]> bbs.cooldavid.org Git - net-next-2.6.git/blame - fs/bio.c
block: fix queue bounce limit setting
[net-next-2.6.git] / fs / bio.c
CommitLineData
1da177e4 1/*
0fe23479 2 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
1da177e4
LT
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>
2056a782 28#include <linux/blktrace_api.h>
5f3ea37c 29#include <trace/block.h>
f1970baf 30#include <scsi/sg.h> /* for struct sg_iovec */
1da177e4 31
0bfc2455
IM
32DEFINE_TRACE(block_split);
33
392ddc32
JA
34/*
35 * Test patch to inline a certain number of bi_io_vec's inside the bio
36 * itself, to shrink a bio data allocation from two mempool calls to one
37 */
38#define BIO_INLINE_VECS 4
39
6feef531 40static mempool_t *bio_split_pool __read_mostly;
1da177e4 41
1da177e4
LT
42/*
43 * if you change this list, also change bvec_alloc or things will
44 * break badly! cannot be bigger than what you can fit into an
45 * unsigned short
46 */
1da177e4 47#define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
bb799ca0 48struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
1da177e4
LT
49 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
50};
51#undef BV
52
1da177e4
LT
53/*
54 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
55 * IO code that does not need private memory pools.
56 */
51d654e1 57struct bio_set *fs_bio_set;
1da177e4 58
bb799ca0
JA
59/*
60 * Our slab pool management
61 */
62struct bio_slab {
63 struct kmem_cache *slab;
64 unsigned int slab_ref;
65 unsigned int slab_size;
66 char name[8];
67};
68static DEFINE_MUTEX(bio_slab_lock);
69static struct bio_slab *bio_slabs;
70static unsigned int bio_slab_nr, bio_slab_max;
71
72static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
73{
74 unsigned int sz = sizeof(struct bio) + extra_size;
75 struct kmem_cache *slab = NULL;
76 struct bio_slab *bslab;
77 unsigned int i, entry = -1;
78
79 mutex_lock(&bio_slab_lock);
80
81 i = 0;
82 while (i < bio_slab_nr) {
83 struct bio_slab *bslab = &bio_slabs[i];
84
85 if (!bslab->slab && entry == -1)
86 entry = i;
87 else if (bslab->slab_size == sz) {
88 slab = bslab->slab;
89 bslab->slab_ref++;
90 break;
91 }
92 i++;
93 }
94
95 if (slab)
96 goto out_unlock;
97
98 if (bio_slab_nr == bio_slab_max && entry == -1) {
99 bio_slab_max <<= 1;
100 bio_slabs = krealloc(bio_slabs,
101 bio_slab_max * sizeof(struct bio_slab),
102 GFP_KERNEL);
103 if (!bio_slabs)
104 goto out_unlock;
105 }
106 if (entry == -1)
107 entry = bio_slab_nr++;
108
109 bslab = &bio_slabs[entry];
110
111 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
112 slab = kmem_cache_create(bslab->name, sz, 0, SLAB_HWCACHE_ALIGN, NULL);
113 if (!slab)
114 goto out_unlock;
115
116 printk("bio: create slab <%s> at %d\n", bslab->name, entry);
117 bslab->slab = slab;
118 bslab->slab_ref = 1;
119 bslab->slab_size = sz;
120out_unlock:
121 mutex_unlock(&bio_slab_lock);
122 return slab;
123}
124
125static void bio_put_slab(struct bio_set *bs)
126{
127 struct bio_slab *bslab = NULL;
128 unsigned int i;
129
130 mutex_lock(&bio_slab_lock);
131
132 for (i = 0; i < bio_slab_nr; i++) {
133 if (bs->bio_slab == bio_slabs[i].slab) {
134 bslab = &bio_slabs[i];
135 break;
136 }
137 }
138
139 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
140 goto out;
141
142 WARN_ON(!bslab->slab_ref);
143
144 if (--bslab->slab_ref)
145 goto out;
146
147 kmem_cache_destroy(bslab->slab);
148 bslab->slab = NULL;
149
150out:
151 mutex_unlock(&bio_slab_lock);
152}
153
7ba1ba12
MP
154unsigned int bvec_nr_vecs(unsigned short idx)
155{
156 return bvec_slabs[idx].nr_vecs;
157}
158
bb799ca0
JA
159void bvec_free_bs(struct bio_set *bs, struct bio_vec *bv, unsigned int idx)
160{
161 BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
162
163 if (idx == BIOVEC_MAX_IDX)
164 mempool_free(bv, bs->bvec_pool);
165 else {
166 struct biovec_slab *bvs = bvec_slabs + idx;
167
168 kmem_cache_free(bvs->slab, bv);
169 }
170}
171
7ff9345f
JA
172struct bio_vec *bvec_alloc_bs(gfp_t gfp_mask, int nr, unsigned long *idx,
173 struct bio_set *bs)
1da177e4
LT
174{
175 struct bio_vec *bvl;
1da177e4
LT
176
177 /*
0a0d96b0
JA
178 * If 'bs' is given, lookup the pool and do the mempool alloc.
179 * If not, this is a bio_kmalloc() allocation and just do a
180 * kzalloc() for the exact number of vecs right away.
1da177e4 181 */
7ff9345f 182 if (!bs)
d3f76110 183 bvl = kmalloc(nr * sizeof(struct bio_vec), gfp_mask);
7ff9345f
JA
184
185 /*
186 * see comment near bvec_array define!
187 */
188 switch (nr) {
189 case 1:
190 *idx = 0;
191 break;
192 case 2 ... 4:
193 *idx = 1;
194 break;
195 case 5 ... 16:
196 *idx = 2;
197 break;
198 case 17 ... 64:
199 *idx = 3;
200 break;
201 case 65 ... 128:
202 *idx = 4;
203 break;
204 case 129 ... BIO_MAX_PAGES:
205 *idx = 5;
206 break;
207 default:
208 return NULL;
209 }
210
211 /*
212 * idx now points to the pool we want to allocate from. only the
213 * 1-vec entry pool is mempool backed.
214 */
215 if (*idx == BIOVEC_MAX_IDX) {
216fallback:
217 bvl = mempool_alloc(bs->bvec_pool, gfp_mask);
218 } else {
219 struct biovec_slab *bvs = bvec_slabs + *idx;
220 gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
221
0a0d96b0 222 /*
7ff9345f
JA
223 * Make this allocation restricted and don't dump info on
224 * allocation failures, since we'll fallback to the mempool
225 * in case of failure.
