1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
9 #include <linux/f2fs_fs.h>
10 #include <linux/mpage.h>
11 #include <linux/backing-dev.h>
12 #include <linux/blkdev.h>
13 #include <linux/pagevec.h>
14 #include <linux/swap.h>
21 #include <trace/events/f2fs.h>
23 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
25 static struct kmem_cache *nat_entry_slab;
26 static struct kmem_cache *free_nid_slab;
27 static struct kmem_cache *nat_entry_set_slab;
28 static struct kmem_cache *fsync_node_entry_slab;
31 * Check whether the given nid is within node id range.
33 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
35 if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
36 set_sbi_flag(sbi, SBI_NEED_FSCK);
37 f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
44 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
46 struct f2fs_nm_info *nm_i = NM_I(sbi);
48 unsigned long avail_ram;
49 unsigned long mem_size = 0;
54 /* only uses low memory */
55 avail_ram = val.totalram - val.totalhigh;
58 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
60 if (type == FREE_NIDS) {
61 mem_size = (nm_i->nid_cnt[FREE_NID] *
62 sizeof(struct free_nid)) >> PAGE_SHIFT;
63 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
64 } else if (type == NAT_ENTRIES) {
65 mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
66 sizeof(struct nat_entry)) >> PAGE_SHIFT;
67 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
68 if (excess_cached_nats(sbi))
70 } else if (type == DIRTY_DENTS) {
71 if (sbi->sb->s_bdi->wb.dirty_exceeded)
73 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
74 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
75 } else if (type == INO_ENTRIES) {
78 for (i = 0; i < MAX_INO_ENTRY; i++)
79 mem_size += sbi->im[i].ino_num *
80 sizeof(struct ino_entry);
81 mem_size >>= PAGE_SHIFT;
82 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
83 } else if (type == EXTENT_CACHE) {
84 mem_size = (atomic_read(&sbi->total_ext_tree) *
85 sizeof(struct extent_tree) +
86 atomic_read(&sbi->total_ext_node) *
87 sizeof(struct extent_node)) >> PAGE_SHIFT;
88 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
89 } else if (type == INMEM_PAGES) {
90 /* it allows 20% / total_ram for inmemory pages */
91 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
92 res = mem_size < (val.totalram / 5);
94 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
100 static void clear_node_page_dirty(struct page *page)
102 if (PageDirty(page)) {
103 f2fs_clear_page_cache_dirty_tag(page);
104 clear_page_dirty_for_io(page);
105 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
107 ClearPageUptodate(page);
110 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
112 return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
115 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
117 struct page *src_page;
118 struct page *dst_page;
122 struct f2fs_nm_info *nm_i = NM_I(sbi);
124 dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
126 /* get current nat block page with lock */
127 src_page = get_current_nat_page(sbi, nid);
128 if (IS_ERR(src_page))
130 dst_page = f2fs_grab_meta_page(sbi, dst_off);
131 f2fs_bug_on(sbi, PageDirty(src_page));
133 src_addr = page_address(src_page);
134 dst_addr = page_address(dst_page);
135 memcpy(dst_addr, src_addr, PAGE_SIZE);
136 set_page_dirty(dst_page);
137 f2fs_put_page(src_page, 1);
139 set_to_next_nat(nm_i, nid);
144 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
146 struct nat_entry *new;
149 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
151 new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
153 nat_set_nid(new, nid);
159 static void __free_nat_entry(struct nat_entry *e)
161 kmem_cache_free(nat_entry_slab, e);
164 /* must be locked by nat_tree_lock */
165 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
166 struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
169 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
170 else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
174 node_info_from_raw_nat(&ne->ni, raw_ne);
176 spin_lock(&nm_i->nat_list_lock);
177 list_add_tail(&ne->list, &nm_i->nat_entries);
178 spin_unlock(&nm_i->nat_list_lock);
180 nm_i->nat_cnt[TOTAL_NAT]++;
181 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
185 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
187 struct nat_entry *ne;
189 ne = radix_tree_lookup(&nm_i->nat_root, n);
191 /* for recent accessed nat entry, move it to tail of lru list */
192 if (ne && !get_nat_flag(ne, IS_DIRTY)) {
193 spin_lock(&nm_i->nat_list_lock);
194 if (!list_empty(&ne->list))
195 list_move_tail(&ne->list, &nm_i->nat_entries);
196 spin_unlock(&nm_i->nat_list_lock);
202 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
203 nid_t start, unsigned int nr, struct nat_entry **ep)
205 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
208 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
210 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
211 nm_i->nat_cnt[TOTAL_NAT]--;
212 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
216 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
217 struct nat_entry *ne)
219 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
220 struct nat_entry_set *head;
222 head = radix_tree_lookup(&nm_i->nat_set_root, set);
224 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
226 INIT_LIST_HEAD(&head->entry_list);
227 INIT_LIST_HEAD(&head->set_list);
230 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
235 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
236 struct nat_entry *ne)
238 struct nat_entry_set *head;
239 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
242 head = __grab_nat_entry_set(nm_i, ne);
245 * update entry_cnt in below condition:
246 * 1. update NEW_ADDR to valid block address;
247 * 2. update old block address to new one;
249 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
250 !get_nat_flag(ne, IS_DIRTY)))
253 set_nat_flag(ne, IS_PREALLOC, new_ne);
255 if (get_nat_flag(ne, IS_DIRTY))
258 nm_i->nat_cnt[DIRTY_NAT]++;
259 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
260 set_nat_flag(ne, IS_DIRTY, true);
262 spin_lock(&nm_i->nat_list_lock);
264 list_del_init(&ne->list);
266 list_move_tail(&ne->list, &head->entry_list);
267 spin_unlock(&nm_i->nat_list_lock);
270 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
271 struct nat_entry_set *set, struct nat_entry *ne)
273 spin_lock(&nm_i->nat_list_lock);
274 list_move_tail(&ne->list, &nm_i->nat_entries);
275 spin_unlock(&nm_i->nat_list_lock);
277 set_nat_flag(ne, IS_DIRTY, false);
279 nm_i->nat_cnt[DIRTY_NAT]--;
280 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
283 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
284 nid_t start, unsigned int nr, struct nat_entry_set **ep)
286 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
290 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
292 return NODE_MAPPING(sbi) == page->mapping &&
293 IS_DNODE(page) && is_cold_node(page);
296 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
298 spin_lock_init(&sbi->fsync_node_lock);
299 INIT_LIST_HEAD(&sbi->fsync_node_list);
300 sbi->fsync_seg_id = 0;
301 sbi->fsync_node_num = 0;
304 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
307 struct fsync_node_entry *fn;
311 fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS);
315 INIT_LIST_HEAD(&fn->list);
317 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
318 list_add_tail(&fn->list, &sbi->fsync_node_list);
319 fn->seq_id = sbi->fsync_seg_id++;
321 sbi->fsync_node_num++;
322 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
327 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
329 struct fsync_node_entry *fn;
332 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
333 list_for_each_entry(fn, &sbi->fsync_node_list, list) {
334 if (fn->page == page) {
336 sbi->fsync_node_num--;
337 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
338 kmem_cache_free(fsync_node_entry_slab, fn);
343 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
347 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
351 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
352 sbi->fsync_seg_id = 0;
353 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
356 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
358 struct f2fs_nm_info *nm_i = NM_I(sbi);
362 down_read(&nm_i->nat_tree_lock);
363 e = __lookup_nat_cache(nm_i, nid);
365 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
366 !get_nat_flag(e, HAS_FSYNCED_INODE))
369 up_read(&nm_i->nat_tree_lock);
373 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
375 struct f2fs_nm_info *nm_i = NM_I(sbi);
379 down_read(&nm_i->nat_tree_lock);
380 e = __lookup_nat_cache(nm_i, nid);
381 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
383 up_read(&nm_i->nat_tree_lock);
387 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
389 struct f2fs_nm_info *nm_i = NM_I(sbi);
391 bool need_update = true;
393 down_read(&nm_i->nat_tree_lock);
394 e = __lookup_nat_cache(nm_i, ino);
395 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
396 (get_nat_flag(e, IS_CHECKPOINTED) ||
397 get_nat_flag(e, HAS_FSYNCED_INODE)))
399 up_read(&nm_i->nat_tree_lock);
403 /* must be locked by nat_tree_lock */
404 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
405 struct f2fs_nat_entry *ne)
407 struct f2fs_nm_info *nm_i = NM_I(sbi);
408 struct nat_entry *new, *e;
410 new = __alloc_nat_entry(nid, false);
414 down_write(&nm_i->nat_tree_lock);
415 e = __lookup_nat_cache(nm_i, nid);
417 e = __init_nat_entry(nm_i, new, ne, false);
419 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
420 nat_get_blkaddr(e) !=
421 le32_to_cpu(ne->block_addr) ||
422 nat_get_version(e) != ne->version);
423 up_write(&nm_i->nat_tree_lock);
425 __free_nat_entry(new);
428 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
429 block_t new_blkaddr, bool fsync_done)
431 struct f2fs_nm_info *nm_i = NM_I(sbi);
433 struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
435 down_write(&nm_i->nat_tree_lock);
436 e = __lookup_nat_cache(nm_i, ni->nid);
438 e = __init_nat_entry(nm_i, new, NULL, true);
439 copy_node_info(&e->ni, ni);
440 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
441 } else if (new_blkaddr == NEW_ADDR) {
443 * when nid is reallocated,
444 * previous nat entry can be remained in nat cache.
