mmu-notifiers: add mm_take_all_locks() operation
[linux-2.6/openmoko-kernel.git] / fs / bio.c
blob25f1af0d81e5bee47ee8487b27ca8e0152b6d156
1 /*
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-
18 #include <linux/mm.h>
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
38 * unsigned short
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),
45 #undef BV
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)
60 struct bio_vec *bvl;
63 * see comment near bvec_array define!
65 switch (nr) {
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;
72 default:
73 return NULL;
76 * idx now points to the pool we want to allocate from
79 bvl = mempool_alloc(bs->bvec_pools[*idx], gfp_mask);
80 if (bvl) {
81 struct biovec_slab *bp = bvec_slabs + *idx;
83 memset(bvl, 0, bp->nr_vecs * sizeof(struct bio_vec));
86 return bvl;
89 void bio_free(struct bio *bio, struct bio_set *bio_set)
91 if (bio->bi_io_vec) {
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
126 * Description:
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
132 * bio_set structure.
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);
138 if (likely(bio)) {
139 struct bio_vec *bvl = NULL;
141 bio_init(bio);
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);
148 bio = NULL;
149 goto out;
151 bio->bi_flags |= idx << BIO_POOL_OFFSET;
152 bio->bi_max_vecs = bvec_slabs[idx].nr_vecs;
154 bio->bi_io_vec = bvl;
156 out:
157 return bio;
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);
164 if (bio)
165 bio->bi_destructor = bio_fs_destructor;
167 return bio;
170 void zero_fill_bio(struct bio *bio)
172 unsigned long flags;
173 struct bio_vec *bv;
174 int i;
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
189 * Description:
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));
198 * last put frees it
200 if (atomic_dec_and_test(&bio->bi_cnt)) {
201 bio->bi_next = NULL;
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
229 * bio will be one.
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
251 * @bio: bio to clone
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);
260 if (!b)
261 return NULL;
263 b->bi_destructor = bio_fs_destructor;
264 __bio_clone(b, bio);
266 if (bio_integrity(bio)) {
267 int ret;
269 ret = bio_integrity_clone(b, bio, fs_bio_set);
271 if (ret < 0)
272 return NULL;
275 return b;
279 * bio_get_nr_vecs - return approx number of vecs
280 * @bdev: I/O target
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
285 * on offset.
287 int bio_get_nr_vecs(struct block_device *bdev)
289 struct request_queue *q = bdev_get_queue(bdev);
290 int nr_pages;
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;
298 return nr_pages;
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)))
312 return 0;
314 if (((bio->bi_size + len) >> 9) > max_sectors)
315 return 0;
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) {
327 prev->bv_len += len;
329 if (q->merge_bvec_fn) {
330 struct bvec_merge_data bvm = {
331 .bi_bdev = bio->bi_bdev,
332 .bi_sector = bio->bi_sector,
333 .bi_size = bio->bi_size,
334 .bi_rw = bio->bi_rw,
337 if (q->merge_bvec_fn(q, &bvm, prev) < len) {
338 prev->bv_len -= len;
339 return 0;
343 goto done;
347 if (bio->bi_vcnt >= bio->bi_max_vecs)
348 return 0;
351 * we might lose a segment or two here, but rather that than
352 * make this too complex.
355 while (bio->bi_phys_segments >= q->max_phys_segments
356 || bio->bi_hw_segments >= q->max_hw_segments
357 || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) {
359 if (retried_segments)
360 return 0;
362 retried_segments = 1;
363 blk_recount_segments(q, bio);
367 * setup the new entry, we might clear it again later if we
368 * cannot add the page
370 bvec = &bio->bi_io_vec[bio->bi_vcnt];
371 bvec->bv_page = page;
372 bvec->bv_len = len;
373 bvec->bv_offset = offset;
376 * if queue has other restrictions (eg varying max sector size
377 * depending on offset), it can specify a merge_bvec_fn in the
378 * queue to get further control
380 if (q->merge_bvec_fn) {
381 struct bvec_merge_data bvm = {
382 .bi_bdev = bio->bi_bdev,
383 .bi_sector = bio->bi_sector,
384 .bi_size = bio->bi_size,
385 .bi_rw = bio->bi_rw,
389 * merge_bvec_fn() returns number of bytes it can accept
390 * at this offset
392 if (q->merge_bvec_fn(q, &bvm, bvec) < len) {
393 bvec->bv_page = NULL;
394 bvec->bv_len = 0;
395 bvec->bv_offset = 0;
396 return 0;
400 /* If we may be able to merge these biovecs, force a recount */
401 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) ||
402 BIOVEC_VIRT_MERGEABLE(bvec-1, bvec)))
403 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
405 bio->bi_vcnt++;
406 bio->bi_phys_segments++;
407 bio->bi_hw_segments++;
408 done:
409 bio->bi_size += len;
410 return len;
414 * bio_add_pc_page - attempt to add page to bio
415 * @q: the target queue
416 * @bio: destination bio
417 * @page: page to add
418 * @len: vec entry length
419 * @offset: vec entry offset
421 * Attempt to add a page to the bio_vec maplist. This can fail for a
422 * number of reasons, such as the bio being full or target block
423 * device limitations. The target block device must allow bio's
424 * smaller than PAGE_SIZE, so it is always possible to add a single
425 * page to an empty bio. This should only be used by REQ_PC bios.
