]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - block/bio.c
3d4a072375efdc8bbf5fd4e7fd17b0f57a97f00f
[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 /* should not use nobvec bioset for nr_iovecs > 0 */
432 if (WARN_ON_ONCE(!bs->bvec_pool && nr_iovecs > 0))
433 return NULL;
434 /*
435 * generic_make_request() converts recursion to iteration; this
436 * means if we're running beneath it, any bios we allocate and
437 * submit will not be submitted (and thus freed) until after we
438 * return.
439 *
440 * This exposes us to a potential deadlock if we allocate
441 * multiple bios from the same bio_set() while running
442 * underneath generic_make_request(). If we were to allocate
443 * multiple bios (say a stacking block driver that was splitting
444 * bios), we would deadlock if we exhausted the mempool's
445 * reserve.
446 *
447 * We solve this, and guarantee forward progress, with a rescuer
448 * workqueue per bio_set. If we go to allocate and there are
449 * bios on current->bio_list, we first try the allocation
450 * without __GFP_WAIT; if that fails, we punt those bios we
451 * would be blocking to the rescuer workqueue before we retry
452 * with the original gfp_flags.
453 */
454
455 if (current->bio_list && !bio_list_empty(current->bio_list))
456 gfp_mask &= ~__GFP_WAIT;
457
458 p = mempool_alloc(bs->bio_pool, gfp_mask);
459 if (!p && gfp_mask != saved_gfp) {
460 punt_bios_to_rescuer(bs);
461 gfp_mask = saved_gfp;
462 p = mempool_alloc(bs->bio_pool, gfp_mask);
463 }
464
465 front_pad = bs->front_pad;
466 inline_vecs = BIO_INLINE_VECS;
467 }
468
469 if (unlikely(!p))
470 return NULL;
471
472 bio = p + front_pad;
473 bio_init(bio);
474
475 if (nr_iovecs > inline_vecs) {
476 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
477 if (!bvl && gfp_mask != saved_gfp) {
478 punt_bios_to_rescuer(bs);
479 gfp_mask = saved_gfp;
480 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
481 }
482
483 if (unlikely(!bvl))
484 goto err_free;
485
486 bio->bi_flags |= 1 << BIO_OWNS_VEC;
487 } else if (nr_iovecs) {
488 bvl = bio->bi_inline_vecs;
489 }
490
491 bio->bi_pool = bs;
492 bio->bi_flags |= idx << BIO_POOL_OFFSET;
493 bio->bi_max_vecs = nr_iovecs;
494 bio->bi_io_vec = bvl;
495 return bio;
496
497 err_free:
498 mempool_free(p, bs->bio_pool);
499 return NULL;
500 }
501 EXPORT_SYMBOL(bio_alloc_bioset);
502
503 void zero_fill_bio(struct bio *bio)
504 {
505 unsigned long flags;
506 struct bio_vec bv;
507 struct bvec_iter iter;
508
509 bio_for_each_segment(bv, bio, iter) {
510 char *data = bvec_kmap_irq(&bv, &flags);
511 memset(data, 0, bv.bv_len);
512 flush_dcache_page(bv.bv_page);
513 bvec_kunmap_irq(data, &flags);
514 }
515 }
516 EXPORT_SYMBOL(zero_fill_bio);
517
518 /**
519 * bio_put - release a reference to a bio
520 * @bio: bio to release reference to
521 *
522 * Description:
523 * Put a reference to a &struct bio, either one you have gotten with
524 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
525 **/
526 void bio_put(struct bio *bio)
527 {
528 BIO_BUG_ON(!atomic_read(&bio->bi_cnt));
529
530 /*
531 * last put frees it
532 */
533 if (atomic_dec_and_test(&bio->bi_cnt))
534 bio_free(bio);
535 }
536 EXPORT_SYMBOL(bio_put);
537
538 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
539 {
540 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
541 blk_recount_segments(q, bio);
542
543 return bio->bi_phys_segments;
544 }
545 EXPORT_SYMBOL(bio_phys_segments);
546
547 /**
548 * __bio_clone_fast - clone a bio that shares the original bio's biovec
549 * @bio: destination bio
550 * @bio_src: bio to clone
551 *
552 * Clone a &bio. Caller will own the returned bio, but not
553 * the actual data it points to. Reference count of returned
554 * bio will be one.
555 *
556 * Caller must ensure that @bio_src is not freed before @bio.
557 */
558 void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
559 {
560 BUG_ON(bio->bi_pool && BIO_POOL_IDX(bio) != BIO_POOL_NONE);
561
562 /*
563 * most users will be overriding ->bi_bdev with a new target,
564 * so we don't set nor calculate new physical/hw segment counts here
565 */
566 bio->bi_bdev = bio_src->bi_bdev;
567 bio->bi_flags |= 1 << BIO_CLONED;
568 bio->bi_rw = bio_src->bi_rw;
569 bio->bi_iter = bio_src->bi_iter;
570 bio->bi_io_vec = bio_src->bi_io_vec;
571 }
572 EXPORT_SYMBOL(__bio_clone_fast);
573
574 /**
575 * bio_clone_fast - clone a bio that shares the original bio's biovec
576 * @bio: bio to clone
577 * @gfp_mask: allocation priority
578 * @bs: bio_set to allocate from
579 *
580 * Like __bio_clone_fast, only also allocates the returned bio
581 */
582 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
583 {
584 struct bio *b;
585
586 b = bio_alloc_bioset(gfp_mask, 0, bs);
587 if (!b)
588 return NULL;
589
590 __bio_clone_fast(b, bio);
591
592 if (bio_integrity(bio)) {
593 int ret;
594
595 ret = bio_integrity_clone(b, bio, gfp_mask);
596
597 if (ret < 0) {
598 bio_put(b);
599 return NULL;
600 }
601 }
602
603 return b;
604 }
605 EXPORT_SYMBOL(bio_clone_fast);
606
607 /**
608 * bio_clone_bioset - clone a bio
609 * @bio_src: bio to clone
610 * @gfp_mask: allocation priority
611 * @bs: bio_set to allocate from
612 *
613 * Clone bio. Caller will own the returned bio, but not the actual data it
614 * points to. Reference count of returned bio will be one.
615 */
616 struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask,
617 struct bio_set *bs)
618 {
619 struct bvec_iter iter;
620 struct bio_vec bv;
621 struct bio *bio;
622
623 /*
624 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
625 * bio_src->bi_io_vec to bio->bi_io_vec.
626 *
627 * We can't do that anymore, because:
628 *
629 * - The point of cloning the biovec is to produce a bio with a biovec
630 * the caller can modify: bi_idx and bi_bvec_done should be 0.
631 *
632 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
633 * we tried to clone the whole thing bio_alloc_bioset() would fail.
