Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
4 */
5#include <linux/mm.h>
6#include <linux/swap.h>
7#include <linux/bio.h>
8#include <linux/blkdev.h>
9#include <linux/uio.h>
10#include <linux/iocontext.h>
11#include <linux/slab.h>
12#include <linux/init.h>
13#include <linux/kernel.h>
14#include <linux/export.h>
15#include <linux/mempool.h>
16#include <linux/workqueue.h>
17#include <linux/cgroup.h>
18#include <linux/highmem.h>
19#include <linux/sched/sysctl.h>
20#include <linux/blk-crypto.h>
21#include <linux/xarray.h>
22
23#include <trace/events/block.h>
24#include "blk.h"
25#include "blk-rq-qos.h"
26#include "blk-cgroup.h"
27
28#define ALLOC_CACHE_THRESHOLD 16
29#define ALLOC_CACHE_SLACK 64
30#define ALLOC_CACHE_MAX 256
31
32struct bio_alloc_cache {
33 struct bio *free_list;
34 struct bio *free_list_irq;
35 unsigned int nr;
36 unsigned int nr_irq;
37};
38
39static struct biovec_slab {
40 int nr_vecs;
41 char *name;
42 struct kmem_cache *slab;
43} bvec_slabs[] __read_mostly = {
44 { .nr_vecs = 16, .name = "biovec-16" },
45 { .nr_vecs = 64, .name = "biovec-64" },
46 { .nr_vecs = 128, .name = "biovec-128" },
47 { .nr_vecs = BIO_MAX_VECS, .name = "biovec-max" },
48};
49
50static struct biovec_slab *biovec_slab(unsigned short nr_vecs)
51{
52 switch (nr_vecs) {
53 /* smaller bios use inline vecs */
54 case 5 ... 16:
55 return &bvec_slabs[0];
56 case 17 ... 64:
57 return &bvec_slabs[1];
58 case 65 ... 128:
59 return &bvec_slabs[2];
60 case 129 ... BIO_MAX_VECS:
61 return &bvec_slabs[3];
62 default:
63 BUG();
64 return NULL;
65 }
66}
67
68/*
69 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
70 * IO code that does not need private memory pools.
71 */
72struct bio_set fs_bio_set;
73EXPORT_SYMBOL(fs_bio_set);
74
75/*
76 * Our slab pool management
77 */
78struct bio_slab {
79 struct kmem_cache *slab;
80 unsigned int slab_ref;
81 unsigned int slab_size;
82 char name[8];
83};
84static DEFINE_MUTEX(bio_slab_lock);
85static DEFINE_XARRAY(bio_slabs);
86
87static struct bio_slab *create_bio_slab(unsigned int size)
88{
89 struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL);
90
91 if (!bslab)
92 return NULL;
93
94 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size);
95 bslab->slab = kmem_cache_create(bslab->name, size,
96 ARCH_KMALLOC_MINALIGN,
97 SLAB_HWCACHE_ALIGN | SLAB_TYPESAFE_BY_RCU, NULL);
98 if (!bslab->slab)
99 goto fail_alloc_slab;
100
101 bslab->slab_ref = 1;
102 bslab->slab_size = size;
103
104 if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL)))
105 return bslab;
106
107 kmem_cache_destroy(bslab->slab);
108
109fail_alloc_slab:
110 kfree(bslab);
111 return NULL;
112}
113
114static inline unsigned int bs_bio_slab_size(struct bio_set *bs)
115{
116 return bs->front_pad + sizeof(struct bio) + bs->back_pad;
117}
118
119static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs)
120{
121 unsigned int size = bs_bio_slab_size(bs);
122 struct bio_slab *bslab;
123
124 mutex_lock(&bio_slab_lock);
125 bslab = xa_load(&bio_slabs, size);
126 if (bslab)
127 bslab->slab_ref++;
128 else
129 bslab = create_bio_slab(size);
130 mutex_unlock(&bio_slab_lock);
131
132 if (bslab)
133 return bslab->slab;
134 return NULL;
135}
136
137static void bio_put_slab(struct bio_set *bs)
138{
139 struct bio_slab *bslab = NULL;
140 unsigned int slab_size = bs_bio_slab_size(bs);
141
142 mutex_lock(&bio_slab_lock);
143
144 bslab = xa_load(&bio_slabs, slab_size);
145 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
146 goto out;
147
148 WARN_ON_ONCE(bslab->slab != bs->bio_slab);
149
150 WARN_ON(!bslab->slab_ref);
151
152 if (--bslab->slab_ref)
153 goto out;
154
155 xa_erase(&bio_slabs, slab_size);
156
157 kmem_cache_destroy(bslab->slab);
158 kfree(bslab);
159
160out:
161 mutex_unlock(&bio_slab_lock);
162}
163
164void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs)
165{
166 BUG_ON(nr_vecs > BIO_MAX_VECS);
167
168 if (nr_vecs == BIO_MAX_VECS)
169 mempool_free(bv, pool);
170 else if (nr_vecs > BIO_INLINE_VECS)
171 kmem_cache_free(biovec_slab(nr_vecs)->slab, bv);
172}
173
174/*
175 * Make the first allocation restricted and don't dump info on allocation
176 * failures, since we'll fall back to the mempool in case of failure.
177 */
178static inline gfp_t bvec_alloc_gfp(gfp_t gfp)
179{
180 return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) |
181 __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
182}
183
184struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
185 gfp_t gfp_mask)
186{
187 struct biovec_slab *bvs = biovec_slab(*nr_vecs);
188
189 if (WARN_ON_ONCE(!bvs))
190 return NULL;
191
192 /*
193 * Upgrade the nr_vecs request to take full advantage of the allocation.
194 * We also rely on this in the bvec_free path.
195 */
196 *nr_vecs = bvs->nr_vecs;
197
198 /*
199 * Try a slab allocation first for all smaller allocations. If that
200 * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool.
201 * The mempool is sized to handle up to BIO_MAX_VECS entries.
202 */
203 if (*nr_vecs < BIO_MAX_VECS) {
204 struct bio_vec *bvl;
205
206 bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask));
207 if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM))
208 return bvl;
209 *nr_vecs = BIO_MAX_VECS;
210 }
211
212 return mempool_alloc(pool, gfp_mask);
213}
214
215void bio_uninit(struct bio *bio)
216{
217#ifdef CONFIG_BLK_CGROUP
218 if (bio->bi_blkg) {
219 blkg_put(bio->bi_blkg);
220 bio->bi_blkg = NULL;
221 }
222#endif
223 if (bio_integrity(bio))
224 bio_integrity_free(bio);
225
226 bio_crypt_free_ctx(bio);
227}
228EXPORT_SYMBOL(bio_uninit);
229
230static void bio_free(struct bio *bio)
231{
232 struct bio_set *bs = bio->bi_pool;
233 void *p = bio;
234
235 WARN_ON_ONCE(!bs);
236
237 bio_uninit(bio);
238 bvec_free(&bs->bvec_pool, bio->bi_io_vec, bio->bi_max_vecs);
239 mempool_free(p - bs->front_pad, &bs->bio_pool);
240}
241
242/*
243 * Users of this function have their own bio allocation. Subsequently,
244 * they must remember to pair any call to bio_init() with bio_uninit()
245 * when IO has completed, or when the bio is released.
246 */
247void bio_init(struct bio *bio, struct block_device *bdev, struct bio_vec *table,
248 unsigned short max_vecs, blk_opf_t opf)
249{
250 bio->bi_next = NULL;
251 bio->bi_bdev = bdev;
252 bio->bi_opf = opf;
253 bio->bi_flags = 0;
254 bio->bi_ioprio = 0;
255 bio->bi_status = 0;
256 bio->bi_iter.bi_sector = 0;
257 bio->bi_iter.bi_size = 0;
258 bio->bi_iter.bi_idx = 0;
259 bio->bi_iter.bi_bvec_done = 0;
260 bio->bi_end_io = NULL;
261 bio->bi_private = NULL;
262#ifdef CONFIG_BLK_CGROUP
263 bio->bi_blkg = NULL;
264 bio->bi_issue.value = 0;
265 if (bdev)
266 bio_associate_blkg(bio);
267#ifdef CONFIG_BLK_CGROUP_IOCOST
268 bio->bi_iocost_cost = 0;
269#endif
270#endif
271#ifdef CONFIG_BLK_INLINE_ENCRYPTION
272 bio->bi_crypt_context = NULL;
273#endif
274#ifdef CONFIG_BLK_DEV_INTEGRITY
275 bio->bi_integrity = NULL;
276#endif
277 bio->bi_vcnt = 0;
278
279 atomic_set(&bio->__bi_remaining, 1);
280 atomic_set(&bio->__bi_cnt, 1);
281 bio->bi_cookie = BLK_QC_T_NONE;
282
283 bio->bi_max_vecs = max_vecs;
284 bio->bi_io_vec = table;
285 bio->bi_pool = NULL;
286}
287EXPORT_SYMBOL(bio_init);
288
289/**
290 * bio_reset - reinitialize a bio
291 * @bio: bio to reset
292 * @bdev: block device to use the bio for
293 * @opf: operation and flags for bio
294 *
295 * Description:
296 * After calling bio_reset(), @bio will be in the same state as a freshly
297 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
298 * preserved are the ones that are initialized by bio_alloc_bioset(). See
299 * comment in struct bio.
300 */
301void bio_reset(struct bio *bio, struct block_device *bdev, blk_opf_t opf)
302{
303 bio_uninit(bio);
304 memset(bio, 0, BIO_RESET_BYTES);
305 atomic_set(&bio->__bi_remaining, 1);
306 bio->bi_bdev = bdev;
307 if (bio->bi_bdev)
308 bio_associate_blkg(bio);
309 bio->bi_opf = opf;
310}
311EXPORT_SYMBOL(bio_reset);
312
313static struct bio *__bio_chain_endio(struct bio *bio)
314{
315 struct bio *parent = bio->bi_private;
316
317 if (bio->bi_status && !parent->bi_status)
318 parent->bi_status = bio->bi_status;
319 bio_put(bio);
320 return parent;
321}
322
323static void bio_chain_endio(struct bio *bio)
324{
325 bio_endio(__bio_chain_endio(bio));
326}
327
328/**
329 * bio_chain - chain bio completions
330 * @bio: the target bio
331 * @parent: the parent bio of @bio
332 *
333 * The caller won't have a bi_end_io called when @bio completes - instead,
334 * @parent's bi_end_io won't be called until both @parent and @bio have
335 * completed; the chained bio will also be freed when it completes.
336 *
337 * The caller must not set bi_private or bi_end_io in @bio.
338 */
339void bio_chain(struct bio *bio, struct bio *parent)
340{
341 BUG_ON(bio->bi_private || bio->bi_end_io);
342
343 bio->bi_private = parent;
344 bio->bi_end_io = bio_chain_endio;
345 bio_inc_remaining(parent);
346}
347EXPORT_SYMBOL(bio_chain);
348
349struct bio *blk_next_bio(struct bio *bio, struct block_device *bdev,
350 unsigned int nr_pages, blk_opf_t opf, gfp_t gfp)
351{
352 struct bio *new = bio_alloc(bdev, nr_pages, opf, gfp);
353
354 if (bio) {
355 bio_chain(bio, new);
356 submit_bio(bio);
357 }
358
359 return new;
360}
361EXPORT_SYMBOL_GPL(blk_next_bio);
362
363static void bio_alloc_rescue(struct work_struct *work)
364{
365 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
366 struct bio *bio;
367
368 while (1) {
369 spin_lock(&bs->rescue_lock);
370 bio = bio_list_pop(&bs->rescue_list);
371 spin_unlock(&bs->rescue_lock);
372
373 if (!bio)
374 break;
375
376 submit_bio_noacct(bio);
377 }
378}
379
380static void punt_bios_to_rescuer(struct bio_set *bs)
381{
382 struct bio_list punt, nopunt;
383 struct bio *bio;
384
385 if (WARN_ON_ONCE(!bs->rescue_workqueue))
386 return;
387 /*
388 * In order to guarantee forward progress we must punt only bios that
389 * were allocated from this bio_set; otherwise, if there was a bio on
390 * there for a stacking driver higher up in the stack, processing it
391 * could require allocating bios from this bio_set, and doing that from
392 * our own rescuer would be bad.
393 *
394 * Since bio lists are singly linked, pop them all instead of trying to
395 * remove from the middle of the list:
396 */
397
398 bio_list_init(&punt);
399 bio_list_init(&nopunt);
400
401 while ((bio = bio_list_pop(¤t->bio_list[0])))
402 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
403 current->bio_list[0] = nopunt;
404
405 bio_list_init(&nopunt);
406 while ((bio = bio_list_pop(¤t->bio_list[1])))
407 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
408 current->bio_list[1] = nopunt;
409
410 spin_lock(&bs->rescue_lock);
411 bio_list_merge(&bs->rescue_list, &punt);
412 spin_unlock(&bs->rescue_lock);
413
414 queue_work(bs->rescue_workqueue, &bs->rescue_work);
415}
416
417static void bio_alloc_irq_cache_splice(struct bio_alloc_cache *cache)
418{
419 unsigned long flags;
420
421 /* cache->free_list must be empty */
422 if (WARN_ON_ONCE(cache->free_list))
423 return;
424
425 local_irq_save(flags);
426 cache->free_list = cache->free_list_irq;
427 cache->free_list_irq = NULL;
428 cache->nr += cache->nr_irq;
429 cache->nr_irq = 0;
430 local_irq_restore(flags);
431}
432
433static struct bio *bio_alloc_percpu_cache(struct block_device *bdev,
434 unsigned short nr_vecs, blk_opf_t opf, gfp_t gfp,
435 struct bio_set *bs)
436{
437 struct bio_alloc_cache *cache;
438 struct bio *bio;
439
440 cache = per_cpu_ptr(bs->cache, get_cpu());
441 if (!cache->free_list) {
442 if (READ_ONCE(cache->nr_irq) >= ALLOC_CACHE_THRESHOLD)
443 bio_alloc_irq_cache_splice(cache);
444 if (!cache->free_list) {
445 put_cpu();
446 return NULL;
447 }
448 }
449 bio = cache->free_list;
450 cache->free_list = bio->bi_next;
451 cache->nr--;
452 put_cpu();
453
454 bio_init(bio, bdev, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs, opf);
455 bio->bi_pool = bs;
456 return bio;
457}
458
459/**
460 * bio_alloc_bioset - allocate a bio for I/O
461 * @bdev: block device to allocate the bio for (can be %NULL)
462 * @nr_vecs: number of bvecs to pre-allocate
463 * @opf: operation and flags for bio
464 * @gfp_mask: the GFP_* mask given to the slab allocator
465 * @bs: the bio_set to allocate from.
