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);
89 f2fs_update_iostat(sbi, FS_META_READ_IO, F2FS_BLKSIZE);
92 if (unlikely(page->mapping != mapping)) {
93 f2fs_put_page(page, 1);
97 if (unlikely(!PageUptodate(page))) {
98 f2fs_put_page(page, 1);
105 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
107 return __get_meta_page(sbi, index, true);
110 struct page *f2fs_get_meta_page_retry(struct f2fs_sb_info *sbi, pgoff_t index)
116 page = __get_meta_page(sbi, index, true);
118 if (PTR_ERR(page) == -EIO &&
119 ++count <= DEFAULT_RETRY_IO_COUNT)
121 f2fs_stop_checkpoint(sbi, false);
127 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
129 return __get_meta_page(sbi, index, false);
132 static bool __is_bitmap_valid(struct f2fs_sb_info *sbi, block_t blkaddr,
135 struct seg_entry *se;
136 unsigned int segno, offset;
139 if (type == DATA_GENERIC)
142 segno = GET_SEGNO(sbi, blkaddr);
143 offset = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
144 se = get_seg_entry(sbi, segno);
146 exist = f2fs_test_bit(offset, se->cur_valid_map);
147 if (exist && type == DATA_GENERIC_ENHANCE_UPDATE) {
148 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
150 set_sbi_flag(sbi, SBI_NEED_FSCK);
154 if (!exist && type == DATA_GENERIC_ENHANCE) {
155 f2fs_err(sbi, "Inconsistent error blkaddr:%u, sit bitmap:%d",
157 set_sbi_flag(sbi, SBI_NEED_FSCK);
163 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
164 block_t blkaddr, int type)
170 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
174 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
175 blkaddr < SM_I(sbi)->ssa_blkaddr))
179 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
180 blkaddr < __start_cp_addr(sbi)))
184 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
185 blkaddr < MAIN_BLKADDR(sbi)))
189 case DATA_GENERIC_ENHANCE:
190 case DATA_GENERIC_ENHANCE_READ:
191 case DATA_GENERIC_ENHANCE_UPDATE:
192 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
193 blkaddr < MAIN_BLKADDR(sbi))) {
194 f2fs_warn(sbi, "access invalid blkaddr:%u",
196 set_sbi_flag(sbi, SBI_NEED_FSCK);
200 return __is_bitmap_valid(sbi, blkaddr, type);
204 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
205 blkaddr >= MAIN_BLKADDR(sbi)))
216 * Readahead CP/NAT/SIT/SSA/POR pages
218 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
222 block_t blkno = start;
223 struct f2fs_io_info fio = {
227 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
228 .encrypted_page = NULL,
230 .is_por = (type == META_POR),
232 struct blk_plug plug;
235 if (unlikely(type == META_POR))
236 fio.op_flags &= ~REQ_META;
238 blk_start_plug(&plug);
239 for (; nrpages-- > 0; blkno++) {
241 if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
246 if (unlikely(blkno >=
247 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
249 /* get nat block addr */
250 fio.new_blkaddr = current_nat_addr(sbi,
251 blkno * NAT_ENTRY_PER_BLOCK);
254 if (unlikely(blkno >= TOTAL_SEGS(sbi)))
256 /* get sit block addr */
257 fio.new_blkaddr = current_sit_addr(sbi,
258 blkno * SIT_ENTRY_PER_BLOCK);
263 fio.new_blkaddr = blkno;
269 page = f2fs_grab_cache_page(META_MAPPING(sbi),
270 fio.new_blkaddr, false);
273 if (PageUptodate(page)) {
274 f2fs_put_page(page, 1);
279 err = f2fs_submit_page_bio(&fio);
280 f2fs_put_page(page, err ? 1 : 0);
283 f2fs_update_iostat(sbi, FS_META_READ_IO, F2FS_BLKSIZE);
286 blk_finish_plug(&plug);
287 return blkno - start;
290 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
293 bool readahead = false;
295 page = find_get_page(META_MAPPING(sbi), index);
296 if (!page || !PageUptodate(page))
298 f2fs_put_page(page, 0);
301 f2fs_ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
304 static int __f2fs_write_meta_page(struct page *page,
305 struct writeback_control *wbc,
306 enum iostat_type io_type)
308 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
310 trace_f2fs_writepage(page, META);
312 if (unlikely(f2fs_cp_error(sbi))) {
313 if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
314 ClearPageUptodate(page);
315 dec_page_count(sbi, F2FS_DIRTY_META);
321 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
323 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
326 f2fs_do_write_meta_page(sbi, page, io_type);
327 dec_page_count(sbi, F2FS_DIRTY_META);
329 if (wbc->for_reclaim)
330 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, META);
334 if (unlikely(f2fs_cp_error(sbi)))
335 f2fs_submit_merged_write(sbi, META);
340 redirty_page_for_writepage(wbc, page);
