bio.c

   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
  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 by
  68 * 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 management
  75 */
  76struct bio_slab {
  77	struct kmem_cache *slab;
  78	unsigned int slab_ref;
  79	unsigned int slab_size;
  80	char name[12];
  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	else
 127		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 allocation
 174 * 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 that
 198	 * 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_CGROUP
 216	if (bio->bi_blkg) {
 217		blkg_put(bio->bi_blkg);
 218		bio->bi_blkg = NULL;
 219	}
 220#endif
 221	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_write_stream = 0;
 255	bio->bi_status = 0;
 256	bio->bi_bvec_gap_bit = 0;
 257	bio->bi_iter.bi_sector = 0;
 258	bio->bi_iter.bi_size = 0;
 259	bio->bi_iter.bi_idx = 0;
 260	bio->bi_iter.bi_bvec_done = 0;
 261	bio->bi_end_io = NULL;
 262	bio->bi_private = NULL;
 263#ifdef CONFIG_BLK_CGROUP
 264	bio->bi_blkg = NULL;
 265	bio->issue_time_ns = 0;
 266	if (bdev)
 267		bio_associate_blkg(bio);
 268#ifdef CONFIG_BLK_CGROUP_IOCOST
 269	bio->bi_iocost_cost = 0;
 270#endif
 271#endif
 272#ifdef CONFIG_BLK_INLINE_ENCRYPTION
 273	bio->bi_crypt_context = NULL;
 274#endif
 275#ifdef CONFIG_BLK_DEV_INTEGRITY
 276	bio->bi_integrity = NULL;
 277#endif
 278	bio->bi_vcnt = 0;
 279
 280	atomic_set(&bio->__bi_remaining, 1);
 281	atomic_set(&bio->__bi_cnt, 1);
 282	bio->bi_cookie = BLK_QC_T_NONE;
 283
 284	bio->bi_max_vecs = max_vecs;
 285	bio->bi_io_vec = table;
 286	bio->bi_pool = NULL;
 287}
 288EXPORT_SYMBOL(bio_init);
 289
 290/**
 291 * bio_reset - reinitialize a bio
 292 * @bio:	bio to reset
 293 * @bdev:	block device to use the bio for
 294 * @opf:	operation and flags for bio
 295 *
 296 * Description:
 297 *   After calling bio_reset(), @bio will be in the same state as a freshly
 298 *   allocated bio returned bio bio_alloc_bioset() - the only fields that are
 299 *   preserved are the ones that are initialized by bio_alloc_bioset(). See
 300 *   comment in struct bio.
 301 */
 302void bio_reset(struct bio *bio, struct block_device *bdev, blk_opf_t opf)
 303{
 304	bio_uninit(bio);
 305	memset(bio, 0, BIO_RESET_BYTES);
 306	atomic_set(&bio->__bi_remaining, 1);
 307	bio->bi_bdev = bdev;
 308	if (bio->bi_bdev)
 309		bio_associate_blkg(bio);
 310	bio->bi_opf = opf;
 311}
 312EXPORT_SYMBOL(bio_reset);
 313
 314static struct bio *__bio_chain_endio(struct bio *bio)
 315{
 316	struct bio *parent = bio->bi_private;
 317
 318	if (bio->bi_status && !parent->bi_status)
 319		parent->bi_status = bio->bi_status;
 320	bio_put(bio);
 321	return parent;
 322}
 323
 324/*
 325 * This function should only be used as a flag and must never be called.
 326 * If execution reaches here, it indicates a serious programming error.
 327 */
 328static void bio_chain_endio(struct bio *bio)
 329{
 330	BUG();
 331}
 332
 333/**
 334 * bio_chain - chain bio completions
 335 * @bio: the target bio
 336 * @parent: the parent bio of @bio
 337 *
 338 * The caller won't have a bi_end_io called when @bio completes - instead,
 339 * @parent's bi_end_io won't be called until both @parent and @bio have
 340 * completed; the chained bio will also be freed when it completes.
 341 *
 342 * The caller must not set bi_private or bi_end_io in @bio.
 343 */
 344void bio_chain(struct bio *bio, struct bio *parent)
 345{
 346	BUG_ON(bio->bi_private || bio->bi_end_io);
 347
 348	bio->bi_private = parent;
 349	bio->bi_end_io	= bio_chain_endio;
 350	bio_inc_remaining(parent);
 351}
 352EXPORT_SYMBOL(bio_chain);
 353
 354/**
 355 * bio_chain_and_submit - submit a bio after chaining it to another one
 356 * @prev: bio to chain and submit
 357 * @new: bio to chain to
 358 *
 359 * If @prev is non-NULL, chain it to @new and submit it.
 360 *
 361 * Return: @new.
 362 */
 363struct bio *bio_chain_and_submit(struct bio *prev, struct bio *new)
 364{
 365	if (prev) {
 366		bio_chain(prev, new);
 367		submit_bio(prev);
 368	}
 369	return new;
 370}
 371
 372struct bio *blk_next_bio(struct bio *bio, struct block_device *bdev,
 373		unsigned int nr_pages, blk_opf_t opf, gfp_t gfp)
 374{
 375	return bio_chain_and_submit(bio, bio_alloc(bdev, nr_pages, opf, gfp));
 376}
 377EXPORT_SYMBOL_GPL(blk_next_bio);
 378
 379static void bio_alloc_rescue(struct work_struct *work)
 380{
 381	struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
 382	struct bio *bio;
 383
 384	while (1) {
 385		spin_lock(&bs->rescue_lock);
 386		bio = bio_list_pop(&bs->rescue_list);
 387		spin_unlock(&bs->rescue_lock);
 388
 389		if (!bio)
 390			break;
 391
 392		submit_bio_noacct(bio);
 393	}
 394}
 395
 396static void punt_bios_to_rescuer(struct bio_set *bs)
 397{
 398	struct bio_list punt, nopunt;
 399	struct bio *bio;
 400
 401	if (WARN_ON_ONCE(!bs->rescue_workqueue))
 402		return;
 403	/*
 404	 * In order to guarantee forward progress we must punt only bios that
 405	 * were allocated from this bio_set; otherwise, if there was a bio on
 406	 * there for a stacking driver higher up in the stack, processing it
 407	 * could require allocating bios from this bio_set, and doing that from
 408	 * our own rescuer would be bad.
