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_BE, 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);
  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 page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
  42{
  43	struct address_space *mapping = META_MAPPING(sbi);
  44	struct page *page;
  45repeat:
  46	page = f2fs_grab_cache_page(mapping, index, false);
  47	if (!page) {
  48		cond_resched();
  49		goto repeat;
  50	}
  51	f2fs_wait_on_page_writeback(page, META, true, true);
  52	if (!PageUptodate(page))
  53		SetPageUptodate(page);
  54	return page;
  55}
  56
  57static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
  58							bool is_meta)
  59{
  60	struct address_space *mapping = META_MAPPING(sbi);
  61	struct page *page;
  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	page = f2fs_grab_cache_page(mapping, index, false);
  78	if (!page) {
  79		cond_resched();
  80		goto repeat;
  81	}
  82	if (PageUptodate(page))
  83		goto out;
  84
  85	fio.page = page;
  86
  87	err = f2fs_submit_page_bio(&fio);
  88	if (err) {
  89		f2fs_put_page(page, 1);
  90		return ERR_PTR(err);
  91	}
  92
  93	f2fs_update_iostat(sbi, NULL, FS_META_READ_IO, F2FS_BLKSIZE);
  94
  95	lock_page(page);
  96	if (unlikely(page->mapping != mapping)) {
  97		f2fs_put_page(page, 1);
  98		goto repeat;
  99	}
 100
 101	if (unlikely(!PageUptodate(page))) {
 102		f2fs_handle_page_eio(sbi, page_folio(page), META);
 103		f2fs_put_page(page, 1);
 104		return ERR_PTR(-EIO);
 105	}
 106out:
 107	return page;
 108}
 109
 110struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
 111{
 112	return __get_meta_page(sbi, index, true);
 113}
 114
 115struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index)
 116{
 117	struct page *page;
 118	int count = 0;
 119
 120retry:
 121	page = __get_meta_page(sbi, index, true);
 122	if (IS_ERR(page)) {
 123		if (PTR_ERR(page) == -EIO &&
 124				++count <= DEFAULT_RETRY_IO_COUNT)
 125			goto retry;
 126		f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_META_PAGE);
 127	}
 128	return page;
 129}
 130
 131/* for POR only */
 132struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
 133{
 134	return __get_meta_page(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	struct page *page;
 256	block_t blkno = start;
 257	struct f2fs_io_info fio = {
 258		.sbi = sbi,
 259		.type = META,
 260		.op = REQ_OP_READ,
 261		.op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
 262		.encrypted_page = NULL,
 263		.in_list = 0,
 264		.is_por = (type == META_POR) ? 1 : 0,
 265	};
 266	struct blk_plug plug;
 267	int err;
 268
 269	if (unlikely(type == META_POR))
 270		fio.op_flags &= ~REQ_META;
 271
 272	blk_start_plug(&plug);
 273	for (; nrpages-- > 0; blkno++) {
 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		page = f2fs_grab_cache_page(META_MAPPING(sbi),
 304						fio.new_blkaddr, false);
 305		if (!page)
 306			continue;
 307		if (PageUptodate(page)) {
 308			f2fs_put_page(page, 1);
 309			continue;
 310		}
 311
 312		fio.page = page;
 313		err = f2fs_submit_page_bio(&fio);
 314		f2fs_put_page(page, err ? 1 : 0);
 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 page *page;
 329	bool readahead = false;
 330
 331	if (ra_blocks == RECOVERY_MIN_RA_BLOCKS)
 332		return;
 333
 334	page = find_get_page(META_MAPPING(sbi), index);
 335	if (!page || !PageUptodate(page))
 336		readahead = true;
 337	f2fs_put_page(page, 0);
 338
 339	if (readahead)
 340		f2fs_ra_meta_pages(sbi, index, ra_blocks, META_POR, true);
 341}
 342
 343static int __f2fs_write_meta_page(struct page *page,
 344				struct writeback_control *wbc,
 345				enum iostat_type io_type)
 346{
 347	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
 348	struct folio *folio = page_folio(page);
 349
 350	trace_f2fs_writepage(folio, META);
 351
 352	if (unlikely(f2fs_cp_error(sbi))) {
 353		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
 354			folio_clear_uptodate(folio);
 355			dec_page_count(sbi, F2FS_DIRTY_META);
 356			folio_unlock(folio);
 357			return 0;
 358		}
 359		goto redirty_out;
 360	}
 361	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 362		goto redirty_out;
 363	if (wbc->for_reclaim && folio->index < GET_SUM_BLOCK(sbi, 0))
 364		goto redirty_out;
 365
 366	f2fs_do_write_meta_page(sbi, folio, io_type);
 367	dec_page_count(sbi, F2FS_DIRTY_META);
 368
 369	if (wbc->for_reclaim)
 370		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
 371
 372	folio_unlock(folio);
 373
 374	if (unlikely(f2fs_cp_error(sbi)))
 375		f2fs_submit_merged_write(sbi, META);
 376
 377	return 0;
 378
 379redirty_out:
 380	redirty_page_for_writepage(wbc, page);
 381	return AOP_WRITEPAGE_ACTIVATE;
 382}
 383
 384static int f2fs_write_meta_page(struct page *page,
 385				struct writeback_control *wbc)
 386{
 387	return __f2fs_write_meta_page(page, wbc, FS_META_IO);
 388}
 389
 390static int f2fs_write_meta_pages(struct address_space *mapping,
 391				struct writeback_control *wbc)
 392{
 393	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
 394	long diff, written;
 395
 396	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
 397		goto skip_write;
 398
 399	/* collect a number of dirty meta pages and write together */
 400	if (wbc->sync_mode != WB_SYNC_ALL &&
 401			get_pages(sbi, F2FS_DIRTY_META) <
 402					nr_pages_to_skip(sbi, META))
 403		goto skip_write;
 404
 405	/* if locked failed, cp will flush dirty pages instead */
 406	if (!f2fs_down_write_trylock(&sbi->cp_global_sem))
 407		goto skip_write;
 408
 409	trace_f2fs_writepages(mapping->host, wbc, META);
 410	diff = nr_pages_to_write(sbi, META, wbc);
 411	written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
 412	f2fs_up_write(&sbi->cp_global_sem);
 413	wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
 414	return 0;
 415
 416skip_write:
 417	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
 418	trace_f2fs_writepages(mapping->host, wbc, META);
 419	return 0;
 420}
 421
 422long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
 423				long nr_to_write, enum iostat_type io_type)
 424{
 425	struct address_space *mapping = META_MAPPING(sbi);
 426	pgoff_t index = 0, prev = ULONG_MAX;
 427	struct folio_batch fbatch;
 428	long nwritten = 0;
 429	int nr_folios;
 430	struct writeback_control wbc = {
 431		.for_reclaim = 0,
 432	};
 433	struct blk_plug plug;
 434
 435	folio_batch_init(&fbatch);
 436
 437	blk_start_plug(&plug);
 438
 439	while ((nr_folios = filemap_get_folios_tag(mapping, &index,
 440					(pgoff_t)-1,
 441					PAGECACHE_TAG_DIRTY, &fbatch))) {
 442		int i;
 443
 444		for (i = 0; i < nr_folios; i++) {
 445			struct folio *folio = fbatch.folios[i];
 446
 447			if (nr_to_write != LONG_MAX && i != 0 &&
 448					folio->index != prev +
 449					folio_nr_pages(fbatch.folios[i-1])) {
