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 *inode_entry_slab;
29 void f2fs_stop_checkpoint(struct f2fs_sb_info *sbi, bool end_io)
31 set_ckpt_flags(sbi, CP_ERROR_FLAG);
33 f2fs_flush_merged_writes(sbi);
37 * We guarantee no failure on the returned page.
39 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
41 struct address_space *mapping = META_MAPPING(sbi);
42 struct page *page = NULL;
44 page = f2fs_grab_cache_page(mapping, index, false);
49 f2fs_wait_on_page_writeback(page, META, true);
50 if (!PageUptodate(page))
51 SetPageUptodate(page);
56 * We guarantee no failure on the returned page.
58 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
61 struct address_space *mapping = META_MAPPING(sbi);
63 struct f2fs_io_info fio = {
67 .op_flags = REQ_META | REQ_PRIO,
70 .encrypted_page = NULL,
73 if (unlikely(!is_meta))
74 fio.op_flags &= ~REQ_META;
76 page = f2fs_grab_cache_page(mapping, index, false);
81 if (PageUptodate(page))
86 if (f2fs_submit_page_bio(&fio)) {
87 f2fs_put_page(page, 1);
92 if (unlikely(page->mapping != mapping)) {
93 f2fs_put_page(page, 1);
98 * if there is any IO error when accessing device, make our filesystem
99 * readonly and make sure do not write checkpoint with non-uptodate
102 if (unlikely(!PageUptodate(page)))
103 f2fs_stop_checkpoint(sbi, false);
108 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
110 return __get_meta_page(sbi, index, true);
114 struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
116 return __get_meta_page(sbi, index, false);
119 bool is_valid_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr, int type)
125 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
129 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
130 blkaddr < SM_I(sbi)->ssa_blkaddr))
134 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
135 blkaddr < __start_cp_addr(sbi)))
139 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
140 blkaddr < MAIN_BLKADDR(sbi)))
151 * Readahead CP/NAT/SIT/SSA pages
153 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
157 block_t blkno = start;
158 struct f2fs_io_info fio = {
162 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
163 .encrypted_page = NULL,
166 struct blk_plug plug;
168 if (unlikely(type == META_POR))
169 fio.op_flags &= ~REQ_META;
171 blk_start_plug(&plug);
172 for (; nrpages-- > 0; blkno++) {
174 if (!is_valid_blkaddr(sbi, blkno, type))
179 if (unlikely(blkno >=
180 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
182 /* get nat block addr */
183 fio.new_blkaddr = current_nat_addr(sbi,
184 blkno * NAT_ENTRY_PER_BLOCK);
187 /* get sit block addr */
188 fio.new_blkaddr = current_sit_addr(sbi,
189 blkno * SIT_ENTRY_PER_BLOCK);
194 fio.new_blkaddr = blkno;
200 page = f2fs_grab_cache_page(META_MAPPING(sbi),
201 fio.new_blkaddr, false);
204 if (PageUptodate(page)) {
205 f2fs_put_page(page, 1);
210 f2fs_submit_page_bio(&fio);
211 f2fs_put_page(page, 0);
214 blk_finish_plug(&plug);
215 return blkno - start;
218 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
221 bool readahead = false;
223 page = find_get_page(META_MAPPING(sbi), index);
224 if (!page || !PageUptodate(page))
226 f2fs_put_page(page, 0);
229 ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
232 static int __f2fs_write_meta_page(struct page *page,
233 struct writeback_control *wbc,
234 enum iostat_type io_type)
236 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
238 trace_f2fs_writepage(page, META);
240 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
242 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
244 if (unlikely(f2fs_cp_error(sbi)))
247 write_meta_page(sbi, page, io_type);
248 dec_page_count(sbi, F2FS_DIRTY_META);
250 if (wbc->for_reclaim)
251 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
252 0, page->index, META);
256 if (unlikely(f2fs_cp_error(sbi)))
257 f2fs_submit_merged_write(sbi, META);
