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 : #define ALLOC_CACHE_THRESHOLD 16
29 : #define ALLOC_CACHE_MAX 256
30 :
31 : struct bio_alloc_cache {
32 : struct bio *free_list;
33 : struct bio *free_list_irq;
34 : unsigned int nr;
35 : unsigned int nr_irq;
36 : };
37 :
38 : static struct biovec_slab {
39 : int nr_vecs;
40 : char *name;
41 : struct kmem_cache *slab;
42 : } bvec_slabs[] __read_mostly = {
43 : { .nr_vecs = 16, .name = "biovec-16" },
44 : { .nr_vecs = 64, .name = "biovec-64" },
45 : { .nr_vecs = 128, .name = "biovec-128" },
46 : { .nr_vecs = BIO_MAX_VECS, .name = "biovec-max" },
47 : };
48 :
49 0 : static struct biovec_slab *biovec_slab(unsigned short nr_vecs)
50 : {
51 0 : switch (nr_vecs) {
52 : /* smaller bios use inline vecs */
53 : case 5 ... 16:
54 : return &bvec_slabs[0];
55 : case 17 ... 64:
56 0 : return &bvec_slabs[1];
57 : case 65 ... 128:
58 0 : return &bvec_slabs[2];
59 : case 129 ... BIO_MAX_VECS:
60 0 : return &bvec_slabs[3];
61 : default:
62 0 : BUG();
63 : return NULL;
64 : }
65 : }
66 :
67 : /*
68 : * fs_bio_set is the bio_set containing bio and iovec memory pools used by
69 : * IO code that does not need private memory pools.
70 : */
71 : struct bio_set fs_bio_set;
72 : EXPORT_SYMBOL(fs_bio_set);
73 :
74 : /*
75 : * Our slab pool management
76 : */
77 : struct bio_slab {
78 : struct kmem_cache *slab;
79 : unsigned int slab_ref;
80 : unsigned int slab_size;
81 : char name[8];
82 : };
83 : static DEFINE_MUTEX(bio_slab_lock);
84 : static DEFINE_XARRAY(bio_slabs);
85 :
86 2 : static struct bio_slab *create_bio_slab(unsigned int size)
87 : {
88 2 : struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL);
89 :
90 2 : if (!bslab)
91 : return NULL;
92 :
93 2 : snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size);
94 2 : bslab->slab = kmem_cache_create(bslab->name, size,
95 : ARCH_KMALLOC_MINALIGN,
96 : SLAB_HWCACHE_ALIGN | SLAB_TYPESAFE_BY_RCU, NULL);
97 2 : if (!bslab->slab)
98 : goto fail_alloc_slab;
99 :
100 2 : bslab->slab_ref = 1;
101 2 : bslab->slab_size = size;
102 :
103 4 : if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL)))
104 : return bslab;
105 :
106 0 : kmem_cache_destroy(bslab->slab);
107 :
108 : fail_alloc_slab:
109 0 : kfree(bslab);
110 0 : return NULL;
111 : }
112 :
113 : static inline unsigned int bs_bio_slab_size(struct bio_set *bs)
114 : {
115 2 : return bs->front_pad + sizeof(struct bio) + bs->back_pad;
116 : }
117 :
118 2 : static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs)
119 : {
120 4 : unsigned int size = bs_bio_slab_size(bs);
121 : struct bio_slab *bslab;
122 :
123 2 : mutex_lock(&bio_slab_lock);
124 2 : bslab = xa_load(&bio_slabs, size);
125 2 : if (bslab)
126 0 : bslab->slab_ref++;
127 : else
128 2 : bslab = create_bio_slab(size);
129 2 : mutex_unlock(&bio_slab_lock);
130 :
131 2 : if (bslab)
132 2 : return bslab->slab;
133 : return NULL;
134 : }
135 :
136 0 : static void bio_put_slab(struct bio_set *bs)
137 : {
138 0 : struct bio_slab *bslab = NULL;
139 0 : unsigned int slab_size = bs_bio_slab_size(bs);
140 :
141 0 : mutex_lock(&bio_slab_lock);
142 :
143 0 : bslab = xa_load(&bio_slabs, slab_size);
144 0 : if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
145 : goto out;
146 :
147 0 : WARN_ON_ONCE(bslab->slab != bs->bio_slab);
148 :
149 0 : WARN_ON(!bslab->slab_ref);
150 :
151 0 : if (--bslab->slab_ref)
152 : goto out;
153 :
154 0 : xa_erase(&bio_slabs, slab_size);
155 :
156 0 : kmem_cache_destroy(bslab->slab);
157 0 : kfree(bslab);
158 :
159 : out:
160 0 : mutex_unlock(&bio_slab_lock);
161 0 : }
162 :
163 0 : void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs)
164 : {
165 0 : BUG_ON(nr_vecs > BIO_MAX_VECS);
166 :
167 0 : if (nr_vecs == BIO_MAX_VECS)
168 0 : mempool_free(bv, pool);
169 0 : else if (nr_vecs > BIO_INLINE_VECS)
170 0 : kmem_cache_free(biovec_slab(nr_vecs)->slab, bv);
171 0 : }
172 :
173 : /*
174 : * Make the first allocation restricted and don't dump info on allocation
175 : * failures, since we'll fall back to the mempool in case of failure.
176 : */
177 : static inline gfp_t bvec_alloc_gfp(gfp_t gfp)
178 : {
179 : return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) |
180 0 : __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
181 : }
182 :
183 0 : struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
184 : gfp_t gfp_mask)
185 : {
186 0 : struct biovec_slab *bvs = biovec_slab(*nr_vecs);
187 :
188 0 : if (WARN_ON_ONCE(!bvs))
189 : return NULL;
190 :
191 : /*
192 : * Upgrade the nr_vecs request to take full advantage of the allocation.
193 : * We also rely on this in the bvec_free path.
194 : */
195 0 : *nr_vecs = bvs->nr_vecs;
196 :
197 : /*
198 : * Try a slab allocation first for all smaller allocations. If that
199 : * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool.
200 : * The mempool is sized to handle up to BIO_MAX_VECS entries.
201 : */
202 0 : if (*nr_vecs < BIO_MAX_VECS) {
203 : struct bio_vec *bvl;
204 :
205 0 : bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask));
206 0 : if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM))
207 : return bvl;
208 0 : *nr_vecs = BIO_MAX_VECS;
209 : }
210 :
211 0 : return mempool_alloc(pool, gfp_mask);
212 : }
213 :
214 0 : void bio_uninit(struct bio *bio)
215 : {
216 : #ifdef CONFIG_BLK_CGROUP
217 : if (bio->bi_blkg) {
218 : blkg_put(bio->bi_blkg);
219 : bio->bi_blkg = NULL;
220 : }
221 : #endif
222 0 : if (bio_integrity(bio))
223 : bio_integrity_free(bio);
224 :
225 0 : bio_crypt_free_ctx(bio);
226 0 : }
227 : EXPORT_SYMBOL(bio_uninit);
228 :
229 0 : static void bio_free(struct bio *bio)
230 : {
231 0 : struct bio_set *bs = bio->bi_pool;
232 0 : void *p = bio;
233 :
234 0 : WARN_ON_ONCE(!bs);
235 :
236 0 : bio_uninit(bio);
237 0 : bvec_free(&bs->bvec_pool, bio->bi_io_vec, bio->bi_max_vecs);
238 0 : mempool_free(p - bs->front_pad, &bs->bio_pool);
239 0 : }
240 :
241 : /*
242 : * Users of this function have their own bio allocation. Subsequently,
243 : * they must remember to pair any call to bio_init() with bio_uninit()
244 : * when IO has completed, or when the bio is released.
