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 #define BIO_POOL_SIZE 2
33 static struct kmem_cache
*bio_slab __read_mostly
;
35 #define BIOVEC_NR_POOLS 6
38 * a small number of entries is fine, not going to be performance critical.
39 * basically we just need to survive
41 #define BIO_SPLIT_ENTRIES 2
42 mempool_t
*bio_split_pool __read_mostly
;
47 struct kmem_cache
*slab
;
51 * if you change this list, also change bvec_alloc or things will
52 * break badly! cannot be bigger than what you can fit into an
56 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
57 static struct biovec_slab bvec_slabs
[BIOVEC_NR_POOLS
] __read_mostly
= {
58 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES
),
63 * bio_set is used to allow other portions of the IO system to
64 * allocate their own private memory pools for bio and iovec structures.
65 * These memory pools in turn all allocate from the bio_slab
66 * and the bvec_slabs[].
70 mempool_t
*bvec_pools
[BIOVEC_NR_POOLS
];
74 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
75 * IO code that does not need private memory pools.
77 static struct bio_set
*fs_bio_set
;
79 static inline struct bio_vec
*bvec_alloc_bs(gfp_t gfp_mask
, int nr
, unsigned long *idx
, struct bio_set
*bs
)
84 * see comment near bvec_array define!
87 case 1 : *idx
= 0; break;
88 case 2 ... 4: *idx
= 1; break;
89 case 5 ... 16: *idx
= 2; break;
90 case 17 ... 64: *idx
= 3; break;
91 case 65 ... 128: *idx
= 4; break;
92 case 129 ... BIO_MAX_PAGES
: *idx
= 5; break;
97 * idx now points to the pool we want to allocate from
100 bvl
= mempool_alloc(bs
->bvec_pools
[*idx
], gfp_mask
);
102 struct biovec_slab
*bp
= bvec_slabs
+ *idx
;
104 memset(bvl
, 0, bp
->nr_vecs
* sizeof(struct bio_vec
));
110 void bio_free(struct bio
*bio
, struct bio_set
*bio_set
)
112 const int pool_idx
= BIO_POOL_IDX(bio
);
114 BIO_BUG_ON(pool_idx
>= BIOVEC_NR_POOLS
);
116 mempool_free(bio
->bi_io_vec
, bio_set
->bvec_pools
[pool_idx
]);
117 mempool_free(bio
, bio_set
->bio_pool
);
121 * default destructor for a bio allocated with bio_alloc_bioset()
123 static void bio_fs_destructor(struct bio
*bio
)
125 bio_free(bio
, fs_bio_set
);
128 void bio_init(struct bio
*bio
)
130 memset(bio
, 0, sizeof(*bio
));
131 bio
->bi_flags
= 1 << BIO_UPTODATE
;
132 atomic_set(&bio
->bi_cnt
, 1);
136 * bio_alloc_bioset - allocate a bio for I/O
137 * @gfp_mask: the GFP_ mask given to the slab allocator
138 * @nr_iovecs: number of iovecs to pre-allocate
139 * @bs: the bio_set to allocate from
142 * bio_alloc_bioset will first try it's on mempool to satisfy the allocation.
143 * If %__GFP_WAIT is set then we will block on the internal pool waiting
144 * for a &struct bio to become free.
