1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
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>
21 #include <trace/events/f2fs.h>
23 static struct kmem_cache *ino_entry_slab;
24 struct kmem_cache *f2fs_inode_entry_slab;
26 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
28 f2fs_build_fault_attr(sbi, 0, 0);
29 set_ckpt_flags(sbi, CP_ERROR_FLAG);
31 f2fs_flush_merged_writes(sbi);
35 * We guarantee no failure on the returned page.
37 struct page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
39 struct address_space *mapping = META_MAPPING(sbi);
40 struct page *page = NULL;
42 page = f2fs_grab_cache_page(mapping, index, false);
47 f2fs_wait_on_page_writeback(page, META, true, true);
48 if (!PageUptodate(page))
49 SetPageUptodate(page);
53 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
56 struct address_space *mapping = META_MAPPING(sbi);
58 struct f2fs_io_info fio = {
62 .op_flags = REQ_META | REQ_PRIO,
65 .encrypted_page = NULL,
70 if (unlikely(!is_meta))
71 fio.op_flags &= ~REQ_META;
73 page = f2fs_grab_cache_page(mapping, index, false);
78 if (PageUptodate(page))
83 err = f2fs_submit_page_bio(&fio);
85 f2fs_put_page(page, 1);
90 if (unlikely(page->mapping != mapping)) {
91 f2fs_put_page(page, 1);
95 if (unlikely(!PageUptodate(page))) {
96 f2fs_put_page(page, 1);
103 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
105 return __get_meta_page(sbi, index, true);
108 struct page *f2fs_get_meta_page_nofail(struct f2fs_sb_info *sbi, pgoff_t index)
114 page = __get_meta_page(sbi, index, true);
116 if (PTR_ERR(page) == -EIO &&
117 ++count <= DEFAULT_RETRY_IO_COUNT)
119 f2fs_stop_checkpoint(sbi, false);
125 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
127 return __get_meta_page(sbi, index, false);
130 static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr,
133 struct seg_entry *se;
134 unsigned int segno, offset;
137 if (type != DATA_GENERIC_ENHANCE && type != DATA_GENERIC_ENHANCE_READ)
140 segno = GET_SEGNO(sbi, blkaddr);
141 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
142 se = get_seg_entry(sbi, segno);
144 exist = f2fs_test_bit(offset, se->cur_valid_map);
145 if (!exist && type == DATA_GENERIC_ENHANCE) {
146 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
148 set_sbi_flag(sbi, SBI_NEED_FSCK);
154 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
155 block_t blkaddr, int type)
161 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
165 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
166 blkaddr < SM_I(sbi)->ssa_blkaddr))
170 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
171 blkaddr < __start_cp_addr(sbi)))
175 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
176 blkaddr < MAIN_BLKADDR(sbi)))
180 case DATA_GENERIC_ENHANCE:
181 case DATA_GENERIC_ENHANCE_READ:
182 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
183 blkaddr < MAIN_BLKADDR(sbi))) {
184 f2fs_warn(sbi, "access invalid blkaddr:%u",
186 set_sbi_flag(sbi, SBI_NEED_FSCK);
190 return __is_bitmap_valid(sbi, blkaddr, type);
194 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
195 blkaddr >= MAIN_BLKADDR(sbi)))
206 * Readahead CP/NAT/SIT/SSA/POR pages
208 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
212 block_t blkno = start;
213 struct f2fs_io_info fio = {
217 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
218 .encrypted_page = NULL,
220 .is_por = (type == META_POR),
222 struct blk_plug plug;
224 if (unlikely(type == META_POR))
225 fio.op_flags &= ~REQ_META;
227 blk_start_plug(&plug);
228 for (; nrpages-- > 0; blkno++) {
230 if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
235 if (unlikely(blkno >=
236 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
238 /* get nat block addr */
239 fio.new_blkaddr = current_nat_addr(sbi,
240 blkno * NAT_ENTRY_PER_BLOCK);
243 /* get sit block addr */
244 fio.new_blkaddr = current_sit_addr(sbi,
245 blkno * SIT_ENTRY_PER_BLOCK);
250 fio.new_blkaddr = blkno;
256 page = f2fs_grab_cache_page(META_MAPPING(sbi),
257 fio.new_blkaddr, false);
260 if (PageUptodate(page)) {
261 f2fs_put_page(page, 1);
266 f2fs_submit_page_bio(&fio);
267 f2fs_put_page(page, 0);
270 blk_finish_plug(&plug);
271 return blkno - start;
