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(unsigned int __nocast 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
));
109 * default destructor for a bio allocated with bio_alloc_bioset()
111 static void bio_destructor(struct bio
*bio
)
113 const int pool_idx
= BIO_POOL_IDX(bio
);
114 struct bio_set
*bs
= bio
->bi_set
;
116 BIO_BUG_ON(pool_idx
>= BIOVEC_NR_POOLS
);
118 mempool_free(bio
->bi_io_vec
, bs
->bvec_pools
[pool_idx
]);
119 mempool_free(bio
, bs
->bio_pool
);
122 inline void bio_init(struct bio
*bio
)
125 bio
->bi_flags
= 1 << BIO_UPTODATE
;
129 bio
->bi_phys_segments
= 0;
130 bio
->bi_hw_segments
= 0;
131 bio
->bi_hw_front_size
= 0;
132 bio
->bi_hw_back_size
= 0;
134 bio
->bi_max_vecs
= 0;
135 bio
->bi_end_io
= NULL
;
136 atomic_set(&bio
->bi_cnt
, 1);
137 bio
->bi_private
= NULL
;
141 * bio_alloc_bioset - allocate a bio for I/O
142 * @gfp_mask: the GFP_ mask given to the slab allocator
143 * @nr_iovecs: number of iovecs to pre-allocate
144 * @bs: the bio_set to allocate from
147 * bio_alloc_bioset will first try it's on mempool to satisfy the allocation.
148 * If %__GFP_WAIT is set then we will block on the internal pool waiting
149 * for a &struct bio to become free.
151 * allocate bio and iovecs from the memory pools specified by the
154 struct bio
*bio_alloc_bioset(unsigned int __nocast gfp_mask
, int nr_iovecs
, struct bio_set
*bs
)
156 struct bio
*bio
= mempool_alloc(bs
->bio_pool
, gfp_mask
);
159 struct bio_vec
*bvl
= NULL
;
162 if (likely(nr_iovecs
)) {
165 bvl
= bvec_alloc_bs(gfp_mask
, nr_iovecs
, &idx
, bs
);
166 if (unlikely(!bvl
)) {
167 mempool_free(bio
, bs
->bio_pool
);
171 bio
->bi_flags
|= idx
<< BIO_POOL_OFFSET
;
172 bio
->bi_max_vecs
= bvec_slabs
[idx
].nr_vecs
;
174 bio
->bi_io_vec
= bvl
;
175 bio
->bi_destructor
= bio_destructor
;
182 struct bio
*bio_alloc(unsigned int __nocast gfp_mask
, int nr_iovecs
)
184 return bio_alloc_bioset(gfp_mask
, nr_iovecs
, fs_bio_set
);
187 void zero_fill_bio(struct bio
*bio
)
193 bio_for_each_segment(bv
, bio
, i
) {
194 char *data
= bvec_kmap_irq(bv
, &flags
);
195 memset(data
, 0, bv
->bv_len
);
196 flush_dcache_page(bv
->bv_page
);
197 bvec_kunmap_irq(data
, &flags
);
200 EXPORT_SYMBOL(zero_fill_bio
);
203 * bio_put - release a reference to a bio
204 * @bio: bio to release reference to
207 * Put a reference to a &struct bio, either one you have gotten with
208 * bio_alloc or bio_get. The last put of a bio will free it.