 409	 *
 410	 * Since bio lists are singly linked, pop them all instead of trying to
 411	 * remove from the middle of the list:
 412	 */
 413
 414	bio_list_init(&punt);
 415	bio_list_init(&nopunt);
 416
 417	while ((bio = bio_list_pop(&current->bio_list[0])))
 418		bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
 419	current->bio_list[0] = nopunt;
 420
 421	bio_list_init(&nopunt);
 422	while ((bio = bio_list_pop(&current->bio_list[1])))
 423		bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
 424	current->bio_list[1] = nopunt;
 425
 426	spin_lock(&bs->rescue_lock);
 427	bio_list_merge(&bs->rescue_list, &punt);
 428	spin_unlock(&bs->rescue_lock);
 429
 430	queue_work(bs->rescue_workqueue, &bs->rescue_work);
 431}
 432
 433static void bio_alloc_irq_cache_splice(struct bio_alloc_cache *cache)
 434{
 435	unsigned long flags;
 436
 437	/* cache->free_list must be empty */
 438	if (WARN_ON_ONCE(cache->free_list))
 439		return;
 440
 441	local_irq_save(flags);
 442	cache->free_list = cache->free_list_irq;
 443	cache->free_list_irq = NULL;
 444	cache->nr += cache->nr_irq;
 445	cache->nr_irq = 0;
 446	local_irq_restore(flags);
 447}
 448
 449static struct bio *bio_alloc_percpu_cache(struct block_device *bdev,
 450		unsigned short nr_vecs, blk_opf_t opf, gfp_t gfp,
 451		struct bio_set *bs)
 452{
 453	struct bio_alloc_cache *cache;
 454	struct bio *bio;
 455
 456	cache = per_cpu_ptr(bs->cache, get_cpu());
 457	if (!cache->free_list) {
 458		if (READ_ONCE(cache->nr_irq) >= ALLOC_CACHE_THRESHOLD)
 459			bio_alloc_irq_cache_splice(cache);
 460		if (!cache->free_list) {
 461			put_cpu();
 462			return NULL;
 463		}
 464	}
 465	bio = cache->free_list;
 466	cache->free_list = bio->bi_next;
 467	cache->nr--;
 468	put_cpu();
 469
 470	if (nr_vecs)
 471		bio_init_inline(bio, bdev, nr_vecs, opf);
 472	else
 473		bio_init(bio, bdev, NULL, nr_vecs, opf);
 474	bio->bi_pool = bs;
 475	return bio;
 476}
 477
 478/**
 479 * bio_alloc_bioset - allocate a bio for I/O
 480 * @bdev:	block device to allocate the bio for (can be %NULL)
 481 * @nr_vecs:	number of bvecs to pre-allocate
 482 * @opf:	operation and flags for bio
 483 * @gfp_mask:   the GFP_* mask given to the slab allocator
 484 * @bs:		the bio_set to allocate from.
 485 *
 486 * Allocate a bio from the mempools in @bs.
 487 *
 488 * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to
 489 * allocate a bio.  This is due to the mempool guarantees.  To make this work,
 490 * callers must never allocate more than 1 bio at a time from the general pool.
 491 * Callers that need to allocate more than 1 bio must always submit the
 492 * previously allocated bio for IO before attempting to allocate a new one.
 493 * Failure to do so can cause deadlocks under memory pressure.
 494 *
 495 * Note that when running under submit_bio_noacct() (i.e. any block driver),
 496 * bios are not submitted until after you return - see the code in
 497 * submit_bio_noacct() that converts recursion into iteration, to prevent
 498 * stack overflows.
 499 *
 500 * This would normally mean allocating multiple bios under submit_bio_noacct()
 501 * would be susceptible to deadlocks, but we have
 502 * deadlock avoidance code that resubmits any blocked bios from a rescuer
 503 * thread.
 504 *
 505 * However, we do not guarantee forward progress for allocations from other
 506 * mempools. Doing multiple allocations from the same mempool under
 507 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad
 508 * for per bio allocations.
 509 *
 510 * Returns: Pointer to new bio on success, NULL on failure.
 511 */
 512struct bio *bio_alloc_bioset(struct block_device *bdev, unsigned short nr_vecs,
 513			     blk_opf_t opf, gfp_t gfp_mask,
 514			     struct bio_set *bs)
 515{
 516	gfp_t saved_gfp = gfp_mask;
 517	struct bio *bio;
 518	void *p;
 519
 520	/* should not use nobvec bioset for nr_vecs > 0 */
 521	if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_vecs > 0))
 522		return NULL;
 523
 524	if (bs->cache && nr_vecs <= BIO_INLINE_VECS) {
 525		opf |= REQ_ALLOC_CACHE;
 526		bio = bio_alloc_percpu_cache(bdev, nr_vecs, opf,
 527					     gfp_mask, bs);
 528		if (bio)
 529			return bio;
 530		/*
 531		 * No cached bio available, bio returned below marked with
 532		 * REQ_ALLOC_CACHE to participate in per-cpu alloc cache.
 533		 */
 534	} else
 535		opf &= ~REQ_ALLOC_CACHE;
 536
 537	/*
 538	 * submit_bio_noacct() converts recursion to iteration; this means if
 539	 * we're running beneath it, any bios we allocate and submit will not be
 540	 * submitted (and thus freed) until after we return.
 541	 *
 542	 * This exposes us to a potential deadlock if we allocate multiple bios
 543	 * from the same bio_set() while running underneath submit_bio_noacct().
 544	 * If we were to allocate multiple bios (say a stacking block driver
 545	 * that was splitting bios), we would deadlock if we exhausted the
 546	 * mempool's reserve.
 547	 *
 548	 * We solve this, and guarantee forward progress, with a rescuer
 549	 * workqueue per bio_set. If we go to allocate and there are bios on
 550	 * current->bio_list, we first try the allocation without
 551	 * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be
 552	 * blocking to the rescuer workqueue before we retry with the original
 553	 * gfp_flags.
