4 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5 * http://www.samsung.com/
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License version 2 as
9 * published by the Free Software Foundation.
12 #include <linux/bio.h>
13 #include <linux/mpage.h>
14 #include <linux/writeback.h>
15 #include <linux/blkdev.h>
16 #include <linux/f2fs_fs.h>
17 #include <linux/pagevec.h>
18 #include <linux/swap.h>
24 #include <trace/events/f2fs.h>
26 static struct kmem_cache *ino_entry_slab;
27 struct kmem_cache *inode_entry_slab;
30 * We guarantee no failure on the returned page.
32 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
34 struct address_space *mapping = META_MAPPING(sbi);
35 struct page *page = NULL;
37 page = f2fs_grab_cache_page(mapping, index, false);
42 f2fs_wait_on_page_writeback(page, META, true);
43 SetPageUptodate(page);
48 * We guarantee no failure on the returned page.
50 static struct page *__get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index,
53 struct address_space *mapping = META_MAPPING(sbi);
55 struct f2fs_io_info fio = {
58 .rw = READ_SYNC | REQ_META | REQ_PRIO,
61 .encrypted_page = NULL,
64 if (unlikely(!is_meta))
67 page = f2fs_grab_cache_page(mapping, index, false);
72 if (PageUptodate(page))
77 if (f2fs_submit_page_bio(&fio)) {
78 f2fs_put_page(page, 1);
83 if (unlikely(page->mapping != mapping)) {
84 f2fs_put_page(page, 1);
89 * if there is any IO error when accessing device, make our filesystem
90 * readonly and make sure do not write checkpoint with non-uptodate
93 if (unlikely(!PageUptodate(page)))
94 f2fs_stop_checkpoint(sbi);
99 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
101 return __get_meta_page(sbi, index, true);
105 struct page *get_tmp_page(struct f2fs_sb_info *sbi, pgoff_t index)
107 return __get_meta_page(sbi, index, false);
110 bool is_valid_blkaddr(struct f2fs_sb_info *sbi, block_t blkaddr, int type)
116 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
120 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
121 blkaddr < SM_I(sbi)->ssa_blkaddr))
125 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
126 blkaddr < __start_cp_addr(sbi)))
130 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
131 blkaddr < MAIN_BLKADDR(sbi)))
142 * Readahead CP/NAT/SIT/SSA pages
144 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages,
148 block_t blkno = start;
149 struct f2fs_io_info fio = {
152 .rw = sync ? (READ_SYNC | REQ_META | REQ_PRIO) : READA,
153 .encrypted_page = NULL,
155 struct blk_plug plug;
157 if (unlikely(type == META_POR))
160 blk_start_plug(&plug);
161 for (; nrpages-- > 0; blkno++) {
163 if (!is_valid_blkaddr(sbi, blkno, type))
168 if (unlikely(blkno >=
169 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
171 /* get nat block addr */
172 fio.new_blkaddr = current_nat_addr(sbi,
173 blkno * NAT_ENTRY_PER_BLOCK);
176 /* get sit block addr */
177 fio.new_blkaddr = current_sit_addr(sbi,
178 blkno * SIT_ENTRY_PER_BLOCK);
183 fio.new_blkaddr = blkno;
189 page = f2fs_grab_cache_page(META_MAPPING(sbi),
190 fio.new_blkaddr, false);
193 if (PageUptodate(page)) {
194 f2fs_put_page(page, 1);
199 fio.old_blkaddr = fio.new_blkaddr;
200 f2fs_submit_page_mbio(&fio);
201 f2fs_put_page(page, 0);
204 f2fs_submit_merged_bio(sbi, META, READ);
205 blk_finish_plug(&plug);
206 return blkno - start;
209 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
212 bool readahead = false;
214 page = find_get_page(META_MAPPING(sbi), index);
215 if (!page || !PageUptodate(page))
217 f2fs_put_page(page, 0);
220 ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR, true);
223 static int f2fs_write_meta_page(struct page *page,
224 struct writeback_control *wbc)
226 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
228 trace_f2fs_writepage(page, META);
