Linux Audio

Check our new training course

Loading...
Note: File does not exist in v3.5.6.
   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * fs/f2fs/data.c
   4 *
   5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6 *             http://www.samsung.com/
   7 */
   8#include <linux/fs.h>
   9#include <linux/f2fs_fs.h>
  10#include <linux/buffer_head.h>
  11#include <linux/sched/mm.h>
  12#include <linux/mpage.h>
  13#include <linux/writeback.h>
  14#include <linux/pagevec.h>
  15#include <linux/blkdev.h>
  16#include <linux/bio.h>
  17#include <linux/blk-crypto.h>
  18#include <linux/swap.h>
  19#include <linux/prefetch.h>
  20#include <linux/uio.h>
  21#include <linux/sched/signal.h>
  22#include <linux/fiemap.h>
  23#include <linux/iomap.h>
  24
  25#include "f2fs.h"
  26#include "node.h"
  27#include "segment.h"
  28#include "iostat.h"
  29#include <trace/events/f2fs.h>
  30
  31#define NUM_PREALLOC_POST_READ_CTXS	128
  32
  33static struct kmem_cache *bio_post_read_ctx_cache;
  34static struct kmem_cache *bio_entry_slab;
  35static mempool_t *bio_post_read_ctx_pool;
  36static struct bio_set f2fs_bioset;
  37
  38#define	F2FS_BIO_POOL_SIZE	NR_CURSEG_TYPE
  39
  40int __init f2fs_init_bioset(void)
  41{
  42	return bioset_init(&f2fs_bioset, F2FS_BIO_POOL_SIZE,
  43					0, BIOSET_NEED_BVECS);
  44}
  45
  46void f2fs_destroy_bioset(void)
  47{
  48	bioset_exit(&f2fs_bioset);
  49}
  50
  51static bool __is_cp_guaranteed(struct page *page)
  52{
  53	struct address_space *mapping = page->mapping;
  54	struct inode *inode;
  55	struct f2fs_sb_info *sbi;
  56
  57	if (!mapping)
  58		return false;
  59
  60	inode = mapping->host;
  61	sbi = F2FS_I_SB(inode);
  62
  63	if (inode->i_ino == F2FS_META_INO(sbi) ||
  64			inode->i_ino == F2FS_NODE_INO(sbi) ||
  65			S_ISDIR(inode->i_mode))
  66		return true;
  67
  68	if (f2fs_is_compressed_page(page))
  69		return false;
  70	if ((S_ISREG(inode->i_mode) && IS_NOQUOTA(inode)) ||
  71			page_private_gcing(page))
  72		return true;
  73	return false;
  74}
  75
  76static enum count_type __read_io_type(struct page *page)
  77{
  78	struct address_space *mapping = page_file_mapping(page);
  79
  80	if (mapping) {
  81		struct inode *inode = mapping->host;
  82		struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
  83
  84		if (inode->i_ino == F2FS_META_INO(sbi))
  85			return F2FS_RD_META;
  86
  87		if (inode->i_ino == F2FS_NODE_INO(sbi))
  88			return F2FS_RD_NODE;
  89	}
  90	return F2FS_RD_DATA;
  91}
  92
  93/* postprocessing steps for read bios */
  94enum bio_post_read_step {
  95#ifdef CONFIG_FS_ENCRYPTION
  96	STEP_DECRYPT	= BIT(0),
  97#else
  98	STEP_DECRYPT	= 0,	/* compile out the decryption-related code */
  99#endif
 100#ifdef CONFIG_F2FS_FS_COMPRESSION
 101	STEP_DECOMPRESS	= BIT(1),
 102#else
 103	STEP_DECOMPRESS	= 0,	/* compile out the decompression-related code */
 104#endif
 105#ifdef CONFIG_FS_VERITY
 106	STEP_VERITY	= BIT(2),
 107#else
 108	STEP_VERITY	= 0,	/* compile out the verity-related code */
 109#endif
 110};
 111
 112struct bio_post_read_ctx {
 113	struct bio *bio;
 114	struct f2fs_sb_info *sbi;
 115	struct work_struct work;
 116	unsigned int enabled_steps;
 117	/*
 118	 * decompression_attempted keeps track of whether
 119	 * f2fs_end_read_compressed_page() has been called on the pages in the
 120	 * bio that belong to a compressed cluster yet.
 121	 */
 122	bool decompression_attempted;
 123	block_t fs_blkaddr;
 124};
 125
 126/*
 127 * Update and unlock a bio's pages, and free the bio.
 128 *
 129 * This marks pages up-to-date only if there was no error in the bio (I/O error,
 130 * decryption error, or verity error), as indicated by bio->bi_status.
 131 *
 132 * "Compressed pages" (pagecache pages backed by a compressed cluster on-disk)
 133 * aren't marked up-to-date here, as decompression is done on a per-compression-
 134 * cluster basis rather than a per-bio basis.  Instead, we only must do two
 135 * things for each compressed page here: call f2fs_end_read_compressed_page()
 136 * with failed=true if an error occurred before it would have normally gotten
 137 * called (i.e., I/O error or decryption error, but *not* verity error), and
 138 * release the bio's reference to the decompress_io_ctx of the page's cluster.
 139 */
 140static void f2fs_finish_read_bio(struct bio *bio, bool in_task)
 141{
 142	struct bio_vec *bv;
 143	struct bvec_iter_all iter_all;
 144	struct bio_post_read_ctx *ctx = bio->bi_private;
 145
 146	bio_for_each_segment_all(bv, bio, iter_all) {
 147		struct page *page = bv->bv_page;
 148
 149		if (f2fs_is_compressed_page(page)) {
 150			if (ctx && !ctx->decompression_attempted)
 151				f2fs_end_read_compressed_page(page, true, 0,
 152							in_task);
 153			f2fs_put_page_dic(page, in_task);
 154			continue;
 155		}
 156
 157		if (bio->bi_status)
 158			ClearPageUptodate(page);
 159		else
 160			SetPageUptodate(page);
 161		dec_page_count(F2FS_P_SB(page), __read_io_type(page));
 162		unlock_page(page);
 163	}
 164
 165	if (ctx)
 166		mempool_free(ctx, bio_post_read_ctx_pool);
 167	bio_put(bio);
 168}
 169
 170static void f2fs_verify_bio(struct work_struct *work)
 171{
 172	struct bio_post_read_ctx *ctx =
 173		container_of(work, struct bio_post_read_ctx, work);
 174	struct bio *bio = ctx->bio;
 175	bool may_have_compressed_pages = (ctx->enabled_steps & STEP_DECOMPRESS);
 176
 177	/*
 178	 * fsverity_verify_bio() may call readahead() again, and while verity
 179	 * will be disabled for this, decryption and/or decompression may still
 180	 * be needed, resulting in another bio_post_read_ctx being allocated.
 181	 * So to prevent deadlocks we need to release the current ctx to the
 182	 * mempool first.  This assumes that verity is the last post-read step.
 183	 */
 184	mempool_free(ctx, bio_post_read_ctx_pool);
 185	bio->bi_private = NULL;
 186
 187	/*
 188	 * Verify the bio's pages with fs-verity.  Exclude compressed pages,
 189	 * as those were handled separately by f2fs_end_read_compressed_page().
 190	 */
 191	if (may_have_compressed_pages) {
 192		struct bio_vec *bv;
 193		struct bvec_iter_all iter_all;
 194
 195		bio_for_each_segment_all(bv, bio, iter_all) {
 196			struct page *page = bv->bv_page;
 197
 198			if (!f2fs_is_compressed_page(page) &&
 199			    !fsverity_verify_page(page)) {
 200				bio->bi_status = BLK_STS_IOERR;
 201				break;
 202			}
 203		}
 204	} else {
 205		fsverity_verify_bio(bio);
 206	}
 207
 208	f2fs_finish_read_bio(bio, true);
 209}
 210
 211/*
 212 * If the bio's data needs to be verified with fs-verity, then enqueue the
 213 * verity work for the bio.  Otherwise finish the bio now.
 214 *
 215 * Note that to avoid deadlocks, the verity work can't be done on the
 216 * decryption/decompression workqueue.  This is because verifying the data pages
 217 * can involve reading verity metadata pages from the file, and these verity
 218 * metadata pages may be encrypted and/or compressed.
 219 */
 220static void f2fs_verify_and_finish_bio(struct bio *bio, bool in_task)
 221{
 222	struct bio_post_read_ctx *ctx = bio->bi_private;
 223
 224	if (ctx && (ctx->enabled_steps & STEP_VERITY)) {
 225		INIT_WORK(&ctx->work, f2fs_verify_bio);
 226		fsverity_enqueue_verify_work(&ctx->work);
 227	} else {
 228		f2fs_finish_read_bio(bio, in_task);
 229	}
 230}
 231
 232/*
 233 * Handle STEP_DECOMPRESS by decompressing any compressed clusters whose last
 234 * remaining page was read by @ctx->bio.
 235 *
 236 * Note that a bio may span clusters (even a mix of compressed and uncompressed
 237 * clusters) or be for just part of a cluster.  STEP_DECOMPRESS just indicates
 238 * that the bio includes at least one compressed page.  The actual decompression
 239 * is done on a per-cluster basis, not a per-bio basis.
 240 */
 241static void f2fs_handle_step_decompress(struct bio_post_read_ctx *ctx,
 242		bool in_task)
 243{
 244	struct bio_vec *bv;
 245	struct bvec_iter_all iter_all;
 246	bool all_compressed = true;
 247	block_t blkaddr = ctx->fs_blkaddr;
 248
 249	bio_for_each_segment_all(bv, ctx->bio, iter_all) {
 250		struct page *page = bv->bv_page;
 251
 252		if (f2fs_is_compressed_page(page))
 253			f2fs_end_read_compressed_page(page, false, blkaddr,
 254						      in_task);
 255		else
 256			all_compressed = false;
 257
 258		blkaddr++;
 259	}
 260
 261	ctx->decompression_attempted = true;
 262
 263	/*
 264	 * Optimization: if all the bio's pages are compressed, then scheduling
 265	 * the per-bio verity work is unnecessary, as verity will be fully
 266	 * handled at the compression cluster level.
 267	 */
 268	if (all_compressed)
 269		ctx->enabled_steps &= ~STEP_VERITY;
 270}
 271
 272static void f2fs_post_read_work(struct work_struct *work)
 273{
 274	struct bio_post_read_ctx *ctx =
 275		container_of(work, struct bio_post_read_ctx, work);
 276	struct bio *bio = ctx->bio;
 277
 278	if ((ctx->enabled_steps & STEP_DECRYPT) && !fscrypt_decrypt_bio(bio)) {
 279		f2fs_finish_read_bio(bio, true);
 280		return;
 281	}
 282
 283	if (ctx->enabled_steps & STEP_DECOMPRESS)
 284		f2fs_handle_step_decompress(ctx, true);
 285
 286	f2fs_verify_and_finish_bio(bio, true);
 287}
 288
 289static void f2fs_read_end_io(struct bio *bio)
 290{
 291	struct f2fs_sb_info *sbi = F2FS_P_SB(bio_first_page_all(bio));
 292	struct bio_post_read_ctx *ctx;
 293	bool intask = in_task();
 294
 295	iostat_update_and_unbind_ctx(bio);
 296	ctx = bio->bi_private;
 297
 298	if (time_to_inject(sbi, FAULT_READ_IO))
 299		bio->bi_status = BLK_STS_IOERR;
 300
 301	if (bio->bi_status) {
 302		f2fs_finish_read_bio(bio, intask);
 303		return;
 304	}
 305
 306	if (ctx) {
 307		unsigned int enabled_steps = ctx->enabled_steps &
 308					(STEP_DECRYPT | STEP_DECOMPRESS);
 309
 310		/*
 311		 * If we have only decompression step between decompression and
 312		 * decrypt, we don't need post processing for this.
 313		 */
 314		if (enabled_steps == STEP_DECOMPRESS &&
 315				!f2fs_low_mem_mode(sbi)) {
 316			f2fs_handle_step_decompress(ctx, intask);
 317		} else if (enabled_steps) {
 318			INIT_WORK(&ctx->work, f2fs_post_read_work);
 319			queue_work(ctx->sbi->post_read_wq, &ctx->work);
 320			return;
 321		}
 322	}
 323
 324	f2fs_verify_and_finish_bio(bio, intask);
 325}
 326
 327static void f2fs_write_end_io(struct bio *bio)
 328{
 329	struct f2fs_sb_info *sbi;
 330	struct bio_vec *bvec;
 331	struct bvec_iter_all iter_all;
 332
 333	iostat_update_and_unbind_ctx(bio);
 334	sbi = bio->bi_private;
 335
 336	if (time_to_inject(sbi, FAULT_WRITE_IO))
 337		bio->bi_status = BLK_STS_IOERR;
 338
 339	bio_for_each_segment_all(bvec, bio, iter_all) {
 340		struct page *page = bvec->bv_page;
 341		enum count_type type = WB_DATA_TYPE(page);
 342
 343		if (page_private_dummy(page)) {
 344			clear_page_private_dummy(page);
 345			unlock_page(page);
 346			mempool_free(page, sbi->write_io_dummy);
 347
 348			if (unlikely(bio->bi_status))
 349				f2fs_stop_checkpoint(sbi, true,
 350						STOP_CP_REASON_WRITE_FAIL);
 351			continue;
 352		}
 353
 354		fscrypt_finalize_bounce_page(&page);
 355
 356#ifdef CONFIG_F2FS_FS_COMPRESSION
 357		if (f2fs_is_compressed_page(page)) {
 358			f2fs_compress_write_end_io(bio, page);
 359			continue;
 360		}
 361#endif
 362
 363		if (unlikely(bio->bi_status)) {
 364			mapping_set_error(page->mapping, -EIO);
 365			if (type == F2FS_WB_CP_DATA)
 366				f2fs_stop_checkpoint(sbi, true,
 367						STOP_CP_REASON_WRITE_FAIL);
 368		}
 369
 370		f2fs_bug_on(sbi, page->mapping == NODE_MAPPING(sbi) &&
 371					page->index != nid_of_node(page));
 372
 373		dec_page_count(sbi, type);
 374		if (f2fs_in_warm_node_list(sbi, page))
 375			f2fs_del_fsync_node_entry(sbi, page);
 376		clear_page_private_gcing(page);
 377		end_page_writeback(page);
 378	}
 379	if (!get_pages(sbi, F2FS_WB_CP_DATA) &&
 380				wq_has_sleeper(&sbi->cp_wait))
 381		wake_up(&sbi->cp_wait);
 382
 383	bio_put(bio);
 384}
 385
 386#ifdef CONFIG_BLK_DEV_ZONED
 387static void f2fs_zone_write_end_io(struct bio *bio)
 388{
 389	struct f2fs_bio_info *io = (struct f2fs_bio_info *)bio->bi_private;
 390
 391	bio->bi_private = io->bi_private;
 392	complete(&io->zone_wait);
 393	f2fs_write_end_io(bio);
 394}
 395#endif
 396
 397struct block_device *f2fs_target_device(struct f2fs_sb_info *sbi,
 398		block_t blk_addr, sector_t *sector)
 399{
 400	struct block_device *bdev = sbi->sb->s_bdev;
 401	int i;
 402
 403	if (f2fs_is_multi_device(sbi)) {
 404		for (i = 0; i < sbi->s_ndevs; i++) {
 405			if (FDEV(i).start_blk <= blk_addr &&
 406			    FDEV(i).end_blk >= blk_addr) {
 407				blk_addr -= FDEV(i).start_blk;
 408				bdev = FDEV(i).bdev;
 409				break;
 410			}
 411		}
 412	}
 413
 414	if (sector)
 415		*sector = SECTOR_FROM_BLOCK(blk_addr);
 416	return bdev;
 417}
 418
 419int f2fs_target_device_index(struct f2fs_sb_info *sbi, block_t blkaddr)
 420{
 421	int i;
 422
 423	if (!f2fs_is_multi_device(sbi))
 424		return 0;
 425
 426	for (i = 0; i < sbi->s_ndevs; i++)
 427		if (FDEV(i).start_blk <= blkaddr && FDEV(i).end_blk >= blkaddr)
 428			return i;
 429	return 0;
 430}
 431
 432static blk_opf_t f2fs_io_flags(struct f2fs_io_info *fio)
 433{
 434	unsigned int temp_mask = GENMASK(NR_TEMP_TYPE - 1, 0);
 435	unsigned int fua_flag, meta_flag, io_flag;
 436	blk_opf_t op_flags = 0;
 437
 438	if (fio->op != REQ_OP_WRITE)
 439		return 0;
 440	if (fio->type == DATA)
 441		io_flag = fio->sbi->data_io_flag;
 442	else if (fio->type == NODE)
 443		io_flag = fio->sbi->node_io_flag;
 444	else
 445		return 0;
 446
 447	fua_flag = io_flag & temp_mask;
 448	meta_flag = (io_flag >> NR_TEMP_TYPE) & temp_mask;
 449
 450	/*
 451	 * data/node io flag bits per temp:
 452	 *      REQ_META     |      REQ_FUA      |
 453	 *    5 |    4 |   3 |    2 |    1 |   0 |
 454	 * Cold | Warm | Hot | Cold | Warm | Hot |
 455	 */
 456	if (BIT(fio->temp) & meta_flag)
 457		op_flags |= REQ_META;
 458	if (BIT(fio->temp) & fua_flag)
 459		op_flags |= REQ_FUA;
 460	return op_flags;
 461}
 462
 463static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages)
 464{
 465	struct f2fs_sb_info *sbi = fio->sbi;
 466	struct block_device *bdev;
 467	sector_t sector;
 468	struct bio *bio;
 469
 470	bdev = f2fs_target_device(sbi, fio->new_blkaddr, &sector);
 471	bio = bio_alloc_bioset(bdev, npages,
 472				fio->op | fio->op_flags | f2fs_io_flags(fio),
 473				GFP_NOIO, &f2fs_bioset);
 474	bio->bi_iter.bi_sector = sector;
 475	if (is_read_io(fio->op)) {
 476		bio->bi_end_io = f2fs_read_end_io;
 477		bio->bi_private = NULL;
 478	} else {
 479		bio->bi_end_io = f2fs_write_end_io;
 480		bio->bi_private = sbi;
 481	}
 482	iostat_alloc_and_bind_ctx(sbi, bio, NULL);
 483
 484	if (fio->io_wbc)
 485		wbc_init_bio(fio->io_wbc, bio);
 486
 487	return bio;
 488}
 489
 490static void f2fs_set_bio_crypt_ctx(struct bio *bio, const struct inode *inode,
 491				  pgoff_t first_idx,
 492				  const struct f2fs_io_info *fio,
 493				  gfp_t gfp_mask)
 494{
 495	/*
 496	 * The f2fs garbage collector sets ->encrypted_page when it wants to
 497	 * read/write raw data without encryption.
 498	 */
 499	if (!fio || !fio->encrypted_page)
 500		fscrypt_set_bio_crypt_ctx(bio, inode, first_idx, gfp_mask);
 501}
 502
 503static bool f2fs_crypt_mergeable_bio(struct bio *bio, const struct inode *inode,
 504				     pgoff_t next_idx,
 505				     const struct f2fs_io_info *fio)
 506{
 507	/*
 508	 * The f2fs garbage collector sets ->encrypted_page when it wants to
 509	 * read/write raw data without encryption.
