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;
372 start_blk = __start_cp_addr(sbi) + 1 +
373 le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
374 orphan_blkaddr = __start_sum_addr(sbi) - 1;
376 ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP);
378 for (i = 0; i < orphan_blkaddr; i++) {
379 struct page *page = get_meta_page(sbi, start_blk + i);
380 struct f2fs_orphan_block *orphan_blk;
382 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
383 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
384 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
385 recover_orphan_inode(sbi, ino);
387 f2fs_put_page(page, 1);
389 /* clear Orphan Flag */
390 clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
391 sbi->por_doing = false;
395 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
397 struct list_head *head;
398 struct f2fs_orphan_block *orphan_blk = NULL;
399 unsigned int nentries = 0;
400 unsigned short index;
401 unsigned short orphan_blocks = (unsigned short)((sbi->n_orphans +
402 (F2FS_ORPHANS_PER_BLOCK - 1)) / F2FS_ORPHANS_PER_BLOCK);
403 struct page *page = NULL;
404 struct orphan_inode_entry *orphan = NULL;
406 for (index = 0; index < orphan_blocks; index++)
407 grab_meta_page(sbi, start_blk + index);
410 spin_lock(&sbi->orphan_inode_lock);
411 head = &sbi->orphan_inode_list;
413 /* loop for each orphan inode entry and write them in Jornal block */
414 list_for_each_entry(orphan, head, list) {
416 page = find_get_page(META_MAPPING(sbi), start_blk++);
419 (struct f2fs_orphan_block *)page_address(page);
420 memset(orphan_blk, 0, sizeof(*orphan_blk));
421 f2fs_put_page(page, 0);
424 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
426 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
428 * an orphan block is full of 1020 entries,
429 * then we need to flush current orphan blocks
430 * and bring another one in memory
432 orphan_blk->blk_addr = cpu_to_le16(index);
433 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
434 orphan_blk->entry_count = cpu_to_le32(nentries);
435 set_page_dirty(page);
436 f2fs_put_page(page, 1);
444 orphan_blk->blk_addr = cpu_to_le16(index);
445 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
446 orphan_blk->entry_count = cpu_to_le32(nentries);
447 set_page_dirty(page);
448 f2fs_put_page(page, 1);
451 spin_unlock(&sbi->orphan_inode_lock);
454 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
455 block_t cp_addr, unsigned long long *version)
457 struct page *cp_page_1, *cp_page_2 = NULL;
458 unsigned long blk_size = sbi->blocksize;
459 struct f2fs_checkpoint *cp_block;
460 unsigned long long cur_version = 0, pre_version = 0;
464 /* Read the 1st cp block in this CP pack */
465 cp_page_1 = get_meta_page(sbi, cp_addr);
467 /* get the version number */
468 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
469 crc_offset = le32_to_cpu(cp_block->checksum_offset);
470 if (crc_offset >= blk_size)
473 crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
474 if (!f2fs_crc_valid(crc, cp_block, crc_offset))
477 pre_version = cur_cp_version(cp_block);
479 /* Read the 2nd cp block in this CP pack */
480 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
481 cp_page_2 = get_meta_page(sbi, cp_addr);
483 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
484 crc_offset = le32_to_cpu(cp_block->checksum_offset);
485 if (crc_offset >= blk_size)
488 crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
489 if (!f2fs_crc_valid(crc, cp_block, crc_offset))
492 cur_version = cur_cp_version(cp_block);
494 if (cur_version == pre_version) {
495 *version = cur_version;
496 f2fs_put_page(cp_page_2, 1);
500 f2fs_put_page(cp_page_2, 1);
502 f2fs_put_page(cp_page_1, 1);
506 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
508 struct f2fs_checkpoint *cp_block;
509 struct f2fs_super_block *fsb = sbi->raw_super;
510 struct page *cp1, *cp2, *cur_page;
511 unsigned long blk_size = sbi->blocksize;
512 unsigned long long cp1_version = 0, cp2_version = 0;
513 unsigned long long cp_start_blk_no;
514 unsigned int cp_blks = 1 + le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
518 sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
522 * Finding out valid cp block involves read both
523 * sets( cp pack1 and cp pack 2)
525 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
526 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
528 /* The second checkpoint pack should start at the next segment */
529 cp_start_blk_no += ((unsigned long long)1) <<
530 le32_to_cpu(fsb->log_blocks_per_seg);
531 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
534 if (ver_after(cp2_version, cp1_version))
