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->ckpt, 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 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 = READ_SYNC | 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 ? (READ_SYNC | REQ_META | REQ_PRIO) : REQ_RAHEAD,
163 .encrypted_page = NULL,
165 struct blk_plug plug;
167 if (unlikely(type == META_POR))
168 fio.op_flags &= ~REQ_META;
170 blk_start_plug(&plug);
171 for (; nrpages-- > 0; blkno++) {
173 if (!is_valid_blkaddr(sbi, blkno, type))
178 if (unlikely(blkno >=
179 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
181 /* get nat block addr */
182 fio.new_blkaddr = current_nat_addr(sbi,
183 blkno * NAT_ENTRY_PER_BLOCK);
186 /* get sit block addr */
187 fio.new_blkaddr = current_sit_addr(sbi,
188 blkno * SIT_ENTRY_PER_BLOCK);
193 fio.new_blkaddr = blkno;
199 page = f2fs_grab_cache_page(META_MAPPING(sbi),
200 fio.new_blkaddr, false);
203 if (PageUptodate(page)) {
204 f2fs_put_page(page, 1);
209 fio.old_blkaddr = fio.new_blkaddr;
210 f2fs_submit_page_mbio(&fio);
211 f2fs_put_page(page, 0);
214 f2fs_submit_merged_bio(sbi, META, READ);
215 blk_finish_plug(&plug);
216 return blkno - start;
219 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
222 bool readahead = false;
224 page = find_get_page(META_MAPPING(sbi), index);
225 if (!page || !PageUptodate(page))
227 f2fs_put_page(page, 0);
230 ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR, true);
233 static int f2fs_write_meta_page(struct page *page,
234 struct writeback_control *wbc)
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);
248 dec_page_count(sbi, F2FS_DIRTY_META);
250 if (wbc->for_reclaim)
251 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, META, WRITE);
255 if (unlikely(f2fs_cp_error(sbi)))
256 f2fs_submit_merged_bio(sbi, META, WRITE);
261 redirty_page_for_writepage(wbc, page);
262 return AOP_WRITEPAGE_ACTIVATE;
265 static int f2fs_write_meta_pages(struct address_space *mapping,
266 struct writeback_control *wbc)
268 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
271 /* collect a number of dirty meta pages and write together */
272 if (wbc->for_kupdate ||
273 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
276 trace_f2fs_writepages(mapping->host, wbc, META);
278 /* if mounting is failed, skip writing node pages */
279 mutex_lock(&sbi->cp_mutex);
280 diff = nr_pages_to_write(sbi, META, wbc);
281 written = sync_meta_pages(sbi, META, wbc->nr_to_write);
282 mutex_unlock(&sbi->cp_mutex);
283 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
287 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
288 trace_f2fs_writepages(mapping->host, wbc, META);
292 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
295 struct address_space *mapping = META_MAPPING(sbi);
296 pgoff_t index = 0, end = ULONG_MAX, prev = ULONG_MAX;
299 struct writeback_control wbc = {
302 struct blk_plug plug;
304 pagevec_init(&pvec, 0);
306 blk_start_plug(&plug);
308 while (index <= end) {
310 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
312 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
313 if (unlikely(nr_pages == 0))
316 for (i = 0; i < nr_pages; i++) {
317 struct page *page = pvec.pages[i];
319 if (prev == ULONG_MAX)
320 prev = page->index - 1;
321 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
322 pagevec_release(&pvec);
328 if (unlikely(page->mapping != mapping)) {
333 if (!PageDirty(page)) {
334 /* someone wrote it for us */
335 goto continue_unlock;
338 f2fs_wait_on_page_writeback(page, META, true);
340 BUG_ON(PageWriteback(page));
341 if (!clear_page_dirty_for_io(page))
342 goto continue_unlock;
344 if (mapping->a_ops->writepage(page, &wbc)) {
350 if (unlikely(nwritten >= nr_to_write))
353 pagevec_release(&pvec);
358 f2fs_submit_merged_bio(sbi, type, WRITE);
360 blk_finish_plug(&plug);
365 static int f2fs_set_meta_page_dirty(struct page *page)
367 trace_f2fs_set_page_dirty(page, META);
369 SetPageUptodate(page);
370 if (!PageDirty(page)) {
371 __set_page_dirty_nobuffers(page);
372 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
373 SetPagePrivate(page);
374 f2fs_trace_pid(page);
380 const struct address_space_operations f2fs_meta_aops = {
381 .writepage = f2fs_write_meta_page,
382 .writepages = f2fs_write_meta_pages,
383 .set_page_dirty = f2fs_set_meta_page_dirty,
384 .invalidatepage = f2fs_invalidate_page,
385 .releasepage = f2fs_release_page,
388 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
390 struct inode_management *im = &sbi->im[type];