427 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
428 unsigned int len, unsigned int offset)
430 return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors);
434 * bio_add_page - attempt to add page to bio
435 * @bio: destination bio
436 * @page: page to add
437 * @len: vec entry length
438 * @offset: vec entry offset
440 * Attempt to add a page to the bio_vec maplist. This can fail for a
441 * number of reasons, such as the bio being full or target block
442 * device limitations. The target block device must allow bio's
443 * smaller than PAGE_SIZE, so it is always possible to add a single
444 * page to an empty bio.
446 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
447 unsigned int offset)
449 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
450 return __bio_add_page(q, bio, page, len, offset, q->max_sectors);
453 struct bio_map_data {
454 struct bio_vec *iovecs;
455 int nr_sgvecs;
456 struct sg_iovec *sgvecs;
459 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
460 struct sg_iovec *iov, int iov_count)
462 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
463 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
464 bmd->nr_sgvecs = iov_count;
465 bio->bi_private = bmd;
468 static void bio_free_map_data(struct bio_map_data *bmd)
470 kfree(bmd->iovecs);
471 kfree(bmd->sgvecs);
472 kfree(bmd);
475 static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count)
477 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), GFP_KERNEL);
479 if (!bmd)
480 return NULL;
482 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, GFP_KERNEL);
483 if (!bmd->iovecs) {
484 kfree(bmd);
485 return NULL;
488 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, GFP_KERNEL);
489 if (bmd->sgvecs)
490 return bmd;
492 kfree(bmd->iovecs);
493 kfree(bmd);
494 return NULL;
497 static int __bio_copy_iov(struct bio *bio, struct sg_iovec *iov, int iov_count,
498 int uncopy)
500 int ret = 0, i;
501 struct bio_vec *bvec;
502 int iov_idx = 0;
503 unsigned int iov_off = 0;
504 int read = bio_data_dir(bio) == READ;
506 __bio_for_each_segment(bvec, bio, i, 0) {
507 char *bv_addr = page_address(bvec->bv_page);
508 unsigned int bv_len = bvec->bv_len;
510 while (bv_len && iov_idx < iov_count) {
511 unsigned int bytes;
512 char *iov_addr;
514 bytes = min_t(unsigned int,
515 iov[iov_idx].iov_len - iov_off, bv_len);
516 iov_addr = iov[iov_idx].iov_base + iov_off;
518 if (!ret) {
519 if (!read && !uncopy)
520 ret = copy_from_user(bv_addr, iov_addr,
521 bytes);
522 if (read && uncopy)
523 ret = copy_to_user(iov_addr, bv_addr,
524 bytes);
526 if (ret)
527 ret = -EFAULT;
530 bv_len -= bytes;
531 bv_addr += bytes;
532 iov_addr += bytes;
533 iov_off += bytes;
535 if (iov[iov_idx].iov_len == iov_off) {
536 iov_idx++;
537 iov_off = 0;
541 if (uncopy)
542 __free_page(bvec->bv_page);
545 return ret;
549 * bio_uncopy_user - finish previously mapped bio
550 * @bio: bio being terminated
552 * Free pages allocated from bio_copy_user() and write back data
553 * to user space in case of a read.
555 int bio_uncopy_user(struct bio *bio)
557 struct bio_map_data *bmd = bio->bi_private;
558 int ret;
560 ret = __bio_copy_iov(bio, bmd->sgvecs, bmd->nr_sgvecs, 1);
562 bio_free_map_data(bmd);
563 bio_put(bio);
564 return ret;
568 * bio_copy_user_iov - copy user data to bio
569 * @q: destination block queue
570 * @iov: the iovec.
