xref: /linux/block/bio.c (revision f990ad67f0febc51274adb604d5bdeab0d06d024)
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-integrity.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/blk-crypto.h>
20 #include <linux/xarray.h>
21 
22 #include <trace/events/block.h>
23 #include "blk.h"
24 #include "blk-rq-qos.h"
25 #include "blk-cgroup.h"
26 
27 #define ALLOC_CACHE_THRESHOLD	16
28 #define ALLOC_CACHE_MAX		256
29 
30 struct bio_alloc_cache {
31 	struct bio		*free_list;
32 	struct bio		*free_list_irq;
33 	unsigned int		nr;
34 	unsigned int		nr_irq;
35 };
36 
37 #define BIO_INLINE_VECS 4
38 
39 static struct biovec_slab {
40 	int nr_vecs;
41 	char *name;
42 	struct kmem_cache *slab;
43 } bvec_slabs[] __read_mostly = {
44 	{ .nr_vecs = 16, .name = "biovec-16" },
45 	{ .nr_vecs = 64, .name = "biovec-64" },
46 	{ .nr_vecs = 128, .name = "biovec-128" },
47 	{ .nr_vecs = BIO_MAX_VECS, .name = "biovec-max" },
48 };
49 
50 static struct biovec_slab *biovec_slab(unsigned short nr_vecs)
51 {
52 	switch (nr_vecs) {
53 	/* smaller bios use inline vecs */
54 	case 5 ... 16:
55 		return &bvec_slabs[0];
56 	case 17 ... 64:
57 		return &bvec_slabs[1];
58 	case 65 ... 128:
59 		return &bvec_slabs[2];
60 	case 129 ... BIO_MAX_VECS:
61 		return &bvec_slabs[3];
62 	default:
63 		BUG();
64 		return NULL;
65 	}
66 }
67 
68 /*
69  * fs_bio_set is the bio_set containing bio and iovec memory pools used by
70  * IO code that does not need private memory pools.
71  */
72 struct bio_set fs_bio_set;
73 EXPORT_SYMBOL(fs_bio_set);
74 
75 /*
76  * Our slab pool management
77  */
78 struct bio_slab {
79 	struct kmem_cache *slab;
80 	unsigned int slab_ref;
81 	unsigned int slab_size;
82 	char name[12];
83 };
84 static DEFINE_MUTEX(bio_slab_lock);
85 static DEFINE_XARRAY(bio_slabs);
86 
87 static struct bio_slab *create_bio_slab(unsigned int size)
88 {
89 	struct bio_slab *bslab = kzalloc_obj(*bslab);
90 
91 	if (!bslab)
92 		return NULL;
93 
94 	snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size);
95 	bslab->slab = kmem_cache_create(bslab->name, size,
96 			ARCH_KMALLOC_MINALIGN,
97 			SLAB_HWCACHE_ALIGN | SLAB_TYPESAFE_BY_RCU, NULL);
98 	if (!bslab->slab)
99 		goto fail_alloc_slab;
100 
101 	bslab->slab_ref = 1;
102 	bslab->slab_size = size;
103 
104 	if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL)))
105 		return bslab;
106 
107 	kmem_cache_destroy(bslab->slab);
108 
109 fail_alloc_slab:
110 	kfree(bslab);
111 	return NULL;
112 }
113 
114 static inline unsigned int bs_bio_slab_size(struct bio_set *bs)
115 {
116 	return bs->front_pad + sizeof(struct bio) + bs->back_pad;
117 }
118 
119 static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs)
120 {
121 	unsigned int size = bs_bio_slab_size(bs);
122 	struct bio_slab *bslab;
123 
124 	mutex_lock(&bio_slab_lock);
125 	bslab = xa_load(&bio_slabs, size);
126 	if (bslab)
127 		bslab->slab_ref++;
128 	else
129 		bslab = create_bio_slab(size);
130 	mutex_unlock(&bio_slab_lock);
131 
132 	if (bslab)
133 		return bslab->slab;
134 	return NULL;
135 }
136 
137 static void bio_put_slab(struct bio_set *bs)
138 {
139 	struct bio_slab *bslab = NULL;
140 	unsigned int slab_size = bs_bio_slab_size(bs);
141 
142 	mutex_lock(&bio_slab_lock);
143 
144 	bslab = xa_load(&bio_slabs, slab_size);
145 	if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
146 		goto out;
147 
148 	WARN_ON_ONCE(bslab->slab != bs->bio_slab);
149 
150 	WARN_ON(!bslab->slab_ref);
151 
152 	if (--bslab->slab_ref)
153 		goto out;
154 
155 	xa_erase(&bio_slabs, slab_size);
156 
157 	kmem_cache_destroy(bslab->slab);
158 	kfree(bslab);
159 
160 out:
161 	mutex_unlock(&bio_slab_lock);
162 }
163 
164 /*
165  * Make the first allocation restricted and don't dump info on allocation
166  * failures, since we'll fall back to the mempool in case of failure.
167  */
168 static inline gfp_t try_alloc_gfp(gfp_t gfp)
169 {
170 	return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) |
171 		__GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
172 }
173 
174 void bio_uninit(struct bio *bio)
175 {
176 #ifdef CONFIG_BLK_CGROUP
177 	if (bio->bi_blkg) {
178 		blkg_put(bio->bi_blkg);
179 		bio->bi_blkg = NULL;
180 	}
181 #endif
182 	if (bio_integrity(bio))
183 		bio_integrity_free(bio);
184 
185 	bio_crypt_free_ctx(bio);
186 }
187 EXPORT_SYMBOL(bio_uninit);
188 
189 static void bio_free(struct bio *bio)
190 {
191 	struct bio_set *bs = bio->bi_pool;
192 	void *p = bio;
193 
194 	WARN_ON_ONCE(!bs);
195 	WARN_ON_ONCE(bio->bi_max_vecs > BIO_MAX_VECS);
196 
197 	bio_uninit(bio);
198 	if (bio->bi_max_vecs == BIO_MAX_VECS)
199 		mempool_free(bio->bi_io_vec, &bs->bvec_pool);
200 	else if (bio->bi_max_vecs > BIO_INLINE_VECS)
201 		kmem_cache_free(biovec_slab(bio->bi_max_vecs)->slab,
202 				bio->bi_io_vec);
203 	mempool_free(p - bs->front_pad, &bs->bio_pool);
204 }
205 
206 /*
207  * Users of this function have their own bio allocation. Subsequently,
208  * they must remember to pair any call to bio_init() with bio_uninit()
209  * when IO has completed, or when the bio is released.
210  */
211 void bio_init(struct bio *bio, struct block_device *bdev, struct bio_vec *table,
212 	      unsigned short max_vecs, blk_opf_t opf)
213 {
214 	bio->bi_next = NULL;
215 	bio->bi_bdev = bdev;
216 	bio->bi_opf = opf;
217 	bio->bi_flags = 0;
218 	bio->bi_ioprio = 0;
219 	bio->bi_write_hint = 0;
220 	bio->bi_write_stream = 0;
221 	bio->bi_status = 0;
222 	bio->bi_bvec_gap_bit = 0;
223 	bio->bi_iter.bi_sector = 0;
224 	bio->bi_iter.bi_size = 0;
225 	bio->bi_iter.bi_idx = 0;
226 	bio->bi_iter.bi_bvec_done = 0;
227 	bio->bi_end_io = NULL;
228 	bio->bi_private = NULL;
229 #ifdef CONFIG_BLK_CGROUP
230 	bio->bi_blkg = NULL;
231 	bio->issue_time_ns = 0;
232 	if (bdev)
233 		bio_associate_blkg(bio);
234 #ifdef CONFIG_BLK_CGROUP_IOCOST
235 	bio->bi_iocost_cost = 0;
236 #endif
237 #endif
238 #ifdef CONFIG_BLK_INLINE_ENCRYPTION
239 	bio->bi_crypt_context = NULL;
240 #endif
241 #ifdef CONFIG_BLK_DEV_INTEGRITY
242 	bio->bi_integrity = NULL;
243 #endif
244 	bio->bi_vcnt = 0;
245 
246 	atomic_set(&bio->__bi_remaining, 1);
247 	atomic_set(&bio->__bi_cnt, 1);
248 	bio->bi_cookie = BLK_QC_T_NONE;
249 
250 	bio->bi_max_vecs = max_vecs;
251 	bio->bi_io_vec = table;
252 	bio->bi_pool = NULL;
253 }
254 EXPORT_SYMBOL(bio_init);
255 
256 /**
257  * bio_reset - reinitialize a bio
258  * @bio:	bio to reset
259  * @bdev:	block device to use the bio for
260  * @opf:	operation and flags for bio
261  *
262  * Description:
263  *   After calling bio_reset(), @bio will be in the same state as a freshly
264  *   allocated bio returned bio bio_alloc_bioset() - the only fields that are
265  *   preserved are the ones that are initialized by bio_alloc_bioset(). See
266  *   comment in struct bio.
