2 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
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 <linux/blktrace_api.h>
29 #include <scsi/sg.h> /* for struct sg_iovec */
31 #define BIO_POOL_SIZE 2
33 static struct kmem_cache *bio_slab __read_mostly;
35 #define BIOVEC_NR_POOLS 6
38 * a small number of entries is fine, not going to be performance critical.
39 * basically we just need to survive
41 #define BIO_SPLIT_ENTRIES 2
42 mempool_t *bio_split_pool __read_mostly;
47 struct kmem_cache *slab;
51 * if you change this list, also change bvec_alloc or things will
52 * break badly! cannot be bigger than what you can fit into an
56 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
57 static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
58 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
63 * bio_set is used to allow other portions of the IO system to
64 * allocate their own private memory pools for bio and iovec structures.
65 * These memory pools in turn all allocate from the bio_slab
66 * and the bvec_slabs[].
70 mempool_t *bvec_pools[BIOVEC_NR_POOLS];
74 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
75 * IO code that does not need private memory pools.
77 static struct bio_set *fs_bio_set;
79 static inline struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, struct bio_set *bs)
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 bvl = mempool_alloc(bs->bvec_pools[*idx], gfp_mask);
102 struct biovec_slab *bp = bvec_slabs + *idx;
104 memset(bvl, 0, bp->nr_vecs * sizeof(struct bio_vec));
110 void bio_free(struct bio *bio, struct bio_set *bio_set)
112 if (bio->bi_io_vec) {
113 const int pool_idx = BIO_POOL_IDX(bio);
115 BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS);
117 mempool_free(bio->bi_io_vec, bio_set->bvec_pools[pool_idx]);
120 mempool_free(bio, bio_set->bio_pool);
124 * default destructor for a bio allocated with bio_alloc_bioset()
126 static void bio_fs_destructor(struct bio *bio)
128 bio_free(bio, fs_bio_set);
131 void bio_init(struct bio *bio)
133 memset(bio, 0, sizeof(*bio));
134 bio->bi_flags = 1 << BIO_UPTODATE;
135 atomic_set(&bio->bi_cnt, 1);
139 * bio_alloc_bioset - allocate a bio for I/O
140 * @gfp_mask: the GFP_ mask given to the slab allocator
141 * @nr_iovecs: number of iovecs to pre-allocate
142 * @bs: the bio_set to allocate from
145 * bio_alloc_bioset will first try it's on mempool to satisfy the allocation.
146 * If %__GFP_WAIT is set then we will block on the internal pool waiting
147 * for a &struct bio to become free.
149 * allocate bio and iovecs from the memory pools specified by the
152 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
154 struct bio *bio = mempool_alloc(bs->bio_pool, gfp_mask);
157 struct bio_vec *bvl = NULL;
160 if (likely(nr_iovecs)) {
161 unsigned long uninitialized_var(idx);
163 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
164 if (unlikely(!bvl)) {
165 mempool_free(bio, bs->bio_pool);
169 bio->bi_flags |= idx << BIO_POOL_OFFSET;
170 bio->bi_max_vecs = bvec_slabs[idx].nr_vecs;
172 bio->bi_io_vec = bvl;
178 struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
180 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
183 bio->bi_destructor = bio_fs_destructor;
188 void zero_fill_bio(struct bio *bio)
194 bio_for_each_segment(bv, bio, i) {
195 char *data = bvec_kmap_irq(bv, &flags);
196 memset(data, 0, bv->bv_len);
197 flush_dcache_page(bv->bv_page);
198 bvec_kunmap_irq(data, &flags);
201 EXPORT_SYMBOL(zero_fill_bio);
204 * bio_put - release a reference to a bio
205 * @bio: bio to release reference to
208 * Put a reference to a &struct bio, either one you have gotten with
209 * bio_alloc or bio_get. The last put of a bio will free it.