466 *
467 * Allocate a bio from the mempools in @bs.
468 *
469 * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to
470 * allocate a bio. This is due to the mempool guarantees. To make this work,
471 * callers must never allocate more than 1 bio at a time from the general pool.
472 * Callers that need to allocate more than 1 bio must always submit the
473 * previously allocated bio for IO before attempting to allocate a new one.
474 * Failure to do so can cause deadlocks under memory pressure.
475 *
476 * Note that when running under submit_bio_noacct() (i.e. any block driver),
477 * bios are not submitted until after you return - see the code in
478 * submit_bio_noacct() that converts recursion into iteration, to prevent
479 * stack overflows.
480 *
481 * This would normally mean allocating multiple bios under submit_bio_noacct()
482 * would be susceptible to deadlocks, but we have
483 * deadlock avoidance code that resubmits any blocked bios from a rescuer
484 * thread.
485 *
486 * However, we do not guarantee forward progress for allocations from other
487 * mempools. Doing multiple allocations from the same mempool under
488 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad
489 * for per bio allocations.
490 *
491 * Returns: Pointer to new bio on success, NULL on failure.
492 */
493struct bio *bio_alloc_bioset(struct block_device *bdev, unsigned short nr_vecs,
494 blk_opf_t opf, gfp_t gfp_mask,
495 struct bio_set *bs)
496{
497 gfp_t saved_gfp = gfp_mask;
498 struct bio *bio;
499 void *p;
500
501 /* should not use nobvec bioset for nr_vecs > 0 */
502 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_vecs > 0))
503 return NULL;
504
505 if (opf & REQ_ALLOC_CACHE) {
506 if (bs->cache && nr_vecs <= BIO_INLINE_VECS) {
507 bio = bio_alloc_percpu_cache(bdev, nr_vecs, opf,
508 gfp_mask, bs);
509 if (bio)
510 return bio;
511 /*
512 * No cached bio available, bio returned below marked with
513 * REQ_ALLOC_CACHE to particpate in per-cpu alloc cache.
514 */
515 } else {
516 opf &= ~REQ_ALLOC_CACHE;
517 }
518 }
519
520 /*
521 * submit_bio_noacct() converts recursion to iteration; this means if
522 * we're running beneath it, any bios we allocate and submit will not be
523 * submitted (and thus freed) until after we return.
524 *
525 * This exposes us to a potential deadlock if we allocate multiple bios
526 * from the same bio_set() while running underneath submit_bio_noacct().
527 * If we were to allocate multiple bios (say a stacking block driver
528 * that was splitting bios), we would deadlock if we exhausted the
529 * mempool's reserve.
530 *
531 * We solve this, and guarantee forward progress, with a rescuer
532 * workqueue per bio_set. If we go to allocate and there are bios on
533 * current->bio_list, we first try the allocation without
534 * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be
535 * blocking to the rescuer workqueue before we retry with the original
536 * gfp_flags.
537 */
538 if (current->bio_list &&
539 (!bio_list_empty(¤t->bio_list[0]) ||
540 !bio_list_empty(¤t->bio_list[1])) &&
541 bs->rescue_workqueue)
542 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
543
544 p = mempool_alloc(&bs->bio_pool, gfp_mask);
545 if (!p && gfp_mask != saved_gfp) {
546 punt_bios_to_rescuer(bs);
547 gfp_mask = saved_gfp;
548 p = mempool_alloc(&bs->bio_pool, gfp_mask);
549 }
550 if (unlikely(!p))
551 return NULL;
552 if (!mempool_is_saturated(&bs->bio_pool))
553 opf &= ~REQ_ALLOC_CACHE;
554
555 bio = p + bs->front_pad;
556 if (nr_vecs > BIO_INLINE_VECS) {
557 struct bio_vec *bvl = NULL;
558
559 bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);
560 if (!bvl && gfp_mask != saved_gfp) {
561 punt_bios_to_rescuer(bs);
562 gfp_mask = saved_gfp;
563 bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);
564 }
565 if (unlikely(!bvl))
566 goto err_free;
567
568 bio_init(bio, bdev, bvl, nr_vecs, opf);
569 } else if (nr_vecs) {
570 bio_init(bio, bdev, bio->bi_inline_vecs, BIO_INLINE_VECS, opf);
571 } else {
572 bio_init(bio, bdev, NULL, 0, opf);
573 }
574
575 bio->bi_pool = bs;
576 return bio;
577
578err_free:
579 mempool_free(p, &bs->bio_pool);
580 return NULL;
581}
582EXPORT_SYMBOL(bio_alloc_bioset);
583
584/**
585 * bio_kmalloc - kmalloc a bio
586 * @nr_vecs: number of bio_vecs to allocate
587 * @gfp_mask: the GFP_* mask given to the slab allocator
588 *
589 * Use kmalloc to allocate a bio (including bvecs). The bio must be initialized
590 * using bio_init() before use. To free a bio returned from this function use
591 * kfree() after calling bio_uninit(). A bio returned from this function can
592 * be reused by calling bio_uninit() before calling bio_init() again.
593 *
594 * Note that unlike bio_alloc() or bio_alloc_bioset() allocations from this
595 * function are not backed by a mempool can fail. Do not use this function
596 * for allocations in the file system I/O path.
597 *
598 * Returns: Pointer to new bio on success, NULL on failure.
599 */
600struct bio *bio_kmalloc(unsigned short nr_vecs, gfp_t gfp_mask)
601{
602 struct bio *bio;
603
604 if (nr_vecs > UIO_MAXIOV)
605 return NULL;
606 return kmalloc(struct_size(bio, bi_inline_vecs, nr_vecs), gfp_mask);
607}
608EXPORT_SYMBOL(bio_kmalloc);
609
610void zero_fill_bio(struct bio *bio)
611{
612 struct bio_vec bv;
613 struct bvec_iter iter;
614
615 bio_for_each_segment(bv, bio, iter)
616 memzero_bvec(&bv);
617}
618EXPORT_SYMBOL(zero_fill_bio);
619
620/**
621 * bio_truncate - truncate the bio to small size of @new_size
622 * @bio: the bio to be truncated
623 * @new_size: new size for truncating the bio
624 *
625 * Description:
626 * Truncate the bio to new size of @new_size. If bio_op(bio) is
627 * REQ_OP_READ, zero the truncated part. This function should only
628 * be used for handling corner cases, such as bio eod.
629 */
630static void bio_truncate(struct bio *bio, unsigned new_size)
631{
632 struct bio_vec bv;
633 struct bvec_iter iter;
634 unsigned int done = 0;
635 bool truncated = false;
636
637 if (new_size >= bio->bi_iter.bi_size)
638 return;
639
640 if (bio_op(bio) != REQ_OP_READ)
641 goto exit;
642
643 bio_for_each_segment(bv, bio, iter) {
644 if (done + bv.bv_len > new_size) {
645 unsigned offset;
646
647 if (!truncated)
648 offset = new_size - done;
649 else
650 offset = 0;
651 zero_user(bv.bv_page, bv.bv_offset + offset,
652 bv.bv_len - offset);
653 truncated = true;
654 }
655 done += bv.bv_len;
656 }
657
658 exit:
659 /*
660 * Don't touch bvec table here and make it really immutable, since
661 * fs bio user has to retrieve all pages via bio_for_each_segment_all
662 * in its .end_bio() callback.
663 *
664 * It is enough to truncate bio by updating .bi_size since we can make
665 * correct bvec with the updated .bi_size for drivers.
666 */
667 bio->bi_iter.bi_size = new_size;
668}
669
670/**
671 * guard_bio_eod - truncate a BIO to fit the block device
672 * @bio: bio to truncate
673 *
674 * This allows us to do IO even on the odd last sectors of a device, even if the
675 * block size is some multiple of the physical sector size.
676 *
677 * We'll just truncate the bio to the size of the device, and clear the end of
678 * the buffer head manually. Truly out-of-range accesses will turn into actual
679 * I/O errors, this only handles the "we need to be able to do I/O at the final
680 * sector" case.
681 */
682void guard_bio_eod(struct bio *bio)
683{
684 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
685
686 if (!maxsector)
687 return;
688
689 /*
690 * If the *whole* IO is past the end of the device,
691 * let it through, and the IO layer will turn it into
692 * an EIO.
693 */
694 if (unlikely(bio->bi_iter.bi_sector >= maxsector))
695 return;
696
697 maxsector -= bio->bi_iter.bi_sector;
698 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
699 return;
700
701 bio_truncate(bio, maxsector << 9);
702}
703
704static int __bio_alloc_cache_prune(struct bio_alloc_cache *cache,
705 unsigned int nr)
706{
707 unsigned int i = 0;
708 struct bio *bio;
709
710 while ((bio = cache->free_list) != NULL) {
711 cache->free_list = bio->bi_next;
712 cache->nr--;
713 bio_free(bio);
714 if (++i == nr)
715 break;
716 }
717 return i;
718}
719
720static void bio_alloc_cache_prune(struct bio_alloc_cache *cache,
721 unsigned int nr)
722{
723 nr -= __bio_alloc_cache_prune(cache, nr);
724 if (!READ_ONCE(cache->free_list)) {
725 bio_alloc_irq_cache_splice(cache);
726 __bio_alloc_cache_prune(cache, nr);
727 }
728}
729
730static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node)
731{
732 struct bio_set *bs;
733
734 bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead);
735 if (bs->cache) {
736 struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu);
737
738 bio_alloc_cache_prune(cache, -1U);
739 }
740 return 0;
741}
742
743static void bio_alloc_cache_destroy(struct bio_set *bs)
744{
745 int cpu;
746
747 if (!bs->cache)
748 return;
749
750 cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
751 for_each_possible_cpu(cpu) {
752 struct bio_alloc_cache *cache;
753
754 cache = per_cpu_ptr(bs->cache, cpu);
755 bio_alloc_cache_prune(cache, -1U);
756 }
757 free_percpu(bs->cache);
758 bs->cache = NULL;
759}
760
761static inline void bio_put_percpu_cache(struct bio *bio)
762{
763 struct bio_alloc_cache *cache;
764
765 cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu());
766 if (READ_ONCE(cache->nr_irq) + cache->nr > ALLOC_CACHE_MAX) {
767 put_cpu();
768 bio_free(bio);
769 return;
770 }
771
772 bio_uninit(bio);
773
774 if ((bio->bi_opf & REQ_POLLED) && !WARN_ON_ONCE(in_interrupt())) {
775 bio->bi_next = cache->free_list;
776 cache->free_list = bio;
777 cache->nr++;
778 } else {
779 unsigned long flags;
780
781 local_irq_save(flags);
782 bio->bi_next = cache->free_list_irq;
783 cache->free_list_irq = bio;
784 cache->nr_irq++;
785 local_irq_restore(flags);
786 }
787 put_cpu();
788}
789
790/**
791 * bio_put - release a reference to a bio
792 * @bio: bio to release reference to
793 *
794 * Description:
795 * Put a reference to a &struct bio, either one you have gotten with
796 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
797 **/
798void bio_put(struct bio *bio)
799{
800 if (unlikely(bio_flagged(bio, BIO_REFFED))) {
801 BUG_ON(!atomic_read(&bio->__bi_cnt));
802 if (!atomic_dec_and_test(&bio->__bi_cnt))
803 return;
804 }
805 if (bio->bi_opf & REQ_ALLOC_CACHE)
806 bio_put_percpu_cache(bio);
807 else
808 bio_free(bio);
809}
810EXPORT_SYMBOL(bio_put);
811
812static int __bio_clone(struct bio *bio, struct bio *bio_src, gfp_t gfp)
813{
814 bio_set_flag(bio, BIO_CLONED);
815 bio->bi_ioprio = bio_src->bi_ioprio;
816 bio->bi_iter = bio_src->bi_iter;
817
818 if (bio->bi_bdev) {
819 if (bio->bi_bdev == bio_src->bi_bdev &&
820 bio_flagged(bio_src, BIO_REMAPPED))
821 bio_set_flag(bio, BIO_REMAPPED);
822 bio_clone_blkg_association(bio, bio_src);
823 }
824
825 if (bio_crypt_clone(bio, bio_src, gfp) < 0)
826 return -ENOMEM;
827 if (bio_integrity(bio_src) &&
828 bio_integrity_clone(bio, bio_src, gfp) < 0)
829 return -ENOMEM;
830 return 0;
831}
832
833/**
834 * bio_alloc_clone - clone a bio that shares the original bio's biovec
835 * @bdev: block_device to clone onto
836 * @bio_src: bio to clone from
837 * @gfp: allocation priority
838 * @bs: bio_set to allocate from
839 *
840 * Allocate a new bio that is a clone of @bio_src. The caller owns the returned
841 * bio, but not the actual data it points to.
842 *
843 * The caller must ensure that the return bio is not freed before @bio_src.
844 */
845struct bio *bio_alloc_clone(struct block_device *bdev, struct bio *bio_src,
846 gfp_t gfp, struct bio_set *bs)
847{
848 struct bio *bio;
849
850 bio = bio_alloc_bioset(bdev, 0, bio_src->bi_opf, gfp, bs);
851 if (!bio)
852 return NULL;
853
854 if (__bio_clone(bio, bio_src, gfp) < 0) {
855 bio_put(bio);
856 return NULL;
857 }
858 bio->bi_io_vec = bio_src->bi_io_vec;
859
860 return bio;
861}
862EXPORT_SYMBOL(bio_alloc_clone);
863
864/**
865 * bio_init_clone - clone a bio that shares the original bio's biovec
866 * @bdev: block_device to clone onto
867 * @bio: bio to clone into
868 * @bio_src: bio to clone from
869 * @gfp: allocation priority
870 *
871 * Initialize a new bio in caller provided memory that is a clone of @bio_src.