210 void bio_put(struct bio
*bio
)
212 BIO_BUG_ON(!atomic_read(&bio
->bi_cnt
));
217 if (atomic_dec_and_test(&bio
->bi_cnt
)) {
219 bio
->bi_destructor(bio
);
223 inline int bio_phys_segments(request_queue_t
*q
, struct bio
*bio
)
225 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
226 blk_recount_segments(q
, bio
);
228 return bio
->bi_phys_segments
;
231 inline int bio_hw_segments(request_queue_t
*q
, struct bio
*bio
)
233 if (unlikely(!bio_flagged(bio
, BIO_SEG_VALID
)))
234 blk_recount_segments(q
, bio
);
236 return bio
->bi_hw_segments
;
240 * __bio_clone - clone a bio
241 * @bio: destination bio
242 * @bio_src: bio to clone
244 * Clone a &bio. Caller will own the returned bio, but not
245 * the actual data it points to. Reference count of returned
248 inline void __bio_clone(struct bio
*bio
, struct bio
*bio_src
)
250 request_queue_t
*q
= bdev_get_queue(bio_src
->bi_bdev
);
252 memcpy(bio
->bi_io_vec
, bio_src
->bi_io_vec
,
253 bio_src
->bi_max_vecs
* sizeof(struct bio_vec
));
255 bio
->bi_sector
= bio_src
->bi_sector
;
256 bio
->bi_bdev
= bio_src
->bi_bdev
;
257 bio
->bi_flags
|= 1 << BIO_CLONED
;
258 bio
->bi_rw
= bio_src
->bi_rw
;
259 bio
->bi_vcnt
= bio_src
->bi_vcnt
;
260 bio
->bi_size
= bio_src
->bi_size
;
261 bio
->bi_idx
= bio_src
->bi_idx
;
262 bio_phys_segments(q
, bio
);
263 bio_hw_segments(q
, bio
);
267 * bio_clone - clone a bio
269 * @gfp_mask: allocation priority
271 * Like __bio_clone, only also allocates the returned bio
273 struct bio
*bio_clone(struct bio
*bio
, unsigned int __nocast gfp_mask
)
275 struct bio
*b
= bio_alloc_bioset(gfp_mask
, bio
->bi_max_vecs
, fs_bio_set
);
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 request_queue_t
*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(request_queue_t
*q
, struct bio
*bio
, struct page
307 *page
, unsigned int len
, unsigned int offset
)
309 int retried_segments
= 0;
310 struct bio_vec
*bvec
;
313 * cloned bio must not modify vec list
315 if (unlikely(bio_flagged(bio
, BIO_CLONED
)))
318 if (bio
->bi_vcnt
>= bio
->bi_max_vecs
)
321 if (((bio
->bi_size
+ len
) >> 9) > q
->max_sectors
)
325 * we might lose a segment or two here, but rather that than
326 * make this too complex.
329 while (bio
->bi_phys_segments
>= q
->max_phys_segments
330 || bio
->bi_hw_segments
>= q
->max_hw_segments
331 || BIOVEC_VIRT_OVERSIZE(bio
->bi_size
)) {
333 if (retried_segments
)
336 retried_segments
= 1;
337 blk_recount_segments(q
, bio
);
341 * setup the new entry, we might clear it again later if we
342 * cannot add the page
344 bvec
= &bio
->bi_io_vec
[bio
->bi_vcnt
];
345 bvec
->bv_page
= page
;
347 bvec
->bv_offset
= offset
;
350 * if queue has other restrictions (eg varying max sector size
351 * depending on offset), it can specify a merge_bvec_fn in the
352 * queue to get further control
354 if (q
->merge_bvec_fn
) {
356 * merge_bvec_fn() returns number of bytes it can accept
359 if (q
->merge_bvec_fn(q
, bio
, bvec
) < len
) {
360 bvec
->bv_page
= NULL
;
367 /* If we may be able to merge these biovecs, force a recount */
368 if (bio
->bi_vcnt
&& (BIOVEC_PHYS_MERGEABLE(bvec
-1, bvec
) ||
369 BIOVEC_VIRT_MERGEABLE(bvec
-1, bvec
)))
370 bio
->bi_flags
&= ~(1 << BIO_SEG_VALID
);
373 bio
->bi_phys_segments
++;
374 bio
->bi_hw_segments
++;
380 * bio_add_page - attempt to add page to bio
381 * @bio: destination bio
383 * @len: vec entry length
384 * @offset: vec entry offset
386 * Attempt to add a page to the bio_vec maplist. This can fail for a
387 * number of reasons, such as the bio being full or target block
388 * device limitations. The target block device must allow bio's
389 * smaller than PAGE_SIZE, so it is always possible to add a single
390 * page to an empty bio.
