checkpoint.c

   1// SPDX-License-Identifier: GPL-2.0
   2/*
   3 * fs/f2fs/checkpoint.c
   4 *
   5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   6 *             http://www.samsung.com/
   7 */
   8#include <linux/fs.h>
   9#include <linux/bio.h>
  10#include <linux/mpage.h>
  11#include <linux/writeback.h>
  12#include <linux/blkdev.h>
  13#include <linux/f2fs_fs.h>
  14#include <linux/pagevec.h>
  15#include <linux/swap.h>
  16#include <linux/kthread.h>
  17
  18#include "f2fs.h"
  19#include "node.h"
  20#include "segment.h"
  21#include "iostat.h"
  22#include <trace/events/f2fs.h>
  23
  24#define DEFAULT_CHECKPOINT_IOPRIO (IOPRIO_PRIO_VALUE(IOPRIO_CLASS_RT, 3))
  25
  26static struct kmem_cache *ino_entry_slab;
  27struct kmem_cache *f2fs_inode_entry_slab;
  28
  29void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io,
  30						unsigned char reason)
  31{
  32	f2fs_build_fault_attr(sbi, 0, 0, FAULT_ALL);
  33	if (!end_io)
  34		f2fs_flush_merged_writes(sbi);
  35	f2fs_handle_critical_error(sbi, reason);
  36}
  37
  38/*
  39 * We guarantee no failure on the returned page.
  40 */
  41struct folio *f2fs_grab_meta_folio(struct f2fs_sb_info *sbi, pgoff_t index)
  42{
  43	struct address_space *mapping = META_MAPPING(sbi);
  44	struct folio *folio;
  45repeat:
  46	folio = f2fs_grab_cache_folio(mapping, index, false);
  47	if (IS_ERR(folio)) {
  48		cond_resched();
  49		goto repeat;
  50	}
  51	f2fs_folio_wait_writeback(folio, META, true, true);
  52	if (!folio_test_uptodate(folio))
  53		folio_mark_uptodate(folio);
  54	return folio;
  55}
  56
  57static struct folio *__get_meta_folio(struct f2fs_sb_info *sbi, pgoff_t index,
  58							bool is_meta)
  59{
  60	struct address_space *mapping = META_MAPPING(sbi);
  61	struct folio *folio;
  62	struct f2fs_io_info fio = {
  63		.sbi = sbi,
  64		.type = META,
  65		.op = REQ_OP_READ,
  66		.op_flags = REQ_META | REQ_PRIO,
  67		.old_blkaddr = index,
  68		.new_blkaddr = index,
  69		.encrypted_page = NULL,
  70		.is_por = !is_meta ? 1 : 0,
  71	};
  72	int err;
  73
  74	if (unlikely(!is_meta))
  75		fio.op_flags &= ~REQ_META;
  76repeat:
  77	folio = f2fs_grab_cache_folio(mapping, index, false);
  78	if (IS_ERR(folio)) {
  79		cond_resched();
  80		goto repeat;
  81	}
  82	if (folio_test_uptodate(folio))
  83		goto out;
  84
  85	fio.folio = folio;
  86
  87	err = f2fs_submit_page_bio(&fio);
  88	if (err) {
  89		f2fs_folio_put(folio, true);
  90		return ERR_PTR(err);
  91	}
  92
  93	f2fs_update_iostat(sbi, NULL, FS_META_READ_IO, F2FS_BLKSIZE);
  94
  95	folio_lock(folio);
  96	if (unlikely(!is_meta_folio(folio))) {
  97		f2fs_folio_put(folio, true);
  98		goto repeat;
  99	}
 100
 101	if (unlikely(!folio_test_uptodate(folio))) {
 102		f2fs_handle_page_eio(sbi, folio, META);
 103		f2fs_folio_put(folio, true);
 104		return ERR_PTR(-EIO);
 105	}
 106out:
 107	return folio;
 108}
 109
 110struct folio *f2fs_get_meta_folio(struct f2fs_sb_info *sbi, pgoff_t index)
 111{
 112	return __get_meta_folio(sbi, index, true);
 113}
 114
 115struct folio *f2fs_get_meta_folio_retry(struct f2fs_sb_info *sbi, pgoff_t index)
 116{
 117	struct folio *folio;
 118	int count = 0;
 119
 120retry:
 121	folio = __get_meta_folio(sbi, index, true);
 122	if (IS_ERR(folio)) {
 123		if (PTR_ERR(folio) == -EIO &&
 124				++count <= DEFAULT_RETRY_IO_COUNT)
 125			goto retry;
 126		f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_META_PAGE);
 127	}
 128	return folio;
 129}
 130
 131/* for POR only */
 132struct folio *f2fs_get_tmp_folio(struct f2fs_sb_info *sbi, pgoff_t index)
 133{
 134	return __get_meta_folio(sbi, index, false);
 135}
 136
 137static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr,
 138							int type)
 139{
 140	struct seg_entry *se;
 141	unsigned int segno, offset;
 142	bool exist;
 143
 144	if (type == DATA_GENERIC)
 145		return true;
 146
 147	segno = GET_SEGNO(sbi, blkaddr);
 148	offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 149	se = get_seg_entry(sbi, segno);
 150
 151	exist = f2fs_test_bit(offset, se->cur_valid_map);
 152
 153	/* skip data, if we already have an error in checkpoint. */
 154	if (unlikely(f2fs_cp_error(sbi)))
 155		return exist;
 156
 157	if ((exist && type == DATA_GENERIC_ENHANCE_UPDATE) ||
 158		(!exist && type == DATA_GENERIC_ENHANCE))
 159		goto out_err;
 160	if (!exist && type != DATA_GENERIC_ENHANCE_UPDATE)
 161		goto out_handle;
 162	return exist;
 163
 164out_err:
 165	f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
 166		 blkaddr, exist);
 167	set_sbi_flag(sbi, SBI_NEED_FSCK);
 168	dump_stack();
 169out_handle:
 170	f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
 171	return exist;
 172}
 173
 174static bool __f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
 175					block_t blkaddr, int type)
 176{
 177	switch (type) {
 178	case META_NAT:
 179		break;
 180	case META_SIT:
 181		if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
 182			goto check_only;
 183		break;
 184	case META_SSA:
 185		if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
 186			blkaddr < SM_I(sbi)->ssa_blkaddr))
 187			goto check_only;
 188		break;
 189	case META_CP:
 190		if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
 191			blkaddr < __start_cp_addr(sbi)))
 192			goto check_only;
 193		break;
 194	case META_POR:
 195		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 196			blkaddr < MAIN_BLKADDR(sbi)))
 197			goto check_only;
 198		break;
 199	case DATA_GENERIC:
 200	case DATA_GENERIC_ENHANCE:
 201	case DATA_GENERIC_ENHANCE_READ:
 202	case DATA_GENERIC_ENHANCE_UPDATE:
 203		if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
 204				blkaddr < MAIN_BLKADDR(sbi))) {
 205
 206			/* Skip to emit an error message. */
 207			if (unlikely(f2fs_cp_error(sbi)))
 208				return false;
 209
 210			f2fs_warn(sbi, "access invalid blkaddr:%u",
 211				  blkaddr);
 212			set_sbi_flag(sbi, SBI_NEED_FSCK);
 213			dump_stack();
 214			goto err;
 215		} else {
 216			return __is_bitmap_valid(sbi, blkaddr, type);
 217		}
 218		break;
 219	case META_GENERIC:
 220		if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
 221			blkaddr >= MAIN_BLKADDR(sbi)))
 222			goto err;
 223		break;
 224	default:
 225		BUG();
 226	}
 227
 228	return true;
 229err:
 230	f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
 231check_only:
 232	return false;
 233}
 234
 235bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
 236					block_t blkaddr, int type)
 237{
 238	if (time_to_inject(sbi, FAULT_BLKADDR_VALIDITY))
 239		return false;
 240	return __f2fs_is_valid_blkaddr(sbi, blkaddr, type);
 241}
 242
 243bool f2fs_is_valid_blkaddr_raw(struct f2fs_sb_info *sbi,
 244					block_t blkaddr, int type)
 245{
 246	return __f2fs_is_valid_blkaddr(sbi, blkaddr, type);
 247}
 248
 249/*
 250 * Readahead CP/NAT/SIT/SSA/POR pages
 251 */
 252int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
 253							int type, bool sync)
 254{
 255	block_t blkno = start;
 256	struct f2fs_io_info fio = {
 257		.sbi = sbi,
 258		.type = META,
 259		.op = REQ_OP_READ,
 260		.op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
 261		.encrypted_page = NULL,
 262		.in_list = 0,
