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(mapping, index);
41 f2fs_wait_on_page_writeback(page, META);
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);
75 static inline int get_max_meta_blks(struct f2fs_sb_info *sbi, int type)
79 return NM_I(sbi)->max_nid / NAT_ENTRY_PER_BLOCK;
81 return SIT_BLK_CNT(sbi);
91 * Readahead CP/NAT/SIT/SSA pages
93 int ra_meta_pages(struct f2fs_sb_info *sbi, int start, int nrpages, int type)
95 block_t prev_blk_addr = 0;
98 int max_blks = get_max_meta_blks(sbi, type);
100 struct f2fs_io_info fio = {
102 .rw = READ_SYNC | REQ_META | REQ_PRIO
105 for (; nrpages-- > 0; blkno++) {
110 /* get nat block addr */
111 if (unlikely(blkno >= max_blks))
113 blk_addr = current_nat_addr(sbi,
114 blkno * NAT_ENTRY_PER_BLOCK);
117 /* get sit block addr */
118 if (unlikely(blkno >= max_blks))
120 blk_addr = current_sit_addr(sbi,
121 blkno * SIT_ENTRY_PER_BLOCK);
122 if (blkno != start && prev_blk_addr + 1 != blk_addr)
124 prev_blk_addr = blk_addr;
128 /* get ssa/cp block addr */
135 page = grab_cache_page(META_MAPPING(sbi), blk_addr);
138 if (PageUptodate(page)) {
139 f2fs_put_page(page, 1);
143 f2fs_submit_page_mbio(sbi, page, blk_addr, &fio);
144 f2fs_put_page(page, 0);
147 f2fs_submit_merged_bio(sbi, META, READ);
148 return blkno - start;
151 static int f2fs_write_meta_page(struct page *page,
152 struct writeback_control *wbc)
154 struct inode *inode = page->mapping->host;
155 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
157 trace_f2fs_writepage(page, META);
159 if (unlikely(sbi->por_doing))
161 if (wbc->for_reclaim)
164 /* Should not write any meta pages, if any IO error was occurred */
165 if (unlikely(is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ERROR_FLAG)))
168 f2fs_wait_on_page_writeback(page, META);
169 write_meta_page(sbi, page);
171 dec_page_count(sbi, F2FS_DIRTY_META);
176 redirty_page_for_writepage(wbc, page);
177 return AOP_WRITEPAGE_ACTIVATE;
180 static int f2fs_write_meta_pages(struct address_space *mapping,
181 struct writeback_control *wbc)
183 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
186 trace_f2fs_writepages(mapping->host, wbc, META);
188 /* collect a number of dirty meta pages and write together */
189 if (wbc->for_kupdate ||
190 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
193 /* if mounting is failed, skip writing node pages */
194 mutex_lock(&sbi->cp_mutex);
195 diff = nr_pages_to_write(sbi, META, wbc);
196 written = sync_meta_pages(sbi, META, wbc->nr_to_write);
197 mutex_unlock(&sbi->cp_mutex);
198 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
202 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
206 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
209 struct address_space *mapping = META_MAPPING(sbi);
210 pgoff_t index = 0, end = LONG_MAX;
213 struct writeback_control wbc = {
217 pagevec_init(&pvec, 0);
219 while (index <= end) {
221 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
223 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
224 if (unlikely(nr_pages == 0))
227 for (i = 0; i < nr_pages; i++) {
228 struct page *page = pvec.pages[i];
232 if (unlikely(page->mapping != mapping)) {
237 if (!PageDirty(page)) {
238 /* someone wrote it for us */
239 goto continue_unlock;
242 if (!clear_page_dirty_for_io(page))
243 goto continue_unlock;
245 if (f2fs_write_meta_page(page, &wbc)) {
250 if (unlikely(nwritten >= nr_to_write))
253 pagevec_release(&pvec);
258 f2fs_submit_merged_bio(sbi, type, WRITE);
263 static int f2fs_set_meta_page_dirty(struct page *page)
265 struct address_space *mapping = page->mapping;
266 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
