xref: /openbmc/linux/fs/f2fs/node.c (revision e5242c5f)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * fs/f2fs/node.c
4  *
5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6  *             http://www.samsung.com/
7  */
8 #include <linux/fs.h>
9 #include <linux/f2fs_fs.h>
10 #include <linux/mpage.h>
11 #include <linux/sched/mm.h>
12 #include <linux/blkdev.h>
13 #include <linux/pagevec.h>
14 #include <linux/swap.h>
15 
16 #include "f2fs.h"
17 #include "node.h"
18 #include "segment.h"
19 #include "xattr.h"
20 #include "iostat.h"
21 #include <trace/events/f2fs.h>
22 
23 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
24 
25 static struct kmem_cache *nat_entry_slab;
26 static struct kmem_cache *free_nid_slab;
27 static struct kmem_cache *nat_entry_set_slab;
28 static struct kmem_cache *fsync_node_entry_slab;
29 
30 /*
31  * Check whether the given nid is within node id range.
32  */
33 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
34 {
35 	if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
36 		set_sbi_flag(sbi, SBI_NEED_FSCK);
37 		f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
38 			  __func__, nid);
39 		f2fs_handle_error(sbi, ERROR_CORRUPTED_INODE);
40 		return -EFSCORRUPTED;
41 	}
42 	return 0;
43 }
44 
45 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
46 {
47 	struct f2fs_nm_info *nm_i = NM_I(sbi);
48 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
49 	struct sysinfo val;
50 	unsigned long avail_ram;
51 	unsigned long mem_size = 0;
52 	bool res = false;
53 
54 	if (!nm_i)
55 		return true;
56 
57 	si_meminfo(&val);
58 
59 	/* only uses low memory */
60 	avail_ram = val.totalram - val.totalhigh;
61 
62 	/*
63 	 * give 25%, 25%, 50%, 50%, 25%, 25% memory for each components respectively
64 	 */
65 	if (type == FREE_NIDS) {
66 		mem_size = (nm_i->nid_cnt[FREE_NID] *
67 				sizeof(struct free_nid)) >> PAGE_SHIFT;
68 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
69 	} else if (type == NAT_ENTRIES) {
70 		mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
71 				sizeof(struct nat_entry)) >> PAGE_SHIFT;
72 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
73 		if (excess_cached_nats(sbi))
74 			res = false;
75 	} else if (type == DIRTY_DENTS) {
76 		if (sbi->sb->s_bdi->wb.dirty_exceeded)
77 			return false;
78 		mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
79 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
80 	} else if (type == INO_ENTRIES) {
81 		int i;
82 
83 		for (i = 0; i < MAX_INO_ENTRY; i++)
84 			mem_size += sbi->im[i].ino_num *
85 						sizeof(struct ino_entry);
86 		mem_size >>= PAGE_SHIFT;
87 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
88 	} else if (type == READ_EXTENT_CACHE || type == AGE_EXTENT_CACHE) {
89 		enum extent_type etype = type == READ_EXTENT_CACHE ?
90 						EX_READ : EX_BLOCK_AGE;
91 		struct extent_tree_info *eti = &sbi->extent_tree[etype];
92 
93 		mem_size = (atomic_read(&eti->total_ext_tree) *
94 				sizeof(struct extent_tree) +
95 				atomic_read(&eti->total_ext_node) *
96 				sizeof(struct extent_node)) >> PAGE_SHIFT;
97 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
98 	} else if (type == DISCARD_CACHE) {
99 		mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
100 				sizeof(struct discard_cmd)) >> PAGE_SHIFT;
101 		res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
102 	} else if (type == COMPRESS_PAGE) {
103 #ifdef CONFIG_F2FS_FS_COMPRESSION
104 		unsigned long free_ram = val.freeram;
105 
106 		/*
107 		 * free memory is lower than watermark or cached page count
108 		 * exceed threshold, deny caching compress page.
109 		 */
110 		res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
111 			(COMPRESS_MAPPING(sbi)->nrpages <
112 			 free_ram * sbi->compress_percent / 100);
113 #else
114 		res = false;
115 #endif
116 	} else {
117 		if (!sbi->sb->s_bdi->wb.dirty_exceeded)
118 			return true;
119 	}
120 	return res;
121 }
122 
123 static void clear_node_page_dirty(struct page *page)
124 {
125 	if (PageDirty(page)) {
126 		f2fs_clear_page_cache_dirty_tag(page);
127 		clear_page_dirty_for_io(page);
128 		dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
129 	}
130 	ClearPageUptodate(page);
131 }
132 
133 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
134 {
135 	return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
136 }
137 
138 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
139 {
140 	struct page *src_page;
141 	struct page *dst_page;
142 	pgoff_t dst_off;
143 	void *src_addr;
144 	void *dst_addr;
145 	struct f2fs_nm_info *nm_i = NM_I(sbi);
146 
147 	dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
148 
149 	/* get current nat block page with lock */
150 	src_page = get_current_nat_page(sbi, nid);
151 	if (IS_ERR(src_page))
152 		return src_page;
153 	dst_page = f2fs_grab_meta_page(sbi, dst_off);
154 	f2fs_bug_on(sbi, PageDirty(src_page));
155 
156 	src_addr = page_address(src_page);
157 	dst_addr = page_address(dst_page);
158 	memcpy(dst_addr, src_addr, PAGE_SIZE);
159 	set_page_dirty(dst_page);
160 	f2fs_put_page(src_page, 1);
161 
162 	set_to_next_nat(nm_i, nid);
163 
164 	return dst_page;
165 }
166 
167 static struct nat_entry *__alloc_nat_entry(struct f2fs_sb_info *sbi,
168 						nid_t nid, bool no_fail)
169 {
170 	struct nat_entry *new;
171 
172 	new = f2fs_kmem_cache_alloc(nat_entry_slab,
173 					GFP_F2FS_ZERO, no_fail, sbi);
174 	if (new) {
175 		nat_set_nid(new, nid);
176 		nat_reset_flag(new);
177 	}
178 	return new;
179 }
180 
181 static void __free_nat_entry(struct nat_entry *e)
182 {
183 	kmem_cache_free(nat_entry_slab, e);
184 }
185 
186 /* must be locked by nat_tree_lock */
187 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
188 	struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
189 {
190 	if (no_fail)
191 		f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
192 	else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
193 		return NULL;
194 
195 	if (raw_ne)
196 		node_info_from_raw_nat(&ne->ni, raw_ne);
197 
198 	spin_lock(&nm_i->nat_list_lock);
199 	list_add_tail(&ne->list, &nm_i->nat_entries);
200 	spin_unlock(&nm_i->nat_list_lock);
201 
202 	nm_i->nat_cnt[TOTAL_NAT]++;
203 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
204 	return ne;
205 }
206 
207 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
208 {
209 	struct nat_entry *ne;
210 
211 	ne = radix_tree_lookup(&nm_i->nat_root, n);
212 
213 	/* for recent accessed nat entry, move it to tail of lru list */
214 	if (ne && !get_nat_flag(ne, IS_DIRTY)) {
215 		spin_lock(&nm_i->nat_list_lock);
216 		if (!list_empty(&ne->list))
217 			list_move_tail(&ne->list, &nm_i->nat_entries);
218 		spin_unlock(&nm_i->nat_list_lock);
219 	}
220 
221 	return ne;
222 }
223 
224 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
225 		nid_t start, unsigned int nr, struct nat_entry **ep)
226 {
227 	return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
228 }
229 
230 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
231 {
232 	radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
233 	nm_i->nat_cnt[TOTAL_NAT]--;
234 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
235 	__free_nat_entry(e);
236 }
237 
238 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
239 							struct nat_entry *ne)
240 {
241 	nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
242 	struct nat_entry_set *head;
243 
244 	head = radix_tree_lookup(&nm_i->nat_set_root, set);
245 	if (!head) {
246 		head = f2fs_kmem_cache_alloc(nat_entry_set_slab,
247 						GFP_NOFS, true, NULL);
248 
249 		INIT_LIST_HEAD(&head->entry_list);
250 		INIT_LIST_HEAD(&head->set_list);
251 		head->set = set;
252 		head->entry_cnt = 0;
253 		f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
254 	}
255 	return head;
256 }
257 
258 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
259 						struct nat_entry *ne)
260 {
261 	struct nat_entry_set *head;
262 	bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
263 
264 	if (!new_ne)
265 		head = __grab_nat_entry_set(nm_i, ne);
266 
267 	/*
268 	 * update entry_cnt in below condition:
269 	 * 1. update NEW_ADDR to valid block address;
270 	 * 2. update old block address to new one;
271 	 */
272 	if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
273 				!get_nat_flag(ne, IS_DIRTY)))
274 		head->entry_cnt++;
275 
276 	set_nat_flag(ne, IS_PREALLOC, new_ne);
277 
278 	if (get_nat_flag(ne, IS_DIRTY))
279 		goto refresh_list;
280 
281 	nm_i->nat_cnt[DIRTY_NAT]++;
282 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
283 	set_nat_flag(ne, IS_DIRTY, true);
284 refresh_list:
285 	spin_lock(&nm_i->nat_list_lock);
286 	if (new_ne)
287 		list_del_init(&ne->list);
288 	else
289 		list_move_tail(&ne->list, &head->entry_list);
290 	spin_unlock(&nm_i->nat_list_lock);
291 }
292 
293 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
294 		struct nat_entry_set *set, struct nat_entry *ne)
295 {
296 	spin_lock(&nm_i->nat_list_lock);
297 	list_move_tail(&ne->list, &nm_i->nat_entries);
298 	spin_unlock(&nm_i->nat_list_lock);
299 
300 	set_nat_flag(ne, IS_DIRTY, false);
301 	set->entry_cnt--;
302 	nm_i->nat_cnt[DIRTY_NAT]--;
303 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
304 }
305 
306 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
307 		nid_t start, unsigned int nr, struct nat_entry_set **ep)
308 {
309 	return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
310 							start, nr);
311 }
312 
313 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
314 {
315 	return NODE_MAPPING(sbi) == page->mapping &&
316 			IS_DNODE(page) && is_cold_node(page);
317 }
318 
319 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
320 {
321 	spin_lock_init(&sbi->fsync_node_lock);
322 	INIT_LIST_HEAD(&sbi->fsync_node_list);
323 	sbi->fsync_seg_id = 0;
324 	sbi->fsync_node_num = 0;
325 }
326 
327 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
328 							struct page *page)
329 {
330 	struct fsync_node_entry *fn;
331 	unsigned long flags;
332 	unsigned int seq_id;
333 
334 	fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab,
335 					GFP_NOFS, true, NULL);
336 
337 	get_page(page);
338 	fn->page = page;
339 	INIT_LIST_HEAD(&fn->list);
