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_writes(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,
167 struct blk_plug plug;
169 if (unlikely(type == META_POR))
170 fio.op_flags &= ~REQ_META;
172 blk_start_plug(&plug);
173 for (; nrpages-- > 0; blkno++) {
175 if (!is_valid_blkaddr(sbi, blkno, type))
180 if (unlikely(blkno >=
181 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
183 /* get nat block addr */
184 fio.new_blkaddr = current_nat_addr(sbi,
185 blkno * NAT_ENTRY_PER_BLOCK);
188 /* get sit block addr */
189 fio.new_blkaddr = current_sit_addr(sbi,
190 blkno * SIT_ENTRY_PER_BLOCK);
195 fio.new_blkaddr = blkno;
201 page = f2fs_grab_cache_page(META_MAPPING(sbi),
202 fio.new_blkaddr, false);
205 if (PageUptodate(page)) {
206 f2fs_put_page(page, 1);
211 f2fs_submit_page_bio(&fio);
212 f2fs_put_page(page, 0);
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, BIO_MAX_PAGES, META_POR, true);
233 static int __f2fs_write_meta_page(struct page *page,
234 struct writeback_control *wbc,
235 enum iostat_type io_type)
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, io_type);
249 dec_page_count(sbi, F2FS_DIRTY_META);
251 if (wbc->for_reclaim)
252 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
253 0, page->index, META);
257 if (unlikely(f2fs_cp_error(sbi)))
258 f2fs_submit_merged_write(sbi, META);
263 redirty_page_for_writepage(wbc, page);
264 return AOP_WRITEPAGE_ACTIVATE;
267 static int f2fs_write_meta_page(struct page *page,
268 struct writeback_control *wbc)
270 return __f2fs_write_meta_page(page, wbc, FS_META_IO);
273 static int f2fs_write_meta_pages(struct address_space *mapping,
274 struct writeback_control *wbc)
276 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
279 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
282 /* collect a number of dirty meta pages and write together */
283 if (wbc->for_kupdate ||
284 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
287 /* if locked failed, cp will flush dirty pages instead */
288 if (!mutex_trylock(&sbi->cp_mutex))
291 trace_f2fs_writepages(mapping->host, wbc, META);
292 diff = nr_pages_to_write(sbi, META, wbc);
293 written = sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
294 mutex_unlock(&sbi->cp_mutex);
295 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
299 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
300 trace_f2fs_writepages(mapping->host, wbc, META);
304 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
305 long nr_to_write, enum iostat_type io_type)
307 struct address_space *mapping = META_MAPPING(sbi);
308 pgoff_t index = 0, prev = ULONG_MAX;
312 struct writeback_control wbc = {
315 struct blk_plug plug;
319 blk_start_plug(&plug);
321 while ((nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
322 PAGECACHE_TAG_DIRTY))) {
325 for (i = 0; i < nr_pages; i++) {
326 struct page *page = pvec.pages[i];
328 if (prev == ULONG_MAX)
329 prev = page->index - 1;
330 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
331 pagevec_release(&pvec);
337 if (unlikely(page->mapping != mapping)) {
342 if (!PageDirty(page)) {
343 /* someone wrote it for us */
344 goto continue_unlock;
347 f2fs_wait_on_page_writeback(page, META, true);
349 BUG_ON(PageWriteback(page));
350 if (!clear_page_dirty_for_io(page))
351 goto continue_unlock;
353 if (__f2fs_write_meta_page(page, &wbc, io_type)) {
359 if (unlikely(nwritten >= nr_to_write))
362 pagevec_release(&pvec);
367 f2fs_submit_merged_write(sbi, type);
369 blk_finish_plug(&plug);
374 static int f2fs_set_meta_page_dirty(struct page *page)
376 trace_f2fs_set_page_dirty(page, META);
378 if (!PageUptodate(page))
379 SetPageUptodate(page);
380 if (!PageDirty(page)) {
381 f2fs_set_page_dirty_nobuffers(page);
382 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
383 SetPagePrivate(page);
384 f2fs_trace_pid(page);
390 const struct address_space_operations f2fs_meta_aops = {
391 .writepage = f2fs_write_meta_page,
392 .writepages = f2fs_write_meta_pages,
