Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * fs/f2fs/node.c
4 *
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
7 */
8#include <linux/fs.h>
9#include <linux/f2fs_fs.h>
10#include <linux/mpage.h>
11#include <linux/sched/mm.h>
12#include <linux/blkdev.h>
13#include <linux/folio_batch.h>
14#include <linux/swap.h>
15
16#include "f2fs.h"
17#include "node.h"
18#include "segment.h"
19#include "xattr.h"
20#include "iostat.h"
21#include <trace/events/f2fs.h>
22
23#define on_f2fs_build_free_nids(nm_i) mutex_is_locked(&(nm_i)->build_lock)
24
25static struct kmem_cache *nat_entry_slab;
26static struct kmem_cache *free_nid_slab;
27static struct kmem_cache *nat_entry_set_slab;
28static struct kmem_cache *fsync_node_entry_slab;
29
30static inline bool is_invalid_nid(struct f2fs_sb_info *sbi, nid_t nid)
31{
32 return nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid;
33}
34
35/*
36 * Check whether the given nid is within node id range.
37 */
38int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
39{
40 if (unlikely(is_invalid_nid(sbi, nid))) {
41 set_sbi_flag(sbi, SBI_NEED_FSCK);
42 f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
43 __func__, nid);
44 f2fs_handle_error(sbi, ERROR_CORRUPTED_INODE);
45 return -EFSCORRUPTED;
46 }
47 return 0;
48}
49
50bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
51{
52 struct f2fs_nm_info *nm_i = NM_I(sbi);
53 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
54 struct sysinfo val;
55 unsigned long avail_ram;
56 unsigned long mem_size = 0;
57 bool res = false;
58
59 if (!nm_i)
60 return true;
61
62 si_meminfo(&val);
63
64 /* only uses low memory */
65 avail_ram = val.totalram - val.totalhigh;
66
67 /*
68 * give 25%, 25%, 50%, 50%, 25%, 25% memory for each components respectively
69 */
70 if (type == FREE_NIDS) {
71 mem_size = (nm_i->nid_cnt[FREE_NID] *
72 sizeof(struct free_nid)) >> PAGE_SHIFT;
73 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
74 } else if (type == NAT_ENTRIES) {
75 mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
76 sizeof(struct nat_entry)) >> PAGE_SHIFT;
77 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
78 if (excess_cached_nats(sbi))
79 res = false;
80 } else if (type == DIRTY_DENTS) {
81 if (bdi_wb_dirty_exceeded(sbi->sb->s_bdi))
82 return false;
83 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
84 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
85 } else if (type == INO_ENTRIES) {
86 int i;
87
88 for (i = 0; i < MAX_INO_ENTRY; i++)
89 mem_size += sbi->im[i].ino_num *
90 sizeof(struct ino_entry);
91 mem_size >>= PAGE_SHIFT;
92 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
93 } else if (type == READ_EXTENT_CACHE || type == AGE_EXTENT_CACHE) {
94 enum extent_type etype = type == READ_EXTENT_CACHE ?
95 EX_READ : EX_BLOCK_AGE;
96 struct extent_tree_info *eti = &sbi->extent_tree[etype];
97
98 mem_size = (atomic_read(&eti->total_ext_tree) *
99 sizeof(struct extent_tree) +
100 atomic_read(&eti->total_ext_node) *
101 sizeof(struct extent_node)) >> PAGE_SHIFT;
102 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
103 } else if (type == DISCARD_CACHE) {
104 mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
105 sizeof(struct discard_cmd)) >> PAGE_SHIFT;
106 res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
107 } else if (type == COMPRESS_PAGE) {
108#ifdef CONFIG_F2FS_FS_COMPRESSION
109 unsigned long free_ram = val.freeram;
110
111 /*
112 * free memory is lower than watermark or cached page count
113 * exceed threshold, deny caching compress page.
114 */
115 res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
116 (COMPRESS_MAPPING(sbi)->nrpages <
117 free_ram * sbi->compress_percent / 100);
118#else
119 res = false;
120#endif
121 } else {
122 if (!bdi_wb_dirty_exceeded(sbi->sb->s_bdi))
123 return true;
124 }
125 return res;
126}
127
128static void clear_node_folio_dirty(struct folio *folio)
129{
130 if (folio_test_dirty(folio)) {
131 f2fs_clear_page_cache_dirty_tag(folio);
132 folio_clear_dirty_for_io(folio);
133 dec_page_count(F2FS_F_SB(folio), F2FS_DIRTY_NODES);
134 }
135 folio_clear_uptodate(folio);
136}
137
138static struct folio *get_current_nat_folio(struct f2fs_sb_info *sbi, nid_t nid)
139{
140 return f2fs_get_meta_folio_retry(sbi, current_nat_addr(sbi, nid));
141}
142
143static struct folio *get_next_nat_folio(struct f2fs_sb_info *sbi, nid_t nid)
144{
145 struct folio *src_folio;
146 struct folio *dst_folio;
147 pgoff_t dst_off;
148 void *src_addr;
149 void *dst_addr;
150 struct f2fs_nm_info *nm_i = NM_I(sbi);
151
152 dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
153
154 /* get current nat block page with lock */
155 src_folio = get_current_nat_folio(sbi, nid);
156 if (IS_ERR(src_folio))
157 return src_folio;
158 dst_folio = f2fs_grab_meta_folio(sbi, dst_off);
159 f2fs_bug_on(sbi, folio_test_dirty(src_folio));
160
161 src_addr = folio_address(src_folio);
162 dst_addr = folio_address(dst_folio);
163 memcpy(dst_addr, src_addr, PAGE_SIZE);
164 folio_mark_dirty(dst_folio);
165 f2fs_folio_put(src_folio, true);
166
167 set_to_next_nat(nm_i, nid);
168
169 return dst_folio;
170}
171
172static struct nat_entry *__alloc_nat_entry(struct f2fs_sb_info *sbi,
173 nid_t nid, bool no_fail)
174{
175 struct nat_entry *new;
176
177 new = f2fs_kmem_cache_alloc(nat_entry_slab,
178 GFP_F2FS_ZERO, no_fail, sbi);
179 if (new) {
180 nat_set_nid(new, nid);
181 nat_reset_flag(new);
182 }
183 return new;
184}
185
186static void __free_nat_entry(struct nat_entry *e)
187{
188 kmem_cache_free(nat_entry_slab, e);
189}
190
191/* must be locked by nat_tree_lock */
192static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
193 struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail, bool init_dirty)
194{
195 if (no_fail)
196 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
197 else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
198 return NULL;
199
200 if (raw_ne)
201 node_info_from_raw_nat(&ne->ni, raw_ne);
202
203 if (init_dirty) {
204 INIT_LIST_HEAD(&ne->list);
205 nm_i->nat_cnt[TOTAL_NAT]++;
206 return ne;
207 }
208
209 spin_lock(&nm_i->nat_list_lock);
210 list_add_tail(&ne->list, &nm_i->nat_entries);
211 spin_unlock(&nm_i->nat_list_lock);
212
213 nm_i->nat_cnt[TOTAL_NAT]++;
214 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
215 return ne;
216}
217
218static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n, bool for_dirty)
219{
220 struct nat_entry *ne;
221
222 ne = radix_tree_lookup(&nm_i->nat_root, n);
223
224 /*
225 * for recent accessed nat entry which will not be dirtied soon
226 * later, move it to tail of lru list.
227 */
228 if (ne && !get_nat_flag(ne, IS_DIRTY) && !for_dirty) {
229 spin_lock(&nm_i->nat_list_lock);
230 if (!list_empty(&ne->list))
231 list_move_tail(&ne->list, &nm_i->nat_entries);
232 spin_unlock(&nm_i->nat_list_lock);
233 }
234
235 return ne;
236}
237
238static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
239 nid_t start, unsigned int nr, struct nat_entry **ep)
240{
241 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
242}
243
244static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
245{
246 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
247 nm_i->nat_cnt[TOTAL_NAT]--;
248 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
249 __free_nat_entry(e);
250}
251
252static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
253 struct nat_entry *ne)
254{
255 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
256 struct nat_entry_set *head;
257
258 head = radix_tree_lookup(&nm_i->nat_set_root, set);
259 if (!head) {
260 head = f2fs_kmem_cache_alloc(nat_entry_set_slab,
261 GFP_NOFS, true, NULL);
262
263 INIT_LIST_HEAD(&head->entry_list);
264 INIT_LIST_HEAD(&head->set_list);
265 head->set = set;
266 head->entry_cnt = 0;
267 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
268 }
269 return head;
270}
271
272static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
273 struct nat_entry *ne, bool init_dirty)
274{
275 struct nat_entry_set *head;
276 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
277
278 if (!new_ne)
279 head = __grab_nat_entry_set(nm_i, ne);
280
281 /*
282 * update entry_cnt in below condition:
283 * 1. update NEW_ADDR to valid block address;
284 * 2. update old block address to new one;
285 */
286 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
287 !get_nat_flag(ne, IS_DIRTY)))
288 head->entry_cnt++;
289
290 set_nat_flag(ne, IS_PREALLOC, new_ne);
291
292 if (get_nat_flag(ne, IS_DIRTY))
293 goto refresh_list;
294
295 nm_i->nat_cnt[DIRTY_NAT]++;
296 if (!init_dirty)
297 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
298 set_nat_flag(ne, IS_DIRTY, true);
299refresh_list:
300 spin_lock(&nm_i->nat_list_lock);
301 if (new_ne)
302 list_del_init(&ne->list);
303 else
304 list_move_tail(&ne->list, &head->entry_list);
305 spin_unlock(&nm_i->nat_list_lock);
306}
307
308static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
309 struct nat_entry_set *set, struct nat_entry *ne)
310{
311 spin_lock(&nm_i->nat_list_lock);
312 list_move_tail(&ne->list, &nm_i->nat_entries);
313 spin_unlock(&nm_i->nat_list_lock);
314
315 set_nat_flag(ne, IS_DIRTY, false);
316 set->entry_cnt--;
317 nm_i->nat_cnt[DIRTY_NAT]--;
318 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
319}
320
321static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
322 nid_t start, unsigned int nr, struct nat_entry_set **ep)
323{
324 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
325 start, nr);
326}
327
328bool f2fs_in_warm_node_list(struct folio *folio)
329{
330 return is_node_folio(folio) && IS_DNODE(folio) && is_cold_node(folio);
331}
332
333void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
334{
335 spin_lock_init(&sbi->fsync_node_lock);
336 INIT_LIST_HEAD(&sbi->fsync_node_list);
337 sbi->fsync_seg_id = 0;
338 sbi->fsync_node_num = 0;
339}
340
341static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
342 struct folio *folio)
343{
344 struct fsync_node_entry *fn;
345 unsigned long flags;
346 unsigned int seq_id;
347
348 fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab,
349 GFP_NOFS, true, NULL);
350
351 folio_get(folio);
352 fn->folio = folio;
353 INIT_LIST_HEAD(&fn->list);
354
355 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
356 list_add_tail(&fn->list, &sbi->fsync_node_list);
357 fn->seq_id = sbi->fsync_seg_id++;
358 seq_id = fn->seq_id;
359 sbi->fsync_node_num++;
360 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
361
362 return seq_id;
363}
364
365void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct folio *folio)
366{
367 struct fsync_node_entry *fn;
368 unsigned long flags;
369
370 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
371 list_for_each_entry(fn, &sbi->fsync_node_list, list) {
372 if (fn->folio == folio) {
373 list_del(&fn->list);
374 sbi->fsync_node_num--;
375 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
376 kmem_cache_free(fsync_node_entry_slab, fn);
377 folio_put(folio);
378 return;
379 }
380 }
381 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
382 f2fs_bug_on(sbi, 1);
383}
384
385void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
386{
387 unsigned long flags;
388
389 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
390 sbi->fsync_seg_id = 0;
391 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
392}
393
394bool f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
395{
396 struct f2fs_nm_info *nm_i = NM_I(sbi);
397 struct nat_entry *e;
398 bool need = false;
399
400 f2fs_down_read(&nm_i->nat_tree_lock);
401 e = __lookup_nat_cache(nm_i, nid, false);
402 if (e) {
403 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
404 !get_nat_flag(e, HAS_FSYNCED_INODE))
405 need = true;
406 }
407 f2fs_up_read(&nm_i->nat_tree_lock);
408 return need;
409}
410
411bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
412{
413 struct f2fs_nm_info *nm_i = NM_I(sbi);
414 struct nat_entry *e;
415 bool is_cp = true;
416
417 f2fs_down_read(&nm_i->nat_tree_lock);
418 e = __lookup_nat_cache(nm_i, nid, false);
419 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
420 is_cp = false;
421 f2fs_up_read(&nm_i->nat_tree_lock);
422 return is_cp;
423}
424
425bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
426{
427 struct f2fs_nm_info *nm_i = NM_I(sbi);
428 struct nat_entry *e;
429 bool need_update = true;
430 struct f2fs_lock_context lc;
431
432 f2fs_down_read_trace(&sbi->node_write, &lc);
433 f2fs_down_read(&nm_i->nat_tree_lock);
434 e = __lookup_nat_cache(nm_i, ino, false);
435 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
436 (get_nat_flag(e, IS_CHECKPOINTED) ||
437 get_nat_flag(e, HAS_FSYNCED_INODE)))
438 need_update = false;
439 f2fs_up_read(&nm_i->nat_tree_lock);
440 f2fs_up_read_trace(&sbi->node_write, &lc);
441 return need_update;
442}
443
444/* must be locked by nat_tree_lock */
445static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
446 struct f2fs_nat_entry *ne)
447{
448 struct f2fs_nm_info *nm_i = NM_I(sbi);
449 struct nat_entry *new, *e;
450
451 /* Let's mitigate lock contention of nat_tree_lock during checkpoint */
452 if (f2fs_rwsem_is_locked(&sbi->cp_global_sem))
453 return;
454
455 new = __alloc_nat_entry(sbi, nid, false);
456 if (!new)
457 return;
458
459 f2fs_down_write(&nm_i->nat_tree_lock);
460 e = __lookup_nat_cache(nm_i, nid, false);
461 if (!e)
462 e = __init_nat_entry(nm_i, new, ne, false, false);
463 else
464 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
465 nat_get_blkaddr(e) !=
466 le32_to_cpu(ne->block_addr) ||
467 nat_get_version(e) != ne->version);
468 f2fs_up_write(&nm_i->nat_tree_lock);
469 if (e != new)
470 __free_nat_entry(new);
471}
472
473static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
474 block_t new_blkaddr, bool fsync_done)
475{
476 struct f2fs_nm_info *nm_i = NM_I(sbi);
477 struct nat_entry *e;
478 struct nat_entry *new = __alloc_nat_entry(sbi, ni->nid, true);
479 bool init_dirty = false;
480
481 f2fs_down_write(&nm_i->nat_tree_lock);
482 e = __lookup_nat_cache(nm_i, ni->nid, true);
483 if (!e) {
484 init_dirty = true;
485 e = __init_nat_entry(nm_i, new, NULL, true, true);
486 copy_node_info(&e->ni, ni);
487 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
488 } else if (new_blkaddr == NEW_ADDR) {
489 /*
490 * when nid is reallocated,
491 * previous nat entry can be remained in nat cache.
