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,
75 if (unlikely(!is_meta))
76 fio.op_flags &= ~REQ_META;
78 page = f2fs_grab_cache_page(mapping, index, false);
83 if (PageUptodate(page))
88 if (f2fs_submit_page_bio(&fio)) {
89 memset(page_address(page), 0, PAGE_SIZE);
90 f2fs_stop_checkpoint(sbi, false);
96 if (unlikely(page->mapping != mapping)) {
97 f2fs_put_page(page, 1);
102 * if there is any IO error when accessing device, make our filesystem
103 * readonly and make sure do not write checkpoint with non-uptodate
106 if (unlikely(!PageUptodate(page)))
107 f2fs_stop_checkpoint(sbi, false);
112 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
114 return __get_meta_page(sbi, index, true);
118 struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
120 return __get_meta_page(sbi, index, false);
123 bool f2fs_is_valid_blkaddr(struct f2fs_sb_info *sbi,
124 block_t blkaddr, int type)
130 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
134 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
135 blkaddr < SM_I(sbi)->ssa_blkaddr))
139 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
140 blkaddr < __start_cp_addr(sbi)))
145 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
146 blkaddr < MAIN_BLKADDR(sbi))) {
147 if (type == DATA_GENERIC) {
148 f2fs_msg(sbi->sb, KERN_WARNING,
149 "access invalid blkaddr:%u", blkaddr);
156 if (unlikely(blkaddr < SEG0_BLKADDR(sbi) ||
157 blkaddr >= MAIN_BLKADDR(sbi)))
168 * Readahead CP/NAT/SIT/SSA pages
170 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
174 block_t blkno = start;
175 struct f2fs_io_info fio = {
179 .op_flags = sync ? (REQ_META | REQ_PRIO) : REQ_RAHEAD,
180 .encrypted_page = NULL,
182 .is_meta = (type != META_POR),
184 struct blk_plug plug;
186 if (unlikely(type == META_POR))
187 fio.op_flags &= ~REQ_META;
189 blk_start_plug(&plug);
190 for (; nrpages-- > 0; blkno++) {
192 if (!f2fs_is_valid_blkaddr(sbi, blkno, type))
197 if (unlikely(blkno >=
198 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
200 /* get nat block addr */
201 fio.new_blkaddr = current_nat_addr(sbi,
202 blkno * NAT_ENTRY_PER_BLOCK);
205 if (unlikely(blkno >= TOTAL_SEGS(sbi)))
207 /* get sit block addr */
208 fio.new_blkaddr = current_sit_addr(sbi,
209 blkno * SIT_ENTRY_PER_BLOCK);
214 fio.new_blkaddr = blkno;
220 page = f2fs_grab_cache_page(META_MAPPING(sbi),
221 fio.new_blkaddr, false);
224 if (PageUptodate(page)) {
225 f2fs_put_page(page, 1);
230 f2fs_submit_page_bio(&fio);
231 f2fs_put_page(page, 0);
234 blk_finish_plug(&plug);
235 return blkno - start;
238 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
241 bool readahead = false;
243 page = find_get_page(META_MAPPING(sbi), index);
244 if (!page || !PageUptodate(page))
246 f2fs_put_page(page, 0);
249 ra_meta_pages(sbi, index, BIO_MAX_PAGES, META_POR, true);
252 static int __f2fs_write_meta_page(struct page *page,
253 struct writeback_control *wbc,
254 enum iostat_type io_type)
256 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
258 trace_f2fs_writepage(page, META);
260 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
262 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
264 if (unlikely(f2fs_cp_error(sbi)))
267 write_meta_page(sbi, page, io_type);
268 dec_page_count(sbi, F2FS_DIRTY_META);
270 if (wbc->for_reclaim)
271 f2fs_submit_merged_write_cond(sbi, page->mapping->host,
272 0, page->index, META);
276 if (unlikely(f2fs_cp_error(sbi)))
277 f2fs_submit_merged_write(sbi, META);
282 redirty_page_for_writepage(wbc, page);
283 return AOP_WRITEPAGE_ACTIVATE;
286 static int f2fs_write_meta_page(struct page *page,
287 struct writeback_control *wbc)
289 return __f2fs_write_meta_page(page, wbc, FS_META_IO);
292 static int f2fs_write_meta_pages(struct address_space *mapping,
293 struct writeback_control *wbc)
295 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
298 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
301 /* collect a number of dirty meta pages and write together */
302 if (wbc->for_kupdate ||
