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 
biovec_slab(unsigned short nr_vecs)49 static struct biovec_slab *biovec_slab(unsigned short nr_vecs)
50 {
51 	switch (nr_vecs) {
52 	/* smaller bios use inline vecs */
53 	case 5 ... 16:
54 		return &bvec_slabs[0];
55 	case 17 ... 64:
56 		return &bvec_slabs[1];
57 	case 65 ... 128:
58 		return &bvec_slabs[2];
59 	case 129 ... BIO_MAX_VECS:
60 		return &bvec_slabs[3];
61 	default:
62 		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 
create_bio_slab(unsigned int size)86 static struct bio_slab *create_bio_slab(unsigned int size)
87 {
88 	struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL);
89 
90 	if (!bslab)
91 		return NULL;
92 
93 	snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size);
94 	bslab->slab = kmem_cache_create(bslab->name, size,
95 			ARCH_KMALLOC_MINALIGN,
96 			SLAB_HWCACHE_ALIGN | SLAB_TYPESAFE_BY_RCU, NULL);
97 	if (!bslab->slab)
98 		goto fail_alloc_slab;
99 
100 	bslab->slab_ref = 1;
101 	bslab->slab_size = size;
102 
103 	if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL)))
104 		return bslab;
105 
106 	kmem_cache_destroy(bslab->slab);
107 
108 fail_alloc_slab:
109 	kfree(bslab);
110 	return NULL;
111 }
112 
bs_bio_slab_size(struct bio_set * bs)113 static inline unsigned int bs_bio_slab_size(struct bio_set *bs)
114 {
115 	return bs->front_pad + sizeof(struct bio) + bs->back_pad;
116 }
117 
bio_find_or_create_slab(struct bio_set * bs)118 static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs)
119 {
120 	unsigned int size = bs_bio_slab_size(bs);
121 	struct bio_slab *bslab;
122 
123 	mutex_lock(&bio_slab_lock);
124 	bslab = xa_load(&bio_slabs, size);
125 	if (bslab)
126 		bslab->slab_ref++;
127 	else
128 		bslab = create_bio_slab(size);
129 	mutex_unlock(&bio_slab_lock);
130 
131 	if (bslab)
132 		return bslab->slab;
133 	return NULL;
134 }
135 
bio_put_slab(struct bio_set * bs)136 static void bio_put_slab(struct bio_set *bs)
137 {
138 	struct bio_slab *bslab = NULL;
139 	unsigned int slab_size = bs_bio_slab_size(bs);
140 
141 	mutex_lock(&bio_slab_lock);
142 
143 	bslab = xa_load(&bio_slabs, slab_size);
144 	if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
145 		goto out;
146 
147 	WARN_ON_ONCE(bslab->slab != bs->bio_slab);
148 
149 	WARN_ON(!bslab->slab_ref);
150 
151 	if (--bslab->slab_ref)
152 		goto out;
153 
154 	xa_erase(&bio_slabs, slab_size);
155 
156 	kmem_cache_destroy(bslab->slab);
157 	kfree(bslab);
158 
159 out:
160 	mutex_unlock(&bio_slab_lock);
161 }
162 
bvec_free(mempool_t * pool,struct bio_vec * bv,unsigned short nr_vecs)163 void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs)
164 {
165 	BUG_ON(nr_vecs > BIO_MAX_VECS);
166 
167 	if (nr_vecs == BIO_MAX_VECS)
168 		mempool_free(bv, pool);
169 	else if (nr_vecs > BIO_INLINE_VECS)
170 		kmem_cache_free(biovec_slab(nr_vecs)->slab, bv);
171 }
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  */
bvec_alloc_gfp(gfp_t gfp)177 static inline gfp_t bvec_alloc_gfp(gfp_t gfp)
178 {
179 	return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) |
180 		__GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
181 }
182 
bvec_alloc(mempool_t * pool,unsigned short * nr_vecs,gfp_t gfp_mask)183 struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
184 		gfp_t gfp_mask)
185 {
186 	struct biovec_slab *bvs = biovec_slab(*nr_vecs);
187 
188 	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 	*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 	if (*nr_vecs < BIO_MAX_VECS) {
203 		struct bio_vec *bvl;
204 
205 		bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask));
206 		if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM))
207 			return bvl;
208 		*nr_vecs = BIO_MAX_VECS;
209 	}
210 
211 	return mempool_alloc(pool, gfp_mask);
212 }
213 
bio_uninit(struct bio * bio)214 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 	if (bio_integrity(bio))
223 		bio_integrity_free(bio);
224 
225 	bio_crypt_free_ctx(bio);
226 }
227 EXPORT_SYMBOL(bio_uninit);
228 
bio_free(struct bio * bio)229 static void bio_free(struct bio *bio)
230 {
231 	struct bio_set *bs = bio->bi_pool;
232 	void *p = bio;
233 
234 	WARN_ON_ONCE(!bs);
235 
236 	bio_uninit(bio);
237 	bvec_free(&bs->bvec_pool, bio->bi_io_vec, bio->bi_max_vecs);
238 	mempool_free(p - bs->front_pad, &bs->bio_pool);
239 }
240 
241 /*
242  * Users of this function have their own bio allocation. Subsequently,
243  * they must remember to pair any call to bio_init() with bio_uninit()
244  * when IO has completed, or when the bio is released.
