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 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 = READ_SYNC | 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 ? (READ_SYNC | REQ_META | REQ_PRIO) : REQ_RAHEAD,
164 .encrypted_page = NULL,
166 struct blk_plug plug;
168 if (unlikely(type == META_POR))
169 fio.op_flags &= ~REQ_META;
171 blk_start_plug(&plug);
172 for (; nrpages-- > 0; blkno++) {
174 if (!is_valid_blkaddr(sbi, blkno, type))
179 if (unlikely(blkno >=
180 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
182 /* get nat block addr */
183 fio.new_blkaddr = current_nat_addr(sbi,
184 blkno * NAT_ENTRY_PER_BLOCK);
187 /* get sit block addr */
188 fio.new_blkaddr = current_sit_addr(sbi,
189 blkno * SIT_ENTRY_PER_BLOCK);
194 fio.new_blkaddr = blkno;
200 page = f2fs_grab_cache_page(META_MAPPING(sbi),
201 fio.new_blkaddr, false);
204 if (PageUptodate(page)) {
205 f2fs_put_page(page, 1);
210 fio.old_blkaddr = fio.new_blkaddr;
211 f2fs_submit_page_mbio(&fio);
212 f2fs_put_page(page, 0);
215 f2fs_submit_merged_bio(sbi, META, READ);
216 blk_finish_plug(&plug);
217 return blkno - start;
220 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
223 bool readahead = false;
225 page = find_get_page(META_MAPPING(sbi), index);
226 if (!page || !PageUptodate(page))
228 f2fs_put_page(page, 0);
231 ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR, true);
234 static int f2fs_write_meta_page(struct page *page,
235 struct writeback_control *wbc)
237 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
239 trace_f2fs_writepage(page, META);
241 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
243 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
245 if (unlikely(f2fs_cp_error(sbi)))
248 write_meta_page(sbi, page);
249 dec_page_count(sbi, F2FS_DIRTY_META);
251 if (wbc->for_reclaim)
252 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, META, WRITE);
256 if (unlikely(f2fs_cp_error(sbi)))
257 f2fs_submit_merged_bio(sbi, META, WRITE);
262 redirty_page_for_writepage(wbc, page);
263 return AOP_WRITEPAGE_ACTIVATE;
266 static int f2fs_write_meta_pages(struct address_space *mapping,
267 struct writeback_control *wbc)
269 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
270 struct blk_plug plug;
273 /* collect a number of dirty meta pages and write together */
274 if (wbc->for_kupdate ||
275 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
278 trace_f2fs_writepages(mapping->host, wbc, META);
280 /* if mounting is failed, skip writing node pages */
281 mutex_lock(&sbi->cp_mutex);
282 diff = nr_pages_to_write(sbi, META, wbc);
283 blk_start_plug(&plug);
284 written = sync_meta_pages(sbi, META, wbc->nr_to_write);
285 blk_finish_plug(&plug);
286 mutex_unlock(&sbi->cp_mutex);
287 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
291 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
292 trace_f2fs_writepages(mapping->host, wbc, META);
296 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
299 struct address_space *mapping = META_MAPPING(sbi);
300 pgoff_t index = 0, end = ULONG_MAX, prev = ULONG_MAX;
303 struct writeback_control wbc = {
306 struct blk_plug plug;
308 pagevec_init(&pvec, 0);
310 blk_start_plug(&plug);
312 while (index <= end) {
314 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
316 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
317 if (unlikely(nr_pages == 0))
320 for (i = 0; i < nr_pages; i++) {
321 struct page *page = pvec.pages[i];
323 if (prev == ULONG_MAX)
324 prev = page->index - 1;
325 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
326 pagevec_release(&pvec);
332 if (unlikely(page->mapping != mapping)) {
337 if (!PageDirty(page)) {
338 /* someone wrote it for us */
339 goto continue_unlock;
342 f2fs_wait_on_page_writeback(page, META, true);
344 BUG_ON(PageWriteback(page));
345 if (!clear_page_dirty_for_io(page))
346 goto continue_unlock;
348 if (mapping->a_ops->writepage(page, &wbc)) {
354 if (unlikely(nwritten >= nr_to_write))
357 pagevec_release(&pvec);