 450				folio_batch_release(&fbatch);
 451				goto stop;
 452			}
 453
 454			folio_lock(folio);
 455
 456			if (unlikely(folio->mapping != mapping)) {
 457continue_unlock:
 458				folio_unlock(folio);
 459				continue;
 460			}
 461			if (!folio_test_dirty(folio)) {
 462				/* someone wrote it for us */
 463				goto continue_unlock;
 464			}
 465
 466			f2fs_wait_on_page_writeback(&folio->page, META,
 467					true, true);
 468
 469			if (!folio_clear_dirty_for_io(folio))
 470				goto continue_unlock;
 471
 472			if (__f2fs_write_meta_page(&folio->page, &wbc,
 473						io_type)) {
 474				folio_unlock(folio);
 475				break;
 476			}
 477			nwritten += folio_nr_pages(folio);
 478			prev = folio->index;
 479			if (unlikely(nwritten >= nr_to_write))
 480				break;
 481		}
 482		folio_batch_release(&fbatch);
 483		cond_resched();
 484	}
 485stop:
 486	if (nwritten)
 487		f2fs_submit_merged_write(sbi, type);
 488
 489	blk_finish_plug(&plug);
 490
 491	return nwritten;
 492}
 493
 494static bool f2fs_dirty_meta_folio(struct address_space *mapping,
 495		struct folio *folio)
 496{
 497	trace_f2fs_set_page_dirty(folio, META);
 498
 499	if (!folio_test_uptodate(folio))
 500		folio_mark_uptodate(folio);
 501	if (filemap_dirty_folio(mapping, folio)) {
 502		inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_META);
 503		set_page_private_reference(&folio->page);
 504		return true;
 505	}
 506	return false;
 507}
 508
 509const struct address_space_operations f2fs_meta_aops = {
 510	.writepage	= f2fs_write_meta_page,
 511	.writepages	= f2fs_write_meta_pages,
 512	.dirty_folio	= f2fs_dirty_meta_folio,
 513	.invalidate_folio = f2fs_invalidate_folio,
 514	.release_folio	= f2fs_release_folio,
 515	.migrate_folio	= filemap_migrate_folio,
 516};
 517
 518static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
 519						unsigned int devidx, int type)
 520{
 521	struct inode_management *im = &sbi->im[type];
 522	struct ino_entry *e = NULL, *new = NULL;
 523
 524	if (type == FLUSH_INO) {
 525		rcu_read_lock();
 526		e = radix_tree_lookup(&im->ino_root, ino);
 527		rcu_read_unlock();
 528	}
 529
 530retry:
 531	if (!e)
 532		new = f2fs_kmem_cache_alloc(ino_entry_slab,
 533						GFP_NOFS, true, NULL);
 534
 535	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
 536
 537	spin_lock(&im->ino_lock);
 538	e = radix_tree_lookup(&im->ino_root, ino);
 539	if (!e) {
 540		if (!new) {
 541			spin_unlock(&im->ino_lock);
 542			radix_tree_preload_end();
 543			goto retry;
 544		}
 545		e = new;
 546		if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
 547			f2fs_bug_on(sbi, 1);
 548
 549		memset(e, 0, sizeof(struct ino_entry));
 550		e->ino = ino;
 551
 552		list_add_tail(&e->list, &im->ino_list);
 553		if (type != ORPHAN_INO)
 554			im->ino_num++;
 555	}
 556
 557	if (type == FLUSH_INO)
 558		f2fs_set_bit(devidx, (char *)&e->dirty_device);
 559
 560	spin_unlock(&im->ino_lock);
 561	radix_tree_preload_end();
 562
 563	if (new && e != new)
 564		kmem_cache_free(ino_entry_slab, new);
 565}
 566
 567static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 568{
 569	struct inode_management *im = &sbi->im[type];
 570	struct ino_entry *e;
 571
 572	spin_lock(&im->ino_lock);
 573	e = radix_tree_lookup(&im->ino_root, ino);
 574	if (e) {
 575		list_del(&e->list);
 576		radix_tree_delete(&im->ino_root, ino);
 577		im->ino_num--;
 578		spin_unlock(&im->ino_lock);
 579		kmem_cache_free(ino_entry_slab, e);
 580		return;
 581	}
 582	spin_unlock(&im->ino_lock);
 583}
 584
 585void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 586{
 587	/* add new dirty ino entry into list */
 588	__add_ino_entry(sbi, ino, 0, type);
 589}
 590
 591void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
 592{
 593	/* remove dirty ino entry from list */
 594	__remove_ino_entry(sbi, ino, type);
 595}
 596
 597/* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
 598bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
 599{
 600	struct inode_management *im = &sbi->im[mode];
 601	struct ino_entry *e;
 602
 603	spin_lock(&im->ino_lock);
 604	e = radix_tree_lookup(&im->ino_root, ino);
 605	spin_unlock(&im->ino_lock);
 606	return e ? true : false;
 607}
 608
 609void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
 610{
 611	struct ino_entry *e, *tmp;
 612	int i;
 613
 614	for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
 615		struct inode_management *im = &sbi->im[i];
 616
 617		spin_lock(&im->ino_lock);
 618		list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
 619			list_del(&e->list);
 620			radix_tree_delete(&im->ino_root, e->ino);
 621			kmem_cache_free(ino_entry_slab, e);
 622			im->ino_num--;
 623		}
 624		spin_unlock(&im->ino_lock);
 625	}
 626}
 627
 628void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 629					unsigned int devidx, int type)
 630{
 631	__add_ino_entry(sbi, ino, devidx, type);
 632}
 633
 634bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
 635					unsigned int devidx, int type)
 636{
 637	struct inode_management *im = &sbi->im[type];
 638	struct ino_entry *e;
 639	bool is_dirty = false;
 640
 641	spin_lock(&im->ino_lock);
 642	e = radix_tree_lookup(&im->ino_root, ino);
 643	if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
 644		is_dirty = true;
 645	spin_unlock(&im->ino_lock);
 646	return is_dirty;
 647}
 648
 649int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
 650{
 651	struct inode_management *im = &sbi->im[ORPHAN_INO];
 652	int err = 0;
 653
 654	spin_lock(&im->ino_lock);
 655
 656	if (time_to_inject(sbi, FAULT_ORPHAN)) {
 657		spin_unlock(&im->ino_lock);
 658		return -ENOSPC;
 659	}
 660
 661	if (unlikely(im->ino_num >= sbi->max_orphans))
 662		err = -ENOSPC;
 663	else
 664		im->ino_num++;
 665	spin_unlock(&im->ino_lock);
 666
 667	return err;
 668}
 669
 670void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
 671{
 672	struct inode_management *im = &sbi->im[ORPHAN_INO];
 673
 674	spin_lock(&im->ino_lock);
 675	f2fs_bug_on(sbi, im->ino_num == 0);
 676	im->ino_num--;
 677	spin_unlock(&im->ino_lock);
 678}
 679
 680void f2fs_add_orphan_inode(struct inode *inode)
 681{
 682	/* add new orphan ino entry into list */
 683	__add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
 684	f2fs_update_inode_page(inode);
 685}
 686
 687void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 688{
 689	/* remove orphan entry from orphan list */
 690	__remove_ino_entry(sbi, ino, ORPHAN_INO);
 691}
 692
 693static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
 694{
 695	struct inode *inode;
 696	struct node_info ni;
 697	int err;
 698
 699	inode = f2fs_iget_retry(sbi->sb, ino);
 700	if (IS_ERR(inode)) {
 701		/*
 702		 * there should be a bug that we can't find the entry
 703		 * to orphan inode.
