brintos

brintos / linux-shallow public Read only

0
0
Text · 51.1 KiB · ac4d77c Raw
1913 lines · c
1// SPDX-License-Identifier: GPL-2.02/*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	1628#define ALLOC_CACHE_MAX		25629 30struct 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 37static struct biovec_slab {38	int nr_vecs;39	char *name;40	struct kmem_cache *slab;41} bvec_slabs[] __read_mostly = {42	{ .nr_vecs = 16, .name = "biovec-16" },43	{ .nr_vecs = 64, .name = "biovec-64" },44	{ .nr_vecs = 128, .name = "biovec-128" },45	{ .nr_vecs = BIO_MAX_VECS, .name = "biovec-max" },46};47 48static struct biovec_slab *biovec_slab(unsigned short nr_vecs)49{50	switch (nr_vecs) {51	/* smaller bios use inline vecs */52	case 5 ... 16:53		return &bvec_slabs[0];54	case 17 ... 64:55		return &bvec_slabs[1];56	case 65 ... 128:57		return &bvec_slabs[2];58	case 129 ... BIO_MAX_VECS:59		return &bvec_slabs[3];60	default:61		BUG();62		return NULL;63	}64}65 66/*67 * fs_bio_set is the bio_set containing bio and iovec memory pools used by68 * IO code that does not need private memory pools.69 */70struct bio_set fs_bio_set;71EXPORT_SYMBOL(fs_bio_set);72 73/*74 * Our slab pool management75 */76struct bio_slab {77	struct kmem_cache *slab;78	unsigned int slab_ref;79	unsigned int slab_size;80	char name[8];81};82static DEFINE_MUTEX(bio_slab_lock);83static DEFINE_XARRAY(bio_slabs);84 85static struct bio_slab *create_bio_slab(unsigned int size)86{87	struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL);88 89	if (!bslab)90		return NULL;91 92	snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size);93	bslab->slab = kmem_cache_create(bslab->name, size,94			ARCH_KMALLOC_MINALIGN,95			SLAB_HWCACHE_ALIGN | SLAB_TYPESAFE_BY_RCU, NULL);96	if (!bslab->slab)97		goto fail_alloc_slab;98 99	bslab->slab_ref = 1;100	bslab->slab_size = size;101 102	if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL)))103		return bslab;104 105	kmem_cache_destroy(bslab->slab);106 107fail_alloc_slab:108	kfree(bslab);109	return NULL;110}111 112static inline unsigned int bs_bio_slab_size(struct bio_set *bs)113{114	return bs->front_pad + sizeof(struct bio) + bs->back_pad;115}116 117static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs)118{119	unsigned int size = bs_bio_slab_size(bs);120	struct bio_slab *bslab;121 122	mutex_lock(&bio_slab_lock);123	bslab = xa_load(&bio_slabs, size);124	if (bslab)125		bslab->slab_ref++;126	else127		bslab = create_bio_slab(size);128	mutex_unlock(&bio_slab_lock);129 130	if (bslab)131		return bslab->slab;132	return NULL;133}134 135static void bio_put_slab(struct bio_set *bs)136{137	struct bio_slab *bslab = NULL;138	unsigned int slab_size = bs_bio_slab_size(bs);139 140	mutex_lock(&bio_slab_lock);141 142	bslab = xa_load(&bio_slabs, slab_size);143	if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))144		goto out;145 146	WARN_ON_ONCE(bslab->slab != bs->bio_slab);147 148	WARN_ON(!bslab->slab_ref);149 150	if (--bslab->slab_ref)151		goto out;152 153	xa_erase(&bio_slabs, slab_size);154 155	kmem_cache_destroy(bslab->slab);156	kfree(bslab);157 158out:159	mutex_unlock(&bio_slab_lock);160}161 162void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs)163{164	BUG_ON(nr_vecs > BIO_MAX_VECS);165 166	if (nr_vecs == BIO_MAX_VECS)167		mempool_free(bv, pool);168	else if (nr_vecs > BIO_INLINE_VECS)169		kmem_cache_free(biovec_slab(nr_vecs)->slab, bv);170}171 172/*173 * Make the first allocation restricted and don't dump info on allocation174 * failures, since we'll fall back to the mempool in case of failure.175 */176static inline gfp_t bvec_alloc_gfp(gfp_t gfp)177{178	return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) |179		__GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;180}181 182struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,183		gfp_t gfp_mask)184{185	struct biovec_slab *bvs = biovec_slab(*nr_vecs);186 187	if (WARN_ON_ONCE(!bvs))188		return NULL;189 190	/*191	 * Upgrade the nr_vecs request to take full advantage of the allocation.192	 * We also rely on this in the bvec_free path.193	 */194	*nr_vecs = bvs->nr_vecs;195 196	/*197	 * Try a slab allocation first for all smaller allocations.  If that198	 * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool.199	 * The mempool is sized to handle up to BIO_MAX_VECS entries.200	 */201	if (*nr_vecs < BIO_MAX_VECS) {202		struct bio_vec *bvl;203 204		bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask));205		if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM))206			return bvl;207		*nr_vecs = BIO_MAX_VECS;208	}209 210	return mempool_alloc(pool, gfp_mask);211}212 213void bio_uninit(struct bio *bio)214{215#ifdef CONFIG_BLK_CGROUP216	if (bio->bi_blkg) {217		blkg_put(bio->bi_blkg);218		bio->bi_blkg = NULL;219	}220#endif221	if (bio_integrity(bio))222		bio_integrity_free(bio);223 224	bio_crypt_free_ctx(bio);225}226EXPORT_SYMBOL(bio_uninit);227 228static void bio_free(struct bio *bio)229{230	struct bio_set *bs = bio->bi_pool;231	void *p = bio;232 233	WARN_ON_ONCE(!bs);234 235	bio_uninit(bio);236	bvec_free(&bs->bvec_pool, bio->bi_io_vec, bio->bi_max_vecs);237	mempool_free(p - bs->front_pad, &bs->bio_pool);238}239 240/*241 * Users of this function have their own bio allocation. Subsequently,242 * they must remember to pair any call to bio_init() with bio_uninit()243 * when IO has completed, or when the bio is released.244 */245void bio_init(struct bio *bio, struct block_device *bdev, struct bio_vec *table,246	      unsigned short max_vecs, blk_opf_t opf)247{248	bio->bi_next = NULL;249	bio->bi_bdev = bdev;250	bio->bi_opf = opf;251	bio->bi_flags = 0;252	bio->bi_ioprio = 0;253	bio->bi_write_hint = 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_CGROUP262	bio->bi_blkg = NULL;263	bio->bi_issue.value = 0;264	if (bdev)265		bio_associate_blkg(bio);266#ifdef CONFIG_BLK_CGROUP_IOCOST267	bio->bi_iocost_cost = 0;268#endif269#endif270#ifdef CONFIG_BLK_INLINE_ENCRYPTION271	bio->bi_crypt_context = NULL;272#endif273#ifdef CONFIG_BLK_DEV_INTEGRITY274	bio->bi_integrity = NULL;275#endif276	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}286EXPORT_SYMBOL(bio_init);287 288/**289 * bio_reset - reinitialize a bio290 * @bio:	bio to reset291 * @bdev:	block device to use the bio for292 * @opf:	operation and flags for bio293 *294 * Description:295 *   After calling bio_reset(), @bio will be in the same state as a freshly296 *   allocated bio returned bio bio_alloc_bioset() - the only fields that are297 *   preserved are the ones that are initialized by bio_alloc_bioset(). See298 *   comment in struct bio.299 */300void 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}310EXPORT_SYMBOL(bio_reset);311 312static struct bio *__bio_chain_endio(struct bio *bio)313{314	struct bio *parent = bio->bi_private;315 316	if (bio->bi_status && !parent->bi_status)317		parent->bi_status = bio->bi_status;318	bio_put(bio);319	return parent;320}321 322static void bio_chain_endio(struct bio *bio)323{324	