230 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
232 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
234 if (unlikely(f2fs_cp_error(sbi)))
237 write_meta_page(sbi, page);
238 dec_page_count(sbi, F2FS_DIRTY_META);
240 if (wbc->for_reclaim)
241 f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, META, WRITE);
245 if (unlikely(f2fs_cp_error(sbi)))
246 f2fs_submit_merged_bio(sbi, META, WRITE);
251 redirty_page_for_writepage(wbc, page);
252 return AOP_WRITEPAGE_ACTIVATE;
255 static int f2fs_write_meta_pages(struct address_space *mapping,
256 struct writeback_control *wbc)
258 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
261 /* collect a number of dirty meta pages and write together */
262 if (wbc->for_kupdate ||
263 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
266 trace_f2fs_writepages(mapping->host, wbc, META);
268 /* if mounting is failed, skip writing node pages */
269 mutex_lock(&sbi->cp_mutex);
270 diff = nr_pages_to_write(sbi, META, wbc);
271 written = sync_meta_pages(sbi, META, wbc->nr_to_write);
272 mutex_unlock(&sbi->cp_mutex);
273 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
277 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
278 trace_f2fs_writepages(mapping->host, wbc, META);
282 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
285 struct address_space *mapping = META_MAPPING(sbi);
286 pgoff_t index = 0, end = ULONG_MAX, prev = ULONG_MAX;
289 struct writeback_control wbc = {
292 struct blk_plug plug;
294 pagevec_init(&pvec, 0);
296 blk_start_plug(&plug);
298 while (index <= end) {
300 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
302 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
303 if (unlikely(nr_pages == 0))
306 for (i = 0; i < nr_pages; i++) {
307 struct page *page = pvec.pages[i];
309 if (prev == ULONG_MAX)
310 prev = page->index - 1;
311 if (nr_to_write != LONG_MAX && page->index != prev + 1) {
312 pagevec_release(&pvec);
318 if (unlikely(page->mapping != mapping)) {
323 if (!PageDirty(page)) {
324 /* someone wrote it for us */
325 goto continue_unlock;
328 f2fs_wait_on_page_writeback(page, META, true);
330 BUG_ON(PageWriteback(page));
331 if (!clear_page_dirty_for_io(page))
332 goto continue_unlock;
334 if (mapping->a_ops->writepage(page, &wbc)) {
340 if (unlikely(nwritten >= nr_to_write))
343 pagevec_release(&pvec);
348 f2fs_submit_merged_bio(sbi, type, WRITE);
350 blk_finish_plug(&plug);
355 static int f2fs_set_meta_page_dirty(struct page *page)
357 trace_f2fs_set_page_dirty(page, META);
359 SetPageUptodate(page);
360 if (!PageDirty(page)) {
361 __set_page_dirty_nobuffers(page);
362 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
363 SetPagePrivate(page);
364 f2fs_trace_pid(page);
370 const struct address_space_operations f2fs_meta_aops = {
371 .writepage = f2fs_write_meta_page,
372 .writepages = f2fs_write_meta_pages,
373 .set_page_dirty = f2fs_set_meta_page_dirty,
374 .invalidatepage = f2fs_invalidate_page,
375 .releasepage = f2fs_release_page,
378 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
380 struct inode_management *im = &sbi->im[type];
381 struct ino_entry *e, *tmp;
383 tmp = f2fs_kmem_cache_alloc(ino_entry_slab, GFP_NOFS);
385 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
387 spin_lock(&im->ino_lock);
388 e = radix_tree_lookup(&im->ino_root, ino);
391 if (radix_tree_insert(&im->ino_root, ino, e)) {
392 spin_unlock(&im->ino_lock);
393 radix_tree_preload_end();
396 memset(e, 0, sizeof(struct ino_entry));
399 list_add_tail(&e->list, &im->ino_list);
400 if (type != ORPHAN_INO)
403 spin_unlock(&im->ino_lock);
404 radix_tree_preload_end();
407 kmem_cache_free(ino_entry_slab, tmp);
410 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
412 struct inode_management *im = &sbi->im[type];