340 
341 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
342 	list_add_tail(&fn->list, &sbi->fsync_node_list);
343 	fn->seq_id = sbi->fsync_seg_id++;
344 	seq_id = fn->seq_id;
345 	sbi->fsync_node_num++;
346 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
347 
348 	return seq_id;
349 }
350 
351 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
352 {
353 	struct fsync_node_entry *fn;
354 	unsigned long flags;
355 
356 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
357 	list_for_each_entry(fn, &sbi->fsync_node_list, list) {
358 		if (fn->page == page) {
359 			list_del(&fn->list);
360 			sbi->fsync_node_num--;
361 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
362 			kmem_cache_free(fsync_node_entry_slab, fn);
363 			put_page(page);
364 			return;
365 		}
366 	}
367 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
368 	f2fs_bug_on(sbi, 1);
369 }
370 
371 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
372 {
373 	unsigned long flags;
374 
375 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
376 	sbi->fsync_seg_id = 0;
377 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
378 }
379 
380 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
381 {
382 	struct f2fs_nm_info *nm_i = NM_I(sbi);
383 	struct nat_entry *e;
384 	bool need = false;
385 
386 	f2fs_down_read(&nm_i->nat_tree_lock);
387 	e = __lookup_nat_cache(nm_i, nid);
388 	if (e) {
389 		if (!get_nat_flag(e, IS_CHECKPOINTED) &&
390 				!get_nat_flag(e, HAS_FSYNCED_INODE))
391 			need = true;
392 	}
393 	f2fs_up_read(&nm_i->nat_tree_lock);
394 	return need;
395 }
396 
397 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
398 {
399 	struct f2fs_nm_info *nm_i = NM_I(sbi);
400 	struct nat_entry *e;
401 	bool is_cp = true;
402 
403 	f2fs_down_read(&nm_i->nat_tree_lock);
404 	e = __lookup_nat_cache(nm_i, nid);
405 	if (e && !get_nat_flag(e, IS_CHECKPOINTED))
406 		is_cp = false;
407 	f2fs_up_read(&nm_i->nat_tree_lock);
408 	return is_cp;
409 }
410 
411 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
412 {
413 	struct f2fs_nm_info *nm_i = NM_I(sbi);
414 	struct nat_entry *e;
415 	bool need_update = true;
416 
417 	f2fs_down_read(&nm_i->nat_tree_lock);
418 	e = __lookup_nat_cache(nm_i, ino);
419 	if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
420 			(get_nat_flag(e, IS_CHECKPOINTED) ||
421 			 get_nat_flag(e, HAS_FSYNCED_INODE)))
422 		need_update = false;
423 	f2fs_up_read(&nm_i->nat_tree_lock);
424 	return need_update;
425 }
426 
427 /* must be locked by nat_tree_lock */
428 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
429 						struct f2fs_nat_entry *ne)
430 {
431 	struct f2fs_nm_info *nm_i = NM_I(sbi);
432 	struct nat_entry *new, *e;
433 
434 	/* Let's mitigate lock contention of nat_tree_lock during checkpoint */
435 	if (f2fs_rwsem_is_locked(&sbi->cp_global_sem))
436 		return;
437 
438 	new = __alloc_nat_entry(sbi, nid, false);
439 	if (!new)
440 		return;
441 
442 	f2fs_down_write(&nm_i->nat_tree_lock);
443 	e = __lookup_nat_cache(nm_i, nid);
444 	if (!e)
445 		e = __init_nat_entry(nm_i, new, ne, false);
446 	else
447 		f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
448 				nat_get_blkaddr(e) !=
449 					le32_to_cpu(ne->block_addr) ||
450 				nat_get_version(e) != ne->version);
451 	f2fs_up_write(&nm_i->nat_tree_lock);
452 	if (e != new)
453 		__free_nat_entry(new);
454 }
455 
456 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
457 			block_t new_blkaddr, bool fsync_done)
458 {
459 	struct f2fs_nm_info *nm_i = NM_I(sbi);
460 	struct nat_entry *e;
461 	struct nat_entry *new = __alloc_nat_entry(sbi, ni->nid, true);
462 
463 	f2fs_down_write(&nm_i->nat_tree_lock);
464 	e = __lookup_nat_cache(nm_i, ni->nid);
465 	if (!e) {
466 		e = __init_nat_entry(nm_i, new, NULL, true);
467 		copy_node_info(&e->ni, ni);
468 		f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
469 	} else if (new_blkaddr == NEW_ADDR) {
470 		/*
471 		 * when nid is reallocated,
472 		 * previous nat entry can be remained in nat cache.
473 		 * So, reinitialize it with new information.
474 		 */
475 		copy_node_info(&e->ni, ni);
476 		f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
477 	}
478 	/* let's free early to reduce memory consumption */
479 	if (e != new)
480 		__free_nat_entry(new);
481 
482 	/* sanity check */
483 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
484 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
485 			new_blkaddr == NULL_ADDR);
486 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
487 			new_blkaddr == NEW_ADDR);
488 	f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
489 			new_blkaddr == NEW_ADDR);
490 
491 	/* increment version no as node is removed */
492 	if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
493 		unsigned char version = nat_get_version(e);
494 
495 		nat_set_version(e, inc_node_version(version));
496 	}
497 
498 	/* change address */
499 	nat_set_blkaddr(e, new_blkaddr);
500 	if (!__is_valid_data_blkaddr(new_blkaddr))
501 		set_nat_flag(e, IS_CHECKPOINTED, false);
502 	__set_nat_cache_dirty(nm_i, e);
503 
504 	/* update fsync_mark if its inode nat entry is still alive */
505 	if (ni->nid != ni->ino)
506 		e = __lookup_nat_cache(nm_i, ni->ino);
507 	if (e) {
508 		if (fsync_done && ni->nid == ni->ino)
509 			set_nat_flag(e, HAS_FSYNCED_INODE, true);
510 		set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
511 	}
512 	f2fs_up_write(&nm_i->nat_tree_lock);
513 }
514 
515 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
516 {
517 	struct f2fs_nm_info *nm_i = NM_I(sbi);
518 	int nr = nr_shrink;
519 
520 	if (!f2fs_down_write_trylock(&nm_i->nat_tree_lock))
521 		return 0;
522 
523 	spin_lock(&nm_i->nat_list_lock);
524 	while (nr_shrink) {
525 		struct nat_entry *ne;
526 
527 		if (list_empty(&nm_i->nat_entries))
528 			break;
529 
530 		ne = list_first_entry(&nm_i->nat_entries,
531 					struct nat_entry, list);
532 		list_del(&ne->list);
533 		spin_unlock(&nm_i->nat_list_lock);
534 
535 		__del_from_nat_cache(nm_i, ne);
536 		nr_shrink--;
537 
538 		spin_lock(&nm_i->nat_list_lock);
539 	}
540 	spin_unlock(&nm_i->nat_list_lock);
541 
542 	f2fs_up_write(&nm_i->nat_tree_lock);
543 	return nr - nr_shrink;
544 }
545 
546 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
547 				struct node_info *ni, bool checkpoint_context)
548 {
549 	struct f2fs_nm_info *nm_i = NM_I(sbi);
550 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
551 	struct f2fs_journal *journal = curseg->journal;
552 	nid_t start_nid = START_NID(nid);
553 	struct f2fs_nat_block *nat_blk;
554 	struct page *page = NULL;
555 	struct f2fs_nat_entry ne;
556 	struct nat_entry *e;
557 	pgoff_t index;
558 	block_t blkaddr;
559 	int i;
560 
561 	ni->nid = nid;
562 retry:
563 	/* Check nat cache */
564 	f2fs_down_read(&nm_i->nat_tree_lock);
565 	e = __lookup_nat_cache(nm_i, nid);
566 	if (e) {
567 		ni->ino = nat_get_ino(e);
568 		ni->blk_addr = nat_get_blkaddr(e);
569 		ni->version = nat_get_version(e);
570 		f2fs_up_read(&nm_i->nat_tree_lock);
571 		return 0;
572 	}
573 
574 	/*
575 	 * Check current segment summary by trying to grab journal_rwsem first.
576 	 * This sem is on the critical path on the checkpoint requiring the above
577 	 * nat_tree_lock. Therefore, we should retry, if we failed to grab here
578 	 * while not bothering checkpoint.
579 	 */
580 	if (!f2fs_rwsem_is_locked(&sbi->cp_global_sem) || checkpoint_context) {
581 		down_read(&curseg->journal_rwsem);
582 	} else if (f2fs_rwsem_is_contended(&nm_i->nat_tree_lock) ||
583 				!down_read_trylock(&curseg->journal_rwsem)) {
584 		f2fs_up_read(&nm_i->nat_tree_lock);
585 		goto retry;
586 	}
587 
588 	i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
589 	if (i >= 0) {
590 		ne = nat_in_journal(journal, i);
591 		node_info_from_raw_nat(ni, &ne);
592 	}
593 	up_read(&curseg->journal_rwsem);
594 	if (i >= 0) {
595 		f2fs_up_read(&nm_i->nat_tree_lock);
596 		goto cache;
597 	}
598 
599 	/* Fill node_info from nat page */
600 	index = current_nat_addr(sbi, nid);
601 	f2fs_up_read(&nm_i->nat_tree_lock);
602 
603 	page = f2fs_get_meta_page(sbi, index);
604 	if (IS_ERR(page))
605 		return PTR_ERR(page);
606 
607 	nat_blk = (struct f2fs_nat_block *)page_address(page);
608 	ne = nat_blk->entries[nid - start_nid];
609 	node_info_from_raw_nat(ni, &ne);
610 	f2fs_put_page(page, 1);
611 cache:
612 	blkaddr = le32_to_cpu(ne.block_addr);
613 	if (__is_valid_data_blkaddr(blkaddr) &&
614 		!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
615 		return -EFAULT;
616 
617 	/* cache nat entry */
618 	cache_nat_entry(sbi, nid, &ne);
619 	return 0;
620 }
621 
622 /*
623  * readahead MAX_RA_NODE number of node pages.
624  */
625 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
626 {
627 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
628 	struct blk_plug plug;
629 	int i, end;
630 	nid_t nid;
631 
632 	blk_start_plug(&plug);
633 
634 	/* Then, try readahead for siblings of the desired node */
635 	end = start + n;
636 	end = min(end, NIDS_PER_BLOCK);
637 	for (i = start; i < end; i++) {
638 		nid = get_nid(parent, i, false);
639 		f2fs_ra_node_page(sbi, nid);
640 	}
641 
642 	blk_finish_plug(&plug);
643 }
644 
645 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
646 {
647 	const long direct_index = ADDRS_PER_INODE(dn->inode);
648 	const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
649 	const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
650 	unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
651 	int cur_level = dn->cur_level;
652 	int max_level = dn->max_level;
653 	pgoff_t base = 0;
654 
655 	if (!dn->max_level)
656 		return pgofs + 1;
657 
658 	while (max_level-- > cur_level)
659 		skipped_unit *= NIDS_PER_BLOCK;
660 
661 	switch (dn->max_level) {
662 	case 3:
663 		base += 2 * indirect_blks;
664 		fallthrough;
665 	case 2:
666 		base += 2 * direct_blks;
667 		fallthrough;
668 	case 1:
669 		base += direct_index;
670 		break;
671 	default:
672 		f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
673 	}
674 
675 	return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
676 }
677 
678 /*
679  * The maximum depth is four.
680  * Offset[0] will have raw inode offset.
