]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - block/bio.c
Merge tag 'hwspinlock-3.17' of git://git.kernel.org/pub/scm/linux/kernel/git/ohad...
[mirror_ubuntu-zesty-kernel.git] / block / bio.c
1 /*
2 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
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/uio.h>
23 #include <linux/iocontext.h>
24 #include <linux/slab.h>
25 #include <linux/init.h>
26 #include <linux/kernel.h>
27 #include <linux/export.h>
28 #include <linux/mempool.h>
29 #include <linux/workqueue.h>
30 #include <linux/cgroup.h>
31 #include <scsi/sg.h> /* for struct sg_iovec */
32
33 #include <trace/events/block.h>
34
35 /*
36 * Test patch to inline a certain number of bi_io_vec's inside the bio
37 * itself, to shrink a bio data allocation from two mempool calls to one
38 */
39 #define BIO_INLINE_VECS 4
40
41 /*
42 * if you change this list, also change bvec_alloc or things will
43 * break badly! cannot be bigger than what you can fit into an
44 * unsigned short
45 */
46 #define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
47 static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
48 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
49 };
50 #undef BV
51
52 /*
53 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
54 * IO code that does not need private memory pools.
55 */
56 struct bio_set *fs_bio_set;
57 EXPORT_SYMBOL(fs_bio_set);
58
59 /*
60 * Our slab pool management
61 */
62 struct bio_slab {
63 struct kmem_cache *slab;
64 unsigned int slab_ref;
65 unsigned int slab_size;
66 char name[8];
67 };
68 static DEFINE_MUTEX(bio_slab_lock);
69 static struct bio_slab *bio_slabs;
70 static unsigned int bio_slab_nr, bio_slab_max;
71
72 static 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, *new_bio_slabs;
77 unsigned int new_bio_slab_max;
78 unsigned int i, entry = -1;
79
80 mutex_lock(&bio_slab_lock);
81
82 i = 0;
83 while (i < bio_slab_nr) {
84 bslab = &bio_slabs[i];
85
86 if (!bslab->slab && entry == -1)
87 entry = i;
88 else if (bslab->slab_size == sz) {
89 slab = bslab->slab;
90 bslab->slab_ref++;
91 break;
92 }
93 i++;
94 }
95
96 if (slab)
97 goto out_unlock;
98
99 if (bio_slab_nr == bio_slab_max && entry == -1) {
100 new_bio_slab_max = bio_slab_max << 1;
101 new_bio_slabs = krealloc(bio_slabs,
102 new_bio_slab_max * sizeof(struct bio_slab),
103 GFP_KERNEL);
104 if (!new_bio_slabs)
105 goto out_unlock;
106 bio_slab_max = new_bio_slab_max;
107 bio_slabs = new_bio_slabs;
108 }
109 if (entry == -1)
110 entry = bio_slab_nr++;
111
112 bslab = &bio_slabs[entry];
113
114 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
115 slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN,
116 SLAB_HWCACHE_ALIGN, NULL);
117 if (!slab)
118 goto out_unlock;
119
120 bslab->slab = slab;
121 bslab->slab_ref = 1;
122 bslab->slab_size = sz;
123 out_unlock:
124 mutex_unlock(&bio_slab_lock);
125 return slab;
126 }
127
128 static void bio_put_slab(struct bio_set *bs)
129 {
130 struct bio_slab *bslab = NULL;
131 unsigned int i;
132
133 mutex_lock(&bio_slab_lock);
134
135 for (i = 0; i < bio_slab_nr; i++) {
136 if (bs->bio_slab == bio_slabs[i].slab) {
137 bslab = &bio_slabs[i];
138 break;
139 }
140 }
141
142 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
143 goto out;
144
145 WARN_ON(!bslab->slab_ref);
146
147 if (--bslab->slab_ref)
148 goto out;
149
150 kmem_cache_destroy(bslab->slab);
151 bslab->slab = NULL;
152
153 out:
154 mutex_unlock(&bio_slab_lock);
155 }
156
157 unsigned int bvec_nr_vecs(unsigned short idx)
158 {
159 return bvec_slabs[idx].nr_vecs;
160 }
161
162 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx)
163 {
164 BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
165
166 if (idx == BIOVEC_MAX_IDX)
167 mempool_free(bv, pool);
168 else {
169 struct biovec_slab *bvs = bvec_slabs + idx;
170
171 kmem_cache_free(bvs->slab, bv);
172 }
173 }
174
175 struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx,
176 mempool_t *pool)
177 {
178 struct bio_vec *bvl;
179
180 /*
181 * see comment near bvec_array define!
182 */
183 switch (nr) {
184 case 1:
185 *idx = 0;
186 break;
187 case 2 ... 4:
188 *idx = 1;
189 break;
190 case 5 ... 16:
191 *idx = 2;
192 break;
193 case 17 ... 64:
194 *idx = 3;
195 break;
196 case 65 ... 128:
197 *idx = 4;
198 break;
199 case 129 ... BIO_MAX_PAGES:
200 *idx = 5;
201 break;
202 default:
203 return NULL;
204 }
205
206 /*
207 * idx now points to the pool we want to allocate from. only the
208 * 1-vec entry pool is mempool backed.
209 */
210 if (*idx == BIOVEC_MAX_IDX) {
211 fallback:
212 bvl = mempool_alloc(pool, gfp_mask);
213 } else {
214 struct biovec_slab *bvs = bvec_slabs + *idx;
215 gfp_t __gfp_mask = gfp_mask & ~(__GFP_WAIT | __GFP_IO);
216
217 /*
218 * Make this allocation restricted and don't dump info on
219 * allocation failures, since we'll fallback to the mempool
220 * in case of failure.
221 */
222 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
223
224 /*
225 * Try a slab allocation. If this fails and __GFP_WAIT
226 * is set, retry with the 1-entry mempool
227 */
228 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
229 if (unlikely(!bvl && (gfp_mask & __GFP_WAIT))) {
230 *idx = BIOVEC_MAX_IDX;
231 goto fallback;
232 }
233 }
234
235 return bvl;
236 }
237
238 static void __bio_free(struct bio *bio)
239 {
240 bio_disassociate_task(bio);
241
242 if (bio_integrity(bio))
243 bio_integrity_free(bio);
244 }
245
246 static void bio_free(struct bio *bio)
247 {
248 struct bio_set *bs = bio->bi_pool;
249 void *p;
250
251 __bio_free(bio);
252
253 if (bs) {
254 if (bio_flagged(bio, BIO_OWNS_VEC))
255 bvec_free(bs->bvec_pool, bio->bi_io_vec, BIO_POOL_IDX(bio));
256
257 /*
258 * If we have front padding, adjust the bio pointer before freeing
259 */
260 p = bio;
261 p -= bs->front_pad;
262
263 mempool_free(p, bs->bio_pool);
264 } else {
265 /* Bio was allocated by bio_kmalloc() */
266 kfree(bio);
267 }
268 }
269
270 void bio_init(struct bio *bio)
271 {
272 memset(bio, 0, sizeof(*bio));
273 bio->bi_flags = 1 << BIO_UPTODATE;
274 atomic_set(&bio->bi_remaining, 1);
275 atomic_set(&bio->bi_cnt, 1);
276 }
277 EXPORT_SYMBOL(bio_init);
278
279 /**
280 * bio_reset - reinitialize a bio
281 * @bio: bio to reset
282 *
283 * Description:
284 * After calling bio_reset(), @bio will be in the same state as a freshly
285 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
286 * preserved are the ones that are initialized by bio_alloc_bioset(). See
287 * comment in struct bio.
