGES
);
540 EXPORT_SYMBOL(bio_get_nr_vecs
);
542 static int __bio_add_page(struct request_queue
*q
, struct bio
*bio
, struct page
543 *page
, unsigned int len
, unsigned int offset
,
544 unsigned short max_sectors
)
546 int retried_segments
= 0;
547 struct bio_vec
*bvec
;
550 * cloned bio must not modify vec list
552 if (unlikely(bio_flagged(bio
, BIO_CLONED
)))
555 if (((bio
->bi_size
+ len
) >> 9) > max_sectors
)
559 * For filesystems with a blocksize smaller than the pagesize
560 * we will often be called with the same page as last time and
561 * a consecutive offset. Optimize this special case.
563 if (bio
->bi_vcnt
> 0) {
564 struct bio_vec
*prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
566 if (page
== prev
->bv_page
&&
567 offset
== prev
->bv_offset
+ prev
->bv_len
) {
568 unsigned int prev_bv_len
= prev
->bv_len
;
571 if (q
->merge_bvec_fn
) {
572 struct bvec_merge_data bvm
= {
573 /* prev_bvec is already charged in
574 bi_size, discharge it in order to
575 simulate merging updated prev_bvec
577 .bi_bdev
= bio
->bi_bdev
,
578 .bi_sector
= bio
->bi_sector
,
579 .bi_size
= bio
->bi_size
- prev_bv_len
,
583 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < prev
->bv_len
) {
593 if (bio
->bi_vcnt
>= bio
->bi_max_vecs
)
597 * we might lose a segment or two here, but rather that than
598 * make this too complex.
601 while (bio
->bi_phys_segments
>= queue_max_segments(q
)) {
603 if (retried_segments
)
606 retried_segments
= 1;
607 blk_recount_segments(q
, bio
);
611 * setup the new entry, we might clear it again later if we
612 * cannot add the page
614 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
615 bvec
->bv_page
= page
;
617 bvec
->bv_offset
= offset
;
620 * if queue has other restrictions (eg varying max sector size
621 * depending on offset), it can specify a merge_bvec_fn in the
622 * queue to get further control
624 if (q
->merge_bvec_fn
) {
625 struct bvec_merge_data bvm
= {
626 .bi_bdev
= bio
->bi_bdev
,
627 .bi_sector
= bio
->bi_sector
,
628 .bi_size
= bio
->bi_size
,
633 * merge_bvec_fn() returns number of bytes it can accept
636 if (q
->merge_bvec_fn(q
, &bvm
, bvec
) < bvec
->bv_len
) {
637 bvec
->bv_page
= NULL
;
644 /* If we may be able to merge these biovecs, force a recount */
645 if (bio
->bi_vcnt
&& (BIOVEC_PHYS_MERGEABLE(bvec
-1, bvec
)))
646 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
649 bio
->bi_phys_segments
++;
656 * bio_add_pc_page - attempt to add page to bio
657 * @q: the target queue
658 * @bio: destination bio
660 * @len: vec entry length
661 * @offset: vec entry offset
663 * Attempt to add a page to the bio_vec maplist. This can fail for a
664 * number of reasons, such as the bio being full or target block device
665 * limitations. The target block device must allow bio's up to PAGE_SIZE,
666 * so it is always possible to add a single page to an empty bio.
668 * This should only be used by REQ_PC bios.
670 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
, struct page
*page
,
671 unsigned int len
, unsigned int offset
)
673 return __bio_add_page(q
, bio
, page
, len
, offset
,
674 queue_max_hw_sectors(q
));
676 EXPORT_SYMBOL(bio_add_pc_page
);
679 * bio_add_page - attempt to add page to bio
680 * @bio: destination bio
682 * @len: vec entry length
683 * @offset: vec entry offset
685 * Attempt to add a page to the bio_vec maplist. This can fail for a
686 * number of reasons, such as the bio being full or target block device
687 * limitations. The target block device must allow bio's up to PAGE_SIZE,
688 * so it is always possible to add a single page to an empty bio.