bio_endio(__bio_chain_endio(bio));325}326 327/**328 * bio_chain - chain bio completions329 * @bio: the target bio330 * @parent: the parent bio of @bio331 *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 have334 * completed; the chained bio will also be freed when it completes.335 *336 * The caller must not set bi_private or bi_end_io in @bio.337 */338void bio_chain(struct bio *bio, struct bio *parent)339{340	BUG_ON(bio->bi_private || bio->bi_end_io);341 342	bio->bi_private = parent;343	bio->bi_end_io	= bio_chain_endio;344	bio_inc_remaining(parent);345}346EXPORT_SYMBOL(bio_chain);347 348/**349 * bio_chain_and_submit - submit a bio after chaining it to another one350 * @prev: bio to chain and submit351 * @new: bio to chain to352 *353 * If @prev is non-NULL, chain it to @new and submit it.354 *355 * Return: @new.356 */357struct bio *bio_chain_and_submit(struct bio *prev, struct bio *new)358{359	if (prev) {360		bio_chain(prev, new);361		submit_bio(prev);362	}363	return new;364}365 366struct bio *blk_next_bio(struct bio *bio, struct block_device *bdev,367		unsigned int nr_pages, blk_opf_t opf, gfp_t gfp)368{369	return bio_chain_and_submit(bio, bio_alloc(bdev, nr_pages, opf, gfp));370}371EXPORT_SYMBOL_GPL(blk_next_bio);372 373static void bio_alloc_rescue(struct work_struct *work)374{375	struct bio_set *bs = container_of(work, struct bio_set, rescue_work);376	struct bio *bio;377 378	while (1) {379		spin_lock(&bs->rescue_lock);380		bio = bio_list_pop(&bs->rescue_list);381		spin_unlock(&bs->rescue_lock);382 383		if (!bio)384			break;385 386		submit_bio_noacct(bio);387	}388}389 390static void punt_bios_to_rescuer(struct bio_set *bs)391{392	struct bio_list punt, nopunt;393	struct bio *bio;394 395	if (WARN_ON_ONCE(!bs->rescue_workqueue))396		return;397	/*398	 * In order to guarantee forward progress we must punt only bios that399	 * were allocated from this bio_set; otherwise, if there was a bio on400	 * there for a stacking driver higher up in the stack, processing it401	 * could require allocating bios from this bio_set, and doing that from402	 * our own rescuer would be bad.403	 *404	 * Since bio lists are singly linked, pop them all instead of trying to405	 * remove from the middle of the list:406	 */407 408	bio_list_init(&punt);409	bio_list_init(&nopunt);410 411	while ((bio = bio_list_pop(&current->bio_list[0])))412		bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);413	current->bio_list[0] = nopunt;414 415	bio_list_init(&nopunt);416	while ((bio = bio_list_pop(&current->bio_list[1])))417		bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);418	current->bio_list[1] = nopunt;419 420	spin_lock(&bs->rescue_lock);421	bio_list_merge(&bs->rescue_list, &punt);422	spin_unlock(&bs->rescue_lock);423 424	queue_work(bs->rescue_workqueue, &bs->rescue_work);425}426 427static void bio_alloc_irq_cache_splice(struct bio_alloc_cache *cache)428{429	unsigned long flags;430 431	/* cache->free_list must be empty */432	if (WARN_ON_ONCE(cache->free_list))433		return;434 435	local_irq_save(flags);436	cache->free_list = cache->free_list_irq;437	cache->free_list_irq = NULL;438	cache->nr += cache->nr_irq;439	cache->nr_irq = 0;440	local_irq_restore(flags);441}442 443static struct bio *bio_alloc_percpu_cache(struct block_device *bdev,444		unsigned short nr_vecs, blk_opf_t opf, gfp_t gfp,445		struct bio_set *bs)446{447	struct bio_alloc_cache *cache;448	struct bio *bio;449 450	cache = per_cpu_ptr(bs->cache, get_cpu());451	if (!cache->free_list) {452		if (READ_ONCE(cache->nr_irq) >= ALLOC_CACHE_THRESHOLD)453			bio_alloc_irq_cache_splice(cache);454		if (!cache->free_list) {455			put_cpu();456			return NULL;457		}458	}459	bio = cache->free_list;460	cache->free_list = bio->bi_next;461	cache->nr--;462	put_cpu();463 464	bio_init(bio, bdev, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs, opf);465	bio->bi_pool = bs;466	return bio;467}468 469/**470 * bio_alloc_bioset - allocate a bio for I/O471 * @bdev:	block device to allocate the bio for (can be %NULL)472 * @nr_vecs:	number of bvecs to pre-allocate473 * @opf:	operation and flags for bio474 * @gfp_mask:   the GFP_* mask given to the slab allocator475 * @bs:		the bio_set to allocate from.476 *477 * Allocate a bio from the mempools in @bs.478 *479 * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to480 * allocate a bio.  This is due to the mempool guarantees.  To make this work,481 * callers must never allocate more than 1 bio at a time from the general pool.482 * Callers that need to allocate more than 1 bio must always submit the483 * previously allocated bio for IO before attempting to allocate a new one.484 * Failure to do so can cause deadlocks under memory pressure.485 *486 * Note that when running under submit_bio_noacct() (i.e. any block driver),487 * bios are not submitted until after you return - see the code in488 * submit_bio_noacct() that converts recursion into iteration, to prevent489 * stack overflows.490 *491 * This would normally mean allocating multiple bios under submit_bio_noacct()492 * would be susceptible to deadlocks, but we have493 * deadlock avoidance code that resubmits any blocked bios from a rescuer494 * thread.495 *496 * However, we do not guarantee forward progress for allocations from other497 * mempools. Doing multiple allocations from the same mempool under498 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad499 * for per bio allocations.500 *501 * Returns: Pointer to new bio on success, NULL on failure.502 */503struct bio *bio_alloc_bioset(struct block_device *bdev, unsigned short nr_vecs,504			     blk_opf_t opf, gfp_t gfp_mask,505			     struct bio_set *bs)506{507	gfp_t saved_gfp = gfp_mask;508	struct bio *bio;509	void *p;510 511	/* should not use nobvec bioset for nr_vecs > 0 */512	if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_vecs > 0))513		return NULL;514 515	if (opf & REQ_ALLOC_CACHE) {516		if (bs->cache && nr_vecs <= BIO_INLINE_VECS) {517			bio = bio_alloc_percpu_cache(bdev, nr_vecs, opf,518						     gfp_mask, bs);519			if (bio)520				return bio;521			/*522			 * No cached bio available, bio returned below marked with523			 * REQ_ALLOC_CACHE to particpate in per-cpu alloc cache.524			 */525		} else {526			opf &= ~REQ_ALLOC_CACHE;527		}528	}529 530	/*531	 * submit_bio_noacct() converts recursion to iteration; this means if532	 * we're running beneath it, any bios we allocate and submit will not be533	 * submitted (and thus freed) until after we return.534	 *535	 * This exposes us to a potential deadlock if we allocate multiple bios536	 * from the same bio_set() while running underneath submit_bio_noacct().537	 * If we were to allocate multiple bios (say a stacking block driver538	 * that was splitting bios), we would deadlock if we exhausted the539	 * mempool's reserve.540	 *541	 * We solve this, and guarantee forward progress, with a rescuer542	 * workqueue per bio_set. If we go to allocate and there are bios on543	 * current->bio_list, we first try the allocation