288 */
289 void bio_reset(struct bio *bio)
290 {
291 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
292
293 __bio_free(bio);
294
295 memset(bio, 0, BIO_RESET_BYTES);
296 bio->bi_flags = flags|(1 << BIO_UPTODATE);
297 atomic_set(&bio->bi_remaining, 1);
298 }
299 EXPORT_SYMBOL(bio_reset);
300
301 static void bio_chain_endio(struct bio *bio, int error)
302 {
303 bio_endio(bio->bi_private, error);
304 bio_put(bio);
305 }
306
307 /**
308 * bio_chain - chain bio completions
309 * @bio: the target bio
310 * @parent: the @bio's parent bio
311 *
312 * The caller won't have a bi_end_io called when @bio completes - instead,
313 * @parent's bi_end_io won't be called until both @parent and @bio have
314 * completed; the chained bio will also be freed when it completes.
315 *
316 * The caller must not set bi_private or bi_end_io in @bio.
317 */
318 void bio_chain(struct bio *bio, struct bio *parent)
319 {
320 BUG_ON(bio->bi_private || bio->bi_end_io);
321
322 bio->bi_private = parent;
323 bio->bi_end_io = bio_chain_endio;
324 atomic_inc(&parent->bi_remaining);
325 }
326 EXPORT_SYMBOL(bio_chain);
327
328 static void bio_alloc_rescue(struct work_struct *work)
329 {
330 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
331 struct bio *bio;
332
333 while (1) {
334 spin_lock(&bs->rescue_lock);
335 bio = bio_list_pop(&bs->rescue_list);
336 spin_unlock(&bs->rescue_lock);
337
338 if (!bio)
339 break;
340
341 generic_make_request(bio);
342 }
343 }
344
345 static void punt_bios_to_rescuer(struct bio_set *bs)
346 {
347 struct bio_list punt, nopunt;
348 struct bio *bio;
349
350 /*
351 * In order to guarantee forward progress we must punt only bios that
352 * were allocated from this bio_set; otherwise, if there was a bio on
353 * there for a stacking driver higher up in the stack, processing it
354 * could require allocating bios from this bio_set, and doing that from
355 * our own rescuer would be bad.
356 *
357 * Since bio lists are singly linked, pop them all instead of trying to
358 * remove from the middle of the list:
359 */
360
361 bio_list_init(&punt);
362 bio_list_init(&nopunt);
363
364 while ((bio = bio_list_pop(current->bio_list)))
365 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
366
367 *current->bio_list = nopunt;
368
369 spin_lock(&bs->rescue_lock);
370 bio_list_merge(&bs->rescue_list, &punt);
371 spin_unlock(&bs->rescue_lock);
372
373 queue_work(bs->rescue_workqueue, &bs->rescue_work);
374 }
375
376 /**
377 * bio_alloc_bioset - allocate a bio for I/O
378 * @gfp_mask: the GFP_ mask given to the slab allocator
379 * @nr_iovecs: number of iovecs to pre-allocate
380 * @bs: the bio_set to allocate from.
381 *
382 * Description:
383 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
384 * backed by the @bs's mempool.
385 *
386 * When @bs is not NULL, if %__GFP_WAIT is set then bio_alloc will always be
387 * able to allocate a bio. This is due to the mempool guarantees. To make this
388 * work, callers must never allocate more than 1 bio at a time from this pool.
389 * Callers that need to allocate more than 1 bio must always submit the
390 * previously allocated bio for IO before attempting to allocate a new one.
391 * Failure to do so can cause deadlocks under memory pressure.
392 *
393 * Note that when running under generic_make_request() (i.e. any block
394 * driver), bios are not submitted until after you return - see the code in
395 * generic_make_request() that converts recursion into iteration, to prevent
396 * stack overflows.
397 *
398 * This would normally mean allocating multiple bios under
399 * generic_make_request() would be susceptible to deadlocks, but we have
400 * deadlock avoidance code that resubmits any blocked bios from a rescuer
401 * thread.
402 *
403 * However, we do not guarantee forward progress for allocations from other
404 * mempools. Doing multiple allocations from the same mempool under
405 * generic_make_request() should be avoided - instead, use bio_set's front_pad
406 * for per bio allocations.
407 *
408 * RETURNS:
409 * Pointer to new bio on success, NULL on failure.
410 */
411 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
412 {
413 gfp_t saved_gfp = gfp_mask;
414 unsigned front_pad;
415 unsigned inline_vecs;
416 unsigned long idx = BIO_POOL_NONE;
417 struct bio_vec *bvl = NULL;
418 struct bio *bio;
419 void *p;
420
421 if (!bs) {
422 if (nr_iovecs > UIO_MAXIOV)
423 return NULL;
424
425 p = kmalloc(sizeof(struct bio) +
426 nr_iovecs * sizeof(struct bio_vec),
427 gfp_mask);
428 front_pad = 0;
429 inline_vecs = nr_iovecs;
430 } else {
431 /*
432 * generic_make_request() converts recursion to iteration; this
433 * means if we're running beneath it, any bios we allocate and
434 * submit will not be submitted (and thus freed) until after we
435 * return.
436 *
437 * This exposes us to a potential deadlock if we allocate
438 * multiple bios from the same bio_set() while running
439 * underneath generic_make_request(). If we were to allocate
440 * multiple bios (say a stacking block driver that was splitting
441 * bios), we would deadlock if we exhausted the mempool's
442 * reserve.
443 *
444 * We solve this, and guarantee forward progress, with a rescuer
445 * workqueue per bio_set. If we go to allocate and there are
446 * bios on current->bio_list, we first try the allocation
447 * without __GFP_WAIT; if that fails, we punt those bios we
448 * would be blocking to the rescuer workqueue before we retry
449 * with the original gfp_flags.
450 */
451
452 if (current->bio_list && !bio_list_empty(current->bio_list))
453 gfp_mask &= ~__GFP_WAIT;
454
455 p = mempool_alloc(bs->bio_pool, gfp_mask);
456 if (!p && gfp_mask != saved_gfp) {
457 punt_bios_to_rescuer(bs);
458 gfp_mask = saved_gfp;
459 p = mempool_alloc(bs->bio_pool, gfp_mask);
460 }
461
462 front_pad = bs->front_pad;
463 inline_vecs = BIO_INLINE_VECS;
464 }
465
466 if (unlikely(!p))
467 return NULL;
468
469 bio = p + front_pad;
470 bio_init(bio);
471
472 if (nr_iovecs > inline_vecs) {
473 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
474 if (!bvl && gfp_mask != saved_gfp) {
475 punt_bios_to_rescuer(bs);
476 gfp_mask = saved_gfp;
477 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
478 }
479
480 if (unlikely(!bvl))
481 goto err_free;
482
483 bio->bi_flags |= 1 << BIO_OWNS_VEC;
484 } else if (nr_iovecs) {
485 bvl = bio->bi_inline_vecs;
486 }
487
488 bio->bi_pool = bs;
489 bio->bi_flags |= idx << BIO_POOL_OFFSET;
490 bio->bi_max_vecs = nr_iovecs;
491 bio->bi_io_vec = bvl;
492 return bio;
493
494 err_free:
495 mempool_free(p, bs->bio_pool);
496 return NULL;
497 }
498 EXPORT_SYMBOL(bio_alloc_bioset);
499
500 void zero_fill_bio(struct bio *bio)
501 {
502 unsigned long flags;
503 struct bio_vec bv;
504 struct bvec_iter iter;
505
506 bio_for_each_segment(bv, bio, iter) {
507 char *data = bvec_kmap_irq(&bv, &flags);
508 memset(data, 0, bv.bv_len);
509 flush_dcache_page(bv.bv_page);
510 bvec_kunmap_irq(data, &flags);
511 }
512 }
513 EXPORT_SYMBOL(zero_fill_bio);
514
515 /**
516 * bio_put - release a reference to a bio
517 * @bio: bio to release reference to
518 *
519 * Description:
520 * Put a reference to a &struct bio, either one you have gotten with
521 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
522 **/
523 void bio_put(struct bio *bio)
524 {
525 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
526
527 /*
528 * last put frees it
529 */
530 if (atomic_dec_and_test(&bio->bi_cnt))
531 bio_free(bio);
532 }
533 EXPORT_SYMBOL(bio_put);
534
535 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
536 {
537 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
538 blk_recount_segments(q, bio);
539
540 return bio->bi_phys_segments;
541 }
542 EXPORT_SYMBOL(bio_phys_segments);
543
544 /**
545 * __bio_clone_fast - clone a bio that shares the original bio's biovec
546 * @bio: destination bio
547 * @bio_src: bio to clone
548 *
549 * Clone a &bio. Caller will own the returned bio, but not
550 * the actual data it points to. Reference count of returned
551 * bio will be one.
