]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - block/bio.c
Merge branch 'for-4.9/block' of git://git.kernel.dk/linux-block
[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
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) }
46 static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __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 */
55 struct bio_set *fs_bio_set;
56 EXPORT_SYMBOL(fs_bio_set);
57
58 /*
59 * Our slab pool management
60 */
61 struct bio_slab {
62 struct kmem_cache *slab;
63 unsigned int slab_ref;
64 unsigned int slab_size;
65 char name[8];
66 };
67 static DEFINE_MUTEX(bio_slab_lock);
68 static struct bio_slab *bio_slabs;
69 static unsigned int bio_slab_nr, bio_slab_max;
70
71 static 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;
122 out_unlock:
123 mutex_unlock(&bio_slab_lock);
124 return slab;
125 }
126
127 static 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
152 out:
153 mutex_unlock(&bio_slab_lock);
154 }
155
156 unsigned int bvec_nr_vecs(unsigned short idx)
157 {
158 return bvec_slabs[idx].nr_vecs;
159 }
160
161 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx)
162 {
163 if (!idx)
164 return;
165 idx--;
166
167 BIO_BUG_ON(idx >= BVEC_POOL_NR);
168
169 if (idx == BVEC_POOL_MAX) {
170 mempool_free(bv, pool);
171 } else {
172 struct biovec_slab *bvs = bvec_slabs + idx;
173
174 kmem_cache_free(bvs->slab, bv);
175 }
176 }
177
178 struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx,
179 mempool_t *pool)
180 {
181 struct bio_vec *bvl;
182
183 /*
184 * see comment near bvec_array define!
185 */
186 switch (nr) {
187 case 1:
188 *idx = 0;
189 break;
190 case 2 ... 4:
191 *idx = 1;
192 break;
193 case 5 ... 16:
194 *idx = 2;
195 break;
196 case 17 ... 64:
197 *idx = 3;
198 break;
199 case 65 ... 128:
200 *idx = 4;
201 break;
202 case 129 ... BIO_MAX_PAGES:
203 *idx = 5;
204 break;
205 default:
206 return NULL;
207 }
208
209 /*
210 * idx now points to the pool we want to allocate from. only the
211 * 1-vec entry pool is mempool backed.
212 */
213 if (*idx == BVEC_POOL_MAX) {
214 fallback:
215 bvl = mempool_alloc(pool, gfp_mask);
216 } else {
217 struct biovec_slab *bvs = bvec_slabs + *idx;
218 gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO);
219
220 /*
221 * Make this allocation restricted and don't dump info on
222 * allocation failures, since we'll fallback to the mempool
223 * in case of failure.
224 */
225 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
226
227 /*
228 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
229 * is set, retry with the 1-entry mempool
230 */
231 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
232 if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) {
233 *idx = BVEC_POOL_MAX;
234 goto fallback;
235 }
236 }
237
238 (*idx)++;
239 return bvl;
240 }
241
242 static void __bio_free(struct bio *bio)
243 {
244 bio_disassociate_task(bio);
245
246 if (bio_integrity(bio))
247 bio_integrity_free(bio);
248 }
249
250 static void bio_free(struct bio *bio)
251 {
252 struct bio_set *bs = bio->bi_pool;
253 void *p;
254
255 __bio_free(bio);
256
257 if (bs) {
258 bvec_free(bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio));
259
260 /*
261 * If we have front padding, adjust the bio pointer before freeing
262 */
263 p = bio;
264 p -= bs->front_pad;
265
266 mempool_free(p, bs->bio_pool);
267 } else {
268 /* Bio was allocated by bio_kmalloc() */
269 kfree(bio);
270 }
271 }
272
273 void bio_init(struct bio *bio)
274 {
275 memset(bio, 0, sizeof(*bio));
276 atomic_set(&bio->__bi_remaining, 1);
277 atomic_set(&bio->__bi_cnt, 1);
278 }
279 EXPORT_SYMBOL(bio_init);
280
281 /**
282 * bio_reset - reinitialize a bio
283 * @bio: bio to reset
284 *
285 * Description:
286 * After calling bio_reset(), @bio will be in the same state as a freshly
287 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
288 * preserved are the ones that are initialized by bio_alloc_bioset(). See
289 * comment in struct bio.
290 */
291 void bio_reset(struct bio *bio)
292 {
293 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
294
295 __bio_free(bio);
296
297 memset(bio, 0, BIO_RESET_BYTES);
298 bio->bi_flags = flags;
299 atomic_set(&bio->__bi_remaining, 1);
300 }
301 EXPORT_SYMBOL(bio_reset);
302
303 static struct bio *__bio_chain_endio(struct bio *bio)
304 {
305 struct bio *parent = bio->bi_private;
306
307 if (!parent->bi_error)
308 parent->bi_error = bio->bi_error;
309 bio_put(bio);
310 return parent;
311 }
312
313 static void bio_chain_endio(struct bio *bio)
314 {
315 bio_endio(__bio_chain_endio(bio));
316 }
317
318 /**
319 * bio_chain - chain bio completions
320 * @bio: the target bio
321 * @parent: the @bio's parent bio
322 *
323 * The caller won't have a bi_end_io called when @bio completes - instead,
324 * @parent's bi_end_io won't be called until both @parent and @bio have
325 * completed; the chained bio will also be freed when it completes.
326 *
327 * The caller must not set bi_private or bi_end_io in @bio.
328 */
329 void bio_chain(struct bio *bio, struct bio *parent)
330 {
331 BUG_ON(bio->bi_private || bio->bi_end_io);
332
333 bio->bi_private = parent;
334 bio->bi_end_io = bio_chain_endio;
335 bio_inc_remaining(parent);
336 }
337 EXPORT_SYMBOL(bio_chain);
338
339 static void bio_alloc_rescue(struct work_struct *work)
340 {
341 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
342 struct bio *bio;
343
344 while (1) {
345 spin_lock(&bs->rescue_lock);
346 bio = bio_list_pop(&bs->rescue_list);
347 spin_unlock(&bs->rescue_lock);
348
349 if (!bio)
350 break;
351
352 generic_make_request(bio);
353 }
354 }
355
356 static void punt_bios_to_rescuer(struct bio_set *bs)
357 {
358 struct bio_list punt, nopunt;
359 struct bio *bio;
360
361 /*
362 * In order to guarantee forward progress we must punt only bios that
363 * were allocated from this bio_set; otherwise, if there was a bio on
364 * there for a stacking driver higher up in the stack, processing it
365 * could require allocating bios from this bio_set, and doing that from
366 * our own rescuer would be bad.
367 *
368 * Since bio lists are singly linked, pop them all instead of trying to
369 * remove from the middle of the list:
370 */
371
372 bio_list_init(&punt);
373 bio_list_init(&nopunt);
374
375 while ((bio = bio_list_pop(current->bio_list)))
376 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
377
378 *current->bio_list = nopunt;
379
380 spin_lock(&bs->rescue_lock);
381 bio_list_merge(&bs->rescue_list, &punt);
382 spin_unlock(&bs->rescue_lock);
383
384 queue_work(bs->rescue_workqueue, &bs->rescue_work);
385 }
386
387 /**
388 * bio_alloc_bioset - allocate a bio for I/O
389 * @gfp_mask: the GFP_ mask given to the slab allocator
390 * @nr_iovecs: number of iovecs to pre-allocate
391 * @bs: the bio_set to allocate from.
