2 * Copyright (C) 2001 Jens Axboe <axboe@suse.de>
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
13 * You should have received a copy of the GNU General Public Licens
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-
19 #include <linux/swap.h>
20 #include <linux/bio.h>
21 #include <linux/blkdev.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/kernel.h>
25 #include <linux/module.h>
26 #include <linux/mempool.h>
27 #include <linux/workqueue.h>
29 #define BIO_POOL_SIZE 256
31 static mempool_t *bio_pool;
32 static kmem_cache_t *bio_slab;
34 #define BIOVEC_NR_POOLS 6
37 * a small number of entries is fine, not going to be performance critical.
38 * basically we just need to survive
40 #define BIO_SPLIT_ENTRIES 8
41 mempool_t *bio_split_pool;
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_pool bvec_array[BIOVEC_NR_POOLS] = {
58 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
62 static inline struct bio_vec *bvec_alloc(int gfp_mask, int nr, unsigned long *idx)
64 struct biovec_pool *bp;
68 * see comment near bvec_array define!
71 case 1 : *idx = 0; break;
72 case 2 ... 4: *idx = 1; break;
73 case 5 ... 16: *idx = 2; break;
74 case 17 ... 64: *idx = 3; break;
75 case 65 ... 128: *idx = 4; break;
76 case 129 ... BIO_MAX_PAGES: *idx = 5; break;
81 * idx now points to the pool we want to allocate from
83 bp = bvec_array + *idx;
85 bvl = mempool_alloc(bp->pool, gfp_mask);
87 memset(bvl, 0, bp->nr_vecs * sizeof(struct bio_vec));
92 * default destructor for a bio allocated with bio_alloc()
94 void bio_destructor(struct bio *bio)
96 const int pool_idx = BIO_POOL_IDX(bio);
97 struct biovec_pool *bp = bvec_array + pool_idx;
99 BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS);
102 * cloned bio doesn't own the veclist
104 if (!bio_flagged(bio, BIO_CLONED))
105 mempool_free(bio->bi_io_vec, bp->pool);
107 mempool_free(bio, bio_pool);
110 inline void bio_init(struct bio *bio)
113 bio->bi_flags = 1 << BIO_UPTODATE;
117 bio->bi_phys_segments = 0;
118 bio->bi_hw_segments = 0;
119 bio->bi_hw_front_size = 0;
120 bio->bi_hw_back_size = 0;
122 bio->bi_max_vecs = 0;
123 bio->bi_end_io = NULL;
124 atomic_set(&bio->bi_cnt, 1);
125 bio->bi_private = NULL;
129 * bio_alloc - allocate a bio for I/O
130 * @gfp_mask: the GFP_ mask given to the slab allocator
131 * @nr_iovecs: number of iovecs to pre-allocate
134 * bio_alloc will first try it's on mempool to satisfy the allocation.
135 * If %__GFP_WAIT is set then we will block on the internal pool waiting
136 * for a &struct bio to become free.
138 struct bio *bio_alloc(int gfp_mask, int nr_iovecs)
140 struct bio_vec *bvl = NULL;
144 bio = mempool_alloc(bio_pool, gfp_mask);
150 if (unlikely(!nr_iovecs))
153 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx);
155 bio->bi_flags |= idx << BIO_POOL_OFFSET;
156 bio->bi_max_vecs = bvec_array[idx].nr_vecs;
158 bio->bi_io_vec = bvl;
159 bio->bi_destructor = bio_destructor;
164 mempool_free(bio, bio_pool);
170 * bio_put - release a reference to a bio
171 * @bio: bio to release reference to
174 * Put a reference to a &struct bio, either one you have gotten with
175 * bio_alloc or bio_get. The last put of a bio will free it.