393 .set_page_dirty = f2fs_set_meta_page_dirty,
394 .invalidatepage = f2fs_invalidate_page,
395 .releasepage = f2fs_release_page,
396 #ifdef CONFIG_MIGRATION
397 .migratepage = f2fs_migrate_page,
401 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, 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 (radix_tree_insert(&im->ino_root, ino, e)) {
415 spin_unlock(&im->ino_lock);
416 radix_tree_preload_end();
419 memset(e, 0, sizeof(struct ino_entry));
422 list_add_tail(&e->list, &im->ino_list);
423 if (type != ORPHAN_INO)
426 spin_unlock(&im->ino_lock);
427 radix_tree_preload_end();
430 kmem_cache_free(ino_entry_slab, tmp);
433 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
435 struct inode_management *im = &sbi->im[type];
438 spin_lock(&im->ino_lock);
439 e = radix_tree_lookup(&im->ino_root, ino);
442 radix_tree_delete(&im->ino_root, ino);
444 spin_unlock(&im->ino_lock);
445 kmem_cache_free(ino_entry_slab, e);
448 spin_unlock(&im->ino_lock);
451 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
453 /* add new dirty ino entry into list */
454 __add_ino_entry(sbi, ino, type);
457 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
459 /* remove dirty ino entry from list */
460 __remove_ino_entry(sbi, ino, type);
463 /* mode should be APPEND_INO or UPDATE_INO */
464 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
466 struct inode_management *im = &sbi->im[mode];
469 spin_lock(&im->ino_lock);
470 e = radix_tree_lookup(&im->ino_root, ino);
471 spin_unlock(&im->ino_lock);
472 return e ? true : false;
475 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
477 struct ino_entry *e, *tmp;
480 for (i = all ? ORPHAN_INO: APPEND_INO; i <= UPDATE_INO; i++) {
481 struct inode_management *im = &sbi->im[i];
483 spin_lock(&im->ino_lock);
484 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
486 radix_tree_delete(&im->ino_root, e->ino);
487 kmem_cache_free(ino_entry_slab, e);
490 spin_unlock(&im->ino_lock);
494 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
496 struct inode_management *im = &sbi->im[ORPHAN_INO];
499 spin_lock(&im->ino_lock);
501 #ifdef CONFIG_F2FS_FAULT_INJECTION
502 if (time_to_inject(sbi, FAULT_ORPHAN)) {
503 spin_unlock(&im->ino_lock);
504 f2fs_show_injection_info(FAULT_ORPHAN);
508 if (unlikely(im->ino_num >= sbi->max_orphans))
512 spin_unlock(&im->ino_lock);
517 void release_orphan_inode(struct f2fs_sb_info *sbi)
519 struct inode_management *im = &sbi->im[ORPHAN_INO];
521 spin_lock(&im->ino_lock);
522 f2fs_bug_on(sbi, im->ino_num == 0);
524 spin_unlock(&im->ino_lock);
527 void add_orphan_inode(struct inode *inode)
529 /* add new orphan ino entry into list */
530 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, ORPHAN_INO);
531 update_inode_page(inode);
534 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
536 /* remove orphan entry from orphan list */
537 __remove_ino_entry(sbi, ino, ORPHAN_INO);
540 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
544 int err = acquire_orphan_inode(sbi);
547 set_sbi_flag(sbi, SBI_NEED_FSCK);
548 f2fs_msg(sbi->sb, KERN_WARNING,
549 "%s: orphan failed (ino=%x), run fsck to fix.",
554 __add_ino_entry(sbi, ino, ORPHAN_INO);
556 inode = f2fs_iget_retry(sbi->sb, ino);
559 * there should be a bug that we can't find the entry
562 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
563 return PTR_ERR(inode);
568 /* truncate all the data during iput */
571 get_node_info(sbi, ino, &ni);
573 /* ENOMEM was fully retried in f2fs_evict_inode. */
574 if (ni.blk_addr != NULL_ADDR) {
575 set_sbi_flag(sbi, SBI_NEED_FSCK);
576 f2fs_msg(sbi->sb, KERN_WARNING,
577 "%s: orphan failed (ino=%x) by kernel, retry mount.",
581 __remove_ino_entry(sbi, ino, ORPHAN_INO);
585 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
587 block_t start_blk, orphan_blocks, i, j;
588 unsigned int s_flags = sbi->sb->s_flags;
591 if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
594 if (s_flags & MS_RDONLY) {
595 f2fs_msg(sbi->sb, KERN_INFO, "orphan cleanup on readonly fs");