634 * But the clone should succeed as long as the number of biovecs we
635 * actually need to allocate is fewer than BIO_MAX_PAGES.
636 *
637 * - Lastly, bi_vcnt should not be looked at or relied upon by code
638 * that does not own the bio - reason being drivers don't use it for
639 * iterating over the biovec anymore, so expecting it to be kept up
640 * to date (i.e. for clones that share the parent biovec) is just
641 * asking for trouble and would force extra work on
642 * __bio_clone_fast() anyways.
643 */
644
645 bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs);
646 if (!bio)
647 return NULL;
648
649 bio->bi_bdev = bio_src->bi_bdev;
650 bio->bi_rw = bio_src->bi_rw;
651 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector;
652 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
653
654 if (bio->bi_rw & REQ_DISCARD)
655 goto integrity_clone;
656
657 if (bio->bi_rw & REQ_WRITE_SAME) {
658 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
659 goto integrity_clone;
660 }
661
662 bio_for_each_segment(bv, bio_src, iter)
663 bio->bi_io_vec[bio->bi_vcnt++] = bv;
664
665 integrity_clone:
666 if (bio_integrity(bio_src)) {
667 int ret;
668
669 ret = bio_integrity_clone(bio, bio_src, gfp_mask);
670 if (ret < 0) {
671 bio_put(bio);
672 return NULL;
673 }
674 }
675
676 return bio;
677 }
678 EXPORT_SYMBOL(bio_clone_bioset);
679
680 /**
681 * bio_get_nr_vecs - return approx number of vecs
682 * @bdev: I/O target
683 *
684 * Return the approximate number of pages we can send to this target.
685 * There's no guarantee that you will be able to fit this number of pages
686 * into a bio, it does not account for dynamic restrictions that vary
687 * on offset.
688 */
689 int bio_get_nr_vecs(struct block_device *bdev)
690 {
691 struct request_queue *q = bdev_get_queue(bdev);
692 int nr_pages;
693
694 nr_pages = min_t(unsigned,
695 queue_max_segments(q),
696 queue_max_sectors(q) / (PAGE_SIZE >> 9) + 1);
697
698 return min_t(unsigned, nr_pages, BIO_MAX_PAGES);
699
700 }
701 EXPORT_SYMBOL(bio_get_nr_vecs);
702
703 static int __bio_add_page(struct request_queue *q, struct bio *bio, struct page
704 *page, unsigned int len, unsigned int offset,
705 unsigned int max_sectors)
706 {
707 int retried_segments = 0;
708 struct bio_vec *bvec;
709
710 /*
711 * cloned bio must not modify vec list
712 */
713 if (unlikely(bio_flagged(bio, BIO_CLONED)))
714 return 0;
715
716 if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors)
717 return 0;
718
719 /*
720 * For filesystems with a blocksize smaller than the pagesize
721 * we will often be called with the same page as last time and
722 * a consecutive offset. Optimize this special case.
723 */
724 if (bio->bi_vcnt > 0) {
725 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
726
727 if (page == prev->bv_page &&
728 offset == prev->bv_offset + prev->bv_len) {
729 unsigned int prev_bv_len = prev->bv_len;
730 prev->bv_len += len;
731
732 if (q->merge_bvec_fn) {
733 struct bvec_merge_data bvm = {
734 /* prev_bvec is already charged in
735 bi_size, discharge it in order to
736 simulate merging updated prev_bvec
737 as new bvec. */
738 .bi_bdev = bio->bi_bdev,
739 .bi_sector = bio->bi_iter.bi_sector,
740 .bi_size = bio->bi_iter.bi_size -
741 prev_bv_len,
742 .bi_rw = bio->bi_rw,
743 };
744
745 if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) {
746 prev->bv_len -= len;
747 return 0;
748 }
749 }
750
751 goto done;
752 }
753
754 /*
755 * If the queue doesn't support SG gaps and adding this
756 * offset would create a gap, disallow it.
757 */
758 if (q->queue_flags & (1 << QUEUE_FLAG_SG_GAPS) &&
759 bvec_gap_to_prev(prev, offset))
760 return 0;
761 }
762
763 if (bio->bi_vcnt >= bio->bi_max_vecs)
764 return 0;
765
766 /*
767 * we might lose a segment or two here, but rather that than
768 * make this too complex.
769 */
770
771 while (bio->bi_phys_segments >= queue_max_segments(q)) {
772
773 if (retried_segments)
774 return 0;
775
776 retried_segments = 1;
777 blk_recount_segments(q, bio);
778 }
779
780 /*
781 * setup the new entry, we might clear it again later if we
782 * cannot add the page
783 */
784 bvec = &bio->bi_io_vec[bio->bi_vcnt];
785 bvec->bv_page = page;
786 bvec->bv_len = len;
787 bvec->bv_offset = offset;
788
789 /*
790 * if queue has other restrictions (eg varying max sector size
791 * depending on offset), it can specify a merge_bvec_fn in the
792 * queue to get further control
793 */
794 if (q->merge_bvec_fn) {
795 struct bvec_merge_data bvm = {
796 .bi_bdev = bio->bi_bdev,
797 .bi_sector = bio->bi_iter.bi_sector,
798 .bi_size = bio->bi_iter.bi_size,
799 .bi_rw = bio->bi_rw,
800 };
801
802 /*
803 * merge_bvec_fn() returns number of bytes it can accept
804 * at this offset
805 */
806 if (q->merge_bvec_fn(q, &bvm, bvec) < bvec->bv_len) {
807 bvec->bv_page = NULL;
808 bvec->bv_len = 0;
809 bvec->bv_offset = 0;
810 return 0;
811 }
812 }
813
814 /* If we may be able to merge these biovecs, force a recount */
815 if (bio->bi_vcnt && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
816 bio->bi_flags &= ~(1 << BIO_SEG_VALID);
817
818 bio->bi_vcnt++;
819 bio->bi_phys_segments++;
820 done:
821 bio->bi_iter.bi_size += len;
822 return len;
823 }
824
825 /**
826 * bio_add_pc_page - attempt to add page to bio
827 * @q: the target queue
828 * @bio: destination bio
829 * @page: page to add
830 * @len: vec entry length
831 * @offset: vec entry offset
832 *
833 * Attempt to add a page to the bio_vec maplist. This can fail for a
834 * number of reasons, such as the bio being full or target block device
835 * limitations. The target block device must allow bio's up to PAGE_SIZE,
836 * so it is always possible to add a single page to an empty bio.
837 *
838 * This should only be used by REQ_PC bios.
