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 <scsi/sg.h> /* for struct sg_iovec */
31 static struct kmem_cache
*bio_slab __read_mostly
;
33 mempool_t
*bio_split_pool __read_mostly
;
36 * if you change this list, also change bvec_alloc or things will
37 * break badly! cannot be bigger than what you can fit into an
41 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
42 static struct biovec_slab bvec_slabs
[BIOVEC_NR_POOLS
] __read_mostly
= {
43 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES
),
48 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
49 * IO code that does not need private memory pools.
51 struct bio_set
*fs_bio_set
;
53 unsigned int bvec_nr_vecs(unsigned short idx
)
55 return bvec_slabs
[idx
].nr_vecs
;
58 struct bio_vec
*bvec_alloc_bs(gfp_t gfp_mask
, int nr
, unsigned long *idx
, struct bio_set
*bs
)
63 * see comment near bvec_array define!
66 case 1 : *idx
= 0; break;
67 case 2 ... 4: *idx
= 1; break;
68 case 5 ... 16: *idx
= 2; break;
69 case 17 ... 64: *idx
= 3; break;
70 case 65 ... 128: *idx
= 4; break;
71 case 129 ... BIO_MAX_PAGES
: *idx
= 5; break;
76 * idx now points to the pool we want to allocate from
79 bvl
= mempool_alloc(bs
->bvec_pools
[*idx
], gfp_mask
);
81 memset(bvl
, 0, bvec_nr_vecs(*idx
) * sizeof(struct bio_vec
));
86 void bio_free(struct bio
*bio
, struct bio_set
*bio_set
)
89 const int pool_idx
= BIO_POOL_IDX(bio
);
91 BIO_BUG_ON(pool_idx
>= BIOVEC_NR_POOLS
);
93 mempool_free(bio
->bi_io_vec
, bio_set
->bvec_pools
[pool_idx
]);
96 if (bio_integrity(bio
))
97 bio_integrity_free(bio
, bio_set
);
99 mempool_free(bio
, bio_set
->bio_pool
);
103 * default destructor for a bio allocated with bio_alloc_bioset()
105 static void bio_fs_destructor(struct bio
*bio
)
107 bio_free(bio
, fs_bio_set
);
110 void bio_init(struct bio
*bio
)
112 memset(bio
, 0, sizeof(*bio
));
113 bio
->bi_flags
= 1 << BIO_UPTODATE
;
114 atomic_set(&bio
->bi_cnt
, 1);
118 * bio_alloc_bioset - allocate a bio for I/O
119 * @gfp_mask: the GFP_ mask given to the slab allocator
120 * @nr_iovecs: number of iovecs to pre-allocate
121 * @bs: the bio_set to allocate from
124 * bio_alloc_bioset will first try it's on mempool to satisfy the allocation.
125 * If %__GFP_WAIT is set then we will block on the internal pool waiting
126 * for a &struct bio to become free.
128 * allocate bio and iovecs from the memory pools specified by the
131 struct bio
*bio_alloc_bioset(gfp_t gfp_mask
, int nr_iovecs
, struct bio_set
*bs
)
133 struct bio
*bio
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
136 struct bio_vec
*bvl
= NULL
;
139 if (likely(nr_iovecs
)) {
140 unsigned long uninitialized_var(idx
);
142 bvl
= bvec_alloc_bs(gfp_mask
, nr_iovecs
, &idx
, bs
);
143 if (unlikely(!bvl
)) {
144 mempool_free(bio
, bs
->bio_pool
);
148 bio
->bi_flags
|= idx
<< BIO_POOL_OFFSET
;
149 bio
->bi_max_vecs
= bvec_nr_vecs(idx
);
151 bio
->bi_io_vec
= bvl
;
157 struct bio
*bio_alloc(gfp_t gfp_mask
, int nr_iovecs
)
159 struct bio
*bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, fs_bio_set
);
162 bio
->bi_destructor
= bio_fs_destructor
;
167 void zero_fill_bio(struct bio
*bio
)
173 bio_for_each_segment(bv
, bio
, i
) {
174 char *data
= bvec_kmap_irq(bv
, &flags
);
175 memset(data
, 0, bv
->bv_len
);
176 flush_dcache_page(bv
->bv_page
);
177 bvec_kunmap_irq(data
, &flags
);
180 EXPORT_SYMBOL(zero_fill_bio
);
183 * bio_put - release a reference to a bio
184 * @bio: bio to release reference to
187 * Put a reference to a &struct bio, either one you have gotten with
188 * bio_alloc or bio_get. The last put of a bio will free it.
