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;
30 * We guarantee no failure on the returned page.
32 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
34 struct address_space *mapping = META_MAPPING(sbi);
35 struct page *page = NULL;
37 page = grab_cache_page(mapping, index);
42 f2fs_wait_on_page_writeback(page, META);
43 SetPageUptodate(page);
48 * We guarantee no failure on the returned page.
50 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
53 struct address_space *mapping = META_MAPPING(sbi);
55 struct f2fs_io_info fio = {
58 .rw = READ_SYNC | REQ_META | REQ_PRIO,
60 .encrypted_page = NULL,
64 if (unlikely(!is_meta))
67 page = grab_cache_page(mapping, index);
72 if (PageUptodate(page))
77 if (f2fs_submit_page_bio(&fio)) {
78 memset(page_address(page), 0, PAGE_SIZE);
79 f2fs_stop_checkpoint(sbi);
85 if (unlikely(page->mapping != mapping)) {
86 f2fs_put_page(page, 1);
91 * if there is any IO error when accessing device, make our filesystem
92 * readonly and make sure do not write checkpoint with non-uptodate
95 if (unlikely(!PageUptodate(page)))
96 f2fs_stop_checkpoint(sbi);
101 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
103 return __get_meta_page(sbi, index, true);
107 struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
109 return __get_meta_page(sbi, index, false);
112 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
113 block_t blkaddr, int type)
119 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
123 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
124 blkaddr < SM_I(sbi)->ssa_blkaddr))
128 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
129 blkaddr < __start_cp_addr(sbi)))
134 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
135 blkaddr < MAIN_BLKADDR(sbi))) {
136 if (type == DATA_GENERIC) {
137 f2fs_msg(sbi->sb, KERN_WARNING,
138 "access invalid blkaddr:%u", blkaddr);
145 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
146 blkaddr >= MAIN_BLKADDR(sbi)))
157 * Readahead CP/NAT/SIT/SSA pages
159 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
162 block_t prev_blk_addr = 0;
164 block_t blkno = start;
165 struct f2fs_io_info fio = {
168 .rw = sync ? (READ_SYNC | REQ_META | REQ_PRIO) : READA,
169 .encrypted_page = NULL,
170 .is_meta = (type != META_POR),
173 if (unlikely(type == META_POR))
176 for (; nrpages-- > 0; blkno++) {
178 if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
183 if (unlikely(blkno >=
184 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
186 /* get nat block addr */
187 fio.blk_addr = current_nat_addr(sbi,
188 blkno * NAT_ENTRY_PER_BLOCK);
191 if (unlikely(blkno >= TOTAL_SEGS(sbi)))
193 /* get sit block addr */
194 fio.blk_addr = current_sit_addr(sbi,
195 blkno * SIT_ENTRY_PER_BLOCK);
196 if (blkno != start && prev_blk_addr + 1 != fio.blk_addr)
198 prev_blk_addr = fio.blk_addr;
203 fio.blk_addr = blkno;
209 page = grab_cache_page(META_MAPPING(sbi), fio.blk_addr);
212 if (PageUptodate(page)) {
213 f2fs_put_page(page, 1);
218 f2fs_submit_page_mbio(&fio);
219 f2fs_put_page(page, 0);
222 f2fs_submit_merged_bio(sbi, META, READ);
223 return blkno - start;
226 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
229 bool readahead = false;
231 page = find_get_page(META_MAPPING(sbi), index);
232 if (!page || (page && !PageUptodate(page)))
234 f2fs_put_page(page, 0);
237 ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR, true);
240 static int f2fs_write_meta_page(struct page *page,
241 struct writeback_control *wbc)
243 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
245 trace_f2fs_writepage(page, META);
