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>
23 #include <trace/events/f2fs.h>
25 static struct kmem_cache *orphan_entry_slab;
26 static struct kmem_cache *inode_entry_slab;
29 * We guarantee no failure on the returned page.
31 struct page *grab_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
33 struct address_space *mapping = META_MAPPING(sbi);
34 struct page *page = NULL;
36 page = grab_cache_page_write_begin(mapping, index, AOP_FLAG_NOFS);
42 SetPageUptodate(page);
47 * We guarantee no failure on the returned page.
49 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
51 struct address_space *mapping = META_MAPPING(sbi);
54 page = grab_cache_page(mapping, index);
59 if (PageUptodate(page))
62 if (f2fs_submit_page_bio(sbi, page, index,
63 READ_SYNC | REQ_META | REQ_PRIO))
67 if (unlikely(page->mapping != mapping)) {
68 f2fs_put_page(page, 1);
72 mark_page_accessed(page);
76 inline int get_max_meta_blks(struct f2fs_sb_info *sbi, int type)
80 return NM_I(sbi)->max_nid / NAT_ENTRY_PER_BLOCK;
82 return SIT_BLK_CNT(sbi);
92 * Readahead CP/NAT/SIT/SSA pages
94 int ra_meta_pages(struct f2fs_sb_info *sbi, int start, int nrpages, int type)
96 block_t prev_blk_addr = 0;
99 int max_blks = get_max_meta_blks(sbi, type);
101 struct f2fs_io_info fio = {
103 .rw = READ_SYNC | REQ_META | REQ_PRIO
106 for (; nrpages-- > 0; blkno++) {
111 /* get nat block addr */
112 if (unlikely(blkno >= max_blks))
114 blk_addr = current_nat_addr(sbi,
115 blkno * NAT_ENTRY_PER_BLOCK);
118 /* get sit block addr */
119 if (unlikely(blkno >= max_blks))
121 blk_addr = current_sit_addr(sbi,
122 blkno * SIT_ENTRY_PER_BLOCK);
123 if (blkno != start && prev_blk_addr + 1 != blk_addr)
125 prev_blk_addr = blk_addr;
129 /* get ssa/cp block addr */
136 page = grab_cache_page(META_MAPPING(sbi), blk_addr);
139 if (PageUptodate(page)) {
140 mark_page_accessed(page);
141 f2fs_put_page(page, 1);
145 f2fs_submit_page_mbio(sbi, page, blk_addr, &fio);
146 mark_page_accessed(page);
147 f2fs_put_page(page, 0);
150 f2fs_submit_merged_bio(sbi, META, READ);
151 return blkno - start;
154 static int f2fs_write_meta_page(struct page *page,
155 struct writeback_control *wbc)
157 struct inode *inode = page->mapping->host;
158 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
160 if (unlikely(sbi->por_doing))
162 if (wbc->for_reclaim)
165 /* Should not write any meta pages, if any IO error was occurred */
166 if (unlikely(is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ERROR_FLAG)))
169 f2fs_wait_on_page_writeback(page, META);
170 write_meta_page(sbi, page);
172 dec_page_count(sbi, F2FS_DIRTY_META);
177 dec_page_count(sbi, F2FS_DIRTY_META);
178 wbc->pages_skipped++;
179 account_page_redirty(page);
180 set_page_dirty(page);
181 return AOP_WRITEPAGE_ACTIVATE;
184 static int f2fs_write_meta_pages(struct address_space *mapping,
185 struct writeback_control *wbc)
187 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
190 /* collect a number of dirty meta pages and write together */
191 if (wbc->for_kupdate ||
192 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
195 /* if mounting is failed, skip writing node pages */
196 mutex_lock(&sbi->cp_mutex);
197 diff = nr_pages_to_write(sbi, META, wbc);
198 written = sync_meta_pages(sbi, META, wbc->nr_to_write);
199 mutex_unlock(&sbi->cp_mutex);
200 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
204 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
208 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
211 struct address_space *mapping = META_MAPPING(sbi);
212 pgoff_t index = 0, end = LONG_MAX;
215 struct writeback_control wbc = {
219 pagevec_init(&pvec, 0);
221 while (index <= end) {
223 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
225 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
226 if (unlikely(nr_pages == 0))
229 for (i = 0; i < nr_pages; i++) {
230 struct page *page = pvec.pages[i];
234 if (unlikely(page->mapping != mapping)) {
239 if (!PageDirty(page)) {
240 /* someone wrote it for us */
241 goto continue_unlock;
244 if (!clear_page_dirty_for_io(page))
245 goto continue_unlock;
247 if (f2fs_write_meta_page(page, &wbc)) {
252 if (unlikely(nwritten >= nr_to_write))
255 pagevec_release(&pvec);
260 f2fs_submit_merged_bio(sbi, type, WRITE);
265 static int f2fs_set_meta_page_dirty(struct page *page)