190 void bio_put(struct bio
*bio
)
192 BIO_BUG_ON(!atomic_read(&bio
->bi_cnt
));
197 if (atomic_dec_and_test(&bio
->bi_cnt
)) {
199 bio
->bi_destructor(bio
);
203 inline int bio_phys_segments(struct request_queue
*q
, struct bio
*bio
)
205 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
206 blk_recount_segments(q
, bio
);
208 return bio
->bi_phys_segments
;
212 * __bio_clone - clone a bio
213 * @bio: destination bio
214 * @bio_src: bio to clone
216 * Clone a &bio. Caller will own the returned bio, but not
217 * the actual data it points to. Reference count of returned
220 void __bio_clone(struct bio
*bio
, struct bio
*bio_src
)
222 memcpy(bio
->bi_io_vec
, bio_src
->bi_io_vec
,
223 bio_src
->bi_max_vecs
* sizeof(struct bio_vec
));
226 * most users will be overriding ->bi_bdev with a new target,
227 * so we don't set nor calculate new physical/hw segment counts here
229 bio
->bi_sector
= bio_src
->bi_sector
;
230 bio
->bi_bdev
= bio_src
->bi_bdev
;
231 bio
->bi_flags
|= 1 << BIO_CLONED
;
232 bio
->bi_rw
= bio_src
->bi_rw
;
233 bio
->bi_vcnt
= bio_src
->bi_vcnt
;
234 bio
->bi_size
= bio_src
->bi_size
;
235 bio
->bi_idx
= bio_src
->bi_idx
;
239 * bio_clone - clone a bio
241 * @gfp_mask: allocation priority
243 * Like __bio_clone, only also allocates the returned bio
245 struct bio
*bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
247 struct bio
*b
= bio_alloc_bioset(gfp_mask
, bio
->bi_max_vecs
, fs_bio_set
);
252 b
->bi_destructor
= bio_fs_destructor
;
255 if (bio_integrity(bio
)) {
258 ret
= bio_integrity_clone(b
, bio
, fs_bio_set
);
268 * bio_get_nr_vecs - return approx number of vecs
271 * Return the approximate number of pages we can send to this target.
272 * There's no guarantee that you will be able to fit this number of pages
273 * into a bio, it does not account for dynamic restrictions that vary
276 int bio_get_nr_vecs(struct block_device
*bdev
)
278 struct request_queue
*q
= bdev_get_queue(bdev
);
281 nr_pages
= ((q
->max_sectors
<< 9) + PAGE_SIZE
- 1) >> PAGE_SHIFT
;
282 if (nr_pages
> q
->max_phys_segments
)
283 nr_pages
= q
->max_phys_segments
;
284 if (nr_pages
> q
->max_hw_segments
)
285 nr_pages
= q
->max_hw_segments
;
290 static int __bio_add_page(struct request_queue
*q
, struct bio
*bio
, struct page
291 *page
, unsigned int len
, unsigned int offset
,
292 unsigned short max_sectors
)
294 int retried_segments
= 0;
295 struct bio_vec
*bvec
;
298 * cloned bio must not modify vec list
300 if (unlikely(bio_flagged(bio
, BIO_CLONED
)))
303 if (((bio
->bi_size
+ len
) >> 9) > max_sectors
)
307 * For filesystems with a blocksize smaller than the pagesize
308 * we will often be called with the same page as last time and
309 * a consecutive offset. Optimize this special case.
311 if (bio
->bi_vcnt
> 0) {
312 struct bio_vec
*prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
314 if (page
== prev
->bv_page
&&
315 offset
== prev
->bv_offset
+ prev
->bv_len
) {
318 if (q
->merge_bvec_fn
) {
319 struct bvec_merge_data bvm
= {
320 .bi_bdev
= bio
->bi_bdev
,
321 .bi_sector
= bio
->bi_sector
,
322 .bi_size
= bio
->bi_size
,
326 if (q
->merge_bvec_fn(q
, &bvm
, prev
) < len
) {
336 if (bio
->bi_vcnt
>= bio
->bi_max_vecs
)
340 * we might lose a segment or two here, but rather that than
341 * make this too complex.