392 *
393 * Description:
394 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
395 * backed by the @bs's mempool.
396 *
397 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
398 * always be able to allocate a bio. This is due to the mempool guarantees.
399 * To make this work, callers must never allocate more than 1 bio at a time
400 * from this pool. Callers that need to allocate more than 1 bio must always
401 * submit the previously allocated bio for IO before attempting to allocate
402 * a new one. Failure to do so can cause deadlocks under memory pressure.
403 *
404 * Note that when running under generic_make_request() (i.e. any block
405 * driver), bios are not submitted until after you return - see the code in
406 * generic_make_request() that converts recursion into iteration, to prevent
407 * stack overflows.
408 *
409 * This would normally mean allocating multiple bios under
410 * generic_make_request() would be susceptible to deadlocks, but we have
411 * deadlock avoidance code that resubmits any blocked bios from a rescuer
412 * thread.
413 *
414 * However, we do not guarantee forward progress for allocations from other
415 * mempools. Doing multiple allocations from the same mempool under
416 * generic_make_request() should be avoided - instead, use bio_set's front_pad
417 * for per bio allocations.
418 *
419 * RETURNS:
420 * Pointer to new bio on success, NULL on failure.
421 */
422 struct bio *bio_alloc_bioset(gfp_t gfp_mask, int nr_iovecs, struct bio_set *bs)
423 {
424 gfp_t saved_gfp = gfp_mask;
425 unsigned front_pad;
426 unsigned inline_vecs;
427 struct bio_vec *bvl = NULL;
428 struct bio *bio;
429 void *p;
430
431 if (!bs) {
432 if (nr_iovecs > UIO_MAXIOV)
433 return NULL;
434
435 p = kmalloc(sizeof(struct bio) +
436 nr_iovecs * sizeof(struct bio_vec),
437 gfp_mask);
438 front_pad = 0;
439 inline_vecs = nr_iovecs;
440 } else {
441 /* should not use nobvec bioset for nr_iovecs > 0 */
442 if (WARN_ON_ONCE(!bs->bvec_pool && nr_iovecs > 0))
443 return NULL;
444 /*
445 * generic_make_request() converts recursion to iteration; this
446 * means if we're running beneath it, any bios we allocate and
447 * submit will not be submitted (and thus freed) until after we
448 * return.
449 *
450 * This exposes us to a potential deadlock if we allocate
451 * multiple bios from the same bio_set() while running
452 * underneath generic_make_request(). If we were to allocate
453 * multiple bios (say a stacking block driver that was splitting
454 * bios), we would deadlock if we exhausted the mempool's
455 * reserve.
456 *
457 * We solve this, and guarantee forward progress, with a rescuer
458 * workqueue per bio_set. If we go to allocate and there are
459 * bios on current->bio_list, we first try the allocation
460 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
461 * bios we would be blocking to the rescuer workqueue before
462 * we retry with the original gfp_flags.
463 */
464
465 if (current->bio_list && !bio_list_empty(current->bio_list))
466 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
467
468 p = mempool_alloc(bs->bio_pool, gfp_mask);
469 if (!p && gfp_mask != saved_gfp) {
470 punt_bios_to_rescuer(bs);
471 gfp_mask = saved_gfp;
472 p = mempool_alloc(bs->bio_pool, gfp_mask);
473 }
474
475 front_pad = bs->front_pad;
476 inline_vecs = BIO_INLINE_VECS;
477 }
478
479 if (unlikely(!p))
480 return NULL;
481
482 bio = p + front_pad;
483 bio_init(bio);
484
485 if (nr_iovecs > inline_vecs) {
486 unsigned long idx = 0;
487
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->bi_flags |= idx << BVEC_POOL_OFFSET;
499 } else if (nr_iovecs) {
500 bvl = bio->bi_inline_vecs;
501 }
502
503 bio->bi_pool = bs;
504 bio->bi_max_vecs = nr_iovecs;
505 bio->bi_io_vec = bvl;
506 return bio;
507
508 err_free:
509 mempool_free(p, bs->bio_pool);
510 return NULL;
511 }
512 EXPORT_SYMBOL(bio_alloc_bioset);
513
514 void zero_fill_bio(struct bio *bio)
515 {
516 unsigned long flags;
517 struct bio_vec bv;
518 struct bvec_iter iter;
519
520 bio_for_each_segment(bv, bio, iter) {
521 char *data = bvec_kmap_irq(&bv, &flags);
522 memset(data, 0, bv.bv_len);
523 flush_dcache_page(bv.bv_page);
524 bvec_kunmap_irq(data, &flags);
525 }
526 }
527 EXPORT_SYMBOL(zero_fill_bio);
528
529 /**
530 * bio_put - release a reference to a bio
531 * @bio: bio to release reference to
532 *
533 * Description:
534 * Put a reference to a &struct bio, either one you have gotten with
535 * bio_alloc, bio_get or bio_clone. The last put of a bio will free it.
536 **/
537 void bio_put(struct bio *bio)
538 {
539 if (!bio_flagged(bio, BIO_REFFED))
540 bio_free(bio);
541 else {
542 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
543
544 /*
545 * last put frees it
546 */
547 if (atomic_dec_and_test(&bio->__bi_cnt))
548 bio_free(bio);
549 }
550 }
551 EXPORT_SYMBOL(bio_put);
552
553 inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
554 {
555 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
556 blk_recount_segments(q, bio);
557
558 return bio->bi_phys_segments;
559 }
560 EXPORT_SYMBOL(bio_phys_segments);
561
562 /**
563 * __bio_clone_fast - clone a bio that shares the original bio's biovec
564 * @bio: destination bio
565 * @bio_src: bio to clone
566 *
567 * Clone a &bio. Caller will own the returned bio, but not
568 * the actual data it points to. Reference count of returned
569 * bio will be one.
570 *
571 * Caller must ensure that @bio_src is not freed before @bio.
572 */
573 void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
574 {
575 BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio));
576
577 /*
578 * most users will be overriding ->bi_bdev with a new target,
579 * so we don't set nor calculate new physical/hw segment counts here
580 */
581 bio->bi_bdev = bio_src->bi_bdev;
582 bio_set_flag(bio, BIO_CLONED);
583 bio->bi_opf = bio_src->bi_opf;
584 bio->bi_iter = bio_src->bi_iter;
585 bio->bi_io_vec = bio_src->bi_io_vec;
586
587 bio_clone_blkcg_association(bio, bio_src);
588 }
589 EXPORT_SYMBOL(__bio_clone_fast);
590
591 /**
592 * bio_clone_fast - clone a bio that shares the original bio's biovec
593 * @bio: bio to clone
594 * @gfp_mask: allocation priority
595 * @bs: bio_set to allocate from
596 *
597 * Like __bio_clone_fast, only also allocates the returned bio
598 */
599 struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
600 {
601 struct bio *b;
602
603 b = bio_alloc_bioset(gfp_mask, 0, bs);
604 if (!b)
605 return NULL;
606
607 __bio_clone_fast(b, bio);
608
609 if (bio_integrity(bio)) {
610 int ret;
611
612 ret = bio_integrity_clone(b, bio, gfp_mask);
613
614 if (ret < 0) {
615 bio_put(b);
616 return NULL;
617 }
618 }
619
620 return b;
621 }
622 EXPORT_SYMBOL(bio_clone_fast);
623
624 /**
625 * bio_clone_bioset - clone a bio
626 * @bio_src: bio to clone
627 * @gfp_mask: allocation priority
628 * @bs: bio_set to allocate from
629 *
630 * Clone bio. Caller will own the returned bio, but not the actual data it
631 * points to. Reference count of returned bio will be one.