 263		.is_por = (type == META_POR) ? 1 : 0,
 264	};
 265	struct blk_plug plug;
 266	int err;
 267
 268	if (unlikely(type == META_POR))
 269		fio.op_flags &= ~REQ_META;
 270
 271	blk_start_plug(&plug);
 272	for (; nrpages-- > 0; blkno++) {
 273		struct folio *folio;
 274
 275		if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
 276			goto out;
 277
 278		switch (type) {
 279		case META_NAT:
 280			if (unlikely(blkno >=
 281					NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
 282				blkno = 0;
 283			/* get nat block addr */
 284			fio.new_blkaddr = current_nat_addr(sbi,
 285					blkno * NAT_ENTRY_PER_BLOCK);
 286			break;
 287		case META_SIT:
 288			if (unlikely(blkno >= TOTAL_SEGS(sbi)))
 289				goto out;
 290			/* get sit block addr */
 291			fio.new_blkaddr = current_sit_addr(sbi,
 292					blkno * SIT_ENTRY_PER_BLOCK);
 293			break;
 294		case META_SSA:
 295		case META_CP:
 296		case META_POR:
 297			fio.new_blkaddr = blkno;
 298			break;
 299		default:
 300			BUG();
 301		}
 302
 303		folio = f2fs_grab_cache_folio(META_MAPPING(sbi),
 304						fio.new_blkaddr, false);
 305		if (IS_ERR(folio))
 306			continue;
 307		if (folio_test_uptodate(folio)) {
 308			f2fs_folio_put(folio, true);
 309			continue;
 310		}
 311
 312		fio.folio = folio;
 313		err = f2fs_submit_page_bio(&fio);
 314		f2fs_folio_put(folio, err ? true : false);
 315
 316		if (!err)
 317			f2fs_update_iostat(sbi, NULL, FS_META_READ_IO,
 318							F2FS_BLKSIZE);
 319	}
 320out:
 321	blk_finish_plug(&plug);
 322	return blkno - start;
 323}
 324
 325void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index,
 326							unsigned int ra_blocks)
 327{
 328	struct folio *folio;
 329	bool readahead = false;
 330
 331	if (ra_blocks == RECOVERY_MIN_RA_BLOCKS)
 332		return;
 333
 334	folio = filemap_get_folio(META_MAPPING(sbi), index);
 335	if (IS_ERR(folio) || !folio_test_uptodate(folio))
 336		readahead = true;
 337	f2fs_folio_put(folio, false);
 338
 339	if (readahead)
 340		f2fs_ra_meta_pages(sbi, index, ra_blocks, META_POR, true);
 341}
 342
 343static bool __f2fs_write_meta_folio(struct folio *folio,
 344				struct writeback_control *wbc,
 345				enum iostat_type io_type)
 346{
 347	struct f2fs_sb_info *sbi = F2FS_F_SB(folio);
 348
 349	trace_f2fs_writepage(folio, META);
 350
 351	if (unlikely(f2fs_cp_error(sbi))) {
 352		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
 353			folio_clear_uptodate(folio);
 354			dec_page_count(sbi, F2FS_DIRTY_META);
 355			folio_unlock(folio);
 356			return true;
 357		}
 358		goto redirty_out;
 359	}
 360	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 361		goto redirty_out;
 362
 363	f2fs_do_write_meta_page(sbi, folio, io_type);
 364	dec_page_count(sbi, F2FS_DIRTY_META);
 365
 366	folio_unlock(folio);
 367
 368	if (unlikely(f2fs_cp_error(sbi)))
 369		f2fs_submit_merged_write(sbi, META);
 370
 371	return true;
 372
 373redirty_out:
 374	folio_redirty_for_writepage(wbc, folio);
 375	return false;
 376}
 377
 378static int f2fs_write_meta_pages(struct address_space *mapping,
 379				struct writeback_control *wbc)
 380{
 381	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 382	long diff, written;
 383
 384	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 385		goto skip_write;
 386
 387	/* collect a number of dirty meta pages and write together */
 388	if (wbc->sync_mode != WB_SYNC_ALL &&
 389			get_pages(sbi, F2FS_DIRTY_META) <
 390					nr_pages_to_skip(sbi, META))
 391		goto skip_write;
 392
 393	/* if locked failed, cp will flush dirty pages instead */
 394	if (!f2fs_down_write_trylock(&sbi->cp_global_sem))
 395		goto skip_write;
 396
 397	trace_f2fs_writepages(mapping->host, wbc, META);
 398	diff = nr_pages_to_write(sbi, META, wbc);
 399	written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
 400	f2fs_up_write(&sbi->cp_global_sem);
 401	wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 402	return 0;
 403
 404skip_write:
 405	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 406	trace_f2fs_writepages(mapping->host, wbc, META);
 407	return 0;
 408}
 409
 410long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 411				long nr_to_write, enum iostat_type io_type)
 412{
 413	struct address_space *mapping = META_MAPPING(sbi);
 414	pgoff_t index = 0, prev = ULONG_MAX;
 415	struct folio_batch fbatch;
 416	long nwritten = 0;
 417	int nr_folios;
 418	struct writeback_control wbc = {};
 419	struct blk_plug plug;
 420
 421	folio_batch_init(&fbatch);
 422
 423	blk_start_plug(&plug);
 424
 425	while ((nr_folios = filemap_get_folios_tag(mapping, &index,
 426					(pgoff_t)-1,
 427					PAGECACHE_TAG_DIRTY, &fbatch))) {
 428		int i;
 429
 430		for (i = 0; i < nr_folios; i++) {
 431			struct folio *folio = fbatch.folios[i];
 432
 433			if (nr_to_write != LONG_MAX && i != 0 &&
 434					folio->index != prev +
 435					folio_nr_pages(fbatch.folios[i-1])) {
 436				folio_batch_release(&fbatch);
 437				goto stop;
 438			}
 439
 440			folio_lock(folio);
 441
 442			if (unlikely(!is_meta_folio(folio))) {
 443continue_unlock:
 444				folio_unlock(folio);
 445				continue;
 446			}
 447			if (!folio_test_dirty(folio)) {
 448				/* someone wrote it for us */
 449				goto continue_unlock;
 450			}
 451
 452			f2fs_folio_wait_writeback(folio, META, true, true);
 453
 454			if (!folio_clear_dirty_for_io(folio))
 455				goto continue_unlock;
 456
 457			if (!__f2fs_write_meta_folio(folio, &wbc,
 458						io_type)) {
 459				folio_unlock(folio);
 460				break;
 461			}
 462			nwritten += folio_nr_pages(folio);
 463			prev = folio->index;
 464			if (unlikely(nwritten >= nr_to_write))
 465				break;
 466		}
 467		folio_batch_release(&fbatch);
 468		cond_resched();
 469	}
 470stop:
 471	if (nwritten)
 472		f2fs_submit_merged_write(sbi, type);
 473
 474	blk_finish_plug(&plug);
 475
 476	return nwritten;
 477}
 478
 479static bool f2fs_dirty_meta_folio(struct address_space *mapping,
 480		struct folio *folio)
 481{
 482	trace_f2fs_set_page_dirty(folio, META);
 483
 484	if (!folio_test_uptodate(folio))
 485		folio_mark_uptodate(folio);
 486	if (filemap_dirty_folio(mapping, folio)) {
 487		inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_META);
 488		folio_set_f2fs_reference(folio);
 489		return true;
 490	}
 491	return false;
 492}
 493
 494const struct address_space_operations f2fs_meta_aops = {
 495	.writepages	= f2fs_write_meta_pages,
 496	.dirty_folio	= f2fs_dirty_meta_folio,
 497	.invalidate_folio = f2fs_invalidate_folio,
 498	.release_folio	= f2fs_release_folio,
 499	.migrate_folio	= filemap_migrate_folio,
 500};
 501
 502static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
 503						unsigned int devidx, int type)
 504{
 505	struct inode_management *im = &sbi->im[type];
 506	struct ino_entry *e = NULL, *new = NULL;
 507	int ret;
 508
 509	if (type == FLUSH_INO) {
 510		rcu_read_lock();
 511		e = radix_tree_lookup(&im->ino_root, ino);
 512		rcu_read_unlock();
 513	}
 514
 515retry:
 516	if (!e)
 517		new = f2fs_kmem_cache_alloc(ino_entry_slab,
 518						GFP_NOFS, true, NULL);
 519
 520	ret = radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 521	f2fs_bug_on(sbi, ret);
 522
 523	spin_lock(&im->ino_lock);
