1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
9 #include <linux/f2fs_fs.h>
10 #include <linux/mpage.h>
11 #include <linux/backing-dev.h>
12 #include <linux/blkdev.h>
13 #include <linux/pagevec.h>
14 #include <linux/swap.h>
20 #include <trace/events/f2fs.h>
22 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
24 static struct kmem_cache *nat_entry_slab;
25 static struct kmem_cache *free_nid_slab;
26 static struct kmem_cache *nat_entry_set_slab;
27 static struct kmem_cache *fsync_node_entry_slab;
30 * Check whether the given nid is within node id range.
32 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
34 if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
35 set_sbi_flag(sbi, SBI_NEED_FSCK);
36 f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
43 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
45 struct f2fs_nm_info *nm_i = NM_I(sbi);
46 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
48 unsigned long avail_ram;
49 unsigned long mem_size = 0;
57 /* only uses low memory */
58 avail_ram = val.totalram - val.totalhigh;
61 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
63 if (type == FREE_NIDS) {
64 mem_size = (nm_i->nid_cnt[FREE_NID] *
65 sizeof(struct free_nid)) >> PAGE_SHIFT;
66 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
67 } else if (type == NAT_ENTRIES) {
68 mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
69 sizeof(struct nat_entry)) >> PAGE_SHIFT;
70 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
71 if (excess_cached_nats(sbi))
73 } else if (type == DIRTY_DENTS) {
74 if (sbi->sb->s_bdi->wb.dirty_exceeded)
76 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
77 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
78 } else if (type == INO_ENTRIES) {
81 for (i = 0; i < MAX_INO_ENTRY; i++)
82 mem_size += sbi->im[i].ino_num *
83 sizeof(struct ino_entry);
84 mem_size >>= PAGE_SHIFT;
85 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
86 } else if (type == EXTENT_CACHE) {
87 mem_size = (atomic_read(&sbi->total_ext_tree) *
88 sizeof(struct extent_tree) +
89 atomic_read(&sbi->total_ext_node) *
90 sizeof(struct extent_node)) >> PAGE_SHIFT;
91 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
92 } else if (type == INMEM_PAGES) {
93 /* it allows 20% / total_ram for inmemory pages */
94 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
95 res = mem_size < (val.totalram / 5);
96 } else if (type == DISCARD_CACHE) {
97 mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
98 sizeof(struct discard_cmd)) >> PAGE_SHIFT;
99 res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
100 } else if (type == COMPRESS_PAGE) {
101 #ifdef CONFIG_F2FS_FS_COMPRESSION
102 unsigned long free_ram = val.freeram;
105 * free memory is lower than watermark or cached page count
106 * exceed threshold, deny caching compress page.
108 res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
109 (COMPRESS_MAPPING(sbi)->nrpages <
110 free_ram * sbi->compress_percent / 100);
115 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
121 static void clear_node_page_dirty(struct page *page)
123 if (PageDirty(page)) {
124 f2fs_clear_page_cache_dirty_tag(page);
125 clear_page_dirty_for_io(page);
126 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
128 ClearPageUptodate(page);
131 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
133 return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
136 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
138 struct page *src_page;
139 struct page *dst_page;
143 struct f2fs_nm_info *nm_i = NM_I(sbi);
145 dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
147 /* get current nat block page with lock */
148 src_page = get_current_nat_page(sbi, nid);
149 if (IS_ERR(src_page))
151 dst_page = f2fs_grab_meta_page(sbi, dst_off);
152 f2fs_bug_on(sbi, PageDirty(src_page));
154 src_addr = page_address(src_page);
155 dst_addr = page_address(dst_page);
156 memcpy(dst_addr, src_addr, PAGE_SIZE);
157 set_page_dirty(dst_page);
158 f2fs_put_page(src_page, 1);
160 set_to_next_nat(nm_i, nid);
165 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
167 struct nat_entry *new;
170 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
172 new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
174 nat_set_nid(new, nid);
180 static void __free_nat_entry(struct nat_entry *e)
182 kmem_cache_free(nat_entry_slab, e);
185 /* must be locked by nat_tree_lock */
186 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
187 struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
190 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
191 else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
195 node_info_from_raw_nat(&ne->ni, raw_ne);
197 spin_lock(&nm_i->nat_list_lock);
198 list_add_tail(&ne->list, &nm_i->nat_entries);
199 spin_unlock(&nm_i->nat_list_lock);
201 nm_i->nat_cnt[TOTAL_NAT]++;
202 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
206 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
208 struct nat_entry *ne;
210 ne = radix_tree_lookup(&nm_i->nat_root, n);
212 /* for recent accessed nat entry, move it to tail of lru list */
213 if (ne && !get_nat_flag(ne, IS_DIRTY)) {
214 spin_lock(&nm_i->nat_list_lock);
215 if (!list_empty(&ne->list))
216 list_move_tail(&ne->list, &nm_i->nat_entries);
217 spin_unlock(&nm_i->nat_list_lock);
223 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
224 nid_t start, unsigned int nr, struct nat_entry **ep)
226 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
229 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
231 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
232 nm_i->nat_cnt[TOTAL_NAT]--;
233 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
237 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
238 struct nat_entry *ne)
240 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
241 struct nat_entry_set *head;
243 head = radix_tree_lookup(&nm_i->nat_set_root, set);
245 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
247 INIT_LIST_HEAD(&head->entry_list);
248 INIT_LIST_HEAD(&head->set_list);
251 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
256 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
257 struct nat_entry *ne)
259 struct nat_entry_set *head;
260 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
263 head = __grab_nat_entry_set(nm_i, ne);
266 * update entry_cnt in below condition:
267 * 1. update NEW_ADDR to valid block address;
268 * 2. update old block address to new one;
270 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
271 !get_nat_flag(ne, IS_DIRTY)))
274 set_nat_flag(ne, IS_PREALLOC, new_ne);
276 if (get_nat_flag(ne, IS_DIRTY))
279 nm_i->nat_cnt[DIRTY_NAT]++;
280 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
281 set_nat_flag(ne, IS_DIRTY, true);
283 spin_lock(&nm_i->nat_list_lock);
285 list_del_init(&ne->list);
287 list_move_tail(&ne->list, &head->entry_list);
288 spin_unlock(&nm_i->nat_list_lock);
291 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
292 struct nat_entry_set *set, struct nat_entry *ne)
294 spin_lock(&nm_i->nat_list_lock);
295 list_move_tail(&ne->list, &nm_i->nat_entries);
296 spin_unlock(&nm_i->nat_list_lock);
298 set_nat_flag(ne, IS_DIRTY, false);
300 nm_i->nat_cnt[DIRTY_NAT]--;
301 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
304 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
305 nid_t start, unsigned int nr, struct nat_entry_set **ep)
307 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
311 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
313 return NODE_MAPPING(sbi) == page->mapping &&
314 IS_DNODE(page) && is_cold_node(page);
317 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
319 spin_lock_init(&sbi->fsync_node_lock);
320 INIT_LIST_HEAD(&sbi->fsync_node_list);
321 sbi->fsync_seg_id = 0;
322 sbi->fsync_node_num = 0;
325 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
328 struct fsync_node_entry *fn;
332 fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS);
336 INIT_LIST_HEAD(&fn->list);
338 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
339 list_add_tail(&fn->list, &sbi->fsync_node_list);
340 fn->seq_id = sbi->fsync_seg_id++;
342 sbi->fsync_node_num++;
343 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
348 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
350 struct fsync_node_entry *fn;
353 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
354 list_for_each_entry(fn, &sbi->fsync_node_list, list) {
355 if (fn->page == page) {
357 sbi->fsync_node_num--;
358 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
359 kmem_cache_free(fsync_node_entry_slab, fn);
364 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
368 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
372 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
373 sbi->fsync_seg_id = 0;
374 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
377 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
379 struct f2fs_nm_info *nm_i = NM_I(sbi);
383 down_read(&nm_i->nat_tree_lock);
384 e = __lookup_nat_cache(nm_i, nid);
386 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
387 !get_nat_flag(e, HAS_FSYNCED_INODE))
390 up_read(&nm_i->nat_tree_lock);
394 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
396 struct f2fs_nm_info *nm_i = NM_I(sbi);
400 down_read(&nm_i->nat_tree_lock);
401 e = __lookup_nat_cache(nm_i, nid);
402 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
404 up_read(&nm_i->nat_tree_lock);
408 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
410 struct f2fs_nm_info *nm_i = NM_I(sbi);
412 bool need_update = true;
414 down_read(&nm_i->nat_tree_lock);
415 e = __lookup_nat_cache(nm_i, ino);
416 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
417 (get_nat_flag(e, IS_CHECKPOINTED) ||
418 get_nat_flag(e, HAS_FSYNCED_INODE)))
420 up_read(&nm_i->nat_tree_lock);
424 /* must be locked by nat_tree_lock */
425 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
426 struct f2fs_nat_entry *ne)
428 struct f2fs_nm_info *nm_i = NM_I(sbi);
429 struct nat_entry *new, *e;
431 new = __alloc_nat_entry(nid, false);
435 down_write(&nm_i->nat_tree_lock);
436 e = __lookup_nat_cache(nm_i, nid);
438 e = __init_nat_entry(nm_i, new, ne, false);
440 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
441 nat_get_blkaddr(e) !=
442 le32_to_cpu(ne->block_addr) ||
443 nat_get_version(e) != ne->version);
444 up_write(&nm_i->nat_tree_lock);
446 __free_nat_entry(new);
449 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
450 block_t new_blkaddr, bool fsync_done)
452 struct f2fs_nm_info *nm_i = NM_I(sbi);
454 struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
456 down_write(&nm_i->nat_tree_lock);
457 e = __lookup_nat_cache(nm_i, ni->nid);
459 e = __init_nat_entry(nm_i, new, NULL, true);
460 copy_node_info(&e->ni, ni);
461 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
462 } else if (new_blkaddr == NEW_ADDR) {
464 * when nid is reallocated,
465 * previous nat entry can be remained in nat cache.