344 while (bio
->bi_phys_segments
>= q
->max_phys_segments
345 || bio
->bi_phys_segments
>= q
->max_hw_segments
) {
347 if (retried_segments
)
350 retried_segments
= 1;
351 blk_recount_segments(q
, bio
);
355 * setup the new entry, we might clear it again later if we
356 * cannot add the page
358 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
359 bvec
->bv_page
= page
;
361 bvec
->bv_offset
= offset
;
364 * if queue has other restrictions (eg varying max sector size
365 * depending on offset), it can specify a merge_bvec_fn in the
366 * queue to get further control
368 if (q
->merge_bvec_fn
) {
369 struct bvec_merge_data bvm
= {
370 .bi_bdev
= bio
->bi_bdev
,
371 .bi_sector
= bio
->bi_sector
,
372 .bi_size
= bio
->bi_size
,
377 * merge_bvec_fn() returns number of bytes it can accept
380 if (q
->merge_bvec_fn(q
, &bvm
, bvec
) < len
) {
381 bvec
->bv_page
= NULL
;
388 /* If we may be able to merge these biovecs, force a recount */
389 if (bio
->bi_vcnt
&& (BIOVEC_PHYS_MERGEABLE(bvec
-1, bvec
)))
390 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
393 bio
->bi_phys_segments
++;
400 * bio_add_pc_page - attempt to add page to bio
401 * @q: the target queue
402 * @bio: destination bio
404 * @len: vec entry length
405 * @offset: vec entry offset
407 * Attempt to add a page to the bio_vec maplist. This can fail for a
408 * number of reasons, such as the bio being full or target block
409 * device limitations. The target block device must allow bio's
410 * smaller than PAGE_SIZE, so it is always possible to add a single
411 * page to an empty bio. This should only be used by REQ_PC bios.
413 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
, struct page
*page
,
414 unsigned int len
, unsigned int offset
)
416 return __bio_add_page(q
, bio
, page
, len
, offset
, q
->max_hw_sectors
);
420 * bio_add_page - attempt to add page to bio
421 * @bio: destination bio
423 * @len: vec entry length
424 * @offset: vec entry offset
426 * Attempt to add a page to the bio_vec maplist. This can fail for a
427 * number of reasons, such as the bio being full or target block
428 * device limitations. The target block device must allow bio's
429 * smaller than PAGE_SIZE, so it is always possible to add a single
430 * page to an empty bio.
432 int bio_add_page(struct bio
*bio
, struct page
*page
, unsigned int len
,
435 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
436 return __bio_add_page(q
, bio
, page
, len
, offset
, q
->max_sectors
);
439 struct bio_map_data
{
440 struct bio_vec
*iovecs
;
442 struct sg_iovec
*sgvecs
;
445 static void bio_set_map_data(struct bio_map_data
*bmd
, struct bio
*bio
,
446 struct sg_iovec
*iov
, int iov_count
)
448 memcpy(bmd
->iovecs
, bio
->bi_io_vec
, sizeof(struct bio_vec
) * bio
->bi_vcnt
);
449 memcpy(bmd
->sgvecs
, iov
, sizeof(struct sg_iovec
) * iov_count
);
450 bmd
->nr_sgvecs
= iov_count
;
451 bio
->bi_private
= bmd
;
454 static void bio_free_map_data(struct bio_map_data
*bmd
)
461 static struct bio_map_data
*bio_alloc_map_data(int nr_segs
, int iov_count
,
464 struct bio_map_data
*bmd
= kmalloc(sizeof(*bmd
), gfp_mask
);
469 bmd
->iovecs
= kmalloc(sizeof(struct bio_vec
) * nr_segs
, gfp_mask
);
475 bmd
->sgvecs
= kmalloc(sizeof(struct sg_iovec
) * iov_count
, gfp_mask
);
484 static int __bio_copy_iov(struct bio
*bio
, struct bio_vec
*iovecs
,
485 struct sg_iovec
*iov
, int iov_count
, int uncopy
)
488 struct bio_vec
*bvec
;
490 unsigned int iov_off
= 0;
491 int read
= bio_data_dir(bio
) == READ
;
493 __bio_for_each_segment(bvec
, bio
, i
, 0) {
494 char *bv_addr
= page_address(bvec
->bv_page
);
495 unsigned int bv_len
= iovecs
[i
].bv_len
;
497 while (bv_len
&& iov_idx
< iov_count
) {
501 bytes
= min_t(unsigned int,
502 iov
[iov_idx
].iov_len
- iov_off
, bv_len
);
503 iov_addr
= iov
[iov_idx
].iov_base
+ iov_off
;
506 if (!read
&& !uncopy
)
507 ret
= copy_from_user(bv_addr
, iov_addr
,
510 ret
= copy_to_user(iov_addr
, bv_addr
,
522 if (iov
[iov_idx
].iov_len
== iov_off
) {
529 __free_page(bvec
->bv_page
);
536 * bio_uncopy_user - finish previously mapped bio
537 * @bio: bio being terminated
539 * Free pages allocated from bio_copy_user() and write back data
540 * to user space in case of a read.
