4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
12 #include <linux/bio.h>
13 #include <linux/mpage.h>
14 #include <linux/writeback.h>
15 #include <linux/blkdev.h>
16 #include <linux/f2fs_fs.h>
17 #include <linux/pagevec.h>
18 #include <linux/swap.h>
24 #include <trace/events/f2fs.h>
26 static struct kmem_cache *ino_entry_slab;
27 struct kmem_cache *f2fs_inode_entry_slab;
29 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
31 f2fs_build_fault_attr(sbi, 0, 0);
32 set_ckpt_flags(sbi, CP_ERROR_FLAG);
34 f2fs_flush_merged_writes(sbi);
38 * We guarantee no failure on the returned page.
40 struct page *f2fs_grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
42 struct address_space *mapping = META_MAPPING(sbi);
43 struct page *page = NULL;
45 page = f2fs_grab_cache_page(mapping, index, false);
50 f2fs_wait_on_page_writeback(page, META, true);
51 if (!PageUptodate(page))
52 SetPageUptodate(page);
57 * We guarantee no failure on the returned page.
59 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
62 struct address_space *mapping = META_MAPPING(sbi);
64 struct f2fs_io_info fio = {
68 .op_flags = REQ_META | REQ_PRIO,
71 .encrypted_page = NULL,
76 if (unlikely(!is_meta))
77 fio.op_flags &= ~REQ_META;
79 page = f2fs_grab_cache_page(mapping, index, false);
84 if (PageUptodate(page))
89 err = f2fs_submit_page_bio(&fio);
91 f2fs_put_page(page, 1);
96 if (unlikely(page->mapping != mapping)) {
97 f2fs_put_page(page, 1);
101 if (unlikely(!PageUptodate(page))) {
102 f2fs_put_page(page, 1);
103 return ERR_PTR(-EIO);
109 struct page *f2fs_get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
111 return __get_meta_page(sbi, index, true);
114 struct page *f2fs_get_meta_page_nofail(struct f2fs_sb_info *sbi, pgoff_t index)
120 page = __get_meta_page(sbi, index, true);
122 if (PTR_ERR(page) == -EIO &&
123 ++count <= DEFAULT_RETRY_IO_COUNT)
126 f2fs_stop_checkpoint(sbi, false);
134 struct page *f2fs_get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
136 return __get_meta_page(sbi, index, false);
139 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
140 block_t blkaddr, int type)
146 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
150 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
151 blkaddr < SM_I(sbi)->ssa_blkaddr))
155 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
156 blkaddr < __start_cp_addr(sbi)))
161 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
162 blkaddr < MAIN_BLKADDR(sbi))) {
163 if (type == DATA_GENERIC) {
164 f2fs_msg(sbi->sb, KERN_WARNING,
165 "access invalid blkaddr:%u", blkaddr);
172 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
173 blkaddr >= MAIN_BLKADDR(sbi)))
184 * Readahead CP/NAT/SIT/SSA pages
186 int f2fs_ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
190 block_t blkno = start;
191 struct f2fs_io_info fio = {
195 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
196 .encrypted_page = NULL,
198 .is_meta = (type != META_POR),
200 struct blk_plug plug;
202 if (unlikely(type == META_POR))
203 fio.op_flags &= ~REQ_META;
205 blk_start_plug(&plug);
206 for (; nrpages-- > 0; blkno++) {
208 if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
213 if (unlikely(blkno >=
214 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
216 /* get nat block addr */
217 fio.new_blkaddr = current_nat_addr(sbi,
218 blkno * NAT_ENTRY_PER_BLOCK);
221 if (unlikely(blkno >= TOTAL_SEGS(sbi)))
223 /* get sit block addr */
224 fio.new_blkaddr = current_sit_addr(sbi,
225 blkno * SIT_ENTRY_PER_BLOCK);
230 fio.new_blkaddr = blkno;
236 page = f2fs_grab_cache_page(META_MAPPING(sbi),
237 fio.new_blkaddr, false);
240 if (PageUptodate(page)) {
241 f2fs_put_page(page, 1);
246 f2fs_submit_page_bio(&fio);
247 f2fs_put_page(page, 0);
250 blk_finish_plug(&plug);
251 return blkno - start;
254 void f2fs_ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
257 bool readahead = false;
259 page = find_get_page(META_MAPPING(sbi), index);
260 if (!page || !PageUptodate(page))
262 f2fs_put_page(page, 0);
265 f2fs_ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