690 int bio_add_page(struct bio
*bio
, struct page
*page
, unsigned int len
,
693 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
694 return __bio_add_page(q
, bio
, page
, len
, offset
, queue_max_sectors(q
));
696 EXPORT_SYMBOL(bio_add_page
);
698 struct bio_map_data
{
699 struct bio_vec
*iovecs
;
700 struct sg_iovec
*sgvecs
;
705 static void bio_set_map_data(struct bio_map_data
*bmd
, struct bio
*bio
,
706 struct sg_iovec
*iov
, int iov_count
,
709 memcpy(bmd
->iovecs
, bio
->bi_io_vec
, sizeof(struct bio_vec
) * bio
->bi_vcnt
);
710 memcpy(bmd
->sgvecs
, iov
, sizeof(struct sg_iovec
) * iov_count
);
711 bmd
->nr_sgvecs
= iov_count
;
712 bmd
->is_our_pages
= is_our_pages
;
713 bio
->bi_private
= bmd
;
716 static void bio_free_map_data(struct bio_map_data
*bmd
)
723 static struct bio_map_data
*bio_alloc_map_data(int nr_segs
,
724 unsigned int iov_count
,
727 struct bio_map_data
*bmd
;
729 if (iov_count
> UIO_MAXIOV
)
732 bmd
= kmalloc(sizeof(*bmd
), gfp_mask
);
736 bmd
->iovecs
= kmalloc(sizeof(struct bio_vec
) * nr_segs
, gfp_mask
);
742 bmd
->sgvecs
= kmalloc(sizeof(struct sg_iovec
) * iov_count
, gfp_mask
);
751 static int __bio_copy_iov(struct bio
*bio
, struct bio_vec
*iovecs
,
752 struct sg_iovec
*iov
, int iov_count
,
753 int to_user
, int from_user
, int do_free_page
)
756 struct bio_vec
*bvec
;
758 unsigned int iov_off
= 0;
760 __bio_for_each_segment(bvec
, bio
, i
, 0) {
761 char *bv_addr
= page_address(bvec
->bv_page
);
762 unsigned int bv_len
= iovecs
[i
].bv_len
;
764 while (bv_len
&& iov_idx
< iov_count
) {
766 char __user
*iov_addr
;
768 bytes
= min_t(unsigned int,
769 iov
[iov_idx
].iov_len
- iov_off
, bv_len
);
770 iov_addr
= iov
[iov_idx
].iov_base
+ iov_off
;
774 ret
= copy_to_user(iov_addr
, bv_addr
,
778 ret
= copy_from_user(bv_addr
, iov_addr
,
790 if (iov
[iov_idx
].iov_len
== iov_off
) {
797 __free_page(bvec
->bv_page
);
804 * bio_uncopy_user - finish previously mapped bio
805 * @bio: bio being terminated
807 * Free pages allocated from bio_copy_user() and write back data
808 * to user space in case of a read.
810 int bio_uncopy_user(struct bio
*bio
)
812 struct bio_map_data
*bmd
= bio
->bi_private
;
815 if (!bio_flagged(bio
, BIO_NULL_MAPPED
))
816 ret
= __bio_copy_iov(bio
, bmd
->iovecs
, bmd
->sgvecs
,
817 bmd
->nr_sgvecs
, bio_data_dir(bio
) == READ
,
818 0, bmd
->is_our_pages
);
819 bio_free_map_data(bmd
);
823 EXPORT_SYMBOL(bio_uncopy_user
);
826 * bio_copy_user_iov - copy user data to bio
827 * @q: destination block queue
828 * @map_data: pointer to the rq_map_data holding pages (if necessary)
830 * @iov_count: number of elements in the iovec
831 * @write_to_vm: bool indicating writing to pages or not
832 * @gfp_mask: memory allocation flags
834 * Prepares and returns a bio for indirect user io, bouncing data
835 * to/from kernel pages as necessary. Must be paired with
836 * call bio_uncopy_user() on io completion.