268 trace_f2fs_set_page_dirty(page, META);
270 SetPageUptodate(page);
271 if (!PageDirty(page)) {
272 __set_page_dirty_nobuffers(page);
273 inc_page_count(sbi, F2FS_DIRTY_META);
279 const struct address_space_operations f2fs_meta_aops = {
280 .writepage = f2fs_write_meta_page,
281 .writepages = f2fs_write_meta_pages,
282 .set_page_dirty = f2fs_set_meta_page_dirty,
285 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
289 spin_lock(&sbi->orphan_inode_lock);
290 if (unlikely(sbi->n_orphans >= sbi->max_orphans))
294 spin_unlock(&sbi->orphan_inode_lock);
299 void release_orphan_inode(struct f2fs_sb_info *sbi)
301 spin_lock(&sbi->orphan_inode_lock);
302 f2fs_bug_on(sbi->n_orphans == 0);
304 spin_unlock(&sbi->orphan_inode_lock);
307 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
309 struct list_head *head;
310 struct orphan_inode_entry *new, *orphan;
312 new = f2fs_kmem_cache_alloc(orphan_entry_slab, GFP_ATOMIC);
315 spin_lock(&sbi->orphan_inode_lock);
316 head = &sbi->orphan_inode_list;
317 list_for_each_entry(orphan, head, list) {
318 if (orphan->ino == ino) {
319 spin_unlock(&sbi->orphan_inode_lock);
320 kmem_cache_free(orphan_entry_slab, new);
324 if (orphan->ino > ino)
328 /* add new orphan entry into list which is sorted by inode number */
329 list_add_tail(&new->list, &orphan->list);
330 spin_unlock(&sbi->orphan_inode_lock);
333 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
335 struct list_head *head;
336 struct orphan_inode_entry *orphan;
338 spin_lock(&sbi->orphan_inode_lock);
339 head = &sbi->orphan_inode_list;
340 list_for_each_entry(orphan, head, list) {
341 if (orphan->ino == ino) {
342 list_del(&orphan->list);
343 f2fs_bug_on(sbi->n_orphans == 0);
345 spin_unlock(&sbi->orphan_inode_lock);
346 kmem_cache_free(orphan_entry_slab, orphan);
350 spin_unlock(&sbi->orphan_inode_lock);
353 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
355 struct inode *inode = f2fs_iget(sbi->sb, ino);
356 f2fs_bug_on(IS_ERR(inode));
359 /* truncate all the data during iput */
363 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
365 block_t start_blk, orphan_blkaddr, i, j;
367 if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
370 sbi->por_doing = true;
371 start_blk = __start_cp_addr(sbi) + 1;
372 orphan_blkaddr = __start_sum_addr(sbi) - 1;
374 ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP);
376 for (i = 0; i < orphan_blkaddr; i++) {
377 struct page *page = get_meta_page(sbi, start_blk + i);
378 struct f2fs_orphan_block *orphan_blk;
380 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
381 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
382 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
383 recover_orphan_inode(sbi, ino);
385 f2fs_put_page(page, 1);
387 /* clear Orphan Flag */
388 clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
389 sbi->por_doing = false;
393 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
395 struct list_head *head;
396 struct f2fs_orphan_block *orphan_blk = NULL;
397 unsigned int nentries = 0;
398 unsigned short index;
399 unsigned short orphan_blocks = (unsigned short)((sbi->n_orphans +
400 (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK);
401 struct page *page = NULL;
402 struct orphan_inode_entry *orphan = NULL;
404 for (index = 0; index < orphan_blocks; index++)
405 grab_meta_page(sbi, start_blk + index);
408 spin_lock(&sbi->orphan_inode_lock);
409 head = &sbi->orphan_inode_list;
411 /* loop for each orphan inode entry and write them in Jornal block */
412 list_for_each_entry(orphan, head, list) {
414 page = find_get_page(META_MAPPING(sbi), start_blk++);
