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>
29 #define BIO_POOL_SIZE 256
31 static kmem_cache_t
*bio_slab
;
33 #define BIOVEC_NR_POOLS 6
36 * a small number of entries is fine, not going to be performance critical.
37 * basically we just need to survive
39 #define BIO_SPLIT_ENTRIES 8
40 mempool_t
*bio_split_pool
;
49 * if you change this list, also change bvec_alloc or things will
50 * break badly! cannot be bigger than what you can fit into an
54 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
55 static struct biovec_slab bvec_slabs
[BIOVEC_NR_POOLS
] __read_mostly
= {
56 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES
),
61 * bio_set is used to allow other portions of the IO system to
62 * allocate their own private memory pools for bio and iovec structures.
63 * These memory pools in turn all allocate from the bio_slab
64 * and the bvec_slabs[].
68 mempool_t
*bvec_pools
[BIOVEC_NR_POOLS
];
72 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
73 * IO code that does not need private memory pools.
75 static struct bio_set
*fs_bio_set
;
77 static inline struct bio_vec
*bvec_alloc_bs(unsigned int __nocast gfp_mask
, int nr
, unsigned long *idx
, struct bio_set
*bs
)
80 struct biovec_slab
*bp
;
83 * see comment near bvec_array define!
86 case 1 : *idx
= 0; break;
87 case 2 ... 4: *idx
= 1; break;
88 case 5 ... 16: *idx
= 2; break;
89 case 17 ... 64: *idx
= 3; break;
90 case 65 ... 128: *idx
= 4; break;
91 case 129 ... BIO_MAX_PAGES
: *idx
= 5; break;
96 * idx now points to the pool we want to allocate from
99 bp
= bvec_slabs
+ *idx
;
100 bvl
= mempool_alloc(bs
->bvec_pools
[*idx
], gfp_mask
);
102 memset(bvl
, 0, bp
->nr_vecs
* sizeof(struct bio_vec
));
107 void bio_free(struct bio
*bio
, struct bio_set
*bio_set
)
109 const int pool_idx
= BIO_POOL_IDX(bio
);
111 BIO_BUG_ON(pool_idx
>= BIOVEC_NR_POOLS
);
113 mempool_free(bio
->bi_io_vec
, bio_set
->bvec_pools
[pool_idx
]);
114 mempool_free(bio
, bio_set
->bio_pool
);
118 * default destructor for a bio allocated with bio_alloc_bioset()
120 static void bio_fs_destructor(struct bio
*bio
)
122 bio_free(bio
, fs_bio_set
);
125 inline void bio_init(struct bio
*bio
)
128 bio
->bi_flags
= 1 << BIO_UPTODATE
;
132 bio
->bi_phys_segments
= 0;
133 bio
->bi_hw_segments
= 0;
134 bio
->bi_hw_front_size
= 0;
135 bio
->bi_hw_back_size
= 0;
137 bio
->bi_max_vecs
= 0;
138 bio
->bi_end_io
= NULL
;
139 atomic_set(&bio
->bi_cnt
, 1);
140 bio
->bi_private
= NULL
;
144 * bio_alloc_bioset - allocate a bio for I/O
145 * @gfp_mask: the GFP_ mask given to the slab allocator
146 * @nr_iovecs: number of iovecs to pre-allocate
147 * @bs: the bio_set to allocate from
150 * bio_alloc_bioset will first try it's on mempool to satisfy the allocation.
151 * If %__GFP_WAIT is set then we will block on the internal pool waiting
152 * for a &struct bio to become free.
