projects/linux·blob v4.4 · refs

2055 lines · 48.5 KiB

1/*
2 * Copyright (C) 2001 Jens Axboe <[email protected]>
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
32#include <trace/events/block.h>
33
34/*
35 * Test patch to inline a certain number of bi_io_vec's inside the bio
36 * itself, to shrink a bio data allocation from two mempool calls to one
37 */
38#define BIO_INLINE_VECS 4
39
40/*
41 * if you change this list, also change bvec_alloc or things will
42 * break badly! cannot be bigger than what you can fit into an
43 * unsigned short
44 */
45#define BV(x) { .nr_vecs = x, .name = "biovec-"__stringify(x) }
46static struct biovec_slab bvec_slabs[BIOVEC_NR_POOLS] __read_mostly = {
47 BV(1), BV(4), BV(16), BV(64), BV(128), BV(BIO_MAX_PAGES),
48};
49#undef BV
50
51/*
52 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
53 * IO code that does not need private memory pools.
54 */
55struct bio_set *fs_bio_set;
56EXPORT_SYMBOL(fs_bio_set);
57
58/*
59 * Our slab pool management
60 */
61struct bio_slab {
62 struct kmem_cache *slab;
63 unsigned int slab_ref;
64 unsigned int slab_size;
65 char name[8];
66};
67static DEFINE_MUTEX(bio_slab_lock);
68static struct bio_slab *bio_slabs;
69static unsigned int bio_slab_nr, bio_slab_max;
70
71static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
72{
73 unsigned int sz = sizeof(struct bio) + extra_size;
74 struct kmem_cache *slab = NULL;
75 struct bio_slab *bslab, *new_bio_slabs;
76 unsigned int new_bio_slab_max;
77 unsigned int i, entry = -1;
78
79 mutex_lock(&bio_slab_lock);
80
81 i = 0;
82 while (i < bio_slab_nr) {
83 bslab = &bio_slabs[i];
84
85 if (!bslab->slab && entry == -1)
86 entry = i;
87 else if (bslab->slab_size == sz) {
88 slab = bslab->slab;
89 bslab->slab_ref++;
90 break;
91 }
92 i++;
93 }
94
95 if (slab)
96 goto out_unlock;
97
98 if (bio_slab_nr == bio_slab_max && entry == -1) {
99 new_bio_slab_max = bio_slab_max << 1;
100 new_bio_slabs = krealloc(bio_slabs,
101 new_bio_slab_max * sizeof(struct bio_slab),
102 GFP_KERNEL);
103 if (!new_bio_slabs)
104 goto out_unlock;
105 bio_slab_max = new_bio_slab_max;
106 bio_slabs = new_bio_slabs;
107 }
108 if (entry == -1)
109 entry = bio_slab_nr++;
110
111 bslab = &bio_slabs[entry];
112
113 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
114 slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN,
115 SLAB_HWCACHE_ALIGN, NULL);
116 if (!slab)
117 goto out_unlock;
118
119 bslab->slab = slab;
120 bslab->slab_ref = 1;
121 bslab->slab_size = sz;
122out_unlock:
123 mutex_unlock(&bio_slab_lock);
124 return slab;
125}
126
127static void bio_put_slab(struct bio_set *bs)
128{
129 struct bio_slab *bslab = NULL;
130 unsigned int i;
131
132 mutex_lock(&bio_slab_lock);
133
134 for (i = 0; i < bio_slab_nr; i++) {
135 if (bs->bio_slab == bio_slabs[i].slab) {
136 bslab = &bio_slabs[i];
137 break;
138 }
139 }
140
141 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
142 goto out;
143
144 WARN_ON(!bslab->slab_ref);
145
146 if (--bslab->slab_ref)
147 goto out;
148
149 kmem_cache_destroy(bslab->slab);
150 bslab->slab = NULL;
151
152out:
153 mutex_unlock(&bio_slab_lock);
154}
155
156unsigned int bvec_nr_vecs(unsigned short idx)
157{
158 return bvec_slabs[idx].nr_vecs;
159}
160
161void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx)
162{
163 BIO_BUG_ON(idx >= BIOVEC_NR_POOLS);
164
165 if (idx == BIOVEC_MAX_IDX)
166 mempool_free(bv, pool);
167 else {
168 struct biovec_slab *bvs = bvec_slabs + idx;
169
170 kmem_cache_free(bvs->slab, bv);
171 }
172}
173
174struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx,
175 mempool_t *pool)
176{
177 struct bio_vec *bvl;
178
179 /*
180 * see comment near bvec_array define!
181 */
182 switch (nr) {
183 case 1:
184 *idx = 0;
185 break;
186 case 2 ... 4:
187 *idx = 1;
188 break;
189 case 5 ... 16:
190 *idx = 2;
191 break;
192 case 17 ... 64:
193 *idx = 3;
194 break;
195 case 65 ... 128:
196 *idx = 4;
197 break;
198 case 129 ... BIO_MAX_PAGES:
199 *idx = 5;
200 break;
201 default:
202 return NULL;
203 }
204
205 /*
206 * idx now points to the pool we want to allocate from. only the
207 * 1-vec entry pool is mempool backed.
208 */
209 if (*idx == BIOVEC_MAX_IDX) {
210fallback:
211 bvl = mempool_alloc(pool, gfp_mask);
212 } else {
213 struct biovec_slab *bvs = bvec_slabs + *idx;
214 gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO);
215
216 /*
217 * Make this allocation restricted and don't dump info on
218 * allocation failures, since we'll fallback to the mempool
219 * in case of failure.
220 */
221 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
222
223 /*
224 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
225 * is set, retry with the 1-entry mempool
226 */
227 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
228 if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) {
229 *idx = BIOVEC_MAX_IDX;
230 goto fallback;
231 }
232 }
233
234 return bvl;
235}
236
237static void __bio_free(struct bio *bio)
238{
239 bio_disassociate_task(bio);
240
241 if (bio_integrity(bio))
242 bio_integrity_free(bio);
243}
244
245static void bio_free(struct bio *bio)
246{
247 struct bio_set *bs = bio->bi_pool;
248 void *p;
249
250 __bio_free(bio);
251
252 if (bs) {
253 if (bio_flagged(bio, BIO_OWNS_VEC))
254 bvec_free(bs->bvec_pool, bio->bi_io_vec, BIO_POOL_IDX(bio));
255
256 /*
257 * If we have front padding, adjust the bio pointer before freeing
258 */
259 p = bio;
260 p -= bs->front_pad;
261
262 mempool_free(p, bs->bio_pool);
263 } else {
264 /* Bio was allocated by bio_kmalloc() */
265 kfree(bio);
266 }
267}
268
269void bio_init(struct bio *bio)
270{
271 memset(bio, 0, sizeof(*bio));
272 atomic_set(&bio->__bi_remaining, 1);
273 atomic_set(&bio->__bi_cnt, 1);
274}
275EXPORT_SYMBOL(bio_init);
276
277/**
278 * bio_reset - reinitialize a bio
279 * @bio: bio to reset
280 *
281 * Description:
282 * After calling bio_reset(), @bio will be in the same state as a freshly
283 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
284 * preserved are the ones that are initialized by bio_alloc_bioset(). See
285 * comment in struct bio.
286 */
287void bio_reset(struct bio *bio)
288{
289 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
290
291 __bio_free(bio);
292
293 memset(bio, 0, BIO_RESET_BYTES);
294 bio->bi_flags = flags;
295 atomic_set(&bio->__bi_remaining, 1);
296}
297EXPORT_SYMBOL(bio_reset);
298
299static void bio_chain_endio(struct bio *bio)
300{
301 struct bio *parent = bio->bi_private;
302
303 parent->bi_error = bio->bi_error;
304 bio_endio(parent);
305 bio_put(bio);
306}
307
308/*
309 * Increment chain count for the bio. Make sure the CHAIN flag update
310 * is visible before the raised count.