146 * allocate bio and iovecs from the memory pools specified by the
149 struct bio
*bio_alloc_bioset(gfp_t gfp_mask
, int nr_iovecs
, struct bio_set
*bs
)
151 struct bio
*bio
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
154 struct bio_vec
*bvl
= NULL
;
157 if (likely(nr_iovecs
)) {
158 unsigned long idx
= 0; /* shut up gcc */
160 bvl
= bvec_alloc_bs(gfp_mask
, nr_iovecs
, &idx
, bs
);
161 if (unlikely(!bvl
)) {
162 mempool_free(bio
, bs
->bio_pool
);
166 bio
->bi_flags
|= idx
<< BIO_POOL_OFFSET
;
167 bio
->bi_max_vecs
= bvec_slabs
[idx
].nr_vecs
;
169 bio
->bi_io_vec
= bvl
;
175 struct bio
*bio_alloc(gfp_t gfp_mask
, int nr_iovecs
)
177 struct bio
*bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, fs_bio_set
);
180 bio
->bi_destructor
= bio_fs_destructor
;
185 void zero_fill_bio(struct bio
*bio
)
191 bio_for_each_segment(bv
, bio
, i
) {
192 char *data
= bvec_kmap_irq(bv
, &flags
);
193 memset(data
, 0, bv
->bv_len
);
194 flush_dcache_page(bv
->bv_page
);
195 bvec_kunmap_irq(data
, &flags
);
198 EXPORT_SYMBOL(zero_fill_bio
);
201 * bio_put - release a reference to a bio
202 * @bio: bio to release reference to
205 * Put a reference to a &struct bio, either one you have gotten with
206 * bio_alloc or bio_get. The last put of a bio will free it.
208 void bio_put(struct bio
*bio
)
210 BIO_BUG_ON(!atomic_read(&bio
->bi_cnt
));
215 if (atomic_dec_and_test(&bio
->bi_cnt
)) {
217 bio
->bi_destructor(bio
);
221 inline int bio_phys_segments(struct request_queue
*q
, struct bio
*bio
)
223 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
224 blk_recount_segments(q
, bio
);
226 return bio
->bi_phys_segments
;
229 inline int bio_hw_segments(struct request_queue
*q
, struct bio
*bio
)
231 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
232 blk_recount_segments(q
, bio
);
234 return bio
->bi_hw_segments
;
238 * __bio_clone - clone a bio
239 * @bio: destination bio
240 * @bio_src: bio to clone
242 * Clone a &bio. Caller will own the returned bio, but not
243 * the actual data it points to. Reference count of returned
246 void __bio_clone(struct bio
*bio
, struct bio
*bio_src
)
248 struct request_queue
*q
= bdev_get_queue(bio_src
->bi_bdev
);
250 memcpy(bio
->bi_io_vec
, bio_src
->bi_io_vec
,
251 bio_src
->bi_max_vecs
* sizeof(struct bio_vec
));
253 bio
->bi_sector
= bio_src
->bi_sector
;
254 bio
->bi_bdev
= bio_src
->bi_bdev
;
255 bio
->bi_flags
|= 1 << BIO_CLONED
;
256 bio
->bi_rw
= bio_src
->bi_rw
;
257 bio
->bi_vcnt
= bio_src
->bi_vcnt
;
258 bio
->bi_size
= bio_src
->bi_size
;
259 bio
->bi_idx
= bio_src
->bi_idx
;
260 bio_phys_segments(q
, bio
);
261 bio_hw_segments(q
, bio
);
265 * bio_clone - clone a bio
267 * @gfp_mask: allocation priority
269 * Like __bio_clone, only also allocates the returned bio
271 struct bio
*bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
273 struct bio
*b
= bio_alloc_bioset(gfp_mask
, bio
->bi_max_vecs
, fs_bio_set
);
276 b
->bi_destructor
= bio_fs_destructor
;
284 * bio_get_nr_vecs - return approx number of vecs
287 * Return the approximate number of pages we can send to this target.