without544	 * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be545	 * blocking to the rescuer workqueue before we retry with the original546	 * gfp_flags.547	 */548	if (current->bio_list &&549	    (!bio_list_empty(&current->bio_list[0]) ||550	     !bio_list_empty(&current->bio_list[1])) &&551	    bs->rescue_workqueue)552		gfp_mask &= ~__GFP_DIRECT_RECLAIM;553 554	p = mempool_alloc(&bs->bio_pool, gfp_mask);555	if (!p && gfp_mask != saved_gfp) {556		punt_bios_to_rescuer(bs);557		gfp_mask = saved_gfp;558		p = mempool_alloc(&bs->bio_pool, gfp_mask);559	}560	if (unlikely(!p))561		return NULL;562	if (!mempool_is_saturated(&bs->bio_pool))563		opf &= ~REQ_ALLOC_CACHE;564 565	bio = p + bs->front_pad;566	if (nr_vecs > BIO_INLINE_VECS) {567		struct bio_vec *bvl = NULL;568 569		bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);570		if (!bvl && gfp_mask != saved_gfp) {571			punt_bios_to_rescuer(bs);572			gfp_mask = saved_gfp;573			bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);574		}575		if (unlikely(!bvl))576			goto err_free;577 578		bio_init(bio, bdev, bvl, nr_vecs, opf);579	} else if (nr_vecs) {580		bio_init(bio, bdev, bio->bi_inline_vecs, BIO_INLINE_VECS, opf);581	} else {582		bio_init(bio, bdev, NULL, 0, opf);583	}584 585	bio->bi_pool = bs;586	return bio;587 588err_free:589	mempool_free(p, &bs->bio_pool);590	return NULL;591}592EXPORT_SYMBOL(bio_alloc_bioset);593 594/**595 * bio_kmalloc - kmalloc a bio596 * @nr_vecs:	number of bio_vecs to allocate597 * @gfp_mask:   the GFP_* mask given to the slab allocator598 *599 * Use kmalloc to allocate a bio (including bvecs).  The bio must be initialized600 * using bio_init() before use.  To free a bio returned from this function use601 * kfree() after calling bio_uninit().  A bio returned from this function can602 * be reused by calling bio_uninit() before calling bio_init() again.603 *604 * Note that unlike bio_alloc() or bio_alloc_bioset() allocations from this605 * function are not backed by a mempool can fail.  Do not use this function606 * for allocations in the file system I/O path.607 *608 * Returns: Pointer to new bio on success, NULL on failure.609 */610struct bio *bio_kmalloc(unsigned short nr_vecs, gfp_t gfp_mask)611{612	struct bio *bio;613 614	if (nr_vecs > UIO_MAXIOV)615		return NULL;616	return kmalloc(struct_size(bio, bi_inline_vecs, nr_vecs), gfp_mask);617}618EXPORT_SYMBOL(bio_kmalloc);619 620void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start)621{622	struct bio_vec bv;623	struct bvec_iter iter;624 625	__bio_for_each_segment(bv, bio, iter, start)626		memzero_bvec(&bv);627}628EXPORT_SYMBOL(zero_fill_bio_iter);629 630/**631 * bio_truncate - truncate the bio to small size of @new_size632 * @bio:	the bio to be truncated633 * @new_size:	new size for truncating the bio634 *635 * Description:636 *   Truncate the bio to new size of @new_size. If bio_op(bio) is637 *   REQ_OP_READ, zero the truncated part. This function should only638 *   be used for handling corner cases, such as bio eod.639 */640static void bio_truncate(struct bio *bio, unsigned new_size)641{642	struct bio_vec bv;643	struct bvec_iter iter;644	unsigned int done = 0;645	bool truncated = false;646 647	if (new_size >= bio->bi_iter.bi_size)648		return;649 650	if (bio_op(bio) != REQ_OP_READ)651		goto exit;652 653	bio_for_each_segment(bv, bio, iter) {654		if (done + bv.bv_len > new_size) {655			unsigned offset;656 657			if (!truncated)658				offset = new_size - done;659			else660				offset = 0;661			zero_user(bv.bv_page, bv.bv_offset + offset,662				  bv.bv_len - offset);663			truncated = true;664		}665		done += bv.bv_len;666	}667 668 exit:669	/*670	 * Don't touch bvec table here and make it really immutable, since671	 * fs bio user has to retrieve all pages via bio_for_each_segment_all672	 * in its .end_bio() callback.673	 *674	 * It is enough to truncate bio by updating .bi_size since we can make675	 * correct bvec with the updated .bi_size for drivers.676	 */677	bio->bi_iter.bi_size = new_size;678}679 680/**681 * guard_bio_eod - truncate a BIO to fit the block device682 * @bio:	bio to truncate683 *684 * This allows us to do IO even on the odd last sectors of a device, even if the685 * block size is some multiple of the physical sector size.686 *687 * We'll just truncate the bio to the size of the device, and clear the end of688 * the buffer head manually.  Truly out-of-range accesses will turn into actual689 * I/O errors, this only handles the "we need to be able to do I/O at the final690 * sector" case.691 */692void guard_bio_eod(struct bio *bio)693{694	sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);695 696	if (!maxsector)697		return;698 699	/*700	 * If the *whole* IO is past the end of the device,701	 * let it through, and the IO layer will turn it into702	 * an EIO.703	 */704	if (unlikely(bio->bi_iter.bi_sector >= maxsector))705		return;706 707	maxsector -= bio->bi_iter.bi_sector;708	if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))709		return;710 711	bio_truncate(bio, maxsector << 9);712}713 714static int __bio_alloc_cache_prune(struct bio_alloc_cache *cache,715				   unsigned int nr)716{717	unsigned int i = 0;718	struct bio *bio;719 720	while ((bio = cache->free_list) != NULL) {721		cache->free_list = bio->bi_next;722		cache->nr--;723		bio_free(bio);724		if (++i == nr)725			break;726	}727	return i;728}729 730static void bio_alloc_cache_prune(struct bio_alloc_cache *cache,731				  unsigned int nr)732{733	nr -= __bio_alloc_cache_prune(cache, nr);734	if (!READ_ONCE(cache->free_list)) {735		bio_alloc_irq_cache_splice(cache);736		__bio_alloc_cache_prune(cache, nr);737	}738}739 740static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node)741{742	struct bio_set *bs;743 744	bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead);745	if (bs->cache) {746		struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu);747 748		bio_alloc_cache_prune(cache, -1U);749	}750	return 0;751}752 753static void bio_alloc_cache_destroy(struct bio_set *bs)754{755	int cpu;756 757	if (!bs->cache)758		return;759 760	cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);761	for_each_possible_cpu(cpu) {762		struct bio_alloc_cache *cache;763 764		cache = per_cpu_ptr(bs->cache, cpu);765		bio_alloc_cache_prune(cache, -1U);766	}767	free_percpu(bs->cache);768	bs->cache = NULL;769}770 771static inline void bio_put_percpu_cache(struct bio *bio)772{773	struct bio_alloc_cache *cache;774 775	cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu());776	if (READ_ONCE(cache->nr_irq) + cache->nr > ALLOC_CACHE_MAX)777		goto out_free;778 779	if (in_task()) {780		bio_uninit(bio);781		bio->bi_next = cache->free_list;782		/* Not necessary but helps not to iopoll already freed bios */783		bio->bi_bdev = NULL;784		cache->free_list = bio;785		cache->nr++;786	} else if (in_hardirq()) {787		lockdep_assert_irqs_disabled();788 789		bio_uninit(bio);790		bio->bi_next = cache->free_list_irq;791		cache->free_list_irq = bio;792		cache->nr_irq++;793	