303 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
306 /* if locked failed, cp will flush dirty pages instead */
307 if (!mutex_trylock(&sbi->cp_mutex))
310 trace_f2fs_writepages(mapping->host, wbc, META);
311 diff = nr_pages_to_write(sbi, META, wbc);
312 written = sync_meta_pages(sbi, META, wbc->nr_to_write, FS_META_IO);
313 mutex_unlock(&sbi->cp_mutex);
314 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
318 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
319 trace_f2fs_writepages(mapping->host, wbc, META);
323 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
324 long nr_to_write, enum iostat_type io_type)
326 struct address_space *mapping = META_MAPPING(sbi);
327 pgoff_t index = 0, end = ULONG_MAX, prev = ULONG_MAX;
330 struct writeback_control wbc = {
333 struct blk_plug plug;
335 pagevec_init(&pvec, 0);
337 blk_start_plug(&plug);
339 while (index <= end) {
341 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
343 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
344 if (unlikely(nr_pages == 0))
347 for (i = 0; i < nr_pages; i++) {
348 struct page *page = pvec.pages[i];
350 if (prev == ULONG_MAX)
351 prev = page->index - 1;
352 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
353 pagevec_release(&pvec);
359 if (unlikely(page->mapping != mapping)) {
364 if (!PageDirty(page)) {
365 /* someone wrote it for us */
366 goto continue_unlock;
369 f2fs_wait_on_page_writeback(page, META, true);
371 BUG_ON(PageWriteback(page));
372 if (!clear_page_dirty_for_io(page))
373 goto continue_unlock;
375 if (__f2fs_write_meta_page(page, &wbc, io_type)) {
381 if (unlikely(nwritten >= nr_to_write))
384 pagevec_release(&pvec);
389 f2fs_submit_merged_write(sbi, type);
391 blk_finish_plug(&plug);
396 static int f2fs_set_meta_page_dirty(struct page *page)
398 trace_f2fs_set_page_dirty(page, META);
400 if (!PageUptodate(page))
401 SetPageUptodate(page);
402 if (!PageDirty(page)) {
403 f2fs_set_page_dirty_nobuffers(page);
404 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
405 SetPagePrivate(page);
406 f2fs_trace_pid(page);
412 const struct address_space_operations f2fs_meta_aops = {
413 .writepage = f2fs_write_meta_page,
414 .writepages = f2fs_write_meta_pages,
415 .set_page_dirty = f2fs_set_meta_page_dirty,
416 .invalidatepage = f2fs_invalidate_page,
417 .releasepage = f2fs_release_page,
418 #ifdef CONFIG_MIGRATION
419 .migratepage = f2fs_migrate_page,
423 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
425 struct inode_management *im = &sbi->im[type];
426 struct ino_entry *e, *tmp;
428 tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
430 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
432 spin_lock(&im->ino_lock);
433 e = radix_tree_lookup(&im->ino_root, ino);
436 if (radix_tree_insert(&im->ino_root, ino, e)) {
437 spin_unlock(&im->ino_lock);
438 radix_tree_preload_end();
441 memset(e, 0, sizeof(struct ino_entry));
444 list_add_tail(&e->list, &im->ino_list);
445 if (type != ORPHAN_INO)
448 spin_unlock(&im->ino_lock);
449 radix_tree_preload_end();
452 kmem_cache_free(ino_entry_slab, tmp);
455 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
457 struct inode_management *im = &sbi->im[type];
460 spin_lock(&im->ino_lock);
461 e = radix_tree_lookup(&im->ino_root, ino);
464 radix_tree_delete(&im->ino_root, ino);
466 spin_unlock(&im->ino_lock);
467 kmem_cache_free(ino_entry_slab, e);
470 spin_unlock(&im->ino_lock);
473 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
475 /* add new dirty ino entry into list */
476 __add_ino_entry(sbi, ino, type);
479 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
481 /* remove dirty ino entry from list */
482 __remove_ino_entry(sbi, ino, type);
485 /* mode should be APPEND_INO or UPDATE_INO */
486 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
488 struct inode_management *im = &sbi->im[mode];
491 spin_lock(&im->ino_lock);
492 e = radix_tree_lookup(&im->ino_root, ino);
493 spin_unlock(&im->ino_lock);
494 return e ? true : false;
497 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
