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>
20 #include <trace/events/f2fs.h>
22 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
24 static struct kmem_cache *nat_entry_slab;
25 static struct kmem_cache *free_nid_slab;
26 static struct kmem_cache *nat_entry_set_slab;
27 static struct kmem_cache *fsync_node_entry_slab;
30 * Check whether the given nid is within node id range.
32 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
34 if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
35 set_sbi_flag(sbi, SBI_NEED_FSCK);
36 f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
43 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
45 struct f2fs_nm_info *nm_i = NM_I(sbi);
47 unsigned long avail_ram;
48 unsigned long mem_size = 0;
53 /* only uses low memory */
54 avail_ram = val.totalram - val.totalhigh;
57 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
59 if (type == FREE_NIDS) {
60 mem_size = (nm_i->nid_cnt[FREE_NID] *
61 sizeof(struct free_nid)) >> PAGE_SHIFT;
62 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
63 } else if (type == NAT_ENTRIES) {
64 mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
65 sizeof(struct nat_entry)) >> PAGE_SHIFT;
66 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
67 if (excess_cached_nats(sbi))
69 } else if (type == DIRTY_DENTS) {
70 if (sbi->sb->s_bdi->wb.dirty_exceeded)
72 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
73 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
74 } else if (type == INO_ENTRIES) {
77 for (i = 0; i < MAX_INO_ENTRY; i++)
78 mem_size += sbi->im[i].ino_num *
79 sizeof(struct ino_entry);
80 mem_size >>= PAGE_SHIFT;
81 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
82 } else if (type == EXTENT_CACHE) {
83 mem_size = (atomic_read(&sbi->total_ext_tree) *
84 sizeof(struct extent_tree) +
85 atomic_read(&sbi->total_ext_node) *
86 sizeof(struct extent_node)) >> PAGE_SHIFT;
87 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
88 } else if (type == INMEM_PAGES) {
89 /* it allows 20% / total_ram for inmemory pages */
90 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
91 res = mem_size < (val.totalram / 5);
93 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
99 static void clear_node_page_dirty(struct page *page)
101 if (PageDirty(page)) {
102 f2fs_clear_page_cache_dirty_tag(page);
103 clear_page_dirty_for_io(page);
104 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
106 ClearPageUptodate(page);
109 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
111 return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
114 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
116 struct page *src_page;
117 struct page *dst_page;
121 struct f2fs_nm_info *nm_i = NM_I(sbi);
123 dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
125 /* get current nat block page with lock */
126 src_page = get_current_nat_page(sbi, nid);
127 if (IS_ERR(src_page))
129 dst_page = f2fs_grab_meta_page(sbi, dst_off);
130 f2fs_bug_on(sbi, PageDirty(src_page));
132 src_addr = page_address(src_page);
133 dst_addr = page_address(dst_page);
134 memcpy(dst_addr, src_addr, PAGE_SIZE);
135 set_page_dirty(dst_page);
136 f2fs_put_page(src_page, 1);
138 set_to_next_nat(nm_i, nid);
143 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
145 struct nat_entry *new;
148 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
150 new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
152 nat_set_nid(new, nid);
158 static void __free_nat_entry(struct nat_entry *e)
160 kmem_cache_free(nat_entry_slab, e);
163 /* must be locked by nat_tree_lock */
164 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
165 struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
168 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
169 else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
173 node_info_from_raw_nat(&ne->ni, raw_ne);
175 spin_lock(&nm_i->nat_list_lock);
176 list_add_tail(&ne->list, &nm_i->nat_entries);
177 spin_unlock(&nm_i->nat_list_lock);
179 nm_i->nat_cnt[TOTAL_NAT]++;
180 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
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));
210 nm_i->nat_cnt[TOTAL_NAT]--;
211 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
215 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
216 struct nat_entry *ne)
218 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
219 struct nat_entry_set *head;
221 head = radix_tree_lookup(&nm_i->nat_set_root, set);
223 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
225 INIT_LIST_HEAD(&head->entry_list);
226 INIT_LIST_HEAD(&head->set_list);
229 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
234 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
235 struct nat_entry *ne)
237 struct nat_entry_set *head;
238 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
241 head = __grab_nat_entry_set(nm_i, ne);
244 * update entry_cnt in below condition:
245 * 1. update NEW_ADDR to valid block address;
246 * 2. update old block address to new one;
248 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
249 !get_nat_flag(ne, IS_DIRTY)))
252 set_nat_flag(ne, IS_PREALLOC, new_ne);
254 if (get_nat_flag(ne, IS_DIRTY))
257 nm_i->nat_cnt[DIRTY_NAT]++;
258 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
259 set_nat_flag(ne, IS_DIRTY, true);
261 spin_lock(&nm_i->nat_list_lock);
263 list_del_init(&ne->list);
265 list_move_tail(&ne->list, &head->entry_list);
266 spin_unlock(&nm_i->nat_list_lock);
269 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
270 struct nat_entry_set *set, struct nat_entry *ne)
272 spin_lock(&nm_i->nat_list_lock);
273 list_move_tail(&ne->list, &nm_i->nat_entries);
274 spin_unlock(&nm_i->nat_list_lock);
276 set_nat_flag(ne, IS_DIRTY, false);
278 nm_i->nat_cnt[DIRTY_NAT]--;
279 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
282 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
283 nid_t start, unsigned int nr, struct nat_entry_set **ep)
285 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
289 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
291 return NODE_MAPPING(sbi) == page->mapping &&
292 IS_DNODE(page) && is_cold_node(page);
295 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
297 spin_lock_init(&sbi->fsync_node_lock);
298 INIT_LIST_HEAD(&sbi->fsync_node_list);
299 sbi->fsync_seg_id = 0;
300 sbi->fsync_node_num = 0;
303 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
306 struct fsync_node_entry *fn;
310 fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS);
314 INIT_LIST_HEAD(&fn->list);
316 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
317 list_add_tail(&fn->list, &sbi->fsync_node_list);
318 fn->seq_id = sbi->fsync_seg_id++;
320 sbi->fsync_node_num++;
321 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
326 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
328 struct fsync_node_entry *fn;
331 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
332 list_for_each_entry(fn, &sbi->fsync_node_list, list) {
333 if (fn->page == page) {
335 sbi->fsync_node_num--;
336 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
337 kmem_cache_free(fsync_node_entry_slab, fn);
342 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
346 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
350 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
351 sbi->fsync_seg_id = 0;
352 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
355 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
357 struct f2fs_nm_info *nm_i = NM_I(sbi);
361 down_read(&nm_i->nat_tree_lock);
362 e = __lookup_nat_cache(nm_i, nid);
364 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
365 !get_nat_flag(e, HAS_FSYNCED_INODE))
368 up_read(&nm_i->nat_tree_lock);
372 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
374 struct f2fs_nm_info *nm_i = NM_I(sbi);
378 down_read(&nm_i->nat_tree_lock);
379 e = __lookup_nat_cache(nm_i, nid);
380 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
382 up_read(&nm_i->nat_tree_lock);
386 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
388 struct f2fs_nm_info *nm_i = NM_I(sbi);
390 bool need_update = true;
392 down_read(&nm_i->nat_tree_lock);
393 e = __lookup_nat_cache(nm_i, ino);
394 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
395 (get_nat_flag(e, IS_CHECKPOINTED) ||
396 get_nat_flag(e, HAS_FSYNCED_INODE)))
398 up_read(&nm_i->nat_tree_lock);
402 /* must be locked by nat_tree_lock */
403 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
404 struct f2fs_nat_entry *ne)
406 struct f2fs_nm_info *nm_i = NM_I(sbi);
407 struct nat_entry *new, *e;
409 new = __alloc_nat_entry(nid, false);
413 down_write(&nm_i->nat_tree_lock);
414 e = __lookup_nat_cache(nm_i, nid);
416 e = __init_nat_entry(nm_i, new, ne, false);
418 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
419 nat_get_blkaddr(e) !=
420 le32_to_cpu(ne->block_addr) ||
421 nat_get_version(e) != ne->version);
422 up_write(&nm_i->nat_tree_lock);
424 __free_nat_entry(new);
427 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
428 block_t new_blkaddr, bool fsync_done)
430 struct f2fs_nm_info *nm_i = NM_I(sbi);
432 struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
434 down_write(&nm_i->nat_tree_lock);
435 e = __lookup_nat_cache(nm_i, ni->nid);
437 e = __init_nat_entry(nm_i, new, NULL, true);
438 copy_node_info(&e->ni, ni);
439 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
440 } else if (new_blkaddr == NEW_ADDR) {
442 * when nid is reallocated,
443 * previous nat entry can be remained in nat cache.