415 spin_lock(&im->ino_lock);
416 e = radix_tree_lookup(&im->ino_root, ino);
419 radix_tree_delete(&im->ino_root, ino);
421 spin_unlock(&im->ino_lock);
422 kmem_cache_free(ino_entry_slab, e);
425 spin_unlock(&im->ino_lock);
428 void add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
430 /* add new dirty ino entry into list */
431 __add_ino_entry(sbi, ino, type);
434 void remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
436 /* remove dirty ino entry from list */
437 __remove_ino_entry(sbi, ino, type);
440 /* mode should be APPEND_INO or UPDATE_INO */
441 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
443 struct inode_management *im = &sbi->im[mode];
446 spin_lock(&im->ino_lock);
447 e = radix_tree_lookup(&im->ino_root, ino);
448 spin_unlock(&im->ino_lock);
449 return e ? true : false;
452 void release_ino_entry(struct f2fs_sb_info *sbi, bool all)
454 struct ino_entry *e, *tmp;
457 for (i = all ? ORPHAN_INO: APPEND_INO; i <= UPDATE_INO; i++) {
458 struct inode_management *im = &sbi->im[i];
460 spin_lock(&im->ino_lock);
461 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
463 radix_tree_delete(&im->ino_root, e->ino);
464 kmem_cache_free(ino_entry_slab, e);
467 spin_unlock(&im->ino_lock);
471 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
473 struct inode_management *im = &sbi->im[ORPHAN_INO];
476 spin_lock(&im->ino_lock);
478 #ifdef CONFIG_F2FS_FAULT_INJECTION
479 if (time_to_inject(FAULT_ORPHAN)) {
480 spin_unlock(&im->ino_lock);
484 if (unlikely(im->ino_num >= sbi->max_orphans))
488 spin_unlock(&im->ino_lock);
493 void release_orphan_inode(struct f2fs_sb_info *sbi)
495 struct inode_management *im = &sbi->im[ORPHAN_INO];
497 spin_lock(&im->ino_lock);
498 f2fs_bug_on(sbi, im->ino_num == 0);
500 spin_unlock(&im->ino_lock);
503 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
505 /* add new orphan ino entry into list */
506 __add_ino_entry(sbi, ino, ORPHAN_INO);
509 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
511 /* remove orphan entry from orphan list */
512 __remove_ino_entry(sbi, ino, ORPHAN_INO);
515 static int recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
519 inode = f2fs_iget(sbi->sb, ino);
522 * there should be a bug that we can't find the entry
525 f2fs_bug_on(sbi, PTR_ERR(inode) == -ENOENT);
526 return PTR_ERR(inode);
531 /* truncate all the data during iput */
536 int recover_orphan_inodes(struct f2fs_sb_info *sbi)
538 block_t start_blk, orphan_blocks, i, j;
541 if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
544 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
545 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
547 ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP, true);
549 for (i = 0; i < orphan_blocks; i++) {
550 struct page *page = get_meta_page(sbi, start_blk + i);
551 struct f2fs_orphan_block *orphan_blk;
553 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
554 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
555 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
556 err = recover_orphan_inode(sbi, ino);
558 f2fs_put_page(page, 1);
562 f2fs_put_page(page, 1);
564 /* clear Orphan Flag */
565 clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
569 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
571 struct list_head *head;
572 struct f2fs_orphan_block *orphan_blk = NULL;
573 unsigned int nentries = 0;
574 unsigned short index = 1;
575 unsigned short orphan_blocks;
576 struct page *page = NULL;
577 struct ino_entry *orphan = NULL;
578 struct inode_management *im = &sbi->im[ORPHAN_INO];
580 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
583 * we don't need to do spin_lock(&im->ino_lock) here, since all the
584 * orphan inode operations are covered under f2fs_lock_op().