 554	 */
 555	if (current->bio_list &&
 556	    (!bio_list_empty(&current->bio_list[0]) ||
 557	     !bio_list_empty(&current->bio_list[1])) &&
 558	    bs->rescue_workqueue)
 559		gfp_mask &= ~__GFP_DIRECT_RECLAIM;
 560
 561	p = mempool_alloc(&bs->bio_pool, gfp_mask);
 562	if (!p && gfp_mask != saved_gfp) {
 563		punt_bios_to_rescuer(bs);
 564		gfp_mask = saved_gfp;
 565		p = mempool_alloc(&bs->bio_pool, gfp_mask);
 566	}
 567	if (unlikely(!p))
 568		return NULL;
 569	if (!mempool_is_saturated(&bs->bio_pool))
 570		opf &= ~REQ_ALLOC_CACHE;
 571
 572	bio = p + bs->front_pad;
 573	if (nr_vecs > BIO_INLINE_VECS) {
 574		struct bio_vec *bvl = NULL;
 575
 576		bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);
 577		if (!bvl && gfp_mask != saved_gfp) {
 578			punt_bios_to_rescuer(bs);
 579			gfp_mask = saved_gfp;
 580			bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);
 581		}
 582		if (unlikely(!bvl))
 583			goto err_free;
 584
 585		bio_init(bio, bdev, bvl, nr_vecs, opf);
 586	} else if (nr_vecs) {
 587		bio_init_inline(bio, bdev, BIO_INLINE_VECS, opf);
 588	} else {
 589		bio_init(bio, bdev, NULL, 0, opf);
 590	}
 591
 592	bio->bi_pool = bs;
 593	return bio;
 594
 595err_free:
 596	mempool_free(p, &bs->bio_pool);
 597	return NULL;
 598}
 599EXPORT_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 */
 617struct 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}
 626EXPORT_SYMBOL(bio_kmalloc);
 627
 628void 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}
 636EXPORT_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 */
 648static 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 */
 700void 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
 722static 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
 738static 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
 748static 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
 761static 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
 779static 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;
 806out_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 **/
 819void 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}
 831EXPORT_SYMBOL(bio_put);
 832
 833static 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. The caller owns the returned
 864 * bio, but not the actual data it points to.
 865 *
 866 * The caller must ensure that the return bio is not freed before @bio_src.
 867 */
 868struct bio *bio_alloc_clone(struct block_device *bdev, struct bio *bio_src,
 869		gfp_t gfp, struct bio_set *bs)
 870{
 871	struct bio *bio;
 872
 873	bio = bio_alloc_bioset(bdev, 0, bio_src->bi_opf, gfp, bs);
 874	if (!bio)
 875		return NULL;
 876
 877	if (__bio_clone(bio, bio_src, gfp) < 0) {
 878		bio_put(bio);
 879		return NULL;
 880	}
 881	bio->bi_io_vec = bio_src->bi_io_vec;
 882
 883	return bio;
 884}
 885EXPORT_SYMBOL(bio_alloc_clone);
 886
 887/**
 888 * bio_init_clone - clone a bio that shares the original bio's biovec
 889 * @bdev: block_device to clone onto
 890 * @bio: bio to clone into
 891 * @bio_src: bio to clone from
 892 * @gfp: allocation priority
 893 *
 894 * Initialize a new bio in caller provided memory that is a clone of @bio_src.
 895 * The caller owns the returned bio, but not the actual data it points to.
 896 *
 897 * The caller must ensure that @bio_src is not freed before @bio.
 898 */
 899int bio_init_clone(struct block_device *bdev, struct bio *bio,
 900		struct bio *bio_src, gfp_t gfp)
 901{
 902	int ret;
 903
 904	bio_init(bio, bdev, bio_src->bi_io_vec, 0, bio_src->bi_opf);
 905	ret = __bio_clone(bio, bio_src, gfp);
 906	if (ret)
 907		bio_uninit(bio);
 908	return ret;
 909}
 910EXPORT_SYMBOL(bio_init_clone);
 911
 912/**
 913 * bio_full - check if the bio is full
 914 * @bio:	bio to check
 915 * @len:	length of one segment to be added
 916 *
 917 * Return true if @bio is full and one segment with @len bytes can't be
 918 * added to the bio, otherwise return false
 919 */
 920static inline bool bio_full(struct bio *bio, unsigned len)
 921{
 922	if (bio->bi_vcnt >= bio->bi_max_vecs)
 923		return true;
 924	if (bio->bi_iter.bi_size > UINT_MAX - len)
 925		return true;
 926	return false;
 927}
 928
 929static bool bvec_try_merge_page(struct bio_vec *bv, struct page *page,
 930		unsigned int len, unsigned int off)
 931{
 932	size_t bv_end = bv->bv_offset + bv->bv_len;
 933	phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1;
 934	phys_addr_t page_addr = page_to_phys(page);
 935
 936	if (vec_end_addr + 1 != page_addr + off)
 937		return false;
 938	if (xen_domain() && !xen_biovec_phys_mergeable(bv, page))
 939		return false;
 940
 941	if ((vec_end_addr & PAGE_MASK) != ((page_addr + off) & PAGE_MASK)) {
 942		if (IS_ENABLED(CONFIG_KMSAN))
 943			return false;
 944		if (bv->bv_page + bv_end / PAGE_SIZE != page + off / PAGE_SIZE)
 945			return false;
 946	}
 947
 948	bv->bv_len += len;
 949	return true;
 950}
 951
 952/*
 953 * Try to merge a page into a segment, while obeying the hardware segment
 954 * size limit.
 955 *
 956 * This is kept around for the integrity metadata, which is still tries
 957 * to build the initial bio to the hardware limit and doesn't have proper
 958 * helpers to split.  Hopefully this will go away soon.
 959 */
 960bool bvec_try_merge_hw_page(struct request_queue *q, struct bio_vec *bv,
 961		struct page *page, unsigned len, unsigned offset)
 962{
 963	unsigned long mask = queue_segment_boundary(q);
 964	phys_addr_t addr1 = bvec_phys(bv);
 965	phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;
 966
 967	if ((addr1 | mask) != (addr2 | mask))
 968		return false;
 969	if (len > queue_max_segment_size(q) - bv->bv_len)
 970		return false;
 971	return bvec_try_merge_page(bv, page, len, offset);
 972}
 973
 974/**
 975 * __bio_add_page - add page(s) to a bio in a new segment
 976 * @bio: destination bio
 977 * @page: start page to add
 978 * @len: length of the data to add, may cross pages
 979 * @off: offset of the data relative to @page, may cross pages
 980 *
 981 * Add the data at @page + @off to @bio as a new bvec.  The caller must ensure
 982 * that @bio has space for another bvec.
