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);
46 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
48 unsigned long avail_ram;
49 unsigned long mem_size = 0;
57 /* only uses low memory */
58 avail_ram = val.totalram - val.totalhigh;
61 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
63 if (type == FREE_NIDS) {
64 mem_size = (nm_i->nid_cnt[FREE_NID] *
65 sizeof(struct free_nid)) >> PAGE_SHIFT;
66 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
67 } else if (type == NAT_ENTRIES) {
68 mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
69 sizeof(struct nat_entry)) >> PAGE_SHIFT;
70 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
71 if (excess_cached_nats(sbi))
73 } else if (type == DIRTY_DENTS) {
74 if (sbi->sb->s_bdi->wb.dirty_exceeded)
76 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
77 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
78 } else if (type == INO_ENTRIES) {
81 for (i = 0; i < MAX_INO_ENTRY; i++)
82 mem_size += sbi->im[i].ino_num *
83 sizeof(struct ino_entry);
84 mem_size >>= PAGE_SHIFT;
85 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
86 } else if (type == EXTENT_CACHE) {
87 mem_size = (atomic_read(&sbi->total_ext_tree) *
88 sizeof(struct extent_tree) +
89 atomic_read(&sbi->total_ext_node) *
90 sizeof(struct extent_node)) >> PAGE_SHIFT;
91 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
92 } else if (type == INMEM_PAGES) {
93 /* it allows 20% / total_ram for inmemory pages */
94 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
95 res = mem_size < (val.totalram / 5);
96 } else if (type == DISCARD_CACHE) {
97 mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
98 sizeof(struct discard_cmd)) >> PAGE_SHIFT;
99 res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
101 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
107 static void clear_node_page_dirty(struct page *page)
109 if (PageDirty(page)) {
110 f2fs_clear_page_cache_dirty_tag(page);
111 clear_page_dirty_for_io(page);
112 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
114 ClearPageUptodate(page);
117 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
119 return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
122 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
124 struct page *src_page;
125 struct page *dst_page;
129 struct f2fs_nm_info *nm_i = NM_I(sbi);
131 dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
133 /* get current nat block page with lock */
134 src_page = get_current_nat_page(sbi, nid);
135 if (IS_ERR(src_page))
137 dst_page = f2fs_grab_meta_page(sbi, dst_off);
138 f2fs_bug_on(sbi, PageDirty(src_page));
140 src_addr = page_address(src_page);
141 dst_addr = page_address(dst_page);
142 memcpy(dst_addr, src_addr, PAGE_SIZE);
143 set_page_dirty(dst_page);
144 f2fs_put_page(src_page, 1);
146 set_to_next_nat(nm_i, nid);
151 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
153 struct nat_entry *new;
156 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
158 new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
160 nat_set_nid(new, nid);
166 static void __free_nat_entry(struct nat_entry *e)
168 kmem_cache_free(nat_entry_slab, e);
171 /* must be locked by nat_tree_lock */
172 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
173 struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
176 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
177 else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
181 node_info_from_raw_nat(&ne->ni, raw_ne);
183 spin_lock(&nm_i->nat_list_lock);
184 list_add_tail(&ne->list, &nm_i->nat_entries);
185 spin_unlock(&nm_i->nat_list_lock);
187 nm_i->nat_cnt[TOTAL_NAT]++;
188 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
192 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
194 struct nat_entry *ne;
196 ne = radix_tree_lookup(&nm_i->nat_root, n);
198 /* for recent accessed nat entry, move it to tail of lru list */
199 if (ne && !get_nat_flag(ne, IS_DIRTY)) {
200 spin_lock(&nm_i->nat_list_lock);
201 if (!list_empty(&ne->list))
202 list_move_tail(&ne->list, &nm_i->nat_entries);
203 spin_unlock(&nm_i->nat_list_lock);
209 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
210 nid_t start, unsigned int nr, struct nat_entry **ep)
212 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
215 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
217 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
218 nm_i->nat_cnt[TOTAL_NAT]--;
219 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
223 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
224 struct nat_entry *ne)
226 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
227 struct nat_entry_set *head;
229 head = radix_tree_lookup(&nm_i->nat_set_root, set);
231 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
233 INIT_LIST_HEAD(&head->entry_list);
234 INIT_LIST_HEAD(&head->set_list);
237 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
242 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
243 struct nat_entry *ne)
245 struct nat_entry_set *head;
246 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
249 head = __grab_nat_entry_set(nm_i, ne);
252 * update entry_cnt in below condition:
253 * 1. update NEW_ADDR to valid block address;
254 * 2. update old block address to new one;
256 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
257 !get_nat_flag(ne, IS_DIRTY)))
260 set_nat_flag(ne, IS_PREALLOC, new_ne);
262 if (get_nat_flag(ne, IS_DIRTY))
265 nm_i->nat_cnt[DIRTY_NAT]++;
266 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
267 set_nat_flag(ne, IS_DIRTY, true);
269 spin_lock(&nm_i->nat_list_lock);
271 list_del_init(&ne->list);
273 list_move_tail(&ne->list, &head->entry_list);
274 spin_unlock(&nm_i->nat_list_lock);
277 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
278 struct nat_entry_set *set, struct nat_entry *ne)
280 spin_lock(&nm_i->nat_list_lock);
281 list_move_tail(&ne->list, &nm_i->nat_entries);
282 spin_unlock(&nm_i->nat_list_lock);
284 set_nat_flag(ne, IS_DIRTY, false);
286 nm_i->nat_cnt[DIRTY_NAT]--;
287 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
290 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
291 nid_t start, unsigned int nr, struct nat_entry_set **ep)
293 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
297 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
299 return NODE_MAPPING(sbi) == page->mapping &&
300 IS_DNODE(page) && is_cold_node(page);
303 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
305 spin_lock_init(&sbi->fsync_node_lock);
306 INIT_LIST_HEAD(&sbi->fsync_node_list);
307 sbi->fsync_seg_id = 0;
308 sbi->fsync_node_num = 0;
311 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
314 struct fsync_node_entry *fn;
318 fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS);
322 INIT_LIST_HEAD(&fn->list);
324 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
325 list_add_tail(&fn->list, &sbi->fsync_node_list);
326 fn->seq_id = sbi->fsync_seg_id++;
328 sbi->fsync_node_num++;
329 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
334 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
336 struct fsync_node_entry *fn;
339 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
340 list_for_each_entry(fn, &sbi->fsync_node_list, list) {
341 if (fn->page == page) {
343 sbi->fsync_node_num--;
344 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
345 kmem_cache_free(fsync_node_entry_slab, fn);
350 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
354 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
358 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
359 sbi->fsync_seg_id = 0;
360 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
363 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
365 struct f2fs_nm_info *nm_i = NM_I(sbi);
369 down_read(&nm_i->nat_tree_lock);
370 e = __lookup_nat_cache(nm_i, nid);
372 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
373 !get_nat_flag(e, HAS_FSYNCED_INODE))
376 up_read(&nm_i->nat_tree_lock);
380 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
382 struct f2fs_nm_info *nm_i = NM_I(sbi);
386 down_read(&nm_i->nat_tree_lock);
387 e = __lookup_nat_cache(nm_i, nid);
388 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
390 up_read(&nm_i->nat_tree_lock);
394 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
396 struct f2fs_nm_info *nm_i = NM_I(sbi);
398 bool need_update = true;
400 down_read(&nm_i->nat_tree_lock);
401 e = __lookup_nat_cache(nm_i, ino);
402 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
403 (get_nat_flag(e, IS_CHECKPOINTED) ||
404 get_nat_flag(e, HAS_FSYNCED_INODE)))
406 up_read(&nm_i->nat_tree_lock);
410 /* must be locked by nat_tree_lock */
411 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
412 struct f2fs_nat_entry *ne)
414 struct f2fs_nm_info *nm_i = NM_I(sbi);
415 struct nat_entry *new, *e;
417 new = __alloc_nat_entry(nid, false);
421 down_write(&nm_i->nat_tree_lock);
422 e = __lookup_nat_cache(nm_i, nid);
424 e = __init_nat_entry(nm_i, new, ne, false);
426 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
427 nat_get_blkaddr(e) !=
428 le32_to_cpu(ne->block_addr) ||
429 nat_get_version(e) != ne->version);
430 up_write(&nm_i->nat_tree_lock);
432 __free_nat_entry(new);
435 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
436 block_t new_blkaddr, bool fsync_done)
438 struct f2fs_nm_info *nm_i = NM_I(sbi);
440 struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
442 down_write(&nm_i->nat_tree_lock);
443 e = __lookup_nat_cache(nm_i, ni->nid);
445 e = __init_nat_entry(nm_i, new, NULL, true);
446 copy_node_info(&e->ni, ni);
447 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
448 } else if (new_blkaddr == NEW_ADDR) {
450 * when nid is reallocated,
451 * previous nat entry can be remained in nat cache.
