xref: /linux/fs/f2fs/node.c (revision 72bea132f3680ee51e7ed2cee62892b6f5121909)
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, (int)NIDS_PER_BLOCK);
637 	for (i = start; i < end; i++) {
638 		nid = get_nid(parent, i, false);
639 		f2fs_ra_node_page(sbi, nid);
640 	}
641 
642 	blk_finish_plug(&plug);
643 }
644 
645 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
646 {
647 	const long direct_index = ADDRS_PER_INODE(dn->inode);
648 	const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
649 	const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
650 	unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
651 	int cur_level = dn->cur_level;
652 	int max_level = dn->max_level;
653 	pgoff_t base = 0;
654 
655 	if (!dn->max_level)
656 		return pgofs + 1;
657 
658 	while (max_level-- > cur_level)
659 		skipped_unit *= NIDS_PER_BLOCK;
660 
661 	switch (dn->max_level) {
662 	case 3:
663 		base += 2 * indirect_blks;
664 		fallthrough;
665 	case 2:
666 		base += 2 * direct_blks;
667 		fallthrough;
668 	case 1:
669 		base += direct_index;
670 		break;
671 	default:
672 		f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
673 	}
674 
675 	return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
676 }
677 
678 /*
679  * The maximum depth is four.
680  * Offset[0] will have raw inode offset.
681  */
682 static int get_node_path(struct inode *inode, long block,
683 				int offset[4], unsigned int noffset[4])
684 {
685 	const long direct_index = ADDRS_PER_INODE(inode);
686 	const long direct_blks = ADDRS_PER_BLOCK(inode);
687 	const long dptrs_per_blk = NIDS_PER_BLOCK;
688 	const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
689 	const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
690 	int n = 0;
691 	int level = 0;
692 
693 	noffset[0] = 0;
694 
695 	if (block < direct_index) {
696 		offset[n] = block;
697 		goto got;
698 	}
699 	block -= direct_index;
700 	if (block < direct_blks) {
701 		offset[n++] = NODE_DIR1_BLOCK;
702 		noffset[n] = 1;
703 		offset[n] = block;
704 		level = 1;
705 		goto got;
706 	}
707 	block -= direct_blks;
708 	if (block < direct_blks) {
709 		offset[n++] = NODE_DIR2_BLOCK;
710 		noffset[n] = 2;
711 		offset[n] = block;
712 		level = 1;
713 		goto got;
714 	}
715 	block -= direct_blks;
716 	if (block < indirect_blks) {
717 		offset[n++] = NODE_IND1_BLOCK;
718 		noffset[n] = 3;
719 		offset[n++] = block / direct_blks;
720 		noffset[n] = 4 + offset[n - 1];
721 		offset[n] = block % direct_blks;
722 		level = 2;
723 		goto got;
724 	}
725 	block -= indirect_blks;
726 	if (block < indirect_blks) {
727 		offset[n++] = NODE_IND2_BLOCK;
728 		noffset[n] = 4 + dptrs_per_blk;
729 		offset[n++] = block / direct_blks;
730 		noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
731 		offset[n] = block % direct_blks;
732 		level = 2;
733 		goto got;
734 	}
735 	block -= indirect_blks;
736 	if (block < dindirect_blks) {
737 		offset[n++] = NODE_DIND_BLOCK;
738 		noffset[n] = 5 + (dptrs_per_blk * 2);
739 		offset[n++] = block / indirect_blks;
740 		noffset[n] = 6 + (dptrs_per_blk * 2) +
741 			      offset[n - 1] * (dptrs_per_blk + 1);
742 		offset[n++] = (block / direct_blks) % dptrs_per_blk;
743 		noffset[n] = 7 + (dptrs_per_blk * 2) +
744 			      offset[n - 2] * (dptrs_per_blk + 1) +
745 			      offset[n - 1];
746 		offset[n] = block % direct_blks;
747 		level = 3;
748 		goto got;
749 	} else {
750 		return -E2BIG;
751 	}
752 got:
753 	return level;
754 }
755 
756 /*
757  * Caller should call f2fs_put_dnode(dn).
758  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
759  * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
760  */
761 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
762 {
763 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
764 	struct page *npage[4];
765 	struct page *parent = NULL;
766 	int offset[4];
767 	unsigned int noffset[4];
768 	nid_t nids[4];
769 	int level, i = 0;
770 	int err = 0;
771 
772 	level = get_node_path(dn->inode, index, offset, noffset);
773 	if (level < 0)
774 		return level;
775 
776 	nids[0] = dn->inode->i_ino;
777 	npage[0] = dn->inode_page;
778 
779 	if (!npage[0]) {
780 		npage[0] = f2fs_get_node_page(sbi, nids[0]);
781 		if (IS_ERR(npage[0]))
782 			return PTR_ERR(npage[0]);
783 	}
784 
785 	/* if inline_data is set, should not report any block indices */
786 	if (f2fs_has_inline_data(dn->inode) && index) {
787 		err = -ENOENT;
788 		f2fs_put_page(npage[0], 1);
789 		goto release_out;
790 	}
791 
792 	parent = npage[0];
793 	if (level != 0)
794 		nids[1] = get_nid(parent, offset[0], true);
795 	dn->inode_page = npage[0];
796 	dn->inode_page_locked = true;
797 
798 	/* get indirect or direct nodes */
799 	for (i = 1; i <= level; i++) {
800 		bool done = false;
801 
802 		if (!nids[i] && mode == ALLOC_NODE) {
803 			/* alloc new node */
804 			if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
805 				err = -ENOSPC;
806 				goto release_pages;
807 			}
808 
809 			dn->nid = nids[i];
810 			npage[i] = f2fs_new_node_page(dn, noffset[i]);
811 			if (IS_ERR(npage[i])) {
812 				f2fs_alloc_nid_failed(sbi, nids[i]);
813 				err = PTR_ERR(npage[i]);
814 				goto release_pages;
815 			}
816 
817 			set_nid(parent, offset[i - 1], nids[i], i == 1);
818 			f2fs_alloc_nid_done(sbi, nids[i]);
819 			done = true;
820 		} else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
821 			npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
822 			if (IS_ERR(npage[i])) {
823 				err = PTR_ERR(npage[i]);
824 				goto release_pages;
825 			}
826 			done = true;
827 		}
828 		if (i == 1) {
829 			dn->inode_page_locked = false;
830 			unlock_page(parent);
831 		} else {
832 			f2fs_put_page(parent, 1);
833 		}
834 
835 		if (!done) {
836 			npage[i] = f2fs_get_node_page(sbi, nids[i]);
837 			if (IS_ERR(npage[i])) {
838 				err = PTR_ERR(npage[i]);
839 				f2fs_put_page(npage[0], 0);
840 				goto release_out;
841 			}
842 		}
843 		if (i < level) {
844 			parent = npage[i];
845 			nids[i + 1] = get_nid(parent, offset[i], false);
846 		}
847 	}
848 	dn->nid = nids[level];
849 	dn->ofs_in_node = offset[level];
850 	dn->node_page = npage[level];
851 	dn->data_blkaddr = f2fs_data_blkaddr(dn);
852 
853 	if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) &&
854 					f2fs_sb_has_readonly(sbi)) {
855 		unsigned int cluster_size = F2FS_I(dn->inode)->i_cluster_size;
856 		unsigned int ofs_in_node = dn->ofs_in_node;
857 		pgoff_t fofs = index;
858 		unsigned int c_len;
859 		block_t blkaddr;
860 
861 		/* should align fofs and ofs_in_node to cluster_size */
862 		if (fofs % cluster_size) {
863 			fofs = round_down(fofs, cluster_size);
864 			ofs_in_node = round_down(ofs_in_node, cluster_size);
865 		}
866 
867 		c_len = f2fs_cluster_blocks_are_contiguous(dn, ofs_in_node);
868 		if (!c_len)
869 			goto out;
870 
871 		blkaddr = data_blkaddr(dn->inode, dn->node_page, ofs_in_node);
872 		if (blkaddr == COMPRESS_ADDR)
873 			blkaddr = data_blkaddr(dn->inode, dn->node_page,
874 						ofs_in_node + 1);
875 
876 		f2fs_update_read_extent_tree_range_compressed(dn->inode,
877 					fofs, blkaddr, cluster_size, c_len);
878 	}
879 out:
880 	return 0;
881 
882 release_pages:
883 	f2fs_put_page(parent, 1);
884 	if (i > 1)
885 		f2fs_put_page(npage[0], 0);
886 release_out:
887 	dn->inode_page = NULL;
888 	dn->node_page = NULL;
889 	if (err == -ENOENT) {
890 		dn->cur_level = i;
891 		dn->max_level = level;
892 		dn->ofs_in_node = offset[level];
893 	}
894 	return err;
895 }
896 
897 static int truncate_node(struct dnode_of_data *dn)
898 {
899 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
900 	struct node_info ni;
901 	int err;
902 	pgoff_t index;
903 
904 	err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
905 	if (err)
906 		return err;
907 
908 	/* Deallocate node address */
909 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
910 	dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
911 	set_node_addr(sbi, &ni, NULL_ADDR, false);
912 
913 	if (dn->nid == dn->inode->i_ino) {
914 		f2fs_remove_orphan_inode(sbi, dn->nid);
915 		dec_valid_inode_count(sbi);
916 		f2fs_inode_synced(dn->inode);
917 	}
918 
919 	clear_node_page_dirty(dn->node_page);
920 	set_sbi_flag(sbi, SBI_IS_DIRTY);
921 
922 	index = dn->node_page->index;
923 	f2fs_put_page(dn->node_page, 1);
924 
925 	invalidate_mapping_pages(NODE_MAPPING(sbi),
926 			index, index);
927 
928 	dn->node_page = NULL;
929 	trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
930 
931 	return 0;
932 }
933 
934 static int truncate_dnode(struct dnode_of_data *dn)
935 {
936 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
937 	struct page *page;
938 	int err;
939 
940 	if (dn->nid == 0)
941 		return 1;
942 
943 	/* get direct node */
944 	page = f2fs_get_node_page(sbi, dn->nid);
945 	if (PTR_ERR(page) == -ENOENT)
946 		return 1;
947 	else if (IS_ERR(page))
948 		return PTR_ERR(page);
949 
950 	if (IS_INODE(page) || ino_of_node(page) != dn->inode->i_ino) {
951 		f2fs_err(sbi, "incorrect node reference, ino: %lu, nid: %u, ino_of_node: %u",
952 				dn->inode->i_ino, dn->nid, ino_of_node(page));
953 		set_sbi_flag(sbi, SBI_NEED_FSCK);
954 		f2fs_handle_error(sbi, ERROR_INVALID_NODE_REFERENCE);
955 		f2fs_put_page(page, 1);
956 		return -EFSCORRUPTED;
957 	}
958 
959 	/* Make dnode_of_data for parameter */
960 	dn->node_page = page;
961 	dn->ofs_in_node = 0;
962 	f2fs_truncate_data_blocks_range(dn, ADDRS_PER_BLOCK(dn->inode));
963 	err = truncate_node(dn);
964 	if (err) {
965 		f2fs_put_page(page, 1);
966 		return err;
967 	}
968 
969 	return 1;
970 }
971 
972 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
973 						int ofs, int depth)
974 {
975 	struct dnode_of_data rdn = *dn;
976 	struct page *page;
977 	struct f2fs_node *rn;
978 	nid_t child_nid;
979 	unsigned int child_nofs;
980 	int freed = 0;
981 	int i, ret;
982 
983 	if (dn->nid == 0)
984 		return NIDS_PER_BLOCK + 1;
985 
986 	trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
987 
988 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
989 	if (IS_ERR(page)) {
990 		trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