177 void bio_put(struct bio *bio)
179 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
184 if (atomic_dec_and_test(&bio->bi_cnt)) {
186 bio->bi_destructor(bio);
190 inline int bio_phys_segments(request_queue_t *q, struct bio *bio)
192 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
193 blk_recount_segments(q, bio);
195 return bio->bi_phys_segments;
198 inline int bio_hw_segments(request_queue_t *q, struct bio *bio)
200 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
201 blk_recount_segments(q, bio);
203 return bio->bi_hw_segments;
207 * __bio_clone - clone a bio
208 * @bio: destination bio
209 * @bio_src: bio to clone
211 * Clone a &bio. Caller will own the returned bio, but not
212 * the actual data it points to. Reference count of returned
215 inline void __bio_clone(struct bio *bio, struct bio *bio_src)
217 bio->bi_io_vec = bio_src->bi_io_vec;
219 bio->bi_sector = bio_src->bi_sector;
220 bio->bi_bdev = bio_src->bi_bdev;
221 bio->bi_flags |= 1 << BIO_CLONED;
222 bio->bi_rw = bio_src->bi_rw;
225 * notes -- maybe just leave bi_idx alone. assume identical mapping
228 bio->bi_vcnt = bio_src->bi_vcnt;
229 bio->bi_idx = bio_src->bi_idx;
230 if (bio_flagged(bio, BIO_SEG_VALID)) {
231 bio->bi_phys_segments = bio_src->bi_phys_segments;
232 bio->bi_hw_segments = bio_src->bi_hw_segments;
233 bio->bi_flags |= (1 << BIO_SEG_VALID);
235 bio->bi_size = bio_src->bi_size;
238 * cloned bio does not own the bio_vec, so users cannot fiddle with
239 * it. clear bi_max_vecs and clear the BIO_POOL_BITS to make this
242 bio->bi_max_vecs = 0;
243 bio->bi_flags &= (BIO_POOL_MASK - 1);
247 * bio_clone - clone a bio
249 * @gfp_mask: allocation priority
251 * Like __bio_clone, only also allocates the returned bio
253 struct bio *bio_clone(struct bio *bio, int gfp_mask)
255 struct bio *b = bio_alloc(gfp_mask, 0);
264 * bio_get_nr_vecs - return approx number of vecs
267 * Return the approximate number of pages we can send to this target.
268 * There's no guarantee that you will be able to fit this number of pages
269 * into a bio, it does not account for dynamic restrictions that vary
272 int bio_get_nr_vecs(struct block_device *bdev)
274 request_queue_t *q = bdev_get_queue(bdev);
277 nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
278 if (nr_pages > q->max_phys_segments)
279 nr_pages = q->max_phys_segments;
280 if (nr_pages > q->max_hw_segments)
281 nr_pages = q->max_hw_segments;
286 static int __bio_add_page(request_queue_t *q, struct bio *bio, struct page
287 *page, unsigned int len, unsigned int offset)
289 int retried_segments = 0;
290 struct bio_vec *bvec;
293 * cloned bio must not modify vec list
295 if (unlikely(bio_flagged(bio, BIO_CLONED)))
298 if (bio->bi_vcnt >= bio->bi_max_vecs)
301 if (((bio->bi_size + len) >> 9) > q->max_sectors)
305 * we might lose a segment or two here, but rather that than
306 * make this too complex.
309 while (bio->bi_phys_segments >= q->max_phys_segments
310 || bio->bi_hw_segments >= q->max_hw_segments
311 || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) {
313 if (retried_segments)
316 retried_segments = 1;
317 blk_recount_segments(q, bio);
321 * setup the new entry, we might clear it again later if we
322 * cannot add the page
324 bvec = &bio->bi_io_vec[bio->bi_vcnt];
325 bvec->bv_page = page;
327 bvec->bv_offset = offset;
330 * if queue has other restrictions (eg varying max sector size
331 * depending on offset), it can specify a merge_bvec_fn in the
332 * queue to get further control
334 if (q->merge_bvec_fn) {
336 * merge_bvec_fn() returns number of bytes it can accept
339 if (q->merge_bvec_fn(q, bio, bvec) < len) {
340 bvec->bv_page = NULL;
347 /* If we may be able to merge these biovecs, force a recount */
348 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) ||
349 BIOVEC_VIRT_MERGEABLE(bvec-1, bvec)))
350 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
353 bio->bi_phys_segments++;
354 bio->bi_hw_segments++;
360 * bio_add_page - attempt to add page to bio
361 * @bio: destination bio
363 * @len: vec entry length
364 * @offset: vec entry offset
366 * Attempt to add a page to the bio_vec maplist. This can fail for a
367 * number of reasons, such as the bio being full or target block
368 * device limitations. The target block device must allow bio's
369 * smaller than PAGE_SIZE, so it is always possible to add a single
370 * page to an empty bio.
