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] = {
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, bio_src->bi_max_vecs * sizeof(struct bio_vec));
254 bio->bi_sector = bio_src->bi_sector;
255 bio->bi_bdev = bio_src->bi_bdev;
256 bio->bi_flags |= 1 << BIO_CLONED;
257 bio->bi_rw = bio_src->bi_rw;
260 * notes -- maybe just leave bi_idx alone. assume identical mapping
263 bio->bi_vcnt = bio_src->bi_vcnt;
264 bio->bi_size = bio_src->bi_size;
265 bio_phys_segments(q, bio);
266 bio_hw_segments(q, bio);
270 * bio_clone - clone a bio
272 * @gfp_mask: allocation priority
274 * Like __bio_clone, only also allocates the returned bio
276 struct bio *bio_clone(struct bio *bio, unsigned int __nocast gfp_mask)
278 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
287 * bio_get_nr_vecs - return approx number of vecs
290 * Return the approximate number of pages we can send to this target.
291 * There's no guarantee that you will be able to fit this number of pages
292 * into a bio, it does not account for dynamic restrictions that vary
295 int bio_get_nr_vecs(struct block_device *bdev)
297 request_queue_t *q = bdev_get_queue(bdev);
300 nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
301 if (nr_pages > q->max_phys_segments)
302 nr_pages = q->max_phys_segments;
303 if (nr_pages > q->max_hw_segments)
304 nr_pages = q->max_hw_segments;
309 static int __bio_add_page(request_queue_t *q, struct bio *bio, struct page
310 *page, unsigned int len, unsigned int offset)
312 int retried_segments = 0;
313 struct bio_vec *bvec;
316 * cloned bio must not modify vec list
318 if (unlikely(bio_flagged(bio, BIO_CLONED)))
321 if (bio->bi_vcnt >= bio->bi_max_vecs)
324 if (((bio->bi_size + len) >> 9) > q->max_sectors)
328 * we might lose a segment or two here, but rather that than
329 * make this too complex.
332 while (bio->bi_phys_segments >= q->max_phys_segments
333 || bio->bi_hw_segments >= q->max_hw_segments
334 || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) {
336 if (retried_segments)
339 retried_segments = 1;
340 blk_recount_segments(q, bio);
344 * setup the new entry, we might clear it again later if we
345 * cannot add the page
347 bvec = &bio->bi_io_vec[bio->bi_vcnt];
348 bvec->bv_page = page;
350 bvec->bv_offset = offset;
353 * if queue has other restrictions (eg varying max sector size
354 * depending on offset), it can specify a merge_bvec_fn in the
355 * queue to get further control
357 if (q->merge_bvec_fn) {
359 * merge_bvec_fn() returns number of bytes it can accept
362 if (q->merge_bvec_fn(q, bio, bvec) < len) {
363 bvec->bv_page = NULL;
370 /* If we may be able to merge these biovecs, force a recount */
371 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) ||
372 BIOVEC_VIRT_MERGEABLE(bvec-1, bvec)))
373 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
376 bio->bi_phys_segments++;
377 bio->bi_hw_segments++;
383 * bio_add_page - attempt to add page to bio
384 * @bio: destination bio
386 * @len: vec entry length
387 * @offset: vec entry offset
389 * Attempt to add a page to the bio_vec maplist. This can fail for a
390 * number of reasons, such as the bio being full or target block
391 * device limitations. The target block device must allow bio's
392 * smaller than PAGE_SIZE, so it is always possible to add a single
393 * page to an empty bio.
395 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
398 return __bio_add_page(bdev_get_queue(bio->bi_bdev), bio, page,
402 struct bio_map_data {
403 struct bio_vec *iovecs;
404 void __user *userptr;
407 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio)
409 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
410 bio->bi_private = bmd;
413 static void bio_free_map_data(struct bio_map_data *bmd)
419 static struct bio_map_data *bio_alloc_map_data(int nr_segs)
421 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), GFP_KERNEL);
426 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, GFP_KERNEL);
435 * bio_uncopy_user - finish previously mapped bio
436 * @bio: bio being terminated
438 * Free pages allocated from bio_copy_user() and write back data
439 * to user space in case of a read.