362 f2fs_submit_merged_bio(sbi, type, WRITE);
364 blk_finish_plug(&plug);
369 static int f2fs_set_meta_page_dirty(struct page *page)
371 trace_f2fs_set_page_dirty(page, META);
373 if (!PageUptodate(page))
374 SetPageUptodate(page);
375 if (!PageDirty(page)) {
376 f2fs_set_page_dirty_nobuffers(page);
377 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
378 SetPagePrivate(page);
379 f2fs_trace_pid(page);
385 const struct address_space_operations f2fs_meta_aops = {
386 .writepage = f2fs_write_meta_page,
387 .writepages = f2fs_write_meta_pages,
388 .set_page_dirty = f2fs_set_meta_page_dirty,
389 .invalidatepage = f2fs_invalidate_page,
390 .releasepage = f2fs_release_page,
393 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
395 struct inode_management *im = &sbi->im[type];
396 struct ino_entry *e, *tmp;
398 tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
400 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
402 spin_lock(&im->ino_lock);
403 e = radix_tree_lookup(&im->ino_root, ino);
406 if (radix_tree_insert(&im->ino_root, ino, e)) {
407 spin_unlock(&im->ino_lock);
408 radix_tree_preload_end();
411 memset(e, 0, sizeof(struct ino_entry));
414 list_add_tail(&e->list, &im->ino_list);
415 if (type != ORPHAN_INO)
418 spin_unlock(&im->ino_lock);
419 radix_tree_preload_end();
422 kmem_cache_free(ino_entry_slab, tmp);
425 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
427 struct inode_management *im = &sbi->im[type];
430 spin_lock(&im->ino_lock);
431 e = radix_tree_lookup(&im->ino_root, ino);
434 radix_tree_delete(&im->ino_root, ino);
436 spin_unlock(&im->ino_lock);
437 kmem_cache_free(ino_entry_slab, e);
440 spin_unlock(&im->ino_lock);
443 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
445 /* add new dirty ino entry into list */
446 __add_ino_entry(sbi, ino, type);
449 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
451 /* remove dirty ino entry from list */
452 __remove_ino_entry(sbi, ino, type);
455 /* mode should be APPEND_INO or UPDATE_INO */
456 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
458 struct inode_management *im = &sbi->im[mode];
461 spin_lock(&im->ino_lock);
462 e = radix_tree_lookup(&im->ino_root, ino);
463 spin_unlock(&im->ino_lock);
464 return e ? true : false;
467 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
469 struct ino_entry *e, *tmp;
472 for (i = all ? ORPHAN_INO: APPEND_INO; i <= UPDATE_INO; i++) {
473 struct inode_management *im = &sbi->im[i];
475 spin_lock(&im->ino_lock);
476 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
478 radix_tree_delete(&im->ino_root, e->ino);
479 kmem_cache_free(ino_entry_slab, e);
482 spin_unlock(&im->ino_lock);
486 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
488 struct inode_management *im = &sbi->im[ORPHAN_INO];
491 spin_lock(&im->ino_lock);
493 #ifdef CONFIG_F2FS_FAULT_INJECTION
494 if (time_to_inject(FAULT_ORPHAN)) {
495 spin_unlock(&im->ino_lock);
499 if (unlikely(im->ino_num >= sbi->max_orphans))
503 spin_unlock(&im->ino_lock);
508 void release_orphan_inode(struct f2fs_sb_info *sbi)
510 struct inode_management *im = &sbi->im[ORPHAN_INO];
512 spin_lock(&im->ino_lock);
513 f2fs_bug_on(sbi, im->ino_num == 0);
515 spin_unlock(&im->ino_lock);
518 void add_orphan_inode(struct inode *inode)
520 /* add new orphan ino entry into list */
521 __add_ino_entry(F2FS_I_SB(inode), inode->i_ino, ORPHAN_INO);
522 update_inode_page(inode);
525 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
527 /* remove orphan entry from orphan list */
528 __remove_ino_entry(sbi, ino, ORPHAN_INO);
531 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
535 inode = f2fs_iget(sbi->sb, ino);
538 * there should be a bug that we can't find the entry
541 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
542 return PTR_ERR(inode);
547 /* truncate all the data during iput */
552 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
554 block_t start_blk, orphan_blocks, i, j;
557 if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
