xref: /linux/fs/f2fs/node.c (revision c8b90d40d5bba8e6fba457b8a7c10d3c0d467e37)
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/pagevec.h>
14 #include <linux/swap.h>
15 
16 #include "f2fs.h"
17 #include "node.h"
18 #include "segment.h"
19 #include "xattr.h"
20 #include "iostat.h"
21 #include <trace/events/f2fs.h>
22 
23 #define on_f2fs_build_free_nids(nm_i) mutex_is_locked(&(nm_i)->build_lock)
24 
25 static struct kmem_cache *nat_entry_slab;
26 static struct kmem_cache *free_nid_slab;
27 static struct kmem_cache *nat_entry_set_slab;
28 static struct kmem_cache *fsync_node_entry_slab;
29 
30 /*
31  * Check whether the given nid is within node id range.
32  */
33 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
34 {
35 	if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
36 		set_sbi_flag(sbi, SBI_NEED_FSCK);
37 		f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
38 			  __func__, nid);
39 		f2fs_handle_error(sbi, ERROR_CORRUPTED_INODE);
40 		return -EFSCORRUPTED;
41 	}
42 	return 0;
43 }
44 
45 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
46 {
47 	struct f2fs_nm_info *nm_i = NM_I(sbi);
48 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
49 	struct sysinfo val;
50 	unsigned long avail_ram;
51 	unsigned long mem_size = 0;
52 	bool res = false;
53 
54 	if (!nm_i)
55 		return true;
56 
57 	si_meminfo(&val);
58 
59 	/* only uses low memory */
60 	avail_ram = val.totalram - val.totalhigh;
61 
62 	/*
63 	 * give 25%, 25%, 50%, 50%, 25%, 25% memory for each components respectively
64 	 */
65 	if (type == FREE_NIDS) {
66 		mem_size = (nm_i->nid_cnt[FREE_NID] *
67 				sizeof(struct free_nid)) >> PAGE_SHIFT;
68 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
69 	} else if (type == NAT_ENTRIES) {
70 		mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
71 				sizeof(struct nat_entry)) >> PAGE_SHIFT;
72 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
73 		if (excess_cached_nats(sbi))
74 			res = false;
75 	} else if (type == DIRTY_DENTS) {
76 		if (sbi->sb->s_bdi->wb.dirty_exceeded)
77 			return false;
78 		mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
79 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
80 	} else if (type == INO_ENTRIES) {
81 		int i;
82 
83 		for (i = 0; i < MAX_INO_ENTRY; i++)
84 			mem_size += sbi->im[i].ino_num *
85 						sizeof(struct ino_entry);
86 		mem_size >>= PAGE_SHIFT;
87 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
88 	} else if (type == READ_EXTENT_CACHE || type == AGE_EXTENT_CACHE) {
89 		enum extent_type etype = type == READ_EXTENT_CACHE ?
90 						EX_READ : EX_BLOCK_AGE;
91 		struct extent_tree_info *eti = &sbi->extent_tree[etype];
92 
93 		mem_size = (atomic_read(&eti->total_ext_tree) *
94 				sizeof(struct extent_tree) +
95 				atomic_read(&eti->total_ext_node) *
96 				sizeof(struct extent_node)) >> PAGE_SHIFT;
97 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
98 	} else if (type == DISCARD_CACHE) {
99 		mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
100 				sizeof(struct discard_cmd)) >> PAGE_SHIFT;
101 		res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
102 	} else if (type == COMPRESS_PAGE) {
103 #ifdef CONFIG_F2FS_FS_COMPRESSION
104 		unsigned long free_ram = val.freeram;
105 
106 		/*
107 		 * free memory is lower than watermark or cached page count
108 		 * exceed threshold, deny caching compress page.
109 		 */
110 		res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
111 			(COMPRESS_MAPPING(sbi)->nrpages <
112 			 free_ram * sbi->compress_percent / 100);
113 #else
114 		res = false;
115 #endif
116 	} else {
117 		if (!sbi->sb->s_bdi->wb.dirty_exceeded)
118 			return true;
119 	}
120 	return res;
121 }
122 
123 static void clear_node_page_dirty(struct page *page)
124 {
125 	if (PageDirty(page)) {
126 		f2fs_clear_page_cache_dirty_tag(page_folio(page));
127 		clear_page_dirty_for_io(page);
128 		dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
129 	}
130 	ClearPageUptodate(page);
131 }
132 
133 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
134 {
135 	return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
136 }
137 
138 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
139 {
140 	struct page *src_page;
141 	struct page *dst_page;
142 	pgoff_t dst_off;
143 	void *src_addr;
144 	void *dst_addr;
145 	struct f2fs_nm_info *nm_i = NM_I(sbi);
146 
147 	dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
148 
149 	/* get current nat block page with lock */
150 	src_page = get_current_nat_page(sbi, nid);
151 	if (IS_ERR(src_page))
152 		return src_page;
153 	dst_page = f2fs_grab_meta_page(sbi, dst_off);
154 	f2fs_bug_on(sbi, PageDirty(src_page));
155 
156 	src_addr = page_address(src_page);
157 	dst_addr = page_address(dst_page);
158 	memcpy(dst_addr, src_addr, PAGE_SIZE);
159 	set_page_dirty(dst_page);
160 	f2fs_put_page(src_page, 1);
161 
162 	set_to_next_nat(nm_i, nid);
163 
164 	return dst_page;
165 }
166 
167 static struct nat_entry *__alloc_nat_entry(struct f2fs_sb_info *sbi,
168 						nid_t nid, bool no_fail)
169 {
170 	struct nat_entry *new;
171 
172 	new = f2fs_kmem_cache_alloc(nat_entry_slab,
173 					GFP_F2FS_ZERO, no_fail, sbi);
174 	if (new) {
175 		nat_set_nid(new, nid);
176 		nat_reset_flag(new);
177 	}
178 	return new;
179 }
180 
181 static void __free_nat_entry(struct nat_entry *e)
182 {
183 	kmem_cache_free(nat_entry_slab, e);
184 }
185 
186 /* must be locked by nat_tree_lock */
187 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
188 	struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
189 {
190 	if (no_fail)
191 		f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
192 	else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
193 		return NULL;
194 
195 	if (raw_ne)
196 		node_info_from_raw_nat(&ne->ni, raw_ne);
197 
198 	spin_lock(&nm_i->nat_list_lock);
199 	list_add_tail(&ne->list, &nm_i->nat_entries);
200 	spin_unlock(&nm_i->nat_list_lock);
201 
202 	nm_i->nat_cnt[TOTAL_NAT]++;
203 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
204 	return ne;
205 }
206 
207 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
208 {
209 	struct nat_entry *ne;
210 
211 	ne = radix_tree_lookup(&nm_i->nat_root, n);
212 
213 	/* for recent accessed nat entry, move it to tail of lru list */
214 	if (ne && !get_nat_flag(ne, IS_DIRTY)) {
215 		spin_lock(&nm_i->nat_list_lock);
216 		if (!list_empty(&ne->list))
217 			list_move_tail(&ne->list, &nm_i->nat_entries);
218 		spin_unlock(&nm_i->nat_list_lock);
219 	}
220 
221 	return ne;
222 }
223 
224 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
225 		nid_t start, unsigned int nr, struct nat_entry **ep)
226 {
227 	return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
228 }
229 
230 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
231 {
232 	radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
233 	nm_i->nat_cnt[TOTAL_NAT]--;
234 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
235 	__free_nat_entry(e);
236 }
237 
238 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
239 							struct nat_entry *ne)
240 {
241 	nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
242 	struct nat_entry_set *head;
243 
244 	head = radix_tree_lookup(&nm_i->nat_set_root, set);
245 	if (!head) {
246 		head = f2fs_kmem_cache_alloc(nat_entry_set_slab,
247 						GFP_NOFS, true, NULL);
248 
249 		INIT_LIST_HEAD(&head->entry_list);
250 		INIT_LIST_HEAD(&head->set_list);
251 		head->set = set;
252 		head->entry_cnt = 0;
253 		f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
254 	}
255 	return head;
256 }
257 
258 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
259 						struct nat_entry *ne)
260 {
261 	struct nat_entry_set *head;
262 	bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
263 
264 	if (!new_ne)
265 		head = __grab_nat_entry_set(nm_i, ne);
266 
267 	/*
268 	 * update entry_cnt in below condition:
269 	 * 1. update NEW_ADDR to valid block address;
270 	 * 2. update old block address to new one;
271 	 */
272 	if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
273 				!get_nat_flag(ne, IS_DIRTY)))
274 		head->entry_cnt++;
275 
276 	set_nat_flag(ne, IS_PREALLOC, new_ne);
277 
278 	if (get_nat_flag(ne, IS_DIRTY))
279 		goto refresh_list;
280 
281 	nm_i->nat_cnt[DIRTY_NAT]++;
282 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
283 	set_nat_flag(ne, IS_DIRTY, true);
284 refresh_list:
285 	spin_lock(&nm_i->nat_list_lock);
286 	if (new_ne)
287 		list_del_init(&ne->list);
288 	else
289 		list_move_tail(&ne->list, &head->entry_list);
290 	spin_unlock(&nm_i->nat_list_lock);
291 }
292 
293 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
294 		struct nat_entry_set *set, struct nat_entry *ne)
295 {
296 	spin_lock(&nm_i->nat_list_lock);
297 	list_move_tail(&ne->list, &nm_i->nat_entries);
298 	spin_unlock(&nm_i->nat_list_lock);
299 
300 	set_nat_flag(ne, IS_DIRTY, false);
301 	set->entry_cnt--;
302 	nm_i->nat_cnt[DIRTY_NAT]--;
303 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
304 }
305 
306 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
307 		nid_t start, unsigned int nr, struct nat_entry_set **ep)
308 {
309 	return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
310 							start, nr);
311 }
312 
313 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
314 {
315 	return NODE_MAPPING(sbi) == page->mapping &&
316 			IS_DNODE(page) && is_cold_node(page);
317 }
318 
319 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
320 {
321 	spin_lock_init(&sbi->fsync_node_lock);
322 	INIT_LIST_HEAD(&sbi->fsync_node_list);
323 	sbi->fsync_seg_id = 0;
324 	sbi->fsync_node_num = 0;
325 }
326 
327 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
328 							struct page *page)
329 {
330 	struct fsync_node_entry *fn;
331 	unsigned long flags;
332 	unsigned int seq_id;
333 
334 	fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab,
335 					GFP_NOFS, true, NULL);
336 
337 	get_page(page);
338 	fn->page = page;
339 	INIT_LIST_HEAD(&fn->list);
340 
341 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
342 	list_add_tail(&fn->list, &sbi->fsync_node_list);
343 	fn->seq_id = sbi->fsync_seg_id++;
344 	seq_id = fn->seq_id;
345 	sbi->fsync_node_num++;
346 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
347 
348 	return seq_id;
349 }
350 
351 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
352 {
353 	struct fsync_node_entry *fn;
354 	unsigned long flags;
355 
356 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
357 	list_for_each_entry(fn, &sbi->fsync_node_list, list) {
358 		if (fn->page == page) {
359 			list_del(&fn->list);
360 			sbi->fsync_node_num--;
361 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
362 			kmem_cache_free(fsync_node_entry_slab, fn);
363 			put_page(page);
364 			return;
365 		}
366 	}
367 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
368 	f2fs_bug_on(sbi, 1);
369 }
370 
371 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
372 {
373 	unsigned long flags;
374 
375 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
376 	sbi->fsync_seg_id = 0;
377 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
378 }
379 
380 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
381 {
382 	struct f2fs_nm_info *nm_i = NM_I(sbi);
383 	struct nat_entry *e;
384 	bool need = false;
385 
386 	f2fs_down_read(&nm_i->nat_tree_lock);
387 	e = __lookup_nat_cache(nm_i, nid);
388 	if (e) {
389 		if (!get_nat_flag(e, IS_CHECKPOINTED) &&
390 				!get_nat_flag(e, HAS_FSYNCED_INODE))
391 			need = true;
392 	}
393 	f2fs_up_read(&nm_i->nat_tree_lock);
394 	return need;
395 }
396 
397 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
398 {
399 	struct f2fs_nm_info *nm_i = NM_I(sbi);
400 	struct nat_entry *e;
401 	bool is_cp = true;
402 
403 	f2fs_down_read(&nm_i->nat_tree_lock);
404 	e = __lookup_nat_cache(nm_i, nid);
405 	if (e && !get_nat_flag(e, IS_CHECKPOINTED))
406 		is_cp = false;
407 	f2fs_up_read(&nm_i->nat_tree_lock);
408 	return is_cp;
409 }
410 
411 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
412 {
413 	struct f2fs_nm_info *nm_i = NM_I(sbi);
414 	struct nat_entry *e;
415 	bool need_update = true;
416 
417 	f2fs_down_read(&nm_i->nat_tree_lock);
418 	e = __lookup_nat_cache(nm_i, ino);
419 	if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
420 			(get_nat_flag(e, IS_CHECKPOINTED) ||
421 			 get_nat_flag(e, HAS_FSYNCED_INODE)))
422 		need_update = false;
423 	f2fs_up_read(&nm_i->nat_tree_lock);
424 	return need_update;
425 }
426 
427 /* must be locked by nat_tree_lock */
428 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
429 						struct f2fs_nat_entry *ne)
430 {
431 	struct f2fs_nm_info *nm_i = NM_I(sbi);
432 	struct nat_entry *new, *e;
433 
434 	/* Let's mitigate lock contention of nat_tree_lock during checkpoint */
435 	if (f2fs_rwsem_is_locked(&sbi->cp_global_sem))
436 		return;
437 
438 	new = __alloc_nat_entry(sbi, nid, false);
439 	if (!new)
440 		return;
441 
442 	f2fs_down_write(&nm_i->nat_tree_lock);
443 	e = __lookup_nat_cache(nm_i, nid);
444 	if (!e)
445 		e = __init_nat_entry(nm_i, new, ne, false);
446 	else
447 		f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
448 				nat_get_blkaddr(e) !=
449 					le32_to_cpu(ne->block_addr) ||
450 				nat_get_version(e) != ne->version);
451 	f2fs_up_write(&nm_i->nat_tree_lock);
452 	if (e != new)
453 		__free_nat_entry(new);
454 }
455 
456 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
457 			block_t new_blkaddr, bool fsync_done)
458 {
459 	struct f2fs_nm_info *nm_i = NM_I(sbi);
460 	struct nat_entry *e;
461 	struct nat_entry *new = __alloc_nat_entry(sbi, ni->nid, true);
462 
463 	f2fs_down_write(&nm_i->nat_tree_lock);
464 	e = __lookup_nat_cache(nm_i, ni->nid);
465 	if (!e) {
466 		e = __init_nat_entry(nm_i, new, NULL, true);
467 		copy_node_info(&e->ni, ni);
468 		f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
469 	} else if (new_blkaddr == NEW_ADDR) {
470 		/*
471 		 * when nid is reallocated,
472 		 * previous nat entry can be remained in nat cache.
