aboutsummaryrefslogtreecommitdiffstats
path: root/fs/bio.c
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 18:20:36 -0400
committerLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 18:20:36 -0400
commit1da177e4c3f41524e886b7f1b8a0c1fc7321cac2 (patch)
tree0bba044c4ce775e45a88a51686b5d9f90697ea9d /fs/bio.c
Linux-2.6.12-rc2v2.6.12-rc2
Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
Diffstat (limited to 'fs/bio.c')
-rw-r--r--fs/bio.c1096
1 files changed, 1096 insertions, 0 deletions
diff --git a/fs/bio.c b/fs/bio.c
new file mode 100644
index 000000000000..e5349e834563
--- /dev/null
+++ b/fs/bio.c
@@ -0,0 +1,1096 @@
1/*
2 * Copyright (C) 2001 Jens Axboe <axboe@suse.de>
3 *
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.
7 *
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.
12 *
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-
16 *
17 */
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
29#define BIO_POOL_SIZE 256
30
31static kmem_cache_t *bio_slab;
32
33#define BIOVEC_NR_POOLS 6
34
35/*
36 * a small number of entries is fine, not going to be performance critical.
37 * basically we just need to survive
38 */
39#define BIO_SPLIT_ENTRIES 8
40mempool_t *bio_split_pool;
41
42struct biovec_slab {
43 int nr_vecs;
44 char *name;
45 kmem_cache_t *slab;
46};
47
48/*
49 * if you change this list, also change bvec_alloc or things will
50 * break badly! cannot be bigger than what you can fit into an
51 * unsigned short
52 */
53
54#define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
55static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] = {
56 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
57};
58#undef BV
59
60/*
61 * bio_set is used to allow other portions of the IO system to
62 * allocate their own private memory pools for bio and iovec structures.
63 * These memory pools in turn all allocate from the bio_slab
64 * and the bvec_slabs[].
65 */
66struct bio_set {
67 mempool_t *bio_pool;
68 mempool_t *bvec_pools[BIOVEC_NR_POOLS];
69};
70
71/*
72 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
73 * IO code that does not need private memory pools.
74 */
75static struct bio_set *fs_bio_set;
76
77static inline struct bio_vec *bvec_alloc_bs(unsigned int __nocast gfp_mask, int nr, unsigned long *idx, struct bio_set *bs)
78{
79 struct bio_vec *bvl;
80 struct biovec_slab *bp;
81
82 /*
83 * see comment near bvec_array define!
84 */
85 switch (nr) {
86 case 1 : *idx = 0; break;
87 case 2 ... 4: *idx = 1; break;
88 case 5 ... 16: *idx = 2; break;
89 case 17 ... 64: *idx = 3; break;
90 case 65 ... 128: *idx = 4; break;
91 case 129 ... BIO_MAX_PAGES: *idx = 5; break;
92 default:
93 return NULL;
94 }
95 /*
96 * idx now points to the pool we want to allocate from
97 */
98
99 bp = bvec_slabs + *idx;
100 bvl = mempool_alloc(bs->bvec_pools[*idx], gfp_mask);
101 if (bvl)
102 memset(bvl, 0, bp->nr_vecs * sizeof(struct bio_vec));
103
104 return bvl;
105}
106
107/*
108 * default destructor for a bio allocated with bio_alloc_bioset()
109 */
110static void bio_destructor(struct bio *bio)
111{
112 const int pool_idx = BIO_POOL_IDX(bio);
113 struct bio_set *bs = bio->bi_set;
114
115 BIO_BUG_ON(pool_idx >= BIOVEC_NR_POOLS);
116
117 mempool_free(bio->bi_io_vec, bs->bvec_pools[pool_idx]);
118 mempool_free(bio, bs->bio_pool);
119}
120
121inline void bio_init(struct bio *bio)
122{
123 bio->bi_next = NULL;
124 bio->bi_flags = 1 << BIO_UPTODATE;
125 bio->bi_rw = 0;
126 bio->bi_vcnt = 0;
127 bio->bi_idx = 0;
128 bio->bi_phys_segments = 0;
129 bio->bi_hw_segments = 0;
130 bio->bi_hw_front_size = 0;
131 bio->bi_hw_back_size = 0;
132 bio->bi_size = 0;
133 bio->bi_max_vecs = 0;
134 bio->bi_end_io = NULL;
135 atomic_set(&bio->bi_cnt, 1);
136 bio->bi_private = NULL;
137}
138
139/**
140 * bio_alloc_bioset - allocate a bio for I/O
141 * @gfp_mask: the GFP_ mask given to the slab allocator
142 * @nr_iovecs: number of iovecs to pre-allocate
143 *
144 * Description:
145 * bio_alloc_bioset will first try it's on mempool to satisfy the allocation.
