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);
32 sbi->sb->s_flags |= MS_RDONLY;
34 f2fs_flush_merged_bios(sbi);
38 * We guarantee no failure on the returned page.
40 struct page *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,
74 if (unlikely(!is_meta))
75 fio.op_flags &= ~REQ_META;
77 page = f2fs_grab_cache_page(mapping, index, false);
82 if (PageUptodate(page))
87 if (f2fs_submit_page_bio(&fio)) {
88 f2fs_put_page(page, 1);
93 if (unlikely(page->mapping != mapping)) {
94 f2fs_put_page(page, 1);
99 * if there is any IO error when accessing device, make our filesystem
100 * readonly and make sure do not write checkpoint with non-uptodate
103 if (unlikely(!PageUptodate(page)))
104 f2fs_stop_checkpoint(sbi, false);
109 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
111 return __get_meta_page(sbi, index, true);
115 struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
117 return __get_meta_page(sbi, index, false);
120 bool is_valid_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr, int type)
126 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
130 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
131 blkaddr < SM_I(sbi)->ssa_blkaddr))
135 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
136 blkaddr < __start_cp_addr(sbi)))
140 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
141 blkaddr < MAIN_BLKADDR(sbi)))
152 * Readahead CP/NAT/SIT/SSA pages
154 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
158 block_t blkno = start;
159 struct f2fs_io_info fio = {
163 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
164 .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 fio.old_blkaddr = fio.new_blkaddr;
211 f2fs_submit_page_mbio(&fio);
212 f2fs_put_page(page, 0);
215 f2fs_submit_merged_bio(sbi, META, READ);
216 blk_finish_plug(&plug);
217 return blkno - start;
220 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
223 bool readahead = false;
225 page = find_get_page(META_MAPPING(sbi), index);
226 if (!page || !PageUptodate(page))
228 f2fs_put_page(page, 0);
231 ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
234 static int f2fs_write_meta_page(struct page *page,
235 struct writeback_control *wbc)
237 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
239 trace_f2fs_writepage(page, META);
241 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
243 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
245 if (unlikely(f2fs_cp_error(sbi)))
248 write_meta_page(sbi, page);
249 dec_page_count(sbi, F2FS_DIRTY_META);
251 if (wbc->for_reclaim)
252 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, META, WRITE);
256 if (unlikely(f2fs_cp_error(sbi)))
257 f2fs_submit_merged_bio(sbi, META, WRITE);
262 redirty_page_for_writepage(wbc, page);
263 return AOP_WRITEPAGE_ACTIVATE;
266 static int f2fs_write_meta_pages(struct address_space *mapping,
267 struct writeback_control *wbc)
269 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
272 /* collect a number of dirty meta pages and write together */
273 if (wbc->for_kupdate ||
274 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
277 trace_f2fs_writepages(mapping->host, wbc, META);
279 /* if mounting is failed, skip writing node pages */
280 mutex_lock(&sbi->cp_mutex);
281 diff = nr_pages_to_write(sbi, META, wbc);
282 written = sync_meta_pages(sbi, META, wbc->nr_to_write);
283 mutex_unlock(&sbi->cp_mutex);
284 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
288 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
289 trace_f2fs_writepages(mapping->host, wbc, META);
293 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
296 struct address_space *mapping = META_MAPPING(sbi);
297 pgoff_t index = 0, end = ULONG_MAX, prev = ULONG_MAX;
300 struct writeback_control wbc = {
303 struct blk_plug plug;
305 pagevec_init(&pvec, 0);
307 blk_start_plug(&plug);
309 while (index <= end) {
311 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
313 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
314 if (unlikely(nr_pages == 0))
317 for (i = 0; i < nr_pages; i++) {
318 struct page *page = pvec.pages[i];
320 if (prev == ULONG_MAX)
321 prev = page->index - 1;
322 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
323 pagevec_release(&pvec);
329 if (unlikely(page->mapping != mapping)) {
334 if (!PageDirty(page)) {
335 /* someone wrote it for us */
336 goto continue_unlock;
339 f2fs_wait_on_page_writeback(page, META, true);
341 BUG_ON(PageWriteback(page));
342 if (!clear_page_dirty_for_io(page))