268 static int __f2fs_write_meta_page(struct page *page,
269 struct writeback_control *wbc,
270 enum iostat_type io_type)
272 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
274 trace_f2fs_writepage(page, META);
276 if (unlikely(f2fs_cp_error(sbi)))
278 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
280 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
283 f2fs_do_write_meta_page(sbi, page, io_type);
284 dec_page_count(sbi, F2FS_DIRTY_META);
286 if (wbc->for_reclaim)
287 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
288 0, page->index, META);
292 if (unlikely(f2fs_cp_error(sbi)))
293 f2fs_submit_merged_write(sbi, META);
298 redirty_page_for_writepage(wbc, page);
299 return AOP_WRITEPAGE_ACTIVATE;
302 static int f2fs_write_meta_page(struct page *page,
303 struct writeback_control *wbc)
305 return __f2fs_write_meta_page(page, wbc, FS_META_IO);
308 static int f2fs_write_meta_pages(struct address_space *mapping,
309 struct writeback_control *wbc)
311 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
314 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
317 /* collect a number of dirty meta pages and write together */
318 if (wbc->for_kupdate ||
319 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
322 /* if locked failed, cp will flush dirty pages instead */
323 if (!mutex_trylock(&sbi->cp_mutex))
326 trace_f2fs_writepages(mapping->host, wbc, META);
327 diff = nr_pages_to_write(sbi, META, wbc);
328 written = f2fs_sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
329 mutex_unlock(&sbi->cp_mutex);
330 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
334 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
335 trace_f2fs_writepages(mapping->host, wbc, META);
339 long f2fs_sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
340 long nr_to_write, enum iostat_type io_type)
342 struct address_space *mapping = META_MAPPING(sbi);
343 pgoff_t index = 0, prev = ULONG_MAX;
347 struct writeback_control wbc = {
350 struct blk_plug plug;
354 blk_start_plug(&plug);
356 while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
357 PAGECACHE_TAG_DIRTY))) {
360 for (i = 0; i < nr_pages; i++) {
361 struct page *page = pvec.pages[i];
363 if (prev == ULONG_MAX)
364 prev = page->index - 1;
365 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
366 pagevec_release(&pvec);
372 if (unlikely(page->mapping != mapping)) {
377 if (!PageDirty(page)) {
378 /* someone wrote it for us */
379 goto continue_unlock;
382 f2fs_wait_on_page_writeback(page, META, true);
384 BUG_ON(PageWriteback(page));
385 if (!clear_page_dirty_for_io(page))
386 goto continue_unlock;
388 if (__f2fs_write_meta_page(page, &wbc, io_type)) {
394 if (unlikely(nwritten >= nr_to_write))
397 pagevec_release(&pvec);
402 f2fs_submit_merged_write(sbi, type);
404 blk_finish_plug(&plug);
409 static int f2fs_set_meta_page_dirty(struct page *page)
411 trace_f2fs_set_page_dirty(page, META);
413 if (!PageUptodate(page))
414 SetPageUptodate(page);
415 if (!PageDirty(page)) {
416 __set_page_dirty_nobuffers(page);
417 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
418 SetPagePrivate(page);
419 f2fs_trace_pid(page);
425 const struct address_space_operations f2fs_meta_aops = {
426 .writepage = f2fs_write_meta_page,
427 .writepages = f2fs_write_meta_pages,
428 .set_page_dirty = f2fs_set_meta_page_dirty,
429 .invalidatepage = f2fs_invalidate_page,
430 .releasepage = f2fs_release_page,
431 #ifdef CONFIG_MIGRATION
432 .migratepage = f2fs_migrate_page,
436 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
437 unsigned int devidx, int type)
439 struct inode_management *im = &sbi->im[type];
440 struct ino_entry *e, *tmp;
442 tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
444 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
446 spin_lock(&im->ino_lock);
447 e = radix_tree_lookup(&im->ino_root, ino);
450 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
453 memset(e, 0, sizeof(struct ino_entry));
456 list_add_tail(&e->list, &im->ino_list);
457 if (type != ORPHAN_INO)
461 if (type == FLUSH_INO)
462 f2fs_set_bit(devidx, (char *)&e->dirty_device);
