]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - fs/bio.c
block: prevent possible io_context->refcount overflow
[mirror_ubuntu-artful-kernel.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 176
7ff9345f
JA
177 /*
178 * see comment near bvec_array define!
179 */
180 switch (nr) {
181 case 1:
182 *idx = 0;
183 break;
184 case 2 ... 4:
185 *idx = 1;
186 break;
187 case 5 ... 16:
188 *idx = 2;
189 break;
190 case 17 ... 64:
191 *idx = 3;
192 break;
193 case 65 ... 128:
194 *idx = 4;
195 break;
196 case 129 ... BIO_MAX_PAGES:
197 *idx = 5;
198 break;
199 default:
200 return NULL;
201 }
202
203 /*
204 * idx now points to the pool we want to allocate from. only the
205 * 1-vec entry pool is mempool backed.
206 */
207 if (*idx == BIOVEC_MAX_IDX) {
208fallback:
209 bvl = mempool_alloc(bs->bvec_pool, gfp_mask);
210 } else {
211 struct biovec_slab *bvs = bvec_slabs + *idx;
212 gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
213
0a0d96b0 214 /*
7ff9345f
JA
215 * Make this allocation restricted and don't dump info on
216 * allocation failures, since we'll fallback to the mempool
217 * in case of failure.
0a0d96b0 218 */
7ff9345f 219 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
1da177e4 220
0a0d96b0 221 /*
7ff9345f
JA
222 * Try a slab allocation. If this fails and __GFP_WAIT
223 * is set, retry with the 1-entry mempool
0a0d96b0 224 */
7ff9345f
JA
225 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
226 if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
227 *idx = BIOVEC_MAX_IDX;
228 goto fallback;
229 }
230 }
231
1da177e4
LT
232 return bvl;
233}
234
7ff9345f 235void bio_free(struct bio *bio, struct bio_set *bs)
1da177e4 236{
bb799ca0 237 void *p;
1da177e4 238
392ddc32 239 if (bio_has_allocated_vec(bio))
bb799ca0 240 bvec_free_bs(bs, bio->bi_io_vec, BIO_POOL_IDX(bio));
1da177e4 241
7ba1ba12 242 if (bio_integrity(bio))
6d2a78e7 243 bio_integrity_free(bio);
7ba1ba12 244
bb799ca0
JA
245 /*
246 * If we have front padding, adjust the bio pointer before freeing
247 */
248 p = bio;
249 if (bs->front_pad)
250 p -= bs->front_pad;
251
252 mempool_free(p, bs->bio_pool);
3676347a
PO
253}
254
858119e1 255void bio_init(struct bio *bio)
1da177e4 256{
2b94de55 257 memset(bio, 0, sizeof(*bio));
1da177e4 258 bio->bi_flags = 1 << BIO_UPTODATE;
c7c22e4d 259 bio->bi_comp_cpu = -1;
1da177e4 260 atomic_set(&bio->bi_cnt, 1);
1da177e4
LT
261}
262
263/**
264 * bio_alloc_bioset - allocate a bio for I/O
265 * @gfp_mask: the GFP_ mask given to the slab allocator
266 * @nr_iovecs: number of iovecs to pre-allocate
0a0d96b0 267 * @bs: the bio_set to allocate from. If %NULL, just use kmalloc
1da177e4
LT
268 *
269 * Description:
0a0d96b0 270 * bio_alloc_bioset will first try its own mempool to satisfy the allocation.
1da177e4 271 * If %__GFP_WAIT is set then we will block on the internal pool waiting
0a0d96b0
JA
272 * for a &struct bio to become free. If a %NULL @bs is passed in, we will
273 * fall back to just using @kmalloc to allocate the required memory.
1da177e4 274 *
bb799ca0
JA
275 * Note that the caller must set ->bi_destructor on succesful return
276 * of a bio, to do the appropriate freeing of the bio once the reference
277 * count drops to zero.
1da177e4 278 **/
dd0fc66f 279struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
1da177e4 280{
451a9ebf 281 unsigned long idx = BIO_POOL_NONE;
34053979 282 struct bio_vec *bvl = NULL;
451a9ebf
TH
283 struct bio *bio;
284 void *p;
285
286 p = mempool_alloc(bs->bio_pool, gfp_mask);
287 if (unlikely(!p))
288 return NULL;
289 bio = p + bs->front_pad;
1da177e4 290
34053979
IM
291 bio_init(bio);
292
293 if (unlikely(!nr_iovecs))
294 goto out_set;
295
296 if (nr_iovecs <= BIO_INLINE_VECS) {
297 bvl = bio->bi_inline_vecs;
298 nr_iovecs = BIO_INLINE_VECS;
299 } else {
300 bvl = bvec_alloc_bs(gfp_mask, nr_iovecs, &idx, bs);
301 if (unlikely(!bvl))
302 goto err_free;
303
304 nr_iovecs = bvec_nr_vecs(idx);
1da177e4 305 }
451a9ebf 306out_set:
34053979
IM
307 bio->bi_flags |= idx << BIO_POOL_OFFSET;
308 bio->bi_max_vecs = nr_iovecs;
34053979 309 bio->bi_io_vec = bvl;
1da177e4 310 return bio;
34053979
IM
311
312err_free:
451a9ebf 313 mempool_free(p, bs->bio_pool);
34053979 314 return NULL;
1da177e4
LT
315}
316
451a9ebf
TH
317static void bio_fs_destructor(struct bio *bio)
318{
319 bio_free(bio, fs_bio_set);
320}
321
322/**
323 * bio_alloc - allocate a new bio, memory pool backed
324 * @gfp_mask: allocation mask to use
325 * @nr_iovecs: number of iovecs
326 *
327 * Allocate a new bio with @nr_iovecs bvecs. If @gfp_mask
328 * contains __GFP_WAIT, the allocation is guaranteed to succeed.
329 *
330 * RETURNS:
331 * Pointer to new bio on success, NULL on failure.
332 */
333struct bio *bio_alloc(gfp_t gfp_mask, int nr_iovecs)
334{
335 struct bio *bio = bio_alloc_bioset(gfp_mask, nr_iovecs, fs_bio_set);
336
337 if (bio)
338 bio->bi_destructor = bio_fs_destructor;
339
340 return bio;
341}
342
343static void bio_kmalloc_destructor(struct bio *bio)
344{
345 if (bio_integrity(bio))
346 bio_integrity_free(bio);
347 kfree(bio);
348}
349
86c824b9
JA
350/**
351 * bio_alloc - allocate a bio for I/O
352 * @gfp_mask: the GFP_ mask given to the slab allocator
353 * @nr_iovecs: number of iovecs to pre-allocate
354 *
355 * Description:
356 * bio_alloc will allocate a bio and associated bio_vec array that can hold
357 * at least @nr_iovecs entries. Allocations will be done from the
358 * fs_bio_set. Also see @bio_alloc_bioset.