571 * @iov_count: number of elements in the iovec
572 * @write_to_vm: bool indicating writing to pages or not
574 * Prepares and returns a bio for indirect user io, bouncing data
575 * to/from kernel pages as necessary. Must be paired with
576 * call bio_uncopy_user() on io completion.
578 struct bio *bio_copy_user_iov(struct request_queue *q, struct sg_iovec *iov,
579 int iov_count, int write_to_vm)
581 struct bio_map_data *bmd;
582 struct bio_vec *bvec;
583 struct page *page;
584 struct bio *bio;
585 int i, ret;
586 int nr_pages = 0;
587 unsigned int len = 0;
589 for (i = 0; i < iov_count; i++) {
590 unsigned long uaddr;
591 unsigned long end;
592 unsigned long start;
594 uaddr = (unsigned long)iov[i].iov_base;
595 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
596 start = uaddr >> PAGE_SHIFT;
598 nr_pages += end - start;
599 len += iov[i].iov_len;
602 bmd = bio_alloc_map_data(nr_pages, iov_count);
603 if (!bmd)
604 return ERR_PTR(-ENOMEM);
606 ret = -ENOMEM;
607 bio = bio_alloc(GFP_KERNEL, nr_pages);
608 if (!bio)
609 goto out_bmd;
611 bio->bi_rw |= (!write_to_vm << BIO_RW);
613 ret = 0;
614 while (len) {
615 unsigned int bytes = PAGE_SIZE;
617 if (bytes > len)
618 bytes = len;
620 page = alloc_page(q->bounce_gfp | GFP_KERNEL);
621 if (!page) {
622 ret = -ENOMEM;
623 break;
626 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
627 break;
629 len -= bytes;
632 if (ret)
633 goto cleanup;
636 * success
638 if (!write_to_vm) {
639 ret = __bio_copy_iov(bio, iov, iov_count, 0);
640 if (ret)
641 goto cleanup;
644 bio_set_map_data(bmd, bio, iov, iov_count);
645 return bio;
646 cleanup:
647 bio_for_each_segment(bvec, bio, i)
648 __free_page(bvec->bv_page);
650 bio_put(bio);
651 out_bmd:
652 bio_free_map_data(bmd);
653 return ERR_PTR(ret);
657 * bio_copy_user - copy user data to bio
658 * @q: destination block queue
659 * @uaddr: start of user address
660 * @len: length in bytes
661 * @write_to_vm: bool indicating writing to pages or not
663 * Prepares and returns a bio for indirect user io, bouncing data
664 * to/from kernel pages as necessary. Must be paired with
665 * call bio_uncopy_user() on io completion.
667 struct bio *bio_copy_user(struct request_queue *q, unsigned long uaddr,
668 unsigned int len, int write_to_vm)
670 struct sg_iovec iov;
672 iov.iov_base = (void __user *)uaddr;
673 iov.iov_len = len;
675 return bio_copy_user_iov(q, &iov, 1, write_to_vm);
678 static struct bio *__bio_map_user_iov(struct request_queue *q,
679 struct block_device *bdev,
680 struct sg_iovec *iov, int iov_count,
681 int write_to_vm)
683 int i, j;
684 int nr_pages = 0;
685 struct page **pages;
686 struct bio *bio;
687 int cur_page = 0;
688 int ret, offset;
690 for (i = 0; i < iov_count; i++) {
691 unsigned long uaddr = (unsigned long)iov[i].iov_base;
692 unsigned long len = iov[i].iov_len;
693 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
694 unsigned long start = uaddr >> PAGE_SHIFT;
696 nr_pages += end - start;
698 * buffer must be aligned to at least hardsector size for now
700 if (uaddr & queue_dma_alignment(q))
701 return ERR_PTR(-EINVAL);
704 if (!nr_pages)
705 return ERR_PTR(-EINVAL);
707 bio = bio_alloc(GFP_KERNEL, nr_pages);
708 if (!bio)
709 return ERR_PTR(-ENOMEM);
711 ret = -ENOMEM;
712 pages = kcalloc(nr_pages, sizeof(struct page *), GFP_KERNEL);
713 if (!pages)
714 goto out;
716 for (i = 0; i < iov_count; i++) {
717 unsigned long uaddr = (unsigned long)iov[i].iov_base;
718 unsigned long len = iov[i].iov_len;
719 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
720 unsigned long start = uaddr >> PAGE_SHIFT;
721 const int local_nr_pages = end - start;
722 const int page_limit = cur_page + local_nr_pages;
724 ret = get_user_pages_fast(uaddr, local_nr_pages,
725 write_to_vm, &pages[cur_page]);
726 if (ret < local_nr_pages) {
727 ret = -EFAULT;
728 goto out_unmap;
731 offset = uaddr & ~PAGE_MASK;
732 for (j = cur_page; j < page_limit; j++) {
733 unsigned int bytes = PAGE_SIZE - offset;
735 if (len <= 0)
736 break;
738 if (bytes > len)
739 bytes = len;
742 * sorry...