452 * So, reinitialize it with new information.
454 copy_node_info(&e->ni, ni);
455 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
457 /* let's free early to reduce memory consumption */
459 __free_nat_entry(new);
462 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
463 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
464 new_blkaddr == NULL_ADDR);
465 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
466 new_blkaddr == NEW_ADDR);
467 f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
468 new_blkaddr == NEW_ADDR);
470 /* increment version no as node is removed */
471 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
472 unsigned char version = nat_get_version(e);
474 nat_set_version(e, inc_node_version(version));
478 nat_set_blkaddr(e, new_blkaddr);
479 if (!__is_valid_data_blkaddr(new_blkaddr))
480 set_nat_flag(e, IS_CHECKPOINTED, false);
481 __set_nat_cache_dirty(nm_i, e);
483 /* update fsync_mark if its inode nat entry is still alive */
484 if (ni->nid != ni->ino)
485 e = __lookup_nat_cache(nm_i, ni->ino);
487 if (fsync_done && ni->nid == ni->ino)
488 set_nat_flag(e, HAS_FSYNCED_INODE, true);
489 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
491 up_write(&nm_i->nat_tree_lock);
494 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
496 struct f2fs_nm_info *nm_i = NM_I(sbi);
499 if (!down_write_trylock(&nm_i->nat_tree_lock))
502 spin_lock(&nm_i->nat_list_lock);
504 struct nat_entry *ne;
506 if (list_empty(&nm_i->nat_entries))
509 ne = list_first_entry(&nm_i->nat_entries,
510 struct nat_entry, list);
512 spin_unlock(&nm_i->nat_list_lock);
514 __del_from_nat_cache(nm_i, ne);
517 spin_lock(&nm_i->nat_list_lock);
519 spin_unlock(&nm_i->nat_list_lock);
521 up_write(&nm_i->nat_tree_lock);
522 return nr - nr_shrink;
525 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
526 struct node_info *ni)
528 struct f2fs_nm_info *nm_i = NM_I(sbi);
529 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
530 struct f2fs_journal *journal = curseg->journal;
531 nid_t start_nid = START_NID(nid);
532 struct f2fs_nat_block *nat_blk;
533 struct page *page = NULL;
534 struct f2fs_nat_entry ne;
542 /* Check nat cache */
543 down_read(&nm_i->nat_tree_lock);
544 e = __lookup_nat_cache(nm_i, nid);
546 ni->ino = nat_get_ino(e);
547 ni->blk_addr = nat_get_blkaddr(e);
548 ni->version = nat_get_version(e);
549 up_read(&nm_i->nat_tree_lock);
553 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
555 /* Check current segment summary */
556 down_read(&curseg->journal_rwsem);
557 i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
559 ne = nat_in_journal(journal, i);
560 node_info_from_raw_nat(ni, &ne);
562 up_read(&curseg->journal_rwsem);
564 up_read(&nm_i->nat_tree_lock);
568 /* Fill node_info from nat page */
569 index = current_nat_addr(sbi, nid);
570 up_read(&nm_i->nat_tree_lock);
572 page = f2fs_get_meta_page(sbi, index);
574 return PTR_ERR(page);
576 nat_blk = (struct f2fs_nat_block *)page_address(page);
577 ne = nat_blk->entries[nid - start_nid];
578 node_info_from_raw_nat(ni, &ne);
579 f2fs_put_page(page, 1);
581 blkaddr = le32_to_cpu(ne.block_addr);
582 if (__is_valid_data_blkaddr(blkaddr) &&
583 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
586 /* cache nat entry */
587 cache_nat_entry(sbi, nid, &ne);
592 * readahead MAX_RA_NODE number of node pages.
594 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
596 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
597 struct blk_plug plug;
601 blk_start_plug(&plug);
603 /* Then, try readahead for siblings of the desired node */
605 end = min(end, NIDS_PER_BLOCK);
606 for (i = start; i < end; i++) {
607 nid = get_nid(parent, i, false);
608 f2fs_ra_node_page(sbi, nid);
611 blk_finish_plug(&plug);
614 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
616 const long direct_index = ADDRS_PER_INODE(dn->inode);
617 const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
618 const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
619 unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
620 int cur_level = dn->cur_level;
621 int max_level = dn->max_level;
627 while (max_level-- > cur_level)
628 skipped_unit *= NIDS_PER_BLOCK;
630 switch (dn->max_level) {
632 base += 2 * indirect_blks;
635 base += 2 * direct_blks;
638 base += direct_index;
641 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
644 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
648 * The maximum depth is four.
649 * Offset[0] will have raw inode offset.
651 static int get_node_path(struct inode *inode, long block,
652 int offset[4], unsigned int noffset[4])
654 const long direct_index = ADDRS_PER_INODE(inode);
655 const long direct_blks = ADDRS_PER_BLOCK(inode);
656 const long dptrs_per_blk = NIDS_PER_BLOCK;
657 const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
658 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
664 if (block < direct_index) {
668 block -= direct_index;
669 if (block < direct_blks) {
670 offset[n++] = NODE_DIR1_BLOCK;
676 block -= direct_blks;
677 if (block < direct_blks) {
678 offset[n++] = NODE_DIR2_BLOCK;
684 block -= direct_blks;
685 if (block < indirect_blks) {
686 offset[n++] = NODE_IND1_BLOCK;
688 offset[n++] = block / direct_blks;
689 noffset[n] = 4 + offset[n - 1];
690 offset[n] = block % direct_blks;
694 block -= indirect_blks;
695 if (block < indirect_blks) {
696 offset[n++] = NODE_IND2_BLOCK;
697 noffset[n] = 4 + dptrs_per_blk;
698 offset[n++] = block / direct_blks;
699 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
700 offset[n] = block % direct_blks;
704 block -= indirect_blks;
705 if (block < dindirect_blks) {
706 offset[n++] = NODE_DIND_BLOCK;
707 noffset[n] = 5 + (dptrs_per_blk * 2);
708 offset[n++] = block / indirect_blks;
709 noffset[n] = 6 + (dptrs_per_blk * 2) +
710 offset[n - 1] * (dptrs_per_blk + 1);
711 offset[n++] = (block / direct_blks) % dptrs_per_blk;
712 noffset[n] = 7 + (dptrs_per_blk * 2) +
713 offset[n - 2] * (dptrs_per_blk + 1) +
715 offset[n] = block % direct_blks;
726 * Caller should call f2fs_put_dnode(dn).