245 : */
246 0 : void bio_init(struct bio *bio, struct block_device *bdev, struct bio_vec *table,
247 : unsigned short max_vecs, blk_opf_t opf)
248 : {
249 0 : bio->bi_next = NULL;
250 0 : bio->bi_bdev = bdev;
251 0 : bio->bi_opf = opf;
252 0 : bio->bi_flags = 0;
253 0 : bio->bi_ioprio = 0;
254 0 : bio->bi_status = 0;
255 0 : bio->bi_iter.bi_sector = 0;
256 0 : bio->bi_iter.bi_size = 0;
257 0 : bio->bi_iter.bi_idx = 0;
258 0 : bio->bi_iter.bi_bvec_done = 0;
259 0 : bio->bi_end_io = NULL;
260 0 : bio->bi_private = NULL;
261 : #ifdef CONFIG_BLK_CGROUP
262 : bio->bi_blkg = NULL;
263 : bio->bi_issue.value = 0;
264 : if (bdev)
265 : bio_associate_blkg(bio);
266 : #ifdef CONFIG_BLK_CGROUP_IOCOST
267 : bio->bi_iocost_cost = 0;
268 : #endif
269 : #endif
270 : #ifdef CONFIG_BLK_INLINE_ENCRYPTION
271 : bio->bi_crypt_context = NULL;
272 : #endif
273 : #ifdef CONFIG_BLK_DEV_INTEGRITY
274 : bio->bi_integrity = NULL;
275 : #endif
276 0 : bio->bi_vcnt = 0;
277 :
278 0 : atomic_set(&bio->__bi_remaining, 1);
279 0 : atomic_set(&bio->__bi_cnt, 1);
280 0 : bio->bi_cookie = BLK_QC_T_NONE;
281 :
282 0 : bio->bi_max_vecs = max_vecs;
283 0 : bio->bi_io_vec = table;
284 0 : bio->bi_pool = NULL;
285 0 : }
286 : EXPORT_SYMBOL(bio_init);
287 :
288 : /**
289 : * bio_reset - reinitialize a bio
290 : * @bio: bio to reset
291 : * @bdev: block device to use the bio for
292 : * @opf: operation and flags for bio
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 : */
300 0 : void bio_reset(struct bio *bio, struct block_device *bdev, blk_opf_t opf)
301 : {
302 0 : bio_uninit(bio);
303 0 : memset(bio, 0, BIO_RESET_BYTES);
304 0 : atomic_set(&bio->__bi_remaining, 1);
305 0 : bio->bi_bdev = bdev;
306 : if (bio->bi_bdev)
307 : bio_associate_blkg(bio);
308 0 : bio->bi_opf = opf;
309 0 : }
310 : EXPORT_SYMBOL(bio_reset);
311 :
312 : static struct bio *__bio_chain_endio(struct bio *bio)
313 : {
314 0 : struct bio *parent = bio->bi_private;
315 :
316 0 : if (bio->bi_status && !parent->bi_status)
317 0 : parent->bi_status = bio->bi_status;
318 0 : bio_put(bio);
319 : return parent;
320 : }
321 :
322 0 : static void bio_chain_endio(struct bio *bio)
323 : {
324 0 : bio_endio(__bio_chain_endio(bio));
325 0 : }
326 :
327 : /**
328 : * bio_chain - chain bio completions
329 : * @bio: the target bio
330 : * @parent: the parent bio of @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 : */
338 0 : void bio_chain(struct bio *bio, struct bio *parent)
339 : {
340 0 : BUG_ON(bio->bi_private || bio->bi_end_io);
341 :
342 0 : bio->bi_private = parent;
343 0 : bio->bi_end_io = bio_chain_endio;
344 0 : bio_inc_remaining(parent);
345 0 : }
346 : EXPORT_SYMBOL(bio_chain);
347 :
348 0 : struct bio *blk_next_bio(struct bio *bio, struct block_device *bdev,
349 : unsigned int nr_pages, blk_opf_t opf, gfp_t gfp)
350 : {
351 0 : struct bio *new = bio_alloc(bdev, nr_pages, opf, gfp);
352 :
353 0 : if (bio) {
354 0 : bio_chain(bio, new);
355 0 : submit_bio(bio);
356 : }
357 :
358 0 : return new;
359 : }
360 : EXPORT_SYMBOL_GPL(blk_next_bio);
361 :
362 0 : static void bio_alloc_rescue(struct work_struct *work)
363 : {
364 0 : struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
365 : struct bio *bio;
366 :
367 : while (1) {
368 0 : spin_lock(&bs->rescue_lock);
369 0 : bio = bio_list_pop(&bs->rescue_list);
370 0 : spin_unlock(&bs->rescue_lock);
371 :
372 0 : if (!bio)
373 : break;
374 :
375 0 : submit_bio_noacct(bio);
376 : }
377 0 : }
378 :
379 0 : static void punt_bios_to_rescuer(struct bio_set *bs)
380 : {
381 : struct bio_list punt, nopunt;
382 : struct bio *bio;
383 :
384 0 : if (WARN_ON_ONCE(!bs->rescue_workqueue))
385 0 : return;
386 : /*
387 : * In order to guarantee forward progress we must punt only bios that
388 : * were allocated from this bio_set; otherwise, if there was a bio on
389 : * there for a stacking driver higher up in the stack, processing it
390 : * could require allocating bios from this bio_set, and doing that from
391 : * our own rescuer would be bad.
392 : *
393 : * Since bio lists are singly linked, pop them all instead of trying to
394 : * remove from the middle of the list:
395 : */
396 :
397 0 : bio_list_init(&punt);
398 : bio_list_init(&nopunt);
399 :
400 0 : while ((bio = bio_list_pop(¤t->bio_list[0])))
401 0 : bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
402 0 : current->bio_list[0] = nopunt;
403 :
404 : bio_list_init(&nopunt);
405 0 : while ((bio = bio_list_pop(¤t->bio_list[1])))
406 0 : bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
407 0 : current->bio_list[1] = nopunt;
408 :
409 0 : spin_lock(&bs->rescue_lock);
410 0 : bio_list_merge(&bs->rescue_list, &punt);
411 0 : spin_unlock(&bs->rescue_lock);
412 :
413 0 : queue_work(bs->rescue_workqueue, &bs->rescue_work);
414 : }
415 :
416 0 : static void bio_alloc_irq_cache_splice(struct bio_alloc_cache *cache)
417 : {
418 : unsigned long flags;
419 :
420 : /* cache->free_list must be empty */
421 0 : if (WARN_ON_ONCE(cache->free_list))
422 : return;
423 :
424 0 : local_irq_save(flags);
425 0 : cache->free_list = cache->free_list_irq;
426 0 : cache->free_list_irq = NULL;
427 0 : cache->nr += cache->nr_irq;
428 0 : cache->nr_irq = 0;
429 0 : local_irq_restore(flags);
430 : }
431 :
432 0 : static struct bio *bio_alloc_percpu_cache(struct block_device *bdev,
433 : unsigned short nr_vecs, blk_opf_t opf, gfp_t gfp,
434 : struct bio_set *bs)
435 : {
436 : struct bio_alloc_cache *cache;
437 : struct bio *bio;
438 :
439 0 : cache = per_cpu_ptr(bs->cache, get_cpu());
440 0 : if (!cache->free_list) {
441 0 : if (READ_ONCE(cache->nr_irq) >= ALLOC_CACHE_THRESHOLD)
442 0 : bio_alloc_irq_cache_splice(cache);
443 0 : if (!cache->free_list) {
444 0 : put_cpu();
445 0 : return NULL;
446 : }
447 : }
448 0 : bio = cache->free_list;
449 0 : cache->free_list = bio->bi_next;
450 0 : cache->nr--;
451 0 : put_cpu();
452 :
453 0 : bio_init(bio, bdev, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs, opf);
454 0 : bio->bi_pool = bs;
455 0 : return bio;
456 : }
457 :
458 : /**
459 : * bio_alloc_bioset - allocate a bio for I/O
460 : * @bdev: block device to allocate the bio for (can be %NULL)
461 : * @nr_vecs: number of bvecs to pre-allocate
462 : * @opf: operation and flags for bio
463 : * @gfp_mask: the GFP_* mask given to the slab allocator
464 : * @bs: the bio_set to allocate from.
465 : *
466 : * Allocate a bio from the mempools in @bs.