267 struct address_space *mapping = page->mapping;
268 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
270 trace_f2fs_set_page_dirty(page, META);
272 SetPageUptodate(page);
273 if (!PageDirty(page)) {
274 __set_page_dirty_nobuffers(page);
275 inc_page_count(sbi, F2FS_DIRTY_META);
281 const struct address_space_operations f2fs_meta_aops = {
282 .writepage = f2fs_write_meta_page,
283 .writepages = f2fs_write_meta_pages,
284 .set_page_dirty = f2fs_set_meta_page_dirty,
287 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
291 spin_lock(&sbi->orphan_inode_lock);
292 if (unlikely(sbi->n_orphans >= sbi->max_orphans))
296 spin_unlock(&sbi->orphan_inode_lock);
301 void release_orphan_inode(struct f2fs_sb_info *sbi)
303 spin_lock(&sbi->orphan_inode_lock);
304 f2fs_bug_on(sbi->n_orphans == 0);
306 spin_unlock(&sbi->orphan_inode_lock);
309 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
311 struct list_head *head, *this;
312 struct orphan_inode_entry *new = NULL, *orphan = NULL;
314 new = f2fs_kmem_cache_alloc(orphan_entry_slab, GFP_ATOMIC);
317 spin_lock(&sbi->orphan_inode_lock);
318 head = &sbi->orphan_inode_list;
319 list_for_each(this, head) {
320 orphan = list_entry(this, struct orphan_inode_entry, list);
321 if (orphan->ino == ino) {
322 spin_unlock(&sbi->orphan_inode_lock);
323 kmem_cache_free(orphan_entry_slab, new);
327 if (orphan->ino > ino)
332 /* add new_oentry into list which is sorted by inode number */
334 list_add(&new->list, this->prev);
336 list_add_tail(&new->list, head);
337 spin_unlock(&sbi->orphan_inode_lock);
340 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
342 struct list_head *head;
343 struct orphan_inode_entry *orphan;
345 spin_lock(&sbi->orphan_inode_lock);
346 head = &sbi->orphan_inode_list;
347 list_for_each_entry(orphan, head, list) {
348 if (orphan->ino == ino) {
349 list_del(&orphan->list);
350 f2fs_bug_on(sbi->n_orphans == 0);
352 spin_unlock(&sbi->orphan_inode_lock);
353 kmem_cache_free(orphan_entry_slab, orphan);
357 spin_unlock(&sbi->orphan_inode_lock);
360 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
362 struct inode *inode = f2fs_iget(sbi->sb, ino);
363 f2fs_bug_on(IS_ERR(inode));
366 /* truncate all the data during iput */
370 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
372 block_t start_blk, orphan_blkaddr, i, j;
374 if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
377 sbi->por_doing = true;
378 start_blk = __start_cp_addr(sbi) + 1;
379 orphan_blkaddr = __start_sum_addr(sbi) - 1;
381 ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP);
383 for (i = 0; i < orphan_blkaddr; i++) {
384 struct page *page = get_meta_page(sbi, start_blk + i);
385 struct f2fs_orphan_block *orphan_blk;
387 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
388 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
389 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
390 recover_orphan_inode(sbi, ino);
392 f2fs_put_page(page, 1);
394 /* clear Orphan Flag */
395 clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
396 sbi->por_doing = false;
400 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
402 struct list_head *head;
403 struct f2fs_orphan_block *orphan_blk = NULL;
404 unsigned int nentries = 0;
405 unsigned short index;
406 unsigned short orphan_blocks = (unsigned short)((sbi->n_orphans +
407 (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK);
408 struct page *page = NULL;
409 struct orphan_inode_entry *orphan = NULL;
411 for (index = 0; index < orphan_blocks; index++)
412 grab_meta_page(sbi, start_blk + index);
415 spin_lock(&sbi->orphan_inode_lock);
416 head = &sbi->orphan_inode_list;
418 /* loop for each orphan inode entry and write them in Jornal block */
419 list_for_each_entry(orphan, head, list) {
421 page = find_get_page(META_MAPPING(sbi), start_blk++);
424 (struct f2fs_orphan_block *)page_address(page);
425 memset(orphan_blk, 0, sizeof(*orphan_blk));
426 f2fs_put_page(page, 0);
429 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
431 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
433 * an orphan block is full of 1020 entries,
434 * then we need to flush current orphan blocks
435 * and bring another one in memory
437 orphan_blk->blk_addr = cpu_to_le16(index);
438 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
439 orphan_blk->entry_count = cpu_to_le32(nentries);