274 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
277 bool readahead = false;
279 page = find_get_page(META_MAPPING(sbi), index);
280 if (!page || !PageUptodate(page))
282 f2fs_put_page(page, 0);
285 f2fs_ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
288 static int __f2fs_write_meta_page(struct page *page,
289 struct writeback_control *wbc,
290 enum iostat_type io_type)
292 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
294 trace_f2fs_writepage(page, META);
296 if (unlikely(f2fs_cp_error(sbi)))
298 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
300 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
303 f2fs_do_write_meta_page(sbi, page, io_type);
304 dec_page_count(sbi, F2FS_DIRTY_META);
306 if (wbc->for_reclaim)
307 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
311 if (unlikely(f2fs_cp_error(sbi)))
312 f2fs_submit_merged_write(sbi, META);
317 redirty_page_for_writepage(wbc, page);
318 return AOP_WRITEPAGE_ACTIVATE;
321 static int f2fs_write_meta_page(struct page *page,
322 struct writeback_control *wbc)
324 return __f2fs_write_meta_page(page, wbc, FS_META_IO);
327 static int f2fs_write_meta_pages(struct address_space *mapping,
328 struct writeback_control *wbc)
330 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
333 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
336 /* collect a number of dirty meta pages and write together */
337 if (wbc->sync_mode != WB_SYNC_ALL &&
338 get_pages(sbi, F2FS_DIRTY_META) <
339 nr_pages_to_skip(sbi, META))
342 /* if locked failed, cp will flush dirty pages instead */
343 if (!mutex_trylock(&sbi->cp_mutex))
346 trace_f2fs_writepages(mapping->host, wbc, META);
347 diff = nr_pages_to_write(sbi, META, wbc);
348 written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
349 mutex_unlock(&sbi->cp_mutex);
350 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
354 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
355 trace_f2fs_writepages(mapping->host, wbc, META);
359 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
360 long nr_to_write, enum iostat_type io_type)
362 struct address_space *mapping = META_MAPPING(sbi);
363 pgoff_t index = 0, prev = ULONG_MAX;
367 struct writeback_control wbc = {
370 struct blk_plug plug;
374 blk_start_plug(&plug);
376 while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
377 PAGECACHE_TAG_DIRTY))) {
380 for (i = 0; i < nr_pages; i++) {
381 struct page *page = pvec.pages[i];
383 if (prev == ULONG_MAX)
384 prev = page->index - 1;
385 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
386 pagevec_release(&pvec);
392 if (unlikely(page->mapping != mapping)) {
397 if (!PageDirty(page)) {
398 /* someone wrote it for us */
399 goto continue_unlock;
402 f2fs_wait_on_page_writeback(page, META, true, true);
404 if (!clear_page_dirty_for_io(page))
405 goto continue_unlock;
407 if (__f2fs_write_meta_page(page, &wbc, io_type)) {
413 if (unlikely(nwritten >= nr_to_write))
416 pagevec_release(&pvec);
421 f2fs_submit_merged_write(sbi, type);
423 blk_finish_plug(&plug);
428 static int f2fs_set_meta_page_dirty(struct page *page)
430 trace_f2fs_set_page_dirty(page, META);
432 if (!PageUptodate(page))
433 SetPageUptodate(page);
434 if (!PageDirty(page)) {
435 __set_page_dirty_nobuffers(page);
436 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
437 f2fs_set_page_private(page, 0);
438 f2fs_trace_pid(page);
444 const struct address_space_operations f2fs_meta_aops = {
445 .writepage = f2fs_write_meta_page,
446 .writepages = f2fs_write_meta_pages,
447 .set_page_dirty = f2fs_set_meta_page_dirty,
448 .invalidatepage = f2fs_invalidate_page,
449 .releasepage = f2fs_release_page,
450 #ifdef CONFIG_MIGRATION
451 .migratepage = f2fs_migrate_page,
455 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
456 unsigned int devidx, int type)
458 struct inode_management *im = &sbi->im[type];
459 struct ino_entry *e, *tmp;
461 tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
463 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
465 spin_lock(&im->ino_lock);
466 e = radix_tree_lookup(&im->ino_root, ino);
469 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
472 memset(e, 0, sizeof(struct ino_entry));