288 * There's no guarantee that you will be able to fit this number of pages
289 * into a bio, it does not account for dynamic restrictions that vary
292 int bio_get_nr_vecs(struct block_device
*bdev
)
294 struct request_queue
*q
= bdev_get_queue(bdev
);
297 nr_pages
= ((q
->max_sectors
<< 9) + PAGE_SIZE
- 1) >> PAGE_SHIFT
;
298 if (nr_pages
> q
->max_phys_segments
)
299 nr_pages
= q
->max_phys_segments
;
300 if (nr_pages
> q
->max_hw_segments
)
301 nr_pages
= q
->max_hw_segments
;
306 static int __bio_add_page(struct request_queue
*q
, struct bio
*bio
, struct page
307 *page
, unsigned int len
, unsigned int offset
,
308 unsigned short max_sectors
)
310 int retried_segments
= 0;
311 struct bio_vec
*bvec
;
314 * cloned bio must not modify vec list
316 if (unlikely(bio_flagged(bio
, BIO_CLONED
)))
319 if (((bio
->bi_size
+ len
) >> 9) > max_sectors
)
323 * For filesystems with a blocksize smaller than the pagesize
324 * we will often be called with the same page as last time and
325 * a consecutive offset. Optimize this special case.
327 if (bio
->bi_vcnt
> 0) {
328 struct bio_vec
*prev
= &bio
->bi_io_vec
[bio
->bi_vcnt
- 1];
330 if (page
== prev
->bv_page
&&
331 offset
== prev
->bv_offset
+ prev
->bv_len
) {
333 if (q
->merge_bvec_fn
&&
334 q
->merge_bvec_fn(q
, bio
, prev
) < len
) {
343 if (bio
->bi_vcnt
>= bio
->bi_max_vecs
)
347 * we might lose a segment or two here, but rather that than
348 * make this too complex.
351 while (bio
->bi_phys_segments
>= q
->max_phys_segments
352 || bio
->bi_hw_segments
>= q
->max_hw_segments
353 || BIOVEC_VIRT_OVERSIZE(bio
->bi_size
)) {
355 if (retried_segments
)
358 retried_segments
= 1;
359 blk_recount_segments(q
, bio
);
363 * setup the new entry, we might clear it again later if we
364 * cannot add the page
366 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
367 bvec
->bv_page
= page
;
369 bvec
->bv_offset
= offset
;
372 * if queue has other restrictions (eg varying max sector size
373 * depending on offset), it can specify a merge_bvec_fn in the
374 * queue to get further control
376 if (q
->merge_bvec_fn
) {
378 * merge_bvec_fn() returns number of bytes it can accept
381 if (q
->merge_bvec_fn(q
, bio
, bvec
) < len
) {
382 bvec
->bv_page
= NULL
;
389 /* If we may be able to merge these biovecs, force a recount */
390 if (bio
->bi_vcnt
&& (BIOVEC_PHYS_MERGEABLE(bvec
-1, bvec
) ||
391 BIOVEC_VIRT_MERGEABLE(bvec
-1, bvec
)))
392 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
395 bio
->bi_phys_segments
++;
396 bio
->bi_hw_segments
++;
403 * bio_add_pc_page - attempt to add page to bio
404 * @q: the target queue
405 * @bio: destination bio
407 * @len: vec entry length
408 * @offset: vec entry offset
410 * Attempt to add a page to the bio_vec maplist. This can fail for a
411 * number of reasons, such as the bio being full or target block
412 * device limitations. The target block device must allow bio's
413 * smaller than PAGE_SIZE, so it is always possible to add a single
414 * page to an empty bio. This should only be used by REQ_PC bios.
416 int bio_add_pc_page(struct request_queue
*q
, struct bio
*bio
, struct page
*page
,
417 unsigned int len
, unsigned int offset
)
419 return __bio_add_page(q
, bio
, page
, len
, offset
, q
->max_hw_sectors
);
423 * bio_add_page - attempt to add page to bio
424 * @bio: destination bio
426 * @len: vec entry length
427 * @offset: vec entry offset
429 * Attempt to add a page to the bio_vec maplist. This can fail for a
430 * number of reasons, such as the bio being full or target block
431 * device limitations. The target block device must allow bio's
432 * smaller than PAGE_SIZE, so it is always possible to add a single
433 * page to an empty bio.
