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;
514 * This function always returns success
516 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
517 struct node_info *ni)
519 struct f2fs_nm_info *nm_i = NM_I(sbi);
520 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
521 struct f2fs_journal *journal = curseg->journal;
522 nid_t start_nid = START_NID(nid);
523 struct f2fs_nat_block *nat_blk;
524 struct page *page = NULL;
525 struct f2fs_nat_entry ne;
533 /* Check nat cache */
534 down_read(&nm_i->nat_tree_lock);
535 e = __lookup_nat_cache(nm_i, nid);
537 ni->ino = nat_get_ino(e);
538 ni->blk_addr = nat_get_blkaddr(e);
539 ni->version = nat_get_version(e);
540 up_read(&nm_i->nat_tree_lock);
544 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
546 /* Check current segment summary */
547 down_read(&curseg->journal_rwsem);
548 i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
550 ne = nat_in_journal(journal, i);
551 node_info_from_raw_nat(ni, &ne);
553 up_read(&curseg->journal_rwsem);
555 up_read(&nm_i->nat_tree_lock);
559 /* Fill node_info from nat page */
560 index = current_nat_addr(sbi, nid);
561 up_read(&nm_i->nat_tree_lock);
563 page = f2fs_get_meta_page(sbi, index);
565 return PTR_ERR(page);
567 nat_blk = (struct f2fs_nat_block *)page_address(page);
568 ne = nat_blk->entries[nid - start_nid];
569 node_info_from_raw_nat(ni, &ne);
570 f2fs_put_page(page, 1);
572 blkaddr = le32_to_cpu(ne.block_addr);
573 if (__is_valid_data_blkaddr(blkaddr) &&
574 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
577 /* cache nat entry */
578 cache_nat_entry(sbi, nid, &ne);
583 * readahead MAX_RA_NODE number of node pages.
585 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
587 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
588 struct blk_plug plug;
592 blk_start_plug(&plug);
594 /* Then, try readahead for siblings of the desired node */
596 end = min(end, NIDS_PER_BLOCK);
597 for (i = start; i < end; i++) {
598 nid = get_nid(parent, i, false);
599 f2fs_ra_node_page(sbi, nid);
602 blk_finish_plug(&plug);
605 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
607 const long direct_index = ADDRS_PER_INODE(dn->inode);
608 const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
609 const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
610 unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
611 int cur_level = dn->cur_level;
612 int max_level = dn->max_level;
618 while (max_level-- > cur_level)
619 skipped_unit *= NIDS_PER_BLOCK;
621 switch (dn->max_level) {
623 base += 2 * indirect_blks;
626 base += 2 * direct_blks;
629 base += direct_index;
632 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
635 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
639 * The maximum depth is four.
640 * Offset[0] will have raw inode offset.
642 static int get_node_path(struct inode *inode, long block,
643 int offset[4], unsigned int noffset[4])
645 const long direct_index = ADDRS_PER_INODE(inode);
646 const long direct_blks = ADDRS_PER_BLOCK(inode);
647 const long dptrs_per_blk = NIDS_PER_BLOCK;
648 const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
649 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
655 if (block < direct_index) {
659 block -= direct_index;
660 if (block < direct_blks) {
661 offset[n++] = NODE_DIR1_BLOCK;
667 block -= direct_blks;
668 if (block < direct_blks) {
669 offset[n++] = NODE_DIR2_BLOCK;
675 block -= direct_blks;
676 if (block < indirect_blks) {
677 offset[n++] = NODE_IND1_BLOCK;
679 offset[n++] = block / direct_blks;
680 noffset[n] = 4 + offset[n - 1];
681 offset[n] = block % direct_blks;
685 block -= indirect_blks;
686 if (block < indirect_blks) {
687 offset[n++] = NODE_IND2_BLOCK;
688 noffset[n] = 4 + dptrs_per_blk;
689 offset[n++] = block / direct_blks;
690 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
691 offset[n] = block % direct_blks;
695 block -= indirect_blks;
696 if (block < dindirect_blks) {
697 offset[n++] = NODE_DIND_BLOCK;
698 noffset[n] = 5 + (dptrs_per_blk * 2);
699 offset[n++] = block / indirect_blks;
700 noffset[n] = 6 + (dptrs_per_blk * 2) +
701 offset[n - 1] * (dptrs_per_blk + 1);
702 offset[n++] = (block / direct_blks) % dptrs_per_blk;
703 noffset[n] = 7 + (dptrs_per_blk * 2) +
704 offset[n - 2] * (dptrs_per_blk + 1) +
706 offset[n] = block % direct_blks;
717 * Caller should call f2fs_put_dnode(dn).
718 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
719 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
720 * In the case of RDONLY_NODE, we don't need to care about mutex.
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 = datablock_addr(dn->inode,
813 dn->node_page, dn->ofs_in_node);
817 f2fs_put_page(parent, 1);
819 f2fs_put_page(npage[0], 0);
821 dn->inode_page = NULL;
822 dn->node_page = NULL;
823 if (err == -ENOENT) {
825 dn->max_level = level;
826 dn->ofs_in_node = offset[level];
831 static int truncate_node(struct dnode_of_data *dn)
833 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
838 err = f2fs_get_node_info(sbi, dn->nid, &ni);
842 /* Deallocate node address */
843 f2fs_invalidate_blocks(sbi, ni.blk_addr);
844 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
845 set_node_addr(sbi, &ni, NULL_ADDR, false);
847 if (dn->nid == dn->inode->i_ino) {
848 f2fs_remove_orphan_inode(sbi, dn->nid);
849 dec_valid_inode_count(sbi);
850 f2fs_inode_synced(dn->inode);
853 clear_node_page_dirty(dn->node_page);
854 set_sbi_flag(sbi, SBI_IS_DIRTY);
856 index = dn->node_page->index;
857 f2fs_put_page(dn->node_page, 1);
859 invalidate_mapping_pages(NODE_MAPPING(sbi),
862 dn->node_page = NULL;
863 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
868 static int truncate_dnode(struct dnode_of_data *dn)
876 /* get direct node */
877 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
878 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
880 else if (IS_ERR(page))
881 return PTR_ERR(page);
883 /* Make dnode_of_data for parameter */
884 dn->node_page = page;
886 f2fs_truncate_data_blocks(dn);
887 err = truncate_node(dn);
894 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
897 struct dnode_of_data rdn = *dn;
899 struct f2fs_node *rn;
901 unsigned int child_nofs;
906 return NIDS_PER_BLOCK + 1;
908 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
910 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
912 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
913 return PTR_ERR(page);
916 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
918 rn = F2FS_NODE(page);
920 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
921 child_nid = le32_to_cpu(rn->in.nid[i]);
925 ret = truncate_dnode(&rdn);
928 if (set_nid(page, i, 0, false))
929 dn->node_changed = true;
932 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
933 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
934 child_nid = le32_to_cpu(rn->in.nid[i]);
935 if (child_nid == 0) {
936 child_nofs += NIDS_PER_BLOCK + 1;
940 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
941 if (ret == (NIDS_PER_BLOCK + 1)) {
942 if (set_nid(page, i, 0, false))
943 dn->node_changed = true;
945 } else if (ret < 0 && ret != -ENOENT) {
953 /* remove current indirect node */
954 dn->node_page = page;
955 ret = truncate_node(dn);
960 f2fs_put_page(page, 1);
962 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
966 f2fs_put_page(page, 1);
967 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
971 static int truncate_partial_nodes(struct dnode_of_data *dn,
972 struct f2fs_inode *ri, int *offset, int depth)
974 struct page *pages[2];
981 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
985 /* get indirect nodes in the path */
986 for (i = 0; i < idx + 1; i++) {
987 /* reference count'll be increased */
988 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
989 if (IS_ERR(pages[i])) {
990 err = PTR_ERR(pages[i]);
994 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
997 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
999 /* free direct nodes linked to a partial indirect node */
1000 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1001 child_nid = get_nid(pages[idx], i, false);
1004 dn->nid = child_nid;
1005 err = truncate_dnode(dn);
1008 if (set_nid(pages[idx], i, 0, false))
1009 dn->node_changed = true;
1012 if (offset[idx + 1] == 0) {
1013 dn->node_page = pages[idx];
1015 err = truncate_node(dn);
1019 f2fs_put_page(pages[idx], 1);
1022 offset[idx + 1] = 0;
1025 for (i = idx; i >= 0; i--)
1026 f2fs_put_page(pages[i], 1);
1028 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1034 * All the block addresses of data and nodes should be nullified.
