Loading...
Note: File does not exist in v3.1.
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#include "blk.h"
34
35/*
36 * Test patch to inline a certain number of bi_io_vec's inside the bio
37 * itself, to shrink a bio data allocation from two mempool calls to one
38 */
39#define BIO_INLINE_VECS 4
40
41/*
42 * if you change this list, also change bvec_alloc or things will
43 * break badly! cannot be bigger than what you can fit into an
44 * unsigned short
45 */
46#define BV(x, n) { .nr_vecs = x, .name = "biovec-"#n }
47static struct biovec_slab bvec_slabs[BVEC_POOL_NR] __read_mostly = {
48 BV(1, 1), BV(4, 4), BV(16, 16), BV(64, 64), BV(128, 128), BV(BIO_MAX_PAGES, max),
49};
50#undef BV
51
52/*
53 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
54 * IO code that does not need private memory pools.
55 */
56struct bio_set *fs_bio_set;
57EXPORT_SYMBOL(fs_bio_set);
58
59/*
60 * Our slab pool management
61 */
62struct bio_slab {
63 struct kmem_cache *slab;
64 unsigned int slab_ref;
65 unsigned int slab_size;
66 char name[8];
67};
68static DEFINE_MUTEX(bio_slab_lock);
69static struct bio_slab *bio_slabs;
70static unsigned int bio_slab_nr, bio_slab_max;
71
72static struct kmem_cache *bio_find_or_create_slab(unsigned int extra_size)
73{
74 unsigned int sz = sizeof(struct bio) + extra_size;
75 struct kmem_cache *slab = NULL;
76 struct bio_slab *bslab, *new_bio_slabs;
77 unsigned int new_bio_slab_max;
78 unsigned int i, entry = -1;
79
80 mutex_lock(&bio_slab_lock);
81
82 i = 0;
83 while (i < bio_slab_nr) {
84 bslab = &bio_slabs[i];
85
86 if (!bslab->slab && entry == -1)
87 entry = i;
88 else if (bslab->slab_size == sz) {
89 slab = bslab->slab;
90 bslab->slab_ref++;
91 break;
92 }
93 i++;
94 }
95
96 if (slab)
97 goto out_unlock;
98
99 if (bio_slab_nr == bio_slab_max && entry == -1) {
100 new_bio_slab_max = bio_slab_max << 1;
101 new_bio_slabs = krealloc(bio_slabs,
102 new_bio_slab_max * sizeof(struct bio_slab),
103 GFP_KERNEL);
104 if (!new_bio_slabs)
105 goto out_unlock;
106 bio_slab_max = new_bio_slab_max;
107 bio_slabs = new_bio_slabs;
108 }
109 if (entry == -1)
110 entry = bio_slab_nr++;
111
112 bslab = &bio_slabs[entry];
113
114 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", entry);
115 slab = kmem_cache_create(bslab->name, sz, ARCH_KMALLOC_MINALIGN,
116 SLAB_HWCACHE_ALIGN, NULL);
117 if (!slab)
118 goto out_unlock;
119
120 bslab->slab = slab;
121 bslab->slab_ref = 1;
122 bslab->slab_size = sz;
123out_unlock:
124 mutex_unlock(&bio_slab_lock);
125 return slab;
126}
127
128static void bio_put_slab(struct bio_set *bs)
129{
130 struct bio_slab *bslab = NULL;
131 unsigned int i;
132
133 mutex_lock(&bio_slab_lock);
134
135 for (i = 0; i < bio_slab_nr; i++) {
136 if (bs->bio_slab == bio_slabs[i].slab) {
137 bslab = &bio_slabs[i];
138 break;
139 }
140 }
141
142 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
143 goto out;
144
145 WARN_ON(!bslab->slab_ref);
146
147 if (--bslab->slab_ref)
148 goto out;
149
150 kmem_cache_destroy(bslab->slab);
151 bslab->slab = NULL;
152
153out:
154 mutex_unlock(&bio_slab_lock);
155}
156
157unsigned int bvec_nr_vecs(unsigned short idx)
158{
159 return bvec_slabs[idx].nr_vecs;
160}
161
162void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned int idx)
163{
164 if (!idx)
165 return;
166 idx--;
167
168 BIO_BUG_ON(idx >= BVEC_POOL_NR);
169
170 if (idx == BVEC_POOL_MAX) {
171 mempool_free(bv, pool);
172 } else {
173 struct biovec_slab *bvs = bvec_slabs + idx;
174
175 kmem_cache_free(bvs->slab, bv);
176 }
177}
178
179struct bio_vec *bvec_alloc(gfp_t gfp_mask, int nr, unsigned long *idx,
180 mempool_t *pool)
181{
182 struct bio_vec *bvl;
183
184 /*
185 * see comment near bvec_array define!
186 */
187 switch (nr) {
188 case 1:
189 *idx = 0;
190 break;
191 case 2 ... 4:
192 *idx = 1;
193 break;
194 case 5 ... 16:
195 *idx = 2;
196 break;
197 case 17 ... 64:
198 *idx = 3;
199 break;
200 case 65 ... 128:
201 *idx = 4;
202 break;
203 case 129 ... BIO_MAX_PAGES:
204 *idx = 5;
205 break;
206 default:
207 return NULL;
208 }
209
210 /*
211 * idx now points to the pool we want to allocate from. only the
212 * 1-vec entry pool is mempool backed.
213 */
214 if (*idx == BVEC_POOL_MAX) {
215fallback:
216 bvl = mempool_alloc(pool, gfp_mask);
217 } else {
218 struct biovec_slab *bvs = bvec_slabs + *idx;
219 gfp_t __gfp_mask = gfp_mask & ~(__GFP_DIRECT_RECLAIM | __GFP_IO);
220
221 /*
222 * Make this allocation restricted and don't dump info on
223 * allocation failures, since we'll fallback to the mempool
224 * in case of failure.
225 */
226 __gfp_mask |= __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
227
228 /*
229 * Try a slab allocation. If this fails and __GFP_DIRECT_RECLAIM
230 * is set, retry with the 1-entry mempool
231 */
232 bvl = kmem_cache_alloc(bvs->slab, __gfp_mask);
233 if (unlikely(!bvl && (gfp_mask & __GFP_DIRECT_RECLAIM))) {
234 *idx = BVEC_POOL_MAX;
235 goto fallback;
236 }
237 }
238
239 (*idx)++;
240 return bvl;
241}
242
243void bio_uninit(struct bio *bio)
244{
245 bio_disassociate_task(bio);
246}
247EXPORT_SYMBOL(bio_uninit);
248
249static void bio_free(struct bio *bio)
250{
251 struct bio_set *bs = bio->bi_pool;
252 void *p;
253
254 bio_uninit(bio);
255
256 if (bs) {
257 bvec_free(bs->bvec_pool, bio->bi_io_vec, BVEC_POOL_IDX(bio));
258
259 /*
260 * If we have front padding, adjust the bio pointer before freeing
261 */
262 p = bio;
263 p -= bs->front_pad;
264
265 mempool_free(p, bs->bio_pool);
266 } else {
267 /* Bio was allocated by bio_kmalloc() */
268 kfree(bio);
269 }
270}
271
272/*
273 * Users of this function have their own bio allocation. Subsequently,
274 * they must remember to pair any call to bio_init() with bio_uninit()
275 * when IO has completed, or when the bio is released.
276 */
277void bio_init(struct bio *bio, struct bio_vec *table,
278 unsigned short max_vecs)
279{
280 memset(bio, 0, sizeof(*bio));
281 atomic_set(&bio->__bi_remaining, 1);
282 atomic_set(&bio->__bi_cnt, 1);
283
284 bio->bi_io_vec = table;
285 bio->bi_max_vecs = max_vecs;
286}
287EXPORT_SYMBOL(bio_init);
288
289/**
290 * bio_reset - reinitialize a bio
291 * @bio: bio to reset
292 *
293 * Description:
294 * After calling bio_reset(), @bio will be in the same state as a freshly
295 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
296 * preserved are the ones that are initialized by bio_alloc_bioset(). See
297 * comment in struct bio.