391 struct ino_entry *e, *tmp;
393 tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
395 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
397 spin_lock(&im->ino_lock);
398 e = radix_tree_lookup(&im->ino_root, ino);
401 if (radix_tree_insert(&im->ino_root, ino, e)) {
402 spin_unlock(&im->ino_lock);
403 radix_tree_preload_end();
406 memset(e, 0, sizeof(struct ino_entry));
409 list_add_tail(&e->list, &im->ino_list);
410 if (type != ORPHAN_INO)
413 spin_unlock(&im->ino_lock);
414 radix_tree_preload_end();
417 kmem_cache_free(ino_entry_slab, tmp);
420 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
422 struct inode_management *im = &sbi->im[type];
425 spin_lock(&im->ino_lock);
426 e = radix_tree_lookup(&im->ino_root, ino);
429 radix_tree_delete(&im->ino_root, ino);
431 spin_unlock(&im->ino_lock);
432 kmem_cache_free(ino_entry_slab, e);
435 spin_unlock(&im->ino_lock);
438 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
440 /* add new dirty ino entry into list */
441 __add_ino_entry(sbi, ino, type);
444 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
446 /* remove dirty ino entry from list */
447 __remove_ino_entry(sbi, ino, type);
450 /* mode should be APPEND_INO or UPDATE_INO */
451 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
453 struct inode_management *im = &sbi->im[mode];
456 spin_lock(&im->ino_lock);
457 e = radix_tree_lookup(&im->ino_root, ino);
458 spin_unlock(&im->ino_lock);
459 return e ? true : false;
462 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
464 struct ino_entry *e, *tmp;
467 for (i = all ? ORPHAN_INO: APPEND_INO; i <= UPDATE_INO; i++) {
468 struct inode_management *im = &sbi->im[i];
470 spin_lock(&im->ino_lock);
471 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
473 radix_tree_delete(&im->ino_root, e->ino);
474 kmem_cache_free(ino_entry_slab, e);
477 spin_unlock(&im->ino_lock);
481 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
483 struct inode_management *im = &sbi->im[ORPHAN_INO];
486 spin_lock(&im->ino_lock);
488 #ifdef CONFIG_F2FS_FAULT_INJECTION
489 if (time_to_inject(FAULT_ORPHAN)) {
490 spin_unlock(&im->ino_lock);
494 if (unlikely(im->ino_num >= sbi->max_orphans))
498 spin_unlock(&im->ino_lock);
503 void release_orphan_inode(struct f2fs_sb_info *sbi)
505 struct inode_management *im = &sbi->im[ORPHAN_INO];
507 spin_lock(&im->ino_lock);
508 f2fs_bug_on(sbi, im->ino_num == 0);
510 spin_unlock(&im->ino_lock);
513 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
515 /* add new orphan ino entry into list */
516 __add_ino_entry(sbi, ino, ORPHAN_INO);
519 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
521 /* remove orphan entry from orphan list */
522 __remove_ino_entry(sbi, ino, ORPHAN_INO);
525 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
529 inode = f2fs_iget(sbi->sb, ino);
532 * there should be a bug that we can't find the entry
535 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
536 return PTR_ERR(inode);
541 /* truncate all the data during iput */
546 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
548 block_t start_blk, orphan_blocks, i, j;
551 if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
554 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
555 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
557 ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
559 for (i = 0; i < orphan_blocks; i++) {
560 struct page *page = get_meta_page(sbi, start_blk + i);
561 struct f2fs_orphan_block *orphan_blk;
563 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
564 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
565 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
566 err = recover_orphan_inode(sbi, ino);
568 f2fs_put_page(page, 1);
572 f2fs_put_page(page, 1);
574 /* clear Orphan Flag */
575 clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
579 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
581 struct list_head *head;
582 struct f2fs_orphan_block *orphan_blk = NULL;
583 unsigned int nentries = 0;
584 unsigned short index = 1;
585 unsigned short orphan_blocks;
586 struct page *page = NULL;
587 struct ino_entry *orphan = NULL;
588 struct inode_management *im = &sbi->im[ORPHAN_INO];
590 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
593 * we don't need to do spin_lock(&im->ino_lock) here, since all the
594 * orphan inode operations are covered under f2fs_lock_op().
595 * And, spin_lock should be avoided due to page operations below.