311 */
312static inline void bio_inc_remaining(struct bio *bio)
313{
314 bio_set_flag(bio, BIO_CHAIN);
315 smp_mb__before_atomic();
316 atomic_inc(&bio->__bi_remaining);
317}
318
319/**
320 * bio_chain - chain bio completions
321 * @bio: the target bio
322 * @parent: the @bio's parent bio
323 *
324 * The caller won't have a bi_end_io called when @bio completes - instead,
325 * @parent's bi_end_io won't be called until both @parent and @bio have
326 * completed; the chained bio will also be freed when it completes.
327 *
328 * The caller must not set bi_private or bi_end_io in @bio.
329 */
330void bio_chain(struct bio *bio, struct bio *parent)
331{
332 BUG_ON(bio->bi_private || bio->bi_end_io);
333
334 bio->bi_private = parent;
335 bio->bi_end_io = bio_chain_endio;
336 bio_inc_remaining(parent);
337}
338EXPORT_SYMBOL(bio_chain);
339
340static void bio_alloc_rescue(struct work_struct *work)
341{
342 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
343 struct bio *bio;
344
345 while (1) {
346 spin_lock(&bs->rescue_lock);
347 bio = bio_list_pop(&bs->rescue_list);
348 spin_unlock(&bs->rescue_lock);
349
350 if (!bio)
351 break;
352
353 generic_make_request(bio);
354 }
355}
356
357static void punt_bios_to_rescuer(struct bio_set *bs)
358{
359 struct bio_list punt, nopunt;
360 struct bio *bio;
361
362 /*
363 * In order to guarantee forward progress we must punt only bios that
364 * were allocated from this bio_set; otherwise, if there was a bio on
365 * there for a stacking driver higher up in the stack, processing it
366 * could require allocating bios from this bio_set, and doing that from
367 * our own rescuer would be bad.
368 *
369 * Since bio lists are singly linked, pop them all instead of trying to
370 * remove from the middle of the list:
371 */
372
373 bio_list_init(&punt);
374 bio_list_init(&nopunt);
375
376 while ((bio = bio_list_pop(current->bio_list)))
377 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
378
379 *current->bio_list = nopunt;
380
381 spin_lock(&bs->rescue_lock);
382 bio_list_merge(&bs->rescue_list, &punt);
383 spin_unlock(&bs->rescue_lock);
384
385 queue_work(bs->rescue_workqueue, &bs->rescue_work);
386}
387
388/**
389 * bio_alloc_bioset - allocate a bio for I/O
390 * @gfp_mask: the GFP_ mask given to the slab allocator
391 * @nr_iovecs: number of iovecs to pre-allocate
392 * @bs: the bio_set to allocate from.
393 *
394 * Description:
395 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
396 * backed by the @bs's mempool.
397 *
398 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
399 * always be able to allocate a bio. This is due to the mempool guarantees.
400 * To make this work, callers must never allocate more than 1 bio at a time
401 * from this pool. Callers that need to allocate more than 1 bio must always
402 * submit the previously allocated bio for IO before attempting to allocate
403 * a new one. Failure to do so can cause deadlocks under memory pressure.
404 *
405 * Note that when running under generic_make_request() (i.e. any block
406 * driver), bios are not submitted until after you return - see the code in
407 * generic_make_request() that converts recursion into iteration, to prevent
408 * stack overflows.
409 *
410 * This would normally mean allocating multiple bios under
411 * generic_make_request() would be susceptible to deadlocks, but we have
412 * deadlock avoidance code that resubmits any blocked bios from a rescuer
413 * thread.
414 *
415 * However, we do not guarantee forward progress for allocations from other
416 * mempools. Doing multiple allocations from the same mempool under
417 * generic_make_request() should be avoided - instead, use bio_set's front_pad
418 * for per bio allocations.
419 *
420 * RETURNS:
421 * Pointer to new bio on success, NULL on failure.
422 */
423struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
424{
425 gfp_t saved_gfp = gfp_mask;
426 unsigned front_pad;
427 unsigned inline_vecs;
428 unsigned long idx = BIO_POOL_NONE;
429 struct bio_vec *bvl = NULL;
430 struct bio *bio;
431 void *p;
432
433 if (!bs) {
434 if (nr_iovecs > UIO_MAXIOV)
435 return NULL;
436
437 p = kmalloc(sizeof(struct bio) +
438 nr_iovecs * sizeof(struct bio_vec),
439 gfp_mask);
440 front_pad = 0;
441 inline_vecs = nr_iovecs;
442 } else {
443 /* should not use nobvec bioset for nr_iovecs > 0 */
444 if (WARN_ON_ONCE(!bs->bvec_pool && nr_iovecs > 0))
445 return NULL;
446 /*
447 * generic_make_request() converts recursion to iteration; this
448 * means if we're running beneath it, any bios we allocate and
449 * submit will not be submitted (and thus freed) until after we
450 * return.
451 *
452 * This exposes us to a potential deadlock if we allocate
453 * multiple bios from the same bio_set() while running
454 * underneath generic_make_request(). If we were to allocate
455 * multiple bios (say a stacking block driver that was splitting
456 * bios), we would deadlock if we exhausted the mempool's
457 * reserve.
458 *
459 * We solve this, and guarantee forward progress, with a rescuer
460 * workqueue per bio_set. If we go to allocate and there are
461 * bios on current->bio_list, we first try the allocation
462 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
463 * bios we would be blocking to the rescuer workqueue before
464 * we retry with the original gfp_flags.
465 */
466
467 if (current->bio_list && !bio_list_empty(current->bio_list))
468 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
469
470 p = mempool_alloc(bs->bio_pool, gfp_mask);
471 if (!p && gfp_mask != saved_gfp) {
472 punt_bios_to_rescuer(bs);
473 gfp_mask = saved_gfp;
474 p = mempool_alloc(bs->bio_pool, gfp_mask);
475 }
476
477 front_pad = bs->front_pad;
478 inline_vecs = BIO_INLINE_VECS;
479 }
480
481 if (unlikely(!p))
482 return NULL;
483
484 bio = p + front_pad;
485 bio_init(bio);
486
487 if (nr_iovecs > inline_vecs) {
488 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
489 if (!bvl && gfp_mask != saved_gfp) {
490 punt_bios_to_rescuer(bs);
491 gfp_mask = saved_gfp;
492 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
493 }
494
495 if (unlikely(!bvl))
496 goto err_free;
497
498 bio_set_flag(bio, BIO_OWNS_VEC);
499 } else if (nr_iovecs) {
500 bvl = bio->bi_inline_vecs;
501 }
502
503 bio->bi_pool = bs;
504 bio->bi_flags |= idx << BIO_POOL_OFFSET;
505 bio->bi_max_vecs = nr_iovecs;
506 bio->bi_io_vec = bvl;
507 return bio;
508
509err_free:
510 mempool_free(p, bs->bio_pool);
511 return NULL;
512}
513EXPORT_SYMBOL(bio_alloc_bioset);
514
515void zero_fill_bio(struct bio *bio)
516{
517 unsigned long flags;
518 struct bio_vec bv;
519 struct bvec_iter iter;
520
521 bio_for_each_segment(bv, bio, iter) {
522 char *data = bvec_kmap_irq(&bv, &flags);
523 memset(data, 0, bv.bv_len);
524 flush_dcache_page(bv.bv_page);
525 bvec_kunmap_irq(data, &flags);
526 }
527}
528EXPORT_SYMBOL(zero_fill_bio);
529
530/**
531 * bio_put - release a reference to a bio
532 * @bio: bio to release reference to
533 *
534 * Description:
535 * Put a reference to a &struct bio, either one you have gotten with
536 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
537 **/
538void bio_put(struct bio *bio)
539{
540 if (!bio_flagged(bio, BIO_REFFED))
541 bio_free(bio);
542 else {
543 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
544
545 /*
546 * last put frees it
547 */
548 if (atomic_dec_and_test(&bio->__bi_cnt))
549 bio_free(bio);
550 }
551}
552EXPORT_SYMBOL(bio_put);
553
554inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
555{
556 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
557 blk_recount_segments(q, bio);
558
559 return bio->bi_phys_segments;
560}
561EXPORT_SYMBOL(bio_phys_segments);
562
563/**
564 * __bio_clone_fast - clone a bio that shares the original bio's biovec
565 * @bio: destination bio
566 * @bio_src: bio to clone
567 *
568 * Clone a &bio. Caller will own the returned bio, but not
569 * the actual data it points to. Reference count of returned
570 * bio will be one.