298 */
299void bio_reset(struct bio *bio)
300{
301 unsigned long flags = bio->bi_flags & (~0UL << BIO_RESET_BITS);
302
303 bio_uninit(bio);
304
305 memset(bio, 0, BIO_RESET_BYTES);
306 bio->bi_flags = flags;
307 atomic_set(&bio->__bi_remaining, 1);
308}
309EXPORT_SYMBOL(bio_reset);
310
311static struct bio *__bio_chain_endio(struct bio *bio)
312{
313 struct bio *parent = bio->bi_private;
314
315 if (!parent->bi_status)
316 parent->bi_status = bio->bi_status;
317 bio_put(bio);
318 return parent;
319}
320
321static void bio_chain_endio(struct bio *bio)
322{
323 bio_endio(__bio_chain_endio(bio));
324}
325
326/**
327 * bio_chain - chain bio completions
328 * @bio: the target bio
329 * @parent: the @bio's parent bio
330 *
331 * The caller won't have a bi_end_io called when @bio completes - instead,
332 * @parent's bi_end_io won't be called until both @parent and @bio have
333 * completed; the chained bio will also be freed when it completes.
334 *
335 * The caller must not set bi_private or bi_end_io in @bio.
336 */
337void bio_chain(struct bio *bio, struct bio *parent)
338{
339 BUG_ON(bio->bi_private || bio->bi_end_io);
340
341 bio->bi_private = parent;
342 bio->bi_end_io = bio_chain_endio;
343 bio_inc_remaining(parent);
344}
345EXPORT_SYMBOL(bio_chain);
346
347static void bio_alloc_rescue(struct work_struct *work)
348{
349 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
350 struct bio *bio;
351
352 while (1) {
353 spin_lock(&bs->rescue_lock);
354 bio = bio_list_pop(&bs->rescue_list);
355 spin_unlock(&bs->rescue_lock);
356
357 if (!bio)
358 break;
359
360 generic_make_request(bio);
361 }
362}
363
364static void punt_bios_to_rescuer(struct bio_set *bs)
365{
366 struct bio_list punt, nopunt;
367 struct bio *bio;
368
369 if (WARN_ON_ONCE(!bs->rescue_workqueue))
370 return;
371 /*
372 * In order to guarantee forward progress we must punt only bios that
373 * were allocated from this bio_set; otherwise, if there was a bio on
374 * there for a stacking driver higher up in the stack, processing it
375 * could require allocating bios from this bio_set, and doing that from
376 * our own rescuer would be bad.
377 *
378 * Since bio lists are singly linked, pop them all instead of trying to
379 * remove from the middle of the list:
380 */
381
382 bio_list_init(&punt);
383 bio_list_init(&nopunt);
384
385 while ((bio = bio_list_pop(¤t->bio_list[0])))
386 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
387 current->bio_list[0] = nopunt;
388
389 bio_list_init(&nopunt);
390 while ((bio = bio_list_pop(¤t->bio_list[1])))
391 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
392 current->bio_list[1] = nopunt;
393
394 spin_lock(&bs->rescue_lock);
395 bio_list_merge(&bs->rescue_list, &punt);
396 spin_unlock(&bs->rescue_lock);
397
398 queue_work(bs->rescue_workqueue, &bs->rescue_work);
399}
400
401/**
402 * bio_alloc_bioset - allocate a bio for I/O
403 * @gfp_mask: the GFP_* mask given to the slab allocator
404 * @nr_iovecs: number of iovecs to pre-allocate
405 * @bs: the bio_set to allocate from.
406 *
407 * Description:
408 * If @bs is NULL, uses kmalloc() to allocate the bio; else the allocation is
409 * backed by the @bs's mempool.
410 *
411 * When @bs is not NULL, if %__GFP_DIRECT_RECLAIM is set then bio_alloc will
412 * always be able to allocate a bio. This is due to the mempool guarantees.
413 * To make this work, callers must never allocate more than 1 bio at a time
414 * from this pool. Callers that need to allocate more than 1 bio must always
415 * submit the previously allocated bio for IO before attempting to allocate
416 * a new one. Failure to do so can cause deadlocks under memory pressure.
417 *
418 * Note that when running under generic_make_request() (i.e. any block
419 * driver), bios are not submitted until after you return - see the code in
420 * generic_make_request() that converts recursion into iteration, to prevent
421 * stack overflows.
422 *
423 * This would normally mean allocating multiple bios under
424 * generic_make_request() would be susceptible to deadlocks, but we have
425 * deadlock avoidance code that resubmits any blocked bios from a rescuer
426 * thread.
427 *
428 * However, we do not guarantee forward progress for allocations from other
429 * mempools. Doing multiple allocations from the same mempool under
430 * generic_make_request() should be avoided - instead, use bio_set's front_pad
431 * for per bio allocations.
432 *
433 * RETURNS:
434 * Pointer to new bio on success, NULL on failure.
435 */
436struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned int nr_iovecs,
437 struct bio_set *bs)
438{
439 gfp_t saved_gfp = gfp_mask;
440 unsigned front_pad;
441 unsigned inline_vecs;
442 struct bio_vec *bvl = NULL;
443 struct bio *bio;
444 void *p;
445
446 if (!bs) {
447 if (nr_iovecs > UIO_MAXIOV)
448 return NULL;
449
450 p = kmalloc(sizeof(struct bio) +
451 nr_iovecs * sizeof(struct bio_vec),
452 gfp_mask);
453 front_pad = 0;
454 inline_vecs = nr_iovecs;
455 } else {
456 /* should not use nobvec bioset for nr_iovecs > 0 */
457 if (WARN_ON_ONCE(!bs->bvec_pool && nr_iovecs > 0))
458 return NULL;
459 /*
460 * generic_make_request() converts recursion to iteration; this
461 * means if we're running beneath it, any bios we allocate and
462 * submit will not be submitted (and thus freed) until after we
463 * return.
464 *
465 * This exposes us to a potential deadlock if we allocate
466 * multiple bios from the same bio_set() while running
467 * underneath generic_make_request(). If we were to allocate
468 * multiple bios (say a stacking block driver that was splitting
469 * bios), we would deadlock if we exhausted the mempool's
470 * reserve.
471 *
472 * We solve this, and guarantee forward progress, with a rescuer
473 * workqueue per bio_set. If we go to allocate and there are
474 * bios on current->bio_list, we first try the allocation
475 * without __GFP_DIRECT_RECLAIM; if that fails, we punt those
476 * bios we would be blocking to the rescuer workqueue before
477 * we retry with the original gfp_flags.