467 : *
468 : * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to
469 : * allocate a bio. This is due to the mempool guarantees. To make this work,
470 : * callers must never allocate more than 1 bio at a time from the general pool.
471 : * Callers that need to allocate more than 1 bio must always submit the
472 : * previously allocated bio for IO before attempting to allocate a new one.
473 : * Failure to do so can cause deadlocks under memory pressure.
474 : *
475 : * Note that when running under submit_bio_noacct() (i.e. any block driver),
476 : * bios are not submitted until after you return - see the code in
477 : * submit_bio_noacct() that converts recursion into iteration, to prevent
478 : * stack overflows.
479 : *
480 : * This would normally mean allocating multiple bios under submit_bio_noacct()
481 : * would be susceptible to deadlocks, but we have
482 : * deadlock avoidance code that resubmits any blocked bios from a rescuer
483 : * thread.
484 : *
485 : * However, we do not guarantee forward progress for allocations from other
486 : * mempools. Doing multiple allocations from the same mempool under
487 : * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad
488 : * for per bio allocations.
489 : *
490 : * Returns: Pointer to new bio on success, NULL on failure.
491 : */
492 0 : struct bio *bio_alloc_bioset(struct block_device *bdev, unsigned short nr_vecs,
493 : blk_opf_t opf, gfp_t gfp_mask,
494 : struct bio_set *bs)
495 : {
496 0 : gfp_t saved_gfp = gfp_mask;
497 : struct bio *bio;
498 : void *p;
499 :
500 : /* should not use nobvec bioset for nr_vecs > 0 */
501 0 : if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_vecs > 0))
502 : return NULL;
503 :
504 0 : if (opf & REQ_ALLOC_CACHE) {
505 0 : if (bs->cache && nr_vecs <= BIO_INLINE_VECS) {
506 0 : bio = bio_alloc_percpu_cache(bdev, nr_vecs, opf,
507 : gfp_mask, bs);
508 0 : if (bio)
509 : return bio;
510 : /*
511 : * No cached bio available, bio returned below marked with
512 : * REQ_ALLOC_CACHE to particpate in per-cpu alloc cache.
513 : */
514 : } else {
515 0 : opf &= ~REQ_ALLOC_CACHE;
516 : }
517 : }
518 :
519 : /*
520 : * submit_bio_noacct() converts recursion to iteration; this means if
521 : * we're running beneath it, any bios we allocate and submit will not be
522 : * submitted (and thus freed) until after we return.
523 : *
524 : * This exposes us to a potential deadlock if we allocate multiple bios
525 : * from the same bio_set() while running underneath submit_bio_noacct().
526 : * If we were to allocate multiple bios (say a stacking block driver
527 : * that was splitting bios), we would deadlock if we exhausted the
528 : * mempool's reserve.
529 : *
530 : * We solve this, and guarantee forward progress, with a rescuer
531 : * workqueue per bio_set. If we go to allocate and there are bios on
532 : * current->bio_list, we first try the allocation without
533 : * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be
534 : * blocking to the rescuer workqueue before we retry with the original
535 : * gfp_flags.
536 : */
537 0 : if (current->bio_list &&
538 0 : (!bio_list_empty(¤t->bio_list[0]) ||
539 0 : !bio_list_empty(¤t->bio_list[1])) &&
540 0 : bs->rescue_workqueue)
541 0 : gfp_mask &= ~__GFP_DIRECT_RECLAIM;
542 :
543 0 : p = mempool_alloc(&bs->bio_pool, gfp_mask);
544 0 : if (!p && gfp_mask != saved_gfp) {
545 0 : punt_bios_to_rescuer(bs);
546 0 : gfp_mask = saved_gfp;
547 0 : p = mempool_alloc(&bs->bio_pool, gfp_mask);
548 : }
549 0 : if (unlikely(!p))
550 : return NULL;
551 0 : if (!mempool_is_saturated(&bs->bio_pool))
552 0 : opf &= ~REQ_ALLOC_CACHE;
553 :
554 0 : bio = p + bs->front_pad;
555 0 : if (nr_vecs > BIO_INLINE_VECS) {
556 0 : struct bio_vec *bvl = NULL;
557 :
558 0 : bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);
559 0 : if (!bvl && gfp_mask != saved_gfp) {
560 0 : punt_bios_to_rescuer(bs);
561 0 : gfp_mask = saved_gfp;
562 0 : bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);
563 : }
564 0 : if (unlikely(!bvl))
565 : goto err_free;
566 :
567 0 : bio_init(bio, bdev, bvl, nr_vecs, opf);
568 0 : } else if (nr_vecs) {
569 0 : bio_init(bio, bdev, bio->bi_inline_vecs, BIO_INLINE_VECS, opf);
570 : } else {
571 : bio_init(bio, bdev, NULL, 0, opf);
572 : }
573 :
574 0 : bio->bi_pool = bs;
575 0 : return bio;
576 :
577 : err_free:
578 0 : mempool_free(p, &bs->bio_pool);
579 0 : return NULL;
580 : }
581 : EXPORT_SYMBOL(bio_alloc_bioset);
582 :
583 : /**
584 : * bio_kmalloc - kmalloc a bio
585 : * @nr_vecs: number of bio_vecs to allocate
586 : * @gfp_mask: the GFP_* mask given to the slab allocator
587 : *
588 : * Use kmalloc to allocate a bio (including bvecs). The bio must be initialized
589 : * using bio_init() before use. To free a bio returned from this function use
590 : * kfree() after calling bio_uninit(). A bio returned from this function can
591 : * be reused by calling bio_uninit() before calling bio_init() again.
592 : *
593 : * Note that unlike bio_alloc() or bio_alloc_bioset() allocations from this
594 : * function are not backed by a mempool can fail. Do not use this function
595 : * for allocations in the file system I/O path.
596 : *
597 : * Returns: Pointer to new bio on success, NULL on failure.
598 : */
599 0 : struct bio *bio_kmalloc(unsigned short nr_vecs, gfp_t gfp_mask)
600 : {
601 : struct bio *bio;
602 :
603 0 : if (nr_vecs > UIO_MAXIOV)
604 : return NULL;
605 0 : return kmalloc(struct_size(bio, bi_inline_vecs, nr_vecs), gfp_mask);
606 : }
607 : EXPORT_SYMBOL(bio_kmalloc);
608 :
609 0 : void zero_fill_bio(struct bio *bio)
610 : {
611 : struct bio_vec bv;
612 : struct bvec_iter iter;
613 :
614 0 : bio_for_each_segment(bv, bio, iter)
615 0 : memzero_bvec(&bv);
616 0 : }
617 : EXPORT_SYMBOL(zero_fill_bio);
618 :
619 : /**
620 : * bio_truncate - truncate the bio to small size of @new_size
621 : * @bio: the bio to be truncated
622 : * @new_size: new size for truncating the bio
623 : *
624 : * Description:
625 : * Truncate the bio to new size of @new_size. If bio_op(bio) is
626 : * REQ_OP_READ, zero the truncated part. This function should only
627 : * be used for handling corner cases, such as bio eod.
628 : */
629 0 : static void bio_truncate(struct bio *bio, unsigned new_size)
630 : {
631 : struct bio_vec bv;
632 : struct bvec_iter iter;
633 0 : unsigned int done = 0;
634 0 : bool truncated = false;
635 :
636 0 : if (new_size >= bio->bi_iter.bi_size)
637 0 : return;
638 :
639 0 : if (bio_op(bio) != REQ_OP_READ)
640 : goto exit;
641 :
642 0 : bio_for_each_segment(bv, bio, iter) {
643 0 : if (done + bv.bv_len > new_size) {
644 : unsigned offset;
645 :
646 0 : if (!truncated)
647 0 : offset = new_size - done;
648 : else
649 : offset = 0;
650 0 : zero_user(bv.bv_page, bv.bv_offset + offset,
651 : bv.bv_len - offset);
652 0 : truncated = true;
653 : }
654 0 : done += bv.bv_len;
655 : }
656 :
657 : exit:
658 : /*
659 : * Don't touch bvec table here and make it really immutable, since
660 : * fs bio user has to retrieve all pages via bio_for_each_segment_all
661 : * in its .end_bio() callback.