445 * So, reinitialize it with new information.
447 copy_node_info(&e->ni, ni);
448 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
450 /* let's free early to reduce memory consumption */
452 __free_nat_entry(new);
455 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
456 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
457 new_blkaddr == NULL_ADDR);
458 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
459 new_blkaddr == NEW_ADDR);
460 f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
461 new_blkaddr == NEW_ADDR);
463 /* increment version no as node is removed */
464 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
465 unsigned char version = nat_get_version(e);
466 nat_set_version(e, inc_node_version(version));
470 nat_set_blkaddr(e, new_blkaddr);
471 if (!__is_valid_data_blkaddr(new_blkaddr))
472 set_nat_flag(e, IS_CHECKPOINTED, false);
473 __set_nat_cache_dirty(nm_i, e);
475 /* update fsync_mark if its inode nat entry is still alive */
476 if (ni->nid != ni->ino)
477 e = __lookup_nat_cache(nm_i, ni->ino);
479 if (fsync_done && ni->nid == ni->ino)
480 set_nat_flag(e, HAS_FSYNCED_INODE, true);
481 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
483 up_write(&nm_i->nat_tree_lock);
486 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
488 struct f2fs_nm_info *nm_i = NM_I(sbi);
491 if (!down_write_trylock(&nm_i->nat_tree_lock))
494 spin_lock(&nm_i->nat_list_lock);
496 struct nat_entry *ne;
498 if (list_empty(&nm_i->nat_entries))
501 ne = list_first_entry(&nm_i->nat_entries,
502 struct nat_entry, list);
504 spin_unlock(&nm_i->nat_list_lock);
506 __del_from_nat_cache(nm_i, ne);
509 spin_lock(&nm_i->nat_list_lock);
511 spin_unlock(&nm_i->nat_list_lock);
513 up_write(&nm_i->nat_tree_lock);
514 return nr - nr_shrink;
518 * This function always returns success
520 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
521 struct node_info *ni)
523 struct f2fs_nm_info *nm_i = NM_I(sbi);
524 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
525 struct f2fs_journal *journal = curseg->journal;
526 nid_t start_nid = START_NID(nid);
527 struct f2fs_nat_block *nat_blk;
528 struct page *page = NULL;
529 struct f2fs_nat_entry ne;
537 /* Check nat cache */
538 down_read(&nm_i->nat_tree_lock);
539 e = __lookup_nat_cache(nm_i, nid);
541 ni->ino = nat_get_ino(e);
542 ni->blk_addr = nat_get_blkaddr(e);
543 ni->version = nat_get_version(e);
544 up_read(&nm_i->nat_tree_lock);
548 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
550 /* Check current segment summary */
551 down_read(&curseg->journal_rwsem);
552 i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
554 ne = nat_in_journal(journal, i);
555 node_info_from_raw_nat(ni, &ne);
557 up_read(&curseg->journal_rwsem);
559 up_read(&nm_i->nat_tree_lock);
563 /* Fill node_info from nat page */
564 index = current_nat_addr(sbi, nid);
565 up_read(&nm_i->nat_tree_lock);
567 page = f2fs_get_meta_page(sbi, index);
569 return PTR_ERR(page);
571 nat_blk = (struct f2fs_nat_block *)page_address(page);
572 ne = nat_blk->entries[nid - start_nid];
573 node_info_from_raw_nat(ni, &ne);
574 f2fs_put_page(page, 1);
576 blkaddr = le32_to_cpu(ne.block_addr);
577 if (__is_valid_data_blkaddr(blkaddr) &&
578 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
581 /* cache nat entry */
582 cache_nat_entry(sbi, nid, &ne);
587 * readahead MAX_RA_NODE number of node pages.
589 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
591 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
592 struct blk_plug plug;
596 blk_start_plug(&plug);
598 /* Then, try readahead for siblings of the desired node */
600 end = min(end, NIDS_PER_BLOCK);
601 for (i = start; i < end; i++) {
602 nid = get_nid(parent, i, false);
603 f2fs_ra_node_page(sbi, nid);
606 blk_finish_plug(&plug);
609 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
611 const long direct_index = ADDRS_PER_INODE(dn->inode);
612 const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
613 const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
614 unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
615 int cur_level = dn->cur_level;
616 int max_level = dn->max_level;
622 while (max_level-- > cur_level)
623 skipped_unit *= NIDS_PER_BLOCK;
625 switch (dn->max_level) {
627 base += 2 * indirect_blks;
630 base += 2 * direct_blks;
633 base += direct_index;
636 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
639 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
643 * The maximum depth is four.
644 * Offset[0] will have raw inode offset.
646 static int get_node_path(struct inode *inode, long block,
647 int offset[4], unsigned int noffset[4])
649 const long direct_index = ADDRS_PER_INODE(inode);
650 const long direct_blks = ADDRS_PER_BLOCK(inode);
651 const long dptrs_per_blk = NIDS_PER_BLOCK;
652 const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
653 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
659 if (block < direct_index) {
663 block -= direct_index;
664 if (block < direct_blks) {
665 offset[n++] = NODE_DIR1_BLOCK;
671 block -= direct_blks;
672 if (block < direct_blks) {
673 offset[n++] = NODE_DIR2_BLOCK;
679 block -= direct_blks;
680 if (block < indirect_blks) {
681 offset[n++] = NODE_IND1_BLOCK;
683 offset[n++] = block / direct_blks;
684 noffset[n] = 4 + offset[n - 1];
685 offset[n] = block % direct_blks;
689 block -= indirect_blks;
690 if (block < indirect_blks) {
691 offset[n++] = NODE_IND2_BLOCK;
692 noffset[n] = 4 + dptrs_per_blk;
693 offset[n++] = block / direct_blks;
694 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
695 offset[n] = block % direct_blks;
699 block -= indirect_blks;
700 if (block < dindirect_blks) {
701 offset[n++] = NODE_DIND_BLOCK;
702 noffset[n] = 5 + (dptrs_per_blk * 2);
703 offset[n++] = block / indirect_blks;
704 noffset[n] = 6 + (dptrs_per_blk * 2) +
705 offset[n - 1] * (dptrs_per_blk + 1);
706 offset[n++] = (block / direct_blks) % dptrs_per_blk;
707 noffset[n] = 7 + (dptrs_per_blk * 2) +
708 offset[n - 2] * (dptrs_per_blk + 1) +
710 offset[n] = block % direct_blks;
721 * Caller should call f2fs_put_dnode(dn).
722 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
723 * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
724 * In the case of RDONLY_NODE, we don't need to care about mutex.