392 int bio_add_page(struct bio
*bio
, struct page
*page
, unsigned int len
,
395 return __bio_add_page(bdev_get_queue(bio
->bi_bdev
), bio
, page
,
399 struct bio_map_data
{
400 struct bio_vec
*iovecs
;
401 void __user
*userptr
;
404 static void bio_set_map_data(struct bio_map_data
*bmd
, struct bio
*bio
)
406 memcpy(bmd
->iovecs
, bio
->bi_io_vec
, sizeof(struct bio_vec
) * bio
->bi_vcnt
);
407 bio
->bi_private
= bmd
;
410 static void bio_free_map_data(struct bio_map_data
*bmd
)
416 static struct bio_map_data
*bio_alloc_map_data(int nr_segs
)
418 struct bio_map_data
*bmd
= kmalloc(sizeof(*bmd
), GFP_KERNEL
);
423 bmd
->iovecs
= kmalloc(sizeof(struct bio_vec
) * nr_segs
, GFP_KERNEL
);
432 * bio_uncopy_user - finish previously mapped bio
433 * @bio: bio being terminated
435 * Free pages allocated from bio_copy_user() and write back data
436 * to user space in case of a read.
438 int bio_uncopy_user(struct bio
*bio
)
440 struct bio_map_data
*bmd
= bio
->bi_private
;
441 const int read
= bio_data_dir(bio
) == READ
;
442 struct bio_vec
*bvec
;
445 __bio_for_each_segment(bvec
, bio
, i
, 0) {
446 char *addr
= page_address(bvec
->bv_page
);
447 unsigned int len
= bmd
->iovecs
[i
].bv_len
;
449 if (read
&& !ret
&& copy_to_user(bmd
->userptr
, addr
, len
))
452 __free_page(bvec
->bv_page
);
455 bio_free_map_data(bmd
);
461 * bio_copy_user - copy user data to bio
462 * @q: destination block queue
463 * @uaddr: start of user address
464 * @len: length in bytes
465 * @write_to_vm: bool indicating writing to pages or not
467 * Prepares and returns a bio for indirect user io, bouncing data
468 * to/from kernel pages as necessary. Must be paired with
469 * call bio_uncopy_user() on io completion.
471 struct bio
*bio_copy_user(request_queue_t
*q
, unsigned long uaddr
,
472 unsigned int len
, int write_to_vm
)
474 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
475 unsigned long start
= uaddr
>> PAGE_SHIFT
;
476 struct bio_map_data
*bmd
;
477 struct bio_vec
*bvec
;
482 bmd
= bio_alloc_map_data(end
- start
);
484 return ERR_PTR(-ENOMEM
);
486 bmd
->userptr
= (void __user
*) uaddr
;
489 bio
= bio_alloc(GFP_KERNEL
, end
- start
);
493 bio
->bi_rw
|= (!write_to_vm
<< BIO_RW
);
497 unsigned int bytes
= PAGE_SIZE
;
502 page
= alloc_page(q
->bounce_gfp
| GFP_KERNEL
);
508 if (__bio_add_page(q
, bio
, page
, bytes
, 0) < bytes
) {
523 char __user
*p
= (char __user
*) uaddr
;
526 * for a write, copy in data to kernel pages
529 bio_for_each_segment(bvec
, bio
, i
) {
530 char *addr
= page_address(bvec
->bv_page
);
532 if (copy_from_user(addr
, p
, bvec
->bv_len
))
538 bio_set_map_data(bmd
, bio
);
541 bio_for_each_segment(bvec
, bio
, i
)
542 __free_page(bvec
->bv_page
);
546 bio_free_map_data(bmd
);
550 static struct bio
*__bio_map_user_iov(request_queue_t
*q
,
551 struct block_device
*bdev
,
552 struct sg_iovec
*iov
, int iov_count
,
562 for (i
= 0; i
< iov_count
; i
++) {
563 unsigned long uaddr
= (unsigned long)iov
[i
].iov_base
;
564 unsigned long len
= iov
[i
].iov_len
;
565 unsigned long end
= (uaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
566 unsigned long start
= uaddr
>> PAGE_SHIFT
;
568 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
