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 = grab_cache_page(mapping, index);
42 f2fs_wait_on_page_writeback(page, META);
43 SetPageUptodate(page);
48 * We guarantee no failure on the returned page.
50 struct page *get_meta_page(struct f2fs_sb_info *sbi, pgoff_t index)
52 struct address_space *mapping = META_MAPPING(sbi);
54 struct f2fs_io_info fio = {
56 .rw = READ_SYNC | REQ_META | REQ_PRIO,
60 page = grab_cache_page(mapping, index);
65 if (PageUptodate(page))
68 if (f2fs_submit_page_bio(sbi, page, &fio))
72 if (unlikely(page->mapping != mapping)) {
73 f2fs_put_page(page, 1);
80 static inline bool is_valid_blkaddr(struct f2fs_sb_info *sbi,
81 block_t blkaddr, int type)
87 if (unlikely(blkaddr >= SIT_BLK_CNT(sbi)))
91 if (unlikely(blkaddr >= MAIN_BLKADDR(sbi) ||
92 blkaddr < SM_I(sbi)->ssa_blkaddr))
96 if (unlikely(blkaddr >= SIT_I(sbi)->sit_base_addr ||
97 blkaddr < __start_cp_addr(sbi)))
101 if (unlikely(blkaddr >= MAX_BLKADDR(sbi) ||
102 blkaddr < MAIN_BLKADDR(sbi)))
113 * Readahead CP/NAT/SIT/SSA pages
115 int ra_meta_pages(struct f2fs_sb_info *sbi, block_t start, int nrpages, int type)
117 block_t prev_blk_addr = 0;
119 block_t blkno = start;
120 struct f2fs_io_info fio = {
122 .rw = READ_SYNC | REQ_META | REQ_PRIO
125 for (; nrpages-- > 0; blkno++) {
127 if (!is_valid_blkaddr(sbi, blkno, type))
132 if (unlikely(blkno >=
133 NAT_BLOCK_OFFSET(NM_I(sbi)->max_nid)))
135 /* get nat block addr */
136 fio.blk_addr = current_nat_addr(sbi,
137 blkno * NAT_ENTRY_PER_BLOCK);
140 /* get sit block addr */
141 fio.blk_addr = current_sit_addr(sbi,
142 blkno * SIT_ENTRY_PER_BLOCK);
143 if (blkno != start && prev_blk_addr + 1 != fio.blk_addr)
145 prev_blk_addr = fio.blk_addr;
150 fio.blk_addr = blkno;
156 page = grab_cache_page(META_MAPPING(sbi), fio.blk_addr);
159 if (PageUptodate(page)) {
160 f2fs_put_page(page, 1);
164 f2fs_submit_page_mbio(sbi, page, &fio);
165 f2fs_put_page(page, 0);
168 f2fs_submit_merged_bio(sbi, META, READ);
169 return blkno - start;
172 void ra_meta_pages_cond(struct f2fs_sb_info *sbi, pgoff_t index)
175 bool readahead = false;
177 page = find_get_page(META_MAPPING(sbi), index);
178 if (!page || (page && !PageUptodate(page)))
180 f2fs_put_page(page, 0);
183 ra_meta_pages(sbi, index, MAX_BIO_BLOCKS(sbi), META_POR);
186 static int f2fs_write_meta_page(struct page *page,
187 struct writeback_control *wbc)
189 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
191 trace_f2fs_writepage(page, META);
193 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
195 if (wbc->for_reclaim && page->index < GET_SUM_BLOCK(sbi, 0))
197 if (unlikely(f2fs_cp_error(sbi)))
200 f2fs_wait_on_page_writeback(page, META);
201 write_meta_page(sbi, page);
202 dec_page_count(sbi, F2FS_DIRTY_META);
205 if (wbc->for_reclaim)
206 f2fs_submit_merged_bio(sbi, META, WRITE);
210 redirty_page_for_writepage(wbc, page);
211 return AOP_WRITEPAGE_ACTIVATE;
214 static int f2fs_write_meta_pages(struct address_space *mapping,
215 struct writeback_control *wbc)
217 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
220 trace_f2fs_writepages(mapping->host, wbc, META);
222 /* collect a number of dirty meta pages and write together */
223 if (wbc->for_kupdate ||
224 get_pages(sbi, F2FS_DIRTY_META) < nr_pages_to_skip(sbi, META))