 704		 */
 705		f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
 706		return PTR_ERR(inode);
 707	}
 708
 709	err = f2fs_dquot_initialize(inode);
 710	if (err) {
 711		iput(inode);
 712		goto err_out;
 713	}
 714
 715	clear_nlink(inode);
 716
 717	/* truncate all the data during iput */
 718	iput(inode);
 719
 720	err = f2fs_get_node_info(sbi, ino, &ni, false);
 721	if (err)
 722		goto err_out;
 723
 724	/* ENOMEM was fully retried in f2fs_evict_inode. */
 725	if (ni.blk_addr != NULL_ADDR) {
 726		err = -EIO;
 727		goto err_out;
 728	}
 729	return 0;
 730
 731err_out:
 732	set_sbi_flag(sbi, SBI_NEED_FSCK);
 733	f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
 734		  __func__, ino);
 735	return err;
 736}
 737
 738int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
 739{
 740	block_t start_blk, orphan_blocks, i, j;
 741	int err = 0;
 742
 743	if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
 744		return 0;
 745
 746	if (f2fs_hw_is_readonly(sbi)) {
 747		f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
 748		return 0;
 749	}
 750
 751	if (is_sbi_flag_set(sbi, SBI_IS_WRITABLE))
 752		f2fs_info(sbi, "orphan cleanup on readonly fs");
 753
 754	start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
 755	orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
 756
 757	f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
 758
 759	for (i = 0; i < orphan_blocks; i++) {
 760		struct page *page;
 761		struct f2fs_orphan_block *orphan_blk;
 762
 763		page = f2fs_get_meta_page(sbi, start_blk + i);
 764		if (IS_ERR(page)) {
 765			err = PTR_ERR(page);
 766			goto out;
 767		}
 768
 769		orphan_blk = (struct f2fs_orphan_block *)page_address(page);
 770		for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
 771			nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
 772
 773			err = recover_orphan_inode(sbi, ino);
 774			if (err) {
 775				f2fs_put_page(page, 1);
 776				goto out;
 777			}
 778		}
 779		f2fs_put_page(page, 1);
 780	}
 781	/* clear Orphan Flag */
 782	clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
 783out:
 784	set_sbi_flag(sbi, SBI_IS_RECOVERED);
 785
 786	return err;
 787}
 788
 789static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
 790{
 791	struct list_head *head;
 792	struct f2fs_orphan_block *orphan_blk = NULL;
 793	unsigned int nentries = 0;
 794	unsigned short index = 1;
 795	unsigned short orphan_blocks;
 796	struct page *page = NULL;
 797	struct ino_entry *orphan = NULL;
 798	struct inode_management *im = &sbi->im[ORPHAN_INO];
 799
 800	orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
 801
 802	/*
 803	 * we don't need to do spin_lock(&im->ino_lock) here, since all the
 804	 * orphan inode operations are covered under f2fs_lock_op().
 805	 * And, spin_lock should be avoided due to page operations below.
 806	 */
 807	head = &im->ino_list;
 808
 809	/* loop for each orphan inode entry and write them in journal block */
 810	list_for_each_entry(orphan, head, list) {
 811		if (!page) {
 812			page = f2fs_grab_meta_page(sbi, start_blk++);
 813			orphan_blk =
 814				(struct f2fs_orphan_block *)page_address(page);
 815			memset(orphan_blk, 0, sizeof(*orphan_blk));
 816		}
 817
 818		orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
 819
 820		if (nentries == F2FS_ORPHANS_PER_BLOCK) {
 821			/*
 822			 * an orphan block is full of 1020 entries,
 823			 * then we need to flush current orphan blocks
 824			 * and bring another one in memory
 825			 */
 826			orphan_blk->blk_addr = cpu_to_le16(index);
 827			orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 828			orphan_blk->entry_count = cpu_to_le32(nentries);
 829			set_page_dirty(page);
 830			f2fs_put_page(page, 1);
 831			index++;
 832			nentries = 0;
 833			page = NULL;
 834		}
 835	}
 836
 837	if (page) {
 838		orphan_blk->blk_addr = cpu_to_le16(index);
 839		orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
 840		orphan_blk->entry_count = cpu_to_le32(nentries);
 841		set_page_dirty(page);
 842		f2fs_put_page(page, 1);
 843	}
 844}
 845
 846static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
 847						struct f2fs_checkpoint *ckpt)
 848{
 849	unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
 850	__u32 chksum;
 851
 852	chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
 853	if (chksum_ofs < CP_CHKSUM_OFFSET) {
 854		chksum_ofs += sizeof(chksum);
 855		chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
 856						F2FS_BLKSIZE - chksum_ofs);
 857	}
 858	return chksum;
 859}
 860
 861static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
 862		struct f2fs_checkpoint **cp_block, struct page **cp_page,
 863		unsigned long long *version)
 864{
 865	size_t crc_offset = 0;
 866	__u32 crc;
 867
 868	*cp_page = f2fs_get_meta_page(sbi, cp_addr);
 869	if (IS_ERR(*cp_page))
 870		return PTR_ERR(*cp_page);
 871
 872	*cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
 873
 874	crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
 875	if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
 876			crc_offset > CP_CHKSUM_OFFSET) {
 877		f2fs_put_page(*cp_page, 1);
 878		f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
 879		return -EINVAL;
 880	}
 881
 882	crc = f2fs_checkpoint_chksum(sbi, *cp_block);
 883	if (crc != cur_cp_crc(*cp_block)) {
 884		f2fs_put_page(*cp_page, 1);
 885		f2fs_warn(sbi, "invalid crc value");
 886		return -EINVAL;
 887	}
 888
 889	*version = cur_cp_version(*cp_block);
 890	return 0;
 891}
 892
 893static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
 894				block_t cp_addr, unsigned long long *version)
 895{
 896	struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
 897	struct f2fs_checkpoint *cp_block = NULL;
 898	unsigned long long cur_version = 0, pre_version = 0;
 899	unsigned int cp_blocks;
 900	int err;
 901
 902	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 903					&cp_page_1, version);
 904	if (err)
 905		return NULL;
 906
 907	cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
 908
 909	if (cp_blocks > BLKS_PER_SEG(sbi) || cp_blocks <= F2FS_CP_PACKS) {
 910		f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
 911			  le32_to_cpu(cp_block->cp_pack_total_block_count));
 912		goto invalid_cp;
 913	}
 914	pre_version = *version;
 915
 916	cp_addr += cp_blocks - 1;
 917	err = get_checkpoint_version(sbi, cp_addr, &cp_block,
 918					&cp_page_2, version);
 919	if (err)
 920		goto invalid_cp;
