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);
48 unsigned long avail_ram;
49 unsigned long mem_size = 0;
54 /* only uses low memory */
55 avail_ram = val.totalram - val.totalhigh;
58 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
60 if (type == FREE_NIDS) {
61 mem_size = (nm_i->nid_cnt[FREE_NID] *
62 sizeof(struct free_nid)) >> PAGE_SHIFT;
63 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
64 } else if (type == NAT_ENTRIES) {
65 mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
66 sizeof(struct nat_entry)) >> PAGE_SHIFT;
67 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
68 if (excess_cached_nats(sbi))
70 } else if (type == DIRTY_DENTS) {
71 if (sbi->sb->s_bdi->wb.dirty_exceeded)
73 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
74 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
75 } else if (type == INO_ENTRIES) {
78 for (i = 0; i < MAX_INO_ENTRY; i++)
79 mem_size += sbi->im[i].ino_num *
80 sizeof(struct ino_entry);
81 mem_size >>= PAGE_SHIFT;
82 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
83 } else if (type == EXTENT_CACHE) {
84 mem_size = (atomic_read(&sbi->total_ext_tree) *
85 sizeof(struct extent_tree) +
86 atomic_read(&sbi->total_ext_node) *
87 sizeof(struct extent_node)) >> PAGE_SHIFT;
88 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
89 } else if (type == INMEM_PAGES) {
90 /* it allows 20% / total_ram for inmemory pages */
91 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
92 res = mem_size < (val.totalram / 5);
94 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
100 static void clear_node_page_dirty(struct page *page)
102 if (PageDirty(page)) {
103 f2fs_clear_page_cache_dirty_tag(page);
104 clear_page_dirty_for_io(page);
105 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
107 ClearPageUptodate(page);
110 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
112 return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
115 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
117 struct page *src_page;
118 struct page *dst_page;
122 struct f2fs_nm_info *nm_i = NM_I(sbi);
124 dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
126 /* get current nat block page with lock */
127 src_page = get_current_nat_page(sbi, nid);
128 if (IS_ERR(src_page))
130 dst_page = f2fs_grab_meta_page(sbi, dst_off);
131 f2fs_bug_on(sbi, PageDirty(src_page));
133 src_addr = page_address(src_page);
134 dst_addr = page_address(dst_page);
135 memcpy(dst_addr, src_addr, PAGE_SIZE);
136 set_page_dirty(dst_page);
137 f2fs_put_page(src_page, 1);
139 set_to_next_nat(nm_i, nid);
144 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
146 struct nat_entry *new;
149 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
151 new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
153 nat_set_nid(new, nid);
159 static void __free_nat_entry(struct nat_entry *e)
161 kmem_cache_free(nat_entry_slab, e);
164 /* must be locked by nat_tree_lock */
165 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
166 struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
169 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
170 else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
174 node_info_from_raw_nat(&ne->ni, raw_ne);
176 spin_lock(&nm_i->nat_list_lock);
177 list_add_tail(&ne->list, &nm_i->nat_entries);
178 spin_unlock(&nm_i->nat_list_lock);
180 nm_i->nat_cnt[TOTAL_NAT]++;
181 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
185 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
187 struct nat_entry *ne;
189 ne = radix_tree_lookup(&nm_i->nat_root, n);
191 /* for recent accessed nat entry, move it to tail of lru list */
192 if (ne && !get_nat_flag(ne, IS_DIRTY)) {
193 spin_lock(&nm_i->nat_list_lock);
194 if (!list_empty(&ne->list))
195 list_move_tail(&ne->list, &nm_i->nat_entries);
196 spin_unlock(&nm_i->nat_list_lock);
202 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
203 nid_t start, unsigned int nr, struct nat_entry **ep)
205 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
208 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
210 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
211 nm_i->nat_cnt[TOTAL_NAT]--;
212 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
216 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
217 struct nat_entry *ne)
219 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
220 struct nat_entry_set *head;
222 head = radix_tree_lookup(&nm_i->nat_set_root, set);
224 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
226 INIT_LIST_HEAD(&head->entry_list);
227 INIT_LIST_HEAD(&head->set_list);
230 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
235 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
236 struct nat_entry *ne)
238 struct nat_entry_set *head;
239 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
242 head = __grab_nat_entry_set(nm_i, ne);
245 * update entry_cnt in below condition:
246 * 1. update NEW_ADDR to valid block address;
247 * 2. update old block address to new one;
249 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
250 !get_nat_flag(ne, IS_DIRTY)))
253 set_nat_flag(ne, IS_PREALLOC, new_ne);
255 if (get_nat_flag(ne, IS_DIRTY))
258 nm_i->nat_cnt[DIRTY_NAT]++;
259 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
260 set_nat_flag(ne, IS_DIRTY, true);
262 spin_lock(&nm_i->nat_list_lock);
264 list_del_init(&ne->list);
266 list_move_tail(&ne->list, &head->entry_list);
267 spin_unlock(&nm_i->nat_list_lock);
270 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
271 struct nat_entry_set *set, struct nat_entry *ne)
273 spin_lock(&nm_i->nat_list_lock);
274 list_move_tail(&ne->list, &nm_i->nat_entries);
275 spin_unlock(&nm_i->nat_list_lock);
277 set_nat_flag(ne, IS_DIRTY, false);
279 nm_i->nat_cnt[DIRTY_NAT]--;
280 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
283 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
284 nid_t start, unsigned int nr, struct nat_entry_set **ep)
286 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
290 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
292 return NODE_MAPPING(sbi) == page->mapping &&
293 IS_DNODE(page) && is_cold_node(page);
296 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
298 spin_lock_init(&sbi->fsync_node_lock);
299 INIT_LIST_HEAD(&sbi->fsync_node_list);
300 sbi->fsync_seg_id = 0;
301 sbi->fsync_node_num = 0;
304 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
307 struct fsync_node_entry *fn;
311 fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS);
315 INIT_LIST_HEAD(&fn->list);
317 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
318 list_add_tail(&fn->list, &sbi->fsync_node_list);
319 fn->seq_id = sbi->fsync_seg_id++;
321 sbi->fsync_node_num++;
322 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
327 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
329 struct fsync_node_entry *fn;
332 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
333 list_for_each_entry(fn, &sbi->fsync_node_list, list) {
334 if (fn->page == page) {
336 sbi->fsync_node_num--;
337 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
338 kmem_cache_free(fsync_node_entry_slab, fn);
343 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
347 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
351 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
352 sbi->fsync_seg_id = 0;
353 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
356 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
358 struct f2fs_nm_info *nm_i = NM_I(sbi);
362 down_read(&nm_i->nat_tree_lock);
363 e = __lookup_nat_cache(nm_i, nid);
365 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
366 !get_nat_flag(e, HAS_FSYNCED_INODE))
369 up_read(&nm_i->nat_tree_lock);
373 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
375 struct f2fs_nm_info *nm_i = NM_I(sbi);
379 down_read(&nm_i->nat_tree_lock);
380 e = __lookup_nat_cache(nm_i, nid);
381 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
383 up_read(&nm_i->nat_tree_lock);
387 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
389 struct f2fs_nm_info *nm_i = NM_I(sbi);
391 bool need_update = true;
393 down_read(&nm_i->nat_tree_lock);
394 e = __lookup_nat_cache(nm_i, ino);
395 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
396 (get_nat_flag(e, IS_CHECKPOINTED) ||
397 get_nat_flag(e, HAS_FSYNCED_INODE)))
399 up_read(&nm_i->nat_tree_lock);
403 /* must be locked by nat_tree_lock */
404 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
405 struct f2fs_nat_entry *ne)
407 struct f2fs_nm_info *nm_i = NM_I(sbi);
408 struct nat_entry *new, *e;
410 new = __alloc_nat_entry(nid, false);
414 down_write(&nm_i->nat_tree_lock);
415 e = __lookup_nat_cache(nm_i, nid);
417 e = __init_nat_entry(nm_i, new, ne, false);
419 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
420 nat_get_blkaddr(e) !=
421 le32_to_cpu(ne->block_addr) ||
422 nat_get_version(e) != ne->version);
423 up_write(&nm_i->nat_tree_lock);
425 __free_nat_entry(new);
428 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
429 block_t new_blkaddr, bool fsync_done)
431 struct f2fs_nm_info *nm_i = NM_I(sbi);
433 struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
435 down_write(&nm_i->nat_tree_lock);
436 e = __lookup_nat_cache(nm_i, ni->nid);
438 e = __init_nat_entry(nm_i, new, NULL, true);
439 copy_node_info(&e->ni, ni);
440 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
441 } else if (new_blkaddr == NEW_ADDR) {
443 * when nid is reallocated,
444 * previous nat entry can be remained in nat cache.