440 set_page_dirty(page);
441 f2fs_put_page(page, 1);
449 orphan_blk->blk_addr = cpu_to_le16(index);
450 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
451 orphan_blk->entry_count = cpu_to_le32(nentries);
452 set_page_dirty(page);
453 f2fs_put_page(page, 1);
456 spin_unlock(&sbi->orphan_inode_lock);
459 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
460 block_t cp_addr, unsigned long long *version)
462 struct page *cp_page_1, *cp_page_2 = NULL;
463 unsigned long blk_size = sbi->blocksize;
464 struct f2fs_checkpoint *cp_block;
465 unsigned long long cur_version = 0, pre_version = 0;
469 /* Read the 1st cp block in this CP pack */
470 cp_page_1 = get_meta_page(sbi, cp_addr);
472 /* get the version number */
473 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
474 crc_offset = le32_to_cpu(cp_block->checksum_offset);
475 if (crc_offset >= blk_size)
478 crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
479 if (!f2fs_crc_valid(crc, cp_block, crc_offset))
482 pre_version = cur_cp_version(cp_block);
484 /* Read the 2nd cp block in this CP pack */
485 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
486 cp_page_2 = get_meta_page(sbi, cp_addr);
488 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
489 crc_offset = le32_to_cpu(cp_block->checksum_offset);
490 if (crc_offset >= blk_size)
493 crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
494 if (!f2fs_crc_valid(crc, cp_block, crc_offset))
497 cur_version = cur_cp_version(cp_block);
499 if (cur_version == pre_version) {
500 *version = cur_version;
501 f2fs_put_page(cp_page_2, 1);
505 f2fs_put_page(cp_page_2, 1);
507 f2fs_put_page(cp_page_1, 1);
511 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
513 struct f2fs_checkpoint *cp_block;
514 struct f2fs_super_block *fsb = sbi->raw_super;
515 struct page *cp1, *cp2, *cur_page;
516 unsigned long blk_size = sbi->blocksize;
517 unsigned long long cp1_version = 0, cp2_version = 0;
518 unsigned long long cp_start_blk_no;
520 sbi->ckpt = kzalloc(blk_size, GFP_KERNEL);
524 * Finding out valid cp block involves read both
525 * sets( cp pack1 and cp pack 2)
527 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
528 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
530 /* The second checkpoint pack should start at the next segment */
531 cp_start_blk_no += ((unsigned long long)1) <<
532 le32_to_cpu(fsb->log_blocks_per_seg);
533 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
536 if (ver_after(cp2_version, cp1_version))
548 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
549 memcpy(sbi->ckpt, cp_block, blk_size);
551 f2fs_put_page(cp1, 1);
552 f2fs_put_page(cp2, 1);
560 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
562 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
563 struct list_head *head = &sbi->dir_inode_list;
564 struct list_head *this;
566 list_for_each(this, head) {
567 struct dir_inode_entry *entry;
568 entry = list_entry(this, struct dir_inode_entry, list);
569 if (unlikely(entry->inode == inode))
572 list_add_tail(&new->list, head);
573 stat_inc_dirty_dir(sbi);
577 void set_dirty_dir_page(struct inode *inode, struct page *page)
579 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
580 struct dir_inode_entry *new;
583 if (!S_ISDIR(inode->i_mode))
586 new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
588 INIT_LIST_HEAD(&new->list);
590 spin_lock(&sbi->dir_inode_lock);
591 ret = __add_dirty_inode(inode, new);
592 inode_inc_dirty_dents(inode);
593 SetPagePrivate(page);
594 spin_unlock(&sbi->dir_inode_lock);
597 kmem_cache_free(inode_entry_slab, new);
600 void add_dirty_dir_inode(struct inode *inode)
602 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
603 struct dir_inode_entry *new =
604 f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
608 INIT_LIST_HEAD(&new->list);
610 spin_lock(&sbi->dir_inode_lock);
611 ret = __add_dirty_inode(inode, new);
612 spin_unlock(&sbi->dir_inode_lock);
615 kmem_cache_free(inode_entry_slab, new);
618 void remove_dirty_dir_inode(struct inode *inode)
620 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
621 struct list_head *this, *head;
623 if (!S_ISDIR(inode->i_mode))
626 spin_lock(&sbi->dir_inode_lock);
627 if (get_dirty_dents(inode)) {
628 spin_unlock(&sbi->dir_inode_lock);
632 head = &sbi->dir_inode_list;