546 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
547 memcpy(sbi->ckpt, cp_block, blk_size);
552 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
554 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
556 for (i = 1; i < cp_blks; i++) {
557 void *sit_bitmap_ptr;
558 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
560 cur_page = get_meta_page(sbi, cp_blk_no + i);
561 sit_bitmap_ptr = page_address(cur_page);
562 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
563 f2fs_put_page(cur_page, 1);
566 f2fs_put_page(cp1, 1);
567 f2fs_put_page(cp2, 1);
575 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
577 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
579 if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
582 set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
583 F2FS_I(inode)->dirty_dir = new;
584 list_add_tail(&new->list, &sbi->dir_inode_list);
585 stat_inc_dirty_dir(sbi);
589 void set_dirty_dir_page(struct inode *inode, struct page *page)
591 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
592 struct dir_inode_entry *new;
595 if (!S_ISDIR(inode->i_mode))
598 new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
600 INIT_LIST_HEAD(&new->list);
602 spin_lock(&sbi->dir_inode_lock);
603 ret = __add_dirty_inode(inode, new);
604 inode_inc_dirty_dents(inode);
605 SetPagePrivate(page);
606 spin_unlock(&sbi->dir_inode_lock);
609 kmem_cache_free(inode_entry_slab, new);
612 void add_dirty_dir_inode(struct inode *inode)
614 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
615 struct dir_inode_entry *new =
616 f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
620 INIT_LIST_HEAD(&new->list);
622 spin_lock(&sbi->dir_inode_lock);
623 ret = __add_dirty_inode(inode, new);
624 spin_unlock(&sbi->dir_inode_lock);
627 kmem_cache_free(inode_entry_slab, new);
630 void remove_dirty_dir_inode(struct inode *inode)
632 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
633 struct dir_inode_entry *entry;
635 if (!S_ISDIR(inode->i_mode))
638 spin_lock(&sbi->dir_inode_lock);
639 if (get_dirty_dents(inode) ||
640 !is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
641 spin_unlock(&sbi->dir_inode_lock);
645 entry = F2FS_I(inode)->dirty_dir;
646 list_del(&entry->list);
647 F2FS_I(inode)->dirty_dir = NULL;
648 clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
649 stat_dec_dirty_dir(sbi);
650 spin_unlock(&sbi->dir_inode_lock);
651 kmem_cache_free(inode_entry_slab, entry);
653 /* Only from the recovery routine */
654 if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
655 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
660 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
662 struct list_head *head;
663 struct dir_inode_entry *entry;
666 spin_lock(&sbi->dir_inode_lock);
668 head = &sbi->dir_inode_list;
669 if (list_empty(head)) {
670 spin_unlock(&sbi->dir_inode_lock);
673 entry = list_entry(head->next, struct dir_inode_entry, list);
674 inode = igrab(entry->inode);
675 spin_unlock(&sbi->dir_inode_lock);
677 filemap_fdatawrite(inode->i_mapping);
681 * We should submit bio, since it exists several
682 * wribacking dentry pages in the freeing inode.
684 f2fs_submit_merged_bio(sbi, DATA, WRITE);
690 * Freeze all the FS-operations for checkpoint.
692 static void block_operations(struct f2fs_sb_info *sbi)
694 struct writeback_control wbc = {
695 .sync_mode = WB_SYNC_ALL,
696 .nr_to_write = LONG_MAX,
699 struct blk_plug plug;
701 blk_start_plug(&plug);
705 /* write all the dirty dentry pages */
706 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
707 f2fs_unlock_all(sbi);
708 sync_dirty_dir_inodes(sbi);
709 goto retry_flush_dents;
713 * POR: we should ensure that there is no dirty node pages
714 * until finishing nat/sit flush.
717 mutex_lock(&sbi->node_write);
719 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
720 mutex_unlock(&sbi->node_write);
721 sync_node_pages(sbi, 0, &wbc);
722 goto retry_flush_nodes;
724 blk_finish_plug(&plug);
727 static void unblock_operations(struct f2fs_sb_info *sbi)
729 mutex_unlock(&sbi->node_write);
730 f2fs_unlock_all(sbi);
733 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
738 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
740 if (!get_pages(sbi, F2FS_WRITEBACK))
745 finish_wait(&sbi->cp_wait, &wait);
748 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
750 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
753 struct page *cp_page;
754 unsigned int data_sum_blocks, orphan_blocks;
758 int cp_payload_blks = le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
761 * This avoids to conduct wrong roll-forward operations and uses
762 * metapages, so should be called prior to sync_meta_pages below.