475 list_add_tail(&e->list, &im->ino_list);
476 if (type != ORPHAN_INO)
480 if (type == FLUSH_INO)
481 f2fs_set_bit(devidx, (char *)&e->dirty_device);
483 spin_unlock(&im->ino_lock);
484 radix_tree_preload_end();
487 kmem_cache_free(ino_entry_slab, tmp);
490 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
492 struct inode_management *im = &sbi->im[type];
495 spin_lock(&im->ino_lock);
496 e = radix_tree_lookup(&im->ino_root, ino);
499 radix_tree_delete(&im->ino_root, ino);
501 spin_unlock(&im->ino_lock);
502 kmem_cache_free(ino_entry_slab, e);
505 spin_unlock(&im->ino_lock);
508 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
510 /* add new dirty ino entry into list */
511 __add_ino_entry(sbi, ino, 0, type);
514 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
516 /* remove dirty ino entry from list */
517 __remove_ino_entry(sbi, ino, type);
520 /* mode should be APPEND_INO or UPDATE_INO */
521 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
523 struct inode_management *im = &sbi->im[mode];
526 spin_lock(&im->ino_lock);
527 e = radix_tree_lookup(&im->ino_root, ino);
528 spin_unlock(&im->ino_lock);
529 return e ? true : false;
532 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
534 struct ino_entry *e, *tmp;
537 for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
538 struct inode_management *im = &sbi->im[i];
540 spin_lock(&im->ino_lock);
541 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
543 radix_tree_delete(&im->ino_root, e->ino);
544 kmem_cache_free(ino_entry_slab, e);
547 spin_unlock(&im->ino_lock);
551 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
552 unsigned int devidx, int type)
554 __add_ino_entry(sbi, ino, devidx, type);
557 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
558 unsigned int devidx, int type)
560 struct inode_management *im = &sbi->im[type];
562 bool is_dirty = false;
564 spin_lock(&im->ino_lock);
565 e = radix_tree_lookup(&im->ino_root, ino);
566 if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
568 spin_unlock(&im->ino_lock);
572 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
574 struct inode_management *im = &sbi->im[ORPHAN_INO];
577 spin_lock(&im->ino_lock);
579 if (time_to_inject(sbi, FAULT_ORPHAN)) {
580 spin_unlock(&im->ino_lock);
581 f2fs_show_injection_info(sbi, FAULT_ORPHAN);
585 if (unlikely(im->ino_num >= sbi->max_orphans))
589 spin_unlock(&im->ino_lock);
594 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
596 struct inode_management *im = &sbi->im[ORPHAN_INO];
598 spin_lock(&im->ino_lock);
599 f2fs_bug_on(sbi, im->ino_num == 0);
601 spin_unlock(&im->ino_lock);
604 void f2fs_add_orphan_inode(struct inode *inode)
606 /* add new orphan ino entry into list */
607 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
608 f2fs_update_inode_page(inode);
611 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
613 /* remove orphan entry from orphan list */
614 __remove_ino_entry(sbi, ino, ORPHAN_INO);
617 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
623 inode = f2fs_iget_retry(sbi->sb, ino);
626 * there should be a bug that we can't find the entry
629 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
630 return PTR_ERR(inode);
633 err = dquot_initialize(inode);
641 /* truncate all the data during iput */
644 err = f2fs_get_node_info(sbi, ino, &ni);
648 /* ENOMEM was fully retried in f2fs_evict_inode. */
649 if (ni.blk_addr != NULL_ADDR) {
656 set_sbi_flag(sbi, SBI_NEED_FSCK);
657 f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
662 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
664 block_t start_blk, orphan_blocks, i, j;
665 unsigned int s_flags = sbi->sb->s_flags;
671 if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
674 if (bdev_read_only(sbi->sb->s_bdev)) {
675 f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
679 if (s_flags & SB_RDONLY) {
680 f2fs_info(sbi, "orphan cleanup on readonly fs");
681 sbi->sb->s_flags &= ~SB_RDONLY;
685 /* Needed for iput() to work correctly and not trash data */
686 sbi->sb->s_flags |= SB_ACTIVE;
689 * Turn on quotas which were not enabled for read-only mounts if
690 * filesystem has quota feature, so that they are updated correctly.