839 */
840 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page *page,
841 unsigned int len, unsigned int offset)
842 {
843 return __bio_add_page(q, bio, page, len, offset,
844 queue_max_hw_sectors(q));
845 }
846 EXPORT_SYMBOL(bio_add_pc_page);
847
848 /**
849 * bio_add_page - attempt to add page to bio
850 * @bio: destination bio
851 * @page: page to add
852 * @len: vec entry length
853 * @offset: vec entry offset
854 *
855 * Attempt to add a page to the bio_vec maplist. This can fail for a
856 * number of reasons, such as the bio being full or target block device
857 * limitations. The target block device must allow bio's up to PAGE_SIZE,
858 * so it is always possible to add a single page to an empty bio.
859 */
860 int bio_add_page(struct bio *bio, struct page *page, unsigned int len,
861 unsigned int offset)
862 {
863 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
864 unsigned int max_sectors;
865
866 max_sectors = blk_max_size_offset(q, bio->bi_iter.bi_sector);
867 if ((max_sectors < (len >> 9)) && !bio->bi_iter.bi_size)
868 max_sectors = len >> 9;
869
870 return __bio_add_page(q, bio, page, len, offset, max_sectors);
871 }
872 EXPORT_SYMBOL(bio_add_page);
873
874 struct submit_bio_ret {
875 struct completion event;
876 int error;
877 };
878
879 static void submit_bio_wait_endio(struct bio *bio, int error)
880 {
881 struct submit_bio_ret *ret = bio->bi_private;
882
883 ret->error = error;
884 complete(&ret->event);
885 }
886
887 /**
888 * submit_bio_wait - submit a bio, and wait until it completes
889 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
890 * @bio: The &struct bio which describes the I/O
891 *
892 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
893 * bio_endio() on failure.
894 */
895 int submit_bio_wait(int rw, struct bio *bio)
896 {
897 struct submit_bio_ret ret;
898
899 rw |= REQ_SYNC;
900 init_completion(&ret.event);
901 bio->bi_private = &ret;
902 bio->bi_end_io = submit_bio_wait_endio;
903 submit_bio(rw, bio);
904 wait_for_completion(&ret.event);
905
906 return ret.error;
907 }
908 EXPORT_SYMBOL(submit_bio_wait);
909
910 /**
911 * bio_advance - increment/complete a bio by some number of bytes
912 * @bio: bio to advance
913 * @bytes: number of bytes to complete
914 *
915 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
916 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
917 * be updated on the last bvec as well.
918 *
919 * @bio will then represent the remaining, uncompleted portion of the io.
920 */
921 void bio_advance(struct bio *bio, unsigned bytes)
922 {
923 if (bio_integrity(bio))
924 bio_integrity_advance(bio, bytes);
925
926 bio_advance_iter(bio, &bio->bi_iter, bytes);
927 }
928 EXPORT_SYMBOL(bio_advance);
929
930 /**
931 * bio_alloc_pages - allocates a single page for each bvec in a bio
932 * @bio: bio to allocate pages for
933 * @gfp_mask: flags for allocation
934 *
935 * Allocates pages up to @bio->bi_vcnt.
936 *
937 * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
938 * freed.
939 */
940 int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask)
941 {
942 int i;
943 struct bio_vec *bv;
944
945 bio_for_each_segment_all(bv, bio, i) {
946 bv->bv_page = alloc_page(gfp_mask);
947 if (!bv->bv_page) {
948 while (--bv >= bio->bi_io_vec)
949 __free_page(bv->bv_page);
950 return -ENOMEM;
951 }
952 }
953
954 return 0;
955 }
956 EXPORT_SYMBOL(bio_alloc_pages);
957
958 /**
959 * bio_copy_data - copy contents of data buffers from one chain of bios to
960 * another
961 * @src: source bio list
962 * @dst: destination bio list
963 *
964 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
965 * @src and @dst as linked lists of bios.
966 *
967 * Stops when it reaches the end of either @src or @dst - that is, copies
968 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
969 */
970 void bio_copy_data(struct bio *dst, struct bio *src)
971 {
972 struct bvec_iter src_iter, dst_iter;
973 struct bio_vec src_bv, dst_bv;
974 void *src_p, *dst_p;
975 unsigned bytes;
976
977 src_iter = src->bi_iter;
978 dst_iter = dst->bi_iter;
979
980 while (1) {
981 if (!src_iter.bi_size) {
982 src = src->bi_next;
983 if (!src)
984 break;
985
986 src_iter = src->bi_iter;
987 }
988
989 if (!dst_iter.bi_size) {
990 dst = dst->bi_next;
991 if (!dst)
992 break;
993
994 dst_iter = dst->bi_iter;
995 }
996
997 src_bv = bio_iter_iovec(src, src_iter);
998 dst_bv = bio_iter_iovec(dst, dst_iter);
999
1000 bytes = min(src_bv.bv_len, dst_bv.bv_len);
1001
1002 src_p = kmap_atomic(src_bv.bv_page);
1003 dst_p = kmap_atomic(dst_bv.bv_page);
1004
1005 memcpy(dst_p + dst_bv.bv_offset,
1006 src_p + src_bv.bv_offset,
1007 bytes);
1008
1009 kunmap_atomic(dst_p);
1010 kunmap_atomic(src_p);
1011
1012 bio_advance_iter(src, &src_iter, bytes);
1013 bio_advance_iter(dst, &dst_iter, bytes);
1014 }
1015 }
1016 EXPORT_SYMBOL(bio_copy_data);
1017
1018 struct bio_map_data {
1019 int nr_sgvecs;
1020 int is_our_pages;
1021 struct sg_iovec sgvecs[];
1022 };
1023
1024 static void bio_set_map_data(struct bio_map_data *bmd, struct bio *bio,
1025 const struct sg_iovec *iov, int iov_count,
1026 int is_our_pages)
1027 {
1028 memcpy(bmd->sgvecs, iov, sizeof(struct sg_iovec) * iov_count);
1029 bmd->nr_sgvecs = iov_count;
1030 bmd->is_our_pages = is_our_pages;
1031 bio->bi_private = bmd;
1032 }
1033
1034 static struct bio_map_data *bio_alloc_map_data(unsigned int iov_count,
1035 gfp_t gfp_mask)
1036 {
1037 if (iov_count > UIO_MAXIOV)
1038 return NULL;
1039
1040 return kmalloc(sizeof(struct bio_map_data) +
1041 sizeof(struct sg_iovec) * iov_count, gfp_mask);
1042 }
1043
1044 static int __bio_copy_iov(struct bio *bio, const struct sg_iovec *iov, int iov_count,
1045 int to_user, int from_user, int do_free_page)
1046 {
1047 int ret = 0, i;
1048 struct bio_vec *bvec;
1049 int iov_idx = 0;
1050 unsigned int iov_off = 0;
1051
1052 bio_for_each_segment_all(bvec, bio, i) {
1053 char *bv_addr = page_address(bvec->bv_page);
1054 unsigned int bv_len = bvec->bv_len;
1055
1056 while (bv_len && iov_idx < iov_count) {
1057 unsigned int bytes;
1058 char __user *iov_addr;
1059
1060 bytes = min_t(unsigned int,
1061 iov[iov_idx].iov_len - iov_off, bv_len);
1062 iov_addr = iov[iov_idx].iov_base + iov_off;
1063
1064 if (!ret) {
1065 if (to_user)
1066 ret = copy_to_user(iov_addr, bv_addr,
1067 bytes);
1068
1069 if (from_user)
1070 ret = copy_from_user(bv_addr, iov_addr,
1071 bytes);
1072
1073 if (ret)
1074 ret = -EFAULT;
1075 }
1076
1077 bv_len -= bytes;
1078 bv_addr += bytes;
1079 iov_addr += bytes;
1080 iov_off += bytes;
1081
1082 if (iov[iov_idx].iov_len == iov_off) {
1083 iov_idx++;
1084 iov_off = 0;
1085 }
1086 }
1087
1088 if (do_free_page)
1089 __free_page(bvec->bv_page);
1090 }
1091
1092 return ret;
1093 }
1094
1095 /**
1096 * bio_uncopy_user - finish previously mapped bio
1097 * @bio: bio being terminated
1098 *
1099 * Free pages allocated from bio_copy_user() and write back data
1100 * to user space in case of a read.