154 * allocate bio and iovecs from the memory pools specified by the
157 struct bio
*bio_alloc_bioset(unsigned int __nocast gfp_mask
, int nr_iovecs
, struct bio_set
*bs
)
159 struct bio
*bio
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
162 struct bio_vec
*bvl
= NULL
;
165 if (likely(nr_iovecs
)) {
168 bvl
= bvec_alloc_bs(gfp_mask
, nr_iovecs
, &idx
, bs
);
169 if (unlikely(!bvl
)) {
170 mempool_free(bio
, bs
->bio_pool
);
174 bio
->bi_flags
|= idx
<< BIO_POOL_OFFSET
;
175 bio
->bi_max_vecs
= bvec_slabs
[idx
].nr_vecs
;
177 bio
->bi_io_vec
= bvl
;
183 struct bio
*bio_alloc(unsigned int __nocast gfp_mask
, int nr_iovecs
)
185 struct bio
*bio
= bio_alloc_bioset(gfp_mask
, nr_iovecs
, fs_bio_set
);
188 bio
->bi_destructor
= bio_fs_destructor
;
193 void zero_fill_bio(struct bio
*bio
)
199 bio_for_each_segment(bv
, bio
, i
) {
200 char *data
= bvec_kmap_irq(bv
, &flags
);
201 memset(data
, 0, bv
->bv_len
);
202 flush_dcache_page(bv
->bv_page
);
203 bvec_kunmap_irq(data
, &flags
);
206 EXPORT_SYMBOL(zero_fill_bio
);
209 * bio_put - release a reference to a bio
210 * @bio: bio to release reference to
213 * Put a reference to a &struct bio, either one you have gotten with
214 * bio_alloc or bio_get. The last put of a bio will free it.
216 void bio_put(struct bio
*bio
)
218 BIO_BUG_ON(!atomic_read(&bio
->bi_cnt
));
223 if (atomic_dec_and_test(&bio
->bi_cnt
)) {
225 bio
->bi_destructor(bio
);
229 inline int bio_phys_segments(request_queue_t
*q
, struct bio
*bio
)
231 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
232 blk_recount_segments(q
, bio
);
234 return bio
->bi_phys_segments
;
237 inline int bio_hw_segments(request_queue_t
*q
, struct bio
*bio
)
239 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
240 blk_recount_segments(q
, bio
);
242 return bio
->bi_hw_segments
;
246 * __bio_clone - clone a bio
247 * @bio: destination bio
248 * @bio_src: bio to clone
250 * Clone a &bio. Caller will own the returned bio, but not
251 * the actual data it points to. Reference count of returned
254 inline void __bio_clone(struct bio
*bio
, struct bio
*bio_src
)
256 request_queue_t
*q
= bdev_get_queue(bio_src
->bi_bdev
);
258 memcpy(bio
->bi_io_vec
, bio_src
->bi_io_vec
,
259 bio_src
->bi_max_vecs
* sizeof(struct bio_vec
));
261 bio
->bi_sector
= bio_src
->bi_sector
;
262 bio
->bi_bdev
= bio_src
->bi_bdev
;
263 bio
->bi_flags
|= 1 << BIO_CLONED
;
264 bio
->bi_rw
= bio_src
->bi_rw
;
265 bio
->bi_vcnt
= bio_src
->bi_vcnt
;
266 bio
->bi_size
= bio_src
->bi_size
;
267 bio
->bi_idx
= bio_src
->bi_idx
;
268 bio_phys_segments(q
, bio
);
269 bio_hw_segments(q
, bio
);
273 * bio_clone - clone a bio
275 * @gfp_mask: allocation priority
277 * Like __bio_clone, only also allocates the returned bio
279 struct bio
*bio_clone(struct bio
*bio
, unsigned int __nocast gfp_mask
)
281 struct bio
*b
= bio_alloc_bioset(gfp_mask
, bio
->bi_max_vecs
, fs_bio_set
);
284 b
->bi_destructor
= bio_fs_destructor
;
292 * bio_get_nr_vecs - return approx number of vecs
295 * Return the approximate number of pages we can send to this target.