262 redirty_page_for_writepage(wbc, page);
263 return AOP_WRITEPAGE_ACTIVATE;
266 static int f2fs_write_meta_page(struct page *page,
267 struct writeback_control *wbc)
269 return __f2fs_write_meta_page(page, wbc, FS_META_IO);
272 static int f2fs_write_meta_pages(struct address_space *mapping,
273 struct writeback_control *wbc)
275 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
278 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
281 /* collect a number of dirty meta pages and write together */
282 if (wbc->for_kupdate ||
283 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
286 /* if locked failed, cp will flush dirty pages instead */
287 if (!mutex_trylock(&sbi->cp_mutex))
290 trace_f2fs_writepages(mapping->host, wbc, META);
291 diff = nr_pages_to_write(sbi, META, wbc);
292 written = sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
293 mutex_unlock(&sbi->cp_mutex);
294 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
298 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
299 trace_f2fs_writepages(mapping->host, wbc, META);
303 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
304 long nr_to_write, enum iostat_type io_type)
306 struct address_space *mapping = META_MAPPING(sbi);
307 pgoff_t index = 0, prev = ULONG_MAX;
311 struct writeback_control wbc = {
314 struct blk_plug plug;
318 blk_start_plug(&plug);
320 while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
321 PAGECACHE_TAG_DIRTY))) {
324 for (i = 0; i < nr_pages; i++) {
325 struct page *page = pvec.pages[i];
327 if (prev == ULONG_MAX)
328 prev = page->index - 1;
329 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
330 pagevec_release(&pvec);
336 if (unlikely(page->mapping != mapping)) {
341 if (!PageDirty(page)) {
342 /* someone wrote it for us */
343 goto continue_unlock;
346 f2fs_wait_on_page_writeback(page, META, true);
348 BUG_ON(PageWriteback(page));
349 if (!clear_page_dirty_for_io(page))
350 goto continue_unlock;
352 if (__f2fs_write_meta_page(page, &wbc, io_type)) {
358 if (unlikely(nwritten >= nr_to_write))
361 pagevec_release(&pvec);
366 f2fs_submit_merged_write(sbi, type);
368 blk_finish_plug(&plug);
373 static int f2fs_set_meta_page_dirty(struct page *page)
375 trace_f2fs_set_page_dirty(page, META);
377 if (!PageUptodate(page))
378 SetPageUptodate(page);
379 if (!PageDirty(page)) {
380 f2fs_set_page_dirty_nobuffers(page);
381 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
382 SetPagePrivate(page);
383 f2fs_trace_pid(page);
389 const struct address_space_operations f2fs_meta_aops = {
390 .writepage = f2fs_write_meta_page,
391 .writepages = f2fs_write_meta_pages,
392 .set_page_dirty = f2fs_set_meta_page_dirty,
393 .invalidatepage = f2fs_invalidate_page,
394 .releasepage = f2fs_release_page,
395 #ifdef CONFIG_MIGRATION
396 .migratepage = f2fs_migrate_page,
400 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino,
401 unsigned int devidx, int type)
403 struct inode_management *im = &sbi->im[type];
404 struct ino_entry *e, *tmp;
406 tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
408 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
410 spin_lock(&im->ino_lock);
411 e = radix_tree_lookup(&im->ino_root, ino);
414 if (unlikely(radix_tree_insert(&im->ino_root, ino, e)))
417 memset(e, 0, sizeof(struct ino_entry));
420 list_add_tail(&e->list, &im->ino_list);
421 if (type != ORPHAN_INO)
425 if (type == FLUSH_INO)
426 f2fs_set_bit(devidx, (char *)&e->dirty_device);
428 spin_unlock(&im->ino_lock);
429 radix_tree_preload_end();
432 kmem_cache_free(ino_entry_slab, tmp);
435 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
437 struct inode_management *im = &sbi->im[type];
440 spin_lock(&im->ino_lock);
441 e = radix_tree_lookup(&im->ino_root, ino);
444 radix_tree_delete(&im->ino_root, ino);
446 spin_unlock(&im->ino_lock);
447 kmem_cache_free(ino_entry_slab, e);