1101 */
1102 int bio_uncopy_user(struct bio *bio)
1103 {
1104 struct bio_map_data *bmd = bio->bi_private;
1105 struct bio_vec *bvec;
1106 int ret = 0, i;
1107
1108 if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1109 /*
1110 * if we're in a workqueue, the request is orphaned, so
1111 * don't copy into a random user address space, just free.
1112 */
1113 if (current->mm)
1114 ret = __bio_copy_iov(bio, bmd->sgvecs, bmd->nr_sgvecs,
1115 bio_data_dir(bio) == READ,
1116 0, bmd->is_our_pages);
1117 else if (bmd->is_our_pages)
1118 bio_for_each_segment_all(bvec, bio, i)
1119 __free_page(bvec->bv_page);
1120 }
1121 kfree(bmd);
1122 bio_put(bio);
1123 return ret;
1124 }
1125 EXPORT_SYMBOL(bio_uncopy_user);
1126
1127 /**
1128 * bio_copy_user_iov - copy user data to bio
1129 * @q: destination block queue
1130 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1131 * @iov: the iovec.
1132 * @iov_count: number of elements in the iovec
1133 * @write_to_vm: bool indicating writing to pages or not
1134 * @gfp_mask: memory allocation flags
1135 *
1136 * Prepares and returns a bio for indirect user io, bouncing data
1137 * to/from kernel pages as necessary. Must be paired with
1138 * call bio_uncopy_user() on io completion.
1139 */
1140 struct bio *bio_copy_user_iov(struct request_queue *q,
1141 struct rq_map_data *map_data,
1142 const struct sg_iovec *iov, int iov_count,
1143 int write_to_vm, gfp_t gfp_mask)
1144 {
1145 struct bio_map_data *bmd;
1146 struct bio_vec *bvec;
1147 struct page *page;
1148 struct bio *bio;
1149 int i, ret;
1150 int nr_pages = 0;
1151 unsigned int len = 0;
1152 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
1153
1154 for (i = 0; i < iov_count; i++) {
1155 unsigned long uaddr;
1156 unsigned long end;
1157 unsigned long start;
1158
1159 uaddr = (unsigned long)iov[i].iov_base;
1160 end = (uaddr + iov[i].iov_len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1161 start = uaddr >> PAGE_SHIFT;
1162
1163 /*
1164 * Overflow, abort
1165 */
1166 if (end < start)
1167 return ERR_PTR(-EINVAL);
1168
1169 nr_pages += end - start;
1170 len += iov[i].iov_len;
1171 }
1172
1173 if (offset)
1174 nr_pages++;
1175
1176 bmd = bio_alloc_map_data(iov_count, gfp_mask);
1177 if (!bmd)
1178 return ERR_PTR(-ENOMEM);
1179
1180 ret = -ENOMEM;
1181 bio = bio_kmalloc(gfp_mask, nr_pages);
1182 if (!bio)
1183 goto out_bmd;
1184
1185 if (!write_to_vm)
1186 bio->bi_rw |= REQ_WRITE;
1187
1188 ret = 0;
1189
1190 if (map_data) {
1191 nr_pages = 1 << map_data->page_order;
1192 i = map_data->offset / PAGE_SIZE;
1193 }
1194 while (len) {
1195 unsigned int bytes = PAGE_SIZE;
1196
1197 bytes -= offset;
1198
1199 if (bytes > len)
1200 bytes = len;
1201
1202 if (map_data) {
1203 if (i == map_data->nr_entries * nr_pages) {
1204 ret = -ENOMEM;
1205 break;
1206 }
1207
1208 page = map_data->pages[i / nr_pages];
1209 page += (i % nr_pages);
1210
1211 i++;
1212 } else {
1213 page = alloc_page(q->bounce_gfp | gfp_mask);
1214 if (!page) {
1215 ret = -ENOMEM;
1216 break;
1217 }
1218 }
1219
1220 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
1221 break;
1222
1223 len -= bytes;
1224 offset = 0;
1225 }
1226
1227 if (ret)
1228 goto cleanup;
1229
1230 /*
1231 * success
1232 */
1233 if ((!write_to_vm && (!map_data || !map_data->null_mapped)) ||
1234 (map_data && map_data->from_user)) {
1235 ret = __bio_copy_iov(bio, iov, iov_count, 0, 1, 0);
1236 if (ret)
1237 goto cleanup;
1238 }
1239
1240 bio_set_map_data(bmd, bio, iov, iov_count, map_data ? 0 : 1);
1241 return bio;
1242 cleanup:
1243 if (!map_data)
1244 bio_for_each_segment_all(bvec, bio, i)
1245 __free_page(bvec->bv_page);
1246
1247 bio_put(bio);
1248 out_bmd:
1249 kfree(bmd);
1250 return ERR_PTR(ret);
1251 }
1252
1253 /**
1254 * bio_copy_user - copy user data to bio
1255 * @q: destination block queue
1256 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1257 * @uaddr: start of user address
1258 * @len: length in bytes
1259 * @write_to_vm: bool indicating writing to pages or not
1260 * @gfp_mask: memory allocation flags
1261 *
1262 * Prepares and returns a bio for indirect user io, bouncing data
1263 * to/from kernel pages as necessary. Must be paired with
1264 * call bio_uncopy_user() on io completion.
