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 *ino_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)
163 if (unlikely(f2fs_cp_error(sbi)))
166 f2fs_wait_on_page_writeback(page, META);
167 write_meta_page(sbi, page);
168 dec_page_count(sbi, F2FS_DIRTY_META);
173 redirty_page_for_writepage(wbc, page);
174 return AOP_WRITEPAGE_ACTIVATE;
177 static int f2fs_write_meta_pages(struct address_space *mapping,
178 struct writeback_control *wbc)
180 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
183 trace_f2fs_writepages(mapping->host, wbc, META);
185 /* collect a number of dirty meta pages and write together */
186 if (wbc->for_kupdate ||
187 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
190 /* if mounting is failed, skip writing node pages */
191 mutex_lock(&sbi->cp_mutex);
192 diff = nr_pages_to_write(sbi, META, wbc);
193 written = sync_meta_pages(sbi, META, wbc->nr_to_write);
194 mutex_unlock(&sbi->cp_mutex);
195 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
199 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
203 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
206 struct address_space *mapping = META_MAPPING(sbi);
207 pgoff_t index = 0, end = LONG_MAX;
210 struct writeback_control wbc = {
214 pagevec_init(&pvec, 0);
216 while (index <= end) {
218 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
220 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
221 if (unlikely(nr_pages == 0))
224 for (i = 0; i < nr_pages; i++) {
225 struct page *page = pvec.pages[i];
229 if (unlikely(page->mapping != mapping)) {
234 if (!PageDirty(page)) {
235 /* someone wrote it for us */
236 goto continue_unlock;
239 if (!clear_page_dirty_for_io(page))
240 goto continue_unlock;
242 if (f2fs_write_meta_page(page, &wbc)) {
247 if (unlikely(nwritten >= nr_to_write))
250 pagevec_release(&pvec);
255 f2fs_submit_merged_bio(sbi, type, WRITE);
260 static int f2fs_set_meta_page_dirty(struct page *page)
262 struct address_space *mapping = page->mapping;
263 struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
265 trace_f2fs_set_page_dirty(page, META);
267 SetPageUptodate(page);
268 if (!PageDirty(page)) {
269 __set_page_dirty_nobuffers(page);
270 inc_page_count(sbi, F2FS_DIRTY_META);
276 const struct address_space_operations f2fs_meta_aops = {
277 .writepage = f2fs_write_meta_page,
278 .writepages = f2fs_write_meta_pages,
279 .set_page_dirty = f2fs_set_meta_page_dirty,
282 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
286 spin_lock(&sbi->ino_lock[type]);
288 e = radix_tree_lookup(&sbi->ino_root[type], ino);
290 e = kmem_cache_alloc(ino_entry_slab, GFP_ATOMIC);
292 spin_unlock(&sbi->ino_lock[type]);
295 if (radix_tree_insert(&sbi->ino_root[type], ino, e)) {
296 spin_unlock(&sbi->ino_lock[type]);
297 kmem_cache_free(ino_entry_slab, e);
300 memset(e, 0, sizeof(struct ino_entry));
303 list_add_tail(&e->list, &sbi->ino_list[type]);
305 spin_unlock(&sbi->ino_lock[type]);
308 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
312 spin_lock(&sbi->ino_lock[type]);
313 e = radix_tree_lookup(&sbi->ino_root[type], ino);
316 radix_tree_delete(&sbi->ino_root[type], ino);
317 if (type == ORPHAN_INO)
319 spin_unlock(&sbi->ino_lock[type]);
320 kmem_cache_free(ino_entry_slab, e);
323 spin_unlock(&sbi->ino_lock[type]);
326 void add_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
328 /* add new dirty ino entry into list */
329 __add_ino_entry(sbi, ino, type);
332 void remove_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
334 /* remove dirty ino entry from list */
335 __remove_ino_entry(sbi, ino, type);
338 /* mode should be APPEND_INO or UPDATE_INO */
339 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
342 spin_lock(&sbi->ino_lock[mode]);
343 e = radix_tree_lookup(&sbi->ino_root[mode], ino);
344 spin_unlock(&sbi->ino_lock[mode]);
345 return e ? true : false;
348 void release_dirty_inode(struct f2fs_sb_info *sbi)
350 struct ino_entry *e, *tmp;
353 for (i = APPEND_INO; i <= UPDATE_INO; i++) {
354 spin_lock(&sbi->ino_lock[i]);
355 list_for_each_entry_safe(e, tmp, &sbi->ino_list[i], list) {
357 radix_tree_delete(&sbi->ino_root[i], e->ino);
358 kmem_cache_free(ino_entry_slab, e);
360 spin_unlock(&sbi->ino_lock[i]);
364 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
368 spin_lock(&sbi->ino_lock[ORPHAN_INO]);
369 if (unlikely(sbi->n_orphans >= sbi->max_orphans))