444 * So, reinitialize it with new information.
446 copy_node_info(&e->ni, ni);
447 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
449 /* let's free early to reduce memory consumption */
451 __free_nat_entry(new);
454 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
455 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
456 new_blkaddr == NULL_ADDR);
457 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
458 new_blkaddr == NEW_ADDR);
459 f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
460 new_blkaddr == NEW_ADDR);
462 /* increment version no as node is removed */
463 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
464 unsigned char version = nat_get_version(e);
465 nat_set_version(e, inc_node_version(version));
469 nat_set_blkaddr(e, new_blkaddr);
470 if (!__is_valid_data_blkaddr(new_blkaddr))
471 set_nat_flag(e, IS_CHECKPOINTED, false);
472 __set_nat_cache_dirty(nm_i, e);
474 /* update fsync_mark if its inode nat entry is still alive */
475 if (ni->nid != ni->ino)
476 e = __lookup_nat_cache(nm_i, ni->ino);
478 if (fsync_done && ni->nid == ni->ino)
479 set_nat_flag(e, HAS_FSYNCED_INODE, true);
480 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
482 up_write(&nm_i->nat_tree_lock);
485 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
487 struct f2fs_nm_info *nm_i = NM_I(sbi);
490 if (!down_write_trylock(&nm_i->nat_tree_lock))
493 spin_lock(&nm_i->nat_list_lock);
495 struct nat_entry *ne;
497 if (list_empty(&nm_i->nat_entries))
500 ne = list_first_entry(&nm_i->nat_entries,
501 struct nat_entry, list);
503 spin_unlock(&nm_i->nat_list_lock);
505 __del_from_nat_cache(nm_i, ne);
508 spin_lock(&nm_i->nat_list_lock);
510 spin_unlock(&nm_i->nat_list_lock);
512 up_write(&nm_i->nat_tree_lock);
513 return nr - nr_shrink;
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 mode is set with ALLOC_NODE.
721 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
723 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
724 struct page *npage[4];
725 struct page *parent = NULL;
727 unsigned int noffset[4];
732 level = get_node_path(dn->inode, index, offset, noffset);
736 nids[0] = dn->inode->i_ino;
737 npage[0] = dn->inode_page;
740 npage[0] = f2fs_get_node_page(sbi, nids[0]);
741 if (IS_ERR(npage[0]))
742 return PTR_ERR(npage[0]);
745 /* if inline_data is set, should not report any block indices */
746 if (f2fs_has_inline_data(dn->inode) && index) {
748 f2fs_put_page(npage[0], 1);
754 nids[1] = get_nid(parent, offset[0], true);
755 dn->inode_page = npage[0];
756 dn->inode_page_locked = true;
758 /* get indirect or direct nodes */
759 for (i = 1; i <= level; i++) {
762 if (!nids[i] && mode == ALLOC_NODE) {
764 if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
770 npage[i] = f2fs_new_node_page(dn, noffset[i]);
771 if (IS_ERR(npage[i])) {
772 f2fs_alloc_nid_failed(sbi, nids[i]);
773 err = PTR_ERR(npage[i]);
777 set_nid(parent, offset[i - 1], nids[i], i == 1);
778 f2fs_alloc_nid_done(sbi, nids[i]);
780 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
781 npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
782 if (IS_ERR(npage[i])) {
783 err = PTR_ERR(npage[i]);
789 dn->inode_page_locked = false;
792 f2fs_put_page(parent, 1);
796 npage[i] = f2fs_get_node_page(sbi, nids[i]);
797 if (IS_ERR(npage[i])) {
798 err = PTR_ERR(npage[i]);
799 f2fs_put_page(npage[0], 0);
805 nids[i + 1] = get_nid(parent, offset[i], false);
808 dn->nid = nids[level];
809 dn->ofs_in_node = offset[level];
810 dn->node_page = npage[level];
811 dn->data_blkaddr = f2fs_data_blkaddr(dn);
815 f2fs_put_page(parent, 1);
817 f2fs_put_page(npage[0], 0);
819 dn->inode_page = NULL;
820 dn->node_page = NULL;
821 if (err == -ENOENT) {
823 dn->max_level = level;
824 dn->ofs_in_node = offset[level];
829 static int truncate_node(struct dnode_of_data *dn)
831 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
836 err = f2fs_get_node_info(sbi, dn->nid, &ni);
840 /* Deallocate node address */
841 f2fs_invalidate_blocks(sbi, ni.blk_addr);
842 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
843 set_node_addr(sbi, &ni, NULL_ADDR, false);
845 if (dn->nid == dn->inode->i_ino) {
846 f2fs_remove_orphan_inode(sbi, dn->nid);
847 dec_valid_inode_count(sbi);
848 f2fs_inode_synced(dn->inode);
851 clear_node_page_dirty(dn->node_page);
852 set_sbi_flag(sbi, SBI_IS_DIRTY);
854 index = dn->node_page->index;
855 f2fs_put_page(dn->node_page, 1);
857 invalidate_mapping_pages(NODE_MAPPING(sbi),
860 dn->node_page = NULL;
861 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
866 static int truncate_dnode(struct dnode_of_data *dn)
874 /* get direct node */
875 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
876 if (PTR_ERR(page) == -ENOENT)
878 else if (IS_ERR(page))
879 return PTR_ERR(page);
881 /* Make dnode_of_data for parameter */
882 dn->node_page = page;
884 f2fs_truncate_data_blocks(dn);
885 err = truncate_node(dn);
892 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
895 struct dnode_of_data rdn = *dn;
897 struct f2fs_node *rn;
899 unsigned int child_nofs;
904 return NIDS_PER_BLOCK + 1;
906 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
908 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
910 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
911 return PTR_ERR(page);
914 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
916 rn = F2FS_NODE(page);
918 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
919 child_nid = le32_to_cpu(rn->in.nid[i]);
923 ret = truncate_dnode(&rdn);
926 if (set_nid(page, i, 0, false))
927 dn->node_changed = true;
930 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
931 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
932 child_nid = le32_to_cpu(rn->in.nid[i]);
933 if (child_nid == 0) {
934 child_nofs += NIDS_PER_BLOCK + 1;
938 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
939 if (ret == (NIDS_PER_BLOCK + 1)) {
940 if (set_nid(page, i, 0, false))
941 dn->node_changed = true;
943 } else if (ret < 0 && ret != -ENOENT) {
951 /* remove current indirect node */
952 dn->node_page = page;
953 ret = truncate_node(dn);
958 f2fs_put_page(page, 1);
960 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
964 f2fs_put_page(page, 1);
965 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
969 static int truncate_partial_nodes(struct dnode_of_data *dn,
970 struct f2fs_inode *ri, int *offset, int depth)
972 struct page *pages[2];
979 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
983 /* get indirect nodes in the path */
984 for (i = 0; i < idx + 1; i++) {
985 /* reference count'll be increased */
986 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
987 if (IS_ERR(pages[i])) {
988 err = PTR_ERR(pages[i]);
992 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
995 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
997 /* free direct nodes linked to a partial indirect node */
998 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
999 child_nid = get_nid(pages[idx], i, false);
1002 dn->nid = child_nid;
1003 err = truncate_dnode(dn);
1006 if (set_nid(pages[idx], i, 0, false))
1007 dn->node_changed = true;
1010 if (offset[idx + 1] == 0) {
1011 dn->node_page = pages[idx];
1013 err = truncate_node(dn);
1017 f2fs_put_page(pages[idx], 1);
1020 offset[idx + 1] = 0;
1023 for (i = idx; i >= 0; i--)
1024 f2fs_put_page(pages[i], 1);
1026 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1032 * All the block addresses of data and nodes should be nullified.