 921	cur_version = *version;
 922
 923	if (cur_version == pre_version) {
 924		*version = cur_version;
 925		f2fs_put_page(cp_page_2, 1);
 926		return cp_page_1;
 927	}
 928	f2fs_put_page(cp_page_2, 1);
 929invalid_cp:
 930	f2fs_put_page(cp_page_1, 1);
 931	return NULL;
 932}
 933
 934int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
 935{
 936	struct f2fs_checkpoint *cp_block;
 937	struct f2fs_super_block *fsb = sbi->raw_super;
 938	struct page *cp1, *cp2, *cur_page;
 939	unsigned long blk_size = sbi->blocksize;
 940	unsigned long long cp1_version = 0, cp2_version = 0;
 941	unsigned long long cp_start_blk_no;
 942	unsigned int cp_blks = 1 + __cp_payload(sbi);
 943	block_t cp_blk_no;
 944	int i;
 945	int err;
 946
 947	sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
 948				  GFP_KERNEL);
 949	if (!sbi->ckpt)
 950		return -ENOMEM;
 951	/*
 952	 * Finding out valid cp block involves read both
 953	 * sets( cp pack 1 and cp pack 2)
 954	 */
 955	cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 956	cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
 957
 958	/* The second checkpoint pack should start at the next segment */
 959	cp_start_blk_no += ((unsigned long long)1) <<
 960				le32_to_cpu(fsb->log_blocks_per_seg);
 961	cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
 962
 963	if (cp1 && cp2) {
 964		if (ver_after(cp2_version, cp1_version))
 965			cur_page = cp2;
 966		else
 967			cur_page = cp1;
 968	} else if (cp1) {
 969		cur_page = cp1;
 970	} else if (cp2) {
 971		cur_page = cp2;
 972	} else {
 973		err = -EFSCORRUPTED;
 974		goto fail_no_cp;
 975	}
 976
 977	cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
 978	memcpy(sbi->ckpt, cp_block, blk_size);
 979
 980	if (cur_page == cp1)
 981		sbi->cur_cp_pack = 1;
 982	else
 983		sbi->cur_cp_pack = 2;
 984
 985	/* Sanity checking of checkpoint */
 986	if (f2fs_sanity_check_ckpt(sbi)) {
 987		err = -EFSCORRUPTED;
 988		goto free_fail_no_cp;
 989	}
 990
 991	if (cp_blks <= 1)
 992		goto done;
 993
 994	cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
 995	if (cur_page == cp2)
 996		cp_blk_no += BIT(le32_to_cpu(fsb->log_blocks_per_seg));
 997
 998	for (i = 1; i < cp_blks; i++) {
 999		void *sit_bitmap_ptr;
1000		unsigned char *ckpt = (unsigned char *)sbi->ckpt;
1001
1002		cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
1003		if (IS_ERR(cur_page)) {
1004			err = PTR_ERR(cur_page);
1005			goto free_fail_no_cp;
1006		}
1007		sit_bitmap_ptr = page_address(cur_page);
1008		memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
1009		f2fs_put_page(cur_page, 1);
1010	}
1011done:
1012	f2fs_put_page(cp1, 1);
1013	f2fs_put_page(cp2, 1);
1014	return 0;
1015
1016free_fail_no_cp:
1017	f2fs_put_page(cp1, 1);
1018	f2fs_put_page(cp2, 1);
1019fail_no_cp:
1020	kvfree(sbi->ckpt);
1021	return err;
1022}
1023
1024static void __add_dirty_inode(struct inode *inode, enum inode_type type)
1025{
1026	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1027	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1028
1029	if (is_inode_flag_set(inode, flag))
1030		return;
1031
1032	set_inode_flag(inode, flag);
1033	list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
1034	stat_inc_dirty_inode(sbi, type);
1035}
1036
1037static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1038{
1039	int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1040
1041	if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1042		return;
1043
1044	list_del_init(&F2FS_I(inode)->dirty_list);
1045	clear_inode_flag(inode, flag);
1046	stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1047}
1048
1049void f2fs_update_dirty_folio(struct inode *inode, struct folio *folio)
1050{
1051	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1052	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1053
1054	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1055			!S_ISLNK(inode->i_mode))
1056		return;
1057
1058	spin_lock(&sbi->inode_lock[type]);
1059	if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1060		__add_dirty_inode(inode, type);
1061	inode_inc_dirty_pages(inode);
1062	spin_unlock(&sbi->inode_lock[type]);
1063
1064	set_page_private_reference(&folio->page);
1065}
1066
1067void f2fs_remove_dirty_inode(struct inode *inode)
1068{
1069	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1070	enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1071
1072	if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1073			!S_ISLNK(inode->i_mode))
1074		return;
1075
1076	if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1077		return;
1078
1079	spin_lock(&sbi->inode_lock[type]);
1080	__remove_dirty_inode(inode, type);
1081	spin_unlock(&sbi->inode_lock[type]);
1082}
1083
1084int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type,
1085						bool from_cp)
1086{
1087	struct list_head *head;
1088	struct inode *inode;
1089	struct f2fs_inode_info *fi;
1090	bool is_dir = (type == DIR_INODE);
1091	unsigned long ino = 0;
1092
1093	trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1094				get_pages(sbi, is_dir ?
1095				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1096retry:
1097	if (unlikely(f2fs_cp_error(sbi))) {
1098		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1099				get_pages(sbi, is_dir ?
1100				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1101		return -EIO;
1102	}
1103
1104	spin_lock(&sbi->inode_lock[type]);
1105
1106	head = &sbi->inode_list[type];
1107	if (list_empty(head)) {
1108		spin_unlock(&sbi->inode_lock[type]);
1109		trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1110				get_pages(sbi, is_dir ?
1111				F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1112		return 0;
1113	}
1114	fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1115	inode = igrab(&fi->vfs_inode);
1116	spin_unlock(&sbi->inode_lock[type]);
1117	if (inode) {
1118		unsigned long cur_ino = inode->i_ino;
1119
1120		if (from_cp)
1121			F2FS_I(inode)->cp_task = current;
1122		F2FS_I(inode)->wb_task = current;
1123
1124		filemap_fdatawrite(inode->i_mapping);
1125
1126		F2FS_I(inode)->wb_task = NULL;
1127		if (from_cp)
1128			F2FS_I(inode)->cp_task = NULL;
1129
1130		iput(inode);
1131		/* We need to give cpu to another writers. */
1132		if (ino == cur_ino)
1133			cond_resched();
1134		else
1135			ino = cur_ino;
1136	} else {
1137		/*
1138		 * We should submit bio, since it exists several
1139		 * writebacking dentry pages in the freeing inode.