 983 */
 984void __bio_add_page(struct bio *bio, struct page *page,
 985		unsigned int len, unsigned int off)
 986{
 987	WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
 988	WARN_ON_ONCE(bio_full(bio, len));
 989
 990	if (is_pci_p2pdma_page(page))
 991		bio->bi_opf |= REQ_NOMERGE;
 992
 993	bvec_set_page(&bio->bi_io_vec[bio->bi_vcnt], page, len, off);
 994	bio->bi_iter.bi_size += len;
 995	bio->bi_vcnt++;
 996}
 997EXPORT_SYMBOL_GPL(__bio_add_page);
 998
 999/**
1000 * bio_add_virt_nofail - add data in the direct kernel mapping to a bio
1001 * @bio: destination bio
1002 * @vaddr: data to add
1003 * @len: length of the data to add, may cross pages
1004 *
1005 * Add the data at @vaddr to @bio.  The caller must have ensure a segment
1006 * is available for the added data.  No merging into an existing segment
1007 * will be performed.
1008 */
1009void bio_add_virt_nofail(struct bio *bio, void *vaddr, unsigned len)
1010{
1011	__bio_add_page(bio, virt_to_page(vaddr), len, offset_in_page(vaddr));
1012}
1013EXPORT_SYMBOL_GPL(bio_add_virt_nofail);
1014
1015/**
1016 *	bio_add_page	-	attempt to add page(s) to bio
1017 *	@bio: destination bio
1018 *	@page: start page to add
1019 *	@len: vec entry length, may cross pages
1020 *	@offset: vec entry offset relative to @page, may cross pages
1021 *
1022 *	Attempt to add page(s) to the bio_vec maplist. This will only fail
1023 *	if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
1024 */
1025int bio_add_page(struct bio *bio, struct page *page,
1026		 unsigned int len, unsigned int offset)
1027{
1028	if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
1029		return 0;
1030	if (bio->bi_iter.bi_size > UINT_MAX - len)
1031		return 0;
1032
1033	if (bio->bi_vcnt > 0) {
1034		struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
1035
1036		if (!zone_device_pages_have_same_pgmap(bv->bv_page, page))
1037			return 0;
1038
1039		if (bvec_try_merge_page(bv, page, len, offset)) {
1040			bio->bi_iter.bi_size += len;
1041			return len;
1042		}
1043	}
1044
1045	if (bio->bi_vcnt >= bio->bi_max_vecs)
1046		return 0;
1047	__bio_add_page(bio, page, len, offset);
1048	return len;
1049}
1050EXPORT_SYMBOL(bio_add_page);
1051
1052void bio_add_folio_nofail(struct bio *bio, struct folio *folio, size_t len,
1053			  size_t off)
1054{
1055	unsigned long nr = off / PAGE_SIZE;
1056
1057	WARN_ON_ONCE(len > UINT_MAX);
1058	__bio_add_page(bio, folio_page(folio, nr), len, off % PAGE_SIZE);
1059}
1060EXPORT_SYMBOL_GPL(bio_add_folio_nofail);
1061
1062/**
1063 * bio_add_folio - Attempt to add part of a folio to a bio.
1064 * @bio: BIO to add to.
1065 * @folio: Folio to add.
1066 * @len: How many bytes from the folio to add.
1067 * @off: First byte in this folio to add.
1068 *
1069 * Filesystems that use folios can call this function instead of calling
1070 * bio_add_page() for each page in the folio.  If @off is bigger than
1071 * PAGE_SIZE, this function can create a bio_vec that starts in a page
1072 * after the bv_page.  BIOs do not support folios that are 4GiB or larger.
1073 *
1074 * Return: Whether the addition was successful.
1075 */
1076bool bio_add_folio(struct bio *bio, struct folio *folio, size_t len,
1077		   size_t off)
1078{
1079	unsigned long nr = off / PAGE_SIZE;
1080
1081	if (len > UINT_MAX)
1082		return false;
1083	return bio_add_page(bio, folio_page(folio, nr), len, off % PAGE_SIZE) > 0;
1084}
1085EXPORT_SYMBOL(bio_add_folio);
1086
1087/**
1088 * bio_add_vmalloc_chunk - add a vmalloc chunk to a bio
1089 * @bio: destination bio
1090 * @vaddr: vmalloc address to add
1091 * @len: total length in bytes of the data to add
1092 *
1093 * Add data starting at @vaddr to @bio and return how many bytes were added.
1094 * This may be less than the amount originally asked.  Returns 0 if no data
1095 * could be added to @bio.
1096 *
1097 * This helper calls flush_kernel_vmap_range() for the range added.  For reads
1098 * the caller still needs to manually call invalidate_kernel_vmap_range() in
1099 * the completion handler.
1100 */
1101unsigned int bio_add_vmalloc_chunk(struct bio *bio, void *vaddr, unsigned len)
1102{
1103	unsigned int offset = offset_in_page(vaddr);
1104
1105	len = min(len, PAGE_SIZE - offset);
1106	if (bio_add_page(bio, vmalloc_to_page(vaddr), len, offset) < len)
1107		return 0;
1108	if (op_is_write(bio_op(bio)))
1109		flush_kernel_vmap_range(vaddr, len);
1110	return len;
1111}
1112EXPORT_SYMBOL_GPL(bio_add_vmalloc_chunk);
1113
1114/**
1115 * bio_add_vmalloc - add a vmalloc region to a bio
1116 * @bio: destination bio
1117 * @vaddr: vmalloc address to add
1118 * @len: total length in bytes of the data to add
1119 *
1120 * Add data starting at @vaddr to @bio.  Return %true on success or %false if
1121 * @bio does not have enough space for the payload.
1122 *
1123 * This helper calls flush_kernel_vmap_range() for the range added.  For reads
1124 * the caller still needs to manually call invalidate_kernel_vmap_range() in
1125 * the completion handler.