245  */
bio_init(struct bio * bio,struct block_device * bdev,struct bio_vec * table,unsigned short max_vecs,blk_opf_t opf)246 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 	bio->bi_next = NULL;
250 	bio->bi_bdev = bdev;
251 	bio->bi_opf = opf;
252 	bio->bi_flags = 0;
253 	bio->bi_ioprio = 0;
254 	bio->bi_status = 0;
255 	bio->bi_iter.bi_sector = 0;
256 	bio->bi_iter.bi_size = 0;
257 	bio->bi_iter.bi_idx = 0;
258 	bio->bi_iter.bi_bvec_done = 0;
259 	bio->bi_end_io = NULL;
260 	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 	bio->bi_vcnt = 0;
277 
278 	atomic_set(&bio->__bi_remaining, 1);
279 	atomic_set(&bio->__bi_cnt, 1);
280 	bio->bi_cookie = BLK_QC_T_NONE;
281 
282 	bio->bi_max_vecs = max_vecs;
283 	bio->bi_io_vec = table;
284 	bio->bi_pool = NULL;
285 }
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  */
bio_reset(struct bio * bio,struct block_device * bdev,blk_opf_t opf)300 void bio_reset(struct bio *bio, struct block_device *bdev, blk_opf_t opf)
301 {
302 	bio_uninit(bio);
303 	memset(bio, 0, BIO_RESET_BYTES);
304 	atomic_set(&bio->__bi_remaining, 1);
305 	bio->bi_bdev = bdev;
306 	if (bio->bi_bdev)
307 		bio_associate_blkg(bio);
308 	bio->bi_opf = opf;
309 }
310 EXPORT_SYMBOL(bio_reset);
311 
__bio_chain_endio(struct bio * bio)312 static struct bio *__bio_chain_endio(struct bio *bio)
313 {
314 	struct bio *parent = bio->bi_private;
315 
316 	if (bio->bi_status && !parent->bi_status)
317 		parent->bi_status = bio->bi_status;
318 	bio_put(bio);
319 	return parent;
320 }
321 
bio_chain_endio(struct bio * bio)322 static void bio_chain_endio(struct bio *bio)
323 {
324 	bio_endio(__bio_chain_endio(bio));
325 }
326 
327 /**
328  * bio_chain - chain bio completions
329  * @bio: the target bio
330  * @parent: the 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  */
bio_chain(struct bio * bio,struct bio * parent)338 void bio_chain(struct bio *bio, struct bio *parent)
339 {
340 	BUG_ON(bio->bi_private || bio->bi_end_io);
341 
342 	bio->bi_private = parent;
343 	bio->bi_end_io	= bio_chain_endio;
344 	bio_inc_remaining(parent);
345 }
346 EXPORT_SYMBOL(bio_chain);
347 
blk_next_bio(struct bio * bio,struct block_device * bdev,unsigned int nr_pages,blk_opf_t opf,gfp_t gfp)348 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 	struct bio *new = bio_alloc(bdev, nr_pages, opf, gfp);
352 
353 	if (bio) {
354 		bio_chain(bio, new);
355 		submit_bio(bio);
356 	}
357 
358 	return new;
359 }
360 EXPORT_SYMBOL_GPL(blk_next_bio);
361 
bio_alloc_rescue(struct work_struct * work)362 static void bio_alloc_rescue(struct work_struct *work)
363 {
364 	struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
365 	struct bio *bio;
366 
367 	while (1) {
368 		spin_lock(&bs->rescue_lock);
369 		bio = bio_list_pop(&bs->rescue_list);
370 		spin_unlock(&bs->rescue_lock);
371 
372 		if (!bio)
373 			break;
374 
375 		submit_bio_noacct(bio);
376 	}
377 }
378 
punt_bios_to_rescuer(struct bio_set * bs)379 static void punt_bios_to_rescuer(struct bio_set *bs)
380 {
381 	struct bio_list punt, nopunt;
382 	struct bio *bio;
383 
384 	if (WARN_ON_ONCE(!bs->rescue_workqueue))
385 		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 	bio_list_init(&punt);
398 	bio_list_init(&nopunt);
399 
400 	while ((bio = bio_list_pop(&current->bio_list[0])))
401 		bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
402 	current->bio_list[0] = nopunt;
403 
404 	bio_list_init(&nopunt);
405 	while ((bio = bio_list_pop(&current->bio_list[1])))
406 		bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
407 	current->bio_list[1] = nopunt;
408 
409 	spin_lock(&bs->rescue_lock);
410 	bio_list_merge(&bs->rescue_list, &punt);
411 	spin_unlock(&bs->rescue_lock);
412 
413 	queue_work(bs->rescue_workqueue, &bs->rescue_work);
414 }
415 
bio_alloc_irq_cache_splice(struct bio_alloc_cache * cache)416 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 	if (WARN_ON_ONCE(cache->free_list))
422 		return;
423 
424 	local_irq_save(flags);
425 	cache->free_list = cache->free_list_irq;
426 	cache->free_list_irq = NULL;
427 	cache->nr += cache->nr_irq;
428 	cache->nr_irq = 0;
429 	local_irq_restore(flags);
430 }
431 
bio_alloc_percpu_cache(struct block_device * bdev,unsigned short nr_vecs,blk_opf_t opf,gfp_t gfp,struct bio_set * bs)432 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 	cache = per_cpu_ptr(bs->cache, get_cpu());
440 	if (!cache->free_list) {
441 		if (READ_ONCE(cache->nr_irq) >= ALLOC_CACHE_THRESHOLD)
442 			bio_alloc_irq_cache_splice(cache);
443 		if (!cache->free_list) {
444 			put_cpu();
445 			return NULL;
446 		}
447 	}
448 	bio = cache->free_list;
449 	cache->free_list = bio->bi_next;
450 	cache->nr--;
451 	put_cpu();
452 
453 	bio_init(bio, bdev, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs, opf);
454 	bio->bi_pool = bs;
455 	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  */
bio_alloc_bioset(struct block_device * bdev,unsigned short nr_vecs,blk_opf_t opf,gfp_t gfp_mask,struct bio_set * bs)492 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 	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 	if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_vecs > 0))
502 		return NULL;
503 
504 	if (opf & REQ_ALLOC_CACHE) {
505 		if (bs->cache && nr_vecs <= BIO_INLINE_VECS) {
506 			bio = bio_alloc_percpu_cache(bdev, nr_vecs, opf,
507 						     gfp_mask, bs);
508 			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 			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 	if (current->bio_list &&
538 	    (!bio_list_empty(&current->bio_list[0]) ||
539 	     !bio_list_empty(&current->bio_list[1])) &&
540 	    bs->rescue_workqueue)
541 		gfp_mask &= ~__GFP_DIRECT_RECLAIM;
542 
543 	p = mempool_alloc(&bs->bio_pool, gfp_mask);
544 	if (!p && gfp_mask != saved_gfp) {
545 		punt_bios_to_rescuer(bs);
546 		gfp_mask = saved_gfp;
547 		p = mempool_alloc(&bs->bio_pool, gfp_mask);
548 	}
549 	if (unlikely(!p))
550 		return NULL;
551 	if (!mempool_is_saturated(&bs->bio_pool))
552 		opf &= ~REQ_ALLOC_CACHE;
553 
554 	bio = p + bs->front_pad;
555 	if (nr_vecs > BIO_INLINE_VECS) {
556 		struct bio_vec *bvl = NULL;
557 
558 		bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);
559 		if (!bvl && gfp_mask != saved_gfp) {
560 			punt_bios_to_rescuer(bs);