692 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
695 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
696 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
698 f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
700 for (i = 0; i < orphan_blocks; i++) {
702 struct f2fs_orphan_block *orphan_blk;
704 page = f2fs_get_meta_page(sbi, start_blk + i);
710 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
711 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
712 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
713 err = recover_orphan_inode(sbi, ino);
715 f2fs_put_page(page, 1);
719 f2fs_put_page(page, 1);
721 /* clear Orphan Flag */
722 clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
724 set_sbi_flag(sbi, SBI_IS_RECOVERED);
727 /* Turn quotas off */
729 f2fs_quota_off_umount(sbi->sb);
731 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
736 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
738 struct list_head *head;
739 struct f2fs_orphan_block *orphan_blk = NULL;
740 unsigned int nentries = 0;
741 unsigned short index = 1;
742 unsigned short orphan_blocks;
743 struct page *page = NULL;
744 struct ino_entry *orphan = NULL;
745 struct inode_management *im = &sbi->im[ORPHAN_INO];
747 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
750 * we don't need to do spin_lock(&im->ino_lock) here, since all the
751 * orphan inode operations are covered under f2fs_lock_op().
752 * And, spin_lock should be avoided due to page operations below.
754 head = &im->ino_list;
756 /* loop for each orphan inode entry and write them in Jornal block */
757 list_for_each_entry(orphan, head, list) {
759 page = f2fs_grab_meta_page(sbi, start_blk++);
761 (struct f2fs_orphan_block *)page_address(page);
762 memset(orphan_blk, 0, sizeof(*orphan_blk));
765 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
767 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
769 * an orphan block is full of 1020 entries,
770 * then we need to flush current orphan blocks
771 * and bring another one in memory
773 orphan_blk->blk_addr = cpu_to_le16(index);
774 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
775 orphan_blk->entry_count = cpu_to_le32(nentries);
776 set_page_dirty(page);
777 f2fs_put_page(page, 1);
785 orphan_blk->blk_addr = cpu_to_le16(index);
786 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
787 orphan_blk->entry_count = cpu_to_le32(nentries);
788 set_page_dirty(page);
789 f2fs_put_page(page, 1);
793 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
794 struct f2fs_checkpoint *ckpt)
796 unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
799 chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
800 if (chksum_ofs < CP_CHKSUM_OFFSET) {
801 chksum_ofs += sizeof(chksum);
802 chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
803 F2FS_BLKSIZE - chksum_ofs);
808 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
809 struct f2fs_checkpoint **cp_block, struct page **cp_page,
810 unsigned long long *version)
812 size_t crc_offset = 0;
815 *cp_page = f2fs_get_meta_page(sbi, cp_addr);
816 if (IS_ERR(*cp_page))
817 return PTR_ERR(*cp_page);
819 *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
821 crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
822 if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
823 crc_offset > CP_CHKSUM_OFFSET) {
824 f2fs_put_page(*cp_page, 1);
825 f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
829 crc = f2fs_checkpoint_chksum(sbi, *cp_block);
830 if (crc != cur_cp_crc(*cp_block)) {
831 f2fs_put_page(*cp_page, 1);
832 f2fs_warn(sbi, "invalid crc value");
836 *version = cur_cp_version(*cp_block);
840 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
841 block_t cp_addr, unsigned long long *version)
843 struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
844 struct f2fs_checkpoint *cp_block = NULL;
845 unsigned long long cur_version = 0, pre_version = 0;
848 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
849 &cp_page_1, version);
853 if (le32_to_cpu(cp_block->cp_pack_total_block_count) >
854 sbi->blocks_per_seg) {
855 f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
856 le32_to_cpu(cp_block->cp_pack_total_block_count));
859 pre_version = *version;
861 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
862 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
863 &cp_page_2, version);
866 cur_version = *version;
868 if (cur_version == pre_version) {
869 *version = cur_version;
870 f2fs_put_page(cp_page_2, 1);
873 f2fs_put_page(cp_page_2, 1);
875 f2fs_put_page(cp_page_1, 1);
879 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