372 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
375 return __bio_add_page(bdev_get_queue(bio->bi_bdev), bio, page,
379 static struct bio *__bio_map_user(request_queue_t *q, struct block_device *bdev,
380 unsigned long uaddr, unsigned int len,
383 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
384 unsigned long start = uaddr >> PAGE_SHIFT;
385 const int nr_pages = end - start;
391 * transfer and buffer must be aligned to at least hardsector
392 * size for now, in the future we can relax this restriction
394 if ((uaddr & queue_dma_alignment(q)) || (len & queue_dma_alignment(q)))
397 bio = bio_alloc(GFP_KERNEL, nr_pages);
401 pages = kmalloc(nr_pages * sizeof(struct page *), GFP_KERNEL);
405 down_read(¤t->mm->mmap_sem);
406 ret = get_user_pages(current, current->mm, uaddr, nr_pages,
407 write_to_vm, 0, pages, NULL);
408 up_read(¤t->mm->mmap_sem);
415 offset = uaddr & ~PAGE_MASK;
416 for (i = 0; i < nr_pages; i++) {
417 unsigned int bytes = PAGE_SIZE - offset;
428 if (__bio_add_page(q, bio, pages[i], bytes, offset) < bytes)
436 * release the pages we didn't map into the bio, if any
439 page_cache_release(pages[i++]);
444 * set data direction, and check if mapped pages need bouncing
447 bio->bi_rw |= (1 << BIO_RW);
449 blk_queue_bounce(q, &bio);
458 * bio_map_user - map user address into bio
459 * @bdev: destination block device
460 * @uaddr: start of user address
461 * @len: length in bytes
462 * @write_to_vm: bool indicating writing to pages or not
464 * Map the user space address into a bio suitable for io to a block
467 struct bio *bio_map_user(request_queue_t *q, struct block_device *bdev,
468 unsigned long uaddr, unsigned int len, int write_to_vm)
472 bio = __bio_map_user(q, bdev, uaddr, len, write_to_vm);
476 * subtle -- if __bio_map_user() ended up bouncing a bio,
477 * it would normally disappear when its bi_end_io is run.
478 * however, we need it for the unmap, so grab an extra
483 if (bio->bi_size < len) {
484 bio_endio(bio, bio->bi_size, 0);
485 bio_unmap_user(bio, 0);
493 static void __bio_unmap_user(struct bio *bio, int write_to_vm)
495 struct bio_vec *bvec;
499 * find original bio if it was bounced
501 if (bio->bi_private) {
503 * someone stole our bio, must not happen
505 BUG_ON(!bio_flagged(bio, BIO_BOUNCED));
507 bio = bio->bi_private;
511 * make sure we dirty pages we wrote to
513 __bio_for_each_segment(bvec, bio, i, 0) {
515 set_page_dirty_lock(bvec->bv_page);
517 page_cache_release(bvec->bv_page);
524 * bio_unmap_user - unmap a bio
525 * @bio: the bio being unmapped
526 * @write_to_vm: bool indicating whether pages were written to
528 * Unmap a bio previously mapped by bio_map_user(). The @write_to_vm
529 * must be the same as passed into bio_map_user(). Must be called with
532 * bio_unmap_user() may sleep.
534 void bio_unmap_user(struct bio *bio, int write_to_vm)
536 __bio_unmap_user(bio, write_to_vm);
541 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
542 * for performing direct-IO in BIOs.
544 * The problem is that we cannot run set_page_dirty() from interrupt context
545 * because the required locks are not interrupt-safe. So what we can do is to
546 * mark the pages dirty _before_ performing IO. And in interrupt context,
547 * check that the pages are still dirty. If so, fine. If not, redirty them
548 * in process context.