267  */
268 void bio_reset(struct bio *bio, struct block_device *bdev, blk_opf_t opf)
269 {
270 	struct bio_vec          *bv = bio->bi_io_vec;
271 
272 	bio_uninit(bio);
273 	memset(bio, 0, BIO_RESET_BYTES);
274 	atomic_set(&bio->__bi_remaining, 1);
275 	bio->bi_io_vec = bv;
276 	bio->bi_bdev = bdev;
277 	if (bio->bi_bdev)
278 		bio_associate_blkg(bio);
279 	bio->bi_opf = opf;
280 }
281 EXPORT_SYMBOL(bio_reset);
282 
283 /**
284  * bio_reuse - reuse a bio with the payload left intact
285  * @bio:	bio to reuse
286  * @opf:	operation and flags for the next I/O
287  *
288  * Allow reusing an existing bio for another operation with all set up
289  * fields including the payload, device and end_io handler left intact.
290  *
291  * Typically used when @bio is first used to read data which is then written
292  * to another location without modification.  @bio must not be in-flight and
293  * owned by the caller.  Can't be used for cloned bios.
294  *
295  * Note: Can't be used when @bio has integrity or blk-crypto contexts for now.
296  * Feel free to add that support when you need it, though.
297  */
298 void bio_reuse(struct bio *bio, blk_opf_t opf)
299 {
300 	unsigned short vcnt = bio->bi_vcnt, i;
301 	bio_end_io_t *end_io = bio->bi_end_io;
302 	void *private = bio->bi_private;
303 
304 	WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
305 	WARN_ON_ONCE(bio_integrity(bio));
306 	WARN_ON_ONCE(bio_has_crypt_ctx(bio));
307 
308 	bio_reset(bio, bio->bi_bdev, opf);
309 	for (i = 0; i < vcnt; i++)
310 		bio->bi_iter.bi_size += bio->bi_io_vec[i].bv_len;
311 	bio->bi_vcnt = vcnt;
312 	bio->bi_private = private;
313 	bio->bi_end_io = end_io;
314 }
315 EXPORT_SYMBOL_GPL(bio_reuse);
316 
317 static struct bio *__bio_chain_endio(struct bio *bio)
318 {
319 	struct bio *parent = bio->bi_private;
320 
321 	if (bio->bi_status && !parent->bi_status)
322 		parent->bi_status = bio->bi_status;
323 	bio_put(bio);
324 	return parent;
325 }
326 
327 /*
328  * This function should only be used as a flag and must never be called.
329  * If execution reaches here, it indicates a serious programming error.
330  */
331 static void bio_chain_endio(struct bio *bio)
332 {
333 	BUG();
334 }
335 
336 /**
337  * bio_chain - chain bio completions
338  * @bio: the target bio
339  * @parent: the parent bio of @bio
340  *
341  * The caller won't have a bi_end_io called when @bio completes - instead,
342  * @parent's bi_end_io won't be called until both @parent and @bio have
343  * completed; the chained bio will also be freed when it completes.
344  *
345  * The caller must not set bi_private or bi_end_io in @bio.
346  */
347 void bio_chain(struct bio *bio, struct bio *parent)
348 {
349 	BUG_ON(bio->bi_private || bio->bi_end_io);
350 
351 	bio->bi_private = parent;
352 	bio->bi_end_io	= bio_chain_endio;
353 	bio_inc_remaining(parent);
354 }
355 EXPORT_SYMBOL(bio_chain);
356 
357 /**
358  * bio_chain_and_submit - submit a bio after chaining it to another one
359  * @prev: bio to chain and submit
360  * @new: bio to chain to
361  *
362  * If @prev is non-NULL, chain it to @new and submit it.
363  *
364  * Return: @new.
365  */
366 struct bio *bio_chain_and_submit(struct bio *prev, struct bio *new)
367 {
368 	if (prev) {
369 		bio_chain(prev, new);
370 		submit_bio(prev);
371 	}
372 	return new;
373 }
374 
375 struct bio *blk_next_bio(struct bio *bio, struct block_device *bdev,
376 		unsigned int nr_pages, blk_opf_t opf, gfp_t gfp)
377 {
378 	return bio_chain_and_submit(bio, bio_alloc(bdev, nr_pages, opf, gfp));
379 }
380 EXPORT_SYMBOL_GPL(blk_next_bio);
381 
382 static void bio_alloc_rescue(struct work_struct *work)
383 {
384 	struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
385 	struct bio *bio;
386 
387 	while (1) {
388 		spin_lock(&bs->rescue_lock);
389 		bio = bio_list_pop(&bs->rescue_list);
390 		spin_unlock(&bs->rescue_lock);
391 
392 		if (!bio)
393 			break;
394 
395 		submit_bio_noacct(bio);
396 	}
397 }
398 
399 /*
400  * submit_bio_noacct() converts recursion to iteration; this means if we're
401  * running beneath it, any bios we allocate and submit will not be submitted
402  * (and thus freed) until after we return.
403  *
404  * This exposes us to a potential deadlock if we allocate multiple bios from the
405  * same bio_set while running underneath submit_bio_noacct().  If we were to
406  * allocate multiple bios (say a stacking block driver that was splitting bios),
407  * we would deadlock if we exhausted the mempool's reserve.
408  *
409  * We solve this, and guarantee forward progress by punting the bios on
410  * current->bio_list to a per bio_set rescuer workqueue before blocking to wait
411  * for elements being returned to the mempool.
412  */
413 static void punt_bios_to_rescuer(struct bio_set *bs)
414 {
415 	struct bio_list punt, nopunt;
416 	struct bio *bio;
417 
418 	if (!current->bio_list || !bs->rescue_workqueue)
419 		return;
420 	if (bio_list_empty(&current->bio_list[0]) &&
421 	    bio_list_empty(&current->bio_list[1]))
422 		return;
423 
424 	/*
425 	 * In order to guarantee forward progress we must punt only bios that
426 	 * were allocated from this bio_set; otherwise, if there was a bio on
427 	 * there for a stacking driver higher up in the stack, processing it
428 	 * could require allocating bios from this bio_set, and doing that from
429 	 * our own rescuer would be bad.
430 	 *
431 	 * Since bio lists are singly linked, pop them all instead of trying to
432 	 * remove from the middle of the list:
433 	 */
434 
435 	bio_list_init(&punt);
436 	bio_list_init(&nopunt);
437 
438 	while ((bio = bio_list_pop(&current->bio_list[0])))
439 		bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
440 	current->bio_list[0] = nopunt;
441 
442 	bio_list_init(&nopunt);
443 	while ((bio = bio_list_pop(&current->bio_list[1])))
444 		bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
445 	current->bio_list[1] = nopunt;
446 
447 	spin_lock(&bs->rescue_lock);
448 	bio_list_merge(&bs->rescue_list, &punt);
449 	spin_unlock(&bs->rescue_lock);
450 
451 	queue_work(bs->rescue_workqueue, &bs->rescue_work);
452 }
453 
454 static void bio_alloc_irq_cache_splice(struct bio_alloc_cache *cache)
455 {
456 	unsigned long flags;
457 
458 	/* cache->free_list must be empty */
459 	if (WARN_ON_ONCE(cache->free_list))
460 		return;
461 
462 	local_irq_save(flags);
463 	cache->free_list = cache->free_list_irq;
464 	cache->free_list_irq = NULL;
465 	cache->nr += cache->nr_irq;
466 	cache->nr_irq = 0;
467 	local_irq_restore(flags);
468 }
469 
470 static struct bio *bio_alloc_percpu_cache(struct bio_set *bs)
471 {
472 	struct bio_alloc_cache *cache;
473 	struct bio *bio;
474 
475 	cache = per_cpu_ptr(bs->cache, get_cpu());
476 	if (!cache->free_list) {
477 		if (READ_ONCE(cache->nr_irq) >= ALLOC_CACHE_THRESHOLD)
478 			bio_alloc_irq_cache_splice(cache);
479 		if (!cache->free_list) {
480 			put_cpu();
481 			return NULL;
482 		}
483 	}
484 	bio = cache->free_list;
485 	cache->free_list = bio->bi_next;
486 	cache->nr--;
487 	put_cpu();
488 	bio->bi_pool = bs;
489 	return bio;
490 }
491 
492 /**
493  * bio_alloc_bioset - allocate a bio for I/O
494  * @bdev:	block device to allocate the bio for (can be %NULL)
495  * @nr_vecs:	number of bvecs to pre-allocate
496  * @opf:	operation and flags for bio
497  * @gfp:	the GFP_* mask given to the slab allocator
498  * @bs:		the bio_set to allocate from.
499  *
500  * Allocate a bio from the mempools in @bs.
501  *
502  * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to
503  * allocate a bio.  This is due to the mempool guarantees.  To make this work,
504  * callers must never allocate more than 1 bio at a time from the general pool.
505  * Callers that need to allocate more than 1 bio must always submit the
506  * previously allocated bio for IO before attempting to allocate a new one.
507  * Failure to do so can cause deadlocks under memory pressure.
508  *
509  * Note that when running under submit_bio_noacct() (i.e. any block driver),
510  * bios are not submitted until after you return - see the code in
511  * submit_bio_noacct() that converts recursion into iteration, to prevent
512  * stack overflows.
513  *
514  * This would normally mean allocating multiple bios under submit_bio_noacct()
515  * would be susceptible to deadlocks, but we have
516  * deadlock avoidance code that resubmits any blocked bios from a rescuer
517  * thread.