445 * So, reinitialize it with new information.
447 copy_node_info(&e->ni, ni);
448 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
450 /* let's free early to reduce memory consumption */
452 __free_nat_entry(new);
455 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
456 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
457 new_blkaddr == NULL_ADDR);
458 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
459 new_blkaddr == NEW_ADDR);
460 f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
461 new_blkaddr == NEW_ADDR);
463 /* increment version no as node is removed */
464 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
465 unsigned char version = nat_get_version(e);
466 nat_set_version(e, inc_node_version(version));
470 nat_set_blkaddr(e, new_blkaddr);
471 if (!__is_valid_data_blkaddr(new_blkaddr))
472 set_nat_flag(e, IS_CHECKPOINTED, false);
473 __set_nat_cache_dirty(nm_i, e);
475 /* update fsync_mark if its inode nat entry is still alive */
476 if (ni->nid != ni->ino)
477 e = __lookup_nat_cache(nm_i, ni->ino);
479 if (fsync_done && ni->nid == ni->ino)
480 set_nat_flag(e, HAS_FSYNCED_INODE, true);
481 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
483 up_write(&nm_i->nat_tree_lock);
486 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
488 struct f2fs_nm_info *nm_i = NM_I(sbi);
491 if (!down_write_trylock(&nm_i->nat_tree_lock))
494 spin_lock(&nm_i->nat_list_lock);
496 struct nat_entry *ne;
498 if (list_empty(&nm_i->nat_entries))
501 ne = list_first_entry(&nm_i->nat_entries,
502 struct nat_entry, list);
504 spin_unlock(&nm_i->nat_list_lock);
506 __del_from_nat_cache(nm_i, ne);
509 spin_lock(&nm_i->nat_list_lock);
511 spin_unlock(&nm_i->nat_list_lock);
513 up_write(&nm_i->nat_tree_lock);
514 return nr - nr_shrink;
517 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
518 struct node_info *ni)
520 struct f2fs_nm_info *nm_i = NM_I(sbi);
521 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
522 struct f2fs_journal *journal = curseg->journal;
523 nid_t start_nid = START_NID(nid);
524 struct f2fs_nat_block *nat_blk;
525 struct page *page = NULL;
526 struct f2fs_nat_entry ne;
534 /* Check nat cache */
535 down_read(&nm_i->nat_tree_lock);
536 e = __lookup_nat_cache(nm_i, nid);
538 ni->ino = nat_get_ino(e);
539 ni->blk_addr = nat_get_blkaddr(e);
540 ni->version = nat_get_version(e);
541 up_read(&nm_i->nat_tree_lock);
545 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
547 /* Check current segment summary */
548 down_read(&curseg->journal_rwsem);
549 i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
551 ne = nat_in_journal(journal, i);
552 node_info_from_raw_nat(ni, &ne);
554 up_read(&curseg->journal_rwsem);
556 up_read(&nm_i->nat_tree_lock);
560 /* Fill node_info from nat page */
561 index = current_nat_addr(sbi, nid);
562 up_read(&nm_i->nat_tree_lock);
564 page = f2fs_get_meta_page(sbi, index);
566 return PTR_ERR(page);
568 nat_blk = (struct f2fs_nat_block *)page_address(page);
569 ne = nat_blk->entries[nid - start_nid];
570 node_info_from_raw_nat(ni, &ne);
571 f2fs_put_page(page, 1);
573 blkaddr = le32_to_cpu(ne.block_addr);
574 if (__is_valid_data_blkaddr(blkaddr) &&
575 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
578 /* cache nat entry */
579 cache_nat_entry(sbi, nid, &ne);
584 * readahead MAX_RA_NODE number of node pages.
586 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
588 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
589 struct blk_plug plug;
593 blk_start_plug(&plug);
595 /* Then, try readahead for siblings of the desired node */
597 end = min(end, NIDS_PER_BLOCK);
598 for (i = start; i < end; i++) {
599 nid = get_nid(parent, i, false);
600 f2fs_ra_node_page(sbi, nid);
603 blk_finish_plug(&plug);
606 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
608 const long direct_index = ADDRS_PER_INODE(dn->inode);
609 const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
610 const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
611 unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
612 int cur_level = dn->cur_level;
613 int max_level = dn->max_level;
619 while (max_level-- > cur_level)
620 skipped_unit *= NIDS_PER_BLOCK;
622 switch (dn->max_level) {
624 base += 2 * indirect_blks;
627 base += 2 * direct_blks;
630 base += direct_index;
633 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
636 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
640 * The maximum depth is four.
641 * Offset[0] will have raw inode offset.
643 static int get_node_path(struct inode *inode, long block,
644 int offset[4], unsigned int noffset[4])
646 const long direct_index = ADDRS_PER_INODE(inode);
647 const long direct_blks = ADDRS_PER_BLOCK(inode);
648 const long dptrs_per_blk = NIDS_PER_BLOCK;
649 const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
650 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
656 if (block < direct_index) {
660 block -= direct_index;
661 if (block < direct_blks) {
662 offset[n++] = NODE_DIR1_BLOCK;
668 block -= direct_blks;
669 if (block < direct_blks) {
670 offset[n++] = NODE_DIR2_BLOCK;
676 block -= direct_blks;
677 if (block < indirect_blks) {
678 offset[n++] = NODE_IND1_BLOCK;
680 offset[n++] = block / direct_blks;
681 noffset[n] = 4 + offset[n - 1];
682 offset[n] = block % direct_blks;
686 block -= indirect_blks;
687 if (block < indirect_blks) {
688 offset[n++] = NODE_IND2_BLOCK;
689 noffset[n] = 4 + dptrs_per_blk;
690 offset[n++] = block / direct_blks;
691 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
692 offset[n] = block % direct_blks;
696 block -= indirect_blks;
697 if (block < dindirect_blks) {
698 offset[n++] = NODE_DIND_BLOCK;
699 noffset[n] = 5 + (dptrs_per_blk * 2);
700 offset[n++] = block / indirect_blks;
701 noffset[n] = 6 + (dptrs_per_blk * 2) +
702 offset[n - 1] * (dptrs_per_blk + 1);
703 offset[n++] = (block / direct_blks) % dptrs_per_blk;
704 noffset[n] = 7 + (dptrs_per_blk * 2) +
705 offset[n - 2] * (dptrs_per_blk + 1) +
707 offset[n] = block % direct_blks;
718 * Caller should call f2fs_put_dnode(dn).
