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 static struct kmem_cache *bio_slab __read_mostly;
33 mempool_t *bio_split_pool __read_mostly;
36 * if you change this list, also change bvec_alloc or things will
37 * break badly! cannot be bigger than what you can fit into an
41 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
42 static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
43 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
48 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
49 * IO code that does not need private memory pools.
51 struct bio_set *fs_bio_set;
53 struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, struct bio_set *bs)
58 * see comment near bvec_array define!
61 case 1 : *idx = 0; break;
62 case 2 ... 4: *idx = 1; break;
63 case 5 ... 16: *idx = 2; break;
64 case 17 ... 64: *idx = 3; break;
65 case 65 ... 128: *idx = 4; break;
66 case 129 ... BIO_MAX_PAGES: *idx = 5; break;
71 * idx now points to the pool we want to allocate from
74 bvl = mempool_alloc(bs->bvec_pools[*idx], gfp_mask);
76 struct biovec_slab *bp = bvec_slabs + *idx;
78 memset(bvl, 0, bp->nr_vecs * sizeof(struct bio_vec));
84 void bio_free(struct bio *bio, struct bio_set *bio_set)
87 const int pool_idx = BIO_POOL_IDX(bio);
89 BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS);
91 mempool_free(bio->bi_io_vec, bio_set->bvec_pools[pool_idx]);
94 mempool_free(bio, bio_set->bio_pool);
98 * default destructor for a bio allocated with bio_alloc_bioset()
100 static void bio_fs_destructor(struct bio *bio)
102 bio_free(bio, fs_bio_set);
105 void bio_init(struct bio *bio)
107 memset(bio, 0, sizeof(*bio));
108 bio->bi_flags = 1 << BIO_UPTODATE;
109 atomic_set(&bio->bi_cnt, 1);
113 * bio_alloc_bioset - allocate a bio for I/O
114 * @gfp_mask: the GFP_ mask given to the slab allocator
115 * @nr_iovecs: number of iovecs to pre-allocate
116 * @bs: the bio_set to allocate from
119 * bio_alloc_bioset will first try it's on mempool to satisfy the allocation.
120 * If %__GFP_WAIT is set then we will block on the internal pool waiting
121 * for a &struct bio to become free.
123 * allocate bio and iovecs from the memory pools specified by the
126 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
128 struct bio *bio = mempool_alloc(bs->bio_pool, gfp_mask);
131 struct bio_vec *bvl = NULL;
134 if (likely(nr_iovecs)) {
135 unsigned long uninitialized_var(idx);
137 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
138 if (unlikely(!bvl)) {
139 mempool_free(bio, bs->bio_pool);
143 bio->bi_flags |= idx << BIO_POOL_OFFSET;
144 bio->bi_max_vecs = bvec_slabs[idx].nr_vecs;
146 bio->bi_io_vec = bvl;
152 struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
154 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
157 bio->bi_destructor = bio_fs_destructor;
162 void zero_fill_bio(struct bio *bio)
168 bio_for_each_segment(bv, bio, i) {
169 char *data = bvec_kmap_irq(bv, &flags);
170 memset(data, 0, bv->bv_len);
171 flush_dcache_page(bv->bv_page);
172 bvec_kunmap_irq(data, &flags);
175 EXPORT_SYMBOL(zero_fill_bio);
178 * bio_put - release a reference to a bio
179 * @bio: bio to release reference to
182 * Put a reference to a &struct bio, either one you have gotten with
183 * bio_alloc or bio_get. The last put of a bio will free it.