518  *
519  * However, we do not guarantee forward progress for allocations from other
520  * mempools. Doing multiple allocations from the same mempool under
521  * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad
522  * for per bio allocations.
523  *
524  * Returns: Pointer to new bio on success, NULL on failure.
525  */
526 struct bio *bio_alloc_bioset(struct block_device *bdev, unsigned short nr_vecs,
527 			     blk_opf_t opf, gfp_t gfp, struct bio_set *bs)
528 {
529 	struct bio_vec *bvecs = NULL;
530 	struct bio *bio = NULL;
531 	gfp_t saved_gfp = gfp;
532 	void *p;
533 
534 	/* should not use nobvec bioset for nr_vecs > 0 */
535 	if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_vecs > 0))
536 		return NULL;
537 
538 	gfp = try_alloc_gfp(gfp);
539 	if (bs->cache && nr_vecs <= BIO_INLINE_VECS) {
540 		/*
541 		 * Set REQ_ALLOC_CACHE even if no cached bio is available to
542 		 * return the allocated bio to the percpu cache when done.
543 		 */
544 		opf |= REQ_ALLOC_CACHE;
545 		bio = bio_alloc_percpu_cache(bs);
546 	} else {
547 		opf &= ~REQ_ALLOC_CACHE;
548 		p = kmem_cache_alloc(bs->bio_slab, gfp);
549 		if (p)
550 			bio = p + bs->front_pad;
551 	}
552 
553 	if (bio && nr_vecs > BIO_INLINE_VECS) {
554 		struct biovec_slab *bvs = biovec_slab(nr_vecs);
555 
556 		/*
557 		 * Upgrade nr_vecs to take full advantage of the allocation.
558 		 * We also rely on this in bio_free().
559 		 */
560 		nr_vecs = bvs->nr_vecs;
561 		bvecs = kmem_cache_alloc(bvs->slab, gfp);
562 		if (unlikely(!bvecs)) {
563 			kmem_cache_free(bs->bio_slab, p);
564 			bio = NULL;
565 		}
566 	}
567 
568 	if (unlikely(!bio)) {
569 		/*
570 		 * Give up if we are not allow to sleep as non-blocking mempool
571 		 * allocations just go back to the slab allocation.
572 		 */
573 		if (!(saved_gfp & __GFP_DIRECT_RECLAIM))
574 			return NULL;
575 
576 		punt_bios_to_rescuer(bs);
577 
578 		/*
579 		 * Don't rob the mempools by returning to the per-CPU cache if
580 		 * we're tight on memory.
581 		 */
582 		opf &= ~REQ_ALLOC_CACHE;
583 
584 		p = mempool_alloc(&bs->bio_pool, saved_gfp);
585 		bio = p + bs->front_pad;
586 		if (nr_vecs > BIO_INLINE_VECS) {
587 			nr_vecs = BIO_MAX_VECS;
588 			bvecs = mempool_alloc(&bs->bvec_pool, saved_gfp);
589 		}
590 	}
591 
592 	if (nr_vecs && nr_vecs <= BIO_INLINE_VECS)
593 		bio_init_inline(bio, bdev, nr_vecs, opf);
594 	else
595 		bio_init(bio, bdev, bvecs, nr_vecs, opf);
596 	bio->bi_pool = bs;
597 	return bio;
598 }
599 EXPORT_SYMBOL(bio_alloc_bioset);
600 
601 /**
602  * bio_kmalloc - kmalloc a bio
603  * @nr_vecs:	number of bio_vecs to allocate
604  * @gfp_mask:   the GFP_* mask given to the slab allocator
605  *
606  * Use kmalloc to allocate a bio (including bvecs).  The bio must be initialized
607  * using bio_init() before use.  To free a bio returned from this function use
608  * kfree() after calling bio_uninit().  A bio returned from this function can
609  * be reused by calling bio_uninit() before calling bio_init() again.
610  *
611  * Note that unlike bio_alloc() or bio_alloc_bioset() allocations from this
612  * function are not backed by a mempool can fail.  Do not use this function
613  * for allocations in the file system I/O path.
614  *
615  * Returns: Pointer to new bio on success, NULL on failure.
616  */
617 struct bio *bio_kmalloc(unsigned short nr_vecs, gfp_t gfp_mask)
618 {
619 	struct bio *bio;
620 
621 	if (nr_vecs > BIO_MAX_INLINE_VECS)
622 		return NULL;
623 	return kmalloc(sizeof(*bio) + nr_vecs * sizeof(struct bio_vec),
624 			gfp_mask);
625 }
626 EXPORT_SYMBOL(bio_kmalloc);
627 
628 void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start)
629 {
630 	struct bio_vec bv;
631 	struct bvec_iter iter;
632 
633 	__bio_for_each_segment(bv, bio, iter, start)
634 		memzero_bvec(&bv);
635 }
636 EXPORT_SYMBOL(zero_fill_bio_iter);
637 
638 /**
639  * bio_truncate - truncate the bio to small size of @new_size
640  * @bio:	the bio to be truncated
641  * @new_size:	new size for truncating the bio
642  *
643  * Description:
644  *   Truncate the bio to new size of @new_size. If bio_op(bio) is
645  *   REQ_OP_READ, zero the truncated part. This function should only
646  *   be used for handling corner cases, such as bio eod.
647  */
648 static void bio_truncate(struct bio *bio, unsigned new_size)
649 {
650 	struct bio_vec bv;
651 	struct bvec_iter iter;
652 	unsigned int done = 0;
653 	bool truncated = false;
654 
655 	if (new_size >= bio->bi_iter.bi_size)
656 		return;
657 
658 	if (bio_op(bio) != REQ_OP_READ)
659 		goto exit;
660 
661 	bio_for_each_segment(bv, bio, iter) {
662 		if (done + bv.bv_len > new_size) {
663 			size_t offset;
664 
665 			if (!truncated)
666 				offset = new_size - done;
667 			else
668 				offset = 0;
669 			memzero_page(bv.bv_page, bv.bv_offset + offset,
670 				  bv.bv_len - offset);
671 			truncated = true;
672 		}
673 		done += bv.bv_len;
674 	}
675 
676  exit:
677 	/*
678 	 * Don't touch bvec table here and make it really immutable, since
679 	 * fs bio user has to retrieve all pages via bio_for_each_segment_all
680 	 * in its .end_bio() callback.
681 	 *
682 	 * It is enough to truncate bio by updating .bi_size since we can make
683 	 * correct bvec with the updated .bi_size for drivers.
684 	 */
685 	bio->bi_iter.bi_size = new_size;
686 }
687 
688 /**
689  * guard_bio_eod - truncate a BIO to fit the block device
690  * @bio:	bio to truncate
691  *
692  * This allows us to do IO even on the odd last sectors of a device, even if the
693  * block size is some multiple of the physical sector size.
694  *
695  * We'll just truncate the bio to the size of the device, and clear the end of
696  * the buffer head manually.  Truly out-of-range accesses will turn into actual
697  * I/O errors, this only handles the "we need to be able to do I/O at the final
698  * sector" case.
699  */
700 void guard_bio_eod(struct bio *bio)
701 {
702 	sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
703 
704 	if (!maxsector)
705 		return;
706 
707 	/*
708 	 * If the *whole* IO is past the end of the device,
709 	 * let it through, and the IO layer will turn it into
710 	 * an EIO.