227 /* if mounting is failed, skip writing node pages */
228 mutex_lock(&sbi->cp_mutex);
229 diff = nr_pages_to_write(sbi, META, wbc);
230 written = sync_meta_pages(sbi, META, wbc->nr_to_write);
231 mutex_unlock(&sbi->cp_mutex);
232 wbc->nr_to_write = max((long)0, wbc->nr_to_write - written - diff);
236 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_META);
240 long sync_meta_pages(struct f2fs_sb_info *sbi, enum page_type type,
243 struct address_space *mapping = META_MAPPING(sbi);
244 pgoff_t index = 0, end = LONG_MAX;
247 struct writeback_control wbc = {
251 pagevec_init(&pvec, 0);
253 while (index <= end) {
255 nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
257 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
258 if (unlikely(nr_pages == 0))
261 for (i = 0; i < nr_pages; i++) {
262 struct page *page = pvec.pages[i];
266 if (unlikely(page->mapping != mapping)) {
271 if (!PageDirty(page)) {
272 /* someone wrote it for us */
273 goto continue_unlock;
276 if (!clear_page_dirty_for_io(page))
277 goto continue_unlock;
279 if (mapping->a_ops->writepage(page, &wbc)) {
284 if (unlikely(nwritten >= nr_to_write))
287 pagevec_release(&pvec);
292 f2fs_submit_merged_bio(sbi, type, WRITE);
297 static int f2fs_set_meta_page_dirty(struct page *page)
299 trace_f2fs_set_page_dirty(page, META);
301 SetPageUptodate(page);
302 if (!PageDirty(page)) {
303 __set_page_dirty_nobuffers(page);
304 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_META);
305 SetPagePrivate(page);
306 f2fs_trace_pid(page);
312 const struct address_space_operations f2fs_meta_aops = {
313 .writepage = f2fs_write_meta_page,
314 .writepages = f2fs_write_meta_pages,
315 .set_page_dirty = f2fs_set_meta_page_dirty,
316 .invalidatepage = f2fs_invalidate_page,
317 .releasepage = f2fs_release_page,
320 static void __add_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
322 struct inode_management *im = &sbi->im[type];
325 if (radix_tree_preload(GFP_NOFS)) {
330 spin_lock(&im->ino_lock);
332 e = radix_tree_lookup(&im->ino_root, ino);
334 e = kmem_cache_alloc(ino_entry_slab, GFP_ATOMIC);
336 spin_unlock(&im->ino_lock);
337 radix_tree_preload_end();
340 if (radix_tree_insert(&im->ino_root, ino, e)) {
341 spin_unlock(&im->ino_lock);
342 kmem_cache_free(ino_entry_slab, e);
343 radix_tree_preload_end();
346 memset(e, 0, sizeof(struct ino_entry));
349 list_add_tail(&e->list, &im->ino_list);
350 if (type != ORPHAN_INO)
353 spin_unlock(&im->ino_lock);
354 radix_tree_preload_end();
357 static void __remove_ino_entry(struct f2fs_sb_info *sbi, nid_t ino, int type)
359 struct inode_management *im = &sbi->im[type];
362 spin_lock(&im->ino_lock);
363 e = radix_tree_lookup(&im->ino_root, ino);
366 radix_tree_delete(&im->ino_root, ino);
368 spin_unlock(&im->ino_lock);
369 kmem_cache_free(ino_entry_slab, e);
372 spin_unlock(&im->ino_lock);
375 void add_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
377 /* add new dirty ino entry into list */
378 __add_ino_entry(sbi, ino, type);
381 void remove_dirty_inode(struct f2fs_sb_info *sbi, nid_t ino, int type)
383 /* remove dirty ino entry from list */
384 __remove_ino_entry(sbi, ino, type);
387 /* mode should be APPEND_INO or UPDATE_INO */