341 return AOP_WRITEPAGE_ACTIVATE;
344 static int f2fs_write_meta_page(struct page *page,
345 struct writeback_control *wbc)
347 return __f2fs_write_meta_page(page, wbc, FS_META_IO);
350 static int f2fs_write_meta_pages(struct address_space *mapping,
351 struct writeback_control *wbc)
353 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
356 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
359 /* collect a number of dirty meta pages and write together */
360 if (wbc->sync_mode != WB_SYNC_ALL &&
361 get_pages(sbi, F2FS_DIRTY_META) <
362 nr_pages_to_skip(sbi, META))
365 /* if locked failed, cp will flush dirty pages instead */
366 if (!mutex_trylock(&sbi->cp_mutex))
369 trace_f2fs_writepages(mapping->host, wbc, META);
370 diff = nr_pages_to_write(sbi, META, wbc);
371 written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
372 mutex_unlock(&sbi->cp_mutex);
373 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
377 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
378 trace_f2fs_writepages(mapping->host, wbc, META);
382 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
383 long nr_to_write, enum iostat_type io_type)
385 struct address_space *mapping = META_MAPPING(sbi);
386 pgoff_t index = 0, prev = ULONG_MAX;
390 struct writeback_control wbc = {
393 struct blk_plug plug;
397 blk_start_plug(&plug);
399 while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
400 PAGECACHE_TAG_DIRTY))) {
403 for (i = 0; i < nr_pages; i++) {
404 struct page *page = pvec.pages[i];
406 if (prev == ULONG_MAX)
407 prev = page->index - 1;
408 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
409 pagevec_release(&pvec);
415 if (unlikely(page->mapping != mapping)) {
420 if (!PageDirty(page)) {
421 /* someone wrote it for us */
422 goto continue_unlock;
425 f2fs_wait_on_page_writeback(page, META, true, true);
427 if (!clear_page_dirty_for_io(page))
428 goto continue_unlock;
430 if (__f2fs_write_meta_page(page, &wbc, io_type)) {
436 if (unlikely(nwritten >= nr_to_write))
439 pagevec_release(&pvec);
444 f2fs_submit_merged_write(sbi, type);
446 blk_finish_plug(&plug);
451 static int f2fs_set_meta_page_dirty(struct page *page)
453 trace_f2fs_set_page_dirty(page, META);
455 if (!PageUptodate(page))
456 SetPageUptodate(page);
457 if (!PageDirty(page)) {
458 __set_page_dirty_nobuffers(page);
459 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
460 f2fs_set_page_private(page, 0);
461 f2fs_trace_pid(page);
467 const struct address_space_operations f2fs_meta_aops = {
468 .writepage = f2fs_write_meta_page,
469 .writepages = f2fs_write_meta_pages,
470 .set_page_dirty = f2fs_set_meta_page_dirty,
471 .invalidatepage = f2fs_invalidate_page,
472 .releasepage = f2fs_release_page,
473 #ifdef CONFIG_MIGRATION
474 .migratepage = f2fs_migrate_page,
478 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
479 unsigned int devidx, int type)
481 struct inode_management *im = &sbi->im[type];
482 struct ino_entry *e, *tmp;
484 tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
486 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
488 spin_lock(&im->ino_lock);
489 e = radix_tree_lookup(&im->ino_root, ino);
492 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
495 memset(e, 0, sizeof(struct ino_entry));
498 list_add_tail(&e->list, &im->ino_list);
499 if (type != ORPHAN_INO)
503 if (type == FLUSH_INO)
504 f2fs_set_bit(devidx, (char *)&e->dirty_device);
506 spin_unlock(&im->ino_lock);
507 radix_tree_preload_end();
510 kmem_cache_free(ino_entry_slab, tmp);
513 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
515 struct inode_management *im = &sbi->im[type];
518 spin_lock(&im->ino_lock);
519 e = radix_tree_lookup(&im->ino_root, ino);
522 radix_tree_delete(&im->ino_root, ino);
524 spin_unlock(&im->ino_lock);
525 kmem_cache_free(ino_entry_slab, e);
528 spin_unlock(&im->ino_lock);
531 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
533 /* add new dirty ino entry into list */
534 __add_ino_entry(sbi, ino, 0, type);
537 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
539 /* remove dirty ino entry from list */
540 __remove_ino_entry(sbi, ino, type);
543 /* mode should be APPEND_INO, UPDATE_INO or TRANS_DIR_INO */
544 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
546 struct inode_management *im = &sbi->im[mode];
549 spin_lock(&im->ino_lock);
550 e = radix_tree_lookup(&im->ino_root, ino);
551 spin_unlock(&im->ino_lock);
552 return e ? true : false;
555 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
