1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * Copyright (C) 2001 Jens Axboe <axboe@kernel.dk>
   4 */
   5#include <linux/mm.h>
   6#include <linux/swap.h>
   7#include <linux/bio.h>
   8#include <linux/blkdev.h>
   9#include <linux/uio.h>
  10#include <linux/iocontext.h>
  11#include <linux/slab.h>
  12#include <linux/init.h>
  13#include <linux/kernel.h>
  14#include <linux/export.h>
  15#include <linux/mempool.h>
  16#include <linux/workqueue.h>
  17#include <linux/cgroup.h>
  18#include <linux/highmem.h>
  19#include <linux/sched/sysctl.h>
  20#include <linux/blk-crypto.h>
  21#include <linux/xarray.h>
  22
  23#include <trace/events/block.h>
  24#include "blk.h"
  25#include "blk-rq-qos.h"
  26#include "blk-cgroup.h"
  27
  28#define ALLOC_CACHE_THRESHOLD	16
 
  29#define ALLOC_CACHE_MAX		256
  30
  31struct bio_alloc_cache {
  32	struct bio		*free_list;
  33	struct bio		*free_list_irq;
  34	unsigned int		nr;
  35	unsigned int		nr_irq;
  36};
  37
  38static struct biovec_slab {
  39	int nr_vecs;
  40	char *name;
  41	struct kmem_cache *slab;
  42} bvec_slabs[] __read_mostly = {
  43	{ .nr_vecs = 16, .name = "biovec-16" },
  44	{ .nr_vecs = 64, .name = "biovec-64" },
  45	{ .nr_vecs = 128, .name = "biovec-128" },
  46	{ .nr_vecs = BIO_MAX_VECS, .name = "biovec-max" },
  47};
  48
  49static struct biovec_slab *biovec_slab(unsigned short nr_vecs)
  50{
  51	switch (nr_vecs) {
  52	/* smaller bios use inline vecs */
  53	case 5 ... 16:
  54		return &bvec_slabs[0];
  55	case 17 ... 64:
  56		return &bvec_slabs[1];
  57	case 65 ... 128:
  58		return &bvec_slabs[2];
  59	case 129 ... BIO_MAX_VECS:
  60		return &bvec_slabs[3];
  61	default:
  62		BUG();
  63		return NULL;
  64	}
  65}
  66
  67/*
  68 * fs_bio_set is the bio_set containing bio and iovec memory pools used by
  69 * IO code that does not need private memory pools.
  70 */
  71struct bio_set fs_bio_set;
  72EXPORT_SYMBOL(fs_bio_set);
  73
  74/*
  75 * Our slab pool management
  76 */
  77struct bio_slab {
  78	struct kmem_cache *slab;
  79	unsigned int slab_ref;
  80	unsigned int slab_size;
  81	char name[8];
  82};
  83static DEFINE_MUTEX(bio_slab_lock);
  84static DEFINE_XARRAY(bio_slabs);
  85
  86static struct bio_slab *create_bio_slab(unsigned int size)
  87{
  88	struct bio_slab *bslab = kzalloc(sizeof(*bslab), GFP_KERNEL);
  89
  90	if (!bslab)
  91		return NULL;
  92
  93	snprintf(bslab->name, sizeof(bslab->name), "bio-%d", size);
  94	bslab->slab = kmem_cache_create(bslab->name, size,
  95			ARCH_KMALLOC_MINALIGN,
  96			SLAB_HWCACHE_ALIGN | SLAB_TYPESAFE_BY_RCU, NULL);
  97	if (!bslab->slab)
  98		goto fail_alloc_slab;
  99
 100	bslab->slab_ref = 1;
 101	bslab->slab_size = size;
 102
 103	if (!xa_err(xa_store(&bio_slabs, size, bslab, GFP_KERNEL)))
 104		return bslab;
 105
 106	kmem_cache_destroy(bslab->slab);
 107
 108fail_alloc_slab:
 109	kfree(bslab);
 110	return NULL;
 111}
 112
 113static inline unsigned int bs_bio_slab_size(struct bio_set *bs)
 114{
 115	return bs->front_pad + sizeof(struct bio) + bs->back_pad;
 116}
 117
 118static struct kmem_cache *bio_find_or_create_slab(struct bio_set *bs)
 119{
 120	unsigned int size = bs_bio_slab_size(bs);
 121	struct bio_slab *bslab;
 122
 123	mutex_lock(&bio_slab_lock);
 124	bslab = xa_load(&bio_slabs, size);
 125	if (bslab)
 126		bslab->slab_ref++;
 127	else
 128		bslab = create_bio_slab(size);
 129	mutex_unlock(&bio_slab_lock);
 130
 131	if (bslab)
 132		return bslab->slab;
 133	return NULL;
 134}
 135
 136static void bio_put_slab(struct bio_set *bs)
 137{
 138	struct bio_slab *bslab = NULL;
 139	unsigned int slab_size = bs_bio_slab_size(bs);
 140
 141	mutex_lock(&bio_slab_lock);
 142
 143	bslab = xa_load(&bio_slabs, slab_size);
 144	if (WARN(!bslab, KERN_ERR "bio: unable to find slab!\n"))
 145		goto out;
 146
 147	WARN_ON_ONCE(bslab->slab != bs->bio_slab);
 148
 149	WARN_ON(!bslab->slab_ref);
 150
 151	if (--bslab->slab_ref)
 152		goto out;
 153
 154	xa_erase(&bio_slabs, slab_size);
 155
 156	kmem_cache_destroy(bslab->slab);
 157	kfree(bslab);
 158
 159out:
 160	mutex_unlock(&bio_slab_lock);
 161}
 162
 163void bvec_free(mempool_t *pool, struct bio_vec *bv, unsigned short nr_vecs)
 164{
 165	BUG_ON(nr_vecs > BIO_MAX_VECS);
 166
 167	if (nr_vecs == BIO_MAX_VECS)
 168		mempool_free(bv, pool);
 169	else if (nr_vecs > BIO_INLINE_VECS)
 170		kmem_cache_free(biovec_slab(nr_vecs)->slab, bv);
 171}
 172
 173/*
 174 * Make the first allocation restricted and don't dump info on allocation
 175 * failures, since we'll fall back to the mempool in case of failure.
 176 */
 177static inline gfp_t bvec_alloc_gfp(gfp_t gfp)
 178{
 179	return (gfp & ~(__GFP_DIRECT_RECLAIM | __GFP_IO)) |
 180		__GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN;
 181}
 182
 183struct bio_vec *bvec_alloc(mempool_t *pool, unsigned short *nr_vecs,
 184		gfp_t gfp_mask)
 185{
 186	struct biovec_slab *bvs = biovec_slab(*nr_vecs);
 187
 188	if (WARN_ON_ONCE(!bvs))
 189		return NULL;
 190
 191	/*
 192	 * Upgrade the nr_vecs request to take full advantage of the allocation.
 193	 * We also rely on this in the bvec_free path.
 194	 */
 195	*nr_vecs = bvs->nr_vecs;
 196
 197	/*
 198	 * Try a slab allocation first for all smaller allocations.  If that
 199	 * fails and __GFP_DIRECT_RECLAIM is set retry with the mempool.
 200	 * The mempool is sized to handle up to BIO_MAX_VECS entries.
 201	 */
 202	if (*nr_vecs < BIO_MAX_VECS) {
 203		struct bio_vec *bvl;
 204
 205		bvl = kmem_cache_alloc(bvs->slab, bvec_alloc_gfp(gfp_mask));
 206		if (likely(bvl) || !(gfp_mask & __GFP_DIRECT_RECLAIM))
 207			return bvl;
 208		*nr_vecs = BIO_MAX_VECS;
 209	}
 210
 211	return mempool_alloc(pool, gfp_mask);
 212}
 213
 214void bio_uninit(struct bio *bio)
 215{
 216#ifdef CONFIG_BLK_CGROUP
 217	if (bio->bi_blkg) {
 218		blkg_put(bio->bi_blkg);
 219		bio->bi_blkg = NULL;
 220	}
 221#endif
 222	if (bio_integrity(bio))
 223		bio_integrity_free(bio);
 224
 225	bio_crypt_free_ctx(bio);
 226}
 227EXPORT_SYMBOL(bio_uninit);
 228
 229static void bio_free(struct bio *bio)
 230{
 231	struct bio_set *bs = bio->bi_pool;
 232	void *p = bio;
 233
 234	WARN_ON_ONCE(!bs);
 235
 236	bio_uninit(bio);
 237	bvec_free(&bs->bvec_pool, bio->bi_io_vec, bio->bi_max_vecs);
 238	mempool_free(p - bs->front_pad, &bs->bio_pool);
 239}
 240
 241/*
 242 * Users of this function have their own bio allocation. Subsequently,
 243 * they must remember to pair any call to bio_init() with bio_uninit()
 244 * when IO has completed, or when the bio is released.
 245 */
 246void bio_init(struct bio *bio, struct block_device *bdev, struct bio_vec *table,
 247	      unsigned short max_vecs, blk_opf_t opf)
 248{
 249	bio->bi_next = NULL;
 250	bio->bi_bdev = bdev;
 251	bio->bi_opf = opf;
 252	bio->bi_flags = 0;
 253	bio->bi_ioprio = 0;
 254	bio->bi_status = 0;
 255	bio->bi_iter.bi_sector = 0;
 256	bio->bi_iter.bi_size = 0;
 257	bio->bi_iter.bi_idx = 0;
 258	bio->bi_iter.bi_bvec_done = 0;
 259	bio->bi_end_io = NULL;
 260	bio->bi_private = NULL;
 261#ifdef CONFIG_BLK_CGROUP
 262	bio->bi_blkg = NULL;
 263	bio->bi_issue.value = 0;
 264	if (bdev)
 265		bio_associate_blkg(bio);
 266#ifdef CONFIG_BLK_CGROUP_IOCOST
 267	bio->bi_iocost_cost = 0;
 268#endif
 269#endif
 270#ifdef CONFIG_BLK_INLINE_ENCRYPTION
 271	bio->bi_crypt_context = NULL;
 272#endif
 273#ifdef CONFIG_BLK_DEV_INTEGRITY
 274	bio->bi_integrity = NULL;
 275#endif
 276	bio->bi_vcnt = 0;
 277
 278	atomic_set(&bio->__bi_remaining, 1);
 279	atomic_set(&bio->__bi_cnt, 1);
 280	bio->bi_cookie = BLK_QC_T_NONE;
 281
 282	bio->bi_max_vecs = max_vecs;
 283	bio->bi_io_vec = table;
 284	bio->bi_pool = NULL;
 285}
 286EXPORT_SYMBOL(bio_init);
 287
 288/**
 289 * bio_reset - reinitialize a bio
 290 * @bio:	bio to reset
 291 * @bdev:	block device to use the bio for
 292 * @opf:	operation and flags for bio
 293 *
 294 * Description:
 295 *   After calling bio_reset(), @bio will be in the same state as a freshly
 296 *   allocated bio returned bio bio_alloc_bioset() - the only fields that are
 297 *   preserved are the ones that are initialized by bio_alloc_bioset(). See
 298 *   comment in struct bio.
 299 */
 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 completions
 329 * @bio: the target bio
 330 * @parent: the parent bio of @bio
 331 *
 332 * The caller won't have a bi_end_io called when @bio completes - instead,
 333 * @parent's bi_end_io won't be called until both @parent and @bio have
 334 * completed; the chained bio will also be freed when it completes.
 335 *
 336 * The caller must not set bi_private or bi_end_io in @bio.
 337 */
 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
 348struct bio *blk_next_bio(struct bio *bio, struct block_device *bdev,
 349		unsigned int nr_pages, blk_opf_t opf, gfp_t gfp)
