2 * Copyright (C) 2001 Jens Axboe <axboe@suse.de>
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 <scsi/sg.h> /* for struct sg_iovec */
30 #define BIO_POOL_SIZE 256
32 static kmem_cache_t
*bio_slab
;
34 #define BIOVEC_NR_POOLS 6
37 * a small number of entries is fine, not going to be performance critical.
38 * basically we just need to survive
40 #define BIO_SPLIT_ENTRIES 8
41 mempool_t
*bio_split_pool
;
50 * if you change this list, also change bvec_alloc or things will
51 * break badly! cannot be bigger than what you can fit into an
55 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
56 static struct biovec_slab bvec_slabs
[BIOVEC_NR_POOLS
] __read_mostly
= {
57 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES
),
62 * bio_set is used to allow other portions of the IO system to
63 * allocate their own private memory pools for bio and iovec structures.
64 * These memory pools in turn all allocate from the bio_slab
65 * and the bvec_slabs[].
69 mempool_t
*bvec_pools
[BIOVEC_NR_POOLS
];
73 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
74 * IO code that does not need private memory pools.
76 static struct bio_set
*fs_bio_set
;
78 static inline struct bio_vec
*bvec_alloc_bs(gfp_t gfp_mask
, int nr
, unsigned long *idx
, struct bio_set
*bs
)
81 struct biovec_slab
*bp
;
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 bp
= bvec_slabs
+ *idx
;
101 bvl
= mempool_alloc(bs
->bvec_pools
[*idx
], gfp_mask
);
103 memset(bvl
, 0, bp
->nr_vecs
* sizeof(struct bio_vec
));
108 void bio_free(struct bio
*bio
, struct bio_set
*bio_set
)
110 const int pool_idx
= BIO_POOL_IDX(bio
);
112 BIO_BUG_ON(pool_idx
>= BIOVEC_NR_POOLS
);
114 mempool_free(bio
->bi_io_vec
, bio_set
->bvec_pools
[pool_idx
]);
115 mempool_free(bio
, bio_set
->bio_pool
);
119 * default destructor for a bio allocated with bio_alloc_bioset()
121 static void bio_fs_destructor(struct bio
*bio
)
123 bio_free(bio
, fs_bio_set
);
126 inline void bio_init(struct bio
*bio
)
129 bio
->bi_flags
= 1 << BIO_UPTODATE
;
133 bio
->bi_phys_segments
= 0;
134 bio
->bi_hw_segments
= 0;
135 bio
->bi_hw_front_size
= 0;
136 bio
->bi_hw_back_size
= 0;
138 bio
->bi_max_vecs
= 0;
139 bio
->bi_end_io
= NULL
;
140 atomic_set(&bio
->bi_cnt
, 1);
141 bio
->bi_private
= NULL
;
145 * bio_alloc_bioset - allocate a bio for I/O
146 * @gfp_mask: the GFP_ mask given to the slab allocator
147 * @nr_iovecs: number of iovecs to pre-allocate
148 * @bs: the bio_set to allocate from
151 * bio_alloc_bioset will first try it's on mempool to satisfy the allocation.
152 * If %__GFP_WAIT is set then we will block on the internal pool waiting
153 * for a &struct bio to become free.
155 * allocate bio and iovecs from the memory pools specified by the
158 struct bio
*bio_alloc_bioset(gfp_t gfp_mask
, int nr_iovecs
, struct bio_set
*bs
)
160 struct bio
*bio
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
163 struct bio_vec
*bvl
= NULL
;
166 if (likely(nr_iovecs
)) {
169 bvl
= bvec_alloc_bs(gfp_mask
, nr_iovecs
, &idx
, bs
);
170 if (unlikely(!bvl
)) {
171 mempool_free(bio
, bs
->bio_pool
);
175 bio
->bi_flags
|= idx
<< BIO_POOL_OFFSET
;
176 bio
->bi_max_vecs
= bvec_slabs
[idx
].nr_vecs
;
178 bio
->bi_io_vec
= bvl
;
184 struct bio
*bio_alloc(gfp_t gfp_mask
, int nr_iovecs
)
186 struct bio
*bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, fs_bio_set
);
189 bio
->bi_destructor
= bio_fs_destructor
;
194 void zero_fill_bio(struct bio
*bio
)
200 bio_for_each_segment(bv
, bio
, i
) {
201 char *data
= bvec_kmap_irq(bv
, &flags
);
202 memset(data
, 0, bv
->bv_len
);
203 flush_dcache_page(bv
->bv_page
);
204 bvec_kunmap_irq(data
, &flags
);
207 EXPORT_SYMBOL(zero_fill_bio
);
210 * bio_put - release a reference to a bio
211 * @bio: bio to release reference to
214 * Put a reference to a &struct bio, either one you have gotten with
215 * bio_alloc or bio_get. The last put of a bio will free it.
