xref: /linux/fs/f2fs/node.c (revision d99ff463ecf651437e9e4abe68f331dfb6b5bd9d)
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(nmi) 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);
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, 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 c_len = f2fs_cluster_blocks_are_contiguous(dn);
856 		block_t blkaddr;
857 
858 		if (!c_len)
859 			goto out;
860 
861 		blkaddr = f2fs_data_blkaddr(dn);
862 		if (blkaddr == COMPRESS_ADDR)
863 			blkaddr = data_blkaddr(dn->inode, dn->node_page,
864 						dn->ofs_in_node + 1);
865 
866 		f2fs_update_read_extent_tree_range_compressed(dn->inode,
867 					index, blkaddr,
868 					F2FS_I(dn->inode)->i_cluster_size,
869 					c_len);
870 	}
871 out:
872 	return 0;
873 
874 release_pages:
875 	f2fs_put_page(parent, 1);
876 	if (i > 1)
877 		f2fs_put_page(npage[0], 0);
878 release_out:
879 	dn->inode_page = NULL;
880 	dn->node_page = NULL;
881 	if (err == -ENOENT) {
882 		dn->cur_level = i;
883 		dn->max_level = level;
884 		dn->ofs_in_node = offset[level];
885 	}
886 	return err;
887 }
888 
889 static int truncate_node(struct dnode_of_data *dn)
890 {
891 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
892 	struct node_info ni;
893 	int err;
894 	pgoff_t index;
895 
896 	err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
897 	if (err)
898 		return err;
899 
900 	/* Deallocate node address */
901 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
902 	dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
903 	set_node_addr(sbi, &ni, NULL_ADDR, false);
904 
905 	if (dn->nid == dn->inode->i_ino) {
906 		f2fs_remove_orphan_inode(sbi, dn->nid);
907 		dec_valid_inode_count(sbi);
908 		f2fs_inode_synced(dn->inode);
909 	}
910 
911 	clear_node_page_dirty(dn->node_page);
912 	set_sbi_flag(sbi, SBI_IS_DIRTY);
913 
914 	index = dn->node_page->index;
915 	f2fs_put_page(dn->node_page, 1);
916 
917 	invalidate_mapping_pages(NODE_MAPPING(sbi),
918 			index, index);
919 
920 	dn->node_page = NULL;
921 	trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
922 
923 	return 0;
924 }
925 
926 static int truncate_dnode(struct dnode_of_data *dn)
927 {
928 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
929 	struct page *page;
930 	int err;
931 
932 	if (dn->nid == 0)
933 		return 1;
934 
935 	/* get direct node */
936 	page = f2fs_get_node_page(sbi, dn->nid);
937 	if (PTR_ERR(page) == -ENOENT)
938 		return 1;
939 	else if (IS_ERR(page))
940 		return PTR_ERR(page);
941 
942 	if (IS_INODE(page) || ino_of_node(page) != dn->inode->i_ino) {
943 		f2fs_err(sbi, "incorrect node reference, ino: %lu, nid: %u, ino_of_node: %u",
944 				dn->inode->i_ino, dn->nid, ino_of_node(page));
945 		set_sbi_flag(sbi, SBI_NEED_FSCK);
946 		f2fs_handle_error(sbi, ERROR_INVALID_NODE_REFERENCE);
947 		f2fs_put_page(page, 1);
948 		return -EFSCORRUPTED;
949 	}
950 
951 	/* Make dnode_of_data for parameter */
952 	dn->node_page = page;
953 	dn->ofs_in_node = 0;
954 	f2fs_truncate_data_blocks_range(dn, ADDRS_PER_BLOCK(dn->inode));
955 	err = truncate_node(dn);
956 	if (err) {
957 		f2fs_put_page(page, 1);
958 		return err;
959 	}
960 
961 	return 1;
962 }
963 
964 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
965 						int ofs, int depth)
966 {
967 	struct dnode_of_data rdn = *dn;
968 	struct page *page;
969 	struct f2fs_node *rn;
970 	nid_t child_nid;
971 	unsigned int child_nofs;
972 	int freed = 0;
973 	int i, ret;
974 
975 	if (dn->nid == 0)
976 		return NIDS_PER_BLOCK + 1;
977 
978 	trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
979 
980 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
981 	if (IS_ERR(page)) {
982 		trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
983 		return PTR_ERR(page);
984 	}
985 
986 	f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
987 
988 	rn = F2FS_NODE(page);
989 	if (depth < 3) {
990 		for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
991 			child_nid = le32_to_cpu(rn->in.nid[i]);
992 			if (child_nid == 0)
993 				continue;
994 			rdn.nid = child_nid;
995 			ret = truncate_dnode(&rdn);
996 			if (ret < 0)
997 				goto out_err;
998 			if (set_nid(page, i, 0, false))
999 				dn->node_changed = true;
1000 		}
1001 	} else {
1002 		child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
1003 		for (i = ofs; i < NIDS_PER_BLOCK; i++) {
1004 			child_nid = le32_to_cpu(rn->in.nid[i]);
1005 			if (child_nid == 0) {
1006 				child_nofs += NIDS_PER_BLOCK + 1;
1007 				continue;
1008 			}
1009 			rdn.nid = child_nid;
1010 			ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
1011 			if (ret == (NIDS_PER_BLOCK + 1)) {
1012 				if (set_nid(page, i, 0, false))
1013 					dn->node_changed = true;
1014 				child_nofs += ret;
1015 			} else if (ret < 0 && ret != -ENOENT) {
1016 				goto out_err;
1017 			}
1018 		}
1019 		freed = child_nofs;
1020 	}
1021 
1022 	if (!ofs) {
1023 		/* remove current indirect node */
1024 		dn->node_page = page;
1025 		ret = truncate_node(dn);
1026 		if (ret)
1027 			goto out_err;
1028 		freed++;
1029 	} else {
1030 		f2fs_put_page(page, 1);
1031 	}
1032 	trace_f2fs_truncate_nodes_exit(dn->inode, freed);
1033 	return freed;
1034 
1035 out_err:
1036 	f2fs_put_page(page, 1);
1037 	trace_f2fs_truncate_nodes_exit(dn->inode, ret);
1038 	return ret;
1039 }
1040 
1041 static int truncate_partial_nodes(struct dnode_of_data *dn,
1042 			struct f2fs_inode *ri, int *offset, int depth)
1043 {
1044 	struct page *pages[2];
1045 	nid_t nid[3];
1046 	nid_t child_nid;
1047 	int err = 0;
1048 	int i;
1049 	int idx = depth - 2;
1050 
1051 	nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1052 	if (!nid[0])
1053 		return 0;
1054 
1055 	/* get indirect nodes in the path */
1056 	for (i = 0; i < idx + 1; i++) {
1057 		/* reference count'll be increased */
1058 		pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
1059 		if (IS_ERR(pages[i])) {
1060 			err = PTR_ERR(pages[i]);
1061 			idx = i - 1;
1062 			goto fail;
1063 		}
1064 		nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1065 	}
1066 
1067 	f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1068 
1069 	/* free direct nodes linked to a partial indirect node */
1070 	for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1071 		child_nid = get_nid(pages[idx], i, false);
1072 		if (!child_nid)
1073 			continue;
1074 		dn->nid = child_nid;
1075 		err = truncate_dnode(dn);
1076 		if (err < 0)
1077 			goto fail;
1078 		if (set_nid(pages[idx], i, 0, false))
1079 			dn->node_changed = true;
1080 	}
1081 
1082 	if (offset[idx + 1] == 0) {
1083 		dn->node_page = pages[idx];
1084 		dn->nid = nid[idx];
1085 		err = truncate_node(dn);
1086 		if (err)
1087 			goto fail;
1088 	} else {
1089 		f2fs_put_page(pages[idx], 1);
1090 	}
1091 	offset[idx]++;
1092 	offset[idx + 1] = 0;
1093 	idx--;
1094 fail:
1095 	for (i = idx; i >= 0; i--)
1096 		f2fs_put_page(pages[i], 1);
1097 
1098 	trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1099 
1100 	return err;
1101 }
1102 
1103 /*
1104  * All the block addresses of data and nodes should be nullified.
1105  */
1106 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1107 {
1108 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1109 	int err = 0, cont = 1;
1110 	int level, offset[4], noffset[4];
1111 	unsigned int nofs = 0;
1112 	struct f2fs_inode *ri;
1113 	struct dnode_of_data dn;
1114 	struct page *page;
1115 
1116 	trace_f2fs_truncate_inode_blocks_enter(inode, from);
1117 
1118 	level = get_node_path(inode, from, offset, noffset);
1119 	if (level < 0) {
1120 		trace_f2fs_truncate_inode_blocks_exit(inode, level);
1121 		return level;
1122 	}
1123 
1124 	page = f2fs_get_node_page(sbi, inode->i_ino);
1125 	if (IS_ERR(page)) {
1126 		trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1127 		return PTR_ERR(page);
1128 	}
1129 
1130 	set_new_dnode(&dn, inode, page, NULL, 0);
1131 	unlock_page(page);
1132 
1133 	ri = F2FS_INODE(page);
1134 	switch (level) {
1135 	case 0:
1136 	case 1:
1137 		nofs = noffset[1];
1138 		break;
1139 	case 2:
1140 		nofs = noffset[1];
1141 		if (!offset[level - 1])
1142 			goto skip_partial;
1143 		err = truncate_partial_nodes(&dn, ri, offset, level);
1144 		if (err < 0 && err != -ENOENT)
1145 			goto fail;
1146 		nofs += 1 + NIDS_PER_BLOCK;
1147 		break;
1148 	case 3:
1149 		nofs = 5 + 2 * NIDS_PER_BLOCK;
1150 		if (!offset[level - 1])
1151 			goto skip_partial;
1152 		err = truncate_partial_nodes(&dn, ri, offset, level);
1153 		if (err < 0 && err != -ENOENT)
1154 			goto fail;
1155 		break;
1156 	default:
1157 		BUG();
1158 	}
1159 
1160 skip_partial:
1161 	while (cont) {
1162 		dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1163 		switch (offset[0]) {
1164 		case NODE_DIR1_BLOCK:
1165 		case NODE_DIR2_BLOCK:
1166 			err = truncate_dnode(&dn);
1167 			break;
1168 
1169 		case NODE_IND1_BLOCK:
1170 		case NODE_IND2_BLOCK:
1171 			err = truncate_nodes(&dn, nofs, offset[1], 2);
1172 			break;
1173 
1174 		case NODE_DIND_BLOCK:
1175 			err = truncate_nodes(&dn, nofs, offset[1], 3);
1176 			cont = 0;
1177 			break;
1178 
1179 		default:
1180 			BUG();
1181 		}
1182 		if (err < 0 && err != -ENOENT)
1183 			goto fail;
1184 		if (offset[1] == 0 &&
1185 				ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1186 			lock_page(page);
1187 			BUG_ON(page->mapping != NODE_MAPPING(sbi));
1188 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1189 			ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1190 			set_page_dirty(page);
1191 			unlock_page(page);
1192 		}
1193 		offset[1] = 0;
1194 		offset[0]++;
1195 		nofs += err;
1196 	}
1197 fail:
1198 	f2fs_put_page(page, 0);
1199 	trace_f2fs_truncate_inode_blocks_exit(inode, err);
1200 	return err > 0 ? 0 : err;
1201 }
1202 
1203 /* caller must lock inode page */
1204 int f2fs_truncate_xattr_node(struct inode *inode)
1205 {
1206 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1207 	nid_t nid = F2FS_I(inode)->i_xattr_nid;
1208 	struct dnode_of_data dn;
1209 	struct page *npage;
1210 	int err;
1211 
1212 	if (!nid)
1213 		return 0;
1214 
1215 	npage = f2fs_get_node_page(sbi, nid);
1216 	if (IS_ERR(npage))
1217 		return PTR_ERR(npage);
1218 
1219 	set_new_dnode(&dn, inode, NULL, npage, nid);
1220 	err = truncate_node(&dn);
1221 	if (err) {
1222 		f2fs_put_page(npage, 1);
1223 		return err;
1224 	}
1225 
1226 	f2fs_i_xnid_write(inode, 0);
1227 
1228 	return 0;
1229 }
1230 
1231 /*
1232  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1233  * f2fs_unlock_op().