1034 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1036 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1037 int err = 0, cont = 1;
1038 int level, offset[4], noffset[4];
1039 unsigned int nofs = 0;
1040 struct f2fs_inode *ri;
1041 struct dnode_of_data dn;
1044 trace_f2fs_truncate_inode_blocks_enter(inode, from);
1046 level = get_node_path(inode, from, offset, noffset);
1048 trace_f2fs_truncate_inode_blocks_exit(inode, level);
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),
1192 "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1193 inode->i_ino, (unsigned long long)inode->i_blocks);
1194 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1197 /* will put inode & node pages */
1198 err = truncate_node(&dn);
1200 f2fs_put_dnode(&dn);
1206 struct page *f2fs_new_inode_page(struct inode *inode)
1208 struct dnode_of_data dn;
1210 /* allocate inode page for new inode */
1211 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1213 /* caller should f2fs_put_page(page, 1); */
1214 return f2fs_new_node_page(&dn, 0);
1217 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1219 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1220 struct node_info new_ni;
1224 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1225 return ERR_PTR(-EPERM);
1227 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1229 return ERR_PTR(-ENOMEM);
1231 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1234 #ifdef CONFIG_F2FS_CHECK_FS
1235 err = f2fs_get_node_info(sbi, dn->nid, &new_ni);
1237 dec_valid_node_count(sbi, dn->inode, !ofs);
1240 f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1242 new_ni.nid = dn->nid;
1243 new_ni.ino = dn->inode->i_ino;
1244 new_ni.blk_addr = NULL_ADDR;
1247 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1249 f2fs_wait_on_page_writeback(page, NODE, true, true);
1250 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1251 set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1252 if (!PageUptodate(page))
1253 SetPageUptodate(page);
1254 if (set_page_dirty(page))
1255 dn->node_changed = true;
1257 if (f2fs_has_xattr_block(ofs))
1258 f2fs_i_xnid_write(dn->inode, dn->nid);
1261 inc_valid_inode_count(sbi);
1265 clear_node_page_dirty(page);
1266 f2fs_put_page(page, 1);
1267 return ERR_PTR(err);
1271 * Caller should do after getting the following values.
1272 * 0: f2fs_put_page(page, 0)
1273 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1275 static int read_node_page(struct page *page, int op_flags)
1277 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1278 struct node_info ni;
1279 struct f2fs_io_info fio = {
1283 .op_flags = op_flags,
1285 .encrypted_page = NULL,
1289 if (PageUptodate(page)) {
1290 if (!f2fs_inode_chksum_verify(sbi, page)) {
1291 ClearPageUptodate(page);
1297 err = f2fs_get_node_info(sbi, page->index, &ni);
1301 if (unlikely(ni.blk_addr == NULL_ADDR) ||
1302 is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1303 ClearPageUptodate(page);
1307 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1309 err = f2fs_submit_page_bio(&fio);
1312 f2fs_update_iostat(sbi, FS_NODE_READ_IO, F2FS_BLKSIZE);
1318 * Readahead a node page
1320 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1327 if (f2fs_check_nid_range(sbi, nid))
1330 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1334 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1338 err = read_node_page(apage, REQ_RAHEAD);
1339 f2fs_put_page(apage, err ? 1 : 0);
1342 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1343 struct page *parent, int start)
1349 return ERR_PTR(-ENOENT);
1350 if (f2fs_check_nid_range(sbi, nid))
1351 return ERR_PTR(-EINVAL);
1353 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1355 return ERR_PTR(-ENOMEM);
1357 err = read_node_page(page, 0);
1359 f2fs_put_page(page, 1);
1360 return ERR_PTR(err);
1361 } else if (err == LOCKED_PAGE) {
1367 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1371 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1372 f2fs_put_page(page, 1);
1376 if (unlikely(!PageUptodate(page))) {
1381 if (!f2fs_inode_chksum_verify(sbi, page)) {
1386 if(unlikely(nid != nid_of_node(page))) {
1387 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1388 nid, nid_of_node(page), ino_of_node(page),
1389 ofs_of_node(page), cpver_of_node(page),
1390 next_blkaddr_of_node(page));
1393 ClearPageUptodate(page);
1394 f2fs_put_page(page, 1);
1395 return ERR_PTR(err);
1400 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1402 return __get_node_page(sbi, nid, NULL, 0);
1405 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1407 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1408 nid_t nid = get_nid(parent, start, false);
1410 return __get_node_page(sbi, nid, parent, start);
1413 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1415 struct inode *inode;
1419 /* should flush inline_data before evict_inode */
1420 inode = ilookup(sbi->sb, ino);
1424 page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1425 FGP_LOCK|FGP_NOWAIT, 0);
1429 if (!PageUptodate(page))
1432 if (!PageDirty(page))
1435 if (!clear_page_dirty_for_io(page))
1438 ret = f2fs_write_inline_data(inode, page);
1439 inode_dec_dirty_pages(inode);
1440 f2fs_remove_dirty_inode(inode);
1442 set_page_dirty(page);
1444 f2fs_put_page(page, 1);
1449 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1452 struct pagevec pvec;
1453 struct page *last_page = NULL;
1456 pagevec_init(&pvec);
1459 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1460 PAGECACHE_TAG_DIRTY))) {
1463 for (i = 0; i < nr_pages; i++) {
1464 struct page *page = pvec.pages[i];
1466 if (unlikely(f2fs_cp_error(sbi))) {
1467 f2fs_put_page(last_page, 0);
1468 pagevec_release(&pvec);
1469 return ERR_PTR(-EIO);
1472 if (!IS_DNODE(page) || !is_cold_node(page))
1474 if (ino_of_node(page) != ino)
1479 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1484 if (ino_of_node(page) != ino)
1485 goto continue_unlock;
1487 if (!PageDirty(page)) {
1488 /* someone wrote it for us */
1489 goto continue_unlock;
1493 f2fs_put_page(last_page, 0);
1499 pagevec_release(&pvec);
1505 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1506 struct writeback_control *wbc, bool do_balance,
1507 enum iostat_type io_type, unsigned int *seq_id)
1509 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1511 struct node_info ni;
1512 struct f2fs_io_info fio = {
1514 .ino = ino_of_node(page),
1517 .op_flags = wbc_to_write_flags(wbc),
1519 .encrypted_page = NULL,
1526 trace_f2fs_writepage(page, NODE);
1528 if (unlikely(f2fs_cp_error(sbi))) {
1529 if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
1530 ClearPageUptodate(page);
1531 dec_page_count(sbi, F2FS_DIRTY_NODES);
1538 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1541 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1542 wbc->sync_mode == WB_SYNC_NONE &&
1543 IS_DNODE(page) && is_cold_node(page))
1546 /* get old block addr of this node page */
1547 nid = nid_of_node(page);
1548 f2fs_bug_on(sbi, page->index != nid);
1550 if (f2fs_get_node_info(sbi, nid, &ni))
1553 if (wbc->for_reclaim) {
1554 if (!down_read_trylock(&sbi->node_write))
1557 down_read(&sbi->node_write);
1560 /* This page is already truncated */
1561 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1562 ClearPageUptodate(page);
1563 dec_page_count(sbi, F2FS_DIRTY_NODES);
1564 up_read(&sbi->node_write);