991 		return PTR_ERR(page);
992 	}
993 
994 	f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
995 
996 	rn = F2FS_NODE(page);
997 	if (depth < 3) {
998 		for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
999 			child_nid = le32_to_cpu(rn->in.nid[i]);
1000 			if (child_nid == 0)
1001 				continue;
1002 			rdn.nid = child_nid;
1003 			ret = truncate_dnode(&rdn);
1004 			if (ret < 0)
1005 				goto out_err;
1006 			if (set_nid(page, i, 0, false))
1007 				dn->node_changed = true;
1008 		}
1009 	} else {
1010 		child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
1011 		for (i = ofs; i < NIDS_PER_BLOCK; i++) {
1012 			child_nid = le32_to_cpu(rn->in.nid[i]);
1013 			if (child_nid == 0) {
1014 				child_nofs += NIDS_PER_BLOCK + 1;
1015 				continue;
1016 			}
1017 			rdn.nid = child_nid;
1018 			ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
1019 			if (ret == (NIDS_PER_BLOCK + 1)) {
1020 				if (set_nid(page, i, 0, false))
1021 					dn->node_changed = true;
1022 				child_nofs += ret;
1023 			} else if (ret < 0 && ret != -ENOENT) {
1024 				goto out_err;
1025 			}
1026 		}
1027 		freed = child_nofs;
1028 	}
1029 
1030 	if (!ofs) {
1031 		/* remove current indirect node */
1032 		dn->node_page = page;
1033 		ret = truncate_node(dn);
1034 		if (ret)
1035 			goto out_err;
1036 		freed++;
1037 	} else {
1038 		f2fs_put_page(page, 1);
1039 	}
1040 	trace_f2fs_truncate_nodes_exit(dn->inode, freed);
1041 	return freed;
1042 
1043 out_err:
1044 	f2fs_put_page(page, 1);
1045 	trace_f2fs_truncate_nodes_exit(dn->inode, ret);
1046 	return ret;
1047 }
1048 
1049 static int truncate_partial_nodes(struct dnode_of_data *dn,
1050 			struct f2fs_inode *ri, int *offset, int depth)
1051 {
1052 	struct page *pages[2];
1053 	nid_t nid[3];
1054 	nid_t child_nid;
1055 	int err = 0;
1056 	int i;
1057 	int idx = depth - 2;
1058 
1059 	nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1060 	if (!nid[0])
1061 		return 0;
1062 
1063 	/* get indirect nodes in the path */
1064 	for (i = 0; i < idx + 1; i++) {
1065 		/* reference count'll be increased */
1066 		pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
1067 		if (IS_ERR(pages[i])) {
1068 			err = PTR_ERR(pages[i]);
1069 			idx = i - 1;
1070 			goto fail;
1071 		}
1072 		nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1073 	}
1074 
1075 	f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1076 
1077 	/* free direct nodes linked to a partial indirect node */
1078 	for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1079 		child_nid = get_nid(pages[idx], i, false);
1080 		if (!child_nid)
1081 			continue;
1082 		dn->nid = child_nid;
1083 		err = truncate_dnode(dn);
1084 		if (err < 0)
1085 			goto fail;
1086 		if (set_nid(pages[idx], i, 0, false))
1087 			dn->node_changed = true;
1088 	}
1089 
1090 	if (offset[idx + 1] == 0) {
1091 		dn->node_page = pages[idx];
1092 		dn->nid = nid[idx];
1093 		err = truncate_node(dn);
1094 		if (err)
1095 			goto fail;
1096 	} else {
1097 		f2fs_put_page(pages[idx], 1);
1098 	}
1099 	offset[idx]++;
1100 	offset[idx + 1] = 0;
1101 	idx--;
1102 fail:
1103 	for (i = idx; i >= 0; i--)
1104 		f2fs_put_page(pages[i], 1);
1105 
1106 	trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1107 
1108 	return err;
1109 }
1110 
1111 /*
1112  * All the block addresses of data and nodes should be nullified.
1113  */
1114 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1115 {
1116 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1117 	int err = 0, cont = 1;
1118 	int level, offset[4], noffset[4];
1119 	unsigned int nofs = 0;
1120 	struct f2fs_inode *ri;
1121 	struct dnode_of_data dn;
1122 	struct page *page;
1123 
1124 	trace_f2fs_truncate_inode_blocks_enter(inode, from);
1125 
1126 	level = get_node_path(inode, from, offset, noffset);
1127 	if (level < 0) {
1128 		trace_f2fs_truncate_inode_blocks_exit(inode, level);
1129 		return level;
1130 	}
1131 
1132 	page = f2fs_get_node_page(sbi, inode->i_ino);
1133 	if (IS_ERR(page)) {
1134 		trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1135 		return PTR_ERR(page);
1136 	}
1137 
1138 	set_new_dnode(&dn, inode, page, NULL, 0);
1139 	unlock_page(page);
1140 
1141 	ri = F2FS_INODE(page);
1142 	switch (level) {
1143 	case 0:
1144 	case 1:
1145 		nofs = noffset[1];
1146 		break;
1147 	case 2:
1148 		nofs = noffset[1];
1149 		if (!offset[level - 1])
1150 			goto skip_partial;
1151 		err = truncate_partial_nodes(&dn, ri, offset, level);
1152 		if (err < 0 && err != -ENOENT)
1153 			goto fail;
1154 		nofs += 1 + NIDS_PER_BLOCK;
1155 		break;
1156 	case 3:
1157 		nofs = 5 + 2 * NIDS_PER_BLOCK;
1158 		if (!offset[level - 1])
1159 			goto skip_partial;
1160 		err = truncate_partial_nodes(&dn, ri, offset, level);
1161 		if (err < 0 && err != -ENOENT)
1162 			goto fail;
1163 		break;
1164 	default:
1165 		BUG();
1166 	}
1167 
1168 skip_partial:
1169 	while (cont) {
1170 		dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1171 		switch (offset[0]) {
1172 		case NODE_DIR1_BLOCK:
1173 		case NODE_DIR2_BLOCK:
1174 			err = truncate_dnode(&dn);
1175 			break;
1176 
1177 		case NODE_IND1_BLOCK:
1178 		case NODE_IND2_BLOCK:
1179 			err = truncate_nodes(&dn, nofs, offset[1], 2);
1180 			break;
1181 
1182 		case NODE_DIND_BLOCK:
1183 			err = truncate_nodes(&dn, nofs, offset[1], 3);
1184 			cont = 0;
1185 			break;
1186 
1187 		default:
1188 			BUG();
1189 		}
1190 		if (err < 0 && err != -ENOENT)
1191 			goto fail;
1192 		if (offset[1] == 0 &&
1193 				ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1194 			lock_page(page);
1195 			BUG_ON(page->mapping != NODE_MAPPING(sbi));
1196 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1197 			ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1198 			set_page_dirty(page);
1199 			unlock_page(page);
1200 		}
1201 		offset[1] = 0;
1202 		offset[0]++;
1203 		nofs += err;
1204 	}
1205 fail:
1206 	f2fs_put_page(page, 0);
1207 	trace_f2fs_truncate_inode_blocks_exit(inode, err);
1208 	return err > 0 ? 0 : err;
1209 }
1210 
1211 /* caller must lock inode page */
1212 int f2fs_truncate_xattr_node(struct inode *inode)
1213 {
1214 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1215 	nid_t nid = F2FS_I(inode)->i_xattr_nid;
1216 	struct dnode_of_data dn;
1217 	struct page *npage;
1218 	int err;
1219 
1220 	if (!nid)
1221 		return 0;
1222 
1223 	npage = f2fs_get_node_page(sbi, nid);
1224 	if (IS_ERR(npage))
1225 		return PTR_ERR(npage);
1226 
1227 	set_new_dnode(&dn, inode, NULL, npage, nid);
1228 	err = truncate_node(&dn);
1229 	if (err) {
1230 		f2fs_put_page(npage, 1);
1231 		return err;
1232 	}
1233 
1234 	f2fs_i_xnid_write(inode, 0);
1235 
1236 	return 0;
1237 }
1238 
1239 /*
1240  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1241  * f2fs_unlock_op().
1242  */
1243 int f2fs_remove_inode_page(struct inode *inode)
1244 {
1245 	struct dnode_of_data dn;
1246 	int err;
1247 
1248 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1249 	err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1250 	if (err)
1251 		return err;
1252 
1253 	err = f2fs_truncate_xattr_node(inode);
1254 	if (err) {
1255 		f2fs_put_dnode(&dn);
1256 		return err;
1257 	}
1258 
1259 	/* remove potential inline_data blocks */
1260 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1261 				S_ISLNK(inode->i_mode))
1262 		f2fs_truncate_data_blocks_range(&dn, 1);
1263 
1264 	/* 0 is possible, after f2fs_new_inode() has failed */
1265 	if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1266 		f2fs_put_dnode(&dn);
1267 		return -EIO;
1268 	}
1269 
1270 	if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1271 		f2fs_warn(F2FS_I_SB(inode),
1272 			"f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1273 			inode->i_ino, (unsigned long long)inode->i_blocks);
1274 		set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1275 	}
1276 
1277 	/* will put inode & node pages */
1278 	err = truncate_node(&dn);
1279 	if (err) {
1280 		f2fs_put_dnode(&dn);
1281 		return err;
1282 	}
1283 	return 0;
1284 }
1285 
1286 struct page *f2fs_new_inode_page(struct inode *inode)
1287 {
1288 	struct dnode_of_data dn;
1289 
1290 	/* allocate inode page for new inode */
1291 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1292 
1293 	/* caller should f2fs_put_page(page, 1); */
1294 	return f2fs_new_node_page(&dn, 0);
1295 }
1296 
1297 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1298 {
1299 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1300 	struct node_info new_ni;
1301 	struct page *page;
1302 	int err;
1303 
1304 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1305 		return ERR_PTR(-EPERM);
1306 
1307 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1308 	if (!page)
1309 		return ERR_PTR(-ENOMEM);
1310 
1311 	if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1312 		goto fail;
1313 
1314 #ifdef CONFIG_F2FS_CHECK_FS
1315 	err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false);
1316 	if (err) {
1317 		dec_valid_node_count(sbi, dn->inode, !ofs);
1318 		goto fail;
1319 	}
1320 	if (unlikely(new_ni.blk_addr != NULL_ADDR)) {
1321 		err = -EFSCORRUPTED;
1322 		set_sbi_flag(sbi, SBI_NEED_FSCK);
1323 		f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1324 		goto fail;
1325 	}
1326 #endif
1327 	new_ni.nid = dn->nid;
1328 	new_ni.ino = dn->inode->i_ino;
1329 	new_ni.blk_addr = NULL_ADDR;
1330 	new_ni.flag = 0;
1331 	new_ni.version = 0;
1332 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1333 
1334 	f2fs_wait_on_page_writeback(page, NODE, true, true);
1335 	fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1336 	set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1337 	if (!PageUptodate(page))
1338 		SetPageUptodate(page);
1339 	if (set_page_dirty(page))
1340 		dn->node_changed = true;
1341 
1342 	if (f2fs_has_xattr_block(ofs))
1343 		f2fs_i_xnid_write(dn->inode, dn->nid);
1344 
1345 	if (ofs == 0)
1346 		inc_valid_inode_count(sbi);
1347 	return page;
1348 
1349 fail:
1350 	clear_node_page_dirty(page);
1351 	f2fs_put_page(page, 1);
1352 	return ERR_PTR(err);
1353 }
1354 
1355 /*
1356  * Caller should do after getting the following values.