571 *
572 * Caller must ensure that @bio_src is not freed before @bio.
573 */
574void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
575{
576 BUG_ON(bio->bi_pool && BIO_POOL_IDX(bio) != BIO_POOL_NONE);
577
578 /*
579 * most users will be overriding ->bi_bdev with a new target,
580 * so we don't set nor calculate new physical/hw segment counts here
581 */
582 bio->bi_bdev = bio_src->bi_bdev;
583 bio_set_flag(bio, BIO_CLONED);
584 bio->bi_rw = bio_src->bi_rw;
585 bio->bi_iter = bio_src->bi_iter;
586 bio->bi_io_vec = bio_src->bi_io_vec;
587}
588EXPORT_SYMBOL(__bio_clone_fast);
589
590/**
591 * bio_clone_fast - clone a bio that shares the original bio's biovec
592 * @bio: bio to clone
593 * @gfp_mask: allocation priority
594 * @bs: bio_set to allocate from
595 *
596 * Like __bio_clone_fast, only also allocates the returned bio
597 */
598struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
599{
600 struct bio *b;
601
602 b = bio_alloc_bioset(gfp_mask, 0, bs);
603 if (!b)
604 return NULL;
605
606 __bio_clone_fast(b, bio);
607
608 if (bio_integrity(bio)) {
609 int ret;
610
611 ret = bio_integrity_clone(b, bio, gfp_mask);
612
613 if (ret < 0) {
614 bio_put(b);
615 return NULL;
616 }
617 }
618
619 return b;
620}
621EXPORT_SYMBOL(bio_clone_fast);
622
623/**
624 * bio_clone_bioset - clone a bio
625 * @bio_src: bio to clone
626 * @gfp_mask: allocation priority
627 * @bs: bio_set to allocate from
628 *
629 * Clone bio. Caller will own the returned bio, but not the actual data it
630 * points to. Reference count of returned bio will be one.
631 */
632struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask,
633 struct bio_set *bs)
634{
635 struct bvec_iter iter;
636 struct bio_vec bv;
637 struct bio *bio;
638
639 /*
640 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
641 * bio_src->bi_io_vec to bio->bi_io_vec.
642 *
643 * We can't do that anymore, because:
644 *
645 * - The point of cloning the biovec is to produce a bio with a biovec
646 * the caller can modify: bi_idx and bi_bvec_done should be 0.
647 *
648 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
649 * we tried to clone the whole thing bio_alloc_bioset() would fail.
650 * But the clone should succeed as long as the number of biovecs we
651 * actually need to allocate is fewer than BIO_MAX_PAGES.
652 *
653 * - Lastly, bi_vcnt should not be looked at or relied upon by code
654 * that does not own the bio - reason being drivers don't use it for
655 * iterating over the biovec anymore, so expecting it to be kept up
656 * to date (i.e. for clones that share the parent biovec) is just
657 * asking for trouble and would force extra work on
658 * __bio_clone_fast() anyways.
659 */
660
661 bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs);
662 if (!bio)
663 return NULL;
664
665 bio->bi_bdev = bio_src->bi_bdev;
666 bio->bi_rw = bio_src->bi_rw;
667 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector;
668 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
669
670 if (bio->bi_rw & REQ_DISCARD)
671 goto integrity_clone;
672
673 if (bio->bi_rw & REQ_WRITE_SAME) {
674 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
675 goto integrity_clone;
676 }
677
678 bio_for_each_segment(bv, bio_src, iter)
679 bio->bi_io_vec[bio->bi_vcnt++] = bv;
680
681integrity_clone:
682 if (bio_integrity(bio_src)) {
683 int ret;
684
685 ret = bio_integrity_clone(bio, bio_src, gfp_mask);
686 if (ret < 0) {
687 bio_put(bio);
688 return NULL;
689 }
690 }
691
692 return bio;
693}
694EXPORT_SYMBOL(bio_clone_bioset);
695
696/**
697 * bio_add_pc_page - attempt to add page to bio
698 * @q: the target queue
699 * @bio: destination bio
700 * @page: page to add
701 * @len: vec entry length
702 * @offset: vec entry offset
703 *
704 * Attempt to add a page to the bio_vec maplist. This can fail for a
705 * number of reasons, such as the bio being full or target block device
706 * limitations. The target block device must allow bio's up to PAGE_SIZE,
707 * so it is always possible to add a single page to an empty bio.
708 *
709 * This should only be used by REQ_PC bios.
710 */
711int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page
712 *page, unsigned int len, unsigned int offset)
713{
714 int retried_segments = 0;
715 struct bio_vec *bvec;
716
717 /*
718 * cloned bio must not modify vec list
719 */
720 if (unlikely(bio_flagged(bio, BIO_CLONED)))
721 return 0;
722
723 if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q))
724 return 0;
725
726 /*
727 * For filesystems with a blocksize smaller than the pagesize
728 * we will often be called with the same page as last time and
729 * a consecutive offset. Optimize this special case.
730 */
731 if (bio->bi_vcnt > 0) {
732 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
733
734 if (page == prev->bv_page &&
735 offset == prev->bv_offset + prev->bv_len) {
736 prev->bv_len += len;
737 bio->bi_iter.bi_size += len;
738 goto done;
739 }
740
741 /*
742 * If the queue doesn't support SG gaps and adding this
743 * offset would create a gap, disallow it.
744 */
745 if (bvec_gap_to_prev(q, prev, offset))
746 return 0;
747 }
748
749 if (bio->bi_vcnt >= bio->bi_max_vecs)
750 return 0;
751
752 /*
753 * setup the new entry, we might clear it again later if we
754 * cannot add the page
755 */
756 bvec = &bio->bi_io_vec[bio->bi_vcnt];
757 bvec->bv_page = page;
758 bvec->bv_len = len;
759 bvec->bv_offset = offset;
760 bio->bi_vcnt++;
761 bio->bi_phys_segments++;
762 bio->bi_iter.bi_size += len;
763
764 /*
765 * Perform a recount if the number of segments is greater
766 * than queue_max_segments(q).