} else {794		goto out_free;795	}796	put_cpu();797	return;798out_free:799	put_cpu();800	bio_free(bio);801}802 803/**804 * bio_put - release a reference to a bio805 * @bio:   bio to release reference to806 *807 * Description:808 *   Put a reference to a &struct bio, either one you have gotten with809 *   bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.810 **/811void bio_put(struct bio *bio)812{813	if (unlikely(bio_flagged(bio, BIO_REFFED))) {814		BUG_ON(!atomic_read(&bio->__bi_cnt));815		if (!atomic_dec_and_test(&bio->__bi_cnt))816			return;817	}818	if (bio->bi_opf & REQ_ALLOC_CACHE)819		bio_put_percpu_cache(bio);820	else821		bio_free(bio);822}823EXPORT_SYMBOL(bio_put);824 825static int __bio_clone(struct bio *bio, struct bio *bio_src, gfp_t gfp)826{827	bio_set_flag(bio, BIO_CLONED);828	bio->bi_ioprio = bio_src->bi_ioprio;829	bio->bi_write_hint = bio_src->bi_write_hint;830	bio->bi_iter = bio_src->bi_iter;831 832	if (bio->bi_bdev) {833		if (bio->bi_bdev == bio_src->bi_bdev &&834		    bio_flagged(bio_src, BIO_REMAPPED))835			bio_set_flag(bio, BIO_REMAPPED);836		bio_clone_blkg_association(bio, bio_src);837	}838 839	if (bio_crypt_clone(bio, bio_src, gfp) < 0)840		return -ENOMEM;841	if (bio_integrity(bio_src) &&842	    bio_integrity_clone(bio, bio_src, gfp) < 0)843		return -ENOMEM;844	return 0;845}846 847/**848 * bio_alloc_clone - clone a bio that shares the original bio's biovec849 * @bdev: block_device to clone onto850 * @bio_src: bio to clone from851 * @gfp: allocation priority852 * @bs: bio_set to allocate from853 *854 * Allocate a new bio that is a clone of @bio_src. The caller owns the returned855 * bio, but not the actual data it points to.856 *857 * The caller must ensure that the return bio is not freed before @bio_src.858 */859struct bio *bio_alloc_clone(struct block_device *bdev, struct bio *bio_src,860		gfp_t gfp, struct bio_set *bs)861{862	struct bio *bio;863 864	bio = bio_alloc_bioset(bdev, 0, bio_src->bi_opf, gfp, bs);865	if (!bio)866		return NULL;867 868	if (__bio_clone(bio, bio_src, gfp) < 0) {869		bio_put(bio);870		return NULL;871	}872	bio->bi_io_vec = bio_src->bi_io_vec;873 874	return bio;875}876EXPORT_SYMBOL(bio_alloc_clone);877 878/**879 * bio_init_clone - clone a bio that shares the original bio's biovec880 * @bdev: block_device to clone onto881 * @bio: bio to clone into882 * @bio_src: bio to clone from883 * @gfp: allocation priority884 *885 * Initialize a new bio in caller provided memory that is a clone of @bio_src.886 * The caller owns the returned bio, but not the actual data it points to.887 *888 * The caller must ensure that @bio_src is not freed before @bio.889 */890int bio_init_clone(struct block_device *bdev, struct bio *bio,891		struct bio *bio_src, gfp_t gfp)892{893	int ret;894 895	bio_init(bio, bdev, bio_src->bi_io_vec, 0, bio_src->bi_opf);896	ret = __bio_clone(bio, bio_src, gfp);897	if (ret)898		bio_uninit(bio);899	return ret;900}901EXPORT_SYMBOL(bio_init_clone);902 903/**904 * bio_full - check if the bio is full905 * @bio:	bio to check906 * @len:	length of one segment to be added907 *908 * Return true if @bio is full and one segment with @len bytes can't be909 * added to the bio, otherwise return false910 */911static inline bool bio_full(struct bio *bio, unsigned len)912{913	if (bio->bi_vcnt >= bio->bi_max_vecs)914		return true;915	if (bio->bi_iter.bi_size > UINT_MAX - len)916		return true;917	return false;918}919 920static bool bvec_try_merge_page(struct bio_vec *bv, struct page *page,921		unsigned int len, unsigned int off, bool *same_page)922{923	size_t bv_end = bv->bv_offset + bv->bv_len;924	phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1;925	phys_addr_t page_addr = page_to_phys(page);926 927	if (vec_end_addr + 1 != page_addr + off)928		return false;929	if (xen_domain() && !xen_biovec_phys_mergeable(bv, page))930		return false;931	if (!zone_device_pages_have_same_pgmap(bv->bv_page, page))932		return false;933 934	*same_page = ((vec_end_addr & PAGE_MASK) == ((page_addr + off) &935		     PAGE_MASK));936	if (!*same_page) {937		if (IS_ENABLED(CONFIG_KMSAN))938			return false;939		if (bv->bv_page + bv_end / PAGE_SIZE != page + off / PAGE_SIZE)940			return false;941	}942 943	bv->bv_len += len;944	return true;945}946 947/*948 * Try to merge a page into a segment, while obeying the hardware segment949 * size limit.  This is not for normal read/write bios, but for passthrough950 * or Zone Append operations that we can't split.951 */952bool bvec_try_merge_hw_page(struct request_queue *q, struct bio_vec *bv,953		struct page *page, unsigned len, unsigned offset,954		bool *same_page)955{956	unsigned long mask = queue_segment_boundary(q);957	phys_addr_t addr1 = bvec_phys(bv);958	phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;959 960	if ((addr1 | mask) != (addr2 | mask))961		return false;962	if (len > queue_max_segment_size(q) - bv->bv_len)963		return false;964	return bvec_try_merge_page(bv, page, len, offset, same_page);965}966 967/**968 * bio_add_hw_page - attempt to add a page to a bio with hw constraints969 * @q: the target queue970 * @bio: destination bio971 * @page: page to add972 * @len: vec entry length973 * @offset: vec entry offset974 * @max_sectors: maximum number of sectors that can be added975 * @same_page: return if the segment has been merged inside the same page976 *977 * Add a page to a bio while respecting the hardware max_sectors, max_segment978 * and gap limitations.979 */980int bio_add_hw_page(struct request_queue *q, struct bio *bio,981		struct page *page, unsigned int len, unsigned int offset,982		unsigned int max_sectors, bool *same_page)983{984	unsigned int max_size = max_sectors << SECTOR_SHIFT;985 986	if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))987		return 0;988 989	len = min3(len, max_size, queue_max_segment_size(q));990	if (len > max_size - bio->bi_iter.bi_size)991		return 0;992 993	if (bio->bi_vcnt > 0) {994		struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];995 996		if (bvec_try_merge_hw_page(q, bv, page, len, offset,997				same_page)) {998			bio->bi_iter.bi_size += len;999			return len;1000		}1001 1002		if (bio->bi_vcnt >=1003		    min(bio->bi_max_vecs, queue_max_segments(q)))1004			return 0;1005 1006		/*1007		 * If the queue doesn't support SG gaps and adding this segment1008		 * would create a gap, disallow it.1009		 */1010		if (bvec_gap_to_prev(&q->limits, bv, offset))1011			return 0;1012	}1013 1014	bvec_set_page(&bio->bi_io_vec[bio->bi_vcnt], page, len, offset);1015	bio->bi_vcnt++;1016	bio->bi_iter.bi_size += len;1017	return len;1018}1019 1020/**1021 * bio_add_hw_folio - attempt to add a folio to a bio with hw constraints1022 * @q: the target queue1023 * @bio: destination bio1024 * @folio: folio to add1025 * @len: vec entry length1026 * @offset: vec entry offset in the folio1027 * @max_sectors: maximum number of sectors that can be added1028 * @same_page: return if the segment has been merged inside the same folio1029 *1030 * Add