211 void bio_put(struct bio *bio)
213 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
218 if (atomic_dec_and_test(&bio->bi_cnt)) {
220 bio->bi_destructor(bio);
224 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
226 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
227 blk_recount_segments(q, bio);
229 return bio->bi_phys_segments;
232 inline int bio_hw_segments(struct request_queue *q, struct bio *bio)
234 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
235 blk_recount_segments(q, bio);
237 return bio->bi_hw_segments;
241 * __bio_clone - clone a bio
242 * @bio: destination bio
243 * @bio_src: bio to clone
245 * Clone a &bio. Caller will own the returned bio, but not
246 * the actual data it points to. Reference count of returned
249 void __bio_clone(struct bio *bio, struct bio *bio_src)
251 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
252 bio_src->bi_max_vecs * sizeof(struct bio_vec));
255 * most users will be overriding ->bi_bdev with a new target,
256 * so we don't set nor calculate new physical/hw segment counts here
258 bio->bi_sector = bio_src->bi_sector;
259 bio->bi_bdev = bio_src->bi_bdev;
260 bio->bi_flags |= 1 << BIO_CLONED;
261 bio->bi_rw = bio_src->bi_rw;
262 bio->bi_vcnt = bio_src->bi_vcnt;
263 bio->bi_size = bio_src->bi_size;
264 bio->bi_idx = bio_src->bi_idx;
268 * bio_clone - clone a bio
270 * @gfp_mask: allocation priority
272 * Like __bio_clone, only also allocates the returned bio
274 struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
276 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
279 b->bi_destructor = bio_fs_destructor;
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 struct request_queue *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(struct request_queue *q, struct bio *bio, struct page
310 *page, unsigned int len, unsigned int offset,
311 unsigned short max_sectors)
313 int retried_segments = 0;
314 struct bio_vec *bvec;
317 * cloned bio must not modify vec list
319 if (unlikely(bio_flagged(bio, BIO_CLONED)))
322 if (((bio->bi_size + len) >> 9) > max_sectors)
326 * For filesystems with a blocksize smaller than the pagesize
327 * we will often be called with the same page as last time and
328 * a consecutive offset. Optimize this special case.
330 if (bio->bi_vcnt > 0) {
331 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
333 if (page == prev->bv_page &&
334 offset == prev->bv_offset + prev->bv_len) {
336 if (q->merge_bvec_fn &&
337 q->merge_bvec_fn(q, bio, prev) < len) {
346 if (bio->bi_vcnt >= bio->bi_max_vecs)
350 * we might lose a segment or two here, but rather that than
351 * make this too complex.
354 while (bio->bi_phys_segments >= q->max_phys_segments
355 || bio->bi_hw_segments >= q->max_hw_segments
356 || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) {
358 if (retried_segments)
361 retried_segments = 1;
362 blk_recount_segments(q, bio);
366 * setup the new entry, we might clear it again later if we
367 * cannot add the page
369 bvec = &bio->bi_io_vec[bio->bi_vcnt];
370 bvec->bv_page = page;
372 bvec->bv_offset = offset;
375 * if queue has other restrictions (eg varying max sector size
376 * depending on offset), it can specify a merge_bvec_fn in the
377 * queue to get further control
379 if (q->merge_bvec_fn) {
381 * merge_bvec_fn() returns number of bytes it can accept
384 if (q->merge_bvec_fn(q, bio, bvec) < len) {
385 bvec->bv_page = NULL;
392 /* If we may be able to merge these biovecs, force a recount */
393 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) ||
394 BIOVEC_VIRT_MERGEABLE(bvec-1, bvec)))
395 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
398 bio->bi_phys_segments++;
399 bio->bi_hw_segments++;
406 * bio_add_pc_page - attempt to add page to bio
407 * @q: the target queue
408 * @bio: destination bio
410 * @len: vec entry length
411 * @offset: vec entry offset
413 * Attempt to add a page to the bio_vec maplist. This can fail for a
414 * number of reasons, such as the bio being full or target block
415 * device limitations. The target block device must allow bio's
416 * smaller than PAGE_SIZE, so it is always possible to add a single
417 * page to an empty bio. This should only be used by REQ_PC bios.