1569 if (__is_valid_data_blkaddr(ni.blk_addr) &&
1570 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1571 DATA_GENERIC_ENHANCE)) {
1572 up_read(&sbi->node_write);
1576 if (atomic && !test_opt(sbi, NOBARRIER))
1577 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1579 /* should add to global list before clearing PAGECACHE status */
1580 if (f2fs_in_warm_node_list(sbi, page)) {
1581 seq = f2fs_add_fsync_node_entry(sbi, page);
1586 set_page_writeback(page);
1587 ClearPageError(page);
1589 fio.old_blkaddr = ni.blk_addr;
1590 f2fs_do_write_node_page(nid, &fio);
1591 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1592 dec_page_count(sbi, F2FS_DIRTY_NODES);
1593 up_read(&sbi->node_write);
1595 if (wbc->for_reclaim) {
1596 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1602 if (unlikely(f2fs_cp_error(sbi))) {
1603 f2fs_submit_merged_write(sbi, NODE);
1607 *submitted = fio.submitted;
1610 f2fs_balance_fs(sbi, false);
1614 redirty_page_for_writepage(wbc, page);
1615 return AOP_WRITEPAGE_ACTIVATE;
1618 int f2fs_move_node_page(struct page *node_page, int gc_type)
1622 if (gc_type == FG_GC) {
1623 struct writeback_control wbc = {
1624 .sync_mode = WB_SYNC_ALL,
1629 f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1631 set_page_dirty(node_page);
1633 if (!clear_page_dirty_for_io(node_page)) {
1638 if (__write_node_page(node_page, false, NULL,
1639 &wbc, false, FS_GC_NODE_IO, NULL)) {
1641 unlock_page(node_page);
1645 /* set page dirty and write it */
1646 if (!PageWriteback(node_page))
1647 set_page_dirty(node_page);
1650 unlock_page(node_page);
1652 f2fs_put_page(node_page, 0);
1656 static int f2fs_write_node_page(struct page *page,
1657 struct writeback_control *wbc)
1659 return __write_node_page(page, false, NULL, wbc, false,
1663 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1664 struct writeback_control *wbc, bool atomic,
1665 unsigned int *seq_id)
1668 struct pagevec pvec;
1670 struct page *last_page = NULL;
1671 bool marked = false;
1672 nid_t ino = inode->i_ino;
1677 last_page = last_fsync_dnode(sbi, ino);
1678 if (IS_ERR_OR_NULL(last_page))
1679 return PTR_ERR_OR_ZERO(last_page);
1682 pagevec_init(&pvec);
1685 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1686 PAGECACHE_TAG_DIRTY))) {
1689 for (i = 0; i < nr_pages; i++) {
1690 struct page *page = pvec.pages[i];
1691 bool submitted = false;
1693 if (unlikely(f2fs_cp_error(sbi))) {
1694 f2fs_put_page(last_page, 0);
1695 pagevec_release(&pvec);
1700 if (!IS_DNODE(page) || !is_cold_node(page))
1702 if (ino_of_node(page) != ino)
1707 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1712 if (ino_of_node(page) != ino)
1713 goto continue_unlock;
1715 if (!PageDirty(page) && page != last_page) {
1716 /* someone wrote it for us */
1717 goto continue_unlock;
1720 f2fs_wait_on_page_writeback(page, NODE, true, true);
1722 set_fsync_mark(page, 0);
1723 set_dentry_mark(page, 0);
1725 if (!atomic || page == last_page) {
1726 set_fsync_mark(page, 1);
1727 if (IS_INODE(page)) {
1728 if (is_inode_flag_set(inode,
1730 f2fs_update_inode(inode, page);
1731 set_dentry_mark(page,
1732 f2fs_need_dentry_mark(sbi, ino));
1734 /* may be written by other thread */
1735 if (!PageDirty(page))
1736 set_page_dirty(page);
1739 if (!clear_page_dirty_for_io(page))
1740 goto continue_unlock;
1742 ret = __write_node_page(page, atomic &&
1744 &submitted, wbc, true,
1745 FS_NODE_IO, seq_id);
1748 f2fs_put_page(last_page, 0);
1750 } else if (submitted) {
1754 if (page == last_page) {
1755 f2fs_put_page(page, 0);
1760 pagevec_release(&pvec);
1766 if (!ret && atomic && !marked) {
1767 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1768 ino, last_page->index);
1769 lock_page(last_page);
1770 f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1771 set_page_dirty(last_page);
1772 unlock_page(last_page);
1777 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1778 return ret ? -EIO: 0;
1781 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1783 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1786 if (inode->i_ino != ino)
1789 if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1792 spin_lock(&sbi->inode_lock[DIRTY_META]);
1793 clean = list_empty(&F2FS_I(inode)->gdirty_list);
1794 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1799 inode = igrab(inode);
1805 static bool flush_dirty_inode(struct page *page)
1807 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1808 struct inode *inode;
1809 nid_t ino = ino_of_node(page);
1811 inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1815 f2fs_update_inode(inode, page);
1822 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1825 struct pagevec pvec;
1828 pagevec_init(&pvec);
1830 while ((nr_pages = pagevec_lookup_tag(&pvec,
1831 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1834 for (i = 0; i < nr_pages; i++) {
1835 struct page *page = pvec.pages[i];
1837 if (!IS_DNODE(page))
1842 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1848 if (!PageDirty(page)) {
1849 /* someone wrote it for us */
1850 goto continue_unlock;
1853 /* flush inline_data, if it's async context. */
1854 if (is_inline_node(page)) {
1855 clear_inline_node(page);
1857 flush_inline_data(sbi, ino_of_node(page));
1862 pagevec_release(&pvec);
1867 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1868 struct writeback_control *wbc,
1869 bool do_balance, enum iostat_type io_type)
1872 struct pagevec pvec;
1876 int nr_pages, done = 0;
1878 pagevec_init(&pvec);
1883 while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1884 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1887 for (i = 0; i < nr_pages; i++) {
1888 struct page *page = pvec.pages[i];
1889 bool submitted = false;
1890 bool may_dirty = true;
1892 /* give a priority to WB_SYNC threads */
1893 if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1894 wbc->sync_mode == WB_SYNC_NONE) {
1900 * flushing sequence with step:
1905 if (step == 0 && IS_DNODE(page))
1907 if (step == 1 && (!IS_DNODE(page) ||
1908 is_cold_node(page)))
1910 if (step == 2 && (!IS_DNODE(page) ||
1911 !is_cold_node(page)))
1914 if (wbc->sync_mode == WB_SYNC_ALL)
1916 else if (!trylock_page(page))
1919 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1925 if (!PageDirty(page)) {
1926 /* someone wrote it for us */
1927 goto continue_unlock;
1930 /* flush inline_data/inode, if it's async context. */
1934 /* flush inline_data */
1935 if (is_inline_node(page)) {
1936 clear_inline_node(page);
1938 flush_inline_data(sbi, ino_of_node(page));
1942 /* flush dirty inode */
1943 if (IS_INODE(page) && may_dirty) {
1945 if (flush_dirty_inode(page))
1949 f2fs_wait_on_page_writeback(page, NODE, true, true);
1951 if (!clear_page_dirty_for_io(page))
1952 goto continue_unlock;
1954 set_fsync_mark(page, 0);
1955 set_dentry_mark(page, 0);
1957 ret = __write_node_page(page, false, &submitted,
1958 wbc, do_balance, io_type, NULL);
1964 if (--wbc->nr_to_write == 0)
1967 pagevec_release(&pvec);
1970 if (wbc->nr_to_write == 0) {
1977 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1978 wbc->sync_mode == WB_SYNC_NONE && step == 1)
1985 f2fs_submit_merged_write(sbi, NODE);
1987 if (unlikely(f2fs_cp_error(sbi)))
1992 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
1993 unsigned int seq_id)
1995 struct fsync_node_entry *fn;
1997 struct list_head *head = &sbi->fsync_node_list;
1998 unsigned long flags;