296 * There's no guarantee that you will be able to fit this number of pages
297 * into a bio, it does not account for dynamic restrictions that vary
300 int bio_get_nr_vecs(struct block_device
*bdev
)
302 request_queue_t
*q
= bdev_get_queue(bdev
);
305 nr_pages
= ((q
->max_sectors
<< 9) + PAGE_SIZE
- 1) >> PAGE_SHIFT
;
306 if (nr_pages
> q
->max_phys_segments
)
307 nr_pages
= q
->max_phys_segments
;
308 if (nr_pages
> q
->max_hw_segments
)
309 nr_pages
= q
->max_hw_segments
;
314 static int __bio_add_page(request_queue_t
*q
, struct bio
*bio
, struct page
315 *page
, unsigned int len
, unsigned int offset
)
317 int retried_segments
= 0;
318 struct bio_vec
*bvec
;
321 * cloned bio must not modify vec list
323 if (unlikely(bio_flagged(bio
, BIO_CLONED
)))
326 if (bio
->bi_vcnt
>= bio
->bi_max_vecs
)
329 if (((bio
->bi_size
+ len
) >> 9) > q
->max_sectors
)
333 * we might lose a segment or two here, but rather that than
334 * make this too complex.
337 while (bio
->bi_phys_segments
>= q
->max_phys_segments
338 || bio
->bi_hw_segments
>= q
->max_hw_segments
339 || BIOVEC_VIRT_OVERSIZE(bio
->bi_size
)) {
341 if (retried_segments
)
344 retried_segments
= 1;
345 blk_recount_segments(q
, bio
);
349 * setup the new entry, we might clear it again later if we
350 * cannot add the page
352 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
353 bvec
->bv_page
= page
;
355 bvec
->bv_offset
= offset
;
358 * if queue has other restrictions (eg varying max sector size
359 * depending on offset), it can specify a merge_bvec_fn in the
360 * queue to get further control
362 if (q
->merge_bvec_fn
) {
364 * merge_bvec_fn() returns number of bytes it can accept
367 if (q
->merge_bvec_fn(q
, bio
, bvec
) < len
) {
368 bvec
->bv_page
= NULL
;
375 /* If we may be able to merge these biovecs, force a recount */
376 if (bio
->bi_vcnt
&& (BIOVEC_PHYS_MERGEABLE(bvec
-1, bvec
) ||
377 BIOVEC_VIRT_MERGEABLE(bvec
-1, bvec
)))
378 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
381 bio
->bi_phys_segments
++;
382 bio
->bi_hw_segments
++;
388 * bio_add_page - attempt to add page to bio
389 * @bio: destination bio
391 * @len: vec entry length
392 * @offset: vec entry offset
394 * Attempt to add a page to the bio_vec maplist. This can fail for a
395 * number of reasons, such as the bio being full or target block
396 * device limitations. The target block device must allow bio's
397 * smaller than PAGE_SIZE, so it is always possible to add a single
398 * page to an empty bio.
400 int bio_add_page(struct bio
*bio
, struct page
*page
, unsigned int len
,
403 return __bio_add_page(bdev_get_queue(bio
->bi_bdev
), bio
, page
,
407 struct bio_map_data
{
408 struct bio_vec
*iovecs
;
409 void __user
*userptr
;
412 static void bio_set_map_data(struct bio_map_data
*bmd
, struct bio
*bio
)
414 memcpy(bmd
->iovecs
, bio
->bi_io_vec
, sizeof(struct bio_vec
) * bio
->bi_vcnt
);
415 bio
->bi_private
= bmd
;
418 static void bio_free_map_data(struct bio_map_data
*bmd
)
424 static struct bio_map_data
*bio_alloc_map_data(int nr_segs
)
426 struct bio_map_data
*bmd
= kmalloc(sizeof(*bmd
), GFP_KERNEL
);
431 bmd
->iovecs
= kmalloc(sizeof(struct bio_vec
) * nr_segs
, GFP_KERNEL
);
440 * bio_uncopy_user - finish previously mapped bio
441 * @bio: bio being terminated
443 * Free pages allocated from bio_copy_user() and write back data
444 * to user space in case of a read.