711 	 */
712 	if (unlikely(bio->bi_iter.bi_sector >= maxsector))
713 		return;
714 
715 	maxsector -= bio->bi_iter.bi_sector;
716 	if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
717 		return;
718 
719 	bio_truncate(bio, maxsector << 9);
720 }
721 
722 static int __bio_alloc_cache_prune(struct bio_alloc_cache *cache,
723 				   unsigned int nr)
724 {
725 	unsigned int i = 0;
726 	struct bio *bio;
727 
728 	while ((bio = cache->free_list) != NULL) {
729 		cache->free_list = bio->bi_next;
730 		cache->nr--;
731 		bio_free(bio);
732 		if (++i == nr)
733 			break;
734 	}
735 	return i;
736 }
737 
738 static void bio_alloc_cache_prune(struct bio_alloc_cache *cache,
739 				  unsigned int nr)
740 {
741 	nr -= __bio_alloc_cache_prune(cache, nr);
742 	if (!READ_ONCE(cache->free_list)) {
743 		bio_alloc_irq_cache_splice(cache);
744 		__bio_alloc_cache_prune(cache, nr);
745 	}
746 }
747 
748 static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node)
749 {
750 	struct bio_set *bs;
751 
752 	bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead);
753 	if (bs->cache) {
754 		struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu);
755 
756 		bio_alloc_cache_prune(cache, -1U);
757 	}
758 	return 0;
759 }
760 
761 static void bio_alloc_cache_destroy(struct bio_set *bs)
762 {
763 	int cpu;
764 
765 	if (!bs->cache)
766 		return;
767 
768 	cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
769 	for_each_possible_cpu(cpu) {
770 		struct bio_alloc_cache *cache;
771 
772 		cache = per_cpu_ptr(bs->cache, cpu);
773 		bio_alloc_cache_prune(cache, -1U);
774 	}
775 	free_percpu(bs->cache);
776 	bs->cache = NULL;
777 }
778 
779 static inline void bio_put_percpu_cache(struct bio *bio)
780 {
781 	struct bio_alloc_cache *cache;
782 
783 	cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu());
784 	if (READ_ONCE(cache->nr_irq) + cache->nr > ALLOC_CACHE_MAX)
785 		goto out_free;
786 
787 	if (in_task()) {
788 		bio_uninit(bio);
789 		bio->bi_next = cache->free_list;
790 		/* Not necessary but helps not to iopoll already freed bios */
791 		bio->bi_bdev = NULL;
792 		cache->free_list = bio;
793 		cache->nr++;
794 	} else if (in_hardirq()) {
795 		lockdep_assert_irqs_disabled();
796 
797 		bio_uninit(bio);
798 		bio->bi_next = cache->free_list_irq;
799 		cache->free_list_irq = bio;
800 		cache->nr_irq++;
801 	} else {
802 		goto out_free;
803 	}
804 	put_cpu();
805 	return;
806 out_free:
807 	put_cpu();
808 	bio_free(bio);
809 }
810 
811 /**
812  * bio_put - release a reference to a bio
813  * @bio:   bio to release reference to
814  *
815  * Description:
816  *   Put a reference to a &struct bio, either one you have gotten with
817  *   bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
818  **/
819 void bio_put(struct bio *bio)
820 {
821 	if (unlikely(bio_flagged(bio, BIO_REFFED))) {
822 		BUG_ON(!atomic_read(&bio->__bi_cnt));
823 		if (!atomic_dec_and_test(&bio->__bi_cnt))
824 			return;
825 	}
826 	if (bio->bi_opf & REQ_ALLOC_CACHE)
827 		bio_put_percpu_cache(bio);
828 	else
829 		bio_free(bio);
830 }
831 EXPORT_SYMBOL(bio_put);
832 
833 static int __bio_clone(struct bio *bio, struct bio *bio_src, gfp_t gfp)
834 {
835 	bio_set_flag(bio, BIO_CLONED);
836 	bio->bi_ioprio = bio_src->bi_ioprio;
837 	bio->bi_write_hint = bio_src->bi_write_hint;
838 	bio->bi_write_stream = bio_src->bi_write_stream;
839 	bio->bi_iter = bio_src->bi_iter;
840 
841 	if (bio->bi_bdev) {
842 		if (bio->bi_bdev == bio_src->bi_bdev &&
843 		    bio_flagged(bio_src, BIO_REMAPPED))
844 			bio_set_flag(bio, BIO_REMAPPED);
845 		bio_clone_blkg_association(bio, bio_src);
846 	}
847 
848 	if (bio_crypt_clone(bio, bio_src, gfp) < 0)
849 		return -ENOMEM;
850 	if (bio_integrity(bio_src) &&
851 	    bio_integrity_clone(bio, bio_src, gfp) < 0)
852 		return -ENOMEM;
853 	return 0;
854 }
855 
856 /**
857  * bio_alloc_clone - clone a bio that shares the original bio's biovec
858  * @bdev: block_device to clone onto
859  * @bio_src: bio to clone from
860  * @gfp: allocation priority
861  * @bs: bio_set to allocate from
862  *
863  * Allocate a new bio that is a clone of @bio_src. This reuses the bio_vecs
864  * pointed to by @bio_src->bi_io_vec, and clones the iterator pointing to
865  * the current position in it.  The caller owns the returned bio, but not
866  * the bio_vecs, and must ensure the bio is freed before the memory
867  * pointed to by @bio_Src->bi_io_vecs.
868  */
869 struct bio *bio_alloc_clone(struct block_device *bdev, struct bio *bio_src,
870 		gfp_t gfp, struct bio_set *bs)
871 {
872 	struct bio *bio;
873 
874 	bio = bio_alloc_bioset(bdev, 0, bio_src->bi_opf, gfp, bs);
875 	if (!bio)
876 		return NULL;
877 
878 	if (__bio_clone(bio, bio_src, gfp) < 0) {
879 		bio_put(bio);
880 		return NULL;
881 	}
882 	bio->bi_io_vec = bio_src->bi_io_vec;
883 
884 	return bio;
885 }
886 EXPORT_SYMBOL(bio_alloc_clone);
887 
888 /**
889  * bio_init_clone - clone a bio that shares the original bio's biovec
890  * @bdev: block_device to clone onto
891  * @bio: bio to clone into
892  * @bio_src: bio to clone from
893  * @gfp: allocation priority
894  *
895  * Initialize a new bio in caller provided memory that is a clone of @bio_src.
896  * The same bio_vecs reuse and bio lifetime rules as bio_alloc_clone() apply.
897  */
898 int bio_init_clone(struct block_device *bdev, struct bio *bio,
899 		struct bio *bio_src, gfp_t gfp)
900 {
901 	int ret;
902 
903 	bio_init(bio, bdev, bio_src->bi_io_vec, 0, bio_src->bi_opf);
904 	ret = __bio_clone(bio, bio_src, gfp);
905 	if (ret)
906 		bio_uninit(bio);
907 	return ret;
908 }
909 EXPORT_SYMBOL(bio_init_clone);
910 
911 /**
912  * bio_full - check if the bio is full
913  * @bio:	bio to check
914  * @len:	length of one segment to be added
915  *
916  * Return true if @bio is full and one segment with @len bytes can't be
917  * added to the bio, otherwise return false
918  */
919 static inline bool bio_full(struct bio *bio, unsigned len)
920 {
921 	if (bio->bi_vcnt >= bio->bi_max_vecs)
922 		return true;
923 	if (bio->bi_iter.bi_size > BIO_MAX_SIZE - len)
924 		return true;
925 	return false;
926 }
927 
928 static bool bvec_try_merge_page(struct bio_vec *bv, struct page *page,
929 		unsigned int len, unsigned int off)
930 {
931 	size_t bv_end = bv->bv_offset + bv->bv_len;
932 	phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1;
933 	phys_addr_t page_addr = page_to_phys(page);
934 
935 	if (vec_end_addr + 1 != page_addr + off)
936 		return false;
937 	if (xen_domain() && !xen_biovec_phys_mergeable(bv, page))
938 		return false;
939 
940 	if ((vec_end_addr & PAGE_MASK) != ((page_addr + off) & PAGE_MASK)) {
941 		if (IS_ENABLED(CONFIG_KMSAN))
942 			return false;
943 		if (bv->bv_page + bv_end / PAGE_SIZE != page + off / PAGE_SIZE)
944 			return false;
945 	}
946 
947 	bv->bv_len += len;
948 	return true;
949 }
950 
951 /*
952  * Try to merge a page into a segment, while obeying the hardware segment
953  * size limit.
954  *
955  * This is kept around for the integrity metadata, which is still tries
956  * to build the initial bio to the hardware limit and doesn't have proper
957  * helpers to split.  Hopefully this will go away soon.
958  */
959 bool bvec_try_merge_hw_page(struct request_queue *q, struct bio_vec *bv,
960 		struct page *page, unsigned len, unsigned offset)
961 {
962 	unsigned long mask = queue_segment_boundary(q);
963 	phys_addr_t addr1 = bvec_phys(bv);
964 	phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;
965 
966 	if ((addr1 | mask) != (addr2 | mask))
967 		return false;
968 	if (len > queue_max_segment_size(q) - bv->bv_len)
969 		return false;
970 	return bvec_try_merge_page(bv, page, len, offset);
971 }
972 
973 /**
974  * __bio_add_page - add page(s) to a bio in a new segment
975  * @bio: destination bio
976  * @page: start page to add
977  * @len: length of the data to add, may cross pages
978  * @off: offset of the data relative to @page, may cross pages
979  *
980  * Add the data at @page + @off to @bio as a new bvec.  The caller must ensure
981  * that @bio has space for another bvec.
982  */
983 void __bio_add_page(struct bio *bio, struct page *page,
984 		unsigned int len, unsigned int off)
985 {
986 	WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
987 	WARN_ON_ONCE(bio_full(bio, len));
988 
989 	if (is_pci_p2pdma_page(page))
990 		bio->bi_opf |= REQ_NOMERGE;
991 
992 	bvec_set_page(&bio->bi_io_vec[bio->bi_vcnt], page, len, off);
993 	bio->bi_iter.bi_size += len;
994 	bio->bi_vcnt++;
995 }
996 EXPORT_SYMBOL_GPL(__bio_add_page);
997 
998 /**
999  * bio_add_virt_nofail - add data in the direct kernel mapping to a bio
1000  * @bio: destination bio
1001  * @vaddr: data to add
1002  * @len: length of the data to add, may cross pages
1003  *
1004  * Add the data at @vaddr to @bio.  The caller must have ensure a segment
1005  * is available for the added data.  No merging into an existing segment
1006  * will be performed.