1140		 */
1141		f2fs_submit_merged_write(sbi, DATA);
1142		cond_resched();
1143	}
1144	goto retry;
1145}
1146
1147static int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1148{
1149	struct list_head *head = &sbi->inode_list[DIRTY_META];
1150	struct inode *inode;
1151	struct f2fs_inode_info *fi;
1152	s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1153
1154	while (total--) {
1155		if (unlikely(f2fs_cp_error(sbi)))
1156			return -EIO;
1157
1158		spin_lock(&sbi->inode_lock[DIRTY_META]);
1159		if (list_empty(head)) {
1160			spin_unlock(&sbi->inode_lock[DIRTY_META]);
1161			return 0;
1162		}
1163		fi = list_first_entry(head, struct f2fs_inode_info,
1164							gdirty_list);
1165		inode = igrab(&fi->vfs_inode);
1166		spin_unlock(&sbi->inode_lock[DIRTY_META]);
1167		if (inode) {
1168			sync_inode_metadata(inode, 0);
1169
1170			/* it's on eviction */
1171			if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1172				f2fs_update_inode_page(inode);
1173			iput(inode);
1174		}
1175	}
1176	return 0;
1177}
1178
1179static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1180{
1181	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1182	struct f2fs_nm_info *nm_i = NM_I(sbi);
1183	nid_t last_nid = nm_i->next_scan_nid;
1184
1185	next_free_nid(sbi, &last_nid);
1186	ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1187	ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1188	ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1189	ckpt->next_free_nid = cpu_to_le32(last_nid);
1190
1191	/* update user_block_counts */
1192	sbi->last_valid_block_count = sbi->total_valid_block_count;
1193	percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1194	percpu_counter_set(&sbi->rf_node_block_count, 0);
1195}
1196
1197static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1198{
1199	bool ret = false;
1200
1201	if (!is_journalled_quota(sbi))
1202		return false;
1203
1204	if (!f2fs_down_write_trylock(&sbi->quota_sem))
1205		return true;
1206	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1207		ret = false;
1208	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1209		ret = false;
1210	} else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1211		clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1212		ret = true;
1213	} else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1214		ret = true;
1215	}
1216	f2fs_up_write(&sbi->quota_sem);
1217	return ret;
1218}
1219
1220/*
1221 * Freeze all the FS-operations for checkpoint.
1222 */
1223static int block_operations(struct f2fs_sb_info *sbi)
1224{
1225	struct writeback_control wbc = {
1226		.sync_mode = WB_SYNC_ALL,
1227		.nr_to_write = LONG_MAX,
1228		.for_reclaim = 0,
1229	};
1230	int err = 0, cnt = 0;
1231
1232	/*
1233	 * Let's flush inline_data in dirty node pages.
1234	 */
1235	f2fs_flush_inline_data(sbi);
1236
1237retry_flush_quotas:
1238	f2fs_lock_all(sbi);
1239	if (__need_flush_quota(sbi)) {
1240		int locked;
1241
1242		if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1243			set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1244			set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1245			goto retry_flush_dents;
1246		}
1247		f2fs_unlock_all(sbi);
1248
1249		/* only failed during mount/umount/freeze/quotactl */
1250		locked = down_read_trylock(&sbi->sb->s_umount);
1251		f2fs_quota_sync(sbi->sb, -1);
1252		if (locked)
1253			up_read(&sbi->sb->s_umount);
1254		cond_resched();
1255		goto retry_flush_quotas;
1256	}
1257
1258retry_flush_dents:
1259	/* write all the dirty dentry pages */
1260	if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1261		f2fs_unlock_all(sbi);
1262		err = f2fs_sync_dirty_inodes(sbi, DIR_INODE, true);
1263		if (err)
1264			return err;
1265		cond_resched();
1266		goto retry_flush_quotas;
1267	}
1268
1269	/*
1270	 * POR: we should ensure that there are no dirty node pages
1271	 * until finishing nat/sit flush. inode->i_blocks can be updated.
1272	 */
1273	f2fs_down_write(&sbi->node_change);
1274
1275	if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1276		f2fs_up_write(&sbi->node_change);
1277		f2fs_unlock_all(sbi);
1278		err = f2fs_sync_inode_meta(sbi);
1279		if (err)
1280			return err;
1281		cond_resched();
1282		goto retry_flush_quotas;
1283	}
1284
1285retry_flush_nodes:
1286	f2fs_down_write(&sbi->node_write);
1287
1288	if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1289		f2fs_up_write(&sbi->node_write);
1290		atomic_inc(&sbi->wb_sync_req[NODE]);
1291		err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1292		atomic_dec(&sbi->wb_sync_req[NODE]);
1293		if (err) {
1294			f2fs_up_write(&sbi->node_change);
1295			f2fs_unlock_all(sbi);
1296			return err;
1297		}
1298		cond_resched();
1299		goto retry_flush_nodes;
1300	}
1301
1302	/*
1303	 * sbi->node_change is used only for AIO write_begin path which produces
1304	 * dirty node blocks and some checkpoint values by block allocation.
1305	 */
1306	__prepare_cp_block(sbi);
1307	f2fs_up_write(&sbi->node_change);
1308	return err;
1309}
1310
1311static void unblock_operations(struct f2fs_sb_info *sbi)
1312{
1313	f2fs_up_write(&sbi->node_write);
1314	f2fs_unlock_all(sbi);
1315}
1316
1317void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1318{
1319	DEFINE_WAIT(wait);
1320
1321	for (;;) {
1322		if (!get_pages(sbi, type))
1323			break;
1324
1325		if (unlikely(f2fs_cp_error(sbi) &&
1326			!is_sbi_flag_set(sbi, SBI_IS_CLOSE)))
1327			break;
1328
1329		if (type == F2FS_DIRTY_META)
1330			f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1331							FS_CP_META_IO);
1332		else if (type == F2FS_WB_CP_DATA)
1333			f2fs_submit_merged_write(sbi, DATA);
1334
1335		prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1336		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1337	}
1338	finish_wait(&sbi->cp_wait, &wait);
1339}
1340
1341static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1342{
1343	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1344	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1345	unsigned long flags;
1346
1347	if (cpc->reason & CP_UMOUNT) {
1348		if (le32_to_cpu(ckpt->cp_pack_total_block_count) +
1349			NM_I(sbi)->nat_bits_blocks > BLKS_PER_SEG(sbi)) {
1350			clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1351			f2fs_notice(sbi, "Disable nat_bits due to no space");
1352		} else if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG) &&
1353						f2fs_nat_bitmap_enabled(sbi)) {
1354			f2fs_enable_nat_bits(sbi);
1355			set_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
1356			f2fs_notice(sbi, "Rebuild and enable nat_bits");
1357		}
1358	}
1359
1360	spin_lock_irqsave(&sbi->cp_lock, flags);
1361
1362	if (cpc->reason & CP_TRIMMED)
1363		__set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1364	else
1365		__clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1366
1367	if (cpc->reason & CP_UMOUNT)
1368		__set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1369	else
1370		__clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1371
1372	if (cpc->reason & CP_FASTBOOT)
1373		__set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1374	else
1375		__clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1376
1377	if (orphan_num)
1378		__set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1379	else
1380		__clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1381
1382	if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1383		__set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1384
1385	if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1386		__set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1387	else
1388		__clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1389
1390	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1391		__set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1392	else
1393		__clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1394
1395	if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1396		__set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1397	else
1398		__clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1399
1400	if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1401		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1402	else
1403		__clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1404
1405	if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1406		__set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1407
1408	/* set this flag to activate crc|cp_ver for recovery */
1409	__set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1410	__clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1411
1412	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1413}
1414
1415static void commit_checkpoint(struct f2fs_sb_info *sbi,
1416	void *src, block_t blk_addr)
1417{
1418	struct writeback_control wbc = {
1419		.for_reclaim = 0,
1420	};
1421
1422	/*
1423	 * filemap_get_folios_tag and lock_page again will take
1424	 * some extra time. Therefore, f2fs_update_meta_pages and
1425	 * f2fs_sync_meta_pages are combined in this function.