146 * If %__GFP_WAIT is set then we will block on the internal pool waiting
147 * for a &struct bio to become free.
148 *
149 * allocate bio and iovecs from the memory pools specified by the
150 * bio_set structure.
151 **/
152struct bio *bio_alloc_bioset(unsigned int __nocast gfp_mask, int nr_iovecs, struct bio_set *bs)
153{
154 struct bio *bio = mempool_alloc(bs->bio_pool, gfp_mask);
155
156 if (likely(bio)) {
157 struct bio_vec *bvl = NULL;
158
159 bio_init(bio);
160 if (likely(nr_iovecs)) {
161 unsigned long idx;
162
163 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
164 if (unlikely(!bvl)) {
165 mempool_free(bio, bs->bio_pool);
166 bio = NULL;
167 goto out;
168 }
169 bio->bi_flags |= idx << BIO_POOL_OFFSET;
170 bio->bi_max_vecs = bvec_slabs[idx].nr_vecs;
171 }
172 bio->bi_io_vec = bvl;
173 bio->bi_destructor = bio_destructor;
174 bio->bi_set = bs;
175 }
176out:
177 return bio;
178}
179
180struct bio *bio_alloc(unsigned int __nocast gfp_mask, int nr_iovecs)
181{
182 return bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
183}
184
185void zero_fill_bio(struct bio *bio)
186{
187 unsigned long flags;
188 struct bio_vec *bv;
189 int i;
190
191 bio_for_each_segment(bv, bio, i) {
192 char *data = bvec_kmap_irq(bv, &flags);
193 memset(data, 0, bv->bv_len);
194 flush_dcache_page(bv->bv_page);
195 bvec_kunmap_irq(data, &flags);
196 }
197}
198EXPORT_SYMBOL(zero_fill_bio);
199
200/**
201 * bio_put - release a reference to a bio
202 * @bio: bio to release reference to
203 *
204 * Description:
205 * Put a reference to a &struct bio, either one you have gotten with
206 * bio_alloc or bio_get. The last put of a bio will free it.
207 **/
208void bio_put(struct bio *bio)
209{
210 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
211
212 /*
213 * last put frees it
214 */
215 if (atomic_dec_and_test(&bio->bi_cnt)) {
216 bio->bi_next = NULL;
217 bio->bi_destructor(bio);
218 }
219}
220
221inline int bio_phys_segments(request_queue_t *q, struct bio *bio)
222{
223 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
224 blk_recount_segments(q, bio);
225
226 return bio->bi_phys_segments;
227}
228
229inline int bio_hw_segments(request_queue_t *q, struct bio *bio)
230{
231 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
232 blk_recount_segments(q, bio);
233
234 return bio->bi_hw_segments;
235}
236
237/**
238 * __bio_clone - clone a bio
239 * @bio: destination bio
240 * @bio_src: bio to clone
241 *
242 * Clone a &bio. Caller will own the returned bio, but not
243 * the actual data it points to. Reference count of returned
244 * bio will be one.
245 */
246inline void __bio_clone(struct bio *bio, struct bio *bio_src)
247{
248 request_queue_t *q = bdev_get_queue(bio_src->bi_bdev);
249
250 memcpy(bio->bi_io_vec, bio_src->bi_io_vec, bio_src->bi_max_vecs * sizeof(struct bio_vec));
251
252 bio->bi_sector = bio_src->bi_sector;
253 bio->bi_bdev = bio_src->bi_bdev;
254 bio->bi_flags |= 1 << BIO_CLONED;
255 bio->bi_rw = bio_src->bi_rw;
256
257 /*
258 * notes -- maybe just leave bi_idx alone. assume identical mapping
259 * for the clone
260 */
261 bio->bi_vcnt = bio_src->bi_vcnt;
262 bio->bi_size = bio_src->bi_size;
263 bio_phys_segments(q, bio);
264 bio_hw_segments(q, bio);
265}
266
267/**
268 * bio_clone - clone a bio
269 * @bio: bio to clone
270 * @gfp_mask: allocation priority
271 *
272 * Like __bio_clone, only also allocates the returned bio
273 */
274struct bio *bio_clone(struct bio *bio, unsigned int __nocast gfp_mask)
275{
276 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
277
278 if (b)
279 __bio_clone(b, bio);
280
281 return b;
282}
283
284/**
285 * bio_get_nr_vecs - return approx number of vecs
286 * @bdev: I/O target
287 *
288 * Return the approximate number of pages we can send to this target.