726 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
728 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
729 struct page *npage[4];
730 struct page *parent = NULL;
732 unsigned int noffset[4];
737 level = get_node_path(dn->inode, index, offset, noffset);
741 nids[0] = dn->inode->i_ino;
742 npage[0] = dn->inode_page;
745 npage[0] = f2fs_get_node_page(sbi, nids[0]);
746 if (IS_ERR(npage[0]))
747 return PTR_ERR(npage[0]);
750 /* if inline_data is set, should not report any block indices */
751 if (f2fs_has_inline_data(dn->inode) && index) {
753 f2fs_put_page(npage[0], 1);
759 nids[1] = get_nid(parent, offset[0], true);
760 dn->inode_page = npage[0];
761 dn->inode_page_locked = true;
763 /* get indirect or direct nodes */
764 for (i = 1; i <= level; i++) {
767 if (!nids[i] && mode == ALLOC_NODE) {
769 if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
775 npage[i] = f2fs_new_node_page(dn, noffset[i]);
776 if (IS_ERR(npage[i])) {
777 f2fs_alloc_nid_failed(sbi, nids[i]);
778 err = PTR_ERR(npage[i]);
782 set_nid(parent, offset[i - 1], nids[i], i == 1);
783 f2fs_alloc_nid_done(sbi, nids[i]);
785 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
786 npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
787 if (IS_ERR(npage[i])) {
788 err = PTR_ERR(npage[i]);
794 dn->inode_page_locked = false;
797 f2fs_put_page(parent, 1);
801 npage[i] = f2fs_get_node_page(sbi, nids[i]);
802 if (IS_ERR(npage[i])) {
803 err = PTR_ERR(npage[i]);
804 f2fs_put_page(npage[0], 0);
810 nids[i + 1] = get_nid(parent, offset[i], false);
813 dn->nid = nids[level];
814 dn->ofs_in_node = offset[level];
815 dn->node_page = npage[level];
816 dn->data_blkaddr = datablock_addr(dn->inode,
817 dn->node_page, dn->ofs_in_node);
821 f2fs_put_page(parent, 1);
823 f2fs_put_page(npage[0], 0);
825 dn->inode_page = NULL;
826 dn->node_page = NULL;
827 if (err == -ENOENT) {
829 dn->max_level = level;
830 dn->ofs_in_node = offset[level];
835 static int truncate_node(struct dnode_of_data *dn)
837 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
842 err = f2fs_get_node_info(sbi, dn->nid, &ni);
846 /* Deallocate node address */
847 f2fs_invalidate_blocks(sbi, ni.blk_addr);
848 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
849 set_node_addr(sbi, &ni, NULL_ADDR, false);
851 if (dn->nid == dn->inode->i_ino) {
852 f2fs_remove_orphan_inode(sbi, dn->nid);
853 dec_valid_inode_count(sbi);
854 f2fs_inode_synced(dn->inode);
857 clear_node_page_dirty(dn->node_page);
858 set_sbi_flag(sbi, SBI_IS_DIRTY);
860 index = dn->node_page->index;
861 f2fs_put_page(dn->node_page, 1);
863 invalidate_mapping_pages(NODE_MAPPING(sbi),
866 dn->node_page = NULL;
867 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
872 static int truncate_dnode(struct dnode_of_data *dn)
880 /* get direct node */
881 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
882 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
884 else if (IS_ERR(page))
885 return PTR_ERR(page);
887 /* Make dnode_of_data for parameter */
888 dn->node_page = page;
890 f2fs_truncate_data_blocks(dn);
891 err = truncate_node(dn);
893 f2fs_put_page(page, 1);
900 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
903 struct dnode_of_data rdn = *dn;
905 struct f2fs_node *rn;
907 unsigned int child_nofs;
912 return NIDS_PER_BLOCK + 1;
914 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
916 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
918 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
919 return PTR_ERR(page);
922 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
924 rn = F2FS_NODE(page);
926 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
927 child_nid = le32_to_cpu(rn->in.nid[i]);
931 ret = truncate_dnode(&rdn);
934 if (set_nid(page, i, 0, false))
935 dn->node_changed = true;
938 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
939 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
940 child_nid = le32_to_cpu(rn->in.nid[i]);
941 if (child_nid == 0) {
942 child_nofs += NIDS_PER_BLOCK + 1;
946 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
947 if (ret == (NIDS_PER_BLOCK + 1)) {
948 if (set_nid(page, i, 0, false))
949 dn->node_changed = true;
951 } else if (ret < 0 && ret != -ENOENT) {
959 /* remove current indirect node */
960 dn->node_page = page;
961 ret = truncate_node(dn);
966 f2fs_put_page(page, 1);
968 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
972 f2fs_put_page(page, 1);
973 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
977 static int truncate_partial_nodes(struct dnode_of_data *dn,
978 struct f2fs_inode *ri, int *offset, int depth)
980 struct page *pages[2];
987 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
991 /* get indirect nodes in the path */
992 for (i = 0; i < idx + 1; i++) {
993 /* reference count'll be increased */
994 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
995 if (IS_ERR(pages[i])) {
996 err = PTR_ERR(pages[i]);
1000 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1003 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1005 /* free direct nodes linked to a partial indirect node */
1006 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1007 child_nid = get_nid(pages[idx], i, false);
1010 dn->nid = child_nid;
1011 err = truncate_dnode(dn);
1014 if (set_nid(pages[idx], i, 0, false))
1015 dn->node_changed = true;
1018 if (offset[idx + 1] == 0) {
1019 dn->node_page = pages[idx];
1021 err = truncate_node(dn);
1025 f2fs_put_page(pages[idx], 1);
1028 offset[idx + 1] = 0;
1031 for (i = idx; i >= 0; i--)
1032 f2fs_put_page(pages[i], 1);
1034 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1040 * All the block addresses of data and nodes should be nullified.
1042 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1044 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1045 int err = 0, cont = 1;
1046 int level, offset[4], noffset[4];
1047 unsigned int nofs = 0;
1048 struct f2fs_inode *ri;
1049 struct dnode_of_data dn;
1052 trace_f2fs_truncate_inode_blocks_enter(inode, from);
1054 level = get_node_path(inode, from, offset, noffset);
1058 page = f2fs_get_node_page(sbi, inode->i_ino);
1060 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1061 return PTR_ERR(page);
1064 set_new_dnode(&dn, inode, page, NULL, 0);
1067 ri = F2FS_INODE(page);
1075 if (!offset[level - 1])
1077 err = truncate_partial_nodes(&dn, ri, offset, level);
1078 if (err < 0 && err != -ENOENT)
1080 nofs += 1 + NIDS_PER_BLOCK;
1083 nofs = 5 + 2 * NIDS_PER_BLOCK;
1084 if (!offset[level - 1])
1086 err = truncate_partial_nodes(&dn, ri, offset, level);
1087 if (err < 0 && err != -ENOENT)
1096 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1097 switch (offset[0]) {
1098 case NODE_DIR1_BLOCK:
1099 case NODE_DIR2_BLOCK:
1100 err = truncate_dnode(&dn);
1103 case NODE_IND1_BLOCK:
1104 case NODE_IND2_BLOCK:
1105 err = truncate_nodes(&dn, nofs, offset[1], 2);
1108 case NODE_DIND_BLOCK:
1109 err = truncate_nodes(&dn, nofs, offset[1], 3);
1116 if (err < 0 && err != -ENOENT)
1118 if (offset[1] == 0 &&
1119 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1121 BUG_ON(page->mapping != NODE_MAPPING(sbi));
1122 f2fs_wait_on_page_writeback(page, NODE, true, true);
1123 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1124 set_page_dirty(page);
1132 f2fs_put_page(page, 0);
1133 trace_f2fs_truncate_inode_blocks_exit(inode, err);
1134 return err > 0 ? 0 : err;
1137 /* caller must lock inode page */
1138 int f2fs_truncate_xattr_node(struct inode *inode)
1140 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1141 nid_t nid = F2FS_I(inode)->i_xattr_nid;
1142 struct dnode_of_data dn;
1149 npage = f2fs_get_node_page(sbi, nid);
1151 return PTR_ERR(npage);
1153 set_new_dnode(&dn, inode, NULL, npage, nid);
1154 err = truncate_node(&dn);
1156 f2fs_put_page(npage, 1);
1160 f2fs_i_xnid_write(inode, 0);
1166 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1169 int f2fs_remove_inode_page(struct inode *inode)
1171 struct dnode_of_data dn;
1174 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1175 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1179 err = f2fs_truncate_xattr_node(inode);
1181 f2fs_put_dnode(&dn);
1185 /* remove potential inline_data blocks */
1186 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1187 S_ISLNK(inode->i_mode))
1188 f2fs_truncate_data_blocks_range(&dn, 1);
1190 /* 0 is possible, after f2fs_new_inode() has failed */
1191 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1192 f2fs_put_dnode(&dn);
1196 if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1197 f2fs_warn(F2FS_I_SB(inode), "Inconsistent i_blocks, ino:%lu, iblocks:%llu",
1198 inode->i_ino, (unsigned long long)inode->i_blocks);
1199 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1202 /* will put inode & node pages */
1203 err = truncate_node(&dn);
1205 f2fs_put_dnode(&dn);
1211 struct page *f2fs_new_inode_page(struct inode *inode)
1213 struct dnode_of_data dn;
1215 /* allocate inode page for new inode */
1216 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1218 /* caller should f2fs_put_page(page, 1); */
1219 return f2fs_new_node_page(&dn, 0);
1222 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1224 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1225 struct node_info new_ni;
1229 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1230 return ERR_PTR(-EPERM);
1232 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1234 return ERR_PTR(-ENOMEM);
1236 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1239 #ifdef CONFIG_F2FS_CHECK_FS
1240 err = f2fs_get_node_info(sbi, dn->nid, &new_ni);
1242 dec_valid_node_count(sbi, dn->inode, !ofs);
1245 if (unlikely(new_ni.blk_addr != NULL_ADDR)) {
1246 err = -EFSCORRUPTED;
1247 set_sbi_flag(sbi, SBI_NEED_FSCK);
1251 new_ni.nid = dn->nid;
1252 new_ni.ino = dn->inode->i_ino;
1253 new_ni.blk_addr = NULL_ADDR;
1256 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1258 f2fs_wait_on_page_writeback(page, NODE, true, true);
1259 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1260 set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1261 if (!PageUptodate(page))
1262 SetPageUptodate(page);
1263 if (set_page_dirty(page))
1264 dn->node_changed = true;
1266 if (f2fs_has_xattr_block(ofs))
1267 f2fs_i_xnid_write(dn->inode, dn->nid);
1270 inc_valid_inode_count(sbi);
1274 clear_node_page_dirty(page);
1275 f2fs_put_page(page, 1);
1276 return ERR_PTR(err);
1280 * Caller should do after getting the following values.
