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 * sizeof(struct nat_entry)) >>
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_nofail(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);
184 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
186 struct nat_entry *ne;
188 ne = radix_tree_lookup(&nm_i->nat_root, n);
190 /* for recent accessed nat entry, move it to tail of lru list */
191 if (ne && !get_nat_flag(ne, IS_DIRTY)) {
192 spin_lock(&nm_i->nat_list_lock);
193 if (!list_empty(&ne->list))
194 list_move_tail(&ne->list, &nm_i->nat_entries);
195 spin_unlock(&nm_i->nat_list_lock);
201 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
202 nid_t start, unsigned int nr, struct nat_entry **ep)
204 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
207 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
209 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
214 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
215 struct nat_entry *ne)
217 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
218 struct nat_entry_set *head;
220 head = radix_tree_lookup(&nm_i->nat_set_root, set);
222 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
224 INIT_LIST_HEAD(&head->entry_list);
225 INIT_LIST_HEAD(&head->set_list);
228 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
233 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
234 struct nat_entry *ne)
236 struct nat_entry_set *head;
237 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
240 head = __grab_nat_entry_set(nm_i, ne);
243 * update entry_cnt in below condition:
244 * 1. update NEW_ADDR to valid block address;
245 * 2. update old block address to new one;
247 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
248 !get_nat_flag(ne, IS_DIRTY)))
251 set_nat_flag(ne, IS_PREALLOC, new_ne);
253 if (get_nat_flag(ne, IS_DIRTY))
256 nm_i->dirty_nat_cnt++;
257 set_nat_flag(ne, IS_DIRTY, true);
259 spin_lock(&nm_i->nat_list_lock);
261 list_del_init(&ne->list);
263 list_move_tail(&ne->list, &head->entry_list);
264 spin_unlock(&nm_i->nat_list_lock);
267 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
268 struct nat_entry_set *set, struct nat_entry *ne)
270 spin_lock(&nm_i->nat_list_lock);
271 list_move_tail(&ne->list, &nm_i->nat_entries);
272 spin_unlock(&nm_i->nat_list_lock);
274 set_nat_flag(ne, IS_DIRTY, false);
276 nm_i->dirty_nat_cnt--;
279 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
280 nid_t start, unsigned int nr, struct nat_entry_set **ep)
282 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
286 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
288 return NODE_MAPPING(sbi) == page->mapping &&
289 IS_DNODE(page) && is_cold_node(page);
292 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
294 spin_lock_init(&sbi->fsync_node_lock);
295 INIT_LIST_HEAD(&sbi->fsync_node_list);
296 sbi->fsync_seg_id = 0;
297 sbi->fsync_node_num = 0;
300 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
303 struct fsync_node_entry *fn;
307 fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS);
311 INIT_LIST_HEAD(&fn->list);
313 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
314 list_add_tail(&fn->list, &sbi->fsync_node_list);
315 fn->seq_id = sbi->fsync_seg_id++;
317 sbi->fsync_node_num++;
318 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
323 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
325 struct fsync_node_entry *fn;
328 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
329 list_for_each_entry(fn, &sbi->fsync_node_list, list) {
330 if (fn->page == page) {
332 sbi->fsync_node_num--;
333 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
334 kmem_cache_free(fsync_node_entry_slab, fn);
339 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
343 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
347 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
348 sbi->fsync_seg_id = 0;
349 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
352 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
354 struct f2fs_nm_info *nm_i = NM_I(sbi);
358 down_read(&nm_i->nat_tree_lock);
359 e = __lookup_nat_cache(nm_i, nid);
361 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
362 !get_nat_flag(e, HAS_FSYNCED_INODE))
365 up_read(&nm_i->nat_tree_lock);
369 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
371 struct f2fs_nm_info *nm_i = NM_I(sbi);
375 down_read(&nm_i->nat_tree_lock);
376 e = __lookup_nat_cache(nm_i, nid);
377 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
379 up_read(&nm_i->nat_tree_lock);
383 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
385 struct f2fs_nm_info *nm_i = NM_I(sbi);
387 bool need_update = true;
389 down_read(&nm_i->nat_tree_lock);
390 e = __lookup_nat_cache(nm_i, ino);
391 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
392 (get_nat_flag(e, IS_CHECKPOINTED) ||
393 get_nat_flag(e, HAS_FSYNCED_INODE)))
395 up_read(&nm_i->nat_tree_lock);
399 /* must be locked by nat_tree_lock */
400 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
401 struct f2fs_nat_entry *ne)
403 struct f2fs_nm_info *nm_i = NM_I(sbi);
404 struct nat_entry *new, *e;
406 new = __alloc_nat_entry(nid, false);
410 down_write(&nm_i->nat_tree_lock);
411 e = __lookup_nat_cache(nm_i, nid);
413 e = __init_nat_entry(nm_i, new, ne, false);
415 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
416 nat_get_blkaddr(e) !=
417 le32_to_cpu(ne->block_addr) ||
418 nat_get_version(e) != ne->version);
419 up_write(&nm_i->nat_tree_lock);
421 __free_nat_entry(new);
424 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
425 block_t new_blkaddr, bool fsync_done)
427 struct f2fs_nm_info *nm_i = NM_I(sbi);
429 struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
431 down_write(&nm_i->nat_tree_lock);
432 e = __lookup_nat_cache(nm_i, ni->nid);
434 e = __init_nat_entry(nm_i, new, NULL, true);
435 copy_node_info(&e->ni, ni);
436 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
437 } else if (new_blkaddr == NEW_ADDR) {
439 * when nid is reallocated,
440 * previous nat entry can be remained in nat cache.
441 * So, reinitialize it with new information.
443 copy_node_info(&e->ni, ni);
444 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
446 /* let's free early to reduce memory consumption */
448 __free_nat_entry(new);
451 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
452 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
453 new_blkaddr == NULL_ADDR);
454 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
455 new_blkaddr == NEW_ADDR);
456 f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
457 new_blkaddr == NEW_ADDR);
459 /* increment version no as node is removed */
460 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
461 unsigned char version = nat_get_version(e);
462 nat_set_version(e, inc_node_version(version));
466 nat_set_blkaddr(e, new_blkaddr);
467 if (!__is_valid_data_blkaddr(new_blkaddr))
468 set_nat_flag(e, IS_CHECKPOINTED, false);
469 __set_nat_cache_dirty(nm_i, e);
471 /* update fsync_mark if its inode nat entry is still alive */
472 if (ni->nid != ni->ino)
473 e = __lookup_nat_cache(nm_i, ni->ino);
475 if (fsync_done && ni->nid == ni->ino)
476 set_nat_flag(e, HAS_FSYNCED_INODE, true);
477 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
479 up_write(&nm_i->nat_tree_lock);
482 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
484 struct f2fs_nm_info *nm_i = NM_I(sbi);
487 if (!down_write_trylock(&nm_i->nat_tree_lock))
490 spin_lock(&nm_i->nat_list_lock);
492 struct nat_entry *ne;
494 if (list_empty(&nm_i->nat_entries))
497 ne = list_first_entry(&nm_i->nat_entries,
498 struct nat_entry, list);
500 spin_unlock(&nm_i->nat_list_lock);
502 __del_from_nat_cache(nm_i, ne);
505 spin_lock(&nm_i->nat_list_lock);
507 spin_unlock(&nm_i->nat_list_lock);
509 up_write(&nm_i->nat_tree_lock);
510 return nr - nr_shrink;
513 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
514 struct node_info *ni)
516 struct f2fs_nm_info *nm_i = NM_I(sbi);
517 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
518 struct f2fs_journal *journal = curseg->journal;
519 nid_t start_nid = START_NID(nid);
520 struct f2fs_nat_block *nat_blk;
521 struct page *page = NULL;
522 struct f2fs_nat_entry ne;
530 /* Check nat cache */
531 down_read(&nm_i->nat_tree_lock);
532 e = __lookup_nat_cache(nm_i, nid);
534 ni->ino = nat_get_ino(e);
535 ni->blk_addr = nat_get_blkaddr(e);
536 ni->version = nat_get_version(e);
537 up_read(&nm_i->nat_tree_lock);
541 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
543 /* Check current segment summary */
544 down_read(&curseg->journal_rwsem);
545 i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
547 ne = nat_in_journal(journal, i);
548 node_info_from_raw_nat(ni, &ne);
550 up_read(&curseg->journal_rwsem);
552 up_read(&nm_i->nat_tree_lock);
556 /* Fill node_info from nat page */
557 index = current_nat_addr(sbi, nid);
558 up_read(&nm_i->nat_tree_lock);
560 page = f2fs_get_meta_page(sbi, index);
562 return PTR_ERR(page);
564 nat_blk = (struct f2fs_nat_block *)page_address(page);
565 ne = nat_blk->entries[nid - start_nid];
566 node_info_from_raw_nat(ni, &ne);
567 f2fs_put_page(page, 1);
569 blkaddr = le32_to_cpu(ne.block_addr);
570 if (__is_valid_data_blkaddr(blkaddr) &&
571 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
574 /* cache nat entry */
575 cache_nat_entry(sbi, nid, &ne);
580 * readahead MAX_RA_NODE number of node pages.