550 * We special-case compound pages here: normally this means reads into hugetlb
551 * pages. The logic in here doesn't really work right for compound pages
552 * because the VM does not uniformly chase down the head page in all cases.
553 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
554 * handle them at all. So we skip compound pages here at an early stage.
556 * Note that this code is very hard to test under normal circumstances because
557 * direct-io pins the pages with get_user_pages(). This makes
558 * is_page_cache_freeable return false, and the VM will not clean the pages.
559 * But other code (eg, pdflush) could clean the pages if they are mapped
562 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
563 * deferred bio dirtying paths.
567 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
569 void bio_set_pages_dirty(struct bio *bio)
571 struct bio_vec *bvec = bio->bi_io_vec;
574 for (i = 0; i < bio->bi_vcnt; i++) {
575 struct page *page = bvec[i].bv_page;
577 if (page && !PageCompound(page))
578 set_page_dirty_lock(page);
582 static void bio_release_pages(struct bio *bio)
584 struct bio_vec *bvec = bio->bi_io_vec;
587 for (i = 0; i < bio->bi_vcnt; i++) {
588 struct page *page = bvec[i].bv_page;
596 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
597 * If they are, then fine. If, however, some pages are clean then they must
598 * have been written out during the direct-IO read. So we take another ref on
599 * the BIO and the offending pages and re-dirty the pages in process context.
601 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
602 * here on. It will run one page_cache_release() against each page and will
603 * run one bio_put() against the BIO.
606 static void bio_dirty_fn(void *data);
608 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn, NULL);
609 static spinlock_t bio_dirty_lock = SPIN_LOCK_UNLOCKED;
610 static struct bio *bio_dirty_list;
613 * This runs in process context
615 static void bio_dirty_fn(void *data)
620 spin_lock_irqsave(&bio_dirty_lock, flags);
621 bio = bio_dirty_list;
622 bio_dirty_list = NULL;
623 spin_unlock_irqrestore(&bio_dirty_lock, flags);
626 struct bio *next = bio->bi_private;
628 bio_set_pages_dirty(bio);
629 bio_release_pages(bio);
635 void bio_check_pages_dirty(struct bio *bio)
637 struct bio_vec *bvec = bio->bi_io_vec;
638 int nr_clean_pages = 0;
641 for (i = 0; i < bio->bi_vcnt; i++) {
642 struct page *page = bvec[i].bv_page;
644 if (PageDirty(page) || PageCompound(page)) {
645 page_cache_release(page);
646 bvec[i].bv_page = NULL;
652 if (nr_clean_pages) {
655 spin_lock_irqsave(&bio_dirty_lock, flags);
656 bio->bi_private = bio_dirty_list;
657 bio_dirty_list = bio;
658 spin_unlock_irqrestore(&bio_dirty_lock, flags);
659 schedule_work(&bio_dirty_work);
666 * bio_endio - end I/O on a bio
668 * @bytes_done: number of bytes completed
669 * @error: error, if any
672 * bio_endio() will end I/O on @bytes_done number of bytes. This may be
673 * just a partial part of the bio, or it may be the whole bio. bio_endio()
674 * is the preferred way to end I/O on a bio, it takes care of decrementing
675 * bi_size and clearing BIO_UPTODATE on error. @error is 0 on success, and
676 * and one of the established -Exxxx (-EIO, for instance) error values in
677 * case something went wrong. Noone should call bi_end_io() directly on
678 * a bio unless they own it and thus know that it has an end_io function.