727 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
728 * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
730 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
732 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
733 struct page *npage[4];
734 struct page *parent = NULL;
736 unsigned int noffset[4];
741 level = get_node_path(dn->inode, index, offset, noffset);
745 nids[0] = dn->inode->i_ino;
746 npage[0] = dn->inode_page;
749 npage[0] = f2fs_get_node_page(sbi, nids[0]);
750 if (IS_ERR(npage[0]))
751 return PTR_ERR(npage[0]);
754 /* if inline_data is set, should not report any block indices */
755 if (f2fs_has_inline_data(dn->inode) && index) {
757 f2fs_put_page(npage[0], 1);
763 nids[1] = get_nid(parent, offset[0], true);
764 dn->inode_page = npage[0];
765 dn->inode_page_locked = true;
767 /* get indirect or direct nodes */
768 for (i = 1; i <= level; i++) {
771 if (!nids[i] && mode == ALLOC_NODE) {
773 if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
779 npage[i] = f2fs_new_node_page(dn, noffset[i]);
780 if (IS_ERR(npage[i])) {
781 f2fs_alloc_nid_failed(sbi, nids[i]);
782 err = PTR_ERR(npage[i]);
786 set_nid(parent, offset[i - 1], nids[i], i == 1);
787 f2fs_alloc_nid_done(sbi, nids[i]);
789 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
790 npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
791 if (IS_ERR(npage[i])) {
792 err = PTR_ERR(npage[i]);
798 dn->inode_page_locked = false;
801 f2fs_put_page(parent, 1);
805 npage[i] = f2fs_get_node_page(sbi, nids[i]);
806 if (IS_ERR(npage[i])) {
807 err = PTR_ERR(npage[i]);
808 f2fs_put_page(npage[0], 0);
814 nids[i + 1] = get_nid(parent, offset[i], false);
817 dn->nid = nids[level];
818 dn->ofs_in_node = offset[level];
819 dn->node_page = npage[level];
820 dn->data_blkaddr = f2fs_data_blkaddr(dn);
824 f2fs_put_page(parent, 1);
826 f2fs_put_page(npage[0], 0);
828 dn->inode_page = NULL;
829 dn->node_page = NULL;
830 if (err == -ENOENT) {
832 dn->max_level = level;
833 dn->ofs_in_node = offset[level];
838 static int truncate_node(struct dnode_of_data *dn)
840 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
845 err = f2fs_get_node_info(sbi, dn->nid, &ni);
849 /* Deallocate node address */
850 f2fs_invalidate_blocks(sbi, ni.blk_addr);
851 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
852 set_node_addr(sbi, &ni, NULL_ADDR, false);
854 if (dn->nid == dn->inode->i_ino) {
855 f2fs_remove_orphan_inode(sbi, dn->nid);
856 dec_valid_inode_count(sbi);
857 f2fs_inode_synced(dn->inode);
860 clear_node_page_dirty(dn->node_page);
861 set_sbi_flag(sbi, SBI_IS_DIRTY);
863 index = dn->node_page->index;
864 f2fs_put_page(dn->node_page, 1);
866 invalidate_mapping_pages(NODE_MAPPING(sbi),
869 dn->node_page = NULL;
870 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
875 static int truncate_dnode(struct dnode_of_data *dn)
883 /* get direct node */
884 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
885 if (PTR_ERR(page) == -ENOENT)
887 else if (IS_ERR(page))
888 return PTR_ERR(page);
890 /* Make dnode_of_data for parameter */
891 dn->node_page = page;
893 f2fs_truncate_data_blocks(dn);
894 err = truncate_node(dn);
901 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
904 struct dnode_of_data rdn = *dn;
906 struct f2fs_node *rn;
908 unsigned int child_nofs;
913 return NIDS_PER_BLOCK + 1;
915 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
917 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
919 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
920 return PTR_ERR(page);
923 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
925 rn = F2FS_NODE(page);
927 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
928 child_nid = le32_to_cpu(rn->in.nid[i]);
932 ret = truncate_dnode(&rdn);
935 if (set_nid(page, i, 0, false))
936 dn->node_changed = true;
939 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
940 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
941 child_nid = le32_to_cpu(rn->in.nid[i]);
942 if (child_nid == 0) {
943 child_nofs += NIDS_PER_BLOCK + 1;
947 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
948 if (ret == (NIDS_PER_BLOCK + 1)) {
949 if (set_nid(page, i, 0, false))
950 dn->node_changed = true;
952 } else if (ret < 0 && ret != -ENOENT) {
960 /* remove current indirect node */
961 dn->node_page = page;
962 ret = truncate_node(dn);
967 f2fs_put_page(page, 1);
969 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
973 f2fs_put_page(page, 1);
974 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
978 static int truncate_partial_nodes(struct dnode_of_data *dn,
979 struct f2fs_inode *ri, int *offset, int depth)
981 struct page *pages[2];
988 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
992 /* get indirect nodes in the path */
993 for (i = 0; i < idx + 1; i++) {
994 /* reference count'll be increased */
995 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
996 if (IS_ERR(pages[i])) {
997 err = PTR_ERR(pages[i]);
1001 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1004 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1006 /* free direct nodes linked to a partial indirect node */
1007 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1008 child_nid = get_nid(pages[idx], i, false);
1011 dn->nid = child_nid;
1012 err = truncate_dnode(dn);
1015 if (set_nid(pages[idx], i, 0, false))
1016 dn->node_changed = true;
1019 if (offset[idx + 1] == 0) {
1020 dn->node_page = pages[idx];
1022 err = truncate_node(dn);
1026 f2fs_put_page(pages[idx], 1);
1029 offset[idx + 1] = 0;
1032 for (i = idx; i >= 0; i--)
1033 f2fs_put_page(pages[i], 1);
1035 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1041 * All the block addresses of data and nodes should be nullified.
1043 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1045 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1046 int err = 0, cont = 1;
1047 int level, offset[4], noffset[4];
1048 unsigned int nofs = 0;
1049 struct f2fs_inode *ri;
1050 struct dnode_of_data dn;
1053 trace_f2fs_truncate_inode_blocks_enter(inode, from);
1055 level = get_node_path(inode, from, offset, noffset);
1057 trace_f2fs_truncate_inode_blocks_exit(inode, level);
1061 page = f2fs_get_node_page(sbi, inode->i_ino);
1063 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1064 return PTR_ERR(page);
1067 set_new_dnode(&dn, inode, page, NULL, 0);
1070 ri = F2FS_INODE(page);
1078 if (!offset[level - 1])
1080 err = truncate_partial_nodes(&dn, ri, offset, level);
1081 if (err < 0 && err != -ENOENT)
1083 nofs += 1 + NIDS_PER_BLOCK;
1086 nofs = 5 + 2 * NIDS_PER_BLOCK;
1087 if (!offset[level - 1])
1089 err = truncate_partial_nodes(&dn, ri, offset, level);
1090 if (err < 0 && err != -ENOENT)
1099 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1100 switch (offset[0]) {
1101 case NODE_DIR1_BLOCK:
1102 case NODE_DIR2_BLOCK:
1103 err = truncate_dnode(&dn);
1106 case NODE_IND1_BLOCK:
1107 case NODE_IND2_BLOCK:
1108 err = truncate_nodes(&dn, nofs, offset[1], 2);
1111 case NODE_DIND_BLOCK:
1112 err = truncate_nodes(&dn, nofs, offset[1], 3);
1119 if (err < 0 && err != -ENOENT)
1121 if (offset[1] == 0 &&
1122 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1124 BUG_ON(page->mapping != NODE_MAPPING(sbi));
1125 f2fs_wait_on_page_writeback(page, NODE, true, true);
1126 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1127 set_page_dirty(page);
1135 f2fs_put_page(page, 0);
1136 trace_f2fs_truncate_inode_blocks_exit(inode, err);
1137 return err > 0 ? 0 : err;
1140 /* caller must lock inode page */
1141 int f2fs_truncate_xattr_node(struct inode *inode)
1143 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1144 nid_t nid = F2FS_I(inode)->i_xattr_nid;
1145 struct dnode_of_data dn;
1152 npage = f2fs_get_node_page(sbi, nid);
1154 return PTR_ERR(npage);
1156 set_new_dnode(&dn, inode, NULL, npage, nid);
1157 err = truncate_node(&dn);
1159 f2fs_put_page(npage, 1);
1163 f2fs_i_xnid_write(inode, 0);
1169 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1172 int f2fs_remove_inode_page(struct inode *inode)
1174 struct dnode_of_data dn;
1177 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1178 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1182 err = f2fs_truncate_xattr_node(inode);