1036 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1038 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1039 int err = 0, cont = 1;
1040 int level, offset[4], noffset[4];
1041 unsigned int nofs = 0;
1042 struct f2fs_inode *ri;
1043 struct dnode_of_data dn;
1046 trace_f2fs_truncate_inode_blocks_enter(inode, from);
1048 level = get_node_path(inode, from, offset, noffset);
1052 page = f2fs_get_node_page(sbi, inode->i_ino);
1054 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1055 return PTR_ERR(page);
1058 set_new_dnode(&dn, inode, page, NULL, 0);
1061 ri = F2FS_INODE(page);
1069 if (!offset[level - 1])
1071 err = truncate_partial_nodes(&dn, ri, offset, level);
1072 if (err < 0 && err != -ENOENT)
1074 nofs += 1 + NIDS_PER_BLOCK;
1077 nofs = 5 + 2 * NIDS_PER_BLOCK;
1078 if (!offset[level - 1])
1080 err = truncate_partial_nodes(&dn, ri, offset, level);
1081 if (err < 0 && err != -ENOENT)
1090 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1091 switch (offset[0]) {
1092 case NODE_DIR1_BLOCK:
1093 case NODE_DIR2_BLOCK:
1094 err = truncate_dnode(&dn);
1097 case NODE_IND1_BLOCK:
1098 case NODE_IND2_BLOCK:
1099 err = truncate_nodes(&dn, nofs, offset[1], 2);
1102 case NODE_DIND_BLOCK:
1103 err = truncate_nodes(&dn, nofs, offset[1], 3);
1110 if (err < 0 && err != -ENOENT)
1112 if (offset[1] == 0 &&
1113 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1115 BUG_ON(page->mapping != NODE_MAPPING(sbi));
1116 f2fs_wait_on_page_writeback(page, NODE, true, true);
1117 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1118 set_page_dirty(page);
1126 f2fs_put_page(page, 0);
1127 trace_f2fs_truncate_inode_blocks_exit(inode, err);
1128 return err > 0 ? 0 : err;
1131 /* caller must lock inode page */
1132 int f2fs_truncate_xattr_node(struct inode *inode)
1134 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1135 nid_t nid = F2FS_I(inode)->i_xattr_nid;
1136 struct dnode_of_data dn;
1143 npage = f2fs_get_node_page(sbi, nid);
1145 return PTR_ERR(npage);
1147 set_new_dnode(&dn, inode, NULL, npage, nid);
1148 err = truncate_node(&dn);
1150 f2fs_put_page(npage, 1);
1154 f2fs_i_xnid_write(inode, 0);
1160 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1163 int f2fs_remove_inode_page(struct inode *inode)
1165 struct dnode_of_data dn;
1168 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1169 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1173 err = f2fs_truncate_xattr_node(inode);
1175 f2fs_put_dnode(&dn);
1179 /* remove potential inline_data blocks */
1180 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1181 S_ISLNK(inode->i_mode))
1182 f2fs_truncate_data_blocks_range(&dn, 1);
1184 /* 0 is possible, after f2fs_new_inode() has failed */
1185 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1186 f2fs_put_dnode(&dn);
1190 if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1191 f2fs_warn(F2FS_I_SB(inode), "Inconsistent i_blocks, ino:%lu, iblocks:%llu",
1192 inode->i_ino, (unsigned long long)inode->i_blocks);
1193 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1196 /* will put inode & node pages */
1197 err = truncate_node(&dn);
1199 f2fs_put_dnode(&dn);
1205 struct page *f2fs_new_inode_page(struct inode *inode)
1207 struct dnode_of_data dn;
1209 /* allocate inode page for new inode */
1210 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1212 /* caller should f2fs_put_page(page, 1); */
1213 return f2fs_new_node_page(&dn, 0);
1216 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1218 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1219 struct node_info new_ni;
1223 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1224 return ERR_PTR(-EPERM);
1226 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1228 return ERR_PTR(-ENOMEM);
1230 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1233 #ifdef CONFIG_F2FS_CHECK_FS
1234 err = f2fs_get_node_info(sbi, dn->nid, &new_ni);
1236 dec_valid_node_count(sbi, dn->inode, !ofs);
1239 f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1241 new_ni.nid = dn->nid;
1242 new_ni.ino = dn->inode->i_ino;
1243 new_ni.blk_addr = NULL_ADDR;
1246 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1248 f2fs_wait_on_page_writeback(page, NODE, true, true);
1249 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1250 set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1251 if (!PageUptodate(page))
1252 SetPageUptodate(page);
1253 if (set_page_dirty(page))
1254 dn->node_changed = true;
1256 if (f2fs_has_xattr_block(ofs))
1257 f2fs_i_xnid_write(dn->inode, dn->nid);
1260 inc_valid_inode_count(sbi);
1264 clear_node_page_dirty(page);
1265 f2fs_put_page(page, 1);
1266 return ERR_PTR(err);
1270 * Caller should do after getting the following values.