492 * So, reinitialize it with new information.
493 */
494 copy_node_info(&e->ni, ni);
495 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
496 }
497 /* let's free early to reduce memory consumption */
498 if (e != new)
499 __free_nat_entry(new);
500
501 /* sanity check */
502 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
503 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
504 new_blkaddr == NULL_ADDR);
505 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
506 new_blkaddr == NEW_ADDR);
507 f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
508 new_blkaddr == NEW_ADDR);
509
510 /* increment version no as node is removed */
511 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
512 unsigned char version = nat_get_version(e);
513
514 nat_set_version(e, inc_node_version(version));
515 }
516
517 /* change address */
518 nat_set_blkaddr(e, new_blkaddr);
519 if (!__is_valid_data_blkaddr(new_blkaddr))
520 set_nat_flag(e, IS_CHECKPOINTED, false);
521 __set_nat_cache_dirty(nm_i, e, init_dirty);
522
523 /* update fsync_mark if its inode nat entry is still alive */
524 if (ni->nid != ni->ino)
525 e = __lookup_nat_cache(nm_i, ni->ino, false);
526 if (e) {
527 if (fsync_done && ni->nid == ni->ino)
528 set_nat_flag(e, HAS_FSYNCED_INODE, true);
529 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
530 }
531 f2fs_up_write(&nm_i->nat_tree_lock);
532}
533
534int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
535{
536 struct f2fs_nm_info *nm_i = NM_I(sbi);
537 int nr = nr_shrink;
538
539 if (!f2fs_down_write_trylock(&nm_i->nat_tree_lock))
540 return 0;
541
542 spin_lock(&nm_i->nat_list_lock);
543 while (nr_shrink) {
544 struct nat_entry *ne;
545
546 if (list_empty(&nm_i->nat_entries))
547 break;
548
549 ne = list_first_entry(&nm_i->nat_entries,
550 struct nat_entry, list);
551 list_del(&ne->list);
552 spin_unlock(&nm_i->nat_list_lock);
553
554 __del_from_nat_cache(nm_i, ne);
555 nr_shrink--;
556
557 spin_lock(&nm_i->nat_list_lock);
558 }
559 spin_unlock(&nm_i->nat_list_lock);
560
561 f2fs_up_write(&nm_i->nat_tree_lock);
562 return nr - nr_shrink;
563}
564
565int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
566 struct node_info *ni, bool checkpoint_context)
567{
568 struct f2fs_nm_info *nm_i = NM_I(sbi);
569 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
570 struct f2fs_journal *journal = curseg->journal;
571 nid_t start_nid = START_NID(nid);
572 struct f2fs_nat_block *nat_blk;
573 struct folio *folio = NULL;
574 struct f2fs_nat_entry ne;
575 struct nat_entry *e;
576 pgoff_t index;
577 int i;
578 bool need_cache = true;
579
580 ni->flag = 0;
581 ni->nid = nid;
582retry:
583 /* Check nat cache */
584 f2fs_down_read(&nm_i->nat_tree_lock);
585 e = __lookup_nat_cache(nm_i, nid, false);
586 if (e) {
587 ni->ino = nat_get_ino(e);
588 ni->blk_addr = nat_get_blkaddr(e);
589 ni->version = nat_get_version(e);
590 f2fs_up_read(&nm_i->nat_tree_lock);
591 if (IS_ENABLED(CONFIG_F2FS_CHECK_FS)) {
592 need_cache = false;
593 goto sanity_check;
594 }
595 return 0;
596 }
597
598 /*
599 * Check current segment summary by trying to grab journal_rwsem first.
600 * This sem is on the critical path on the checkpoint requiring the above
601 * nat_tree_lock. Therefore, we should retry, if we failed to grab here
602 * while not bothering checkpoint.
603 */
604 if (!f2fs_rwsem_is_locked(&sbi->cp_global_sem) || checkpoint_context) {
605 down_read(&curseg->journal_rwsem);
606 } else if (f2fs_rwsem_is_contended(&nm_i->nat_tree_lock) ||
607 !down_read_trylock(&curseg->journal_rwsem)) {
608 f2fs_up_read(&nm_i->nat_tree_lock);
609 goto retry;
610 }
611
612 i = f2fs_lookup_journal_in_cursum(sbi, journal, NAT_JOURNAL, nid, 0);
613 if (i >= 0) {
614 ne = nat_in_journal(journal, i);
615 node_info_from_raw_nat(ni, &ne);
616 }
617 up_read(&curseg->journal_rwsem);
618 if (i >= 0) {
619 f2fs_up_read(&nm_i->nat_tree_lock);
620 goto sanity_check;
621 }
622
623 /* Fill node_info from nat page */
624 index = current_nat_addr(sbi, nid);
625 f2fs_up_read(&nm_i->nat_tree_lock);
626
627 folio = f2fs_get_meta_folio(sbi, index);
628 if (IS_ERR(folio))
629 return PTR_ERR(folio);
630
631 nat_blk = folio_address(folio);
632 ne = nat_blk->entries[nid - start_nid];
633 node_info_from_raw_nat(ni, &ne);
634 f2fs_folio_put(folio, true);
635sanity_check:
636 if (__is_valid_data_blkaddr(ni->blk_addr) &&
637 !f2fs_is_valid_blkaddr(sbi, ni->blk_addr,
638 DATA_GENERIC_ENHANCE)) {
639 set_sbi_flag(sbi, SBI_NEED_FSCK);
640 f2fs_err_ratelimited(sbi,
641 "f2fs_get_node_info of %pS: inconsistent nat entry, "
642 "ino:%u, nid:%u, blkaddr:%u, ver:%u, flag:%u",
643 __builtin_return_address(0),
644 ni->ino, ni->nid, ni->blk_addr, ni->version, ni->flag);
645 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NAT);
646 return -EFSCORRUPTED;
647 }
648
649 if (unlikely(f2fs_quota_file(sbi, ni->nid) &&
650 !__is_valid_data_blkaddr(ni->blk_addr))) {
651 set_sbi_flag(sbi, SBI_NEED_FSCK);
652 f2fs_err_ratelimited(sbi,
653 "f2fs_get_node_info of %pS: inconsistent nat entry from qf_ino, "
654 "ino:%u, nid:%u, blkaddr:%u, ver:%u, flag:%u",
655 __builtin_return_address(0),
656 ni->ino, ni->nid, ni->blk_addr, ni->version, ni->flag);
657 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NAT);
658 }
659
660 /* cache nat entry */
661 if (need_cache)
662 cache_nat_entry(sbi, nid, &ne);
663 return 0;
664}
665
666/*
667 * readahead MAX_RA_NODE number of node pages.
668 */
669static void f2fs_ra_node_pages(struct folio *parent, int start, int n)
670{
671 struct f2fs_sb_info *sbi = F2FS_F_SB(parent);
672 struct blk_plug plug;
673 int i, end;
674 nid_t nid;
675
676 blk_start_plug(&plug);
677
678 /* Then, try readahead for siblings of the desired node */
679 end = start + n;
680 end = min(end, (int)NIDS_PER_BLOCK);
681 for (i = start; i < end; i++) {
682 nid = get_nid(parent, i, false);
683 f2fs_ra_node_page(sbi, nid);
684 }
685
686 blk_finish_plug(&plug);
687}
688
689pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
690{
691 const long direct_index = ADDRS_PER_INODE(dn->inode);
692 const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
693 const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
694 unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
695 int cur_level = dn->cur_level;
696 int max_level = dn->max_level;
697 pgoff_t base = 0;
698
699 if (!dn->max_level)
700 return pgofs + 1;
701
702 while (max_level-- > cur_level)
703 skipped_unit *= NIDS_PER_BLOCK;
704
705 switch (dn->max_level) {
706 case 3:
707 base += 2 * indirect_blks;
708 fallthrough;
709 case 2:
710 base += 2 * direct_blks;
711 fallthrough;
712 case 1:
713 base += direct_index;
714 break;
715 default:
716 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
717 }
718
719 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
720}
721
722/*
723 * The maximum depth is four.
724 * Offset[0] will have raw inode offset.
725 */
726static int get_node_path(struct inode *inode, long block,
727 int offset[4], unsigned int noffset[4])
728{
729 const long direct_index = ADDRS_PER_INODE(inode);
730 const long direct_blks = ADDRS_PER_BLOCK(inode);
731 const long dptrs_per_blk = NIDS_PER_BLOCK;
732 const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
733 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
734 int n = 0;
735 int level = 0;
736
737 noffset[0] = 0;
738
739 if (block < direct_index) {
740 offset[n] = block;
741 goto got;
742 }
743 block -= direct_index;
744 if (block < direct_blks) {
745 offset[n++] = NODE_DIR1_BLOCK;
746 noffset[n] = 1;
747 offset[n] = block;
748 level = 1;
749 goto got;
750 }
751 block -= direct_blks;
752 if (block < direct_blks) {
753 offset[n++] = NODE_DIR2_BLOCK;
754 noffset[n] = 2;
755 offset[n] = block;
756 level = 1;
757 goto got;
758 }
759 block -= direct_blks;
760 if (block < indirect_blks) {
761 offset[n++] = NODE_IND1_BLOCK;
762 noffset[n] = 3;
763 offset[n++] = block / direct_blks;
764 noffset[n] = 4 + offset[n - 1];
765 offset[n] = block % direct_blks;
766 level = 2;
767 goto got;
768 }
769 block -= indirect_blks;
770 if (block < indirect_blks) {
771 offset[n++] = NODE_IND2_BLOCK;
772 noffset[n] = 4 + dptrs_per_blk;
773 offset[n++] = block / direct_blks;
774 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
775 offset[n] = block % direct_blks;
776 level = 2;
777 goto got;
778 }
779 block -= indirect_blks;
780 if (block < dindirect_blks) {
781 offset[n++] = NODE_DIND_BLOCK;
782 noffset[n] = 5 + (dptrs_per_blk * 2);
783 offset[n++] = block / indirect_blks;
784 noffset[n] = 6 + (dptrs_per_blk * 2) +
785 offset[n - 1] * (dptrs_per_blk + 1);
786 offset[n++] = (block / direct_blks) % dptrs_per_blk;
787 noffset[n] = 7 + (dptrs_per_blk * 2) +
788 offset[n - 2] * (dptrs_per_blk + 1) +
789 offset[n - 1];
790 offset[n] = block % direct_blks;
791 level = 3;
792 goto got;
793 } else {
794 return -E2BIG;
795 }
796got:
797 return level;
798}
799
800static struct folio *f2fs_get_node_folio_ra(struct folio *parent, int start);
801
802/*
803 * Caller should call f2fs_put_dnode(dn).