872 * The caller owns the returned bio, but not the actual data it points to.
873 *
874 * The caller must ensure that @bio_src is not freed before @bio.
875 */
876int bio_init_clone(struct block_device *bdev, struct bio *bio,
877 struct bio *bio_src, gfp_t gfp)
878{
879 int ret;
880
881 bio_init(bio, bdev, bio_src->bi_io_vec, 0, bio_src->bi_opf);
882 ret = __bio_clone(bio, bio_src, gfp);
883 if (ret)
884 bio_uninit(bio);
885 return ret;
886}
887EXPORT_SYMBOL(bio_init_clone);
888
889/**
890 * bio_full - check if the bio is full
891 * @bio: bio to check
892 * @len: length of one segment to be added
893 *
894 * Return true if @bio is full and one segment with @len bytes can't be
895 * added to the bio, otherwise return false
896 */
897static inline bool bio_full(struct bio *bio, unsigned len)
898{
899 if (bio->bi_vcnt >= bio->bi_max_vecs)
900 return true;
901 if (bio->bi_iter.bi_size > UINT_MAX - len)
902 return true;
903 return false;
904}
905
906static inline bool page_is_mergeable(const struct bio_vec *bv,
907 struct page *page, unsigned int len, unsigned int off,
908 bool *same_page)
909{
910 size_t bv_end = bv->bv_offset + bv->bv_len;
911 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1;
912 phys_addr_t page_addr = page_to_phys(page);
913
914 if (vec_end_addr + 1 != page_addr + off)
915 return false;
916 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page))
917 return false;
918 if (!zone_device_pages_have_same_pgmap(bv->bv_page, page))
919 return false;
920
921 *same_page = ((vec_end_addr & PAGE_MASK) == page_addr);
922 if (*same_page)
923 return true;
924 else if (IS_ENABLED(CONFIG_KMSAN))
925 return false;
926 return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE);
927}
928
929/**
930 * __bio_try_merge_page - try appending data to an existing bvec.
931 * @bio: destination bio
932 * @page: start page to add
933 * @len: length of the data to add
934 * @off: offset of the data relative to @page
935 * @same_page: return if the segment has been merged inside the same page
936 *
937 * Try to add the data at @page + @off to the last bvec of @bio. This is a
938 * useful optimisation for file systems with a block size smaller than the
939 * page size.
940 *
941 * Warn if (@len, @off) crosses pages in case that @same_page is true.
942 *
943 * Return %true on success or %false on failure.
944 */
945static bool __bio_try_merge_page(struct bio *bio, struct page *page,
946 unsigned int len, unsigned int off, bool *same_page)
947{
948 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
949 return false;
950
951 if (bio->bi_vcnt > 0) {
952 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
953
954 if (page_is_mergeable(bv, page, len, off, same_page)) {
955 if (bio->bi_iter.bi_size > UINT_MAX - len) {
956 *same_page = false;
957 return false;
958 }
959 bv->bv_len += len;
960 bio->bi_iter.bi_size += len;
961 return true;
962 }
963 }
964 return false;
965}
966
967/*
968 * Try to merge a page into a segment, while obeying the hardware segment
969 * size limit. This is not for normal read/write bios, but for passthrough
970 * or Zone Append operations that we can't split.
971 */
972static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio,
973 struct page *page, unsigned len,
974 unsigned offset, bool *same_page)
975{
976 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
977 unsigned long mask = queue_segment_boundary(q);
978 phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset;
979 phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;
980
981 if ((addr1 | mask) != (addr2 | mask))
982 return false;
983 if (bv->bv_len + len > queue_max_segment_size(q))
984 return false;
985 return __bio_try_merge_page(bio, page, len, offset, same_page);
986}
987
988/**
989 * bio_add_hw_page - attempt to add a page to a bio with hw constraints
990 * @q: the target queue
991 * @bio: destination bio
992 * @page: page to add
993 * @len: vec entry length
994 * @offset: vec entry offset
995 * @max_sectors: maximum number of sectors that can be added
996 * @same_page: return if the segment has been merged inside the same page
997 *
998 * Add a page to a bio while respecting the hardware max_sectors, max_segment
999 * and gap limitations.
1000 */
1001int bio_add_hw_page(struct request_queue *q, struct bio *bio,
1002 struct page *page, unsigned int len, unsigned int offset,
1003 unsigned int max_sectors, bool *same_page)
1004{
1005 struct bio_vec *bvec;
1006
1007 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
1008 return 0;
1009
1010 if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors)
1011 return 0;
1012
1013 if (bio->bi_vcnt > 0) {
1014 if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page))
1015 return len;
1016
1017 /*
1018 * If the queue doesn't support SG gaps and adding this segment
1019 * would create a gap, disallow it.
1020 */
1021 bvec = &bio->bi_io_vec[bio->bi_vcnt - 1];
1022 if (bvec_gap_to_prev(&q->limits, bvec, offset))
1023 return 0;
1024 }
1025
1026 if (bio_full(bio, len))
1027 return 0;
1028
1029 if (bio->bi_vcnt >= queue_max_segments(q))
1030 return 0;
1031
1032 bvec = &bio->bi_io_vec[bio->bi_vcnt];
1033 bvec->bv_page = page;
1034 bvec->bv_len = len;
1035 bvec->bv_offset = offset;
1036 bio->bi_vcnt++;
1037 bio->bi_iter.bi_size += len;
1038 return len;
1039}
1040
1041/**
1042 * bio_add_pc_page - attempt to add page to passthrough bio
1043 * @q: the target queue
1044 * @bio: destination bio
1045 * @page: page to add
1046 * @len: vec entry length
1047 * @offset: vec entry offset
1048 *
1049 * Attempt to add a page to the bio_vec maplist. This can fail for a
1050 * number of reasons, such as the bio being full or target block device
1051 * limitations. The target block device must allow bio's up to PAGE_SIZE,
1052 * so it is always possible to add a single page to an empty bio.
1053 *
1054 * This should only be used by passthrough bios.
1055 */
1056int bio_add_pc_page(struct request_queue *q, struct bio *bio,
1057 struct page *page, unsigned int len, unsigned int offset)
1058{
1059 bool same_page = false;
1060 return bio_add_hw_page(q, bio, page, len, offset,
1061 queue_max_hw_sectors(q), &same_page);
1062}
1063EXPORT_SYMBOL(bio_add_pc_page);
1064
1065/**
1066 * bio_add_zone_append_page - attempt to add page to zone-append bio
1067 * @bio: destination bio
1068 * @page: page to add
1069 * @len: vec entry length
1070 * @offset: vec entry offset
1071 *
1072 * Attempt to add a page to the bio_vec maplist of a bio that will be submitted
1073 * for a zone-append request. This can fail for a number of reasons, such as the
1074 * bio being full or the target block device is not a zoned block device or
1075 * other limitations of the target block device. The target block device must
1076 * allow bio's up to PAGE_SIZE, so it is always possible to add a single page
1077 * to an empty bio.
1078 *
1079 * Returns: number of bytes added to the bio, or 0 in case of a failure.
1080 */
1081int bio_add_zone_append_page(struct bio *bio, struct page *page,
1082 unsigned int len, unsigned int offset)
1083{
1084 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1085 bool same_page = false;
1086
1087 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND))
1088 return 0;
1089
1090 if (WARN_ON_ONCE(!bdev_is_zoned(bio->bi_bdev)))
1091 return 0;
1092
1093 return bio_add_hw_page(q, bio, page, len, offset,
1094 queue_max_zone_append_sectors(q), &same_page);
1095}
1096EXPORT_SYMBOL_GPL(bio_add_zone_append_page);
1097
1098/**
1099 * __bio_add_page - add page(s) to a bio in a new segment
1100 * @bio: destination bio
1101 * @page: start page to add
1102 * @len: length of the data to add, may cross pages
1103 * @off: offset of the data relative to @page, may cross pages
1104 *
1105 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
1106 * that @bio has space for another bvec.
1107 */
1108void __bio_add_page(struct bio *bio, struct page *page,
1109 unsigned int len, unsigned int off)
1110{
1111 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt];
1112
1113 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
1114 WARN_ON_ONCE(bio_full(bio, len));
1115
1116 bv->bv_page = page;
1117 bv->bv_offset = off;
1118 bv->bv_len = len;
1119
1120 bio->bi_iter.bi_size += len;
1121 bio->bi_vcnt++;
1122}
1123EXPORT_SYMBOL_GPL(__bio_add_page);
1124
1125/**
1126 * bio_add_page - attempt to add page(s) to bio
1127 * @bio: destination bio
1128 * @page: start page to add
1129 * @len: vec entry length, may cross pages
1130 * @offset: vec entry offset relative to @page, may cross pages
1131 *
1132 * Attempt to add page(s) to the bio_vec maplist. This will only fail
1133 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
1134 */
1135int bio_add_page(struct bio *bio, struct page *page,
1136 unsigned int len, unsigned int offset)
1137{
1138 bool same_page = false;
1139
1140 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
1141 if (bio_full(bio, len))
1142 return 0;
1143 __bio_add_page(bio, page, len, offset);
1144 }
1145 return len;
1146}
1147EXPORT_SYMBOL(bio_add_page);
1148
1149/**
1150 * bio_add_folio - Attempt to add part of a folio to a bio.
1151 * @bio: BIO to add to.
1152 * @folio: Folio to add.
1153 * @len: How many bytes from the folio to add.
1154 * @off: First byte in this folio to add.
1155 *
1156 * Filesystems that use folios can call this function instead of calling
1157 * bio_add_page() for each page in the folio. If @off is bigger than
1158 * PAGE_SIZE, this function can create a bio_vec that starts in a page
1159 * after the bv_page. BIOs do not support folios that are 4GiB or larger.
1160 *
1161 * Return: Whether the addition was successful.
1162 */
1163bool bio_add_folio(struct bio *bio, struct folio *folio, size_t len,
1164 size_t off)
1165{
1166 if (len > UINT_MAX || off > UINT_MAX)
1167 return false;
1168 return bio_add_page(bio, &folio->page, len, off) > 0;
1169}
1170
1171void __bio_release_pages(struct bio *bio, bool mark_dirty)
1172{
1173 struct bvec_iter_all iter_all;
1174 struct bio_vec *bvec;
1175
1176 bio_for_each_segment_all(bvec, bio, iter_all) {
1177 if (mark_dirty && !PageCompound(bvec->bv_page))
1178 set_page_dirty_lock(bvec->bv_page);
1179 put_page(bvec->bv_page);
1180 }
1181}
1182EXPORT_SYMBOL_GPL(__bio_release_pages);
1183
1184void bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter)
1185{
1186 size_t size = iov_iter_count(iter);
1187
1188 WARN_ON_ONCE(bio->bi_max_vecs);
1189
1190 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1191 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1192 size_t max_sectors = queue_max_zone_append_sectors(q);
1193
1194 size = min(size, max_sectors << SECTOR_SHIFT);
1195 }
1196
1197 bio->bi_vcnt = iter->nr_segs;
1198 bio->bi_io_vec = (struct bio_vec *)iter->bvec;
1199 bio->bi_iter.bi_bvec_done = iter->iov_offset;
1200 bio->bi_iter.bi_size = size;
1201 bio_set_flag(bio, BIO_NO_PAGE_REF);
1202 bio_set_flag(bio, BIO_CLONED);
1203}
1204
1205static int bio_iov_add_page(struct bio *bio, struct page *page,
1206 unsigned int len, unsigned int offset)
1207{
1208 bool same_page = false;
1209
1210 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
1211 __bio_add_page(bio, page, len, offset);
1212 return 0;
1213 }
1214
1215 if (same_page)
1216 put_page(page);
1217 return 0;
1218}
1219
1220static int bio_iov_add_zone_append_page(struct bio *bio, struct page *page,
1221 unsigned int len, unsigned int offset)
1222{
1223 struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1224 bool same_page = false;
1225
1226 if (bio_add_hw_page(q, bio, page, len, offset,
1227 queue_max_zone_append_sectors(q), &same_page) != len)
1228 return -EINVAL;
1229 if (same_page)
1230 put_page(page);
1231 return 0;
1232}
1233
1234#define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
1235
1236/**
1237 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
1238 * @bio: bio to add pages to
1239 * @iter: iov iterator describing the region to be mapped
1240 *
1241 * Pins pages from *iter and appends them to @bio's bvec array. The
1242 * pages will have to be released using put_page() when done.
1243 * For multi-segment *iter, this function only adds pages from the
1244 * next non-empty segment of the iov iterator.
1245 */
1246static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1247{
1248 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1249 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1250 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1251 struct page **pages = (struct page **)bv;
1252 unsigned int gup_flags = 0;
1253 ssize_t size, left;
1254 unsigned len, i = 0;
1255 size_t offset, trim;
1256 int ret = 0;
1257
1258 /*
1259 * Move page array up in the allocated memory for the bio vecs as far as
1260 * possible so that we can start filling biovecs from the beginning
1261 * without overwriting the temporary page array.
1262 */
1263 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1264 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1265
1266 if (bio->bi_bdev && blk_queue_pci_p2pdma(bio->bi_bdev->bd_disk->queue))
1267 gup_flags |= FOLL_PCI_P2PDMA;
1268
1269 /*
1270 * Each segment in the iov is required to be a block size multiple.
1271 * However, we may not be able to get the entire segment if it spans
1272 * more pages than bi_max_vecs allows, so we have to ALIGN_DOWN the
1273 * result to ensure the bio's total size is correct. The remainder of
1274 * the iov data will be picked up in the next bio iteration.