719 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
720 * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
722 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
724 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
725 struct page *npage[4];
726 struct page *parent = NULL;
728 unsigned int noffset[4];
733 level = get_node_path(dn->inode, index, offset, noffset);
737 nids[0] = dn->inode->i_ino;
738 npage[0] = dn->inode_page;
741 npage[0] = f2fs_get_node_page(sbi, nids[0]);
742 if (IS_ERR(npage[0]))
743 return PTR_ERR(npage[0]);
746 /* if inline_data is set, should not report any block indices */
747 if (f2fs_has_inline_data(dn->inode) && index) {
749 f2fs_put_page(npage[0], 1);
755 nids[1] = get_nid(parent, offset[0], true);
756 dn->inode_page = npage[0];
757 dn->inode_page_locked = true;
759 /* get indirect or direct nodes */
760 for (i = 1; i <= level; i++) {
763 if (!nids[i] && mode == ALLOC_NODE) {
765 if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
771 npage[i] = f2fs_new_node_page(dn, noffset[i]);
772 if (IS_ERR(npage[i])) {
773 f2fs_alloc_nid_failed(sbi, nids[i]);
774 err = PTR_ERR(npage[i]);
778 set_nid(parent, offset[i - 1], nids[i], i == 1);
779 f2fs_alloc_nid_done(sbi, nids[i]);
781 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
782 npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
783 if (IS_ERR(npage[i])) {
784 err = PTR_ERR(npage[i]);
790 dn->inode_page_locked = false;
793 f2fs_put_page(parent, 1);
797 npage[i] = f2fs_get_node_page(sbi, nids[i]);
798 if (IS_ERR(npage[i])) {
799 err = PTR_ERR(npage[i]);
800 f2fs_put_page(npage[0], 0);
806 nids[i + 1] = get_nid(parent, offset[i], false);
809 dn->nid = nids[level];
810 dn->ofs_in_node = offset[level];
811 dn->node_page = npage[level];
812 dn->data_blkaddr = f2fs_data_blkaddr(dn);
816 f2fs_put_page(parent, 1);
818 f2fs_put_page(npage[0], 0);
820 dn->inode_page = NULL;
821 dn->node_page = NULL;
822 if (err == -ENOENT) {
824 dn->max_level = level;
825 dn->ofs_in_node = offset[level];
830 static int truncate_node(struct dnode_of_data *dn)
832 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
837 err = f2fs_get_node_info(sbi, dn->nid, &ni);
841 /* Deallocate node address */
842 f2fs_invalidate_blocks(sbi, ni.blk_addr);
843 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
844 set_node_addr(sbi, &ni, NULL_ADDR, false);
846 if (dn->nid == dn->inode->i_ino) {
847 f2fs_remove_orphan_inode(sbi, dn->nid);
848 dec_valid_inode_count(sbi);
849 f2fs_inode_synced(dn->inode);
852 clear_node_page_dirty(dn->node_page);
853 set_sbi_flag(sbi, SBI_IS_DIRTY);
855 index = dn->node_page->index;
856 f2fs_put_page(dn->node_page, 1);
858 invalidate_mapping_pages(NODE_MAPPING(sbi),
861 dn->node_page = NULL;
862 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
867 static int truncate_dnode(struct dnode_of_data *dn)
875 /* get direct node */
876 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
877 if (PTR_ERR(page) == -ENOENT)
879 else if (IS_ERR(page))
880 return PTR_ERR(page);
882 /* Make dnode_of_data for parameter */
883 dn->node_page = page;
885 f2fs_truncate_data_blocks(dn);
886 err = truncate_node(dn);
888 f2fs_put_page(page, 1);
895 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
898 struct dnode_of_data rdn = *dn;
900 struct f2fs_node *rn;
902 unsigned int child_nofs;
907 return NIDS_PER_BLOCK + 1;
909 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
911 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
913 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
914 return PTR_ERR(page);
917 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
919 rn = F2FS_NODE(page);
921 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
922 child_nid = le32_to_cpu(rn->in.nid[i]);
926 ret = truncate_dnode(&rdn);
929 if (set_nid(page, i, 0, false))
930 dn->node_changed = true;
933 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
934 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
935 child_nid = le32_to_cpu(rn->in.nid[i]);
936 if (child_nid == 0) {
937 child_nofs += NIDS_PER_BLOCK + 1;
941 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
942 if (ret == (NIDS_PER_BLOCK + 1)) {
943 if (set_nid(page, i, 0, false))
944 dn->node_changed = true;
946 } else if (ret < 0 && ret != -ENOENT) {
954 /* remove current indirect node */
955 dn->node_page = page;
956 ret = truncate_node(dn);
961 f2fs_put_page(page, 1);
963 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
967 f2fs_put_page(page, 1);
968 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
972 static int truncate_partial_nodes(struct dnode_of_data *dn,
973 struct f2fs_inode *ri, int *offset, int depth)
975 struct page *pages[2];
982 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
986 /* get indirect nodes in the path */
987 for (i = 0; i < idx + 1; i++) {
988 /* reference count'll be increased */
989 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
990 if (IS_ERR(pages[i])) {
991 err = PTR_ERR(pages[i]);
995 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
998 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1000 /* free direct nodes linked to a partial indirect node */
1001 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1002 child_nid = get_nid(pages[idx], i, false);
1005 dn->nid = child_nid;
1006 err = truncate_dnode(dn);
1009 if (set_nid(pages[idx], i, 0, false))
1010 dn->node_changed = true;
1013 if (offset[idx + 1] == 0) {
1014 dn->node_page = pages[idx];
1016 err = truncate_node(dn);
1020 f2fs_put_page(pages[idx], 1);
1023 offset[idx + 1] = 0;
1026 for (i = idx; i >= 0; i--)
1027 f2fs_put_page(pages[i], 1);
1029 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1035 * All the block addresses of data and nodes should be nullified.
1037 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1039 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1040 int err = 0, cont = 1;
1041 int level, offset[4], noffset[4];
1042 unsigned int nofs = 0;
1043 struct f2fs_inode *ri;
1044 struct dnode_of_data dn;
1047 trace_f2fs_truncate_inode_blocks_enter(inode, from);
1049 level = get_node_path(inode, from, offset, noffset);
1051 trace_f2fs_truncate_inode_blocks_exit(inode, level);
1055 page = f2fs_get_node_page(sbi, inode->i_ino);
1057 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1058 return PTR_ERR(page);
1061 set_new_dnode(&dn, inode, page, NULL, 0);
1064 ri = F2FS_INODE(page);
1072 if (!offset[level - 1])
1074 err = truncate_partial_nodes(&dn, ri, offset, level);
1075 if (err < 0 && err != -ENOENT)
1077 nofs += 1 + NIDS_PER_BLOCK;
1080 nofs = 5 + 2 * NIDS_PER_BLOCK;
1081 if (!offset[level - 1])
1083 err = truncate_partial_nodes(&dn, ri, offset, level);
1084 if (err < 0 && err != -ENOENT)
1093 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1094 switch (offset[0]) {
1095 case NODE_DIR1_BLOCK:
1096 case NODE_DIR2_BLOCK:
1097 err = truncate_dnode(&dn);
1100 case NODE_IND1_BLOCK:
1101 case NODE_IND2_BLOCK:
1102 err = truncate_nodes(&dn, nofs, offset[1], 2);
1105 case NODE_DIND_BLOCK:
1106 err = truncate_nodes(&dn, nofs, offset[1], 3);
1113 if (err < 0 && err != -ENOENT)
1115 if (offset[1] == 0 &&
1116 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1118 BUG_ON(page->mapping != NODE_MAPPING(sbi));
1119 f2fs_wait_on_page_writeback(page, NODE, true, true);
1120 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1121 set_page_dirty(page);
1129 f2fs_put_page(page, 0);
1130 trace_f2fs_truncate_inode_blocks_exit(inode, err);
1131 return err > 0 ? 0 : err;
1134 /* caller must lock inode page */
1135 int f2fs_truncate_xattr_node(struct inode *inode)
1137 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1138 nid_t nid = F2FS_I(inode)->i_xattr_nid;
1139 struct dnode_of_data dn;
1146 npage = f2fs_get_node_page(sbi, nid);
1148 return PTR_ERR(npage);
1150 set_new_dnode(&dn, inode, NULL, npage, nid);
1151 err = truncate_node(&dn);
1153 f2fs_put_page(npage, 1);
1157 f2fs_i_xnid_write(inode, 0);
1163 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1166 int f2fs_remove_inode_page(struct inode *inode)
1168 struct dnode_of_data dn;
1171 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1172 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1176 err = f2fs_truncate_xattr_node(inode);
1178 f2fs_put_dnode(&dn);
1182 /* remove potential inline_data blocks */
1183 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1184 S_ISLNK(inode->i_mode))
1185 f2fs_truncate_data_blocks_range(&dn, 1);
1187 /* 0 is possible, after f2fs_new_inode() has failed */
1188 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1189 f2fs_put_dnode(&dn);
1193 if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1194 f2fs_warn(F2FS_I_SB(inode),
1195 "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1196 inode->i_ino, (unsigned long long)inode->i_blocks);
1197 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1200 /* will put inode & node pages */
1201 err = truncate_node(&dn);
1203 f2fs_put_dnode(&dn);
1209 struct page *f2fs_new_inode_page(struct inode *inode)
1211 struct dnode_of_data dn;
1213 /* allocate inode page for new inode */
1214 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1216 /* caller should f2fs_put_page(page, 1); */
1217 return f2fs_new_node_page(&dn, 0);
1220 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1222 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1223 struct node_info new_ni;
1227 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1228 return ERR_PTR(-EPERM);
1230 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1232 return ERR_PTR(-ENOMEM);
1234 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1237 #ifdef CONFIG_F2FS_CHECK_FS
1238 err = f2fs_get_node_info(sbi, dn->nid, &new_ni);
1240 dec_valid_node_count(sbi, dn->inode, !ofs);
1243 if (unlikely(new_ni.blk_addr != NULL_ADDR)) {
1244 err = -EFSCORRUPTED;
1245 set_sbi_flag(sbi, SBI_NEED_FSCK);
1249 new_ni.nid = dn->nid;
1250 new_ni.ino = dn->inode->i_ino;
1251 new_ni.blk_addr = NULL_ADDR;
1254 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1256 f2fs_wait_on_page_writeback(page, NODE, true, true);
1257 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1258 set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1259 if (!PageUptodate(page))
1260 SetPageUptodate(page);
1261 if (set_page_dirty(page))
1262 dn->node_changed = true;
1264 if (f2fs_has_xattr_block(ofs))
1265 f2fs_i_xnid_write(dn->inode, dn->nid);
1268 inc_valid_inode_count(sbi);
1272 clear_node_page_dirty(page);
1273 f2fs_put_page(page, 1);
1274 return ERR_PTR(err);
1278 * Caller should do after getting the following values.