744 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
745 bytes)
746 break;
748 len -= bytes;
749 offset = 0;
752 cur_page = j;
754 * release the pages we didn't map into the bio, if any
756 while (j < page_limit)
757 page_cache_release(pages[j++]);
760 kfree(pages);
763 * set data direction, and check if mapped pages need bouncing
765 if (!write_to_vm)
766 bio->bi_rw |= (1 << BIO_RW);
768 bio->bi_bdev = bdev;
769 bio->bi_flags |= (1 << BIO_USER_MAPPED);
770 return bio;
772 out_unmap:
773 for (i = 0; i < nr_pages; i++) {
774 if(!pages[i])
775 break;
776 page_cache_release(pages[i]);
778 out:
779 kfree(pages);
780 bio_put(bio);
781 return ERR_PTR(ret);
785 * bio_map_user - map user address into bio
786 * @q: the struct request_queue for the bio
787 * @bdev: destination block device
788 * @uaddr: start of user address
789 * @len: length in bytes
790 * @write_to_vm: bool indicating writing to pages or not
792 * Map the user space address into a bio suitable for io to a block
793 * device. Returns an error pointer in case of error.
795 struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
796 unsigned long uaddr, unsigned int len, int write_to_vm)
798 struct sg_iovec iov;
800 iov.iov_base = (void __user *)uaddr;
801 iov.iov_len = len;
803 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm);
807 * bio_map_user_iov - map user sg_iovec table into bio
808 * @q: the struct request_queue for the bio
809 * @bdev: destination block device
810 * @iov: the iovec.
811 * @iov_count: number of elements in the iovec
812 * @write_to_vm: bool indicating writing to pages or not
814 * Map the user space address into a bio suitable for io to a block
815 * device. Returns an error pointer in case of error.
817 struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
818 struct sg_iovec *iov, int iov_count,
819 int write_to_vm)
821 struct bio *bio;
823 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm);
825 if (IS_ERR(bio))
826 return bio;
829 * subtle -- if __bio_map_user() ended up bouncing a bio,
830 * it would normally disappear when its bi_end_io is run.
831 * however, we need it for the unmap, so grab an extra
832 * reference to it
834 bio_get(bio);
836 return bio;
839 static void __bio_unmap_user(struct bio *bio)
841 struct bio_vec *bvec;
842 int i;
845 * make sure we dirty pages we wrote to
847 __bio_for_each_segment(bvec, bio, i, 0) {
848 if (bio_data_dir(bio) == READ)
849 set_page_dirty_lock(bvec->bv_page);
851 page_cache_release(bvec->bv_page);
854 bio_put(bio);
858 * bio_unmap_user - unmap a bio
859 * @bio: the bio being unmapped
861 * Unmap a bio previously mapped by bio_map_user(). Must be called with
862 * a process context.
864 * bio_unmap_user() may sleep.
866 void bio_unmap_user(struct bio *bio)
868 __bio_unmap_user(bio);
869 bio_put(bio);
872 static void bio_map_kern_endio(struct bio *bio, int err)
874 bio_put(bio);
878 static struct bio *__bio_map_kern(struct request_queue *q, void *data,
879 unsigned int len, gfp_t gfp_mask)
881 unsigned long kaddr = (unsigned long)data;
882 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
883 unsigned long start = kaddr >> PAGE_SHIFT;
884 const int nr_pages = end - start;
885 int offset, i;
886 struct bio *bio;
888 bio = bio_alloc(gfp_mask, nr_pages);
889 if (!bio)
890 return ERR_PTR(-ENOMEM);
892 offset = offset_in_page(kaddr);
893 for (i = 0; i < nr_pages; i++) {
894 unsigned int bytes = PAGE_SIZE - offset;
896 if (len <= 0)
897 break;
899 if (bytes > len)
900 bytes = len;
902 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
903 offset) < bytes)
904 break;
906 data += bytes;
907 len -= bytes;
908 offset = 0;
911 bio->bi_end_io = bio_map_kern_endio;
912 return bio;
916 * bio_map_kern - map kernel address into bio
917 * @q: the struct request_queue for the bio
918 * @data: pointer to buffer to map
919 * @len: length in bytes
920 * @gfp_mask: allocation flags for bio allocation
922 * Map the kernel address into a bio suitable for io to a block
923 * device. Returns an error pointer in case of error.