289 * There's no guarantee that you will be able to fit this number of pages
290 * into a bio, it does not account for dynamic restrictions that vary
291 * on offset.
292 */
293int bio_get_nr_vecs(struct block_device *bdev)
294{
295 request_queue_t *q = bdev_get_queue(bdev);
296 int nr_pages;
297
298 nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
299 if (nr_pages > q->max_phys_segments)
300 nr_pages = q->max_phys_segments;
301 if (nr_pages > q->max_hw_segments)
302 nr_pages = q->max_hw_segments;
303
304 return nr_pages;
305}
306
307static int __bio_add_page(request_queue_t *q, struct bio *bio, struct page
308 *page, unsigned int len, unsigned int offset)
309{
310 int retried_segments = 0;
311 struct bio_vec *bvec;
312
313 /*
314 * cloned bio must not modify vec list
315 */
316 if (unlikely(bio_flagged(bio, BIO_CLONED)))
317 return 0;
318
319 if (bio->bi_vcnt >= bio->bi_max_vecs)
320 return 0;
321
322 if (((bio->bi_size + len) >> 9) > q->max_sectors)
323 return 0;
324
325 /*
326 * we might lose a segment or two here, but rather that than
327 * make this too complex.
328 */
329
330 while (bio->bi_phys_segments >= q->max_phys_segments
331 || bio->bi_hw_segments >= q->max_hw_segments
332 || BIOVEC_VIRT_OVERSIZE(bio->bi_size)) {
333
334 if (retried_segments)
335 return 0;
336
337 retried_segments = 1;
338 blk_recount_segments(q, bio);
339 }
340
341 /*
342 * setup the new entry, we might clear it again later if we
343 * cannot add the page
344 */
345 bvec = &bio->bi_io_vec[bio->bi_vcnt];
346 bvec->bv_page = page;
347 bvec->bv_len = len;
348 bvec->bv_offset = offset;
349
350 /*
351 * if queue has other restrictions (eg varying max sector size
352 * depending on offset), it can specify a merge_bvec_fn in the
353 * queue to get further control
354 */
355 if (q->merge_bvec_fn) {
356 /*
357 * merge_bvec_fn() returns number of bytes it can accept
358 * at this offset
359 */
360 if (q->merge_bvec_fn(q, bio, bvec) < len) {
361 bvec->bv_page = NULL;
362 bvec->bv_len = 0;
363 bvec->bv_offset = 0;
364 return 0;
365 }
366 }
367
368 /* If we may be able to merge these biovecs, force a recount */
369 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec) ||
370 BIOVEC_VIRT_MERGEABLE(bvec-1, bvec)))
371 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
372
373 bio->bi_vcnt++;
374 bio->bi_phys_segments++;
375 bio->bi_hw_segments++;
376 bio->bi_size += len;
377 return len;
378}
379
380/**
381 * bio_add_page - attempt to add page to bio
382 * @bio: destination bio
383 * @page: page to add
384 * @len: vec entry length
385 * @offset: vec entry offset
386 *
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.