466 * So, reinitialize it with new information.
468 copy_node_info(&e->ni, ni);
469 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
471 /* let's free early to reduce memory consumption */
473 __free_nat_entry(new);
476 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
477 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
478 new_blkaddr == NULL_ADDR);
479 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
480 new_blkaddr == NEW_ADDR);
481 f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
482 new_blkaddr == NEW_ADDR);
484 /* increment version no as node is removed */
485 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
486 unsigned char version = nat_get_version(e);
488 nat_set_version(e, inc_node_version(version));
492 nat_set_blkaddr(e, new_blkaddr);
493 if (!__is_valid_data_blkaddr(new_blkaddr))
494 set_nat_flag(e, IS_CHECKPOINTED, false);
495 __set_nat_cache_dirty(nm_i, e);
497 /* update fsync_mark if its inode nat entry is still alive */
498 if (ni->nid != ni->ino)
499 e = __lookup_nat_cache(nm_i, ni->ino);
501 if (fsync_done && ni->nid == ni->ino)
502 set_nat_flag(e, HAS_FSYNCED_INODE, true);
503 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
505 up_write(&nm_i->nat_tree_lock);
508 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
510 struct f2fs_nm_info *nm_i = NM_I(sbi);
513 if (!down_write_trylock(&nm_i->nat_tree_lock))
516 spin_lock(&nm_i->nat_list_lock);
518 struct nat_entry *ne;
520 if (list_empty(&nm_i->nat_entries))
523 ne = list_first_entry(&nm_i->nat_entries,
524 struct nat_entry, list);
526 spin_unlock(&nm_i->nat_list_lock);
528 __del_from_nat_cache(nm_i, ne);
531 spin_lock(&nm_i->nat_list_lock);
533 spin_unlock(&nm_i->nat_list_lock);
535 up_write(&nm_i->nat_tree_lock);
536 return nr - nr_shrink;
539 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
540 struct node_info *ni)
542 struct f2fs_nm_info *nm_i = NM_I(sbi);
543 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
544 struct f2fs_journal *journal = curseg->journal;
545 nid_t start_nid = START_NID(nid);
546 struct f2fs_nat_block *nat_blk;
547 struct page *page = NULL;
548 struct f2fs_nat_entry ne;
556 /* Check nat cache */
557 down_read(&nm_i->nat_tree_lock);
558 e = __lookup_nat_cache(nm_i, nid);
560 ni->ino = nat_get_ino(e);
561 ni->blk_addr = nat_get_blkaddr(e);
562 ni->version = nat_get_version(e);
563 up_read(&nm_i->nat_tree_lock);
567 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
569 /* Check current segment summary */
570 down_read(&curseg->journal_rwsem);
571 i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
573 ne = nat_in_journal(journal, i);
574 node_info_from_raw_nat(ni, &ne);
576 up_read(&curseg->journal_rwsem);
578 up_read(&nm_i->nat_tree_lock);
582 /* Fill node_info from nat page */
583 index = current_nat_addr(sbi, nid);
584 up_read(&nm_i->nat_tree_lock);
586 page = f2fs_get_meta_page(sbi, index);
588 return PTR_ERR(page);
590 nat_blk = (struct f2fs_nat_block *)page_address(page);
591 ne = nat_blk->entries[nid - start_nid];
592 node_info_from_raw_nat(ni, &ne);
593 f2fs_put_page(page, 1);
595 blkaddr = le32_to_cpu(ne.block_addr);
596 if (__is_valid_data_blkaddr(blkaddr) &&
597 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
600 /* cache nat entry */
601 cache_nat_entry(sbi, nid, &ne);
606 * readahead MAX_RA_NODE number of node pages.
608 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
610 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
611 struct blk_plug plug;
615 blk_start_plug(&plug);
617 /* Then, try readahead for siblings of the desired node */
619 end = min(end, NIDS_PER_BLOCK);
620 for (i = start; i < end; i++) {
621 nid = get_nid(parent, i, false);
622 f2fs_ra_node_page(sbi, nid);
625 blk_finish_plug(&plug);
628 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
630 const long direct_index = ADDRS_PER_INODE(dn->inode);
631 const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
632 const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
633 unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
634 int cur_level = dn->cur_level;
635 int max_level = dn->max_level;
641 while (max_level-- > cur_level)
642 skipped_unit *= NIDS_PER_BLOCK;
644 switch (dn->max_level) {
646 base += 2 * indirect_blks;
649 base += 2 * direct_blks;
652 base += direct_index;
655 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
658 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
662 * The maximum depth is four.
663 * Offset[0] will have raw inode offset.
665 static int get_node_path(struct inode *inode, long block,
666 int offset[4], unsigned int noffset[4])
668 const long direct_index = ADDRS_PER_INODE(inode);
669 const long direct_blks = ADDRS_PER_BLOCK(inode);
670 const long dptrs_per_blk = NIDS_PER_BLOCK;
671 const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
672 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
678 if (block < direct_index) {
682 block -= direct_index;
683 if (block < direct_blks) {
684 offset[n++] = NODE_DIR1_BLOCK;
690 block -= direct_blks;
691 if (block < direct_blks) {
692 offset[n++] = NODE_DIR2_BLOCK;
698 block -= direct_blks;
699 if (block < indirect_blks) {
700 offset[n++] = NODE_IND1_BLOCK;
702 offset[n++] = block / direct_blks;
703 noffset[n] = 4 + offset[n - 1];
704 offset[n] = block % direct_blks;
708 block -= indirect_blks;
709 if (block < indirect_blks) {
710 offset[n++] = NODE_IND2_BLOCK;
711 noffset[n] = 4 + dptrs_per_blk;
712 offset[n++] = block / direct_blks;
713 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
714 offset[n] = block % direct_blks;
718 block -= indirect_blks;
719 if (block < dindirect_blks) {
720 offset[n++] = NODE_DIND_BLOCK;
721 noffset[n] = 5 + (dptrs_per_blk * 2);
722 offset[n++] = block / indirect_blks;
723 noffset[n] = 6 + (dptrs_per_blk * 2) +
724 offset[n - 1] * (dptrs_per_blk + 1);
725 offset[n++] = (block / direct_blks) % dptrs_per_blk;
726 noffset[n] = 7 + (dptrs_per_blk * 2) +
727 offset[n - 2] * (dptrs_per_blk + 1) +
729 offset[n] = block % direct_blks;
740 * Caller should call f2fs_put_dnode(dn).