557 struct ino_entry *e, *tmp;
560 for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
561 struct inode_management *im = &sbi->im[i];
563 spin_lock(&im->ino_lock);
564 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
566 radix_tree_delete(&im->ino_root, e->ino);
567 kmem_cache_free(ino_entry_slab, e);
570 spin_unlock(&im->ino_lock);
574 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
575 unsigned int devidx, int type)
577 __add_ino_entry(sbi, ino, devidx, type);
580 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
581 unsigned int devidx, int type)
583 struct inode_management *im = &sbi->im[type];
585 bool is_dirty = false;
587 spin_lock(&im->ino_lock);
588 e = radix_tree_lookup(&im->ino_root, ino);
589 if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
591 spin_unlock(&im->ino_lock);
595 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
597 struct inode_management *im = &sbi->im[ORPHAN_INO];
600 spin_lock(&im->ino_lock);
602 if (time_to_inject(sbi, FAULT_ORPHAN)) {
603 spin_unlock(&im->ino_lock);
604 f2fs_show_injection_info(sbi, FAULT_ORPHAN);
608 if (unlikely(im->ino_num >= sbi->max_orphans))
612 spin_unlock(&im->ino_lock);
617 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
619 struct inode_management *im = &sbi->im[ORPHAN_INO];
621 spin_lock(&im->ino_lock);
622 f2fs_bug_on(sbi, im->ino_num == 0);
624 spin_unlock(&im->ino_lock);
627 void f2fs_add_orphan_inode(struct inode *inode)
629 /* add new orphan ino entry into list */
630 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
631 f2fs_update_inode_page(inode);
634 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
636 /* remove orphan entry from orphan list */
637 __remove_ino_entry(sbi, ino, ORPHAN_INO);
640 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
646 inode = f2fs_iget_retry(sbi->sb, ino);
649 * there should be a bug that we can't find the entry
652 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
653 return PTR_ERR(inode);
656 err = dquot_initialize(inode);
664 /* truncate all the data during iput */
667 err = f2fs_get_node_info(sbi, ino, &ni);
671 /* ENOMEM was fully retried in f2fs_evict_inode. */
672 if (ni.blk_addr != NULL_ADDR) {
679 set_sbi_flag(sbi, SBI_NEED_FSCK);
680 f2fs_warn(sbi, "%s: orphan failed (ino=%x), run fsck to fix.",
685 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
687 block_t start_blk, orphan_blocks, i, j;
688 unsigned int s_flags = sbi->sb->s_flags;
694 if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
697 if (bdev_read_only(sbi->sb->s_bdev)) {
698 f2fs_info(sbi, "write access unavailable, skipping orphan cleanup");
702 if (s_flags & SB_RDONLY) {
703 f2fs_info(sbi, "orphan cleanup on readonly fs");
704 sbi->sb->s_flags &= ~SB_RDONLY;
708 /* Needed for iput() to work correctly and not trash data */
709 sbi->sb->s_flags |= SB_ACTIVE;
712 * Turn on quotas which were not enabled for read-only mounts if
713 * filesystem has quota feature, so that they are updated correctly.
715 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
718 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
719 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
721 f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
723 for (i = 0; i < orphan_blocks; i++) {
725 struct f2fs_orphan_block *orphan_blk;
727 page = f2fs_get_meta_page(sbi, start_blk + i);
733 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
734 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
735 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
736 err = recover_orphan_inode(sbi, ino);
738 f2fs_put_page(page, 1);
742 f2fs_put_page(page, 1);
744 /* clear Orphan Flag */
745 clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
747 set_sbi_flag(sbi, SBI_IS_RECOVERED);
750 /* Turn quotas off */
752 f2fs_quota_off_umount(sbi->sb);
754 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
759 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
761 struct list_head *head;
762 struct f2fs_orphan_block *orphan_blk = NULL;
763 unsigned int nentries = 0;
764 unsigned short index = 1;
765 unsigned short orphan_blocks;
766 struct page *page = NULL;
767 struct ino_entry *orphan = NULL;
768 struct inode_management *im = &sbi->im[ORPHAN_INO];
770 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
773 * we don't need to do spin_lock(&im->ino_lock) here, since all the
774 * orphan inode operations are covered under f2fs_lock_op().
775 * And, spin_lock should be avoided due to page operations below.