499 struct ino_entry *e, *tmp;
502 for (i = all ? ORPHAN_INO: APPEND_INO; i <= UPDATE_INO; i++) {
503 struct inode_management *im = &sbi->im[i];
505 spin_lock(&im->ino_lock);
506 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
508 radix_tree_delete(&im->ino_root, e->ino);
509 kmem_cache_free(ino_entry_slab, e);
512 spin_unlock(&im->ino_lock);
516 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
518 struct inode_management *im = &sbi->im[ORPHAN_INO];
521 spin_lock(&im->ino_lock);
523 #ifdef CONFIG_F2FS_FAULT_INJECTION
524 if (time_to_inject(sbi, FAULT_ORPHAN)) {
525 spin_unlock(&im->ino_lock);
526 f2fs_show_injection_info(FAULT_ORPHAN);
530 if (unlikely(im->ino_num >= sbi->max_orphans))
534 spin_unlock(&im->ino_lock);
539 void release_orphan_inode(struct f2fs_sb_info *sbi)
541 struct inode_management *im = &sbi->im[ORPHAN_INO];
543 spin_lock(&im->ino_lock);
544 f2fs_bug_on(sbi, im->ino_num == 0);
546 spin_unlock(&im->ino_lock);
549 void add_orphan_inode(struct inode *inode)
551 /* add new orphan ino entry into list */
552 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, ORPHAN_INO);
553 update_inode_page(inode);
556 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
558 /* remove orphan entry from orphan list */
559 __remove_ino_entry(sbi, ino, ORPHAN_INO);
562 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
566 int err = acquire_orphan_inode(sbi);
569 set_sbi_flag(sbi, SBI_NEED_FSCK);
570 f2fs_msg(sbi->sb, KERN_WARNING,
571 "%s: orphan failed (ino=%x), run fsck to fix.",
576 __add_ino_entry(sbi, ino, ORPHAN_INO);
578 inode = f2fs_iget_retry(sbi->sb, ino);
581 * there should be a bug that we can't find the entry
584 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
585 return PTR_ERR(inode);
590 /* truncate all the data during iput */
593 get_node_info(sbi, ino, &ni);
595 /* ENOMEM was fully retried in f2fs_evict_inode. */
596 if (ni.blk_addr != NULL_ADDR) {
597 set_sbi_flag(sbi, SBI_NEED_FSCK);
598 f2fs_msg(sbi->sb, KERN_WARNING,
599 "%s: orphan failed (ino=%x) by kernel, retry mount.",
603 __remove_ino_entry(sbi, ino, ORPHAN_INO);
607 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
609 block_t start_blk, orphan_blocks, i, j;
610 unsigned int s_flags = sbi->sb->s_flags;
613 if (!is_set_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG))
616 if (s_flags & MS_RDONLY) {
617 f2fs_msg(sbi->sb, KERN_INFO, "orphan cleanup on readonly fs");
618 sbi->sb->s_flags &= ~MS_RDONLY;
622 /* Needed for iput() to work correctly and not trash data */
623 sbi->sb->s_flags |= MS_ACTIVE;
624 /* Turn on quotas so that they are updated correctly */
625 f2fs_enable_quota_files(sbi);
628 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
629 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
631 ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
633 for (i = 0; i < orphan_blocks; i++) {
634 struct page *page = get_meta_page(sbi, start_blk + i);
635 struct f2fs_orphan_block *orphan_blk;
637 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
638 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
639 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
640 err = recover_orphan_inode(sbi, ino);
642 f2fs_put_page(page, 1);
646 f2fs_put_page(page, 1);
648 /* clear Orphan Flag */
649 clear_ckpt_flags(sbi, CP_ORPHAN_PRESENT_FLAG);
652 /* Turn quotas off */
653 f2fs_quota_off_umount(sbi->sb);
655 sbi->sb->s_flags = s_flags; /* Restore MS_RDONLY status */
660 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
662 struct list_head *head;
663 struct f2fs_orphan_block *orphan_blk = NULL;
664 unsigned int nentries = 0;
665 unsigned short index = 1;
666 unsigned short orphan_blocks;
667 struct page *page = NULL;
668 struct ino_entry *orphan = NULL;
669 struct inode_management *im = &sbi->im[ORPHAN_INO];
671 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
674 * we don't need to do spin_lock(&im->ino_lock) here, since all the
675 * orphan inode operations are covered under f2fs_lock_op().