741 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
742 * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
744 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
746 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
747 struct page *npage[4];
748 struct page *parent = NULL;
750 unsigned int noffset[4];
755 level = get_node_path(dn->inode, index, offset, noffset);
759 nids[0] = dn->inode->i_ino;
760 npage[0] = dn->inode_page;
763 npage[0] = f2fs_get_node_page(sbi, nids[0]);
764 if (IS_ERR(npage[0]))
765 return PTR_ERR(npage[0]);
768 /* if inline_data is set, should not report any block indices */
769 if (f2fs_has_inline_data(dn->inode) && index) {
771 f2fs_put_page(npage[0], 1);
777 nids[1] = get_nid(parent, offset[0], true);
778 dn->inode_page = npage[0];
779 dn->inode_page_locked = true;
781 /* get indirect or direct nodes */
782 for (i = 1; i <= level; i++) {
785 if (!nids[i] && mode == ALLOC_NODE) {
787 if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
793 npage[i] = f2fs_new_node_page(dn, noffset[i]);
794 if (IS_ERR(npage[i])) {
795 f2fs_alloc_nid_failed(sbi, nids[i]);
796 err = PTR_ERR(npage[i]);
800 set_nid(parent, offset[i - 1], nids[i], i == 1);
801 f2fs_alloc_nid_done(sbi, nids[i]);
803 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
804 npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
805 if (IS_ERR(npage[i])) {
806 err = PTR_ERR(npage[i]);
812 dn->inode_page_locked = false;
815 f2fs_put_page(parent, 1);
819 npage[i] = f2fs_get_node_page(sbi, nids[i]);
820 if (IS_ERR(npage[i])) {
821 err = PTR_ERR(npage[i]);
822 f2fs_put_page(npage[0], 0);
828 nids[i + 1] = get_nid(parent, offset[i], false);
831 dn->nid = nids[level];
832 dn->ofs_in_node = offset[level];
833 dn->node_page = npage[level];
834 dn->data_blkaddr = f2fs_data_blkaddr(dn);
838 f2fs_put_page(parent, 1);
840 f2fs_put_page(npage[0], 0);
842 dn->inode_page = NULL;
843 dn->node_page = NULL;
844 if (err == -ENOENT) {
846 dn->max_level = level;
847 dn->ofs_in_node = offset[level];
852 static int truncate_node(struct dnode_of_data *dn)
854 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
859 err = f2fs_get_node_info(sbi, dn->nid, &ni);
863 /* Deallocate node address */
864 f2fs_invalidate_blocks(sbi, ni.blk_addr);
865 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
866 set_node_addr(sbi, &ni, NULL_ADDR, false);
868 if (dn->nid == dn->inode->i_ino) {
869 f2fs_remove_orphan_inode(sbi, dn->nid);
870 dec_valid_inode_count(sbi);
871 f2fs_inode_synced(dn->inode);
874 clear_node_page_dirty(dn->node_page);
875 set_sbi_flag(sbi, SBI_IS_DIRTY);
877 index = dn->node_page->index;
878 f2fs_put_page(dn->node_page, 1);
880 invalidate_mapping_pages(NODE_MAPPING(sbi),
883 dn->node_page = NULL;
884 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
889 static int truncate_dnode(struct dnode_of_data *dn)
897 /* get direct node */
898 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
899 if (PTR_ERR(page) == -ENOENT)
901 else if (IS_ERR(page))
902 return PTR_ERR(page);
904 /* Make dnode_of_data for parameter */
905 dn->node_page = page;
907 f2fs_truncate_data_blocks(dn);
908 err = truncate_node(dn);
915 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
918 struct dnode_of_data rdn = *dn;
920 struct f2fs_node *rn;
922 unsigned int child_nofs;
927 return NIDS_PER_BLOCK + 1;
929 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
931 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
933 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
934 return PTR_ERR(page);
937 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
939 rn = F2FS_NODE(page);
941 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
942 child_nid = le32_to_cpu(rn->in.nid[i]);
946 ret = truncate_dnode(&rdn);
949 if (set_nid(page, i, 0, false))
950 dn->node_changed = true;
953 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
954 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
955 child_nid = le32_to_cpu(rn->in.nid[i]);
956 if (child_nid == 0) {
957 child_nofs += NIDS_PER_BLOCK + 1;
961 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
962 if (ret == (NIDS_PER_BLOCK + 1)) {
963 if (set_nid(page, i, 0, false))
964 dn->node_changed = true;
966 } else if (ret < 0 && ret != -ENOENT) {
974 /* remove current indirect node */
975 dn->node_page = page;
976 ret = truncate_node(dn);
981 f2fs_put_page(page, 1);
983 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
987 f2fs_put_page(page, 1);
988 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
992 static int truncate_partial_nodes(struct dnode_of_data *dn,
993 struct f2fs_inode *ri, int *offset, int depth)
995 struct page *pages[2];
1000 int idx = depth - 2;
1002 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1006 /* get indirect nodes in the path */
1007 for (i = 0; i < idx + 1; i++) {
1008 /* reference count'll be increased */
1009 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
1010 if (IS_ERR(pages[i])) {
1011 err = PTR_ERR(pages[i]);
1015 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1018 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1020 /* free direct nodes linked to a partial indirect node */
1021 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1022 child_nid = get_nid(pages[idx], i, false);
1025 dn->nid = child_nid;
1026 err = truncate_dnode(dn);
1029 if (set_nid(pages[idx], i, 0, false))
1030 dn->node_changed = true;
1033 if (offset[idx + 1] == 0) {
1034 dn->node_page = pages[idx];
1036 err = truncate_node(dn);
1040 f2fs_put_page(pages[idx], 1);
1043 offset[idx + 1] = 0;
1046 for (i = idx; i >= 0; i--)
1047 f2fs_put_page(pages[i], 1);
1049 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1055 * All the block addresses of data and nodes should be nullified.
1057 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1059 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1060 int err = 0, cont = 1;
1061 int level, offset[4], noffset[4];
1062 unsigned int nofs = 0;
1063 struct f2fs_inode *ri;
1064 struct dnode_of_data dn;
1067 trace_f2fs_truncate_inode_blocks_enter(inode, from);
1069 level = get_node_path(inode, from, offset, noffset);
1071 trace_f2fs_truncate_inode_blocks_exit(inode, level);
1075 page = f2fs_get_node_page(sbi, inode->i_ino);
1077 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1078 return PTR_ERR(page);
1081 set_new_dnode(&dn, inode, page, NULL, 0);
1084 ri = F2FS_INODE(page);
1092 if (!offset[level - 1])
1094 err = truncate_partial_nodes(&dn, ri, offset, level);
1095 if (err < 0 && err != -ENOENT)
1097 nofs += 1 + NIDS_PER_BLOCK;
1100 nofs = 5 + 2 * NIDS_PER_BLOCK;
1101 if (!offset[level - 1])
1103 err = truncate_partial_nodes(&dn, ri, offset, level);
1104 if (err < 0 && err != -ENOENT)
1113 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1114 switch (offset[0]) {
1115 case NODE_DIR1_BLOCK:
1116 case NODE_DIR2_BLOCK:
1117 err = truncate_dnode(&dn);
1120 case NODE_IND1_BLOCK:
1121 case NODE_IND2_BLOCK:
1122 err = truncate_nodes(&dn, nofs, offset[1], 2);
1125 case NODE_DIND_BLOCK:
1126 err = truncate_nodes(&dn, nofs, offset[1], 3);
1133 if (err < 0 && err != -ENOENT)
1135 if (offset[1] == 0 &&
1136 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1138 BUG_ON(page->mapping != NODE_MAPPING(sbi));
1139 f2fs_wait_on_page_writeback(page, NODE, true, true);
1140 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1141 set_page_dirty(page);
1149 f2fs_put_page(page, 0);
1150 trace_f2fs_truncate_inode_blocks_exit(inode, err);
1151 return err > 0 ? 0 : err;
1154 /* caller must lock inode page */
1155 int f2fs_truncate_xattr_node(struct inode *inode)
1157 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1158 nid_t nid = F2FS_I(inode)->i_xattr_nid;
1159 struct dnode_of_data dn;
1166 npage = f2fs_get_node_page(sbi, nid);
1168 return PTR_ERR(npage);
1170 set_new_dnode(&dn, inode, NULL, npage, nid);
1171 err = truncate_node(&dn);
1173 f2fs_put_page(npage, 1);
1177 f2fs_i_xnid_write(inode, 0);
1183 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1186 int f2fs_remove_inode_page(struct inode *inode)
1188 struct dnode_of_data dn;
1191 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1192 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1196 err = f2fs_truncate_xattr_node(inode);
1198 f2fs_put_dnode(&dn);
1202 /* remove potential inline_data blocks */
1203 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1204 S_ISLNK(inode->i_mode))
1205 f2fs_truncate_data_blocks_range(&dn, 1);
1207 /* 0 is possible, after f2fs_new_inode() has failed */
1208 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1209 f2fs_put_dnode(&dn);
1213 if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1214 f2fs_warn(F2FS_I_SB(inode),
1215 "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1216 inode->i_ino, (unsigned long long)inode->i_blocks);
1217 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1220 /* will put inode & node pages */
1221 err = truncate_node(&dn);
1223 f2fs_put_dnode(&dn);
1229 struct page *f2fs_new_inode_page(struct inode *inode)
1231 struct dnode_of_data dn;
1233 /* allocate inode page for new inode */
1234 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1236 /* caller should f2fs_put_page(page, 1); */
1237 return f2fs_new_node_page(&dn, 0);
1240 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1242 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1243 struct node_info new_ni;
1247 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1248 return ERR_PTR(-EPERM);
1250 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1252 return ERR_PTR(-ENOMEM);
1254 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1257 #ifdef CONFIG_F2FS_CHECK_FS
1258 err = f2fs_get_node_info(sbi, dn->nid, &new_ni);
1260 dec_valid_node_count(sbi, dn->inode, !ofs);
1263 f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1265 new_ni.nid = dn->nid;
1266 new_ni.ino = dn->inode->i_ino;
1267 new_ni.blk_addr = NULL_ADDR;
1270 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1272 f2fs_wait_on_page_writeback(page, NODE, true, true);
1273 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1274 set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1275 if (!PageUptodate(page))
1276 SetPageUptodate(page);
1277 if (set_page_dirty(page))
1278 dn->node_changed = true;
1280 if (f2fs_has_xattr_block(ofs))
1281 f2fs_i_xnid_write(dn->inode, dn->nid);
1284 inc_valid_inode_count(sbi);
1288 clear_node_page_dirty(page);
1289 f2fs_put_page(page, 1);
1290 return ERR_PTR(err);
1294 * Caller should do after getting the following values.
