projects/linux·blob master · refs

3600 lines · 88.4 KiB

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#include <linux/fserror.h>
16
17#include "f2fs.h"
18#include "node.h"
19#include "segment.h"
20#include "xattr.h"
21#include "iostat.h"
22#include <trace/events/f2fs.h>
23
24#define on_f2fs_build_free_nids(nm_i) mutex_is_locked(&(nm_i)->build_lock)
25
26static struct kmem_cache *nat_entry_slab;
27static struct kmem_cache *free_nid_slab;
28static struct kmem_cache *nat_entry_set_slab;
29static struct kmem_cache *fsync_node_entry_slab;
30
31static inline bool is_invalid_nid(struct f2fs_sb_info *sbi, nid_t nid)
32{
33 return nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid;
34}
35
36/*
37 * Check whether the given nid is within node id range.
38 */
39int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
40{
41 if (unlikely(is_invalid_nid(sbi, nid))) {
42 set_sbi_flag(sbi, SBI_NEED_FSCK);
43 f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
44 __func__, nid);
45 f2fs_handle_error(sbi, ERROR_CORRUPTED_INODE);
46 return -EFSCORRUPTED;
47 }
48 return 0;
49}
50
51bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
52{
53 struct f2fs_nm_info *nm_i = NM_I(sbi);
54 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
55 struct sysinfo val;
56 unsigned long avail_ram;
57 unsigned long mem_size = 0;
58 bool res = false;
59
60 if (!nm_i)
61 return true;
62
63 si_meminfo(&val);
64
65 /* only uses low memory */
66 avail_ram = val.totalram - val.totalhigh;
67
68 /*
69 * give 25%, 25%, 50%, 50%, 25%, 25% memory for each components respectively
70 */
71 if (type == FREE_NIDS) {
72 mem_size = (nm_i->nid_cnt[FREE_NID] *
73 sizeof(struct free_nid)) >> PAGE_SHIFT;
74 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
75 } else if (type == NAT_ENTRIES) {
76 /*
77 * nat_cnt[] is heuristic accounting. Sample it locklessly here
78 * to avoid taking nat_tree_lock in the balance path.
79 */
80 mem_size = (data_race(READ_ONCE(nm_i->nat_cnt[TOTAL_NAT])) *
81 sizeof(struct nat_entry)) >> PAGE_SHIFT;
82 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
83 if (excess_cached_nats(sbi))
84 res = false;
85 } else if (type == DIRTY_DENTS) {
86 if (bdi_wb_dirty_exceeded(sbi->sb->s_bdi))
87 return false;
88 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
89 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
90 } else if (type == INO_ENTRIES) {
91 int i;
92
93 for (i = 0; i < MAX_INO_ENTRY; i++)
94 mem_size += sbi->im[i].ino_num *
95 sizeof(struct ino_entry);
96 mem_size >>= PAGE_SHIFT;
97 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
98 } else if (type == READ_EXTENT_CACHE || type == AGE_EXTENT_CACHE) {
99 enum extent_type etype = type == READ_EXTENT_CACHE ?
100 EX_READ : EX_BLOCK_AGE;
101 struct extent_tree_info *eti = &sbi->extent_tree[etype];
102
103 mem_size = (atomic_read(&eti->total_ext_tree) *
104 sizeof(struct extent_tree) +
105 atomic_read(&eti->total_ext_node) *
106 sizeof(struct extent_node)) >> PAGE_SHIFT;
107 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
108 } else if (type == DISCARD_CACHE) {
109 mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
110 sizeof(struct discard_cmd)) >> PAGE_SHIFT;
111 res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
112 } else if (type == COMPRESS_PAGE) {
113#ifdef CONFIG_F2FS_FS_COMPRESSION
114 unsigned long free_ram = val.freeram;
115
116 /*
117 * free memory is lower than watermark or cached page count
118 * exceed threshold, deny caching compress page.
119 */
120 res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
121 (COMPRESS_MAPPING(sbi)->nrpages <
122 free_ram * sbi->compress_percent / 100);
123#else
124 res = false;
125#endif
126 } else {
127 if (!bdi_wb_dirty_exceeded(sbi->sb->s_bdi))
128 return true;
129 }
130 return res;
131}
132
133static void clear_node_folio_dirty(struct folio *folio)
134{
135 if (folio_test_dirty(folio)) {
136 f2fs_clear_page_cache_dirty_tag(folio);
137 folio_clear_dirty_for_io(folio);
138 dec_page_count(F2FS_F_SB(folio), F2FS_DIRTY_NODES);
139 }
140 folio_clear_uptodate(folio);
141}
142
143static struct folio *get_current_nat_folio(struct f2fs_sb_info *sbi, nid_t nid)
144{
145 return f2fs_get_meta_folio_retry(sbi, current_nat_addr(sbi, nid));
146}
147
148static struct folio *get_next_nat_folio(struct f2fs_sb_info *sbi, nid_t nid)
149{
150 struct folio *src_folio;
151 struct folio *dst_folio;
152 pgoff_t dst_off;
153 void *src_addr;
154 void *dst_addr;
155 struct f2fs_nm_info *nm_i = NM_I(sbi);
156
157 dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
158
159 /* get current nat block page with lock */
160 src_folio = get_current_nat_folio(sbi, nid);
161 if (IS_ERR(src_folio))
162 return src_folio;
163 dst_folio = f2fs_grab_meta_folio(sbi, dst_off);
164 f2fs_bug_on(sbi, folio_test_dirty(src_folio));
165
166 src_addr = folio_address(src_folio);
167 dst_addr = folio_address(dst_folio);
168 memcpy(dst_addr, src_addr, PAGE_SIZE);
169 folio_mark_dirty(dst_folio);
170 f2fs_folio_put(src_folio, true);
171
172 set_to_next_nat(nm_i, nid);
173
174 return dst_folio;
175}
176
177static struct nat_entry *__alloc_nat_entry(struct f2fs_sb_info *sbi,
178 nid_t nid, bool no_fail)
179{
180 struct nat_entry *new;
181
182 new = f2fs_kmem_cache_alloc(nat_entry_slab,
183 GFP_F2FS_ZERO, no_fail, sbi);
184 if (new) {
185 nat_set_nid(new, nid);
186 nat_reset_flag(new);
187 }
188 return new;
189}
190
191static void __free_nat_entry(struct nat_entry *e)
192{
193 kmem_cache_free(nat_entry_slab, e);
194}
195
196/* must be locked by nat_tree_lock */
197static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
198 struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail, bool init_dirty)
199{
200 if (no_fail)
201 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
202 else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
203 return NULL;
204
205 if (raw_ne)
206 node_info_from_raw_nat(&ne->ni, raw_ne);
207
208 if (init_dirty) {
209 INIT_LIST_HEAD(&ne->list);
210 nm_i->nat_cnt[TOTAL_NAT]++;
211 return ne;
212 }
213
214 spin_lock(&nm_i->nat_list_lock);
215 list_add_tail(&ne->list, &nm_i->nat_entries);
216 spin_unlock(&nm_i->nat_list_lock);
217
218 nm_i->nat_cnt[TOTAL_NAT]++;
219 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
220 return ne;
221}
222
223static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n, bool for_dirty)
224{
225 struct nat_entry *ne;
226
227 ne = radix_tree_lookup(&nm_i->nat_root, n);
228
229 /*
230 * for recent accessed nat entry which will not be dirtied soon
231 * later, move it to tail of lru list.
232 */
233 if (ne && !get_nat_flag(ne, IS_DIRTY) && !for_dirty) {
234 spin_lock(&nm_i->nat_list_lock);
235 if (!list_empty(&ne->list))
236 list_move_tail(&ne->list, &nm_i->nat_entries);
237 spin_unlock(&nm_i->nat_list_lock);
238 }
239
240 return ne;
241}
242
243static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
244 nid_t start, unsigned int nr, struct nat_entry **ep)
245{
246 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
247}
248
249static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
250{
251 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
252 nm_i->nat_cnt[TOTAL_NAT]--;
253 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
254 __free_nat_entry(e);
255}
256
257static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
258 struct nat_entry *ne)
259{
260 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
261 struct nat_entry_set *head;
262
263 head = radix_tree_lookup(&nm_i->nat_set_root, set);
264 if (!head) {
265 head = f2fs_kmem_cache_alloc(nat_entry_set_slab,
266 GFP_NOFS, true, NULL);
267
268 INIT_LIST_HEAD(&head->entry_list);
269 INIT_LIST_HEAD(&head->set_list);
270 head->set = set;
271 head->entry_cnt = 0;
272 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
273 }
274 return head;
275}
276
277static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
278 struct nat_entry *ne, bool init_dirty)
279{
280 struct nat_entry_set *head;
281 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
282
283 if (!new_ne)
284 head = __grab_nat_entry_set(nm_i, ne);
285
286 /*
287 * update entry_cnt in below condition:
288 * 1. update NEW_ADDR to valid block address;
289 * 2. update old block address to new one;
290 */
291 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
292 !get_nat_flag(ne, IS_DIRTY)))
293 head->entry_cnt++;
294
295 set_nat_flag(ne, IS_PREALLOC, new_ne);
296
297 if (get_nat_flag(ne, IS_DIRTY))
298 goto refresh_list;
299
300 nm_i->nat_cnt[DIRTY_NAT]++;
301 if (!init_dirty)
302 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
303 set_nat_flag(ne, IS_DIRTY, true);
304refresh_list:
305 spin_lock(&nm_i->nat_list_lock);
306 if (new_ne)
307 list_del_init(&ne->list);
308 else
309 list_move_tail(&ne->list, &head->entry_list);
310 spin_unlock(&nm_i->nat_list_lock);
311}
312
313static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
314 struct nat_entry_set *set, struct nat_entry *ne)
315{
316 spin_lock(&nm_i->nat_list_lock);
317 list_move_tail(&ne->list, &nm_i->nat_entries);
318 spin_unlock(&nm_i->nat_list_lock);
319
320 set_nat_flag(ne, IS_DIRTY, false);
321 set->entry_cnt--;
322 nm_i->nat_cnt[DIRTY_NAT]--;
323 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
324}
325
326static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
327 nid_t start, unsigned int nr, struct nat_entry_set **ep)
328{
329 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
330 start, nr);
331}
332
333bool f2fs_in_warm_node_list(struct folio *folio)
334{
335 return is_node_folio(folio) && IS_DNODE(folio) && is_cold_node(folio);
336}
337
338void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
339{
340 spin_lock_init(&sbi->fsync_node_lock);
341 INIT_LIST_HEAD(&sbi->fsync_node_list);
342 sbi->fsync_seg_id = 0;
343 sbi->fsync_node_num = 0;
344}
345
346static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
347 struct folio *folio)
348{
349 struct fsync_node_entry *fn;
350 unsigned long flags;
351 unsigned int seq_id;
352
353 fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab,
354 GFP_NOFS, true, NULL);
355
356 folio_get(folio);
357 fn->folio = folio;
358 INIT_LIST_HEAD(&fn->list);
359
360 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
361 list_add_tail(&fn->list, &sbi->fsync_node_list);
362 fn->seq_id = sbi->fsync_seg_id++;
363 seq_id = fn->seq_id;
364 sbi->fsync_node_num++;
365 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
366
367 return seq_id;
368}
369
370void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct folio *folio)
371{
372 struct fsync_node_entry *fn;
373 unsigned long flags;
374
375 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
376 list_for_each_entry(fn, &sbi->fsync_node_list, list) {
377 if (fn->folio == folio) {
378 list_del(&fn->list);
379 sbi->fsync_node_num--;
380 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
381 kmem_cache_free(fsync_node_entry_slab, fn);
382 folio_put(folio);
383 return;
384 }
385 }
386 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
387 f2fs_bug_on(sbi, 1);
388}
389
390void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
391{
392 unsigned long flags;
393
394 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
395 sbi->fsync_seg_id = 0;
396 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
397}
398
399bool f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
400{
401 struct f2fs_nm_info *nm_i = NM_I(sbi);
402 struct nat_entry *e;
403 bool need = false;
404 struct f2fs_lock_context lc;
405
406 f2fs_down_read_trace(&nm_i->nat_tree_lock, &lc);
407 e = __lookup_nat_cache(nm_i, nid, false);
408 if (e) {
409 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
410 !get_nat_flag(e, HAS_FSYNCED_INODE))
411 need = true;
412 }
413 f2fs_up_read_trace(&nm_i->nat_tree_lock, &lc);
414 return need;
415}
416
417bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
418{
419 struct f2fs_nm_info *nm_i = NM_I(sbi);
420 struct nat_entry *e;
421 bool is_cp = true;
422 struct f2fs_lock_context lc;
423
424 f2fs_down_read_trace(&nm_i->nat_tree_lock, &lc);
425 e = __lookup_nat_cache(nm_i, nid, false);
426 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
427 is_cp = false;
428 f2fs_up_read_trace(&nm_i->nat_tree_lock, &lc);
429 return is_cp;
430}
431
432bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
433{
434 struct f2fs_nm_info *nm_i = NM_I(sbi);
435 struct nat_entry *e;
436 bool need_update = true;
437 struct f2fs_lock_context lc, nlc;
438
439 f2fs_down_read_trace(&sbi->node_write, &lc);
440 f2fs_down_read_trace(&nm_i->nat_tree_lock, &nlc);
441 e = __lookup_nat_cache(nm_i, ino, false);
442 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
443 (get_nat_flag(e, IS_CHECKPOINTED) ||
444 get_nat_flag(e, HAS_FSYNCED_INODE)))
445 need_update = false;
446 f2fs_up_read_trace(&nm_i->nat_tree_lock, &nlc);
447 f2fs_up_read_trace(&sbi->node_write, &lc);
448 return need_update;
449}
450
451/* must be locked by nat_tree_lock */
452static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
453 struct f2fs_nat_entry *ne)
454{
455 struct f2fs_nm_info *nm_i = NM_I(sbi);
456 struct nat_entry *new, *e;
457 struct f2fs_lock_context lc;
458
459 /* Let's mitigate lock contention of nat_tree_lock during checkpoint */
460 if (f2fs_rwsem_is_locked(&sbi->cp_global_sem))
461 return;
462
463 new = __alloc_nat_entry(sbi, nid, false);
464 if (!new)
465 return;
466
467 f2fs_down_write_trace(&nm_i->nat_tree_lock, &lc);
468 e = __lookup_nat_cache(nm_i, nid, false);
469 if (!e)
470 e = __init_nat_entry(nm_i, new, ne, false, false);
471 else
472 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
473 nat_get_blkaddr(e) !=
474 le32_to_cpu(ne->block_addr) ||
475 nat_get_version(e) != ne->version);
476 f2fs_up_write_trace(&nm_i->nat_tree_lock, &lc);
477 if (e != new)
478 __free_nat_entry(new);
479}
480
481static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
482 block_t new_blkaddr, bool fsync_done)
483{
484 struct f2fs_nm_info *nm_i = NM_I(sbi);
485 struct nat_entry *e;
486 struct nat_entry *new = __alloc_nat_entry(sbi, ni->nid, true);
487 bool init_dirty = false;
488 struct f2fs_lock_context lc;
489
490 f2fs_down_write_trace(&nm_i->nat_tree_lock, &lc);
491 e = __lookup_nat_cache(nm_i, ni->nid, true);
492 if (!e) {
493 init_dirty = true;
494 e = __init_nat_entry(nm_i, new, NULL, true, true);
495 copy_node_info(&e->ni, ni);
496 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
497 } else if (new_blkaddr == NEW_ADDR) {
498 /*
499 * when nid is reallocated,
500 * previous nat entry can be remained in nat cache.