419 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
420 unsigned int len, unsigned int offset)
422 return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors);
426 * bio_add_page - attempt to add page to bio
427 * @bio: destination bio
429 * @len: vec entry length
430 * @offset: vec entry offset
432 * Attempt to add a page to the bio_vec maplist. This can fail for a
433 * number of reasons, such as the bio being full or target block
434 * device limitations. The target block device must allow bio's
435 * smaller than PAGE_SIZE, so it is always possible to add a single
436 * page to an empty bio.
438 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
441 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
442 return __bio_add_page(q, bio, page, len, offset, q->max_sectors);
445 struct bio_map_data {
446 struct bio_vec *iovecs;
448 struct sg_iovec *sgvecs;
451 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
452 struct sg_iovec *iov, int iov_count)
454 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
455 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
456 bmd->nr_sgvecs = iov_count;
457 bio->bi_private = bmd;
460 static void bio_free_map_data(struct bio_map_data *bmd)
467 static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count,
470 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask);
475 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
481 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
490 static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
491 struct sg_iovec *iov, int iov_count, int uncopy)
494 struct bio_vec *bvec;
496 unsigned int iov_off = 0;
497 int read = bio_data_dir(bio) == READ;
499 __bio_for_each_segment(bvec, bio, i, 0) {
500 char *bv_addr = page_address(bvec->bv_page);
501 unsigned int bv_len = iovecs[i].bv_len;
503 while (bv_len && iov_idx < iov_count) {
507 bytes = min_t(unsigned int,
508 iov[iov_idx].iov_len - iov_off, bv_len);
509 iov_addr = iov[iov_idx].iov_base + iov_off;
512 if (!read && !uncopy)
513 ret = copy_from_user(bv_addr, iov_addr,
516 ret = copy_to_user(iov_addr, bv_addr,
528 if (iov[iov_idx].iov_len == iov_off) {
535 __free_page(bvec->bv_page);
542 * bio_uncopy_user - finish previously mapped bio
543 * @bio: bio being terminated
545 * Free pages allocated from bio_copy_user() and write back data
546 * to user space in case of a read.
548 int bio_uncopy_user(struct bio *bio)
550 struct bio_map_data *bmd = bio->bi_private;
553 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs, bmd->nr_sgvecs, 1);
555 bio_free_map_data(bmd);
561 * bio_copy_user_iov - copy user data to bio
562 * @q: destination block queue
564 * @iov_count: number of elements in the iovec
565 * @write_to_vm: bool indicating writing to pages or not
567 * Prepares and returns a bio for indirect user io, bouncing data
568 * to/from kernel pages as necessary. Must be paired with
569 * call bio_uncopy_user() on io completion.
571 struct bio *bio_copy_user_iov(struct request_queue *q, struct sg_iovec *iov,
572 int iov_count, int write_to_vm)
574 struct bio_map_data *bmd;
575 struct bio_vec *bvec;
580 unsigned int len = 0;
582 for (i = 0; i < iov_count; i++) {
587 uaddr = (unsigned long)iov[i].iov_base;
588 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
589 start = uaddr >> PAGE_SHIFT;
591 nr_pages += end - start;
592 len += iov[i].iov_len;
595 bmd = bio_alloc_map_data(nr_pages, iov_count, GFP_KERNEL);
597 return ERR_PTR(-ENOMEM);
600 bio = bio_alloc(GFP_KERNEL, nr_pages);
604 bio->bi_rw |= (!write_to_vm << BIO_RW);
608 unsigned int bytes = PAGE_SIZE;
613 page = alloc_page(q->bounce_gfp | GFP_KERNEL);
619 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
632 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0);
637 bio_set_map_data(bmd, bio, iov, iov_count);
640 bio_for_each_segment(bvec, bio, i)
641 __free_page(bvec->bv_page);
645 bio_free_map_data(bmd);
650 * bio_copy_user - copy user data to bio
651 * @q: destination block queue
652 * @uaddr: start of user address
653 * @len: length in bytes
654 * @write_to_vm: bool indicating writing to pages or not
656 * Prepares and returns a bio for indirect user io, bouncing data
657 * to/from kernel pages as necessary. Must be paired with
658 * call bio_uncopy_user() on io completion.