1007  */
1008 void bio_add_virt_nofail(struct bio *bio, void *vaddr, unsigned len)
1009 {
1010 	__bio_add_page(bio, virt_to_page(vaddr), len, offset_in_page(vaddr));
1011 }
1012 EXPORT_SYMBOL_GPL(bio_add_virt_nofail);
1013 
1014 /**
1015  *	bio_add_page	-	attempt to add page(s) to bio
1016  *	@bio: destination bio
1017  *	@page: start page to add
1018  *	@len: vec entry length, may cross pages
1019  *	@offset: vec entry offset relative to @page, may cross pages
1020  *
1021  *	Attempt to add page(s) to the bio_vec maplist. This will only fail
1022  *	if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
1023  */
1024 int bio_add_page(struct bio *bio, struct page *page,
1025 		 unsigned int len, unsigned int offset)
1026 {
1027 	if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
1028 		return 0;
1029 	if (WARN_ON_ONCE(len == 0))
1030 		return 0;
1031 	if (bio->bi_iter.bi_size > BIO_MAX_SIZE - len)
1032 		return 0;
1033 
1034 	if (bio->bi_vcnt > 0) {
1035 		struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
1036 
1037 		if (!zone_device_pages_have_same_pgmap(bv->bv_page, page))
1038 			return 0;
1039 
1040 		if (bvec_try_merge_page(bv, page, len, offset)) {
1041 			bio->bi_iter.bi_size += len;
1042 			return len;
1043 		}
1044 	}
1045 
1046 	if (bio->bi_vcnt >= bio->bi_max_vecs)
1047 		return 0;
1048 	__bio_add_page(bio, page, len, offset);
1049 	return len;
1050 }
1051 EXPORT_SYMBOL(bio_add_page);
1052 
1053 void bio_add_folio_nofail(struct bio *bio, struct folio *folio, size_t len,
1054 			  size_t off)
1055 {
1056 	unsigned long nr = off / PAGE_SIZE;
1057 
1058 	WARN_ON_ONCE(len > BIO_MAX_SIZE);
1059 	__bio_add_page(bio, folio_page(folio, nr), len, off % PAGE_SIZE);
1060 }
1061 EXPORT_SYMBOL_GPL(bio_add_folio_nofail);
1062 
1063 /**
1064  * bio_add_folio - Attempt to add part of a folio to a bio.
1065  * @bio: BIO to add to.
1066  * @folio: Folio to add.
1067  * @len: How many bytes from the folio to add.
1068  * @off: First byte in this folio to add.
1069  *
1070  * Filesystems that use folios can call this function instead of calling
1071  * bio_add_page() for each page in the folio.  If @off is bigger than
1072  * PAGE_SIZE, this function can create a bio_vec that starts in a page
1073  * after the bv_page.  BIOs do not support folios that are 4GiB or larger.
1074  *
1075  * Return: Whether the addition was successful.
1076  */
1077 bool bio_add_folio(struct bio *bio, struct folio *folio, size_t len,
1078 		   size_t off)
1079 {
1080 	unsigned long nr = off / PAGE_SIZE;
1081 
1082 	if (len > BIO_MAX_SIZE)
1083 		return false;
1084 	return bio_add_page(bio, folio_page(folio, nr), len, off % PAGE_SIZE) > 0;
1085 }
1086 EXPORT_SYMBOL(bio_add_folio);
1087 
1088 /**
1089  * bio_add_vmalloc_chunk - add a vmalloc chunk to a bio
1090  * @bio: destination bio
1091  * @vaddr: vmalloc address to add
1092  * @len: total length in bytes of the data to add
1093  *
1094  * Add data starting at @vaddr to @bio and return how many bytes were added.
1095  * This may be less than the amount originally asked.  Returns 0 if no data
1096  * could be added to @bio.
1097  *
1098  * This helper calls flush_kernel_vmap_range() for the range added.  For reads
1099  * the caller still needs to manually call invalidate_kernel_vmap_range() in
1100  * the completion handler.
1101  */
1102 unsigned int bio_add_vmalloc_chunk(struct bio *bio, void *vaddr, unsigned len)
1103 {
1104 	unsigned int offset = offset_in_page(vaddr);
1105 
1106 	len = min(len, PAGE_SIZE - offset);
1107 	if (bio_add_page(bio, vmalloc_to_page(vaddr), len, offset) < len)
1108 		return 0;
1109 	if (op_is_write(bio_op(bio)))
1110 		flush_kernel_vmap_range(vaddr, len);
1111 	return len;
1112 }
1113 EXPORT_SYMBOL_GPL(bio_add_vmalloc_chunk);
1114 
1115 /**
1116  * bio_add_vmalloc - add a vmalloc region to a bio
1117  * @bio: destination bio
1118  * @vaddr: vmalloc address to add
1119  * @len: total length in bytes of the data to add
1120  *
1121  * Add data starting at @vaddr to @bio.  Return %true on success or %false if
1122  * @bio does not have enough space for the payload.
1123  *
1124  * This helper calls flush_kernel_vmap_range() for the range added.  For reads
1125  * the caller still needs to manually call invalidate_kernel_vmap_range() in
1126  * the completion handler.
1127  */
1128 bool bio_add_vmalloc(struct bio *bio, void *vaddr, unsigned int len)
1129 {
1130 	do {
1131 		unsigned int added = bio_add_vmalloc_chunk(bio, vaddr, len);
1132 
1133 		if (!added)
1134 			return false;
1135 		vaddr += added;
1136 		len -= added;
1137 	} while (len);
1138 
1139 	return true;
1140 }
1141 EXPORT_SYMBOL_GPL(bio_add_vmalloc);
1142 
1143 void __bio_release_pages(struct bio *bio, bool mark_dirty)
1144 {
1145 	struct folio_iter fi;
1146 
1147 	bio_for_each_folio_all(fi, bio) {
1148 		size_t nr_pages;
1149 
1150 		if (mark_dirty) {
1151 			folio_lock(fi.folio);
1152 			folio_mark_dirty(fi.folio);
1153 			folio_unlock(fi.folio);
1154 		}
1155 		nr_pages = (fi.offset + fi.length - 1) / PAGE_SIZE -
1156 			   fi.offset / PAGE_SIZE + 1;
1157 		unpin_user_folio(fi.folio, nr_pages);
1158 	}
1159 }
1160 EXPORT_SYMBOL_GPL(__bio_release_pages);
1161 
1162 void bio_iov_bvec_set(struct bio *bio, const struct iov_iter *iter)
1163 {
1164 	WARN_ON_ONCE(bio->bi_max_vecs);
1165 
1166 	bio->bi_io_vec = (struct bio_vec *)iter->bvec;
1167 	bio->bi_iter.bi_idx = 0;
1168 	bio->bi_iter.bi_bvec_done = iter->iov_offset;
1169 	bio->bi_iter.bi_size = iov_iter_count(iter);
1170 	bio_set_flag(bio, BIO_CLONED);
1171 }
1172 
1173 /*
1174  * Aligns the bio size to the len_align_mask, releasing excessive bio vecs that
1175  * __bio_iov_iter_get_pages may have inserted, and reverts the trimmed length
1176  * for the next iteration.
1177  */
1178 static int bio_iov_iter_align_down(struct bio *bio, struct iov_iter *iter,
1179 			    unsigned len_align_mask)
1180 {
1181 	size_t nbytes = bio->bi_iter.bi_size & len_align_mask;
1182 
1183 	if (!nbytes)
1184 		return 0;
1185 
1186 	iov_iter_revert(iter, nbytes);
1187 	bio->bi_iter.bi_size -= nbytes;
1188 	do {
1189 		struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
1190 
1191 		if (nbytes < bv->bv_len) {
1192 			bv->bv_len -= nbytes;
1193 			break;
1194 		}
1195 
1196 		if (bio_flagged(bio, BIO_PAGE_PINNED))
1197 			unpin_user_page(bv->bv_page);
1198 
1199 		bio->bi_vcnt--;
1200 		nbytes -= bv->bv_len;
1201 	} while (nbytes);
1202 
1203 	if (!bio->bi_vcnt)
1204 		return -EFAULT;
1205 	return 0;
1206 }
1207 
1208 /**
1209  * bio_iov_iter_get_pages - add user or kernel pages to a bio
1210  * @bio: bio to add pages to
1211  * @iter: iov iterator describing the region to be added
1212  * @len_align_mask: the mask to align the total size to, 0 for any length
1213  *
1214  * This takes either an iterator pointing to user memory, or one pointing to
1215  * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1216  * map them into the kernel. On IO completion, the caller should put those
1217  * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided
1218  * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs
1219  * to ensure the bvecs and pages stay referenced until the submitted I/O is
1220  * completed by a call to ->ki_complete() or returns with an error other than
1221  * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF
1222  * on IO completion. If it isn't, then pages should be released.
1223  *
1224  * The function tries, but does not guarantee, to pin as many pages as
1225  * fit into the bio, or are requested in @iter, whatever is smaller. If
1226  * MM encounters an error pinning the requested pages, it stops. Error
1227  * is returned only if 0 pages could be pinned.