501 * So, reinitialize it with new information.
502 */
503 copy_node_info(&e->ni, ni);
504 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
505 }
506 /* let's free early to reduce memory consumption */
507 if (e != new)
508 __free_nat_entry(new);
509
510 /* sanity check */
511 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
512 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
513 new_blkaddr == NULL_ADDR);
514 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
515 new_blkaddr == NEW_ADDR);
516 f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
517 new_blkaddr == NEW_ADDR);
518
519 /* increment version no as node is removed */
520 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
521 unsigned char version = nat_get_version(e);
522
523 nat_set_version(e, inc_node_version(version));
524 }
525
526 /* change address */
527 nat_set_blkaddr(e, new_blkaddr);
528 if (!__is_valid_data_blkaddr(new_blkaddr))
529 set_nat_flag(e, IS_CHECKPOINTED, false);
530 __set_nat_cache_dirty(nm_i, e, init_dirty);
531
532 /* update fsync_mark if its inode nat entry is still alive */
533 if (ni->nid != ni->ino)
534 e = __lookup_nat_cache(nm_i, ni->ino, false);
535 if (e) {
536 if (fsync_done && ni->nid == ni->ino)
537 set_nat_flag(e, HAS_FSYNCED_INODE, true);
538 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
539 }
540 f2fs_up_write_trace(&nm_i->nat_tree_lock, &lc);
541}
542
543int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
544{
545 struct f2fs_nm_info *nm_i = NM_I(sbi);
546 int nr = nr_shrink;
547 struct f2fs_lock_context lc;
548
549 if (!f2fs_down_write_trylock_trace(&nm_i->nat_tree_lock, &lc))
550 return 0;
551
552 spin_lock(&nm_i->nat_list_lock);
553 while (nr_shrink) {
554 struct nat_entry *ne;
555
556 if (list_empty(&nm_i->nat_entries))
557 break;
558
559 ne = list_first_entry(&nm_i->nat_entries,
560 struct nat_entry, list);
561 list_del(&ne->list);
562 spin_unlock(&nm_i->nat_list_lock);
563
564 __del_from_nat_cache(nm_i, ne);
565 nr_shrink--;
566
567 spin_lock(&nm_i->nat_list_lock);
568 }
569 spin_unlock(&nm_i->nat_list_lock);
570
571 f2fs_up_write_trace(&nm_i->nat_tree_lock, &lc);
572 return nr - nr_shrink;
573}
574
575int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
576 struct node_info *ni, bool checkpoint_context)
577{
578 struct f2fs_nm_info *nm_i = NM_I(sbi);
579 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
580 struct f2fs_journal *journal = curseg->journal;
581 nid_t start_nid = START_NID(nid);
582 struct f2fs_nat_block *nat_blk;
583 struct folio *folio = NULL;
584 struct f2fs_nat_entry ne;
585 struct nat_entry *e;
586 pgoff_t index;
587 int i;
588 bool need_cache = true;
589 struct f2fs_lock_context lc;
590
591 ni->flag = 0;
592 ni->nid = nid;
593retry:
594 /* Check nat cache */
595 f2fs_down_read_trace(&nm_i->nat_tree_lock, &lc);
596 e = __lookup_nat_cache(nm_i, nid, false);
597 if (e) {
598 ni->ino = nat_get_ino(e);
599 ni->blk_addr = nat_get_blkaddr(e);
600 ni->version = nat_get_version(e);
601 f2fs_up_read_trace(&nm_i->nat_tree_lock, &lc);
602 if (IS_ENABLED(CONFIG_F2FS_CHECK_FS)) {
603 need_cache = false;
604 goto sanity_check;
605 }
606 return 0;
607 }
608
609 /*
610 * Check current segment summary by trying to grab journal_rwsem first.
611 * This sem is on the critical path on the checkpoint requiring the above
612 * nat_tree_lock. Therefore, we should retry, if we failed to grab here
613 * while not bothering checkpoint.
614 */
615 if (!f2fs_rwsem_is_locked(&sbi->cp_global_sem) || checkpoint_context) {
616 down_read(&curseg->journal_rwsem);
617 } else if (f2fs_rwsem_is_contended(&nm_i->nat_tree_lock) ||
618 !down_read_trylock(&curseg->journal_rwsem)) {
619 f2fs_up_read_trace(&nm_i->nat_tree_lock, &lc);
620 goto retry;
621 }
622
623 i = f2fs_lookup_journal_in_cursum(sbi, journal, NAT_JOURNAL, nid, 0);
624 if (i >= 0) {
625 ne = nat_in_journal(journal, i);
626 node_info_from_raw_nat(ni, &ne);
627 }
628 up_read(&curseg->journal_rwsem);
629 if (i >= 0) {
630 f2fs_up_read_trace(&nm_i->nat_tree_lock, &lc);
631 goto sanity_check;
632 }
633
634 /* Fill node_info from nat page */
635 index = current_nat_addr(sbi, nid);
636 f2fs_up_read_trace(&nm_i->nat_tree_lock, &lc);
637
638 folio = f2fs_get_meta_folio(sbi, index);
639 if (IS_ERR(folio))
640 return PTR_ERR(folio);
641
642 nat_blk = folio_address(folio);
643 ne = nat_blk->entries[nid - start_nid];
644 node_info_from_raw_nat(ni, &ne);
645 f2fs_folio_put(folio, true);
646sanity_check:
647 if (__is_valid_data_blkaddr(ni->blk_addr) &&
648 !f2fs_is_valid_blkaddr(sbi, ni->blk_addr,
649 DATA_GENERIC_ENHANCE)) {
650 set_sbi_flag(sbi, SBI_NEED_FSCK);
651 f2fs_err_ratelimited(sbi,
652 "f2fs_get_node_info of %pS: inconsistent nat entry, "
653 "ino:%u, nid:%u, blkaddr:%u, ver:%u, flag:%u",
654 __builtin_return_address(0),
655 ni->ino, ni->nid, ni->blk_addr, ni->version, ni->flag);
656 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NAT);
657 return -EFSCORRUPTED;
658 }
659
660 if (unlikely(f2fs_quota_file(sbi, ni->nid) &&
661 !__is_valid_data_blkaddr(ni->blk_addr))) {
662 set_sbi_flag(sbi, SBI_NEED_FSCK);
663 f2fs_err_ratelimited(sbi,
664 "f2fs_get_node_info of %pS: inconsistent nat entry from qf_ino, "
665 "ino:%u, nid:%u, blkaddr:%u, ver:%u, flag:%u",
666 __builtin_return_address(0),
667 ni->ino, ni->nid, ni->blk_addr, ni->version, ni->flag);
668 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NAT);
669 return -EFSCORRUPTED;
670 }
671
672 /* cache nat entry */
673 if (need_cache)
674 cache_nat_entry(sbi, nid, &ne);
675 return 0;
676}
677
678/*
679 * readahead MAX_RA_NODE number of node pages.
680 */
681static void f2fs_ra_node_pages(struct folio *parent, int start, int n)
682{
683 struct f2fs_sb_info *sbi = F2FS_F_SB(parent);
684 struct blk_plug plug;
685 int i, end;
686 nid_t nid;
687
688 blk_start_plug(&plug);
689
690 /* Then, try readahead for siblings of the desired node */
691 end = start + n;
692 end = min(end, (int)NIDS_PER_BLOCK);
693 for (i = start; i < end; i++) {
694 nid = get_nid(parent, i, false);
695 f2fs_ra_node_page(sbi, nid);
696 }
697
698 blk_finish_plug(&plug);
699}
700
701pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
702{
703 const long direct_index = ADDRS_PER_INODE(dn->inode);
704 const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
705 const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
706 unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
707 int cur_level = dn->cur_level;
708 int max_level = dn->max_level;
709 pgoff_t base = 0;
710
711 if (!dn->max_level)
712 return pgofs + 1;
713
714 while (max_level-- > cur_level)
715 skipped_unit *= NIDS_PER_BLOCK;
716
717 switch (dn->max_level) {
718 case 3:
719 base += 2 * indirect_blks;
720 fallthrough;
721 case 2:
722 base += 2 * direct_blks;
723 fallthrough;
724 case 1:
725 base += direct_index;
726 break;
727 default:
728 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
729 }
730
731 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
732}
733
734/*
735 * The maximum depth is four.
736 * Offset[0] will have raw inode offset.
737 */
738static int get_node_path(struct inode *inode, long block,
739 int offset[4], unsigned int noffset[4])
740{
741 const long direct_index = ADDRS_PER_INODE(inode);
742 const long direct_blks = ADDRS_PER_BLOCK(inode);
743 const long dptrs_per_blk = NIDS_PER_BLOCK;
744 const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
745 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
746 int n = 0;
747 int level = 0;
748
749 noffset[0] = 0;
750
751 if (block < direct_index) {
752 offset[n] = block;
753 goto got;
754 }
755 block -= direct_index;
756 if (block < direct_blks) {
757 offset[n++] = NODE_DIR1_BLOCK;
758 noffset[n] = 1;
759 offset[n] = block;
760 level = 1;
761 goto got;
762 }
763 block -= direct_blks;
764 if (block < direct_blks) {
765 offset[n++] = NODE_DIR2_BLOCK;
766 noffset[n] = 2;
767 offset[n] = block;
768 level = 1;
769 goto got;
770 }
771 block -= direct_blks;
772 if (block < indirect_blks) {
773 offset[n++] = NODE_IND1_BLOCK;
774 noffset[n] = 3;
775 offset[n++] = block / direct_blks;
776 noffset[n] = 4 + offset[n - 1];
777 offset[n] = block % direct_blks;
778 level = 2;
779 goto got;
780 }
781 block -= indirect_blks;
782 if (block < indirect_blks) {
783 offset[n++] = NODE_IND2_BLOCK;
784 noffset[n] = 4 + dptrs_per_blk;
785 offset[n++] = block / direct_blks;
786 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
787 offset[n] = block % direct_blks;
788 level = 2;
789 goto got;
790 }
791 block -= indirect_blks;
792 if (block < dindirect_blks) {
793 offset[n++] = NODE_DIND_BLOCK;
794 noffset[n] = 5 + (dptrs_per_blk * 2);
795 offset[n++] = block / indirect_blks;
796 noffset[n] = 6 + (dptrs_per_blk * 2) +
797 offset[n - 1] * (dptrs_per_blk + 1);
798 offset[n++] = (block / direct_blks) % dptrs_per_blk;
799 noffset[n] = 7 + (dptrs_per_blk * 2) +
800 offset[n - 2] * (dptrs_per_blk + 1) +
801 offset[n - 1];
802 offset[n] = block % direct_blks;
803 level = 3;
804 goto got;
805 } else {
806 return -E2BIG;
807 }
808got:
809 return level;
810}
811
812static struct folio *f2fs_get_node_folio_ra(struct folio *parent, int start);
813
814/*
815 * Caller should call f2fs_put_dnode(dn).