633 list_for_each(this, head) {
634 struct dir_inode_entry *entry;
635 entry = list_entry(this, struct dir_inode_entry, list);
636 if (entry->inode == inode) {
637 list_del(&entry->list);
638 stat_dec_dirty_dir(sbi);
639 spin_unlock(&sbi->dir_inode_lock);
640 kmem_cache_free(inode_entry_slab, entry);
644 spin_unlock(&sbi->dir_inode_lock);
647 /* Only from the recovery routine */
648 if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
649 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
654 struct inode *check_dirty_dir_inode(struct f2fs_sb_info *sbi, nid_t ino)
657 struct list_head *this, *head;
658 struct inode *inode = NULL;
660 spin_lock(&sbi->dir_inode_lock);
662 head = &sbi->dir_inode_list;
663 list_for_each(this, head) {
664 struct dir_inode_entry *entry;
665 entry = list_entry(this, struct dir_inode_entry, list);
666 if (entry->inode->i_ino == ino) {
667 inode = entry->inode;
671 spin_unlock(&sbi->dir_inode_lock);
675 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
677 struct list_head *head;
678 struct dir_inode_entry *entry;
681 spin_lock(&sbi->dir_inode_lock);
683 head = &sbi->dir_inode_list;
684 if (list_empty(head)) {
685 spin_unlock(&sbi->dir_inode_lock);
688 entry = list_entry(head->next, struct dir_inode_entry, list);
689 inode = igrab(entry->inode);
690 spin_unlock(&sbi->dir_inode_lock);
692 filemap_fdatawrite(inode->i_mapping);
696 * We should submit bio, since it exists several
697 * wribacking dentry pages in the freeing inode.
699 f2fs_submit_merged_bio(sbi, DATA, WRITE);
705 * Freeze all the FS-operations for checkpoint.
707 static void block_operations(struct f2fs_sb_info *sbi)
709 struct writeback_control wbc = {
710 .sync_mode = WB_SYNC_ALL,
711 .nr_to_write = LONG_MAX,
714 struct blk_plug plug;
716 blk_start_plug(&plug);
720 /* write all the dirty dentry pages */
721 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
722 f2fs_unlock_all(sbi);
723 sync_dirty_dir_inodes(sbi);
724 goto retry_flush_dents;
728 * POR: we should ensure that there is no dirty node pages
729 * until finishing nat/sit flush.
732 mutex_lock(&sbi->node_write);
734 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
735 mutex_unlock(&sbi->node_write);
736 sync_node_pages(sbi, 0, &wbc);
737 goto retry_flush_nodes;
739 blk_finish_plug(&plug);
742 static void unblock_operations(struct f2fs_sb_info *sbi)
744 mutex_unlock(&sbi->node_write);
745 f2fs_unlock_all(sbi);
748 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
753 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
755 if (!get_pages(sbi, F2FS_WRITEBACK))
760 finish_wait(&sbi->cp_wait, &wait);
763 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
765 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
768 struct page *cp_page;
769 unsigned int data_sum_blocks, orphan_blocks;
774 /* Flush all the NAT/SIT pages */
775 while (get_pages(sbi, F2FS_DIRTY_META))
776 sync_meta_pages(sbi, META, LONG_MAX);
778 next_free_nid(sbi, &last_nid);
782 * version number is already updated
784 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
785 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
786 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
787 for (i = 0; i < 3; i++) {
788 ckpt->cur_node_segno[i] =
789 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
790 ckpt->cur_node_blkoff[i] =
791 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
792 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
793 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
795 for (i = 0; i < 3; i++) {
796 ckpt->cur_data_segno[i] =
797 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
798 ckpt->cur_data_blkoff[i] =
799 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
800 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
801 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
804 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
805 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
806 ckpt->next_free_nid = cpu_to_le32(last_nid);
808 /* 2 cp + n data seg summary + orphan inode blocks */
809 data_sum_blocks = npages_for_summary_flush(sbi);
810 if (data_sum_blocks < 3)
811 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
813 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
815 orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
816 / F2FS_ORPHANS_PER_BLOCK;