676 * And, spin_lock should be avoided due to page operations below.
678 head = &im->ino_list;
680 /* loop for each orphan inode entry and write them in Jornal block */
681 list_for_each_entry(orphan, head, list) {
683 page = grab_meta_page(sbi, start_blk++);
685 (struct f2fs_orphan_block *)page_address(page);
686 memset(orphan_blk, 0, sizeof(*orphan_blk));
689 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
691 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
693 * an orphan block is full of 1020 entries,
694 * then we need to flush current orphan blocks
695 * and bring another one in memory
697 orphan_blk->blk_addr = cpu_to_le16(index);
698 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
699 orphan_blk->entry_count = cpu_to_le32(nentries);
700 set_page_dirty(page);
701 f2fs_put_page(page, 1);
709 orphan_blk->blk_addr = cpu_to_le16(index);
710 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
711 orphan_blk->entry_count = cpu_to_le32(nentries);
712 set_page_dirty(page);
713 f2fs_put_page(page, 1);
717 static int get_checkpoint_version(struct f2fs_sb_info *sbi, block_t cp_addr,
718 struct f2fs_checkpoint **cp_block, struct page **cp_page,
719 unsigned long long *version)
721 unsigned long blk_size = sbi->blocksize;
722 size_t crc_offset = 0;
725 *cp_page = get_meta_page(sbi, cp_addr);
726 *cp_block = (struct f2fs_checkpoint *)page_address(*cp_page);
728 crc_offset = le32_to_cpu((*cp_block)->checksum_offset);
729 if (crc_offset > (blk_size - sizeof(__le32))) {
730 f2fs_put_page(*cp_page, 1);
731 f2fs_msg(sbi->sb, KERN_WARNING,
732 "invalid crc_offset: %zu", crc_offset);
736 crc = cur_cp_crc(*cp_block);
737 if (!f2fs_crc_valid(sbi, crc, *cp_block, crc_offset)) {
738 f2fs_put_page(*cp_page, 1);
739 f2fs_msg(sbi->sb, KERN_WARNING, "invalid crc value");
743 *version = cur_cp_version(*cp_block);
747 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
748 block_t cp_addr, unsigned long long *version)
750 struct page *cp_page_1 = NULL, *cp_page_2 = NULL;
751 struct f2fs_checkpoint *cp_block = NULL;
752 unsigned long long cur_version = 0, pre_version = 0;
755 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
756 &cp_page_1, version);
760 if (le32_to_cpu(cp_block->cp_pack_total_block_count) >
761 sbi->blocks_per_seg) {
762 f2fs_msg(sbi->sb, KERN_WARNING,
763 "invalid cp_pack_total_block_count:%u",
764 le32_to_cpu(cp_block->cp_pack_total_block_count));
767 pre_version = *version;
769 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
770 err = get_checkpoint_version(sbi, cp_addr, &cp_block,
771 &cp_page_2, version);
774 cur_version = *version;
776 if (cur_version == pre_version) {
777 *version = cur_version;
778 f2fs_put_page(cp_page_2, 1);
781 f2fs_put_page(cp_page_2, 1);
783 f2fs_put_page(cp_page_1, 1);
787 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
789 struct f2fs_checkpoint *cp_block;
790 struct f2fs_super_block *fsb = sbi->raw_super;
791 struct page *cp1, *cp2, *cur_page;
792 unsigned long blk_size = sbi->blocksize;
793 unsigned long long cp1_version = 0, cp2_version = 0;
794 unsigned long long cp_start_blk_no;
795 unsigned int cp_blks = 1 + __cp_payload(sbi);
800 sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
804 * Finding out valid cp block involves read both
805 * sets( cp pack1 and cp pack 2)
807 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
808 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
810 /* The second checkpoint pack should start at the next segment */
811 cp_start_blk_no += ((unsigned long long)1) <<
812 le32_to_cpu(fsb->log_blocks_per_seg);
813 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
816 if (ver_after(cp2_version, cp1_version))
829 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