1295 * 0: f2fs_put_page(page, 0)
1296 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1298 static int read_node_page(struct page *page, int op_flags)
1300 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1301 struct node_info ni;
1302 struct f2fs_io_info fio = {
1306 .op_flags = op_flags,
1308 .encrypted_page = NULL,
1312 if (PageUptodate(page)) {
1313 if (!f2fs_inode_chksum_verify(sbi, page)) {
1314 ClearPageUptodate(page);
1320 err = f2fs_get_node_info(sbi, page->index, &ni);
1324 if (unlikely(ni.blk_addr == NULL_ADDR) ||
1325 is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1326 ClearPageUptodate(page);
1330 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1332 err = f2fs_submit_page_bio(&fio);
1335 f2fs_update_iostat(sbi, FS_NODE_READ_IO, F2FS_BLKSIZE);
1341 * Readahead a node page
1343 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1350 if (f2fs_check_nid_range(sbi, nid))
1353 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1357 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1361 err = read_node_page(apage, REQ_RAHEAD);
1362 f2fs_put_page(apage, err ? 1 : 0);
1365 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1366 struct page *parent, int start)
1372 return ERR_PTR(-ENOENT);
1373 if (f2fs_check_nid_range(sbi, nid))
1374 return ERR_PTR(-EINVAL);
1376 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1378 return ERR_PTR(-ENOMEM);
1380 err = read_node_page(page, 0);
1382 f2fs_put_page(page, 1);
1383 return ERR_PTR(err);
1384 } else if (err == LOCKED_PAGE) {
1390 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1394 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1395 f2fs_put_page(page, 1);
1399 if (unlikely(!PageUptodate(page))) {
1404 if (!f2fs_inode_chksum_verify(sbi, page)) {
1409 if (unlikely(nid != nid_of_node(page))) {
1410 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1411 nid, nid_of_node(page), ino_of_node(page),
1412 ofs_of_node(page), cpver_of_node(page),
1413 next_blkaddr_of_node(page));
1416 ClearPageUptodate(page);
1417 f2fs_put_page(page, 1);
1418 return ERR_PTR(err);
1423 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1425 return __get_node_page(sbi, nid, NULL, 0);
1428 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1430 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1431 nid_t nid = get_nid(parent, start, false);
1433 return __get_node_page(sbi, nid, parent, start);
1436 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1438 struct inode *inode;
1442 /* should flush inline_data before evict_inode */
1443 inode = ilookup(sbi->sb, ino);
1447 page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1448 FGP_LOCK|FGP_NOWAIT, 0);
1452 if (!PageUptodate(page))
1455 if (!PageDirty(page))
1458 if (!clear_page_dirty_for_io(page))
1461 ret = f2fs_write_inline_data(inode, page);
1462 inode_dec_dirty_pages(inode);
1463 f2fs_remove_dirty_inode(inode);
1465 set_page_dirty(page);
1467 f2fs_put_page(page, 1);
1472 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1475 struct pagevec pvec;
1476 struct page *last_page = NULL;
1479 pagevec_init(&pvec);
1482 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1483 PAGECACHE_TAG_DIRTY))) {
1486 for (i = 0; i < nr_pages; i++) {
1487 struct page *page = pvec.pages[i];
1489 if (unlikely(f2fs_cp_error(sbi))) {
1490 f2fs_put_page(last_page, 0);
1491 pagevec_release(&pvec);
1492 return ERR_PTR(-EIO);
1495 if (!IS_DNODE(page) || !is_cold_node(page))
1497 if (ino_of_node(page) != ino)
1502 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1507 if (ino_of_node(page) != ino)
1508 goto continue_unlock;
1510 if (!PageDirty(page)) {
1511 /* someone wrote it for us */
1512 goto continue_unlock;
1516 f2fs_put_page(last_page, 0);
1522 pagevec_release(&pvec);
1528 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1529 struct writeback_control *wbc, bool do_balance,
1530 enum iostat_type io_type, unsigned int *seq_id)
1532 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1534 struct node_info ni;
1535 struct f2fs_io_info fio = {
1537 .ino = ino_of_node(page),
1540 .op_flags = wbc_to_write_flags(wbc),
1542 .encrypted_page = NULL,
1549 trace_f2fs_writepage(page, NODE);
1551 if (unlikely(f2fs_cp_error(sbi))) {
1552 ClearPageUptodate(page);
1553 dec_page_count(sbi, F2FS_DIRTY_NODES);
1558 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1561 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1562 wbc->sync_mode == WB_SYNC_NONE &&
1563 IS_DNODE(page) && is_cold_node(page))
1566 /* get old block addr of this node page */
1567 nid = nid_of_node(page);
1568 f2fs_bug_on(sbi, page->index != nid);
1570 if (f2fs_get_node_info(sbi, nid, &ni))
1573 if (wbc->for_reclaim) {
1574 if (!down_read_trylock(&sbi->node_write))
1577 down_read(&sbi->node_write);
1580 /* This page is already truncated */
1581 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1582 ClearPageUptodate(page);
1583 dec_page_count(sbi, F2FS_DIRTY_NODES);
1584 up_read(&sbi->node_write);
1589 if (__is_valid_data_blkaddr(ni.blk_addr) &&
1590 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1591 DATA_GENERIC_ENHANCE)) {
1592 up_read(&sbi->node_write);
1596 if (atomic && !test_opt(sbi, NOBARRIER))
1597 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1599 /* should add to global list before clearing PAGECACHE status */
1600 if (f2fs_in_warm_node_list(sbi, page)) {
1601 seq = f2fs_add_fsync_node_entry(sbi, page);
1606 set_page_writeback(page);
1607 ClearPageError(page);
1609 fio.old_blkaddr = ni.blk_addr;
1610 f2fs_do_write_node_page(nid, &fio);
1611 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1612 dec_page_count(sbi, F2FS_DIRTY_NODES);
1613 up_read(&sbi->node_write);
1615 if (wbc->for_reclaim) {
1616 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1622 if (unlikely(f2fs_cp_error(sbi))) {
1623 f2fs_submit_merged_write(sbi, NODE);
1627 *submitted = fio.submitted;
1630 f2fs_balance_fs(sbi, false);
1634 redirty_page_for_writepage(wbc, page);
1635 return AOP_WRITEPAGE_ACTIVATE;
1638 int f2fs_move_node_page(struct page *node_page, int gc_type)
1642 if (gc_type == FG_GC) {
1643 struct writeback_control wbc = {
1644 .sync_mode = WB_SYNC_ALL,
1649 f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1651 set_page_dirty(node_page);
1653 if (!clear_page_dirty_for_io(node_page)) {
1658 if (__write_node_page(node_page, false, NULL,
1659 &wbc, false, FS_GC_NODE_IO, NULL)) {
1661 unlock_page(node_page);
1665 /* set page dirty and write it */
1666 if (!PageWriteback(node_page))
1667 set_page_dirty(node_page);
1670 unlock_page(node_page);
1672 f2fs_put_page(node_page, 0);
1676 static int f2fs_write_node_page(struct page *page,
1677 struct writeback_control *wbc)
1679 return __write_node_page(page, false, NULL, wbc, false,
1683 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1684 struct writeback_control *wbc, bool atomic,
1685 unsigned int *seq_id)
1688 struct pagevec pvec;
1690 struct page *last_page = NULL;
1691 bool marked = false;
1692 nid_t ino = inode->i_ino;
1697 last_page = last_fsync_dnode(sbi, ino);
1698 if (IS_ERR_OR_NULL(last_page))
1699 return PTR_ERR_OR_ZERO(last_page);
1702 pagevec_init(&pvec);
1705 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1706 PAGECACHE_TAG_DIRTY))) {
1709 for (i = 0; i < nr_pages; i++) {
1710 struct page *page = pvec.pages[i];
1711 bool submitted = false;
1713 if (unlikely(f2fs_cp_error(sbi))) {
1714 f2fs_put_page(last_page, 0);
1715 pagevec_release(&pvec);
1720 if (!IS_DNODE(page) || !is_cold_node(page))
1722 if (ino_of_node(page) != ino)
1727 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1732 if (ino_of_node(page) != ino)
1733 goto continue_unlock;
1735 if (!PageDirty(page) && page != last_page) {
1736 /* someone wrote it for us */
1737 goto continue_unlock;
1740 f2fs_wait_on_page_writeback(page, NODE, true, true);
1742 set_fsync_mark(page, 0);
1743 set_dentry_mark(page, 0);