441 int bio_uncopy_user(struct bio *bio)
443 struct bio_map_data *bmd = bio->bi_private;
444 const int read = bio_data_dir(bio) == READ;
445 struct bio_vec *bvec;
448 __bio_for_each_segment(bvec, bio, i, 0) {
449 char *addr = page_address(bvec->bv_page);
450 unsigned int len = bmd->iovecs[i].bv_len;
452 if (read && !ret && copy_to_user(bmd->userptr, addr, len))
455 __free_page(bvec->bv_page);
458 bio_free_map_data(bmd);
464 * bio_copy_user - copy user data to bio
465 * @q: destination block queue
466 * @uaddr: start of user address
467 * @len: length in bytes
468 * @write_to_vm: bool indicating writing to pages or not
470 * Prepares and returns a bio for indirect user io, bouncing data
471 * to/from kernel pages as necessary. Must be paired with
472 * call bio_uncopy_user() on io completion.
474 struct bio *bio_copy_user(request_queue_t *q, unsigned long uaddr,
475 unsigned int len, int write_to_vm)
477 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
478 unsigned long start = uaddr >> PAGE_SHIFT;
479 struct bio_map_data *bmd;
480 struct bio_vec *bvec;
485 bmd = bio_alloc_map_data(end - start);
487 return ERR_PTR(-ENOMEM);
489 bmd->userptr = (void __user *) uaddr;
492 bio = bio_alloc(GFP_KERNEL, end - start);
496 bio->bi_rw |= (!write_to_vm << BIO_RW);
500 unsigned int bytes = PAGE_SIZE;
505 page = alloc_page(q->bounce_gfp | GFP_KERNEL);
511 if (__bio_add_page(q, bio, page, bytes, 0) < bytes) {
526 char __user *p = (char __user *) uaddr;
529 * for a write, copy in data to kernel pages
532 bio_for_each_segment(bvec, bio, i) {
533 char *addr = page_address(bvec->bv_page);
535 if (copy_from_user(addr, p, bvec->bv_len))
541 bio_set_map_data(bmd, bio);
544 bio_for_each_segment(bvec, bio, i)
545 __free_page(bvec->bv_page);
549 bio_free_map_data(bmd);
553 static struct bio *__bio_map_user_iov(request_queue_t *q,
554 struct block_device *bdev,
555 struct sg_iovec *iov, int iov_count,
565 for (i = 0; i < iov_count; i++) {
566 unsigned long uaddr = (unsigned long)iov[i].iov_base;
567 unsigned long len = iov[i].iov_len;
568 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
569 unsigned long start = uaddr >> PAGE_SHIFT;
571 nr_pages += end - start;
573 * transfer and buffer must be aligned to at least hardsector
574 * size for now, in the future we can relax this restriction
576 if ((uaddr & queue_dma_alignment(q)) || (len & queue_dma_alignment(q)))
577 return ERR_PTR(-EINVAL);
581 return ERR_PTR(-EINVAL);
583 bio = bio_alloc(GFP_KERNEL, nr_pages);
585 return ERR_PTR(-ENOMEM);
588 pages = kmalloc(nr_pages * sizeof(struct page *), GFP_KERNEL);
592 memset(pages, 0, nr_pages * sizeof(struct page *));
594 for (i = 0; i < iov_count; i++) {
595 unsigned long uaddr = (unsigned long)iov[i].iov_base;
596 unsigned long len = iov[i].iov_len;
597 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
598 unsigned long start = uaddr >> PAGE_SHIFT;
599 const int local_nr_pages = end - start;
600 const int page_limit = cur_page + local_nr_pages;
602 down_read(¤t->mm->mmap_sem);
603 ret = get_user_pages(current, current->mm, uaddr,
605 write_to_vm, 0, &pages[cur_page], NULL);
606 up_read(¤t->mm->mmap_sem);
608 if (ret < local_nr_pages)
612 offset = uaddr & ~PAGE_MASK;
613 for (j = cur_page; j < page_limit; j++) {
614 unsigned int bytes = PAGE_SIZE - offset;
625 if (__bio_add_page(q, bio, pages[j], bytes, offset) < bytes)
634 * release the pages we didn't map into the bio, if any
636 while (j < page_limit)
637 page_cache_release(pages[j++]);
643 * set data direction, and check if mapped pages need bouncing
646 bio->bi_rw |= (1 << BIO_RW);
649 bio->bi_flags |= (1 << BIO_USER_MAPPED);
653 for (i = 0; i < nr_pages; i++) {
656 page_cache_release(pages[i]);
665 * bio_map_user - map user address into bio
666 * @q: the request_queue_t for the bio
667 * @bdev: destination block device
668 * @uaddr: start of user address
669 * @len: length in bytes
670 * @write_to_vm: bool indicating writing to pages or not
672 * Map the user space address into a bio suitable for io to a block
673 * device. Returns an error pointer in case of error.