804 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
805 * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
806 */
807int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
808{
809 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
810 struct folio *nfolio[4];
811 struct folio *parent = NULL;
812 int offset[4];
813 unsigned int noffset[4];
814 nid_t nids[4];
815 int level, i = 0;
816 int err = 0;
817
818 level = get_node_path(dn->inode, index, offset, noffset);
819 if (level < 0)
820 return level;
821
822 nids[0] = dn->inode->i_ino;
823
824 if (!dn->inode_folio) {
825 nfolio[0] = f2fs_get_inode_folio(sbi, nids[0]);
826 if (IS_ERR(nfolio[0]))
827 return PTR_ERR(nfolio[0]);
828 } else {
829 nfolio[0] = dn->inode_folio;
830 }
831
832 /* if inline_data is set, should not report any block indices */
833 if (f2fs_has_inline_data(dn->inode) && index) {
834 err = -ENOENT;
835 f2fs_folio_put(nfolio[0], true);
836 goto release_out;
837 }
838
839 parent = nfolio[0];
840 if (level != 0)
841 nids[1] = get_nid(parent, offset[0], true);
842 dn->inode_folio = nfolio[0];
843 dn->inode_folio_locked = true;
844
845 /* get indirect or direct nodes */
846 for (i = 1; i <= level; i++) {
847 bool done = false;
848
849 if (nids[i] && nids[i] == dn->inode->i_ino) {
850 err = -EFSCORRUPTED;
851 f2fs_err_ratelimited(sbi,
852 "inode mapping table is corrupted, run fsck to fix it, "
853 "ino:%llu, nid:%u, level:%d, offset:%d",
854 dn->inode->i_ino, nids[i], level, offset[level]);
855 set_sbi_flag(sbi, SBI_NEED_FSCK);
856 goto release_pages;
857 }
858
859 if (!nids[i] && mode == ALLOC_NODE) {
860 /* alloc new node */
861 if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
862 err = -ENOSPC;
863 goto release_pages;
864 }
865
866 dn->nid = nids[i];
867 nfolio[i] = f2fs_new_node_folio(dn, noffset[i]);
868 if (IS_ERR(nfolio[i])) {
869 f2fs_alloc_nid_failed(sbi, nids[i]);
870 err = PTR_ERR(nfolio[i]);
871 goto release_pages;
872 }
873
874 set_nid(parent, offset[i - 1], nids[i], i == 1);
875 f2fs_alloc_nid_done(sbi, nids[i]);
876 done = true;
877 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
878 nfolio[i] = f2fs_get_node_folio_ra(parent, offset[i - 1]);
879 if (IS_ERR(nfolio[i])) {
880 err = PTR_ERR(nfolio[i]);
881 goto release_pages;
882 }
883 done = true;
884 }
885 if (i == 1) {
886 dn->inode_folio_locked = false;
887 folio_unlock(parent);
888 } else {
889 f2fs_folio_put(parent, true);
890 }
891
892 if (!done) {
893 nfolio[i] = f2fs_get_node_folio(sbi, nids[i],
894 NODE_TYPE_NON_INODE);
895 if (IS_ERR(nfolio[i])) {
896 err = PTR_ERR(nfolio[i]);
897 f2fs_folio_put(nfolio[0], false);
898 goto release_out;
899 }
900 }
901 if (i < level) {
902 parent = nfolio[i];
903 nids[i + 1] = get_nid(parent, offset[i], false);
904 }
905 }
906 dn->nid = nids[level];
907 dn->ofs_in_node = offset[level];
908 dn->node_folio = nfolio[level];
909 dn->data_blkaddr = f2fs_data_blkaddr(dn);
910
911 if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) &&
912 f2fs_sb_has_readonly(sbi)) {
913 unsigned int cluster_size = F2FS_I(dn->inode)->i_cluster_size;
914 unsigned int ofs_in_node = dn->ofs_in_node;
915 pgoff_t fofs = index;
916 unsigned int c_len;
917 block_t blkaddr;
918
919 /* should align fofs and ofs_in_node to cluster_size */
920 if (fofs % cluster_size) {
921 fofs = round_down(fofs, cluster_size);
922 ofs_in_node = round_down(ofs_in_node, cluster_size);
923 }
924
925 c_len = f2fs_cluster_blocks_are_contiguous(dn, ofs_in_node);
926 if (!c_len)
927 goto out;
928
929 blkaddr = data_blkaddr(dn->inode, dn->node_folio, ofs_in_node);
930 if (blkaddr == COMPRESS_ADDR)
931 blkaddr = data_blkaddr(dn->inode, dn->node_folio,
932 ofs_in_node + 1);
933
934 f2fs_update_read_extent_tree_range_compressed(dn->inode,
935 fofs, blkaddr, cluster_size, c_len);
936 }
937out:
938 return 0;
939
940release_pages:
941 f2fs_folio_put(parent, true);
942 if (i > 1)
943 f2fs_folio_put(nfolio[0], false);
944release_out:
945 dn->inode_folio = NULL;
946 dn->node_folio = NULL;
947 if (err == -ENOENT) {
948 dn->cur_level = i;
949 dn->max_level = level;
950 dn->ofs_in_node = offset[level];
951 }
952 return err;
953}
954
955static int truncate_node(struct dnode_of_data *dn)
956{
957 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
958 struct node_info ni;
959 int err;
960 pgoff_t index;
961
962 err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
963 if (err)
964 return err;
965
966 if (ni.blk_addr != NEW_ADDR &&
967 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr, DATA_GENERIC_ENHANCE)) {
968 f2fs_err_ratelimited(sbi,
969 "nat entry is corrupted, run fsck to fix it, ino:%u, "
970 "nid:%u, blkaddr:%u", ni.ino, ni.nid, ni.blk_addr);
971 set_sbi_flag(sbi, SBI_NEED_FSCK);
972 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NAT);
973 return -EFSCORRUPTED;
974 }
975
976 /* Deallocate node address */
977 f2fs_invalidate_blocks(sbi, ni.blk_addr, 1);
978 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
979 set_node_addr(sbi, &ni, NULL_ADDR, false);
980
981 if (dn->nid == dn->inode->i_ino) {
982 f2fs_remove_orphan_inode(sbi, dn->nid);
983 dec_valid_inode_count(sbi);
984 f2fs_inode_synced(dn->inode);
985 }
986
987 clear_node_folio_dirty(dn->node_folio);
988 set_sbi_flag(sbi, SBI_IS_DIRTY);
989
990 index = dn->node_folio->index;
991 f2fs_folio_put(dn->node_folio, true);
992
993 invalidate_mapping_pages(NODE_MAPPING(sbi),
994 index, index);
995
996 dn->node_folio = NULL;
997 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
998
999 return 0;
1000}
1001
1002static int truncate_dnode(struct dnode_of_data *dn)
1003{
1004 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1005 struct folio *folio;
1006 int err;
1007
1008 if (dn->nid == 0)
1009 return 1;
1010
1011 /* get direct node */
1012 folio = f2fs_get_node_folio(sbi, dn->nid, NODE_TYPE_NON_INODE);
1013 if (PTR_ERR(folio) == -ENOENT)
1014 return 1;
1015 else if (IS_ERR(folio))
1016 return PTR_ERR(folio);
1017
1018 if (IS_INODE(folio) || ino_of_node(folio) != dn->inode->i_ino) {
1019 f2fs_err(sbi, "incorrect node reference, ino: %llu, nid: %u, ino_of_node: %u",
1020 dn->inode->i_ino, dn->nid, ino_of_node(folio));
1021 set_sbi_flag(sbi, SBI_NEED_FSCK);
1022 f2fs_handle_error(sbi, ERROR_INVALID_NODE_REFERENCE);
1023 f2fs_folio_put(folio, true);
1024 return -EFSCORRUPTED;
1025 }
1026
1027 /* Make dnode_of_data for parameter */
1028 dn->node_folio = folio;
1029 dn->ofs_in_node = 0;
1030 f2fs_truncate_data_blocks_range(dn, ADDRS_PER_BLOCK(dn->inode));
1031 err = truncate_node(dn);
1032 if (err) {
1033 f2fs_folio_put(folio, true);
1034 return err;
1035 }
1036
1037 return 1;
1038}
1039
1040static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
1041 int ofs, int depth)
1042{
1043 struct dnode_of_data rdn = *dn;
1044 struct folio *folio;
1045 struct f2fs_node *rn;
1046 nid_t child_nid;
1047 unsigned int child_nofs;
1048 int freed = 0;
1049 int i, ret;
1050
1051 if (dn->nid == 0)
1052 return NIDS_PER_BLOCK + 1;
1053
1054 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
1055
1056 folio = f2fs_get_node_folio(F2FS_I_SB(dn->inode), dn->nid,
1057 NODE_TYPE_NON_INODE);
1058 if (IS_ERR(folio)) {
1059 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(folio));
1060 return PTR_ERR(folio);
1061 }
1062
1063 f2fs_ra_node_pages(folio, ofs, NIDS_PER_BLOCK);
1064
1065 rn = F2FS_NODE(folio);
1066 if (depth < 3) {
1067 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
1068 child_nid = le32_to_cpu(rn->in.nid[i]);
1069 if (child_nid == 0)
1070 continue;
1071 rdn.nid = child_nid;
1072 ret = truncate_dnode(&rdn);
1073 if (ret < 0)
1074 goto out_err;
1075 if (set_nid(folio, i, 0, false))
1076 dn->node_changed = true;
1077 }
1078 } else {
1079 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
1080 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
1081 child_nid = le32_to_cpu(rn->in.nid[i]);
1082 if (child_nid == 0) {
1083 child_nofs += NIDS_PER_BLOCK + 1;
1084 continue;
1085 }
1086 rdn.nid = child_nid;
1087 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
1088 if (ret == (NIDS_PER_BLOCK + 1)) {
1089 if (set_nid(folio, i, 0, false))
1090 dn->node_changed = true;
1091 child_nofs += ret;
1092 } else if (ret < 0 && ret != -ENOENT) {
1093 goto out_err;
1094 }
1095 }
1096 freed = child_nofs;
1097 }
1098
1099 if (!ofs) {
1100 /* remove current indirect node */
1101 dn->node_folio = folio;
1102 ret = truncate_node(dn);
1103 if (ret)
1104 goto out_err;
1105 freed++;
1106 } else {
1107 f2fs_folio_put(folio, true);
1108 }
1109 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
1110 return freed;
1111
1112out_err:
1113 f2fs_folio_put(folio, true);
1114 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
1115 return ret;
1116}
1117
1118static int truncate_partial_nodes(struct dnode_of_data *dn,
1119 int *offset, int depth)
1120{
1121 struct folio *folios[2];
1122 nid_t nid[3];
1123 nid_t child_nid;
1124 int err = 0;
1125 int i;
1126 int idx = depth - 2;
1127
1128 nid[0] = get_nid(dn->inode_folio, offset[0], true);
1129 if (!nid[0])
1130 return 0;
1131
1132 /* get indirect nodes in the path */
1133 for (i = 0; i < idx + 1; i++) {
1134 /* reference count'll be increased */
1135 folios[i] = f2fs_get_node_folio(F2FS_I_SB(dn->inode), nid[i],
1136 NODE_TYPE_NON_INODE);
1137 if (IS_ERR(folios[i])) {
1138 err = PTR_ERR(folios[i]);
1139 idx = i - 1;
1140 goto fail;
1141 }
1142 nid[i + 1] = get_nid(folios[i], offset[i + 1], false);
1143 }
1144
1145 f2fs_ra_node_pages(folios[idx], offset[idx + 1], NIDS_PER_BLOCK);
1146
1147 /* free direct nodes linked to a partial indirect node */
1148 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1149 child_nid = get_nid(folios[idx], i, false);
1150 if (!child_nid)
1151 continue;
1152 dn->nid = child_nid;
1153 err = truncate_dnode(dn);
1154 if (err < 0)
1155 goto fail;
1156 if (set_nid(folios[idx], i, 0, false))
1157 dn->node_changed = true;
1158 }
1159
1160 if (offset[idx + 1] == 0) {
1161 dn->node_folio = folios[idx];
1162 dn->nid = nid[idx];
1163 err = truncate_node(dn);
1164 if (err)
1165 goto fail;
1166 } else {
1167 f2fs_folio_put(folios[idx], true);
1168 }
1169 offset[idx]++;
1170 offset[idx + 1] = 0;
1171 idx--;
1172fail:
1173 for (i = idx; i >= 0; i--)
1174 f2fs_folio_put(folios[i], true);
1175
1176 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1177
1178 return err;
1179}
1180
1181/*
1182 * All the block addresses of data and nodes should be nullified.
1183 */
1184int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1185{
1186 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1187 int err = 0, cont = 1;
1188 int level, offset[4], noffset[4];
1189 unsigned int nofs = 0;
1190 struct dnode_of_data dn;
1191 struct folio *folio;
1192
1193 trace_f2fs_truncate_inode_blocks_enter(inode, from);
1194
1195 level = get_node_path(inode, from, offset, noffset);
1196 if (level <= 0) {
1197 if (!level) {
1198 level = -EFSCORRUPTED;
1199 f2fs_err(sbi, "%s: inode ino=%llx has corrupted node block, from:%lu addrs:%u",
1200 __func__, inode->i_ino,
1201 from, ADDRS_PER_INODE(inode));
1202 set_sbi_flag(sbi, SBI_NEED_FSCK);
1203 }
1204 trace_f2fs_truncate_inode_blocks_exit(inode, level);
1205 return level;
1206 }
1207
1208 folio = f2fs_get_inode_folio(sbi, inode->i_ino);
1209 if (IS_ERR(folio)) {
1210 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(folio));
1211 return PTR_ERR(folio);
1212 }
1213
1214 set_new_dnode(&dn, inode, folio, NULL, 0);
1215 folio_unlock(folio);
1216
1217 switch (level) {
1218 case 0:
1219 case 1:
1220 nofs = noffset[1];
1221 break;
1222 case 2:
1223 nofs = noffset[1];
1224 if (!offset[level - 1])
1225 goto skip_partial;
1226 err = truncate_partial_nodes(&dn, offset, level);
1227 if (err < 0 && err != -ENOENT)
1228 goto fail;
1229 nofs += 1 + NIDS_PER_BLOCK;
1230 break;
1231 case 3:
1232 nofs = 5 + 2 * NIDS_PER_BLOCK;
1233 if (!offset[level - 1])
1234 goto skip_partial;
1235 err = truncate_partial_nodes(&dn, offset, level);
1236 if (err < 0 && err != -ENOENT)
1237 goto fail;
1238 break;
1239 default:
1240 BUG();
1241 }
1242
1243skip_partial:
1244 while (cont) {
1245 dn.nid = get_nid(folio, offset[0], true);
1246 switch (offset[0]) {
1247 case NODE_DIR1_BLOCK:
1248 case NODE_DIR2_BLOCK:
1249 err = truncate_dnode(&dn);
1250 break;
1251
1252 case NODE_IND1_BLOCK:
1253 case NODE_IND2_BLOCK:
1254 err = truncate_nodes(&dn, nofs, offset[1], 2);
1255 break;
1256
1257 case NODE_DIND_BLOCK:
1258 err = truncate_nodes(&dn, nofs, offset[1], 3);
1259 cont = 0;
1260 break;
1261
1262 default:
1263 BUG();
1264 }
1265 if (err == -ENOENT) {
1266 set_sbi_flag(F2FS_F_SB(folio), SBI_NEED_FSCK);
1267 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1268 f2fs_err_ratelimited(sbi,
1269 "truncate node fail, ino:%llu, nid:%u, "
1270 "offset[0]:%d, offset[1]:%d, nofs:%d",
1271 inode->i_ino, dn.nid, offset[0],
1272 offset[1], nofs);
1273 err = 0;
1274 }
1275 if (err < 0)
1276 goto fail;
1277 if (offset[1] == 0 && get_nid(folio, offset[0], true)) {
1278 folio_lock(folio);
1279 BUG_ON(!is_node_folio(folio));
1280 set_nid(folio, offset[0], 0, true);
1281 folio_unlock(folio);
1282 }
1283 offset[1] = 0;
1284 offset[0]++;
1285 nofs += err;
1286 }
1287fail:
1288 f2fs_folio_put(folio, false);
1289 trace_f2fs_truncate_inode_blocks_exit(inode, err);
1290 return err > 0 ? 0 : err;
1291}
1292
1293/* caller must lock inode page */
1294int f2fs_truncate_xattr_node(struct inode *inode)
1295{
1296 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1297 nid_t nid = F2FS_I(inode)->i_xattr_nid;
1298 struct dnode_of_data dn;
1299 struct folio *nfolio;
1300 int err;
1301
1302 if (!nid)
1303 return 0;
1304
1305 nfolio = f2fs_get_xnode_folio(sbi, nid);
1306 if (IS_ERR(nfolio))
1307 return PTR_ERR(nfolio);
1308
1309 set_new_dnode(&dn, inode, NULL, nfolio, nid);
1310 err = truncate_node(&dn);
1311 if (err) {
1312 f2fs_folio_put(nfolio, true);
1313 return err;
1314 }
1315
1316 f2fs_i_xnid_write(inode, 0);
1317
1318 return 0;
1319}
1320
1321/*
1322 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1323 * f2fs_unlock_op().