597 head = &im->ino_list;
599 /* loop for each orphan inode entry and write them in Jornal block */
600 list_for_each_entry(orphan, head, list) {
602 page = grab_meta_page(sbi, start_blk++);
604 (struct f2fs_orphan_block *)page_address(page);
605 memset(orphan_blk, 0, sizeof(*orphan_blk));
608 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
610 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
612 * an orphan block is full of 1020 entries,
613 * then we need to flush current orphan blocks
614 * and bring another one in memory
616 orphan_blk->blk_addr = cpu_to_le16(index);
617 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
618 orphan_blk->entry_count = cpu_to_le32(nentries);
619 set_page_dirty(page);
620 f2fs_put_page(page, 1);
628 orphan_blk->blk_addr = cpu_to_le16(index);
629 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
630 orphan_blk->entry_count = cpu_to_le32(nentries);
631 set_page_dirty(page);
632 f2fs_put_page(page, 1);
636 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
637 block_t cp_addr, unsigned long long *version)
639 struct page *cp_page_1, *cp_page_2 = NULL;
640 unsigned long blk_size = sbi->blocksize;
641 struct f2fs_checkpoint *cp_block;
642 unsigned long long cur_version = 0, pre_version = 0;
646 /* Read the 1st cp block in this CP pack */
647 cp_page_1 = get_meta_page(sbi, cp_addr);
649 /* get the version number */
650 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
651 crc_offset = le32_to_cpu(cp_block->checksum_offset);
652 if (crc_offset >= blk_size)
655 crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
656 if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
659 pre_version = cur_cp_version(cp_block);
661 /* Read the 2nd cp block in this CP pack */
662 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
663 cp_page_2 = get_meta_page(sbi, cp_addr);
665 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
666 crc_offset = le32_to_cpu(cp_block->checksum_offset);
667 if (crc_offset >= blk_size)
670 crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
671 if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
674 cur_version = cur_cp_version(cp_block);
676 if (cur_version == pre_version) {
677 *version = cur_version;
678 f2fs_put_page(cp_page_2, 1);
682 f2fs_put_page(cp_page_2, 1);
684 f2fs_put_page(cp_page_1, 1);
688 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
690 struct f2fs_checkpoint *cp_block;
691 struct f2fs_super_block *fsb = sbi->raw_super;
692 struct page *cp1, *cp2, *cur_page;
693 unsigned long blk_size = sbi->blocksize;
694 unsigned long long cp1_version = 0, cp2_version = 0;
695 unsigned long long cp_start_blk_no;
696 unsigned int cp_blks = 1 + __cp_payload(sbi);
700 sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
704 * Finding out valid cp block involves read both
705 * sets( cp pack1 and cp pack 2)
707 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
708 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
710 /* The second checkpoint pack should start at the next segment */
711 cp_start_blk_no += ((unsigned long long)1) <<
712 le32_to_cpu(fsb->log_blocks_per_seg);
713 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
716 if (ver_after(cp2_version, cp1_version))
728 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
729 memcpy(sbi->ckpt, cp_block, blk_size);
731 /* Sanity checking of checkpoint */
732 if (sanity_check_ckpt(sbi))
738 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
740 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
742 for (i = 1; i < cp_blks; i++) {
743 void *sit_bitmap_ptr;
744 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
746 cur_page = get_meta_page(sbi, cp_blk_no + i);
747 sit_bitmap_ptr = page_address(cur_page);
748 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
749 f2fs_put_page(cur_page, 1);
752 f2fs_put_page(cp1, 1);
753 f2fs_put_page(cp2, 1);
761 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
763 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
764 struct f2fs_inode_info *fi = F2FS_I(inode);
765 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
767 if (is_inode_flag_set(fi, flag))
770 set_inode_flag(fi, flag);
771 list_add_tail(&fi->dirty_list, &sbi->inode_list[type]);
772 stat_inc_dirty_inode(sbi, type);
775 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
777 struct f2fs_inode_info *fi = F2FS_I(inode);
778 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
780 if (get_dirty_pages(inode) ||
781 !is_inode_flag_set(F2FS_I(inode), flag))
784 list_del_init(&fi->dirty_list);
785 clear_inode_flag(fi, flag);
786 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
789 void update_dirty_page(struct inode *inode, struct page *page)
791 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
792 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
794 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
795 !S_ISLNK(inode->i_mode))
798 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH)) {
799 spin_lock(&sbi->inode_lock[type]);
800 __add_dirty_inode(inode, type);
801 spin_unlock(&sbi->inode_lock[type]);
804 inode_inc_dirty_pages(inode);
805 SetPagePrivate(page);
806 f2fs_trace_pid(page);
809 void remove_dirty_inode(struct inode *inode)
811 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
812 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
814 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
815 !S_ISLNK(inode->i_mode))
818 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
821 spin_lock(&sbi->inode_lock[type]);
822 __remove_dirty_inode(inode, type);
823 spin_unlock(&sbi->inode_lock[type]);
826 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
828 struct list_head *head;
830 struct f2fs_inode_info *fi;
831 bool is_dir = (type == DIR_INODE);
833 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
834 get_pages(sbi, is_dir ?