816 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
817 * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
818 */
819int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
820{
821 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
822 struct folio *nfolio[4];
823 struct folio *parent = NULL;
824 int offset[4];
825 unsigned int noffset[4];
826 nid_t nids[4];
827 int level, i = 0;
828 int err = 0;
829
830 level = get_node_path(dn->inode, index, offset, noffset);
831 if (level < 0)
832 return level;
833
834 nids[0] = dn->inode->i_ino;
835
836 if (!dn->inode_folio) {
837 nfolio[0] = f2fs_get_inode_folio(sbi, nids[0]);
838 if (IS_ERR(nfolio[0]))
839 return PTR_ERR(nfolio[0]);
840 } else {
841 nfolio[0] = dn->inode_folio;
842 }
843
844 /* if inline_data is set, should not report any block indices */
845 if (f2fs_has_inline_data(dn->inode) && index) {
846 err = -ENOENT;
847 f2fs_folio_put(nfolio[0], true);
848 goto release_out;
849 }
850
851 parent = nfolio[0];
852 if (level != 0)
853 nids[1] = get_nid(parent, offset[0], true);
854 dn->inode_folio = nfolio[0];
855 dn->inode_folio_locked = true;
856
857 /* get indirect or direct nodes */
858 for (i = 1; i <= level; i++) {
859 bool done = false;
860
861 if (nids[i] && nids[i] == dn->inode->i_ino) {
862 err = -EFSCORRUPTED;
863 f2fs_err_ratelimited(sbi,
864 "inode mapping table is corrupted, run fsck to fix it, "
865 "ino:%llu, nid:%u, level:%d, offset:%d",
866 dn->inode->i_ino, nids[i], level, offset[level]);
867 set_sbi_flag(sbi, SBI_NEED_FSCK);
868 goto release_pages;
869 }
870
871 if (!nids[i] && mode == ALLOC_NODE) {
872 /* alloc new node */
873 if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
874 err = -ENOSPC;
875 goto release_pages;
876 }
877
878 dn->nid = nids[i];
879 nfolio[i] = f2fs_new_node_folio(dn, noffset[i]);
880 if (IS_ERR(nfolio[i])) {
881 f2fs_alloc_nid_failed(sbi, nids[i]);
882 err = PTR_ERR(nfolio[i]);
883 goto release_pages;
884 }
885
886 set_nid(parent, offset[i - 1], nids[i], i == 1);
887 f2fs_alloc_nid_done(sbi, nids[i]);
888 done = true;
889 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
890 nfolio[i] = f2fs_get_node_folio_ra(parent, offset[i - 1]);
891 if (IS_ERR(nfolio[i])) {
892 err = PTR_ERR(nfolio[i]);
893 goto release_pages;
894 }
895 done = true;
896 }
897 if (i == 1) {
898 dn->inode_folio_locked = false;
899 folio_unlock(parent);
900 } else {
901 f2fs_folio_put(parent, true);
902 }
903
904 if (!done) {
905 nfolio[i] = f2fs_get_node_folio(sbi, nids[i],
906 NODE_TYPE_NON_INODE);
907 if (IS_ERR(nfolio[i])) {
908 err = PTR_ERR(nfolio[i]);
909 f2fs_folio_put(nfolio[0], false);
910 goto release_out;
911 }
912 }
913 if (i < level) {
914 parent = nfolio[i];
915 nids[i + 1] = get_nid(parent, offset[i], false);
916 }
917 }
918 dn->nid = nids[level];
919 dn->ofs_in_node = offset[level];
920 dn->node_folio = nfolio[level];
921 dn->data_blkaddr = f2fs_data_blkaddr(dn);
922
923 if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) &&
924 f2fs_sb_has_readonly(sbi)) {
925 unsigned int cluster_size = F2FS_I(dn->inode)->i_cluster_size;
926 unsigned int ofs_in_node = dn->ofs_in_node;
927 pgoff_t fofs = index;
928 unsigned int c_len;
929 block_t blkaddr;
930
931 /* should align fofs and ofs_in_node to cluster_size */
932 if (fofs % cluster_size) {
933 fofs = round_down(fofs, cluster_size);
934 ofs_in_node = round_down(ofs_in_node, cluster_size);
935 }
936
937 c_len = f2fs_cluster_blocks_are_contiguous(dn, ofs_in_node);
938 if (!c_len)
939 goto out;
940
941 blkaddr = data_blkaddr(dn->inode, dn->node_folio, ofs_in_node);
942 if (blkaddr == COMPRESS_ADDR)
943 blkaddr = data_blkaddr(dn->inode, dn->node_folio,
944 ofs_in_node + 1);
945
946 f2fs_update_read_extent_tree_range_compressed(dn->inode,
947 fofs, blkaddr, cluster_size, c_len);
948 }
949out:
950 return 0;
951
952release_pages:
953 f2fs_folio_put(parent, true);
954 if (i > 1)
955 f2fs_folio_put(nfolio[0], false);
956release_out:
957 dn->inode_folio = NULL;
958 dn->node_folio = NULL;
959 if (err == -ENOENT) {
960 dn->cur_level = i;
961 dn->max_level = level;
962 dn->ofs_in_node = offset[level];
963 }
964 return err;
965}
966
967static int truncate_node(struct dnode_of_data *dn)
968{
969 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
970 struct node_info ni;
971 int err;
972 pgoff_t index;
973
974 err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
975 if (err)
976 return err;
977
978 if (ni.blk_addr != NEW_ADDR &&
979 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr, DATA_GENERIC_ENHANCE)) {
980 f2fs_err_ratelimited(sbi,
981 "nat entry is corrupted, run fsck to fix it, ino:%u, "
982 "nid:%u, blkaddr:%u", ni.ino, ni.nid, ni.blk_addr);
983 set_sbi_flag(sbi, SBI_NEED_FSCK);
984 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NAT);
985 return -EFSCORRUPTED;
986 }
987
988 /* Deallocate node address */
989 f2fs_invalidate_blocks(sbi, ni.blk_addr, 1);
990 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
991 set_node_addr(sbi, &ni, NULL_ADDR, false);
992
993 if (dn->nid == dn->inode->i_ino) {
994 f2fs_remove_orphan_inode(sbi, dn->nid);
995 dec_valid_inode_count(sbi);
996 f2fs_inode_synced(dn->inode);
997 }
998
999 clear_node_folio_dirty(dn->node_folio);
1000 set_sbi_flag(sbi, SBI_IS_DIRTY);
1001
1002 index = dn->node_folio->index;
1003 f2fs_folio_put(dn->node_folio, true);
1004
1005 invalidate_mapping_pages(NODE_MAPPING(sbi),
1006 index, index);
1007
1008 dn->node_folio = NULL;
1009 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
1010
1011 return 0;
1012}
1013
1014static int truncate_dnode(struct dnode_of_data *dn)
1015{
1016 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1017 struct folio *folio;
1018 int err;
1019
1020 if (dn->nid == 0)
1021 return 1;
1022
1023 /* get direct node */
1024 folio = f2fs_get_node_folio(sbi, dn->nid, NODE_TYPE_NON_INODE);
1025 if (PTR_ERR(folio) == -ENOENT)
1026 return 1;
1027 else if (IS_ERR(folio))
1028 return PTR_ERR(folio);
1029
1030 if (IS_INODE(folio) || ino_of_node(folio) != dn->inode->i_ino) {
1031 f2fs_err(sbi, "incorrect node reference, ino: %llu, nid: %u, ino_of_node: %u",
1032 dn->inode->i_ino, dn->nid, ino_of_node(folio));
1033 set_sbi_flag(sbi, SBI_NEED_FSCK);
1034 f2fs_handle_error(sbi, ERROR_INVALID_NODE_REFERENCE);
1035 f2fs_folio_put(folio, true);
1036 return -EFSCORRUPTED;
1037 }
1038
1039 /* Make dnode_of_data for parameter */
1040 dn->node_folio = folio;
1041 dn->ofs_in_node = 0;
1042 f2fs_truncate_data_blocks_range(dn, ADDRS_PER_BLOCK(dn->inode));
1043 err = truncate_node(dn);
1044 if (err) {
1045 f2fs_folio_put(folio, true);
1046 return err;
1047 }
1048
1049 return 1;
1050}
1051
1052static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
1053 int ofs, int depth)
1054{
1055 struct dnode_of_data rdn = *dn;
1056 struct folio *folio;
1057 struct f2fs_node *rn;
1058 nid_t child_nid;
1059 unsigned int child_nofs;
1060 int freed = 0;
1061 int i, ret;
1062
1063 if (dn->nid == 0)
1064 return NIDS_PER_BLOCK + 1;
1065
1066 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
1067
1068 folio = f2fs_get_node_folio(F2FS_I_SB(dn->inode), dn->nid,
1069 NODE_TYPE_NON_INODE);
1070 if (IS_ERR(folio)) {
1071 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(folio));
1072 return PTR_ERR(folio);
1073 }
1074
1075 f2fs_ra_node_pages(folio, ofs, NIDS_PER_BLOCK);
1076
1077 rn = F2FS_NODE(folio);
1078 if (depth < 3) {
1079 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
1080 child_nid = le32_to_cpu(rn->in.nid[i]);
1081 if (child_nid == 0)
1082 continue;
1083 rdn.nid = child_nid;
1084 ret = truncate_dnode(&rdn);
1085 if (ret < 0)
1086 goto out_err;
1087 if (set_nid(folio, i, 0, false))
1088 dn->node_changed = true;
1089 }
1090 } else {
1091 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
1092 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
1093 child_nid = le32_to_cpu(rn->in.nid[i]);
1094 if (child_nid == 0) {
1095 child_nofs += NIDS_PER_BLOCK + 1;
1096 continue;
1097 }
1098 rdn.nid = child_nid;
1099 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
1100 if (ret == (NIDS_PER_BLOCK + 1)) {
1101 if (set_nid(folio, i, 0, false))
1102 dn->node_changed = true;
1103 child_nofs += ret;
1104 } else if (ret < 0 && ret != -ENOENT) {
1105 goto out_err;
1106 }
1107 }
1108 freed = child_nofs;
1109 }
1110
1111 if (!ofs) {
1112 /* remove current indirect node */
1113 dn->node_folio = folio;
1114 ret = truncate_node(dn);
1115 if (ret)
1116 goto out_err;
1117 freed++;
1118 } else {
1119 f2fs_folio_put(folio, true);
1120 }
1121 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
1122 return freed;
1123
1124out_err:
1125 f2fs_folio_put(folio, true);
1126 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
1127 return ret;
1128}
1129
1130static int truncate_partial_nodes(struct dnode_of_data *dn,
1131 int *offset, int depth)
1132{
1133 struct folio *folios[2];
1134 nid_t nid[3];
1135 nid_t child_nid;
1136 int err = 0;
1137 int i;
1138 int idx = depth - 2;
1139
1140 nid[0] = get_nid(dn->inode_folio, offset[0], true);
1141 if (!nid[0])
1142 return 0;
1143
1144 /* get indirect nodes in the path */
1145 for (i = 0; i < idx + 1; i++) {
1146 /* reference count'll be increased */
1147 folios[i] = f2fs_get_node_folio(F2FS_I_SB(dn->inode), nid[i],
1148 NODE_TYPE_NON_INODE);
1149 if (IS_ERR(folios[i])) {
1150 err = PTR_ERR(folios[i]);
1151 idx = i - 1;
1152 goto fail;
1153 }
1154 nid[i + 1] = get_nid(folios[i], offset[i + 1], false);
1155 }
1156
1157 f2fs_ra_node_pages(folios[idx], offset[idx + 1], NIDS_PER_BLOCK);
1158
1159 /* free direct nodes linked to a partial indirect node */
1160 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1161 child_nid = get_nid(folios[idx], i, false);
1162 if (!child_nid)
1163 continue;
1164 dn->nid = child_nid;
1165 err = truncate_dnode(dn);
1166 if (err < 0)
1167 goto fail;
1168 if (set_nid(folios[idx], i, 0, false))
1169 dn->node_changed = true;
1170 }
1171
1172 if (offset[idx + 1] == 0) {
1173 dn->node_folio = folios[idx];
1174 dn->nid = nid[idx];
1175 err = truncate_node(dn);
1176 if (err)
1177 goto fail;
1178 } else {
1179 f2fs_folio_put(folios[idx], true);
1180 }
1181 offset[idx]++;
1182 offset[idx + 1] = 0;
1183 idx--;
1184fail:
1185 for (i = idx; i >= 0; i--)
1186 f2fs_folio_put(folios[i], true);
1187
1188 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1189
1190 return err;
1191}
1192
1193/*
1194 * All the block addresses of data and nodes should be nullified.
1195 */
1196int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1197{
1198 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1199 int err = 0, cont = 1;
1200 int level, offset[4], noffset[4];
1201 unsigned int nofs = 0;
1202 struct dnode_of_data dn;
1203 struct folio *folio;
1204
1205 trace_f2fs_truncate_inode_blocks_enter(inode, from);
1206
1207 level = get_node_path(inode, from, offset, noffset);
1208 if (level <= 0) {
1209 if (!level) {
1210 level = -EFSCORRUPTED;
1211 f2fs_err(sbi, "%s: inode ino=%llx has corrupted node block, from:%lu addrs:%u",
1212 __func__, inode->i_ino,
1213 from, ADDRS_PER_INODE(inode));
1214 set_sbi_flag(sbi, SBI_NEED_FSCK);
1215 }
1216 trace_f2fs_truncate_inode_blocks_exit(inode, level);
1217 return level;
1218 }
1219
1220 folio = f2fs_get_inode_folio(sbi, inode->i_ino);
1221 if (IS_ERR(folio)) {
1222 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(folio));
1223 return PTR_ERR(folio);
1224 }
1225
1226 set_new_dnode(&dn, inode, folio, NULL, 0);
1227 folio_unlock(folio);
1228
1229 switch (level) {
1230 case 0:
1231 case 1:
1232 nofs = noffset[1];
1233 break;
1234 case 2:
1235 nofs = noffset[1];
1236 if (!offset[level - 1])
1237 goto skip_partial;
1238 err = truncate_partial_nodes(&dn, offset, level);
1239 if (err < 0 && err != -ENOENT)
1240 goto fail;
1241 nofs += 1 + NIDS_PER_BLOCK;
1242 break;
1243 case 3:
1244 nofs = 5 + 2 * NIDS_PER_BLOCK;
1245 if (!offset[level - 1])
1246 goto skip_partial;
1247 err = truncate_partial_nodes(&dn, offset, level);
1248 if (err < 0 && err != -ENOENT)
1249 goto fail;
1250 break;
1251 default:
1252 BUG();
1253 }
1254
1255skip_partial:
1256 while (cont) {
1257 dn.nid = get_nid(folio, offset[0], true);
1258 switch (offset[0]) {
1259 case NODE_DIR1_BLOCK:
1260 case NODE_DIR2_BLOCK:
1261 err = truncate_dnode(&dn);
1262 break;
1263
1264 case NODE_IND1_BLOCK:
1265 case NODE_IND2_BLOCK:
1266 err = truncate_nodes(&dn, nofs, offset[1], 2);
1267 break;
1268
1269 case NODE_DIND_BLOCK:
1270 err = truncate_nodes(&dn, nofs, offset[1], 3);
1271 cont = 0;
1272 break;
1273
1274 default:
1275 BUG();
1276 }
1277 if (err == -ENOENT) {
1278 set_sbi_flag(F2FS_F_SB(folio), SBI_NEED_FSCK);
1279 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1280 fserror_report_file_metadata(dn.inode, -EFSCORRUPTED,
1281 GFP_NOFS);
1282 f2fs_err_ratelimited(sbi,
1283 "truncate node fail, ino:%llu, nid:%u, "
1284 "offset[0]:%d, offset[1]:%d, nofs:%d",
1285 inode->i_ino, dn.nid, offset[0],
1286 offset[1], nofs);
1287 err = 0;
1288 }
1289 if (err < 0)
1290 goto fail;
1291 if (offset[1] == 0 && get_nid(folio, offset[0], true)) {
1292 folio_lock(folio);
1293 BUG_ON(!is_node_folio(folio));
1294 set_nid(folio, offset[0], 0, true);
1295 folio_unlock(folio);
1296 }
1297 offset[1] = 0;
1298 offset[0]++;
1299 nofs += err;
1300 }
1301fail:
1302 f2fs_folio_put(folio, false);
1303 trace_f2fs_truncate_inode_blocks_exit(inode, err);
1304 return err > 0 ? 0 : err;
1305}
1306
1307/* caller must lock inode page */
1308int f2fs_truncate_xattr_node(struct inode *inode)
1309{
1310 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1311 nid_t nid = F2FS_I(inode)->i_xattr_nid;
1312 struct dnode_of_data dn;
1313 struct folio *nfolio;
1314 int err;
1315
1316 if (!nid)
1317 return 0;
1318
1319 nfolio = f2fs_get_xnode_folio(sbi, nid);
1320 if (IS_ERR(nfolio))
1321 return PTR_ERR(nfolio);
1322
1323 set_new_dnode(&dn, inode, NULL, nfolio, nid);
1324 err = truncate_node(&dn);
1325 if (err) {
1326 f2fs_folio_put(nfolio, true);
1327 return err;
1328 }
1329
1330 f2fs_i_xnid_write(inode, 0);
1331
1332 return 0;
1333}
1334
1335/*
1336 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1337 * f2fs_unlock_op().