675 struct bio *bio_map_user(request_queue_t *q, struct block_device *bdev,
676 unsigned long uaddr, unsigned int len, int write_to_vm)
680 iov.iov_base = (__user void *)uaddr;
683 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm);
687 * bio_map_user_iov - map user sg_iovec table into bio
688 * @q: the request_queue_t for the bio
689 * @bdev: destination block device
691 * @iov_count: number of elements in the iovec
692 * @write_to_vm: bool indicating writing to pages or not
694 * Map the user space address into a bio suitable for io to a block
695 * device. Returns an error pointer in case of error.
697 struct bio *bio_map_user_iov(request_queue_t *q, struct block_device *bdev,
698 struct sg_iovec *iov, int iov_count,
704 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm);
710 * subtle -- if __bio_map_user() ended up bouncing a bio,
711 * it would normally disappear when its bi_end_io is run.
712 * however, we need it for the unmap, so grab an extra
717 for (i = 0; i < iov_count; i++)
718 len += iov[i].iov_len;
720 if (bio->bi_size == len)
724 * don't support partial mappings
726 bio_endio(bio, bio->bi_size, 0);
728 return ERR_PTR(-EINVAL);
731 static void __bio_unmap_user(struct bio *bio)
733 struct bio_vec *bvec;
737 * make sure we dirty pages we wrote to
739 __bio_for_each_segment(bvec, bio, i, 0) {
740 if (bio_data_dir(bio) == READ)
741 set_page_dirty_lock(bvec->bv_page);
743 page_cache_release(bvec->bv_page);
750 * bio_unmap_user - unmap a bio
751 * @bio: the bio being unmapped
753 * Unmap a bio previously mapped by bio_map_user(). Must be called with
756 * bio_unmap_user() may sleep.
758 void bio_unmap_user(struct bio *bio)
760 __bio_unmap_user(bio);
764 static int bio_map_kern_endio(struct bio *bio, unsigned int bytes_done, int err)
774 static struct bio *__bio_map_kern(request_queue_t *q, void *data,
775 unsigned int len, unsigned int gfp_mask)
777 unsigned long kaddr = (unsigned long)data;
778 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
779 unsigned long start = kaddr >> PAGE_SHIFT;
780 const int nr_pages = end - start;
784 bio = bio_alloc(gfp_mask, nr_pages);
786 return ERR_PTR(-ENOMEM);
788 offset = offset_in_page(kaddr);
789 for (i = 0; i < nr_pages; i++) {
790 unsigned int bytes = PAGE_SIZE - offset;
798 if (__bio_add_page(q, bio, virt_to_page(data), bytes,
807 bio->bi_end_io = bio_map_kern_endio;
812 * bio_map_kern - map kernel address into bio
813 * @q: the request_queue_t for the bio
814 * @data: pointer to buffer to map
815 * @len: length in bytes
816 * @gfp_mask: allocation flags for bio allocation
818 * Map the kernel address into a bio suitable for io to a block
819 * device. Returns an error pointer in case of error.
821 struct bio *bio_map_kern(request_queue_t *q, void *data, unsigned int len,
822 unsigned int gfp_mask)
826 bio = __bio_map_kern(q, data, len, gfp_mask);
830 if (bio->bi_size == len)
834 * Don't support partial mappings.
837 return ERR_PTR(-EINVAL);
841 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
842 * for performing direct-IO in BIOs.
844 * The problem is that we cannot run set_page_dirty() from interrupt context
845 * because the required locks are not interrupt-safe. So what we can do is to
846 * mark the pages dirty _before_ performing IO. And in interrupt context,
847 * check that the pages are still dirty. If so, fine. If not, redirty them
848 * in process context.
850 * We special-case compound pages here: normally this means reads into hugetlb
851 * pages. The logic in here doesn't really work right for compound pages
852 * because the VM does not uniformly chase down the head page in all cases.
853 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
854 * handle them at all. So we skip compound pages here at an early stage.
856 * Note that this code is very hard to test under normal circumstances because
857 * direct-io pins the pages with get_user_pages(). This makes
858 * is_page_cache_freeable return false, and the VM will not clean the pages.
859 * But other code (eg, pdflush) could clean the pages if they are mapped
862 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
863 * deferred bio dirtying paths.