 524	e = radix_tree_lookup(&im->ino_root, ino);
 525	if (!e) {
 526		if (!new) {
 527			spin_unlock(&im->ino_lock);
 528			radix_tree_preload_end();
 529			goto retry;
 530		}
 531		e = new;
 532		if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
 533			f2fs_bug_on(sbi, 1);
 534
 535		memset(e, 0, sizeof(struct ino_entry));
 536		e->ino = ino;
 537
 538		list_add_tail(&e->list, &im->ino_list);
 539		if (type != ORPHAN_INO)
 540			im->ino_num++;
 541	}
 542
 543	if (type == FLUSH_INO)
 544		f2fs_set_bit(devidx, (char *)&e->dirty_device);
 545
 546	spin_unlock(&im->ino_lock);
 547	radix_tree_preload_end();
 548
 549	if (new && e != new)
 550		kmem_cache_free(ino_entry_slab, new);
 551}
 552
 553static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 554{
 555	struct inode_management *im = &sbi->im[type];
 556	struct ino_entry *e;
 557
 558	spin_lock(&im->ino_lock);
 559	e = radix_tree_lookup(&im->ino_root, ino);
 560	if (e) {
 561		list_del(&e->list);
 562		radix_tree_delete(&im->ino_root, ino);
 563		im->ino_num--;
 564		spin_unlock(&im->ino_lock);
 565		kmem_cache_free(ino_entry_slab, e);
 566		return;
 567	}
 568	spin_unlock(&im->ino_lock);
 569}
 570
 571void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 572{
 573	/* add new dirty ino entry into list */
 574	__add_ino_entry(sbi, ino, 0, type);
 575}
 576
 577void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 578{
 579	/* remove dirty ino entry from list */
 580	__remove_ino_entry(sbi, ino, type);
 581}
 582
 583/* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
 584bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 585{
 586	struct inode_management *im = &sbi->im[mode];
 587	struct ino_entry *e;
 588
 589	spin_lock(&im->ino_lock);
 590	e = radix_tree_lookup(&im->ino_root, ino);
 591	spin_unlock(&im->ino_lock);
 592	return e ? true : false;
 593}
 594
 595void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
 596{
 597	struct ino_entry *e, *tmp;
 598	int i;
 599
 600	for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
 601		struct inode_management *im = &sbi->im[i];
 602
 603		spin_lock(&im->ino_lock);
 604		list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 605			list_del(&e->list);
 606			radix_tree_delete(&im->ino_root, e->ino);
 607			kmem_cache_free(ino_entry_slab, e);
 608			im->ino_num--;
 609		}
 610		spin_unlock(&im->ino_lock);
 611	}
 612}
 613
 614void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 615					unsigned int devidx, int type)
 616{
 617	__add_ino_entry(sbi, ino, devidx, type);
 618}
 619
 620bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 621					unsigned int devidx, int type)
 622{
 623	struct inode_management *im = &sbi->im[type];
 624	struct ino_entry *e;
 625	bool is_dirty = false;
 626
 627	spin_lock(&im->ino_lock);
 628	e = radix_tree_lookup(&im->ino_root, ino);
 629	if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
 630		is_dirty = true;
 631	spin_unlock(&im->ino_lock);
 632	return is_dirty;
 633}
 634
 635int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
 636{
 637	struct inode_management *im = &sbi->im[ORPHAN_INO];
 638	int err = 0;
 639
 640	spin_lock(&im->ino_lock);
 641
 642	if (time_to_inject(sbi, FAULT_ORPHAN)) {
 643		spin_unlock(&im->ino_lock);
 644		return -ENOSPC;
 645	}
 646
 647	if (unlikely(im->ino_num >= sbi->max_orphans))
 648		err = -ENOSPC;
 649	else
 650		im->ino_num++;
 651	spin_unlock(&im->ino_lock);
 652
 653	return err;
 654}
 655
 656void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
 657{
 658	struct inode_management *im = &sbi->im[ORPHAN_INO];
 659
 660	spin_lock(&im->ino_lock);
 661	f2fs_bug_on(sbi, im->ino_num == 0);
 662	im->ino_num--;
 663	spin_unlock(&im->ino_lock);
 664}
 665
 666void f2fs_add_orphan_inode(struct inode *inode)
 667{
 668	/* add new orphan ino entry into list */
 669	__add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
 670	f2fs_update_inode_page(inode);
 671}
 672
 673void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 674{
 675	/* remove orphan entry from orphan list */
 676	__remove_ino_entry(sbi, ino, ORPHAN_INO);
 677}
 678
 679static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 680{
 681	struct inode *inode;
 682	struct node_info ni;
 683	int err;
 684
 685	inode = f2fs_iget_retry(sbi->sb, ino);
 686	if (IS_ERR(inode)) {
 687		/*
 688		 * there should be a bug that we can't find the entry
 689		 * to orphan inode.
 690		 */
 691		f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 692		return PTR_ERR(inode);
 693	}
 694
 695	err = f2fs_dquot_initialize(inode);
 696	if (err) {
 697		iput(inode);
 698		goto err_out;
 699	}
 700
 701	clear_nlink(inode);
 702
 703	/* truncate all the data during iput */
 704	iput(inode);
 705
 706	err = f2fs_get_node_info(sbi, ino, &ni, false);
 707	if (err)
 708		goto err_out;
 709
 710	/* ENOMEM was fully retried in f2fs_evict_inode. */
 711	if (ni.blk_addr != NULL_ADDR) {
 712		err = -EIO;
 713		goto err_out;
 714	}
 715	return 0;
 716
 717err_out:
 718	set_sbi_flag(sbi, SBI_NEED_FSCK);
 719	f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
 720		  __func__, ino);
 721	return err;
 722}
 723
 724int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
 725{
 726	block_t start_blk, orphan_blocks, i, j;
 727	int err = 0;
 728
 729	if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
 730		return 0;
 731
 732	if (f2fs_hw_is_readonly(sbi)) {
 733		f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
 734		return 0;
 735	}
 736
 737	if (is_sbi_flag_set(sbi, SBI_IS_WRITABLE))
 738		f2fs_info(sbi, "orphan cleanup on readonly fs");
 739
 740	start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 741	orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 742
 743	f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 744
 745	for (i = 0; i < orphan_blocks; i++) {
 746		struct folio *folio;
 747		struct f2fs_orphan_block *orphan_blk;
 748
 749		folio = f2fs_get_meta_folio(sbi, start_blk + i);
 750		if (IS_ERR(folio)) {
 751			err = PTR_ERR(folio);
 752			goto out;
 753		}
 754
 755		orphan_blk = folio_address(folio);
 756		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 757			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 758
 759			err = recover_orphan_inode(sbi, ino);
 760			if (err) {
 761				f2fs_folio_put(folio, true);
 762				goto out;
 763			}
 764		}
 765		f2fs_folio_put(folio, true);
 766	}
 767	/* clear Orphan Flag */
 768	clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
 769out:
 770	set_sbi_flag(sbi, SBI_IS_RECOVERED);
 771
 772	return err;
 773}
 774
 775static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 776{
 777	struct list_head *head;
 778	struct f2fs_orphan_block *orphan_blk = NULL;
 779	unsigned int nentries = 0;
 780	unsigned short index = 1;
 781	unsigned short orphan_blocks;
 782	struct folio *folio = NULL;
 783	struct ino_entry *orphan = NULL;
 784	struct inode_management *im = &sbi->im[ORPHAN_INO];
 785
 786	orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 787
 788	/*
 789	 * we don't need to do spin_lock(&im->ino_lock) here, since all the
 790	 * orphan inode operations are covered under f2fs_lock_op().
