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