1357  * 0: f2fs_put_page(page, 0)
1358  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1359  */
1360 static int read_node_page(struct page *page, blk_opf_t op_flags)
1361 {
1362 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1363 	struct node_info ni;
1364 	struct f2fs_io_info fio = {
1365 		.sbi = sbi,
1366 		.type = NODE,
1367 		.op = REQ_OP_READ,
1368 		.op_flags = op_flags,
1369 		.page = page,
1370 		.encrypted_page = NULL,
1371 	};
1372 	int err;
1373 
1374 	if (PageUptodate(page)) {
1375 		if (!f2fs_inode_chksum_verify(sbi, page)) {
1376 			ClearPageUptodate(page);
1377 			return -EFSBADCRC;
1378 		}
1379 		return LOCKED_PAGE;
1380 	}
1381 
1382 	err = f2fs_get_node_info(sbi, page->index, &ni, false);
1383 	if (err)
1384 		return err;
1385 
1386 	/* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1387 	if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR)) {
1388 		ClearPageUptodate(page);
1389 		return -ENOENT;
1390 	}
1391 
1392 	fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1393 
1394 	err = f2fs_submit_page_bio(&fio);
1395 
1396 	if (!err)
1397 		f2fs_update_iostat(sbi, NULL, FS_NODE_READ_IO, F2FS_BLKSIZE);
1398 
1399 	return err;
1400 }
1401 
1402 /*
1403  * Readahead a node page
1404  */
1405 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1406 {
1407 	struct page *apage;
1408 	int err;
1409 
1410 	if (!nid)
1411 		return;
1412 	if (f2fs_check_nid_range(sbi, nid))
1413 		return;
1414 
1415 	apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1416 	if (apage)
1417 		return;
1418 
1419 	apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1420 	if (!apage)
1421 		return;
1422 
1423 	err = read_node_page(apage, REQ_RAHEAD);
1424 	f2fs_put_page(apage, err ? 1 : 0);
1425 }
1426 
1427 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1428 					struct page *parent, int start)
1429 {
1430 	struct page *page;
1431 	int err;
1432 
1433 	if (!nid)
1434 		return ERR_PTR(-ENOENT);
1435 	if (f2fs_check_nid_range(sbi, nid))
1436 		return ERR_PTR(-EINVAL);
1437 repeat:
1438 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1439 	if (!page)
1440 		return ERR_PTR(-ENOMEM);
1441 
1442 	err = read_node_page(page, 0);
1443 	if (err < 0) {
1444 		goto out_put_err;
1445 	} else if (err == LOCKED_PAGE) {
1446 		err = 0;
1447 		goto page_hit;
1448 	}
1449 
1450 	if (parent)
1451 		f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1452 
1453 	lock_page(page);
1454 
1455 	if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1456 		f2fs_put_page(page, 1);
1457 		goto repeat;
1458 	}
1459 
1460 	if (unlikely(!PageUptodate(page))) {
1461 		err = -EIO;
1462 		goto out_err;
1463 	}
1464 
1465 	if (!f2fs_inode_chksum_verify(sbi, page)) {
1466 		err = -EFSBADCRC;
1467 		goto out_err;
1468 	}
1469 page_hit:
1470 	if (likely(nid == nid_of_node(page)))
1471 		return page;
1472 
1473 	f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1474 			  nid, nid_of_node(page), ino_of_node(page),
1475 			  ofs_of_node(page), cpver_of_node(page),
1476 			  next_blkaddr_of_node(page));
1477 	set_sbi_flag(sbi, SBI_NEED_FSCK);
1478 	f2fs_handle_error(sbi, ERROR_INCONSISTENT_FOOTER);
1479 	err = -EFSCORRUPTED;
1480 out_err:
1481 	ClearPageUptodate(page);
1482 out_put_err:
1483 	/* ENOENT comes from read_node_page which is not an error. */
1484 	if (err != -ENOENT)
1485 		f2fs_handle_page_eio(sbi, page->index, NODE);
1486 	f2fs_put_page(page, 1);
1487 	return ERR_PTR(err);
1488 }
1489 
1490 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1491 {
1492 	return __get_node_page(sbi, nid, NULL, 0);
1493 }
1494 
1495 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1496 {
1497 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1498 	nid_t nid = get_nid(parent, start, false);
1499 
1500 	return __get_node_page(sbi, nid, parent, start);
1501 }
1502 
1503 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1504 {
1505 	struct inode *inode;
1506 	struct page *page;
1507 	int ret;
1508 
1509 	/* should flush inline_data before evict_inode */
1510 	inode = ilookup(sbi->sb, ino);
1511 	if (!inode)
1512 		return;
1513 
1514 	page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1515 					FGP_LOCK|FGP_NOWAIT, 0);
1516 	if (!page)
1517 		goto iput_out;
1518 
1519 	if (!PageUptodate(page))
1520 		goto page_out;
1521 
1522 	if (!PageDirty(page))
1523 		goto page_out;
1524 
1525 	if (!clear_page_dirty_for_io(page))
1526 		goto page_out;
1527 
1528 	ret = f2fs_write_inline_data(inode, page);
1529 	inode_dec_dirty_pages(inode);
1530 	f2fs_remove_dirty_inode(inode);
1531 	if (ret)
1532 		set_page_dirty(page);
1533 page_out:
1534 	f2fs_put_page(page, 1);
1535 iput_out:
1536 	iput(inode);
1537 }
1538 
1539 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1540 {
1541 	pgoff_t index;
1542 	struct folio_batch fbatch;
1543 	struct page *last_page = NULL;
1544 	int nr_folios;
1545 
1546 	folio_batch_init(&fbatch);
1547 	index = 0;
1548 
1549 	while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1550 					(pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1551 					&fbatch))) {
1552 		int i;
1553 
1554 		for (i = 0; i < nr_folios; i++) {
1555 			struct page *page = &fbatch.folios[i]->page;
1556 
1557 			if (unlikely(f2fs_cp_error(sbi))) {
1558 				f2fs_put_page(last_page, 0);
1559 				folio_batch_release(&fbatch);
1560 				return ERR_PTR(-EIO);
1561 			}
1562 
1563 			if (!IS_DNODE(page) || !is_cold_node(page))
1564 				continue;
1565 			if (ino_of_node(page) != ino)
1566 				continue;
1567 
1568 			lock_page(page);
1569 
1570 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1571 continue_unlock:
1572 				unlock_page(page);
1573 				continue;
1574 			}
1575 			if (ino_of_node(page) != ino)
1576 				goto continue_unlock;
1577 
1578 			if (!PageDirty(page)) {
1579 				/* someone wrote it for us */
1580 				goto continue_unlock;
1581 			}
1582 
1583 			if (last_page)
1584 				f2fs_put_page(last_page, 0);
1585 
1586 			get_page(page);
1587 			last_page = page;
1588 			unlock_page(page);
1589 		}
1590 		folio_batch_release(&fbatch);
1591 		cond_resched();
1592 	}
1593 	return last_page;
1594 }
1595 
1596 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1597 				struct writeback_control *wbc, bool do_balance,
1598 				enum iostat_type io_type, unsigned int *seq_id)
1599 {
1600 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1601 	nid_t nid;
1602 	struct node_info ni;
1603 	struct f2fs_io_info fio = {
1604 		.sbi = sbi,
1605 		.ino = ino_of_node(page),
1606 		.type = NODE,
1607 		.op = REQ_OP_WRITE,
1608 		.op_flags = wbc_to_write_flags(wbc),
1609 		.page = page,
1610 		.encrypted_page = NULL,
1611 		.submitted = 0,
1612 		.io_type = io_type,
1613 		.io_wbc = wbc,
1614 	};
1615 	unsigned int seq;
1616 
1617 	trace_f2fs_writepage(page, NODE);
1618 
1619 	if (unlikely(f2fs_cp_error(sbi))) {
1620 		/* keep node pages in remount-ro mode */
1621 		if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY)
1622 			goto redirty_out;
1623 		ClearPageUptodate(page);
1624 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1625 		unlock_page(page);
1626 		return 0;
1627 	}
1628 
1629 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1630 		goto redirty_out;
1631 
1632 	if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1633 			wbc->sync_mode == WB_SYNC_NONE &&
1634 			IS_DNODE(page) && is_cold_node(page))
1635 		goto redirty_out;
1636 
1637 	/* get old block addr of this node page */
1638 	nid = nid_of_node(page);
1639 	f2fs_bug_on(sbi, page->index != nid);
1640 
1641 	if (f2fs_get_node_info(sbi, nid, &ni, !do_balance))
1642 		goto redirty_out;
1643 
1644 	if (wbc->for_reclaim) {
1645 		if (!f2fs_down_read_trylock(&sbi->node_write))
1646 			goto redirty_out;
1647 	} else {
1648 		f2fs_down_read(&sbi->node_write);
1649 	}
1650 
1651 	/* This page is already truncated */
1652 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1653 		ClearPageUptodate(page);
1654 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1655 		f2fs_up_read(&sbi->node_write);
1656 		unlock_page(page);
1657 		return 0;
1658 	}
1659 
1660 	if (__is_valid_data_blkaddr(ni.blk_addr) &&
1661 		!f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1662 					DATA_GENERIC_ENHANCE)) {
1663 		f2fs_up_read(&sbi->node_write);
1664 		goto redirty_out;
1665 	}
1666 
1667 	if (atomic && !test_opt(sbi, NOBARRIER) && !f2fs_sb_has_blkzoned(sbi))
1668 		fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1669 
1670 	/* should add to global list before clearing PAGECACHE status */
1671 	if (f2fs_in_warm_node_list(sbi, page)) {
1672 		seq = f2fs_add_fsync_node_entry(sbi, page);
1673 		if (seq_id)
1674 			*seq_id = seq;
1675 	}
1676 
1677 	set_page_writeback(page);
1678 
1679 	fio.old_blkaddr = ni.blk_addr;
1680 	f2fs_do_write_node_page(nid, &fio);
1681 	set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1682 	dec_page_count(sbi, F2FS_DIRTY_NODES);
1683 	f2fs_up_read(&sbi->node_write);
1684 
1685 	if (wbc->for_reclaim) {
1686 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1687 		submitted = NULL;
1688 	}
1689 
1690 	unlock_page(page);
1691 
1692 	if (unlikely(f2fs_cp_error(sbi))) {
1693 		f2fs_submit_merged_write(sbi, NODE);
1694 		submitted = NULL;
1695 	}
1696 	if (submitted)
1697 		*submitted = fio.submitted;
1698 
1699 	if (do_balance)
1700 		f2fs_balance_fs(sbi, false);
1701 	return 0;
1702 
1703 redirty_out:
1704 	redirty_page_for_writepage(wbc, page);
1705 	return AOP_WRITEPAGE_ACTIVATE;
1706 }
1707 
1708 int f2fs_move_node_page(struct page *node_page, int gc_type)
1709 {
1710 	int err = 0;
1711 
1712 	if (gc_type == FG_GC) {
1713 		struct writeback_control wbc = {
1714 			.sync_mode = WB_SYNC_ALL,
1715 			.nr_to_write = 1,
1716 			.for_reclaim = 0,
1717 		};
1718 
1719 		f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1720 
1721 		set_page_dirty(node_page);
1722 
1723 		if (!clear_page_dirty_for_io(node_page)) {
1724 			err = -EAGAIN;
1725 			goto out_page;
1726 		}
1727 
1728 		if (__write_node_page(node_page, false, NULL,
1729 					&wbc, false, FS_GC_NODE_IO, NULL)) {