388 bool exist_written_data(struct f2fs_sb_info *sbi, nid_t ino, int mode)
390 struct inode_management *im = &sbi->im[mode];
393 spin_lock(&im->ino_lock);
394 e = radix_tree_lookup(&im->ino_root, ino);
395 spin_unlock(&im->ino_lock);
396 return e ? true : false;
399 void release_dirty_inode(struct f2fs_sb_info *sbi)
401 struct ino_entry *e, *tmp;
404 for (i = APPEND_INO; i <= UPDATE_INO; i++) {
405 struct inode_management *im = &sbi->im[i];
407 spin_lock(&im->ino_lock);
408 list_for_each_entry_safe(e, tmp, &im->ino_list, list) {
410 radix_tree_delete(&im->ino_root, e->ino);
411 kmem_cache_free(ino_entry_slab, e);
414 spin_unlock(&im->ino_lock);
418 int acquire_orphan_inode(struct f2fs_sb_info *sbi)
420 struct inode_management *im = &sbi->im[ORPHAN_INO];
423 spin_lock(&im->ino_lock);
424 if (unlikely(im->ino_num >= sbi->max_orphans))
428 spin_unlock(&im->ino_lock);
433 void release_orphan_inode(struct f2fs_sb_info *sbi)
435 struct inode_management *im = &sbi->im[ORPHAN_INO];
437 spin_lock(&im->ino_lock);
438 f2fs_bug_on(sbi, im->ino_num == 0);
440 spin_unlock(&im->ino_lock);
443 void add_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
445 /* add new orphan ino entry into list */
446 __add_ino_entry(sbi, ino, ORPHAN_INO);
449 void remove_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
451 /* remove orphan entry from orphan list */
452 __remove_ino_entry(sbi, ino, ORPHAN_INO);
455 static void recover_orphan_inode(struct f2fs_sb_info *sbi, nid_t ino)
457 struct inode *inode = f2fs_iget(sbi->sb, ino);
458 f2fs_bug_on(sbi, IS_ERR(inode));
461 /* truncate all the data during iput */
465 void recover_orphan_inodes(struct f2fs_sb_info *sbi)
467 block_t start_blk, orphan_blocks, i, j;
469 if (!is_set_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG))
472 set_sbi_flag(sbi, SBI_POR_DOING);
474 start_blk = __start_cp_addr(sbi) + 1 + __cp_payload(sbi);
475 orphan_blocks = __start_sum_addr(sbi) - 1 - __cp_payload(sbi);
477 ra_meta_pages(sbi, start_blk, orphan_blocks, META_CP);
479 for (i = 0; i < orphan_blocks; i++) {
480 struct page *page = get_meta_page(sbi, start_blk + i);
481 struct f2fs_orphan_block *orphan_blk;
483 orphan_blk = (struct f2fs_orphan_block *)page_address(page);
484 for (j = 0; j < le32_to_cpu(orphan_blk->entry_count); j++) {
485 nid_t ino = le32_to_cpu(orphan_blk->ino[j]);
486 recover_orphan_inode(sbi, ino);
488 f2fs_put_page(page, 1);
490 /* clear Orphan Flag */
491 clear_ckpt_flags(F2FS_CKPT(sbi), CP_ORPHAN_PRESENT_FLAG);
492 clear_sbi_flag(sbi, SBI_POR_DOING);
496 static void write_orphan_inodes(struct f2fs_sb_info *sbi, block_t start_blk)
498 struct list_head *head;
499 struct f2fs_orphan_block *orphan_blk = NULL;
500 unsigned int nentries = 0;
501 unsigned short index;
502 unsigned short orphan_blocks;
503 struct page *page = NULL;
504 struct ino_entry *orphan = NULL;
505 struct inode_management *im = &sbi->im[ORPHAN_INO];
507 orphan_blocks = GET_ORPHAN_BLOCKS(im->ino_num);
509 for (index = 0; index < orphan_blocks; index++)
510 grab_meta_page(sbi, start_blk + index);
513 spin_lock(&im->ino_lock);
514 head = &im->ino_list;
516 /* loop for each orphan inode entry and write them in Jornal block */
517 list_for_each_entry(orphan, head, list) {
519 page = find_get_page(META_MAPPING(sbi), start_blk++);