681  */
682 static int get_node_path(struct inode *inode, long block,
683 				int offset[4], unsigned int noffset[4])
684 {
685 	const long direct_index = ADDRS_PER_INODE(inode);
686 	const long direct_blks = ADDRS_PER_BLOCK(inode);
687 	const long dptrs_per_blk = NIDS_PER_BLOCK;
688 	const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
689 	const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
690 	int n = 0;
691 	int level = 0;
692 
693 	noffset[0] = 0;
694 
695 	if (block < direct_index) {
696 		offset[n] = block;
697 		goto got;
698 	}
699 	block -= direct_index;
700 	if (block < direct_blks) {
701 		offset[n++] = NODE_DIR1_BLOCK;
702 		noffset[n] = 1;
703 		offset[n] = block;
704 		level = 1;
705 		goto got;
706 	}
707 	block -= direct_blks;
708 	if (block < direct_blks) {
709 		offset[n++] = NODE_DIR2_BLOCK;
710 		noffset[n] = 2;
711 		offset[n] = block;
712 		level = 1;
713 		goto got;
714 	}
715 	block -= direct_blks;
716 	if (block < indirect_blks) {
717 		offset[n++] = NODE_IND1_BLOCK;
718 		noffset[n] = 3;
719 		offset[n++] = block / direct_blks;
720 		noffset[n] = 4 + offset[n - 1];
721 		offset[n] = block % direct_blks;
722 		level = 2;
723 		goto got;
724 	}
725 	block -= indirect_blks;
726 	if (block < indirect_blks) {
727 		offset[n++] = NODE_IND2_BLOCK;
728 		noffset[n] = 4 + dptrs_per_blk;
729 		offset[n++] = block / direct_blks;
730 		noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
731 		offset[n] = block % direct_blks;
732 		level = 2;
733 		goto got;
734 	}
735 	block -= indirect_blks;
736 	if (block < dindirect_blks) {
737 		offset[n++] = NODE_DIND_BLOCK;
738 		noffset[n] = 5 + (dptrs_per_blk * 2);
739 		offset[n++] = block / indirect_blks;
740 		noffset[n] = 6 + (dptrs_per_blk * 2) +
741 			      offset[n - 1] * (dptrs_per_blk + 1);
742 		offset[n++] = (block / direct_blks) % dptrs_per_blk;
743 		noffset[n] = 7 + (dptrs_per_blk * 2) +
744 			      offset[n - 2] * (dptrs_per_blk + 1) +
745 			      offset[n - 1];
746 		offset[n] = block % direct_blks;
747 		level = 3;
748 		goto got;
749 	} else {
750 		return -E2BIG;
751 	}
752 got:
753 	return level;
754 }
755 
756 /*
757  * Caller should call f2fs_put_dnode(dn).
758  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
759  * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
760  */
761 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
762 {
763 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
764 	struct page *npage[4];
765 	struct page *parent = NULL;
766 	int offset[4];
767 	unsigned int noffset[4];
768 	nid_t nids[4];
769 	int level, i = 0;
770 	int err = 0;
771 
772 	level = get_node_path(dn->inode, index, offset, noffset);
773 	if (level < 0)
774 		return level;
775 
776 	nids[0] = dn->inode->i_ino;
777 	npage[0] = dn->inode_page;
778 
779 	if (!npage[0]) {
780 		npage[0] = f2fs_get_node_page(sbi, nids[0]);
781 		if (IS_ERR(npage[0]))
782 			return PTR_ERR(npage[0]);
783 	}
784 
785 	/* if inline_data is set, should not report any block indices */
786 	if (f2fs_has_inline_data(dn->inode) && index) {
787 		err = -ENOENT;
788 		f2fs_put_page(npage[0], 1);
789 		goto release_out;
790 	}
791 
792 	parent = npage[0];
793 	if (level != 0)
794 		nids[1] = get_nid(parent, offset[0], true);
795 	dn->inode_page = npage[0];
796 	dn->inode_page_locked = true;
797 
798 	/* get indirect or direct nodes */
799 	for (i = 1; i <= level; i++) {
800 		bool done = false;
801 
802 		if (!nids[i] && mode == ALLOC_NODE) {
803 			/* alloc new node */
804 			if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
805 				err = -ENOSPC;
806 				goto release_pages;
807 			}
808 
809 			dn->nid = nids[i];
810 			npage[i] = f2fs_new_node_page(dn, noffset[i]);
811 			if (IS_ERR(npage[i])) {
812 				f2fs_alloc_nid_failed(sbi, nids[i]);
813 				err = PTR_ERR(npage[i]);
814 				goto release_pages;
815 			}
816 
817 			set_nid(parent, offset[i - 1], nids[i], i == 1);
818 			f2fs_alloc_nid_done(sbi, nids[i]);
819 			done = true;
820 		} else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
821 			npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
822 			if (IS_ERR(npage[i])) {
823 				err = PTR_ERR(npage[i]);
824 				goto release_pages;
825 			}
826 			done = true;
827 		}
828 		if (i == 1) {
829 			dn->inode_page_locked = false;
830 			unlock_page(parent);
831 		} else {
832 			f2fs_put_page(parent, 1);
833 		}
834 
835 		if (!done) {
836 			npage[i] = f2fs_get_node_page(sbi, nids[i]);
837 			if (IS_ERR(npage[i])) {
838 				err = PTR_ERR(npage[i]);
839 				f2fs_put_page(npage[0], 0);
840 				goto release_out;
841 			}
842 		}
843 		if (i < level) {
844 			parent = npage[i];
845 			nids[i + 1] = get_nid(parent, offset[i], false);
846 		}
847 	}
848 	dn->nid = nids[level];
849 	dn->ofs_in_node = offset[level];
850 	dn->node_page = npage[level];
851 	dn->data_blkaddr = f2fs_data_blkaddr(dn);
852 
853 	if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) &&
854 					f2fs_sb_has_readonly(sbi)) {
855 		unsigned int cluster_size = F2FS_I(dn->inode)->i_cluster_size;
856 		unsigned int ofs_in_node = dn->ofs_in_node;
857 		pgoff_t fofs = index;
858 		unsigned int c_len;
859 		block_t blkaddr;
860 
861 		/* should align fofs and ofs_in_node to cluster_size */
862 		if (fofs % cluster_size) {
863 			fofs = round_down(fofs, cluster_size);
864 			ofs_in_node = round_down(ofs_in_node, cluster_size);
865 		}
866 
867 		c_len = f2fs_cluster_blocks_are_contiguous(dn, ofs_in_node);
868 		if (!c_len)
869 			goto out;
870 
871 		blkaddr = data_blkaddr(dn->inode, dn->node_page, ofs_in_node);
872 		if (blkaddr == COMPRESS_ADDR)
873 			blkaddr = data_blkaddr(dn->inode, dn->node_page,
874 						ofs_in_node + 1);
875 
876 		f2fs_update_read_extent_tree_range_compressed(dn->inode,
877 					fofs, blkaddr, cluster_size, c_len);
878 	}
879 out:
880 	return 0;
881 
882 release_pages:
883 	f2fs_put_page(parent, 1);
884 	if (i > 1)
885 		f2fs_put_page(npage[0], 0);
886 release_out:
887 	dn->inode_page = NULL;
888 	dn->node_page = NULL;
889 	if (err == -ENOENT) {
890 		dn->cur_level = i;
891 		dn->max_level = level;
892 		dn->ofs_in_node = offset[level];
893 	}
894 	return err;
895 }
896 
897 static int truncate_node(struct dnode_of_data *dn)
898 {
899 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
900 	struct node_info ni;
901 	int err;
902 	pgoff_t index;
903 
904 	err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
905 	if (err)
906 		return err;
907 
908 	if (ni.blk_addr != NEW_ADDR &&
909 		!f2fs_is_valid_blkaddr(sbi, ni.blk_addr, DATA_GENERIC_ENHANCE)) {
910 		f2fs_err_ratelimited(sbi,
911 			"nat entry is corrupted, run fsck to fix it, ino:%u, "
912 			"nid:%u, blkaddr:%u", ni.ino, ni.nid, ni.blk_addr);
913 		set_sbi_flag(sbi, SBI_NEED_FSCK);
914 		f2fs_handle_error(sbi, ERROR_INCONSISTENT_NAT);
915 		return -EFSCORRUPTED;
916 	}
917 
918 	/* Deallocate node address */
919 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
920 	dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
921 	set_node_addr(sbi, &ni, NULL_ADDR, false);
922 
923 	if (dn->nid == dn->inode->i_ino) {
924 		f2fs_remove_orphan_inode(sbi, dn->nid);
925 		dec_valid_inode_count(sbi);
926 		f2fs_inode_synced(dn->inode);
927 	}
928 
929 	clear_node_page_dirty(dn->node_page);
930 	set_sbi_flag(sbi, SBI_IS_DIRTY);
931 
932 	index = dn->node_page->index;
933 	f2fs_put_page(dn->node_page, 1);
934 
935 	invalidate_mapping_pages(NODE_MAPPING(sbi),
936 			index, index);
937 
938 	dn->node_page = NULL;
939 	trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
940 
941 	return 0;
942 }
943 
944 static int truncate_dnode(struct dnode_of_data *dn)
945 {
946 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
947 	struct page *page;
948 	int err;
949 
950 	if (dn->nid == 0)
951 		return 1;
952 
953 	/* get direct node */
954 	page = f2fs_get_node_page(sbi, dn->nid);
955 	if (PTR_ERR(page) == -ENOENT)
956 		return 1;
957 	else if (IS_ERR(page))
958 		return PTR_ERR(page);
959 
960 	if (IS_INODE(page) || ino_of_node(page) != dn->inode->i_ino) {
961 		f2fs_err(sbi, "incorrect node reference, ino: %lu, nid: %u, ino_of_node: %u",
962 				dn->inode->i_ino, dn->nid, ino_of_node(page));
963 		set_sbi_flag(sbi, SBI_NEED_FSCK);
964 		f2fs_handle_error(sbi, ERROR_INVALID_NODE_REFERENCE);
965 		f2fs_put_page(page, 1);
966 		return -EFSCORRUPTED;
967 	}
968 
969 	/* Make dnode_of_data for parameter */
970 	dn->node_page = page;
971 	dn->ofs_in_node = 0;
972 	f2fs_truncate_data_blocks_range(dn, ADDRS_PER_BLOCK(dn->inode));
973 	err = truncate_node(dn);
974 	if (err) {
975 		f2fs_put_page(page, 1);
976 		return err;
977 	}
978 
979 	return 1;
980 }
981 
982 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
983 						int ofs, int depth)
984 {
985 	struct dnode_of_data rdn = *dn;
986 	struct page *page;
987 	struct f2fs_node *rn;
988 	nid_t child_nid;
989 	unsigned int child_nofs;
990 	int freed = 0;
991 	int i, ret;
992 
993 	if (dn->nid == 0)
994 		return NIDS_PER_BLOCK + 1;
995 
996 	trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
997 
998 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
999 	if (IS_ERR(page)) {
1000 		trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
1001 		return PTR_ERR(page);
1002 	}
1003 
1004 	f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
1005 
1006 	rn = F2FS_NODE(page);
1007 	if (depth < 3) {
1008 		for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
1009 			child_nid = le32_to_cpu(rn->in.nid[i]);
1010 			if (child_nid == 0)
1011 				continue;
1012 			rdn.nid = child_nid;
1013 			ret = truncate_dnode(&rdn);
1014 			if (ret < 0)
1015 				goto out_err;
1016 			if (set_nid(page, i, 0, false))
1017 				dn->node_changed = true;
