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 unsigned int bvec_nr_vecs(unsigned short idx)
55 return bvec_slabs[idx].nr_vecs;
58 struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx, struct bio_set *bs)
63 * see comment near bvec_array define!
66 case 1 : *idx = 0; break;
67 case 2 ... 4: *idx = 1; break;
68 case 5 ... 16: *idx = 2; break;
69 case 17 ... 64: *idx = 3; break;
70 case 65 ... 128: *idx = 4; break;
71 case 129 ... BIO_MAX_PAGES: *idx = 5; break;
76 * idx now points to the pool we want to allocate from
79 bvl = mempool_alloc(bs->bvec_pools[*idx], gfp_mask);
81 struct biovec_slab *bp = bvec_slabs + *idx;
83 memset(bvl, 0, bp->nr_vecs * sizeof(struct bio_vec));
89 void bio_free(struct bio *bio, struct bio_set *bio_set)
92 const int pool_idx = BIO_POOL_IDX(bio);
94 BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS);
96 mempool_free(bio->bi_io_vec, bio_set->bvec_pools[pool_idx]);
99 if (bio_integrity(bio))
100 bio_integrity_free(bio, bio_set);
102 mempool_free(bio, bio_set->bio_pool);
106 * default destructor for a bio allocated with bio_alloc_bioset()
108 static void bio_fs_destructor(struct bio *bio)
110 bio_free(bio, fs_bio_set);
113 void bio_init(struct bio *bio)
115 memset(bio, 0, sizeof(*bio));
116 bio->bi_flags = 1 << BIO_UPTODATE;
117 atomic_set(&bio->bi_cnt, 1);
121 * bio_alloc_bioset - allocate a bio for I/O
122 * @gfp_mask: the GFP_ mask given to the slab allocator
123 * @nr_iovecs: number of iovecs to pre-allocate
124 * @bs: the bio_set to allocate from
127 * bio_alloc_bioset will first try it's on mempool to satisfy the allocation.
128 * If %__GFP_WAIT is set then we will block on the internal pool waiting
129 * for a &struct bio to become free.
131 * allocate bio and iovecs from the memory pools specified by the
134 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
136 struct bio *bio = mempool_alloc(bs->bio_pool, gfp_mask);
139 struct bio_vec *bvl = NULL;
142 if (likely(nr_iovecs)) {
143 unsigned long uninitialized_var(idx);
145 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
146 if (unlikely(!bvl)) {
147 mempool_free(bio, bs->bio_pool);
151 bio->bi_flags |= idx << BIO_POOL_OFFSET;
152 bio->bi_max_vecs = bvec_slabs[idx].nr_vecs;
154 bio->bi_io_vec = bvl;
160 struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
162 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
165 bio->bi_destructor = bio_fs_destructor;
170 void zero_fill_bio(struct bio *bio)
176 bio_for_each_segment(bv, bio, i) {
177 char *data = bvec_kmap_irq(bv, &flags);
178 memset(data, 0, bv->bv_len);
179 flush_dcache_page(bv->bv_page);
180 bvec_kunmap_irq(data, &flags);
183 EXPORT_SYMBOL(zero_fill_bio);
186 * bio_put - release a reference to a bio
187 * @bio: bio to release reference to
190 * Put a reference to a &struct bio, either one you have gotten with
191 * bio_alloc or bio_get. The last put of a bio will free it.