373 spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
378 void release_orphan_inode(struct f2fs_sb_info *sbi)
380 spin_lock(&sbi->ino_lock[ORPHAN_INO]);
381 f2fs_bug_on(sbi->n_orphans == 0);
383 spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
386 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
388 /* add new orphan ino entry into list */
389 __add_ino_entry(sbi, ino, ORPHAN_INO);
392 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
394 /* remove orphan entry from orphan list */
395 __remove_ino_entry(sbi, ino, ORPHAN_INO);
398 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
400 struct inode *inode = f2fs_iget(sbi->sb, ino);
401 f2fs_bug_on(IS_ERR(inode));
404 /* truncate all the data during iput */
408 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
410 block_t start_blk, orphan_blkaddr, i, j;
412 if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
415 sbi->por_doing = true;
417 start_blk = __start_cp_addr(sbi) + 1 +
418 le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
419 orphan_blkaddr = __start_sum_addr(sbi) - 1;
421 ra_meta_pages(sbi, start_blk, orphan_blkaddr, META_CP);
423 for (i = 0; i < orphan_blkaddr; i++) {
424 struct page *page = get_meta_page(sbi, start_blk + i);
425 struct f2fs_orphan_block *orphan_blk;
427 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
428 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
429 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
430 recover_orphan_inode(sbi, ino);
432 f2fs_put_page(page, 1);
434 /* clear Orphan Flag */
435 clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
436 sbi->por_doing = false;
440 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
442 struct list_head *head;
443 struct f2fs_orphan_block *orphan_blk = NULL;
444 unsigned int nentries = 0;
445 unsigned short index;
446 unsigned short orphan_blocks =
447 (unsigned short)GET_ORPHAN_BLOCKS(sbi->n_orphans);
448 struct page *page = NULL;
449 struct ino_entry *orphan = NULL;
451 for (index = 0; index < orphan_blocks; index++)
452 grab_meta_page(sbi, start_blk + index);
455 spin_lock(&sbi->ino_lock[ORPHAN_INO]);
456 head = &sbi->ino_list[ORPHAN_INO];
458 /* loop for each orphan inode entry and write them in Jornal block */
459 list_for_each_entry(orphan, head, list) {
461 page = find_get_page(META_MAPPING(sbi), start_blk++);
464 (struct f2fs_orphan_block *)page_address(page);
465 memset(orphan_blk, 0, sizeof(*orphan_blk));
466 f2fs_put_page(page, 0);
469 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
471 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
473 * an orphan block is full of 1020 entries,
474 * then we need to flush current orphan blocks
475 * and bring another one in memory
477 orphan_blk->blk_addr = cpu_to_le16(index);
478 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
479 orphan_blk->entry_count = cpu_to_le32(nentries);
480 set_page_dirty(page);
481 f2fs_put_page(page, 1);
489 orphan_blk->blk_addr = cpu_to_le16(index);
490 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
491 orphan_blk->entry_count = cpu_to_le32(nentries);
492 set_page_dirty(page);
493 f2fs_put_page(page, 1);
496 spin_unlock(&sbi->ino_lock[ORPHAN_INO]);
499 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
500 block_t cp_addr, unsigned long long *version)
502 struct page *cp_page_1, *cp_page_2 = NULL;
503 unsigned long blk_size = sbi->blocksize;
504 struct f2fs_checkpoint *cp_block;
505 unsigned long long cur_version = 0, pre_version = 0;
509 /* Read the 1st cp block in this CP pack */
510 cp_page_1 = get_meta_page(sbi, cp_addr);
512 /* get the version number */
513 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
514 crc_offset = le32_to_cpu(cp_block->checksum_offset);
515 if (crc_offset >= blk_size)
518 crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
519 if (!f2fs_crc_valid(crc, cp_block, crc_offset))
522 pre_version = cur_cp_version(cp_block);
524 /* Read the 2nd cp block in this CP pack */
525 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
526 cp_page_2 = get_meta_page(sbi, cp_addr);
528 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
529 crc_offset = le32_to_cpu(cp_block->checksum_offset);
530 if (crc_offset >= blk_size)
533 crc = le32_to_cpu(*((__u32 *)((unsigned char *)cp_block + crc_offset)));
534 if (!f2fs_crc_valid(crc, cp_block, crc_offset))