777 head = &im->ino_list;
779 /* loop for each orphan inode entry and write them in Jornal block */
780 list_for_each_entry(orphan, head, list) {
782 page = f2fs_grab_meta_page(sbi, start_blk++);
784 (struct f2fs_orphan_block *)page_address(page);
785 memset(orphan_blk, 0, sizeof(*orphan_blk));
788 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
790 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
792 * an orphan block is full of 1020 entries,
793 * then we need to flush current orphan blocks
794 * and bring another one in memory
796 orphan_blk->blk_addr = cpu_to_le16(index);
797 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
798 orphan_blk->entry_count = cpu_to_le32(nentries);
799 set_page_dirty(page);
800 f2fs_put_page(page, 1);
808 orphan_blk->blk_addr = cpu_to_le16(index);
809 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
810 orphan_blk->entry_count = cpu_to_le32(nentries);
811 set_page_dirty(page);
812 f2fs_put_page(page, 1);
816 static __u32 f2fs_checkpoint_chksum(struct f2fs_sb_info *sbi,
817 struct f2fs_checkpoint *ckpt)
819 unsigned int chksum_ofs = le32_to_cpu(ckpt->checksum_offset);
822 chksum = f2fs_crc32(sbi, ckpt, chksum_ofs);
823 if (chksum_ofs < CP_CHKSUM_OFFSET) {
824 chksum_ofs += sizeof(chksum);
825 chksum = f2fs_chksum(sbi, chksum, (__u8 *)ckpt + chksum_ofs,
826 F2FS_BLKSIZE - chksum_ofs);
831 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
832 struct f2fs_checkpoint **cp_block, struct page **cp_page,
833 unsigned long long *version)
835 size_t crc_offset = 0;
838 *cp_page = f2fs_get_meta_page(sbi, cp_addr);
839 if (IS_ERR(*cp_page))
840 return PTR_ERR(*cp_page);
842 *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
844 crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
845 if (crc_offset < CP_MIN_CHKSUM_OFFSET ||
846 crc_offset > CP_CHKSUM_OFFSET) {
847 f2fs_put_page(*cp_page, 1);
848 f2fs_warn(sbi, "invalid crc_offset: %zu", crc_offset);
852 crc = f2fs_checkpoint_chksum(sbi, *cp_block);
853 if (crc != cur_cp_crc(*cp_block)) {
854 f2fs_put_page(*cp_page, 1);
855 f2fs_warn(sbi, "invalid crc value");
859 *version = cur_cp_version(*cp_block);
863 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
864 block_t cp_addr, unsigned long long *version)
866 struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
867 struct f2fs_checkpoint *cp_block = NULL;
868 unsigned long long cur_version = 0, pre_version = 0;
869 unsigned int cp_blocks;
872 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
873 &cp_page_1, version);
877 cp_blocks = le32_to_cpu(cp_block->cp_pack_total_block_count);
879 if (cp_blocks > sbi->blocks_per_seg || cp_blocks <= F2FS_CP_PACKS) {
880 f2fs_warn(sbi, "invalid cp_pack_total_block_count:%u",
881 le32_to_cpu(cp_block->cp_pack_total_block_count));
884 pre_version = *version;
886 cp_addr += cp_blocks - 1;
887 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
888 &cp_page_2, version);
891 cur_version = *version;
893 if (cur_version == pre_version) {
894 *version = cur_version;
895 f2fs_put_page(cp_page_2, 1);
898 f2fs_put_page(cp_page_2, 1);
900 f2fs_put_page(cp_page_1, 1);
904 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
906 struct f2fs_checkpoint *cp_block;
907 struct f2fs_super_block *fsb = sbi->raw_super;
908 struct page *cp1, *cp2, *cur_page;
909 unsigned long blk_size = sbi->blocksize;
910 unsigned long long cp1_version = 0, cp2_version = 0;
911 unsigned long long cp_start_blk_no;
912 unsigned int cp_blks = 1 + __cp_payload(sbi);
917 sbi->ckpt = f2fs_kvzalloc(sbi, array_size(blk_size, cp_blks),
922 * Finding out valid cp block involves read both
923 * sets( cp pack 1 and cp pack 2)
925 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
926 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
928 /* The second checkpoint pack should start at the next segment */
929 cp_start_blk_no += ((unsigned long long)1) <<
930 le32_to_cpu(fsb->log_blocks_per_seg);
931 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
934 if (ver_after(cp2_version, cp1_version))
947 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
948 memcpy(sbi->ckpt, cp_block, blk_size);
951 sbi->cur_cp_pack = 1;
953 sbi->cur_cp_pack = 2;
955 /* Sanity checking of checkpoint */
956 if (f2fs_sanity_check_ckpt(sbi)) {
958 goto free_fail_no_cp;
964 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
966 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
968 for (i = 1; i < cp_blks; i++) {
969 void *sit_bitmap_ptr;
970 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
972 cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
973 if (IS_ERR(cur_page)) {
974 err = PTR_ERR(cur_page);
975 goto free_fail_no_cp;
977 sit_bitmap_ptr = page_address(cur_page);
978 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
979 f2fs_put_page(cur_page, 1);
982 f2fs_put_page(cp1, 1);
983 f2fs_put_page(cp2, 1);
987 f2fs_put_page(cp1, 1);
