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)
318 int retried_segments = 0;
319 struct bio_vec *bvec;
322 * cloned bio must not modify vec list
324 if (unlikely(bio_flagged(bio, BIO_CLONED)))
327 if (bio->bi_vcnt >= bio->bi_max_vecs)
330 if (((bio->bi_size + len) >> 9) > q->max_sectors)
334 * we might lose a segment or two here, but rather that than
335 * make this too complex.
338 while (bio->bi_phys_segments >= q->max_phys_segments
339 || bio->bi_hw_segments >= q->max_hw_segments
340 || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) {
342 if (retried_segments)
345 retried_segments = 1;
346 blk_recount_segments(q, bio);
350 * setup the new entry, we might clear it again later if we
351 * cannot add the page
353 bvec = &bio->bi_io_vec[bio->bi_vcnt];
354 bvec->bv_page = page;
356 bvec->bv_offset = offset;
359 * if queue has other restrictions (eg varying max sector size
360 * depending on offset), it can specify a merge_bvec_fn in the
361 * queue to get further control
363 if (q->merge_bvec_fn) {
365 * merge_bvec_fn() returns number of bytes it can accept
368 if (q->merge_bvec_fn(q, bio, bvec) < len) {
369 bvec->bv_page = NULL;
376 /* If we may be able to merge these biovecs, force a recount */
377 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) ||
378 BIOVEC_VIRT_MERGEABLE(bvec-1, bvec)))
379 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
382 bio->bi_phys_segments++;
383 bio->bi_hw_segments++;
389 * bio_add_pc_page - attempt to add page to bio
390 * @bio: destination bio
392 * @len: vec entry length
393 * @offset: vec entry offset
395 * Attempt to add a page to the bio_vec maplist. This can fail for a
396 * number of reasons, such as the bio being full or target block
397 * device limitations. The target block device must allow bio's
398 * smaller than PAGE_SIZE, so it is always possible to add a single
399 * page to an empty bio. This should only be used by REQ_PC bios.
401 int bio_add_pc_page(request_queue_t *q, struct bio *bio, struct page *page,
402 unsigned int len, unsigned int offset)
404 return __bio_add_page(q, bio, page, len, offset);
408 * bio_add_page - attempt to add page to bio
409 * @bio: destination bio
411 * @len: vec entry length
412 * @offset: vec entry offset
414 * Attempt to add a page to the bio_vec maplist. This can fail for a
415 * number of reasons, such as the bio being full or target block
416 * device limitations. The target block device must allow bio's
417 * smaller than PAGE_SIZE, so it is always possible to add a single
418 * page to an empty bio.
420 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
423 return __bio_add_page(bdev_get_queue(bio->bi_bdev), bio, page,
427 struct bio_map_data {
428 struct bio_vec *iovecs;
429 void __user *userptr;
432 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio)
434 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
435 bio->bi_private = bmd;
438 static void bio_free_map_data(struct bio_map_data *bmd)
444 static struct bio_map_data *bio_alloc_map_data(int nr_segs)
446 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), GFP_KERNEL);
451 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, GFP_KERNEL);
460 * bio_uncopy_user - finish previously mapped bio
461 * @bio: bio being terminated
463 * Free pages allocated from bio_copy_user() and write back data
464 * to user space in case of a read.
466 int bio_uncopy_user(struct bio *bio)
468 struct bio_map_data *bmd = bio->bi_private;
469 const int read = bio_data_dir(bio) == READ;
470 struct bio_vec *bvec;
473 __bio_for_each_segment(bvec, bio, i, 0) {
474 char *addr = page_address(bvec->bv_page);
475 unsigned int len = bmd->iovecs[i].bv_len;
477 if (read && !ret && copy_to_user(bmd->userptr, addr, len))
480 __free_page(bvec->bv_page);
483 bio_free_map_data(bmd);
489 * bio_copy_user - copy user data to bio
490 * @q: destination block queue
491 * @uaddr: start of user address
492 * @len: length in bytes
493 * @write_to_vm: bool indicating writing to pages or not
495 * Prepares and returns a bio for indirect user io, bouncing data
496 * to/from kernel pages as necessary. Must be paired with
497 * call bio_uncopy_user() on io completion.