662 : *
663 : * It is enough to truncate bio by updating .bi_size since we can make
664 : * correct bvec with the updated .bi_size for drivers.
665 : */
666 0 : bio->bi_iter.bi_size = new_size;
667 : }
668 :
669 : /**
670 : * guard_bio_eod - truncate a BIO to fit the block device
671 : * @bio: bio to truncate
672 : *
673 : * This allows us to do IO even on the odd last sectors of a device, even if the
674 : * block size is some multiple of the physical sector size.
675 : *
676 : * We'll just truncate the bio to the size of the device, and clear the end of
677 : * the buffer head manually. Truly out-of-range accesses will turn into actual
678 : * I/O errors, this only handles the "we need to be able to do I/O at the final
679 : * sector" case.
680 : */
681 0 : void guard_bio_eod(struct bio *bio)
682 : {
683 0 : sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
684 :
685 0 : if (!maxsector)
686 : return;
687 :
688 : /*
689 : * If the *whole* IO is past the end of the device,
690 : * let it through, and the IO layer will turn it into
691 : * an EIO.
692 : */
693 0 : if (unlikely(bio->bi_iter.bi_sector >= maxsector))
694 : return;
695 :
696 0 : maxsector -= bio->bi_iter.bi_sector;
697 0 : if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
698 : return;
699 :
700 0 : bio_truncate(bio, maxsector << 9);
701 : }
702 :
703 : static int __bio_alloc_cache_prune(struct bio_alloc_cache *cache,
704 : unsigned int nr)
705 : {
706 : unsigned int i = 0;
707 : struct bio *bio;
708 :
709 0 : while ((bio = cache->free_list) != NULL) {
710 0 : cache->free_list = bio->bi_next;
711 0 : cache->nr--;
712 0 : bio_free(bio);
713 0 : if (++i == nr)
714 : break;
715 : }
716 : return i;
717 : }
718 :
719 0 : static void bio_alloc_cache_prune(struct bio_alloc_cache *cache,
720 : unsigned int nr)
721 : {
722 0 : nr -= __bio_alloc_cache_prune(cache, nr);
723 0 : if (!READ_ONCE(cache->free_list)) {
724 0 : bio_alloc_irq_cache_splice(cache);
725 0 : __bio_alloc_cache_prune(cache, nr);
726 : }
727 0 : }
728 :
729 0 : static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node)
730 : {
731 : struct bio_set *bs;
732 :
733 0 : bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead);
734 0 : if (bs->cache) {
735 0 : struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu);
736 :
737 0 : bio_alloc_cache_prune(cache, -1U);
738 : }
739 0 : return 0;
740 : }
741 :
742 0 : static void bio_alloc_cache_destroy(struct bio_set *bs)
743 : {
744 : int cpu;
745 :
746 0 : if (!bs->cache)
747 : return;
748 :
749 0 : cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
750 0 : for_each_possible_cpu(cpu) {
751 : struct bio_alloc_cache *cache;
752 :
753 0 : cache = per_cpu_ptr(bs->cache, cpu);
754 0 : bio_alloc_cache_prune(cache, -1U);
755 : }
756 0 : free_percpu(bs->cache);
757 0 : bs->cache = NULL;
758 : }
759 :
760 0 : static inline void bio_put_percpu_cache(struct bio *bio)
761 : {
762 : struct bio_alloc_cache *cache;
763 :
764 0 : cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu());
765 0 : if (READ_ONCE(cache->nr_irq) + cache->nr > ALLOC_CACHE_MAX) {
766 0 : put_cpu();
767 0 : bio_free(bio);
768 0 : return;
769 : }
770 :
771 0 : bio_uninit(bio);
772 :
773 0 : if ((bio->bi_opf & REQ_POLLED) && !WARN_ON_ONCE(in_interrupt())) {
774 0 : bio->bi_next = cache->free_list;
775 0 : bio->bi_bdev = NULL;
776 0 : cache->free_list = bio;
777 0 : cache->nr++;
778 : } else {
779 : unsigned long flags;
780 :
781 0 : local_irq_save(flags);
782 0 : bio->bi_next = cache->free_list_irq;
783 0 : cache->free_list_irq = bio;
784 0 : cache->nr_irq++;
785 0 : local_irq_restore(flags);
786 : }
787 0 : 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 : **/
798 0 : void bio_put(struct bio *bio)
799 : {
800 0 : if (unlikely(bio_flagged(bio, BIO_REFFED))) {
801 0 : BUG_ON(!atomic_read(&bio->__bi_cnt));
802 0 : if (!atomic_dec_and_test(&bio->__bi_cnt))
803 : return;
804 : }
805 0 : if (bio->bi_opf & REQ_ALLOC_CACHE)
806 0 : bio_put_percpu_cache(bio);
807 : else
808 0 : bio_free(bio);
809 : }
810 : EXPORT_SYMBOL(bio_put);
811 :
812 : static int __bio_clone(struct bio *bio, struct bio *bio_src, gfp_t gfp)
813 : {
814 0 : bio_set_flag(bio, BIO_CLONED);
815 0 : bio->bi_ioprio = bio_src->bi_ioprio;
816 0 : bio->bi_iter = bio_src->bi_iter;
817 :
818 0 : if (bio->bi_bdev) {
819 0 : if (bio->bi_bdev == bio_src->bi_bdev &&
820 0 : bio_flagged(bio_src, BIO_REMAPPED))
821 : bio_set_flag(bio, BIO_REMAPPED);
822 : bio_clone_blkg_association(bio, bio_src);
823 : }
824 :
825 0 : if (bio_crypt_clone(bio, bio_src, gfp) < 0)
826 : return -ENOMEM;
827 0 : 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 : */
845 0 : struct 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 0 : bio = bio_alloc_bioset(bdev, 0, bio_src->bi_opf, gfp, bs);
851 0 : if (!bio)
852 : return NULL;
853 :
854 0 : if (__bio_clone(bio, bio_src, gfp) < 0) {
855 : bio_put(bio);
856 : return NULL;
857 : }
858 0 : bio->bi_io_vec = bio_src->bi_io_vec;
859 :
860 0 : return bio;
861 : }
862 : EXPORT_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 : */
876 0 : int bio_init_clone(struct block_device *bdev, struct bio *bio,
877 : struct bio *bio_src, gfp_t gfp)
878 : {
879 : int ret;
880 :
881 0 : bio_init(bio, bdev, bio_src->bi_io_vec, 0, bio_src->bi_opf);
882 0 : ret = __bio_clone(bio, bio_src, gfp);
883 : if (ret)
884 : bio_uninit(bio);
885 0 : return ret;
886 : }
887 : EXPORT_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 : */
897 : static inline bool bio_full(struct bio *bio, unsigned len)
898 : {
899 0 : if (bio->bi_vcnt >= bio->bi_max_vecs)
900 : return true;
901 0 : if (bio->bi_iter.bi_size > UINT_MAX - len)
902 : return true;
903 : return false;
904 : }
905 :
906 0 : static 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 0 : size_t bv_end = bv->bv_offset + bv->bv_len;
911 0 : phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1;
912 0 : phys_addr_t page_addr = page_to_phys(page);
913 :
914 0 : 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 0 : if (!zone_device_pages_have_same_pgmap(bv->bv_page, page))
919 : return false;
920 :
921 0 : *same_page = ((vec_end_addr & PAGE_MASK) == page_addr);
922 0 : if (*same_page)
923 : return true;
924 : else if (IS_ENABLED(CONFIG_KMSAN))
925 : return false;
926 0 : 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 : */
945 0 : static bool __bio_try_merge_page(struct bio *bio, struct page *page,
946 : unsigned int len, unsigned int off, bool *same_page)
947 : {
948 0 : if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
949 : return false;
950 :
951 0 : if (bio->bi_vcnt > 0) {
952 0 : struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
953 :
954 0 : if (page_is_mergeable(bv, page, len, off, same_page)) {
955 0 : if (bio->bi_iter.bi_size > UINT_MAX - len) {
956 0 : *same_page = false;
957 0 : return false;
958 : }
959 0 : bv->bv_len += len;
960 0 : bio->bi_iter.bi_size += len;
961 0 : 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 : */
972 0 : static 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 0 : struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
977 0 : unsigned long mask = queue_segment_boundary(q);
978 0 : phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset;
979 0 : phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;
980 :
981 0 : if ((addr1 | mask) != (addr2 | mask))
982 : return false;
983 0 : if (bv->bv_len + len > queue_max_segment_size(q))
984 : return false;
985 0 : 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 : */
1001 0 : int 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 0 : if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
1008 : return 0;
1009 :
1010 0 : if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors)
1011 : return 0;
1012 :
1013 0 : if (bio->bi_vcnt > 0) {
1014 0 : if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page))
1015 0 : 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 0 : bvec = &bio->bi_io_vec[bio->bi_vcnt - 1];
1022 0 : if (bvec_gap_to_prev(&q->limits, bvec, offset))
1023 : return 0;
1024 : }
1025 :
1026 0 : if (bio_full(bio, len))
1027 : return 0;
1028 :
1029 0 : if (bio->bi_vcnt >= queue_max_segments(q))
1030 : return 0;
1031 :
1032 0 : bvec_set_page(&bio->bi_io_vec[bio->bi_vcnt], page, len, offset);
1033 0 : bio->bi_vcnt++;
1034 0 : bio->bi_iter.bi_size += len;
1035 0 : return len;
1036 : }
1037 :
1038 : /**
1039 : * bio_add_pc_page - attempt to add page to passthrough bio
1040 : * @q: the target queue
1041 : * @bio: destination bio
1042 : * @page: page to add
1043 : * @len: vec entry length
1044 : * @offset: vec entry offset
1045 : *
1046 : * Attempt to add a page to the bio_vec maplist. This can fail for a
1047 : * number of reasons, such as the bio being full or target block device
1048 : * limitations. The target block device must allow bio's up to PAGE_SIZE,
1049 : * so it is always possible to add a single page to an empty bio.