473 		 * So, reinitialize it with new information.
474 		 */
475 		copy_node_info(&e->ni, ni);
476 		f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
477 	}
478 	/* let's free early to reduce memory consumption */
479 	if (e != new)
480 		__free_nat_entry(new);
481 
482 	/* sanity check */
483 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
484 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
485 			new_blkaddr == NULL_ADDR);
486 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
487 			new_blkaddr == NEW_ADDR);
488 	f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
489 			new_blkaddr == NEW_ADDR);
490 
491 	/* increment version no as node is removed */
492 	if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
493 		unsigned char version = nat_get_version(e);
494 
495 		nat_set_version(e, inc_node_version(version));
496 	}
497 
498 	/* change address */
499 	nat_set_blkaddr(e, new_blkaddr);
500 	if (!__is_valid_data_blkaddr(new_blkaddr))
501 		set_nat_flag(e, IS_CHECKPOINTED, false);
502 	__set_nat_cache_dirty(nm_i, e);
503 
504 	/* update fsync_mark if its inode nat entry is still alive */
505 	if (ni->nid != ni->ino)
506 		e = __lookup_nat_cache(nm_i, ni->ino);
507 	if (e) {
508 		if (fsync_done && ni->nid == ni->ino)
509 			set_nat_flag(e, HAS_FSYNCED_INODE, true);
510 		set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
511 	}
512 	f2fs_up_write(&nm_i->nat_tree_lock);
513 }
514 
515 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
516 {
517 	struct f2fs_nm_info *nm_i = NM_I(sbi);
518 	int nr = nr_shrink;
519 
520 	if (!f2fs_down_write_trylock(&nm_i->nat_tree_lock))
521 		return 0;
522 
523 	spin_lock(&nm_i->nat_list_lock);
524 	while (nr_shrink) {
525 		struct nat_entry *ne;
526 
527 		if (list_empty(&nm_i->nat_entries))
528 			break;
529 
530 		ne = list_first_entry(&nm_i->nat_entries,
531 					struct nat_entry, list);
532 		list_del(&ne->list);
533 		spin_unlock(&nm_i->nat_list_lock);
534 
535 		__del_from_nat_cache(nm_i, ne);
536 		nr_shrink--;
537 
538 		spin_lock(&nm_i->nat_list_lock);
539 	}
540 	spin_unlock(&nm_i->nat_list_lock);
541 
542 	f2fs_up_write(&nm_i->nat_tree_lock);
543 	return nr - nr_shrink;
544 }
545 
546 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
547 				struct node_info *ni, bool checkpoint_context)
548 {
549 	struct f2fs_nm_info *nm_i = NM_I(sbi);
550 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
551 	struct f2fs_journal *journal = curseg->journal;
552 	nid_t start_nid = START_NID(nid);
553 	struct f2fs_nat_block *nat_blk;
554 	struct page *page = NULL;
555 	struct f2fs_nat_entry ne;
556 	struct nat_entry *e;
557 	pgoff_t index;
558 	block_t blkaddr;
559 	int i;
560 
561 	ni->nid = nid;
562 retry:
563 	/* Check nat cache */
564 	f2fs_down_read(&nm_i->nat_tree_lock);
565 	e = __lookup_nat_cache(nm_i, nid);
566 	if (e) {
567 		ni->ino = nat_get_ino(e);
568 		ni->blk_addr = nat_get_blkaddr(e);
569 		ni->version = nat_get_version(e);
570 		f2fs_up_read(&nm_i->nat_tree_lock);
571 		return 0;
572 	}
573 
574 	/*
575 	 * Check current segment summary by trying to grab journal_rwsem first.
576 	 * This sem is on the critical path on the checkpoint requiring the above
577 	 * nat_tree_lock. Therefore, we should retry, if we failed to grab here
578 	 * while not bothering checkpoint.
579 	 */
580 	if (!f2fs_rwsem_is_locked(&sbi->cp_global_sem) || checkpoint_context) {
581 		down_read(&curseg->journal_rwsem);
582 	} else if (f2fs_rwsem_is_contended(&nm_i->nat_tree_lock) ||
583 				!down_read_trylock(&curseg->journal_rwsem)) {
584 		f2fs_up_read(&nm_i->nat_tree_lock);
585 		goto retry;
586 	}
587 
588 	i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
589 	if (i >= 0) {
590 		ne = nat_in_journal(journal, i);
591 		node_info_from_raw_nat(ni, &ne);
592 	}
593 	up_read(&curseg->journal_rwsem);
594 	if (i >= 0) {
595 		f2fs_up_read(&nm_i->nat_tree_lock);
596 		goto cache;
597 	}
598 
599 	/* Fill node_info from nat page */
600 	index = current_nat_addr(sbi, nid);
601 	f2fs_up_read(&nm_i->nat_tree_lock);
602 
603 	page = f2fs_get_meta_page(sbi, index);
604 	if (IS_ERR(page))
605 		return PTR_ERR(page);
606 
607 	nat_blk = (struct f2fs_nat_block *)page_address(page);
608 	ne = nat_blk->entries[nid - start_nid];
609 	node_info_from_raw_nat(ni, &ne);
610 	f2fs_put_page(page, 1);
611 cache:
612 	blkaddr = le32_to_cpu(ne.block_addr);
613 	if (__is_valid_data_blkaddr(blkaddr) &&
614 		!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
615 		return -EFAULT;
616 
617 	/* cache nat entry */
618 	cache_nat_entry(sbi, nid, &ne);
619 	return 0;
620 }
621 
622 /*
623  * readahead MAX_RA_NODE number of node pages.
624  */
625 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
626 {
627 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
628 	struct blk_plug plug;
629 	int i, end;
630 	nid_t nid;
631 
632 	blk_start_plug(&plug);
633 
634 	/* Then, try readahead for siblings of the desired node */
635 	end = start + n;
636 	end = min(end, (int)NIDS_PER_BLOCK);
637 	for (i = start; i < end; i++) {
638 		nid = get_nid(parent, i, false);
639 		f2fs_ra_node_page(sbi, nid);
640 	}
641 
642 	blk_finish_plug(&plug);
643 }
644 
645 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
646 {
647 	const long direct_index = ADDRS_PER_INODE(dn->inode);
648 	const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
649 	const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
650 	unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
651 	int cur_level = dn->cur_level;
652 	int max_level = dn->max_level;
653 	pgoff_t base = 0;
654 
655 	if (!dn->max_level)
656 		return pgofs + 1;
657 
658 	while (max_level-- > cur_level)
659 		skipped_unit *= NIDS_PER_BLOCK;
660 
661 	switch (dn->max_level) {
662 	case 3:
663 		base += 2 * indirect_blks;
664 		fallthrough;
665 	case 2:
666 		base += 2 * direct_blks;
667 		fallthrough;
668 	case 1:
669 		base += direct_index;
670 		break;
671 	default:
672 		f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
673 	}
674 
675 	return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
676 }
677 
678 /*
679  * The maximum depth is four.
680  * Offset[0] will have raw inode offset.
681  */
682 static int get_node_path(struct inode *inode, long block,
683 				int offset[4], unsigned int noffset[4])
684 {
685 	const long direct_index = ADDRS_PER_INODE(inode);
686 	const long direct_blks = ADDRS_PER_BLOCK(inode);
687 	const long dptrs_per_blk = NIDS_PER_BLOCK;
688 	const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
689 	const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
690 	int n = 0;
691 	int level = 0;
692 
693 	noffset[0] = 0;
694 
695 	if (block < direct_index) {
696 		offset[n] = block;
697 		goto got;
698 	}
699 	block -= direct_index;
700 	if (block < direct_blks) {
701 		offset[n++] = NODE_DIR1_BLOCK;
702 		noffset[n] = 1;
703 		offset[n] = block;
704 		level = 1;
705 		goto got;
706 	}
707 	block -= direct_blks;
708 	if (block < direct_blks) {
709 		offset[n++] = NODE_DIR2_BLOCK;
710 		noffset[n] = 2;
711 		offset[n] = block;
712 		level = 1;
713 		goto got;
714 	}
715 	block -= direct_blks;
716 	if (block < indirect_blks) {
717 		offset[n++] = NODE_IND1_BLOCK;
718 		noffset[n] = 3;
719 		offset[n++] = block / direct_blks;
720 		noffset[n] = 4 + offset[n - 1];
721 		offset[n] = block % direct_blks;
722 		level = 2;
723 		goto got;
724 	}
725 	block -= indirect_blks;
726 	if (block < indirect_blks) {
727 		offset[n++] = NODE_IND2_BLOCK;
728 		noffset[n] = 4 + dptrs_per_blk;
729 		offset[n++] = block / direct_blks;
730 		noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
731 		offset[n] = block % direct_blks;
732 		level = 2;
733 		goto got;
734 	}
735 	block -= indirect_blks;
736 	if (block < dindirect_blks) {
737 		offset[n++] = NODE_DIND_BLOCK;
738 		noffset[n] = 5 + (dptrs_per_blk * 2);
739 		offset[n++] = block / indirect_blks;
740 		noffset[n] = 6 + (dptrs_per_blk * 2) +
741 			      offset[n - 1] * (dptrs_per_blk + 1);
742 		offset[n++] = (block / direct_blks) % dptrs_per_blk;
743 		noffset[n] = 7 + (dptrs_per_blk * 2) +
744 			      offset[n - 2] * (dptrs_per_blk + 1) +
745 			      offset[n - 1];
746 		offset[n] = block % direct_blks;
747 		level = 3;
748 		goto got;
749 	} else {
750 		return -E2BIG;
751 	}
752 got:
753 	return level;
754 }
755 
756 /*
757  * Caller should call f2fs_put_dnode(dn).