1234  */
1235 int f2fs_remove_inode_page(struct inode *inode)
1236 {
1237 	struct dnode_of_data dn;
1238 	int err;
1239 
1240 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1241 	err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1242 	if (err)
1243 		return err;
1244 
1245 	err = f2fs_truncate_xattr_node(inode);
1246 	if (err) {
1247 		f2fs_put_dnode(&dn);
1248 		return err;
1249 	}
1250 
1251 	/* remove potential inline_data blocks */
1252 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1253 				S_ISLNK(inode->i_mode))
1254 		f2fs_truncate_data_blocks_range(&dn, 1);
1255 
1256 	/* 0 is possible, after f2fs_new_inode() has failed */
1257 	if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1258 		f2fs_put_dnode(&dn);
1259 		return -EIO;
1260 	}
1261 
1262 	if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1263 		f2fs_warn(F2FS_I_SB(inode),
1264 			"f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1265 			inode->i_ino, (unsigned long long)inode->i_blocks);
1266 		set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1267 	}
1268 
1269 	/* will put inode & node pages */
1270 	err = truncate_node(&dn);
1271 	if (err) {
1272 		f2fs_put_dnode(&dn);
1273 		return err;
1274 	}
1275 	return 0;
1276 }
1277 
1278 struct page *f2fs_new_inode_page(struct inode *inode)
1279 {
1280 	struct dnode_of_data dn;
1281 
1282 	/* allocate inode page for new inode */
1283 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1284 
1285 	/* caller should f2fs_put_page(page, 1); */
1286 	return f2fs_new_node_page(&dn, 0);
1287 }
1288 
1289 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1290 {
1291 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1292 	struct node_info new_ni;
1293 	struct page *page;
1294 	int err;
1295 
1296 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1297 		return ERR_PTR(-EPERM);
1298 
1299 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1300 	if (!page)
1301 		return ERR_PTR(-ENOMEM);
1302 
1303 	if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1304 		goto fail;
1305 
1306 #ifdef CONFIG_F2FS_CHECK_FS
1307 	err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false);
1308 	if (err) {
1309 		dec_valid_node_count(sbi, dn->inode, !ofs);
1310 		goto fail;
1311 	}
1312 	if (unlikely(new_ni.blk_addr != NULL_ADDR)) {
1313 		err = -EFSCORRUPTED;
1314 		set_sbi_flag(sbi, SBI_NEED_FSCK);
1315 		f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1316 		goto fail;
1317 	}
1318 #endif
1319 	new_ni.nid = dn->nid;
1320 	new_ni.ino = dn->inode->i_ino;
1321 	new_ni.blk_addr = NULL_ADDR;
1322 	new_ni.flag = 0;
1323 	new_ni.version = 0;
1324 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1325 
1326 	f2fs_wait_on_page_writeback(page, NODE, true, true);
1327 	fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1328 	set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1329 	if (!PageUptodate(page))
1330 		SetPageUptodate(page);
1331 	if (set_page_dirty(page))
1332 		dn->node_changed = true;
1333 
1334 	if (f2fs_has_xattr_block(ofs))
1335 		f2fs_i_xnid_write(dn->inode, dn->nid);
1336 
1337 	if (ofs == 0)
1338 		inc_valid_inode_count(sbi);
1339 	return page;
1340 
1341 fail:
1342 	clear_node_page_dirty(page);
1343 	f2fs_put_page(page, 1);
1344 	return ERR_PTR(err);
1345 }
1346 
1347 /*
1348  * Caller should do after getting the following values.
1349  * 0: f2fs_put_page(page, 0)
1350  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1351  */
1352 static int read_node_page(struct page *page, blk_opf_t op_flags)
1353 {
1354 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1355 	struct node_info ni;
1356 	struct f2fs_io_info fio = {
1357 		.sbi = sbi,
1358 		.type = NODE,
1359 		.op = REQ_OP_READ,
1360 		.op_flags = op_flags,
1361 		.page = page,
1362 		.encrypted_page = NULL,
1363 	};
1364 	int err;
1365 
1366 	if (PageUptodate(page)) {
1367 		if (!f2fs_inode_chksum_verify(sbi, page)) {
1368 			ClearPageUptodate(page);
1369 			return -EFSBADCRC;
1370 		}
1371 		return LOCKED_PAGE;
1372 	}
1373 
1374 	err = f2fs_get_node_info(sbi, page->index, &ni, false);
1375 	if (err)
1376 		return err;
1377 
1378 	/* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1379 	if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR)) {
1380 		ClearPageUptodate(page);
1381 		return -ENOENT;
1382 	}
1383 
1384 	fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1385 
1386 	err = f2fs_submit_page_bio(&fio);
1387 
1388 	if (!err)
1389 		f2fs_update_iostat(sbi, NULL, FS_NODE_READ_IO, F2FS_BLKSIZE);
1390 
1391 	return err;
1392 }
1393 
1394 /*
1395  * Readahead a node page
1396  */
1397 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1398 {
1399 	struct page *apage;
1400 	int err;
1401 
1402 	if (!nid)
1403 		return;
1404 	if (f2fs_check_nid_range(sbi, nid))
1405 		return;
1406 
1407 	apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1408 	if (apage)
1409 		return;
1410 
1411 	apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1412 	if (!apage)
1413 		return;
1414 
1415 	err = read_node_page(apage, REQ_RAHEAD);
1416 	f2fs_put_page(apage, err ? 1 : 0);
1417 }
1418 
1419 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1420 					struct page *parent, int start)
1421 {
1422 	struct page *page;
1423 	int err;
1424 
1425 	if (!nid)
1426 		return ERR_PTR(-ENOENT);
1427 	if (f2fs_check_nid_range(sbi, nid))
1428 		return ERR_PTR(-EINVAL);
1429 repeat:
1430 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1431 	if (!page)
1432 		return ERR_PTR(-ENOMEM);
1433 
1434 	err = read_node_page(page, 0);
1435 	if (err < 0) {
1436 		goto out_put_err;
1437 	} else if (err == LOCKED_PAGE) {
1438 		err = 0;
1439 		goto page_hit;
1440 	}
1441 
1442 	if (parent)
1443 		f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1444 
1445 	lock_page(page);
1446 
1447 	if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1448 		f2fs_put_page(page, 1);
1449 		goto repeat;
1450 	}
1451 
1452 	if (unlikely(!PageUptodate(page))) {
1453 		err = -EIO;
1454 		goto out_err;
1455 	}
1456 
1457 	if (!f2fs_inode_chksum_verify(sbi, page)) {
1458 		err = -EFSBADCRC;
1459 		goto out_err;
1460 	}
1461 page_hit:
1462 	if (likely(nid == nid_of_node(page)))
1463 		return page;
1464 
1465 	f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1466 			  nid, nid_of_node(page), ino_of_node(page),
1467 			  ofs_of_node(page), cpver_of_node(page),
1468 			  next_blkaddr_of_node(page));
1469 	set_sbi_flag(sbi, SBI_NEED_FSCK);
1470 	err = -EINVAL;
1471 out_err:
1472 	ClearPageUptodate(page);
1473 out_put_err:
1474 	/* ENOENT comes from read_node_page which is not an error. */
1475 	if (err != -ENOENT)
1476 		f2fs_handle_page_eio(sbi, page->index, NODE);
1477 	f2fs_put_page(page, 1);
1478 	return ERR_PTR(err);
1479 }
1480 
1481 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1482 {
1483 	return __get_node_page(sbi, nid, NULL, 0);
1484 }
1485 
1486 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1487 {
1488 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1489 	nid_t nid = get_nid(parent, start, false);
1490 
1491 	return __get_node_page(sbi, nid, parent, start);
1492 }
1493 
1494 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1495 {
1496 	struct inode *inode;
1497 	struct page *page;
1498 	int ret;
1499 
1500 	/* should flush inline_data before evict_inode */
1501 	inode = ilookup(sbi->sb, ino);
1502 	if (!inode)
1503 		return;
1504 
1505 	page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1506 					FGP_LOCK|FGP_NOWAIT, 0);
1507 	if (!page)
1508 		goto iput_out;
1509 
1510 	if (!PageUptodate(page))
1511 		goto page_out;
1512 
1513 	if (!PageDirty(page))
1514 		goto page_out;
1515 
1516 	if (!clear_page_dirty_for_io(page))
1517 		goto page_out;
1518 
1519 	ret = f2fs_write_inline_data(inode, page);
1520 	inode_dec_dirty_pages(inode);
1521 	f2fs_remove_dirty_inode(inode);
1522 	if (ret)
1523 		set_page_dirty(page);
1524 page_out:
1525 	f2fs_put_page(page, 1);
1526 iput_out:
1527 	iput(inode);
1528 }
1529 
1530 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1531 {
1532 	pgoff_t index;
1533 	struct folio_batch fbatch;
1534 	struct page *last_page = NULL;
1535 	int nr_folios;
1536 
1537 	folio_batch_init(&fbatch);
1538 	index = 0;
1539 
1540 	while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1541 					(pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1542 					&fbatch))) {
1543 		int i;
1544 
1545 		for (i = 0; i < nr_folios; i++) {
1546 			struct page *page = &fbatch.folios[i]->page;
1547 
1548 			if (unlikely(f2fs_cp_error(sbi))) {
1549 				f2fs_put_page(last_page, 0);
1550 				folio_batch_release(&fbatch);
1551 				return ERR_PTR(-EIO);
1552 			}
1553 
1554 			if (!IS_DNODE(page) || !is_cold_node(page))
1555 				continue;
1556 			if (ino_of_node(page) != ino)
1557 				continue;
1558 
1559 			lock_page(page);
1560 
1561 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1562 continue_unlock:
1563 				unlock_page(page);
1564 				continue;
1565 			}
1566 			if (ino_of_node(page) != ino)
1567 				goto continue_unlock;
1568 
1569 			if (!PageDirty(page)) {
1570 				/* someone wrote it for us */
1571 				goto continue_unlock;
1572 			}
1573 
1574 			if (last_page)