359 *
360 * If %__GFP_WAIT is set, then bio_alloc will always be able to allocate
361 * a bio. This is due to the mempool guarantees. To make this work, callers
362 * must never allocate more than 1 bio at the time from this pool. Callers
363 * that need to allocate more than 1 bio must always submit the previously
364 * allocate bio for IO before attempting to allocate a new one. Failure to
365 * do so can cause livelocks under memory pressure.
366 *
367 **/
0a0d96b0
JA
368struct bio *bio_kmalloc(gfp_t gfp_mask, int nr_iovecs)
369{
451a9ebf 370 struct bio *bio;
0a0d96b0 371
451a9ebf
TH
372 bio = kmalloc(sizeof(struct bio) + nr_iovecs * sizeof(struct bio_vec),
373 gfp_mask);
374 if (unlikely(!bio))
375 return NULL;
376
377 bio_init(bio);
378 bio->bi_flags |= BIO_POOL_NONE << BIO_POOL_OFFSET;
379 bio->bi_max_vecs = nr_iovecs;
380 bio->bi_io_vec = bio->bi_inline_vecs;
381 bio->bi_destructor = bio_kmalloc_destructor;
0a0d96b0
JA
382
383 return bio;
384}
385
1da177e4
LT
386void zero_fill_bio(struct bio *bio)
387{
388 unsigned long flags;
389 struct bio_vec *bv;
390 int i;
391
392 bio_for_each_segment(bv, bio, i) {
393 char *data = bvec_kmap_irq(bv, &flags);
394 memset(data, 0, bv->bv_len);
395 flush_dcache_page(bv->bv_page);
396 bvec_kunmap_irq(data, &flags);
397 }
398}
399EXPORT_SYMBOL(zero_fill_bio);
400
401/**
402 * bio_put - release a reference to a bio
403 * @bio: bio to release reference to
404 *
405 * Description:
406 * Put a reference to a &struct bio, either one you have gotten with
407 * bio_alloc or bio_get. The last put of a bio will free it.
408 **/
409void bio_put(struct bio *bio)
410{
411 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
412
413 /*
414 * last put frees it
415 */
416 if (atomic_dec_and_test(&bio->bi_cnt)) {
417 bio->bi_next = NULL;
418 bio->bi_destructor(bio);
419 }
420}
421
165125e1 422inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
1da177e4
LT
423{
424 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
425 blk_recount_segments(q, bio);
426
427 return bio->bi_phys_segments;
428}
429
1da177e4
LT
430/**
431 * __bio_clone - clone a bio
432 * @bio: destination bio
433 * @bio_src: bio to clone
434 *
435 * Clone a &bio. Caller will own the returned bio, but not
436 * the actual data it points to. Reference count of returned
437 * bio will be one.
438 */
858119e1 439void __bio_clone(struct bio *bio, struct bio *bio_src)
1da177e4 440{
e525e153
AM
441 memcpy(bio->bi_io_vec, bio_src->bi_io_vec,
442 bio_src->bi_max_vecs * sizeof(struct bio_vec));
1da177e4 443
5d84070e
JA
444 /*
445 * most users will be overriding ->bi_bdev with a new target,
446 * so we don't set nor calculate new physical/hw segment counts here
447 */
1da177e4
LT
448 bio->bi_sector = bio_src->bi_sector;
449 bio->bi_bdev = bio_src->bi_bdev;
450 bio->bi_flags |= 1 << BIO_CLONED;
451 bio->bi_rw = bio_src->bi_rw;
1da177e4
LT
452 bio->bi_vcnt = bio_src->bi_vcnt;
453 bio->bi_size = bio_src->bi_size;
a5453be4 454 bio->bi_idx = bio_src->bi_idx;
1da177e4
LT
455}
456
457/**
458 * bio_clone - clone a bio
459 * @bio: bio to clone
460 * @gfp_mask: allocation priority
461 *
462 * Like __bio_clone, only also allocates the returned bio
463 */
dd0fc66f 464struct bio *bio_clone(struct bio *bio, gfp_t gfp_mask)
1da177e4
LT
465{
466 struct bio *b = bio_alloc_bioset(gfp_mask, bio->bi_max_vecs, fs_bio_set);
467
7ba1ba12
MP
468 if (!b)
469 return NULL;
470
471 b->bi_destructor = bio_fs_destructor;
472 __bio_clone(b, bio);
473
474 if (bio_integrity(bio)) {
475 int ret;
476
6d2a78e7 477 ret = bio_integrity_clone(b, bio, gfp_mask);
7ba1ba12 478
059ea331
LZ
479 if (ret < 0) {
480 bio_put(b);
7ba1ba12 481 return NULL;
059ea331 482 }
3676347a 483 }
1da177e4
LT
484
485 return b;
486}
487
488/**
489 * bio_get_nr_vecs - return approx number of vecs
490 * @bdev: I/O target
491 *
492 * Return the approximate number of pages we can send to this target.
493 * There's no guarantee that you will be able to fit this number of pages
494 * into a bio, it does not account for dynamic restrictions that vary
495 * on offset.
496 */
497int bio_get_nr_vecs(struct block_device *bdev)
498{
165125e1 499 struct request_queue *q = bdev_get_queue(bdev);
1da177e4
LT
500 int nr_pages;
501
ae03bf63
MP
502 nr_pages = ((queue_max_sectors(q) << 9) + PAGE_SIZE - 1) >> PAGE_SHIFT;
503 if (nr_pages > queue_max_phys_segments(q))
504 nr_pages = queue_max_phys_segments(q);
505 if (nr_pages > queue_max_hw_segments(q))
506 nr_pages = queue_max_hw_segments(q);
1da177e4
LT
507
508 return nr_pages;
509}
510
165125e1 511static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
defd94b7
MC
512 *page, unsigned int len, unsigned int offset,
513 unsigned short max_sectors)
1da177e4
LT
514{
515 int retried_segments = 0;
516 struct bio_vec *bvec;
517
518 /*
519 * cloned bio must not modify vec list
520 */
521 if (unlikely(bio_flagged(bio, BIO_CLONED)))
522 return 0;
523
80cfd548 524 if (((bio->bi_size + len) >> 9) > max_sectors)
1da177e4
LT
525 return 0;
526
80cfd548
JA
527 /*
528 * For filesystems with a blocksize smaller than the pagesize
529 * we will often be called with the same page as last time and
530 * a consecutive offset. Optimize this special case.