758  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
759  * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
760  */
761 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
762 {
763 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
764 	struct page *npage[4];
765 	struct page *parent = NULL;
766 	int offset[4];
767 	unsigned int noffset[4];
768 	nid_t nids[4];
769 	int level, i = 0;
770 	int err = 0;
771 
772 	level = get_node_path(dn->inode, index, offset, noffset);
773 	if (level < 0)
774 		return level;
775 
776 	nids[0] = dn->inode->i_ino;
777 	npage[0] = dn->inode_page;
778 
779 	if (!npage[0]) {
780 		npage[0] = f2fs_get_node_page(sbi, nids[0]);
781 		if (IS_ERR(npage[0]))
782 			return PTR_ERR(npage[0]);
783 	}
784 
785 	/* if inline_data is set, should not report any block indices */
786 	if (f2fs_has_inline_data(dn->inode) && index) {
787 		err = -ENOENT;
788 		f2fs_put_page(npage[0], 1);
789 		goto release_out;
790 	}
791 
792 	parent = npage[0];
793 	if (level != 0)
794 		nids[1] = get_nid(parent, offset[0], true);
795 	dn->inode_page = npage[0];
796 	dn->inode_page_locked = true;
797 
798 	/* get indirect or direct nodes */
799 	for (i = 1; i <= level; i++) {
800 		bool done = false;
801 
802 		if (!nids[i] && mode == ALLOC_NODE) {
803 			/* alloc new node */
804 			if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
805 				err = -ENOSPC;
806 				goto release_pages;
807 			}
808 
809 			dn->nid = nids[i];
810 			npage[i] = f2fs_new_node_page(dn, noffset[i]);
811 			if (IS_ERR(npage[i])) {
812 				f2fs_alloc_nid_failed(sbi, nids[i]);
813 				err = PTR_ERR(npage[i]);
814 				goto release_pages;
815 			}
816 
817 			set_nid(parent, offset[i - 1], nids[i], i == 1);
818 			f2fs_alloc_nid_done(sbi, nids[i]);
819 			done = true;
820 		} else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
821 			npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
822 			if (IS_ERR(npage[i])) {
823 				err = PTR_ERR(npage[i]);
824 				goto release_pages;
825 			}
826 			done = true;
827 		}
828 		if (i == 1) {
829 			dn->inode_page_locked = false;
830 			unlock_page(parent);
831 		} else {
832 			f2fs_put_page(parent, 1);
833 		}
834 
835 		if (!done) {
836 			npage[i] = f2fs_get_node_page(sbi, nids[i]);
837 			if (IS_ERR(npage[i])) {
838 				err = PTR_ERR(npage[i]);
839 				f2fs_put_page(npage[0], 0);
840 				goto release_out;
841 			}
842 		}
843 		if (i < level) {
844 			parent = npage[i];
845 			nids[i + 1] = get_nid(parent, offset[i], false);
846 		}
847 	}
848 	dn->nid = nids[level];
849 	dn->ofs_in_node = offset[level];
850 	dn->node_page = npage[level];
851 	dn->data_blkaddr = f2fs_data_blkaddr(dn);
852 
853 	if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) &&
854 					f2fs_sb_has_readonly(sbi)) {
855 		unsigned int cluster_size = F2FS_I(dn->inode)->i_cluster_size;
856 		unsigned int ofs_in_node = dn->ofs_in_node;
857 		pgoff_t fofs = index;
858 		unsigned int c_len;
859 		block_t blkaddr;
860 
861 		/* should align fofs and ofs_in_node to cluster_size */
862 		if (fofs % cluster_size) {
863 			fofs = round_down(fofs, cluster_size);
864 			ofs_in_node = round_down(ofs_in_node, cluster_size);
865 		}
866 
867 		c_len = f2fs_cluster_blocks_are_contiguous(dn, ofs_in_node);
868 		if (!c_len)
869 			goto out;
870 
871 		blkaddr = data_blkaddr(dn->inode, dn->node_page, ofs_in_node);
872 		if (blkaddr == COMPRESS_ADDR)
873 			blkaddr = data_blkaddr(dn->inode, dn->node_page,
874 						ofs_in_node + 1);
875 
876 		f2fs_update_read_extent_tree_range_compressed(dn->inode,
877 					fofs, blkaddr, cluster_size, c_len);
878 	}
879 out:
880 	return 0;
881 
882 release_pages:
883 	f2fs_put_page(parent, 1);
884 	if (i > 1)
885 		f2fs_put_page(npage[0], 0);
886 release_out:
887 	dn->inode_page = NULL;
888 	dn->node_page = NULL;
889 	if (err == -ENOENT) {
890 		dn->cur_level = i;
891 		dn->max_level = level;
892 		dn->ofs_in_node = offset[level];
893 	}
894 	return err;
895 }
896 
897 static int truncate_node(struct dnode_of_data *dn)
898 {
899 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
900 	struct node_info ni;
901 	int err;
902 	pgoff_t index;
903 
904 	err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
905 	if (err)
906 		return err;
907 
908 	if (ni.blk_addr != NEW_ADDR &&
909 		!f2fs_is_valid_blkaddr(sbi, ni.blk_addr, DATA_GENERIC_ENHANCE)) {
910 		f2fs_err_ratelimited(sbi,
911 			"nat entry is corrupted, run fsck to fix it, ino:%u, "
912 			"nid:%u, blkaddr:%u", ni.ino, ni.nid, ni.blk_addr);
913 		set_sbi_flag(sbi, SBI_NEED_FSCK);
914 		f2fs_handle_error(sbi, ERROR_INCONSISTENT_NAT);
915 		return -EFSCORRUPTED;
916 	}
917 
918 	/* Deallocate node address */
919 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
920 	dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
921 	set_node_addr(sbi, &ni, NULL_ADDR, false);
922 
923 	if (dn->nid == dn->inode->i_ino) {
924 		f2fs_remove_orphan_inode(sbi, dn->nid);
925 		dec_valid_inode_count(sbi);
926 		f2fs_inode_synced(dn->inode);
927 	}
928 
929 	clear_node_page_dirty(dn->node_page);
930 	set_sbi_flag(sbi, SBI_IS_DIRTY);
931 
932 	index = page_folio(dn->node_page)->index;
933 	f2fs_put_page(dn->node_page, 1);
934 
935 	invalidate_mapping_pages(NODE_MAPPING(sbi),
936 			index, index);
937 
938 	dn->node_page = NULL;
939 	trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
940 
941 	return 0;
942 }
943 
944 static int truncate_dnode(struct dnode_of_data *dn)
945 {
946 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
947 	struct page *page;
948 	int err;
949 
950 	if (dn->nid == 0)
951 		return 1;
952 
953 	/* get direct node */
954 	page = f2fs_get_node_page(sbi, dn->nid);
955 	if (PTR_ERR(page) == -ENOENT)
956 		return 1;
957 	else if (IS_ERR(page))
958 		return PTR_ERR(page);
959 
960 	if (IS_INODE(page) || ino_of_node(page) != dn->inode->i_ino) {
961 		f2fs_err(sbi, "incorrect node reference, ino: %lu, nid: %u, ino_of_node: %u",
962 				dn->inode->i_ino, dn->nid, ino_of_node(page));
963 		set_sbi_flag(sbi, SBI_NEED_FSCK);
964 		f2fs_handle_error(sbi, ERROR_INVALID_NODE_REFERENCE);
965 		f2fs_put_page(page, 1);
966 		return -EFSCORRUPTED;
967 	}
968 
969 	/* Make dnode_of_data for parameter */
970 	dn->node_page = page;
971 	dn->ofs_in_node = 0;
972 	f2fs_truncate_data_blocks_range(dn, ADDRS_PER_BLOCK(dn->inode));
973 	err = truncate_node(dn);
974 	if (err) {
975 		f2fs_put_page(page, 1);
976 		return err;
977 	}
978 
979 	return 1;
980 }
981 
982 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
983 						int ofs, int depth)
984 {
985 	struct dnode_of_data rdn = *dn;
986 	struct page *page;
987 	struct f2fs_node *rn;
988 	nid_t child_nid;
989 	unsigned int child_nofs;
990 	int freed = 0;
991 	int i, ret;
992 
993 	if (dn->nid == 0)
994 		return NIDS_PER_BLOCK + 1;
995 
996 	trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
997 
998 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
999 	if (IS_ERR(page)) {
1000 		trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
1001 		return PTR_ERR(page);
1002 	}
1003 
1004 	f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
1005 
1006 	rn = F2FS_NODE(page);
1007 	if (depth < 3) {
1008 		for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
1009 			child_nid = le32_to_cpu(rn->in.nid[i]);
1010 			if (child_nid == 0)
1011 				continue;
1012 			rdn.nid = child_nid;
1013 			ret = truncate_dnode(&rdn);
1014 			if (ret < 0)
1015 				goto out_err;
1016 			if (set_nid(page, i, 0, false))
1017 				dn->node_changed = true;
1018 		}
1019 	} else {
1020 		child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
1021 		for (i = ofs; i < NIDS_PER_BLOCK; i++) {
1022 			child_nid = le32_to_cpu(rn->in.nid[i]);
1023 			if (child_nid == 0) {
1024 				child_nofs += NIDS_PER_BLOCK + 1;
1025 				continue;
1026 			}
1027 			rdn.nid = child_nid;
1028 			ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
1029 			if (ret == (NIDS_PER_BLOCK + 1)) {
1030 				if (set_nid(page, i, 0, false))
1031 					dn->node_changed = true;
1032 				child_nofs += ret;
1033 			} else if (ret < 0 && ret != -ENOENT) {
1034 				goto out_err;
1035 			}
1036 		}
1037 		freed = child_nofs;
1038 	}
1039 
1040 	if (!ofs) {
1041 		/* remove current indirect node */
1042 		dn->node_page = page;
1043 		ret = truncate_node(dn);
1044 		if (ret)
1045 			goto out_err;
1046 		freed++;
1047 	} else {
1048 		f2fs_put_page(page, 1);
1049 	}
1050 	trace_f2fs_truncate_nodes_exit(dn->inode, freed);
1051 	return freed;
1052 
1053 out_err:
1054 	f2fs_put_page(page, 1);
1055 	trace_f2fs_truncate_nodes_exit(dn->inode, ret);
1056 	return ret;
1057 }
1058 
1059 static int truncate_partial_nodes(struct dnode_of_data *dn,
1060 			struct f2fs_inode *ri, int *offset, int depth)
1061 {
1062 	struct page *pages[2];
1063 	nid_t nid[3];
1064 	nid_t child_nid;
1065 	int err = 0;
1066 	int i;
1067 	int idx = depth - 2;
1068 
1069 	nid[0] = get_nid(dn->inode_page, offset[0], true);
1070 	if (!nid[0])
1071 		return 0;
1072 
1073 	/* get indirect nodes in the path */
1074 	for (i = 0; i < idx + 1; i++) {
1075 		/* reference count'll be increased */
1076 		pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
1077 		if (IS_ERR(pages[i])) {
1078 			err = PTR_ERR(pages[i]);
1079 			idx = i - 1;
1080 			goto fail;
1081 		}
1082 		nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1083 	}
1084 
1085 	f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1086 
1087 	/* free direct nodes linked to a partial indirect node */
1088 	for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1089 		child_nid = get_nid(pages[idx], i, false);
1090 		if (!child_nid)
1091 			continue;
1092 		dn->nid = child_nid;
1093 		err = truncate_dnode(dn);
1094 		if (err < 0)
1095 			goto fail;
1096 		if (set_nid(pages[idx], i, 0, false))
1097 			dn->node_changed = true;
1098 	}
1099 
1100 	if (offset[idx + 1] == 0) {
1101 		dn->node_page = pages[idx];
1102 		dn->nid = nid[idx];
1103 		err = truncate_node(dn);
1104 		if (err)
1105 			goto fail;
1106 	} else {
1107 		f2fs_put_page(pages[idx], 1);
1108 	}
1109 	offset[idx]++;
1110 	offset[idx + 1] = 0;
1111 	idx--;
1112 fail:
1113 	for (i = idx; i >= 0; i--)
1114 		f2fs_put_page(pages[i], 1);
1115 
1116 	trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1117 
1118 	return err;
1119 }
1120 
1121 /*
1122  * All the block addresses of data and nodes should be nullified.
1123  */
1124 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1125 {
1126 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1127 	int err = 0, cont = 1;
1128 	int level, offset[4], noffset[4];
1129 	unsigned int nofs = 0;
1130 	struct f2fs_inode *ri;
1131 	struct dnode_of_data dn;
1132 	struct page *page;
1133 
1134 	trace_f2fs_truncate_inode_blocks_enter(inode, from);
1135 
1136 	level = get_node_path(inode, from, offset, noffset);
1137 	if (level < 0) {
1138 		trace_f2fs_truncate_inode_blocks_exit(inode, level);
1139 		return level;
1140 	}
1141 
1142 	page = f2fs_get_node_page(sbi, inode->i_ino);
1143 	if (IS_ERR(page)) {
1144 		trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1145 		return PTR_ERR(page);
1146 	}
1147 
1148 	set_new_dnode(&dn, inode, page, NULL, 0);
1149 	unlock_page(page);
1150 
1151 	ri = F2FS_INODE(page);
1152 	switch (level) {
1153 	case 0:
1154 	case 1:
1155 		nofs = noffset[1];
1156 		break;
1157 	case 2:
1158 		nofs = noffset[1];
1159 		if (!offset[level - 1])
1160 			goto skip_partial;
1161 		err = truncate_partial_nodes(&dn, ri, offset, level);
1162 		if (err < 0 && err != -ENOENT)
1163 			goto fail;
1164 		nofs += 1 + NIDS_PER_BLOCK;
1165 		break;
1166 	case 3:
1167 		nofs = 5 + 2 * NIDS_PER_BLOCK;
1168 		if (!offset[level - 1])
1169 			goto skip_partial;
1170 		err = truncate_partial_nodes(&dn, ri, offset, level);
1171 		if (err < 0 && err != -ENOENT)
1172 			goto fail;
1173 		break;
1174 	default:
1175 		BUG();
1176 	}
1177 
1178 skip_partial:
1179 	while (cont) {
1180 		dn.nid = get_nid(page, offset[0], true);
1181 		switch (offset[0]) {
1182 		case NODE_DIR1_BLOCK:
1183 		case NODE_DIR2_BLOCK:
1184 			err = truncate_dnode(&dn);
1185 			break;
1186 
1187 		case NODE_IND1_BLOCK:
1188 		case NODE_IND2_BLOCK:
1189 			err = truncate_nodes(&dn, nofs, offset[1], 2);
1190 			break;
1191 
1192 		case NODE_DIND_BLOCK:
1193 			err = truncate_nodes(&dn, nofs, offset[1], 3);
1194 			cont = 0;
1195 			break;
1196 
1197 		default:
1198 			BUG();
1199 		}
1200 		if (err == -ENOENT) {
1201 			set_sbi_flag(F2FS_P_SB(page), SBI_NEED_FSCK);
1202 			f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1203 			f2fs_err_ratelimited(sbi,
1204 				"truncate node fail, ino:%lu, nid:%u, "
1205 				"offset[0]:%d, offset[1]:%d, nofs:%d",
1206 				inode->i_ino, dn.nid, offset[0],
1207 				offset[1], nofs);
1208 			err = 0;
1209 		}
1210 		if (err < 0)
1211 			goto fail;
1212 		if (offset[1] == 0 && get_nid(page, offset[0], true)) {
1213 			lock_page(page);
1214 			BUG_ON(page->mapping != NODE_MAPPING(sbi));
1215 			set_nid(page, offset[0], 0, true);
1216 			unlock_page(page);
1217 		}
1218 		offset[1] = 0;
1219 		offset[0]++;
1220 		nofs += err;
1221 	}
1222 fail:
1223 	f2fs_put_page(page, 0);
1224 	trace_f2fs_truncate_inode_blocks_exit(inode, err);
1225 	return err > 0 ? 0 : err;
1226 }
1227 
1228 /* caller must lock inode page */
1229 int f2fs_truncate_xattr_node(struct inode *inode)
1230 {
1231 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1232 	nid_t nid = F2FS_I(inode)->i_xattr_nid;
1233 	struct dnode_of_data dn;
1234 	struct page *npage;
1235 	int err;
1236 
1237 	if (!nid)
1238 		return 0;
1239 
1240 	npage = f2fs_get_node_page(sbi, nid);
1241 	if (IS_ERR(npage))
1242 		return PTR_ERR(npage);
1243 
1244 	set_new_dnode(&dn, inode, NULL, npage, nid);
1245 	err = truncate_node(&dn);
1246 	if (err) {
1247 		f2fs_put_page(npage, 1);
1248 		return err;
1249 	}
1250 
1251 	f2fs_i_xnid_write(inode, 0);
1252 
1253 	return 0;
1254 }
1255 
1256 /*
1257  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1258  * f2fs_unlock_op().