596 sbi->sb->s_flags &= ~MS_RDONLY;
600 /* Needed for iput() to work correctly and not trash data */
601 sbi->sb->s_flags |= MS_ACTIVE;
602 /* Turn on quotas so that they are updated correctly */
603 f2fs_enable_quota_files(sbi);
606 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
607 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
609 ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
611 for (i = 0; i < orphan_blocks; i++) {
612 struct page *page = get_meta_page(sbi, start_blk + i);
613 struct f2fs_orphan_block *orphan_blk;
615 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
616 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
617 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
618 err = recover_orphan_inode(sbi, ino);
620 f2fs_put_page(page, 1);
624 f2fs_put_page(page, 1);
626 /* clear Orphan Flag */
627 clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
630 /* Turn quotas off */
631 f2fs_quota_off_umount(sbi->sb);
633 sbi->sb->s_flags = s_flags; /* Restore MS_RDONLY status */
638 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
640 struct list_head *head;
641 struct f2fs_orphan_block *orphan_blk = NULL;
642 unsigned int nentries = 0;
643 unsigned short index = 1;
644 unsigned short orphan_blocks;
645 struct page *page = NULL;
646 struct ino_entry *orphan = NULL;
647 struct inode_management *im = &sbi->im[ORPHAN_INO];
649 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
652 * we don't need to do spin_lock(&im->ino_lock) here, since all the
653 * orphan inode operations are covered under f2fs_lock_op().
654 * And, spin_lock should be avoided due to page operations below.
656 head = &im->ino_list;
658 /* loop for each orphan inode entry and write them in Jornal block */
659 list_for_each_entry(orphan, head, list) {
661 page = grab_meta_page(sbi, start_blk++);
663 (struct f2fs_orphan_block *)page_address(page);
664 memset(orphan_blk, 0, sizeof(*orphan_blk));
667 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
669 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
671 * an orphan block is full of 1020 entries,
672 * then we need to flush current orphan blocks
673 * and bring another one in memory
675 orphan_blk->blk_addr = cpu_to_le16(index);
676 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
677 orphan_blk->entry_count = cpu_to_le32(nentries);
678 set_page_dirty(page);
679 f2fs_put_page(page, 1);
687 orphan_blk->blk_addr = cpu_to_le16(index);
688 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
689 orphan_blk->entry_count = cpu_to_le32(nentries);
690 set_page_dirty(page);
691 f2fs_put_page(page, 1);
695 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
696 struct f2fs_checkpoint **cp_block, struct page **cp_page,
697 unsigned long long *version)
699 unsigned long blk_size = sbi->blocksize;
700 size_t crc_offset = 0;
703 *cp_page = get_meta_page(sbi, cp_addr);
704 *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
706 crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
707 if (crc_offset > (blk_size - sizeof(__le32))) {
708 f2fs_msg(sbi->sb, KERN_WARNING,
709 "invalid crc_offset: %zu", crc_offset);
713 crc = cur_cp_crc(*cp_block);
714 if (!f2fs_crc_valid(sbi, crc, *cp_block, crc_offset)) {
715 f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
719 *version = cur_cp_version(*cp_block);
723 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
724 block_t cp_addr, unsigned long long *version)
726 struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
727 struct f2fs_checkpoint *cp_block = NULL;
728 unsigned long long cur_version = 0, pre_version = 0;
731 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
732 &cp_page_1, version);
735 pre_version = *version;
737 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
738 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
739 &cp_page_2, version);
742 cur_version = *version;
744 if (cur_version == pre_version) {
745 *version = cur_version;
746 f2fs_put_page(cp_page_2, 1);
750 f2fs_put_page(cp_page_2, 1);