838 struct bio
*bio_copy_user_iov(struct request_queue
*q
,
839 struct rq_map_data
*map_data
,
840 struct sg_iovec
*iov
, int iov_count
,
841 int write_to_vm
, gfp_t gfp_mask
)
843 struct bio_map_data
*bmd
;
844 struct bio_vec
*bvec
;
849 unsigned int len
= 0;
850 unsigned int offset
= map_data
? map_data
->offset
& ~PAGE_MASK
: 0;
852 for (i
= 0; i
< iov_count
; i
++) {
857 uaddr
= (unsigned long)iov
[i
].iov_base
;
858 end
= (uaddr
+ iov
[i
].iov_len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
859 start
= uaddr
>> PAGE_SHIFT
;
865 return ERR_PTR(-EINVAL
);
867 nr_pages
+= end
- start
;
868 len
+= iov
[i
].iov_len
;
874 bmd
= bio_alloc_map_data(nr_pages
, iov_count
, gfp_mask
);
876 return ERR_PTR(-ENOMEM
);
879 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
884 bio
->bi_rw
|= REQ_WRITE
;
889 nr_pages
= 1 << map_data
->page_order
;
890 i
= map_data
->offset
/ PAGE_SIZE
;
893 unsigned int bytes
= PAGE_SIZE
;
901 if (i
== map_data
->nr_entries
* nr_pages
) {
906 page
= map_data
->pages
[i
/ nr_pages
];
907 page
+= (i
% nr_pages
);
911 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
918 if (bio_add_pc_page(q
, bio
, page
, bytes
, offset
) < bytes
)
931 if ((!write_to_vm
&& (!map_data
|| !map_data
->null_mapped
)) ||
932 (map_data
&& map_data
->from_user
)) {
933 ret
= __bio_copy_iov(bio
, bio
->bi_io_vec
, iov
, iov_count
, 0, 1, 0);
938 bio_set_map_data(bmd
, bio
, iov
, iov_count
, map_data
? 0 : 1);
942 bio_for_each_segment(bvec
, bio
, i
)
943 __free_page(bvec
->bv_page
);
947 bio_free_map_data(bmd
);
952 * bio_copy_user - copy user data to bio
953 * @q: destination block queue
954 * @map_data: pointer to the rq_map_data holding pages (if necessary)
955 * @uaddr: start of user address
956 * @len: length in bytes
957 * @write_to_vm: bool indicating writing to pages or not
958 * @gfp_mask: memory allocation flags
960 * Prepares and returns a bio for indirect user io, bouncing data
961 * to/from kernel pages as necessary. Must be paired with
962 * call bio_uncopy_user() on io completion.
964 struct bio
*bio_copy_user(struct request_queue
*q
, struct rq_map_data
*map_data
,
965 unsigned long uaddr
, unsigned int len
,
966 int write_to_vm
, gfp_t gfp_mask
)
970 iov
.iov_base
= (void __user
*)uaddr
;
973 return bio_copy_user_iov(q
, map_data
, &iov
, 1, write_to_vm
, gfp_mask
);
975 EXPORT_SYMBOL(bio_copy_user
);
977 static struct bio
*__bio_map_user_iov(struct request_queue
*q
,
978 struct block_device
*bdev
,
979 struct sg_iovec
*iov
, int iov_count
,
980 int write_to_vm
, gfp_t gfp_mask
)
989 for (i
= 0; i
< iov_count
; i
++) {
990 unsigned long uaddr
= (unsigned long)iov
[i
].iov_base
;
991 unsigned long len
= iov
[i
].iov_len
;
992 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
993 unsigned long start
= uaddr
>> PAGE_SHIFT
;
999 return ERR_PTR(-EINVAL
);
1001 nr_pages
+= end
- start
;
1003 * buffer must be aligned to at least hardsector size for now
1005 if (uaddr
& queue_dma_alignment(q
))
1006 return ERR_PTR(-EINVAL
);
1010 return ERR_PTR(-EINVAL
);
1012 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1014 return ERR_PTR(-ENOMEM