1281 * 0: f2fs_put_page(page, 0)
1282 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1284 static int read_node_page(struct page *page, int op_flags)
1286 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1287 struct node_info ni;
1288 struct f2fs_io_info fio = {
1292 .op_flags = op_flags,
1294 .encrypted_page = NULL,
1298 if (PageUptodate(page)) {
1299 if (!f2fs_inode_chksum_verify(sbi, page)) {
1300 ClearPageUptodate(page);
1306 err = f2fs_get_node_info(sbi, page->index, &ni);
1310 if (unlikely(ni.blk_addr == NULL_ADDR) ||
1311 is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1312 ClearPageUptodate(page);
1316 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1317 return f2fs_submit_page_bio(&fio);
1321 * Readahead a node page
1323 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1330 if (f2fs_check_nid_range(sbi, nid))
1333 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1337 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1341 err = read_node_page(apage, REQ_RAHEAD);
1342 f2fs_put_page(apage, err ? 1 : 0);
1345 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1346 struct page *parent, int start)
1352 return ERR_PTR(-ENOENT);
1353 if (f2fs_check_nid_range(sbi, nid))
1354 return ERR_PTR(-EINVAL);
1356 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1358 return ERR_PTR(-ENOMEM);
1360 err = read_node_page(page, 0);
1362 f2fs_put_page(page, 1);
1363 return ERR_PTR(err);
1364 } else if (err == LOCKED_PAGE) {
1370 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1374 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1375 f2fs_put_page(page, 1);
1379 if (unlikely(!PageUptodate(page))) {
1384 if (!f2fs_inode_chksum_verify(sbi, page)) {
1389 if(unlikely(nid != nid_of_node(page))) {
1390 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1391 nid, nid_of_node(page), ino_of_node(page),
1392 ofs_of_node(page), cpver_of_node(page),
1393 next_blkaddr_of_node(page));
1394 set_sbi_flag(sbi, SBI_NEED_FSCK);
1397 ClearPageUptodate(page);
1398 f2fs_put_page(page, 1);
1399 return ERR_PTR(err);
1404 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1406 return __get_node_page(sbi, nid, NULL, 0);
1409 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1411 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1412 nid_t nid = get_nid(parent, start, false);
1414 return __get_node_page(sbi, nid, parent, start);
1417 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1419 struct inode *inode;
1423 /* should flush inline_data before evict_inode */
1424 inode = ilookup(sbi->sb, ino);
1428 page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1429 FGP_LOCK|FGP_NOWAIT, 0);
1433 if (!PageUptodate(page))
1436 if (!PageDirty(page))
1439 if (!clear_page_dirty_for_io(page))
1442 ret = f2fs_write_inline_data(inode, page);
1443 inode_dec_dirty_pages(inode);
1444 f2fs_remove_dirty_inode(inode);
1446 set_page_dirty(page);
1448 f2fs_put_page(page, 1);
1453 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1456 struct pagevec pvec;
1457 struct page *last_page = NULL;
1460 pagevec_init(&pvec);
1463 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1464 PAGECACHE_TAG_DIRTY))) {
1467 for (i = 0; i < nr_pages; i++) {
1468 struct page *page = pvec.pages[i];
1470 if (unlikely(f2fs_cp_error(sbi))) {
1471 f2fs_put_page(last_page, 0);
1472 pagevec_release(&pvec);
1473 return ERR_PTR(-EIO);
1476 if (!IS_DNODE(page) || !is_cold_node(page))
1478 if (ino_of_node(page) != ino)
1483 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1488 if (ino_of_node(page) != ino)
1489 goto continue_unlock;
1491 if (!PageDirty(page)) {
1492 /* someone wrote it for us */
1493 goto continue_unlock;
1497 f2fs_put_page(last_page, 0);
1503 pagevec_release(&pvec);
1509 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1510 struct writeback_control *wbc, bool do_balance,
1511 enum iostat_type io_type, unsigned int *seq_id)
1513 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1515 struct node_info ni;
1516 struct f2fs_io_info fio = {
1518 .ino = ino_of_node(page),
1521 .op_flags = wbc_to_write_flags(wbc),
1523 .encrypted_page = NULL,
1530 trace_f2fs_writepage(page, NODE);
1532 if (unlikely(f2fs_cp_error(sbi)))
1535 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1538 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1539 wbc->sync_mode == WB_SYNC_NONE &&
1540 IS_DNODE(page) && is_cold_node(page))
1543 /* get old block addr of this node page */
1544 nid = nid_of_node(page);
1545 f2fs_bug_on(sbi, page->index != nid);
1547 if (f2fs_get_node_info(sbi, nid, &ni))
1550 if (wbc->for_reclaim) {
1551 if (!down_read_trylock(&sbi->node_write))
1554 down_read(&sbi->node_write);
1557 /* This page is already truncated */
1558 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1559 ClearPageUptodate(page);
1560 dec_page_count(sbi, F2FS_DIRTY_NODES);
1561 up_read(&sbi->node_write);
1566 if (__is_valid_data_blkaddr(ni.blk_addr) &&
1567 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1568 DATA_GENERIC_ENHANCE)) {
1569 up_read(&sbi->node_write);
1573 if (atomic && !test_opt(sbi, NOBARRIER))
1574 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1576 /* should add to global list before clearing PAGECACHE status */
1577 if (f2fs_in_warm_node_list(sbi, page)) {
1578 seq = f2fs_add_fsync_node_entry(sbi, page);
1583 set_page_writeback(page);
1584 ClearPageError(page);
1586 fio.old_blkaddr = ni.blk_addr;
1587 f2fs_do_write_node_page(nid, &fio);
1588 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1589 dec_page_count(sbi, F2FS_DIRTY_NODES);
1590 up_read(&sbi->node_write);
1592 if (wbc->for_reclaim) {
1593 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1599 if (unlikely(f2fs_cp_error(sbi))) {
1600 f2fs_submit_merged_write(sbi, NODE);
1604 *submitted = fio.submitted;
1607 f2fs_balance_fs(sbi, false);
1611 redirty_page_for_writepage(wbc, page);
1612 return AOP_WRITEPAGE_ACTIVATE;
1615 int f2fs_move_node_page(struct page *node_page, int gc_type)
1619 if (gc_type == FG_GC) {
1620 struct writeback_control wbc = {
1621 .sync_mode = WB_SYNC_ALL,
1626 f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1628 set_page_dirty(node_page);
1630 if (!clear_page_dirty_for_io(node_page)) {
1635 if (__write_node_page(node_page, false, NULL,
1636 &wbc, false, FS_GC_NODE_IO, NULL)) {
1638 unlock_page(node_page);
1642 /* set page dirty and write it */
1643 if (!PageWriteback(node_page))
1644 set_page_dirty(node_page);
1647 unlock_page(node_page);
1649 f2fs_put_page(node_page, 0);
1653 static int f2fs_write_node_page(struct page *page,
1654 struct writeback_control *wbc)
1656 return __write_node_page(page, false, NULL, wbc, false,
1660 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1661 struct writeback_control *wbc, bool atomic,
1662 unsigned int *seq_id)
1665 struct pagevec pvec;
1667 struct page *last_page = NULL;
1668 bool marked = false;
1669 nid_t ino = inode->i_ino;
1674 last_page = last_fsync_dnode(sbi, ino);
1675 if (IS_ERR_OR_NULL(last_page))
1676 return PTR_ERR_OR_ZERO(last_page);
1679 pagevec_init(&pvec);
1682 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1683 PAGECACHE_TAG_DIRTY))) {
1686 for (i = 0; i < nr_pages; i++) {
1687 struct page *page = pvec.pages[i];
1688 bool submitted = false;
1690 if (unlikely(f2fs_cp_error(sbi))) {
1691 f2fs_put_page(last_page, 0);
1692 pagevec_release(&pvec);
1697 if (!IS_DNODE(page) || !is_cold_node(page))
1699 if (ino_of_node(page) != ino)
1704 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1709 if (ino_of_node(page) != ino)
1710 goto continue_unlock;
1712 if (!PageDirty(page) && page != last_page) {
1713 /* someone wrote it for us */
1714 goto continue_unlock;