185 void bio_put(struct bio *bio)
187 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
192 if (atomic_dec_and_test(&bio->bi_cnt)) {
194 bio->bi_destructor(bio);
198 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
200 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
201 blk_recount_segments(q, bio);
203 return bio->bi_phys_segments;
206 inline int bio_hw_segments(struct request_queue *q, struct bio *bio)
208 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
209 blk_recount_segments(q, bio);
211 return bio->bi_hw_segments;
215 * __bio_clone - clone a bio
216 * @bio: destination bio
217 * @bio_src: bio to clone
219 * Clone a &bio. Caller will own the returned bio, but not
220 * the actual data it points to. Reference count of returned
223 void __bio_clone(struct bio *bio, struct bio *bio_src)
225 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
226 bio_src->bi_max_vecs * sizeof(struct bio_vec));
229 * most users will be overriding ->bi_bdev with a new target,
230 * so we don't set nor calculate new physical/hw segment counts here
232 bio->bi_sector = bio_src->bi_sector;
233 bio->bi_bdev = bio_src->bi_bdev;
234 bio->bi_flags |= 1 << BIO_CLONED;
235 bio->bi_rw = bio_src->bi_rw;
236 bio->bi_vcnt = bio_src->bi_vcnt;
237 bio->bi_size = bio_src->bi_size;
238 bio->bi_idx = bio_src->bi_idx;
242 * bio_clone - clone a bio
244 * @gfp_mask: allocation priority
246 * Like __bio_clone, only also allocates the returned bio
248 struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
250 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
253 b->bi_destructor = bio_fs_destructor;
261 * bio_get_nr_vecs - return approx number of vecs
264 * Return the approximate number of pages we can send to this target.
265 * There's no guarantee that you will be able to fit this number of pages
266 * into a bio, it does not account for dynamic restrictions that vary
269 int bio_get_nr_vecs(struct block_device *bdev)
271 struct request_queue *q = bdev_get_queue(bdev);
274 nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
275 if (nr_pages > q->max_phys_segments)
276 nr_pages = q->max_phys_segments;
277 if (nr_pages > q->max_hw_segments)
278 nr_pages = q->max_hw_segments;
283 static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
284 *page, unsigned int len, unsigned int offset,
285 unsigned short max_sectors)
287 int retried_segments = 0;
288 struct bio_vec *bvec;
291 * cloned bio must not modify vec list
293 if (unlikely(bio_flagged(bio, BIO_CLONED)))
296 if (((bio->bi_size + len) >> 9) > max_sectors)
300 * For filesystems with a blocksize smaller than the pagesize
301 * we will often be called with the same page as last time and
302 * a consecutive offset. Optimize this special case.
304 if (bio->bi_vcnt > 0) {
305 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
307 if (page == prev->bv_page &&
308 offset == prev->bv_offset + prev->bv_len) {
310 if (q->merge_bvec_fn &&
311 q->merge_bvec_fn(q, bio, prev) < len) {
320 if (bio->bi_vcnt >= bio->bi_max_vecs)
324 * we might lose a segment or two here, but rather that than
325 * make this too complex.
328 while (bio->bi_phys_segments >= q->max_phys_segments
329 || bio->bi_hw_segments >= q->max_hw_segments
330 || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) {
332 if (retried_segments)
335 retried_segments = 1;
336 blk_recount_segments(q, bio);
340 * setup the new entry, we might clear it again later if we
341 * cannot add the page
343 bvec = &bio->bi_io_vec[bio->bi_vcnt];
344 bvec->bv_page = page;
346 bvec->bv_offset = offset;
349 * if queue has other restrictions (eg varying max sector size
350 * depending on offset), it can specify a merge_bvec_fn in the
351 * queue to get further control
353 if (q->merge_bvec_fn) {
355 * merge_bvec_fn() returns number of bytes it can accept
358 if (q->merge_bvec_fn(q, bio, bvec) < len) {
359 bvec->bv_page = NULL;
366 /* If we may be able to merge these biovecs, force a recount */
367 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) ||
368 BIOVEC_VIRT_MERGEABLE(bvec-1, bvec)))
369 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
372 bio->bi_phys_segments++;
373 bio->bi_hw_segments++;
380 * bio_add_pc_page - attempt to add page to bio
381 * @q: the target queue
382 * @bio: destination bio
384 * @len: vec entry length
385 * @offset: vec entry offset
387 * Attempt to add a page to the bio_vec maplist. This can fail for a
388 * number of reasons, such as the bio being full or target block
389 * device limitations. The target block device must allow bio's
390 * smaller than PAGE_SIZE, so it is always possible to add a single
391 * page to an empty bio. This should only be used by REQ_PC bios.