1745 if (!atomic || page == last_page) {
1746 set_fsync_mark(page, 1);
1747 if (IS_INODE(page)) {
1748 if (is_inode_flag_set(inode,
1750 f2fs_update_inode(inode, page);
1751 set_dentry_mark(page,
1752 f2fs_need_dentry_mark(sbi, ino));
1754 /* may be written by other thread */
1755 if (!PageDirty(page))
1756 set_page_dirty(page);
1759 if (!clear_page_dirty_for_io(page))
1760 goto continue_unlock;
1762 ret = __write_node_page(page, atomic &&
1764 &submitted, wbc, true,
1765 FS_NODE_IO, seq_id);
1768 f2fs_put_page(last_page, 0);
1770 } else if (submitted) {
1774 if (page == last_page) {
1775 f2fs_put_page(page, 0);
1780 pagevec_release(&pvec);
1786 if (!ret && atomic && !marked) {
1787 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1788 ino, last_page->index);
1789 lock_page(last_page);
1790 f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1791 set_page_dirty(last_page);
1792 unlock_page(last_page);
1797 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1798 return ret ? -EIO : 0;
1801 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1803 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1806 if (inode->i_ino != ino)
1809 if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1812 spin_lock(&sbi->inode_lock[DIRTY_META]);
1813 clean = list_empty(&F2FS_I(inode)->gdirty_list);
1814 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1819 inode = igrab(inode);
1825 static bool flush_dirty_inode(struct page *page)
1827 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1828 struct inode *inode;
1829 nid_t ino = ino_of_node(page);
1831 inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1835 f2fs_update_inode(inode, page);
1842 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1845 struct pagevec pvec;
1848 pagevec_init(&pvec);
1850 while ((nr_pages = pagevec_lookup_tag(&pvec,
1851 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1854 for (i = 0; i < nr_pages; i++) {
1855 struct page *page = pvec.pages[i];
1857 if (!IS_DNODE(page))
1862 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1868 if (!PageDirty(page)) {
1869 /* someone wrote it for us */
1870 goto continue_unlock;
1873 /* flush inline_data, if it's async context. */
1874 if (page_private_inline(page)) {
1875 clear_page_private_inline(page);
1877 flush_inline_data(sbi, ino_of_node(page));
1882 pagevec_release(&pvec);
1887 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1888 struct writeback_control *wbc,
1889 bool do_balance, enum iostat_type io_type)
1892 struct pagevec pvec;
1896 int nr_pages, done = 0;
1898 pagevec_init(&pvec);
1903 while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1904 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1907 for (i = 0; i < nr_pages; i++) {
1908 struct page *page = pvec.pages[i];
1909 bool submitted = false;
1910 bool may_dirty = true;
1912 /* give a priority to WB_SYNC threads */
1913 if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1914 wbc->sync_mode == WB_SYNC_NONE) {
1920 * flushing sequence with step:
1925 if (step == 0 && IS_DNODE(page))
1927 if (step == 1 && (!IS_DNODE(page) ||
1928 is_cold_node(page)))
1930 if (step == 2 && (!IS_DNODE(page) ||
1931 !is_cold_node(page)))
1934 if (wbc->sync_mode == WB_SYNC_ALL)
1936 else if (!trylock_page(page))
1939 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1945 if (!PageDirty(page)) {
1946 /* someone wrote it for us */
1947 goto continue_unlock;
1950 /* flush inline_data/inode, if it's async context. */
1954 /* flush inline_data */
1955 if (page_private_inline(page)) {
1956 clear_page_private_inline(page);
1958 flush_inline_data(sbi, ino_of_node(page));
1962 /* flush dirty inode */
1963 if (IS_INODE(page) && may_dirty) {
1965 if (flush_dirty_inode(page))
1969 f2fs_wait_on_page_writeback(page, NODE, true, true);
1971 if (!clear_page_dirty_for_io(page))
1972 goto continue_unlock;
1974 set_fsync_mark(page, 0);
1975 set_dentry_mark(page, 0);
1977 ret = __write_node_page(page, false, &submitted,
1978 wbc, do_balance, io_type, NULL);
1984 if (--wbc->nr_to_write == 0)
1987 pagevec_release(&pvec);
1990 if (wbc->nr_to_write == 0) {
1997 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1998 wbc->sync_mode == WB_SYNC_NONE && step == 1)
2005 f2fs_submit_merged_write(sbi, NODE);
2007 if (unlikely(f2fs_cp_error(sbi)))
2012 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2013 unsigned int seq_id)
2015 struct fsync_node_entry *fn;
2017 struct list_head *head = &sbi->fsync_node_list;
2018 unsigned long flags;
2019 unsigned int cur_seq_id = 0;
2022 while (seq_id && cur_seq_id < seq_id) {
2023 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2024 if (list_empty(head)) {
2025 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2028 fn = list_first_entry(head, struct fsync_node_entry, list);
2029 if (fn->seq_id > seq_id) {
2030 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2033 cur_seq_id = fn->seq_id;
2036 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2038 f2fs_wait_on_page_writeback(page, NODE, true, false);
2039 if (TestClearPageError(page))
2048 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2055 static int f2fs_write_node_pages(struct address_space *mapping,
2056 struct writeback_control *wbc)
2058 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2059 struct blk_plug plug;
2062 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2065 /* balancing f2fs's metadata in background */
2066 f2fs_balance_fs_bg(sbi, true);
2068 /* collect a number of dirty node pages and write together */
2069 if (wbc->sync_mode != WB_SYNC_ALL &&
2070 get_pages(sbi, F2FS_DIRTY_NODES) <
2071 nr_pages_to_skip(sbi, NODE))
2074 if (wbc->sync_mode == WB_SYNC_ALL)
2075 atomic_inc(&sbi->wb_sync_req[NODE]);
2076 else if (atomic_read(&sbi->wb_sync_req[NODE]))
2079 trace_f2fs_writepages(mapping->host, wbc, NODE);
2081 diff = nr_pages_to_write(sbi, NODE, wbc);
2082 blk_start_plug(&plug);
2083 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2084 blk_finish_plug(&plug);
2085 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2087 if (wbc->sync_mode == WB_SYNC_ALL)
2088 atomic_dec(&sbi->wb_sync_req[NODE]);
2092 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2093 trace_f2fs_writepages(mapping->host, wbc, NODE);
2097 static int f2fs_set_node_page_dirty(struct page *page)
2099 trace_f2fs_set_page_dirty(page, NODE);
2101 if (!PageUptodate(page))
2102 SetPageUptodate(page);
2103 #ifdef CONFIG_F2FS_CHECK_FS
2105 f2fs_inode_chksum_set(F2FS_P_SB(page), page);
2107 if (!PageDirty(page)) {
2108 __set_page_dirty_nobuffers(page);
2109 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
2110 set_page_private_reference(page);
2117 * Structure of the f2fs node operations
2119 const struct address_space_operations f2fs_node_aops = {
2120 .writepage = f2fs_write_node_page,
2121 .writepages = f2fs_write_node_pages,
2122 .set_page_dirty = f2fs_set_node_page_dirty,
2123 .invalidatepage = f2fs_invalidate_page,
2124 .releasepage = f2fs_release_page,
2125 #ifdef CONFIG_MIGRATION
2126 .migratepage = f2fs_migrate_page,
2130 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2133 return radix_tree_lookup(&nm_i->free_nid_root, n);
2136 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2139 struct f2fs_nm_info *nm_i = NM_I(sbi);
2140 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2145 nm_i->nid_cnt[FREE_NID]++;
2146 list_add_tail(&i->list, &nm_i->free_nid_list);
2150 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2151 struct free_nid *i, enum nid_state state)
2153 struct f2fs_nm_info *nm_i = NM_I(sbi);
2155 f2fs_bug_on(sbi, state != i->state);
2156 nm_i->nid_cnt[state]--;
2157 if (state == FREE_NID)
2159 radix_tree_delete(&nm_i->free_nid_root, i->nid);
2162 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2163 enum nid_state org_state, enum nid_state dst_state)
2165 struct f2fs_nm_info *nm_i = NM_I(sbi);
2167 f2fs_bug_on(sbi, org_state != i->state);