 510	 */
 511	if (fio && fio->encrypted_page)
 512		return !bio_has_crypt_ctx(bio);
 513
 514	return fscrypt_mergeable_bio(bio, inode, next_idx);
 515}
 516
 517void f2fs_submit_read_bio(struct f2fs_sb_info *sbi, struct bio *bio,
 518				 enum page_type type)
 519{
 520	WARN_ON_ONCE(!is_read_io(bio_op(bio)));
 521	trace_f2fs_submit_read_bio(sbi->sb, type, bio);
 522
 523	iostat_update_submit_ctx(bio, type);
 524	submit_bio(bio);
 525}
 526
 527static void f2fs_align_write_bio(struct f2fs_sb_info *sbi, struct bio *bio)
 528{
 529	unsigned int start =
 530		(bio->bi_iter.bi_size >> F2FS_BLKSIZE_BITS) % F2FS_IO_SIZE(sbi);
 531
 532	if (start == 0)
 533		return;
 534
 535	/* fill dummy pages */
 536	for (; start < F2FS_IO_SIZE(sbi); start++) {
 537		struct page *page =
 538			mempool_alloc(sbi->write_io_dummy,
 539				      GFP_NOIO | __GFP_NOFAIL);
 540		f2fs_bug_on(sbi, !page);
 541
 542		lock_page(page);
 543
 544		zero_user_segment(page, 0, PAGE_SIZE);
 545		set_page_private_dummy(page);
 546
 547		if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE)
 548			f2fs_bug_on(sbi, 1);
 549	}
 550}
 551
 552static void f2fs_submit_write_bio(struct f2fs_sb_info *sbi, struct bio *bio,
 553				  enum page_type type)
 554{
 555	WARN_ON_ONCE(is_read_io(bio_op(bio)));
 556
 557	if (type == DATA || type == NODE) {
 558		if (f2fs_lfs_mode(sbi) && current->plug)
 559			blk_finish_plug(current->plug);
 560
 561		if (F2FS_IO_ALIGNED(sbi)) {
 562			f2fs_align_write_bio(sbi, bio);
 563			/*
 564			 * In the NODE case, we lose next block address chain.
 565			 * So, we need to do checkpoint in f2fs_sync_file.
 566			 */
 567			if (type == NODE)
 568				set_sbi_flag(sbi, SBI_NEED_CP);
 569		}
 570	}
 571
 572	trace_f2fs_submit_write_bio(sbi->sb, type, bio);
 573	iostat_update_submit_ctx(bio, type);
 574	submit_bio(bio);
 575}
 576
 577static void __submit_merged_bio(struct f2fs_bio_info *io)
 578{
 579	struct f2fs_io_info *fio = &io->fio;
 580
 581	if (!io->bio)
 582		return;
 583
 584	if (is_read_io(fio->op)) {
 585		trace_f2fs_prepare_read_bio(io->sbi->sb, fio->type, io->bio);
 586		f2fs_submit_read_bio(io->sbi, io->bio, fio->type);
 587	} else {
 588		trace_f2fs_prepare_write_bio(io->sbi->sb, fio->type, io->bio);
 589		f2fs_submit_write_bio(io->sbi, io->bio, fio->type);
 590	}
 591	io->bio = NULL;
 592}
 593
 594static bool __has_merged_page(struct bio *bio, struct inode *inode,
 595						struct page *page, nid_t ino)
 596{
 597	struct bio_vec *bvec;
 598	struct bvec_iter_all iter_all;
 599
 600	if (!bio)
 601		return false;
 602
 603	if (!inode && !page && !ino)
 604		return true;
 605
 606	bio_for_each_segment_all(bvec, bio, iter_all) {
 607		struct page *target = bvec->bv_page;
 608
 609		if (fscrypt_is_bounce_page(target)) {
 610			target = fscrypt_pagecache_page(target);
 611			if (IS_ERR(target))
 612				continue;
 613		}
 614		if (f2fs_is_compressed_page(target)) {
 615			target = f2fs_compress_control_page(target);
 616			if (IS_ERR(target))
 617				continue;
 618		}
 619
 620		if (inode && inode == target->mapping->host)
 621			return true;
 622		if (page && page == target)
 623			return true;
 624		if (ino && ino == ino_of_node(target))
 625			return true;
 626	}
 627
 628	return false;
 629}
 630
 631int f2fs_init_write_merge_io(struct f2fs_sb_info *sbi)
 632{
 633	int i;
 634
 635	for (i = 0; i < NR_PAGE_TYPE; i++) {
 636		int n = (i == META) ? 1 : NR_TEMP_TYPE;
 637		int j;
 638
 639		sbi->write_io[i] = f2fs_kmalloc(sbi,
 640				array_size(n, sizeof(struct f2fs_bio_info)),
 641				GFP_KERNEL);
 642		if (!sbi->write_io[i])
 643			return -ENOMEM;
 644
 645		for (j = HOT; j < n; j++) {
 646			init_f2fs_rwsem(&sbi->write_io[i][j].io_rwsem);
 647			sbi->write_io[i][j].sbi = sbi;
 648			sbi->write_io[i][j].bio = NULL;
 649			spin_lock_init(&sbi->write_io[i][j].io_lock);
 650			INIT_LIST_HEAD(&sbi->write_io[i][j].io_list);
 651			INIT_LIST_HEAD(&sbi->write_io[i][j].bio_list);
 652			init_f2fs_rwsem(&sbi->write_io[i][j].bio_list_lock);
 653#ifdef CONFIG_BLK_DEV_ZONED
 654			init_completion(&sbi->write_io[i][j].zone_wait);
 655			sbi->write_io[i][j].zone_pending_bio = NULL;
 656			sbi->write_io[i][j].bi_private = NULL;
 657#endif
 658		}
 659	}
 660
 661	return 0;
 662}
 663
 664static void __f2fs_submit_merged_write(struct f2fs_sb_info *sbi,
 665				enum page_type type, enum temp_type temp)
 666{
 667	enum page_type btype = PAGE_TYPE_OF_BIO(type);
 668	struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 669
 670	f2fs_down_write(&io->io_rwsem);
 671
 672	if (!io->bio)
 673		goto unlock_out;
 674
 675	/* change META to META_FLUSH in the checkpoint procedure */
 676	if (type >= META_FLUSH) {
 677		io->fio.type = META_FLUSH;
 678		io->bio->bi_opf |= REQ_META | REQ_PRIO | REQ_SYNC;
 679		if (!test_opt(sbi, NOBARRIER))
 680			io->bio->bi_opf |= REQ_PREFLUSH | REQ_FUA;
 681	}
 682	__submit_merged_bio(io);
 683unlock_out:
 684	f2fs_up_write(&io->io_rwsem);
 685}
 686
 687static void __submit_merged_write_cond(struct f2fs_sb_info *sbi,
 688				struct inode *inode, struct page *page,
 689				nid_t ino, enum page_type type, bool force)
 690{
 691	enum temp_type temp;
 692	bool ret = true;
 693
 694	for (temp = HOT; temp < NR_TEMP_TYPE; temp++) {
 695		if (!force)	{
 696			enum page_type btype = PAGE_TYPE_OF_BIO(type);
 697			struct f2fs_bio_info *io = sbi->write_io[btype] + temp;
 698
 699			f2fs_down_read(&io->io_rwsem);
 700			ret = __has_merged_page(io->bio, inode, page, ino);
 701			f2fs_up_read(&io->io_rwsem);
 702		}
 703		if (ret)
 704			__f2fs_submit_merged_write(sbi, type, temp);
 705
 706		/* TODO: use HOT temp only for meta pages now. */
 707		if (type >= META)
 708			break;
 709	}
 710}
 711
 712void f2fs_submit_merged_write(struct f2fs_sb_info *sbi, enum page_type type)
 713{
 714	__submit_merged_write_cond(sbi, NULL, NULL, 0, type, true);
 715}
 716
 717void f2fs_submit_merged_write_cond(struct f2fs_sb_info *sbi,
 718				struct inode *inode, struct page *page,
 719				nid_t ino, enum page_type type)
 720{
 721	__submit_merged_write_cond(sbi, inode, page, ino, type, false);
 722}
 723
 724void f2fs_flush_merged_writes(struct f2fs_sb_info *sbi)
 725{
 726	f2fs_submit_merged_write(sbi, DATA);
 727	f2fs_submit_merged_write(sbi, NODE);
 728	f2fs_submit_merged_write(sbi, META);
 729}
 730
 731/*
 732 * Fill the locked page with data located in the block address.
 733 * A caller needs to unlock the page on failure.
 734 */
 735int f2fs_submit_page_bio(struct f2fs_io_info *fio)
 736{
 737	struct bio *bio;
 738	struct page *page = fio->encrypted_page ?
 739			fio->encrypted_page : fio->page;
 740
 741	if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 742			fio->is_por ? META_POR : (__is_meta_io(fio) ?
 743			META_GENERIC : DATA_GENERIC_ENHANCE))) {
 744		f2fs_handle_error(fio->sbi, ERROR_INVALID_BLKADDR);
 745		return -EFSCORRUPTED;
 746	}
 747
 748	trace_f2fs_submit_page_bio(page, fio);
 749
 750	/* Allocate a new bio */
 751	bio = __bio_alloc(fio, 1);
 752
 753	f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 754			       fio->page->index, fio, GFP_NOIO);
 755
 756	if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
 757		bio_put(bio);
 758		return -EFAULT;
 759	}
 760
 761	if (fio->io_wbc && !is_read_io(fio->op))
 762		wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE);
 763
 764	inc_page_count(fio->sbi, is_read_io(fio->op) ?
 765			__read_io_type(page) : WB_DATA_TYPE(fio->page));
 766
 767	if (is_read_io(bio_op(bio)))
 768		f2fs_submit_read_bio(fio->sbi, bio, fio->type);
 769	else
 770		f2fs_submit_write_bio(fio->sbi, bio, fio->type);
 771	return 0;
 772}
 773
 774static bool page_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 775				block_t last_blkaddr, block_t cur_blkaddr)
 776{
 777	if (unlikely(sbi->max_io_bytes &&
 778			bio->bi_iter.bi_size >= sbi->max_io_bytes))
 779		return false;
 780	if (last_blkaddr + 1 != cur_blkaddr)
 781		return false;
 782	return bio->bi_bdev == f2fs_target_device(sbi, cur_blkaddr, NULL);
 783}
 784
 785static bool io_type_is_mergeable(struct f2fs_bio_info *io,
 786						struct f2fs_io_info *fio)
 787{
 788	if (io->fio.op != fio->op)
 789		return false;
 790	return io->fio.op_flags == fio->op_flags;
 791}
 792
 793static bool io_is_mergeable(struct f2fs_sb_info *sbi, struct bio *bio,
 794					struct f2fs_bio_info *io,
 795					struct f2fs_io_info *fio,
 796					block_t last_blkaddr,
 797					block_t cur_blkaddr)
 798{
 799	if (F2FS_IO_ALIGNED(sbi) && (fio->type == DATA || fio->type == NODE)) {
 800		unsigned int filled_blocks =
 801				F2FS_BYTES_TO_BLK(bio->bi_iter.bi_size);
 802		unsigned int io_size = F2FS_IO_SIZE(sbi);
 803		unsigned int left_vecs = bio->bi_max_vecs - bio->bi_vcnt;
 804
 805		/* IOs in bio is aligned and left space of vectors is not enough */
 806		if (!(filled_blocks % io_size) && left_vecs < io_size)
 807			return false;
 808	}
 809	if (!page_is_mergeable(sbi, bio, last_blkaddr, cur_blkaddr))
 810		return false;
 811	return io_type_is_mergeable(io, fio);
 812}
 813
 814static void add_bio_entry(struct f2fs_sb_info *sbi, struct bio *bio,
 815				struct page *page, enum temp_type temp)
 816{
 817	struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 818	struct bio_entry *be;
 819
 820	be = f2fs_kmem_cache_alloc(bio_entry_slab, GFP_NOFS, true, NULL);
 821	be->bio = bio;
 822	bio_get(bio);
 823
 824	if (bio_add_page(bio, page, PAGE_SIZE, 0) != PAGE_SIZE)
 825		f2fs_bug_on(sbi, 1);
 826
 827	f2fs_down_write(&io->bio_list_lock);
 828	list_add_tail(&be->list, &io->bio_list);
 829	f2fs_up_write(&io->bio_list_lock);
 830}
 831
 832static void del_bio_entry(struct bio_entry *be)
 833{
 834	list_del(&be->list);
 835	kmem_cache_free(bio_entry_slab, be);
 836}
 837
 838static int add_ipu_page(struct f2fs_io_info *fio, struct bio **bio,
 839							struct page *page)
 840{
 841	struct f2fs_sb_info *sbi = fio->sbi;
 842	enum temp_type temp;
 843	bool found = false;
 844	int ret = -EAGAIN;
 845
 846	for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 847		struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 848		struct list_head *head = &io->bio_list;
 849		struct bio_entry *be;
 850
 851		f2fs_down_write(&io->bio_list_lock);
 852		list_for_each_entry(be, head, list) {
 853			if (be->bio != *bio)
 854				continue;
 855
 856			found = true;
 857
 858			f2fs_bug_on(sbi, !page_is_mergeable(sbi, *bio,
 859							    *fio->last_block,
 860							    fio->new_blkaddr));
 861			if (f2fs_crypt_mergeable_bio(*bio,
 862					fio->page->mapping->host,
 863					fio->page->index, fio) &&
 864			    bio_add_page(*bio, page, PAGE_SIZE, 0) ==
 865					PAGE_SIZE) {
 866				ret = 0;
 867				break;
 868			}
 869
 870			/* page can't be merged into bio; submit the bio */
 871			del_bio_entry(be);
 872			f2fs_submit_write_bio(sbi, *bio, DATA);
 873			break;
 874		}
 875		f2fs_up_write(&io->bio_list_lock);
 876	}
 877
 878	if (ret) {
 879		bio_put(*bio);
 880		*bio = NULL;
 881	}
 882
 883	return ret;
 884}
 885
 886void f2fs_submit_merged_ipu_write(struct f2fs_sb_info *sbi,
 887					struct bio **bio, struct page *page)
 888{
 889	enum temp_type temp;
 890	bool found = false;
 891	struct bio *target = bio ? *bio : NULL;
 892
 893	f2fs_bug_on(sbi, !target && !page);
 894
 895	for (temp = HOT; temp < NR_TEMP_TYPE && !found; temp++) {
 896		struct f2fs_bio_info *io = sbi->write_io[DATA] + temp;
 897		struct list_head *head = &io->bio_list;
 898		struct bio_entry *be;
 899
 900		if (list_empty(head))
 901			continue;
 902
 903		f2fs_down_read(&io->bio_list_lock);
 904		list_for_each_entry(be, head, list) {
 905			if (target)
 906				found = (target == be->bio);
 907			else
 908				found = __has_merged_page(be->bio, NULL,
 909								page, 0);
 910			if (found)
 911				break;
 912		}
 913		f2fs_up_read(&io->bio_list_lock);
 914
 915		if (!found)
 916			continue;
 917
 918		found = false;
 919
 920		f2fs_down_write(&io->bio_list_lock);
 921		list_for_each_entry(be, head, list) {
 922			if (target)
 923				found = (target == be->bio);
 924			else
 925				found = __has_merged_page(be->bio, NULL,
 926								page, 0);
 927			if (found) {
 928				target = be->bio;
 929				del_bio_entry(be);
 930				break;
 931			}
 932		}
 933		f2fs_up_write(&io->bio_list_lock);
 934	}
 935
 936	if (found)
 937		f2fs_submit_write_bio(sbi, target, DATA);
 938	if (bio && *bio) {
 939		bio_put(*bio);
 940		*bio = NULL;
 941	}
 942}
 943
 944int f2fs_merge_page_bio(struct f2fs_io_info *fio)
 945{
 946	struct bio *bio = *fio->bio;
 947	struct page *page = fio->encrypted_page ?
 948			fio->encrypted_page : fio->page;
 949
 950	if (!f2fs_is_valid_blkaddr(fio->sbi, fio->new_blkaddr,
 951			__is_meta_io(fio) ? META_GENERIC : DATA_GENERIC)) {
 952		f2fs_handle_error(fio->sbi, ERROR_INVALID_BLKADDR);
 953		return -EFSCORRUPTED;
 954	}
 955
 956	trace_f2fs_submit_page_bio(page, fio);
 957
 958	if (bio && !page_is_mergeable(fio->sbi, bio, *fio->last_block,
 959						fio->new_blkaddr))
 960		f2fs_submit_merged_ipu_write(fio->sbi, &bio, NULL);
 961alloc_new:
 962	if (!bio) {
 963		bio = __bio_alloc(fio, BIO_MAX_VECS);
 964		f2fs_set_bio_crypt_ctx(bio, fio->page->mapping->host,
 965				       fio->page->index, fio, GFP_NOIO);
 966
 967		add_bio_entry(fio->sbi, bio, page, fio->temp);
 968	} else {
 969		if (add_ipu_page(fio, &bio, page))
 970			goto alloc_new;
 971	}
 972
 973	if (fio->io_wbc)
 974		wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE);
 975
 976	inc_page_count(fio->sbi, WB_DATA_TYPE(page));
 977
 978	*fio->last_block = fio->new_blkaddr;
 979	*fio->bio = bio;
 980
 981	return 0;
 982}
 983
 984#ifdef CONFIG_BLK_DEV_ZONED
 985static bool is_end_zone_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr)
 986{
 987	int devi = 0;
 988
 989	if (f2fs_is_multi_device(sbi)) {
 990		devi = f2fs_target_device_index(sbi, blkaddr);
 991		if (blkaddr < FDEV(devi).start_blk ||
 992		    blkaddr > FDEV(devi).end_blk) {
 993			f2fs_err(sbi, "Invalid block %x", blkaddr);
 994			return false;
 995		}
 996		blkaddr -= FDEV(devi).start_blk;
 997	}
 998	return bdev_is_zoned(FDEV(devi).bdev) &&
 999		f2fs_blkz_is_seq(sbi, devi, blkaddr) &&
1000		(blkaddr % sbi->blocks_per_blkz == sbi->blocks_per_blkz - 1);
1001}
1002#endif
1003
1004void f2fs_submit_page_write(struct f2fs_io_info *fio)
1005{
1006	struct f2fs_sb_info *sbi = fio->sbi;
1007	enum page_type btype = PAGE_TYPE_OF_BIO(fio->type);
1008	struct f2fs_bio_info *io = sbi->write_io[btype] + fio->temp;
1009	struct page *bio_page;
1010
1011	f2fs_bug_on(sbi, is_read_io(fio->op));
1012
1013	f2fs_down_write(&io->io_rwsem);
1014
1015#ifdef CONFIG_BLK_DEV_ZONED
1016	if (f2fs_sb_has_blkzoned(sbi) && btype < META && io->zone_pending_bio) {
1017		wait_for_completion_io(&io->zone_wait);
1018		bio_put(io->zone_pending_bio);
1019		io->zone_pending_bio = NULL;
1020		io->bi_private = NULL;
1021	}
1022#endif
1023
1024next:
1025	if (fio->in_list) {
1026		spin_lock(&io->io_lock);
1027		if (list_empty(&io->io_list)) {
1028			spin_unlock(&io->io_lock);
1029			goto out;
1030		}
1031		fio = list_first_entry(&io->io_list,
1032						struct f2fs_io_info, list);
1033		list_del(&fio->list);
1034		spin_unlock(&io->io_lock);
1035	}
1036
1037	verify_fio_blkaddr(fio);
1038
1039	if (fio->encrypted_page)
1040		bio_page = fio->encrypted_page;
1041	else if (fio->compressed_page)
1042		bio_page = fio->compressed_page;
1043	else
1044		bio_page = fio->page;
1045
1046	/* set submitted = true as a return value */
1047	fio->submitted = 1;
1048
1049	inc_page_count(sbi, WB_DATA_TYPE(bio_page));
1050
1051	if (io->bio &&
1052	    (!io_is_mergeable(sbi, io->bio, io, fio, io->last_block_in_bio,
1053			      fio->new_blkaddr) ||
1054	     !f2fs_crypt_mergeable_bio(io->bio, fio->page->mapping->host,
1055				       bio_page->index, fio)))
1056		__submit_merged_bio(io);
1057alloc_new:
1058	if (io->bio == NULL) {
1059		if (F2FS_IO_ALIGNED(sbi) &&
1060				(fio->type == DATA || fio->type == NODE) &&
1061				fio->new_blkaddr & F2FS_IO_SIZE_MASK(sbi)) {
1062			dec_page_count(sbi, WB_DATA_TYPE(bio_page));
1063			fio->retry = 1;
1064			goto skip;
1065		}
1066		io->bio = __bio_alloc(fio, BIO_MAX_VECS);
1067		f2fs_set_bio_crypt_ctx(io->bio, fio->page->mapping->host,
1068				       bio_page->index, fio, GFP_NOIO);
1069		io->fio = *fio;
1070	}
1071
1072	if (bio_add_page(io->bio, bio_page, PAGE_SIZE, 0) < PAGE_SIZE) {
1073		__submit_merged_bio(io);
1074		goto alloc_new;
1075	}
1076
1077	if (fio->io_wbc)
1078		wbc_account_cgroup_owner(fio->io_wbc, fio->page, PAGE_SIZE);
1079
1080	io->last_block_in_bio = fio->new_blkaddr;
1081
1082	trace_f2fs_submit_page_write(fio->page, fio);
1083skip:
1084	if (fio->in_list)
1085		goto next;
1086out:
1087#ifdef CONFIG_BLK_DEV_ZONED
1088	if (f2fs_sb_has_blkzoned(sbi) && btype < META &&
1089			is_end_zone_blkaddr(sbi, fio->new_blkaddr)) {
1090		bio_get(io->bio);
1091		reinit_completion(&io->zone_wait);
1092		io->bi_private = io->bio->bi_private;
1093		io->bio->bi_private = io;
1094		io->bio->bi_end_io = f2fs_zone_write_end_io;
1095		io->zone_pending_bio = io->bio;
1096		__submit_merged_bio(io);
1097	}
1098#endif
1099	if (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) ||
1100				!f2fs_is_checkpoint_ready(sbi))
1101		__submit_merged_bio(io);
1102	f2fs_up_write(&io->io_rwsem);
1103}
1104
1105static struct bio *f2fs_grab_read_bio(struct inode *inode, block_t blkaddr,
1106				      unsigned nr_pages, blk_opf_t op_flag,
1107				      pgoff_t first_idx, bool for_write)
1108{
1109	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1110	struct bio *bio;
1111	struct bio_post_read_ctx *ctx = NULL;
1112	unsigned int post_read_steps = 0;
1113	sector_t sector;
1114	struct block_device *bdev = f2fs_target_device(sbi, blkaddr, &sector);
1115
1116	bio = bio_alloc_bioset(bdev, bio_max_segs(nr_pages),
1117			       REQ_OP_READ | op_flag,
1118			       for_write ? GFP_NOIO : GFP_KERNEL, &f2fs_bioset);
1119	if (!bio)
1120		return ERR_PTR(-ENOMEM);
1121	bio->bi_iter.bi_sector = sector;
1122	f2fs_set_bio_crypt_ctx(bio, inode, first_idx, NULL, GFP_NOFS);
1123	bio->bi_end_io = f2fs_read_end_io;
1124
1125	if (fscrypt_inode_uses_fs_layer_crypto(inode))
1126		post_read_steps |= STEP_DECRYPT;
1127
1128	if (f2fs_need_verity(inode, first_idx))
1129		post_read_steps |= STEP_VERITY;
1130
1131	/*
1132	 * STEP_DECOMPRESS is handled specially, since a compressed file might
1133	 * contain both compressed and uncompressed clusters.  We'll allocate a
1134	 * bio_post_read_ctx if the file is compressed, but the caller is
1135	 * responsible for enabling STEP_DECOMPRESS if it's actually needed.