1184 f2fs_put_dnode(&dn);
1188 /* remove potential inline_data blocks */
1189 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1190 S_ISLNK(inode->i_mode))
1191 f2fs_truncate_data_blocks_range(&dn, 1);
1193 /* 0 is possible, after f2fs_new_inode() has failed */
1194 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1195 f2fs_put_dnode(&dn);
1199 if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1200 f2fs_warn(F2FS_I_SB(inode),
1201 "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1202 inode->i_ino, (unsigned long long)inode->i_blocks);
1203 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1206 /* will put inode & node pages */
1207 err = truncate_node(&dn);
1209 f2fs_put_dnode(&dn);
1215 struct page *f2fs_new_inode_page(struct inode *inode)
1217 struct dnode_of_data dn;
1219 /* allocate inode page for new inode */
1220 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1222 /* caller should f2fs_put_page(page, 1); */
1223 return f2fs_new_node_page(&dn, 0);
1226 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1228 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1229 struct node_info new_ni;
1233 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1234 return ERR_PTR(-EPERM);
1236 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1238 return ERR_PTR(-ENOMEM);
1240 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1243 #ifdef CONFIG_F2FS_CHECK_FS
1244 err = f2fs_get_node_info(sbi, dn->nid, &new_ni);
1246 dec_valid_node_count(sbi, dn->inode, !ofs);
1249 f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
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;
1318 err = f2fs_submit_page_bio(&fio);
1321 f2fs_update_iostat(sbi, FS_NODE_READ_IO, F2FS_BLKSIZE);
1327 * Readahead a node page
1329 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1336 if (f2fs_check_nid_range(sbi, nid))
1339 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1343 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1347 err = read_node_page(apage, REQ_RAHEAD);
1348 f2fs_put_page(apage, err ? 1 : 0);
1351 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1352 struct page *parent, int start)
1358 return ERR_PTR(-ENOENT);
1359 if (f2fs_check_nid_range(sbi, nid))
1360 return ERR_PTR(-EINVAL);
1362 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1364 return ERR_PTR(-ENOMEM);
1366 err = read_node_page(page, 0);
1368 f2fs_put_page(page, 1);
1369 return ERR_PTR(err);
1370 } else if (err == LOCKED_PAGE) {
1376 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1380 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1381 f2fs_put_page(page, 1);
1385 if (unlikely(!PageUptodate(page))) {
1390 if (!f2fs_inode_chksum_verify(sbi, page)) {
1395 if (unlikely(nid != nid_of_node(page))) {
1396 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1397 nid, nid_of_node(page), ino_of_node(page),
1398 ofs_of_node(page), cpver_of_node(page),
1399 next_blkaddr_of_node(page));
1402 ClearPageUptodate(page);
1403 f2fs_put_page(page, 1);
1404 return ERR_PTR(err);
1409 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1411 return __get_node_page(sbi, nid, NULL, 0);
1414 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1416 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1417 nid_t nid = get_nid(parent, start, false);
1419 return __get_node_page(sbi, nid, parent, start);
1422 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1424 struct inode *inode;
1428 /* should flush inline_data before evict_inode */
1429 inode = ilookup(sbi->sb, ino);
1433 page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1434 FGP_LOCK|FGP_NOWAIT, 0);
1438 if (!PageUptodate(page))
1441 if (!PageDirty(page))
1444 if (!clear_page_dirty_for_io(page))
1447 ret = f2fs_write_inline_data(inode, page);
1448 inode_dec_dirty_pages(inode);
1449 f2fs_remove_dirty_inode(inode);
1451 set_page_dirty(page);
1453 f2fs_put_page(page, 1);
1458 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1461 struct pagevec pvec;
1462 struct page *last_page = NULL;
1465 pagevec_init(&pvec);
1468 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1469 PAGECACHE_TAG_DIRTY))) {
1472 for (i = 0; i < nr_pages; i++) {
1473 struct page *page = pvec.pages[i];
1475 if (unlikely(f2fs_cp_error(sbi))) {
1476 f2fs_put_page(last_page, 0);
1477 pagevec_release(&pvec);
1478 return ERR_PTR(-EIO);
1481 if (!IS_DNODE(page) || !is_cold_node(page))
1483 if (ino_of_node(page) != ino)
1488 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1493 if (ino_of_node(page) != ino)
1494 goto continue_unlock;
1496 if (!PageDirty(page)) {
1497 /* someone wrote it for us */
1498 goto continue_unlock;
1502 f2fs_put_page(last_page, 0);
1508 pagevec_release(&pvec);
1514 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1515 struct writeback_control *wbc, bool do_balance,
1516 enum iostat_type io_type, unsigned int *seq_id)
1518 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1520 struct node_info ni;
1521 struct f2fs_io_info fio = {
1523 .ino = ino_of_node(page),
1526 .op_flags = wbc_to_write_flags(wbc),
1528 .encrypted_page = NULL,
1535 trace_f2fs_writepage(page, NODE);
1537 if (unlikely(f2fs_cp_error(sbi))) {
1538 if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
1539 ClearPageUptodate(page);
1540 dec_page_count(sbi, F2FS_DIRTY_NODES);
1547 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1550 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1551 wbc->sync_mode == WB_SYNC_NONE &&
1552 IS_DNODE(page) && is_cold_node(page))
1555 /* get old block addr of this node page */
1556 nid = nid_of_node(page);
1557 f2fs_bug_on(sbi, page->index != nid);
1559 if (f2fs_get_node_info(sbi, nid, &ni))
1562 if (wbc->for_reclaim) {
1563 if (!down_read_trylock(&sbi->node_write))
1566 down_read(&sbi->node_write);
1569 /* This page is already truncated */
1570 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1571 ClearPageUptodate(page);
1572 dec_page_count(sbi, F2FS_DIRTY_NODES);
1573 up_read(&sbi->node_write);
1578 if (__is_valid_data_blkaddr(ni.blk_addr) &&
1579 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1580 DATA_GENERIC_ENHANCE)) {
1581 up_read(&sbi->node_write);
1585 if (atomic && !test_opt(sbi, NOBARRIER))
1586 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1588 /* should add to global list before clearing PAGECACHE status */
1589 if (f2fs_in_warm_node_list(sbi, page)) {
1590 seq = f2fs_add_fsync_node_entry(sbi, page);
1595 set_page_writeback(page);
1596 ClearPageError(page);
1598 fio.old_blkaddr = ni.blk_addr;
1599 f2fs_do_write_node_page(nid, &fio);
1600 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1601 dec_page_count(sbi, F2FS_DIRTY_NODES);
1602 up_read(&sbi->node_write);
1604 if (wbc->for_reclaim) {
1605 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1611 if (unlikely(f2fs_cp_error(sbi))) {
1612 f2fs_submit_merged_write(sbi, NODE);
1616 *submitted = fio.submitted;
1619 f2fs_balance_fs(sbi, false);
1623 redirty_page_for_writepage(wbc, page);
1624 return AOP_WRITEPAGE_ACTIVATE;
1627 int f2fs_move_node_page(struct page *node_page, int gc_type)
1631 if (gc_type == FG_GC) {
1632 struct writeback_control wbc = {
1633 .sync_mode = WB_SYNC_ALL,
1638 f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1640 set_page_dirty(node_page);
1642 if (!clear_page_dirty_for_io(node_page)) {
1647 if (__write_node_page(node_page, false, NULL,
1648 &wbc, false, FS_GC_NODE_IO, NULL)) {
1650 unlock_page(node_page);
1654 /* set page dirty and write it */
1655 if (!PageWriteback(node_page))
1656 set_page_dirty(node_page);
1659 unlock_page(node_page);
1661 f2fs_put_page(node_page, 0);
1665 static int f2fs_write_node_page(struct page *page,
1666 struct writeback_control *wbc)
1668 return __write_node_page(page, false, NULL, wbc, false,
1672 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1673 struct writeback_control *wbc, bool atomic,
1674 unsigned int *seq_id)
1677 struct pagevec pvec;
1679 struct page *last_page = NULL;
1680 bool marked = false;
1681 nid_t ino = inode->i_ino;
1686 last_page = last_fsync_dnode(sbi, ino);
1687 if (IS_ERR_OR_NULL(last_page))
1688 return PTR_ERR_OR_ZERO(last_page);
1691 pagevec_init(&pvec);
1694 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1695 PAGECACHE_TAG_DIRTY))) {