767 */
768
769 while (bio->bi_phys_segments > queue_max_segments(q)) {
770
771 if (retried_segments)
772 goto failed;
773
774 retried_segments = 1;
775 blk_recount_segments(q, bio);
776 }
777
778 /* If we may be able to merge these biovecs, force a recount */
779 if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
780 bio_clear_flag(bio, BIO_SEG_VALID);
781
782 done:
783 return len;
784
785 failed:
786 bvec->bv_page = NULL;
787 bvec->bv_len = 0;
788 bvec->bv_offset = 0;
789 bio->bi_vcnt--;
790 bio->bi_iter.bi_size -= len;
791 blk_recount_segments(q, bio);
792 return 0;
793}
794EXPORT_SYMBOL(bio_add_pc_page);
795
796/**
797 * bio_add_page - attempt to add page to bio
798 * @bio: destination bio
799 * @page: page to add
800 * @len: vec entry length
801 * @offset: vec entry offset
802 *
803 * Attempt to add a page to the bio_vec maplist. This will only fail
804 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
805 */
806int bio_add_page(struct bio *bio, struct page *page,
807 unsigned int len, unsigned int offset)
808{
809 struct bio_vec *bv;
810
811 /*
812 * cloned bio must not modify vec list
813 */
814 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
815 return 0;
816
817 /*
818 * For filesystems with a blocksize smaller than the pagesize
819 * we will often be called with the same page as last time and
820 * a consecutive offset. Optimize this special case.
821 */
822 if (bio->bi_vcnt > 0) {
823 bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
824
825 if (page == bv->bv_page &&
826 offset == bv->bv_offset + bv->bv_len) {
827 bv->bv_len += len;
828 goto done;
829 }
830 }
831
832 if (bio->bi_vcnt >= bio->bi_max_vecs)
833 return 0;
834
835 bv = &bio->bi_io_vec[bio->bi_vcnt];
836 bv->bv_page = page;
837 bv->bv_len = len;
838 bv->bv_offset = offset;
839
840 bio->bi_vcnt++;
841done:
842 bio->bi_iter.bi_size += len;
843 return len;
844}
845EXPORT_SYMBOL(bio_add_page);
846
847struct submit_bio_ret {
848 struct completion event;
849 int error;
850};
851
852static void submit_bio_wait_endio(struct bio *bio)
853{
854 struct submit_bio_ret *ret = bio->bi_private;
855
856 ret->error = bio->bi_error;
857 complete(&ret->event);
858}
859
860/**
861 * submit_bio_wait - submit a bio, and wait until it completes
862 * @rw: whether to %READ or %WRITE, or maybe to %READA (read ahead)
863 * @bio: The &struct bio which describes the I/O
864 *
865 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
866 * bio_endio() on failure.
867 */
868int submit_bio_wait(int rw, struct bio *bio)
869{
870 struct submit_bio_ret ret;
871
872 rw |= REQ_SYNC;
873 init_completion(&ret.event);
874 bio->bi_private = &ret;
875 bio->bi_end_io = submit_bio_wait_endio;
876 submit_bio(rw, bio);
877 wait_for_completion(&ret.event);
878
879 return ret.error;
880}
881EXPORT_SYMBOL(submit_bio_wait);
882
883/**
884 * bio_advance - increment/complete a bio by some number of bytes
885 * @bio: bio to advance
886 * @bytes: number of bytes to complete
887 *
888 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
889 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
890 * be updated on the last bvec as well.
891 *
892 * @bio will then represent the remaining, uncompleted portion of the io.
893 */
894void bio_advance(struct bio *bio, unsigned bytes)
895{
896 if (bio_integrity(bio))
897 bio_integrity_advance(bio, bytes);
898
899 bio_advance_iter(bio, &bio->bi_iter, bytes);
900}
901EXPORT_SYMBOL(bio_advance);
902
903/**
904 * bio_alloc_pages - allocates a single page for each bvec in a bio
905 * @bio: bio to allocate pages for
906 * @gfp_mask: flags for allocation
907 *
908 * Allocates pages up to @bio->bi_vcnt.
909 *
910 * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
911 * freed.
912 */
913int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask)
914{
915 int i;
916 struct bio_vec *bv;
917
918 bio_for_each_segment_all(bv, bio, i) {
919 bv->bv_page = alloc_page(gfp_mask);
920 if (!bv->bv_page) {
921 while (--bv >= bio->bi_io_vec)
922 __free_page(bv->bv_page);
923 return -ENOMEM;
924 }
925 }
926
927 return 0;
928}
929EXPORT_SYMBOL(bio_alloc_pages);
930
931/**
932 * bio_copy_data - copy contents of data buffers from one chain of bios to
933 * another
934 * @src: source bio list
935 * @dst: destination bio list
936 *
937 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
938 * @src and @dst as linked lists of bios.
939 *
940 * Stops when it reaches the end of either @src or @dst - that is, copies
941 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
942 */
943void bio_copy_data(struct bio *dst, struct bio *src)
944{
945 struct bvec_iter src_iter, dst_iter;
946 struct bio_vec src_bv, dst_bv;
947 void *src_p, *dst_p;
948 unsigned bytes;
949
950 src_iter = src->bi_iter;
951 dst_iter = dst->bi_iter;
952
953 while (1) {
954 if (!src_iter.bi_size) {
955 src = src->bi_next;
956 if (!src)
957 break;
958
959 src_iter = src->bi_iter;
960 }
961
962 if (!dst_iter.bi_size) {
963 dst = dst->bi_next;
964 if (!dst)
965 break;
966
967 dst_iter = dst->bi_iter;
968 }
969
970 src_bv = bio_iter_iovec(src, src_iter);
971 dst_bv = bio_iter_iovec(dst, dst_iter);
972
973 bytes = min(src_bv.bv_len, dst_bv.bv_len);
974
975 src_p = kmap_atomic(src_bv.bv_page);
976 dst_p = kmap_atomic(dst_bv.bv_page);
977
978 memcpy(dst_p + dst_bv.bv_offset,
979 src_p + src_bv.bv_offset,
980 bytes);
981
982 kunmap_atomic(dst_p);
983 kunmap_atomic(src_p);
984
985 bio_advance_iter(src, &src_iter, bytes);
986 bio_advance_iter(dst, &dst_iter, bytes);
987 }
988}
989EXPORT_SYMBOL(bio_copy_data);
990
991struct bio_map_data {
992 int is_our_pages;
993 struct iov_iter iter;
994 struct iovec iov[];
995};
996
997static struct bio_map_data *bio_alloc_map_data(unsigned int iov_count,
998 gfp_t gfp_mask)
999{
1000 if (iov_count > UIO_MAXIOV)
1001 return NULL;
1002
1003 return kmalloc(sizeof(struct bio_map_data) +
1004 sizeof(struct iovec) * iov_count, gfp_mask);
1005}
1006
1007/**
1008 * bio_copy_from_iter - copy all pages from iov_iter to bio
1009 * @bio: The &struct bio which describes the I/O as destination
1010 * @iter: iov_iter as source
1011 *
1012 * Copy all pages from iov_iter to bio.
1013 * Returns 0 on success, or error on failure.
1014 */
1015static int bio_copy_from_iter(struct bio *bio, struct iov_iter iter)
1016{
1017 int i;
1018 struct bio_vec *bvec;
1019
1020 bio_for_each_segment_all(bvec, bio, i) {
1021 ssize_t ret;
1022
1023 ret = copy_page_from_iter(bvec->bv_page,
1024 bvec->bv_offset,
1025 bvec->bv_len,
1026 &iter);
1027
1028 if (!iov_iter_count(&iter))
1029 break;
1030
1031 if (ret < bvec->bv_len)
1032 return -EFAULT;
1033 }
1034
1035 return 0;
1036}
1037
1038/**
1039 * bio_copy_to_iter - copy all pages from bio to iov_iter
1040 * @bio: The &struct bio which describes the I/O as source
1041 * @iter: iov_iter as destination
1042 *
1043 * Copy all pages from bio to iov_iter.
1044 * Returns 0 on success, or error on failure.