1136	 */
1137
1138	if (post_read_steps || f2fs_compressed_file(inode)) {
1139		/* Due to the mempool, this never fails. */
1140		ctx = mempool_alloc(bio_post_read_ctx_pool, GFP_NOFS);
1141		ctx->bio = bio;
1142		ctx->sbi = sbi;
1143		ctx->enabled_steps = post_read_steps;
1144		ctx->fs_blkaddr = blkaddr;
1145		ctx->decompression_attempted = false;
1146		bio->bi_private = ctx;
1147	}
1148	iostat_alloc_and_bind_ctx(sbi, bio, ctx);
1149
1150	return bio;
1151}
1152
1153/* This can handle encryption stuffs */
1154static int f2fs_submit_page_read(struct inode *inode, struct page *page,
1155				 block_t blkaddr, blk_opf_t op_flags,
1156				 bool for_write)
1157{
1158	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1159	struct bio *bio;
1160
1161	bio = f2fs_grab_read_bio(inode, blkaddr, 1, op_flags,
1162					page->index, for_write);
1163	if (IS_ERR(bio))
1164		return PTR_ERR(bio);
1165
1166	/* wait for GCed page writeback via META_MAPPING */
1167	f2fs_wait_on_block_writeback(inode, blkaddr);
1168
1169	if (bio_add_page(bio, page, PAGE_SIZE, 0) < PAGE_SIZE) {
1170		iostat_update_and_unbind_ctx(bio);
1171		if (bio->bi_private)
1172			mempool_free(bio->bi_private, bio_post_read_ctx_pool);
1173		bio_put(bio);
1174		return -EFAULT;
1175	}
1176	inc_page_count(sbi, F2FS_RD_DATA);
1177	f2fs_update_iostat(sbi, NULL, FS_DATA_READ_IO, F2FS_BLKSIZE);
1178	f2fs_submit_read_bio(sbi, bio, DATA);
1179	return 0;
1180}
1181
1182static void __set_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1183{
1184	__le32 *addr = get_dnode_addr(dn->inode, dn->node_page);
1185
1186	dn->data_blkaddr = blkaddr;
1187	addr[dn->ofs_in_node] = cpu_to_le32(dn->data_blkaddr);
1188}
1189
1190/*
1191 * Lock ordering for the change of data block address:
1192 * ->data_page
1193 *  ->node_page
1194 *    update block addresses in the node page
1195 */
1196void f2fs_set_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1197{
1198	f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1199	__set_data_blkaddr(dn, blkaddr);
1200	if (set_page_dirty(dn->node_page))
1201		dn->node_changed = true;
1202}
1203
1204void f2fs_update_data_blkaddr(struct dnode_of_data *dn, block_t blkaddr)
1205{
1206	f2fs_set_data_blkaddr(dn, blkaddr);
1207	f2fs_update_read_extent_cache(dn);
1208}
1209
1210/* dn->ofs_in_node will be returned with up-to-date last block pointer */
1211int f2fs_reserve_new_blocks(struct dnode_of_data *dn, blkcnt_t count)
1212{
1213	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1214	int err;
1215
1216	if (!count)
1217		return 0;
1218
1219	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1220		return -EPERM;
1221	if (unlikely((err = inc_valid_block_count(sbi, dn->inode, &count))))
1222		return err;
1223
1224	trace_f2fs_reserve_new_blocks(dn->inode, dn->nid,
1225						dn->ofs_in_node, count);
1226
1227	f2fs_wait_on_page_writeback(dn->node_page, NODE, true, true);
1228
1229	for (; count > 0; dn->ofs_in_node++) {
1230		block_t blkaddr = f2fs_data_blkaddr(dn);
1231
1232		if (blkaddr == NULL_ADDR) {
1233			__set_data_blkaddr(dn, NEW_ADDR);
1234			count--;
1235		}
1236	}
1237
1238	if (set_page_dirty(dn->node_page))
1239		dn->node_changed = true;
1240	return 0;
1241}
1242
1243/* Should keep dn->ofs_in_node unchanged */
1244int f2fs_reserve_new_block(struct dnode_of_data *dn)
1245{
1246	unsigned int ofs_in_node = dn->ofs_in_node;
1247	int ret;
1248
1249	ret = f2fs_reserve_new_blocks(dn, 1);
1250	dn->ofs_in_node = ofs_in_node;
1251	return ret;
1252}
1253
1254int f2fs_reserve_block(struct dnode_of_data *dn, pgoff_t index)
1255{
1256	bool need_put = dn->inode_page ? false : true;
1257	int err;
1258
1259	err = f2fs_get_dnode_of_data(dn, index, ALLOC_NODE);
1260	if (err)
1261		return err;
1262
1263	if (dn->data_blkaddr == NULL_ADDR)
1264		err = f2fs_reserve_new_block(dn);
1265	if (err || need_put)
1266		f2fs_put_dnode(dn);
1267	return err;
1268}
1269
1270struct page *f2fs_get_read_data_page(struct inode *inode, pgoff_t index,
1271				     blk_opf_t op_flags, bool for_write,
1272				     pgoff_t *next_pgofs)
1273{
1274	struct address_space *mapping = inode->i_mapping;
1275	struct dnode_of_data dn;
1276	struct page *page;
1277	int err;
1278
1279	page = f2fs_grab_cache_page(mapping, index, for_write);
1280	if (!page)
1281		return ERR_PTR(-ENOMEM);
1282
1283	if (f2fs_lookup_read_extent_cache_block(inode, index,
1284						&dn.data_blkaddr)) {
1285		if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), dn.data_blkaddr,
1286						DATA_GENERIC_ENHANCE_READ)) {
1287			err = -EFSCORRUPTED;
1288			f2fs_handle_error(F2FS_I_SB(inode),
1289						ERROR_INVALID_BLKADDR);
1290			goto put_err;
1291		}
1292		goto got_it;
1293	}
1294
1295	set_new_dnode(&dn, inode, NULL, NULL, 0);
1296	err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
1297	if (err) {
1298		if (err == -ENOENT && next_pgofs)
1299			*next_pgofs = f2fs_get_next_page_offset(&dn, index);
1300		goto put_err;
1301	}
1302	f2fs_put_dnode(&dn);
1303
1304	if (unlikely(dn.data_blkaddr == NULL_ADDR)) {
1305		err = -ENOENT;
1306		if (next_pgofs)
1307			*next_pgofs = index + 1;
1308		goto put_err;
1309	}
1310	if (dn.data_blkaddr != NEW_ADDR &&
1311			!f2fs_is_valid_blkaddr(F2FS_I_SB(inode),
1312						dn.data_blkaddr,
1313						DATA_GENERIC_ENHANCE)) {
1314		err = -EFSCORRUPTED;
1315		f2fs_handle_error(F2FS_I_SB(inode),
1316					ERROR_INVALID_BLKADDR);
1317		goto put_err;
1318	}
1319got_it:
1320	if (PageUptodate(page)) {
1321		unlock_page(page);
1322		return page;
1323	}
1324
1325	/*
1326	 * A new dentry page is allocated but not able to be written, since its
1327	 * new inode page couldn't be allocated due to -ENOSPC.
1328	 * In such the case, its blkaddr can be remained as NEW_ADDR.
1329	 * see, f2fs_add_link -> f2fs_get_new_data_page ->
1330	 * f2fs_init_inode_metadata.
1331	 */
1332	if (dn.data_blkaddr == NEW_ADDR) {
1333		zero_user_segment(page, 0, PAGE_SIZE);
1334		if (!PageUptodate(page))
1335			SetPageUptodate(page);
1336		unlock_page(page);
1337		return page;
1338	}
1339
1340	err = f2fs_submit_page_read(inode, page, dn.data_blkaddr,
1341						op_flags, for_write);
1342	if (err)
1343		goto put_err;
1344	return page;
1345
1346put_err:
1347	f2fs_put_page(page, 1);
1348	return ERR_PTR(err);
1349}
1350
1351struct page *f2fs_find_data_page(struct inode *inode, pgoff_t index,
1352					pgoff_t *next_pgofs)
1353{
1354	struct address_space *mapping = inode->i_mapping;
1355	struct page *page;
1356
1357	page = find_get_page(mapping, index);
1358	if (page && PageUptodate(page))
1359		return page;
1360	f2fs_put_page(page, 0);
1361
1362	page = f2fs_get_read_data_page(inode, index, 0, false, next_pgofs);
1363	if (IS_ERR(page))
1364		return page;
1365
1366	if (PageUptodate(page))
1367		return page;
1368
1369	wait_on_page_locked(page);
1370	if (unlikely(!PageUptodate(page))) {
1371		f2fs_put_page(page, 0);
1372		return ERR_PTR(-EIO);
1373	}
1374	return page;
1375}
1376
1377/*
1378 * If it tries to access a hole, return an error.
1379 * Because, the callers, functions in dir.c and GC, should be able to know
1380 * whether this page exists or not.
1381 */
1382struct page *f2fs_get_lock_data_page(struct inode *inode, pgoff_t index,
1383							bool for_write)
1384{
1385	struct address_space *mapping = inode->i_mapping;
1386	struct page *page;
1387
1388	page = f2fs_get_read_data_page(inode, index, 0, for_write, NULL);
1389	if (IS_ERR(page))
1390		return page;
1391
1392	/* wait for read completion */
1393	lock_page(page);
1394	if (unlikely(page->mapping != mapping || !PageUptodate(page))) {
1395		f2fs_put_page(page, 1);
1396		return ERR_PTR(-EIO);
1397	}
1398	return page;
1399}
1400
1401/*
1402 * Caller ensures that this data page is never allocated.
1403 * A new zero-filled data page is allocated in the page cache.
1404 *
1405 * Also, caller should grab and release a rwsem by calling f2fs_lock_op() and
1406 * f2fs_unlock_op().
1407 * Note that, ipage is set only by make_empty_dir, and if any error occur,
1408 * ipage should be released by this function.
1409 */
1410struct page *f2fs_get_new_data_page(struct inode *inode,
1411		struct page *ipage, pgoff_t index, bool new_i_size)
1412{
1413	struct address_space *mapping = inode->i_mapping;
1414	struct page *page;
1415	struct dnode_of_data dn;
1416	int err;
1417
1418	page = f2fs_grab_cache_page(mapping, index, true);
1419	if (!page) {
1420		/*
1421		 * before exiting, we should make sure ipage will be released
1422		 * if any error occur.
1423		 */
1424		f2fs_put_page(ipage, 1);
1425		return ERR_PTR(-ENOMEM);
1426	}
1427
1428	set_new_dnode(&dn, inode, ipage, NULL, 0);
1429	err = f2fs_reserve_block(&dn, index);
1430	if (err) {
1431		f2fs_put_page(page, 1);
1432		return ERR_PTR(err);
1433	}
1434	if (!ipage)
1435		f2fs_put_dnode(&dn);
1436
1437	if (PageUptodate(page))
1438		goto got_it;
1439
1440	if (dn.data_blkaddr == NEW_ADDR) {
1441		zero_user_segment(page, 0, PAGE_SIZE);
1442		if (!PageUptodate(page))
1443			SetPageUptodate(page);
1444	} else {
1445		f2fs_put_page(page, 1);
1446
1447		/* if ipage exists, blkaddr should be NEW_ADDR */
1448		f2fs_bug_on(F2FS_I_SB(inode), ipage);
1449		page = f2fs_get_lock_data_page(inode, index, true);
1450		if (IS_ERR(page))
1451			return page;
1452	}
1453got_it:
1454	if (new_i_size && i_size_read(inode) <
1455				((loff_t)(index + 1) << PAGE_SHIFT))
1456		f2fs_i_size_write(inode, ((loff_t)(index + 1) << PAGE_SHIFT));
1457	return page;
1458}
1459
1460static int __allocate_data_block(struct dnode_of_data *dn, int seg_type)
1461{
1462	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1463	struct f2fs_summary sum;
1464	struct node_info ni;
1465	block_t old_blkaddr;
1466	blkcnt_t count = 1;
1467	int err;
1468
1469	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1470		return -EPERM;
1471
1472	err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
1473	if (err)
1474		return err;
1475
1476	dn->data_blkaddr = f2fs_data_blkaddr(dn);
1477	if (dn->data_blkaddr == NULL_ADDR) {
1478		err = inc_valid_block_count(sbi, dn->inode, &count);
1479		if (unlikely(err))
1480			return err;
1481	}
1482
1483	set_summary(&sum, dn->nid, dn->ofs_in_node, ni.version);
1484	old_blkaddr = dn->data_blkaddr;
1485	f2fs_allocate_data_block(sbi, NULL, old_blkaddr, &dn->data_blkaddr,
1486				&sum, seg_type, NULL);
1487	if (GET_SEGNO(sbi, old_blkaddr) != NULL_SEGNO)
1488		f2fs_invalidate_internal_cache(sbi, old_blkaddr);
1489
1490	f2fs_update_data_blkaddr(dn, dn->data_blkaddr);
1491	return 0;
1492}
1493
1494static void f2fs_map_lock(struct f2fs_sb_info *sbi, int flag)
1495{
1496	if (flag == F2FS_GET_BLOCK_PRE_AIO)
1497		f2fs_down_read(&sbi->node_change);
1498	else
1499		f2fs_lock_op(sbi);
1500}
1501
1502static void f2fs_map_unlock(struct f2fs_sb_info *sbi, int flag)
1503{
1504	if (flag == F2FS_GET_BLOCK_PRE_AIO)
1505		f2fs_up_read(&sbi->node_change);
1506	else
1507		f2fs_unlock_op(sbi);
1508}
1509
1510int f2fs_get_block_locked(struct dnode_of_data *dn, pgoff_t index)
1511{
1512	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1513	int err = 0;
1514
1515	f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO);
1516	if (!f2fs_lookup_read_extent_cache_block(dn->inode, index,
1517						&dn->data_blkaddr))
1518		err = f2fs_reserve_block(dn, index);
1519	f2fs_map_unlock(sbi, F2FS_GET_BLOCK_PRE_AIO);
1520
1521	return err;
1522}
1523
1524static int f2fs_map_no_dnode(struct inode *inode,
1525		struct f2fs_map_blocks *map, struct dnode_of_data *dn,
1526		pgoff_t pgoff)
1527{
1528	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1529
1530	/*
1531	 * There is one exceptional case that read_node_page() may return
1532	 * -ENOENT due to filesystem has been shutdown or cp_error, return
1533	 * -EIO in that case.
1534	 */
1535	if (map->m_may_create &&
1536	    (is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN) || f2fs_cp_error(sbi)))
1537		return -EIO;
1538
1539	if (map->m_next_pgofs)
1540		*map->m_next_pgofs = f2fs_get_next_page_offset(dn, pgoff);
1541	if (map->m_next_extent)
1542		*map->m_next_extent = f2fs_get_next_page_offset(dn, pgoff);
1543	return 0;
1544}
1545
1546static bool f2fs_map_blocks_cached(struct inode *inode,
1547		struct f2fs_map_blocks *map, int flag)
1548{
1549	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1550	unsigned int maxblocks = map->m_len;
1551	pgoff_t pgoff = (pgoff_t)map->m_lblk;
1552	struct extent_info ei = {};
1553
1554	if (!f2fs_lookup_read_extent_cache(inode, pgoff, &ei))
1555		return false;
1556
1557	map->m_pblk = ei.blk + pgoff - ei.fofs;
1558	map->m_len = min((pgoff_t)maxblocks, ei.fofs + ei.len - pgoff);
1559	map->m_flags = F2FS_MAP_MAPPED;
1560	if (map->m_next_extent)
1561		*map->m_next_extent = pgoff + map->m_len;
1562
1563	/* for hardware encryption, but to avoid potential issue in future */
1564	if (flag == F2FS_GET_BLOCK_DIO)
1565		f2fs_wait_on_block_writeback_range(inode,
1566					map->m_pblk, map->m_len);
1567
1568	if (f2fs_allow_multi_device_dio(sbi, flag)) {
1569		int bidx = f2fs_target_device_index(sbi, map->m_pblk);
1570		struct f2fs_dev_info *dev = &sbi->devs[bidx];
1571
1572		map->m_bdev = dev->bdev;
1573		map->m_pblk -= dev->start_blk;
1574		map->m_len = min(map->m_len, dev->end_blk + 1 - map->m_pblk);
1575	} else {
1576		map->m_bdev = inode->i_sb->s_bdev;
1577	}
1578	return true;
1579}
1580
1581/*
1582 * f2fs_map_blocks() tries to find or build mapping relationship which
1583 * maps continuous logical blocks to physical blocks, and return such
1584 * info via f2fs_map_blocks structure.