);
578 return ERR_PTR(-EINVAL
);
580 bio
= bio_alloc(GFP_KERNEL
, nr_pages
);
582 return ERR_PTR(-ENOMEM
);
585 pages
= kmalloc(nr_pages
* sizeof(struct page
*), GFP_KERNEL
);
589 memset(pages
, 0, nr_pages
* sizeof(struct page
*));
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
;
596 const int local_nr_pages
= end
- start
;
597 const int page_limit
= cur_page
+ local_nr_pages
;
599 down_read(¤t
->mm
->mmap_sem
);
600 ret
= get_user_pages(current
, current
->mm
, uaddr
,
602 write_to_vm
, 0, &pages
[cur_page
], NULL
);
603 up_read(¤t
->mm
->mmap_sem
);
605 if (ret
< local_nr_pages
)
609 offset
= uaddr
& ~PAGE_MASK
;
610 for (j
= cur_page
; j
< page_limit
; j
++) {
611 unsigned int bytes
= PAGE_SIZE
- offset
;
622 if (__bio_add_page(q
, bio
, pages
[j
], bytes
, offset
) < bytes
)
631 * release the pages we didn't map into the bio, if any
633 while (j
< page_limit
)
634 page_cache_release(pages
[j
++]);
640 * set data direction, and check if mapped pages need bouncing
643 bio
->bi_rw
|= (1 << BIO_RW
);
646 bio
->bi_flags
|= (1 << BIO_USER_MAPPED
);
650 for (i
= 0; i
< nr_pages
; i
++) {
653 page_cache_release(pages
[i
]);
662 * bio_map_user - map user address into bio
663 * @q: the request_queue_t for the bio
664 * @bdev: destination block device
665 * @uaddr: start of user address
666 * @len: length in bytes
667 * @write_to_vm: bool indicating writing to pages or not
669 * Map the user space address into a bio suitable for io to a block
670 * device. Returns an error pointer in case of error.
672 struct bio
*bio_map_user(request_queue_t
*q
, struct block_device
*bdev
,
673 unsigned long uaddr
, unsigned int len
, int write_to_vm
)
677 iov
.iov_base
= (__user
void *)uaddr
;
680 return bio_map_user_iov(q
, bdev
, &iov
, 1, write_to_vm
);
684 * bio_map_user_iov - map user sg_iovec table into bio
685 * @q: the request_queue_t for the bio
686 * @bdev: destination block device
688 * @iov_count: number of elements in the iovec
689 * @write_to_vm: bool indicating writing to pages or not
691 * Map the user space address into a bio suitable for io to a block
692 * device. Returns an error pointer in case of error.
694 struct bio
*bio_map_user_iov(request_queue_t
*q
, struct block_device
*bdev
,
695 struct sg_iovec
*iov
, int iov_count
,
701 bio
= __bio_map_user_iov(q
, bdev
, iov
, iov_count
, write_to_vm
);
707 * subtle -- if __bio_map_user() ended up bouncing a bio,
708 * it would normally disappear when its bi_end_io is run.
709 * however, we need it for the unmap, so grab an extra
714 for (i
= 0; i
< iov_count
; i
++)
715 len
+= iov
[i
].iov_len
;
717 if (bio
->bi_size
== len
)
721 * don't support partial mappings
723 bio_endio(bio
, bio
->bi_size
, 0);
725 return ERR_PTR(-EINVAL
);
728 static void __bio_unmap_user(struct bio
*bio
)
730 struct bio_vec
*bvec
;
734 * make sure we dirty pages we wrote to
736 __bio_for_each_segment(bvec
, bio
, i
, 0) {
737 if (bio_data_dir(bio
) == READ
)
738 set_page_dirty_lock(bvec
->bv_page
);
740 page_cache_release(bvec
->bv_page
);
747 * bio_unmap_user - unmap a bio
748 * @bio: the bio being unmapped
750 * Unmap a bio previously mapped by bio_map_user(). Must be called with
753 * bio_unmap_user() may sleep.