552 *
553 * Caller must ensure that @bio_src is not freed before @bio.
554 */
555 void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
556 {
557 BUG_ON(bio->bi_pool && BIO_POOL_IDX(bio) != BIO_POOL_NONE);
558
559 /*
560 * most users will be overriding ->bi_bdev with a new target,
561 * so we don't set nor calculate new physical/hw segment counts here
562 */
563 bio->bi_bdev = bio_src->bi_bdev;
564 bio->bi_flags |= 1 << BIO_CLONED;
565 bio->bi_rw = bio_src->bi_rw;
566 bio->bi_iter = bio_src->bi_iter;
567 bio->bi_io_vec = bio_src->bi_io_vec;
568 }
569 EXPORT_SYMBOL(__bio_clone_fast);
570
571 /**
572 * bio_clone_fast - clone a bio that shares the original bio's biovec
573 * @bio: bio to clone
574 * @gfp_mask: allocation priority
575 * @bs: bio_set to allocate from
576 *
577 * Like __bio_clone_fast, only also allocates the returned bio
578 */
579 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
580 {
581 struct bio *b;
582
583 b = bio_alloc_bioset(gfp_mask, 0, bs);
584 if (!b)
585 return NULL;
586
587 __bio_clone_fast(b, bio);
588
589 if (bio_integrity(bio)) {
590 int ret;
591
592 ret = bio_integrity_clone(b, bio, gfp_mask);
593
594 if (ret < 0) {
595 bio_put(b);
596 return NULL;
597 }
598 }
599
600 return b;
601 }
602 EXPORT_SYMBOL(bio_clone_fast);
603
604 /**
605 * bio_clone_bioset - clone a bio
606 * @bio_src: bio to clone
607 * @gfp_mask: allocation priority
608 * @bs: bio_set to allocate from
609 *
610 * Clone bio. Caller will own the returned bio, but not the actual data it
611 * points to. Reference count of returned bio will be one.
612 */
613 struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask,
614 struct bio_set *bs)
615 {
616 struct bvec_iter iter;
617 struct bio_vec bv;
618 struct bio *bio;
619
620 /*
621 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
622 * bio_src->bi_io_vec to bio->bi_io_vec.
623 *
624 * We can't do that anymore, because:
625 *
626 * - The point of cloning the biovec is to produce a bio with a biovec
627 * the caller can modify: bi_idx and bi_bvec_done should be 0.
628 *
629 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
630 * we tried to clone the whole thing bio_alloc_bioset() would fail.
631 * But the clone should succeed as long as the number of biovecs we
632 * actually need to allocate is fewer than BIO_MAX_PAGES.
633 *
634 * - Lastly, bi_vcnt should not be looked at or relied upon by code
635 * that does not own the bio - reason being drivers don't use it for
636 * iterating over the biovec anymore, so expecting it to be kept up
637 * to date (i.e. for clones that share the parent biovec) is just
638 * asking for trouble and would force extra work on
639 * __bio_clone_fast() anyways.
640 */
641
642 bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs);
643 if (!bio)
644 return NULL;
645
646 bio->bi_bdev = bio_src->bi_bdev;
647 bio->bi_rw = bio_src->bi_rw;
648 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector;
649 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
650
651 if (bio->bi_rw & REQ_DISCARD)
652 goto integrity_clone;
653
654 if (bio->bi_rw & REQ_WRITE_SAME) {
655 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
656 goto integrity_clone;
657 }
658
659 bio_for_each_segment(bv, bio_src, iter)
660 bio->bi_io_vec[bio->bi_vcnt++] = bv;
661
662 integrity_clone:
663 if (bio_integrity(bio_src)) {
664 int ret;
665
666 ret = bio_integrity_clone(bio, bio_src, gfp_mask);
667 if (ret < 0) {
668 bio_put(bio);
669 return NULL;
670 }
671 }
672
673 return bio;
674 }
675 EXPORT_SYMBOL(bio_clone_bioset);
676
677 /**
678 * bio_get_nr_vecs - return approx number of vecs
679 * @bdev: I/O target
680 *
681 * Return the approximate number of pages we can send to this target.
682 * There's no guarantee that you will be able to fit this number of pages
683 * into a bio, it does not account for dynamic restrictions that vary
684 * on offset.
685 */
686 int bio_get_nr_vecs(struct block_device *bdev)
687 {
688 struct request_queue *q = bdev_get_queue(bdev);
689 int nr_pages;
690
691 nr_pages = min_t(unsigned,
692 queue_max_segments(q),
693 queue_max_sectors(q) / (PAGE_SIZE >> 9) + 1);
694
695 return min_t(unsigned, nr_pages, BIO_MAX_PAGES);
696
697 }
698 EXPORT_SYMBOL(bio_get_nr_vecs);
699
700 static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
701 *page, unsigned int len, unsigned int offset,
702 unsigned int max_sectors)
703 {
704 int retried_segments = 0;
705 struct bio_vec *bvec;
706
707 /*
708 * cloned bio must not modify vec list
709 */
710 if (unlikely(bio_flagged(bio, BIO_CLONED)))
711 return 0;
712
713 if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors)
714 return 0;
715
716 /*
717 * For filesystems with a blocksize smaller than the pagesize
718 * we will often be called with the same page as last time and
719 * a consecutive offset. Optimize this special case.
720 */
721 if (bio->bi_vcnt > 0) {
722 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
723
724 if (page == prev->bv_page &&
725 offset == prev->bv_offset + prev->bv_len) {
726 unsigned int prev_bv_len = prev->bv_len;
727 prev->bv_len += len;
728
729 if (q->merge_bvec_fn) {
730 struct bvec_merge_data bvm = {
731 /* prev_bvec is already charged in
732 bi_size, discharge it in order to
733 simulate merging updated prev_bvec
734 as new bvec. */
735 .bi_bdev = bio->bi_bdev,
736 .bi_sector = bio->bi_iter.bi_sector,
737 .bi_size = bio->bi_iter.bi_size -
738 prev_bv_len,
739 .bi_rw = bio->bi_rw,
740 };
741
742 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) {
743 prev->bv_len -= len;
744 return 0;
745 }
746 }
747
748 goto done;
749 }
750
751 /*
752 * If the queue doesn't support SG gaps and adding this
753 * offset would create a gap, disallow it.
754 */
755 if (q->queue_flags & (1 << QUEUE_FLAG_SG_GAPS) &&
756 bvec_gap_to_prev(prev, offset))
757 return 0;
758 }
759
760 if (bio->bi_vcnt >= bio->bi_max_vecs)
761 return 0;
762
763 /*
764 * we might lose a segment or two here, but rather that than
765 * make this too complex.