1585 */
1586int f2fs_map_blocks(struct inode *inode, struct f2fs_map_blocks *map, int flag)
1587{
1588	unsigned int maxblocks = map->m_len;
1589	struct dnode_of_data dn;
1590	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1591	int mode = map->m_may_create ? ALLOC_NODE : LOOKUP_NODE;
1592	pgoff_t pgofs, end_offset, end;
1593	int err = 0, ofs = 1;
1594	unsigned int ofs_in_node, last_ofs_in_node;
1595	blkcnt_t prealloc;
1596	block_t blkaddr;
1597	unsigned int start_pgofs;
1598	int bidx = 0;
1599	bool is_hole;
1600
1601	if (!maxblocks)
1602		return 0;
1603
1604	if (!map->m_may_create && f2fs_map_blocks_cached(inode, map, flag))
1605		goto out;
1606
1607	map->m_bdev = inode->i_sb->s_bdev;
1608	map->m_multidev_dio =
1609		f2fs_allow_multi_device_dio(F2FS_I_SB(inode), flag);
1610
1611	map->m_len = 0;
1612	map->m_flags = 0;
1613
1614	/* it only supports block size == page size */
1615	pgofs =	(pgoff_t)map->m_lblk;
1616	end = pgofs + maxblocks;
1617
1618next_dnode:
1619	if (map->m_may_create)
1620		f2fs_map_lock(sbi, flag);
1621
1622	/* When reading holes, we need its node page */
1623	set_new_dnode(&dn, inode, NULL, NULL, 0);
1624	err = f2fs_get_dnode_of_data(&dn, pgofs, mode);
1625	if (err) {
1626		if (flag == F2FS_GET_BLOCK_BMAP)
1627			map->m_pblk = 0;
1628		if (err == -ENOENT)
1629			err = f2fs_map_no_dnode(inode, map, &dn, pgofs);
1630		goto unlock_out;
1631	}
1632
1633	start_pgofs = pgofs;
1634	prealloc = 0;
1635	last_ofs_in_node = ofs_in_node = dn.ofs_in_node;
1636	end_offset = ADDRS_PER_PAGE(dn.node_page, inode);
1637
1638next_block:
1639	blkaddr = f2fs_data_blkaddr(&dn);
1640	is_hole = !__is_valid_data_blkaddr(blkaddr);
1641	if (!is_hole &&
1642	    !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE)) {
1643		err = -EFSCORRUPTED;
1644		f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1645		goto sync_out;
1646	}
1647
1648	/* use out-place-update for direct IO under LFS mode */
1649	if (map->m_may_create &&
1650	    (is_hole || (f2fs_lfs_mode(sbi) && flag == F2FS_GET_BLOCK_DIO))) {
1651		if (unlikely(f2fs_cp_error(sbi))) {
1652			err = -EIO;
1653			goto sync_out;
1654		}
1655
1656		switch (flag) {
1657		case F2FS_GET_BLOCK_PRE_AIO:
1658			if (blkaddr == NULL_ADDR) {
1659				prealloc++;
1660				last_ofs_in_node = dn.ofs_in_node;
1661			}
1662			break;
1663		case F2FS_GET_BLOCK_PRE_DIO:
1664		case F2FS_GET_BLOCK_DIO:
1665			err = __allocate_data_block(&dn, map->m_seg_type);
1666			if (err)
1667				goto sync_out;
1668			if (flag == F2FS_GET_BLOCK_PRE_DIO)
1669				file_need_truncate(inode);
1670			set_inode_flag(inode, FI_APPEND_WRITE);
1671			break;
1672		default:
1673			WARN_ON_ONCE(1);
1674			err = -EIO;
1675			goto sync_out;
1676		}
1677
1678		blkaddr = dn.data_blkaddr;
1679		if (is_hole)
1680			map->m_flags |= F2FS_MAP_NEW;
1681	} else if (is_hole) {
1682		if (f2fs_compressed_file(inode) &&
1683		    f2fs_sanity_check_cluster(&dn)) {
1684			err = -EFSCORRUPTED;
1685			f2fs_handle_error(sbi,
1686					ERROR_CORRUPTED_CLUSTER);
1687			goto sync_out;
1688		}
1689
1690		switch (flag) {
1691		case F2FS_GET_BLOCK_PRECACHE:
1692			goto sync_out;
1693		case F2FS_GET_BLOCK_BMAP:
1694			map->m_pblk = 0;
1695			goto sync_out;
1696		case F2FS_GET_BLOCK_FIEMAP:
1697			if (blkaddr == NULL_ADDR) {
1698				if (map->m_next_pgofs)
1699					*map->m_next_pgofs = pgofs + 1;
1700				goto sync_out;
1701			}
1702			break;
1703		default:
1704			/* for defragment case */
1705			if (map->m_next_pgofs)
1706				*map->m_next_pgofs = pgofs + 1;
1707			goto sync_out;
1708		}
1709	}
1710
1711	if (flag == F2FS_GET_BLOCK_PRE_AIO)
1712		goto skip;
1713
1714	if (map->m_multidev_dio)
1715		bidx = f2fs_target_device_index(sbi, blkaddr);
1716
1717	if (map->m_len == 0) {
1718		/* reserved delalloc block should be mapped for fiemap. */
1719		if (blkaddr == NEW_ADDR)
1720			map->m_flags |= F2FS_MAP_DELALLOC;
1721		map->m_flags |= F2FS_MAP_MAPPED;
1722
1723		map->m_pblk = blkaddr;
1724		map->m_len = 1;
1725
1726		if (map->m_multidev_dio)
1727			map->m_bdev = FDEV(bidx).bdev;
1728	} else if ((map->m_pblk != NEW_ADDR &&
1729			blkaddr == (map->m_pblk + ofs)) ||
1730			(map->m_pblk == NEW_ADDR && blkaddr == NEW_ADDR) ||
1731			flag == F2FS_GET_BLOCK_PRE_DIO) {
1732		if (map->m_multidev_dio && map->m_bdev != FDEV(bidx).bdev)
1733			goto sync_out;
1734		ofs++;
1735		map->m_len++;
1736	} else {
1737		goto sync_out;
1738	}
1739
1740skip:
1741	dn.ofs_in_node++;
1742	pgofs++;
1743
1744	/* preallocate blocks in batch for one dnode page */
1745	if (flag == F2FS_GET_BLOCK_PRE_AIO &&
1746			(pgofs == end || dn.ofs_in_node == end_offset)) {
1747
1748		dn.ofs_in_node = ofs_in_node;
1749		err = f2fs_reserve_new_blocks(&dn, prealloc);
1750		if (err)
1751			goto sync_out;
1752
1753		map->m_len += dn.ofs_in_node - ofs_in_node;
1754		if (prealloc && dn.ofs_in_node != last_ofs_in_node + 1) {
1755			err = -ENOSPC;
1756			goto sync_out;
1757		}
1758		dn.ofs_in_node = end_offset;
1759	}
1760
1761	if (pgofs >= end)
1762		goto sync_out;
1763	else if (dn.ofs_in_node < end_offset)
1764		goto next_block;
1765
1766	if (flag == F2FS_GET_BLOCK_PRECACHE) {
1767		if (map->m_flags & F2FS_MAP_MAPPED) {
1768			unsigned int ofs = start_pgofs - map->m_lblk;
1769
1770			f2fs_update_read_extent_cache_range(&dn,
1771				start_pgofs, map->m_pblk + ofs,
1772				map->m_len - ofs);
1773		}
1774	}
1775
1776	f2fs_put_dnode(&dn);
1777
1778	if (map->m_may_create) {
1779		f2fs_map_unlock(sbi, flag);
1780		f2fs_balance_fs(sbi, dn.node_changed);
1781	}
1782	goto next_dnode;
1783
1784sync_out:
1785
1786	if (flag == F2FS_GET_BLOCK_DIO && map->m_flags & F2FS_MAP_MAPPED) {
1787		/*
1788		 * for hardware encryption, but to avoid potential issue
1789		 * in future
1790		 */
1791		f2fs_wait_on_block_writeback_range(inode,
1792						map->m_pblk, map->m_len);
1793
1794		if (map->m_multidev_dio) {
1795			block_t blk_addr = map->m_pblk;
1796
1797			bidx = f2fs_target_device_index(sbi, map->m_pblk);
1798
1799			map->m_bdev = FDEV(bidx).bdev;
1800			map->m_pblk -= FDEV(bidx).start_blk;
1801
1802			if (map->m_may_create)
1803				f2fs_update_device_state(sbi, inode->i_ino,
1804							blk_addr, map->m_len);
1805
1806			f2fs_bug_on(sbi, blk_addr + map->m_len >
1807						FDEV(bidx).end_blk + 1);
1808		}
1809	}
1810
1811	if (flag == F2FS_GET_BLOCK_PRECACHE) {
1812		if (map->m_flags & F2FS_MAP_MAPPED) {
1813			unsigned int ofs = start_pgofs - map->m_lblk;
1814
1815			f2fs_update_read_extent_cache_range(&dn,
1816				start_pgofs, map->m_pblk + ofs,
1817				map->m_len - ofs);
1818		}
1819		if (map->m_next_extent)
1820			*map->m_next_extent = pgofs + 1;
1821	}
1822	f2fs_put_dnode(&dn);
1823unlock_out:
1824	if (map->m_may_create) {
1825		f2fs_map_unlock(sbi, flag);
1826		f2fs_balance_fs(sbi, dn.node_changed);
1827	}
1828out:
1829	trace_f2fs_map_blocks(inode, map, flag, err);
1830	return err;
1831}
1832
1833bool f2fs_overwrite_io(struct inode *inode, loff_t pos, size_t len)
1834{
1835	struct f2fs_map_blocks map;
1836	block_t last_lblk;
1837	int err;
1838
1839	if (pos + len > i_size_read(inode))
1840		return false;
1841
1842	map.m_lblk = F2FS_BYTES_TO_BLK(pos);
1843	map.m_next_pgofs = NULL;
1844	map.m_next_extent = NULL;
1845	map.m_seg_type = NO_CHECK_TYPE;
1846	map.m_may_create = false;
1847	last_lblk = F2FS_BLK_ALIGN(pos + len);
1848
1849	while (map.m_lblk < last_lblk) {
1850		map.m_len = last_lblk - map.m_lblk;
1851		err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_DEFAULT);
1852		if (err || map.m_len == 0)
1853			return false;
1854		map.m_lblk += map.m_len;
1855	}
1856	return true;
1857}
1858
1859static inline u64 bytes_to_blks(struct inode *inode, u64 bytes)
1860{
1861	return (bytes >> inode->i_blkbits);
1862}
1863
1864static inline u64 blks_to_bytes(struct inode *inode, u64 blks)
1865{
1866	return (blks << inode->i_blkbits);
1867}
1868
1869static int f2fs_xattr_fiemap(struct inode *inode,
1870				struct fiemap_extent_info *fieinfo)
1871{
1872	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1873	struct page *page;
1874	struct node_info ni;
1875	__u64 phys = 0, len;
1876	__u32 flags;
1877	nid_t xnid = F2FS_I(inode)->i_xattr_nid;
1878	int err = 0;
1879
1880	if (f2fs_has_inline_xattr(inode)) {
1881		int offset;
1882
1883		page = f2fs_grab_cache_page(NODE_MAPPING(sbi),
1884						inode->i_ino, false);
1885		if (!page)
1886			return -ENOMEM;
1887
1888		err = f2fs_get_node_info(sbi, inode->i_ino, &ni, false);
1889		if (err) {
1890			f2fs_put_page(page, 1);
1891			return err;
1892		}
1893
1894		phys = blks_to_bytes(inode, ni.blk_addr);
1895		offset = offsetof(struct f2fs_inode, i_addr) +
1896					sizeof(__le32) * (DEF_ADDRS_PER_INODE -
1897					get_inline_xattr_addrs(inode));
1898
1899		phys += offset;
1900		len = inline_xattr_size(inode);
1901
1902		f2fs_put_page(page, 1);
1903
1904		flags = FIEMAP_EXTENT_DATA_INLINE | FIEMAP_EXTENT_NOT_ALIGNED;
1905
1906		if (!xnid)
1907			flags |= FIEMAP_EXTENT_LAST;
1908
1909		err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1910		trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1911		if (err)
1912			return err;
1913	}
1914
1915	if (xnid) {
1916		page = f2fs_grab_cache_page(NODE_MAPPING(sbi), xnid, false);
1917		if (!page)
1918			return -ENOMEM;
1919
1920		err = f2fs_get_node_info(sbi, xnid, &ni, false);
1921		if (err) {
1922			f2fs_put_page(page, 1);
1923			return err;
1924		}
1925
1926		phys = blks_to_bytes(inode, ni.blk_addr);
1927		len = inode->i_sb->s_blocksize;
1928
1929		f2fs_put_page(page, 1);
1930
1931		flags = FIEMAP_EXTENT_LAST;
1932	}
1933
1934	if (phys) {
1935		err = fiemap_fill_next_extent(fieinfo, 0, phys, len, flags);
1936		trace_f2fs_fiemap(inode, 0, phys, len, flags, err);
1937	}
1938
1939	return (err < 0 ? err : 0);
1940}
1941
1942static loff_t max_inode_blocks(struct inode *inode)
1943{
1944	loff_t result = ADDRS_PER_INODE(inode);
1945	loff_t leaf_count = ADDRS_PER_BLOCK(inode);
1946
1947	/* two direct node blocks */
1948	result += (leaf_count * 2);
1949
1950	/* two indirect node blocks */
1951	leaf_count *= NIDS_PER_BLOCK;
1952	result += (leaf_count * 2);
1953
1954	/* one double indirect node block */
1955	leaf_count *= NIDS_PER_BLOCK;
1956	result += leaf_count;
1957
1958	return result;
1959}
1960
1961int f2fs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
1962		u64 start, u64 len)
1963{
1964	struct f2fs_map_blocks map;
1965	sector_t start_blk, last_blk;
1966	pgoff_t next_pgofs;
1967	u64 logical = 0, phys = 0, size = 0;
1968	u32 flags = 0;
1969	int ret = 0;
1970	bool compr_cluster = false, compr_appended;
1971	unsigned int cluster_size = F2FS_I(inode)->i_cluster_size;
1972	unsigned int count_in_cluster = 0;
1973	loff_t maxbytes;
1974
1975	if (fieinfo->fi_flags & FIEMAP_FLAG_CACHE) {
1976		ret = f2fs_precache_extents(inode);
1977		if (ret)
1978			return ret;
1979	}
1980
1981	ret = fiemap_prep(inode, fieinfo, start, &len, FIEMAP_FLAG_XATTR);
1982	if (ret)
1983		return ret;
1984
1985	inode_lock_shared(inode);
1986
1987	maxbytes = max_file_blocks(inode) << F2FS_BLKSIZE_BITS;
1988	if (start > maxbytes) {
1989		ret = -EFBIG;
1990		goto out;
1991	}
1992
1993	if (len > maxbytes || (maxbytes - len) < start)
1994		len = maxbytes - start;
1995
1996	if (fieinfo->fi_flags & FIEMAP_FLAG_XATTR) {
1997		ret = f2fs_xattr_fiemap(inode, fieinfo);
1998		goto out;
1999	}
2000
2001	if (f2fs_has_inline_data(inode) || f2fs_has_inline_dentry(inode)) {
2002		ret = f2fs_inline_data_fiemap(inode, fieinfo, start, len);
2003		if (ret != -EAGAIN)
2004			goto out;
2005	}
2006
2007	if (bytes_to_blks(inode, len) == 0)
2008		len = blks_to_bytes(inode, 1);
2009
2010	start_blk = bytes_to_blks(inode, start);
2011	last_blk = bytes_to_blks(inode, start + len - 1);
2012
2013next:
2014	memset(&map, 0, sizeof(map));
2015	map.m_lblk = start_blk;
2016	map.m_len = bytes_to_blks(inode, len);
2017	map.m_next_pgofs = &next_pgofs;
2018	map.m_seg_type = NO_CHECK_TYPE;
2019
2020	if (compr_cluster) {
2021		map.m_lblk += 1;
2022		map.m_len = cluster_size - count_in_cluster;
2023	}
2024
2025	ret = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_FIEMAP);
2026	if (ret)
2027		goto out;
2028
2029	/* HOLE */
2030	if (!compr_cluster && !(map.m_flags & F2FS_MAP_FLAGS)) {
2031		start_blk = next_pgofs;
2032
2033		if (blks_to_bytes(inode, start_blk) < blks_to_bytes(inode,
2034						max_inode_blocks(inode)))
2035			goto prep_next;
2036
2037		flags |= FIEMAP_EXTENT_LAST;
2038	}
2039
2040	compr_appended = false;
2041	/* In a case of compressed cluster, append this to the last extent */
2042	if (compr_cluster && ((map.m_flags & F2FS_MAP_DELALLOC) ||
2043			!(map.m_flags & F2FS_MAP_FLAGS))) {
2044		compr_appended = true;
2045		goto skip_fill;
2046	}
2047
2048	if (size) {
2049		flags |= FIEMAP_EXTENT_MERGED;
2050		if (IS_ENCRYPTED(inode))
2051			flags |= FIEMAP_EXTENT_DATA_ENCRYPTED;
2052
2053		ret = fiemap_fill_next_extent(fieinfo, logical,
2054				phys, size, flags);
2055		trace_f2fs_fiemap(inode, logical, phys, size, flags, ret);
2056		if (ret)
2057			goto out;
2058		size = 0;
2059	}
2060
2061	if (start_blk > last_blk)
2062		goto out;
2063
2064skip_fill:
2065	if (map.m_pblk == COMPRESS_ADDR) {
2066		compr_cluster = true;
2067		count_in_cluster = 1;
2068	} else if (compr_appended) {
2069		unsigned int appended_blks = cluster_size -
2070						count_in_cluster + 1;
2071		size += blks_to_bytes(inode, appended_blks);
2072		start_blk += appended_blks;
2073		compr_cluster = false;
2074	} else {
2075		logical = blks_to_bytes(inode, start_blk);
2076		phys = __is_valid_data_blkaddr(map.m_pblk) ?