1265 */
1266 struct bio *bio_copy_user(struct request_queue *q, struct rq_map_data *map_data,
1267 unsigned long uaddr, unsigned int len,
1268 int write_to_vm, gfp_t gfp_mask)
1269 {
1270 struct sg_iovec iov;
1271
1272 iov.iov_base = (void __user *)uaddr;
1273 iov.iov_len = len;
1274
1275 return bio_copy_user_iov(q, map_data, &iov, 1, write_to_vm, gfp_mask);
1276 }
1277 EXPORT_SYMBOL(bio_copy_user);
1278
1279 static struct bio *__bio_map_user_iov(struct request_queue *q,
1280 struct block_device *bdev,
1281 const struct sg_iovec *iov, int iov_count,
1282 int write_to_vm, gfp_t gfp_mask)
1283 {
1284 int i, j;
1285 int nr_pages = 0;
1286 struct page **pages;
1287 struct bio *bio;
1288 int cur_page = 0;
1289 int ret, offset;
1290
1291 for (i = 0; i < iov_count; i++) {
1292 unsigned long uaddr = (unsigned long)iov[i].iov_base;
1293 unsigned long len = iov[i].iov_len;
1294 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1295 unsigned long start = uaddr >> PAGE_SHIFT;
1296
1297 /*
1298 * Overflow, abort
1299 */
1300 if (end < start)
1301 return ERR_PTR(-EINVAL);
1302
1303 nr_pages += end - start;
1304 /*
1305 * buffer must be aligned to at least hardsector size for now
1306 */
1307 if (uaddr & queue_dma_alignment(q))
1308 return ERR_PTR(-EINVAL);
1309 }
1310
1311 if (!nr_pages)
1312 return ERR_PTR(-EINVAL);
1313
1314 bio = bio_kmalloc(gfp_mask, nr_pages);
1315 if (!bio)
1316 return ERR_PTR(-ENOMEM);
1317
1318 ret = -ENOMEM;
1319 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
1320 if (!pages)
1321 goto out;
1322
1323 for (i = 0; i < iov_count; i++) {
1324 unsigned long uaddr = (unsigned long)iov[i].iov_base;
1325 unsigned long len = iov[i].iov_len;
1326 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1327 unsigned long start = uaddr >> PAGE_SHIFT;
1328 const int local_nr_pages = end - start;
1329 const int page_limit = cur_page + local_nr_pages;
1330
1331 ret = get_user_pages_fast(uaddr, local_nr_pages,
1332 write_to_vm, &pages[cur_page]);
1333 if (ret < local_nr_pages) {
1334 ret = -EFAULT;
1335 goto out_unmap;
1336 }
1337
1338 offset = uaddr & ~PAGE_MASK;
1339 for (j = cur_page; j < page_limit; j++) {
1340 unsigned int bytes = PAGE_SIZE - offset;
1341
1342 if (len <= 0)
1343 break;
1344
1345 if (bytes > len)
1346 bytes = len;
1347
1348 /*
1349 * sorry...
1350 */
1351 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1352 bytes)
1353 break;
1354
1355 len -= bytes;
1356 offset = 0;
1357 }
1358
1359 cur_page = j;
1360 /*
1361 * release the pages we didn't map into the bio, if any
1362 */
1363 while (j < page_limit)
1364 page_cache_release(pages[j++]);
1365 }
1366
1367 kfree(pages);
1368
1369 /*
1370 * set data direction, and check if mapped pages need bouncing
1371 */
1372 if (!write_to_vm)
1373 bio->bi_rw |= REQ_WRITE;
1374
1375 bio->bi_bdev = bdev;
1376 bio->bi_flags |= (1 << BIO_USER_MAPPED);
1377 return bio;
1378
1379 out_unmap:
1380 for (i = 0; i < nr_pages; i++) {
1381 if(!pages[i])
1382 break;
1383 page_cache_release(pages[i]);
1384 }
1385 out:
1386 kfree(pages);
1387 bio_put(bio);
1388 return ERR_PTR(ret);
1389 }
1390
1391 /**
1392 * bio_map_user - map user address into bio
1393 * @q: the struct request_queue for the bio
1394 * @bdev: destination block device
1395 * @uaddr: start of user address
1396 * @len: length in bytes
1397 * @write_to_vm: bool indicating writing to pages or not
1398 * @gfp_mask: memory allocation flags
1399 *
1400 * Map the user space address into a bio suitable for io to a block
1401 * device. Returns an error pointer in case of error.
1402 */
1403 struct bio *bio_map_user(struct request_queue *q, struct block_device *bdev,
1404 unsigned long uaddr, unsigned int len, int write_to_vm,
1405 gfp_t gfp_mask)
1406 {
1407 struct sg_iovec iov;
1408
1409 iov.iov_base = (void __user *)uaddr;
1410 iov.iov_len = len;
1411
1412 return bio_map_user_iov(q, bdev, &iov, 1, write_to_vm, gfp_mask);
1413 }
1414 EXPORT_SYMBOL(bio_map_user);
1415
1416 /**
1417 * bio_map_user_iov - map user sg_iovec table into bio
1418 * @q: the struct request_queue for the bio
1419 * @bdev: destination block device
1420 * @iov: the iovec.
1421 * @iov_count: number of elements in the iovec
1422 * @write_to_vm: bool indicating writing to pages or not
1423 * @gfp_mask: memory allocation flags
1424 *
1425 * Map the user space address into a bio suitable for io to a block
1426 * device. Returns an error pointer in case of error.
1427 */
1428 struct bio *bio_map_user_iov(struct request_queue *q, struct block_device *bdev,
1429 const struct sg_iovec *iov, int iov_count,
1430 int write_to_vm, gfp_t gfp_mask)
1431 {
1432 struct bio *bio;
1433
1434 bio = __bio_map_user_iov(q, bdev, iov, iov_count, write_to_vm,
1435 gfp_mask);
1436 if (IS_ERR(bio))
1437 return bio;
1438
1439 /*
1440 * subtle -- if __bio_map_user() ended up bouncing a bio,
1441 * it would normally disappear when its bi_end_io is run.
1442 * however, we need it for the unmap, so grab an extra
1443 * reference to it
1444 */
1445 bio_get(bio);
1446
1447 return bio;
1448 }
1449
1450 static void __bio_unmap_user(struct bio *bio)
1451 {
1452 struct bio_vec *bvec;
1453 int i;
1454
1455 /*
1456 * make sure we dirty pages we wrote to
1457 */
1458 bio_for_each_segment_all(bvec, bio, i) {
1459 if (bio_data_dir(bio) == READ)
1460 set_page_dirty_lock(bvec->bv_page);
1461
1462 page_cache_release(bvec->bv_page);
1463 }
1464
1465 bio_put(bio);
1466 }
1467
1468 /**
1469 * bio_unmap_user - unmap a bio
1470 * @bio: the bio being unmapped
1471 *
1472 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1473 * a process context.
1474 *
1475 * bio_unmap_user() may sleep.
