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>
21 #include <trace/events/f2fs.h>
23 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
25 static struct kmem_cache *nat_entry_slab;
26 static struct kmem_cache *free_nid_slab;
27 static struct kmem_cache *nat_entry_set_slab;
28 static struct kmem_cache *fsync_node_entry_slab;
31 * Check whether the given nid is within node id range.
33 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
35 if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
36 set_sbi_flag(sbi, SBI_NEED_FSCK);
37 f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
44 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
46 struct f2fs_nm_info *nm_i = NM_I(sbi);
47 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
49 unsigned long avail_ram;
50 unsigned long mem_size = 0;
58 /* only uses low memory */
59 avail_ram = val.totalram - val.totalhigh;
62 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
64 if (type == FREE_NIDS) {
65 mem_size = (nm_i->nid_cnt[FREE_NID] *
66 sizeof(struct free_nid)) >> PAGE_SHIFT;
67 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
68 } else if (type == NAT_ENTRIES) {
69 mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
70 sizeof(struct nat_entry)) >> PAGE_SHIFT;
71 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
72 if (excess_cached_nats(sbi))
74 } else if (type == DIRTY_DENTS) {
75 if (sbi->sb->s_bdi->wb.dirty_exceeded)
77 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
78 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
79 } else if (type == INO_ENTRIES) {
82 for (i = 0; i < MAX_INO_ENTRY; i++)
83 mem_size += sbi->im[i].ino_num *
84 sizeof(struct ino_entry);
85 mem_size >>= PAGE_SHIFT;
86 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
87 } else if (type == EXTENT_CACHE) {
88 mem_size = (atomic_read(&sbi->total_ext_tree) *
89 sizeof(struct extent_tree) +
90 atomic_read(&sbi->total_ext_node) *
91 sizeof(struct extent_node)) >> PAGE_SHIFT;
92 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
93 } else if (type == INMEM_PAGES) {
94 /* it allows 20% / total_ram for inmemory pages */
95 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
96 res = mem_size < (val.totalram / 5);
97 } else if (type == DISCARD_CACHE) {
98 mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
99 sizeof(struct discard_cmd)) >> PAGE_SHIFT;
100 res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
101 } else if (type == COMPRESS_PAGE) {
102 #ifdef CONFIG_F2FS_FS_COMPRESSION
103 unsigned long free_ram = val.freeram;
106 * free memory is lower than watermark or cached page count
107 * exceed threshold, deny caching compress page.
109 res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
110 (COMPRESS_MAPPING(sbi)->nrpages <
111 free_ram * sbi->compress_percent / 100);
116 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
122 static void clear_node_page_dirty(struct page *page)
124 if (PageDirty(page)) {
125 f2fs_clear_page_cache_dirty_tag(page);
126 clear_page_dirty_for_io(page);
127 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
129 ClearPageUptodate(page);
132 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
134 return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
137 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
139 struct page *src_page;
140 struct page *dst_page;
144 struct f2fs_nm_info *nm_i = NM_I(sbi);
146 dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
148 /* get current nat block page with lock */
149 src_page = get_current_nat_page(sbi, nid);
150 if (IS_ERR(src_page))
152 dst_page = f2fs_grab_meta_page(sbi, dst_off);
153 f2fs_bug_on(sbi, PageDirty(src_page));
155 src_addr = page_address(src_page);
156 dst_addr = page_address(dst_page);
157 memcpy(dst_addr, src_addr, PAGE_SIZE);
158 set_page_dirty(dst_page);
159 f2fs_put_page(src_page, 1);
161 set_to_next_nat(nm_i, nid);
166 static struct nat_entry *__alloc_nat_entry(struct f2fs_sb_info *sbi,
167 nid_t nid, bool no_fail)
169 struct nat_entry *new;
171 new = f2fs_kmem_cache_alloc(nat_entry_slab,
172 GFP_F2FS_ZERO, no_fail, sbi);
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,
246 GFP_NOFS, true, NULL);
248 INIT_LIST_HEAD(&head->entry_list);
249 INIT_LIST_HEAD(&head->set_list);
252 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
257 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
258 struct nat_entry *ne)
260 struct nat_entry_set *head;
261 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
264 head = __grab_nat_entry_set(nm_i, ne);
267 * update entry_cnt in below condition:
268 * 1. update NEW_ADDR to valid block address;
269 * 2. update old block address to new one;
271 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
272 !get_nat_flag(ne, IS_DIRTY)))
275 set_nat_flag(ne, IS_PREALLOC, new_ne);
277 if (get_nat_flag(ne, IS_DIRTY))
280 nm_i->nat_cnt[DIRTY_NAT]++;
281 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
282 set_nat_flag(ne, IS_DIRTY, true);
284 spin_lock(&nm_i->nat_list_lock);
286 list_del_init(&ne->list);
288 list_move_tail(&ne->list, &head->entry_list);
289 spin_unlock(&nm_i->nat_list_lock);
292 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
293 struct nat_entry_set *set, struct nat_entry *ne)
295 spin_lock(&nm_i->nat_list_lock);
296 list_move_tail(&ne->list, &nm_i->nat_entries);
297 spin_unlock(&nm_i->nat_list_lock);
299 set_nat_flag(ne, IS_DIRTY, false);
301 nm_i->nat_cnt[DIRTY_NAT]--;
302 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
305 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
306 nid_t start, unsigned int nr, struct nat_entry_set **ep)
308 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
312 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
314 return NODE_MAPPING(sbi) == page->mapping &&
315 IS_DNODE(page) && is_cold_node(page);
318 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
320 spin_lock_init(&sbi->fsync_node_lock);
321 INIT_LIST_HEAD(&sbi->fsync_node_list);
322 sbi->fsync_seg_id = 0;
323 sbi->fsync_node_num = 0;
326 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
329 struct fsync_node_entry *fn;
333 fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab,
334 GFP_NOFS, true, NULL);
338 INIT_LIST_HEAD(&fn->list);
340 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
341 list_add_tail(&fn->list, &sbi->fsync_node_list);
342 fn->seq_id = sbi->fsync_seg_id++;
344 sbi->fsync_node_num++;
345 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
350 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
352 struct fsync_node_entry *fn;
355 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
356 list_for_each_entry(fn, &sbi->fsync_node_list, list) {
357 if (fn->page == page) {
359 sbi->fsync_node_num--;
360 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
361 kmem_cache_free(fsync_node_entry_slab, fn);
366 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
370 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
374 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
375 sbi->fsync_seg_id = 0;
376 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
379 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
381 struct f2fs_nm_info *nm_i = NM_I(sbi);
385 down_read(&nm_i->nat_tree_lock);
386 e = __lookup_nat_cache(nm_i, nid);
388 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
389 !get_nat_flag(e, HAS_FSYNCED_INODE))
392 up_read(&nm_i->nat_tree_lock);
396 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
398 struct f2fs_nm_info *nm_i = NM_I(sbi);
402 down_read(&nm_i->nat_tree_lock);
403 e = __lookup_nat_cache(nm_i, nid);
404 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
406 up_read(&nm_i->nat_tree_lock);
410 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
412 struct f2fs_nm_info *nm_i = NM_I(sbi);
414 bool need_update = true;
416 down_read(&nm_i->nat_tree_lock);
417 e = __lookup_nat_cache(nm_i, ino);
418 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
419 (get_nat_flag(e, IS_CHECKPOINTED) ||
420 get_nat_flag(e, HAS_FSYNCED_INODE)))
422 up_read(&nm_i->nat_tree_lock);
426 /* must be locked by nat_tree_lock */
427 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
428 struct f2fs_nat_entry *ne)
430 struct f2fs_nm_info *nm_i = NM_I(sbi);
431 struct nat_entry *new, *e;
433 new = __alloc_nat_entry(sbi, nid, false);
437 down_write(&nm_i->nat_tree_lock);
438 e = __lookup_nat_cache(nm_i, nid);
440 e = __init_nat_entry(nm_i, new, ne, false);
442 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
443 nat_get_blkaddr(e) !=
444 le32_to_cpu(ne->block_addr) ||
445 nat_get_version(e) != ne->version);
446 up_write(&nm_i->nat_tree_lock);
448 __free_nat_entry(new);
451 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
452 block_t new_blkaddr, bool fsync_done)
454 struct f2fs_nm_info *nm_i = NM_I(sbi);
456 struct nat_entry *new = __alloc_nat_entry(sbi, ni->nid, true);
458 down_write(&nm_i->nat_tree_lock);
459 e = __lookup_nat_cache(nm_i, ni->nid);
461 e = __init_nat_entry(nm_i, new, NULL, true);
462 copy_node_info(&e->ni, ni);
463 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
464 } else if (new_blkaddr == NEW_ADDR) {
466 * when nid is reallocated,
467 * previous nat entry can be remained in nat cache.
468 * So, reinitialize it with new information.