435 int bio_add_page(struct bio
*bio
, struct page
*page
, unsigned int len
,
438 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
439 return __bio_add_page(q
, bio
, page
, len
, offset
, q
->max_sectors
);
442 struct bio_map_data
{
443 struct bio_vec
*iovecs
;
444 void __user
*userptr
;
447 static void bio_set_map_data(struct bio_map_data
*bmd
, struct bio
*bio
)
449 memcpy(bmd
->iovecs
, bio
->bi_io_vec
, sizeof(struct bio_vec
) * bio
->bi_vcnt
);
450 bio
->bi_private
= bmd
;
453 static void bio_free_map_data(struct bio_map_data
*bmd
)
459 static struct bio_map_data
*bio_alloc_map_data(int nr_segs
)
461 struct bio_map_data
*bmd
= kmalloc(sizeof(*bmd
), GFP_KERNEL
);
466 bmd
->iovecs
= kmalloc(sizeof(struct bio_vec
) * nr_segs
, GFP_KERNEL
);
475 * bio_uncopy_user - finish previously mapped bio
476 * @bio: bio being terminated
478 * Free pages allocated from bio_copy_user() and write back data
479 * to user space in case of a read.
481 int bio_uncopy_user(struct bio
*bio
)
483 struct bio_map_data
*bmd
= bio
->bi_private
;
484 const int read
= bio_data_dir(bio
) == READ
;
485 struct bio_vec
*bvec
;
488 __bio_for_each_segment(bvec
, bio
, i
, 0) {
489 char *addr
= page_address(bvec
->bv_page
);
490 unsigned int len
= bmd
->iovecs
[i
].bv_len
;
492 if (read
&& !ret
&& copy_to_user(bmd
->userptr
, addr
, len
))
495 __free_page(bvec
->bv_page
);
498 bio_free_map_data(bmd
);
504 * bio_copy_user - copy user data to bio
505 * @q: destination block queue
506 * @uaddr: start of user address
507 * @len: length in bytes
508 * @write_to_vm: bool indicating writing to pages or not
510 * Prepares and returns a bio for indirect user io, bouncing data
511 * to/from kernel pages as necessary. Must be paired with
512 * call bio_uncopy_user() on io completion.
514 struct bio
*bio_copy_user(struct request_queue
*q
, unsigned long uaddr
,
515 unsigned int len
, int write_to_vm
)
517 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
518 unsigned long start
= uaddr
>> PAGE_SHIFT
;
519 struct bio_map_data
*bmd
;
520 struct bio_vec
*bvec
;
525 bmd
= bio_alloc_map_data(end
- start
);
527 return ERR_PTR(-ENOMEM
);
529 bmd
->userptr
= (void __user
*) uaddr
;
532 bio
= bio_alloc(GFP_KERNEL
, end
- start
);
536 bio
->bi_rw
|= (!write_to_vm
<< BIO_RW
);
540 unsigned int bytes
= PAGE_SIZE
;
545 page
= alloc_page(q
->bounce_gfp
| GFP_KERNEL
);
551 if (bio_add_pc_page(q
, bio
, page
, bytes
, 0) < bytes
)
564 char __user
*p
= (char __user
*) uaddr
;
567 * for a write, copy in data to kernel pages
570 bio_for_each_segment(bvec
, bio
, i
) {
571 char *addr
= page_address(bvec
->bv_page
);
573 if (copy_from_user(addr
, p
, bvec
->bv_len
))
579 bio_set_map_data(bmd
, bio
);
582 bio_for_each_segment(bvec
, bio
, i
)
583 __free_page(bvec
->bv_page
);
587 bio_free_map_data(bmd
);
591 static struct bio
*__bio_map_user_iov(struct request_queue
*q
,
592 struct block_device
*bdev
,
593 struct sg_iovec
*iov
, int iov_count
,
603 for (i
= 0; i
< iov_count
; i
++) {