585 * And, spin_lock should be avoided due to page operations below.
587 head = &im->ino_list;
589 /* loop for each orphan inode entry and write them in Jornal block */
590 list_for_each_entry(orphan, head, list) {
592 page = grab_meta_page(sbi, start_blk++);
594 (struct f2fs_orphan_block *)page_address(page);
595 memset(orphan_blk, 0, sizeof(*orphan_blk));
598 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
600 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
602 * an orphan block is full of 1020 entries,
603 * then we need to flush current orphan blocks
604 * and bring another one in memory
606 orphan_blk->blk_addr = cpu_to_le16(index);
607 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
608 orphan_blk->entry_count = cpu_to_le32(nentries);
609 set_page_dirty(page);
610 f2fs_put_page(page, 1);
618 orphan_blk->blk_addr = cpu_to_le16(index);
619 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
620 orphan_blk->entry_count = cpu_to_le32(nentries);
621 set_page_dirty(page);
622 f2fs_put_page(page, 1);
626 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
627 block_t cp_addr, unsigned long long *version)
629 struct page *cp_page_1, *cp_page_2 = NULL;
630 unsigned long blk_size = sbi->blocksize;
631 struct f2fs_checkpoint *cp_block;
632 unsigned long long cur_version = 0, pre_version = 0;
636 /* Read the 1st cp block in this CP pack */
637 cp_page_1 = get_meta_page(sbi, cp_addr);
639 /* get the version number */
640 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
641 crc_offset = le32_to_cpu(cp_block->checksum_offset);
642 if (crc_offset >= blk_size)
645 crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
646 if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
649 pre_version = cur_cp_version(cp_block);
651 /* Read the 2nd cp block in this CP pack */
652 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
653 cp_page_2 = get_meta_page(sbi, cp_addr);
655 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
656 crc_offset = le32_to_cpu(cp_block->checksum_offset);
657 if (crc_offset >= blk_size)
660 crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
661 if (!f2fs_crc_valid(sbi, crc, cp_block, crc_offset))
664 cur_version = cur_cp_version(cp_block);
666 if (cur_version == pre_version) {
667 *version = cur_version;
668 f2fs_put_page(cp_page_2, 1);
672 f2fs_put_page(cp_page_2, 1);
674 f2fs_put_page(cp_page_1, 1);
678 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
680 struct f2fs_checkpoint *cp_block;
681 struct f2fs_super_block *fsb = sbi->raw_super;
682 struct page *cp1, *cp2, *cur_page;
683 unsigned long blk_size = sbi->blocksize;
684 unsigned long long cp1_version = 0, cp2_version = 0;
685 unsigned long long cp_start_blk_no;
686 unsigned int cp_blks = 1 + __cp_payload(sbi);
690 sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
694 * Finding out valid cp block involves read both
695 * sets( cp pack1 and cp pack 2)
697 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
698 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
700 /* The second checkpoint pack should start at the next segment */
701 cp_start_blk_no += ((unsigned long long)1) <<
702 le32_to_cpu(fsb->log_blocks_per_seg);
703 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
706 if (ver_after(cp2_version, cp1_version))
718 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
719 memcpy(sbi->ckpt, cp_block, blk_size);
721 /* Sanity checking of checkpoint */
722 if (sanity_check_ckpt(sbi))
728 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
730 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