 350{
 351	struct bio *new = bio_alloc(bdev, nr_pages, opf, gfp);
 352
 353	if (bio) {
 354		bio_chain(bio, new);
 355		submit_bio(bio);
 356	}
 357
 358	return new;
 359}
 360EXPORT_SYMBOL_GPL(blk_next_bio);
 361
 362static void bio_alloc_rescue(struct work_struct *work)
 363{
 364	struct bio_set *bs = container_of(work, struct bio_set, rescue_work);
 365	struct bio *bio;
 366
 367	while (1) {
 368		spin_lock(&bs->rescue_lock);
 369		bio = bio_list_pop(&bs->rescue_list);
 370		spin_unlock(&bs->rescue_lock);
 371
 372		if (!bio)
 373			break;
 374
 375		submit_bio_noacct(bio);
 376	}
 377}
 378
 379static void punt_bios_to_rescuer(struct bio_set *bs)
 380{
 381	struct bio_list punt, nopunt;
 382	struct bio *bio;
 383
 384	if (WARN_ON_ONCE(!bs->rescue_workqueue))
 385		return;
 386	/*
 387	 * In order to guarantee forward progress we must punt only bios that
 388	 * were allocated from this bio_set; otherwise, if there was a bio on
 389	 * there for a stacking driver higher up in the stack, processing it
 390	 * could require allocating bios from this bio_set, and doing that from
 391	 * our own rescuer would be bad.
 392	 *
 393	 * Since bio lists are singly linked, pop them all instead of trying to
 394	 * remove from the middle of the list:
 395	 */
 396
 397	bio_list_init(&punt);
 398	bio_list_init(&nopunt);
 399
 400	while ((bio = bio_list_pop(&current->bio_list[0])))
 401		bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
 402	current->bio_list[0] = nopunt;
 403
 404	bio_list_init(&nopunt);
 405	while ((bio = bio_list_pop(&current->bio_list[1])))
 406		bio_list_add(bio->bi_pool == bs ? &punt : &nopunt, bio);
 407	current->bio_list[1] = nopunt;
 408
 409	spin_lock(&bs->rescue_lock);
 410	bio_list_merge(&bs->rescue_list, &punt);
 411	spin_unlock(&bs->rescue_lock);
 412
 413	queue_work(bs->rescue_workqueue, &bs->rescue_work);
 414}
 415
 416static void bio_alloc_irq_cache_splice(struct bio_alloc_cache *cache)
 417{
 418	unsigned long flags;
 419
 420	/* cache->free_list must be empty */
 421	if (WARN_ON_ONCE(cache->free_list))
 422		return;
 423
 424	local_irq_save(flags);
 425	cache->free_list = cache->free_list_irq;
 426	cache->free_list_irq = NULL;
 427	cache->nr += cache->nr_irq;
 428	cache->nr_irq = 0;
 429	local_irq_restore(flags);
 430}
 431
 432static struct bio *bio_alloc_percpu_cache(struct block_device *bdev,
 433		unsigned short nr_vecs, blk_opf_t opf, gfp_t gfp,
 434		struct bio_set *bs)
 435{
 436	struct bio_alloc_cache *cache;
 437	struct bio *bio;
 438
 439	cache = per_cpu_ptr(bs->cache, get_cpu());
 440	if (!cache->free_list) {
 441		if (READ_ONCE(cache->nr_irq) >= ALLOC_CACHE_THRESHOLD)
 442			bio_alloc_irq_cache_splice(cache);
 443		if (!cache->free_list) {
 444			put_cpu();
 445			return NULL;
 446		}
 447	}
 448	bio = cache->free_list;
 449	cache->free_list = bio->bi_next;
 450	cache->nr--;
 451	put_cpu();
 452
 453	bio_init(bio, bdev, nr_vecs ? bio->bi_inline_vecs : NULL, nr_vecs, opf);
 454	bio->bi_pool = bs;
 455	return bio;
 456}
 457
 458/**
 459 * bio_alloc_bioset - allocate a bio for I/O
 460 * @bdev:	block device to allocate the bio for (can be %NULL)
 461 * @nr_vecs:	number of bvecs to pre-allocate
 462 * @opf:	operation and flags for bio
 463 * @gfp_mask:   the GFP_* mask given to the slab allocator
 464 * @bs:		the bio_set to allocate from.
 465 *
 466 * Allocate a bio from the mempools in @bs.
 467 *
 468 * If %__GFP_DIRECT_RECLAIM is set then bio_alloc will always be able to
 469 * allocate a bio.  This is due to the mempool guarantees.  To make this work,
 470 * callers must never allocate more than 1 bio at a time from the general pool.
 471 * Callers that need to allocate more than 1 bio must always submit the
 472 * previously allocated bio for IO before attempting to allocate a new one.
 473 * Failure to do so can cause deadlocks under memory pressure.
 474 *
 475 * Note that when running under submit_bio_noacct() (i.e. any block driver),
 476 * bios are not submitted until after you return - see the code in
 477 * submit_bio_noacct() that converts recursion into iteration, to prevent
 478 * stack overflows.
 479 *
 480 * This would normally mean allocating multiple bios under submit_bio_noacct()
 481 * would be susceptible to deadlocks, but we have
 482 * deadlock avoidance code that resubmits any blocked bios from a rescuer
 483 * thread.
 484 *
 485 * However, we do not guarantee forward progress for allocations from other
 486 * mempools. Doing multiple allocations from the same mempool under
 487 * submit_bio_noacct() should be avoided - instead, use bio_set's front_pad
 488 * for per bio allocations.
 489 *
 490 * Returns: Pointer to new bio on success, NULL on failure.
 491 */
 492struct bio *bio_alloc_bioset(struct block_device *bdev, unsigned short nr_vecs,
 493			     blk_opf_t opf, gfp_t gfp_mask,
 494			     struct bio_set *bs)
 495{
 496	gfp_t saved_gfp = gfp_mask;
 497	struct bio *bio;
 498	void *p;
 499
 500	/* should not use nobvec bioset for nr_vecs > 0 */
 501	if (WARN_ON_ONCE(!mempool_initialized(&bs->bvec_pool) && nr_vecs > 0))
 502		return NULL;
 503
 504	if (opf & REQ_ALLOC_CACHE) {
 505		if (bs->cache && nr_vecs <= BIO_INLINE_VECS) {
 506			bio = bio_alloc_percpu_cache(bdev, nr_vecs, opf,
 507						     gfp_mask, bs);
 508			if (bio)
 509				return bio;
 510			/*
 511			 * No cached bio available, bio returned below marked with
 512			 * REQ_ALLOC_CACHE to particpate in per-cpu alloc cache.
 513			 */
 514		} else {
 515			opf &= ~REQ_ALLOC_CACHE;
 516		}
 517	}
 518
 519	/*
 520	 * submit_bio_noacct() converts recursion to iteration; this means if
 521	 * we're running beneath it, any bios we allocate and submit will not be
 522	 * submitted (and thus freed) until after we return.
 523	 *
 524	 * This exposes us to a potential deadlock if we allocate multiple bios
 525	 * from the same bio_set() while running underneath submit_bio_noacct().
 526	 * If we were to allocate multiple bios (say a stacking block driver
 527	 * that was splitting bios), we would deadlock if we exhausted the
 528	 * mempool's reserve.
 529	 *
 530	 * We solve this, and guarantee forward progress, with a rescuer
 531	 * workqueue per bio_set. If we go to allocate and there are bios on
 532	 * current->bio_list, we first try the allocation without
 533	 * __GFP_DIRECT_RECLAIM; if that fails, we punt those bios we would be
 534	 * blocking to the rescuer workqueue before we retry with the original
 535	 * gfp_flags.
 536	 */
 537	if (current->bio_list &&
 538	    (!bio_list_empty(&current->bio_list[0]) ||
 539	     !bio_list_empty(&current->bio_list[1])) &&
 540	    bs->rescue_workqueue)
 541		gfp_mask &= ~__GFP_DIRECT_RECLAIM;
 542
 543	p = mempool_alloc(&bs->bio_pool, gfp_mask);
 544	if (!p && gfp_mask != saved_gfp) {
 545		punt_bios_to_rescuer(bs);
 546		gfp_mask = saved_gfp;
 547		p = mempool_alloc(&bs->bio_pool, gfp_mask);
 548	}
 549	if (unlikely(!p))
 550		return NULL;
 551	if (!mempool_is_saturated(&bs->bio_pool))
 552		opf &= ~REQ_ALLOC_CACHE;
 553
 554	bio = p + bs->front_pad;
 555	if (nr_vecs > BIO_INLINE_VECS) {
 556		struct bio_vec *bvl = NULL;
 557
 558		bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);
 559		if (!bvl && gfp_mask != saved_gfp) {
 560			punt_bios_to_rescuer(bs);
 561			gfp_mask = saved_gfp;
 562			bvl = bvec_alloc(&bs->bvec_pool, &nr_vecs, gfp_mask);
 563		}
 564		if (unlikely(!bvl))
 565			goto err_free;
 566
 567		bio_init(bio, bdev, bvl, nr_vecs, opf);
 568	} else if (nr_vecs) {
 569		bio_init(bio, bdev, bio->bi_inline_vecs, BIO_INLINE_VECS, opf);
 570	} else {
 571		bio_init(bio, bdev, NULL, 0, opf);
 572	}
 573
 574	bio->bi_pool = bs;
 575	return bio;
 576
 577err_free:
 578	mempool_free(p, &bs->bio_pool);
 579	return NULL;
 580}
 581EXPORT_SYMBOL(bio_alloc_bioset);
 582
 583/**
 584 * bio_kmalloc - kmalloc a bio
 585 * @nr_vecs:	number of bio_vecs to allocate
 586 * @gfp_mask:   the GFP_* mask given to the slab allocator
 587 *
 588 * Use kmalloc to allocate a bio (including bvecs).  The bio must be initialized
 589 * using bio_init() before use.  To free a bio returned from this function use
 590 * kfree() after calling bio_uninit().  A bio returned from this function can
 591 * be reused by calling bio_uninit() before calling bio_init() again.
 592 *
 593 * Note that unlike bio_alloc() or bio_alloc_bioset() allocations from this
 594 * function are not backed by a mempool can fail.  Do not use this function
 595 * for allocations in the file system I/O path.
 596 *
 597 * Returns: Pointer to new bio on success, NULL on failure.
 598 */
 599struct bio *bio_kmalloc(unsigned short nr_vecs, gfp_t gfp_mask)
 600{
 601	struct bio *bio;
 602
 603	if (nr_vecs > UIO_MAXIOV)