632 */
633 struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask,
634 struct bio_set *bs)
635 {
636 struct bvec_iter iter;
637 struct bio_vec bv;
638 struct bio *bio;
639
640 /*
641 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
642 * bio_src->bi_io_vec to bio->bi_io_vec.
643 *
644 * We can't do that anymore, because:
645 *
646 * - The point of cloning the biovec is to produce a bio with a biovec
647 * the caller can modify: bi_idx and bi_bvec_done should be 0.
648 *
649 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
650 * we tried to clone the whole thing bio_alloc_bioset() would fail.
651 * But the clone should succeed as long as the number of biovecs we
652 * actually need to allocate is fewer than BIO_MAX_PAGES.
653 *
654 * - Lastly, bi_vcnt should not be looked at or relied upon by code
655 * that does not own the bio - reason being drivers don't use it for
656 * iterating over the biovec anymore, so expecting it to be kept up
657 * to date (i.e. for clones that share the parent biovec) is just
658 * asking for trouble and would force extra work on
659 * __bio_clone_fast() anyways.
660 */
661
662 bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs);
663 if (!bio)
664 return NULL;
665 bio->bi_bdev = bio_src->bi_bdev;
666 bio->bi_opf = bio_src->bi_opf;
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 switch (bio_op(bio)) {
671 case REQ_OP_DISCARD:
672 case REQ_OP_SECURE_ERASE:
673 break;
674 case REQ_OP_WRITE_SAME:
675 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
676 break;
677 default:
678 bio_for_each_segment(bv, bio_src, iter)
679 bio->bi_io_vec[bio->bi_vcnt++] = bv;
680 break;
681 }
682
683 if (bio_integrity(bio_src)) {
684 int ret;
685
686 ret = bio_integrity_clone(bio, bio_src, gfp_mask);
687 if (ret < 0) {
688 bio_put(bio);
689 return NULL;
690 }
691 }
692
693 bio_clone_blkcg_association(bio, bio_src);
694
695 return bio;
696 }
697 EXPORT_SYMBOL(bio_clone_bioset);
698
699 /**
700 * bio_add_pc_page - attempt to add page to bio
701 * @q: the target queue
702 * @bio: destination bio
703 * @page: page to add
704 * @len: vec entry length
705 * @offset: vec entry offset
706 *
707 * Attempt to add a page to the bio_vec maplist. This can fail for a
708 * number of reasons, such as the bio being full or target block device
709 * limitations. The target block device must allow bio's up to PAGE_SIZE,
710 * so it is always possible to add a single page to an empty bio.
711 *
712 * This should only be used by REQ_PC bios.
713 */
714 int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page
715 *page, unsigned int len, unsigned int offset)
716 {
717 int retried_segments = 0;
718 struct bio_vec *bvec;
719
720 /*
721 * cloned bio must not modify vec list
722 */
723 if (unlikely(bio_flagged(bio, BIO_CLONED)))
724 return 0;
725
726 if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q))
727 return 0;
728
729 /*
730 * For filesystems with a blocksize smaller than the pagesize
731 * we will often be called with the same page as last time and
732 * a consecutive offset. Optimize this special case.
733 */
734 if (bio->bi_vcnt > 0) {
735 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
736
737 if (page == prev->bv_page &&
738 offset == prev->bv_offset + prev->bv_len) {
739 prev->bv_len += len;
740 bio->bi_iter.bi_size += len;
741 goto done;
742 }
743
744 /*
745 * If the queue doesn't support SG gaps and adding this
746 * offset would create a gap, disallow it.
747 */
748 if (bvec_gap_to_prev(q, prev, offset))
749 return 0;
750 }
751
752 if (bio->bi_vcnt >= bio->bi_max_vecs)
753 return 0;
754
755 /*
756 * setup the new entry, we might clear it again later if we
757 * cannot add the page
758 */
759 bvec = &bio->bi_io_vec[bio->bi_vcnt];
760 bvec->bv_page = page;
761 bvec->bv_len = len;
762 bvec->bv_offset = offset;
763 bio->bi_vcnt++;
764 bio->bi_phys_segments++;
765 bio->bi_iter.bi_size += len;
766
767 /*
768 * Perform a recount if the number of segments is greater
769 * than queue_max_segments(q).
770 */
771
772 while (bio->bi_phys_segments > queue_max_segments(q)) {
773
774 if (retried_segments)
775 goto failed;
776
777 retried_segments = 1;
778 blk_recount_segments(q, bio);
779 }
780
781 /* If we may be able to merge these biovecs, force a recount */
782 if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
783 bio_clear_flag(bio, BIO_SEG_VALID);
784
785 done:
786 return len;
787
788 failed:
789 bvec->bv_page = NULL;
790 bvec->bv_len = 0;
791 bvec->bv_offset = 0;
792 bio->bi_vcnt--;
793 bio->bi_iter.bi_size -= len;
794 blk_recount_segments(q, bio);
795 return 0;
796 }
797 EXPORT_SYMBOL(bio_add_pc_page);
798
799 /**
800 * bio_add_page - attempt to add page to bio
801 * @bio: destination bio
802 * @page: page to add
803 * @len: vec entry length
804 * @offset: vec entry offset
805 *
806 * Attempt to add a page to the bio_vec maplist. This will only fail
807 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
808 */
809 int bio_add_page(struct bio *bio, struct page *page,
810 unsigned int len, unsigned int offset)
811 {
812 struct bio_vec *bv;
813
814 /*
815 * cloned bio must not modify vec list
816 */
817 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
818 return 0;
819
820 /*
821 * For filesystems with a blocksize smaller than the pagesize
822 * we will often be called with the same page as last time and
823 * a consecutive offset. Optimize this special case.
824 */
825 if (bio->bi_vcnt > 0) {
826 bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
827
828 if (page == bv->bv_page &&
829 offset == bv->bv_offset + bv->bv_len) {
830 bv->bv_len += len;
831 goto done;
832 }
833 }
834
835 if (bio->bi_vcnt >= bio->bi_max_vecs)
836 return 0;
837
838 bv = &bio->bi_io_vec[bio->bi_vcnt];
839 bv->bv_page = page;
840 bv->bv_len = len;
841 bv->bv_offset = offset;
842
843 bio->bi_vcnt++;
844 done:
845 bio->bi_iter.bi_size += len;
846 return len;
847 }
848 EXPORT_SYMBOL(bio_add_page);
849
850 struct submit_bio_ret {
851 struct completion event;
852 int error;
853 };
854
855 static void submit_bio_wait_endio(struct bio *bio)
856 {
857 struct submit_bio_ret *ret = bio->bi_private;
858
859 ret->error = bio->bi_error;
860 complete(&ret->event);
861 }
862
863 /**
864 * submit_bio_wait - submit a bio, and wait until it completes
865 * @bio: The &struct bio which describes the I/O
866 *
867 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
868 * bio_endio() on failure.
869 */
870 int submit_bio_wait(struct bio *bio)
871 {
872 struct submit_bio_ret ret;
873
874 init_completion(&ret.event);
875 bio->bi_private = &ret;
876 bio->bi_end_io = submit_bio_wait_endio;
877 bio->bi_opf |= REQ_SYNC;
878 submit_bio(bio);
879 wait_for_completion_io(&ret.event);
880
881 return ret.error;
882 }
883 EXPORT_SYMBOL(submit_bio_wait);
884
885 /**
886 * bio_advance - increment/complete a bio by some number of bytes
887 * @bio: bio to advance
888 * @bytes: number of bytes to complete
889 *
890 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
891 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
892 * be updated on the last bvec as well.