1324 */
1325int f2fs_remove_inode_page(struct inode *inode)
1326{
1327 struct dnode_of_data dn;
1328 int err;
1329
1330 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1331 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1332 if (err)
1333 return err;
1334
1335 err = f2fs_truncate_xattr_node(inode);
1336 if (err) {
1337 f2fs_put_dnode(&dn);
1338 return err;
1339 }
1340
1341 /* remove potential inline_data blocks */
1342 if (!IS_DEVICE_ALIASING(inode) &&
1343 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1344 S_ISLNK(inode->i_mode)))
1345 f2fs_truncate_data_blocks_range(&dn, 1);
1346
1347 /* 0 is possible, after f2fs_new_inode() has failed */
1348 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1349 f2fs_put_dnode(&dn);
1350 return -EIO;
1351 }
1352
1353 if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1354 f2fs_warn(F2FS_I_SB(inode),
1355 "f2fs_remove_inode_page: inconsistent i_blocks, ino:%llu, iblocks:%llu",
1356 inode->i_ino, (unsigned long long)inode->i_blocks);
1357 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1358 }
1359
1360 /* will put inode & node pages */
1361 err = truncate_node(&dn);
1362 if (err) {
1363 f2fs_put_dnode(&dn);
1364 return err;
1365 }
1366 return 0;
1367}
1368
1369struct folio *f2fs_new_inode_folio(struct inode *inode)
1370{
1371 struct dnode_of_data dn;
1372
1373 /* allocate inode page for new inode */
1374 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1375
1376 /* caller should f2fs_folio_put(folio, true); */
1377 return f2fs_new_node_folio(&dn, 0);
1378}
1379
1380struct folio *f2fs_new_node_folio(struct dnode_of_data *dn, unsigned int ofs)
1381{
1382 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1383 struct node_info new_ni;
1384 struct folio *folio;
1385 int err;
1386
1387 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1388 return ERR_PTR(-EPERM);
1389
1390 folio = f2fs_grab_cache_folio(NODE_MAPPING(sbi), dn->nid, false);
1391 if (IS_ERR(folio))
1392 return folio;
1393
1394 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1395 goto fail;
1396
1397#ifdef CONFIG_F2FS_CHECK_FS
1398 err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false);
1399 if (err) {
1400 dec_valid_node_count(sbi, dn->inode, !ofs);
1401 goto fail;
1402 }
1403 if (unlikely(new_ni.blk_addr != NULL_ADDR)) {
1404 err = -EFSCORRUPTED;
1405 dec_valid_node_count(sbi, dn->inode, !ofs);
1406 set_sbi_flag(sbi, SBI_NEED_FSCK);
1407 f2fs_warn_ratelimited(sbi,
1408 "f2fs_new_node_folio: inconsistent nat entry, "
1409 "ino:%u, nid:%u, blkaddr:%u, ver:%u, flag:%u",
1410 new_ni.ino, new_ni.nid, new_ni.blk_addr,
1411 new_ni.version, new_ni.flag);
1412 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NAT);
1413 goto fail;
1414 }
1415#endif
1416 new_ni.nid = dn->nid;
1417 new_ni.ino = dn->inode->i_ino;
1418 new_ni.blk_addr = NULL_ADDR;
1419 new_ni.flag = 0;
1420 new_ni.version = 0;
1421 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1422
1423 f2fs_folio_wait_writeback(folio, NODE, true, true);
1424 fill_node_footer(folio, dn->nid, dn->inode->i_ino, ofs, true);
1425 set_cold_node(folio, S_ISDIR(dn->inode->i_mode));
1426 if (!folio_test_uptodate(folio))
1427 folio_mark_uptodate(folio);
1428 if (folio_mark_dirty(folio))
1429 dn->node_changed = true;
1430
1431 if (f2fs_has_xattr_block(ofs))
1432 f2fs_i_xnid_write(dn->inode, dn->nid);
1433
1434 if (ofs == 0)
1435 inc_valid_inode_count(sbi);
1436 return folio;
1437fail:
1438 clear_node_folio_dirty(folio);
1439 f2fs_folio_put(folio, true);
1440 return ERR_PTR(err);
1441}
1442
1443/*
1444 * Caller should do after getting the following values.
1445 * 0: f2fs_folio_put(folio, false)
1446 * LOCKED_PAGE or error: f2fs_folio_put(folio, true)
1447 */
1448static int read_node_folio(struct folio *folio, blk_opf_t op_flags)
1449{
1450 struct f2fs_sb_info *sbi = F2FS_F_SB(folio);
1451 struct node_info ni;
1452 struct f2fs_io_info fio = {
1453 .sbi = sbi,
1454 .type = NODE,
1455 .op = REQ_OP_READ,
1456 .op_flags = op_flags,
1457 .folio = folio,
1458 .encrypted_page = NULL,
1459 };
1460 int err;
1461
1462 if (folio_test_uptodate(folio)) {
1463 if (!f2fs_inode_chksum_verify(sbi, folio)) {
1464 folio_clear_uptodate(folio);
1465 return -EFSBADCRC;
1466 }
1467 return LOCKED_PAGE;
1468 }
1469
1470 err = f2fs_get_node_info(sbi, folio->index, &ni, false);
1471 if (err)
1472 return err;
1473
1474 /* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1475 if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR)) {
1476 folio_clear_uptodate(folio);
1477 return -ENOENT;
1478 }
1479
1480 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1481
1482 err = f2fs_submit_page_bio(&fio);
1483
1484 if (!err)
1485 f2fs_update_iostat(sbi, NULL, FS_NODE_READ_IO, F2FS_BLKSIZE);
1486
1487 return err;
1488}
1489
1490/*
1491 * Readahead a node page
1492 */
1493void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1494{
1495 struct folio *afolio;
1496 int err;
1497
1498 if (!nid)
1499 return;
1500 if (f2fs_check_nid_range(sbi, nid))
1501 return;
1502
1503 afolio = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1504 if (afolio)
1505 return;
1506
1507 afolio = f2fs_grab_cache_folio(NODE_MAPPING(sbi), nid, false);
1508 if (IS_ERR(afolio))
1509 return;
1510
1511 err = read_node_folio(afolio, REQ_RAHEAD);
1512 f2fs_folio_put(afolio, err ? true : false);
1513}
1514
1515int f2fs_sanity_check_node_footer(struct f2fs_sb_info *sbi,
1516 struct folio *folio, pgoff_t nid,
1517 enum node_type ntype, bool in_irq)
1518{
1519 bool is_inode, is_xnode;
1520
1521 if (unlikely(nid != nid_of_node(folio)))
1522 goto out_err;
1523
1524 is_inode = IS_INODE(folio);
1525 is_xnode = f2fs_has_xattr_block(ofs_of_node(folio));
1526
1527 switch (ntype) {
1528 case NODE_TYPE_REGULAR:
1529 if (is_inode && is_xnode)
1530 goto out_err;
1531 break;
1532 case NODE_TYPE_INODE:
1533 if (!is_inode || is_xnode)
1534 goto out_err;
1535 break;
1536 case NODE_TYPE_XATTR:
1537 if (is_inode || !is_xnode)
1538 goto out_err;
1539 break;
1540 case NODE_TYPE_NON_INODE:
1541 if (is_inode)
1542 goto out_err;
1543 break;
1544 default:
1545 break;
1546 }
1547 if (time_to_inject(sbi, FAULT_INCONSISTENT_FOOTER))
1548 goto out_err;
1549 return 0;
1550out_err:
1551 set_sbi_flag(sbi, SBI_NEED_FSCK);
1552 f2fs_warn_ratelimited(sbi, "inconsistent node block, node_type:%d, nid:%lu, "
1553 "node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1554 ntype, nid, nid_of_node(folio), ino_of_node(folio),
1555 ofs_of_node(folio), cpver_of_node(folio),
1556 next_blkaddr_of_node(folio));
1557
1558 f2fs_handle_error(sbi, ERROR_INCONSISTENT_FOOTER);
1559 return -EFSCORRUPTED;
1560}
1561
1562static struct folio *__get_node_folio(struct f2fs_sb_info *sbi, pgoff_t nid,
1563 struct folio *parent, int start, enum node_type ntype)
1564{
1565 struct folio *folio;
1566 int err;
1567
1568 if (!nid)
1569 return ERR_PTR(-ENOENT);
1570 if (f2fs_check_nid_range(sbi, nid))
1571 return ERR_PTR(-EINVAL);
1572repeat:
1573 folio = f2fs_grab_cache_folio(NODE_MAPPING(sbi), nid, false);
1574 if (IS_ERR(folio))
1575 return folio;
1576
1577 err = read_node_folio(folio, 0);
1578 if (err < 0)
1579 goto out_put_err;
1580 if (err == LOCKED_PAGE)
1581 goto page_hit;
1582
1583 if (parent)
1584 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1585
1586 folio_lock(folio);
1587
1588 if (unlikely(!is_node_folio(folio))) {
1589 f2fs_folio_put(folio, true);
1590 goto repeat;
1591 }
1592
1593 if (unlikely(!folio_test_uptodate(folio))) {
1594 err = -EIO;
1595 goto out_put_err;
1596 }
1597
1598 if (!f2fs_inode_chksum_verify(sbi, folio)) {
1599 err = -EFSBADCRC;
1600 goto out_err;
1601 }
1602page_hit:
1603 err = f2fs_sanity_check_node_footer(sbi, folio, nid, ntype, false);
1604 if (!err)
1605 return folio;
1606out_err:
1607 folio_clear_uptodate(folio);
1608out_put_err:
1609 /* ENOENT comes from read_node_folio which is not an error. */
1610 if (err != -ENOENT)
1611 f2fs_handle_page_eio(sbi, folio, NODE);
1612 f2fs_folio_put(folio, true);
1613 return ERR_PTR(err);
1614}
1615
1616struct folio *f2fs_get_node_folio(struct f2fs_sb_info *sbi, pgoff_t nid,
1617 enum node_type node_type)
1618{
1619 return __get_node_folio(sbi, nid, NULL, 0, node_type);
1620}
1621
1622struct folio *f2fs_get_inode_folio(struct f2fs_sb_info *sbi, pgoff_t ino)
1623{
1624 return __get_node_folio(sbi, ino, NULL, 0, NODE_TYPE_INODE);
1625}
1626
1627struct folio *f2fs_get_xnode_folio(struct f2fs_sb_info *sbi, pgoff_t xnid)
1628{
1629 return __get_node_folio(sbi, xnid, NULL, 0, NODE_TYPE_XATTR);
1630}
1631
1632static struct folio *f2fs_get_node_folio_ra(struct folio *parent, int start)
1633{
1634 struct f2fs_sb_info *sbi = F2FS_F_SB(parent);
1635 nid_t nid = get_nid(parent, start, false);
1636
1637 return __get_node_folio(sbi, nid, parent, start, NODE_TYPE_REGULAR);
1638}
1639
1640static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1641{
1642 struct inode *inode;
1643 struct folio *folio;
1644 int ret;
1645
1646 /* should flush inline_data before evict_inode */
1647 inode = ilookup(sbi->sb, ino);
1648 if (!inode)
1649 return;
1650
1651 folio = f2fs_filemap_get_folio(inode->i_mapping, 0,
1652 FGP_LOCK|FGP_NOWAIT, 0);
1653 if (IS_ERR(folio))
1654 goto iput_out;
1655
1656 if (!folio_test_uptodate(folio))
1657 goto folio_out;
1658
1659 if (!folio_test_dirty(folio))
1660 goto folio_out;
1661
1662 if (!folio_clear_dirty_for_io(folio))
1663 goto folio_out;
1664
1665 ret = f2fs_write_inline_data(inode, folio);
1666 inode_dec_dirty_pages(inode);
1667 f2fs_remove_dirty_inode(inode);
1668 if (ret)
1669 folio_mark_dirty(folio);
1670folio_out:
1671 f2fs_folio_put(folio, true);
1672iput_out:
1673 iput(inode);
1674}
1675
1676static struct folio *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1677{
1678 pgoff_t index;
1679 struct folio_batch fbatch;
1680 struct folio *last_folio = NULL;
1681 int nr_folios;
1682
1683 folio_batch_init(&fbatch);
1684 index = 0;
1685
1686 while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1687 (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1688 &fbatch))) {
1689 int i;
1690
1691 for (i = 0; i < nr_folios; i++) {
1692 struct folio *folio = fbatch.folios[i];
1693
1694 if (unlikely(f2fs_cp_error(sbi))) {
1695 f2fs_folio_put(last_folio, false);
1696 folio_batch_release(&fbatch);
1697 return ERR_PTR(-EIO);
1698 }
1699
1700 if (!IS_DNODE(folio) || !is_cold_node(folio))
1701 continue;
1702 if (ino_of_node(folio) != ino)
1703 continue;
1704
1705 folio_lock(folio);
1706
1707 if (unlikely(!is_node_folio(folio))) {
1708continue_unlock:
1709 folio_unlock(folio);
1710 continue;
1711 }
1712 if (ino_of_node(folio) != ino)
1713 goto continue_unlock;
1714
1715 if (!folio_test_dirty(folio)) {
1716 /* someone wrote it for us */
1717 goto continue_unlock;
1718 }
1719
1720 if (last_folio)
1721 f2fs_folio_put(last_folio, false);
1722
1723 folio_get(folio);
1724 last_folio = folio;
1725 folio_unlock(folio);
1726 }
1727 folio_batch_release(&fbatch);
1728 cond_resched();
1729 }
1730 return last_folio;
1731}
1732
1733static bool __write_node_folio(struct folio *folio, bool atomic, bool do_fsync,
1734 bool *submitted, struct writeback_control *wbc,
1735 bool do_balance, enum iostat_type io_type,
1736 unsigned int *seq_id)
1737{
1738 struct f2fs_sb_info *sbi = F2FS_F_SB(folio);
1739 nid_t nid;
1740 struct node_info ni;
1741 struct f2fs_io_info fio = {
1742 .sbi = sbi,
1743 .ino = ino_of_node(folio),
1744 .type = NODE,
1745 .op = REQ_OP_WRITE,
1746 .op_flags = wbc_to_write_flags(wbc),
1747 .folio = folio,
1748 .encrypted_page = NULL,
1749 .submitted = 0,
1750 .io_type = io_type,
1751 .io_wbc = wbc,
1752 };
1753 struct f2fs_lock_context lc;
1754 unsigned int seq;
1755
1756 trace_f2fs_writepage(folio, NODE);
1757
1758 if (unlikely(f2fs_cp_error(sbi))) {
1759 /* keep node pages in remount-ro mode */
1760 if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY)
1761 goto redirty_out;
1762 folio_clear_uptodate(folio);
1763 dec_page_count(sbi, F2FS_DIRTY_NODES);
1764 folio_unlock(folio);
1765 return true;
1766 }
1767
1768 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1769 goto redirty_out;
1770
1771 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1772 wbc->sync_mode == WB_SYNC_NONE &&
1773 IS_DNODE(folio) && is_cold_node(folio))
1774 goto redirty_out;
1775
1776 /* get old block addr of this node page */
1777 nid = nid_of_node(folio);
1778
1779 if (f2fs_sanity_check_node_footer(sbi, folio, nid,
1780 NODE_TYPE_REGULAR, false)) {
1781 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_CORRUPTED_NID);
1782 goto redirty_out;
1783 }
1784