1050 : *
1051 : * This should only be used by passthrough bios.
1052 : */
1053 0 : int bio_add_pc_page(struct request_queue *q, struct bio *bio,
1054 : struct page *page, unsigned int len, unsigned int offset)
1055 : {
1056 0 : bool same_page = false;
1057 0 : return bio_add_hw_page(q, bio, page, len, offset,
1058 : queue_max_hw_sectors(q), &same_page);
1059 : }
1060 : EXPORT_SYMBOL(bio_add_pc_page);
1061 :
1062 : /**
1063 : * bio_add_zone_append_page - attempt to add page to zone-append bio
1064 : * @bio: destination bio
1065 : * @page: page to add
1066 : * @len: vec entry length
1067 : * @offset: vec entry offset
1068 : *
1069 : * Attempt to add a page to the bio_vec maplist of a bio that will be submitted
1070 : * for a zone-append request. This can fail for a number of reasons, such as the
1071 : * bio being full or the target block device is not a zoned block device or
1072 : * other limitations of the target block device. The target block device must
1073 : * allow bio's up to PAGE_SIZE, so it is always possible to add a single page
1074 : * to an empty bio.
1075 : *
1076 : * Returns: number of bytes added to the bio, or 0 in case of a failure.
1077 : */
1078 0 : int bio_add_zone_append_page(struct bio *bio, struct page *page,
1079 : unsigned int len, unsigned int offset)
1080 : {
1081 0 : struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1082 0 : bool same_page = false;
1083 :
1084 0 : if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND))
1085 : return 0;
1086 :
1087 0 : if (WARN_ON_ONCE(!bdev_is_zoned(bio->bi_bdev)))
1088 : return 0;
1089 :
1090 : return bio_add_hw_page(q, bio, page, len, offset,
1091 : queue_max_zone_append_sectors(q), &same_page);
1092 : }
1093 : EXPORT_SYMBOL_GPL(bio_add_zone_append_page);
1094 :
1095 : /**
1096 : * __bio_add_page - add page(s) to a bio in a new segment
1097 : * @bio: destination bio
1098 : * @page: start page to add
1099 : * @len: length of the data to add, may cross pages
1100 : * @off: offset of the data relative to @page, may cross pages
1101 : *
1102 : * Add the data at @page + @off to @bio as a new bvec. The caller must ensure
1103 : * that @bio has space for another bvec.
1104 : */
1105 0 : void __bio_add_page(struct bio *bio, struct page *page,
1106 : unsigned int len, unsigned int off)
1107 : {
1108 0 : WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
1109 0 : WARN_ON_ONCE(bio_full(bio, len));
1110 :
1111 0 : bvec_set_page(&bio->bi_io_vec[bio->bi_vcnt], page, len, off);
1112 0 : bio->bi_iter.bi_size += len;
1113 0 : bio->bi_vcnt++;
1114 0 : }
1115 : EXPORT_SYMBOL_GPL(__bio_add_page);
1116 :
1117 : /**
1118 : * bio_add_page - attempt to add page(s) to bio
1119 : * @bio: destination bio
1120 : * @page: start page to add
1121 : * @len: vec entry length, may cross pages
1122 : * @offset: vec entry offset relative to @page, may cross pages
1123 : *
1124 : * Attempt to add page(s) to the bio_vec maplist. This will only fail
1125 : * if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
1126 : */
1127 0 : int bio_add_page(struct bio *bio, struct page *page,
1128 : unsigned int len, unsigned int offset)
1129 : {
1130 0 : bool same_page = false;
1131 :
1132 0 : if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
1133 0 : if (bio_full(bio, len))
1134 : return 0;
1135 0 : __bio_add_page(bio, page, len, offset);
1136 : }
1137 0 : return len;
1138 : }
1139 : EXPORT_SYMBOL(bio_add_page);
1140 :
1141 : /**
1142 : * bio_add_folio - Attempt to add part of a folio to a bio.
1143 : * @bio: BIO to add to.
1144 : * @folio: Folio to add.
1145 : * @len: How many bytes from the folio to add.
1146 : * @off: First byte in this folio to add.
1147 : *
1148 : * Filesystems that use folios can call this function instead of calling
1149 : * bio_add_page() for each page in the folio. If @off is bigger than
1150 : * PAGE_SIZE, this function can create a bio_vec that starts in a page
1151 : * after the bv_page. BIOs do not support folios that are 4GiB or larger.
1152 : *
1153 : * Return: Whether the addition was successful.