392 */
393int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
394 unsigned int offset)
395{
396 return __bio_add_page(bdev_get_queue(bio->bi_bdev), bio, page,
397 len, offset);
398}
399
400struct bio_map_data {
401 struct bio_vec *iovecs;
402 void __user *userptr;
403};
404
405static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio)
406{
407 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
408 bio->bi_private = bmd;
409}
410
411static void bio_free_map_data(struct bio_map_data *bmd)
412{
413 kfree(bmd->iovecs);
414 kfree(bmd);
415}
416
417static struct bio_map_data *bio_alloc_map_data(int nr_segs)
418{
419 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), GFP_KERNEL);
420
421 if (!bmd)
422 return NULL;
423
424 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, GFP_KERNEL);
425 if (bmd->iovecs)
426 return bmd;
427
428 kfree(bmd);
429 return NULL;
430}
431
432/**
433 * bio_uncopy_user - finish previously mapped bio
434 * @bio: bio being terminated
435 *
436 * Free pages allocated from bio_copy_user() and write back data
437 * to user space in case of a read.
438 */
439int bio_uncopy_user(struct bio *bio)
440{
441 struct bio_map_data *bmd = bio->bi_private;
442 const int read = bio_data_dir(bio) == READ;
443 struct bio_vec *bvec;
444 int i, ret = 0;
445
446 __bio_for_each_segment(bvec, bio, i, 0) {
447 char *addr = page_address(bvec->bv_page);
448 unsigned int len = bmd->iovecs[i].bv_len;
449
450 if (read && !ret && copy_to_user(bmd->userptr, addr, len))
451 ret = -EFAULT;
452
453 __free_page(bvec->bv_page);
454 bmd->userptr += len;
455 }
456 bio_free_map_data(bmd);
457 bio_put(bio);
458 return ret;
459}
460
461/**
462 * bio_copy_user - copy user data to bio
463 * @q: destination block queue
464 * @uaddr: start of user address
465 * @len: length in bytes
466 * @write_to_vm: bool indicating writing to pages or not
467 *
468 * Prepares and returns a bio for indirect user io, bouncing data
469 * to/from kernel pages as necessary. Must be paired with
470 * call bio_uncopy_user() on io completion.
471 */
472struct bio *bio_copy_user(request_queue_t *q, unsigned long uaddr,
473 unsigned int len, int write_to_vm)
474{
475 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
476 unsigned long start = uaddr >> PAGE_SHIFT;
477 struct bio_map_data *bmd;
478 struct bio_vec *bvec;
479 struct page *page;
480 struct bio *bio;
481 int i, ret;
482
483 bmd = bio_alloc_map_data(end - start);
484 if (!bmd)
485 return ERR_PTR(-ENOMEM);
486
487 bmd->userptr = (void __user *) uaddr;
488
489 ret = -ENOMEM;
490 bio = bio_alloc(GFP_KERNEL, end - start);
491 if (!bio)
492 goto out_bmd;
493
494 bio->bi_rw |= (!write_to_vm << BIO_RW);
495
496 ret = 0;
497 while (len) {
498 unsigned int bytes = PAGE_SIZE;
499
500 if (bytes > len)
501 bytes = len;
502
503 page = alloc_page(q->bounce_gfp | GFP_KERNEL);
504 if (!page) {
505 ret = -ENOMEM;
506 break;
507 }
508
509 if (__bio_add_page(q, bio, page, bytes, 0) < bytes) {
510 ret = -EINVAL;
511 break;
512 }
513
514 len -= bytes;
515 }
516
517 if (ret)
518 goto cleanup;
519
520 /*
521 * success
522 */
523 if (!write_to_vm) {
524 char __user *p = (char __user *) uaddr;
525
526 /*
527 * for a write, copy in data to kernel pages
528 */
529 ret = -EFAULT;
530 bio_for_each_segment(bvec, bio, i) {
531 char *addr = page_address(bvec->bv_page);
532
533 if (copy_from_user(addr, p, bvec->bv_len))
534 goto cleanup;
535 p += bvec->bv_len;
536 }
537 }
538
539 bio_set_map_data(bmd, bio);
540 return bio;
541cleanup:
542 bio_for_each_segment(bvec, bio, i)
543 __free_page(bvec->bv_page);
544
545 bio_put(bio);
546out_bmd:
547 bio_free_map_data(bmd);
548 return ERR_PTR(ret);
549}
550
551static struct bio *__bio_map_user(request_queue_t *q, struct block_device *bdev,
552 unsigned long uaddr, unsigned int len,
553 int write_to_vm)
554{
555 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
556 unsigned long start = uaddr >> PAGE_SHIFT;
557 const int nr_pages = end - start;
558 int ret, offset, i;
559 struct page **pages;
560 struct bio *bio;
561
562 /*
563 * transfer and buffer must be aligned to at least hardsector
564 * size for now, in the future we can relax this restriction
565 */
566 if ((uaddr & queue_dma_alignment(q)) || (len & queue_dma_alignment(q)))
567 return ERR_PTR(-EINVAL);
568
569 bio = bio_alloc(GFP_KERNEL, nr_pages);
570 if (!bio)
571 return ERR_PTR(-ENOMEM);
572
573 ret = -ENOMEM;
574 pages = kmalloc(nr_pages * sizeof(struct page *), GFP_KERNEL);
575 if (!pages)
576 goto out;
577
578 down_read(&current->mm->mmap_sem);
579 ret = get_user_pages(current, current->mm, uaddr, nr_pages,
580 write_to_vm, 0, pages, NULL);
581 up_read(&current->mm->mmap_sem);
582
583 if (ret < nr_pages)
584 goto out;
585
586 bio->bi_bdev = bdev;
587
588 offset = uaddr & ~PAGE_MASK;
589 for (i = 0; i < nr_pages; i++) {
590 unsigned int bytes = PAGE_SIZE - offset;
591
592 if (len <= 0)
593 break;
594
595 if (bytes > len)
596 bytes = len;
597
598 /*
599 * sorry...