);
1017 pages
= kcalloc(nr_pages
, sizeof(struct page
*), gfp_mask
);
1021 for (i
= 0; i
< iov_count
; i
++) {
1022 unsigned long uaddr
= (unsigned long)iov
[i
].iov_base
;
1023 unsigned long len
= iov
[i
].iov_len
;
1024 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1025 unsigned long start
= uaddr
>> PAGE_SHIFT
;
1026 const int local_nr_pages
= end
- start
;
1027 const int page_limit
= cur_page
+ local_nr_pages
;
1029 ret
= get_user_pages_fast(uaddr
, local_nr_pages
,
1030 write_to_vm
, &pages
[cur_page
]);
1031 if (ret
< local_nr_pages
) {
1036 offset
= uaddr
& ~PAGE_MASK
;
1037 for (j
= cur_page
; j
< page_limit
; j
++) {
1038 unsigned int bytes
= PAGE_SIZE
- offset
;
1049 if (bio_add_pc_page(q
, bio
, pages
[j
], bytes
, offset
) <
1059 * release the pages we didn't map into the bio, if any
1061 while (j
< page_limit
)
1062 page_cache_release(pages
[j
++]);
1068 * set data direction, and check if mapped pages need bouncing
1071 bio
->bi_rw
|= REQ_WRITE
;
1073 bio
->bi_bdev
= bdev
;
1074 bio
->bi_flags
|= (1 << BIO_USER_MAPPED
);
1078 for (i
= 0; i
< nr_pages
; i
++) {
1081 page_cache_release(pages
[i
]);
1086 return ERR_PTR(ret
);
1090 * bio_map_user - map user address into bio
1091 * @q: the struct request_queue for the bio
1092 * @bdev: destination block device
1093 * @uaddr: start of user address
1094 * @len: length in bytes
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(struct request_queue
*q
, struct block_device
*bdev
,
1102 unsigned long uaddr
, unsigned int len
, int write_to_vm
,
1105 struct sg_iovec iov
;
1107 iov
.iov_base
= (void __user
*)uaddr
;
1110 return bio_map_user_iov(q
, bdev
, &iov
, 1, write_to_vm
, gfp_mask
);
1112 EXPORT_SYMBOL(bio_map_user
);
1115 * bio_map_user_iov - map user sg_iovec table into bio
1116 * @q: the struct request_queue for the bio
1117 * @bdev: destination block device
1119 * @iov_count: number of elements in the iovec
1120 * @write_to_vm: bool indicating writing to pages or not
1121 * @gfp_mask: memory allocation flags
1123 * Map the user space address into a bio suitable for io to a block
1124 * device. Returns an error pointer in case of error.
1126 struct bio
*bio_map_user_iov(struct request_queue
*q
, struct block_device
*bdev
,
1127 struct sg_iovec
*iov
, int iov_count
,
1128 int write_to_vm
, gfp_t gfp_mask
)
1132 bio
= __bio_map_user_iov(q
, bdev
, iov
, iov_count
, write_to_vm
,
1138 * subtle -- if __bio_map_user() ended up bouncing a bio,
1139 * it would normally disappear when its bi_end_io is run.
1140 * however, we need it for the unmap, so grab an extra
1148 static void __bio_unmap_user(struct bio
*bio
)
1150 struct bio_vec
*bvec
;
1154 * make sure we dirty pages we wrote to
1156 __bio_for_each_segment(bvec
, bio
, i
, 0) {
1157 if (bio_data_dir(bio
) == READ
)
1158 set_page_dirty_lock(bvec
->bv_page
);
1160 page_cache_release(bvec
->bv_page
);
1167 * bio_unmap_user - unmap a bio
1168 * @bio: the bio being unmapped
1170 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1171 * a process context.
1173 * bio_unmap_user() may sleep.