560 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
561 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
563 ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
565 for (i = 0; i < orphan_blocks; i++) {
566 struct page *page = get_meta_page(sbi, start_blk + i);
567 struct f2fs_orphan_block *orphan_blk;
569 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
570 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
571 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
572 err = recover_orphan_inode(sbi, ino);
574 f2fs_put_page(page, 1);
578 f2fs_put_page(page, 1);
580 /* clear Orphan Flag */
581 clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
585 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
587 struct list_head *head;
588 struct f2fs_orphan_block *orphan_blk = NULL;
589 unsigned int nentries = 0;
590 unsigned short index = 1;
591 unsigned short orphan_blocks;
592 struct page *page = NULL;
593 struct ino_entry *orphan = NULL;
594 struct inode_management *im = &sbi->im[ORPHAN_INO];
596 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
599 * we don't need to do spin_lock(&im->ino_lock) here, since all the
600 * orphan inode operations are covered under f2fs_lock_op().
601 * And, spin_lock should be avoided due to page operations below.
603 head = &im->ino_list;
605 /* loop for each orphan inode entry and write them in Jornal block */
606 list_for_each_entry(orphan, head, list) {
608 page = grab_meta_page(sbi, start_blk++);
610 (struct f2fs_orphan_block *)page_address(page);
611 memset(orphan_blk, 0, sizeof(*orphan_blk));
614 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
616 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
618 * an orphan block is full of 1020 entries,
619 * then we need to flush current orphan blocks
620 * and bring another one in memory
622 orphan_blk->blk_addr = cpu_to_le16(index);
623 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
624 orphan_blk->entry_count = cpu_to_le32(nentries);
625 set_page_dirty(page);
626 f2fs_put_page(page, 1);
634 orphan_blk->blk_addr = cpu_to_le16(index);
635 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
636 orphan_blk->entry_count = cpu_to_le32(nentries);
637 set_page_dirty(page);
638 f2fs_put_page(page, 1);
642 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
643 block_t cp_addr, unsigned long long *version)
645 struct page *cp_page_1, *cp_page_2 = NULL;
646 unsigned long blk_size = sbi->blocksize;
647 struct f2fs_checkpoint *cp_block;
648 unsigned long long cur_version = 0, pre_version = 0;
652 /* Read the 1st cp block in this CP pack */
653 cp_page_1 = get_meta_page(sbi, cp_addr);
655 /* get the version number */
656 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
657 crc_offset = le32_to_cpu(cp_block->checksum_offset);
658 if (crc_offset >= blk_size)
661 crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
662 if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
665 pre_version = cur_cp_version(cp_block);
667 /* Read the 2nd cp block in this CP pack */
668 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
669 cp_page_2 = get_meta_page(sbi, cp_addr);
671 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
672 crc_offset = le32_to_cpu(cp_block->checksum_offset);
673 if (crc_offset >= blk_size)
676 crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
677 if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
680 cur_version = cur_cp_version(cp_block);
682 if (cur_version == pre_version) {
683 *version = cur_version;
684 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);
737 /* Sanity checking of checkpoint */
738 if (sanity_check_ckpt(sbi))
744 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
746 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
748 for (i = 1; i < cp_blks; i++) {
749 void *sit_bitmap_ptr;
750 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
752 cur_page = get_meta_page(sbi, cp_blk_no + i);