830 memcpy(sbi->ckpt, cp_block, blk_size);
833 sbi->cur_cp_pack = 1;
835 sbi->cur_cp_pack = 2;
837 /* Sanity checking of checkpoint */
838 if (sanity_check_ckpt(sbi)) {
840 goto free_fail_no_cp;
846 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
848 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
850 for (i = 1; i < cp_blks; i++) {
851 void *sit_bitmap_ptr;
852 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
854 cur_page = get_meta_page(sbi, cp_blk_no + i);
855 sit_bitmap_ptr = page_address(cur_page);
856 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
857 f2fs_put_page(cur_page, 1);
860 f2fs_put_page(cp1, 1);
861 f2fs_put_page(cp2, 1);
865 f2fs_put_page(cp1, 1);
866 f2fs_put_page(cp2, 1);
872 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
874 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
875 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
877 if (is_inode_flag_set(inode, flag))
880 set_inode_flag(inode, flag);
881 if (!f2fs_is_volatile_file(inode))
882 list_add_tail(&F2FS_I(inode)->dirty_list,
883 &sbi->inode_list[type]);
884 stat_inc_dirty_inode(sbi, type);
887 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
889 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
891 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
894 list_del_init(&F2FS_I(inode)->dirty_list);
895 clear_inode_flag(inode, flag);
896 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
899 void update_dirty_page(struct inode *inode, struct page *page)
901 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
902 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
904 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
905 !S_ISLNK(inode->i_mode))
908 spin_lock(&sbi->inode_lock[type]);
909 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
910 __add_dirty_inode(inode, type);
911 inode_inc_dirty_pages(inode);
912 spin_unlock(&sbi->inode_lock[type]);
914 SetPagePrivate(page);
915 f2fs_trace_pid(page);
918 void remove_dirty_inode(struct inode *inode)
920 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
921 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
923 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
924 !S_ISLNK(inode->i_mode))
927 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
930 spin_lock(&sbi->inode_lock[type]);
931 __remove_dirty_inode(inode, type);
932 spin_unlock(&sbi->inode_lock[type]);
935 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
937 struct list_head *head;
939 struct f2fs_inode_info *fi;
940 bool is_dir = (type == DIR_INODE);
941 unsigned long ino = 0;
943 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
944 get_pages(sbi, is_dir ?
945 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
947 if (unlikely(f2fs_cp_error(sbi))) {
948 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
949 get_pages(sbi, is_dir ?
950 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
954 spin_lock(&sbi->inode_lock[type]);
956 head = &sbi->inode_list[type];
957 if (list_empty(head)) {
958 spin_unlock(&sbi->inode_lock[type]);
959 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
960 get_pages(sbi, is_dir ?
961 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
964 fi = list_first_entry(head, struct f2fs_inode_info, dirty_list);
965 inode = igrab(&fi->vfs_inode);
966 spin_unlock(&sbi->inode_lock[type]);
968 unsigned long cur_ino = inode->i_ino;
971 F2FS_I(inode)->cp_task = current;
973 filemap_fdatawrite(inode->i_mapping);
976 F2FS_I(inode)->cp_task = NULL;
979 /* We need to give cpu to another writers. */
980 if (ino == cur_ino) {
981 congestion_wait(BLK_RW_ASYNC, HZ/50);
988 * We should submit bio, since it exists several
989 * wribacking dentry pages in the freeing inode.