582 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
584 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
585 struct blk_plug plug;
589 blk_start_plug(&plug);
591 /* Then, try readahead for siblings of the desired node */
593 end = min(end, NIDS_PER_BLOCK);
594 for (i = start; i < end; i++) {
595 nid = get_nid(parent, i, false);
596 f2fs_ra_node_page(sbi, nid);
599 blk_finish_plug(&plug);
602 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
604 const long direct_index = ADDRS_PER_INODE(dn->inode);
605 const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
606 const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
607 unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
608 int cur_level = dn->cur_level;
609 int max_level = dn->max_level;
615 while (max_level-- > cur_level)
616 skipped_unit *= NIDS_PER_BLOCK;
618 switch (dn->max_level) {
620 base += 2 * indirect_blks;
623 base += 2 * direct_blks;
626 base += direct_index;
629 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
632 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
636 * The maximum depth is four.
637 * Offset[0] will have raw inode offset.
639 static int get_node_path(struct inode *inode, long block,
640 int offset[4], unsigned int noffset[4])
642 const long direct_index = ADDRS_PER_INODE(inode);
643 const long direct_blks = ADDRS_PER_BLOCK(inode);
644 const long dptrs_per_blk = NIDS_PER_BLOCK;
645 const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
646 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
652 if (block < direct_index) {
656 block -= direct_index;
657 if (block < direct_blks) {
658 offset[n++] = NODE_DIR1_BLOCK;
664 block -= direct_blks;
665 if (block < direct_blks) {
666 offset[n++] = NODE_DIR2_BLOCK;
672 block -= direct_blks;
673 if (block < indirect_blks) {
674 offset[n++] = NODE_IND1_BLOCK;
676 offset[n++] = block / direct_blks;
677 noffset[n] = 4 + offset[n - 1];
678 offset[n] = block % direct_blks;
682 block -= indirect_blks;
683 if (block < indirect_blks) {
684 offset[n++] = NODE_IND2_BLOCK;
685 noffset[n] = 4 + dptrs_per_blk;
686 offset[n++] = block / direct_blks;
687 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
688 offset[n] = block % direct_blks;
692 block -= indirect_blks;
693 if (block < dindirect_blks) {
694 offset[n++] = NODE_DIND_BLOCK;
695 noffset[n] = 5 + (dptrs_per_blk * 2);
696 offset[n++] = block / indirect_blks;
697 noffset[n] = 6 + (dptrs_per_blk * 2) +
698 offset[n - 1] * (dptrs_per_blk + 1);
699 offset[n++] = (block / direct_blks) % dptrs_per_blk;
700 noffset[n] = 7 + (dptrs_per_blk * 2) +
701 offset[n - 2] * (dptrs_per_blk + 1) +
703 offset[n] = block % direct_blks;
714 * Caller should call f2fs_put_dnode(dn).
715 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
716 * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
718 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
720 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
721 struct page *npage[4];
722 struct page *parent = NULL;
724 unsigned int noffset[4];
729 level = get_node_path(dn->inode, index, offset, noffset);
733 nids[0] = dn->inode->i_ino;
734 npage[0] = dn->inode_page;
737 npage[0] = f2fs_get_node_page(sbi, nids[0]);
738 if (IS_ERR(npage[0]))
739 return PTR_ERR(npage[0]);
742 /* if inline_data is set, should not report any block indices */
743 if (f2fs_has_inline_data(dn->inode) && index) {
745 f2fs_put_page(npage[0], 1);
751 nids[1] = get_nid(parent, offset[0], true);
752 dn->inode_page = npage[0];
753 dn->inode_page_locked = true;
755 /* get indirect or direct nodes */
756 for (i = 1; i <= level; i++) {
759 if (!nids[i] && mode == ALLOC_NODE) {
761 if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
767 npage[i] = f2fs_new_node_page(dn, noffset[i]);
768 if (IS_ERR(npage[i])) {
769 f2fs_alloc_nid_failed(sbi, nids[i]);
770 err = PTR_ERR(npage[i]);
774 set_nid(parent, offset[i - 1], nids[i], i == 1);
775 f2fs_alloc_nid_done(sbi, nids[i]);
777 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
778 npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
779 if (IS_ERR(npage[i])) {
780 err = PTR_ERR(npage[i]);
786 dn->inode_page_locked = false;
789 f2fs_put_page(parent, 1);
793 npage[i] = f2fs_get_node_page(sbi, nids[i]);
794 if (IS_ERR(npage[i])) {
795 err = PTR_ERR(npage[i]);
796 f2fs_put_page(npage[0], 0);
802 nids[i + 1] = get_nid(parent, offset[i], false);
805 dn->nid = nids[level];
806 dn->ofs_in_node = offset[level];
807 dn->node_page = npage[level];
808 dn->data_blkaddr = f2fs_data_blkaddr(dn);
812 f2fs_put_page(parent, 1);
814 f2fs_put_page(npage[0], 0);
816 dn->inode_page = NULL;
817 dn->node_page = NULL;
818 if (err == -ENOENT) {
820 dn->max_level = level;
821 dn->ofs_in_node = offset[level];
826 static int truncate_node(struct dnode_of_data *dn)
828 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
833 err = f2fs_get_node_info(sbi, dn->nid, &ni);
837 /* Deallocate node address */
838 f2fs_invalidate_blocks(sbi, ni.blk_addr);
839 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
840 set_node_addr(sbi, &ni, NULL_ADDR, false);
842 if (dn->nid == dn->inode->i_ino) {
843 f2fs_remove_orphan_inode(sbi, dn->nid);
844 dec_valid_inode_count(sbi);
845 f2fs_inode_synced(dn->inode);
848 clear_node_page_dirty(dn->node_page);
849 set_sbi_flag(sbi, SBI_IS_DIRTY);
851 index = dn->node_page->index;
852 f2fs_put_page(dn->node_page, 1);
854 invalidate_mapping_pages(NODE_MAPPING(sbi),
857 dn->node_page = NULL;
858 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
863 static int truncate_dnode(struct dnode_of_data *dn)
871 /* get direct node */
872 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
873 if (PTR_ERR(page) == -ENOENT)
875 else if (IS_ERR(page))
876 return PTR_ERR(page);
878 /* Make dnode_of_data for parameter */
879 dn->node_page = page;
881 f2fs_truncate_data_blocks(dn);
882 err = truncate_node(dn);
889 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
892 struct dnode_of_data rdn = *dn;
894 struct f2fs_node *rn;
896 unsigned int child_nofs;
901 return NIDS_PER_BLOCK + 1;
903 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
905 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
907 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
908 return PTR_ERR(page);
911 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
913 rn = F2FS_NODE(page);
915 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
916 child_nid = le32_to_cpu(rn->in.nid[i]);
920 ret = truncate_dnode(&rdn);
923 if (set_nid(page, i, 0, false))
924 dn->node_changed = true;
927 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
928 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
929 child_nid = le32_to_cpu(rn->in.nid[i]);
930 if (child_nid == 0) {
931 child_nofs += NIDS_PER_BLOCK + 1;
935 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
936 if (ret == (NIDS_PER_BLOCK + 1)) {
937 if (set_nid(page, i, 0, false))
938 dn->node_changed = true;
940 } else if (ret < 0 && ret != -ENOENT) {
948 /* remove current indirect node */
949 dn->node_page = page;
950 ret = truncate_node(dn);
955 f2fs_put_page(page, 1);
957 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
961 f2fs_put_page(page, 1);
962 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
966 static int truncate_partial_nodes(struct dnode_of_data *dn,
967 struct f2fs_inode *ri, int *offset, int depth)
969 struct page *pages[2];
976 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
980 /* get indirect nodes in the path */
981 for (i = 0; i < idx + 1; i++) {
982 /* reference count'll be increased */
983 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
984 if (IS_ERR(pages[i])) {
985 err = PTR_ERR(pages[i]);
989 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
992 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
994 /* free direct nodes linked to a partial indirect node */
995 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
996 child_nid = get_nid(pages[idx], i, false);
1000 err = truncate_dnode(dn);
1003 if (set_nid(pages[idx], i, 0, false))
1004 dn->node_changed = true;
1007 if (offset[idx + 1] == 0) {
1008 dn->node_page = pages[idx];
1010 err = truncate_node(dn);
1014 f2fs_put_page(pages[idx], 1);
1017 offset[idx + 1] = 0;
1020 for (i = idx; i >= 0; i--)
1021 f2fs_put_page(pages[i], 1);
1023 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1029 * All the block addresses of data and nodes should be nullified.