193 void bio_put(struct bio *bio)
195 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
200 if (atomic_dec_and_test(&bio->bi_cnt)) {
202 bio->bi_destructor(bio);
206 inline int bio_phys_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_phys_segments;
214 inline int bio_hw_segments(struct request_queue *q, struct bio *bio)
216 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
217 blk_recount_segments(q, bio);
219 return bio->bi_hw_segments;
223 * __bio_clone - clone a bio
224 * @bio: destination bio
225 * @bio_src: bio to clone
227 * Clone a &bio. Caller will own the returned bio, but not
228 * the actual data it points to. Reference count of returned
231 void __bio_clone(struct bio *bio, struct bio *bio_src)
233 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
234 bio_src->bi_max_vecs * sizeof(struct bio_vec));
237 * most users will be overriding ->bi_bdev with a new target,
238 * so we don't set nor calculate new physical/hw segment counts here
240 bio->bi_sector = bio_src->bi_sector;
241 bio->bi_bdev = bio_src->bi_bdev;
242 bio->bi_flags |= 1 << BIO_CLONED;
243 bio->bi_rw = bio_src->bi_rw;
244 bio->bi_vcnt = bio_src->bi_vcnt;
245 bio->bi_size = bio_src->bi_size;
246 bio->bi_idx = bio_src->bi_idx;
250 * bio_clone - clone a bio
252 * @gfp_mask: allocation priority
254 * Like __bio_clone, only also allocates the returned bio
256 struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
258 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
263 b->bi_destructor = bio_fs_destructor;
266 if (bio_integrity(bio)) {
269 ret = bio_integrity_clone(b, bio, fs_bio_set);
279 * bio_get_nr_vecs - return approx number of vecs
282 * Return the approximate number of pages we can send to this target.
283 * There's no guarantee that you will be able to fit this number of pages
284 * into a bio, it does not account for dynamic restrictions that vary
287 int bio_get_nr_vecs(struct block_device *bdev)
289 struct request_queue *q = bdev_get_queue(bdev);
292 nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
293 if (nr_pages > q->max_phys_segments)
294 nr_pages = q->max_phys_segments;
295 if (nr_pages > q->max_hw_segments)
296 nr_pages = q->max_hw_segments;
301 static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
302 *page, unsigned int len, unsigned int offset,
303 unsigned short max_sectors)
305 int retried_segments = 0;
306 struct bio_vec *bvec;
309 * cloned bio must not modify vec list
311 if (unlikely(bio_flagged(bio, BIO_CLONED)))
314 if (((bio->bi_size + len) >> 9) > max_sectors)
318 * For filesystems with a blocksize smaller than the pagesize
319 * we will often be called with the same page as last time and
320 * a consecutive offset. Optimize this special case.
322 if (bio->bi_vcnt > 0) {
323 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
325 if (page == prev->bv_page &&
326 offset == prev->bv_offset + prev->bv_len) {
328 if (q->merge_bvec_fn &&
329 q->merge_bvec_fn(q, bio, prev) < len) {
338 if (bio->bi_vcnt >= bio->bi_max_vecs)
342 * we might lose a segment or two here, but rather that than
343 * make this too complex.
346 while (bio->bi_phys_segments >= q->max_phys_segments
347 || bio->bi_hw_segments >= q->max_hw_segments
348 || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) {
350 if (retried_segments)
353 retried_segments = 1;
354 blk_recount_segments(q, bio);
358 * setup the new entry, we might clear it again later if we
359 * cannot add the page
361 bvec = &bio->bi_io_vec[bio->bi_vcnt];
362 bvec->bv_page = page;
364 bvec->bv_offset = offset;
367 * if queue has other restrictions (eg varying max sector size
368 * depending on offset), it can specify a merge_bvec_fn in the
369 * queue to get further control
371 if (q->merge_bvec_fn) {
373 * merge_bvec_fn() returns number of bytes it can accept
376 if (q->merge_bvec_fn(q, bio, bvec) < len) {
377 bvec->bv_page = NULL;
384 /* If we may be able to merge these biovecs, force a recount */
385 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) ||
386 BIOVEC_VIRT_MERGEABLE(bvec-1, bvec)))
387 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
390 bio->bi_phys_segments++;
391 bio->bi_hw_segments++;
398 * bio_add_pc_page - attempt to add page to bio
399 * @q: the target queue
400 * @bio: destination bio
402 * @len: vec entry length
403 * @offset: vec entry offset
405 * Attempt to add a page to the bio_vec maplist. This can fail for a
406 * number of reasons, such as the bio being full or target block
407 * device limitations. The target block device must allow bio's
408 * smaller than PAGE_SIZE, so it is always possible to add a single
409 * page to an empty bio. This should only be used by REQ_PC bios.