 791	 * And, spin_lock should be avoided due to page operations below.
 792	 */
 793	head = &im->ino_list;
 794
 795	/* loop for each orphan inode entry and write them in journal block */
 796	list_for_each_entry(orphan, head, list) {
 797		if (!folio) {
 798			folio = f2fs_grab_meta_folio(sbi, start_blk++);
 799			orphan_blk = folio_address(folio);
 800			memset(orphan_blk, 0, sizeof(*orphan_blk));
 801		}
 802
 803		orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 804
 805		if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 806			/*
 807			 * an orphan block is full of 1020 entries,
 808			 * then we need to flush current orphan blocks
 809			 * and bring another one in memory
 810			 */
 811			orphan_blk->blk_addr = cpu_to_le16(index);
 812			orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 813			orphan_blk->entry_count = cpu_to_le32(nentries);
 814			folio_mark_dirty(folio);
 815			f2fs_folio_put(folio, true);
 816			index++;
 817			nentries = 0;
 818			folio = NULL;
 819		}
 820	}
 821
 822	if (folio) {
 823		orphan_blk->blk_addr = cpu_to_le16(index);
 824		orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 825		orphan_blk->entry_count = cpu_to_le32(nentries);
 826		folio_mark_dirty(folio);
 827		f2fs_folio_put(folio, true);
 828	}
 829}
 830
 831static __u32 f2fs_checkpoint_chksum(struct f2fs_checkpoint *ckpt)
 832{
 833	unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
 834	__u32 chksum;
 835
 836	chksum = f2fs_crc32(ckpt, chksum_ofs);
 837	if (chksum_ofs < CP_CHKSUM_OFFSET) {
 838		chksum_ofs += sizeof(chksum);
 839		chksum = f2fs_chksum(chksum, (__u8 *)ckpt + chksum_ofs,
 840				     F2FS_BLKSIZE - chksum_ofs);
 841	}
 842	return chksum;
 843}
 844
 845static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
 846		struct f2fs_checkpoint **cp_block, struct folio **cp_folio,
 847		unsigned long long *version)
 848{
 849	size_t crc_offset = 0;
 850	__u32 crc;
 851
 852	*cp_folio = f2fs_get_meta_folio(sbi, cp_addr);
 853	if (IS_ERR(*cp_folio))
 854		return PTR_ERR(*cp_folio);
 855
 856	*cp_block = folio_address(*cp_folio);
 857
 858	crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
 859	if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
 860			crc_offset > CP_CHKSUM_OFFSET) {
 861		f2fs_folio_put(*cp_folio, true);
 862		f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
 863		return -EINVAL;
 864	}
 865
 866	crc = f2fs_checkpoint_chksum(*cp_block);
 867	if (crc != cur_cp_crc(*cp_block)) {
 868		f2fs_folio_put(*cp_folio, true);
 869		f2fs_warn(sbi, "invalid crc value");
 870		return -EINVAL;
 871	}
 872
 873	*version = cur_cp_version(*cp_block);
 874	return 0;
 875}
 876
 877static struct folio *validate_checkpoint(struct f2fs_sb_info *sbi,
 878				block_t cp_addr, unsigned long long *version)
 879{
 880	struct folio *cp_folio_1 = NULL, *cp_folio_2 = NULL;
 881	struct f2fs_checkpoint *cp_block = NULL;
 882	unsigned long long cur_version = 0, pre_version = 0;
 883	unsigned int cp_blocks;
 884	int err;
 885
 886	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 887					&cp_folio_1, version);
 888	if (err)
 889		return NULL;
 890
 891	cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
 892
 893	if (cp_blocks > BLKS_PER_SEG(sbi) || cp_blocks <= F2FS_CP_PACKS) {
 894		f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
 895			  le32_to_cpu(cp_block->cp_pack_total_block_count));
 896		goto invalid_cp;
 897	}
 898	pre_version = *version;
 899
 900	cp_addr += cp_blocks - 1;
 901	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 902					&cp_folio_2, version);
 903	if (err)
 904		goto invalid_cp;
 905	cur_version = *version;
 906
 907	if (cur_version == pre_version) {
 908		*version = cur_version;
 909		f2fs_folio_put(cp_folio_2, true);
 910		return cp_folio_1;
 911	}
 912	f2fs_folio_put(cp_folio_2, true);
 913invalid_cp:
 914	f2fs_folio_put(cp_folio_1, true);
 915	return NULL;
 916}
 917
 918int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
 919{
 920	struct f2fs_checkpoint *cp_block;
 921	struct f2fs_super_block *fsb = sbi->raw_super;
 922	struct folio *cp1, *cp2, *cur_folio;
 923	unsigned long blk_size = sbi->blocksize;
 924	unsigned long long cp1_version = 0, cp2_version = 0;
 925	unsigned long long cp_start_blk_no;
 926	unsigned int cp_blks = 1 + __cp_payload(sbi);
 927	block_t cp_blk_no;
 928	int i;
 929	int err;
 930
 931	sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
 932				  GFP_KERNEL);
 933	if (!sbi->ckpt)
 934		return -ENOMEM;
 935	/*
 936	 * Finding out valid cp block involves read both
 937	 * sets( cp pack 1 and cp pack 2)
 938	 */
 939	cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 940	cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 941
 942	/* The second checkpoint pack should start at the next segment */
 943	cp_start_blk_no += ((unsigned long long)1) <<
 944				le32_to_cpu(fsb->log_blocks_per_seg);
 945	cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 946
 947	if (cp1 && cp2) {
 948		if (ver_after(cp2_version, cp1_version))
 949			cur_folio = cp2;
 950		else
 951			cur_folio = cp1;
 952	} else if (cp1) {
 953		cur_folio = cp1;
 954	} else if (cp2) {
 955		cur_folio = cp2;
 956	} else {
 957		err = -EFSCORRUPTED;
 958		goto fail_no_cp;
 959	}
 960
 961	cp_block = folio_address(cur_folio);
 962	memcpy(sbi->ckpt, cp_block, blk_size);
 963
 964	if (cur_folio == cp1)
 965		sbi->cur_cp_pack = 1;
 966	else
 967		sbi->cur_cp_pack = 2;
 968
 969	/* Sanity checking of checkpoint */
 970	if (f2fs_sanity_check_ckpt(sbi)) {
 971		err = -EFSCORRUPTED;
 972		goto free_fail_no_cp;
 973	}
 974
 975	if (cp_blks <= 1)
 976		goto done;
 977
 978	cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 979	if (cur_folio == cp2)
 980		cp_blk_no += BIT(le32_to_cpu(fsb->log_blocks_per_seg));
 981
 982	for (i = 1; i < cp_blks; i++) {
 983		void *sit_bitmap_ptr;
 984		unsigned char *ckpt = (unsigned char *)sbi->ckpt;
 985
 986		cur_folio = f2fs_get_meta_folio(sbi, cp_blk_no + i);
 987		if (IS_ERR(cur_folio)) {
 988			err = PTR_ERR(cur_folio);
 989			goto free_fail_no_cp;
 990		}
 991		sit_bitmap_ptr = folio_address(cur_folio);
 992		memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
 993		f2fs_folio_put(cur_folio, true);
 994	}
 995done:
 996	f2fs_folio_put(cp1, true);
 997	f2fs_folio_put(cp2, true);
 998	return 0;
 999
1000free_fail_no_cp:
1001	f2fs_folio_put(cp1, true);
1002	f2fs_folio_put(cp2, true);
1003fail_no_cp:
1004	kvfree(sbi->ckpt);
1005	return err;
1006}
1007
1008static void __add_dirty_inode(struct inode *inode, enum inode_type type)
1009{
1010	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1011	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1012
1013	if (is_inode_flag_set(inode, flag))
1014		return;
1015
1016	set_inode_flag(inode, flag);
1017	list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
1018	stat_inc_dirty_inode(sbi, type);
1019}
1020
1021static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1022{
1023	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1024
1025	if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1026		return;
1027
1028	list_del_init(&F2FS_I(inode)->dirty_list);
1029	clear_inode_flag(inode, flag);
1030	stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1031}
1032
1033void f2fs_update_dirty_folio(struct inode *inode, struct folio *folio)
1034{
1035	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1036	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1037
1038	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1039			!S_ISLNK(inode->i_mode))
1040		return;
1041
1042	spin_lock(&sbi->inode_lock[type]);
1043	if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1044		__add_dirty_inode(inode, type);
1045	inode_inc_dirty_pages(inode);
1046	spin_unlock(&sbi->inode_lock[type]);
1047
1048	folio_set_f2fs_reference(folio);
1049}
1050
1051void f2fs_remove_dirty_inode(struct inode *inode)
1052{
1053	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1054	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1055
1056	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1057			!S_ISLNK(inode->i_mode))
1058		return;
1059
1060	if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1061		return;
1062
1063	spin_lock(&sbi->inode_lock[type]);
1064	__remove_dirty_inode(inode, type);
1065	spin_unlock(&sbi->inode_lock[type]);
1066}
1067
1068int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type,
1069						bool from_cp)
1070{
1071	struct list_head *head;
1072	struct inode *inode;
1073	struct f2fs_inode_info *fi;
1074	bool is_dir = (type == DIR_INODE);
1075	unsigned long ino = 0;
1076
1077	trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1078				get_pages(sbi, is_dir ?