446 int bio_uncopy_user(struct bio
*bio
)
448 struct bio_map_data
*bmd
= bio
->bi_private
;
449 const int read
= bio_data_dir(bio
) == READ
;
450 struct bio_vec
*bvec
;
453 __bio_for_each_segment(bvec
, bio
, i
, 0) {
454 char *addr
= page_address(bvec
->bv_page
);
455 unsigned int len
= bmd
->iovecs
[i
].bv_len
;
457 if (read
&& !ret
&& copy_to_user(bmd
->userptr
, addr
, len
))
460 __free_page(bvec
->bv_page
);
463 bio_free_map_data(bmd
);
469 * bio_copy_user - copy user data to bio
470 * @q: destination block queue
471 * @uaddr: start of user address
472 * @len: length in bytes
473 * @write_to_vm: bool indicating writing to pages or not
475 * Prepares and returns a bio for indirect user io, bouncing data
476 * to/from kernel pages as necessary. Must be paired with
477 * call bio_uncopy_user() on io completion.
479 struct bio
*bio_copy_user(request_queue_t
*q
, unsigned long uaddr
,
480 unsigned int len
, int write_to_vm
)
482 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
483 unsigned long start
= uaddr
>> PAGE_SHIFT
;
484 struct bio_map_data
*bmd
;
485 struct bio_vec
*bvec
;
490 bmd
= bio_alloc_map_data(end
- start
);
492 return ERR_PTR(-ENOMEM
);
494 bmd
->userptr
= (void __user
*) uaddr
;
497 bio
= bio_alloc(GFP_KERNEL
, end
- start
);
501 bio
->bi_rw
|= (!write_to_vm
<< BIO_RW
);
505 unsigned int bytes
= PAGE_SIZE
;
510 page
= alloc_page(q
->bounce_gfp
| GFP_KERNEL
);
516 if (__bio_add_page(q
, bio
, page
, bytes
, 0) < bytes
) {
531 char __user
*p
= (char __user
*) uaddr
;
534 * for a write, copy in data to kernel pages
537 bio_for_each_segment(bvec
, bio
, i
) {
538 char *addr
= page_address(bvec
->bv_page
);
540 if (copy_from_user(addr
, p
, bvec
->bv_len
))
546 bio_set_map_data(bmd
, bio
);
549 bio_for_each_segment(bvec
, bio
, i
)
550 __free_page(bvec
->bv_page
);
554 bio_free_map_data(bmd
);
558 static struct bio
*__bio_map_user(request_queue_t
*q
, struct block_device
*bdev
,
559 unsigned long uaddr
, unsigned int len
,
562 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
563 unsigned long start
= uaddr
>> PAGE_SHIFT
;
564 const int nr_pages
= end
- start
;
570 * transfer and buffer must be aligned to at least hardsector
571 * size for now, in the future we can relax this restriction
573 if ((uaddr
& queue_dma_alignment(q
)) || (len
& queue_dma_alignment(q
)))
574 return ERR_PTR(-EINVAL
);
576 bio
= bio_alloc(GFP_KERNEL
, nr_pages
);
578 return ERR_PTR(-ENOMEM
);
581 pages
= kmalloc(nr_pages
* sizeof(struct page
*), GFP_KERNEL
);
585 down_read(¤t
->mm
->mmap_sem
);
586 ret
= get_user_pages(current
, current
->mm
, uaddr
, nr_pages
,
587 write_to_vm
, 0, pages
, NULL
);
588 up_read(¤t
->mm
->mmap_sem
);
595 offset
= uaddr
& ~PAGE_MASK
;
596 for (i
= 0; i
< nr_pages
; i
++) {
597 unsigned int bytes
= PAGE_SIZE
- offset
;
608 if (__bio_add_page(q
, bio
, pages
[i
], bytes
, offset
) < bytes
)
616 * release the pages we didn't map into the bio, if any
619 page_cache_release(pages
[i
++]);
624 * set data direction, and check if mapped pages need bouncing
627 bio
->bi_rw
|= (1 << BIO_RW
);
629 bio
->bi_flags
|= (1 << BIO_USER_MAPPED
);
638 * bio_map_user - map user address into bio
639 * @q: the request_queue_t for the bio
640 * @bdev: destination block device
641 * @uaddr: start of user address
642 * @len: length in bytes
643 * @write_to_vm: bool indicating writing to pages or not
645 * Map the user space address into a bio suitable for io to a block
646 * device. Returns an error pointer in case of error.