1228  */
1229 int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter,
1230 			   unsigned len_align_mask)
1231 {
1232 	iov_iter_extraction_t flags = 0;
1233 
1234 	if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
1235 		return -EIO;
1236 
1237 	if (iov_iter_is_bvec(iter)) {
1238 		bio_iov_bvec_set(bio, iter);
1239 		iov_iter_advance(iter, bio->bi_iter.bi_size);
1240 		return 0;
1241 	}
1242 
1243 	if (iov_iter_extract_will_pin(iter))
1244 		bio_set_flag(bio, BIO_PAGE_PINNED);
1245 	if (bio->bi_bdev && blk_queue_pci_p2pdma(bio->bi_bdev->bd_disk->queue))
1246 		flags |= ITER_ALLOW_P2PDMA;
1247 
1248 	do {
1249 		ssize_t ret;
1250 
1251 		ret = iov_iter_extract_bvecs(iter, bio->bi_io_vec,
1252 				BIO_MAX_SIZE - bio->bi_iter.bi_size,
1253 				&bio->bi_vcnt, bio->bi_max_vecs, flags);
1254 		if (ret <= 0) {
1255 			if (!bio->bi_vcnt)
1256 				return ret;
1257 			break;
1258 		}
1259 		bio->bi_iter.bi_size += ret;
1260 	} while (iov_iter_count(iter) && !bio_full(bio, 0));
1261 
1262 	if (is_pci_p2pdma_page(bio->bi_io_vec->bv_page))
1263 		bio->bi_opf |= REQ_NOMERGE;
1264 	return bio_iov_iter_align_down(bio, iter, len_align_mask);
1265 }
1266 
1267 static struct folio *folio_alloc_greedy(gfp_t gfp, size_t *size)
1268 {
1269 	struct folio *folio;
1270 
1271 	while (*size > PAGE_SIZE) {
1272 		folio = folio_alloc(gfp | __GFP_NORETRY, get_order(*size));
1273 		if (folio)
1274 			return folio;
1275 		*size = rounddown_pow_of_two(*size - 1);
1276 	}
1277 
1278 	return folio_alloc(gfp, get_order(*size));
1279 }
1280 
1281 static void bio_free_folios(struct bio *bio)
1282 {
1283 	struct bio_vec *bv;
1284 	int i;
1285 
1286 	bio_for_each_bvec_all(bv, bio, i) {
1287 		struct folio *folio = page_folio(bv->bv_page);
1288 
1289 		if (!is_zero_folio(folio))
1290 			folio_put(folio);
1291 	}
1292 }
1293 
1294 static int bio_iov_iter_bounce_write(struct bio *bio, struct iov_iter *iter,
1295 		size_t maxlen)
1296 {
1297 	size_t total_len = min(maxlen, iov_iter_count(iter));
1298 
1299 	if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
1300 		return -EINVAL;
1301 	if (WARN_ON_ONCE(bio->bi_iter.bi_size))
1302 		return -EINVAL;
1303 	if (WARN_ON_ONCE(bio->bi_vcnt >= bio->bi_max_vecs))
1304 		return -EINVAL;
1305 
1306 	do {
1307 		size_t this_len = min(total_len, SZ_1M);
1308 		struct folio *folio;
1309 
1310 		if (this_len > PAGE_SIZE * 2)
1311 			this_len = rounddown_pow_of_two(this_len);
1312 
1313 		if (bio->bi_iter.bi_size > BIO_MAX_SIZE - this_len)
1314 			break;
1315 
1316 		folio = folio_alloc_greedy(GFP_KERNEL, &this_len);
1317 		if (!folio)
1318 			break;
1319 		bio_add_folio_nofail(bio, folio, this_len, 0);
1320 
1321 		if (copy_from_iter(folio_address(folio), this_len, iter) !=
1322 				this_len) {
1323 			bio_free_folios(bio);
1324 			return -EFAULT;
1325 		}
1326 
1327 		total_len -= this_len;
1328 	} while (total_len && bio->bi_vcnt < bio->bi_max_vecs);
1329 
1330 	if (!bio->bi_iter.bi_size)
1331 		return -ENOMEM;
1332 	return 0;
1333 }
1334 
1335 static int bio_iov_iter_bounce_read(struct bio *bio, struct iov_iter *iter,
1336 		size_t maxlen)
1337 {
1338 	size_t len = min3(iov_iter_count(iter), maxlen, SZ_1M);
1339 	struct folio *folio;
1340 
1341 	folio = folio_alloc_greedy(GFP_KERNEL, &len);
1342 	if (!folio)
1343 		return -ENOMEM;
1344 
1345 	do {
1346 		ssize_t ret;
1347 
1348 		ret = iov_iter_extract_bvecs(iter, bio->bi_io_vec + 1, len,
1349 				&bio->bi_vcnt, bio->bi_max_vecs - 1, 0);
1350 		if (ret <= 0) {
1351 			if (!bio->bi_vcnt) {
1352 				folio_put(folio);
1353 				return ret;
1354 			}
1355 			break;
1356 		}
1357 		len -= ret;
1358 		bio->bi_iter.bi_size += ret;
1359 	} while (len && bio->bi_vcnt < bio->bi_max_vecs - 1);
1360 
1361 	/*
1362 	 * Set the folio directly here.  The above loop has already calculated
1363 	 * the correct bi_size, and we use bi_vcnt for the user buffers.  That
1364 	 * is safe as bi_vcnt is only used by the submitter and not the actual
1365 	 * I/O path.
1366 	 */
1367 	bvec_set_folio(&bio->bi_io_vec[0], folio, bio->bi_iter.bi_size, 0);
1368 	if (iov_iter_extract_will_pin(iter))
1369 		bio_set_flag(bio, BIO_PAGE_PINNED);
1370 	return 0;
1371 }
1372 
1373 /**
1374  * bio_iov_iter_bounce - bounce buffer data from an iter into a bio
1375  * @bio:	bio to send
1376  * @iter:	iter to read from / write into
1377  * @maxlen:	maximum size to bounce
1378  *
1379  * Helper for direct I/O implementations that need to bounce buffer because
1380  * we need to checksum the data or perform other operations that require
1381  * consistency.  Allocates folios to back the bounce buffer, and for writes
1382  * copies the data into it.  Needs to be paired with bio_iov_iter_unbounce()
1383  * called on completion.
1384  */
1385 int bio_iov_iter_bounce(struct bio *bio, struct iov_iter *iter, size_t maxlen)
1386 {
1387 	if (op_is_write(bio_op(bio)))
1388 		return bio_iov_iter_bounce_write(bio, iter, maxlen);
1389 	return bio_iov_iter_bounce_read(bio, iter, maxlen);
1390 }
1391 
1392 static void bvec_unpin(struct bio_vec *bv, bool mark_dirty)
1393 {
1394 	struct folio *folio = page_folio(bv->bv_page);
1395 	size_t nr_pages = (bv->bv_offset + bv->bv_len - 1) / PAGE_SIZE -
1396 			bv->bv_offset / PAGE_SIZE + 1;
1397 
1398 	if (mark_dirty)
1399 		folio_mark_dirty_lock(folio);
1400 	unpin_user_folio(folio, nr_pages);
1401 }
1402 
1403 static void bio_iov_iter_unbounce_read(struct bio *bio, bool is_error,
1404 		bool mark_dirty)
1405 {
1406 	unsigned int len = bio->bi_io_vec[0].bv_len;
1407 
1408 	if (likely(!is_error)) {
1409 		void *buf = bvec_virt(&bio->bi_io_vec[0]);
1410 		struct iov_iter to;
1411 
1412 		iov_iter_bvec(&to, ITER_DEST, bio->bi_io_vec + 1, bio->bi_vcnt,
1413 				len);
1414 		/* copying to pinned pages should always work */
1415 		WARN_ON_ONCE(copy_to_iter(buf, len, &to) != len);
1416 	} else {
1417 		/* No need to mark folios dirty if never copied to them */
1418 		mark_dirty = false;
1419 	}
1420 
1421 	if (bio_flagged(bio, BIO_PAGE_PINNED)) {
1422 		int i;
1423 
1424 		for (i = 0; i < bio->bi_vcnt; i++)
1425 			bvec_unpin(&bio->bi_io_vec[1 + i], mark_dirty);
1426 	}
1427 
1428 	folio_put(page_folio(bio->bi_io_vec[0].bv_page));
1429 }
1430 
1431 /**
1432  * bio_iov_iter_unbounce - finish a bounce buffer operation
1433  * @bio:	completed bio
1434  * @is_error:	%true if an I/O error occurred and data should not be copied
1435  * @mark_dirty:	If %true, folios will be marked dirty.
1436  *
1437  * Helper for direct I/O implementations that need to bounce buffer because
1438  * we need to checksum the data or perform other operations that require
1439  * consistency.  Called to complete a bio set up by bio_iov_iter_bounce().
1440  * Copies data back for reads, and marks the original folios dirty if
1441  * requested and then frees the bounce buffer.
1442  */
1443 void bio_iov_iter_unbounce(struct bio *bio, bool is_error, bool mark_dirty)
1444 {
1445 	if (op_is_write(bio_op(bio)))
1446 		bio_free_folios(bio);
1447 	else
1448 		bio_iov_iter_unbounce_read(bio, is_error, mark_dirty);
1449 }
1450 
1451 static void submit_bio_wait_endio(struct bio *bio)
1452 {
1453 	complete(bio->bi_private);
1454 }
1455 
1456 /**
1457  * submit_bio_wait - submit a bio, and wait until it completes
1458  * @bio: The &struct bio which describes the I/O
1459  *
1460  * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1461  * bio_endio() on failure.