1018 		}
1019 	} else {
1020 		child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
1021 		for (i = ofs; i < NIDS_PER_BLOCK; i++) {
1022 			child_nid = le32_to_cpu(rn->in.nid[i]);
1023 			if (child_nid == 0) {
1024 				child_nofs += NIDS_PER_BLOCK + 1;
1025 				continue;
1026 			}
1027 			rdn.nid = child_nid;
1028 			ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
1029 			if (ret == (NIDS_PER_BLOCK + 1)) {
1030 				if (set_nid(page, i, 0, false))
1031 					dn->node_changed = true;
1032 				child_nofs += ret;
1033 			} else if (ret < 0 && ret != -ENOENT) {
1034 				goto out_err;
1035 			}
1036 		}
1037 		freed = child_nofs;
1038 	}
1039 
1040 	if (!ofs) {
1041 		/* remove current indirect node */
1042 		dn->node_page = page;
1043 		ret = truncate_node(dn);
1044 		if (ret)
1045 			goto out_err;
1046 		freed++;
1047 	} else {
1048 		f2fs_put_page(page, 1);
1049 	}
1050 	trace_f2fs_truncate_nodes_exit(dn->inode, freed);
1051 	return freed;
1052 
1053 out_err:
1054 	f2fs_put_page(page, 1);
1055 	trace_f2fs_truncate_nodes_exit(dn->inode, ret);
1056 	return ret;
1057 }
1058 
1059 static int truncate_partial_nodes(struct dnode_of_data *dn,
1060 			struct f2fs_inode *ri, int *offset, int depth)
1061 {
1062 	struct page *pages[2];
1063 	nid_t nid[3];
1064 	nid_t child_nid;
1065 	int err = 0;
1066 	int i;
1067 	int idx = depth - 2;
1068 
1069 	nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1070 	if (!nid[0])
1071 		return 0;
1072 
1073 	/* get indirect nodes in the path */
1074 	for (i = 0; i < idx + 1; i++) {
1075 		/* reference count'll be increased */
1076 		pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
1077 		if (IS_ERR(pages[i])) {
1078 			err = PTR_ERR(pages[i]);
1079 			idx = i - 1;
1080 			goto fail;
1081 		}
1082 		nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1083 	}
1084 
1085 	f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1086 
1087 	/* free direct nodes linked to a partial indirect node */
1088 	for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1089 		child_nid = get_nid(pages[idx], i, false);
1090 		if (!child_nid)
1091 			continue;
1092 		dn->nid = child_nid;
1093 		err = truncate_dnode(dn);
1094 		if (err < 0)
1095 			goto fail;
1096 		if (set_nid(pages[idx], i, 0, false))
1097 			dn->node_changed = true;
1098 	}
1099 
1100 	if (offset[idx + 1] == 0) {
1101 		dn->node_page = pages[idx];
1102 		dn->nid = nid[idx];
1103 		err = truncate_node(dn);
1104 		if (err)
1105 			goto fail;
1106 	} else {
1107 		f2fs_put_page(pages[idx], 1);
1108 	}
1109 	offset[idx]++;
1110 	offset[idx + 1] = 0;
1111 	idx--;
1112 fail:
1113 	for (i = idx; i >= 0; i--)
1114 		f2fs_put_page(pages[i], 1);
1115 
1116 	trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1117 
1118 	return err;
1119 }
1120 
1121 /*
1122  * All the block addresses of data and nodes should be nullified.
1123  */
1124 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1125 {
1126 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1127 	int err = 0, cont = 1;
1128 	int level, offset[4], noffset[4];
1129 	unsigned int nofs = 0;
1130 	struct f2fs_inode *ri;
1131 	struct dnode_of_data dn;
1132 	struct page *page;
1133 
1134 	trace_f2fs_truncate_inode_blocks_enter(inode, from);
1135 
1136 	level = get_node_path(inode, from, offset, noffset);
1137 	if (level < 0) {
1138 		trace_f2fs_truncate_inode_blocks_exit(inode, level);
1139 		return level;
1140 	}
1141 
1142 	page = f2fs_get_node_page(sbi, inode->i_ino);
1143 	if (IS_ERR(page)) {
1144 		trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1145 		return PTR_ERR(page);
1146 	}
1147 
1148 	set_new_dnode(&dn, inode, page, NULL, 0);
1149 	unlock_page(page);
1150 
1151 	ri = F2FS_INODE(page);
1152 	switch (level) {
1153 	case 0:
1154 	case 1:
1155 		nofs = noffset[1];
1156 		break;
1157 	case 2:
1158 		nofs = noffset[1];
1159 		if (!offset[level - 1])
1160 			goto skip_partial;
1161 		err = truncate_partial_nodes(&dn, ri, offset, level);
1162 		if (err < 0 && err != -ENOENT)
1163 			goto fail;
1164 		nofs += 1 + NIDS_PER_BLOCK;
1165 		break;
1166 	case 3:
1167 		nofs = 5 + 2 * NIDS_PER_BLOCK;
1168 		if (!offset[level - 1])
1169 			goto skip_partial;
1170 		err = truncate_partial_nodes(&dn, ri, offset, level);
1171 		if (err < 0 && err != -ENOENT)
1172 			goto fail;
1173 		break;
1174 	default:
1175 		BUG();
1176 	}
1177 
1178 skip_partial:
1179 	while (cont) {
1180 		dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1181 		switch (offset[0]) {
1182 		case NODE_DIR1_BLOCK:
1183 		case NODE_DIR2_BLOCK:
1184 			err = truncate_dnode(&dn);
1185 			break;
1186 
1187 		case NODE_IND1_BLOCK:
1188 		case NODE_IND2_BLOCK:
1189 			err = truncate_nodes(&dn, nofs, offset[1], 2);
1190 			break;
1191 
1192 		case NODE_DIND_BLOCK:
1193 			err = truncate_nodes(&dn, nofs, offset[1], 3);
1194 			cont = 0;
1195 			break;
1196 
1197 		default:
1198 			BUG();
1199 		}
1200 		if (err < 0 && err != -ENOENT)
1201 			goto fail;
1202 		if (offset[1] == 0 &&
1203 				ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1204 			lock_page(page);
1205 			BUG_ON(page->mapping != NODE_MAPPING(sbi));
1206 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1207 			ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1208 			set_page_dirty(page);
1209 			unlock_page(page);
1210 		}
1211 		offset[1] = 0;
1212 		offset[0]++;
1213 		nofs += err;
1214 	}
1215 fail:
1216 	f2fs_put_page(page, 0);
1217 	trace_f2fs_truncate_inode_blocks_exit(inode, err);
1218 	return err > 0 ? 0 : err;
1219 }
1220 
1221 /* caller must lock inode page */
1222 int f2fs_truncate_xattr_node(struct inode *inode)
1223 {
1224 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1225 	nid_t nid = F2FS_I(inode)->i_xattr_nid;
1226 	struct dnode_of_data dn;
1227 	struct page *npage;
1228 	int err;
1229 
1230 	if (!nid)
1231 		return 0;
1232 
1233 	npage = f2fs_get_node_page(sbi, nid);
1234 	if (IS_ERR(npage))
1235 		return PTR_ERR(npage);
1236 
1237 	set_new_dnode(&dn, inode, NULL, npage, nid);
1238 	err = truncate_node(&dn);
1239 	if (err) {
1240 		f2fs_put_page(npage, 1);
1241 		return err;
1242 	}
1243 
1244 	f2fs_i_xnid_write(inode, 0);
1245 
1246 	return 0;
1247 }
1248 
1249 /*
1250  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1251  * f2fs_unlock_op().
1252  */
1253 int f2fs_remove_inode_page(struct inode *inode)
1254 {
1255 	struct dnode_of_data dn;
1256 	int err;
1257 
1258 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1259 	err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1260 	if (err)
1261 		return err;
1262 
1263 	err = f2fs_truncate_xattr_node(inode);
1264 	if (err) {
1265 		f2fs_put_dnode(&dn);
1266 		return err;
1267 	}
1268 
1269 	/* remove potential inline_data blocks */
1270 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1271 				S_ISLNK(inode->i_mode))
1272 		f2fs_truncate_data_blocks_range(&dn, 1);
1273 
1274 	/* 0 is possible, after f2fs_new_inode() has failed */
1275 	if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1276 		f2fs_put_dnode(&dn);
1277 		return -EIO;
1278 	}
1279 
1280 	if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1281 		f2fs_warn(F2FS_I_SB(inode),
1282 			"f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1283 			inode->i_ino, (unsigned long long)inode->i_blocks);
1284 		set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1285 	}
1286 
1287 	/* will put inode & node pages */
1288 	err = truncate_node(&dn);
1289 	if (err) {
1290 		f2fs_put_dnode(&dn);
1291 		return err;
1292 	}
1293 	return 0;
1294 }
1295 
1296 struct page *f2fs_new_inode_page(struct inode *inode)
1297 {
1298 	struct dnode_of_data dn;
1299 
1300 	/* allocate inode page for new inode */
1301 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1302 
1303 	/* caller should f2fs_put_page(page, 1); */
1304 	return f2fs_new_node_page(&dn, 0);
1305 }
1306 
1307 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1308 {
1309 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1310 	struct node_info new_ni;
1311 	struct page *page;
1312 	int err;
1313 
1314 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1315 		return ERR_PTR(-EPERM);
1316 
1317 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1318 	if (!page)
1319 		return ERR_PTR(-ENOMEM);
1320 
1321 	if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1322 		goto fail;
1323 
1324 #ifdef CONFIG_F2FS_CHECK_FS
1325 	err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false);
1326 	if (err) {
1327 		dec_valid_node_count(sbi, dn->inode, !ofs);
1328 		goto fail;
1329 	}
1330 	if (unlikely(new_ni.blk_addr != NULL_ADDR)) {
1331 		err = -EFSCORRUPTED;
1332 		dec_valid_node_count(sbi, dn->inode, !ofs);
1333 		set_sbi_flag(sbi, SBI_NEED_FSCK);
1334 		f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1335 		goto fail;
1336 	}
1337 #endif
1338 	new_ni.nid = dn->nid;
1339 	new_ni.ino = dn->inode->i_ino;
1340 	new_ni.blk_addr = NULL_ADDR;
1341 	new_ni.flag = 0;
1342 	new_ni.version = 0;
1343 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1344 
1345 	f2fs_wait_on_page_writeback(page, NODE, true, true);
1346 	fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1347 	set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1348 	if (!PageUptodate(page))
1349 		SetPageUptodate(page);
1350 	if (set_page_dirty(page))
1351 		dn->node_changed = true;
1352 
1353 	if (f2fs_has_xattr_block(ofs))
1354 		f2fs_i_xnid_write(dn->inode, dn->nid);
1355 
1356 	if (ofs == 0)
1357 		inc_valid_inode_count(sbi);
1358 	return page;
1359 fail:
1360 	clear_node_page_dirty(page);
1361 	f2fs_put_page(page, 1);
1362 	return ERR_PTR(err);
1363 }
1364 
1365 /*
1366  * Caller should do after getting the following values.