1045 */
1046static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
1047{
1048 int i;
1049 struct bio_vec *bvec;
1050
1051 bio_for_each_segment_all(bvec, bio, i) {
1052 ssize_t ret;
1053
1054 ret = copy_page_to_iter(bvec->bv_page,
1055 bvec->bv_offset,
1056 bvec->bv_len,
1057 &iter);
1058
1059 if (!iov_iter_count(&iter))
1060 break;
1061
1062 if (ret < bvec->bv_len)
1063 return -EFAULT;
1064 }
1065
1066 return 0;
1067}
1068
1069static void bio_free_pages(struct bio *bio)
1070{
1071 struct bio_vec *bvec;
1072 int i;
1073
1074 bio_for_each_segment_all(bvec, bio, i)
1075 __free_page(bvec->bv_page);
1076}
1077
1078/**
1079 * bio_uncopy_user - finish previously mapped bio
1080 * @bio: bio being terminated
1081 *
1082 * Free pages allocated from bio_copy_user_iov() and write back data
1083 * to user space in case of a read.
1084 */
1085int bio_uncopy_user(struct bio *bio)
1086{
1087 struct bio_map_data *bmd = bio->bi_private;
1088 int ret = 0;
1089
1090 if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1091 /*
1092 * if we're in a workqueue, the request is orphaned, so
1093 * don't copy into a random user address space, just free.
1094 */
1095 if (current->mm && bio_data_dir(bio) == READ)
1096 ret = bio_copy_to_iter(bio, bmd->iter);
1097 if (bmd->is_our_pages)
1098 bio_free_pages(bio);
1099 }
1100 kfree(bmd);
1101 bio_put(bio);
1102 return ret;
1103}
1104EXPORT_SYMBOL(bio_uncopy_user);
1105
1106/**
1107 * bio_copy_user_iov - copy user data to bio
1108 * @q: destination block queue
1109 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1110 * @iter: iovec iterator
1111 * @gfp_mask: memory allocation flags
1112 *
1113 * Prepares and returns a bio for indirect user io, bouncing data
1114 * to/from kernel pages as necessary. Must be paired with
1115 * call bio_uncopy_user() on io completion.
1116 */
1117struct bio *bio_copy_user_iov(struct request_queue *q,
1118 struct rq_map_data *map_data,
1119 const struct iov_iter *iter,
1120 gfp_t gfp_mask)
1121{
1122 struct bio_map_data *bmd;
1123 struct page *page;
1124 struct bio *bio;
1125 int i, ret;
1126 int nr_pages = 0;
1127 unsigned int len = iter->count;
1128 unsigned int offset = map_data ? map_data->offset & ~PAGE_MASK : 0;
1129
1130 for (i = 0; i < iter->nr_segs; i++) {
1131 unsigned long uaddr;
1132 unsigned long end;
1133 unsigned long start;
1134
1135 uaddr = (unsigned long) iter->iov[i].iov_base;
1136 end = (uaddr + iter->iov[i].iov_len + PAGE_SIZE - 1)
1137 >> PAGE_SHIFT;
1138 start = uaddr >> PAGE_SHIFT;
1139
1140 /*
1141 * Overflow, abort
1142 */
1143 if (end < start)
1144 return ERR_PTR(-EINVAL);
1145
1146 nr_pages += end - start;
1147 }
1148
1149 if (offset)
1150 nr_pages++;
1151
1152 bmd = bio_alloc_map_data(iter->nr_segs, gfp_mask);
1153 if (!bmd)
1154 return ERR_PTR(-ENOMEM);
1155
1156 /*
1157 * We need to do a deep copy of the iov_iter including the iovecs.
1158 * The caller provided iov might point to an on-stack or otherwise
1159 * shortlived one.
1160 */
1161 bmd->is_our_pages = map_data ? 0 : 1;
1162 memcpy(bmd->iov, iter->iov, sizeof(struct iovec) * iter->nr_segs);
1163 iov_iter_init(&bmd->iter, iter->type, bmd->iov,
1164 iter->nr_segs, iter->count);
1165
1166 ret = -ENOMEM;
1167 bio = bio_kmalloc(gfp_mask, nr_pages);
1168 if (!bio)
1169 goto out_bmd;
1170
1171 if (iter->type & WRITE)
1172 bio->bi_rw |= REQ_WRITE;
1173
1174 ret = 0;
1175
1176 if (map_data) {
1177 nr_pages = 1 << map_data->page_order;
1178 i = map_data->offset / PAGE_SIZE;
1179 }
1180 while (len) {
1181 unsigned int bytes = PAGE_SIZE;
1182
1183 bytes -= offset;
1184
1185 if (bytes > len)
1186 bytes = len;
1187
1188 if (map_data) {
1189 if (i == map_data->nr_entries * nr_pages) {
1190 ret = -ENOMEM;
1191 break;
1192 }
1193
1194 page = map_data->pages[i / nr_pages];
1195 page += (i % nr_pages);
1196
1197 i++;
1198 } else {
1199 page = alloc_page(q->bounce_gfp | gfp_mask);
1200 if (!page) {
1201 ret = -ENOMEM;
1202 break;
1203 }
1204 }
1205
1206 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
1207 break;
1208
1209 len -= bytes;
1210 offset = 0;
1211 }
1212
1213 if (ret)
1214 goto cleanup;
1215
1216 /*
1217 * success
1218 */
1219 if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) ||
1220 (map_data && map_data->from_user)) {
1221 ret = bio_copy_from_iter(bio, *iter);
1222 if (ret)
1223 goto cleanup;
1224 }
1225
1226 bio->bi_private = bmd;
1227 return bio;
1228cleanup:
1229 if (!map_data)
1230 bio_free_pages(bio);
1231 bio_put(bio);
1232out_bmd:
1233 kfree(bmd);
1234 return ERR_PTR(ret);
1235}
1236
1237/**
1238 * bio_map_user_iov - map user iovec into bio
1239 * @q: the struct request_queue for the bio
1240 * @iter: iovec iterator
1241 * @gfp_mask: memory allocation flags
1242 *
1243 * Map the user space address into a bio suitable for io to a block
1244 * device. Returns an error pointer in case of error.
1245 */
1246struct bio *bio_map_user_iov(struct request_queue *q,
1247 const struct iov_iter *iter,
1248 gfp_t gfp_mask)
1249{
1250 int j;
1251 int nr_pages = 0;
1252 struct page **pages;
1253 struct bio *bio;
1254 int cur_page = 0;
1255 int ret, offset;
1256 struct iov_iter i;
1257 struct iovec iov;
1258
1259 iov_for_each(iov, i, *iter) {
1260 unsigned long uaddr = (unsigned long) iov.iov_base;
1261 unsigned long len = iov.iov_len;
1262 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1263 unsigned long start = uaddr >> PAGE_SHIFT;
1264
1265 /*
1266 * Overflow, abort
1267 */
1268 if (end < start)
1269 return ERR_PTR(-EINVAL);
1270
1271 nr_pages += end - start;
1272 /*
1273 * buffer must be aligned to at least hardsector size for now
1274 */
1275 if (uaddr & queue_dma_alignment(q))
1276 return ERR_PTR(-EINVAL);
1277 }
1278
1279 if (!nr_pages)
1280 return ERR_PTR(-EINVAL);
1281
1282 bio = bio_kmalloc(gfp_mask, nr_pages);
1283 if (!bio)
1284 return ERR_PTR(-ENOMEM);
1285
1286 ret = -ENOMEM;
1287 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
1288 if (!pages)
1289 goto out;
1290
1291 iov_for_each(iov, i, *iter) {
1292 unsigned long uaddr = (unsigned long) iov.iov_base;
1293 unsigned long len = iov.iov_len;
1294 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1295 unsigned long start = uaddr >> PAGE_SHIFT;
1296 const int local_nr_pages = end - start;
1297 const int page_limit = cur_page + local_nr_pages;
1298
1299 ret = get_user_pages_fast(uaddr, local_nr_pages,
1300 (iter->type & WRITE) != WRITE,
1301 &pages[cur_page]);
1302 if (ret < local_nr_pages) {
1303 ret = -EFAULT;
1304 goto out_unmap;
1305 }
1306
1307 offset = uaddr & ~PAGE_MASK;
1308 for (j = cur_page; j < page_limit; j++) {
1309 unsigned int bytes = PAGE_SIZE - offset;
1310
1311 if (len <= 0)
1312 break;
1313
1314 if (bytes > len)
1315 bytes = len;
1316
1317 /*
1318 * sorry...