1999 unsigned int cur_seq_id = 0;
2002 while (seq_id && cur_seq_id < seq_id) {
2003 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2004 if (list_empty(head)) {
2005 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2008 fn = list_first_entry(head, struct fsync_node_entry, list);
2009 if (fn->seq_id > seq_id) {
2010 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2013 cur_seq_id = fn->seq_id;
2016 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2018 f2fs_wait_on_page_writeback(page, NODE, true, false);
2019 if (TestClearPageError(page))
2028 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2035 static int f2fs_write_node_pages(struct address_space *mapping,
2036 struct writeback_control *wbc)
2038 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2039 struct blk_plug plug;
2042 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2045 /* balancing f2fs's metadata in background */
2046 f2fs_balance_fs_bg(sbi, true);
2048 /* collect a number of dirty node pages and write together */
2049 if (wbc->sync_mode != WB_SYNC_ALL &&
2050 get_pages(sbi, F2FS_DIRTY_NODES) <
2051 nr_pages_to_skip(sbi, NODE))
2054 if (wbc->sync_mode == WB_SYNC_ALL)
2055 atomic_inc(&sbi->wb_sync_req[NODE]);
2056 else if (atomic_read(&sbi->wb_sync_req[NODE]))
2059 trace_f2fs_writepages(mapping->host, wbc, NODE);
2061 diff = nr_pages_to_write(sbi, NODE, wbc);
2062 blk_start_plug(&plug);
2063 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2064 blk_finish_plug(&plug);
2065 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2067 if (wbc->sync_mode == WB_SYNC_ALL)
2068 atomic_dec(&sbi->wb_sync_req[NODE]);
2072 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2073 trace_f2fs_writepages(mapping->host, wbc, NODE);
2077 static int f2fs_set_node_page_dirty(struct page *page)
2079 trace_f2fs_set_page_dirty(page, NODE);
2081 if (!PageUptodate(page))
2082 SetPageUptodate(page);
2083 #ifdef CONFIG_F2FS_CHECK_FS
2085 f2fs_inode_chksum_set(F2FS_P_SB(page), page);
2087 if (!PageDirty(page)) {
2088 __set_page_dirty_nobuffers(page);
2089 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
2090 f2fs_set_page_private(page, 0);
2097 * Structure of the f2fs node operations
2099 const struct address_space_operations f2fs_node_aops = {
2100 .writepage = f2fs_write_node_page,
2101 .writepages = f2fs_write_node_pages,
2102 .set_page_dirty = f2fs_set_node_page_dirty,
2103 .invalidatepage = f2fs_invalidate_page,
2104 .releasepage = f2fs_release_page,
2105 #ifdef CONFIG_MIGRATION
2106 .migratepage = f2fs_migrate_page,
2110 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2113 return radix_tree_lookup(&nm_i->free_nid_root, n);
2116 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2119 struct f2fs_nm_info *nm_i = NM_I(sbi);
2121 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2125 nm_i->nid_cnt[FREE_NID]++;
2126 list_add_tail(&i->list, &nm_i->free_nid_list);
2130 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2131 struct free_nid *i, enum nid_state state)
2133 struct f2fs_nm_info *nm_i = NM_I(sbi);
2135 f2fs_bug_on(sbi, state != i->state);
2136 nm_i->nid_cnt[state]--;
2137 if (state == FREE_NID)
2139 radix_tree_delete(&nm_i->free_nid_root, i->nid);
2142 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2143 enum nid_state org_state, enum nid_state dst_state)
2145 struct f2fs_nm_info *nm_i = NM_I(sbi);
2147 f2fs_bug_on(sbi, org_state != i->state);
2148 i->state = dst_state;
2149 nm_i->nid_cnt[org_state]--;
2150 nm_i->nid_cnt[dst_state]++;
2152 switch (dst_state) {
2157 list_add_tail(&i->list, &nm_i->free_nid_list);
2164 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2165 bool set, bool build)
2167 struct f2fs_nm_info *nm_i = NM_I(sbi);
2168 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2169 unsigned int nid_ofs = nid - START_NID(nid);
2171 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2175 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2177 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2178 nm_i->free_nid_count[nat_ofs]++;
2180 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2182 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2184 nm_i->free_nid_count[nat_ofs]--;
2188 /* return if the nid is recognized as free */
2189 static bool add_free_nid(struct f2fs_sb_info *sbi,
2190 nid_t nid, bool build, bool update)
2192 struct f2fs_nm_info *nm_i = NM_I(sbi);
2193 struct free_nid *i, *e;
2194 struct nat_entry *ne;
2198 /* 0 nid should not be used */
2199 if (unlikely(nid == 0))
2202 if (unlikely(f2fs_check_nid_range(sbi, nid)))
2205 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2207 i->state = FREE_NID;
2209 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2211 spin_lock(&nm_i->nid_list_lock);
2219 * - __insert_nid_to_list(PREALLOC_NID)
2220 * - f2fs_balance_fs_bg
2221 * - f2fs_build_free_nids
2222 * - __f2fs_build_free_nids
2225 * - __lookup_nat_cache
2227 * - f2fs_init_inode_metadata
2228 * - f2fs_new_inode_page
2229 * - f2fs_new_node_page
2231 * - f2fs_alloc_nid_done
2232 * - __remove_nid_from_list(PREALLOC_NID)
2233 * - __insert_nid_to_list(FREE_NID)
2235 ne = __lookup_nat_cache(nm_i, nid);
2236 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2237 nat_get_blkaddr(ne) != NULL_ADDR))
2240 e = __lookup_free_nid_list(nm_i, nid);
2242 if (e->state == FREE_NID)
2248 err = __insert_free_nid(sbi, i);
2251 update_free_nid_bitmap(sbi, nid, ret, build);
2253 nm_i->available_nids++;
2255 spin_unlock(&nm_i->nid_list_lock);
2256 radix_tree_preload_end();
2259 kmem_cache_free(free_nid_slab, i);
2263 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2265 struct f2fs_nm_info *nm_i = NM_I(sbi);
2267 bool need_free = false;
2269 spin_lock(&nm_i->nid_list_lock);
2270 i = __lookup_free_nid_list(nm_i, nid);
2271 if (i && i->state == FREE_NID) {
2272 __remove_free_nid(sbi, i, FREE_NID);
2275 spin_unlock(&nm_i->nid_list_lock);
2278 kmem_cache_free(free_nid_slab, i);
2281 static int scan_nat_page(struct f2fs_sb_info *sbi,
2282 struct page *nat_page, nid_t start_nid)
2284 struct f2fs_nm_info *nm_i = NM_I(sbi);
2285 struct f2fs_nat_block *nat_blk = page_address(nat_page);
2287 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2290 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2292 i = start_nid % NAT_ENTRY_PER_BLOCK;
2294 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2295 if (unlikely(start_nid >= nm_i->max_nid))
2298 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2300 if (blk_addr == NEW_ADDR)
2303 if (blk_addr == NULL_ADDR) {
2304 add_free_nid(sbi, start_nid, true, true);
2306 spin_lock(&NM_I(sbi)->nid_list_lock);
2307 update_free_nid_bitmap(sbi, start_nid, false, true);
2308 spin_unlock(&NM_I(sbi)->nid_list_lock);
2315 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2317 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2318 struct f2fs_journal *journal = curseg->journal;
2321 down_read(&curseg->journal_rwsem);
2322 for (i = 0; i < nats_in_cursum(journal); i++) {
2326 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2327 nid = le32_to_cpu(nid_in_journal(journal, i));
2328 if (addr == NULL_ADDR)
2329 add_free_nid(sbi, nid, true, false);
2331 remove_free_nid(sbi, nid);
2333 up_read(&curseg->journal_rwsem);
2336 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2338 struct f2fs_nm_info *nm_i = NM_I(sbi);
2339 unsigned int i, idx;