1271 * 0: f2fs_put_page(page, 0)
1272 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1274 static int read_node_page(struct page *page, int op_flags)
1276 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1277 struct node_info ni;
1278 struct f2fs_io_info fio = {
1282 .op_flags = op_flags,
1284 .encrypted_page = NULL,
1288 if (PageUptodate(page)) {
1289 if (!f2fs_inode_chksum_verify(sbi, page)) {
1290 ClearPageUptodate(page);
1296 err = f2fs_get_node_info(sbi, page->index, &ni);
1300 if (unlikely(ni.blk_addr == NULL_ADDR) ||
1301 is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1302 ClearPageUptodate(page);
1306 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1307 return f2fs_submit_page_bio(&fio);
1311 * Readahead a node page
1313 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1320 if (f2fs_check_nid_range(sbi, nid))
1323 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1327 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1331 err = read_node_page(apage, REQ_RAHEAD);
1332 f2fs_put_page(apage, err ? 1 : 0);
1335 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1336 struct page *parent, int start)
1342 return ERR_PTR(-ENOENT);
1343 if (f2fs_check_nid_range(sbi, nid))
1344 return ERR_PTR(-EINVAL);
1346 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1348 return ERR_PTR(-ENOMEM);
1350 err = read_node_page(page, 0);
1352 f2fs_put_page(page, 1);
1353 return ERR_PTR(err);
1354 } else if (err == LOCKED_PAGE) {
1360 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1364 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1365 f2fs_put_page(page, 1);
1369 if (unlikely(!PageUptodate(page))) {
1374 if (!f2fs_inode_chksum_verify(sbi, page)) {
1379 if(unlikely(nid != nid_of_node(page))) {
1380 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1381 nid, nid_of_node(page), ino_of_node(page),
1382 ofs_of_node(page), cpver_of_node(page),
1383 next_blkaddr_of_node(page));
1386 ClearPageUptodate(page);
1387 f2fs_put_page(page, 1);
1388 return ERR_PTR(err);
1393 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1395 return __get_node_page(sbi, nid, NULL, 0);
1398 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1400 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1401 nid_t nid = get_nid(parent, start, false);
1403 return __get_node_page(sbi, nid, parent, start);
1406 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1408 struct inode *inode;
1412 /* should flush inline_data before evict_inode */
1413 inode = ilookup(sbi->sb, ino);
1417 page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1418 FGP_LOCK|FGP_NOWAIT, 0);
1422 if (!PageUptodate(page))
1425 if (!PageDirty(page))
1428 if (!clear_page_dirty_for_io(page))
1431 ret = f2fs_write_inline_data(inode, page);
1432 inode_dec_dirty_pages(inode);
1433 f2fs_remove_dirty_inode(inode);
1435 set_page_dirty(page);
1437 f2fs_put_page(page, 1);
1442 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1445 struct pagevec pvec;
1446 struct page *last_page = NULL;
1449 pagevec_init(&pvec);
1452 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1453 PAGECACHE_TAG_DIRTY))) {
1456 for (i = 0; i < nr_pages; i++) {
1457 struct page *page = pvec.pages[i];
1459 if (unlikely(f2fs_cp_error(sbi))) {
1460 f2fs_put_page(last_page, 0);
1461 pagevec_release(&pvec);
1462 return ERR_PTR(-EIO);
1465 if (!IS_DNODE(page) || !is_cold_node(page))
1467 if (ino_of_node(page) != ino)
1472 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1477 if (ino_of_node(page) != ino)
1478 goto continue_unlock;
1480 if (!PageDirty(page)) {
1481 /* someone wrote it for us */
1482 goto continue_unlock;
1486 f2fs_put_page(last_page, 0);
1492 pagevec_release(&pvec);
1498 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1499 struct writeback_control *wbc, bool do_balance,
1500 enum iostat_type io_type, unsigned int *seq_id)
1502 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1504 struct node_info ni;
1505 struct f2fs_io_info fio = {
1507 .ino = ino_of_node(page),
1510 .op_flags = wbc_to_write_flags(wbc),
1512 .encrypted_page = NULL,
1519 trace_f2fs_writepage(page, NODE);
1521 if (unlikely(f2fs_cp_error(sbi)))
1524 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1527 if (wbc->sync_mode == WB_SYNC_NONE &&
1528 IS_DNODE(page) && is_cold_node(page))
1531 /* get old block addr of this node page */
1532 nid = nid_of_node(page);
1533 f2fs_bug_on(sbi, page->index != nid);
1535 if (f2fs_get_node_info(sbi, nid, &ni))
1538 if (wbc->for_reclaim) {
1539 if (!down_read_trylock(&sbi->node_write))
1542 down_read(&sbi->node_write);
1545 /* This page is already truncated */
1546 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1547 ClearPageUptodate(page);
1548 dec_page_count(sbi, F2FS_DIRTY_NODES);
1549 up_read(&sbi->node_write);
1554 if (__is_valid_data_blkaddr(ni.blk_addr) &&
1555 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1556 DATA_GENERIC_ENHANCE)) {
1557 up_read(&sbi->node_write);
1561 if (atomic && !test_opt(sbi, NOBARRIER))
1562 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1564 set_page_writeback(page);
1565 ClearPageError(page);
1567 if (f2fs_in_warm_node_list(sbi, page)) {
1568 seq = f2fs_add_fsync_node_entry(sbi, page);
1573 fio.old_blkaddr = ni.blk_addr;
1574 f2fs_do_write_node_page(nid, &fio);
1575 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1576 dec_page_count(sbi, F2FS_DIRTY_NODES);
1577 up_read(&sbi->node_write);
1579 if (wbc->for_reclaim) {
1580 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1586 if (unlikely(f2fs_cp_error(sbi))) {
1587 f2fs_submit_merged_write(sbi, NODE);
1591 *submitted = fio.submitted;
1594 f2fs_balance_fs(sbi, false);
1598 redirty_page_for_writepage(wbc, page);
1599 return AOP_WRITEPAGE_ACTIVATE;
1602 int f2fs_move_node_page(struct page *node_page, int gc_type)
1606 if (gc_type == FG_GC) {
1607 struct writeback_control wbc = {
1608 .sync_mode = WB_SYNC_ALL,
1613 f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1615 set_page_dirty(node_page);
1617 if (!clear_page_dirty_for_io(node_page)) {
1622 if (__write_node_page(node_page, false, NULL,
1623 &wbc, false, FS_GC_NODE_IO, NULL)) {
1625 unlock_page(node_page);
1629 /* set page dirty and write it */
1630 if (!PageWriteback(node_page))
1631 set_page_dirty(node_page);
1634 unlock_page(node_page);
1636 f2fs_put_page(node_page, 0);
1640 static int f2fs_write_node_page(struct page *page,
1641 struct writeback_control *wbc)
1643 return __write_node_page(page, false, NULL, wbc, false,
1647 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1648 struct writeback_control *wbc, bool atomic,
1649 unsigned int *seq_id)
1652 struct pagevec pvec;
1654 struct page *last_page = NULL;
1655 bool marked = false;
1656 nid_t ino = inode->i_ino;
1661 last_page = last_fsync_dnode(sbi, ino);