 604		return NULL;
 605	return kmalloc(struct_size(bio, bi_inline_vecs, nr_vecs), gfp_mask);
 606}
 607EXPORT_SYMBOL(bio_kmalloc);
 608
 609void zero_fill_bio_iter(struct bio *bio, struct bvec_iter start)
 610{
 611	struct bio_vec bv;
 612	struct bvec_iter iter;
 613
 614	__bio_for_each_segment(bv, bio, iter, start)
 615		memzero_bvec(&bv);
 616}
 617EXPORT_SYMBOL(zero_fill_bio_iter);
 618
 619/**
 620 * bio_truncate - truncate the bio to small size of @new_size
 621 * @bio:	the bio to be truncated
 622 * @new_size:	new size for truncating the bio
 623 *
 624 * Description:
 625 *   Truncate the bio to new size of @new_size. If bio_op(bio) is
 626 *   REQ_OP_READ, zero the truncated part. This function should only
 627 *   be used for handling corner cases, such as bio eod.
 628 */
 629static void bio_truncate(struct bio *bio, unsigned new_size)
 630{
 631	struct bio_vec bv;
 632	struct bvec_iter iter;
 633	unsigned int done = 0;
 634	bool truncated = false;
 635
 636	if (new_size >= bio->bi_iter.bi_size)
 637		return;
 638
 639	if (bio_op(bio) != REQ_OP_READ)
 640		goto exit;
 641
 642	bio_for_each_segment(bv, bio, iter) {
 643		if (done + bv.bv_len > new_size) {
 644			unsigned offset;
 645
 646			if (!truncated)
 647				offset = new_size - done;
 648			else
 649				offset = 0;
 650			zero_user(bv.bv_page, bv.bv_offset + offset,
 651				  bv.bv_len - offset);
 652			truncated = true;
 653		}
 654		done += bv.bv_len;
 655	}
 656
 657 exit:
 658	/*
 659	 * Don't touch bvec table here and make it really immutable, since
 660	 * fs bio user has to retrieve all pages via bio_for_each_segment_all
 661	 * in its .end_bio() callback.
 662	 *
 663	 * It is enough to truncate bio by updating .bi_size since we can make
 664	 * correct bvec with the updated .bi_size for drivers.
 665	 */
 666	bio->bi_iter.bi_size = new_size;
 667}
 668
 669/**
 670 * guard_bio_eod - truncate a BIO to fit the block device
 671 * @bio:	bio to truncate
 672 *
 673 * This allows us to do IO even on the odd last sectors of a device, even if the
 674 * block size is some multiple of the physical sector size.
 675 *
 676 * We'll just truncate the bio to the size of the device, and clear the end of
 677 * the buffer head manually.  Truly out-of-range accesses will turn into actual
 678 * I/O errors, this only handles the "we need to be able to do I/O at the final
 679 * sector" case.
 680 */
 681void guard_bio_eod(struct bio *bio)
 682{
 683	sector_t maxsector = bdev_nr_sectors(bio->bi_bdev);
 684
 685	if (!maxsector)
 686		return;
 687
 688	/*
 689	 * If the *whole* IO is past the end of the device,
 690	 * let it through, and the IO layer will turn it into
 691	 * an EIO.
 692	 */
 693	if (unlikely(bio->bi_iter.bi_sector >= maxsector))
 694		return;
 695
 696	maxsector -= bio->bi_iter.bi_sector;
 697	if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
 698		return;
 699
 700	bio_truncate(bio, maxsector << 9);
 701}
 702
 703static int __bio_alloc_cache_prune(struct bio_alloc_cache *cache,
 704				   unsigned int nr)
 705{
 706	unsigned int i = 0;
 707	struct bio *bio;
 708
 709	while ((bio = cache->free_list) != NULL) {
 710		cache->free_list = bio->bi_next;
 711		cache->nr--;
 712		bio_free(bio);
 713		if (++i == nr)
 714			break;
 715	}
 716	return i;
 717}
 718
 719static void bio_alloc_cache_prune(struct bio_alloc_cache *cache,
 720				  unsigned int nr)
 721{
 722	nr -= __bio_alloc_cache_prune(cache, nr);
 723	if (!READ_ONCE(cache->free_list)) {
 724		bio_alloc_irq_cache_splice(cache);
 725		__bio_alloc_cache_prune(cache, nr);
 726	}
 727}
 728
 729static int bio_cpu_dead(unsigned int cpu, struct hlist_node *node)
 730{
 731	struct bio_set *bs;
 732
 733	bs = hlist_entry_safe(node, struct bio_set, cpuhp_dead);
 734	if (bs->cache) {
 735		struct bio_alloc_cache *cache = per_cpu_ptr(bs->cache, cpu);
 736
 737		bio_alloc_cache_prune(cache, -1U);
 738	}
 739	return 0;
 740}
 741
 742static void bio_alloc_cache_destroy(struct bio_set *bs)
 743{
 744	int cpu;
 745
 746	if (!bs->cache)
 747		return;
 748
 749	cpuhp_state_remove_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
 750	for_each_possible_cpu(cpu) {
 751		struct bio_alloc_cache *cache;
 752
 753		cache = per_cpu_ptr(bs->cache, cpu);
 754		bio_alloc_cache_prune(cache, -1U);
 755	}
 756	free_percpu(bs->cache);
 757	bs->cache = NULL;
 758}
 759
 760static inline void bio_put_percpu_cache(struct bio *bio)
 761{
 762	struct bio_alloc_cache *cache;
 763
 764	cache = per_cpu_ptr(bio->bi_pool->cache, get_cpu());
 765	if (READ_ONCE(cache->nr_irq) + cache->nr > ALLOC_CACHE_MAX) {
 766		put_cpu();
 767		bio_free(bio);
 768		return;
 769	}
 770
 771	bio_uninit(bio);
 772
 773	if ((bio->bi_opf & REQ_POLLED) && !WARN_ON_ONCE(in_interrupt())) {
 774		bio->bi_next = cache->free_list;
 775		bio->bi_bdev = NULL;
 776		cache->free_list = bio;
 777		cache->nr++;
 778	} else {
 779		unsigned long flags;
 780
 781		local_irq_save(flags);
 782		bio->bi_next = cache->free_list_irq;
 783		cache->free_list_irq = bio;
 784		cache->nr_irq++;
 785		local_irq_restore(flags);
 786	}
 787	put_cpu();
 788}
 789
 790/**
 791 * bio_put - release a reference to a bio
 792 * @bio:   bio to release reference to
 793 *
 794 * Description:
 795 *   Put a reference to a &struct bio, either one you have gotten with
 796 *   bio_alloc, bio_get or bio_clone_*. The last put of a bio will free it.
 797 **/
 798void bio_put(struct bio *bio)
 799{
 800	if (unlikely(bio_flagged(bio, BIO_REFFED))) {
 801		BUG_ON(!atomic_read(&bio->__bi_cnt));
 802		if (!atomic_dec_and_test(&bio->__bi_cnt))
 803			return;
 804	}
 805	if (bio->bi_opf & REQ_ALLOC_CACHE)
 806		bio_put_percpu_cache(bio);
 807	else
 808		bio_free(bio);
 809}
 810EXPORT_SYMBOL(bio_put);
 811
 812static int __bio_clone(struct bio *bio, struct bio *bio_src, gfp_t gfp)
 813{
 814	bio_set_flag(bio, BIO_CLONED);
 815	bio->bi_ioprio = bio_src->bi_ioprio;
 816	bio->bi_iter = bio_src->bi_iter;
 817
 818	if (bio->bi_bdev) {
 819		if (bio->bi_bdev == bio_src->bi_bdev &&
 820		    bio_flagged(bio_src, BIO_REMAPPED))
 821			bio_set_flag(bio, BIO_REMAPPED);
 822		bio_clone_blkg_association(bio, bio_src);
 823	}
 824
 825	if (bio_crypt_clone(bio, bio_src, gfp) < 0)
 826		return -ENOMEM;
 827	if (bio_integrity(bio_src) &&
 828	    bio_integrity_clone(bio, bio_src, gfp) < 0)
 829		return -ENOMEM;
 830	return 0;
 831}
 832
 833/**
 834 * bio_alloc_clone - clone a bio that shares the original bio's biovec
 835 * @bdev: block_device to clone onto
 836 * @bio_src: bio to clone from
 837 * @gfp: allocation priority
 838 * @bs: bio_set to allocate from
 839 *
 840 * Allocate a new bio that is a clone of @bio_src. The caller owns the returned
 841 * bio, but not the actual data it points to.
 842 *
 843 * The caller must ensure that the return bio is not freed before @bio_src.
 844 */
 845struct bio *bio_alloc_clone(struct block_device *bdev, struct bio *bio_src,
 846		gfp_t gfp, struct bio_set *bs)
 847{
 848	struct bio *bio;
 849
 850	bio = bio_alloc_bioset(bdev, 0, bio_src->bi_opf, gfp, bs);
 851	if (!bio)
 852		return NULL;
 853
 854	if (__bio_clone(bio, bio_src, gfp) < 0) {
 855		bio_put(bio);
 856		return NULL;
 857	}
 858	bio->bi_io_vec = bio_src->bi_io_vec;
 859
 860	return bio;
 861}
 862EXPORT_SYMBOL(bio_alloc_clone);
 863
 864/**
 865 * bio_init_clone - clone a bio that shares the original bio's biovec
 866 * @bdev: block_device to clone onto
 867 * @bio: bio to clone into
 868 * @bio_src: bio to clone from
 869 * @gfp: allocation priority
 870 *
 871 * Initialize a new bio in caller provided memory that is a clone of @bio_src.
 872 * The caller owns the returned bio, but not the actual data it points to.
 873 *
 874 * The caller must ensure that @bio_src is not freed before @bio.
 875 */
 876int bio_init_clone(struct block_device *bdev, struct bio *bio,
 877		struct bio *bio_src, gfp_t gfp)
 878{
 879	int ret;
 880
 881	bio_init(bio, bdev, bio_src->bi_io_vec, 0, bio_src->bi_opf);
 882	ret = __bio_clone(bio, bio_src, gfp);
 883	if (ret)
 884		bio_uninit(bio);
 885	return ret;
 886}
 887EXPORT_SYMBOL(bio_init_clone);
 888
 889/**
 890 * bio_full - check if the bio is full
 891 * @bio:	bio to check
 892 * @len:	length of one segment to be added
 893 *
 894 * Return true if @bio is full and one segment with @len bytes can't be
 895 * added to the bio, otherwise return false
 896 */
 897static inline bool bio_full(struct bio *bio, unsigned len)
 898{
 899	if (bio->bi_vcnt >= bio->bi_max_vecs)
 900		return true;
 901	if (bio->bi_iter.bi_size > UINT_MAX - len)
 902		return true;
 903	return false;
 904}
 905
 906static bool bvec_try_merge_page(struct bio_vec *bv, struct page *page,
 907		unsigned int len, unsigned int off, bool *same_page)
 