0a0d96b0 226 */
7ff9345f 227 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
1da177e4 228
0a0d96b0 229 /*
7ff9345f
JA
230 * Try a slab allocation. If this fails and __GFP_WAIT
231 * is set, retry with the 1-entry mempool
0a0d96b0 232 */
7ff9345f
JA
233 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
234 if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
235 *idx = BIOVEC_MAX_IDX;
236 goto fallback;
237 }
238 }
239
1da177e4
LT
240 return bvl;
241}
242
7ff9345f 243void bio_free(struct bio *bio, struct bio_set *bs)
1da177e4 244{
bb799ca0 245 void *p;
1da177e4 246
392ddc32 247 if (bio_has_allocated_vec(bio))
bb799ca0 248 bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio));
1da177e4 249
7ba1ba12 250 if (bio_integrity(bio))
6d2a78e7 251 bio_integrity_free(bio);
7ba1ba12 252
bb799ca0
JA
253 /*
254 * If we have front padding, adjust the bio pointer before freeing
255 */
256 p = bio;
257 if (bs->front_pad)
258 p -= bs->front_pad;
259
260 mempool_free(p, bs->bio_pool);
3676347a
PO
261}
262
263/*
264 * default destructor for a bio allocated with bio_alloc_bioset()
265 */
266static void bio_fs_destructor(struct bio *bio)
267{
268 bio_free(bio, fs_bio_set);
1da177e4
LT
269}
270
0a0d96b0
JA
271static void bio_kmalloc_destructor(struct bio *bio)
272{
392ddc32
JA
273 if (bio_has_allocated_vec(bio))
274 kfree(bio->bi_io_vec);
0a0d96b0
JA
275 kfree(bio);
276}
277
858119e1 278void bio_init(struct bio *bio)
1da177e4 279{
2b94de55 280 memset(bio, 0, sizeof(*bio));
1da177e4 281 bio->bi_flags = 1 << BIO_UPTODATE;
c7c22e4d 282 bio->bi_comp_cpu = -1;
1da177e4 283 atomic_set(&bio->bi_cnt, 1);
1da177e4
LT
284}
285
286/**
287 * bio_alloc_bioset - allocate a bio for I/O
288 * @gfp_mask: the GFP_ mask given to the slab allocator
289 * @nr_iovecs: number of iovecs to pre-allocate
0a0d96b0 290 * @bs: the bio_set to allocate from. If %NULL, just use kmalloc
1da177e4
LT
291 *
292 * Description:
0a0d96b0 293 * bio_alloc_bioset will first try its own mempool to satisfy the allocation.
1da177e4 294 * If %__GFP_WAIT is set then we will block on the internal pool waiting
0a0d96b0
JA
295 * for a &struct bio to become free. If a %NULL @bs is passed in, we will
296 * fall back to just using @kmalloc to allocate the required memory.
1da177e4 297 *
bb799ca0
JA
298 * Note that the caller must set ->bi_destructor on succesful return
299 * of a bio, to do the appropriate freeing of the bio once the reference
300 * count drops to zero.
1da177e4 301 **/
dd0fc66f 302struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
1da177e4 303{
34053979 304 struct bio_vec *bvl = NULL;
bb799ca0 305 struct bio *bio = NULL;
34053979
IM
306 unsigned long idx = 0;
307 void *p = NULL;
bb799ca0
JA
308
309 if (bs) {
a60e78e5 310 p = mempool_alloc(bs->bio_pool, gfp_mask);
34053979
IM
311 if (!p)
312 goto err;
313 bio = p + bs->front_pad;
314 } else {
0a0d96b0 315 bio = kmalloc(sizeof(*bio), gfp_mask);
34053979
IM
316 if (!bio)
317 goto err;
318 }
1da177e4 319
34053979
IM
320 bio_init(bio);
321
322 if (unlikely(!nr_iovecs))
323 goto out_set;
324
325 if (nr_iovecs <= BIO_INLINE_VECS) {
326 bvl = bio->bi_inline_vecs;
327 nr_iovecs = BIO_INLINE_VECS;
328 } else {
329 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
330 if (unlikely(!bvl))
331 goto err_free;
332
333 nr_iovecs = bvec_nr_vecs(idx);
1da177e4 334 }
34053979
IM
335 bio->bi_flags |= idx << BIO_POOL_OFFSET;
336 bio->bi_max_vecs = nr_iovecs;
337out_set:
338 bio->bi_io_vec = bvl;
339
1da177e4 340 return bio;
34053979
IM
341
342err_free:
343 if (bs)
344 mempool_free(p, bs->bio_pool);
345 else
346 kfree(bio);
347err:
348 return NULL;
1da177e4
LT
349}
350
86c824b9
JA
351/**
352 * bio_alloc - allocate a bio for I/O
353 * @gfp_mask: the GFP_ mask given to the slab allocator
354 * @nr_iovecs: number of iovecs to pre-allocate
355 *
356 * Description:
357 * bio_alloc will allocate a bio and associated bio_vec array that can hold
358 * at least @nr_iovecs entries. Allocations will be done from the
359 * fs_bio_set. Also see @bio_alloc_bioset.
360 *
361 * If %__GFP_WAIT is set, then bio_alloc will always be able to allocate
362 * a bio. This is due to the mempool guarantees. To make this work, callers
363 * must never allocate more than 1 bio at the time from this pool. Callers
364 * that need to allocate more than 1 bio must always submit the previously
365 * allocate bio for IO before attempting to allocate a new one. Failure to
366 * do so can cause livelocks under memory pressure.
367 *
368 **/
dd0fc66f 369struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
1da177e4 370{
3676347a
PO
371 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
372
373 if (bio)
374 bio->bi_destructor = bio_fs_destructor;
375
376 return bio;
1da177e4
LT
377}
378
0a0d96b0
JA
379/*
380 * Like bio_alloc(), but doesn't use a mempool backing. This means that
381 * it CAN fail, but while bio_alloc() can only be used for allocations
382 * that have a short (finite) life span, bio_kmalloc() should be used
383 * for more permanent bio allocations (like allocating some bio's for
384 * initalization or setup purposes).
385 */
386struct bio *bio_kmalloc(gfp_t gfp_mask, int nr_iovecs)
387{
388 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, NULL);
389
390 if (bio)
391 bio->bi_destructor = bio_kmalloc_destructor;
392
393 return bio;
394}
395
1da177e4
LT
396void zero_fill_bio(struct bio *bio)
397{
398 unsigned long flags;
399 struct bio_vec *bv;
400 int i;
401
402 bio_for_each_segment(bv, bio, i) {
403 char *data = bvec_kmap_irq(bv, &flags);
404 memset(data, 0, bv->bv_len);
405 flush_dcache_page(bv->bv_page);
406 bvec_kunmap_irq(data, &flags);
407 }
408}
409EXPORT_SYMBOL(zero_fill_bio);
410
411/**
412 * bio_put - release a reference to a bio
413 * @bio: bio to release reference to
414 *
415 * Description:
416 * Put a reference to a &struct bio, either one you have gotten with
417 * bio_alloc or bio_get. The last put of a bio will free it.
418 **/
419void bio_put(struct bio *bio)
420{
421 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
422
423 /*
424 * last put frees it
425 */
426 if (atomic_dec_and_test(&bio->bi_cnt)) {
427 bio->bi_next = NULL;
428 bio->bi_destructor(bio);
429 }
430}
431
165125e1 432inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
1da177e4
LT
433{
434 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
435 blk_recount_segments(q, bio);
436
437 return bio->bi_phys_segments;
438}
439
1da177e4
LT
440/**
441 * __bio_clone - clone a bio
442 * @bio: destination bio
443 * @bio_src: bio to clone
444 *
445 * Clone a &bio. Caller will own the returned bio, but not
446 * the actual data it points to. Reference count of returned
447 * bio will be one.