1462  *
1463  * WARNING: Unlike to how submit_bio() is usually used, this function does not
1464  * result in bio reference to be consumed. The caller must drop the reference
1465  * on his own.
1466  */
1467 int submit_bio_wait(struct bio *bio)
1468 {
1469 	DECLARE_COMPLETION_ONSTACK_MAP(done,
1470 			bio->bi_bdev->bd_disk->lockdep_map);
1471 
1472 	bio->bi_private = &done;
1473 	bio->bi_end_io = submit_bio_wait_endio;
1474 	bio->bi_opf |= REQ_SYNC;
1475 	submit_bio(bio);
1476 	blk_wait_io(&done);
1477 
1478 	return blk_status_to_errno(bio->bi_status);
1479 }
1480 EXPORT_SYMBOL(submit_bio_wait);
1481 
1482 /**
1483  * bdev_rw_virt - synchronously read into / write from kernel mapping
1484  * @bdev:	block device to access
1485  * @sector:	sector to access
1486  * @data:	data to read/write
1487  * @len:	length in byte to read/write
1488  * @op:		operation (e.g. REQ_OP_READ/REQ_OP_WRITE)
1489  *
1490  * Performs synchronous I/O to @bdev for @data/@len.  @data must be in
1491  * the kernel direct mapping and not a vmalloc address.
1492  */
1493 int bdev_rw_virt(struct block_device *bdev, sector_t sector, void *data,
1494 		size_t len, enum req_op op)
1495 {
1496 	struct bio_vec bv;
1497 	struct bio bio;
1498 	int error;
1499 
1500 	if (WARN_ON_ONCE(is_vmalloc_addr(data)))
1501 		return -EIO;
1502 
1503 	bio_init(&bio, bdev, &bv, 1, op);
1504 	bio.bi_iter.bi_sector = sector;
1505 	bio_add_virt_nofail(&bio, data, len);
1506 	error = submit_bio_wait(&bio);
1507 	bio_uninit(&bio);
1508 	return error;
1509 }
1510 EXPORT_SYMBOL_GPL(bdev_rw_virt);
1511 
1512 static void bio_wait_end_io(struct bio *bio)
1513 {
1514 	complete(bio->bi_private);
1515 	bio_put(bio);
1516 }
1517 
1518 /*
1519  * bio_await_chain - ends @bio and waits for every chained bio to complete
1520  */
1521 void bio_await_chain(struct bio *bio)
1522 {
1523 	DECLARE_COMPLETION_ONSTACK_MAP(done,
1524 			bio->bi_bdev->bd_disk->lockdep_map);
1525 
1526 	bio->bi_private = &done;
1527 	bio->bi_end_io = bio_wait_end_io;
1528 	bio_endio(bio);
1529 	blk_wait_io(&done);
1530 }
1531 
1532 void __bio_advance(struct bio *bio, unsigned bytes)
1533 {
1534 	if (bio_integrity(bio))
1535 		bio_integrity_advance(bio, bytes);
1536 
1537 	bio_crypt_advance(bio, bytes);
1538 	bio_advance_iter(bio, &bio->bi_iter, bytes);
1539 }
1540 EXPORT_SYMBOL(__bio_advance);
1541 
1542 void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1543 			struct bio *src, struct bvec_iter *src_iter)
1544 {
1545 	while (src_iter->bi_size && dst_iter->bi_size) {
1546 		struct bio_vec src_bv = bio_iter_iovec(src, *src_iter);
1547 		struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter);
1548 		unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len);
1549 		void *src_buf = bvec_kmap_local(&src_bv);
1550 		void *dst_buf = bvec_kmap_local(&dst_bv);
1551 
1552 		memcpy(dst_buf, src_buf, bytes);
1553 
1554 		kunmap_local(dst_buf);
1555 		kunmap_local(src_buf);
1556 
1557 		bio_advance_iter_single(src, src_iter, bytes);
1558 		bio_advance_iter_single(dst, dst_iter, bytes);
1559 	}
1560 }
1561 EXPORT_SYMBOL(bio_copy_data_iter);
1562 
1563 /**
1564  * bio_copy_data - copy contents of data buffers from one bio to another
1565  * @src: source bio
1566  * @dst: destination bio
1567  *
1568  * Stops when it reaches the end of either @src or @dst - that is, copies
1569  * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1570  */
1571 void bio_copy_data(struct bio *dst, struct bio *src)
1572 {
1573 	struct bvec_iter src_iter = src->bi_iter;
1574 	struct bvec_iter dst_iter = dst->bi_iter;
1575 
1576 	bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1577 }
1578 EXPORT_SYMBOL(bio_copy_data);
1579 
1580 void bio_free_pages(struct bio *bio)
1581 {
1582 	struct bio_vec *bvec;
1583 	struct bvec_iter_all iter_all;
1584 
1585 	bio_for_each_segment_all(bvec, bio, iter_all)
1586 		__free_page(bvec->bv_page);
1587 }
1588 EXPORT_SYMBOL(bio_free_pages);
1589 
1590 /*
1591  * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1592  * for performing direct-IO in BIOs.
1593  *
1594  * The problem is that we cannot run folio_mark_dirty() from interrupt context
1595  * because the required locks are not interrupt-safe.  So what we can do is to
1596  * mark the pages dirty _before_ performing IO.  And in interrupt context,
1597  * check that the pages are still dirty.   If so, fine.  If not, redirty them
1598  * in process context.
1599  *
1600  * Note that this code is very hard to test under normal circumstances because
1601  * direct-io pins the pages with get_user_pages().  This makes
1602  * is_page_cache_freeable return false, and the VM will not clean the pages.
1603  * But other code (eg, flusher threads) could clean the pages if they are mapped
1604  * pagecache.
1605  *
1606  * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1607  * deferred bio dirtying paths.
1608  */
1609 
1610 /*
1611  * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1612  */
1613 void bio_set_pages_dirty(struct bio *bio)
1614 {
1615 	struct folio_iter fi;
1616 
1617 	bio_for_each_folio_all(fi, bio) {
1618 		folio_lock(fi.folio);
1619 		folio_mark_dirty(fi.folio);
1620 		folio_unlock(fi.folio);
1621 	}
1622 }
1623 EXPORT_SYMBOL_GPL(bio_set_pages_dirty);
1624 
1625 /*
1626  * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1627  * If they are, then fine.  If, however, some pages are clean then they must
1628  * have been written out during the direct-IO read.  So we take another ref on
1629  * the BIO and re-dirty the pages in process context.
1630  *
1631  * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1632  * here on.  It will unpin each page and will run one bio_put() against the
1633  * BIO.
1634  */
1635 
1636 static void bio_dirty_fn(struct work_struct *work);
1637 
1638 static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1639 static DEFINE_SPINLOCK(bio_dirty_lock);
1640 static struct bio *bio_dirty_list;
1641 
1642 /*
1643  * This runs in process context
1644  */
1645 static void bio_dirty_fn(struct work_struct *work)
1646 {
1647 	struct bio *bio, *next;
1648 
1649 	spin_lock_irq(&bio_dirty_lock);
1650 	next = bio_dirty_list;
1651 	bio_dirty_list = NULL;
1652 	spin_unlock_irq(&bio_dirty_lock);
1653 
1654 	while ((bio = next) != NULL) {
1655 		next = bio->bi_private;
1656 
1657 		bio_release_pages(bio, true);
1658 		bio_put(bio);
1659 	}
1660 }
1661 
1662 void bio_check_pages_dirty(struct bio *bio)
1663 {
1664 	struct folio_iter fi;
1665 	unsigned long flags;
1666 
1667 	bio_for_each_folio_all(fi, bio) {
1668 		if (!folio_test_dirty(fi.folio))
1669 			goto defer;
1670 	}
1671 
1672 	bio_release_pages(bio, false);
1673 	bio_put(bio);
1674 	return;
1675 defer:
1676 	spin_lock_irqsave(&bio_dirty_lock, flags);
1677 	bio->bi_private = bio_dirty_list;
1678 	bio_dirty_list = bio;
1679 	spin_unlock_irqrestore(&bio_dirty_lock, flags);
1680 	schedule_work(&bio_dirty_work);
1681 }
1682 EXPORT_SYMBOL_GPL(bio_check_pages_dirty);
1683 
1684 static inline bool bio_remaining_done(struct bio *bio)
1685 {
1686 	/*
1687 	 * If we're not chaining, then ->__bi_remaining is always 1 and
1688 	 * we always end io on the first invocation.
1689 	 */
1690 	if (!bio_flagged(bio, BIO_CHAIN))
1691 		return true;
1692 
1693 	BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1694 
1695 	if (atomic_dec_and_test(&bio->__bi_remaining)) {
1696 		bio_clear_flag(bio, BIO_CHAIN);
1697 		return true;
1698 	}
1699 
1700 	return false;
1701 }
1702 
1703 /**
1704  * bio_endio - end I/O on a bio
1705  * @bio:	bio
1706  *
1707  * Description:
1708  *   bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1709  *   way to end I/O on a bio. No one should call bi_end_io() directly on a
1710  *   bio unless they own it and thus know that it has an end_io function.
1711  *
1712  *   bio_endio() can be called several times on a bio that has been chained
1713  *   using bio_chain().  The ->bi_end_io() function will only be called the
1714  *   last time.