1662 if (IS_ERR_OR_NULL(last_page))
1663 return PTR_ERR_OR_ZERO(last_page);
1666 pagevec_init(&pvec);
1669 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1670 PAGECACHE_TAG_DIRTY))) {
1673 for (i = 0; i < nr_pages; i++) {
1674 struct page *page = pvec.pages[i];
1675 bool submitted = false;
1677 if (unlikely(f2fs_cp_error(sbi))) {
1678 f2fs_put_page(last_page, 0);
1679 pagevec_release(&pvec);
1684 if (!IS_DNODE(page) || !is_cold_node(page))
1686 if (ino_of_node(page) != ino)
1691 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1696 if (ino_of_node(page) != ino)
1697 goto continue_unlock;
1699 if (!PageDirty(page) && page != last_page) {
1700 /* someone wrote it for us */
1701 goto continue_unlock;
1704 f2fs_wait_on_page_writeback(page, NODE, true, true);
1706 set_fsync_mark(page, 0);
1707 set_dentry_mark(page, 0);
1709 if (!atomic || page == last_page) {
1710 set_fsync_mark(page, 1);
1711 if (IS_INODE(page)) {
1712 if (is_inode_flag_set(inode,
1714 f2fs_update_inode(inode, page);
1715 set_dentry_mark(page,
1716 f2fs_need_dentry_mark(sbi, ino));
1718 /* may be written by other thread */
1719 if (!PageDirty(page))
1720 set_page_dirty(page);
1723 if (!clear_page_dirty_for_io(page))
1724 goto continue_unlock;
1726 ret = __write_node_page(page, atomic &&
1728 &submitted, wbc, true,
1729 FS_NODE_IO, seq_id);
1732 f2fs_put_page(last_page, 0);
1734 } else if (submitted) {
1738 if (page == last_page) {
1739 f2fs_put_page(page, 0);
1744 pagevec_release(&pvec);
1750 if (!ret && atomic && !marked) {
1751 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1752 ino, last_page->index);
1753 lock_page(last_page);
1754 f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1755 set_page_dirty(last_page);
1756 unlock_page(last_page);
1761 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1762 return ret ? -EIO: 0;
1765 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1766 struct writeback_control *wbc,
1767 bool do_balance, enum iostat_type io_type)
1770 struct pagevec pvec;
1774 int nr_pages, done = 0;
1776 pagevec_init(&pvec);
1781 while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1782 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1785 for (i = 0; i < nr_pages; i++) {
1786 struct page *page = pvec.pages[i];
1787 bool submitted = false;
1789 /* give a priority to WB_SYNC threads */
1790 if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1791 wbc->sync_mode == WB_SYNC_NONE) {
1797 * flushing sequence with step:
1802 if (step == 0 && IS_DNODE(page))
1804 if (step == 1 && (!IS_DNODE(page) ||
1805 is_cold_node(page)))
1807 if (step == 2 && (!IS_DNODE(page) ||
1808 !is_cold_node(page)))
1811 if (wbc->sync_mode == WB_SYNC_ALL)
1813 else if (!trylock_page(page))
1816 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1822 if (!PageDirty(page)) {
1823 /* someone wrote it for us */
1824 goto continue_unlock;
1827 /* flush inline_data */
1828 if (is_inline_node(page)) {
1829 clear_inline_node(page);
1831 flush_inline_data(sbi, ino_of_node(page));
1835 f2fs_wait_on_page_writeback(page, NODE, true, true);
1837 if (!clear_page_dirty_for_io(page))
1838 goto continue_unlock;
1840 set_fsync_mark(page, 0);
1841 set_dentry_mark(page, 0);
1843 ret = __write_node_page(page, false, &submitted,
1844 wbc, do_balance, io_type, NULL);
1850 if (--wbc->nr_to_write == 0)
1853 pagevec_release(&pvec);
1856 if (wbc->nr_to_write == 0) {
1863 if (wbc->sync_mode == WB_SYNC_NONE && step == 1)
1870 f2fs_submit_merged_write(sbi, NODE);
1872 if (unlikely(f2fs_cp_error(sbi)))
1877 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
1878 unsigned int seq_id)
1880 struct fsync_node_entry *fn;
1882 struct list_head *head = &sbi->fsync_node_list;
1883 unsigned long flags;
1884 unsigned int cur_seq_id = 0;
1887 while (seq_id && cur_seq_id < seq_id) {
1888 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
1889 if (list_empty(head)) {
1890 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1893 fn = list_first_entry(head, struct fsync_node_entry, list);
1894 if (fn->seq_id > seq_id) {
1895 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1898 cur_seq_id = fn->seq_id;
1901 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1903 f2fs_wait_on_page_writeback(page, NODE, true, false);
1904 if (TestClearPageError(page))
1913 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1920 static int f2fs_write_node_pages(struct address_space *mapping,
1921 struct writeback_control *wbc)
1923 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1924 struct blk_plug plug;
1927 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1930 /* balancing f2fs's metadata in background */
1931 f2fs_balance_fs_bg(sbi);
1933 /* collect a number of dirty node pages and write together */
1934 if (wbc->sync_mode != WB_SYNC_ALL &&
1935 get_pages(sbi, F2FS_DIRTY_NODES) <
1936 nr_pages_to_skip(sbi, NODE))
1939 if (wbc->sync_mode == WB_SYNC_ALL)
1940 atomic_inc(&sbi->wb_sync_req[NODE]);
1941 else if (atomic_read(&sbi->wb_sync_req[NODE]))
1944 trace_f2fs_writepages(mapping->host, wbc, NODE);
1946 diff = nr_pages_to_write(sbi, NODE, wbc);
1947 blk_start_plug(&plug);
1948 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
1949 blk_finish_plug(&plug);
1950 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1952 if (wbc->sync_mode == WB_SYNC_ALL)
1953 atomic_dec(&sbi->wb_sync_req[NODE]);
1957 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1958 trace_f2fs_writepages(mapping->host, wbc, NODE);
1962 static int f2fs_set_node_page_dirty(struct page *page)
1964 trace_f2fs_set_page_dirty(page, NODE);
1966 if (!PageUptodate(page))
1967 SetPageUptodate(page);
1968 #ifdef CONFIG_F2FS_CHECK_FS
1970 f2fs_inode_chksum_set(F2FS_P_SB(page), page);
1972 if (!PageDirty(page)) {
1973 __set_page_dirty_nobuffers(page);
1974 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1975 f2fs_set_page_private(page, 0);
1976 f2fs_trace_pid(page);
1983 * Structure of the f2fs node operations
1985 const struct address_space_operations f2fs_node_aops = {
1986 .writepage = f2fs_write_node_page,
1987 .writepages = f2fs_write_node_pages,
1988 .set_page_dirty = f2fs_set_node_page_dirty,
1989 .invalidatepage = f2fs_invalidate_page,
1990 .releasepage = f2fs_release_page,
1991 #ifdef CONFIG_MIGRATION
1992 .migratepage = f2fs_migrate_page,
1996 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1999 return radix_tree_lookup(&nm_i->free_nid_root, n);
2002 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2003 struct free_nid *i, enum nid_state state)
2005 struct f2fs_nm_info *nm_i = NM_I(sbi);
2007 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2011 f2fs_bug_on(sbi, state != i->state);
2012 nm_i->nid_cnt[state]++;
2013 if (state == FREE_NID)
2014 list_add_tail(&i->list, &nm_i->free_nid_list);