1079				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1080retry:
1081	if (unlikely(f2fs_cp_error(sbi))) {
1082		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1083				get_pages(sbi, is_dir ?
1084				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1085		return -EIO;
1086	}
1087
1088	spin_lock(&sbi->inode_lock[type]);
1089
1090	head = &sbi->inode_list[type];
1091	if (list_empty(head)) {
1092		spin_unlock(&sbi->inode_lock[type]);
1093		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1094				get_pages(sbi, is_dir ?
1095				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1096		return 0;
1097	}
1098	fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1099	inode = igrab(&fi->vfs_inode);
1100	spin_unlock(&sbi->inode_lock[type]);
1101	if (inode) {
1102		unsigned long cur_ino = inode->i_ino;
1103
1104		if (from_cp)
1105			F2FS_I(inode)->cp_task = current;
1106		F2FS_I(inode)->wb_task = current;
1107
1108		filemap_fdatawrite(inode->i_mapping);
1109
1110		F2FS_I(inode)->wb_task = NULL;
1111		if (from_cp)
1112			F2FS_I(inode)->cp_task = NULL;
1113
1114		iput(inode);
1115		/* We need to give cpu to another writers. */
1116		if (ino == cur_ino)
1117			cond_resched();
1118		else
1119			ino = cur_ino;
1120	} else {
1121		/*
1122		 * We should submit bio, since it exists several
1123		 * writebacking dentry pages in the freeing inode.
1124		 */
1125		f2fs_submit_merged_write(sbi, DATA);
1126		cond_resched();
1127	}
1128	goto retry;
1129}
1130
1131static int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1132{
1133	struct list_head *head = &sbi->inode_list[DIRTY_META];
1134	struct inode *inode;
1135	struct f2fs_inode_info *fi;
1136	s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1137
1138	while (total--) {
1139		if (unlikely(f2fs_cp_error(sbi)))
1140			return -EIO;
1141
1142		spin_lock(&sbi->inode_lock[DIRTY_META]);
1143		if (list_empty(head)) {
1144			spin_unlock(&sbi->inode_lock[DIRTY_META]);
1145			return 0;
1146		}
1147		fi = list_first_entry(head, struct f2fs_inode_info,
1148							gdirty_list);
1149		inode = igrab(&fi->vfs_inode);
1150		spin_unlock(&sbi->inode_lock[DIRTY_META]);
1151		if (inode) {
1152			sync_inode_metadata(inode, 0);
1153
1154			/* it's on eviction */
1155			if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1156				f2fs_update_inode_page(inode);
1157			iput(inode);
1158		}
1159	}
1160	return 0;
1161}
1162
1163static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1164{
1165	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1166	struct f2fs_nm_info *nm_i = NM_I(sbi);
1167	nid_t last_nid = nm_i->next_scan_nid;
1168
1169	next_free_nid(sbi, &last_nid);
1170	ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1171	ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1172	ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1173	ckpt->next_free_nid = cpu_to_le32(last_nid);
1174
1175	/* update user_block_counts */
1176	sbi->last_valid_block_count = sbi->total_valid_block_count;
1177	percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1178	percpu_counter_set(&sbi->rf_node_block_count, 0);
1179}
1180
1181static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1182{
1183	bool ret = false;
1184
1185	if (!is_journalled_quota(sbi))
1186		return false;
1187
1188	if (!f2fs_down_write_trylock(&sbi->quota_sem))
1189		return true;
1190	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1191		ret = false;
1192	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1193		ret = false;
1194	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1195		clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1196		ret = true;
1197	} else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1198		ret = true;
1199	}
1200	f2fs_up_write(&sbi->quota_sem);
1201	return ret;
1202}
1203
1204/*
1205 * Freeze all the FS-operations for checkpoint.
1206 */
1207static int block_operations(struct f2fs_sb_info *sbi)
1208{
1209	struct writeback_control wbc = {
1210		.sync_mode = WB_SYNC_ALL,
1211		.nr_to_write = LONG_MAX,
1212	};
1213	int err = 0, cnt = 0;
1214
1215	/*
1216	 * Let's flush inline_data in dirty node pages.
1217	 */
1218	f2fs_flush_inline_data(sbi);
1219
1220retry_flush_quotas:
1221	f2fs_lock_all(sbi);
1222	if (__need_flush_quota(sbi)) {
1223		bool need_lock = sbi->umount_lock_holder != current;
1224
1225		if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1226			set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1227			set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1228			goto retry_flush_dents;
1229		}
1230		f2fs_unlock_all(sbi);
1231
1232		/* don't grab s_umount lock during mount/umount/remount/freeze/quotactl */
1233		if (!need_lock) {
1234			f2fs_do_quota_sync(sbi->sb, -1);
1235		} else if (down_read_trylock(&sbi->sb->s_umount)) {
1236			f2fs_do_quota_sync(sbi->sb, -1);
1237			up_read(&sbi->sb->s_umount);
1238		}
1239		cond_resched();
1240		goto retry_flush_quotas;
1241	}
1242
1243retry_flush_dents:
1244	/* write all the dirty dentry pages */
1245	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1246		f2fs_unlock_all(sbi);
1247		err = f2fs_sync_dirty_inodes(sbi, DIR_INODE, true);
1248		if (err)
1249			return err;
1250		cond_resched();
1251		goto retry_flush_quotas;
1252	}
1253
1254	/*
1255	 * POR: we should ensure that there are no dirty node pages
1256	 * until finishing nat/sit flush. inode->i_blocks can be updated.