1575 				f2fs_put_page(last_page, 0);
1576 
1577 			get_page(page);
1578 			last_page = page;
1579 			unlock_page(page);
1580 		}
1581 		folio_batch_release(&fbatch);
1582 		cond_resched();
1583 	}
1584 	return last_page;
1585 }
1586 
1587 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1588 				struct writeback_control *wbc, bool do_balance,
1589 				enum iostat_type io_type, unsigned int *seq_id)
1590 {
1591 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1592 	nid_t nid;
1593 	struct node_info ni;
1594 	struct f2fs_io_info fio = {
1595 		.sbi = sbi,
1596 		.ino = ino_of_node(page),
1597 		.type = NODE,
1598 		.op = REQ_OP_WRITE,
1599 		.op_flags = wbc_to_write_flags(wbc),
1600 		.page = page,
1601 		.encrypted_page = NULL,
1602 		.submitted = 0,
1603 		.io_type = io_type,
1604 		.io_wbc = wbc,
1605 	};
1606 	unsigned int seq;
1607 
1608 	trace_f2fs_writepage(page, NODE);
1609 
1610 	if (unlikely(f2fs_cp_error(sbi))) {
1611 		/* keep node pages in remount-ro mode */
1612 		if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY)
1613 			goto redirty_out;
1614 		ClearPageUptodate(page);
1615 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1616 		unlock_page(page);
1617 		return 0;
1618 	}
1619 
1620 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1621 		goto redirty_out;
1622 
1623 	if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1624 			wbc->sync_mode == WB_SYNC_NONE &&
1625 			IS_DNODE(page) && is_cold_node(page))
1626 		goto redirty_out;
1627 
1628 	/* get old block addr of this node page */
1629 	nid = nid_of_node(page);
1630 	f2fs_bug_on(sbi, page->index != nid);
1631 
1632 	if (f2fs_get_node_info(sbi, nid, &ni, !do_balance))
1633 		goto redirty_out;
1634 
1635 	if (wbc->for_reclaim) {
1636 		if (!f2fs_down_read_trylock(&sbi->node_write))
1637 			goto redirty_out;
1638 	} else {
1639 		f2fs_down_read(&sbi->node_write);
1640 	}
1641 
1642 	/* This page is already truncated */
1643 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1644 		ClearPageUptodate(page);
1645 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1646 		f2fs_up_read(&sbi->node_write);
1647 		unlock_page(page);
1648 		return 0;
1649 	}
1650 
1651 	if (__is_valid_data_blkaddr(ni.blk_addr) &&
1652 		!f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1653 					DATA_GENERIC_ENHANCE)) {
1654 		f2fs_up_read(&sbi->node_write);
1655 		goto redirty_out;
1656 	}
1657 
1658 	if (atomic && !test_opt(sbi, NOBARRIER) && !f2fs_sb_has_blkzoned(sbi))
1659 		fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1660 
1661 	/* should add to global list before clearing PAGECACHE status */
1662 	if (f2fs_in_warm_node_list(sbi, page)) {
1663 		seq = f2fs_add_fsync_node_entry(sbi, page);
1664 		if (seq_id)
1665 			*seq_id = seq;
1666 	}
1667 
1668 	set_page_writeback(page);
1669 
1670 	fio.old_blkaddr = ni.blk_addr;
1671 	f2fs_do_write_node_page(nid, &fio);
1672 	set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1673 	dec_page_count(sbi, F2FS_DIRTY_NODES);
1674 	f2fs_up_read(&sbi->node_write);
1675 
1676 	if (wbc->for_reclaim) {
1677 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1678 		submitted = NULL;
1679 	}
1680 
1681 	unlock_page(page);
1682 
1683 	if (unlikely(f2fs_cp_error(sbi))) {
1684 		f2fs_submit_merged_write(sbi, NODE);
1685 		submitted = NULL;
1686 	}
1687 	if (submitted)
1688 		*submitted = fio.submitted;
1689 
1690 	if (do_balance)
1691 		f2fs_balance_fs(sbi, false);
1692 	return 0;
1693 
1694 redirty_out:
1695 	redirty_page_for_writepage(wbc, page);
1696 	return AOP_WRITEPAGE_ACTIVATE;
1697 }
1698 
1699 int f2fs_move_node_page(struct page *node_page, int gc_type)
1700 {
1701 	int err = 0;
1702 
1703 	if (gc_type == FG_GC) {
1704 		struct writeback_control wbc = {
1705 			.sync_mode = WB_SYNC_ALL,
1706 			.nr_to_write = 1,
1707 			.for_reclaim = 0,
1708 		};
1709 
1710 		f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1711 
1712 		set_page_dirty(node_page);
1713 
1714 		if (!clear_page_dirty_for_io(node_page)) {
1715 			err = -EAGAIN;
1716 			goto out_page;
1717 		}
1718 
1719 		if (__write_node_page(node_page, false, NULL,
1720 					&wbc, false, FS_GC_NODE_IO, NULL)) {
1721 			err = -EAGAIN;
1722 			unlock_page(node_page);
1723 		}
1724 		goto release_page;
1725 	} else {
1726 		/* set page dirty and write it */
1727 		if (!PageWriteback(node_page))
1728 			set_page_dirty(node_page);
1729 	}
1730 out_page:
1731 	unlock_page(node_page);
1732 release_page:
1733 	f2fs_put_page(node_page, 0);
1734 	return err;
1735 }
1736 
1737 static int f2fs_write_node_page(struct page *page,
1738 				struct writeback_control *wbc)
1739 {
1740 	return __write_node_page(page, false, NULL, wbc, false,
1741 						FS_NODE_IO, NULL);
1742 }
1743 
1744 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1745 			struct writeback_control *wbc, bool atomic,
1746 			unsigned int *seq_id)
1747 {
1748 	pgoff_t index;
1749 	struct folio_batch fbatch;
1750 	int ret = 0;
1751 	struct page *last_page = NULL;
1752 	bool marked = false;
1753 	nid_t ino = inode->i_ino;
1754 	int nr_folios;
1755 	int nwritten = 0;
1756 
1757 	if (atomic) {
1758 		last_page = last_fsync_dnode(sbi, ino);
1759 		if (IS_ERR_OR_NULL(last_page))
1760 			return PTR_ERR_OR_ZERO(last_page);
1761 	}
1762 retry:
1763 	folio_batch_init(&fbatch);
1764 	index = 0;
1765 
1766 	while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1767 					(pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1768 					&fbatch))) {
1769 		int i;
1770 
1771 		for (i = 0; i < nr_folios; i++) {
1772 			struct page *page = &fbatch.folios[i]->page;
1773 			bool submitted = false;
1774 
1775 			if (unlikely(f2fs_cp_error(sbi))) {
1776 				f2fs_put_page(last_page, 0);
1777 				folio_batch_release(&fbatch);
1778 				ret = -EIO;
1779 				goto out;
1780 			}
1781 
1782 			if (!IS_DNODE(page) || !is_cold_node(page))
1783 				continue;
1784 			if (ino_of_node(page) != ino)
1785 				continue;
1786 
1787 			lock_page(page);
1788 
1789 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1790 continue_unlock:
1791 				unlock_page(page);
1792 				continue;
1793 			}
1794 			if (ino_of_node(page) != ino)
1795 				goto continue_unlock;
1796 
1797 			if (!PageDirty(page) && page != last_page) {
1798 				/* someone wrote it for us */
1799 				goto continue_unlock;
1800 			}
1801 
1802 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1803 
1804 			set_fsync_mark(page, 0);
1805 			set_dentry_mark(page, 0);
1806 
1807 			if (!atomic || page == last_page) {
1808 				set_fsync_mark(page, 1);
1809 				percpu_counter_inc(&sbi->rf_node_block_count);
1810 				if (IS_INODE(page)) {
1811 					if (is_inode_flag_set(inode,
1812 								FI_DIRTY_INODE))
1813 						f2fs_update_inode(inode, page);
1814 					set_dentry_mark(page,
1815 						f2fs_need_dentry_mark(sbi, ino));
1816 				}
1817 				/* may be written by other thread */
1818 				if (!PageDirty(page))
1819 					set_page_dirty(page);
1820 			}
1821 
1822 			if (!clear_page_dirty_for_io(page))
1823 				goto continue_unlock;
1824 
1825 			ret = __write_node_page(page, atomic &&
1826 						page == last_page,
1827 						&submitted, wbc, true,
1828 						FS_NODE_IO, seq_id);
1829 			if (ret) {
1830 				unlock_page(page);
1831 				f2fs_put_page(last_page, 0);
1832 				break;
1833 			} else if (submitted) {
1834 				nwritten++;
1835 			}
1836 
1837 			if (page == last_page) {
1838 				f2fs_put_page(page, 0);
1839 				marked = true;
1840 				break;
1841 			}
1842 		}
1843 		folio_batch_release(&fbatch);
1844 		cond_resched();
1845 
1846 		if (ret || marked)
1847 			break;
1848 	}
1849 	if (!ret && atomic && !marked) {
1850 		f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1851 			   ino, last_page->index);
1852 		lock_page(last_page);
1853 		f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1854 		set_page_dirty(last_page);
1855 		unlock_page(last_page);
1856 		goto retry;
1857 	}
1858 out:
1859 	if (nwritten)
1860 		f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1861 	return ret ? -EIO : 0;
1862 }
1863 
1864 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1865 {
1866 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1867 	bool clean;
1868 
1869 	if (inode->i_ino != ino)
1870 		return 0;
1871 
1872 	if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1873 		return 0;
1874 
1875 	spin_lock(&sbi->inode_lock[DIRTY_META]);
1876 	clean = list_empty(&F2FS_I(inode)->gdirty_list);
1877 	spin_unlock(&sbi->inode_lock[DIRTY_META]);
1878 
1879 	if (clean)
1880 		return 0;
1881 
1882 	inode = igrab(inode);
1883 	if (!inode)
1884 		return 0;
1885 	return 1;
1886 }
1887 
1888 static bool flush_dirty_inode(struct page *page)
1889 {
1890 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1891 	struct inode *inode;
1892 	nid_t ino = ino_of_node(page);
1893 
1894 	inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1895 	if (!inode)
1896 		return false;
1897 
1898 	f2fs_update_inode(inode, page);
1899 	unlock_page(page);
1900 
1901 	iput(inode);
1902 	return true;
1903 }
1904 
1905 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1906 {
1907 	pgoff_t index = 0;
1908 	struct folio_batch fbatch;
1909 	int nr_folios;