660 struct bio *bio_copy_user(struct request_queue *q, unsigned long uaddr,
661 unsigned int len, int write_to_vm)
665 iov.iov_base = (void __user *)uaddr;
668 return bio_copy_user_iov(q, &iov, 1, write_to_vm);
671 static struct bio *__bio_map_user_iov(struct request_queue *q,
672 struct block_device *bdev,
673 struct sg_iovec *iov, int iov_count,
683 for (i = 0; i < iov_count; i++) {
684 unsigned long uaddr = (unsigned long)iov[i].iov_base;
685 unsigned long len = iov[i].iov_len;
686 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
687 unsigned long start = uaddr >> PAGE_SHIFT;
689 nr_pages += end - start;
691 * buffer must be aligned to at least hardsector size for now
693 if (uaddr & queue_dma_alignment(q))
694 return ERR_PTR(-EINVAL);
698 return ERR_PTR(-EINVAL);
700 bio = bio_alloc(GFP_KERNEL, nr_pages);
702 return ERR_PTR(-ENOMEM);
705 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL);
709 for (i = 0; i < iov_count; i++) {
710 unsigned long uaddr = (unsigned long)iov[i].iov_base;
711 unsigned long len = iov[i].iov_len;
712 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
713 unsigned long start = uaddr >> PAGE_SHIFT;
714 const int local_nr_pages = end - start;
715 const int page_limit = cur_page + local_nr_pages;
717 down_read(¤t->mm->mmap_sem);
718 ret = get_user_pages(current, current->mm, uaddr,
720 write_to_vm, 0, &pages[cur_page], NULL);
721 up_read(¤t->mm->mmap_sem);
723 if (ret < local_nr_pages) {
728 offset = uaddr & ~PAGE_MASK;
729 for (j = cur_page; j < page_limit; j++) {
730 unsigned int bytes = PAGE_SIZE - offset;
741 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
751 * release the pages we didn't map into the bio, if any
753 while (j < page_limit)
754 page_cache_release(pages[j++]);
760 * set data direction, and check if mapped pages need bouncing
763 bio->bi_rw |= (1 << BIO_RW);
766 bio->bi_flags |= (1 << BIO_USER_MAPPED);
770 for (i = 0; i < nr_pages; i++) {
773 page_cache_release(pages[i]);
782 * bio_map_user - map user address into bio
783 * @q: the struct request_queue for the bio
784 * @bdev: destination block device
785 * @uaddr: start of user address
786 * @len: length in bytes
787 * @write_to_vm: bool indicating writing to pages or not
789 * Map the user space address into a bio suitable for io to a block
790 * device. Returns an error pointer in case of error.
792 struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
793 unsigned long uaddr, unsigned int len, int write_to_vm)
797 iov.iov_base = (void __user *)uaddr;
800 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm);
804 * bio_map_user_iov - map user sg_iovec table into bio
805 * @q: the struct request_queue for the bio
806 * @bdev: destination block device
808 * @iov_count: number of elements in the iovec
809 * @write_to_vm: bool indicating writing to pages or not
811 * Map the user space address into a bio suitable for io to a block
812 * device. Returns an error pointer in case of error.
814 struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
815 struct sg_iovec *iov, int iov_count,
820 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm);
826 * subtle -- if __bio_map_user() ended up bouncing a bio,
827 * it would normally disappear when its bi_end_io is run.
828 * however, we need it for the unmap, so grab an extra
836 static void __bio_unmap_user(struct bio *bio)
838 struct bio_vec *bvec;
842 * make sure we dirty pages we wrote to
844 __bio_for_each_segment(bvec, bio, i, 0) {
845 if (bio_data_dir(bio) == READ)
846 set_page_dirty_lock(bvec->bv_page);
848 page_cache_release(bvec->bv_page);
855 * bio_unmap_user - unmap a bio
856 * @bio: the bio being unmapped
858 * Unmap a bio previously mapped by bio_map_user(). Must be called with
861 * bio_unmap_user() may sleep.