531 */
532 if (bio->bi_vcnt > 0) {
533 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
534
535 if (page == prev->bv_page &&
536 offset == prev->bv_offset + prev->bv_len) {
537 prev->bv_len += len;
cc371e66
AK
538
539 if (q->merge_bvec_fn) {
540 struct bvec_merge_data bvm = {
541 .bi_bdev = bio->bi_bdev,
542 .bi_sector = bio->bi_sector,
543 .bi_size = bio->bi_size,
544 .bi_rw = bio->bi_rw,
545 };
546
547 if (q->merge_bvec_fn(q, &bvm, prev) < len) {
548 prev->bv_len -= len;
549 return 0;
550 }
80cfd548
JA
551 }
552
553 goto done;
554 }
555 }
556
557 if (bio->bi_vcnt >= bio->bi_max_vecs)
1da177e4
LT
558 return 0;
559
560 /*
561 * we might lose a segment or two here, but rather that than
562 * make this too complex.
563 */
564
ae03bf63
MP
565 while (bio->bi_phys_segments >= queue_max_phys_segments(q)
566 || bio->bi_phys_segments >= queue_max_hw_segments(q)) {
1da177e4
LT
567
568 if (retried_segments)
569 return 0;
570
571 retried_segments = 1;
572 blk_recount_segments(q, bio);
573 }
574
575 /*
576 * setup the new entry, we might clear it again later if we
577 * cannot add the page
578 */
579 bvec = &bio->bi_io_vec[bio->bi_vcnt];
580 bvec->bv_page = page;
581 bvec->bv_len = len;
582 bvec->bv_offset = offset;
583
584 /*
585 * if queue has other restrictions (eg varying max sector size
586 * depending on offset), it can specify a merge_bvec_fn in the
587 * queue to get further control
588 */
589 if (q->merge_bvec_fn) {
cc371e66
AK
590 struct bvec_merge_data bvm = {
591 .bi_bdev = bio->bi_bdev,
592 .bi_sector = bio->bi_sector,
593 .bi_size = bio->bi_size,
594 .bi_rw = bio->bi_rw,
595 };
596
1da177e4
LT
597 /*
598 * merge_bvec_fn() returns number of bytes it can accept
599 * at this offset
600 */
cc371e66 601 if (q->merge_bvec_fn(q, &bvm, bvec) < len) {
1da177e4
LT
602 bvec->bv_page = NULL;
603 bvec->bv_len = 0;
604 bvec->bv_offset = 0;
605 return 0;
606 }
607 }
608
609 /* If we may be able to merge these biovecs, force a recount */
b8b3e16c 610 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
1da177e4
LT
611 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
612
613 bio->bi_vcnt++;
614 bio->bi_phys_segments++;
80cfd548 615 done:
1da177e4
LT
616 bio->bi_size += len;
617 return len;
618}
619
6e68af66
MC
620/**
621 * bio_add_pc_page - attempt to add page to bio
fddfdeaf 622 * @q: the target queue
6e68af66
MC
623 * @bio: destination bio
624 * @page: page to add
625 * @len: vec entry length
626 * @offset: vec entry offset
627 *
628 * Attempt to add a page to the bio_vec maplist. This can fail for a
629 * number of reasons, such as the bio being full or target block
630 * device limitations. The target block device must allow bio's
631 * smaller than PAGE_SIZE, so it is always possible to add a single
632 * page to an empty bio. This should only be used by REQ_PC bios.
633 */
165125e1 634int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
6e68af66
MC
635 unsigned int len, unsigned int offset)
636{
ae03bf63
MP
637 return __bio_add_page(q, bio, page, len, offset,
638 queue_max_hw_sectors(q));
6e68af66
MC
639}
640
1da177e4
LT
641/**
642 * bio_add_page - attempt to add page to bio
643 * @bio: destination bio
644 * @page: page to add
645 * @len: vec entry length
646 * @offset: vec entry offset
647 *
648 * Attempt to add a page to the bio_vec maplist. This can fail for a
649 * number of reasons, such as the bio being full or target block
650 * device limitations. The target block device must allow bio's
651 * smaller than PAGE_SIZE, so it is always possible to add a single
652 * page to an empty bio.
653 */
654int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
655 unsigned int offset)
656{
defd94b7 657 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
ae03bf63 658 return __bio_add_page(q, bio, page, len, offset, queue_max_sectors(q));
1da177e4
LT
659}
660
661struct bio_map_data {
662 struct bio_vec *iovecs;
c5dec1c3 663 struct sg_iovec *sgvecs;
152e283f
FT
664 int nr_sgvecs;
665 int is_our_pages;
1da177e4
LT
666};
667
c5dec1c3 668static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
152e283f
FT
669 struct sg_iovec *iov, int iov_count,
670 int is_our_pages)
1da177e4
LT
671{
672 memcpy(bmd->iovecs, bio->bi_io_vec, sizeof(struct bio_vec) * bio->bi_vcnt);
c5dec1c3
FT
673 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
674 bmd->nr_sgvecs = iov_count;
152e283f 675 bmd->is_our_pages = is_our_pages;
1da177e4
LT
676 bio->bi_private = bmd;
677}
678
679static void bio_free_map_data(struct bio_map_data *bmd)
680{
681 kfree(bmd->iovecs);
c5dec1c3 682 kfree(bmd->sgvecs);
1da177e4
LT
683 kfree(bmd);
684}
685
76029ff3
FT
686static struct bio_map_data *bio_alloc_map_data(int nr_segs, int iov_count,
687 gfp_t gfp_mask)
1da177e4 688{
76029ff3 689 struct bio_map_data *bmd = kmalloc(sizeof(*bmd), gfp_mask);
1da177e4
LT
690
691 if (!bmd)
692 return NULL;
693
76029ff3 694 bmd->iovecs = kmalloc(sizeof(struct bio_vec) * nr_segs, gfp_mask);
c5dec1c3
FT
695 if (!bmd->iovecs) {
696 kfree(bmd);
697 return NULL;
698 }
699
76029ff3 700 bmd->sgvecs = kmalloc(sizeof(struct sg_iovec) * iov_count, gfp_mask);
c5dec1c3 701 if (bmd->sgvecs)
1da177e4
LT
702 return bmd;
703
c5dec1c3 704 kfree(bmd->iovecs);
1da177e4
LT
705 kfree(bmd);
706 return NULL;
707}
708
aefcc28a 709static int __bio_copy_iov(struct bio *bio, struct bio_vec *iovecs,
152e283f
FT
710 struct sg_iovec *iov, int iov_count, int uncopy,
711 int do_free_page)
c5dec1c3
FT
712{
713 int ret = 0, i;
714 struct bio_vec *bvec;
715 int iov_idx = 0;
716 unsigned int iov_off = 0;
717 int read = bio_data_dir(bio) == READ;
718
719 __bio_for_each_segment(bvec, bio, i, 0) {
720 char *bv_addr = page_address(bvec->bv_page);
aefcc28a 721 unsigned int bv_len = iovecs[i].bv_len;
c5dec1c3
FT
722
723 while (bv_len && iov_idx < iov_count) {
724 unsigned int bytes;
725 char *iov_addr;
726
727 bytes = min_t(unsigned int,
728 iov[iov_idx].iov_len - iov_off, bv_len);
729 iov_addr = iov[iov_idx].iov_base + iov_off;
730
731 if (!ret) {
732 if (!read && !uncopy)
733 ret = copy_from_user(bv_addr, iov_addr,
734 bytes);
735 if (read && uncopy)
736 ret = copy_to_user(iov_addr, bv_addr,
737 bytes);
738
739 if (ret)
740 ret = -EFAULT;
741 }
742
743 bv_len -= bytes;
744 bv_addr += bytes;
745 iov_addr += bytes;
746 iov_off += bytes;
747
748 if (iov[iov_idx].iov_len == iov_off) {
749 iov_idx++;
750 iov_off = 0;
751 }
752 }
753
152e283f 754 if (do_free_page)
c5dec1c3
FT
755 __free_page(bvec->bv_page);
756 }
757
758 return ret;
759}
760
1da177e4
LT
761/**
762 * bio_uncopy_user - finish previously mapped bio
763 * @bio: bio being terminated
764 *
765 * Free pages allocated from bio_copy_user() and write back data
766 * to user space in case of a read.