343 goto continue_unlock;
345 if (mapping->a_ops->writepage(page, &wbc)) {
351 if (unlikely(nwritten >= nr_to_write))
354 pagevec_release(&pvec);
359 f2fs_submit_merged_bio(sbi, type, WRITE);
361 blk_finish_plug(&plug);
366 static int f2fs_set_meta_page_dirty(struct page *page)
368 trace_f2fs_set_page_dirty(page, META);
370 if (!PageUptodate(page))
371 SetPageUptodate(page);
372 if (!PageDirty(page)) {
373 f2fs_set_page_dirty_nobuffers(page);
374 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
375 SetPagePrivate(page);
376 f2fs_trace_pid(page);
382 const struct address_space_operations f2fs_meta_aops = {
383 .writepage = f2fs_write_meta_page,
384 .writepages = f2fs_write_meta_pages,
385 .set_page_dirty = f2fs_set_meta_page_dirty,
386 .invalidatepage = f2fs_invalidate_page,
387 .releasepage = f2fs_release_page,
388 #ifdef CONFIG_MIGRATION
389 .migratepage = f2fs_migrate_page,
393 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
395 struct inode_management *im = &sbi->im[type];
396 struct ino_entry *e, *tmp;
398 tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
400 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
402 spin_lock(&im->ino_lock);
403 e = radix_tree_lookup(&im->ino_root, ino);
406 if (radix_tree_insert(&im->ino_root, ino, e)) {
407 spin_unlock(&im->ino_lock);
408 radix_tree_preload_end();
411 memset(e, 0, sizeof(struct ino_entry));
414 list_add_tail(&e->list, &im->ino_list);
415 if (type != ORPHAN_INO)
418 spin_unlock(&im->ino_lock);
419 radix_tree_preload_end();
422 kmem_cache_free(ino_entry_slab, tmp);
425 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
427 struct inode_management *im = &sbi->im[type];
430 spin_lock(&im->ino_lock);
431 e = radix_tree_lookup(&im->ino_root, ino);
434 radix_tree_delete(&im->ino_root, ino);
436 spin_unlock(&im->ino_lock);
437 kmem_cache_free(ino_entry_slab, e);
440 spin_unlock(&im->ino_lock);
443 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
445 /* add new dirty ino entry into list */
446 __add_ino_entry(sbi, ino, type);
449 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
451 /* remove dirty ino entry from list */
452 __remove_ino_entry(sbi, ino, type);
455 /* mode should be APPEND_INO or UPDATE_INO */
456 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
458 struct inode_management *im = &sbi->im[mode];
461 spin_lock(&im->ino_lock);
462 e = radix_tree_lookup(&im->ino_root, ino);
463 spin_unlock(&im->ino_lock);
464 return e ? true : false;
467 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
469 struct ino_entry *e, *tmp;
472 for (i = all ? ORPHAN_INO: APPEND_INO; i <= UPDATE_INO; i++) {
473 struct inode_management *im = &sbi->im[i];
475 spin_lock(&im->ino_lock);
476 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
478 radix_tree_delete(&im->ino_root, e->ino);
479 kmem_cache_free(ino_entry_slab, e);
482 spin_unlock(&im->ino_lock);
486 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
488 struct inode_management *im = &sbi->im[ORPHAN_INO];
491 spin_lock(&im->ino_lock);
493 #ifdef CONFIG_F2FS_FAULT_INJECTION
494 if (time_to_inject(sbi, FAULT_ORPHAN)) {
495 spin_unlock(&im->ino_lock);
499 if (unlikely(im->ino_num >= sbi->max_orphans))
503 spin_unlock(&im->ino_lock);
508 void release_orphan_inode(struct f2fs_sb_info *sbi)
510 struct inode_management *im = &sbi->im[ORPHAN_INO];
512 spin_lock(&im->ino_lock);
513 f2fs_bug_on(sbi, im->ino_num == 0);
515 spin_unlock(&im->ino_lock);
518 void add_orphan_inode(struct inode *inode)
520 /* add new orphan ino entry into list */
521 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, ORPHAN_INO);
522 update_inode_page(inode);
525 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
527 /* remove orphan entry from orphan list */
528 __remove_ino_entry(sbi, ino, ORPHAN_INO);
531 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
535 int err = acquire_orphan_inode(sbi);
538 set_sbi_flag(sbi, SBI_NEED_FSCK);
539 f2fs_msg(sbi->sb, KERN_WARNING,
540 "%s: orphan failed (ino=%x), run fsck to fix.",
545 __add_ino_entry(sbi, ino, ORPHAN_INO);
547 inode = f2fs_iget_retry(sbi->sb, ino);
550 * there should be a bug that we can't find the entry
553 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
554 return PTR_ERR(inode);
559 /* truncate all the data during iput */
562 get_node_info(sbi, ino, &ni);
564 /* ENOMEM was fully retried in f2fs_evict_inode. */
565 if (ni.blk_addr != NULL_ADDR) {
566 set_sbi_flag(sbi, SBI_NEED_FSCK);