2018 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2019 struct free_nid *i, enum nid_state state)
2021 struct f2fs_nm_info *nm_i = NM_I(sbi);
2023 f2fs_bug_on(sbi, state != i->state);
2024 nm_i->nid_cnt[state]--;
2025 if (state == FREE_NID)
2027 radix_tree_delete(&nm_i->free_nid_root, i->nid);
2030 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2031 enum nid_state org_state, enum nid_state dst_state)
2033 struct f2fs_nm_info *nm_i = NM_I(sbi);
2035 f2fs_bug_on(sbi, org_state != i->state);
2036 i->state = dst_state;
2037 nm_i->nid_cnt[org_state]--;
2038 nm_i->nid_cnt[dst_state]++;
2040 switch (dst_state) {
2045 list_add_tail(&i->list, &nm_i->free_nid_list);
2052 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2053 bool set, bool build)
2055 struct f2fs_nm_info *nm_i = NM_I(sbi);
2056 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2057 unsigned int nid_ofs = nid - START_NID(nid);
2059 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2063 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2065 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2066 nm_i->free_nid_count[nat_ofs]++;
2068 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2070 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2072 nm_i->free_nid_count[nat_ofs]--;
2076 /* return if the nid is recognized as free */
2077 static bool add_free_nid(struct f2fs_sb_info *sbi,
2078 nid_t nid, bool build, bool update)
2080 struct f2fs_nm_info *nm_i = NM_I(sbi);
2081 struct free_nid *i, *e;
2082 struct nat_entry *ne;
2086 /* 0 nid should not be used */
2087 if (unlikely(nid == 0))
2090 if (unlikely(f2fs_check_nid_range(sbi, nid)))
2093 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2095 i->state = FREE_NID;
2097 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2099 spin_lock(&nm_i->nid_list_lock);
2107 * - __insert_nid_to_list(PREALLOC_NID)
2108 * - f2fs_balance_fs_bg
2109 * - f2fs_build_free_nids
2110 * - __f2fs_build_free_nids
2113 * - __lookup_nat_cache
2115 * - f2fs_init_inode_metadata
2116 * - f2fs_new_inode_page
2117 * - f2fs_new_node_page
2119 * - f2fs_alloc_nid_done
2120 * - __remove_nid_from_list(PREALLOC_NID)
2121 * - __insert_nid_to_list(FREE_NID)
2123 ne = __lookup_nat_cache(nm_i, nid);
2124 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2125 nat_get_blkaddr(ne) != NULL_ADDR))
2128 e = __lookup_free_nid_list(nm_i, nid);
2130 if (e->state == FREE_NID)
2136 err = __insert_free_nid(sbi, i, FREE_NID);
2139 update_free_nid_bitmap(sbi, nid, ret, build);
2141 nm_i->available_nids++;
2143 spin_unlock(&nm_i->nid_list_lock);
2144 radix_tree_preload_end();
2147 kmem_cache_free(free_nid_slab, i);
2151 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2153 struct f2fs_nm_info *nm_i = NM_I(sbi);
2155 bool need_free = false;
2157 spin_lock(&nm_i->nid_list_lock);
2158 i = __lookup_free_nid_list(nm_i, nid);
2159 if (i && i->state == FREE_NID) {
2160 __remove_free_nid(sbi, i, FREE_NID);
2163 spin_unlock(&nm_i->nid_list_lock);
2166 kmem_cache_free(free_nid_slab, i);
2169 static int scan_nat_page(struct f2fs_sb_info *sbi,
2170 struct page *nat_page, nid_t start_nid)
2172 struct f2fs_nm_info *nm_i = NM_I(sbi);
2173 struct f2fs_nat_block *nat_blk = page_address(nat_page);
2175 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2178 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2180 i = start_nid % NAT_ENTRY_PER_BLOCK;
2182 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2183 if (unlikely(start_nid >= nm_i->max_nid))
2186 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2188 if (blk_addr == NEW_ADDR)
2191 if (blk_addr == NULL_ADDR) {
2192 add_free_nid(sbi, start_nid, true, true);
2194 spin_lock(&NM_I(sbi)->nid_list_lock);
2195 update_free_nid_bitmap(sbi, start_nid, false, true);
2196 spin_unlock(&NM_I(sbi)->nid_list_lock);
2203 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2205 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2206 struct f2fs_journal *journal = curseg->journal;
2209 down_read(&curseg->journal_rwsem);
2210 for (i = 0; i < nats_in_cursum(journal); i++) {
2214 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2215 nid = le32_to_cpu(nid_in_journal(journal, i));
2216 if (addr == NULL_ADDR)
2217 add_free_nid(sbi, nid, true, false);
2219 remove_free_nid(sbi, nid);
2221 up_read(&curseg->journal_rwsem);
2224 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2226 struct f2fs_nm_info *nm_i = NM_I(sbi);
2227 unsigned int i, idx;
2230 down_read(&nm_i->nat_tree_lock);
2232 for (i = 0; i < nm_i->nat_blocks; i++) {
2233 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2235 if (!nm_i->free_nid_count[i])
2237 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2238 idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2239 NAT_ENTRY_PER_BLOCK, idx);
2240 if (idx >= NAT_ENTRY_PER_BLOCK)
2243 nid = i * NAT_ENTRY_PER_BLOCK + idx;
2244 add_free_nid(sbi, nid, true, false);
2246 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2251 scan_curseg_cache(sbi);
2253 up_read(&nm_i->nat_tree_lock);
2256 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2257 bool sync, bool mount)
2259 struct f2fs_nm_info *nm_i = NM_I(sbi);
2261 nid_t nid = nm_i->next_scan_nid;
2263 if (unlikely(nid >= nm_i->max_nid))
2266 /* Enough entries */
2267 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2270 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2274 /* try to find free nids in free_nid_bitmap */
2275 scan_free_nid_bits(sbi);
2277 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2281 /* readahead nat pages to be scanned */
2282 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2285 down_read(&nm_i->nat_tree_lock);
2288 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2289 nm_i->nat_block_bitmap)) {
2290 struct page *page = get_current_nat_page(sbi, nid);
2293 ret = PTR_ERR(page);
2295 ret = scan_nat_page(sbi, page, nid);
2296 f2fs_put_page(page, 1);
2300 up_read(&nm_i->nat_tree_lock);
2301 f2fs_bug_on(sbi, !mount);
2302 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2307 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2308 if (unlikely(nid >= nm_i->max_nid))
2311 if (++i >= FREE_NID_PAGES)
2315 /* go to the next free nat pages to find free nids abundantly */
2316 nm_i->next_scan_nid = nid;
2318 /* find free nids from current sum_pages */
2319 scan_curseg_cache(sbi);
2321 up_read(&nm_i->nat_tree_lock);
2323 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2324 nm_i->ra_nid_pages, META_NAT, false);
2329 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2333 mutex_lock(&NM_I(sbi)->build_lock);
2334 ret = __f2fs_build_free_nids(sbi, sync, mount);
2335 mutex_unlock(&NM_I(sbi)->build_lock);
2341 * If this function returns success, caller can obtain a new nid
2342 * from second parameter of this function.