520 f2fs_bug_on(sbi, !page);
522 (struct f2fs_orphan_block *)page_address(page);
523 memset(orphan_blk, 0, sizeof(*orphan_blk));
524 f2fs_put_page(page, 0);
527 orphan_blk->ino[nentries++] = cpu_to_le32(orphan->ino);
529 if (nentries == F2FS_ORPHANS_PER_BLOCK) {
531 * an orphan block is full of 1020 entries,
532 * then we need to flush current orphan blocks
533 * and bring another one in memory
535 orphan_blk->blk_addr = cpu_to_le16(index);
536 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
537 orphan_blk->entry_count = cpu_to_le32(nentries);
538 set_page_dirty(page);
539 f2fs_put_page(page, 1);
547 orphan_blk->blk_addr = cpu_to_le16(index);
548 orphan_blk->blk_count = cpu_to_le16(orphan_blocks);
549 orphan_blk->entry_count = cpu_to_le32(nentries);
550 set_page_dirty(page);
551 f2fs_put_page(page, 1);
554 spin_unlock(&im->ino_lock);
557 static struct page *validate_checkpoint(struct f2fs_sb_info *sbi,
558 block_t cp_addr, unsigned long long *version)
560 struct page *cp_page_1, *cp_page_2 = NULL;
561 unsigned long blk_size = sbi->blocksize;
562 struct f2fs_checkpoint *cp_block;
563 unsigned long long cur_version = 0, pre_version = 0;
567 /* Read the 1st cp block in this CP pack */
568 cp_page_1 = get_meta_page(sbi, cp_addr);
570 /* get the version number */
571 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_1);
572 crc_offset = le32_to_cpu(cp_block->checksum_offset);
573 if (crc_offset >= blk_size)
576 crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
577 if (!f2fs_crc_valid(crc, cp_block, crc_offset))
580 pre_version = cur_cp_version(cp_block);
582 /* Read the 2nd cp block in this CP pack */
583 cp_addr += le32_to_cpu(cp_block->cp_pack_total_block_count) - 1;
584 cp_page_2 = get_meta_page(sbi, cp_addr);
586 cp_block = (struct f2fs_checkpoint *)page_address(cp_page_2);
587 crc_offset = le32_to_cpu(cp_block->checksum_offset);
588 if (crc_offset >= blk_size)
591 crc = le32_to_cpu(*((__le32 *)((unsigned char *)cp_block + crc_offset)));
592 if (!f2fs_crc_valid(crc, cp_block, crc_offset))
595 cur_version = cur_cp_version(cp_block);
597 if (cur_version == pre_version) {
598 *version = cur_version;
599 f2fs_put_page(cp_page_2, 1);
603 f2fs_put_page(cp_page_2, 1);
605 f2fs_put_page(cp_page_1, 1);
609 int get_valid_checkpoint(struct f2fs_sb_info *sbi)
611 struct f2fs_checkpoint *cp_block;
612 struct f2fs_super_block *fsb = sbi->raw_super;
613 struct page *cp1, *cp2, *cur_page;
614 unsigned long blk_size = sbi->blocksize;
615 unsigned long long cp1_version = 0, cp2_version = 0;
616 unsigned long long cp_start_blk_no;
617 unsigned int cp_blks = 1 + __cp_payload(sbi);
621 sbi->ckpt = kzalloc(cp_blks * blk_size, GFP_KERNEL);
625 * Finding out valid cp block involves read both
626 * sets( cp pack1 and cp pack 2)
628 cp_start_blk_no = le32_to_cpu(fsb->cp_blkaddr);
629 cp1 = validate_checkpoint(sbi, cp_start_blk_no, &cp1_version);
631 /* The second checkpoint pack should start at the next segment */
632 cp_start_blk_no += ((unsigned long long)1) <<
633 le32_to_cpu(fsb->log_blocks_per_seg);
634 cp2 = validate_checkpoint(sbi, cp_start_blk_no, &cp2_version);
637 if (ver_after(cp2_version, cp1_version))
649 cp_block = (struct f2fs_checkpoint *)page_address(cur_page);