1257	 */
1258	f2fs_down_write(&sbi->node_change);
1259
1260	if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1261		f2fs_up_write(&sbi->node_change);
1262		f2fs_unlock_all(sbi);
1263		err = f2fs_sync_inode_meta(sbi);
1264		if (err)
1265			return err;
1266		cond_resched();
1267		goto retry_flush_quotas;
1268	}
1269
1270retry_flush_nodes:
1271	f2fs_down_write(&sbi->node_write);
1272
1273	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1274		f2fs_up_write(&sbi->node_write);
1275		atomic_inc(&sbi->wb_sync_req[NODE]);
1276		err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1277		atomic_dec(&sbi->wb_sync_req[NODE]);
1278		if (err) {
1279			f2fs_up_write(&sbi->node_change);
1280			f2fs_unlock_all(sbi);
1281			return err;
1282		}
1283		cond_resched();
1284		goto retry_flush_nodes;
1285	}
1286
1287	/*
1288	 * sbi->node_change is used only for AIO write_begin path which produces
1289	 * dirty node blocks and some checkpoint values by block allocation.
1290	 */
1291	__prepare_cp_block(sbi);
1292	f2fs_up_write(&sbi->node_change);
1293	return err;
1294}
1295
1296static void unblock_operations(struct f2fs_sb_info *sbi)
1297{
1298	f2fs_up_write(&sbi->node_write);
1299	f2fs_unlock_all(sbi);
1300}
1301
1302void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1303{
1304	DEFINE_WAIT(wait);
1305
1306	for (;;) {
1307		if (!get_pages(sbi, type))
1308			break;
1309
1310		if (unlikely(f2fs_cp_error(sbi) &&
1311			!is_sbi_flag_set(sbi, SBI_IS_CLOSE)))
1312			break;
1313
1314		if (type == F2FS_DIRTY_META)
1315			f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1316							FS_CP_META_IO);
1317		else if (type == F2FS_WB_CP_DATA)
1318			f2fs_submit_merged_write(sbi, DATA);
1319
1320		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1321		io_schedule_timeout(DEFAULT_SCHEDULE_TIMEOUT);
1322	}
1323	finish_wait(&sbi->cp_wait, &wait);
1324}
1325
1326static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1327{
1328	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1329	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1330	unsigned long flags;
1331
1332	spin_lock_irqsave(&sbi->cp_lock, flags);
1333
1334	if ((cpc->reason & CP_UMOUNT) &&
1335			le32_to_cpu(ckpt->cp_pack_total_block_count) >
1336			sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1337		disable_nat_bits(sbi, false);
1338
1339	if (cpc->reason & CP_TRIMMED)
1340		__set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1341	else
1342		__clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1343
1344	if (cpc->reason & CP_UMOUNT)
1345		__set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1346	else
1347		__clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1348
1349	if (cpc->reason & CP_FASTBOOT)
1350		__set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1351	else
1352		__clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1353
1354	if (orphan_num)
1355		__set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1356	else
1357		__clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1358
1359	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1360		__set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1361
1362	if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1363		__set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1364	else
1365		__clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1366
1367	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1368		__set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1369	else
1370		__clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1371
1372	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1373		__set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1374	else
1375		__clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1376
1377	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1378		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1379	else
1380		__clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1381
1382	if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1383		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1384
1385	/* set this flag to activate crc|cp_ver for recovery */
1386	__set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1387	__clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1388
1389	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1390}
1391
1392static void commit_checkpoint(struct f2fs_sb_info *sbi,
1393	void *src, block_t blk_addr)
1394{
1395	struct writeback_control wbc = {};
1396
1397	/*
1398	 * filemap_get_folios_tag and folio_lock again will take
1399	 * some extra time. Therefore, f2fs_update_meta_pages and
1400	 * f2fs_sync_meta_pages are combined in this function.
1401	 */
1402	struct folio *folio = f2fs_grab_meta_folio(sbi, blk_addr);
1403
1404	memcpy(folio_address(folio), src, PAGE_SIZE);
1405
1406	folio_mark_dirty(folio);
1407	if (unlikely(!folio_clear_dirty_for_io(folio)))
1408		f2fs_bug_on(sbi, 1);
1409
1410	/* writeout cp pack 2 page */
1411	if (unlikely(!__f2fs_write_meta_folio(folio, &wbc, FS_CP_META_IO))) {
1412		if (f2fs_cp_error(sbi)) {
1413			f2fs_folio_put(folio, true);
1414			return;
1415		}
1416		f2fs_bug_on(sbi, true);
1417	}
1418
1419	f2fs_folio_put(folio, false);
1420
1421	/* submit checkpoint (with barrier if NOBARRIER is not set) */
1422	f2fs_submit_merged_write(sbi, META_FLUSH);
1423}
1424
1425static inline u64 get_sectors_written(struct block_device *bdev)
1426{
1427	return (u64)part_stat_read(bdev, sectors[STAT_WRITE]);
1428}
1429
1430u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1431{
1432	if (f2fs_is_multi_device(sbi)) {
1433		u64 sectors = 0;
1434		int i;
1435
1436		for (i = 0; i < sbi->s_ndevs; i++)
1437			sectors += get_sectors_written(FDEV(i).bdev);
1438
1439		return sectors;
1440	}
1441
1442	return get_sectors_written(sbi->sb->s_bdev);
1443}
1444
1445static inline void stat_cp_time(struct cp_control *cpc, enum cp_time type)
1446{
1447	cpc->stats.times[type] = ktime_get();
1448}
1449
1450static inline void check_cp_time(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1451{
1452	unsigned long long sb_diff, cur_diff;
1453	enum cp_time ct;
1454
1455	sb_diff = (u64)ktime_ms_delta(sbi->cp_stats.times[CP_TIME_END],
1456					sbi->cp_stats.times[CP_TIME_START]);
1457	cur_diff = (u64)ktime_ms_delta(cpc->stats.times[CP_TIME_END],
1458					cpc->stats.times[CP_TIME_START]);
1459
1460	if (cur_diff > sb_diff) {
1461		sbi->cp_stats = cpc->stats;
1462		if (cur_diff < CP_LONG_LATENCY_THRESHOLD)
1463			return;
1464
1465		f2fs_warn(sbi, "checkpoint was blocked for %llu ms", cur_diff);
1466		for (ct = CP_TIME_START; ct < CP_TIME_MAX - 1; ct++)
1467			f2fs_warn(sbi, "Step#%d: %llu ms", ct,
1468				(u64)ktime_ms_delta(cpc->stats.times[ct + 1],
1469						cpc->stats.times[ct]));
1470	}
1471}
1472
1473static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1474{
1475	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1476	struct f2fs_nm_info *nm_i = NM_I(sbi);
1477	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1478	block_t start_blk;
1479	unsigned int data_sum_blocks, orphan_blocks;
1480	__u32 crc32 = 0;
1481	int i;
1482	int cp_payload_blks = __cp_payload(sbi);
1483	struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1484	u64 kbytes_written;
1485	int err;
1486
1487	/* Flush all the NAT/SIT pages */
1488	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1489
1490	stat_cp_time(cpc, CP_TIME_SYNC_META);
1491
1492	/* start to update checkpoint, cp ver is already updated previously */
1493	ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1494	ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1495	for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1496		struct curseg_info *curseg = CURSEG_I(sbi, i + CURSEG_HOT_NODE);