1275 */
1276 size = iov_iter_get_pages(iter, pages,
1277 UINT_MAX - bio->bi_iter.bi_size,
1278 nr_pages, &offset, gup_flags);
1279 if (unlikely(size <= 0))
1280 return size ? size : -EFAULT;
1281
1282 nr_pages = DIV_ROUND_UP(offset + size, PAGE_SIZE);
1283
1284 trim = size & (bdev_logical_block_size(bio->bi_bdev) - 1);
1285 iov_iter_revert(iter, trim);
1286
1287 size -= trim;
1288 if (unlikely(!size)) {
1289 ret = -EFAULT;
1290 goto out;
1291 }
1292
1293 for (left = size, i = 0; left > 0; left -= len, i++) {
1294 struct page *page = pages[i];
1295
1296 len = min_t(size_t, PAGE_SIZE - offset, left);
1297 if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1298 ret = bio_iov_add_zone_append_page(bio, page, len,
1299 offset);
1300 if (ret)
1301 break;
1302 } else
1303 bio_iov_add_page(bio, page, len, offset);
1304
1305 offset = 0;
1306 }
1307
1308 iov_iter_revert(iter, left);
1309out:
1310 while (i < nr_pages)
1311 put_page(pages[i++]);
1312
1313 return ret;
1314}
1315
1316/**
1317 * bio_iov_iter_get_pages - add user or kernel pages to a bio
1318 * @bio: bio to add pages to
1319 * @iter: iov iterator describing the region to be added
1320 *
1321 * This takes either an iterator pointing to user memory, or one pointing to
1322 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1323 * map them into the kernel. On IO completion, the caller should put those
1324 * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided
1325 * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs
1326 * to ensure the bvecs and pages stay referenced until the submitted I/O is
1327 * completed by a call to ->ki_complete() or returns with an error other than
1328 * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF
1329 * on IO completion. If it isn't, then pages should be released.
1330 *
1331 * The function tries, but does not guarantee, to pin as many pages as
1332 * fit into the bio, or are requested in @iter, whatever is smaller. If
1333 * MM encounters an error pinning the requested pages, it stops. Error
1334 * is returned only if 0 pages could be pinned.
1335 */
1336int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1337{
1338 int ret = 0;
1339
1340 if (iov_iter_is_bvec(iter)) {
1341 bio_iov_bvec_set(bio, iter);
1342 iov_iter_advance(iter, bio->bi_iter.bi_size);
1343 return 0;
1344 }
1345
1346 do {
1347 ret = __bio_iov_iter_get_pages(bio, iter);
1348 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));
1349
1350 return bio->bi_vcnt ? 0 : ret;
1351}
1352EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
1353
1354static void submit_bio_wait_endio(struct bio *bio)
1355{
1356 complete(bio->bi_private);
1357}
1358
1359/**
1360 * submit_bio_wait - submit a bio, and wait until it completes
1361 * @bio: The &struct bio which describes the I/O
1362 *
1363 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1364 * bio_endio() on failure.
1365 *
1366 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1367 * result in bio reference to be consumed. The caller must drop the reference
1368 * on his own.
1369 */
1370int submit_bio_wait(struct bio *bio)
1371{
1372 DECLARE_COMPLETION_ONSTACK_MAP(done,
1373 bio->bi_bdev->bd_disk->lockdep_map);
1374 unsigned long hang_check;
1375
1376 bio->bi_private = &done;
1377 bio->bi_end_io = submit_bio_wait_endio;
1378 bio->bi_opf |= REQ_SYNC;
1379 submit_bio(bio);
1380
1381 /* Prevent hang_check timer from firing at us during very long I/O */
1382 hang_check = sysctl_hung_task_timeout_secs;
1383 if (hang_check)
1384 while (!wait_for_completion_io_timeout(&done,
1385 hang_check * (HZ/2)))
1386 ;
1387 else
1388 wait_for_completion_io(&done);
1389
1390 return blk_status_to_errno(bio->bi_status);
1391}
1392EXPORT_SYMBOL(submit_bio_wait);
1393
1394void __bio_advance(struct bio *bio, unsigned bytes)
1395{
1396 if (bio_integrity(bio))
1397 bio_integrity_advance(bio, bytes);
1398
1399 bio_crypt_advance(bio, bytes);
1400 bio_advance_iter(bio, &bio->bi_iter, bytes);
1401}
1402EXPORT_SYMBOL(__bio_advance);
1403
1404void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1405 struct bio *src, struct bvec_iter *src_iter)
1406{
1407 while (src_iter->bi_size && dst_iter->bi_size) {
1408 struct bio_vec src_bv = bio_iter_iovec(src, *src_iter);
1409 struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter);
1410 unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len);
1411 void *src_buf = bvec_kmap_local(&src_bv);
1412 void *dst_buf = bvec_kmap_local(&dst_bv);
1413
1414 memcpy(dst_buf, src_buf, bytes);
1415
1416 kunmap_local(dst_buf);
1417 kunmap_local(src_buf);
1418
1419 bio_advance_iter_single(src, src_iter, bytes);
1420 bio_advance_iter_single(dst, dst_iter, bytes);
1421 }
1422}
1423EXPORT_SYMBOL(bio_copy_data_iter);
1424
1425/**
1426 * bio_copy_data - copy contents of data buffers from one bio to another
1427 * @src: source bio
1428 * @dst: destination bio
1429 *
1430 * Stops when it reaches the end of either @src or @dst - that is, copies
1431 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1432 */
1433void bio_copy_data(struct bio *dst, struct bio *src)
1434{
1435 struct bvec_iter src_iter = src->bi_iter;
1436 struct bvec_iter dst_iter = dst->bi_iter;
1437
1438 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1439}
1440EXPORT_SYMBOL(bio_copy_data);
1441
1442void bio_free_pages(struct bio *bio)
1443{
1444 struct bio_vec *bvec;
1445 struct bvec_iter_all iter_all;
1446
1447 bio_for_each_segment_all(bvec, bio, iter_all)
1448 __free_page(bvec->bv_page);
1449}
1450EXPORT_SYMBOL(bio_free_pages);
1451
1452/*
1453 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1454 * for performing direct-IO in BIOs.
1455 *
1456 * The problem is that we cannot run set_page_dirty() from interrupt context
1457 * because the required locks are not interrupt-safe. So what we can do is to
1458 * mark the pages dirty _before_ performing IO. And in interrupt context,
1459 * check that the pages are still dirty. If so, fine. If not, redirty them
1460 * in process context.
1461 *
1462 * We special-case compound pages here: normally this means reads into hugetlb
1463 * pages. The logic in here doesn't really work right for compound pages
1464 * because the VM does not uniformly chase down the head page in all cases.
1465 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1466 * handle them at all. So we skip compound pages here at an early stage.
1467 *
1468 * Note that this code is very hard to test under normal circumstances because
1469 * direct-io pins the pages with get_user_pages(). This makes
1470 * is_page_cache_freeable return false, and the VM will not clean the pages.
1471 * But other code (eg, flusher threads) could clean the pages if they are mapped
1472 * pagecache.
1473 *
1474 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1475 * deferred bio dirtying paths.
1476 */
1477
1478/*
1479 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1480 */
1481void bio_set_pages_dirty(struct bio *bio)
1482{
1483 struct bio_vec *bvec;
1484 struct bvec_iter_all iter_all;
1485
1486 bio_for_each_segment_all(bvec, bio, iter_all) {
1487 if (!PageCompound(bvec->bv_page))
1488 set_page_dirty_lock(bvec->bv_page);
1489 }
1490}
1491
1492/*
1493 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1494 * If they are, then fine. If, however, some pages are clean then they must
1495 * have been written out during the direct-IO read. So we take another ref on
1496 * the BIO and re-dirty the pages in process context.
1497 *
1498 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1499 * here on. It will run one put_page() against each page and will run one
1500 * bio_put() against the BIO.
1501 */
1502
1503static void bio_dirty_fn(struct work_struct *work);
1504
1505static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1506static DEFINE_SPINLOCK(bio_dirty_lock);
1507static struct bio *bio_dirty_list;
1508
1509/*
1510 * This runs in process context
1511 */
1512static void bio_dirty_fn(struct work_struct *work)
1513{
1514 struct bio *bio, *next;
1515
1516 spin_lock_irq(&bio_dirty_lock);
1517 next = bio_dirty_list;
1518 bio_dirty_list = NULL;
1519 spin_unlock_irq(&bio_dirty_lock);
1520
1521 while ((bio = next) != NULL) {
1522 next = bio->bi_private;
1523
1524 bio_release_pages(bio, true);
1525 bio_put(bio);
1526 }
1527}
1528
1529void bio_check_pages_dirty(struct bio *bio)
1530{
1531 struct bio_vec *bvec;
1532 unsigned long flags;
1533 struct bvec_iter_all iter_all;
1534
1535 bio_for_each_segment_all(bvec, bio, iter_all) {
1536 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page))
1537 goto defer;
1538 }
1539
1540 bio_release_pages(bio, false);
1541 bio_put(bio);
1542 return;
1543defer:
1544 spin_lock_irqsave(&bio_dirty_lock, flags);
1545 bio->bi_private = bio_dirty_list;
1546 bio_dirty_list = bio;
1547 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1548 schedule_work(&bio_dirty_work);
1549}
1550
1551static inline bool bio_remaining_done(struct bio *bio)
1552{
1553 /*
1554 * If we're not chaining, then ->__bi_remaining is always 1 and
1555 * we always end io on the first invocation.
1556 */
1557 if (!bio_flagged(bio, BIO_CHAIN))
1558 return true;
1559
1560 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1561
1562 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1563 bio_clear_flag(bio, BIO_CHAIN);
1564 return true;
1565 }
1566
1567 return false;
1568}
1569
1570/**
1571 * bio_endio - end I/O on a bio
1572 * @bio: bio
1573 *
1574 * Description:
1575 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1576 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1577 * bio unless they own it and thus know that it has an end_io function.
1578 *
1579 * bio_endio() can be called several times on a bio that has been chained
1580 * using bio_chain(). The ->bi_end_io() function will only be called the
1581 * last time.
1582 **/
1583void bio_endio(struct bio *bio)
1584{
1585again:
1586 if (!bio_remaining_done(bio))
1587 return;
1588 if (!bio_integrity_endio(bio))
1589 return;
1590
1591 rq_qos_done_bio(bio);
1592
1593 if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1594 trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio);
1595 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1596 }
1597
1598 /*
1599 * Need to have a real endio function for chained bios, otherwise
1600 * various corner cases will break (like stacking block devices that
1601 * save/restore bi_end_io) - however, we want to avoid unbounded
1602 * recursion and blowing the stack. Tail call optimization would
1603 * handle this, but compiling with frame pointers also disables
1604 * gcc's sibling call optimization.
1605 */
1606 if (bio->bi_end_io == bio_chain_endio) {
1607 bio = __bio_chain_endio(bio);
1608 goto again;
1609 }
1610
1611 blk_throtl_bio_endio(bio);
1612 /* release cgroup info */
1613 bio_uninit(bio);
1614 if (bio->bi_end_io)
1615 bio->bi_end_io(bio);
1616}
1617EXPORT_SYMBOL(bio_endio);
1618
1619/**
1620 * bio_split - split a bio
1621 * @bio: bio to split
1622 * @sectors: number of sectors to split from the front of @bio
1623 * @gfp: gfp mask
1624 * @bs: bio set to allocate from
1625 *
1626 * Allocates and returns a new bio which represents @sectors from the start of
1627 * @bio, and updates @bio to represent the remaining sectors.
1628 *
1629 * Unless this is a discard request the newly allocated bio will point
1630 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1631 * neither @bio nor @bs are freed before the split bio.
1632 */
1633struct bio *bio_split(struct bio *bio, int sectors,
1634 gfp_t gfp, struct bio_set *bs)
1635{
1636 struct bio *split;
1637
1638 BUG_ON(sectors <= 0);
1639 BUG_ON(sectors >= bio_sectors(bio));
1640
1641 /* Zone append commands cannot be split */
1642 if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))
1643 return NULL;
1644
1645 split = bio_alloc_clone(bio->bi_bdev, bio, gfp, bs);
1646 if (!split)
1647 return NULL;
1648
1649 split->bi_iter.bi_size = sectors << 9;
1650
1651 if (bio_integrity(split))
1652 bio_integrity_trim(split);
1653
1654 bio_advance(bio, split->bi_iter.bi_size);
1655
1656 if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1657 bio_set_flag(split, BIO_TRACE_COMPLETION);
1658
1659 return split;
1660}
1661EXPORT_SYMBOL(bio_split);
1662
1663/**
1664 * bio_trim - trim a bio
1665 * @bio: bio to trim
1666 * @offset: number of sectors to trim from the front of @bio
1667 * @size: size we want to trim @bio to, in sectors
1668 *
1669 * This function is typically used for bios that are cloned and submitted
1670 * to the underlying device in parts.
1671 */
1672void bio_trim(struct bio *bio, sector_t offset, sector_t size)
1673{
1674 if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS ||
1675 offset + size > bio_sectors(bio)))
1676 return;
1677
1678 size <<= 9;
1679 if (offset == 0 && size == bio->bi_iter.bi_size)
1680 return;
1681
1682 bio_advance(bio, offset << 9);
1683 bio->bi_iter.bi_size = size;
1684
1685 if (bio_integrity(bio))
1686 bio_integrity_trim(bio);
1687}
1688EXPORT_SYMBOL_GPL(bio_trim);
1689
1690/*
1691 * create memory pools for biovec's in a bio_set.
1692 * use the global biovec slabs created for general use.
1693 */
1694int biovec_init_pool(mempool_t *pool, int pool_entries)
1695{
1696 struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1;
1697
1698 return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1699}
1700
1701/*
1702 * bioset_exit - exit a bioset initialized with bioset_init()
1703 *
1704 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1705 * kzalloc()).
1706 */
1707void bioset_exit(struct bio_set *bs)
1708{
1709 bio_alloc_cache_destroy(bs);
1710 if (bs->rescue_workqueue)
1711 destroy_workqueue(bs->rescue_workqueue);
1712 bs->rescue_workqueue = NULL;
1713
1714 mempool_exit(&bs->bio_pool);
1715 mempool_exit(&bs->bvec_pool);
1716
1717 bioset_integrity_free(bs);
1718 if (bs->bio_slab)
1719 bio_put_slab(bs);
1720 bs->bio_slab = NULL;
1721}
1722EXPORT_SYMBOL(bioset_exit);
1723
1724/**
1725 * bioset_init - Initialize a bio_set
1726 * @bs: pool to initialize
1727 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1728 * @front_pad: Number of bytes to allocate in front of the returned bio
1729 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1730 * and %BIOSET_NEED_RESCUER
1731 *
1732 * Description:
1733 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1734 * to ask for a number of bytes to be allocated in front of the bio.