764 discard_next_dnode(sbi);
766 /* Flush all the NAT/SIT pages */
767 while (get_pages(sbi, F2FS_DIRTY_META))
768 sync_meta_pages(sbi, META, LONG_MAX);
770 next_free_nid(sbi, &last_nid);
774 * version number is already updated
776 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
777 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
778 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
779 for (i = 0; i < 3; i++) {
780 ckpt->cur_node_segno[i] =
781 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
782 ckpt->cur_node_blkoff[i] =
783 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
784 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
785 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
787 for (i = 0; i < 3; i++) {
788 ckpt->cur_data_segno[i] =
789 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
790 ckpt->cur_data_blkoff[i] =
791 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
792 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
793 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
796 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
797 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
798 ckpt->next_free_nid = cpu_to_le32(last_nid);
800 /* 2 cp + n data seg summary + orphan inode blocks */
801 data_sum_blocks = npages_for_summary_flush(sbi);
802 if (data_sum_blocks < 3)
803 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
805 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
807 orphan_blocks = (sbi->n_orphans + F2FS_ORPHANS_PER_BLOCK - 1)
808 / F2FS_ORPHANS_PER_BLOCK;
809 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
813 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
814 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
815 cp_payload_blks + data_sum_blocks +
816 orphan_blocks + NR_CURSEG_NODE_TYPE);
818 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
819 ckpt->cp_pack_total_block_count = cpu_to_le32(2 +
820 cp_payload_blks + data_sum_blocks +
825 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
827 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
829 /* update SIT/NAT bitmap */
830 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
831 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
833 crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
834 *((__le32 *)((unsigned char *)ckpt +
835 le32_to_cpu(ckpt->checksum_offset)))
836 = cpu_to_le32(crc32);
838 start_blk = __start_cp_addr(sbi);
840 /* write out checkpoint buffer at block 0 */
841 cp_page = grab_meta_page(sbi, start_blk++);
842 kaddr = page_address(cp_page);
843 memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
844 set_page_dirty(cp_page);
845 f2fs_put_page(cp_page, 1);
847 for (i = 1; i < 1 + cp_payload_blks; i++) {
848 cp_page = grab_meta_page(sbi, start_blk++);
849 kaddr = page_address(cp_page);
850 memcpy(kaddr, (char *)ckpt + i * F2FS_BLKSIZE,
851 (1 << sbi->log_blocksize));
852 set_page_dirty(cp_page);
853 f2fs_put_page(cp_page, 1);
856 if (sbi->n_orphans) {
857 write_orphan_inodes(sbi, start_blk);
858 start_blk += orphan_blocks;
861 write_data_summaries(sbi, start_blk);
862 start_blk += data_sum_blocks;
864 write_node_summaries(sbi, start_blk);
865 start_blk += NR_CURSEG_NODE_TYPE;
868 /* writeout checkpoint block */
869 cp_page = grab_meta_page(sbi, start_blk);
870 kaddr = page_address(cp_page);
871 memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
872 set_page_dirty(cp_page);
873 f2fs_put_page(cp_page, 1);
875 /* wait for previous submitted node/meta pages writeback */
876 wait_on_all_pages_writeback(sbi);
878 filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
879 filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
881 /* update user_block_counts */
882 sbi->last_valid_block_count = sbi->total_valid_block_count;
883 sbi->alloc_valid_block_count = 0;
885 /* Here, we only have one bio having CP pack */
886 sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
888 if (unlikely(!is_set_ckpt_flags(ckpt, CP_ERROR_FLAG))) {
889 clear_prefree_segments(sbi);
890 F2FS_RESET_SB_DIRT(sbi);
895 * We guarantee that this checkpoint procedure should not fail.
897 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
899 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
900 unsigned long long ckpt_ver;
902 trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
904 mutex_lock(&sbi->cp_mutex);
905 block_operations(sbi);
907 trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
909 f2fs_submit_merged_bio(sbi, DATA, WRITE);
910 f2fs_submit_merged_bio(sbi, NODE, WRITE);
911 f2fs_submit_merged_bio(sbi, META, WRITE);
914 * update checkpoint pack index
915 * Increase the version number so that
916 * SIT entries and seg summaries are written at correct place
918 ckpt_ver = cur_cp_version(ckpt);
919 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
921 /* write cached NAT/SIT entries to NAT/SIT area */
922 flush_nat_entries(sbi);
923 flush_sit_entries(sbi);
925 /* unlock all the fs_lock[] in do_checkpoint() */
926 do_checkpoint(sbi, is_umount);
928 unblock_operations(sbi);
929 mutex_unlock(&sbi->cp_mutex);
931 stat_inc_cp_count(sbi->stat_info);
932 trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
935 void init_orphan_info(struct f2fs_sb_info *sbi)
937 spin_lock_init(&sbi->orphan_inode_lock);
938 INIT_LIST_HEAD(&sbi->orphan_inode_list);
941 * considering 512 blocks in a segment 8 blocks are needed for cp
942 * and log segment summaries. Remaining blocks are used to keep
943 * orphan entries with the limitation one reserved segment
944 * for cp pack we can have max 1020*504 orphan entries
946 sbi->max_orphans = (sbi->blocks_per_seg - 2 - NR_CURSEG_TYPE)
947 * F2FS_ORPHANS_PER_BLOCK;
950 int __init create_checkpoint_caches(void)
952 orphan_entry_slab = f2fs_kmem_cache_create("f2fs_orphan_entry",
953 sizeof(struct orphan_inode_entry));
954 if (!orphan_entry_slab)
956 inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
957 sizeof(struct dir_inode_entry));
958 if (!inode_entry_slab) {
959 kmem_cache_destroy(orphan_entry_slab);
965 void destroy_checkpoint_caches(void)
967 kmem_cache_destroy(orphan_entry_slab);
968 kmem_cache_destroy(inode_entry_slab);