1910 
1911 	folio_batch_init(&fbatch);
1912 
1913 	while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1914 					(pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1915 					&fbatch))) {
1916 		int i;
1917 
1918 		for (i = 0; i < nr_folios; i++) {
1919 			struct page *page = &fbatch.folios[i]->page;
1920 
1921 			if (!IS_DNODE(page))
1922 				continue;
1923 
1924 			lock_page(page);
1925 
1926 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1927 continue_unlock:
1928 				unlock_page(page);
1929 				continue;
1930 			}
1931 
1932 			if (!PageDirty(page)) {
1933 				/* someone wrote it for us */
1934 				goto continue_unlock;
1935 			}
1936 
1937 			/* flush inline_data, if it's async context. */
1938 			if (page_private_inline(page)) {
1939 				clear_page_private_inline(page);
1940 				unlock_page(page);
1941 				flush_inline_data(sbi, ino_of_node(page));
1942 				continue;
1943 			}
1944 			unlock_page(page);
1945 		}
1946 		folio_batch_release(&fbatch);
1947 		cond_resched();
1948 	}
1949 }
1950 
1951 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1952 				struct writeback_control *wbc,
1953 				bool do_balance, enum iostat_type io_type)
1954 {
1955 	pgoff_t index;
1956 	struct folio_batch fbatch;
1957 	int step = 0;
1958 	int nwritten = 0;
1959 	int ret = 0;
1960 	int nr_folios, done = 0;
1961 
1962 	folio_batch_init(&fbatch);
1963 
1964 next_step:
1965 	index = 0;
1966 
1967 	while (!done && (nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi),
1968 				&index, (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1969 				&fbatch))) {
1970 		int i;
1971 
1972 		for (i = 0; i < nr_folios; i++) {
1973 			struct page *page = &fbatch.folios[i]->page;
1974 			bool submitted = false;
1975 
1976 			/* give a priority to WB_SYNC threads */
1977 			if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1978 					wbc->sync_mode == WB_SYNC_NONE) {
1979 				done = 1;
1980 				break;
1981 			}
1982 
1983 			/*
1984 			 * flushing sequence with step:
1985 			 * 0. indirect nodes
1986 			 * 1. dentry dnodes
1987 			 * 2. file dnodes
1988 			 */
1989 			if (step == 0 && IS_DNODE(page))
1990 				continue;
1991 			if (step == 1 && (!IS_DNODE(page) ||
1992 						is_cold_node(page)))
1993 				continue;
1994 			if (step == 2 && (!IS_DNODE(page) ||
1995 						!is_cold_node(page)))
1996 				continue;
1997 lock_node:
1998 			if (wbc->sync_mode == WB_SYNC_ALL)
1999 				lock_page(page);
2000 			else if (!trylock_page(page))
2001 				continue;
2002 
2003 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
2004 continue_unlock:
2005 				unlock_page(page);
2006 				continue;
2007 			}
2008 
2009 			if (!PageDirty(page)) {
2010 				/* someone wrote it for us */
2011 				goto continue_unlock;
2012 			}
2013 
2014 			/* flush inline_data/inode, if it's async context. */
2015 			if (!do_balance)
2016 				goto write_node;
2017 
2018 			/* flush inline_data */
2019 			if (page_private_inline(page)) {
2020 				clear_page_private_inline(page);
2021 				unlock_page(page);
2022 				flush_inline_data(sbi, ino_of_node(page));
2023 				goto lock_node;
2024 			}
2025 
2026 			/* flush dirty inode */
2027 			if (IS_INODE(page) && flush_dirty_inode(page))
2028 				goto lock_node;
2029 write_node:
2030 			f2fs_wait_on_page_writeback(page, NODE, true, true);
2031 
2032 			if (!clear_page_dirty_for_io(page))
2033 				goto continue_unlock;
2034 
2035 			set_fsync_mark(page, 0);
2036 			set_dentry_mark(page, 0);
2037 
2038 			ret = __write_node_page(page, false, &submitted,
2039 						wbc, do_balance, io_type, NULL);
2040 			if (ret)
2041 				unlock_page(page);
2042 			else if (submitted)
2043 				nwritten++;
2044 
2045 			if (--wbc->nr_to_write == 0)
2046 				break;
2047 		}
2048 		folio_batch_release(&fbatch);
2049 		cond_resched();
2050 
2051 		if (wbc->nr_to_write == 0) {
2052 			step = 2;
2053 			break;
2054 		}
2055 	}
2056 
2057 	if (step < 2) {
2058 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2059 				wbc->sync_mode == WB_SYNC_NONE && step == 1)
2060 			goto out;
2061 		step++;
2062 		goto next_step;
2063 	}
2064 out:
2065 	if (nwritten)
2066 		f2fs_submit_merged_write(sbi, NODE);
2067 
2068 	if (unlikely(f2fs_cp_error(sbi)))
2069 		return -EIO;
2070 	return ret;
2071 }
2072 
2073 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2074 						unsigned int seq_id)
2075 {
2076 	struct fsync_node_entry *fn;
2077 	struct page *page;
2078 	struct list_head *head = &sbi->fsync_node_list;
2079 	unsigned long flags;
2080 	unsigned int cur_seq_id = 0;
2081 
2082 	while (seq_id && cur_seq_id < seq_id) {
2083 		spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2084 		if (list_empty(head)) {
2085 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2086 			break;
2087 		}
2088 		fn = list_first_entry(head, struct fsync_node_entry, list);
2089 		if (fn->seq_id > seq_id) {
2090 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2091 			break;
2092 		}
2093 		cur_seq_id = fn->seq_id;
2094 		page = fn->page;
2095 		get_page(page);
2096 		spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2097 
2098 		f2fs_wait_on_page_writeback(page, NODE, true, false);
2099 
2100 		put_page(page);
2101 	}
2102 
2103 	return filemap_check_errors(NODE_MAPPING(sbi));
2104 }
2105 
2106 static int f2fs_write_node_pages(struct address_space *mapping,
2107 			    struct writeback_control *wbc)
2108 {
2109 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2110 	struct blk_plug plug;
2111 	long diff;
2112 
2113 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2114 		goto skip_write;
2115 
2116 	/* balancing f2fs's metadata in background */
2117 	f2fs_balance_fs_bg(sbi, true);
2118 
2119 	/* collect a number of dirty node pages and write together */
2120 	if (wbc->sync_mode != WB_SYNC_ALL &&
2121 			get_pages(sbi, F2FS_DIRTY_NODES) <
2122 					nr_pages_to_skip(sbi, NODE))
2123 		goto skip_write;
2124 
2125 	if (wbc->sync_mode == WB_SYNC_ALL)
2126 		atomic_inc(&sbi->wb_sync_req[NODE]);
2127 	else if (atomic_read(&sbi->wb_sync_req[NODE])) {
2128 		/* to avoid potential deadlock */
2129 		if (current->plug)
2130 			blk_finish_plug(current->plug);
2131 		goto skip_write;
2132 	}
2133 
2134 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2135 
2136 	diff = nr_pages_to_write(sbi, NODE, wbc);
2137 	blk_start_plug(&plug);
2138 	f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2139 	blk_finish_plug(&plug);
2140 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2141 
2142 	if (wbc->sync_mode == WB_SYNC_ALL)
2143 		atomic_dec(&sbi->wb_sync_req[NODE]);
2144 	return 0;
2145 
2146 skip_write:
2147 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2148 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2149 	return 0;
2150 }
2151 
2152 static bool f2fs_dirty_node_folio(struct address_space *mapping,
2153 		struct folio *folio)
2154 {
2155 	trace_f2fs_set_page_dirty(&folio->page, NODE);
2156 
2157 	if (!folio_test_uptodate(folio))
2158 		folio_mark_uptodate(folio);
2159 #ifdef CONFIG_F2FS_CHECK_FS
2160 	if (IS_INODE(&folio->page))
2161 		f2fs_inode_chksum_set(F2FS_M_SB(mapping), &folio->page);
2162 #endif
2163 	if (filemap_dirty_folio(mapping, folio)) {
2164 		inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
2165 		set_page_private_reference(&folio->page);
2166 		return true;
2167 	}
2168 	return false;
2169 }
2170 
2171 /*
2172  * Structure of the f2fs node operations
2173  */
2174 const struct address_space_operations f2fs_node_aops = {
2175 	.writepage	= f2fs_write_node_page,
2176 	.writepages	= f2fs_write_node_pages,
2177 	.dirty_folio	= f2fs_dirty_node_folio,
2178 	.invalidate_folio = f2fs_invalidate_folio,
2179 	.release_folio	= f2fs_release_folio,
2180 	.migrate_folio	= filemap_migrate_folio,
2181 };
2182 
2183 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2184 						nid_t n)
2185 {
2186 	return radix_tree_lookup(&nm_i->free_nid_root, n);
2187 }
2188 
2189 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2190 				struct free_nid *i)
2191 {
2192 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2193 	int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2194 
2195 	if (err)
2196 		return err;
2197 
2198 	nm_i->nid_cnt[FREE_NID]++;
2199 	list_add_tail(&i->list, &nm_i->free_nid_list);
2200 	return 0;
2201 }
2202 
2203 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2204 			struct free_nid *i, enum nid_state state)
2205 {
2206 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2207 
2208 	f2fs_bug_on(sbi, state != i->state);
2209 	nm_i->nid_cnt[state]--;
2210 	if (state == FREE_NID)
2211 		list_del(&i->list);
2212 	radix_tree_delete(&nm_i->free_nid_root, i->nid);
2213 }
2214 
2215 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2216 			enum nid_state org_state, enum nid_state dst_state)
2217 {
2218 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2219 
2220 	f2fs_bug_on(sbi, org_state != i->state);
2221 	i->state = dst_state;
2222 	nm_i->nid_cnt[org_state]--;
2223 	nm_i->nid_cnt[dst_state]++;
2224 
2225 	switch (dst_state) {
2226 	case PREALLOC_NID:
2227 		list_del(&i->list);
2228 		break;
2229 	case FREE_NID:
2230 		list_add_tail(&i->list, &nm_i->free_nid_list);
2231 		break;