1730 			err = -EAGAIN;
1731 			unlock_page(node_page);
1732 		}
1733 		goto release_page;
1734 	} else {
1735 		/* set page dirty and write it */
1736 		if (!PageWriteback(node_page))
1737 			set_page_dirty(node_page);
1738 	}
1739 out_page:
1740 	unlock_page(node_page);
1741 release_page:
1742 	f2fs_put_page(node_page, 0);
1743 	return err;
1744 }
1745 
1746 static int f2fs_write_node_page(struct page *page,
1747 				struct writeback_control *wbc)
1748 {
1749 	return __write_node_page(page, false, NULL, wbc, false,
1750 						FS_NODE_IO, NULL);
1751 }
1752 
1753 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1754 			struct writeback_control *wbc, bool atomic,
1755 			unsigned int *seq_id)
1756 {
1757 	pgoff_t index;
1758 	struct folio_batch fbatch;
1759 	int ret = 0;
1760 	struct page *last_page = NULL;
1761 	bool marked = false;
1762 	nid_t ino = inode->i_ino;
1763 	int nr_folios;
1764 	int nwritten = 0;
1765 
1766 	if (atomic) {
1767 		last_page = last_fsync_dnode(sbi, ino);
1768 		if (IS_ERR_OR_NULL(last_page))
1769 			return PTR_ERR_OR_ZERO(last_page);
1770 	}
1771 retry:
1772 	folio_batch_init(&fbatch);
1773 	index = 0;
1774 
1775 	while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1776 					(pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1777 					&fbatch))) {
1778 		int i;
1779 
1780 		for (i = 0; i < nr_folios; i++) {
1781 			struct page *page = &fbatch.folios[i]->page;
1782 			bool submitted = false;
1783 
1784 			if (unlikely(f2fs_cp_error(sbi))) {
1785 				f2fs_put_page(last_page, 0);
1786 				folio_batch_release(&fbatch);
1787 				ret = -EIO;
1788 				goto out;
1789 			}
1790 
1791 			if (!IS_DNODE(page) || !is_cold_node(page))
1792 				continue;
1793 			if (ino_of_node(page) != ino)
1794 				continue;
1795 
1796 			lock_page(page);
1797 
1798 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1799 continue_unlock:
1800 				unlock_page(page);
1801 				continue;
1802 			}
1803 			if (ino_of_node(page) != ino)
1804 				goto continue_unlock;
1805 
1806 			if (!PageDirty(page) && page != last_page) {
1807 				/* someone wrote it for us */
1808 				goto continue_unlock;
1809 			}
1810 
1811 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1812 
1813 			set_fsync_mark(page, 0);
1814 			set_dentry_mark(page, 0);
1815 
1816 			if (!atomic || page == last_page) {
1817 				set_fsync_mark(page, 1);
1818 				percpu_counter_inc(&sbi->rf_node_block_count);
1819 				if (IS_INODE(page)) {
1820 					if (is_inode_flag_set(inode,
1821 								FI_DIRTY_INODE))
1822 						f2fs_update_inode(inode, page);
1823 					set_dentry_mark(page,
1824 						f2fs_need_dentry_mark(sbi, ino));
1825 				}
1826 				/* may be written by other thread */
1827 				if (!PageDirty(page))
1828 					set_page_dirty(page);
1829 			}
1830 
1831 			if (!clear_page_dirty_for_io(page))
1832 				goto continue_unlock;
1833 
1834 			ret = __write_node_page(page, atomic &&
1835 						page == last_page,
1836 						&submitted, wbc, true,
1837 						FS_NODE_IO, seq_id);
1838 			if (ret) {
1839 				unlock_page(page);
1840 				f2fs_put_page(last_page, 0);
1841 				break;
1842 			} else if (submitted) {
1843 				nwritten++;
1844 			}
1845 
1846 			if (page == last_page) {
1847 				f2fs_put_page(page, 0);
1848 				marked = true;
1849 				break;
1850 			}
1851 		}
1852 		folio_batch_release(&fbatch);
1853 		cond_resched();
1854 
1855 		if (ret || marked)
1856 			break;
1857 	}
1858 	if (!ret && atomic && !marked) {
1859 		f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1860 			   ino, last_page->index);
1861 		lock_page(last_page);
1862 		f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1863 		set_page_dirty(last_page);
1864 		unlock_page(last_page);
1865 		goto retry;
1866 	}
1867 out:
1868 	if (nwritten)
1869 		f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1870 	return ret ? -EIO : 0;
1871 }
1872 
1873 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1874 {
1875 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1876 	bool clean;
1877 
1878 	if (inode->i_ino != ino)
1879 		return 0;
1880 
1881 	if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1882 		return 0;
1883 
1884 	spin_lock(&sbi->inode_lock[DIRTY_META]);
1885 	clean = list_empty(&F2FS_I(inode)->gdirty_list);
1886 	spin_unlock(&sbi->inode_lock[DIRTY_META]);
1887 
1888 	if (clean)
1889 		return 0;
1890 
1891 	inode = igrab(inode);
1892 	if (!inode)
1893 		return 0;
1894 	return 1;
1895 }
1896 
1897 static bool flush_dirty_inode(struct page *page)
1898 {
1899 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1900 	struct inode *inode;
1901 	nid_t ino = ino_of_node(page);
1902 
1903 	inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1904 	if (!inode)
1905 		return false;
1906 
1907 	f2fs_update_inode(inode, page);
1908 	unlock_page(page);
1909 
1910 	iput(inode);
1911 	return true;
1912 }
1913 
1914 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1915 {
1916 	pgoff_t index = 0;
1917 	struct folio_batch fbatch;
1918 	int nr_folios;
1919 
1920 	folio_batch_init(&fbatch);
1921 
1922 	while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1923 					(pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1924 					&fbatch))) {
1925 		int i;
1926 
1927 		for (i = 0; i < nr_folios; i++) {
1928 			struct page *page = &fbatch.folios[i]->page;
1929 
1930 			if (!IS_INODE(page))
1931 				continue;
1932 
1933 			lock_page(page);
1934 
1935 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1936 continue_unlock:
1937 				unlock_page(page);
1938 				continue;
1939 			}
1940 
1941 			if (!PageDirty(page)) {
1942 				/* someone wrote it for us */
1943 				goto continue_unlock;
1944 			}
1945 
1946 			/* flush inline_data, if it's async context. */
1947 			if (page_private_inline(page)) {
1948 				clear_page_private_inline(page);
1949 				unlock_page(page);
1950 				flush_inline_data(sbi, ino_of_node(page));
1951 				continue;
1952 			}
1953 			unlock_page(page);
1954 		}
1955 		folio_batch_release(&fbatch);
1956 		cond_resched();
1957 	}
1958 }
1959 
1960 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1961 				struct writeback_control *wbc,
1962 				bool do_balance, enum iostat_type io_type)
1963 {
1964 	pgoff_t index;
1965 	struct folio_batch fbatch;
1966 	int step = 0;
1967 	int nwritten = 0;
1968 	int ret = 0;
1969 	int nr_folios, done = 0;
1970 
1971 	folio_batch_init(&fbatch);
1972 
1973 next_step:
1974 	index = 0;
1975 
1976 	while (!done && (nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi),
1977 				&index, (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1978 				&fbatch))) {
1979 		int i;
1980 
1981 		for (i = 0; i < nr_folios; i++) {
1982 			struct page *page = &fbatch.folios[i]->page;
1983 			bool submitted = false;
1984 
1985 			/* give a priority to WB_SYNC threads */
1986 			if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1987 					wbc->sync_mode == WB_SYNC_NONE) {
1988 				done = 1;
1989 				break;
1990 			}
1991 
1992 			/*
1993 			 * flushing sequence with step:
1994 			 * 0. indirect nodes
1995 			 * 1. dentry dnodes
1996 			 * 2. file dnodes
1997 			 */
1998 			if (step == 0 && IS_DNODE(page))
1999 				continue;
2000 			if (step == 1 && (!IS_DNODE(page) ||
2001 						is_cold_node(page)))
2002 				continue;
2003 			if (step == 2 && (!IS_DNODE(page) ||
2004 						!is_cold_node(page)))
2005 				continue;
2006 lock_node:
2007 			if (wbc->sync_mode == WB_SYNC_ALL)
2008 				lock_page(page);
2009 			else if (!trylock_page(page))
2010 				continue;
2011 
2012 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
2013 continue_unlock:
2014 				unlock_page(page);
2015 				continue;
2016 			}
2017 
2018 			if (!PageDirty(page)) {
2019 				/* someone wrote it for us */
2020 				goto continue_unlock;
2021 			}
2022 
2023 			/* flush inline_data/inode, if it's async context. */
2024 			if (!do_balance)
2025 				goto write_node;
2026 
2027 			/* flush inline_data */
2028 			if (page_private_inline(page)) {
2029 				clear_page_private_inline(page);
2030 				unlock_page(page);
2031 				flush_inline_data(sbi, ino_of_node(page));
2032 				goto lock_node;
2033 			}
2034 
2035 			/* flush dirty inode */
2036 			if (IS_INODE(page) && flush_dirty_inode(page))
2037 				goto lock_node;
2038 write_node:
2039 			f2fs_wait_on_page_writeback(page, NODE, true, true);
2040 
2041 			if (!clear_page_dirty_for_io(page))
2042 				goto continue_unlock;
2043 
2044 			set_fsync_mark(page, 0);
2045 			set_dentry_mark(page, 0);
2046 
2047 			ret = __write_node_page(page, false, &submitted,
2048 						wbc, do_balance, io_type, NULL);
2049 			if (ret)
2050 				unlock_page(page);
2051 			else if (submitted)
2052 				nwritten++;
2053 
2054 			if (--wbc->nr_to_write == 0)
2055 				break;
2056 		}
2057 		folio_batch_release(&fbatch);
2058 		cond_resched();
2059 
2060 		if (wbc->nr_to_write == 0) {
2061 			step = 2;
2062 			break;
2063 		}
2064 	}
2065 
2066 	if (step < 2) {
2067 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2068 				wbc->sync_mode == WB_SYNC_NONE && step == 1)
2069 			goto out;
2070 		step++;
2071 		goto next_step;
2072 	}
2073 out:
2074 	if (nwritten)
2075 		f2fs_submit_merged_write(sbi, NODE);
2076 
2077 	if (unlikely(f2fs_cp_error(sbi)))
2078 		return -EIO;
2079 	return ret;
2080 }
2081 
2082 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2083 						unsigned int seq_id)
2084 {
2085 	struct fsync_node_entry *fn;
2086 	struct page *page;
2087 	struct list_head *head = &sbi->fsync_node_list;
2088 	unsigned long flags;
2089 	unsigned int cur_seq_id = 0;
2090 
2091 	while (seq_id && cur_seq_id < seq_id) {
2092 		spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2093 		if (list_empty(head)) {
2094 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2095 			break;
2096 		}
2097 		fn = list_first_entry(head, struct fsync_node_entry, list);
2098 		if (fn->seq_id > seq_id) {
2099 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2100 			break;
2101 		}
2102 		cur_seq_id = fn->seq_id;
2103 		page = fn->page;
2104 		get_page(page);
2105 		spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2106 