1476 */
1477 void bio_unmap_user(struct bio *bio)
1478 {
1479 __bio_unmap_user(bio);
1480 bio_put(bio);
1481 }
1482 EXPORT_SYMBOL(bio_unmap_user);
1483
1484 static void bio_map_kern_endio(struct bio *bio, int err)
1485 {
1486 bio_put(bio);
1487 }
1488
1489 static struct bio *__bio_map_kern(struct request_queue *q, void *data,
1490 unsigned int len, gfp_t gfp_mask)
1491 {
1492 unsigned long kaddr = (unsigned long)data;
1493 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1494 unsigned long start = kaddr >> PAGE_SHIFT;
1495 const int nr_pages = end - start;
1496 int offset, i;
1497 struct bio *bio;
1498
1499 bio = bio_kmalloc(gfp_mask, nr_pages);
1500 if (!bio)
1501 return ERR_PTR(-ENOMEM);
1502
1503 offset = offset_in_page(kaddr);
1504 for (i = 0; i < nr_pages; i++) {
1505 unsigned int bytes = PAGE_SIZE - offset;
1506
1507 if (len <= 0)
1508 break;
1509
1510 if (bytes > len)
1511 bytes = len;
1512
1513 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1514 offset) < bytes)
1515 break;
1516
1517 data += bytes;
1518 len -= bytes;
1519 offset = 0;
1520 }
1521
1522 bio->bi_end_io = bio_map_kern_endio;
1523 return bio;
1524 }
1525
1526 /**
1527 * bio_map_kern - map kernel address into bio
1528 * @q: the struct request_queue for the bio
1529 * @data: pointer to buffer to map
1530 * @len: length in bytes
1531 * @gfp_mask: allocation flags for bio allocation
1532 *
1533 * Map the kernel address into a bio suitable for io to a block
1534 * device. Returns an error pointer in case of error.
1535 */
1536 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1537 gfp_t gfp_mask)
1538 {
1539 struct bio *bio;
1540
1541 bio = __bio_map_kern(q, data, len, gfp_mask);
1542 if (IS_ERR(bio))
1543 return bio;
1544
1545 if (bio->bi_iter.bi_size == len)
1546 return bio;
1547
1548 /*
1549 * Don't support partial mappings.
1550 */
1551 bio_put(bio);
1552 return ERR_PTR(-EINVAL);
1553 }
1554 EXPORT_SYMBOL(bio_map_kern);
1555
1556 static void bio_copy_kern_endio(struct bio *bio, int err)
1557 {
1558 struct bio_vec *bvec;
1559 const int read = bio_data_dir(bio) == READ;
1560 struct bio_map_data *bmd = bio->bi_private;
1561 int i;
1562 char *p = bmd->sgvecs[0].iov_base;
1563
1564 bio_for_each_segment_all(bvec, bio, i) {
1565 char *addr = page_address(bvec->bv_page);
1566
1567 if (read)
1568 memcpy(p, addr, bvec->bv_len);
1569
1570 __free_page(bvec->bv_page);
1571 p += bvec->bv_len;
1572 }
1573
1574 kfree(bmd);
1575 bio_put(bio);
1576 }
1577
1578 /**
1579 * bio_copy_kern - copy kernel address into bio
1580 * @q: the struct request_queue for the bio
1581 * @data: pointer to buffer to copy
1582 * @len: length in bytes
1583 * @gfp_mask: allocation flags for bio and page allocation
1584 * @reading: data direction is READ
1585 *
1586 * copy the kernel address into a bio suitable for io to a block
1587 * device. Returns an error pointer in case of error.
1588 */
1589 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1590 gfp_t gfp_mask, int reading)
1591 {
1592 struct bio *bio;
1593 struct bio_vec *bvec;
1594 int i;
1595
1596 bio = bio_copy_user(q, NULL, (unsigned long)data, len, 1, gfp_mask);
1597 if (IS_ERR(bio))
1598 return bio;
1599
1600 if (!reading) {
1601 void *p = data;
1602
1603 bio_for_each_segment_all(bvec, bio, i) {
1604 char *addr = page_address(bvec->bv_page);
1605
1606 memcpy(addr, p, bvec->bv_len);
1607 p += bvec->bv_len;
1608 }
1609 }
1610
1611 bio->bi_end_io = bio_copy_kern_endio;
1612
1613 return bio;
1614 }
1615 EXPORT_SYMBOL(bio_copy_kern);
1616
1617 /*
1618 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1619 * for performing direct-IO in BIOs.
1620 *
1621 * The problem is that we cannot run set_page_dirty() from interrupt context
1622 * because the required locks are not interrupt-safe. So what we can do is to
1623 * mark the pages dirty _before_ performing IO. And in interrupt context,
1624 * check that the pages are still dirty. If so, fine. If not, redirty them
1625 * in process context.
1626 *
1627 * We special-case compound pages here: normally this means reads into hugetlb
1628 * pages. The logic in here doesn't really work right for compound pages
1629 * because the VM does not uniformly chase down the head page in all cases.
1630 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1631 * handle them at all. So we skip compound pages here at an early stage.
1632 *
1633 * Note that this code is very hard to test under normal circumstances because
1634 * direct-io pins the pages with get_user_pages(). This makes
1635 * is_page_cache_freeable return false, and the VM will not clean the pages.
1636 * But other code (eg, flusher threads) could clean the pages if they are mapped
1637 * pagecache.
1638 *
1639 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1640 * deferred bio dirtying paths.
1641 */
1642
1643 /*
1644 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1645 */
1646 void bio_set_pages_dirty(struct bio *bio)
1647 {
1648 struct bio_vec *bvec;
1649 int i;
1650
1651 bio_for_each_segment_all(bvec, bio, i) {
1652 struct page *page = bvec->bv_page;
1653
1654 if (page && !PageCompound(page))
1655 set_page_dirty_lock(page);
1656 }
1657 }
1658
1659 static void bio_release_pages(struct bio *bio)
1660 {
1661 struct bio_vec *bvec;
1662 int i;
1663
1664 bio_for_each_segment_all(bvec, bio, i) {
1665 struct page *page = bvec->bv_page;
1666
1667 if (page)
1668 put_page(page);
1669 }
1670 }
1671
1672 /*
1673 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1674 * If they are, then fine. If, however, some pages are clean then they must
1675 * have been written out during the direct-IO read. So we take another ref on
1676 * the BIO and the offending pages and re-dirty the pages in process context.
1677 *
1678 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1679 * here on. It will run one page_cache_release() against each page and will
1680 * run one bio_put() against the BIO.