2232 	default:
2233 		BUG_ON(1);
2234 	}
2235 }
2236 
2237 bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi)
2238 {
2239 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2240 	unsigned int i;
2241 	bool ret = true;
2242 
2243 	f2fs_down_read(&nm_i->nat_tree_lock);
2244 	for (i = 0; i < nm_i->nat_blocks; i++) {
2245 		if (!test_bit_le(i, nm_i->nat_block_bitmap)) {
2246 			ret = false;
2247 			break;
2248 		}
2249 	}
2250 	f2fs_up_read(&nm_i->nat_tree_lock);
2251 
2252 	return ret;
2253 }
2254 
2255 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2256 							bool set, bool build)
2257 {
2258 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2259 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2260 	unsigned int nid_ofs = nid - START_NID(nid);
2261 
2262 	if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2263 		return;
2264 
2265 	if (set) {
2266 		if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2267 			return;
2268 		__set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2269 		nm_i->free_nid_count[nat_ofs]++;
2270 	} else {
2271 		if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2272 			return;
2273 		__clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2274 		if (!build)
2275 			nm_i->free_nid_count[nat_ofs]--;
2276 	}
2277 }
2278 
2279 /* return if the nid is recognized as free */
2280 static bool add_free_nid(struct f2fs_sb_info *sbi,
2281 				nid_t nid, bool build, bool update)
2282 {
2283 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2284 	struct free_nid *i, *e;
2285 	struct nat_entry *ne;
2286 	int err = -EINVAL;
2287 	bool ret = false;
2288 
2289 	/* 0 nid should not be used */
2290 	if (unlikely(nid == 0))
2291 		return false;
2292 
2293 	if (unlikely(f2fs_check_nid_range(sbi, nid)))
2294 		return false;
2295 
2296 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL);
2297 	i->nid = nid;
2298 	i->state = FREE_NID;
2299 
2300 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2301 
2302 	spin_lock(&nm_i->nid_list_lock);
2303 
2304 	if (build) {
2305 		/*
2306 		 *   Thread A             Thread B
2307 		 *  - f2fs_create
2308 		 *   - f2fs_new_inode
2309 		 *    - f2fs_alloc_nid
2310 		 *     - __insert_nid_to_list(PREALLOC_NID)
2311 		 *                     - f2fs_balance_fs_bg
2312 		 *                      - f2fs_build_free_nids
2313 		 *                       - __f2fs_build_free_nids
2314 		 *                        - scan_nat_page
2315 		 *                         - add_free_nid
2316 		 *                          - __lookup_nat_cache
2317 		 *  - f2fs_add_link
2318 		 *   - f2fs_init_inode_metadata
2319 		 *    - f2fs_new_inode_page
2320 		 *     - f2fs_new_node_page
2321 		 *      - set_node_addr
2322 		 *  - f2fs_alloc_nid_done
2323 		 *   - __remove_nid_from_list(PREALLOC_NID)
2324 		 *                         - __insert_nid_to_list(FREE_NID)
2325 		 */
2326 		ne = __lookup_nat_cache(nm_i, nid);
2327 		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2328 				nat_get_blkaddr(ne) != NULL_ADDR))
2329 			goto err_out;
2330 
2331 		e = __lookup_free_nid_list(nm_i, nid);
2332 		if (e) {
2333 			if (e->state == FREE_NID)
2334 				ret = true;
2335 			goto err_out;
2336 		}
2337 	}
2338 	ret = true;
2339 	err = __insert_free_nid(sbi, i);
2340 err_out:
2341 	if (update) {
2342 		update_free_nid_bitmap(sbi, nid, ret, build);
2343 		if (!build)
2344 			nm_i->available_nids++;
2345 	}
2346 	spin_unlock(&nm_i->nid_list_lock);
2347 	radix_tree_preload_end();
2348 
2349 	if (err)
2350 		kmem_cache_free(free_nid_slab, i);
2351 	return ret;
2352 }
2353 
2354 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2355 {
2356 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2357 	struct free_nid *i;
2358 	bool need_free = false;
2359 
2360 	spin_lock(&nm_i->nid_list_lock);
2361 	i = __lookup_free_nid_list(nm_i, nid);
2362 	if (i && i->state == FREE_NID) {
2363 		__remove_free_nid(sbi, i, FREE_NID);
2364 		need_free = true;
2365 	}
2366 	spin_unlock(&nm_i->nid_list_lock);
2367 
2368 	if (need_free)
2369 		kmem_cache_free(free_nid_slab, i);
2370 }
2371 
2372 static int scan_nat_page(struct f2fs_sb_info *sbi,
2373 			struct page *nat_page, nid_t start_nid)
2374 {
2375 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2376 	struct f2fs_nat_block *nat_blk = page_address(nat_page);
2377 	block_t blk_addr;
2378 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2379 	int i;
2380 
2381 	__set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2382 
2383 	i = start_nid % NAT_ENTRY_PER_BLOCK;
2384 
2385 	for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2386 		if (unlikely(start_nid >= nm_i->max_nid))
2387 			break;
2388 
2389 		blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2390 
2391 		if (blk_addr == NEW_ADDR)
2392 			return -EINVAL;
2393 
2394 		if (blk_addr == NULL_ADDR) {
2395 			add_free_nid(sbi, start_nid, true, true);
2396 		} else {
2397 			spin_lock(&NM_I(sbi)->nid_list_lock);
2398 			update_free_nid_bitmap(sbi, start_nid, false, true);
2399 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2400 		}
2401 	}
2402 
2403 	return 0;
2404 }
2405 
2406 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2407 {
2408 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2409 	struct f2fs_journal *journal = curseg->journal;
2410 	int i;
2411 
2412 	down_read(&curseg->journal_rwsem);
2413 	for (i = 0; i < nats_in_cursum(journal); i++) {
2414 		block_t addr;
2415 		nid_t nid;
2416 
2417 		addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2418 		nid = le32_to_cpu(nid_in_journal(journal, i));
2419 		if (addr == NULL_ADDR)
2420 			add_free_nid(sbi, nid, true, false);
2421 		else
2422 			remove_free_nid(sbi, nid);
2423 	}
2424 	up_read(&curseg->journal_rwsem);
2425 }
2426 
2427 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2428 {
2429 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2430 	unsigned int i, idx;
2431 	nid_t nid;
2432 
2433 	f2fs_down_read(&nm_i->nat_tree_lock);
2434 
2435 	for (i = 0; i < nm_i->nat_blocks; i++) {
2436 		if (!test_bit_le(i, nm_i->nat_block_bitmap))
2437 			continue;
2438 		if (!nm_i->free_nid_count[i])
2439 			continue;
2440 		for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2441 			idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2442 						NAT_ENTRY_PER_BLOCK, idx);
2443 			if (idx >= NAT_ENTRY_PER_BLOCK)
2444 				break;
2445 
2446 			nid = i * NAT_ENTRY_PER_BLOCK + idx;
2447 			add_free_nid(sbi, nid, true, false);
2448 
2449 			if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2450 				goto out;
2451 		}
2452 	}
2453 out:
2454 	scan_curseg_cache(sbi);
2455 
2456 	f2fs_up_read(&nm_i->nat_tree_lock);
2457 }
2458 
2459 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2460 						bool sync, bool mount)
2461 {
2462 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2463 	int i = 0, ret;
2464 	nid_t nid = nm_i->next_scan_nid;
2465 
2466 	if (unlikely(nid >= nm_i->max_nid))
2467 		nid = 0;
2468 
2469 	if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2470 		nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2471 
2472 	/* Enough entries */
2473 	if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2474 		return 0;
2475 
2476 	if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2477 		return 0;
2478 
2479 	if (!mount) {
2480 		/* try to find free nids in free_nid_bitmap */
2481 		scan_free_nid_bits(sbi);
2482 
2483 		if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2484 			return 0;
2485 	}
2486 
2487 	/* readahead nat pages to be scanned */
2488 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2489 							META_NAT, true);
2490 
2491 	f2fs_down_read(&nm_i->nat_tree_lock);
2492 
2493 	while (1) {
2494 		if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2495 						nm_i->nat_block_bitmap)) {
2496 			struct page *page = get_current_nat_page(sbi, nid);
2497 
2498 			if (IS_ERR(page)) {
2499 				ret = PTR_ERR(page);
2500 			} else {
2501 				ret = scan_nat_page(sbi, page, nid);
2502 				f2fs_put_page(page, 1);
2503 			}
2504 
2505 			if (ret) {
2506 				f2fs_up_read(&nm_i->nat_tree_lock);
2507 				f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2508 				return ret;
2509 			}
2510 		}
2511 
2512 		nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2513 		if (unlikely(nid >= nm_i->max_nid))
2514 			nid = 0;
2515 
2516 		if (++i >= FREE_NID_PAGES)
2517 			break;
2518 	}
2519 
2520 	/* go to the next free nat pages to find free nids abundantly */
2521 	nm_i->next_scan_nid = nid;
2522 
2523 	/* find free nids from current sum_pages */
2524 	scan_curseg_cache(sbi);
2525 
2526 	f2fs_up_read(&nm_i->nat_tree_lock);
2527 
2528 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2529 					nm_i->ra_nid_pages, META_NAT, false);
2530 
2531 	return 0;
2532 }
2533 
2534 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2535 {
2536 	int ret;
2537 
2538 	mutex_lock(&NM_I(sbi)->build_lock);
2539 	ret = __f2fs_build_free_nids(sbi, sync, mount);
2540 	mutex_unlock(&NM_I(sbi)->build_lock);
2541 
2542 	return ret;
2543 }
2544 
2545 /*
2546  * If this function returns success, caller can obtain a new nid
2547  * from second parameter of this function.