2342 down_read(&nm_i->nat_tree_lock);
2344 for (i = 0; i < nm_i->nat_blocks; i++) {
2345 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2347 if (!nm_i->free_nid_count[i])
2349 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2350 idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2351 NAT_ENTRY_PER_BLOCK, idx);
2352 if (idx >= NAT_ENTRY_PER_BLOCK)
2355 nid = i * NAT_ENTRY_PER_BLOCK + idx;
2356 add_free_nid(sbi, nid, true, false);
2358 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2363 scan_curseg_cache(sbi);
2365 up_read(&nm_i->nat_tree_lock);
2368 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2369 bool sync, bool mount)
2371 struct f2fs_nm_info *nm_i = NM_I(sbi);
2373 nid_t nid = nm_i->next_scan_nid;
2375 if (unlikely(nid >= nm_i->max_nid))
2378 if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2379 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2381 /* Enough entries */
2382 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2385 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2389 /* try to find free nids in free_nid_bitmap */
2390 scan_free_nid_bits(sbi);
2392 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2396 /* readahead nat pages to be scanned */
2397 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2400 down_read(&nm_i->nat_tree_lock);
2403 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2404 nm_i->nat_block_bitmap)) {
2405 struct page *page = get_current_nat_page(sbi, nid);
2408 ret = PTR_ERR(page);
2410 ret = scan_nat_page(sbi, page, nid);
2411 f2fs_put_page(page, 1);
2415 up_read(&nm_i->nat_tree_lock);
2416 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2421 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2422 if (unlikely(nid >= nm_i->max_nid))
2425 if (++i >= FREE_NID_PAGES)
2429 /* go to the next free nat pages to find free nids abundantly */
2430 nm_i->next_scan_nid = nid;
2432 /* find free nids from current sum_pages */
2433 scan_curseg_cache(sbi);
2435 up_read(&nm_i->nat_tree_lock);
2437 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2438 nm_i->ra_nid_pages, META_NAT, false);
2443 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2447 mutex_lock(&NM_I(sbi)->build_lock);
2448 ret = __f2fs_build_free_nids(sbi, sync, mount);
2449 mutex_unlock(&NM_I(sbi)->build_lock);
2455 * If this function returns success, caller can obtain a new nid
2456 * from second parameter of this function.
2457 * The returned nid could be used ino as well as nid when inode is created.
2459 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2461 struct f2fs_nm_info *nm_i = NM_I(sbi);
2462 struct free_nid *i = NULL;
2464 if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2465 f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2469 spin_lock(&nm_i->nid_list_lock);
2471 if (unlikely(nm_i->available_nids == 0)) {
2472 spin_unlock(&nm_i->nid_list_lock);
2476 /* We should not use stale free nids created by f2fs_build_free_nids */
2477 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2478 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2479 i = list_first_entry(&nm_i->free_nid_list,
2480 struct free_nid, list);
2483 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2484 nm_i->available_nids--;
2486 update_free_nid_bitmap(sbi, *nid, false, false);
2488 spin_unlock(&nm_i->nid_list_lock);
2491 spin_unlock(&nm_i->nid_list_lock);
2493 /* Let's scan nat pages and its caches to get free nids */
2494 if (!f2fs_build_free_nids(sbi, true, false))
2500 * f2fs_alloc_nid() should be called prior to this function.
2502 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2504 struct f2fs_nm_info *nm_i = NM_I(sbi);
2507 spin_lock(&nm_i->nid_list_lock);
2508 i = __lookup_free_nid_list(nm_i, nid);
2509 f2fs_bug_on(sbi, !i);
2510 __remove_free_nid(sbi, i, PREALLOC_NID);
2511 spin_unlock(&nm_i->nid_list_lock);
2513 kmem_cache_free(free_nid_slab, i);
2517 * f2fs_alloc_nid() should be called prior to this function.
2519 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2521 struct f2fs_nm_info *nm_i = NM_I(sbi);
2523 bool need_free = false;
2528 spin_lock(&nm_i->nid_list_lock);
2529 i = __lookup_free_nid_list(nm_i, nid);
2530 f2fs_bug_on(sbi, !i);
2532 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2533 __remove_free_nid(sbi, i, PREALLOC_NID);
2536 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2539 nm_i->available_nids++;
2541 update_free_nid_bitmap(sbi, nid, true, false);
2543 spin_unlock(&nm_i->nid_list_lock);
2546 kmem_cache_free(free_nid_slab, i);
2549 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2551 struct f2fs_nm_info *nm_i = NM_I(sbi);
2554 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2557 if (!mutex_trylock(&nm_i->build_lock))
2560 while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2561 struct free_nid *i, *next;
2562 unsigned int batch = SHRINK_NID_BATCH_SIZE;
2564 spin_lock(&nm_i->nid_list_lock);
2565 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2566 if (!nr_shrink || !batch ||
2567 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2569 __remove_free_nid(sbi, i, FREE_NID);
2570 kmem_cache_free(free_nid_slab, i);
2574 spin_unlock(&nm_i->nid_list_lock);
2577 mutex_unlock(&nm_i->build_lock);
2579 return nr - nr_shrink;
2582 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2584 void *src_addr, *dst_addr;
2587 struct f2fs_inode *ri;
2589 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2591 return PTR_ERR(ipage);
2593 ri = F2FS_INODE(page);
2594 if (ri->i_inline & F2FS_INLINE_XATTR) {
2595 if (!f2fs_has_inline_xattr(inode)) {
2596 set_inode_flag(inode, FI_INLINE_XATTR);
2597 stat_inc_inline_xattr(inode);
2600 if (f2fs_has_inline_xattr(inode)) {
2601 stat_dec_inline_xattr(inode);
2602 clear_inode_flag(inode, FI_INLINE_XATTR);
2607 dst_addr = inline_xattr_addr(inode, ipage);
2608 src_addr = inline_xattr_addr(inode, page);
2609 inline_size = inline_xattr_size(inode);
2611 f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2612 memcpy(dst_addr, src_addr, inline_size);
2614 f2fs_update_inode(inode, ipage);
2615 f2fs_put_page(ipage, 1);
2619 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2621 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2622 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2624 struct dnode_of_data dn;
2625 struct node_info ni;
2632 /* 1: invalidate the previous xattr nid */
2633 err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2637 f2fs_invalidate_blocks(sbi, ni.blk_addr);
2638 dec_valid_node_count(sbi, inode, false);
2639 set_node_addr(sbi, &ni, NULL_ADDR, false);
2642 /* 2: update xattr nid in inode */
2643 if (!f2fs_alloc_nid(sbi, &new_xnid))
2646 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2647 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2648 if (IS_ERR(xpage)) {
2649 f2fs_alloc_nid_failed(sbi, new_xnid);
2650 return PTR_ERR(xpage);
2653 f2fs_alloc_nid_done(sbi, new_xnid);
2654 f2fs_update_inode_page(inode);
2656 /* 3: update and set xattr node page dirty */
2657 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2659 set_page_dirty(xpage);
2660 f2fs_put_page(xpage, 1);
2665 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2667 struct f2fs_inode *src, *dst;
2668 nid_t ino = ino_of_node(page);
2669 struct node_info old_ni, new_ni;
2673 err = f2fs_get_node_info(sbi, ino, &old_ni);