393 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
394 unsigned int len, unsigned int offset)
396 return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors);
400 * bio_add_page - attempt to add page to bio
401 * @bio: destination bio
403 * @len: vec entry length
404 * @offset: vec entry offset
406 * Attempt to add a page to the bio_vec maplist. This can fail for a
407 * number of reasons, such as the bio being full or target block
408 * device limitations. The target block device must allow bio's
409 * smaller than PAGE_SIZE, so it is always possible to add a single
410 * page to an empty bio.
412 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
415 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
416 return __bio_add_page(q, bio, page, len, offset, q->max_sectors);
419 struct bio_map_data {
420 struct bio_vec *iovecs;
422 struct sg_iovec *sgvecs;
425 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
426 struct sg_iovec *iov, int iov_count)
428 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
429 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
430 bmd->nr_sgvecs = iov_count;
431 bio->bi_private = bmd;
434 static void bio_free_map_data(struct bio_map_data *bmd)
441 static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count)
443 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), GFP_KERNEL);
448 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, GFP_KERNEL);
454 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, GFP_KERNEL);
463 static int __bio_copy_iov(struct bio *bio, struct sg_iovec *iov, int iov_count,
467 struct bio_vec *bvec;
469 unsigned int iov_off = 0;
470 int read = bio_data_dir(bio) == READ;
472 __bio_for_each_segment(bvec, bio, i, 0) {
473 char *bv_addr = page_address(bvec->bv_page);
474 unsigned int bv_len = bvec->bv_len;
476 while (bv_len && iov_idx < iov_count) {
480 bytes = min_t(unsigned int,
481 iov[iov_idx].iov_len - iov_off, bv_len);
482 iov_addr = iov[iov_idx].iov_base + iov_off;
485 if (!read && !uncopy)
486 ret = copy_from_user(bv_addr, iov_addr,
489 ret = copy_to_user(iov_addr, bv_addr,
501 if (iov[iov_idx].iov_len == iov_off) {
508 __free_page(bvec->bv_page);
515 * bio_uncopy_user - finish previously mapped bio
516 * @bio: bio being terminated
518 * Free pages allocated from bio_copy_user() and write back data
519 * to user space in case of a read.
521 int bio_uncopy_user(struct bio *bio)
523 struct bio_map_data *bmd = bio->bi_private;
526 ret = __bio_copy_iov(bio, bmd->sgvecs, bmd->nr_sgvecs, 1);
528 bio_free_map_data(bmd);
534 * bio_copy_user_iov - copy user data to bio
535 * @q: destination block queue
537 * @iov_count: number of elements in the iovec
538 * @write_to_vm: bool indicating writing to pages or not
540 * Prepares and returns a bio for indirect user io, bouncing data
541 * to/from kernel pages as necessary. Must be paired with
542 * call bio_uncopy_user() on io completion.
544 struct bio *bio_copy_user_iov(struct request_queue *q, struct sg_iovec *iov,
545 int iov_count, int write_to_vm)
547 struct bio_map_data *bmd;
548 struct bio_vec *bvec;
553 unsigned int len = 0;
555 for (i = 0; i < iov_count; i++) {
560 uaddr = (unsigned long)iov[i].iov_base;
561 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
562 start = uaddr >> PAGE_SHIFT;
564 nr_pages += end - start;
565 len += iov[i].iov_len;
568 bmd = bio_alloc_map_data(nr_pages, iov_count);
570 return ERR_PTR(-ENOMEM);
573 bio = bio_alloc(GFP_KERNEL, nr_pages);
577 bio->bi_rw |= (!write_to_vm << BIO_RW);
581 unsigned int bytes = PAGE_SIZE;
586 page = alloc_page(q->bounce_gfp | GFP_KERNEL);
592 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
605 ret = __bio_copy_iov(bio, iov, iov_count, 0);
610 bio_set_map_data(bmd, bio, iov, iov_count);
613 bio_for_each_segment(bvec, bio, i)
614 __free_page(bvec->bv_page);
618 bio_free_map_data(bmd);
623 * bio_copy_user - copy user data to bio
624 * @q: destination block queue
625 * @uaddr: start of user address
626 * @len: length in bytes
627 * @write_to_vm: bool indicating writing to pages or not
629 * Prepares and returns a bio for indirect user io, bouncing data
630 * to/from kernel pages as necessary. Must be paired with
631 * call bio_uncopy_user() on io completion.