991 f2fs_submit_merged_write(sbi, DATA);
997 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
999 struct list_head *head = &sbi->inode_list[DIRTY_META];
1000 struct inode *inode;
1001 struct f2fs_inode_info *fi;
1002 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
1005 if (unlikely(f2fs_cp_error(sbi)))
1008 spin_lock(&sbi->inode_lock[DIRTY_META]);
1009 if (list_empty(head)) {
1010 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1013 fi = list_first_entry(head, struct f2fs_inode_info,
1015 inode = igrab(&fi->vfs_inode);
1016 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1018 sync_inode_metadata(inode, 0);
1020 /* it's on eviction */
1021 if (is_inode_flag_set(inode, FI_DIRTY_INODE))
1022 update_inode_page(inode);
1029 static void __prepare_cp_block(struct f2fs_sb_info *sbi)
1031 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1032 struct f2fs_nm_info *nm_i = NM_I(sbi);
1033 nid_t last_nid = nm_i->next_scan_nid;
1035 next_free_nid(sbi, &last_nid);
1036 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1037 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1038 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1039 ckpt->next_free_nid = cpu_to_le32(last_nid);
1043 * Freeze all the FS-operations for checkpoint.
1045 static int block_operations(struct f2fs_sb_info *sbi)
1047 struct writeback_control wbc = {
1048 .sync_mode = WB_SYNC_ALL,
1049 .nr_to_write = LONG_MAX,
1052 struct blk_plug plug;
1055 blk_start_plug(&plug);
1059 /* write all the dirty dentry pages */
1060 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
1061 f2fs_unlock_all(sbi);
1062 err = sync_dirty_inodes(sbi, DIR_INODE);
1066 goto retry_flush_dents;
1070 * POR: we should ensure that there are no dirty node pages
1071 * until finishing nat/sit flush. inode->i_blocks can be updated.
1073 down_write(&sbi->node_change);
1075 if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
1076 up_write(&sbi->node_change);
1077 f2fs_unlock_all(sbi);
1078 err = f2fs_sync_inode_meta(sbi);
1082 goto retry_flush_dents;
1086 down_write(&sbi->node_write);
1088 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
1089 up_write(&sbi->node_write);
1090 err = sync_node_pages(sbi, &wbc, false, FS_CP_NODE_IO);
1092 up_write(&sbi->node_change);
1093 f2fs_unlock_all(sbi);
1097 goto retry_flush_nodes;
1101 * sbi->node_change is used only for AIO write_begin path which produces
1102 * dirty node blocks and some checkpoint values by block allocation.
1104 __prepare_cp_block(sbi);
1105 up_write(&sbi->node_change);
1107 blk_finish_plug(&plug);
1111 static void unblock_operations(struct f2fs_sb_info *sbi)
1113 up_write(&sbi->node_write);
1114 f2fs_unlock_all(sbi);
1117 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
1122 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
1124 if (!get_pages(sbi, F2FS_WB_CP_DATA))
1127 io_schedule_timeout(5*HZ);
1129 finish_wait(&sbi->cp_wait, &wait);
1132 static void update_ckpt_flags(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1134 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
1135 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1136 unsigned long flags;
1138 spin_lock_irqsave(&sbi->cp_lock, flags);
1140 if ((cpc->reason & CP_UMOUNT) &&
1141 le32_to_cpu(ckpt->cp_pack_total_block_count) >
1142 sbi->blocks_per_seg - NM_I(sbi)->nat_bits_blocks)
1143 disable_nat_bits(sbi, false);
1145 if (cpc->reason & CP_TRIMMED)
1146 __set_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1148 __clear_ckpt_flags(ckpt, CP_TRIMMED_FLAG);
1150 if (cpc->reason & CP_UMOUNT)
1151 __set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1153 __clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1155 if (cpc->reason & CP_FASTBOOT)
1156 __set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1158 __clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1161 __set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1163 __clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1165 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1166 __set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1168 /* set this flag to activate crc|cp_ver for recovery */
1169 __set_ckpt_flags(ckpt, CP_CRC_RECOVERY_FLAG);
1171 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1174 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1176 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1177 struct f2fs_nm_info *nm_i = NM_I(sbi);
1178 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num, flags;
1180 unsigned int data_sum_blocks, orphan_blocks;
1183 int cp_payload_blks = __cp_payload(sbi);
1184 struct super_block *sb = sbi->sb;
1185 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
1188 /* Flush all the NAT/SIT pages */