1126 */
1127bool bio_add_vmalloc(struct bio *bio, void *vaddr, unsigned int len)
1128{
1129	do {
1130		unsigned int added = bio_add_vmalloc_chunk(bio, vaddr, len);
1131
1132		if (!added)
1133			return false;
1134		vaddr += added;
1135		len -= added;
1136	} while (len);
1137
1138	return true;
1139}
1140EXPORT_SYMBOL_GPL(bio_add_vmalloc);
1141
1142void __bio_release_pages(struct bio *bio, bool mark_dirty)
1143{
1144	struct folio_iter fi;
1145
1146	bio_for_each_folio_all(fi, bio) {
1147		size_t nr_pages;
1148
1149		if (mark_dirty) {
1150			folio_lock(fi.folio);
1151			folio_mark_dirty(fi.folio);
1152			folio_unlock(fi.folio);
1153		}
1154		nr_pages = (fi.offset + fi.length - 1) / PAGE_SIZE -
1155			   fi.offset / PAGE_SIZE + 1;
1156		unpin_user_folio(fi.folio, nr_pages);
1157	}
1158}
1159EXPORT_SYMBOL_GPL(__bio_release_pages);
1160
1161void bio_iov_bvec_set(struct bio *bio, const struct iov_iter *iter)
1162{
1163	WARN_ON_ONCE(bio->bi_max_vecs);
1164
1165	bio->bi_vcnt = iter->nr_segs;
1166	bio->bi_io_vec = (struct bio_vec *)iter->bvec;
1167	bio->bi_iter.bi_bvec_done = iter->iov_offset;
1168	bio->bi_iter.bi_size = iov_iter_count(iter);
1169	bio_set_flag(bio, BIO_CLONED);
1170}
1171
1172static unsigned int get_contig_folio_len(unsigned int *num_pages,
1173					 struct page **pages, unsigned int i,
1174					 struct folio *folio, size_t left,
1175					 size_t offset)
1176{
1177	size_t bytes = left;
1178	size_t contig_sz = min_t(size_t, PAGE_SIZE - offset, bytes);
1179	unsigned int j;
1180
1181	/*
1182	 * We might COW a single page in the middle of
1183	 * a large folio, so we have to check that all
1184	 * pages belong to the same folio.
1185	 */
1186	bytes -= contig_sz;
1187	for (j = i + 1; j < i + *num_pages; j++) {
1188		size_t next = min_t(size_t, PAGE_SIZE, bytes);
1189
1190		if (page_folio(pages[j]) != folio ||
1191		    pages[j] != pages[j - 1] + 1) {
1192			break;
1193		}
1194		contig_sz += next;
1195		bytes -= next;
1196	}
1197	*num_pages = j - i;
1198
1199	return contig_sz;
1200}
1201
1202#define PAGE_PTRS_PER_BVEC     (sizeof(struct bio_vec) / sizeof(struct page *))
1203
1204/**
1205 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
1206 * @bio: bio to add pages to
1207 * @iter: iov iterator describing the region to be mapped
1208 *
1209 * Extracts pages from *iter and appends them to @bio's bvec array.  The pages
1210 * will have to be cleaned up in the way indicated by the BIO_PAGE_PINNED flag.
1211 * For a multi-segment *iter, this function only adds pages from the next
1212 * non-empty segment of the iov iterator.
1213 */
1214static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1215{
1216	iov_iter_extraction_t extraction_flags = 0;
1217	unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1218	unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1219	struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1220	struct page **pages = (struct page **)bv;
1221	ssize_t size;
1222	unsigned int num_pages, i = 0;
1223	size_t offset, folio_offset, left, len;
1224	int ret = 0;
1225
1226	/*
1227	 * Move page array up in the allocated memory for the bio vecs as far as
1228	 * possible so that we can start filling biovecs from the beginning
1229	 * without overwriting the temporary page array.
1230	 */
1231	BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1232	pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1233
1234	if (bio->bi_bdev && blk_queue_pci_p2pdma(bio->bi_bdev->bd_disk->queue))
1235		extraction_flags |= ITER_ALLOW_P2PDMA;
1236
1237	size = iov_iter_extract_pages(iter, &pages,
1238				      UINT_MAX - bio->bi_iter.bi_size,
1239				      nr_pages, extraction_flags, &offset);
1240	if (unlikely(size <= 0))
1241		return size ? size : -EFAULT;
1242
1243	nr_pages = DIV_ROUND_UP(offset + size, PAGE_SIZE);
1244	for (left = size, i = 0; left > 0; left -= len, i += num_pages) {
1245		struct page *page = pages[i];
1246		struct folio *folio = page_folio(page);
1247		unsigned int old_vcnt = bio->bi_vcnt;
1248
1249		folio_offset = ((size_t)folio_page_idx(folio, page) <<
1250			       PAGE_SHIFT) + offset;
1251
1252		len = min(folio_size(folio) - folio_offset, left);
1253
1254		num_pages = DIV_ROUND_UP(offset + len, PAGE_SIZE);
1255
1256		if (num_pages > 1)
1257			len = get_contig_folio_len(&num_pages, pages, i,
1258						   folio, left, offset);
1259
1260		if (!bio_add_folio(bio, folio, len, folio_offset)) {
1261			WARN_ON_ONCE(1);
1262			ret = -EINVAL;
1263			goto out;
1264		}
1265
1266		if (bio_flagged(bio, BIO_PAGE_PINNED)) {
1267			/*
1268			 * We're adding another fragment of a page that already
1269			 * was part of the last segment.  Undo our pin as the
1270			 * page was pinned when an earlier fragment of it was
1271			 * added to the bio and __bio_release_pages expects a
1272			 * single pin per page.
1273			 */
1274			if (offset && bio->bi_vcnt == old_vcnt)
1275				unpin_user_folio(folio, 1);
1276		}
1277		offset = 0;
1278	}
1279
1280	iov_iter_revert(iter, left);
1281out:
1282	while (i < nr_pages)
1283		bio_release_page(bio, pages[i++]);
1284
1285	return ret;
1286}
1287
1288/*
1289 * Aligns the bio size to the len_align_mask, releasing excessive bio vecs that
1290 * __bio_iov_iter_get_pages may have inserted, and reverts the trimmed length
1291 * for the next iteration.
1292 */
1293static int bio_iov_iter_align_down(struct bio *bio, struct iov_iter *iter,
1294			    unsigned len_align_mask)
1295{
1296	size_t nbytes = bio->bi_iter.bi_size & len_align_mask;
1297
1298	if (!nbytes)
1299		return 0;
1300
1301	iov_iter_revert(iter, nbytes);
1302	bio->bi_iter.bi_size -= nbytes;
1303	do {
1304		struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
1305
1306		if (nbytes < bv->bv_len) {
1307			bv->bv_len -= nbytes;
1308			break;
1309		}
1310
1311		bio_release_page(bio, bv->bv_page);
1312		bio->bi_vcnt--;
1313		nbytes -= bv->bv_len;
1314	} while (nbytes);
1315
1316	if (!bio->bi_vcnt)
1317		return -EFAULT;
1318	return 0;
1319}
1320
1321/**
1322 * bio_iov_iter_get_pages - add user or kernel pages to a bio
1323 * @bio: bio to add pages to
1324 * @iter: iov iterator describing the region to be added
1325 * @len_align_mask: the mask to align the total size to, 0 for any length
1326 *
1327 * This takes either an iterator pointing to user memory, or one pointing to
1328 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1329 * map them into the kernel. On IO completion, the caller should put those
1330 * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided
1331 * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs
1332 * to ensure the bvecs and pages stay referenced until the submitted I/O is
1333 * completed by a call to ->ki_complete() or returns with an error other than
1334 * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF
1335 * on IO completion. If it isn't, then pages should be released.