835 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
837 if (unlikely(f2fs_cp_error(sbi)))
840 spin_lock(&sbi->inode_lock[type]);
842 head = &sbi->inode_list[type];
843 if (list_empty(head)) {
844 spin_unlock(&sbi->inode_lock[type]);
845 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
846 get_pages(sbi, is_dir ?
847 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
850 fi = list_entry(head->next, struct f2fs_inode_info, dirty_list);
851 inode = igrab(&fi->vfs_inode);
852 spin_unlock(&sbi->inode_lock[type]);
854 filemap_fdatawrite(inode->i_mapping);
858 * We should submit bio, since it exists several
859 * wribacking dentry pages in the freeing inode.
861 f2fs_submit_merged_bio(sbi, DATA, WRITE);
868 * Freeze all the FS-operations for checkpoint.
870 static int block_operations(struct f2fs_sb_info *sbi)
872 struct writeback_control wbc = {
873 .sync_mode = WB_SYNC_ALL,
874 .nr_to_write = LONG_MAX,
877 struct blk_plug plug;
880 blk_start_plug(&plug);
884 /* write all the dirty dentry pages */
885 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
886 f2fs_unlock_all(sbi);
887 err = sync_dirty_inodes(sbi, DIR_INODE);
890 goto retry_flush_dents;
894 * POR: we should ensure that there are no dirty node pages
895 * until finishing nat/sit flush.
898 down_write(&sbi->node_write);
900 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
901 up_write(&sbi->node_write);
902 err = sync_node_pages(sbi, &wbc);
904 f2fs_unlock_all(sbi);
907 goto retry_flush_nodes;
910 blk_finish_plug(&plug);
914 static void unblock_operations(struct f2fs_sb_info *sbi)
916 up_write(&sbi->node_write);
917 f2fs_unlock_all(sbi);
920 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
925 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
927 if (!atomic_read(&sbi->nr_wb_bios))
930 io_schedule_timeout(5*HZ);
932 finish_wait(&sbi->cp_wait, &wait);
935 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
937 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
938 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
939 struct f2fs_nm_info *nm_i = NM_I(sbi);
940 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
941 nid_t last_nid = nm_i->next_scan_nid;
943 unsigned int data_sum_blocks, orphan_blocks;
946 int cp_payload_blks = __cp_payload(sbi);
947 block_t discard_blk = NEXT_FREE_BLKADDR(sbi, curseg);
948 bool invalidate = false;
949 struct super_block *sb = sbi->sb;
950 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
954 * This avoids to conduct wrong roll-forward operations and uses
955 * metapages, so should be called prior to sync_meta_pages below.
957 if (discard_next_dnode(sbi, discard_blk))
960 /* Flush all the NAT/SIT pages */
961 while (get_pages(sbi, F2FS_DIRTY_META)) {
962 sync_meta_pages(sbi, META, LONG_MAX);
963 if (unlikely(f2fs_cp_error(sbi)))
967 next_free_nid(sbi, &last_nid);
971 * version number is already updated
973 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
974 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
975 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
976 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
977 ckpt->cur_node_segno[i] =
978 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
979 ckpt->cur_node_blkoff[i] =
980 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
981 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
982 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
984 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
985 ckpt->cur_data_segno[i] =
986 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
987 ckpt->cur_data_blkoff[i] =
988 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
989 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
990 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
993 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
994 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
995 ckpt->next_free_nid = cpu_to_le32(last_nid);
997 /* 2 cp + n data seg summary + orphan inode blocks */
998 data_sum_blocks = npages_for_summary_flush(sbi, false);
999 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1000 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1002 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1004 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1005 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1008 if (__remain_node_summaries(cpc->reason))
1009 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1010 cp_payload_blks + data_sum_blocks +
1011 orphan_blocks + NR_CURSEG_NODE_TYPE);
1013 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1014 cp_payload_blks + data_sum_blocks +
1017 if (cpc->reason == CP_UMOUNT)
1018 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1020 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1022 if (cpc->reason == CP_FASTBOOT)