1175 void bio_unmap_user(struct bio
*bio
)
1177 __bio_unmap_user(bio
);
1180 EXPORT_SYMBOL(bio_unmap_user
);
1182 static void bio_map_kern_endio(struct bio
*bio
, int err
)
1187 static struct bio
*__bio_map_kern(struct request_queue
*q
, void *data
,
1188 unsigned int len
, gfp_t gfp_mask
)
1190 unsigned long kaddr
= (unsigned long)data
;
1191 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
1192 unsigned long start
= kaddr
>> PAGE_SHIFT
;
1193 const int nr_pages
= end
- start
;
1197 bio
= bio_kmalloc(gfp_mask
, nr_pages
);
1199 return ERR_PTR(-ENOMEM
);
1201 offset
= offset_in_page(kaddr
);
1202 for (i
= 0; i
< nr_pages
; i
++) {
1203 unsigned int bytes
= PAGE_SIZE
- offset
;
1211 if (bio_add_pc_page(q
, bio
, virt_to_page(data
), bytes
,
1220 bio
->bi_end_io
= bio_map_kern_endio
;
1225 * bio_map_kern - map kernel address into bio
1226 * @q: the struct request_queue for the bio
1227 * @data: pointer to buffer to map
1228 * @len: length in bytes
1229 * @gfp_mask: allocation flags for bio allocation
1231 * Map the kernel address into a bio suitable for io to a block
1232 * device. Returns an error pointer in case of error.
1234 struct bio
*bio_map_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1239 bio
= __bio_map_kern(q
, data
, len
, gfp_mask
);
1243 if (bio
->bi_size
== len
)
1247 * Don't support partial mappings.
1250 return ERR_PTR(-EINVAL
);
1252 EXPORT_SYMBOL(bio_map_kern
);
1254 static void bio_copy_kern_endio(struct bio
*bio
, int err
)
1256 struct bio_vec
*bvec
;
1257 const int read
= bio_data_dir(bio
) == READ
;
1258 struct bio_map_data
*bmd
= bio
->bi_private
;
1260 char *p
= bmd
->sgvecs
[0].iov_base
;
1262 __bio_for_each_segment(bvec
, bio
, i
, 0) {
1263 char *addr
= page_address(bvec
->bv_page
);
1264 int len
= bmd
->iovecs
[i
].bv_len
;
1267 memcpy(p
, addr
, len
);
1269 __free_page(bvec
->bv_page
);
1273 bio_free_map_data(bmd
);
1278 * bio_copy_kern - copy kernel address into bio
1279 * @q: the struct request_queue for the bio
1280 * @data: pointer to buffer to copy
1281 * @len: length in bytes
1282 * @gfp_mask: allocation flags for bio and page allocation
1283 * @reading: data direction is READ
1285 * copy the kernel address into a bio suitable for io to a block
1286 * device. Returns an error pointer in case of error.
1288 struct bio
*bio_copy_kern(struct request_queue
*q
, void *data
, unsigned int len
,
1289 gfp_t gfp_mask
, int reading
)
1292 struct bio_vec
*bvec
;
1295 bio
= bio_copy_user(q
, NULL
, (unsigned long)data
, len
, 1, gfp_mask
);
1302 bio_for_each_segment(bvec
, bio
, i
) {
1303 char *addr
= page_address(bvec
->bv_page
);
1305 memcpy(addr
, p
, bvec
->bv_len
);
1310 bio
->bi_end_io
= bio_copy_kern_endio
;
1314 EXPORT_SYMBOL(bio_copy_kern
);
1317 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1318 * for performing direct-IO in BIOs.
1320 * The problem is that we cannot run set_page_dirty() from interrupt context
1321 * because the required locks are not interrupt-safe. So what we can do is to
1322 * mark the pages dirty _before_ performing IO. And in interrupt context,
1323 * check that the pages are still dirty. If so, fine. If not, redirty them
1324 * in process context.