417 (struct f2fs_orphan_block *)page_address(page);
418 memset(orphan_blk, 0, sizeof(*orphan_blk));
419 f2fs_put_page(page, 0);
422 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
424 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
426 * an orphan block is full of 1020 entries,
427 * then we need to flush current orphan blocks
428 * and bring another one in memory
430 orphan_blk->blk_addr = cpu_to_le16(index);
431 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
432 orphan_blk->entry_count = cpu_to_le32(nentries);
433 set_page_dirty(page);
434 f2fs_put_page(page, 1);
442 orphan_blk->blk_addr = cpu_to_le16(index);
443 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
444 orphan_blk->entry_count = cpu_to_le32(nentries);
445 set_page_dirty(page);
446 f2fs_put_page(page, 1);
449 spin_unlock(&sbi->orphan_inode_lock);
452 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
453 block_t cp_addr, unsigned long long *version)
455 struct page *cp_page_1, *cp_page_2 = NULL;
456 unsigned long blk_size = sbi->blocksize;
457 struct f2fs_checkpoint *cp_block;
458 unsigned long long cur_version = 0, pre_version = 0;
462 /* Read the 1st cp block in this CP pack */
463 cp_page_1 = get_meta_page(sbi, cp_addr);
465 /* get the version number */
466 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
467 crc_offset = le32_to_cpu(cp_block->checksum_offset);
468 if (crc_offset >= blk_size)
471 crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
472 if (!f2fs_crc_valid(crc, cp_block, crc_offset))
475 pre_version = cur_cp_version(cp_block);
477 /* Read the 2nd cp block in this CP pack */
478 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
479 cp_page_2 = get_meta_page(sbi, cp_addr);
481 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
482 crc_offset = le32_to_cpu(cp_block->checksum_offset);
483 if (crc_offset >= blk_size)
486 crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
487 if (!f2fs_crc_valid(crc, cp_block, crc_offset))
490 cur_version = cur_cp_version(cp_block);
492 if (cur_version == pre_version) {
493 *version = cur_version;
494 f2fs_put_page(cp_page_2, 1);
498 f2fs_put_page(cp_page_2, 1);
500 f2fs_put_page(cp_page_1, 1);
504 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
506 struct f2fs_checkpoint *cp_block;
507 struct f2fs_super_block *fsb = sbi->raw_super;
508 struct page *cp1, *cp2, *cur_page;
509 unsigned long blk_size = sbi->blocksize;
510 unsigned long long cp1_version = 0, cp2_version = 0;
511 unsigned long long cp_start_blk_no;
513 sbi->ckpt = kzalloc(blk_size, GFP_KERNEL);
517 * Finding out valid cp block involves read both
518 * sets( cp pack1 and cp pack 2)
520 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
521 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
523 /* The second checkpoint pack should start at the next segment */
524 cp_start_blk_no += ((unsigned long long)1) <<
525 le32_to_cpu(fsb->log_blocks_per_seg);
526 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
529 if (ver_after(cp2_version, cp1_version))
541 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
542 memcpy(sbi->ckpt, cp_block, blk_size);
544 f2fs_put_page(cp1, 1);
545 f2fs_put_page(cp2, 1);
553 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
555 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
557 if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
560 set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
561 F2FS_I(inode)->dirty_dir = new;
562 list_add_tail(&new->list, &sbi->dir_inode_list);
563 stat_inc_dirty_dir(sbi);
567 void set_dirty_dir_page(struct inode *inode, struct page *page)