600 */
601 if (__bio_add_page(q, bio, pages[i], bytes, offset) < bytes)
602 break;
603
604 len -= bytes;
605 offset = 0;
606 }
607
608 /*
609 * release the pages we didn't map into the bio, if any
610 */
611 while (i < nr_pages)
612 page_cache_release(pages[i++]);
613
614 kfree(pages);
615
616 /*
617 * set data direction, and check if mapped pages need bouncing
618 */
619 if (!write_to_vm)
620 bio->bi_rw |= (1 << BIO_RW);
621
622 bio->bi_flags |= (1 << BIO_USER_MAPPED);
623 return bio;
624out:
625 kfree(pages);
626 bio_put(bio);
627 return ERR_PTR(ret);
628}
629
630/**
631 * bio_map_user - map user address into bio
632 * @bdev: destination block device
633 * @uaddr: start of user address
634 * @len: length in bytes
635 * @write_to_vm: bool indicating writing to pages or not
636 *
637 * Map the user space address into a bio suitable for io to a block
638 * device. Returns an error pointer in case of error.
639 */
640struct bio *bio_map_user(request_queue_t *q, struct block_device *bdev,
641 unsigned long uaddr, unsigned int len, int write_to_vm)
642{
643 struct bio *bio;
644
645 bio = __bio_map_user(q, bdev, uaddr, len, write_to_vm);
646
647 if (IS_ERR(bio))
648 return bio;
649
650 /*
651 * subtle -- if __bio_map_user() ended up bouncing a bio,
652 * it would normally disappear when its bi_end_io is run.
653 * however, we need it for the unmap, so grab an extra
654 * reference to it
655 */
656 bio_get(bio);
657
658 if (bio->bi_size == len)
659 return bio;
660
661 /*
662 * don't support partial mappings
663 */
664 bio_endio(bio, bio->bi_size, 0);
665 bio_unmap_user(bio);
666 return ERR_PTR(-EINVAL);
667}
668
669static void __bio_unmap_user(struct bio *bio)
670{
671 struct bio_vec *bvec;
672 int i;
673
674 /*
675 * make sure we dirty pages we wrote to
676 */
677 __bio_for_each_segment(bvec, bio, i, 0) {
678 if (bio_data_dir(bio) == READ)
679 set_page_dirty_lock(bvec->bv_page);
680
681 page_cache_release(bvec->bv_page);
682 }
683
684 bio_put(bio);
685}
686
687/**
688 * bio_unmap_user - unmap a bio
689 * @bio: the bio being unmapped
690 *
691 * Unmap a bio previously mapped by bio_map_user(). Must be called with
692 * a process context.
693 *
694 * bio_unmap_user() may sleep.
695 */
696void bio_unmap_user(struct bio *bio)
697{
698 __bio_unmap_user(bio);
699 bio_put(bio);
700}
701
702/*
703 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
704 * for performing direct-IO in BIOs.