542 int bio_uncopy_user(struct bio
*bio
)
544 struct bio_map_data
*bmd
= bio
->bi_private
;
547 ret
= __bio_copy_iov(bio
, bmd
->iovecs
, bmd
->sgvecs
, bmd
->nr_sgvecs
, 1);
549 bio_free_map_data(bmd
);
555 * bio_copy_user_iov - copy user data to bio
556 * @q: destination block queue
558 * @iov_count: number of elements in the iovec
559 * @write_to_vm: bool indicating writing to pages or not
561 * Prepares and returns a bio for indirect user io, bouncing data
562 * to/from kernel pages as necessary. Must be paired with
563 * call bio_uncopy_user() on io completion.
565 struct bio
*bio_copy_user_iov(struct request_queue
*q
, struct sg_iovec
*iov
,
566 int iov_count
, int write_to_vm
)
568 struct bio_map_data
*bmd
;
569 struct bio_vec
*bvec
;
574 unsigned int len
= 0;
576 for (i
= 0; i
< iov_count
; i
++) {
581 uaddr
= (unsigned long)iov
[i
].iov_base
;
582 end
= (uaddr
+ iov
[i
].iov_len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
583 start
= uaddr
>> PAGE_SHIFT
;
585 nr_pages
+= end
- start
;
586 len
+= iov
[i
].iov_len
;
589 bmd
= bio_alloc_map_data(nr_pages
, iov_count
, GFP_KERNEL
);
591 return ERR_PTR(-ENOMEM
);
594 bio
= bio_alloc(GFP_KERNEL
, nr_pages
);
598 bio
->bi_rw
|= (!write_to_vm
<< BIO_RW
);
602 unsigned int bytes
= PAGE_SIZE
;
607 page
= alloc_page(q
->bounce_gfp
| GFP_KERNEL
);
613 if (bio_add_pc_page(q
, bio
, page
, bytes
, 0) < bytes
)
626 ret
= __bio_copy_iov(bio
, bio
->bi_io_vec
, iov
, iov_count
, 0);
631 bio_set_map_data(bmd
, bio
, iov
, iov_count
);
634 bio_for_each_segment(bvec
, bio
, i
)
635 __free_page(bvec
->bv_page
);
639 bio_free_map_data(bmd
);
644 * bio_copy_user - copy user data to bio
645 * @q: destination block queue
646 * @uaddr: start of user address
647 * @len: length in bytes
648 * @write_to_vm: bool indicating writing to pages or not
650 * Prepares and returns a bio for indirect user io, bouncing data
651 * to/from kernel pages as necessary. Must be paired with
652 * call bio_uncopy_user() on io completion.