755 void bio_unmap_user(struct bio
*bio
)
757 __bio_unmap_user(bio
);
761 static int bio_map_kern_endio(struct bio
*bio
, unsigned int bytes_done
, int err
)
771 static struct bio
*__bio_map_kern(request_queue_t
*q
, void *data
,
772 unsigned int len
, unsigned int gfp_mask
)
774 unsigned long kaddr
= (unsigned long)data
;
775 unsigned long end
= (kaddr
+ len
+ PAGE_SIZE
- 1) >> PAGE_SHIFT
;
776 unsigned long start
= kaddr
>> PAGE_SHIFT
;
777 const int nr_pages
= end
- start
;
781 bio
= bio_alloc(gfp_mask
, nr_pages
);
783 return ERR_PTR(-ENOMEM
);
785 offset
= offset_in_page(kaddr
);
786 for (i
= 0; i
< nr_pages
; i
++) {
787 unsigned int bytes
= PAGE_SIZE
- offset
;
795 if (__bio_add_page(q
, bio
, virt_to_page(data
), bytes
,
804 bio
->bi_end_io
= bio_map_kern_endio
;
809 * bio_map_kern - map kernel address into bio
810 * @q: the request_queue_t for the bio
811 * @data: pointer to buffer to map
812 * @len: length in bytes
813 * @gfp_mask: allocation flags for bio allocation
815 * Map the kernel address into a bio suitable for io to a block
816 * device. Returns an error pointer in case of error.
818 struct bio
*bio_map_kern(request_queue_t
*q
, void *data
, unsigned int len
,
819 unsigned int gfp_mask
)
823 bio
= __bio_map_kern(q
, data
, len
, gfp_mask
);
827 if (bio
->bi_size
== len
)
831 * Don't support partial mappings.
834 return ERR_PTR(-EINVAL
);
838 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
839 * for performing direct-IO in BIOs.
841 * The problem is that we cannot run set_page_dirty() from interrupt context
842 * because the required locks are not interrupt-safe. So what we can do is to
843 * mark the pages dirty _before_ performing IO. And in interrupt context,
844 * check that the pages are still dirty. If so, fine. If not, redirty them
845 * in process context.
847 * We special-case compound pages here: normally this means reads into hugetlb
848 * pages. The logic in here doesn't really work right for compound pages
849 * because the VM does not uniformly chase down the head page in all cases.
850 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
851 * handle them at all. So we skip compound pages here at an early stage.
853 * Note that this code is very hard to test under normal circumstances because
854 * direct-io pins the pages with get_user_pages(). This makes
855 * is_page_cache_freeable return false, and the VM will not clean the pages.
856 * But other code (eg, pdflush) could clean the pages if they are mapped
859 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
860 * deferred bio dirtying paths.
864 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
866 void bio_set_pages_dirty(struct bio
*bio
)
868 struct bio_vec
*bvec
= bio
->bi_io_vec
;
871 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
872 struct page
*page
= bvec
[i
].bv_page
;
874 if (page
&& !PageCompound(page
))
875 set_page_dirty_lock(page
);
879 static void bio_release_pages(struct bio
*bio
)
881 struct bio_vec
*bvec
= bio
->bi_io_vec
;
884 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
885 struct page
*page
= bvec
[i
].bv_page
;
893 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
894 * If they are, then fine. If, however, some pages are clean then they must
895 * have been written out during the direct-IO read. So we take another ref on
896 * the BIO and the offending pages and re-dirty the pages in process context.
898 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
899 * here on. It will run one page_cache_release() against each page and will
900 * run one bio_put() against the BIO.