705 *
706 * The problem is that we cannot run set_page_dirty() from interrupt context
707 * because the required locks are not interrupt-safe. So what we can do is to
708 * mark the pages dirty _before_ performing IO. And in interrupt context,
709 * check that the pages are still dirty. If so, fine. If not, redirty them
710 * in process context.
711 *
712 * We special-case compound pages here: normally this means reads into hugetlb
713 * pages. The logic in here doesn't really work right for compound pages
714 * because the VM does not uniformly chase down the head page in all cases.
715 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
716 * handle them at all. So we skip compound pages here at an early stage.
717 *
718 * Note that this code is very hard to test under normal circumstances because
719 * direct-io pins the pages with get_user_pages(). This makes
720 * is_page_cache_freeable return false, and the VM will not clean the pages.
721 * But other code (eg, pdflush) could clean the pages if they are mapped
722 * pagecache.
723 *
724 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
725 * deferred bio dirtying paths.
726 */
727
728/*
729 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
730 */
731void bio_set_pages_dirty(struct bio *bio)
732{
733 struct bio_vec *bvec = bio->bi_io_vec;
734 int i;
735
736 for (i = 0; i < bio->bi_vcnt; i++) {
737 struct page *page = bvec[i].bv_page;
738
739 if (page && !PageCompound(page))
740 set_page_dirty_lock(page);
741 }
742}
743
744static void bio_release_pages(struct bio *bio)
745{
746 struct bio_vec *bvec = bio->bi_io_vec;
747 int i;
748
749 for (i = 0; i < bio->bi_vcnt; i++) {
750 struct page *page = bvec[i].bv_page;
751
752 if (page)
753 put_page(page);
754 }
755}
756
757/*
758 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
759 * If they are, then fine. If, however, some pages are clean then they must
760 * have been written out during the direct-IO read. So we take another ref on
761 * the BIO and the offending pages and re-dirty the pages in process context.
762 *
763 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
764 * here on. It will run one page_cache_release() against each page and will
765 * run one bio_put() against the BIO.
766 */
767
768static void bio_dirty_fn(void *data);
769
770static DECLARE_WORK(bio_dirty_work, bio_dirty_fn, NULL);
771static DEFINE_SPINLOCK(bio_dirty_lock);
772static struct bio *bio_dirty_list;
773
774/*
775 * This runs in process context
776 */
777static void bio_dirty_fn(void *data)
778{
779 unsigned long flags;
780 struct bio *bio;
781
782 spin_lock_irqsave(&bio_dirty_lock, flags);
783 bio = bio_dirty_list;
784 bio_dirty_list = NULL;
785 spin_unlock_irqrestore(&bio_dirty_lock, flags);
786
787 while (bio) {
788 struct bio *next = bio->bi_private;
789
790 bio_set_pages_dirty(bio);
791 bio_release_pages(bio);
792 bio_put(bio);
793 bio = next;
794 }
795}
796
797void bio_check_pages_dirty(struct bio *bio)
798{
799 struct bio_vec *bvec = bio->bi_io_vec;
800 int nr_clean_pages = 0;
801 int i;
802
803 for (i = 0; i < bio->bi_vcnt; i++) {
804 struct page *page = bvec[i].bv_page;
805
806 if (PageDirty(page) || PageCompound(page)) {
807 page_cache_release(page);
808 bvec[i].bv_page = NULL;
809 } else {
810 nr_clean_pages++;
811 }
812 }
813
814 if (nr_clean_pages) {
815 unsigned long flags;
816
817 spin_lock_irqsave(&bio_dirty_lock, flags);
818 bio->bi_private = bio_dirty_list;
819 bio_dirty_list = bio;
820 spin_unlock_irqrestore(&bio_dirty_lock, flags);
821 schedule_work(&bio_dirty_work);
822 } else {
823 bio_put(bio);
824 }
825}
826
827/**
828 * bio_endio - end I/O on a bio
829 * @bio: bio
830 * @bytes_done: number of bytes completed
831 * @error: error, if any
832 *
833 * Description:
834 * bio_endio() will end I/O on @bytes_done number of bytes. This may be
835 * just a partial part of the bio, or it may be the whole bio. bio_endio()
836 * is the preferred way to end I/O on a bio, it takes care of decrementing
837 * bi_size and clearing BIO_UPTODATE on error. @error is 0 on success, and
838 * and one of the established -Exxxx (-EIO, for instance) error values in
839 * case something went wrong. Noone should call bi_end_io() directly on
840 * a bio unless they own it and thus know that it has an end_io function.