1189 while (get_pages(sbi, F2FS_DIRTY_META)) {
1190 sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1191 if (unlikely(f2fs_cp_error(sbi)))
1197 * version number is already updated
1199 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
1200 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1201 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1202 ckpt->cur_node_segno[i] =
1203 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1204 ckpt->cur_node_blkoff[i] =
1205 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1206 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1207 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1209 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1210 ckpt->cur_data_segno[i] =
1211 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1212 ckpt->cur_data_blkoff[i] =
1213 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1214 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1215 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1218 /* 2 cp + n data seg summary + orphan inode blocks */
1219 data_sum_blocks = npages_for_summary_flush(sbi, false);
1220 spin_lock_irqsave(&sbi->cp_lock, flags);
1221 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1222 __set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1224 __clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1225 spin_unlock_irqrestore(&sbi->cp_lock, flags);
1227 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1228 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1231 if (__remain_node_summaries(cpc->reason))
1232 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1233 cp_payload_blks + data_sum_blocks +
1234 orphan_blocks + NR_CURSEG_NODE_TYPE);
1236 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1237 cp_payload_blks + data_sum_blocks +
1240 /* update ckpt flag for checkpoint */
1241 update_ckpt_flags(sbi, cpc);
1243 /* update SIT/NAT bitmap */
1244 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1245 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1247 crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1248 *((__le32 *)((unsigned char *)ckpt +
1249 le32_to_cpu(ckpt->checksum_offset)))
1250 = cpu_to_le32(crc32);
1252 start_blk = __start_cp_next_addr(sbi);
1254 /* write nat bits */
1255 if (enabled_nat_bits(sbi, cpc)) {
1256 __u64 cp_ver = cur_cp_version(ckpt);
1259 cp_ver |= ((__u64)crc32 << 32);
1260 *(__le64 *)nm_i->nat_bits = cpu_to_le64(cp_ver);
1262 blk = start_blk + sbi->blocks_per_seg - nm_i->nat_bits_blocks;
1263 for (i = 0; i < nm_i->nat_bits_blocks; i++)
1264 update_meta_page(sbi, nm_i->nat_bits +
1265 (i << F2FS_BLKSIZE_BITS), blk + i);
1267 /* Flush all the NAT BITS pages */
1268 while (get_pages(sbi, F2FS_DIRTY_META)) {
1269 sync_meta_pages(sbi, META, LONG_MAX, FS_CP_META_IO);
1270 if (unlikely(f2fs_cp_error(sbi)))
1275 /* need to wait for end_io results */
1276 wait_on_all_pages_writeback(sbi);
1277 if (unlikely(f2fs_cp_error(sbi)))
1280 /* write out checkpoint buffer at block 0 */
1281 update_meta_page(sbi, ckpt, start_blk++);
1283 for (i = 1; i < 1 + cp_payload_blks; i++)
1284 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1288 write_orphan_inodes(sbi, start_blk);
1289 start_blk += orphan_blocks;
1292 write_data_summaries(sbi, start_blk);
1293 start_blk += data_sum_blocks;
1295 /* Record write statistics in the hot node summary */
1296 kbytes_written = sbi->kbytes_written;
1297 if (sb->s_bdev->bd_part)
1298 kbytes_written += BD_PART_WRITTEN(sbi);
1300 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1302 if (__remain_node_summaries(cpc->reason)) {
1303 write_node_summaries(sbi, start_blk);
1304 start_blk += NR_CURSEG_NODE_TYPE;
1307 /* writeout checkpoint block */
1308 update_meta_page(sbi, ckpt, start_blk);
1310 /* wait for previous submitted node/meta pages writeback */
1311 wait_on_all_pages_writeback(sbi);
1313 if (unlikely(f2fs_cp_error(sbi)))
1316 filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1317 filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1319 /* update user_block_counts */
1320 sbi->last_valid_block_count = sbi->total_valid_block_count;
1321 percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1323 /* Here, we only have one bio having CP pack */
1324 sync_meta_pages(sbi, META_FLUSH, LONG_MAX, FS_CP_META_IO);
1326 /* wait for previous submitted meta pages writeback */
1327 wait_on_all_pages_writeback(sbi);
1329 release_ino_entry(sbi, false);
1331 if (unlikely(f2fs_cp_error(sbi)))
1334 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1335 clear_sbi_flag(sbi, SBI_NEED_CP);
1336 __set_cp_next_pack(sbi);
1339 * redirty superblock if metadata like node page or inode cache is
1340 * updated during writing checkpoint.