247 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
249 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
251 if (unlikely(f2fs_cp_error(sbi)))
254 f2fs_wait_on_page_writeback(page, META);
255 write_meta_page(sbi, page);
256 dec_page_count(sbi, F2FS_DIRTY_META);
259 if (wbc->for_reclaim)
260 f2fs_submit_merged_bio(sbi, META, WRITE);
264 redirty_page_for_writepage(wbc, page);
265 return AOP_WRITEPAGE_ACTIVATE;
268 static int f2fs_write_meta_pages(struct address_space *mapping,
269 struct writeback_control *wbc)
271 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
274 trace_f2fs_writepages(mapping->host, wbc, META);
276 /* collect a number of dirty meta pages and write together */
277 if (wbc->for_kupdate ||
278 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
281 /* if mounting is failed, skip writing node pages */
282 mutex_lock(&sbi->cp_mutex);
283 diff = nr_pages_to_write(sbi, META, wbc);
284 written = sync_meta_pages(sbi, META, wbc->nr_to_write);
285 mutex_unlock(&sbi->cp_mutex);
286 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
290 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
294 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
297 struct address_space *mapping = META_MAPPING(sbi);
298 pgoff_t index = 0, end = LONG_MAX, prev = LONG_MAX;
301 struct writeback_control wbc = {
305 pagevec_init(&pvec, 0);
307 while (index <= end) {
309 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
311 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
312 if (unlikely(nr_pages == 0))
315 for (i = 0; i < nr_pages; i++) {
316 struct page *page = pvec.pages[i];
318 if (prev == LONG_MAX)
319 prev = page->index - 1;
320 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
321 pagevec_release(&pvec);
327 if (unlikely(page->mapping != mapping)) {
332 if (!PageDirty(page)) {
333 /* someone wrote it for us */
334 goto continue_unlock;
337 if (!clear_page_dirty_for_io(page))
338 goto continue_unlock;
340 if (mapping->a_ops->writepage(page, &wbc)) {
346 if (unlikely(nwritten >= nr_to_write))
349 pagevec_release(&pvec);
354 f2fs_submit_merged_bio(sbi, type, WRITE);
359 static int f2fs_set_meta_page_dirty(struct page *page)
361 trace_f2fs_set_page_dirty(page, META);
363 SetPageUptodate(page);
364 if (!PageDirty(page)) {
365 __set_page_dirty_nobuffers(page);
366 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
367 SetPagePrivate(page);
368 f2fs_trace_pid(page);
374 const struct address_space_operations f2fs_meta_aops = {
375 .writepage = f2fs_write_meta_page,
376 .writepages = f2fs_write_meta_pages,
377 .set_page_dirty = f2fs_set_meta_page_dirty,
378 .invalidatepage = f2fs_invalidate_page,
379 .releasepage = f2fs_release_page,
382 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
384 struct inode_management *im = &sbi->im[type];
385 struct ino_entry *e, *tmp;
387 tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
389 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
391 spin_lock(&im->ino_lock);
392 e = radix_tree_lookup(&im->ino_root, ino);
395 if (radix_tree_insert(&im->ino_root, ino, e)) {
396 spin_unlock(&im->ino_lock);
397 radix_tree_preload_end();
400 memset(e, 0, sizeof(struct ino_entry));
403 list_add_tail(&e->list, &im->ino_list);
404 if (type != ORPHAN_INO)
407 spin_unlock(&im->ino_lock);
408 radix_tree_preload_end();
411 kmem_cache_free(ino_entry_slab, tmp);
414 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
416 struct inode_management *im = &sbi->im[type];
419 spin_lock(&im->ino_lock);
420 e = radix_tree_lookup(&im->ino_root, ino);
423 radix_tree_delete(&im->ino_root, ino);