633 struct bio *bio_copy_user(struct request_queue *q, unsigned long uaddr,
634 unsigned int len, int write_to_vm)
638 iov.iov_base = (void __user *)uaddr;
641 return bio_copy_user_iov(q, &iov, 1, write_to_vm);
644 static struct bio *__bio_map_user_iov(struct request_queue *q,
645 struct block_device *bdev,
646 struct sg_iovec *iov, int iov_count,
656 for (i = 0; i < iov_count; i++) {
657 unsigned long uaddr = (unsigned long)iov[i].iov_base;
658 unsigned long len = iov[i].iov_len;
659 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
660 unsigned long start = uaddr >> PAGE_SHIFT;
662 nr_pages += end - start;
664 * buffer must be aligned to at least hardsector size for now
666 if (uaddr & queue_dma_alignment(q))
667 return ERR_PTR(-EINVAL);
671 return ERR_PTR(-EINVAL);
673 bio = bio_alloc(GFP_KERNEL, nr_pages);
675 return ERR_PTR(-ENOMEM);
678 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL);
682 for (i = 0; i < iov_count; i++) {
683 unsigned long uaddr = (unsigned long)iov[i].iov_base;
684 unsigned long len = iov[i].iov_len;
685 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
686 unsigned long start = uaddr >> PAGE_SHIFT;
687 const int local_nr_pages = end - start;
688 const int page_limit = cur_page + local_nr_pages;
690 down_read(¤t->mm->mmap_sem);
691 ret = get_user_pages(current, current->mm, uaddr,
693 write_to_vm, 0, &pages[cur_page], NULL);
694 up_read(¤t->mm->mmap_sem);
696 if (ret < local_nr_pages) {
701 offset = uaddr & ~PAGE_MASK;
702 for (j = cur_page; j < page_limit; j++) {
703 unsigned int bytes = PAGE_SIZE - offset;
714 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
724 * release the pages we didn't map into the bio, if any
726 while (j < page_limit)
727 page_cache_release(pages[j++]);
733 * set data direction, and check if mapped pages need bouncing
736 bio->bi_rw |= (1 << BIO_RW);
739 bio->bi_flags |= (1 << BIO_USER_MAPPED);
743 for (i = 0; i < nr_pages; i++) {
746 page_cache_release(pages[i]);
755 * bio_map_user - map user address into bio
756 * @q: the struct request_queue for the bio
757 * @bdev: destination block device
758 * @uaddr: start of user address
759 * @len: length in bytes
760 * @write_to_vm: bool indicating writing to pages or not
762 * Map the user space address into a bio suitable for io to a block
763 * device. Returns an error pointer in case of error.
765 struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
766 unsigned long uaddr, unsigned int len, int write_to_vm)
770 iov.iov_base = (void __user *)uaddr;
773 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm);
777 * bio_map_user_iov - map user sg_iovec table into bio
778 * @q: the struct request_queue for the bio
779 * @bdev: destination block device
781 * @iov_count: number of elements in the iovec
782 * @write_to_vm: bool indicating writing to pages or not
784 * Map the user space address into a bio suitable for io to a block
785 * device. Returns an error pointer in case of error.
787 struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
788 struct sg_iovec *iov, int iov_count,
793 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm);
799 * subtle -- if __bio_map_user() ended up bouncing a bio,
800 * it would normally disappear when its bi_end_io is run.
801 * however, we need it for the unmap, so grab an extra
809 static void __bio_unmap_user(struct bio *bio)
811 struct bio_vec *bvec;
815 * make sure we dirty pages we wrote to
817 __bio_for_each_segment(bvec, bio, i, 0) {
818 if (bio_data_dir(bio) == READ)
819 set_page_dirty_lock(bvec->bv_page);
821 page_cache_release(bvec->bv_page);
828 * bio_unmap_user - unmap a bio
829 * @bio: the bio being unmapped
831 * Unmap a bio previously mapped by bio_map_user(). Must be called with
834 * bio_unmap_user() may sleep.