925 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
926 gfp_t gfp_mask)
928 struct bio *bio;
930 bio = __bio_map_kern(q, data, len, gfp_mask);
931 if (IS_ERR(bio))
932 return bio;
934 if (bio->bi_size == len)
935 return bio;
938 * Don't support partial mappings.
940 bio_put(bio);
941 return ERR_PTR(-EINVAL);
944 static void bio_copy_kern_endio(struct bio *bio, int err)
946 struct bio_vec *bvec;
947 const int read = bio_data_dir(bio) == READ;
948 char *p = bio->bi_private;
949 int i;
951 __bio_for_each_segment(bvec, bio, i, 0) {
952 char *addr = page_address(bvec->bv_page);
954 if (read && !err)
955 memcpy(p, addr, bvec->bv_len);
957 __free_page(bvec->bv_page);
958 p += bvec->bv_len;
961 bio_put(bio);
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;
982 struct bio *bio;
983 struct bio_vec *bvec;
984 int i, ret;
986 bio = bio_alloc(gfp_mask, nr_pages);
987 if (!bio)
988 return ERR_PTR(-ENOMEM);
990 while (len) {
991 struct page *page;
992 unsigned int bytes = PAGE_SIZE;
994 if (bytes > len)
995 bytes = len;
997 page = alloc_page(q->bounce_gfp | gfp_mask);
998 if (!page) {
999 ret = -ENOMEM;
1000 goto cleanup;
1003 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes) {
1004 ret = -EINVAL;
1005 goto cleanup;
1008 len -= bytes;
1011 if (!reading) {
1012 void *p = data;
1014 bio_for_each_segment(bvec, bio, i) {
1015 char *addr = page_address(bvec->bv_page);
1017 memcpy(addr, p, bvec->bv_len);
1018 p += bvec->bv_len;
1022 bio->bi_private = data;
1023 bio->bi_end_io = bio_copy_kern_endio;
1024 return bio;
1025 cleanup:
1026 bio_for_each_segment(bvec, bio, i)
1027 __free_page(bvec->bv_page);
1029 bio_put(bio);
1031 return ERR_PTR(ret);
1035 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1036 * for performing direct-IO in BIOs.
1038 * The problem is that we cannot run set_page_dirty() from interrupt context
1039 * because the required locks are not interrupt-safe. So what we can do is to
1040 * mark the pages dirty _before_ performing IO. And in interrupt context,
1041 * check that the pages are still dirty. If so, fine. If not, redirty them
1042 * in process context.
1044 * We special-case compound pages here: normally this means reads into hugetlb
1045 * pages. The logic in here doesn't really work right for compound pages
1046 * because the VM does not uniformly chase down the head page in all cases.
1047 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1048 * handle them at all. So we skip compound pages here at an early stage.
1050 * Note that this code is very hard to test under normal circumstances because
1051 * direct-io pins the pages with get_user_pages(). This makes
1052 * is_page_cache_freeable return false, and the VM will not clean the pages.
1053 * But other code (eg, pdflush) could clean the pages if they are mapped
1054 * pagecache.
1056 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1057 * deferred bio dirtying paths.
1061 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1063 void bio_set_pages_dirty(struct bio *bio)
1065 struct bio_vec *bvec = bio->bi_io_vec;
1066 int i;
1068 for (i = 0; i < bio->bi_vcnt; i++) {
1069 struct page *page = bvec[i].bv_page;
1071 if (page && !PageCompound(page))
1072 set_page_dirty_lock(page);
1076 static void bio_release_pages(struct bio *bio)
1078 struct bio_vec *bvec = bio->bi_io_vec;
1079 int i;
1081 for (i = 0; i < bio->bi_vcnt; i++) {
1082 struct page *page = bvec[i].bv_page;
1084 if (page)
1085 put_page(page);
1090 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1091 * If they are, then fine. If, however, some pages are clean then they must
1092 * have been written out during the direct-IO read. So we take another ref on
1093 * the BIO and the offending pages and re-dirty the pages in process context.