537 cur_version = cur_cp_version(cp_block);
539 if (cur_version == pre_version) {
540 *version = cur_version;
541 f2fs_put_page(cp_page_2, 1);
545 f2fs_put_page(cp_page_2, 1);
547 f2fs_put_page(cp_page_1, 1);
551 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
553 struct f2fs_checkpoint *cp_block;
554 struct f2fs_super_block *fsb = sbi->raw_super;
555 struct page *cp1, *cp2, *cur_page;
556 unsigned long blk_size = sbi->blocksize;
557 unsigned long long cp1_version = 0, cp2_version = 0;
558 unsigned long long cp_start_blk_no;
559 unsigned int cp_blks = 1 + le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
563 sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
567 * Finding out valid cp block involves read both
568 * sets( cp pack1 and cp pack 2)
570 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
571 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
573 /* The second checkpoint pack should start at the next segment */
574 cp_start_blk_no += ((unsigned long long)1) <<
575 le32_to_cpu(fsb->log_blocks_per_seg);
576 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
579 if (ver_after(cp2_version, cp1_version))
591 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
592 memcpy(sbi->ckpt, cp_block, blk_size);
597 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
599 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
601 for (i = 1; i < cp_blks; i++) {
602 void *sit_bitmap_ptr;
603 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
605 cur_page = get_meta_page(sbi, cp_blk_no + i);
606 sit_bitmap_ptr = page_address(cur_page);
607 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
608 f2fs_put_page(cur_page, 1);
611 f2fs_put_page(cp1, 1);
612 f2fs_put_page(cp2, 1);
620 static int __add_dirty_inode(struct inode *inode, struct dir_inode_entry *new)
622 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
624 if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
627 set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
628 F2FS_I(inode)->dirty_dir = new;
629 list_add_tail(&new->list, &sbi->dir_inode_list);
630 stat_inc_dirty_dir(sbi);
634 void set_dirty_dir_page(struct inode *inode, struct page *page)
636 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
637 struct dir_inode_entry *new;
640 if (!S_ISDIR(inode->i_mode))
643 new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
645 INIT_LIST_HEAD(&new->list);
647 spin_lock(&sbi->dir_inode_lock);
648 ret = __add_dirty_inode(inode, new);
649 inode_inc_dirty_dents(inode);
650 SetPagePrivate(page);
651 spin_unlock(&sbi->dir_inode_lock);
654 kmem_cache_free(inode_entry_slab, new);
657 void add_dirty_dir_inode(struct inode *inode)
659 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
660 struct dir_inode_entry *new =
661 f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
665 INIT_LIST_HEAD(&new->list);
667 spin_lock(&sbi->dir_inode_lock);
668 ret = __add_dirty_inode(inode, new);
669 spin_unlock(&sbi->dir_inode_lock);
672 kmem_cache_free(inode_entry_slab, new);
675 void remove_dirty_dir_inode(struct inode *inode)
677 struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
678 struct dir_inode_entry *entry;
680 if (!S_ISDIR(inode->i_mode))
683 spin_lock(&sbi->dir_inode_lock);
684 if (get_dirty_dents(inode) ||
685 !is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
686 spin_unlock(&sbi->dir_inode_lock);
690 entry = F2FS_I(inode)->dirty_dir;
691 list_del(&entry->list);
692 F2FS_I(inode)->dirty_dir = NULL;
693 clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
694 stat_dec_dirty_dir(sbi);
695 spin_unlock(&sbi->dir_inode_lock);
696 kmem_cache_free(inode_entry_slab, entry);
698 /* Only from the recovery routine */
699 if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
700 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
705 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
707 struct list_head *head;
708 struct dir_inode_entry *entry;
711 spin_lock(&sbi->dir_inode_lock);
713 head = &sbi->dir_inode_list;
714 if (list_empty(head)) {
715 spin_unlock(&sbi->dir_inode_lock);
718 entry = list_entry(head->next, struct dir_inode_entry, list);
719 inode = igrab(entry->inode);
720 spin_unlock(&sbi->dir_inode_lock);
722 filemap_fdatawrite(inode->i_mapping);
726 * We should submit bio, since it exists several
727 * wribacking dentry pages in the freeing inode.