648 struct bio
*bio_map_user(request_queue_t
*q
, struct block_device
*bdev
,
649 unsigned long uaddr
, unsigned int len
, int write_to_vm
)
653 bio
= __bio_map_user(q
, bdev
, uaddr
, len
, write_to_vm
);
659 * subtle -- if __bio_map_user() ended up bouncing a bio,
660 * it would normally disappear when its bi_end_io is run.
661 * however, we need it for the unmap, so grab an extra
666 if (bio
->bi_size
== len
)
670 * don't support partial mappings
672 bio_endio(bio
, bio
->bi_size
, 0);
674 return ERR_PTR(-EINVAL
);
677 static void __bio_unmap_user(struct bio
*bio
)
679 struct bio_vec
*bvec
;
683 * make sure we dirty pages we wrote to
685 __bio_for_each_segment(bvec
, bio
, i
, 0) {
686 if (bio_data_dir(bio
) == READ
)
687 set_page_dirty_lock(bvec
->bv_page
);
689 page_cache_release(bvec
->bv_page
);
696 * bio_unmap_user - unmap a bio
697 * @bio: the bio being unmapped
699 * Unmap a bio previously mapped by bio_map_user(). Must be called with
702 * bio_unmap_user() may sleep.
704 void bio_unmap_user(struct bio
*bio
)
706 __bio_unmap_user(bio
);
711 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
712 * for performing direct-IO in BIOs.
714 * The problem is that we cannot run set_page_dirty() from interrupt context
715 * because the required locks are not interrupt-safe. So what we can do is to
716 * mark the pages dirty _before_ performing IO. And in interrupt context,
717 * check that the pages are still dirty. If so, fine. If not, redirty them
718 * in process context.
720 * We special-case compound pages here: normally this means reads into hugetlb
721 * pages. The logic in here doesn't really work right for compound pages
722 * because the VM does not uniformly chase down the head page in all cases.
723 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
724 * handle them at all. So we skip compound pages here at an early stage.
726 * Note that this code is very hard to test under normal circumstances because
727 * direct-io pins the pages with get_user_pages(). This makes
728 * is_page_cache_freeable return false, and the VM will not clean the pages.
729 * But other code (eg, pdflush) could clean the pages if they are mapped
732 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
733 * deferred bio dirtying paths.
737 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
739 void bio_set_pages_dirty(struct bio
*bio
)
741 struct bio_vec
*bvec
= bio
->bi_io_vec
;
744 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
745 struct page
*page
= bvec
[i
].bv_page
;
747 if (page
&& !PageCompound(page
))
748 set_page_dirty_lock(page
);
752 static void bio_release_pages(struct bio
*bio
)
754 struct bio_vec
*bvec
= bio
->bi_io_vec
;
757 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
758 struct page
*page
= bvec
[i
].bv_page
;
766 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
767 * If they are, then fine. If, however, some pages are clean then they must
768 * have been written out during the direct-IO read. So we take another ref on
769 * the BIO and the offending pages and re-dirty the pages in process context.
771 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
772 * here on. It will run one page_cache_release() against each page and will
773 * run one bio_put() against the BIO.
776 static void bio_dirty_fn(void *data
);
778 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
, NULL
);
779 static DEFINE_SPINLOCK(bio_dirty_lock
);
780 static struct bio
*bio_dirty_list
;
783 * This runs in process context
785 static void bio_dirty_fn(void *data
)
790 spin_lock_irqsave(&bio_dirty_lock
, flags
);
791 bio
= bio_dirty_list
;
792 bio_dirty_list
= NULL
;
793 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
796 struct bio
*next
= bio
->bi_private
;
798 bio_set_pages_dirty(bio
);
799 bio_release_pages(bio
);
805 void bio_check_pages_dirty(struct bio
*bio
)
807 struct bio_vec
*bvec
= bio
->bi_io_vec
;
808 int nr_clean_pages
= 0;
811 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
812 struct page
*page
= bvec
[i
].bv_page
;
814 if (PageDirty(page
) || PageCompound(page
)) {
815 page_cache_release(page
);
816 bvec
[i
].bv_page
= NULL
;
822 if (nr_clean_pages
) {
825 spin_lock_irqsave(&bio_dirty_lock
, flags
);
826 bio
->bi_private
= bio_dirty_list
;
827 bio_dirty_list
= bio
;
828 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
829 schedule_work(&bio_dirty_work
);
836 * bio_endio - end I/O on a bio
838 * @bytes_done: number of bytes completed
839 * @error: error, if any
842 * bio_endio() will end I/O on @bytes_done number of bytes. This may be
843 * just a partial part of the bio, or it may be the whole bio. bio_endio()
844 * is the preferred way to end I/O on a bio, it takes care of decrementing
845 * bi_size and clearing BIO_UPTODATE on error. @error is 0 on success, and
846 * and one of the established -Exxxx (-EIO, for instance) error values in
847 * case something went wrong. Noone should call bi_end_io() directly on
848 * a bio unless they own it and thus know that it has an end_io function.