a folio to a bio while respecting the hardware max_sectors, max_segment1031 * and gap limitations.1032 */1033int bio_add_hw_folio(struct request_queue *q, struct bio *bio,1034		struct folio *folio, size_t len, size_t offset,1035		unsigned int max_sectors, bool *same_page)1036{1037	if (len > UINT_MAX || offset > UINT_MAX)1038		return 0;1039	return bio_add_hw_page(q, bio, folio_page(folio, 0), len, offset,1040			       max_sectors, same_page);1041}1042 1043/**1044 * bio_add_pc_page	- attempt to add page to passthrough bio1045 * @q: the target queue1046 * @bio: destination bio1047 * @page: page to add1048 * @len: vec entry length1049 * @offset: vec entry offset1050 *1051 * Attempt to add a page to the bio_vec maplist. This can fail for a1052 * number of reasons, such as the bio being full or target block device1053 * limitations. The target block device must allow bio's up to PAGE_SIZE,1054 * so it is always possible to add a single page to an empty bio.1055 *1056 * This should only be used by passthrough bios.1057 */1058int bio_add_pc_page(struct request_queue *q, struct bio *bio,1059		struct page *page, unsigned int len, unsigned int offset)1060{1061	bool same_page = false;1062	return bio_add_hw_page(q, bio, page, len, offset,1063			queue_max_hw_sectors(q), &same_page);1064}1065EXPORT_SYMBOL(bio_add_pc_page);1066 1067/**1068 * bio_add_zone_append_page - attempt to add page to zone-append bio1069 * @bio: destination bio1070 * @page: page to add1071 * @len: vec entry length1072 * @offset: vec entry offset1073 *1074 * Attempt to add a page to the bio_vec maplist of a bio that will be submitted1075 * for a zone-append request. This can fail for a number of reasons, such as the1076 * bio being full or the target block device is not a zoned block device or1077 * other limitations of the target block device. The target block device must1078 * allow bio's up to PAGE_SIZE, so it is always possible to add a single page1079 * to an empty bio.1080 *1081 * Returns: number of bytes added to the bio, or 0 in case of a failure.1082 */1083int bio_add_zone_append_page(struct bio *bio, struct page *page,1084			     unsigned int len, unsigned int offset)1085{1086	struct request_queue *q = bdev_get_queue(bio->bi_bdev);1087	bool same_page = false;1088 1089	if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND))1090		return 0;1091 1092	if (WARN_ON_ONCE(!bdev_is_zoned(bio->bi_bdev)))1093		return 0;1094 1095	return bio_add_hw_page(q, bio, page, len, offset,1096			       queue_max_zone_append_sectors(q), &same_page);1097}1098EXPORT_SYMBOL_GPL(bio_add_zone_append_page);1099 1100/**1101 * __bio_add_page - add page(s) to a bio in a new segment1102 * @bio: destination bio1103 * @page: start page to add1104 * @len: length of the data to add, may cross pages1105 * @off: offset of the data relative to @page, may cross pages1106 *1107 * Add the data at @page + @off to @bio as a new bvec.  The caller must ensure1108 * that @bio has space for another bvec.1109 */1110void __bio_add_page(struct bio *bio, struct page *page,1111		unsigned int len, unsigned int off)1112{1113	WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));1114	WARN_ON_ONCE(bio_full(bio, len));1115 1116	bvec_set_page(&bio->bi_io_vec[bio->bi_vcnt], page, len, off);1117	bio->bi_iter.bi_size += len;1118	bio->bi_vcnt++;1119}1120EXPORT_SYMBOL_GPL(__bio_add_page);1121 1122/**1123 *	bio_add_page	-	attempt to add page(s) to bio1124 *	@bio: destination bio1125 *	@page: start page to add1126 *	@len: vec entry length, may cross pages1127 *	@offset: vec entry offset relative to @page, may cross pages1128 *1129 *	Attempt to add page(s) to the bio_vec maplist. This will only fail1130 *	if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.1131 */1132int bio_add_page(struct bio *bio, struct page *page,1133		 unsigned int len, unsigned int offset)1134{1135	bool same_page = false;1136 1137	if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))1138		return 0;1139	if (bio->bi_iter.bi_size > UINT_MAX - len)1140		return 0;1141 1142	if (bio->bi_vcnt > 0 &&1143	    bvec_try_merge_page(&bio->bi_io_vec[bio->bi_vcnt - 1],1144				page, len, offset, &same_page)) {1145		bio->bi_iter.bi_size += len;1146		return len;1147	}1148 1149	if (bio->bi_vcnt >= bio->bi_max_vecs)1150		return 0;1151	__bio_add_page(bio, page, len, offset);1152	return len;1153}1154EXPORT_SYMBOL(bio_add_page);1155 1156void bio_add_folio_nofail(struct bio *bio, struct folio *folio, size_t len,1157			  size_t off)1158{1159	WARN_ON_ONCE(len > UINT_MAX);1160	WARN_ON_ONCE(off > UINT_MAX);1161	__bio_add_page(bio, &folio->page, len, off);1162}1163EXPORT_SYMBOL_GPL(bio_add_folio_nofail);1164 1165/**1166 * bio_add_folio - Attempt to add part of a folio to a bio.1167 * @bio: BIO to add to.1168 * @folio: Folio to add.1169 * @len: How many bytes from the folio to add.1170 * @off: First byte in this folio to add.1171 *1172 * Filesystems that use folios can call this function instead of calling1173 * bio_add_page() for each page in the folio.  If @off is bigger than1174 * PAGE_SIZE, this function can create a bio_vec that starts in a page1175 * after the bv_page.  BIOs do not support folios that are 4GiB or larger.1176 *1177 * Return: Whether the addition was successful.1178 */1179bool bio_add_folio(struct bio *bio, struct folio *folio, size_t len,1180		   size_t off)1181{1182	if (len > UINT_MAX || off > UINT_MAX)1183		return false;1184	return bio_add_page(bio, &folio->page, len, off) > 0;1185}1186EXPORT_SYMBOL(bio_add_folio);1187 1188void __bio_release_pages(struct bio *bio, bool mark_dirty)1189{1190	struct folio_iter fi;1191 1192	bio_for_each_folio_all(fi, bio) {1193		size_t nr_pages;1194 1195		if (mark_dirty) {1196			folio_lock(fi.folio);1197			folio_mark_dirty(fi.folio);1198			folio_unlock(fi.folio);1199		}1200		nr_pages = (fi.offset + fi.length - 1) / PAGE_SIZE -1201			   fi.offset / PAGE_SIZE + 1;1202		unpin_user_folio(fi.folio, nr_pages);1203	}1204}1205EXPORT_SYMBOL_GPL(__bio_release_pages);1206 1207void bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter)1208{1209	size_t size = iov_iter_count(iter);1210 1211	WARN_ON_ONCE(bio->bi_max_vecs);1212 1213	if (bio_op(bio) == REQ_OP_ZONE_APPEND) {1214		struct request_queue *q = bdev_get_queue(bio->bi_bdev);1215		size_t max_sectors = queue_max_zone_append_sectors(q);1216 1217		size = min(size, max_sectors << SECTOR_SHIFT);1218	}1219 1220	bio->bi_vcnt = iter->nr_segs;1221	bio->bi_io_vec = (struct bio_vec *)iter->bvec;1222	bio->bi_iter.bi_bvec_done = iter->iov_offset;1223	bio->bi_iter.bi_size = size;1224	bio_set_flag(bio, BIO_CLONED);1225}1226 1227static int bio_iov_add_folio(struct bio *bio, struct folio *folio, size_t len,1228			     size_t offset)1229{1230	bool same_page = false;1231 1232	if (WARN_ON_ONCE(bio->bi_iter.bi_size > UINT_MAX - len))1233		return -EIO;1234 1235	if (bio->bi_vcnt > 0 &&1236	    bvec_try_merge_page(&bio->bi_io_vec[bio->bi_vcnt - 1],1237				folio_page(folio, 0), len, offset,1238				&same_page)) {1239		