217 void bio_put(struct bio
*bio
)
219 BIO_BUG_ON(!atomic_read(&bio
->bi_cnt
));
224 if (atomic_dec_and_test(&bio
->bi_cnt
)) {
226 bio
->bi_destructor(bio
);
230 inline int bio_phys_segments(request_queue_t
*q
, struct bio
*bio
)
232 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
233 blk_recount_segments(q
, bio
);
235 return bio
->bi_phys_segments
;
238 inline int bio_hw_segments(request_queue_t
*q
, struct bio
*bio
)
240 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
241 blk_recount_segments(q
, bio
);
243 return bio
->bi_hw_segments
;
247 * __bio_clone - clone a bio
248 * @bio: destination bio
249 * @bio_src: bio to clone
251 * Clone a &bio. Caller will own the returned bio, but not
252 * the actual data it points to. Reference count of returned
255 inline void __bio_clone(struct bio
*bio
, struct bio
*bio_src
)
257 request_queue_t
*q
= bdev_get_queue(bio_src
->bi_bdev
);
259 memcpy(bio
->bi_io_vec
, bio_src
->bi_io_vec
,
260 bio_src
->bi_max_vecs
* sizeof(struct bio_vec
));
262 bio
->bi_sector
= bio_src
->bi_sector
;
263 bio
->bi_bdev
= bio_src
->bi_bdev
;
264 bio
->bi_flags
|= 1 << BIO_CLONED
;
265 bio
->bi_rw
= bio_src
->bi_rw
;
266 bio
->bi_vcnt
= bio_src
->bi_vcnt
;
267 bio
->bi_size
= bio_src
->bi_size
;
268 bio
->bi_idx
= bio_src
->bi_idx
;
269 bio_phys_segments(q
, bio
);
270 bio_hw_segments(q
, bio
);
274 * bio_clone - clone a bio
276 * @gfp_mask: allocation priority
278 * Like __bio_clone, only also allocates the returned bio
280 struct bio
*bio_clone(struct bio
*bio
, gfp_t gfp_mask
)
282 struct bio
*b
= bio_alloc_bioset(gfp_mask
, bio
->bi_max_vecs
, fs_bio_set
);
285 b
->bi_destructor
= bio_fs_destructor
;
293 * bio_get_nr_vecs - return approx number of vecs
296 * Return the approximate number of pages we can send to this target.
297 * There's no guarantee that you will be able to fit this number of pages
298 * into a bio, it does not account for dynamic restrictions that vary
301 int bio_get_nr_vecs(struct block_device
*bdev
)
303 request_queue_t
*q
= bdev_get_queue(bdev
);
306 nr_pages
= ((q
->max_sectors
<< 9) + PAGE_SIZE
- 1) >> PAGE_SHIFT
;
307 if (nr_pages
> q
->max_phys_segments
)
308 nr_pages
= q
->max_phys_segments
;
309 if (nr_pages
> q
->max_hw_segments
)
310 nr_pages
= q
->max_hw_segments
;
315 static int __bio_add_page(request_queue_t
*q
, struct bio
*bio
, struct page
316 *page
, unsigned int len
, unsigned int offset
,
317 unsigned short max_sectors
)
319 int retried_segments
= 0;
320 struct bio_vec
*bvec
;
323 * cloned bio must not modify vec list
325 if (unlikely(bio_flagged(bio
, BIO_CLONED
)))
328 if (bio
->bi_vcnt
>= bio
->bi_max_vecs
)
331 if (((bio
->bi_size
+ len
) >> 9) > max_sectors
)
335 * we might lose a segment or two here, but rather that than
336 * make this too complex.