867 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
869 void bio_set_pages_dirty(struct bio *bio)
871 struct bio_vec *bvec = bio->bi_io_vec;
874 for (i = 0; i < bio->bi_vcnt; i++) {
875 struct page *page = bvec[i].bv_page;
877 if (page && !PageCompound(page))
878 set_page_dirty_lock(page);
882 static void bio_release_pages(struct bio *bio)
884 struct bio_vec *bvec = bio->bi_io_vec;
887 for (i = 0; i < bio->bi_vcnt; i++) {
888 struct page *page = bvec[i].bv_page;
896 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
897 * If they are, then fine. If, however, some pages are clean then they must
898 * have been written out during the direct-IO read. So we take another ref on
899 * the BIO and the offending pages and re-dirty the pages in process context.
901 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
902 * here on. It will run one page_cache_release() against each page and will
903 * run one bio_put() against the BIO.
906 static void bio_dirty_fn(void *data);
908 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn, NULL);
909 static DEFINE_SPINLOCK(bio_dirty_lock);
910 static struct bio *bio_dirty_list;
913 * This runs in process context
915 static void bio_dirty_fn(void *data)
920 spin_lock_irqsave(&bio_dirty_lock, flags);
921 bio = bio_dirty_list;
922 bio_dirty_list = NULL;
923 spin_unlock_irqrestore(&bio_dirty_lock, flags);
926 struct bio *next = bio->bi_private;
928 bio_set_pages_dirty(bio);
929 bio_release_pages(bio);
935 void bio_check_pages_dirty(struct bio *bio)
937 struct bio_vec *bvec = bio->bi_io_vec;
938 int nr_clean_pages = 0;
941 for (i = 0; i < bio->bi_vcnt; i++) {
942 struct page *page = bvec[i].bv_page;
944 if (PageDirty(page) || PageCompound(page)) {
945 page_cache_release(page);
946 bvec[i].bv_page = NULL;
952 if (nr_clean_pages) {
955 spin_lock_irqsave(&bio_dirty_lock, flags);
956 bio->bi_private = bio_dirty_list;
957 bio_dirty_list = bio;
958 spin_unlock_irqrestore(&bio_dirty_lock, flags);
959 schedule_work(&bio_dirty_work);
966 * bio_endio - end I/O on a bio
968 * @bytes_done: number of bytes completed
969 * @error: error, if any
972 * bio_endio() will end I/O on @bytes_done number of bytes. This may be
973 * just a partial part of the bio, or it may be the whole bio. bio_endio()
974 * is the preferred way to end I/O on a bio, it takes care of decrementing
975 * bi_size and clearing BIO_UPTODATE on error. @error is 0 on success, and
976 * and one of the established -Exxxx (-EIO, for instance) error values in
977 * case something went wrong. Noone should call bi_end_io() directly on
978 * a bio unless they own it and thus know that it has an end_io function.
980 void bio_endio(struct bio *bio, unsigned int bytes_done, int error)
983 clear_bit(BIO_UPTODATE, &bio->bi_flags);
985 if (unlikely(bytes_done > bio->bi_size)) {
986 printk("%s: want %u bytes done, only %u left\n", __FUNCTION__,
987 bytes_done, bio->bi_size);
988 bytes_done = bio->bi_size;
991 bio->bi_size -= bytes_done;
992 bio->bi_sector += (bytes_done >> 9);
995 bio->bi_end_io(bio, bytes_done, error);
998 void bio_pair_release(struct bio_pair *bp)
1000 if (atomic_dec_and_test(&bp->cnt)) {
1001 struct bio *master = bp->bio1.bi_private;
1003 bio_endio(master, master->bi_size, bp->error);
1004 mempool_free(bp, bp->bio2.bi_private);
1008 static int bio_pair_end_1(struct bio * bi, unsigned int done, int err)
1010 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1018 bio_pair_release(bp);
1022 static int bio_pair_end_2(struct bio * bi, unsigned int done, int err)
1024 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1032 bio_pair_release(bp);
1037 * split a bio - only worry about a bio with a single page
1040 struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors)
1042 struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO);
1047 BUG_ON(bi->bi_vcnt != 1);
1048 BUG_ON(bi->bi_idx != 0);
1049 atomic_set(&bp->cnt, 3);
1053 bp->bio2.bi_sector += first_sectors;
1054 bp->bio2.bi_size -= first_sectors << 9;
1055 bp->bio1.bi_size = first_sectors << 9;
1057 bp->bv1 = bi->bi_io_vec[0];
1058 bp->bv2 = bi->bi_io_vec[0];
1059 bp->bv2.bv_offset += first_sectors << 9;
1060 bp->bv2.bv_len -= first_sectors << 9;
1061 bp->bv1.bv_len = first_sectors << 9;
1063 bp->bio1.bi_io_vec = &bp->bv1;
1064 bp->bio2.bi_io_vec = &bp->bv2;
1066 bp->bio1.bi_end_io = bio_pair_end_1;
1067 bp->bio2.bi_end_io = bio_pair_end_2;
1069 bp->bio1.bi_private = bi;
1070 bp->bio2.bi_private = pool;
1075 static void *bio_pair_alloc(unsigned int __nocast gfp_flags, void *data)
1077 return kmalloc(sizeof(struct bio_pair), gfp_flags);
1080 static void bio_pair_free(void *bp, void *data)
1087 * create memory pools for biovec's in a bio_set.