732 for (i = 1; i < cp_blks; i++) {
733 void *sit_bitmap_ptr;
734 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
736 cur_page = get_meta_page(sbi, cp_blk_no + i);
737 sit_bitmap_ptr = page_address(cur_page);
738 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
739 f2fs_put_page(cur_page, 1);
742 f2fs_put_page(cp1, 1);
743 f2fs_put_page(cp2, 1);
751 static void __add_dirty_inode(struct inode *inode, enum inode_type type)
753 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
754 struct f2fs_inode_info *fi = F2FS_I(inode);
755 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
757 if (is_inode_flag_set(fi, flag))
760 set_inode_flag(fi, flag);
761 list_add_tail(&fi->dirty_list, &sbi->inode_list[type]);
762 stat_inc_dirty_inode(sbi, type);
765 static void __remove_dirty_inode(struct inode *inode, enum inode_type type)
767 struct f2fs_inode_info *fi = F2FS_I(inode);
768 int flag = (type == DIR_INODE) ? FI_DIRTY_DIR : FI_DIRTY_FILE;
770 if (get_dirty_pages(inode) ||
771 !is_inode_flag_set(F2FS_I(inode), flag))
774 list_del_init(&fi->dirty_list);
775 clear_inode_flag(fi, flag);
776 stat_dec_dirty_inode(F2FS_I_SB(inode), type);
779 void update_dirty_page(struct inode *inode, struct page *page)
781 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
782 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
784 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
785 !S_ISLNK(inode->i_mode))
788 spin_lock(&sbi->inode_lock[type]);
789 __add_dirty_inode(inode, type);
790 spin_unlock(&sbi->inode_lock[type]);
792 inode_inc_dirty_pages(inode);
793 SetPagePrivate(page);
794 f2fs_trace_pid(page);
797 void remove_dirty_inode(struct inode *inode)
799 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
800 enum inode_type type = S_ISDIR(inode->i_mode) ? DIR_INODE : FILE_INODE;
802 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode) &&
803 !S_ISLNK(inode->i_mode))
806 spin_lock(&sbi->inode_lock[type]);
807 __remove_dirty_inode(inode, type);
808 spin_unlock(&sbi->inode_lock[type]);
811 int sync_dirty_inodes(struct f2fs_sb_info *sbi, enum inode_type type)
813 struct list_head *head;
815 struct f2fs_inode_info *fi;
816 bool is_dir = (type == DIR_INODE);
818 trace_f2fs_sync_dirty_inodes_enter(sbi->sb, is_dir,
819 get_pages(sbi, is_dir ?
820 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
822 if (unlikely(f2fs_cp_error(sbi)))
825 spin_lock(&sbi->inode_lock[type]);
827 head = &sbi->inode_list[type];
828 if (list_empty(head)) {
829 spin_unlock(&sbi->inode_lock[type]);
830 trace_f2fs_sync_dirty_inodes_exit(sbi->sb, is_dir,
831 get_pages(sbi, is_dir ?
832 F2FS_DIRTY_DENTS : F2FS_DIRTY_DATA));
835 fi = list_entry(head->next, struct f2fs_inode_info, dirty_list);
836 inode = igrab(&fi->vfs_inode);
837 spin_unlock(&sbi->inode_lock[type]);
839 filemap_fdatawrite(inode->i_mapping);
843 * We should submit bio, since it exists several
844 * wribacking dentry pages in the freeing inode.
846 f2fs_submit_merged_bio(sbi, DATA, WRITE);
853 * Freeze all the FS-operations for checkpoint.
855 static int block_operations(struct f2fs_sb_info *sbi)
857 struct writeback_control wbc = {
858 .sync_mode = WB_SYNC_ALL,
859 .nr_to_write = LONG_MAX,
862 struct blk_plug plug;
865 blk_start_plug(&plug);
869 /* write all the dirty dentry pages */
870 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
871 f2fs_unlock_all(sbi);
872 err = sync_dirty_inodes(sbi, DIR_INODE);
875 goto retry_flush_dents;
879 * POR: we should ensure that there are no dirty node pages
880 * until finishing nat/sit flush.