2548  * The returned nid could be used ino as well as nid when inode is created.
2549  */
2550 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2551 {
2552 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2553 	struct free_nid *i = NULL;
2554 retry:
2555 	if (time_to_inject(sbi, FAULT_ALLOC_NID))
2556 		return false;
2557 
2558 	spin_lock(&nm_i->nid_list_lock);
2559 
2560 	if (unlikely(nm_i->available_nids == 0)) {
2561 		spin_unlock(&nm_i->nid_list_lock);
2562 		return false;
2563 	}
2564 
2565 	/* We should not use stale free nids created by f2fs_build_free_nids */
2566 	if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2567 		f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2568 		i = list_first_entry(&nm_i->free_nid_list,
2569 					struct free_nid, list);
2570 		*nid = i->nid;
2571 
2572 		__move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2573 		nm_i->available_nids--;
2574 
2575 		update_free_nid_bitmap(sbi, *nid, false, false);
2576 
2577 		spin_unlock(&nm_i->nid_list_lock);
2578 		return true;
2579 	}
2580 	spin_unlock(&nm_i->nid_list_lock);
2581 
2582 	/* Let's scan nat pages and its caches to get free nids */
2583 	if (!f2fs_build_free_nids(sbi, true, false))
2584 		goto retry;
2585 	return false;
2586 }
2587 
2588 /*
2589  * f2fs_alloc_nid() should be called prior to this function.
2590  */
2591 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2592 {
2593 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2594 	struct free_nid *i;
2595 
2596 	spin_lock(&nm_i->nid_list_lock);
2597 	i = __lookup_free_nid_list(nm_i, nid);
2598 	f2fs_bug_on(sbi, !i);
2599 	__remove_free_nid(sbi, i, PREALLOC_NID);
2600 	spin_unlock(&nm_i->nid_list_lock);
2601 
2602 	kmem_cache_free(free_nid_slab, i);
2603 }
2604 
2605 /*
2606  * f2fs_alloc_nid() should be called prior to this function.
2607  */
2608 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2609 {
2610 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2611 	struct free_nid *i;
2612 	bool need_free = false;
2613 
2614 	if (!nid)
2615 		return;
2616 
2617 	spin_lock(&nm_i->nid_list_lock);
2618 	i = __lookup_free_nid_list(nm_i, nid);
2619 	f2fs_bug_on(sbi, !i);
2620 
2621 	if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2622 		__remove_free_nid(sbi, i, PREALLOC_NID);
2623 		need_free = true;
2624 	} else {
2625 		__move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2626 	}
2627 
2628 	nm_i->available_nids++;
2629 
2630 	update_free_nid_bitmap(sbi, nid, true, false);
2631 
2632 	spin_unlock(&nm_i->nid_list_lock);
2633 
2634 	if (need_free)
2635 		kmem_cache_free(free_nid_slab, i);
2636 }
2637 
2638 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2639 {
2640 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2641 	int nr = nr_shrink;
2642 
2643 	if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2644 		return 0;
2645 
2646 	if (!mutex_trylock(&nm_i->build_lock))
2647 		return 0;
2648 
2649 	while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2650 		struct free_nid *i, *next;
2651 		unsigned int batch = SHRINK_NID_BATCH_SIZE;
2652 
2653 		spin_lock(&nm_i->nid_list_lock);
2654 		list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2655 			if (!nr_shrink || !batch ||
2656 				nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2657 				break;
2658 			__remove_free_nid(sbi, i, FREE_NID);
2659 			kmem_cache_free(free_nid_slab, i);
2660 			nr_shrink--;
2661 			batch--;
2662 		}
2663 		spin_unlock(&nm_i->nid_list_lock);
2664 	}
2665 
2666 	mutex_unlock(&nm_i->build_lock);
2667 
2668 	return nr - nr_shrink;
2669 }
2670 
2671 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2672 {
2673 	void *src_addr, *dst_addr;
2674 	size_t inline_size;
2675 	struct page *ipage;
2676 	struct f2fs_inode *ri;
2677 
2678 	ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2679 	if (IS_ERR(ipage))
2680 		return PTR_ERR(ipage);
2681 
2682 	ri = F2FS_INODE(page);
2683 	if (ri->i_inline & F2FS_INLINE_XATTR) {
2684 		if (!f2fs_has_inline_xattr(inode)) {
2685 			set_inode_flag(inode, FI_INLINE_XATTR);
2686 			stat_inc_inline_xattr(inode);
2687 		}
2688 	} else {
2689 		if (f2fs_has_inline_xattr(inode)) {
2690 			stat_dec_inline_xattr(inode);
2691 			clear_inode_flag(inode, FI_INLINE_XATTR);
2692 		}
2693 		goto update_inode;
2694 	}
2695 
2696 	dst_addr = inline_xattr_addr(inode, ipage);
2697 	src_addr = inline_xattr_addr(inode, page);
2698 	inline_size = inline_xattr_size(inode);
2699 
2700 	f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2701 	memcpy(dst_addr, src_addr, inline_size);
2702 update_inode:
2703 	f2fs_update_inode(inode, ipage);
2704 	f2fs_put_page(ipage, 1);
2705 	return 0;
2706 }
2707 
2708 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2709 {
2710 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2711 	nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2712 	nid_t new_xnid;
2713 	struct dnode_of_data dn;
2714 	struct node_info ni;
2715 	struct page *xpage;
2716 	int err;
2717 
2718 	if (!prev_xnid)
2719 		goto recover_xnid;
2720 
2721 	/* 1: invalidate the previous xattr nid */
2722 	err = f2fs_get_node_info(sbi, prev_xnid, &ni, false);
2723 	if (err)
2724 		return err;
2725 
2726 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
2727 	dec_valid_node_count(sbi, inode, false);
2728 	set_node_addr(sbi, &ni, NULL_ADDR, false);
2729 
2730 recover_xnid:
2731 	/* 2: update xattr nid in inode */
2732 	if (!f2fs_alloc_nid(sbi, &new_xnid))
2733 		return -ENOSPC;
2734 
2735 	set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2736 	xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2737 	if (IS_ERR(xpage)) {
2738 		f2fs_alloc_nid_failed(sbi, new_xnid);
2739 		return PTR_ERR(xpage);
2740 	}
2741 
2742 	f2fs_alloc_nid_done(sbi, new_xnid);
2743 	f2fs_update_inode_page(inode);
2744 
2745 	/* 3: update and set xattr node page dirty */
2746 	memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2747 
2748 	set_page_dirty(xpage);
2749 	f2fs_put_page(xpage, 1);
2750 
2751 	return 0;
2752 }
2753 
2754 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2755 {
2756 	struct f2fs_inode *src, *dst;
2757 	nid_t ino = ino_of_node(page);
2758 	struct node_info old_ni, new_ni;
2759 	struct page *ipage;
2760 	int err;
2761 
2762 	err = f2fs_get_node_info(sbi, ino, &old_ni, false);
2763 	if (err)
2764 		return err;
2765 
2766 	if (unlikely(old_ni.blk_addr != NULL_ADDR))
2767 		return -EINVAL;
2768 retry:
2769 	ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2770 	if (!ipage) {
2771 		memalloc_retry_wait(GFP_NOFS);
2772 		goto retry;
2773 	}
2774 
2775 	/* Should not use this inode from free nid list */
2776 	remove_free_nid(sbi, ino);
2777 
2778 	if (!PageUptodate(ipage))
2779 		SetPageUptodate(ipage);
2780 	fill_node_footer(ipage, ino, ino, 0, true);
2781 	set_cold_node(ipage, false);
2782 
2783 	src = F2FS_INODE(page);
2784 	dst = F2FS_INODE(ipage);
2785 
2786 	memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2787 	dst->i_size = 0;
2788 	dst->i_blocks = cpu_to_le64(1);
2789 	dst->i_links = cpu_to_le32(1);
2790 	dst->i_xattr_nid = 0;
2791 	dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2792 	if (dst->i_inline & F2FS_EXTRA_ATTR) {
2793 		dst->i_extra_isize = src->i_extra_isize;
2794 
2795 		if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2796 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2797 							i_inline_xattr_size))
2798 			dst->i_inline_xattr_size = src->i_inline_xattr_size;
2799 
2800 		if (f2fs_sb_has_project_quota(sbi) &&
2801 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2802 								i_projid))
2803 			dst->i_projid = src->i_projid;
2804 
2805 		if (f2fs_sb_has_inode_crtime(sbi) &&
2806 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2807 							i_crtime_nsec)) {
2808 			dst->i_crtime = src->i_crtime;
2809 			dst->i_crtime_nsec = src->i_crtime_nsec;
2810 		}
2811 	}
2812 
2813 	new_ni = old_ni;
2814 	new_ni.ino = ino;
2815 
2816 	if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2817 		WARN_ON(1);
2818 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2819 	inc_valid_inode_count(sbi);
2820 	set_page_dirty(ipage);
2821 	f2fs_put_page(ipage, 1);
2822 	return 0;
2823 }
2824 
2825 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2826 			unsigned int segno, struct f2fs_summary_block *sum)
2827 {
2828 	struct f2fs_node *rn;
2829 	struct f2fs_summary *sum_entry;
2830 	block_t addr;
2831 	int i, idx, last_offset, nrpages;
2832 
2833 	/* scan the node segment */
2834 	last_offset = sbi->blocks_per_seg;
2835 	addr = START_BLOCK(sbi, segno);
2836 	sum_entry = &sum->entries[0];
2837 
2838 	for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2839 		nrpages = bio_max_segs(last_offset - i);
2840 
2841 		/* readahead node pages */
2842 		f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2843 
2844 		for (idx = addr; idx < addr + nrpages; idx++) {
2845 			struct page *page = f2fs_get_tmp_page(sbi, idx);
2846 
2847 			if (IS_ERR(page))
2848 				return PTR_ERR(page);
2849 
2850 			rn = F2FS_NODE(page);
2851 			sum_entry->nid = rn->footer.nid;
2852 			sum_entry->version = 0;
2853 			sum_entry->ofs_in_node = 0;
2854 			sum_entry++;
2855 			f2fs_put_page(page, 1);
2856 		}
2857 
2858 		invalidate_mapping_pages(META_MAPPING(sbi), addr,
2859 							addr + nrpages);
2860 	}
2861 	return 0;
2862 }
2863 
2864 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2865 {
2866 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2867 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2868 	struct f2fs_journal *journal = curseg->journal;
2869 	int i;
2870 
2871 	down_write(&curseg->journal_rwsem);
2872 	for (i = 0; i < nats_in_cursum(journal); i++) {
2873 		struct nat_entry *ne;
2874 		struct f2fs_nat_entry raw_ne;
2875 		nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2876 
2877 		if (f2fs_check_nid_range(sbi, nid))
2878 			continue;
2879 
2880 		raw_ne = nat_in_journal(journal, i);
2881 
2882 		ne = __lookup_nat_cache(nm_i, nid);
2883 		if (!ne) {
2884 			ne = __alloc_nat_entry(sbi, nid, true);
2885 			__init_nat_entry(nm_i, ne, &raw_ne, true);
2886 		}
2887 
2888 		/*
2889 		 * if a free nat in journal has not been used after last
2890 		 * checkpoint, we should remove it from available nids,
2891 		 * since later we will add it again.