1326 * We special-case compound pages here: normally this means reads into hugetlb
1327 * pages. The logic in here doesn't really work right for compound pages
1328 * because the VM does not uniformly chase down the head page in all cases.
1329 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1330 * handle them at all. So we skip compound pages here at an early stage.
1332 * Note that this code is very hard to test under normal circumstances because
1333 * direct-io pins the pages with get_user_pages(). This makes
1334 * is_page_cache_freeable return false, and the VM will not clean the pages.
1335 * But other code (eg, flusher threads) could clean the pages if they are mapped
1338 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1339 * deferred bio dirtying paths.
1343 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1345 void bio_set_pages_dirty(struct bio
*bio
)
1347 struct bio_vec
*bvec
= bio
->bi_io_vec
;
1350 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
1351 struct page
*page
= bvec
[i
].bv_page
;
1353 if (page
&& !PageCompound(page
))
1354 set_page_dirty_lock(page
);
1358 static void bio_release_pages(struct bio
*bio
)
1360 struct bio_vec
*bvec
= bio
->bi_io_vec
;
1363 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
1364 struct page
*page
= bvec
[i
].bv_page
;
1372 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1373 * If they are, then fine. If, however, some pages are clean then they must
1374 * have been written out during the direct-IO read. So we take another ref on
1375 * the BIO and the offending pages and re-dirty the pages in process context.
1377 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1378 * here on. It will run one page_cache_release() against each page and will
1379 * run one bio_put() against the BIO.
1382 static void bio_dirty_fn(struct work_struct
*work
);
1384 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
1385 static DEFINE_SPINLOCK(bio_dirty_lock
);
1386 static struct bio
*bio_dirty_list
;
1389 * This runs in process context
1391 static void bio_dirty_fn(struct work_struct
*work
)
1393 unsigned long flags
;
1396 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1397 bio
= bio_dirty_list
;
1398 bio_dirty_list
= NULL
;
1399 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1402 struct bio
*next
= bio
->bi_private
;
1404 bio_set_pages_dirty(bio
);
1405 bio_release_pages(bio
);
1411 void bio_check_pages_dirty(struct bio
*bio
)
1413 struct bio_vec
*bvec
= bio
->bi_io_vec
;
1414 int nr_clean_pages
= 0;
1417 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
1418 struct page
*page
= bvec
[i
].bv_page
;
1420 if (PageDirty(page
) || PageCompound(page
)) {
1421 page_cache_release(page
);
1422 bvec
[i
].bv_page
= NULL
;
1428 if (nr_clean_pages
) {
1429 unsigned long flags
;
1431 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1432 bio
->bi_private
= bio_dirty_list
;
1433 bio_dirty_list
= bio
;
1434 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1435 schedule_work(&bio_dirty_work
);
1441 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1442 void bio_flush_dcache_pages(struct bio
*bi
)
1445 struct bio_vec
*bvec
;
1447 bio_for_each_segment(bvec
, bi
, i
)
1448 flush_dcache_page(bvec
->bv_page
);
1450 EXPORT_SYMBOL(bio_flush_dcache_pages
);
1454 * bio_endio - end I/O on a bio
1456 * @error: error, if any