2343 * The returned nid could be used ino as well as nid when inode is created.
2345 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2347 struct f2fs_nm_info *nm_i = NM_I(sbi);
2348 struct free_nid *i = NULL;
2350 if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2351 f2fs_show_injection_info(FAULT_ALLOC_NID);
2355 spin_lock(&nm_i->nid_list_lock);
2357 if (unlikely(nm_i->available_nids == 0)) {
2358 spin_unlock(&nm_i->nid_list_lock);
2362 /* We should not use stale free nids created by f2fs_build_free_nids */
2363 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2364 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2365 i = list_first_entry(&nm_i->free_nid_list,
2366 struct free_nid, list);
2369 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2370 nm_i->available_nids--;
2372 update_free_nid_bitmap(sbi, *nid, false, false);
2374 spin_unlock(&nm_i->nid_list_lock);
2377 spin_unlock(&nm_i->nid_list_lock);
2379 /* Let's scan nat pages and its caches to get free nids */
2380 if (!f2fs_build_free_nids(sbi, true, false))
2386 * f2fs_alloc_nid() should be called prior to this function.
2388 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2390 struct f2fs_nm_info *nm_i = NM_I(sbi);
2393 spin_lock(&nm_i->nid_list_lock);
2394 i = __lookup_free_nid_list(nm_i, nid);
2395 f2fs_bug_on(sbi, !i);
2396 __remove_free_nid(sbi, i, PREALLOC_NID);
2397 spin_unlock(&nm_i->nid_list_lock);
2399 kmem_cache_free(free_nid_slab, i);
2403 * f2fs_alloc_nid() should be called prior to this function.
2405 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2407 struct f2fs_nm_info *nm_i = NM_I(sbi);
2409 bool need_free = false;
2414 spin_lock(&nm_i->nid_list_lock);
2415 i = __lookup_free_nid_list(nm_i, nid);
2416 f2fs_bug_on(sbi, !i);
2418 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2419 __remove_free_nid(sbi, i, PREALLOC_NID);
2422 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2425 nm_i->available_nids++;
2427 update_free_nid_bitmap(sbi, nid, true, false);
2429 spin_unlock(&nm_i->nid_list_lock);
2432 kmem_cache_free(free_nid_slab, i);
2435 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2437 struct f2fs_nm_info *nm_i = NM_I(sbi);
2438 struct free_nid *i, *next;
2441 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2444 if (!mutex_trylock(&nm_i->build_lock))
2447 spin_lock(&nm_i->nid_list_lock);
2448 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2449 if (nr_shrink <= 0 ||
2450 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2453 __remove_free_nid(sbi, i, FREE_NID);
2454 kmem_cache_free(free_nid_slab, i);
2457 spin_unlock(&nm_i->nid_list_lock);
2458 mutex_unlock(&nm_i->build_lock);
2460 return nr - nr_shrink;
2463 void f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2465 void *src_addr, *dst_addr;
2468 struct f2fs_inode *ri;
2470 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2471 f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2473 ri = F2FS_INODE(page);
2474 if (ri->i_inline & F2FS_INLINE_XATTR) {
2475 set_inode_flag(inode, FI_INLINE_XATTR);
2477 clear_inode_flag(inode, FI_INLINE_XATTR);
2481 dst_addr = inline_xattr_addr(inode, ipage);
2482 src_addr = inline_xattr_addr(inode, page);
2483 inline_size = inline_xattr_size(inode);
2485 f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2486 memcpy(dst_addr, src_addr, inline_size);
2488 f2fs_update_inode(inode, ipage);
2489 f2fs_put_page(ipage, 1);
2492 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2494 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2495 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2497 struct dnode_of_data dn;
2498 struct node_info ni;
2505 /* 1: invalidate the previous xattr nid */
2506 err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2510 f2fs_invalidate_blocks(sbi, ni.blk_addr);
2511 dec_valid_node_count(sbi, inode, false);
2512 set_node_addr(sbi, &ni, NULL_ADDR, false);
2515 /* 2: update xattr nid in inode */
2516 if (!f2fs_alloc_nid(sbi, &new_xnid))
2519 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2520 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2521 if (IS_ERR(xpage)) {
2522 f2fs_alloc_nid_failed(sbi, new_xnid);
2523 return PTR_ERR(xpage);
2526 f2fs_alloc_nid_done(sbi, new_xnid);
2527 f2fs_update_inode_page(inode);
2529 /* 3: update and set xattr node page dirty */
2530 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2532 set_page_dirty(xpage);
2533 f2fs_put_page(xpage, 1);
2538 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2540 struct f2fs_inode *src, *dst;
2541 nid_t ino = ino_of_node(page);
2542 struct node_info old_ni, new_ni;
2546 err = f2fs_get_node_info(sbi, ino, &old_ni);
2550 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2553 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2555 congestion_wait(BLK_RW_ASYNC, HZ/50);
2559 /* Should not use this inode from free nid list */
2560 remove_free_nid(sbi, ino);
2562 if (!PageUptodate(ipage))
2563 SetPageUptodate(ipage);
2564 fill_node_footer(ipage, ino, ino, 0, true);
2565 set_cold_node(ipage, false);
2567 src = F2FS_INODE(page);
2568 dst = F2FS_INODE(ipage);
2570 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2572 dst->i_blocks = cpu_to_le64(1);
2573 dst->i_links = cpu_to_le32(1);
2574 dst->i_xattr_nid = 0;
2575 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2576 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2577 dst->i_extra_isize = src->i_extra_isize;
2579 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2580 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2581 i_inline_xattr_size))
2582 dst->i_inline_xattr_size = src->i_inline_xattr_size;
2584 if (f2fs_sb_has_project_quota(sbi) &&
2585 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2587 dst->i_projid = src->i_projid;
2589 if (f2fs_sb_has_inode_crtime(sbi) &&
2590 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2592 dst->i_crtime = src->i_crtime;
2593 dst->i_crtime_nsec = src->i_crtime_nsec;
2600 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2602 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2603 inc_valid_inode_count(sbi);
2604 set_page_dirty(ipage);
2605 f2fs_put_page(ipage, 1);
2609 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2610 unsigned int segno, struct f2fs_summary_block *sum)
2612 struct f2fs_node *rn;
2613 struct f2fs_summary *sum_entry;
2615 int i, idx, last_offset, nrpages;
2617 /* scan the node segment */
2618 last_offset = sbi->blocks_per_seg;
2619 addr = START_BLOCK(sbi, segno);
2620 sum_entry = &sum->entries[0];
2622 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2623 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2625 /* readahead node pages */
2626 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2628 for (idx = addr; idx < addr + nrpages; idx++) {
2629 struct page *page = f2fs_get_tmp_page(sbi, idx);
2632 return PTR_ERR(page);
2634 rn = F2FS_NODE(page);
2635 sum_entry->nid = rn->footer.nid;
2636 sum_entry->version = 0;
2637 sum_entry->ofs_in_node = 0;
2639 f2fs_put_page(page, 1);
2642 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2648 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2650 struct f2fs_nm_info *nm_i = NM_I(sbi);
2651 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2652 struct f2fs_journal *journal = curseg->journal;
2655 down_write(&curseg->journal_rwsem);
2656 for (i = 0; i < nats_in_cursum(journal); i++) {
2657 struct nat_entry *ne;
2658 struct f2fs_nat_entry raw_ne;
2659 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2661 raw_ne = nat_in_journal(journal, i);
2663 ne = __lookup_nat_cache(nm_i, nid);
2665 ne = __alloc_nat_entry(nid, true);
2666 __init_nat_entry(nm_i, ne, &raw_ne, true);
2670 * if a free nat in journal has not been used after last
2671 * checkpoint, we should remove it from available nids,
2672 * since later we will add it again.