729 f2fs_submit_merged_bio(sbi, DATA, WRITE);
735 * Freeze all the FS-operations for checkpoint.
737 static int block_operations(struct f2fs_sb_info *sbi)
739 struct writeback_control wbc = {
740 .sync_mode = WB_SYNC_ALL,
741 .nr_to_write = LONG_MAX,
744 struct blk_plug plug;
747 blk_start_plug(&plug);
751 /* write all the dirty dentry pages */
752 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
753 f2fs_unlock_all(sbi);
754 sync_dirty_dir_inodes(sbi);
755 if (unlikely(f2fs_cp_error(sbi))) {
759 goto retry_flush_dents;
763 * POR: we should ensure that there are no dirty node pages
764 * until finishing nat/sit flush.
767 down_write(&sbi->node_write);
769 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
770 up_write(&sbi->node_write);
771 sync_node_pages(sbi, 0, &wbc);
772 if (unlikely(f2fs_cp_error(sbi))) {
773 f2fs_unlock_all(sbi);
777 goto retry_flush_nodes;
780 blk_finish_plug(&plug);
784 static void unblock_operations(struct f2fs_sb_info *sbi)
786 up_write(&sbi->node_write);
787 f2fs_unlock_all(sbi);
790 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
795 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
797 if (!get_pages(sbi, F2FS_WRITEBACK))
802 finish_wait(&sbi->cp_wait, &wait);
805 static void do_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
807 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
808 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
811 struct page *cp_page;
812 unsigned int data_sum_blocks, orphan_blocks;
816 int cp_payload_blks = le32_to_cpu(F2FS_RAW_SUPER(sbi)->cp_payload);
819 * This avoids to conduct wrong roll-forward operations and uses
820 * metapages, so should be called prior to sync_meta_pages below.
822 discard_next_dnode(sbi, NEXT_FREE_BLKADDR(sbi, curseg));
824 /* Flush all the NAT/SIT pages */
825 while (get_pages(sbi, F2FS_DIRTY_META)) {
826 sync_meta_pages(sbi, META, LONG_MAX);
827 if (unlikely(f2fs_cp_error(sbi)))
831 next_free_nid(sbi, &last_nid);
835 * version number is already updated
837 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
838 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
839 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
840 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
841 ckpt->cur_node_segno[i] =
842 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
843 ckpt->cur_node_blkoff[i] =
844 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
845 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
846 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
848 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
849 ckpt->cur_data_segno[i] =
850 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
851 ckpt->cur_data_blkoff[i] =
852 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
853 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
854 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
857 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
858 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
859 ckpt->next_free_nid = cpu_to_le32(last_nid);
861 /* 2 cp + n data seg summary + orphan inode blocks */
862 data_sum_blocks = npages_for_summary_flush(sbi);
863 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
864 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
866 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
868 orphan_blocks = GET_ORPHAN_BLOCKS(sbi->n_orphans);
869 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
873 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
874 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
875 cp_payload_blks + data_sum_blocks +
876 orphan_blocks + NR_CURSEG_NODE_TYPE);
878 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
879 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
880 cp_payload_blks + data_sum_blocks +
885 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
887 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
889 /* update SIT/NAT bitmap */
890 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
891 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
893 crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