1279 * 0: f2fs_put_page(page, 0)
1280 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1282 static int read_node_page(struct page *page, int op_flags)
1284 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1285 struct node_info ni;
1286 struct f2fs_io_info fio = {
1290 .op_flags = op_flags,
1292 .encrypted_page = NULL,
1296 if (PageUptodate(page)) {
1297 if (!f2fs_inode_chksum_verify(sbi, page)) {
1298 ClearPageUptodate(page);
1304 err = f2fs_get_node_info(sbi, page->index, &ni);
1308 if (unlikely(ni.blk_addr == NULL_ADDR) ||
1309 is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1310 ClearPageUptodate(page);
1314 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1316 err = f2fs_submit_page_bio(&fio);
1319 f2fs_update_iostat(sbi, FS_NODE_READ_IO, F2FS_BLKSIZE);
1325 * Readahead a node page
1327 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1334 if (f2fs_check_nid_range(sbi, nid))
1337 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1341 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1345 err = read_node_page(apage, REQ_RAHEAD);
1346 f2fs_put_page(apage, err ? 1 : 0);
1349 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1350 struct page *parent, int start)
1356 return ERR_PTR(-ENOENT);
1357 if (f2fs_check_nid_range(sbi, nid))
1358 return ERR_PTR(-EINVAL);
1360 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1362 return ERR_PTR(-ENOMEM);
1364 err = read_node_page(page, 0);
1366 f2fs_put_page(page, 1);
1367 return ERR_PTR(err);
1368 } else if (err == LOCKED_PAGE) {
1374 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1378 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1379 f2fs_put_page(page, 1);
1383 if (unlikely(!PageUptodate(page))) {
1388 if (!f2fs_inode_chksum_verify(sbi, page)) {
1393 if(unlikely(nid != nid_of_node(page))) {
1394 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1395 nid, nid_of_node(page), ino_of_node(page),
1396 ofs_of_node(page), cpver_of_node(page),
1397 next_blkaddr_of_node(page));
1398 set_sbi_flag(sbi, SBI_NEED_FSCK);
1401 ClearPageUptodate(page);
1402 f2fs_put_page(page, 1);
1403 return ERR_PTR(err);
1408 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1410 return __get_node_page(sbi, nid, NULL, 0);
1413 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1415 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1416 nid_t nid = get_nid(parent, start, false);
1418 return __get_node_page(sbi, nid, parent, start);
1421 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1423 struct inode *inode;
1427 /* should flush inline_data before evict_inode */
1428 inode = ilookup(sbi->sb, ino);
1432 page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1433 FGP_LOCK|FGP_NOWAIT, 0);
1437 if (!PageUptodate(page))
1440 if (!PageDirty(page))
1443 if (!clear_page_dirty_for_io(page))
1446 ret = f2fs_write_inline_data(inode, page);
1447 inode_dec_dirty_pages(inode);
1448 f2fs_remove_dirty_inode(inode);
1450 set_page_dirty(page);
1452 f2fs_put_page(page, 1);
1457 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1460 struct pagevec pvec;
1461 struct page *last_page = NULL;
1464 pagevec_init(&pvec);
1467 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1468 PAGECACHE_TAG_DIRTY))) {
1471 for (i = 0; i < nr_pages; i++) {
1472 struct page *page = pvec.pages[i];
1474 if (unlikely(f2fs_cp_error(sbi))) {
1475 f2fs_put_page(last_page, 0);
1476 pagevec_release(&pvec);
1477 return ERR_PTR(-EIO);
1480 if (!IS_DNODE(page) || !is_cold_node(page))
1482 if (ino_of_node(page) != ino)
1487 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1492 if (ino_of_node(page) != ino)
1493 goto continue_unlock;
1495 if (!PageDirty(page)) {
1496 /* someone wrote it for us */
1497 goto continue_unlock;
1501 f2fs_put_page(last_page, 0);
1507 pagevec_release(&pvec);
1513 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1514 struct writeback_control *wbc, bool do_balance,
1515 enum iostat_type io_type, unsigned int *seq_id)
1517 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1519 struct node_info ni;
1520 struct f2fs_io_info fio = {
1522 .ino = ino_of_node(page),
1525 .op_flags = wbc_to_write_flags(wbc),
1527 .encrypted_page = NULL,
1534 trace_f2fs_writepage(page, NODE);
1536 if (unlikely(f2fs_cp_error(sbi))) {
1537 if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
1538 ClearPageUptodate(page);
1539 dec_page_count(sbi, F2FS_DIRTY_NODES);
1546 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1549 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1550 wbc->sync_mode == WB_SYNC_NONE &&
1551 IS_DNODE(page) && is_cold_node(page))
1554 /* get old block addr of this node page */
1555 nid = nid_of_node(page);
1556 f2fs_bug_on(sbi, page->index != nid);
1558 if (f2fs_get_node_info(sbi, nid, &ni))
1561 if (wbc->for_reclaim) {
1562 if (!down_read_trylock(&sbi->node_write))
1565 down_read(&sbi->node_write);
1568 /* This page is already truncated */
1569 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1570 ClearPageUptodate(page);
1571 dec_page_count(sbi, F2FS_DIRTY_NODES);
1572 up_read(&sbi->node_write);
1577 if (__is_valid_data_blkaddr(ni.blk_addr) &&
1578 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1579 DATA_GENERIC_ENHANCE)) {
1580 up_read(&sbi->node_write);
1584 if (atomic && !test_opt(sbi, NOBARRIER))
1585 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1587 /* should add to global list before clearing PAGECACHE status */
1588 if (f2fs_in_warm_node_list(sbi, page)) {
1589 seq = f2fs_add_fsync_node_entry(sbi, page);
1594 set_page_writeback(page);
1595 ClearPageError(page);
1597 fio.old_blkaddr = ni.blk_addr;
1598 f2fs_do_write_node_page(nid, &fio);
1599 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1600 dec_page_count(sbi, F2FS_DIRTY_NODES);
1601 up_read(&sbi->node_write);
1603 if (wbc->for_reclaim) {
1604 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1610 if (unlikely(f2fs_cp_error(sbi))) {
1611 f2fs_submit_merged_write(sbi, NODE);
1615 *submitted = fio.submitted;
1618 f2fs_balance_fs(sbi, false);
1622 redirty_page_for_writepage(wbc, page);
1623 return AOP_WRITEPAGE_ACTIVATE;
1626 int f2fs_move_node_page(struct page *node_page, int gc_type)
1630 if (gc_type == FG_GC) {
1631 struct writeback_control wbc = {
1632 .sync_mode = WB_SYNC_ALL,
1637 f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1639 set_page_dirty(node_page);
1641 if (!clear_page_dirty_for_io(node_page)) {
1646 if (__write_node_page(node_page, false, NULL,
1647 &wbc, false, FS_GC_NODE_IO, NULL)) {
1649 unlock_page(node_page);
1653 /* set page dirty and write it */
1654 if (!PageWriteback(node_page))
1655 set_page_dirty(node_page);
1658 unlock_page(node_page);
1660 f2fs_put_page(node_page, 0);
1664 static int f2fs_write_node_page(struct page *page,
1665 struct writeback_control *wbc)
1667 return __write_node_page(page, false, NULL, wbc, false,
1671 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1672 struct writeback_control *wbc, bool atomic,
1673 unsigned int *seq_id)
1676 struct pagevec pvec;
1678 struct page *last_page = NULL;
1679 bool marked = false;
1680 nid_t ino = inode->i_ino;
1685 last_page = last_fsync_dnode(sbi, ino);
1686 if (IS_ERR_OR_NULL(last_page))
1687 return PTR_ERR_OR_ZERO(last_page);
1690 pagevec_init(&pvec);
1693 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1694 PAGECACHE_TAG_DIRTY))) {
1697 for (i = 0; i < nr_pages; i++) {
1698 struct page *page = pvec.pages[i];
1699 bool submitted = false;
1701 if (unlikely(f2fs_cp_error(sbi))) {
1702 f2fs_put_page(last_page, 0);
1703 pagevec_release(&pvec);
1708 if (!IS_DNODE(page) || !is_cold_node(page))
1710 if (ino_of_node(page) != ino)
1715 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1720 if (ino_of_node(page) != ino)
1721 goto continue_unlock;
1723 if (!PageDirty(page) && page != last_page) {
1724 /* someone wrote it for us */
1725 goto continue_unlock;
1728 f2fs_wait_on_page_writeback(page, NODE, true, true);
1730 set_fsync_mark(page, 0);