883 down_write(&sbi->node_write);
885 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
886 up_write(&sbi->node_write);
887 err = sync_node_pages(sbi, &wbc);
889 f2fs_unlock_all(sbi);
892 goto retry_flush_nodes;
895 blk_finish_plug(&plug);
899 static void unblock_operations(struct f2fs_sb_info *sbi)
901 up_write(&sbi->node_write);
902 f2fs_unlock_all(sbi);
905 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
910 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
912 if (!get_pages(sbi, F2FS_WRITEBACK))
915 io_schedule_timeout(5*HZ);
917 finish_wait(&sbi->cp_wait, &wait);
920 static int do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
922 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
923 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
924 struct f2fs_nm_info *nm_i = NM_I(sbi);
925 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
926 nid_t last_nid = nm_i->next_scan_nid;
928 unsigned int data_sum_blocks, orphan_blocks;
931 int cp_payload_blks = __cp_payload(sbi);
932 block_t discard_blk = NEXT_FREE_BLKADDR(sbi, curseg);
933 bool invalidate = false;
934 struct super_block *sb = sbi->sb;
935 struct curseg_info *seg_i = CURSEG_I(sbi, CURSEG_HOT_NODE);
939 * This avoids to conduct wrong roll-forward operations and uses
940 * metapages, so should be called prior to sync_meta_pages below.
942 if (discard_next_dnode(sbi, discard_blk))
945 /* Flush all the NAT/SIT pages */
946 while (get_pages(sbi, F2FS_DIRTY_META)) {
947 sync_meta_pages(sbi, META, LONG_MAX);
948 if (unlikely(f2fs_cp_error(sbi)))
952 next_free_nid(sbi, &last_nid);
956 * version number is already updated
958 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
959 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
960 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
961 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
962 ckpt->cur_node_segno[i] =
963 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
964 ckpt->cur_node_blkoff[i] =
965 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
966 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
967 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
969 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
970 ckpt->cur_data_segno[i] =
971 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
972 ckpt->cur_data_blkoff[i] =
973 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
974 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
975 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
978 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
979 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
980 ckpt->next_free_nid = cpu_to_le32(last_nid);
982 /* 2 cp + n data seg summary + orphan inode blocks */
983 data_sum_blocks = npages_for_summary_flush(sbi, false);
984 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
985 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
987 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
989 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
990 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
993 if (__remain_node_summaries(cpc->reason))
994 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
995 cp_payload_blks + data_sum_blocks +
996 orphan_blocks + NR_CURSEG_NODE_TYPE);
998 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
999 cp_payload_blks + data_sum_blocks +
1002 if (cpc->reason == CP_UMOUNT)
1003 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1005 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
1007 if (cpc->reason == CP_FASTBOOT)
1008 set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1010 clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
1013 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1015 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
1017 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
1018 set_ckpt_flags(ckpt, CP_FSCK_FLAG);
1020 /* update SIT/NAT bitmap */
1021 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
1022 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
1024 crc32 = f2fs_crc32(sbi, ckpt, le32_to_cpu(ckpt->checksum_offset));
1025 *((__le32 *)((unsigned char *)ckpt +
1026 le32_to_cpu(ckpt->checksum_offset)))
1027 = cpu_to_le32(crc32);
1029 start_blk = __start_cp_addr(sbi);