411 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
412 unsigned int len, unsigned int offset)
414 return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors);
418 * bio_add_page - attempt to add page to bio
419 * @bio: destination bio
421 * @len: vec entry length
422 * @offset: vec entry offset
424 * Attempt to add a page to the bio_vec maplist. This can fail for a
425 * number of reasons, such as the bio being full or target block
426 * device limitations. The target block device must allow bio's
427 * smaller than PAGE_SIZE, so it is always possible to add a single
428 * page to an empty bio.
430 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
433 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
434 return __bio_add_page(q, bio, page, len, offset, q->max_sectors);
437 struct bio_map_data {
438 struct bio_vec *iovecs;
440 struct sg_iovec *sgvecs;
443 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
444 struct sg_iovec *iov, int iov_count)
446 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
447 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
448 bmd->nr_sgvecs = iov_count;
449 bio->bi_private = bmd;
452 static void bio_free_map_data(struct bio_map_data *bmd)
459 static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count)
461 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), GFP_KERNEL);
466 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, GFP_KERNEL);
472 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, GFP_KERNEL);
481 static int __bio_copy_iov(struct bio *bio, struct sg_iovec *iov, int iov_count,
485 struct bio_vec *bvec;
487 unsigned int iov_off = 0;
488 int read = bio_data_dir(bio) == READ;
490 __bio_for_each_segment(bvec, bio, i, 0) {
491 char *bv_addr = page_address(bvec->bv_page);
492 unsigned int bv_len = bvec->bv_len;
494 while (bv_len && iov_idx < iov_count) {
498 bytes = min_t(unsigned int,
499 iov[iov_idx].iov_len - iov_off, bv_len);
500 iov_addr = iov[iov_idx].iov_base + iov_off;
503 if (!read && !uncopy)
504 ret = copy_from_user(bv_addr, iov_addr,
507 ret = copy_to_user(iov_addr, bv_addr,
519 if (iov[iov_idx].iov_len == iov_off) {
526 __free_page(bvec->bv_page);
533 * bio_uncopy_user - finish previously mapped bio
534 * @bio: bio being terminated
536 * Free pages allocated from bio_copy_user() and write back data
537 * to user space in case of a read.
539 int bio_uncopy_user(struct bio *bio)
541 struct bio_map_data *bmd = bio->bi_private;
544 ret = __bio_copy_iov(bio, bmd->sgvecs, bmd->nr_sgvecs, 1);
546 bio_free_map_data(bmd);
552 * bio_copy_user_iov - copy user data to bio
553 * @q: destination block queue
555 * @iov_count: number of elements in the iovec
556 * @write_to_vm: bool indicating writing to pages or not
558 * Prepares and returns a bio for indirect user io, bouncing data
559 * to/from kernel pages as necessary. Must be paired with
560 * call bio_uncopy_user() on io completion.
562 struct bio *bio_copy_user_iov(struct request_queue *q, struct sg_iovec *iov,
563 int iov_count, int write_to_vm)
565 struct bio_map_data *bmd;
566 struct bio_vec *bvec;
571 unsigned int len = 0;
573 for (i = 0; i < iov_count; i++) {
578 uaddr = (unsigned long)iov[i].iov_base;
579 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
580 start = uaddr >> PAGE_SHIFT;
582 nr_pages += end - start;
583 len += iov[i].iov_len;
586 bmd = bio_alloc_map_data(nr_pages, iov_count);
588 return ERR_PTR(-ENOMEM);
591 bio = bio_alloc(GFP_KERNEL, nr_pages);
595 bio->bi_rw |= (!write_to_vm << BIO_RW);
599 unsigned int bytes = PAGE_SIZE;
604 page = alloc_page(q->bounce_gfp | GFP_KERNEL);
610 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
623 ret = __bio_copy_iov(bio, iov, iov_count, 0);
628 bio_set_map_data(bmd, bio, iov, iov_count);
631 bio_for_each_segment(bvec, bio, i)
632 __free_page(bvec->bv_page);
636 bio_free_map_data(bmd);
641 * bio_copy_user - copy user data to bio
642 * @q: destination block queue
643 * @uaddr: start of user address
644 * @len: length in bytes
645 * @write_to_vm: bool indicating writing to pages or not
647 * Prepares and returns a bio for indirect user io, bouncing data
648 * to/from kernel pages as necessary. Must be paired with
649 * call bio_uncopy_user() on io completion.