450 spin_unlock(&im->ino_lock);
453 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
455 /* add new dirty ino entry into list */
456 __add_ino_entry(sbi, ino, 0, type);
459 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
461 /* remove dirty ino entry from list */
462 __remove_ino_entry(sbi, ino, type);
465 /* mode should be APPEND_INO or UPDATE_INO */
466 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
468 struct inode_management *im = &sbi->im[mode];
471 spin_lock(&im->ino_lock);
472 e = radix_tree_lookup(&im->ino_root, ino);
473 spin_unlock(&im->ino_lock);
474 return e ? true : false;
477 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
479 struct ino_entry *e, *tmp;
482 for (i = all ? ORPHAN_INO : APPEND_INO; i < MAX_INO_ENTRY; i++) {
483 struct inode_management *im = &sbi->im[i];
485 spin_lock(&im->ino_lock);
486 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
488 radix_tree_delete(&im->ino_root, e->ino);
489 kmem_cache_free(ino_entry_slab, e);
492 spin_unlock(&im->ino_lock);
496 void set_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
497 unsigned int devidx, int type)
499 __add_ino_entry(sbi, ino, devidx, type);
502 bool is_dirty_device(struct f2fs_sb_info *sbi, nid_t ino,
503 unsigned int devidx, int type)
505 struct inode_management *im = &sbi->im[type];
507 bool is_dirty = false;
509 spin_lock(&im->ino_lock);
510 e = radix_tree_lookup(&im->ino_root, ino);
511 if (e && f2fs_test_bit(devidx, (char *)&e->dirty_device))
513 spin_unlock(&im->ino_lock);
517 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
519 struct inode_management *im = &sbi->im[ORPHAN_INO];
522 spin_lock(&im->ino_lock);
524 #ifdef CONFIG_F2FS_FAULT_INJECTION
525 if (time_to_inject(sbi, FAULT_ORPHAN)) {
526 spin_unlock(&im->ino_lock);
527 f2fs_show_injection_info(FAULT_ORPHAN);
531 if (unlikely(im->ino_num >= sbi->max_orphans))
535 spin_unlock(&im->ino_lock);
540 void release_orphan_inode(struct f2fs_sb_info *sbi)
542 struct inode_management *im = &sbi->im[ORPHAN_INO];
544 spin_lock(&im->ino_lock);
545 f2fs_bug_on(sbi, im->ino_num == 0);
547 spin_unlock(&im->ino_lock);
550 void add_orphan_inode(struct inode *inode)
552 /* add new orphan ino entry into list */
553 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, 0, ORPHAN_INO);
554 update_inode_page(inode);
557 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
559 /* remove orphan entry from orphan list */
560 __remove_ino_entry(sbi, ino, ORPHAN_INO);
563 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
567 int err = acquire_orphan_inode(sbi);
570 set_sbi_flag(sbi, SBI_NEED_FSCK);
571 f2fs_msg(sbi->sb, KERN_WARNING,
572 "%s: orphan failed (ino=%x), run fsck to fix.",
577 __add_ino_entry(sbi, ino, 0, ORPHAN_INO);
579 inode = f2fs_iget_retry(sbi->sb, ino);
582 * there should be a bug that we can't find the entry
585 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
586 return PTR_ERR(inode);
591 /* truncate all the data during iput */
594 get_node_info(sbi, ino, &ni);
596 /* ENOMEM was fully retried in f2fs_evict_inode. */
597 if (ni.blk_addr != NULL_ADDR) {
598 set_sbi_flag(sbi, SBI_NEED_FSCK);
599 f2fs_msg(sbi->sb, KERN_WARNING,
600 "%s: orphan failed (ino=%x) by kernel, retry mount.",
604 __remove_ino_entry(sbi, ino, ORPHAN_INO);
608 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
610 block_t start_blk, orphan_blocks, i, j;
611 unsigned int s_flags = sbi->sb->s_flags;
617 if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
620 if (s_flags & MS_RDONLY) {
621 f2fs_msg(sbi->sb, KERN_INFO, "orphan cleanup on readonly fs");
622 sbi->sb->s_flags &= ~MS_RDONLY;
626 /* Needed for iput() to work correctly and not trash data */
627 sbi->sb->s_flags |= MS_ACTIVE;
629 /* Turn on quotas so that they are updated correctly */
630 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & MS_RDONLY);
633 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