1785 if (f2fs_get_node_info(sbi, nid, &ni, !do_balance))
1786 goto redirty_out;
1787
1788 f2fs_down_read_trace(&sbi->node_write, &lc);
1789
1790 /* This page is already truncated */
1791 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1792 folio_clear_uptodate(folio);
1793 dec_page_count(sbi, F2FS_DIRTY_NODES);
1794 f2fs_up_read_trace(&sbi->node_write, &lc);
1795 folio_unlock(folio);
1796 return true;
1797 }
1798
1799 if (__is_valid_data_blkaddr(ni.blk_addr) &&
1800 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1801 DATA_GENERIC_ENHANCE)) {
1802 f2fs_up_read_trace(&sbi->node_write, &lc);
1803 goto redirty_out;
1804 }
1805
1806 if (atomic && !test_opt(sbi, NOBARRIER))
1807 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1808
1809 set_dentry_mark(folio, false);
1810 set_fsync_mark(folio, do_fsync);
1811 if (IS_INODE(folio) && (atomic || is_fsync_dnode(folio)))
1812 set_dentry_mark(folio,
1813 f2fs_need_dentry_mark(sbi, ino_of_node(folio)));
1814
1815 /* should add to global list before clearing PAGECACHE status */
1816 if (f2fs_in_warm_node_list(folio)) {
1817 seq = f2fs_add_fsync_node_entry(sbi, folio);
1818 if (seq_id)
1819 *seq_id = seq;
1820 }
1821
1822 folio_start_writeback(folio);
1823
1824 fio.old_blkaddr = ni.blk_addr;
1825 f2fs_do_write_node_page(nid, &fio);
1826 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(folio));
1827 dec_page_count(sbi, F2FS_DIRTY_NODES);
1828 f2fs_up_read_trace(&sbi->node_write, &lc);
1829
1830 folio_unlock(folio);
1831
1832 if (unlikely(f2fs_cp_error(sbi))) {
1833 f2fs_submit_merged_write(sbi, NODE);
1834 submitted = NULL;
1835 }
1836 if (submitted)
1837 *submitted = fio.submitted;
1838
1839 if (do_balance)
1840 f2fs_balance_fs(sbi, false);
1841 return true;
1842
1843redirty_out:
1844 folio_redirty_for_writepage(wbc, folio);
1845 folio_unlock(folio);
1846 return false;
1847}
1848
1849int f2fs_write_single_node_folio(struct folio *node_folio, int sync_mode,
1850 bool mark_dirty, enum iostat_type io_type)
1851{
1852 int err = 0;
1853 struct writeback_control wbc = {
1854 .sync_mode = WB_SYNC_ALL,
1855 .nr_to_write = 1,
1856 };
1857
1858 if (!sync_mode) {
1859 /* set page dirty and write it */
1860 if (!folio_test_writeback(node_folio))
1861 folio_mark_dirty(node_folio);
1862 goto out_folio;
1863 }
1864
1865 f2fs_folio_wait_writeback(node_folio, NODE, true, true);
1866
1867 if (mark_dirty)
1868 folio_mark_dirty(node_folio);
1869 else if (!folio_test_dirty(node_folio))
1870 goto out_folio;
1871
1872 if (!folio_clear_dirty_for_io(node_folio)) {
1873 err = -EAGAIN;
1874 goto out_folio;
1875 }
1876
1877 if (!__write_node_folio(node_folio, false, false, NULL,
1878 &wbc, false, FS_GC_NODE_IO, NULL))
1879 err = -EAGAIN;
1880 goto release_folio;
1881out_folio:
1882 folio_unlock(node_folio);
1883release_folio:
1884 f2fs_folio_put(node_folio, false);
1885 return err;
1886}
1887
1888int f2fs_move_node_folio(struct folio *node_folio, int gc_type)
1889{
1890 return f2fs_write_single_node_folio(node_folio, gc_type == FG_GC,
1891 true, FS_GC_NODE_IO);
1892}
1893
1894int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1895 struct writeback_control *wbc, bool atomic,
1896 unsigned int *seq_id)
1897{
1898 pgoff_t index;
1899 struct folio_batch fbatch;
1900 int ret = 0;
1901 struct folio *last_folio = NULL;
1902 bool marked = false;
1903 nid_t ino = inode->i_ino;
1904 int nr_folios;
1905 int nwritten = 0;
1906
1907 if (atomic) {
1908 last_folio = last_fsync_dnode(sbi, ino);
1909 if (IS_ERR_OR_NULL(last_folio))
1910 return PTR_ERR_OR_ZERO(last_folio);
1911 }
1912retry:
1913 folio_batch_init(&fbatch);
1914 index = 0;
1915
1916 while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1917 (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1918 &fbatch))) {
1919 int i;
1920
1921 for (i = 0; i < nr_folios; i++) {
1922 struct folio *folio = fbatch.folios[i];
1923 bool submitted = false;
1924 bool do_fsync = false;
1925
1926 if (unlikely(f2fs_cp_error(sbi))) {
1927 f2fs_folio_put(last_folio, false);
1928 folio_batch_release(&fbatch);
1929 ret = -EIO;
1930 goto out;
1931 }
1932
1933 if (!IS_DNODE(folio) || !is_cold_node(folio))
1934 continue;
1935 if (ino_of_node(folio) != ino)
1936 continue;
1937
1938 folio_lock(folio);
1939
1940 if (unlikely(!is_node_folio(folio))) {
1941continue_unlock:
1942 folio_unlock(folio);
1943 continue;
1944 }
1945 if (ino_of_node(folio) != ino)
1946 goto continue_unlock;
1947
1948 if (!folio_test_dirty(folio) && folio != last_folio) {
1949 /* someone wrote it for us */
1950 goto continue_unlock;
1951 }
1952
1953 f2fs_folio_wait_writeback(folio, NODE, true, true);
1954
1955 if (!atomic || folio == last_folio) {
1956 do_fsync = true;
1957 percpu_counter_inc(&sbi->rf_node_block_count);
1958 if (IS_INODE(folio)) {
1959 if (is_inode_flag_set(inode,
1960 FI_DIRTY_INODE))
1961 f2fs_update_inode(inode, folio);
1962 }
1963 /* may be written by other thread */
1964 if (!folio_test_dirty(folio))
1965 folio_mark_dirty(folio);
1966 }
1967
1968 if (!folio_clear_dirty_for_io(folio))
1969 goto continue_unlock;
1970
1971 if (!__write_node_folio(folio, atomic &&
1972 folio == last_folio,
1973 do_fsync, &submitted,
1974 wbc, true, FS_NODE_IO,
1975 seq_id)) {
1976 f2fs_folio_put(last_folio, false);
1977 folio_batch_release(&fbatch);
1978 ret = -EIO;
1979 goto out;
1980 }
1981 if (submitted)
1982 nwritten++;
1983
1984 if (folio == last_folio) {
1985 f2fs_folio_put(folio, false);
1986 folio_batch_release(&fbatch);
1987 marked = true;
1988 goto out;
1989 }
1990 }
1991 folio_batch_release(&fbatch);
1992 cond_resched();
1993 }
1994 if (atomic && !marked) {
1995 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1996 ino, last_folio->index);
1997 folio_lock(last_folio);
1998 f2fs_folio_wait_writeback(last_folio, NODE, true, true);
1999 folio_mark_dirty(last_folio);
2000 folio_unlock(last_folio);
2001 goto retry;
2002 }
2003out:
2004 if (nwritten)
2005 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
2006 return ret;
2007}
2008
2009static int f2fs_match_ino(struct inode *inode, u64 ino, void *data)
2010{
2011 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2012 bool clean;
2013
2014 if (inode->i_ino != ino)
2015 return 0;
2016
2017 if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
2018 return 0;
2019
2020 spin_lock(&sbi->inode_lock[DIRTY_META]);
2021 clean = list_empty(&F2FS_I(inode)->gdirty_list);
2022 spin_unlock(&sbi->inode_lock[DIRTY_META]);
2023
2024 if (clean)
2025 return 0;
2026
2027 inode = igrab(inode);
2028 if (!inode)
2029 return 0;
2030 return 1;
2031}
2032
2033static bool flush_dirty_inode(struct folio *folio)
2034{
2035 struct f2fs_sb_info *sbi = F2FS_F_SB(folio);
2036 struct inode *inode;
2037 nid_t ino = ino_of_node(folio);
2038
2039 inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
2040 if (!inode)
2041 return false;
2042
2043 f2fs_update_inode(inode, folio);
2044 folio_unlock(folio);
2045
2046 iput(inode);
2047 return true;
2048}
2049
2050void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
2051{
2052 pgoff_t index = 0;
2053 struct folio_batch fbatch;
2054 int nr_folios;
2055
2056 folio_batch_init(&fbatch);
2057
2058 while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
2059 (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
2060 &fbatch))) {
2061 int i;
2062
2063 for (i = 0; i < nr_folios; i++) {
2064 struct folio *folio = fbatch.folios[i];
2065
2066 if (!IS_INODE(folio))
2067 continue;
2068
2069 folio_lock(folio);
2070
2071 if (unlikely(!is_node_folio(folio)))
2072 goto unlock;
2073 if (!folio_test_dirty(folio))
2074 goto unlock;
2075
2076 /* flush inline_data, if it's async context. */
2077 if (folio_test_f2fs_inline(folio)) {
2078 folio_clear_f2fs_inline(folio);
2079 folio_unlock(folio);
2080 flush_inline_data(sbi, ino_of_node(folio));
2081 continue;
2082 }
2083unlock:
2084 folio_unlock(folio);
2085 }
2086 folio_batch_release(&fbatch);
2087 cond_resched();
2088 }
2089}
2090
2091int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
2092 struct writeback_control *wbc,
2093 bool do_balance, enum iostat_type io_type)
2094{
2095 pgoff_t index;
2096 struct folio_batch fbatch;
2097 int step = 0;
2098 int nwritten = 0;
2099 int ret = 0;
2100 int nr_folios, done = 0;
2101
2102 folio_batch_init(&fbatch);
2103
2104next_step:
2105 index = 0;
2106
2107 while (!done && (nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi),
2108 &index, (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
2109 &fbatch))) {
2110 int i;
2111
2112 for (i = 0; i < nr_folios; i++) {
2113 struct folio *folio = fbatch.folios[i];
2114 bool submitted = false;
2115
2116 /* give a priority to WB_SYNC threads */
2117 if (atomic_read(&sbi->wb_sync_req[NODE]) &&
2118 wbc->sync_mode == WB_SYNC_NONE) {
2119 done = 1;
2120 break;
2121 }
2122
2123 /*
2124 * flushing sequence with step:
2125 * 0. indirect nodes
2126 * 1. dentry dnodes
2127 * 2. file dnodes
2128 */
2129 if (step == 0 && IS_DNODE(folio))
2130 continue;
2131 if (step == 1 && (!IS_DNODE(folio) ||
2132 is_cold_node(folio)))
2133 continue;
2134 if (step == 2 && (!IS_DNODE(folio) ||
2135 !is_cold_node(folio)))
2136 continue;
2137lock_node:
2138 if (wbc->sync_mode == WB_SYNC_ALL)
2139 folio_lock(folio);
2140 else if (!folio_trylock(folio))
2141 continue;
2142
2143 if (unlikely(!is_node_folio(folio))) {
2144continue_unlock:
2145 folio_unlock(folio);
2146 continue;
2147 }
2148
2149 if (!folio_test_dirty(folio)) {
2150 /* someone wrote it for us */
2151 goto continue_unlock;
2152 }
2153
2154 /* flush inline_data/inode, if it's async context. */
2155 if (!do_balance)
2156 goto write_node;
2157
2158 /* flush inline_data */
2159 if (folio_test_f2fs_inline(folio)) {
2160 folio_clear_f2fs_inline(folio);
2161 folio_unlock(folio);
2162 flush_inline_data(sbi, ino_of_node(folio));
2163 goto lock_node;
2164 }
2165
2166 /* flush dirty inode */
2167 if (IS_INODE(folio) && flush_dirty_inode(folio))
2168 goto lock_node;
2169write_node:
2170 f2fs_folio_wait_writeback(folio, NODE, true, true);
2171
2172 if (!folio_clear_dirty_for_io(folio))
2173 goto continue_unlock;
2174
2175 if (!__write_node_folio(folio, false, false, &submitted,
2176 wbc, do_balance, io_type, NULL)) {
2177 folio_batch_release(&fbatch);
2178 ret = -EIO;
2179 goto out;
2180 }
2181 if (submitted)
2182 nwritten++;
2183
2184 if (--wbc->nr_to_write == 0)
2185 break;
2186 }
2187 folio_batch_release(&fbatch);
2188 cond_resched();
2189
2190 if (wbc->nr_to_write == 0) {
2191 step = 2;
2192 break;
2193 }
2194 }
2195
2196 if (step < 2) {
2197 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2198 wbc->sync_mode == WB_SYNC_NONE && step == 1)
2199 goto out;
2200 step++;
2201 goto next_step;
2202 }
2203out:
2204 if (nwritten)
2205 f2fs_submit_merged_write(sbi, NODE);
2206
2207 if (unlikely(f2fs_cp_error(sbi)))
2208 return -EIO;
2209 return ret;
2210}
2211
2212int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2213 unsigned int seq_id)
2214{
2215 struct fsync_node_entry *fn;
2216 struct list_head *head = &sbi->fsync_node_list;
2217 unsigned long flags;
2218 unsigned int cur_seq_id = 0;
2219
2220 while (seq_id && cur_seq_id < seq_id) {
2221 struct folio *folio;
2222
2223 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2224 if (list_empty(head)) {
2225 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2226 break;
2227 }
2228 fn = list_first_entry(head, struct fsync_node_entry, list);
2229 if (fn->seq_id > seq_id) {
2230 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2231 break;
2232 }
2233 cur_seq_id = fn->seq_id;
2234 folio = fn->folio;
2235 folio_get(folio);
2236 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2237
2238 f2fs_folio_wait_writeback(folio, NODE, true, false);
2239
2240 folio_put(folio);
2241 }
2242
2243 return filemap_check_errors(NODE_MAPPING(sbi));
2244}
2245
2246static int f2fs_write_node_pages(struct address_space *mapping,
2247 struct writeback_control *wbc)
2248{
2249 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2250 struct blk_plug plug;
2251 long diff;
2252
2253 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2254 goto skip_write;
2255
2256 /* balancing f2fs's metadata in background */
2257 f2fs_balance_fs_bg(sbi, true);
2258
2259 /* collect a number of dirty node pages and write together */
2260 if (wbc->sync_mode != WB_SYNC_ALL &&
2261 get_pages(sbi, F2FS_DIRTY_NODES) <
2262 nr_pages_to_skip(sbi, NODE))
2263 goto skip_write;
2264
2265 if (wbc->sync_mode == WB_SYNC_ALL)
2266 atomic_inc(&sbi->wb_sync_req[NODE]);
2267 else if (atomic_read(&sbi->wb_sync_req[NODE])) {