766 */
767
768 while (bio->bi_phys_segments >= queue_max_segments(q)) {
769
770 if (retried_segments)
771 return 0;
772
773 retried_segments = 1;
774 blk_recount_segments(q, bio);
775 }
776
777 /*
778 * setup the new entry, we might clear it again later if we
779 * cannot add the page
780 */
781 bvec = &bio->bi_io_vec[bio->bi_vcnt];
782 bvec->bv_page = page;
783 bvec->bv_len = len;
784 bvec->bv_offset = offset;
785
786 /*
787 * if queue has other restrictions (eg varying max sector size
788 * depending on offset), it can specify a merge_bvec_fn in the
789 * queue to get further control
790 */
791 if (q->merge_bvec_fn) {
792 struct bvec_merge_data bvm = {
793 .bi_bdev = bio->bi_bdev,
794 .bi_sector = bio->bi_iter.bi_sector,
795 .bi_size = bio->bi_iter.bi_size,
796 .bi_rw = bio->bi_rw,
797 };
798
799 /*
800 * merge_bvec_fn() returns number of bytes it can accept
801 * at this offset
802 */
803 if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len) {
804 bvec->bv_page = NULL;
805 bvec->bv_len = 0;
806 bvec->bv_offset = 0;
807 return 0;
808 }
809 }
810
811 /* If we may be able to merge these biovecs, force a recount */
812 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
813 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
814
815 bio->bi_vcnt++;
816 bio->bi_phys_segments++;
817 done:
818 bio->bi_iter.bi_size += len;
819 return len;
820 }
821
822 /**
823 * bio_add_pc_page - attempt to add page to bio
824 * @q: the target queue
825 * @bio: destination bio
826 * @page: page to add
827 * @len: vec entry length
828 * @offset: vec entry offset
829 *
830 * Attempt to add a page to the bio_vec maplist. This can fail for a
831 * number of reasons, such as the bio being full or target block device
832 * limitations. The target block device must allow bio's up to PAGE_SIZE,
833 * so it is always possible to add a single page to an empty bio.
834 *
835 * This should only be used by REQ_PC bios.
836 */
837 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
838 unsigned int len, unsigned int offset)
839 {
840 return __bio_add_page(q, bio, page, len, offset,
841 queue_max_hw_sectors(q));
842 }
843 EXPORT_SYMBOL(bio_add_pc_page);
844
845 /**
846 * bio_add_page - attempt to add page to bio
847 * @bio: destination bio
848 * @page: page to add
849 * @len: vec entry length
850 * @offset: vec entry offset
851 *
852 * Attempt to add a page to the bio_vec maplist. This can fail for a
853 * number of reasons, such as the bio being full or target block device
854 * limitations. The target block device must allow bio's up to PAGE_SIZE,
855 * so it is always possible to add a single page to an empty bio.
856 */
857 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
858 unsigned int offset)
859 {
860 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
861 unsigned int max_sectors;
862
863 max_sectors = blk_max_size_offset(q, bio->bi_iter.bi_sector);
864 if ((max_sectors < (len >> 9)) && !bio->bi_iter.bi_size)
865 max_sectors = len >> 9;
866
867 return __bio_add_page(q, bio, page, len, offset, max_sectors);
868 }
869 EXPORT_SYMBOL(bio_add_page);
870
871 struct submit_bio_ret {
872 struct completion event;
873 int error;
874 };
875
876 static void submit_bio_wait_endio(struct bio *bio, int error)
877 {
878 struct submit_bio_ret *ret = bio->bi_private;
879
880 ret->error = error;
881 complete(&ret->event);
882 }
883
884 /**
885 * submit_bio_wait - submit a bio, and wait until it completes
886 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
887 * @bio: The &struct bio which describes the I/O
888 *
889 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
890 * bio_endio() on failure.
891 */
892 int submit_bio_wait(int rw, struct bio *bio)
893 {
894 struct submit_bio_ret ret;
895
896 rw |= REQ_SYNC;
897 init_completion(&ret.event);
898 bio->bi_private = &ret;
899 bio->bi_end_io = submit_bio_wait_endio;
900 submit_bio(rw, bio);
901 wait_for_completion(&ret.event);
902
903 return ret.error;
904 }
905 EXPORT_SYMBOL(submit_bio_wait);
906
907 /**
908 * bio_advance - increment/complete a bio by some number of bytes
909 * @bio: bio to advance
910 * @bytes: number of bytes to complete
911 *
912 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
913 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
914 * be updated on the last bvec as well.
915 *
916 * @bio will then represent the remaining, uncompleted portion of the io.
917 */
918 void bio_advance(struct bio *bio, unsigned bytes)
919 {
920 if (bio_integrity(bio))
921 bio_integrity_advance(bio, bytes);
922
923 bio_advance_iter(bio, &bio->bi_iter, bytes);
924 }
925 EXPORT_SYMBOL(bio_advance);
926
927 /**
928 * bio_alloc_pages - allocates a single page for each bvec in a bio
929 * @bio: bio to allocate pages for
930 * @gfp_mask: flags for allocation
931 *
932 * Allocates pages up to @bio->bi_vcnt.
933 *
934 * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
935 * freed.
936 */
937 int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask)
938 {
939 int i;
940 struct bio_vec *bv;
941
942 bio_for_each_segment_all(bv, bio, i) {
943 bv->bv_page = alloc_page(gfp_mask);
944 if (!bv->bv_page) {
945 while (--bv >= bio->bi_io_vec)
946 __free_page(bv->bv_page);
947 return -ENOMEM;
948 }
949 }
950
951 return 0;
952 }
953 EXPORT_SYMBOL(bio_alloc_pages);
954
955 /**
956 * bio_copy_data - copy contents of data buffers from one chain of bios to
957 * another
958 * @src: source bio list
959 * @dst: destination bio list
960 *
961 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
962 * @src and @dst as linked lists of bios.
963 *
964 * Stops when it reaches the end of either @src or @dst - that is, copies
965 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
966 */
967 void bio_copy_data(struct bio *dst, struct bio *src)
968 {
969 struct bvec_iter src_iter, dst_iter;
970 struct bio_vec src_bv, dst_bv;
971 void *src_p, *dst_p;
972 unsigned bytes;
973
974 src_iter = src->bi_iter;
975 dst_iter = dst->bi_iter;
976
977 while (1) {
978 if (!src_iter.bi_size) {
979 src = src->bi_next;
980 if (!src)
981 break;
982
983 src_iter = src->bi_iter;
984 }
985
986 if (!dst_iter.bi_size) {
987 dst = dst->bi_next;
988 if (!dst)
989 break;
990
991 dst_iter = dst->bi_iter;
992 }
993
994 src_bv = bio_iter_iovec(src, src_iter);
995 dst_bv = bio_iter_iovec(dst, dst_iter);
996
997 bytes = min(src_bv.bv_len, dst_bv.bv_len);
998
999 src_p = kmap_atomic(src_bv.bv_page);
1000 dst_p = kmap_atomic(dst_bv.bv_page);
1001
1002 memcpy(dst_p + dst_bv.bv_offset,
1003 src_p + src_bv.bv_offset,
1004 bytes);
1005
1006 kunmap_atomic(dst_p);
1007 kunmap_atomic(src_p);
1008
1009 bio_advance_iter(src, &src_iter, bytes);
1010 bio_advance_iter(dst, &dst_iter, bytes);
1011 }
1012 }
1013 EXPORT_SYMBOL(bio_copy_data);
1014
1015 struct bio_map_data {
1016 int nr_sgvecs;
1017 int is_our_pages;
1018 struct sg_iovec sgvecs[];
1019 };
1020
1021 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
1022 const struct sg_iovec *iov, int iov_count,
1023 int is_our_pages)
1024 {
1025 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
1026 bmd->nr_sgvecs = iov_count;
1027 bmd->is_our_pages = is_our_pages;
1028 bio->bi_private = bmd;
1029 }
1030
1031 static struct bio_map_data *bio_alloc_map_data(unsigned int iov_count,
1032 gfp_t gfp_mask)
1033 {
1034 if (iov_count > UIO_MAXIOV)
1035 return NULL;
1036
1037 return kmalloc(sizeof(struct bio_map_data) +
1038 sizeof(struct sg_iovec) * iov_count, gfp_mask);
1039 }
1040
1041 static int __bio_copy_iov(struct bio *bio, const struct sg_iovec *iov, int iov_count,
1042 int to_user, int from_user, int do_free_page)
1043 {
1044 int ret = 0, i;
1045 struct bio_vec *bvec;
1046 int iov_idx = 0;
1047 unsigned int iov_off = 0;
1048
1049 bio_for_each_segment_all(bvec, bio, i) {
1050 char *bv_addr = page_address(bvec->bv_page);
1051 unsigned int bv_len = bvec->bv_len;
1052
1053 while (bv_len && iov_idx < iov_count) {
1054 unsigned int bytes;
1055 char __user *iov_addr;
1056
1057 bytes = min_t(unsigned int,
1058 iov[iov_idx].iov_len - iov_off, bv_len);
1059 iov_addr = iov[iov_idx].iov_base + iov_off;
1060
1061 if (!ret) {
1062 if (to_user)
1063 ret = copy_to_user(iov_addr, bv_addr,
1064 bytes);
1065
1066 if (from_user)
1067 ret = copy_from_user(bv_addr, iov_addr,
1068 bytes);
1069
1070 if (ret)
1071 ret = -EFAULT;
1072 }
1073
1074 bv_len -= bytes;
1075 bv_addr += bytes;
1076 iov_addr += bytes;
1077 iov_off += bytes;
1078
1079 if (iov[iov_idx].iov_len == iov_off) {
1080 iov_idx++;
1081 iov_off = 0;
1082 }
1083 }
1084
1085 if (do_free_page)
1086 __free_page(bvec->bv_page);
1087 }
1088
1089 return ret;
1090 }
1091
1092 /**
1093 * bio_uncopy_user - finish previously mapped bio
1094 * @bio: bio being terminated
1095 *
1096 * Free pages allocated from bio_copy_user() and write back data
1097 * to user space in case of a read.