1154 : */
1155 0 : bool bio_add_folio(struct bio *bio, struct folio *folio, size_t len,
1156 : size_t off)
1157 : {
1158 0 : if (len > UINT_MAX || off > UINT_MAX)
1159 : return false;
1160 0 : return bio_add_page(bio, &folio->page, len, off) > 0;
1161 : }
1162 : EXPORT_SYMBOL(bio_add_folio);
1163 :
1164 0 : void __bio_release_pages(struct bio *bio, bool mark_dirty)
1165 : {
1166 : struct bvec_iter_all iter_all;
1167 : struct bio_vec *bvec;
1168 :
1169 0 : bio_for_each_segment_all(bvec, bio, iter_all) {
1170 0 : if (mark_dirty && !PageCompound(bvec->bv_page))
1171 0 : set_page_dirty_lock(bvec->bv_page);
1172 0 : put_page(bvec->bv_page);
1173 : }
1174 0 : }
1175 : EXPORT_SYMBOL_GPL(__bio_release_pages);
1176 :
1177 0 : void bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter)
1178 : {
1179 0 : size_t size = iov_iter_count(iter);
1180 :
1181 0 : WARN_ON_ONCE(bio->bi_max_vecs);
1182 :
1183 0 : if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1184 0 : struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1185 0 : size_t max_sectors = queue_max_zone_append_sectors(q);
1186 :
1187 0 : size = min(size, max_sectors << SECTOR_SHIFT);
1188 : }
1189 :
1190 0 : bio->bi_vcnt = iter->nr_segs;
1191 0 : bio->bi_io_vec = (struct bio_vec *)iter->bvec;
1192 0 : bio->bi_iter.bi_bvec_done = iter->iov_offset;
1193 0 : bio->bi_iter.bi_size = size;
1194 0 : bio_set_flag(bio, BIO_NO_PAGE_REF);
1195 0 : bio_set_flag(bio, BIO_CLONED);
1196 0 : }
1197 :
1198 0 : static int bio_iov_add_page(struct bio *bio, struct page *page,
1199 : unsigned int len, unsigned int offset)
1200 : {
1201 0 : bool same_page = false;
1202 :
1203 0 : if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
1204 0 : __bio_add_page(bio, page, len, offset);
1205 0 : return 0;
1206 : }
1207 :
1208 0 : if (same_page)
1209 0 : put_page(page);
1210 : return 0;
1211 : }
1212 :
1213 0 : static int bio_iov_add_zone_append_page(struct bio *bio, struct page *page,
1214 : unsigned int len, unsigned int offset)
1215 : {
1216 0 : struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1217 0 : bool same_page = false;
1218 :
1219 0 : if (bio_add_hw_page(q, bio, page, len, offset,
1220 : queue_max_zone_append_sectors(q), &same_page) != len)
1221 : return -EINVAL;
1222 0 : if (same_page)
1223 0 : put_page(page);
1224 : return 0;
1225 : }
1226 :
1227 : #define PAGE_PTRS_PER_BVEC (sizeof(struct bio_vec) / sizeof(struct page *))
1228 :
1229 : /**
1230 : * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
1231 : * @bio: bio to add pages to
1232 : * @iter: iov iterator describing the region to be mapped
1233 : *
1234 : * Pins pages from *iter and appends them to @bio's bvec array. The
1235 : * pages will have to be released using put_page() when done.
1236 : * For multi-segment *iter, this function only adds pages from the
1237 : * next non-empty segment of the iov iterator.
1238 : */
1239 0 : static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1240 : {
1241 0 : iov_iter_extraction_t extraction_flags = 0;
1242 0 : unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1243 0 : unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1244 0 : struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1245 0 : struct page **pages = (struct page **)bv;
1246 : ssize_t size, left;
1247 0 : unsigned len, i = 0;
1248 : size_t offset, trim;
1249 0 : int ret = 0;
1250 :
1251 : /*
1252 : * Move page array up in the allocated memory for the bio vecs as far as
1253 : * possible so that we can start filling biovecs from the beginning
1254 : * without overwriting the temporary page array.
1255 : */
1256 : BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1257 0 : pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1258 :
1259 0 : if (bio->bi_bdev && blk_queue_pci_p2pdma(bio->bi_bdev->bd_disk->queue))
1260 0 : extraction_flags |= ITER_ALLOW_P2PDMA;
1261 :
1262 : /*
1263 : * Each segment in the iov is required to be a block size multiple.
1264 : * However, we may not be able to get the entire segment if it spans
1265 : * more pages than bi_max_vecs allows, so we have to ALIGN_DOWN the
1266 : * result to ensure the bio's total size is correct. The remainder of
1267 : * the iov data will be picked up in the next bio iteration.
1268 : */
1269 0 : size = iov_iter_get_pages(iter, pages,
1270 0 : UINT_MAX - bio->bi_iter.bi_size,
1271 : nr_pages, &offset, extraction_flags);
1272 0 : if (unlikely(size <= 0))
1273 0 : return size ? size : -EFAULT;
1274 :
1275 0 : nr_pages = DIV_ROUND_UP(offset + size, PAGE_SIZE);
1276 :
1277 0 : trim = size & (bdev_logical_block_size(bio->bi_bdev) - 1);
1278 0 : iov_iter_revert(iter, trim);
1279 :
1280 0 : size -= trim;
1281 0 : if (unlikely(!size)) {
1282 : ret = -EFAULT;
1283 : goto out;
1284 : }
1285 :
1286 0 : for (left = size, i = 0; left > 0; left -= len, i++) {
1287 0 : struct page *page = pages[i];
1288 :
1289 0 : len = min_t(size_t, PAGE_SIZE - offset, left);
1290 0 : if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1291 0 : ret = bio_iov_add_zone_append_page(bio, page, len,
1292 : offset);
1293 0 : if (ret)
1294 : break;
1295 : } else
1296 0 : bio_iov_add_page(bio, page, len, offset);
1297 :
1298 0 : offset = 0;
1299 : }
1300 :
1301 0 : iov_iter_revert(iter, left);
1302 : out:
1303 0 : while (i < nr_pages)
1304 0 : put_page(pages[i++]);
1305 :
1306 : return ret;
1307 : }
1308 :
1309 : /**
1310 : * bio_iov_iter_get_pages - add user or kernel pages to a bio
1311 : * @bio: bio to add pages to
1312 : * @iter: iov iterator describing the region to be added
1313 : *
1314 : * This takes either an iterator pointing to user memory, or one pointing to
1315 : * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1316 : * map them into the kernel. On IO completion, the caller should put those
1317 : * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided
1318 : * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs
1319 : * to ensure the bvecs and pages stay referenced until the submitted I/O is
1320 : * completed by a call to ->ki_complete() or returns with an error other than
1321 : * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF
1322 : * on IO completion. If it isn't, then pages should be released.
1323 : *
1324 : * The function tries, but does not guarantee, to pin as many pages as
1325 : * fit into the bio, or are requested in @iter, whatever is smaller. If
1326 : * MM encounters an error pinning the requested pages, it stops. Error
1327 : * is returned only if 0 pages could be pinned.
1328 : */
1329 0 : int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1330 : {
1331 0 : int ret = 0;
1332 :
1333 0 : if (iov_iter_is_bvec(iter)) {
1334 0 : bio_iov_bvec_set(bio, iter);
1335 0 : iov_iter_advance(iter, bio->bi_iter.bi_size);
1336 0 : return 0;
1337 : }
1338 :
1339 : do {
1340 0 : ret = __bio_iov_iter_get_pages(bio, iter);
1341 0 : } while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));
1342 :
1343 0 : return bio->bi_vcnt ? 0 : ret;
1344 : }
1345 : EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
1346 :
1347 0 : static void submit_bio_wait_endio(struct bio *bio)
1348 : {
1349 0 : complete(bio->bi_private);
1350 0 : }
1351 :
1352 : /**
1353 : * submit_bio_wait - submit a bio, and wait until it completes
1354 : * @bio: The &struct bio which describes the I/O
1355 : *
1356 : * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1357 : * bio_endio() on failure.
1358 : *
1359 : * WARNING: Unlike to how submit_bio() is usually used, this function does not
1360 : * result in bio reference to be consumed. The caller must drop the reference
1361 : * on his own.