988 f2fs_put_page(cp2, 1);
994 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
996 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
997 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
999 if (is_inode_flag_set(inode, flag))
1002 set_inode_flag(inode, flag);
1003 if (!f2fs_is_volatile_file(inode))
1004 list_add_tail(&F2FS_I(inode)->dirty_list,
1005 &sbi->inode_list[type]);
1006 stat_inc_dirty_inode(sbi, type);
1009 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
1011 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
1013 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
1016 list_del_init(&F2FS_I(inode)->dirty_list);
1017 clear_inode_flag(inode, flag);
1018 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
1021 void f2fs_update_dirty_page(struct inode *inode, struct page *page)
1023 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1024 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1026 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1027 !S_ISLNK(inode->i_mode))
1030 spin_lock(&sbi->inode_lock[type]);
1031 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
1032 __add_dirty_inode(inode, type);
1033 inode_inc_dirty_pages(inode);
1034 spin_unlock(&sbi->inode_lock[type]);
1036 f2fs_set_page_private(page, 0);
1037 f2fs_trace_pid(page);
1040 void f2fs_remove_dirty_inode(struct inode *inode)
1042 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1043 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
1045 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
1046 !S_ISLNK(inode->i_mode))
1049 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
1052 spin_lock(&sbi->inode_lock[type]);
1053 __remove_dirty_inode(inode, type);
1054 spin_unlock(&sbi->inode_lock[type]);
1057 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type,
1060 struct list_head *head;
1061 struct inode *inode;
1062 struct f2fs_inode_info *fi;
1063 bool is_dir = (type == DIR_INODE);
1064 unsigned long ino = 0;
1066 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1067 get_pages(sbi, is_dir ?
1068 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1070 if (unlikely(f2fs_cp_error(sbi))) {
1071 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1072 get_pages(sbi, is_dir ?
1073 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1077 spin_lock(&sbi->inode_lock[type]);
1079 head = &sbi->inode_list[type];
1080 if (list_empty(head)) {
1081 spin_unlock(&sbi->inode_lock[type]);
1082 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1083 get_pages(sbi, is_dir ?
1084 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1087 fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1088 inode = igrab(&fi->vfs_inode);
1089 spin_unlock(&sbi->inode_lock[type]);
1091 unsigned long cur_ino = inode->i_ino;
1094 F2FS_I(inode)->cp_task = current;
1095 F2FS_I(inode)->wb_task = current;
1097 filemap_fdatawrite(inode->i_mapping);
1099 F2FS_I(inode)->wb_task = NULL;
1101 F2FS_I(inode)->cp_task = NULL;
1104 /* We need to give cpu to another writers. */
1111 * We should submit bio, since it exists several
1112 * wribacking dentry pages in the freeing inode.
1114 f2fs_submit_merged_write(sbi, DATA);
1120 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1122 struct list_head *head = &sbi->inode_list[DIRTY_META];
1123 struct inode *inode;
1124 struct f2fs_inode_info *fi;
1125 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1128 if (unlikely(f2fs_cp_error(sbi)))
1131 spin_lock(&sbi->inode_lock[DIRTY_META]);
1132 if (list_empty(head)) {
1133 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1136 fi = list_first_entry(head, struct f2fs_inode_info,
1138 inode = igrab(&fi->vfs_inode);
1139 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1141 sync_inode_metadata(inode, 0);
1143 /* it's on eviction */
1144 if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1145 f2fs_update_inode_page(inode);
1152 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1154 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1155 struct f2fs_nm_info *nm_i = NM_I(sbi);
1156 nid_t last_nid = nm_i->next_scan_nid;
1158 next_free_nid(sbi, &last_nid);
1159 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1160 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1161 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1162 ckpt->next_free_nid = cpu_to_le32(last_nid);
1165 static bool __need_flush_quota(struct f2fs_sb_info *sbi)
1169 if (!is_journalled_quota(sbi))
1172 if (!down_write_trylock(&sbi->quota_sem))
1174 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH)) {
1176 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR)) {
1178 } else if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_FLUSH)) {
1179 clear_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1181 } else if (get_pages(sbi, F2FS_DIRTY_QDATA)) {
1184 up_write(&sbi->quota_sem);
1189 * Freeze all the FS-operations for checkpoint.