753 sit_bitmap_ptr = page_address(cur_page);
754 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
755 f2fs_put_page(cur_page, 1);
758 f2fs_put_page(cp1, 1);
759 f2fs_put_page(cp2, 1);
767 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
769 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
770 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
772 if (is_inode_flag_set(inode, flag))
775 set_inode_flag(inode, flag);
776 list_add_tail(&F2FS_I(inode)->dirty_list, &sbi->inode_list[type]);
777 stat_inc_dirty_inode(sbi, type);
780 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
782 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
784 if (get_dirty_pages(inode) || !is_inode_flag_set(inode, flag))
787 list_del_init(&F2FS_I(inode)->dirty_list);
788 clear_inode_flag(inode, flag);
789 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
792 void update_dirty_page(struct inode *inode, struct page *page)
794 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
795 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
797 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
798 !S_ISLNK(inode->i_mode))
801 spin_lock(&sbi->inode_lock[type]);
802 if (type != FILE_INODE || test_opt(sbi, DATA_FLUSH))
803 __add_dirty_inode(inode, type);
804 inode_inc_dirty_pages(inode);
805 spin_unlock(&sbi->inode_lock[type]);
807 SetPagePrivate(page);
808 f2fs_trace_pid(page);
811 void remove_dirty_inode(struct inode *inode)
813 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
814 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
816 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
817 !S_ISLNK(inode->i_mode))
820 if (type == FILE_INODE && !test_opt(sbi, DATA_FLUSH))
823 spin_lock(&sbi->inode_lock[type]);
824 __remove_dirty_inode(inode, type);
825 spin_unlock(&sbi->inode_lock[type]);
828 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
830 struct list_head *head;
832 struct f2fs_inode_info *fi;
833 bool is_dir = (type == DIR_INODE);
835 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
836 get_pages(sbi, is_dir ?
837 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
839 if (unlikely(f2fs_cp_error(sbi)))
842 spin_lock(&sbi->inode_lock[type]);
844 head = &sbi->inode_list[type];
845 if (list_empty(head)) {
846 spin_unlock(&sbi->inode_lock[type]);
847 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
848 get_pages(sbi, is_dir ?
849 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
852 fi = list_entry(head->next, struct f2fs_inode_info, dirty_list);
853 inode = igrab(&fi->vfs_inode);
854 spin_unlock(&sbi->inode_lock[type]);
856 filemap_fdatawrite(inode->i_mapping);
860 * We should submit bio, since it exists several
861 * wribacking dentry pages in the freeing inode.
863 f2fs_submit_merged_bio(sbi, DATA, WRITE);
869 int f2fs_sync_inode_meta(struct f2fs_sb_info *sbi)
871 struct list_head *head = &sbi->inode_list[DIRTY_META];
873 struct f2fs_inode_info *fi;
874 s64 total = get_pages(sbi, F2FS_DIRTY_IMETA);
877 if (unlikely(f2fs_cp_error(sbi)))
880 spin_lock(&sbi->inode_lock[DIRTY_META]);
881 if (list_empty(head)) {
882 spin_unlock(&sbi->inode_lock[DIRTY_META]);
885 fi = list_entry(head->next, struct f2fs_inode_info,
887 inode = igrab(&fi->vfs_inode);
888 spin_unlock(&sbi->inode_lock[DIRTY_META]);
890 update_inode_page(inode);
898 * Freeze all the FS-operations for checkpoint.
900 static int block_operations(struct f2fs_sb_info *sbi)
902 struct writeback_control wbc = {
903 .sync_mode = WB_SYNC_ALL,
904 .nr_to_write = LONG_MAX,
907 struct blk_plug plug;
910 blk_start_plug(&plug);
914 /* write all the dirty dentry pages */
915 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
916 f2fs_unlock_all(sbi);
917 err = sync_dirty_inodes(sbi, DIR_INODE);
920 goto retry_flush_dents;
923 if (get_pages(sbi, F2FS_DIRTY_IMETA)) {
924 f2fs_unlock_all(sbi);
925 err = f2fs_sync_inode_meta(sbi);
928 goto retry_flush_dents;
932 * POR: we should ensure that there are no dirty node pages
933 * until finishing nat/sit flush.