 908{
 909	size_t bv_end = bv->bv_offset + bv->bv_len;
 910	phys_addr_t vec_end_addr = page_to_phys(bv->bv_page) + bv_end - 1;
 911	phys_addr_t page_addr = page_to_phys(page);
 912
 913	if (vec_end_addr + 1 != page_addr + off)
 914		return false;
 915	if (xen_domain() && !xen_biovec_phys_mergeable(bv, page))
 916		return false;
 917	if (!zone_device_pages_have_same_pgmap(bv->bv_page, page))
 918		return false;
 919
 920	*same_page = ((vec_end_addr & PAGE_MASK) == page_addr);
 921	if (!*same_page) {
 922		if (IS_ENABLED(CONFIG_KMSAN))
 923			return false;
 924		if (bv->bv_page + bv_end / PAGE_SIZE != page + off / PAGE_SIZE)
 925			return false;
 926	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 927
 928	bv->bv_len += len;
 929	return true;
 
 
 
 
 
 
 
 
 
 930}
 931
 932/*
 933 * Try to merge a page into a segment, while obeying the hardware segment
 934 * size limit.  This is not for normal read/write bios, but for passthrough
 935 * or Zone Append operations that we can't split.
 936 */
 937bool bvec_try_merge_hw_page(struct request_queue *q, struct bio_vec *bv,
 938		struct page *page, unsigned len, unsigned offset,
 939		bool *same_page)
 940{
 