893 *
894 * @bio will then represent the remaining, uncompleted portion of the io.
895 */
896 void bio_advance(struct bio *bio, unsigned bytes)
897 {
898 if (bio_integrity(bio))
899 bio_integrity_advance(bio, bytes);
900
901 bio_advance_iter(bio, &bio->bi_iter, bytes);
902 }
903 EXPORT_SYMBOL(bio_advance);
904
905 /**
906 * bio_alloc_pages - allocates a single page for each bvec in a bio
907 * @bio: bio to allocate pages for
908 * @gfp_mask: flags for allocation
909 *
910 * Allocates pages up to @bio->bi_vcnt.
911 *
912 * Returns 0 on success, -ENOMEM on failure. On failure, any allocated pages are
913 * freed.
914 */
915 int bio_alloc_pages(struct bio *bio, gfp_t gfp_mask)
916 {
917 int i;
918 struct bio_vec *bv;
919
920 bio_for_each_segment_all(bv, bio, i) {
921 bv->bv_page = alloc_page(gfp_mask);
922 if (!bv->bv_page) {
923 while (--bv >= bio->bi_io_vec)
924 __free_page(bv->bv_page);
925 return -ENOMEM;
926 }
927 }
928
929 return 0;
930 }
931 EXPORT_SYMBOL(bio_alloc_pages);
932
933 /**
934 * bio_copy_data - copy contents of data buffers from one chain of bios to
935 * another
936 * @src: source bio list
937 * @dst: destination bio list
938 *
939 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
940 * @src and @dst as linked lists of bios.
941 *
942 * Stops when it reaches the end of either @src or @dst - that is, copies
943 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
944 */
945 void bio_copy_data(struct bio *dst, struct bio *src)
946 {
947 struct bvec_iter src_iter, dst_iter;
948 struct bio_vec src_bv, dst_bv;
949 void *src_p, *dst_p;
950 unsigned bytes;
951
952 src_iter = src->bi_iter;
953 dst_iter = dst->bi_iter;
954
955 while (1) {
956 if (!src_iter.bi_size) {
957 src = src->bi_next;
958 if (!src)
959 break;
960
961 src_iter = src->bi_iter;
962 }
963
964 if (!dst_iter.bi_size) {
965 dst = dst->bi_next;
966 if (!dst)
967 break;
968
969 dst_iter = dst->bi_iter;
970 }
971
972 src_bv = bio_iter_iovec(src, src_iter);
973 dst_bv = bio_iter_iovec(dst, dst_iter);
974
975 bytes = min(src_bv.bv_len, dst_bv.bv_len);
976
977 src_p = kmap_atomic(src_bv.bv_page);
978 dst_p = kmap_atomic(dst_bv.bv_page);
979
980 memcpy(dst_p + dst_bv.bv_offset,
981 src_p + src_bv.bv_offset,
982 bytes);
983
984 kunmap_atomic(dst_p);
985 kunmap_atomic(src_p);
986
987 bio_advance_iter(src, &src_iter, bytes);
988 bio_advance_iter(dst, &dst_iter, bytes);
989 }
990 }
991 EXPORT_SYMBOL(bio_copy_data);
992
993 struct bio_map_data {
994 int is_our_pages;
995 struct iov_iter iter;
996 struct iovec iov[];
997 };
998
999 static struct bio_map_data *bio_alloc_map_data(unsigned int iov_count,
1000 gfp_t gfp_mask)
1001 {
1002 if (iov_count > UIO_MAXIOV)
1003 return NULL;
1004
1005 return kmalloc(sizeof(struct bio_map_data) +
1006 sizeof(struct iovec) * iov_count, gfp_mask);
1007 }
1008
1009 /**
1010 * bio_copy_from_iter - copy all pages from iov_iter to bio
1011 * @bio: The &struct bio which describes the I/O as destination
1012 * @iter: iov_iter as source
1013 *
1014 * Copy all pages from iov_iter to bio.
1015 * Returns 0 on success, or error on failure.
1016 */
1017 static int bio_copy_from_iter(struct bio *bio, struct iov_iter iter)
1018 {
1019 int i;
1020 struct bio_vec *bvec;
1021
1022 bio_for_each_segment_all(bvec, bio, i) {
1023 ssize_t ret;
1024
1025 ret = copy_page_from_iter(bvec->bv_page,
1026 bvec->bv_offset,
1027 bvec->bv_len,
1028 &iter);
1029
1030 if (!iov_iter_count(&iter))
1031 break;
1032
1033 if (ret < bvec->bv_len)
1034 return -EFAULT;
1035 }
1036
1037 return 0;
1038 }
1039
1040 /**
1041 * bio_copy_to_iter - copy all pages from bio to iov_iter
1042 * @bio: The &struct bio which describes the I/O as source
1043 * @iter: iov_iter as destination
1044 *
1045 * Copy all pages from bio to iov_iter.
1046 * Returns 0 on success, or error on failure.
1047 */
1048 static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
1049 {
1050 int i;
1051 struct bio_vec *bvec;
1052
1053 bio_for_each_segment_all(bvec, bio, i) {
1054 ssize_t ret;
1055
1056 ret = copy_page_to_iter(bvec->bv_page,
1057 bvec->bv_offset,
1058 bvec->bv_len,
1059 &iter);
1060
1061 if (!iov_iter_count(&iter))
1062 break;
1063
1064 if (ret < bvec->bv_len)
1065 return -EFAULT;
1066 }
1067
1068 return 0;
1069 }
1070
1071 void bio_free_pages(struct bio *bio)
1072 {
1073 struct bio_vec *bvec;
1074 int i;
1075
1076 bio_for_each_segment_all(bvec, bio, i)
1077 __free_page(bvec->bv_page);
1078 }
1079 EXPORT_SYMBOL(bio_free_pages);
1080
1081 /**
1082 * bio_uncopy_user - finish previously mapped bio
1083 * @bio: bio being terminated
1084 *
1085 * Free pages allocated from bio_copy_user_iov() and write back data
1086 * to user space in case of a read.
1087 */
1088 int bio_uncopy_user(struct bio *bio)
1089 {
1090 struct bio_map_data *bmd = bio->bi_private;
1091 int ret = 0;
1092
1093 if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1094 /*
1095 * if we're in a workqueue, the request is orphaned, so
1096 * don't copy into a random user address space, just free
1097 * and return -EINTR so user space doesn't expect any data.
1098 */
1099 if (!current->mm)
1100 ret = -EINTR;
1101 else if (bio_data_dir(bio) == READ)
1102 ret = bio_copy_to_iter(bio, bmd->iter);
1103 if (bmd->is_our_pages)
1104 bio_free_pages(bio);
1105 }
1106 kfree(bmd);
1107 bio_put(bio);
1108 return ret;
1109 }
1110
1111 /**
1112 * bio_copy_user_iov - copy user data to bio
1113 * @q: destination block queue
1114 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1115 * @iter: iovec iterator
1116 * @gfp_mask: memory allocation flags
1117 *
1118 * Prepares and returns a bio for indirect user io, bouncing data
1119 * to/from kernel pages as necessary. Must be paired with
1120 * call bio_uncopy_user() on io completion.