478 */
479
480 if (current->bio_list &&
481 (!bio_list_empty(¤t->bio_list[0]) ||
482 !bio_list_empty(¤t->bio_list[1])) &&
483 bs->rescue_workqueue)
484 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
485
486 p = mempool_alloc(bs->bio_pool, gfp_mask);
487 if (!p && gfp_mask != saved_gfp) {
488 punt_bios_to_rescuer(bs);
489 gfp_mask = saved_gfp;
490 p = mempool_alloc(bs->bio_pool, gfp_mask);
491 }
492
493 front_pad = bs->front_pad;
494 inline_vecs = BIO_INLINE_VECS;
495 }
496
497 if (unlikely(!p))
498 return NULL;
499
500 bio = p + front_pad;
501 bio_init(bio, NULL, 0);
502
503 if (nr_iovecs > inline_vecs) {
504 unsigned long idx = 0;
505
506 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
507 if (!bvl && gfp_mask != saved_gfp) {
508 punt_bios_to_rescuer(bs);
509 gfp_mask = saved_gfp;
510 bvl = bvec_alloc(gfp_mask, nr_iovecs, &idx, bs->bvec_pool);
511 }
512
513 if (unlikely(!bvl))
514 goto err_free;
515
516 bio->bi_flags |= idx << BVEC_POOL_OFFSET;
517 } else if (nr_iovecs) {
518 bvl = bio->bi_inline_vecs;
519 }
520
521 bio->bi_pool = bs;
522 bio->bi_max_vecs = nr_iovecs;
523 bio->bi_io_vec = bvl;
524 return bio;
525
526err_free:
527 mempool_free(p, bs->bio_pool);
528 return NULL;
529}
530EXPORT_SYMBOL(bio_alloc_bioset);
531
532void zero_fill_bio(struct bio *bio)
533{
534 unsigned long flags;
535 struct bio_vec bv;
536 struct bvec_iter iter;
537
538 bio_for_each_segment(bv, bio, iter) {
539 char *data = bvec_kmap_irq(&bv, &flags);
540 memset(data, 0, bv.bv_len);
541 flush_dcache_page(bv.bv_page);
542 bvec_kunmap_irq(data, &flags);
543 }
544}
545EXPORT_SYMBOL(zero_fill_bio);
546
547/**
548 * bio_put - release a reference to a bio
549 * @bio: bio to release reference to
550 *
551 * Description:
552 * Put a reference to a &struct bio, either one you have gotten with
553 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
554 **/
555void bio_put(struct bio *bio)
556{
557 if (!bio_flagged(bio, BIO_REFFED))
558 bio_free(bio);
559 else {
560 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
561
562 /*
563 * last put frees it
564 */
565 if (atomic_dec_and_test(&bio->__bi_cnt))
566 bio_free(bio);
567 }
568}
569EXPORT_SYMBOL(bio_put);
570
571inline int bio_phys_segments(struct request_queue *q, struct bio *bio)
572{
573 if (unlikely(!bio_flagged(bio, BIO_SEG_VALID)))
574 blk_recount_segments(q, bio);
575
576 return bio->bi_phys_segments;
577}
578EXPORT_SYMBOL(bio_phys_segments);
579
580/**
581 * __bio_clone_fast - clone a bio that shares the original bio's biovec
582 * @bio: destination bio
583 * @bio_src: bio to clone
584 *
585 * Clone a &bio. Caller will own the returned bio, but not
586 * the actual data it points to. Reference count of returned
587 * bio will be one.
588 *
589 * Caller must ensure that @bio_src is not freed before @bio.
590 */
591void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
592{
593 BUG_ON(bio->bi_pool && BVEC_POOL_IDX(bio));
594
595 /*
596 * most users will be overriding ->bi_disk with a new target,
597 * so we don't set nor calculate new physical/hw segment counts here
598 */
599 bio->bi_disk = bio_src->bi_disk;
600 bio->bi_partno = bio_src->bi_partno;
601 bio_set_flag(bio, BIO_CLONED);
602 if (bio_flagged(bio_src, BIO_THROTTLED))
603 bio_set_flag(bio, BIO_THROTTLED);
604 bio->bi_opf = bio_src->bi_opf;
605 bio->bi_write_hint = bio_src->bi_write_hint;
606 bio->bi_iter = bio_src->bi_iter;
607 bio->bi_io_vec = bio_src->bi_io_vec;
608
609 bio_clone_blkcg_association(bio, bio_src);
610}
611EXPORT_SYMBOL(__bio_clone_fast);
612
613/**
614 * bio_clone_fast - clone a bio that shares the original bio's biovec
615 * @bio: bio to clone
616 * @gfp_mask: allocation priority
617 * @bs: bio_set to allocate from
618 *
619 * Like __bio_clone_fast, only also allocates the returned bio
620 */
621struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
622{
623 struct bio *b;
624
625 b = bio_alloc_bioset(gfp_mask, 0, bs);
626 if (!b)
627 return NULL;
628
629 __bio_clone_fast(b, bio);
630
631 if (bio_integrity(bio)) {
632 int ret;
633
634 ret = bio_integrity_clone(b, bio, gfp_mask);
635
636 if (ret < 0) {
637 bio_put(b);
638 return NULL;
639 }
640 }
641
642 return b;
643}
644EXPORT_SYMBOL(bio_clone_fast);
645
646/**
647 * bio_clone_bioset - clone a bio
648 * @bio_src: bio to clone
649 * @gfp_mask: allocation priority
650 * @bs: bio_set to allocate from
651 *
652 * Clone bio. Caller will own the returned bio, but not the actual data it
653 * points to. Reference count of returned bio will be one.
654 */
655struct bio *bio_clone_bioset(struct bio *bio_src, gfp_t gfp_mask,
656 struct bio_set *bs)
657{
658 struct bvec_iter iter;
659 struct bio_vec bv;
660 struct bio *bio;
661
662 /*
663 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
664 * bio_src->bi_io_vec to bio->bi_io_vec.
665 *
666 * We can't do that anymore, because:
667 *
668 * - The point of cloning the biovec is to produce a bio with a biovec
669 * the caller can modify: bi_idx and bi_bvec_done should be 0.
670 *
671 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
672 * we tried to clone the whole thing bio_alloc_bioset() would fail.
673 * But the clone should succeed as long as the number of biovecs we
674 * actually need to allocate is fewer than BIO_MAX_PAGES.
675 *
676 * - Lastly, bi_vcnt should not be looked at or relied upon by code
677 * that does not own the bio - reason being drivers don't use it for
678 * iterating over the biovec anymore, so expecting it to be kept up
679 * to date (i.e. for clones that share the parent biovec) is just
680 * asking for trouble and would force extra work on
681 * __bio_clone_fast() anyways.
682 */
683
684 bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs);
685 if (!bio)
686 return NULL;
687 bio->bi_disk = bio_src->bi_disk;
688 bio->bi_opf = bio_src->bi_opf;
689 bio->bi_write_hint = bio_src->bi_write_hint;
690 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector;
691 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
692
693 switch (bio_op(bio)) {
694 case REQ_OP_DISCARD:
695 case REQ_OP_SECURE_ERASE:
696 case REQ_OP_WRITE_ZEROES:
697 break;
698 case REQ_OP_WRITE_SAME:
699 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
700 break;
701 default:
702 bio_for_each_segment(bv, bio_src, iter)
703 bio->bi_io_vec[bio->bi_vcnt++] = bv;
704 break;
705 }
706
707 if (bio_integrity(bio_src)) {
708 int ret;
709
710 ret = bio_integrity_clone(bio, bio_src, gfp_mask);
711 if (ret < 0) {
712 bio_put(bio);
713 return NULL;
714 }
715 }
716
717 bio_clone_blkcg_association(bio, bio_src);
718
719 return bio;
720}
721EXPORT_SYMBOL(bio_clone_bioset);
722
723/**
724 * bio_add_pc_page - attempt to add page to bio
725 * @q: the target queue
726 * @bio: destination bio
727 * @page: page to add
728 * @len: vec entry length
729 * @offset: vec entry offset
730 *
731 * Attempt to add a page to the bio_vec maplist. This can fail for a
732 * number of reasons, such as the bio being full or target block device
733 * limitations. The target block device must allow bio's up to PAGE_SIZE,
734 * so it is always possible to add a single page to an empty bio.
735 *
736 * This should only be used by REQ_PC bios.
737 */
738int bio_add_pc_page(struct request_queue *q, struct bio *bio, struct page
739 *page, unsigned int len, unsigned int offset)
740{
741 int retried_segments = 0;
742 struct bio_vec *bvec;
743
744 /*
745 * cloned bio must not modify vec list
746 */
747 if (unlikely(bio_flagged(bio, BIO_CLONED)))
748 return 0;
749
750 if (((bio->bi_iter.bi_size + len) >> 9) > queue_max_hw_sectors(q))
751 return 0;
752
753 /*
754 * For filesystems with a blocksize smaller than the pagesize
755 * we will often be called with the same page as last time and
756 * a consecutive offset. Optimize this special case.
757 */
758 if (bio->bi_vcnt > 0) {
759 struct bio_vec *prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
760
761 if (page == prev->bv_page &&
762 offset == prev->bv_offset + prev->bv_len) {
763 prev->bv_len += len;
764 bio->bi_iter.bi_size += len;
765 goto done;
766 }
767
768 /*
769 * If the queue doesn't support SG gaps and adding this
770 * offset would create a gap, disallow it.