836 void bio_unmap_user(struct bio *bio)
838 __bio_unmap_user(bio);
842 static void bio_map_kern_endio(struct bio *bio, int err)
848 static struct bio *__bio_map_kern(struct request_queue *q, void *data,
849 unsigned int len, gfp_t gfp_mask)
851 unsigned long kaddr = (unsigned long)data;
852 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
853 unsigned long start = kaddr >> PAGE_SHIFT;
854 const int nr_pages = end - start;
858 bio = bio_alloc(gfp_mask, nr_pages);
860 return ERR_PTR(-ENOMEM);
862 offset = offset_in_page(kaddr);
863 for (i = 0; i < nr_pages; i++) {
864 unsigned int bytes = PAGE_SIZE - offset;
872 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
881 bio->bi_end_io = bio_map_kern_endio;
886 * bio_map_kern - map kernel address into bio
887 * @q: the struct request_queue for the bio
888 * @data: pointer to buffer to map
889 * @len: length in bytes
890 * @gfp_mask: allocation flags for bio allocation
892 * Map the kernel address into a bio suitable for io to a block
893 * device. Returns an error pointer in case of error.
895 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
900 bio = __bio_map_kern(q, data, len, gfp_mask);
904 if (bio->bi_size == len)
908 * Don't support partial mappings.
911 return ERR_PTR(-EINVAL);
914 static void bio_copy_kern_endio(struct bio *bio, int err)
916 struct bio_vec *bvec;
917 const int read = bio_data_dir(bio) == READ;
918 char *p = bio->bi_private;
921 __bio_for_each_segment(bvec, bio, i, 0) {
922 char *addr = page_address(bvec->bv_page);
925 memcpy(p, addr, bvec->bv_len);
927 __free_page(bvec->bv_page);
935 * bio_copy_kern - copy kernel address into bio
936 * @q: the struct request_queue for the bio
937 * @data: pointer to buffer to copy
938 * @len: length in bytes
939 * @gfp_mask: allocation flags for bio and page allocation
940 * @reading: data direction is READ
942 * copy the kernel address into a bio suitable for io to a block
943 * device. Returns an error pointer in case of error.
945 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
946 gfp_t gfp_mask, int reading)
948 unsigned long kaddr = (unsigned long)data;
949 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
950 unsigned long start = kaddr >> PAGE_SHIFT;
951 const int nr_pages = end - start;
953 struct bio_vec *bvec;
956 bio = bio_alloc(gfp_mask, nr_pages);
958 return ERR_PTR(-ENOMEM);
962 unsigned int bytes = PAGE_SIZE;
967 page = alloc_page(q->bounce_gfp | gfp_mask);
973 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) {
984 bio_for_each_segment(bvec, bio, i) {
985 char *addr = page_address(bvec->bv_page);
987 memcpy(addr, p, bvec->bv_len);
992 bio->bi_private = data;
993 bio->bi_end_io = bio_copy_kern_endio;
996 bio_for_each_segment(bvec, bio, i)
997 __free_page(bvec->bv_page);
1001 return ERR_PTR(ret);
1005 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1006 * for performing direct-IO in BIOs.
1008 * The problem is that we cannot run set_page_dirty() from interrupt context
1009 * because the required locks are not interrupt-safe. So what we can do is to
1010 * mark the pages dirty _before_ performing IO. And in interrupt context,
1011 * check that the pages are still dirty. If so, fine. If not, redirty them
1012 * in process context.
1014 * We special-case compound pages here: normally this means reads into hugetlb
1015 * pages. The logic in here doesn't really work right for compound pages
1016 * because the VM does not uniformly chase down the head page in all cases.
1017 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1018 * handle them at all. So we skip compound pages here at an early stage.