561 			gfp_mask = saved_gfp;
562 			bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);
563 		}
564 		if (unlikely(!bvl))
565 			goto err_free;
566 
567 		bio_init(bio, bdev, bvl, nr_vecs, opf);
568 	} else if (nr_vecs) {
569 		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 	bio->bi_pool = bs;
575 	return bio;
576 
577 err_free:
578 	mempool_free(p, &bs->bio_pool);
579 	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  */
bio_kmalloc(unsigned short nr_vecs,gfp_t gfp_mask)599 struct bio *bio_kmalloc(unsigned short nr_vecs, gfp_t gfp_mask)
600 {
601 	struct bio *bio;
602 
603 	if (nr_vecs > UIO_MAXIOV)
604 		return NULL;
605 	return kmalloc(struct_size(bio, bi_inline_vecs, nr_vecs), gfp_mask);
606 }
607 EXPORT_SYMBOL(bio_kmalloc);
608 
zero_fill_bio(struct bio * bio)609 void zero_fill_bio(struct bio *bio)
610 {
611 	struct bio_vec bv;
612 	struct bvec_iter iter;
613 
614 	bio_for_each_segment(bv, bio, iter)
615 		memzero_bvec(&bv);
616 }
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  */
bio_truncate(struct bio * bio,unsigned new_size)629 static void bio_truncate(struct bio *bio, unsigned new_size)
630 {
631 	struct bio_vec bv;
632 	struct bvec_iter iter;
633 	unsigned int done = 0;
634 	bool truncated = false;
635 
636 	if (new_size >= bio->bi_iter.bi_size)
637 		return;
638 
639 	if (bio_op(bio) != REQ_OP_READ)
640 		goto exit;
641 
642 	bio_for_each_segment(bv, bio, iter) {
643 		if (done + bv.bv_len > new_size) {
644 			unsigned offset;
645 
646 			if (!truncated)
647 				offset = new_size - done;
648 			else
649 				offset = 0;
650 			zero_user(bv.bv_page, bv.bv_offset + offset,
651 				  bv.bv_len - offset);
652 			truncated = true;
653 		}
654 		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 	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  */
guard_bio_eod(struct bio * bio)681 void guard_bio_eod(struct bio *bio)
682 {
683 	sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
684 
685 	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 	if (unlikely(bio->bi_iter.bi_sector >= maxsector))
694 		return;
695 
696 	maxsector -= bio->bi_iter.bi_sector;
697 	if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
698 		return;
699 
700 	bio_truncate(bio, maxsector << 9);
701 }
702 
__bio_alloc_cache_prune(struct bio_alloc_cache * cache,unsigned int nr)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 	while ((bio = cache->free_list) != NULL) {
710 		cache->free_list = bio->bi_next;
711 		cache->nr--;
712 		bio_free(bio);
713 		if (++i == nr)
714 			break;
715 	}
716 	return i;
717 }
718 
bio_alloc_cache_prune(struct bio_alloc_cache * cache,unsigned int nr)719 static void bio_alloc_cache_prune(struct bio_alloc_cache *cache,
720 				  unsigned int nr)
721 {
722 	nr -= __bio_alloc_cache_prune(cache, nr);
723 	if (!READ_ONCE(cache->free_list)) {
724 		bio_alloc_irq_cache_splice(cache);
725 		__bio_alloc_cache_prune(cache, nr);
726 	}
727 }
728 
bio_cpu_dead(unsigned int cpu,struct hlist_node * node)729 static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node)
730 {
731 	struct bio_set *bs;
732 
733 	bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead);
734 	if (bs->cache) {
735 		struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu);
736 
737 		bio_alloc_cache_prune(cache, -1U);
738 	}
739 	return 0;
740 }
741 
bio_alloc_cache_destroy(struct bio_set * bs)742 static void bio_alloc_cache_destroy(struct bio_set *bs)
743 {
744 	int cpu;
745 
746 	if (!bs->cache)
747 		return;
748 
749 	cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
750 	for_each_possible_cpu(cpu) {
751 		struct bio_alloc_cache *cache;
752 
753 		cache = per_cpu_ptr(bs->cache, cpu);
754 		bio_alloc_cache_prune(cache, -1U);
755 	}
756 	free_percpu(bs->cache);
757 	bs->cache = NULL;
758 }
759 
bio_put_percpu_cache(struct bio * bio)760 static inline void bio_put_percpu_cache(struct bio *bio)
761 {
762 	struct bio_alloc_cache *cache;
763 
764 	cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu());
765 	if (READ_ONCE(cache->nr_irq) + cache->nr > ALLOC_CACHE_MAX) {
766 		put_cpu();
767 		bio_free(bio);
768 		return;
769 	}
770 
771 	bio_uninit(bio);
772 
773 	if ((bio->bi_opf & REQ_POLLED) && !WARN_ON_ONCE(in_interrupt())) {
774 		bio->bi_next = cache->free_list;
775 		bio->bi_bdev = NULL;
776 		cache->free_list = bio;
777 		cache->nr++;
778 	} else {
779 		unsigned long flags;
780 
781 		local_irq_save(flags);
782 		bio->bi_next = cache->free_list_irq;
783 		cache->free_list_irq = bio;
784 		cache->nr_irq++;
785 		local_irq_restore(flags);
786 	}
787 	put_cpu();
788 }
789 
790 /**
791  * bio_put - release a reference to a bio
792  * @bio:   bio to release reference to
793  *
794  * Description:
795  *   Put a reference to a &struct bio, either one you have gotten with
796  *   bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
797  **/
bio_put(struct bio * bio)798 void bio_put(struct bio *bio)
799 {
800 	if (unlikely(bio_flagged(bio, BIO_REFFED))) {
801 		BUG_ON(!atomic_read(&bio->__bi_cnt));
802 		if (!atomic_dec_and_test(&bio->__bi_cnt))
803 			return;
804 	}
805 	if (bio->bi_opf & REQ_ALLOC_CACHE)
806 		bio_put_percpu_cache(bio);
807 	else
808 		bio_free(bio);
809 }
810 EXPORT_SYMBOL(bio_put);
811 
__bio_clone(struct bio * bio,struct bio * bio_src,gfp_t gfp)812 static int __bio_clone(struct bio *bio, struct bio *bio_src, gfp_t gfp)
813 {
814 	bio_set_flag(bio, BIO_CLONED);
815 	bio->bi_ioprio = bio_src->bi_ioprio;
816 	bio->bi_iter = bio_src->bi_iter;
817 
818 	if (bio->bi_bdev) {
819 		if (bio->bi_bdev == bio_src->bi_bdev &&
820 		    bio_flagged(bio_src, BIO_REMAPPED))
821 			bio_set_flag(bio, BIO_REMAPPED);
822 		bio_clone_blkg_association(bio, bio_src);
823 	}
824 
825 	if (bio_crypt_clone(bio, bio_src, gfp) < 0)
826 		return -ENOMEM;
827 	if (bio_integrity(bio_src) &&
828 	    bio_integrity_clone(bio, bio_src, gfp) < 0)
829 		return -ENOMEM;
830 	return 0;
831 }
832 
833 /**
834  * bio_alloc_clone - clone a bio that shares the original bio's biovec
835  * @bdev: block_device to clone onto
836  * @bio_src: bio to clone from
837  * @gfp: allocation priority
838  * @bs: bio_set to allocate from
839  *
840  * Allocate a new bio that is a clone of @bio_src. The caller owns the returned
841  * bio, but not the actual data it points to.