894 *((__le32 *)((unsigned char *)ckpt +
895 le32_to_cpu(ckpt->checksum_offset)))
896 = cpu_to_le32(crc32);
898 start_blk = __start_cp_addr(sbi);
900 /* write out checkpoint buffer at block 0 */
901 cp_page = grab_meta_page(sbi, start_blk++);
902 kaddr = page_address(cp_page);
903 memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
904 set_page_dirty(cp_page);
905 f2fs_put_page(cp_page, 1);
907 for (i = 1; i < 1 + cp_payload_blks; i++) {
908 cp_page = grab_meta_page(sbi, start_blk++);
909 kaddr = page_address(cp_page);
910 memcpy(kaddr, (char *)ckpt + i * F2FS_BLKSIZE,
911 (1 << sbi->log_blocksize));
912 set_page_dirty(cp_page);
913 f2fs_put_page(cp_page, 1);
916 if (sbi->n_orphans) {
917 write_orphan_inodes(sbi, start_blk);
918 start_blk += orphan_blocks;
921 write_data_summaries(sbi, start_blk);
922 start_blk += data_sum_blocks;
924 write_node_summaries(sbi, start_blk);
925 start_blk += NR_CURSEG_NODE_TYPE;
928 /* writeout checkpoint block */
929 cp_page = grab_meta_page(sbi, start_blk);
930 kaddr = page_address(cp_page);
931 memcpy(kaddr, ckpt, (1 << sbi->log_blocksize));
932 set_page_dirty(cp_page);
933 f2fs_put_page(cp_page, 1);
935 /* wait for previous submitted node/meta pages writeback */
936 wait_on_all_pages_writeback(sbi);
938 if (unlikely(f2fs_cp_error(sbi)))
941 filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
942 filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
944 /* update user_block_counts */
945 sbi->last_valid_block_count = sbi->total_valid_block_count;
946 sbi->alloc_valid_block_count = 0;
948 /* Here, we only have one bio having CP pack */
949 sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
951 release_dirty_inode(sbi);
953 if (unlikely(f2fs_cp_error(sbi)))
956 clear_prefree_segments(sbi);
957 F2FS_RESET_SB_DIRT(sbi);
961 * We guarantee that this checkpoint procedure will not fail.
963 void write_checkpoint(struct f2fs_sb_info *sbi, bool is_umount)
965 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
966 unsigned long long ckpt_ver;
968 trace_f2fs_write_checkpoint(sbi->sb, is_umount, "start block_ops");
970 mutex_lock(&sbi->cp_mutex);
974 if (unlikely(f2fs_cp_error(sbi)))
976 if (block_operations(sbi))
979 trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish block_ops");
981 f2fs_submit_merged_bio(sbi, DATA, WRITE);
982 f2fs_submit_merged_bio(sbi, NODE, WRITE);
983 f2fs_submit_merged_bio(sbi, META, WRITE);
986 * update checkpoint pack index
987 * Increase the version number so that
988 * SIT entries and seg summaries are written at correct place
990 ckpt_ver = cur_cp_version(ckpt);
991 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
993 /* write cached NAT/SIT entries to NAT/SIT area */
994 flush_nat_entries(sbi);
995 flush_sit_entries(sbi);
997 /* unlock all the fs_lock[] in do_checkpoint() */
998 do_checkpoint(sbi, is_umount);
1000 unblock_operations(sbi);
1001 stat_inc_cp_count(sbi->stat_info);
1003 mutex_unlock(&sbi->cp_mutex);
1004 trace_f2fs_write_checkpoint(sbi->sb, is_umount, "finish checkpoint");
1007 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1011 for (i = 0; i < MAX_INO_ENTRY; i++) {
1012 INIT_RADIX_TREE(&sbi->ino_root[i], GFP_ATOMIC);
1013 spin_lock_init(&sbi->ino_lock[i]);
1014 INIT_LIST_HEAD(&sbi->ino_list[i]);
1018 * considering 512 blocks in a segment 8 blocks are needed for cp
1019 * and log segment summaries. Remaining blocks are used to keep
1020 * orphan entries with the limitation one reserved segment
1021 * for cp pack we can have max 1020*504 orphan entries
1024 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1025 NR_CURSEG_TYPE) * F2FS_ORPHANS_PER_BLOCK;
1028 int __init create_checkpoint_caches(void)
1030 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1031 sizeof(struct ino_entry));
1032 if (!ino_entry_slab)
1034 inode_entry_slab = f2fs_kmem_cache_create("f2fs_dirty_dir_entry",
1035 sizeof(struct dir_inode_entry));
1036 if (!inode_entry_slab) {
1037 kmem_cache_destroy(ino_entry_slab);
1043 void destroy_checkpoint_caches(void)
1045 kmem_cache_destroy(ino_entry_slab);
1046 kmem_cache_destroy(inode_entry_slab);