448 */
858119e1 449void __bio_clone(struct bio *bio, struct bio *bio_src)
1da177e4 450{
e525e153
AM
451 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
452 bio_src->bi_max_vecs * sizeof(struct bio_vec));
1da177e4 453
5d84070e
JA
454 /*
455 * most users will be overriding ->bi_bdev with a new target,
456 * so we don't set nor calculate new physical/hw segment counts here
457 */
1da177e4
LT
458 bio->bi_sector = bio_src->bi_sector;
459 bio->bi_bdev = bio_src->bi_bdev;
460 bio->bi_flags |= 1 << BIO_CLONED;
461 bio->bi_rw = bio_src->bi_rw;
1da177e4
LT
462 bio->bi_vcnt = bio_src->bi_vcnt;
463 bio->bi_size = bio_src->bi_size;
a5453be4 464 bio->bi_idx = bio_src->bi_idx;
1da177e4
LT
465}
466
467/**
468 * bio_clone - clone a bio
469 * @bio: bio to clone
470 * @gfp_mask: allocation priority
471 *
472 * Like __bio_clone, only also allocates the returned bio
473 */
dd0fc66f 474struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
1da177e4
LT
475{
476 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
477
7ba1ba12
MP
478 if (!b)
479 return NULL;
480
481 b->bi_destructor = bio_fs_destructor;
482 __bio_clone(b, bio);
483
484 if (bio_integrity(bio)) {
485 int ret;
486
6d2a78e7 487 ret = bio_integrity_clone(b, bio, gfp_mask);
7ba1ba12 488
059ea331
LZ
489 if (ret < 0) {
490 bio_put(b);
7ba1ba12 491 return NULL;
059ea331 492 }
3676347a 493 }
1da177e4
LT
494
495 return b;
496}
497
498/**
499 * bio_get_nr_vecs - return approx number of vecs
500 * @bdev: I/O target
501 *
502 * Return the approximate number of pages we can send to this target.
503 * There's no guarantee that you will be able to fit this number of pages
504 * into a bio, it does not account for dynamic restrictions that vary
505 * on offset.
506 */
507int bio_get_nr_vecs(struct block_device *bdev)
508{
165125e1 509 struct request_queue *q = bdev_get_queue(bdev);
1da177e4
LT
510 int nr_pages;
511
512 nr_pages = ((q->max_sectors << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
513 if (nr_pages > q->max_phys_segments)
514 nr_pages = q->max_phys_segments;
515 if (nr_pages > q->max_hw_segments)
516 nr_pages = q->max_hw_segments;
517
518 return nr_pages;
519}
520
165125e1 521static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
defd94b7
MC
522 *page, unsigned int len, unsigned int offset,
523 unsigned short max_sectors)
1da177e4
LT
524{
525 int retried_segments = 0;
526 struct bio_vec *bvec;
527
528 /*
529 * cloned bio must not modify vec list
530 */
531 if (unlikely(bio_flagged(bio, BIO_CLONED)))
532 return 0;
533
80cfd548 534 if (((bio->bi_size + len) >> 9) > max_sectors)
1da177e4
LT
535 return 0;
536
80cfd548
JA
537 /*
538 * For filesystems with a blocksize smaller than the pagesize
539 * we will often be called with the same page as last time and
540 * a consecutive offset. Optimize this special case.
541 */
542 if (bio->bi_vcnt > 0) {
543 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
544
545 if (page == prev->bv_page &&
546 offset == prev->bv_offset + prev->bv_len) {
547 prev->bv_len += len;
cc371e66
AK
548
549 if (q->merge_bvec_fn) {
550 struct bvec_merge_data bvm = {
551 .bi_bdev = bio->bi_bdev,
552 .bi_sector = bio->bi_sector,
553 .bi_size = bio->bi_size,
554 .bi_rw = bio->bi_rw,
555 };
556
557 if (q->merge_bvec_fn(q, &bvm, prev) < len) {
558 prev->bv_len -= len;
559 return 0;
560 }
80cfd548
JA
561 }
562
563 goto done;
564 }
565 }
566
567 if (bio->bi_vcnt >= bio->bi_max_vecs)
1da177e4
LT
568 return 0;
569
570 /*
571 * we might lose a segment or two here, but rather that than
572 * make this too complex.
573 */
574
575 while (bio->bi_phys_segments >= q->max_phys_segments
5df97b91 576 || bio->bi_phys_segments >= q->max_hw_segments) {
1da177e4
LT
577
578 if (retried_segments)
579 return 0;
580
581 retried_segments = 1;
582 blk_recount_segments(q, bio);
583 }
584
585 /*
586 * setup the new entry, we might clear it again later if we
587 * cannot add the page
588 */
589 bvec = &bio->bi_io_vec[bio->bi_vcnt];
590 bvec->bv_page = page;
591 bvec->bv_len = len;
592 bvec->bv_offset = offset;
593
594 /*
595 * if queue has other restrictions (eg varying max sector size
596 * depending on offset), it can specify a merge_bvec_fn in the
597 * queue to get further control
598 */
599 if (q->merge_bvec_fn) {
cc371e66
AK
600 struct bvec_merge_data bvm = {
601 .bi_bdev = bio->bi_bdev,
602 .bi_sector = bio->bi_sector,
603 .bi_size = bio->bi_size,
604 .bi_rw = bio->bi_rw,
605 };
606
1da177e4
LT
607 /*
608 * merge_bvec_fn() returns number of bytes it can accept
609 * at this offset
610 */
cc371e66 611 if (q->merge_bvec_fn(q, &bvm, bvec) < len) {
1da177e4
LT
612 bvec->bv_page = NULL;
613 bvec->bv_len = 0;
614 bvec->bv_offset = 0;
615 return 0;
616 }
617 }
618
619 /* If we may be able to merge these biovecs, force a recount */
b8b3e16c 620 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
1da177e4
LT
621 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
622
623 bio->bi_vcnt++;
624 bio->bi_phys_segments++;
80cfd548 625 done:
1da177e4
LT
626 bio->bi_size += len;
627 return len;
628}
629
6e68af66
MC
630/**
631 * bio_add_pc_page - attempt to add page to bio
fddfdeaf 632 * @q: the target queue
6e68af66
MC
633 * @bio: destination bio
634 * @page: page to add
635 * @len: vec entry length
636 * @offset: vec entry offset
637 *
638 * Attempt to add a page to the bio_vec maplist. This can fail for a
639 * number of reasons, such as the bio being full or target block
640 * device limitations. The target block device must allow bio's
641 * smaller than PAGE_SIZE, so it is always possible to add a single
642 * page to an empty bio. This should only be used by REQ_PC bios.
643 */
165125e1 644int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
6e68af66
MC
645 unsigned int len, unsigned int offset)
646{
defd94b7 647 return __bio_add_page(q, bio, page, len, offset, q->max_hw_sectors);
6e68af66
MC
648}
649
1da177e4
LT
650/**
651 * bio_add_page - attempt to add page to bio
652 * @bio: destination bio
653 * @page: page to add
654 * @len: vec entry length
655 * @offset: vec entry offset
656 *
657 * Attempt to add a page to the bio_vec maplist. This can fail for a
658 * number of reasons, such as the bio being full or target block
659 * device limitations. The target block device must allow bio's
660 * smaller than PAGE_SIZE, so it is always possible to add a single
661 * page to an empty bio.