654 struct bio
*bio_copy_user(struct request_queue
*q
, unsigned long uaddr
,
655 unsigned int len
, int write_to_vm
)
659 iov
.iov_base
= (void __user
*)uaddr
;
662 return bio_copy_user_iov(q
, &iov
, 1, write_to_vm
);
665 static struct bio
*__bio_map_user_iov(struct request_queue
*q
,
666 struct block_device
*bdev
,
667 struct sg_iovec
*iov
, int iov_count
,
677 for (i
= 0; i
< iov_count
; i
++) {
678 unsigned long uaddr
= (unsigned long)iov
[i
].iov_base
;
679 unsigned long len
= iov
[i
].iov_len
;
680 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
681 unsigned long start
= uaddr
>> PAGE_SHIFT
;
683 nr_pages
+= end
- start
;
685 * buffer must be aligned to at least hardsector size for now
687 if (uaddr
& queue_dma_alignment(q
))
688 return ERR_PTR(-EINVAL
);
692 return ERR_PTR(-EINVAL
);
694 bio
= bio_alloc(GFP_KERNEL
, nr_pages
);
696 return ERR_PTR(-ENOMEM
);
699 pages
= kcalloc(nr_pages
, sizeof(struct page
*), GFP_KERNEL
);
703 for (i
= 0; i
< iov_count
; i
++) {
704 unsigned long uaddr
= (unsigned long)iov
[i
].iov_base
;
705 unsigned long len
= iov
[i
].iov_len
;
706 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
707 unsigned long start
= uaddr
>> PAGE_SHIFT
;
708 const int local_nr_pages
= end
- start
;
709 const int page_limit
= cur_page
+ local_nr_pages
;
711 ret
= get_user_pages_fast(uaddr
, local_nr_pages
,
712 write_to_vm
, &pages
[cur_page
]);
713 if (ret
< local_nr_pages
) {
718 offset
= uaddr
& ~PAGE_MASK
;
719 for (j
= cur_page
; j
< page_limit
; j
++) {
720 unsigned int bytes
= PAGE_SIZE
- offset
;
731 if (bio_add_pc_page(q
, bio
, pages
[j
], bytes
, offset
) <
741 * release the pages we didn't map into the bio, if any
743 while (j
< page_limit
)
744 page_cache_release(pages
[j
++]);
750 * set data direction, and check if mapped pages need bouncing
753 bio
->bi_rw
|= (1 << BIO_RW
);
756 bio
->bi_flags
|= (1 << BIO_USER_MAPPED
);
760 for (i
= 0; i
< nr_pages
; i
++) {
763 page_cache_release(pages
[i
]);
772 * bio_map_user - map user address into bio
773 * @q: the struct request_queue for the bio
774 * @bdev: destination block device
775 * @uaddr: start of user address
776 * @len: length in bytes
777 * @write_to_vm: bool indicating writing to pages or not
779 * Map the user space address into a bio suitable for io to a block
780 * device. Returns an error pointer in case of error.
782 struct bio
*bio_map_user(struct request_queue
*q
, struct block_device
*bdev
,
783 unsigned long uaddr
, unsigned int len
, int write_to_vm
)
787 iov
.iov_base
= (void __user
*)uaddr
;
790 return bio_map_user_iov(q
, bdev
, &iov
, 1, write_to_vm
);
794 * bio_map_user_iov - map user sg_iovec table into bio
795 * @q: the struct request_queue for the bio
796 * @bdev: destination block device
798 * @iov_count: number of elements in the iovec
799 * @write_to_vm: bool indicating writing to pages or not
801 * Map the user space address into a bio suitable for io to a block
802 * device. Returns an error pointer in case of error.
804 struct bio
*bio_map_user_iov(struct request_queue
*q
, struct block_device
*bdev
,
805 struct sg_iovec
*iov
, int iov_count
,
810 bio
= __bio_map_user_iov(q
, bdev
, iov
, iov_count
, write_to_vm
);
816 * subtle -- if __bio_map_user() ended up bouncing a bio,
817 * it would normally disappear when its bi_end_io is run.
818 * however, we need it for the unmap, so grab an extra
826 static void __bio_unmap_user(struct bio
*bio
)
828 struct bio_vec
*bvec
;
832 * make sure we dirty pages we wrote to
834 __bio_for_each_segment(bvec
, bio
, i
, 0) {
835 if (bio_data_dir(bio
) == READ
)
836 set_page_dirty_lock(bvec
->bv_page
);
838 page_cache_release(bvec
->bv_page
);
845 * bio_unmap_user - unmap a bio
846 * @bio: the bio being unmapped
848 * Unmap a bio previously mapped by bio_map_user(). Must be called with
851 * bio_unmap_user() may sleep.
853 void bio_unmap_user(struct bio
*bio
)
855 __bio_unmap_user(bio
);
859 static void bio_map_kern_endio(struct bio
*bio
, int err
)
865 static struct bio
*__bio_map_kern(struct request_queue
*q
, void *data
,
866 unsigned int len
, gfp_t gfp_mask
)
868 unsigned long kaddr
= (unsigned long)data
;
869 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
870 unsigned long start
= kaddr
>> PAGE_SHIFT
;
871 const int nr_pages
= end
- start
;
875 bio
= bio_alloc(gfp_mask
, nr_pages
);
877 return ERR_PTR(-ENOMEM
);
879 offset
= offset_in_page(kaddr
);
880 for (i
= 0; i
< nr_pages
; i
++) {
881 unsigned int bytes
= PAGE_SIZE
- offset
;
889 if (bio_add_pc_page(q
, bio
, virt_to_page(data
), bytes
,
898 bio
->bi_end_io
= bio_map_kern_endio
;
903 * bio_map_kern - map kernel address into bio
904 * @q: the struct request_queue for the bio
905 * @data: pointer to buffer to map
906 * @len: length in bytes
907 * @gfp_mask: allocation flags for bio allocation
909 * Map the kernel address into a bio suitable for io to a block
910 * device. Returns an error pointer in case of error.