1020 * Note that this code is very hard to test under normal circumstances because
1021 * direct-io pins the pages with get_user_pages(). This makes
1022 * is_page_cache_freeable return false, and the VM will not clean the pages.
1023 * But other code (eg, pdflush) could clean the pages if they are mapped
1026 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1027 * deferred bio dirtying paths.
1031 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1033 void bio_set_pages_dirty(struct bio *bio)
1035 struct bio_vec *bvec = bio->bi_io_vec;
1038 for (i = 0; i < bio->bi_vcnt; i++) {
1039 struct page *page = bvec[i].bv_page;
1041 if (page && !PageCompound(page))
1042 set_page_dirty_lock(page);
1046 static void bio_release_pages(struct bio *bio)
1048 struct bio_vec *bvec = bio->bi_io_vec;
1051 for (i = 0; i < bio->bi_vcnt; i++) {
1052 struct page *page = bvec[i].bv_page;
1060 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1061 * If they are, then fine. If, however, some pages are clean then they must
1062 * have been written out during the direct-IO read. So we take another ref on
1063 * the BIO and the offending pages and re-dirty the pages in process context.
1065 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1066 * here on. It will run one page_cache_release() against each page and will
1067 * run one bio_put() against the BIO.
1070 static void bio_dirty_fn(struct work_struct *work);
1072 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1073 static DEFINE_SPINLOCK(bio_dirty_lock);
1074 static struct bio *bio_dirty_list;
1077 * This runs in process context
1079 static void bio_dirty_fn(struct work_struct *work)
1081 unsigned long flags;
1084 spin_lock_irqsave(&bio_dirty_lock, flags);
1085 bio = bio_dirty_list;
1086 bio_dirty_list = NULL;
1087 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1090 struct bio *next = bio->bi_private;
1092 bio_set_pages_dirty(bio);
1093 bio_release_pages(bio);
1099 void bio_check_pages_dirty(struct bio *bio)
1101 struct bio_vec *bvec = bio->bi_io_vec;
1102 int nr_clean_pages = 0;
1105 for (i = 0; i < bio->bi_vcnt; i++) {
1106 struct page *page = bvec[i].bv_page;
1108 if (PageDirty(page) || PageCompound(page)) {
1109 page_cache_release(page);
1110 bvec[i].bv_page = NULL;
1116 if (nr_clean_pages) {
1117 unsigned long flags;
1119 spin_lock_irqsave(&bio_dirty_lock, flags);
1120 bio->bi_private = bio_dirty_list;
1121 bio_dirty_list = bio;
1122 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1123 schedule_work(&bio_dirty_work);
1130 * bio_endio - end I/O on a bio
1132 * @error: error, if any
1135 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1136 * preferred way to end I/O on a bio, it takes care of clearing
1137 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1138 * established -Exxxx (-EIO, for instance) error values in case
1139 * something went wrong. Noone should call bi_end_io() directly on a
1140 * bio unless they own it and thus know that it has an end_io
1143 void bio_endio(struct bio *bio, int error)
1146 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1147 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1151 bio->bi_end_io(bio, error);
1154 void bio_pair_release(struct bio_pair *bp)
1156 if (atomic_dec_and_test(&bp->cnt)) {
1157 struct bio *master = bp->bio1.bi_private;
1159 bio_endio(master, bp->error);
1160 mempool_free(bp, bp->bio2.bi_private);
1164 static void bio_pair_end_1(struct bio *bi, int err)
1166 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1171 bio_pair_release(bp);
1174 static void bio_pair_end_2(struct bio *bi, int err)
1176 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1181 bio_pair_release(bp);
1185 * split a bio - only worry about a bio with a single page
1188 struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors)
1190 struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO);
1195 blk_add_trace_pdu_int(bdev_get_queue(bi->bi_bdev), BLK_TA_SPLIT, bi,
1196 bi->bi_sector + first_sectors);
1198 BUG_ON(bi->bi_vcnt != 1);
1199 BUG_ON(bi->bi_idx != 0);
1200 atomic_set(&bp->cnt, 3);
1204 bp->bio2.bi_sector += first_sectors;
1205 bp->bio2.bi_size -= first_sectors << 9;
1206 bp->bio1.bi_size = first_sectors << 9;
1208 bp->bv1 = bi->bi_io_vec[0];
1209 bp->bv2 = bi->bi_io_vec[0];
1210 bp->bv2.bv_offset += first_sectors << 9;
1211 bp->bv2.bv_len -= first_sectors << 9;
1212 bp->bv1.bv_len = first_sectors << 9;
1214 bp->bio1.bi_io_vec = &bp->bv1;
1215 bp->bio2.bi_io_vec = &bp->bv2;
1217 bp->bio1.bi_max_vecs = 1;
1218 bp->bio2.bi_max_vecs = 1;
1220 bp->bio1.bi_end_io = bio_pair_end_1;
1221 bp->bio2.bi_end_io = bio_pair_end_2;
1223 bp->bio1.bi_private = bi;
1224 bp->bio2.bi_private = pool;
1231 * create memory pools for biovec's in a bio_set.