1717 f2fs_wait_on_page_writeback(page, NODE, true, true);
1719 set_fsync_mark(page, 0);
1720 set_dentry_mark(page, 0);
1722 if (!atomic || page == last_page) {
1723 set_fsync_mark(page, 1);
1724 if (IS_INODE(page)) {
1725 if (is_inode_flag_set(inode,
1727 f2fs_update_inode(inode, page);
1728 set_dentry_mark(page,
1729 f2fs_need_dentry_mark(sbi, ino));
1731 /* may be written by other thread */
1732 if (!PageDirty(page))
1733 set_page_dirty(page);
1736 if (!clear_page_dirty_for_io(page))
1737 goto continue_unlock;
1739 ret = __write_node_page(page, atomic &&
1741 &submitted, wbc, true,
1742 FS_NODE_IO, seq_id);
1745 f2fs_put_page(last_page, 0);
1747 } else if (submitted) {
1751 if (page == last_page) {
1752 f2fs_put_page(page, 0);
1757 pagevec_release(&pvec);
1763 if (!ret && atomic && !marked) {
1764 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1765 ino, last_page->index);
1766 lock_page(last_page);
1767 f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1768 set_page_dirty(last_page);
1769 unlock_page(last_page);
1774 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1775 return ret ? -EIO: 0;
1778 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1780 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1783 if (inode->i_ino != ino)
1786 if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1789 spin_lock(&sbi->inode_lock[DIRTY_META]);
1790 clean = list_empty(&F2FS_I(inode)->gdirty_list);
1791 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1796 inode = igrab(inode);
1802 static bool flush_dirty_inode(struct page *page)
1804 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1805 struct inode *inode;
1806 nid_t ino = ino_of_node(page);
1808 inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1812 f2fs_update_inode(inode, page);
1819 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1820 struct writeback_control *wbc,
1821 bool do_balance, enum iostat_type io_type)
1824 struct pagevec pvec;
1828 int nr_pages, done = 0;
1830 pagevec_init(&pvec);
1835 while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1836 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1839 for (i = 0; i < nr_pages; i++) {
1840 struct page *page = pvec.pages[i];
1841 bool submitted = false;
1842 bool may_dirty = true;
1844 /* give a priority to WB_SYNC threads */
1845 if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1846 wbc->sync_mode == WB_SYNC_NONE) {
1852 * flushing sequence with step:
1857 if (step == 0 && IS_DNODE(page))
1859 if (step == 1 && (!IS_DNODE(page) ||
1860 is_cold_node(page)))
1862 if (step == 2 && (!IS_DNODE(page) ||
1863 !is_cold_node(page)))
1866 if (wbc->sync_mode == WB_SYNC_ALL)
1868 else if (!trylock_page(page))
1871 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1877 if (!PageDirty(page)) {
1878 /* someone wrote it for us */
1879 goto continue_unlock;
1882 /* flush inline_data */
1883 if (is_inline_node(page)) {
1884 clear_inline_node(page);
1886 flush_inline_data(sbi, ino_of_node(page));
1890 /* flush dirty inode */
1891 if (IS_INODE(page) && may_dirty) {
1893 if (flush_dirty_inode(page))
1897 f2fs_wait_on_page_writeback(page, NODE, true, true);
1899 if (!clear_page_dirty_for_io(page))
1900 goto continue_unlock;
1902 set_fsync_mark(page, 0);
1903 set_dentry_mark(page, 0);
1905 ret = __write_node_page(page, false, &submitted,
1906 wbc, do_balance, io_type, NULL);
1912 if (--wbc->nr_to_write == 0)
1915 pagevec_release(&pvec);
1918 if (wbc->nr_to_write == 0) {
1925 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1926 wbc->sync_mode == WB_SYNC_NONE && step == 1)
1933 f2fs_submit_merged_write(sbi, NODE);
1935 if (unlikely(f2fs_cp_error(sbi)))
1940 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
1941 unsigned int seq_id)
1943 struct fsync_node_entry *fn;
1945 struct list_head *head = &sbi->fsync_node_list;
1946 unsigned long flags;
1947 unsigned int cur_seq_id = 0;
1950 while (seq_id && cur_seq_id < seq_id) {
1951 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
1952 if (list_empty(head)) {
1953 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1956 fn = list_first_entry(head, struct fsync_node_entry, list);
1957 if (fn->seq_id > seq_id) {
1958 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1961 cur_seq_id = fn->seq_id;
1964 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
1966 f2fs_wait_on_page_writeback(page, NODE, true, false);
1967 if (TestClearPageError(page))
1976 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1983 static int f2fs_write_node_pages(struct address_space *mapping,
1984 struct writeback_control *wbc)
1986 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1987 struct blk_plug plug;
1990 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1993 /* balancing f2fs's metadata in background */
1994 f2fs_balance_fs_bg(sbi);
1996 /* collect a number of dirty node pages and write together */
1997 if (wbc->sync_mode != WB_SYNC_ALL &&
1998 get_pages(sbi, F2FS_DIRTY_NODES) <
1999 nr_pages_to_skip(sbi, NODE))
2002 if (wbc->sync_mode == WB_SYNC_ALL)
2003 atomic_inc(&sbi->wb_sync_req[NODE]);
2004 else if (atomic_read(&sbi->wb_sync_req[NODE])) {
2005 /* to avoid potential deadlock */
2007 blk_finish_plug(current->plug);
2011 trace_f2fs_writepages(mapping->host, wbc, NODE);
2013 diff = nr_pages_to_write(sbi, NODE, wbc);
2014 blk_start_plug(&plug);
2015 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2016 blk_finish_plug(&plug);
2017 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2019 if (wbc->sync_mode == WB_SYNC_ALL)
2020 atomic_dec(&sbi->wb_sync_req[NODE]);
2024 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2025 trace_f2fs_writepages(mapping->host, wbc, NODE);
2029 static int f2fs_set_node_page_dirty(struct page *page)
2031 trace_f2fs_set_page_dirty(page, NODE);
2033 if (!PageUptodate(page))
2034 SetPageUptodate(page);
2035 #ifdef CONFIG_F2FS_CHECK_FS
2037 f2fs_inode_chksum_set(F2FS_P_SB(page), page);
2039 if (!PageDirty(page)) {
2040 __set_page_dirty_nobuffers(page);
2041 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
2042 f2fs_set_page_private(page, 0);
2043 f2fs_trace_pid(page);
2050 * Structure of the f2fs node operations
2052 const struct address_space_operations f2fs_node_aops = {
2053 .writepage = f2fs_write_node_page,
2054 .writepages = f2fs_write_node_pages,
2055 .set_page_dirty = f2fs_set_node_page_dirty,
2056 .invalidatepage = f2fs_invalidate_page,
2057 .releasepage = f2fs_release_page,
2058 #ifdef CONFIG_MIGRATION
2059 .migratepage = f2fs_migrate_page,
2063 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2066 return radix_tree_lookup(&nm_i->free_nid_root, n);
2069 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2070 struct free_nid *i, enum nid_state state)
2072 struct f2fs_nm_info *nm_i = NM_I(sbi);
2074 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2078 f2fs_bug_on(sbi, state != i->state);
2079 nm_i->nid_cnt[state]++;
2080 if (state == FREE_NID)
2081 list_add_tail(&i->list, &nm_i->free_nid_list);
2085 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2086 struct free_nid *i, enum nid_state state)
2088 struct f2fs_nm_info *nm_i = NM_I(sbi);
2090 f2fs_bug_on(sbi, state != i->state);
2091 nm_i->nid_cnt[state]--;
2092 if (state == FREE_NID)
2094 radix_tree_delete(&nm_i->free_nid_root, i->nid);
2097 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2098 enum nid_state org_state, enum nid_state dst_state)
2100 struct f2fs_nm_info *nm_i = NM_I(sbi);
2102 f2fs_bug_on(sbi, org_state != i->state);
2103 i->state = dst_state;
2104 nm_i->nid_cnt[org_state]--;
2105 nm_i->nid_cnt[dst_state]++;
2107 switch (dst_state) {
2112 list_add_tail(&i->list, &nm_i->free_nid_list);
2119 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2120 bool set, bool build)