1426	 */
1427	struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1428	int err;
1429
1430	f2fs_wait_on_page_writeback(page, META, true, true);
1431
1432	memcpy(page_address(page), src, PAGE_SIZE);
1433
1434	set_page_dirty(page);
1435	if (unlikely(!clear_page_dirty_for_io(page)))
1436		f2fs_bug_on(sbi, 1);
1437
1438	/* writeout cp pack 2 page */
1439	err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1440	if (unlikely(err && f2fs_cp_error(sbi))) {
1441		f2fs_put_page(page, 1);
1442		return;
1443	}
1444
1445	f2fs_bug_on(sbi, err);
1446	f2fs_put_page(page, 0);
1447
1448	/* submit checkpoint (with barrier if NOBARRIER is not set) */
1449	f2fs_submit_merged_write(sbi, META_FLUSH);
1450}
1451
1452static inline u64 get_sectors_written(struct block_device *bdev)
1453{
1454	return (u64)part_stat_read(bdev, sectors[STAT_WRITE]);
1455}
1456
1457u64 f2fs_get_sectors_written(struct f2fs_sb_info *sbi)
1458{
1459	if (f2fs_is_multi_device(sbi)) {
1460		u64 sectors = 0;
1461		int i;
1462
1463		for (i = 0; i < sbi->s_ndevs; i++)
1464			sectors += get_sectors_written(FDEV(i).bdev);
1465
1466		return sectors;
1467	}
1468
1469	return get_sectors_written(sbi->sb->s_bdev);
1470}
1471
1472static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1473{
1474	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1475	struct f2fs_nm_info *nm_i = NM_I(sbi);
1476	unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1477	block_t start_blk;
1478	unsigned int data_sum_blocks, orphan_blocks;
1479	__u32 crc32 = 0;
1480	int i;
1481	int cp_payload_blks = __cp_payload(sbi);
1482	struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1483	u64 kbytes_written;
1484	int err;
1485
1486	/* Flush all the NAT/SIT pages */
1487	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1488
1489	/* start to update checkpoint, cp ver is already updated previously */
1490	ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1491	ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1492	for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1493		struct curseg_info *curseg = CURSEG_I(sbi, i + CURSEG_HOT_NODE);
1494
1495		ckpt->cur_node_segno[i] = cpu_to_le32(curseg->segno);
1496		ckpt->cur_node_blkoff[i] = cpu_to_le16(curseg->next_blkoff);
1497		ckpt->alloc_type[i + CURSEG_HOT_NODE] = curseg->alloc_type;
1498	}
1499	for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1500		struct curseg_info *curseg = CURSEG_I(sbi, i + CURSEG_HOT_DATA);
1501
1502		ckpt->cur_data_segno[i] = cpu_to_le32(curseg->segno);
1503		ckpt->cur_data_blkoff[i] = cpu_to_le16(curseg->next_blkoff);
1504		ckpt->alloc_type[i + CURSEG_HOT_DATA] = curseg->alloc_type;
1505	}
1506
1507	/* 2 cp + n data seg summary + orphan inode blocks */
1508	data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1509	spin_lock_irqsave(&sbi->cp_lock, flags);
1510	if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1511		__set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1512	else
1513		__clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1514	spin_unlock_irqrestore(&sbi->cp_lock, flags);
1515
1516	orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1517	ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1518			orphan_blocks);
1519
1520	if (__remain_node_summaries(cpc->reason))
1521		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1522				cp_payload_blks + data_sum_blocks +
1523				orphan_blocks + NR_CURSEG_NODE_TYPE);
1524	else
1525		ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1526				cp_payload_blks + data_sum_blocks +
1527				orphan_blocks);
1528
1529	/* update ckpt flag for checkpoint */
1530	update_ckpt_flags(sbi, cpc);
1531
1532	/* update SIT/NAT bitmap */
1533	get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1534	get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1535
1536	crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1537	*((__le32 *)((unsigned char *)ckpt +
1538				le32_to_cpu(ckpt->checksum_offset)))
1539				= cpu_to_le32(crc32);
1540
1541	start_blk = __start_cp_next_addr(sbi);
1542
1543	/* write nat bits */
1544	if ((cpc->reason & CP_UMOUNT) &&
1545			is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG)) {
1546		__u64 cp_ver = cur_cp_version(ckpt);
1547		block_t blk;
1548
1549		cp_ver |= ((__u64)crc32 << 32);
1550		*(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1551
1552		blk = start_blk + BLKS_PER_SEG(sbi) - nm_i->nat_bits_blocks;
1553		for (i = 0; i < nm_i->nat_bits_blocks; i++)
1554			f2fs_update_meta_page(sbi, nm_i->nat_bits +
1555					F2FS_BLK_TO_BYTES(i), blk + i);
1556	}
1557
1558	/* write out checkpoint buffer at block 0 */
1559	f2fs_update_meta_page(sbi, ckpt, start_blk++);
1560
1561	for (i = 1; i < 1 + cp_payload_blks; i++)
1562		f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1563							start_blk++);
1564
1565	if (orphan_num) {
1566		write_orphan_inodes(sbi, start_blk);
1567		start_blk += orphan_blocks;
1568	}
1569
1570	f2fs_write_data_summaries(sbi, start_blk);
1571	start_blk += data_sum_blocks;
1572
1573	/* Record write statistics in the hot node summary */
1574	kbytes_written = sbi->kbytes_written;
1575	kbytes_written += (f2fs_get_sectors_written(sbi) -
1576				sbi->sectors_written_start) >> 1;
1577	seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1578
1579	if (__remain_node_summaries(cpc->reason)) {
1580		f2fs_write_node_summaries(sbi, start_blk);
1581		start_blk += NR_CURSEG_NODE_TYPE;
1582	}
1583
 
 
 
 
 
1584	/* Here, we have one bio having CP pack except cp pack 2 page */
1585	f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1586	/* Wait for all dirty meta pages to be submitted for IO */
1587	f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1588
1589	/* wait for previous submitted meta pages writeback */
1590	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1591
1592	/* flush all device cache */
1593	err = f2fs_flush_device_cache(sbi);
1594	if (err)
1595		return err;
1596
1597	/* barrier and flush checkpoint cp pack 2 page if it can */
1598	commit_checkpoint(sbi, ckpt, start_blk);
1599	f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1600
1601	/*
1602	 * invalidate intermediate page cache borrowed from meta inode which are
1603	 * used for migration of encrypted, verity or compressed inode's blocks.