470 copy_node_info(&e->ni, ni);
471 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
473 /* let's free early to reduce memory consumption */
475 __free_nat_entry(new);
478 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
479 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
480 new_blkaddr == NULL_ADDR);
481 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
482 new_blkaddr == NEW_ADDR);
483 f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
484 new_blkaddr == NEW_ADDR);
486 /* increment version no as node is removed */
487 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
488 unsigned char version = nat_get_version(e);
490 nat_set_version(e, inc_node_version(version));
494 nat_set_blkaddr(e, new_blkaddr);
495 if (!__is_valid_data_blkaddr(new_blkaddr))
496 set_nat_flag(e, IS_CHECKPOINTED, false);
497 __set_nat_cache_dirty(nm_i, e);
499 /* update fsync_mark if its inode nat entry is still alive */
500 if (ni->nid != ni->ino)
501 e = __lookup_nat_cache(nm_i, ni->ino);
503 if (fsync_done && ni->nid == ni->ino)
504 set_nat_flag(e, HAS_FSYNCED_INODE, true);
505 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
507 up_write(&nm_i->nat_tree_lock);
510 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
512 struct f2fs_nm_info *nm_i = NM_I(sbi);
515 if (!down_write_trylock(&nm_i->nat_tree_lock))
518 spin_lock(&nm_i->nat_list_lock);
520 struct nat_entry *ne;
522 if (list_empty(&nm_i->nat_entries))
525 ne = list_first_entry(&nm_i->nat_entries,
526 struct nat_entry, list);
528 spin_unlock(&nm_i->nat_list_lock);
530 __del_from_nat_cache(nm_i, ne);
533 spin_lock(&nm_i->nat_list_lock);
535 spin_unlock(&nm_i->nat_list_lock);
537 up_write(&nm_i->nat_tree_lock);
538 return nr - nr_shrink;
541 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
542 struct node_info *ni)
544 struct f2fs_nm_info *nm_i = NM_I(sbi);
545 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
546 struct f2fs_journal *journal = curseg->journal;
547 nid_t start_nid = START_NID(nid);
548 struct f2fs_nat_block *nat_blk;
549 struct page *page = NULL;
550 struct f2fs_nat_entry ne;
558 /* Check nat cache */
559 down_read(&nm_i->nat_tree_lock);
560 e = __lookup_nat_cache(nm_i, nid);
562 ni->ino = nat_get_ino(e);
563 ni->blk_addr = nat_get_blkaddr(e);
564 ni->version = nat_get_version(e);
565 up_read(&nm_i->nat_tree_lock);
570 * Check current segment summary by trying to grab journal_rwsem first.
571 * This sem is on the critical path on the checkpoint requiring the above
572 * nat_tree_lock. Therefore, we should retry, if we failed to grab here
573 * while not bothering checkpoint.
575 if (!rwsem_is_locked(&sbi->cp_global_sem)) {
576 down_read(&curseg->journal_rwsem);
577 } else if (!down_read_trylock(&curseg->journal_rwsem)) {
578 up_read(&nm_i->nat_tree_lock);
582 i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
584 ne = nat_in_journal(journal, i);
585 node_info_from_raw_nat(ni, &ne);
587 up_read(&curseg->journal_rwsem);
589 up_read(&nm_i->nat_tree_lock);
593 /* Fill node_info from nat page */
594 index = current_nat_addr(sbi, nid);
595 up_read(&nm_i->nat_tree_lock);
597 page = f2fs_get_meta_page(sbi, index);
599 return PTR_ERR(page);
601 nat_blk = (struct f2fs_nat_block *)page_address(page);
602 ne = nat_blk->entries[nid - start_nid];
603 node_info_from_raw_nat(ni, &ne);
604 f2fs_put_page(page, 1);
606 blkaddr = le32_to_cpu(ne.block_addr);
607 if (__is_valid_data_blkaddr(blkaddr) &&
608 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
611 /* cache nat entry */
612 cache_nat_entry(sbi, nid, &ne);
617 * readahead MAX_RA_NODE number of node pages.
619 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
621 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
622 struct blk_plug plug;
626 blk_start_plug(&plug);
628 /* Then, try readahead for siblings of the desired node */
630 end = min(end, NIDS_PER_BLOCK);
631 for (i = start; i < end; i++) {
632 nid = get_nid(parent, i, false);
633 f2fs_ra_node_page(sbi, nid);
636 blk_finish_plug(&plug);
639 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
641 const long direct_index = ADDRS_PER_INODE(dn->inode);
642 const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
643 const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
644 unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
645 int cur_level = dn->cur_level;
646 int max_level = dn->max_level;
652 while (max_level-- > cur_level)
653 skipped_unit *= NIDS_PER_BLOCK;
655 switch (dn->max_level) {
657 base += 2 * indirect_blks;
660 base += 2 * direct_blks;
663 base += direct_index;
666 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
669 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
673 * The maximum depth is four.
674 * Offset[0] will have raw inode offset.
676 static int get_node_path(struct inode *inode, long block,
677 int offset[4], unsigned int noffset[4])
679 const long direct_index = ADDRS_PER_INODE(inode);
680 const long direct_blks = ADDRS_PER_BLOCK(inode);
681 const long dptrs_per_blk = NIDS_PER_BLOCK;
682 const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
683 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
689 if (block < direct_index) {
693 block -= direct_index;
694 if (block < direct_blks) {
695 offset[n++] = NODE_DIR1_BLOCK;
701 block -= direct_blks;
702 if (block < direct_blks) {
703 offset[n++] = NODE_DIR2_BLOCK;
709 block -= direct_blks;
710 if (block < indirect_blks) {
711 offset[n++] = NODE_IND1_BLOCK;
713 offset[n++] = block / direct_blks;
714 noffset[n] = 4 + offset[n - 1];
715 offset[n] = block % direct_blks;
719 block -= indirect_blks;
720 if (block < indirect_blks) {
721 offset[n++] = NODE_IND2_BLOCK;
722 noffset[n] = 4 + dptrs_per_blk;
723 offset[n++] = block / direct_blks;
724 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
725 offset[n] = block % direct_blks;
729 block -= indirect_blks;
730 if (block < dindirect_blks) {
731 offset[n++] = NODE_DIND_BLOCK;
732 noffset[n] = 5 + (dptrs_per_blk * 2);
733 offset[n++] = block / indirect_blks;
734 noffset[n] = 6 + (dptrs_per_blk * 2) +
735 offset[n - 1] * (dptrs_per_blk + 1);
736 offset[n++] = (block / direct_blks) % dptrs_per_blk;
737 noffset[n] = 7 + (dptrs_per_blk * 2) +
738 offset[n - 2] * (dptrs_per_blk + 1) +
740 offset[n] = block % direct_blks;
751 * Caller should call f2fs_put_dnode(dn).