903 static void bio_dirty_fn(void *data
);
905 static DECLARE_WORK(bio_dirty_work
, bio_dirty_fn
, NULL
);
906 static DEFINE_SPINLOCK(bio_dirty_lock
);
907 static struct bio
*bio_dirty_list
;
910 * This runs in process context
912 static void bio_dirty_fn(void *data
)
917 spin_lock_irqsave(&bio_dirty_lock
, flags
);
918 bio
= bio_dirty_list
;
919 bio_dirty_list
= NULL
;
920 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
923 struct bio
*next
= bio
->bi_private
;
925 bio_set_pages_dirty(bio
);
926 bio_release_pages(bio
);
932 void bio_check_pages_dirty(struct bio
*bio
)
934 struct bio_vec
*bvec
= bio
->bi_io_vec
;
935 int nr_clean_pages
= 0;
938 for (i
= 0; i
< bio
->bi_vcnt
; i
++) {
939 struct page
*page
= bvec
[i
].bv_page
;
941 if (PageDirty(page
) || PageCompound(page
)) {
942 page_cache_release(page
);
943 bvec
[i
].bv_page
= NULL
;
949 if (nr_clean_pages
) {
952 spin_lock_irqsave(&bio_dirty_lock
, flags
);
953 bio
->bi_private
= bio_dirty_list
;
954 bio_dirty_list
= bio
;
955 spin_unlock_irqrestore(&bio_dirty_lock
, flags
);
956 schedule_work(&bio_dirty_work
);
963 * bio_endio - end I/O on a bio
965 * @bytes_done: number of bytes completed
966 * @error: error, if any
969 * bio_endio() will end I/O on @bytes_done number of bytes. This may be
970 * just a partial part of the bio, or it may be the whole bio. bio_endio()
971 * is the preferred way to end I/O on a bio, it takes care of decrementing
972 * bi_size and clearing BIO_UPTODATE on error. @error is 0 on success, and
973 * and one of the established -Exxxx (-EIO, for instance) error values in
974 * case something went wrong. Noone should call bi_end_io() directly on
975 * a bio unless they own it and thus know that it has an end_io function.
977 void bio_endio(struct bio
*bio
, unsigned int bytes_done
, int error
)
980 clear_bit(BIO_UPTODATE
, &bio
->bi_flags
);
982 if (unlikely(bytes_done
> bio
->bi_size
)) {
983 printk("%s: want %u bytes done, only %u left\n", __FUNCTION__
,
984 bytes_done
, bio
->bi_size
);
985 bytes_done
= bio
->bi_size
;
988 bio
->bi_size
-= bytes_done
;
989 bio
->bi_sector
+= (bytes_done
>> 9);
992 bio
->bi_end_io(bio
, bytes_done
, error
);
995 void bio_pair_release(struct bio_pair
*bp
)
997 if (atomic_dec_and_test(&bp
->cnt
)) {
998 struct bio
*master
= bp
->bio1
.bi_private
;
1000 bio_endio(master
, master
->bi_size
, bp
->error
);
1001 mempool_free(bp
, bp
->bio2
.bi_private
);
1005 static int bio_pair_end_1(struct bio
* bi
, unsigned int done
, int err
)
1007 struct bio_pair
*bp
= container_of(bi
, struct bio_pair
, bio1
);
1015 bio_pair_release(bp
);
1019 static int bio_pair_end_2(struct bio
* bi
, unsigned int done
, int err
)
1021 struct bio_pair
*bp
= container_of(bi
, struct bio_pair
, bio2
);
1029 bio_pair_release(bp
);
1034 * split a bio - only worry about a bio with a single page
1037 struct bio_pair
*bio_split(struct bio
*bi
, mempool_t
*pool
, int first_sectors
)
1039 struct bio_pair
*bp
= mempool_alloc(pool
, GFP_NOIO
);
1044 BUG_ON(bi
->bi_vcnt
!= 1);
1045 BUG_ON(bi
->bi_idx
!= 0);
1046 atomic_set(&bp
->cnt
, 3);
1050 bp
->bio2
.bi_sector
+= first_sectors
;
1051 bp
->bio2
.bi_size
-= first_sectors
<< 9;
1052 bp
->bio1
.bi_size
= first_sectors
<< 9;
1054 bp
->bv1
= bi
->bi_io_vec
[0];
1055 bp
->bv2
= bi
->bi_io_vec
[0];
1056 bp
->bv2
.bv_offset
+= first_sectors
<< 9;
1057 bp
->bv2
.bv_len
-= first_sectors
<< 9;
1058 bp
->bv1
.bv_len
= first_sectors
<< 9;
1060 bp
->bio1
.bi_io_vec
= &bp
->bv1
;
1061 bp
->bio2
.bi_io_vec
= &bp
->bv2
;
1063 bp
->bio1
.bi_end_io
= bio_pair_end_1
;
1064 bp
->bio2
.bi_end_io
= bio_pair_end_2
;
1066 bp
->bio1
.bi_private
= bi
;
1067 bp
->bio2
.bi_private
= pool
;
1072 static void *bio_pair_alloc(unsigned int __nocast gfp_flags
, void *data
)
1074 return kmalloc(sizeof(struct bio_pair
), gfp_flags
);
1077 static void bio_pair_free(void *bp
, void *data
)
1084 * create memory pools for biovec's in a bio_set.