1098 */
1099 int bio_uncopy_user(struct bio *bio)
1100 {
1101 struct bio_map_data *bmd = bio->bi_private;
1102 struct bio_vec *bvec;
1103 int ret = 0, i;
1104
1105 if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1106 /*
1107 * if we're in a workqueue, the request is orphaned, so
1108 * don't copy into a random user address space, just free.
1109 */
1110 if (current->mm)
1111 ret = __bio_copy_iov(bio, bmd->sgvecs, bmd->nr_sgvecs,
1112 bio_data_dir(bio) == READ,
1113 0, bmd->is_our_pages);
1114 else if (bmd->is_our_pages)
1115 bio_for_each_segment_all(bvec, bio, i)
1116 __free_page(bvec->bv_page);
1117 }
1118 kfree(bmd);
1119 bio_put(bio);
1120 return ret;
1121 }
1122 EXPORT_SYMBOL(bio_uncopy_user);
1123
1124 /**
1125 * bio_copy_user_iov - copy user data to bio
1126 * @q: destination block queue
1127 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1128 * @iov: the iovec.
1129 * @iov_count: number of elements in the iovec
1130 * @write_to_vm: bool indicating writing to pages or not
1131 * @gfp_mask: memory allocation flags
1132 *
1133 * Prepares and returns a bio for indirect user io, bouncing data
1134 * to/from kernel pages as necessary. Must be paired with
1135 * call bio_uncopy_user() on io completion.
1136 */
1137 struct bio *bio_copy_user_iov(struct request_queue *q,
1138 struct rq_map_data *map_data,
1139 const struct sg_iovec *iov, int iov_count,
1140 int write_to_vm, gfp_t gfp_mask)
1141 {
1142 struct bio_map_data *bmd;
1143 struct bio_vec *bvec;
1144 struct page *page;
1145 struct bio *bio;
1146 int i, ret;
1147 int nr_pages = 0;
1148 unsigned int len = 0;
1149 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
1150
1151 for (i = 0; i < iov_count; i++) {
1152 unsigned long uaddr;
1153 unsigned long end;
1154 unsigned long start;
1155
1156 uaddr = (unsigned long)iov[i].iov_base;
1157 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1158 start = uaddr >> PAGE_SHIFT;
1159
1160 /*
1161 * Overflow, abort
1162 */
1163 if (end < start)
1164 return ERR_PTR(-EINVAL);
1165
1166 nr_pages += end - start;
1167 len += iov[i].iov_len;
1168 }
1169
1170 if (offset)
1171 nr_pages++;
1172
1173 bmd = bio_alloc_map_data(iov_count, gfp_mask);
1174 if (!bmd)
1175 return ERR_PTR(-ENOMEM);
1176
1177 ret = -ENOMEM;
1178 bio = bio_kmalloc(gfp_mask, nr_pages);
1179 if (!bio)
1180 goto out_bmd;
1181
1182 if (!write_to_vm)
1183 bio->bi_rw |= REQ_WRITE;
1184
1185 ret = 0;
1186
1187 if (map_data) {
1188 nr_pages = 1 << map_data->page_order;
1189 i = map_data->offset / PAGE_SIZE;
1190 }
1191 while (len) {
1192 unsigned int bytes = PAGE_SIZE;
1193
1194 bytes -= offset;
1195
1196 if (bytes > len)
1197 bytes = len;
1198
1199 if (map_data) {
1200 if (i == map_data->nr_entries * nr_pages) {
1201 ret = -ENOMEM;
1202 break;
1203 }
1204
1205 page = map_data->pages[i / nr_pages];
1206 page += (i % nr_pages);
1207
1208 i++;
1209 } else {
1210 page = alloc_page(q->bounce_gfp | gfp_mask);
1211 if (!page) {
1212 ret = -ENOMEM;
1213 break;
1214 }
1215 }
1216
1217 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
1218 break;
1219
1220 len -= bytes;
1221 offset = 0;
1222 }
1223
1224 if (ret)
1225 goto cleanup;
1226
1227 /*
1228 * success
1229 */
1230 if ((!write_to_vm && (!map_data || !map_data->null_mapped)) ||
1231 (map_data && map_data->from_user)) {
1232 ret = __bio_copy_iov(bio, iov, iov_count, 0, 1, 0);
1233 if (ret)
1234 goto cleanup;
1235 }
1236
1237 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
1238 return bio;
1239 cleanup:
1240 if (!map_data)
1241 bio_for_each_segment_all(bvec, bio, i)
1242 __free_page(bvec->bv_page);
1243
1244 bio_put(bio);
1245 out_bmd:
1246 kfree(bmd);
1247 return ERR_PTR(ret);
1248 }
1249
1250 /**
1251 * bio_copy_user - copy user data to bio
1252 * @q: destination block queue
1253 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1254 * @uaddr: start of user address
1255 * @len: length in bytes
1256 * @write_to_vm: bool indicating writing to pages or not
1257 * @gfp_mask: memory allocation flags
1258 *
1259 * Prepares and returns a bio for indirect user io, bouncing data
1260 * to/from kernel pages as necessary. Must be paired with
1261 * call bio_uncopy_user() on io completion.