842  *
843  * The caller must ensure that the return bio is not freed before @bio_src.
844  */
bio_alloc_clone(struct block_device * bdev,struct bio * bio_src,gfp_t gfp,struct bio_set * bs)845 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 	bio = bio_alloc_bioset(bdev, 0, bio_src->bi_opf, gfp, bs);
851 	if (!bio)
852 		return NULL;
853 
854 	if (__bio_clone(bio, bio_src, gfp) < 0) {
855 		bio_put(bio);
856 		return NULL;
857 	}
858 	bio->bi_io_vec = bio_src->bi_io_vec;
859 
860 	return bio;
861 }
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  */
bio_init_clone(struct block_device * bdev,struct bio * bio,struct bio * bio_src,gfp_t gfp)876 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 	bio_init(bio, bdev, bio_src->bi_io_vec, 0, bio_src->bi_opf);
882 	ret = __bio_clone(bio, bio_src, gfp);
883 	if (ret)
884 		bio_uninit(bio);
885 	return ret;
886 }
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  */
bio_full(struct bio * bio,unsigned len)897 static inline bool bio_full(struct bio *bio, unsigned len)
898 {
899 	if (bio->bi_vcnt >= bio->bi_max_vecs)
900 		return true;
901 	if (bio->bi_iter.bi_size > UINT_MAX - len)
902 		return true;
903 	return false;
904 }
905 
page_is_mergeable(const struct bio_vec * bv,struct page * page,unsigned int len,unsigned int off,bool * same_page)906 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 	size_t bv_end = bv->bv_offset + bv->bv_len;
911 	phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1;
912 	phys_addr_t page_addr = page_to_phys(page);
913 
914 	if (vec_end_addr + 1 != page_addr + off)
915 		return false;
916 	if (xen_domain() && !xen_biovec_phys_mergeable(bv, page))
917 		return false;
918 	if (!zone_device_pages_have_same_pgmap(bv->bv_page, page))
919 		return false;
920 
921 	*same_page = ((vec_end_addr & PAGE_MASK) == page_addr);
922 	if (*same_page)
923 		return true;
924 	else if (IS_ENABLED(CONFIG_KMSAN))
925 		return false;
926 	return (bv->bv_page + bv_end / PAGE_SIZE) == (page + off / PAGE_SIZE);
927 }
928 
929 /**
930  * __bio_try_merge_page - try appending data to an existing bvec.
931  * @bio: destination bio
932  * @page: start page to add
933  * @len: length of the data to add
934  * @off: offset of the data relative to @page
935  * @same_page: return if the segment has been merged inside the same page
936  *
937  * Try to add the data at @page + @off to the last bvec of @bio.  This is a
938  * useful optimisation for file systems with a block size smaller than the
939  * page size.
940  *
941  * Warn if (@len, @off) crosses pages in case that @same_page is true.
942  *
943  * Return %true on success or %false on failure.
944  */
__bio_try_merge_page(struct bio * bio,struct page * page,unsigned int len,unsigned int off,bool * same_page)945 static bool __bio_try_merge_page(struct bio *bio, struct page *page,
946 		unsigned int len, unsigned int off, bool *same_page)
947 {
948 	if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
949 		return false;
950 
951 	if (bio->bi_vcnt > 0) {
952 		struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
953 
954 		if (page_is_mergeable(bv, page, len, off, same_page)) {
955 			if (bio->bi_iter.bi_size > UINT_MAX - len) {
956 				*same_page = false;
957 				return false;
958 			}
959 			bv->bv_len += len;
960 			bio->bi_iter.bi_size += len;
961 			return true;
962 		}
963 	}
964 	return false;
965 }
966 
967 /*
968  * Try to merge a page into a segment, while obeying the hardware segment
969  * size limit.  This is not for normal read/write bios, but for passthrough
970  * or Zone Append operations that we can't split.
971  */
bio_try_merge_hw_seg(struct request_queue * q,struct bio * bio,struct page * page,unsigned len,unsigned offset,bool * same_page)972 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 	struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
977 	unsigned long mask = queue_segment_boundary(q);
978 	phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset;
979 	phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;
980 
981 	if ((addr1 | mask) != (addr2 | mask))
982 		return false;
983 	if (bv->bv_len + len > queue_max_segment_size(q))
984 		return false;
985 	return __bio_try_merge_page(bio, page, len, offset, same_page);
986 }
987 
988 /**
989  * bio_add_hw_page - attempt to add a page to a bio with hw constraints
990  * @q: the target queue
991  * @bio: destination bio
992  * @page: page to add
993  * @len: vec entry length
994  * @offset: vec entry offset
995  * @max_sectors: maximum number of sectors that can be added
996  * @same_page: return if the segment has been merged inside the same page
997  *
998  * Add a page to a bio while respecting the hardware max_sectors, max_segment
999  * and gap limitations.
1000  */
bio_add_hw_page(struct request_queue * q,struct bio * bio,struct page * page,unsigned int len,unsigned int offset,unsigned int max_sectors,bool * same_page)1001 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 	if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
1008 		return 0;
1009 
1010 	if (((bio->bi_iter.bi_size + len) >> 9) > max_sectors)
1011 		return 0;
1012 
1013 	if (bio->bi_vcnt > 0) {
1014 		if (bio_try_merge_hw_seg(q, bio, page, len, offset, same_page))
1015 			return len;
1016 
1017 		/*
1018 		 * If the queue doesn't support SG gaps and adding this segment
1019 		 * would create a gap, disallow it.