1121 */
1122 struct bio *bio_copy_user_iov(struct request_queue *q,
1123 struct rq_map_data *map_data,
1124 const struct iov_iter *iter,
1125 gfp_t gfp_mask)
1126 {
1127 struct bio_map_data *bmd;
1128 struct page *page;
1129 struct bio *bio;
1130 int i, ret;
1131 int nr_pages = 0;
1132 unsigned int len = iter->count;
1133 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
1134
1135 for (i = 0; i < iter->nr_segs; i++) {
1136 unsigned long uaddr;
1137 unsigned long end;
1138 unsigned long start;
1139
1140 uaddr = (unsigned long) iter->iov[i].iov_base;
1141 end = (uaddr + iter->iov[i].iov_len + PAGE_SIZE - 1)
1142 >> PAGE_SHIFT;
1143 start = uaddr >> PAGE_SHIFT;
1144
1145 /*
1146 * Overflow, abort
1147 */
1148 if (end < start)
1149 return ERR_PTR(-EINVAL);
1150
1151 nr_pages += end - start;
1152 }
1153
1154 if (offset)
1155 nr_pages++;
1156
1157 bmd = bio_alloc_map_data(iter->nr_segs, gfp_mask);
1158 if (!bmd)
1159 return ERR_PTR(-ENOMEM);
1160
1161 /*
1162 * We need to do a deep copy of the iov_iter including the iovecs.
1163 * The caller provided iov might point to an on-stack or otherwise
1164 * shortlived one.
1165 */
1166 bmd->is_our_pages = map_data ? 0 : 1;
1167 memcpy(bmd->iov, iter->iov, sizeof(struct iovec) * iter->nr_segs);
1168 iov_iter_init(&bmd->iter, iter->type, bmd->iov,
1169 iter->nr_segs, iter->count);
1170
1171 ret = -ENOMEM;
1172 bio = bio_kmalloc(gfp_mask, nr_pages);
1173 if (!bio)
1174 goto out_bmd;
1175
1176 if (iter->type & WRITE)
1177 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
1178
1179 ret = 0;
1180
1181 if (map_data) {
1182 nr_pages = 1 << map_data->page_order;
1183 i = map_data->offset / PAGE_SIZE;
1184 }
1185 while (len) {
1186 unsigned int bytes = PAGE_SIZE;
1187
1188 bytes -= offset;
1189
1190 if (bytes > len)
1191 bytes = len;
1192
1193 if (map_data) {
1194 if (i == map_data->nr_entries * nr_pages) {
1195 ret = -ENOMEM;
1196 break;
1197 }
1198
1199 page = map_data->pages[i / nr_pages];
1200 page += (i % nr_pages);
1201
1202 i++;
1203 } else {
1204 page = alloc_page(q->bounce_gfp | gfp_mask);
1205 if (!page) {
1206 ret = -ENOMEM;
1207 break;
1208 }
1209 }
1210
1211 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
1212 break;
1213
1214 len -= bytes;
1215 offset = 0;
1216 }
1217
1218 if (ret)
1219 goto cleanup;
1220
1221 /*
1222 * success
1223 */
1224 if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) ||
1225 (map_data && map_data->from_user)) {
1226 ret = bio_copy_from_iter(bio, *iter);
1227 if (ret)
1228 goto cleanup;
1229 }
1230
1231 bio->bi_private = bmd;
1232 return bio;
1233 cleanup:
1234 if (!map_data)
1235 bio_free_pages(bio);
1236 bio_put(bio);
1237 out_bmd:
1238 kfree(bmd);
1239 return ERR_PTR(ret);
1240 }
1241
1242 /**
1243 * bio_map_user_iov - map user iovec into bio
1244 * @q: the struct request_queue for the bio
1245 * @iter: iovec iterator
1246 * @gfp_mask: memory allocation flags
1247 *
1248 * Map the user space address into a bio suitable for io to a block
1249 * device. Returns an error pointer in case of error.
1250 */
1251 struct bio *bio_map_user_iov(struct request_queue *q,
1252 const struct iov_iter *iter,
1253 gfp_t gfp_mask)
1254 {
1255 int j;
1256 int nr_pages = 0;
1257 struct page **pages;
1258 struct bio *bio;
1259 int cur_page = 0;
1260 int ret, offset;
1261 struct iov_iter i;
1262 struct iovec iov;
1263
1264 iov_for_each(iov, i, *iter) {
1265 unsigned long uaddr = (unsigned long) iov.iov_base;
1266 unsigned long len = iov.iov_len;
1267 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1268 unsigned long start = uaddr >> PAGE_SHIFT;
1269
1270 /*
1271 * Overflow, abort
1272 */
1273 if (end < start)
1274 return ERR_PTR(-EINVAL);
1275
1276 nr_pages += end - start;
1277 /*
1278 * buffer must be aligned to at least logical block size for now
1279 */
1280 if (uaddr & queue_dma_alignment(q))
1281 return ERR_PTR(-EINVAL);
1282 }
1283
1284 if (!nr_pages)
1285 return ERR_PTR(-EINVAL);
1286
1287 bio = bio_kmalloc(gfp_mask, nr_pages);
1288 if (!bio)
1289 return ERR_PTR(-ENOMEM);
1290
1291 ret = -ENOMEM;
1292 pages = kcalloc(nr_pages, sizeof(struct page *), gfp_mask);
1293 if (!pages)
1294 goto out;
1295
1296 iov_for_each(iov, i, *iter) {
1297 unsigned long uaddr = (unsigned long) iov.iov_base;
1298 unsigned long len = iov.iov_len;
1299 unsigned long end = (uaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1300 unsigned long start = uaddr >> PAGE_SHIFT;
1301 const int local_nr_pages = end - start;
1302 const int page_limit = cur_page + local_nr_pages;
1303
1304 ret = get_user_pages_fast(uaddr, local_nr_pages,
1305 (iter->type & WRITE) != WRITE,
1306 &pages[cur_page]);
1307 if (ret < local_nr_pages) {
1308 ret = -EFAULT;
1309 goto out_unmap;
1310 }
1311
1312 offset = offset_in_page(uaddr);
1313 for (j = cur_page; j < page_limit; j++) {
1314 unsigned int bytes = PAGE_SIZE - offset;
1315
1316 if (len <= 0)
1317 break;
1318
1319 if (bytes > len)
1320 bytes = len;
1321
1322 /*
1323 * sorry...
1324 */
1325 if (bio_add_pc_page(q, bio, pages[j], bytes, offset) <
1326 bytes)
1327 break;
1328
1329 len -= bytes;
1330 offset = 0;
1331 }
1332
1333 cur_page = j;
1334 /*
1335 * release the pages we didn't map into the bio, if any
1336 */
1337 while (j < page_limit)
1338 put_page(pages[j++]);
1339 }
1340
1341 kfree(pages);
1342
1343 /*
1344 * set data direction, and check if mapped pages need bouncing
1345 */
1346 if (iter->type & WRITE)
1347 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
1348
1349 bio_set_flag(bio, BIO_USER_MAPPED);
1350
1351 /*
1352 * subtle -- if __bio_map_user() ended up bouncing a bio,
1353 * it would normally disappear when its bi_end_io is run.