1095 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1096 * here on. It will run one page_cache_release() against each page and will
1097 * run one bio_put() against the BIO.
1100 static void bio_dirty_fn(struct work_struct *work);
1102 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1103 static DEFINE_SPINLOCK(bio_dirty_lock);
1104 static struct bio *bio_dirty_list;
1107 * This runs in process context
1109 static void bio_dirty_fn(struct work_struct *work)
1111 unsigned long flags;
1112 struct bio *bio;
1114 spin_lock_irqsave(&bio_dirty_lock, flags);
1115 bio = bio_dirty_list;
1116 bio_dirty_list = NULL;
1117 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1119 while (bio) {
1120 struct bio *next = bio->bi_private;
1122 bio_set_pages_dirty(bio);
1123 bio_release_pages(bio);
1124 bio_put(bio);
1125 bio = next;
1129 void bio_check_pages_dirty(struct bio *bio)
1131 struct bio_vec *bvec = bio->bi_io_vec;
1132 int nr_clean_pages = 0;
1133 int i;
1135 for (i = 0; i < bio->bi_vcnt; i++) {
1136 struct page *page = bvec[i].bv_page;
1138 if (PageDirty(page) || PageCompound(page)) {
1139 page_cache_release(page);
1140 bvec[i].bv_page = NULL;
1141 } else {
1142 nr_clean_pages++;
1146 if (nr_clean_pages) {
1147 unsigned long flags;
1149 spin_lock_irqsave(&bio_dirty_lock, flags);
1150 bio->bi_private = bio_dirty_list;
1151 bio_dirty_list = bio;
1152 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1153 schedule_work(&bio_dirty_work);
1154 } else {
1155 bio_put(bio);
1160 * bio_endio - end I/O on a bio
1161 * @bio: bio
1162 * @error: error, if any
1164 * Description:
1165 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1166 * preferred way to end I/O on a bio, it takes care of clearing
1167 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1168 * established -Exxxx (-EIO, for instance) error values in case
1169 * something went wrong. Noone should call bi_end_io() directly on a
1170 * bio unless they own it and thus know that it has an end_io
1171 * function.
1173 void bio_endio(struct bio *bio, int error)
1175 if (error)
1176 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1177 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1178 error = -EIO;
1180 if (bio->bi_end_io)
1181 bio->bi_end_io(bio, error);
1184 void bio_pair_release(struct bio_pair *bp)
1186 if (atomic_dec_and_test(&bp->cnt)) {
1187 struct bio *master = bp->bio1.bi_private;
1189 bio_endio(master, bp->error);
1190 mempool_free(bp, bp->bio2.bi_private);
1194 static void bio_pair_end_1(struct bio *bi, int err)
1196 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1198 if (err)
1199 bp->error = err;
1201 bio_pair_release(bp);
1204 static void bio_pair_end_2(struct bio *bi, int err)
1206 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1208 if (err)
1209 bp->error = err;
1211 bio_pair_release(bp);
1215 * split a bio - only worry about a bio with a single page
1216 * in it's iovec
1218 struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors)
1220 struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO);
1222 if (!bp)
1223 return bp;
1225 blk_add_trace_pdu_int(bdev_get_queue(bi->bi_bdev), BLK_TA_SPLIT, bi,
1226 bi->bi_sector + first_sectors);
1228 BUG_ON(bi->bi_vcnt != 1);
1229 BUG_ON(bi->bi_idx != 0);
1230 atomic_set(&bp->cnt, 3);
1231 bp->error = 0;
1232 bp->bio1 = *bi;
1233 bp->bio2 = *bi;
1234 bp->bio2.bi_sector += first_sectors;
1235 bp->bio2.bi_size -= first_sectors << 9;
1236 bp->bio1.bi_size = first_sectors << 9;
1238 bp->bv1 = bi->bi_io_vec[0];
1239 bp->bv2 = bi->bi_io_vec[0];
1240 bp->bv2.bv_offset += first_sectors << 9;
1241 bp->bv2.bv_len -= first_sectors << 9;
1242 bp->bv1.bv_len = first_sectors << 9;
1244 bp->bio1.bi_io_vec = &bp->bv1;
1245 bp->bio2.bi_io_vec = &bp->bv2;
1247 bp->bio1.bi_max_vecs = 1;
1248 bp->bio2.bi_max_vecs = 1;
1250 bp->bio1.bi_end_io = bio_pair_end_1;
1251 bp->bio2.bi_end_io = bio_pair_end_2;
1253 bp->bio1.bi_private = bi;
1254 bp->bio2.bi_private = pool;
1256 if (bio_integrity(bi))
1257 bio_integrity_split(bi, bp, first_sectors);
1259 return bp;
1264 * create memory pools for biovec's in a bio_set.