752 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
753 * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
755 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
757 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
758 struct page *npage[4];
759 struct page *parent = NULL;
761 unsigned int noffset[4];
766 level = get_node_path(dn->inode, index, offset, noffset);
770 nids[0] = dn->inode->i_ino;
771 npage[0] = dn->inode_page;
774 npage[0] = f2fs_get_node_page(sbi, nids[0]);
775 if (IS_ERR(npage[0]))
776 return PTR_ERR(npage[0]);
779 /* if inline_data is set, should not report any block indices */
780 if (f2fs_has_inline_data(dn->inode) && index) {
782 f2fs_put_page(npage[0], 1);
788 nids[1] = get_nid(parent, offset[0], true);
789 dn->inode_page = npage[0];
790 dn->inode_page_locked = true;
792 /* get indirect or direct nodes */
793 for (i = 1; i <= level; i++) {
796 if (!nids[i] && mode == ALLOC_NODE) {
798 if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
804 npage[i] = f2fs_new_node_page(dn, noffset[i]);
805 if (IS_ERR(npage[i])) {
806 f2fs_alloc_nid_failed(sbi, nids[i]);
807 err = PTR_ERR(npage[i]);
811 set_nid(parent, offset[i - 1], nids[i], i == 1);
812 f2fs_alloc_nid_done(sbi, nids[i]);
814 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
815 npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
816 if (IS_ERR(npage[i])) {
817 err = PTR_ERR(npage[i]);
823 dn->inode_page_locked = false;
826 f2fs_put_page(parent, 1);
830 npage[i] = f2fs_get_node_page(sbi, nids[i]);
831 if (IS_ERR(npage[i])) {
832 err = PTR_ERR(npage[i]);
833 f2fs_put_page(npage[0], 0);
839 nids[i + 1] = get_nid(parent, offset[i], false);
842 dn->nid = nids[level];
843 dn->ofs_in_node = offset[level];
844 dn->node_page = npage[level];
845 dn->data_blkaddr = f2fs_data_blkaddr(dn);
847 if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) &&
848 f2fs_sb_has_readonly(sbi)) {
849 unsigned int c_len = f2fs_cluster_blocks_are_contiguous(dn);
855 blkaddr = f2fs_data_blkaddr(dn);
856 if (blkaddr == COMPRESS_ADDR)
857 blkaddr = data_blkaddr(dn->inode, dn->node_page,
858 dn->ofs_in_node + 1);
860 f2fs_update_extent_tree_range_compressed(dn->inode,
862 F2FS_I(dn->inode)->i_cluster_size,
869 f2fs_put_page(parent, 1);
871 f2fs_put_page(npage[0], 0);
873 dn->inode_page = NULL;
874 dn->node_page = NULL;
875 if (err == -ENOENT) {
877 dn->max_level = level;
878 dn->ofs_in_node = offset[level];
883 static int truncate_node(struct dnode_of_data *dn)
885 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
890 err = f2fs_get_node_info(sbi, dn->nid, &ni);
894 /* Deallocate node address */
895 f2fs_invalidate_blocks(sbi, ni.blk_addr);
896 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
897 set_node_addr(sbi, &ni, NULL_ADDR, false);
899 if (dn->nid == dn->inode->i_ino) {
900 f2fs_remove_orphan_inode(sbi, dn->nid);
901 dec_valid_inode_count(sbi);
902 f2fs_inode_synced(dn->inode);
905 clear_node_page_dirty(dn->node_page);
906 set_sbi_flag(sbi, SBI_IS_DIRTY);
908 index = dn->node_page->index;
909 f2fs_put_page(dn->node_page, 1);
911 invalidate_mapping_pages(NODE_MAPPING(sbi),
914 dn->node_page = NULL;
915 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
920 static int truncate_dnode(struct dnode_of_data *dn)
928 /* get direct node */
929 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
930 if (PTR_ERR(page) == -ENOENT)
932 else if (IS_ERR(page))
933 return PTR_ERR(page);
935 /* Make dnode_of_data for parameter */
936 dn->node_page = page;
938 f2fs_truncate_data_blocks(dn);
939 err = truncate_node(dn);
946 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
949 struct dnode_of_data rdn = *dn;
951 struct f2fs_node *rn;
953 unsigned int child_nofs;
958 return NIDS_PER_BLOCK + 1;
960 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
962 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
964 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
965 return PTR_ERR(page);
968 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
970 rn = F2FS_NODE(page);
972 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
973 child_nid = le32_to_cpu(rn->in.nid[i]);
977 ret = truncate_dnode(&rdn);
980 if (set_nid(page, i, 0, false))
981 dn->node_changed = true;
984 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
985 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
986 child_nid = le32_to_cpu(rn->in.nid[i]);
987 if (child_nid == 0) {
988 child_nofs += NIDS_PER_BLOCK + 1;
992 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
993 if (ret == (NIDS_PER_BLOCK + 1)) {
994 if (set_nid(page, i, 0, false))
995 dn->node_changed = true;
997 } else if (ret < 0 && ret != -ENOENT) {
1005 /* remove current indirect node */
1006 dn->node_page = page;
1007 ret = truncate_node(dn);
1012 f2fs_put_page(page, 1);
1014 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
1018 f2fs_put_page(page, 1);
1019 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
1023 static int truncate_partial_nodes(struct dnode_of_data *dn,
1024 struct f2fs_inode *ri, int *offset, int depth)
1026 struct page *pages[2];
1031 int idx = depth - 2;
1033 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1037 /* get indirect nodes in the path */
1038 for (i = 0; i < idx + 1; i++) {
1039 /* reference count'll be increased */
1040 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
1041 if (IS_ERR(pages[i])) {
1042 err = PTR_ERR(pages[i]);
1046 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1049 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1051 /* free direct nodes linked to a partial indirect node */
1052 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1053 child_nid = get_nid(pages[idx], i, false);
1056 dn->nid = child_nid;
1057 err = truncate_dnode(dn);
1060 if (set_nid(pages[idx], i, 0, false))
1061 dn->node_changed = true;
1064 if (offset[idx + 1] == 0) {
1065 dn->node_page = pages[idx];
1067 err = truncate_node(dn);
1071 f2fs_put_page(pages[idx], 1);
1074 offset[idx + 1] = 0;
1077 for (i = idx; i >= 0; i--)
1078 f2fs_put_page(pages[i], 1);
1080 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1086 * All the block addresses of data and nodes should be nullified.
1088 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1090 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1091 int err = 0, cont = 1;
1092 int level, offset[4], noffset[4];
1093 unsigned int nofs = 0;
1094 struct f2fs_inode *ri;
1095 struct dnode_of_data dn;
1098 trace_f2fs_truncate_inode_blocks_enter(inode, from);
1100 level = get_node_path(inode, from, offset, noffset);
1102 trace_f2fs_truncate_inode_blocks_exit(inode, level);
1106 page = f2fs_get_node_page(sbi, inode->i_ino);
1108 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1109 return PTR_ERR(page);
1112 set_new_dnode(&dn, inode, page, NULL, 0);
1115 ri = F2FS_INODE(page);
1123 if (!offset[level - 1])
1125 err = truncate_partial_nodes(&dn, ri, offset, level);
1126 if (err < 0 && err != -ENOENT)
1128 nofs += 1 + NIDS_PER_BLOCK;
1131 nofs = 5 + 2 * NIDS_PER_BLOCK;
1132 if (!offset[level - 1])
1134 err = truncate_partial_nodes(&dn, ri, offset, level);
1135 if (err < 0 && err != -ENOENT)
1144 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1145 switch (offset[0]) {
1146 case NODE_DIR1_BLOCK:
1147 case NODE_DIR2_BLOCK:
1148 err = truncate_dnode(&dn);
1151 case NODE_IND1_BLOCK:
1152 case NODE_IND2_BLOCK:
1153 err = truncate_nodes(&dn, nofs, offset[1], 2);
1156 case NODE_DIND_BLOCK:
1157 err = truncate_nodes(&dn, nofs, offset[1], 3);
1164 if (err < 0 && err != -ENOENT)
1166 if (offset[1] == 0 &&
1167 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1169 BUG_ON(page->mapping != NODE_MAPPING(sbi));
1170 f2fs_wait_on_page_writeback(page, NODE, true, true);
1171 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1172 set_page_dirty(page);
1180 f2fs_put_page(page, 0);
1181 trace_f2fs_truncate_inode_blocks_exit(inode, err);
1182 return err > 0 ? 0 : err;
1185 /* caller must lock inode page */
1186 int f2fs_truncate_xattr_node(struct inode *inode)
1188 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1189 nid_t nid = F2FS_I(inode)->i_xattr_nid;
1190 struct dnode_of_data dn;
1197 npage = f2fs_get_node_page(sbi, nid);
1199 return PTR_ERR(npage);
1201 set_new_dnode(&dn, inode, NULL, npage, nid);
1202 err = truncate_node(&dn);
1204 f2fs_put_page(npage, 1);
1208 f2fs_i_xnid_write(inode, 0);
1214 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1217 int f2fs_remove_inode_page(struct inode *inode)
1219 struct dnode_of_data dn;
1222 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1223 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1227 err = f2fs_truncate_xattr_node(inode);
1229 f2fs_put_dnode(&dn);
1233 /* remove potential inline_data blocks */
1234 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1235 S_ISLNK(inode->i_mode))
1236 f2fs_truncate_data_blocks_range(&dn, 1);
1238 /* 0 is possible, after f2fs_new_inode() has failed */
1239 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1240 f2fs_put_dnode(&dn);
1244 if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1245 f2fs_warn(F2FS_I_SB(inode),
1246 "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1247 inode->i_ino, (unsigned long long)inode->i_blocks);
1248 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1251 /* will put inode & node pages */
1252 err = truncate_node(&dn);
1254 f2fs_put_dnode(&dn);
1260 struct page *f2fs_new_inode_page(struct inode *inode)
1262 struct dnode_of_data dn;
1264 /* allocate inode page for new inode */
1265 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1267 /* caller should f2fs_put_page(page, 1); */
1268 return f2fs_new_node_page(&dn, 0);
1271 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1273 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1274 struct node_info new_ni;
1278 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1279 return ERR_PTR(-EPERM);
1281 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1283 return ERR_PTR(-ENOMEM);
1285 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1288 #ifdef CONFIG_F2FS_CHECK_FS
1289 err = f2fs_get_node_info(sbi, dn->nid, &new_ni);
1291 dec_valid_node_count(sbi, dn->inode, !ofs);
1294 f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1296 new_ni.nid = dn->nid;
1297 new_ni.ino = dn->inode->i_ino;
1298 new_ni.blk_addr = NULL_ADDR;
1301 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1303 f2fs_wait_on_page_writeback(page, NODE, true, true);
1304 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1305 set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1306 if (!PageUptodate(page))
1307 SetPageUptodate(page);
1308 if (set_page_dirty(page))
1309 dn->node_changed = true;
1311 if (f2fs_has_xattr_block(ofs))
1312 f2fs_i_xnid_write(dn->inode, dn->nid);
1315 inc_valid_inode_count(sbi);
1319 clear_node_page_dirty(page);
1320 f2fs_put_page(page, 1);
1321 return ERR_PTR(err);
1325 * Caller should do after getting the following values.