767 */
768int bio_uncopy_user(struct bio *bio)
769{
770 struct bio_map_data *bmd = bio->bi_private;
81882766 771 int ret = 0;
1da177e4 772
81882766
FT
773 if (!bio_flagged(bio, BIO_NULL_MAPPED))
774 ret = __bio_copy_iov(bio, bmd->iovecs, bmd->sgvecs,
775 bmd->nr_sgvecs, 1, bmd->is_our_pages);
1da177e4
LT
776 bio_free_map_data(bmd);
777 bio_put(bio);
778 return ret;
779}
780
781/**
c5dec1c3 782 * bio_copy_user_iov - copy user data to bio
1da177e4 783 * @q: destination block queue
152e283f 784 * @map_data: pointer to the rq_map_data holding pages (if necessary)
c5dec1c3
FT
785 * @iov: the iovec.
786 * @iov_count: number of elements in the iovec
1da177e4 787 * @write_to_vm: bool indicating writing to pages or not
a3bce90e 788 * @gfp_mask: memory allocation flags
1da177e4
LT
789 *
790 * Prepares and returns a bio for indirect user io, bouncing data
791 * to/from kernel pages as necessary. Must be paired with
792 * call bio_uncopy_user() on io completion.
793 */
152e283f
FT
794struct bio *bio_copy_user_iov(struct request_queue *q,
795 struct rq_map_data *map_data,
796 struct sg_iovec *iov, int iov_count,
797 int write_to_vm, gfp_t gfp_mask)
1da177e4 798{
1da177e4
LT
799 struct bio_map_data *bmd;
800 struct bio_vec *bvec;
801 struct page *page;
802 struct bio *bio;
803 int i, ret;
c5dec1c3
FT
804 int nr_pages = 0;
805 unsigned int len = 0;
56c451f4 806 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
1da177e4 807
c5dec1c3
FT
808 for (i = 0; i < iov_count; i++) {
809 unsigned long uaddr;
810 unsigned long end;
811 unsigned long start;
812
813 uaddr = (unsigned long)iov[i].iov_base;
814 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
815 start = uaddr >> PAGE_SHIFT;
816
817 nr_pages += end - start;
818 len += iov[i].iov_len;
819 }
820
69838727
FT
821 if (offset)
822 nr_pages++;
823
a3bce90e 824 bmd = bio_alloc_map_data(nr_pages, iov_count, gfp_mask);
1da177e4
LT
825 if (!bmd)
826 return ERR_PTR(-ENOMEM);
827
1da177e4 828 ret = -ENOMEM;
a9e9dc24 829 bio = bio_kmalloc(gfp_mask, nr_pages);
1da177e4
LT
830 if (!bio)
831 goto out_bmd;
832
833 bio->bi_rw |= (!write_to_vm << BIO_RW);
834
835 ret = 0;
56c451f4
FT
836
837 if (map_data) {
e623ddb4 838 nr_pages = 1 << map_data->page_order;
56c451f4
FT
839 i = map_data->offset / PAGE_SIZE;
840 }
1da177e4 841 while (len) {
e623ddb4 842 unsigned int bytes = PAGE_SIZE;
1da177e4 843
56c451f4
FT
844 bytes -= offset;
845
1da177e4
LT
846 if (bytes > len)
847 bytes = len;
848
152e283f 849 if (map_data) {
e623ddb4 850 if (i == map_data->nr_entries * nr_pages) {
152e283f
FT
851 ret = -ENOMEM;
852 break;
853 }
e623ddb4
FT
854
855 page = map_data->pages[i / nr_pages];
856 page += (i % nr_pages);
857
858 i++;
859 } else {
152e283f 860 page = alloc_page(q->bounce_gfp | gfp_mask);
e623ddb4
FT
861 if (!page) {
862 ret = -ENOMEM;
863 break;
864 }
1da177e4
LT
865 }
866
56c451f4 867 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
1da177e4 868 break;
1da177e4
LT
869
870 len -= bytes;
56c451f4 871 offset = 0;
1da177e4
LT
872 }
873
874 if (ret)
875 goto cleanup;
876
877 /*
878 * success
879 */
97ae77a1 880 if (!write_to_vm && (!map_data || !map_data->null_mapped)) {
152e283f 881 ret = __bio_copy_iov(bio, bio->bi_io_vec, iov, iov_count, 0, 0);
c5dec1c3
FT
882 if (ret)
883 goto cleanup;
1da177e4
LT
884 }
885
152e283f 886 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
1da177e4
LT
887 return bio;
888cleanup:
152e283f
FT
889 if (!map_data)
890 bio_for_each_segment(bvec, bio, i)
891 __free_page(bvec->bv_page);
1da177e4
LT
892
893 bio_put(bio);
894out_bmd:
895 bio_free_map_data(bmd);
896 return ERR_PTR(ret);
897}
898
c5dec1c3
FT
899/**
900 * bio_copy_user - copy user data to bio
901 * @q: destination block queue
152e283f 902 * @map_data: pointer to the rq_map_data holding pages (if necessary)
c5dec1c3
FT
903 * @uaddr: start of user address
904 * @len: length in bytes
905 * @write_to_vm: bool indicating writing to pages or not
a3bce90e 906 * @gfp_mask: memory allocation flags
c5dec1c3
FT
907 *
908 * Prepares and returns a bio for indirect user io, bouncing data
909 * to/from kernel pages as necessary. Must be paired with
910 * call bio_uncopy_user() on io completion.