1031 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1033 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1034 int err = 0, cont = 1;
1035 int level, offset[4], noffset[4];
1036 unsigned int nofs = 0;
1037 struct f2fs_inode *ri;
1038 struct dnode_of_data dn;
1041 trace_f2fs_truncate_inode_blocks_enter(inode, from);
1043 level = get_node_path(inode, from, offset, noffset);
1047 page = f2fs_get_node_page(sbi, inode->i_ino);
1049 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1050 return PTR_ERR(page);
1053 set_new_dnode(&dn, inode, page, NULL, 0);
1056 ri = F2FS_INODE(page);
1064 if (!offset[level - 1])
1066 err = truncate_partial_nodes(&dn, ri, offset, level);
1067 if (err < 0 && err != -ENOENT)
1069 nofs += 1 + NIDS_PER_BLOCK;
1072 nofs = 5 + 2 * NIDS_PER_BLOCK;
1073 if (!offset[level - 1])
1075 err = truncate_partial_nodes(&dn, ri, offset, level);
1076 if (err < 0 && err != -ENOENT)
1085 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1086 switch (offset[0]) {
1087 case NODE_DIR1_BLOCK:
1088 case NODE_DIR2_BLOCK:
1089 err = truncate_dnode(&dn);
1092 case NODE_IND1_BLOCK:
1093 case NODE_IND2_BLOCK:
1094 err = truncate_nodes(&dn, nofs, offset[1], 2);
1097 case NODE_DIND_BLOCK:
1098 err = truncate_nodes(&dn, nofs, offset[1], 3);
1105 if (err < 0 && err != -ENOENT)
1107 if (offset[1] == 0 &&
1108 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1110 BUG_ON(page->mapping != NODE_MAPPING(sbi));
1111 f2fs_wait_on_page_writeback(page, NODE, true, true);
1112 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1113 set_page_dirty(page);
1121 f2fs_put_page(page, 0);
1122 trace_f2fs_truncate_inode_blocks_exit(inode, err);
1123 return err > 0 ? 0 : err;
1126 /* caller must lock inode page */
1127 int f2fs_truncate_xattr_node(struct inode *inode)
1129 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1130 nid_t nid = F2FS_I(inode)->i_xattr_nid;
1131 struct dnode_of_data dn;
1138 npage = f2fs_get_node_page(sbi, nid);
1140 return PTR_ERR(npage);
1142 set_new_dnode(&dn, inode, NULL, npage, nid);
1143 err = truncate_node(&dn);
1145 f2fs_put_page(npage, 1);
1149 f2fs_i_xnid_write(inode, 0);
1155 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1158 int f2fs_remove_inode_page(struct inode *inode)
1160 struct dnode_of_data dn;
1163 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1164 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1168 err = f2fs_truncate_xattr_node(inode);
1170 f2fs_put_dnode(&dn);
1174 /* remove potential inline_data blocks */
1175 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1176 S_ISLNK(inode->i_mode))
1177 f2fs_truncate_data_blocks_range(&dn, 1);
1179 /* 0 is possible, after f2fs_new_inode() has failed */
1180 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1181 f2fs_put_dnode(&dn);
1185 if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1186 f2fs_warn(F2FS_I_SB(inode),
1187 "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1188 inode->i_ino, (unsigned long long)inode->i_blocks);
1189 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1192 /* will put inode & node pages */
1193 err = truncate_node(&dn);
1195 f2fs_put_dnode(&dn);
1201 struct page *f2fs_new_inode_page(struct inode *inode)
1203 struct dnode_of_data dn;
1205 /* allocate inode page for new inode */
1206 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1208 /* caller should f2fs_put_page(page, 1); */
1209 return f2fs_new_node_page(&dn, 0);
1212 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1214 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1215 struct node_info new_ni;
1219 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1220 return ERR_PTR(-EPERM);
1222 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1224 return ERR_PTR(-ENOMEM);
1226 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1229 #ifdef CONFIG_F2FS_CHECK_FS
1230 err = f2fs_get_node_info(sbi, dn->nid, &new_ni);
1232 dec_valid_node_count(sbi, dn->inode, !ofs);
1235 f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1237 new_ni.nid = dn->nid;
1238 new_ni.ino = dn->inode->i_ino;
1239 new_ni.blk_addr = NULL_ADDR;
1242 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1244 f2fs_wait_on_page_writeback(page, NODE, true, true);
1245 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1246 set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1247 if (!PageUptodate(page))
1248 SetPageUptodate(page);
1249 if (set_page_dirty(page))
1250 dn->node_changed = true;
1252 if (f2fs_has_xattr_block(ofs))
1253 f2fs_i_xnid_write(dn->inode, dn->nid);
1256 inc_valid_inode_count(sbi);
1260 clear_node_page_dirty(page);
1261 f2fs_put_page(page, 1);
1262 return ERR_PTR(err);
1266 * Caller should do after getting the following values.