1336 *
1337 * The function tries, but does not guarantee, to pin as many pages as
1338 * fit into the bio, or are requested in @iter, whatever is smaller. If
1339 * MM encounters an error pinning the requested pages, it stops. Error
1340 * is returned only if 0 pages could be pinned.
1341 */
1342int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter,
1343			   unsigned len_align_mask)
1344{
1345	int ret = 0;
1346
1347	if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
1348		return -EIO;
1349
1350	if (iov_iter_is_bvec(iter)) {
1351		bio_iov_bvec_set(bio, iter);
1352		iov_iter_advance(iter, bio->bi_iter.bi_size);
1353		return 0;
1354	}
1355
1356	if (iov_iter_extract_will_pin(iter))
1357		bio_set_flag(bio, BIO_PAGE_PINNED);
1358	do {
1359		ret = __bio_iov_iter_get_pages(bio, iter);
1360	} while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));
1361
1362	if (bio->bi_vcnt)
1363		return bio_iov_iter_align_down(bio, iter, len_align_mask);
1364	return ret;
1365}
1366
1367static void submit_bio_wait_endio(struct bio *bio)
1368{
1369	complete(bio->bi_private);
1370}
1371
1372/**
1373 * submit_bio_wait - submit a bio, and wait until it completes
1374 * @bio: The &struct bio which describes the I/O
1375 *
1376 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1377 * bio_endio() on failure.
1378 *
1379 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1380 * result in bio reference to be consumed. The caller must drop the reference
1381 * on his own.
1382 */
1383int submit_bio_wait(struct bio *bio)
1384{
1385	DECLARE_COMPLETION_ONSTACK_MAP(done,
1386			bio->bi_bdev->bd_disk->lockdep_map);
1387
1388	bio->bi_private = &done;
1389	bio->bi_end_io = submit_bio_wait_endio;
1390	bio->bi_opf |= REQ_SYNC;
1391	submit_bio(bio);
1392	blk_wait_io(&done);
1393
1394	return blk_status_to_errno(bio->bi_status);
1395}
1396EXPORT_SYMBOL(submit_bio_wait);
1397
1398/**
1399 * bdev_rw_virt - synchronously read into / write from kernel mapping
1400 * @bdev:	block device to access
1401 * @sector:	sector to access
1402 * @data:	data to read/write
1403 * @len:	length in byte to read/write
1404 * @op:		operation (e.g. REQ_OP_READ/REQ_OP_WRITE)
1405 *
1406 * Performs synchronous I/O to @bdev for @data/@len.  @data must be in
1407 * the kernel direct mapping and not a vmalloc address.
1408 */
1409int bdev_rw_virt(struct block_device *bdev, sector_t sector, void *data,
1410		size_t len, enum req_op op)
1411{
1412	struct bio_vec bv;
1413	struct bio bio;
1414	int error;
1415
1416	if (WARN_ON_ONCE(is_vmalloc_addr(data)))
1417		return -EIO;
1418
1419	bio_init(&bio, bdev, &bv, 1, op);
1420	bio.bi_iter.bi_sector = sector;
1421	bio_add_virt_nofail(&bio, data, len);
1422	error = submit_bio_wait(&bio);
1423	bio_uninit(&bio);
1424	return error;
1425}
1426EXPORT_SYMBOL_GPL(bdev_rw_virt);
1427
1428static void bio_wait_end_io(struct bio *bio)
1429{
1430	complete(bio->bi_private);
1431	bio_put(bio);
1432}
1433
1434/*
1435 * bio_await_chain - ends @bio and waits for every chained bio to complete
1436 */
1437void bio_await_chain(struct bio *bio)
1438{
1439	DECLARE_COMPLETION_ONSTACK_MAP(done,
1440			bio->bi_bdev->bd_disk->lockdep_map);
1441
1442	bio->bi_private = &done;
1443	bio->bi_end_io = bio_wait_end_io;
1444	bio_endio(bio);
1445	blk_wait_io(&done);
1446}
1447
1448void __bio_advance(struct bio *bio, unsigned bytes)
1449{
1450	if (bio_integrity(bio))
1451		bio_integrity_advance(bio, bytes);
1452
1453	bio_crypt_advance(bio, bytes);
1454	bio_advance_iter(bio, &bio->bi_iter, bytes);
1455}
1456EXPORT_SYMBOL(__bio_advance);
1457
1458void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1459			struct bio *src, struct bvec_iter *src_iter)
1460{
1461	while (src_iter->bi_size && dst_iter->bi_size) {
1462		struct bio_vec src_bv = bio_iter_iovec(src, *src_iter);
1463		struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter);
1464		unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len);
1465		void *src_buf = bvec_kmap_local(&src_bv);
1466		void *dst_buf = bvec_kmap_local(&dst_bv);
1467
1468		memcpy(dst_buf, src_buf, bytes);
1469
1470		kunmap_local(dst_buf);
1471		kunmap_local(src_buf);
1472
1473		bio_advance_iter_single(src, src_iter, bytes);
1474		bio_advance_iter_single(dst, dst_iter, bytes);
1475	}
1476}
1477EXPORT_SYMBOL(bio_copy_data_iter);
1478
1479/**
1480 * bio_copy_data - copy contents of data buffers from one bio to another
1481 * @src: source bio
1482 * @dst: destination bio
1483 *
1484 * Stops when it reaches the end of either @src or @dst - that is, copies
1485 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1486 */
1487void bio_copy_data(struct bio *dst, struct bio *src)
1488{
1489	struct bvec_iter src_iter = src->bi_iter;
1490	struct bvec_iter dst_iter = dst->bi_iter;
1491
1492	bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1493}
1494EXPORT_SYMBOL(bio_copy_data);
1495
1496void bio_free_pages(struct bio *bio)
1497{
1498	struct bio_vec *bvec;
1499	struct bvec_iter_all iter_all;
1500
1501	bio_for_each_segment_all(bvec, bio, iter_all)
1502		__free_page(bvec->bv_page);
1503}
1504EXPORT_SYMBOL(bio_free_pages);
1505
1506/*
1507 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1508 * for performing direct-IO in BIOs.