1735 * Front pad allocation is useful for embedding the bio inside
1736 * another structure, to avoid allocating extra data to go with the bio.
1737 * Note that the bio must be embedded at the END of that structure always,
1738 * or things will break badly.
1739 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1740 * for allocating iovecs. This pool is not needed e.g. for bio_init_clone().
1741 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used
1742 * to dispatch queued requests when the mempool runs out of space.
1743 *
1744 */
1745int bioset_init(struct bio_set *bs,
1746 unsigned int pool_size,
1747 unsigned int front_pad,
1748 int flags)
1749{
1750 bs->front_pad = front_pad;
1751 if (flags & BIOSET_NEED_BVECS)
1752 bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1753 else
1754 bs->back_pad = 0;
1755
1756 spin_lock_init(&bs->rescue_lock);
1757 bio_list_init(&bs->rescue_list);
1758 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1759
1760 bs->bio_slab = bio_find_or_create_slab(bs);
1761 if (!bs->bio_slab)
1762 return -ENOMEM;
1763
1764 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
1765 goto bad;
1766
1767 if ((flags & BIOSET_NEED_BVECS) &&
1768 biovec_init_pool(&bs->bvec_pool, pool_size))
1769 goto bad;
1770
1771 if (flags & BIOSET_NEED_RESCUER) {
1772 bs->rescue_workqueue = alloc_workqueue("bioset",
1773 WQ_MEM_RECLAIM, 0);
1774 if (!bs->rescue_workqueue)
1775 goto bad;
1776 }
1777 if (flags & BIOSET_PERCPU_CACHE) {
1778 bs->cache = alloc_percpu(struct bio_alloc_cache);
1779 if (!bs->cache)
1780 goto bad;
1781 cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
1782 }
1783
1784 return 0;
1785bad:
1786 bioset_exit(bs);
1787 return -ENOMEM;
1788}
1789EXPORT_SYMBOL(bioset_init);
1790
1791static int __init init_bio(void)
1792{
1793 int i;
1794
1795 bio_integrity_init();
1796
1797 for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) {
1798 struct biovec_slab *bvs = bvec_slabs + i;
1799
1800 bvs->slab = kmem_cache_create(bvs->name,
1801 bvs->nr_vecs * sizeof(struct bio_vec), 0,
1802 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1803 }
1804
1805 cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL,
1806 bio_cpu_dead);
1807
1808 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0,
1809 BIOSET_NEED_BVECS | BIOSET_PERCPU_CACHE))
1810 panic("bio: can't allocate bios\n");
1811
1812 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
1813 panic("bio: can't create integrity pool\n");
1814
1815 return 0;
1816}
1817subsys_initcall(init_bio);
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
4 */
5#include <linux/mm.h>
6#include <linux/swap.h>
7#include <linux/bio.h>
8#include <linux/blkdev.h>
9#include <linux/uio.h>
10#include <linux/iocontext.h>
11#include <linux/slab.h>
12#include <linux/init.h>
13#include <linux/kernel.h>
14#include <linux/export.h>
15#include <linux/mempool.h>
16#include <linux/workqueue.h>
17#include <linux/cgroup.h>
18#include <linux/blk-cgroup.h>
19#include <linux/highmem.h>
20#include <linux/sched/sysctl.h>
21#include <linux/blk-crypto.h>
22#include <linux/xarray.h>
23
24#include <trace/events/block.h>
25#include "blk.h"
26#include "blk-rq-qos.h"
27
28static struct biovec_slab {
29 int nr_vecs;
30 char *name;
31 struct kmem_cache *slab;
32} bvec_slabs[] __read_mostly = {
33 { .nr_vecs = 16, .name = "biovec-16" },
34 { .nr_vecs = 64, .name = "biovec-64" },
35 { .nr_vecs = 128, .name = "biovec-128" },
36 { .nr_vecs = BIO_MAX_VECS, .name = "biovec-max" },
37};
38
39static struct biovec_slab *biovec_slab(unsigned short nr_vecs)
40{
41 switch (nr_vecs) {
42 /* smaller bios use inline vecs */
43 case 5 ... 16:
44 return &bvec_slabs[0];
45 case 17 ... 64:
46 return &bvec_slabs[1];
47 case 65 ... 128:
48 return &bvec_slabs[2];
49 case 129 ... BIO_MAX_VECS:
50 return &bvec_slabs[3];
51 default:
52 BUG();
53 return NULL;
54 }
55}
56
57/*
58 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
59 * IO code that does not need private memory pools.
60 */
61struct bio_set fs_bio_set;
62EXPORT_SYMBOL(fs_bio_set);
63
64/*
65 * Our slab pool management
66 */
67struct bio_slab {
68 struct kmem_cache *slab;
69 unsigned int slab_ref;
70 unsigned int slab_size;
71 char name[8];
72};
73static DEFINE_MUTEX(bio_slab_lock);
74static DEFINE_XARRAY(bio_slabs);
75
76static struct bio_slab *create_bio_slab(unsigned int size)
77{
78 struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL);
79
80 if (!bslab)
81 return NULL;
82
83 snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size);
84 bslab->slab = kmem_cache_create(bslab->name, size,
85 ARCH_KMALLOC_MINALIGN, SLAB_HWCACHE_ALIGN, NULL);
86 if (!bslab->slab)
87 goto fail_alloc_slab;
88
89 bslab->slab_ref = 1;
90 bslab->slab_size = size;
91
92 if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL)))
93 return bslab;
94
95 kmem_cache_destroy(bslab->slab);
96
97fail_alloc_slab:
98 kfree(bslab);
99 return NULL;
100}
101
102static inline unsigned int bs_bio_slab_size(struct bio_set *bs)
103{
104 return bs->front_pad + sizeof(struct bio) + bs->back_pad;
105}
106
107static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs)
108{
109 unsigned int size = bs_bio_slab_size(bs);
110 struct bio_slab *bslab;
111
112 mutex_lock(&bio_slab_lock);
113 bslab = xa_load(&bio_slabs, size);
114 if (bslab)
115 bslab->slab_ref++;
116 else
117 bslab = create_bio_slab(size);
118 mutex_unlock(&bio_slab_lock);
119
120 if (bslab)
121 return bslab->slab;
122 return NULL;
123}
124
125static void bio_put_slab(struct bio_set *bs)
126{
127 struct bio_slab *bslab = NULL;
128 unsigned int slab_size = bs_bio_slab_size(bs);
129
130 mutex_lock(&bio_slab_lock);
131
132 bslab = xa_load(&bio_slabs, slab_size);
133 if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
134 goto out;
135
136 WARN_ON_ONCE(bslab->slab != bs->bio_slab);
137
138 WARN_ON(!bslab->slab_ref);
139
140 if (--bslab->slab_ref)
141 goto out;
142
143 xa_erase(&bio_slabs, slab_size);
144
145 kmem_cache_destroy(bslab->slab);
146 kfree(bslab);
147
148out:
149 mutex_unlock(&bio_slab_lock);
150}
151
152void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs)
153{
154 BIO_BUG_ON(nr_vecs > BIO_MAX_VECS);
155
156 if (nr_vecs == BIO_MAX_VECS)
157 mempool_free(bv, pool);
158 else if (nr_vecs > BIO_INLINE_VECS)
159 kmem_cache_free(biovec_slab(nr_vecs)->slab, bv);
160}
161
162/*
163 * Make the first allocation restricted and don't dump info on allocation
164 * failures, since we'll fall back to the mempool in case of failure.
165 */
166static inline gfp_t bvec_alloc_gfp(gfp_t gfp)
167{
168 return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) |
169 __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
170}
171
172struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
173 gfp_t gfp_mask)
174{
175 struct biovec_slab *bvs = biovec_slab(*nr_vecs);
176
177 if (WARN_ON_ONCE(!bvs))
178 return NULL;
179
180 /*
181 * Upgrade the nr_vecs request to take full advantage of the allocation.
182 * We also rely on this in the bvec_free path.
183 */
184 *nr_vecs = bvs->nr_vecs;
185
186 /*
187 * Try a slab allocation first for all smaller allocations. If that
188 * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool.
189 * The mempool is sized to handle up to BIO_MAX_VECS entries.
190 */
191 if (*nr_vecs < BIO_MAX_VECS) {
192 struct bio_vec *bvl;
193
194 bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask));
195 if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM))
196 return bvl;
197 *nr_vecs = BIO_MAX_VECS;
198 }
199
200 return mempool_alloc(pool, gfp_mask);
201}
202
203void bio_uninit(struct bio *bio)
204{
205#ifdef CONFIG_BLK_CGROUP
206 if (bio->bi_blkg) {
207 blkg_put(bio->bi_blkg);
208 bio->bi_blkg = NULL;
209 }
210#endif
211 if (bio_integrity(bio))
212 bio_integrity_free(bio);
213
214 bio_crypt_free_ctx(bio);
215}
216EXPORT_SYMBOL(bio_uninit);
217
218static void bio_free(struct bio *bio)
219{
220 struct bio_set *bs = bio->bi_pool;
221 void *p;
222
223 bio_uninit(bio);
224
225 if (bs) {
226 bvec_free(&bs->bvec_pool, bio->bi_io_vec, bio->bi_max_vecs);
227
228 /*
229 * If we have front padding, adjust the bio pointer before freeing
230 */
231 p = bio;
232 p -= bs->front_pad;
233
234 mempool_free(p, &bs->bio_pool);
235 } else {
236 /* Bio was allocated by bio_kmalloc() */
237 kfree(bio);
238 }
239}
240
241/*
242 * Users of this function have their own bio allocation. Subsequently,
243 * they must remember to pair any call to bio_init() with bio_uninit()
244 * when IO has completed, or when the bio is released.
245 */
246void bio_init(struct bio *bio, struct bio_vec *table,
247 unsigned short max_vecs)
248{
249 memset(bio, 0, sizeof(*bio));
250 atomic_set(&bio->__bi_remaining, 1);
251 atomic_set(&bio->__bi_cnt, 1);
252
253 bio->bi_io_vec = table;
254 bio->bi_max_vecs = max_vecs;
255}
256EXPORT_SYMBOL(bio_init);
257
258/**
259 * bio_reset - reinitialize a bio
260 * @bio: bio to reset
261 *
262 * Description:
263 * After calling bio_reset(), @bio will be in the same state as a freshly
264 * allocated bio returned bio bio_alloc_bioset() - the only fields that are
265 * preserved are the ones that are initialized by bio_alloc_bioset(). See
266 * comment in struct bio.
267 */
268void bio_reset(struct bio *bio)
269{
270 bio_uninit(bio);
271 memset(bio, 0, BIO_RESET_BYTES);
272 atomic_set(&bio->__bi_remaining, 1);
273}
274EXPORT_SYMBOL(bio_reset);
275
276static struct bio *__bio_chain_endio(struct bio *bio)
277{
278 struct bio *parent = bio->bi_private;
279
280 if (bio->bi_status && !parent->bi_status)
281 parent->bi_status = bio->bi_status;
282 bio_put(bio);
283 return parent;
284}
285
286static void bio_chain_endio(struct bio *bio)
287{
288 bio_endio(__bio_chain_endio(bio));
289}
290
291/**
292 * bio_chain - chain bio completions
293 * @bio: the target bio
294 * @parent: the parent bio of @bio
295 *
296 * The caller won't have a bi_end_io called when @bio completes - instead,
297 * @parent's bi_end_io won't be called until both @parent and @bio have
298 * completed; the chained bio will also be freed when it completes.
299 *
300 * The caller must not set bi_private or bi_end_io in @bio.
301 */
302void bio_chain(struct bio *bio, struct bio *parent)
303{
304 BUG_ON(bio->bi_private || bio->bi_end_io);
305
306 bio->bi_private = parent;
307 bio->bi_end_io = bio_chain_endio;
308 bio_inc_remaining(parent);
309}
310EXPORT_SYMBOL(bio_chain);
311
312static void bio_alloc_rescue(struct work_struct *work)
313{
314 struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
315 struct bio *bio;
316
317 while (1) {
318 spin_lock(&bs->rescue_lock);
319 bio = bio_list_pop(&bs->rescue_list);
320 spin_unlock(&bs->rescue_lock);
321
322 if (!bio)
323 break;
324
325 submit_bio_noacct(bio);
326 }
327}
328
329static void punt_bios_to_rescuer(struct bio_set *bs)
330{
331 struct bio_list punt, nopunt;
332 struct bio *bio;
333
334 if (WARN_ON_ONCE(!bs->rescue_workqueue))
335 return;
336 /*
337 * In order to guarantee forward progress we must punt only bios that
338 * were allocated from this bio_set; otherwise, if there was a bio on
339 * there for a stacking driver higher up in the stack, processing it
340 * could require allocating bios from this bio_set, and doing that from
341 * our own rescuer would be bad.
342 *
343 * Since bio lists are singly linked, pop them all instead of trying to
344 * remove from the middle of the list:
345 */
346
347 bio_list_init(&punt);
348 bio_list_init(&nopunt);
349
350 while ((bio = bio_list_pop(¤t->bio_list[0])))
351 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
352 current->bio_list[0] = nopunt;
353
354 bio_list_init(&nopunt);
355 while ((bio = bio_list_pop(¤t->bio_list[1])))
356 bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
357 current->bio_list[1] = nopunt;
358
359 spin_lock(&bs->rescue_lock);
360 bio_list_merge(&bs->rescue_list, &punt);
361 spin_unlock(&bs->rescue_lock);
362
363 queue_work(bs->rescue_workqueue, &bs->rescue_work);
364}
365
366/**
367 * bio_alloc_bioset - allocate a bio for I/O
368 * @gfp_mask: the GFP_* mask given to the slab allocator
369 * @nr_iovecs: number of iovecs to pre-allocate
370 * @bs: the bio_set to allocate from.
371 *
372 * Allocate a bio from the mempools in @bs.