850 void bio_endio(struct bio
*bio
, unsigned int bytes_done
, int error
)
853 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
855 if (unlikely(bytes_done
> bio
->bi_size
)) {
856 printk("%s: want %u bytes done, only %u left\n", __FUNCTION__
,
857 bytes_done
, bio
->bi_size
);
858 bytes_done
= bio
->bi_size
;
861 bio
->bi_size
-= bytes_done
;
862 bio
->bi_sector
+= (bytes_done
>> 9);
865 bio
->bi_end_io(bio
, bytes_done
, error
);
868 void bio_pair_release(struct bio_pair
*bp
)
870 if (atomic_dec_and_test(&bp
->cnt
)) {
871 struct bio
*master
= bp
->bio1
.bi_private
;
873 bio_endio(master
, master
->bi_size
, bp
->error
);
874 mempool_free(bp
, bp
->bio2
.bi_private
);
878 static int bio_pair_end_1(struct bio
* bi
, unsigned int done
, int err
)
880 struct bio_pair
*bp
= container_of(bi
, struct bio_pair
, bio1
);
888 bio_pair_release(bp
);
892 static int bio_pair_end_2(struct bio
* bi
, unsigned int done
, int err
)
894 struct bio_pair
*bp
= container_of(bi
, struct bio_pair
, bio2
);
902 bio_pair_release(bp
);
907 * split a bio - only worry about a bio with a single page
910 struct bio_pair
*bio_split(struct bio
*bi
, mempool_t
*pool
, int first_sectors
)
912 struct bio_pair
*bp
= mempool_alloc(pool
, GFP_NOIO
);
917 BUG_ON(bi
->bi_vcnt
!= 1);
918 BUG_ON(bi
->bi_idx
!= 0);
919 atomic_set(&bp
->cnt
, 3);
923 bp
->bio2
.bi_sector
+= first_sectors
;
924 bp
->bio2
.bi_size
-= first_sectors
<< 9;
925 bp
->bio1
.bi_size
= first_sectors
<< 9;
927 bp
->bv1
= bi
->bi_io_vec
[0];
928 bp
->bv2
= bi
->bi_io_vec
[0];
929 bp
->bv2
.bv_offset
+= first_sectors
<< 9;
930 bp
->bv2
.bv_len
-= first_sectors
<< 9;
931 bp
->bv1
.bv_len
= first_sectors
<< 9;
933 bp
->bio1
.bi_io_vec
= &bp
->bv1
;
934 bp
->bio2
.bi_io_vec
= &bp
->bv2
;
936 bp
->bio1
.bi_end_io
= bio_pair_end_1
;
937 bp
->bio2
.bi_end_io
= bio_pair_end_2
;
939 bp
->bio1
.bi_private
= bi
;
940 bp
->bio2
.bi_private
= pool
;
945 static void *bio_pair_alloc(unsigned int __nocast gfp_flags
, void *data
)
947 return kmalloc(sizeof(struct bio_pair
), gfp_flags
);
950 static void bio_pair_free(void *bp
, void *data
)
957 * create memory pools for biovec's in a bio_set.
958 * use the global biovec slabs created for general use.