650 memcpy(sbi->ckpt, cp_block, blk_size);
655 cp_blk_no = le32_to_cpu(fsb->cp_blkaddr);
657 cp_blk_no += 1 << le32_to_cpu(fsb->log_blocks_per_seg);
659 for (i = 1; i < cp_blks; i++) {
660 void *sit_bitmap_ptr;
661 unsigned char *ckpt = (unsigned char *)sbi->ckpt;
663 cur_page = get_meta_page(sbi, cp_blk_no + i);
664 sit_bitmap_ptr = page_address(cur_page);
665 memcpy(ckpt + i * blk_size, sit_bitmap_ptr, blk_size);
666 f2fs_put_page(cur_page, 1);
669 f2fs_put_page(cp1, 1);
670 f2fs_put_page(cp2, 1);
678 static int __add_dirty_inode(struct inode *inode, struct inode_entry *new)
680 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
682 if (is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR))
685 set_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
686 F2FS_I(inode)->dirty_dir = new;
687 list_add_tail(&new->list, &sbi->dir_inode_list);
688 stat_inc_dirty_dir(sbi);
692 void update_dirty_page(struct inode *inode, struct page *page)
694 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
695 struct inode_entry *new;
698 if (!S_ISDIR(inode->i_mode) && !S_ISREG(inode->i_mode))
701 if (!S_ISDIR(inode->i_mode)) {
702 inode_inc_dirty_pages(inode);
706 new = f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
708 INIT_LIST_HEAD(&new->list);
710 spin_lock(&sbi->dir_inode_lock);
711 ret = __add_dirty_inode(inode, new);
712 inode_inc_dirty_pages(inode);
713 spin_unlock(&sbi->dir_inode_lock);
716 kmem_cache_free(inode_entry_slab, new);
718 SetPagePrivate(page);
719 f2fs_trace_pid(page);
722 void add_dirty_dir_inode(struct inode *inode)
724 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
725 struct inode_entry *new =
726 f2fs_kmem_cache_alloc(inode_entry_slab, GFP_NOFS);
730 INIT_LIST_HEAD(&new->list);
732 spin_lock(&sbi->dir_inode_lock);
733 ret = __add_dirty_inode(inode, new);
734 spin_unlock(&sbi->dir_inode_lock);
737 kmem_cache_free(inode_entry_slab, new);
740 void remove_dirty_dir_inode(struct inode *inode)
742 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
743 struct inode_entry *entry;
745 if (!S_ISDIR(inode->i_mode))
748 spin_lock(&sbi->dir_inode_lock);
749 if (get_dirty_pages(inode) ||
750 !is_inode_flag_set(F2FS_I(inode), FI_DIRTY_DIR)) {
751 spin_unlock(&sbi->dir_inode_lock);
755 entry = F2FS_I(inode)->dirty_dir;
756 list_del(&entry->list);
757 F2FS_I(inode)->dirty_dir = NULL;
758 clear_inode_flag(F2FS_I(inode), FI_DIRTY_DIR);
759 stat_dec_dirty_dir(sbi);
760 spin_unlock(&sbi->dir_inode_lock);
761 kmem_cache_free(inode_entry_slab, entry);
763 /* Only from the recovery routine */
764 if (is_inode_flag_set(F2FS_I(inode), FI_DELAY_IPUT)) {
765 clear_inode_flag(F2FS_I(inode), FI_DELAY_IPUT);
770 void sync_dirty_dir_inodes(struct f2fs_sb_info *sbi)
772 struct list_head *head;
773 struct inode_entry *entry;
776 if (unlikely(f2fs_cp_error(sbi)))
779 spin_lock(&sbi->dir_inode_lock);
781 head = &sbi->dir_inode_list;
782 if (list_empty(head)) {
783 spin_unlock(&sbi->dir_inode_lock);
786 entry = list_entry(head->next, struct inode_entry, list);
787 inode = igrab(entry->inode);
788 spin_unlock(&sbi->dir_inode_lock);
790 filemap_fdatawrite(inode->i_mapping);
794 * We should submit bio, since it exists several
795 * wribacking dentry pages in the freeing inode.