1367  * 0: f2fs_put_page(page, 0)
1368  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1369  */
1370 static int read_node_page(struct page *page, blk_opf_t op_flags)
1371 {
1372 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1373 	struct node_info ni;
1374 	struct f2fs_io_info fio = {
1375 		.sbi = sbi,
1376 		.type = NODE,
1377 		.op = REQ_OP_READ,
1378 		.op_flags = op_flags,
1379 		.page = page,
1380 		.encrypted_page = NULL,
1381 	};
1382 	int err;
1383 
1384 	if (PageUptodate(page)) {
1385 		if (!f2fs_inode_chksum_verify(sbi, page)) {
1386 			ClearPageUptodate(page);
1387 			return -EFSBADCRC;
1388 		}
1389 		return LOCKED_PAGE;
1390 	}
1391 
1392 	err = f2fs_get_node_info(sbi, page->index, &ni, false);
1393 	if (err)
1394 		return err;
1395 
1396 	/* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1397 	if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR)) {
1398 		ClearPageUptodate(page);
1399 		return -ENOENT;
1400 	}
1401 
1402 	fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1403 
1404 	err = f2fs_submit_page_bio(&fio);
1405 
1406 	if (!err)
1407 		f2fs_update_iostat(sbi, NULL, FS_NODE_READ_IO, F2FS_BLKSIZE);
1408 
1409 	return err;
1410 }
1411 
1412 /*
1413  * Readahead a node page
1414  */
1415 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1416 {
1417 	struct page *apage;
1418 	int err;
1419 
1420 	if (!nid)
1421 		return;
1422 	if (f2fs_check_nid_range(sbi, nid))
1423 		return;
1424 
1425 	apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1426 	if (apage)
1427 		return;
1428 
1429 	apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1430 	if (!apage)
1431 		return;
1432 
1433 	err = read_node_page(apage, REQ_RAHEAD);
1434 	f2fs_put_page(apage, err ? 1 : 0);
1435 }
1436 
1437 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1438 					struct page *parent, int start)
1439 {
1440 	struct page *page;
1441 	int err;
1442 
1443 	if (!nid)
1444 		return ERR_PTR(-ENOENT);
1445 	if (f2fs_check_nid_range(sbi, nid))
1446 		return ERR_PTR(-EINVAL);
1447 repeat:
1448 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1449 	if (!page)
1450 		return ERR_PTR(-ENOMEM);
1451 
1452 	err = read_node_page(page, 0);
1453 	if (err < 0) {
1454 		goto out_put_err;
1455 	} else if (err == LOCKED_PAGE) {
1456 		err = 0;
1457 		goto page_hit;
1458 	}
1459 
1460 	if (parent)
1461 		f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1462 
1463 	lock_page(page);
1464 
1465 	if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1466 		f2fs_put_page(page, 1);
1467 		goto repeat;
1468 	}
1469 
1470 	if (unlikely(!PageUptodate(page))) {
1471 		err = -EIO;
1472 		goto out_err;
1473 	}
1474 
1475 	if (!f2fs_inode_chksum_verify(sbi, page)) {
1476 		err = -EFSBADCRC;
1477 		goto out_err;
1478 	}
1479 page_hit:
1480 	if (likely(nid == nid_of_node(page)))
1481 		return page;
1482 
1483 	f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1484 			  nid, nid_of_node(page), ino_of_node(page),
1485 			  ofs_of_node(page), cpver_of_node(page),
1486 			  next_blkaddr_of_node(page));
1487 	set_sbi_flag(sbi, SBI_NEED_FSCK);
1488 	f2fs_handle_error(sbi, ERROR_INCONSISTENT_FOOTER);
1489 	err = -EFSCORRUPTED;
1490 out_err:
1491 	ClearPageUptodate(page);
1492 out_put_err:
1493 	/* ENOENT comes from read_node_page which is not an error. */
1494 	if (err != -ENOENT)
1495 		f2fs_handle_page_eio(sbi, page->index, NODE);
1496 	f2fs_put_page(page, 1);
1497 	return ERR_PTR(err);
1498 }
1499 
1500 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1501 {
1502 	return __get_node_page(sbi, nid, NULL, 0);
1503 }
1504 
1505 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1506 {
1507 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1508 	nid_t nid = get_nid(parent, start, false);
1509 
1510 	return __get_node_page(sbi, nid, parent, start);
1511 }
1512 
1513 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1514 {
1515 	struct inode *inode;
1516 	struct page *page;
1517 	int ret;
1518 
1519 	/* should flush inline_data before evict_inode */
1520 	inode = ilookup(sbi->sb, ino);
1521 	if (!inode)
1522 		return;
1523 
1524 	page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1525 					FGP_LOCK|FGP_NOWAIT, 0);
1526 	if (!page)
1527 		goto iput_out;
1528 
1529 	if (!PageUptodate(page))
1530 		goto page_out;
1531 
1532 	if (!PageDirty(page))
1533 		goto page_out;
1534 
1535 	if (!clear_page_dirty_for_io(page))
1536 		goto page_out;
1537 
1538 	ret = f2fs_write_inline_data(inode, page);
1539 	inode_dec_dirty_pages(inode);
1540 	f2fs_remove_dirty_inode(inode);
1541 	if (ret)
1542 		set_page_dirty(page);
1543 page_out:
1544 	f2fs_put_page(page, 1);
1545 iput_out:
1546 	iput(inode);
1547 }
1548 
1549 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1550 {
1551 	pgoff_t index;
1552 	struct folio_batch fbatch;
1553 	struct page *last_page = NULL;
1554 	int nr_folios;
1555 
1556 	folio_batch_init(&fbatch);
1557 	index = 0;
1558 
1559 	while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1560 					(pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1561 					&fbatch))) {
1562 		int i;
1563 
1564 		for (i = 0; i < nr_folios; i++) {
1565 			struct page *page = &fbatch.folios[i]->page;
1566 
1567 			if (unlikely(f2fs_cp_error(sbi))) {
1568 				f2fs_put_page(last_page, 0);
1569 				folio_batch_release(&fbatch);
1570 				return ERR_PTR(-EIO);
1571 			}
1572 
1573 			if (!IS_DNODE(page) || !is_cold_node(page))
1574 				continue;
1575 			if (ino_of_node(page) != ino)
1576 				continue;
1577 
1578 			lock_page(page);
1579 
1580 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1581 continue_unlock:
1582 				unlock_page(page);
1583 				continue;
1584 			}
1585 			if (ino_of_node(page) != ino)
1586 				goto continue_unlock;
1587 
1588 			if (!PageDirty(page)) {
1589 				/* someone wrote it for us */
1590 				goto continue_unlock;
1591 			}
1592 
1593 			if (last_page)
1594 				f2fs_put_page(last_page, 0);
1595 
1596 			get_page(page);
1597 			last_page = page;
1598 			unlock_page(page);
1599 		}
1600 		folio_batch_release(&fbatch);
1601 		cond_resched();
1602 	}
1603 	return last_page;
1604 }
1605 
1606 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1607 				struct writeback_control *wbc, bool do_balance,
1608 				enum iostat_type io_type, unsigned int *seq_id)
1609 {
1610 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1611 	nid_t nid;
1612 	struct node_info ni;
1613 	struct f2fs_io_info fio = {
1614 		.sbi = sbi,
1615 		.ino = ino_of_node(page),
1616 		.type = NODE,
1617 		.op = REQ_OP_WRITE,
1618 		.op_flags = wbc_to_write_flags(wbc),
1619 		.page = page,
1620 		.encrypted_page = NULL,
1621 		.submitted = 0,
1622 		.io_type = io_type,
1623 		.io_wbc = wbc,
1624 	};
1625 	unsigned int seq;
1626 
1627 	trace_f2fs_writepage(page, NODE);
1628 
1629 	if (unlikely(f2fs_cp_error(sbi))) {
1630 		/* keep node pages in remount-ro mode */
1631 		if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY)
1632 			goto redirty_out;
1633 		ClearPageUptodate(page);
1634 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1635 		unlock_page(page);
1636 		return 0;
1637 	}
1638 
1639 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1640 		goto redirty_out;
1641 
1642 	if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1643 			wbc->sync_mode == WB_SYNC_NONE &&
1644 			IS_DNODE(page) && is_cold_node(page))
1645 		goto redirty_out;
1646 
1647 	/* get old block addr of this node page */
1648 	nid = nid_of_node(page);
1649 	f2fs_bug_on(sbi, page->index != nid);
1650 
1651 	if (f2fs_get_node_info(sbi, nid, &ni, !do_balance))
1652 		goto redirty_out;
1653 
1654 	if (wbc->for_reclaim) {
1655 		if (!f2fs_down_read_trylock(&sbi->node_write))
1656 			goto redirty_out;
1657 	} else {
1658 		f2fs_down_read(&sbi->node_write);
1659 	}
1660 
1661 	/* This page is already truncated */
1662 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1663 		ClearPageUptodate(page);
1664 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1665 		f2fs_up_read(&sbi->node_write);
1666 		unlock_page(page);
1667 		return 0;
1668 	}
1669 
1670 	if (__is_valid_data_blkaddr(ni.blk_addr) &&
1671 		!f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1672 					DATA_GENERIC_ENHANCE)) {
1673 		f2fs_up_read(&sbi->node_write);
1674 		goto redirty_out;
1675 	}
1676 
1677 	if (atomic && !test_opt(sbi, NOBARRIER) && !f2fs_sb_has_blkzoned(sbi))
1678 		fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1679 
1680 	/* should add to global list before clearing PAGECACHE status */
1681 	if (f2fs_in_warm_node_list(sbi, page)) {
1682 		seq = f2fs_add_fsync_node_entry(sbi, page);
1683 		if (seq_id)
1684 			*seq_id = seq;
1685 	}
1686 
1687 	set_page_writeback(page);
1688 
1689 	fio.old_blkaddr = ni.blk_addr;
1690 	f2fs_do_write_node_page(nid, &fio);
1691 	set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1692 	dec_page_count(sbi, F2FS_DIRTY_NODES);
1693 	f2fs_up_read(&sbi->node_write);
1694 
1695 	if (wbc->for_reclaim) {
1696 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1697 		submitted = NULL;
1698 	}
1699 
1700 	unlock_page(page);
1701 
1702 	if (unlikely(f2fs_cp_error(sbi))) {
1703 		f2fs_submit_merged_write(sbi, NODE);
1704 		submitted = NULL;
1705 	}
1706 	if (submitted)
1707 		*submitted = fio.submitted;
1708 
1709 	if (do_balance)
1710 		f2fs_balance_fs(sbi, false);
1711 	return 0;
1712 
1713 redirty_out:
1714 	redirty_page_for_writepage(wbc, page);
1715 	return AOP_WRITEPAGE_ACTIVATE;
1716 }
1717 
1718 int f2fs_move_node_page(struct page *node_page, int gc_type)
1719 {
1720 	int err = 0;
1721 
1722 	if (gc_type == FG_GC) {
1723 		struct writeback_control wbc = {
1724 			.sync_mode = WB_SYNC_ALL,
1725 			.nr_to_write = 1,
1726 			.for_reclaim = 0,
1727 		};
1728 
1729 		f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1730 
1731 		set_page_dirty(node_page);
1732 
1733 		if (!clear_page_dirty_for_io(node_page)) {
1734 			err = -EAGAIN;
1735 			goto out_page;
1736 		}
1737 
1738 		if (__write_node_page(node_page, false, NULL,
1739 					&wbc, false, FS_GC_NODE_IO, NULL)) {
1740 			err = -EAGAIN;
1741 			unlock_page(node_page);
1742 		}