1338 */
1339int f2fs_remove_inode_page(struct inode *inode)
1340{
1341 struct dnode_of_data dn;
1342 int err;
1343
1344 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1345 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1346 if (err)
1347 return err;
1348
1349 err = f2fs_truncate_xattr_node(inode);
1350 if (err) {
1351 f2fs_put_dnode(&dn);
1352 return err;
1353 }
1354
1355 /* remove potential inline_data blocks */
1356 if (!IS_DEVICE_ALIASING(inode) &&
1357 (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1358 S_ISLNK(inode->i_mode)))
1359 f2fs_truncate_data_blocks_range(&dn, 1);
1360
1361 /* 0 is possible, after f2fs_new_inode() has failed */
1362 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1363 f2fs_put_dnode(&dn);
1364 return -EIO;
1365 }
1366
1367 if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1368 f2fs_warn(F2FS_I_SB(inode),
1369 "f2fs_remove_inode_page: inconsistent i_blocks, ino:%llu, iblocks:%llu",
1370 inode->i_ino, (unsigned long long)inode->i_blocks);
1371 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1372 }
1373
1374 /* will put inode & node pages */
1375 err = truncate_node(&dn);
1376 if (err) {
1377 f2fs_put_dnode(&dn);
1378 return err;
1379 }
1380 return 0;
1381}
1382
1383struct folio *f2fs_new_inode_folio(struct inode *inode)
1384{
1385 struct dnode_of_data dn;
1386
1387 /* allocate inode page for new inode */
1388 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1389
1390 /* caller should f2fs_folio_put(folio, true); */
1391 return f2fs_new_node_folio(&dn, 0);
1392}
1393
1394struct folio *f2fs_new_node_folio(struct dnode_of_data *dn, unsigned int ofs)
1395{
1396 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1397 struct node_info new_ni;
1398 struct folio *folio;
1399 int err;
1400
1401 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1402 return ERR_PTR(-EPERM);
1403
1404 folio = f2fs_grab_cache_folio(NODE_MAPPING(sbi), dn->nid, false);
1405 if (IS_ERR(folio))
1406 return folio;
1407
1408 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1409 goto fail;
1410
1411#ifdef CONFIG_F2FS_CHECK_FS
1412 err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false);
1413 if (err) {
1414 dec_valid_node_count(sbi, dn->inode, !ofs);
1415 goto fail;
1416 }
1417 if (unlikely(new_ni.blk_addr != NULL_ADDR)) {
1418 err = -EFSCORRUPTED;
1419 dec_valid_node_count(sbi, dn->inode, !ofs);
1420 set_sbi_flag(sbi, SBI_NEED_FSCK);
1421 f2fs_warn_ratelimited(sbi,
1422 "f2fs_new_node_folio: inconsistent nat entry, "
1423 "ino:%u, nid:%u, blkaddr:%u, ver:%u, flag:%u",
1424 new_ni.ino, new_ni.nid, new_ni.blk_addr,
1425 new_ni.version, new_ni.flag);
1426 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NAT);
1427 goto fail;
1428 }
1429#endif
1430 new_ni.nid = dn->nid;
1431 new_ni.ino = dn->inode->i_ino;
1432 new_ni.blk_addr = NULL_ADDR;
1433 new_ni.flag = 0;
1434 new_ni.version = 0;
1435 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1436
1437 f2fs_folio_wait_writeback(folio, NODE, true, true);
1438 fill_node_footer(folio, dn->nid, dn->inode->i_ino, ofs, true);
1439 set_cold_node(folio, S_ISDIR(dn->inode->i_mode));
1440 if (!folio_test_uptodate(folio))
1441 folio_mark_uptodate(folio);
1442 if (folio_mark_dirty(folio))
1443 dn->node_changed = true;
1444
1445 if (f2fs_has_xattr_block(ofs))
1446 f2fs_i_xnid_write(dn->inode, dn->nid);
1447
1448 if (ofs == 0)
1449 inc_valid_inode_count(sbi);
1450 return folio;
1451fail:
1452 clear_node_folio_dirty(folio);
1453 f2fs_folio_put(folio, true);
1454 return ERR_PTR(err);
1455}
1456
1457/*
1458 * Caller should do after getting the following values.
1459 * 0: f2fs_folio_put(folio, false)
1460 * LOCKED_PAGE or error: f2fs_folio_put(folio, true)
1461 */
1462static int read_node_folio(struct folio *folio, blk_opf_t op_flags)
1463{
1464 struct f2fs_sb_info *sbi = F2FS_F_SB(folio);
1465 struct node_info ni;
1466 struct f2fs_io_info fio = {
1467 .sbi = sbi,
1468 .type = NODE,
1469 .op = REQ_OP_READ,
1470 .op_flags = op_flags,
1471 .folio = folio,
1472 .encrypted_page = NULL,
1473 };
1474 int err;
1475
1476 if (folio_test_uptodate(folio)) {
1477 if (!f2fs_inode_chksum_verify(sbi, folio)) {
1478 folio_clear_uptodate(folio);
1479 return -EFSBADCRC;
1480 }
1481 return LOCKED_PAGE;
1482 }
1483
1484 err = f2fs_get_node_info(sbi, folio->index, &ni, false);
1485 if (err)
1486 return err;
1487
1488 /* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1489 if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR)) {
1490 folio_clear_uptodate(folio);
1491 return -ENOENT;
1492 }
1493
1494 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1495
1496 err = f2fs_submit_page_bio(&fio);
1497
1498 if (!err)
1499 f2fs_update_iostat(sbi, NULL, FS_NODE_READ_IO, F2FS_BLKSIZE);
1500
1501 return err;
1502}
1503
1504/*
1505 * Readahead a node page
1506 */
1507void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1508{
1509 struct folio *afolio;
1510 int err;
1511
1512 if (!nid)
1513 return;
1514 if (f2fs_check_nid_range(sbi, nid))
1515 return;
1516
1517 afolio = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1518 if (afolio)
1519 return;
1520
1521 afolio = f2fs_grab_cache_folio(NODE_MAPPING(sbi), nid, false);
1522 if (IS_ERR(afolio))
1523 return;
1524
1525 err = read_node_folio(afolio, REQ_RAHEAD);
1526 f2fs_folio_put(afolio, err ? true : false);
1527}
1528
1529int f2fs_sanity_check_node_footer(struct f2fs_sb_info *sbi,
1530 struct folio *folio, pgoff_t nid,
1531 enum node_type ntype, bool in_irq)
1532{
1533 bool is_inode, is_xnode;
1534
1535 if (unlikely(nid != nid_of_node(folio)))
1536 goto out_err;
1537
1538 is_inode = IS_INODE(folio);
1539 is_xnode = f2fs_has_xattr_block(ofs_of_node(folio));
1540
1541 switch (ntype) {
1542 case NODE_TYPE_REGULAR:
1543 if (is_inode && is_xnode)
1544 goto out_err;
1545 break;
1546 case NODE_TYPE_INODE:
1547 if (!is_inode || is_xnode)
1548 goto out_err;
1549 break;
1550 case NODE_TYPE_XATTR:
1551 if (is_inode || !is_xnode)
1552 goto out_err;
1553 break;
1554 case NODE_TYPE_NON_INODE:
1555 if (is_inode)
1556 goto out_err;
1557 break;
1558 case NODE_TYPE_NON_IXNODE:
1559 if (is_inode || is_xnode)
1560 goto out_err;
1561 break;
1562 default:
1563 break;
1564 }
1565 if (time_to_inject(sbi, FAULT_INCONSISTENT_FOOTER))
1566 goto out_err;
1567 return 0;
1568out_err:
1569 set_sbi_flag(sbi, SBI_NEED_FSCK);
1570 f2fs_warn_ratelimited(sbi, "inconsistent node block, node_type:%d, nid:%lu, "
1571 "node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1572 ntype, nid, nid_of_node(folio), ino_of_node(folio),
1573 ofs_of_node(folio), cpver_of_node(folio),
1574 next_blkaddr_of_node(folio));
1575
1576 f2fs_handle_error(sbi, ERROR_INCONSISTENT_FOOTER);
1577 fserror_report_file_metadata(folio->mapping->host,
1578 -EFSCORRUPTED, in_irq ? GFP_NOWAIT : GFP_NOFS);
1579 return -EFSCORRUPTED;
1580}
1581
1582static struct folio *__get_node_folio(struct f2fs_sb_info *sbi, pgoff_t nid,
1583 struct folio *parent, int start, enum node_type ntype)
1584{
1585 struct folio *folio;
1586 int err;
1587
1588 if (!nid)
1589 return ERR_PTR(-ENOENT);
1590 if (f2fs_check_nid_range(sbi, nid))
1591 return ERR_PTR(-EINVAL);
1592repeat:
1593 folio = f2fs_grab_cache_folio(NODE_MAPPING(sbi), nid, false);
1594 if (IS_ERR(folio))
1595 return folio;
1596
1597 err = read_node_folio(folio, 0);
1598 if (err < 0)
1599 goto out_put_err;
1600 if (err == LOCKED_PAGE)
1601 goto page_hit;
1602
1603 if (parent)
1604 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1605
1606 folio_lock(folio);
1607
1608 if (unlikely(!is_node_folio(folio))) {
1609 f2fs_folio_put(folio, true);
1610 goto repeat;
1611 }
1612
1613 if (unlikely(!folio_test_uptodate(folio))) {
1614 err = -EIO;
1615 goto out_put_err;
1616 }
1617
1618 if (!f2fs_inode_chksum_verify(sbi, folio)) {
1619 err = -EFSBADCRC;
1620 goto out_err;
1621 }
1622page_hit:
1623 err = f2fs_sanity_check_node_footer(sbi, folio, nid, ntype, false);
1624 if (!err)
1625 return folio;
1626out_err:
1627 clear_node_folio_dirty(folio);
1628out_put_err:
1629 /* ENOENT comes from read_node_folio which is not an error. */
1630 if (err != -ENOENT)
1631 f2fs_handle_page_eio(sbi, folio, NODE);
1632 f2fs_folio_put(folio, true);
1633 return ERR_PTR(err);
1634}
1635
1636struct folio *f2fs_get_node_folio(struct f2fs_sb_info *sbi, pgoff_t nid,
1637 enum node_type node_type)
1638{
1639 return __get_node_folio(sbi, nid, NULL, 0, node_type);
1640}
1641
1642struct folio *f2fs_get_inode_folio(struct f2fs_sb_info *sbi, pgoff_t ino)
1643{
1644 return __get_node_folio(sbi, ino, NULL, 0, NODE_TYPE_INODE);
1645}
1646
1647struct folio *f2fs_get_xnode_folio(struct f2fs_sb_info *sbi, pgoff_t xnid)
1648{
1649 return __get_node_folio(sbi, xnid, NULL, 0, NODE_TYPE_XATTR);
1650}
1651
1652static struct folio *f2fs_get_node_folio_ra(struct folio *parent, int start)
1653{
1654 struct f2fs_sb_info *sbi = F2FS_F_SB(parent);
1655 nid_t nid = get_nid(parent, start, false);
1656
1657 return __get_node_folio(sbi, nid, parent, start, NODE_TYPE_NON_IXNODE);
1658}
1659
1660static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1661{
1662 struct inode *inode;
1663 struct folio *folio;
1664 int ret;
1665
1666 /* should flush inline_data before evict_inode */
1667 inode = ilookup(sbi->sb, ino);
1668 if (!inode)
1669 return;
1670
1671 folio = f2fs_filemap_get_folio(inode->i_mapping, 0,
1672 FGP_LOCK|FGP_NOWAIT, 0);
1673 if (IS_ERR(folio))
1674 goto iput_out;
1675
1676 if (!folio_test_uptodate(folio))
1677 goto folio_out;
1678
1679 if (!folio_test_dirty(folio))
1680 goto folio_out;
1681
1682 if (!folio_clear_dirty_for_io(folio))
1683 goto folio_out;
1684
1685 ret = f2fs_write_inline_data(inode, folio);
1686 inode_dec_dirty_pages(inode);
1687 f2fs_remove_dirty_inode(inode);
1688 if (ret)
1689 folio_mark_dirty(folio);
1690folio_out:
1691 f2fs_folio_put(folio, true);
1692iput_out:
1693 iput(inode);
1694}
1695
1696static struct folio *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1697{
1698 pgoff_t index;
1699 struct folio_batch fbatch;
1700 struct folio *last_folio = NULL;
1701 int nr_folios;
1702
1703 folio_batch_init(&fbatch);
1704 index = 0;
1705
1706 while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1707 (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1708 &fbatch))) {
1709 int i;
1710
1711 for (i = 0; i < nr_folios; i++) {
1712 struct folio *folio = fbatch.folios[i];
1713
1714 if (unlikely(f2fs_cp_error(sbi))) {
1715 f2fs_folio_put(last_folio, false);
1716 folio_batch_release(&fbatch);
1717 return ERR_PTR(-EIO);
1718 }
1719
1720 if (!IS_DNODE(folio) || !is_cold_node(folio))
1721 continue;
1722 if (ino_of_node(folio) != ino)
1723 continue;
1724
1725 folio_lock(folio);
1726
1727 if (unlikely(!is_node_folio(folio))) {
1728continue_unlock:
1729 folio_unlock(folio);
1730 continue;
1731 }
1732 if (ino_of_node(folio) != ino)
1733 goto continue_unlock;
1734
1735 if (!folio_test_dirty(folio)) {
1736 /* someone wrote it for us */
1737 goto continue_unlock;
1738 }
1739
1740 if (last_folio)
1741 f2fs_folio_put(last_folio, false);
1742
1743 folio_get(folio);
1744 last_folio = folio;
1745 folio_unlock(folio);
1746 }
1747 folio_batch_release(&fbatch);
1748 cond_resched();
1749 }
1750 return last_folio;
1751}
1752
1753static bool __write_node_folio(struct folio *folio, bool atomic, bool do_fsync,
1754 bool *submitted, struct writeback_control *wbc,
1755 bool do_balance, enum iostat_type io_type,
1756 unsigned int *seq_id)
1757{
1758 struct f2fs_sb_info *sbi = F2FS_F_SB(folio);
1759 nid_t nid;
1760 struct node_info ni;
1761 struct f2fs_io_info fio = {
1762 .sbi = sbi,
1763 .ino = ino_of_node(folio),
1764 .type = NODE,
1765 .op = REQ_OP_WRITE,
1766 .op_flags = wbc_to_write_flags(wbc),
1767 .folio = folio,
1768 .encrypted_page = NULL,
1769 .submitted = 0,
1770 .io_type = io_type,
1771 .io_wbc = wbc,
1772 };
1773 struct f2fs_lock_context lc;
1774 unsigned int seq;
1775
1776 trace_f2fs_writepage(folio, NODE);
1777
1778 if (unlikely(f2fs_cp_error(sbi))) {
1779 /* keep node pages in remount-ro mode */
1780 if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY)
1781 goto redirty_out;
1782 folio_clear_uptodate(folio);
1783 dec_page_count(sbi, F2FS_DIRTY_NODES);
1784 folio_unlock(folio);
1785 return true;
1786 }
1787