662 */
663int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
664 unsigned int offset)
665{
defd94b7
MC
666 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
667 return __bio_add_page(q, bio, page, len, offset, q->max_sectors);
1da177e4
LT
668}
669
670struct bio_map_data {
671 struct bio_vec *iovecs;
c5dec1c3 672 struct sg_iovec *sgvecs;
152e283f
FT
673 int nr_sgvecs;
674 int is_our_pages;
1da177e4
LT
675};
676
c5dec1c3 677static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
152e283f
FT
678 struct sg_iovec *iov, int iov_count,
679 int is_our_pages)
1da177e4
LT
680{
681 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
c5dec1c3
FT
682 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
683 bmd->nr_sgvecs = iov_count;
152e283f 684 bmd->is_our_pages = is_our_pages;
1da177e4
LT
685 bio->bi_private = bmd;
686}
687
688static void bio_free_map_data(struct bio_map_data *bmd)
689{
690 kfree(bmd->iovecs);
c5dec1c3 691 kfree(bmd->sgvecs);
1da177e4
LT
692 kfree(bmd);
693}
694
76029ff3
FT
695static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count,
696 gfp_t gfp_mask)
1da177e4 697{
76029ff3 698 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask);
1da177e4
LT
699
700 if (!bmd)
701 return NULL;
702
76029ff3 703 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
c5dec1c3
FT
704 if (!bmd->iovecs) {
705 kfree(bmd);
706 return NULL;
707 }
708
76029ff3 709 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
c5dec1c3 710 if (bmd->sgvecs)
1da177e4
LT
711 return bmd;
712
c5dec1c3 713 kfree(bmd->iovecs);
1da177e4
LT
714 kfree(bmd);
715 return NULL;
716}
717
aefcc28a 718static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
152e283f
FT
719 struct sg_iovec *iov, int iov_count, int uncopy,
720 int do_free_page)
c5dec1c3
FT
721{
722 int ret = 0, i;
723 struct bio_vec *bvec;
724 int iov_idx = 0;
725 unsigned int iov_off = 0;
726 int read = bio_data_dir(bio) == READ;
727
728 __bio_for_each_segment(bvec, bio, i, 0) {
729 char *bv_addr = page_address(bvec->bv_page);
aefcc28a 730 unsigned int bv_len = iovecs[i].bv_len;
c5dec1c3
FT
731
732 while (bv_len && iov_idx < iov_count) {
733 unsigned int bytes;
734 char *iov_addr;
735
736 bytes = min_t(unsigned int,
737 iov[iov_idx].iov_len - iov_off, bv_len);
738 iov_addr = iov[iov_idx].iov_base + iov_off;
739
740 if (!ret) {
741 if (!read && !uncopy)
742 ret = copy_from_user(bv_addr, iov_addr,
743 bytes);
744 if (read && uncopy)
745 ret = copy_to_user(iov_addr, bv_addr,
746 bytes);
747
748 if (ret)
749 ret = -EFAULT;
750 }
751
752 bv_len -= bytes;
753 bv_addr += bytes;
754 iov_addr += bytes;
755 iov_off += bytes;
756
757 if (iov[iov_idx].iov_len == iov_off) {
758 iov_idx++;
759 iov_off = 0;
760 }
761 }
762
152e283f 763 if (do_free_page)
c5dec1c3
FT
764 __free_page(bvec->bv_page);
765 }
766
767 return ret;
768}
769
1da177e4
LT
770/**
771 * bio_uncopy_user - finish previously mapped bio
772 * @bio: bio being terminated
773 *
774 * Free pages allocated from bio_copy_user() and write back data
775 * to user space in case of a read.
776 */
777int bio_uncopy_user(struct bio *bio)
778{
779 struct bio_map_data *bmd = bio->bi_private;
81882766 780 int ret = 0;
1da177e4 781
81882766
FT
782 if (!bio_flagged(bio, BIO_NULL_MAPPED))
783 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
784 bmd->nr_sgvecs, 1, bmd->is_our_pages);
1da177e4
LT
785 bio_free_map_data(bmd);
786 bio_put(bio);
787 return ret;
788}
789
790/**
c5dec1c3 791 * bio_copy_user_iov - copy user data to bio
1da177e4 792 * @q: destination block queue
152e283f 793 * @map_data: pointer to the rq_map_data holding pages (if necessary)
c5dec1c3
FT
794 * @iov: the iovec.
795 * @iov_count: number of elements in the iovec
1da177e4 796 * @write_to_vm: bool indicating writing to pages or not
a3bce90e 797 * @gfp_mask: memory allocation flags
1da177e4
LT
798 *
799 * Prepares and returns a bio for indirect user io, bouncing data
800 * to/from kernel pages as necessary. Must be paired with
801 * call bio_uncopy_user() on io completion.
802 */
152e283f
FT
803struct bio *bio_copy_user_iov(struct request_queue *q,
804 struct rq_map_data *map_data,
805 struct sg_iovec *iov, int iov_count,
806 int write_to_vm, gfp_t gfp_mask)
1da177e4 807{
1da177e4
LT
808 struct bio_map_data *bmd;
809 struct bio_vec *bvec;
810 struct page *page;
811 struct bio *bio;
812 int i, ret;
c5dec1c3
FT
813 int nr_pages = 0;
814 unsigned int len = 0;
56c451f4 815 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
1da177e4 816
c5dec1c3
FT
817 for (i = 0; i < iov_count; i++) {
818 unsigned long uaddr;
819 unsigned long end;
820 unsigned long start;
821
822 uaddr = (unsigned long)iov[i].iov_base;
823 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
824 start = uaddr >> PAGE_SHIFT;
825
826 nr_pages += end - start;
827 len += iov[i].iov_len;
828 }
829
a3bce90e 830 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
1da177e4
LT
831 if (!bmd)
832 return ERR_PTR(-ENOMEM);
833
1da177e4 834 ret = -ENOMEM;
a3bce90e 835 bio = bio_alloc(gfp_mask, nr_pages);
1da177e4
LT
836 if (!bio)
837 goto out_bmd;
838
839 bio->bi_rw |= (!write_to_vm << BIO_RW);
840
841 ret = 0;
56c451f4
FT
842
843 if (map_data) {
e623ddb4 844 nr_pages = 1 << map_data->page_order;
56c451f4
FT
845 i = map_data->offset / PAGE_SIZE;
846 }
1da177e4 847 while (len) {
e623ddb4 848 unsigned int bytes = PAGE_SIZE;
1da177e4 849
56c451f4
FT
850 bytes -= offset;
851
1da177e4
LT
852 if (bytes > len)
853 bytes = len;
854
152e283f 855 if (map_data) {
e623ddb4 856 if (i == map_data->nr_entries * nr_pages) {
152e283f
FT
857 ret = -ENOMEM;
858 break;
859 }
e623ddb4
FT
860
861 page = map_data->pages[i / nr_pages];
862 page += (i % nr_pages);
863
864 i++;
865 } else {
152e283f 866 page = alloc_page(q->bounce_gfp | gfp_mask);
e623ddb4
FT
867 if (!page) {
868 ret = -ENOMEM;
869 break;
870 }
1da177e4
LT
871 }
872
56c451f4 873 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
1da177e4 874 break;
1da177e4
LT
875
876 len -= bytes;
56c451f4 877 offset = 0;
1da177e4
LT
878 }
879
880 if (ret)
881 goto cleanup;
882
883 /*
884 * success
885 */
97ae77a1 886 if (!write_to_vm && (!map_data || !map_data->null_mapped)) {
152e283f 887 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 0);
c5dec1c3
FT
888 if (ret)
889 goto cleanup;
1da177e4
LT
890 }
891
152e283f 892 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
1da177e4
LT
893 return bio;
894cleanup:
152e283f
FT
895 if (!map_data)
896 bio_for_each_segment(bvec, bio, i)
897 __free_page(bvec->bv_page);
1da177e4
LT
898
899 bio_put(bio);
900out_bmd:
901 bio_free_map_data(bmd);
902 return ERR_PTR(ret);
903}
904
c5dec1c3
FT
905/**
906 * bio_copy_user - copy user data to bio
907 * @q: destination block queue
152e283f 908 * @map_data: pointer to the rq_map_data holding pages (if necessary)
c5dec1c3
FT
909 * @uaddr: start of user address
910 * @len: length in bytes
911 * @write_to_vm: bool indicating writing to pages or not
a3bce90e 912 * @gfp_mask: memory allocation flags
c5dec1c3
FT
913 *
914 * Prepares and returns a bio for indirect user io, bouncing data
915 * to/from kernel pages as necessary. Must be paired with
916 * call bio_uncopy_user() on io completion.