2674 if (!get_nat_flag(ne, IS_DIRTY) &&
2675 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2676 spin_lock(&nm_i->nid_list_lock);
2677 nm_i->available_nids--;
2678 spin_unlock(&nm_i->nid_list_lock);
2681 __set_nat_cache_dirty(nm_i, ne);
2683 update_nats_in_cursum(journal, -i);
2684 up_write(&curseg->journal_rwsem);
2687 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2688 struct list_head *head, int max)
2690 struct nat_entry_set *cur;
2692 if (nes->entry_cnt >= max)
2695 list_for_each_entry(cur, head, set_list) {
2696 if (cur->entry_cnt >= nes->entry_cnt) {
2697 list_add(&nes->set_list, cur->set_list.prev);
2702 list_add_tail(&nes->set_list, head);
2705 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2708 struct f2fs_nm_info *nm_i = NM_I(sbi);
2709 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2710 struct f2fs_nat_block *nat_blk = page_address(page);
2714 if (!enabled_nat_bits(sbi, NULL))
2717 if (nat_index == 0) {
2721 for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2722 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2726 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2727 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2731 __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2732 if (valid == NAT_ENTRY_PER_BLOCK)
2733 __set_bit_le(nat_index, nm_i->full_nat_bits);
2735 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2738 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2739 struct nat_entry_set *set, struct cp_control *cpc)
2741 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2742 struct f2fs_journal *journal = curseg->journal;
2743 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2744 bool to_journal = true;
2745 struct f2fs_nat_block *nat_blk;
2746 struct nat_entry *ne, *cur;
2747 struct page *page = NULL;
2750 * there are two steps to flush nat entries:
2751 * #1, flush nat entries to journal in current hot data summary block.
2752 * #2, flush nat entries to nat page.
2754 if (enabled_nat_bits(sbi, cpc) ||
2755 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2759 down_write(&curseg->journal_rwsem);
2761 page = get_next_nat_page(sbi, start_nid);
2763 return PTR_ERR(page);
2765 nat_blk = page_address(page);
2766 f2fs_bug_on(sbi, !nat_blk);
2769 /* flush dirty nats in nat entry set */
2770 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2771 struct f2fs_nat_entry *raw_ne;
2772 nid_t nid = nat_get_nid(ne);
2775 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2778 offset = f2fs_lookup_journal_in_cursum(journal,
2779 NAT_JOURNAL, nid, 1);
2780 f2fs_bug_on(sbi, offset < 0);
2781 raw_ne = &nat_in_journal(journal, offset);
2782 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2784 raw_ne = &nat_blk->entries[nid - start_nid];
2786 raw_nat_from_node_info(raw_ne, &ne->ni);
2788 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2789 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2790 add_free_nid(sbi, nid, false, true);
2792 spin_lock(&NM_I(sbi)->nid_list_lock);
2793 update_free_nid_bitmap(sbi, nid, false, false);
2794 spin_unlock(&NM_I(sbi)->nid_list_lock);
2799 up_write(&curseg->journal_rwsem);
2801 __update_nat_bits(sbi, start_nid, page);
2802 f2fs_put_page(page, 1);
2805 /* Allow dirty nats by node block allocation in write_begin */
2806 if (!set->entry_cnt) {
2807 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2808 kmem_cache_free(nat_entry_set_slab, set);
2814 * This function is called during the checkpointing process.
2816 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2818 struct f2fs_nm_info *nm_i = NM_I(sbi);
2819 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2820 struct f2fs_journal *journal = curseg->journal;
2821 struct nat_entry_set *setvec[SETVEC_SIZE];
2822 struct nat_entry_set *set, *tmp;
2828 /* during unmount, let's flush nat_bits before checking dirty_nat_cnt */
2829 if (enabled_nat_bits(sbi, cpc)) {
2830 down_write(&nm_i->nat_tree_lock);
2831 remove_nats_in_journal(sbi);
2832 up_write(&nm_i->nat_tree_lock);
2835 if (!nm_i->dirty_nat_cnt)
2838 down_write(&nm_i->nat_tree_lock);
2841 * if there are no enough space in journal to store dirty nat
2842 * entries, remove all entries from journal and merge them
2843 * into nat entry set.