464 spin_unlock(&im->ino_lock);
465 radix_tree_preload_end();
468 kmem_cache_free(ino_entry_slab, tmp);
471 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
473 struct inode_management *im = &sbi->im[type];
476 spin_lock(&im->ino_lock);
477 e = radix_tree_lookup(&im->ino_root, ino);
480 radix_tree_delete(&im->ino_root, ino);
482 spin_unlock(&im->ino_lock);
483 kmem_cache_free(ino_entry_slab, e);
486 spin_unlock(&im->ino_lock);
489 void f2fs_add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
491 /* add new dirty ino entry into list */
492 __add_ino_entry(sbi, ino, 0, type);
495 void f2fs_remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
497 /* remove dirty ino entry from list */
498 __remove_ino_entry(sbi, ino, type);
501 /* mode should be APPEND_INO or UPDATE_INO */
502 bool f2fs_exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
504 struct inode_management *im = &sbi->im[mode];
507 spin_lock(&im->ino_lock);
508 e = radix_tree_lookup(&im->ino_root, ino);
509 spin_unlock(&im->ino_lock);
510 return e ? true : false;
513 void f2fs_release_ino_entry(struct f2fs_sb_info *sbi, bool all)
515 struct ino_entry *e, *tmp;
518 for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
519 struct inode_management *im = &sbi->im[i];
521 spin_lock(&im->ino_lock);
522 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
524 radix_tree_delete(&im->ino_root, e->ino);
525 kmem_cache_free(ino_entry_slab, e);
528 spin_unlock(&im->ino_lock);
532 void f2fs_set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
533 unsigned int devidx, int type)
535 __add_ino_entry(sbi, ino, devidx, type);
538 bool f2fs_is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
539 unsigned int devidx, int type)
541 struct inode_management *im = &sbi->im[type];
543 bool is_dirty = false;
545 spin_lock(&im->ino_lock);
546 e = radix_tree_lookup(&im->ino_root, ino);
547 if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
549 spin_unlock(&im->ino_lock);
553 int f2fs_acquire_orphan_inode(struct f2fs_sb_info *sbi)
555 struct inode_management *im = &sbi->im[ORPHAN_INO];
558 spin_lock(&im->ino_lock);
560 if (time_to_inject(sbi, FAULT_ORPHAN)) {
561 spin_unlock(&im->ino_lock);
562 f2fs_show_injection_info(FAULT_ORPHAN);
566 if (unlikely(im->ino_num >= sbi->max_orphans))
570 spin_unlock(&im->ino_lock);
575 void f2fs_release_orphan_inode(struct f2fs_sb_info *sbi)
577 struct inode_management *im = &sbi->im[ORPHAN_INO];
579 spin_lock(&im->ino_lock);
580 f2fs_bug_on(sbi, im->ino_num == 0);
582 spin_unlock(&im->ino_lock);
585 void f2fs_add_orphan_inode(struct inode *inode)
587 /* add new orphan ino entry into list */
588 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
589 f2fs_update_inode_page(inode);
592 void f2fs_remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
594 /* remove orphan entry from orphan list */
595 __remove_ino_entry(sbi, ino, ORPHAN_INO);
598 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
604 inode = f2fs_iget_retry(sbi->sb, ino);
607 * there should be a bug that we can't find the entry
610 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
611 return PTR_ERR(inode);
614 err = dquot_initialize(inode);
622 /* truncate all the data during iput */
625 err = f2fs_get_node_info(sbi, ino, &ni);
629 /* ENOMEM was fully retried in f2fs_evict_inode. */
630 if (ni.blk_addr != NULL_ADDR) {
637 set_sbi_flag(sbi, SBI_NEED_FSCK);
638 f2fs_msg(sbi->sb, KERN_WARNING,
639 "%s: orphan failed (ino=%x), run fsck to fix.",
644 int f2fs_recover_orphan_inodes(struct f2fs_sb_info *sbi)
646 block_t start_blk, orphan_blocks, i, j;
647 unsigned int s_flags = sbi->sb->s_flags;
653 if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
656 if (s_flags & SB_RDONLY) {
657 f2fs_msg(sbi->sb, KERN_INFO, "orphan cleanup on readonly fs");
658 sbi->sb->s_flags &= ~SB_RDONLY;
662 /* Needed for iput() to work correctly and not trash data */
663 sbi->sb->s_flags |= SB_ACTIVE;
666 * Turn on quotas which were not enabled for read-only mounts if
667 * filesystem has quota feature, so that they are updated correctly.