1698 for (i = 0; i < nr_pages; i++) {
1699 struct page *page = pvec.pages[i];
1700 bool submitted = false;
1702 if (unlikely(f2fs_cp_error(sbi))) {
1703 f2fs_put_page(last_page, 0);
1704 pagevec_release(&pvec);
1709 if (!IS_DNODE(page) || !is_cold_node(page))
1711 if (ino_of_node(page) != ino)
1716 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1721 if (ino_of_node(page) != ino)
1722 goto continue_unlock;
1724 if (!PageDirty(page) && page != last_page) {
1725 /* someone wrote it for us */
1726 goto continue_unlock;
1729 f2fs_wait_on_page_writeback(page, NODE, true, true);
1731 set_fsync_mark(page, 0);
1732 set_dentry_mark(page, 0);
1734 if (!atomic || page == last_page) {
1735 set_fsync_mark(page, 1);
1736 if (IS_INODE(page)) {
1737 if (is_inode_flag_set(inode,
1739 f2fs_update_inode(inode, page);
1740 set_dentry_mark(page,
1741 f2fs_need_dentry_mark(sbi, ino));
1743 /* may be written by other thread */
1744 if (!PageDirty(page))
1745 set_page_dirty(page);
1748 if (!clear_page_dirty_for_io(page))
1749 goto continue_unlock;
1751 ret = __write_node_page(page, atomic &&
1753 &submitted, wbc, true,
1754 FS_NODE_IO, seq_id);
1757 f2fs_put_page(last_page, 0);
1759 } else if (submitted) {
1763 if (page == last_page) {
1764 f2fs_put_page(page, 0);
1769 pagevec_release(&pvec);
1775 if (!ret && atomic && !marked) {
1776 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1777 ino, last_page->index);
1778 lock_page(last_page);
1779 f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1780 set_page_dirty(last_page);
1781 unlock_page(last_page);
1786 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1787 return ret ? -EIO : 0;
1790 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1792 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1795 if (inode->i_ino != ino)
1798 if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1801 spin_lock(&sbi->inode_lock[DIRTY_META]);
1802 clean = list_empty(&F2FS_I(inode)->gdirty_list);
1803 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1808 inode = igrab(inode);
1814 static bool flush_dirty_inode(struct page *page)
1816 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1817 struct inode *inode;
1818 nid_t ino = ino_of_node(page);
1820 inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1824 f2fs_update_inode(inode, page);
1831 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1834 struct pagevec pvec;
1837 pagevec_init(&pvec);
1839 while ((nr_pages = pagevec_lookup_tag(&pvec,
1840 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1843 for (i = 0; i < nr_pages; i++) {
1844 struct page *page = pvec.pages[i];
1846 if (!IS_DNODE(page))
1851 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1857 if (!PageDirty(page)) {
1858 /* someone wrote it for us */
1859 goto continue_unlock;
1862 /* flush inline_data, if it's async context. */
1863 if (is_inline_node(page)) {
1864 clear_inline_node(page);
1866 flush_inline_data(sbi, ino_of_node(page));
1871 pagevec_release(&pvec);
1876 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1877 struct writeback_control *wbc,
1878 bool do_balance, enum iostat_type io_type)
1881 struct pagevec pvec;
1885 int nr_pages, done = 0;
1887 pagevec_init(&pvec);
1892 while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1893 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1896 for (i = 0; i < nr_pages; i++) {
1897 struct page *page = pvec.pages[i];
1898 bool submitted = false;
1899 bool may_dirty = true;
1901 /* give a priority to WB_SYNC threads */
1902 if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1903 wbc->sync_mode == WB_SYNC_NONE) {
1909 * flushing sequence with step:
1914 if (step == 0 && IS_DNODE(page))
1916 if (step == 1 && (!IS_DNODE(page) ||
1917 is_cold_node(page)))
1919 if (step == 2 && (!IS_DNODE(page) ||
1920 !is_cold_node(page)))
1923 if (wbc->sync_mode == WB_SYNC_ALL)
1925 else if (!trylock_page(page))
1928 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1934 if (!PageDirty(page)) {
1935 /* someone wrote it for us */
1936 goto continue_unlock;
1939 /* flush inline_data/inode, if it's async context. */
1943 /* flush inline_data */
1944 if (is_inline_node(page)) {
1945 clear_inline_node(page);
1947 flush_inline_data(sbi, ino_of_node(page));
1951 /* flush dirty inode */
1952 if (IS_INODE(page) && may_dirty) {
1954 if (flush_dirty_inode(page))
1958 f2fs_wait_on_page_writeback(page, NODE, true, true);
1960 if (!clear_page_dirty_for_io(page))
1961 goto continue_unlock;
1963 set_fsync_mark(page, 0);
1964 set_dentry_mark(page, 0);
1966 ret = __write_node_page(page, false, &submitted,
1967 wbc, do_balance, io_type, NULL);
1973 if (--wbc->nr_to_write == 0)
1976 pagevec_release(&pvec);
1979 if (wbc->nr_to_write == 0) {
1986 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1987 wbc->sync_mode == WB_SYNC_NONE && step == 1)
1994 f2fs_submit_merged_write(sbi, NODE);
1996 if (unlikely(f2fs_cp_error(sbi)))
2001 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2002 unsigned int seq_id)
2004 struct fsync_node_entry *fn;
2006 struct list_head *head = &sbi->fsync_node_list;
2007 unsigned long flags;
2008 unsigned int cur_seq_id = 0;
2011 while (seq_id && cur_seq_id < seq_id) {
2012 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2013 if (list_empty(head)) {
2014 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2017 fn = list_first_entry(head, struct fsync_node_entry, list);
2018 if (fn->seq_id > seq_id) {
2019 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2022 cur_seq_id = fn->seq_id;
2025 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2027 f2fs_wait_on_page_writeback(page, NODE, true, false);
2028 if (TestClearPageError(page))
2037 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2044 static int f2fs_write_node_pages(struct address_space *mapping,
2045 struct writeback_control *wbc)
2047 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2048 struct blk_plug plug;
2051 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2054 /* balancing f2fs's metadata in background */
2055 f2fs_balance_fs_bg(sbi, true);
2057 /* collect a number of dirty node pages and write together */
2058 if (wbc->sync_mode != WB_SYNC_ALL &&
2059 get_pages(sbi, F2FS_DIRTY_NODES) <
2060 nr_pages_to_skip(sbi, NODE))
2063 if (wbc->sync_mode == WB_SYNC_ALL)
2064 atomic_inc(&sbi->wb_sync_req[NODE]);
2065 else if (atomic_read(&sbi->wb_sync_req[NODE]))
2068 trace_f2fs_writepages(mapping->host, wbc, NODE);
2070 diff = nr_pages_to_write(sbi, NODE, wbc);
2071 blk_start_plug(&plug);
2072 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2073 blk_finish_plug(&plug);
2074 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2076 if (wbc->sync_mode == WB_SYNC_ALL)
2077 atomic_dec(&sbi->wb_sync_req[NODE]);
2081 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2082 trace_f2fs_writepages(mapping->host, wbc, NODE);
2086 static int f2fs_set_node_page_dirty(struct page *page)
2088 trace_f2fs_set_page_dirty(page, NODE);
2090 if (!PageUptodate(page))
2091 SetPageUptodate(page);
2092 #ifdef CONFIG_F2FS_CHECK_FS
2094 f2fs_inode_chksum_set(F2FS_P_SB(page), page);
2096 if (!PageDirty(page)) {
2097 __set_page_dirty_nobuffers(page);
2098 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
2099 f2fs_set_page_private(page, 0);
2106 * Structure of the f2fs node operations
2108 const struct address_space_operations f2fs_node_aops = {
2109 .writepage = f2fs_write_node_page,
2110 .writepages = f2fs_write_node_pages,
2111 .set_page_dirty = f2fs_set_node_page_dirty,
2112 .invalidatepage = f2fs_invalidate_page,
2113 .releasepage = f2fs_release_page,
2114 #ifdef CONFIG_MIGRATION
2115 .migratepage = f2fs_migrate_page,
2119 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2122 return radix_tree_lookup(&nm_i->free_nid_root, n);
2125 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2128 struct f2fs_nm_info *nm_i = NM_I(sbi);
2129 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2134 nm_i->nid_cnt[FREE_NID]++;
2135 list_add_tail(&i->list, &nm_i->free_nid_list);
2139 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2140 struct free_nid *i, enum nid_state state)
2142 struct f2fs_nm_info *nm_i = NM_I(sbi);
2144 f2fs_bug_on(sbi, state != i->state);