1497
1498		ckpt->cur_node_segno[i] = cpu_to_le32(curseg->segno);
1499		ckpt->cur_node_blkoff[i] = cpu_to_le16(curseg->next_blkoff);
1500		ckpt->alloc_type[i + CURSEG_HOT_NODE] = curseg->alloc_type;
1501	}
1502	for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1503		struct curseg_info *curseg = CURSEG_I(sbi, i + CURSEG_HOT_DATA);
1504
1505		ckpt->cur_data_segno[i] = cpu_to_le32(curseg->segno);
1506		ckpt->cur_data_blkoff[i] = cpu_to_le16(curseg->next_blkoff);
1507		ckpt->alloc_type[i + CURSEG_HOT_DATA] = curseg->alloc_type;
1508	}
1509
1510	/* 2 cp + n data seg summary + orphan inode blocks */
1511	data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1512	spin_lock_irqsave(&sbi->cp_lock, flags);
1513	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1514		__set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1515	else
1516		__clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1517	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1518
1519	orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1520	ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1521			orphan_blocks);
1522
1523	if (__remain_node_summaries(cpc->reason))
1524		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1525				cp_payload_blks + data_sum_blocks +
1526				orphan_blocks + NR_CURSEG_NODE_TYPE);
1527	else
1528		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1529				cp_payload_blks + data_sum_blocks +
1530				orphan_blocks);
1531
1532	/* update ckpt flag for checkpoint */
1533	update_ckpt_flags(sbi, cpc);
1534
1535	/* update SIT/NAT bitmap */
1536	get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1537	get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1538
1539	crc32 = f2fs_checkpoint_chksum(ckpt);
1540	*((__le32 *)((unsigned char *)ckpt +
1541				le32_to_cpu(ckpt->checksum_offset)))
1542				= cpu_to_le32(crc32);
1543
1544	start_blk = __start_cp_next_addr(sbi);
1545
1546	/* write nat bits */
1547	if (enabled_nat_bits(sbi, cpc)) {
1548		__u64 cp_ver = cur_cp_version(ckpt);
1549		block_t blk;
1550
1551		cp_ver |= ((__u64)crc32 << 32);
1552		*(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1553
1554		blk = start_blk + BLKS_PER_SEG(sbi) - nm_i->nat_bits_blocks;
1555		for (i = 0; i < nm_i->nat_bits_blocks; i++)
1556			f2fs_update_meta_page(sbi, nm_i->nat_bits +
1557					F2FS_BLK_TO_BYTES(i), blk + i);
1558	}
1559
1560	/* write out checkpoint buffer at block 0 */
1561	f2fs_update_meta_page(sbi, ckpt, start_blk++);
1562
1563	for (i = 1; i < 1 + cp_payload_blks; i++)
1564		f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1565							start_blk++);
1566
1567	if (orphan_num) {
1568		write_orphan_inodes(sbi, start_blk);
1569		start_blk += orphan_blocks;
1570	}
1571
1572	f2fs_write_data_summaries(sbi, start_blk);
1573	start_blk += data_sum_blocks;
1574
1575	/* Record write statistics in the hot node summary */
1576	kbytes_written = sbi->kbytes_written;
1577	kbytes_written += (f2fs_get_sectors_written(sbi) -
1578				sbi->sectors_written_start) >> 1;
1579	seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1580
1581	if (__remain_node_summaries(cpc->reason)) {
1582		f2fs_write_node_summaries(sbi, start_blk);
1583		start_blk += NR_CURSEG_NODE_TYPE;
1584	}
1585
1586	/* Here, we have one bio having CP pack except cp pack 2 page */
1587	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1588	stat_cp_time(cpc, CP_TIME_SYNC_CP_META);
1589
1590	/* Wait for all dirty meta pages to be submitted for IO */
1591	f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1592	stat_cp_time(cpc, CP_TIME_WAIT_DIRTY_META);
1593
1594	/* wait for previous submitted meta pages writeback */
1595	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1596	stat_cp_time(cpc, CP_TIME_WAIT_CP_DATA);
1597
1598	/* flush all device cache */
1599	err = f2fs_flush_device_cache(sbi);
1600	if (err)
1601		return err;
1602	stat_cp_time(cpc, CP_TIME_FLUSH_DEVICE);
1603
1604	/* barrier and flush checkpoint cp pack 2 page if it can */
1605	commit_checkpoint(sbi, ckpt, start_blk);
1606	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1607	stat_cp_time(cpc, CP_TIME_WAIT_LAST_CP);
1608
1609	/*
1610	 * invalidate intermediate page cache borrowed from meta inode which are
1611	 * used for migration of encrypted, verity or compressed inode's blocks.
1612	 */
1613	if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1614		f2fs_sb_has_compression(sbi))
1615		f2fs_bug_on(sbi,
1616			invalidate_inode_pages2_range(META_MAPPING(sbi),
1617				MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1));
1618
1619	f2fs_release_ino_entry(sbi, false);
1620
1621	f2fs_reset_fsync_node_info(sbi);
1622
1623	clear_sbi_flag(sbi, SBI_IS_DIRTY);
1624	clear_sbi_flag(sbi, SBI_NEED_CP);
1625	clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1626
1627	spin_lock(&sbi->stat_lock);
1628	sbi->unusable_block_count = 0;
1629	spin_unlock(&sbi->stat_lock);
1630
1631	__set_cp_next_pack(sbi);
1632
1633	/*
1634	 * redirty superblock if metadata like node page or inode cache is
1635	 * updated during writing checkpoint.
1636	 */
1637	if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1638			get_pages(sbi, F2FS_DIRTY_IMETA))
1639		set_sbi_flag(sbi, SBI_IS_DIRTY);
1640
1641	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1642
1643	return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1644}
1645
1646int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1647{
1648	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1649	unsigned long long ckpt_ver;
1650	int err = 0;
1651
1652	stat_cp_time(cpc, CP_TIME_START);
1653
1654	if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1655		return -EROFS;
1656
1657	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1658		if (cpc->reason != CP_PAUSE)
1659			return 0;
1660		f2fs_warn(sbi, "Start checkpoint disabled!");
1661	}
1662	if (cpc->reason != CP_RESIZE)
1663		f2fs_down_write(&sbi->cp_global_sem);
1664
1665	stat_cp_time(cpc, CP_TIME_LOCK);
1666
1667	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1668		((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1669		((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1670		goto out;
1671	if (unlikely(f2fs_cp_error(sbi))) {
1672		err = -EIO;
1673		goto out;
1674	}
1675
1676	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, CP_PHASE_START_BLOCK_OPS);
1677
1678	err = block_operations(sbi);
1679	if (err)
1680		goto out;
1681
1682	stat_cp_time(cpc, CP_TIME_OP_LOCK);
1683
1684	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, CP_PHASE_FINISH_BLOCK_OPS);
1685
1686	f2fs_flush_merged_writes(sbi);
1687
1688	/* this is the case of multiple fstrims without any changes */
1689	if (cpc->reason & CP_DISCARD) {
1690		if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1691			unblock_operations(sbi);
1692			goto out;
1693		}
1694
1695		if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1696				SIT_I(sbi)->dirty_sentries == 0 &&
1697				prefree_segments(sbi) == 0) {
1698			f2fs_flush_sit_entries(sbi, cpc);
1699			f2fs_clear_prefree_segments(sbi, cpc);
1700			unblock_operations(sbi);
1701			goto out;
1702		}
1703	}
1704
1705	/*
1706	 * update checkpoint pack index
1707	 * Increase the version number so that
1708	 * SIT entries and seg summaries are written at correct place
1709	 */
1710	ckpt_ver = cur_cp_version(ckpt);
1711	ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1712
1713	/* write cached NAT/SIT entries to NAT/SIT area */
1714	err = f2fs_flush_nat_entries(sbi, cpc);
1715	if (err) {
1716		f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err);
1717		f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1718		goto stop;
1719	}
1720
1721	f2fs_flush_sit_entries(sbi, cpc);