669 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
672 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
673 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
675 f2fs_ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
677 for (i = 0; i < orphan_blocks; i++) {
679 struct f2fs_orphan_block *orphan_blk;
681 page = f2fs_get_meta_page(sbi, start_blk + i);
687 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
688 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
689 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
690 err = recover_orphan_inode(sbi, ino);
692 f2fs_put_page(page, 1);
696 f2fs_put_page(page, 1);
698 /* clear Orphan Flag */
699 clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
701 set_sbi_flag(sbi, SBI_IS_RECOVERED);
704 /* Turn quotas off */
706 f2fs_quota_off_umount(sbi->sb);
708 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
713 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
715 struct list_head *head;
716 struct f2fs_orphan_block *orphan_blk = NULL;
717 unsigned int nentries = 0;
718 unsigned short index = 1;
719 unsigned short orphan_blocks;
720 struct page *page = NULL;
721 struct ino_entry *orphan = NULL;
722 struct inode_management *im = &sbi->im[ORPHAN_INO];
724 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
727 * we don't need to do spin_lock(&im->ino_lock) here, since all the
728 * orphan inode operations are covered under f2fs_lock_op().
729 * And, spin_lock should be avoided due to page operations below.
731 head = &im->ino_list;
733 /* loop for each orphan inode entry and write them in Jornal block */
734 list_for_each_entry(orphan, head, list) {
736 page = f2fs_grab_meta_page(sbi, start_blk++);
738 (struct f2fs_orphan_block *)page_address(page);
739 memset(orphan_blk, 0, sizeof(*orphan_blk));
742 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
744 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
746 * an orphan block is full of 1020 entries,
747 * then we need to flush current orphan blocks
748 * and bring another one in memory
750 orphan_blk->blk_addr = cpu_to_le16(index);
751 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
752 orphan_blk->entry_count = cpu_to_le32(nentries);
753 set_page_dirty(page);
754 f2fs_put_page(page, 1);
762 orphan_blk->blk_addr = cpu_to_le16(index);
763 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
764 orphan_blk->entry_count = cpu_to_le32(nentries);
765 set_page_dirty(page);
766 f2fs_put_page(page, 1);
770 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
771 struct f2fs_checkpoint **cp_block, struct page **cp_page,
772 unsigned long long *version)
774 unsigned long blk_size = sbi->blocksize;
775 size_t crc_offset = 0;
778 *cp_page = f2fs_get_meta_page(sbi, cp_addr);
779 if (IS_ERR(*cp_page))
780 return PTR_ERR(*cp_page);
782 *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
784 crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
785 if (crc_offset > (blk_size - sizeof(__le32))) {
786 f2fs_put_page(*cp_page, 1);
787 f2fs_msg(sbi->sb, KERN_WARNING,
788 "invalid crc_offset: %zu", crc_offset);
792 crc = cur_cp_crc(*cp_block);
793 if (!f2fs_crc_valid(sbi, crc, *cp_block, crc_offset)) {
794 f2fs_put_page(*cp_page, 1);
795 f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
799 *version = cur_cp_version(*cp_block);
803 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
804 block_t cp_addr, unsigned long long *version)
806 struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
807 struct f2fs_checkpoint *cp_block = NULL;
808 unsigned long long cur_version = 0, pre_version = 0;
811 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
812 &cp_page_1, version);
816 if (le32_to_cpu(cp_block->cp_pack_total_block_count) >
817 sbi->blocks_per_seg) {
818 f2fs_msg(sbi->sb, KERN_WARNING,
819 "invalid cp_pack_total_block_count:%u",
820 le32_to_cpu(cp_block->cp_pack_total_block_count));
823 pre_version = *version;
825 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
826 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
827 &cp_page_2, version);
830 cur_version = *version;
832 if (cur_version == pre_version) {
833 *version = cur_version;
834 f2fs_put_page(cp_page_2, 1);
837 f2fs_put_page(cp_page_2, 1);
839 f2fs_put_page(cp_page_1, 1);
843 int f2fs_get_valid_checkpoint(struct f2fs_sb_info *sbi)
845 struct f2fs_checkpoint *cp_block;
846 struct f2fs_super_block *fsb = sbi->raw_super;
847 struct page *cp1, *cp2, *cur_page;
848 unsigned long blk_size = sbi->blocksize;