797 f2fs_submit_merged_bio(sbi, DATA, WRITE);
803 * Freeze all the FS-operations for checkpoint.
805 static int block_operations(struct f2fs_sb_info *sbi)
807 struct writeback_control wbc = {
808 .sync_mode = WB_SYNC_ALL,
809 .nr_to_write = LONG_MAX,
812 struct blk_plug plug;
815 blk_start_plug(&plug);
819 /* write all the dirty dentry pages */
820 if (get_pages(sbi, F2FS_DIRTY_DENTS)) {
821 f2fs_unlock_all(sbi);
822 sync_dirty_dir_inodes(sbi);
823 if (unlikely(f2fs_cp_error(sbi))) {
827 goto retry_flush_dents;
831 * POR: we should ensure that there are no dirty node pages
832 * until finishing nat/sit flush.
835 down_write(&sbi->node_write);
837 if (get_pages(sbi, F2FS_DIRTY_NODES)) {
838 up_write(&sbi->node_write);
839 sync_node_pages(sbi, 0, &wbc);
840 if (unlikely(f2fs_cp_error(sbi))) {
841 f2fs_unlock_all(sbi);
845 goto retry_flush_nodes;
848 blk_finish_plug(&plug);
852 static void unblock_operations(struct f2fs_sb_info *sbi)
854 up_write(&sbi->node_write);
855 f2fs_unlock_all(sbi);
858 static void wait_on_all_pages_writeback(struct f2fs_sb_info *sbi)
863 prepare_to_wait(&sbi->cp_wait, &wait, TASK_UNINTERRUPTIBLE);
865 if (!get_pages(sbi, F2FS_WRITEBACK))
870 finish_wait(&sbi->cp_wait, &wait);
873 static void do_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
875 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
876 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
877 struct f2fs_nm_info *nm_i = NM_I(sbi);
878 unsigned long orphan_num = sbi->im[ORPHAN_INO].ino_num;
879 nid_t last_nid = nm_i->next_scan_nid;
881 struct page *cp_page;
882 unsigned int data_sum_blocks, orphan_blocks;
886 int cp_payload_blks = __cp_payload(sbi);
889 * This avoids to conduct wrong roll-forward operations and uses
890 * metapages, so should be called prior to sync_meta_pages below.
892 discard_next_dnode(sbi, NEXT_FREE_BLKADDR(sbi, curseg));
894 /* Flush all the NAT/SIT pages */
895 while (get_pages(sbi, F2FS_DIRTY_META)) {
896 sync_meta_pages(sbi, META, LONG_MAX);
897 if (unlikely(f2fs_cp_error(sbi)))
901 next_free_nid(sbi, &last_nid);
905 * version number is already updated
907 ckpt->elapsed_time = cpu_to_le64(get_mtime(sbi));
908 ckpt->valid_block_count = cpu_to_le64(valid_user_blocks(sbi));
909 ckpt->free_segment_count = cpu_to_le32(free_segments(sbi));
910 for (i = 0; i < NR_CURSEG_NODE_TYPE; i++) {
911 ckpt->cur_node_segno[i] =
912 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_NODE));
913 ckpt->cur_node_blkoff[i] =
914 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_NODE));
915 ckpt->alloc_type[i + CURSEG_HOT_NODE] =
916 curseg_alloc_type(sbi, i + CURSEG_HOT_NODE);
918 for (i = 0; i < NR_CURSEG_DATA_TYPE; i++) {
919 ckpt->cur_data_segno[i] =
920 cpu_to_le32(curseg_segno(sbi, i + CURSEG_HOT_DATA));
921 ckpt->cur_data_blkoff[i] =
922 cpu_to_le16(curseg_blkoff(sbi, i + CURSEG_HOT_DATA));
923 ckpt->alloc_type[i + CURSEG_HOT_DATA] =
924 curseg_alloc_type(sbi, i + CURSEG_HOT_DATA);
927 ckpt->valid_node_count = cpu_to_le32(valid_node_count(sbi));
928 ckpt->valid_inode_count = cpu_to_le32(valid_inode_count(sbi));
929 ckpt->next_free_nid = cpu_to_le32(last_nid);
931 /* 2 cp + n data seg summary + orphan inode blocks */
932 data_sum_blocks = npages_for_summary_flush(sbi, false);