1020 		 */
1021 		bvec = &bio->bi_io_vec[bio->bi_vcnt - 1];
1022 		if (bvec_gap_to_prev(&q->limits, bvec, offset))
1023 			return 0;
1024 	}
1025 
1026 	if (bio_full(bio, len))
1027 		return 0;
1028 
1029 	if (bio->bi_vcnt >= queue_max_segments(q))
1030 		return 0;
1031 
1032 	bvec_set_page(&bio->bi_io_vec[bio->bi_vcnt], page, len, offset);
1033 	bio->bi_vcnt++;
1034 	bio->bi_iter.bi_size += len;
1035 	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  */
bio_add_pc_page(struct request_queue * q,struct bio * bio,struct page * page,unsigned int len,unsigned int offset)1053 int bio_add_pc_page(struct request_queue *q, struct bio *bio,
1054 		struct page *page, unsigned int len, unsigned int offset)
1055 {
1056 	bool same_page = false;
1057 	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  */
bio_add_zone_append_page(struct bio * bio,struct page * page,unsigned int len,unsigned int offset)1078 int bio_add_zone_append_page(struct bio *bio, struct page *page,
1079 			     unsigned int len, unsigned int offset)
1080 {
1081 	struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1082 	bool same_page = false;
1083 
1084 	if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND))
1085 		return 0;
1086 
1087 	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  */
__bio_add_page(struct bio * bio,struct page * page,unsigned int len,unsigned int off)1105 void __bio_add_page(struct bio *bio, struct page *page,
1106 		unsigned int len, unsigned int off)
1107 {
1108 	WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
1109 	WARN_ON_ONCE(bio_full(bio, len));
1110 
1111 	bvec_set_page(&bio->bi_io_vec[bio->bi_vcnt], page, len, off);
1112 	bio->bi_iter.bi_size += len;
1113 	bio->bi_vcnt++;
1114 }
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  */
bio_add_page(struct bio * bio,struct page * page,unsigned int len,unsigned int offset)1127 int bio_add_page(struct bio *bio, struct page *page,
1128 		 unsigned int len, unsigned int offset)
1129 {
1130 	bool same_page = false;
1131 
1132 	if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
1133 		if (bio_full(bio, len))
1134 			return 0;
1135 		__bio_add_page(bio, page, len, offset);
1136 	}
1137 	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  */
bio_add_folio(struct bio * bio,struct folio * folio,size_t len,size_t off)1155 bool bio_add_folio(struct bio *bio, struct folio *folio, size_t len,
1156 		   size_t off)
1157 {
1158 	if (len > UINT_MAX || off > UINT_MAX)
1159 		return false;
1160 	return bio_add_page(bio, &folio->page, len, off) > 0;
1161 }
1162 
__bio_release_pages(struct bio * bio,bool mark_dirty)1163 void __bio_release_pages(struct bio *bio, bool mark_dirty)
1164 {
1165 	struct bvec_iter_all iter_all;
1166 	struct bio_vec *bvec;
1167 
1168 	bio_for_each_segment_all(bvec, bio, iter_all) {
1169 		if (mark_dirty && !PageCompound(bvec->bv_page))
1170 			set_page_dirty_lock(bvec->bv_page);
1171 		put_page(bvec->bv_page);
1172 	}
1173 }
1174 EXPORT_SYMBOL_GPL(__bio_release_pages);
1175 
bio_iov_bvec_set(struct bio * bio,struct iov_iter * iter)1176 void bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter)
1177 {
1178 	size_t size = iov_iter_count(iter);
1179 
1180 	WARN_ON_ONCE(bio->bi_max_vecs);
1181 
1182 	if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1183 		struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1184 		size_t max_sectors = queue_max_zone_append_sectors(q);
1185 
1186 		size = min(size, max_sectors << SECTOR_SHIFT);
1187 	}
1188 
1189 	bio->bi_vcnt = iter->nr_segs;
1190 	bio->bi_io_vec = (struct bio_vec *)iter->bvec;
1191 	bio->bi_iter.bi_bvec_done = iter->iov_offset;
1192 	bio->bi_iter.bi_size = size;
1193 	bio_set_flag(bio, BIO_NO_PAGE_REF);
1194 	bio_set_flag(bio, BIO_CLONED);
1195 }
1196 
bio_iov_add_page(struct bio * bio,struct page * page,unsigned int len,unsigned int offset)1197 static int bio_iov_add_page(struct bio *bio, struct page *page,
1198 		unsigned int len, unsigned int offset)
1199 {
1200 	bool same_page = false;
1201 
1202 	if (!__bio_try_merge_page(bio, page, len, offset, &same_page)) {
1203 		__bio_add_page(bio, page, len, offset);
1204 		return 0;
1205 	}
1206 
1207 	if (same_page)
1208 		put_page(page);
1209 	return 0;
1210 }
1211 
bio_iov_add_zone_append_page(struct bio * bio,struct page * page,unsigned int len,unsigned int offset)1212 static int bio_iov_add_zone_append_page(struct bio *bio, struct page *page,
1213 		unsigned int len, unsigned int offset)
1214 {
1215 	struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1216 	bool same_page = false;
1217 
1218 	if (bio_add_hw_page(q, bio, page, len, offset,
1219 			queue_max_zone_append_sectors(q), &same_page) != len)
1220 		return -EINVAL;
1221 	if (same_page)
1222 		put_page(page);
1223 	return 0;
1224 }
1225 
1226 #define PAGE_PTRS_PER_BVEC     (sizeof(struct bio_vec) / sizeof(struct page *))
1227 
1228 /**
1229  * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
1230  * @bio: bio to add pages to
1231  * @iter: iov iterator describing the region to be mapped
1232  *
1233  * Pins pages from *iter and appends them to @bio's bvec array. The
1234  * pages will have to be released using put_page() when done.
1235  * For multi-segment *iter, this function only adds pages from the
1236  * next non-empty segment of the iov iterator.
1237  */
__bio_iov_iter_get_pages(struct bio * bio,struct iov_iter * iter)1238 static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1239 {
1240 	iov_iter_extraction_t extraction_flags = 0;
1241 	unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1242 	unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1243 	struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1244 	struct page **pages = (struct page **)bv;
1245 	ssize_t size, left;
1246 	unsigned len, i = 0;
1247 	size_t offset, trim;
1248 	int ret = 0;
1249 
1250 	/*
1251 	 * Move page array up in the allocated memory for the bio vecs as far as
1252 	 * possible so that we can start filling biovecs from the beginning
1253 	 * without overwriting the temporary page array.
1254 	 */
1255 	BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1256 	pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1257 
1258 	if (bio->bi_bdev && blk_queue_pci_p2pdma(bio->bi_bdev->bd_disk->queue))
1259 		extraction_flags |= ITER_ALLOW_P2PDMA;
1260 
1261 	/*
1262 	 * Each segment in the iov is required to be a block size multiple.
1263 	 * However, we may not be able to get the entire segment if it spans
1264 	 * more pages than bi_max_vecs allows, so we have to ALIGN_DOWN the
1265 	 * result to ensure the bio's total size is correct. The remainder of
1266 	 * the iov data will be picked up in the next bio iteration.