1265 * use the global biovec slabs created for general use.
1267 static int biovec_create_pools(struct bio_set *bs, int pool_entries)
1269 int i;
1271 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1272 struct biovec_slab *bp = bvec_slabs + i;
1273 mempool_t **bvp = bs->bvec_pools + i;
1275 *bvp = mempool_create_slab_pool(pool_entries, bp->slab);
1276 if (!*bvp)
1277 return -ENOMEM;
1279 return 0;
1282 static void biovec_free_pools(struct bio_set *bs)
1284 int i;
1286 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1287 mempool_t *bvp = bs->bvec_pools[i];
1289 if (bvp)
1290 mempool_destroy(bvp);
1295 void bioset_free(struct bio_set *bs)
1297 if (bs->bio_pool)
1298 mempool_destroy(bs->bio_pool);
1300 bioset_integrity_free(bs);
1301 biovec_free_pools(bs);
1303 kfree(bs);
1306 struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size)
1308 struct bio_set *bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1310 if (!bs)
1311 return NULL;
1313 bs->bio_pool = mempool_create_slab_pool(bio_pool_size, bio_slab);
1314 if (!bs->bio_pool)
1315 goto bad;
1317 if (bioset_integrity_create(bs, bio_pool_size))
1318 goto bad;
1320 if (!biovec_create_pools(bs, bvec_pool_size))
1321 return bs;
1323 bad:
1324 bioset_free(bs);
1325 return NULL;
1328 static void __init biovec_init_slabs(void)
1330 int i;
1332 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1333 int size;
1334 struct biovec_slab *bvs = bvec_slabs + i;
1336 size = bvs->nr_vecs * sizeof(struct bio_vec);
1337 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1338 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1342 static int __init init_bio(void)
1344 bio_slab = KMEM_CACHE(bio, SLAB_HWCACHE_ALIGN|SLAB_PANIC);
1346 bio_integrity_init_slab();
1347 biovec_init_slabs();
1349 fs_bio_set = bioset_create(BIO_POOL_SIZE, 2);
1350 if (!fs_bio_set)
1351 panic("bio: can't allocate bios\n");
1353 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1354 sizeof(struct bio_pair));
1355 if (!bio_split_pool)
1356 panic("bio: can't create split pool\n");
1358 return 0;
1361 subsys_initcall(init_bio);
1363 EXPORT_SYMBOL(bio_alloc);
1364 EXPORT_SYMBOL(bio_put);
1365 EXPORT_SYMBOL(bio_free);
1366 EXPORT_SYMBOL(bio_endio);
1367 EXPORT_SYMBOL(bio_init);
1368 EXPORT_SYMBOL(__bio_clone);
1369 EXPORT_SYMBOL(bio_clone);
1370 EXPORT_SYMBOL(bio_phys_segments);
1371 EXPORT_SYMBOL(bio_hw_segments);
1372 EXPORT_SYMBOL(bio_add_page);
1373 EXPORT_SYMBOL(bio_add_pc_page);
1374 EXPORT_SYMBOL(bio_get_nr_vecs);
1375 EXPORT_SYMBOL(bio_map_user);
1376 EXPORT_SYMBOL(bio_unmap_user);
1377 EXPORT_SYMBOL(bio_map_kern);
1378 EXPORT_SYMBOL(bio_copy_kern);
1379 EXPORT_SYMBOL(bio_pair_release);
1380 EXPORT_SYMBOL(bio_split);
1381 EXPORT_SYMBOL(bio_split_pool);
1382 EXPORT_SYMBOL(bio_copy_user);
1383 EXPORT_SYMBOL(bio_uncopy_user);
1384 EXPORT_SYMBOL(bioset_create);
1385 EXPORT_SYMBOL(bioset_free);
1386 EXPORT_SYMBOL(bio_alloc_bioset);