1743 		goto release_page;
1744 	} else {
1745 		/* set page dirty and write it */
1746 		if (!PageWriteback(node_page))
1747 			set_page_dirty(node_page);
1748 	}
1749 out_page:
1750 	unlock_page(node_page);
1751 release_page:
1752 	f2fs_put_page(node_page, 0);
1753 	return err;
1754 }
1755 
1756 static int f2fs_write_node_page(struct page *page,
1757 				struct writeback_control *wbc)
1758 {
1759 	return __write_node_page(page, false, NULL, wbc, false,
1760 						FS_NODE_IO, NULL);
1761 }
1762 
1763 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1764 			struct writeback_control *wbc, bool atomic,
1765 			unsigned int *seq_id)
1766 {
1767 	pgoff_t index;
1768 	struct folio_batch fbatch;
1769 	int ret = 0;
1770 	struct page *last_page = NULL;
1771 	bool marked = false;
1772 	nid_t ino = inode->i_ino;
1773 	int nr_folios;
1774 	int nwritten = 0;
1775 
1776 	if (atomic) {
1777 		last_page = last_fsync_dnode(sbi, ino);
1778 		if (IS_ERR_OR_NULL(last_page))
1779 			return PTR_ERR_OR_ZERO(last_page);
1780 	}
1781 retry:
1782 	folio_batch_init(&fbatch);
1783 	index = 0;
1784 
1785 	while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1786 					(pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1787 					&fbatch))) {
1788 		int i;
1789 
1790 		for (i = 0; i < nr_folios; i++) {
1791 			struct page *page = &fbatch.folios[i]->page;
1792 			bool submitted = false;
1793 
1794 			if (unlikely(f2fs_cp_error(sbi))) {
1795 				f2fs_put_page(last_page, 0);
1796 				folio_batch_release(&fbatch);
1797 				ret = -EIO;
1798 				goto out;
1799 			}
1800 
1801 			if (!IS_DNODE(page) || !is_cold_node(page))
1802 				continue;
1803 			if (ino_of_node(page) != ino)
1804 				continue;
1805 
1806 			lock_page(page);
1807 
1808 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1809 continue_unlock:
1810 				unlock_page(page);
1811 				continue;
1812 			}
1813 			if (ino_of_node(page) != ino)
1814 				goto continue_unlock;
1815 
1816 			if (!PageDirty(page) && page != last_page) {
1817 				/* someone wrote it for us */
1818 				goto continue_unlock;
1819 			}
1820 
1821 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1822 
1823 			set_fsync_mark(page, 0);
1824 			set_dentry_mark(page, 0);
1825 
1826 			if (!atomic || page == last_page) {
1827 				set_fsync_mark(page, 1);
1828 				percpu_counter_inc(&sbi->rf_node_block_count);
1829 				if (IS_INODE(page)) {
1830 					if (is_inode_flag_set(inode,
1831 								FI_DIRTY_INODE))
1832 						f2fs_update_inode(inode, page);
1833 					set_dentry_mark(page,
1834 						f2fs_need_dentry_mark(sbi, ino));
1835 				}
1836 				/* may be written by other thread */
1837 				if (!PageDirty(page))
1838 					set_page_dirty(page);
1839 			}
1840 
1841 			if (!clear_page_dirty_for_io(page))
1842 				goto continue_unlock;
1843 
1844 			ret = __write_node_page(page, atomic &&
1845 						page == last_page,
1846 						&submitted, wbc, true,
1847 						FS_NODE_IO, seq_id);
1848 			if (ret) {
1849 				unlock_page(page);
1850 				f2fs_put_page(last_page, 0);
1851 				break;
1852 			} else if (submitted) {
1853 				nwritten++;
1854 			}
1855 
1856 			if (page == last_page) {
1857 				f2fs_put_page(page, 0);
1858 				marked = true;
1859 				break;
1860 			}
1861 		}
1862 		folio_batch_release(&fbatch);
1863 		cond_resched();
1864 
1865 		if (ret || marked)
1866 			break;
1867 	}
1868 	if (!ret && atomic && !marked) {
1869 		f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1870 			   ino, last_page->index);
1871 		lock_page(last_page);
1872 		f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1873 		set_page_dirty(last_page);
1874 		unlock_page(last_page);
1875 		goto retry;
1876 	}
1877 out:
1878 	if (nwritten)
1879 		f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1880 	return ret ? -EIO : 0;
1881 }
1882 
1883 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1884 {
1885 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1886 	bool clean;
1887 
1888 	if (inode->i_ino != ino)
1889 		return 0;
1890 
1891 	if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1892 		return 0;
1893 
1894 	spin_lock(&sbi->inode_lock[DIRTY_META]);
1895 	clean = list_empty(&F2FS_I(inode)->gdirty_list);
1896 	spin_unlock(&sbi->inode_lock[DIRTY_META]);
1897 
1898 	if (clean)
1899 		return 0;
1900 
1901 	inode = igrab(inode);
1902 	if (!inode)
1903 		return 0;
1904 	return 1;
1905 }
1906 
1907 static bool flush_dirty_inode(struct page *page)
1908 {
1909 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1910 	struct inode *inode;
1911 	nid_t ino = ino_of_node(page);
1912 
1913 	inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1914 	if (!inode)
1915 		return false;
1916 
1917 	f2fs_update_inode(inode, page);
1918 	unlock_page(page);
1919 
1920 	iput(inode);
1921 	return true;
1922 }
1923 
1924 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1925 {
1926 	pgoff_t index = 0;
1927 	struct folio_batch fbatch;
1928 	int nr_folios;
1929 
1930 	folio_batch_init(&fbatch);
1931 
1932 	while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1933 					(pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1934 					&fbatch))) {
1935 		int i;
1936 
1937 		for (i = 0; i < nr_folios; i++) {
1938 			struct page *page = &fbatch.folios[i]->page;
1939 
1940 			if (!IS_DNODE(page))
1941 				continue;
1942 
1943 			lock_page(page);
1944 
1945 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1946 continue_unlock:
1947 				unlock_page(page);
1948 				continue;
1949 			}
1950 
1951 			if (!PageDirty(page)) {
1952 				/* someone wrote it for us */
1953 				goto continue_unlock;
1954 			}
1955 
1956 			/* flush inline_data, if it's async context. */
1957 			if (page_private_inline(page)) {
1958 				clear_page_private_inline(page);
1959 				unlock_page(page);
1960 				flush_inline_data(sbi, ino_of_node(page));
1961 				continue;
1962 			}
1963 			unlock_page(page);
1964 		}
1965 		folio_batch_release(&fbatch);
1966 		cond_resched();
1967 	}
1968 }
1969 
1970 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1971 				struct writeback_control *wbc,
1972 				bool do_balance, enum iostat_type io_type)
1973 {
1974 	pgoff_t index;
1975 	struct folio_batch fbatch;
1976 	int step = 0;
1977 	int nwritten = 0;
1978 	int ret = 0;
1979 	int nr_folios, done = 0;
1980 
1981 	folio_batch_init(&fbatch);
1982 
1983 next_step:
1984 	index = 0;
1985 
1986 	while (!done && (nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi),
1987 				&index, (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1988 				&fbatch))) {
1989 		int i;
1990 
1991 		for (i = 0; i < nr_folios; i++) {
1992 			struct page *page = &fbatch.folios[i]->page;
1993 			bool submitted = false;
1994 
1995 			/* give a priority to WB_SYNC threads */
1996 			if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1997 					wbc->sync_mode == WB_SYNC_NONE) {
1998 				done = 1;
1999 				break;
2000 			}
2001 
2002 			/*
2003 			 * flushing sequence with step:
2004 			 * 0. indirect nodes
2005 			 * 1. dentry dnodes
2006 			 * 2. file dnodes
2007 			 */
2008 			if (step == 0 && IS_DNODE(page))
2009 				continue;
2010 			if (step == 1 && (!IS_DNODE(page) ||
2011 						is_cold_node(page)))
2012 				continue;
2013 			if (step == 2 && (!IS_DNODE(page) ||
2014 						!is_cold_node(page)))
2015 				continue;
2016 lock_node:
2017 			if (wbc->sync_mode == WB_SYNC_ALL)
2018 				lock_page(page);
2019 			else if (!trylock_page(page))
2020 				continue;
2021 
2022 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
2023 continue_unlock:
2024 				unlock_page(page);
2025 				continue;
2026 			}
2027 
2028 			if (!PageDirty(page)) {
2029 				/* someone wrote it for us */
2030 				goto continue_unlock;
2031 			}
2032 
2033 			/* flush inline_data/inode, if it's async context. */
2034 			if (!do_balance)
2035 				goto write_node;
2036 
2037 			/* flush inline_data */
2038 			if (page_private_inline(page)) {
2039 				clear_page_private_inline(page);
2040 				unlock_page(page);
2041 				flush_inline_data(sbi, ino_of_node(page));
2042 				goto lock_node;
2043 			}
2044 
2045 			/* flush dirty inode */
2046 			if (IS_INODE(page) && flush_dirty_inode(page))
2047 				goto lock_node;
2048 write_node:
2049 			f2fs_wait_on_page_writeback(page, NODE, true, true);
2050 
2051 			if (!clear_page_dirty_for_io(page))
2052 				goto continue_unlock;
2053 
2054 			set_fsync_mark(page, 0);
2055 			set_dentry_mark(page, 0);
2056 
2057 			ret = __write_node_page(page, false, &submitted,
2058 						wbc, do_balance, io_type, NULL);
2059 			if (ret)
2060 				unlock_page(page);
2061 			else if (submitted)
2062 				nwritten++;
2063 
2064 			if (--wbc->nr_to_write == 0)
2065 				break;
2066 		}
2067 		folio_batch_release(&fbatch);
2068 		cond_resched();
2069 
2070 		if (wbc->nr_to_write == 0) {
2071 			step = 2;
2072 			break;
2073 		}
2074 	}
2075 
2076 	if (step < 2) {
2077 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2078 				wbc->sync_mode == WB_SYNC_NONE && step == 1)
2079 			goto out;
2080 		step++;
2081 		goto next_step;
2082 	}
2083 out:
2084 	if (nwritten)
2085 		f2fs_submit_merged_write(sbi, NODE);
2086 
2087 	if (unlikely(f2fs_cp_error(sbi)))
2088 		return -EIO;
2089 	return ret;
2090 }
2091 
2092 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2093 						unsigned int seq_id)
2094 {
2095 	struct fsync_node_entry *fn;
2096 	struct page *page;
2097 	struct list_head *head = &sbi->fsync_node_list;
2098 	unsigned long flags;
2099 	unsigned int cur_seq_id = 0;
2100 
2101 	while (seq_id && cur_seq_id < seq_id) {
2102 		spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2103 		if (list_empty(head)) {
2104 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2105 			break;
2106 		}
2107 		fn = list_first_entry(head, struct fsync_node_entry, list);
2108 		if (fn->seq_id > seq_id) {
2109 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2110 			break;
2111 		}
2112 		cur_seq_id = fn->seq_id;
2113 		page = fn->page;
2114 		get_page(page);
2115 		spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2116 
2117 		f2fs_wait_on_page_writeback(page, NODE, true, false);
2118 
2119 		put_page(page);
2120 	}
2121 