1604	 */
1605	if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1606		f2fs_sb_has_compression(sbi))
1607		f2fs_bug_on(sbi,
1608			invalidate_inode_pages2_range(META_MAPPING(sbi),
1609				MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1));
1610
1611	f2fs_release_ino_entry(sbi, false);
1612
1613	f2fs_reset_fsync_node_info(sbi);
1614
1615	clear_sbi_flag(sbi, SBI_IS_DIRTY);
1616	clear_sbi_flag(sbi, SBI_NEED_CP);
1617	clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1618
1619	spin_lock(&sbi->stat_lock);
1620	sbi->unusable_block_count = 0;
1621	spin_unlock(&sbi->stat_lock);
1622
1623	__set_cp_next_pack(sbi);
1624
1625	/*
1626	 * redirty superblock if metadata like node page or inode cache is
1627	 * updated during writing checkpoint.
1628	 */
1629	if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1630			get_pages(sbi, F2FS_DIRTY_IMETA))
1631		set_sbi_flag(sbi, SBI_IS_DIRTY);
1632
1633	f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1634
1635	return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1636}
1637
1638int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1639{
1640	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1641	unsigned long long ckpt_ver;
1642	int err = 0;
1643
1644	if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1645		return -EROFS;
1646
1647	if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1648		if (cpc->reason != CP_PAUSE)
1649			return 0;
1650		f2fs_warn(sbi, "Start checkpoint disabled!");
1651	}
1652	if (cpc->reason != CP_RESIZE)
1653		f2fs_down_write(&sbi->cp_global_sem);
1654
1655	if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1656		((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1657		((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1658		goto out;
1659	if (unlikely(f2fs_cp_error(sbi))) {
1660		err = -EIO;
1661		goto out;
1662	}
1663
1664	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1665
1666	err = block_operations(sbi);
1667	if (err)
1668		goto out;
1669
1670	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1671
1672	f2fs_flush_merged_writes(sbi);
1673
1674	/* this is the case of multiple fstrims without any changes */
1675	if (cpc->reason & CP_DISCARD) {
1676		if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1677			unblock_operations(sbi);
1678			goto out;
1679		}
1680
1681		if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1682				SIT_I(sbi)->dirty_sentries == 0 &&
1683				prefree_segments(sbi) == 0) {
1684			f2fs_flush_sit_entries(sbi, cpc);
1685			f2fs_clear_prefree_segments(sbi, cpc);
1686			unblock_operations(sbi);
1687			goto out;
1688		}
1689	}
1690
1691	/*
1692	 * update checkpoint pack index
1693	 * Increase the version number so that
1694	 * SIT entries and seg summaries are written at correct place
1695	 */
1696	ckpt_ver = cur_cp_version(ckpt);
1697	ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1698
1699	/* write cached NAT/SIT entries to NAT/SIT area */
1700	err = f2fs_flush_nat_entries(sbi, cpc);
1701	if (err) {
1702		f2fs_err(sbi, "f2fs_flush_nat_entries failed err:%d, stop checkpoint", err);
1703		f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1704		goto stop;
1705	}
1706
1707	f2fs_flush_sit_entries(sbi, cpc);
1708
1709	/* save inmem log status */
1710	f2fs_save_inmem_curseg(sbi);
1711
1712	err = do_checkpoint(sbi, cpc);
1713	if (err) {
1714		f2fs_err(sbi, "do_checkpoint failed err:%d, stop checkpoint", err);
1715		f2fs_bug_on(sbi, !f2fs_cp_error(sbi));
1716		f2fs_release_discard_addrs(sbi);
1717	} else {
1718		f2fs_clear_prefree_segments(sbi, cpc);
1719	}
1720
1721	f2fs_restore_inmem_curseg(sbi);
1722	f2fs_reinit_atgc_curseg(sbi);
1723	stat_inc_cp_count(sbi);
1724stop:
1725	unblock_operations(sbi);
1726
1727	if (cpc->reason & CP_RECOVERY)
1728		f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1729
1730	/* update CP_TIME to trigger checkpoint periodically */
1731	f2fs_update_time(sbi, CP_TIME);
1732	trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1733out:
1734	if (cpc->reason != CP_RESIZE)
1735		f2fs_up_write(&sbi->cp_global_sem);
1736	return err;
1737}
1738
1739void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1740{
1741	int i;
1742
1743	for (i = 0; i < MAX_INO_ENTRY; i++) {
1744		struct inode_management *im = &sbi->im[i];
1745
1746		INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1747		spin_lock_init(&im->ino_lock);
1748		INIT_LIST_HEAD(&im->ino_list);
1749		im->ino_num = 0;
1750	}
1751
1752	sbi->max_orphans = (BLKS_PER_SEG(sbi) - F2FS_CP_PACKS -
1753			NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1754			F2FS_ORPHANS_PER_BLOCK;
1755}
1756
1757int __init f2fs_create_checkpoint_caches(void)
1758{
1759	ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1760			sizeof(struct ino_entry));
1761	if (!ino_entry_slab)
1762		return -ENOMEM;
1763	f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1764			sizeof(struct inode_entry));
1765	if (!f2fs_inode_entry_slab) {
1766		kmem_cache_destroy(ino_entry_slab);
1767		return -ENOMEM;
1768	}
1769	return 0;
1770}
1771
1772void f2fs_destroy_checkpoint_caches(void)
1773{
1774	kmem_cache_destroy(ino_entry_slab);
1775	kmem_cache_destroy(f2fs_inode_entry_slab);
1776}
1777
1778static int __write_checkpoint_sync(struct f2fs_sb_info *sbi)
1779{