2268 /* to avoid potential deadlock */
2269 if (current->plug)
2270 blk_finish_plug(current->plug);
2271 goto skip_write;
2272 }
2273
2274 trace_f2fs_writepages(mapping->host, wbc, NODE);
2275
2276 diff = nr_pages_to_write(sbi, NODE, wbc);
2277 blk_start_plug(&plug);
2278 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2279 blk_finish_plug(&plug);
2280 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2281
2282 if (wbc->sync_mode == WB_SYNC_ALL)
2283 atomic_dec(&sbi->wb_sync_req[NODE]);
2284 return 0;
2285
2286skip_write:
2287 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2288 trace_f2fs_writepages(mapping->host, wbc, NODE);
2289 return 0;
2290}
2291
2292static bool f2fs_dirty_node_folio(struct address_space *mapping,
2293 struct folio *folio)
2294{
2295 trace_f2fs_set_page_dirty(folio, NODE);
2296
2297 if (!folio_test_uptodate(folio))
2298 folio_mark_uptodate(folio);
2299#ifdef CONFIG_F2FS_CHECK_FS
2300 if (IS_INODE(folio))
2301 f2fs_inode_chksum_set(F2FS_M_SB(mapping), folio);
2302#endif
2303 if (filemap_dirty_folio(mapping, folio)) {
2304 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
2305 folio_set_f2fs_reference(folio);
2306 return true;
2307 }
2308 return false;
2309}
2310
2311/*
2312 * Structure of the f2fs node operations
2313 */
2314const struct address_space_operations f2fs_node_aops = {
2315 .writepages = f2fs_write_node_pages,
2316 .dirty_folio = f2fs_dirty_node_folio,
2317 .invalidate_folio = f2fs_invalidate_folio,
2318 .release_folio = f2fs_release_folio,
2319 .migrate_folio = filemap_migrate_folio,
2320};
2321
2322static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2323 nid_t n)
2324{
2325 return radix_tree_lookup(&nm_i->free_nid_root, n);
2326}
2327
2328static int __insert_free_nid(struct f2fs_sb_info *sbi,
2329 struct free_nid *i)
2330{
2331 struct f2fs_nm_info *nm_i = NM_I(sbi);
2332 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2333
2334 if (err)
2335 return err;
2336
2337 nm_i->nid_cnt[FREE_NID]++;
2338 list_add_tail(&i->list, &nm_i->free_nid_list);
2339 return 0;
2340}
2341
2342static void __remove_free_nid(struct f2fs_sb_info *sbi,
2343 struct free_nid *i, enum nid_state state)
2344{
2345 struct f2fs_nm_info *nm_i = NM_I(sbi);
2346
2347 f2fs_bug_on(sbi, state != i->state);
2348 nm_i->nid_cnt[state]--;
2349 if (state == FREE_NID)
2350 list_del(&i->list);
2351 radix_tree_delete(&nm_i->free_nid_root, i->nid);
2352}
2353
2354static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2355 enum nid_state org_state, enum nid_state dst_state)
2356{
2357 struct f2fs_nm_info *nm_i = NM_I(sbi);
2358
2359 f2fs_bug_on(sbi, org_state != i->state);
2360 i->state = dst_state;
2361 nm_i->nid_cnt[org_state]--;
2362 nm_i->nid_cnt[dst_state]++;
2363
2364 switch (dst_state) {
2365 case PREALLOC_NID:
2366 list_del(&i->list);
2367 break;
2368 case FREE_NID:
2369 list_add_tail(&i->list, &nm_i->free_nid_list);
2370 break;
2371 default:
2372 BUG_ON(1);
2373 }
2374}
2375
2376static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2377 bool set, bool build)
2378{
2379 struct f2fs_nm_info *nm_i = NM_I(sbi);
2380 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2381 unsigned int nid_ofs = nid - START_NID(nid);
2382
2383 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2384 return;
2385
2386 if (set) {
2387 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2388 return;
2389 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2390 nm_i->free_nid_count[nat_ofs]++;
2391 } else {
2392 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2393 return;
2394 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2395 if (!build)
2396 nm_i->free_nid_count[nat_ofs]--;
2397 }
2398}
2399
2400/* return if the nid is recognized as free */
2401static bool add_free_nid(struct f2fs_sb_info *sbi,
2402 nid_t nid, bool build, bool update)
2403{
2404 struct f2fs_nm_info *nm_i = NM_I(sbi);
2405 struct free_nid *i, *e;
2406 struct nat_entry *ne;
2407 int err;
2408 bool ret = false;
2409
2410 /* 0 nid should not be used */
2411 if (unlikely(nid == 0))
2412 return false;
2413
2414 if (unlikely(f2fs_check_nid_range(sbi, nid)))
2415 return false;
2416
2417 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL);
2418 i->nid = nid;
2419 i->state = FREE_NID;
2420
2421 err = radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2422 f2fs_bug_on(sbi, err);
2423
2424 err = -EINVAL;
2425
2426 spin_lock(&nm_i->nid_list_lock);
2427
2428 if (build) {
2429 /*
2430 * Thread A Thread B
2431 * - f2fs_create
2432 * - f2fs_new_inode
2433 * - f2fs_alloc_nid
2434 * - __insert_nid_to_list(PREALLOC_NID)
2435 * - f2fs_balance_fs_bg
2436 * - f2fs_build_free_nids
2437 * - __f2fs_build_free_nids
2438 * - scan_nat_page
2439 * - add_free_nid
2440 * - __lookup_nat_cache
2441 * - f2fs_add_link
2442 * - f2fs_init_inode_metadata
2443 * - f2fs_new_inode_folio
2444 * - f2fs_new_node_folio
2445 * - set_node_addr
2446 * - f2fs_alloc_nid_done
2447 * - __remove_nid_from_list(PREALLOC_NID)
2448 * - __insert_nid_to_list(FREE_NID)
2449 */
2450 ne = __lookup_nat_cache(nm_i, nid, false);
2451 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2452 nat_get_blkaddr(ne) != NULL_ADDR))
2453 goto err_out;
2454
2455 e = __lookup_free_nid_list(nm_i, nid);
2456 if (e) {
2457 if (e->state == FREE_NID)
2458 ret = true;
2459 goto err_out;
2460 }
2461 }
2462 ret = true;
2463 err = __insert_free_nid(sbi, i);
2464err_out:
2465 if (update) {
2466 update_free_nid_bitmap(sbi, nid, ret, build);
2467 if (!build)
2468 nm_i->available_nids++;
2469 }
2470 spin_unlock(&nm_i->nid_list_lock);
2471 radix_tree_preload_end();
2472
2473 if (err)
2474 kmem_cache_free(free_nid_slab, i);
2475 return ret;
2476}
2477
2478static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2479{
2480 struct f2fs_nm_info *nm_i = NM_I(sbi);
2481 struct free_nid *i;
2482 bool need_free = false;
2483
2484 spin_lock(&nm_i->nid_list_lock);
2485 i = __lookup_free_nid_list(nm_i, nid);
2486 if (i && i->state == FREE_NID) {
2487 __remove_free_nid(sbi, i, FREE_NID);
2488 need_free = true;
2489 }
2490 spin_unlock(&nm_i->nid_list_lock);
2491
2492 if (need_free)
2493 kmem_cache_free(free_nid_slab, i);
2494}
2495
2496static int scan_nat_page(struct f2fs_sb_info *sbi,
2497 struct f2fs_nat_block *nat_blk, nid_t start_nid)
2498{
2499 struct f2fs_nm_info *nm_i = NM_I(sbi);
2500 block_t blk_addr;
2501 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2502 int i;
2503
2504 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2505
2506 i = start_nid % NAT_ENTRY_PER_BLOCK;
2507
2508 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2509 if (unlikely(start_nid >= nm_i->max_nid))
2510 break;
2511
2512 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2513
2514 if (blk_addr == NEW_ADDR)
2515 return -EFSCORRUPTED;
2516
2517 if (blk_addr == NULL_ADDR) {
2518 add_free_nid(sbi, start_nid, true, true);
2519 } else {
2520 spin_lock(&NM_I(sbi)->nid_list_lock);
2521 update_free_nid_bitmap(sbi, start_nid, false, true);
2522 spin_unlock(&NM_I(sbi)->nid_list_lock);
2523 }
2524 }
2525
2526 return 0;
2527}
2528
2529static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2530{
2531 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2532 struct f2fs_journal *journal = curseg->journal;
2533 int i;
2534
2535 down_read(&curseg->journal_rwsem);
2536 for (i = 0; i < nats_in_cursum(journal); i++) {
2537 block_t addr;
2538 nid_t nid;
2539
2540 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2541 nid = le32_to_cpu(nid_in_journal(journal, i));
2542 if (addr == NULL_ADDR)
2543 add_free_nid(sbi, nid, true, false);
2544 else
2545 remove_free_nid(sbi, nid);
2546 }
2547 up_read(&curseg->journal_rwsem);
2548}
2549
2550static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2551{
2552 struct f2fs_nm_info *nm_i = NM_I(sbi);
2553 unsigned int i, idx;
2554 nid_t nid;
2555
2556 f2fs_down_read(&nm_i->nat_tree_lock);
2557
2558 for (i = 0; i < nm_i->nat_blocks; i++) {
2559 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2560 continue;
2561 if (!nm_i->free_nid_count[i])
2562 continue;
2563 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2564 idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2565 NAT_ENTRY_PER_BLOCK, idx);
2566 if (idx >= NAT_ENTRY_PER_BLOCK)
2567 break;
2568
2569 nid = i * NAT_ENTRY_PER_BLOCK + idx;
2570 add_free_nid(sbi, nid, true, false);
2571
2572 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2573 goto out;
2574 }
2575 }
2576out:
2577 scan_curseg_cache(sbi);
2578
2579 f2fs_up_read(&nm_i->nat_tree_lock);
2580}
2581
2582static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2583 bool sync, bool mount)
2584{
2585 struct f2fs_nm_info *nm_i = NM_I(sbi);
2586 int i = 0, ret;
2587 nid_t nid = nm_i->next_scan_nid;
2588
2589 if (unlikely(nid >= nm_i->max_nid))
2590 nid = 0;
2591
2592 if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2593 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2594
2595 /* Enough entries */
2596 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2597 return 0;
2598
2599 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2600 return 0;
2601
2602 if (!mount) {
2603 /* try to find free nids in free_nid_bitmap */
2604 scan_free_nid_bits(sbi);
2605
2606 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2607 return 0;
2608 }
2609
2610 /* readahead nat pages to be scanned */
2611 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2612 META_NAT, true);
2613
2614 f2fs_down_read(&nm_i->nat_tree_lock);
2615
2616 while (1) {
2617 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2618 nm_i->nat_block_bitmap)) {
2619 struct folio *folio = get_current_nat_folio(sbi, nid);
2620
2621 if (IS_ERR(folio)) {
2622 ret = PTR_ERR(folio);
2623 } else {
2624 ret = scan_nat_page(sbi, folio_address(folio),
2625 nid);
2626 f2fs_folio_put(folio, true);
2627 }
2628
2629 if (ret) {
2630 f2fs_up_read(&nm_i->nat_tree_lock);
2631
2632 if (ret == -EFSCORRUPTED) {
2633 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2634 set_sbi_flag(sbi, SBI_NEED_FSCK);
2635 f2fs_handle_error(sbi,
2636 ERROR_INCONSISTENT_NAT);
2637 }
2638
2639 return ret;
2640 }
2641 }
2642
2643 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2644 if (unlikely(nid >= nm_i->max_nid))
2645 nid = 0;
2646
2647 if (++i >= FREE_NID_PAGES)
2648 break;
2649 }
2650
2651 /* go to the next free nat pages to find free nids abundantly */
2652 nm_i->next_scan_nid = nid;
2653
2654 /* find free nids from current sum_pages */
2655 scan_curseg_cache(sbi);
2656
2657 f2fs_up_read(&nm_i->nat_tree_lock);
2658
2659 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2660 nm_i->ra_nid_pages, META_NAT, false);
2661
2662 return 0;
2663}
2664
2665int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2666{
2667 int ret;
2668
2669 mutex_lock(&NM_I(sbi)->build_lock);
2670 ret = __f2fs_build_free_nids(sbi, sync, mount);
2671 mutex_unlock(&NM_I(sbi)->build_lock);
2672
2673 return ret;
2674}
2675
2676/*
2677 * If this function returns success, caller can obtain a new nid
2678 * from second parameter of this function.
2679 * The returned nid could be used ino as well as nid when inode is created.
2680 */
2681bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2682{
2683 struct f2fs_nm_info *nm_i = NM_I(sbi);
2684 struct free_nid *i = NULL;
2685retry:
2686 if (time_to_inject(sbi, FAULT_ALLOC_NID))
2687 return false;
2688
2689 spin_lock(&nm_i->nid_list_lock);
2690
2691 if (unlikely(nm_i->available_nids == 0)) {
2692 spin_unlock(&nm_i->nid_list_lock);
2693 return false;
2694 }
2695
2696 /* We should not use stale free nids created by f2fs_build_free_nids */
2697 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2698 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2699 i = list_first_entry(&nm_i->free_nid_list,
2700 struct free_nid, list);
2701
2702 if (unlikely(is_invalid_nid(sbi, i->nid))) {
2703 spin_unlock(&nm_i->nid_list_lock);
2704 f2fs_err(sbi, "Corrupted nid %u in free_nid_list",
2705 i->nid);
2706 f2fs_stop_checkpoint(sbi, false,
2707 STOP_CP_REASON_CORRUPTED_NID);
2708 return false;
2709 }
2710
2711 *nid = i->nid;
2712
2713 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2714 nm_i->available_nids--;
2715
2716 update_free_nid_bitmap(sbi, *nid, false, false);
2717
2718 spin_unlock(&nm_i->nid_list_lock);
2719 return true;
2720 }
2721 spin_unlock(&nm_i->nid_list_lock);
2722
2723 /* Let's scan nat pages and its caches to get free nids */
2724 if (!f2fs_build_free_nids(sbi, true, false))
2725 goto retry;
2726 return false;
2727}
2728
2729/*
2730 * f2fs_alloc_nid() should be called prior to this function.