933 if (data_sum_blocks < NR_CURSEG_DATA_TYPE)
934 set_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
936 clear_ckpt_flags(ckpt, CP_COMPACT_SUM_FLAG);
938 orphan_blocks = GET_ORPHAN_BLOCKS(orphan_num);
939 ckpt->cp_pack_start_sum = cpu_to_le32(1 + cp_payload_blks +
942 if (__remain_node_summaries(cpc->reason))
943 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS+
944 cp_payload_blks + data_sum_blocks +
945 orphan_blocks + NR_CURSEG_NODE_TYPE);
947 ckpt->cp_pack_total_block_count = cpu_to_le32(F2FS_CP_PACKS +
948 cp_payload_blks + data_sum_blocks +
951 if (cpc->reason == CP_UMOUNT)
952 set_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
954 clear_ckpt_flags(ckpt, CP_UMOUNT_FLAG);
956 if (cpc->reason == CP_FASTBOOT)
957 set_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
959 clear_ckpt_flags(ckpt, CP_FASTBOOT_FLAG);
962 set_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
964 clear_ckpt_flags(ckpt, CP_ORPHAN_PRESENT_FLAG);
966 if (is_sbi_flag_set(sbi, SBI_NEED_FSCK))
967 set_ckpt_flags(ckpt, CP_FSCK_FLAG);
969 /* update SIT/NAT bitmap */
970 get_sit_bitmap(sbi, __bitmap_ptr(sbi, SIT_BITMAP));
971 get_nat_bitmap(sbi, __bitmap_ptr(sbi, NAT_BITMAP));
973 crc32 = f2fs_crc32(ckpt, le32_to_cpu(ckpt->checksum_offset));
974 *((__le32 *)((unsigned char *)ckpt +
975 le32_to_cpu(ckpt->checksum_offset)))
976 = cpu_to_le32(crc32);
978 start_blk = __start_cp_addr(sbi);
980 /* write out checkpoint buffer at block 0 */
981 cp_page = grab_meta_page(sbi, start_blk++);
982 kaddr = page_address(cp_page);
983 memcpy(kaddr, ckpt, F2FS_BLKSIZE);
984 set_page_dirty(cp_page);
985 f2fs_put_page(cp_page, 1);
987 for (i = 1; i < 1 + cp_payload_blks; i++) {
988 cp_page = grab_meta_page(sbi, start_blk++);
989 kaddr = page_address(cp_page);
990 memcpy(kaddr, (char *)ckpt + i * F2FS_BLKSIZE, F2FS_BLKSIZE);
991 set_page_dirty(cp_page);
992 f2fs_put_page(cp_page, 1);
996 write_orphan_inodes(sbi, start_blk);
997 start_blk += orphan_blocks;
1000 write_data_summaries(sbi, start_blk);
1001 start_blk += data_sum_blocks;
1002 if (__remain_node_summaries(cpc->reason)) {
1003 write_node_summaries(sbi, start_blk);
1004 start_blk += NR_CURSEG_NODE_TYPE;
1007 /* writeout checkpoint block */
1008 cp_page = grab_meta_page(sbi, start_blk);
1009 kaddr = page_address(cp_page);
1010 memcpy(kaddr, ckpt, F2FS_BLKSIZE);
1011 set_page_dirty(cp_page);
1012 f2fs_put_page(cp_page, 1);
1014 /* wait for previous submitted node/meta pages writeback */
1015 wait_on_all_pages_writeback(sbi);
1017 if (unlikely(f2fs_cp_error(sbi)))
1020 filemap_fdatawait_range(NODE_MAPPING(sbi), 0, LONG_MAX);
1021 filemap_fdatawait_range(META_MAPPING(sbi), 0, LONG_MAX);
1023 /* update user_block_counts */
1024 sbi->last_valid_block_count = sbi->total_valid_block_count;
1025 sbi->alloc_valid_block_count = 0;
1027 /* Here, we only have one bio having CP pack */
1028 sync_meta_pages(sbi, META_FLUSH, LONG_MAX);
1030 /* wait for previous submitted meta pages writeback */
1031 wait_on_all_pages_writeback(sbi);
1033 release_dirty_inode(sbi);
1035 if (unlikely(f2fs_cp_error(sbi)))
1038 clear_prefree_segments(sbi);
1039 clear_sbi_flag(sbi, SBI_IS_DIRTY);
1043 * We guarantee that this checkpoint procedure will not fail.