911 */
152e283f
FT
912struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
913 unsigned long uaddr, unsigned int len,
914 int write_to_vm, gfp_t gfp_mask)
c5dec1c3
FT
915{
916 struct sg_iovec iov;
917
918 iov.iov_base = (void __user *)uaddr;
919 iov.iov_len = len;
920
152e283f 921 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
c5dec1c3
FT
922}
923
165125e1 924static struct bio *__bio_map_user_iov(struct request_queue *q,
f1970baf
JB
925 struct block_device *bdev,
926 struct sg_iovec *iov, int iov_count,
a3bce90e 927 int write_to_vm, gfp_t gfp_mask)
1da177e4 928{
f1970baf
JB
929 int i, j;
930 int nr_pages = 0;
1da177e4
LT
931 struct page **pages;
932 struct bio *bio;
f1970baf
JB
933 int cur_page = 0;
934 int ret, offset;
1da177e4 935
f1970baf
JB
936 for (i = 0; i < iov_count; i++) {
937 unsigned long uaddr = (unsigned long)iov[i].iov_base;
938 unsigned long len = iov[i].iov_len;
939 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
940 unsigned long start = uaddr >> PAGE_SHIFT;
941
942 nr_pages += end - start;
943 /*
ad2d7225 944 * buffer must be aligned to at least hardsector size for now
f1970baf 945 */
ad2d7225 946 if (uaddr & queue_dma_alignment(q))
f1970baf
JB
947 return ERR_PTR(-EINVAL);
948 }
949
950 if (!nr_pages)
1da177e4
LT
951 return ERR_PTR(-EINVAL);
952
a9e9dc24 953 bio = bio_kmalloc(gfp_mask, nr_pages);
1da177e4
LT
954 if (!bio)
955 return ERR_PTR(-ENOMEM);
956
957 ret = -ENOMEM;
a3bce90e 958 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
1da177e4
LT
959 if (!pages)
960 goto out;
961
f1970baf
JB
962 for (i = 0; i < iov_count; i++) {
963 unsigned long uaddr = (unsigned long)iov[i].iov_base;
964 unsigned long len = iov[i].iov_len;
965 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
966 unsigned long start = uaddr >> PAGE_SHIFT;
967 const int local_nr_pages = end - start;
968 const int page_limit = cur_page + local_nr_pages;
969
f5dd33c4
NP
970 ret = get_user_pages_fast(uaddr, local_nr_pages,
971 write_to_vm, &pages[cur_page]);
99172157
JA
972 if (ret < local_nr_pages) {
973 ret = -EFAULT;
f1970baf 974 goto out_unmap;
99172157 975 }
f1970baf
JB
976
977 offset = uaddr & ~PAGE_MASK;
978 for (j = cur_page; j < page_limit; j++) {
979 unsigned int bytes = PAGE_SIZE - offset;
980
981 if (len <= 0)
982 break;
983
984 if (bytes > len)
985 bytes = len;
986
987 /*
988 * sorry...
989 */
defd94b7
MC
990 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
991 bytes)
f1970baf
JB
992 break;
993
994 len -= bytes;
995 offset = 0;
996 }
1da177e4 997
f1970baf 998 cur_page = j;
1da177e4 999 /*
f1970baf 1000 * release the pages we didn't map into the bio, if any
1da177e4 1001 */
f1970baf
JB
1002 while (j < page_limit)
1003 page_cache_release(pages[j++]);
1da177e4
LT
1004 }
1005
1da177e4
LT
1006 kfree(pages);
1007
1008 /*
1009 * set data direction, and check if mapped pages need bouncing
1010 */
1011 if (!write_to_vm)
1012 bio->bi_rw |= (1 << BIO_RW);
1013
f1970baf 1014 bio->bi_bdev = bdev;
1da177e4
LT
1015 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1016 return bio;
f1970baf
JB
1017
1018 out_unmap:
1019 for (i = 0; i < nr_pages; i++) {
1020 if(!pages[i])
1021 break;
1022 page_cache_release(pages[i]);
1023 }
1024 out:
1da177e4
LT
1025 kfree(pages);
1026 bio_put(bio);
1027 return ERR_PTR(ret);
1028}
1029
1030/**
1031 * bio_map_user - map user address into bio
165125e1 1032 * @q: the struct request_queue for the bio
1da177e4
LT
1033 * @bdev: destination block device
1034 * @uaddr: start of user address
1035 * @len: length in bytes
1036 * @write_to_vm: bool indicating writing to pages or not
a3bce90e 1037 * @gfp_mask: memory allocation flags
1da177e4
LT
1038 *
1039 * Map the user space address into a bio suitable for io to a block
1040 * device. Returns an error pointer in case of error.
1041 */
165125e1 1042struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
a3bce90e
FT
1043 unsigned long uaddr, unsigned int len, int write_to_vm,
1044 gfp_t gfp_mask)
f1970baf
JB
1045{
1046 struct sg_iovec iov;
1047
3f70353e 1048 iov.iov_base = (void __user *)uaddr;
f1970baf
JB
1049 iov.iov_len = len;
1050
a3bce90e 1051 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
f1970baf
JB
1052}
1053
1054/**
1055 * bio_map_user_iov - map user sg_iovec table into bio
165125e1 1056 * @q: the struct request_queue for the bio
f1970baf
JB
1057 * @bdev: destination block device
1058 * @iov: the iovec.
1059 * @iov_count: number of elements in the iovec
1060 * @write_to_vm: bool indicating writing to pages or not
a3bce90e 1061 * @gfp_mask: memory allocation flags
f1970baf
JB
1062 *
1063 * Map the user space address into a bio suitable for io to a block
1064 * device. Returns an error pointer in case of error.
1065 */
165125e1 1066struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
f1970baf 1067 struct sg_iovec *iov, int iov_count,
a3bce90e 1068 int write_to_vm, gfp_t gfp_mask)
1da177e4
LT
1069{
1070 struct bio *bio;
1071
a3bce90e
FT
1072 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1073 gfp_mask);
1da177e4
LT
1074 if (IS_ERR(bio))
1075 return bio;
1076
1077 /*
1078 * subtle -- if __bio_map_user() ended up bouncing a bio,
1079 * it would normally disappear when its bi_end_io is run.
1080 * however, we need it for the unmap, so grab an extra
1081 * reference to it
1082 */
1083 bio_get(bio);
1084
0e75f906 1085 return bio;
1da177e4
LT
1086}
1087
1088static void __bio_unmap_user(struct bio *bio)
1089{
1090 struct bio_vec *bvec;
1091 int i;
1092
1093 /*
1094 * make sure we dirty pages we wrote to
1095 */
1096 __bio_for_each_segment(bvec, bio, i, 0) {
1097 if (bio_data_dir(bio) == READ)
1098 set_page_dirty_lock(bvec->bv_page);
1099
1100 page_cache_release(bvec->bv_page);
1101 }
1102
1103 bio_put(bio);
1104}
1105
1106/**
1107 * bio_unmap_user - unmap a bio
1108 * @bio: the bio being unmapped
1109 *
1110 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1111 * a process context.