499 struct bio *bio_copy_user(request_queue_t *q, unsigned long uaddr,
500 unsigned int len, int write_to_vm)
502 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
503 unsigned long start = uaddr >> PAGE_SHIFT;
504 struct bio_map_data *bmd;
505 struct bio_vec *bvec;
510 bmd = bio_alloc_map_data(end - start);
512 return ERR_PTR(-ENOMEM);
514 bmd->userptr = (void __user *) uaddr;
517 bio = bio_alloc(GFP_KERNEL, end - start);
521 bio->bi_rw |= (!write_to_vm << BIO_RW);
525 unsigned int bytes = PAGE_SIZE;
530 page = alloc_page(q->bounce_gfp | GFP_KERNEL);
536 if (__bio_add_page(q, bio, page, bytes, 0) < bytes) {
551 char __user *p = (char __user *) uaddr;
554 * for a write, copy in data to kernel pages
557 bio_for_each_segment(bvec, bio, i) {
558 char *addr = page_address(bvec->bv_page);
560 if (copy_from_user(addr, p, bvec->bv_len))
566 bio_set_map_data(bmd, bio);
569 bio_for_each_segment(bvec, bio, i)
570 __free_page(bvec->bv_page);
574 bio_free_map_data(bmd);
578 static struct bio *__bio_map_user_iov(request_queue_t *q,
579 struct block_device *bdev,
580 struct sg_iovec *iov, int iov_count,
590 for (i = 0; i < iov_count; i++) {
591 unsigned long uaddr = (unsigned long)iov[i].iov_base;
592 unsigned long len = iov[i].iov_len;
593 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
594 unsigned long start = uaddr >> PAGE_SHIFT;
596 nr_pages += end - start;
598 * transfer and buffer must be aligned to at least hardsector
599 * size for now, in the future we can relax this restriction
601 if ((uaddr & queue_dma_alignment(q)) || (len & queue_dma_alignment(q)))
602 return ERR_PTR(-EINVAL);
606 return ERR_PTR(-EINVAL);
608 bio = bio_alloc(GFP_KERNEL, nr_pages);
610 return ERR_PTR(-ENOMEM);
613 pages = kmalloc(nr_pages * sizeof(struct page *), GFP_KERNEL);
617 memset(pages, 0, nr_pages * sizeof(struct page *));
619 for (i = 0; i < iov_count; i++) {
620 unsigned long uaddr = (unsigned long)iov[i].iov_base;
621 unsigned long len = iov[i].iov_len;
622 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
623 unsigned long start = uaddr >> PAGE_SHIFT;
624 const int local_nr_pages = end - start;
625 const int page_limit = cur_page + local_nr_pages;
627 down_read(¤t->mm->mmap_sem);
628 ret = get_user_pages(current, current->mm, uaddr,
630 write_to_vm, 0, &pages[cur_page], NULL);
631 up_read(¤t->mm->mmap_sem);
633 if (ret < local_nr_pages)
637 offset = uaddr & ~PAGE_MASK;
638 for (j = cur_page; j < page_limit; j++) {
639 unsigned int bytes = PAGE_SIZE - offset;
650 if (__bio_add_page(q, bio, pages[j], bytes, offset) < bytes)
659 * release the pages we didn't map into the bio, if any
661 while (j < page_limit)
662 page_cache_release(pages[j++]);
668 * set data direction, and check if mapped pages need bouncing
671 bio->bi_rw |= (1 << BIO_RW);
674 bio->bi_flags |= (1 << BIO_USER_MAPPED);
678 for (i = 0; i < nr_pages; i++) {
681 page_cache_release(pages[i]);
690 * bio_map_user - map user address into bio
691 * @q: the request_queue_t for the bio
692 * @bdev: destination block device
693 * @uaddr: start of user address
694 * @len: length in bytes
695 * @write_to_vm: bool indicating writing to pages or not
697 * Map the user space address into a bio suitable for io to a block
698 * device. Returns an error pointer in case of error.
700 struct bio *bio_map_user(request_queue_t *q, struct block_device *bdev,
701 unsigned long uaddr, unsigned int len, int write_to_vm)
705 iov.iov_base = (void __user *)uaddr;
708 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm);
712 * bio_map_user_iov - map user sg_iovec table into bio
713 * @q: the request_queue_t for the bio
714 * @bdev: destination block device
716 * @iov_count: number of elements in the iovec
717 * @write_to_vm: bool indicating writing to pages or not
719 * Map the user space address into a bio suitable for io to a block
720 * device. Returns an error pointer in case of error.