917 */
152e283f
FT
918struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
919 unsigned long uaddr, unsigned int len,
920 int write_to_vm, gfp_t gfp_mask)
c5dec1c3
FT
921{
922 struct sg_iovec iov;
923
924 iov.iov_base = (void __user *)uaddr;
925 iov.iov_len = len;
926
152e283f 927 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
c5dec1c3
FT
928}
929
165125e1 930static struct bio *__bio_map_user_iov(struct request_queue *q,
f1970baf
JB
931 struct block_device *bdev,
932 struct sg_iovec *iov, int iov_count,
a3bce90e 933 int write_to_vm, gfp_t gfp_mask)
1da177e4 934{
f1970baf
JB
935 int i, j;
936 int nr_pages = 0;
1da177e4
LT
937 struct page **pages;
938 struct bio *bio;
f1970baf
JB
939 int cur_page = 0;
940 int ret, offset;
1da177e4 941
f1970baf
JB
942 for (i = 0; i < iov_count; i++) {
943 unsigned long uaddr = (unsigned long)iov[i].iov_base;
944 unsigned long len = iov[i].iov_len;
945 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
946 unsigned long start = uaddr >> PAGE_SHIFT;
947
948 nr_pages += end - start;
949 /*
ad2d7225 950 * buffer must be aligned to at least hardsector size for now
f1970baf 951 */
ad2d7225 952 if (uaddr & queue_dma_alignment(q))
f1970baf
JB
953 return ERR_PTR(-EINVAL);
954 }
955
956 if (!nr_pages)
1da177e4
LT
957 return ERR_PTR(-EINVAL);
958
a3bce90e 959 bio = bio_alloc(gfp_mask, nr_pages);
1da177e4
LT
960 if (!bio)
961 return ERR_PTR(-ENOMEM);
962
963 ret = -ENOMEM;
a3bce90e 964 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
1da177e4
LT
965 if (!pages)
966 goto out;
967
f1970baf
JB
968 for (i = 0; i < iov_count; i++) {
969 unsigned long uaddr = (unsigned long)iov[i].iov_base;
970 unsigned long len = iov[i].iov_len;
971 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
972 unsigned long start = uaddr >> PAGE_SHIFT;
973 const int local_nr_pages = end - start;
974 const int page_limit = cur_page + local_nr_pages;
975
f5dd33c4
NP
976 ret = get_user_pages_fast(uaddr, local_nr_pages,
977 write_to_vm, &pages[cur_page]);
99172157
JA
978 if (ret < local_nr_pages) {
979 ret = -EFAULT;
f1970baf 980 goto out_unmap;
99172157 981 }
f1970baf
JB
982
983 offset = uaddr & ~PAGE_MASK;
984 for (j = cur_page; j < page_limit; j++) {
985 unsigned int bytes = PAGE_SIZE - offset;
986
987 if (len <= 0)
988 break;
989
990 if (bytes > len)
991 bytes = len;
992
993 /*
994 * sorry...
995 */
defd94b7
MC
996 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
997 bytes)
f1970baf
JB
998 break;
999
1000 len -= bytes;
1001 offset = 0;
1002 }
1da177e4 1003
f1970baf 1004 cur_page = j;
1da177e4 1005 /*
f1970baf 1006 * release the pages we didn't map into the bio, if any
1da177e4 1007 */
f1970baf
JB
1008 while (j < page_limit)
1009 page_cache_release(pages[j++]);
1da177e4
LT
1010 }
1011
1da177e4
LT
1012 kfree(pages);
1013
1014 /*
1015 * set data direction, and check if mapped pages need bouncing
1016 */
1017 if (!write_to_vm)
1018 bio->bi_rw |= (1 << BIO_RW);
1019
f1970baf 1020 bio->bi_bdev = bdev;
1da177e4
LT
1021 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1022 return bio;
f1970baf
JB
1023
1024 out_unmap:
1025 for (i = 0; i < nr_pages; i++) {
1026 if(!pages[i])
1027 break;
1028 page_cache_release(pages[i]);
1029 }
1030 out:
1da177e4
LT
1031 kfree(pages);
1032 bio_put(bio);
1033 return ERR_PTR(ret);
1034}
1035
1036/**
1037 * bio_map_user - map user address into bio
165125e1 1038 * @q: the struct request_queue for the bio
1da177e4
LT
1039 * @bdev: destination block device
1040 * @uaddr: start of user address
1041 * @len: length in bytes
1042 * @write_to_vm: bool indicating writing to pages or not
a3bce90e 1043 * @gfp_mask: memory allocation flags
1da177e4
LT
1044 *
1045 * Map the user space address into a bio suitable for io to a block
1046 * device. Returns an error pointer in case of error.
1047 */
165125e1 1048struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
a3bce90e
FT
1049 unsigned long uaddr, unsigned int len, int write_to_vm,
1050 gfp_t gfp_mask)
f1970baf
JB
1051{
1052 struct sg_iovec iov;
1053
3f70353e 1054 iov.iov_base = (void __user *)uaddr;
f1970baf
JB
1055 iov.iov_len = len;
1056
a3bce90e 1057 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
f1970baf
JB
1058}
1059
1060/**
1061 * bio_map_user_iov - map user sg_iovec table into bio
165125e1 1062 * @q: the struct request_queue for the bio
f1970baf
JB
1063 * @bdev: destination block device
1064 * @iov: the iovec.
