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