722 struct bio *bio_map_user_iov(request_queue_t *q, struct block_device *bdev,
723 struct sg_iovec *iov, int iov_count,
729 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm);
735 * subtle -- if __bio_map_user() ended up bouncing a bio,
736 * it would normally disappear when its bi_end_io is run.
737 * however, we need it for the unmap, so grab an extra
742 for (i = 0; i < iov_count; i++)
743 len += iov[i].iov_len;
745 if (bio->bi_size == len)
749 * don't support partial mappings
751 bio_endio(bio, bio->bi_size, 0);
753 return ERR_PTR(-EINVAL);
756 static void __bio_unmap_user(struct bio *bio)
758 struct bio_vec *bvec;
762 * make sure we dirty pages we wrote to
764 __bio_for_each_segment(bvec, bio, i, 0) {
765 if (bio_data_dir(bio) == READ)
766 set_page_dirty_lock(bvec->bv_page);
768 page_cache_release(bvec->bv_page);
775 * bio_unmap_user - unmap a bio
776 * @bio: the bio being unmapped
778 * Unmap a bio previously mapped by bio_map_user(). Must be called with
781 * bio_unmap_user() may sleep.
783 void bio_unmap_user(struct bio *bio)
785 __bio_unmap_user(bio);
789 static int bio_map_kern_endio(struct bio *bio, unsigned int bytes_done, int err)
799 static struct bio *__bio_map_kern(request_queue_t *q, void *data,
800 unsigned int len, gfp_t gfp_mask)
802 unsigned long kaddr = (unsigned long)data;
803 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
804 unsigned long start = kaddr >> PAGE_SHIFT;
805 const int nr_pages = end - start;
809 bio = bio_alloc(gfp_mask, nr_pages);
811 return ERR_PTR(-ENOMEM);
813 offset = offset_in_page(kaddr);
814 for (i = 0; i < nr_pages; i++) {
815 unsigned int bytes = PAGE_SIZE - offset;
823 if (__bio_add_page(q, bio, virt_to_page(data), bytes,
832 bio->bi_end_io = bio_map_kern_endio;
837 * bio_map_kern - map kernel address into bio
838 * @q: the request_queue_t for the bio
839 * @data: pointer to buffer to map
840 * @len: length in bytes
841 * @gfp_mask: allocation flags for bio allocation
843 * Map the kernel address into a bio suitable for io to a block
844 * device. Returns an error pointer in case of error.
846 struct bio *bio_map_kern(request_queue_t *q, void *data, unsigned int len,
851 bio = __bio_map_kern(q, data, len, gfp_mask);
855 if (bio->bi_size == len)
859 * Don't support partial mappings.
862 return ERR_PTR(-EINVAL);
866 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
867 * for performing direct-IO in BIOs.
869 * The problem is that we cannot run set_page_dirty() from interrupt context
870 * because the required locks are not interrupt-safe. So what we can do is to
871 * mark the pages dirty _before_ performing IO. And in interrupt context,
872 * check that the pages are still dirty. If so, fine. If not, redirty them
873 * in process context.
875 * We special-case compound pages here: normally this means reads into hugetlb
876 * pages. The logic in here doesn't really work right for compound pages
877 * because the VM does not uniformly chase down the head page in all cases.
878 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
879 * handle them at all. So we skip compound pages here at an early stage.
881 * Note that this code is very hard to test under normal circumstances because
882 * direct-io pins the pages with get_user_pages(). This makes
883 * is_page_cache_freeable return false, and the VM will not clean the pages.
884 * But other code (eg, pdflush) could clean the pages if they are mapped
887 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
888 * deferred bio dirtying paths.
892 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
894 void bio_set_pages_dirty(struct bio *bio)
896 struct bio_vec *bvec = bio->bi_io_vec;
899 for (i = 0; i < bio->bi_vcnt; i++) {
900 struct page *page = bvec[i].bv_page;
902 if (page && !PageCompound(page))
903 set_page_dirty_lock(page);
907 static void bio_release_pages(struct bio *bio)
909 struct bio_vec *bvec = bio->bi_io_vec;
912 for (i = 0; i < bio->bi_vcnt; i++) {
913 struct page *page = bvec[i].bv_page;
921 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
922 * If they are, then fine. If, however, some pages are clean then they must
923 * have been written out during the direct-IO read. So we take another ref on
924 * the BIO and the offending pages and re-dirty the pages in process context.