1065 * @iov_count: number of elements in the iovec
1066 * @write_to_vm: bool indicating writing to pages or not
a3bce90e 1067 * @gfp_mask: memory allocation flags
f1970baf
JB
1068 *
1069 * Map the user space address into a bio suitable for io to a block
1070 * device. Returns an error pointer in case of error.
1071 */
165125e1 1072struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
f1970baf 1073 struct sg_iovec *iov, int iov_count,
a3bce90e 1074 int write_to_vm, gfp_t gfp_mask)
1da177e4
LT
1075{
1076 struct bio *bio;
1077
a3bce90e
FT
1078 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1079 gfp_mask);
1da177e4
LT
1080 if (IS_ERR(bio))
1081 return bio;
1082
1083 /*
1084 * subtle -- if __bio_map_user() ended up bouncing a bio,
1085 * it would normally disappear when its bi_end_io is run.
1086 * however, we need it for the unmap, so grab an extra
1087 * reference to it
1088 */
1089 bio_get(bio);
1090
0e75f906 1091 return bio;
1da177e4
LT
1092}
1093
1094static void __bio_unmap_user(struct bio *bio)
1095{
1096 struct bio_vec *bvec;
1097 int i;
1098
1099 /*
1100 * make sure we dirty pages we wrote to
1101 */
1102 __bio_for_each_segment(bvec, bio, i, 0) {
1103 if (bio_data_dir(bio) == READ)
1104 set_page_dirty_lock(bvec->bv_page);
1105
1106 page_cache_release(bvec->bv_page);
1107 }
1108
1109 bio_put(bio);
1110}
1111
1112/**
1113 * bio_unmap_user - unmap a bio
1114 * @bio: the bio being unmapped
1115 *
1116 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1117 * a process context.
1118 *
1119 * bio_unmap_user() may sleep.
1120 */
1121void bio_unmap_user(struct bio *bio)
1122{
1123 __bio_unmap_user(bio);
1124 bio_put(bio);
1125}
1126
6712ecf8 1127static void bio_map_kern_endio(struct bio *bio, int err)
b823825e 1128{
b823825e 1129 bio_put(bio);
b823825e
JA
1130}
1131
1132
165125e1 1133static struct bio *__bio_map_kern(struct request_queue *q, void *data,
27496a8c 1134 unsigned int len, gfp_t gfp_mask)
df46b9a4
MC
1135{
1136 unsigned long kaddr = (unsigned long)data;
1137 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1138 unsigned long start = kaddr >> PAGE_SHIFT;
1139 const int nr_pages = end - start;
1140 int offset, i;
1141 struct bio *bio;
1142
1143 bio = bio_alloc(gfp_mask, nr_pages);
1144 if (!bio)
1145 return ERR_PTR(-ENOMEM);
1146
1147 offset = offset_in_page(kaddr);
1148 for (i = 0; i < nr_pages; i++) {
1149 unsigned int bytes = PAGE_SIZE - offset;
1150
1151 if (len <= 0)
1152 break;
1153
1154 if (bytes > len)
1155 bytes = len;
1156
defd94b7
MC
1157 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1158 offset) < bytes)
df46b9a4
MC
1159 break;
1160
1161 data += bytes;
1162 len -= bytes;
1163 offset = 0;
1164 }
1165
b823825e 1166 bio->bi_end_io = bio_map_kern_endio;
df46b9a4
MC
1167 return bio;
1168}
1169
1170/**
1171 * bio_map_kern - map kernel address into bio
165125e1 1172 * @q: the struct request_queue for the bio
df46b9a4
MC
1173 * @data: pointer to buffer to map
1174 * @len: length in bytes
1175 * @gfp_mask: allocation flags for bio allocation
1176 *
1177 * Map the kernel address into a bio suitable for io to a block
1178 * device. Returns an error pointer in case of error.
1179 */
165125e1 1180struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
27496a8c 1181 gfp_t gfp_mask)
df46b9a4
MC
1182{
1183 struct bio *bio;
1184
1185 bio = __bio_map_kern(q, data, len, gfp_mask);
1186 if (IS_ERR(bio))
1187 return bio;
1188
1189 if (bio->bi_size == len)
1190 return bio;
1191
1192 /*
1193 * Don't support partial mappings.
1194 */
1195 bio_put(bio);
1196 return ERR_PTR(-EINVAL);
1197}
1198
68154e90
FT
1199static void bio_copy_kern_endio(struct bio *bio, int err)
1200{
1201 struct bio_vec *bvec;
1202 const int read = bio_data_dir(bio) == READ;
76029ff3 1203 struct bio_map_data *bmd = bio->bi_private;
68154e90 1204 int i;
76029ff3 1205 char *p = bmd->sgvecs[0].iov_base;
68154e90
FT
1206
1207 __bio_for_each_segment(bvec, bio, i, 0) {
1208 char *addr = page_address(bvec->bv_page);
76029ff3 1209 int len = bmd->iovecs[i].bv_len;
68154e90
FT
1210
1211 if (read && !err)
76029ff3 1212 memcpy(p, addr, len);
68154e90
FT
1213
1214 __free_page(bvec->bv_page);
76029ff3 1215 p += len;
68154e90
FT
1216 }
1217
76029ff3 1218 bio_free_map_data(bmd);
68154e90
FT
1219 bio_put(bio);
1220}
1221
1222/**
1223 * bio_copy_kern - copy kernel address into bio
1224 * @q: the struct request_queue for the bio
1225 * @data: pointer to buffer to copy
1226 * @len: length in bytes
1227 * @gfp_mask: allocation flags for bio and page allocation
ffee0259 1228 * @reading: data direction is READ
68154e90
FT
1229 *
1230 * copy the kernel address into a bio suitable for io to a block
1231 * device. Returns an error pointer in case of error.
1232 */
1233struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1234 gfp_t gfp_mask, int reading)
1235{
68154e90
FT
1236 struct bio *bio;
1237 struct bio_vec *bvec;
4d8ab62e 1238 int i;
68154e90 1239
4d8ab62e
FT
1240 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1241 if (IS_ERR(bio))
1242 return bio;
68154e90
FT
1243
1244 if (!reading) {
1245 void *p = data;
1246
1247 bio_for_each_segment(bvec, bio, i) {
1248 char *addr = page_address(bvec->bv_page);
1249
1250 memcpy(addr, p, bvec->bv_len);
1251 p += bvec->bv_len;
1252 }
1253 }
1254
68154e90 1255 bio->bi_end_io = bio_copy_kern_endio;
76029ff3 1256
68154e90 1257 return bio;
68154e90
FT
1258}
1259
1da177e4
LT
1260/*
1261 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1262 * for performing direct-IO in BIOs.
1263 *
1264 * The problem is that we cannot run set_page_dirty() from interrupt context
1265 * because the required locks are not interrupt-safe. So what we can do is to
1266 * mark the pages dirty _before_ performing IO. And in interrupt context,
1267 * check that the pages are still dirty. If so, fine. If not, redirty them
1268 * in process context.