1326 * 0: f2fs_put_page(page, 0)
1327 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1329 static int read_node_page(struct page *page, int op_flags)
1331 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1332 struct node_info ni;
1333 struct f2fs_io_info fio = {
1337 .op_flags = op_flags,
1339 .encrypted_page = NULL,
1343 if (PageUptodate(page)) {
1344 if (!f2fs_inode_chksum_verify(sbi, page)) {
1345 ClearPageUptodate(page);
1351 err = f2fs_get_node_info(sbi, page->index, &ni);
1355 /* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1356 if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR) ||
1357 is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1358 ClearPageUptodate(page);
1362 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1364 err = f2fs_submit_page_bio(&fio);
1367 f2fs_update_iostat(sbi, FS_NODE_READ_IO, F2FS_BLKSIZE);
1373 * Readahead a node page
1375 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1382 if (f2fs_check_nid_range(sbi, nid))
1385 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1389 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1393 err = read_node_page(apage, REQ_RAHEAD);
1394 f2fs_put_page(apage, err ? 1 : 0);
1397 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1398 struct page *parent, int start)
1404 return ERR_PTR(-ENOENT);
1405 if (f2fs_check_nid_range(sbi, nid))
1406 return ERR_PTR(-EINVAL);
1408 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1410 return ERR_PTR(-ENOMEM);
1412 err = read_node_page(page, 0);
1414 f2fs_put_page(page, 1);
1415 return ERR_PTR(err);
1416 } else if (err == LOCKED_PAGE) {
1422 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1426 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1427 f2fs_put_page(page, 1);
1431 if (unlikely(!PageUptodate(page))) {
1436 if (!f2fs_inode_chksum_verify(sbi, page)) {
1441 if (unlikely(nid != nid_of_node(page))) {
1442 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1443 nid, nid_of_node(page), ino_of_node(page),
1444 ofs_of_node(page), cpver_of_node(page),
1445 next_blkaddr_of_node(page));
1446 set_sbi_flag(sbi, SBI_NEED_FSCK);
1449 ClearPageUptodate(page);
1450 f2fs_put_page(page, 1);
1451 return ERR_PTR(err);
1456 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1458 return __get_node_page(sbi, nid, NULL, 0);
1461 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1463 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1464 nid_t nid = get_nid(parent, start, false);
1466 return __get_node_page(sbi, nid, parent, start);
1469 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1471 struct inode *inode;
1475 /* should flush inline_data before evict_inode */
1476 inode = ilookup(sbi->sb, ino);
1480 page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1481 FGP_LOCK|FGP_NOWAIT, 0);
1485 if (!PageUptodate(page))
1488 if (!PageDirty(page))
1491 if (!clear_page_dirty_for_io(page))
1494 ret = f2fs_write_inline_data(inode, page);
1495 inode_dec_dirty_pages(inode);
1496 f2fs_remove_dirty_inode(inode);
1498 set_page_dirty(page);
1500 f2fs_put_page(page, 1);
1505 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1508 struct pagevec pvec;
1509 struct page *last_page = NULL;
1512 pagevec_init(&pvec);
1515 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1516 PAGECACHE_TAG_DIRTY))) {
1519 for (i = 0; i < nr_pages; i++) {
1520 struct page *page = pvec.pages[i];
1522 if (unlikely(f2fs_cp_error(sbi))) {
1523 f2fs_put_page(last_page, 0);
1524 pagevec_release(&pvec);
1525 return ERR_PTR(-EIO);
1528 if (!IS_DNODE(page) || !is_cold_node(page))
1530 if (ino_of_node(page) != ino)
1535 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1540 if (ino_of_node(page) != ino)
1541 goto continue_unlock;
1543 if (!PageDirty(page)) {
1544 /* someone wrote it for us */
1545 goto continue_unlock;
1549 f2fs_put_page(last_page, 0);
1555 pagevec_release(&pvec);
1561 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1562 struct writeback_control *wbc, bool do_balance,
1563 enum iostat_type io_type, unsigned int *seq_id)
1565 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1567 struct node_info ni;
1568 struct f2fs_io_info fio = {
1570 .ino = ino_of_node(page),
1573 .op_flags = wbc_to_write_flags(wbc),
1575 .encrypted_page = NULL,
1582 trace_f2fs_writepage(page, NODE);
1584 if (unlikely(f2fs_cp_error(sbi))) {
1585 ClearPageUptodate(page);
1586 dec_page_count(sbi, F2FS_DIRTY_NODES);
1591 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1594 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1595 wbc->sync_mode == WB_SYNC_NONE &&
1596 IS_DNODE(page) && is_cold_node(page))
1599 /* get old block addr of this node page */
1600 nid = nid_of_node(page);
1601 f2fs_bug_on(sbi, page->index != nid);
1603 if (f2fs_get_node_info(sbi, nid, &ni))
1606 if (wbc->for_reclaim) {
1607 if (!down_read_trylock(&sbi->node_write))
1610 down_read(&sbi->node_write);
1613 /* This page is already truncated */
1614 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1615 ClearPageUptodate(page);
1616 dec_page_count(sbi, F2FS_DIRTY_NODES);
1617 up_read(&sbi->node_write);
1622 if (__is_valid_data_blkaddr(ni.blk_addr) &&
1623 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1624 DATA_GENERIC_ENHANCE)) {
1625 up_read(&sbi->node_write);
1629 if (atomic && !test_opt(sbi, NOBARRIER))
1630 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1632 /* should add to global list before clearing PAGECACHE status */
1633 if (f2fs_in_warm_node_list(sbi, page)) {
1634 seq = f2fs_add_fsync_node_entry(sbi, page);
1639 set_page_writeback(page);
1640 ClearPageError(page);
1642 fio.old_blkaddr = ni.blk_addr;
1643 f2fs_do_write_node_page(nid, &fio);
1644 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1645 dec_page_count(sbi, F2FS_DIRTY_NODES);
1646 up_read(&sbi->node_write);
1648 if (wbc->for_reclaim) {
1649 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1655 if (unlikely(f2fs_cp_error(sbi))) {
1656 f2fs_submit_merged_write(sbi, NODE);
1660 *submitted = fio.submitted;
1663 f2fs_balance_fs(sbi, false);
1667 redirty_page_for_writepage(wbc, page);
1668 return AOP_WRITEPAGE_ACTIVATE;
1671 int f2fs_move_node_page(struct page *node_page, int gc_type)
1675 if (gc_type == FG_GC) {
1676 struct writeback_control wbc = {
1677 .sync_mode = WB_SYNC_ALL,
1682 f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1684 set_page_dirty(node_page);
1686 if (!clear_page_dirty_for_io(node_page)) {
1691 if (__write_node_page(node_page, false, NULL,
1692 &wbc, false, FS_GC_NODE_IO, NULL)) {
1694 unlock_page(node_page);
1698 /* set page dirty and write it */
1699 if (!PageWriteback(node_page))
1700 set_page_dirty(node_page);
1703 unlock_page(node_page);
1705 f2fs_put_page(node_page, 0);
1709 static int f2fs_write_node_page(struct page *page,
1710 struct writeback_control *wbc)
1712 return __write_node_page(page, false, NULL, wbc, false,
1716 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1717 struct writeback_control *wbc, bool atomic,
1718 unsigned int *seq_id)
1721 struct pagevec pvec;
1723 struct page *last_page = NULL;
1724 bool marked = false;
1725 nid_t ino = inode->i_ino;
1730 last_page = last_fsync_dnode(sbi, ino);
1731 if (IS_ERR_OR_NULL(last_page))
1732 return PTR_ERR_OR_ZERO(last_page);
1735 pagevec_init(&pvec);
1738 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1739 PAGECACHE_TAG_DIRTY))) {
1742 for (i = 0; i < nr_pages; i++) {
1743 struct page *page = pvec.pages[i];
1744 bool submitted = false;
1746 if (unlikely(f2fs_cp_error(sbi))) {
1747 f2fs_put_page(last_page, 0);
1748 pagevec_release(&pvec);
1753 if (!IS_DNODE(page) || !is_cold_node(page))
1755 if (ino_of_node(page) != ino)
1760 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1765 if (ino_of_node(page) != ino)
1766 goto continue_unlock;
1768 if (!PageDirty(page) && page != last_page) {
1769 /* someone wrote it for us */
1770 goto continue_unlock;
1773 f2fs_wait_on_page_writeback(page, NODE, true, true);
1775 set_fsync_mark(page, 0);
1776 set_dentry_mark(page, 0);
1778 if (!atomic || page == last_page) {