752 f2fs_put_page(cp_page_1, 1);
756 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
758 struct f2fs_checkpoint *cp_block;
759 struct f2fs_super_block *fsb = sbi->raw_super;
760 struct page *cp1, *cp2, *cur_page;
761 unsigned long blk_size = sbi->blocksize;
762 unsigned long long cp1_version = 0, cp2_version = 0;
763 unsigned long long cp_start_blk_no;
764 unsigned int cp_blks = 1 + __cp_payload(sbi);
768 sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
772 * Finding out valid cp block involves read both
773 * sets( cp pack1 and cp pack 2)
775 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
776 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
778 /* The second checkpoint pack should start at the next segment */
779 cp_start_blk_no += ((unsigned long long)1) <<
780 le32_to_cpu(fsb->log_blocks_per_seg);
781 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
784 if (ver_after(cp2_version, cp1_version))
796 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
797 memcpy(sbi->ckpt, cp_block, blk_size);
799 /* Sanity checking of checkpoint */
800 if (sanity_check_ckpt(sbi))
801 goto free_fail_no_cp;
804 sbi->cur_cp_pack = 1;
806 sbi->cur_cp_pack = 2;
811 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
813 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
815 for (i = 1; i < cp_blks; i++) {
816 void *sit_bitmap_ptr;
817 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
819 cur_page = get_meta_page(sbi, cp_blk_no + i);
820 sit_bitmap_ptr = page_address(cur_page);
821 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
822 f2fs_put_page(cur_page, 1);
825 f2fs_put_page(cp1, 1);
826 f2fs_put_page(cp2, 1);
830 f2fs_put_page(cp1, 1);
831 f2fs_put_page(cp2, 1);
837 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
839 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
840 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
842 if (is_inode_flag_set(inode, flag))
845 set_inode_flag(inode, flag);
846 if (!f2fs_is_volatile_file(inode))
847 list_add_tail(&F2FS_I(inode)->dirty_list,
848 &sbi->inode_list[type]);
849 stat_inc_dirty_inode(sbi, type);
852 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
854 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
856 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
859 list_del_init(&F2FS_I(inode)->dirty_list);
860 clear_inode_flag(inode, flag);
861 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
864 void update_dirty_page(struct inode *inode, struct page *page)
866 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
867 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
869 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
870 !S_ISLNK(inode->i_mode))
873 spin_lock(&sbi->inode_lock[type]);
874 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
875 __add_dirty_inode(inode, type);
876 inode_inc_dirty_pages(inode);
877 spin_unlock(&sbi->inode_lock[type]);
879 SetPagePrivate(page);
880 f2fs_trace_pid(page);
883 void remove_dirty_inode(struct inode *inode)
885 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
886 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
888 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
889 !S_ISLNK(inode->i_mode))
892 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
895 spin_lock(&sbi->inode_lock[type]);
896 __remove_dirty_inode(inode, type);
897 spin_unlock(&sbi->inode_lock[type]);
900 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
902 struct list_head *head;
904 struct f2fs_inode_info *fi;
905 bool is_dir = (type == DIR_INODE);
906 unsigned long ino = 0;
908 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
909 get_pages(sbi, is_dir ?
910 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
912 if (unlikely(f2fs_cp_error(sbi)))
915 spin_lock(&sbi->inode_lock[type]);
917 head = &sbi->inode_list[type];
918 if (list_empty(head)) {
919 spin_unlock(&sbi->inode_lock[type]);
920 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
921 get_pages(sbi, is_dir ?