651 struct bio *bio_copy_user(struct request_queue *q, unsigned long uaddr,
652 unsigned int len, int write_to_vm)
656 iov.iov_base = (void __user *)uaddr;
659 return bio_copy_user_iov(q, &iov, 1, write_to_vm);
662 static struct bio *__bio_map_user_iov(struct request_queue *q,
663 struct block_device *bdev,
664 struct sg_iovec *iov, int iov_count,
674 for (i = 0; i < iov_count; i++) {
675 unsigned long uaddr = (unsigned long)iov[i].iov_base;
676 unsigned long len = iov[i].iov_len;
677 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
678 unsigned long start = uaddr >> PAGE_SHIFT;
680 nr_pages += end - start;
682 * buffer must be aligned to at least hardsector size for now
684 if (uaddr & queue_dma_alignment(q))
685 return ERR_PTR(-EINVAL);
689 return ERR_PTR(-EINVAL);
691 bio = bio_alloc(GFP_KERNEL, nr_pages);
693 return ERR_PTR(-ENOMEM);
696 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL);
700 for (i = 0; i < iov_count; i++) {
701 unsigned long uaddr = (unsigned long)iov[i].iov_base;
702 unsigned long len = iov[i].iov_len;
703 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
704 unsigned long start = uaddr >> PAGE_SHIFT;
705 const int local_nr_pages = end - start;
706 const int page_limit = cur_page + local_nr_pages;
708 down_read(¤t->mm->mmap_sem);
709 ret = get_user_pages(current, current->mm, uaddr,
711 write_to_vm, 0, &pages[cur_page], NULL);
712 up_read(¤t->mm->mmap_sem);
714 if (ret < local_nr_pages) {
719 offset = uaddr & ~PAGE_MASK;
720 for (j = cur_page; j < page_limit; j++) {
721 unsigned int bytes = PAGE_SIZE - offset;
732 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
742 * release the pages we didn't map into the bio, if any
744 while (j < page_limit)
745 page_cache_release(pages[j++]);
751 * set data direction, and check if mapped pages need bouncing
754 bio->bi_rw |= (1 << BIO_RW);
757 bio->bi_flags |= (1 << BIO_USER_MAPPED);
761 for (i = 0; i < nr_pages; i++) {
764 page_cache_release(pages[i]);
773 * bio_map_user - map user address into bio
774 * @q: the struct request_queue for the bio
775 * @bdev: destination block device
776 * @uaddr: start of user address
777 * @len: length in bytes
778 * @write_to_vm: bool indicating writing to pages or not
780 * Map the user space address into a bio suitable for io to a block
781 * device. Returns an error pointer in case of error.
783 struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
784 unsigned long uaddr, unsigned int len, int write_to_vm)
788 iov.iov_base = (void __user *)uaddr;
791 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm);
795 * bio_map_user_iov - map user sg_iovec table into bio
796 * @q: the struct request_queue for the bio
797 * @bdev: destination block device
799 * @iov_count: number of elements in the iovec
800 * @write_to_vm: bool indicating writing to pages or not
802 * Map the user space address into a bio suitable for io to a block
803 * device. Returns an error pointer in case of error.
805 struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
806 struct sg_iovec *iov, int iov_count,
811 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm);
817 * subtle -- if __bio_map_user() ended up bouncing a bio,
818 * it would normally disappear when its bi_end_io is run.
819 * however, we need it for the unmap, so grab an extra
827 static void __bio_unmap_user(struct bio *bio)
829 struct bio_vec *bvec;
833 * make sure we dirty pages we wrote to
835 __bio_for_each_segment(bvec, bio, i, 0) {
836 if (bio_data_dir(bio) == READ)
837 set_page_dirty_lock(bvec->bv_page);
839 page_cache_release(bvec->bv_page);
846 * bio_unmap_user - unmap a bio
847 * @bio: the bio being unmapped
849 * Unmap a bio previously mapped by bio_map_user(). Must be called with
852 * bio_unmap_user() may sleep.