1269 *
1270 * We special-case compound pages here: normally this means reads into hugetlb
1271 * pages. The logic in here doesn't really work right for compound pages
1272 * because the VM does not uniformly chase down the head page in all cases.
1273 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1274 * handle them at all. So we skip compound pages here at an early stage.
1275 *
1276 * Note that this code is very hard to test under normal circumstances because
1277 * direct-io pins the pages with get_user_pages(). This makes
1278 * is_page_cache_freeable return false, and the VM will not clean the pages.
1279 * But other code (eg, pdflush) could clean the pages if they are mapped
1280 * pagecache.
1281 *
1282 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1283 * deferred bio dirtying paths.
1284 */
1285
1286/*
1287 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1288 */
1289void bio_set_pages_dirty(struct bio *bio)
1290{
1291 struct bio_vec *bvec = bio->bi_io_vec;
1292 int i;
1293
1294 for (i = 0; i < bio->bi_vcnt; i++) {
1295 struct page *page = bvec[i].bv_page;
1296
1297 if (page && !PageCompound(page))
1298 set_page_dirty_lock(page);
1299 }
1300}
1301
86b6c7a7 1302static void bio_release_pages(struct bio *bio)
1da177e4
LT
1303{
1304 struct bio_vec *bvec = bio->bi_io_vec;
1305 int i;
1306
1307 for (i = 0; i < bio->bi_vcnt; i++) {
1308 struct page *page = bvec[i].bv_page;
1309
1310 if (page)
1311 put_page(page);
1312 }
1313}
1314
1315/*
1316 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1317 * If they are, then fine. If, however, some pages are clean then they must
1318 * have been written out during the direct-IO read. So we take another ref on
1319 * the BIO and the offending pages and re-dirty the pages in process context.
1320 *
1321 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1322 * here on. It will run one page_cache_release() against each page and will
1323 * run one bio_put() against the BIO.
1324 */
1325
65f27f38 1326static void bio_dirty_fn(struct work_struct *work);
1da177e4 1327
65f27f38 1328static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1da177e4
LT
1329static DEFINE_SPINLOCK(bio_dirty_lock);
1330static struct bio *bio_dirty_list;
1331
1332/*
1333 * This runs in process context
1334 */
65f27f38 1335static void bio_dirty_fn(struct work_struct *work)
1da177e4
LT
1336{
1337 unsigned long flags;
1338 struct bio *bio;
1339
1340 spin_lock_irqsave(&bio_dirty_lock, flags);
1341 bio = bio_dirty_list;
1342 bio_dirty_list = NULL;
1343 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1344
1345 while (bio) {
1346 struct bio *next = bio->bi_private;
1347
1348 bio_set_pages_dirty(bio);
1349 bio_release_pages(bio);
1350 bio_put(bio);
1351 bio = next;
1352 }
1353}
1354
1355void bio_check_pages_dirty(struct bio *bio)
1356{
1357 struct bio_vec *bvec = bio->bi_io_vec;
1358 int nr_clean_pages = 0;
1359 int i;
1360
1361 for (i = 0; i < bio->bi_vcnt; i++) {
1362 struct page *page = bvec[i].bv_page;
1363
1364 if (PageDirty(page) || PageCompound(page)) {
1365 page_cache_release(page);
1366 bvec[i].bv_page = NULL;
1367 } else {
1368 nr_clean_pages++;
1369 }
1370 }
1371
1372 if (nr_clean_pages) {
1373 unsigned long flags;
1374
1375 spin_lock_irqsave(&bio_dirty_lock, flags);
1376 bio->bi_private = bio_dirty_list;
1377 bio_dirty_list = bio;
1378 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1379 schedule_work(&bio_dirty_work);
1380 } else {
1381 bio_put(bio);
1382 }
1383}
1384
1385/**
1386 * bio_endio - end I/O on a bio
1387 * @bio: bio
1da177e4
LT
1388 * @error: error, if any
1389 *
1390 * Description:
6712ecf8 1391 * bio_endio() will end I/O on the whole bio. bio_endio() is the
5bb23a68
N
1392 * preferred way to end I/O on a bio, it takes care of clearing
1393 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1394 * established -Exxxx (-EIO, for instance) error values in case
1395 * something went wrong. Noone should call bi_end_io() directly on a
1396 * bio unless they own it and thus know that it has an end_io
1397 * function.
1da177e4 1398 **/
6712ecf8 1399void bio_endio(struct bio *bio, int error)
1da177e4
LT
1400{
1401 if (error)
1402 clear_bit(BIO_UPTODATE, &bio->bi_flags);
9cc54d40
N
1403 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1404 error = -EIO;
1da177e4 1405
5bb23a68 1406 if (bio->bi_end_io)
6712ecf8 1407 bio->bi_end_io(bio, error);
1da177e4
LT
1408}
1409
1410void bio_pair_release(struct bio_pair *bp)
1411{
1412 if (atomic_dec_and_test(&bp->cnt)) {
1413 struct bio *master = bp->bio1.bi_private;
1414
6712ecf8 1415 bio_endio(master, bp->error);
1da177e4
LT
1416 mempool_free(bp, bp->bio2.bi_private);
1417 }
1418}
1419
6712ecf8 1420static void bio_pair_end_1(struct bio *bi, int err)
1da177e4
LT
1421{
1422 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1423
1424 if (err)
1425 bp->error = err;
1426
1da177e4 1427 bio_pair_release(bp);
1da177e4
LT
1428}
1429
6712ecf8 1430static void bio_pair_end_2(struct bio *bi, int err)
1da177e4
LT
1431{
1432 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1433
1434 if (err)
1435 bp->error = err;
1436
1da177e4 1437 bio_pair_release(bp);
1da177e4
LT
1438}
1439
1440/*
c7eee1b8 1441 * split a bio - only worry about a bio with a single page in its iovec
1da177e4 1442 */
6feef531 1443struct bio_pair *bio_split(struct bio *bi, int first_sectors)
1da177e4 1444{
6feef531 1445 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
1da177e4
LT
1446
1447 if (!bp)
1448 return bp;
1449
5f3ea37c 1450 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
2056a782
JA
1451 bi->bi_sector + first_sectors);
1452
1da177e4
LT
1453 BUG_ON(bi->bi_vcnt != 1);
1454 BUG_ON(bi->bi_idx != 0);
1455 atomic_set(&bp->cnt, 3);
1456 bp->error = 0;
1457 bp->bio1 = *bi;
1458 bp->bio2 = *bi;
1459 bp->bio2.bi_sector += first_sectors;
1460 bp->bio2.bi_size -= first_sectors << 9;
1461 bp->bio1.bi_size = first_sectors << 9;
1462
1463 bp->bv1 = bi->bi_io_vec[0];
1464 bp->bv2 = bi->bi_io_vec[0];
1465 bp->bv2.bv_offset += first_sectors << 9;
1466 bp->bv2.bv_len -= first_sectors << 9;
1467 bp->bv1.bv_len = first_sectors << 9;
1468
1469 bp->bio1.bi_io_vec = &bp->bv1;
1470 bp->bio2.bi_io_vec = &bp->bv2;
1471
a2eb0c10
N
1472 bp->bio1.bi_max_vecs = 1;
1473 bp->bio2.bi_max_vecs = 1;
1474
1da177e4
LT
1475 bp->bio1.bi_end_io = bio_pair_end_1;
1476 bp->bio2.bi_end_io = bio_pair_end_2;
1477
1478 bp->bio1.bi_private = bi;
6feef531 1479 bp->bio2.bi_private = bio_split_pool;
1da177e4 1480
7ba1ba12
MP
1481 if (bio_integrity(bi))
1482 bio_integrity_split(bi, bp, first_sectors);
1483
1da177e4
LT
1484 return bp;
1485}
1486
ad3316bf
MP
1487/**
1488 * bio_sector_offset - Find hardware sector offset in bio
1489 * @bio: bio to inspect
1490 * @index: bio_vec index
1491 * @offset: offset in bv_page
1492 *
1493 * Return the number of hardware sectors between beginning of bio
1494 * and an end point indicated by a bio_vec index and an offset
1495 * within that vector's page.