2731 */
2732void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2733{
2734 struct f2fs_nm_info *nm_i = NM_I(sbi);
2735 struct free_nid *i;
2736
2737 spin_lock(&nm_i->nid_list_lock);
2738 i = __lookup_free_nid_list(nm_i, nid);
2739 f2fs_bug_on(sbi, !i);
2740 __remove_free_nid(sbi, i, PREALLOC_NID);
2741 spin_unlock(&nm_i->nid_list_lock);
2742
2743 kmem_cache_free(free_nid_slab, i);
2744}
2745
2746/*
2747 * f2fs_alloc_nid() should be called prior to this function.
2748 */
2749void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2750{
2751 struct f2fs_nm_info *nm_i = NM_I(sbi);
2752 struct free_nid *i;
2753 bool need_free = false;
2754
2755 if (!nid)
2756 return;
2757
2758 spin_lock(&nm_i->nid_list_lock);
2759 i = __lookup_free_nid_list(nm_i, nid);
2760 f2fs_bug_on(sbi, !i);
2761
2762 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2763 __remove_free_nid(sbi, i, PREALLOC_NID);
2764 need_free = true;
2765 } else {
2766 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2767 }
2768
2769 nm_i->available_nids++;
2770
2771 update_free_nid_bitmap(sbi, nid, true, false);
2772
2773 spin_unlock(&nm_i->nid_list_lock);
2774
2775 if (need_free)
2776 kmem_cache_free(free_nid_slab, i);
2777}
2778
2779int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2780{
2781 struct f2fs_nm_info *nm_i = NM_I(sbi);
2782 int nr = nr_shrink;
2783
2784 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2785 return 0;
2786
2787 if (!mutex_trylock(&nm_i->build_lock))
2788 return 0;
2789
2790 while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2791 struct free_nid *i, *next;
2792 unsigned int batch = SHRINK_NID_BATCH_SIZE;
2793
2794 spin_lock(&nm_i->nid_list_lock);
2795 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2796 if (!nr_shrink || !batch ||
2797 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2798 break;
2799 __remove_free_nid(sbi, i, FREE_NID);
2800 kmem_cache_free(free_nid_slab, i);
2801 nr_shrink--;
2802 batch--;
2803 }
2804 spin_unlock(&nm_i->nid_list_lock);
2805 }
2806
2807 mutex_unlock(&nm_i->build_lock);
2808
2809 return nr - nr_shrink;
2810}
2811
2812int f2fs_recover_inline_xattr(struct inode *inode, struct folio *folio)
2813{
2814 void *src_addr, *dst_addr;
2815 size_t inline_size;
2816 struct folio *ifolio;
2817 struct f2fs_inode *ri;
2818
2819 ifolio = f2fs_get_inode_folio(F2FS_I_SB(inode), inode->i_ino);
2820 if (IS_ERR(ifolio))
2821 return PTR_ERR(ifolio);
2822
2823 ri = F2FS_INODE(folio);
2824 if (ri->i_inline & F2FS_INLINE_XATTR) {
2825 if (!f2fs_has_inline_xattr(inode)) {
2826 set_inode_flag(inode, FI_INLINE_XATTR);
2827 stat_inc_inline_xattr(inode);
2828 }
2829 } else {
2830 if (f2fs_has_inline_xattr(inode)) {
2831 stat_dec_inline_xattr(inode);
2832 clear_inode_flag(inode, FI_INLINE_XATTR);
2833 }
2834 goto update_inode;
2835 }
2836
2837 dst_addr = inline_xattr_addr(inode, ifolio);
2838 src_addr = inline_xattr_addr(inode, folio);
2839 inline_size = inline_xattr_size(inode);
2840
2841 f2fs_folio_wait_writeback(ifolio, NODE, true, true);
2842 memcpy(dst_addr, src_addr, inline_size);
2843update_inode:
2844 f2fs_update_inode(inode, ifolio);
2845 f2fs_folio_put(ifolio, true);
2846 return 0;
2847}
2848
2849int f2fs_recover_xattr_data(struct inode *inode, struct folio *folio)
2850{
2851 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2852 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2853 nid_t new_xnid;
2854 struct dnode_of_data dn;
2855 struct node_info ni;
2856 struct folio *xfolio;
2857 int err;
2858
2859 if (!prev_xnid)
2860 goto recover_xnid;
2861
2862 /* 1: invalidate the previous xattr nid */
2863 err = f2fs_get_node_info(sbi, prev_xnid, &ni, false);
2864 if (err)
2865 return err;
2866
2867 f2fs_invalidate_blocks(sbi, ni.blk_addr, 1);
2868 dec_valid_node_count(sbi, inode, false);
2869 set_node_addr(sbi, &ni, NULL_ADDR, false);
2870
2871recover_xnid:
2872 /* 2: update xattr nid in inode */
2873 if (!f2fs_alloc_nid(sbi, &new_xnid))
2874 return -ENOSPC;
2875
2876 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2877 xfolio = f2fs_new_node_folio(&dn, XATTR_NODE_OFFSET);
2878 if (IS_ERR(xfolio)) {
2879 f2fs_alloc_nid_failed(sbi, new_xnid);
2880 return PTR_ERR(xfolio);
2881 }
2882
2883 f2fs_alloc_nid_done(sbi, new_xnid);
2884 f2fs_update_inode_page(inode);
2885
2886 /* 3: update and set xattr node page dirty */
2887 if (folio) {
2888 memcpy(F2FS_NODE(xfolio), F2FS_NODE(folio),
2889 VALID_XATTR_BLOCK_SIZE);
2890 folio_mark_dirty(xfolio);
2891 }
2892 f2fs_folio_put(xfolio, true);
2893
2894 return 0;
2895}
2896
2897int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct folio *folio)
2898{
2899 struct f2fs_inode *src, *dst;
2900 nid_t ino = ino_of_node(folio);
2901 struct node_info old_ni, new_ni;
2902 struct folio *ifolio;
2903 int err;
2904
2905 err = f2fs_get_node_info(sbi, ino, &old_ni, false);
2906 if (err)
2907 return err;
2908
2909 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2910 return -EINVAL;
2911retry:
2912 ifolio = f2fs_grab_cache_folio(NODE_MAPPING(sbi), ino, false);
2913 if (IS_ERR(ifolio)) {
2914 memalloc_retry_wait(GFP_NOFS);
2915 goto retry;
2916 }
2917
2918 /* Should not use this inode from free nid list */
2919 remove_free_nid(sbi, ino);
2920
2921 if (!folio_test_uptodate(ifolio))
2922 folio_mark_uptodate(ifolio);
2923 fill_node_footer(ifolio, ino, ino, 0, true);
2924 set_cold_node(ifolio, false);
2925
2926 src = F2FS_INODE(folio);
2927 dst = F2FS_INODE(ifolio);
2928
2929 memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2930 dst->i_size = 0;
2931 dst->i_blocks = cpu_to_le64(1);
2932 dst->i_links = cpu_to_le32(1);
2933 dst->i_xattr_nid = 0;
2934 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2935 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2936 dst->i_extra_isize = src->i_extra_isize;
2937
2938 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2939 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2940 i_inline_xattr_size))
2941 dst->i_inline_xattr_size = src->i_inline_xattr_size;
2942
2943 if (f2fs_sb_has_project_quota(sbi) &&
2944 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2945 i_projid))
2946 dst->i_projid = src->i_projid;
2947
2948 if (f2fs_sb_has_inode_crtime(sbi) &&
2949 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2950 i_crtime_nsec)) {
2951 dst->i_crtime = src->i_crtime;
2952 dst->i_crtime_nsec = src->i_crtime_nsec;
2953 }
2954 }
2955
2956 new_ni = old_ni;
2957 new_ni.ino = ino;
2958
2959 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2960 WARN_ON(1);
2961 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2962 inc_valid_inode_count(sbi);
2963 folio_mark_dirty(ifolio);
2964 f2fs_folio_put(ifolio, true);
2965 return 0;
2966}
2967
2968int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2969 unsigned int segno, struct f2fs_summary_block *sum)
2970{
2971 struct f2fs_node *rn;
2972 struct f2fs_summary *sum_entry;
2973 block_t addr;
2974 int i, idx, last_offset, nrpages;
2975
2976 /* scan the node segment */
2977 last_offset = BLKS_PER_SEG(sbi);
2978 addr = START_BLOCK(sbi, segno);
2979 sum_entry = sum_entries(sum);
2980
2981 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2982 nrpages = bio_max_segs(last_offset - i);
2983
2984 /* readahead node pages */
2985 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2986
2987 for (idx = addr; idx < addr + nrpages; idx++) {
2988 struct folio *folio = f2fs_get_tmp_folio(sbi, idx);
2989
2990 if (IS_ERR(folio))
2991 return PTR_ERR(folio);
2992
2993 rn = F2FS_NODE(folio);
2994 sum_entry->nid = rn->footer.nid;
2995 sum_entry->version = 0;
2996 sum_entry->ofs_in_node = 0;
2997 sum_entry++;
2998 f2fs_folio_put(folio, true);
2999 }
3000
3001 invalidate_mapping_pages(META_MAPPING(sbi), addr,
3002 addr + nrpages);
3003 }
3004 return 0;
3005}
3006
3007static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
3008{
3009 struct f2fs_nm_info *nm_i = NM_I(sbi);
3010 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3011 struct f2fs_journal *journal = curseg->journal;
3012 int i;
3013 bool init_dirty;
3014
3015 down_write(&curseg->journal_rwsem);
3016 for (i = 0; i < nats_in_cursum(journal); i++) {
3017 struct nat_entry *ne;
3018 struct f2fs_nat_entry raw_ne;
3019 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
3020
3021 if (f2fs_check_nid_range(sbi, nid))
3022 continue;
3023
3024 init_dirty = false;
3025
3026 raw_ne = nat_in_journal(journal, i);
3027
3028 ne = __lookup_nat_cache(nm_i, nid, true);
3029 if (!ne) {
3030 init_dirty = true;
3031 ne = __alloc_nat_entry(sbi, nid, true);
3032 __init_nat_entry(nm_i, ne, &raw_ne, true, true);
3033 }
3034
3035 /*
3036 * if a free nat in journal has not been used after last
3037 * checkpoint, we should remove it from available nids,
3038 * since later we will add it again.
3039 */
3040 if (!get_nat_flag(ne, IS_DIRTY) &&
3041 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
3042 spin_lock(&nm_i->nid_list_lock);
3043 nm_i->available_nids--;
3044 spin_unlock(&nm_i->nid_list_lock);
3045 }
3046
3047 __set_nat_cache_dirty(nm_i, ne, init_dirty);
3048 }
3049 update_nats_in_cursum(journal, -i);
3050 up_write(&curseg->journal_rwsem);
3051}
3052
3053static void __adjust_nat_entry_set(struct nat_entry_set *nes,
3054 struct list_head *head, int max)
3055{
3056 struct nat_entry_set *cur;
3057
3058 if (nes->entry_cnt >= max)
3059 goto add_out;
3060
3061 list_for_each_entry(cur, head, set_list) {
3062 if (cur->entry_cnt >= nes->entry_cnt) {
3063 list_add(&nes->set_list, cur->set_list.prev);
3064 return;
3065 }
3066 }
3067add_out:
3068 list_add_tail(&nes->set_list, head);
3069}
3070
3071static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
3072 const struct f2fs_nat_block *nat_blk)
3073{
3074 struct f2fs_nm_info *nm_i = NM_I(sbi);
3075 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
3076 int valid = 0;
3077 int i = 0;
3078
3079 if (!enabled_nat_bits(sbi, NULL))
3080 return;
3081
3082 if (nat_index == 0) {
3083 valid = 1;
3084 i = 1;
3085 }
3086 for (; i < NAT_ENTRY_PER_BLOCK; i++) {
3087 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
3088 valid++;
3089 }
3090 if (valid == 0) {
3091 __set_bit_le(nat_index, nm_i->empty_nat_bits);
3092 __clear_bit_le(nat_index, nm_i->full_nat_bits);
3093 return;
3094 }
3095
3096 __clear_bit_le(nat_index, nm_i->empty_nat_bits);
3097 if (valid == NAT_ENTRY_PER_BLOCK)
3098 __set_bit_le(nat_index, nm_i->full_nat_bits);
3099 else
3100 __clear_bit_le(nat_index, nm_i->full_nat_bits);
3101}
3102
3103static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
3104 struct nat_entry_set *set, struct cp_control *cpc)
3105{
3106 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3107 struct f2fs_journal *journal = curseg->journal;
3108 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
3109 bool to_journal = true;
3110 struct f2fs_nat_block *nat_blk;
3111 struct nat_entry *ne, *cur;
3112 struct folio *folio = NULL;
3113
3114 /*
3115 * there are two steps to flush nat entries:
3116 * #1, flush nat entries to journal in current hot data summary block.
3117 * #2, flush nat entries to nat page.