912 struct bio
*bio_map_kern(struct request_queue
*q
, void *data
, unsigned int len
,
917 bio
= __bio_map_kern(q
, data
, len
, gfp_mask
);
921 if (bio
->bi_size
== len
)
925 * Don't support partial mappings.
928 return ERR_PTR(-EINVAL
);
931 static void bio_copy_kern_endio(struct bio
*bio
, int err
)
933 struct bio_vec
*bvec
;
934 const int read
= bio_data_dir(bio
) == READ
;
935 struct bio_map_data
*bmd
= bio
->bi_private
;
937 char *p
= bmd
->sgvecs
[0].iov_base
;
939 __bio_for_each_segment(bvec
, bio
, i
, 0) {
940 char *addr
= page_address(bvec
->bv_page
);
941 int len
= bmd
->iovecs
[i
].bv_len
;
944 memcpy(p
, addr
, len
);
946 __free_page(bvec
->bv_page
);
950 bio_free_map_data(bmd
);
955 * bio_copy_kern - copy kernel address into bio
956 * @q: the struct request_queue for the bio
957 * @data: pointer to buffer to copy
958 * @len: length in bytes
959 * @gfp_mask: allocation flags for bio and page allocation
960 * @reading: data direction is READ
962 * copy the kernel address into a bio suitable for io to a block
963 * device. Returns an error pointer in case of error.
965 struct bio
*bio_copy_kern(struct request_queue
*q
, void *data
, unsigned int len
,
966 gfp_t gfp_mask
, int reading
)
968 unsigned long kaddr
= (unsigned long)data
;
969 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
970 unsigned long start
= kaddr
>> PAGE_SHIFT
;
971 const int nr_pages
= end
- start
;
973 struct bio_vec
*bvec
;
974 struct bio_map_data
*bmd
;
981 bmd
= bio_alloc_map_data(nr_pages
, 1, gfp_mask
);
983 return ERR_PTR(-ENOMEM
);
986 bio
= bio_alloc(gfp_mask
, nr_pages
);
992 unsigned int bytes
= PAGE_SIZE
;
997 page
= alloc_page(q
->bounce_gfp
| gfp_mask
);
1003 if (bio_add_pc_page(q
, bio
, page
, bytes
, 0) < bytes
) {
1014 bio_for_each_segment(bvec
, bio
, i
) {
1015 char *addr
= page_address(bvec
->bv_page
);
1017 memcpy(addr
, p
, bvec
->bv_len
);
1022 bio
->bi_private
= bmd
;
1023 bio
->bi_end_io
= bio_copy_kern_endio
;
1025 bio_set_map_data(bmd
, bio
, &iov
, 1);
1028 bio_for_each_segment(bvec
, bio
, i
)
1029 __free_page(bvec
->bv_page
);
1033 bio_free_map_data(bmd
);
1035 return ERR_PTR(ret
);
1039 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1040 * for performing direct-IO in BIOs.
1042 * The problem is that we cannot run set_page_dirty() from interrupt context
1043 * because the required locks are not interrupt-safe. So what we can do is to
1044 * mark the pages dirty _before_ performing IO. And in interrupt context,
1045 * check that the pages are still dirty. If so, fine. If not, redirty them
1046 * in process context.
1048 * We special-case compound pages here: normally this means reads into hugetlb
1049 * pages. The logic in here doesn't really work right for compound pages
1050 * because the VM does not uniformly chase down the head page in all cases.
1051 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1052 * handle them at all. So we skip compound pages here at an early stage.
1054 * Note that this code is very hard to test under normal circumstances because
1055 * direct-io pins the pages with get_user_pages(). This makes
1056 * is_page_cache_freeable return false, and the VM will not clean the pages.
1057 * But other code (eg, pdflush) could clean the pages if they are mapped
1060 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1061 * deferred bio dirtying paths.