2122 struct f2fs_nm_info *nm_i = NM_I(sbi);
2123 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2124 unsigned int nid_ofs = nid - START_NID(nid);
2126 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2130 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2132 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2133 nm_i->free_nid_count[nat_ofs]++;
2135 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2137 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2139 nm_i->free_nid_count[nat_ofs]--;
2143 /* return if the nid is recognized as free */
2144 static bool add_free_nid(struct f2fs_sb_info *sbi,
2145 nid_t nid, bool build, bool update)
2147 struct f2fs_nm_info *nm_i = NM_I(sbi);
2148 struct free_nid *i, *e;
2149 struct nat_entry *ne;
2153 /* 0 nid should not be used */
2154 if (unlikely(nid == 0))
2157 if (unlikely(f2fs_check_nid_range(sbi, nid)))
2160 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2162 i->state = FREE_NID;
2164 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2166 spin_lock(&nm_i->nid_list_lock);
2174 * - __insert_nid_to_list(PREALLOC_NID)
2175 * - f2fs_balance_fs_bg
2176 * - f2fs_build_free_nids
2177 * - __f2fs_build_free_nids
2180 * - __lookup_nat_cache
2182 * - f2fs_init_inode_metadata
2183 * - f2fs_new_inode_page
2184 * - f2fs_new_node_page
2186 * - f2fs_alloc_nid_done
2187 * - __remove_nid_from_list(PREALLOC_NID)
2188 * - __insert_nid_to_list(FREE_NID)
2190 ne = __lookup_nat_cache(nm_i, nid);
2191 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2192 nat_get_blkaddr(ne) != NULL_ADDR))
2195 e = __lookup_free_nid_list(nm_i, nid);
2197 if (e->state == FREE_NID)
2203 err = __insert_free_nid(sbi, i, FREE_NID);
2206 update_free_nid_bitmap(sbi, nid, ret, build);
2208 nm_i->available_nids++;
2210 spin_unlock(&nm_i->nid_list_lock);
2211 radix_tree_preload_end();
2214 kmem_cache_free(free_nid_slab, i);
2218 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2220 struct f2fs_nm_info *nm_i = NM_I(sbi);
2222 bool need_free = false;
2224 spin_lock(&nm_i->nid_list_lock);
2225 i = __lookup_free_nid_list(nm_i, nid);
2226 if (i && i->state == FREE_NID) {
2227 __remove_free_nid(sbi, i, FREE_NID);
2230 spin_unlock(&nm_i->nid_list_lock);
2233 kmem_cache_free(free_nid_slab, i);
2236 static int scan_nat_page(struct f2fs_sb_info *sbi,
2237 struct page *nat_page, nid_t start_nid)
2239 struct f2fs_nm_info *nm_i = NM_I(sbi);
2240 struct f2fs_nat_block *nat_blk = page_address(nat_page);
2242 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2245 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2247 i = start_nid % NAT_ENTRY_PER_BLOCK;
2249 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2250 if (unlikely(start_nid >= nm_i->max_nid))
2253 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2255 if (blk_addr == NEW_ADDR)
2258 if (blk_addr == NULL_ADDR) {
2259 add_free_nid(sbi, start_nid, true, true);
2261 spin_lock(&NM_I(sbi)->nid_list_lock);
2262 update_free_nid_bitmap(sbi, start_nid, false, true);
2263 spin_unlock(&NM_I(sbi)->nid_list_lock);
2270 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2272 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2273 struct f2fs_journal *journal = curseg->journal;
2276 down_read(&curseg->journal_rwsem);
2277 for (i = 0; i < nats_in_cursum(journal); i++) {
2281 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2282 nid = le32_to_cpu(nid_in_journal(journal, i));
2283 if (addr == NULL_ADDR)
2284 add_free_nid(sbi, nid, true, false);
2286 remove_free_nid(sbi, nid);
2288 up_read(&curseg->journal_rwsem);
2291 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2293 struct f2fs_nm_info *nm_i = NM_I(sbi);
2294 unsigned int i, idx;
2297 down_read(&nm_i->nat_tree_lock);
2299 for (i = 0; i < nm_i->nat_blocks; i++) {
2300 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2302 if (!nm_i->free_nid_count[i])
2304 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2305 idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2306 NAT_ENTRY_PER_BLOCK, idx);
2307 if (idx >= NAT_ENTRY_PER_BLOCK)
2310 nid = i * NAT_ENTRY_PER_BLOCK + idx;
2311 add_free_nid(sbi, nid, true, false);
2313 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2318 scan_curseg_cache(sbi);
2320 up_read(&nm_i->nat_tree_lock);
2323 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2324 bool sync, bool mount)
2326 struct f2fs_nm_info *nm_i = NM_I(sbi);
2328 nid_t nid = nm_i->next_scan_nid;
2330 if (unlikely(nid >= nm_i->max_nid))
2333 if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2334 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2336 /* Enough entries */
2337 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2340 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2344 /* try to find free nids in free_nid_bitmap */
2345 scan_free_nid_bits(sbi);
2347 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2351 /* readahead nat pages to be scanned */
2352 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2355 down_read(&nm_i->nat_tree_lock);
2358 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2359 nm_i->nat_block_bitmap)) {
2360 struct page *page = get_current_nat_page(sbi, nid);
2363 ret = PTR_ERR(page);
2365 ret = scan_nat_page(sbi, page, nid);
2366 f2fs_put_page(page, 1);
2370 up_read(&nm_i->nat_tree_lock);
2371 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2376 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2377 if (unlikely(nid >= nm_i->max_nid))
2380 if (++i >= FREE_NID_PAGES)
2384 /* go to the next free nat pages to find free nids abundantly */
2385 nm_i->next_scan_nid = nid;
2387 /* find free nids from current sum_pages */
2388 scan_curseg_cache(sbi);
2390 up_read(&nm_i->nat_tree_lock);
2392 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2393 nm_i->ra_nid_pages, META_NAT, false);
2398 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2402 mutex_lock(&NM_I(sbi)->build_lock);
2403 ret = __f2fs_build_free_nids(sbi, sync, mount);
2404 mutex_unlock(&NM_I(sbi)->build_lock);
2410 * If this function returns success, caller can obtain a new nid
2411 * from second parameter of this function.
2412 * The returned nid could be used ino as well as nid when inode is created.
2414 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2416 struct f2fs_nm_info *nm_i = NM_I(sbi);
2417 struct free_nid *i = NULL;
2419 if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2420 f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2424 spin_lock(&nm_i->nid_list_lock);
2426 if (unlikely(nm_i->available_nids == 0)) {
2427 spin_unlock(&nm_i->nid_list_lock);
2431 /* We should not use stale free nids created by f2fs_build_free_nids */
2432 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2433 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2434 i = list_first_entry(&nm_i->free_nid_list,
2435 struct free_nid, list);
2438 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2439 nm_i->available_nids--;
2441 update_free_nid_bitmap(sbi, *nid, false, false);
2443 spin_unlock(&nm_i->nid_list_lock);
2446 spin_unlock(&nm_i->nid_list_lock);
2448 /* Let's scan nat pages and its caches to get free nids */
2449 if (!f2fs_build_free_nids(sbi, true, false))
2455 * f2fs_alloc_nid() should be called prior to this function.
2457 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2459 struct f2fs_nm_info *nm_i = NM_I(sbi);
2462 spin_lock(&nm_i->nid_list_lock);
2463 i = __lookup_free_nid_list(nm_i, nid);
2464 f2fs_bug_on(sbi, !i);
2465 __remove_free_nid(sbi, i, PREALLOC_NID);
2466 spin_unlock(&nm_i->nid_list_lock);
2468 kmem_cache_free(free_nid_slab, i);
2472 * f2fs_alloc_nid() should be called prior to this function.