bio->bi_iter.bi_size += len;1240		if (same_page && bio_flagged(bio, BIO_PAGE_PINNED))1241			unpin_user_folio(folio, 1);1242		return 0;1243	}1244	bio_add_folio_nofail(bio, folio, len, offset);1245	return 0;1246}1247 1248static int bio_iov_add_zone_append_folio(struct bio *bio, struct folio *folio,1249					 size_t len, size_t offset)1250{1251	struct request_queue *q = bdev_get_queue(bio->bi_bdev);1252	bool same_page = false;1253 1254	if (bio_add_hw_folio(q, bio, folio, len, offset,1255			queue_max_zone_append_sectors(q), &same_page) != len)1256		return -EINVAL;1257	if (same_page && bio_flagged(bio, BIO_PAGE_PINNED))1258		unpin_user_folio(folio, 1);1259	return 0;1260}1261 1262static unsigned int get_contig_folio_len(unsigned int *num_pages,1263					 struct page **pages, unsigned int i,1264					 struct folio *folio, size_t left,1265					 size_t offset)1266{1267	size_t bytes = left;1268	size_t contig_sz = min_t(size_t, PAGE_SIZE - offset, bytes);1269	unsigned int j;1270 1271	/*1272	 * We might COW a single page in the middle of1273	 * a large folio, so we have to check that all1274	 * pages belong to the same folio.1275	 */1276	bytes -= contig_sz;1277	for (j = i + 1; j < i + *num_pages; j++) {1278		size_t next = min_t(size_t, PAGE_SIZE, bytes);1279 1280		if (page_folio(pages[j]) != folio ||1281		    pages[j] != pages[j - 1] + 1) {1282			break;1283		}1284		contig_sz += next;1285		bytes -= next;1286	}1287	*num_pages = j - i;1288 1289	return contig_sz;1290}1291 1292#define PAGE_PTRS_PER_BVEC     (sizeof(struct bio_vec) / sizeof(struct page *))1293 1294/**1295 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio1296 * @bio: bio to add pages to1297 * @iter: iov iterator describing the region to be mapped1298 *1299 * Extracts pages from *iter and appends them to @bio's bvec array.  The pages1300 * will have to be cleaned up in the way indicated by the BIO_PAGE_PINNED flag.1301 * For a multi-segment *iter, this function only adds pages from the next1302 * non-empty segment of the iov iterator.1303 */1304static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)1305{1306	iov_iter_extraction_t extraction_flags = 0;1307	unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;1308	unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;1309	struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;1310	struct page **pages = (struct page **)bv;1311	ssize_t size;1312	unsigned int num_pages, i = 0;1313	size_t offset, folio_offset, left, len;1314	int ret = 0;1315 1316	/*1317	 * Move page array up in the allocated memory for the bio vecs as far as1318	 * possible so that we can start filling biovecs from the beginning1319	 * without overwriting the temporary page array.1320	 */1321	BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);1322	pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);1323 1324	if (bio->bi_bdev && blk_queue_pci_p2pdma(bio->bi_bdev->bd_disk->queue))1325		extraction_flags |= ITER_ALLOW_P2PDMA;1326 1327	/*1328	 * Each segment in the iov is required to be a block size multiple.1329	 * However, we may not be able to get the entire segment if it spans1330	 * more pages than bi_max_vecs allows, so we have to ALIGN_DOWN the1331	 * result to ensure the bio's total size is correct. The remainder of1332	 * the iov data will be picked up in the next bio iteration.1333	 */1334	size = iov_iter_extract_pages(iter, &pages,1335				      UINT_MAX - bio->bi_iter.bi_size,1336				      nr_pages, extraction_flags, &offset);1337	if (unlikely(size <= 0))1338		return size ? size : -EFAULT;1339 1340	nr_pages = DIV_ROUND_UP(offset + size, PAGE_SIZE);1341 1342	if (bio->bi_bdev) {1343		size_t trim = size & (bdev_logical_block_size(bio->bi_bdev) - 1);1344		iov_iter_revert(iter, trim);1345		size -= trim;1346	}1347 1348	if (unlikely(!size)) {1349		ret = -EFAULT;1350		goto out;1351	}1352 1353	for (left = size, i = 0; left > 0; left -= len, i += num_pages) {1354		struct page *page = pages[i];1355		struct folio *folio = page_folio(page);1356 1357		folio_offset = ((size_t)folio_page_idx(folio, page) <<1358			       PAGE_SHIFT) + offset;1359 1360		len = min(folio_size(folio) - folio_offset, left);1361 1362		num_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE);1363 1364		if (num_pages > 1)1365			len = get_contig_folio_len(&num_pages, pages, i,1366						   folio, left, offset);1367 1368		if (bio_op(bio) == REQ_OP_ZONE_APPEND) {1369			ret = bio_iov_add_zone_append_folio(bio, folio, len,1370					folio_offset);1371			if (ret)1372				break;1373		} else1374			bio_iov_add_folio(bio, folio, len, folio_offset);1375 1376		offset = 0;1377	}1378 1379	iov_iter_revert(iter, left);1380out:1381	while (i < nr_pages)1382		bio_release_page(bio, pages[i++]);1383 1384	return ret;1385}1386 1387/**1388 * bio_iov_iter_get_pages - add user or kernel pages to a bio1389 * @bio: bio to add pages to1390 * @iter: iov iterator describing the region to be added1391 *1392 * This takes either an iterator pointing to user memory, or one pointing to1393 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and1394 * map them into the kernel. On IO completion, the caller should put those1395 * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided1396 * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs1397 * to ensure the bvecs and pages stay referenced until the submitted I/O is1398 * completed by a call to ->ki_complete() or returns with an error other than1399 * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF1400 * on IO completion. If it isn't, then pages should be released.1401 *1402 * The function tries, but does not guarantee, to pin as many pages as1403 * fit into the bio, or are requested in @iter, whatever is smaller. If1404 * MM encounters an error pinning the requested pages, it stops. Error1405 * is returned only if 0 pages could be pinned.1406 */1407int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)1408{1409	int ret = 0;1410 1411	if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))1412		return -EIO;1413 1414	if (iov_iter_is_bvec(iter)) {1415		bio_iov_bvec_set(bio, iter);1416		iov_iter_advance(iter, bio->bi_iter.bi_size);1417		return 0;1418	}1419 1420	if (iov_iter_extract_will_pin(iter))1421		bio_set_flag(bio, BIO_PAGE_PINNED);1422	do {1423		ret = __bio_iov_iter_get_pages(bio, iter);1424	} while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));1425 1426	return bio->bi_vcnt ? 