1779 set_fsync_mark(page, 1);
1780 if (IS_INODE(page)) {
1781 if (is_inode_flag_set(inode,
1783 f2fs_update_inode(inode, page);
1784 set_dentry_mark(page,
1785 f2fs_need_dentry_mark(sbi, ino));
1787 /* may be written by other thread */
1788 if (!PageDirty(page))
1789 set_page_dirty(page);
1792 if (!clear_page_dirty_for_io(page))
1793 goto continue_unlock;
1795 ret = __write_node_page(page, atomic &&
1797 &submitted, wbc, true,
1798 FS_NODE_IO, seq_id);
1801 f2fs_put_page(last_page, 0);
1803 } else if (submitted) {
1807 if (page == last_page) {
1808 f2fs_put_page(page, 0);
1813 pagevec_release(&pvec);
1819 if (!ret && atomic && !marked) {
1820 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1821 ino, last_page->index);
1822 lock_page(last_page);
1823 f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1824 set_page_dirty(last_page);
1825 unlock_page(last_page);
1830 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1831 return ret ? -EIO : 0;
1834 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1836 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1839 if (inode->i_ino != ino)
1842 if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1845 spin_lock(&sbi->inode_lock[DIRTY_META]);
1846 clean = list_empty(&F2FS_I(inode)->gdirty_list);
1847 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1852 inode = igrab(inode);
1858 static bool flush_dirty_inode(struct page *page)
1860 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1861 struct inode *inode;
1862 nid_t ino = ino_of_node(page);
1864 inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1868 f2fs_update_inode(inode, page);
1875 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1878 struct pagevec pvec;
1881 pagevec_init(&pvec);
1883 while ((nr_pages = pagevec_lookup_tag(&pvec,
1884 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1887 for (i = 0; i < nr_pages; i++) {
1888 struct page *page = pvec.pages[i];
1890 if (!IS_DNODE(page))
1895 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1901 if (!PageDirty(page)) {
1902 /* someone wrote it for us */
1903 goto continue_unlock;
1906 /* flush inline_data, if it's async context. */
1907 if (page_private_inline(page)) {
1908 clear_page_private_inline(page);
1910 flush_inline_data(sbi, ino_of_node(page));
1915 pagevec_release(&pvec);
1920 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1921 struct writeback_control *wbc,
1922 bool do_balance, enum iostat_type io_type)
1925 struct pagevec pvec;
1929 int nr_pages, done = 0;
1931 pagevec_init(&pvec);
1936 while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1937 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1940 for (i = 0; i < nr_pages; i++) {
1941 struct page *page = pvec.pages[i];
1942 bool submitted = false;
1943 bool may_dirty = true;
1945 /* give a priority to WB_SYNC threads */
1946 if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1947 wbc->sync_mode == WB_SYNC_NONE) {
1953 * flushing sequence with step:
1958 if (step == 0 && IS_DNODE(page))
1960 if (step == 1 && (!IS_DNODE(page) ||
1961 is_cold_node(page)))
1963 if (step == 2 && (!IS_DNODE(page) ||
1964 !is_cold_node(page)))
1967 if (wbc->sync_mode == WB_SYNC_ALL)
1969 else if (!trylock_page(page))
1972 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1978 if (!PageDirty(page)) {
1979 /* someone wrote it for us */
1980 goto continue_unlock;
1983 /* flush inline_data/inode, if it's async context. */
1987 /* flush inline_data */
1988 if (page_private_inline(page)) {
1989 clear_page_private_inline(page);
1991 flush_inline_data(sbi, ino_of_node(page));
1995 /* flush dirty inode */
1996 if (IS_INODE(page) && may_dirty) {
1998 if (flush_dirty_inode(page))
2002 f2fs_wait_on_page_writeback(page, NODE, true, true);
2004 if (!clear_page_dirty_for_io(page))
2005 goto continue_unlock;
2007 set_fsync_mark(page, 0);
2008 set_dentry_mark(page, 0);
2010 ret = __write_node_page(page, false, &submitted,
2011 wbc, do_balance, io_type, NULL);
2017 if (--wbc->nr_to_write == 0)
2020 pagevec_release(&pvec);
2023 if (wbc->nr_to_write == 0) {
2030 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2031 wbc->sync_mode == WB_SYNC_NONE && step == 1)
2038 f2fs_submit_merged_write(sbi, NODE);
2040 if (unlikely(f2fs_cp_error(sbi)))
2045 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2046 unsigned int seq_id)
2048 struct fsync_node_entry *fn;
2050 struct list_head *head = &sbi->fsync_node_list;
2051 unsigned long flags;
2052 unsigned int cur_seq_id = 0;
2055 while (seq_id && cur_seq_id < seq_id) {
2056 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2057 if (list_empty(head)) {
2058 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2061 fn = list_first_entry(head, struct fsync_node_entry, list);
2062 if (fn->seq_id > seq_id) {
2063 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2066 cur_seq_id = fn->seq_id;
2069 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2071 f2fs_wait_on_page_writeback(page, NODE, true, false);
2072 if (TestClearPageError(page))
2081 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2088 static int f2fs_write_node_pages(struct address_space *mapping,
2089 struct writeback_control *wbc)
2091 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2092 struct blk_plug plug;
2095 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2098 /* balancing f2fs's metadata in background */
2099 f2fs_balance_fs_bg(sbi, true);
2101 /* collect a number of dirty node pages and write together */
2102 if (wbc->sync_mode != WB_SYNC_ALL &&
2103 get_pages(sbi, F2FS_DIRTY_NODES) <
2104 nr_pages_to_skip(sbi, NODE))
2107 if (wbc->sync_mode == WB_SYNC_ALL)
2108 atomic_inc(&sbi->wb_sync_req[NODE]);
2109 else if (atomic_read(&sbi->wb_sync_req[NODE]))
2112 trace_f2fs_writepages(mapping->host, wbc, NODE);
2114 diff = nr_pages_to_write(sbi, NODE, wbc);
2115 blk_start_plug(&plug);
2116 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2117 blk_finish_plug(&plug);
2118 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2120 if (wbc->sync_mode == WB_SYNC_ALL)
2121 atomic_dec(&sbi->wb_sync_req[NODE]);
2125 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2126 trace_f2fs_writepages(mapping->host, wbc, NODE);
2130 static int f2fs_set_node_page_dirty(struct page *page)
2132 trace_f2fs_set_page_dirty(page, NODE);
2134 if (!PageUptodate(page))
2135 SetPageUptodate(page);
2136 #ifdef CONFIG_F2FS_CHECK_FS
2138 f2fs_inode_chksum_set(F2FS_P_SB(page), page);
2140 if (!PageDirty(page)) {
2141 __set_page_dirty_nobuffers(page);
2142 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
2143 set_page_private_reference(page);
2150 * Structure of the f2fs node operations
2152 const struct address_space_operations f2fs_node_aops = {
2153 .writepage = f2fs_write_node_page,
2154 .writepages = f2fs_write_node_pages,
2155 .set_page_dirty = f2fs_set_node_page_dirty,
2156 .invalidatepage = f2fs_invalidate_page,
2157 .releasepage = f2fs_release_page,
2158 #ifdef CONFIG_MIGRATION
2159 .migratepage = f2fs_migrate_page,
2163 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2166 return radix_tree_lookup(&nm_i->free_nid_root, n);
2169 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2172 struct f2fs_nm_info *nm_i = NM_I(sbi);
2173 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2178 nm_i->nid_cnt[FREE_NID]++;
2179 list_add_tail(&i->list, &nm_i->free_nid_list);
2183 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2184 struct free_nid *i, enum nid_state state)
2186 struct f2fs_nm_info *nm_i = NM_I(sbi);
2188 f2fs_bug_on(sbi, state != i->state);
2189 nm_i->nid_cnt[state]--;
2190 if (state == FREE_NID)
2192 radix_tree_delete(&nm_i->free_nid_root, i->nid);
2195 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2196 enum nid_state org_state, enum nid_state dst_state)
2198 struct f2fs_nm_info *nm_i = NM_I(sbi);
2200 f2fs_bug_on(sbi, org_state != i->state);
2201 i->state = dst_state;
2202 nm_i->nid_cnt[org_state]--;
2203 nm_i->nid_cnt[dst_state]++;
2205 switch (dst_state) {
2210 list_add_tail(&i->list, &nm_i->free_nid_list);