960 static int biovec_create_pools(struct bio_set
*bs
, int pool_entries
, int scale
)
964 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
965 struct biovec_slab
*bp
= bvec_slabs
+ i
;
966 mempool_t
**bvp
= bs
->bvec_pools
+ i
;
971 *bvp
= mempool_create(pool_entries
, mempool_alloc_slab
,
972 mempool_free_slab
, bp
->slab
);
979 static void biovec_free_pools(struct bio_set
*bs
)
983 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
984 mempool_t
*bvp
= bs
->bvec_pools
[i
];
987 mempool_destroy(bvp
);
992 void bioset_free(struct bio_set
*bs
)
995 mempool_destroy(bs
->bio_pool
);
997 biovec_free_pools(bs
);
1002 struct bio_set
*bioset_create(int bio_pool_size
, int bvec_pool_size
, int scale
)
1004 struct bio_set
*bs
= kmalloc(sizeof(*bs
), GFP_KERNEL
);
1009 memset(bs
, 0, sizeof(*bs
));
1010 bs
->bio_pool
= mempool_create(bio_pool_size
, mempool_alloc_slab
,
1011 mempool_free_slab
, bio_slab
);
1016 if (!biovec_create_pools(bs
, bvec_pool_size
, scale
))
1024 static void __init
biovec_init_slabs(void)
1028 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
1030 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
1032 size
= bvs
->nr_vecs
* sizeof(struct bio_vec
);
1033 bvs
->slab
= kmem_cache_create(bvs
->name
, size
, 0,
1034 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
, NULL
);
1038 static int __init
init_bio(void)
1040 int megabytes
, bvec_pool_entries
;
1041 int scale
= BIOVEC_NR_POOLS
;
1043 bio_slab
= kmem_cache_create("bio", sizeof(struct bio
), 0,
1044 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
, NULL
);
1046 biovec_init_slabs();
1048 megabytes
= nr_free_pages() >> (20 - PAGE_SHIFT
);
1051 * find out where to start scaling
1053 if (megabytes
<= 16)
1055 else if (megabytes
<= 32)
1057 else if (megabytes
<= 64)
1059 else if (megabytes
<= 96)
1061 else if (megabytes
<= 128)
1065 * scale number of entries
1067 bvec_pool_entries
= megabytes
* 2;
1068 if (bvec_pool_entries
> 256)
1069 bvec_pool_entries
= 256;
1071 fs_bio_set
= bioset_create(BIO_POOL_SIZE
, bvec_pool_entries
, scale
);
1073 panic("bio: can't allocate bios\n");
1075 bio_split_pool
= mempool_create(BIO_SPLIT_ENTRIES
,
1076 bio_pair_alloc
, bio_pair_free
, NULL
);
1077 if (!bio_split_pool
)
1078 panic("bio: can't create split pool\n");
1083 subsys_initcall(init_bio
);
1085 EXPORT_SYMBOL(bio_alloc
);
1086 EXPORT_SYMBOL(bio_put
);
1087 EXPORT_SYMBOL(bio_free
);
1088 EXPORT_SYMBOL(bio_endio
);
1089 EXPORT_SYMBOL(bio_init
);
1090 EXPORT_SYMBOL(__bio_clone
);
1091 EXPORT_SYMBOL(bio_clone
);
1092 EXPORT_SYMBOL(bio_phys_segments
);
1093 EXPORT_SYMBOL(bio_hw_segments
);
1094 EXPORT_SYMBOL(bio_add_page
);
1095 EXPORT_SYMBOL(bio_get_nr_vecs
);
1096 EXPORT_SYMBOL(bio_map_user
);
1097 EXPORT_SYMBOL(bio_unmap_user
);
1098 EXPORT_SYMBOL(bio_pair_release
);
1099 EXPORT_SYMBOL(bio_split
);
1100 EXPORT_SYMBOL(bio_split_pool
);
1101 EXPORT_SYMBOL(bio_copy_user
);
1102 EXPORT_SYMBOL(bio_uncopy_user
);
1103 EXPORT_SYMBOL(bioset_create
);
1104 EXPORT_SYMBOL(bioset_free
);
1105 EXPORT_SYMBOL(bio_alloc_bioset
);