2845 if (enabled_nat_bits(sbi, cpc) ||
2846 !__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2847 remove_nats_in_journal(sbi);
2849 while ((found = __gang_lookup_nat_set(nm_i,
2850 set_idx, SETVEC_SIZE, setvec))) {
2852 set_idx = setvec[found - 1]->set + 1;
2853 for (idx = 0; idx < found; idx++)
2854 __adjust_nat_entry_set(setvec[idx], &sets,
2855 MAX_NAT_JENTRIES(journal));
2858 /* flush dirty nats in nat entry set */
2859 list_for_each_entry_safe(set, tmp, &sets, set_list) {
2860 err = __flush_nat_entry_set(sbi, set, cpc);
2865 up_write(&nm_i->nat_tree_lock);
2866 /* Allow dirty nats by node block allocation in write_begin */
2871 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2873 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2874 struct f2fs_nm_info *nm_i = NM_I(sbi);
2875 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2877 __u64 cp_ver = cur_cp_version(ckpt);
2878 block_t nat_bits_addr;
2880 if (!enabled_nat_bits(sbi, NULL))
2883 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
2884 nm_i->nat_bits = f2fs_kzalloc(sbi,
2885 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
2886 if (!nm_i->nat_bits)
2889 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2890 nm_i->nat_bits_blocks;
2891 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2894 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
2896 return PTR_ERR(page);
2898 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2899 page_address(page), F2FS_BLKSIZE);
2900 f2fs_put_page(page, 1);
2903 cp_ver |= (cur_cp_crc(ckpt) << 32);
2904 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2905 disable_nat_bits(sbi, true);
2909 nm_i->full_nat_bits = nm_i->nat_bits + 8;
2910 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2912 f2fs_notice(sbi, "Found nat_bits in checkpoint");
2916 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2918 struct f2fs_nm_info *nm_i = NM_I(sbi);
2920 nid_t nid, last_nid;
2922 if (!enabled_nat_bits(sbi, NULL))
2925 for (i = 0; i < nm_i->nat_blocks; i++) {
2926 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2927 if (i >= nm_i->nat_blocks)
2930 __set_bit_le(i, nm_i->nat_block_bitmap);
2932 nid = i * NAT_ENTRY_PER_BLOCK;
2933 last_nid = nid + NAT_ENTRY_PER_BLOCK;
2935 spin_lock(&NM_I(sbi)->nid_list_lock);
2936 for (; nid < last_nid; nid++)
2937 update_free_nid_bitmap(sbi, nid, true, true);
2938 spin_unlock(&NM_I(sbi)->nid_list_lock);
2941 for (i = 0; i < nm_i->nat_blocks; i++) {
2942 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2943 if (i >= nm_i->nat_blocks)
2946 __set_bit_le(i, nm_i->nat_block_bitmap);
2950 static int init_node_manager(struct f2fs_sb_info *sbi)
2952 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2953 struct f2fs_nm_info *nm_i = NM_I(sbi);
2954 unsigned char *version_bitmap;
2955 unsigned int nat_segs;
2958 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2960 /* segment_count_nat includes pair segment so divide to 2. */
2961 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2962 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2963 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
2965 /* not used nids: 0, node, meta, (and root counted as valid node) */
2966 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2967 sbi->nquota_files - F2FS_RESERVED_NODE_NUM;
2968 nm_i->nid_cnt[FREE_NID] = 0;
2969 nm_i->nid_cnt[PREALLOC_NID] = 0;
2971 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2972 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2973 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2975 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2976 INIT_LIST_HEAD(&nm_i->free_nid_list);
2977 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2978 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2979 INIT_LIST_HEAD(&nm_i->nat_entries);
2980 spin_lock_init(&nm_i->nat_list_lock);
2982 mutex_init(&nm_i->build_lock);
2983 spin_lock_init(&nm_i->nid_list_lock);
2984 init_rwsem(&nm_i->nat_tree_lock);
2986 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2987 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2988 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2989 if (!version_bitmap)
2992 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2994 if (!nm_i->nat_bitmap)
2997 err = __get_nat_bitmaps(sbi);
3001 #ifdef CONFIG_F2FS_CHECK_FS
3002 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3004 if (!nm_i->nat_bitmap_mir)
3011 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3013 struct f2fs_nm_info *nm_i = NM_I(sbi);
3016 nm_i->free_nid_bitmap =
3017 f2fs_kzalloc(sbi, array_size(sizeof(unsigned char *),
3020 if (!nm_i->free_nid_bitmap)
3023 for (i = 0; i < nm_i->nat_blocks; i++) {
3024 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3025 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3026 if (!nm_i->free_nid_bitmap[i])
3030 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3032 if (!nm_i->nat_block_bitmap)
3035 nm_i->free_nid_count =
3036 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3039 if (!nm_i->free_nid_count)
3044 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3048 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3053 err = init_node_manager(sbi);
3057 err = init_free_nid_cache(sbi);
3061 /* load free nid status from nat_bits table */
3062 load_free_nid_bitmap(sbi);
3064 return f2fs_build_free_nids(sbi, true, true);
3067 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3069 struct f2fs_nm_info *nm_i = NM_I(sbi);
3070 struct free_nid *i, *next_i;
3071 struct nat_entry *natvec[NATVEC_SIZE];
3072 struct nat_entry_set *setvec[SETVEC_SIZE];
3079 /* destroy free nid list */
3080 spin_lock(&nm_i->nid_list_lock);
3081 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3082 __remove_free_nid(sbi, i, FREE_NID);
3083 spin_unlock(&nm_i->nid_list_lock);
3084 kmem_cache_free(free_nid_slab, i);
3085 spin_lock(&nm_i->nid_list_lock);
3087 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3088 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3089 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3090 spin_unlock(&nm_i->nid_list_lock);
3092 /* destroy nat cache */
3093 down_write(&nm_i->nat_tree_lock);
3094 while ((found = __gang_lookup_nat_cache(nm_i,
3095 nid, NATVEC_SIZE, natvec))) {
3098 nid = nat_get_nid(natvec[found - 1]) + 1;
3099 for (idx = 0; idx < found; idx++) {
3100 spin_lock(&nm_i->nat_list_lock);
3101 list_del(&natvec[idx]->list);
3102 spin_unlock(&nm_i->nat_list_lock);
3104 __del_from_nat_cache(nm_i, natvec[idx]);
3107 f2fs_bug_on(sbi, nm_i->nat_cnt);
3109 /* destroy nat set cache */
3111 while ((found = __gang_lookup_nat_set(nm_i,
3112 nid, SETVEC_SIZE, setvec))) {
3115 nid = setvec[found - 1]->set + 1;
3116 for (idx = 0; idx < found; idx++) {
3117 /* entry_cnt is not zero, when cp_error was occurred */
3118 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3119 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3120 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3123 up_write(&nm_i->nat_tree_lock);
3125 kvfree(nm_i->nat_block_bitmap);
3126 if (nm_i->free_nid_bitmap) {
3129 for (i = 0; i < nm_i->nat_blocks; i++)
3130 kvfree(nm_i->free_nid_bitmap[i]);
3131 kvfree(nm_i->free_nid_bitmap);
3133 kvfree(nm_i->free_nid_count);
3135 kvfree(nm_i->nat_bitmap);
3136 kvfree(nm_i->nat_bits);
3137 #ifdef CONFIG_F2FS_CHECK_FS
3138 kvfree(nm_i->nat_bitmap_mir);
3140 sbi->nm_info = NULL;
3144 int __init f2fs_create_node_manager_caches(void)
3146 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
3147 sizeof(struct nat_entry));
3148 if (!nat_entry_slab)
3151 free_nid_slab = f2fs_kmem_cache_create("free_nid",
3152 sizeof(struct free_nid));
3154 goto destroy_nat_entry;
3156 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
3157 sizeof(struct nat_entry_set));
3158 if (!nat_entry_set_slab)
3159 goto destroy_free_nid;
3161 fsync_node_entry_slab = f2fs_kmem_cache_create("fsync_node_entry",
3162 sizeof(struct fsync_node_entry));
3163 if (!fsync_node_entry_slab)
3164 goto destroy_nat_entry_set;
3167 destroy_nat_entry_set:
3168 kmem_cache_destroy(nat_entry_set_slab);
3170 kmem_cache_destroy(free_nid_slab);
3172 kmem_cache_destroy(nat_entry_slab);
3177 void f2fs_destroy_node_manager_caches(void)
3179 kmem_cache_destroy(fsync_node_entry_slab);
3180 kmem_cache_destroy(nat_entry_set_slab);
3181 kmem_cache_destroy(free_nid_slab);
3182 kmem_cache_destroy(nat_entry_slab);