2107 		f2fs_wait_on_page_writeback(page, NODE, true, false);
2108 
2109 		put_page(page);
2110 	}
2111 
2112 	return filemap_check_errors(NODE_MAPPING(sbi));
2113 }
2114 
2115 static int f2fs_write_node_pages(struct address_space *mapping,
2116 			    struct writeback_control *wbc)
2117 {
2118 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2119 	struct blk_plug plug;
2120 	long diff;
2121 
2122 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2123 		goto skip_write;
2124 
2125 	/* balancing f2fs's metadata in background */
2126 	f2fs_balance_fs_bg(sbi, true);
2127 
2128 	/* collect a number of dirty node pages and write together */
2129 	if (wbc->sync_mode != WB_SYNC_ALL &&
2130 			get_pages(sbi, F2FS_DIRTY_NODES) <
2131 					nr_pages_to_skip(sbi, NODE))
2132 		goto skip_write;
2133 
2134 	if (wbc->sync_mode == WB_SYNC_ALL)
2135 		atomic_inc(&sbi->wb_sync_req[NODE]);
2136 	else if (atomic_read(&sbi->wb_sync_req[NODE])) {
2137 		/* to avoid potential deadlock */
2138 		if (current->plug)
2139 			blk_finish_plug(current->plug);
2140 		goto skip_write;
2141 	}
2142 
2143 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2144 
2145 	diff = nr_pages_to_write(sbi, NODE, wbc);
2146 	blk_start_plug(&plug);
2147 	f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2148 	blk_finish_plug(&plug);
2149 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2150 
2151 	if (wbc->sync_mode == WB_SYNC_ALL)
2152 		atomic_dec(&sbi->wb_sync_req[NODE]);
2153 	return 0;
2154 
2155 skip_write:
2156 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2157 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2158 	return 0;
2159 }
2160 
2161 static bool f2fs_dirty_node_folio(struct address_space *mapping,
2162 		struct folio *folio)
2163 {
2164 	trace_f2fs_set_page_dirty(&folio->page, NODE);
2165 
2166 	if (!folio_test_uptodate(folio))
2167 		folio_mark_uptodate(folio);
2168 #ifdef CONFIG_F2FS_CHECK_FS
2169 	if (IS_INODE(&folio->page))
2170 		f2fs_inode_chksum_set(F2FS_M_SB(mapping), &folio->page);
2171 #endif
2172 	if (filemap_dirty_folio(mapping, folio)) {
2173 		inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
2174 		set_page_private_reference(&folio->page);
2175 		return true;
2176 	}
2177 	return false;
2178 }
2179 
2180 /*
2181  * Structure of the f2fs node operations
2182  */
2183 const struct address_space_operations f2fs_node_aops = {
2184 	.writepage	= f2fs_write_node_page,
2185 	.writepages	= f2fs_write_node_pages,
2186 	.dirty_folio	= f2fs_dirty_node_folio,
2187 	.invalidate_folio = f2fs_invalidate_folio,
2188 	.release_folio	= f2fs_release_folio,
2189 	.migrate_folio	= filemap_migrate_folio,
2190 };
2191 
2192 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2193 						nid_t n)
2194 {
2195 	return radix_tree_lookup(&nm_i->free_nid_root, n);
2196 }
2197 
2198 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2199 				struct free_nid *i)
2200 {
2201 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2202 	int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2203 
2204 	if (err)
2205 		return err;
2206 
2207 	nm_i->nid_cnt[FREE_NID]++;
2208 	list_add_tail(&i->list, &nm_i->free_nid_list);
2209 	return 0;
2210 }
2211 
2212 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2213 			struct free_nid *i, enum nid_state state)
2214 {
2215 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2216 
2217 	f2fs_bug_on(sbi, state != i->state);
2218 	nm_i->nid_cnt[state]--;
2219 	if (state == FREE_NID)
2220 		list_del(&i->list);
2221 	radix_tree_delete(&nm_i->free_nid_root, i->nid);
2222 }
2223 
2224 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2225 			enum nid_state org_state, enum nid_state dst_state)
2226 {
2227 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2228 
2229 	f2fs_bug_on(sbi, org_state != i->state);
2230 	i->state = dst_state;
2231 	nm_i->nid_cnt[org_state]--;
2232 	nm_i->nid_cnt[dst_state]++;
2233 
2234 	switch (dst_state) {
2235 	case PREALLOC_NID:
2236 		list_del(&i->list);
2237 		break;
2238 	case FREE_NID:
2239 		list_add_tail(&i->list, &nm_i->free_nid_list);
2240 		break;
2241 	default:
2242 		BUG_ON(1);
2243 	}
2244 }
2245 
2246 bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi)
2247 {
2248 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2249 	unsigned int i;
2250 	bool ret = true;
2251 
2252 	f2fs_down_read(&nm_i->nat_tree_lock);
2253 	for (i = 0; i < nm_i->nat_blocks; i++) {
2254 		if (!test_bit_le(i, nm_i->nat_block_bitmap)) {
2255 			ret = false;
2256 			break;
2257 		}
2258 	}
2259 	f2fs_up_read(&nm_i->nat_tree_lock);
2260 
2261 	return ret;
2262 }
2263 
2264 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2265 							bool set, bool build)
2266 {
2267 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2268 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2269 	unsigned int nid_ofs = nid - START_NID(nid);
2270 
2271 	if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2272 		return;
2273 
2274 	if (set) {
2275 		if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2276 			return;
2277 		__set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2278 		nm_i->free_nid_count[nat_ofs]++;
2279 	} else {
2280 		if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2281 			return;
2282 		__clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2283 		if (!build)
2284 			nm_i->free_nid_count[nat_ofs]--;
2285 	}
2286 }
2287 
2288 /* return if the nid is recognized as free */
2289 static bool add_free_nid(struct f2fs_sb_info *sbi,
2290 				nid_t nid, bool build, bool update)
2291 {
2292 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2293 	struct free_nid *i, *e;
2294 	struct nat_entry *ne;
2295 	int err = -EINVAL;
2296 	bool ret = false;
2297 
2298 	/* 0 nid should not be used */
2299 	if (unlikely(nid == 0))
2300 		return false;
2301 
2302 	if (unlikely(f2fs_check_nid_range(sbi, nid)))
2303 		return false;
2304 
2305 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL);
2306 	i->nid = nid;
2307 	i->state = FREE_NID;
2308 
2309 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2310 
2311 	spin_lock(&nm_i->nid_list_lock);
2312 
2313 	if (build) {
2314 		/*
2315 		 *   Thread A             Thread B
2316 		 *  - f2fs_create
2317 		 *   - f2fs_new_inode
2318 		 *    - f2fs_alloc_nid
2319 		 *     - __insert_nid_to_list(PREALLOC_NID)
2320 		 *                     - f2fs_balance_fs_bg
2321 		 *                      - f2fs_build_free_nids
2322 		 *                       - __f2fs_build_free_nids
2323 		 *                        - scan_nat_page
2324 		 *                         - add_free_nid
2325 		 *                          - __lookup_nat_cache
2326 		 *  - f2fs_add_link
2327 		 *   - f2fs_init_inode_metadata
2328 		 *    - f2fs_new_inode_page
2329 		 *     - f2fs_new_node_page
2330 		 *      - set_node_addr
2331 		 *  - f2fs_alloc_nid_done
2332 		 *   - __remove_nid_from_list(PREALLOC_NID)
2333 		 *                         - __insert_nid_to_list(FREE_NID)
2334 		 */
2335 		ne = __lookup_nat_cache(nm_i, nid);
2336 		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2337 				nat_get_blkaddr(ne) != NULL_ADDR))
2338 			goto err_out;
2339 
2340 		e = __lookup_free_nid_list(nm_i, nid);
2341 		if (e) {
2342 			if (e->state == FREE_NID)
2343 				ret = true;
2344 			goto err_out;
2345 		}
2346 	}
2347 	ret = true;
2348 	err = __insert_free_nid(sbi, i);
2349 err_out:
2350 	if (update) {
2351 		update_free_nid_bitmap(sbi, nid, ret, build);
2352 		if (!build)
2353 			nm_i->available_nids++;
2354 	}
2355 	spin_unlock(&nm_i->nid_list_lock);
2356 	radix_tree_preload_end();
2357 
2358 	if (err)
2359 		kmem_cache_free(free_nid_slab, i);
2360 	return ret;
2361 }
2362 
2363 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2364 {
2365 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2366 	struct free_nid *i;
2367 	bool need_free = false;
2368 
2369 	spin_lock(&nm_i->nid_list_lock);
2370 	i = __lookup_free_nid_list(nm_i, nid);
2371 	if (i && i->state == FREE_NID) {
2372 		__remove_free_nid(sbi, i, FREE_NID);
2373 		need_free = true;
2374 	}
2375 	spin_unlock(&nm_i->nid_list_lock);
2376 
2377 	if (need_free)
2378 		kmem_cache_free(free_nid_slab, i);
2379 }
2380 
2381 static int scan_nat_page(struct f2fs_sb_info *sbi,
2382 			struct page *nat_page, nid_t start_nid)
2383 {
2384 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2385 	struct f2fs_nat_block *nat_blk = page_address(nat_page);
2386 	block_t blk_addr;
2387 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2388 	int i;
2389 
2390 	__set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2391 
2392 	i = start_nid % NAT_ENTRY_PER_BLOCK;
2393 
2394 	for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2395 		if (unlikely(start_nid >= nm_i->max_nid))
2396 			break;
2397 
2398 		blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2399 
2400 		if (blk_addr == NEW_ADDR)
2401 			return -EFSCORRUPTED;
2402 
2403 		if (blk_addr == NULL_ADDR) {
2404 			add_free_nid(sbi, start_nid, true, true);
2405 		} else {
2406 			spin_lock(&NM_I(sbi)->nid_list_lock);
2407 			update_free_nid_bitmap(sbi, start_nid, false, true);
2408 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2409 		}
2410 	}
2411 
2412 	return 0;
2413 }
2414 
2415 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2416 {
2417 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2418 	struct f2fs_journal *journal = curseg->journal;
2419 	int i;
2420 
2421 	down_read(&curseg->journal_rwsem);
2422 	for (i = 0; i < nats_in_cursum(journal); i++) {
2423 		block_t addr;
2424 		nid_t nid;
2425 
2426 		addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2427 		nid = le32_to_cpu(nid_in_journal(journal, i));
2428 		if (addr == NULL_ADDR)
2429 			add_free_nid(sbi, nid, true, false);
2430 		else
2431 			remove_free_nid(sbi, nid);
2432 	}
2433 	up_read(&curseg->journal_rwsem);
2434 }
2435 
2436 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2437 {
2438 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2439 	unsigned int i, idx;
2440 	nid_t nid;
2441 