1267 * 0: f2fs_put_page(page, 0)
1268 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1270 static int read_node_page(struct page *page, int op_flags)
1272 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1273 struct node_info ni;
1274 struct f2fs_io_info fio = {
1278 .op_flags = op_flags,
1280 .encrypted_page = NULL,
1284 if (PageUptodate(page)) {
1285 if (!f2fs_inode_chksum_verify(sbi, page)) {
1286 ClearPageUptodate(page);
1292 err = f2fs_get_node_info(sbi, page->index, &ni);
1296 if (unlikely(ni.blk_addr == NULL_ADDR) ||
1297 is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1298 ClearPageUptodate(page);
1302 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1303 return f2fs_submit_page_bio(&fio);
1307 * Readahead a node page
1309 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1316 if (f2fs_check_nid_range(sbi, nid))
1319 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1323 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1327 err = read_node_page(apage, REQ_RAHEAD);
1328 f2fs_put_page(apage, err ? 1 : 0);
1331 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1332 struct page *parent, int start)
1338 return ERR_PTR(-ENOENT);
1339 if (f2fs_check_nid_range(sbi, nid))
1340 return ERR_PTR(-EINVAL);
1342 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1344 return ERR_PTR(-ENOMEM);
1346 err = read_node_page(page, 0);
1348 f2fs_put_page(page, 1);
1349 return ERR_PTR(err);
1350 } else if (err == LOCKED_PAGE) {
1356 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1360 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1361 f2fs_put_page(page, 1);
1365 if (unlikely(!PageUptodate(page))) {
1370 if (!f2fs_inode_chksum_verify(sbi, page)) {
1375 if(unlikely(nid != nid_of_node(page))) {
1376 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1377 nid, nid_of_node(page), ino_of_node(page),
1378 ofs_of_node(page), cpver_of_node(page),
1379 next_blkaddr_of_node(page));
1382 ClearPageUptodate(page);
1383 f2fs_put_page(page, 1);
1384 return ERR_PTR(err);
1389 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1391 return __get_node_page(sbi, nid, NULL, 0);
1394 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1396 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1397 nid_t nid = get_nid(parent, start, false);
1399 return __get_node_page(sbi, nid, parent, start);
1402 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1404 struct inode *inode;
1408 /* should flush inline_data before evict_inode */
1409 inode = ilookup(sbi->sb, ino);
1413 page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1414 FGP_LOCK|FGP_NOWAIT, 0);
1418 if (!PageUptodate(page))
1421 if (!PageDirty(page))
1424 if (!clear_page_dirty_for_io(page))
1427 ret = f2fs_write_inline_data(inode, page);
1428 inode_dec_dirty_pages(inode);
1429 f2fs_remove_dirty_inode(inode);
1431 set_page_dirty(page);
1433 f2fs_put_page(page, 1);
1438 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1441 struct pagevec pvec;
1442 struct page *last_page = NULL;
1445 pagevec_init(&pvec);
1448 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1449 PAGECACHE_TAG_DIRTY))) {
1452 for (i = 0; i < nr_pages; i++) {
1453 struct page *page = pvec.pages[i];
1455 if (unlikely(f2fs_cp_error(sbi))) {
1456 f2fs_put_page(last_page, 0);
1457 pagevec_release(&pvec);
1458 return ERR_PTR(-EIO);
1461 if (!IS_DNODE(page) || !is_cold_node(page))
1463 if (ino_of_node(page) != ino)
1468 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1473 if (ino_of_node(page) != ino)
1474 goto continue_unlock;
1476 if (!PageDirty(page)) {
1477 /* someone wrote it for us */
1478 goto continue_unlock;
1482 f2fs_put_page(last_page, 0);
1488 pagevec_release(&pvec);
1494 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1495 struct writeback_control *wbc, bool do_balance,
1496 enum iostat_type io_type, unsigned int *seq_id)
1498 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1500 struct node_info ni;
1501 struct f2fs_io_info fio = {
1503 .ino = ino_of_node(page),
1506 .op_flags = wbc_to_write_flags(wbc),
1508 .encrypted_page = NULL,
1515 trace_f2fs_writepage(page, NODE);
1517 if (unlikely(f2fs_cp_error(sbi)))
1520 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1523 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1524 wbc->sync_mode == WB_SYNC_NONE &&
1525 IS_DNODE(page) && is_cold_node(page))
1528 /* get old block addr of this node page */
1529 nid = nid_of_node(page);
1530 f2fs_bug_on(sbi, page->index != nid);
1532 if (f2fs_get_node_info(sbi, nid, &ni))
1535 if (wbc->for_reclaim) {
1536 if (!down_read_trylock(&sbi->node_write))
1539 down_read(&sbi->node_write);
1542 /* This page is already truncated */
1543 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1544 ClearPageUptodate(page);
1545 dec_page_count(sbi, F2FS_DIRTY_NODES);
1546 up_read(&sbi->node_write);
1551 if (__is_valid_data_blkaddr(ni.blk_addr) &&
1552 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1553 DATA_GENERIC_ENHANCE)) {
1554 up_read(&sbi->node_write);
1558 if (atomic && !test_opt(sbi, NOBARRIER))
1559 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1561 /* should add to global list before clearing PAGECACHE status */
1562 if (f2fs_in_warm_node_list(sbi, page)) {
1563 seq = f2fs_add_fsync_node_entry(sbi, page);
1568 set_page_writeback(page);
1569 ClearPageError(page);
1571 fio.old_blkaddr = ni.blk_addr;
1572 f2fs_do_write_node_page(nid, &fio);
1573 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1574 dec_page_count(sbi, F2FS_DIRTY_NODES);
1575 up_read(&sbi->node_write);
1577 if (wbc->for_reclaim) {
1578 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1584 if (unlikely(f2fs_cp_error(sbi))) {
1585 f2fs_submit_merged_write(sbi, NODE);
1589 *submitted = fio.submitted;
1592 f2fs_balance_fs(sbi, false);
1596 redirty_page_for_writepage(wbc, page);
1597 return AOP_WRITEPAGE_ACTIVATE;
1600 int f2fs_move_node_page(struct page *node_page, int gc_type)
1604 if (gc_type == FG_GC) {
1605 struct writeback_control wbc = {
1606 .sync_mode = WB_SYNC_ALL,
1611 f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1613 set_page_dirty(node_page);
1615 if (!clear_page_dirty_for_io(node_page)) {
1620 if (__write_node_page(node_page, false, NULL,
1621 &wbc, false, FS_GC_NODE_IO, NULL)) {
1623 unlock_page(node_page);
1627 /* set page dirty and write it */
1628 if (!PageWriteback(node_page))
1629 set_page_dirty(node_page);
1632 unlock_page(node_page);
1634 f2fs_put_page(node_page, 0);
1638 static int f2fs_write_node_page(struct page *page,
1639 struct writeback_control *wbc)
1641 return __write_node_page(page, false, NULL, wbc, false,
1645 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1646 struct writeback_control *wbc, bool atomic,
1647 unsigned int *seq_id)
1650 struct pagevec pvec;
1652 struct page *last_page = NULL;
1653 bool marked = false;
1654 nid_t ino = inode->i_ino;
1659 last_page = last_fsync_dnode(sbi, ino);
1660 if (IS_ERR_OR_NULL(last_page))
1661 return PTR_ERR_OR_ZERO(last_page);
1664 pagevec_init(&pvec);
1667 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1668 PAGECACHE_TAG_DIRTY))) {
1671 for (i = 0; i < nr_pages; i++) {
1672 struct page *page = pvec.pages[i];
1673 bool submitted = false;
1675 if (unlikely(f2fs_cp_error(sbi))) {
1676 f2fs_put_page(last_page, 0);
1677 pagevec_release(&pvec);
1682 if (!IS_DNODE(page) || !is_cold_node(page))
1684 if (ino_of_node(page) != ino)
1689 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1694 if (ino_of_node(page) != ino)
1695 goto continue_unlock;
1697 if (!PageDirty(page) && page != last_page) {
1698 /* someone wrote it for us */