1259  */
1260 int f2fs_remove_inode_page(struct inode *inode)
1261 {
1262 	struct dnode_of_data dn;
1263 	int err;
1264 
1265 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1266 	err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1267 	if (err)
1268 		return err;
1269 
1270 	err = f2fs_truncate_xattr_node(inode);
1271 	if (err) {
1272 		f2fs_put_dnode(&dn);
1273 		return err;
1274 	}
1275 
1276 	/* remove potential inline_data blocks */
1277 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1278 				S_ISLNK(inode->i_mode))
1279 		f2fs_truncate_data_blocks_range(&dn, 1);
1280 
1281 	/* 0 is possible, after f2fs_new_inode() has failed */
1282 	if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1283 		f2fs_put_dnode(&dn);
1284 		return -EIO;
1285 	}
1286 
1287 	if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1288 		f2fs_warn(F2FS_I_SB(inode),
1289 			"f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1290 			inode->i_ino, (unsigned long long)inode->i_blocks);
1291 		set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1292 	}
1293 
1294 	/* will put inode & node pages */
1295 	err = truncate_node(&dn);
1296 	if (err) {
1297 		f2fs_put_dnode(&dn);
1298 		return err;
1299 	}
1300 	return 0;
1301 }
1302 
1303 struct page *f2fs_new_inode_page(struct inode *inode)
1304 {
1305 	struct dnode_of_data dn;
1306 
1307 	/* allocate inode page for new inode */
1308 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1309 
1310 	/* caller should f2fs_put_page(page, 1); */
1311 	return f2fs_new_node_page(&dn, 0);
1312 }
1313 
1314 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1315 {
1316 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1317 	struct node_info new_ni;
1318 	struct page *page;
1319 	int err;
1320 
1321 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1322 		return ERR_PTR(-EPERM);
1323 
1324 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1325 	if (!page)
1326 		return ERR_PTR(-ENOMEM);
1327 
1328 	if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1329 		goto fail;
1330 
1331 #ifdef CONFIG_F2FS_CHECK_FS
1332 	err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false);
1333 	if (err) {
1334 		dec_valid_node_count(sbi, dn->inode, !ofs);
1335 		goto fail;
1336 	}
1337 	if (unlikely(new_ni.blk_addr != NULL_ADDR)) {
1338 		err = -EFSCORRUPTED;
1339 		dec_valid_node_count(sbi, dn->inode, !ofs);
1340 		set_sbi_flag(sbi, SBI_NEED_FSCK);
1341 		f2fs_warn_ratelimited(sbi,
1342 			"f2fs_new_node_page: inconsistent nat entry, "
1343 			"ino:%u, nid:%u, blkaddr:%u, ver:%u, flag:%u",
1344 			new_ni.ino, new_ni.nid, new_ni.blk_addr,
1345 			new_ni.version, new_ni.flag);
1346 		f2fs_handle_error(sbi, ERROR_INCONSISTENT_NAT);
1347 		goto fail;
1348 	}
1349 #endif
1350 	new_ni.nid = dn->nid;
1351 	new_ni.ino = dn->inode->i_ino;
1352 	new_ni.blk_addr = NULL_ADDR;
1353 	new_ni.flag = 0;
1354 	new_ni.version = 0;
1355 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1356 
1357 	f2fs_wait_on_page_writeback(page, NODE, true, true);
1358 	fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1359 	set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1360 	if (!PageUptodate(page))
1361 		SetPageUptodate(page);
1362 	if (set_page_dirty(page))
1363 		dn->node_changed = true;
1364 
1365 	if (f2fs_has_xattr_block(ofs))
1366 		f2fs_i_xnid_write(dn->inode, dn->nid);
1367 
1368 	if (ofs == 0)
1369 		inc_valid_inode_count(sbi);
1370 	return page;
1371 fail:
1372 	clear_node_page_dirty(page);
1373 	f2fs_put_page(page, 1);
1374 	return ERR_PTR(err);
1375 }
1376 
1377 /*
1378  * Caller should do after getting the following values.
1379  * 0: f2fs_put_page(page, 0)
1380  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1381  */
1382 static int read_node_page(struct page *page, blk_opf_t op_flags)
1383 {
1384 	struct folio *folio = page_folio(page);
1385 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1386 	struct node_info ni;
1387 	struct f2fs_io_info fio = {
1388 		.sbi = sbi,
1389 		.type = NODE,
1390 		.op = REQ_OP_READ,
1391 		.op_flags = op_flags,
1392 		.page = page,
1393 		.encrypted_page = NULL,
1394 	};
1395 	int err;
1396 
1397 	if (folio_test_uptodate(folio)) {
1398 		if (!f2fs_inode_chksum_verify(sbi, page)) {
1399 			folio_clear_uptodate(folio);
1400 			return -EFSBADCRC;
1401 		}
1402 		return LOCKED_PAGE;
1403 	}
1404 
1405 	err = f2fs_get_node_info(sbi, folio->index, &ni, false);
1406 	if (err)
1407 		return err;
1408 
1409 	/* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1410 	if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR)) {
1411 		folio_clear_uptodate(folio);
1412 		return -ENOENT;
1413 	}
1414 
1415 	fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1416 
1417 	err = f2fs_submit_page_bio(&fio);
1418 
1419 	if (!err)
1420 		f2fs_update_iostat(sbi, NULL, FS_NODE_READ_IO, F2FS_BLKSIZE);
1421 
1422 	return err;
1423 }
1424 
1425 /*
1426  * Readahead a node page
1427  */
1428 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1429 {
1430 	struct page *apage;
1431 	int err;
1432 
1433 	if (!nid)
1434 		return;
1435 	if (f2fs_check_nid_range(sbi, nid))
1436 		return;
1437 
1438 	apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1439 	if (apage)
1440 		return;
1441 
1442 	apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1443 	if (!apage)
1444 		return;
1445 
1446 	err = read_node_page(apage, REQ_RAHEAD);
1447 	f2fs_put_page(apage, err ? 1 : 0);
1448 }
1449 
1450 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1451 					struct page *parent, int start)
1452 {
1453 	struct page *page;
1454 	int err;
1455 
1456 	if (!nid)
1457 		return ERR_PTR(-ENOENT);
1458 	if (f2fs_check_nid_range(sbi, nid))
1459 		return ERR_PTR(-EINVAL);
1460 repeat:
1461 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1462 	if (!page)
1463 		return ERR_PTR(-ENOMEM);
1464 
1465 	err = read_node_page(page, 0);
1466 	if (err < 0) {
1467 		goto out_put_err;
1468 	} else if (err == LOCKED_PAGE) {
1469 		err = 0;
1470 		goto page_hit;
1471 	}
1472 
1473 	if (parent)
1474 		f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1475 
1476 	lock_page(page);
1477 
1478 	if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1479 		f2fs_put_page(page, 1);
1480 		goto repeat;
1481 	}
1482 
1483 	if (unlikely(!PageUptodate(page))) {
1484 		err = -EIO;
1485 		goto out_err;
1486 	}
1487 
1488 	if (!f2fs_inode_chksum_verify(sbi, page)) {
1489 		err = -EFSBADCRC;
1490 		goto out_err;
1491 	}
1492 page_hit:
1493 	if (likely(nid == nid_of_node(page)))
1494 		return page;
1495 
1496 	f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1497 			  nid, nid_of_node(page), ino_of_node(page),
1498 			  ofs_of_node(page), cpver_of_node(page),
1499 			  next_blkaddr_of_node(page));
1500 	set_sbi_flag(sbi, SBI_NEED_FSCK);
1501 	f2fs_handle_error(sbi, ERROR_INCONSISTENT_FOOTER);
1502 	err = -EFSCORRUPTED;
1503 out_err:
1504 	ClearPageUptodate(page);
1505 out_put_err:
1506 	/* ENOENT comes from read_node_page which is not an error. */
1507 	if (err != -ENOENT)
1508 		f2fs_handle_page_eio(sbi, page_folio(page), NODE);
1509 	f2fs_put_page(page, 1);
1510 	return ERR_PTR(err);
1511 }
1512 
1513 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1514 {
1515 	return __get_node_page(sbi, nid, NULL, 0);
1516 }
1517 
1518 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1519 {
1520 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1521 	nid_t nid = get_nid(parent, start, false);
1522 
1523 	return __get_node_page(sbi, nid, parent, start);
1524 }
1525 
1526 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1527 {
1528 	struct inode *inode;
1529 	struct page *page;
1530 	int ret;
1531 
1532 	/* should flush inline_data before evict_inode */
1533 	inode = ilookup(sbi->sb, ino);
1534 	if (!inode)
1535 		return;
1536 
1537 	page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1538 					FGP_LOCK|FGP_NOWAIT, 0);
1539 	if (!page)
1540 		goto iput_out;
1541 
1542 	if (!PageUptodate(page))
1543 		goto page_out;
1544 
1545 	if (!PageDirty(page))
1546 		goto page_out;
1547 
1548 	if (!clear_page_dirty_for_io(page))
1549 		goto page_out;
1550 
1551 	ret = f2fs_write_inline_data(inode, page_folio(page));
1552 	inode_dec_dirty_pages(inode);
1553 	f2fs_remove_dirty_inode(inode);
1554 	if (ret)
1555 		set_page_dirty(page);
1556 page_out:
1557 	f2fs_put_page(page, 1);
1558 iput_out:
1559 	iput(inode);
1560 }
1561 
1562 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1563 {
1564 	pgoff_t index;
1565 	struct folio_batch fbatch;
1566 	struct page *last_page = NULL;
1567 	int nr_folios;
1568 
1569 	folio_batch_init(&fbatch);
1570 	index = 0;
1571 
1572 	while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1573 					(pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1574 					&fbatch))) {
1575 		int i;
1576 
1577 		for (i = 0; i < nr_folios; i++) {
1578 			struct page *page = &fbatch.folios[i]->page;
1579 
1580 			if (unlikely(f2fs_cp_error(sbi))) {
1581 				f2fs_put_page(last_page, 0);
1582 				folio_batch_release(&fbatch);
1583 				return ERR_PTR(-EIO);
1584 			}
1585 
1586 			if (!IS_DNODE(page) || !is_cold_node(page))
1587 				continue;
1588 			if (ino_of_node(page) != ino)
1589 				continue;
1590 
1591 			lock_page(page);
1592 
1593 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1594 continue_unlock:
1595 				unlock_page(page);
1596 				continue;
1597 			}
1598 			if (ino_of_node(page) != ino)
1599 				goto continue_unlock;
1600 
1601 			if (!PageDirty(page)) {
1602 				/* someone wrote it for us */
1603 				goto continue_unlock;
1604 			}
1605 
1606 			if (last_page)
1607 				f2fs_put_page(last_page, 0);
1608 
1609 			get_page(page);
1610 			last_page = page;
1611 			unlock_page(page);
1612 		}
1613 		folio_batch_release(&fbatch);
1614 		cond_resched();
1615 	}
1616 	return last_page;
1617 }
1618 
1619 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1620 				struct writeback_control *wbc, bool do_balance,
1621 				enum iostat_type io_type, unsigned int *seq_id)
1622 {
1623 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1624 	struct folio *folio = page_folio(page);
1625 	nid_t nid;
1626 	struct node_info ni;
1627 	struct f2fs_io_info fio = {
1628 		.sbi = sbi,
1629 		.ino = ino_of_node(page),
1630 		.type = NODE,
1631 		.op = REQ_OP_WRITE,
1632 		.op_flags = wbc_to_write_flags(wbc),
1633 		.page = page,
1634 		.encrypted_page = NULL,
1635 		.submitted = 0,
1636 		.io_type = io_type,
1637 		.io_wbc = wbc,
1638 	};
1639 	unsigned int seq;
1640 
1641 	trace_f2fs_writepage(folio, NODE);
1642 
1643 	if (unlikely(f2fs_cp_error(sbi))) {
1644 		/* keep node pages in remount-ro mode */
1645 		if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY)
1646 			goto redirty_out;
1647 		folio_clear_uptodate(folio);
1648 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1649 		folio_unlock(folio);
1650 		return 0;
1651 	}
1652 
1653 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1654 		goto redirty_out;
1655 
1656 	if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1657 			wbc->sync_mode == WB_SYNC_NONE &&
1658 			IS_DNODE(page) && is_cold_node(page))
1659 		goto redirty_out;
1660 
1661 	/* get old block addr of this node page */
1662 	nid = nid_of_node(page);
1663 	f2fs_bug_on(sbi, folio->index != nid);
1664 
1665 	if (f2fs_get_node_info(sbi, nid, &ni, !do_balance))
1666 		goto redirty_out;
1667 
1668 	if (wbc->for_reclaim) {
1669 		if (!f2fs_down_read_trylock(&sbi->node_write))
1670 			goto redirty_out;
1671 	} else {
1672 		f2fs_down_read(&sbi->node_write);
1673 	}
1674 
1675 	/* This page is already truncated */
1676 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1677 		folio_clear_uptodate(folio);
1678 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1679 		f2fs_up_read(&sbi->node_write);
1680 		folio_unlock(folio);
1681 		return 0;
1682 	}
1683 
1684 	if (__is_valid_data_blkaddr(ni.blk_addr) &&
1685 		!f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1686 					DATA_GENERIC_ENHANCE)) {
1687 		f2fs_up_read(&sbi->node_write);
1688 		goto redirty_out;
1689 	}
1690 
1691 	if (atomic && !test_opt(sbi, NOBARRIER))
1692 		fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1693 
1694 	/* should add to global list before clearing PAGECACHE status */
1695 	if (f2fs_in_warm_node_list(sbi, page)) {
1696 		seq = f2fs_add_fsync_node_entry(sbi, page);
1697 		if (seq_id)
1698 			*seq_id = seq;
1699 	}
1700 
1701 	folio_start_writeback(folio);
1702 
1703 	fio.old_blkaddr = ni.blk_addr;
1704 	f2fs_do_write_node_page(nid, &fio);
1705 	set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1706 	dec_page_count(sbi, F2FS_DIRTY_NODES);
1707 	f2fs_up_read(&sbi->node_write);
1708 
1709 	if (wbc->for_reclaim) {
1710 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1711 		submitted = NULL;
1712 	}
1713 
1714 	folio_unlock(folio);
1715 
1716 	if (unlikely(f2fs_cp_error(sbi))) {
1717 		f2fs_submit_merged_write(sbi, NODE);
1718 		submitted = NULL;
1719 	}
1720 	if (submitted)
1721 		*submitted = fio.submitted;
1722 
1723 	if (do_balance)
1724 		f2fs_balance_fs(sbi, false);