1045 void write_checkpoint(struct f2fs_sb_info *sbi, struct cp_control *cpc)
1047 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
1048 unsigned long long ckpt_ver;
1050 mutex_lock(&sbi->cp_mutex);
1052 if (!is_sbi_flag_set(sbi, SBI_IS_DIRTY) &&
1053 cpc->reason != CP_DISCARD && cpc->reason != CP_UMOUNT)
1055 if (unlikely(f2fs_cp_error(sbi)))
1057 if (f2fs_readonly(sbi->sb))
1060 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "start block_ops");
1062 if (block_operations(sbi))
1065 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish block_ops");
1067 f2fs_submit_merged_bio(sbi, DATA, WRITE);
1068 f2fs_submit_merged_bio(sbi, NODE, WRITE);
1069 f2fs_submit_merged_bio(sbi, META, WRITE);
1072 * update checkpoint pack index
1073 * Increase the version number so that
1074 * SIT entries and seg summaries are written at correct place
1076 ckpt_ver = cur_cp_version(ckpt);
1077 ckpt->checkpoint_ver = cpu_to_le64(++ckpt_ver);
1079 /* write cached NAT/SIT entries to NAT/SIT area */
1080 flush_nat_entries(sbi);
1081 flush_sit_entries(sbi, cpc);
1083 /* unlock all the fs_lock[] in do_checkpoint() */
1084 do_checkpoint(sbi, cpc);
1086 unblock_operations(sbi);
1087 stat_inc_cp_count(sbi->stat_info);
1089 mutex_unlock(&sbi->cp_mutex);
1090 trace_f2fs_write_checkpoint(sbi->sb, cpc->reason, "finish checkpoint");
1093 void init_ino_entry_info(struct f2fs_sb_info *sbi)
1097 for (i = 0; i < MAX_INO_ENTRY; i++) {
1098 struct inode_management *im = &sbi->im[i];
1100 INIT_RADIX_TREE(&im->ino_root, GFP_ATOMIC);
1101 spin_lock_init(&im->ino_lock);
1102 INIT_LIST_HEAD(&im->ino_list);
1107 * considering 512 blocks in a segment 8 blocks are needed for cp
1108 * and log segment summaries. Remaining blocks are used to keep
1109 * orphan entries with the limitation one reserved segment
1110 * for cp pack we can have max 1020*504 orphan entries
1112 sbi->max_orphans = (sbi->blocks_per_seg - F2FS_CP_PACKS -
1113 NR_CURSEG_TYPE) * F2FS_ORPHANS_PER_BLOCK;
1116 int __init create_checkpoint_caches(void)
1118 ino_entry_slab = f2fs_kmem_cache_create("f2fs_ino_entry",
1119 sizeof(struct ino_entry));
1120 if (!ino_entry_slab)
1122 inode_entry_slab = f2fs_kmem_cache_create("f2fs_inode_entry",
1123 sizeof(struct inode_entry));
1124 if (!inode_entry_slab) {
1125 kmem_cache_destroy(ino_entry_slab);
1131 void destroy_checkpoint_caches(void)
1133 kmem_cache_destroy(ino_entry_slab);
1134 kmem_cache_destroy(inode_entry_slab);