854 void bio_unmap_user(struct bio *bio)
856 __bio_unmap_user(bio);
860 static void bio_map_kern_endio(struct bio *bio, int err)
866 static struct bio *__bio_map_kern(struct request_queue *q, void *data,
867 unsigned int len, gfp_t gfp_mask)
869 unsigned long kaddr = (unsigned long)data;
870 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
871 unsigned long start = kaddr >> PAGE_SHIFT;
872 const int nr_pages = end - start;
876 bio = bio_alloc(gfp_mask, nr_pages);
878 return ERR_PTR(-ENOMEM);
880 offset = offset_in_page(kaddr);
881 for (i = 0; i < nr_pages; i++) {
882 unsigned int bytes = PAGE_SIZE - offset;
890 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
899 bio->bi_end_io = bio_map_kern_endio;
904 * bio_map_kern - map kernel address into bio
905 * @q: the struct request_queue for the bio
906 * @data: pointer to buffer to map
907 * @len: length in bytes
908 * @gfp_mask: allocation flags for bio allocation
910 * Map the kernel address into a bio suitable for io to a block
911 * device. Returns an error pointer in case of error.
913 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
918 bio = __bio_map_kern(q, data, len, gfp_mask);
922 if (bio->bi_size == len)
926 * Don't support partial mappings.
929 return ERR_PTR(-EINVAL);
932 static void bio_copy_kern_endio(struct bio *bio, int err)
934 struct bio_vec *bvec;
935 const int read = bio_data_dir(bio) == READ;
936 char *p = bio->bi_private;
939 __bio_for_each_segment(bvec, bio, i, 0) {
940 char *addr = page_address(bvec->bv_page);
943 memcpy(p, addr, bvec->bv_len);
945 __free_page(bvec->bv_page);
953 * bio_copy_kern - copy kernel address into bio
954 * @q: the struct request_queue for the bio
955 * @data: pointer to buffer to copy
956 * @len: length in bytes
957 * @gfp_mask: allocation flags for bio and page allocation
958 * @reading: data direction is READ
960 * copy the kernel address into a bio suitable for io to a block
961 * device. Returns an error pointer in case of error.
963 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
964 gfp_t gfp_mask, int reading)
966 unsigned long kaddr = (unsigned long)data;
967 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
968 unsigned long start = kaddr >> PAGE_SHIFT;
969 const int nr_pages = end - start;
971 struct bio_vec *bvec;
974 bio = bio_alloc(gfp_mask, nr_pages);
976 return ERR_PTR(-ENOMEM);
980 unsigned int bytes = PAGE_SIZE;
985 page = alloc_page(q->bounce_gfp | gfp_mask);
991 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) {
1002 bio_for_each_segment(bvec, bio, i) {
1003 char *addr = page_address(bvec->bv_page);
1005 memcpy(addr, p, bvec->bv_len);
1010 bio->bi_private = data;
1011 bio->bi_end_io = bio_copy_kern_endio;
1014 bio_for_each_segment(bvec, bio, i)
1015 __free_page(bvec->bv_page);
1019 return ERR_PTR(ret);
1023 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1024 * for performing direct-IO in BIOs.
1026 * The problem is that we cannot run set_page_dirty() from interrupt context
1027 * because the required locks are not interrupt-safe. So what we can do is to
1028 * mark the pages dirty _before_ performing IO. And in interrupt context,
1029 * check that the pages are still dirty. If so, fine. If not, redirty them
1030 * in process context.
1032 * We special-case compound pages here: normally this means reads into hugetlb
1033 * pages. The logic in here doesn't really work right for compound pages
1034 * because the VM does not uniformly chase down the head page in all cases.
1035 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1036 * handle them at all. So we skip compound pages here at an early stage.
1038 * Note that this code is very hard to test under normal circumstances because
1039 * direct-io pins the pages with get_user_pages(). This makes
1040 * is_page_cache_freeable return false, and the VM will not clean the pages.
1041 * But other code (eg, pdflush) could clean the pages if they are mapped
1044 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1045 * deferred bio dirtying paths.