0 : ret;1427}1428EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);1429 1430static void submit_bio_wait_endio(struct bio *bio)1431{1432	complete(bio->bi_private);1433}1434 1435/**1436 * submit_bio_wait - submit a bio, and wait until it completes1437 * @bio: The &struct bio which describes the I/O1438 *1439 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from1440 * bio_endio() on failure.1441 *1442 * WARNING: Unlike to how submit_bio() is usually used, this function does not1443 * result in bio reference to be consumed. The caller must drop the reference1444 * on his own.1445 */1446int submit_bio_wait(struct bio *bio)1447{1448	DECLARE_COMPLETION_ONSTACK_MAP(done,1449			bio->bi_bdev->bd_disk->lockdep_map);1450 1451	bio->bi_private = &done;1452	bio->bi_end_io = submit_bio_wait_endio;1453	bio->bi_opf |= REQ_SYNC;1454	submit_bio(bio);1455	blk_wait_io(&done);1456 1457	return blk_status_to_errno(bio->bi_status);1458}1459EXPORT_SYMBOL(submit_bio_wait);1460 1461static void bio_wait_end_io(struct bio *bio)1462{1463	complete(bio->bi_private);1464	bio_put(bio);1465}1466 1467/*1468 * bio_await_chain - ends @bio and waits for every chained bio to complete1469 */1470void bio_await_chain(struct bio *bio)1471{1472	DECLARE_COMPLETION_ONSTACK_MAP(done,1473			bio->bi_bdev->bd_disk->lockdep_map);1474 1475	bio->bi_private = &done;1476	bio->bi_end_io = bio_wait_end_io;1477	bio_endio(bio);1478	blk_wait_io(&done);1479}1480 1481void __bio_advance(struct bio *bio, unsigned bytes)1482{1483	if (bio_integrity(bio))1484		bio_integrity_advance(bio, bytes);1485 1486	bio_crypt_advance(bio, bytes);1487	bio_advance_iter(bio, &bio->bi_iter, bytes);1488}1489EXPORT_SYMBOL(__bio_advance);1490 1491void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,1492			struct bio *src, struct bvec_iter *src_iter)1493{1494	while (src_iter->bi_size && dst_iter->bi_size) {1495		struct bio_vec src_bv = bio_iter_iovec(src, *src_iter);1496		struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter);1497		unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len);1498		void *src_buf = bvec_kmap_local(&src_bv);1499		void *dst_buf = bvec_kmap_local(&dst_bv);1500 1501		memcpy(dst_buf, src_buf, bytes);1502 1503		kunmap_local(dst_buf);1504		kunmap_local(src_buf);1505 1506		bio_advance_iter_single(src, src_iter, bytes);1507		bio_advance_iter_single(dst, dst_iter, bytes);1508	}1509}1510EXPORT_SYMBOL(bio_copy_data_iter);1511 1512/**1513 * bio_copy_data - copy contents of data buffers from one bio to another1514 * @src: source bio1515 * @dst: destination bio1516 *1517 * Stops when it reaches the end of either @src or @dst - that is, copies1518 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).1519 */1520void bio_copy_data(struct bio *dst, struct bio *src)1521{1522	struct bvec_iter src_iter = src->bi_iter;1523	struct bvec_iter dst_iter = dst->bi_iter;1524 1525	bio_copy_data_iter(dst, &dst_iter, src, &src_iter);1526}1527EXPORT_SYMBOL(bio_copy_data);1528 1529void bio_free_pages(struct bio *bio)1530{1531	struct bio_vec *bvec;1532	struct bvec_iter_all iter_all;1533 1534	bio_for_each_segment_all(bvec, bio, iter_all)1535		__free_page(bvec->bv_page);1536}1537EXPORT_SYMBOL(bio_free_pages);1538 1539/*1540 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions1541 * for performing direct-IO in BIOs.1542 *1543 * The problem is that we cannot run folio_mark_dirty() from interrupt context1544 * because the required locks are not interrupt-safe.  So what we can do is to1545 * mark the pages dirty _before_ performing IO.  And in interrupt context,1546 * check that the pages are still dirty.   If so, fine.  If not, redirty them1547 * in process context.1548 *1549 * Note that this code is very hard to test under normal circumstances because1550 * direct-io pins the pages with get_user_pages().  This makes1551 * is_page_cache_freeable return false, and the VM will not clean the pages.1552 * But other code (eg, flusher threads) could clean the pages if they are mapped1553 * pagecache.1554 *1555 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the1556 * deferred bio dirtying paths.1557 */1558 1559/*1560 * bio_set_pages_dirty() will mark all the bio's pages as dirty.1561 */1562void bio_set_pages_dirty(struct bio *bio)1563{1564	struct folio_iter fi;1565 1566	bio_for_each_folio_all(fi, bio) {1567		folio_lock(fi.folio);1568		folio_mark_dirty(fi.folio);1569		folio_unlock(fi.folio);1570	}1571}1572EXPORT_SYMBOL_GPL(bio_set_pages_dirty);1573 1574/*1575 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.1576 * If they are, then fine.  If, however, some pages are clean then they must1577 * have been written out during the direct-IO read.  So we take another ref on1578 * the BIO and re-dirty the pages in process context.1579 *1580 * It is expected that bio_check_pages_dirty() will wholly own the BIO from1581 * here on.  It will unpin each page and will run one bio_put() against the1582 * BIO.1583 */1584 1585static void bio_dirty_fn(struct work_struct *work);1586 1587static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);1588static DEFINE_SPINLOCK(bio_dirty_lock);1589static struct bio *bio_dirty_list;1590 1591/*1592 * This runs in process context1593 */1594static void bio_dirty_fn(struct work_struct *work)1595{1596	struct bio *bio, *next;1597 1598	spin_lock_irq(&bio_dirty_lock);1599	next = bio_dirty_list;1600	bio_dirty_list = NULL;1601	spin_unlock_irq(&bio_dirty_lock);1602 1603	while ((bio = next) != NULL) {1604		next = bio->bi_private;1605 1606		bio_release_pages(bio, true);1607		bio_put(bio);1608	}1609}1610 1611void bio_check_pages_dirty(struct bio *bio)1612{1613	struct folio_iter fi;1614	unsigned long flags;1615 1616	bio_for_each_folio_all(fi, bio) {1617		if (!folio_test_dirty(fi.folio))1618			goto defer;1619	}1620 1621	bio_release_pages(bio, false);1622	bio_put(bio);1623	return;1624defer:1625	spin_lock_irqsave(&bio_dirty_lock, flags);1626	bio->bi_private = bio_dirty_list;1627	bio_dirty_list = bio;1628	spin_unlock_irqrestore(&bio_dirty_lock, flags);1629	schedule_work(&bio_dirty_work);1630}1631EXPORT_SYMBOL_GPL(bio_check_pages_dirty);1632 1633static inline bool bio_remaining_done(struct bio *bio)1634{1635	/*1636	 * If we're not chaining, then ->__bi_remaining is always 1 and1637	 * we always end io on the first invocation.1638	 */1639	if (!bio_flagged(bio, BIO_CHAIN))1640		return true;1641 1642	BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);1643 1644	if (atomic_dec_and_test(&bio->__bi_remaining)) {1645		bio_clear_flag(bio, BIO_CHAIN);1646		return true;1647	}1648 1649	return false;1650}1651 1652/**1653 * bio_endio - end I/O on a bio1654 * @bio:	bio1655 *1656 * Description:1657 *   bio_endio() will end I/O on the whole bio. bio_endio() is the preferred1658 *   way to end I/O on a bio. No one should call bi_end_io() directly on a1659 *   bio unless they own it and thus know that it has an end_io function.1660 *1661 *   bio_endio() can be called several times on a bio that has been chained1662 *   using bio_chain().  The ->bi_end_io() function will only be called the1663 *   last time.1664 **/1665void bio_endio(struct bio *bio)1666{1667again:1668	if (!bio_remaining_done(bio))1669		return;1670	if (!bio_integrity_endio(bio))1671		return;1672 1673	blk_zone_bio_endio(bio);1674 1675	rq_qos_done_bio(bio);1676 1677	if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) {1678		trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio);1679		bio_clear_flag(bio, BIO_TRACE_COMPLETION);1680	}1681 1682	/*1683	 * Need to have a real endio function for chained bios, otherwise1684	 * various corner cases will break (like stacking block devices that1685	 * save/restore bi_end_io) - however, we want to avoid unbounded1686	 * recursion and blowing the stack. Tail call optimization would1687	 * handle this, but compiling with frame pointers also disables1688	 * gcc's sibling call optimization.1689	 */1690	if (bio->bi_end_io == bio_chain_endio) {1691		bio = __bio_chain_endio(bio);1692		goto again;1693	}1694 1695#ifdef CONFIG_BLK_CGROUP1696	/*1697	 * Release cgroup info.  