1354 * however, we need it for the unmap, so grab an extra
1355 * reference to it
1356 */
1357 bio_get(bio);
1358 return bio;
1359
1360 out_unmap:
1361 for (j = 0; j < nr_pages; j++) {
1362 if (!pages[j])
1363 break;
1364 put_page(pages[j]);
1365 }
1366 out:
1367 kfree(pages);
1368 bio_put(bio);
1369 return ERR_PTR(ret);
1370 }
1371
1372 static void __bio_unmap_user(struct bio *bio)
1373 {
1374 struct bio_vec *bvec;
1375 int i;
1376
1377 /*
1378 * make sure we dirty pages we wrote to
1379 */
1380 bio_for_each_segment_all(bvec, bio, i) {
1381 if (bio_data_dir(bio) == READ)
1382 set_page_dirty_lock(bvec->bv_page);
1383
1384 put_page(bvec->bv_page);
1385 }
1386
1387 bio_put(bio);
1388 }
1389
1390 /**
1391 * bio_unmap_user - unmap a bio
1392 * @bio: the bio being unmapped
1393 *
1394 * Unmap a bio previously mapped by bio_map_user(). Must be called with
1395 * a process context.
1396 *
1397 * bio_unmap_user() may sleep.
1398 */
1399 void bio_unmap_user(struct bio *bio)
1400 {
1401 __bio_unmap_user(bio);
1402 bio_put(bio);
1403 }
1404
1405 static void bio_map_kern_endio(struct bio *bio)
1406 {
1407 bio_put(bio);
1408 }
1409
1410 /**
1411 * bio_map_kern - map kernel address into bio
1412 * @q: the struct request_queue for the bio
1413 * @data: pointer to buffer to map
1414 * @len: length in bytes
1415 * @gfp_mask: allocation flags for bio allocation
1416 *
1417 * Map the kernel address into a bio suitable for io to a block
1418 * device. Returns an error pointer in case of error.
1419 */
1420 struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1421 gfp_t gfp_mask)
1422 {
1423 unsigned long kaddr = (unsigned long)data;
1424 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1425 unsigned long start = kaddr >> PAGE_SHIFT;
1426 const int nr_pages = end - start;
1427 int offset, i;
1428 struct bio *bio;
1429
1430 bio = bio_kmalloc(gfp_mask, nr_pages);
1431 if (!bio)
1432 return ERR_PTR(-ENOMEM);
1433
1434 offset = offset_in_page(kaddr);
1435 for (i = 0; i < nr_pages; i++) {
1436 unsigned int bytes = PAGE_SIZE - offset;
1437
1438 if (len <= 0)
1439 break;
1440
1441 if (bytes > len)
1442 bytes = len;
1443
1444 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1445 offset) < bytes) {
1446 /* we don't support partial mappings */
1447 bio_put(bio);
1448 return ERR_PTR(-EINVAL);
1449 }
1450
1451 data += bytes;
1452 len -= bytes;
1453 offset = 0;
1454 }
1455
1456 bio->bi_end_io = bio_map_kern_endio;
1457 return bio;
1458 }
1459 EXPORT_SYMBOL(bio_map_kern);
1460
1461 static void bio_copy_kern_endio(struct bio *bio)
1462 {
1463 bio_free_pages(bio);
1464 bio_put(bio);
1465 }
1466
1467 static void bio_copy_kern_endio_read(struct bio *bio)
1468 {
1469 char *p = bio->bi_private;
1470 struct bio_vec *bvec;
1471 int i;
1472
1473 bio_for_each_segment_all(bvec, bio, i) {
1474 memcpy(p, page_address(bvec->bv_page), bvec->bv_len);
1475 p += bvec->bv_len;
1476 }
1477
1478 bio_copy_kern_endio(bio);
1479 }
1480
1481 /**
1482 * bio_copy_kern - copy kernel address into bio
1483 * @q: the struct request_queue for the bio
1484 * @data: pointer to buffer to copy
1485 * @len: length in bytes
1486 * @gfp_mask: allocation flags for bio and page allocation
1487 * @reading: data direction is READ
1488 *
1489 * copy the kernel address into a bio suitable for io to a block
1490 * device. Returns an error pointer in case of error.
1491 */
1492 struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1493 gfp_t gfp_mask, int reading)
1494 {
1495 unsigned long kaddr = (unsigned long)data;
1496 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1497 unsigned long start = kaddr >> PAGE_SHIFT;
1498 struct bio *bio;
1499 void *p = data;
1500 int nr_pages = 0;
1501
1502 /*
1503 * Overflow, abort
1504 */
1505 if (end < start)
1506 return ERR_PTR(-EINVAL);
1507
1508 nr_pages = end - start;
1509 bio = bio_kmalloc(gfp_mask, nr_pages);
1510 if (!bio)
1511 return ERR_PTR(-ENOMEM);
1512
1513 while (len) {
1514 struct page *page;
1515 unsigned int bytes = PAGE_SIZE;
1516
1517 if (bytes > len)
1518 bytes = len;
1519
1520 page = alloc_page(q->bounce_gfp | gfp_mask);
1521 if (!page)
1522 goto cleanup;
1523
1524 if (!reading)
1525 memcpy(page_address(page), p, bytes);
1526
1527 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
1528 break;
1529
1530 len -= bytes;
1531 p += bytes;
1532 }
1533
1534 if (reading) {
1535 bio->bi_end_io = bio_copy_kern_endio_read;
1536 bio->bi_private = data;
1537 } else {
1538 bio->bi_end_io = bio_copy_kern_endio;
1539 bio_set_op_attrs(bio, REQ_OP_WRITE, 0);
1540 }
1541
1542 return bio;
1543
1544 cleanup:
1545 bio_free_pages(bio);
1546 bio_put(bio);
1547 return ERR_PTR(-ENOMEM);
1548 }
1549
1550 /*
1551 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1552 * for performing direct-IO in BIOs.
1553 *
1554 * The problem is that we cannot run set_page_dirty() from interrupt context
1555 * because the required locks are not interrupt-safe. So what we can do is to
1556 * mark the pages dirty _before_ performing IO. And in interrupt context,
1557 * check that the pages are still dirty. If so, fine. If not, redirty them
1558 * in process context.
1559 *
1560 * We special-case compound pages here: normally this means reads into hugetlb
1561 * pages. The logic in here doesn't really work right for compound pages
1562 * because the VM does not uniformly chase down the head page in all cases.
1563 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1564 * handle them at all. So we skip compound pages here at an early stage.
1565 *
1566 * Note that this code is very hard to test under normal circumstances because
1567 * direct-io pins the pages with get_user_pages(). This makes
1568 * is_page_cache_freeable return false, and the VM will not clean the pages.
1569 * But other code (eg, flusher threads) could clean the pages if they are mapped
1570 * pagecache.
1571 *
1572 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1573 * deferred bio dirtying paths.
1574 */
1575
1576 /*
1577 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1578 */
1579 void bio_set_pages_dirty(struct bio *bio)
1580 {
1581 struct bio_vec *bvec;
1582 int i;
1583
1584 bio_for_each_segment_all(bvec, bio, i) {
1585 struct page *page = bvec->bv_page;
1586
1587 if (page && !PageCompound(page))
1588 set_page_dirty_lock(page);
1589 }
1590 }
1591
1592 static void bio_release_pages(struct bio *bio)
1593 {
1594 struct bio_vec *bvec;
1595 int i;
1596
1597 bio_for_each_segment_all(bvec, bio, i) {
1598 struct page *page = bvec->bv_page;
1599
1600 if (page)
1601 put_page(page);
1602 }
1603 }
1604
1605 /*
1606 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1607 * If they are, then fine. If, however, some pages are clean then they must
1608 * have been written out during the direct-IO read. So we take another ref on
1609 * the BIO and the offending pages and re-dirty the pages in process context.