1699 goto continue_unlock;
1702 f2fs_wait_on_page_writeback(page, NODE, true, true);
1704 set_fsync_mark(page, 0);
1705 set_dentry_mark(page, 0);
1707 if (!atomic || page == last_page) {
1708 set_fsync_mark(page, 1);
1709 if (IS_INODE(page)) {
1710 if (is_inode_flag_set(inode,
1712 f2fs_update_inode(inode, page);
1713 set_dentry_mark(page,
1714 f2fs_need_dentry_mark(sbi, ino));
1716 /* may be written by other thread */
1717 if (!PageDirty(page))
1718 set_page_dirty(page);
1721 if (!clear_page_dirty_for_io(page))
1722 goto continue_unlock;
1724 ret = __write_node_page(page, atomic &&
1726 &submitted, wbc, true,
1727 FS_NODE_IO, seq_id);
1730 f2fs_put_page(last_page, 0);
1732 } else if (submitted) {
1736 if (page == last_page) {
1737 f2fs_put_page(page, 0);
1742 pagevec_release(&pvec);
1748 if (!ret && atomic && !marked) {
1749 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1750 ino, last_page->index);
1751 lock_page(last_page);
1752 f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1753 set_page_dirty(last_page);
1754 unlock_page(last_page);
1759 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1760 return ret ? -EIO: 0;
1763 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1765 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1768 if (inode->i_ino != ino)
1771 if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1774 spin_lock(&sbi->inode_lock[DIRTY_META]);
1775 clean = list_empty(&F2FS_I(inode)->gdirty_list);
1776 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1781 inode = igrab(inode);
1787 static bool flush_dirty_inode(struct page *page)
1789 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1790 struct inode *inode;
1791 nid_t ino = ino_of_node(page);
1793 inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1797 f2fs_update_inode(inode, page);
1804 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1805 struct writeback_control *wbc,
1806 bool do_balance, enum iostat_type io_type)
1809 struct pagevec pvec;
1813 int nr_pages, done = 0;
1815 pagevec_init(&pvec);
1820 while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1821 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1824 for (i = 0; i < nr_pages; i++) {
1825 struct page *page = pvec.pages[i];
1826 bool submitted = false;
1827 bool may_dirty = true;
1829 /* give a priority to WB_SYNC threads */
1830 if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1831 wbc->sync_mode == WB_SYNC_NONE) {
1837 * flushing sequence with step:
1842 if (step == 0 && IS_DNODE(page))
1844 if (step == 1 && (!IS_DNODE(page) ||
1845 is_cold_node(page)))
1847 if (step == 2 && (!IS_DNODE(page) ||
1848 !is_cold_node(page)))
1851 if (wbc->sync_mode == WB_SYNC_ALL)
1853 else if (!trylock_page(page))
1856 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1862 if (!PageDirty(page)) {
1863 /* someone wrote it for us */
1864 goto continue_unlock;
1867 /* flush inline_data */
1868 if (is_inline_node(page)) {
1869 clear_inline_node(page);
1871 flush_inline_data(sbi, ino_of_node(page));
1875 /* flush dirty inode */
1876 if (IS_INODE(page) && may_dirty) {
1878 if (flush_dirty_inode(page))
1882 f2fs_wait_on_page_writeback(page, NODE, true, true);
1884 if (!clear_page_dirty_for_io(page))
1885 goto continue_unlock;
1887 set_fsync_mark(page, 0);
1888 set_dentry_mark(page, 0);
1890 ret = __write_node_page(page, false, &submitted,
1891 wbc, do_balance, io_type, NULL);
1897 if (--wbc->nr_to_write == 0)
1900 pagevec_release(&pvec);
1903 if (wbc->nr_to_write == 0) {
1910 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1911 wbc->sync_mode == WB_SYNC_NONE && step == 1)
1918 f2fs_submit_merged_write(sbi, NODE);
1920 if (unlikely(f2fs_cp_error(sbi)))
1925 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
1926 unsigned int seq_id)
1928 struct fsync_node_entry *fn;
1930 struct list_head *head = &sbi->fsync_node_list;
1931 unsigned long flags;
1932 unsigned int cur_seq_id = 0;
1935 while (seq_id && cur_seq_id < seq_id) {
1936 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
1937 if (list_empty(head)) {
1938 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1941 fn = list_first_entry(head, struct fsync_node_entry, list);
1942 if (fn->seq_id > seq_id) {
1943 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1946 cur_seq_id = fn->seq_id;
1949 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1951 f2fs_wait_on_page_writeback(page, NODE, true, false);
1952 if (TestClearPageError(page))
1961 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1968 static int f2fs_write_node_pages(struct address_space *mapping,
1969 struct writeback_control *wbc)
1971 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1972 struct blk_plug plug;
1975 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1978 /* balancing f2fs's metadata in background */
1979 f2fs_balance_fs_bg(sbi, true);
1981 /* collect a number of dirty node pages and write together */
1982 if (wbc->sync_mode != WB_SYNC_ALL &&
1983 get_pages(sbi, F2FS_DIRTY_NODES) <
1984 nr_pages_to_skip(sbi, NODE))
1987 if (wbc->sync_mode == WB_SYNC_ALL)
1988 atomic_inc(&sbi->wb_sync_req[NODE]);
1989 else if (atomic_read(&sbi->wb_sync_req[NODE]))
1992 trace_f2fs_writepages(mapping->host, wbc, NODE);
1994 diff = nr_pages_to_write(sbi, NODE, wbc);
1995 blk_start_plug(&plug);
1996 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
1997 blk_finish_plug(&plug);
1998 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2000 if (wbc->sync_mode == WB_SYNC_ALL)
2001 atomic_dec(&sbi->wb_sync_req[NODE]);
2005 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2006 trace_f2fs_writepages(mapping->host, wbc, NODE);
2010 static int f2fs_set_node_page_dirty(struct page *page)
2012 trace_f2fs_set_page_dirty(page, NODE);
2014 if (!PageUptodate(page))
2015 SetPageUptodate(page);
2016 #ifdef CONFIG_F2FS_CHECK_FS
2018 f2fs_inode_chksum_set(F2FS_P_SB(page), page);
2020 if (!PageDirty(page)) {
2021 __set_page_dirty_nobuffers(page);
2022 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
2023 f2fs_set_page_private(page, 0);
2024 f2fs_trace_pid(page);
2031 * Structure of the f2fs node operations
2033 const struct address_space_operations f2fs_node_aops = {
2034 .writepage = f2fs_write_node_page,
2035 .writepages = f2fs_write_node_pages,
2036 .set_page_dirty = f2fs_set_node_page_dirty,
2037 .invalidatepage = f2fs_invalidate_page,
2038 .releasepage = f2fs_release_page,
2039 #ifdef CONFIG_MIGRATION
2040 .migratepage = f2fs_migrate_page,
2044 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2047 return radix_tree_lookup(&nm_i->free_nid_root, n);
2050 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2051 struct free_nid *i, enum nid_state state)
2053 struct f2fs_nm_info *nm_i = NM_I(sbi);
2055 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2059 f2fs_bug_on(sbi, state != i->state);
2060 nm_i->nid_cnt[state]++;
2061 if (state == FREE_NID)
2062 list_add_tail(&i->list, &nm_i->free_nid_list);
2066 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2067 struct free_nid *i, enum nid_state state)
2069 struct f2fs_nm_info *nm_i = NM_I(sbi);
2071 f2fs_bug_on(sbi, state != i->state);
2072 nm_i->nid_cnt[state]--;
2073 if (state == FREE_NID)
2075 radix_tree_delete(&nm_i->free_nid_root, i->nid);
2078 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2079 enum nid_state org_state, enum nid_state dst_state)
2081 struct f2fs_nm_info *nm_i = NM_I(sbi);
2083 f2fs_bug_on(sbi, org_state != i->state);
2084 i->state = dst_state;
2085 nm_i->nid_cnt[org_state]--;
2086 nm_i->nid_cnt[dst_state]++;
2088 switch (dst_state) {
2093 list_add_tail(&i->list, &nm_i->free_nid_list);