1715  **/
1716 void bio_endio(struct bio *bio)
1717 {
1718 again:
1719 	if (!bio_remaining_done(bio))
1720 		return;
1721 	if (!bio_integrity_endio(bio))
1722 		return;
1723 
1724 	blk_zone_bio_endio(bio);
1725 
1726 	rq_qos_done_bio(bio);
1727 
1728 	if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1729 		trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio);
1730 		bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1731 	}
1732 
1733 	/*
1734 	 * Need to have a real endio function for chained bios, otherwise
1735 	 * various corner cases will break (like stacking block devices that
1736 	 * save/restore bi_end_io) - however, we want to avoid unbounded
1737 	 * recursion and blowing the stack. Tail call optimization would
1738 	 * handle this, but compiling with frame pointers also disables
1739 	 * gcc's sibling call optimization.
1740 	 */
1741 	if (bio->bi_end_io == bio_chain_endio) {
1742 		bio = __bio_chain_endio(bio);
1743 		goto again;
1744 	}
1745 
1746 #ifdef CONFIG_BLK_CGROUP
1747 	/*
1748 	 * Release cgroup info.  We shouldn't have to do this here, but quite
1749 	 * a few callers of bio_init fail to call bio_uninit, so we cover up
1750 	 * for that here at least for now.
1751 	 */
1752 	if (bio->bi_blkg) {
1753 		blkg_put(bio->bi_blkg);
1754 		bio->bi_blkg = NULL;
1755 	}
1756 #endif
1757 
1758 	if (bio->bi_end_io)
1759 		bio->bi_end_io(bio);
1760 }
1761 EXPORT_SYMBOL(bio_endio);
1762 
1763 /**
1764  * bio_split - split a bio
1765  * @bio:	bio to split
1766  * @sectors:	number of sectors to split from the front of @bio
1767  * @gfp:	gfp mask
1768  * @bs:		bio set to allocate from
1769  *
1770  * Allocates and returns a new bio which represents @sectors from the start of
1771  * @bio, and updates @bio to represent the remaining sectors.
1772  *
1773  * Unless this is a discard request the newly allocated bio will point
1774  * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1775  * neither @bio nor @bs are freed before the split bio.
1776  */
1777 struct bio *bio_split(struct bio *bio, int sectors,
1778 		      gfp_t gfp, struct bio_set *bs)
1779 {
1780 	struct bio *split;
1781 
1782 	if (WARN_ON_ONCE(sectors <= 0))
1783 		return ERR_PTR(-EINVAL);
1784 	if (WARN_ON_ONCE(sectors >= bio_sectors(bio)))
1785 		return ERR_PTR(-EINVAL);
1786 
1787 	/* Zone append commands cannot be split */
1788 	if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))
1789 		return ERR_PTR(-EINVAL);
1790 
1791 	/* atomic writes cannot be split */
1792 	if (bio->bi_opf & REQ_ATOMIC)
1793 		return ERR_PTR(-EINVAL);
1794 
1795 	split = bio_alloc_clone(bio->bi_bdev, bio, gfp, bs);
1796 	if (!split)
1797 		return ERR_PTR(-ENOMEM);
1798 
1799 	split->bi_iter.bi_size = sectors << 9;
1800 
1801 	if (bio_integrity(split))
1802 		bio_integrity_trim(split);
1803 
1804 	bio_advance(bio, split->bi_iter.bi_size);
1805 
1806 	if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1807 		bio_set_flag(split, BIO_TRACE_COMPLETION);
1808 
1809 	return split;
1810 }
1811 EXPORT_SYMBOL(bio_split);
1812 
1813 /**
1814  * bio_trim - trim a bio
1815  * @bio:	bio to trim
1816  * @offset:	number of sectors to trim from the front of @bio
1817  * @size:	size we want to trim @bio to, in sectors
1818  *
1819  * This function is typically used for bios that are cloned and submitted
1820  * to the underlying device in parts.
1821  */
1822 void bio_trim(struct bio *bio, sector_t offset, sector_t size)
1823 {
1824 	/* We should never trim an atomic write */
1825 	if (WARN_ON_ONCE(bio->bi_opf & REQ_ATOMIC && size))
1826 		return;
1827 
1828 	if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS ||
1829 			 offset + size > bio_sectors(bio)))
1830 		return;
1831 
1832 	size <<= 9;
1833 	if (offset == 0 && size == bio->bi_iter.bi_size)
1834 		return;
1835 
1836 	bio_advance(bio, offset << 9);
1837 	bio->bi_iter.bi_size = size;
1838 
1839 	if (bio_integrity(bio))
1840 		bio_integrity_trim(bio);
1841 }
1842 EXPORT_SYMBOL_GPL(bio_trim);
1843 
1844 /*
1845  * create memory pools for biovec's in a bio_set.
1846  * use the global biovec slabs created for general use.
1847  */
1848 int biovec_init_pool(mempool_t *pool, int pool_entries)
1849 {
1850 	struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1;
1851 
1852 	return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1853 }
1854 
1855 /*
1856  * bioset_exit - exit a bioset initialized with bioset_init()
1857  *
1858  * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1859  * kzalloc()).
1860  */
1861 void bioset_exit(struct bio_set *bs)
1862 {
1863 	bio_alloc_cache_destroy(bs);
1864 	if (bs->rescue_workqueue)
1865 		destroy_workqueue(bs->rescue_workqueue);
1866 	bs->rescue_workqueue = NULL;
1867 
1868 	mempool_exit(&bs->bio_pool);
1869 	mempool_exit(&bs->bvec_pool);
1870 
1871 	if (bs->bio_slab)
1872 		bio_put_slab(bs);
1873 	bs->bio_slab = NULL;
1874 }
1875 EXPORT_SYMBOL(bioset_exit);
1876 
1877 /**
1878  * bioset_init - Initialize a bio_set
1879  * @bs:		pool to initialize
1880  * @pool_size:	Number of bio and bio_vecs to cache in the mempool
1881  * @front_pad:	Number of bytes to allocate in front of the returned bio
1882  * @flags:	Flags to modify behavior, currently %BIOSET_NEED_BVECS
1883  *              and %BIOSET_NEED_RESCUER
1884  *
1885  * Description:
1886  *    Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1887  *    to ask for a number of bytes to be allocated in front of the bio.
1888  *    Front pad allocation is useful for embedding the bio inside
1889  *    another structure, to avoid allocating extra data to go with the bio.
1890  *    Note that the bio must be embedded at the END of that structure always,
1891  *    or things will break badly.
1892  *    If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1893  *    for allocating iovecs.  This pool is not needed e.g. for bio_init_clone().
1894  *    If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used
1895  *    to dispatch queued requests when the mempool runs out of space.
1896  *
1897  */
1898 int bioset_init(struct bio_set *bs,
1899 		unsigned int pool_size,
1900 		unsigned int front_pad,
1901 		int flags)
1902 {
1903 	bs->front_pad = front_pad;
1904 	if (flags & BIOSET_NEED_BVECS)
1905 		bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1906 	else
1907 		bs->back_pad = 0;
1908 
1909 	spin_lock_init(&bs->rescue_lock);
1910 	bio_list_init(&bs->rescue_list);
1911 	INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1912 
1913 	bs->bio_slab = bio_find_or_create_slab(bs);
1914 	if (!bs->bio_slab)
1915 		return -ENOMEM;
1916 
1917 	if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
1918 		goto bad;
1919 
1920 	if ((flags & BIOSET_NEED_BVECS) &&
1921 	    biovec_init_pool(&bs->bvec_pool, pool_size))
1922 		goto bad;
1923 
1924 	if (flags & BIOSET_NEED_RESCUER) {
1925 		bs->rescue_workqueue = alloc_workqueue("bioset",
1926 							WQ_MEM_RECLAIM, 0);
1927 		if (!bs->rescue_workqueue)
1928 			goto bad;
1929 	}
1930 	if (flags & BIOSET_PERCPU_CACHE) {
1931 		bs->cache = alloc_percpu(struct bio_alloc_cache);
1932 		if (!bs->cache)
1933 			goto bad;
1934 		cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
1935 	}
1936 
1937 	return 0;
1938 bad:
1939 	bioset_exit(bs);
1940 	return -ENOMEM;
1941 }
1942 EXPORT_SYMBOL(bioset_init);
1943 
1944 static int __init init_bio(void)
1945 {
1946 	int i;
1947 
1948 	BUILD_BUG_ON(BIO_FLAG_LAST > 8 * sizeof_field(struct bio, bi_flags));
1949 
1950 	for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) {
1951 		struct biovec_slab *bvs = bvec_slabs + i;
1952 
1953 		bvs->slab = kmem_cache_create(bvs->name,
1954 				bvs->nr_vecs * sizeof(struct bio_vec), 0,
1955 				SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1956 	}
1957 
1958 	cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL,
1959 					bio_cpu_dead);
1960 
1961 	if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0,
1962 			BIOSET_NEED_BVECS | BIOSET_PERCPU_CACHE))
1963 		panic("bio: can't allocate bios\n");
1964 
1965 	return 0;
1966 }
1967 subsys_initcall(init_bio);
1968