425 spin_unlock(&im->ino_lock);
426 kmem_cache_free(ino_entry_slab, e);
429 spin_unlock(&im->ino_lock);
432 void add_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
434 /* add new dirty ino entry into list */
435 __add_ino_entry(sbi, ino, type);
438 void remove_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
440 /* remove dirty ino entry from list */
441 __remove_ino_entry(sbi, ino, type);
444 /* mode should be APPEND_INO or UPDATE_INO */
445 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
447 struct inode_management *im = &sbi->im[mode];
450 spin_lock(&im->ino_lock);
451 e = radix_tree_lookup(&im->ino_root, ino);
452 spin_unlock(&im->ino_lock);
453 return e ? true : false;
456 void release_dirty_inode(struct f2fs_sb_info *sbi)
458 struct ino_entry *e, *tmp;
461 for (i = APPEND_INO; i <= UPDATE_INO; i++) {
462 struct inode_management *im = &sbi->im[i];
464 spin_lock(&im->ino_lock);
465 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
467 radix_tree_delete(&im->ino_root, e->ino);
468 kmem_cache_free(ino_entry_slab, e);
471 spin_unlock(&im->ino_lock);
475 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
477 struct inode_management *im = &sbi->im[ORPHAN_INO];
480 spin_lock(&im->ino_lock);
481 if (unlikely(im->ino_num >= sbi->max_orphans))
485 spin_unlock(&im->ino_lock);
490 void release_orphan_inode(struct f2fs_sb_info *sbi)
492 struct inode_management *im = &sbi->im[ORPHAN_INO];
494 spin_lock(&im->ino_lock);
495 f2fs_bug_on(sbi, im->ino_num == 0);
497 spin_unlock(&im->ino_lock);
500 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
502 /* add new orphan ino entry into list */
503 __add_ino_entry(sbi, ino, ORPHAN_INO);
506 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
508 /* remove orphan entry from orphan list */
509 __remove_ino_entry(sbi, ino, ORPHAN_INO);
512 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
516 inode = f2fs_iget(sbi->sb, ino);
519 * there should be a bug that we can't find the entry
522 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
523 return PTR_ERR(inode);
528 /* truncate all the data during iput */
533 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
535 block_t start_blk, orphan_blocks, i, j;
538 if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
541 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
542 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
544 ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
546 for (i = 0; i < orphan_blocks; i++) {
547 struct page *page = get_meta_page(sbi, start_blk + i);
548 struct f2fs_orphan_block *orphan_blk;
550 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
551 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
552 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
553 err = recover_orphan_inode(sbi, ino);
555 f2fs_put_page(page, 1);
559 f2fs_put_page(page, 1);
561 /* clear Orphan Flag */
562 clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
566 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
568 struct list_head *head;
569 struct f2fs_orphan_block *orphan_blk = NULL;
570 unsigned int nentries = 0;
571 unsigned short index = 1;
572 unsigned short orphan_blocks;
573 struct page *page = NULL;
574 struct ino_entry *orphan = NULL;
575 struct inode_management *im = &sbi->im[ORPHAN_INO];
577 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
580 * we don't need to do spin_lock(&im->ino_lock) here, since all the
581 * orphan inode operations are covered under f2fs_lock_op().
582 * And, spin_lock should be avoided due to page operations below.