849 unsigned long long cp1_version = 0, cp2_version = 0;
850 unsigned long long cp_start_blk_no;
851 unsigned int cp_blks = 1 + __cp_payload(sbi);
856 sbi->ckpt = f2fs_kzalloc(sbi, array_size(blk_size, cp_blks),
861 * Finding out valid cp block involves read both
862 * sets( cp pack1 and cp pack 2)
864 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
865 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
867 /* The second checkpoint pack should start at the next segment */
868 cp_start_blk_no += ((unsigned long long)1) <<
869 le32_to_cpu(fsb->log_blocks_per_seg);
870 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
873 if (ver_after(cp2_version, cp1_version))
886 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
887 memcpy(sbi->ckpt, cp_block, blk_size);
890 sbi->cur_cp_pack = 1;
892 sbi->cur_cp_pack = 2;
894 /* Sanity checking of checkpoint */
895 if (f2fs_sanity_check_ckpt(sbi)) {
897 goto free_fail_no_cp;
903 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
905 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
907 for (i = 1; i < cp_blks; i++) {
908 void *sit_bitmap_ptr;
909 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
911 cur_page = f2fs_get_meta_page(sbi, cp_blk_no + i);
912 if (IS_ERR(cur_page)) {
913 err = PTR_ERR(cur_page);
914 goto free_fail_no_cp;
916 sit_bitmap_ptr = page_address(cur_page);
917 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
918 f2fs_put_page(cur_page, 1);
921 f2fs_put_page(cp1, 1);
922 f2fs_put_page(cp2, 1);
926 f2fs_put_page(cp1, 1);
927 f2fs_put_page(cp2, 1);
933 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
935 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
936 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
938 if (is_inode_flag_set(inode, flag))
941 set_inode_flag(inode, flag);
942 if (!f2fs_is_volatile_file(inode))
943 list_add_tail(&F2FS_I(inode)->dirty_list,
944 &sbi->inode_list[type]);
945 stat_inc_dirty_inode(sbi, type);
948 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
950 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
952 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
955 list_del_init(&F2FS_I(inode)->dirty_list);
956 clear_inode_flag(inode, flag);
957 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
960 void f2fs_update_dirty_page(struct inode *inode, struct page *page)
962 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
963 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
965 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
966 !S_ISLNK(inode->i_mode))
969 spin_lock(&sbi->inode_lock[type]);
970 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
971 __add_dirty_inode(inode, type);
972 inode_inc_dirty_pages(inode);
973 spin_unlock(&sbi->inode_lock[type]);
975 SetPagePrivate(page);
976 f2fs_trace_pid(page);
979 void f2fs_remove_dirty_inode(struct inode *inode)
981 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
982 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
984 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
985 !S_ISLNK(inode->i_mode))
988 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
991 spin_lock(&sbi->inode_lock[type]);
992 __remove_dirty_inode(inode, type);
993 spin_unlock(&sbi->inode_lock[type]);
996 int f2fs_sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
998 struct list_head *head;
1000 struct f2fs_inode_info *fi;
1001 bool is_dir = (type == DIR_INODE);
1002 unsigned long ino = 0;
1004 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
1005 get_pages(sbi, is_dir ?
1006 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1008 if (unlikely(f2fs_cp_error(sbi))) {
1009 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1010 get_pages(sbi, is_dir ?
1011 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1015 spin_lock(&sbi->inode_lock[type]);
1017 head = &sbi->inode_list[type];
1018 if (list_empty(head)) {
1019 spin_unlock(&sbi->inode_lock[type]);
1020 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
1021 get_pages(sbi, is_dir ?
1022 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
1025 fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
1026 inode = igrab(&fi->vfs_inode);
1027 spin_unlock(&sbi->inode_lock[type]);
1029 unsigned long cur_ino = inode->i_ino;
1032 F2FS_I(inode)->cp_task = current;
1034 filemap_fdatawrite(inode->i_mapping);
1037 F2FS_I(inode)->cp_task = NULL;
1040 /* We need to give cpu to another writers. */
1047 * We should submit bio, since it exists several
1048 * wribacking dentry pages in the freeing inode.