1725 	return 0;
1726 
1727 redirty_out:
1728 	folio_redirty_for_writepage(wbc, folio);
1729 	return AOP_WRITEPAGE_ACTIVATE;
1730 }
1731 
1732 int f2fs_move_node_page(struct page *node_page, int gc_type)
1733 {
1734 	int err = 0;
1735 
1736 	if (gc_type == FG_GC) {
1737 		struct writeback_control wbc = {
1738 			.sync_mode = WB_SYNC_ALL,
1739 			.nr_to_write = 1,
1740 			.for_reclaim = 0,
1741 		};
1742 
1743 		f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1744 
1745 		set_page_dirty(node_page);
1746 
1747 		if (!clear_page_dirty_for_io(node_page)) {
1748 			err = -EAGAIN;
1749 			goto out_page;
1750 		}
1751 
1752 		if (__write_node_page(node_page, false, NULL,
1753 					&wbc, false, FS_GC_NODE_IO, NULL)) {
1754 			err = -EAGAIN;
1755 			unlock_page(node_page);
1756 		}
1757 		goto release_page;
1758 	} else {
1759 		/* set page dirty and write it */
1760 		if (!folio_test_writeback(page_folio(node_page)))
1761 			set_page_dirty(node_page);
1762 	}
1763 out_page:
1764 	unlock_page(node_page);
1765 release_page:
1766 	f2fs_put_page(node_page, 0);
1767 	return err;
1768 }
1769 
1770 static int f2fs_write_node_page(struct page *page,
1771 				struct writeback_control *wbc)
1772 {
1773 	return __write_node_page(page, false, NULL, wbc, false,
1774 						FS_NODE_IO, NULL);
1775 }
1776 
1777 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1778 			struct writeback_control *wbc, bool atomic,
1779 			unsigned int *seq_id)
1780 {
1781 	pgoff_t index;
1782 	struct folio_batch fbatch;
1783 	int ret = 0;
1784 	struct page *last_page = NULL;
1785 	bool marked = false;
1786 	nid_t ino = inode->i_ino;
1787 	int nr_folios;
1788 	int nwritten = 0;
1789 
1790 	if (atomic) {
1791 		last_page = last_fsync_dnode(sbi, ino);
1792 		if (IS_ERR_OR_NULL(last_page))
1793 			return PTR_ERR_OR_ZERO(last_page);
1794 	}
1795 retry:
1796 	folio_batch_init(&fbatch);
1797 	index = 0;
1798 
1799 	while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1800 					(pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1801 					&fbatch))) {
1802 		int i;
1803 
1804 		for (i = 0; i < nr_folios; i++) {
1805 			struct page *page = &fbatch.folios[i]->page;
1806 			bool submitted = false;
1807 
1808 			if (unlikely(f2fs_cp_error(sbi))) {
1809 				f2fs_put_page(last_page, 0);
1810 				folio_batch_release(&fbatch);
1811 				ret = -EIO;
1812 				goto out;
1813 			}
1814 
1815 			if (!IS_DNODE(page) || !is_cold_node(page))
1816 				continue;
1817 			if (ino_of_node(page) != ino)
1818 				continue;
1819 
1820 			lock_page(page);
1821 
1822 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1823 continue_unlock:
1824 				unlock_page(page);
1825 				continue;
1826 			}
1827 			if (ino_of_node(page) != ino)
1828 				goto continue_unlock;
1829 
1830 			if (!PageDirty(page) && page != last_page) {
1831 				/* someone wrote it for us */
1832 				goto continue_unlock;
1833 			}
1834 
1835 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1836 
1837 			set_fsync_mark(page, 0);
1838 			set_dentry_mark(page, 0);
1839 
1840 			if (!atomic || page == last_page) {
1841 				set_fsync_mark(page, 1);
1842 				percpu_counter_inc(&sbi->rf_node_block_count);
1843 				if (IS_INODE(page)) {
1844 					if (is_inode_flag_set(inode,
1845 								FI_DIRTY_INODE))
1846 						f2fs_update_inode(inode, page);
1847 					set_dentry_mark(page,
1848 						f2fs_need_dentry_mark(sbi, ino));
1849 				}
1850 				/* may be written by other thread */
1851 				if (!PageDirty(page))
1852 					set_page_dirty(page);
1853 			}
1854 
1855 			if (!clear_page_dirty_for_io(page))
1856 				goto continue_unlock;
1857 
1858 			ret = __write_node_page(page, atomic &&
1859 						page == last_page,
1860 						&submitted, wbc, true,
1861 						FS_NODE_IO, seq_id);
1862 			if (ret) {
1863 				unlock_page(page);
1864 				f2fs_put_page(last_page, 0);
1865 				break;
1866 			} else if (submitted) {
1867 				nwritten++;
1868 			}
1869 
1870 			if (page == last_page) {
1871 				f2fs_put_page(page, 0);
1872 				marked = true;
1873 				break;
1874 			}
1875 		}
1876 		folio_batch_release(&fbatch);
1877 		cond_resched();
1878 
1879 		if (ret || marked)
1880 			break;
1881 	}
1882 	if (!ret && atomic && !marked) {
1883 		f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1884 			   ino, page_folio(last_page)->index);
1885 		lock_page(last_page);
1886 		f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1887 		set_page_dirty(last_page);
1888 		unlock_page(last_page);
1889 		goto retry;
1890 	}
1891 out:
1892 	if (nwritten)
1893 		f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1894 	return ret ? -EIO : 0;
1895 }
1896 
1897 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1898 {
1899 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1900 	bool clean;
1901 
1902 	if (inode->i_ino != ino)
1903 		return 0;
1904 
1905 	if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1906 		return 0;
1907 
1908 	spin_lock(&sbi->inode_lock[DIRTY_META]);
1909 	clean = list_empty(&F2FS_I(inode)->gdirty_list);
1910 	spin_unlock(&sbi->inode_lock[DIRTY_META]);
1911 
1912 	if (clean)
1913 		return 0;
1914 
1915 	inode = igrab(inode);
1916 	if (!inode)
1917 		return 0;
1918 	return 1;
1919 }
1920 
1921 static bool flush_dirty_inode(struct page *page)
1922 {
1923 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1924 	struct inode *inode;
1925 	nid_t ino = ino_of_node(page);
1926 
1927 	inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1928 	if (!inode)
1929 		return false;
1930 
1931 	f2fs_update_inode(inode, page);
1932 	unlock_page(page);
1933 
1934 	iput(inode);
1935 	return true;
1936 }
1937 
1938 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1939 {
1940 	pgoff_t index = 0;
1941 	struct folio_batch fbatch;
1942 	int nr_folios;
1943 
1944 	folio_batch_init(&fbatch);
1945 
1946 	while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1947 					(pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1948 					&fbatch))) {
1949 		int i;
1950 
1951 		for (i = 0; i < nr_folios; i++) {
1952 			struct page *page = &fbatch.folios[i]->page;
1953 
1954 			if (!IS_INODE(page))
1955 				continue;
1956 
1957 			lock_page(page);
1958 
1959 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1960 continue_unlock:
1961 				unlock_page(page);
1962 				continue;
1963 			}
1964 
1965 			if (!PageDirty(page)) {
1966 				/* someone wrote it for us */
1967 				goto continue_unlock;
1968 			}
1969 
1970 			/* flush inline_data, if it's async context. */
1971 			if (page_private_inline(page)) {
1972 				clear_page_private_inline(page);
1973 				unlock_page(page);
1974 				flush_inline_data(sbi, ino_of_node(page));
1975 				continue;
1976 			}
1977 			unlock_page(page);
1978 		}
1979 		folio_batch_release(&fbatch);
1980 		cond_resched();
1981 	}
1982 }
1983 
1984 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1985 				struct writeback_control *wbc,
1986 				bool do_balance, enum iostat_type io_type)
1987 {
1988 	pgoff_t index;
1989 	struct folio_batch fbatch;
1990 	int step = 0;
1991 	int nwritten = 0;
1992 	int ret = 0;
1993 	int nr_folios, done = 0;
1994 
1995 	folio_batch_init(&fbatch);
1996 
1997 next_step:
1998 	index = 0;
1999 
2000 	while (!done && (nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi),
2001 				&index, (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
2002 				&fbatch))) {
2003 		int i;
2004 
2005 		for (i = 0; i < nr_folios; i++) {
2006 			struct page *page = &fbatch.folios[i]->page;
2007 			bool submitted = false;
2008 
2009 			/* give a priority to WB_SYNC threads */
2010 			if (atomic_read(&sbi->wb_sync_req[NODE]) &&
2011 					wbc->sync_mode == WB_SYNC_NONE) {
2012 				done = 1;
2013 				break;
2014 			}
2015 
2016 			/*
2017 			 * flushing sequence with step:
2018 			 * 0. indirect nodes
2019 			 * 1. dentry dnodes
2020 			 * 2. file dnodes
2021 			 */
2022 			if (step == 0 && IS_DNODE(page))
2023 				continue;
2024 			if (step == 1 && (!IS_DNODE(page) ||
2025 						is_cold_node(page)))
2026 				continue;
2027 			if (step == 2 && (!IS_DNODE(page) ||
2028 						!is_cold_node(page)))
2029 				continue;
2030 lock_node:
2031 			if (wbc->sync_mode == WB_SYNC_ALL)
2032 				lock_page(page);
2033 			else if (!trylock_page(page))
2034 				continue;
2035 
2036 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
2037 continue_unlock:
2038 				unlock_page(page);
2039 				continue;
2040 			}
2041 
2042 			if (!PageDirty(page)) {
2043 				/* someone wrote it for us */
2044 				goto continue_unlock;
2045 			}
2046 
2047 			/* flush inline_data/inode, if it's async context. */
2048 			if (!do_balance)
2049 				goto write_node;
2050 
2051 			/* flush inline_data */
2052 			if (page_private_inline(page)) {
2053 				clear_page_private_inline(page);
2054 				unlock_page(page);
2055 				flush_inline_data(sbi, ino_of_node(page));
2056 				goto lock_node;
2057 			}
2058 
2059 			/* flush dirty inode */
2060 			if (IS_INODE(page) && flush_dirty_inode(page))
2061 				goto lock_node;
2062 write_node:
2063 			f2fs_wait_on_page_writeback(page, NODE, true, true);
2064 
2065 			if (!clear_page_dirty_for_io(page))
2066 				goto continue_unlock;
2067 
2068 			set_fsync_mark(page, 0);
2069 			set_dentry_mark(page, 0);
2070 
2071 			ret = __write_node_page(page, false, &submitted,
2072 						wbc, do_balance, io_type, NULL);
2073 			if (ret)
2074 				unlock_page(page);
2075 			else if (submitted)
2076 				nwritten++;
2077 
2078 			if (--wbc->nr_to_write == 0)
2079 				break;
2080 		}
2081 		folio_batch_release(&fbatch);
2082 		cond_resched();
2083 
2084 		if (wbc->nr_to_write == 0) {
2085 			step = 2;
2086 			break;
2087 		}
2088 	}
2089 
2090 	if (step < 2) {
2091 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2092 				wbc->sync_mode == WB_SYNC_NONE && step == 1)
2093 			goto out;
2094 		step++;
2095 		goto next_step;
2096 	}
2097 out:
2098 	if (nwritten)
2099 		f2fs_submit_merged_write(sbi, NODE);
2100 
2101 	if (unlikely(f2fs_cp_error(sbi)))
2102 		return -EIO;
2103 	return ret;
2104 }
2105 
2106 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2107 						unsigned int seq_id)
2108 {
2109 	struct fsync_node_entry *fn;
2110 	struct page *page;
2111 	struct list_head *head = &sbi->fsync_node_list;
2112 	unsigned long flags;
2113 	unsigned int cur_seq_id = 0;
2114 
2115 	while (seq_id && cur_seq_id < seq_id) {
2116 		spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2117 		if (list_empty(head)) {
2118 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2119 			break;
2120 		}
2121 		fn = list_first_entry(head, struct fsync_node_entry, list);
2122 		if (fn->seq_id > seq_id) {
2123 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2124 			break;
2125 		}
2126 		cur_seq_id = fn->seq_id;
2127 		page = fn->page;
2128 		get_page(page);
2129 		spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2130 
2131 		f2fs_wait_on_page_writeback(page, NODE, true, false);
2132 
2133 		put_page(page);
2134 	}
2135 
2136 	return filemap_check_errors(NODE_MAPPING(sbi));
2137 }
2138 
2139 static int f2fs_write_node_pages(struct address_space *mapping,
2140 			    struct writeback_control *wbc)
2141 {
2142 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2143 	struct blk_plug plug;
2144 	long diff;
2145 
2146 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2147 		goto skip_write;
2148 
2149 	/* balancing f2fs's metadata in background */
2150 	f2fs_balance_fs_bg(sbi, true);
2151 
2152 	/* collect a number of dirty node pages and write together */
2153 	if (wbc->sync_mode != WB_SYNC_ALL &&
2154 			get_pages(sbi, F2FS_DIRTY_NODES) <
2155 					nr_pages_to_skip(sbi, NODE))
2156 		goto skip_write;
2157 
2158 	if (wbc->sync_mode == WB_SYNC_ALL)
2159 		atomic_inc(&sbi->wb_sync_req[NODE]);
2160 	else if (atomic_read(&sbi->wb_sync_req[NODE])) {
2161 		/* to avoid potential deadlock */
2162 		if (current->plug)
2163 			blk_finish_plug(current->plug);
2164 		goto skip_write;
2165 	}
2166 
2167 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2168 
2169 	diff = nr_pages_to_write(sbi, NODE, wbc);
2170 	blk_start_plug(&plug);
2171 	f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2172 	blk_finish_plug(&plug);
2173 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2174 
2175 	if (wbc->sync_mode == WB_SYNC_ALL)
2176 		atomic_dec(&sbi->wb_sync_req[NODE]);
2177 	return 0;
2178 
2179 skip_write:
2180 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2181 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2182 	return 0;
2183 }
2184 
2185 static bool f2fs_dirty_node_folio(struct address_space *mapping,
2186 		struct folio *folio)
2187 {
2188 	trace_f2fs_set_page_dirty(folio, NODE);
2189 
2190 	if (!folio_test_uptodate(folio))
2191 		folio_mark_uptodate(folio);
2192 #ifdef CONFIG_F2FS_CHECK_FS
2193 	if (IS_INODE(&folio->page))
2194 		f2fs_inode_chksum_set(F2FS_M_SB(mapping), &folio->page);
2195 #endif
2196 	if (filemap_dirty_folio(mapping, folio)) {
2197 		inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
2198 		set_page_private_reference(&folio->page);
2199 		return true;