569 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
570 struct dir_inode_entry *new;
573 if (!S_ISDIR(inode->i_mode))
576 new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
578 INIT_LIST_HEAD(&new->list);
580 spin_lock(&sbi->dir_inode_lock);
581 ret = __add_dirty_inode(inode, new);
582 inode_inc_dirty_dents(inode);
583 SetPagePrivate(page);
584 spin_unlock(&sbi->dir_inode_lock);
587 kmem_cache_free(inode_entry_slab, new);
590 void add_dirty_dir_inode(struct inode *inode)
592 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
593 struct dir_inode_entry *new =
594 f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
598 INIT_LIST_HEAD(&new->list);
600 spin_lock(&sbi->dir_inode_lock);
601 ret = __add_dirty_inode(inode, new);
602 spin_unlock(&sbi->dir_inode_lock);
605 kmem_cache_free(inode_entry_slab, new);
608 void remove_dirty_dir_inode(struct inode *inode)
610 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
611 struct dir_inode_entry *entry;
613 if (!S_ISDIR(inode->i_mode))
616 spin_lock(&sbi->dir_inode_lock);
617 if (get_dirty_dents(inode) ||
618 !is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
619 spin_unlock(&sbi->dir_inode_lock);
623 entry = F2FS_I(inode)->dirty_dir;
624 list_del(&entry->list);
625 F2FS_I(inode)->dirty_dir = NULL;
626 clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
627 stat_dec_dirty_dir(sbi);
628 spin_unlock(&sbi->dir_inode_lock);
629 kmem_cache_free(inode_entry_slab, entry);
631 /* Only from the recovery routine */
632 if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
633 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
638 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
640 struct list_head *head;
641 struct dir_inode_entry *entry;
644 spin_lock(&sbi->dir_inode_lock);
646 head = &sbi->dir_inode_list;
647 if (list_empty(head)) {
648 spin_unlock(&sbi->dir_inode_lock);
651 entry = list_entry(head->next, struct dir_inode_entry, list);
652 inode = igrab(entry->inode);
653 spin_unlock(&sbi->dir_inode_lock);
655 filemap_fdatawrite(inode->i_mapping);
659 * We should submit bio, since it exists several
660 * wribacking dentry pages in the freeing inode.
662 f2fs_submit_merged_bio(sbi, DATA, WRITE);
668 * Freeze all the FS-operations for checkpoint.
670 static void block_operations(struct f2fs_sb_info *sbi)
672 struct writeback_control wbc = {
673 .sync_mode = WB_SYNC_ALL,
674 .nr_to_write = LONG_MAX,
677 struct blk_plug plug;
679 blk_start_plug(&plug);
683 /* write all the dirty dentry pages */
684 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
685 f2fs_unlock_all(sbi);
686 sync_dirty_dir_inodes(sbi);
687 goto retry_flush_dents;
691 * POR: we should ensure that there is no dirty node pages
692 * until finishing nat/sit flush.
695 mutex_lock(&sbi->node_write);
697 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
698 mutex_unlock(&sbi->node_write);
699 sync_node_pages(sbi, 0, &wbc);
700 goto retry_flush_nodes;
702 blk_finish_plug(&plug);
705 static void unblock_operations(struct f2fs_sb_info *sbi)
707 mutex_unlock(&sbi->node_write);
708 f2fs_unlock_all(sbi);
711 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
716 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
718 if (!get_pages(sbi, F2FS_WRITEBACK))
723 finish_wait(&sbi->cp_wait, &wait);
726 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
728 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
731 struct page *cp_page;
732 unsigned int data_sum_blocks, orphan_blocks;
738 * This avoids to conduct wrong roll-forward operations and uses
739 * metapages, so should be called prior to sync_meta_pages below.