1267 	 */
1268 	size = iov_iter_get_pages(iter, pages,
1269 				  UINT_MAX - bio->bi_iter.bi_size,
1270 				  nr_pages, &offset, extraction_flags);
1271 	if (unlikely(size <= 0))
1272 		return size ? size : -EFAULT;
1273 
1274 	nr_pages = DIV_ROUND_UP(offset + size, PAGE_SIZE);
1275 
1276 	trim = size & (bdev_logical_block_size(bio->bi_bdev) - 1);
1277 	iov_iter_revert(iter, trim);
1278 
1279 	size -= trim;
1280 	if (unlikely(!size)) {
1281 		ret = -EFAULT;
1282 		goto out;
1283 	}
1284 
1285 	for (left = size, i = 0; left > 0; left -= len, i++) {
1286 		struct page *page = pages[i];
1287 
1288 		len = min_t(size_t, PAGE_SIZE - offset, left);
1289 		if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1290 			ret = bio_iov_add_zone_append_page(bio, page, len,
1291 					offset);
1292 			if (ret)
1293 				break;
1294 		} else
1295 			bio_iov_add_page(bio, page, len, offset);
1296 
1297 		offset = 0;
1298 	}
1299 
1300 	iov_iter_revert(iter, left);
1301 out:
1302 	while (i < nr_pages)
1303 		put_page(pages[i++]);
1304 
1305 	return ret;
1306 }
1307 
1308 /**
1309  * bio_iov_iter_get_pages - add user or kernel pages to a bio
1310  * @bio: bio to add pages to
1311  * @iter: iov iterator describing the region to be added
1312  *
1313  * This takes either an iterator pointing to user memory, or one pointing to
1314  * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1315  * map them into the kernel. On IO completion, the caller should put those
1316  * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided
1317  * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs
1318  * to ensure the bvecs and pages stay referenced until the submitted I/O is
1319  * completed by a call to ->ki_complete() or returns with an error other than
1320  * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF
1321  * on IO completion. If it isn't, then pages should be released.
1322  *
1323  * The function tries, but does not guarantee, to pin as many pages as
1324  * fit into the bio, or are requested in @iter, whatever is smaller. If
1325  * MM encounters an error pinning the requested pages, it stops. Error
1326  * is returned only if 0 pages could be pinned.
1327  */
bio_iov_iter_get_pages(struct bio * bio,struct iov_iter * iter)1328 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1329 {
1330 	int ret = 0;
1331 
1332 	if (iov_iter_is_bvec(iter)) {
1333 		bio_iov_bvec_set(bio, iter);
1334 		iov_iter_advance(iter, bio->bi_iter.bi_size);
1335 		return 0;
1336 	}
1337 
1338 	do {
1339 		ret = __bio_iov_iter_get_pages(bio, iter);
1340 	} while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));
1341 
1342 	return bio->bi_vcnt ? 0 : ret;
1343 }
1344 EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
1345 
submit_bio_wait_endio(struct bio * bio)1346 static void submit_bio_wait_endio(struct bio *bio)
1347 {
1348 	complete(bio->bi_private);
1349 }
1350 
1351 /**
1352  * submit_bio_wait - submit a bio, and wait until it completes
1353  * @bio: The &struct bio which describes the I/O
1354  *
1355  * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1356  * bio_endio() on failure.
1357  *
1358  * WARNING: Unlike to how submit_bio() is usually used, this function does not
1359  * result in bio reference to be consumed. The caller must drop the reference
1360  * on his own.
1361  */
submit_bio_wait(struct bio * bio)1362 int submit_bio_wait(struct bio *bio)
1363 {
1364 	DECLARE_COMPLETION_ONSTACK_MAP(done,
1365 			bio->bi_bdev->bd_disk->lockdep_map);
1366 	unsigned long hang_check;
1367 
1368 	bio->bi_private = &done;
1369 	bio->bi_end_io = submit_bio_wait_endio;
1370 	bio->bi_opf |= REQ_SYNC;
1371 	submit_bio(bio);
1372 
1373 	/* Prevent hang_check timer from firing at us during very long I/O */
1374 	hang_check = sysctl_hung_task_timeout_secs;
1375 	if (hang_check)
1376 		while (!wait_for_completion_io_timeout(&done,
1377 					hang_check * (HZ/2)))
1378 			;
1379 	else
1380 		wait_for_completion_io(&done);
1381 
1382 	return blk_status_to_errno(bio->bi_status);
1383 }
1384 EXPORT_SYMBOL(submit_bio_wait);
1385 
__bio_advance(struct bio * bio,unsigned bytes)1386 void __bio_advance(struct bio *bio, unsigned bytes)
1387 {
1388 	if (bio_integrity(bio))
1389 		bio_integrity_advance(bio, bytes);
1390 
1391 	bio_crypt_advance(bio, bytes);
1392 	bio_advance_iter(bio, &bio->bi_iter, bytes);
1393 }
1394 EXPORT_SYMBOL(__bio_advance);
1395 
bio_copy_data_iter(struct bio * dst,struct bvec_iter * dst_iter,struct bio * src,struct bvec_iter * src_iter)1396 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1397 			struct bio *src, struct bvec_iter *src_iter)
1398 {
1399 	while (src_iter->bi_size && dst_iter->bi_size) {
1400 		struct bio_vec src_bv = bio_iter_iovec(src, *src_iter);
1401 		struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter);
1402 		unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len);
1403 		void *src_buf = bvec_kmap_local(&src_bv);
1404 		void *dst_buf = bvec_kmap_local(&dst_bv);
1405 
1406 		memcpy(dst_buf, src_buf, bytes);
1407 
1408 		kunmap_local(dst_buf);
1409 		kunmap_local(src_buf);
1410 
1411 		bio_advance_iter_single(src, src_iter, bytes);
1412 		bio_advance_iter_single(dst, dst_iter, bytes);
1413 	}
1414 }
1415 EXPORT_SYMBOL(bio_copy_data_iter);
1416 
1417 /**
1418  * bio_copy_data - copy contents of data buffers from one bio to another
1419  * @src: source bio
1420  * @dst: destination bio
1421  *
1422  * Stops when it reaches the end of either @src or @dst - that is, copies
1423  * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1424  */
bio_copy_data(struct bio * dst,struct bio * src)1425 void bio_copy_data(struct bio *dst, struct bio *src)
1426 {
1427 	struct bvec_iter src_iter = src->bi_iter;
1428 	struct bvec_iter dst_iter = dst->bi_iter;
1429 
1430 	bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1431 }
1432 EXPORT_SYMBOL(bio_copy_data);
1433 
bio_free_pages(struct bio * bio)1434 void bio_free_pages(struct bio *bio)
1435 {
1436 	struct bio_vec *bvec;
1437 	struct bvec_iter_all iter_all;
1438 
1439 	bio_for_each_segment_all(bvec, bio, iter_all)
1440 		__free_page(bvec->bv_page);
1441 }
1442 EXPORT_SYMBOL(bio_free_pages);
1443 
1444 /*
1445  * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1446  * for performing direct-IO in BIOs.
1447  *
1448  * The problem is that we cannot run set_page_dirty() from interrupt context
1449  * because the required locks are not interrupt-safe.  So what we can do is to
1450  * mark the pages dirty _before_ performing IO.  And in interrupt context,
1451  * check that the pages are still dirty.   If so, fine.  If not, redirty them
1452  * in process context.