2200 	}
2201 	return false;
2202 }
2203 
2204 /*
2205  * Structure of the f2fs node operations
2206  */
2207 const struct address_space_operations f2fs_node_aops = {
2208 	.writepage	= f2fs_write_node_page,
2209 	.writepages	= f2fs_write_node_pages,
2210 	.dirty_folio	= f2fs_dirty_node_folio,
2211 	.invalidate_folio = f2fs_invalidate_folio,
2212 	.release_folio	= f2fs_release_folio,
2213 	.migrate_folio	= filemap_migrate_folio,
2214 };
2215 
2216 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2217 						nid_t n)
2218 {
2219 	return radix_tree_lookup(&nm_i->free_nid_root, n);
2220 }
2221 
2222 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2223 				struct free_nid *i)
2224 {
2225 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2226 	int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2227 
2228 	if (err)
2229 		return err;
2230 
2231 	nm_i->nid_cnt[FREE_NID]++;
2232 	list_add_tail(&i->list, &nm_i->free_nid_list);
2233 	return 0;
2234 }
2235 
2236 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2237 			struct free_nid *i, enum nid_state state)
2238 {
2239 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2240 
2241 	f2fs_bug_on(sbi, state != i->state);
2242 	nm_i->nid_cnt[state]--;
2243 	if (state == FREE_NID)
2244 		list_del(&i->list);
2245 	radix_tree_delete(&nm_i->free_nid_root, i->nid);
2246 }
2247 
2248 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2249 			enum nid_state org_state, enum nid_state dst_state)
2250 {
2251 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2252 
2253 	f2fs_bug_on(sbi, org_state != i->state);
2254 	i->state = dst_state;
2255 	nm_i->nid_cnt[org_state]--;
2256 	nm_i->nid_cnt[dst_state]++;
2257 
2258 	switch (dst_state) {
2259 	case PREALLOC_NID:
2260 		list_del(&i->list);
2261 		break;
2262 	case FREE_NID:
2263 		list_add_tail(&i->list, &nm_i->free_nid_list);
2264 		break;
2265 	default:
2266 		BUG_ON(1);
2267 	}
2268 }
2269 
2270 bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi)
2271 {
2272 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2273 	unsigned int i;
2274 	bool ret = true;
2275 
2276 	f2fs_down_read(&nm_i->nat_tree_lock);
2277 	for (i = 0; i < nm_i->nat_blocks; i++) {
2278 		if (!test_bit_le(i, nm_i->nat_block_bitmap)) {
2279 			ret = false;
2280 			break;
2281 		}
2282 	}
2283 	f2fs_up_read(&nm_i->nat_tree_lock);
2284 
2285 	return ret;
2286 }
2287 
2288 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2289 							bool set, bool build)
2290 {
2291 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2292 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2293 	unsigned int nid_ofs = nid - START_NID(nid);
2294 
2295 	if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2296 		return;
2297 
2298 	if (set) {
2299 		if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2300 			return;
2301 		__set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2302 		nm_i->free_nid_count[nat_ofs]++;
2303 	} else {
2304 		if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2305 			return;
2306 		__clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2307 		if (!build)
2308 			nm_i->free_nid_count[nat_ofs]--;
2309 	}
2310 }
2311 
2312 /* return if the nid is recognized as free */
2313 static bool add_free_nid(struct f2fs_sb_info *sbi,
2314 				nid_t nid, bool build, bool update)
2315 {
2316 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2317 	struct free_nid *i, *e;
2318 	struct nat_entry *ne;
2319 	int err = -EINVAL;
2320 	bool ret = false;
2321 
2322 	/* 0 nid should not be used */
2323 	if (unlikely(nid == 0))
2324 		return false;
2325 
2326 	if (unlikely(f2fs_check_nid_range(sbi, nid)))
2327 		return false;
2328 
2329 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL);
2330 	i->nid = nid;
2331 	i->state = FREE_NID;
2332 
2333 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2334 
2335 	spin_lock(&nm_i->nid_list_lock);
2336 
2337 	if (build) {
2338 		/*
2339 		 *   Thread A             Thread B
2340 		 *  - f2fs_create
2341 		 *   - f2fs_new_inode
2342 		 *    - f2fs_alloc_nid
2343 		 *     - __insert_nid_to_list(PREALLOC_NID)
2344 		 *                     - f2fs_balance_fs_bg
2345 		 *                      - f2fs_build_free_nids
2346 		 *                       - __f2fs_build_free_nids
2347 		 *                        - scan_nat_page
2348 		 *                         - add_free_nid
2349 		 *                          - __lookup_nat_cache
2350 		 *  - f2fs_add_link
2351 		 *   - f2fs_init_inode_metadata
2352 		 *    - f2fs_new_inode_page
2353 		 *     - f2fs_new_node_page
2354 		 *      - set_node_addr
2355 		 *  - f2fs_alloc_nid_done
2356 		 *   - __remove_nid_from_list(PREALLOC_NID)
2357 		 *                         - __insert_nid_to_list(FREE_NID)
2358 		 */
2359 		ne = __lookup_nat_cache(nm_i, nid);
2360 		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2361 				nat_get_blkaddr(ne) != NULL_ADDR))
2362 			goto err_out;
2363 
2364 		e = __lookup_free_nid_list(nm_i, nid);
2365 		if (e) {
2366 			if (e->state == FREE_NID)
2367 				ret = true;
2368 			goto err_out;
2369 		}
2370 	}
2371 	ret = true;
2372 	err = __insert_free_nid(sbi, i);
2373 err_out:
2374 	if (update) {
2375 		update_free_nid_bitmap(sbi, nid, ret, build);
2376 		if (!build)
2377 			nm_i->available_nids++;
2378 	}
2379 	spin_unlock(&nm_i->nid_list_lock);
2380 	radix_tree_preload_end();
2381 
2382 	if (err)
2383 		kmem_cache_free(free_nid_slab, i);
2384 	return ret;
2385 }
2386 
2387 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2388 {
2389 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2390 	struct free_nid *i;
2391 	bool need_free = false;
2392 
2393 	spin_lock(&nm_i->nid_list_lock);
2394 	i = __lookup_free_nid_list(nm_i, nid);
2395 	if (i && i->state == FREE_NID) {
2396 		__remove_free_nid(sbi, i, FREE_NID);
2397 		need_free = true;
2398 	}
2399 	spin_unlock(&nm_i->nid_list_lock);
2400 
2401 	if (need_free)
2402 		kmem_cache_free(free_nid_slab, i);
2403 }
2404 
2405 static int scan_nat_page(struct f2fs_sb_info *sbi,
2406 			struct page *nat_page, nid_t start_nid)
2407 {
2408 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2409 	struct f2fs_nat_block *nat_blk = page_address(nat_page);
2410 	block_t blk_addr;
2411 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2412 	int i;
2413 
2414 	__set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2415 
2416 	i = start_nid % NAT_ENTRY_PER_BLOCK;
2417 
2418 	for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2419 		if (unlikely(start_nid >= nm_i->max_nid))
2420 			break;
2421 
2422 		blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2423 
2424 		if (blk_addr == NEW_ADDR)
2425 			return -EFSCORRUPTED;
2426 
2427 		if (blk_addr == NULL_ADDR) {
2428 			add_free_nid(sbi, start_nid, true, true);
2429 		} else {
2430 			spin_lock(&NM_I(sbi)->nid_list_lock);
2431 			update_free_nid_bitmap(sbi, start_nid, false, true);
2432 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2433 		}
2434 	}
2435 
2436 	return 0;
2437 }
2438 
2439 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2440 {
2441 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2442 	struct f2fs_journal *journal = curseg->journal;
2443 	int i;
2444 
2445 	down_read(&curseg->journal_rwsem);
2446 	for (i = 0; i < nats_in_cursum(journal); i++) {
2447 		block_t addr;
2448 		nid_t nid;
2449 
2450 		addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2451 		nid = le32_to_cpu(nid_in_journal(journal, i));
2452 		if (addr == NULL_ADDR)
2453 			add_free_nid(sbi, nid, true, false);
2454 		else
2455 			remove_free_nid(sbi, nid);
2456 	}
2457 	up_read(&curseg->journal_rwsem);
2458 }
2459 
2460 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2461 {
2462 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2463 	unsigned int i, idx;
2464 	nid_t nid;
2465 
2466 	f2fs_down_read(&nm_i->nat_tree_lock);
2467 
2468 	for (i = 0; i < nm_i->nat_blocks; i++) {
2469 		if (!test_bit_le(i, nm_i->nat_block_bitmap))
2470 			continue;
2471 		if (!nm_i->free_nid_count[i])
2472 			continue;
2473 		for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2474 			idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2475 						NAT_ENTRY_PER_BLOCK, idx);
2476 			if (idx >= NAT_ENTRY_PER_BLOCK)
2477 				break;
2478 
2479 			nid = i * NAT_ENTRY_PER_BLOCK + idx;
2480 			add_free_nid(sbi, nid, true, false);
2481 
2482 			if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2483 				goto out;
2484 		}
2485 	}
2486 out:
2487 	scan_curseg_cache(sbi);
2488 
2489 	f2fs_up_read(&nm_i->nat_tree_lock);
2490 }
2491 
2492 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2493 						bool sync, bool mount)
2494 {
2495 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2496 	int i = 0, ret;
2497 	nid_t nid = nm_i->next_scan_nid;
2498 
2499 	if (unlikely(nid >= nm_i->max_nid))
2500 		nid = 0;
2501 
2502 	if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2503 		nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2504 
2505 	/* Enough entries */
2506 	if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2507 		return 0;
2508 
2509 	if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2510 		return 0;
2511 
2512 	if (!mount) {
2513 		/* try to find free nids in free_nid_bitmap */
2514 		scan_free_nid_bits(sbi);
2515 
2516 		if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2517 			return 0;
2518 	}
2519 
2520 	/* readahead nat pages to be scanned */
2521 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2522 							META_NAT, true);
2523 
2524 	f2fs_down_read(&nm_i->nat_tree_lock);
2525 
2526 	while (1) {
2527 		if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2528 						nm_i->nat_block_bitmap)) {
2529 			struct page *page = get_current_nat_page(sbi, nid);
2530 
2531 			if (IS_ERR(page)) {
2532 				ret = PTR_ERR(page);
2533 			} else {
2534 				ret = scan_nat_page(sbi, page, nid);
2535 				f2fs_put_page(page, 1);
2536 			}
2537 
2538 			if (ret) {
2539 				f2fs_up_read(&nm_i->nat_tree_lock);
2540 
2541 				if (ret == -EFSCORRUPTED) {
2542 					f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2543 					set_sbi_flag(sbi, SBI_NEED_FSCK);
2544 					f2fs_handle_error(sbi,
2545 						ERROR_INCONSISTENT_NAT);
2546 				}
2547 
2548 				return ret;
2549 			}
2550 		}
2551 
2552 		nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2553 		if (unlikely(nid >= nm_i->max_nid))
2554 			nid = 0;
2555 
2556 		if (++i >= FREE_NID_PAGES)
2557 			break;
2558 	}
2559 
2560 	/* go to the next free nat pages to find free nids abundantly */
2561 	nm_i->next_scan_nid = nid;
2562 
2563 	/* find free nids from current sum_pages */
2564 	scan_curseg_cache(sbi);
2565 
2566 	f2fs_up_read(&nm_i->nat_tree_lock);
2567 
2568 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2569 					nm_i->ra_nid_pages, META_NAT, false);
2570 
2571 	return 0;
2572 }
2573 
2574 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2575 {
2576 	int ret;
2577 
2578 	mutex_lock(&NM_I(sbi)->build_lock);
2579 	ret = __f2fs_build_free_nids(sbi, sync, mount);
2580 	mutex_unlock(&NM_I(sbi)->build_lock);
2581 
2582 	return ret;
2583 }
2584 
2585 /*
2586  * If this function returns success, caller can obtain a new nid
2587  * from second parameter of this function.
2588  * The returned nid could be used ino as well as nid when inode is created.
2589  */
2590 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2591 {
2592 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2593 	struct free_nid *i = NULL;
2594 retry:
2595 	if (time_to_inject(sbi, FAULT_ALLOC_NID))
2596 		return false;
2597 
2598 	spin_lock(&nm_i->nid_list_lock);
2599 
2600 	if (unlikely(nm_i->available_nids == 0)) {
2601 		spin_unlock(&nm_i->nid_list_lock);
2602 		return false;
2603 	}
2604 
2605 	/* We should not use stale free nids created by f2fs_build_free_nids */
2606 	if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2607 		f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2608 		i = list_first_entry(&nm_i->free_nid_list,
2609 					struct free_nid, list);
2610 		*nid = i->nid;
2611 
2612 		__move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2613 		nm_i->available_nids--;
2614 
2615 		update_free_nid_bitmap(sbi, *nid, false, false);
2616 
2617 		spin_unlock(&nm_i->nid_list_lock);
2618 		return true;
2619 	}
2620 	spin_unlock(&nm_i->nid_list_lock);
2621 
2622 	/* Let's scan nat pages and its caches to get free nids */
2623 	if (!f2fs_build_free_nids(sbi, true, false))
2624 		goto retry;
2625 	return false;
2626 }
2627 
2628 /*
2629  * f2fs_alloc_nid() should be called prior to this function.
2630  */
2631 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2632 {
2633 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2634 	struct free_nid *i;
2635 
2636 	spin_lock(&nm_i->nid_list_lock);
2637 	i = __lookup_free_nid_list(nm_i, nid);
2638 	f2fs_bug_on(sbi, !i);
2639 	__remove_free_nid(sbi, i, PREALLOC_NID);
2640 	spin_unlock(&nm_i->nid_list_lock);
2641 
2642 	kmem_cache_free(free_nid_slab, i);
2643 }
2644 
2645 /*
2646  * f2fs_alloc_nid() should be called prior to this function.