1780	struct cp_control cpc = { .reason = CP_SYNC, };
1781	int err;
1782
1783	f2fs_down_write(&sbi->gc_lock);
1784	err = f2fs_write_checkpoint(sbi, &cpc);
1785	f2fs_up_write(&sbi->gc_lock);
1786
1787	return err;
1788}
1789
1790static void __checkpoint_and_complete_reqs(struct f2fs_sb_info *sbi)
1791{
1792	struct ckpt_req_control *cprc = &sbi->cprc_info;
1793	struct ckpt_req *req, *next;
1794	struct llist_node *dispatch_list;
1795	u64 sum_diff = 0, diff, count = 0;
1796	int ret;
1797
1798	dispatch_list = llist_del_all(&cprc->issue_list);
1799	if (!dispatch_list)
1800		return;
1801	dispatch_list = llist_reverse_order(dispatch_list);
1802
1803	ret = __write_checkpoint_sync(sbi);
1804	atomic_inc(&cprc->issued_ckpt);
1805
1806	llist_for_each_entry_safe(req, next, dispatch_list, llnode) {
1807		diff = (u64)ktime_ms_delta(ktime_get(), req->queue_time);
1808		req->ret = ret;
1809		complete(&req->wait);
1810
1811		sum_diff += diff;
1812		count++;
1813	}
1814	atomic_sub(count, &cprc->queued_ckpt);
1815	atomic_add(count, &cprc->total_ckpt);
1816
1817	spin_lock(&cprc->stat_lock);
1818	cprc->cur_time = (unsigned int)div64_u64(sum_diff, count);
1819	if (cprc->peak_time < cprc->cur_time)
1820		cprc->peak_time = cprc->cur_time;
1821	spin_unlock(&cprc->stat_lock);
1822}
1823
1824static int issue_checkpoint_thread(void *data)
1825{
1826	struct f2fs_sb_info *sbi = data;
1827	struct ckpt_req_control *cprc = &sbi->cprc_info;
1828	wait_queue_head_t *q = &cprc->ckpt_wait_queue;
1829repeat:
1830	if (kthread_should_stop())
1831		return 0;
1832
1833	if (!llist_empty(&cprc->issue_list))
1834		__checkpoint_and_complete_reqs(sbi);
1835
1836	wait_event_interruptible(*q,
1837		kthread_should_stop() || !llist_empty(&cprc->issue_list));
1838	goto repeat;
1839}
1840
1841static void flush_remained_ckpt_reqs(struct f2fs_sb_info *sbi,
1842		struct ckpt_req *wait_req)
1843{
1844	struct ckpt_req_control *cprc = &sbi->cprc_info;
1845
1846	if (!llist_empty(&cprc->issue_list)) {
1847		__checkpoint_and_complete_reqs(sbi);
1848	} else {
1849		/* already dispatched by issue_checkpoint_thread */
1850		if (wait_req)
1851			wait_for_completion(&wait_req->wait);
1852	}
1853}
1854
1855static void init_ckpt_req(struct ckpt_req *req)
1856{
1857	memset(req, 0, sizeof(struct ckpt_req));
1858
1859	init_completion(&req->wait);
1860	req->queue_time = ktime_get();
1861}
1862
1863int f2fs_issue_checkpoint(struct f2fs_sb_info *sbi)
1864{
1865	struct ckpt_req_control *cprc = &sbi->cprc_info;
1866	struct ckpt_req req;
1867	struct cp_control cpc;
1868
1869	cpc.reason = __get_cp_reason(sbi);
1870	if (!test_opt(sbi, MERGE_CHECKPOINT) || cpc.reason != CP_SYNC) {
1871		int ret;
1872
1873		f2fs_down_write(&sbi->gc_lock);
1874		ret = f2fs_write_checkpoint(sbi, &cpc);
1875		f2fs_up_write(&sbi->gc_lock);
1876
1877		return ret;
1878	}
1879
1880	if (!cprc->f2fs_issue_ckpt)
1881		return __write_checkpoint_sync(sbi);
1882
1883	init_ckpt_req(&req);
1884
1885	llist_add(&req.llnode, &cprc->issue_list);
1886	atomic_inc(&cprc->queued_ckpt);
1887
1888	/*
1889	 * update issue_list before we wake up issue_checkpoint thread,
1890	 * this smp_mb() pairs with another barrier in ___wait_event(),
1891	 * see more details in comments of waitqueue_active().
1892	 */
1893	smp_mb();
1894
1895	if (waitqueue_active(&cprc->ckpt_wait_queue))
1896		wake_up(&cprc->ckpt_wait_queue);
1897
1898	if (cprc->f2fs_issue_ckpt)
1899		wait_for_completion(&req.wait);
1900	else
1901		flush_remained_ckpt_reqs(sbi, &req);
1902
1903	return req.ret;
1904}
1905
1906int f2fs_start_ckpt_thread(struct f2fs_sb_info *sbi)
1907{
1908	dev_t dev = sbi->sb->s_bdev->bd_dev;
1909	struct ckpt_req_control *cprc = &sbi->cprc_info;
1910
1911	if (cprc->f2fs_issue_ckpt)
1912		return 0;
1913
1914	cprc->f2fs_issue_ckpt = kthread_run(issue_checkpoint_thread, sbi,
1915			"f2fs_ckpt-%u:%u", MAJOR(dev), MINOR(dev));
1916	if (IS_ERR(cprc->f2fs_issue_ckpt)) {
1917		int err = PTR_ERR(cprc->f2fs_issue_ckpt);
1918
1919		cprc->f2fs_issue_ckpt = NULL;
1920		return err;
1921	}
1922
1923	set_task_ioprio(cprc->f2fs_issue_ckpt, cprc->ckpt_thread_ioprio);
1924
1925	return 0;
1926}
1927
1928void f2fs_stop_ckpt_thread(struct f2fs_sb_info *sbi)
1929{
1930	struct ckpt_req_control *cprc = &sbi->cprc_info;
1931	struct task_struct *ckpt_task;
1932
1933	if (!cprc->f2fs_issue_ckpt)
1934		return;
1935
1936	ckpt_task = cprc->f2fs_issue_ckpt;
1937	cprc->f2fs_issue_ckpt = NULL;
1938	kthread_stop(ckpt_task);
1939
1940	f2fs_flush_ckpt_thread(sbi);
1941}
1942
1943void f2fs_flush_ckpt_thread(struct f2fs_sb_info *sbi)
1944{
1945	struct ckpt_req_control *cprc = &sbi->cprc_info;
1946
1947	flush_remained_ckpt_reqs(sbi, NULL);
1948
1949	/* Let's wait for the previous dispatched checkpoint. */
1950	while (atomic_read(&cprc->queued_ckpt))
1951		io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1952}
1953
1954void f2fs_init_ckpt_req_control(struct f2fs_sb_info *sbi)
1955{
1956	struct ckpt_req_control *cprc = &sbi->cprc_info;
1957
1958	atomic_set(&cprc->issued_ckpt, 0);
1959	atomic_set(&cprc->total_ckpt, 0);
1960	atomic_set(&cprc->queued_ckpt, 0);
1961	cprc->ckpt_thread_ioprio = DEFAULT_CHECKPOINT_IOPRIO;
1962	init_waitqueue_head(&cprc->ckpt_wait_queue);
1963	init_llist_head(&cprc->issue_list);
1964	spin_lock_init(&cprc->stat_lock);
1965}