1232 * use the global biovec slabs created for general use.
1234 static int biovec_create_pools(struct bio_set *bs, int pool_entries)
1238 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1239 struct biovec_slab *bp = bvec_slabs + i;
1240 mempool_t **bvp = bs->bvec_pools + i;
1242 *bvp = mempool_create_slab_pool(pool_entries, bp->slab);
1249 static void biovec_free_pools(struct bio_set *bs)
1253 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1254 mempool_t *bvp = bs->bvec_pools[i];
1257 mempool_destroy(bvp);
1262 void bioset_free(struct bio_set *bs)
1265 mempool_destroy(bs->bio_pool);
1267 biovec_free_pools(bs);
1272 struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size)
1274 struct bio_set *bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1279 bs->bio_pool = mempool_create_slab_pool(bio_pool_size, bio_slab);
1283 if (!biovec_create_pools(bs, bvec_pool_size))
1291 static void __init biovec_init_slabs(void)
1295 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1297 struct biovec_slab *bvs = bvec_slabs + i;
1299 size = bvs->nr_vecs * sizeof(struct bio_vec);
1300 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1301 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1305 static int __init init_bio(void)
1307 bio_slab = KMEM_CACHE(bio, SLAB_HWCACHE_ALIGN|SLAB_PANIC);
1309 biovec_init_slabs();
1311 fs_bio_set = bioset_create(BIO_POOL_SIZE, 2);
1313 panic("bio: can't allocate bios\n");
1315 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1316 sizeof(struct bio_pair));
1317 if (!bio_split_pool)
1318 panic("bio: can't create split pool\n");
1323 subsys_initcall(init_bio);
1325 EXPORT_SYMBOL(bio_alloc);
1326 EXPORT_SYMBOL(bio_put);
1327 EXPORT_SYMBOL(bio_free);
1328 EXPORT_SYMBOL(bio_endio);
1329 EXPORT_SYMBOL(bio_init);
1330 EXPORT_SYMBOL(__bio_clone);
1331 EXPORT_SYMBOL(bio_clone);
1332 EXPORT_SYMBOL(bio_phys_segments);
1333 EXPORT_SYMBOL(bio_hw_segments);
1334 EXPORT_SYMBOL(bio_add_page);
1335 EXPORT_SYMBOL(bio_add_pc_page);
1336 EXPORT_SYMBOL(bio_get_nr_vecs);
1337 EXPORT_SYMBOL(bio_map_user);
1338 EXPORT_SYMBOL(bio_unmap_user);
1339 EXPORT_SYMBOL(bio_map_kern);
1340 EXPORT_SYMBOL(bio_copy_kern);
1341 EXPORT_SYMBOL(bio_pair_release);
1342 EXPORT_SYMBOL(bio_split);
1343 EXPORT_SYMBOL(bio_split_pool);
1344 EXPORT_SYMBOL(bio_copy_user);
1345 EXPORT_SYMBOL(bio_uncopy_user);
1346 EXPORT_SYMBOL(bioset_create);
1347 EXPORT_SYMBOL(bioset_free);
1348 EXPORT_SYMBOL(bio_alloc_bioset);