1319 */
1320 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1321 bytes)
1322 break;
1323
1324 len -= bytes;
1325 offset = 0;
1326 }
1327
1328 cur_page = j;
1329 /*
1330 * release the pages we didn't map into the bio, if any
1331 */
1332 while (j < page_limit)
1333 page_cache_release(pages[j++]);
1334 }
1335
1336 kfree(pages);
1337
1338 /*
1339 * set data direction, and check if mapped pages need bouncing
1340 */
1341 if (iter->type & WRITE)
1342 bio->bi_rw |= REQ_WRITE;
1343
1344 bio_set_flag(bio, BIO_USER_MAPPED);
1345
1346 /*
1347 * subtle -- if __bio_map_user() ended up bouncing a bio,
1348 * it would normally disappear when its bi_end_io is run.
1349 * however, we need it for the unmap, so grab an extra
1350 * reference to it
1351 */
1352 bio_get(bio);
1353 return bio;
1354
1355 out_unmap:
1356 for (j = 0; j < nr_pages; j++) {
1357 if (!pages[j])
1358 break;
1359 page_cache_release(pages[j]);
1360 }
1361 out:
1362 kfree(pages);
1363 bio_put(bio);
1364 return ERR_PTR(ret);
1365}
1366
1367static void __bio_unmap_user(struct bio *bio)
1368{
1369 struct bio_vec *bvec;
1370 int i;
1371
1372 /*
1373 * make sure we dirty pages we wrote to
1374 */
1375 bio_for_each_segment_all(bvec, bio, i) {
1376 if (bio_data_dir(bio) == READ)
1377 set_page_dirty_lock(bvec->bv_page);
1378
1379 page_cache_release(bvec->bv_page);
1380 }
1381
1382 bio_put(bio);
1383}
1384
1385/**
1386 * bio_unmap_user - unmap a bio
1387 * @bio: the bio being unmapped
1388 *
1389 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1390 * a process context.
1391 *
1392 * bio_unmap_user() may sleep.
1393 */
1394void bio_unmap_user(struct bio *bio)
1395{
1396 __bio_unmap_user(bio);
1397 bio_put(bio);
1398}
1399EXPORT_SYMBOL(bio_unmap_user);
1400
1401static void bio_map_kern_endio(struct bio *bio)
1402{
1403 bio_put(bio);
1404}
1405
1406/**
1407 * bio_map_kern - map kernel address into bio
1408 * @q: the struct request_queue for the bio
1409 * @data: pointer to buffer to map
1410 * @len: length in bytes
1411 * @gfp_mask: allocation flags for bio allocation
1412 *
1413 * Map the kernel address into a bio suitable for io to a block
1414 * device. Returns an error pointer in case of error.
1415 */
1416struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1417 gfp_t gfp_mask)
1418{
1419 unsigned long kaddr = (unsigned long)data;
1420 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1421 unsigned long start = kaddr >> PAGE_SHIFT;
1422 const int nr_pages = end - start;
1423 int offset, i;
1424 struct bio *bio;
1425
1426 bio = bio_kmalloc(gfp_mask, nr_pages);
1427 if (!bio)
1428 return ERR_PTR(-ENOMEM);
1429
1430 offset = offset_in_page(kaddr);
1431 for (i = 0; i < nr_pages; i++) {
1432 unsigned int bytes = PAGE_SIZE - offset;
1433
1434 if (len <= 0)
1435 break;
1436
1437 if (bytes > len)
1438 bytes = len;
1439
1440 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1441 offset) < bytes) {
1442 /* we don't support partial mappings */
1443 bio_put(bio);
1444 return ERR_PTR(-EINVAL);
1445 }
1446
1447 data += bytes;
1448 len -= bytes;
1449 offset = 0;
1450 }
1451
1452 bio->bi_end_io = bio_map_kern_endio;
1453 return bio;
1454}
1455EXPORT_SYMBOL(bio_map_kern);
1456
1457static void bio_copy_kern_endio(struct bio *bio)
1458{
1459 bio_free_pages(bio);
1460 bio_put(bio);
1461}
1462
1463static void bio_copy_kern_endio_read(struct bio *bio)
1464{
1465 char *p = bio->bi_private;
1466 struct bio_vec *bvec;
1467 int i;
1468
1469 bio_for_each_segment_all(bvec, bio, i) {
1470 memcpy(p, page_address(bvec->bv_page), bvec->bv_len);
1471 p += bvec->bv_len;
1472 }
1473
1474 bio_copy_kern_endio(bio);
1475}
1476
1477/**
1478 * bio_copy_kern - copy kernel address into bio
1479 * @q: the struct request_queue for the bio
1480 * @data: pointer to buffer to copy
1481 * @len: length in bytes
1482 * @gfp_mask: allocation flags for bio and page allocation
1483 * @reading: data direction is READ
1484 *
1485 * copy the kernel address into a bio suitable for io to a block
1486 * device. Returns an error pointer in case of error.
1487 */
1488struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1489 gfp_t gfp_mask, int reading)
1490{
1491 unsigned long kaddr = (unsigned long)data;
1492 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1493 unsigned long start = kaddr >> PAGE_SHIFT;
1494 struct bio *bio;
1495 void *p = data;
1496 int nr_pages = 0;
1497
1498 /*
1499 * Overflow, abort
1500 */
1501 if (end < start)
1502 return ERR_PTR(-EINVAL);
1503
1504 nr_pages = end - start;
1505 bio = bio_kmalloc(gfp_mask, nr_pages);
1506 if (!bio)
1507 return ERR_PTR(-ENOMEM);
1508
1509 while (len) {
1510 struct page *page;
1511 unsigned int bytes = PAGE_SIZE;
1512
1513 if (bytes > len)
1514 bytes = len;
1515
1516 page = alloc_page(q->bounce_gfp | gfp_mask);
1517 if (!page)
1518 goto cleanup;
1519
1520 if (!reading)
1521 memcpy(page_address(page), p, bytes);
1522
1523 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
1524 break;
1525
1526 len -= bytes;
1527 p += bytes;
1528 }
1529
1530 if (reading) {
1531 bio->bi_end_io = bio_copy_kern_endio_read;
1532 bio->bi_private = data;
1533 } else {
1534 bio->bi_end_io = bio_copy_kern_endio;
1535 bio->bi_rw |= REQ_WRITE;
1536 }
1537
1538 return bio;
1539
1540cleanup:
1541 bio_free_pages(bio);
1542 bio_put(bio);
1543 return ERR_PTR(-ENOMEM);
1544}
1545EXPORT_SYMBOL(bio_copy_kern);
1546
1547/*
1548 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1549 * for performing direct-IO in BIOs.
1550 *
1551 * The problem is that we cannot run set_page_dirty() from interrupt context
1552 * because the required locks are not interrupt-safe. So what we can do is to
1553 * mark the pages dirty _before_ performing IO. And in interrupt context,
1554 * check that the pages are still dirty. If so, fine. If not, redirty them
1555 * in process context.
1556 *
1557 * We special-case compound pages here: normally this means reads into hugetlb
1558 * pages. The logic in here doesn't really work right for compound pages
1559 * because the VM does not uniformly chase down the head page in all cases.
1560 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1561 * handle them at all. So we skip compound pages here at an early stage.