771 */
772 if (bvec_gap_to_prev(q, prev, offset))
773 return 0;
774 }
775
776 if (bio->bi_vcnt >= bio->bi_max_vecs)
777 return 0;
778
779 /*
780 * setup the new entry, we might clear it again later if we
781 * cannot add the page
782 */
783 bvec = &bio->bi_io_vec[bio->bi_vcnt];
784 bvec->bv_page = page;
785 bvec->bv_len = len;
786 bvec->bv_offset = offset;
787 bio->bi_vcnt++;
788 bio->bi_phys_segments++;
789 bio->bi_iter.bi_size += len;
790
791 /*
792 * Perform a recount if the number of segments is greater
793 * than queue_max_segments(q).
794 */
795
796 while (bio->bi_phys_segments > queue_max_segments(q)) {
797
798 if (retried_segments)
799 goto failed;
800
801 retried_segments = 1;
802 blk_recount_segments(q, bio);
803 }
804
805 /* If we may be able to merge these biovecs, force a recount */
806 if (bio->bi_vcnt > 1 && (BIOVEC_PHYS_MERGEABLE(bvec-1, bvec)))
807 bio_clear_flag(bio, BIO_SEG_VALID);
808
809 done:
810 return len;
811
812 failed:
813 bvec->bv_page = NULL;
814 bvec->bv_len = 0;
815 bvec->bv_offset = 0;
816 bio->bi_vcnt--;
817 bio->bi_iter.bi_size -= len;
818 blk_recount_segments(q, bio);
819 return 0;
820}
821EXPORT_SYMBOL(bio_add_pc_page);
822
823/**
824 * bio_add_page - attempt to add page to bio
825 * @bio: destination bio
826 * @page: page to add
827 * @len: vec entry length
828 * @offset: vec entry offset
829 *
830 * Attempt to add a page to the bio_vec maplist. This will only fail
831 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
832 */
833int bio_add_page(struct bio *bio, struct page *page,
834 unsigned int len, unsigned int offset)
835{
836 struct bio_vec *bv;
837
838 /*
839 * cloned bio must not modify vec list
840 */
841 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
842 return 0;
843
844 /*
845 * For filesystems with a blocksize smaller than the pagesize
846 * we will often be called with the same page as last time and
847 * a consecutive offset. Optimize this special case.
848 */
849 if (bio->bi_vcnt > 0) {
850 bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
851
852 if (page == bv->bv_page &&
853 offset == bv->bv_offset + bv->bv_len) {
854 bv->bv_len += len;
855 goto done;
856 }
857 }
858
859 if (bio->bi_vcnt >= bio->bi_max_vecs)
860 return 0;
861
862 bv = &bio->bi_io_vec[bio->bi_vcnt];
863 bv->bv_page = page;
864 bv->bv_len = len;
865 bv->bv_offset = offset;
866
867 bio->bi_vcnt++;
868done:
869 bio->bi_iter.bi_size += len;
870 return len;
871}
872EXPORT_SYMBOL(bio_add_page);
873
874/**
875 * bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
876 * @bio: bio to add pages to
877 * @iter: iov iterator describing the region to be mapped
878 *
879 * Pins as many pages from *iter and appends them to @bio's bvec array. The
880 * pages will have to be released using put_page() when done.
881 */
882int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
883{
884 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
885 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
886 struct page **pages = (struct page **)bv;
887 size_t offset, diff;
888 ssize_t size;
889
890 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
891 if (unlikely(size <= 0))
892 return size ? size : -EFAULT;
893 nr_pages = (size + offset + PAGE_SIZE - 1) / PAGE_SIZE;
894
895 /*
896 * Deep magic below: We need to walk the pinned pages backwards
897 * because we are abusing the space allocated for the bio_vecs
898 * for the page array. Because the bio_vecs are larger than the
899 * page pointers by definition this will always work. But it also
900 * means we can't use bio_add_page, so any changes to it's semantics
901 * need to be reflected here as well.
902 */
903 bio->bi_iter.bi_size += size;
904 bio->bi_vcnt += nr_pages;
905
906 diff = (nr_pages * PAGE_SIZE - offset) - size;
907 while (nr_pages--) {
908 bv[nr_pages].bv_page = pages[nr_pages];
909 bv[nr_pages].bv_len = PAGE_SIZE;
910 bv[nr_pages].bv_offset = 0;
911 }
912
913 bv[0].bv_offset += offset;
914 bv[0].bv_len -= offset;
915 if (diff)
916 bv[bio->bi_vcnt - 1].bv_len -= diff;
917
918 iov_iter_advance(iter, size);
919 return 0;
920}
921EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
922
923static void submit_bio_wait_endio(struct bio *bio)
924{
925 complete(bio->bi_private);
926}
927
928/**
929 * submit_bio_wait - submit a bio, and wait until it completes
930 * @bio: The &struct bio which describes the I/O
931 *
932 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
933 * bio_endio() on failure.
934 *
935 * WARNING: Unlike to how submit_bio() is usually used, this function does not
936 * result in bio reference to be consumed. The caller must drop the reference
937 * on his own.
938 */
939int submit_bio_wait(struct bio *bio)
940{
941 DECLARE_COMPLETION_ONSTACK_MAP(done, bio->bi_disk->lockdep_map);
942
943 bio->bi_private = &done;
944 bio->bi_end_io = submit_bio_wait_endio;
945 bio->bi_opf |= REQ_SYNC;
946 submit_bio(bio);
947 wait_for_completion_io(&done);
948
949 return blk_status_to_errno(bio->bi_status);
950}
951EXPORT_SYMBOL(submit_bio_wait);
952
953/**
954 * bio_advance - increment/complete a bio by some number of bytes
955 * @bio: bio to advance
956 * @bytes: number of bytes to complete
957 *
958 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
959 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
960 * be updated on the last bvec as well.
961 *
962 * @bio will then represent the remaining, uncompleted portion of the io.
963 */
964void bio_advance(struct bio *bio, unsigned bytes)
965{
966 if (bio_integrity(bio))
967 bio_integrity_advance(bio, bytes);
968
969 bio_advance_iter(bio, &bio->bi_iter, bytes);
970}
971EXPORT_SYMBOL(bio_advance);
972
973/**
974 * bio_copy_data - copy contents of data buffers from one chain of bios to
975 * another
976 * @src: source bio list
977 * @dst: destination bio list
978 *
979 * If @src and @dst are single bios, bi_next must be NULL - otherwise, treats
980 * @src and @dst as linked lists of bios.
981 *
982 * Stops when it reaches the end of either @src or @dst - that is, copies
983 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
984 */
985void bio_copy_data(struct bio *dst, struct bio *src)
986{
987 struct bvec_iter src_iter, dst_iter;
988 struct bio_vec src_bv, dst_bv;
989 void *src_p, *dst_p;
990 unsigned bytes;
991
992 src_iter = src->bi_iter;
993 dst_iter = dst->bi_iter;
994
995 while (1) {
996 if (!src_iter.bi_size) {
997 src = src->bi_next;
998 if (!src)
999 break;
1000
1001 src_iter = src->bi_iter;
1002 }
1003
1004 if (!dst_iter.bi_size) {
1005 dst = dst->bi_next;
1006 if (!dst)
1007 break;
1008
1009 dst_iter = dst->bi_iter;
1010 }
1011
1012 src_bv = bio_iter_iovec(src, src_iter);
1013 dst_bv = bio_iter_iovec(dst, dst_iter);
1014
1015 bytes = min(src_bv.bv_len, dst_bv.bv_len);
1016
1017 src_p = kmap_atomic(src_bv.bv_page);
1018 dst_p = kmap_atomic(dst_bv.bv_page);
1019
1020 memcpy(dst_p + dst_bv.bv_offset,
1021 src_p + src_bv.bv_offset,
1022 bytes);
1023
1024 kunmap_atomic(dst_p);
1025 kunmap_atomic(src_p);
1026
1027 bio_advance_iter(src, &src_iter, bytes);
1028 bio_advance_iter(dst, &dst_iter, bytes);
1029 }
1030}
1031EXPORT_SYMBOL(bio_copy_data);
1032
1033struct bio_map_data {
1034 int is_our_pages;
1035 struct iov_iter iter;
1036 struct iovec iov[];
1037};
1038
1039static struct bio_map_data *bio_alloc_map_data(struct iov_iter *data,
1040 gfp_t gfp_mask)
1041{
1042 struct bio_map_data *bmd;
1043 if (data->nr_segs > UIO_MAXIOV)
1044 return NULL;
1045
1046 bmd = kmalloc(sizeof(struct bio_map_data) +
1047 sizeof(struct iovec) * data->nr_segs, gfp_mask);
1048 if (!bmd)
1049 return NULL;
1050 memcpy(bmd->iov, data->iov, sizeof(struct iovec) * data->nr_segs);
1051 bmd->iter = *data;
1052 bmd->iter.iov = bmd->iov;
1053 return bmd;
1054}
1055
1056/**
1057 * bio_copy_from_iter - copy all pages from iov_iter to bio
1058 * @bio: The &struct bio which describes the I/O as destination
1059 * @iter: iov_iter as source
1060 *
1061 * Copy all pages from iov_iter to bio.