1342 if (get_pages(sbi, F2FS_DIRTY_NODES) ||
1343 get_pages(sbi, F2FS_DIRTY_IMETA))
1344 set_sbi_flag(sbi, SBI_IS_DIRTY);
1346 f2fs_bug_on(sbi, get_pages(sbi, F2FS_DIRTY_DENTS));
1352 * We guarantee that this checkpoint procedure will not fail.
1354 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1356 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1357 unsigned long long ckpt_ver;
1360 mutex_lock(&sbi->cp_mutex);
1362 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1363 ((cpc->reason & CP_FASTBOOT) || (cpc->reason & CP_SYNC) ||
1364 ((cpc->reason & CP_DISCARD) && !sbi->discard_blks)))
1366 if (unlikely(f2fs_cp_error(sbi))) {
1370 if (f2fs_readonly(sbi->sb)) {
1375 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1377 err = block_operations(sbi);
1381 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1383 f2fs_flush_merged_writes(sbi);
1385 /* this is the case of multiple fstrims without any changes */
1386 if (cpc->reason & CP_DISCARD) {
1387 if (!exist_trim_candidates(sbi, cpc)) {
1388 unblock_operations(sbi);
1392 if (NM_I(sbi)->dirty_nat_cnt == 0 &&
1393 SIT_I(sbi)->dirty_sentries == 0 &&
1394 prefree_segments(sbi) == 0) {
1395 flush_sit_entries(sbi, cpc);
1396 clear_prefree_segments(sbi, cpc);
1397 unblock_operations(sbi);
1403 * update checkpoint pack index
1404 * Increase the version number so that
1405 * SIT entries and seg summaries are written at correct place
1407 ckpt_ver = cur_cp_version(ckpt);
1408 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1410 /* write cached NAT/SIT entries to NAT/SIT area */
1411 flush_nat_entries(sbi, cpc);
1412 flush_sit_entries(sbi, cpc);
1414 /* unlock all the fs_lock[] in do_checkpoint() */
1415 err = do_checkpoint(sbi, cpc);
1417 release_discard_addrs(sbi);
1419 clear_prefree_segments(sbi, cpc);
1421 unblock_operations(sbi);
1422 stat_inc_cp_count(sbi->stat_info);
1424 if (cpc->reason & CP_RECOVERY)
1425 f2fs_msg(sbi->sb, KERN_NOTICE,
1426 "checkpoint: version = %llx", ckpt_ver);
1428 /* do checkpoint periodically */
1429 f2fs_update_time(sbi, CP_TIME);
1430 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1432 mutex_unlock(&sbi->cp_mutex);
1436 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1440 for (i = 0; i < MAX_INO_ENTRY; i++) {
1441 struct inode_management *im = &sbi->im[i];
1443 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1444 spin_lock_init(&im->ino_lock);
1445 INIT_LIST_HEAD(&im->ino_list);
1449 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1450 NR_CURSEG_TYPE - __cp_payload(sbi)) *
1451 F2FS_ORPHANS_PER_BLOCK;
1454 int __init create_checkpoint_caches(void)
1456 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1457 sizeof(struct ino_entry));
1458 if (!ino_entry_slab)
1460 inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1461 sizeof(struct inode_entry));
1462 if (!inode_entry_slab) {
1463 kmem_cache_destroy(ino_entry_slab);
1469 void destroy_checkpoint_caches(void)
1471 kmem_cache_destroy(ino_entry_slab);
1472 kmem_cache_destroy(inode_entry_slab);