1050 f2fs_submit_merged_write(sbi, DATA);
1056 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
1058 struct list_head *head = &sbi->inode_list[DIRTY_META];
1059 struct inode *inode;
1060 struct f2fs_inode_info *fi;
1061 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1064 if (unlikely(f2fs_cp_error(sbi)))
1067 spin_lock(&sbi->inode_lock[DIRTY_META]);
1068 if (list_empty(head)) {
1069 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1072 fi = list_first_entry(head, struct f2fs_inode_info,
1074 inode = igrab(&fi->vfs_inode);
1075 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1077 sync_inode_metadata(inode, 0);
1079 /* it's on eviction */
1080 if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1081 f2fs_update_inode_page(inode);
1088 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1090 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1091 struct f2fs_nm_info *nm_i = NM_I(sbi);
1092 nid_t last_nid = nm_i->next_scan_nid;
1094 next_free_nid(sbi, &last_nid);
1095 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1096 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1097 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1098 ckpt->next_free_nid = cpu_to_le32(last_nid);
1102 * Freeze all the FS-operations for checkpoint.
1104 static int block_operations(struct f2fs_sb_info *sbi)
1106 struct writeback_control wbc = {
1107 .sync_mode = WB_SYNC_ALL,
1108 .nr_to_write = LONG_MAX,
1111 struct blk_plug plug;
1114 blk_start_plug(&plug);
1118 /* write all the dirty dentry pages */
1119 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1120 f2fs_unlock_all(sbi);
1121 err = f2fs_sync_dirty_inodes(sbi, DIR_INODE);
1125 goto retry_flush_dents;
1129 * POR: we should ensure that there are no dirty node pages
1130 * until finishing nat/sit flush. inode->i_blocks can be updated.
1132 down_write(&sbi->node_change);
1134 if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1135 up_write(&sbi->node_change);
1136 f2fs_unlock_all(sbi);
1137 err = f2fs_sync_inode_meta(sbi);
1141 goto retry_flush_dents;
1145 down_write(&sbi->node_write);
1147 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1148 up_write(&sbi->node_write);
1149 atomic_inc(&sbi->wb_sync_req[NODE]);
1150 err = f2fs_sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1151 atomic_dec(&sbi->wb_sync_req[NODE]);
1153 up_write(&sbi->node_change);
1154 f2fs_unlock_all(sbi);
1158 goto retry_flush_nodes;
1162 * sbi->node_change is used only for AIO write_begin path which produces
1163 * dirty node blocks and some checkpoint values by block allocation.
1165 __prepare_cp_block(sbi);
1166 up_write(&sbi->node_change);
1168 blk_finish_plug(&plug);
1172 static void unblock_operations(struct f2fs_sb_info *sbi)
1174 up_write(&sbi->node_write);
1175 f2fs_unlock_all(sbi);
1178 void f2fs_wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
1183 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1185 if (!get_pages(sbi, F2FS_WB_CP_DATA))
1188 if (unlikely(f2fs_cp_error(sbi)))
1191 io_schedule_timeout(5*HZ);
1193 finish_wait(&sbi->cp_wait, &wait);
1196 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1198 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1199 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1200 unsigned long flags;
1202 spin_lock_irqsave(&sbi->cp_lock, flags);
1204 if ((cpc->reason & CP_UMOUNT) &&
1205 le32_to_cpu(ckpt->cp_pack_total_block_count) >
1206 sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1207 disable_nat_bits(sbi, false);
1209 if (cpc->reason & CP_TRIMMED)
1210 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1212 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1214 if (cpc->reason & CP_UMOUNT)
1215 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1217 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1219 if (cpc->reason & CP_FASTBOOT)
1220 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1222 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1225 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1227 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1229 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1230 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1232 /* set this flag to activate crc|cp_ver for recovery */
1233 __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1234 __clear_ckpt_flags(ckpt, CP_NOCRC_RECOVERY_FLAG);
1236 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1239 static void commit_checkpoint(struct f2fs_sb_info *sbi,
1240 void *src, block_t blk_addr)
1242 struct writeback_control wbc = {
1247 * pagevec_lookup_tag and lock_page again will take
1248 * some extra time. Therefore, f2fs_update_meta_pages and
1249 * f2fs_sync_meta_pages are combined in this function.