373 *
374 * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to
375 * allocate a bio. This is due to the mempool guarantees. To make this work,
376 * callers must never allocate more than 1 bio at a time from the general pool.
377 * Callers that need to allocate more than 1 bio must always submit the
378 * previously allocated bio for IO before attempting to allocate a new one.
379 * Failure to do so can cause deadlocks under memory pressure.
380 *
381 * Note that when running under submit_bio_noacct() (i.e. any block driver),
382 * bios are not submitted until after you return - see the code in
383 * submit_bio_noacct() that converts recursion into iteration, to prevent
384 * stack overflows.
385 *
386 * This would normally mean allocating multiple bios under submit_bio_noacct()
387 * would be susceptible to deadlocks, but we have
388 * deadlock avoidance code that resubmits any blocked bios from a rescuer
389 * thread.
390 *
391 * However, we do not guarantee forward progress for allocations from other
392 * mempools. Doing multiple allocations from the same mempool under
393 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad
394 * for per bio allocations.
395 *
396 * Returns: Pointer to new bio on success, NULL on failure.
397 */
398struct bio *bio_alloc_bioset(gfp_t gfp_mask, unsigned short nr_iovecs,
399 struct bio_set *bs)
400{
401 gfp_t saved_gfp = gfp_mask;
402 struct bio *bio;
403 void *p;
404
405 /* should not use nobvec bioset for nr_iovecs > 0 */
406 if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_iovecs > 0))
407 return NULL;
408
409 /*
410 * submit_bio_noacct() converts recursion to iteration; this means if
411 * we're running beneath it, any bios we allocate and submit will not be
412 * submitted (and thus freed) until after we return.
413 *
414 * This exposes us to a potential deadlock if we allocate multiple bios
415 * from the same bio_set() while running underneath submit_bio_noacct().
416 * If we were to allocate multiple bios (say a stacking block driver
417 * that was splitting bios), we would deadlock if we exhausted the
418 * mempool's reserve.
419 *
420 * We solve this, and guarantee forward progress, with a rescuer
421 * workqueue per bio_set. If we go to allocate and there are bios on
422 * current->bio_list, we first try the allocation without
423 * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be
424 * blocking to the rescuer workqueue before we retry with the original
425 * gfp_flags.
426 */
427 if (current->bio_list &&
428 (!bio_list_empty(¤t->bio_list[0]) ||
429 !bio_list_empty(¤t->bio_list[1])) &&
430 bs->rescue_workqueue)
431 gfp_mask &= ~__GFP_DIRECT_RECLAIM;
432
433 p = mempool_alloc(&bs->bio_pool, gfp_mask);
434 if (!p && gfp_mask != saved_gfp) {
435 punt_bios_to_rescuer(bs);
436 gfp_mask = saved_gfp;
437 p = mempool_alloc(&bs->bio_pool, gfp_mask);
438 }
439 if (unlikely(!p))
440 return NULL;
441
442 bio = p + bs->front_pad;
443 if (nr_iovecs > BIO_INLINE_VECS) {
444 struct bio_vec *bvl = NULL;
445
446 bvl = bvec_alloc(&bs->bvec_pool, &nr_iovecs, gfp_mask);
447 if (!bvl && gfp_mask != saved_gfp) {
448 punt_bios_to_rescuer(bs);
449 gfp_mask = saved_gfp;
450 bvl = bvec_alloc(&bs->bvec_pool, &nr_iovecs, gfp_mask);
451 }
452 if (unlikely(!bvl))
453 goto err_free;
454
455 bio_init(bio, bvl, nr_iovecs);
456 } else if (nr_iovecs) {
457 bio_init(bio, bio->bi_inline_vecs, BIO_INLINE_VECS);
458 } else {
459 bio_init(bio, NULL, 0);
460 }
461
462 bio->bi_pool = bs;
463 return bio;
464
465err_free:
466 mempool_free(p, &bs->bio_pool);
467 return NULL;
468}
469EXPORT_SYMBOL(bio_alloc_bioset);
470
471/**
472 * bio_kmalloc - kmalloc a bio for I/O
473 * @gfp_mask: the GFP_* mask given to the slab allocator
474 * @nr_iovecs: number of iovecs to pre-allocate
475 *
476 * Use kmalloc to allocate and initialize a bio.
477 *
478 * Returns: Pointer to new bio on success, NULL on failure.
479 */
480struct bio *bio_kmalloc(gfp_t gfp_mask, unsigned short nr_iovecs)
481{
482 struct bio *bio;
483
484 if (nr_iovecs > UIO_MAXIOV)
485 return NULL;
486
487 bio = kmalloc(struct_size(bio, bi_inline_vecs, nr_iovecs), gfp_mask);
488 if (unlikely(!bio))
489 return NULL;
490 bio_init(bio, nr_iovecs ? bio->bi_inline_vecs : NULL, nr_iovecs);
491 bio->bi_pool = NULL;
492 return bio;
493}
494EXPORT_SYMBOL(bio_kmalloc);
495
496void zero_fill_bio(struct bio *bio)
497{
498 unsigned long flags;
499 struct bio_vec bv;
500 struct bvec_iter iter;
501
502 bio_for_each_segment(bv, bio, iter) {
503 char *data = bvec_kmap_irq(&bv, &flags);
504 memset(data, 0, bv.bv_len);
505 flush_dcache_page(bv.bv_page);
506 bvec_kunmap_irq(data, &flags);
507 }
508}
509EXPORT_SYMBOL(zero_fill_bio);
510
511/**
512 * bio_truncate - truncate the bio to small size of @new_size
513 * @bio: the bio to be truncated
514 * @new_size: new size for truncating the bio
515 *
516 * Description:
517 * Truncate the bio to new size of @new_size. If bio_op(bio) is
518 * REQ_OP_READ, zero the truncated part. This function should only
519 * be used for handling corner cases, such as bio eod.
520 */
521void bio_truncate(struct bio *bio, unsigned new_size)
522{
523 struct bio_vec bv;
524 struct bvec_iter iter;
525 unsigned int done = 0;
526 bool truncated = false;
527
528 if (new_size >= bio->bi_iter.bi_size)
529 return;
530
531 if (bio_op(bio) != REQ_OP_READ)
532 goto exit;
533
534 bio_for_each_segment(bv, bio, iter) {
535 if (done + bv.bv_len > new_size) {
536 unsigned offset;
537
538 if (!truncated)
539 offset = new_size - done;
540 else
541 offset = 0;
542 zero_user(bv.bv_page, offset, bv.bv_len - offset);
543 truncated = true;
544 }
545 done += bv.bv_len;
546 }
547
548 exit:
549 /*
550 * Don't touch bvec table here and make it really immutable, since
551 * fs bio user has to retrieve all pages via bio_for_each_segment_all
552 * in its .end_bio() callback.
553 *
554 * It is enough to truncate bio by updating .bi_size since we can make
555 * correct bvec with the updated .bi_size for drivers.
556 */
557 bio->bi_iter.bi_size = new_size;
558}
559
560/**
561 * guard_bio_eod - truncate a BIO to fit the block device
562 * @bio: bio to truncate
563 *
564 * This allows us to do IO even on the odd last sectors of a device, even if the
565 * block size is some multiple of the physical sector size.
566 *
567 * We'll just truncate the bio to the size of the device, and clear the end of
568 * the buffer head manually. Truly out-of-range accesses will turn into actual
569 * I/O errors, this only handles the "we need to be able to do I/O at the final
570 * sector" case.
571 */
572void guard_bio_eod(struct bio *bio)
573{
574 sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
575
576 if (!maxsector)
577 return;
578
579 /*
580 * If the *whole* IO is past the end of the device,
581 * let it through, and the IO layer will turn it into
582 * an EIO.
583 */
584 if (unlikely(bio->bi_iter.bi_sector >= maxsector))
585 return;
586
587 maxsector -= bio->bi_iter.bi_sector;
588 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
589 return;
590
591 bio_truncate(bio, maxsector << 9);
592}
593
594/**
595 * bio_put - release a reference to a bio
596 * @bio: bio to release reference to
597 *
598 * Description:
599 * Put a reference to a &struct bio, either one you have gotten with
600 * bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
601 **/
602void bio_put(struct bio *bio)
603{
604 if (!bio_flagged(bio, BIO_REFFED))
605 bio_free(bio);
606 else {
607 BIO_BUG_ON(!atomic_read(&bio->__bi_cnt));
608
609 /*
610 * last put frees it
611 */
612 if (atomic_dec_and_test(&bio->__bi_cnt))
613 bio_free(bio);
614 }
615}
616EXPORT_SYMBOL(bio_put);
617
618/**
619 * __bio_clone_fast - clone a bio that shares the original bio's biovec
620 * @bio: destination bio
621 * @bio_src: bio to clone
622 *
623 * Clone a &bio. Caller will own the returned bio, but not
624 * the actual data it points to. Reference count of returned
625 * bio will be one.
626 *
627 * Caller must ensure that @bio_src is not freed before @bio.
628 */
629void __bio_clone_fast(struct bio *bio, struct bio *bio_src)
630{
631 WARN_ON_ONCE(bio->bi_pool && bio->bi_max_vecs);
632
633 /*
634 * most users will be overriding ->bi_bdev with a new target,
635 * so we don't set nor calculate new physical/hw segment counts here
636 */
637 bio->bi_bdev = bio_src->bi_bdev;
638 bio_set_flag(bio, BIO_CLONED);
639 if (bio_flagged(bio_src, BIO_THROTTLED))
640 bio_set_flag(bio, BIO_THROTTLED);
641 if (bio_flagged(bio_src, BIO_REMAPPED))
642 bio_set_flag(bio, BIO_REMAPPED);
643 bio->bi_opf = bio_src->bi_opf;
644 bio->bi_ioprio = bio_src->bi_ioprio;
645 bio->bi_write_hint = bio_src->bi_write_hint;
646 bio->bi_iter = bio_src->bi_iter;
647 bio->bi_io_vec = bio_src->bi_io_vec;
648
649 bio_clone_blkg_association(bio, bio_src);
650 blkcg_bio_issue_init(bio);
651}
652EXPORT_SYMBOL(__bio_clone_fast);
653
654/**
655 * bio_clone_fast - clone a bio that shares the original bio's biovec
656 * @bio: bio to clone
657 * @gfp_mask: allocation priority
658 * @bs: bio_set to allocate from
659 *
660 * Like __bio_clone_fast, only also allocates the returned bio
661 */
662struct bio *bio_clone_fast(struct bio *bio, gfp_t gfp_mask, struct bio_set *bs)
663{
664 struct bio *b;
665
666 b = bio_alloc_bioset(gfp_mask, 0, bs);
667 if (!b)
668 return NULL;
669
670 __bio_clone_fast(b, bio);
671
672 if (bio_crypt_clone(b, bio, gfp_mask) < 0)
673 goto err_put;
674
675 if (bio_integrity(bio) &&
676 bio_integrity_clone(b, bio, gfp_mask) < 0)
677 goto err_put;
678
679 return b;
680
681err_put:
682 bio_put(b);
683 return NULL;
684}
685EXPORT_SYMBOL(bio_clone_fast);
686
687const char *bio_devname(struct bio *bio, char *buf)
688{
689 return bdevname(bio->bi_bdev, buf);
690}
691EXPORT_SYMBOL(bio_devname);
692
693static inline bool page_is_mergeable(const struct bio_vec *bv,
694 struct page *page, unsigned int len, unsigned int off,
695 bool *same_page)
696{
697 size_t bv_end = bv->bv_offset + bv->bv_len;
698 phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1;
699 phys_addr_t page_addr = page_to_phys(page);
700
701 if (vec_end_addr + 1 != page_addr + off)
702 return false;
703 if (xen_domain() && !xen_biovec_phys_mergeable(bv, page))
704 return false;
705
706 *same_page = ((vec_end_addr & PAGE_MASK) == page_addr);
707 if (*same_page)
708 return true;
709 return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE);
710}
711
712/*
713 * Try to merge a page into a segment, while obeying the hardware segment
714 * size limit. This is not for normal read/write bios, but for passthrough
715 * or Zone Append operations that we can't split.
716 */
717static bool bio_try_merge_hw_seg(struct request_queue *q, struct bio *bio,
718 struct page *page, unsigned len,
719 unsigned offset, bool *same_page)
720{
721 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
722 unsigned long mask = queue_segment_boundary(q);
723 phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset;
724 phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;
725
726 if ((addr1 | mask) != (addr2 | mask))
727 return false;
728 if (bv->bv_len + len > queue_max_segment_size(q))
729 return false;
730 return __bio_try_merge_page(bio, page, len, offset, same_page);
731}
732
733/**
734 * bio_add_hw_page - attempt to add a page to a bio with hw constraints
735 * @q: the target queue
736 * @bio: destination bio
737 * @page: page to add
738 * @len: vec entry length
739 * @offset: vec entry offset
740 * @max_sectors: maximum number of sectors that can be added
741 * @same_page: return if the segment has been merged inside the same page
742 *
743 * Add a page to a bio while respecting the hardware max_sectors, max_segment
744 * and gap limitations.
745 */
746int bio_add_hw_page(struct request_queue *q, struct bio *bio,
747 struct page *page, unsigned int len, unsigned int offset,
748 unsigned int max_sectors, bool *same_page)
749{
750 struct bio_vec *bvec;
751
752 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
753 return 0;
754
755 if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors)
756 return 0;
757
758 if (bio->bi_vcnt > 0) {
759 if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page))
760 return len;
761
762 /*
763 * If the queue doesn't support SG gaps and adding this segment
764 * would create a gap, disallow it.