We shouldn't have to do this here, but quite1698	 * a few callers of bio_init fail to call bio_uninit, so we cover up1699	 * for that here at least for now.1700	 */1701	if (bio->bi_blkg) {1702		blkg_put(bio->bi_blkg);1703		bio->bi_blkg = NULL;1704	}1705#endif1706 1707	if (bio->bi_end_io)1708		bio->bi_end_io(bio);1709}1710EXPORT_SYMBOL(bio_endio);1711 1712/**1713 * bio_split - split a bio1714 * @bio:	bio to split1715 * @sectors:	number of sectors to split from the front of @bio1716 * @gfp:	gfp mask1717 * @bs:		bio set to allocate from1718 *1719 * Allocates and returns a new bio which represents @sectors from the start of1720 * @bio, and updates @bio to represent the remaining sectors.1721 *1722 * Unless this is a discard request the newly allocated bio will point1723 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that1724 * neither @bio nor @bs are freed before the split bio.1725 */1726struct bio *bio_split(struct bio *bio, int sectors,1727		      gfp_t gfp, struct bio_set *bs)1728{1729	struct bio *split;1730 1731	BUG_ON(sectors <= 0);1732	BUG_ON(sectors >= bio_sectors(bio));1733 1734	/* Zone append commands cannot be split */1735	if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))1736		return NULL;1737 1738	split = bio_alloc_clone(bio->bi_bdev, bio, gfp, bs);1739	if (!split)1740		return NULL;1741 1742	split->bi_iter.bi_size = sectors << 9;1743 1744	if (bio_integrity(split))1745		bio_integrity_trim(split);1746 1747	bio_advance(bio, split->bi_iter.bi_size);1748 1749	if (bio_flagged(bio, BIO_TRACE_COMPLETION))1750		bio_set_flag(split, BIO_TRACE_COMPLETION);1751 1752	return split;1753}1754EXPORT_SYMBOL(bio_split);1755 1756/**1757 * bio_trim - trim a bio1758 * @bio:	bio to trim1759 * @offset:	number of sectors to trim from the front of @bio1760 * @size:	size we want to trim @bio to, in sectors1761 *1762 * This function is typically used for bios that are cloned and submitted1763 * to the underlying device in parts.1764 */1765void bio_trim(struct bio *bio, sector_t offset, sector_t size)1766{1767	if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS ||1768			 offset + size > bio_sectors(bio)))1769		return;1770 1771	size <<= 9;1772	if (offset == 0 && size == bio->bi_iter.bi_size)1773		return;1774 1775	bio_advance(bio, offset << 9);1776	bio->bi_iter.bi_size = size;1777 1778	if (bio_integrity(bio))1779		bio_integrity_trim(bio);1780}1781EXPORT_SYMBOL_GPL(bio_trim);1782 1783/*1784 * create memory pools for biovec's in a bio_set.1785 * use the global biovec slabs created for general use.1786 */1787int biovec_init_pool(mempool_t *pool, int pool_entries)1788{1789	struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1;1790 1791	return mempool_init_slab_pool(pool, pool_entries, bp->slab);1792}1793 1794/*1795 * bioset_exit - exit a bioset initialized with bioset_init()1796 *1797 * May be called on a zeroed but uninitialized bioset (i.e. allocated with1798 * kzalloc()).1799 */1800void bioset_exit(struct bio_set *bs)1801{1802	bio_alloc_cache_destroy(bs);1803	if (bs->rescue_workqueue)1804		destroy_workqueue(bs->rescue_workqueue);1805	bs->rescue_workqueue = NULL;1806 1807	mempool_exit(&bs->bio_pool);1808	mempool_exit(&bs->bvec_pool);1809 1810	bioset_integrity_free(bs);1811	if (bs->bio_slab)1812		bio_put_slab(bs);1813	bs->bio_slab = NULL;1814}1815EXPORT_SYMBOL(bioset_exit);1816 1817/**1818 * bioset_init - Initialize a bio_set1819 * @bs:		pool to initialize1820 * @pool_size:	Number of bio and bio_vecs to cache in the mempool1821 * @front_pad:	Number of bytes to allocate in front of the returned bio1822 * @flags:	Flags to modify behavior, currently %BIOSET_NEED_BVECS1823 *              and %BIOSET_NEED_RESCUER1824 *1825 * Description:1826 *    Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller1827 *    to ask for a number of bytes to be allocated in front of the bio.1828 *    Front pad allocation is useful for embedding the bio inside1829 *    another structure, to avoid allocating extra data to go with the bio.1830 *    Note that the bio must be embedded at the END of that structure always,1831 *    or things will break badly.1832 *    If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated1833 *    for allocating iovecs.  This pool is not needed e.g. for bio_init_clone().1834 *    If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used1835 *    to dispatch queued requests when the mempool runs out of space.1836 *1837 */1838int bioset_init(struct bio_set *bs,1839		unsigned int pool_size,1840		unsigned int front_pad,1841		int flags)1842{1843	bs->front_pad = front_pad;1844	if (flags & BIOSET_NEED_BVECS)1845		bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);1846	else1847		bs->back_pad = 0;1848 1849	spin_lock_init(&bs->rescue_lock);1850	bio_list_init(&bs->rescue_list);1851	INIT_WORK(&bs->rescue_work, bio_alloc_rescue);1852 1853	bs->bio_slab = bio_find_or_create_slab(bs);1854	if (!bs->bio_slab)1855		return -ENOMEM;1856 1857	if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))1858		goto bad;1859 1860	if ((flags & BIOSET_NEED_BVECS) &&1861	    biovec_init_pool(&bs->bvec_pool, pool_size))1862		goto bad;1863 1864	if (flags & BIOSET_NEED_RESCUER) {1865		bs->rescue_workqueue = alloc_workqueue("bioset",1866							WQ_MEM_RECLAIM, 0);1867		if (!bs->rescue_workqueue)1868			goto bad;1869	}1870	if (flags & BIOSET_PERCPU_CACHE) {1871		bs->cache = alloc_percpu(struct bio_alloc_cache);1872		if (!bs->cache)1873			goto bad;1874		cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);1875	}1876 1877	return 0;1878bad:1879	bioset_exit(bs);1880	return -ENOMEM;1881}1882EXPORT_SYMBOL(bioset_init);1883 1884static int __init init_bio(void)1885{1886	int i;1887 1888	BUILD_BUG_ON(BIO_FLAG_LAST > 8 * sizeof_field(struct bio, bi_flags));1889 1890	bio_integrity_init();1891 1892	for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) {1893		struct biovec_slab *bvs = bvec_slabs + i;1894 1895		bvs->slab = kmem_cache_create(bvs->name,1896				bvs->nr_vecs * sizeof(struct bio_vec), 0,1897				SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);1898	}1899 1900	cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL,1901					bio_cpu_dead);1902 1903	if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0,1904			BIOSET_NEED_BVECS | BIOSET_PERCPU_CACHE))1905		panic("bio: can't allocate bios\n");1906 1907	if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))1908		panic("bio: can't create integrity pool\n");1909 1910	return 0;1911}1912subsys_initcall(init_bio);1913