2122 	return filemap_check_errors(NODE_MAPPING(sbi));
2123 }
2124 
2125 static int f2fs_write_node_pages(struct address_space *mapping,
2126 			    struct writeback_control *wbc)
2127 {
2128 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2129 	struct blk_plug plug;
2130 	long diff;
2131 
2132 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2133 		goto skip_write;
2134 
2135 	/* balancing f2fs's metadata in background */
2136 	f2fs_balance_fs_bg(sbi, true);
2137 
2138 	/* collect a number of dirty node pages and write together */
2139 	if (wbc->sync_mode != WB_SYNC_ALL &&
2140 			get_pages(sbi, F2FS_DIRTY_NODES) <
2141 					nr_pages_to_skip(sbi, NODE))
2142 		goto skip_write;
2143 
2144 	if (wbc->sync_mode == WB_SYNC_ALL)
2145 		atomic_inc(&sbi->wb_sync_req[NODE]);
2146 	else if (atomic_read(&sbi->wb_sync_req[NODE])) {
2147 		/* to avoid potential deadlock */
2148 		if (current->plug)
2149 			blk_finish_plug(current->plug);
2150 		goto skip_write;
2151 	}
2152 
2153 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2154 
2155 	diff = nr_pages_to_write(sbi, NODE, wbc);
2156 	blk_start_plug(&plug);
2157 	f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2158 	blk_finish_plug(&plug);
2159 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2160 
2161 	if (wbc->sync_mode == WB_SYNC_ALL)
2162 		atomic_dec(&sbi->wb_sync_req[NODE]);
2163 	return 0;
2164 
2165 skip_write:
2166 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2167 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2168 	return 0;
2169 }
2170 
2171 static bool f2fs_dirty_node_folio(struct address_space *mapping,
2172 		struct folio *folio)
2173 {
2174 	trace_f2fs_set_page_dirty(&folio->page, NODE);
2175 
2176 	if (!folio_test_uptodate(folio))
2177 		folio_mark_uptodate(folio);
2178 #ifdef CONFIG_F2FS_CHECK_FS
2179 	if (IS_INODE(&folio->page))
2180 		f2fs_inode_chksum_set(F2FS_M_SB(mapping), &folio->page);
2181 #endif
2182 	if (filemap_dirty_folio(mapping, folio)) {
2183 		inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
2184 		set_page_private_reference(&folio->page);
2185 		return true;
2186 	}
2187 	return false;
2188 }
2189 
2190 /*
2191  * Structure of the f2fs node operations
2192  */
2193 const struct address_space_operations f2fs_node_aops = {
2194 	.writepage	= f2fs_write_node_page,
2195 	.writepages	= f2fs_write_node_pages,
2196 	.dirty_folio	= f2fs_dirty_node_folio,
2197 	.invalidate_folio = f2fs_invalidate_folio,
2198 	.release_folio	= f2fs_release_folio,
2199 	.migrate_folio	= filemap_migrate_folio,
2200 };
2201 
2202 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2203 						nid_t n)
2204 {
2205 	return radix_tree_lookup(&nm_i->free_nid_root, n);
2206 }
2207 
2208 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2209 				struct free_nid *i)
2210 {
2211 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2212 	int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2213 
2214 	if (err)
2215 		return err;
2216 
2217 	nm_i->nid_cnt[FREE_NID]++;
2218 	list_add_tail(&i->list, &nm_i->free_nid_list);
2219 	return 0;
2220 }
2221 
2222 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2223 			struct free_nid *i, enum nid_state state)
2224 {
2225 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2226 
2227 	f2fs_bug_on(sbi, state != i->state);
2228 	nm_i->nid_cnt[state]--;
2229 	if (state == FREE_NID)
2230 		list_del(&i->list);
2231 	radix_tree_delete(&nm_i->free_nid_root, i->nid);
2232 }
2233 
2234 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2235 			enum nid_state org_state, enum nid_state dst_state)
2236 {
2237 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2238 
2239 	f2fs_bug_on(sbi, org_state != i->state);
2240 	i->state = dst_state;
2241 	nm_i->nid_cnt[org_state]--;
2242 	nm_i->nid_cnt[dst_state]++;
2243 
2244 	switch (dst_state) {
2245 	case PREALLOC_NID:
2246 		list_del(&i->list);
2247 		break;
2248 	case FREE_NID:
2249 		list_add_tail(&i->list, &nm_i->free_nid_list);
2250 		break;
2251 	default:
2252 		BUG_ON(1);
2253 	}
2254 }
2255 
2256 bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi)
2257 {
2258 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2259 	unsigned int i;
2260 	bool ret = true;
2261 
2262 	f2fs_down_read(&nm_i->nat_tree_lock);
2263 	for (i = 0; i < nm_i->nat_blocks; i++) {
2264 		if (!test_bit_le(i, nm_i->nat_block_bitmap)) {
2265 			ret = false;
2266 			break;
2267 		}
2268 	}
2269 	f2fs_up_read(&nm_i->nat_tree_lock);
2270 
2271 	return ret;
2272 }
2273 
2274 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2275 							bool set, bool build)
2276 {
2277 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2278 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2279 	unsigned int nid_ofs = nid - START_NID(nid);
2280 
2281 	if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2282 		return;
2283 
2284 	if (set) {
2285 		if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2286 			return;
2287 		__set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2288 		nm_i->free_nid_count[nat_ofs]++;
2289 	} else {
2290 		if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2291 			return;
2292 		__clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2293 		if (!build)
2294 			nm_i->free_nid_count[nat_ofs]--;
2295 	}
2296 }
2297 
2298 /* return if the nid is recognized as free */
2299 static bool add_free_nid(struct f2fs_sb_info *sbi,
2300 				nid_t nid, bool build, bool update)
2301 {
2302 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2303 	struct free_nid *i, *e;
2304 	struct nat_entry *ne;
2305 	int err = -EINVAL;
2306 	bool ret = false;
2307 
2308 	/* 0 nid should not be used */
2309 	if (unlikely(nid == 0))
2310 		return false;
2311 
2312 	if (unlikely(f2fs_check_nid_range(sbi, nid)))
2313 		return false;
2314 
2315 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL);
2316 	i->nid = nid;
2317 	i->state = FREE_NID;
2318 
2319 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2320 
2321 	spin_lock(&nm_i->nid_list_lock);
2322 
2323 	if (build) {
2324 		/*
2325 		 *   Thread A             Thread B
2326 		 *  - f2fs_create
2327 		 *   - f2fs_new_inode
2328 		 *    - f2fs_alloc_nid
2329 		 *     - __insert_nid_to_list(PREALLOC_NID)
2330 		 *                     - f2fs_balance_fs_bg
2331 		 *                      - f2fs_build_free_nids
2332 		 *                       - __f2fs_build_free_nids
2333 		 *                        - scan_nat_page
2334 		 *                         - add_free_nid
2335 		 *                          - __lookup_nat_cache
2336 		 *  - f2fs_add_link
2337 		 *   - f2fs_init_inode_metadata
2338 		 *    - f2fs_new_inode_page
2339 		 *     - f2fs_new_node_page
2340 		 *      - set_node_addr
2341 		 *  - f2fs_alloc_nid_done
2342 		 *   - __remove_nid_from_list(PREALLOC_NID)
2343 		 *                         - __insert_nid_to_list(FREE_NID)
2344 		 */
2345 		ne = __lookup_nat_cache(nm_i, nid);
2346 		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2347 				nat_get_blkaddr(ne) != NULL_ADDR))
2348 			goto err_out;
2349 
2350 		e = __lookup_free_nid_list(nm_i, nid);
2351 		if (e) {
2352 			if (e->state == FREE_NID)
2353 				ret = true;
2354 			goto err_out;
2355 		}
2356 	}
2357 	ret = true;
2358 	err = __insert_free_nid(sbi, i);
2359 err_out:
2360 	if (update) {
2361 		update_free_nid_bitmap(sbi, nid, ret, build);
2362 		if (!build)
2363 			nm_i->available_nids++;
2364 	}
2365 	spin_unlock(&nm_i->nid_list_lock);
2366 	radix_tree_preload_end();
2367 
2368 	if (err)
2369 		kmem_cache_free(free_nid_slab, i);
2370 	return ret;
2371 }
2372 
2373 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2374 {
2375 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2376 	struct free_nid *i;
2377 	bool need_free = false;
2378 
2379 	spin_lock(&nm_i->nid_list_lock);
2380 	i = __lookup_free_nid_list(nm_i, nid);
2381 	if (i && i->state == FREE_NID) {
2382 		__remove_free_nid(sbi, i, FREE_NID);
2383 		need_free = true;
2384 	}
2385 	spin_unlock(&nm_i->nid_list_lock);
2386 
2387 	if (need_free)
2388 		kmem_cache_free(free_nid_slab, i);
2389 }
2390 
2391 static int scan_nat_page(struct f2fs_sb_info *sbi,
2392 			struct page *nat_page, nid_t start_nid)
2393 {
2394 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2395 	struct f2fs_nat_block *nat_blk = page_address(nat_page);
2396 	block_t blk_addr;
2397 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2398 	int i;
2399 
2400 	__set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2401 
2402 	i = start_nid % NAT_ENTRY_PER_BLOCK;
2403 
2404 	for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2405 		if (unlikely(start_nid >= nm_i->max_nid))
2406 			break;
2407 
2408 		blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2409 
2410 		if (blk_addr == NEW_ADDR)
2411 			return -EFSCORRUPTED;
2412 
2413 		if (blk_addr == NULL_ADDR) {
2414 			add_free_nid(sbi, start_nid, true, true);
2415 		} else {
2416 			spin_lock(&NM_I(sbi)->nid_list_lock);
2417 			update_free_nid_bitmap(sbi, start_nid, false, true);
2418 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2419 		}
2420 	}
2421 
2422 	return 0;
2423 }
2424 
2425 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2426 {
2427 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2428 	struct f2fs_journal *journal = curseg->journal;
2429 	int i;
2430 
2431 	down_read(&curseg->journal_rwsem);
2432 	for (i = 0; i < nats_in_cursum(journal); i++) {
2433 		block_t addr;
2434 		nid_t nid;
2435 
2436 		addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2437 		nid = le32_to_cpu(nid_in_journal(journal, i));
2438 		if (addr == NULL_ADDR)
2439 			add_free_nid(sbi, nid, true, false);
2440 		else
2441 			remove_free_nid(sbi, nid);
2442 	}
2443 	up_read(&curseg->journal_rwsem);
2444 }
2445 
2446 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2447 {
2448 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2449 	unsigned int i, idx;
2450 	nid_t nid;
2451 
2452 	f2fs_down_read(&nm_i->nat_tree_lock);
2453 
2454 	for (i = 0; i < nm_i->nat_blocks; i++) {
2455 		if (!test_bit_le(i, nm_i->nat_block_bitmap))
2456 			continue;
2457 		if (!nm_i->free_nid_count[i])