1262 */
1263 struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
1264 unsigned long uaddr, unsigned int len,
1265 int write_to_vm, gfp_t gfp_mask)
1266 {
1267 struct sg_iovec iov;
1268
1269 iov.iov_base = (void __user *)uaddr;
1270 iov.iov_len = len;
1271
1272 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
1273 }
1274 EXPORT_SYMBOL(bio_copy_user);
1275
1276 static struct bio *__bio_map_user_iov(struct request_queue *q,
1277 struct block_device *bdev,
1278 const struct sg_iovec *iov, int iov_count,
1279 int write_to_vm, gfp_t gfp_mask)
1280 {
1281 int i, j;
1282 int nr_pages = 0;
1283 struct page **pages;
1284 struct bio *bio;
1285 int cur_page = 0;
1286 int ret, offset;
1287
1288 for (i = 0; i < iov_count; i++) {
1289 unsigned long uaddr = (unsigned long)iov[i].iov_base;
1290 unsigned long len = iov[i].iov_len;
1291 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1292 unsigned long start = uaddr >> PAGE_SHIFT;
1293
1294 /*
1295 * Overflow, abort
1296 */
1297 if (end < start)
1298 return ERR_PTR(-EINVAL);
1299
1300 nr_pages += end - start;
1301 /*
1302 * buffer must be aligned to at least hardsector size for now
1303 */
1304 if (uaddr & queue_dma_alignment(q))
1305 return ERR_PTR(-EINVAL);
1306 }
1307
1308 if (!nr_pages)
1309 return ERR_PTR(-EINVAL);
1310
1311 bio = bio_kmalloc(gfp_mask, nr_pages);
1312 if (!bio)
1313 return ERR_PTR(-ENOMEM);
1314
1315 ret = -ENOMEM;
1316 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
1317 if (!pages)
1318 goto out;
1319
1320 for (i = 0; i < iov_count; i++) {
1321 unsigned long uaddr = (unsigned long)iov[i].iov_base;
1322 unsigned long len = iov[i].iov_len;
1323 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1324 unsigned long start = uaddr >> PAGE_SHIFT;
1325 const int local_nr_pages = end - start;
1326 const int page_limit = cur_page + local_nr_pages;
1327
1328 ret = get_user_pages_fast(uaddr, local_nr_pages,
1329 write_to_vm, &pages[cur_page]);
1330 if (ret < local_nr_pages) {
1331 ret = -EFAULT;
1332 goto out_unmap;
1333 }
1334
1335 offset = uaddr & ~PAGE_MASK;
1336 for (j = cur_page; j < page_limit; j++) {
1337 unsigned int bytes = PAGE_SIZE - offset;
1338
1339 if (len <= 0)
1340 break;
1341
1342 if (bytes > len)
1343 bytes = len;
1344
1345 /*
1346 * sorry...
1347 */
1348 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1349 bytes)
1350 break;
1351
1352 len -= bytes;
1353 offset = 0;
1354 }
1355
1356 cur_page = j;
1357 /*
1358 * release the pages we didn't map into the bio, if any
1359 */
1360 while (j < page_limit)
1361 page_cache_release(pages[j++]);
1362 }
1363
1364 kfree(pages);
1365
1366 /*
1367 * set data direction, and check if mapped pages need bouncing
1368 */
1369 if (!write_to_vm)
1370 bio->bi_rw |= REQ_WRITE;
1371
1372 bio->bi_bdev = bdev;
1373 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1374 return bio;
1375
1376 out_unmap:
1377 for (i = 0; i < nr_pages; i++) {
1378 if(!pages[i])
1379 break;
1380 page_cache_release(pages[i]);
1381 }
1382 out:
1383 kfree(pages);
1384 bio_put(bio);
1385 return ERR_PTR(ret);
1386 }
1387
1388 /**
1389 * bio_map_user - map user address into bio
1390 * @q: the struct request_queue for the bio
1391 * @bdev: destination block device
1392 * @uaddr: start of user address
1393 * @len: length in bytes
1394 * @write_to_vm: bool indicating writing to pages or not
1395 * @gfp_mask: memory allocation flags
1396 *
1397 * Map the user space address into a bio suitable for io to a block
1398 * device. Returns an error pointer in case of error.
1399 */
1400 struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
1401 unsigned long uaddr, unsigned int len, int write_to_vm,
1402 gfp_t gfp_mask)
1403 {
1404 struct sg_iovec iov;
1405
1406 iov.iov_base = (void __user *)uaddr;
1407 iov.iov_len = len;
1408
1409 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
1410 }
1411 EXPORT_SYMBOL(bio_map_user);
1412
1413 /**
1414 * bio_map_user_iov - map user sg_iovec table into bio
1415 * @q: the struct request_queue for the bio
1416 * @bdev: destination block device
1417 * @iov: the iovec.
1418 * @iov_count: number of elements in the iovec
1419 * @write_to_vm: bool indicating writing to pages or not
1420 * @gfp_mask: memory allocation flags
1421 *
1422 * Map the user space address into a bio suitable for io to a block
1423 * device. Returns an error pointer in case of error.
1424 */
1425 struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
1426 const struct sg_iovec *iov, int iov_count,
1427 int write_to_vm, gfp_t gfp_mask)
1428 {
1429 struct bio *bio;
1430
1431 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1432 gfp_mask);
1433 if (IS_ERR(bio))
1434 return bio;
1435
1436 /*
1437 * subtle -- if __bio_map_user() ended up bouncing a bio,
1438 * it would normally disappear when its bi_end_io is run.
1439 * however, we need it for the unmap, so grab an extra
1440 * reference to it
1441 */
1442 bio_get(bio);
1443
1444 return bio;
1445 }
1446
1447 static void __bio_unmap_user(struct bio *bio)
1448 {
1449 struct bio_vec *bvec;
1450 int i;
1451
1452 /*
1453 * make sure we dirty pages we wrote to
1454 */
1455 bio_for_each_segment_all(bvec, bio, i) {
1456 if (bio_data_dir(bio) == READ)
1457 set_page_dirty_lock(bvec->bv_page);
1458
1459 page_cache_release(bvec->bv_page);
1460 }
1461
1462 bio_put(bio);
1463 }
1464
1465 /**
1466 * bio_unmap_user - unmap a bio
1467 * @bio: the bio being unmapped
1468 *
1469 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1470 * a process context.
1471 *
1472 * bio_unmap_user() may sleep.
1473 */
1474 void bio_unmap_user(struct bio *bio)
1475 {
1476 __bio_unmap_user(bio);
1477 bio_put(bio);
1478 }
1479 EXPORT_SYMBOL(bio_unmap_user);
1480
1481 static void bio_map_kern_endio(struct bio *bio, int err)
1482 {
1483 bio_put(bio);
1484 }
1485
1486 static struct bio *__bio_map_kern(struct request_queue *q, void *data,
1487 unsigned int len, gfp_t gfp_mask)
1488 {
1489 unsigned long kaddr = (unsigned long)data;
1490 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1491 unsigned long start = kaddr >> PAGE_SHIFT;
1492 const int nr_pages = end - start;
1493 int offset, i;
1494 struct bio *bio;
1495
1496 bio = bio_kmalloc(gfp_mask, nr_pages);
1497 if (!bio)
1498 return ERR_PTR(-ENOMEM);
1499
1500 offset = offset_in_page(kaddr);
1501 for (i = 0; i < nr_pages; i++) {
1502 unsigned int bytes = PAGE_SIZE - offset;
1503
1504 if (len <= 0)
1505 break;
1506
1507 if (bytes > len)
1508 bytes = len;
1509
1510 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1511 offset) < bytes)
1512 break;
1513
1514 data += bytes;
1515 len -= bytes;
1516 offset = 0;
1517 }
1518
1519 bio->bi_end_io = bio_map_kern_endio;
1520 return bio;
1521 }
1522
1523 /**
1524 * bio_map_kern - map kernel address into bio
1525 * @q: the struct request_queue for the bio
1526 * @data: pointer to buffer to map
1527 * @len: length in bytes
1528 * @gfp_mask: allocation flags for bio allocation
1529 *
1530 * Map the kernel address into a bio suitable for io to a block
1531 * device. Returns an error pointer in case of error.
1532 */
1533 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1534 gfp_t gfp_mask)
1535 {
1536 struct bio *bio;
1537
1538 bio = __bio_map_kern(q, data, len, gfp_mask);
1539 if (IS_ERR(bio))
1540 return bio;
1541
1542 if (bio->bi_iter.bi_size == len)
1543 return bio;
1544
1545 /*
1546 * Don't support partial mappings.