926 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
927 * here on. It will run one page_cache_release() against each page and will
928 * run one bio_put() against the BIO.
931 static void bio_dirty_fn(void *data);
933 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn, NULL);
934 static DEFINE_SPINLOCK(bio_dirty_lock);
935 static struct bio *bio_dirty_list;
938 * This runs in process context
940 static void bio_dirty_fn(void *data)
945 spin_lock_irqsave(&bio_dirty_lock, flags);
946 bio = bio_dirty_list;
947 bio_dirty_list = NULL;
948 spin_unlock_irqrestore(&bio_dirty_lock, flags);
951 struct bio *next = bio->bi_private;
953 bio_set_pages_dirty(bio);
954 bio_release_pages(bio);
960 void bio_check_pages_dirty(struct bio *bio)
962 struct bio_vec *bvec = bio->bi_io_vec;
963 int nr_clean_pages = 0;
966 for (i = 0; i < bio->bi_vcnt; i++) {
967 struct page *page = bvec[i].bv_page;
969 if (PageDirty(page) || PageCompound(page)) {
970 page_cache_release(page);
971 bvec[i].bv_page = NULL;
977 if (nr_clean_pages) {
980 spin_lock_irqsave(&bio_dirty_lock, flags);
981 bio->bi_private = bio_dirty_list;
982 bio_dirty_list = bio;
983 spin_unlock_irqrestore(&bio_dirty_lock, flags);
984 schedule_work(&bio_dirty_work);
991 * bio_endio - end I/O on a bio
993 * @bytes_done: number of bytes completed
994 * @error: error, if any
997 * bio_endio() will end I/O on @bytes_done number of bytes. This may be
998 * just a partial part of the bio, or it may be the whole bio. bio_endio()
999 * is the preferred way to end I/O on a bio, it takes care of decrementing
1000 * bi_size and clearing BIO_UPTODATE on error. @error is 0 on success, and
1001 * and one of the established -Exxxx (-EIO, for instance) error values in
1002 * case something went wrong. Noone should call bi_end_io() directly on
1003 * a bio unless they own it and thus know that it has an end_io function.
1005 void bio_endio(struct bio *bio, unsigned int bytes_done, int error)
1008 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1010 if (unlikely(bytes_done > bio->bi_size)) {
1011 printk("%s: want %u bytes done, only %u left\n", __FUNCTION__,
1012 bytes_done, bio->bi_size);
1013 bytes_done = bio->bi_size;
1016 bio->bi_size -= bytes_done;
1017 bio->bi_sector += (bytes_done >> 9);
1020 bio->bi_end_io(bio, bytes_done, error);
1023 void bio_pair_release(struct bio_pair *bp)
1025 if (atomic_dec_and_test(&bp->cnt)) {
1026 struct bio *master = bp->bio1.bi_private;
1028 bio_endio(master, master->bi_size, bp->error);
1029 mempool_free(bp, bp->bio2.bi_private);
1033 static int bio_pair_end_1(struct bio * bi, unsigned int done, int err)
1035 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1043 bio_pair_release(bp);
1047 static int bio_pair_end_2(struct bio * bi, unsigned int done, int err)
1049 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1057 bio_pair_release(bp);
1062 * split a bio - only worry about a bio with a single page
1065 struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors)
1067 struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO);
1072 BUG_ON(bi->bi_vcnt != 1);
1073 BUG_ON(bi->bi_idx != 0);
1074 atomic_set(&bp->cnt, 3);
1078 bp->bio2.bi_sector += first_sectors;
1079 bp->bio2.bi_size -= first_sectors << 9;
1080 bp->bio1.bi_size = first_sectors << 9;
1082 bp->bv1 = bi->bi_io_vec[0];
1083 bp->bv2 = bi->bi_io_vec[0];
1084 bp->bv2.bv_offset += first_sectors << 9;
1085 bp->bv2.bv_len -= first_sectors << 9;
1086 bp->bv1.bv_len = first_sectors << 9;
1088 bp->bio1.bi_io_vec = &bp->bv1;
1089 bp->bio2.bi_io_vec = &bp->bv2;
1091 bp->bio1.bi_end_io = bio_pair_end_1;
1092 bp->bio2.bi_end_io = bio_pair_end_2;
1094 bp->bio1.bi_private = bi;
1095 bp->bio2.bi_private = pool;
1100 static void *bio_pair_alloc(gfp_t gfp_flags, void *data)
1102 return kmalloc(sizeof(struct bio_pair), gfp_flags);
1105 static void bio_pair_free(void *bp, void *data)
1112 * create memory pools for biovec's in a bio_set.