1788 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1789 goto redirty_out;
1790
1791 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1792 wbc->sync_mode == WB_SYNC_NONE &&
1793 IS_DNODE(folio) && is_cold_node(folio))
1794 goto redirty_out;
1795
1796 /* get old block addr of this node page */
1797 nid = nid_of_node(folio);
1798
1799 if (f2fs_sanity_check_node_footer(sbi, folio, folio->index,
1800 NODE_TYPE_REGULAR, false)) {
1801 fserror_report_metadata(sbi->sb, -EFSCORRUPTED, GFP_NOFS);
1802 f2fs_stop_checkpoint(sbi, false, STOP_CP_REASON_CORRUPTED_NID);
1803 goto redirty_out;
1804 }
1805
1806 if (f2fs_get_node_info(sbi, nid, &ni, !do_balance))
1807 goto redirty_out;
1808
1809 f2fs_down_read_trace(&sbi->node_write, &lc);
1810
1811 /* This page is already truncated */
1812 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1813 folio_clear_uptodate(folio);
1814 dec_page_count(sbi, F2FS_DIRTY_NODES);
1815 f2fs_up_read_trace(&sbi->node_write, &lc);
1816 folio_unlock(folio);
1817 return true;
1818 }
1819
1820 if (__is_valid_data_blkaddr(ni.blk_addr) &&
1821 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1822 DATA_GENERIC_ENHANCE)) {
1823 f2fs_up_read_trace(&sbi->node_write, &lc);
1824 goto redirty_out;
1825 }
1826
1827 if (atomic && !test_opt(sbi, NOBARRIER))
1828 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1829
1830 set_dentry_mark(folio, false);
1831 set_fsync_mark(folio, do_fsync);
1832 if (IS_INODE(folio) && (atomic || is_fsync_dnode(folio)))
1833 set_dentry_mark(folio,
1834 f2fs_need_dentry_mark(sbi, ino_of_node(folio)));
1835
1836 /* should add to global list before clearing PAGECACHE status */
1837 if (f2fs_in_warm_node_list(folio)) {
1838 seq = f2fs_add_fsync_node_entry(sbi, folio);
1839 if (seq_id)
1840 *seq_id = seq;
1841 }
1842
1843 folio_start_writeback(folio);
1844
1845 fio.old_blkaddr = ni.blk_addr;
1846 f2fs_do_write_node_page(nid, &fio);
1847 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(folio));
1848 dec_page_count(sbi, F2FS_DIRTY_NODES);
1849 f2fs_up_read_trace(&sbi->node_write, &lc);
1850
1851 folio_unlock(folio);
1852
1853 if (unlikely(f2fs_cp_error(sbi))) {
1854 f2fs_submit_merged_write(sbi, NODE);
1855 submitted = NULL;
1856 }
1857 if (submitted)
1858 *submitted = fio.submitted;
1859
1860 if (do_balance)
1861 f2fs_balance_fs(sbi, false);
1862 return true;
1863
1864redirty_out:
1865 folio_redirty_for_writepage(wbc, folio);
1866 folio_unlock(folio);
1867 return false;
1868}
1869
1870int f2fs_write_single_node_folio(struct folio *node_folio, int sync_mode,
1871 bool mark_dirty, enum iostat_type io_type)
1872{
1873 int err = 0;
1874 struct writeback_control wbc = {
1875 .sync_mode = WB_SYNC_ALL,
1876 .nr_to_write = 1,
1877 };
1878
1879 if (!sync_mode) {
1880 /* set page dirty and write it */
1881 if (!folio_test_writeback(node_folio))
1882 folio_mark_dirty(node_folio);
1883 goto out_folio;
1884 }
1885
1886 f2fs_folio_wait_writeback(node_folio, NODE, true, true);
1887
1888 if (mark_dirty)
1889 folio_mark_dirty(node_folio);
1890 else if (!folio_test_dirty(node_folio))
1891 goto out_folio;
1892
1893 if (!folio_clear_dirty_for_io(node_folio)) {
1894 err = -EAGAIN;
1895 goto out_folio;
1896 }
1897
1898 if (!__write_node_folio(node_folio, false, false, NULL,
1899 &wbc, false, io_type, NULL))
1900 err = -EAGAIN;
1901 goto release_folio;
1902out_folio:
1903 folio_unlock(node_folio);
1904release_folio:
1905 f2fs_folio_put(node_folio, false);
1906 return err;
1907}
1908
1909int f2fs_move_node_folio(struct folio *node_folio, int gc_type)
1910{
1911 return f2fs_write_single_node_folio(node_folio, gc_type == FG_GC,
1912 true, FS_GC_NODE_IO);
1913}
1914
1915int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1916 struct writeback_control *wbc, bool atomic,
1917 unsigned int *seq_id)
1918{
1919 pgoff_t index;
1920 struct folio_batch fbatch;
1921 int ret = 0;
1922 struct folio *last_folio = NULL;
1923 bool marked = false;
1924 nid_t ino = inode->i_ino;
1925 int nr_folios;
1926 int nwritten = 0;
1927
1928 if (atomic) {
1929 last_folio = last_fsync_dnode(sbi, ino);
1930 if (IS_ERR_OR_NULL(last_folio))
1931 return PTR_ERR_OR_ZERO(last_folio);
1932 }
1933retry:
1934 folio_batch_init(&fbatch);
1935 index = 0;
1936
1937 while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1938 (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1939 &fbatch))) {
1940 int i;
1941
1942 for (i = 0; i < nr_folios; i++) {
1943 struct folio *folio = fbatch.folios[i];
1944 bool submitted = false;
1945 bool do_fsync = false;
1946
1947 if (unlikely(f2fs_cp_error(sbi))) {
1948 f2fs_folio_put(last_folio, false);
1949 folio_batch_release(&fbatch);
1950 ret = -EIO;
1951 goto out;
1952 }
1953
1954 if (!IS_DNODE(folio) || !is_cold_node(folio))
1955 continue;
1956 if (ino_of_node(folio) != ino)
1957 continue;
1958
1959 folio_lock(folio);
1960
1961 if (unlikely(!is_node_folio(folio))) {
1962continue_unlock:
1963 folio_unlock(folio);
1964 continue;
1965 }
1966 if (ino_of_node(folio) != ino)
1967 goto continue_unlock;
1968
1969 if (!folio_test_dirty(folio) && folio != last_folio) {
1970 /* someone wrote it for us */
1971 goto continue_unlock;
1972 }
1973
1974 f2fs_folio_wait_writeback(folio, NODE, true, true);
1975
1976 if (!atomic || folio == last_folio) {
1977 do_fsync = true;
1978 percpu_counter_inc(&sbi->rf_node_block_count);
1979 if (IS_INODE(folio)) {
1980 if (is_inode_flag_set(inode,
1981 FI_DIRTY_INODE))
1982 f2fs_update_inode(inode, folio);
1983 }
1984 /* may be written by other thread */
1985 if (!folio_test_dirty(folio))
1986 folio_mark_dirty(folio);
1987 }
1988
1989 if (!folio_clear_dirty_for_io(folio))
1990 goto continue_unlock;
1991
1992 if (!__write_node_folio(folio, atomic &&
1993 folio == last_folio,
1994 do_fsync, &submitted,
1995 wbc, true, FS_NODE_IO,
1996 seq_id)) {
1997 f2fs_folio_put(last_folio, false);
1998 folio_batch_release(&fbatch);
1999 ret = -EIO;
2000 goto out;
2001 }
2002 if (submitted)
2003 nwritten++;
2004
2005 if (folio == last_folio) {
2006 f2fs_folio_put(folio, false);
2007 folio_batch_release(&fbatch);
2008 marked = true;
2009 goto out;
2010 }
2011 }
2012 folio_batch_release(&fbatch);
2013 cond_resched();
2014 }
2015 if (atomic && !marked) {
2016 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
2017 ino, last_folio->index);
2018 folio_lock(last_folio);
2019 if (unlikely(!is_node_folio(last_folio))) {
2020 f2fs_folio_put(last_folio, true);
2021 ret = -EAGAIN;
2022 goto out;
2023 }
2024 f2fs_folio_wait_writeback(last_folio, NODE, true, true);
2025 folio_mark_dirty(last_folio);
2026 folio_unlock(last_folio);
2027 goto retry;
2028 }
2029out:
2030 if (nwritten)
2031 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
2032 return ret;
2033}
2034
2035static int f2fs_match_ino(struct inode *inode, u64 ino, void *data)
2036{
2037 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2038 bool clean;
2039
2040 if (inode->i_ino != ino)
2041 return 0;
2042
2043 if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
2044 return 0;
2045
2046 spin_lock(&sbi->inode_lock[DIRTY_META]);
2047 clean = list_empty(&F2FS_I(inode)->gdirty_list);
2048 spin_unlock(&sbi->inode_lock[DIRTY_META]);
2049
2050 if (clean)
2051 return 0;
2052
2053 inode = igrab(inode);
2054 if (!inode)
2055 return 0;
2056 return 1;
2057}
2058
2059static bool flush_dirty_inode(struct folio *folio)
2060{
2061 struct f2fs_sb_info *sbi = F2FS_F_SB(folio);
2062 struct inode *inode;
2063 nid_t ino = ino_of_node(folio);
2064
2065 inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
2066 if (!inode)
2067 return false;
2068
2069 f2fs_update_inode(inode, folio);
2070 folio_unlock(folio);
2071
2072 iput(inode);
2073 return true;
2074}
2075
2076void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
2077{
2078 pgoff_t index = 0;
2079 struct folio_batch fbatch;
2080 int nr_folios;
2081
2082 folio_batch_init(&fbatch);
2083
2084 while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
2085 (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
2086 &fbatch))) {
2087 int i;
2088
2089 for (i = 0; i < nr_folios; i++) {
2090 struct folio *folio = fbatch.folios[i];
2091
2092 if (!IS_INODE(folio))
2093 continue;
2094
2095 folio_lock(folio);
2096
2097 if (unlikely(!is_node_folio(folio)))
2098 goto unlock;
2099 if (!folio_test_dirty(folio))
2100 goto unlock;
2101
2102 /* flush inline_data, if it's async context. */
2103 if (folio_test_f2fs_inline(folio)) {
2104 folio_clear_f2fs_inline(folio);
2105 folio_unlock(folio);
2106 flush_inline_data(sbi, ino_of_node(folio));
2107 continue;
2108 }
2109unlock:
2110 folio_unlock(folio);
2111 }
2112 folio_batch_release(&fbatch);
2113 cond_resched();
2114 }
2115}
2116
2117int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
2118 struct writeback_control *wbc,
2119 bool do_balance, enum iostat_type io_type)
2120{
2121 pgoff_t index;
2122 struct folio_batch fbatch;
2123 int step = 0;
2124 int nwritten = 0;
2125 int ret = 0;
2126 int nr_folios, done = 0;
2127
2128 folio_batch_init(&fbatch);
2129
2130next_step:
2131 index = 0;
2132
2133 while (!done && (nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi),
2134 &index, (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
2135 &fbatch))) {
2136 int i;
2137
2138 for (i = 0; i < nr_folios; i++) {
2139 struct folio *folio = fbatch.folios[i];
2140 bool submitted = false;
2141
2142 /* give a priority to WB_SYNC threads */
2143 if (atomic_read(&sbi->wb_sync_req[NODE]) &&
2144 wbc->sync_mode == WB_SYNC_NONE) {
2145 done = 1;
2146 break;
2147 }
2148
2149 /*
2150 * flushing sequence with step:
2151 * 0. indirect nodes
2152 * 1. dentry dnodes
2153 * 2. file dnodes
2154 */
2155 if (step == 0 && IS_DNODE(folio))
2156 continue;
2157 if (step == 1 && (!IS_DNODE(folio) ||
2158 is_cold_node(folio)))
2159 continue;
2160 if (step == 2 && (!IS_DNODE(folio) ||
2161 !is_cold_node(folio)))
2162 continue;
2163lock_node:
2164 if (wbc->sync_mode == WB_SYNC_ALL)
2165 folio_lock(folio);
2166 else if (!folio_trylock(folio))
2167 continue;
2168
2169 if (unlikely(!is_node_folio(folio))) {
2170continue_unlock:
2171 folio_unlock(folio);
2172 continue;
2173 }
2174
2175 if (!folio_test_dirty(folio)) {
2176 /* someone wrote it for us */
2177 goto continue_unlock;
2178 }
2179
2180 /* flush inline_data/inode, if it's async context. */
2181 if (!do_balance)
2182 goto write_node;
2183
2184 /* flush inline_data */
2185 if (folio_test_f2fs_inline(folio)) {
2186 folio_clear_f2fs_inline(folio);
2187 folio_unlock(folio);
2188 flush_inline_data(sbi, ino_of_node(folio));
2189 goto lock_node;
2190 }
2191
2192 /* flush dirty inode */
2193 if (IS_INODE(folio) && flush_dirty_inode(folio))
2194 goto lock_node;
2195write_node:
2196 f2fs_folio_wait_writeback(folio, NODE, true, true);
2197
2198 if (!folio_clear_dirty_for_io(folio))
2199 goto continue_unlock;
2200
2201 if (!__write_node_folio(folio, false, false, &submitted,
2202 wbc, do_balance, io_type, NULL)) {
2203 folio_batch_release(&fbatch);
2204 ret = -EIO;
2205 goto out;
2206 }
2207 if (submitted)
2208 nwritten++;
2209
2210 if (--wbc->nr_to_write == 0)
2211 break;
2212 }
2213 folio_batch_release(&fbatch);
2214 cond_resched();
2215
2216 if (wbc->nr_to_write == 0) {
2217 step = 2;
2218 break;
2219 }
2220 }
2221
2222 if (step < 2) {
2223 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2224 wbc->sync_mode == WB_SYNC_NONE && step == 1)
2225 goto out;
2226 step++;
2227 goto next_step;
2228 }
2229out:
2230 if (nwritten)
2231 f2fs_submit_merged_write(sbi, NODE);
2232
2233 if (unlikely(f2fs_cp_error(sbi)))
2234 return -EIO;
2235 return ret;
2236}
2237
2238int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2239 unsigned int seq_id)
2240{
2241 struct fsync_node_entry *fn;
2242 struct list_head *head = &sbi->fsync_node_list;
2243 unsigned long flags;
2244 unsigned int cur_seq_id = 0;
2245
2246 while (seq_id && cur_seq_id < seq_id) {
2247 struct folio *folio;
2248
2249 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2250 if (list_empty(head)) {
2251 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2252 break;
2253 }
2254 fn = list_first_entry(head, struct fsync_node_entry, list);
2255 if (fn->seq_id > seq_id) {
2256 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2257 break;
2258 }
2259 cur_seq_id = fn->seq_id;