604 unsigned long uaddr
= (unsigned long)iov
[i
].iov_base
;
605 unsigned long len
= iov
[i
].iov_len
;
606 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
607 unsigned long start
= uaddr
>> PAGE_SHIFT
;
609 nr_pages
+= end
- start
;
611 * buffer must be aligned to at least hardsector size for now
613 if (uaddr
& queue_dma_alignment(q
))
614 return ERR_PTR(-EINVAL
);
618 return ERR_PTR(-EINVAL
);
620 bio
= bio_alloc(GFP_KERNEL
, nr_pages
);
622 return ERR_PTR(-ENOMEM
);
625 pages
= kcalloc(nr_pages
, sizeof(struct page
*), GFP_KERNEL
);
629 for (i
= 0; i
< iov_count
; i
++) {
630 unsigned long uaddr
= (unsigned long)iov
[i
].iov_base
;
631 unsigned long len
= iov
[i
].iov_len
;
632 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
633 unsigned long start
= uaddr
>> PAGE_SHIFT
;
634 const int local_nr_pages
= end
- start
;
635 const int page_limit
= cur_page
+ local_nr_pages
;
637 down_read(¤t
->mm
->mmap_sem
);
638 ret
= get_user_pages(current
, current
->mm
, uaddr
,
640 write_to_vm
, 0, &pages
[cur_page
], NULL
);
641 up_read(¤t
->mm
->mmap_sem
);
643 if (ret
< local_nr_pages
) {
648 offset
= uaddr
& ~PAGE_MASK
;
649 for (j
= cur_page
; j
< page_limit
; j
++) {
650 unsigned int bytes
= PAGE_SIZE
- offset
;
661 if (bio_add_pc_page(q
, bio
, pages
[j
], bytes
, offset
) <
671 * release the pages we didn't map into the bio, if any
673 while (j
< page_limit
)
674 page_cache_release(pages
[j
++]);
680 * set data direction, and check if mapped pages need bouncing
683 bio
->bi_rw
|= (1 << BIO_RW
);
686 bio
->bi_flags
|= (1 << BIO_USER_MAPPED
);
690 for (i
= 0; i
< nr_pages
; i
++) {
693 page_cache_release(pages
[i
]);
702 * bio_map_user - map user address into bio
703 * @q: the struct request_queue for the bio
704 * @bdev: destination block device
705 * @uaddr: start of user address
706 * @len: length in bytes
707 * @write_to_vm: bool indicating writing to pages or not
709 * Map the user space address into a bio suitable for io to a block
710 * device. Returns an error pointer in case of error.
712 struct bio
*bio_map_user(struct request_queue
*q
, struct block_device
*bdev
,
713 unsigned long uaddr
, unsigned int len
, int write_to_vm
)
717 iov
.iov_base
= (void __user
*)uaddr
;
720 return bio_map_user_iov(q
, bdev
, &iov
, 1, write_to_vm
);
724 * bio_map_user_iov - map user sg_iovec table into bio
725 * @q: the struct request_queue for the bio
726 * @bdev: destination block device
728 * @iov_count: number of elements in the iovec
729 * @write_to_vm: bool indicating writing to pages or not
731 * Map the user space address into a bio suitable for io to a block
732 * device. Returns an error pointer in case of error.
734 struct bio
*bio_map_user_iov(struct request_queue
*q
, struct block_device
*bdev
,
735 struct sg_iovec
*iov
, int iov_count
,
740 bio
= __bio_map_user_iov(q
, bdev
, iov
, iov_count
, write_to_vm
);
746 * subtle -- if __bio_map_user() ended up bouncing a bio,
747 * it would normally disappear when its bi_end_io is run.