2145 nm_i->nid_cnt[state]--;
2146 if (state == FREE_NID)
2148 radix_tree_delete(&nm_i->free_nid_root, i->nid);
2151 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2152 enum nid_state org_state, enum nid_state dst_state)
2154 struct f2fs_nm_info *nm_i = NM_I(sbi);
2156 f2fs_bug_on(sbi, org_state != i->state);
2157 i->state = dst_state;
2158 nm_i->nid_cnt[org_state]--;
2159 nm_i->nid_cnt[dst_state]++;
2161 switch (dst_state) {
2166 list_add_tail(&i->list, &nm_i->free_nid_list);
2173 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2174 bool set, bool build)
2176 struct f2fs_nm_info *nm_i = NM_I(sbi);
2177 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2178 unsigned int nid_ofs = nid - START_NID(nid);
2180 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2184 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2186 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2187 nm_i->free_nid_count[nat_ofs]++;
2189 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2191 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2193 nm_i->free_nid_count[nat_ofs]--;
2197 /* return if the nid is recognized as free */
2198 static bool add_free_nid(struct f2fs_sb_info *sbi,
2199 nid_t nid, bool build, bool update)
2201 struct f2fs_nm_info *nm_i = NM_I(sbi);
2202 struct free_nid *i, *e;
2203 struct nat_entry *ne;
2207 /* 0 nid should not be used */
2208 if (unlikely(nid == 0))
2211 if (unlikely(f2fs_check_nid_range(sbi, nid)))
2214 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2216 i->state = FREE_NID;
2218 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2220 spin_lock(&nm_i->nid_list_lock);
2228 * - __insert_nid_to_list(PREALLOC_NID)
2229 * - f2fs_balance_fs_bg
2230 * - f2fs_build_free_nids
2231 * - __f2fs_build_free_nids
2234 * - __lookup_nat_cache
2236 * - f2fs_init_inode_metadata
2237 * - f2fs_new_inode_page
2238 * - f2fs_new_node_page
2240 * - f2fs_alloc_nid_done
2241 * - __remove_nid_from_list(PREALLOC_NID)
2242 * - __insert_nid_to_list(FREE_NID)
2244 ne = __lookup_nat_cache(nm_i, nid);
2245 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2246 nat_get_blkaddr(ne) != NULL_ADDR))
2249 e = __lookup_free_nid_list(nm_i, nid);
2251 if (e->state == FREE_NID)
2257 err = __insert_free_nid(sbi, i);
2260 update_free_nid_bitmap(sbi, nid, ret, build);
2262 nm_i->available_nids++;
2264 spin_unlock(&nm_i->nid_list_lock);
2265 radix_tree_preload_end();
2268 kmem_cache_free(free_nid_slab, i);
2272 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2274 struct f2fs_nm_info *nm_i = NM_I(sbi);
2276 bool need_free = false;
2278 spin_lock(&nm_i->nid_list_lock);
2279 i = __lookup_free_nid_list(nm_i, nid);
2280 if (i && i->state == FREE_NID) {
2281 __remove_free_nid(sbi, i, FREE_NID);
2284 spin_unlock(&nm_i->nid_list_lock);
2287 kmem_cache_free(free_nid_slab, i);
2290 static int scan_nat_page(struct f2fs_sb_info *sbi,
2291 struct page *nat_page, nid_t start_nid)
2293 struct f2fs_nm_info *nm_i = NM_I(sbi);
2294 struct f2fs_nat_block *nat_blk = page_address(nat_page);
2296 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2299 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2301 i = start_nid % NAT_ENTRY_PER_BLOCK;
2303 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2304 if (unlikely(start_nid >= nm_i->max_nid))
2307 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2309 if (blk_addr == NEW_ADDR)
2312 if (blk_addr == NULL_ADDR) {
2313 add_free_nid(sbi, start_nid, true, true);
2315 spin_lock(&NM_I(sbi)->nid_list_lock);
2316 update_free_nid_bitmap(sbi, start_nid, false, true);
2317 spin_unlock(&NM_I(sbi)->nid_list_lock);
2324 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2326 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2327 struct f2fs_journal *journal = curseg->journal;
2330 down_read(&curseg->journal_rwsem);
2331 for (i = 0; i < nats_in_cursum(journal); i++) {
2335 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2336 nid = le32_to_cpu(nid_in_journal(journal, i));
2337 if (addr == NULL_ADDR)
2338 add_free_nid(sbi, nid, true, false);
2340 remove_free_nid(sbi, nid);
2342 up_read(&curseg->journal_rwsem);
2345 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2347 struct f2fs_nm_info *nm_i = NM_I(sbi);
2348 unsigned int i, idx;
2351 down_read(&nm_i->nat_tree_lock);
2353 for (i = 0; i < nm_i->nat_blocks; i++) {
2354 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2356 if (!nm_i->free_nid_count[i])
2358 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2359 idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2360 NAT_ENTRY_PER_BLOCK, idx);
2361 if (idx >= NAT_ENTRY_PER_BLOCK)
2364 nid = i * NAT_ENTRY_PER_BLOCK + idx;
2365 add_free_nid(sbi, nid, true, false);
2367 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2372 scan_curseg_cache(sbi);
2374 up_read(&nm_i->nat_tree_lock);
2377 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2378 bool sync, bool mount)
2380 struct f2fs_nm_info *nm_i = NM_I(sbi);
2382 nid_t nid = nm_i->next_scan_nid;
2384 if (unlikely(nid >= nm_i->max_nid))
2387 if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2388 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2390 /* Enough entries */
2391 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2394 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2398 /* try to find free nids in free_nid_bitmap */
2399 scan_free_nid_bits(sbi);
2401 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2405 /* readahead nat pages to be scanned */
2406 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2409 down_read(&nm_i->nat_tree_lock);
2412 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2413 nm_i->nat_block_bitmap)) {
2414 struct page *page = get_current_nat_page(sbi, nid);
2417 ret = PTR_ERR(page);
2419 ret = scan_nat_page(sbi, page, nid);
2420 f2fs_put_page(page, 1);
2424 up_read(&nm_i->nat_tree_lock);
2425 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2430 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2431 if (unlikely(nid >= nm_i->max_nid))
2434 if (++i >= FREE_NID_PAGES)
2438 /* go to the next free nat pages to find free nids abundantly */
2439 nm_i->next_scan_nid = nid;
2441 /* find free nids from current sum_pages */
2442 scan_curseg_cache(sbi);
2444 up_read(&nm_i->nat_tree_lock);
2446 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2447 nm_i->ra_nid_pages, META_NAT, false);
2452 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2456 mutex_lock(&NM_I(sbi)->build_lock);
2457 ret = __f2fs_build_free_nids(sbi, sync, mount);
2458 mutex_unlock(&NM_I(sbi)->build_lock);
2464 * If this function returns success, caller can obtain a new nid
2465 * from second parameter of this function.
2466 * The returned nid could be used ino as well as nid when inode is created.
2468 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2470 struct f2fs_nm_info *nm_i = NM_I(sbi);
2471 struct free_nid *i = NULL;
2473 if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2474 f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2478 spin_lock(&nm_i->nid_list_lock);
2480 if (unlikely(nm_i->available_nids == 0)) {
2481 spin_unlock(&nm_i->nid_list_lock);
2485 /* We should not use stale free nids created by f2fs_build_free_nids */
2486 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2487 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2488 i = list_first_entry(&nm_i->free_nid_list,
2489 struct free_nid, list);
2492 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2493 nm_i->available_nids--;
2495 update_free_nid_bitmap(sbi, *nid, false, false);
2497 spin_unlock(&nm_i->nid_list_lock);
2500 spin_unlock(&nm_i->nid_list_lock);
2502 /* Let's scan nat pages and its caches to get free nids */
2503 if (!f2fs_build_free_nids(sbi, true, false))
2509 * f2fs_alloc_nid() should be called prior to this function.
2511 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2513 struct f2fs_nm_info *nm_i = NM_I(sbi);
2516 spin_lock(&nm_i->nid_list_lock);
2517 i = __lookup_free_nid_list(nm_i, nid);
2518 f2fs_bug_on(sbi, !i);
2519 __remove_free_nid(sbi, i, PREALLOC_NID);
2520 spin_unlock(&nm_i->nid_list_lock);
2522 kmem_cache_free(free_nid_slab, i);
2526 * f2fs_alloc_nid() should be called prior to this function.