2474 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2476 struct f2fs_nm_info *nm_i = NM_I(sbi);
2478 bool need_free = false;
2483 spin_lock(&nm_i->nid_list_lock);
2484 i = __lookup_free_nid_list(nm_i, nid);
2485 f2fs_bug_on(sbi, !i);
2487 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2488 __remove_free_nid(sbi, i, PREALLOC_NID);
2491 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2494 nm_i->available_nids++;
2496 update_free_nid_bitmap(sbi, nid, true, false);
2498 spin_unlock(&nm_i->nid_list_lock);
2501 kmem_cache_free(free_nid_slab, i);
2504 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2506 struct f2fs_nm_info *nm_i = NM_I(sbi);
2507 struct free_nid *i, *next;
2510 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2513 if (!mutex_trylock(&nm_i->build_lock))
2516 spin_lock(&nm_i->nid_list_lock);
2517 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2518 if (nr_shrink <= 0 ||
2519 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2522 __remove_free_nid(sbi, i, FREE_NID);
2523 kmem_cache_free(free_nid_slab, i);
2526 spin_unlock(&nm_i->nid_list_lock);
2527 mutex_unlock(&nm_i->build_lock);
2529 return nr - nr_shrink;
2532 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2534 void *src_addr, *dst_addr;
2537 struct f2fs_inode *ri;
2539 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2541 return PTR_ERR(ipage);
2543 ri = F2FS_INODE(page);
2544 if (ri->i_inline & F2FS_INLINE_XATTR) {
2545 set_inode_flag(inode, FI_INLINE_XATTR);
2547 clear_inode_flag(inode, FI_INLINE_XATTR);
2551 dst_addr = inline_xattr_addr(inode, ipage);
2552 src_addr = inline_xattr_addr(inode, page);
2553 inline_size = inline_xattr_size(inode);
2555 f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2556 memcpy(dst_addr, src_addr, inline_size);
2558 f2fs_update_inode(inode, ipage);
2559 f2fs_put_page(ipage, 1);
2563 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2565 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2566 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2568 struct dnode_of_data dn;
2569 struct node_info ni;
2576 /* 1: invalidate the previous xattr nid */
2577 err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2581 f2fs_invalidate_blocks(sbi, ni.blk_addr);
2582 dec_valid_node_count(sbi, inode, false);
2583 set_node_addr(sbi, &ni, NULL_ADDR, false);
2586 /* 2: update xattr nid in inode */
2587 if (!f2fs_alloc_nid(sbi, &new_xnid))
2590 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2591 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2592 if (IS_ERR(xpage)) {
2593 f2fs_alloc_nid_failed(sbi, new_xnid);
2594 return PTR_ERR(xpage);
2597 f2fs_alloc_nid_done(sbi, new_xnid);
2598 f2fs_update_inode_page(inode);
2600 /* 3: update and set xattr node page dirty */
2601 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2603 set_page_dirty(xpage);
2604 f2fs_put_page(xpage, 1);
2609 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2611 struct f2fs_inode *src, *dst;
2612 nid_t ino = ino_of_node(page);
2613 struct node_info old_ni, new_ni;
2617 err = f2fs_get_node_info(sbi, ino, &old_ni);
2621 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2624 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2626 congestion_wait(BLK_RW_ASYNC, HZ/50);
2630 /* Should not use this inode from free nid list */
2631 remove_free_nid(sbi, ino);
2633 if (!PageUptodate(ipage))
2634 SetPageUptodate(ipage);
2635 fill_node_footer(ipage, ino, ino, 0, true);
2636 set_cold_node(ipage, false);
2638 src = F2FS_INODE(page);
2639 dst = F2FS_INODE(ipage);
2641 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2643 dst->i_blocks = cpu_to_le64(1);
2644 dst->i_links = cpu_to_le32(1);
2645 dst->i_xattr_nid = 0;
2646 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2647 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2648 dst->i_extra_isize = src->i_extra_isize;
2650 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2651 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2652 i_inline_xattr_size))
2653 dst->i_inline_xattr_size = src->i_inline_xattr_size;
2655 if (f2fs_sb_has_project_quota(sbi) &&
2656 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2658 dst->i_projid = src->i_projid;
2660 if (f2fs_sb_has_inode_crtime(sbi) &&
2661 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2663 dst->i_crtime = src->i_crtime;
2664 dst->i_crtime_nsec = src->i_crtime_nsec;
2671 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2673 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2674 inc_valid_inode_count(sbi);
2675 set_page_dirty(ipage);
2676 f2fs_put_page(ipage, 1);
2680 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2681 unsigned int segno, struct f2fs_summary_block *sum)
2683 struct f2fs_node *rn;
2684 struct f2fs_summary *sum_entry;
2686 int i, idx, last_offset, nrpages;
2688 /* scan the node segment */
2689 last_offset = sbi->blocks_per_seg;
2690 addr = START_BLOCK(sbi, segno);
2691 sum_entry = &sum->entries[0];
2693 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2694 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2696 /* readahead node pages */
2697 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2699 for (idx = addr; idx < addr + nrpages; idx++) {
2700 struct page *page = f2fs_get_tmp_page(sbi, idx);
2703 return PTR_ERR(page);
2705 rn = F2FS_NODE(page);
2706 sum_entry->nid = rn->footer.nid;
2707 sum_entry->version = 0;
2708 sum_entry->ofs_in_node = 0;
2710 f2fs_put_page(page, 1);
2713 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2719 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2721 struct f2fs_nm_info *nm_i = NM_I(sbi);
2722 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2723 struct f2fs_journal *journal = curseg->journal;
2726 down_write(&curseg->journal_rwsem);
2727 for (i = 0; i < nats_in_cursum(journal); i++) {
2728 struct nat_entry *ne;
2729 struct f2fs_nat_entry raw_ne;
2730 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2732 if (f2fs_check_nid_range(sbi, nid))
2735 raw_ne = nat_in_journal(journal, i);
2737 ne = __lookup_nat_cache(nm_i, nid);
2739 ne = __alloc_nat_entry(nid, true);
2740 __init_nat_entry(nm_i, ne, &raw_ne, true);
2744 * if a free nat in journal has not been used after last
2745 * checkpoint, we should remove it from available nids,
2746 * since later we will add it again.
2748 if (!get_nat_flag(ne, IS_DIRTY) &&
2749 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2750 spin_lock(&nm_i->nid_list_lock);
2751 nm_i->available_nids--;
2752 spin_unlock(&nm_i->nid_list_lock);
2755 __set_nat_cache_dirty(nm_i, ne);
2757 update_nats_in_cursum(journal, -i);
2758 up_write(&curseg->journal_rwsem);
2761 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2762 struct list_head *head, int max)
2764 struct nat_entry_set *cur;
2766 if (nes->entry_cnt >= max)
2769 list_for_each_entry(cur, head, set_list) {
2770 if (cur->entry_cnt >= nes->entry_cnt) {
2771 list_add(&nes->set_list, cur->set_list.prev);
2776 list_add_tail(&nes->set_list, head);
2779 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2782 struct f2fs_nm_info *nm_i = NM_I(sbi);
2783 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2784 struct f2fs_nat_block *nat_blk = page_address(page);
2788 if (!enabled_nat_bits(sbi, NULL))
2791 if (nat_index == 0) {
2795 for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2796 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2800 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2801 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2805 __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2806 if (valid == NAT_ENTRY_PER_BLOCK)
2807 __set_bit_le(nat_index, nm_i->full_nat_bits);
2809 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2812 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2813 struct nat_entry_set *set, struct cp_control *cpc)
2815 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2816 struct f2fs_journal *journal = curseg->journal;
2817 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2818 bool to_journal = true;
2819 struct f2fs_nat_block *nat_blk;
2820 struct nat_entry *ne, *cur;
2821 struct page *page = NULL;
2824 * there are two steps to flush nat entries:
2825 * #1, flush nat entries to journal in current hot data summary block.
2826 * #2, flush nat entries to nat page.
2828 if (enabled_nat_bits(sbi, cpc) ||
2829 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2833 down_write(&curseg->journal_rwsem);
2835 page = get_next_nat_page(sbi, start_nid);
2837 return PTR_ERR(page);
2839 nat_blk = page_address(page);
2840 f2fs_bug_on(sbi, !nat_blk);
2843 /* flush dirty nats in nat entry set */
2844 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2845 struct f2fs_nat_entry *raw_ne;
2846 nid_t nid = nat_get_nid(ne);
2849 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2852 offset = f2fs_lookup_journal_in_cursum(journal,
2853 NAT_JOURNAL, nid, 1);
2854 f2fs_bug_on(sbi, offset < 0);
2855 raw_ne = &nat_in_journal(journal, offset);
2856 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2858 raw_ne = &nat_blk->entries[nid - start_nid];
2860 raw_nat_from_node_info(raw_ne, &ne->ni);
2862 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2863 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2864 add_free_nid(sbi, nid, false, true);
2866 spin_lock(&NM_I(sbi)->nid_list_lock);
2867 update_free_nid_bitmap(sbi, nid, false, false);
2868 spin_unlock(&NM_I(sbi)->nid_list_lock);
2873 up_write(&curseg->journal_rwsem);
2875 __update_nat_bits(sbi, start_nid, page);
2876 f2fs_put_page(page, 1);
2879 /* Allow dirty nats by node block allocation in write_begin */
2880 if (!set->entry_cnt) {
2881 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2882 kmem_cache_free(nat_entry_set_slab, set);
2888 * This function is called during the checkpointing process.
2890 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2892 struct f2fs_nm_info *nm_i = NM_I(sbi);
2893 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2894 struct f2fs_journal *journal = curseg->journal;
2895 struct nat_entry_set *setvec[SETVEC_SIZE];
2896 struct nat_entry_set *set, *tmp;
2903 * during unmount, let's flush nat_bits before checking
2904 * nat_cnt[DIRTY_NAT].
2906 if (enabled_nat_bits(sbi, cpc)) {
2907 down_write(&nm_i->nat_tree_lock);
2908 remove_nats_in_journal(sbi);
2909 up_write(&nm_i->nat_tree_lock);
2912 if (!nm_i->nat_cnt[DIRTY_NAT])
2915 down_write(&nm_i->nat_tree_lock);
2918 * if there are no enough space in journal to store dirty nat
2919 * entries, remove all entries from journal and merge them
2920 * into nat entry set.