881 struct f2fs_checkpoint *cp_block;
882 struct f2fs_super_block *fsb = sbi->raw_super;
883 struct page *cp1, *cp2, *cur_page;
884 unsigned long blk_size = sbi->blocksize;
885 unsigned long long cp1_version = 0, cp2_version = 0;
886 unsigned long long cp_start_blk_no;
887 unsigned int cp_blks = 1 + __cp_payload(sbi);
892 sbi->ckpt = f2fs_kzalloc(sbi, array_size(blk_size, cp_blks),
897 * Finding out valid cp block involves read both
898 * sets( cp pack 1 and cp pack 2)
900 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
901 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
903 /* The second checkpoint pack should start at the next segment */
904 cp_start_blk_no += ((unsigned long long)1) <<
905 le32_to_cpu(fsb->log_blocks_per_seg);
906 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
909 if (ver_after(cp2_version, cp1_version))
922 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
923 memcpy(sbi->ckpt, cp_block, blk_size);
926 sbi->cur_cp_pack = 1;
928 sbi->cur_cp_pack = 2;
930 /* Sanity checking of checkpoint */
931 if (f2fs_sanity_check_ckpt(sbi)) {
933 goto free_fail_no_cp;
939 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
941 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
943 for (i = 1; i < cp_blks; i++) {
944 void *sit_bitmap_ptr;
945 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
947 cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
948 if (IS_ERR(cur_page)) {
949 err = PTR_ERR(cur_page);
950 goto free_fail_no_cp;
952 sit_bitmap_ptr = page_address(cur_page);
953 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
954 f2fs_put_page(cur_page, 1);
957 f2fs_put_page(cp1, 1);
958 f2fs_put_page(cp2, 1);
962 f2fs_put_page(cp1, 1);
963 f2fs_put_page(cp2, 1);
969 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
971 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
972 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
974 if (is_inode_flag_set(inode, flag))
977 set_inode_flag(inode, flag);
978 if (!f2fs_is_volatile_file(inode))
979 list_add_tail(&F2FS_I(inode)->dirty_list,
980 &sbi->inode_list[type]);
981 stat_inc_dirty_inode(sbi, type);
984 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
986 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
988 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
991 list_del_init(&F2FS_I(inode)->dirty_list);
992 clear_inode_flag(inode, flag);
993 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
996 void f2fs_update_dirty_page(struct inode *inode, struct page *page)
998 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
999 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1001 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1002 !S_ISLNK(inode->i_mode))
1005 spin_lock(&sbi->inode_lock[type]);
1006 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1007 __add_dirty_inode(inode, type);
1008 inode_inc_dirty_pages(inode);
1009 spin_unlock(&sbi->inode_lock[type]);
1011 f2fs_set_page_private(page, 0);
1012 f2fs_trace_pid(page);
1015 void f2fs_remove_dirty_inode(struct inode *inode)
1017 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1018 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1020 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1021 !S_ISLNK(inode->i_mode))
1024 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1027 spin_lock(&sbi->inode_lock[type]);
1028 __remove_dirty_inode(inode, type);
1029 spin_unlock(&sbi->inode_lock[type]);
1032 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
1034 struct list_head *head;
1035 struct inode *inode;
1036 struct f2fs_inode_info *fi;
1037 bool is_dir = (type == DIR_INODE);
1038 unsigned long ino = 0;
1040 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1041 get_pages(sbi, is_dir ?
1042 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1044 if (unlikely(f2fs_cp_error(sbi)))
1047 spin_lock(&sbi->inode_lock[type]);
1049 head = &sbi->inode_list[type];
1050 if (list_empty(head)) {
1051 spin_unlock(&sbi->inode_lock[type]);
1052 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1053 get_pages(sbi, is_dir ?
1054 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1057 fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1058 inode = igrab(&fi->vfs_inode);
1059 spin_unlock(&sbi->inode_lock[type]);
1061 unsigned long cur_ino = inode->i_ino;
1063 F2FS_I(inode)->cp_task = current;
1065 filemap_fdatawrite(inode->i_mapping);
1067 F2FS_I(inode)->cp_task = NULL;
1070 /* We need to give cpu to another writers. */
1077 * We should submit bio, since it exists several
1078 * wribacking dentry pages in the freeing inode.