863 void bio_unmap_user(struct bio *bio)
865 __bio_unmap_user(bio);
869 static void bio_map_kern_endio(struct bio *bio, int err)
875 static struct bio *__bio_map_kern(struct request_queue *q, void *data,
876 unsigned int len, gfp_t gfp_mask)
878 unsigned long kaddr = (unsigned long)data;
879 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
880 unsigned long start = kaddr >> PAGE_SHIFT;
881 const int nr_pages = end - start;
885 bio = bio_alloc(gfp_mask, nr_pages);
887 return ERR_PTR(-ENOMEM);
889 offset = offset_in_page(kaddr);
890 for (i = 0; i < nr_pages; i++) {
891 unsigned int bytes = PAGE_SIZE - offset;
899 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
908 bio->bi_end_io = bio_map_kern_endio;
913 * bio_map_kern - map kernel address into bio
914 * @q: the struct request_queue for the bio
915 * @data: pointer to buffer to map
916 * @len: length in bytes
917 * @gfp_mask: allocation flags for bio allocation
919 * Map the kernel address into a bio suitable for io to a block
920 * device. Returns an error pointer in case of error.
922 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
927 bio = __bio_map_kern(q, data, len, gfp_mask);
931 if (bio->bi_size == len)
935 * Don't support partial mappings.
938 return ERR_PTR(-EINVAL);
941 static void bio_copy_kern_endio(struct bio *bio, int err)
943 struct bio_vec *bvec;
944 const int read = bio_data_dir(bio) == READ;
945 struct bio_map_data *bmd = bio->bi_private;
947 char *p = bmd->sgvecs[0].iov_base;
949 __bio_for_each_segment(bvec, bio, i, 0) {
950 char *addr = page_address(bvec->bv_page);
951 int len = bmd->iovecs[i].bv_len;
954 memcpy(p, addr, len);
956 __free_page(bvec->bv_page);
960 bio_free_map_data(bmd);
965 * bio_copy_kern - copy kernel address into bio
966 * @q: the struct request_queue for the bio
967 * @data: pointer to buffer to copy
968 * @len: length in bytes
969 * @gfp_mask: allocation flags for bio and page allocation
970 * @reading: data direction is READ
972 * copy the kernel address into a bio suitable for io to a block
973 * device. Returns an error pointer in case of error.
975 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
976 gfp_t gfp_mask, int reading)
978 unsigned long kaddr = (unsigned long)data;
979 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
980 unsigned long start = kaddr >> PAGE_SHIFT;
981 const int nr_pages = end - start;
983 struct bio_vec *bvec;
984 struct bio_map_data *bmd;
991 bmd = bio_alloc_map_data(nr_pages, 1, gfp_mask);
993 return ERR_PTR(-ENOMEM);
996 bio = bio_alloc(gfp_mask, nr_pages);
1002 unsigned int bytes = PAGE_SIZE;
1007 page = alloc_page(q->bounce_gfp | gfp_mask);
1013 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) {
1024 bio_for_each_segment(bvec, bio, i) {
1025 char *addr = page_address(bvec->bv_page);
1027 memcpy(addr, p, bvec->bv_len);
1032 bio->bi_private = bmd;
1033 bio->bi_end_io = bio_copy_kern_endio;
1035 bio_set_map_data(bmd, bio, &iov, 1);
1038 bio_for_each_segment(bvec, bio, i)
1039 __free_page(bvec->bv_page);
1043 bio_free_map_data(bmd);
1045 return ERR_PTR(ret);
1049 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1050 * for performing direct-IO in BIOs.
1052 * The problem is that we cannot run set_page_dirty() from interrupt context
1053 * because the required locks are not interrupt-safe. So what we can do is to
1054 * mark the pages dirty _before_ performing IO. And in interrupt context,
1055 * check that the pages are still dirty. If so, fine. If not, redirty them
1056 * in process context.
1058 * We special-case compound pages here: normally this means reads into hugetlb
1059 * pages. The logic in here doesn't really work right for compound pages
1060 * because the VM does not uniformly chase down the head page in all cases.
1061 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1062 * handle them at all. So we skip compound pages here at an early stage.