1062 * Returns 0 on success, or error on failure.
1063 */
1064static int bio_copy_from_iter(struct bio *bio, struct iov_iter *iter)
1065{
1066 int i;
1067 struct bio_vec *bvec;
1068
1069 bio_for_each_segment_all(bvec, bio, i) {
1070 ssize_t ret;
1071
1072 ret = copy_page_from_iter(bvec->bv_page,
1073 bvec->bv_offset,
1074 bvec->bv_len,
1075 iter);
1076
1077 if (!iov_iter_count(iter))
1078 break;
1079
1080 if (ret < bvec->bv_len)
1081 return -EFAULT;
1082 }
1083
1084 return 0;
1085}
1086
1087/**
1088 * bio_copy_to_iter - copy all pages from bio to iov_iter
1089 * @bio: The &struct bio which describes the I/O as source
1090 * @iter: iov_iter as destination
1091 *
1092 * Copy all pages from bio to iov_iter.
1093 * Returns 0 on success, or error on failure.
1094 */
1095static int bio_copy_to_iter(struct bio *bio, struct iov_iter iter)
1096{
1097 int i;
1098 struct bio_vec *bvec;
1099
1100 bio_for_each_segment_all(bvec, bio, i) {
1101 ssize_t ret;
1102
1103 ret = copy_page_to_iter(bvec->bv_page,
1104 bvec->bv_offset,
1105 bvec->bv_len,
1106 &iter);
1107
1108 if (!iov_iter_count(&iter))
1109 break;
1110
1111 if (ret < bvec->bv_len)
1112 return -EFAULT;
1113 }
1114
1115 return 0;
1116}
1117
1118void bio_free_pages(struct bio *bio)
1119{
1120 struct bio_vec *bvec;
1121 int i;
1122
1123 bio_for_each_segment_all(bvec, bio, i)
1124 __free_page(bvec->bv_page);
1125}
1126EXPORT_SYMBOL(bio_free_pages);
1127
1128/**
1129 * bio_uncopy_user - finish previously mapped bio
1130 * @bio: bio being terminated
1131 *
1132 * Free pages allocated from bio_copy_user_iov() and write back data
1133 * to user space in case of a read.
1134 */
1135int bio_uncopy_user(struct bio *bio)
1136{
1137 struct bio_map_data *bmd = bio->bi_private;
1138 int ret = 0;
1139
1140 if (!bio_flagged(bio, BIO_NULL_MAPPED)) {
1141 /*
1142 * if we're in a workqueue, the request is orphaned, so
1143 * don't copy into a random user address space, just free
1144 * and return -EINTR so user space doesn't expect any data.
1145 */
1146 if (!current->mm)
1147 ret = -EINTR;
1148 else if (bio_data_dir(bio) == READ)
1149 ret = bio_copy_to_iter(bio, bmd->iter);
1150 if (bmd->is_our_pages)
1151 bio_free_pages(bio);
1152 }
1153 kfree(bmd);
1154 bio_put(bio);
1155 return ret;
1156}
1157
1158/**
1159 * bio_copy_user_iov - copy user data to bio
1160 * @q: destination block queue
1161 * @map_data: pointer to the rq_map_data holding pages (if necessary)
1162 * @iter: iovec iterator
1163 * @gfp_mask: memory allocation flags
1164 *
1165 * Prepares and returns a bio for indirect user io, bouncing data
1166 * to/from kernel pages as necessary. Must be paired with
1167 * call bio_uncopy_user() on io completion.
1168 */
1169struct bio *bio_copy_user_iov(struct request_queue *q,
1170 struct rq_map_data *map_data,
1171 struct iov_iter *iter,
1172 gfp_t gfp_mask)
1173{
1174 struct bio_map_data *bmd;
1175 struct page *page;
1176 struct bio *bio;
1177 int i = 0, ret;
1178 int nr_pages;
1179 unsigned int len = iter->count;
1180 unsigned int offset = map_data ? offset_in_page(map_data->offset) : 0;
1181
1182 bmd = bio_alloc_map_data(iter, gfp_mask);
1183 if (!bmd)
1184 return ERR_PTR(-ENOMEM);
1185
1186 /*
1187 * We need to do a deep copy of the iov_iter including the iovecs.
1188 * The caller provided iov might point to an on-stack or otherwise
1189 * shortlived one.
1190 */
1191 bmd->is_our_pages = map_data ? 0 : 1;
1192
1193 nr_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE);
1194 if (nr_pages > BIO_MAX_PAGES)
1195 nr_pages = BIO_MAX_PAGES;
1196
1197 ret = -ENOMEM;
1198 bio = bio_kmalloc(gfp_mask, nr_pages);
1199 if (!bio)
1200 goto out_bmd;
1201
1202 ret = 0;
1203
1204 if (map_data) {
1205 nr_pages = 1 << map_data->page_order;
1206 i = map_data->offset / PAGE_SIZE;
1207 }
1208 while (len) {
1209 unsigned int bytes = PAGE_SIZE;
1210
1211 bytes -= offset;
1212
1213 if (bytes > len)
1214 bytes = len;
1215
1216 if (map_data) {
1217 if (i == map_data->nr_entries * nr_pages) {
1218 ret = -ENOMEM;
1219 break;
1220 }
1221
1222 page = map_data->pages[i / nr_pages];
1223 page += (i % nr_pages);
1224
1225 i++;
1226 } else {
1227 page = alloc_page(q->bounce_gfp | gfp_mask);
1228 if (!page) {
1229 ret = -ENOMEM;
1230 break;
1231 }
1232 }
1233
1234 if (bio_add_pc_page(q, bio, page, bytes, offset) < bytes)
1235 break;
1236
1237 len -= bytes;
1238 offset = 0;
1239 }
1240
1241 if (ret)
1242 goto cleanup;
1243
1244 if (map_data)
1245 map_data->offset += bio->bi_iter.bi_size;
1246
1247 /*
1248 * success
1249 */
1250 if (((iter->type & WRITE) && (!map_data || !map_data->null_mapped)) ||
1251 (map_data && map_data->from_user)) {
1252 ret = bio_copy_from_iter(bio, iter);
1253 if (ret)
1254 goto cleanup;
1255 } else {
1256 iov_iter_advance(iter, bio->bi_iter.bi_size);
1257 }
1258
1259 bio->bi_private = bmd;
1260 if (map_data && map_data->null_mapped)
1261 bio_set_flag(bio, BIO_NULL_MAPPED);
1262 return bio;
1263cleanup:
1264 if (!map_data)
1265 bio_free_pages(bio);
1266 bio_put(bio);
1267out_bmd:
1268 kfree(bmd);
1269 return ERR_PTR(ret);
1270}
1271
1272/**
1273 * bio_map_user_iov - map user iovec into bio
1274 * @q: the struct request_queue for the bio
1275 * @iter: iovec iterator
1276 * @gfp_mask: memory allocation flags
1277 *
1278 * Map the user space address into a bio suitable for io to a block
1279 * device. Returns an error pointer in case of error.