2528 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2530 struct f2fs_nm_info *nm_i = NM_I(sbi);
2532 bool need_free = false;
2537 spin_lock(&nm_i->nid_list_lock);
2538 i = __lookup_free_nid_list(nm_i, nid);
2539 f2fs_bug_on(sbi, !i);
2541 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2542 __remove_free_nid(sbi, i, PREALLOC_NID);
2545 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2548 nm_i->available_nids++;
2550 update_free_nid_bitmap(sbi, nid, true, false);
2552 spin_unlock(&nm_i->nid_list_lock);
2555 kmem_cache_free(free_nid_slab, i);
2558 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2560 struct f2fs_nm_info *nm_i = NM_I(sbi);
2563 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2566 if (!mutex_trylock(&nm_i->build_lock))
2569 while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2570 struct free_nid *i, *next;
2571 unsigned int batch = SHRINK_NID_BATCH_SIZE;
2573 spin_lock(&nm_i->nid_list_lock);
2574 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2575 if (!nr_shrink || !batch ||
2576 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2578 __remove_free_nid(sbi, i, FREE_NID);
2579 kmem_cache_free(free_nid_slab, i);
2583 spin_unlock(&nm_i->nid_list_lock);
2586 mutex_unlock(&nm_i->build_lock);
2588 return nr - nr_shrink;
2591 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2593 void *src_addr, *dst_addr;
2596 struct f2fs_inode *ri;
2598 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2600 return PTR_ERR(ipage);
2602 ri = F2FS_INODE(page);
2603 if (ri->i_inline & F2FS_INLINE_XATTR) {
2604 if (!f2fs_has_inline_xattr(inode)) {
2605 set_inode_flag(inode, FI_INLINE_XATTR);
2606 stat_inc_inline_xattr(inode);
2609 if (f2fs_has_inline_xattr(inode)) {
2610 stat_dec_inline_xattr(inode);
2611 clear_inode_flag(inode, FI_INLINE_XATTR);
2616 dst_addr = inline_xattr_addr(inode, ipage);
2617 src_addr = inline_xattr_addr(inode, page);
2618 inline_size = inline_xattr_size(inode);
2620 f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2621 memcpy(dst_addr, src_addr, inline_size);
2623 f2fs_update_inode(inode, ipage);
2624 f2fs_put_page(ipage, 1);
2628 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2630 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2631 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2633 struct dnode_of_data dn;
2634 struct node_info ni;
2641 /* 1: invalidate the previous xattr nid */
2642 err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2646 f2fs_invalidate_blocks(sbi, ni.blk_addr);
2647 dec_valid_node_count(sbi, inode, false);
2648 set_node_addr(sbi, &ni, NULL_ADDR, false);
2651 /* 2: update xattr nid in inode */
2652 if (!f2fs_alloc_nid(sbi, &new_xnid))
2655 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2656 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2657 if (IS_ERR(xpage)) {
2658 f2fs_alloc_nid_failed(sbi, new_xnid);
2659 return PTR_ERR(xpage);
2662 f2fs_alloc_nid_done(sbi, new_xnid);
2663 f2fs_update_inode_page(inode);
2665 /* 3: update and set xattr node page dirty */
2666 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2668 set_page_dirty(xpage);
2669 f2fs_put_page(xpage, 1);
2674 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2676 struct f2fs_inode *src, *dst;
2677 nid_t ino = ino_of_node(page);
2678 struct node_info old_ni, new_ni;
2682 err = f2fs_get_node_info(sbi, ino, &old_ni);
2686 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2689 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2691 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2695 /* Should not use this inode from free nid list */
2696 remove_free_nid(sbi, ino);
2698 if (!PageUptodate(ipage))
2699 SetPageUptodate(ipage);
2700 fill_node_footer(ipage, ino, ino, 0, true);
2701 set_cold_node(ipage, false);
2703 src = F2FS_INODE(page);
2704 dst = F2FS_INODE(ipage);
2706 memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2708 dst->i_blocks = cpu_to_le64(1);
2709 dst->i_links = cpu_to_le32(1);
2710 dst->i_xattr_nid = 0;
2711 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2712 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2713 dst->i_extra_isize = src->i_extra_isize;
2715 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2716 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2717 i_inline_xattr_size))
2718 dst->i_inline_xattr_size = src->i_inline_xattr_size;
2720 if (f2fs_sb_has_project_quota(sbi) &&
2721 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2723 dst->i_projid = src->i_projid;
2725 if (f2fs_sb_has_inode_crtime(sbi) &&
2726 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2728 dst->i_crtime = src->i_crtime;
2729 dst->i_crtime_nsec = src->i_crtime_nsec;
2736 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2738 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2739 inc_valid_inode_count(sbi);
2740 set_page_dirty(ipage);
2741 f2fs_put_page(ipage, 1);
2745 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2746 unsigned int segno, struct f2fs_summary_block *sum)
2748 struct f2fs_node *rn;
2749 struct f2fs_summary *sum_entry;
2751 int i, idx, last_offset, nrpages;
2753 /* scan the node segment */
2754 last_offset = sbi->blocks_per_seg;
2755 addr = START_BLOCK(sbi, segno);
2756 sum_entry = &sum->entries[0];
2758 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2759 nrpages = bio_max_segs(last_offset - i);
2761 /* readahead node pages */
2762 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2764 for (idx = addr; idx < addr + nrpages; idx++) {
2765 struct page *page = f2fs_get_tmp_page(sbi, idx);
2768 return PTR_ERR(page);
2770 rn = F2FS_NODE(page);
2771 sum_entry->nid = rn->footer.nid;
2772 sum_entry->version = 0;
2773 sum_entry->ofs_in_node = 0;
2775 f2fs_put_page(page, 1);
2778 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2784 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2786 struct f2fs_nm_info *nm_i = NM_I(sbi);
2787 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2788 struct f2fs_journal *journal = curseg->journal;
2791 down_write(&curseg->journal_rwsem);
2792 for (i = 0; i < nats_in_cursum(journal); i++) {
2793 struct nat_entry *ne;
2794 struct f2fs_nat_entry raw_ne;
2795 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2797 if (f2fs_check_nid_range(sbi, nid))
2800 raw_ne = nat_in_journal(journal, i);
2802 ne = __lookup_nat_cache(nm_i, nid);
2804 ne = __alloc_nat_entry(nid, true);
2805 __init_nat_entry(nm_i, ne, &raw_ne, true);
2809 * if a free nat in journal has not been used after last
2810 * checkpoint, we should remove it from available nids,
2811 * since later we will add it again.
2813 if (!get_nat_flag(ne, IS_DIRTY) &&
2814 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2815 spin_lock(&nm_i->nid_list_lock);
2816 nm_i->available_nids--;
2817 spin_unlock(&nm_i->nid_list_lock);
2820 __set_nat_cache_dirty(nm_i, ne);
2822 update_nats_in_cursum(journal, -i);
2823 up_write(&curseg->journal_rwsem);
2826 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2827 struct list_head *head, int max)
2829 struct nat_entry_set *cur;
2831 if (nes->entry_cnt >= max)
2834 list_for_each_entry(cur, head, set_list) {
2835 if (cur->entry_cnt >= nes->entry_cnt) {
2836 list_add(&nes->set_list, cur->set_list.prev);
2841 list_add_tail(&nes->set_list, head);
2844 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2847 struct f2fs_nm_info *nm_i = NM_I(sbi);
2848 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2849 struct f2fs_nat_block *nat_blk = page_address(page);
2853 if (!enabled_nat_bits(sbi, NULL))
2856 if (nat_index == 0) {
2860 for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2861 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2865 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2866 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2870 __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2871 if (valid == NAT_ENTRY_PER_BLOCK)
2872 __set_bit_le(nat_index, nm_i->full_nat_bits);
2874 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2877 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2878 struct nat_entry_set *set, struct cp_control *cpc)
2880 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2881 struct f2fs_journal *journal = curseg->journal;
2882 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2883 bool to_journal = true;
2884 struct f2fs_nat_block *nat_blk;
2885 struct nat_entry *ne, *cur;
2886 struct page *page = NULL;
2889 * there are two steps to flush nat entries:
2890 * #1, flush nat entries to journal in current hot data summary block.
2891 * #2, flush nat entries to nat page.
2893 if (enabled_nat_bits(sbi, cpc) ||
2894 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2898 down_write(&curseg->journal_rwsem);
2900 page = get_next_nat_page(sbi, start_nid);
2902 return PTR_ERR(page);
2904 nat_blk = page_address(page);
2905 f2fs_bug_on(sbi, !nat_blk);
2908 /* flush dirty nats in nat entry set */
2909 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2910 struct f2fs_nat_entry *raw_ne;
2911 nid_t nid = nat_get_nid(ne);
2914 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2917 offset = f2fs_lookup_journal_in_cursum(journal,
2918 NAT_JOURNAL, nid, 1);
2919 f2fs_bug_on(sbi, offset < 0);
2920 raw_ne = &nat_in_journal(journal, offset);
2921 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2923 raw_ne = &nat_blk->entries[nid - start_nid];
2925 raw_nat_from_node_info(raw_ne, &ne->ni);
2927 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2928 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2929 add_free_nid(sbi, nid, false, true);
2931 spin_lock(&NM_I(sbi)->nid_list_lock);
2932 update_free_nid_bitmap(sbi, nid, false, false);
2933 spin_unlock(&NM_I(sbi)->nid_list_lock);
2938 up_write(&curseg->journal_rwsem);
2940 __update_nat_bits(sbi, start_nid, page);
2941 f2fs_put_page(page, 1);
2944 /* Allow dirty nats by node block allocation in write_begin */
2945 if (!set->entry_cnt) {
2946 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2947 kmem_cache_free(nat_entry_set_slab, set);
2953 * This function is called during the checkpointing process.
2955 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2957 struct f2fs_nm_info *nm_i = NM_I(sbi);
2958 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2959 struct f2fs_journal *journal = curseg->journal;
2960 struct nat_entry_set *setvec[SETVEC_SIZE];
2961 struct nat_entry_set *set, *tmp;
2968 * during unmount, let's flush nat_bits before checking
2969 * nat_cnt[DIRTY_NAT].
2971 if (enabled_nat_bits(sbi, cpc)) {
2972 down_write(&nm_i->nat_tree_lock);
2973 remove_nats_in_journal(sbi);
2974 up_write(&nm_i->nat_tree_lock);
2977 if (!nm_i->nat_cnt[DIRTY_NAT])
2980 down_write(&nm_i->nat_tree_lock);
2983 * if there are no enough space in journal to store dirty nat
2984 * entries, remove all entries from journal and merge them
2985 * into nat entry set.