2260 folio = fn->folio;
2261 folio_get(folio);
2262 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2263
2264 f2fs_folio_wait_writeback(folio, NODE, true, false);
2265
2266 folio_put(folio);
2267 }
2268
2269 return filemap_check_errors(NODE_MAPPING(sbi));
2270}
2271
2272static int f2fs_write_node_pages(struct address_space *mapping,
2273 struct writeback_control *wbc)
2274{
2275 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2276 struct blk_plug plug;
2277 long diff;
2278
2279 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2280 goto skip_write;
2281
2282 /* balancing f2fs's metadata in background */
2283 f2fs_balance_fs_bg(sbi, true);
2284
2285 /* collect a number of dirty node pages and write together */
2286 if (wbc->sync_mode != WB_SYNC_ALL &&
2287 get_pages(sbi, F2FS_DIRTY_NODES) <
2288 nr_pages_to_skip(sbi, NODE))
2289 goto skip_write;
2290
2291 if (wbc->sync_mode == WB_SYNC_ALL)
2292 atomic_inc(&sbi->wb_sync_req[NODE]);
2293 else if (atomic_read(&sbi->wb_sync_req[NODE])) {
2294 /* to avoid potential deadlock */
2295 if (current->plug)
2296 blk_finish_plug(current->plug);
2297 goto skip_write;
2298 }
2299
2300 trace_f2fs_writepages(mapping->host, wbc, NODE);
2301
2302 diff = nr_pages_to_write(sbi, NODE, wbc);
2303 blk_start_plug(&plug);
2304 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2305 blk_finish_plug(&plug);
2306 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2307
2308 if (wbc->sync_mode == WB_SYNC_ALL)
2309 atomic_dec(&sbi->wb_sync_req[NODE]);
2310 return 0;
2311
2312skip_write:
2313 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2314 trace_f2fs_writepages(mapping->host, wbc, NODE);
2315 return 0;
2316}
2317
2318static bool f2fs_dirty_node_folio(struct address_space *mapping,
2319 struct folio *folio)
2320{
2321 trace_f2fs_set_page_dirty(folio, NODE);
2322
2323 if (!folio_test_uptodate(folio))
2324 folio_mark_uptodate(folio);
2325#ifdef CONFIG_F2FS_CHECK_FS
2326 if (IS_INODE(folio))
2327 f2fs_inode_chksum_set(F2FS_M_SB(mapping), folio);
2328#endif
2329 if (filemap_dirty_folio(mapping, folio)) {
2330 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
2331 folio_set_f2fs_reference(folio);
2332 return true;
2333 }
2334 return false;
2335}
2336
2337/*
2338 * Structure of the f2fs node operations
2339 */
2340const struct address_space_operations f2fs_node_aops = {
2341 .writepages = f2fs_write_node_pages,
2342 .dirty_folio = f2fs_dirty_node_folio,
2343 .invalidate_folio = f2fs_invalidate_folio,
2344 .release_folio = f2fs_release_folio,
2345 .migrate_folio = filemap_migrate_folio,
2346};
2347
2348static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2349 nid_t n)
2350{
2351 return radix_tree_lookup(&nm_i->free_nid_root, n);
2352}
2353
2354static int __insert_free_nid(struct f2fs_sb_info *sbi,
2355 struct free_nid *i)
2356{
2357 struct f2fs_nm_info *nm_i = NM_I(sbi);
2358 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2359
2360 if (err)
2361 return err;
2362
2363 nm_i->nid_cnt[FREE_NID]++;
2364 list_add_tail(&i->list, &nm_i->free_nid_list);
2365 return 0;
2366}
2367
2368static void __remove_free_nid(struct f2fs_sb_info *sbi,
2369 struct free_nid *i, enum nid_state state)
2370{
2371 struct f2fs_nm_info *nm_i = NM_I(sbi);
2372
2373 f2fs_bug_on(sbi, state != i->state);
2374 nm_i->nid_cnt[state]--;
2375 if (state == FREE_NID)
2376 list_del(&i->list);
2377 radix_tree_delete(&nm_i->free_nid_root, i->nid);
2378}
2379
2380static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2381 enum nid_state org_state, enum nid_state dst_state)
2382{
2383 struct f2fs_nm_info *nm_i = NM_I(sbi);
2384
2385 f2fs_bug_on(sbi, org_state != i->state);
2386 i->state = dst_state;
2387 nm_i->nid_cnt[org_state]--;
2388 nm_i->nid_cnt[dst_state]++;
2389
2390 switch (dst_state) {
2391 case PREALLOC_NID:
2392 list_del(&i->list);
2393 break;
2394 case FREE_NID:
2395 list_add_tail(&i->list, &nm_i->free_nid_list);
2396 break;
2397 default:
2398 BUG_ON(1);
2399 }
2400}
2401
2402static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2403 bool set, bool build)
2404{
2405 struct f2fs_nm_info *nm_i = NM_I(sbi);
2406 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2407 unsigned int nid_ofs = nid - START_NID(nid);
2408
2409 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2410 return;
2411
2412 if (set) {
2413 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2414 return;
2415 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2416 nm_i->free_nid_count[nat_ofs]++;
2417 } else {
2418 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2419 return;
2420 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2421 if (!build)
2422 nm_i->free_nid_count[nat_ofs]--;
2423 }
2424}
2425
2426/* return if the nid is recognized as free */
2427static bool add_free_nid(struct f2fs_sb_info *sbi,
2428 nid_t nid, bool build, bool update)
2429{
2430 struct f2fs_nm_info *nm_i = NM_I(sbi);
2431 struct free_nid *i, *e;
2432 struct nat_entry *ne;
2433 int err;
2434 bool ret = false;
2435
2436 /* 0 nid should not be used */
2437 if (unlikely(nid == 0))
2438 return false;
2439
2440 if (unlikely(f2fs_check_nid_range(sbi, nid)))
2441 return false;
2442
2443 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL);
2444 i->nid = nid;
2445 i->state = FREE_NID;
2446
2447 err = radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2448 f2fs_bug_on(sbi, err);
2449
2450 err = -EINVAL;
2451
2452 spin_lock(&nm_i->nid_list_lock);
2453
2454 if (build) {
2455 /*
2456 * Thread A Thread B
2457 * - f2fs_create
2458 * - f2fs_new_inode
2459 * - f2fs_alloc_nid
2460 * - __insert_nid_to_list(PREALLOC_NID)
2461 * - f2fs_balance_fs_bg
2462 * - f2fs_build_free_nids
2463 * - __f2fs_build_free_nids
2464 * - scan_nat_page
2465 * - add_free_nid
2466 * - __lookup_nat_cache
2467 * - f2fs_add_link
2468 * - f2fs_init_inode_metadata
2469 * - f2fs_new_inode_folio
2470 * - f2fs_new_node_folio
2471 * - set_node_addr
2472 * - f2fs_alloc_nid_done
2473 * - __remove_nid_from_list(PREALLOC_NID)
2474 * - __insert_nid_to_list(FREE_NID)
2475 */
2476 ne = __lookup_nat_cache(nm_i, nid, false);
2477 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2478 nat_get_blkaddr(ne) != NULL_ADDR))
2479 goto err_out;
2480
2481 e = __lookup_free_nid_list(nm_i, nid);
2482 if (e) {
2483 if (e->state == FREE_NID)
2484 ret = true;
2485 goto err_out;
2486 }
2487 }
2488 ret = true;
2489 err = __insert_free_nid(sbi, i);
2490err_out:
2491 if (update) {
2492 update_free_nid_bitmap(sbi, nid, ret, build);
2493 if (!build)
2494 nm_i->available_nids++;
2495 }
2496 spin_unlock(&nm_i->nid_list_lock);
2497 radix_tree_preload_end();
2498
2499 if (err)
2500 kmem_cache_free(free_nid_slab, i);
2501 return ret;
2502}
2503
2504static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2505{
2506 struct f2fs_nm_info *nm_i = NM_I(sbi);
2507 struct free_nid *i;
2508 bool need_free = false;
2509
2510 spin_lock(&nm_i->nid_list_lock);
2511 i = __lookup_free_nid_list(nm_i, nid);
2512 if (i && i->state == FREE_NID) {
2513 __remove_free_nid(sbi, i, FREE_NID);
2514 need_free = true;
2515 }
2516 spin_unlock(&nm_i->nid_list_lock);
2517
2518 if (need_free)
2519 kmem_cache_free(free_nid_slab, i);
2520}
2521
2522static int scan_nat_page(struct f2fs_sb_info *sbi,
2523 struct f2fs_nat_block *nat_blk, nid_t start_nid)
2524{
2525 struct f2fs_nm_info *nm_i = NM_I(sbi);
2526 block_t blk_addr;
2527 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2528 int i;
2529
2530 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2531
2532 i = start_nid % NAT_ENTRY_PER_BLOCK;
2533
2534 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2535 if (unlikely(start_nid >= nm_i->max_nid))
2536 break;
2537
2538 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2539
2540 if (blk_addr == NEW_ADDR)
2541 return -EFSCORRUPTED;
2542
2543 if (blk_addr == NULL_ADDR) {
2544 add_free_nid(sbi, start_nid, true, true);
2545 } else {
2546 spin_lock(&NM_I(sbi)->nid_list_lock);
2547 update_free_nid_bitmap(sbi, start_nid, false, true);
2548 spin_unlock(&NM_I(sbi)->nid_list_lock);
2549 }
2550 }
2551
2552 return 0;
2553}
2554
2555static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2556{
2557 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2558 struct f2fs_journal *journal = curseg->journal;
2559 int i;
2560
2561 down_read(&curseg->journal_rwsem);
2562 for (i = 0; i < nats_in_cursum(journal); i++) {
2563 block_t addr;
2564 nid_t nid;
2565
2566 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2567 nid = le32_to_cpu(nid_in_journal(journal, i));
2568 if (addr == NULL_ADDR)
2569 add_free_nid(sbi, nid, true, false);
2570 else
2571 remove_free_nid(sbi, nid);
2572 }
2573 up_read(&curseg->journal_rwsem);
2574}
2575
2576static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2577{
2578 struct f2fs_nm_info *nm_i = NM_I(sbi);
2579 unsigned int i, idx;
2580 nid_t nid;
2581 struct f2fs_lock_context lc;
2582
2583 f2fs_down_read_trace(&nm_i->nat_tree_lock, &lc);
2584
2585 for (i = 0; i < nm_i->nat_blocks; i++) {
2586 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2587 continue;
2588 if (!nm_i->free_nid_count[i])
2589 continue;
2590 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2591 idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2592 NAT_ENTRY_PER_BLOCK, idx);
2593 if (idx >= NAT_ENTRY_PER_BLOCK)
2594 break;
2595
2596 nid = i * NAT_ENTRY_PER_BLOCK + idx;
2597 add_free_nid(sbi, nid, true, false);
2598
2599 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2600 goto out;
2601 }
2602 }
2603out:
2604 scan_curseg_cache(sbi);
2605
2606 f2fs_up_read_trace(&nm_i->nat_tree_lock, &lc);
2607}
2608
2609static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2610 bool sync, bool mount)
2611{
2612 struct f2fs_nm_info *nm_i = NM_I(sbi);
2613 int i = 0, ret;
2614 nid_t nid = nm_i->next_scan_nid;
2615 struct f2fs_lock_context lc;
2616
2617 if (unlikely(nid >= nm_i->max_nid))
2618 nid = 0;
2619
2620 if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2621 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2622
2623 /* Enough entries */
2624 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2625 return 0;
2626
2627 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2628 return 0;
2629
2630 if (!mount) {
2631 /* try to find free nids in free_nid_bitmap */
2632 scan_free_nid_bits(sbi);
2633
2634 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2635 return 0;
2636 }
2637
2638 /* readahead nat pages to be scanned */
2639 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2640 META_NAT, true);
2641
2642 f2fs_down_read_trace(&nm_i->nat_tree_lock, &lc);
2643
2644 while (1) {
2645 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2646 nm_i->nat_block_bitmap)) {
2647 struct folio *folio = get_current_nat_folio(sbi, nid);
2648
2649 if (IS_ERR(folio)) {
2650 ret = PTR_ERR(folio);
2651 } else {
2652 ret = scan_nat_page(sbi, folio_address(folio),
2653 nid);
2654 f2fs_folio_put(folio, true);
2655 }
2656
2657 if (ret) {
2658 f2fs_up_read_trace(&nm_i->nat_tree_lock, &lc);
2659
2660 if (ret == -EFSCORRUPTED) {
2661 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2662 set_sbi_flag(sbi, SBI_NEED_FSCK);
2663 f2fs_handle_error(sbi,
2664 ERROR_INCONSISTENT_NAT);
2665 }
2666
2667 return ret;
2668 }
2669 }
2670
2671 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2672 if (unlikely(nid >= nm_i->max_nid))
2673 nid = 0;
2674
2675 if (++i >= FREE_NID_PAGES)
2676 break;
2677 }
2678
2679 /* go to the next free nat pages to find free nids abundantly */
2680 nm_i->next_scan_nid = nid;
2681
2682 /* find free nids from current sum_pages */
2683 scan_curseg_cache(sbi);
2684
2685 f2fs_up_read_trace(&nm_i->nat_tree_lock, &lc);
2686
2687 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2688 nm_i->ra_nid_pages, META_NAT, false);
2689
2690 return 0;
2691}
2692
2693int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2694{
2695 int ret;
2696
2697 mutex_lock(&NM_I(sbi)->build_lock);
2698 ret = __f2fs_build_free_nids(sbi, sync, mount);
2699 mutex_unlock(&NM_I(sbi)->build_lock);
2700
2701 return ret;
2702}
2703
2704/*
2705 * If this function returns success, caller can obtain a new nid
2706 * from second parameter of this function.
2707 * The returned nid could be used ino as well as nid when inode is created.