1459 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1460 * preferred way to end I/O on a bio, it takes care of clearing
1461 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1462 * established -Exxxx (-EIO, for instance) error values in case
1463 * something went wrong. No one should call bi_end_io() directly on a
1464 * bio unless they own it and thus know that it has an end_io
1467 void bio_endio(struct bio
*bio
, int error
)
1470 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1471 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
1475 bio
->bi_end_io(bio
, error
);
1477 EXPORT_SYMBOL(bio_endio
);
1479 void bio_pair_release(struct bio_pair
*bp
)
1481 if (atomic_dec_and_test(&bp
->cnt
)) {
1482 struct bio
*master
= bp
->bio1
.bi_private
;
1484 bio_endio(master
, bp
->error
);
1485 mempool_free(bp
, bp
->bio2
.bi_private
);
1488 EXPORT_SYMBOL(bio_pair_release
);
1490 static void bio_pair_end_1(struct bio
*bi
, int err
)
1492 struct bio_pair
*bp
= container_of(bi
, struct bio_pair
, bio1
);
1497 bio_pair_release(bp
);
1500 static void bio_pair_end_2(struct bio
*bi
, int err
)
1502 struct bio_pair
*bp
= container_of(bi
, struct bio_pair
, bio2
);
1507 bio_pair_release(bp
);
1511 * split a bio - only worry about a bio with a single page in its iovec
1513 struct bio_pair
*bio_split(struct bio
*bi
, int first_sectors
)
1515 struct bio_pair
*bp
= mempool_alloc(bio_split_pool
, GFP_NOIO
);
1520 trace_block_split(bdev_get_queue(bi
->bi_bdev
), bi
,
1521 bi
->bi_sector
+ first_sectors
);
1523 BUG_ON(bi
->bi_vcnt
!= 1);
1524 BUG_ON(bi
->bi_idx
!= 0);
1525 atomic_set(&bp
->cnt
, 3);
1529 bp
->bio2
.bi_sector
+= first_sectors
;
1530 bp
->bio2
.bi_size
-= first_sectors
<< 9;
1531 bp
->bio1
.bi_size
= first_sectors
<< 9;
1533 bp
->bv1
= bi
->bi_io_vec
[0];
1534 bp
->bv2
= bi
->bi_io_vec
[0];
1535 bp
->bv2
.bv_offset
+= first_sectors
<< 9;
1536 bp
->bv2
.bv_len
-= first_sectors
<< 9;
1537 bp
->bv1
.bv_len
= first_sectors
<< 9;
1539 bp
->bio1
.bi_io_vec
= &bp
->bv1
;
1540 bp
->bio2
.bi_io_vec
= &bp
->bv2
;
1542 bp
->bio1
.bi_max_vecs
= 1;
1543 bp
->bio2
.bi_max_vecs
= 1;
1545 bp
->bio1
.bi_end_io
= bio_pair_end_1
;
1546 bp
->bio2
.bi_end_io
= bio_pair_end_2
;
1548 bp
->bio1
.bi_private
= bi
;
1549 bp
->bio2
.bi_private
= bio_split_pool
;
1551 if (bio_integrity(bi
))
1552 bio_integrity_split(bi
, bp
, first_sectors
);
1556 EXPORT_SYMBOL(bio_split
);
1559 * bio_sector_offset - Find hardware sector offset in bio
1560 * @bio: bio to inspect
1561 * @index: bio_vec index
1562 * @offset: offset in bv_page
1564 * Return the number of hardware sectors between beginning of bio
1565 * and an end point indicated by a bio_vec index and an offset
1566 * within that vector's page.
1568 sector_t
bio_sector_offset(struct bio
*bio
, unsigned short index
,
1569 unsigned int offset
)
1571 unsigned int sector_sz
;
1576 sector_sz
= queue_logical_block_size(bio
->bi_bdev
->bd_disk
->queue
);
1579 if (index
>= bio
->bi_idx
)
1580 index
= bio
->bi_vcnt
- 1;
1582 __bio_for_each_segment(bv
, bio
, i
, 0) {
1584 if (offset
> bv
->bv_offset
)
1585 sectors
+= (offset
- bv
->bv_offset
) / sector_sz
;
1589 sectors
+= bv
->bv_len
/ sector_sz
;
1594 EXPORT_SYMBOL(bio_sector_offset
);
1597 * create memory pools for biovec's in a bio_set.
1598 * use the global biovec slabs created for general use.