339 while (bio
->bi_phys_segments
>= q
->max_phys_segments
340 || bio
->bi_hw_segments
>= q
->max_hw_segments
341 || BIOVEC_VIRT_OVERSIZE(bio
->bi_size
)) {
343 if (retried_segments
)
346 retried_segments
= 1;
347 blk_recount_segments(q
, bio
);
351 * setup the new entry, we might clear it again later if we
352 * cannot add the page
354 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
355 bvec
->bv_page
= page
;
357 bvec
->bv_offset
= offset
;
360 * if queue has other restrictions (eg varying max sector size
361 * depending on offset), it can specify a merge_bvec_fn in the
362 * queue to get further control
364 if (q
->merge_bvec_fn
) {
366 * merge_bvec_fn() returns number of bytes it can accept
369 if (q
->merge_bvec_fn(q
, bio
, bvec
) < len
) {
370 bvec
->bv_page
= NULL
;
377 /* If we may be able to merge these biovecs, force a recount */
378 if (bio
->bi_vcnt
&& (BIOVEC_PHYS_MERGEABLE(bvec
-1, bvec
) ||
379 BIOVEC_VIRT_MERGEABLE(bvec
-1, bvec
)))
380 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
383 bio
->bi_phys_segments
++;
384 bio
->bi_hw_segments
++;
390 * bio_add_pc_page - attempt to add page to bio
391 * @bio: destination bio
393 * @len: vec entry length
394 * @offset: vec entry offset
396 * Attempt to add a page to the bio_vec maplist. This can fail for a
397 * number of reasons, such as the bio being full or target block
398 * device limitations. The target block device must allow bio's
399 * smaller than PAGE_SIZE, so it is always possible to add a single
400 * page to an empty bio. This should only be used by REQ_PC bios.
402 int bio_add_pc_page(request_queue_t
*q
, struct bio
*bio
, struct page
*page
,
403 unsigned int len
, unsigned int offset
)
405 return __bio_add_page(q
, bio
, page
, len
, offset
, q
->max_hw_sectors
);
409 * bio_add_page - attempt to add page to bio
410 * @bio: destination bio
412 * @len: vec entry length
413 * @offset: vec entry offset
415 * Attempt to add a page to the bio_vec maplist. This can fail for a
416 * number of reasons, such as the bio being full or target block
417 * device limitations. The target block device must allow bio's
418 * smaller than PAGE_SIZE, so it is always possible to add a single
419 * page to an empty bio.
421 int bio_add_page(struct bio
*bio
, struct page
*page
, unsigned int len
,
424 struct request_queue
*q
= bdev_get_queue(bio
->bi_bdev
);
425 return __bio_add_page(q
, bio
, page
, len
, offset
, q
->max_sectors
);
428 struct bio_map_data
{
429 struct bio_vec
*iovecs
;
430 void __user
*userptr
;
433 static void bio_set_map_data(struct bio_map_data
*bmd
, struct bio
*bio
)
435 memcpy(bmd
->iovecs
, bio
->bi_io_vec
, sizeof(struct bio_vec
) * bio
->bi_vcnt
);
436 bio
->bi_private
= bmd
;
439 static void bio_free_map_data(struct bio_map_data
*bmd
)
445 static struct bio_map_data
*bio_alloc_map_data(int nr_segs
)
447 struct bio_map_data
*bmd
= kmalloc(sizeof(*bmd
), GFP_KERNEL
);
452 bmd
->iovecs
= kmalloc(sizeof(struct bio_vec
) * nr_segs
, GFP_KERNEL
);
461 * bio_uncopy_user - finish previously mapped bio
462 * @bio: bio being terminated
464 * Free pages allocated from bio_copy_user() and write back data
465 * to user space in case of a read.