2922 if (enabled_nat_bits(sbi, cpc) ||
2923 !__has_cursum_space(journal,
2924 nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
2925 remove_nats_in_journal(sbi);
2927 while ((found = __gang_lookup_nat_set(nm_i,
2928 set_idx, SETVEC_SIZE, setvec))) {
2930 set_idx = setvec[found - 1]->set + 1;
2931 for (idx = 0; idx < found; idx++)
2932 __adjust_nat_entry_set(setvec[idx], &sets,
2933 MAX_NAT_JENTRIES(journal));
2936 /* flush dirty nats in nat entry set */
2937 list_for_each_entry_safe(set, tmp, &sets, set_list) {
2938 err = __flush_nat_entry_set(sbi, set, cpc);
2943 up_write(&nm_i->nat_tree_lock);
2944 /* Allow dirty nats by node block allocation in write_begin */
2949 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2951 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2952 struct f2fs_nm_info *nm_i = NM_I(sbi);
2953 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2955 __u64 cp_ver = cur_cp_version(ckpt);
2956 block_t nat_bits_addr;
2958 if (!enabled_nat_bits(sbi, NULL))
2961 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
2962 nm_i->nat_bits = f2fs_kvzalloc(sbi,
2963 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
2964 if (!nm_i->nat_bits)
2967 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2968 nm_i->nat_bits_blocks;
2969 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2972 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
2974 return PTR_ERR(page);
2976 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2977 page_address(page), F2FS_BLKSIZE);
2978 f2fs_put_page(page, 1);
2981 cp_ver |= (cur_cp_crc(ckpt) << 32);
2982 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2983 disable_nat_bits(sbi, true);
2987 nm_i->full_nat_bits = nm_i->nat_bits + 8;
2988 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2990 f2fs_notice(sbi, "Found nat_bits in checkpoint");
2994 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2996 struct f2fs_nm_info *nm_i = NM_I(sbi);
2998 nid_t nid, last_nid;
3000 if (!enabled_nat_bits(sbi, NULL))
3003 for (i = 0; i < nm_i->nat_blocks; i++) {
3004 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3005 if (i >= nm_i->nat_blocks)
3008 __set_bit_le(i, nm_i->nat_block_bitmap);
3010 nid = i * NAT_ENTRY_PER_BLOCK;
3011 last_nid = nid + NAT_ENTRY_PER_BLOCK;
3013 spin_lock(&NM_I(sbi)->nid_list_lock);
3014 for (; nid < last_nid; nid++)
3015 update_free_nid_bitmap(sbi, nid, true, true);
3016 spin_unlock(&NM_I(sbi)->nid_list_lock);
3019 for (i = 0; i < nm_i->nat_blocks; i++) {
3020 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3021 if (i >= nm_i->nat_blocks)
3024 __set_bit_le(i, nm_i->nat_block_bitmap);
3028 static int init_node_manager(struct f2fs_sb_info *sbi)
3030 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3031 struct f2fs_nm_info *nm_i = NM_I(sbi);
3032 unsigned char *version_bitmap;
3033 unsigned int nat_segs;
3036 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3038 /* segment_count_nat includes pair segment so divide to 2. */
3039 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3040 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3041 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3043 /* not used nids: 0, node, meta, (and root counted as valid node) */
3044 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3045 F2FS_RESERVED_NODE_NUM;
3046 nm_i->nid_cnt[FREE_NID] = 0;
3047 nm_i->nid_cnt[PREALLOC_NID] = 0;
3048 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3049 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3050 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3052 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3053 INIT_LIST_HEAD(&nm_i->free_nid_list);
3054 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3055 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3056 INIT_LIST_HEAD(&nm_i->nat_entries);
3057 spin_lock_init(&nm_i->nat_list_lock);
3059 mutex_init(&nm_i->build_lock);
3060 spin_lock_init(&nm_i->nid_list_lock);
3061 init_rwsem(&nm_i->nat_tree_lock);
3063 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3064 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3065 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3066 if (!version_bitmap)
3069 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3071 if (!nm_i->nat_bitmap)
3074 err = __get_nat_bitmaps(sbi);
3078 #ifdef CONFIG_F2FS_CHECK_FS
3079 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3081 if (!nm_i->nat_bitmap_mir)
3088 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3090 struct f2fs_nm_info *nm_i = NM_I(sbi);
3093 nm_i->free_nid_bitmap =
3094 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3097 if (!nm_i->free_nid_bitmap)
3100 for (i = 0; i < nm_i->nat_blocks; i++) {
3101 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3102 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3103 if (!nm_i->free_nid_bitmap[i])
3107 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3109 if (!nm_i->nat_block_bitmap)
3112 nm_i->free_nid_count =
3113 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3116 if (!nm_i->free_nid_count)
3121 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3125 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3130 err = init_node_manager(sbi);
3134 err = init_free_nid_cache(sbi);
3138 /* load free nid status from nat_bits table */
3139 load_free_nid_bitmap(sbi);
3141 return f2fs_build_free_nids(sbi, true, true);
3144 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3146 struct f2fs_nm_info *nm_i = NM_I(sbi);
3147 struct free_nid *i, *next_i;
3148 struct nat_entry *natvec[NATVEC_SIZE];
3149 struct nat_entry_set *setvec[SETVEC_SIZE];
3156 /* destroy free nid list */
3157 spin_lock(&nm_i->nid_list_lock);
3158 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3159 __remove_free_nid(sbi, i, FREE_NID);
3160 spin_unlock(&nm_i->nid_list_lock);
3161 kmem_cache_free(free_nid_slab, i);
3162 spin_lock(&nm_i->nid_list_lock);
3164 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3165 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3166 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3167 spin_unlock(&nm_i->nid_list_lock);
3169 /* destroy nat cache */
3170 down_write(&nm_i->nat_tree_lock);
3171 while ((found = __gang_lookup_nat_cache(nm_i,
3172 nid, NATVEC_SIZE, natvec))) {
3175 nid = nat_get_nid(natvec[found - 1]) + 1;
3176 for (idx = 0; idx < found; idx++) {
3177 spin_lock(&nm_i->nat_list_lock);
3178 list_del(&natvec[idx]->list);
3179 spin_unlock(&nm_i->nat_list_lock);
3181 __del_from_nat_cache(nm_i, natvec[idx]);
3184 f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3186 /* destroy nat set cache */
3188 while ((found = __gang_lookup_nat_set(nm_i,
3189 nid, SETVEC_SIZE, setvec))) {
3192 nid = setvec[found - 1]->set + 1;
3193 for (idx = 0; idx < found; idx++) {
3194 /* entry_cnt is not zero, when cp_error was occurred */
3195 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3196 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3197 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3200 up_write(&nm_i->nat_tree_lock);
3202 kvfree(nm_i->nat_block_bitmap);
3203 if (nm_i->free_nid_bitmap) {
3206 for (i = 0; i < nm_i->nat_blocks; i++)
3207 kvfree(nm_i->free_nid_bitmap[i]);
3208 kvfree(nm_i->free_nid_bitmap);
3210 kvfree(nm_i->free_nid_count);
3212 kvfree(nm_i->nat_bitmap);
3213 kvfree(nm_i->nat_bits);
3214 #ifdef CONFIG_F2FS_CHECK_FS
3215 kvfree(nm_i->nat_bitmap_mir);
3217 sbi->nm_info = NULL;
3221 int __init f2fs_create_node_manager_caches(void)
3223 nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
3224 sizeof(struct nat_entry));
3225 if (!nat_entry_slab)
3228 free_nid_slab = f2fs_kmem_cache_create("free_nid",
3229 sizeof(struct free_nid));
3231 goto destroy_nat_entry;
3233 nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
3234 sizeof(struct nat_entry_set));
3235 if (!nat_entry_set_slab)
3236 goto destroy_free_nid;
3238 fsync_node_entry_slab = f2fs_kmem_cache_create("fsync_node_entry",
3239 sizeof(struct fsync_node_entry));
3240 if (!fsync_node_entry_slab)
3241 goto destroy_nat_entry_set;
3244 destroy_nat_entry_set:
3245 kmem_cache_destroy(nat_entry_set_slab);
3247 kmem_cache_destroy(free_nid_slab);
3249 kmem_cache_destroy(nat_entry_slab);
3254 void f2fs_destroy_node_manager_caches(void)
3256 kmem_cache_destroy(fsync_node_entry_slab);
3257 kmem_cache_destroy(nat_entry_set_slab);
3258 kmem_cache_destroy(free_nid_slab);
3259 kmem_cache_destroy(nat_entry_slab);