1722
1723	stat_cp_time(cpc, CP_TIME_FLUSH_META);
1724
1725	/* save inmem log status */
1726	f2fs_save_inmem_curseg(sbi);
1727
1728	err = do_checkpoint(sbi, cpc);
1729	if (err) {
1730		f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err);
1731		f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1732		f2fs_release_discard_addrs(sbi);
1733	} else {
1734		f2fs_clear_prefree_segments(sbi, cpc);
1735	}
1736
1737	f2fs_restore_inmem_curseg(sbi);
1738	f2fs_reinit_atgc_curseg(sbi);
1739	stat_inc_cp_count(sbi);
1740stop:
1741	unblock_operations(sbi);
1742	stat_cp_time(cpc, CP_TIME_END);
1743	check_cp_time(sbi, cpc);
1744
1745	if (cpc->reason & CP_RECOVERY)
1746		f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1747
1748	/* update CP_TIME to trigger checkpoint periodically */
1749	f2fs_update_time(sbi, CP_TIME);
1750	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, CP_PHASE_FINISH_CHECKPOINT);
1751out:
1752	if (cpc->reason != CP_RESIZE)
1753		f2fs_up_write(&sbi->cp_global_sem);
1754	return err;
1755}
1756
1757void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1758{
1759	int i;
1760
1761	for (i = 0; i < MAX_INO_ENTRY; i++) {
1762		struct inode_management *im = &sbi->im[i];
1763
1764		INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1765		spin_lock_init(&im->ino_lock);
1766		INIT_LIST_HEAD(&im->ino_list);
1767		im->ino_num = 0;
1768	}
1769
1770	sbi->max_orphans = (BLKS_PER_SEG(sbi) - F2FS_CP_PACKS -
1771			NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1772			F2FS_ORPHANS_PER_BLOCK;
1773}
1774
1775int __init f2fs_create_checkpoint_caches(void)
1776{
1777	ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1778			sizeof(struct ino_entry));
1779	if (!ino_entry_slab)
1780		return -ENOMEM;
1781	f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1782			sizeof(struct inode_entry));
1783	if (!f2fs_inode_entry_slab) {
1784		kmem_cache_destroy(ino_entry_slab);
1785		return -ENOMEM;
1786	}
1787	return 0;
1788}
1789
1790void f2fs_destroy_checkpoint_caches(void)
1791{
1792	kmem_cache_destroy(ino_entry_slab);
1793	kmem_cache_destroy(f2fs_inode_entry_slab);
1794}
1795
1796static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1797{
1798	struct cp_control cpc = { .reason = CP_SYNC, };
1799	int err;
1800
1801	f2fs_down_write(&sbi->gc_lock);
1802	err = f2fs_write_checkpoint(sbi, &cpc);
1803	f2fs_up_write(&sbi->gc_lock);
1804
1805	return err;
1806}
1807
1808static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1809{
1810	struct ckpt_req_control *cprc = &sbi->cprc_info;
1811	struct ckpt_req *req, *next;
1812	struct llist_node *dispatch_list;
1813	u64 sum_diff = 0, diff, count = 0;
1814	int ret;
1815
1816	dispatch_list = llist_del_all(&cprc->issue_list);
1817	if (!dispatch_list)
1818		return;
1819	dispatch_list = llist_reverse_order(dispatch_list);
1820
1821	ret = __write_checkpoint_sync(sbi);
1822	atomic_inc(&cprc->issued_ckpt);
1823
1824	llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1825		diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1826		req->ret = ret;
1827		req->delta_time = diff;
1828		complete(&req->wait);
1829
1830		sum_diff += diff;
1831		count++;
1832	}
1833	atomic_sub(count, &cprc->queued_ckpt);
1834	atomic_add(count, &cprc->total_ckpt);
1835
1836	spin_lock(&cprc->stat_lock);
1837	cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1838	if (cprc->peak_time < cprc->cur_time)
1839		cprc->peak_time = cprc->cur_time;
1840	spin_unlock(&cprc->stat_lock);
1841}
1842
1843static int issue_checkpoint_thread(void *data)
1844{
1845	struct f2fs_sb_info *sbi = data;
1846	struct ckpt_req_control *cprc = &sbi->cprc_info;
1847	wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1848repeat:
1849	if (kthread_should_stop())
1850		return 0;
1851
1852	if (!llist_empty(&cprc->issue_list))
1853		__checkpoint_and_complete_reqs(sbi);
1854
1855	wait_event_interruptible(*q,
1856		kthread_should_stop() || !llist_empty(&cprc->issue_list));
1857	goto repeat;
1858}
1859
1860static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1861		struct ckpt_req *wait_req)
1862{
1863	struct ckpt_req_control *cprc = &sbi->cprc_info;
1864
1865	if (!llist_empty(&cprc->issue_list)) {
1866		__checkpoint_and_complete_reqs(sbi);
1867	} else {
1868		/* already dispatched by issue_checkpoint_thread */
1869		if (wait_req)
1870			wait_for_completion(&wait_req->wait);
1871	}
1872}
1873
1874static void init_ckpt_req(struct ckpt_req *req)
1875{
1876	memset(req, 0, sizeof(struct ckpt_req));
1877
1878	init_completion(&req->wait);
1879	req->queue_time = ktime_get();
1880}
1881
1882int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1883{
1884	struct ckpt_req_control *cprc = &sbi->cprc_info;
1885	struct ckpt_req req;
1886	struct cp_control cpc;
1887
1888	cpc.reason = __get_cp_reason(sbi);
1889	if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC ||
1890		sbi->umount_lock_holder == current) {
1891		int ret;
1892
1893		f2fs_down_write(&sbi->gc_lock);
1894		ret = f2fs_write_checkpoint(sbi, &cpc);
1895		f2fs_up_write(&sbi->gc_lock);
1896
1897		return ret;
1898	}
1899
1900	if (!cprc->f2fs_issue_ckpt)
1901		return __write_checkpoint_sync(sbi);
1902
1903	init_ckpt_req(&req);
1904
1905	llist_add(&req.llnode, &cprc->issue_list);
1906	atomic_inc(&cprc->queued_ckpt);
1907
1908	/*
1909	 * update issue_list before we wake up issue_checkpoint thread,
1910	 * this smp_mb() pairs with another barrier in ___wait_event(),
1911	 * see more details in comments of waitqueue_active().
1912	 */
1913	smp_mb();
1914
1915	if (waitqueue_active(&cprc->ckpt_wait_queue))
1916		wake_up(&cprc->ckpt_wait_queue);
1917
1918	if (cprc->f2fs_issue_ckpt)
1919		wait_for_completion(&req.wait);
1920	else
1921		flush_remained_ckpt_reqs(sbi, &req);
1922
1923	if (unlikely(req.delta_time >= CP_LONG_LATENCY_THRESHOLD)) {
1924		f2fs_warn_ratelimited(sbi,
1925			"blocked on checkpoint for %u ms", cprc->peak_time);
1926		dump_stack();
1927	}
1928
1929	return req.ret;
1930}
1931
1932int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1933{
1934	dev_t dev = sbi->sb->s_bdev->bd_dev;
1935	struct ckpt_req_control *cprc = &sbi->cprc_info;
1936
1937	if (cprc->f2fs_issue_ckpt)
1938		return 0;
1939
1940	cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1941			"f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1942	if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1943		int err = PTR_ERR(cprc->f2fs_issue_ckpt);
1944
1945		cprc->f2fs_issue_ckpt = NULL;
1946		return err;
1947	}
1948
1949	set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1950
1951	return 0;
1952}
1953
1954void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1955{
1956	struct ckpt_req_control *cprc = &sbi->cprc_info;
1957	struct task_struct *ckpt_task;
1958
1959	if (!cprc->f2fs_issue_ckpt)
1960		return;
1961
1962	ckpt_task = cprc->f2fs_issue_ckpt;
1963	cprc->f2fs_issue_ckpt = NULL;
1964	kthread_stop(ckpt_task);
1965
1966	f2fs_flush_ckpt_thread(sbi);
1967}
1968
1969void f2fs_flush_ckpt_thread(struct f2fs_sb_info *sbi)
1970{
1971	struct ckpt_req_control *cprc = &sbi->cprc_info;
1972
1973	flush_remained_ckpt_reqs(sbi, NULL);
1974
1975	/* Let's wait for the previous dispatched checkpoint. */
1976	while (atomic_read(&cprc->queued_ckpt))
1977		io_schedule_timeout(DEFAULT_SCHEDULE_TIMEOUT);
1978}
1979
1980void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1981{
1982	struct ckpt_req_control *cprc = &sbi->cprc_info;
1983
1984	atomic_set(&cprc->issued_ckpt, 0);
1985	atomic_set(&cprc->total_ckpt, 0);
1986	atomic_set(&cprc->queued_ckpt, 0);
1987	cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1988	init_waitqueue_head(&cprc->ckpt_wait_queue);
1989	init_llist_head(&cprc->issue_list);
1990	spin_lock_init(&cprc->stat_lock);
1991}