1610 *
1611 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1612 * here on. It will run one put_page() against each page and will run one
1613 * bio_put() against the BIO.
1614 */
1615
1616 static void bio_dirty_fn(struct work_struct *work);
1617
1618 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1619 static DEFINE_SPINLOCK(bio_dirty_lock);
1620 static struct bio *bio_dirty_list;
1621
1622 /*
1623 * This runs in process context
1624 */
1625 static void bio_dirty_fn(struct work_struct *work)
1626 {
1627 unsigned long flags;
1628 struct bio *bio;
1629
1630 spin_lock_irqsave(&bio_dirty_lock, flags);
1631 bio = bio_dirty_list;
1632 bio_dirty_list = NULL;
1633 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1634
1635 while (bio) {
1636 struct bio *next = bio->bi_private;
1637
1638 bio_set_pages_dirty(bio);
1639 bio_release_pages(bio);
1640 bio_put(bio);
1641 bio = next;
1642 }
1643 }
1644
1645 void bio_check_pages_dirty(struct bio *bio)
1646 {
1647 struct bio_vec *bvec;
1648 int nr_clean_pages = 0;
1649 int i;
1650
1651 bio_for_each_segment_all(bvec, bio, i) {
1652 struct page *page = bvec->bv_page;
1653
1654 if (PageDirty(page) || PageCompound(page)) {
1655 put_page(page);
1656 bvec->bv_page = NULL;
1657 } else {
1658 nr_clean_pages++;
1659 }
1660 }
1661
1662 if (nr_clean_pages) {
1663 unsigned long flags;
1664
1665 spin_lock_irqsave(&bio_dirty_lock, flags);
1666 bio->bi_private = bio_dirty_list;
1667 bio_dirty_list = bio;
1668 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1669 schedule_work(&bio_dirty_work);
1670 } else {
1671 bio_put(bio);
1672 }
1673 }
1674
1675 void generic_start_io_acct(int rw, unsigned long sectors,
1676 struct hd_struct *part)
1677 {
1678 int cpu = part_stat_lock();
1679
1680 part_round_stats(cpu, part);
1681 part_stat_inc(cpu, part, ios[rw]);
1682 part_stat_add(cpu, part, sectors[rw], sectors);
1683 part_inc_in_flight(part, rw);
1684
1685 part_stat_unlock();
1686 }
1687 EXPORT_SYMBOL(generic_start_io_acct);
1688
1689 void generic_end_io_acct(int rw, struct hd_struct *part,
1690 unsigned long start_time)
1691 {
1692 unsigned long duration = jiffies - start_time;
1693 int cpu = part_stat_lock();
1694
1695 part_stat_add(cpu, part, ticks[rw], duration);
1696 part_round_stats(cpu, part);
1697 part_dec_in_flight(part, rw);
1698
1699 part_stat_unlock();
1700 }
1701 EXPORT_SYMBOL(generic_end_io_acct);
1702
1703 #if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1704 void bio_flush_dcache_pages(struct bio *bi)
1705 {
1706 struct bio_vec bvec;
1707 struct bvec_iter iter;
1708
1709 bio_for_each_segment(bvec, bi, iter)
1710 flush_dcache_page(bvec.bv_page);
1711 }
1712 EXPORT_SYMBOL(bio_flush_dcache_pages);
1713 #endif
1714
1715 static inline bool bio_remaining_done(struct bio *bio)
1716 {
1717 /*
1718 * If we're not chaining, then ->__bi_remaining is always 1 and
1719 * we always end io on the first invocation.
1720 */
1721 if (!bio_flagged(bio, BIO_CHAIN))
1722 return true;
1723
1724 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1725
1726 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1727 bio_clear_flag(bio, BIO_CHAIN);
1728 return true;
1729 }
1730
1731 return false;
1732 }
1733
1734 /**
1735 * bio_endio - end I/O on a bio
1736 * @bio: bio
1737 *
1738 * Description:
1739 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1740 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1741 * bio unless they own it and thus know that it has an end_io function.
1742 **/
1743 void bio_endio(struct bio *bio)
1744 {
1745 again:
1746 if (!bio_remaining_done(bio))
1747 return;
1748
1749 /*
1750 * Need to have a real endio function for chained bios, otherwise
1751 * various corner cases will break (like stacking block devices that
1752 * save/restore bi_end_io) - however, we want to avoid unbounded
1753 * recursion and blowing the stack. Tail call optimization would
1754 * handle this, but compiling with frame pointers also disables
1755 * gcc's sibling call optimization.
1756 */
1757 if (bio->bi_end_io == bio_chain_endio) {
1758 bio = __bio_chain_endio(bio);
1759 goto again;
1760 }
1761
1762 if (bio->bi_end_io)
1763 bio->bi_end_io(bio);
1764 }
1765 EXPORT_SYMBOL(bio_endio);
1766
1767 /**
1768 * bio_split - split a bio
1769 * @bio: bio to split
1770 * @sectors: number of sectors to split from the front of @bio
1771 * @gfp: gfp mask
1772 * @bs: bio set to allocate from
1773 *
1774 * Allocates and returns a new bio which represents @sectors from the start of
1775 * @bio, and updates @bio to represent the remaining sectors.
1776 *
1777 * Unless this is a discard request the newly allocated bio will point
1778 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1779 * @bio is not freed before the split.
1780 */
1781 struct bio *bio_split(struct bio *bio, int sectors,
1782 gfp_t gfp, struct bio_set *bs)
1783 {
1784 struct bio *split = NULL;
1785
1786 BUG_ON(sectors <= 0);
1787 BUG_ON(sectors >= bio_sectors(bio));
1788
1789 /*
1790 * Discards need a mutable bio_vec to accommodate the payload
1791 * required by the DSM TRIM and UNMAP commands.
1792 */
1793 if (bio_op(bio) == REQ_OP_DISCARD || bio_op(bio) == REQ_OP_SECURE_ERASE)
1794 split = bio_clone_bioset(bio, gfp, bs);
1795 else
1796 split = bio_clone_fast(bio, gfp, bs);
1797
1798 if (!split)
1799 return NULL;
1800
1801 split->bi_iter.bi_size = sectors << 9;
1802
1803 if (bio_integrity(split))
1804 bio_integrity_trim(split, 0, sectors);
1805
1806 bio_advance(bio, split->bi_iter.bi_size);
1807
1808 return split;
1809 }
1810 EXPORT_SYMBOL(bio_split);
1811
1812 /**
1813 * bio_trim - trim a bio
1814 * @bio: bio to trim
1815 * @offset: number of sectors to trim from the front of @bio
1816 * @size: size we want to trim @bio to, in sectors
1817 */
1818 void bio_trim(struct bio *bio, int offset, int size)
1819 {
1820 /* 'bio' is a cloned bio which we need to trim to match
1821 * the given offset and size.
1822 */
1823
1824 size <<= 9;
1825 if (offset == 0 && size == bio->bi_iter.bi_size)
1826 return;
1827
1828 bio_clear_flag(bio, BIO_SEG_VALID);
1829
1830 bio_advance(bio, offset << 9);
1831
1832 bio->bi_iter.bi_size = size;
1833 }
1834 EXPORT_SYMBOL_GPL(bio_trim);
1835
1836 /*
1837 * create memory pools for biovec's in a bio_set.