817 ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks);
820 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
821 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
822 data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE);
824 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
825 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
826 data_sum_blocks + orphan_blocks);
830 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
832 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
834 /* update SIT/NAT bitmap */
835 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
836 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
838 crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
839 *((__le32 *)((unsigned char *)ckpt +
840 le32_to_cpu(ckpt->checksum_offset)))
841 = cpu_to_le32(crc32);
843 start_blk = __start_cp_addr(sbi);
845 /* write out checkpoint buffer at block 0 */
846 cp_page = grab_meta_page(sbi, start_blk++);
847 kaddr = page_address(cp_page);
848 memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
849 set_page_dirty(cp_page);
850 f2fs_put_page(cp_page, 1);
852 if (sbi->n_orphans) {
853 write_orphan_inodes(sbi, start_blk);
854 start_blk += orphan_blocks;
857 write_data_summaries(sbi, start_blk);
858 start_blk += data_sum_blocks;
860 write_node_summaries(sbi, start_blk);
861 start_blk += NR_CURSEG_NODE_TYPE;
864 /* writeout checkpoint block */
865 cp_page = grab_meta_page(sbi, start_blk);
866 kaddr = page_address(cp_page);
867 memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
868 set_page_dirty(cp_page);
869 f2fs_put_page(cp_page, 1);
871 /* wait for previous submitted node/meta pages writeback */
872 wait_on_all_pages_writeback(sbi);
874 filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
875 filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
877 /* update user_block_counts */
878 sbi->last_valid_block_count = sbi->total_valid_block_count;
879 sbi->alloc_valid_block_count = 0;
881 /* Here, we only have one bio having CP pack */
882 sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
884 if (unlikely(!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG))) {
885 clear_prefree_segments(sbi);
886 F2FS_RESET_SB_DIRT(sbi);
891 * We guarantee that this checkpoint procedure should not fail.
893 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
895 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
896 unsigned long long ckpt_ver;
898 trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
900 mutex_lock(&sbi->cp_mutex);
901 block_operations(sbi);
903 trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
905 f2fs_submit_merged_bio(sbi, DATA, WRITE);
906 f2fs_submit_merged_bio(sbi, NODE, WRITE);
907 f2fs_submit_merged_bio(sbi, META, WRITE);
910 * update checkpoint pack index
911 * Increase the version number so that
912 * SIT entries and seg summaries are written at correct place
914 ckpt_ver = cur_cp_version(ckpt);
915 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
917 /* write cached NAT/SIT entries to NAT/SIT area */
918 flush_nat_entries(sbi);
919 flush_sit_entries(sbi);
921 /* unlock all the fs_lock[] in do_checkpoint() */
922 do_checkpoint(sbi, is_umount);
924 unblock_operations(sbi);
925 mutex_unlock(&sbi->cp_mutex);
927 stat_inc_cp_count(sbi->stat_info);
928 trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
931 void init_orphan_info(struct f2fs_sb_info *sbi)
933 spin_lock_init(&sbi->orphan_inode_lock);
934 INIT_LIST_HEAD(&sbi->orphan_inode_list);
937 * considering 512 blocks in a segment 8 blocks are needed for cp
938 * and log segment summaries. Remaining blocks are used to keep
939 * orphan entries with the limitation one reserved segment
940 * for cp pack we can have max 1020*504 orphan entries
942 sbi->max_orphans = (sbi->blocks_per_seg - 2 - NR_CURSEG_TYPE)
943 * F2FS_ORPHANS_PER_BLOCK;
946 int __init create_checkpoint_caches(void)
948 orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry",
949 sizeof(struct orphan_inode_entry));
950 if (!orphan_entry_slab)
952 inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
953 sizeof(struct dir_inode_entry));
954 if (!inode_entry_slab) {
955 kmem_cache_destroy(orphan_entry_slab);
961 void destroy_checkpoint_caches(void)
963 kmem_cache_destroy(orphan_entry_slab);
964 kmem_cache_destroy(inode_entry_slab);