1251 struct page *page = f2fs_grab_meta_page(sbi, blk_addr);
1254 memcpy(page_address(page), src, PAGE_SIZE);
1255 set_page_dirty(page);
1257 f2fs_wait_on_page_writeback(page, META, true);
1258 f2fs_bug_on(sbi, PageWriteback(page));
1259 if (unlikely(!clear_page_dirty_for_io(page)))
1260 f2fs_bug_on(sbi, 1);
1262 /* writeout cp pack 2 page */
1263 err = __f2fs_write_meta_page(page, &wbc, FS_CP_META_IO);
1264 if (unlikely(err && f2fs_cp_error(sbi))) {
1265 f2fs_put_page(page, 1);
1269 f2fs_bug_on(sbi, err);
1270 f2fs_put_page(page, 0);
1272 /* submit checkpoint (with barrier if NOBARRIER is not set) */
1273 f2fs_submit_merged_write(sbi, META_FLUSH);
1276 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1278 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1279 struct f2fs_nm_info *nm_i = NM_I(sbi);
1280 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1282 unsigned int data_sum_blocks, orphan_blocks;
1285 int cp_payload_blks = __cp_payload(sbi);
1286 struct super_block *sb = sbi->sb;
1287 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1291 /* Flush all the NAT/SIT pages */
1292 while (get_pages(sbi, F2FS_DIRTY_META)) {
1293 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1294 if (unlikely(f2fs_cp_error(sbi)))
1300 * version number is already updated
1302 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi, true));
1303 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1304 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1305 ckpt->cur_node_segno[i] =
1306 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1307 ckpt->cur_node_blkoff[i] =
1308 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1309 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1310 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1312 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1313 ckpt->cur_data_segno[i] =
1314 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1315 ckpt->cur_data_blkoff[i] =
1316 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1317 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1318 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1321 /* 2 cp + n data seg summary + orphan inode blocks */
1322 data_sum_blocks = f2fs_npages_for_summary_flush(sbi, false);
1323 spin_lock_irqsave(&sbi->cp_lock, flags);
1324 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1325 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1327 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1328 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1330 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1331 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1334 if (__remain_node_summaries(cpc->reason))
1335 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1336 cp_payload_blks + data_sum_blocks +
1337 orphan_blocks + NR_CURSEG_NODE_TYPE);
1339 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1340 cp_payload_blks + data_sum_blocks +
1343 /* update ckpt flag for checkpoint */
1344 update_ckpt_flags(sbi, cpc);
1346 /* update SIT/NAT bitmap */
1347 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1348 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1350 crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1351 *((__le32 *)((unsigned char *)ckpt +
1352 le32_to_cpu(ckpt->checksum_offset)))
1353 = cpu_to_le32(crc32);
1355 start_blk = __start_cp_next_addr(sbi);
1357 /* write nat bits */
1358 if (enabled_nat_bits(sbi, cpc)) {
1359 __u64 cp_ver = cur_cp_version(ckpt);
1362 cp_ver |= ((__u64)crc32 << 32);
1363 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1365 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1366 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1367 f2fs_update_meta_page(sbi, nm_i->nat_bits +
1368 (i << F2FS_BLKSIZE_BITS), blk + i);
1370 /* Flush all the NAT BITS pages */
1371 while (get_pages(sbi, F2FS_DIRTY_META)) {
1372 f2fs_sync_meta_pages(sbi, META, LONG_MAX,
1374 if (unlikely(f2fs_cp_error(sbi)))
1379 /* write out checkpoint buffer at block 0 */
1380 f2fs_update_meta_page(sbi, ckpt, start_blk++);
1382 for (i = 1; i < 1 + cp_payload_blks; i++)
1383 f2fs_update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1387 write_orphan_inodes(sbi, start_blk);
1388 start_blk += orphan_blocks;
1391 f2fs_write_data_summaries(sbi, start_blk);
1392 start_blk += data_sum_blocks;
1394 /* Record write statistics in the hot node summary */
1395 kbytes_written = sbi->kbytes_written;
1396 if (sb->s_bdev->bd_part)
1397 kbytes_written += BD_PART_WRITTEN(sbi);
1399 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1401 if (__remain_node_summaries(cpc->reason)) {
1402 f2fs_write_node_summaries(sbi, start_blk);
1403 start_blk += NR_CURSEG_NODE_TYPE;
1406 /* update user_block_counts */
1407 sbi->last_valid_block_count = sbi->total_valid_block_count;
1408 percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1410 /* Here, we have one bio having CP pack except cp pack 2 page */
1411 f2fs_sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1413 /* wait for previous submitted meta pages writeback */
1414 f2fs_wait_on_all_pages_writeback(sbi);
1416 /* flush all device cache */
1417 err = f2fs_flush_device_cache(sbi);
1421 /* barrier and flush checkpoint cp pack 2 page if it can */
1422 commit_checkpoint(sbi, ckpt, start_blk);
1423 f2fs_wait_on_all_pages_writeback(sbi);
1426 * invalidate intermediate page cache borrowed from meta inode
1427 * which are used for migration of encrypted inode's blocks.