1496 */
1497sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1498 unsigned int offset)
1499{
1500 unsigned int sector_sz = queue_hardsect_size(bio->bi_bdev->bd_disk->queue);
1501 struct bio_vec *bv;
1502 sector_t sectors;
1503 int i;
1504
1505 sectors = 0;
1506
1507 if (index >= bio->bi_idx)
1508 index = bio->bi_vcnt - 1;
1509
1510 __bio_for_each_segment(bv, bio, i, 0) {
1511 if (i == index) {
1512 if (offset > bv->bv_offset)
1513 sectors += (offset - bv->bv_offset) / sector_sz;
1514 break;
1515 }
1516
1517 sectors += bv->bv_len / sector_sz;
1518 }
1519
1520 return sectors;
1521}
1522EXPORT_SYMBOL(bio_sector_offset);
1da177e4
LT
1523
1524/*
1525 * create memory pools for biovec's in a bio_set.
1526 * use the global biovec slabs created for general use.
1527 */
5972511b 1528static int biovec_create_pools(struct bio_set *bs, int pool_entries)
1da177e4 1529{
7ff9345f 1530 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1da177e4 1531
7ff9345f
JA
1532 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1533 if (!bs->bvec_pool)
1534 return -ENOMEM;
1da177e4 1535
1da177e4
LT
1536 return 0;
1537}
1538
1539static void biovec_free_pools(struct bio_set *bs)
1540{
7ff9345f 1541 mempool_destroy(bs->bvec_pool);
1da177e4
LT
1542}
1543
1544void bioset_free(struct bio_set *bs)
1545{
1546 if (bs->bio_pool)
1547 mempool_destroy(bs->bio_pool);
1548
1549 biovec_free_pools(bs);
bb799ca0 1550 bio_put_slab(bs);
1da177e4
LT
1551
1552 kfree(bs);
1553}
1554
bb799ca0
JA
1555/**
1556 * bioset_create - Create a bio_set
1557 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1558 * @front_pad: Number of bytes to allocate in front of the returned bio
1559 *
1560 * Description:
1561 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1562 * to ask for a number of bytes to be allocated in front of the bio.
1563 * Front pad allocation is useful for embedding the bio inside
1564 * another structure, to avoid allocating extra data to go with the bio.
1565 * Note that the bio must be embedded at the END of that structure always,
1566 * or things will break badly.
1567 */
1568struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1da177e4 1569{
392ddc32 1570 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1b434498 1571 struct bio_set *bs;
1da177e4 1572
1b434498 1573 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1da177e4
LT
1574 if (!bs)
1575 return NULL;
1576
bb799ca0 1577 bs->front_pad = front_pad;
1b434498 1578
392ddc32 1579 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
bb799ca0
JA
1580 if (!bs->bio_slab) {
1581 kfree(bs);
1582 return NULL;
1583 }
1584
1585 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1da177e4
LT
1586 if (!bs->bio_pool)
1587 goto bad;
1588
bb799ca0 1589 if (!biovec_create_pools(bs, pool_size))
1da177e4
LT
1590 return bs;
1591
1592bad:
1593 bioset_free(bs);
1594 return NULL;
1595}
1596
1597static void __init biovec_init_slabs(void)
1598{
1599 int i;
1600
1601 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1602 int size;
1603 struct biovec_slab *bvs = bvec_slabs + i;
1604
a7fcd37c
JA
1605#ifndef CONFIG_BLK_DEV_INTEGRITY
1606 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1607 bvs->slab = NULL;
1608 continue;
1609 }
1610#endif
1611
1da177e4
LT
1612 size = bvs->nr_vecs * sizeof(struct bio_vec);
1613 bvs->slab = kmem_cache_create(bvs->name, size, 0,
20c2df83 1614 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1da177e4
LT
1615 }
1616}
1617
1618static int __init init_bio(void)
1619{
bb799ca0
JA
1620 bio_slab_max = 2;
1621 bio_slab_nr = 0;
1622 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1623 if (!bio_slabs)
1624 panic("bio: can't allocate bios\n");
1da177e4
LT
1625
1626 biovec_init_slabs();
1627
bb799ca0 1628 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
1da177e4
LT
1629 if (!fs_bio_set)
1630 panic("bio: can't allocate bios\n");
1631
0eaae62a
MD
1632 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1633 sizeof(struct bio_pair));
1da177e4
LT
1634 if (!bio_split_pool)
1635 panic("bio: can't create split pool\n");
1636
1637 return 0;
1638}
1639
1640subsys_initcall(init_bio);
1641
1642EXPORT_SYMBOL(bio_alloc);
0a0d96b0 1643EXPORT_SYMBOL(bio_kmalloc);
1da177e4 1644EXPORT_SYMBOL(bio_put);
3676347a 1645EXPORT_SYMBOL(bio_free);
1da177e4
LT
1646EXPORT_SYMBOL(bio_endio);
1647EXPORT_SYMBOL(bio_init);
1648EXPORT_SYMBOL(__bio_clone);
1649EXPORT_SYMBOL(bio_clone);
1650EXPORT_SYMBOL(bio_phys_segments);
1da177e4 1651EXPORT_SYMBOL(bio_add_page);
6e68af66 1652EXPORT_SYMBOL(bio_add_pc_page);
1da177e4 1653EXPORT_SYMBOL(bio_get_nr_vecs);
40044ce0
JA
1654EXPORT_SYMBOL(bio_map_user);
1655EXPORT_SYMBOL(bio_unmap_user);
df46b9a4 1656EXPORT_SYMBOL(bio_map_kern);
68154e90 1657EXPORT_SYMBOL(bio_copy_kern);
1da177e4
LT
1658EXPORT_SYMBOL(bio_pair_release);
1659EXPORT_SYMBOL(bio_split);
1da177e4
LT
1660EXPORT_SYMBOL(bio_copy_user);
1661EXPORT_SYMBOL(bio_uncopy_user);
1662EXPORT_SYMBOL(bioset_create);
1663EXPORT_SYMBOL(bioset_free);
1664EXPORT_SYMBOL(bio_alloc_bioset);