1562 *
1563 * Note that this code is very hard to test under normal circumstances because
1564 * direct-io pins the pages with get_user_pages(). This makes
1565 * is_page_cache_freeable return false, and the VM will not clean the pages.
1566 * But other code (eg, flusher threads) could clean the pages if they are mapped
1567 * pagecache.
1568 *
1569 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1570 * deferred bio dirtying paths.
1571 */
1572
1573/*
1574 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1575 */
1576void bio_set_pages_dirty(struct bio *bio)
1577{
1578 struct bio_vec *bvec;
1579 int i;
1580
1581 bio_for_each_segment_all(bvec, bio, i) {
1582 struct page *page = bvec->bv_page;
1583
1584 if (page && !PageCompound(page))
1585 set_page_dirty_lock(page);
1586 }
1587}
1588
1589static void bio_release_pages(struct bio *bio)
1590{
1591 struct bio_vec *bvec;
1592 int i;
1593
1594 bio_for_each_segment_all(bvec, bio, i) {
1595 struct page *page = bvec->bv_page;
1596
1597 if (page)
1598 put_page(page);
1599 }
1600}
1601
1602/*
1603 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1604 * If they are, then fine. If, however, some pages are clean then they must
1605 * have been written out during the direct-IO read. So we take another ref on
1606 * the BIO and the offending pages and re-dirty the pages in process context.
1607 *
1608 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1609 * here on. It will run one page_cache_release() against each page and will
1610 * run one bio_put() against the BIO.
1611 */
1612
1613static void bio_dirty_fn(struct work_struct *work);
1614
1615static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1616static DEFINE_SPINLOCK(bio_dirty_lock);
1617static struct bio *bio_dirty_list;
1618
1619/*
1620 * This runs in process context
1621 */
1622static void bio_dirty_fn(struct work_struct *work)
1623{
1624 unsigned long flags;
1625 struct bio *bio;
1626
1627 spin_lock_irqsave(&bio_dirty_lock, flags);
1628 bio = bio_dirty_list;
1629 bio_dirty_list = NULL;
1630 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1631
1632 while (bio) {
1633 struct bio *next = bio->bi_private;
1634
1635 bio_set_pages_dirty(bio);
1636 bio_release_pages(bio);
1637 bio_put(bio);
1638 bio = next;
1639 }
1640}
1641
1642void bio_check_pages_dirty(struct bio *bio)
1643{
1644 struct bio_vec *bvec;
1645 int nr_clean_pages = 0;
1646 int i;
1647
1648 bio_for_each_segment_all(bvec, bio, i) {
1649 struct page *page = bvec->bv_page;
1650
1651 if (PageDirty(page) || PageCompound(page)) {
1652 page_cache_release(page);
1653 bvec->bv_page = NULL;
1654 } else {
1655 nr_clean_pages++;
1656 }
1657 }
1658
1659 if (nr_clean_pages) {
1660 unsigned long flags;
1661
1662 spin_lock_irqsave(&bio_dirty_lock, flags);
1663 bio->bi_private = bio_dirty_list;
1664 bio_dirty_list = bio;
1665 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1666 schedule_work(&bio_dirty_work);
1667 } else {
1668 bio_put(bio);
1669 }
1670}
1671
1672void generic_start_io_acct(int rw, unsigned long sectors,
1673 struct hd_struct *part)
1674{
1675 int cpu = part_stat_lock();
1676
1677 part_round_stats(cpu, part);
1678 part_stat_inc(cpu, part, ios[rw]);
1679 part_stat_add(cpu, part, sectors[rw], sectors);
1680 part_inc_in_flight(part, rw);
1681
1682 part_stat_unlock();
1683}
1684EXPORT_SYMBOL(generic_start_io_acct);
1685
1686void generic_end_io_acct(int rw, struct hd_struct *part,
1687 unsigned long start_time)
1688{
1689 unsigned long duration = jiffies - start_time;
1690 int cpu = part_stat_lock();
1691
1692 part_stat_add(cpu, part, ticks[rw], duration);
1693 part_round_stats(cpu, part);
1694 part_dec_in_flight(part, rw);
1695
1696 part_stat_unlock();
1697}
1698EXPORT_SYMBOL(generic_end_io_acct);
1699
1700#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1701void bio_flush_dcache_pages(struct bio *bi)
1702{
1703 struct bio_vec bvec;
1704 struct bvec_iter iter;
1705
1706 bio_for_each_segment(bvec, bi, iter)
1707 flush_dcache_page(bvec.bv_page);
1708}
1709EXPORT_SYMBOL(bio_flush_dcache_pages);
1710#endif
1711
1712static inline bool bio_remaining_done(struct bio *bio)
1713{
1714 /*
1715 * If we're not chaining, then ->__bi_remaining is always 1 and
1716 * we always end io on the first invocation.
1717 */
1718 if (!bio_flagged(bio, BIO_CHAIN))
1719 return true;
1720
1721 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1722
1723 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1724 bio_clear_flag(bio, BIO_CHAIN);
1725 return true;
1726 }
1727
1728 return false;
1729}
1730
1731/**
1732 * bio_endio - end I/O on a bio
1733 * @bio: bio
1734 *
1735 * Description:
1736 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1737 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1738 * bio unless they own it and thus know that it has an end_io function.
1739 **/
1740void bio_endio(struct bio *bio)
1741{
1742 while (bio) {
1743 if (unlikely(!bio_remaining_done(bio)))
1744 break;
1745
1746 /*
1747 * Need to have a real endio function for chained bios,
1748 * otherwise various corner cases will break (like stacking
1749 * block devices that save/restore bi_end_io) - however, we want
1750 * to avoid unbounded recursion and blowing the stack. Tail call
1751 * optimization would handle this, but compiling with frame
1752 * pointers also disables gcc's sibling call optimization.
1753 */
1754 if (bio->bi_end_io == bio_chain_endio) {
1755 struct bio *parent = bio->bi_private;
1756 parent->bi_error = bio->bi_error;
1757 bio_put(bio);
1758 bio = parent;
1759 } else {
1760 if (bio->bi_end_io)
1761 bio->bi_end_io(bio);
1762 bio = NULL;
1763 }
1764 }
1765}
1766EXPORT_SYMBOL(bio_endio);
1767
1768/**
1769 * bio_split - split a bio
1770 * @bio: bio to split
1771 * @sectors: number of sectors to split from the front of @bio
1772 * @gfp: gfp mask
1773 * @bs: bio set to allocate from
1774 *
1775 * Allocates and returns a new bio which represents @sectors from the start of
1776 * @bio, and updates @bio to represent the remaining sectors.
1777 *
1778 * Unless this is a discard request the newly allocated bio will point
1779 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1780 * @bio is not freed before the split.
1781 */
1782struct bio *bio_split(struct bio *bio, int sectors,
1783 gfp_t gfp, struct bio_set *bs)
1784{
1785 struct bio *split = NULL;
1786
1787 BUG_ON(sectors <= 0);
1788 BUG_ON(sectors >= bio_sectors(bio));
1789
1790 /*
1791 * Discards need a mutable bio_vec to accommodate the payload
1792 * required by the DSM TRIM and UNMAP commands.
1793 */
1794 if (bio->bi_rw & REQ_DISCARD)
1795 split = bio_clone_bioset(bio, gfp, bs);
1796 else
1797 split = bio_clone_fast(bio, gfp, bs);
1798
1799 if (!split)
1800 return NULL;
1801
1802 split->bi_iter.bi_size = sectors << 9;
1803
1804 if (bio_integrity(split))
1805 bio_integrity_trim(split, 0, sectors);
1806
1807 bio_advance(bio, split->bi_iter.bi_size);
1808
1809 return split;
1810}
1811EXPORT_SYMBOL(bio_split);
1812
1813/**
1814 * bio_trim - trim a bio
1815 * @bio: bio to trim
1816 * @offset: number of sectors to trim from the front of @bio
1817 * @size: size we want to trim @bio to, in sectors
1818 */
1819void bio_trim(struct bio *bio, int offset, int size)
1820{
1821 /* 'bio' is a cloned bio which we need to trim to match
1822 * the given offset and size.