922 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
925 fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
926 inode = igrab(&fi->vfs_inode);
927 spin_unlock(&sbi->inode_lock[type]);
929 unsigned long cur_ino = inode->i_ino;
932 F2FS_I(inode)->cp_task = current;
934 filemap_fdatawrite(inode->i_mapping);
937 F2FS_I(inode)->cp_task = NULL;
940 /* We need to give cpu to another writers. */
941 if (ino == cur_ino) {
942 congestion_wait(BLK_RW_ASYNC, HZ/50);
949 * We should submit bio, since it exists several
950 * wribacking dentry pages in the freeing inode.
952 f2fs_submit_merged_write(sbi, DATA);
958 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
960 struct list_head *head = &sbi->inode_list[DIRTY_META];
962 struct f2fs_inode_info *fi;
963 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
966 if (unlikely(f2fs_cp_error(sbi)))
969 spin_lock(&sbi->inode_lock[DIRTY_META]);
970 if (list_empty(head)) {
971 spin_unlock(&sbi->inode_lock[DIRTY_META]);
974 fi = list_first_entry(head, struct f2fs_inode_info,
976 inode = igrab(&fi->vfs_inode);
977 spin_unlock(&sbi->inode_lock[DIRTY_META]);
979 sync_inode_metadata(inode, 0);
981 /* it's on eviction */
982 if (is_inode_flag_set(inode, FI_DIRTY_INODE))
983 update_inode_page(inode);
990 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
992 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
993 struct f2fs_nm_info *nm_i = NM_I(sbi);
994 nid_t last_nid = nm_i->next_scan_nid;
996 next_free_nid(sbi, &last_nid);
997 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
998 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
999 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1000 ckpt->next_free_nid = cpu_to_le32(last_nid);
1004 * Freeze all the FS-operations for checkpoint.
1006 static int block_operations(struct f2fs_sb_info *sbi)
1008 struct writeback_control wbc = {
1009 .sync_mode = WB_SYNC_ALL,
1010 .nr_to_write = LONG_MAX,
1013 struct blk_plug plug;
1016 blk_start_plug(&plug);
1020 /* write all the dirty dentry pages */
1021 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1022 f2fs_unlock_all(sbi);
1023 err = sync_dirty_inodes(sbi, DIR_INODE);
1027 goto retry_flush_dents;
1031 * POR: we should ensure that there are no dirty node pages
1032 * until finishing nat/sit flush. inode->i_blocks can be updated.
1034 down_write(&sbi->node_change);
1036 if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1037 up_write(&sbi->node_change);
1038 f2fs_unlock_all(sbi);
1039 err = f2fs_sync_inode_meta(sbi);
1043 goto retry_flush_dents;
1047 down_write(&sbi->node_write);
1049 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1050 up_write(&sbi->node_write);
1051 err = sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1053 up_write(&sbi->node_change);
1054 f2fs_unlock_all(sbi);
1058 goto retry_flush_nodes;
1062 * sbi->node_change is used only for AIO write_begin path which produces
1063 * dirty node blocks and some checkpoint values by block allocation.