1112 *
1113 * bio_unmap_user() may sleep.
1114 */
1115void bio_unmap_user(struct bio *bio)
1116{
1117 __bio_unmap_user(bio);
1118 bio_put(bio);
1119}
1120
6712ecf8 1121static void bio_map_kern_endio(struct bio *bio, int err)
b823825e 1122{
b823825e 1123 bio_put(bio);
b823825e
JA
1124}
1125
1126
165125e1 1127static struct bio *__bio_map_kern(struct request_queue *q, void *data,
27496a8c 1128 unsigned int len, gfp_t gfp_mask)
df46b9a4
MC
1129{
1130 unsigned long kaddr = (unsigned long)data;
1131 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1132 unsigned long start = kaddr >> PAGE_SHIFT;
1133 const int nr_pages = end - start;
1134 int offset, i;
1135 struct bio *bio;
1136
a9e9dc24 1137 bio = bio_kmalloc(gfp_mask, nr_pages);
df46b9a4
MC
1138 if (!bio)
1139 return ERR_PTR(-ENOMEM);
1140
1141 offset = offset_in_page(kaddr);
1142 for (i = 0; i < nr_pages; i++) {
1143 unsigned int bytes = PAGE_SIZE - offset;
1144
1145 if (len <= 0)
1146 break;
1147
1148 if (bytes > len)
1149 bytes = len;
1150
defd94b7
MC
1151 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1152 offset) < bytes)
df46b9a4
MC
1153 break;
1154
1155 data += bytes;
1156 len -= bytes;
1157 offset = 0;
1158 }
1159
b823825e 1160 bio->bi_end_io = bio_map_kern_endio;
df46b9a4
MC
1161 return bio;
1162}
1163
1164/**
1165 * bio_map_kern - map kernel address into bio
165125e1 1166 * @q: the struct request_queue for the bio
df46b9a4
MC
1167 * @data: pointer to buffer to map
1168 * @len: length in bytes
1169 * @gfp_mask: allocation flags for bio allocation
1170 *
1171 * Map the kernel address into a bio suitable for io to a block
1172 * device. Returns an error pointer in case of error.
1173 */
165125e1 1174struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
27496a8c 1175 gfp_t gfp_mask)
df46b9a4
MC
1176{
1177 struct bio *bio;
1178
1179 bio = __bio_map_kern(q, data, len, gfp_mask);
1180 if (IS_ERR(bio))
1181 return bio;
1182
1183 if (bio->bi_size == len)
1184 return bio;
1185
1186 /*
1187 * Don't support partial mappings.
1188 */
1189 bio_put(bio);
1190 return ERR_PTR(-EINVAL);
1191}
1192
68154e90
FT
1193static void bio_copy_kern_endio(struct bio *bio, int err)
1194{
1195 struct bio_vec *bvec;
1196 const int read = bio_data_dir(bio) == READ;
76029ff3 1197 struct bio_map_data *bmd = bio->bi_private;
68154e90 1198 int i;
76029ff3 1199 char *p = bmd->sgvecs[0].iov_base;
68154e90
FT
1200
1201 __bio_for_each_segment(bvec, bio, i, 0) {
1202 char *addr = page_address(bvec->bv_page);
76029ff3 1203 int len = bmd->iovecs[i].bv_len;
68154e90 1204
4fc981ef 1205 if (read)
76029ff3 1206 memcpy(p, addr, len);
68154e90
FT
1207
1208 __free_page(bvec->bv_page);
76029ff3 1209 p += len;
68154e90
FT
1210 }
1211
76029ff3 1212 bio_free_map_data(bmd);
68154e90
FT
1213 bio_put(bio);
1214}
1215
1216/**
1217 * bio_copy_kern - copy kernel address into bio
1218 * @q: the struct request_queue for the bio
1219 * @data: pointer to buffer to copy
1220 * @len: length in bytes
1221 * @gfp_mask: allocation flags for bio and page allocation
ffee0259 1222 * @reading: data direction is READ
68154e90
FT
1223 *
1224 * copy the kernel address into a bio suitable for io to a block
1225 * device. Returns an error pointer in case of error.
1226 */
1227struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1228 gfp_t gfp_mask, int reading)
1229{
68154e90
FT
1230 struct bio *bio;
1231 struct bio_vec *bvec;
4d8ab62e 1232 int i;
68154e90 1233
4d8ab62e
FT
1234 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1235 if (IS_ERR(bio))
1236 return bio;
68154e90
FT
1237
1238 if (!reading) {
1239 void *p = data;
1240
1241 bio_for_each_segment(bvec, bio, i) {
1242 char *addr = page_address(bvec->bv_page);
1243
1244 memcpy(addr, p, bvec->bv_len);
1245 p += bvec->bv_len;
1246 }
1247 }
1248
68154e90 1249 bio->bi_end_io = bio_copy_kern_endio;
76029ff3 1250
68154e90 1251 return bio;
68154e90
FT
1252}
1253
1da177e4
LT
1254/*
1255 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1256 * for performing direct-IO in BIOs.
1257 *
1258 * The problem is that we cannot run set_page_dirty() from interrupt context
1259 * because the required locks are not interrupt-safe. So what we can do is to
1260 * mark the pages dirty _before_ performing IO. And in interrupt context,
1261 * check that the pages are still dirty. If so, fine. If not, redirty them
1262 * in process context.
1263 *
1264 * We special-case compound pages here: normally this means reads into hugetlb
1265 * pages. The logic in here doesn't really work right for compound pages
1266 * because the VM does not uniformly chase down the head page in all cases.
1267 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1268 * handle them at all. So we skip compound pages here at an early stage.
1269 *
1270 * Note that this code is very hard to test under normal circumstances because
1271 * direct-io pins the pages with get_user_pages(). This makes
1272 * is_page_cache_freeable return false, and the VM will not clean the pages.
1273 * But other code (eg, pdflush) could clean the pages if they are mapped
1274 * pagecache.
1275 *
1276 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1277 * deferred bio dirtying paths.
1278 */
1279
1280/*
1281 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1282 */
1283void bio_set_pages_dirty(struct bio *bio)
1284{
1285 struct bio_vec *bvec = bio->bi_io_vec;
1286 int i;
1287
1288 for (i = 0; i < bio->bi_vcnt; i++) {
1289 struct page *page = bvec[i].bv_page;
1290
1291 if (page && !PageCompound(page))
1292 set_page_dirty_lock(page);
1293 }
1294}
1295
86b6c7a7 1296static void bio_release_pages(struct bio *bio)
1da177e4
LT
1297{
1298 struct bio_vec *bvec = bio->bi_io_vec;
1299 int i;
1300
1301 for (i = 0; i < bio->bi_vcnt; i++) {
1302 struct page *page = bvec[i].bv_page;
1303
1304 if (page)
1305 put_page(page);
1306 }
1307}
1308
1309/*
1310 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1311 * If they are, then fine. If, however, some pages are clean then they must
1312 * have been written out during the direct-IO read. So we take another ref on
1313 * the BIO and the offending pages and re-dirty the pages in process context.