2442 	f2fs_down_read(&nm_i->nat_tree_lock);
2443 
2444 	for (i = 0; i < nm_i->nat_blocks; i++) {
2445 		if (!test_bit_le(i, nm_i->nat_block_bitmap))
2446 			continue;
2447 		if (!nm_i->free_nid_count[i])
2448 			continue;
2449 		for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2450 			idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2451 						NAT_ENTRY_PER_BLOCK, idx);
2452 			if (idx >= NAT_ENTRY_PER_BLOCK)
2453 				break;
2454 
2455 			nid = i * NAT_ENTRY_PER_BLOCK + idx;
2456 			add_free_nid(sbi, nid, true, false);
2457 
2458 			if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2459 				goto out;
2460 		}
2461 	}
2462 out:
2463 	scan_curseg_cache(sbi);
2464 
2465 	f2fs_up_read(&nm_i->nat_tree_lock);
2466 }
2467 
2468 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2469 						bool sync, bool mount)
2470 {
2471 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2472 	int i = 0, ret;
2473 	nid_t nid = nm_i->next_scan_nid;
2474 
2475 	if (unlikely(nid >= nm_i->max_nid))
2476 		nid = 0;
2477 
2478 	if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2479 		nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2480 
2481 	/* Enough entries */
2482 	if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2483 		return 0;
2484 
2485 	if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2486 		return 0;
2487 
2488 	if (!mount) {
2489 		/* try to find free nids in free_nid_bitmap */
2490 		scan_free_nid_bits(sbi);
2491 
2492 		if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2493 			return 0;
2494 	}
2495 
2496 	/* readahead nat pages to be scanned */
2497 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2498 							META_NAT, true);
2499 
2500 	f2fs_down_read(&nm_i->nat_tree_lock);
2501 
2502 	while (1) {
2503 		if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2504 						nm_i->nat_block_bitmap)) {
2505 			struct page *page = get_current_nat_page(sbi, nid);
2506 
2507 			if (IS_ERR(page)) {
2508 				ret = PTR_ERR(page);
2509 			} else {
2510 				ret = scan_nat_page(sbi, page, nid);
2511 				f2fs_put_page(page, 1);
2512 			}
2513 
2514 			if (ret) {
2515 				f2fs_up_read(&nm_i->nat_tree_lock);
2516 
2517 				if (ret == -EFSCORRUPTED) {
2518 					f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2519 					set_sbi_flag(sbi, SBI_NEED_FSCK);
2520 					f2fs_handle_error(sbi,
2521 						ERROR_INCONSISTENT_NAT);
2522 				}
2523 
2524 				return ret;
2525 			}
2526 		}
2527 
2528 		nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2529 		if (unlikely(nid >= nm_i->max_nid))
2530 			nid = 0;
2531 
2532 		if (++i >= FREE_NID_PAGES)
2533 			break;
2534 	}
2535 
2536 	/* go to the next free nat pages to find free nids abundantly */
2537 	nm_i->next_scan_nid = nid;
2538 
2539 	/* find free nids from current sum_pages */
2540 	scan_curseg_cache(sbi);
2541 
2542 	f2fs_up_read(&nm_i->nat_tree_lock);
2543 
2544 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2545 					nm_i->ra_nid_pages, META_NAT, false);
2546 
2547 	return 0;
2548 }
2549 
2550 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2551 {
2552 	int ret;
2553 
2554 	mutex_lock(&NM_I(sbi)->build_lock);
2555 	ret = __f2fs_build_free_nids(sbi, sync, mount);
2556 	mutex_unlock(&NM_I(sbi)->build_lock);
2557 
2558 	return ret;
2559 }
2560 
2561 /*
2562  * If this function returns success, caller can obtain a new nid
2563  * from second parameter of this function.
2564  * The returned nid could be used ino as well as nid when inode is created.
2565  */
2566 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2567 {
2568 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2569 	struct free_nid *i = NULL;
2570 retry:
2571 	if (time_to_inject(sbi, FAULT_ALLOC_NID))
2572 		return false;
2573 
2574 	spin_lock(&nm_i->nid_list_lock);
2575 
2576 	if (unlikely(nm_i->available_nids == 0)) {
2577 		spin_unlock(&nm_i->nid_list_lock);
2578 		return false;
2579 	}
2580 
2581 	/* We should not use stale free nids created by f2fs_build_free_nids */
2582 	if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2583 		f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2584 		i = list_first_entry(&nm_i->free_nid_list,
2585 					struct free_nid, list);
2586 		*nid = i->nid;
2587 
2588 		__move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2589 		nm_i->available_nids--;
2590 
2591 		update_free_nid_bitmap(sbi, *nid, false, false);
2592 
2593 		spin_unlock(&nm_i->nid_list_lock);
2594 		return true;
2595 	}
2596 	spin_unlock(&nm_i->nid_list_lock);
2597 
2598 	/* Let's scan nat pages and its caches to get free nids */
2599 	if (!f2fs_build_free_nids(sbi, true, false))
2600 		goto retry;
2601 	return false;
2602 }
2603 
2604 /*
2605  * f2fs_alloc_nid() should be called prior to this function.
2606  */
2607 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2608 {
2609 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2610 	struct free_nid *i;
2611 
2612 	spin_lock(&nm_i->nid_list_lock);
2613 	i = __lookup_free_nid_list(nm_i, nid);
2614 	f2fs_bug_on(sbi, !i);
2615 	__remove_free_nid(sbi, i, PREALLOC_NID);
2616 	spin_unlock(&nm_i->nid_list_lock);
2617 
2618 	kmem_cache_free(free_nid_slab, i);
2619 }
2620 
2621 /*
2622  * f2fs_alloc_nid() should be called prior to this function.
2623  */
2624 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2625 {
2626 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2627 	struct free_nid *i;
2628 	bool need_free = false;
2629 
2630 	if (!nid)
2631 		return;
2632 
2633 	spin_lock(&nm_i->nid_list_lock);
2634 	i = __lookup_free_nid_list(nm_i, nid);
2635 	f2fs_bug_on(sbi, !i);
2636 
2637 	if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2638 		__remove_free_nid(sbi, i, PREALLOC_NID);
2639 		need_free = true;
2640 	} else {
2641 		__move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2642 	}
2643 
2644 	nm_i->available_nids++;
2645 
2646 	update_free_nid_bitmap(sbi, nid, true, false);
2647 
2648 	spin_unlock(&nm_i->nid_list_lock);
2649 
2650 	if (need_free)
2651 		kmem_cache_free(free_nid_slab, i);
2652 }
2653 
2654 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2655 {
2656 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2657 	int nr = nr_shrink;
2658 
2659 	if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2660 		return 0;
2661 
2662 	if (!mutex_trylock(&nm_i->build_lock))
2663 		return 0;
2664 
2665 	while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2666 		struct free_nid *i, *next;
2667 		unsigned int batch = SHRINK_NID_BATCH_SIZE;
2668 
2669 		spin_lock(&nm_i->nid_list_lock);
2670 		list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2671 			if (!nr_shrink || !batch ||
2672 				nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2673 				break;
2674 			__remove_free_nid(sbi, i, FREE_NID);
2675 			kmem_cache_free(free_nid_slab, i);
2676 			nr_shrink--;
2677 			batch--;
2678 		}
2679 		spin_unlock(&nm_i->nid_list_lock);
2680 	}
2681 
2682 	mutex_unlock(&nm_i->build_lock);
2683 
2684 	return nr - nr_shrink;
2685 }
2686 
2687 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2688 {
2689 	void *src_addr, *dst_addr;
2690 	size_t inline_size;
2691 	struct page *ipage;
2692 	struct f2fs_inode *ri;
2693 
2694 	ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2695 	if (IS_ERR(ipage))
2696 		return PTR_ERR(ipage);
2697 
2698 	ri = F2FS_INODE(page);
2699 	if (ri->i_inline & F2FS_INLINE_XATTR) {
2700 		if (!f2fs_has_inline_xattr(inode)) {
2701 			set_inode_flag(inode, FI_INLINE_XATTR);
2702 			stat_inc_inline_xattr(inode);
2703 		}
2704 	} else {
2705 		if (f2fs_has_inline_xattr(inode)) {
2706 			stat_dec_inline_xattr(inode);
2707 			clear_inode_flag(inode, FI_INLINE_XATTR);
2708 		}
2709 		goto update_inode;
2710 	}
2711 
2712 	dst_addr = inline_xattr_addr(inode, ipage);
2713 	src_addr = inline_xattr_addr(inode, page);
2714 	inline_size = inline_xattr_size(inode);
2715 
2716 	f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2717 	memcpy(dst_addr, src_addr, inline_size);
2718 update_inode:
2719 	f2fs_update_inode(inode, ipage);
2720 	f2fs_put_page(ipage, 1);
2721 	return 0;
2722 }
2723 
2724 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2725 {
2726 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2727 	nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2728 	nid_t new_xnid;
2729 	struct dnode_of_data dn;
2730 	struct node_info ni;
2731 	struct page *xpage;
2732 	int err;
2733 
2734 	if (!prev_xnid)
2735 		goto recover_xnid;
2736 
2737 	/* 1: invalidate the previous xattr nid */
2738 	err = f2fs_get_node_info(sbi, prev_xnid, &ni, false);
2739 	if (err)
2740 		return err;
2741 
2742 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
2743 	dec_valid_node_count(sbi, inode, false);
2744 	set_node_addr(sbi, &ni, NULL_ADDR, false);
2745 
2746 recover_xnid:
2747 	/* 2: update xattr nid in inode */
2748 	if (!f2fs_alloc_nid(sbi, &new_xnid))
2749 		return -ENOSPC;
2750 
2751 	set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2752 	xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2753 	if (IS_ERR(xpage)) {
2754 		f2fs_alloc_nid_failed(sbi, new_xnid);
2755 		return PTR_ERR(xpage);
2756 	}
2757 
2758 	f2fs_alloc_nid_done(sbi, new_xnid);
2759 	f2fs_update_inode_page(inode);
2760 
2761 	/* 3: update and set xattr node page dirty */
2762 	if (page) {
2763 		memcpy(F2FS_NODE(xpage), F2FS_NODE(page),
2764 				VALID_XATTR_BLOCK_SIZE);
2765 		set_page_dirty(xpage);
2766 	}
2767 	f2fs_put_page(xpage, 1);
2768 
2769 	return 0;
2770 }
2771 
2772 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2773 {
2774 	struct f2fs_inode *src, *dst;
2775 	nid_t ino = ino_of_node(page);
2776 	struct node_info old_ni, new_ni;
2777 	struct page *ipage;
2778 	int err;
2779 
2780 	err = f2fs_get_node_info(sbi, ino, &old_ni, false);
2781 	if (err)
2782 		return err;
2783 
2784 	if (unlikely(old_ni.blk_addr != NULL_ADDR))
2785 		return -EINVAL;
2786 retry:
2787 	ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2788 	if (!ipage) {
2789 		memalloc_retry_wait(GFP_NOFS);
2790 		goto retry;
2791 	}
2792 