1453  *
1454  * We special-case compound pages here: normally this means reads into hugetlb
1455  * pages.  The logic in here doesn't really work right for compound pages
1456  * because the VM does not uniformly chase down the head page in all cases.
1457  * But dirtiness of compound pages is pretty meaningless anyway: the VM doesn't
1458  * handle them at all.  So we skip compound pages here at an early stage.
1459  *
1460  * Note that this code is very hard to test under normal circumstances because
1461  * direct-io pins the pages with get_user_pages().  This makes
1462  * is_page_cache_freeable return false, and the VM will not clean the pages.
1463  * But other code (eg, flusher threads) could clean the pages if they are mapped
1464  * pagecache.
1465  *
1466  * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1467  * deferred bio dirtying paths.
1468  */
1469 
1470 /*
1471  * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1472  */
bio_set_pages_dirty(struct bio * bio)1473 void bio_set_pages_dirty(struct bio *bio)
1474 {
1475 	struct bio_vec *bvec;
1476 	struct bvec_iter_all iter_all;
1477 
1478 	bio_for_each_segment_all(bvec, bio, iter_all) {
1479 		if (!PageCompound(bvec->bv_page))
1480 			set_page_dirty_lock(bvec->bv_page);
1481 	}
1482 }
1483 
1484 /*
1485  * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1486  * If they are, then fine.  If, however, some pages are clean then they must
1487  * have been written out during the direct-IO read.  So we take another ref on
1488  * the BIO and re-dirty the pages in process context.
1489  *
1490  * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1491  * here on.  It will run one put_page() against each page and will run one
1492  * bio_put() against the BIO.
1493  */
1494 
1495 static void bio_dirty_fn(struct work_struct *work);
1496 
1497 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1498 static DEFINE_SPINLOCK(bio_dirty_lock);
1499 static struct bio *bio_dirty_list;
1500 
1501 /*
1502  * This runs in process context
1503  */
bio_dirty_fn(struct work_struct * work)1504 static void bio_dirty_fn(struct work_struct *work)
1505 {
1506 	struct bio *bio, *next;
1507 
1508 	spin_lock_irq(&bio_dirty_lock);
1509 	next = bio_dirty_list;
1510 	bio_dirty_list = NULL;
1511 	spin_unlock_irq(&bio_dirty_lock);
1512 
1513 	while ((bio = next) != NULL) {
1514 		next = bio->bi_private;
1515 
1516 		bio_release_pages(bio, true);
1517 		bio_put(bio);
1518 	}
1519 }
1520 
bio_check_pages_dirty(struct bio * bio)1521 void bio_check_pages_dirty(struct bio *bio)
1522 {
1523 	struct bio_vec *bvec;
1524 	unsigned long flags;
1525 	struct bvec_iter_all iter_all;
1526 
1527 	bio_for_each_segment_all(bvec, bio, iter_all) {
1528 		if (!PageDirty(bvec->bv_page) && !PageCompound(bvec->bv_page))
1529 			goto defer;
1530 	}
1531 
1532 	bio_release_pages(bio, false);
1533 	bio_put(bio);
1534 	return;
1535 defer:
1536 	spin_lock_irqsave(&bio_dirty_lock, flags);
1537 	bio->bi_private = bio_dirty_list;
1538 	bio_dirty_list = bio;
1539 	spin_unlock_irqrestore(&bio_dirty_lock, flags);
1540 	schedule_work(&bio_dirty_work);
1541 }
1542 
bio_remaining_done(struct bio * bio)1543 static inline bool bio_remaining_done(struct bio *bio)
1544 {
1545 	/*
1546 	 * If we're not chaining, then ->__bi_remaining is always 1 and
1547 	 * we always end io on the first invocation.
1548 	 */
1549 	if (!bio_flagged(bio, BIO_CHAIN))
1550 		return true;
1551 
1552 	BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1553 
1554 	if (atomic_dec_and_test(&bio->__bi_remaining)) {
1555 		bio_clear_flag(bio, BIO_CHAIN);
1556 		return true;
1557 	}
1558 
1559 	return false;
1560 }
1561 
1562 /**
1563  * bio_endio - end I/O on a bio
1564  * @bio:	bio
1565  *
1566  * Description:
1567  *   bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1568  *   way to end I/O on a bio. No one should call bi_end_io() directly on a
1569  *   bio unless they own it and thus know that it has an end_io function.
1570  *
1571  *   bio_endio() can be called several times on a bio that has been chained
1572  *   using bio_chain().  The ->bi_end_io() function will only be called the
1573  *   last time.
1574  **/
bio_endio(struct bio * bio)1575 void bio_endio(struct bio *bio)
1576 {
1577 again:
1578 	if (!bio_remaining_done(bio))
1579 		return;
1580 	if (!bio_integrity_endio(bio))
1581 		return;
1582 
1583 	rq_qos_done_bio(bio);
1584 
1585 	if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1586 		trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio);
1587 		bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1588 	}
1589 
1590 	/*
1591 	 * Need to have a real endio function for chained bios, otherwise
1592 	 * various corner cases will break (like stacking block devices that
1593 	 * save/restore bi_end_io) - however, we want to avoid unbounded
1594 	 * recursion and blowing the stack. Tail call optimization would
1595 	 * handle this, but compiling with frame pointers also disables
1596 	 * gcc's sibling call optimization.
1597 	 */
1598 	if (bio->bi_end_io == bio_chain_endio) {
1599 		bio = __bio_chain_endio(bio);
1600 		goto again;
1601 	}
1602 
1603 	blk_throtl_bio_endio(bio);
1604 	/* release cgroup info */
1605 	bio_uninit(bio);
1606 	if (bio->bi_end_io)
1607 		bio->bi_end_io(bio);
1608 }
1609 EXPORT_SYMBOL(bio_endio);
1610 
1611 /**
1612  * bio_split - split a bio
1613  * @bio:	bio to split
1614  * @sectors:	number of sectors to split from the front of @bio
1615  * @gfp:	gfp mask
1616  * @bs:		bio set to allocate from
1617  *
1618  * Allocates and returns a new bio which represents @sectors from the start of
1619  * @bio, and updates @bio to represent the remaining sectors.
1620  *
1621  * Unless this is a discard request the newly allocated bio will point
1622  * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1623  * neither @bio nor @bs are freed before the split bio.