1362 : */
1363 0 : int submit_bio_wait(struct bio *bio)
1364 : {
1365 0 : DECLARE_COMPLETION_ONSTACK_MAP(done,
1366 : bio->bi_bdev->bd_disk->lockdep_map);
1367 : unsigned long hang_check;
1368 :
1369 0 : bio->bi_private = &done;
1370 0 : bio->bi_end_io = submit_bio_wait_endio;
1371 0 : bio->bi_opf |= REQ_SYNC;
1372 0 : submit_bio(bio);
1373 :
1374 : /* Prevent hang_check timer from firing at us during very long I/O */
1375 0 : hang_check = sysctl_hung_task_timeout_secs;
1376 : if (hang_check)
1377 : while (!wait_for_completion_io_timeout(&done,
1378 : hang_check * (HZ/2)))
1379 : ;
1380 : else
1381 0 : wait_for_completion_io(&done);
1382 :
1383 0 : return blk_status_to_errno(bio->bi_status);
1384 : }
1385 : EXPORT_SYMBOL(submit_bio_wait);
1386 :
1387 0 : void __bio_advance(struct bio *bio, unsigned bytes)
1388 : {
1389 0 : if (bio_integrity(bio))
1390 : bio_integrity_advance(bio, bytes);
1391 :
1392 0 : bio_crypt_advance(bio, bytes);
1393 0 : bio_advance_iter(bio, &bio->bi_iter, bytes);
1394 0 : }
1395 : EXPORT_SYMBOL(__bio_advance);
1396 :
1397 0 : void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1398 : struct bio *src, struct bvec_iter *src_iter)
1399 : {
1400 0 : while (src_iter->bi_size && dst_iter->bi_size) {
1401 0 : struct bio_vec src_bv = bio_iter_iovec(src, *src_iter);
1402 0 : struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter);
1403 0 : unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len);
1404 0 : void *src_buf = bvec_kmap_local(&src_bv);
1405 0 : void *dst_buf = bvec_kmap_local(&dst_bv);
1406 :
1407 0 : memcpy(dst_buf, src_buf, bytes);
1408 :
1409 : kunmap_local(dst_buf);
1410 : kunmap_local(src_buf);
1411 :
1412 0 : bio_advance_iter_single(src, src_iter, bytes);
1413 0 : bio_advance_iter_single(dst, dst_iter, bytes);
1414 : }
1415 0 : }
1416 : EXPORT_SYMBOL(bio_copy_data_iter);
1417 :
1418 : /**
1419 : * bio_copy_data - copy contents of data buffers from one bio to another
1420 : * @src: source bio
1421 : * @dst: destination bio
1422 : *
1423 : * Stops when it reaches the end of either @src or @dst - that is, copies
1424 : * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1425 : */
1426 0 : void bio_copy_data(struct bio *dst, struct bio *src)
1427 : {
1428 0 : struct bvec_iter src_iter = src->bi_iter;
1429 0 : struct bvec_iter dst_iter = dst->bi_iter;
1430 :
1431 0 : bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1432 0 : }
1433 : EXPORT_SYMBOL(bio_copy_data);
1434 :
1435 0 : void bio_free_pages(struct bio *bio)
1436 : {
1437 : struct bio_vec *bvec;
1438 : struct bvec_iter_all iter_all;
1439 :
1440 0 : bio_for_each_segment_all(bvec, bio, iter_all)
1441 0 : __free_page(bvec->bv_page);
1442 0 : }
1443 : EXPORT_SYMBOL(bio_free_pages);
1444 :
1445 : /*
1446 : * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1447 : * for performing direct-IO in BIOs.
1448 : *
1449 : * The problem is that we cannot run set_page_dirty() from interrupt context
1450 : * because the required locks are not interrupt-safe. So what we can do is to
1451 : * mark the pages dirty _before_ performing IO. And in interrupt context,
1452 : * check that the pages are still dirty. If so, fine. If not, redirty them
1453 : * in process context.
1454 : *
1455 : * We special-case compound pages here: normally this means reads into hugetlb
1456 : * pages. The logic in here doesn't really work right for compound pages
1457 : * because the VM does not uniformly chase down the head page in all cases.
1458 : * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1459 : * handle them at all. So we skip compound pages here at an early stage.
1460 : *
1461 : * Note that this code is very hard to test under normal circumstances because
1462 : * direct-io pins the pages with get_user_pages(). This makes
1463 : * is_page_cache_freeable return false, and the VM will not clean the pages.
1464 : * But other code (eg, flusher threads) could clean the pages if they are mapped
1465 : * pagecache.
1466 : *
1467 : * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1468 : * deferred bio dirtying paths.
1469 : */
1470 :
1471 : /*
1472 : * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1473 : */
1474 0 : void bio_set_pages_dirty(struct bio *bio)
1475 : {
1476 : struct bio_vec *bvec;
1477 : struct bvec_iter_all iter_all;
1478 :
1479 0 : bio_for_each_segment_all(bvec, bio, iter_all) {
1480 0 : if (!PageCompound(bvec->bv_page))
1481 0 : set_page_dirty_lock(bvec->bv_page);
1482 : }
1483 0 : }
1484 :
1485 : /*
1486 : * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1487 : * If they are, then fine. If, however, some pages are clean then they must
1488 : * have been written out during the direct-IO read. So we take another ref on
1489 : * the BIO and re-dirty the pages in process context.
1490 : *
1491 : * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1492 : * here on. It will run one put_page() against each page and will run one
1493 : * bio_put() against the BIO.
1494 : */
1495 :
1496 : static void bio_dirty_fn(struct work_struct *work);
1497 :
1498 : static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1499 : static DEFINE_SPINLOCK(bio_dirty_lock);
1500 : static struct bio *bio_dirty_list;
1501 :
1502 : /*
1503 : * This runs in process context
1504 : */
1505 0 : static void bio_dirty_fn(struct work_struct *work)
1506 : {
1507 : struct bio *bio, *next;
1508 :
1509 0 : spin_lock_irq(&bio_dirty_lock);
1510 0 : next = bio_dirty_list;
1511 0 : bio_dirty_list = NULL;
1512 : spin_unlock_irq(&bio_dirty_lock);
1513 :
1514 0 : while ((bio = next) != NULL) {
1515 0 : next = bio->bi_private;
1516 :
1517 0 : bio_release_pages(bio, true);
1518 0 : bio_put(bio);
1519 : }
1520 0 : }
1521 :
1522 0 : void bio_check_pages_dirty(struct bio *bio)
1523 : {
1524 : struct bio_vec *bvec;
1525 : unsigned long flags;
1526 : struct bvec_iter_all iter_all;
1527 :
1528 0 : bio_for_each_segment_all(bvec, bio, iter_all) {
1529 0 : if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page))
1530 : goto defer;
1531 : }
1532 :
1533 0 : bio_release_pages(bio, false);
1534 0 : bio_put(bio);
1535 0 : return;
1536 : defer:
1537 0 : spin_lock_irqsave(&bio_dirty_lock, flags);
1538 0 : bio->bi_private = bio_dirty_list;
1539 0 : bio_dirty_list = bio;
1540 0 : spin_unlock_irqrestore(&bio_dirty_lock, flags);
1541 0 : schedule_work(&bio_dirty_work);
1542 : }
1543 :
1544 0 : static inline bool bio_remaining_done(struct bio *bio)
1545 : {
1546 : /*
1547 : * If we're not chaining, then ->__bi_remaining is always 1 and
1548 : * we always end io on the first invocation.
1549 : */
1550 0 : if (!bio_flagged(bio, BIO_CHAIN))
1551 : return true;
1552 :
1553 0 : BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1554 :
1555 0 : if (atomic_dec_and_test(&bio->__bi_remaining)) {
1556 0 : bio_clear_flag(bio, BIO_CHAIN);
1557 0 : return true;
1558 : }
1559 :
1560 : return false;
1561 : }
1562 :
1563 : /**
1564 : * bio_endio - end I/O on a bio
1565 : * @bio: bio
1566 : *
1567 : * Description:
1568 : * bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1569 : * way to end I/O on a bio. No one should call bi_end_io() directly on a
1570 : * bio unless they own it and thus know that it has an end_io function.
1571 : *
1572 : * bio_endio() can be called several times on a bio that has been chained
1573 : * using bio_chain(). The ->bi_end_io() function will only be called the
1574 : * last time.
1575 : **/
1576 0 : void bio_endio(struct bio *bio)
1577 : {
1578 : again:
1579 0 : if (!bio_remaining_done(bio))
1580 : return;
1581 0 : if (!bio_integrity_endio(bio))
1582 : return;
1583 :
1584 0 : rq_qos_done_bio(bio);
1585 :
1586 0 : if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1587 0 : trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio);
1588 : bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1589 : }
1590 :
1591 : /*
1592 : * Need to have a real endio function for chained bios, otherwise
1593 : * various corner cases will break (like stacking block devices that
1594 : * save/restore bi_end_io) - however, we want to avoid unbounded
1595 : * recursion and blowing the stack. Tail call optimization would
1596 : * handle this, but compiling with frame pointers also disables
1597 : * gcc's sibling call optimization.