584 head = &im->ino_list;
586 /* loop for each orphan inode entry and write them in Jornal block */
587 list_for_each_entry(orphan, head, list) {
589 page = grab_meta_page(sbi, start_blk++);
591 (struct f2fs_orphan_block *)page_address(page);
592 memset(orphan_blk, 0, sizeof(*orphan_blk));
595 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
597 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
599 * an orphan block is full of 1020 entries,
600 * then we need to flush current orphan blocks
601 * and bring another one in memory
603 orphan_blk->blk_addr = cpu_to_le16(index);
604 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
605 orphan_blk->entry_count = cpu_to_le32(nentries);
606 set_page_dirty(page);
607 f2fs_put_page(page, 1);
615 orphan_blk->blk_addr = cpu_to_le16(index);
616 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
617 orphan_blk->entry_count = cpu_to_le32(nentries);
618 set_page_dirty(page);
619 f2fs_put_page(page, 1);
623 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
624 struct f2fs_checkpoint **cp_block, struct page **cp_page,
625 unsigned long long *version)
627 unsigned long blk_size = sbi->blocksize;
628 size_t crc_offset = 0;
631 *cp_page = get_meta_page(sbi, cp_addr);
632 *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
634 crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
635 if (crc_offset >= blk_size) {
636 f2fs_put_page(*cp_page, 1);
637 f2fs_msg(sbi->sb, KERN_WARNING,
638 "invalid crc_offset: %zu", crc_offset);
642 crc = le32_to_cpu(*((__le32 *)((unsigned char *)*cp_block
644 if (!f2fs_crc_valid(crc, *cp_block, crc_offset)) {
645 f2fs_put_page(*cp_page, 1);
646 f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
650 *version = cur_cp_version(*cp_block);
654 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
655 block_t cp_addr, unsigned long long *version)
657 struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
658 struct f2fs_checkpoint *cp_block = NULL;
659 unsigned long long cur_version = 0, pre_version = 0;
662 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
663 &cp_page_1, version);
667 if (le32_to_cpu(cp_block->cp_pack_total_block_count) >
668 sbi->blocks_per_seg) {
669 f2fs_msg(sbi->sb, KERN_WARNING,
670 "invalid cp_pack_total_block_count:%u",
671 le32_to_cpu(cp_block->cp_pack_total_block_count));
674 pre_version = *version;
676 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
677 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
678 &cp_page_2, version);
681 cur_version = *version;
683 if (cur_version == pre_version) {
684 *version = cur_version;
685 f2fs_put_page(cp_page_2, 1);
688 f2fs_put_page(cp_page_2, 1);
690 f2fs_put_page(cp_page_1, 1);
694 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
696 struct f2fs_checkpoint *cp_block;
697 struct f2fs_super_block *fsb = sbi->raw_super;
698 struct page *cp1, *cp2, *cur_page;
699 unsigned long blk_size = sbi->blocksize;
700 unsigned long long cp1_version = 0, cp2_version = 0;
701 unsigned long long cp_start_blk_no;
702 unsigned int cp_blks = 1 + __cp_payload(sbi);
706 sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
710 * Finding out valid cp block involves read both
711 * sets( cp pack1 and cp pack 2)
713 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
714 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
716 /* The second checkpoint pack should start at the next segment */
717 cp_start_blk_no += ((unsigned long long)1) <<
718 le32_to_cpu(fsb->log_blocks_per_seg);
719 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
722 if (ver_after(cp2_version, cp1_version))
734 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
735 memcpy(sbi->ckpt, cp_block, blk_size);
738 sbi->cur_cp_pack = 1;
740 sbi->cur_cp_pack = 2;
742 /* Sanity checking of checkpoint */
743 if (sanity_check_ckpt(sbi))
744 goto free_fail_no_cp;
749 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
751 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
753 for (i = 1; i < cp_blks; i++) {
754 void *sit_bitmap_ptr;
755 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
757 cur_page = get_meta_page(sbi, cp_blk_no + i);
758 sit_bitmap_ptr = page_address(cur_page);
759 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