1838 * use the global biovec slabs created for general use.
1839 */
1840 mempool_t *biovec_create_pool(int pool_entries)
1841 {
1842 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX;
1843
1844 return mempool_create_slab_pool(pool_entries, bp->slab);
1845 }
1846
1847 void bioset_free(struct bio_set *bs)
1848 {
1849 if (bs->rescue_workqueue)
1850 destroy_workqueue(bs->rescue_workqueue);
1851
1852 if (bs->bio_pool)
1853 mempool_destroy(bs->bio_pool);
1854
1855 if (bs->bvec_pool)
1856 mempool_destroy(bs->bvec_pool);
1857
1858 bioset_integrity_free(bs);
1859 bio_put_slab(bs);
1860
1861 kfree(bs);
1862 }
1863 EXPORT_SYMBOL(bioset_free);
1864
1865 static struct bio_set *__bioset_create(unsigned int pool_size,
1866 unsigned int front_pad,
1867 bool create_bvec_pool)
1868 {
1869 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1870 struct bio_set *bs;
1871
1872 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1873 if (!bs)
1874 return NULL;
1875
1876 bs->front_pad = front_pad;
1877
1878 spin_lock_init(&bs->rescue_lock);
1879 bio_list_init(&bs->rescue_list);
1880 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1881
1882 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1883 if (!bs->bio_slab) {
1884 kfree(bs);
1885 return NULL;
1886 }
1887
1888 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1889 if (!bs->bio_pool)
1890 goto bad;
1891
1892 if (create_bvec_pool) {
1893 bs->bvec_pool = biovec_create_pool(pool_size);
1894 if (!bs->bvec_pool)
1895 goto bad;
1896 }
1897
1898 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1899 if (!bs->rescue_workqueue)
1900 goto bad;
1901
1902 return bs;
1903 bad:
1904 bioset_free(bs);
1905 return NULL;
1906 }
1907
1908 /**
1909 * bioset_create - Create a bio_set
1910 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1911 * @front_pad: Number of bytes to allocate in front of the returned bio
1912 *
1913 * Description:
1914 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1915 * to ask for a number of bytes to be allocated in front of the bio.
1916 * Front pad allocation is useful for embedding the bio inside
1917 * another structure, to avoid allocating extra data to go with the bio.
1918 * Note that the bio must be embedded at the END of that structure always,
1919 * or things will break badly.
1920 */
1921 struct bio_set *bioset_create(unsigned int pool_size, unsigned int front_pad)
1922 {
1923 return __bioset_create(pool_size, front_pad, true);
1924 }
1925 EXPORT_SYMBOL(bioset_create);
1926
1927 /**
1928 * bioset_create_nobvec - Create a bio_set without bio_vec mempool
1929 * @pool_size: Number of bio to cache in the mempool
1930 * @front_pad: Number of bytes to allocate in front of the returned bio
1931 *
1932 * Description:
1933 * Same functionality as bioset_create() except that mempool is not
1934 * created for bio_vecs. Saving some memory for bio_clone_fast() users.
1935 */
1936 struct bio_set *bioset_create_nobvec(unsigned int pool_size, unsigned int front_pad)
1937 {
1938 return __bioset_create(pool_size, front_pad, false);
1939 }
1940 EXPORT_SYMBOL(bioset_create_nobvec);
1941
1942 #ifdef CONFIG_BLK_CGROUP
1943
1944 /**
1945 * bio_associate_blkcg - associate a bio with the specified blkcg
1946 * @bio: target bio
1947 * @blkcg_css: css of the blkcg to associate
1948 *
1949 * Associate @bio with the blkcg specified by @blkcg_css. Block layer will
1950 * treat @bio as if it were issued by a task which belongs to the blkcg.
1951 *
1952 * This function takes an extra reference of @blkcg_css which will be put
1953 * when @bio is released. The caller must own @bio and is responsible for
1954 * synchronizing calls to this function.
1955 */
1956 int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css)
1957 {
1958 if (unlikely(bio->bi_css))
1959 return -EBUSY;
1960 css_get(blkcg_css);
1961 bio->bi_css = blkcg_css;
1962 return 0;
1963 }
1964 EXPORT_SYMBOL_GPL(bio_associate_blkcg);
1965
1966 /**
1967 * bio_associate_current - associate a bio with %current
1968 * @bio: target bio
1969 *
1970 * Associate @bio with %current if it hasn't been associated yet. Block
1971 * layer will treat @bio as if it were issued by %current no matter which
1972 * task actually issues it.
1973 *
1974 * This function takes an extra reference of @task's io_context and blkcg
1975 * which will be put when @bio is released. The caller must own @bio,
1976 * ensure %current->io_context exists, and is responsible for synchronizing
1977 * calls to this function.
1978 */
1979 int bio_associate_current(struct bio *bio)
1980 {
1981 struct io_context *ioc;
1982
1983 if (bio->bi_css)
1984 return -EBUSY;
1985
1986 ioc = current->io_context;
1987 if (!ioc)
1988 return -ENOENT;
1989
1990 get_io_context_active(ioc);
1991 bio->bi_ioc = ioc;
1992 bio->bi_css = task_get_css(current, io_cgrp_id);
1993 return 0;
1994 }
1995 EXPORT_SYMBOL_GPL(bio_associate_current);
1996
1997 /**
1998 * bio_disassociate_task - undo bio_associate_current()
1999 * @bio: target bio
2000 */
2001 void bio_disassociate_task(struct bio *bio)
2002 {
2003 if (bio->bi_ioc) {
2004 put_io_context(bio->bi_ioc);
2005 bio->bi_ioc = NULL;
2006 }
2007 if (bio->bi_css) {
2008 css_put(bio->bi_css);
2009 bio->bi_css = NULL;
2010 }
2011 }
2012
2013 /**
2014 * bio_clone_blkcg_association - clone blkcg association from src to dst bio
2015 * @dst: destination bio
2016 * @src: source bio
2017 */
2018 void bio_clone_blkcg_association(struct bio *dst, struct bio *src)
2019 {
2020 if (src->bi_css)
2021 WARN_ON(bio_associate_blkcg(dst, src->bi_css));
2022 }
2023
2024 #endif /* CONFIG_BLK_CGROUP */
2025
2026 static void __init biovec_init_slabs(void)
2027 {
2028 int i;
2029
2030 for (i = 0; i < BVEC_POOL_NR; i++) {
2031 int size;
2032 struct biovec_slab *bvs = bvec_slabs + i;
2033
2034 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2035 bvs->slab = NULL;
2036 continue;
2037 }
2038
2039 size = bvs->nr_vecs * sizeof(struct bio_vec);
2040 bvs->slab = kmem_cache_create(bvs->name, size, 0,
2041 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2042 }
2043 }
2044
2045 static int __init init_bio(void)
2046 {
2047 bio_slab_max = 2;
2048 bio_slab_nr = 0;
2049 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
2050 if (!bio_slabs)
2051 panic("bio: can't allocate bios\n");
2052
2053 bio_integrity_init();
2054 biovec_init_slabs();
2055
2056 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0);
2057 if (!fs_bio_set)
2058 panic("bio: can't allocate bios\n");
2059
2060 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
2061 panic("bio: can't create integrity pool\n");
2062
2063 return 0;
2064 }
2065 subsys_initcall(init_bio);