2077			blks_to_bytes(inode, map.m_pblk) : 0;
2078		size = blks_to_bytes(inode, map.m_len);
2079		flags = 0;
2080
2081		if (compr_cluster) {
2082			flags = FIEMAP_EXTENT_ENCODED;
2083			count_in_cluster += map.m_len;
2084			if (count_in_cluster == cluster_size) {
2085				compr_cluster = false;
2086				size += blks_to_bytes(inode, 1);
2087			}
2088		} else if (map.m_flags & F2FS_MAP_DELALLOC) {
2089			flags = FIEMAP_EXTENT_UNWRITTEN;
2090		}
2091
2092		start_blk += bytes_to_blks(inode, size);
2093	}
2094
2095prep_next:
2096	cond_resched();
2097	if (fatal_signal_pending(current))
2098		ret = -EINTR;
2099	else
2100		goto next;
2101out:
2102	if (ret == 1)
2103		ret = 0;
2104
2105	inode_unlock_shared(inode);
2106	return ret;
2107}
2108
2109static inline loff_t f2fs_readpage_limit(struct inode *inode)
2110{
2111	if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode))
2112		return inode->i_sb->s_maxbytes;
2113
2114	return i_size_read(inode);
2115}
2116
2117static int f2fs_read_single_page(struct inode *inode, struct page *page,
2118					unsigned nr_pages,
2119					struct f2fs_map_blocks *map,
2120					struct bio **bio_ret,
2121					sector_t *last_block_in_bio,
2122					bool is_readahead)
2123{
2124	struct bio *bio = *bio_ret;
2125	const unsigned blocksize = blks_to_bytes(inode, 1);
2126	sector_t block_in_file;
2127	sector_t last_block;
2128	sector_t last_block_in_file;
2129	sector_t block_nr;
2130	int ret = 0;
2131
2132	block_in_file = (sector_t)page_index(page);
2133	last_block = block_in_file + nr_pages;
2134	last_block_in_file = bytes_to_blks(inode,
2135			f2fs_readpage_limit(inode) + blocksize - 1);
2136	if (last_block > last_block_in_file)
2137		last_block = last_block_in_file;
2138
2139	/* just zeroing out page which is beyond EOF */
2140	if (block_in_file >= last_block)
2141		goto zero_out;
2142	/*
2143	 * Map blocks using the previous result first.
2144	 */
2145	if ((map->m_flags & F2FS_MAP_MAPPED) &&
2146			block_in_file > map->m_lblk &&
2147			block_in_file < (map->m_lblk + map->m_len))
2148		goto got_it;
2149
2150	/*
2151	 * Then do more f2fs_map_blocks() calls until we are
2152	 * done with this page.
2153	 */
2154	map->m_lblk = block_in_file;
2155	map->m_len = last_block - block_in_file;
2156
2157	ret = f2fs_map_blocks(inode, map, F2FS_GET_BLOCK_DEFAULT);
2158	if (ret)
2159		goto out;
2160got_it:
2161	if ((map->m_flags & F2FS_MAP_MAPPED)) {
2162		block_nr = map->m_pblk + block_in_file - map->m_lblk;
2163		SetPageMappedToDisk(page);
2164
2165		if (!f2fs_is_valid_blkaddr(F2FS_I_SB(inode), block_nr,
2166						DATA_GENERIC_ENHANCE_READ)) {
2167			ret = -EFSCORRUPTED;
2168			f2fs_handle_error(F2FS_I_SB(inode),
2169						ERROR_INVALID_BLKADDR);
2170			goto out;
2171		}
2172	} else {
2173zero_out:
2174		zero_user_segment(page, 0, PAGE_SIZE);
2175		if (f2fs_need_verity(inode, page->index) &&
2176		    !fsverity_verify_page(page)) {
2177			ret = -EIO;
2178			goto out;
2179		}
2180		if (!PageUptodate(page))
2181			SetPageUptodate(page);
2182		unlock_page(page);
2183		goto out;
2184	}
2185
2186	/*
2187	 * This page will go to BIO.  Do we need to send this
2188	 * BIO off first?
2189	 */
2190	if (bio && (!page_is_mergeable(F2FS_I_SB(inode), bio,
2191				       *last_block_in_bio, block_nr) ||
2192		    !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2193submit_and_realloc:
2194		f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA);
2195		bio = NULL;
2196	}
2197	if (bio == NULL) {
2198		bio = f2fs_grab_read_bio(inode, block_nr, nr_pages,
2199				is_readahead ? REQ_RAHEAD : 0, page->index,
2200				false);
2201		if (IS_ERR(bio)) {
2202			ret = PTR_ERR(bio);
2203			bio = NULL;
2204			goto out;
2205		}
2206	}
2207
2208	/*
2209	 * If the page is under writeback, we need to wait for
2210	 * its completion to see the correct decrypted data.
2211	 */
2212	f2fs_wait_on_block_writeback(inode, block_nr);
2213
2214	if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2215		goto submit_and_realloc;
2216
2217	inc_page_count(F2FS_I_SB(inode), F2FS_RD_DATA);
2218	f2fs_update_iostat(F2FS_I_SB(inode), NULL, FS_DATA_READ_IO,
2219							F2FS_BLKSIZE);
2220	*last_block_in_bio = block_nr;
2221out:
2222	*bio_ret = bio;
2223	return ret;
2224}
2225
2226#ifdef CONFIG_F2FS_FS_COMPRESSION
2227int f2fs_read_multi_pages(struct compress_ctx *cc, struct bio **bio_ret,
2228				unsigned nr_pages, sector_t *last_block_in_bio,
2229				bool is_readahead, bool for_write)
2230{
2231	struct dnode_of_data dn;
2232	struct inode *inode = cc->inode;
2233	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2234	struct bio *bio = *bio_ret;
2235	unsigned int start_idx = cc->cluster_idx << cc->log_cluster_size;
2236	sector_t last_block_in_file;
2237	const unsigned blocksize = blks_to_bytes(inode, 1);
2238	struct decompress_io_ctx *dic = NULL;
2239	struct extent_info ei = {};
2240	bool from_dnode = true;
2241	int i;
2242	int ret = 0;
2243
2244	f2fs_bug_on(sbi, f2fs_cluster_is_empty(cc));
2245
2246	last_block_in_file = bytes_to_blks(inode,
2247			f2fs_readpage_limit(inode) + blocksize - 1);
2248
2249	/* get rid of pages beyond EOF */
2250	for (i = 0; i < cc->cluster_size; i++) {
2251		struct page *page = cc->rpages[i];
2252
2253		if (!page)
2254			continue;
2255		if ((sector_t)page->index >= last_block_in_file) {
2256			zero_user_segment(page, 0, PAGE_SIZE);
2257			if (!PageUptodate(page))
2258				SetPageUptodate(page);
2259		} else if (!PageUptodate(page)) {
2260			continue;
2261		}
2262		unlock_page(page);
2263		if (for_write)
2264			put_page(page);
2265		cc->rpages[i] = NULL;
2266		cc->nr_rpages--;
2267	}
2268
2269	/* we are done since all pages are beyond EOF */
2270	if (f2fs_cluster_is_empty(cc))
2271		goto out;
2272
2273	if (f2fs_lookup_read_extent_cache(inode, start_idx, &ei))
2274		from_dnode = false;
2275
2276	if (!from_dnode)
2277		goto skip_reading_dnode;
2278
2279	set_new_dnode(&dn, inode, NULL, NULL, 0);
2280	ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
2281	if (ret)
2282		goto out;
2283
2284	if (unlikely(f2fs_cp_error(sbi))) {
2285		ret = -EIO;
2286		goto out_put_dnode;
2287	}
2288	f2fs_bug_on(sbi, dn.data_blkaddr != COMPRESS_ADDR);
2289
2290skip_reading_dnode:
2291	for (i = 1; i < cc->cluster_size; i++) {
2292		block_t blkaddr;
2293
2294		blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2295					dn.ofs_in_node + i) :
2296					ei.blk + i - 1;
2297
2298		if (!__is_valid_data_blkaddr(blkaddr))
2299			break;
2300
2301		if (!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC)) {
2302			ret = -EFAULT;
2303			goto out_put_dnode;
2304		}
2305		cc->nr_cpages++;
2306
2307		if (!from_dnode && i >= ei.c_len)
2308			break;
2309	}
2310
2311	/* nothing to decompress */
2312	if (cc->nr_cpages == 0) {
2313		ret = 0;
2314		goto out_put_dnode;
2315	}
2316
2317	dic = f2fs_alloc_dic(cc);
2318	if (IS_ERR(dic)) {
2319		ret = PTR_ERR(dic);
2320		goto out_put_dnode;
2321	}
2322
2323	for (i = 0; i < cc->nr_cpages; i++) {
2324		struct page *page = dic->cpages[i];
2325		block_t blkaddr;
2326		struct bio_post_read_ctx *ctx;
2327
2328		blkaddr = from_dnode ? data_blkaddr(dn.inode, dn.node_page,
2329					dn.ofs_in_node + i + 1) :
2330					ei.blk + i;
2331
2332		f2fs_wait_on_block_writeback(inode, blkaddr);
2333
2334		if (f2fs_load_compressed_page(sbi, page, blkaddr)) {
2335			if (atomic_dec_and_test(&dic->remaining_pages)) {
2336				f2fs_decompress_cluster(dic, true);
2337				break;
2338			}
2339			continue;
2340		}
2341
2342		if (bio && (!page_is_mergeable(sbi, bio,
2343					*last_block_in_bio, blkaddr) ||
2344		    !f2fs_crypt_mergeable_bio(bio, inode, page->index, NULL))) {
2345submit_and_realloc:
2346			f2fs_submit_read_bio(sbi, bio, DATA);
2347			bio = NULL;
2348		}
2349
2350		if (!bio) {
2351			bio = f2fs_grab_read_bio(inode, blkaddr, nr_pages,
2352					is_readahead ? REQ_RAHEAD : 0,
2353					page->index, for_write);
2354			if (IS_ERR(bio)) {
2355				ret = PTR_ERR(bio);
2356				f2fs_decompress_end_io(dic, ret, true);
2357				f2fs_put_dnode(&dn);
2358				*bio_ret = NULL;
2359				return ret;
2360			}
2361		}
2362
2363		if (bio_add_page(bio, page, blocksize, 0) < blocksize)
2364			goto submit_and_realloc;
2365
2366		ctx = get_post_read_ctx(bio);
2367		ctx->enabled_steps |= STEP_DECOMPRESS;
2368		refcount_inc(&dic->refcnt);
2369
2370		inc_page_count(sbi, F2FS_RD_DATA);
2371		f2fs_update_iostat(sbi, inode, FS_DATA_READ_IO, F2FS_BLKSIZE);
2372		*last_block_in_bio = blkaddr;
2373	}
2374
2375	if (from_dnode)
2376		f2fs_put_dnode(&dn);
2377
2378	*bio_ret = bio;
2379	return 0;
2380
2381out_put_dnode:
2382	if (from_dnode)
2383		f2fs_put_dnode(&dn);
2384out:
2385	for (i = 0; i < cc->cluster_size; i++) {
2386		if (cc->rpages[i]) {
2387			ClearPageUptodate(cc->rpages[i]);
2388			unlock_page(cc->rpages[i]);
2389		}
2390	}
2391	*bio_ret = bio;
2392	return ret;
2393}
2394#endif
2395
2396/*
2397 * This function was originally taken from fs/mpage.c, and customized for f2fs.
2398 * Major change was from block_size == page_size in f2fs by default.
2399 */
2400static int f2fs_mpage_readpages(struct inode *inode,
2401		struct readahead_control *rac, struct page *page)
2402{
2403	struct bio *bio = NULL;
2404	sector_t last_block_in_bio = 0;
2405	struct f2fs_map_blocks map;
2406#ifdef CONFIG_F2FS_FS_COMPRESSION
2407	struct compress_ctx cc = {
2408		.inode = inode,
2409		.log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
2410		.cluster_size = F2FS_I(inode)->i_cluster_size,
2411		.cluster_idx = NULL_CLUSTER,
2412		.rpages = NULL,
2413		.cpages = NULL,
2414		.nr_rpages = 0,
2415		.nr_cpages = 0,
2416	};
2417	pgoff_t nc_cluster_idx = NULL_CLUSTER;
2418#endif
2419	unsigned nr_pages = rac ? readahead_count(rac) : 1;
2420	unsigned max_nr_pages = nr_pages;
2421	int ret = 0;
2422
2423	map.m_pblk = 0;
2424	map.m_lblk = 0;
2425	map.m_len = 0;
2426	map.m_flags = 0;
2427	map.m_next_pgofs = NULL;
2428	map.m_next_extent = NULL;
2429	map.m_seg_type = NO_CHECK_TYPE;
2430	map.m_may_create = false;
2431
2432	for (; nr_pages; nr_pages--) {
2433		if (rac) {
2434			page = readahead_page(rac);
2435			prefetchw(&page->flags);
2436		}
2437
2438#ifdef CONFIG_F2FS_FS_COMPRESSION
2439		if (f2fs_compressed_file(inode)) {
2440			/* there are remained compressed pages, submit them */
2441			if (!f2fs_cluster_can_merge_page(&cc, page->index)) {
2442				ret = f2fs_read_multi_pages(&cc, &bio,
2443							max_nr_pages,
2444							&last_block_in_bio,
2445							rac != NULL, false);
2446				f2fs_destroy_compress_ctx(&cc, false);
2447				if (ret)
2448					goto set_error_page;
2449			}
2450			if (cc.cluster_idx == NULL_CLUSTER) {
2451				if (nc_cluster_idx ==
2452					page->index >> cc.log_cluster_size) {
2453					goto read_single_page;
2454				}
2455
2456				ret = f2fs_is_compressed_cluster(inode, page->index);
2457				if (ret < 0)
2458					goto set_error_page;
2459				else if (!ret) {
2460					nc_cluster_idx =
2461						page->index >> cc.log_cluster_size;
2462					goto read_single_page;
2463				}
2464
2465				nc_cluster_idx = NULL_CLUSTER;
2466			}
2467			ret = f2fs_init_compress_ctx(&cc);
2468			if (ret)
2469				goto set_error_page;
2470
2471			f2fs_compress_ctx_add_page(&cc, page);
2472
2473			goto next_page;
2474		}
2475read_single_page:
2476#endif
2477
2478		ret = f2fs_read_single_page(inode, page, max_nr_pages, &map,
2479					&bio, &last_block_in_bio, rac);
2480		if (ret) {
2481#ifdef CONFIG_F2FS_FS_COMPRESSION
2482set_error_page:
2483#endif
2484			zero_user_segment(page, 0, PAGE_SIZE);
2485			unlock_page(page);
2486		}
2487#ifdef CONFIG_F2FS_FS_COMPRESSION
2488next_page:
2489#endif
2490		if (rac)
2491			put_page(page);
2492
2493#ifdef CONFIG_F2FS_FS_COMPRESSION
2494		if (f2fs_compressed_file(inode)) {
2495			/* last page */
2496			if (nr_pages == 1 && !f2fs_cluster_is_empty(&cc)) {
2497				ret = f2fs_read_multi_pages(&cc, &bio,
2498							max_nr_pages,
2499							&last_block_in_bio,
2500							rac != NULL, false);
2501				f2fs_destroy_compress_ctx(&cc, false);
2502			}
2503		}
2504#endif
2505	}
2506	if (bio)
2507		f2fs_submit_read_bio(F2FS_I_SB(inode), bio, DATA);
2508	return ret;
2509}
2510
2511static int f2fs_read_data_folio(struct file *file, struct folio *folio)
2512{
2513	struct page *page = &folio->page;
2514	struct inode *inode = page_file_mapping(page)->host;
2515	int ret = -EAGAIN;
2516
2517	trace_f2fs_readpage(page, DATA);
2518
2519	if (!f2fs_is_compress_backend_ready(inode)) {
2520		unlock_page(page);
2521		return -EOPNOTSUPP;
2522	}
2523
2524	/* If the file has inline data, try to read it directly */
2525	if (f2fs_has_inline_data(inode))
2526		ret = f2fs_read_inline_data(inode, page);
2527	if (ret == -EAGAIN)
2528		ret = f2fs_mpage_readpages(inode, NULL, page);
2529	return ret;
2530}
2531
2532static void f2fs_readahead(struct readahead_control *rac)
2533{
2534	struct inode *inode = rac->mapping->host;
2535
2536	trace_f2fs_readpages(inode, readahead_index(rac), readahead_count(rac));
2537
2538	if (!f2fs_is_compress_backend_ready(inode))
2539		return;
2540
2541	/* If the file has inline data, skip readahead */
2542	if (f2fs_has_inline_data(inode))
2543		return;
2544
2545	f2fs_mpage_readpages(inode, rac, NULL);
2546}
2547
2548int f2fs_encrypt_one_page(struct f2fs_io_info *fio)
2549{
2550	struct inode *inode = fio->page->mapping->host;
2551	struct page *mpage, *page;
2552	gfp_t gfp_flags = GFP_NOFS;
2553
2554	if (!f2fs_encrypted_file(inode))
2555		return 0;
2556
2557	page = fio->compressed_page ? fio->compressed_page : fio->page;
2558
2559	if (fscrypt_inode_uses_inline_crypto(inode))
2560		return 0;
2561
2562retry_encrypt:
2563	fio->encrypted_page = fscrypt_encrypt_pagecache_blocks(page,
2564					PAGE_SIZE, 0, gfp_flags);
2565	if (IS_ERR(fio->encrypted_page)) {
2566		/* flush pending IOs and wait for a while in the ENOMEM case */
2567		if (PTR_ERR(fio->encrypted_page) == -ENOMEM) {
2568			f2fs_flush_merged_writes(fio->sbi);
2569			memalloc_retry_wait(GFP_NOFS);
2570			gfp_flags |= __GFP_NOFAIL;
2571			goto retry_encrypt;
2572		}
2573		return PTR_ERR(fio->encrypted_page);
2574	}
2575
2576	mpage = find_lock_page(META_MAPPING(fio->sbi), fio->old_blkaddr);
2577	if (mpage) {
2578		if (PageUptodate(mpage))
2579			memcpy(page_address(mpage),
2580				page_address(fio->encrypted_page), PAGE_SIZE);
2581		f2fs_put_page(mpage, 1);
2582	}
2583	return 0;
2584}
2585
2586static inline bool check_inplace_update_policy(struct inode *inode,
2587				struct f2fs_io_info *fio)
2588{
2589	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2590
2591	if (IS_F2FS_IPU_HONOR_OPU_WRITE(sbi) &&
2592	    is_inode_flag_set(inode, FI_OPU_WRITE))
2593		return false;
2594	if (IS_F2FS_IPU_FORCE(sbi))
2595		return true;
2596	if (IS_F2FS_IPU_SSR(sbi) && f2fs_need_SSR(sbi))
2597		return true;
2598	if (IS_F2FS_IPU_UTIL(sbi) && utilization(sbi) > SM_I(sbi)->min_ipu_util)
2599		return true;
2600	if (IS_F2FS_IPU_SSR_UTIL(sbi) && f2fs_need_SSR(sbi) &&
2601	    utilization(sbi) > SM_I(sbi)->min_ipu_util)
2602		return true;
2603
2604	/*
2605	 * IPU for rewrite async pages
2606	 */
2607	if (IS_F2FS_IPU_ASYNC(sbi) && fio && fio->op == REQ_OP_WRITE &&
2608	    !(fio->op_flags & REQ_SYNC) && !IS_ENCRYPTED(inode))
2609		return true;
2610
2611	/* this is only set during fdatasync */
2612	if (IS_F2FS_IPU_FSYNC(sbi) && is_inode_flag_set(inode, FI_NEED_IPU))
2613		return true;
2614
2615	if (unlikely(fio && is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2616			!f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2617		return true;
2618
2619	return false;
2620}
2621
2622bool f2fs_should_update_inplace(struct inode *inode, struct f2fs_io_info *fio)
2623{
2624	/* swap file is migrating in aligned write mode */
2625	if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2626		return false;
2627
2628	if (f2fs_is_pinned_file(inode))
2629		return true;
2630
2631	/* if this is cold file, we should overwrite to avoid fragmentation */
2632	if (file_is_cold(inode) && !is_inode_flag_set(inode, FI_OPU_WRITE))
2633		return true;
2634
2635	return check_inplace_update_policy(inode, fio);
2636}
2637
2638bool f2fs_should_update_outplace(struct inode *inode, struct f2fs_io_info *fio)
2639{
2640	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2641
2642	/* The below cases were checked when setting it. */
2643	if (f2fs_is_pinned_file(inode))
2644		return false;
2645	if (fio && is_sbi_flag_set(sbi, SBI_NEED_FSCK))
2646		return true;
2647	if (f2fs_lfs_mode(sbi))
2648		return true;
2649	if (S_ISDIR(inode->i_mode))
2650		return true;
2651	if (IS_NOQUOTA(inode))
2652		return true;
2653	if (f2fs_is_atomic_file(inode))
2654		return true;
2655	/* rewrite low ratio compress data w/ OPU mode to avoid fragmentation */
2656	if (f2fs_compressed_file(inode) &&
2657		F2FS_OPTION(sbi).compress_mode == COMPR_MODE_USER &&
2658		is_inode_flag_set(inode, FI_ENABLE_COMPRESS))
2659		return true;
2660
2661	/* swap file is migrating in aligned write mode */
2662	if (is_inode_flag_set(inode, FI_ALIGNED_WRITE))
2663		return true;
2664
2665	if (is_inode_flag_set(inode, FI_OPU_WRITE))
2666		return true;
2667
2668	if (fio) {
2669		if (page_private_gcing(fio->page))
2670			return true;
2671		if (page_private_dummy(fio->page))
2672			return true;
2673		if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2674			f2fs_is_checkpointed_data(sbi, fio->old_blkaddr)))
2675			return true;
2676	}
2677	return false;
2678}
2679
2680static inline bool need_inplace_update(struct f2fs_io_info *fio)
2681{
2682	struct inode *inode = fio->page->mapping->host;
2683
2684	if (f2fs_should_update_outplace(inode, fio))
2685		return false;
2686
2687	return f2fs_should_update_inplace(inode, fio);
2688}
2689
2690int f2fs_do_write_data_page(struct f2fs_io_info *fio)
2691{
2692	struct page *page = fio->page;
2693	struct inode *inode = page->mapping->host;
2694	struct dnode_of_data dn;
2695	struct node_info ni;
2696	bool ipu_force = false;
2697	int err = 0;
2698
2699	/* Use COW inode to make dnode_of_data for atomic write */
2700	if (f2fs_is_atomic_file(inode))
2701		set_new_dnode(&dn, F2FS_I(inode)->cow_inode, NULL, NULL, 0);
2702	else
2703		set_new_dnode(&dn, inode, NULL, NULL, 0);
2704
2705	if (need_inplace_update(fio) &&
2706	    f2fs_lookup_read_extent_cache_block(inode, page->index,
2707						&fio->old_blkaddr)) {
2708		if (!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2709						DATA_GENERIC_ENHANCE)) {
2710			f2fs_handle_error(fio->sbi,
2711						ERROR_INVALID_BLKADDR);
2712			return -EFSCORRUPTED;
2713		}
2714
2715		ipu_force = true;
2716		fio->need_lock = LOCK_DONE;
2717		goto got_it;
2718	}
2719
2720	/* Deadlock due to between page->lock and f2fs_lock_op */
2721	if (fio->need_lock == LOCK_REQ && !f2fs_trylock_op(fio->sbi))
2722		return -EAGAIN;
2723
2724	err = f2fs_get_dnode_of_data(&dn, page->index, LOOKUP_NODE);
2725	if (err)
2726		goto out;
2727
2728	fio->old_blkaddr = dn.data_blkaddr;
2729
2730	/* This page is already truncated */
2731	if (fio->old_blkaddr == NULL_ADDR) {
2732		ClearPageUptodate(page);
2733		clear_page_private_gcing(page);
2734		goto out_writepage;
2735	}
2736got_it:
2737	if (__is_valid_data_blkaddr(fio->old_blkaddr) &&
2738		!f2fs_is_valid_blkaddr(fio->sbi, fio->old_blkaddr,
2739						DATA_GENERIC_ENHANCE)) {
2740		err = -EFSCORRUPTED;
2741		f2fs_handle_error(fio->sbi, ERROR_INVALID_BLKADDR);
2742		goto out_writepage;
2743	}
2744
2745	/* wait for GCed page writeback via META_MAPPING */
2746	if (fio->post_read)
2747		f2fs_wait_on_block_writeback(inode, fio->old_blkaddr);
2748
2749	/*
2750	 * If current allocation needs SSR,
2751	 * it had better in-place writes for updated data.