1280 */
1281struct bio *bio_map_user_iov(struct request_queue *q,
1282 struct iov_iter *iter,
1283 gfp_t gfp_mask)
1284{
1285 int j;
1286 struct bio *bio;
1287 int ret;
1288 struct bio_vec *bvec;
1289
1290 if (!iov_iter_count(iter))
1291 return ERR_PTR(-EINVAL);
1292
1293 bio = bio_kmalloc(gfp_mask, iov_iter_npages(iter, BIO_MAX_PAGES));
1294 if (!bio)
1295 return ERR_PTR(-ENOMEM);
1296
1297 while (iov_iter_count(iter)) {
1298 struct page **pages;
1299 ssize_t bytes;
1300 size_t offs, added = 0;
1301 int npages;
1302
1303 bytes = iov_iter_get_pages_alloc(iter, &pages, LONG_MAX, &offs);
1304 if (unlikely(bytes <= 0)) {
1305 ret = bytes ? bytes : -EFAULT;
1306 goto out_unmap;
1307 }
1308
1309 npages = DIV_ROUND_UP(offs + bytes, PAGE_SIZE);
1310
1311 if (unlikely(offs & queue_dma_alignment(q))) {
1312 ret = -EINVAL;
1313 j = 0;
1314 } else {
1315 for (j = 0; j < npages; j++) {
1316 struct page *page = pages[j];
1317 unsigned int n = PAGE_SIZE - offs;
1318 unsigned short prev_bi_vcnt = bio->bi_vcnt;
1319
1320 if (n > bytes)
1321 n = bytes;
1322
1323 if (!bio_add_pc_page(q, bio, page, n, offs))
1324 break;
1325
1326 /*
1327 * check if vector was merged with previous
1328 * drop page reference if needed
1329 */
1330 if (bio->bi_vcnt == prev_bi_vcnt)
1331 put_page(page);
1332
1333 added += n;
1334 bytes -= n;
1335 offs = 0;
1336 }
1337 iov_iter_advance(iter, added);
1338 }
1339 /*
1340 * release the pages we didn't map into the bio, if any
1341 */
1342 while (j < npages)
1343 put_page(pages[j++]);
1344 kvfree(pages);
1345 /* couldn't stuff something into bio? */
1346 if (bytes)
1347 break;
1348 }
1349
1350 bio_set_flag(bio, BIO_USER_MAPPED);
1351
1352 /*
1353 * subtle -- if bio_map_user_iov() ended up bouncing a bio,
1354 * it would normally disappear when its bi_end_io is run.
1355 * however, we need it for the unmap, so grab an extra
1356 * reference to it
1357 */
1358 bio_get(bio);
1359 return bio;
1360
1361 out_unmap:
1362 bio_for_each_segment_all(bvec, bio, j) {
1363 put_page(bvec->bv_page);
1364 }
1365 bio_put(bio);
1366 return ERR_PTR(ret);
1367}
1368
1369static void __bio_unmap_user(struct bio *bio)
1370{
1371 struct bio_vec *bvec;
1372 int i;
1373
1374 /*
1375 * make sure we dirty pages we wrote to
1376 */
1377 bio_for_each_segment_all(bvec, bio, i) {
1378 if (bio_data_dir(bio) == READ)
1379 set_page_dirty_lock(bvec->bv_page);
1380
1381 put_page(bvec->bv_page);
1382 }
1383
1384 bio_put(bio);
1385}
1386
1387/**
1388 * bio_unmap_user - unmap a bio
1389 * @bio: the bio being unmapped
1390 *
1391 * Unmap a bio previously mapped by bio_map_user_iov(). Must be called from
1392 * process context.
1393 *
1394 * bio_unmap_user() may sleep.
1395 */
1396void bio_unmap_user(struct bio *bio)
1397{
1398 __bio_unmap_user(bio);
1399 bio_put(bio);
1400}
1401
1402static void bio_map_kern_endio(struct bio *bio)
1403{
1404 bio_put(bio);
1405}
1406
1407/**
1408 * bio_map_kern - map kernel address into bio
1409 * @q: the struct request_queue for the bio
1410 * @data: pointer to buffer to map
1411 * @len: length in bytes
1412 * @gfp_mask: allocation flags for bio allocation
1413 *
1414 * Map the kernel address into a bio suitable for io to a block
1415 * device. Returns an error pointer in case of error.
1416 */
1417struct bio *bio_map_kern(struct request_queue *q, void *data, unsigned int len,
1418 gfp_t gfp_mask)
1419{
1420 unsigned long kaddr = (unsigned long)data;
1421 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1422 unsigned long start = kaddr >> PAGE_SHIFT;
1423 const int nr_pages = end - start;
1424 int offset, i;
1425 struct bio *bio;
1426
1427 bio = bio_kmalloc(gfp_mask, nr_pages);
1428 if (!bio)
1429 return ERR_PTR(-ENOMEM);
1430
1431 offset = offset_in_page(kaddr);
1432 for (i = 0; i < nr_pages; i++) {
1433 unsigned int bytes = PAGE_SIZE - offset;
1434
1435 if (len <= 0)
1436 break;
1437
1438 if (bytes > len)
1439 bytes = len;
1440
1441 if (bio_add_pc_page(q, bio, virt_to_page(data), bytes,
1442 offset) < bytes) {
1443 /* we don't support partial mappings */
1444 bio_put(bio);
1445 return ERR_PTR(-EINVAL);
1446 }
1447
1448 data += bytes;
1449 len -= bytes;
1450 offset = 0;
1451 }
1452
1453 bio->bi_end_io = bio_map_kern_endio;
1454 return bio;
1455}
1456EXPORT_SYMBOL(bio_map_kern);
1457
1458static void bio_copy_kern_endio(struct bio *bio)
1459{
1460 bio_free_pages(bio);
1461 bio_put(bio);
1462}
1463
1464static void bio_copy_kern_endio_read(struct bio *bio)
1465{
1466 char *p = bio->bi_private;
1467 struct bio_vec *bvec;
1468 int i;
1469
1470 bio_for_each_segment_all(bvec, bio, i) {
1471 memcpy(p, page_address(bvec->bv_page), bvec->bv_len);
1472 p += bvec->bv_len;
1473 }
1474
1475 bio_copy_kern_endio(bio);
1476}
1477
1478/**
1479 * bio_copy_kern - copy kernel address into bio
1480 * @q: the struct request_queue for the bio
1481 * @data: pointer to buffer to copy
1482 * @len: length in bytes
1483 * @gfp_mask: allocation flags for bio and page allocation
1484 * @reading: data direction is READ
1485 *
1486 * copy the kernel address into a bio suitable for io to a block
1487 * device. Returns an error pointer in case of error.
1488 */
1489struct bio *bio_copy_kern(struct request_queue *q, void *data, unsigned int len,
1490 gfp_t gfp_mask, int reading)
1491{
1492 unsigned long kaddr = (unsigned long)data;
1493 unsigned long end = (kaddr + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
1494 unsigned long start = kaddr >> PAGE_SHIFT;
1495 struct bio *bio;
1496 void *p = data;
1497 int nr_pages = 0;
1498
1499 /*
1500 * Overflow, abort
1501 */
1502 if (end < start)
1503 return ERR_PTR(-EINVAL);
1504
1505 nr_pages = end - start;
1506 bio = bio_kmalloc(gfp_mask, nr_pages);
1507 if (!bio)
1508 return ERR_PTR(-ENOMEM);
1509
1510 while (len) {
1511 struct page *page;
1512 unsigned int bytes = PAGE_SIZE;
1513
1514 if (bytes > len)
1515 bytes = len;
1516
1517 page = alloc_page(q->bounce_gfp | gfp_mask);
1518 if (!page)
1519 goto cleanup;
1520
1521 if (!reading)
1522 memcpy(page_address(page), p, bytes);
1523
1524 if (bio_add_pc_page(q, bio, page, bytes, 0) < bytes)
1525 break;
1526
1527 len -= bytes;
1528 p += bytes;
1529 }
1530
1531 if (reading) {
1532 bio->bi_end_io = bio_copy_kern_endio_read;
1533 bio->bi_private = data;
1534 } else {
1535 bio->bi_end_io = bio_copy_kern_endio;
1536 }
1537
1538 return bio;
1539
1540cleanup:
1541 bio_free_pages(bio);
1542 bio_put(bio);
1543 return ERR_PTR(-ENOMEM);
1544}
1545
1546/*
1547 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1548 * for performing direct-IO in BIOs.