 941	unsigned long mask = queue_segment_boundary(q);
 942	phys_addr_t addr1 = page_to_phys(bv->bv_page) + bv->bv_offset;
 943	phys_addr_t addr2 = page_to_phys(page) + offset + len - 1;
 944
 945	if ((addr1 | mask) != (addr2 | mask))
 946		return false;
 947	if (len > queue_max_segment_size(q) - bv->bv_len)
 948		return false;
 949	return bvec_try_merge_page(bv, page, len, offset, same_page);
 950}
 951
 952/**
 953 * bio_add_hw_page - attempt to add a page to a bio with hw constraints
 954 * @q: the target queue
 955 * @bio: destination bio
 956 * @page: page to add
 957 * @len: vec entry length
 958 * @offset: vec entry offset
 959 * @max_sectors: maximum number of sectors that can be added
 960 * @same_page: return if the segment has been merged inside the same page
 961 *
 962 * Add a page to a bio while respecting the hardware max_sectors, max_segment
 963 * and gap limitations.
 964 */
 965int bio_add_hw_page(struct request_queue *q, struct bio *bio,
 966		struct page *page, unsigned int len, unsigned int offset,
 967		unsigned int max_sectors, bool *same_page)
 968{
 969	unsigned int max_size = max_sectors << SECTOR_SHIFT;
 970
 971	if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
 972		return 0;
 973
 974	len = min3(len, max_size, queue_max_segment_size(q));
 975	if (len > max_size - bio->bi_iter.bi_size)
 976		return 0;
 977
 978	if (bio->bi_vcnt > 0) {
 979		struct bio_vec *bv = &bio->bi_io_vec[bio->bi_vcnt - 1];
 980
 981		if (bvec_try_merge_hw_page(q, bv, page, len, offset,
 982				same_page)) {
 983			bio->bi_iter.bi_size += len;
 984			return len;
 985		}
 986
 987		if (bio->bi_vcnt >=
 988		    min(bio->bi_max_vecs, queue_max_segments(q)))
 989			return 0;
 990
 991		/*
 992		 * If the queue doesn't support SG gaps and adding this segment
 993		 * would create a gap, disallow it.
 994		 */
 995		if (bvec_gap_to_prev(&q->limits, bv, offset))
 
 996			return 0;
 997	}
 998
 999	bvec_set_page(&bio->bi_io_vec[bio->bi_vcnt], page, len, offset);
 
 
 
 
 
 
 
 
 
1000	bio->bi_vcnt++;
1001	bio->bi_iter.bi_size += len;
1002	return len;
1003}
1004
1005/**
1006 * bio_add_pc_page	- attempt to add page to passthrough bio
1007 * @q: the target queue
1008 * @bio: destination bio
1009 * @page: page to add
1010 * @len: vec entry length
1011 * @offset: vec entry offset
1012 *
1013 * Attempt to add a page to the bio_vec maplist. This can fail for a
1014 * number of reasons, such as the bio being full or target block device
1015 * limitations. The target block device must allow bio's up to PAGE_SIZE,
1016 * so it is always possible to add a single page to an empty bio.
1017 *
1018 * This should only be used by passthrough bios.
1019 */
1020int bio_add_pc_page(struct request_queue *q, struct bio *bio,
1021		struct page *page, unsigned int len, unsigned int offset)
1022{
1023	bool same_page = false;
1024	return bio_add_hw_page(q, bio, page, len, offset,
1025			queue_max_hw_sectors(q), &same_page);
1026}
1027EXPORT_SYMBOL(bio_add_pc_page);
1028
1029/**
1030 * bio_add_zone_append_page - attempt to add page to zone-append bio
1031 * @bio: destination bio
1032 * @page: page to add
1033 * @len: vec entry length
1034 * @offset: vec entry offset
1035 *
1036 * Attempt to add a page to the bio_vec maplist of a bio that will be submitted
1037 * for a zone-append request. This can fail for a number of reasons, such as the
1038 * bio being full or the target block device is not a zoned block device or
1039 * other limitations of the target block device. The target block device must
1040 * allow bio's up to PAGE_SIZE, so it is always possible to add a single page
1041 * to an empty bio.
1042 *
1043 * Returns: number of bytes added to the bio, or 0 in case of a failure.
1044 */
1045int bio_add_zone_append_page(struct bio *bio, struct page *page,
1046			     unsigned int len, unsigned int offset)
1047{
1048	struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1049	bool same_page = false;
1050
1051	if (WARN_ON_ONCE(bio_op(bio) != REQ_OP_ZONE_APPEND))
1052		return 0;
1053
1054	if (WARN_ON_ONCE(!bdev_is_zoned(bio->bi_bdev)))
1055		return 0;
1056
1057	return bio_add_hw_page(q, bio, page, len, offset,
1058			       queue_max_zone_append_sectors(q), &same_page);
1059}
1060EXPORT_SYMBOL_GPL(bio_add_zone_append_page);
1061
1062/**
1063 * __bio_add_page - add page(s) to a bio in a new segment
1064 * @bio: destination bio
1065 * @page: start page to add
1066 * @len: length of the data to add, may cross pages
1067 * @off: offset of the data relative to @page, may cross pages
1068 *
1069 * Add the data at @page + @off to @bio as a new bvec.  The caller must ensure
1070 * that @bio has space for another bvec.
1071 */
1072void __bio_add_page(struct bio *bio, struct page *page,
1073		unsigned int len, unsigned int off)
1074{
 
 
1075	WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED));
1076	WARN_ON_ONCE(bio_full(bio, len));
1077
1078	bvec_set_page(&bio->bi_io_vec[bio->bi_vcnt], page, len, off);
 
 
 
1079	bio->bi_iter.bi_size += len;
1080	bio->bi_vcnt++;
1081}
1082EXPORT_SYMBOL_GPL(__bio_add_page);
1083
1084/**
1085 *	bio_add_page	-	attempt to add page(s) to bio
1086 *	@bio: destination bio
1087 *	@page: start page to add
1088 *	@len: vec entry length, may cross pages
1089 *	@offset: vec entry offset relative to @page, may cross pages
1090 *
1091 *	Attempt to add page(s) to the bio_vec maplist. This will only fail
1092 *	if either bio->bi_vcnt == bio->bi_max_vecs or it's a cloned bio.
1093 */
1094int bio_add_page(struct bio *bio, struct page *page,
1095		 unsigned int len, unsigned int offset)
1096{
1097	bool same_page = false;
1098
1099	if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
1100		return 0;
1101	if (bio->bi_iter.bi_size > UINT_MAX - len)
1102		return 0;
1103
1104	if (bio->bi_vcnt > 0 &&
1105	    bvec_try_merge_page(&bio->bi_io_vec[bio->bi_vcnt - 1],
1106				page, len, offset, &same_page)) {
1107		bio->bi_iter.bi_size += len;
1108		return len;
1109	}
1110
1111	if (bio->bi_vcnt >= bio->bi_max_vecs)
1112		return 0;
1113	__bio_add_page(bio, page, len, offset);
1114	return len;
1115}
1116EXPORT_SYMBOL(bio_add_page);
1117
1118void bio_add_folio_nofail(struct bio *bio, struct folio *folio, size_t len,
1119			  size_t off)
1120{
1121	WARN_ON_ONCE(len > UINT_MAX);
1122	WARN_ON_ONCE(off > UINT_MAX);
1123	__bio_add_page(bio, &folio->page, len, off);
1124}
1125
1126/**
1127 * bio_add_folio - Attempt to add part of a folio to a bio.
1128 * @bio: BIO to add to.
1129 * @folio: Folio to add.
1130 * @len: How many bytes from the folio to add.
1131 * @off: First byte in this folio to add.
1132 *
1133 * Filesystems that use folios can call this function instead of calling
1134 * bio_add_page() for each page in the folio.  If @off is bigger than
1135 * PAGE_SIZE, this function can create a bio_vec that starts in a page
1136 * after the bv_page.  BIOs do not support folios that are 4GiB or larger.
1137 *
1138 * Return: Whether the addition was successful.
1139 */
1140bool bio_add_folio(struct bio *bio, struct folio *folio, size_t len,
1141		   size_t off)
1142{
1143	if (len > UINT_MAX || off > UINT_MAX)
1144		return false;
1145	return bio_add_page(bio, &folio->page, len, off) > 0;
1146}
1147EXPORT_SYMBOL(bio_add_folio);
1148
1149void __bio_release_pages(struct bio *bio, bool mark_dirty)
1150{
1151	struct folio_iter fi;
 