1080 f2fs_submit_merged_write(sbi, DATA);
1086 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1088 struct list_head *head = &sbi->inode_list[DIRTY_META];
1089 struct inode *inode;
1090 struct f2fs_inode_info *fi;
1091 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1094 if (unlikely(f2fs_cp_error(sbi)))
1097 spin_lock(&sbi->inode_lock[DIRTY_META]);
1098 if (list_empty(head)) {
1099 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1102 fi = list_first_entry(head, struct f2fs_inode_info,
1104 inode = igrab(&fi->vfs_inode);
1105 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1107 sync_inode_metadata(inode, 0);
1109 /* it's on eviction */
1110 if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1111 f2fs_update_inode_page(inode);
1118 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1120 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1121 struct f2fs_nm_info *nm_i = NM_I(sbi);
1122 nid_t last_nid = nm_i->next_scan_nid;
1124 next_free_nid(sbi, &last_nid);
1125 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1126 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1127 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1128 ckpt->next_free_nid = cpu_to_le32(last_nid);
1131 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1135 if (!is_journalled_quota(sbi))
1138 down_write(&sbi->quota_sem);
1139 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1141 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1143 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1144 clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1146 } else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1149 up_write(&sbi->quota_sem);
1154 * Freeze all the FS-operations for checkpoint.
1156 static int block_operations(struct f2fs_sb_info *sbi)
1158 struct writeback_control wbc = {
1159 .sync_mode = WB_SYNC_ALL,
1160 .nr_to_write = LONG_MAX,
1163 struct blk_plug plug;
1164 int err = 0, cnt = 0;
1166 blk_start_plug(&plug);
1170 if (__need_flush_quota(sbi)) {
1173 if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1174 set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1175 set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1176 goto retry_flush_dents;
1178 f2fs_unlock_all(sbi);
1180 /* only failed during mount/umount/freeze/quotactl */
1181 locked = down_read_trylock(&sbi->sb->s_umount);
1182 f2fs_quota_sync(sbi->sb, -1);
1184 up_read(&sbi->sb->s_umount);
1186 goto retry_flush_quotas;
1190 /* write all the dirty dentry pages */
1191 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1192 f2fs_unlock_all(sbi);
1193 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE);
1197 goto retry_flush_quotas;
1201 * POR: we should ensure that there are no dirty node pages
1202 * until finishing nat/sit flush. inode->i_blocks can be updated.
1204 down_write(&sbi->node_change);
1206 if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1207 up_write(&sbi->node_change);
1208 f2fs_unlock_all(sbi);
1209 err = f2fs_sync_inode_meta(sbi);
1213 goto retry_flush_quotas;
1217 down_write(&sbi->node_write);
1219 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1220 up_write(&sbi->node_write);
1221 atomic_inc(&sbi->wb_sync_req[NODE]);
1222 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1223 atomic_dec(&sbi->wb_sync_req[NODE]);
1225 up_write(&sbi->node_change);
1226 f2fs_unlock_all(sbi);
1230 goto retry_flush_nodes;
1234 * sbi->node_change is used only for AIO write_begin path which produces
1235 * dirty node blocks and some checkpoint values by block allocation.
1237 __prepare_cp_block(sbi);
1238 up_write(&sbi->node_change);
1240 blk_finish_plug(&plug);
1244 static void unblock_operations(struct f2fs_sb_info *sbi)
1246 up_write(&sbi->node_write);
1247 f2fs_unlock_all(sbi);
1250 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1255 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1257 if (!get_pages(sbi, type))
1260 if (unlikely(f2fs_cp_error(sbi)))
1263 io_schedule_timeout(HZ/50);
1265 finish_wait(&sbi->cp_wait, &wait);
1268 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1270 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1271 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1272 unsigned long flags;
1274 spin_lock_irqsave(&sbi->cp_lock, flags);
1276 if ((cpc->reason & CP_UMOUNT) &&
1277 le32_to_cpu(ckpt->cp_pack_total_block_count) >
1278 sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1279 disable_nat_bits(sbi, false);
1281 if (cpc->reason & CP_TRIMMED)
1282 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1284 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1286 if (cpc->reason & CP_UMOUNT)
1287 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1289 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1291 if (cpc->reason & CP_FASTBOOT)
1292 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1294 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1297 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1299 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1301 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK) ||
1302 is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1303 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1305 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1306 __set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1308 __clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1310 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1311 __set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1313 __clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1315 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1316 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1318 __clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1320 if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1321 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1323 /* set this flag to activate crc|cp_ver for recovery */
1324 __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1325 __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1327 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1330 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1331 void *src, block_t blk_addr)
1333 struct writeback_control wbc = {
1338 * pagevec_lookup_tag and lock_page again will take
1339 * some extra time. Therefore, f2fs_update_meta_pages and
1340 * f2fs_sync_meta_pages are combined in this function.