2708 */
2709bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2710{
2711 struct f2fs_nm_info *nm_i = NM_I(sbi);
2712 struct free_nid *i = NULL;
2713retry:
2714 if (time_to_inject(sbi, FAULT_ALLOC_NID))
2715 return false;
2716
2717 spin_lock(&nm_i->nid_list_lock);
2718
2719 if (unlikely(nm_i->available_nids == 0)) {
2720 spin_unlock(&nm_i->nid_list_lock);
2721 return false;
2722 }
2723
2724 /* We should not use stale free nids created by f2fs_build_free_nids */
2725 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2726 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2727 i = list_first_entry(&nm_i->free_nid_list,
2728 struct free_nid, list);
2729
2730 if (unlikely(is_invalid_nid(sbi, i->nid))) {
2731 spin_unlock(&nm_i->nid_list_lock);
2732 f2fs_err(sbi, "Corrupted nid %u in free_nid_list",
2733 i->nid);
2734 fserror_report_metadata(sbi->sb, -EFSCORRUPTED,
2735 GFP_NOFS);
2736 f2fs_stop_checkpoint(sbi, false,
2737 STOP_CP_REASON_CORRUPTED_NID);
2738 return false;
2739 }
2740
2741 *nid = i->nid;
2742
2743 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2744 nm_i->available_nids--;
2745
2746 update_free_nid_bitmap(sbi, *nid, false, false);
2747
2748 spin_unlock(&nm_i->nid_list_lock);
2749 return true;
2750 }
2751 spin_unlock(&nm_i->nid_list_lock);
2752
2753 /* Let's scan nat pages and its caches to get free nids */
2754 if (!f2fs_build_free_nids(sbi, true, false))
2755 goto retry;
2756 return false;
2757}
2758
2759/*
2760 * f2fs_alloc_nid() should be called prior to this function.
2761 */
2762void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2763{
2764 struct f2fs_nm_info *nm_i = NM_I(sbi);
2765 struct free_nid *i;
2766
2767 spin_lock(&nm_i->nid_list_lock);
2768 i = __lookup_free_nid_list(nm_i, nid);
2769 f2fs_bug_on(sbi, !i);
2770 __remove_free_nid(sbi, i, PREALLOC_NID);
2771 spin_unlock(&nm_i->nid_list_lock);
2772
2773 kmem_cache_free(free_nid_slab, i);
2774}
2775
2776/*
2777 * f2fs_alloc_nid() should be called prior to this function.
2778 */
2779void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2780{
2781 struct f2fs_nm_info *nm_i = NM_I(sbi);
2782 struct free_nid *i;
2783 bool need_free = false;
2784
2785 if (!nid)
2786 return;
2787
2788 spin_lock(&nm_i->nid_list_lock);
2789 i = __lookup_free_nid_list(nm_i, nid);
2790 f2fs_bug_on(sbi, !i);
2791
2792 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2793 __remove_free_nid(sbi, i, PREALLOC_NID);
2794 need_free = true;
2795 } else {
2796 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2797 }
2798
2799 nm_i->available_nids++;
2800
2801 update_free_nid_bitmap(sbi, nid, true, false);
2802
2803 spin_unlock(&nm_i->nid_list_lock);
2804
2805 if (need_free)
2806 kmem_cache_free(free_nid_slab, i);
2807}
2808
2809int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2810{
2811 struct f2fs_nm_info *nm_i = NM_I(sbi);
2812 int nr = nr_shrink;
2813
2814 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2815 return 0;
2816
2817 if (!mutex_trylock(&nm_i->build_lock))
2818 return 0;
2819
2820 while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2821 struct free_nid *i, *next;
2822 unsigned int batch = SHRINK_NID_BATCH_SIZE;
2823
2824 spin_lock(&nm_i->nid_list_lock);
2825 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2826 if (!nr_shrink || !batch ||
2827 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2828 break;
2829 __remove_free_nid(sbi, i, FREE_NID);
2830 kmem_cache_free(free_nid_slab, i);
2831 nr_shrink--;
2832 batch--;
2833 }
2834 spin_unlock(&nm_i->nid_list_lock);
2835 }
2836
2837 mutex_unlock(&nm_i->build_lock);
2838
2839 return nr - nr_shrink;
2840}
2841
2842int f2fs_recover_inline_xattr(struct inode *inode, struct folio *folio)
2843{
2844 void *src_addr, *dst_addr;
2845 size_t inline_size;
2846 struct folio *ifolio;
2847 struct f2fs_inode *ri;
2848
2849 ifolio = f2fs_get_inode_folio(F2FS_I_SB(inode), inode->i_ino);
2850 if (IS_ERR(ifolio))
2851 return PTR_ERR(ifolio);
2852
2853 ri = F2FS_INODE(folio);
2854 if (ri->i_inline & F2FS_INLINE_XATTR) {
2855 if (!f2fs_has_inline_xattr(inode)) {
2856 set_inode_flag(inode, FI_INLINE_XATTR);
2857 stat_inc_inline_xattr(inode);
2858 }
2859 } else {
2860 if (f2fs_has_inline_xattr(inode)) {
2861 stat_dec_inline_xattr(inode);
2862 clear_inode_flag(inode, FI_INLINE_XATTR);
2863 }
2864 goto update_inode;
2865 }
2866
2867 dst_addr = inline_xattr_addr(inode, ifolio);
2868 src_addr = inline_xattr_addr(inode, folio);
2869 inline_size = inline_xattr_size(inode);
2870
2871 f2fs_folio_wait_writeback(ifolio, NODE, true, true);
2872 memcpy(dst_addr, src_addr, inline_size);
2873update_inode:
2874 f2fs_update_inode(inode, ifolio);
2875 f2fs_folio_put(ifolio, true);
2876 return 0;
2877}
2878
2879int f2fs_recover_xattr_data(struct inode *inode, struct folio *folio)
2880{
2881 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2882 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2883 nid_t new_xnid;
2884 struct dnode_of_data dn;
2885 struct node_info ni;
2886 struct folio *xfolio;
2887 int err;
2888
2889 if (!prev_xnid)
2890 goto recover_xnid;
2891
2892 /* 1: invalidate the previous xattr nid */
2893 err = f2fs_get_node_info(sbi, prev_xnid, &ni, false);
2894 if (err)
2895 return err;
2896
2897 f2fs_invalidate_blocks(sbi, ni.blk_addr, 1);
2898 dec_valid_node_count(sbi, inode, false);
2899 set_node_addr(sbi, &ni, NULL_ADDR, false);
2900
2901recover_xnid:
2902 /* 2: update xattr nid in inode */
2903 if (!f2fs_alloc_nid(sbi, &new_xnid))
2904 return -ENOSPC;
2905
2906 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2907 xfolio = f2fs_new_node_folio(&dn, XATTR_NODE_OFFSET);
2908 if (IS_ERR(xfolio)) {
2909 f2fs_alloc_nid_failed(sbi, new_xnid);
2910 return PTR_ERR(xfolio);
2911 }
2912
2913 f2fs_alloc_nid_done(sbi, new_xnid);
2914 f2fs_update_inode_page(inode);
2915
2916 /* 3: update and set xattr node page dirty */
2917 if (folio) {
2918 memcpy(F2FS_NODE(xfolio), F2FS_NODE(folio),
2919 VALID_XATTR_BLOCK_SIZE);
2920 folio_mark_dirty(xfolio);
2921 }
2922 f2fs_folio_put(xfolio, true);
2923
2924 return 0;
2925}
2926
2927int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct folio *folio)
2928{
2929 struct f2fs_inode *src, *dst;
2930 nid_t ino = ino_of_node(folio);
2931 struct node_info old_ni, new_ni;
2932 struct folio *ifolio;
2933 int err;
2934
2935 err = f2fs_get_node_info(sbi, ino, &old_ni, false);
2936 if (err)
2937 return err;
2938
2939 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2940 return -EINVAL;
2941retry:
2942 ifolio = f2fs_grab_cache_folio(NODE_MAPPING(sbi), ino, false);
2943 if (IS_ERR(ifolio)) {
2944 memalloc_retry_wait(GFP_NOFS);
2945 goto retry;
2946 }
2947
2948 /* Should not use this inode from free nid list */
2949 remove_free_nid(sbi, ino);
2950
2951 if (!folio_test_uptodate(ifolio))
2952 folio_mark_uptodate(ifolio);
2953 fill_node_footer(ifolio, ino, ino, 0, true);
2954 set_cold_node(ifolio, false);
2955
2956 src = F2FS_INODE(folio);
2957 dst = F2FS_INODE(ifolio);
2958
2959 memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2960 dst->i_size = 0;
2961 dst->i_blocks = cpu_to_le64(1);
2962 dst->i_links = cpu_to_le32(1);
2963 dst->i_xattr_nid = 0;
2964 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2965 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2966 dst->i_extra_isize = src->i_extra_isize;
2967
2968 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2969 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2970 i_inline_xattr_size))
2971 dst->i_inline_xattr_size = src->i_inline_xattr_size;
2972
2973 if (f2fs_sb_has_project_quota(sbi) &&
2974 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2975 i_projid))
2976 dst->i_projid = src->i_projid;
2977
2978 if (f2fs_sb_has_inode_crtime(sbi) &&
2979 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2980 i_crtime_nsec)) {
2981 dst->i_crtime = src->i_crtime;
2982 dst->i_crtime_nsec = src->i_crtime_nsec;
2983 }
2984 }
2985
2986 new_ni = old_ni;
2987 new_ni.ino = ino;
2988
2989 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2990 WARN_ON(1);
2991 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2992 inc_valid_inode_count(sbi);
2993 folio_mark_dirty(ifolio);
2994 f2fs_folio_put(ifolio, true);
2995 return 0;
2996}
2997
2998int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2999 unsigned int segno, struct f2fs_summary_block *sum)
3000{
3001 struct f2fs_node *rn;
3002 struct f2fs_summary *sum_entry;
3003 block_t addr;
3004 int i, idx, last_offset, nrpages;
3005
3006 /* scan the node segment */
3007 last_offset = BLKS_PER_SEG(sbi);
3008 addr = START_BLOCK(sbi, segno);
3009 sum_entry = sum_entries(sum);
3010
3011 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
3012 nrpages = bio_max_segs(last_offset - i);
3013
3014 /* readahead node pages */
3015 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
3016
3017 for (idx = addr; idx < addr + nrpages; idx++) {
3018 struct folio *folio = f2fs_get_tmp_folio(sbi, idx);
3019
3020 if (IS_ERR(folio))
3021 return PTR_ERR(folio);
3022
3023 rn = F2FS_NODE(folio);
3024 sum_entry->nid = rn->footer.nid;
3025 sum_entry->version = 0;
3026 sum_entry->ofs_in_node = 0;
3027 sum_entry++;
3028 f2fs_folio_put(folio, true);
3029 }
3030
3031 invalidate_mapping_pages(META_MAPPING(sbi), addr,
3032 addr + nrpages);
3033 }
3034 return 0;
3035}
3036
3037static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
3038{
3039 struct f2fs_nm_info *nm_i = NM_I(sbi);
3040 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3041 struct f2fs_journal *journal = curseg->journal;
3042 int i;
3043 bool init_dirty;
3044
3045 down_write(&curseg->journal_rwsem);
3046 for (i = 0; i < nats_in_cursum(journal); i++) {
3047 struct nat_entry *ne;
3048 struct f2fs_nat_entry raw_ne;
3049 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
3050
3051 if (f2fs_check_nid_range(sbi, nid))
3052 continue;
3053
3054 init_dirty = false;
3055
3056 raw_ne = nat_in_journal(journal, i);
3057
3058 ne = __lookup_nat_cache(nm_i, nid, true);
3059 if (!ne) {
3060 init_dirty = true;
3061 ne = __alloc_nat_entry(sbi, nid, true);
3062 __init_nat_entry(nm_i, ne, &raw_ne, true, true);
3063 }
3064
3065 /*
3066 * if a free nat in journal has not been used after last
3067 * checkpoint, we should remove it from available nids,
3068 * since later we will add it again.
3069 */
3070 if (!get_nat_flag(ne, IS_DIRTY) &&
3071 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
3072 spin_lock(&nm_i->nid_list_lock);
3073 nm_i->available_nids--;
3074 spin_unlock(&nm_i->nid_list_lock);
3075 }
3076
3077 __set_nat_cache_dirty(nm_i, ne, init_dirty);
3078 }
3079 update_nats_in_cursum(journal, -i);
3080 up_write(&curseg->journal_rwsem);
3081}
3082
3083static void __adjust_nat_entry_set(struct nat_entry_set *nes,
3084 struct list_head *head, int max)
3085{
3086 struct nat_entry_set *cur;
3087
3088 if (nes->entry_cnt >= max)
3089 goto add_out;
3090
3091 list_for_each_entry(cur, head, set_list) {
3092 if (cur->entry_cnt >= nes->entry_cnt) {
3093 list_add(&nes->set_list, cur->set_list.prev);
3094 return;
3095 }
3096 }
3097add_out:
3098 list_add_tail(&nes->set_list, head);
3099}
3100
3101static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
3102 const struct f2fs_nat_block *nat_blk)
3103{
3104 struct f2fs_nm_info *nm_i = NM_I(sbi);
3105 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
3106 int valid = 0;
3107 int i = 0;
3108
3109 if (!enabled_nat_bits(sbi, NULL))
3110 return;
3111
3112 if (nat_index == 0) {
3113 valid = 1;
3114 i = 1;
3115 }
3116 for (; i < NAT_ENTRY_PER_BLOCK; i++) {
3117 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
3118 valid++;
3119 }
3120 if (valid == 0) {
3121 __set_bit_le(nat_index, nm_i->empty_nat_bits);
3122 __clear_bit_le(nat_index, nm_i->full_nat_bits);
3123 return;
3124 }
3125
3126 __clear_bit_le(nat_index, nm_i->empty_nat_bits);
3127 if (valid == NAT_ENTRY_PER_BLOCK)
3128 __set_bit_le(nat_index, nm_i->full_nat_bits);
3129 else
3130 __clear_bit_le(nat_index, nm_i->full_nat_bits);
3131}
3132
3133static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
3134 struct nat_entry_set *set, struct cp_control *cpc)
3135{
3136 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3137 struct f2fs_journal *journal = curseg->journal;
3138 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
3139 bool to_journal = true;
3140 struct f2fs_nat_block *nat_blk;
3141 struct nat_entry *ne, *cur;
3142 struct folio *folio = NULL;
3143
3144 /*
3145 * there are two steps to flush nat entries:
3146 * #1, flush nat entries to journal in current hot data summary block.
3147 * #2, flush nat entries to nat page.