1624  */
bio_split(struct bio * bio,int sectors,gfp_t gfp,struct bio_set * bs)1625 struct bio *bio_split(struct bio *bio, int sectors,
1626 		      gfp_t gfp, struct bio_set *bs)
1627 {
1628 	struct bio *split;
1629 
1630 	BUG_ON(sectors <= 0);
1631 	BUG_ON(sectors >= bio_sectors(bio));
1632 
1633 	/* Zone append commands cannot be split */
1634 	if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))
1635 		return NULL;
1636 
1637 	split = bio_alloc_clone(bio->bi_bdev, bio, gfp, bs);
1638 	if (!split)
1639 		return NULL;
1640 
1641 	split->bi_iter.bi_size = sectors << 9;
1642 
1643 	if (bio_integrity(split))
1644 		bio_integrity_trim(split);
1645 
1646 	bio_advance(bio, split->bi_iter.bi_size);
1647 
1648 	if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1649 		bio_set_flag(split, BIO_TRACE_COMPLETION);
1650 
1651 	return split;
1652 }
1653 EXPORT_SYMBOL(bio_split);
1654 
1655 /**
1656  * bio_trim - trim a bio
1657  * @bio:	bio to trim
1658  * @offset:	number of sectors to trim from the front of @bio
1659  * @size:	size we want to trim @bio to, in sectors
1660  *
1661  * This function is typically used for bios that are cloned and submitted
1662  * to the underlying device in parts.
1663  */
bio_trim(struct bio * bio,sector_t offset,sector_t size)1664 void bio_trim(struct bio *bio, sector_t offset, sector_t size)
1665 {
1666 	if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS ||
1667 			 offset + size > bio_sectors(bio)))
1668 		return;
1669 
1670 	size <<= 9;
1671 	if (offset == 0 && size == bio->bi_iter.bi_size)
1672 		return;
1673 
1674 	bio_advance(bio, offset << 9);
1675 	bio->bi_iter.bi_size = size;
1676 
1677 	if (bio_integrity(bio))
1678 		bio_integrity_trim(bio);
1679 }
1680 EXPORT_SYMBOL_GPL(bio_trim);
1681 
1682 /*
1683  * create memory pools for biovec's in a bio_set.
1684  * use the global biovec slabs created for general use.
1685  */
biovec_init_pool(mempool_t * pool,int pool_entries)1686 int biovec_init_pool(mempool_t *pool, int pool_entries)
1687 {
1688 	struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1;
1689 
1690 	return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1691 }
1692 
1693 /*
1694  * bioset_exit - exit a bioset initialized with bioset_init()
1695  *
1696  * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1697  * kzalloc()).
1698  */
bioset_exit(struct bio_set * bs)1699 void bioset_exit(struct bio_set *bs)
1700 {
1701 	bio_alloc_cache_destroy(bs);
1702 	if (bs->rescue_workqueue)
1703 		destroy_workqueue(bs->rescue_workqueue);
1704 	bs->rescue_workqueue = NULL;
1705 
1706 	mempool_exit(&bs->bio_pool);
1707 	mempool_exit(&bs->bvec_pool);
1708 
1709 	bioset_integrity_free(bs);
1710 	if (bs->bio_slab)
1711 		bio_put_slab(bs);
1712 	bs->bio_slab = NULL;
1713 }
1714 EXPORT_SYMBOL(bioset_exit);
1715 
1716 /**
1717  * bioset_init - Initialize a bio_set
1718  * @bs:		pool to initialize
1719  * @pool_size:	Number of bio and bio_vecs to cache in the mempool
1720  * @front_pad:	Number of bytes to allocate in front of the returned bio
1721  * @flags:	Flags to modify behavior, currently %BIOSET_NEED_BVECS
1722  *              and %BIOSET_NEED_RESCUER
1723  *
1724  * Description:
1725  *    Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1726  *    to ask for a number of bytes to be allocated in front of the bio.
1727  *    Front pad allocation is useful for embedding the bio inside
1728  *    another structure, to avoid allocating extra data to go with the bio.
1729  *    Note that the bio must be embedded at the END of that structure always,
1730  *    or things will break badly.
1731  *    If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1732  *    for allocating iovecs.  This pool is not needed e.g. for bio_init_clone().
1733  *    If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used
1734  *    to dispatch queued requests when the mempool runs out of space.
1735  *
1736  */
bioset_init(struct bio_set * bs,unsigned int pool_size,unsigned int front_pad,int flags)1737 int bioset_init(struct bio_set *bs,
1738 		unsigned int pool_size,
1739 		unsigned int front_pad,
1740 		int flags)
1741 {
1742 	bs->front_pad = front_pad;
1743 	if (flags & BIOSET_NEED_BVECS)
1744 		bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1745 	else
1746 		bs->back_pad = 0;
1747 
1748 	spin_lock_init(&bs->rescue_lock);
1749 	bio_list_init(&bs->rescue_list);
1750 	INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1751 
1752 	bs->bio_slab = bio_find_or_create_slab(bs);
1753 	if (!bs->bio_slab)
1754 		return -ENOMEM;
1755 
1756 	if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
1757 		goto bad;
1758 
1759 	if ((flags & BIOSET_NEED_BVECS) &&
1760 	    biovec_init_pool(&bs->bvec_pool, pool_size))
1761 		goto bad;
1762 
1763 	if (flags & BIOSET_NEED_RESCUER) {
1764 		bs->rescue_workqueue = alloc_workqueue("bioset",
1765 							WQ_MEM_RECLAIM, 0);
1766 		if (!bs->rescue_workqueue)
1767 			goto bad;
1768 	}
1769 	if (flags & BIOSET_PERCPU_CACHE) {
1770 		bs->cache = alloc_percpu(struct bio_alloc_cache);
1771 		if (!bs->cache)
1772 			goto bad;
1773 		cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
1774 	}
1775 
1776 	return 0;
1777 bad:
1778 	bioset_exit(bs);
1779 	return -ENOMEM;
1780 }
1781 EXPORT_SYMBOL(bioset_init);
1782 
init_bio(void)1783 static int __init init_bio(void)
1784 {
1785 	int i;
1786 
1787 	BUILD_BUG_ON(BIO_FLAG_LAST > 8 * sizeof_field(struct bio, bi_flags));
1788 
1789 	bio_integrity_init();
1790 
1791 	for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) {
1792 		struct biovec_slab *bvs = bvec_slabs + i;
1793 
1794 		bvs->slab = kmem_cache_create(bvs->name,
1795 				bvs->nr_vecs * sizeof(struct bio_vec), 0,
1796 				SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1797 	}
1798 
1799 	cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL,
1800 					bio_cpu_dead);
1801 
1802 	if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0,
1803 			BIOSET_NEED_BVECS | BIOSET_PERCPU_CACHE))
1804 		panic("bio: can't allocate bios\n");
1805 
1806 	if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
1807 		panic("bio: can't create integrity pool\n");
1808 
1809 	return 0;
1810 }
1811 subsys_initcall(init_bio);
1812