1509 *
1510 * The problem is that we cannot run folio_mark_dirty() from interrupt context
1511 * because the required locks are not interrupt-safe.  So what we can do is to
1512 * mark the pages dirty _before_ performing IO.  And in interrupt context,
1513 * check that the pages are still dirty.   If so, fine.  If not, redirty them
1514 * in process context.
1515 *
1516 * Note that this code is very hard to test under normal circumstances because
1517 * direct-io pins the pages with get_user_pages().  This makes
1518 * is_page_cache_freeable return false, and the VM will not clean the pages.
1519 * But other code (eg, flusher threads) could clean the pages if they are mapped
1520 * pagecache.
1521 *
1522 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1523 * deferred bio dirtying paths.
1524 */
1525
1526/*
1527 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1528 */
1529void bio_set_pages_dirty(struct bio *bio)
1530{
1531	struct folio_iter fi;
1532
1533	bio_for_each_folio_all(fi, bio) {
1534		folio_lock(fi.folio);
1535		folio_mark_dirty(fi.folio);
1536		folio_unlock(fi.folio);
1537	}
1538}
1539EXPORT_SYMBOL_GPL(bio_set_pages_dirty);
1540
1541/*
1542 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1543 * If they are, then fine.  If, however, some pages are clean then they must
1544 * have been written out during the direct-IO read.  So we take another ref on
1545 * the BIO and re-dirty the pages in process context.
1546 *
1547 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1548 * here on.  It will unpin each page and will run one bio_put() against the
1549 * BIO.
1550 */
1551
1552static void bio_dirty_fn(struct work_struct *work);
1553
1554static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1555static DEFINE_SPINLOCK(bio_dirty_lock);
1556static struct bio *bio_dirty_list;
1557
1558/*
1559 * This runs in process context
1560 */
1561static void bio_dirty_fn(struct work_struct *work)
1562{
1563	struct bio *bio, *next;
1564
1565	spin_lock_irq(&bio_dirty_lock);
1566	next = bio_dirty_list;
1567	bio_dirty_list = NULL;
1568	spin_unlock_irq(&bio_dirty_lock);
1569
1570	while ((bio = next) != NULL) {
1571		next = bio->bi_private;
1572
1573		bio_release_pages(bio, true);
1574		bio_put(bio);
1575	}
1576}
1577
1578void bio_check_pages_dirty(struct bio *bio)
1579{
1580	struct folio_iter fi;
1581	unsigned long flags;
1582
1583	bio_for_each_folio_all(fi, bio) {
1584		if (!folio_test_dirty(fi.folio))
1585			goto defer;
1586	}
1587
1588	bio_release_pages(bio, false);
1589	bio_put(bio);
1590	return;
1591defer:
1592	spin_lock_irqsave(&bio_dirty_lock, flags);
1593	bio->bi_private = bio_dirty_list;
1594	bio_dirty_list = bio;
1595	spin_unlock_irqrestore(&bio_dirty_lock, flags);
1596	schedule_work(&bio_dirty_work);
1597}
1598EXPORT_SYMBOL_GPL(bio_check_pages_dirty);
1599
1600static inline bool bio_remaining_done(struct bio *bio)
1601{
1602	/*
1603	 * If we're not chaining, then ->__bi_remaining is always 1 and
1604	 * we always end io on the first invocation.
1605	 */
1606	if (!bio_flagged(bio, BIO_CHAIN))
1607		return true;
1608
1609	BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1610
1611	if (atomic_dec_and_test(&bio->__bi_remaining)) {
1612		bio_clear_flag(bio, BIO_CHAIN);
1613		return true;
1614	}
1615
1616	return false;
1617}
1618
1619/**
1620 * bio_endio - end I/O on a bio
1621 * @bio:	bio
1622 *
1623 * Description:
1624 *   bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1625 *   way to end I/O on a bio. No one should call bi_end_io() directly on a
1626 *   bio unless they own it and thus know that it has an end_io function.
1627 *
1628 *   bio_endio() can be called several times on a bio that has been chained
1629 *   using bio_chain().  The ->bi_end_io() function will only be called the
1630 *   last time.
1631 **/
1632void bio_endio(struct bio *bio)
1633{
1634again:
1635	if (!bio_remaining_done(bio))
1636		return;
1637	if (!bio_integrity_endio(bio))
1638		return;
1639
1640	blk_zone_bio_endio(bio);
1641
1642	rq_qos_done_bio(bio);
1643
1644	if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1645		trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio);
1646		bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1647	}
1648
1649	/*
1650	 * Need to have a real endio function for chained bios, otherwise
1651	 * various corner cases will break (like stacking block devices that
1652	 * save/restore bi_end_io) - however, we want to avoid unbounded
1653	 * recursion and blowing the stack. Tail call optimization would
1654	 * handle this, but compiling with frame pointers also disables
1655	 * gcc's sibling call optimization.
1656	 */
1657	if (bio->bi_end_io == bio_chain_endio) {
1658		bio = __bio_chain_endio(bio);
1659		goto again;
1660	}
1661
1662#ifdef CONFIG_BLK_CGROUP
1663	/*
1664	 * Release cgroup info.  We shouldn't have to do this here, but quite
1665	 * a few callers of bio_init fail to call bio_uninit, so we cover up
1666	 * for that here at least for now.
1667	 */
1668	if (bio->bi_blkg) {
1669		blkg_put(bio->bi_blkg);
1670		bio->bi_blkg = NULL;
1671	}
1672#endif
1673
1674	if (bio->bi_end_io)
1675		bio->bi_end_io(bio);
1676}
1677EXPORT_SYMBOL(bio_endio);
1678
1679/**
1680 * bio_split - split a bio
1681 * @bio:	bio to split
1682 * @sectors:	number of sectors to split from the front of @bio
1683 * @gfp:	gfp mask
1684 * @bs:		bio set to allocate from
1685 *
1686 * Allocates and returns a new bio which represents @sectors from the start of
1687 * @bio, and updates @bio to represent the remaining sectors.
1688 *
1689 * Unless this is a discard request the newly allocated bio will point
1690 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1691 * neither @bio nor @bs are freed before the split bio.