467 int bio_uncopy_user(struct bio
*bio
)
469 struct bio_map_data
*bmd
= bio
->bi_private
;
470 const int read
= bio_data_dir(bio
) == READ
;
471 struct bio_vec
*bvec
;
474 __bio_for_each_segment(bvec
, bio
, i
, 0) {
475 char *addr
= page_address(bvec
->bv_page
);
476 unsigned int len
= bmd
->iovecs
[i
].bv_len
;
478 if (read
&& !ret
&& copy_to_user(bmd
->userptr
, addr
, len
))
481 __free_page(bvec
->bv_page
);
484 bio_free_map_data(bmd
);
490 * bio_copy_user - copy user data to bio
491 * @q: destination block queue
492 * @uaddr: start of user address
493 * @len: length in bytes
494 * @write_to_vm: bool indicating writing to pages or not
496 * Prepares and returns a bio for indirect user io, bouncing data
497 * to/from kernel pages as necessary. Must be paired with
498 * call bio_uncopy_user() on io completion.
500 struct bio
*bio_copy_user(request_queue_t
*q
, unsigned long uaddr
,
501 unsigned int len
, int write_to_vm
)
503 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
504 unsigned long start
= uaddr
>> PAGE_SHIFT
;
505 struct bio_map_data
*bmd
;
506 struct bio_vec
*bvec
;
511 bmd
= bio_alloc_map_data(end
- start
);
513 return ERR_PTR(-ENOMEM
);
515 bmd
->userptr
= (void __user
*) uaddr
;
518 bio
= bio_alloc(GFP_KERNEL
, end
- start
);
522 bio
->bi_rw
|= (!write_to_vm
<< BIO_RW
);
526 unsigned int bytes
= PAGE_SIZE
;
531 page
= alloc_page(q
->bounce_gfp
| GFP_KERNEL
);
537 if (bio_add_pc_page(q
, bio
, page
, bytes
, 0) < bytes
) {
552 char __user
*p
= (char __user
*) uaddr
;
555 * for a write, copy in data to kernel pages
558 bio_for_each_segment(bvec
, bio
, i
) {
559 char *addr
= page_address(bvec
->bv_page
);
561 if (copy_from_user(addr
, p
, bvec
->bv_len
))
567 bio_set_map_data(bmd
, bio
);
570 bio_for_each_segment(bvec
, bio
, i
)
571 __free_page(bvec
->bv_page
);
575 bio_free_map_data(bmd
);
579 static struct bio
*__bio_map_user_iov(request_queue_t
*q
,
580 struct block_device
*bdev
,
581 struct sg_iovec
*iov
, int iov_count
,
591 for (i
= 0; i
< iov_count
; i
++) {
592 unsigned long uaddr
= (unsigned long)iov
[i
].iov_base
;
593 unsigned long len
= iov
[i
].iov_len
;
594 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
595 unsigned long start
= uaddr
>> PAGE_SHIFT
;
597 nr_pages
+= end
- start
;
599 * transfer and buffer must be aligned to at least hardsector
600 * size for now, in the future we can relax this restriction
602 if ((uaddr
& queue_dma_alignment(q
)) || (len
& queue_dma_alignment(q
)))
603 return ERR_PTR(-EINVAL
);
607 return ERR_PTR(-EINVAL
);
609 bio
= bio_alloc(GFP_KERNEL
, nr_pages
);
611 return ERR_PTR(-ENOMEM
);
614 pages
= kmalloc(nr_pages
* sizeof(struct page
*), GFP_KERNEL
);
618 memset(pages
, 0, nr_pages
* sizeof(struct page
*));
620 for (i
= 0; i
< iov_count
; i
++) {
621 unsigned long uaddr
= (unsigned long)iov
[i
].iov_base
;
622 unsigned long len
= iov
[i
].iov_len
;
623 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
624 unsigned long start
= uaddr
>> PAGE_SHIFT
;
625 const int local_nr_pages
= end
- start
;
626 const int page_limit
= cur_page
+ local_nr_pages
;
628 down_read(¤t
->mm
->mmap_sem
);
629 ret
= get_user_pages(current
, current
->mm
, uaddr
,
631 write_to_vm
, 0, &pages
[cur_page
], NULL
);
632 up_read(¤t
->mm
->mmap_sem
);
634 if (ret
< local_nr_pages
)
638 offset
= uaddr
& ~PAGE_MASK
;
639 for (j
= cur_page
; j
< page_limit
; j
++) {
640 unsigned int bytes
= PAGE_SIZE
- offset
;
651 if (bio_add_pc_page(q
, bio
, pages
[j
], bytes
, offset
) <