1065 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1067 void bio_set_pages_dirty(struct bio
*bio
)
1069 struct bio_vec
*bvec
= bio
->bi_io_vec
;
1072 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
1073 struct page
*page
= bvec
[i
].bv_page
;
1075 if (page
&& !PageCompound(page
))
1076 set_page_dirty_lock(page
);
1080 static void bio_release_pages(struct bio
*bio
)
1082 struct bio_vec
*bvec
= bio
->bi_io_vec
;
1085 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
1086 struct page
*page
= bvec
[i
].bv_page
;
1094 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1095 * If they are, then fine. If, however, some pages are clean then they must
1096 * have been written out during the direct-IO read. So we take another ref on
1097 * the BIO and the offending pages and re-dirty the pages in process context.
1099 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1100 * here on. It will run one page_cache_release() against each page and will
1101 * run one bio_put() against the BIO.
1104 static void bio_dirty_fn(struct work_struct
*work
);
1106 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
1107 static DEFINE_SPINLOCK(bio_dirty_lock
);
1108 static struct bio
*bio_dirty_list
;
1111 * This runs in process context
1113 static void bio_dirty_fn(struct work_struct
*work
)
1115 unsigned long flags
;
1118 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1119 bio
= bio_dirty_list
;
1120 bio_dirty_list
= NULL
;
1121 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1124 struct bio
*next
= bio
->bi_private
;
1126 bio_set_pages_dirty(bio
);
1127 bio_release_pages(bio
);
1133 void bio_check_pages_dirty(struct bio
*bio
)
1135 struct bio_vec
*bvec
= bio
->bi_io_vec
;
1136 int nr_clean_pages
= 0;
1139 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
1140 struct page
*page
= bvec
[i
].bv_page
;
1142 if (PageDirty(page
) || PageCompound(page
)) {
1143 page_cache_release(page
);
1144 bvec
[i
].bv_page
= NULL
;
1150 if (nr_clean_pages
) {
1151 unsigned long flags
;
1153 spin_lock_irqsave(&bio_dirty_lock
, flags
);
1154 bio
->bi_private
= bio_dirty_list
;
1155 bio_dirty_list
= bio
;
1156 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
1157 schedule_work(&bio_dirty_work
);
1164 * bio_endio - end I/O on a bio
1166 * @error: error, if any
1169 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1170 * preferred way to end I/O on a bio, it takes care of clearing
1171 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1172 * established -Exxxx (-EIO, for instance) error values in case
1173 * something went wrong. Noone should call bi_end_io() directly on a
1174 * bio unless they own it and thus know that it has an end_io
1177 void bio_endio(struct bio
*bio
, int error
)
1180 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1181 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
1185 bio
->bi_end_io(bio
, error
);
1188 void bio_pair_release(struct bio_pair
*bp
)
1190 if (atomic_dec_and_test(&bp
->cnt
)) {
1191 struct bio
*master
= bp
->bio1
.bi_private
;
1193 bio_endio(master
, bp
->error
);
1194 mempool_free(bp
, bp
->bio2
.bi_private
);
1198 static void bio_pair_end_1(struct bio
*bi
, int err
)
1200 struct bio_pair
*bp
= container_of(bi
, struct bio_pair
, bio1
);
1205 bio_pair_release(bp
);
1208 static void bio_pair_end_2(struct bio
*bi
, int err
)
1210 struct bio_pair
*bp
= container_of(bi
, struct bio_pair
, bio2
);
1215 bio_pair_release(bp
);
1219 * split a bio - only worry about a bio with a single page
1222 struct bio_pair
*bio_split(struct bio
*bi
, mempool_t
*pool
, int first_sectors
)
1224 struct bio_pair
*bp
= mempool_alloc(pool
, GFP_NOIO
);
1229 blk_add_trace_pdu_int(bdev_get_queue(bi
->bi_bdev
), BLK_TA_SPLIT
, bi
,
1230 bi
->bi_sector
+ first_sectors
);
1232 BUG_ON(bi
->bi_vcnt
!= 1);
1233 BUG_ON(bi
->bi_idx
!= 0);
1234 atomic_set(&bp
->cnt
, 3);
1238 bp
->bio2
.bi_sector
+= first_sectors
;
1239 bp
->bio2
.bi_size
-= first_sectors
<< 9;
1240 bp
->bio1
.bi_size
= first_sectors
<< 9;
1242 bp
->bv1
= bi
->bi_io_vec
[0];
1243 bp
->bv2
= bi
->bi_io_vec
[0];
1244 bp
->bv2
.bv_offset
+= first_sectors
<< 9;
1245 bp
->bv2
.bv_len
-= first_sectors
<< 9;
1246 bp
->bv1
.bv_len
= first_sectors
<< 9;
1248 bp
->bio1
.bi_io_vec
= &bp
->bv1
;
1249 bp
->bio2
.bi_io_vec
= &bp
->bv2
;
1251 bp
->bio1
.bi_max_vecs
= 1;
1252 bp
->bio2
.bi_max_vecs
= 1;
1254 bp
->bio1
.bi_end_io
= bio_pair_end_1
;
1255 bp
->bio2
.bi_end_io
= bio_pair_end_2
;
1257 bp
->bio1
.bi_private
= bi
;
1258 bp
->bio2
.bi_private
= pool
;
1260 if (bio_integrity(bi
))
1261 bio_integrity_split(bi
, bp
, first_sectors
);
1268 * create memory pools for biovec's in a bio_set.