680 void bio_endio(struct bio *bio, unsigned int bytes_done, int error)
683 clear_bit(BIO_UPTODATE, &bio->bi_flags);
685 if (unlikely(bytes_done > bio->bi_size)) {
686 printk("%s: want %u bytes done, only %u left\n", __FUNCTION__,
687 bytes_done, bio->bi_size);
688 bytes_done = bio->bi_size;
691 bio->bi_size -= bytes_done;
692 bio->bi_sector += (bytes_done >> 9);
695 bio->bi_end_io(bio, bytes_done, error);
698 void bio_pair_release(struct bio_pair *bp)
700 if (atomic_dec_and_test(&bp->cnt)) {
701 struct bio *master = bp->bio1.bi_private;
703 bio_endio(master, master->bi_size, bp->error);
704 mempool_free(bp, bp->bio2.bi_private);
708 static int bio_pair_end_1(struct bio * bi, unsigned int done, int err)
710 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
718 bio_pair_release(bp);
722 static int bio_pair_end_2(struct bio * bi, unsigned int done, int err)
724 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
732 bio_pair_release(bp);
737 * split a bio - only worry about a bio with a single page
740 struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors)
742 struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO);
747 BUG_ON(bi->bi_vcnt != 1);
748 BUG_ON(bi->bi_idx != 0);
749 atomic_set(&bp->cnt, 3);
753 bp->bio2.bi_sector += first_sectors;
754 bp->bio2.bi_size -= first_sectors << 9;
755 bp->bio1.bi_size = first_sectors << 9;
757 bp->bv1 = bi->bi_io_vec[0];
758 bp->bv2 = bi->bi_io_vec[0];
759 bp->bv2.bv_offset += first_sectors << 9;
760 bp->bv2.bv_len -= first_sectors << 9;
761 bp->bv1.bv_len = first_sectors << 9;
763 bp->bio1.bi_io_vec = &bp->bv1;
764 bp->bio2.bi_io_vec = &bp->bv2;
766 bp->bio1.bi_end_io = bio_pair_end_1;
767 bp->bio2.bi_end_io = bio_pair_end_2;
769 bp->bio1.bi_private = bi;
770 bp->bio2.bi_private = pool;
775 static void *bio_pair_alloc(int gfp_flags, void *data)
777 return kmalloc(sizeof(struct bio_pair), gfp_flags);
780 static void bio_pair_free(void *bp, void *data)
785 static void __init biovec_init_pools(void)
787 int i, size, megabytes, pool_entries = BIO_POOL_SIZE;
788 int scale = BIOVEC_NR_POOLS;
790 megabytes = nr_free_pages() >> (20 - PAGE_SHIFT);
793 * find out where to start scaling
797 else if (megabytes <= 32)
799 else if (megabytes <= 64)
801 else if (megabytes <= 96)
803 else if (megabytes <= 128)
807 * scale number of entries
809 pool_entries = megabytes * 2;
810 if (pool_entries > 256)
813 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
814 struct biovec_pool *bp = bvec_array + i;
816 size = bp->nr_vecs * sizeof(struct bio_vec);
818 bp->slab = kmem_cache_create(bp->name, size, 0,
819 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
824 bp->pool = mempool_create(pool_entries, mempool_alloc_slab,
825 mempool_free_slab, bp->slab);
827 panic("biovec: can't init mempool\n");
831 static int __init init_bio(void)
833 bio_slab = kmem_cache_create("bio", sizeof(struct bio), 0,
834 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
835 bio_pool = mempool_create(BIO_POOL_SIZE, mempool_alloc_slab,
836 mempool_free_slab, bio_slab);
838 panic("bio: can't create mempool\n");
842 bio_split_pool = mempool_create(BIO_SPLIT_ENTRIES,
843 bio_pair_alloc, bio_pair_free, NULL);
845 panic("bio: can't create split pool\n");
850 subsys_initcall(init_bio);
852 EXPORT_SYMBOL(bio_alloc);
853 EXPORT_SYMBOL(bio_put);
854 EXPORT_SYMBOL(bio_endio);
855 EXPORT_SYMBOL(bio_init);
856 EXPORT_SYMBOL(__bio_clone);
857 EXPORT_SYMBOL(bio_clone);
858 EXPORT_SYMBOL(bio_phys_segments);
859 EXPORT_SYMBOL(bio_hw_segments);
860 EXPORT_SYMBOL(bio_add_page);
861 EXPORT_SYMBOL(bio_get_nr_vecs);
862 EXPORT_SYMBOL(bio_map_user);
863 EXPORT_SYMBOL(bio_unmap_user);
864 EXPORT_SYMBOL(bio_pair_release);
865 EXPORT_SYMBOL(bio_split);
866 EXPORT_SYMBOL(bio_split_pool);