2892 		 */
2893 		if (!get_nat_flag(ne, IS_DIRTY) &&
2894 				le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2895 			spin_lock(&nm_i->nid_list_lock);
2896 			nm_i->available_nids--;
2897 			spin_unlock(&nm_i->nid_list_lock);
2898 		}
2899 
2900 		__set_nat_cache_dirty(nm_i, ne);
2901 	}
2902 	update_nats_in_cursum(journal, -i);
2903 	up_write(&curseg->journal_rwsem);
2904 }
2905 
2906 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2907 						struct list_head *head, int max)
2908 {
2909 	struct nat_entry_set *cur;
2910 
2911 	if (nes->entry_cnt >= max)
2912 		goto add_out;
2913 
2914 	list_for_each_entry(cur, head, set_list) {
2915 		if (cur->entry_cnt >= nes->entry_cnt) {
2916 			list_add(&nes->set_list, cur->set_list.prev);
2917 			return;
2918 		}
2919 	}
2920 add_out:
2921 	list_add_tail(&nes->set_list, head);
2922 }
2923 
2924 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs,
2925 							unsigned int valid)
2926 {
2927 	if (valid == 0) {
2928 		__set_bit_le(nat_ofs, nm_i->empty_nat_bits);
2929 		__clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2930 		return;
2931 	}
2932 
2933 	__clear_bit_le(nat_ofs, nm_i->empty_nat_bits);
2934 	if (valid == NAT_ENTRY_PER_BLOCK)
2935 		__set_bit_le(nat_ofs, nm_i->full_nat_bits);
2936 	else
2937 		__clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2938 }
2939 
2940 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2941 						struct page *page)
2942 {
2943 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2944 	unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2945 	struct f2fs_nat_block *nat_blk = page_address(page);
2946 	int valid = 0;
2947 	int i = 0;
2948 
2949 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
2950 		return;
2951 
2952 	if (nat_index == 0) {
2953 		valid = 1;
2954 		i = 1;
2955 	}
2956 	for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2957 		if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2958 			valid++;
2959 	}
2960 
2961 	__update_nat_bits(nm_i, nat_index, valid);
2962 }
2963 
2964 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi)
2965 {
2966 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2967 	unsigned int nat_ofs;
2968 
2969 	f2fs_down_read(&nm_i->nat_tree_lock);
2970 
2971 	for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) {
2972 		unsigned int valid = 0, nid_ofs = 0;
2973 
2974 		/* handle nid zero due to it should never be used */
2975 		if (unlikely(nat_ofs == 0)) {
2976 			valid = 1;
2977 			nid_ofs = 1;
2978 		}
2979 
2980 		for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) {
2981 			if (!test_bit_le(nid_ofs,
2982 					nm_i->free_nid_bitmap[nat_ofs]))
2983 				valid++;
2984 		}
2985 
2986 		__update_nat_bits(nm_i, nat_ofs, valid);
2987 	}
2988 
2989 	f2fs_up_read(&nm_i->nat_tree_lock);
2990 }
2991 
2992 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2993 		struct nat_entry_set *set, struct cp_control *cpc)
2994 {
2995 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2996 	struct f2fs_journal *journal = curseg->journal;
2997 	nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2998 	bool to_journal = true;
2999 	struct f2fs_nat_block *nat_blk;
3000 	struct nat_entry *ne, *cur;
3001 	struct page *page = NULL;
3002 
3003 	/*
3004 	 * there are two steps to flush nat entries:
3005 	 * #1, flush nat entries to journal in current hot data summary block.
3006 	 * #2, flush nat entries to nat page.
3007 	 */
3008 	if ((cpc->reason & CP_UMOUNT) ||
3009 		!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
3010 		to_journal = false;
3011 
3012 	if (to_journal) {
3013 		down_write(&curseg->journal_rwsem);
3014 	} else {
3015 		page = get_next_nat_page(sbi, start_nid);
3016 		if (IS_ERR(page))
3017 			return PTR_ERR(page);
3018 
3019 		nat_blk = page_address(page);
3020 		f2fs_bug_on(sbi, !nat_blk);
3021 	}
3022 
3023 	/* flush dirty nats in nat entry set */
3024 	list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
3025 		struct f2fs_nat_entry *raw_ne;
3026 		nid_t nid = nat_get_nid(ne);
3027 		int offset;
3028 
3029 		f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
3030 
3031 		if (to_journal) {
3032 			offset = f2fs_lookup_journal_in_cursum(journal,
3033 							NAT_JOURNAL, nid, 1);
3034 			f2fs_bug_on(sbi, offset < 0);
3035 			raw_ne = &nat_in_journal(journal, offset);
3036 			nid_in_journal(journal, offset) = cpu_to_le32(nid);
3037 		} else {
3038 			raw_ne = &nat_blk->entries[nid - start_nid];
3039 		}
3040 		raw_nat_from_node_info(raw_ne, &ne->ni);
3041 		nat_reset_flag(ne);
3042 		__clear_nat_cache_dirty(NM_I(sbi), set, ne);
3043 		if (nat_get_blkaddr(ne) == NULL_ADDR) {
3044 			add_free_nid(sbi, nid, false, true);
3045 		} else {
3046 			spin_lock(&NM_I(sbi)->nid_list_lock);
3047 			update_free_nid_bitmap(sbi, nid, false, false);
3048 			spin_unlock(&NM_I(sbi)->nid_list_lock);
3049 		}
3050 	}
3051 
3052 	if (to_journal) {
3053 		up_write(&curseg->journal_rwsem);
3054 	} else {
3055 		update_nat_bits(sbi, start_nid, page);
3056 		f2fs_put_page(page, 1);
3057 	}
3058 
3059 	/* Allow dirty nats by node block allocation in write_begin */
3060 	if (!set->entry_cnt) {
3061 		radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
3062 		kmem_cache_free(nat_entry_set_slab, set);
3063 	}
3064 	return 0;
3065 }
3066 
3067 /*
3068  * This function is called during the checkpointing process.
3069  */
3070 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3071 {
3072 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3073 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3074 	struct f2fs_journal *journal = curseg->journal;
3075 	struct nat_entry_set *setvec[NAT_VEC_SIZE];
3076 	struct nat_entry_set *set, *tmp;
3077 	unsigned int found;
3078 	nid_t set_idx = 0;
3079 	LIST_HEAD(sets);
3080 	int err = 0;
3081 
3082 	/*
3083 	 * during unmount, let's flush nat_bits before checking
3084 	 * nat_cnt[DIRTY_NAT].
3085 	 */
3086 	if (cpc->reason & CP_UMOUNT) {
3087 		f2fs_down_write(&nm_i->nat_tree_lock);
3088 		remove_nats_in_journal(sbi);
3089 		f2fs_up_write(&nm_i->nat_tree_lock);
3090 	}
3091 
3092 	if (!nm_i->nat_cnt[DIRTY_NAT])
3093 		return 0;
3094 
3095 	f2fs_down_write(&nm_i->nat_tree_lock);
3096 
3097 	/*
3098 	 * if there are no enough space in journal to store dirty nat
3099 	 * entries, remove all entries from journal and merge them
3100 	 * into nat entry set.