1152
1153	bio_for_each_folio_all(fi, bio) {
1154		struct page *page;
1155		size_t done = 0;
1156
1157		if (mark_dirty) {
1158			folio_lock(fi.folio);
1159			folio_mark_dirty(fi.folio);
1160			folio_unlock(fi.folio);
1161		}
1162		page = folio_page(fi.folio, fi.offset / PAGE_SIZE);
1163		do {
1164			bio_release_page(bio, page++);
1165			done += PAGE_SIZE;
1166		} while (done < fi.length);
1167	}
1168}
1169EXPORT_SYMBOL_GPL(__bio_release_pages);
1170
1171void bio_iov_bvec_set(struct bio *bio, struct iov_iter *iter)
1172{
1173	size_t size = iov_iter_count(iter);
1174
1175	WARN_ON_ONCE(bio->bi_max_vecs);
1176
1177	if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1178		struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1179		size_t max_sectors = queue_max_zone_append_sectors(q);
1180
1181		size = min(size, max_sectors << SECTOR_SHIFT);
1182	}
1183
1184	bio->bi_vcnt = iter->nr_segs;
1185	bio->bi_io_vec = (struct bio_vec *)iter->bvec;
1186	bio->bi_iter.bi_bvec_done = iter->iov_offset;
1187	bio->bi_iter.bi_size = size;
 
1188	bio_set_flag(bio, BIO_CLONED);
1189}
1190
1191static int bio_iov_add_page(struct bio *bio, struct page *page,
1192		unsigned int len, unsigned int offset)
1193{
1194	bool same_page = false;
1195
1196	if (WARN_ON_ONCE(bio->bi_iter.bi_size > UINT_MAX - len))
1197		return -EIO;
1198
1199	if (bio->bi_vcnt > 0 &&
1200	    bvec_try_merge_page(&bio->bi_io_vec[bio->bi_vcnt - 1],
1201				page, len, offset, &same_page)) {
1202		bio->bi_iter.bi_size += len;
1203		if (same_page)
1204			bio_release_page(bio, page);
1205		return 0;
1206	}
1207	__bio_add_page(bio, page, len, offset);
 
 
1208	return 0;
1209}
1210
1211static int bio_iov_add_zone_append_page(struct bio *bio, struct page *page,
1212		unsigned int len, unsigned int offset)
1213{
1214	struct request_queue *q = bdev_get_queue(bio->bi_bdev);
1215	bool same_page = false;
1216
1217	if (bio_add_hw_page(q, bio, page, len, offset,
1218			queue_max_zone_append_sectors(q), &same_page) != len)
1219		return -EINVAL;
1220	if (same_page)
1221		bio_release_page(bio, page);
1222	return 0;
1223}
1224
1225#define PAGE_PTRS_PER_BVEC     (sizeof(struct bio_vec) / sizeof(struct page *))
1226
1227/**
1228 * __bio_iov_iter_get_pages - pin user or kernel pages and add them to a bio
1229 * @bio: bio to add pages to
1230 * @iter: iov iterator describing the region to be mapped
1231 *
1232 * Extracts pages from *iter and appends them to @bio's bvec array.  The pages
1233 * will have to be cleaned up in the way indicated by the BIO_PAGE_PINNED flag.
1234 * For a multi-segment *iter, this function only adds pages from the next
1235 * non-empty segment of the iov iterator.
1236 */
1237static int __bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1238{
1239	iov_iter_extraction_t extraction_flags = 0;
1240	unsigned short nr_pages = bio->bi_max_vecs - bio->bi_vcnt;
1241	unsigned short entries_left = bio->bi_max_vecs - bio->bi_vcnt;
1242	struct bio_vec *bv = bio->bi_io_vec + bio->bi_vcnt;
1243	struct page **pages = (struct page **)bv;
 
1244	ssize_t size, left;
1245	unsigned len, i = 0;
1246	size_t offset;
1247	int ret = 0;
1248
1249	/*
1250	 * Move page array up in the allocated memory for the bio vecs as far as
1251	 * possible so that we can start filling biovecs from the beginning
1252	 * without overwriting the temporary page array.
1253	 */
1254	BUILD_BUG_ON(PAGE_PTRS_PER_BVEC < 2);
1255	pages += entries_left * (PAGE_PTRS_PER_BVEC - 1);
1256
1257	if (bio->bi_bdev && blk_queue_pci_p2pdma(bio->bi_bdev->bd_disk->queue))
1258		extraction_flags |= ITER_ALLOW_P2PDMA;
1259
1260	/*
1261	 * Each segment in the iov is required to be a block size multiple.
1262	 * However, we may not be able to get the entire segment if it spans
1263	 * more pages than bi_max_vecs allows, so we have to ALIGN_DOWN the
1264	 * result to ensure the bio's total size is correct. The remainder of
1265	 * the iov data will be picked up in the next bio iteration.
1266	 */
1267	size = iov_iter_extract_pages(iter, &pages,
1268				      UINT_MAX - bio->bi_iter.bi_size,
1269				      nr_pages, extraction_flags, &offset);
1270	if (unlikely(size <= 0))
1271		return size ? size : -EFAULT;
1272
1273	nr_pages = DIV_ROUND_UP(offset + size, PAGE_SIZE);
1274
1275	if (bio->bi_bdev) {
1276		size_t trim = size & (bdev_logical_block_size(bio->bi_bdev) - 1);
1277		iov_iter_revert(iter, trim);
1278		size -= trim;
1279	}
1280
 
1281	if (unlikely(!size)) {
1282		ret = -EFAULT;
1283		goto out;
1284	}
1285
1286	for (left = size, i = 0; left > 0; left -= len, i++) {
1287		struct page *page = pages[i];
1288
1289		len = min_t(size_t, PAGE_SIZE - offset, left);
1290		if (bio_op(bio) == REQ_OP_ZONE_APPEND) {
1291			ret = bio_iov_add_zone_append_page(bio, page, len,
1292					offset);
1293			if (ret)
1294				break;
1295		} else
1296			bio_iov_add_page(bio, page, len, offset);
1297
1298		offset = 0;
1299	}
1300
1301	iov_iter_revert(iter, left);
1302out:
1303	while (i < nr_pages)
1304		bio_release_page(bio, pages[i++]);
1305
1306	return ret;
1307}
1308
1309/**
1310 * bio_iov_iter_get_pages - add user or kernel pages to a bio
1311 * @bio: bio to add pages to
1312 * @iter: iov iterator describing the region to be added
1313 *
1314 * This takes either an iterator pointing to user memory, or one pointing to
1315 * kernel pages (BVEC iterator). If we're adding user pages, we pin them and
1316 * map them into the kernel. On IO completion, the caller should put those
1317 * pages. For bvec based iterators bio_iov_iter_get_pages() uses the provided
1318 * bvecs rather than copying them. Hence anyone issuing kiocb based IO needs
1319 * to ensure the bvecs and pages stay referenced until the submitted I/O is
1320 * completed by a call to ->ki_complete() or returns with an error other than
1321 * -EIOCBQUEUED. The caller needs to check if the bio is flagged BIO_NO_PAGE_REF
1322 * on IO completion. If it isn't, then pages should be released.
1323 *
1324 * The function tries, but does not guarantee, to pin as many pages as
1325 * fit into the bio, or are requested in @iter, whatever is smaller. If
1326 * MM encounters an error pinning the requested pages, it stops. Error
1327 * is returned only if 0 pages could be pinned.
1328 */
1329int bio_iov_iter_get_pages(struct bio *bio, struct iov_iter *iter)
1330{
1331	int ret = 0;
1332
1333	if (WARN_ON_ONCE(bio_flagged(bio, BIO_CLONED)))
1334		return -EIO;
1335
1336	if (iov_iter_is_bvec(iter)) {
1337		bio_iov_bvec_set(bio, iter);
1338		iov_iter_advance(iter, bio->bi_iter.bi_size);
1339		return 0;
1340	}
1341
1342	if (iov_iter_extract_will_pin(iter))
1343		bio_set_flag(bio, BIO_PAGE_PINNED);
1344	do {
1345		ret = __bio_iov_iter_get_pages(bio, iter);
1346	} while (!ret && iov_iter_count(iter) && !bio_full(bio, 0));
1347
1348	return bio->bi_vcnt ? 0 : ret;
1349}
1350EXPORT_SYMBOL_GPL(bio_iov_iter_get_pages);
1351
1352static void submit_bio_wait_endio(struct bio *bio)
1353{
1354	complete(bio->bi_private);
1355}
1356
1357/**
1358 * submit_bio_wait - submit a bio, and wait until it completes
1359 * @bio: The &struct bio which describes the I/O
1360 *
1361 * Simple wrapper around submit_bio(). Returns 0 on success, or the error from
1362 * bio_endio() on failure.
1363 *
1364 * WARNING: Unlike to how submit_bio() is usually used, this function does not
1365 * result in bio reference to be consumed. The caller must drop the reference
1366 * on his own.
1367 */
1368int submit_bio_wait(struct bio *bio)
1369{
1370	DECLARE_COMPLETION_ONSTACK_MAP(done,
1371			bio->bi_bdev->bd_disk->lockdep_map);
1372	unsigned long hang_check;
1373
1374	bio->bi_private = &done;
1375	bio->bi_end_io = submit_bio_wait_endio;
1376	bio->bi_opf |= REQ_SYNC;
1377	submit_bio(bio);
1378
1379	/* Prevent hang_check timer from firing at us during very long I/O */
1380	hang_check = sysctl_hung_task_timeout_secs;
1381	if (hang_check)
1382		while (!wait_for_completion_io_timeout(&done,
1383					hang_check * (HZ/2)))
1384			;
1385	else
1386		wait_for_completion_io(&done);
1387
1388	return blk_status_to_errno(bio->bi_status);
1389}
1390EXPORT_SYMBOL(submit_bio_wait);
1391
1392void __bio_advance(struct bio *bio, unsigned bytes)
1393{
1394	if (bio_integrity(bio))
1395		bio_integrity_advance(bio, bytes);
1396
1397	bio_crypt_advance(bio, bytes);
1398	bio_advance_iter(bio, &bio->bi_iter, bytes);
1399}
1400EXPORT_SYMBOL(__bio_advance);
1401
1402void bio_copy_data_iter(struct bio *dst, struct bvec_iter *dst_iter,
1403			struct bio *src, struct bvec_iter *src_iter)
1404{
1405	while (src_iter->bi_size && dst_iter->bi_size) {
1406		struct bio_vec src_bv = bio_iter_iovec(src, *src_iter);
1407		struct bio_vec dst_bv = bio_iter_iovec(dst, *dst_iter);
1408		unsigned int bytes = min(src_bv.bv_len, dst_bv.bv_len);
1409		void *src_buf = bvec_kmap_local(&src_bv);
1410		void *dst_buf = bvec_kmap_local(&dst_bv);
1411
1412		memcpy(dst_buf, src_buf, bytes);
1413
1414		kunmap_local(dst_buf);
1415		kunmap_local(src_buf);
1416
1417		bio_advance_iter_single(src, src_iter, bytes);
1418		bio_advance_iter_single(dst, dst_iter, bytes);
1419	}
1420}
1421EXPORT_SYMBOL(bio_copy_data_iter);
1422
1423/**
1424 * bio_copy_data - copy contents of data buffers from one bio to another
1425 * @src: source bio
1426 * @dst: destination bio
1427 *
1428 * Stops when it reaches the end of either @src or @dst - that is, copies
1429 * min(src->bi_size, dst->bi_size) bytes (or the equivalent for lists of bios).
1430 */
1431void bio_copy_data(struct bio *dst, struct bio *src)
1432{
1433	struct bvec_iter src_iter = src->bi_iter;
1434	struct bvec_iter dst_iter = dst->bi_iter;
1435
1436	bio_copy_data_iter(dst, &dst_iter, src, &src_iter);
1437}
1438EXPORT_SYMBOL(bio_copy_data);
1439
1440void bio_free_pages(struct bio *bio)
1441{
1442	struct bio_vec *bvec;
1443	struct bvec_iter_all iter_all;
1444
1445	bio_for_each_segment_all(bvec, bio, iter_all)
1446		__free_page(bvec->bv_page);
1447}
1448EXPORT_SYMBOL(bio_free_pages);
1449
1450/*
1451 * bio_set_pages_dirty() and bio_check_pages_dirty() are support functions
1452 * for performing direct-IO in BIOs.
1453 *
1454 * The problem is that we cannot run folio_mark_dirty() from interrupt context
1455 * because the required locks are not interrupt-safe.  So what we can do is to
1456 * mark the pages dirty _before_ performing IO.  And in interrupt context,
1457 * check that the pages are still dirty.   If so, fine.  If not, redirty them
1458 * in process context.
1459 *
 