1681 */
1682
1683 static void bio_dirty_fn(struct work_struct *work);
1684
1685 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1686 static DEFINE_SPINLOCK(bio_dirty_lock);
1687 static struct bio *bio_dirty_list;
1688
1689 /*
1690 * This runs in process context
1691 */
1692 static void bio_dirty_fn(struct work_struct *work)
1693 {
1694 unsigned long flags;
1695 struct bio *bio;
1696
1697 spin_lock_irqsave(&bio_dirty_lock, flags);
1698 bio = bio_dirty_list;
1699 bio_dirty_list = NULL;
1700 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1701
1702 while (bio) {
1703 struct bio *next = bio->bi_private;
1704
1705 bio_set_pages_dirty(bio);
1706 bio_release_pages(bio);
1707 bio_put(bio);
1708 bio = next;
1709 }
1710 }
1711
1712 void bio_check_pages_dirty(struct bio *bio)
1713 {
1714 struct bio_vec *bvec;
1715 int nr_clean_pages = 0;
1716 int i;
1717
1718 bio_for_each_segment_all(bvec, bio, i) {
1719 struct page *page = bvec->bv_page;
1720
1721 if (PageDirty(page) || PageCompound(page)) {
1722 page_cache_release(page);
1723 bvec->bv_page = NULL;
1724 } else {
1725 nr_clean_pages++;
1726 }
1727 }
1728
1729 if (nr_clean_pages) {
1730 unsigned long flags;
1731
1732 spin_lock_irqsave(&bio_dirty_lock, flags);
1733 bio->bi_private = bio_dirty_list;
1734 bio_dirty_list = bio;
1735 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1736 schedule_work(&bio_dirty_work);
1737 } else {
1738 bio_put(bio);
1739 }
1740 }
1741
1742 void generic_start_io_acct(int rw, unsigned long sectors,
1743 struct hd_struct *part)
1744 {
1745 int cpu = part_stat_lock();
1746
1747 part_round_stats(cpu, part);
1748 part_stat_inc(cpu, part, ios[rw]);
1749 part_stat_add(cpu, part, sectors[rw], sectors);
1750 part_inc_in_flight(part, rw);
1751
1752 part_stat_unlock();
1753 }
1754 EXPORT_SYMBOL(generic_start_io_acct);
1755
1756 void generic_end_io_acct(int rw, struct hd_struct *part,
1757 unsigned long start_time)
1758 {
1759 unsigned long duration = jiffies - start_time;
1760 int cpu = part_stat_lock();
1761
1762 part_stat_add(cpu, part, ticks[rw], duration);
1763 part_round_stats(cpu, part);
1764 part_dec_in_flight(part, rw);
1765
1766 part_stat_unlock();
1767 }
1768 EXPORT_SYMBOL(generic_end_io_acct);
1769
1770 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1771 void bio_flush_dcache_pages(struct bio *bi)
1772 {
1773 struct bio_vec bvec;
1774 struct bvec_iter iter;
1775
1776 bio_for_each_segment(bvec, bi, iter)
1777 flush_dcache_page(bvec.bv_page);
1778 }
1779 EXPORT_SYMBOL(bio_flush_dcache_pages);
1780 #endif
1781
1782 /**
1783 * bio_endio - end I/O on a bio
1784 * @bio: bio
1785 * @error: error, if any
1786 *
1787 * Description:
1788 * bio_endio() will end I/O on the whole bio. bio_endio() is the
1789 * preferred way to end I/O on a bio, it takes care of clearing
1790 * BIO_UPTODATE on error. @error is 0 on success, and and one of the
1791 * established -Exxxx (-EIO, for instance) error values in case
1792 * something went wrong. No one should call bi_end_io() directly on a
1793 * bio unless they own it and thus know that it has an end_io
1794 * function.
1795 **/
1796 void bio_endio(struct bio *bio, int error)
1797 {
1798 while (bio) {
1799 BUG_ON(atomic_read(&bio->bi_remaining) <= 0);
1800
1801 if (error)
1802 clear_bit(BIO_UPTODATE, &bio->bi_flags);
1803 else if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
1804 error = -EIO;
1805
1806 if (!atomic_dec_and_test(&bio->bi_remaining))
1807 return;
1808
1809 /*
1810 * Need to have a real endio function for chained bios,
1811 * otherwise various corner cases will break (like stacking
1812 * block devices that save/restore bi_end_io) - however, we want
1813 * to avoid unbounded recursion and blowing the stack. Tail call
1814 * optimization would handle this, but compiling with frame
1815 * pointers also disables gcc's sibling call optimization.
1816 */
1817 if (bio->bi_end_io == bio_chain_endio) {
1818 struct bio *parent = bio->bi_private;
1819 bio_put(bio);
1820 bio = parent;
1821 } else {
1822 if (bio->bi_end_io)
1823 bio->bi_end_io(bio, error);
1824 bio = NULL;
1825 }
1826 }
1827 }
1828 EXPORT_SYMBOL(bio_endio);
1829
1830 /**
1831 * bio_endio_nodec - end I/O on a bio, without decrementing bi_remaining
1832 * @bio: bio
1833 * @error: error, if any
1834 *
1835 * For code that has saved and restored bi_end_io; thing hard before using this
1836 * function, probably you should've cloned the entire bio.
1837 **/
1838 void bio_endio_nodec(struct bio *bio, int error)
1839 {
1840 atomic_inc(&bio->bi_remaining);
1841 bio_endio(bio, error);
1842 }
1843 EXPORT_SYMBOL(bio_endio_nodec);
1844
1845 /**
1846 * bio_split - split a bio
1847 * @bio: bio to split
1848 * @sectors: number of sectors to split from the front of @bio
1849 * @gfp: gfp mask
1850 * @bs: bio set to allocate from
1851 *
1852 * Allocates and returns a new bio which represents @sectors from the start of
1853 * @bio, and updates @bio to represent the remaining sectors.
1854 *
1855 * The newly allocated bio will point to @bio's bi_io_vec; it is the caller's
1856 * responsibility to ensure that @bio is not freed before the split.
1857 */
1858 struct bio *bio_split(struct bio *bio, int sectors,
1859 gfp_t gfp, struct bio_set *bs)
1860 {
1861 struct bio *split = NULL;
1862
1863 BUG_ON(sectors <= 0);
1864 BUG_ON(sectors >= bio_sectors(bio));
1865
1866 split = bio_clone_fast(bio, gfp, bs);
1867 if (!split)
1868 return NULL;
1869
1870 split->bi_iter.bi_size = sectors << 9;
1871
1872 if (bio_integrity(split))
1873 bio_integrity_trim(split, 0, sectors);
1874
1875 bio_advance(bio, split->bi_iter.bi_size);
1876
1877 return split;
1878 }
1879 EXPORT_SYMBOL(bio_split);
1880
1881 /**
1882 * bio_trim - trim a bio
1883 * @bio: bio to trim
1884 * @offset: number of sectors to trim from the front of @bio
1885 * @size: size we want to trim @bio to, in sectors
1886 */
1887 void bio_trim(struct bio *bio, int offset, int size)
1888 {
1889 /* 'bio' is a cloned bio which we need to trim to match
1890 * the given offset and size.