567 f2fs_msg(sbi->sb, KERN_WARNING,
568 "%s: orphan failed (ino=%x), run fsck to fix.",
572 __remove_ino_entry(sbi, ino, ORPHAN_INO);
576 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
578 block_t start_blk, orphan_blocks, i, j;
581 if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
584 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
585 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
587 ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
589 for (i = 0; i < orphan_blocks; i++) {
590 struct page *page = get_meta_page(sbi, start_blk + i);
591 struct f2fs_orphan_block *orphan_blk;
593 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
594 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
595 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
596 err = recover_orphan_inode(sbi, ino);
598 f2fs_put_page(page, 1);
602 f2fs_put_page(page, 1);
604 /* clear Orphan Flag */
605 clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
609 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
611 struct list_head *head;
612 struct f2fs_orphan_block *orphan_blk = NULL;
613 unsigned int nentries = 0;
614 unsigned short index = 1;
615 unsigned short orphan_blocks;
616 struct page *page = NULL;
617 struct ino_entry *orphan = NULL;
618 struct inode_management *im = &sbi->im[ORPHAN_INO];
620 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
623 * we don't need to do spin_lock(&im->ino_lock) here, since all the
624 * orphan inode operations are covered under f2fs_lock_op().
625 * And, spin_lock should be avoided due to page operations below.
627 head = &im->ino_list;
629 /* loop for each orphan inode entry and write them in Jornal block */
630 list_for_each_entry(orphan, head, list) {
632 page = grab_meta_page(sbi, start_blk++);
634 (struct f2fs_orphan_block *)page_address(page);
635 memset(orphan_blk, 0, sizeof(*orphan_blk));
638 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
640 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
642 * an orphan block is full of 1020 entries,
643 * then we need to flush current orphan blocks
644 * and bring another one in memory
646 orphan_blk->blk_addr = cpu_to_le16(index);
647 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
648 orphan_blk->entry_count = cpu_to_le32(nentries);
649 set_page_dirty(page);
650 f2fs_put_page(page, 1);
658 orphan_blk->blk_addr = cpu_to_le16(index);
659 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
660 orphan_blk->entry_count = cpu_to_le32(nentries);
661 set_page_dirty(page);
662 f2fs_put_page(page, 1);
666 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
667 struct f2fs_checkpoint **cp_block, struct page **cp_page,
668 unsigned long long *version)
670 unsigned long blk_size = sbi->blocksize;
671 size_t crc_offset = 0;
674 *cp_page = get_meta_page(sbi, cp_addr);
675 *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
677 crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
678 if (crc_offset >= blk_size) {
679 f2fs_msg(sbi->sb, KERN_WARNING,
680 "invalid crc_offset: %zu", crc_offset);
684 crc = le32_to_cpu(*((__le32 *)((unsigned char *)*cp_block
686 if (!f2fs_crc_valid(sbi, crc, *cp_block, crc_offset)) {
687 f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
691 *version = cur_cp_version(*cp_block);
695 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
696 block_t cp_addr, unsigned long long *version)
698 struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
699 struct f2fs_checkpoint *cp_block = NULL;
700 unsigned long long cur_version = 0, pre_version = 0;
703 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
704 &cp_page_1, version);
707 pre_version = *version;
709 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
710 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
711 &cp_page_2, version);
714 cur_version = *version;
716 if (cur_version == pre_version) {
717 *version = cur_version;
718 f2fs_put_page(cp_page_2, 1);
722 f2fs_put_page(cp_page_2, 1);
724 f2fs_put_page(cp_page_1, 1);
728 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
730 struct f2fs_checkpoint *cp_block;
731 struct f2fs_super_block *fsb = sbi->raw_super;
732 struct page *cp1, *cp2, *cur_page;
733 unsigned long blk_size = sbi->blocksize;
734 unsigned long long cp1_version = 0, cp2_version = 0;
735 unsigned long long cp_start_blk_no;
736 unsigned int cp_blks = 1 + __cp_payload(sbi);
740 sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
744 * Finding out valid cp block involves read both