661 * release the pages we didn't map into the bio, if any
663 while (j
< page_limit
)
664 page_cache_release(pages
[j
++]);
670 * set data direction, and check if mapped pages need bouncing
673 bio
->bi_rw
|= (1 << BIO_RW
);
676 bio
->bi_flags
|= (1 << BIO_USER_MAPPED
);
680 for (i
= 0; i
< nr_pages
; i
++) {
683 page_cache_release(pages
[i
]);
692 * bio_map_user - map user address into bio
693 * @q: the request_queue_t for the bio
694 * @bdev: destination block device
695 * @uaddr: start of user address
696 * @len: length in bytes
697 * @write_to_vm: bool indicating writing to pages or not
699 * Map the user space address into a bio suitable for io to a block
700 * device. Returns an error pointer in case of error.
702 struct bio
*bio_map_user(request_queue_t
*q
, struct block_device
*bdev
,
703 unsigned long uaddr
, unsigned int len
, int write_to_vm
)
707 iov
.iov_base
= (void __user
*)uaddr
;
710 return bio_map_user_iov(q
, bdev
, &iov
, 1, write_to_vm
);
714 * bio_map_user_iov - map user sg_iovec table into bio
715 * @q: the request_queue_t for the bio
716 * @bdev: destination block device
718 * @iov_count: number of elements in the iovec
719 * @write_to_vm: bool indicating writing to pages or not
721 * Map the user space address into a bio suitable for io to a block
722 * device. Returns an error pointer in case of error.
724 struct bio
*bio_map_user_iov(request_queue_t
*q
, struct block_device
*bdev
,
725 struct sg_iovec
*iov
, int iov_count
,
731 bio
= __bio_map_user_iov(q
, bdev
, iov
, iov_count
, write_to_vm
);
737 * subtle -- if __bio_map_user() ended up bouncing a bio,
738 * it would normally disappear when its bi_end_io is run.
739 * however, we need it for the unmap, so grab an extra
744 for (i
= 0; i
< iov_count
; i
++)
745 len
+= iov
[i
].iov_len
;
747 if (bio
->bi_size
== len
)
751 * don't support partial mappings
753 bio_endio(bio
, bio
->bi_size
, 0);
755 return ERR_PTR(-EINVAL
);
758 static void __bio_unmap_user(struct bio
*bio
)
760 struct bio_vec
*bvec
;
764 * make sure we dirty pages we wrote to
766 __bio_for_each_segment(bvec
, bio
, i
, 0) {
767 if (bio_data_dir(bio
) == READ
)
768 set_page_dirty_lock(bvec
->bv_page
);
770 page_cache_release(bvec
->bv_page
);
777 * bio_unmap_user - unmap a bio
778 * @bio: the bio being unmapped
780 * Unmap a bio previously mapped by bio_map_user(). Must be called with
783 * bio_unmap_user() may sleep.
785 void bio_unmap_user(struct bio
*bio
)
787 __bio_unmap_user(bio
);
791 static int bio_map_kern_endio(struct bio
*bio
, unsigned int bytes_done
, int err
)
801 static struct bio
*__bio_map_kern(request_queue_t
*q
, void *data
,
802 unsigned int len
, gfp_t gfp_mask
)
804 unsigned long kaddr
= (unsigned long)data
;
805 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
806 unsigned long start
= kaddr
>> PAGE_SHIFT
;
807 const int nr_pages
= end
- start
;
811 bio
= bio_alloc(gfp_mask
, nr_pages
);
813 return ERR_PTR(-ENOMEM
);
815 offset
= offset_in_page(kaddr
);
816 for (i
= 0; i
< nr_pages
; i
++) {
817 unsigned int bytes
= PAGE_SIZE
- offset
;
825 if (bio_add_pc_page(q
, bio
, virt_to_page(data
), bytes
,
834 bio
->bi_end_io
= bio_map_kern_endio
;
839 * bio_map_kern - map kernel address into bio
840 * @q: the request_queue_t for the bio
841 * @data: pointer to buffer to map
842 * @len: length in bytes
843 * @gfp_mask: allocation flags for bio allocation
845 * Map the kernel address into a bio suitable for io to a block
846 * device. Returns an error pointer in case of error.