2217 bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi)
2219 struct f2fs_nm_info *nm_i = NM_I(sbi);
2223 down_read(&nm_i->nat_tree_lock);
2224 for (i = 0; i < nm_i->nat_blocks; i++) {
2225 if (!test_bit_le(i, nm_i->nat_block_bitmap)) {
2230 up_read(&nm_i->nat_tree_lock);
2235 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2236 bool set, bool build)
2238 struct f2fs_nm_info *nm_i = NM_I(sbi);
2239 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2240 unsigned int nid_ofs = nid - START_NID(nid);
2242 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2246 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2248 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2249 nm_i->free_nid_count[nat_ofs]++;
2251 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2253 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2255 nm_i->free_nid_count[nat_ofs]--;
2259 /* return if the nid is recognized as free */
2260 static bool add_free_nid(struct f2fs_sb_info *sbi,
2261 nid_t nid, bool build, bool update)
2263 struct f2fs_nm_info *nm_i = NM_I(sbi);
2264 struct free_nid *i, *e;
2265 struct nat_entry *ne;
2269 /* 0 nid should not be used */
2270 if (unlikely(nid == 0))
2273 if (unlikely(f2fs_check_nid_range(sbi, nid)))
2276 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL);
2278 i->state = FREE_NID;
2280 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2282 spin_lock(&nm_i->nid_list_lock);
2290 * - __insert_nid_to_list(PREALLOC_NID)
2291 * - f2fs_balance_fs_bg
2292 * - f2fs_build_free_nids
2293 * - __f2fs_build_free_nids
2296 * - __lookup_nat_cache
2298 * - f2fs_init_inode_metadata
2299 * - f2fs_new_inode_page
2300 * - f2fs_new_node_page
2302 * - f2fs_alloc_nid_done
2303 * - __remove_nid_from_list(PREALLOC_NID)
2304 * - __insert_nid_to_list(FREE_NID)
2306 ne = __lookup_nat_cache(nm_i, nid);
2307 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2308 nat_get_blkaddr(ne) != NULL_ADDR))
2311 e = __lookup_free_nid_list(nm_i, nid);
2313 if (e->state == FREE_NID)
2319 err = __insert_free_nid(sbi, i);
2322 update_free_nid_bitmap(sbi, nid, ret, build);
2324 nm_i->available_nids++;
2326 spin_unlock(&nm_i->nid_list_lock);
2327 radix_tree_preload_end();
2330 kmem_cache_free(free_nid_slab, i);
2334 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2336 struct f2fs_nm_info *nm_i = NM_I(sbi);
2338 bool need_free = false;
2340 spin_lock(&nm_i->nid_list_lock);
2341 i = __lookup_free_nid_list(nm_i, nid);
2342 if (i && i->state == FREE_NID) {
2343 __remove_free_nid(sbi, i, FREE_NID);
2346 spin_unlock(&nm_i->nid_list_lock);
2349 kmem_cache_free(free_nid_slab, i);
2352 static int scan_nat_page(struct f2fs_sb_info *sbi,
2353 struct page *nat_page, nid_t start_nid)
2355 struct f2fs_nm_info *nm_i = NM_I(sbi);
2356 struct f2fs_nat_block *nat_blk = page_address(nat_page);
2358 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2361 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2363 i = start_nid % NAT_ENTRY_PER_BLOCK;
2365 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2366 if (unlikely(start_nid >= nm_i->max_nid))
2369 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2371 if (blk_addr == NEW_ADDR)
2374 if (blk_addr == NULL_ADDR) {
2375 add_free_nid(sbi, start_nid, true, true);
2377 spin_lock(&NM_I(sbi)->nid_list_lock);
2378 update_free_nid_bitmap(sbi, start_nid, false, true);
2379 spin_unlock(&NM_I(sbi)->nid_list_lock);
2386 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2388 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2389 struct f2fs_journal *journal = curseg->journal;
2392 down_read(&curseg->journal_rwsem);
2393 for (i = 0; i < nats_in_cursum(journal); i++) {
2397 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2398 nid = le32_to_cpu(nid_in_journal(journal, i));
2399 if (addr == NULL_ADDR)
2400 add_free_nid(sbi, nid, true, false);
2402 remove_free_nid(sbi, nid);
2404 up_read(&curseg->journal_rwsem);
2407 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2409 struct f2fs_nm_info *nm_i = NM_I(sbi);
2410 unsigned int i, idx;
2413 down_read(&nm_i->nat_tree_lock);
2415 for (i = 0; i < nm_i->nat_blocks; i++) {
2416 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2418 if (!nm_i->free_nid_count[i])
2420 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2421 idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2422 NAT_ENTRY_PER_BLOCK, idx);
2423 if (idx >= NAT_ENTRY_PER_BLOCK)
2426 nid = i * NAT_ENTRY_PER_BLOCK + idx;
2427 add_free_nid(sbi, nid, true, false);
2429 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2434 scan_curseg_cache(sbi);
2436 up_read(&nm_i->nat_tree_lock);
2439 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2440 bool sync, bool mount)
2442 struct f2fs_nm_info *nm_i = NM_I(sbi);
2444 nid_t nid = nm_i->next_scan_nid;
2446 if (unlikely(nid >= nm_i->max_nid))
2449 if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2450 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2452 /* Enough entries */
2453 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2456 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2460 /* try to find free nids in free_nid_bitmap */
2461 scan_free_nid_bits(sbi);
2463 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2467 /* readahead nat pages to be scanned */
2468 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2471 down_read(&nm_i->nat_tree_lock);
2474 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2475 nm_i->nat_block_bitmap)) {
2476 struct page *page = get_current_nat_page(sbi, nid);
2479 ret = PTR_ERR(page);
2481 ret = scan_nat_page(sbi, page, nid);
2482 f2fs_put_page(page, 1);
2486 up_read(&nm_i->nat_tree_lock);
2487 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2492 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2493 if (unlikely(nid >= nm_i->max_nid))
2496 if (++i >= FREE_NID_PAGES)
2500 /* go to the next free nat pages to find free nids abundantly */
2501 nm_i->next_scan_nid = nid;
2503 /* find free nids from current sum_pages */
2504 scan_curseg_cache(sbi);
2506 up_read(&nm_i->nat_tree_lock);
2508 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2509 nm_i->ra_nid_pages, META_NAT, false);
2514 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2518 mutex_lock(&NM_I(sbi)->build_lock);
2519 ret = __f2fs_build_free_nids(sbi, sync, mount);
2520 mutex_unlock(&NM_I(sbi)->build_lock);
2526 * If this function returns success, caller can obtain a new nid
2527 * from second parameter of this function.
2528 * The returned nid could be used ino as well as nid when inode is created.
2530 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2532 struct f2fs_nm_info *nm_i = NM_I(sbi);
2533 struct free_nid *i = NULL;
2535 if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2536 f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2540 spin_lock(&nm_i->nid_list_lock);
2542 if (unlikely(nm_i->available_nids == 0)) {
2543 spin_unlock(&nm_i->nid_list_lock);
2547 /* We should not use stale free nids created by f2fs_build_free_nids */
2548 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2549 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2550 i = list_first_entry(&nm_i->free_nid_list,
2551 struct free_nid, list);
2554 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2555 nm_i->available_nids--;
2557 update_free_nid_bitmap(sbi, *nid, false, false);
2559 spin_unlock(&nm_i->nid_list_lock);
2562 spin_unlock(&nm_i->nid_list_lock);
2564 /* Let's scan nat pages and its caches to get free nids */
2565 if (!f2fs_build_free_nids(sbi, true, false))
2571 * f2fs_alloc_nid() should be called prior to this function.
2573 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2575 struct f2fs_nm_info *nm_i = NM_I(sbi);
2578 spin_lock(&nm_i->nid_list_lock);
2579 i = __lookup_free_nid_list(nm_i, nid);
2580 f2fs_bug_on(sbi, !i);
2581 __remove_free_nid(sbi, i, PREALLOC_NID);
2582 spin_unlock(&nm_i->nid_list_lock);
2584 kmem_cache_free(free_nid_slab, i);
2588 * f2fs_alloc_nid() should be called prior to this function.