1731 set_dentry_mark(page, 0);
1733 if (!atomic || page == last_page) {
1734 set_fsync_mark(page, 1);
1735 if (IS_INODE(page)) {
1736 if (is_inode_flag_set(inode,
1738 f2fs_update_inode(inode, page);
1739 set_dentry_mark(page,
1740 f2fs_need_dentry_mark(sbi, ino));
1742 /* may be written by other thread */
1743 if (!PageDirty(page))
1744 set_page_dirty(page);
1747 if (!clear_page_dirty_for_io(page))
1748 goto continue_unlock;
1750 ret = __write_node_page(page, atomic &&
1752 &submitted, wbc, true,
1753 FS_NODE_IO, seq_id);
1756 f2fs_put_page(last_page, 0);
1758 } else if (submitted) {
1762 if (page == last_page) {
1763 f2fs_put_page(page, 0);
1768 pagevec_release(&pvec);
1774 if (!ret && atomic && !marked) {
1775 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1776 ino, last_page->index);
1777 lock_page(last_page);
1778 f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1779 set_page_dirty(last_page);
1780 unlock_page(last_page);
1785 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1786 return ret ? -EIO: 0;
1789 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1791 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1794 if (inode->i_ino != ino)
1797 if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1800 spin_lock(&sbi->inode_lock[DIRTY_META]);
1801 clean = list_empty(&F2FS_I(inode)->gdirty_list);
1802 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1807 inode = igrab(inode);
1813 static bool flush_dirty_inode(struct page *page)
1815 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1816 struct inode *inode;
1817 nid_t ino = ino_of_node(page);
1819 inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1823 f2fs_update_inode(inode, page);
1830 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1833 struct pagevec pvec;
1836 pagevec_init(&pvec);
1838 while ((nr_pages = pagevec_lookup_tag(&pvec,
1839 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1842 for (i = 0; i < nr_pages; i++) {
1843 struct page *page = pvec.pages[i];
1845 if (!IS_DNODE(page))
1850 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1856 if (!PageDirty(page)) {
1857 /* someone wrote it for us */
1858 goto continue_unlock;
1861 /* flush inline_data, if it's async context. */
1862 if (is_inline_node(page)) {
1863 clear_inline_node(page);
1865 flush_inline_data(sbi, ino_of_node(page));
1870 pagevec_release(&pvec);
1875 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1876 struct writeback_control *wbc,
1877 bool do_balance, enum iostat_type io_type)
1880 struct pagevec pvec;
1884 int nr_pages, done = 0;
1886 pagevec_init(&pvec);
1891 while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1892 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1895 for (i = 0; i < nr_pages; i++) {
1896 struct page *page = pvec.pages[i];
1897 bool submitted = false;
1898 bool may_dirty = true;
1900 /* give a priority to WB_SYNC threads */
1901 if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1902 wbc->sync_mode == WB_SYNC_NONE) {
1908 * flushing sequence with step:
1913 if (step == 0 && IS_DNODE(page))
1915 if (step == 1 && (!IS_DNODE(page) ||
1916 is_cold_node(page)))
1918 if (step == 2 && (!IS_DNODE(page) ||
1919 !is_cold_node(page)))
1922 if (wbc->sync_mode == WB_SYNC_ALL)
1924 else if (!trylock_page(page))
1927 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1933 if (!PageDirty(page)) {
1934 /* someone wrote it for us */
1935 goto continue_unlock;
1938 /* flush inline_data/inode, if it's async context. */
1942 /* flush inline_data */
1943 if (is_inline_node(page)) {
1944 clear_inline_node(page);
1946 flush_inline_data(sbi, ino_of_node(page));
1950 /* flush dirty inode */
1951 if (IS_INODE(page) && may_dirty) {
1953 if (flush_dirty_inode(page))
1957 f2fs_wait_on_page_writeback(page, NODE, true, true);
1959 if (!clear_page_dirty_for_io(page))
1960 goto continue_unlock;
1962 set_fsync_mark(page, 0);
1963 set_dentry_mark(page, 0);
1965 ret = __write_node_page(page, false, &submitted,
1966 wbc, do_balance, io_type, NULL);
1972 if (--wbc->nr_to_write == 0)
1975 pagevec_release(&pvec);
1978 if (wbc->nr_to_write == 0) {
1985 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1986 wbc->sync_mode == WB_SYNC_NONE && step == 1)
1993 f2fs_submit_merged_write(sbi, NODE);
1995 if (unlikely(f2fs_cp_error(sbi)))
2000 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2001 unsigned int seq_id)
2003 struct fsync_node_entry *fn;
2005 struct list_head *head = &sbi->fsync_node_list;
2006 unsigned long flags;
2007 unsigned int cur_seq_id = 0;
2010 while (seq_id && cur_seq_id < seq_id) {
2011 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2012 if (list_empty(head)) {
2013 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2016 fn = list_first_entry(head, struct fsync_node_entry, list);
2017 if (fn->seq_id > seq_id) {
2018 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2021 cur_seq_id = fn->seq_id;
2024 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2026 f2fs_wait_on_page_writeback(page, NODE, true, false);
2027 if (TestClearPageError(page))
2036 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2043 static int f2fs_write_node_pages(struct address_space *mapping,
2044 struct writeback_control *wbc)
2046 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2047 struct blk_plug plug;
2050 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2053 /* balancing f2fs's metadata in background */
2054 f2fs_balance_fs_bg(sbi, true);
2056 /* collect a number of dirty node pages and write together */
2057 if (wbc->sync_mode != WB_SYNC_ALL &&
2058 get_pages(sbi, F2FS_DIRTY_NODES) <
2059 nr_pages_to_skip(sbi, NODE))
2062 if (wbc->sync_mode == WB_SYNC_ALL)
2063 atomic_inc(&sbi->wb_sync_req[NODE]);
2064 else if (atomic_read(&sbi->wb_sync_req[NODE])) {
2065 /* to avoid potential deadlock */
2067 blk_finish_plug(current->plug);
2071 trace_f2fs_writepages(mapping->host, wbc, NODE);
2073 diff = nr_pages_to_write(sbi, NODE, wbc);
2074 blk_start_plug(&plug);
2075 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2076 blk_finish_plug(&plug);
2077 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2079 if (wbc->sync_mode == WB_SYNC_ALL)
2080 atomic_dec(&sbi->wb_sync_req[NODE]);
2084 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2085 trace_f2fs_writepages(mapping->host, wbc, NODE);
2089 static int f2fs_set_node_page_dirty(struct page *page)
2091 trace_f2fs_set_page_dirty(page, NODE);
2093 if (!PageUptodate(page))
2094 SetPageUptodate(page);
2095 #ifdef CONFIG_F2FS_CHECK_FS
2097 f2fs_inode_chksum_set(F2FS_P_SB(page), page);
2099 if (!PageDirty(page)) {
2100 __set_page_dirty_nobuffers(page);
2101 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
2102 f2fs_set_page_private(page, 0);
2103 f2fs_trace_pid(page);
2110 * Structure of the f2fs node operations
2112 const struct address_space_operations f2fs_node_aops = {
2113 .writepage = f2fs_write_node_page,
2114 .writepages = f2fs_write_node_pages,
2115 .set_page_dirty = f2fs_set_node_page_dirty,
2116 .invalidatepage = f2fs_invalidate_page,
2117 .releasepage = f2fs_release_page,
2118 #ifdef CONFIG_MIGRATION
2119 .migratepage = f2fs_migrate_page,
2123 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2126 return radix_tree_lookup(&nm_i->free_nid_root, n);
2129 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2132 struct f2fs_nm_info *nm_i = NM_I(sbi);
2134 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2138 nm_i->nid_cnt[FREE_NID]++;
2139 list_add_tail(&i->list, &nm_i->free_nid_list);
2143 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2144 struct free_nid *i, enum nid_state state)
2146 struct f2fs_nm_info *nm_i = NM_I(sbi);
2148 f2fs_bug_on(sbi, state != i->state);
2149 nm_i->nid_cnt[state]--;
2150 if (state == FREE_NID)
2152 radix_tree_delete(&nm_i->free_nid_root, i->nid);
2155 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2156 enum nid_state org_state, enum nid_state dst_state)
2158 struct f2fs_nm_info *nm_i = NM_I(sbi);