2100 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2101 bool set, bool build)
2103 struct f2fs_nm_info *nm_i = NM_I(sbi);
2104 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2105 unsigned int nid_ofs = nid - START_NID(nid);
2107 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2111 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2113 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2114 nm_i->free_nid_count[nat_ofs]++;
2116 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2118 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2120 nm_i->free_nid_count[nat_ofs]--;
2124 /* return if the nid is recognized as free */
2125 static bool add_free_nid(struct f2fs_sb_info *sbi,
2126 nid_t nid, bool build, bool update)
2128 struct f2fs_nm_info *nm_i = NM_I(sbi);
2129 struct free_nid *i, *e;
2130 struct nat_entry *ne;
2134 /* 0 nid should not be used */
2135 if (unlikely(nid == 0))
2138 if (unlikely(f2fs_check_nid_range(sbi, nid)))
2141 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2143 i->state = FREE_NID;
2145 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2147 spin_lock(&nm_i->nid_list_lock);
2155 * - __insert_nid_to_list(PREALLOC_NID)
2156 * - f2fs_balance_fs_bg
2157 * - f2fs_build_free_nids
2158 * - __f2fs_build_free_nids
2161 * - __lookup_nat_cache
2163 * - f2fs_init_inode_metadata
2164 * - f2fs_new_inode_page
2165 * - f2fs_new_node_page
2167 * - f2fs_alloc_nid_done
2168 * - __remove_nid_from_list(PREALLOC_NID)
2169 * - __insert_nid_to_list(FREE_NID)
2171 ne = __lookup_nat_cache(nm_i, nid);
2172 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2173 nat_get_blkaddr(ne) != NULL_ADDR))
2176 e = __lookup_free_nid_list(nm_i, nid);
2178 if (e->state == FREE_NID)
2184 err = __insert_free_nid(sbi, i, FREE_NID);
2187 update_free_nid_bitmap(sbi, nid, ret, build);
2189 nm_i->available_nids++;
2191 spin_unlock(&nm_i->nid_list_lock);
2192 radix_tree_preload_end();
2195 kmem_cache_free(free_nid_slab, i);
2199 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2201 struct f2fs_nm_info *nm_i = NM_I(sbi);
2203 bool need_free = false;
2205 spin_lock(&nm_i->nid_list_lock);
2206 i = __lookup_free_nid_list(nm_i, nid);
2207 if (i && i->state == FREE_NID) {
2208 __remove_free_nid(sbi, i, FREE_NID);
2211 spin_unlock(&nm_i->nid_list_lock);
2214 kmem_cache_free(free_nid_slab, i);
2217 static int scan_nat_page(struct f2fs_sb_info *sbi,
2218 struct page *nat_page, nid_t start_nid)
2220 struct f2fs_nm_info *nm_i = NM_I(sbi);
2221 struct f2fs_nat_block *nat_blk = page_address(nat_page);
2223 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2226 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2228 i = start_nid % NAT_ENTRY_PER_BLOCK;
2230 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2231 if (unlikely(start_nid >= nm_i->max_nid))
2234 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2236 if (blk_addr == NEW_ADDR)
2239 if (blk_addr == NULL_ADDR) {
2240 add_free_nid(sbi, start_nid, true, true);
2242 spin_lock(&NM_I(sbi)->nid_list_lock);
2243 update_free_nid_bitmap(sbi, start_nid, false, true);
2244 spin_unlock(&NM_I(sbi)->nid_list_lock);
2251 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2253 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2254 struct f2fs_journal *journal = curseg->journal;
2257 down_read(&curseg->journal_rwsem);
2258 for (i = 0; i < nats_in_cursum(journal); i++) {
2262 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2263 nid = le32_to_cpu(nid_in_journal(journal, i));
2264 if (addr == NULL_ADDR)
2265 add_free_nid(sbi, nid, true, false);
2267 remove_free_nid(sbi, nid);
2269 up_read(&curseg->journal_rwsem);
2272 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2274 struct f2fs_nm_info *nm_i = NM_I(sbi);
2275 unsigned int i, idx;
2278 down_read(&nm_i->nat_tree_lock);
2280 for (i = 0; i < nm_i->nat_blocks; i++) {
2281 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2283 if (!nm_i->free_nid_count[i])
2285 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2286 idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2287 NAT_ENTRY_PER_BLOCK, idx);
2288 if (idx >= NAT_ENTRY_PER_BLOCK)
2291 nid = i * NAT_ENTRY_PER_BLOCK + idx;
2292 add_free_nid(sbi, nid, true, false);
2294 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2299 scan_curseg_cache(sbi);
2301 up_read(&nm_i->nat_tree_lock);
2304 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2305 bool sync, bool mount)
2307 struct f2fs_nm_info *nm_i = NM_I(sbi);
2309 nid_t nid = nm_i->next_scan_nid;
2311 if (unlikely(nid >= nm_i->max_nid))
2314 /* Enough entries */
2315 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2318 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2322 /* try to find free nids in free_nid_bitmap */
2323 scan_free_nid_bits(sbi);
2325 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2329 /* readahead nat pages to be scanned */
2330 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2333 down_read(&nm_i->nat_tree_lock);
2336 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2337 nm_i->nat_block_bitmap)) {
2338 struct page *page = get_current_nat_page(sbi, nid);
2341 ret = PTR_ERR(page);
2343 ret = scan_nat_page(sbi, page, nid);
2344 f2fs_put_page(page, 1);
2348 up_read(&nm_i->nat_tree_lock);
2349 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2354 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2355 if (unlikely(nid >= nm_i->max_nid))
2358 if (++i >= FREE_NID_PAGES)
2362 /* go to the next free nat pages to find free nids abundantly */
2363 nm_i->next_scan_nid = nid;
2365 /* find free nids from current sum_pages */
2366 scan_curseg_cache(sbi);
2368 up_read(&nm_i->nat_tree_lock);
2370 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2371 nm_i->ra_nid_pages, META_NAT, false);
2376 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2380 mutex_lock(&NM_I(sbi)->build_lock);
2381 ret = __f2fs_build_free_nids(sbi, sync, mount);
2382 mutex_unlock(&NM_I(sbi)->build_lock);
2388 * If this function returns success, caller can obtain a new nid
2389 * from second parameter of this function.
2390 * The returned nid could be used ino as well as nid when inode is created.
2392 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2394 struct f2fs_nm_info *nm_i = NM_I(sbi);
2395 struct free_nid *i = NULL;
2397 if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2398 f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2402 spin_lock(&nm_i->nid_list_lock);
2404 if (unlikely(nm_i->available_nids == 0)) {
2405 spin_unlock(&nm_i->nid_list_lock);
2409 /* We should not use stale free nids created by f2fs_build_free_nids */
2410 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2411 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2412 i = list_first_entry(&nm_i->free_nid_list,
2413 struct free_nid, list);
2416 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2417 nm_i->available_nids--;
2419 update_free_nid_bitmap(sbi, *nid, false, false);
2421 spin_unlock(&nm_i->nid_list_lock);
2424 spin_unlock(&nm_i->nid_list_lock);
2426 /* Let's scan nat pages and its caches to get free nids */
2427 if (!f2fs_build_free_nids(sbi, true, false))
2433 * f2fs_alloc_nid() should be called prior to this function.
2435 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2437 struct f2fs_nm_info *nm_i = NM_I(sbi);
2440 spin_lock(&nm_i->nid_list_lock);
2441 i = __lookup_free_nid_list(nm_i, nid);
2442 f2fs_bug_on(sbi, !i);
2443 __remove_free_nid(sbi, i, PREALLOC_NID);
2444 spin_unlock(&nm_i->nid_list_lock);
2446 kmem_cache_free(free_nid_slab, i);
2450 * f2fs_alloc_nid() should be called prior to this function.