2677 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2680 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2682 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2686 /* Should not use this inode from free nid list */
2687 remove_free_nid(sbi, ino);
2689 if (!PageUptodate(ipage))
2690 SetPageUptodate(ipage);
2691 fill_node_footer(ipage, ino, ino, 0, true);
2692 set_cold_node(ipage, false);
2694 src = F2FS_INODE(page);
2695 dst = F2FS_INODE(ipage);
2697 memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2699 dst->i_blocks = cpu_to_le64(1);
2700 dst->i_links = cpu_to_le32(1);
2701 dst->i_xattr_nid = 0;
2702 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2703 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2704 dst->i_extra_isize = src->i_extra_isize;
2706 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2707 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2708 i_inline_xattr_size))
2709 dst->i_inline_xattr_size = src->i_inline_xattr_size;
2711 if (f2fs_sb_has_project_quota(sbi) &&
2712 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2714 dst->i_projid = src->i_projid;
2716 if (f2fs_sb_has_inode_crtime(sbi) &&
2717 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2719 dst->i_crtime = src->i_crtime;
2720 dst->i_crtime_nsec = src->i_crtime_nsec;
2727 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2729 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2730 inc_valid_inode_count(sbi);
2731 set_page_dirty(ipage);
2732 f2fs_put_page(ipage, 1);
2736 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2737 unsigned int segno, struct f2fs_summary_block *sum)
2739 struct f2fs_node *rn;
2740 struct f2fs_summary *sum_entry;
2742 int i, idx, last_offset, nrpages;
2744 /* scan the node segment */
2745 last_offset = sbi->blocks_per_seg;
2746 addr = START_BLOCK(sbi, segno);
2747 sum_entry = &sum->entries[0];
2749 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2750 nrpages = bio_max_segs(last_offset - i);
2752 /* readahead node pages */
2753 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2755 for (idx = addr; idx < addr + nrpages; idx++) {
2756 struct page *page = f2fs_get_tmp_page(sbi, idx);
2759 return PTR_ERR(page);
2761 rn = F2FS_NODE(page);
2762 sum_entry->nid = rn->footer.nid;
2763 sum_entry->version = 0;
2764 sum_entry->ofs_in_node = 0;
2766 f2fs_put_page(page, 1);
2769 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2775 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2777 struct f2fs_nm_info *nm_i = NM_I(sbi);
2778 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2779 struct f2fs_journal *journal = curseg->journal;
2782 down_write(&curseg->journal_rwsem);
2783 for (i = 0; i < nats_in_cursum(journal); i++) {
2784 struct nat_entry *ne;
2785 struct f2fs_nat_entry raw_ne;
2786 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2788 raw_ne = nat_in_journal(journal, i);
2790 ne = __lookup_nat_cache(nm_i, nid);
2792 ne = __alloc_nat_entry(nid, true);
2793 __init_nat_entry(nm_i, ne, &raw_ne, true);
2797 * if a free nat in journal has not been used after last
2798 * checkpoint, we should remove it from available nids,
2799 * since later we will add it again.
2801 if (!get_nat_flag(ne, IS_DIRTY) &&
2802 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2803 spin_lock(&nm_i->nid_list_lock);
2804 nm_i->available_nids--;
2805 spin_unlock(&nm_i->nid_list_lock);
2808 __set_nat_cache_dirty(nm_i, ne);
2810 update_nats_in_cursum(journal, -i);
2811 up_write(&curseg->journal_rwsem);
2814 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2815 struct list_head *head, int max)
2817 struct nat_entry_set *cur;
2819 if (nes->entry_cnt >= max)
2822 list_for_each_entry(cur, head, set_list) {
2823 if (cur->entry_cnt >= nes->entry_cnt) {
2824 list_add(&nes->set_list, cur->set_list.prev);
2829 list_add_tail(&nes->set_list, head);
2832 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2835 struct f2fs_nm_info *nm_i = NM_I(sbi);
2836 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2837 struct f2fs_nat_block *nat_blk = page_address(page);
2841 if (!enabled_nat_bits(sbi, NULL))
2844 if (nat_index == 0) {
2848 for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2849 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2853 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2854 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2858 __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2859 if (valid == NAT_ENTRY_PER_BLOCK)
2860 __set_bit_le(nat_index, nm_i->full_nat_bits);
2862 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2865 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2866 struct nat_entry_set *set, struct cp_control *cpc)
2868 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2869 struct f2fs_journal *journal = curseg->journal;
2870 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2871 bool to_journal = true;
2872 struct f2fs_nat_block *nat_blk;
2873 struct nat_entry *ne, *cur;
2874 struct page *page = NULL;
2877 * there are two steps to flush nat entries:
2878 * #1, flush nat entries to journal in current hot data summary block.
2879 * #2, flush nat entries to nat page.
2881 if (enabled_nat_bits(sbi, cpc) ||
2882 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2886 down_write(&curseg->journal_rwsem);
2888 page = get_next_nat_page(sbi, start_nid);
2890 return PTR_ERR(page);
2892 nat_blk = page_address(page);
2893 f2fs_bug_on(sbi, !nat_blk);
2896 /* flush dirty nats in nat entry set */
2897 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2898 struct f2fs_nat_entry *raw_ne;
2899 nid_t nid = nat_get_nid(ne);
2902 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2905 offset = f2fs_lookup_journal_in_cursum(journal,
2906 NAT_JOURNAL, nid, 1);
2907 f2fs_bug_on(sbi, offset < 0);
2908 raw_ne = &nat_in_journal(journal, offset);
2909 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2911 raw_ne = &nat_blk->entries[nid - start_nid];
2913 raw_nat_from_node_info(raw_ne, &ne->ni);
2915 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2916 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2917 add_free_nid(sbi, nid, false, true);
2919 spin_lock(&NM_I(sbi)->nid_list_lock);
2920 update_free_nid_bitmap(sbi, nid, false, false);
2921 spin_unlock(&NM_I(sbi)->nid_list_lock);
2926 up_write(&curseg->journal_rwsem);
2928 __update_nat_bits(sbi, start_nid, page);
2929 f2fs_put_page(page, 1);
2932 /* Allow dirty nats by node block allocation in write_begin */
2933 if (!set->entry_cnt) {
2934 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2935 kmem_cache_free(nat_entry_set_slab, set);
2941 * This function is called during the checkpointing process.
2943 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2945 struct f2fs_nm_info *nm_i = NM_I(sbi);
2946 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2947 struct f2fs_journal *journal = curseg->journal;
2948 struct nat_entry_set *setvec[SETVEC_SIZE];
2949 struct nat_entry_set *set, *tmp;
2956 * during unmount, let's flush nat_bits before checking
2957 * nat_cnt[DIRTY_NAT].
2959 if (enabled_nat_bits(sbi, cpc)) {
2960 down_write(&nm_i->nat_tree_lock);
2961 remove_nats_in_journal(sbi);
2962 up_write(&nm_i->nat_tree_lock);
2965 if (!nm_i->nat_cnt[DIRTY_NAT])
2968 down_write(&nm_i->nat_tree_lock);
2971 * if there are no enough space in journal to store dirty nat
2972 * entries, remove all entries from journal and merge them
2973 * into nat entry set.