1549 *
1550 * The problem is that we cannot run set_page_dirty() from interrupt context
1551 * because the required locks are not interrupt-safe. So what we can do is to
1552 * mark the pages dirty _before_ performing IO. And in interrupt context,
1553 * check that the pages are still dirty. If so, fine. If not, redirty them
1554 * in process context.
1555 *
1556 * We special-case compound pages here: normally this means reads into hugetlb
1557 * pages. The logic in here doesn't really work right for compound pages
1558 * because the VM does not uniformly chase down the head page in all cases.
1559 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1560 * handle them at all. So we skip compound pages here at an early stage.
1561 *
1562 * Note that this code is very hard to test under normal circumstances because
1563 * direct-io pins the pages with get_user_pages(). This makes
1564 * is_page_cache_freeable return false, and the VM will not clean the pages.
1565 * But other code (eg, flusher threads) could clean the pages if they are mapped
1566 * pagecache.
1567 *
1568 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1569 * deferred bio dirtying paths.
1570 */
1571
1572/*
1573 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1574 */
1575void bio_set_pages_dirty(struct bio *bio)
1576{
1577 struct bio_vec *bvec;
1578 int i;
1579
1580 bio_for_each_segment_all(bvec, bio, i) {
1581 struct page *page = bvec->bv_page;
1582
1583 if (page && !PageCompound(page))
1584 set_page_dirty_lock(page);
1585 }
1586}
1587
1588static void bio_release_pages(struct bio *bio)
1589{
1590 struct bio_vec *bvec;
1591 int i;
1592
1593 bio_for_each_segment_all(bvec, bio, i) {
1594 struct page *page = bvec->bv_page;
1595
1596 if (page)
1597 put_page(page);
1598 }
1599}
1600
1601/*
1602 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1603 * If they are, then fine. If, however, some pages are clean then they must
1604 * have been written out during the direct-IO read. So we take another ref on
1605 * the BIO and the offending pages and re-dirty the pages in process context.
1606 *
1607 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1608 * here on. It will run one put_page() against each page and will run one
1609 * bio_put() against the BIO.
1610 */
1611
1612static void bio_dirty_fn(struct work_struct *work);
1613
1614static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1615static DEFINE_SPINLOCK(bio_dirty_lock);
1616static struct bio *bio_dirty_list;
1617
1618/*
1619 * This runs in process context
1620 */
1621static void bio_dirty_fn(struct work_struct *work)
1622{
1623 unsigned long flags;
1624 struct bio *bio;
1625
1626 spin_lock_irqsave(&bio_dirty_lock, flags);
1627 bio = bio_dirty_list;
1628 bio_dirty_list = NULL;
1629 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1630
1631 while (bio) {
1632 struct bio *next = bio->bi_private;
1633
1634 bio_set_pages_dirty(bio);
1635 bio_release_pages(bio);
1636 bio_put(bio);
1637 bio = next;
1638 }
1639}
1640
1641void bio_check_pages_dirty(struct bio *bio)
1642{
1643 struct bio_vec *bvec;
1644 int nr_clean_pages = 0;
1645 int i;
1646
1647 bio_for_each_segment_all(bvec, bio, i) {
1648 struct page *page = bvec->bv_page;
1649
1650 if (PageDirty(page) || PageCompound(page)) {
1651 put_page(page);
1652 bvec->bv_page = NULL;
1653 } else {
1654 nr_clean_pages++;
1655 }
1656 }
1657
1658 if (nr_clean_pages) {
1659 unsigned long flags;
1660
1661 spin_lock_irqsave(&bio_dirty_lock, flags);
1662 bio->bi_private = bio_dirty_list;
1663 bio_dirty_list = bio;
1664 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1665 schedule_work(&bio_dirty_work);
1666 } else {
1667 bio_put(bio);
1668 }
1669}
1670
1671void generic_start_io_acct(struct request_queue *q, int rw,
1672 unsigned long sectors, struct hd_struct *part)
1673{
1674 int cpu = part_stat_lock();
1675
1676 part_round_stats(q, cpu, part);
1677 part_stat_inc(cpu, part, ios[rw]);
1678 part_stat_add(cpu, part, sectors[rw], sectors);
1679 part_inc_in_flight(q, part, rw);
1680
1681 part_stat_unlock();
1682}
1683EXPORT_SYMBOL(generic_start_io_acct);
1684
1685void generic_end_io_acct(struct request_queue *q, int rw,
1686 struct hd_struct *part, unsigned long start_time)
1687{
1688 unsigned long duration = jiffies - start_time;
1689 int cpu = part_stat_lock();
1690
1691 part_stat_add(cpu, part, ticks[rw], duration);
1692 part_round_stats(q, cpu, part);
1693 part_dec_in_flight(q, part, rw);
1694
1695 part_stat_unlock();
1696}
1697EXPORT_SYMBOL(generic_end_io_acct);
1698
1699#if ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE
1700void bio_flush_dcache_pages(struct bio *bi)
1701{
1702 struct bio_vec bvec;
1703 struct bvec_iter iter;
1704
1705 bio_for_each_segment(bvec, bi, iter)
1706 flush_dcache_page(bvec.bv_page);
1707}
1708EXPORT_SYMBOL(bio_flush_dcache_pages);
1709#endif
1710
1711static inline bool bio_remaining_done(struct bio *bio)
1712{
1713 /*
1714 * If we're not chaining, then ->__bi_remaining is always 1 and
1715 * we always end io on the first invocation.
1716 */
1717 if (!bio_flagged(bio, BIO_CHAIN))
1718 return true;
1719
1720 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1721
1722 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1723 bio_clear_flag(bio, BIO_CHAIN);
1724 return true;
1725 }
1726
1727 return false;
1728}
1729
1730/**
1731 * bio_endio - end I/O on a bio
1732 * @bio: bio
1733 *
1734 * Description:
1735 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1736 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1737 * bio unless they own it and thus know that it has an end_io function.
1738 *
1739 * bio_endio() can be called several times on a bio that has been chained
1740 * using bio_chain(). The ->bi_end_io() function will only be called the
1741 * last time. At this point the BLK_TA_COMPLETE tracing event will be
1742 * generated if BIO_TRACE_COMPLETION is set.
1743 **/
1744void bio_endio(struct bio *bio)
1745{
1746again:
1747 if (!bio_remaining_done(bio))
1748 return;
1749 if (!bio_integrity_endio(bio))
1750 return;
1751
1752 /*
1753 * Need to have a real endio function for chained bios, otherwise
1754 * various corner cases will break (like stacking block devices that
1755 * save/restore bi_end_io) - however, we want to avoid unbounded
1756 * recursion and blowing the stack. Tail call optimization would
1757 * handle this, but compiling with frame pointers also disables
1758 * gcc's sibling call optimization.
1759 */
1760 if (bio->bi_end_io == bio_chain_endio) {
1761 bio = __bio_chain_endio(bio);
1762 goto again;
1763 }
1764
1765 if (bio->bi_disk && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1766 trace_block_bio_complete(bio->bi_disk->queue, bio,
1767 blk_status_to_errno(bio->bi_status));
1768 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1769 }
1770
1771 blk_throtl_bio_endio(bio);
1772 /* release cgroup info */
1773 bio_uninit(bio);
1774 if (bio->bi_end_io)
1775 bio->bi_end_io(bio);
1776}
1777EXPORT_SYMBOL(bio_endio);
1778
1779/**
1780 * bio_split - split a bio
1781 * @bio: bio to split
1782 * @sectors: number of sectors to split from the front of @bio
1783 * @gfp: gfp mask
1784 * @bs: bio set to allocate from
1785 *
1786 * Allocates and returns a new bio which represents @sectors from the start of
1787 * @bio, and updates @bio to represent the remaining sectors.