1342 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1345 f2fs_wait_on_page_writeback(page, META, true, true);
1347 memcpy(page_address(page), src, PAGE_SIZE);
1349 set_page_dirty(page);
1350 if (unlikely(!clear_page_dirty_for_io(page)))
1351 f2fs_bug_on(sbi, 1);
1353 /* writeout cp pack 2 page */
1354 err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1355 if (unlikely(err && f2fs_cp_error(sbi))) {
1356 f2fs_put_page(page, 1);
1360 f2fs_bug_on(sbi, err);
1361 f2fs_put_page(page, 0);
1363 /* submit checkpoint (with barrier if NOBARRIER is not set) */
1364 f2fs_submit_merged_write(sbi, META_FLUSH);
1367 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1369 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1370 struct f2fs_nm_info *nm_i = NM_I(sbi);
1371 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1373 unsigned int data_sum_blocks, orphan_blocks;
1376 int cp_payload_blks = __cp_payload(sbi);
1377 struct super_block *sb = sbi->sb;
1378 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1382 /* Flush all the NAT/SIT pages */
1383 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1385 /* start to update checkpoint, cp ver is already updated previously */
1386 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1387 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1388 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1389 ckpt->cur_node_segno[i] =
1390 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1391 ckpt->cur_node_blkoff[i] =
1392 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1393 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1394 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1396 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1397 ckpt->cur_data_segno[i] =
1398 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1399 ckpt->cur_data_blkoff[i] =
1400 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1401 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1402 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1405 /* 2 cp + n data seg summary + orphan inode blocks */
1406 data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1407 spin_lock_irqsave(&sbi->cp_lock, flags);
1408 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1409 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1411 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1412 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1414 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1415 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1418 if (__remain_node_summaries(cpc->reason))
1419 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1420 cp_payload_blks + data_sum_blocks +
1421 orphan_blocks + NR_CURSEG_NODE_TYPE);
1423 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1424 cp_payload_blks + data_sum_blocks +
1427 /* update ckpt flag for checkpoint */
1428 update_ckpt_flags(sbi, cpc);
1430 /* update SIT/NAT bitmap */
1431 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1432 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1434 crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1435 *((__le32 *)((unsigned char *)ckpt +
1436 le32_to_cpu(ckpt->checksum_offset)))
1437 = cpu_to_le32(crc32);
1439 start_blk = __start_cp_next_addr(sbi);
1441 /* write nat bits */
1442 if (enabled_nat_bits(sbi, cpc)) {
1443 __u64 cp_ver = cur_cp_version(ckpt);
1446 cp_ver |= ((__u64)crc32 << 32);
1447 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1449 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1450 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1451 f2fs_update_meta_page(sbi, nm_i->nat_bits +
1452 (i << F2FS_BLKSIZE_BITS), blk + i);
1455 /* write out checkpoint buffer at block 0 */
1456 f2fs_update_meta_page(sbi, ckpt, start_blk++);
1458 for (i = 1; i < 1 + cp_payload_blks; i++)
1459 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1463 write_orphan_inodes(sbi, start_blk);
1464 start_blk += orphan_blocks;
1467 f2fs_write_data_summaries(sbi, start_blk);
1468 start_blk += data_sum_blocks;
1470 /* Record write statistics in the hot node summary */
1471 kbytes_written = sbi->kbytes_written;
1472 if (sb->s_bdev->bd_part)
1473 kbytes_written += BD_PART_WRITTEN(sbi);
1475 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1477 if (__remain_node_summaries(cpc->reason)) {
1478 f2fs_write_node_summaries(sbi, start_blk);
1479 start_blk += NR_CURSEG_NODE_TYPE;
1482 /* update user_block_counts */
1483 sbi->last_valid_block_count = sbi->total_valid_block_count;
1484 percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1486 /* Here, we have one bio having CP pack except cp pack 2 page */
1487 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1488 /* Wait for all dirty meta pages to be submitted for IO */
1489 f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1491 /* wait for previous submitted meta pages writeback */
1492 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1494 /* flush all device cache */
1495 err = f2fs_flush_device_cache(sbi);
1499 /* barrier and flush checkpoint cp pack 2 page if it can */
1500 commit_checkpoint(sbi, ckpt, start_blk);
1501 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1504 * invalidate intermediate page cache borrowed from meta inode which are
1505 * used for migration of encrypted or verity inode's blocks.