2160 f2fs_bug_on(sbi, org_state != i->state);
2161 i->state = dst_state;
2162 nm_i->nid_cnt[org_state]--;
2163 nm_i->nid_cnt[dst_state]++;
2165 switch (dst_state) {
2170 list_add_tail(&i->list, &nm_i->free_nid_list);
2177 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2178 bool set, bool build)
2180 struct f2fs_nm_info *nm_i = NM_I(sbi);
2181 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2182 unsigned int nid_ofs = nid - START_NID(nid);
2184 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2188 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2190 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2191 nm_i->free_nid_count[nat_ofs]++;
2193 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2195 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2197 nm_i->free_nid_count[nat_ofs]--;
2201 /* return if the nid is recognized as free */
2202 static bool add_free_nid(struct f2fs_sb_info *sbi,
2203 nid_t nid, bool build, bool update)
2205 struct f2fs_nm_info *nm_i = NM_I(sbi);
2206 struct free_nid *i, *e;
2207 struct nat_entry *ne;
2211 /* 0 nid should not be used */
2212 if (unlikely(nid == 0))
2215 if (unlikely(f2fs_check_nid_range(sbi, nid)))
2218 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2220 i->state = FREE_NID;
2222 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2224 spin_lock(&nm_i->nid_list_lock);
2232 * - __insert_nid_to_list(PREALLOC_NID)
2233 * - f2fs_balance_fs_bg
2234 * - f2fs_build_free_nids
2235 * - __f2fs_build_free_nids
2238 * - __lookup_nat_cache
2240 * - f2fs_init_inode_metadata
2241 * - f2fs_new_inode_page
2242 * - f2fs_new_node_page
2244 * - f2fs_alloc_nid_done
2245 * - __remove_nid_from_list(PREALLOC_NID)
2246 * - __insert_nid_to_list(FREE_NID)
2248 ne = __lookup_nat_cache(nm_i, nid);
2249 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2250 nat_get_blkaddr(ne) != NULL_ADDR))
2253 e = __lookup_free_nid_list(nm_i, nid);
2255 if (e->state == FREE_NID)
2261 err = __insert_free_nid(sbi, i);
2264 update_free_nid_bitmap(sbi, nid, ret, build);
2266 nm_i->available_nids++;
2268 spin_unlock(&nm_i->nid_list_lock);
2269 radix_tree_preload_end();
2272 kmem_cache_free(free_nid_slab, i);
2276 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2278 struct f2fs_nm_info *nm_i = NM_I(sbi);
2280 bool need_free = false;
2282 spin_lock(&nm_i->nid_list_lock);
2283 i = __lookup_free_nid_list(nm_i, nid);
2284 if (i && i->state == FREE_NID) {
2285 __remove_free_nid(sbi, i, FREE_NID);
2288 spin_unlock(&nm_i->nid_list_lock);
2291 kmem_cache_free(free_nid_slab, i);
2294 static int scan_nat_page(struct f2fs_sb_info *sbi,
2295 struct page *nat_page, nid_t start_nid)
2297 struct f2fs_nm_info *nm_i = NM_I(sbi);
2298 struct f2fs_nat_block *nat_blk = page_address(nat_page);
2300 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2303 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2305 i = start_nid % NAT_ENTRY_PER_BLOCK;
2307 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2308 if (unlikely(start_nid >= nm_i->max_nid))
2311 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2313 if (blk_addr == NEW_ADDR)
2316 if (blk_addr == NULL_ADDR) {
2317 add_free_nid(sbi, start_nid, true, true);
2319 spin_lock(&NM_I(sbi)->nid_list_lock);
2320 update_free_nid_bitmap(sbi, start_nid, false, true);
2321 spin_unlock(&NM_I(sbi)->nid_list_lock);
2328 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2330 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2331 struct f2fs_journal *journal = curseg->journal;
2334 down_read(&curseg->journal_rwsem);
2335 for (i = 0; i < nats_in_cursum(journal); i++) {
2339 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2340 nid = le32_to_cpu(nid_in_journal(journal, i));
2341 if (addr == NULL_ADDR)
2342 add_free_nid(sbi, nid, true, false);
2344 remove_free_nid(sbi, nid);
2346 up_read(&curseg->journal_rwsem);
2349 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2351 struct f2fs_nm_info *nm_i = NM_I(sbi);
2352 unsigned int i, idx;
2355 down_read(&nm_i->nat_tree_lock);
2357 for (i = 0; i < nm_i->nat_blocks; i++) {
2358 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2360 if (!nm_i->free_nid_count[i])
2362 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2363 idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2364 NAT_ENTRY_PER_BLOCK, idx);
2365 if (idx >= NAT_ENTRY_PER_BLOCK)
2368 nid = i * NAT_ENTRY_PER_BLOCK + idx;
2369 add_free_nid(sbi, nid, true, false);
2371 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2376 scan_curseg_cache(sbi);
2378 up_read(&nm_i->nat_tree_lock);
2381 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2382 bool sync, bool mount)
2384 struct f2fs_nm_info *nm_i = NM_I(sbi);
2386 nid_t nid = nm_i->next_scan_nid;
2388 if (unlikely(nid >= nm_i->max_nid))
2391 if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2392 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2394 /* Enough entries */
2395 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2398 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2402 /* try to find free nids in free_nid_bitmap */
2403 scan_free_nid_bits(sbi);
2405 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2409 /* readahead nat pages to be scanned */
2410 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2413 down_read(&nm_i->nat_tree_lock);
2416 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2417 nm_i->nat_block_bitmap)) {
2418 struct page *page = get_current_nat_page(sbi, nid);
2421 ret = PTR_ERR(page);
2423 ret = scan_nat_page(sbi, page, nid);
2424 f2fs_put_page(page, 1);
2428 up_read(&nm_i->nat_tree_lock);
2429 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2434 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2435 if (unlikely(nid >= nm_i->max_nid))
2438 if (++i >= FREE_NID_PAGES)
2442 /* go to the next free nat pages to find free nids abundantly */
2443 nm_i->next_scan_nid = nid;
2445 /* find free nids from current sum_pages */
2446 scan_curseg_cache(sbi);
2448 up_read(&nm_i->nat_tree_lock);
2450 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2451 nm_i->ra_nid_pages, META_NAT, false);
2456 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2460 mutex_lock(&NM_I(sbi)->build_lock);
2461 ret = __f2fs_build_free_nids(sbi, sync, mount);
2462 mutex_unlock(&NM_I(sbi)->build_lock);
2468 * If this function returns success, caller can obtain a new nid
2469 * from second parameter of this function.
2470 * The returned nid could be used ino as well as nid when inode is created.
2472 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2474 struct f2fs_nm_info *nm_i = NM_I(sbi);
2475 struct free_nid *i = NULL;
2477 if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2478 f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2482 spin_lock(&nm_i->nid_list_lock);
2484 if (unlikely(nm_i->available_nids == 0)) {
2485 spin_unlock(&nm_i->nid_list_lock);
2489 /* We should not use stale free nids created by f2fs_build_free_nids */
2490 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2491 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2492 i = list_first_entry(&nm_i->free_nid_list,
2493 struct free_nid, list);
2496 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2497 nm_i->available_nids--;
2499 update_free_nid_bitmap(sbi, *nid, false, false);
2501 spin_unlock(&nm_i->nid_list_lock);
2504 spin_unlock(&nm_i->nid_list_lock);
2506 /* Let's scan nat pages and its caches to get free nids */
2507 if (!f2fs_build_free_nids(sbi, true, false))
2513 * f2fs_alloc_nid() should be called prior to this function.
2515 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2517 struct f2fs_nm_info *nm_i = NM_I(sbi);
2520 spin_lock(&nm_i->nid_list_lock);
2521 i = __lookup_free_nid_list(nm_i, nid);
2522 f2fs_bug_on(sbi, !i);
2523 __remove_free_nid(sbi, i, PREALLOC_NID);
2524 spin_unlock(&nm_i->nid_list_lock);
2526 kmem_cache_free(free_nid_slab, i);
2530 * f2fs_alloc_nid() should be called prior to this function.