1314 *
1315 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1316 * here on. It will run one page_cache_release() against each page and will
1317 * run one bio_put() against the BIO.
1318 */
1319
65f27f38 1320static void bio_dirty_fn(struct work_struct *work);
1da177e4 1321
65f27f38 1322static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1da177e4
LT
1323static DEFINE_SPINLOCK(bio_dirty_lock);
1324static struct bio *bio_dirty_list;
1325
1326/*
1327 * This runs in process context
1328 */
65f27f38 1329static void bio_dirty_fn(struct work_struct *work)
1da177e4
LT
1330{
1331 unsigned long flags;
1332 struct bio *bio;
1333
1334 spin_lock_irqsave(&bio_dirty_lock, flags);
1335 bio = bio_dirty_list;
1336 bio_dirty_list = NULL;
1337 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1338
1339 while (bio) {
1340 struct bio *next = bio->bi_private;
1341
1342 bio_set_pages_dirty(bio);
1343 bio_release_pages(bio);
1344 bio_put(bio);
1345 bio = next;
1346 }
1347}
1348
1349void bio_check_pages_dirty(struct bio *bio)
1350{
1351 struct bio_vec *bvec = bio->bi_io_vec;
1352 int nr_clean_pages = 0;
1353 int i;
1354
1355 for (i = 0; i < bio->bi_vcnt; i++) {
1356 struct page *page = bvec[i].bv_page;
1357
1358 if (PageDirty(page) || PageCompound(page)) {
1359 page_cache_release(page);
1360 bvec[i].bv_page = NULL;
1361 } else {
1362 nr_clean_pages++;
1363 }
1364 }
1365
1366 if (nr_clean_pages) {
1367 unsigned long flags;
1368
1369 spin_lock_irqsave(&bio_dirty_lock, flags);
1370 bio->bi_private = bio_dirty_list;
1371 bio_dirty_list = bio;
1372 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1373 schedule_work(&bio_dirty_work);
1374 } else {
1375 bio_put(bio);
1376 }
1377}
1378
1379/**
1380 * bio_endio - end I/O on a bio
1381 * @bio: bio
1da177e4
LT
1382 * @error: error, if any
1383 *
1384 * Description:
6712ecf8 1385 * bio_endio() will end I/O on the whole bio. bio_endio() is the
5bb23a68
N
1386 * preferred way to end I/O on a bio, it takes care of clearing
1387 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1388 * established -Exxxx (-EIO, for instance) error values in case
1389 * something went wrong. Noone should call bi_end_io() directly on a
1390 * bio unless they own it and thus know that it has an end_io
1391 * function.
1da177e4 1392 **/
6712ecf8 1393void bio_endio(struct bio *bio, int error)
1da177e4
LT
1394{
1395 if (error)
1396 clear_bit(BIO_UPTODATE, &bio->bi_flags);
9cc54d40
N
1397 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1398 error = -EIO;
1da177e4 1399
5bb23a68 1400 if (bio->bi_end_io)
6712ecf8 1401 bio->bi_end_io(bio, error);
1da177e4
LT
1402}
1403
1404void bio_pair_release(struct bio_pair *bp)
1405{
1406 if (atomic_dec_and_test(&bp->cnt)) {
1407 struct bio *master = bp->bio1.bi_private;
1408
6712ecf8 1409 bio_endio(master, bp->error);
1da177e4
LT
1410 mempool_free(bp, bp->bio2.bi_private);
1411 }
1412}
1413
6712ecf8 1414static void bio_pair_end_1(struct bio *bi, int err)
1da177e4
LT
1415{
1416 struct bio_pair *bp = container_of(bi, struct bio_pair, bio1);
1417
1418 if (err)
1419 bp->error = err;
1420
1da177e4 1421 bio_pair_release(bp);
1da177e4
LT
1422}
1423
6712ecf8 1424static void bio_pair_end_2(struct bio *bi, int err)
1da177e4
LT
1425{
1426 struct bio_pair *bp = container_of(bi, struct bio_pair, bio2);
1427
1428 if (err)
1429 bp->error = err;
1430
1da177e4 1431 bio_pair_release(bp);
1da177e4
LT
1432}
1433
1434/*
c7eee1b8 1435 * split a bio - only worry about a bio with a single page in its iovec
1da177e4 1436 */
6feef531 1437struct bio_pair *bio_split(struct bio *bi, int first_sectors)
1da177e4 1438{
6feef531 1439 struct bio_pair *bp = mempool_alloc(bio_split_pool, GFP_NOIO);
1da177e4
LT
1440
1441 if (!bp)
1442 return bp;
1443
5f3ea37c 1444 trace_block_split(bdev_get_queue(bi->bi_bdev), bi,
2056a782
JA
1445 bi->bi_sector + first_sectors);
1446
1da177e4
LT
1447 BUG_ON(bi->bi_vcnt != 1);
1448 BUG_ON(bi->bi_idx != 0);
1449 atomic_set(&bp->cnt, 3);
1450 bp->error = 0;
1451 bp->bio1 = *bi;
1452 bp->bio2 = *bi;
1453 bp->bio2.bi_sector += first_sectors;
1454 bp->bio2.bi_size -= first_sectors << 9;
1455 bp->bio1.bi_size = first_sectors << 9;
1456
1457 bp->bv1 = bi->bi_io_vec[0];
1458 bp->bv2 = bi->bi_io_vec[0];
1459 bp->bv2.bv_offset += first_sectors << 9;
1460 bp->bv2.bv_len -= first_sectors << 9;
1461 bp->bv1.bv_len = first_sectors << 9;
1462
1463 bp->bio1.bi_io_vec = &bp->bv1;
1464 bp->bio2.bi_io_vec = &bp->bv2;
1465
a2eb0c10
N
1466 bp->bio1.bi_max_vecs = 1;
1467 bp->bio2.bi_max_vecs = 1;
1468
1da177e4
LT
1469 bp->bio1.bi_end_io = bio_pair_end_1;
1470 bp->bio2.bi_end_io = bio_pair_end_2;
1471
1472 bp->bio1.bi_private = bi;
6feef531 1473 bp->bio2.bi_private = bio_split_pool;
1da177e4 1474
7ba1ba12
MP
1475 if (bio_integrity(bi))
1476 bio_integrity_split(bi, bp, first_sectors);
1477
1da177e4
LT
1478 return bp;
1479}
1480
ad3316bf
MP
1481/**
1482 * bio_sector_offset - Find hardware sector offset in bio
1483 * @bio: bio to inspect
1484 * @index: bio_vec index
1485 * @offset: offset in bv_page
1486 *
1487 * Return the number of hardware sectors between beginning of bio
1488 * and an end point indicated by a bio_vec index and an offset
1489 * within that vector's page.