1269 * use the global biovec slabs created for general use.
1271 static int biovec_create_pools(struct bio_set
*bs
, int pool_entries
)
1275 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
1276 struct biovec_slab
*bp
= bvec_slabs
+ i
;
1277 mempool_t
**bvp
= bs
->bvec_pools
+ i
;
1279 *bvp
= mempool_create_slab_pool(pool_entries
, bp
->slab
);
1286 static void biovec_free_pools(struct bio_set
*bs
)
1290 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
1291 mempool_t
*bvp
= bs
->bvec_pools
[i
];
1294 mempool_destroy(bvp
);
1299 void bioset_free(struct bio_set
*bs
)
1302 mempool_destroy(bs
->bio_pool
);
1304 bioset_integrity_free(bs
);
1305 biovec_free_pools(bs
);
1310 struct bio_set
*bioset_create(int bio_pool_size
, int bvec_pool_size
)
1312 struct bio_set
*bs
= kzalloc(sizeof(*bs
), GFP_KERNEL
);
1317 bs
->bio_pool
= mempool_create_slab_pool(bio_pool_size
, bio_slab
);
1321 if (bioset_integrity_create(bs
, bio_pool_size
))
1324 if (!biovec_create_pools(bs
, bvec_pool_size
))
1332 static void __init
biovec_init_slabs(void)
1336 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
1338 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
1340 size
= bvs
->nr_vecs
* sizeof(struct bio_vec
);
1341 bvs
->slab
= kmem_cache_create(bvs
->name
, size
, 0,
1342 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
1346 static int __init
init_bio(void)
1348 bio_slab
= KMEM_CACHE(bio
, SLAB_HWCACHE_ALIGN
|SLAB_PANIC
);
1350 bio_integrity_init_slab();
1351 biovec_init_slabs();
1353 fs_bio_set
= bioset_create(BIO_POOL_SIZE
, 2);
1355 panic("bio: can't allocate bios\n");
1357 bio_split_pool
= mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES
,
1358 sizeof(struct bio_pair
));
1359 if (!bio_split_pool
)
1360 panic("bio: can't create split pool\n");
1365 subsys_initcall(init_bio
);
1367 EXPORT_SYMBOL(bio_alloc
);
1368 EXPORT_SYMBOL(bio_put
);
1369 EXPORT_SYMBOL(bio_free
);
1370 EXPORT_SYMBOL(bio_endio
);
1371 EXPORT_SYMBOL(bio_init
);
1372 EXPORT_SYMBOL(__bio_clone
);
1373 EXPORT_SYMBOL(bio_clone
);
1374 EXPORT_SYMBOL(bio_phys_segments
);
1375 EXPORT_SYMBOL(bio_add_page
);
1376 EXPORT_SYMBOL(bio_add_pc_page
);
1377 EXPORT_SYMBOL(bio_get_nr_vecs
);
1378 EXPORT_SYMBOL(bio_map_user
);
1379 EXPORT_SYMBOL(bio_unmap_user
);
1380 EXPORT_SYMBOL(bio_map_kern
);
1381 EXPORT_SYMBOL(bio_copy_kern
);
1382 EXPORT_SYMBOL(bio_pair_release
);
1383 EXPORT_SYMBOL(bio_split
);
1384 EXPORT_SYMBOL(bio_split_pool
);
1385 EXPORT_SYMBOL(bio_copy_user
);
1386 EXPORT_SYMBOL(bio_uncopy_user
);
1387 EXPORT_SYMBOL(bioset_create
);
1388 EXPORT_SYMBOL(bioset_free
);
1389 EXPORT_SYMBOL(bio_alloc_bioset
);