1692 */
1693struct bio *bio_split(struct bio *bio, int sectors,
1694		      gfp_t gfp, struct bio_set *bs)
1695{
1696	struct bio *split;
1697
1698	if (WARN_ON_ONCE(sectors <= 0))
1699		return ERR_PTR(-EINVAL);
1700	if (WARN_ON_ONCE(sectors >= bio_sectors(bio)))
1701		return ERR_PTR(-EINVAL);
1702
1703	/* Zone append commands cannot be split */
1704	if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))
1705		return ERR_PTR(-EINVAL);
1706
1707	/* atomic writes cannot be split */
1708	if (bio->bi_opf & REQ_ATOMIC)
1709		return ERR_PTR(-EINVAL);
1710
1711	split = bio_alloc_clone(bio->bi_bdev, bio, gfp, bs);
1712	if (!split)
1713		return ERR_PTR(-ENOMEM);
1714
1715	split->bi_iter.bi_size = sectors << 9;
1716
1717	if (bio_integrity(split))
1718		bio_integrity_trim(split);
1719
1720	bio_advance(bio, split->bi_iter.bi_size);
1721
1722	if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1723		bio_set_flag(split, BIO_TRACE_COMPLETION);
1724
1725	return split;
1726}
1727EXPORT_SYMBOL(bio_split);
1728
1729/**
1730 * bio_trim - trim a bio
1731 * @bio:	bio to trim
1732 * @offset:	number of sectors to trim from the front of @bio
1733 * @size:	size we want to trim @bio to, in sectors
1734 *
1735 * This function is typically used for bios that are cloned and submitted
1736 * to the underlying device in parts.
1737 */
1738void bio_trim(struct bio *bio, sector_t offset, sector_t size)
1739{
1740	/* We should never trim an atomic write */
1741	if (WARN_ON_ONCE(bio->bi_opf & REQ_ATOMIC && size))
1742		return;
1743
1744	if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS ||
1745			 offset + size > bio_sectors(bio)))
1746		return;
1747
1748	size <<= 9;
1749	if (offset == 0 && size == bio->bi_iter.bi_size)
1750		return;
1751
1752	bio_advance(bio, offset << 9);
1753	bio->bi_iter.bi_size = size;
1754
1755	if (bio_integrity(bio))
1756		bio_integrity_trim(bio);
1757}
1758EXPORT_SYMBOL_GPL(bio_trim);
1759
1760/*
1761 * create memory pools for biovec's in a bio_set.
1762 * use the global biovec slabs created for general use.
1763 */
1764int biovec_init_pool(mempool_t *pool, int pool_entries)
1765{
1766	struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1;
1767
1768	return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1769}
1770
1771/*
1772 * bioset_exit - exit a bioset initialized with bioset_init()
1773 *
1774 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1775 * kzalloc()).
1776 */
1777void bioset_exit(struct bio_set *bs)
1778{
1779	bio_alloc_cache_destroy(bs);
1780	if (bs->rescue_workqueue)
1781		destroy_workqueue(bs->rescue_workqueue);
1782	bs->rescue_workqueue = NULL;
1783
1784	mempool_exit(&bs->bio_pool);
1785	mempool_exit(&bs->bvec_pool);
1786
1787	if (bs->bio_slab)
1788		bio_put_slab(bs);
1789	bs->bio_slab = NULL;
1790}
1791EXPORT_SYMBOL(bioset_exit);
1792
1793/**
1794 * bioset_init - Initialize a bio_set
1795 * @bs:		pool to initialize
1796 * @pool_size:	Number of bio and bio_vecs to cache in the mempool
1797 * @front_pad:	Number of bytes to allocate in front of the returned bio
1798 * @flags:	Flags to modify behavior, currently %BIOSET_NEED_BVECS
1799 *              and %BIOSET_NEED_RESCUER
1800 *
1801 * Description:
1802 *    Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1803 *    to ask for a number of bytes to be allocated in front of the bio.
1804 *    Front pad allocation is useful for embedding the bio inside
1805 *    another structure, to avoid allocating extra data to go with the bio.
1806 *    Note that the bio must be embedded at the END of that structure always,
1807 *    or things will break badly.
1808 *    If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1809 *    for allocating iovecs.  This pool is not needed e.g. for bio_init_clone().
1810 *    If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used
1811 *    to dispatch queued requests when the mempool runs out of space.
1812 *
1813 */
1814int bioset_init(struct bio_set *bs,
1815		unsigned int pool_size,
1816		unsigned int front_pad,
1817		int flags)
1818{
1819	bs->front_pad = front_pad;
1820	if (flags & BIOSET_NEED_BVECS)
1821		bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1822	else
1823		bs->back_pad = 0;
1824
1825	spin_lock_init(&bs->rescue_lock);
1826	bio_list_init(&bs->rescue_list);
1827	INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1828
1829	bs->bio_slab = bio_find_or_create_slab(bs);
1830	if (!bs->bio_slab)
1831		return -ENOMEM;
1832
1833	if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
1834		goto bad;
1835
1836	if ((flags & BIOSET_NEED_BVECS) &&
1837	    biovec_init_pool(&bs->bvec_pool, pool_size))
1838		goto bad;
1839
1840	if (flags & BIOSET_NEED_RESCUER) {
1841		bs->rescue_workqueue = alloc_workqueue("bioset",
1842							WQ_MEM_RECLAIM, 0);
1843		if (!bs->rescue_workqueue)
1844			goto bad;
1845	}
1846	if (flags & BIOSET_PERCPU_CACHE) {
1847		bs->cache = alloc_percpu(struct bio_alloc_cache);
1848		if (!bs->cache)
1849			goto bad;
1850		cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
1851	}
1852
1853	return 0;
1854bad:
1855	bioset_exit(bs);
1856	return -ENOMEM;
1857}
1858EXPORT_SYMBOL(bioset_init);
1859
1860static int __init init_bio(void)
1861{
1862	int i;
1863
1864	BUILD_BUG_ON(BIO_FLAG_LAST > 8 * sizeof_field(struct bio, bi_flags));
1865
1866	for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) {
1867		struct biovec_slab *bvs = bvec_slabs + i;
1868
1869		bvs->slab = kmem_cache_create(bvs->name,
1870				bvs->nr_vecs * sizeof(struct bio_vec), 0,
1871				SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1872	}
1873
1874	cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL,
1875					bio_cpu_dead);
1876
1877	if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0,
1878			BIOSET_NEED_BVECS | BIOSET_PERCPU_CACHE))
1879		panic("bio: can't allocate bios\n");
1880
1881	return 0;
1882}
1883subsys_initcall(init_bio);