1823 */
1824
1825 size <<= 9;
1826 if (offset == 0 && size == bio->bi_iter.bi_size)
1827 return;
1828
1829 bio_clear_flag(bio, BIO_SEG_VALID);
1830
1831 bio_advance(bio, offset << 9);
1832
1833 bio->bi_iter.bi_size = size;
1834}
1835EXPORT_SYMBOL_GPL(bio_trim);
1836
1837/*
1838 * create memory pools for biovec's in a bio_set.
1839 * use the global biovec slabs created for general use.
1840 */
1841mempool_t *biovec_create_pool(int pool_entries)
1842{
1843 struct biovec_slab *bp = bvec_slabs + BIOVEC_MAX_IDX;
1844
1845 return mempool_create_slab_pool(pool_entries, bp->slab);
1846}
1847
1848void bioset_free(struct bio_set *bs)
1849{
1850 if (bs->rescue_workqueue)
1851 destroy_workqueue(bs->rescue_workqueue);
1852
1853 if (bs->bio_pool)
1854 mempool_destroy(bs->bio_pool);
1855
1856 if (bs->bvec_pool)
1857 mempool_destroy(bs->bvec_pool);
1858
1859 bioset_integrity_free(bs);
1860 bio_put_slab(bs);
1861
1862 kfree(bs);
1863}
1864EXPORT_SYMBOL(bioset_free);
1865
1866static struct bio_set *__bioset_create(unsigned int pool_size,
1867 unsigned int front_pad,
1868 bool create_bvec_pool)
1869{
1870 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1871 struct bio_set *bs;
1872
1873 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1874 if (!bs)
1875 return NULL;
1876
1877 bs->front_pad = front_pad;
1878
1879 spin_lock_init(&bs->rescue_lock);
1880 bio_list_init(&bs->rescue_list);
1881 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1882
1883 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1884 if (!bs->bio_slab) {
1885 kfree(bs);
1886 return NULL;
1887 }
1888
1889 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1890 if (!bs->bio_pool)
1891 goto bad;
1892
1893 if (create_bvec_pool) {
1894 bs->bvec_pool = biovec_create_pool(pool_size);
1895 if (!bs->bvec_pool)
1896 goto bad;
1897 }
1898
1899 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1900 if (!bs->rescue_workqueue)
1901 goto bad;
1902
1903 return bs;
1904bad:
1905 bioset_free(bs);
1906 return NULL;
1907}
1908
1909/**
1910 * bioset_create - Create a bio_set
1911 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1912 * @front_pad: Number of bytes to allocate in front of the returned bio
1913 *
1914 * Description:
1915 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1916 * to ask for a number of bytes to be allocated in front of the bio.
1917 * Front pad allocation is useful for embedding the bio inside
1918 * another structure, to avoid allocating extra data to go with the bio.
1919 * Note that the bio must be embedded at the END of that structure always,
1920 * or things will break badly.
1921 */
1922struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1923{
1924 return __bioset_create(pool_size, front_pad, true);
1925}
1926EXPORT_SYMBOL(bioset_create);
1927
1928/**
1929 * bioset_create_nobvec - Create a bio_set without bio_vec mempool
1930 * @pool_size: Number of bio to cache in the mempool
1931 * @front_pad: Number of bytes to allocate in front of the returned bio
1932 *
1933 * Description:
1934 * Same functionality as bioset_create() except that mempool is not
1935 * created for bio_vecs. Saving some memory for bio_clone_fast() users.
1936 */
1937struct bio_set *bioset_create_nobvec(unsigned int pool_size, unsigned int front_pad)
1938{
1939 return __bioset_create(pool_size, front_pad, false);
1940}
1941EXPORT_SYMBOL(bioset_create_nobvec);
1942
1943#ifdef CONFIG_BLK_CGROUP
1944
1945/**
1946 * bio_associate_blkcg - associate a bio with the specified blkcg
1947 * @bio: target bio
1948 * @blkcg_css: css of the blkcg to associate
1949 *
1950 * Associate @bio with the blkcg specified by @blkcg_css. Block layer will
1951 * treat @bio as if it were issued by a task which belongs to the blkcg.
1952 *
1953 * This function takes an extra reference of @blkcg_css which will be put
1954 * when @bio is released. The caller must own @bio and is responsible for
1955 * synchronizing calls to this function.
1956 */
1957int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css)
1958{
1959 if (unlikely(bio->bi_css))
1960 return -EBUSY;
1961 css_get(blkcg_css);
1962 bio->bi_css = blkcg_css;
1963 return 0;
1964}
1965EXPORT_SYMBOL_GPL(bio_associate_blkcg);
1966
1967/**
1968 * bio_associate_current - associate a bio with %current
1969 * @bio: target bio
1970 *
1971 * Associate @bio with %current if it hasn't been associated yet. Block
1972 * layer will treat @bio as if it were issued by %current no matter which
1973 * task actually issues it.
1974 *
1975 * This function takes an extra reference of @task's io_context and blkcg
1976 * which will be put when @bio is released. The caller must own @bio,
1977 * ensure %current->io_context exists, and is responsible for synchronizing
1978 * calls to this function.
1979 */
1980int bio_associate_current(struct bio *bio)
1981{
1982 struct io_context *ioc;
1983
1984 if (bio->bi_css)
1985 return -EBUSY;
1986
1987 ioc = current->io_context;
1988 if (!ioc)
1989 return -ENOENT;
1990
1991 get_io_context_active(ioc);
1992 bio->bi_ioc = ioc;
1993 bio->bi_css = task_get_css(current, io_cgrp_id);
1994 return 0;
1995}
1996EXPORT_SYMBOL_GPL(bio_associate_current);
1997
1998/**
1999 * bio_disassociate_task - undo bio_associate_current()
2000 * @bio: target bio
2001 */
2002void bio_disassociate_task(struct bio *bio)
2003{
2004 if (bio->bi_ioc) {
2005 put_io_context(bio->bi_ioc);
2006 bio->bi_ioc = NULL;
2007 }
2008 if (bio->bi_css) {
2009 css_put(bio->bi_css);
2010 bio->bi_css = NULL;
2011 }
2012}
2013
2014#endif /* CONFIG_BLK_CGROUP */
2015
2016static void __init biovec_init_slabs(void)
2017{
2018 int i;
2019
2020 for (i = 0; i < BIOVEC_NR_POOLS; i++) {
2021 int size;
2022 struct biovec_slab *bvs = bvec_slabs + i;
2023
2024 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2025 bvs->slab = NULL;
2026 continue;
2027 }
2028
2029 size = bvs->nr_vecs * sizeof(struct bio_vec);
2030 bvs->slab = kmem_cache_create(bvs->name, size, 0,
2031 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2032 }
2033}
2034
2035static int __init init_bio(void)
2036{
2037 bio_slab_max = 2;
2038 bio_slab_nr = 0;
2039 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
2040 if (!bio_slabs)
2041 panic("bio: can't allocate bios\n");
2042
2043 bio_integrity_init();
2044 biovec_init_slabs();
2045
2046 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
2047 if (!fs_bio_set)
2048 panic("bio: can't allocate bios\n");
2049
2050 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
2051 panic("bio: can't create integrity pool\n");
2052
2053 return 0;
2054}
2055subsys_initcall(init_bio);