2590 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2592 struct f2fs_nm_info *nm_i = NM_I(sbi);
2594 bool need_free = false;
2599 spin_lock(&nm_i->nid_list_lock);
2600 i = __lookup_free_nid_list(nm_i, nid);
2601 f2fs_bug_on(sbi, !i);
2603 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2604 __remove_free_nid(sbi, i, PREALLOC_NID);
2607 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2610 nm_i->available_nids++;
2612 update_free_nid_bitmap(sbi, nid, true, false);
2614 spin_unlock(&nm_i->nid_list_lock);
2617 kmem_cache_free(free_nid_slab, i);
2620 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2622 struct f2fs_nm_info *nm_i = NM_I(sbi);
2625 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2628 if (!mutex_trylock(&nm_i->build_lock))
2631 while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2632 struct free_nid *i, *next;
2633 unsigned int batch = SHRINK_NID_BATCH_SIZE;
2635 spin_lock(&nm_i->nid_list_lock);
2636 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2637 if (!nr_shrink || !batch ||
2638 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2640 __remove_free_nid(sbi, i, FREE_NID);
2641 kmem_cache_free(free_nid_slab, i);
2645 spin_unlock(&nm_i->nid_list_lock);
2648 mutex_unlock(&nm_i->build_lock);
2650 return nr - nr_shrink;
2653 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2655 void *src_addr, *dst_addr;
2658 struct f2fs_inode *ri;
2660 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2662 return PTR_ERR(ipage);
2664 ri = F2FS_INODE(page);
2665 if (ri->i_inline & F2FS_INLINE_XATTR) {
2666 if (!f2fs_has_inline_xattr(inode)) {
2667 set_inode_flag(inode, FI_INLINE_XATTR);
2668 stat_inc_inline_xattr(inode);
2671 if (f2fs_has_inline_xattr(inode)) {
2672 stat_dec_inline_xattr(inode);
2673 clear_inode_flag(inode, FI_INLINE_XATTR);
2678 dst_addr = inline_xattr_addr(inode, ipage);
2679 src_addr = inline_xattr_addr(inode, page);
2680 inline_size = inline_xattr_size(inode);
2682 f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2683 memcpy(dst_addr, src_addr, inline_size);
2685 f2fs_update_inode(inode, ipage);
2686 f2fs_put_page(ipage, 1);
2690 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2692 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2693 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2695 struct dnode_of_data dn;
2696 struct node_info ni;
2703 /* 1: invalidate the previous xattr nid */
2704 err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2708 f2fs_invalidate_blocks(sbi, ni.blk_addr);
2709 dec_valid_node_count(sbi, inode, false);
2710 set_node_addr(sbi, &ni, NULL_ADDR, false);
2713 /* 2: update xattr nid in inode */
2714 if (!f2fs_alloc_nid(sbi, &new_xnid))
2717 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2718 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2719 if (IS_ERR(xpage)) {
2720 f2fs_alloc_nid_failed(sbi, new_xnid);
2721 return PTR_ERR(xpage);
2724 f2fs_alloc_nid_done(sbi, new_xnid);
2725 f2fs_update_inode_page(inode);
2727 /* 3: update and set xattr node page dirty */
2728 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2730 set_page_dirty(xpage);
2731 f2fs_put_page(xpage, 1);
2736 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2738 struct f2fs_inode *src, *dst;
2739 nid_t ino = ino_of_node(page);
2740 struct node_info old_ni, new_ni;
2744 err = f2fs_get_node_info(sbi, ino, &old_ni);
2748 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2751 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2753 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2757 /* Should not use this inode from free nid list */
2758 remove_free_nid(sbi, ino);
2760 if (!PageUptodate(ipage))
2761 SetPageUptodate(ipage);
2762 fill_node_footer(ipage, ino, ino, 0, true);
2763 set_cold_node(ipage, false);
2765 src = F2FS_INODE(page);
2766 dst = F2FS_INODE(ipage);
2768 memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2770 dst->i_blocks = cpu_to_le64(1);
2771 dst->i_links = cpu_to_le32(1);
2772 dst->i_xattr_nid = 0;
2773 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2774 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2775 dst->i_extra_isize = src->i_extra_isize;
2777 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2778 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2779 i_inline_xattr_size))
2780 dst->i_inline_xattr_size = src->i_inline_xattr_size;
2782 if (f2fs_sb_has_project_quota(sbi) &&
2783 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2785 dst->i_projid = src->i_projid;
2787 if (f2fs_sb_has_inode_crtime(sbi) &&
2788 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2790 dst->i_crtime = src->i_crtime;
2791 dst->i_crtime_nsec = src->i_crtime_nsec;
2798 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2800 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2801 inc_valid_inode_count(sbi);
2802 set_page_dirty(ipage);
2803 f2fs_put_page(ipage, 1);
2807 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2808 unsigned int segno, struct f2fs_summary_block *sum)
2810 struct f2fs_node *rn;
2811 struct f2fs_summary *sum_entry;
2813 int i, idx, last_offset, nrpages;
2815 /* scan the node segment */
2816 last_offset = sbi->blocks_per_seg;
2817 addr = START_BLOCK(sbi, segno);
2818 sum_entry = &sum->entries[0];
2820 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2821 nrpages = bio_max_segs(last_offset - i);
2823 /* readahead node pages */
2824 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2826 for (idx = addr; idx < addr + nrpages; idx++) {
2827 struct page *page = f2fs_get_tmp_page(sbi, idx);
2830 return PTR_ERR(page);
2832 rn = F2FS_NODE(page);
2833 sum_entry->nid = rn->footer.nid;
2834 sum_entry->version = 0;
2835 sum_entry->ofs_in_node = 0;
2837 f2fs_put_page(page, 1);
2840 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2846 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2848 struct f2fs_nm_info *nm_i = NM_I(sbi);
2849 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2850 struct f2fs_journal *journal = curseg->journal;
2853 down_write(&curseg->journal_rwsem);
2854 for (i = 0; i < nats_in_cursum(journal); i++) {
2855 struct nat_entry *ne;
2856 struct f2fs_nat_entry raw_ne;
2857 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2859 if (f2fs_check_nid_range(sbi, nid))
2862 raw_ne = nat_in_journal(journal, i);
2864 ne = __lookup_nat_cache(nm_i, nid);
2866 ne = __alloc_nat_entry(sbi, nid, true);
2867 __init_nat_entry(nm_i, ne, &raw_ne, true);
2871 * if a free nat in journal has not been used after last
2872 * checkpoint, we should remove it from available nids,
2873 * since later we will add it again.
2875 if (!get_nat_flag(ne, IS_DIRTY) &&
2876 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2877 spin_lock(&nm_i->nid_list_lock);
2878 nm_i->available_nids--;
2879 spin_unlock(&nm_i->nid_list_lock);
2882 __set_nat_cache_dirty(nm_i, ne);
2884 update_nats_in_cursum(journal, -i);
2885 up_write(&curseg->journal_rwsem);
2888 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2889 struct list_head *head, int max)
2891 struct nat_entry_set *cur;
2893 if (nes->entry_cnt >= max)
2896 list_for_each_entry(cur, head, set_list) {
2897 if (cur->entry_cnt >= nes->entry_cnt) {
2898 list_add(&nes->set_list, cur->set_list.prev);
2903 list_add_tail(&nes->set_list, head);
2906 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs,
2910 __set_bit_le(nat_ofs, nm_i->empty_nat_bits);
2911 __clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2915 __clear_bit_le(nat_ofs, nm_i->empty_nat_bits);
2916 if (valid == NAT_ENTRY_PER_BLOCK)
2917 __set_bit_le(nat_ofs, nm_i->full_nat_bits);
2919 __clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2922 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2925 struct f2fs_nm_info *nm_i = NM_I(sbi);
2926 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2927 struct f2fs_nat_block *nat_blk = page_address(page);
2931 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
2934 if (nat_index == 0) {
2938 for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2939 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2943 __update_nat_bits(nm_i, nat_index, valid);
2946 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi)
2948 struct f2fs_nm_info *nm_i = NM_I(sbi);
2949 unsigned int nat_ofs;
2951 down_read(&nm_i->nat_tree_lock);
2953 for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) {
2954 unsigned int valid = 0, nid_ofs = 0;
2956 /* handle nid zero due to it should never be used */
2957 if (unlikely(nat_ofs == 0)) {
2962 for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) {
2963 if (!test_bit_le(nid_ofs,
2964 nm_i->free_nid_bitmap[nat_ofs]))
2968 __update_nat_bits(nm_i, nat_ofs, valid);
2971 up_read(&nm_i->nat_tree_lock);
2974 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2975 struct nat_entry_set *set, struct cp_control *cpc)
2977 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2978 struct f2fs_journal *journal = curseg->journal;
2979 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2980 bool to_journal = true;
2981 struct f2fs_nat_block *nat_blk;
2982 struct nat_entry *ne, *cur;
2983 struct page *page = NULL;
2986 * there are two steps to flush nat entries:
2987 * #1, flush nat entries to journal in current hot data summary block.
2988 * #2, flush nat entries to nat page.
2990 if ((cpc->reason & CP_UMOUNT) ||
2991 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2995 down_write(&curseg->journal_rwsem);
2997 page = get_next_nat_page(sbi, start_nid);
2999 return PTR_ERR(page);
3001 nat_blk = page_address(page);
3002 f2fs_bug_on(sbi, !nat_blk);
3005 /* flush dirty nats in nat entry set */
3006 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
3007 struct f2fs_nat_entry *raw_ne;
3008 nid_t nid = nat_get_nid(ne);
3011 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
3014 offset = f2fs_lookup_journal_in_cursum(journal,
3015 NAT_JOURNAL, nid, 1);
3016 f2fs_bug_on(sbi, offset < 0);
3017 raw_ne = &nat_in_journal(journal, offset);
3018 nid_in_journal(journal, offset) = cpu_to_le32(nid);
3020 raw_ne = &nat_blk->entries[nid - start_nid];
3022 raw_nat_from_node_info(raw_ne, &ne->ni);
3024 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
3025 if (nat_get_blkaddr(ne) == NULL_ADDR) {
3026 add_free_nid(sbi, nid, false, true);
3028 spin_lock(&NM_I(sbi)->nid_list_lock);
3029 update_free_nid_bitmap(sbi, nid, false, false);
3030 spin_unlock(&NM_I(sbi)->nid_list_lock);
3035 up_write(&curseg->journal_rwsem);
3037 update_nat_bits(sbi, start_nid, page);
3038 f2fs_put_page(page, 1);
3041 /* Allow dirty nats by node block allocation in write_begin */
3042 if (!set->entry_cnt) {
3043 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
3044 kmem_cache_free(nat_entry_set_slab, set);
3050 * This function is called during the checkpointing process.