748 * however, we need it for the unmap, so grab an extra
756 static void __bio_unmap_user(struct bio
*bio
)
758 struct bio_vec
*bvec
;
762 * make sure we dirty pages we wrote to
764 __bio_for_each_segment(bvec
, bio
, i
, 0) {
765 if (bio_data_dir(bio
) == READ
)
766 set_page_dirty_lock(bvec
->bv_page
);
768 page_cache_release(bvec
->bv_page
);
775 * bio_unmap_user - unmap a bio
776 * @bio: the bio being unmapped
778 * Unmap a bio previously mapped by bio_map_user(). Must be called with
781 * bio_unmap_user() may sleep.
783 void bio_unmap_user(struct bio
*bio
)
785 __bio_unmap_user(bio
);
789 static void bio_map_kern_endio(struct bio
*bio
, int err
)
795 static struct bio
*__bio_map_kern(struct request_queue
*q
, void *data
,
796 unsigned int len
, gfp_t gfp_mask
)
798 unsigned long kaddr
= (unsigned long)data
;
799 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
800 unsigned long start
= kaddr
>> PAGE_SHIFT
;
801 const int nr_pages
= end
- start
;
805 bio
= bio_alloc(gfp_mask
, nr_pages
);
807 return ERR_PTR(-ENOMEM
);
809 offset
= offset_in_page(kaddr
);
810 for (i
= 0; i
< nr_pages
; i
++) {
811 unsigned int bytes
= PAGE_SIZE
- offset
;
819 if (bio_add_pc_page(q
, bio
, virt_to_page(data
), bytes
,
828 bio
->bi_end_io
= bio_map_kern_endio
;
833 * bio_map_kern - map kernel address into bio
834 * @q: the struct request_queue for the bio
835 * @data: pointer to buffer to map
836 * @len: length in bytes
837 * @gfp_mask: allocation flags for bio allocation
839 * Map the kernel address into a bio suitable for io to a block
840 * device. Returns an error pointer in case of error.
842 struct bio
*bio_map_kern(struct request_queue
*q
, void *data
, unsigned int len
,
847 bio
= __bio_map_kern(q
, data
, len
, gfp_mask
);
851 if (bio
->bi_size
== len
)
855 * Don't support partial mappings.
858 return ERR_PTR(-EINVAL
);
862 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
863 * for performing direct-IO in BIOs.
865 * The problem is that we cannot run set_page_dirty() from interrupt context
866 * because the required locks are not interrupt-safe. So what we can do is to
867 * mark the pages dirty _before_ performing IO. And in interrupt context,
868 * check that the pages are still dirty. If so, fine. If not, redirty them
869 * in process context.
871 * We special-case compound pages here: normally this means reads into hugetlb
872 * pages. The logic in here doesn't really work right for compound pages
873 * because the VM does not uniformly chase down the head page in all cases.
874 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
875 * handle them at all. So we skip compound pages here at an early stage.
877 * Note that this code is very hard to test under normal circumstances because
878 * direct-io pins the pages with get_user_pages(). This makes
879 * is_page_cache_freeable return false, and the VM will not clean the pages.
880 * But other code (eg, pdflush) could clean the pages if they are mapped
883 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
884 * deferred bio dirtying paths.
888 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
890 void bio_set_pages_dirty(struct bio
*bio
)
892 struct bio_vec
*bvec
= bio
->bi_io_vec
;
895 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
896 struct page
*page
= bvec
[i
].bv_page
;
898 if (page
&& !PageCompound(page
))
899 set_page_dirty_lock(page
);
903 void bio_release_pages(struct bio
*bio
)
905 struct bio_vec
*bvec
= bio
->bi_io_vec
;
908 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
909 struct page
*page
= bvec
[i
].bv_page
;
917 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
918 * If they are, then fine. If, however, some pages are clean then they must
919 * have been written out during the direct-IO read. So we take another ref on
920 * the BIO and the offending pages and re-dirty the pages in process context.
922 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
923 * here on. It will run one page_cache_release() against each page and will
924 * run one bio_put() against the BIO.