1191 static int block_operations(struct f2fs_sb_info *sbi)
1193 struct writeback_control wbc = {
1194 .sync_mode = WB_SYNC_ALL,
1195 .nr_to_write = LONG_MAX,
1198 int err = 0, cnt = 0;
1201 * Let's flush inline_data in dirty node pages.
1203 f2fs_flush_inline_data(sbi);
1207 if (__need_flush_quota(sbi)) {
1210 if (++cnt > DEFAULT_RETRY_QUOTA_FLUSH_COUNT) {
1211 set_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1212 set_sbi_flag(sbi, SBI_QUOTA_NEED_FLUSH);
1213 goto retry_flush_dents;
1215 f2fs_unlock_all(sbi);
1217 /* only failed during mount/umount/freeze/quotactl */
1218 locked = down_read_trylock(&sbi->sb->s_umount);
1219 f2fs_quota_sync(sbi->sb, -1);
1221 up_read(&sbi->sb->s_umount);
1223 goto retry_flush_quotas;
1227 /* write all the dirty dentry pages */
1228 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1229 f2fs_unlock_all(sbi);
1230 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE, true);
1234 goto retry_flush_quotas;
1238 * POR: we should ensure that there are no dirty node pages
1239 * until finishing nat/sit flush. inode->i_blocks can be updated.
1241 down_write(&sbi->node_change);
1243 if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1244 up_write(&sbi->node_change);
1245 f2fs_unlock_all(sbi);
1246 err = f2fs_sync_inode_meta(sbi);
1250 goto retry_flush_quotas;
1254 down_write(&sbi->node_write);
1256 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1257 up_write(&sbi->node_write);
1258 atomic_inc(&sbi->wb_sync_req[NODE]);
1259 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1260 atomic_dec(&sbi->wb_sync_req[NODE]);
1262 up_write(&sbi->node_change);
1263 f2fs_unlock_all(sbi);
1267 goto retry_flush_nodes;
1271 * sbi->node_change is used only for AIO write_begin path which produces
1272 * dirty node blocks and some checkpoint values by block allocation.
1274 __prepare_cp_block(sbi);
1275 up_write(&sbi->node_change);
1279 static void unblock_operations(struct f2fs_sb_info *sbi)
1281 up_write(&sbi->node_write);
1282 f2fs_unlock_all(sbi);
1285 void f2fs_wait_on_all_pages(struct f2fs_sb_info *sbi, int type)
1290 if (!get_pages(sbi, type))
1293 if (unlikely(f2fs_cp_error(sbi) &&
1294 !is_sbi_flag_set(sbi, SBI_IS_CLOSE)))
1297 if (type == F2FS_DIRTY_META)
1298 f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1300 else if (type == F2FS_WB_CP_DATA)
1301 f2fs_submit_merged_write(sbi, DATA);
1303 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1304 io_schedule_timeout(DEFAULT_IO_TIMEOUT);
1306 finish_wait(&sbi->cp_wait, &wait);
1309 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1311 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1312 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1313 unsigned long flags;
1315 spin_lock_irqsave(&sbi->cp_lock, flags);
1317 if ((cpc->reason & CP_UMOUNT) &&
1318 le32_to_cpu(ckpt->cp_pack_total_block_count) >
1319 sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1320 disable_nat_bits(sbi, false);
1322 if (cpc->reason & CP_TRIMMED)
1323 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1325 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1327 if (cpc->reason & CP_UMOUNT)
1328 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1330 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1332 if (cpc->reason & CP_FASTBOOT)
1333 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1335 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1338 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1340 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1342 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1343 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1345 if (is_sbi_flag_set(sbi, SBI_IS_RESIZEFS))
1346 __set_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1348 __clear_ckpt_flags(ckpt, CP_RESIZEFS_FLAG);
1350 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED))
1351 __set_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1353 __clear_ckpt_flags(ckpt, CP_DISABLED_FLAG);
1355 if (is_sbi_flag_set(sbi, SBI_CP_DISABLED_QUICK))
1356 __set_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1358 __clear_ckpt_flags(ckpt, CP_DISABLED_QUICK_FLAG);
1360 if (is_sbi_flag_set(sbi, SBI_QUOTA_SKIP_FLUSH))
1361 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1363 __clear_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1365 if (is_sbi_flag_set(sbi, SBI_QUOTA_NEED_REPAIR))
1366 __set_ckpt_flags(ckpt, CP_QUOTA_NEED_FSCK_FLAG);
1368 /* set this flag to activate crc|cp_ver for recovery */
1369 __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1370 __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1372 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1375 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1376 void *src, block_t blk_addr)
1378 struct writeback_control wbc = {
1383 * pagevec_lookup_tag and lock_page again will take
1384 * some extra time. Therefore, f2fs_update_meta_pages and
1385 * f2fs_sync_meta_pages are combined in this function.