2752	 */
2753	if (ipu_force ||
2754		(__is_valid_data_blkaddr(fio->old_blkaddr) &&
2755					need_inplace_update(fio))) {
2756		err = f2fs_encrypt_one_page(fio);
2757		if (err)
2758			goto out_writepage;
2759
2760		set_page_writeback(page);
2761		f2fs_put_dnode(&dn);
2762		if (fio->need_lock == LOCK_REQ)
2763			f2fs_unlock_op(fio->sbi);
2764		err = f2fs_inplace_write_data(fio);
2765		if (err) {
2766			if (fscrypt_inode_uses_fs_layer_crypto(inode))
2767				fscrypt_finalize_bounce_page(&fio->encrypted_page);
2768			if (PageWriteback(page))
2769				end_page_writeback(page);
2770		} else {
2771			set_inode_flag(inode, FI_UPDATE_WRITE);
2772		}
2773		trace_f2fs_do_write_data_page(fio->page, IPU);
2774		return err;
2775	}
2776
2777	if (fio->need_lock == LOCK_RETRY) {
2778		if (!f2fs_trylock_op(fio->sbi)) {
2779			err = -EAGAIN;
2780			goto out_writepage;
2781		}
2782		fio->need_lock = LOCK_REQ;
2783	}
2784
2785	err = f2fs_get_node_info(fio->sbi, dn.nid, &ni, false);
2786	if (err)
2787		goto out_writepage;
2788
2789	fio->version = ni.version;
2790
2791	err = f2fs_encrypt_one_page(fio);
2792	if (err)
2793		goto out_writepage;
2794
2795	set_page_writeback(page);
2796
2797	if (fio->compr_blocks && fio->old_blkaddr == COMPRESS_ADDR)
2798		f2fs_i_compr_blocks_update(inode, fio->compr_blocks - 1, false);
2799
2800	/* LFS mode write path */
2801	f2fs_outplace_write_data(&dn, fio);
2802	trace_f2fs_do_write_data_page(page, OPU);
2803	set_inode_flag(inode, FI_APPEND_WRITE);
2804out_writepage:
2805	f2fs_put_dnode(&dn);
2806out:
2807	if (fio->need_lock == LOCK_REQ)
2808		f2fs_unlock_op(fio->sbi);
2809	return err;
2810}
2811
2812int f2fs_write_single_data_page(struct page *page, int *submitted,
2813				struct bio **bio,
2814				sector_t *last_block,
2815				struct writeback_control *wbc,
2816				enum iostat_type io_type,
2817				int compr_blocks,
2818				bool allow_balance)
2819{
2820	struct inode *inode = page->mapping->host;
2821	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2822	loff_t i_size = i_size_read(inode);
2823	const pgoff_t end_index = ((unsigned long long)i_size)
2824							>> PAGE_SHIFT;
2825	loff_t psize = (loff_t)(page->index + 1) << PAGE_SHIFT;
2826	unsigned offset = 0;
2827	bool need_balance_fs = false;
2828	bool quota_inode = IS_NOQUOTA(inode);
2829	int err = 0;
2830	struct f2fs_io_info fio = {
2831		.sbi = sbi,
2832		.ino = inode->i_ino,
2833		.type = DATA,
2834		.op = REQ_OP_WRITE,
2835		.op_flags = wbc_to_write_flags(wbc),
2836		.old_blkaddr = NULL_ADDR,
2837		.page = page,
2838		.encrypted_page = NULL,
2839		.submitted = 0,
2840		.compr_blocks = compr_blocks,
2841		.need_lock = LOCK_RETRY,
2842		.post_read = f2fs_post_read_required(inode) ? 1 : 0,
2843		.io_type = io_type,
2844		.io_wbc = wbc,
2845		.bio = bio,
2846		.last_block = last_block,
2847	};
2848
2849	trace_f2fs_writepage(page, DATA);
2850
2851	/* we should bypass data pages to proceed the kworker jobs */
2852	if (unlikely(f2fs_cp_error(sbi))) {
2853		mapping_set_error(page->mapping, -EIO);
2854		/*
2855		 * don't drop any dirty dentry pages for keeping lastest
2856		 * directory structure.
2857		 */
2858		if (S_ISDIR(inode->i_mode) &&
2859				!is_sbi_flag_set(sbi, SBI_IS_CLOSE))
2860			goto redirty_out;
2861
2862		/* keep data pages in remount-ro mode */
2863		if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY)
2864			goto redirty_out;
2865		goto out;
2866	}
2867
2868	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2869		goto redirty_out;
2870
2871	if (page->index < end_index ||
2872			f2fs_verity_in_progress(inode) ||
2873			compr_blocks)
2874		goto write;
2875
2876	/*
2877	 * If the offset is out-of-range of file size,
2878	 * this page does not have to be written to disk.
2879	 */
2880	offset = i_size & (PAGE_SIZE - 1);
2881	if ((page->index >= end_index + 1) || !offset)
2882		goto out;
2883
2884	zero_user_segment(page, offset, PAGE_SIZE);
2885write:
2886	/* Dentry/quota blocks are controlled by checkpoint */
2887	if (S_ISDIR(inode->i_mode) || quota_inode) {
2888		/*
2889		 * We need to wait for node_write to avoid block allocation during
2890		 * checkpoint. This can only happen to quota writes which can cause
2891		 * the below discard race condition.
2892		 */
2893		if (quota_inode)
2894			f2fs_down_read(&sbi->node_write);
2895
2896		fio.need_lock = LOCK_DONE;
2897		err = f2fs_do_write_data_page(&fio);
2898
2899		if (quota_inode)
2900			f2fs_up_read(&sbi->node_write);
2901
2902		goto done;
2903	}
2904
2905	if (!wbc->for_reclaim)
2906		need_balance_fs = true;
2907	else if (has_not_enough_free_secs(sbi, 0, 0))
2908		goto redirty_out;
2909	else
2910		set_inode_flag(inode, FI_HOT_DATA);
2911
2912	err = -EAGAIN;
2913	if (f2fs_has_inline_data(inode)) {
2914		err = f2fs_write_inline_data(inode, page);
2915		if (!err)
2916			goto out;
2917	}
2918
2919	if (err == -EAGAIN) {
2920		err = f2fs_do_write_data_page(&fio);
2921		if (err == -EAGAIN) {
2922			fio.need_lock = LOCK_REQ;
2923			err = f2fs_do_write_data_page(&fio);
2924		}
2925	}
2926
2927	if (err) {
2928		file_set_keep_isize(inode);
2929	} else {
2930		spin_lock(&F2FS_I(inode)->i_size_lock);
2931		if (F2FS_I(inode)->last_disk_size < psize)
2932			F2FS_I(inode)->last_disk_size = psize;
2933		spin_unlock(&F2FS_I(inode)->i_size_lock);
2934	}
2935
2936done:
2937	if (err && err != -ENOENT)
2938		goto redirty_out;
2939
2940out:
2941	inode_dec_dirty_pages(inode);
2942	if (err) {
2943		ClearPageUptodate(page);
2944		clear_page_private_gcing(page);
2945	}
2946
2947	if (wbc->for_reclaim) {
2948		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, DATA);
2949		clear_inode_flag(inode, FI_HOT_DATA);
2950		f2fs_remove_dirty_inode(inode);
2951		submitted = NULL;
2952	}
2953	unlock_page(page);
2954	if (!S_ISDIR(inode->i_mode) && !IS_NOQUOTA(inode) &&
2955			!F2FS_I(inode)->wb_task && allow_balance)
2956		f2fs_balance_fs(sbi, need_balance_fs);
2957
2958	if (unlikely(f2fs_cp_error(sbi))) {
2959		f2fs_submit_merged_write(sbi, DATA);
2960		if (bio && *bio)
2961			f2fs_submit_merged_ipu_write(sbi, bio, NULL);
2962		submitted = NULL;
2963	}
2964
2965	if (submitted)
2966		*submitted = fio.submitted;
2967
2968	return 0;
2969
2970redirty_out:
2971	redirty_page_for_writepage(wbc, page);
2972	/*
2973	 * pageout() in MM translates EAGAIN, so calls handle_write_error()
2974	 * -> mapping_set_error() -> set_bit(AS_EIO, ...).
2975	 * file_write_and_wait_range() will see EIO error, which is critical
2976	 * to return value of fsync() followed by atomic_write failure to user.
2977	 */
2978	if (!err || wbc->for_reclaim)
2979		return AOP_WRITEPAGE_ACTIVATE;
2980	unlock_page(page);
2981	return err;
2982}
2983
2984static int f2fs_write_data_page(struct page *page,
2985					struct writeback_control *wbc)
2986{
2987#ifdef CONFIG_F2FS_FS_COMPRESSION
2988	struct inode *inode = page->mapping->host;
2989
2990	if (unlikely(f2fs_cp_error(F2FS_I_SB(inode))))
2991		goto out;
2992
2993	if (f2fs_compressed_file(inode)) {
2994		if (f2fs_is_compressed_cluster(inode, page->index)) {
2995			redirty_page_for_writepage(wbc, page);
2996			return AOP_WRITEPAGE_ACTIVATE;
2997		}
2998	}
2999out:
3000#endif
3001
3002	return f2fs_write_single_data_page(page, NULL, NULL, NULL,
3003						wbc, FS_DATA_IO, 0, true);
3004}
3005
3006/*
3007 * This function was copied from write_cache_pages from mm/page-writeback.c.
3008 * The major change is making write step of cold data page separately from
3009 * warm/hot data page.
3010 */
3011static int f2fs_write_cache_pages(struct address_space *mapping,
3012					struct writeback_control *wbc,
3013					enum iostat_type io_type)
3014{
3015	int ret = 0;
3016	int done = 0, retry = 0;
3017	struct page *pages_local[F2FS_ONSTACK_PAGES];
3018	struct page **pages = pages_local;
3019	struct folio_batch fbatch;
3020	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
3021	struct bio *bio = NULL;
3022	sector_t last_block;
3023#ifdef CONFIG_F2FS_FS_COMPRESSION
3024	struct inode *inode = mapping->host;
3025	struct compress_ctx cc = {
3026		.inode = inode,
3027		.log_cluster_size = F2FS_I(inode)->i_log_cluster_size,
3028		.cluster_size = F2FS_I(inode)->i_cluster_size,
3029		.cluster_idx = NULL_CLUSTER,
3030		.rpages = NULL,
3031		.nr_rpages = 0,
3032		.cpages = NULL,
3033		.valid_nr_cpages = 0,
3034		.rbuf = NULL,
3035		.cbuf = NULL,
3036		.rlen = PAGE_SIZE * F2FS_I(inode)->i_cluster_size,
3037		.private = NULL,
3038	};
3039#endif
3040	int nr_folios, p, idx;
3041	int nr_pages;
3042	unsigned int max_pages = F2FS_ONSTACK_PAGES;
3043	pgoff_t index;
3044	pgoff_t end;		/* Inclusive */
3045	pgoff_t done_index;
3046	int range_whole = 0;
3047	xa_mark_t tag;
3048	int nwritten = 0;
3049	int submitted = 0;
3050	int i;
3051
3052#ifdef CONFIG_F2FS_FS_COMPRESSION
3053	if (f2fs_compressed_file(inode) &&
3054		1 << cc.log_cluster_size > F2FS_ONSTACK_PAGES) {
3055		pages = f2fs_kzalloc(sbi, sizeof(struct page *) <<
3056				cc.log_cluster_size, GFP_NOFS | __GFP_NOFAIL);
3057		max_pages = 1 << cc.log_cluster_size;
3058	}
3059#endif
3060
3061	folio_batch_init(&fbatch);
3062
3063	if (get_dirty_pages(mapping->host) <=
3064				SM_I(F2FS_M_SB(mapping))->min_hot_blocks)
3065		set_inode_flag(mapping->host, FI_HOT_DATA);
3066	else
3067		clear_inode_flag(mapping->host, FI_HOT_DATA);
3068
3069	if (wbc->range_cyclic) {
3070		index = mapping->writeback_index; /* prev offset */
3071		end = -1;
3072	} else {
3073		index = wbc->range_start >> PAGE_SHIFT;
3074		end = wbc->range_end >> PAGE_SHIFT;
3075		if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
3076			range_whole = 1;
3077	}
3078	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3079		tag = PAGECACHE_TAG_TOWRITE;
3080	else
3081		tag = PAGECACHE_TAG_DIRTY;
3082retry:
3083	retry = 0;
3084	if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
3085		tag_pages_for_writeback(mapping, index, end);
3086	done_index = index;
3087	while (!done && !retry && (index <= end)) {
3088		nr_pages = 0;
3089again:
3090		nr_folios = filemap_get_folios_tag(mapping, &index, end,
3091				tag, &fbatch);
3092		if (nr_folios == 0) {
3093			if (nr_pages)
3094				goto write;
3095			break;
3096		}
3097
3098		for (i = 0; i < nr_folios; i++) {
3099			struct folio *folio = fbatch.folios[i];
3100
3101			idx = 0;
3102			p = folio_nr_pages(folio);
3103add_more:
3104			pages[nr_pages] = folio_page(folio, idx);
3105			folio_get(folio);
3106			if (++nr_pages == max_pages) {
3107				index = folio->index + idx + 1;
3108				folio_batch_release(&fbatch);
3109				goto write;
3110			}
3111			if (++idx < p)
3112				goto add_more;
3113		}
3114		folio_batch_release(&fbatch);
3115		goto again;
3116write:
3117		for (i = 0; i < nr_pages; i++) {
3118			struct page *page = pages[i];
3119			struct folio *folio = page_folio(page);
3120			bool need_readd;
3121readd:
3122			need_readd = false;
3123#ifdef CONFIG_F2FS_FS_COMPRESSION
3124			if (f2fs_compressed_file(inode)) {
3125				void *fsdata = NULL;
3126				struct page *pagep;
3127				int ret2;
3128
3129				ret = f2fs_init_compress_ctx(&cc);
3130				if (ret) {
3131					done = 1;
3132					break;
3133				}
3134
3135				if (!f2fs_cluster_can_merge_page(&cc,
3136								folio->index)) {
3137					ret = f2fs_write_multi_pages(&cc,
3138						&submitted, wbc, io_type);
3139					if (!ret)
3140						need_readd = true;
3141					goto result;
3142				}
3143
3144				if (unlikely(f2fs_cp_error(sbi)))
3145					goto lock_folio;
3146
3147				if (!f2fs_cluster_is_empty(&cc))
3148					goto lock_folio;
3149
3150				if (f2fs_all_cluster_page_ready(&cc,
3151					pages, i, nr_pages, true))
3152					goto lock_folio;
3153
3154				ret2 = f2fs_prepare_compress_overwrite(
3155							inode, &pagep,
3156							folio->index, &fsdata);
3157				if (ret2 < 0) {
3158					ret = ret2;
3159					done = 1;
3160					break;
3161				} else if (ret2 &&
3162					(!f2fs_compress_write_end(inode,
3163						fsdata, folio->index, 1) ||
3164					 !f2fs_all_cluster_page_ready(&cc,
3165						pages, i, nr_pages,
3166						false))) {
3167					retry = 1;
3168					break;
3169				}
3170			}
3171#endif
3172			/* give a priority to WB_SYNC threads */
3173			if (atomic_read(&sbi->wb_sync_req[DATA]) &&
3174					wbc->sync_mode == WB_SYNC_NONE) {
3175				done = 1;
3176				break;
3177			}
3178#ifdef CONFIG_F2FS_FS_COMPRESSION
3179lock_folio:
3180#endif
3181			done_index = folio->index;
3182retry_write:
3183			folio_lock(folio);
3184
3185			if (unlikely(folio->mapping != mapping)) {
3186continue_unlock:
3187				folio_unlock(folio);
3188				continue;
3189			}
3190
3191			if (!folio_test_dirty(folio)) {
3192				/* someone wrote it for us */
3193				goto continue_unlock;
3194			}
3195
3196			if (folio_test_writeback(folio)) {
3197				if (wbc->sync_mode == WB_SYNC_NONE)
3198					goto continue_unlock;
3199				f2fs_wait_on_page_writeback(&folio->page, DATA, true, true);
3200			}
3201
3202			if (!folio_clear_dirty_for_io(folio))
3203				goto continue_unlock;
3204
3205#ifdef CONFIG_F2FS_FS_COMPRESSION
3206			if (f2fs_compressed_file(inode)) {
3207				folio_get(folio);
3208				f2fs_compress_ctx_add_page(&cc, &folio->page);
3209				continue;
3210			}
3211#endif
3212			ret = f2fs_write_single_data_page(&folio->page,
3213					&submitted, &bio, &last_block,
3214					wbc, io_type, 0, true);
3215			if (ret == AOP_WRITEPAGE_ACTIVATE)
3216				folio_unlock(folio);
3217#ifdef CONFIG_F2FS_FS_COMPRESSION
3218result:
3219#endif
3220			nwritten += submitted;
3221			wbc->nr_to_write -= submitted;
3222
3223			if (unlikely(ret)) {
3224				/*
3225				 * keep nr_to_write, since vfs uses this to
3226				 * get # of written pages.