1547 */
1548 bio_put(bio);
1549 return ERR_PTR(-EINVAL);
1550 }
1551 EXPORT_SYMBOL(bio_map_kern);
1552
1553 static void bio_copy_kern_endio(struct bio *bio, int err)
1554 {
1555 struct bio_vec *bvec;
1556 const int read = bio_data_dir(bio) == READ;
1557 struct bio_map_data *bmd = bio->bi_private;
1558 int i;
1559 char *p = bmd->sgvecs[0].iov_base;
1560
1561 bio_for_each_segment_all(bvec, bio, i) {
1562 char *addr = page_address(bvec->bv_page);
1563
1564 if (read)
1565 memcpy(p, addr, bvec->bv_len);
1566
1567 __free_page(bvec->bv_page);
1568 p += bvec->bv_len;
1569 }
1570
1571 kfree(bmd);
1572 bio_put(bio);
1573 }
1574
1575 /**
1576 * bio_copy_kern - copy kernel address into bio
1577 * @q: the struct request_queue for the bio
1578 * @data: pointer to buffer to copy
1579 * @len: length in bytes
1580 * @gfp_mask: allocation flags for bio and page allocation
1581 * @reading: data direction is READ
1582 *
1583 * copy the kernel address into a bio suitable for io to a block
1584 * device. Returns an error pointer in case of error.
1585 */
1586 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1587 gfp_t gfp_mask, int reading)
1588 {
1589 struct bio *bio;
1590 struct bio_vec *bvec;
1591 int i;
1592
1593 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1594 if (IS_ERR(bio))
1595 return bio;
1596
1597 if (!reading) {
1598 void *p = data;
1599
1600 bio_for_each_segment_all(bvec, bio, i) {
1601 char *addr = page_address(bvec->bv_page);
1602
1603 memcpy(addr, p, bvec->bv_len);
1604 p += bvec->bv_len;
1605 }
1606 }
1607
1608 bio->bi_end_io = bio_copy_kern_endio;
1609
1610 return bio;
1611 }
1612 EXPORT_SYMBOL(bio_copy_kern);
1613
1614 /*
1615 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1616 * for performing direct-IO in BIOs.
1617 *
1618 * The problem is that we cannot run set_page_dirty() from interrupt context
1619 * because the required locks are not interrupt-safe. So what we can do is to
1620 * mark the pages dirty _before_ performing IO. And in interrupt context,
1621 * check that the pages are still dirty. If so, fine. If not, redirty them
1622 * in process context.
1623 *
1624 * We special-case compound pages here: normally this means reads into hugetlb
1625 * pages. The logic in here doesn't really work right for compound pages
1626 * because the VM does not uniformly chase down the head page in all cases.
1627 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1628 * handle them at all. So we skip compound pages here at an early stage.
1629 *
1630 * Note that this code is very hard to test under normal circumstances because
1631 * direct-io pins the pages with get_user_pages(). This makes
1632 * is_page_cache_freeable return false, and the VM will not clean the pages.
1633 * But other code (eg, flusher threads) could clean the pages if they are mapped
1634 * pagecache.
1635 *
1636 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1637 * deferred bio dirtying paths.
1638 */
1639
1640 /*
1641 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1642 */
1643 void bio_set_pages_dirty(struct bio *bio)
1644 {
1645 struct bio_vec *bvec;
1646 int i;
1647
1648 bio_for_each_segment_all(bvec, bio, i) {
1649 struct page *page = bvec->bv_page;
1650
1651 if (page && !PageCompound(page))
1652 set_page_dirty_lock(page);
1653 }
1654 }
1655
1656 static void bio_release_pages(struct bio *bio)
1657 {
1658 struct bio_vec *bvec;
1659 int i;
1660
1661 bio_for_each_segment_all(bvec, bio, i) {
1662 struct page *page = bvec->bv_page;
1663
1664 if (page)
1665 put_page(page);
1666 }
1667 }
1668
1669 /*
1670 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1671 * If they are, then fine. If, however, some pages are clean then they must
1672 * have been written out during the direct-IO read. So we take another ref on
1673 * the BIO and the offending pages and re-dirty the pages in process context.
1674 *
1675 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1676 * here on. It will run one page_cache_release() against each page and will
1677 * run one bio_put() against the BIO.
1678 */
1679
1680 static void bio_dirty_fn(struct work_struct *work);
1681
1682 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1683 static DEFINE_SPINLOCK(bio_dirty_lock);
1684 static struct bio *bio_dirty_list;
1685
1686 /*
1687 * This runs in process context
1688 */
1689 static void bio_dirty_fn(struct work_struct *work)
1690 {
1691 unsigned long flags;
1692 struct bio *bio;
1693
1694 spin_lock_irqsave(&bio_dirty_lock, flags);
1695 bio = bio_dirty_list;
1696 bio_dirty_list = NULL;
1697 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1698
1699 while (bio) {
1700 struct bio *next = bio->bi_private;
1701
1702 bio_set_pages_dirty(bio);
1703 bio_release_pages(bio);
1704 bio_put(bio);
1705 bio = next;
1706 }
1707 }
1708
1709 void bio_check_pages_dirty(struct bio *bio)
1710 {
1711 struct bio_vec *bvec;
1712 int nr_clean_pages = 0;
1713 int i;
1714
1715 bio_for_each_segment_all(bvec, bio, i) {
1716 struct page *page = bvec->bv_page;
1717
1718 if (PageDirty(page) || PageCompound(page)) {
1719 page_cache_release(page);
1720 bvec->bv_page = NULL;
1721 } else {
1722 nr_clean_pages++;
1723 }
1724 }
1725
1726 if (nr_clean_pages) {
1727 unsigned long flags;
1728
1729 spin_lock_irqsave(&bio_dirty_lock, flags);
1730 bio->bi_private = bio_dirty_list;
1731 bio_dirty_list = bio;
1732 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1733 schedule_work(&bio_dirty_work);
1734 } else {
1735 bio_put(bio);
1736 }
1737 }
1738
1739 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1740 void bio_flush_dcache_pages(struct bio *bi)
1741 {
1742 struct bio_vec bvec;
1743 struct bvec_iter iter;
1744
1745 bio_for_each_segment(bvec, bi, iter)
1746 flush_dcache_page(bvec.bv_page);
1747 }
1748 EXPORT_SYMBOL(bio_flush_dcache_pages);
1749 #endif
1750
1751 /**
1752 * bio_endio - end I/O on a bio
1753 * @bio: bio
1754 * @error: error, if any
1755 *
1756 * Description:
1757 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1758 * preferred way to end I/O on a bio, it takes care of clearing
1759 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1760 * established -Exxxx (-EIO, for instance) error values in case
1761 * something went wrong. No one should call bi_end_io() directly on a
1762 * bio unless they own it and thus know that it has an end_io
1763 * function.
1764 **/
1765 void bio_endio(struct bio *bio, int error)
1766 {
1767 while (bio) {
1768 BUG_ON(atomic_read(&bio->bi_remaining) <= 0);
1769
1770 if (error)
1771 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1772 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1773 error = -EIO;
1774
1775 if (!atomic_dec_and_test(&bio->bi_remaining))
1776 return;
1777
1778 /*
1779 * Need to have a real endio function for chained bios,
1780 * otherwise various corner cases will break (like stacking
1781 * block devices that save/restore bi_end_io) - however, we want
1782 * to avoid unbounded recursion and blowing the stack. Tail call
1783 * optimization would handle this, but compiling with frame
1784 * pointers also disables gcc's sibling call optimization.
1785 */
1786 if (bio->bi_end_io == bio_chain_endio) {
1787 struct bio *parent = bio->bi_private;
1788 bio_put(bio);
1789 bio = parent;
1790 } else {
1791 if (bio->bi_end_io)
1792 bio->bi_end_io(bio, error);
1793 bio = NULL;
1794 }
1795 }
1796 }
1797 EXPORT_SYMBOL(bio_endio);
1798
1799 /**
1800 * bio_endio_nodec - end I/O on a bio, without decrementing bi_remaining
1801 * @bio: bio
1802 * @error: error, if any
1803 *
1804 * For code that has saved and restored bi_end_io; thing hard before using this
1805 * function, probably you should've cloned the entire bio.