1387 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1390 f2fs_wait_on_page_writeback(page, META, true, true);
1392 memcpy(page_address(page), src, PAGE_SIZE);
1394 set_page_dirty(page);
1395 if (unlikely(!clear_page_dirty_for_io(page)))
1396 f2fs_bug_on(sbi, 1);
1398 /* writeout cp pack 2 page */
1399 err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1400 if (unlikely(err && f2fs_cp_error(sbi))) {
1401 f2fs_put_page(page, 1);
1405 f2fs_bug_on(sbi, err);
1406 f2fs_put_page(page, 0);
1408 /* submit checkpoint (with barrier if NOBARRIER is not set) */
1409 f2fs_submit_merged_write(sbi, META_FLUSH);
1412 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1414 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1415 struct f2fs_nm_info *nm_i = NM_I(sbi);
1416 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1418 unsigned int data_sum_blocks, orphan_blocks;
1421 int cp_payload_blks = __cp_payload(sbi);
1422 struct super_block *sb = sbi->sb;
1423 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1427 /* Flush all the NAT/SIT pages */
1428 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1430 /* start to update checkpoint, cp ver is already updated previously */
1431 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1432 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1433 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1434 ckpt->cur_node_segno[i] =
1435 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1436 ckpt->cur_node_blkoff[i] =
1437 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1438 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1439 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1441 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1442 ckpt->cur_data_segno[i] =
1443 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1444 ckpt->cur_data_blkoff[i] =
1445 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1446 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1447 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1450 /* 2 cp + n data seg summary + orphan inode blocks */
1451 data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1452 spin_lock_irqsave(&sbi->cp_lock, flags);
1453 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1454 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1456 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1457 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1459 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1460 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1463 if (__remain_node_summaries(cpc->reason))
1464 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1465 cp_payload_blks + data_sum_blocks +
1466 orphan_blocks + NR_CURSEG_NODE_TYPE);
1468 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1469 cp_payload_blks + data_sum_blocks +
1472 /* update ckpt flag for checkpoint */
1473 update_ckpt_flags(sbi, cpc);
1475 /* update SIT/NAT bitmap */
1476 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1477 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1479 crc32 = f2fs_checkpoint_chksum(sbi, ckpt);
1480 *((__le32 *)((unsigned char *)ckpt +
1481 le32_to_cpu(ckpt->checksum_offset)))
1482 = cpu_to_le32(crc32);
1484 start_blk = __start_cp_next_addr(sbi);
1486 /* write nat bits */
1487 if (enabled_nat_bits(sbi, cpc)) {
1488 __u64 cp_ver = cur_cp_version(ckpt);
1491 cp_ver |= ((__u64)crc32 << 32);
1492 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1494 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1495 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1496 f2fs_update_meta_page(sbi, nm_i->nat_bits +
1497 (i << F2FS_BLKSIZE_BITS), blk + i);
1500 /* write out checkpoint buffer at block 0 */
1501 f2fs_update_meta_page(sbi, ckpt, start_blk++);
1503 for (i = 1; i < 1 + cp_payload_blks; i++)
1504 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1508 write_orphan_inodes(sbi, start_blk);
1509 start_blk += orphan_blocks;
1512 f2fs_write_data_summaries(sbi, start_blk);
1513 start_blk += data_sum_blocks;
1515 /* Record write statistics in the hot node summary */
1516 kbytes_written = sbi->kbytes_written;
1517 if (sb->s_bdev->bd_part)
1518 kbytes_written += BD_PART_WRITTEN(sbi);
1520 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1522 if (__remain_node_summaries(cpc->reason)) {
1523 f2fs_write_node_summaries(sbi, start_blk);
1524 start_blk += NR_CURSEG_NODE_TYPE;
1527 /* update user_block_counts */
1528 sbi->last_valid_block_count = sbi->total_valid_block_count;
1529 percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1531 /* Here, we have one bio having CP pack except cp pack 2 page */
1532 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1533 /* Wait for all dirty meta pages to be submitted for IO */
1534 f2fs_wait_on_all_pages(sbi, F2FS_DIRTY_META);
1536 /* wait for previous submitted meta pages writeback */
1537 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1539 /* flush all device cache */
1540 err = f2fs_flush_device_cache(sbi);
1544 /* barrier and flush checkpoint cp pack 2 page if it can */
1545 commit_checkpoint(sbi, ckpt, start_blk);
1546 f2fs_wait_on_all_pages(sbi, F2FS_WB_CP_DATA);
1549 * invalidate intermediate page cache borrowed from meta inode which are
1550 * used for migration of encrypted, verity or compressed inode's blocks.