1490 */
1491sector_t bio_sector_offset(struct bio *bio, unsigned short index,
1492 unsigned int offset)
1493{
e1defc4f 1494 unsigned int sector_sz;
ad3316bf
MP
1495 struct bio_vec *bv;
1496 sector_t sectors;
1497 int i;
1498
e1defc4f 1499 sector_sz = queue_logical_block_size(bio->bi_bdev->bd_disk->queue);
ad3316bf
MP
1500 sectors = 0;
1501
1502 if (index >= bio->bi_idx)
1503 index = bio->bi_vcnt - 1;
1504
1505 __bio_for_each_segment(bv, bio, i, 0) {
1506 if (i == index) {
1507 if (offset > bv->bv_offset)
1508 sectors += (offset - bv->bv_offset) / sector_sz;
1509 break;
1510 }
1511
1512 sectors += bv->bv_len / sector_sz;
1513 }
1514
1515 return sectors;
1516}
1517EXPORT_SYMBOL(bio_sector_offset);
1da177e4
LT
1518
1519/*
1520 * create memory pools for biovec's in a bio_set.
1521 * use the global biovec slabs created for general use.
1522 */
5972511b 1523static int biovec_create_pools(struct bio_set *bs, int pool_entries)
1da177e4 1524{
7ff9345f 1525 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1da177e4 1526
7ff9345f
JA
1527 bs->bvec_pool = mempool_create_slab_pool(pool_entries, bp->slab);
1528 if (!bs->bvec_pool)
1529 return -ENOMEM;
1da177e4 1530
1da177e4
LT
1531 return 0;
1532}
1533
1534static void biovec_free_pools(struct bio_set *bs)
1535{
7ff9345f 1536 mempool_destroy(bs->bvec_pool);
1da177e4
LT
1537}
1538
1539void bioset_free(struct bio_set *bs)
1540{
1541 if (bs->bio_pool)
1542 mempool_destroy(bs->bio_pool);
1543
1544 biovec_free_pools(bs);
bb799ca0 1545 bio_put_slab(bs);
1da177e4
LT
1546
1547 kfree(bs);
1548}
1549
bb799ca0
JA
1550/**
1551 * bioset_create - Create a bio_set
1552 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1553 * @front_pad: Number of bytes to allocate in front of the returned bio
1554 *
1555 * Description:
1556 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1557 * to ask for a number of bytes to be allocated in front of the bio.
1558 * Front pad allocation is useful for embedding the bio inside
1559 * another structure, to avoid allocating extra data to go with the bio.
1560 * Note that the bio must be embedded at the END of that structure always,
1561 * or things will break badly.
1562 */
1563struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1da177e4 1564{
392ddc32 1565 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1b434498 1566 struct bio_set *bs;
1da177e4 1567
1b434498 1568 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1da177e4
LT
1569 if (!bs)
1570 return NULL;
1571
bb799ca0 1572 bs->front_pad = front_pad;
1b434498 1573
392ddc32 1574 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
bb799ca0
JA
1575 if (!bs->bio_slab) {
1576 kfree(bs);
1577 return NULL;
1578 }
1579
1580 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1da177e4
LT
1581 if (!bs->bio_pool)
1582 goto bad;
1583
bb799ca0 1584 if (!biovec_create_pools(bs, pool_size))
1da177e4
LT
1585 return bs;
1586
1587bad:
1588 bioset_free(bs);
1589 return NULL;
1590}
1591
1592static void __init biovec_init_slabs(void)
1593{
1594 int i;
1595
1596 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
1597 int size;
1598 struct biovec_slab *bvs = bvec_slabs + i;
1599
a7fcd37c
JA
1600#ifndef CONFIG_BLK_DEV_INTEGRITY
1601 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
1602 bvs->slab = NULL;
1603 continue;
1604 }
1605#endif
1606
1da177e4
LT
1607 size = bvs->nr_vecs * sizeof(struct bio_vec);
1608 bvs->slab = kmem_cache_create(bvs->name, size, 0,
20c2df83 1609 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1da177e4
LT
1610 }
1611}
1612
1613static int __init init_bio(void)
1614{
bb799ca0
JA
1615 bio_slab_max = 2;
1616 bio_slab_nr = 0;
1617 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
1618 if (!bio_slabs)
1619 panic("bio: can't allocate bios\n");
1da177e4
LT
1620
1621 biovec_init_slabs();
1622
bb799ca0 1623 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
1da177e4
LT
1624 if (!fs_bio_set)
1625 panic("bio: can't allocate bios\n");
1626
0eaae62a
MD
1627 bio_split_pool = mempool_create_kmalloc_pool(BIO_SPLIT_ENTRIES,
1628 sizeof(struct bio_pair));
1da177e4
LT
1629 if (!bio_split_pool)
1630 panic("bio: can't create split pool\n");
1631
1632 return 0;
1633}
1634
1635subsys_initcall(init_bio);
1636
1637EXPORT_SYMBOL(bio_alloc);
0a0d96b0 1638EXPORT_SYMBOL(bio_kmalloc);
1da177e4 1639EXPORT_SYMBOL(bio_put);
3676347a 1640EXPORT_SYMBOL(bio_free);
1da177e4
LT
1641EXPORT_SYMBOL(bio_endio);
1642EXPORT_SYMBOL(bio_init);
1643EXPORT_SYMBOL(__bio_clone);
1644EXPORT_SYMBOL(bio_clone);
1645EXPORT_SYMBOL(bio_phys_segments);
1da177e4 1646EXPORT_SYMBOL(bio_add_page);
6e68af66 1647EXPORT_SYMBOL(bio_add_pc_page);
1da177e4 1648EXPORT_SYMBOL(bio_get_nr_vecs);
40044ce0
JA
1649EXPORT_SYMBOL(bio_map_user);
1650EXPORT_SYMBOL(bio_unmap_user);
df46b9a4 1651EXPORT_SYMBOL(bio_map_kern);
68154e90 1652EXPORT_SYMBOL(bio_copy_kern);
1da177e4
LT
1653EXPORT_SYMBOL(bio_pair_release);
1654EXPORT_SYMBOL(bio_split);
1da177e4
LT
1655EXPORT_SYMBOL(bio_copy_user);
1656EXPORT_SYMBOL(bio_uncopy_user);
1657EXPORT_SYMBOL(bioset_create);
1658EXPORT_SYMBOL(bioset_free);
1659EXPORT_SYMBOL(bio_alloc_bioset);