2168 i->state = dst_state;
2169 nm_i->nid_cnt[org_state]--;
2170 nm_i->nid_cnt[dst_state]++;
2172 switch (dst_state) {
2177 list_add_tail(&i->list, &nm_i->free_nid_list);
2184 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2185 bool set, bool build)
2187 struct f2fs_nm_info *nm_i = NM_I(sbi);
2188 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2189 unsigned int nid_ofs = nid - START_NID(nid);
2191 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2195 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2197 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2198 nm_i->free_nid_count[nat_ofs]++;
2200 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2202 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2204 nm_i->free_nid_count[nat_ofs]--;
2208 /* return if the nid is recognized as free */
2209 static bool add_free_nid(struct f2fs_sb_info *sbi,
2210 nid_t nid, bool build, bool update)
2212 struct f2fs_nm_info *nm_i = NM_I(sbi);
2213 struct free_nid *i, *e;
2214 struct nat_entry *ne;
2218 /* 0 nid should not be used */
2219 if (unlikely(nid == 0))
2222 if (unlikely(f2fs_check_nid_range(sbi, nid)))
2225 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2227 i->state = FREE_NID;
2229 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2231 spin_lock(&nm_i->nid_list_lock);
2239 * - __insert_nid_to_list(PREALLOC_NID)
2240 * - f2fs_balance_fs_bg
2241 * - f2fs_build_free_nids
2242 * - __f2fs_build_free_nids
2245 * - __lookup_nat_cache
2247 * - f2fs_init_inode_metadata
2248 * - f2fs_new_inode_page
2249 * - f2fs_new_node_page
2251 * - f2fs_alloc_nid_done
2252 * - __remove_nid_from_list(PREALLOC_NID)
2253 * - __insert_nid_to_list(FREE_NID)
2255 ne = __lookup_nat_cache(nm_i, nid);
2256 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2257 nat_get_blkaddr(ne) != NULL_ADDR))
2260 e = __lookup_free_nid_list(nm_i, nid);
2262 if (e->state == FREE_NID)
2268 err = __insert_free_nid(sbi, i);
2271 update_free_nid_bitmap(sbi, nid, ret, build);
2273 nm_i->available_nids++;
2275 spin_unlock(&nm_i->nid_list_lock);
2276 radix_tree_preload_end();
2279 kmem_cache_free(free_nid_slab, i);
2283 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2285 struct f2fs_nm_info *nm_i = NM_I(sbi);
2287 bool need_free = false;
2289 spin_lock(&nm_i->nid_list_lock);
2290 i = __lookup_free_nid_list(nm_i, nid);
2291 if (i && i->state == FREE_NID) {
2292 __remove_free_nid(sbi, i, FREE_NID);
2295 spin_unlock(&nm_i->nid_list_lock);
2298 kmem_cache_free(free_nid_slab, i);
2301 static int scan_nat_page(struct f2fs_sb_info *sbi,
2302 struct page *nat_page, nid_t start_nid)
2304 struct f2fs_nm_info *nm_i = NM_I(sbi);
2305 struct f2fs_nat_block *nat_blk = page_address(nat_page);
2307 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2310 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2312 i = start_nid % NAT_ENTRY_PER_BLOCK;
2314 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2315 if (unlikely(start_nid >= nm_i->max_nid))
2318 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2320 if (blk_addr == NEW_ADDR)
2323 if (blk_addr == NULL_ADDR) {
2324 add_free_nid(sbi, start_nid, true, true);
2326 spin_lock(&NM_I(sbi)->nid_list_lock);
2327 update_free_nid_bitmap(sbi, start_nid, false, true);
2328 spin_unlock(&NM_I(sbi)->nid_list_lock);
2335 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2337 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2338 struct f2fs_journal *journal = curseg->journal;
2341 down_read(&curseg->journal_rwsem);
2342 for (i = 0; i < nats_in_cursum(journal); i++) {
2346 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2347 nid = le32_to_cpu(nid_in_journal(journal, i));
2348 if (addr == NULL_ADDR)
2349 add_free_nid(sbi, nid, true, false);
2351 remove_free_nid(sbi, nid);
2353 up_read(&curseg->journal_rwsem);
2356 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2358 struct f2fs_nm_info *nm_i = NM_I(sbi);
2359 unsigned int i, idx;
2362 down_read(&nm_i->nat_tree_lock);
2364 for (i = 0; i < nm_i->nat_blocks; i++) {
2365 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2367 if (!nm_i->free_nid_count[i])
2369 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2370 idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2371 NAT_ENTRY_PER_BLOCK, idx);
2372 if (idx >= NAT_ENTRY_PER_BLOCK)
2375 nid = i * NAT_ENTRY_PER_BLOCK + idx;
2376 add_free_nid(sbi, nid, true, false);
2378 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2383 scan_curseg_cache(sbi);
2385 up_read(&nm_i->nat_tree_lock);
2388 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2389 bool sync, bool mount)
2391 struct f2fs_nm_info *nm_i = NM_I(sbi);
2393 nid_t nid = nm_i->next_scan_nid;
2395 if (unlikely(nid >= nm_i->max_nid))
2398 if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2399 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2401 /* Enough entries */
2402 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2405 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2409 /* try to find free nids in free_nid_bitmap */
2410 scan_free_nid_bits(sbi);
2412 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2416 /* readahead nat pages to be scanned */
2417 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2420 down_read(&nm_i->nat_tree_lock);
2423 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2424 nm_i->nat_block_bitmap)) {
2425 struct page *page = get_current_nat_page(sbi, nid);
2428 ret = PTR_ERR(page);
2430 ret = scan_nat_page(sbi, page, nid);
2431 f2fs_put_page(page, 1);
2435 up_read(&nm_i->nat_tree_lock);
2436 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2441 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2442 if (unlikely(nid >= nm_i->max_nid))
2445 if (++i >= FREE_NID_PAGES)
2449 /* go to the next free nat pages to find free nids abundantly */
2450 nm_i->next_scan_nid = nid;
2452 /* find free nids from current sum_pages */
2453 scan_curseg_cache(sbi);
2455 up_read(&nm_i->nat_tree_lock);
2457 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2458 nm_i->ra_nid_pages, META_NAT, false);
2463 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2467 mutex_lock(&NM_I(sbi)->build_lock);
2468 ret = __f2fs_build_free_nids(sbi, sync, mount);
2469 mutex_unlock(&NM_I(sbi)->build_lock);
2475 * If this function returns success, caller can obtain a new nid
2476 * from second parameter of this function.
2477 * The returned nid could be used ino as well as nid when inode is created.
2479 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2481 struct f2fs_nm_info *nm_i = NM_I(sbi);
2482 struct free_nid *i = NULL;
2484 if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2485 f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2489 spin_lock(&nm_i->nid_list_lock);
2491 if (unlikely(nm_i->available_nids == 0)) {
2492 spin_unlock(&nm_i->nid_list_lock);
2496 /* We should not use stale free nids created by f2fs_build_free_nids */
2497 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2498 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2499 i = list_first_entry(&nm_i->free_nid_list,
2500 struct free_nid, list);
2503 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2504 nm_i->available_nids--;
2506 update_free_nid_bitmap(sbi, *nid, false, false);
2508 spin_unlock(&nm_i->nid_list_lock);
2511 spin_unlock(&nm_i->nid_list_lock);
2513 /* Let's scan nat pages and its caches to get free nids */
2514 if (!f2fs_build_free_nids(sbi, true, false))
2520 * f2fs_alloc_nid() should be called prior to this function.
2522 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2524 struct f2fs_nm_info *nm_i = NM_I(sbi);
2527 spin_lock(&nm_i->nid_list_lock);
2528 i = __lookup_free_nid_list(nm_i, nid);
2529 f2fs_bug_on(sbi, !i);
2530 __remove_free_nid(sbi, i, PREALLOC_NID);
2531 spin_unlock(&nm_i->nid_list_lock);
2533 kmem_cache_free(free_nid_slab, i);
2537 * f2fs_alloc_nid() should be called prior to this function.