3052 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3054 struct f2fs_nm_info *nm_i = NM_I(sbi);
3055 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3056 struct f2fs_journal *journal = curseg->journal;
3057 struct nat_entry_set *setvec[SETVEC_SIZE];
3058 struct nat_entry_set *set, *tmp;
3065 * during unmount, let's flush nat_bits before checking
3066 * nat_cnt[DIRTY_NAT].
3068 if (cpc->reason & CP_UMOUNT) {
3069 down_write(&nm_i->nat_tree_lock);
3070 remove_nats_in_journal(sbi);
3071 up_write(&nm_i->nat_tree_lock);
3074 if (!nm_i->nat_cnt[DIRTY_NAT])
3077 down_write(&nm_i->nat_tree_lock);
3080 * if there are no enough space in journal to store dirty nat
3081 * entries, remove all entries from journal and merge them
3082 * into nat entry set.
3084 if (cpc->reason & CP_UMOUNT ||
3085 !__has_cursum_space(journal,
3086 nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3087 remove_nats_in_journal(sbi);
3089 while ((found = __gang_lookup_nat_set(nm_i,
3090 set_idx, SETVEC_SIZE, setvec))) {
3093 set_idx = setvec[found - 1]->set + 1;
3094 for (idx = 0; idx < found; idx++)
3095 __adjust_nat_entry_set(setvec[idx], &sets,
3096 MAX_NAT_JENTRIES(journal));
3099 /* flush dirty nats in nat entry set */
3100 list_for_each_entry_safe(set, tmp, &sets, set_list) {
3101 err = __flush_nat_entry_set(sbi, set, cpc);
3106 up_write(&nm_i->nat_tree_lock);
3107 /* Allow dirty nats by node block allocation in write_begin */
3112 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3114 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3115 struct f2fs_nm_info *nm_i = NM_I(sbi);
3116 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3118 __u64 cp_ver = cur_cp_version(ckpt);
3119 block_t nat_bits_addr;
3121 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3122 nm_i->nat_bits = f2fs_kvzalloc(sbi,
3123 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3124 if (!nm_i->nat_bits)
3127 nm_i->full_nat_bits = nm_i->nat_bits + 8;
3128 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3130 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3133 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3134 nm_i->nat_bits_blocks;
3135 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3138 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3140 return PTR_ERR(page);
3142 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3143 page_address(page), F2FS_BLKSIZE);
3144 f2fs_put_page(page, 1);
3147 cp_ver |= (cur_cp_crc(ckpt) << 32);
3148 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3149 clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
3150 f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)",
3151 cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits));
3155 f2fs_notice(sbi, "Found nat_bits in checkpoint");
3159 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3161 struct f2fs_nm_info *nm_i = NM_I(sbi);
3163 nid_t nid, last_nid;
3165 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3168 for (i = 0; i < nm_i->nat_blocks; i++) {
3169 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3170 if (i >= nm_i->nat_blocks)
3173 __set_bit_le(i, nm_i->nat_block_bitmap);
3175 nid = i * NAT_ENTRY_PER_BLOCK;
3176 last_nid = nid + NAT_ENTRY_PER_BLOCK;
3178 spin_lock(&NM_I(sbi)->nid_list_lock);
3179 for (; nid < last_nid; nid++)
3180 update_free_nid_bitmap(sbi, nid, true, true);
3181 spin_unlock(&NM_I(sbi)->nid_list_lock);
3184 for (i = 0; i < nm_i->nat_blocks; i++) {
3185 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3186 if (i >= nm_i->nat_blocks)
3189 __set_bit_le(i, nm_i->nat_block_bitmap);
3193 static int init_node_manager(struct f2fs_sb_info *sbi)
3195 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3196 struct f2fs_nm_info *nm_i = NM_I(sbi);
3197 unsigned char *version_bitmap;
3198 unsigned int nat_segs;
3201 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3203 /* segment_count_nat includes pair segment so divide to 2. */
3204 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3205 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3206 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3208 /* not used nids: 0, node, meta, (and root counted as valid node) */
3209 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3210 F2FS_RESERVED_NODE_NUM;
3211 nm_i->nid_cnt[FREE_NID] = 0;
3212 nm_i->nid_cnt[PREALLOC_NID] = 0;
3213 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3214 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3215 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3217 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3218 INIT_LIST_HEAD(&nm_i->free_nid_list);
3219 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3220 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3221 INIT_LIST_HEAD(&nm_i->nat_entries);
3222 spin_lock_init(&nm_i->nat_list_lock);
3224 mutex_init(&nm_i->build_lock);
3225 spin_lock_init(&nm_i->nid_list_lock);
3226 init_rwsem(&nm_i->nat_tree_lock);
3228 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3229 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3230 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3231 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3233 if (!nm_i->nat_bitmap)
3236 err = __get_nat_bitmaps(sbi);
3240 #ifdef CONFIG_F2FS_CHECK_FS
3241 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3243 if (!nm_i->nat_bitmap_mir)
3250 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3252 struct f2fs_nm_info *nm_i = NM_I(sbi);
3255 nm_i->free_nid_bitmap =
3256 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3259 if (!nm_i->free_nid_bitmap)
3262 for (i = 0; i < nm_i->nat_blocks; i++) {
3263 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3264 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3265 if (!nm_i->free_nid_bitmap[i])
3269 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3271 if (!nm_i->nat_block_bitmap)
3274 nm_i->free_nid_count =
3275 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3278 if (!nm_i->free_nid_count)
3283 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3287 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3292 err = init_node_manager(sbi);
3296 err = init_free_nid_cache(sbi);
3300 /* load free nid status from nat_bits table */
3301 load_free_nid_bitmap(sbi);
3303 return f2fs_build_free_nids(sbi, true, true);
3306 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3308 struct f2fs_nm_info *nm_i = NM_I(sbi);
3309 struct free_nid *i, *next_i;
3310 struct nat_entry *natvec[NATVEC_SIZE];
3311 struct nat_entry_set *setvec[SETVEC_SIZE];
3318 /* destroy free nid list */
3319 spin_lock(&nm_i->nid_list_lock);
3320 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3321 __remove_free_nid(sbi, i, FREE_NID);
3322 spin_unlock(&nm_i->nid_list_lock);
3323 kmem_cache_free(free_nid_slab, i);
3324 spin_lock(&nm_i->nid_list_lock);
3326 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3327 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3328 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3329 spin_unlock(&nm_i->nid_list_lock);
3331 /* destroy nat cache */
3332 down_write(&nm_i->nat_tree_lock);
3333 while ((found = __gang_lookup_nat_cache(nm_i,
3334 nid, NATVEC_SIZE, natvec))) {
3337 nid = nat_get_nid(natvec[found - 1]) + 1;
3338 for (idx = 0; idx < found; idx++) {
3339 spin_lock(&nm_i->nat_list_lock);
3340 list_del(&natvec[idx]->list);
3341 spin_unlock(&nm_i->nat_list_lock);
3343 __del_from_nat_cache(nm_i, natvec[idx]);
3346 f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3348 /* destroy nat set cache */
3350 while ((found = __gang_lookup_nat_set(nm_i,
3351 nid, SETVEC_SIZE, setvec))) {
3354 nid = setvec[found - 1]->set + 1;
3355 for (idx = 0; idx < found; idx++) {
3356 /* entry_cnt is not zero, when cp_error was occurred */
3357 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3358 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3359 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3362 up_write(&nm_i->nat_tree_lock);
3364 kvfree(nm_i->nat_block_bitmap);
3365 if (nm_i->free_nid_bitmap) {
3368 for (i = 0; i < nm_i->nat_blocks; i++)
3369 kvfree(nm_i->free_nid_bitmap[i]);
3370 kvfree(nm_i->free_nid_bitmap);
3372 kvfree(nm_i->free_nid_count);
3374 kvfree(nm_i->nat_bitmap);
3375 kvfree(nm_i->nat_bits);
3376 #ifdef CONFIG_F2FS_CHECK_FS
3377 kvfree(nm_i->nat_bitmap_mir);
3379 sbi->nm_info = NULL;
3383 int __init f2fs_create_node_manager_caches(void)
3385 nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3386 sizeof(struct nat_entry));
3387 if (!nat_entry_slab)
3390 free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3391 sizeof(struct free_nid));
3393 goto destroy_nat_entry;
3395 nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3396 sizeof(struct nat_entry_set));
3397 if (!nat_entry_set_slab)
3398 goto destroy_free_nid;
3400 fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3401 sizeof(struct fsync_node_entry));
3402 if (!fsync_node_entry_slab)
3403 goto destroy_nat_entry_set;
3406 destroy_nat_entry_set:
3407 kmem_cache_destroy(nat_entry_set_slab);
3409 kmem_cache_destroy(free_nid_slab);
3411 kmem_cache_destroy(nat_entry_slab);
3416 void f2fs_destroy_node_manager_caches(void)
3418 kmem_cache_destroy(fsync_node_entry_slab);
3419 kmem_cache_destroy(nat_entry_set_slab);
3420 kmem_cache_destroy(free_nid_slab);
3421 kmem_cache_destroy(nat_entry_slab);