841 **/
842void bio_endio(struct bio *bio, unsigned int bytes_done, int error)
843{
844 if (error)
845 clear_bit(BIO_UPTODATE, &bio->bi_flags);
846
847 if (unlikely(bytes_done > bio->bi_size)) {
848 printk("%s: want %u bytes done, only %u left\n", __FUNCTION__,
849 bytes_done, bio->bi_size);
850 bytes_done = bio->bi_size;
851 }
852
853 bio->bi_size -= bytes_done;
854 bio->bi_sector += (bytes_done >> 9);
855
856 if (bio->bi_end_io)
857 bio->bi_end_io(bio, bytes_done, error);
858}
859
860void bio_pair_release(struct bio_pair *bp)
861{
862 if (atomic_dec_and_test(&bp->cnt)) {
863 struct bio *master = bp->bio1.bi_private;
864
865 bio_endio(master, master->bi_size, bp->error);
866 mempool_free(bp, bp->bio2.bi_private);
867 }
868}
869
870static int bio_pair_end_1(struct bio * bi, unsigned int done, int err)
871{
872 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
873
874 if (err)
875 bp->error = err;
876
877 if (bi->bi_size)
878 return 1;
879
880 bio_pair_release(bp);
881 return 0;
882}
883
884static int bio_pair_end_2(struct bio * bi, unsigned int done, int err)
885{
886 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
887
888 if (err)
889 bp->error = err;
890
891 if (bi->bi_size)
892 return 1;
893
894 bio_pair_release(bp);
895 return 0;
896}
897
898/*
899 * split a bio - only worry about a bio with a single page
900 * in it's iovec
901 */
902struct bio_pair *bio_split(struct bio *bi, mempool_t *pool, int first_sectors)
903{
904 struct bio_pair *bp = mempool_alloc(pool, GFP_NOIO);
905
906 if (!bp)
907 return bp;
908
909 BUG_ON(bi->bi_vcnt != 1);
910 BUG_ON(bi->bi_idx != 0);
911 atomic_set(&bp->cnt, 3);
912 bp->error = 0;
913 bp->bio1 = *bi;
914 bp->bio2 = *bi;
915 bp->bio2.bi_sector += first_sectors;
916 bp->bio2.bi_size -= first_sectors << 9;
917 bp->bio1.bi_size = first_sectors << 9;
918
919 bp->bv1 = bi->bi_io_vec[0];
920 bp->bv2 = bi->bi_io_vec[0];
921 bp->bv2.bv_offset += first_sectors << 9;
922 bp->bv2.bv_len -= first_sectors << 9;
923 bp->bv1.bv_len = first_sectors << 9;
924
925 bp->bio1.bi_io_vec = &bp->bv1;
926 bp->bio2.bi_io_vec = &bp->bv2;
927
928 bp->bio1.bi_end_io = bio_pair_end_1;
929 bp->bio2.bi_end_io = bio_pair_end_2;
930
931 bp->bio1.bi_private = bi;
932 bp->bio2.bi_private = pool;
933
934 return bp;
935}
936
937static void *bio_pair_alloc(unsigned int __nocast gfp_flags, void *data)
938{
939 return kmalloc(sizeof(struct bio_pair), gfp_flags);
940}
941
942static void bio_pair_free(void *bp, void *data)
943{
944 kfree(bp);
945}
946
947
948/*
949 * create memory pools for biovec's in a bio_set.
950 * use the global biovec slabs created for general use.