2532 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2534 struct f2fs_nm_info *nm_i = NM_I(sbi);
2536 bool need_free = false;
2541 spin_lock(&nm_i->nid_list_lock);
2542 i = __lookup_free_nid_list(nm_i, nid);
2543 f2fs_bug_on(sbi, !i);
2545 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2546 __remove_free_nid(sbi, i, PREALLOC_NID);
2549 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2552 nm_i->available_nids++;
2554 update_free_nid_bitmap(sbi, nid, true, false);
2556 spin_unlock(&nm_i->nid_list_lock);
2559 kmem_cache_free(free_nid_slab, i);
2562 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2564 struct f2fs_nm_info *nm_i = NM_I(sbi);
2567 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2570 if (!mutex_trylock(&nm_i->build_lock))
2573 while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2574 struct free_nid *i, *next;
2575 unsigned int batch = SHRINK_NID_BATCH_SIZE;
2577 spin_lock(&nm_i->nid_list_lock);
2578 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2579 if (!nr_shrink || !batch ||
2580 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2582 __remove_free_nid(sbi, i, FREE_NID);
2583 kmem_cache_free(free_nid_slab, i);
2587 spin_unlock(&nm_i->nid_list_lock);
2590 mutex_unlock(&nm_i->build_lock);
2592 return nr - nr_shrink;
2595 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2597 void *src_addr, *dst_addr;
2600 struct f2fs_inode *ri;
2602 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2604 return PTR_ERR(ipage);
2606 ri = F2FS_INODE(page);
2607 if (ri->i_inline & F2FS_INLINE_XATTR) {
2608 set_inode_flag(inode, FI_INLINE_XATTR);
2610 clear_inode_flag(inode, FI_INLINE_XATTR);
2614 dst_addr = inline_xattr_addr(inode, ipage);
2615 src_addr = inline_xattr_addr(inode, page);
2616 inline_size = inline_xattr_size(inode);
2618 f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2619 memcpy(dst_addr, src_addr, inline_size);
2621 f2fs_update_inode(inode, ipage);
2622 f2fs_put_page(ipage, 1);
2626 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2628 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2629 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2631 struct dnode_of_data dn;
2632 struct node_info ni;
2639 /* 1: invalidate the previous xattr nid */
2640 err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2644 f2fs_invalidate_blocks(sbi, ni.blk_addr);
2645 dec_valid_node_count(sbi, inode, false);
2646 set_node_addr(sbi, &ni, NULL_ADDR, false);
2649 /* 2: update xattr nid in inode */
2650 if (!f2fs_alloc_nid(sbi, &new_xnid))
2653 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2654 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2655 if (IS_ERR(xpage)) {
2656 f2fs_alloc_nid_failed(sbi, new_xnid);
2657 return PTR_ERR(xpage);
2660 f2fs_alloc_nid_done(sbi, new_xnid);
2661 f2fs_update_inode_page(inode);
2663 /* 3: update and set xattr node page dirty */
2664 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2666 set_page_dirty(xpage);
2667 f2fs_put_page(xpage, 1);
2672 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2674 struct f2fs_inode *src, *dst;
2675 nid_t ino = ino_of_node(page);
2676 struct node_info old_ni, new_ni;
2680 err = f2fs_get_node_info(sbi, ino, &old_ni);
2684 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2687 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2689 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2693 /* Should not use this inode from free nid list */
2694 remove_free_nid(sbi, ino);
2696 if (!PageUptodate(ipage))
2697 SetPageUptodate(ipage);
2698 fill_node_footer(ipage, ino, ino, 0, true);
2699 set_cold_node(ipage, false);
2701 src = F2FS_INODE(page);
2702 dst = F2FS_INODE(ipage);
2704 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2706 dst->i_blocks = cpu_to_le64(1);
2707 dst->i_links = cpu_to_le32(1);
2708 dst->i_xattr_nid = 0;
2709 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2710 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2711 dst->i_extra_isize = src->i_extra_isize;
2713 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2714 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2715 i_inline_xattr_size))
2716 dst->i_inline_xattr_size = src->i_inline_xattr_size;
2718 if (f2fs_sb_has_project_quota(sbi) &&
2719 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2721 dst->i_projid = src->i_projid;
2723 if (f2fs_sb_has_inode_crtime(sbi) &&
2724 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2726 dst->i_crtime = src->i_crtime;
2727 dst->i_crtime_nsec = src->i_crtime_nsec;
2734 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2736 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2737 inc_valid_inode_count(sbi);
2738 set_page_dirty(ipage);
2739 f2fs_put_page(ipage, 1);
2743 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2744 unsigned int segno, struct f2fs_summary_block *sum)
2746 struct f2fs_node *rn;
2747 struct f2fs_summary *sum_entry;
2749 int i, idx, last_offset, nrpages;
2751 /* scan the node segment */
2752 last_offset = sbi->blocks_per_seg;
2753 addr = START_BLOCK(sbi, segno);
2754 sum_entry = &sum->entries[0];
2756 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2757 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2759 /* readahead node pages */
2760 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2762 for (idx = addr; idx < addr + nrpages; idx++) {
2763 struct page *page = f2fs_get_tmp_page(sbi, idx);
2766 return PTR_ERR(page);
2768 rn = F2FS_NODE(page);
2769 sum_entry->nid = rn->footer.nid;
2770 sum_entry->version = 0;
2771 sum_entry->ofs_in_node = 0;
2773 f2fs_put_page(page, 1);
2776 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2782 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2784 struct f2fs_nm_info *nm_i = NM_I(sbi);
2785 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2786 struct f2fs_journal *journal = curseg->journal;
2789 down_write(&curseg->journal_rwsem);
2790 for (i = 0; i < nats_in_cursum(journal); i++) {
2791 struct nat_entry *ne;
2792 struct f2fs_nat_entry raw_ne;
2793 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2795 if (f2fs_check_nid_range(sbi, nid))
2798 raw_ne = nat_in_journal(journal, i);
2800 ne = __lookup_nat_cache(nm_i, nid);
2802 ne = __alloc_nat_entry(nid, true);
2803 __init_nat_entry(nm_i, ne, &raw_ne, true);
2807 * if a free nat in journal has not been used after last
2808 * checkpoint, we should remove it from available nids,
2809 * since later we will add it again.
2811 if (!get_nat_flag(ne, IS_DIRTY) &&
2812 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2813 spin_lock(&nm_i->nid_list_lock);
2814 nm_i->available_nids--;
2815 spin_unlock(&nm_i->nid_list_lock);
2818 __set_nat_cache_dirty(nm_i, ne);
2820 update_nats_in_cursum(journal, -i);
2821 up_write(&curseg->journal_rwsem);
2824 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2825 struct list_head *head, int max)
2827 struct nat_entry_set *cur;
2829 if (nes->entry_cnt >= max)
2832 list_for_each_entry(cur, head, set_list) {
2833 if (cur->entry_cnt >= nes->entry_cnt) {
2834 list_add(&nes->set_list, cur->set_list.prev);
2839 list_add_tail(&nes->set_list, head);
2842 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2845 struct f2fs_nm_info *nm_i = NM_I(sbi);
2846 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2847 struct f2fs_nat_block *nat_blk = page_address(page);
2851 if (!enabled_nat_bits(sbi, NULL))
2854 if (nat_index == 0) {
2858 for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2859 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2863 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2864 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2868 __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2869 if (valid == NAT_ENTRY_PER_BLOCK)
2870 __set_bit_le(nat_index, nm_i->full_nat_bits);
2872 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2875 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2876 struct nat_entry_set *set, struct cp_control *cpc)
2878 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2879 struct f2fs_journal *journal = curseg->journal;
2880 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2881 bool to_journal = true;
2882 struct f2fs_nat_block *nat_blk;
2883 struct nat_entry *ne, *cur;
2884 struct page *page = NULL;
2887 * there are two steps to flush nat entries:
2888 * #1, flush nat entries to journal in current hot data summary block.
2889 * #2, flush nat entries to nat page.
2891 if (enabled_nat_bits(sbi, cpc) ||
2892 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2896 down_write(&curseg->journal_rwsem);
2898 page = get_next_nat_page(sbi, start_nid);
2900 return PTR_ERR(page);
2902 nat_blk = page_address(page);
2903 f2fs_bug_on(sbi, !nat_blk);
2906 /* flush dirty nats in nat entry set */
2907 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2908 struct f2fs_nat_entry *raw_ne;
2909 nid_t nid = nat_get_nid(ne);
2912 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2915 offset = f2fs_lookup_journal_in_cursum(journal,
2916 NAT_JOURNAL, nid, 1);
2917 f2fs_bug_on(sbi, offset < 0);
2918 raw_ne = &nat_in_journal(journal, offset);
2919 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2921 raw_ne = &nat_blk->entries[nid - start_nid];
2923 raw_nat_from_node_info(raw_ne, &ne->ni);
2925 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2926 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2927 add_free_nid(sbi, nid, false, true);
2929 spin_lock(&NM_I(sbi)->nid_list_lock);
2930 update_free_nid_bitmap(sbi, nid, false, false);
2931 spin_unlock(&NM_I(sbi)->nid_list_lock);
2936 up_write(&curseg->journal_rwsem);
2938 __update_nat_bits(sbi, start_nid, page);
2939 f2fs_put_page(page, 1);
2942 /* Allow dirty nats by node block allocation in write_begin */
2943 if (!set->entry_cnt) {
2944 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2945 kmem_cache_free(nat_entry_set_slab, set);
2951 * This function is called during the checkpointing process.
2953 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2955 struct f2fs_nm_info *nm_i = NM_I(sbi);
2956 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2957 struct f2fs_journal *journal = curseg->journal;
2958 struct nat_entry_set *setvec[SETVEC_SIZE];
2959 struct nat_entry_set *set, *tmp;
2966 * during unmount, let's flush nat_bits before checking
2967 * nat_cnt[DIRTY_NAT].