3148 */
3149 if (enabled_nat_bits(sbi, cpc) ||
3150 !__has_cursum_space(sbi, journal, set->entry_cnt, NAT_JOURNAL))
3151 to_journal = false;
3152
3153 if (to_journal) {
3154 down_write(&curseg->journal_rwsem);
3155 } else {
3156 folio = get_next_nat_folio(sbi, start_nid);
3157 if (IS_ERR(folio))
3158 return PTR_ERR(folio);
3159
3160 nat_blk = folio_address(folio);
3161 f2fs_bug_on(sbi, !nat_blk);
3162 }
3163
3164 /* flush dirty nats in nat entry set */
3165 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
3166 struct f2fs_nat_entry *raw_ne;
3167 nid_t nid = nat_get_nid(ne);
3168 int offset;
3169
3170 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
3171
3172 if (to_journal) {
3173 offset = f2fs_lookup_journal_in_cursum(sbi, journal,
3174 NAT_JOURNAL, nid, 1);
3175 f2fs_bug_on(sbi, offset < 0);
3176 raw_ne = &nat_in_journal(journal, offset);
3177 nid_in_journal(journal, offset) = cpu_to_le32(nid);
3178 } else {
3179 raw_ne = &nat_blk->entries[nid - start_nid];
3180 }
3181 raw_nat_from_node_info(raw_ne, &ne->ni);
3182 nat_reset_flag(ne);
3183 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
3184 if (nat_get_blkaddr(ne) == NULL_ADDR) {
3185 add_free_nid(sbi, nid, false, true);
3186 } else {
3187 spin_lock(&NM_I(sbi)->nid_list_lock);
3188 update_free_nid_bitmap(sbi, nid, false, false);
3189 spin_unlock(&NM_I(sbi)->nid_list_lock);
3190 }
3191 }
3192
3193 if (to_journal) {
3194 up_write(&curseg->journal_rwsem);
3195 } else {
3196 __update_nat_bits(sbi, start_nid, nat_blk);
3197 f2fs_folio_put(folio, true);
3198 }
3199
3200 /* Allow dirty nats by node block allocation in write_begin */
3201 if (!set->entry_cnt) {
3202 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
3203 kmem_cache_free(nat_entry_set_slab, set);
3204 }
3205 return 0;
3206}
3207
3208/*
3209 * This function is called during the checkpointing process.
3210 */
3211int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3212{
3213 struct f2fs_nm_info *nm_i = NM_I(sbi);
3214 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3215 struct f2fs_journal *journal = curseg->journal;
3216 struct nat_entry_set *setvec[NAT_VEC_SIZE];
3217 struct nat_entry_set *set, *tmp;
3218 unsigned int found, entry_count = 0;
3219 nid_t set_idx = 0;
3220 LIST_HEAD(sets);
3221 int err = 0;
3222 struct f2fs_lock_context lc;
3223
3224 /*
3225 * during unmount, let's flush nat_bits before checking
3226 * nat_cnt[DIRTY_NAT].
3227 */
3228 if (enabled_nat_bits(sbi, cpc)) {
3229 f2fs_down_write_trace(&nm_i->nat_tree_lock, &lc);
3230 remove_nats_in_journal(sbi);
3231 f2fs_up_write_trace(&nm_i->nat_tree_lock, &lc);
3232 }
3233
3234 if (!nm_i->nat_cnt[DIRTY_NAT])
3235 return 0;
3236
3237 f2fs_down_write_trace(&nm_i->nat_tree_lock, &lc);
3238
3239 /*
3240 * if there are no enough space in journal to store dirty nat
3241 * entries, remove all entries from journal and merge them
3242 * into nat entry set.
3243 */
3244 if (enabled_nat_bits(sbi, cpc) ||
3245 !__has_cursum_space(sbi, journal,
3246 nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3247 remove_nats_in_journal(sbi);
3248
3249 while ((found = __gang_lookup_nat_set(nm_i,
3250 set_idx, NAT_VEC_SIZE, setvec))) {
3251 unsigned idx;
3252
3253 set_idx = setvec[found - 1]->set + 1;
3254 for (idx = 0; idx < found; idx++)
3255 __adjust_nat_entry_set(setvec[idx], &sets,
3256 MAX_NAT_JENTRIES(sbi, journal));
3257 }
3258
3259 /*
3260 * Readahead the current NAT block to prevent read requests from
3261 * being issued and waited on one by one.
3262 */
3263 list_for_each_entry(set, &sets, set_list) {
3264 entry_count += set->entry_cnt;
3265 if (!enabled_nat_bits(sbi, cpc) &&
3266 __has_cursum_space(sbi, journal,
3267 entry_count, NAT_JOURNAL))
3268 continue;
3269 f2fs_ra_meta_pages(sbi, set->set, 1, META_NAT, true);
3270 }
3271 /* flush dirty nats in nat entry set */
3272 list_for_each_entry_safe(set, tmp, &sets, set_list) {
3273 err = __flush_nat_entry_set(sbi, set, cpc);
3274 if (err)
3275 break;
3276 }
3277
3278 f2fs_up_write_trace(&nm_i->nat_tree_lock, &lc);
3279 /* Allow dirty nats by node block allocation in write_begin */
3280
3281 return err;
3282}
3283
3284static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3285{
3286 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3287 struct f2fs_nm_info *nm_i = NM_I(sbi);
3288 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3289 unsigned int i;
3290 __u64 cp_ver = cur_cp_version(ckpt);
3291 block_t nat_bits_addr;
3292
3293 if (!enabled_nat_bits(sbi, NULL))
3294 return 0;
3295
3296 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3297 nm_i->nat_bits = f2fs_kvzalloc(sbi,
3298 F2FS_BLK_TO_BYTES(nm_i->nat_bits_blocks), GFP_KERNEL);
3299 if (!nm_i->nat_bits)
3300 return -ENOMEM;
3301
3302 nat_bits_addr = __start_cp_addr(sbi) + BLKS_PER_SEG(sbi) -
3303 nm_i->nat_bits_blocks;
3304 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3305 struct folio *folio;
3306
3307 folio = f2fs_get_meta_folio(sbi, nat_bits_addr++);
3308 if (IS_ERR(folio))
3309 return PTR_ERR(folio);
3310
3311 memcpy(nm_i->nat_bits + F2FS_BLK_TO_BYTES(i),
3312 folio_address(folio), F2FS_BLKSIZE);
3313 f2fs_folio_put(folio, true);
3314 }
3315
3316 cp_ver |= (cur_cp_crc(ckpt) << 32);
3317 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3318 disable_nat_bits(sbi, true);
3319 return 0;
3320 }
3321
3322 nm_i->full_nat_bits = nm_i->nat_bits + 8;
3323 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3324
3325 f2fs_notice(sbi, "Found nat_bits in checkpoint");
3326 return 0;
3327}
3328
3329static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3330{
3331 struct f2fs_nm_info *nm_i = NM_I(sbi);
3332 unsigned int i = 0;
3333 nid_t nid, last_nid;
3334
3335 if (!enabled_nat_bits(sbi, NULL))
3336 return;
3337
3338 for (i = 0; i < nm_i->nat_blocks; i++) {
3339 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3340 if (i >= nm_i->nat_blocks)
3341 break;
3342
3343 __set_bit_le(i, nm_i->nat_block_bitmap);
3344
3345 nid = i * NAT_ENTRY_PER_BLOCK;
3346 last_nid = nid + NAT_ENTRY_PER_BLOCK;
3347
3348 spin_lock(&NM_I(sbi)->nid_list_lock);
3349 for (; nid < last_nid; nid++)
3350 update_free_nid_bitmap(sbi, nid, true, true);
3351 spin_unlock(&NM_I(sbi)->nid_list_lock);
3352 }
3353
3354 for (i = 0; i < nm_i->nat_blocks; i++) {
3355 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3356 if (i >= nm_i->nat_blocks)
3357 break;
3358
3359 __set_bit_le(i, nm_i->nat_block_bitmap);
3360 }
3361}
3362
3363static int init_node_manager(struct f2fs_sb_info *sbi)
3364{
3365 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3366 struct f2fs_nm_info *nm_i = NM_I(sbi);
3367 unsigned char *version_bitmap;
3368 unsigned int nat_segs;
3369 int err;
3370
3371 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3372
3373 /* segment_count_nat includes pair segment so divide to 2. */
3374 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3375 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3376 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3377
3378 /* not used nids: 0, node, meta, (and root counted as valid node) */
3379 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3380 F2FS_RESERVED_NODE_NUM;
3381 nm_i->nid_cnt[FREE_NID] = 0;
3382 nm_i->nid_cnt[PREALLOC_NID] = 0;
3383 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3384 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3385 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3386 nm_i->max_rf_node_blocks = DEF_RF_NODE_BLOCKS;
3387
3388 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3389 INIT_LIST_HEAD(&nm_i->free_nid_list);
3390 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3391 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3392 INIT_LIST_HEAD(&nm_i->nat_entries);
3393 spin_lock_init(&nm_i->nat_list_lock);
3394
3395 mutex_init(&nm_i->build_lock);
3396 spin_lock_init(&nm_i->nid_list_lock);
3397 init_f2fs_rwsem_trace(&nm_i->nat_tree_lock, sbi,
3398 LOCK_NAME_NAT_TREE_LOCK);
3399
3400 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3401 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3402 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3403 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3404 GFP_KERNEL);
3405 if (!nm_i->nat_bitmap)
3406 return -ENOMEM;
3407
3408 if (!test_opt(sbi, NAT_BITS))
3409 disable_nat_bits(sbi, true);
3410
3411 err = __get_nat_bitmaps(sbi);
3412 if (err)
3413 return err;
3414
3415#ifdef CONFIG_F2FS_CHECK_FS
3416 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3417 GFP_KERNEL);
3418 if (!nm_i->nat_bitmap_mir)
3419 return -ENOMEM;
3420#endif
3421
3422 return 0;
3423}
3424
3425static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3426{
3427 struct f2fs_nm_info *nm_i = NM_I(sbi);
3428 int i;
3429
3430 nm_i->free_nid_bitmap =
3431 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3432 nm_i->nat_blocks),
3433 GFP_KERNEL);
3434 if (!nm_i->free_nid_bitmap)
3435 return -ENOMEM;
3436
3437 for (i = 0; i < nm_i->nat_blocks; i++) {
3438 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3439 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3440 if (!nm_i->free_nid_bitmap[i])
3441 return -ENOMEM;
3442 }
3443
3444 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3445 GFP_KERNEL);
3446 if (!nm_i->nat_block_bitmap)
3447 return -ENOMEM;
3448
3449 nm_i->free_nid_count =
3450 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3451 nm_i->nat_blocks),
3452 GFP_KERNEL);
3453 if (!nm_i->free_nid_count)
3454 return -ENOMEM;
3455 return 0;
3456}
3457
3458int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3459{
3460 int err;
3461
3462 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3463 GFP_KERNEL);
3464 if (!sbi->nm_info)
3465 return -ENOMEM;
3466
3467 err = init_node_manager(sbi);
3468 if (err)
3469 return err;
3470
3471 err = init_free_nid_cache(sbi);
3472 if (err)
3473 return err;
3474
3475 /* load free nid status from nat_bits table */
3476 load_free_nid_bitmap(sbi);
3477
3478 return f2fs_build_free_nids(sbi, true, true);
3479}
3480
3481void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3482{
3483 struct f2fs_nm_info *nm_i = NM_I(sbi);
3484 struct free_nid *i, *next_i;
3485 void *vec[NAT_VEC_SIZE];
3486 struct nat_entry **natvec = (struct nat_entry **)vec;
3487 struct nat_entry_set **setvec = (struct nat_entry_set **)vec;
3488 nid_t nid = 0;
3489 unsigned int found;
3490 struct f2fs_lock_context lc;
3491
3492 if (!nm_i)
3493 return;
3494
3495 /* destroy free nid list */
3496 spin_lock(&nm_i->nid_list_lock);
3497 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3498 __remove_free_nid(sbi, i, FREE_NID);
3499 spin_unlock(&nm_i->nid_list_lock);
3500 kmem_cache_free(free_nid_slab, i);
3501 spin_lock(&nm_i->nid_list_lock);
3502 }
3503 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3504 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3505 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3506 spin_unlock(&nm_i->nid_list_lock);
3507
3508 /* destroy nat cache */
3509 f2fs_down_write_trace(&nm_i->nat_tree_lock, &lc);
3510 while ((found = __gang_lookup_nat_cache(nm_i,
3511 nid, NAT_VEC_SIZE, natvec))) {
3512 unsigned idx;
3513
3514 nid = nat_get_nid(natvec[found - 1]) + 1;
3515 for (idx = 0; idx < found; idx++) {
3516 spin_lock(&nm_i->nat_list_lock);
3517 list_del(&natvec[idx]->list);
3518 spin_unlock(&nm_i->nat_list_lock);
3519
3520 __del_from_nat_cache(nm_i, natvec[idx]);
3521 }
3522 }
3523 f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3524
3525 /* destroy nat set cache */
3526 nid = 0;
3527 memset(vec, 0, sizeof(void *) * NAT_VEC_SIZE);
3528 while ((found = __gang_lookup_nat_set(nm_i,
3529 nid, NAT_VEC_SIZE, setvec))) {
3530 unsigned idx;
3531
3532 nid = setvec[found - 1]->set + 1;
3533 for (idx = 0; idx < found; idx++) {
3534 /* entry_cnt is not zero, when cp_error was occurred */
3535 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3536 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3537 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3538 }
3539 }
3540 f2fs_up_write_trace(&nm_i->nat_tree_lock, &lc);
3541
3542 kvfree(nm_i->nat_block_bitmap);
3543 if (nm_i->free_nid_bitmap) {
3544 int i;
3545
3546 for (i = 0; i < nm_i->nat_blocks; i++)
3547 kvfree(nm_i->free_nid_bitmap[i]);
3548 kvfree(nm_i->free_nid_bitmap);
3549 }
3550 kvfree(nm_i->free_nid_count);
3551
3552 kfree(nm_i->nat_bitmap);
3553 kvfree(nm_i->nat_bits);
3554#ifdef CONFIG_F2FS_CHECK_FS
3555 kfree(nm_i->nat_bitmap_mir);
3556#endif
3557 sbi->nm_info = NULL;
3558 kfree(nm_i);
3559}
3560
3561int __init f2fs_create_node_manager_caches(void)
3562{
3563 nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3564 sizeof(struct nat_entry));
3565 if (!nat_entry_slab)
3566 goto fail;
3567
3568 free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3569 sizeof(struct free_nid));
3570 if (!free_nid_slab)
3571 goto destroy_nat_entry;
3572
3573 nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3574 sizeof(struct nat_entry_set));
3575 if (!nat_entry_set_slab)
3576 goto destroy_free_nid;
3577
3578 fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3579 sizeof(struct fsync_node_entry));
3580 if (!fsync_node_entry_slab)
3581 goto destroy_nat_entry_set;
3582 return 0;
3583
3584destroy_nat_entry_set:
3585 kmem_cache_destroy(nat_entry_set_slab);
3586destroy_free_nid:
3587 kmem_cache_destroy(free_nid_slab);
3588destroy_nat_entry:
3589 kmem_cache_destroy(nat_entry_slab);
3590fail:
3591 return -ENOMEM;
3592}
3593
3594void f2fs_destroy_node_manager_caches(void)
3595{
3596 kmem_cache_destroy(fsync_node_entry_slab);
3597 kmem_cache_destroy(nat_entry_set_slab);
3598 kmem_cache_destroy(free_nid_slab);
3599 kmem_cache_destroy(nat_entry_slab);
3600}