634 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
636 ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
638 for (i = 0; i < orphan_blocks; i++) {
639 struct page *page = get_meta_page(sbi, start_blk + i);
640 struct f2fs_orphan_block *orphan_blk;
642 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
643 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
644 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
645 err = recover_orphan_inode(sbi, ino);
647 f2fs_put_page(page, 1);
651 f2fs_put_page(page, 1);
653 /* clear Orphan Flag */
654 clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
657 /* Turn quotas off */
659 f2fs_quota_off_umount(sbi->sb);
661 sbi->sb->s_flags = s_flags; /* Restore MS_RDONLY status */
666 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
668 struct list_head *head;
669 struct f2fs_orphan_block *orphan_blk = NULL;
670 unsigned int nentries = 0;
671 unsigned short index = 1;
672 unsigned short orphan_blocks;
673 struct page *page = NULL;
674 struct ino_entry *orphan = NULL;
675 struct inode_management *im = &sbi->im[ORPHAN_INO];
677 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
680 * we don't need to do spin_lock(&im->ino_lock) here, since all the
681 * orphan inode operations are covered under f2fs_lock_op().
682 * And, spin_lock should be avoided due to page operations below.
684 head = &im->ino_list;
686 /* loop for each orphan inode entry and write them in Jornal block */
687 list_for_each_entry(orphan, head, list) {
689 page = grab_meta_page(sbi, start_blk++);
691 (struct f2fs_orphan_block *)page_address(page);
692 memset(orphan_blk, 0, sizeof(*orphan_blk));
695 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
697 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
699 * an orphan block is full of 1020 entries,
700 * then we need to flush current orphan blocks
701 * and bring another one in memory
703 orphan_blk->blk_addr = cpu_to_le16(index);
704 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
705 orphan_blk->entry_count = cpu_to_le32(nentries);
706 set_page_dirty(page);
707 f2fs_put_page(page, 1);
715 orphan_blk->blk_addr = cpu_to_le16(index);
716 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
717 orphan_blk->entry_count = cpu_to_le32(nentries);
718 set_page_dirty(page);
719 f2fs_put_page(page, 1);
723 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
724 struct f2fs_checkpoint **cp_block, struct page **cp_page,
725 unsigned long long *version)
727 unsigned long blk_size = sbi->blocksize;
728 size_t crc_offset = 0;
731 *cp_page = get_meta_page(sbi, cp_addr);
732 *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
734 crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
735 if (crc_offset > (blk_size - sizeof(__le32))) {
736 f2fs_msg(sbi->sb, KERN_WARNING,
737 "invalid crc_offset: %zu", crc_offset);
741 crc = cur_cp_crc(*cp_block);
742 if (!f2fs_crc_valid(sbi, crc, *cp_block, crc_offset)) {
743 f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
747 *version = cur_cp_version(*cp_block);
751 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
752 block_t cp_addr, unsigned long long *version)
754 struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
755 struct f2fs_checkpoint *cp_block = NULL;
756 unsigned long long cur_version = 0, pre_version = 0;
759 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
760 &cp_page_1, version);
763 pre_version = *version;
765 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
766 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
767 &cp_page_2, version);
770 cur_version = *version;
772 if (cur_version == pre_version) {
773 *version = cur_version;
774 f2fs_put_page(cp_page_2, 1);
778 f2fs_put_page(cp_page_2, 1);
780 f2fs_put_page(cp_page_1, 1);
784 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
786 struct f2fs_checkpoint *cp_block;