3118 */
3119 if (enabled_nat_bits(sbi, cpc) ||
3120 !__has_cursum_space(sbi, journal, set->entry_cnt, NAT_JOURNAL))
3121 to_journal = false;
3122
3123 if (to_journal) {
3124 down_write(&curseg->journal_rwsem);
3125 } else {
3126 folio = get_next_nat_folio(sbi, start_nid);
3127 if (IS_ERR(folio))
3128 return PTR_ERR(folio);
3129
3130 nat_blk = folio_address(folio);
3131 f2fs_bug_on(sbi, !nat_blk);
3132 }
3133
3134 /* flush dirty nats in nat entry set */
3135 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
3136 struct f2fs_nat_entry *raw_ne;
3137 nid_t nid = nat_get_nid(ne);
3138 int offset;
3139
3140 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
3141
3142 if (to_journal) {
3143 offset = f2fs_lookup_journal_in_cursum(sbi, journal,
3144 NAT_JOURNAL, nid, 1);
3145 f2fs_bug_on(sbi, offset < 0);
3146 raw_ne = &nat_in_journal(journal, offset);
3147 nid_in_journal(journal, offset) = cpu_to_le32(nid);
3148 } else {
3149 raw_ne = &nat_blk->entries[nid - start_nid];
3150 }
3151 raw_nat_from_node_info(raw_ne, &ne->ni);
3152 nat_reset_flag(ne);
3153 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
3154 if (nat_get_blkaddr(ne) == NULL_ADDR) {
3155 add_free_nid(sbi, nid, false, true);
3156 } else {
3157 spin_lock(&NM_I(sbi)->nid_list_lock);
3158 update_free_nid_bitmap(sbi, nid, false, false);
3159 spin_unlock(&NM_I(sbi)->nid_list_lock);
3160 }
3161 }
3162
3163 if (to_journal) {
3164 up_write(&curseg->journal_rwsem);
3165 } else {
3166 __update_nat_bits(sbi, start_nid, nat_blk);
3167 f2fs_folio_put(folio, true);
3168 }
3169
3170 /* Allow dirty nats by node block allocation in write_begin */
3171 if (!set->entry_cnt) {
3172 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
3173 kmem_cache_free(nat_entry_set_slab, set);
3174 }
3175 return 0;
3176}
3177
3178/*
3179 * This function is called during the checkpointing process.
3180 */
3181int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3182{
3183 struct f2fs_nm_info *nm_i = NM_I(sbi);
3184 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3185 struct f2fs_journal *journal = curseg->journal;
3186 struct nat_entry_set *setvec[NAT_VEC_SIZE];
3187 struct nat_entry_set *set, *tmp;
3188 unsigned int found, entry_count = 0;
3189 nid_t set_idx = 0;
3190 LIST_HEAD(sets);
3191 int err = 0;
3192
3193 /*
3194 * during unmount, let's flush nat_bits before checking
3195 * nat_cnt[DIRTY_NAT].
3196 */
3197 if (enabled_nat_bits(sbi, cpc)) {
3198 f2fs_down_write(&nm_i->nat_tree_lock);
3199 remove_nats_in_journal(sbi);
3200 f2fs_up_write(&nm_i->nat_tree_lock);
3201 }
3202
3203 if (!nm_i->nat_cnt[DIRTY_NAT])
3204 return 0;
3205
3206 f2fs_down_write(&nm_i->nat_tree_lock);
3207
3208 /*
3209 * if there are no enough space in journal to store dirty nat
3210 * entries, remove all entries from journal and merge them
3211 * into nat entry set.
3212 */
3213 if (enabled_nat_bits(sbi, cpc) ||
3214 !__has_cursum_space(sbi, journal,
3215 nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3216 remove_nats_in_journal(sbi);
3217
3218 while ((found = __gang_lookup_nat_set(nm_i,
3219 set_idx, NAT_VEC_SIZE, setvec))) {
3220 unsigned idx;
3221
3222 set_idx = setvec[found - 1]->set + 1;
3223 for (idx = 0; idx < found; idx++)
3224 __adjust_nat_entry_set(setvec[idx], &sets,
3225 MAX_NAT_JENTRIES(sbi, journal));
3226 }
3227
3228 /*
3229 * Readahead the current NAT block to prevent read requests from
3230 * being issued and waited on one by one.
3231 */
3232 list_for_each_entry(set, &sets, set_list) {
3233 entry_count += set->entry_cnt;
3234 if (!enabled_nat_bits(sbi, cpc) &&
3235 __has_cursum_space(sbi, journal,
3236 entry_count, NAT_JOURNAL))
3237 continue;
3238 f2fs_ra_meta_pages(sbi, set->set, 1, META_NAT, true);
3239 }
3240 /* flush dirty nats in nat entry set */
3241 list_for_each_entry_safe(set, tmp, &sets, set_list) {
3242 err = __flush_nat_entry_set(sbi, set, cpc);
3243 if (err)
3244 break;
3245 }
3246
3247 f2fs_up_write(&nm_i->nat_tree_lock);
3248 /* Allow dirty nats by node block allocation in write_begin */
3249
3250 return err;
3251}
3252
3253static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3254{
3255 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3256 struct f2fs_nm_info *nm_i = NM_I(sbi);
3257 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3258 unsigned int i;
3259 __u64 cp_ver = cur_cp_version(ckpt);
3260 block_t nat_bits_addr;
3261
3262 if (!enabled_nat_bits(sbi, NULL))
3263 return 0;
3264
3265 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3266 nm_i->nat_bits = f2fs_kvzalloc(sbi,
3267 F2FS_BLK_TO_BYTES(nm_i->nat_bits_blocks), GFP_KERNEL);
3268 if (!nm_i->nat_bits)
3269 return -ENOMEM;
3270
3271 nat_bits_addr = __start_cp_addr(sbi) + BLKS_PER_SEG(sbi) -
3272 nm_i->nat_bits_blocks;
3273 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3274 struct folio *folio;
3275
3276 folio = f2fs_get_meta_folio(sbi, nat_bits_addr++);
3277 if (IS_ERR(folio))
3278 return PTR_ERR(folio);
3279
3280 memcpy(nm_i->nat_bits + F2FS_BLK_TO_BYTES(i),
3281 folio_address(folio), F2FS_BLKSIZE);
3282 f2fs_folio_put(folio, true);
3283 }
3284
3285 cp_ver |= (cur_cp_crc(ckpt) << 32);
3286 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3287 disable_nat_bits(sbi, true);
3288 return 0;
3289 }
3290
3291 nm_i->full_nat_bits = nm_i->nat_bits + 8;
3292 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3293
3294 f2fs_notice(sbi, "Found nat_bits in checkpoint");
3295 return 0;
3296}
3297
3298static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3299{
3300 struct f2fs_nm_info *nm_i = NM_I(sbi);
3301 unsigned int i = 0;
3302 nid_t nid, last_nid;
3303
3304 if (!enabled_nat_bits(sbi, NULL))
3305 return;
3306
3307 for (i = 0; i < nm_i->nat_blocks; i++) {
3308 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3309 if (i >= nm_i->nat_blocks)
3310 break;
3311
3312 __set_bit_le(i, nm_i->nat_block_bitmap);
3313
3314 nid = i * NAT_ENTRY_PER_BLOCK;
3315 last_nid = nid + NAT_ENTRY_PER_BLOCK;
3316
3317 spin_lock(&NM_I(sbi)->nid_list_lock);
3318 for (; nid < last_nid; nid++)
3319 update_free_nid_bitmap(sbi, nid, true, true);
3320 spin_unlock(&NM_I(sbi)->nid_list_lock);
3321 }
3322
3323 for (i = 0; i < nm_i->nat_blocks; i++) {
3324 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3325 if (i >= nm_i->nat_blocks)
3326 break;
3327
3328 __set_bit_le(i, nm_i->nat_block_bitmap);
3329 }
3330}
3331
3332static int init_node_manager(struct f2fs_sb_info *sbi)
3333{
3334 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3335 struct f2fs_nm_info *nm_i = NM_I(sbi);
3336 unsigned char *version_bitmap;
3337 unsigned int nat_segs;
3338 int err;
3339
3340 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3341
3342 /* segment_count_nat includes pair segment so divide to 2. */
3343 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3344 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3345 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3346
3347 /* not used nids: 0, node, meta, (and root counted as valid node) */
3348 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3349 F2FS_RESERVED_NODE_NUM;
3350 nm_i->nid_cnt[FREE_NID] = 0;
3351 nm_i->nid_cnt[PREALLOC_NID] = 0;
3352 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3353 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3354 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3355 nm_i->max_rf_node_blocks = DEF_RF_NODE_BLOCKS;
3356
3357 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3358 INIT_LIST_HEAD(&nm_i->free_nid_list);
3359 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3360 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3361 INIT_LIST_HEAD(&nm_i->nat_entries);
3362 spin_lock_init(&nm_i->nat_list_lock);
3363
3364 mutex_init(&nm_i->build_lock);
3365 spin_lock_init(&nm_i->nid_list_lock);
3366 init_f2fs_rwsem(&nm_i->nat_tree_lock);
3367
3368 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3369 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3370 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3371 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3372 GFP_KERNEL);
3373 if (!nm_i->nat_bitmap)
3374 return -ENOMEM;
3375
3376 if (!test_opt(sbi, NAT_BITS))
3377 disable_nat_bits(sbi, true);
3378
3379 err = __get_nat_bitmaps(sbi);
3380 if (err)
3381 return err;
3382
3383#ifdef CONFIG_F2FS_CHECK_FS
3384 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3385 GFP_KERNEL);
3386 if (!nm_i->nat_bitmap_mir)
3387 return -ENOMEM;
3388#endif
3389
3390 return 0;
3391}
3392
3393static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3394{
3395 struct f2fs_nm_info *nm_i = NM_I(sbi);
3396 int i;
3397
3398 nm_i->free_nid_bitmap =
3399 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3400 nm_i->nat_blocks),
3401 GFP_KERNEL);
3402 if (!nm_i->free_nid_bitmap)
3403 return -ENOMEM;
3404
3405 for (i = 0; i < nm_i->nat_blocks; i++) {
3406 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3407 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3408 if (!nm_i->free_nid_bitmap[i])
3409 return -ENOMEM;
3410 }
3411
3412 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3413 GFP_KERNEL);
3414 if (!nm_i->nat_block_bitmap)
3415 return -ENOMEM;
3416
3417 nm_i->free_nid_count =
3418 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3419 nm_i->nat_blocks),
3420 GFP_KERNEL);
3421 if (!nm_i->free_nid_count)
3422 return -ENOMEM;
3423 return 0;
3424}
3425
3426int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3427{
3428 int err;
3429
3430 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3431 GFP_KERNEL);
3432 if (!sbi->nm_info)
3433 return -ENOMEM;
3434
3435 err = init_node_manager(sbi);
3436 if (err)
3437 return err;
3438
3439 err = init_free_nid_cache(sbi);
3440 if (err)
3441 return err;
3442
3443 /* load free nid status from nat_bits table */
3444 load_free_nid_bitmap(sbi);
3445
3446 return f2fs_build_free_nids(sbi, true, true);
3447}
3448
3449void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3450{
3451 struct f2fs_nm_info *nm_i = NM_I(sbi);
3452 struct free_nid *i, *next_i;
3453 void *vec[NAT_VEC_SIZE];
3454 struct nat_entry **natvec = (struct nat_entry **)vec;
3455 struct nat_entry_set **setvec = (struct nat_entry_set **)vec;
3456 nid_t nid = 0;
3457 unsigned int found;
3458
3459 if (!nm_i)
3460 return;
3461
3462 /* destroy free nid list */
3463 spin_lock(&nm_i->nid_list_lock);
3464 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3465 __remove_free_nid(sbi, i, FREE_NID);
3466 spin_unlock(&nm_i->nid_list_lock);
3467 kmem_cache_free(free_nid_slab, i);
3468 spin_lock(&nm_i->nid_list_lock);
3469 }
3470 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3471 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3472 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3473 spin_unlock(&nm_i->nid_list_lock);
3474
3475 /* destroy nat cache */
3476 f2fs_down_write(&nm_i->nat_tree_lock);
3477 while ((found = __gang_lookup_nat_cache(nm_i,
3478 nid, NAT_VEC_SIZE, natvec))) {
3479 unsigned idx;
3480
3481 nid = nat_get_nid(natvec[found - 1]) + 1;
3482 for (idx = 0; idx < found; idx++) {
3483 spin_lock(&nm_i->nat_list_lock);
3484 list_del(&natvec[idx]->list);
3485 spin_unlock(&nm_i->nat_list_lock);
3486
3487 __del_from_nat_cache(nm_i, natvec[idx]);
3488 }
3489 }
3490 f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3491
3492 /* destroy nat set cache */
3493 nid = 0;
3494 memset(vec, 0, sizeof(void *) * NAT_VEC_SIZE);
3495 while ((found = __gang_lookup_nat_set(nm_i,
3496 nid, NAT_VEC_SIZE, setvec))) {
3497 unsigned idx;
3498
3499 nid = setvec[found - 1]->set + 1;
3500 for (idx = 0; idx < found; idx++) {
3501 /* entry_cnt is not zero, when cp_error was occurred */
3502 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3503 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3504 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3505 }
3506 }
3507 f2fs_up_write(&nm_i->nat_tree_lock);
3508
3509 kvfree(nm_i->nat_block_bitmap);
3510 if (nm_i->free_nid_bitmap) {
3511 int i;
3512
3513 for (i = 0; i < nm_i->nat_blocks; i++)
3514 kvfree(nm_i->free_nid_bitmap[i]);
3515 kvfree(nm_i->free_nid_bitmap);
3516 }
3517 kvfree(nm_i->free_nid_count);
3518
3519 kfree(nm_i->nat_bitmap);
3520 kvfree(nm_i->nat_bits);
3521#ifdef CONFIG_F2FS_CHECK_FS
3522 kfree(nm_i->nat_bitmap_mir);
3523#endif
3524 sbi->nm_info = NULL;
3525 kfree(nm_i);
3526}
3527
3528int __init f2fs_create_node_manager_caches(void)
3529{
3530 nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3531 sizeof(struct nat_entry));
3532 if (!nat_entry_slab)
3533 goto fail;
3534
3535 free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3536 sizeof(struct free_nid));
3537 if (!free_nid_slab)
3538 goto destroy_nat_entry;
3539
3540 nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3541 sizeof(struct nat_entry_set));
3542 if (!nat_entry_set_slab)
3543 goto destroy_free_nid;
3544
3545 fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3546 sizeof(struct fsync_node_entry));
3547 if (!fsync_node_entry_slab)
3548 goto destroy_nat_entry_set;
3549 return 0;
3550
3551destroy_nat_entry_set:
3552 kmem_cache_destroy(nat_entry_set_slab);
3553destroy_free_nid:
3554 kmem_cache_destroy(free_nid_slab);
3555destroy_nat_entry:
3556 kmem_cache_destroy(nat_entry_slab);
3557fail:
3558 return -ENOMEM;
3559}
3560
3561void f2fs_destroy_node_manager_caches(void)
3562{
3563 kmem_cache_destroy(fsync_node_entry_slab);
3564 kmem_cache_destroy(nat_entry_set_slab);
3565 kmem_cache_destroy(free_nid_slab);
3566 kmem_cache_destroy(nat_entry_slab);
3567}