2793 	/* Should not use this inode from free nid list */
2794 	remove_free_nid(sbi, ino);
2795 
2796 	if (!PageUptodate(ipage))
2797 		SetPageUptodate(ipage);
2798 	fill_node_footer(ipage, ino, ino, 0, true);
2799 	set_cold_node(ipage, false);
2800 
2801 	src = F2FS_INODE(page);
2802 	dst = F2FS_INODE(ipage);
2803 
2804 	memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2805 	dst->i_size = 0;
2806 	dst->i_blocks = cpu_to_le64(1);
2807 	dst->i_links = cpu_to_le32(1);
2808 	dst->i_xattr_nid = 0;
2809 	dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2810 	if (dst->i_inline & F2FS_EXTRA_ATTR) {
2811 		dst->i_extra_isize = src->i_extra_isize;
2812 
2813 		if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2814 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2815 							i_inline_xattr_size))
2816 			dst->i_inline_xattr_size = src->i_inline_xattr_size;
2817 
2818 		if (f2fs_sb_has_project_quota(sbi) &&
2819 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2820 								i_projid))
2821 			dst->i_projid = src->i_projid;
2822 
2823 		if (f2fs_sb_has_inode_crtime(sbi) &&
2824 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2825 							i_crtime_nsec)) {
2826 			dst->i_crtime = src->i_crtime;
2827 			dst->i_crtime_nsec = src->i_crtime_nsec;
2828 		}
2829 	}
2830 
2831 	new_ni = old_ni;
2832 	new_ni.ino = ino;
2833 
2834 	if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2835 		WARN_ON(1);
2836 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2837 	inc_valid_inode_count(sbi);
2838 	set_page_dirty(ipage);
2839 	f2fs_put_page(ipage, 1);
2840 	return 0;
2841 }
2842 
2843 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2844 			unsigned int segno, struct f2fs_summary_block *sum)
2845 {
2846 	struct f2fs_node *rn;
2847 	struct f2fs_summary *sum_entry;
2848 	block_t addr;
2849 	int i, idx, last_offset, nrpages;
2850 
2851 	/* scan the node segment */
2852 	last_offset = BLKS_PER_SEG(sbi);
2853 	addr = START_BLOCK(sbi, segno);
2854 	sum_entry = &sum->entries[0];
2855 
2856 	for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2857 		nrpages = bio_max_segs(last_offset - i);
2858 
2859 		/* readahead node pages */
2860 		f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2861 
2862 		for (idx = addr; idx < addr + nrpages; idx++) {
2863 			struct page *page = f2fs_get_tmp_page(sbi, idx);
2864 
2865 			if (IS_ERR(page))
2866 				return PTR_ERR(page);
2867 
2868 			rn = F2FS_NODE(page);
2869 			sum_entry->nid = rn->footer.nid;
2870 			sum_entry->version = 0;
2871 			sum_entry->ofs_in_node = 0;
2872 			sum_entry++;
2873 			f2fs_put_page(page, 1);
2874 		}
2875 
2876 		invalidate_mapping_pages(META_MAPPING(sbi), addr,
2877 							addr + nrpages);
2878 	}
2879 	return 0;
2880 }
2881 
2882 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2883 {
2884 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2885 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2886 	struct f2fs_journal *journal = curseg->journal;
2887 	int i;
2888 
2889 	down_write(&curseg->journal_rwsem);
2890 	for (i = 0; i < nats_in_cursum(journal); i++) {
2891 		struct nat_entry *ne;
2892 		struct f2fs_nat_entry raw_ne;
2893 		nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2894 
2895 		if (f2fs_check_nid_range(sbi, nid))
2896 			continue;
2897 
2898 		raw_ne = nat_in_journal(journal, i);
2899 
2900 		ne = __lookup_nat_cache(nm_i, nid);
2901 		if (!ne) {
2902 			ne = __alloc_nat_entry(sbi, nid, true);
2903 			__init_nat_entry(nm_i, ne, &raw_ne, true);
2904 		}
2905 
2906 		/*
2907 		 * if a free nat in journal has not been used after last
2908 		 * checkpoint, we should remove it from available nids,
2909 		 * since later we will add it again.
2910 		 */
2911 		if (!get_nat_flag(ne, IS_DIRTY) &&
2912 				le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2913 			spin_lock(&nm_i->nid_list_lock);
2914 			nm_i->available_nids--;
2915 			spin_unlock(&nm_i->nid_list_lock);
2916 		}
2917 
2918 		__set_nat_cache_dirty(nm_i, ne);
2919 	}
2920 	update_nats_in_cursum(journal, -i);
2921 	up_write(&curseg->journal_rwsem);
2922 }
2923 
2924 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2925 						struct list_head *head, int max)
2926 {
2927 	struct nat_entry_set *cur;
2928 
2929 	if (nes->entry_cnt >= max)
2930 		goto add_out;
2931 
2932 	list_for_each_entry(cur, head, set_list) {
2933 		if (cur->entry_cnt >= nes->entry_cnt) {
2934 			list_add(&nes->set_list, cur->set_list.prev);
2935 			return;
2936 		}
2937 	}
2938 add_out:
2939 	list_add_tail(&nes->set_list, head);
2940 }
2941 
2942 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs,
2943 							unsigned int valid)
2944 {
2945 	if (valid == 0) {
2946 		__set_bit_le(nat_ofs, nm_i->empty_nat_bits);
2947 		__clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2948 		return;
2949 	}
2950 
2951 	__clear_bit_le(nat_ofs, nm_i->empty_nat_bits);
2952 	if (valid == NAT_ENTRY_PER_BLOCK)
2953 		__set_bit_le(nat_ofs, nm_i->full_nat_bits);
2954 	else
2955 		__clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2956 }
2957 
2958 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2959 						struct page *page)
2960 {
2961 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2962 	unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2963 	struct f2fs_nat_block *nat_blk = page_address(page);
2964 	int valid = 0;
2965 	int i = 0;
2966 
2967 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
2968 		return;
2969 
2970 	if (nat_index == 0) {
2971 		valid = 1;
2972 		i = 1;
2973 	}
2974 	for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2975 		if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2976 			valid++;
2977 	}
2978 
2979 	__update_nat_bits(nm_i, nat_index, valid);
2980 }
2981 
2982 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi)
2983 {
2984 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2985 	unsigned int nat_ofs;
2986 
2987 	f2fs_down_read(&nm_i->nat_tree_lock);
2988 
2989 	for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) {
2990 		unsigned int valid = 0, nid_ofs = 0;
2991 
2992 		/* handle nid zero due to it should never be used */
2993 		if (unlikely(nat_ofs == 0)) {
2994 			valid = 1;
2995 			nid_ofs = 1;
2996 		}
2997 
2998 		for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) {
2999 			if (!test_bit_le(nid_ofs,
3000 					nm_i->free_nid_bitmap[nat_ofs]))
3001 				valid++;
3002 		}
3003 
3004 		__update_nat_bits(nm_i, nat_ofs, valid);
3005 	}
3006 
3007 	f2fs_up_read(&nm_i->nat_tree_lock);
3008 }
3009 
3010 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
3011 		struct nat_entry_set *set, struct cp_control *cpc)
3012 {
3013 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3014 	struct f2fs_journal *journal = curseg->journal;
3015 	nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
3016 	bool to_journal = true;
3017 	struct f2fs_nat_block *nat_blk;
3018 	struct nat_entry *ne, *cur;
3019 	struct page *page = NULL;
3020 
3021 	/*
3022 	 * there are two steps to flush nat entries:
3023 	 * #1, flush nat entries to journal in current hot data summary block.
3024 	 * #2, flush nat entries to nat page.
3025 	 */
3026 	if ((cpc->reason & CP_UMOUNT) ||
3027 		!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
3028 		to_journal = false;
3029 
3030 	if (to_journal) {
3031 		down_write(&curseg->journal_rwsem);
3032 	} else {
3033 		page = get_next_nat_page(sbi, start_nid);
3034 		if (IS_ERR(page))
3035 			return PTR_ERR(page);
3036 
3037 		nat_blk = page_address(page);
3038 		f2fs_bug_on(sbi, !nat_blk);
3039 	}
3040 
3041 	/* flush dirty nats in nat entry set */
3042 	list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
3043 		struct f2fs_nat_entry *raw_ne;
3044 		nid_t nid = nat_get_nid(ne);
3045 		int offset;
3046 
3047 		f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
3048 
3049 		if (to_journal) {
3050 			offset = f2fs_lookup_journal_in_cursum(journal,
3051 							NAT_JOURNAL, nid, 1);
3052 			f2fs_bug_on(sbi, offset < 0);
3053 			raw_ne = &nat_in_journal(journal, offset);
3054 			nid_in_journal(journal, offset) = cpu_to_le32(nid);
3055 		} else {
3056 			raw_ne = &nat_blk->entries[nid - start_nid];
3057 		}
3058 		raw_nat_from_node_info(raw_ne, &ne->ni);
3059 		nat_reset_flag(ne);
3060 		__clear_nat_cache_dirty(NM_I(sbi), set, ne);
3061 		if (nat_get_blkaddr(ne) == NULL_ADDR) {
3062 			add_free_nid(sbi, nid, false, true);
3063 		} else {
3064 			spin_lock(&NM_I(sbi)->nid_list_lock);
3065 			update_free_nid_bitmap(sbi, nid, false, false);
3066 			spin_unlock(&NM_I(sbi)->nid_list_lock);
3067 		}
3068 	}
3069 
3070 	if (to_journal) {
3071 		up_write(&curseg->journal_rwsem);
3072 	} else {
3073 		update_nat_bits(sbi, start_nid, page);
3074 		f2fs_put_page(page, 1);
3075 	}
3076 
3077 	/* Allow dirty nats by node block allocation in write_begin */
3078 	if (!set->entry_cnt) {
3079 		radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
3080 		kmem_cache_free(nat_entry_set_slab, set);
3081 	}
3082 	return 0;
3083 }
3084 
3085 /*
3086  * This function is called during the checkpointing process.
3087  */
3088 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3089 {
3090 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3091 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3092 	struct f2fs_journal *journal = curseg->journal;
3093 	struct nat_entry_set *setvec[NAT_VEC_SIZE];
3094 	struct nat_entry_set *set, *tmp;
3095 	unsigned int found;
3096 	nid_t set_idx = 0;
3097 	LIST_HEAD(sets);
3098 	int err = 0;
3099 
3100 	/*
3101 	 * during unmount, let's flush nat_bits before checking
3102 	 * nat_cnt[DIRTY_NAT].
3103 	 */
3104 	if (cpc->reason & CP_UMOUNT) {
3105 		f2fs_down_write(&nm_i->nat_tree_lock);
3106 		remove_nats_in_journal(sbi);
3107 		f2fs_up_write(&nm_i->nat_tree_lock);
3108 	}
3109 
3110 	if (!nm_i->nat_cnt[DIRTY_NAT])
3111 		return 0;
3112 
3113 	f2fs_down_write(&nm_i->nat_tree_lock);
3114 
3115 	/*
3116 	 * if there are no enough space in journal to store dirty nat
3117 	 * entries, remove all entries from journal and merge them
3118 	 * into nat entry set.