787 struct f2fs_super_block *fsb = sbi->raw_super;
788 struct page *cp1, *cp2, *cur_page;
789 unsigned long blk_size = sbi->blocksize;
790 unsigned long long cp1_version = 0, cp2_version = 0;
791 unsigned long long cp_start_blk_no;
792 unsigned int cp_blks = 1 + __cp_payload(sbi);
796 sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
800 * Finding out valid cp block involves read both
801 * sets( cp pack1 and cp pack 2)
803 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
804 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
806 /* The second checkpoint pack should start at the next segment */
807 cp_start_blk_no += ((unsigned long long)1) <<
808 le32_to_cpu(fsb->log_blocks_per_seg);
809 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
812 if (ver_after(cp2_version, cp1_version))
824 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
825 memcpy(sbi->ckpt, cp_block, blk_size);
827 /* Sanity checking of checkpoint */
828 if (sanity_check_ckpt(sbi))
829 goto free_fail_no_cp;
832 sbi->cur_cp_pack = 1;
834 sbi->cur_cp_pack = 2;
839 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
841 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
843 for (i = 1; i < cp_blks; i++) {
844 void *sit_bitmap_ptr;
845 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
847 cur_page = get_meta_page(sbi, cp_blk_no + i);
848 sit_bitmap_ptr = page_address(cur_page);
849 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
850 f2fs_put_page(cur_page, 1);
853 f2fs_put_page(cp1, 1);
854 f2fs_put_page(cp2, 1);
858 f2fs_put_page(cp1, 1);
859 f2fs_put_page(cp2, 1);
865 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
867 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
868 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
870 if (is_inode_flag_set(inode, flag))
873 set_inode_flag(inode, flag);
874 if (!f2fs_is_volatile_file(inode))
875 list_add_tail(&F2FS_I(inode)->dirty_list,
876 &sbi->inode_list[type]);
877 stat_inc_dirty_inode(sbi, type);
880 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
882 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
884 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
887 list_del_init(&F2FS_I(inode)->dirty_list);
888 clear_inode_flag(inode, flag);
889 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
892 void update_dirty_page(struct inode *inode, struct page *page)
894 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
895 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
897 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
898 !S_ISLNK(inode->i_mode))
901 spin_lock(&sbi->inode_lock[type]);
902 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
903 __add_dirty_inode(inode, type);
904 inode_inc_dirty_pages(inode);
905 spin_unlock(&sbi->inode_lock[type]);
907 SetPagePrivate(page);
908 f2fs_trace_pid(page);
911 void remove_dirty_inode(struct inode *inode)
913 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
914 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
916 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
917 !S_ISLNK(inode->i_mode))
920 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
923 spin_lock(&sbi->inode_lock[type]);
924 __remove_dirty_inode(inode, type);
925 spin_unlock(&sbi->inode_lock[type]);
928 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
930 struct list_head *head;
932 struct f2fs_inode_info *fi;
933 bool is_dir = (type == DIR_INODE);
934 unsigned long ino = 0;
936 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
937 get_pages(sbi, is_dir ?
938 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
940 if (unlikely(f2fs_cp_error(sbi)))
943 spin_lock(&sbi->inode_lock[type]);
945 head = &sbi->inode_list[type];
946 if (list_empty(head)) {
947 spin_unlock(&sbi->inode_lock[type]);
948 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
949 get_pages(sbi, is_dir ?