 
 
 
 
 
1460 * Note that this code is very hard to test under normal circumstances because
1461 * direct-io pins the pages with get_user_pages().  This makes
1462 * is_page_cache_freeable return false, and the VM will not clean the pages.
1463 * But other code (eg, flusher threads) could clean the pages if they are mapped
1464 * pagecache.
1465 *
1466 * Simply disabling the call to bio_set_pages_dirty() is a good way to test the
1467 * deferred bio dirtying paths.
1468 */
1469
1470/*
1471 * bio_set_pages_dirty() will mark all the bio's pages as dirty.
1472 */
1473void bio_set_pages_dirty(struct bio *bio)
1474{
1475	struct folio_iter fi;
 
1476
1477	bio_for_each_folio_all(fi, bio) {
1478		folio_lock(fi.folio);
1479		folio_mark_dirty(fi.folio);
1480		folio_unlock(fi.folio);
1481	}
1482}
1483EXPORT_SYMBOL_GPL(bio_set_pages_dirty);
1484
1485/*
1486 * bio_check_pages_dirty() will check that all the BIO's pages are still dirty.
1487 * If they are, then fine.  If, however, some pages are clean then they must
1488 * have been written out during the direct-IO read.  So we take another ref on
1489 * the BIO and re-dirty the pages in process context.
1490 *
1491 * It is expected that bio_check_pages_dirty() will wholly own the BIO from
1492 * here on.  It will unpin each page and will run one bio_put() against the
1493 * BIO.
1494 */
1495
1496static void bio_dirty_fn(struct work_struct *work);
1497
1498static DECLARE_WORK(bio_dirty_work, bio_dirty_fn);
1499static DEFINE_SPINLOCK(bio_dirty_lock);
1500static struct bio *bio_dirty_list;
1501
1502/*
1503 * This runs in process context
1504 */
1505static void bio_dirty_fn(struct work_struct *work)
1506{
1507	struct bio *bio, *next;
1508
1509	spin_lock_irq(&bio_dirty_lock);
1510	next = bio_dirty_list;
1511	bio_dirty_list = NULL;
1512	spin_unlock_irq(&bio_dirty_lock);
1513
1514	while ((bio = next) != NULL) {
1515		next = bio->bi_private;
1516
1517		bio_release_pages(bio, true);
1518		bio_put(bio);
1519	}
1520}
1521
1522void bio_check_pages_dirty(struct bio *bio)
1523{
1524	struct folio_iter fi;
1525	unsigned long flags;
 