1023 set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1025 clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1028 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1030 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1032 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1033 set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1035 /* update SIT/NAT bitmap */
1036 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1037 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1039 crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1040 *((__le32 *)((unsigned char *)ckpt +
1041 le32_to_cpu(ckpt->checksum_offset)))
1042 = cpu_to_le32(crc32);
1044 start_blk = __start_cp_addr(sbi);
1046 /* need to wait for end_io results */
1047 wait_on_all_pages_writeback(sbi);
1048 if (unlikely(f2fs_cp_error(sbi)))
1051 /* write out checkpoint buffer at block 0 */
1052 update_meta_page(sbi, ckpt, start_blk++);
1054 for (i = 1; i < 1 + cp_payload_blks; i++)
1055 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1059 write_orphan_inodes(sbi, start_blk);
1060 start_blk += orphan_blocks;
1063 write_data_summaries(sbi, start_blk);
1064 start_blk += data_sum_blocks;
1066 /* Record write statistics in the hot node summary */
1067 kbytes_written = sbi->kbytes_written;
1068 if (sb->s_bdev->bd_part)
1069 kbytes_written += BD_PART_WRITTEN(sbi);
1071 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1073 if (__remain_node_summaries(cpc->reason)) {
1074 write_node_summaries(sbi, start_blk);
1075 start_blk += NR_CURSEG_NODE_TYPE;
1078 /* writeout checkpoint block */
1079 update_meta_page(sbi, ckpt, start_blk);
1081 /* wait for previous submitted node/meta pages writeback */
1082 wait_on_all_pages_writeback(sbi);
1084 if (unlikely(f2fs_cp_error(sbi)))
1087 filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1088 filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1090 /* update user_block_counts */
1091 sbi->last_valid_block_count = sbi->total_valid_block_count;
1092 percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1094 /* Here, we only have one bio having CP pack */
1095 sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1097 /* wait for previous submitted meta pages writeback */
1098 wait_on_all_pages_writeback(sbi);
1101 * invalidate meta page which is used temporarily for zeroing out
1102 * block at the end of warm node chain.
1105 invalidate_mapping_pages(META_MAPPING(sbi), discard_blk,
1108 release_ino_entry(sbi, false);
1110 if (unlikely(f2fs_cp_error(sbi)))
1113 clear_prefree_segments(sbi, cpc);
1114 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1120 * We guarantee that this checkpoint procedure will not fail.
1122 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1124 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1125 unsigned long long ckpt_ver;
1128 mutex_lock(&sbi->cp_mutex);
1130 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1131 (cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1132 (cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1134 if (unlikely(f2fs_cp_error(sbi))) {
1138 if (f2fs_readonly(sbi->sb)) {
1143 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1145 err = block_operations(sbi);
1149 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1151 f2fs_flush_merged_bios(sbi);
1154 * update checkpoint pack index
1155 * Increase the version number so that
1156 * SIT entries and seg summaries are written at correct place
1158 ckpt_ver = cur_cp_version(ckpt);
1159 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1161 /* write cached NAT/SIT entries to NAT/SIT area */
1162 flush_nat_entries(sbi);
1163 flush_sit_entries(sbi, cpc);
1165 /* unlock all the fs_lock[] in do_checkpoint() */
1166 err = do_checkpoint(sbi, cpc);
1168 unblock_operations(sbi);
1169 stat_inc_cp_count(sbi->stat_info);
1171 if (cpc->reason == CP_RECOVERY)
1172 f2fs_msg(sbi->sb, KERN_NOTICE,
1173 "checkpoint: version = %llx", ckpt_ver);
1175 /* do checkpoint periodically */
1176 f2fs_update_time(sbi, CP_TIME);
1177 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1179 mutex_unlock(&sbi->cp_mutex);
1183 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1187 for (i = 0; i < MAX_INO_ENTRY; i++) {
1188 struct inode_management *im = &sbi->im[i];
1190 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1191 spin_lock_init(&im->ino_lock);
1192 INIT_LIST_HEAD(&im->ino_list);
1196 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1197 NR_CURSEG_TYPE - __cp_payload(sbi)) *
1198 F2FS_ORPHANS_PER_BLOCK;
1201 int __init create_checkpoint_caches(void)
1203 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1204 sizeof(struct ino_entry));
1205 if (!ino_entry_slab)
1207 inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1208 sizeof(struct inode_entry));
1209 if (!inode_entry_slab) {
1210 kmem_cache_destroy(ino_entry_slab);
1216 void destroy_checkpoint_caches(void)
1218 kmem_cache_destroy(ino_entry_slab);
1219 kmem_cache_destroy(inode_entry_slab);