3101 	 */
3102 	if (cpc->reason & CP_UMOUNT ||
3103 		!__has_cursum_space(journal,
3104 			nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3105 		remove_nats_in_journal(sbi);
3106 
3107 	while ((found = __gang_lookup_nat_set(nm_i,
3108 					set_idx, NAT_VEC_SIZE, setvec))) {
3109 		unsigned idx;
3110 
3111 		set_idx = setvec[found - 1]->set + 1;
3112 		for (idx = 0; idx < found; idx++)
3113 			__adjust_nat_entry_set(setvec[idx], &sets,
3114 						MAX_NAT_JENTRIES(journal));
3115 	}
3116 
3117 	/* flush dirty nats in nat entry set */
3118 	list_for_each_entry_safe(set, tmp, &sets, set_list) {
3119 		err = __flush_nat_entry_set(sbi, set, cpc);
3120 		if (err)
3121 			break;
3122 	}
3123 
3124 	f2fs_up_write(&nm_i->nat_tree_lock);
3125 	/* Allow dirty nats by node block allocation in write_begin */
3126 
3127 	return err;
3128 }
3129 
3130 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3131 {
3132 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3133 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3134 	unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3135 	unsigned int i;
3136 	__u64 cp_ver = cur_cp_version(ckpt);
3137 	block_t nat_bits_addr;
3138 
3139 	nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3140 	nm_i->nat_bits = f2fs_kvzalloc(sbi,
3141 			nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3142 	if (!nm_i->nat_bits)
3143 		return -ENOMEM;
3144 
3145 	nm_i->full_nat_bits = nm_i->nat_bits + 8;
3146 	nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3147 
3148 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3149 		return 0;
3150 
3151 	nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3152 						nm_i->nat_bits_blocks;
3153 	for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3154 		struct page *page;
3155 
3156 		page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3157 		if (IS_ERR(page))
3158 			return PTR_ERR(page);
3159 
3160 		memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3161 					page_address(page), F2FS_BLKSIZE);
3162 		f2fs_put_page(page, 1);
3163 	}
3164 
3165 	cp_ver |= (cur_cp_crc(ckpt) << 32);
3166 	if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3167 		clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
3168 		f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)",
3169 			cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits));
3170 		return 0;
3171 	}
3172 
3173 	f2fs_notice(sbi, "Found nat_bits in checkpoint");
3174 	return 0;
3175 }
3176 
3177 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3178 {
3179 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3180 	unsigned int i = 0;
3181 	nid_t nid, last_nid;
3182 
3183 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3184 		return;
3185 
3186 	for (i = 0; i < nm_i->nat_blocks; i++) {
3187 		i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3188 		if (i >= nm_i->nat_blocks)
3189 			break;
3190 
3191 		__set_bit_le(i, nm_i->nat_block_bitmap);
3192 
3193 		nid = i * NAT_ENTRY_PER_BLOCK;
3194 		last_nid = nid + NAT_ENTRY_PER_BLOCK;
3195 
3196 		spin_lock(&NM_I(sbi)->nid_list_lock);
3197 		for (; nid < last_nid; nid++)
3198 			update_free_nid_bitmap(sbi, nid, true, true);
3199 		spin_unlock(&NM_I(sbi)->nid_list_lock);
3200 	}
3201 
3202 	for (i = 0; i < nm_i->nat_blocks; i++) {
3203 		i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3204 		if (i >= nm_i->nat_blocks)
3205 			break;
3206 
3207 		__set_bit_le(i, nm_i->nat_block_bitmap);
3208 	}
3209 }
3210 
3211 static int init_node_manager(struct f2fs_sb_info *sbi)
3212 {
3213 	struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3214 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3215 	unsigned char *version_bitmap;
3216 	unsigned int nat_segs;
3217 	int err;
3218 
3219 	nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3220 
3221 	/* segment_count_nat includes pair segment so divide to 2. */
3222 	nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3223 	nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3224 	nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3225 
3226 	/* not used nids: 0, node, meta, (and root counted as valid node) */
3227 	nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3228 						F2FS_RESERVED_NODE_NUM;
3229 	nm_i->nid_cnt[FREE_NID] = 0;
3230 	nm_i->nid_cnt[PREALLOC_NID] = 0;
3231 	nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3232 	nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3233 	nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3234 	nm_i->max_rf_node_blocks = DEF_RF_NODE_BLOCKS;
3235 
3236 	INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3237 	INIT_LIST_HEAD(&nm_i->free_nid_list);
3238 	INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3239 	INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3240 	INIT_LIST_HEAD(&nm_i->nat_entries);
3241 	spin_lock_init(&nm_i->nat_list_lock);
3242 
3243 	mutex_init(&nm_i->build_lock);
3244 	spin_lock_init(&nm_i->nid_list_lock);
3245 	init_f2fs_rwsem(&nm_i->nat_tree_lock);
3246 
3247 	nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3248 	nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3249 	version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3250 	nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3251 					GFP_KERNEL);
3252 	if (!nm_i->nat_bitmap)
3253 		return -ENOMEM;
3254 
3255 	err = __get_nat_bitmaps(sbi);
3256 	if (err)
3257 		return err;
3258 
3259 #ifdef CONFIG_F2FS_CHECK_FS
3260 	nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3261 					GFP_KERNEL);
3262 	if (!nm_i->nat_bitmap_mir)
3263 		return -ENOMEM;
3264 #endif
3265 
3266 	return 0;
3267 }
3268 
3269 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3270 {
3271 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3272 	int i;
3273 
3274 	nm_i->free_nid_bitmap =
3275 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3276 					      nm_i->nat_blocks),
3277 			      GFP_KERNEL);
3278 	if (!nm_i->free_nid_bitmap)
3279 		return -ENOMEM;
3280 
3281 	for (i = 0; i < nm_i->nat_blocks; i++) {
3282 		nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3283 			f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3284 		if (!nm_i->free_nid_bitmap[i])
3285 			return -ENOMEM;
3286 	}
3287 
3288 	nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3289 								GFP_KERNEL);
3290 	if (!nm_i->nat_block_bitmap)
3291 		return -ENOMEM;
3292 
3293 	nm_i->free_nid_count =
3294 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3295 					      nm_i->nat_blocks),
3296 			      GFP_KERNEL);
3297 	if (!nm_i->free_nid_count)
3298 		return -ENOMEM;
3299 	return 0;
3300 }
3301 
3302 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3303 {
3304 	int err;
3305 
3306 	sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3307 							GFP_KERNEL);
3308 	if (!sbi->nm_info)
3309 		return -ENOMEM;
3310 
3311 	err = init_node_manager(sbi);
3312 	if (err)
3313 		return err;
3314 
3315 	err = init_free_nid_cache(sbi);
3316 	if (err)
3317 		return err;
3318 
3319 	/* load free nid status from nat_bits table */
3320 	load_free_nid_bitmap(sbi);
3321 
3322 	return f2fs_build_free_nids(sbi, true, true);
3323 }
3324 
3325 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3326 {
3327 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3328 	struct free_nid *i, *next_i;
3329 	void *vec[NAT_VEC_SIZE];
3330 	struct nat_entry **natvec = (struct nat_entry **)vec;
3331 	struct nat_entry_set **setvec = (struct nat_entry_set **)vec;
3332 	nid_t nid = 0;
3333 	unsigned int found;
3334 
3335 	if (!nm_i)
3336 		return;
3337 
3338 	/* destroy free nid list */
3339 	spin_lock(&nm_i->nid_list_lock);
3340 	list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3341 		__remove_free_nid(sbi, i, FREE_NID);
3342 		spin_unlock(&nm_i->nid_list_lock);
3343 		kmem_cache_free(free_nid_slab, i);
3344 		spin_lock(&nm_i->nid_list_lock);
3345 	}
3346 	f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3347 	f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3348 	f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3349 	spin_unlock(&nm_i->nid_list_lock);
3350 
3351 	/* destroy nat cache */
3352 	f2fs_down_write(&nm_i->nat_tree_lock);
3353 	while ((found = __gang_lookup_nat_cache(nm_i,
3354 					nid, NAT_VEC_SIZE, natvec))) {
3355 		unsigned idx;
3356 
3357 		nid = nat_get_nid(natvec[found - 1]) + 1;
3358 		for (idx = 0; idx < found; idx++) {
3359 			spin_lock(&nm_i->nat_list_lock);
3360 			list_del(&natvec[idx]->list);
3361 			spin_unlock(&nm_i->nat_list_lock);
3362 
3363 			__del_from_nat_cache(nm_i, natvec[idx]);
3364 		}
3365 	}
3366 	f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3367 
3368 	/* destroy nat set cache */
3369 	nid = 0;
3370 	memset(vec, 0, sizeof(void *) * NAT_VEC_SIZE);
3371 	while ((found = __gang_lookup_nat_set(nm_i,
3372 					nid, NAT_VEC_SIZE, setvec))) {
3373 		unsigned idx;
3374 
3375 		nid = setvec[found - 1]->set + 1;
3376 		for (idx = 0; idx < found; idx++) {
3377 			/* entry_cnt is not zero, when cp_error was occurred */
3378 			f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3379 			radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3380 			kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3381 		}
3382 	}
3383 	f2fs_up_write(&nm_i->nat_tree_lock);
3384 
3385 	kvfree(nm_i->nat_block_bitmap);
3386 	if (nm_i->free_nid_bitmap) {
3387 		int i;
3388 
3389 		for (i = 0; i < nm_i->nat_blocks; i++)
3390 			kvfree(nm_i->free_nid_bitmap[i]);
3391 		kvfree(nm_i->free_nid_bitmap);
3392 	}
3393 	kvfree(nm_i->free_nid_count);
3394 
3395 	kvfree(nm_i->nat_bitmap);
3396 	kvfree(nm_i->nat_bits);
3397 #ifdef CONFIG_F2FS_CHECK_FS
3398 	kvfree(nm_i->nat_bitmap_mir);
3399 #endif
3400 	sbi->nm_info = NULL;
3401 	kfree(nm_i);
3402 }
3403 
3404 int __init f2fs_create_node_manager_caches(void)
3405 {
3406 	nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3407 			sizeof(struct nat_entry));
3408 	if (!nat_entry_slab)
3409 		goto fail;
3410 
3411 	free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3412 			sizeof(struct free_nid));
3413 	if (!free_nid_slab)
3414 		goto destroy_nat_entry;
3415 
3416 	nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3417 			sizeof(struct nat_entry_set));
3418 	if (!nat_entry_set_slab)
3419 		goto destroy_free_nid;
3420 
3421 	fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3422 			sizeof(struct fsync_node_entry));
3423 	if (!fsync_node_entry_slab)
3424 		goto destroy_nat_entry_set;
3425 	return 0;
3426 
3427 destroy_nat_entry_set:
3428 	kmem_cache_destroy(nat_entry_set_slab);
3429 destroy_free_nid:
3430 	kmem_cache_destroy(free_nid_slab);
3431 destroy_nat_entry:
3432 	kmem_cache_destroy(nat_entry_slab);
3433 fail:
3434 	return -ENOMEM;
3435 }
3436 
3437 void f2fs_destroy_node_manager_caches(void)
3438 {
3439 	kmem_cache_destroy(fsync_node_entry_slab);
3440 	kmem_cache_destroy(nat_entry_set_slab);
3441 	kmem_cache_destroy(free_nid_slab);
3442 	kmem_cache_destroy(nat_entry_slab);
3443 }
3444