745 * sets( cp pack1 and cp pack 2)
747 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
748 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
750 /* The second checkpoint pack should start at the next segment */
751 cp_start_blk_no += ((unsigned long long)1) <<
752 le32_to_cpu(fsb->log_blocks_per_seg);
753 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
756 if (ver_after(cp2_version, cp1_version))
768 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
769 memcpy(sbi->ckpt, cp_block, blk_size);
771 /* Sanity checking of checkpoint */
772 if (sanity_check_ckpt(sbi))
773 goto free_fail_no_cp;
776 sbi->cur_cp_pack = 1;
778 sbi->cur_cp_pack = 2;
783 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
785 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
787 for (i = 1; i < cp_blks; i++) {
788 void *sit_bitmap_ptr;
789 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
791 cur_page = get_meta_page(sbi, cp_blk_no + i);
792 sit_bitmap_ptr = page_address(cur_page);
793 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
794 f2fs_put_page(cur_page, 1);
797 f2fs_put_page(cp1, 1);
798 f2fs_put_page(cp2, 1);
802 f2fs_put_page(cp1, 1);
803 f2fs_put_page(cp2, 1);
809 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
811 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
812 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
814 if (is_inode_flag_set(inode, flag))
817 set_inode_flag(inode, flag);
818 list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
819 stat_inc_dirty_inode(sbi, type);
822 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
824 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
826 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
829 list_del_init(&F2FS_I(inode)->dirty_list);
830 clear_inode_flag(inode, flag);
831 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
834 void update_dirty_page(struct inode *inode, struct page *page)
836 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
837 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
839 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
840 !S_ISLNK(inode->i_mode))
843 spin_lock(&sbi->inode_lock[type]);
844 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
845 __add_dirty_inode(inode, type);
846 inode_inc_dirty_pages(inode);
847 spin_unlock(&sbi->inode_lock[type]);
849 SetPagePrivate(page);
850 f2fs_trace_pid(page);
853 void remove_dirty_inode(struct inode *inode)
855 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
856 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
858 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
859 !S_ISLNK(inode->i_mode))
862 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
865 spin_lock(&sbi->inode_lock[type]);
866 __remove_dirty_inode(inode, type);
867 spin_unlock(&sbi->inode_lock[type]);
870 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
872 struct list_head *head;
874 struct f2fs_inode_info *fi;
875 bool is_dir = (type == DIR_INODE);
877 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
878 get_pages(sbi, is_dir ?
879 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
881 if (unlikely(f2fs_cp_error(sbi)))
884 spin_lock(&sbi->inode_lock[type]);
886 head = &sbi->inode_list[type];
887 if (list_empty(head)) {
888 spin_unlock(&sbi->inode_lock[type]);
889 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
890 get_pages(sbi, is_dir ?
891 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
894 fi = list_entry(head->next, struct f2fs_inode_info, dirty_list);
895 inode = igrab(&fi->vfs_inode);
896 spin_unlock(&sbi->inode_lock[type]);
898 filemap_fdatawrite(inode->i_mapping);
902 * We should submit bio, since it exists several
903 * wribacking dentry pages in the freeing inode.
905 f2fs_submit_merged_bio(sbi, DATA, WRITE);
911 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
913 struct list_head *head = &sbi->inode_list[DIRTY_META];
915 struct f2fs_inode_info *fi;
916 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
919 if (unlikely(f2fs_cp_error(sbi)))
922 spin_lock(&sbi->inode_lock[DIRTY_META]);
923 if (list_empty(head)) {
924 spin_unlock(&sbi->inode_lock[DIRTY_META]);
927 fi = list_entry(head->next, struct f2fs_inode_info,
929 inode = igrab(&fi->vfs_inode);
930 spin_unlock(&sbi->inode_lock[DIRTY_META]);
932 sync_inode_metadata(inode, 0);
934 /* it's on eviction */
935 if (is_inode_flag_set(inode, FI_DIRTY_INODE))
936 update_inode_page(inode);
944 * Freeze all the FS-operations for checkpoint.