1429 if (f2fs_sb_has_encrypt(sbi->sb))
1430 invalidate_mapping_pages(META_MAPPING(sbi),
1431 MAIN_BLKADDR(sbi), MAX_BLKADDR(sbi) - 1);
1433 f2fs_release_ino_entry(sbi, false);
1435 f2fs_reset_fsync_node_info(sbi);
1437 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1438 clear_sbi_flag(sbi, SBI_NEED_CP);
1439 __set_cp_next_pack(sbi);
1442 * redirty superblock if metadata like node page or inode cache is
1443 * updated during writing checkpoint.
1445 if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1446 get_pages(sbi, F2FS_DIRTY_IMETA))
1447 set_sbi_flag(sbi, SBI_IS_DIRTY);
1449 f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1451 return unlikely(f2fs_cp_error(sbi)) ? -EIO : 0;
1455 * We guarantee that this checkpoint procedure will not fail.
1457 int f2fs_write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1459 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1460 unsigned long long ckpt_ver;
1463 mutex_lock(&sbi->cp_mutex);
1465 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1466 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1467 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1469 if (unlikely(f2fs_cp_error(sbi))) {
1473 if (f2fs_readonly(sbi->sb)) {
1478 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1480 err = block_operations(sbi);
1484 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1486 f2fs_flush_merged_writes(sbi);
1488 /* this is the case of multiple fstrims without any changes */
1489 if (cpc->reason & CP_DISCARD) {
1490 if (!f2fs_exist_trim_candidates(sbi, cpc)) {
1491 unblock_operations(sbi);
1495 if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1496 SIT_I(sbi)->dirty_sentries == 0 &&
1497 prefree_segments(sbi) == 0) {
1498 f2fs_flush_sit_entries(sbi, cpc);
1499 f2fs_clear_prefree_segments(sbi, cpc);
1500 unblock_operations(sbi);
1506 * update checkpoint pack index
1507 * Increase the version number so that
1508 * SIT entries and seg summaries are written at correct place
1510 ckpt_ver = cur_cp_version(ckpt);
1511 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1513 /* write cached NAT/SIT entries to NAT/SIT area */
1514 f2fs_flush_nat_entries(sbi, cpc);
1515 f2fs_flush_sit_entries(sbi, cpc);
1517 /* unlock all the fs_lock[] in do_checkpoint() */
1518 err = do_checkpoint(sbi, cpc);
1520 f2fs_release_discard_addrs(sbi);
1522 f2fs_clear_prefree_segments(sbi, cpc);
1524 unblock_operations(sbi);
1525 stat_inc_cp_count(sbi->stat_info);
1527 if (cpc->reason & CP_RECOVERY)
1528 f2fs_msg(sbi->sb, KERN_NOTICE,
1529 "checkpoint: version = %llx", ckpt_ver);
1531 /* do checkpoint periodically */
1532 f2fs_update_time(sbi, CP_TIME);
1533 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1535 mutex_unlock(&sbi->cp_mutex);
1539 void f2fs_init_ino_entry_info(struct f2fs_sb_info *sbi)
1543 for (i = 0; i < MAX_INO_ENTRY; i++) {
1544 struct inode_management *im = &sbi->im[i];
1546 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1547 spin_lock_init(&im->ino_lock);
1548 INIT_LIST_HEAD(&im->ino_list);
1552 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1553 NR_CURSEG_TYPE - __cp_payload(sbi)) *
1554 F2FS_ORPHANS_PER_BLOCK;
1557 int __init f2fs_create_checkpoint_caches(void)
1559 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1560 sizeof(struct ino_entry));
1561 if (!ino_entry_slab)
1563 f2fs_inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1564 sizeof(struct inode_entry));
1565 if (!f2fs_inode_entry_slab) {
1566 kmem_cache_destroy(ino_entry_slab);
1572 void f2fs_destroy_checkpoint_caches(void)
1574 kmem_cache_destroy(ino_entry_slab);
1575 kmem_cache_destroy(f2fs_inode_entry_slab);