950 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
953 fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
954 inode = igrab(&fi->vfs_inode);
955 spin_unlock(&sbi->inode_lock[type]);
957 unsigned long cur_ino = inode->i_ino;
960 F2FS_I(inode)->cp_task = current;
962 filemap_fdatawrite(inode->i_mapping);
965 F2FS_I(inode)->cp_task = NULL;
968 /* We need to give cpu to another writers. */
969 if (ino == cur_ino) {
970 congestion_wait(BLK_RW_ASYNC, HZ/50);
977 * We should submit bio, since it exists several
978 * wribacking dentry pages in the freeing inode.
980 f2fs_submit_merged_write(sbi, DATA);
986 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
988 struct list_head *head = &sbi->inode_list[DIRTY_META];
990 struct f2fs_inode_info *fi;
991 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
994 if (unlikely(f2fs_cp_error(sbi)))
997 spin_lock(&sbi->inode_lock[DIRTY_META]);
998 if (list_empty(head)) {
999 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1002 fi = list_first_entry(head, struct f2fs_inode_info,
1004 inode = igrab(&fi->vfs_inode);
1005 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1007 sync_inode_metadata(inode, 0);
1009 /* it's on eviction */
1010 if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1011 update_inode_page(inode);
1018 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1020 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1021 struct f2fs_nm_info *nm_i = NM_I(sbi);
1022 nid_t last_nid = nm_i->next_scan_nid;
1024 next_free_nid(sbi, &last_nid);
1025 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1026 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1027 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1028 ckpt->next_free_nid = cpu_to_le32(last_nid);
1032 * Freeze all the FS-operations for checkpoint.
1034 static int block_operations(struct f2fs_sb_info *sbi)
1036 struct writeback_control wbc = {
1037 .sync_mode = WB_SYNC_ALL,
1038 .nr_to_write = LONG_MAX,
1041 struct blk_plug plug;
1044 blk_start_plug(&plug);
1048 /* write all the dirty dentry pages */
1049 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1050 f2fs_unlock_all(sbi);
1051 err = sync_dirty_inodes(sbi, DIR_INODE);
1055 goto retry_flush_dents;
1059 * POR: we should ensure that there are no dirty node pages
1060 * until finishing nat/sit flush. inode->i_blocks can be updated.
1062 down_write(&sbi->node_change);
1064 if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1065 up_write(&sbi->node_change);
1066 f2fs_unlock_all(sbi);
1067 err = f2fs_sync_inode_meta(sbi);
1071 goto retry_flush_dents;
1075 down_write(&sbi->node_write);
1077 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1078 up_write(&sbi->node_write);
1079 err = sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1081 up_write(&sbi->node_change);
1082 f2fs_unlock_all(sbi);
1086 goto retry_flush_nodes;
1090 * sbi->node_change is used only for AIO write_begin path which produces
1091 * dirty node blocks and some checkpoint values by block allocation.