1049 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1051 void bio_set_pages_dirty(struct bio *bio)
1053 struct bio_vec *bvec = bio->bi_io_vec;
1056 for (i = 0; i < bio->bi_vcnt; i++) {
1057 struct page *page = bvec[i].bv_page;
1059 if (page && !PageCompound(page))
1060 set_page_dirty_lock(page);
1064 static void bio_release_pages(struct bio *bio)
1066 struct bio_vec *bvec = bio->bi_io_vec;
1069 for (i = 0; i < bio->bi_vcnt; i++) {
1070 struct page *page = bvec[i].bv_page;
1078 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1079 * If they are, then fine. If, however, some pages are clean then they must
1080 * have been written out during the direct-IO read. So we take another ref on
1081 * the BIO and the offending pages and re-dirty the pages in process context.
1083 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1084 * here on. It will run one page_cache_release() against each page and will
1085 * run one bio_put() against the BIO.
1088 static void bio_dirty_fn(struct work_struct *work);
1090 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1091 static DEFINE_SPINLOCK(bio_dirty_lock);
1092 static struct bio *bio_dirty_list;
1095 * This runs in process context
1097 static void bio_dirty_fn(struct work_struct *work)
1099 unsigned long flags;
1102 spin_lock_irqsave(&bio_dirty_lock, flags);
1103 bio = bio_dirty_list;
1104 bio_dirty_list = NULL;
1105 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1108 struct bio *next = bio->bi_private;
1110 bio_set_pages_dirty(bio);
1111 bio_release_pages(bio);
1117 void bio_check_pages_dirty(struct bio *bio)
1119 struct bio_vec *bvec = bio->bi_io_vec;
1120 int nr_clean_pages = 0;
1123 for (i = 0; i < bio->bi_vcnt; i++) {
1124 struct page *page = bvec[i].bv_page;
1126 if (PageDirty(page) || PageCompound(page)) {
1127 page_cache_release(page);
1128 bvec[i].bv_page = NULL;
1134 if (nr_clean_pages) {
1135 unsigned long flags;
1137 spin_lock_irqsave(&bio_dirty_lock, flags);
1138 bio->bi_private = bio_dirty_list;
1139 bio_dirty_list = bio;
1140 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1141 schedule_work(&bio_dirty_work);
1148 * bio_endio - end I/O on a bio
1150 * @error: error, if any
1153 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1154 * preferred way to end I/O on a bio, it takes care of clearing
1155 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1156 * established -Exxxx (-EIO, for instance) error values in case
1157 * something went wrong. Noone should call bi_end_io() directly on a
1158 * bio unless they own it and thus know that it has an end_io
1161 void bio_endio(struct bio *bio, int error)
1164 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1165 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1169 bio->bi_end_io(bio, error);
1172 void bio_pair_release(struct bio_pair *bp)
1174 if (atomic_dec_and_test(&bp->cnt)) {
1175 struct bio *master = bp->bio1.bi_private;
1177 bio_endio(master, bp->error);
1178 mempool_free(bp, bp->bio2.bi_private);
1182 static void bio_pair_end_1(struct bio *bi, int err)
1184 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1189 bio_pair_release(bp);
1192 static void bio_pair_end_2(struct bio *bi, int err)
1194 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1199 bio_pair_release(bp);
1203 * split a bio - only worry about a bio with a single page
1206 struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors)
1208 struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO);
1213 blk_add_trace_pdu_int(bdev_get_queue(bi->bi_bdev), BLK_TA_SPLIT, bi,
1214 bi->bi_sector + first_sectors);
1216 BUG_ON(bi->bi_vcnt != 1);
1217 BUG_ON(bi->bi_idx != 0);
1218 atomic_set(&bp->cnt, 3);
1222 bp->bio2.bi_sector += first_sectors;
1223 bp->bio2.bi_size -= first_sectors << 9;
1224 bp->bio1.bi_size = first_sectors << 9;
1226 bp->bv1 = bi->bi_io_vec[0];
1227 bp->bv2 = bi->bi_io_vec[0];
1228 bp->bv2.bv_offset += first_sectors << 9;
1229 bp->bv2.bv_len -= first_sectors << 9;
1230 bp->bv1.bv_len = first_sectors << 9;
1232 bp->bio1.bi_io_vec = &bp->bv1;
1233 bp->bio2.bi_io_vec = &bp->bv2;
1235 bp->bio1.bi_max_vecs = 1;
1236 bp->bio2.bi_max_vecs = 1;
1238 bp->bio1.bi_end_io = bio_pair_end_1;
1239 bp->bio2.bi_end_io = bio_pair_end_2;
1241 bp->bio1.bi_private = bi;
1242 bp->bio2.bi_private = pool;
1244 if (bio_integrity(bi))
1245 bio_integrity_split(bi, bp, first_sectors);
1252 * create memory pools for biovec's in a bio_set.