1031 /* need to wait for end_io results */
1032 wait_on_all_pages_writeback(sbi);
1033 if (unlikely(f2fs_cp_error(sbi)))
1036 /* write out checkpoint buffer at block 0 */
1037 update_meta_page(sbi, ckpt, start_blk++);
1039 for (i = 1; i < 1 + cp_payload_blks; i++)
1040 update_meta_page(sbi, (char *)ckpt + i * F2FS_BLKSIZE,
1044 write_orphan_inodes(sbi, start_blk);
1045 start_blk += orphan_blocks;
1048 write_data_summaries(sbi, start_blk);
1049 start_blk += data_sum_blocks;
1051 /* Record write statistics in the hot node summary */
1052 kbytes_written = sbi->kbytes_written;
1053 if (sb->s_bdev->bd_part)
1054 kbytes_written += BD_PART_WRITTEN(sbi);
1056 seg_i->journal->info.kbytes_written = cpu_to_le64(kbytes_written);
1058 if (__remain_node_summaries(cpc->reason)) {
1059 write_node_summaries(sbi, start_blk);
1060 start_blk += NR_CURSEG_NODE_TYPE;
1063 /* writeout checkpoint block */
1064 update_meta_page(sbi, ckpt, start_blk);
1066 /* wait for previous submitted node/meta pages writeback */
1067 wait_on_all_pages_writeback(sbi);
1069 if (unlikely(f2fs_cp_error(sbi)))
1072 filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LLONG_MAX);
1073 filemap_fdatawait_range(META_MAPPING(sbi), 0, LLONG_MAX);
1075 /* update user_block_counts */
1076 sbi->last_valid_block_count = sbi->total_valid_block_count;
1077 sbi->alloc_valid_block_count = 0;
1079 /* Here, we only have one bio having CP pack */
1080 sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1082 /* wait for previous submitted meta pages writeback */
1083 wait_on_all_pages_writeback(sbi);
1086 * invalidate meta page which is used temporarily for zeroing out
1087 * block at the end of warm node chain.
1090 invalidate_mapping_pages(META_MAPPING(sbi), discard_blk,
1093 release_ino_entry(sbi, false);
1095 if (unlikely(f2fs_cp_error(sbi)))
1098 clear_prefree_segments(sbi, cpc);
1099 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1105 * We guarantee that this checkpoint procedure will not fail.
1107 int write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1109 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1110 unsigned long long ckpt_ver;
1113 mutex_lock(&sbi->cp_mutex);
1115 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1116 (cpc->reason == CP_FASTBOOT || cpc->reason == CP_SYNC ||
1117 (cpc->reason == CP_DISCARD && !sbi->discard_blks)))
1119 if (unlikely(f2fs_cp_error(sbi))) {
1123 if (f2fs_readonly(sbi->sb)) {
1128 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1130 err = block_operations(sbi);
1134 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1136 f2fs_flush_merged_bios(sbi);
1139 * update checkpoint pack index
1140 * Increase the version number so that
1141 * SIT entries and seg summaries are written at correct place
1143 ckpt_ver = cur_cp_version(ckpt);
1144 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1146 /* write cached NAT/SIT entries to NAT/SIT area */
1147 flush_nat_entries(sbi);
1148 flush_sit_entries(sbi, cpc);
1150 /* unlock all the fs_lock[] in do_checkpoint() */
1151 err = do_checkpoint(sbi, cpc);
1153 unblock_operations(sbi);
1154 stat_inc_cp_count(sbi->stat_info);
1156 if (cpc->reason == CP_RECOVERY)
1157 f2fs_msg(sbi->sb, KERN_NOTICE,
1158 "checkpoint: version = %llx", ckpt_ver);
1160 /* do checkpoint periodically */
1161 f2fs_update_time(sbi, CP_TIME);
1162 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1164 mutex_unlock(&sbi->cp_mutex);
1168 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1172 for (i = 0; i < MAX_INO_ENTRY; i++) {
1173 struct inode_management *im = &sbi->im[i];
1175 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1176 spin_lock_init(&im->ino_lock);
1177 INIT_LIST_HEAD(&im->ino_list);
1181 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1182 NR_CURSEG_TYPE - __cp_payload(sbi)) *
1183 F2FS_ORPHANS_PER_BLOCK;
1186 int __init create_checkpoint_caches(void)
1188 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1189 sizeof(struct ino_entry));
1190 if (!ino_entry_slab)
1192 inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1193 sizeof(struct inode_entry));
1194 if (!inode_entry_slab) {
1195 kmem_cache_destroy(ino_entry_slab);
1201 void destroy_checkpoint_caches(void)
1203 kmem_cache_destroy(ino_entry_slab);
1204 kmem_cache_destroy(inode_entry_slab);