927 static void bio_dirty_fn(struct work_struct
*work
);
929 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
);
930 static DEFINE_SPINLOCK(bio_dirty_lock
);
931 static struct bio
*bio_dirty_list
;
934 * This runs in process context
936 static void bio_dirty_fn(struct work_struct
*work
)
941 spin_lock_irqsave(&bio_dirty_lock
, flags
);
942 bio
= bio_dirty_list
;
943 bio_dirty_list
= NULL
;
944 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
947 struct bio
*next
= bio
->bi_private
;
949 bio_set_pages_dirty(bio
);
950 bio_release_pages(bio
);
956 void bio_check_pages_dirty(struct bio
*bio
)
958 struct bio_vec
*bvec
= bio
->bi_io_vec
;
959 int nr_clean_pages
= 0;
962 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
963 struct page
*page
= bvec
[i
].bv_page
;
965 if (PageDirty(page
) || PageCompound(page
)) {
966 page_cache_release(page
);
967 bvec
[i
].bv_page
= NULL
;
973 if (nr_clean_pages
) {
976 spin_lock_irqsave(&bio_dirty_lock
, flags
);
977 bio
->bi_private
= bio_dirty_list
;
978 bio_dirty_list
= bio
;
979 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
980 schedule_work(&bio_dirty_work
);
987 * bio_endio - end I/O on a bio
989 * @error: error, if any
992 * bio_endio() will end I/O on the whole bio. bio_endio() is the
993 * preferred way to end I/O on a bio, it takes care of clearing
994 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
995 * established -Exxxx (-EIO, for instance) error values in case
996 * something went wrong. Noone should call bi_end_io() directly on a
997 * bio unless they own it and thus know that it has an end_io
1000 void bio_endio(struct bio
*bio
, int error
)
1003 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1004 else if (!test_bit(BIO_UPTODATE
, &bio
->bi_flags
))
1008 bio
->bi_end_io(bio
, error
);
1011 void bio_pair_release(struct bio_pair
*bp
)
1013 if (atomic_dec_and_test(&bp
->cnt
)) {
1014 struct bio
*master
= bp
->bio1
.bi_private
;
1016 bio_endio(master
, bp
->error
);
1017 mempool_free(bp
, bp
->bio2
.bi_private
);
1021 static void bio_pair_end_1(struct bio
*bi
, int err
)
1023 struct bio_pair
*bp
= container_of(bi
, struct bio_pair
, bio1
);
1028 bio_pair_release(bp
);
1031 static void bio_pair_end_2(struct bio
*bi
, int err
)
1033 struct bio_pair
*bp
= container_of(bi
, struct bio_pair
, bio2
);
1038 bio_pair_release(bp
);
1042 * split a bio - only worry about a bio with a single page
1045 struct bio_pair
*bio_split(struct bio
*bi
, mempool_t
*pool
, int first_sectors
)
1047 struct bio_pair
*bp
= mempool_alloc(pool
, GFP_NOIO
);
1052 blk_add_trace_pdu_int(bdev_get_queue(bi
->bi_bdev
), BLK_TA_SPLIT
, bi
,
1053 bi
->bi_sector
+ first_sectors
);
1055 BUG_ON(bi
->bi_vcnt
!= 1);
1056 BUG_ON(bi
->bi_idx
!= 0);
1057 atomic_set(&bp
->cnt
, 3);
1061 bp
->bio2
.bi_sector
+= first_sectors
;
1062 bp
->bio2
.bi_size
-= first_sectors
<< 9;
1063 bp
->bio1
.bi_size
= first_sectors
<< 9;
1065 bp
->bv1
= bi
->bi_io_vec
[0];
1066 bp
->bv2
= bi
->bi_io_vec
[0];
1067 bp
->bv2
.bv_offset
+= first_sectors
<< 9;
1068 bp
->bv2
.bv_len
-= first_sectors
<< 9;
1069 bp
->bv1
.bv_len
= first_sectors
<< 9;
1071 bp
->bio1
.bi_io_vec
= &bp
->bv1
;
1072 bp
->bio2
.bi_io_vec
= &bp
->bv2
;
1074 bp
->bio1
.bi_max_vecs
= 1;
1075 bp
->bio2
.bi_max_vecs
= 1;
1077 bp
->bio1
.bi_end_io
= bio_pair_end_1
;
1078 bp
->bio2
.bi_end_io
= bio_pair_end_2
;
1080 bp
->bio1
.bi_private
= bi
;
1081 bp
->bio2
.bi_private
= pool
;
1088 * create memory pools for biovec's in a bio_set.