848 struct bio
*bio_map_kern(request_queue_t
*q
, void *data
, unsigned int len
,
853 bio
= __bio_map_kern(q
, data
, len
, gfp_mask
);
857 if (bio
->bi_size
== len
)
861 * Don't support partial mappings.
864 return ERR_PTR(-EINVAL
);
868 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
869 * for performing direct-IO in BIOs.
871 * The problem is that we cannot run set_page_dirty() from interrupt context
872 * because the required locks are not interrupt-safe. So what we can do is to
873 * mark the pages dirty _before_ performing IO. And in interrupt context,
874 * check that the pages are still dirty. If so, fine. If not, redirty them
875 * in process context.
877 * We special-case compound pages here: normally this means reads into hugetlb
878 * pages. The logic in here doesn't really work right for compound pages
879 * because the VM does not uniformly chase down the head page in all cases.
880 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
881 * handle them at all. So we skip compound pages here at an early stage.
883 * Note that this code is very hard to test under normal circumstances because
884 * direct-io pins the pages with get_user_pages(). This makes
885 * is_page_cache_freeable return false, and the VM will not clean the pages.
886 * But other code (eg, pdflush) could clean the pages if they are mapped
889 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
890 * deferred bio dirtying paths.
894 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
896 void bio_set_pages_dirty(struct bio
*bio
)
898 struct bio_vec
*bvec
= bio
->bi_io_vec
;
901 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
902 struct page
*page
= bvec
[i
].bv_page
;
904 if (page
&& !PageCompound(page
))
905 set_page_dirty_lock(page
);
909 static void bio_release_pages(struct bio
*bio
)
911 struct bio_vec
*bvec
= bio
->bi_io_vec
;
914 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
915 struct page
*page
= bvec
[i
].bv_page
;
923 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
924 * If they are, then fine. If, however, some pages are clean then they must
925 * have been written out during the direct-IO read. So we take another ref on
926 * the BIO and the offending pages and re-dirty the pages in process context.
928 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
929 * here on. It will run one page_cache_release() against each page and will
930 * run one bio_put() against the BIO.
933 static void bio_dirty_fn(void *data
);
935 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
, NULL
);
936 static DEFINE_SPINLOCK(bio_dirty_lock
);
937 static struct bio
*bio_dirty_list
;
940 * This runs in process context
942 static void bio_dirty_fn(void *data
)
947 spin_lock_irqsave(&bio_dirty_lock
, flags
);
948 bio
= bio_dirty_list
;
949 bio_dirty_list
= NULL
;
950 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
953 struct bio
*next
= bio
->bi_private
;
955 bio_set_pages_dirty(bio
);
956 bio_release_pages(bio
);
962 void bio_check_pages_dirty(struct bio
*bio
)
964 struct bio_vec
*bvec
= bio
->bi_io_vec
;
965 int nr_clean_pages
= 0;
968 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
969 struct page
*page
= bvec
[i
].bv_page
;
971 if (PageDirty(page
) || PageCompound(page
)) {
972 page_cache_release(page
);
973 bvec
[i
].bv_page
= NULL
;
979 if (nr_clean_pages
) {
982 spin_lock_irqsave(&bio_dirty_lock
, flags
);
983 bio
->bi_private
= bio_dirty_list
;
984 bio_dirty_list
= bio
;
985 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
986 schedule_work(&bio_dirty_work
);
993 * bio_endio - end I/O on a bio
995 * @bytes_done: number of bytes completed
996 * @error: error, if any
999 * bio_endio() will end I/O on @bytes_done number of bytes. This may be
1000 * just a partial part of the bio, or it may be the whole bio. bio_endio()
1001 * is the preferred way to end I/O on a bio, it takes care of decrementing
1002 * bi_size and clearing BIO_UPTODATE on error. @error is 0 on success, and
1003 * and one of the established -Exxxx (-EIO, for instance) error values in
1004 * case something went wrong. Noone should call bi_end_io() directly on
1005 * a bio unless they own it and thus know that it has an end_io function.