1093 __prepare_cp_block(sbi);
1094 up_write(&sbi->node_change);
1096 blk_finish_plug(&plug);
1100 static void unblock_operations(struct f2fs_sb_info *sbi)
1102 up_write(&sbi->node_write);
1103 f2fs_unlock_all(sbi);
1106 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
1111 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1113 if (!get_pages(sbi, F2FS_WB_CP_DATA))
1116 io_schedule_timeout(5*HZ);
1118 finish_wait(&sbi->cp_wait, &wait);
1121 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1123 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1124 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1125 unsigned long flags;
1127 spin_lock_irqsave(&sbi->cp_lock, flags);
1129 if ((cpc->reason & CP_UMOUNT) &&
1130 le32_to_cpu(ckpt->cp_pack_total_block_count) >
1131 sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1132 disable_nat_bits(sbi, false);
1134 if (cpc->reason & CP_TRIMMED)
1135 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1137 if (cpc->reason & CP_UMOUNT)
1138 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1140 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1142 if (cpc->reason & CP_FASTBOOT)
1143 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1145 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1148 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1150 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1152 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1153 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1155 /* set this flag to activate crc|cp_ver for recovery */
1156 __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1158 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1161 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1163 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1164 struct f2fs_nm_info *nm_i = NM_I(sbi);
1165 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1167 unsigned int data_sum_blocks, orphan_blocks;
1170 int cp_payload_blks = __cp_payload(sbi);
1171 struct super_block *sb = sbi->sb;
1172 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1176 /* Flush all the NAT/SIT pages */
1177 while (get_pages(sbi, F2FS_DIRTY_META)) {
1178 sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1179 if (unlikely(f2fs_cp_error(sbi)))
1185 * version number is already updated
1187 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
1188 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1189 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1190 ckpt->cur_node_segno[i] =
1191 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1192 ckpt->cur_node_blkoff[i] =
1193 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1194 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1195 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1197 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1198 ckpt->cur_data_segno[i] =
1199 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1200 ckpt->cur_data_blkoff[i] =
1201 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1202 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1203 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1206 /* 2 cp + n data seg summary + orphan inode blocks */
1207 data_sum_blocks = npages_for_summary_flush(sbi, false);
1208 spin_lock_irqsave(&sbi->cp_lock, flags);
1209 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1210 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1212 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1213 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1215 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1216 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1219 if (__remain_node_summaries(cpc->reason))
1220 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1221 cp_payload_blks + data_sum_blocks +
1222 orphan_blocks + NR_CURSEG_NODE_TYPE);
1224 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1225 cp_payload_blks + data_sum_blocks +
1228 /* update ckpt flag for checkpoint */
1229 update_ckpt_flags(sbi, cpc);
1231 /* update SIT/NAT bitmap */
1232 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1233 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1235 crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1236 *((__le32 *)((unsigned char *)ckpt +
1237 le32_to_cpu(ckpt->checksum_offset)))
1238 = cpu_to_le32(crc32);
1240 start_blk = __start_cp_next_addr(sbi);
1242 /* write nat bits */
1243 if (enabled_nat_bits(sbi, cpc)) {
1244 __u64 cp_ver = cur_cp_version(ckpt);
1247 cp_ver |= ((__u64)crc32 << 32);
1248 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1250 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1251 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1252 update_meta_page(sbi, nm_i->nat_bits +
1253 (i << F2FS_BLKSIZE_BITS), blk + i);
1255 /* Flush all the NAT BITS pages */
1256 while (get_pages(sbi, F2FS_DIRTY_META)) {
1257 sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1258 if (unlikely(f2fs_cp_error(sbi)))
1263 /* need to wait for end_io results */
1264 wait_on_all_pages_writeback(sbi);
1265 if (unlikely(f2fs_cp_error(sbi)))
1268 /* flush all device cache */
1269 err = f2fs_flush_device_cache(sbi);
1273 /* write out checkpoint buffer at block 0 */
1274 update_meta_page(sbi, ckpt, start_blk++);
1276 for (i = 1; i < 1 + cp_payload_blks; i++)
1277 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1281 write_orphan_inodes(sbi, start_blk);
1282 start_blk += orphan_blocks;
1285 write_data_summaries(sbi, start_blk);
1286 start_blk += data_sum_blocks;
1288 /* Record write statistics in the hot node summary */
1289 kbytes_written = sbi->kbytes_written;
1290 if (sb->s_bdev->bd_part)
1291 kbytes_written += BD_PART_WRITTEN(sbi);
1293 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1295 if (__remain_node_summaries(cpc->reason)) {
1296 write_node_summaries(sbi, start_blk);
1297 start_blk += NR_CURSEG_NODE_TYPE;
1300 /* writeout checkpoint block */
1301 update_meta_page(sbi, ckpt, start_blk);
1303 /* wait for previous submitted node/meta pages writeback */
1304 wait_on_all_pages_writeback(sbi);
1306 if (unlikely(f2fs_cp_error(sbi)))
1309 filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1310 filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1312 /* update user_block_counts */
1313 sbi->last_valid_block_count = sbi->total_valid_block_count;
1314 percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1316 /* Here, we only have one bio having CP pack */
1317 sync_meta_pages(sbi, META_FLUSH, LONG_MAX, FS_CP_META_IO);
1319 /* wait for previous submitted meta pages writeback */
1320 wait_on_all_pages_writeback(sbi);
1322 release_ino_entry(sbi, false);
1324 if (unlikely(f2fs_cp_error(sbi)))
1327 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1328 clear_sbi_flag(sbi, SBI_NEED_CP);
1329 __set_cp_next_pack(sbi);
1332 * redirty superblock if metadata like node page or inode cache is
1333 * updated during writing checkpoint.