760 f2fs_put_page(cur_page, 1);
763 f2fs_put_page(cp1, 1);
764 f2fs_put_page(cp2, 1);
768 f2fs_put_page(cp1, 1);
769 f2fs_put_page(cp2, 1);
775 static int __add_dirty_inode(struct inode *inode, struct inode_entry *new)
777 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
779 if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
782 set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
783 F2FS_I(inode)->dirty_dir = new;
784 list_add_tail(&new->list, &sbi->dir_inode_list);
785 stat_inc_dirty_dir(sbi);
789 void update_dirty_page(struct inode *inode, struct page *page)
791 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
792 struct inode_entry *new;
795 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
796 !S_ISLNK(inode->i_mode))
799 if (!S_ISDIR(inode->i_mode)) {
800 inode_inc_dirty_pages(inode);
804 new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
806 INIT_LIST_HEAD(&new->list);
808 spin_lock(&sbi->dir_inode_lock);
809 ret = __add_dirty_inode(inode, new);
810 inode_inc_dirty_pages(inode);
811 spin_unlock(&sbi->dir_inode_lock);
814 kmem_cache_free(inode_entry_slab, new);
816 SetPagePrivate(page);
817 f2fs_trace_pid(page);
820 void remove_dirty_dir_inode(struct inode *inode)
822 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
823 struct inode_entry *entry;
825 if (!S_ISDIR(inode->i_mode))
828 spin_lock(&sbi->dir_inode_lock);
829 if (get_dirty_pages(inode) ||
830 !is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
831 spin_unlock(&sbi->dir_inode_lock);
835 entry = F2FS_I(inode)->dirty_dir;
836 list_del(&entry->list);
837 F2FS_I(inode)->dirty_dir = NULL;
838 clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
839 stat_dec_dirty_dir(sbi);
840 spin_unlock(&sbi->dir_inode_lock);
841 kmem_cache_free(inode_entry_slab, entry);
844 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
846 struct list_head *head;
847 struct inode_entry *entry;
850 if (unlikely(f2fs_cp_error(sbi)))
853 spin_lock(&sbi->dir_inode_lock);
855 head = &sbi->dir_inode_list;
856 if (list_empty(head)) {
857 spin_unlock(&sbi->dir_inode_lock);
860 entry = list_entry(head->next, struct inode_entry, list);
861 inode = igrab(entry->inode);
862 spin_unlock(&sbi->dir_inode_lock);
864 filemap_fdatawrite(inode->i_mapping);
868 * We should submit bio, since it exists several
869 * wribacking dentry pages in the freeing inode.
871 f2fs_submit_merged_bio(sbi, DATA, WRITE);
878 * Freeze all the FS-operations for checkpoint.
880 static int block_operations(struct f2fs_sb_info *sbi)
882 struct writeback_control wbc = {
883 .sync_mode = WB_SYNC_ALL,
884 .nr_to_write = LONG_MAX,
887 struct blk_plug plug;
890 blk_start_plug(&plug);
894 /* write all the dirty dentry pages */
895 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
896 f2fs_unlock_all(sbi);
897 sync_dirty_dir_inodes(sbi);
898 if (unlikely(f2fs_cp_error(sbi))) {
902 goto retry_flush_dents;
906 * POR: we should ensure that there are no dirty node pages
907 * until finishing nat/sit flush.
910 down_write(&sbi->node_write);
912 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
913 up_write(&sbi->node_write);
914 sync_node_pages(sbi, 0, &wbc);
915 if (unlikely(f2fs_cp_error(sbi))) {
916 f2fs_unlock_all(sbi);
920 goto retry_flush_nodes;
923 blk_finish_plug(&plug);
927 static void unblock_operations(struct f2fs_sb_info *sbi)
929 up_write(&sbi->node_write);
930 f2fs_unlock_all(sbi);
933 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
938 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
940 if (!get_pages(sbi, F2FS_WRITEBACK))
945 finish_wait(&sbi->cp_wait, &wait);
948 static void do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
950 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
951 struct f2fs_nm_info *nm_i = NM_I(sbi);
952 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
953 nid_t last_nid = nm_i->next_scan_nid;
955 unsigned int data_sum_blocks, orphan_blocks;
958 int cp_payload_blks = __cp_payload(sbi);
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 /* set this flag to activate crc|cp_ver for recovery */
1036 set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1038 /* update SIT/NAT bitmap */
1039 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1040 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1042 crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