1089 * use the global biovec slabs created for general use.
1091 static int biovec_create_pools(struct bio_set
*bs
, int pool_entries
)
1095 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
1096 struct biovec_slab
*bp
= bvec_slabs
+ i
;
1097 mempool_t
**bvp
= bs
->bvec_pools
+ i
;
1099 *bvp
= mempool_create_slab_pool(pool_entries
, bp
->slab
);
1106 static void biovec_free_pools(struct bio_set
*bs
)
1110 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
1111 mempool_t
*bvp
= bs
->bvec_pools
[i
];
1114 mempool_destroy(bvp
);
1119 void bioset_free(struct bio_set
*bs
)
1122 mempool_destroy(bs
->bio_pool
);
1124 biovec_free_pools(bs
);
1129 struct bio_set
*bioset_create(int bio_pool_size
, int bvec_pool_size
)
1131 struct bio_set
*bs
= kzalloc(sizeof(*bs
), GFP_KERNEL
);
1136 bs
->bio_pool
= mempool_create_slab_pool(bio_pool_size
, bio_slab
);
1140 if (!biovec_create_pools(bs
, bvec_pool_size
))
1148 static void __init
biovec_init_slabs(void)
1152 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
1154 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
1156 size
= bvs
->nr_vecs
* sizeof(struct bio_vec
);
1157 bvs
->slab
= kmem_cache_create(bvs
->name
, size
, 0,
1158 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
1162 static int __init
init_bio(void)
1164 bio_slab
= KMEM_CACHE(bio
, SLAB_HWCACHE_ALIGN
|SLAB_PANIC
);
1166 biovec_init_slabs();
1168 fs_bio_set
= bioset_create(BIO_POOL_SIZE
, 2);
1170 panic("bio: can't allocate bios\n");
1172 bio_split_pool
= mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES
,
1173 sizeof(struct bio_pair
));
1174 if (!bio_split_pool
)
1175 panic("bio: can't create split pool\n");
1180 subsys_initcall(init_bio
);
1182 EXPORT_SYMBOL(bio_alloc
);
1183 EXPORT_SYMBOL(bio_put
);
1184 EXPORT_SYMBOL(bio_free
);
1185 EXPORT_SYMBOL(bio_endio
);
1186 EXPORT_SYMBOL(bio_init
);
1187 EXPORT_SYMBOL(__bio_clone
);
1188 EXPORT_SYMBOL(bio_clone
);
1189 EXPORT_SYMBOL(bio_phys_segments
);
1190 EXPORT_SYMBOL(bio_hw_segments
);
1191 EXPORT_SYMBOL(bio_add_page
);
1192 EXPORT_SYMBOL(bio_add_pc_page
);
1193 EXPORT_SYMBOL(bio_get_nr_vecs
);
1194 EXPORT_SYMBOL(bio_map_kern
);
1195 EXPORT_SYMBOL(bio_pair_release
);
1196 EXPORT_SYMBOL(bio_split
);
1197 EXPORT_SYMBOL(bio_split_pool
);
1198 EXPORT_SYMBOL(bio_copy_user
);
1199 EXPORT_SYMBOL(bio_uncopy_user
);
1200 EXPORT_SYMBOL(bioset_create
);
1201 EXPORT_SYMBOL(bioset_free
);
1202 EXPORT_SYMBOL(bio_alloc_bioset
);