2647  */
2648 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2649 {
2650 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2651 	struct free_nid *i;
2652 	bool need_free = false;
2653 
2654 	if (!nid)
2655 		return;
2656 
2657 	spin_lock(&nm_i->nid_list_lock);
2658 	i = __lookup_free_nid_list(nm_i, nid);
2659 	f2fs_bug_on(sbi, !i);
2660 
2661 	if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2662 		__remove_free_nid(sbi, i, PREALLOC_NID);
2663 		need_free = true;
2664 	} else {
2665 		__move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2666 	}
2667 
2668 	nm_i->available_nids++;
2669 
2670 	update_free_nid_bitmap(sbi, nid, true, false);
2671 
2672 	spin_unlock(&nm_i->nid_list_lock);
2673 
2674 	if (need_free)
2675 		kmem_cache_free(free_nid_slab, i);
2676 }
2677 
2678 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2679 {
2680 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2681 	int nr = nr_shrink;
2682 
2683 	if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2684 		return 0;
2685 
2686 	if (!mutex_trylock(&nm_i->build_lock))
2687 		return 0;
2688 
2689 	while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2690 		struct free_nid *i, *next;
2691 		unsigned int batch = SHRINK_NID_BATCH_SIZE;
2692 
2693 		spin_lock(&nm_i->nid_list_lock);
2694 		list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2695 			if (!nr_shrink || !batch ||
2696 				nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2697 				break;
2698 			__remove_free_nid(sbi, i, FREE_NID);
2699 			kmem_cache_free(free_nid_slab, i);
2700 			nr_shrink--;
2701 			batch--;
2702 		}
2703 		spin_unlock(&nm_i->nid_list_lock);
2704 	}
2705 
2706 	mutex_unlock(&nm_i->build_lock);
2707 
2708 	return nr - nr_shrink;
2709 }
2710 
2711 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2712 {
2713 	void *src_addr, *dst_addr;
2714 	size_t inline_size;
2715 	struct page *ipage;
2716 	struct f2fs_inode *ri;
2717 
2718 	ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2719 	if (IS_ERR(ipage))
2720 		return PTR_ERR(ipage);
2721 
2722 	ri = F2FS_INODE(page);
2723 	if (ri->i_inline & F2FS_INLINE_XATTR) {
2724 		if (!f2fs_has_inline_xattr(inode)) {
2725 			set_inode_flag(inode, FI_INLINE_XATTR);
2726 			stat_inc_inline_xattr(inode);
2727 		}
2728 	} else {
2729 		if (f2fs_has_inline_xattr(inode)) {
2730 			stat_dec_inline_xattr(inode);
2731 			clear_inode_flag(inode, FI_INLINE_XATTR);
2732 		}
2733 		goto update_inode;
2734 	}
2735 
2736 	dst_addr = inline_xattr_addr(inode, ipage);
2737 	src_addr = inline_xattr_addr(inode, page);
2738 	inline_size = inline_xattr_size(inode);
2739 
2740 	f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2741 	memcpy(dst_addr, src_addr, inline_size);
2742 update_inode:
2743 	f2fs_update_inode(inode, ipage);
2744 	f2fs_put_page(ipage, 1);
2745 	return 0;
2746 }
2747 
2748 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2749 {
2750 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2751 	nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2752 	nid_t new_xnid;
2753 	struct dnode_of_data dn;
2754 	struct node_info ni;
2755 	struct page *xpage;
2756 	int err;
2757 
2758 	if (!prev_xnid)
2759 		goto recover_xnid;
2760 
2761 	/* 1: invalidate the previous xattr nid */
2762 	err = f2fs_get_node_info(sbi, prev_xnid, &ni, false);
2763 	if (err)
2764 		return err;
2765 
2766 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
2767 	dec_valid_node_count(sbi, inode, false);
2768 	set_node_addr(sbi, &ni, NULL_ADDR, false);
2769 
2770 recover_xnid:
2771 	/* 2: update xattr nid in inode */
2772 	if (!f2fs_alloc_nid(sbi, &new_xnid))
2773 		return -ENOSPC;
2774 
2775 	set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2776 	xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2777 	if (IS_ERR(xpage)) {
2778 		f2fs_alloc_nid_failed(sbi, new_xnid);
2779 		return PTR_ERR(xpage);
2780 	}
2781 
2782 	f2fs_alloc_nid_done(sbi, new_xnid);
2783 	f2fs_update_inode_page(inode);
2784 
2785 	/* 3: update and set xattr node page dirty */
2786 	if (page) {
2787 		memcpy(F2FS_NODE(xpage), F2FS_NODE(page),
2788 				VALID_XATTR_BLOCK_SIZE);
2789 		set_page_dirty(xpage);
2790 	}
2791 	f2fs_put_page(xpage, 1);
2792 
2793 	return 0;
2794 }
2795 
2796 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2797 {
2798 	struct f2fs_inode *src, *dst;
2799 	nid_t ino = ino_of_node(page);
2800 	struct node_info old_ni, new_ni;
2801 	struct page *ipage;
2802 	int err;
2803 
2804 	err = f2fs_get_node_info(sbi, ino, &old_ni, false);
2805 	if (err)
2806 		return err;
2807 
2808 	if (unlikely(old_ni.blk_addr != NULL_ADDR))
2809 		return -EINVAL;
2810 retry:
2811 	ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2812 	if (!ipage) {
2813 		memalloc_retry_wait(GFP_NOFS);
2814 		goto retry;
2815 	}
2816 
2817 	/* Should not use this inode from free nid list */
2818 	remove_free_nid(sbi, ino);
2819 
2820 	if (!PageUptodate(ipage))
2821 		SetPageUptodate(ipage);
2822 	fill_node_footer(ipage, ino, ino, 0, true);
2823 	set_cold_node(ipage, false);
2824 
2825 	src = F2FS_INODE(page);
2826 	dst = F2FS_INODE(ipage);
2827 
2828 	memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2829 	dst->i_size = 0;
2830 	dst->i_blocks = cpu_to_le64(1);
2831 	dst->i_links = cpu_to_le32(1);
2832 	dst->i_xattr_nid = 0;
2833 	dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2834 	if (dst->i_inline & F2FS_EXTRA_ATTR) {
2835 		dst->i_extra_isize = src->i_extra_isize;
2836 
2837 		if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2838 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2839 							i_inline_xattr_size))
2840 			dst->i_inline_xattr_size = src->i_inline_xattr_size;
2841 
2842 		if (f2fs_sb_has_project_quota(sbi) &&
2843 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2844 								i_projid))
2845 			dst->i_projid = src->i_projid;
2846 
2847 		if (f2fs_sb_has_inode_crtime(sbi) &&
2848 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2849 							i_crtime_nsec)) {
2850 			dst->i_crtime = src->i_crtime;
2851 			dst->i_crtime_nsec = src->i_crtime_nsec;
2852 		}
2853 	}
2854 
2855 	new_ni = old_ni;
2856 	new_ni.ino = ino;
2857 
2858 	if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2859 		WARN_ON(1);
2860 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2861 	inc_valid_inode_count(sbi);
2862 	set_page_dirty(ipage);
2863 	f2fs_put_page(ipage, 1);
2864 	return 0;
2865 }
2866 
2867 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2868 			unsigned int segno, struct f2fs_summary_block *sum)
2869 {
2870 	struct f2fs_node *rn;
2871 	struct f2fs_summary *sum_entry;
2872 	block_t addr;
2873 	int i, idx, last_offset, nrpages;
2874 
2875 	/* scan the node segment */
2876 	last_offset = BLKS_PER_SEG(sbi);
2877 	addr = START_BLOCK(sbi, segno);
2878 	sum_entry = &sum->entries[0];
2879 
2880 	for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2881 		nrpages = bio_max_segs(last_offset - i);
2882 
2883 		/* readahead node pages */
2884 		f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2885 
2886 		for (idx = addr; idx < addr + nrpages; idx++) {
2887 			struct page *page = f2fs_get_tmp_page(sbi, idx);
2888 
2889 			if (IS_ERR(page))
2890 				return PTR_ERR(page);
2891 
2892 			rn = F2FS_NODE(page);
2893 			sum_entry->nid = rn->footer.nid;
2894 			sum_entry->version = 0;
2895 			sum_entry->ofs_in_node = 0;
2896 			sum_entry++;
2897 			f2fs_put_page(page, 1);
2898 		}
2899 
2900 		invalidate_mapping_pages(META_MAPPING(sbi), addr,
2901 							addr + nrpages);
2902 	}
2903 	return 0;
2904 }
2905 
2906 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2907 {
2908 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2909 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2910 	struct f2fs_journal *journal = curseg->journal;
2911 	int i;
2912 
2913 	down_write(&curseg->journal_rwsem);
2914 	for (i = 0; i < nats_in_cursum(journal); i++) {
2915 		struct nat_entry *ne;
2916 		struct f2fs_nat_entry raw_ne;
2917 		nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2918 
2919 		if (f2fs_check_nid_range(sbi, nid))
2920 			continue;
2921 
2922 		raw_ne = nat_in_journal(journal, i);
2923 
2924 		ne = __lookup_nat_cache(nm_i, nid);
2925 		if (!ne) {
2926 			ne = __alloc_nat_entry(sbi, nid, true);
2927 			__init_nat_entry(nm_i, ne, &raw_ne, true);
2928 		}
2929 
2930 		/*
2931 		 * if a free nat in journal has not been used after last
2932 		 * checkpoint, we should remove it from available nids,
2933 		 * since later we will add it again.
2934 		 */
2935 		if (!get_nat_flag(ne, IS_DIRTY) &&
2936 				le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2937 			spin_lock(&nm_i->nid_list_lock);
2938 			nm_i->available_nids--;
2939 			spin_unlock(&nm_i->nid_list_lock);
2940 		}
2941 
2942 		__set_nat_cache_dirty(nm_i, ne);
2943 	}
2944 	update_nats_in_cursum(journal, -i);
2945 	up_write(&curseg->journal_rwsem);
2946 }
2947 
2948 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2949 						struct list_head *head, int max)
2950 {
2951 	struct nat_entry_set *cur;
2952 
2953 	if (nes->entry_cnt >= max)
2954 		goto add_out;
2955 
2956 	list_for_each_entry(cur, head, set_list) {
2957 		if (cur->entry_cnt >= nes->entry_cnt) {
2958 			list_add(&nes->set_list, cur->set_list.prev);
2959 			return;
2960 		}
2961 	}
2962 add_out:
2963 	list_add_tail(&nes->set_list, head);
2964 }
2965 
2966 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs,
2967 							unsigned int valid)
2968 {
2969 	if (valid == 0) {
2970 		__set_bit_le(nat_ofs, nm_i->empty_nat_bits);
2971 		__clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2972 		return;
2973 	}
2974 
2975 	__clear_bit_le(nat_ofs, nm_i->empty_nat_bits);
2976 	if (valid == NAT_ENTRY_PER_BLOCK)
2977 		__set_bit_le(nat_ofs, nm_i->full_nat_bits);
2978 	else
2979 		__clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2980 }
2981 
2982 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2983 						struct page *page)
2984 {
2985 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2986 	unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2987 	struct f2fs_nat_block *nat_blk = page_address(page);
2988 	int valid = 0;
2989 	int i = 0;
2990 
2991 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
2992 		return;
2993 
2994 	if (nat_index == 0) {
2995 		valid = 1;
2996 		i = 1;
2997 	}
2998 	for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2999 		if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
3000 			valid++;
3001 	}
3002 
3003 	__update_nat_bits(nm_i, nat_index, valid);
3004 }
3005 
3006 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi)
3007 {
3008 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3009 	unsigned int nat_ofs;
3010 
3011 	f2fs_down_read(&nm_i->nat_tree_lock);
3012 
3013 	for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) {
3014 		unsigned int valid = 0, nid_ofs = 0;
3015 
3016 		/* handle nid zero due to it should never be used */
3017 		if (unlikely(nat_ofs == 0)) {
3018 			valid = 1;
3019 			nid_ofs = 1;
3020 		}
3021 
3022 		for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) {
3023 			if (!test_bit_le(nid_ofs,
3024 					nm_i->free_nid_bitmap[nat_ofs]))
3025 				valid++;
3026 		}
3027 
3028 		__update_nat_bits(nm_i, nat_ofs, valid);
3029 	}
3030 
3031 	f2fs_up_read(&nm_i->nat_tree_lock);
3032 }
3033 
3034 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
3035 		struct nat_entry_set *set, struct cp_control *cpc)
3036 {
3037 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3038 	struct f2fs_journal *journal = curseg->journal;
3039 	nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
3040 	bool to_journal = true;
3041 	struct f2fs_nat_block *nat_blk;
3042 	struct nat_entry *ne, *cur;
3043 	struct page *page = NULL;
3044 
3045 	/*
3046 	 * there are two steps to flush nat entries:
3047 	 * #1, flush nat entries to journal in current hot data summary block.
3048 	 * #2, flush nat entries to nat page.