1007 void bio_endio(struct bio
*bio
, unsigned int bytes_done
, int error
)
1010 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
1012 if (unlikely(bytes_done
> bio
->bi_size
)) {
1013 printk("%s: want %u bytes done, only %u left\n", __FUNCTION__
,
1014 bytes_done
, bio
->bi_size
);
1015 bytes_done
= bio
->bi_size
;
1018 bio
->bi_size
-= bytes_done
;
1019 bio
->bi_sector
+= (bytes_done
>> 9);
1022 bio
->bi_end_io(bio
, bytes_done
, error
);
1025 void bio_pair_release(struct bio_pair
*bp
)
1027 if (atomic_dec_and_test(&bp
->cnt
)) {
1028 struct bio
*master
= bp
->bio1
.bi_private
;
1030 bio_endio(master
, master
->bi_size
, bp
->error
);
1031 mempool_free(bp
, bp
->bio2
.bi_private
);
1035 static int bio_pair_end_1(struct bio
* bi
, unsigned int done
, int err
)
1037 struct bio_pair
*bp
= container_of(bi
, struct bio_pair
, bio1
);
1045 bio_pair_release(bp
);
1049 static int bio_pair_end_2(struct bio
* bi
, unsigned int done
, int err
)
1051 struct bio_pair
*bp
= container_of(bi
, struct bio_pair
, bio2
);
1059 bio_pair_release(bp
);
1064 * split a bio - only worry about a bio with a single page
1067 struct bio_pair
*bio_split(struct bio
*bi
, mempool_t
*pool
, int first_sectors
)
1069 struct bio_pair
*bp
= mempool_alloc(pool
, GFP_NOIO
);
1074 BUG_ON(bi
->bi_vcnt
!= 1);
1075 BUG_ON(bi
->bi_idx
!= 0);
1076 atomic_set(&bp
->cnt
, 3);
1080 bp
->bio2
.bi_sector
+= first_sectors
;
1081 bp
->bio2
.bi_size
-= first_sectors
<< 9;
1082 bp
->bio1
.bi_size
= first_sectors
<< 9;
1084 bp
->bv1
= bi
->bi_io_vec
[0];
1085 bp
->bv2
= bi
->bi_io_vec
[0];
1086 bp
->bv2
.bv_offset
+= first_sectors
<< 9;
1087 bp
->bv2
.bv_len
-= first_sectors
<< 9;
1088 bp
->bv1
.bv_len
= first_sectors
<< 9;
1090 bp
->bio1
.bi_io_vec
= &bp
->bv1
;
1091 bp
->bio2
.bi_io_vec
= &bp
->bv2
;
1093 bp
->bio1
.bi_end_io
= bio_pair_end_1
;
1094 bp
->bio2
.bi_end_io
= bio_pair_end_2
;
1096 bp
->bio1
.bi_private
= bi
;
1097 bp
->bio2
.bi_private
= pool
;
1102 static void *bio_pair_alloc(gfp_t gfp_flags
, void *data
)
1104 return kmalloc(sizeof(struct bio_pair
), gfp_flags
);
1107 static void bio_pair_free(void *bp
, void *data
)
1114 * create memory pools for biovec's in a bio_set.
1115 * use the global biovec slabs created for general use.