3227				 */
3228				if (ret == AOP_WRITEPAGE_ACTIVATE) {
3229					ret = 0;
3230					goto next;
3231				} else if (ret == -EAGAIN) {
3232					ret = 0;
3233					if (wbc->sync_mode == WB_SYNC_ALL) {
3234						f2fs_io_schedule_timeout(
3235							DEFAULT_IO_TIMEOUT);
3236						goto retry_write;
3237					}
3238					goto next;
3239				}
3240				done_index = folio_next_index(folio);
3241				done = 1;
3242				break;
3243			}
3244
3245			if (wbc->nr_to_write <= 0 &&
3246					wbc->sync_mode == WB_SYNC_NONE) {
3247				done = 1;
3248				break;
3249			}
3250next:
3251			if (need_readd)
3252				goto readd;
3253		}
3254		release_pages(pages, nr_pages);
3255		cond_resched();
3256	}
3257#ifdef CONFIG_F2FS_FS_COMPRESSION
3258	/* flush remained pages in compress cluster */
3259	if (f2fs_compressed_file(inode) && !f2fs_cluster_is_empty(&cc)) {
3260		ret = f2fs_write_multi_pages(&cc, &submitted, wbc, io_type);
3261		nwritten += submitted;
3262		wbc->nr_to_write -= submitted;
3263		if (ret) {
3264			done = 1;
3265			retry = 0;
3266		}
3267	}
3268	if (f2fs_compressed_file(inode))
3269		f2fs_destroy_compress_ctx(&cc, false);
3270#endif
3271	if (retry) {
3272		index = 0;
3273		end = -1;
3274		goto retry;
3275	}
3276	if (wbc->range_cyclic && !done)
3277		done_index = 0;
3278	if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
3279		mapping->writeback_index = done_index;
3280
3281	if (nwritten)
3282		f2fs_submit_merged_write_cond(F2FS_M_SB(mapping), mapping->host,
3283								NULL, 0, DATA);
3284	/* submit cached bio of IPU write */
3285	if (bio)
3286		f2fs_submit_merged_ipu_write(sbi, &bio, NULL);
3287
3288#ifdef CONFIG_F2FS_FS_COMPRESSION
3289	if (pages != pages_local)
3290		kfree(pages);
3291#endif
3292
3293	return ret;
3294}
3295
3296static inline bool __should_serialize_io(struct inode *inode,
3297					struct writeback_control *wbc)
3298{
3299	/* to avoid deadlock in path of data flush */
3300	if (F2FS_I(inode)->wb_task)
3301		return false;
3302
3303	if (!S_ISREG(inode->i_mode))
3304		return false;
3305	if (IS_NOQUOTA(inode))
3306		return false;
3307
3308	if (f2fs_need_compress_data(inode))
3309		return true;
3310	if (wbc->sync_mode != WB_SYNC_ALL)
3311		return true;
3312	if (get_dirty_pages(inode) >= SM_I(F2FS_I_SB(inode))->min_seq_blocks)
3313		return true;
3314	return false;
3315}
3316
3317static int __f2fs_write_data_pages(struct address_space *mapping,
3318						struct writeback_control *wbc,
3319						enum iostat_type io_type)
3320{
3321	struct inode *inode = mapping->host;
3322	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3323	struct blk_plug plug;
3324	int ret;
3325	bool locked = false;
3326
3327	/* deal with chardevs and other special file */
3328	if (!mapping->a_ops->writepage)
3329		return 0;
3330
3331	/* skip writing if there is no dirty page in this inode */
3332	if (!get_dirty_pages(inode) && wbc->sync_mode == WB_SYNC_NONE)
3333		return 0;
3334
3335	/* during POR, we don't need to trigger writepage at all. */
3336	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
3337		goto skip_write;
3338
3339	if ((S_ISDIR(inode->i_mode) || IS_NOQUOTA(inode)) &&
3340			wbc->sync_mode == WB_SYNC_NONE &&
3341			get_dirty_pages(inode) < nr_pages_to_skip(sbi, DATA) &&
3342			f2fs_available_free_memory(sbi, DIRTY_DENTS))
3343		goto skip_write;
3344
3345	/* skip writing in file defragment preparing stage */
3346	if (is_inode_flag_set(inode, FI_SKIP_WRITES))
3347		goto skip_write;
3348
3349	trace_f2fs_writepages(mapping->host, wbc, DATA);
3350
3351	/* to avoid spliting IOs due to mixed WB_SYNC_ALL and WB_SYNC_NONE */
3352	if (wbc->sync_mode == WB_SYNC_ALL)
3353		atomic_inc(&sbi->wb_sync_req[DATA]);
3354	else if (atomic_read(&sbi->wb_sync_req[DATA])) {
3355		/* to avoid potential deadlock */
3356		if (current->plug)
3357			blk_finish_plug(current->plug);
3358		goto skip_write;
3359	}
3360
3361	if (__should_serialize_io(inode, wbc)) {
3362		mutex_lock(&sbi->writepages);
3363		locked = true;
3364	}
3365
3366	blk_start_plug(&plug);
3367	ret = f2fs_write_cache_pages(mapping, wbc, io_type);
3368	blk_finish_plug(&plug);
3369
3370	if (locked)
3371		mutex_unlock(&sbi->writepages);
3372
3373	if (wbc->sync_mode == WB_SYNC_ALL)
3374		atomic_dec(&sbi->wb_sync_req[DATA]);
3375	/*
3376	 * if some pages were truncated, we cannot guarantee its mapping->host
3377	 * to detect pending bios.
3378	 */
3379
3380	f2fs_remove_dirty_inode(inode);
3381	return ret;
3382
3383skip_write:
3384	wbc->pages_skipped += get_dirty_pages(inode);
3385	trace_f2fs_writepages(mapping->host, wbc, DATA);
3386	return 0;
3387}
3388
3389static int f2fs_write_data_pages(struct address_space *mapping,
3390			    struct writeback_control *wbc)
3391{
3392	struct inode *inode = mapping->host;
3393
3394	return __f2fs_write_data_pages(mapping, wbc,
3395			F2FS_I(inode)->cp_task == current ?
3396			FS_CP_DATA_IO : FS_DATA_IO);
3397}
3398
3399void f2fs_write_failed(struct inode *inode, loff_t to)
3400{
3401	loff_t i_size = i_size_read(inode);
3402
3403	if (IS_NOQUOTA(inode))
3404		return;
3405
3406	/* In the fs-verity case, f2fs_end_enable_verity() does the truncate */
3407	if (to > i_size && !f2fs_verity_in_progress(inode)) {
3408		f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3409		filemap_invalidate_lock(inode->i_mapping);
3410
3411		truncate_pagecache(inode, i_size);
3412		f2fs_truncate_blocks(inode, i_size, true);
3413
3414		filemap_invalidate_unlock(inode->i_mapping);
3415		f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3416	}
3417}
3418
3419static int prepare_write_begin(struct f2fs_sb_info *sbi,
3420			struct page *page, loff_t pos, unsigned len,
3421			block_t *blk_addr, bool *node_changed)
3422{
3423	struct inode *inode = page->mapping->host;
3424	pgoff_t index = page->index;
3425	struct dnode_of_data dn;
3426	struct page *ipage;
3427	bool locked = false;
3428	int flag = F2FS_GET_BLOCK_PRE_AIO;
3429	int err = 0;
3430
3431	/*
3432	 * If a whole page is being written and we already preallocated all the
3433	 * blocks, then there is no need to get a block address now.
3434	 */
3435	if (len == PAGE_SIZE && is_inode_flag_set(inode, FI_PREALLOCATED_ALL))
3436		return 0;
3437
3438	/* f2fs_lock_op avoids race between write CP and convert_inline_page */
3439	if (f2fs_has_inline_data(inode)) {
3440		if (pos + len > MAX_INLINE_DATA(inode))
3441			flag = F2FS_GET_BLOCK_DEFAULT;
3442		f2fs_map_lock(sbi, flag);
3443		locked = true;
3444	} else if ((pos & PAGE_MASK) >= i_size_read(inode)) {
3445		f2fs_map_lock(sbi, flag);
3446		locked = true;
3447	}
3448
3449restart:
3450	/* check inline_data */
3451	ipage = f2fs_get_node_page(sbi, inode->i_ino);
3452	if (IS_ERR(ipage)) {
3453		err = PTR_ERR(ipage);
3454		goto unlock_out;
3455	}
3456
3457	set_new_dnode(&dn, inode, ipage, ipage, 0);
3458
3459	if (f2fs_has_inline_data(inode)) {
3460		if (pos + len <= MAX_INLINE_DATA(inode)) {
3461			f2fs_do_read_inline_data(page, ipage);
3462			set_inode_flag(inode, FI_DATA_EXIST);
3463			if (inode->i_nlink)
3464				set_page_private_inline(ipage);
3465			goto out;
3466		}
3467		err = f2fs_convert_inline_page(&dn, page);
3468		if (err || dn.data_blkaddr != NULL_ADDR)
3469			goto out;
3470	}
3471
3472	if (!f2fs_lookup_read_extent_cache_block(inode, index,
3473						 &dn.data_blkaddr)) {
3474		if (locked) {
3475			err = f2fs_reserve_block(&dn, index);
3476			goto out;
3477		}
3478
3479		/* hole case */
3480		err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3481		if (!err && dn.data_blkaddr != NULL_ADDR)
3482			goto out;
3483		f2fs_put_dnode(&dn);
3484		f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO);
3485		WARN_ON(flag != F2FS_GET_BLOCK_PRE_AIO);
3486		locked = true;
3487		goto restart;
3488	}
3489out:
3490	if (!err) {
3491		/* convert_inline_page can make node_changed */
3492		*blk_addr = dn.data_blkaddr;
3493		*node_changed = dn.node_changed;
3494	}
3495	f2fs_put_dnode(&dn);
3496unlock_out:
3497	if (locked)
3498		f2fs_map_unlock(sbi, flag);
3499	return err;
3500}
3501
3502static int __find_data_block(struct inode *inode, pgoff_t index,
3503				block_t *blk_addr)
3504{
3505	struct dnode_of_data dn;
3506	struct page *ipage;
3507	int err = 0;
3508
3509	ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
3510	if (IS_ERR(ipage))
3511		return PTR_ERR(ipage);
3512
3513	set_new_dnode(&dn, inode, ipage, ipage, 0);
3514
3515	if (!f2fs_lookup_read_extent_cache_block(inode, index,
3516						 &dn.data_blkaddr)) {
3517		/* hole case */
3518		err = f2fs_get_dnode_of_data(&dn, index, LOOKUP_NODE);
3519		if (err) {
3520			dn.data_blkaddr = NULL_ADDR;
3521			err = 0;
3522		}
3523	}
3524	*blk_addr = dn.data_blkaddr;
3525	f2fs_put_dnode(&dn);
3526	return err;
3527}
3528
3529static int __reserve_data_block(struct inode *inode, pgoff_t index,
3530				block_t *blk_addr, bool *node_changed)
3531{
3532	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3533	struct dnode_of_data dn;
3534	struct page *ipage;
3535	int err = 0;
3536
3537	f2fs_map_lock(sbi, F2FS_GET_BLOCK_PRE_AIO);
3538
3539	ipage = f2fs_get_node_page(sbi, inode->i_ino);
3540	if (IS_ERR(ipage)) {
3541		err = PTR_ERR(ipage);
3542		goto unlock_out;
3543	}
3544	set_new_dnode(&dn, inode, ipage, ipage, 0);
3545
3546	if (!f2fs_lookup_read_extent_cache_block(dn.inode, index,
3547						&dn.data_blkaddr))
3548		err = f2fs_reserve_block(&dn, index);
3549
3550	*blk_addr = dn.data_blkaddr;
3551	*node_changed = dn.node_changed;
3552	f2fs_put_dnode(&dn);
3553
3554unlock_out:
3555	f2fs_map_unlock(sbi, F2FS_GET_BLOCK_PRE_AIO);
3556	return err;
3557}
3558
3559static int prepare_atomic_write_begin(struct f2fs_sb_info *sbi,
3560			struct page *page, loff_t pos, unsigned int len,
3561			block_t *blk_addr, bool *node_changed, bool *use_cow)
3562{
3563	struct inode *inode = page->mapping->host;
3564	struct inode *cow_inode = F2FS_I(inode)->cow_inode;
3565	pgoff_t index = page->index;
3566	int err = 0;
3567	block_t ori_blk_addr = NULL_ADDR;
3568
3569	/* If pos is beyond the end of file, reserve a new block in COW inode */
3570	if ((pos & PAGE_MASK) >= i_size_read(inode))
3571		goto reserve_block;
3572
3573	/* Look for the block in COW inode first */
3574	err = __find_data_block(cow_inode, index, blk_addr);
3575	if (err) {
3576		return err;
3577	} else if (*blk_addr != NULL_ADDR) {
3578		*use_cow = true;
3579		return 0;
3580	}
3581
3582	if (is_inode_flag_set(inode, FI_ATOMIC_REPLACE))
3583		goto reserve_block;
3584
3585	/* Look for the block in the original inode */
3586	err = __find_data_block(inode, index, &ori_blk_addr);
3587	if (err)
3588		return err;
3589
3590reserve_block:
3591	/* Finally, we should reserve a new block in COW inode for the update */
3592	err = __reserve_data_block(cow_inode, index, blk_addr, node_changed);
3593	if (err)
3594		return err;
3595	inc_atomic_write_cnt(inode);
3596
3597	if (ori_blk_addr != NULL_ADDR)
3598		*blk_addr = ori_blk_addr;
3599	return 0;
3600}
3601
3602static int f2fs_write_begin(struct file *file, struct address_space *mapping,
3603		loff_t pos, unsigned len, struct page **pagep, void **fsdata)
3604{
3605	struct inode *inode = mapping->host;
3606	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3607	struct page *page = NULL;
3608	pgoff_t index = ((unsigned long long) pos) >> PAGE_SHIFT;
3609	bool need_balance = false;
3610	bool use_cow = false;
3611	block_t blkaddr = NULL_ADDR;
3612	int err = 0;
3613
3614	trace_f2fs_write_begin(inode, pos, len);
3615
3616	if (!f2fs_is_checkpoint_ready(sbi)) {
3617		err = -ENOSPC;
3618		goto fail;
3619	}
3620
3621	/*
3622	 * We should check this at this moment to avoid deadlock on inode page
3623	 * and #0 page. The locking rule for inline_data conversion should be:
3624	 * lock_page(page #0) -> lock_page(inode_page)
3625	 */
3626	if (index != 0) {
3627		err = f2fs_convert_inline_inode(inode);
3628		if (err)
3629			goto fail;
3630	}
3631
3632#ifdef CONFIG_F2FS_FS_COMPRESSION
3633	if (f2fs_compressed_file(inode)) {
3634		int ret;
3635
3636		*fsdata = NULL;
3637
3638		if (len == PAGE_SIZE && !(f2fs_is_atomic_file(inode)))
3639			goto repeat;
3640
3641		ret = f2fs_prepare_compress_overwrite(inode, pagep,
3642							index, fsdata);
3643		if (ret < 0) {
3644			err = ret;
3645			goto fail;
3646		} else if (ret) {
3647			return 0;
3648		}
3649	}
3650#endif
3651
3652repeat:
3653	/*
3654	 * Do not use grab_cache_page_write_begin() to avoid deadlock due to
3655	 * wait_for_stable_page. Will wait that below with our IO control.
3656	 */
3657	page = f2fs_pagecache_get_page(mapping, index,
3658				FGP_LOCK | FGP_WRITE | FGP_CREAT, GFP_NOFS);
3659	if (!page) {
3660		err = -ENOMEM;
3661		goto fail;
3662	}
3663
3664	/* TODO: cluster can be compressed due to race with .writepage */
3665
3666	*pagep = page;
3667
3668	if (f2fs_is_atomic_file(inode))
3669		err = prepare_atomic_write_begin(sbi, page, pos, len,
3670					&blkaddr, &need_balance, &use_cow);
3671	else
3672		err = prepare_write_begin(sbi, page, pos, len,
3673					&blkaddr, &need_balance);
3674	if (err)
3675		goto fail;
3676
3677	if (need_balance && !IS_NOQUOTA(inode) &&
3678			has_not_enough_free_secs(sbi, 0, 0)) {
3679		unlock_page(page);
3680		f2fs_balance_fs(sbi, true);
3681		lock_page(page);
3682		if (page->mapping != mapping) {
3683			/* The page got truncated from under us */
3684			f2fs_put_page(page, 1);
3685			goto repeat;
3686		}
3687	}
3688
3689	f2fs_wait_on_page_writeback(page, DATA, false, true);
3690
3691	if (len == PAGE_SIZE || PageUptodate(page))
3692		return 0;
3693
3694	if (!(pos & (PAGE_SIZE - 1)) && (pos + len) >= i_size_read(inode) &&
3695	    !f2fs_verity_in_progress(inode)) {
3696		zero_user_segment(page, len, PAGE_SIZE);
3697		return 0;
3698	}
3699
3700	if (blkaddr == NEW_ADDR) {
3701		zero_user_segment(page, 0, PAGE_SIZE);
3702		SetPageUptodate(page);
3703	} else {
3704		if (!f2fs_is_valid_blkaddr(sbi, blkaddr,
3705				DATA_GENERIC_ENHANCE_READ)) {
3706			err = -EFSCORRUPTED;
3707			f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
3708			goto fail;
3709		}
3710		err = f2fs_submit_page_read(use_cow ?
3711				F2FS_I(inode)->cow_inode : inode, page,
3712				blkaddr, 0, true);
3713		if (err)
3714			goto fail;
3715
3716		lock_page(page);
3717		if (unlikely(page->mapping != mapping)) {
3718			f2fs_put_page(page, 1);
3719			goto repeat;
3720		}
3721		if (unlikely(!PageUptodate(page))) {
3722			err = -EIO;
3723			goto fail;
3724		}
3725	}
3726	return 0;
3727
3728fail:
3729	f2fs_put_page(page, 1);
3730	f2fs_write_failed(inode, pos + len);
3731	return err;
3732}
3733
3734static int f2fs_write_end(struct file *file,
3735			struct address_space *mapping,
3736			loff_t pos, unsigned len, unsigned copied,
3737			struct page *page, void *fsdata)
3738{
3739	struct inode *inode = page->mapping->host;
3740
3741	trace_f2fs_write_end(inode, pos, len, copied);
3742
3743	/*
3744	 * This should be come from len == PAGE_SIZE, and we expect copied
3745	 * should be PAGE_SIZE. Otherwise, we treat it with zero copied and
3746	 * let generic_perform_write() try to copy data again through copied=0.