1788 *
1789 * Unless this is a discard request the newly allocated bio will point
1790 * to @bio's bi_io_vec; it is the caller's responsibility to ensure that
1791 * @bio is not freed before the split.
1792 */
1793struct bio *bio_split(struct bio *bio, int sectors,
1794 gfp_t gfp, struct bio_set *bs)
1795{
1796 struct bio *split;
1797
1798 BUG_ON(sectors <= 0);
1799 BUG_ON(sectors >= bio_sectors(bio));
1800
1801 split = bio_clone_fast(bio, gfp, bs);
1802 if (!split)
1803 return NULL;
1804
1805 split->bi_iter.bi_size = sectors << 9;
1806
1807 if (bio_integrity(split))
1808 bio_integrity_trim(split);
1809
1810 bio_advance(bio, split->bi_iter.bi_size);
1811
1812 if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1813 bio_set_flag(split, BIO_TRACE_COMPLETION);
1814
1815 return split;
1816}
1817EXPORT_SYMBOL(bio_split);
1818
1819/**
1820 * bio_trim - trim a bio
1821 * @bio: bio to trim
1822 * @offset: number of sectors to trim from the front of @bio
1823 * @size: size we want to trim @bio to, in sectors
1824 */
1825void bio_trim(struct bio *bio, int offset, int size)
1826{
1827 /* 'bio' is a cloned bio which we need to trim to match
1828 * the given offset and size.
1829 */
1830
1831 size <<= 9;
1832 if (offset == 0 && size == bio->bi_iter.bi_size)
1833 return;
1834
1835 bio_clear_flag(bio, BIO_SEG_VALID);
1836
1837 bio_advance(bio, offset << 9);
1838
1839 bio->bi_iter.bi_size = size;
1840
1841 if (bio_integrity(bio))
1842 bio_integrity_trim(bio);
1843
1844}
1845EXPORT_SYMBOL_GPL(bio_trim);
1846
1847/*
1848 * create memory pools for biovec's in a bio_set.
1849 * use the global biovec slabs created for general use.
1850 */
1851mempool_t *biovec_create_pool(int pool_entries)
1852{
1853 struct biovec_slab *bp = bvec_slabs + BVEC_POOL_MAX;
1854
1855 return mempool_create_slab_pool(pool_entries, bp->slab);
1856}
1857
1858void bioset_free(struct bio_set *bs)
1859{
1860 if (bs->rescue_workqueue)
1861 destroy_workqueue(bs->rescue_workqueue);
1862
1863 mempool_destroy(bs->bio_pool);
1864 mempool_destroy(bs->bvec_pool);
1865
1866 bioset_integrity_free(bs);
1867 bio_put_slab(bs);
1868
1869 kfree(bs);
1870}
1871EXPORT_SYMBOL(bioset_free);
1872
1873/**
1874 * bioset_create - Create a bio_set
1875 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1876 * @front_pad: Number of bytes to allocate in front of the returned bio
1877 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1878 * and %BIOSET_NEED_RESCUER
1879 *
1880 * Description:
1881 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1882 * to ask for a number of bytes to be allocated in front of the bio.
1883 * Front pad allocation is useful for embedding the bio inside
1884 * another structure, to avoid allocating extra data to go with the bio.
1885 * Note that the bio must be embedded at the END of that structure always,
1886 * or things will break badly.
1887 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1888 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast().
1889 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
1890 * dispatch queued requests when the mempool runs out of space.
1891 *
1892 */
1893struct bio_set *bioset_create(unsigned int pool_size,
1894 unsigned int front_pad,
1895 int flags)
1896{
1897 unsigned int back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1898 struct bio_set *bs;
1899
1900 bs = kzalloc(sizeof(*bs), GFP_KERNEL);
1901 if (!bs)
1902 return NULL;
1903
1904 bs->front_pad = front_pad;
1905
1906 spin_lock_init(&bs->rescue_lock);
1907 bio_list_init(&bs->rescue_list);
1908 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1909
1910 bs->bio_slab = bio_find_or_create_slab(front_pad + back_pad);
1911 if (!bs->bio_slab) {
1912 kfree(bs);
1913 return NULL;
1914 }
1915
1916 bs->bio_pool = mempool_create_slab_pool(pool_size, bs->bio_slab);
1917 if (!bs->bio_pool)
1918 goto bad;
1919
1920 if (flags & BIOSET_NEED_BVECS) {
1921 bs->bvec_pool = biovec_create_pool(pool_size);
1922 if (!bs->bvec_pool)
1923 goto bad;
1924 }
1925
1926 if (!(flags & BIOSET_NEED_RESCUER))
1927 return bs;
1928
1929 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1930 if (!bs->rescue_workqueue)
1931 goto bad;
1932
1933 return bs;
1934bad:
1935 bioset_free(bs);
1936 return NULL;
1937}
1938EXPORT_SYMBOL(bioset_create);
1939
1940#ifdef CONFIG_BLK_CGROUP
1941
1942/**
1943 * bio_associate_blkcg - associate a bio with the specified blkcg
1944 * @bio: target bio
1945 * @blkcg_css: css of the blkcg to associate
1946 *
1947 * Associate @bio with the blkcg specified by @blkcg_css. Block layer will
1948 * treat @bio as if it were issued by a task which belongs to the blkcg.
1949 *
1950 * This function takes an extra reference of @blkcg_css which will be put
1951 * when @bio is released. The caller must own @bio and is responsible for
1952 * synchronizing calls to this function.
1953 */
1954int bio_associate_blkcg(struct bio *bio, struct cgroup_subsys_state *blkcg_css)
1955{
1956 if (unlikely(bio->bi_css))
1957 return -EBUSY;
1958 css_get(blkcg_css);
1959 bio->bi_css = blkcg_css;
1960 return 0;
1961}
1962EXPORT_SYMBOL_GPL(bio_associate_blkcg);
1963
1964/**
1965 * bio_disassociate_task - undo bio_associate_current()
1966 * @bio: target bio
1967 */
1968void bio_disassociate_task(struct bio *bio)
1969{
1970 if (bio->bi_ioc) {
1971 put_io_context(bio->bi_ioc);
1972 bio->bi_ioc = NULL;
1973 }
1974 if (bio->bi_css) {
1975 css_put(bio->bi_css);
1976 bio->bi_css = NULL;
1977 }
1978}
1979
1980/**
1981 * bio_clone_blkcg_association - clone blkcg association from src to dst bio
1982 * @dst: destination bio
1983 * @src: source bio
1984 */
1985void bio_clone_blkcg_association(struct bio *dst, struct bio *src)
1986{
1987 if (src->bi_css)
1988 WARN_ON(bio_associate_blkcg(dst, src->bi_css));
1989}
1990EXPORT_SYMBOL_GPL(bio_clone_blkcg_association);
1991#endif /* CONFIG_BLK_CGROUP */
1992
1993static void __init biovec_init_slabs(void)
1994{
1995 int i;
1996
1997 for (i = 0; i < BVEC_POOL_NR; i++) {
1998 int size;
1999 struct biovec_slab *bvs = bvec_slabs + i;
2000
2001 if (bvs->nr_vecs <= BIO_INLINE_VECS) {
2002 bvs->slab = NULL;
2003 continue;
2004 }
2005
2006 size = bvs->nr_vecs * sizeof(struct bio_vec);
2007 bvs->slab = kmem_cache_create(bvs->name, size, 0,
2008 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2009 }
2010}
2011
2012static int __init init_bio(void)
2013{
2014 bio_slab_max = 2;
2015 bio_slab_nr = 0;
2016 bio_slabs = kzalloc(bio_slab_max * sizeof(struct bio_slab), GFP_KERNEL);
2017 if (!bio_slabs)
2018 panic("bio: can't allocate bios\n");
2019
2020 bio_integrity_init();
2021 biovec_init_slabs();
2022
2023 fs_bio_set = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
2024 if (!fs_bio_set)
2025 panic("bio: can't allocate bios\n");
2026
2027 if (bioset_integrity_create(fs_bio_set, BIO_POOL_SIZE))
2028 panic("bio: can't create integrity pool\n");
2029
2030 return 0;
2031}
2032subsys_initcall(init_bio);