936 down_write(&sbi->node_write);
938 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
939 up_write(&sbi->node_write);
940 err = sync_node_pages(sbi, &wbc);
942 f2fs_unlock_all(sbi);
945 goto retry_flush_nodes;
948 blk_finish_plug(&plug);
952 static void unblock_operations(struct f2fs_sb_info *sbi)
954 up_write(&sbi->node_write);
956 build_free_nids(sbi);
957 f2fs_unlock_all(sbi);
960 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
965 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
967 if (!atomic_read(&sbi->nr_wb_bios))
970 io_schedule_timeout(5*HZ);
972 finish_wait(&sbi->cp_wait, &wait);
975 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
977 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
978 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
979 struct f2fs_nm_info *nm_i = NM_I(sbi);
980 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
981 nid_t last_nid = nm_i->next_scan_nid;
983 unsigned int data_sum_blocks, orphan_blocks;
986 int cp_payload_blks = __cp_payload(sbi);
987 block_t discard_blk = NEXT_FREE_BLKADDR(sbi, curseg);
988 bool invalidate = false;
989 struct super_block *sb = sbi->sb;
990 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
994 * This avoids to conduct wrong roll-forward operations and uses
995 * metapages, so should be called prior to sync_meta_pages below.
997 if (!test_opt(sbi, LFS) && discard_next_dnode(sbi, discard_blk))
1000 /* Flush all the NAT/SIT pages */
1001 while (get_pages(sbi, F2FS_DIRTY_META)) {
1002 sync_meta_pages(sbi, META, LONG_MAX);
1003 if (unlikely(f2fs_cp_error(sbi)))
1007 next_free_nid(sbi, &last_nid);
1011 * version number is already updated
1013 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
1014 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
1015 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
1016 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
1017 ckpt->cur_node_segno[i] =
1018 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
1019 ckpt->cur_node_blkoff[i] =
1020 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
1021 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
1022 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
1024 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
1025 ckpt->cur_data_segno[i] =
1026 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
1027 ckpt->cur_data_blkoff[i] =
1028 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
1029 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
1030 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
1033 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
1034 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
1035 ckpt->next_free_nid = cpu_to_le32(last_nid);
1037 /* 2 cp + n data seg summary + orphan inode blocks */
1038 data_sum_blocks = npages_for_summary_flush(sbi, false);
1039 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
1040 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1042 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
1044 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
1045 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
1048 if (__remain_node_summaries(cpc->reason))
1049 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
1050 cp_payload_blks + data_sum_blocks +
1051 orphan_blocks + NR_CURSEG_NODE_TYPE);
1053 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
1054 cp_payload_blks + data_sum_blocks +
1057 if (cpc->reason == CP_UMOUNT)
1058 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1060 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1062 if (cpc->reason == CP_FASTBOOT)
1063 set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1065 clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1068 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1070 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1072 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1073 set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1075 /* update SIT/NAT bitmap */
1076 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1077 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1079 crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1080 *((__le32 *)((unsigned char *)ckpt +
1081 le32_to_cpu(ckpt->checksum_offset)))
1082 = cpu_to_le32(crc32);