3119 	 */
3120 	if (cpc->reason & CP_UMOUNT ||
3121 		!__has_cursum_space(journal,
3122 			nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3123 		remove_nats_in_journal(sbi);
3124 
3125 	while ((found = __gang_lookup_nat_set(nm_i,
3126 					set_idx, NAT_VEC_SIZE, setvec))) {
3127 		unsigned idx;
3128 
3129 		set_idx = setvec[found - 1]->set + 1;
3130 		for (idx = 0; idx < found; idx++)
3131 			__adjust_nat_entry_set(setvec[idx], &sets,
3132 						MAX_NAT_JENTRIES(journal));
3133 	}
3134 
3135 	/* flush dirty nats in nat entry set */
3136 	list_for_each_entry_safe(set, tmp, &sets, set_list) {
3137 		err = __flush_nat_entry_set(sbi, set, cpc);
3138 		if (err)
3139 			break;
3140 	}
3141 
3142 	f2fs_up_write(&nm_i->nat_tree_lock);
3143 	/* Allow dirty nats by node block allocation in write_begin */
3144 
3145 	return err;
3146 }
3147 
3148 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3149 {
3150 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3151 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3152 	unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3153 	unsigned int i;
3154 	__u64 cp_ver = cur_cp_version(ckpt);
3155 	block_t nat_bits_addr;
3156 
3157 	nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3158 	nm_i->nat_bits = f2fs_kvzalloc(sbi,
3159 			nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3160 	if (!nm_i->nat_bits)
3161 		return -ENOMEM;
3162 
3163 	nm_i->full_nat_bits = nm_i->nat_bits + 8;
3164 	nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3165 
3166 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3167 		return 0;
3168 
3169 	nat_bits_addr = __start_cp_addr(sbi) + BLKS_PER_SEG(sbi) -
3170 						nm_i->nat_bits_blocks;
3171 	for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3172 		struct page *page;
3173 
3174 		page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3175 		if (IS_ERR(page))
3176 			return PTR_ERR(page);
3177 
3178 		memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3179 					page_address(page), F2FS_BLKSIZE);
3180 		f2fs_put_page(page, 1);
3181 	}
3182 
3183 	cp_ver |= (cur_cp_crc(ckpt) << 32);
3184 	if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3185 		clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
3186 		f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)",
3187 			cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits));
3188 		return 0;
3189 	}
3190 
3191 	f2fs_notice(sbi, "Found nat_bits in checkpoint");
3192 	return 0;
3193 }
3194 
3195 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3196 {
3197 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3198 	unsigned int i = 0;
3199 	nid_t nid, last_nid;
3200 
3201 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3202 		return;
3203 
3204 	for (i = 0; i < nm_i->nat_blocks; i++) {
3205 		i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3206 		if (i >= nm_i->nat_blocks)
3207 			break;
3208 
3209 		__set_bit_le(i, nm_i->nat_block_bitmap);
3210 
3211 		nid = i * NAT_ENTRY_PER_BLOCK;
3212 		last_nid = nid + NAT_ENTRY_PER_BLOCK;
3213 
3214 		spin_lock(&NM_I(sbi)->nid_list_lock);
3215 		for (; nid < last_nid; nid++)
3216 			update_free_nid_bitmap(sbi, nid, true, true);
3217 		spin_unlock(&NM_I(sbi)->nid_list_lock);
3218 	}
3219 
3220 	for (i = 0; i < nm_i->nat_blocks; i++) {
3221 		i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3222 		if (i >= nm_i->nat_blocks)
3223 			break;
3224 
3225 		__set_bit_le(i, nm_i->nat_block_bitmap);
3226 	}
3227 }
3228 
3229 static int init_node_manager(struct f2fs_sb_info *sbi)
3230 {
3231 	struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3232 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3233 	unsigned char *version_bitmap;
3234 	unsigned int nat_segs;
3235 	int err;
3236 
3237 	nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3238 
3239 	/* segment_count_nat includes pair segment so divide to 2. */
3240 	nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3241 	nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3242 	nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3243 
3244 	/* not used nids: 0, node, meta, (and root counted as valid node) */
3245 	nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3246 						F2FS_RESERVED_NODE_NUM;
3247 	nm_i->nid_cnt[FREE_NID] = 0;
3248 	nm_i->nid_cnt[PREALLOC_NID] = 0;
3249 	nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3250 	nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3251 	nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3252 	nm_i->max_rf_node_blocks = DEF_RF_NODE_BLOCKS;
3253 
3254 	INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3255 	INIT_LIST_HEAD(&nm_i->free_nid_list);
3256 	INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3257 	INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3258 	INIT_LIST_HEAD(&nm_i->nat_entries);
3259 	spin_lock_init(&nm_i->nat_list_lock);
3260 
3261 	mutex_init(&nm_i->build_lock);
3262 	spin_lock_init(&nm_i->nid_list_lock);
3263 	init_f2fs_rwsem(&nm_i->nat_tree_lock);
3264 
3265 	nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3266 	nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3267 	version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3268 	nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3269 					GFP_KERNEL);
3270 	if (!nm_i->nat_bitmap)
3271 		return -ENOMEM;
3272 
3273 	err = __get_nat_bitmaps(sbi);
3274 	if (err)
3275 		return err;
3276 
3277 #ifdef CONFIG_F2FS_CHECK_FS
3278 	nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3279 					GFP_KERNEL);
3280 	if (!nm_i->nat_bitmap_mir)
3281 		return -ENOMEM;
3282 #endif
3283 
3284 	return 0;
3285 }
3286 
3287 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3288 {
3289 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3290 	int i;
3291 
3292 	nm_i->free_nid_bitmap =
3293 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3294 					      nm_i->nat_blocks),
3295 			      GFP_KERNEL);
3296 	if (!nm_i->free_nid_bitmap)
3297 		return -ENOMEM;
3298 
3299 	for (i = 0; i < nm_i->nat_blocks; i++) {
3300 		nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3301 			f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3302 		if (!nm_i->free_nid_bitmap[i])
3303 			return -ENOMEM;
3304 	}
3305 
3306 	nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3307 								GFP_KERNEL);
3308 	if (!nm_i->nat_block_bitmap)
3309 		return -ENOMEM;
3310 
3311 	nm_i->free_nid_count =
3312 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3313 					      nm_i->nat_blocks),
3314 			      GFP_KERNEL);
3315 	if (!nm_i->free_nid_count)
3316 		return -ENOMEM;
3317 	return 0;
3318 }
3319 
3320 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3321 {
3322 	int err;
3323 
3324 	sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3325 							GFP_KERNEL);
3326 	if (!sbi->nm_info)
3327 		return -ENOMEM;
3328 
3329 	err = init_node_manager(sbi);
3330 	if (err)
3331 		return err;
3332 
3333 	err = init_free_nid_cache(sbi);
3334 	if (err)
3335 		return err;
3336 
3337 	/* load free nid status from nat_bits table */
3338 	load_free_nid_bitmap(sbi);
3339 
3340 	return f2fs_build_free_nids(sbi, true, true);
3341 }
3342 
3343 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3344 {
3345 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3346 	struct free_nid *i, *next_i;
3347 	void *vec[NAT_VEC_SIZE];
3348 	struct nat_entry **natvec = (struct nat_entry **)vec;
3349 	struct nat_entry_set **setvec = (struct nat_entry_set **)vec;
3350 	nid_t nid = 0;
3351 	unsigned int found;
3352 
3353 	if (!nm_i)
3354 		return;
3355 
3356 	/* destroy free nid list */
3357 	spin_lock(&nm_i->nid_list_lock);
3358 	list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3359 		__remove_free_nid(sbi, i, FREE_NID);
3360 		spin_unlock(&nm_i->nid_list_lock);
3361 		kmem_cache_free(free_nid_slab, i);
3362 		spin_lock(&nm_i->nid_list_lock);
3363 	}
3364 	f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3365 	f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3366 	f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3367 	spin_unlock(&nm_i->nid_list_lock);
3368 
3369 	/* destroy nat cache */
3370 	f2fs_down_write(&nm_i->nat_tree_lock);
3371 	while ((found = __gang_lookup_nat_cache(nm_i,
3372 					nid, NAT_VEC_SIZE, natvec))) {
3373 		unsigned idx;
3374 
3375 		nid = nat_get_nid(natvec[found - 1]) + 1;
3376 		for (idx = 0; idx < found; idx++) {
3377 			spin_lock(&nm_i->nat_list_lock);
3378 			list_del(&natvec[idx]->list);
3379 			spin_unlock(&nm_i->nat_list_lock);
3380 
3381 			__del_from_nat_cache(nm_i, natvec[idx]);
3382 		}
3383 	}
3384 	f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3385 
3386 	/* destroy nat set cache */
3387 	nid = 0;
3388 	memset(vec, 0, sizeof(void *) * NAT_VEC_SIZE);
3389 	while ((found = __gang_lookup_nat_set(nm_i,
3390 					nid, NAT_VEC_SIZE, setvec))) {
3391 		unsigned idx;
3392 
3393 		nid = setvec[found - 1]->set + 1;
3394 		for (idx = 0; idx < found; idx++) {
3395 			/* entry_cnt is not zero, when cp_error was occurred */
3396 			f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3397 			radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3398 			kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3399 		}
3400 	}
3401 	f2fs_up_write(&nm_i->nat_tree_lock);
3402 
3403 	kvfree(nm_i->nat_block_bitmap);
3404 	if (nm_i->free_nid_bitmap) {
3405 		int i;
3406 
3407 		for (i = 0; i < nm_i->nat_blocks; i++)
3408 			kvfree(nm_i->free_nid_bitmap[i]);
3409 		kvfree(nm_i->free_nid_bitmap);
3410 	}
3411 	kvfree(nm_i->free_nid_count);
3412 
3413 	kvfree(nm_i->nat_bitmap);
3414 	kvfree(nm_i->nat_bits);
3415 #ifdef CONFIG_F2FS_CHECK_FS
3416 	kvfree(nm_i->nat_bitmap_mir);
3417 #endif
3418 	sbi->nm_info = NULL;
3419 	kfree(nm_i);
3420 }
3421 
3422 int __init f2fs_create_node_manager_caches(void)
3423 {
3424 	nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3425 			sizeof(struct nat_entry));
3426 	if (!nat_entry_slab)
3427 		goto fail;
3428 
3429 	free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3430 			sizeof(struct free_nid));
3431 	if (!free_nid_slab)
3432 		goto destroy_nat_entry;
3433 
3434 	nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3435 			sizeof(struct nat_entry_set));
3436 	if (!nat_entry_set_slab)
3437 		goto destroy_free_nid;
3438 
3439 	fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3440 			sizeof(struct fsync_node_entry));
3441 	if (!fsync_node_entry_slab)
3442 		goto destroy_nat_entry_set;
3443 	return 0;
3444 
3445 destroy_nat_entry_set:
3446 	kmem_cache_destroy(nat_entry_set_slab);
3447 destroy_free_nid:
3448 	kmem_cache_destroy(free_nid_slab);
3449 destroy_nat_entry:
3450 	kmem_cache_destroy(nat_entry_slab);
3451 fail:
3452 	return -ENOMEM;
3453 }
3454 
3455 void f2fs_destroy_node_manager_caches(void)
3456 {
3457 	kmem_cache_destroy(fsync_node_entry_slab);
3458 	kmem_cache_destroy(nat_entry_set_slab);
3459 	kmem_cache_destroy(free_nid_slab);
3460 	kmem_cache_destroy(nat_entry_slab);
3461 }
3462