1891 */
1892
1893 size <<= 9;
1894 if (offset == 0 && size == bio->bi_iter.bi_size)
1895 return;
1896
1897 clear_bit(BIO_SEG_VALID, &bio->bi_flags);
1898
1899 bio_advance(bio, offset << 9);
1900
1901 bio->bi_iter.bi_size = size;
1902 }
1903 EXPORT_SYMBOL_GPL(bio_trim);
1904
1905 /*
1906 * create memory pools for biovec's in a bio_set.
1907 * use the global biovec slabs created for general use.
1908 */
1909 mempool_t *biovec_create_pool(int pool_entries)
1910 {
1911 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1912
1913 return mempool_create_slab_pool(pool_entries, bp->slab);
1914 }
1915
1916 void bioset_free(struct bio_set *bs)
1917 {
1918 if (bs->rescue_workqueue)
1919 destroy_workqueue(bs->rescue_workqueue);
1920
1921 if (bs->bio_pool)
1922 mempool_destroy(bs->bio_pool);
1923
1924 if (bs->bvec_pool)
1925 mempool_destroy(bs->bvec_pool);
1926
1927 bioset_integrity_free(bs);
1928 bio_put_slab(bs);
1929
1930 kfree(bs);
1931 }
1932 EXPORT_SYMBOL(bioset_free);
1933
1934 static struct bio_set *__bioset_create(unsigned int pool_size,
1935 unsigned int front_pad,
1936 bool create_bvec_pool)
1937 {
1938 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1939 struct bio_set *bs;
1940
1941 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1942 if (!bs)
1943 return NULL;
1944
1945 bs->front_pad = front_pad;
1946
1947 spin_lock_init(&bs->rescue_lock);
1948 bio_list_init(&bs->rescue_list);
1949 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1950
1951 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1952 if (!bs->bio_slab) {
1953 kfree(bs);
1954 return NULL;
1955 }
1956
1957 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1958 if (!bs->bio_pool)
1959 goto bad;
1960
1961 if (create_bvec_pool) {
1962 bs->bvec_pool = biovec_create_pool(pool_size);
1963 if (!bs->bvec_pool)
1964 goto bad;
1965 }
1966
1967 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1968 if (!bs->rescue_workqueue)
1969 goto bad;
1970
1971 return bs;
1972 bad:
1973 bioset_free(bs);
1974 return NULL;
1975 }
1976
1977 /**
1978 * bioset_create - Create a bio_set
1979 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1980 * @front_pad: Number of bytes to allocate in front of the returned bio
1981 *
1982 * Description:
1983 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1984 * to ask for a number of bytes to be allocated in front of the bio.
1985 * Front pad allocation is useful for embedding the bio inside
1986 * another structure, to avoid allocating extra data to go with the bio.
1987 * Note that the bio must be embedded at the END of that structure always,
1988 * or things will break badly.
1989 */
1990 struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1991 {
1992 return __bioset_create(pool_size, front_pad, true);
1993 }
1994 EXPORT_SYMBOL(bioset_create);
1995
1996 /**
1997 * bioset_create_nobvec - Create a bio_set without bio_vec mempool
1998 * @pool_size: Number of bio to cache in the mempool
1999 * @front_pad: Number of bytes to allocate in front of the returned bio
2000 *
2001 * Description:
2002 * Same functionality as bioset_create() except that mempool is not
2003 * created for bio_vecs. Saving some memory for bio_clone_fast() users.
2004 */
2005 struct bio_set *bioset_create_nobvec(unsigned int pool_size, unsigned int front_pad)
2006 {
2007 return __bioset_create(pool_size, front_pad, false);
2008 }
2009 EXPORT_SYMBOL(bioset_create_nobvec);
2010
2011 #ifdef CONFIG_BLK_CGROUP
2012 /**
2013 * bio_associate_current - associate a bio with %current
2014 * @bio: target bio
2015 *
2016 * Associate @bio with %current if it hasn't been associated yet. Block
2017 * layer will treat @bio as if it were issued by %current no matter which
2018 * task actually issues it.
2019 *
2020 * This function takes an extra reference of @task's io_context and blkcg
2021 * which will be put when @bio is released. The caller must own @bio,
2022 * ensure %current->io_context exists, and is responsible for synchronizing
2023 * calls to this function.
2024 */
2025 int bio_associate_current(struct bio *bio)
2026 {
2027 struct io_context *ioc;
2028 struct cgroup_subsys_state *css;
2029
2030 if (bio->bi_ioc)
2031 return -EBUSY;
2032
2033 ioc = current->io_context;
2034 if (!ioc)
2035 return -ENOENT;
2036
2037 /* acquire active ref on @ioc and associate */
2038 get_io_context_active(ioc);
2039 bio->bi_ioc = ioc;
2040
2041 /* associate blkcg if exists */
2042 rcu_read_lock();
2043 css = task_css(current, blkio_cgrp_id);
2044 if (css && css_tryget_online(css))
2045 bio->bi_css = css;
2046 rcu_read_unlock();
2047
2048 return 0;
2049 }
2050
2051 /**
2052 * bio_disassociate_task - undo bio_associate_current()
2053 * @bio: target bio
2054 */
2055 void bio_disassociate_task(struct bio *bio)
2056 {
2057 if (bio->bi_ioc) {
2058 put_io_context(bio->bi_ioc);
2059 bio->bi_ioc = NULL;
2060 }
2061 if (bio->bi_css) {
2062 css_put(bio->bi_css);
2063 bio->bi_css = NULL;
2064 }
2065 }
2066
2067 #endif /* CONFIG_BLK_CGROUP */
2068
2069 static void __init biovec_init_slabs(void)
2070 {
2071 int i;
2072
2073 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
2074 int size;
2075 struct biovec_slab *bvs = bvec_slabs + i;
2076
2077 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2078 bvs->slab = NULL;
2079 continue;
2080 }
2081
2082 size = bvs->nr_vecs * sizeof(struct bio_vec);
2083 bvs->slab = kmem_cache_create(bvs->name, size, 0,
2084 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2085 }
2086 }
2087
2088 static int __init init_bio(void)
2089 {
2090 bio_slab_max = 2;
2091 bio_slab_nr = 0;
2092 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
2093 if (!bio_slabs)
2094 panic("bio: can't allocate bios\n");
2095
2096 bio_integrity_init();
2097 biovec_init_slabs();
2098
2099 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
2100 if (!fs_bio_set)
2101 panic("bio: can't allocate bios\n");
2102
2103 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
2104 panic("bio: can't create integrity pool\n");
2105
2106 return 0;
2107 }
2108 subsys_initcall(init_bio);