2539 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2541 struct f2fs_nm_info *nm_i = NM_I(sbi);
2543 bool need_free = false;
2548 spin_lock(&nm_i->nid_list_lock);
2549 i = __lookup_free_nid_list(nm_i, nid);
2550 f2fs_bug_on(sbi, !i);
2552 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2553 __remove_free_nid(sbi, i, PREALLOC_NID);
2556 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2559 nm_i->available_nids++;
2561 update_free_nid_bitmap(sbi, nid, true, false);
2563 spin_unlock(&nm_i->nid_list_lock);
2566 kmem_cache_free(free_nid_slab, i);
2569 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2571 struct f2fs_nm_info *nm_i = NM_I(sbi);
2574 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2577 if (!mutex_trylock(&nm_i->build_lock))
2580 while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2581 struct free_nid *i, *next;
2582 unsigned int batch = SHRINK_NID_BATCH_SIZE;
2584 spin_lock(&nm_i->nid_list_lock);
2585 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2586 if (!nr_shrink || !batch ||
2587 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2589 __remove_free_nid(sbi, i, FREE_NID);
2590 kmem_cache_free(free_nid_slab, i);
2594 spin_unlock(&nm_i->nid_list_lock);
2597 mutex_unlock(&nm_i->build_lock);
2599 return nr - nr_shrink;
2602 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2604 void *src_addr, *dst_addr;
2607 struct f2fs_inode *ri;
2609 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2611 return PTR_ERR(ipage);
2613 ri = F2FS_INODE(page);
2614 if (ri->i_inline & F2FS_INLINE_XATTR) {
2615 if (!f2fs_has_inline_xattr(inode)) {
2616 set_inode_flag(inode, FI_INLINE_XATTR);
2617 stat_inc_inline_xattr(inode);
2620 if (f2fs_has_inline_xattr(inode)) {
2621 stat_dec_inline_xattr(inode);
2622 clear_inode_flag(inode, FI_INLINE_XATTR);
2627 dst_addr = inline_xattr_addr(inode, ipage);
2628 src_addr = inline_xattr_addr(inode, page);
2629 inline_size = inline_xattr_size(inode);
2631 f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2632 memcpy(dst_addr, src_addr, inline_size);
2634 f2fs_update_inode(inode, ipage);
2635 f2fs_put_page(ipage, 1);
2639 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2641 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2642 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2644 struct dnode_of_data dn;
2645 struct node_info ni;
2652 /* 1: invalidate the previous xattr nid */
2653 err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2657 f2fs_invalidate_blocks(sbi, ni.blk_addr);
2658 dec_valid_node_count(sbi, inode, false);
2659 set_node_addr(sbi, &ni, NULL_ADDR, false);
2662 /* 2: update xattr nid in inode */
2663 if (!f2fs_alloc_nid(sbi, &new_xnid))
2666 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2667 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2668 if (IS_ERR(xpage)) {
2669 f2fs_alloc_nid_failed(sbi, new_xnid);
2670 return PTR_ERR(xpage);
2673 f2fs_alloc_nid_done(sbi, new_xnid);
2674 f2fs_update_inode_page(inode);
2676 /* 3: update and set xattr node page dirty */
2677 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2679 set_page_dirty(xpage);
2680 f2fs_put_page(xpage, 1);
2685 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2687 struct f2fs_inode *src, *dst;
2688 nid_t ino = ino_of_node(page);
2689 struct node_info old_ni, new_ni;
2693 err = f2fs_get_node_info(sbi, ino, &old_ni);
2697 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2700 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2702 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2706 /* Should not use this inode from free nid list */
2707 remove_free_nid(sbi, ino);
2709 if (!PageUptodate(ipage))
2710 SetPageUptodate(ipage);
2711 fill_node_footer(ipage, ino, ino, 0, true);
2712 set_cold_node(ipage, false);
2714 src = F2FS_INODE(page);
2715 dst = F2FS_INODE(ipage);
2717 memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2719 dst->i_blocks = cpu_to_le64(1);
2720 dst->i_links = cpu_to_le32(1);
2721 dst->i_xattr_nid = 0;
2722 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2723 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2724 dst->i_extra_isize = src->i_extra_isize;
2726 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2727 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2728 i_inline_xattr_size))
2729 dst->i_inline_xattr_size = src->i_inline_xattr_size;
2731 if (f2fs_sb_has_project_quota(sbi) &&
2732 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2734 dst->i_projid = src->i_projid;
2736 if (f2fs_sb_has_inode_crtime(sbi) &&
2737 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2739 dst->i_crtime = src->i_crtime;
2740 dst->i_crtime_nsec = src->i_crtime_nsec;
2747 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2749 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2750 inc_valid_inode_count(sbi);
2751 set_page_dirty(ipage);
2752 f2fs_put_page(ipage, 1);
2756 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2757 unsigned int segno, struct f2fs_summary_block *sum)
2759 struct f2fs_node *rn;
2760 struct f2fs_summary *sum_entry;
2762 int i, idx, last_offset, nrpages;
2764 /* scan the node segment */
2765 last_offset = sbi->blocks_per_seg;
2766 addr = START_BLOCK(sbi, segno);
2767 sum_entry = &sum->entries[0];
2769 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2770 nrpages = bio_max_segs(last_offset - i);
2772 /* readahead node pages */
2773 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2775 for (idx = addr; idx < addr + nrpages; idx++) {
2776 struct page *page = f2fs_get_tmp_page(sbi, idx);
2779 return PTR_ERR(page);
2781 rn = F2FS_NODE(page);
2782 sum_entry->nid = rn->footer.nid;
2783 sum_entry->version = 0;
2784 sum_entry->ofs_in_node = 0;
2786 f2fs_put_page(page, 1);
2789 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2795 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2797 struct f2fs_nm_info *nm_i = NM_I(sbi);
2798 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2799 struct f2fs_journal *journal = curseg->journal;
2802 down_write(&curseg->journal_rwsem);
2803 for (i = 0; i < nats_in_cursum(journal); i++) {
2804 struct nat_entry *ne;
2805 struct f2fs_nat_entry raw_ne;
2806 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2808 if (f2fs_check_nid_range(sbi, nid))
2811 raw_ne = nat_in_journal(journal, i);
2813 ne = __lookup_nat_cache(nm_i, nid);
2815 ne = __alloc_nat_entry(nid, true);
2816 __init_nat_entry(nm_i, ne, &raw_ne, true);
2820 * if a free nat in journal has not been used after last
2821 * checkpoint, we should remove it from available nids,
2822 * since later we will add it again.
2824 if (!get_nat_flag(ne, IS_DIRTY) &&
2825 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2826 spin_lock(&nm_i->nid_list_lock);
2827 nm_i->available_nids--;
2828 spin_unlock(&nm_i->nid_list_lock);
2831 __set_nat_cache_dirty(nm_i, ne);
2833 update_nats_in_cursum(journal, -i);
2834 up_write(&curseg->journal_rwsem);
2837 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2838 struct list_head *head, int max)
2840 struct nat_entry_set *cur;
2842 if (nes->entry_cnt >= max)
2845 list_for_each_entry(cur, head, set_list) {
2846 if (cur->entry_cnt >= nes->entry_cnt) {
2847 list_add(&nes->set_list, cur->set_list.prev);
2852 list_add_tail(&nes->set_list, head);
2855 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2858 struct f2fs_nm_info *nm_i = NM_I(sbi);
2859 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2860 struct f2fs_nat_block *nat_blk = page_address(page);
2864 if (!enabled_nat_bits(sbi, NULL))
2867 if (nat_index == 0) {
2871 for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2872 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2876 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2877 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2881 __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2882 if (valid == NAT_ENTRY_PER_BLOCK)
2883 __set_bit_le(nat_index, nm_i->full_nat_bits);
2885 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2888 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2889 struct nat_entry_set *set, struct cp_control *cpc)
2891 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2892 struct f2fs_journal *journal = curseg->journal;
2893 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2894 bool to_journal = true;
2895 struct f2fs_nat_block *nat_blk;
2896 struct nat_entry *ne, *cur;
2897 struct page *page = NULL;
2900 * there are two steps to flush nat entries:
2901 * #1, flush nat entries to journal in current hot data summary block.
2902 * #2, flush nat entries to nat page.
2904 if (enabled_nat_bits(sbi, cpc) ||
2905 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2909 down_write(&curseg->journal_rwsem);
2911 page = get_next_nat_page(sbi, start_nid);
2913 return PTR_ERR(page);
2915 nat_blk = page_address(page);
2916 f2fs_bug_on(sbi, !nat_blk);
2919 /* flush dirty nats in nat entry set */
2920 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2921 struct f2fs_nat_entry *raw_ne;
2922 nid_t nid = nat_get_nid(ne);
2925 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2928 offset = f2fs_lookup_journal_in_cursum(journal,
2929 NAT_JOURNAL, nid, 1);
2930 f2fs_bug_on(sbi, offset < 0);
2931 raw_ne = &nat_in_journal(journal, offset);
2932 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2934 raw_ne = &nat_blk->entries[nid - start_nid];
2936 raw_nat_from_node_info(raw_ne, &ne->ni);
2938 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2939 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2940 add_free_nid(sbi, nid, false, true);
2942 spin_lock(&NM_I(sbi)->nid_list_lock);
2943 update_free_nid_bitmap(sbi, nid, false, false);
2944 spin_unlock(&NM_I(sbi)->nid_list_lock);
2949 up_write(&curseg->journal_rwsem);
2951 __update_nat_bits(sbi, start_nid, page);
2952 f2fs_put_page(page, 1);
2955 /* Allow dirty nats by node block allocation in write_begin */
2956 if (!set->entry_cnt) {
2957 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2958 kmem_cache_free(nat_entry_set_slab, set);
2964 * This function is called during the checkpointing process.
2966 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2968 struct f2fs_nm_info *nm_i = NM_I(sbi);
2969 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2970 struct f2fs_journal *journal = curseg->journal;
2971 struct nat_entry_set *setvec[SETVEC_SIZE];
2972 struct nat_entry_set *set, *tmp;
2979 * during unmount, let's flush nat_bits before checking
2980 * nat_cnt[DIRTY_NAT].