946 static int block_operations(struct f2fs_sb_info *sbi)
948 struct writeback_control wbc = {
949 .sync_mode = WB_SYNC_ALL,
950 .nr_to_write = LONG_MAX,
953 struct blk_plug plug;
956 blk_start_plug(&plug);
960 /* write all the dirty dentry pages */
961 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
962 f2fs_unlock_all(sbi);
963 err = sync_dirty_inodes(sbi, DIR_INODE);
966 goto retry_flush_dents;
969 if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
970 f2fs_unlock_all(sbi);
971 err = f2fs_sync_inode_meta(sbi);
974 goto retry_flush_dents;
978 * POR: we should ensure that there are no dirty node pages
979 * until finishing nat/sit flush.
982 down_write(&sbi->node_write);
984 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
985 up_write(&sbi->node_write);
986 err = sync_node_pages(sbi, &wbc);
988 f2fs_unlock_all(sbi);
991 goto retry_flush_nodes;
994 blk_finish_plug(&plug);
998 static void unblock_operations(struct f2fs_sb_info *sbi)
1000 up_write(&sbi->node_write);
1002 build_free_nids(sbi, false);
1003 f2fs_unlock_all(sbi);
1006 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
1011 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1013 if (!get_pages(sbi, F2FS_WB_CP_DATA))
1016 io_schedule_timeout(5*HZ);
1018 finish_wait(&sbi->cp_wait, &wait);
1021 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1023 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1024 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1026 spin_lock(&sbi->cp_lock);
1028 if (cpc->reason == CP_UMOUNT)
1029 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1031 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1033 if (cpc->reason == CP_FASTBOOT)
1034 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1036 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1039 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1041 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1043 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1044 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1046 /* set this flag to activate crc|cp_ver for recovery */
1047 __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1049 spin_unlock(&sbi->cp_lock);
1052 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1054 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1055 struct f2fs_nm_info *nm_i = NM_I(sbi);
1056 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1057 nid_t last_nid = nm_i->next_scan_nid;
1059 unsigned int data_sum_blocks, orphan_blocks;
1062 int cp_payload_blks = __cp_payload(sbi);
1063 struct super_block *sb = sbi->sb;
1064 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1067 /* Flush all the NAT/SIT pages */
1068 while (get_pages(sbi, F2FS_DIRTY_META)) {
1069 sync_meta_pages(sbi, META, LONG_MAX);
1070 if (unlikely(f2fs_cp_error(sbi)))
1074 next_free_nid(sbi, &last_nid);
1078 * version number is already updated
1080 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
1081 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1082 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1083 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1084 ckpt->cur_node_segno[i] =
1085 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1086 ckpt->cur_node_blkoff[i] =
1087 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1088 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1089 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1091 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1092 ckpt->cur_data_segno[i] =
1093 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1094 ckpt->cur_data_blkoff[i] =
1095 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1096 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1097 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1100 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1101 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1102 ckpt->next_free_nid = cpu_to_le32(last_nid);
1104 /* 2 cp + n data seg summary + orphan inode blocks */
1105 data_sum_blocks = npages_for_summary_flush(sbi, false);
1106 spin_lock(&sbi->cp_lock);
1107 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1108 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1110 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1111 spin_unlock(&sbi->cp_lock);
1113 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1114 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1117 if (__remain_node_summaries(cpc->reason))
1118 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1119 cp_payload_blks + data_sum_blocks +
1120 orphan_blocks + NR_CURSEG_NODE_TYPE);
1122 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1123 cp_payload_blks + data_sum_blocks +
1126 /* update ckpt flag for checkpoint */
1127 update_ckpt_flags(sbi, cpc);
1129 /* update SIT/NAT bitmap */
1130 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1131 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1133 crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1134 *((__le32 *)((unsigned char *)ckpt +