1335 if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1336 get_pages(sbi, F2FS_DIRTY_IMETA))
1337 set_sbi_flag(sbi, SBI_IS_DIRTY);
1339 f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1345 * We guarantee that this checkpoint procedure will not fail.
1347 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1349 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1350 unsigned long long ckpt_ver;
1353 mutex_lock(&sbi->cp_mutex);
1355 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1356 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1357 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1359 if (unlikely(f2fs_cp_error(sbi))) {
1363 if (f2fs_readonly(sbi->sb)) {
1368 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1370 err = block_operations(sbi);
1374 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1376 f2fs_flush_merged_writes(sbi);
1378 /* this is the case of multiple fstrims without any changes */
1379 if (cpc->reason & CP_DISCARD) {
1380 if (!exist_trim_candidates(sbi, cpc)) {
1381 unblock_operations(sbi);
1385 if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1386 SIT_I(sbi)->dirty_sentries == 0 &&
1387 prefree_segments(sbi) == 0) {
1388 flush_sit_entries(sbi, cpc);
1389 clear_prefree_segments(sbi, cpc);
1390 unblock_operations(sbi);
1396 * update checkpoint pack index
1397 * Increase the version number so that
1398 * SIT entries and seg summaries are written at correct place
1400 ckpt_ver = cur_cp_version(ckpt);
1401 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1403 /* write cached NAT/SIT entries to NAT/SIT area */
1404 flush_nat_entries(sbi, cpc);
1405 flush_sit_entries(sbi, cpc);
1407 /* unlock all the fs_lock[] in do_checkpoint() */
1408 err = do_checkpoint(sbi, cpc);
1410 release_discard_addrs(sbi);
1412 clear_prefree_segments(sbi, cpc);
1414 unblock_operations(sbi);
1415 stat_inc_cp_count(sbi->stat_info);
1417 if (cpc->reason & CP_RECOVERY)
1418 f2fs_msg(sbi->sb, KERN_NOTICE,
1419 "checkpoint: version = %llx", ckpt_ver);
1421 /* do checkpoint periodically */
1422 f2fs_update_time(sbi, CP_TIME);
1423 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1425 mutex_unlock(&sbi->cp_mutex);
1429 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1433 for (i = 0; i < MAX_INO_ENTRY; i++) {
1434 struct inode_management *im = &sbi->im[i];
1436 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1437 spin_lock_init(&im->ino_lock);
1438 INIT_LIST_HEAD(&im->ino_list);
1442 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1443 NR_CURSEG_TYPE - __cp_payload(sbi)) *
1444 F2FS_ORPHANS_PER_BLOCK;
1447 int __init create_checkpoint_caches(void)
1449 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1450 sizeof(struct ino_entry));
1451 if (!ino_entry_slab)
1453 inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1454 sizeof(struct inode_entry));
1455 if (!inode_entry_slab) {
1456 kmem_cache_destroy(ino_entry_slab);
1462 void destroy_checkpoint_caches(void)
1464 kmem_cache_destroy(ino_entry_slab);
1465 kmem_cache_destroy(inode_entry_slab);