951 */
952static int biovec_create_pools(struct bio_set *bs, int pool_entries, int scale)
953{
954 int i;
955
956 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
957 struct biovec_slab *bp = bvec_slabs + i;
958 mempool_t **bvp = bs->bvec_pools + i;
959
960 if (i >= scale)
961 pool_entries >>= 1;
962
963 *bvp = mempool_create(pool_entries, mempool_alloc_slab,
964 mempool_free_slab, bp->slab);
965 if (!*bvp)
966 return -ENOMEM;
967 }
968 return 0;
969}
970
971static void biovec_free_pools(struct bio_set *bs)
972{
973 int i;
974
975 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
976 mempool_t *bvp = bs->bvec_pools[i];
977
978 if (bvp)
979 mempool_destroy(bvp);
980 }
981
982}
983
984void bioset_free(struct bio_set *bs)
985{
986 if (bs->bio_pool)
987 mempool_destroy(bs->bio_pool);
988
989 biovec_free_pools(bs);
990
991 kfree(bs);
992}
993
994struct bio_set *bioset_create(int bio_pool_size, int bvec_pool_size, int scale)
995{
996 struct bio_set *bs = kmalloc(sizeof(*bs), GFP_KERNEL);
997
998 if (!bs)
999 return NULL;
1000
1001 memset(bs, 0, sizeof(*bs));
1002 bs->bio_pool = mempool_create(bio_pool_size, mempool_alloc_slab,
1003 mempool_free_slab, bio_slab);
1004
1005 if (!bs->bio_pool)
1006 goto bad;
1007
1008 if (!biovec_create_pools(bs, bvec_pool_size, scale))
1009 return bs;
1010
1011bad:
1012 bioset_free(bs);
1013 return NULL;
1014}
1015
1016static void __init biovec_init_slabs(void)
1017{
1018 int i;
1019
1020 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1021 int size;
1022 struct biovec_slab *bvs = bvec_slabs + i;
1023
1024 size = bvs->nr_vecs * sizeof(struct bio_vec);
1025 bvs->slab = kmem_cache_create(bvs->name, size, 0,
1026 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1027 }
1028}
1029
1030static int __init init_bio(void)
1031{
1032 int megabytes, bvec_pool_entries;
1033 int scale = BIOVEC_NR_POOLS;
1034
1035 bio_slab = kmem_cache_create("bio", sizeof(struct bio), 0,
1036 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL);
1037
1038 biovec_init_slabs();
1039
1040 megabytes = nr_free_pages() >> (20 - PAGE_SHIFT);
1041
1042 /*
1043 * find out where to start scaling
1044 */
1045 if (megabytes <= 16)
1046 scale = 0;
1047 else if (megabytes <= 32)
1048 scale = 1;
1049 else if (megabytes <= 64)
1050 scale = 2;
1051 else if (megabytes <= 96)
1052 scale = 3;
1053 else if (megabytes <= 128)
1054 scale = 4;
1055
1056 /*
1057 * scale number of entries
1058 */
1059 bvec_pool_entries = megabytes * 2;
1060 if (bvec_pool_entries > 256)
1061 bvec_pool_entries = 256;
1062
1063 fs_bio_set = bioset_create(BIO_POOL_SIZE, bvec_pool_entries, scale);
1064 if (!fs_bio_set)
1065 panic("bio: can't allocate bios\n");
1066
1067 bio_split_pool = mempool_create(BIO_SPLIT_ENTRIES,
1068 bio_pair_alloc, bio_pair_free, NULL);
1069 if (!bio_split_pool)
1070 panic("bio: can't create split pool\n");
1071
1072 return 0;
1073}
1074
1075subsys_initcall(init_bio);
1076
1077EXPORT_SYMBOL(bio_alloc);
1078EXPORT_SYMBOL(bio_put);
1079EXPORT_SYMBOL(bio_endio);
1080EXPORT_SYMBOL(bio_init);
1081EXPORT_SYMBOL(__bio_clone);
1082EXPORT_SYMBOL(bio_clone);
1083EXPORT_SYMBOL(bio_phys_segments);
1084EXPORT_SYMBOL(bio_hw_segments);
1085EXPORT_SYMBOL(bio_add_page);
1086EXPORT_SYMBOL(bio_get_nr_vecs);
1087EXPORT_SYMBOL(bio_map_user);
1088EXPORT_SYMBOL(bio_unmap_user);
1089EXPORT_SYMBOL(bio_pair_release);
1090EXPORT_SYMBOL(bio_split);
1091EXPORT_SYMBOL(bio_split_pool);
1092EXPORT_SYMBOL(bio_copy_user);
1093EXPORT_SYMBOL(bio_uncopy_user);
1094EXPORT_SYMBOL(bioset_create);
1095EXPORT_SYMBOL(bioset_free);
1096EXPORT_SYMBOL(bio_alloc_bioset);