3747	 */
3748	if (!PageUptodate(page)) {
3749		if (unlikely(copied != len))
3750			copied = 0;
3751		else
3752			SetPageUptodate(page);
3753	}
3754
3755#ifdef CONFIG_F2FS_FS_COMPRESSION
3756	/* overwrite compressed file */
3757	if (f2fs_compressed_file(inode) && fsdata) {
3758		f2fs_compress_write_end(inode, fsdata, page->index, copied);
3759		f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3760
3761		if (pos + copied > i_size_read(inode) &&
3762				!f2fs_verity_in_progress(inode))
3763			f2fs_i_size_write(inode, pos + copied);
3764		return copied;
3765	}
3766#endif
3767
3768	if (!copied)
3769		goto unlock_out;
3770
3771	set_page_dirty(page);
3772
3773	if (pos + copied > i_size_read(inode) &&
3774	    !f2fs_verity_in_progress(inode)) {
3775		f2fs_i_size_write(inode, pos + copied);
3776		if (f2fs_is_atomic_file(inode))
3777			f2fs_i_size_write(F2FS_I(inode)->cow_inode,
3778					pos + copied);
3779	}
3780unlock_out:
3781	f2fs_put_page(page, 1);
3782	f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
3783	return copied;
3784}
3785
3786void f2fs_invalidate_folio(struct folio *folio, size_t offset, size_t length)
3787{
3788	struct inode *inode = folio->mapping->host;
3789	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3790
3791	if (inode->i_ino >= F2FS_ROOT_INO(sbi) &&
3792				(offset || length != folio_size(folio)))
3793		return;
3794
3795	if (folio_test_dirty(folio)) {
3796		if (inode->i_ino == F2FS_META_INO(sbi)) {
3797			dec_page_count(sbi, F2FS_DIRTY_META);
3798		} else if (inode->i_ino == F2FS_NODE_INO(sbi)) {
3799			dec_page_count(sbi, F2FS_DIRTY_NODES);
3800		} else {
3801			inode_dec_dirty_pages(inode);
3802			f2fs_remove_dirty_inode(inode);
3803		}
3804	}
3805	clear_page_private_all(&folio->page);
3806}
3807
3808bool f2fs_release_folio(struct folio *folio, gfp_t wait)
3809{
3810	/* If this is dirty folio, keep private data */
3811	if (folio_test_dirty(folio))
3812		return false;
3813
3814	clear_page_private_all(&folio->page);
3815	return true;
3816}
3817
3818static bool f2fs_dirty_data_folio(struct address_space *mapping,
3819		struct folio *folio)
3820{
3821	struct inode *inode = mapping->host;
3822
3823	trace_f2fs_set_page_dirty(&folio->page, DATA);
3824
3825	if (!folio_test_uptodate(folio))
3826		folio_mark_uptodate(folio);
3827	BUG_ON(folio_test_swapcache(folio));
3828
3829	if (filemap_dirty_folio(mapping, folio)) {
3830		f2fs_update_dirty_folio(inode, folio);
3831		return true;
3832	}
3833	return false;
3834}
3835
3836
3837static sector_t f2fs_bmap_compress(struct inode *inode, sector_t block)
3838{
3839#ifdef CONFIG_F2FS_FS_COMPRESSION
3840	struct dnode_of_data dn;
3841	sector_t start_idx, blknr = 0;
3842	int ret;
3843
3844	start_idx = round_down(block, F2FS_I(inode)->i_cluster_size);
3845
3846	set_new_dnode(&dn, inode, NULL, NULL, 0);
3847	ret = f2fs_get_dnode_of_data(&dn, start_idx, LOOKUP_NODE);
3848	if (ret)
3849		return 0;
3850
3851	if (dn.data_blkaddr != COMPRESS_ADDR) {
3852		dn.ofs_in_node += block - start_idx;
3853		blknr = f2fs_data_blkaddr(&dn);
3854		if (!__is_valid_data_blkaddr(blknr))
3855			blknr = 0;
3856	}
3857
3858	f2fs_put_dnode(&dn);
3859	return blknr;
3860#else
3861	return 0;
3862#endif
3863}
3864
3865
3866static sector_t f2fs_bmap(struct address_space *mapping, sector_t block)
3867{
3868	struct inode *inode = mapping->host;
3869	sector_t blknr = 0;
3870
3871	if (f2fs_has_inline_data(inode))
3872		goto out;
3873
3874	/* make sure allocating whole blocks */
3875	if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
3876		filemap_write_and_wait(mapping);
3877
3878	/* Block number less than F2FS MAX BLOCKS */
3879	if (unlikely(block >= max_file_blocks(inode)))
3880		goto out;
3881
3882	if (f2fs_compressed_file(inode)) {
3883		blknr = f2fs_bmap_compress(inode, block);
3884	} else {
3885		struct f2fs_map_blocks map;
3886
3887		memset(&map, 0, sizeof(map));
3888		map.m_lblk = block;
3889		map.m_len = 1;
3890		map.m_next_pgofs = NULL;
3891		map.m_seg_type = NO_CHECK_TYPE;
3892
3893		if (!f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_BMAP))
3894			blknr = map.m_pblk;
3895	}
3896out:
3897	trace_f2fs_bmap(inode, block, blknr);
3898	return blknr;
3899}
3900
3901#ifdef CONFIG_SWAP
3902static int f2fs_migrate_blocks(struct inode *inode, block_t start_blk,
3903							unsigned int blkcnt)
3904{
3905	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3906	unsigned int blkofs;
3907	unsigned int blk_per_sec = BLKS_PER_SEC(sbi);
3908	unsigned int secidx = start_blk / blk_per_sec;
3909	unsigned int end_sec = secidx + blkcnt / blk_per_sec;
3910	int ret = 0;
3911
3912	f2fs_down_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3913	filemap_invalidate_lock(inode->i_mapping);
3914
3915	set_inode_flag(inode, FI_ALIGNED_WRITE);
3916	set_inode_flag(inode, FI_OPU_WRITE);
3917
3918	for (; secidx < end_sec; secidx++) {
3919		f2fs_down_write(&sbi->pin_sem);
3920
3921		f2fs_lock_op(sbi);
3922		f2fs_allocate_new_section(sbi, CURSEG_COLD_DATA_PINNED, false);
3923		f2fs_unlock_op(sbi);
3924
3925		set_inode_flag(inode, FI_SKIP_WRITES);
3926
3927		for (blkofs = 0; blkofs < blk_per_sec; blkofs++) {
3928			struct page *page;
3929			unsigned int blkidx = secidx * blk_per_sec + blkofs;
3930
3931			page = f2fs_get_lock_data_page(inode, blkidx, true);
3932			if (IS_ERR(page)) {
3933				f2fs_up_write(&sbi->pin_sem);
3934				ret = PTR_ERR(page);
3935				goto done;
3936			}
3937
3938			set_page_dirty(page);
3939			f2fs_put_page(page, 1);
3940		}
3941
3942		clear_inode_flag(inode, FI_SKIP_WRITES);
3943
3944		ret = filemap_fdatawrite(inode->i_mapping);
3945
3946		f2fs_up_write(&sbi->pin_sem);
3947
3948		if (ret)
3949			break;
3950	}
3951
3952done:
3953	clear_inode_flag(inode, FI_SKIP_WRITES);
3954	clear_inode_flag(inode, FI_OPU_WRITE);
3955	clear_inode_flag(inode, FI_ALIGNED_WRITE);
3956
3957	filemap_invalidate_unlock(inode->i_mapping);
3958	f2fs_up_write(&F2FS_I(inode)->i_gc_rwsem[WRITE]);
3959
3960	return ret;
3961}
3962
3963static int check_swap_activate(struct swap_info_struct *sis,
3964				struct file *swap_file, sector_t *span)
3965{
3966	struct address_space *mapping = swap_file->f_mapping;
3967	struct inode *inode = mapping->host;
3968	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
3969	sector_t cur_lblock;
3970	sector_t last_lblock;
3971	sector_t pblock;
3972	sector_t lowest_pblock = -1;
3973	sector_t highest_pblock = 0;
3974	int nr_extents = 0;
3975	unsigned long nr_pblocks;
3976	unsigned int blks_per_sec = BLKS_PER_SEC(sbi);
3977	unsigned int sec_blks_mask = BLKS_PER_SEC(sbi) - 1;
3978	unsigned int not_aligned = 0;
3979	int ret = 0;
3980
3981	/*
3982	 * Map all the blocks into the extent list.  This code doesn't try
3983	 * to be very smart.
3984	 */
3985	cur_lblock = 0;
3986	last_lblock = bytes_to_blks(inode, i_size_read(inode));
3987
3988	while (cur_lblock < last_lblock && cur_lblock < sis->max) {
3989		struct f2fs_map_blocks map;
3990retry:
3991		cond_resched();
3992
3993		memset(&map, 0, sizeof(map));
3994		map.m_lblk = cur_lblock;
3995		map.m_len = last_lblock - cur_lblock;
3996		map.m_next_pgofs = NULL;
3997		map.m_next_extent = NULL;
3998		map.m_seg_type = NO_CHECK_TYPE;
3999		map.m_may_create = false;
4000
4001		ret = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_FIEMAP);
4002		if (ret)
4003			goto out;
4004
4005		/* hole */
4006		if (!(map.m_flags & F2FS_MAP_FLAGS)) {
4007			f2fs_err(sbi, "Swapfile has holes");
4008			ret = -EINVAL;
4009			goto out;
4010		}
4011
4012		pblock = map.m_pblk;
4013		nr_pblocks = map.m_len;
4014
4015		if ((pblock - SM_I(sbi)->main_blkaddr) & sec_blks_mask ||
4016				nr_pblocks & sec_blks_mask) {
4017			not_aligned++;
4018
4019			nr_pblocks = roundup(nr_pblocks, blks_per_sec);
4020			if (cur_lblock + nr_pblocks > sis->max)
4021				nr_pblocks -= blks_per_sec;
4022
4023			if (!nr_pblocks) {
4024				/* this extent is last one */
4025				nr_pblocks = map.m_len;
4026				f2fs_warn(sbi, "Swapfile: last extent is not aligned to section");
4027				goto next;
4028			}
4029
4030			ret = f2fs_migrate_blocks(inode, cur_lblock,
4031							nr_pblocks);
4032			if (ret)
4033				goto out;
4034			goto retry;
4035		}
4036next:
4037		if (cur_lblock + nr_pblocks >= sis->max)
4038			nr_pblocks = sis->max - cur_lblock;
4039
4040		if (cur_lblock) {	/* exclude the header page */
4041			if (pblock < lowest_pblock)
4042				lowest_pblock = pblock;
4043			if (pblock + nr_pblocks - 1 > highest_pblock)
4044				highest_pblock = pblock + nr_pblocks - 1;
4045		}
4046
4047		/*
4048		 * We found a PAGE_SIZE-length, PAGE_SIZE-aligned run of blocks
4049		 */
4050		ret = add_swap_extent(sis, cur_lblock, nr_pblocks, pblock);
4051		if (ret < 0)
4052			goto out;
4053		nr_extents += ret;
4054		cur_lblock += nr_pblocks;
4055	}
4056	ret = nr_extents;
4057	*span = 1 + highest_pblock - lowest_pblock;
4058	if (cur_lblock == 0)
4059		cur_lblock = 1;	/* force Empty message */
4060	sis->max = cur_lblock;
4061	sis->pages = cur_lblock - 1;
4062	sis->highest_bit = cur_lblock - 1;
4063out:
4064	if (not_aligned)
4065		f2fs_warn(sbi, "Swapfile (%u) is not align to section: 1) creat(), 2) ioctl(F2FS_IOC_SET_PIN_FILE), 3) fallocate(%lu * N)",
4066			  not_aligned, blks_per_sec * F2FS_BLKSIZE);
4067	return ret;
4068}
4069
4070static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4071				sector_t *span)
4072{
4073	struct inode *inode = file_inode(file);
4074	int ret;
4075
4076	if (!S_ISREG(inode->i_mode))
4077		return -EINVAL;
4078
4079	if (f2fs_readonly(F2FS_I_SB(inode)->sb))
4080		return -EROFS;
4081
4082	if (f2fs_lfs_mode(F2FS_I_SB(inode))) {
4083		f2fs_err(F2FS_I_SB(inode),
4084			"Swapfile not supported in LFS mode");
4085		return -EINVAL;
4086	}
4087
4088	ret = f2fs_convert_inline_inode(inode);
4089	if (ret)
4090		return ret;
4091
4092	if (!f2fs_disable_compressed_file(inode))
4093		return -EINVAL;
4094
4095	f2fs_precache_extents(inode);
4096
4097	ret = check_swap_activate(sis, file, span);
4098	if (ret < 0)
4099		return ret;
4100
4101	stat_inc_swapfile_inode(inode);
4102	set_inode_flag(inode, FI_PIN_FILE);
4103	f2fs_update_time(F2FS_I_SB(inode), REQ_TIME);
4104	return ret;
4105}
4106
4107static void f2fs_swap_deactivate(struct file *file)
4108{
4109	struct inode *inode = file_inode(file);
4110
4111	stat_dec_swapfile_inode(inode);
4112	clear_inode_flag(inode, FI_PIN_FILE);
4113}
4114#else
4115static int f2fs_swap_activate(struct swap_info_struct *sis, struct file *file,
4116				sector_t *span)
4117{
4118	return -EOPNOTSUPP;
4119}
4120
4121static void f2fs_swap_deactivate(struct file *file)
4122{
4123}
4124#endif
4125
4126const struct address_space_operations f2fs_dblock_aops = {
4127	.read_folio	= f2fs_read_data_folio,
4128	.readahead	= f2fs_readahead,
4129	.writepage	= f2fs_write_data_page,
4130	.writepages	= f2fs_write_data_pages,
4131	.write_begin	= f2fs_write_begin,
4132	.write_end	= f2fs_write_end,
4133	.dirty_folio	= f2fs_dirty_data_folio,
4134	.migrate_folio	= filemap_migrate_folio,
4135	.invalidate_folio = f2fs_invalidate_folio,
4136	.release_folio	= f2fs_release_folio,
4137	.bmap		= f2fs_bmap,
4138	.swap_activate  = f2fs_swap_activate,
4139	.swap_deactivate = f2fs_swap_deactivate,
4140};
4141
4142void f2fs_clear_page_cache_dirty_tag(struct page *page)
4143{
4144	struct address_space *mapping = page_mapping(page);
4145	unsigned long flags;
4146
4147	xa_lock_irqsave(&mapping->i_pages, flags);
4148	__xa_clear_mark(&mapping->i_pages, page_index(page),
4149						PAGECACHE_TAG_DIRTY);
4150	xa_unlock_irqrestore(&mapping->i_pages, flags);
4151}
4152
4153int __init f2fs_init_post_read_processing(void)
4154{
4155	bio_post_read_ctx_cache =
4156		kmem_cache_create("f2fs_bio_post_read_ctx",
4157				  sizeof(struct bio_post_read_ctx), 0, 0, NULL);
4158	if (!bio_post_read_ctx_cache)
4159		goto fail;
4160	bio_post_read_ctx_pool =
4161		mempool_create_slab_pool(NUM_PREALLOC_POST_READ_CTXS,
4162					 bio_post_read_ctx_cache);
4163	if (!bio_post_read_ctx_pool)
4164		goto fail_free_cache;
4165	return 0;
4166
4167fail_free_cache:
4168	kmem_cache_destroy(bio_post_read_ctx_cache);
4169fail:
4170	return -ENOMEM;
4171}
4172
4173void f2fs_destroy_post_read_processing(void)
4174{
4175	mempool_destroy(bio_post_read_ctx_pool);
4176	kmem_cache_destroy(bio_post_read_ctx_cache);
4177}
4178
4179int f2fs_init_post_read_wq(struct f2fs_sb_info *sbi)
4180{
4181	if (!f2fs_sb_has_encrypt(sbi) &&
4182		!f2fs_sb_has_verity(sbi) &&
4183		!f2fs_sb_has_compression(sbi))
4184		return 0;
4185
4186	sbi->post_read_wq = alloc_workqueue("f2fs_post_read_wq",
4187						 WQ_UNBOUND | WQ_HIGHPRI,
4188						 num_online_cpus());
4189	return sbi->post_read_wq ? 0 : -ENOMEM;
4190}
4191
4192void f2fs_destroy_post_read_wq(struct f2fs_sb_info *sbi)
4193{
4194	if (sbi->post_read_wq)
4195		destroy_workqueue(sbi->post_read_wq);
4196}
4197
4198int __init f2fs_init_bio_entry_cache(void)
4199{
4200	bio_entry_slab = f2fs_kmem_cache_create("f2fs_bio_entry_slab",
4201			sizeof(struct bio_entry));
4202	return bio_entry_slab ? 0 : -ENOMEM;
4203}
4204
4205void f2fs_destroy_bio_entry_cache(void)
4206{
4207	kmem_cache_destroy(bio_entry_slab);
4208}
4209
4210static int f2fs_iomap_begin(struct inode *inode, loff_t offset, loff_t length,
4211			    unsigned int flags, struct iomap *iomap,
4212			    struct iomap *srcmap)
4213{
4214	struct f2fs_map_blocks map = {};
4215	pgoff_t next_pgofs = 0;
4216	int err;
4217
4218	map.m_lblk = bytes_to_blks(inode, offset);
4219	map.m_len = bytes_to_blks(inode, offset + length - 1) - map.m_lblk + 1;
4220	map.m_next_pgofs = &next_pgofs;
4221	map.m_seg_type = f2fs_rw_hint_to_seg_type(inode->i_write_hint);
4222	if (flags & IOMAP_WRITE)
4223		map.m_may_create = true;
4224
4225	err = f2fs_map_blocks(inode, &map, F2FS_GET_BLOCK_DIO);
4226	if (err)
4227		return err;
4228
4229	iomap->offset = blks_to_bytes(inode, map.m_lblk);
4230
4231	/*
4232	 * When inline encryption is enabled, sometimes I/O to an encrypted file
4233	 * has to be broken up to guarantee DUN contiguity.  Handle this by
4234	 * limiting the length of the mapping returned.
4235	 */
4236	map.m_len = fscrypt_limit_io_blocks(inode, map.m_lblk, map.m_len);
4237
4238	/*
4239	 * We should never see delalloc or compressed extents here based on
4240	 * prior flushing and checks.
4241	 */
4242	if (WARN_ON_ONCE(map.m_pblk == NEW_ADDR))
4243		return -EINVAL;
4244	if (WARN_ON_ONCE(map.m_pblk == COMPRESS_ADDR))
4245		return -EINVAL;
4246
4247	if (map.m_pblk != NULL_ADDR) {
4248		iomap->length = blks_to_bytes(inode, map.m_len);
4249		iomap->type = IOMAP_MAPPED;
4250		iomap->flags |= IOMAP_F_MERGED;
4251		iomap->bdev = map.m_bdev;
4252		iomap->addr = blks_to_bytes(inode, map.m_pblk);
4253	} else {
4254		if (flags & IOMAP_WRITE)
4255			return -ENOTBLK;
4256		iomap->length = blks_to_bytes(inode, next_pgofs) -
4257				iomap->offset;
4258		iomap->type = IOMAP_HOLE;
4259		iomap->addr = IOMAP_NULL_ADDR;
4260	}
4261
4262	if (map.m_flags & F2FS_MAP_NEW)
4263		iomap->flags |= IOMAP_F_NEW;
4264	if ((inode->i_state & I_DIRTY_DATASYNC) ||
4265	    offset + length > i_size_read(inode))
4266		iomap->flags |= IOMAP_F_DIRTY;
4267
4268	return 0;
4269}
4270
4271const struct iomap_ops f2fs_iomap_ops = {
4272	.iomap_begin	= f2fs_iomap_begin,
4273};