1600 static int biovec_create_pools(struct bio_set
*bs
, int pool_entries
)
1602 struct biovec_slab
*bp
= bvec_slabs
+ BIOVEC_MAX_IDX
;
1604 bs
->bvec_pool
= mempool_create_slab_pool(pool_entries
, bp
->slab
);
1611 static void biovec_free_pools(struct bio_set
*bs
)
1613 mempool_destroy(bs
->bvec_pool
);
1616 void bioset_free(struct bio_set
*bs
)
1619 mempool_destroy(bs
->bio_pool
);
1621 bioset_integrity_free(bs
);
1622 biovec_free_pools(bs
);
1627 EXPORT_SYMBOL(bioset_free
);
1630 * bioset_create - Create a bio_set
1631 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1632 * @front_pad: Number of bytes to allocate in front of the returned bio
1635 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1636 * to ask for a number of bytes to be allocated in front of the bio.
1637 * Front pad allocation is useful for embedding the bio inside
1638 * another structure, to avoid allocating extra data to go with the bio.
1639 * Note that the bio must be embedded at the END of that structure always,
1640 * or things will break badly.
1642 struct bio_set
*bioset_create(unsigned int pool_size
, unsigned int front_pad
)
1644 unsigned int back_pad
= BIO_INLINE_VECS
* sizeof(struct bio_vec
);
1647 bs
= kzalloc(sizeof(*bs
), GFP_KERNEL
);
1651 bs
->front_pad
= front_pad
;
1653 bs
->bio_slab
= bio_find_or_create_slab(front_pad
+ back_pad
);
1654 if (!bs
->bio_slab
) {
1659 bs
->bio_pool
= mempool_create_slab_pool(pool_size
, bs
->bio_slab
);
1663 if (!biovec_create_pools(bs
, pool_size
))
1670 EXPORT_SYMBOL(bioset_create
);
1672 #ifdef CONFIG_BLK_CGROUP
1674 * bio_associate_current - associate a bio with %current
1677 * Associate @bio with %current if it hasn't been associated yet. Block
1678 * layer will treat @bio as if it were issued by %current no matter which
1679 * task actually issues it.
1681 * This function takes an extra reference of @task's io_context and blkcg
1682 * which will be put when @bio is released. The caller must own @bio,
1683 * ensure %current->io_context exists, and is responsible for synchronizing
1684 * calls to this function.
1686 int bio_associate_current(struct bio
*bio
)
1688 struct io_context
*ioc
;
1689 struct cgroup_subsys_state
*css
;
1694 ioc
= current
->io_context
;
1698 /* acquire active ref on @ioc and associate */
1699 get_io_context_active(ioc
);
1702 /* associate blkcg if exists */
1704 css
= task_subsys_state(current
, blkio_subsys_id
);
1705 if (css
&& css_tryget(css
))
1713 * bio_disassociate_task - undo bio_associate_current()
1716 void bio_disassociate_task(struct bio
*bio
)
1719 put_io_context(bio
->bi_ioc
);
1723 css_put(bio
->bi_css
);
1728 #endif /* CONFIG_BLK_CGROUP */
1730 static void __init
biovec_init_slabs(void)
1734 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
1736 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
1738 if (bvs
->nr_vecs
<= BIO_INLINE_VECS
) {
1743 size
= bvs
->nr_vecs
* sizeof(struct bio_vec
);
1744 bvs
->slab
= kmem_cache_create(bvs
->name
, size
, 0,
1745 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
1749 static int __init
init_bio(void)
1753 bio_slabs
= kzalloc(bio_slab_max
* sizeof(struct bio_slab
), GFP_KERNEL
);
1755 panic("bio: can't allocate bios\n");
1757 bio_integrity_init();
1758 biovec_init_slabs();
1760 fs_bio_set
= bioset_create(BIO_POOL_SIZE
, 0);
1762 panic("bio: can't allocate bios\n");
1764 if (bioset_integrity_create(fs_bio_set
, BIO_POOL_SIZE
))
1765 panic("bio: can't create integrity pool\n");
1767 bio_split_pool
= mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES
,
1768 sizeof(struct bio_pair
));
1769 if (!bio_split_pool
)
1770 panic("bio: can't create split pool\n");
1774 subsys_initcall(init_bio
);