1088 * use the global biovec slabs created for general use.
1090 static int biovec_create_pools(struct bio_set *bs, int pool_entries, int scale)
1094 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1095 struct biovec_slab *bp = bvec_slabs + i;
1096 mempool_t **bvp = bs->bvec_pools + i;
1101 *bvp = mempool_create(pool_entries, mempool_alloc_slab,
1102 mempool_free_slab, bp->slab);
1109 static void biovec_free_pools(struct bio_set *bs)
1113 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1114 mempool_t *bvp = bs->bvec_pools[i];
1117 mempool_destroy(bvp);
1122 void bioset_free(struct bio_set *bs)
1125 mempool_destroy(bs->bio_pool);
1127 biovec_free_pools(bs);
1132 struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size, int scale)
1134 struct bio_set *bs = kmalloc(sizeof(*bs), GFP_KERNEL);
1139 memset(bs, 0, sizeof(*bs));
1140 bs->bio_pool = mempool_create(bio_pool_size, mempool_alloc_slab,
1141 mempool_free_slab, bio_slab);
1146 if (!biovec_create_pools(bs, bvec_pool_size, scale))
1154 static void __init biovec_init_slabs(void)
1158 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1160 struct biovec_slab *bvs = bvec_slabs + i;
1162 size = bvs->nr_vecs * sizeof(struct bio_vec);
1163 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1164 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1168 static int __init init_bio(void)
1170 int megabytes, bvec_pool_entries;
1171 int scale = BIOVEC_NR_POOLS;
1173 bio_slab = kmem_cache_create("bio", sizeof(struct bio), 0,
1174 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1176 biovec_init_slabs();
1178 megabytes = nr_free_pages() >> (20 - PAGE_SHIFT);
1181 * find out where to start scaling
1183 if (megabytes <= 16)
1185 else if (megabytes <= 32)
1187 else if (megabytes <= 64)
1189 else if (megabytes <= 96)
1191 else if (megabytes <= 128)
1195 * scale number of entries
1197 bvec_pool_entries = megabytes * 2;
1198 if (bvec_pool_entries > 256)
1199 bvec_pool_entries = 256;
1201 fs_bio_set = bioset_create(BIO_POOL_SIZE, bvec_pool_entries, scale);
1203 panic("bio: can't allocate bios\n");
1205 bio_split_pool = mempool_create(BIO_SPLIT_ENTRIES,
1206 bio_pair_alloc, bio_pair_free, NULL);
1207 if (!bio_split_pool)
1208 panic("bio: can't create split pool\n");
1213 subsys_initcall(init_bio);
1215 EXPORT_SYMBOL(bio_alloc);
1216 EXPORT_SYMBOL(bio_put);
1217 EXPORT_SYMBOL(bio_endio);
1218 EXPORT_SYMBOL(bio_init);
1219 EXPORT_SYMBOL(__bio_clone);
1220 EXPORT_SYMBOL(bio_clone);
1221 EXPORT_SYMBOL(bio_phys_segments);
1222 EXPORT_SYMBOL(bio_hw_segments);
1223 EXPORT_SYMBOL(bio_add_page);
1224 EXPORT_SYMBOL(bio_get_nr_vecs);
1225 EXPORT_SYMBOL(bio_map_user);
1226 EXPORT_SYMBOL(bio_unmap_user);
1227 EXPORT_SYMBOL(bio_map_kern);
1228 EXPORT_SYMBOL(bio_pair_release);
1229 EXPORT_SYMBOL(bio_split);
1230 EXPORT_SYMBOL(bio_split_pool);
1231 EXPORT_SYMBOL(bio_copy_user);
1232 EXPORT_SYMBOL(bio_uncopy_user);
1233 EXPORT_SYMBOL(bioset_create);
1234 EXPORT_SYMBOL(bioset_free);
1235 EXPORT_SYMBOL(bio_alloc_bioset);