765 */
766 bvec = &bio->bi_io_vec[bio->bi_vcnt - 1];
767 if (bvec_gap_to_prev(q, bvec, offset))
768 return 0;
769 }
770
771 if (bio_full(bio, len))
772 return 0;
773
774 if (bio->bi_vcnt >= queue_max_segments(q))
775 return 0;
776
777 bvec = &bio->bi_io_vec[bio->bi_vcnt];
778 bvec->bv_page = page;
779 bvec->bv_len = len;
780 bvec->bv_offset = offset;
781 bio->bi_vcnt++;
782 bio->bi_iter.bi_size += len;
783 return len;
784}
785
786/**
787 * bio_add_pc_page - attempt to add page to passthrough bio
788 * @q: the target queue
789 * @bio: destination bio
790 * @page: page to add
791 * @len: vec entry length
792 * @offset: vec entry offset
793 *
794 * Attempt to add a page to the bio_vec maplist. This can fail for a
795 * number of reasons, such as the bio being full or target block device
796 * limitations. The target block device must allow bio's up to PAGE_SIZE,
797 * so it is always possible to add a single page to an empty bio.
798 *
799 * This should only be used by passthrough bios.
800 */
801int bio_add_pc_page(struct request_queue *q, struct bio *bio,
802 struct page *page, unsigned int len, unsigned int offset)
803{
804 bool same_page = false;
805 return bio_add_hw_page(q, bio, page, len, offset,
806 queue_max_hw_sectors(q), &same_page);
807}
808EXPORT_SYMBOL(bio_add_pc_page);
809
810/**
811 * bio_add_zone_append_page - attempt to add page to zone-append bio
812 * @bio: destination bio
813 * @page: page to add
814 * @len: vec entry length
815 * @offset: vec entry offset
816 *
817 * Attempt to add a page to the bio_vec maplist of a bio that will be submitted
818 * for a zone-append request. This can fail for a number of reasons, such as the
819 * bio being full or the target block device is not a zoned block device or
820 * other limitations of the target block device. The target block device must
821 * allow bio's up to PAGE_SIZE, so it is always possible to add a single page
822 * to an empty bio.
823 *
824 * Returns: number of bytes added to the bio, or 0 in case of a failure.
825 */
826int bio_add_zone_append_page(struct bio *bio, struct page *page,
827 unsigned int len, unsigned int offset)
828{
829 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
830 bool same_page = false;
831
832 if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND))
833 return 0;
834
835 if (WARN_ON_ONCE(!blk_queue_is_zoned(q)))
836 return 0;
837
838 return bio_add_hw_page(q, bio, page, len, offset,
839 queue_max_zone_append_sectors(q), &same_page);
840}
841EXPORT_SYMBOL_GPL(bio_add_zone_append_page);
842
843/**
844 * __bio_try_merge_page - try appending data to an existing bvec.
845 * @bio: destination bio
846 * @page: start page to add
847 * @len: length of the data to add
848 * @off: offset of the data relative to @page
849 * @same_page: return if the segment has been merged inside the same page
850 *
851 * Try to add the data at @page + @off to the last bvec of @bio. This is a
852 * useful optimisation for file systems with a block size smaller than the
853 * page size.
854 *
855 * Warn if (@len, @off) crosses pages in case that @same_page is true.
856 *
857 * Return %true on success or %false on failure.
858 */
859bool __bio_try_merge_page(struct bio *bio, struct page *page,
860 unsigned int len, unsigned int off, bool *same_page)
861{
862 if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
863 return false;
864
865 if (bio->bi_vcnt > 0) {
866 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
867
868 if (page_is_mergeable(bv, page, len, off, same_page)) {
869 if (bio->bi_iter.bi_size > UINT_MAX - len) {
870 *same_page = false;
871 return false;
872 }
873 bv->bv_len += len;
874 bio->bi_iter.bi_size += len;
875 return true;
876 }
877 }
878 return false;
879}
880EXPORT_SYMBOL_GPL(__bio_try_merge_page);
881
882/**
883 * __bio_add_page - add page(s) to a bio in a new segment
884 * @bio: destination bio
885 * @page: start page to add
886 * @len: length of the data to add, may cross pages
887 * @off: offset of the data relative to @page, may cross pages
888 *
889 * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
890 * that @bio has space for another bvec.
891 */
892void __bio_add_page(struct bio *bio, struct page *page,
893 unsigned int len, unsigned int off)
894{
895 struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt];
896
897 WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
898 WARN_ON_ONCE(bio_full(bio, len));
899
900 bv->bv_page = page;
901 bv->bv_offset = off;
902 bv->bv_len = len;
903
904 bio->bi_iter.bi_size += len;
905 bio->bi_vcnt++;
906
907 if (!bio_flagged(bio, BIO_WORKINGSET) && unlikely(PageWorkingset(page)))
908 bio_set_flag(bio, BIO_WORKINGSET);
909}
910EXPORT_SYMBOL_GPL(__bio_add_page);
911
912/**
913 * bio_add_page - attempt to add page(s) to bio
914 * @bio: destination bio
915 * @page: start page to add
916 * @len: vec entry length, may cross pages
917 * @offset: vec entry offset relative to @page, may cross pages
918 *
919 * Attempt to add page(s) to the bio_vec maplist. This will only fail
920 * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
921 */
922int bio_add_page(struct bio *bio, struct page *page,
923 unsigned int len, unsigned int offset)
924{
925 bool same_page = false;
926
927 if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
928 if (bio_full(bio, len))
929 return 0;
930 __bio_add_page(bio, page, len, offset);
931 }
932 return len;
933}
934EXPORT_SYMBOL(bio_add_page);
935
936void bio_release_pages(struct bio *bio, bool mark_dirty)
937{
938 struct bvec_iter_all iter_all;
939 struct bio_vec *bvec;
940
941 if (bio_flagged(bio, BIO_NO_PAGE_REF))
942 return;
943
944 bio_for_each_segment_all(bvec, bio, iter_all) {
945 if (mark_dirty && !PageCompound(bvec->bv_page))
946 set_page_dirty_lock(bvec->bv_page);
947 put_page(bvec->bv_page);
948 }
949}
950EXPORT_SYMBOL_GPL(bio_release_pages);
951
952static void __bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter)
953{
954 WARN_ON_ONCE(bio->bi_max_vecs);
955
956 bio->bi_vcnt = iter->nr_segs;
957 bio->bi_io_vec = (struct bio_vec *)iter->bvec;
958 bio->bi_iter.bi_bvec_done = iter->iov_offset;
959 bio->bi_iter.bi_size = iter->count;
960 bio_set_flag(bio, BIO_NO_PAGE_REF);
961 bio_set_flag(bio, BIO_CLONED);
962}
963
964static int bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter)
965{
966 __bio_iov_bvec_set(bio, iter);
967 iov_iter_advance(iter, iter->count);
968 return 0;
969}
970
971static int bio_iov_bvec_set_append(struct bio *bio, struct iov_iter *iter)
972{
973 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
974 struct iov_iter i = *iter;
975
976 iov_iter_truncate(&i, queue_max_zone_append_sectors(q) << 9);
977 __bio_iov_bvec_set(bio, &i);
978 iov_iter_advance(iter, i.count);
979 return 0;
980}
981
982static void bio_put_pages(struct page **pages, size_t size, size_t off)
983{
984 size_t i, nr = DIV_ROUND_UP(size + (off & ~PAGE_MASK), PAGE_SIZE);
985
986 for (i = 0; i < nr; i++)
987 put_page(pages[i]);
988}
989
990#define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
991
992/**
993 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
994 * @bio: bio to add pages to
995 * @iter: iov iterator describing the region to be mapped
996 *
997 * Pins pages from *iter and appends them to @bio's bvec array. The
998 * pages will have to be released using put_page() when done.
999 * For multi-segment *iter, this function only adds pages from the
1000 * next non-empty segment of the iov iterator.
1001 */
1002static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1003{
1004 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1005 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1006 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1007 struct page **pages = (struct page **)bv;
1008 bool same_page = false;
1009 ssize_t size, left;
1010 unsigned len, i;
1011 size_t offset;
1012
1013 /*
1014 * Move page array up in the allocated memory for the bio vecs as far as
1015 * possible so that we can start filling biovecs from the beginning
1016 * without overwriting the temporary page array.
1017 */
1018 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1019 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1020
1021 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
1022 if (unlikely(size <= 0))
1023 return size ? size : -EFAULT;
1024
1025 for (left = size, i = 0; left > 0; left -= len, i++) {
1026 struct page *page = pages[i];
1027
1028 len = min_t(size_t, PAGE_SIZE - offset, left);
1029
1030 if (__bio_try_merge_page(bio, page, len, offset, &same_page)) {
1031 if (same_page)
1032 put_page(page);
1033 } else {
1034 if (WARN_ON_ONCE(bio_full(bio, len))) {
1035 bio_put_pages(pages + i, left, offset);
1036 return -EINVAL;
1037 }
1038 __bio_add_page(bio, page, len, offset);
1039 }
1040 offset = 0;
1041 }
1042
1043 iov_iter_advance(iter, size);
1044 return 0;
1045}
1046
1047static int __bio_iov_append_get_pages(struct bio *bio, struct iov_iter *iter)
1048{
1049 unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1050 unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1051 struct request_queue *q = bio->bi_bdev->bd_disk->queue;
1052 unsigned int max_append_sectors = queue_max_zone_append_sectors(q);
1053 struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1054 struct page **pages = (struct page **)bv;
1055 ssize_t size, left;
1056 unsigned len, i;
1057 size_t offset;
1058 int ret = 0;
1059
1060 if (WARN_ON_ONCE(!max_append_sectors))
1061 return 0;
1062
1063 /*
1064 * Move page array up in the allocated memory for the bio vecs as far as
1065 * possible so that we can start filling biovecs from the beginning
1066 * without overwriting the temporary page array.
1067 */
1068 BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1069 pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1070
1071 size = iov_iter_get_pages(iter, pages, LONG_MAX, nr_pages, &offset);
1072 if (unlikely(size <= 0))
1073 return size ? size : -EFAULT;
1074
1075 for (left = size, i = 0; left > 0; left -= len, i++) {
1076 struct page *page = pages[i];
1077 bool same_page = false;
1078
1079 len = min_t(size_t, PAGE_SIZE - offset, left);
1080 if (bio_add_hw_page(q, bio, page, len, offset,
1081 max_append_sectors, &same_page) != len) {
1082 bio_put_pages(pages + i, left, offset);
1083 ret = -EINVAL;
1084 break;
1085 }
1086 if (same_page)
1087 put_page(page);
1088 offset = 0;
1089 }
1090
1091 iov_iter_advance(iter, size - left);
1092 return ret;
1093}
1094
1095/**
1096 * bio_iov_iter_get_pages - add user or kernel pages to a bio
1097 * @bio: bio to add pages to
1098 * @iter: iov iterator describing the region to be added
1099 *
1100 * This takes either an iterator pointing to user memory, or one pointing to
1101 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1102 * map them into the kernel. On IO completion, the caller should put those
1103 * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided
1104 * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs
1105 * to ensure the bvecs and pages stay referenced until the submitted I/O is
1106 * completed by a call to ->ki_complete() or returns with an error other than
1107 * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF
1108 * on IO completion. If it isn't, then pages should be released.
1109 *
1110 * The function tries, but does not guarantee, to pin as many pages as
1111 * fit into the bio, or are requested in @iter, whatever is smaller. If
1112 * MM encounters an error pinning the requested pages, it stops. Error
1113 * is returned only if 0 pages could be pinned.
1114 *
1115 * It's intended for direct IO, so doesn't do PSI tracking, the caller is
1116 * responsible for setting BIO_WORKINGSET if necessary.
1117 */
1118int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1119{
1120 int ret = 0;
1121
1122 if (iov_iter_is_bvec(iter)) {
1123 if (bio_op(bio) == REQ_OP_ZONE_APPEND)
1124 return bio_iov_bvec_set_append(bio, iter);
1125 return bio_iov_bvec_set(bio, iter);
1126 }
1127
1128 do {
1129 if (bio_op(bio) == REQ_OP_ZONE_APPEND)
1130 ret = __bio_iov_append_get_pages(bio, iter);
1131 else
1132 ret = __bio_iov_iter_get_pages(bio, iter);
1133 } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));
1134
1135 /* don't account direct I/O as memory stall */
1136 bio_clear_flag(bio, BIO_WORKINGSET);
1137 return bio->bi_vcnt ? 0 : ret;
1138}
1139EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
1140
1141static void submit_bio_wait_endio(struct bio *bio)
1142{
1143 complete(bio->bi_private);
1144}
1145
1146/**
1147 * submit_bio_wait - submit a bio, and wait until it completes
1148 * @bio: The &struct bio which describes the I/O
1149 *
1150 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1151 * bio_endio() on failure.
1152 *
1153 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1154 * result in bio reference to be consumed. The caller must drop the reference
1155 * on his own.
1156 */
1157int submit_bio_wait(struct bio *bio)
1158{
1159 DECLARE_COMPLETION_ONSTACK_MAP(done,
1160 bio->bi_bdev->bd_disk->lockdep_map);
1161 unsigned long hang_check;
1162
1163 bio->bi_private = &done;
1164 bio->bi_end_io = submit_bio_wait_endio;
1165 bio->bi_opf |= REQ_SYNC;
1166 submit_bio(bio);
1167
1168 /* Prevent hang_check timer from firing at us during very long I/O */
1169 hang_check = sysctl_hung_task_timeout_secs;
1170 if (hang_check)
1171 while (!wait_for_completion_io_timeout(&done,
1172 hang_check * (HZ/2)))
1173 ;
1174 else
1175 wait_for_completion_io(&done);
1176
1177 return blk_status_to_errno(bio->bi_status);
1178}
1179EXPORT_SYMBOL(submit_bio_wait);
1180
1181/**
1182 * bio_advance - increment/complete a bio by some number of bytes
1183 * @bio: bio to advance
1184 * @bytes: number of bytes to complete
1185 *
1186 * This updates bi_sector, bi_size and bi_idx; if the number of bytes to
1187 * complete doesn't align with a bvec boundary, then bv_len and bv_offset will
1188 * be updated on the last bvec as well.
1189 *
1190 * @bio will then represent the remaining, uncompleted portion of the io.