1085 * use the global biovec slabs created for general use.
1087 static int biovec_create_pools(struct bio_set
*bs
, int pool_entries
, int scale
)
1091 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
1092 struct biovec_slab
*bp
= bvec_slabs
+ i
;
1093 mempool_t
**bvp
= bs
->bvec_pools
+ i
;
1098 *bvp
= mempool_create(pool_entries
, mempool_alloc_slab
,
1099 mempool_free_slab
, 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
, int scale
)
1131 struct bio_set
*bs
= kmalloc(sizeof(*bs
), GFP_KERNEL
);
1136 memset(bs
, 0, sizeof(*bs
));
1137 bs
->bio_pool
= mempool_create(bio_pool_size
, mempool_alloc_slab
,
1138 mempool_free_slab
, bio_slab
);
1143 if (!biovec_create_pools(bs
, bvec_pool_size
, scale
))
1151 static void __init
biovec_init_slabs(void)
1155 for (i
= 0; i
< BIOVEC_NR_POOLS
; i
++) {
1157 struct biovec_slab
*bvs
= bvec_slabs
+ i
;
1159 size
= bvs
->nr_vecs
* sizeof(struct bio_vec
);
1160 bvs
->slab
= kmem_cache_create(bvs
->name
, size
, 0,
1161 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
, NULL
);
1165 static int __init
init_bio(void)
1167 int megabytes
, bvec_pool_entries
;
1168 int scale
= BIOVEC_NR_POOLS
;
1170 bio_slab
= kmem_cache_create("bio", sizeof(struct bio
), 0,
1171 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
, NULL
);
1173 biovec_init_slabs();
1175 megabytes
= nr_free_pages() >> (20 - PAGE_SHIFT
);
1178 * find out where to start scaling
1180 if (megabytes
<= 16)
1182 else if (megabytes
<= 32)
1184 else if (megabytes
<= 64)
1186 else if (megabytes
<= 96)
1188 else if (megabytes
<= 128)
1192 * scale number of entries
1194 bvec_pool_entries
= megabytes
* 2;
1195 if (bvec_pool_entries
> 256)
1196 bvec_pool_entries
= 256;
1198 fs_bio_set
= bioset_create(BIO_POOL_SIZE
, bvec_pool_entries
, scale
);
1200 panic("bio: can't allocate bios\n");
1202 bio_split_pool
= mempool_create(BIO_SPLIT_ENTRIES
,
1203 bio_pair_alloc
, bio_pair_free
, NULL
);
1204 if (!bio_split_pool
)
1205 panic("bio: can't create split pool\n");
1210 subsys_initcall(init_bio
);
1212 EXPORT_SYMBOL(bio_alloc
);
1213 EXPORT_SYMBOL(bio_put
);
1214 EXPORT_SYMBOL(bio_endio
);
1215 EXPORT_SYMBOL(bio_init
);
1216 EXPORT_SYMBOL(__bio_clone
);
1217 EXPORT_SYMBOL(bio_clone
);
1218 EXPORT_SYMBOL(bio_phys_segments
);
1219 EXPORT_SYMBOL(bio_hw_segments
);
1220 EXPORT_SYMBOL(bio_add_page
);
1221 EXPORT_SYMBOL(bio_get_nr_vecs
);
1222 EXPORT_SYMBOL(bio_map_user
);
1223 EXPORT_SYMBOL(bio_unmap_user
);
1224 EXPORT_SYMBOL(bio_map_kern
);
1225 EXPORT_SYMBOL(bio_pair_release
);
1226 EXPORT_SYMBOL(bio_split
);
1227 EXPORT_SYMBOL(bio_split_pool
);
1228 EXPORT_SYMBOL(bio_copy_user
);
1229 EXPORT_SYMBOL(bio_uncopy_user
);
1230 EXPORT_SYMBOL(bioset_create
);
1231 EXPORT_SYMBOL(bioset_free
);
1232 EXPORT_SYMBOL(bio_alloc_bioset
);