1507 if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi))
1508 invalidate_mapping_pages(META_MAPPING(sbi),
1509 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1511 f2fs_release_ino_entry(sbi, false);
1513 f2fs_reset_fsync_node_info(sbi);
1515 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1516 clear_sbi_flag(sbi, SBI_NEED_CP);
1517 clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1519 spin_lock(&sbi->stat_lock);
1520 sbi->unusable_block_count = 0;
1521 spin_unlock(&sbi->stat_lock);
1523 __set_cp_next_pack(sbi);
1526 * redirty superblock if metadata like node page or inode cache is
1527 * updated during writing checkpoint.
1529 if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1530 get_pages(sbi, F2FS_DIRTY_IMETA))
1531 set_sbi_flag(sbi, SBI_IS_DIRTY);
1533 f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1535 return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1538 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1540 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1541 unsigned long long ckpt_ver;
1544 if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1547 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1548 if (cpc->reason != CP_PAUSE)
1550 f2fs_warn(sbi, "Start checkpoint disabled!");
1552 mutex_lock(&sbi->cp_mutex);
1554 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1555 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1556 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1558 if (unlikely(f2fs_cp_error(sbi))) {
1563 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1565 err = block_operations(sbi);
1569 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1571 f2fs_flush_merged_writes(sbi);
1573 /* this is the case of multiple fstrims without any changes */
1574 if (cpc->reason & CP_DISCARD) {
1575 if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1576 unblock_operations(sbi);
1580 if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1581 SIT_I(sbi)->dirty_sentries == 0 &&
1582 prefree_segments(sbi) == 0) {
1583 f2fs_flush_sit_entries(sbi, cpc);
1584 f2fs_clear_prefree_segments(sbi, cpc);
1585 unblock_operations(sbi);
1591 * update checkpoint pack index
1592 * Increase the version number so that
1593 * SIT entries and seg summaries are written at correct place
1595 ckpt_ver = cur_cp_version(ckpt);
1596 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1598 /* write cached NAT/SIT entries to NAT/SIT area */
1599 err = f2fs_flush_nat_entries(sbi, cpc);
1603 f2fs_flush_sit_entries(sbi, cpc);
1605 err = do_checkpoint(sbi, cpc);
1607 f2fs_release_discard_addrs(sbi);
1609 f2fs_clear_prefree_segments(sbi, cpc);
1611 unblock_operations(sbi);
1612 stat_inc_cp_count(sbi->stat_info);
1614 if (cpc->reason & CP_RECOVERY)
1615 f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1617 /* update CP_TIME to trigger checkpoint periodically */
1618 f2fs_update_time(sbi, CP_TIME);
1619 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1621 mutex_unlock(&sbi->cp_mutex);
1625 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1629 for (i = 0; i < MAX_INO_ENTRY; i++) {
1630 struct inode_management *im = &sbi->im[i];
1632 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1633 spin_lock_init(&im->ino_lock);
1634 INIT_LIST_HEAD(&im->ino_list);
1638 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1639 NR_CURSEG_TYPE - __cp_payload(sbi)) *
1640 F2FS_ORPHANS_PER_BLOCK;
1643 int __init f2fs_create_checkpoint_caches(void)
1645 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1646 sizeof(struct ino_entry));
1647 if (!ino_entry_slab)
1649 f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1650 sizeof(struct inode_entry));
1651 if (!f2fs_inode_entry_slab) {
1652 kmem_cache_destroy(ino_entry_slab);
1658 void f2fs_destroy_checkpoint_caches(void)
1660 kmem_cache_destroy(ino_entry_slab);
1661 kmem_cache_destroy(f2fs_inode_entry_slab);