2969 if (enabled_nat_bits(sbi, cpc)) {
2970 down_write(&nm_i->nat_tree_lock);
2971 remove_nats_in_journal(sbi);
2972 up_write(&nm_i->nat_tree_lock);
2975 if (!nm_i->nat_cnt[DIRTY_NAT])
2978 down_write(&nm_i->nat_tree_lock);
2981 * if there are no enough space in journal to store dirty nat
2982 * entries, remove all entries from journal and merge them
2983 * into nat entry set.
2985 if (enabled_nat_bits(sbi, cpc) ||
2986 !__has_cursum_space(journal,
2987 nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
2988 remove_nats_in_journal(sbi);
2990 while ((found = __gang_lookup_nat_set(nm_i,
2991 set_idx, SETVEC_SIZE, setvec))) {
2993 set_idx = setvec[found - 1]->set + 1;
2994 for (idx = 0; idx < found; idx++)
2995 __adjust_nat_entry_set(setvec[idx], &sets,
2996 MAX_NAT_JENTRIES(journal));
2999 /* flush dirty nats in nat entry set */
3000 list_for_each_entry_safe(set, tmp, &sets, set_list) {
3001 err = __flush_nat_entry_set(sbi, set, cpc);
3006 up_write(&nm_i->nat_tree_lock);
3007 /* Allow dirty nats by node block allocation in write_begin */
3012 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3014 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3015 struct f2fs_nm_info *nm_i = NM_I(sbi);
3016 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3018 __u64 cp_ver = cur_cp_version(ckpt);
3019 block_t nat_bits_addr;
3021 if (!enabled_nat_bits(sbi, NULL))
3024 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3025 nm_i->nat_bits = f2fs_kvzalloc(sbi,
3026 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3027 if (!nm_i->nat_bits)
3030 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3031 nm_i->nat_bits_blocks;
3032 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3035 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3037 return PTR_ERR(page);
3039 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3040 page_address(page), F2FS_BLKSIZE);
3041 f2fs_put_page(page, 1);
3044 cp_ver |= (cur_cp_crc(ckpt) << 32);
3045 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3046 disable_nat_bits(sbi, true);
3050 nm_i->full_nat_bits = nm_i->nat_bits + 8;
3051 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3053 f2fs_notice(sbi, "Found nat_bits in checkpoint");
3057 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3059 struct f2fs_nm_info *nm_i = NM_I(sbi);
3061 nid_t nid, last_nid;
3063 if (!enabled_nat_bits(sbi, NULL))
3066 for (i = 0; i < nm_i->nat_blocks; i++) {
3067 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3068 if (i >= nm_i->nat_blocks)
3071 __set_bit_le(i, nm_i->nat_block_bitmap);
3073 nid = i * NAT_ENTRY_PER_BLOCK;
3074 last_nid = nid + NAT_ENTRY_PER_BLOCK;
3076 spin_lock(&NM_I(sbi)->nid_list_lock);
3077 for (; nid < last_nid; nid++)
3078 update_free_nid_bitmap(sbi, nid, true, true);
3079 spin_unlock(&NM_I(sbi)->nid_list_lock);
3082 for (i = 0; i < nm_i->nat_blocks; i++) {
3083 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3084 if (i >= nm_i->nat_blocks)
3087 __set_bit_le(i, nm_i->nat_block_bitmap);
3091 static int init_node_manager(struct f2fs_sb_info *sbi)
3093 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3094 struct f2fs_nm_info *nm_i = NM_I(sbi);
3095 unsigned char *version_bitmap;
3096 unsigned int nat_segs;
3099 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3101 /* segment_count_nat includes pair segment so divide to 2. */
3102 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3103 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3104 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3106 /* not used nids: 0, node, meta, (and root counted as valid node) */
3107 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3108 F2FS_RESERVED_NODE_NUM;
3109 nm_i->nid_cnt[FREE_NID] = 0;
3110 nm_i->nid_cnt[PREALLOC_NID] = 0;
3111 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3112 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3113 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3115 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3116 INIT_LIST_HEAD(&nm_i->free_nid_list);
3117 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3118 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3119 INIT_LIST_HEAD(&nm_i->nat_entries);
3120 spin_lock_init(&nm_i->nat_list_lock);
3122 mutex_init(&nm_i->build_lock);
3123 spin_lock_init(&nm_i->nid_list_lock);
3124 init_rwsem(&nm_i->nat_tree_lock);
3126 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3127 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3128 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3129 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3131 if (!nm_i->nat_bitmap)
3134 err = __get_nat_bitmaps(sbi);
3138 #ifdef CONFIG_F2FS_CHECK_FS
3139 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3141 if (!nm_i->nat_bitmap_mir)
3148 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3150 struct f2fs_nm_info *nm_i = NM_I(sbi);
3153 nm_i->free_nid_bitmap =
3154 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3157 if (!nm_i->free_nid_bitmap)
3160 for (i = 0; i < nm_i->nat_blocks; i++) {
3161 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3162 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3163 if (!nm_i->free_nid_bitmap[i])
3167 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3169 if (!nm_i->nat_block_bitmap)
3172 nm_i->free_nid_count =
3173 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3176 if (!nm_i->free_nid_count)
3181 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3185 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3190 err = init_node_manager(sbi);
3194 err = init_free_nid_cache(sbi);
3198 /* load free nid status from nat_bits table */
3199 load_free_nid_bitmap(sbi);
3201 return f2fs_build_free_nids(sbi, true, true);
3204 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3206 struct f2fs_nm_info *nm_i = NM_I(sbi);
3207 struct free_nid *i, *next_i;
3208 struct nat_entry *natvec[NATVEC_SIZE];
3209 struct nat_entry_set *setvec[SETVEC_SIZE];
3216 /* destroy free nid list */
3217 spin_lock(&nm_i->nid_list_lock);
3218 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3219 __remove_free_nid(sbi, i, FREE_NID);
3220 spin_unlock(&nm_i->nid_list_lock);
3221 kmem_cache_free(free_nid_slab, i);
3222 spin_lock(&nm_i->nid_list_lock);
3224 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3225 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3226 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3227 spin_unlock(&nm_i->nid_list_lock);
3229 /* destroy nat cache */
3230 down_write(&nm_i->nat_tree_lock);
3231 while ((found = __gang_lookup_nat_cache(nm_i,
3232 nid, NATVEC_SIZE, natvec))) {
3235 nid = nat_get_nid(natvec[found - 1]) + 1;
3236 for (idx = 0; idx < found; idx++) {
3237 spin_lock(&nm_i->nat_list_lock);
3238 list_del(&natvec[idx]->list);
3239 spin_unlock(&nm_i->nat_list_lock);
3241 __del_from_nat_cache(nm_i, natvec[idx]);
3244 f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3246 /* destroy nat set cache */
3248 while ((found = __gang_lookup_nat_set(nm_i,
3249 nid, SETVEC_SIZE, setvec))) {
3252 nid = setvec[found - 1]->set + 1;
3253 for (idx = 0; idx < found; idx++) {
3254 /* entry_cnt is not zero, when cp_error was occurred */
3255 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3256 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3257 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3260 up_write(&nm_i->nat_tree_lock);
3262 kvfree(nm_i->nat_block_bitmap);
3263 if (nm_i->free_nid_bitmap) {
3266 for (i = 0; i < nm_i->nat_blocks; i++)
3267 kvfree(nm_i->free_nid_bitmap[i]);
3268 kvfree(nm_i->free_nid_bitmap);
3270 kvfree(nm_i->free_nid_count);
3272 kvfree(nm_i->nat_bitmap);
3273 kvfree(nm_i->nat_bits);
3274 #ifdef CONFIG_F2FS_CHECK_FS
3275 kvfree(nm_i->nat_bitmap_mir);
3277 sbi->nm_info = NULL;
3281 int __init f2fs_create_node_manager_caches(void)
3283 nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3284 sizeof(struct nat_entry));
3285 if (!nat_entry_slab)
3288 free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3289 sizeof(struct free_nid));
3291 goto destroy_nat_entry;
3293 nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3294 sizeof(struct nat_entry_set));
3295 if (!nat_entry_set_slab)
3296 goto destroy_free_nid;
3298 fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3299 sizeof(struct fsync_node_entry));
3300 if (!fsync_node_entry_slab)
3301 goto destroy_nat_entry_set;
3304 destroy_nat_entry_set:
3305 kmem_cache_destroy(nat_entry_set_slab);
3307 kmem_cache_destroy(free_nid_slab);
3309 kmem_cache_destroy(nat_entry_slab);
3314 void f2fs_destroy_node_manager_caches(void)
3316 kmem_cache_destroy(fsync_node_entry_slab);
3317 kmem_cache_destroy(nat_entry_set_slab);
3318 kmem_cache_destroy(free_nid_slab);
3319 kmem_cache_destroy(nat_entry_slab);