1043 *((__le32 *)((unsigned char *)ckpt +
1044 le32_to_cpu(ckpt->checksum_offset)))
1045 = cpu_to_le32(crc32);
1047 start_blk = __start_cp_next_addr(sbi);
1049 /* need to wait for end_io results */
1050 wait_on_all_pages_writeback(sbi);
1051 if (unlikely(f2fs_cp_error(sbi)))
1054 /* write out checkpoint buffer at block 0 */
1055 update_meta_page(sbi, ckpt, start_blk++);
1057 for (i = 1; i < 1 + cp_payload_blks; i++)
1058 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1062 write_orphan_inodes(sbi, start_blk);
1063 start_blk += orphan_blocks;
1066 write_data_summaries(sbi, start_blk);
1067 start_blk += data_sum_blocks;
1068 if (__remain_node_summaries(cpc->reason)) {
1069 write_node_summaries(sbi, start_blk);
1070 start_blk += NR_CURSEG_NODE_TYPE;
1073 /* writeout checkpoint block */
1074 update_meta_page(sbi, ckpt, start_blk);
1076 /* wait for previous submitted node/meta pages writeback */
1077 wait_on_all_pages_writeback(sbi);
1079 if (unlikely(f2fs_cp_error(sbi)))
1082 filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
1083 filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
1085 /* update user_block_counts */
1086 sbi->last_valid_block_count = sbi->total_valid_block_count;
1087 sbi->alloc_valid_block_count = 0;
1089 /* Here, we only have one bio having CP pack */
1090 sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1092 /* wait for previous submitted meta pages writeback */
1093 wait_on_all_pages_writeback(sbi);
1095 release_dirty_inode(sbi);
1097 if (unlikely(f2fs_cp_error(sbi)))
1100 clear_prefree_segments(sbi, cpc);
1101 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1102 __set_cp_next_pack(sbi);
1106 * We guarantee that this checkpoint procedure will not fail.
1108 void write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1110 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1111 unsigned long long ckpt_ver;
1113 mutex_lock(&sbi->cp_mutex);
1115 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1116 (cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1117 (cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1119 if (unlikely(f2fs_cp_error(sbi)))
1121 if (f2fs_readonly(sbi->sb))
1124 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1126 if (block_operations(sbi))
1129 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1131 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1132 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1133 f2fs_submit_merged_bio(sbi, META, WRITE);
1136 * update checkpoint pack index
1137 * Increase the version number so that
1138 * SIT entries and seg summaries are written at correct place
1140 ckpt_ver = cur_cp_version(ckpt);
1141 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1143 /* write cached NAT/SIT entries to NAT/SIT area */
1144 flush_nat_entries(sbi);
1145 flush_sit_entries(sbi, cpc);
1147 /* unlock all the fs_lock[] in do_checkpoint() */
1148 do_checkpoint(sbi, cpc);
1150 unblock_operations(sbi);
1151 stat_inc_cp_count(sbi->stat_info);
1153 if (cpc->reason == CP_RECOVERY)
1154 f2fs_msg(sbi->sb, KERN_NOTICE,
1155 "checkpoint: version = %llx", ckpt_ver);
1157 /* do checkpoint periodically */
1158 sbi->cp_expires = round_jiffies_up(jiffies + HZ * sbi->cp_interval);
1160 mutex_unlock(&sbi->cp_mutex);
1161 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1164 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1168 for (i = 0; i < MAX_INO_ENTRY; i++) {
1169 struct inode_management *im = &sbi->im[i];
1171 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1172 spin_lock_init(&im->ino_lock);
1173 INIT_LIST_HEAD(&im->ino_list);
1177 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1178 NR_CURSEG_TYPE - __cp_payload(sbi)) *
1179 F2FS_ORPHANS_PER_BLOCK;
1182 int __init create_checkpoint_caches(void)
1184 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1185 sizeof(struct ino_entry));
1186 if (!ino_entry_slab)
1188 inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1189 sizeof(struct inode_entry));
1190 if (!inode_entry_slab) {
1191 kmem_cache_destroy(ino_entry_slab);
1197 void destroy_checkpoint_caches(void)
1199 kmem_cache_destroy(ino_entry_slab);
1200 kmem_cache_destroy(inode_entry_slab);