741 discard_next_dnode(sbi);
743 /* Flush all the NAT/SIT pages */
744 while (get_pages(sbi, F2FS_DIRTY_META))
745 sync_meta_pages(sbi, META, LONG_MAX);
747 next_free_nid(sbi, &last_nid);
751 * version number is already updated
753 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
754 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
755 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
756 for (i = 0; i < 3; i++) {
757 ckpt->cur_node_segno[i] =
758 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
759 ckpt->cur_node_blkoff[i] =
760 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
761 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
762 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
764 for (i = 0; i < 3; i++) {
765 ckpt->cur_data_segno[i] =
766 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
767 ckpt->cur_data_blkoff[i] =
768 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
769 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
770 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
773 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
774 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
775 ckpt->next_free_nid = cpu_to_le32(last_nid);
777 /* 2 cp + n data seg summary + orphan inode blocks */
778 data_sum_blocks = npages_for_summary_flush(sbi);
779 if (data_sum_blocks < 3)
780 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
782 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
784 orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
785 / F2FS_ORPHANS_PER_BLOCK;
786 ckpt->cp_pack_start_sum = cpu_to_le32(1 + orphan_blocks);
789 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
790 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
791 data_sum_blocks + orphan_blocks + NR_CURSEG_NODE_TYPE);
793 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
794 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
795 data_sum_blocks + orphan_blocks);
799 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
801 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
803 /* update SIT/NAT bitmap */
804 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
805 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
807 crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
808 *((__le32 *)((unsigned char *)ckpt +
809 le32_to_cpu(ckpt->checksum_offset)))
810 = cpu_to_le32(crc32);
812 start_blk = __start_cp_addr(sbi);
814 /* write out checkpoint buffer at block 0 */
815 cp_page = grab_meta_page(sbi, start_blk++);
816 kaddr = page_address(cp_page);
817 memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
818 set_page_dirty(cp_page);
819 f2fs_put_page(cp_page, 1);
821 if (sbi->n_orphans) {
822 write_orphan_inodes(sbi, start_blk);
823 start_blk += orphan_blocks;
826 write_data_summaries(sbi, start_blk);
827 start_blk += data_sum_blocks;
829 write_node_summaries(sbi, start_blk);
830 start_blk += NR_CURSEG_NODE_TYPE;
833 /* writeout checkpoint block */
834 cp_page = grab_meta_page(sbi, start_blk);
835 kaddr = page_address(cp_page);
836 memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
837 set_page_dirty(cp_page);
838 f2fs_put_page(cp_page, 1);
840 /* wait for previous submitted node/meta pages writeback */
841 wait_on_all_pages_writeback(sbi);
843 filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
844 filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
846 /* update user_block_counts */
847 sbi->last_valid_block_count = sbi->total_valid_block_count;
848 sbi->alloc_valid_block_count = 0;
850 /* Here, we only have one bio having CP pack */
851 sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
853 if (unlikely(!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG))) {
854 clear_prefree_segments(sbi);
855 F2FS_RESET_SB_DIRT(sbi);
860 * We guarantee that this checkpoint procedure should not fail.
862 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
864 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
865 unsigned long long ckpt_ver;
867 trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
869 mutex_lock(&sbi->cp_mutex);
870 block_operations(sbi);
872 trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
874 f2fs_submit_merged_bio(sbi, DATA, WRITE);
875 f2fs_submit_merged_bio(sbi, NODE, WRITE);
876 f2fs_submit_merged_bio(sbi, META, WRITE);
879 * update checkpoint pack index
880 * Increase the version number so that
881 * SIT entries and seg summaries are written at correct place
883 ckpt_ver = cur_cp_version(ckpt);
884 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
886 /* write cached NAT/SIT entries to NAT/SIT area */
887 flush_nat_entries(sbi);
888 flush_sit_entries(sbi);
890 /* unlock all the fs_lock[] in do_checkpoint() */
891 do_checkpoint(sbi, is_umount);
893 unblock_operations(sbi);
894 mutex_unlock(&sbi->cp_mutex);
896 stat_inc_cp_count(sbi->stat_info);
897 trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
900 void init_orphan_info(struct f2fs_sb_info *sbi)
902 spin_lock_init(&sbi->orphan_inode_lock);
903 INIT_LIST_HEAD(&sbi->orphan_inode_list);
906 * considering 512 blocks in a segment 8 blocks are needed for cp
907 * and log segment summaries. Remaining blocks are used to keep
908 * orphan entries with the limitation one reserved segment
909 * for cp pack we can have max 1020*504 orphan entries
911 sbi->max_orphans = (sbi->blocks_per_seg - 2 - NR_CURSEG_TYPE)
912 * F2FS_ORPHANS_PER_BLOCK;
915 int __init create_checkpoint_caches(void)
917 orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry",
918 sizeof(struct orphan_inode_entry));
919 if (!orphan_entry_slab)
921 inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
922 sizeof(struct dir_inode_entry));
923 if (!inode_entry_slab) {
924 kmem_cache_destroy(orphan_entry_slab);
930 void destroy_checkpoint_caches(void)
932 kmem_cache_destroy(orphan_entry_slab);
933 kmem_cache_destroy(inode_entry_slab);