3049 	 */
3050 	if ((cpc->reason & CP_UMOUNT) ||
3051 		!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
3052 		to_journal = false;
3053 
3054 	if (to_journal) {
3055 		down_write(&curseg->journal_rwsem);
3056 	} else {
3057 		page = get_next_nat_page(sbi, start_nid);
3058 		if (IS_ERR(page))
3059 			return PTR_ERR(page);
3060 
3061 		nat_blk = page_address(page);
3062 		f2fs_bug_on(sbi, !nat_blk);
3063 	}
3064 
3065 	/* flush dirty nats in nat entry set */
3066 	list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
3067 		struct f2fs_nat_entry *raw_ne;
3068 		nid_t nid = nat_get_nid(ne);
3069 		int offset;
3070 
3071 		f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
3072 
3073 		if (to_journal) {
3074 			offset = f2fs_lookup_journal_in_cursum(journal,
3075 							NAT_JOURNAL, nid, 1);
3076 			f2fs_bug_on(sbi, offset < 0);
3077 			raw_ne = &nat_in_journal(journal, offset);
3078 			nid_in_journal(journal, offset) = cpu_to_le32(nid);
3079 		} else {
3080 			raw_ne = &nat_blk->entries[nid - start_nid];
3081 		}
3082 		raw_nat_from_node_info(raw_ne, &ne->ni);
3083 		nat_reset_flag(ne);
3084 		__clear_nat_cache_dirty(NM_I(sbi), set, ne);
3085 		if (nat_get_blkaddr(ne) == NULL_ADDR) {
3086 			add_free_nid(sbi, nid, false, true);
3087 		} else {
3088 			spin_lock(&NM_I(sbi)->nid_list_lock);
3089 			update_free_nid_bitmap(sbi, nid, false, false);
3090 			spin_unlock(&NM_I(sbi)->nid_list_lock);
3091 		}
3092 	}
3093 
3094 	if (to_journal) {
3095 		up_write(&curseg->journal_rwsem);
3096 	} else {
3097 		update_nat_bits(sbi, start_nid, page);
3098 		f2fs_put_page(page, 1);
3099 	}
3100 
3101 	/* Allow dirty nats by node block allocation in write_begin */
3102 	if (!set->entry_cnt) {
3103 		radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
3104 		kmem_cache_free(nat_entry_set_slab, set);
3105 	}
3106 	return 0;
3107 }
3108 
3109 /*
3110  * This function is called during the checkpointing process.
3111  */
3112 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3113 {
3114 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3115 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3116 	struct f2fs_journal *journal = curseg->journal;
3117 	struct nat_entry_set *setvec[NAT_VEC_SIZE];
3118 	struct nat_entry_set *set, *tmp;
3119 	unsigned int found;
3120 	nid_t set_idx = 0;
3121 	LIST_HEAD(sets);
3122 	int err = 0;
3123 
3124 	/*
3125 	 * during unmount, let's flush nat_bits before checking
3126 	 * nat_cnt[DIRTY_NAT].
3127 	 */
3128 	if (cpc->reason & CP_UMOUNT) {
3129 		f2fs_down_write(&nm_i->nat_tree_lock);
3130 		remove_nats_in_journal(sbi);
3131 		f2fs_up_write(&nm_i->nat_tree_lock);
3132 	}
3133 
3134 	if (!nm_i->nat_cnt[DIRTY_NAT])
3135 		return 0;
3136 
3137 	f2fs_down_write(&nm_i->nat_tree_lock);
3138 
3139 	/*
3140 	 * if there are no enough space in journal to store dirty nat
3141 	 * entries, remove all entries from journal and merge them
3142 	 * into nat entry set.
3143 	 */
3144 	if (cpc->reason & CP_UMOUNT ||
3145 		!__has_cursum_space(journal,
3146 			nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3147 		remove_nats_in_journal(sbi);
3148 
3149 	while ((found = __gang_lookup_nat_set(nm_i,
3150 					set_idx, NAT_VEC_SIZE, setvec))) {
3151 		unsigned idx;
3152 
3153 		set_idx = setvec[found - 1]->set + 1;
3154 		for (idx = 0; idx < found; idx++)
3155 			__adjust_nat_entry_set(setvec[idx], &sets,
3156 						MAX_NAT_JENTRIES(journal));
3157 	}
3158 
3159 	/* flush dirty nats in nat entry set */
3160 	list_for_each_entry_safe(set, tmp, &sets, set_list) {
3161 		err = __flush_nat_entry_set(sbi, set, cpc);
3162 		if (err)
3163 			break;
3164 	}
3165 
3166 	f2fs_up_write(&nm_i->nat_tree_lock);
3167 	/* Allow dirty nats by node block allocation in write_begin */
3168 
3169 	return err;
3170 }
3171 
3172 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3173 {
3174 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3175 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3176 	unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3177 	unsigned int i;
3178 	__u64 cp_ver = cur_cp_version(ckpt);
3179 	block_t nat_bits_addr;
3180 
3181 	nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3182 	nm_i->nat_bits = f2fs_kvzalloc(sbi,
3183 			F2FS_BLK_TO_BYTES(nm_i->nat_bits_blocks), GFP_KERNEL);
3184 	if (!nm_i->nat_bits)
3185 		return -ENOMEM;
3186 
3187 	nm_i->full_nat_bits = nm_i->nat_bits + 8;
3188 	nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3189 
3190 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3191 		return 0;
3192 
3193 	nat_bits_addr = __start_cp_addr(sbi) + BLKS_PER_SEG(sbi) -
3194 						nm_i->nat_bits_blocks;
3195 	for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3196 		struct page *page;
3197 
3198 		page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3199 		if (IS_ERR(page))
3200 			return PTR_ERR(page);
3201 
3202 		memcpy(nm_i->nat_bits + F2FS_BLK_TO_BYTES(i),
3203 					page_address(page), F2FS_BLKSIZE);
3204 		f2fs_put_page(page, 1);
3205 	}
3206 
3207 	cp_ver |= (cur_cp_crc(ckpt) << 32);
3208 	if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3209 		clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
3210 		f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)",
3211 			cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits));
3212 		return 0;
3213 	}
3214 
3215 	f2fs_notice(sbi, "Found nat_bits in checkpoint");
3216 	return 0;
3217 }
3218 
3219 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3220 {
3221 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3222 	unsigned int i = 0;
3223 	nid_t nid, last_nid;
3224 
3225 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3226 		return;
3227 
3228 	for (i = 0; i < nm_i->nat_blocks; i++) {
3229 		i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3230 		if (i >= nm_i->nat_blocks)
3231 			break;
3232 
3233 		__set_bit_le(i, nm_i->nat_block_bitmap);
3234 
3235 		nid = i * NAT_ENTRY_PER_BLOCK;
3236 		last_nid = nid + NAT_ENTRY_PER_BLOCK;
3237 
3238 		spin_lock(&NM_I(sbi)->nid_list_lock);
3239 		for (; nid < last_nid; nid++)
3240 			update_free_nid_bitmap(sbi, nid, true, true);
3241 		spin_unlock(&NM_I(sbi)->nid_list_lock);
3242 	}
3243 
3244 	for (i = 0; i < nm_i->nat_blocks; i++) {
3245 		i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3246 		if (i >= nm_i->nat_blocks)
3247 			break;
3248 
3249 		__set_bit_le(i, nm_i->nat_block_bitmap);
3250 	}
3251 }
3252 
3253 static int init_node_manager(struct f2fs_sb_info *sbi)
3254 {
3255 	struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3256 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3257 	unsigned char *version_bitmap;
3258 	unsigned int nat_segs;
3259 	int err;
3260 
3261 	nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3262 
3263 	/* segment_count_nat includes pair segment so divide to 2. */
3264 	nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3265 	nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3266 	nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3267 
3268 	/* not used nids: 0, node, meta, (and root counted as valid node) */
3269 	nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3270 						F2FS_RESERVED_NODE_NUM;
3271 	nm_i->nid_cnt[FREE_NID] = 0;
3272 	nm_i->nid_cnt[PREALLOC_NID] = 0;
3273 	nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3274 	nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3275 	nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3276 	nm_i->max_rf_node_blocks = DEF_RF_NODE_BLOCKS;
3277 
3278 	INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3279 	INIT_LIST_HEAD(&nm_i->free_nid_list);
3280 	INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3281 	INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3282 	INIT_LIST_HEAD(&nm_i->nat_entries);
3283 	spin_lock_init(&nm_i->nat_list_lock);
3284 
3285 	mutex_init(&nm_i->build_lock);
3286 	spin_lock_init(&nm_i->nid_list_lock);
3287 	init_f2fs_rwsem(&nm_i->nat_tree_lock);
3288 
3289 	nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3290 	nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3291 	version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3292 	nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3293 					GFP_KERNEL);
3294 	if (!nm_i->nat_bitmap)
3295 		return -ENOMEM;
3296 
3297 	err = __get_nat_bitmaps(sbi);
3298 	if (err)
3299 		return err;
3300 
3301 #ifdef CONFIG_F2FS_CHECK_FS
3302 	nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3303 					GFP_KERNEL);
3304 	if (!nm_i->nat_bitmap_mir)
3305 		return -ENOMEM;
3306 #endif
3307 
3308 	return 0;
3309 }
3310 
3311 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3312 {
3313 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3314 	int i;
3315 
3316 	nm_i->free_nid_bitmap =
3317 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3318 					      nm_i->nat_blocks),
3319 			      GFP_KERNEL);
3320 	if (!nm_i->free_nid_bitmap)
3321 		return -ENOMEM;
3322 
3323 	for (i = 0; i < nm_i->nat_blocks; i++) {
3324 		nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3325 			f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3326 		if (!nm_i->free_nid_bitmap[i])
3327 			return -ENOMEM;
3328 	}
3329 
3330 	nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3331 								GFP_KERNEL);
3332 	if (!nm_i->nat_block_bitmap)
3333 		return -ENOMEM;
3334 
3335 	nm_i->free_nid_count =
3336 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3337 					      nm_i->nat_blocks),
3338 			      GFP_KERNEL);
3339 	if (!nm_i->free_nid_count)
3340 		return -ENOMEM;
3341 	return 0;
3342 }
3343 
3344 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3345 {
3346 	int err;
3347 
3348 	sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3349 							GFP_KERNEL);
3350 	if (!sbi->nm_info)
3351 		return -ENOMEM;
3352 
3353 	err = init_node_manager(sbi);
3354 	if (err)
3355 		return err;
3356 
3357 	err = init_free_nid_cache(sbi);
3358 	if (err)
3359 		return err;
3360 
3361 	/* load free nid status from nat_bits table */
3362 	load_free_nid_bitmap(sbi);
3363 
3364 	return f2fs_build_free_nids(sbi, true, true);
3365 }
3366 
3367 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3368 {
3369 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3370 	struct free_nid *i, *next_i;
3371 	void *vec[NAT_VEC_SIZE];
3372 	struct nat_entry **natvec = (struct nat_entry **)vec;
3373 	struct nat_entry_set **setvec = (struct nat_entry_set **)vec;
3374 	nid_t nid = 0;
3375 	unsigned int found;
3376 
3377 	if (!nm_i)
3378 		return;
3379 
3380 	/* destroy free nid list */
3381 	spin_lock(&nm_i->nid_list_lock);
3382 	list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3383 		__remove_free_nid(sbi, i, FREE_NID);
3384 		spin_unlock(&nm_i->nid_list_lock);
3385 		kmem_cache_free(free_nid_slab, i);
3386 		spin_lock(&nm_i->nid_list_lock);
3387 	}
3388 	f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3389 	f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3390 	f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3391 	spin_unlock(&nm_i->nid_list_lock);
3392 
3393 	/* destroy nat cache */
3394 	f2fs_down_write(&nm_i->nat_tree_lock);
3395 	while ((found = __gang_lookup_nat_cache(nm_i,
3396 					nid, NAT_VEC_SIZE, natvec))) {
3397 		unsigned idx;
3398 
3399 		nid = nat_get_nid(natvec[found - 1]) + 1;
3400 		for (idx = 0; idx < found; idx++) {
3401 			spin_lock(&nm_i->nat_list_lock);
3402 			list_del(&natvec[idx]->list);
3403 			spin_unlock(&nm_i->nat_list_lock);
3404 
3405 			__del_from_nat_cache(nm_i, natvec[idx]);
3406 		}
3407 	}
3408 	f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3409 
3410 	/* destroy nat set cache */
3411 	nid = 0;
3412 	memset(vec, 0, sizeof(void *) * NAT_VEC_SIZE);
3413 	while ((found = __gang_lookup_nat_set(nm_i,
3414 					nid, NAT_VEC_SIZE, setvec))) {
3415 		unsigned idx;
3416 
3417 		nid = setvec[found - 1]->set + 1;
3418 		for (idx = 0; idx < found; idx++) {
3419 			/* entry_cnt is not zero, when cp_error was occurred */
3420 			f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3421 			radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3422 			kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3423 		}
3424 	}
3425 	f2fs_up_write(&nm_i->nat_tree_lock);
3426 
3427 	kvfree(nm_i->nat_block_bitmap);
3428 	if (nm_i->free_nid_bitmap) {
3429 		int i;
3430 
3431 		for (i = 0; i < nm_i->nat_blocks; i++)
3432 			kvfree(nm_i->free_nid_bitmap[i]);
3433 		kvfree(nm_i->free_nid_bitmap);
3434 	}
3435 	kvfree(nm_i->free_nid_count);
3436 
3437 	kvfree(nm_i->nat_bitmap);
3438 	kvfree(nm_i->nat_bits);
3439 #ifdef CONFIG_F2FS_CHECK_FS
3440 	kvfree(nm_i->nat_bitmap_mir);
3441 #endif
3442 	sbi->nm_info = NULL;
3443 	kfree(nm_i);
3444 }
3445 
3446 int __init f2fs_create_node_manager_caches(void)
3447 {
3448 	nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3449 			sizeof(struct nat_entry));
3450 	if (!nat_entry_slab)
3451 		goto fail;
3452 
3453 	free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3454 			sizeof(struct free_nid));
3455 	if (!free_nid_slab)
3456 		goto destroy_nat_entry;
3457 
3458 	nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3459 			sizeof(struct nat_entry_set));
3460 	if (!nat_entry_set_slab)
3461 		goto destroy_free_nid;
3462 
3463 	fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3464 			sizeof(struct fsync_node_entry));
3465 	if (!fsync_node_entry_slab)
3466 		goto destroy_nat_entry_set;
3467 	return 0;
3468 
3469 destroy_nat_entry_set:
3470 	kmem_cache_destroy(nat_entry_set_slab);
3471 destroy_free_nid:
3472 	kmem_cache_destroy(free_nid_slab);
3473 destroy_nat_entry:
3474 	kmem_cache_destroy(nat_entry_slab);
3475 fail:
3476 	return -ENOMEM;
3477 }
3478 
3479 void f2fs_destroy_node_manager_caches(void)
3480 {
3481 	kmem_cache_destroy(fsync_node_entry_slab);
3482 	kmem_cache_destroy(nat_entry_set_slab);
3483 	kmem_cache_destroy(free_nid_slab);
3484 	kmem_cache_destroy(nat_entry_slab);
3485 }
3486