1806 **/
1807 void bio_endio_nodec(struct bio *bio, int error)
1808 {
1809 atomic_inc(&bio->bi_remaining);
1810 bio_endio(bio, error);
1811 }
1812 EXPORT_SYMBOL(bio_endio_nodec);
1813
1814 /**
1815 * bio_split - split a bio
1816 * @bio: bio to split
1817 * @sectors: number of sectors to split from the front of @bio
1818 * @gfp: gfp mask
1819 * @bs: bio set to allocate from
1820 *
1821 * Allocates and returns a new bio which represents @sectors from the start of
1822 * @bio, and updates @bio to represent the remaining sectors.
1823 *
1824 * The newly allocated bio will point to @bio's bi_io_vec; it is the caller's
1825 * responsibility to ensure that @bio is not freed before the split.
1826 */
1827 struct bio *bio_split(struct bio *bio, int sectors,
1828 gfp_t gfp, struct bio_set *bs)
1829 {
1830 struct bio *split = NULL;
1831
1832 BUG_ON(sectors <= 0);
1833 BUG_ON(sectors >= bio_sectors(bio));
1834
1835 split = bio_clone_fast(bio, gfp, bs);
1836 if (!split)
1837 return NULL;
1838
1839 split->bi_iter.bi_size = sectors << 9;
1840
1841 if (bio_integrity(split))
1842 bio_integrity_trim(split, 0, sectors);
1843
1844 bio_advance(bio, split->bi_iter.bi_size);
1845
1846 return split;
1847 }
1848 EXPORT_SYMBOL(bio_split);
1849
1850 /**
1851 * bio_trim - trim a bio
1852 * @bio: bio to trim
1853 * @offset: number of sectors to trim from the front of @bio
1854 * @size: size we want to trim @bio to, in sectors
1855 */
1856 void bio_trim(struct bio *bio, int offset, int size)
1857 {
1858 /* 'bio' is a cloned bio which we need to trim to match
1859 * the given offset and size.
1860 */
1861
1862 size <<= 9;
1863 if (offset == 0 && size == bio->bi_iter.bi_size)
1864 return;
1865
1866 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
1867
1868 bio_advance(bio, offset << 9);
1869
1870 bio->bi_iter.bi_size = size;
1871 }
1872 EXPORT_SYMBOL_GPL(bio_trim);
1873
1874 /*
1875 * create memory pools for biovec's in a bio_set.
1876 * use the global biovec slabs created for general use.
1877 */
1878 mempool_t *biovec_create_pool(int pool_entries)
1879 {
1880 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1881
1882 return mempool_create_slab_pool(pool_entries, bp->slab);
1883 }
1884
1885 void bioset_free(struct bio_set *bs)
1886 {
1887 if (bs->rescue_workqueue)
1888 destroy_workqueue(bs->rescue_workqueue);
1889
1890 if (bs->bio_pool)
1891 mempool_destroy(bs->bio_pool);
1892
1893 if (bs->bvec_pool)
1894 mempool_destroy(bs->bvec_pool);
1895
1896 bioset_integrity_free(bs);
1897 bio_put_slab(bs);
1898
1899 kfree(bs);
1900 }
1901 EXPORT_SYMBOL(bioset_free);
1902
1903 /**
1904 * bioset_create - Create a bio_set
1905 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1906 * @front_pad: Number of bytes to allocate in front of the returned bio
1907 *
1908 * Description:
1909 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1910 * to ask for a number of bytes to be allocated in front of the bio.
1911 * Front pad allocation is useful for embedding the bio inside
1912 * another structure, to avoid allocating extra data to go with the bio.
1913 * Note that the bio must be embedded at the END of that structure always,
1914 * or things will break badly.
1915 */
1916 struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1917 {
1918 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1919 struct bio_set *bs;
1920
1921 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1922 if (!bs)
1923 return NULL;
1924
1925 bs->front_pad = front_pad;
1926
1927 spin_lock_init(&bs->rescue_lock);
1928 bio_list_init(&bs->rescue_list);
1929 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1930
1931 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1932 if (!bs->bio_slab) {
1933 kfree(bs);
1934 return NULL;
1935 }
1936
1937 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1938 if (!bs->bio_pool)
1939 goto bad;
1940
1941 bs->bvec_pool = biovec_create_pool(pool_size);
1942 if (!bs->bvec_pool)
1943 goto bad;
1944
1945 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1946 if (!bs->rescue_workqueue)
1947 goto bad;
1948
1949 return bs;
1950 bad:
1951 bioset_free(bs);
1952 return NULL;
1953 }
1954 EXPORT_SYMBOL(bioset_create);
1955
1956 #ifdef CONFIG_BLK_CGROUP
1957 /**
1958 * bio_associate_current - associate a bio with %current
1959 * @bio: target bio
1960 *
1961 * Associate @bio with %current if it hasn't been associated yet. Block
1962 * layer will treat @bio as if it were issued by %current no matter which
1963 * task actually issues it.
1964 *
1965 * This function takes an extra reference of @task's io_context and blkcg
1966 * which will be put when @bio is released. The caller must own @bio,
1967 * ensure %current->io_context exists, and is responsible for synchronizing
1968 * calls to this function.
1969 */
1970 int bio_associate_current(struct bio *bio)
1971 {
1972 struct io_context *ioc;
1973 struct cgroup_subsys_state *css;
1974
1975 if (bio->bi_ioc)
1976 return -EBUSY;
1977
1978 ioc = current->io_context;
1979 if (!ioc)
1980 return -ENOENT;
1981
1982 /* acquire active ref on @ioc and associate */
1983 get_io_context_active(ioc);
1984 bio->bi_ioc = ioc;
1985
1986 /* associate blkcg if exists */
1987 rcu_read_lock();
1988 css = task_css(current, blkio_cgrp_id);
1989 if (css && css_tryget_online(css))
1990 bio->bi_css = css;
1991 rcu_read_unlock();
1992
1993 return 0;
1994 }
1995
1996 /**
1997 * bio_disassociate_task - undo bio_associate_current()
1998 * @bio: target bio
1999 */
2000 void bio_disassociate_task(struct bio *bio)
2001 {
2002 if (bio->bi_ioc) {
2003 put_io_context(bio->bi_ioc);
2004 bio->bi_ioc = NULL;
2005 }
2006 if (bio->bi_css) {
2007 css_put(bio->bi_css);
2008 bio->bi_css = NULL;
2009 }
2010 }
2011
2012 #endif /* CONFIG_BLK_CGROUP */
2013
2014 static void __init biovec_init_slabs(void)
2015 {
2016 int i;
2017
2018 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
2019 int size;
2020 struct biovec_slab *bvs = bvec_slabs + i;
2021
2022 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2023 bvs->slab = NULL;
2024 continue;
2025 }
2026
2027 size = bvs->nr_vecs * sizeof(struct bio_vec);
2028 bvs->slab = kmem_cache_create(bvs->name, size, 0,
2029 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2030 }
2031 }
2032
2033 static int __init init_bio(void)
2034 {
2035 bio_slab_max = 2;
2036 bio_slab_nr = 0;
2037 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
2038 if (!bio_slabs)
2039 panic("bio: can't allocate bios\n");
2040
2041 bio_integrity_init();
2042 biovec_init_slabs();
2043
2044 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
2045 if (!fs_bio_set)
2046 panic("bio: can't allocate bios\n");
2047
2048 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
2049 panic("bio: can't create integrity pool\n");
2050
2051 return 0;
2052 }
2053 subsys_initcall(init_bio);