1064 * Note that this code is very hard to test under normal circumstances because
1065 * direct-io pins the pages with get_user_pages(). This makes
1066 * is_page_cache_freeable return false, and the VM will not clean the pages.
1067 * But other code (eg, pdflush) could clean the pages if they are mapped
1070 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1071 * deferred bio dirtying paths.
1075 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1077 void bio_set_pages_dirty(struct bio *bio)
1079 struct bio_vec *bvec = bio->bi_io_vec;
1082 for (i = 0; i < bio->bi_vcnt; i++) {
1083 struct page *page = bvec[i].bv_page;
1085 if (page && !PageCompound(page))
1086 set_page_dirty_lock(page);
1090 static void bio_release_pages(struct bio *bio)
1092 struct bio_vec *bvec = bio->bi_io_vec;
1095 for (i = 0; i < bio->bi_vcnt; i++) {
1096 struct page *page = bvec[i].bv_page;
1104 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1105 * If they are, then fine. If, however, some pages are clean then they must
1106 * have been written out during the direct-IO read. So we take another ref on
1107 * the BIO and the offending pages and re-dirty the pages in process context.
1109 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1110 * here on. It will run one page_cache_release() against each page and will
1111 * run one bio_put() against the BIO.
1114 static void bio_dirty_fn(struct work_struct *work);
1116 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1117 static DEFINE_SPINLOCK(bio_dirty_lock);
1118 static struct bio *bio_dirty_list;
1121 * This runs in process context
1123 static void bio_dirty_fn(struct work_struct *work)
1125 unsigned long flags;
1128 spin_lock_irqsave(&bio_dirty_lock, flags);
1129 bio = bio_dirty_list;
1130 bio_dirty_list = NULL;
1131 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1134 struct bio *next = bio->bi_private;
1136 bio_set_pages_dirty(bio);
1137 bio_release_pages(bio);
1143 void bio_check_pages_dirty(struct bio *bio)
1145 struct bio_vec *bvec = bio->bi_io_vec;
1146 int nr_clean_pages = 0;
1149 for (i = 0; i < bio->bi_vcnt; i++) {
1150 struct page *page = bvec[i].bv_page;
1152 if (PageDirty(page) || PageCompound(page)) {
1153 page_cache_release(page);
1154 bvec[i].bv_page = NULL;
1160 if (nr_clean_pages) {
1161 unsigned long flags;
1163 spin_lock_irqsave(&bio_dirty_lock, flags);
1164 bio->bi_private = bio_dirty_list;
1165 bio_dirty_list = bio;
1166 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1167 schedule_work(&bio_dirty_work);
1174 * bio_endio - end I/O on a bio
1176 * @error: error, if any
1179 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1180 * preferred way to end I/O on a bio, it takes care of clearing
1181 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1182 * established -Exxxx (-EIO, for instance) error values in case
1183 * something went wrong. Noone should call bi_end_io() directly on a
1184 * bio unless they own it and thus know that it has an end_io
1187 void bio_endio(struct bio *bio, int error)
1190 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1191 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1195 bio->bi_end_io(bio, error);
1198 void bio_pair_release(struct bio_pair *bp)
1200 if (atomic_dec_and_test(&bp->cnt)) {
1201 struct bio *master = bp->bio1.bi_private;
1203 bio_endio(master, bp->error);
1204 mempool_free(bp, bp->bio2.bi_private);
1208 static void bio_pair_end_1(struct bio *bi, int err)
1210 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1215 bio_pair_release(bp);
1218 static void bio_pair_end_2(struct bio *bi, int err)
1220 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1225 bio_pair_release(bp);
1229 * split a bio - only worry about a bio with a single page
1232 struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors)
1234 struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO);
1239 blk_add_trace_pdu_int(bdev_get_queue(bi->bi_bdev), BLK_TA_SPLIT, bi,
1240 bi->bi_sector + first_sectors);
1242 BUG_ON(bi->bi_vcnt != 1);
1243 BUG_ON(bi->bi_idx != 0);
1244 atomic_set(&bp->cnt, 3);
1248 bp->bio2.bi_sector += first_sectors;
1249 bp->bio2.bi_size -= first_sectors << 9;
1250 bp->bio1.bi_size = first_sectors << 9;
1252 bp->bv1 = bi->bi_io_vec[0];
1253 bp->bv2 = bi->bi_io_vec[0];
1254 bp->bv2.bv_offset += first_sectors << 9;
1255 bp->bv2.bv_len -= first_sectors << 9;
1256 bp->bv1.bv_len = first_sectors << 9;
1258 bp->bio1.bi_io_vec = &bp->bv1;
1259 bp->bio2.bi_io_vec = &bp->bv2;
1261 bp->bio1.bi_max_vecs = 1;
1262 bp->bio2.bi_max_vecs = 1;
1264 bp->bio1.bi_end_io = bio_pair_end_1;
1265 bp->bio2.bi_end_io = bio_pair_end_2;
1267 bp->bio1.bi_private = bi;
1268 bp->bio2.bi_private = pool;
1275 * create memory pools for biovec's in a bio_set.