2982 if (enabled_nat_bits(sbi, cpc)) {
2983 down_write(&nm_i->nat_tree_lock);
2984 remove_nats_in_journal(sbi);
2985 up_write(&nm_i->nat_tree_lock);
2988 if (!nm_i->nat_cnt[DIRTY_NAT])
2991 down_write(&nm_i->nat_tree_lock);
2994 * if there are no enough space in journal to store dirty nat
2995 * entries, remove all entries from journal and merge them
2996 * into nat entry set.
2998 if (enabled_nat_bits(sbi, cpc) ||
2999 !__has_cursum_space(journal,
3000 nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3001 remove_nats_in_journal(sbi);
3003 while ((found = __gang_lookup_nat_set(nm_i,
3004 set_idx, SETVEC_SIZE, setvec))) {
3007 set_idx = setvec[found - 1]->set + 1;
3008 for (idx = 0; idx < found; idx++)
3009 __adjust_nat_entry_set(setvec[idx], &sets,
3010 MAX_NAT_JENTRIES(journal));
3013 /* flush dirty nats in nat entry set */
3014 list_for_each_entry_safe(set, tmp, &sets, set_list) {
3015 err = __flush_nat_entry_set(sbi, set, cpc);
3020 up_write(&nm_i->nat_tree_lock);
3021 /* Allow dirty nats by node block allocation in write_begin */
3026 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3028 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3029 struct f2fs_nm_info *nm_i = NM_I(sbi);
3030 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3032 __u64 cp_ver = cur_cp_version(ckpt);
3033 block_t nat_bits_addr;
3035 if (!enabled_nat_bits(sbi, NULL))
3038 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3039 nm_i->nat_bits = f2fs_kvzalloc(sbi,
3040 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3041 if (!nm_i->nat_bits)
3044 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3045 nm_i->nat_bits_blocks;
3046 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3049 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3051 return PTR_ERR(page);
3053 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3054 page_address(page), F2FS_BLKSIZE);
3055 f2fs_put_page(page, 1);
3058 cp_ver |= (cur_cp_crc(ckpt) << 32);
3059 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3060 disable_nat_bits(sbi, true);
3064 nm_i->full_nat_bits = nm_i->nat_bits + 8;
3065 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3067 f2fs_notice(sbi, "Found nat_bits in checkpoint");
3071 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3073 struct f2fs_nm_info *nm_i = NM_I(sbi);
3075 nid_t nid, last_nid;
3077 if (!enabled_nat_bits(sbi, NULL))
3080 for (i = 0; i < nm_i->nat_blocks; i++) {
3081 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3082 if (i >= nm_i->nat_blocks)
3085 __set_bit_le(i, nm_i->nat_block_bitmap);
3087 nid = i * NAT_ENTRY_PER_BLOCK;
3088 last_nid = nid + NAT_ENTRY_PER_BLOCK;
3090 spin_lock(&NM_I(sbi)->nid_list_lock);
3091 for (; nid < last_nid; nid++)
3092 update_free_nid_bitmap(sbi, nid, true, true);
3093 spin_unlock(&NM_I(sbi)->nid_list_lock);
3096 for (i = 0; i < nm_i->nat_blocks; i++) {
3097 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3098 if (i >= nm_i->nat_blocks)
3101 __set_bit_le(i, nm_i->nat_block_bitmap);
3105 static int init_node_manager(struct f2fs_sb_info *sbi)
3107 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3108 struct f2fs_nm_info *nm_i = NM_I(sbi);
3109 unsigned char *version_bitmap;
3110 unsigned int nat_segs;
3113 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3115 /* segment_count_nat includes pair segment so divide to 2. */
3116 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3117 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3118 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3120 /* not used nids: 0, node, meta, (and root counted as valid node) */
3121 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3122 F2FS_RESERVED_NODE_NUM;
3123 nm_i->nid_cnt[FREE_NID] = 0;
3124 nm_i->nid_cnt[PREALLOC_NID] = 0;
3125 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3126 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3127 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3129 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3130 INIT_LIST_HEAD(&nm_i->free_nid_list);
3131 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3132 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3133 INIT_LIST_HEAD(&nm_i->nat_entries);
3134 spin_lock_init(&nm_i->nat_list_lock);
3136 mutex_init(&nm_i->build_lock);
3137 spin_lock_init(&nm_i->nid_list_lock);
3138 init_rwsem(&nm_i->nat_tree_lock);
3140 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3141 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3142 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3143 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3145 if (!nm_i->nat_bitmap)
3148 err = __get_nat_bitmaps(sbi);
3152 #ifdef CONFIG_F2FS_CHECK_FS
3153 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3155 if (!nm_i->nat_bitmap_mir)
3162 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3164 struct f2fs_nm_info *nm_i = NM_I(sbi);
3167 nm_i->free_nid_bitmap =
3168 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3171 if (!nm_i->free_nid_bitmap)
3174 for (i = 0; i < nm_i->nat_blocks; i++) {
3175 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3176 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3177 if (!nm_i->free_nid_bitmap[i])
3181 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3183 if (!nm_i->nat_block_bitmap)
3186 nm_i->free_nid_count =
3187 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3190 if (!nm_i->free_nid_count)
3195 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3199 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3204 err = init_node_manager(sbi);
3208 err = init_free_nid_cache(sbi);
3212 /* load free nid status from nat_bits table */
3213 load_free_nid_bitmap(sbi);
3215 return f2fs_build_free_nids(sbi, true, true);
3218 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3220 struct f2fs_nm_info *nm_i = NM_I(sbi);
3221 struct free_nid *i, *next_i;
3222 struct nat_entry *natvec[NATVEC_SIZE];
3223 struct nat_entry_set *setvec[SETVEC_SIZE];
3230 /* destroy free nid list */
3231 spin_lock(&nm_i->nid_list_lock);
3232 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3233 __remove_free_nid(sbi, i, FREE_NID);
3234 spin_unlock(&nm_i->nid_list_lock);
3235 kmem_cache_free(free_nid_slab, i);
3236 spin_lock(&nm_i->nid_list_lock);
3238 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3239 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3240 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3241 spin_unlock(&nm_i->nid_list_lock);
3243 /* destroy nat cache */
3244 down_write(&nm_i->nat_tree_lock);
3245 while ((found = __gang_lookup_nat_cache(nm_i,
3246 nid, NATVEC_SIZE, natvec))) {
3249 nid = nat_get_nid(natvec[found - 1]) + 1;
3250 for (idx = 0; idx < found; idx++) {
3251 spin_lock(&nm_i->nat_list_lock);
3252 list_del(&natvec[idx]->list);
3253 spin_unlock(&nm_i->nat_list_lock);
3255 __del_from_nat_cache(nm_i, natvec[idx]);
3258 f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3260 /* destroy nat set cache */
3262 while ((found = __gang_lookup_nat_set(nm_i,
3263 nid, SETVEC_SIZE, setvec))) {
3266 nid = setvec[found - 1]->set + 1;
3267 for (idx = 0; idx < found; idx++) {
3268 /* entry_cnt is not zero, when cp_error was occurred */
3269 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3270 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3271 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3274 up_write(&nm_i->nat_tree_lock);
3276 kvfree(nm_i->nat_block_bitmap);
3277 if (nm_i->free_nid_bitmap) {
3280 for (i = 0; i < nm_i->nat_blocks; i++)
3281 kvfree(nm_i->free_nid_bitmap[i]);
3282 kvfree(nm_i->free_nid_bitmap);
3284 kvfree(nm_i->free_nid_count);
3286 kvfree(nm_i->nat_bitmap);
3287 kvfree(nm_i->nat_bits);
3288 #ifdef CONFIG_F2FS_CHECK_FS
3289 kvfree(nm_i->nat_bitmap_mir);
3291 sbi->nm_info = NULL;
3295 int __init f2fs_create_node_manager_caches(void)
3297 nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3298 sizeof(struct nat_entry));
3299 if (!nat_entry_slab)
3302 free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3303 sizeof(struct free_nid));
3305 goto destroy_nat_entry;
3307 nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3308 sizeof(struct nat_entry_set));
3309 if (!nat_entry_set_slab)
3310 goto destroy_free_nid;
3312 fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3313 sizeof(struct fsync_node_entry));
3314 if (!fsync_node_entry_slab)
3315 goto destroy_nat_entry_set;
3318 destroy_nat_entry_set:
3319 kmem_cache_destroy(nat_entry_set_slab);
3321 kmem_cache_destroy(free_nid_slab);
3323 kmem_cache_destroy(nat_entry_slab);
3328 void f2fs_destroy_node_manager_caches(void)
3330 kmem_cache_destroy(fsync_node_entry_slab);
3331 kmem_cache_destroy(nat_entry_set_slab);
3332 kmem_cache_destroy(free_nid_slab);
3333 kmem_cache_destroy(nat_entry_slab);