1526
1527	bio_for_each_folio_all(fi, bio) {
1528		if (!folio_test_dirty(fi.folio))
1529			goto defer;
1530	}
1531
1532	bio_release_pages(bio, false);
1533	bio_put(bio);
1534	return;
1535defer:
1536	spin_lock_irqsave(&bio_dirty_lock, flags);
1537	bio->bi_private = bio_dirty_list;
1538	bio_dirty_list = bio;
1539	spin_unlock_irqrestore(&bio_dirty_lock, flags);
1540	schedule_work(&bio_dirty_work);
1541}
1542EXPORT_SYMBOL_GPL(bio_check_pages_dirty);
1543
1544static inline bool bio_remaining_done(struct bio *bio)
1545{
1546	/*
1547	 * If we're not chaining, then ->__bi_remaining is always 1 and
1548	 * we always end io on the first invocation.
1549	 */
1550	if (!bio_flagged(bio, BIO_CHAIN))
1551		return true;
1552
1553	BUG_ON(atomic_read(&bio->__bi_remaining) <= 0);
1554
1555	if (atomic_dec_and_test(&bio->__bi_remaining)) {
1556		bio_clear_flag(bio, BIO_CHAIN);
1557		return true;
1558	}
1559
1560	return false;
1561}
1562
1563/**
1564 * bio_endio - end I/O on a bio
1565 * @bio:	bio
1566 *
1567 * Description:
1568 *   bio_endio() will end I/O on the whole bio. bio_endio() is the preferred
1569 *   way to end I/O on a bio. No one should call bi_end_io() directly on a
1570 *   bio unless they own it and thus know that it has an end_io function.
1571 *
1572 *   bio_endio() can be called several times on a bio that has been chained
1573 *   using bio_chain().  The ->bi_end_io() function will only be called the
1574 *   last time.
1575 **/
1576void bio_endio(struct bio *bio)
1577{
1578again:
1579	if (!bio_remaining_done(bio))
1580		return;
1581	if (!bio_integrity_endio(bio))
1582		return;
1583
1584	rq_qos_done_bio(bio);
1585
1586	if (bio->bi_bdev && bio_flagged(bio, BIO_TRACE_COMPLETION)) {
1587		trace_block_bio_complete(bdev_get_queue(bio->bi_bdev), bio);
1588		bio_clear_flag(bio, BIO_TRACE_COMPLETION);
1589	}
1590
1591	/*
1592	 * Need to have a real endio function for chained bios, otherwise
1593	 * various corner cases will break (like stacking block devices that
1594	 * save/restore bi_end_io) - however, we want to avoid unbounded
1595	 * recursion and blowing the stack. Tail call optimization would
1596	 * handle this, but compiling with frame pointers also disables
1597	 * gcc's sibling call optimization.
1598	 */
1599	if (bio->bi_end_io == bio_chain_endio) {
1600		bio = __bio_chain_endio(bio);
1601		goto again;
1602	}
1603
1604	blk_throtl_bio_endio(bio);
1605	/* release cgroup info */
1606	bio_uninit(bio);
1607	if (bio->bi_end_io)
1608		bio->bi_end_io(bio);
1609}
1610EXPORT_SYMBOL(bio_endio);
1611
1612/**
1613 * bio_split - split a bio
1614 * @bio:	bio to split
1615 * @sectors:	number of sectors to split from the front of @bio
1616 * @gfp:	gfp mask
1617 * @bs:		bio set to allocate from
1618 *
1619 * Allocates and returns a new bio which represents @sectors from the start of
1620 * @bio, and updates @bio to represent the remaining sectors.
1621 *
1622 * Unless this is a discard request the newly allocated bio will point
1623 * to @bio's bi_io_vec. It is the caller's responsibility to ensure that
1624 * neither @bio nor @bs are freed before the split bio.
1625 */
1626struct bio *bio_split(struct bio *bio, int sectors,
1627		      gfp_t gfp, struct bio_set *bs)
1628{
1629	struct bio *split;
1630
1631	BUG_ON(sectors <= 0);
1632	BUG_ON(sectors >= bio_sectors(bio));
1633
1634	/* Zone append commands cannot be split */
1635	if (WARN_ON_ONCE(bio_op(bio) == REQ_OP_ZONE_APPEND))
1636		return NULL;
1637
1638	split = bio_alloc_clone(bio->bi_bdev, bio, gfp, bs);
1639	if (!split)
1640		return NULL;
1641
1642	split->bi_iter.bi_size = sectors << 9;
1643
1644	if (bio_integrity(split))
1645		bio_integrity_trim(split);
1646
1647	bio_advance(bio, split->bi_iter.bi_size);
1648
1649	if (bio_flagged(bio, BIO_TRACE_COMPLETION))
1650		bio_set_flag(split, BIO_TRACE_COMPLETION);
1651
1652	return split;
1653}
1654EXPORT_SYMBOL(bio_split);
1655
1656/**
1657 * bio_trim - trim a bio
1658 * @bio:	bio to trim
1659 * @offset:	number of sectors to trim from the front of @bio
1660 * @size:	size we want to trim @bio to, in sectors
1661 *
1662 * This function is typically used for bios that are cloned and submitted
1663 * to the underlying device in parts.
1664 */
1665void bio_trim(struct bio *bio, sector_t offset, sector_t size)
1666{
1667	if (WARN_ON_ONCE(offset > BIO_MAX_SECTORS || size > BIO_MAX_SECTORS ||
1668			 offset + size > bio_sectors(bio)))
1669		return;
1670
1671	size <<= 9;
1672	if (offset == 0 && size == bio->bi_iter.bi_size)
1673		return;
1674
1675	bio_advance(bio, offset << 9);
1676	bio->bi_iter.bi_size = size;
1677
1678	if (bio_integrity(bio))
1679		bio_integrity_trim(bio);
1680}
1681EXPORT_SYMBOL_GPL(bio_trim);
1682
1683/*
1684 * create memory pools for biovec's in a bio_set.
1685 * use the global biovec slabs created for general use.
1686 */
1687int biovec_init_pool(mempool_t *pool, int pool_entries)
1688{
1689	struct biovec_slab *bp = bvec_slabs + ARRAY_SIZE(bvec_slabs) - 1;
1690
1691	return mempool_init_slab_pool(pool, pool_entries, bp->slab);
1692}
1693
1694/*
1695 * bioset_exit - exit a bioset initialized with bioset_init()
1696 *
1697 * May be called on a zeroed but uninitialized bioset (i.e. allocated with
1698 * kzalloc()).
1699 */
1700void bioset_exit(struct bio_set *bs)
1701{
1702	bio_alloc_cache_destroy(bs);
1703	if (bs->rescue_workqueue)
1704		destroy_workqueue(bs->rescue_workqueue);
1705	bs->rescue_workqueue = NULL;
1706
1707	mempool_exit(&bs->bio_pool);
1708	mempool_exit(&bs->bvec_pool);
1709
1710	bioset_integrity_free(bs);
1711	if (bs->bio_slab)
1712		bio_put_slab(bs);
1713	bs->bio_slab = NULL;
1714}
1715EXPORT_SYMBOL(bioset_exit);
1716
1717/**
1718 * bioset_init - Initialize a bio_set
1719 * @bs:		pool to initialize
1720 * @pool_size:	Number of bio and bio_vecs to cache in the mempool
1721 * @front_pad:	Number of bytes to allocate in front of the returned bio
1722 * @flags:	Flags to modify behavior, currently %BIOSET_NEED_BVECS
1723 *              and %BIOSET_NEED_RESCUER
1724 *
1725 * Description:
1726 *    Set up a bio_set to be used with @bio_alloc_bioset. Allows the caller
1727 *    to ask for a number of bytes to be allocated in front of the bio.
1728 *    Front pad allocation is useful for embedding the bio inside
1729 *    another structure, to avoid allocating extra data to go with the bio.
1730 *    Note that the bio must be embedded at the END of that structure always,
1731 *    or things will break badly.
1732 *    If %BIOSET_NEED_BVECS is set in @flags, a separate pool will be allocated
1733 *    for allocating iovecs.  This pool is not needed e.g. for bio_init_clone().
1734 *    If %BIOSET_NEED_RESCUER is set, a workqueue is created which can be used
1735 *    to dispatch queued requests when the mempool runs out of space.
1736 *
1737 */
1738int bioset_init(struct bio_set *bs,
1739		unsigned int pool_size,
1740		unsigned int front_pad,
1741		int flags)
1742{
1743	bs->front_pad = front_pad;
1744	if (flags & BIOSET_NEED_BVECS)
1745		bs->back_pad = BIO_INLINE_VECS * sizeof(struct bio_vec);
1746	else
1747		bs->back_pad = 0;
1748
1749	spin_lock_init(&bs->rescue_lock);
1750	bio_list_init(&bs->rescue_list);
1751	INIT_WORK(&bs->rescue_work, bio_alloc_rescue);
1752
1753	bs->bio_slab = bio_find_or_create_slab(bs);
1754	if (!bs->bio_slab)
1755		return -ENOMEM;
1756
1757	if (mempool_init_slab_pool(&bs->bio_pool, pool_size, bs->bio_slab))
1758		goto bad;
1759
1760	if ((flags & BIOSET_NEED_BVECS) &&
1761	    biovec_init_pool(&bs->bvec_pool, pool_size))
1762		goto bad;
1763
1764	if (flags & BIOSET_NEED_RESCUER) {
1765		bs->rescue_workqueue = alloc_workqueue("bioset",
1766							WQ_MEM_RECLAIM, 0);
1767		if (!bs->rescue_workqueue)
1768			goto bad;
1769	}
1770	if (flags & BIOSET_PERCPU_CACHE) {
1771		bs->cache = alloc_percpu(struct bio_alloc_cache);
1772		if (!bs->cache)
1773			goto bad;
1774		cpuhp_state_add_instance_nocalls(CPUHP_BIO_DEAD, &bs->cpuhp_dead);
1775	}
1776
1777	return 0;
1778bad:
1779	bioset_exit(bs);
1780	return -ENOMEM;
1781}
1782EXPORT_SYMBOL(bioset_init);
1783
1784static int __init init_bio(void)
1785{
1786	int i;
1787
1788	BUILD_BUG_ON(BIO_FLAG_LAST > 8 * sizeof_field(struct bio, bi_flags));
1789
1790	bio_integrity_init();
1791
1792	for (i = 0; i < ARRAY_SIZE(bvec_slabs); i++) {
1793		struct biovec_slab *bvs = bvec_slabs + i;
1794
1795		bvs->slab = kmem_cache_create(bvs->name,
1796				bvs->nr_vecs * sizeof(struct bio_vec), 0,
1797				SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
1798	}
1799
1800	cpuhp_setup_state_multi(CPUHP_BIO_DEAD, "block/bio:dead", NULL,
1801					bio_cpu_dead);
1802
1803	if (bioset_init(&fs_bio_set, BIO_POOL_SIZE, 0,
1804			BIOSET_NEED_BVECS | BIOSET_PERCPU_CACHE))
1805		panic("bio: can't allocate bios\n");
1806
1807	if (bioset_integrity_create(&fs_bio_set, BIO_POOL_SIZE))
1808		panic("bio: can't create integrity pool\n");
1809
1810	return 0;
1811}
1812subsys_initcall(init_bio);