1065 __prepare_cp_block(sbi);
1066 up_write(&sbi->node_change);
1068 blk_finish_plug(&plug);
1072 static void unblock_operations(struct f2fs_sb_info *sbi)
1074 up_write(&sbi->node_write);
1075 f2fs_unlock_all(sbi);
1078 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
1083 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1085 if (!get_pages(sbi, F2FS_WB_CP_DATA))
1088 io_schedule_timeout(5*HZ);
1090 finish_wait(&sbi->cp_wait, &wait);
1093 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1095 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1096 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1097 unsigned long flags;
1099 spin_lock_irqsave(&sbi->cp_lock, flags);
1101 if ((cpc->reason & CP_UMOUNT) &&
1102 le32_to_cpu(ckpt->cp_pack_total_block_count) >
1103 sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1104 disable_nat_bits(sbi, false);
1106 if (cpc->reason & CP_TRIMMED)
1107 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1109 if (cpc->reason & CP_UMOUNT)
1110 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1112 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1114 if (cpc->reason & CP_FASTBOOT)
1115 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1117 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1120 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1122 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1124 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1125 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1127 /* set this flag to activate crc|cp_ver for recovery */
1128 __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1130 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1133 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1135 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1136 struct f2fs_nm_info *nm_i = NM_I(sbi);
1137 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1139 unsigned int data_sum_blocks, orphan_blocks;
1142 int cp_payload_blks = __cp_payload(sbi);
1143 struct super_block *sb = sbi->sb;
1144 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1147 /* Flush all the NAT/SIT pages */
1148 while (get_pages(sbi, F2FS_DIRTY_META)) {
1149 sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1150 if (unlikely(f2fs_cp_error(sbi)))
1156 * version number is already updated
1158 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
1159 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1160 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1161 ckpt->cur_node_segno[i] =
1162 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1163 ckpt->cur_node_blkoff[i] =
1164 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1165 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1166 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1168 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1169 ckpt->cur_data_segno[i] =
1170 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1171 ckpt->cur_data_blkoff[i] =
1172 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1173 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1174 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1177 /* 2 cp + n data seg summary + orphan inode blocks */
1178 data_sum_blocks = npages_for_summary_flush(sbi, false);
1179 spin_lock_irqsave(&sbi->cp_lock, flags);
1180 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1181 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1183 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1184 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1186 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1187 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1190 if (__remain_node_summaries(cpc->reason))
1191 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1192 cp_payload_blks + data_sum_blocks +
1193 orphan_blocks + NR_CURSEG_NODE_TYPE);
1195 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1196 cp_payload_blks + data_sum_blocks +
1199 /* update ckpt flag for checkpoint */
1200 update_ckpt_flags(sbi, cpc);
1202 /* update SIT/NAT bitmap */
1203 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1204 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1206 crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1207 *((__le32 *)((unsigned char *)ckpt +
1208 le32_to_cpu(ckpt->checksum_offset)))
1209 = cpu_to_le32(crc32);
1211 start_blk = __start_cp_next_addr(sbi);
1213 /* write nat bits */
1214 if (enabled_nat_bits(sbi, cpc)) {
1215 __u64 cp_ver = cur_cp_version(ckpt);
1218 cp_ver |= ((__u64)crc32 << 32);
1219 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1221 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1222 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1223 update_meta_page(sbi, nm_i->nat_bits +
1224 (i << F2FS_BLKSIZE_BITS), blk + i);
1226 /* Flush all the NAT BITS pages */
1227 while (get_pages(sbi, F2FS_DIRTY_META)) {
1228 sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1229 if (unlikely(f2fs_cp_error(sbi)))
1234 /* need to wait for end_io results */
1235 wait_on_all_pages_writeback(sbi);
1236 if (unlikely(f2fs_cp_error(sbi)))
1239 /* write out checkpoint buffer at block 0 */
1240 update_meta_page(sbi, ckpt, start_blk++);
1242 for (i = 1; i < 1 + cp_payload_blks; i++)
1243 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1247 write_orphan_inodes(sbi, start_blk);
1248 start_blk += orphan_blocks;
1251 write_data_summaries(sbi, start_blk);
1252 start_blk += data_sum_blocks;
1254 /* Record write statistics in the hot node summary */
1255 kbytes_written = sbi->kbytes_written;
1256 if (sb->s_bdev->bd_part)
1257 kbytes_written += BD_PART_WRITTEN(sbi);
1259 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1261 if (__remain_node_summaries(cpc->reason)) {
1262 write_node_summaries(sbi, start_blk);
1263 start_blk += NR_CURSEG_NODE_TYPE;
1266 /* writeout checkpoint block */
1267 update_meta_page(sbi, ckpt, start_blk);
1269 /* wait for previous submitted node/meta pages writeback */
1270 wait_on_all_pages_writeback(sbi);
1272 if (unlikely(f2fs_cp_error(sbi)))
1275 filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1276 filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1278 /* update user_block_counts */
1279 sbi->last_valid_block_count = sbi->total_valid_block_count;
1280 percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1282 /* Here, we only have one bio having CP pack */
1283 sync_meta_pages(sbi, META_FLUSH, LONG_MAX, FS_CP_META_IO);
1285 /* wait for previous submitted meta pages writeback */
1286 wait_on_all_pages_writeback(sbi);
1288 release_ino_entry(sbi, false);
1290 if (unlikely(f2fs_cp_error(sbi)))
1293 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1294 clear_sbi_flag(sbi, SBI_NEED_CP);
1295 __set_cp_next_pack(sbi);
1298 * redirty superblock if metadata like node page or inode cache is
1299 * updated during writing checkpoint.