1191 */
1192void bio_advance(struct bio *bio, unsigned bytes)
1193{
1194 if (bio_integrity(bio))
1195 bio_integrity_advance(bio, bytes);
1196
1197 bio_crypt_advance(bio, bytes);
1198 bio_advance_iter(bio, &bio->bi_iter, bytes);
1199}
1200EXPORT_SYMBOL(bio_advance);
1201
1202void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1203 struct bio *src, struct bvec_iter *src_iter)
1204{
1205 struct bio_vec src_bv, dst_bv;
1206 void *src_p, *dst_p;
1207 unsigned bytes;
1208
1209 while (src_iter->bi_size && dst_iter->bi_size) {
1210 src_bv = bio_iter_iovec(src, *src_iter);
1211 dst_bv = bio_iter_iovec(dst, *dst_iter);
1212
1213 bytes = min(src_bv.bv_len, dst_bv.bv_len);
1214
1215 src_p = kmap_atomic(src_bv.bv_page);
1216 dst_p = kmap_atomic(dst_bv.bv_page);
1217
1218 memcpy(dst_p + dst_bv.bv_offset,
1219 src_p + src_bv.bv_offset,
1220 bytes);
1221
1222 kunmap_atomic(dst_p);
1223 kunmap_atomic(src_p);
1224
1225 flush_dcache_page(dst_bv.bv_page);
1226
1227 bio_advance_iter_single(src, src_iter, bytes);
1228 bio_advance_iter_single(dst, dst_iter, bytes);
1229 }
1230}
1231EXPORT_SYMBOL(bio_copy_data_iter);
1232
1233/**
1234 * bio_copy_data - copy contents of data buffers from one bio to another
1235 * @src: source bio
1236 * @dst: destination bio
1237 *
1238 * Stops when it reaches the end of either @src or @dst - that is, copies
1239 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1240 */
1241void bio_copy_data(struct bio *dst, struct bio *src)
1242{
1243 struct bvec_iter src_iter = src->bi_iter;
1244 struct bvec_iter dst_iter = dst->bi_iter;
1245
1246 bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1247}
1248EXPORT_SYMBOL(bio_copy_data);
1249
1250void bio_free_pages(struct bio *bio)
1251{
1252 struct bio_vec *bvec;
1253 struct bvec_iter_all iter_all;
1254
1255 bio_for_each_segment_all(bvec, bio, iter_all)
1256 __free_page(bvec->bv_page);
1257}
1258EXPORT_SYMBOL(bio_free_pages);
1259
1260/*
1261 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1262 * for performing direct-IO in BIOs.
1263 *
1264 * The problem is that we cannot run set_page_dirty() from interrupt context
1265 * because the required locks are not interrupt-safe. So what we can do is to
1266 * mark the pages dirty _before_ performing IO. And in interrupt context,
1267 * check that the pages are still dirty. If so, fine. If not, redirty them
1268 * in process context.
1269 *
1270 * We special-case compound pages here: normally this means reads into hugetlb
1271 * pages. The logic in here doesn't really work right for compound pages
1272 * because the VM does not uniformly chase down the head page in all cases.
1273 * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1274 * handle them at all. So we skip compound pages here at an early stage.
1275 *
1276 * Note that this code is very hard to test under normal circumstances because
1277 * direct-io pins the pages with get_user_pages(). This makes
1278 * is_page_cache_freeable return false, and the VM will not clean the pages.
1279 * But other code (eg, flusher threads) could clean the pages if they are mapped
1280 * pagecache.
1281 *
1282 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1283 * deferred bio dirtying paths.
1284 */
1285
1286/*
1287 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1288 */
1289void bio_set_pages_dirty(struct bio *bio)
1290{
1291 struct bio_vec *bvec;
1292 struct bvec_iter_all iter_all;
1293
1294 bio_for_each_segment_all(bvec, bio, iter_all) {
1295 if (!PageCompound(bvec->bv_page))
1296 set_page_dirty_lock(bvec->bv_page);
1297 }
1298}
1299
1300/*
1301 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1302 * If they are, then fine. If, however, some pages are clean then they must
1303 * have been written out during the direct-IO read. So we take another ref on
1304 * the BIO and re-dirty the pages in process context.
1305 *
1306 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1307 * here on. It will run one put_page() against each page and will run one
1308 * bio_put() against the BIO.
1309 */
1310
1311static void bio_dirty_fn(struct work_struct *work);
1312
1313static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1314static DEFINE_SPINLOCK(bio_dirty_lock);
1315static struct bio *bio_dirty_list;
1316
1317/*
1318 * This runs in process context
1319 */
1320static void bio_dirty_fn(struct work_struct *work)
1321{
1322 struct bio *bio, *next;
1323
1324 spin_lock_irq(&bio_dirty_lock);
1325 next = bio_dirty_list;
1326 bio_dirty_list = NULL;
1327 spin_unlock_irq(&bio_dirty_lock);
1328
1329 while ((bio = next) != NULL) {
1330 next = bio->bi_private;
1331
1332 bio_release_pages(bio, true);
1333 bio_put(bio);
1334 }
1335}
1336
1337void bio_check_pages_dirty(struct bio *bio)
1338{
1339 struct bio_vec *bvec;
1340 unsigned long flags;
1341 struct bvec_iter_all iter_all;
1342
1343 bio_for_each_segment_all(bvec, bio, iter_all) {
1344 if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page))
1345 goto defer;
1346 }
1347
1348 bio_release_pages(bio, false);
1349 bio_put(bio);
1350 return;
1351defer:
1352 spin_lock_irqsave(&bio_dirty_lock, flags);
1353 bio->bi_private = bio_dirty_list;
1354 bio_dirty_list = bio;
1355 spin_unlock_irqrestore(&bio_dirty_lock, flags);
1356 schedule_work(&bio_dirty_work);
1357}
1358
1359static inline bool bio_remaining_done(struct bio *bio)
1360{
1361 /*
1362 * If we're not chaining, then ->__bi_remaining is always 1 and
1363 * we always end io on the first invocation.
1364 */
1365 if (!bio_flagged(bio, BIO_CHAIN))
1366 return true;
1367
1368 BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1369
1370 if (atomic_dec_and_test(&bio->__bi_remaining)) {
1371 bio_clear_flag(bio, BIO_CHAIN);
1372 return true;
1373 }
1374
1375 return false;
1376}
1377
1378/**
1379 * bio_endio - end I/O on a bio
1380 * @bio: bio
1381 *
1382 * Description:
1383 * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1384 * way to end I/O on a bio. No one should call bi_end_io() directly on a
1385 * bio unless they own it and thus know that it has an end_io function.
1386 *
1387 * bio_endio() can be called several times on a bio that has been chained
1388 * using bio_chain(). The ->bi_end_io() function will only be called the
1389 * last time.
1390 **/
1391void bio_endio(struct bio *bio)
1392{
1393again:
1394 if (!bio_remaining_done(bio))
1395 return;
1396 if (!bio_integrity_endio(bio))
1397 return;
1398
1399 if (bio->bi_bdev && bio_flagged(bio, BIO_TRACKED))
1400 rq_qos_done_bio(bio->bi_bdev->bd_disk->queue, bio);
1401
1402 if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1403 trace_block_bio_complete(bio->bi_bdev->bd_disk->queue, bio);
1404 bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1405 }
1406
1407 /*
1408 * Need to have a real endio function for chained bios, otherwise
1409 * various corner cases will break (like stacking block devices that
1410 * save/restore bi_end_io) - however, we want to avoid unbounded
1411 * recursion and blowing the stack. Tail call optimization would
1412 * handle this, but compiling with frame pointers also disables
1413 * gcc's sibling call optimization.
1414 */
1415 if (bio->bi_end_io == bio_chain_endio) {
1416 bio = __bio_chain_endio(bio);
1417 goto again;
1418 }
1419
1420 blk_throtl_bio_endio(bio);
1421 /* release cgroup info */
1422 bio_uninit(bio);
1423 if (bio->bi_end_io)
1424 bio->bi_end_io(bio);
1425}
1426EXPORT_SYMBOL(bio_endio);
1427
1428/**
1429 * bio_split - split a bio
1430 * @bio: bio to split
1431 * @sectors: number of sectors to split from the front of @bio
1432 * @gfp: gfp mask
1433 * @bs: bio set to allocate from
1434 *
1435 * Allocates and returns a new bio which represents @sectors from the start of
1436 * @bio, and updates @bio to represent the remaining sectors.
1437 *
1438 * Unless this is a discard request the newly allocated bio will point
1439 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1440 * neither @bio nor @bs are freed before the split bio.
1441 */
1442struct bio *bio_split(struct bio *bio, int sectors,
1443 gfp_t gfp, struct bio_set *bs)
1444{
1445 struct bio *split;
1446
1447 BUG_ON(sectors <= 0);
1448 BUG_ON(sectors >= bio_sectors(bio));
1449
1450 /* Zone append commands cannot be split */
1451 if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))
1452 return NULL;
1453
1454 split = bio_clone_fast(bio, gfp, bs);
1455 if (!split)
1456 return NULL;
1457
1458 split->bi_iter.bi_size = sectors << 9;
1459
1460 if (bio_integrity(split))
1461 bio_integrity_trim(split);
1462
1463 bio_advance(bio, split->bi_iter.bi_size);
1464
1465 if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1466 bio_set_flag(split, BIO_TRACE_COMPLETION);
1467
1468 return split;
1469}
1470EXPORT_SYMBOL(bio_split);
1471
1472/**
1473 * bio_trim - trim a bio
1474 * @bio: bio to trim
1475 * @offset: number of sectors to trim from the front of @bio
1476 * @size: size we want to trim @bio to, in sectors
1477 */
1478void bio_trim(struct bio *bio, int offset, int size)
1479{
1480 /* 'bio' is a cloned bio which we need to trim to match
1481 * the given offset and size.
1482 */
1483
1484 size <<= 9;
1485 if (offset == 0 && size == bio->bi_iter.bi_size)
1486 return;
1487
1488 bio_advance(bio, offset << 9);
1489 bio->bi_iter.bi_size = size;
1490
1491 if (bio_integrity(bio))
1492 bio_integrity_trim(bio);
1493
1494}
1495EXPORT_SYMBOL_GPL(bio_trim);
1496
1497/*
1498 * create memory pools for biovec's in a bio_set.
1499 * use the global biovec slabs created for general use.
1500 */
1501int biovec_init_pool(mempool_t *pool, int pool_entries)
1502{
1503 struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1;
1504
1505 return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1506}
1507
1508/*
1509 * bioset_exit - exit a bioset initialized with bioset_init()
1510 *
1511 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1512 * kzalloc()).
1513 */
1514void bioset_exit(struct bio_set *bs)
1515{
1516 if (bs->rescue_workqueue)
1517 destroy_workqueue(bs->rescue_workqueue);
1518 bs->rescue_workqueue = NULL;
1519
1520 mempool_exit(&bs->bio_pool);
1521 mempool_exit(&bs->bvec_pool);
1522
1523 bioset_integrity_free(bs);
1524 if (bs->bio_slab)
1525 bio_put_slab(bs);
1526 bs->bio_slab = NULL;
1527}
1528EXPORT_SYMBOL(bioset_exit);
1529
1530/**
1531 * bioset_init - Initialize a bio_set
1532 * @bs: pool to initialize
1533 * @pool_size: Number of bio and bio_vecs to cache in the mempool
1534 * @front_pad: Number of bytes to allocate in front of the returned bio
1535 * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1536 * and %BIOSET_NEED_RESCUER
1537 *
1538 * Description:
1539 * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1540 * to ask for a number of bytes to be allocated in front of the bio.
1541 * Front pad allocation is useful for embedding the bio inside
1542 * another structure, to avoid allocating extra data to go with the bio.
1543 * Note that the bio must be embedded at the END of that structure always,
1544 * or things will break badly.
1545 * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1546 * for allocating iovecs. This pool is not needed e.g. for bio_clone_fast().
1547 * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used to
1548 * dispatch queued requests when the mempool runs out of space.
1549 *
1550 */
1551int bioset_init(struct bio_set *bs,
1552 unsigned int pool_size,
1553 unsigned int front_pad,
1554 int flags)
1555{
1556 bs->front_pad = front_pad;
1557 if (flags & BIOSET_NEED_BVECS)
1558 bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1559 else
1560 bs->back_pad = 0;
1561
1562 spin_lock_init(&bs->rescue_lock);
1563 bio_list_init(&bs->rescue_list);
1564 INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1565
1566 bs->bio_slab = bio_find_or_create_slab(bs);
1567 if (!bs->bio_slab)
1568 return -ENOMEM;
1569
1570 if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
1571 goto bad;
1572
1573 if ((flags & BIOSET_NEED_BVECS) &&
1574 biovec_init_pool(&bs->bvec_pool, pool_size))
1575 goto bad;
1576
1577 if (!(flags & BIOSET_NEED_RESCUER))
1578 return 0;
1579
1580 bs->rescue_workqueue = alloc_workqueue("bioset", WQ_MEM_RECLAIM, 0);
1581 if (!bs->rescue_workqueue)
1582 goto bad;
1583
1584 return 0;
1585bad:
1586 bioset_exit(bs);
1587 return -ENOMEM;
1588}
1589EXPORT_SYMBOL(bioset_init);
1590
1591/*
1592 * Initialize and setup a new bio_set, based on the settings from
1593 * another bio_set.
1594 */
1595int bioset_init_from_src(struct bio_set *bs, struct bio_set *src)
1596{
1597 int flags;
1598
1599 flags = 0;
1600 if (src->bvec_pool.min_nr)
1601 flags |= BIOSET_NEED_BVECS;
1602 if (src->rescue_workqueue)
1603 flags |= BIOSET_NEED_RESCUER;
1604
1605 return bioset_init(bs, src->bio_pool.min_nr, src->front_pad, flags);
1606}
1607EXPORT_SYMBOL(bioset_init_from_src);
1608
1609static int __init init_bio(void)
1610{
1611 int i;
1612
1613 bio_integrity_init();
1614
1615 for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) {
1616 struct biovec_slab *bvs = bvec_slabs + i;
1617
1618 bvs->slab = kmem_cache_create(bvs->name,
1619 bvs->nr_vecs * sizeof(struct bio_vec), 0,
1620 SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1621 }
1622
1623 if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS))
1624 panic("bio: can't allocate bios\n");
1625
1626 if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
1627 panic("bio: can't create integrity pool\n");
1628
1629 return 0;
1630}
1631subsys_initcall(init_bio);