2987 if (enabled_nat_bits(sbi, cpc) ||
2988 !__has_cursum_space(journal,
2989 nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
2990 remove_nats_in_journal(sbi);
2992 while ((found = __gang_lookup_nat_set(nm_i,
2993 set_idx, SETVEC_SIZE, setvec))) {
2996 set_idx = setvec[found - 1]->set + 1;
2997 for (idx = 0; idx < found; idx++)
2998 __adjust_nat_entry_set(setvec[idx], &sets,
2999 MAX_NAT_JENTRIES(journal));
3002 /* flush dirty nats in nat entry set */
3003 list_for_each_entry_safe(set, tmp, &sets, set_list) {
3004 err = __flush_nat_entry_set(sbi, set, cpc);
3009 up_write(&nm_i->nat_tree_lock);
3010 /* Allow dirty nats by node block allocation in write_begin */
3015 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3017 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3018 struct f2fs_nm_info *nm_i = NM_I(sbi);
3019 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3021 __u64 cp_ver = cur_cp_version(ckpt);
3022 block_t nat_bits_addr;
3024 if (!enabled_nat_bits(sbi, NULL))
3027 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3028 nm_i->nat_bits = f2fs_kvzalloc(sbi,
3029 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3030 if (!nm_i->nat_bits)
3033 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3034 nm_i->nat_bits_blocks;
3035 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3038 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3040 return PTR_ERR(page);
3042 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3043 page_address(page), F2FS_BLKSIZE);
3044 f2fs_put_page(page, 1);
3047 cp_ver |= (cur_cp_crc(ckpt) << 32);
3048 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3049 disable_nat_bits(sbi, true);
3053 nm_i->full_nat_bits = nm_i->nat_bits + 8;
3054 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3056 f2fs_notice(sbi, "Found nat_bits in checkpoint");
3060 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3062 struct f2fs_nm_info *nm_i = NM_I(sbi);
3064 nid_t nid, last_nid;
3066 if (!enabled_nat_bits(sbi, NULL))
3069 for (i = 0; i < nm_i->nat_blocks; i++) {
3070 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3071 if (i >= nm_i->nat_blocks)
3074 __set_bit_le(i, nm_i->nat_block_bitmap);
3076 nid = i * NAT_ENTRY_PER_BLOCK;
3077 last_nid = nid + NAT_ENTRY_PER_BLOCK;
3079 spin_lock(&NM_I(sbi)->nid_list_lock);
3080 for (; nid < last_nid; nid++)
3081 update_free_nid_bitmap(sbi, nid, true, true);
3082 spin_unlock(&NM_I(sbi)->nid_list_lock);
3085 for (i = 0; i < nm_i->nat_blocks; i++) {
3086 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3087 if (i >= nm_i->nat_blocks)
3090 __set_bit_le(i, nm_i->nat_block_bitmap);
3094 static int init_node_manager(struct f2fs_sb_info *sbi)
3096 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3097 struct f2fs_nm_info *nm_i = NM_I(sbi);
3098 unsigned char *version_bitmap;
3099 unsigned int nat_segs;
3102 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3104 /* segment_count_nat includes pair segment so divide to 2. */
3105 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3106 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3107 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3109 /* not used nids: 0, node, meta, (and root counted as valid node) */
3110 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3111 F2FS_RESERVED_NODE_NUM;
3112 nm_i->nid_cnt[FREE_NID] = 0;
3113 nm_i->nid_cnt[PREALLOC_NID] = 0;
3114 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3115 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3116 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3118 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3119 INIT_LIST_HEAD(&nm_i->free_nid_list);
3120 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3121 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3122 INIT_LIST_HEAD(&nm_i->nat_entries);
3123 spin_lock_init(&nm_i->nat_list_lock);
3125 mutex_init(&nm_i->build_lock);
3126 spin_lock_init(&nm_i->nid_list_lock);
3127 init_rwsem(&nm_i->nat_tree_lock);
3129 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3130 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3131 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3132 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3134 if (!nm_i->nat_bitmap)
3137 err = __get_nat_bitmaps(sbi);
3141 #ifdef CONFIG_F2FS_CHECK_FS
3142 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3144 if (!nm_i->nat_bitmap_mir)
3151 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3153 struct f2fs_nm_info *nm_i = NM_I(sbi);
3156 nm_i->free_nid_bitmap =
3157 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3160 if (!nm_i->free_nid_bitmap)
3163 for (i = 0; i < nm_i->nat_blocks; i++) {
3164 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3165 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3166 if (!nm_i->free_nid_bitmap[i])
3170 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3172 if (!nm_i->nat_block_bitmap)
3175 nm_i->free_nid_count =
3176 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3179 if (!nm_i->free_nid_count)
3184 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3188 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3193 err = init_node_manager(sbi);
3197 err = init_free_nid_cache(sbi);
3201 /* load free nid status from nat_bits table */
3202 load_free_nid_bitmap(sbi);
3204 return f2fs_build_free_nids(sbi, true, true);
3207 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3209 struct f2fs_nm_info *nm_i = NM_I(sbi);
3210 struct free_nid *i, *next_i;
3211 struct nat_entry *natvec[NATVEC_SIZE];
3212 struct nat_entry_set *setvec[SETVEC_SIZE];
3219 /* destroy free nid list */
3220 spin_lock(&nm_i->nid_list_lock);
3221 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3222 __remove_free_nid(sbi, i, FREE_NID);
3223 spin_unlock(&nm_i->nid_list_lock);
3224 kmem_cache_free(free_nid_slab, i);
3225 spin_lock(&nm_i->nid_list_lock);
3227 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3228 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3229 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3230 spin_unlock(&nm_i->nid_list_lock);
3232 /* destroy nat cache */
3233 down_write(&nm_i->nat_tree_lock);
3234 while ((found = __gang_lookup_nat_cache(nm_i,
3235 nid, NATVEC_SIZE, natvec))) {
3238 nid = nat_get_nid(natvec[found - 1]) + 1;
3239 for (idx = 0; idx < found; idx++) {
3240 spin_lock(&nm_i->nat_list_lock);
3241 list_del(&natvec[idx]->list);
3242 spin_unlock(&nm_i->nat_list_lock);
3244 __del_from_nat_cache(nm_i, natvec[idx]);
3247 f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3249 /* destroy nat set cache */
3251 while ((found = __gang_lookup_nat_set(nm_i,
3252 nid, SETVEC_SIZE, setvec))) {
3255 nid = setvec[found - 1]->set + 1;
3256 for (idx = 0; idx < found; idx++) {
3257 /* entry_cnt is not zero, when cp_error was occurred */
3258 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3259 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3260 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3263 up_write(&nm_i->nat_tree_lock);
3265 kvfree(nm_i->nat_block_bitmap);
3266 if (nm_i->free_nid_bitmap) {
3269 for (i = 0; i < nm_i->nat_blocks; i++)
3270 kvfree(nm_i->free_nid_bitmap[i]);
3271 kvfree(nm_i->free_nid_bitmap);
3273 kvfree(nm_i->free_nid_count);
3275 kvfree(nm_i->nat_bitmap);
3276 kvfree(nm_i->nat_bits);
3277 #ifdef CONFIG_F2FS_CHECK_FS
3278 kvfree(nm_i->nat_bitmap_mir);
3280 sbi->nm_info = NULL;
3284 int __init f2fs_create_node_manager_caches(void)
3286 nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3287 sizeof(struct nat_entry));
3288 if (!nat_entry_slab)
3291 free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3292 sizeof(struct free_nid));
3294 goto destroy_nat_entry;
3296 nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3297 sizeof(struct nat_entry_set));
3298 if (!nat_entry_set_slab)
3299 goto destroy_free_nid;
3301 fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3302 sizeof(struct fsync_node_entry));
3303 if (!fsync_node_entry_slab)
3304 goto destroy_nat_entry_set;
3307 destroy_nat_entry_set:
3308 kmem_cache_destroy(nat_entry_set_slab);
3310 kmem_cache_destroy(free_nid_slab);
3312 kmem_cache_destroy(nat_entry_slab);
3317 void f2fs_destroy_node_manager_caches(void)
3319 kmem_cache_destroy(fsync_node_entry_slab);
3320 kmem_cache_destroy(nat_entry_set_slab);
3321 kmem_cache_destroy(free_nid_slab);
3322 kmem_cache_destroy(nat_entry_slab);