1117 static int biovec_create_pools(struct bio_set
*bs
, int pool_entries
, int scale
)
1121 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
1122 struct biovec_slab
*bp
= bvec_slabs
+ i
;
1123 mempool_t
**bvp
= bs
->bvec_pools
+ i
;
1128 *bvp
= mempool_create(pool_entries
, mempool_alloc_slab
,
1129 mempool_free_slab
, bp
->slab
);
1136 static void biovec_free_pools(struct bio_set
*bs
)
1140 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
1141 mempool_t
*bvp
= bs
->bvec_pools
[i
];
1144 mempool_destroy(bvp
);
1149 void bioset_free(struct bio_set
*bs
)
1152 mempool_destroy(bs
->bio_pool
);
1154 biovec_free_pools(bs
);
1159 struct bio_set
*bioset_create(int bio_pool_size
, int bvec_pool_size
, int scale
)
1161 struct bio_set
*bs
= kmalloc(sizeof(*bs
), GFP_KERNEL
);
1166 memset(bs
, 0, sizeof(*bs
));
1167 bs
->bio_pool
= mempool_create(bio_pool_size
, mempool_alloc_slab
,
1168 mempool_free_slab
, bio_slab
);
1173 if (!biovec_create_pools(bs
, bvec_pool_size
, scale
))
1181 static void __init
biovec_init_slabs(void)
1185 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
1187 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
1189 size
= bvs
->nr_vecs
* sizeof(struct bio_vec
);
1190 bvs
->slab
= kmem_cache_create(bvs
->name
, size
, 0,
1191 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
, NULL
);
1195 static int __init
init_bio(void)
1197 int megabytes
, bvec_pool_entries
;
1198 int scale
= BIOVEC_NR_POOLS
;
1200 bio_slab
= kmem_cache_create("bio", sizeof(struct bio
), 0,
1201 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
, NULL
);
1203 biovec_init_slabs();
1205 megabytes
= nr_free_pages() >> (20 - PAGE_SHIFT
);
1208 * find out where to start scaling
1210 if (megabytes
<= 16)
1212 else if (megabytes
<= 32)
1214 else if (megabytes
<= 64)
1216 else if (megabytes
<= 96)
1218 else if (megabytes
<= 128)
1222 * scale number of entries
1224 bvec_pool_entries
= megabytes
* 2;
1225 if (bvec_pool_entries
> 256)
1226 bvec_pool_entries
= 256;
1228 fs_bio_set
= bioset_create(BIO_POOL_SIZE
, bvec_pool_entries
, scale
);
1230 panic("bio: can't allocate bios\n");
1232 bio_split_pool
= mempool_create(BIO_SPLIT_ENTRIES
,
1233 bio_pair_alloc
, bio_pair_free
, NULL
);
1234 if (!bio_split_pool
)
1235 panic("bio: can't create split pool\n");
1240 subsys_initcall(init_bio
);
1242 EXPORT_SYMBOL(bio_alloc
);
1243 EXPORT_SYMBOL(bio_put
);
1244 EXPORT_SYMBOL(bio_free
);
1245 EXPORT_SYMBOL(bio_endio
);
1246 EXPORT_SYMBOL(bio_init
);
1247 EXPORT_SYMBOL(__bio_clone
);
1248 EXPORT_SYMBOL(bio_clone
);
1249 EXPORT_SYMBOL(bio_phys_segments
);
1250 EXPORT_SYMBOL(bio_hw_segments
);
1251 EXPORT_SYMBOL(bio_add_page
);
1252 EXPORT_SYMBOL(bio_add_pc_page
);
1253 EXPORT_SYMBOL(bio_get_nr_vecs
);
1254 EXPORT_SYMBOL(bio_map_user
);
1255 EXPORT_SYMBOL(bio_unmap_user
);
1256 EXPORT_SYMBOL(bio_map_kern
);
1257 EXPORT_SYMBOL(bio_pair_release
);
1258 EXPORT_SYMBOL(bio_split
);
1259 EXPORT_SYMBOL(bio_split_pool
);
1260 EXPORT_SYMBOL(bio_copy_user
);
1261 EXPORT_SYMBOL(bio_uncopy_user
);
1262 EXPORT_SYMBOL(bioset_create
);
1263 EXPORT_SYMBOL(bioset_free
);
1264 EXPORT_SYMBOL(bio_alloc_bioset
);