1135 le32_to_cpu(ckpt->checksum_offset)))
1136 = cpu_to_le32(crc32);
1138 start_blk = __start_cp_next_addr(sbi);
1140 /* need to wait for end_io results */
1141 wait_on_all_pages_writeback(sbi);
1142 if (unlikely(f2fs_cp_error(sbi)))
1145 /* write out checkpoint buffer at block 0 */
1146 update_meta_page(sbi, ckpt, start_blk++);
1148 for (i = 1; i < 1 + cp_payload_blks; i++)
1149 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1153 write_orphan_inodes(sbi, start_blk);
1154 start_blk += orphan_blocks;
1157 write_data_summaries(sbi, start_blk);
1158 start_blk += data_sum_blocks;
1160 /* Record write statistics in the hot node summary */
1161 kbytes_written = sbi->kbytes_written;
1162 if (sb->s_bdev->bd_part)
1163 kbytes_written += BD_PART_WRITTEN(sbi);
1165 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1167 if (__remain_node_summaries(cpc->reason)) {
1168 write_node_summaries(sbi, start_blk);
1169 start_blk += NR_CURSEG_NODE_TYPE;
1172 /* writeout checkpoint block */
1173 update_meta_page(sbi, ckpt, start_blk);
1175 /* wait for previous submitted node/meta pages writeback */
1176 wait_on_all_pages_writeback(sbi);
1178 if (unlikely(f2fs_cp_error(sbi)))
1181 filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1182 filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1184 /* update user_block_counts */
1185 sbi->last_valid_block_count = sbi->total_valid_block_count;
1186 percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1188 /* Here, we only have one bio having CP pack */
1189 sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1191 /* wait for previous submitted meta pages writeback */
1192 wait_on_all_pages_writeback(sbi);
1194 release_ino_entry(sbi, false);
1196 if (unlikely(f2fs_cp_error(sbi)))
1199 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1200 clear_sbi_flag(sbi, SBI_NEED_CP);
1201 __set_cp_next_pack(sbi);
1204 * redirty superblock if metadata like node page or inode cache is
1205 * updated during writing checkpoint.
1207 if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1208 get_pages(sbi, F2FS_DIRTY_IMETA))
1209 set_sbi_flag(sbi, SBI_IS_DIRTY);
1211 f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1217 * We guarantee that this checkpoint procedure will not fail.
1219 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1221 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1222 unsigned long long ckpt_ver;
1225 mutex_lock(&sbi->cp_mutex);
1227 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1228 (cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1229 (cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1231 if (unlikely(f2fs_cp_error(sbi))) {
1235 if (f2fs_readonly(sbi->sb)) {
1240 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1242 err = block_operations(sbi);
1246 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1248 f2fs_flush_merged_bios(sbi);
1250 /* this is the case of multiple fstrims without any changes */
1251 if (cpc->reason == CP_DISCARD) {
1252 if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1253 SIT_I(sbi)->dirty_sentries == 0 &&
1254 prefree_segments(sbi) == 0) {
1255 flush_sit_entries(sbi, cpc);
1256 clear_prefree_segments(sbi, cpc);
1257 unblock_operations(sbi);
1263 * update checkpoint pack index
1264 * Increase the version number so that
1265 * SIT entries and seg summaries are written at correct place
1267 ckpt_ver = cur_cp_version(ckpt);
1268 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1270 /* write cached NAT/SIT entries to NAT/SIT area */
1271 flush_nat_entries(sbi);
1272 flush_sit_entries(sbi, cpc);
1274 /* unlock all the fs_lock[] in do_checkpoint() */
1275 err = do_checkpoint(sbi, cpc);
1277 release_discard_addrs(sbi);
1279 clear_prefree_segments(sbi, cpc);
1281 unblock_operations(sbi);
1282 stat_inc_cp_count(sbi->stat_info);
1284 if (cpc->reason == CP_RECOVERY)
1285 f2fs_msg(sbi->sb, KERN_NOTICE,
1286 "checkpoint: version = %llx", ckpt_ver);
1288 /* do checkpoint periodically */
1289 f2fs_update_time(sbi, CP_TIME);
1290 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1292 mutex_unlock(&sbi->cp_mutex);
1296 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1300 for (i = 0; i < MAX_INO_ENTRY; i++) {
1301 struct inode_management *im = &sbi->im[i];
1303 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1304 spin_lock_init(&im->ino_lock);
1305 INIT_LIST_HEAD(&im->ino_list);
1309 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1310 NR_CURSEG_TYPE - __cp_payload(sbi)) *
1311 F2FS_ORPHANS_PER_BLOCK;
1314 int __init create_checkpoint_caches(void)
1316 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1317 sizeof(struct ino_entry));
1318 if (!ino_entry_slab)
1320 inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1321 sizeof(struct inode_entry));
1322 if (!inode_entry_slab) {
1323 kmem_cache_destroy(ino_entry_slab);
1329 void destroy_checkpoint_caches(void)
1331 kmem_cache_destroy(ino_entry_slab);
1332 kmem_cache_destroy(inode_entry_slab);