1113 * use the global biovec slabs created for general use.
1115 static int biovec_create_pools(struct bio_set *bs, int pool_entries, int scale)
1119 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1120 struct biovec_slab *bp = bvec_slabs + i;
1121 mempool_t **bvp = bs->bvec_pools + i;
1126 *bvp = mempool_create(pool_entries, mempool_alloc_slab,
1127 mempool_free_slab, bp->slab);
1134 static void biovec_free_pools(struct bio_set *bs)
1138 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1139 mempool_t *bvp = bs->bvec_pools[i];
1142 mempool_destroy(bvp);
1147 void bioset_free(struct bio_set *bs)
1150 mempool_destroy(bs->bio_pool);
1152 biovec_free_pools(bs);
1157 struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size, int scale)
1159 struct bio_set *bs = kmalloc(sizeof(*bs), GFP_KERNEL);
1164 memset(bs, 0, sizeof(*bs));
1165 bs->bio_pool = mempool_create(bio_pool_size, mempool_alloc_slab,
1166 mempool_free_slab, bio_slab);
1171 if (!biovec_create_pools(bs, bvec_pool_size, scale))
1179 static void __init biovec_init_slabs(void)
1183 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1185 struct biovec_slab *bvs = bvec_slabs + i;
1187 size = bvs->nr_vecs * sizeof(struct bio_vec);
1188 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1189 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1193 static int __init init_bio(void)
1195 int megabytes, bvec_pool_entries;
1196 int scale = BIOVEC_NR_POOLS;
1198 bio_slab = kmem_cache_create("bio", sizeof(struct bio), 0,
1199 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1201 biovec_init_slabs();
1203 megabytes = nr_free_pages() >> (20 - PAGE_SHIFT);
1206 * find out where to start scaling
1208 if (megabytes <= 16)
1210 else if (megabytes <= 32)
1212 else if (megabytes <= 64)
1214 else if (megabytes <= 96)
1216 else if (megabytes <= 128)
1220 * scale number of entries
1222 bvec_pool_entries = megabytes * 2;
1223 if (bvec_pool_entries > 256)
1224 bvec_pool_entries = 256;
1226 fs_bio_set = bioset_create(BIO_POOL_SIZE, bvec_pool_entries, scale);
1228 panic("bio: can't allocate bios\n");
1230 bio_split_pool = mempool_create(BIO_SPLIT_ENTRIES,
1231 bio_pair_alloc, bio_pair_free, NULL);
1232 if (!bio_split_pool)
1233 panic("bio: can't create split pool\n");
1238 subsys_initcall(init_bio);
1240 EXPORT_SYMBOL(bio_alloc);
1241 EXPORT_SYMBOL(bio_put);
1242 EXPORT_SYMBOL(bio_free);
1243 EXPORT_SYMBOL(bio_endio);
1244 EXPORT_SYMBOL(bio_init);
1245 EXPORT_SYMBOL(__bio_clone);
1246 EXPORT_SYMBOL(bio_clone);
1247 EXPORT_SYMBOL(bio_phys_segments);
1248 EXPORT_SYMBOL(bio_hw_segments);
1249 EXPORT_SYMBOL(bio_add_page);
1250 EXPORT_SYMBOL(bio_add_pc_page);
1251 EXPORT_SYMBOL(bio_get_nr_vecs);
1252 EXPORT_SYMBOL(bio_map_user);
1253 EXPORT_SYMBOL(bio_unmap_user);
1254 EXPORT_SYMBOL(bio_map_kern);
1255 EXPORT_SYMBOL(bio_pair_release);
1256 EXPORT_SYMBOL(bio_split);
1257 EXPORT_SYMBOL(bio_split_pool);
1258 EXPORT_SYMBOL(bio_copy_user);
1259 EXPORT_SYMBOL(bio_uncopy_user);
1260 EXPORT_SYMBOL(bioset_create);
1261 EXPORT_SYMBOL(bioset_free);
1262 EXPORT_SYMBOL(bio_alloc_bioset);