1276 * use the global biovec slabs created for general use.
1278 static int biovec_create_pools(struct bio_set *bs, int pool_entries)
1282 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1283 struct biovec_slab *bp = bvec_slabs + i;
1284 mempool_t **bvp = bs->bvec_pools + i;
1286 *bvp = mempool_create_slab_pool(pool_entries, bp->slab);
1293 static void biovec_free_pools(struct bio_set *bs)
1297 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1298 mempool_t *bvp = bs->bvec_pools[i];
1301 mempool_destroy(bvp);
1306 void bioset_free(struct bio_set *bs)
1309 mempool_destroy(bs->bio_pool);
1311 biovec_free_pools(bs);
1316 struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size)
1318 struct bio_set *bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1323 bs->bio_pool = mempool_create_slab_pool(bio_pool_size, bio_slab);
1327 if (!biovec_create_pools(bs, bvec_pool_size))
1335 static void __init biovec_init_slabs(void)
1339 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1341 struct biovec_slab *bvs = bvec_slabs + i;
1343 size = bvs->nr_vecs * sizeof(struct bio_vec);
1344 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1345 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1349 static int __init init_bio(void)
1351 bio_slab = KMEM_CACHE(bio, SLAB_HWCACHE_ALIGN|SLAB_PANIC);
1353 biovec_init_slabs();
1355 fs_bio_set = bioset_create(BIO_POOL_SIZE, 2);
1357 panic("bio: can't allocate bios\n");
1359 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1360 sizeof(struct bio_pair));
1361 if (!bio_split_pool)
1362 panic("bio: can't create split pool\n");
1367 subsys_initcall(init_bio);
1369 EXPORT_SYMBOL(bio_alloc);
1370 EXPORT_SYMBOL(bio_put);
1371 EXPORT_SYMBOL(bio_free);
1372 EXPORT_SYMBOL(bio_endio);
1373 EXPORT_SYMBOL(bio_init);
1374 EXPORT_SYMBOL(__bio_clone);
1375 EXPORT_SYMBOL(bio_clone);
1376 EXPORT_SYMBOL(bio_phys_segments);
1377 EXPORT_SYMBOL(bio_hw_segments);
1378 EXPORT_SYMBOL(bio_add_page);
1379 EXPORT_SYMBOL(bio_add_pc_page);
1380 EXPORT_SYMBOL(bio_get_nr_vecs);
1381 EXPORT_SYMBOL(bio_map_user);
1382 EXPORT_SYMBOL(bio_unmap_user);
1383 EXPORT_SYMBOL(bio_map_kern);
1384 EXPORT_SYMBOL(bio_copy_kern);
1385 EXPORT_SYMBOL(bio_pair_release);
1386 EXPORT_SYMBOL(bio_split);
1387 EXPORT_SYMBOL(bio_split_pool);
1388 EXPORT_SYMBOL(bio_copy_user);
1389 EXPORT_SYMBOL(bio_uncopy_user);
1390 EXPORT_SYMBOL(bioset_create);
1391 EXPORT_SYMBOL(bioset_free);
1392 EXPORT_SYMBOL(bio_alloc_bioset);