1552 if (f2fs_sb_has_encrypt(sbi) || f2fs_sb_has_verity(sbi) ||
1553 f2fs_sb_has_compression(sbi))
1554 invalidate_mapping_pages(META_MAPPING(sbi),
1555 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1557 f2fs_release_ino_entry(sbi, false);
1559 f2fs_reset_fsync_node_info(sbi);
1561 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1562 clear_sbi_flag(sbi, SBI_NEED_CP);
1563 clear_sbi_flag(sbi, SBI_QUOTA_SKIP_FLUSH);
1565 spin_lock(&sbi->stat_lock);
1566 sbi->unusable_block_count = 0;
1567 spin_unlock(&sbi->stat_lock);
1569 __set_cp_next_pack(sbi);
1572 * redirty superblock if metadata like node page or inode cache is
1573 * updated during writing checkpoint.
1575 if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1576 get_pages(sbi, F2FS_DIRTY_IMETA))
1577 set_sbi_flag(sbi, SBI_IS_DIRTY);
1579 f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1581 return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1584 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1586 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1587 unsigned long long ckpt_ver;
1590 if (f2fs_readonly(sbi->sb) || f2fs_hw_is_readonly(sbi))
1593 if (unlikely(is_sbi_flag_set(sbi, SBI_CP_DISABLED))) {
1594 if (cpc->reason != CP_PAUSE)
1596 f2fs_warn(sbi, "Start checkpoint disabled!");
1598 if (cpc->reason != CP_RESIZE)
1599 mutex_lock(&sbi->cp_mutex);
1601 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1602 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1603 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1605 if (unlikely(f2fs_cp_error(sbi))) {
1610 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1612 err = block_operations(sbi);
1616 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1618 f2fs_flush_merged_writes(sbi);
1620 /* this is the case of multiple fstrims without any changes */
1621 if (cpc->reason & CP_DISCARD) {
1622 if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1623 unblock_operations(sbi);
1627 if (NM_I(sbi)->nat_cnt[DIRTY_NAT] == 0 &&
1628 SIT_I(sbi)->dirty_sentries == 0 &&
1629 prefree_segments(sbi) == 0) {
1630 f2fs_flush_sit_entries(sbi, cpc);
1631 f2fs_clear_prefree_segments(sbi, cpc);
1632 unblock_operations(sbi);
1638 * update checkpoint pack index
1639 * Increase the version number so that
1640 * SIT entries and seg summaries are written at correct place
1642 ckpt_ver = cur_cp_version(ckpt);
1643 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1645 /* write cached NAT/SIT entries to NAT/SIT area */
1646 err = f2fs_flush_nat_entries(sbi, cpc);
1650 f2fs_flush_sit_entries(sbi, cpc);
1652 /* save inmem log status */
1653 f2fs_save_inmem_curseg(sbi);
1655 err = do_checkpoint(sbi, cpc);
1657 f2fs_release_discard_addrs(sbi);
1659 f2fs_clear_prefree_segments(sbi, cpc);
1661 f2fs_restore_inmem_curseg(sbi);
1663 unblock_operations(sbi);
1664 stat_inc_cp_count(sbi->stat_info);
1666 if (cpc->reason & CP_RECOVERY)
1667 f2fs_notice(sbi, "checkpoint: version = %llx", ckpt_ver);
1669 /* update CP_TIME to trigger checkpoint periodically */
1670 f2fs_update_time(sbi, CP_TIME);
1671 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1673 if (cpc->reason != CP_RESIZE)
1674 mutex_unlock(&sbi->cp_mutex);
1678 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1682 for (i = 0; i < MAX_INO_ENTRY; i++) {
1683 struct inode_management *im = &sbi->im[i];
1685 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1686 spin_lock_init(&im->ino_lock);
1687 INIT_LIST_HEAD(&im->ino_list);
1691 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1692 NR_CURSEG_PERSIST_TYPE - __cp_payload(sbi)) *
1693 F2FS_ORPHANS_PER_BLOCK;
1696 int __init f2fs_create_checkpoint_caches(void)
1698 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1699 sizeof(struct ino_entry));
1700 if (!ino_entry_slab)
1702 f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1703 sizeof(struct inode_entry));
1704 if (!f2fs_inode_entry_slab) {
1705 kmem_cache_destroy(ino_entry_slab);
1711 void f2fs_destroy_checkpoint_caches(void)
1713 kmem_cache_destroy(ino_entry_slab);
1714 kmem_cache_destroy(f2fs_inode_entry_slab);