2452 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2454 struct f2fs_nm_info *nm_i = NM_I(sbi);
2456 bool need_free = false;
2461 spin_lock(&nm_i->nid_list_lock);
2462 i = __lookup_free_nid_list(nm_i, nid);
2463 f2fs_bug_on(sbi, !i);
2465 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2466 __remove_free_nid(sbi, i, PREALLOC_NID);
2469 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2472 nm_i->available_nids++;
2474 update_free_nid_bitmap(sbi, nid, true, false);
2476 spin_unlock(&nm_i->nid_list_lock);
2479 kmem_cache_free(free_nid_slab, i);
2482 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2484 struct f2fs_nm_info *nm_i = NM_I(sbi);
2485 struct free_nid *i, *next;
2488 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2491 if (!mutex_trylock(&nm_i->build_lock))
2494 spin_lock(&nm_i->nid_list_lock);
2495 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2496 if (nr_shrink <= 0 ||
2497 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2500 __remove_free_nid(sbi, i, FREE_NID);
2501 kmem_cache_free(free_nid_slab, i);
2504 spin_unlock(&nm_i->nid_list_lock);
2505 mutex_unlock(&nm_i->build_lock);
2507 return nr - nr_shrink;
2510 void f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2512 void *src_addr, *dst_addr;
2515 struct f2fs_inode *ri;
2517 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2518 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2520 ri = F2FS_INODE(page);
2521 if (ri->i_inline & F2FS_INLINE_XATTR) {
2522 set_inode_flag(inode, FI_INLINE_XATTR);
2524 clear_inode_flag(inode, FI_INLINE_XATTR);
2528 dst_addr = inline_xattr_addr(inode, ipage);
2529 src_addr = inline_xattr_addr(inode, page);
2530 inline_size = inline_xattr_size(inode);
2532 f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2533 memcpy(dst_addr, src_addr, inline_size);
2535 f2fs_update_inode(inode, ipage);
2536 f2fs_put_page(ipage, 1);
2539 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2541 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2542 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2544 struct dnode_of_data dn;
2545 struct node_info ni;
2552 /* 1: invalidate the previous xattr nid */
2553 err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2557 f2fs_invalidate_blocks(sbi, ni.blk_addr);
2558 dec_valid_node_count(sbi, inode, false);
2559 set_node_addr(sbi, &ni, NULL_ADDR, false);
2562 /* 2: update xattr nid in inode */
2563 if (!f2fs_alloc_nid(sbi, &new_xnid))
2566 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2567 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2568 if (IS_ERR(xpage)) {
2569 f2fs_alloc_nid_failed(sbi, new_xnid);
2570 return PTR_ERR(xpage);
2573 f2fs_alloc_nid_done(sbi, new_xnid);
2574 f2fs_update_inode_page(inode);
2576 /* 3: update and set xattr node page dirty */
2577 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2579 set_page_dirty(xpage);
2580 f2fs_put_page(xpage, 1);
2585 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2587 struct f2fs_inode *src, *dst;
2588 nid_t ino = ino_of_node(page);
2589 struct node_info old_ni, new_ni;
2593 err = f2fs_get_node_info(sbi, ino, &old_ni);
2597 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2600 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2602 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2606 /* Should not use this inode from free nid list */
2607 remove_free_nid(sbi, ino);
2609 if (!PageUptodate(ipage))
2610 SetPageUptodate(ipage);
2611 fill_node_footer(ipage, ino, ino, 0, true);
2612 set_cold_node(ipage, false);
2614 src = F2FS_INODE(page);
2615 dst = F2FS_INODE(ipage);
2617 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2619 dst->i_blocks = cpu_to_le64(1);
2620 dst->i_links = cpu_to_le32(1);
2621 dst->i_xattr_nid = 0;
2622 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2623 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2624 dst->i_extra_isize = src->i_extra_isize;
2626 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2627 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2628 i_inline_xattr_size))
2629 dst->i_inline_xattr_size = src->i_inline_xattr_size;
2631 if (f2fs_sb_has_project_quota(sbi) &&
2632 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2634 dst->i_projid = src->i_projid;
2636 if (f2fs_sb_has_inode_crtime(sbi) &&
2637 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2639 dst->i_crtime = src->i_crtime;
2640 dst->i_crtime_nsec = src->i_crtime_nsec;
2647 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2649 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2650 inc_valid_inode_count(sbi);
2651 set_page_dirty(ipage);
2652 f2fs_put_page(ipage, 1);
2656 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2657 unsigned int segno, struct f2fs_summary_block *sum)
2659 struct f2fs_node *rn;
2660 struct f2fs_summary *sum_entry;
2662 int i, idx, last_offset, nrpages;
2664 /* scan the node segment */
2665 last_offset = sbi->blocks_per_seg;
2666 addr = START_BLOCK(sbi, segno);
2667 sum_entry = &sum->entries[0];
2669 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2670 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2672 /* readahead node pages */
2673 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2675 for (idx = addr; idx < addr + nrpages; idx++) {
2676 struct page *page = f2fs_get_tmp_page(sbi, idx);
2679 return PTR_ERR(page);
2681 rn = F2FS_NODE(page);
2682 sum_entry->nid = rn->footer.nid;
2683 sum_entry->version = 0;
2684 sum_entry->ofs_in_node = 0;
2686 f2fs_put_page(page, 1);
2689 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2695 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2697 struct f2fs_nm_info *nm_i = NM_I(sbi);
2698 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2699 struct f2fs_journal *journal = curseg->journal;
2702 down_write(&curseg->journal_rwsem);
2703 for (i = 0; i < nats_in_cursum(journal); i++) {
2704 struct nat_entry *ne;
2705 struct f2fs_nat_entry raw_ne;
2706 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2708 raw_ne = nat_in_journal(journal, i);
2710 ne = __lookup_nat_cache(nm_i, nid);
2712 ne = __alloc_nat_entry(nid, true);
2713 __init_nat_entry(nm_i, ne, &raw_ne, true);
2717 * if a free nat in journal has not been used after last
2718 * checkpoint, we should remove it from available nids,
2719 * since later we will add it again.
2721 if (!get_nat_flag(ne, IS_DIRTY) &&
2722 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2723 spin_lock(&nm_i->nid_list_lock);
2724 nm_i->available_nids--;
2725 spin_unlock(&nm_i->nid_list_lock);
2728 __set_nat_cache_dirty(nm_i, ne);
2730 update_nats_in_cursum(journal, -i);
2731 up_write(&curseg->journal_rwsem);
2734 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2735 struct list_head *head, int max)
2737 struct nat_entry_set *cur;
2739 if (nes->entry_cnt >= max)
2742 list_for_each_entry(cur, head, set_list) {
2743 if (cur->entry_cnt >= nes->entry_cnt) {
2744 list_add(&nes->set_list, cur->set_list.prev);
2749 list_add_tail(&nes->set_list, head);
2752 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2755 struct f2fs_nm_info *nm_i = NM_I(sbi);
2756 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2757 struct f2fs_nat_block *nat_blk = page_address(page);
2761 if (!enabled_nat_bits(sbi, NULL))
2764 if (nat_index == 0) {
2768 for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2769 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2773 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2774 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2778 __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2779 if (valid == NAT_ENTRY_PER_BLOCK)
2780 __set_bit_le(nat_index, nm_i->full_nat_bits);
2782 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2785 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2786 struct nat_entry_set *set, struct cp_control *cpc)
2788 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2789 struct f2fs_journal *journal = curseg->journal;
2790 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2791 bool to_journal = true;
2792 struct f2fs_nat_block *nat_blk;
2793 struct nat_entry *ne, *cur;
2794 struct page *page = NULL;
2797 * there are two steps to flush nat entries:
2798 * #1, flush nat entries to journal in current hot data summary block.
2799 * #2, flush nat entries to nat page.
2801 if (enabled_nat_bits(sbi, cpc) ||
2802 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2806 down_write(&curseg->journal_rwsem);
2808 page = get_next_nat_page(sbi, start_nid);
2810 return PTR_ERR(page);
2812 nat_blk = page_address(page);
2813 f2fs_bug_on(sbi, !nat_blk);
2816 /* flush dirty nats in nat entry set */
2817 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2818 struct f2fs_nat_entry *raw_ne;
2819 nid_t nid = nat_get_nid(ne);
2822 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2825 offset = f2fs_lookup_journal_in_cursum(journal,
2826 NAT_JOURNAL, nid, 1);
2827 f2fs_bug_on(sbi, offset < 0);
2828 raw_ne = &nat_in_journal(journal, offset);
2829 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2831 raw_ne = &nat_blk->entries[nid - start_nid];
2833 raw_nat_from_node_info(raw_ne, &ne->ni);
2835 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2836 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2837 add_free_nid(sbi, nid, false, true);
2839 spin_lock(&NM_I(sbi)->nid_list_lock);
2840 update_free_nid_bitmap(sbi, nid, false, false);
2841 spin_unlock(&NM_I(sbi)->nid_list_lock);
2846 up_write(&curseg->journal_rwsem);
2848 __update_nat_bits(sbi, start_nid, page);
2849 f2fs_put_page(page, 1);
2852 /* Allow dirty nats by node block allocation in write_begin */
2853 if (!set->entry_cnt) {
2854 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2855 kmem_cache_free(nat_entry_set_slab, set);
2861 * This function is called during the checkpointing process.
2863 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2865 struct f2fs_nm_info *nm_i = NM_I(sbi);
2866 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2867 struct f2fs_journal *journal = curseg->journal;
2868 struct nat_entry_set *setvec[SETVEC_SIZE];
2869 struct nat_entry_set *set, *tmp;
2875 /* during unmount, let's flush nat_bits before checking dirty_nat_cnt */
2876 if (enabled_nat_bits(sbi, cpc)) {
2877 down_write(&nm_i->nat_tree_lock);
2878 remove_nats_in_journal(sbi);
2879 up_write(&nm_i->nat_tree_lock);
2882 if (!nm_i->dirty_nat_cnt)
2885 down_write(&nm_i->nat_tree_lock);
2888 * if there are no enough space in journal to store dirty nat
2889 * entries, remove all entries from journal and merge them
2890 * into nat entry set.