1598 : */
1599 0 : if (bio->bi_end_io == bio_chain_endio) {
1600 0 : bio = __bio_chain_endio(bio);
1601 0 : goto again;
1602 : }
1603 :
1604 0 : blk_throtl_bio_endio(bio);
1605 : /* release cgroup info */
1606 0 : bio_uninit(bio);
1607 0 : if (bio->bi_end_io)
1608 0 : bio->bi_end_io(bio);
1609 : }
1610 : EXPORT_SYMBOL(bio_endio);
1611 :
1612 : /**
1613 : * bio_split - split a bio
1614 : * @bio: bio to split
1615 : * @sectors: number of sectors to split from the front of @bio
1616 : * @gfp: gfp mask
1617 : * @bs: bio set to allocate from
1618 : *
1619 : * Allocates and returns a new bio which represents @sectors from the start of
1620 : * @bio, and updates @bio to represent the remaining sectors.
1621 : *
1622 : * Unless this is a discard request the newly allocated bio will point
1623 : * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1624 : * neither @bio nor @bs are freed before the split bio.
1625 : */
1626 0 : struct bio *bio_split(struct bio *bio, int sectors,
1627 : gfp_t gfp, struct bio_set *bs)
1628 : {
1629 : struct bio *split;
1630 :
1631 0 : BUG_ON(sectors <= 0);
1632 0 : BUG_ON(sectors >= bio_sectors(bio));
1633 :
1634 : /* Zone append commands cannot be split */
1635 0 : if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))
1636 : return NULL;
1637 :
1638 0 : split = bio_alloc_clone(bio->bi_bdev, bio, gfp, bs);
1639 0 : if (!split)
1640 : return NULL;
1641 :
1642 0 : split->bi_iter.bi_size = sectors << 9;
1643 :
1644 0 : if (bio_integrity(split))
1645 : bio_integrity_trim(split);
1646 :
1647 0 : bio_advance(bio, split->bi_iter.bi_size);
1648 :
1649 0 : if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1650 : bio_set_flag(split, BIO_TRACE_COMPLETION);
1651 :
1652 : return split;
1653 : }
1654 : EXPORT_SYMBOL(bio_split);
1655 :
1656 : /**
1657 : * bio_trim - trim a bio
1658 : * @bio: bio to trim
1659 : * @offset: number of sectors to trim from the front of @bio
1660 : * @size: size we want to trim @bio to, in sectors
1661 : *
1662 : * This function is typically used for bios that are cloned and submitted
1663 : * to the underlying device in parts.
1664 : */
1665 0 : void bio_trim(struct bio *bio, sector_t offset, sector_t size)
1666 : {
1667 0 : if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS ||
1668 : offset + size > bio_sectors(bio)))
1669 : return;
1670 :
1671 0 : size <<= 9;
1672 0 : if (offset == 0 && size == bio->bi_iter.bi_size)
1673 : return;
1674 :
1675 0 : bio_advance(bio, offset << 9);
1676 0 : bio->bi_iter.bi_size = size;
1677 :
1678 0 : if (bio_integrity(bio))
1679 : bio_integrity_trim(bio);
1680 : }
1681 : EXPORT_SYMBOL_GPL(bio_trim);
1682 :
1683 : /*
1684 : * create memory pools for biovec's in a bio_set.
1685 : * use the global biovec slabs created for general use.
1686 : */
1687 0 : int biovec_init_pool(mempool_t *pool, int pool_entries)
1688 : {
1689 2 : struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1;
1690 :
1691 4 : return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1692 : }
1693 :
1694 : /*
1695 : * bioset_exit - exit a bioset initialized with bioset_init()
1696 : *
1697 : * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1698 : * kzalloc()).
1699 : */
1700 0 : void bioset_exit(struct bio_set *bs)
1701 : {
1702 0 : bio_alloc_cache_destroy(bs);
1703 0 : if (bs->rescue_workqueue)
1704 0 : destroy_workqueue(bs->rescue_workqueue);
1705 0 : bs->rescue_workqueue = NULL;
1706 :
1707 0 : mempool_exit(&bs->bio_pool);
1708 0 : mempool_exit(&bs->bvec_pool);
1709 :
1710 0 : bioset_integrity_free(bs);
1711 0 : if (bs->bio_slab)
1712 0 : bio_put_slab(bs);
1713 0 : bs->bio_slab = NULL;
1714 0 : }
1715 : EXPORT_SYMBOL(bioset_exit);
1716 :
1717 : /**
1718 : * bioset_init - Initialize a bio_set
1719 : * @bs: pool to initialize
1720 : * @pool_size: Number of bio and bio_vecs to cache in the mempool
1721 : * @front_pad: Number of bytes to allocate in front of the returned bio
1722 : * @flags: Flags to modify behavior, currently %BIOSET_NEED_BVECS
1723 : * and %BIOSET_NEED_RESCUER
1724 : *
1725 : * Description:
1726 : * Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1727 : * to ask for a number of bytes to be allocated in front of the bio.
1728 : * Front pad allocation is useful for embedding the bio inside
1729 : * another structure, to avoid allocating extra data to go with the bio.
1730 : * Note that the bio must be embedded at the END of that structure always,
1731 : * or things will break badly.
1732 : * If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1733 : * for allocating iovecs. This pool is not needed e.g. for bio_init_clone().
1734 : * If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used
1735 : * to dispatch queued requests when the mempool runs out of space.
1736 : *
1737 : */
1738 2 : int bioset_init(struct bio_set *bs,
1739 : unsigned int pool_size,
1740 : unsigned int front_pad,
1741 : int flags)
1742 : {
1743 2 : bs->front_pad = front_pad;
1744 2 : if (flags & BIOSET_NEED_BVECS)
1745 2 : bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1746 : else
1747 0 : bs->back_pad = 0;
1748 :
1749 2 : spin_lock_init(&bs->rescue_lock);
1750 4 : bio_list_init(&bs->rescue_list);
1751 4 : INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1752 :
1753 2 : bs->bio_slab = bio_find_or_create_slab(bs);
1754 2 : if (!bs->bio_slab)
1755 : return -ENOMEM;
1756 :
1757 4 : if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
1758 : goto bad;
1759 :
1760 4 : if ((flags & BIOSET_NEED_BVECS) &&
1761 4 : biovec_init_pool(&bs->bvec_pool, pool_size))
1762 : goto bad;
1763 :
1764 2 : if (flags & BIOSET_NEED_RESCUER) {
1765 0 : bs->rescue_workqueue = alloc_workqueue("bioset",
1766 : WQ_MEM_RECLAIM, 0);
1767 0 : if (!bs->rescue_workqueue)
1768 : goto bad;
1769 : }
1770 2 : if (flags & BIOSET_PERCPU_CACHE) {
1771 2 : bs->cache = alloc_percpu(struct bio_alloc_cache);
1772 2 : if (!bs->cache)
1773 : goto bad;
1774 2 : cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
1775 : }
1776 :
1777 : return 0;
1778 : bad:
1779 0 : bioset_exit(bs);
1780 0 : return -ENOMEM;
1781 : }
1782 : EXPORT_SYMBOL(bioset_init);
1783 :
1784 1 : static int __init init_bio(void)
1785 : {
1786 : int i;
1787 :
1788 : BUILD_BUG_ON(BIO_FLAG_LAST > 8 * sizeof_field(struct bio, bi_flags));
1789 :
1790 : bio_integrity_init();
1791 :
1792 5 : for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) {
1793 4 : struct biovec_slab *bvs = bvec_slabs + i;
1794 :
1795 4 : bvs->slab = kmem_cache_create(bvs->name,
1796 4 : bvs->nr_vecs * sizeof(struct bio_vec), 0,
1797 : SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1798 : }
1799 :
1800 1 : cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL,
1801 : bio_cpu_dead);
1802 :
1803 1 : if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0,
1804 : BIOSET_NEED_BVECS | BIOSET_PERCPU_CACHE))
1805 0 : panic("bio: can't allocate bios\n");
1806 :
1807 1 : if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
1808 : panic("bio: can't create integrity pool\n");
1809 :
1810 1 : return 0;
1811 : }
1812 : subsys_initcall(init_bio);
|