1301 if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1302 get_pages(sbi, F2FS_DIRTY_IMETA))
1303 set_sbi_flag(sbi, SBI_IS_DIRTY);
1305 f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1311 * We guarantee that this checkpoint procedure will not fail.
1313 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1315 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1316 unsigned long long ckpt_ver;
1319 mutex_lock(&sbi->cp_mutex);
1321 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1322 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1323 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1325 if (unlikely(f2fs_cp_error(sbi))) {
1329 if (f2fs_readonly(sbi->sb)) {
1334 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1336 err = block_operations(sbi);
1340 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1342 f2fs_flush_merged_writes(sbi);
1344 /* this is the case of multiple fstrims without any changes */
1345 if (cpc->reason & CP_DISCARD) {
1346 if (!exist_trim_candidates(sbi, cpc)) {
1347 unblock_operations(sbi);
1351 if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1352 SIT_I(sbi)->dirty_sentries == 0 &&
1353 prefree_segments(sbi) == 0) {
1354 flush_sit_entries(sbi, cpc);
1355 clear_prefree_segments(sbi, cpc);
1356 unblock_operations(sbi);
1362 * update checkpoint pack index
1363 * Increase the version number so that
1364 * SIT entries and seg summaries are written at correct place
1366 ckpt_ver = cur_cp_version(ckpt);
1367 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1369 /* write cached NAT/SIT entries to NAT/SIT area */
1370 flush_nat_entries(sbi, cpc);
1371 flush_sit_entries(sbi, cpc);
1373 /* unlock all the fs_lock[] in do_checkpoint() */
1374 err = do_checkpoint(sbi, cpc);
1376 release_discard_addrs(sbi);
1378 clear_prefree_segments(sbi, cpc);
1380 unblock_operations(sbi);
1381 stat_inc_cp_count(sbi->stat_info);
1383 if (cpc->reason & CP_RECOVERY)
1384 f2fs_msg(sbi->sb, KERN_NOTICE,
1385 "checkpoint: version = %llx", ckpt_ver);
1387 /* do checkpoint periodically */
1388 f2fs_update_time(sbi, CP_TIME);
1389 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1391 mutex_unlock(&sbi->cp_mutex);
1395 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1399 for (i = 0; i < MAX_INO_ENTRY; i++) {
1400 struct inode_management *im = &sbi->im[i];
1402 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1403 spin_lock_init(&im->ino_lock);
1404 INIT_LIST_HEAD(&im->ino_list);
1408 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1409 NR_CURSEG_TYPE - __cp_payload(sbi)) *
1410 F2FS_ORPHANS_PER_BLOCK;
1413 int __init create_checkpoint_caches(void)
1415 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1416 sizeof(struct ino_entry));
1417 if (!ino_entry_slab)
1419 inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1420 sizeof(struct inode_entry));
1421 if (!inode_entry_slab) {
1422 kmem_cache_destroy(ino_entry_slab);
1428 void destroy_checkpoint_caches(void)
1430 kmem_cache_destroy(ino_entry_slab);
1431 kmem_cache_destroy(inode_entry_slab);