2892 if (enabled_nat_bits(sbi, cpc) ||
2893 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2894 remove_nats_in_journal(sbi);
2896 while ((found = __gang_lookup_nat_set(nm_i,
2897 set_idx, SETVEC_SIZE, setvec))) {
2899 set_idx = setvec[found - 1]->set + 1;
2900 for (idx = 0; idx < found; idx++)
2901 __adjust_nat_entry_set(setvec[idx], &sets,
2902 MAX_NAT_JENTRIES(journal));
2905 /* flush dirty nats in nat entry set */
2906 list_for_each_entry_safe(set, tmp, &sets, set_list) {
2907 err = __flush_nat_entry_set(sbi, set, cpc);
2912 up_write(&nm_i->nat_tree_lock);
2913 /* Allow dirty nats by node block allocation in write_begin */
2918 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2920 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2921 struct f2fs_nm_info *nm_i = NM_I(sbi);
2922 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2924 __u64 cp_ver = cur_cp_version(ckpt);
2925 block_t nat_bits_addr;
2927 if (!enabled_nat_bits(sbi, NULL))
2930 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
2931 nm_i->nat_bits = f2fs_kzalloc(sbi,
2932 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
2933 if (!nm_i->nat_bits)
2936 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2937 nm_i->nat_bits_blocks;
2938 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2941 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
2943 return PTR_ERR(page);
2945 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2946 page_address(page), F2FS_BLKSIZE);
2947 f2fs_put_page(page, 1);
2950 cp_ver |= (cur_cp_crc(ckpt) << 32);
2951 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2952 disable_nat_bits(sbi, true);
2956 nm_i->full_nat_bits = nm_i->nat_bits + 8;
2957 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2959 f2fs_notice(sbi, "Found nat_bits in checkpoint");
2963 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2965 struct f2fs_nm_info *nm_i = NM_I(sbi);
2967 nid_t nid, last_nid;
2969 if (!enabled_nat_bits(sbi, NULL))
2972 for (i = 0; i < nm_i->nat_blocks; i++) {
2973 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2974 if (i >= nm_i->nat_blocks)
2977 __set_bit_le(i, nm_i->nat_block_bitmap);
2979 nid = i * NAT_ENTRY_PER_BLOCK;
2980 last_nid = nid + NAT_ENTRY_PER_BLOCK;
2982 spin_lock(&NM_I(sbi)->nid_list_lock);
2983 for (; nid < last_nid; nid++)
2984 update_free_nid_bitmap(sbi, nid, true, true);
2985 spin_unlock(&NM_I(sbi)->nid_list_lock);
2988 for (i = 0; i < nm_i->nat_blocks; i++) {
2989 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2990 if (i >= nm_i->nat_blocks)
2993 __set_bit_le(i, nm_i->nat_block_bitmap);
2997 static int init_node_manager(struct f2fs_sb_info *sbi)
2999 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3000 struct f2fs_nm_info *nm_i = NM_I(sbi);
3001 unsigned char *version_bitmap;
3002 unsigned int nat_segs;
3005 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3007 /* segment_count_nat includes pair segment so divide to 2. */
3008 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3009 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3010 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3012 /* not used nids: 0, node, meta, (and root counted as valid node) */
3013 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3014 F2FS_RESERVED_NODE_NUM;
3015 nm_i->nid_cnt[FREE_NID] = 0;
3016 nm_i->nid_cnt[PREALLOC_NID] = 0;
3018 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3019 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3020 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3022 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3023 INIT_LIST_HEAD(&nm_i->free_nid_list);
3024 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3025 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3026 INIT_LIST_HEAD(&nm_i->nat_entries);
3027 spin_lock_init(&nm_i->nat_list_lock);
3029 mutex_init(&nm_i->build_lock);
3030 spin_lock_init(&nm_i->nid_list_lock);
3031 init_rwsem(&nm_i->nat_tree_lock);
3033 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3034 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3035 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3036 if (!version_bitmap)
3039 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3041 if (!nm_i->nat_bitmap)
3044 err = __get_nat_bitmaps(sbi);
3048 #ifdef CONFIG_F2FS_CHECK_FS
3049 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3051 if (!nm_i->nat_bitmap_mir)
3058 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3060 struct f2fs_nm_info *nm_i = NM_I(sbi);
3063 nm_i->free_nid_bitmap =
3064 f2fs_kzalloc(sbi, array_size(sizeof(unsigned char *),
3067 if (!nm_i->free_nid_bitmap)
3070 for (i = 0; i < nm_i->nat_blocks; i++) {
3071 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3072 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3073 if (!nm_i->free_nid_bitmap[i])
3077 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3079 if (!nm_i->nat_block_bitmap)
3082 nm_i->free_nid_count =
3083 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3086 if (!nm_i->free_nid_count)
3091 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3095 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3100 err = init_node_manager(sbi);
3104 err = init_free_nid_cache(sbi);
3108 /* load free nid status from nat_bits table */
3109 load_free_nid_bitmap(sbi);
3111 return f2fs_build_free_nids(sbi, true, true);
3114 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3116 struct f2fs_nm_info *nm_i = NM_I(sbi);
3117 struct free_nid *i, *next_i;
3118 struct nat_entry *natvec[NATVEC_SIZE];
3119 struct nat_entry_set *setvec[SETVEC_SIZE];
3126 /* destroy free nid list */
3127 spin_lock(&nm_i->nid_list_lock);
3128 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3129 __remove_free_nid(sbi, i, FREE_NID);
3130 spin_unlock(&nm_i->nid_list_lock);
3131 kmem_cache_free(free_nid_slab, i);
3132 spin_lock(&nm_i->nid_list_lock);
3134 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3135 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3136 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3137 spin_unlock(&nm_i->nid_list_lock);
3139 /* destroy nat cache */
3140 down_write(&nm_i->nat_tree_lock);
3141 while ((found = __gang_lookup_nat_cache(nm_i,
3142 nid, NATVEC_SIZE, natvec))) {
3145 nid = nat_get_nid(natvec[found - 1]) + 1;
3146 for (idx = 0; idx < found; idx++) {
3147 spin_lock(&nm_i->nat_list_lock);
3148 list_del(&natvec[idx]->list);
3149 spin_unlock(&nm_i->nat_list_lock);
3151 __del_from_nat_cache(nm_i, natvec[idx]);
3154 f2fs_bug_on(sbi, nm_i->nat_cnt);
3156 /* destroy nat set cache */
3158 while ((found = __gang_lookup_nat_set(nm_i,
3159 nid, SETVEC_SIZE, setvec))) {
3162 nid = setvec[found - 1]->set + 1;
3163 for (idx = 0; idx < found; idx++) {
3164 /* entry_cnt is not zero, when cp_error was occurred */
3165 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3166 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3167 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3170 up_write(&nm_i->nat_tree_lock);
3172 kvfree(nm_i->nat_block_bitmap);
3173 if (nm_i->free_nid_bitmap) {
3176 for (i = 0; i < nm_i->nat_blocks; i++)
3177 kvfree(nm_i->free_nid_bitmap[i]);
3178 kvfree(nm_i->free_nid_bitmap);
3180 kvfree(nm_i->free_nid_count);
3182 kvfree(nm_i->nat_bitmap);
3183 kvfree(nm_i->nat_bits);
3184 #ifdef CONFIG_F2FS_CHECK_FS
3185 kvfree(nm_i->nat_bitmap_mir);
3187 sbi->nm_info = NULL;
3191 int __init f2fs_create_node_manager_caches(void)
3193 nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3194 sizeof(struct nat_entry));
3195 if (!nat_entry_slab)
3198 free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3199 sizeof(struct free_nid));
3201 goto destroy_nat_entry;
3203 nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3204 sizeof(struct nat_entry_set));
3205 if (!nat_entry_set_slab)
3206 goto destroy_free_nid;
3208 fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3209 sizeof(struct fsync_node_entry));
3210 if (!fsync_node_entry_slab)
3211 goto destroy_nat_entry_set;
3214 destroy_nat_entry_set:
3215 kmem_cache_destroy(nat_entry_set_slab);
3217 kmem_cache_destroy(free_nid_slab);
3219 kmem_cache_destroy(nat_entry_slab);
3224 void f2fs_destroy_node_manager_caches(void)
3226 kmem_cache_destroy(fsync_node_entry_slab);
3227 kmem_cache_destroy(nat_entry_set_slab);
3228 kmem_cache_destroy(free_nid_slab);
3229 kmem_cache_destroy(nat_entry_slab);