1253 * use the global biovec slabs created for general use.
1255 static int biovec_create_pools(struct bio_set *bs, int pool_entries)
1259 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1260 struct biovec_slab *bp = bvec_slabs + i;
1261 mempool_t **bvp = bs->bvec_pools + i;
1263 *bvp = mempool_create_slab_pool(pool_entries, bp->slab);
1270 static void biovec_free_pools(struct bio_set *bs)
1274 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1275 mempool_t *bvp = bs->bvec_pools[i];
1278 mempool_destroy(bvp);
1283 void bioset_free(struct bio_set *bs)
1286 mempool_destroy(bs->bio_pool);
1288 bioset_integrity_free(bs);
1289 biovec_free_pools(bs);
1294 struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size)
1296 struct bio_set *bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1301 bs->bio_pool = mempool_create_slab_pool(bio_pool_size, bio_slab);
1305 if (bioset_integrity_create(bs, bio_pool_size))
1308 if (!biovec_create_pools(bs, bvec_pool_size))
1316 static void __init biovec_init_slabs(void)
1320 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1322 struct biovec_slab *bvs = bvec_slabs + i;
1324 size = bvs->nr_vecs * sizeof(struct bio_vec);
1325 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1326 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1330 static int __init init_bio(void)
1332 bio_slab = KMEM_CACHE(bio, SLAB_HWCACHE_ALIGN|SLAB_PANIC);
1334 bio_integrity_init_slab();
1335 biovec_init_slabs();
1337 fs_bio_set = bioset_create(BIO_POOL_SIZE, 2);
1339 panic("bio: can't allocate bios\n");
1341 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1342 sizeof(struct bio_pair));
1343 if (!bio_split_pool)
1344 panic("bio: can't create split pool\n");
1349 subsys_initcall(init_bio);
1351 EXPORT_SYMBOL(bio_alloc);
1352 EXPORT_SYMBOL(bio_put);
1353 EXPORT_SYMBOL(bio_free);
1354 EXPORT_SYMBOL(bio_endio);
1355 EXPORT_SYMBOL(bio_init);
1356 EXPORT_SYMBOL(__bio_clone);
1357 EXPORT_SYMBOL(bio_clone);
1358 EXPORT_SYMBOL(bio_phys_segments);
1359 EXPORT_SYMBOL(bio_hw_segments);
1360 EXPORT_SYMBOL(bio_add_page);
1361 EXPORT_SYMBOL(bio_add_pc_page);
1362 EXPORT_SYMBOL(bio_get_nr_vecs);
1363 EXPORT_SYMBOL(bio_map_user);
1364 EXPORT_SYMBOL(bio_unmap_user);
1365 EXPORT_SYMBOL(bio_map_kern);
1366 EXPORT_SYMBOL(bio_copy_kern);
1367 EXPORT_SYMBOL(bio_pair_release);
1368 EXPORT_SYMBOL(bio_split);
1369 EXPORT_SYMBOL(bio_split_pool);
1370 EXPORT_SYMBOL(bio_copy_user);
1371 EXPORT_SYMBOL(bio_uncopy_user);
1372 EXPORT_SYMBOL(bioset_create);
1373 EXPORT_SYMBOL(bioset_free);
1374 EXPORT_SYMBOL(bio_alloc_bioset);