2975 if (enabled_nat_bits(sbi, cpc) ||
2976 !__has_cursum_space(journal,
2977 nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
2978 remove_nats_in_journal(sbi);
2980 while ((found = __gang_lookup_nat_set(nm_i,
2981 set_idx, SETVEC_SIZE, setvec))) {
2983 set_idx = setvec[found - 1]->set + 1;
2984 for (idx = 0; idx < found; idx++)
2985 __adjust_nat_entry_set(setvec[idx], &sets,
2986 MAX_NAT_JENTRIES(journal));
2989 /* flush dirty nats in nat entry set */
2990 list_for_each_entry_safe(set, tmp, &sets, set_list) {
2991 err = __flush_nat_entry_set(sbi, set, cpc);
2996 up_write(&nm_i->nat_tree_lock);
2997 /* Allow dirty nats by node block allocation in write_begin */
3002 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3004 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3005 struct f2fs_nm_info *nm_i = NM_I(sbi);
3006 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3008 __u64 cp_ver = cur_cp_version(ckpt);
3009 block_t nat_bits_addr;
3011 if (!enabled_nat_bits(sbi, NULL))
3014 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3015 nm_i->nat_bits = f2fs_kvzalloc(sbi,
3016 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3017 if (!nm_i->nat_bits)
3020 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3021 nm_i->nat_bits_blocks;
3022 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3025 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3027 return PTR_ERR(page);
3029 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3030 page_address(page), F2FS_BLKSIZE);
3031 f2fs_put_page(page, 1);
3034 cp_ver |= (cur_cp_crc(ckpt) << 32);
3035 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3036 disable_nat_bits(sbi, true);
3040 nm_i->full_nat_bits = nm_i->nat_bits + 8;
3041 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3043 f2fs_notice(sbi, "Found nat_bits in checkpoint");
3047 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3049 struct f2fs_nm_info *nm_i = NM_I(sbi);
3051 nid_t nid, last_nid;
3053 if (!enabled_nat_bits(sbi, NULL))
3056 for (i = 0; i < nm_i->nat_blocks; i++) {
3057 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3058 if (i >= nm_i->nat_blocks)
3061 __set_bit_le(i, nm_i->nat_block_bitmap);
3063 nid = i * NAT_ENTRY_PER_BLOCK;
3064 last_nid = nid + NAT_ENTRY_PER_BLOCK;
3066 spin_lock(&NM_I(sbi)->nid_list_lock);
3067 for (; nid < last_nid; nid++)
3068 update_free_nid_bitmap(sbi, nid, true, true);
3069 spin_unlock(&NM_I(sbi)->nid_list_lock);
3072 for (i = 0; i < nm_i->nat_blocks; i++) {
3073 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3074 if (i >= nm_i->nat_blocks)
3077 __set_bit_le(i, nm_i->nat_block_bitmap);
3081 static int init_node_manager(struct f2fs_sb_info *sbi)
3083 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3084 struct f2fs_nm_info *nm_i = NM_I(sbi);
3085 unsigned char *version_bitmap;
3086 unsigned int nat_segs;
3089 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3091 /* segment_count_nat includes pair segment so divide to 2. */
3092 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3093 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3094 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3096 /* not used nids: 0, node, meta, (and root counted as valid node) */
3097 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3098 F2FS_RESERVED_NODE_NUM;
3099 nm_i->nid_cnt[FREE_NID] = 0;
3100 nm_i->nid_cnt[PREALLOC_NID] = 0;
3101 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3102 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3103 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3105 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3106 INIT_LIST_HEAD(&nm_i->free_nid_list);
3107 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3108 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3109 INIT_LIST_HEAD(&nm_i->nat_entries);
3110 spin_lock_init(&nm_i->nat_list_lock);
3112 mutex_init(&nm_i->build_lock);
3113 spin_lock_init(&nm_i->nid_list_lock);
3114 init_rwsem(&nm_i->nat_tree_lock);
3116 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3117 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3118 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3119 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3121 if (!nm_i->nat_bitmap)
3124 err = __get_nat_bitmaps(sbi);
3128 #ifdef CONFIG_F2FS_CHECK_FS
3129 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3131 if (!nm_i->nat_bitmap_mir)
3138 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3140 struct f2fs_nm_info *nm_i = NM_I(sbi);
3143 nm_i->free_nid_bitmap =
3144 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3147 if (!nm_i->free_nid_bitmap)
3150 for (i = 0; i < nm_i->nat_blocks; i++) {
3151 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3152 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3153 if (!nm_i->free_nid_bitmap[i])
3157 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3159 if (!nm_i->nat_block_bitmap)
3162 nm_i->free_nid_count =
3163 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3166 if (!nm_i->free_nid_count)
3171 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3175 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3180 err = init_node_manager(sbi);
3184 err = init_free_nid_cache(sbi);
3188 /* load free nid status from nat_bits table */
3189 load_free_nid_bitmap(sbi);
3191 return f2fs_build_free_nids(sbi, true, true);
3194 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3196 struct f2fs_nm_info *nm_i = NM_I(sbi);
3197 struct free_nid *i, *next_i;
3198 struct nat_entry *natvec[NATVEC_SIZE];
3199 struct nat_entry_set *setvec[SETVEC_SIZE];
3206 /* destroy free nid list */
3207 spin_lock(&nm_i->nid_list_lock);
3208 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3209 __remove_free_nid(sbi, i, FREE_NID);
3210 spin_unlock(&nm_i->nid_list_lock);
3211 kmem_cache_free(free_nid_slab, i);
3212 spin_lock(&nm_i->nid_list_lock);
3214 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3215 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3216 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3217 spin_unlock(&nm_i->nid_list_lock);
3219 /* destroy nat cache */
3220 down_write(&nm_i->nat_tree_lock);
3221 while ((found = __gang_lookup_nat_cache(nm_i,
3222 nid, NATVEC_SIZE, natvec))) {
3225 nid = nat_get_nid(natvec[found - 1]) + 1;
3226 for (idx = 0; idx < found; idx++) {
3227 spin_lock(&nm_i->nat_list_lock);
3228 list_del(&natvec[idx]->list);
3229 spin_unlock(&nm_i->nat_list_lock);
3231 __del_from_nat_cache(nm_i, natvec[idx]);
3234 f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3236 /* destroy nat set cache */
3238 while ((found = __gang_lookup_nat_set(nm_i,
3239 nid, SETVEC_SIZE, setvec))) {
3242 nid = setvec[found - 1]->set + 1;
3243 for (idx = 0; idx < found; idx++) {
3244 /* entry_cnt is not zero, when cp_error was occurred */
3245 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3246 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3247 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3250 up_write(&nm_i->nat_tree_lock);
3252 kvfree(nm_i->nat_block_bitmap);
3253 if (nm_i->free_nid_bitmap) {
3256 for (i = 0; i < nm_i->nat_blocks; i++)
3257 kvfree(nm_i->free_nid_bitmap[i]);
3258 kvfree(nm_i->free_nid_bitmap);
3260 kvfree(nm_i->free_nid_count);
3262 kvfree(nm_i->nat_bitmap);
3263 kvfree(nm_i->nat_bits);
3264 #ifdef CONFIG_F2FS_CHECK_FS
3265 kvfree(nm_i->nat_bitmap_mir);
3267 sbi->nm_info = NULL;
3271 int __init f2fs_create_node_manager_caches(void)
3273 nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3274 sizeof(struct nat_entry));
3275 if (!nat_entry_slab)
3278 free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3279 sizeof(struct free_nid));
3281 goto destroy_nat_entry;
3283 nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3284 sizeof(struct nat_entry_set));
3285 if (!nat_entry_set_slab)
3286 goto destroy_free_nid;
3288 fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3289 sizeof(struct fsync_node_entry));
3290 if (!fsync_node_entry_slab)
3291 goto destroy_nat_entry_set;
3294 destroy_nat_entry_set:
3295 kmem_cache_destroy(nat_entry_set_slab);
3297 kmem_cache_destroy(free_nid_slab);
3299 kmem_cache_destroy(nat_entry_slab);
3304 void f2fs_destroy_node_manager_caches(void)
3306 kmem_cache_destroy(fsync_node_entry_slab);
3307 kmem_cache_destroy(nat_entry_set_slab);
3308 kmem_cache_destroy(free_nid_slab);
3309 kmem_cache_destroy(nat_entry_slab);