1084 start_blk = __start_cp_addr(sbi);
1086 /* need to wait for end_io results */
1087 wait_on_all_pages_writeback(sbi);
1088 if (unlikely(f2fs_cp_error(sbi)))
1091 /* write out checkpoint buffer at block 0 */
1092 update_meta_page(sbi, ckpt, start_blk++);
1094 for (i = 1; i < 1 + cp_payload_blks; i++)
1095 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1099 write_orphan_inodes(sbi, start_blk);
1100 start_blk += orphan_blocks;
1103 write_data_summaries(sbi, start_blk);
1104 start_blk += data_sum_blocks;
1106 /* Record write statistics in the hot node summary */
1107 kbytes_written = sbi->kbytes_written;
1108 if (sb->s_bdev->bd_part)
1109 kbytes_written += BD_PART_WRITTEN(sbi);
1111 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1113 if (__remain_node_summaries(cpc->reason)) {
1114 write_node_summaries(sbi, start_blk);
1115 start_blk += NR_CURSEG_NODE_TYPE;
1118 /* writeout checkpoint block */
1119 update_meta_page(sbi, ckpt, start_blk);
1121 /* wait for previous submitted node/meta pages writeback */
1122 wait_on_all_pages_writeback(sbi);
1124 if (unlikely(f2fs_cp_error(sbi)))
1127 filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1128 filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1130 /* update user_block_counts */
1131 sbi->last_valid_block_count = sbi->total_valid_block_count;
1132 percpu_counter_set(&sbi->alloc_valid_block_count, 0);
1134 /* Here, we only have one bio having CP pack */
1135 sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1137 /* wait for previous submitted meta pages writeback */
1138 wait_on_all_pages_writeback(sbi);
1141 * invalidate meta page which is used temporarily for zeroing out
1142 * block at the end of warm node chain.
1145 invalidate_mapping_pages(META_MAPPING(sbi), discard_blk,
1148 release_ino_entry(sbi, false);
1150 if (unlikely(f2fs_cp_error(sbi)))
1153 clear_prefree_segments(sbi, cpc);
1154 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1160 * We guarantee that this checkpoint procedure will not fail.
1162 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1164 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1165 unsigned long long ckpt_ver;
1168 mutex_lock(&sbi->cp_mutex);
1170 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1171 (cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1172 (cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1174 if (unlikely(f2fs_cp_error(sbi))) {
1178 if (f2fs_readonly(sbi->sb)) {
1183 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1185 err = block_operations(sbi);
1189 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1191 f2fs_flush_merged_bios(sbi);
1194 * update checkpoint pack index
1195 * Increase the version number so that
1196 * SIT entries and seg summaries are written at correct place
1198 ckpt_ver = cur_cp_version(ckpt);
1199 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1201 /* write cached NAT/SIT entries to NAT/SIT area */
1202 flush_nat_entries(sbi);
1203 flush_sit_entries(sbi, cpc);
1205 /* unlock all the fs_lock[] in do_checkpoint() */
1206 err = do_checkpoint(sbi, cpc);
1208 unblock_operations(sbi);
1209 stat_inc_cp_count(sbi->stat_info);
1211 if (cpc->reason == CP_RECOVERY)
1212 f2fs_msg(sbi->sb, KERN_NOTICE,
1213 "checkpoint: version = %llx", ckpt_ver);
1215 /* do checkpoint periodically */
1216 f2fs_update_time(sbi, CP_TIME);
1217 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1219 mutex_unlock(&sbi->cp_mutex);
1223 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1227 for (i = 0; i < MAX_INO_ENTRY; i++) {
1228 struct inode_management *im = &sbi->im[i];
1230 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1231 spin_lock_init(&im->ino_lock);
1232 INIT_LIST_HEAD(&im->ino_list);
1236 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1237 NR_CURSEG_TYPE - __cp_payload(sbi)) *
1238 F2FS_ORPHANS_PER_BLOCK;
1241 int __init create_checkpoint_caches(void)
1243 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1244 sizeof(struct ino_entry));
1245 if (!ino_entry_slab)
1247 inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1248 sizeof(struct inode_entry));
1249 if (!inode_entry_slab) {
1250 kmem_cache_destroy(ino_entry_slab);
1256 void destroy_checkpoint_caches(void)
1258 kmem_cache_destroy(ino_entry_slab);
1259 kmem_cache_destroy(inode_entry_slab);