2458 			continue;
2459 		for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2460 			idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2461 						NAT_ENTRY_PER_BLOCK, idx);
2462 			if (idx >= NAT_ENTRY_PER_BLOCK)
2463 				break;
2464 
2465 			nid = i * NAT_ENTRY_PER_BLOCK + idx;
2466 			add_free_nid(sbi, nid, true, false);
2467 
2468 			if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2469 				goto out;
2470 		}
2471 	}
2472 out:
2473 	scan_curseg_cache(sbi);
2474 
2475 	f2fs_up_read(&nm_i->nat_tree_lock);
2476 }
2477 
2478 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2479 						bool sync, bool mount)
2480 {
2481 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2482 	int i = 0, ret;
2483 	nid_t nid = nm_i->next_scan_nid;
2484 
2485 	if (unlikely(nid >= nm_i->max_nid))
2486 		nid = 0;
2487 
2488 	if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2489 		nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2490 
2491 	/* Enough entries */
2492 	if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2493 		return 0;
2494 
2495 	if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2496 		return 0;
2497 
2498 	if (!mount) {
2499 		/* try to find free nids in free_nid_bitmap */
2500 		scan_free_nid_bits(sbi);
2501 
2502 		if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2503 			return 0;
2504 	}
2505 
2506 	/* readahead nat pages to be scanned */
2507 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2508 							META_NAT, true);
2509 
2510 	f2fs_down_read(&nm_i->nat_tree_lock);
2511 
2512 	while (1) {
2513 		if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2514 						nm_i->nat_block_bitmap)) {
2515 			struct page *page = get_current_nat_page(sbi, nid);
2516 
2517 			if (IS_ERR(page)) {
2518 				ret = PTR_ERR(page);
2519 			} else {
2520 				ret = scan_nat_page(sbi, page, nid);
2521 				f2fs_put_page(page, 1);
2522 			}
2523 
2524 			if (ret) {
2525 				f2fs_up_read(&nm_i->nat_tree_lock);
2526 
2527 				if (ret == -EFSCORRUPTED) {
2528 					f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2529 					set_sbi_flag(sbi, SBI_NEED_FSCK);
2530 					f2fs_handle_error(sbi,
2531 						ERROR_INCONSISTENT_NAT);
2532 				}
2533 
2534 				return ret;
2535 			}
2536 		}
2537 
2538 		nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2539 		if (unlikely(nid >= nm_i->max_nid))
2540 			nid = 0;
2541 
2542 		if (++i >= FREE_NID_PAGES)
2543 			break;
2544 	}
2545 
2546 	/* go to the next free nat pages to find free nids abundantly */
2547 	nm_i->next_scan_nid = nid;
2548 
2549 	/* find free nids from current sum_pages */
2550 	scan_curseg_cache(sbi);
2551 
2552 	f2fs_up_read(&nm_i->nat_tree_lock);
2553 
2554 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2555 					nm_i->ra_nid_pages, META_NAT, false);
2556 
2557 	return 0;
2558 }
2559 
2560 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2561 {
2562 	int ret;
2563 
2564 	mutex_lock(&NM_I(sbi)->build_lock);
2565 	ret = __f2fs_build_free_nids(sbi, sync, mount);
2566 	mutex_unlock(&NM_I(sbi)->build_lock);
2567 
2568 	return ret;
2569 }
2570 
2571 /*
2572  * If this function returns success, caller can obtain a new nid
2573  * from second parameter of this function.
2574  * The returned nid could be used ino as well as nid when inode is created.
2575  */
2576 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2577 {
2578 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2579 	struct free_nid *i = NULL;
2580 retry:
2581 	if (time_to_inject(sbi, FAULT_ALLOC_NID))
2582 		return false;
2583 
2584 	spin_lock(&nm_i->nid_list_lock);
2585 
2586 	if (unlikely(nm_i->available_nids == 0)) {
2587 		spin_unlock(&nm_i->nid_list_lock);
2588 		return false;
2589 	}
2590 
2591 	/* We should not use stale free nids created by f2fs_build_free_nids */
2592 	if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2593 		f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2594 		i = list_first_entry(&nm_i->free_nid_list,
2595 					struct free_nid, list);
2596 		*nid = i->nid;
2597 
2598 		__move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2599 		nm_i->available_nids--;
2600 
2601 		update_free_nid_bitmap(sbi, *nid, false, false);
2602 
2603 		spin_unlock(&nm_i->nid_list_lock);
2604 		return true;
2605 	}
2606 	spin_unlock(&nm_i->nid_list_lock);
2607 
2608 	/* Let's scan nat pages and its caches to get free nids */
2609 	if (!f2fs_build_free_nids(sbi, true, false))
2610 		goto retry;
2611 	return false;
2612 }
2613 
2614 /*
2615  * f2fs_alloc_nid() should be called prior to this function.
2616  */
2617 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2618 {
2619 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2620 	struct free_nid *i;
2621 
2622 	spin_lock(&nm_i->nid_list_lock);
2623 	i = __lookup_free_nid_list(nm_i, nid);
2624 	f2fs_bug_on(sbi, !i);
2625 	__remove_free_nid(sbi, i, PREALLOC_NID);
2626 	spin_unlock(&nm_i->nid_list_lock);
2627 
2628 	kmem_cache_free(free_nid_slab, i);
2629 }
2630 
2631 /*
2632  * f2fs_alloc_nid() should be called prior to this function.
2633  */
2634 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2635 {
2636 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2637 	struct free_nid *i;
2638 	bool need_free = false;
2639 
2640 	if (!nid)
2641 		return;
2642 
2643 	spin_lock(&nm_i->nid_list_lock);
2644 	i = __lookup_free_nid_list(nm_i, nid);
2645 	f2fs_bug_on(sbi, !i);
2646 
2647 	if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2648 		__remove_free_nid(sbi, i, PREALLOC_NID);
2649 		need_free = true;
2650 	} else {
2651 		__move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2652 	}
2653 
2654 	nm_i->available_nids++;
2655 
2656 	update_free_nid_bitmap(sbi, nid, true, false);
2657 
2658 	spin_unlock(&nm_i->nid_list_lock);
2659 
2660 	if (need_free)
2661 		kmem_cache_free(free_nid_slab, i);
2662 }
2663 
2664 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2665 {
2666 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2667 	int nr = nr_shrink;
2668 
2669 	if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2670 		return 0;
2671 
2672 	if (!mutex_trylock(&nm_i->build_lock))
2673 		return 0;
2674 
2675 	while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2676 		struct free_nid *i, *next;
2677 		unsigned int batch = SHRINK_NID_BATCH_SIZE;
2678 
2679 		spin_lock(&nm_i->nid_list_lock);
2680 		list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2681 			if (!nr_shrink || !batch ||
2682 				nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2683 				break;
2684 			__remove_free_nid(sbi, i, FREE_NID);
2685 			kmem_cache_free(free_nid_slab, i);
2686 			nr_shrink--;
2687 			batch--;
2688 		}
2689 		spin_unlock(&nm_i->nid_list_lock);
2690 	}
2691 
2692 	mutex_unlock(&nm_i->build_lock);
2693 
2694 	return nr - nr_shrink;
2695 }
2696 
2697 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2698 {
2699 	void *src_addr, *dst_addr;
2700 	size_t inline_size;
2701 	struct page *ipage;
2702 	struct f2fs_inode *ri;
2703 
2704 	ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2705 	if (IS_ERR(ipage))
2706 		return PTR_ERR(ipage);
2707 
2708 	ri = F2FS_INODE(page);
2709 	if (ri->i_inline & F2FS_INLINE_XATTR) {
2710 		if (!f2fs_has_inline_xattr(inode)) {
2711 			set_inode_flag(inode, FI_INLINE_XATTR);
2712 			stat_inc_inline_xattr(inode);
2713 		}
2714 	} else {
2715 		if (f2fs_has_inline_xattr(inode)) {
2716 			stat_dec_inline_xattr(inode);
2717 			clear_inode_flag(inode, FI_INLINE_XATTR);
2718 		}
2719 		goto update_inode;
2720 	}
2721 
2722 	dst_addr = inline_xattr_addr(inode, ipage);
2723 	src_addr = inline_xattr_addr(inode, page);
2724 	inline_size = inline_xattr_size(inode);
2725 
2726 	f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2727 	memcpy(dst_addr, src_addr, inline_size);
2728 update_inode:
2729 	f2fs_update_inode(inode, ipage);
2730 	f2fs_put_page(ipage, 1);
2731 	return 0;
2732 }
2733 
2734 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2735 {
2736 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2737 	nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2738 	nid_t new_xnid;
2739 	struct dnode_of_data dn;
2740 	struct node_info ni;
2741 	struct page *xpage;
2742 	int err;
2743 
2744 	if (!prev_xnid)
2745 		goto recover_xnid;
2746 
2747 	/* 1: invalidate the previous xattr nid */
2748 	err = f2fs_get_node_info(sbi, prev_xnid, &ni, false);
2749 	if (err)
2750 		return err;
2751 
2752 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
2753 	dec_valid_node_count(sbi, inode, false);
2754 	set_node_addr(sbi, &ni, NULL_ADDR, false);
2755 
2756 recover_xnid:
2757 	/* 2: update xattr nid in inode */
2758 	if (!f2fs_alloc_nid(sbi, &new_xnid))
2759 		return -ENOSPC;
2760 
2761 	set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2762 	xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2763 	if (IS_ERR(xpage)) {
2764 		f2fs_alloc_nid_failed(sbi, new_xnid);
2765 		return PTR_ERR(xpage);
2766 	}
2767 
2768 	f2fs_alloc_nid_done(sbi, new_xnid);
2769 	f2fs_update_inode_page(inode);
2770 
2771 	/* 3: update and set xattr node page dirty */
2772 	if (page) {
2773 		memcpy(F2FS_NODE(xpage), F2FS_NODE(page),
2774 				VALID_XATTR_BLOCK_SIZE);
2775 		set_page_dirty(xpage);
2776 	}
2777 	f2fs_put_page(xpage, 1);
2778 
2779 	return 0;
2780 }
2781 
2782 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2783 {
2784 	struct f2fs_inode *src, *dst;
2785 	nid_t ino = ino_of_node(page);
2786 	struct node_info old_ni, new_ni;
2787 	struct page *ipage;
2788 	int err;
2789 
2790 	err = f2fs_get_node_info(sbi, ino, &old_ni, false);
2791 	if (err)
2792 		return err;
2793 
2794 	if (unlikely(old_ni.blk_addr != NULL_ADDR))
2795 		return -EINVAL;
2796 retry:
2797 	ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2798 	if (!ipage) {
2799 		memalloc_retry_wait(GFP_NOFS);
2800 		goto retry;
2801 	}
2802 
2803 	/* Should not use this inode from free nid list */
2804 	remove_free_nid(sbi, ino);
2805 
2806 	if (!PageUptodate(ipage))
2807 		SetPageUptodate(ipage);
2808 	fill_node_footer(ipage, ino, ino, 0, true);
2809 	set_cold_node(ipage, false);
2810 
2811 	src = F2FS_INODE(page);
2812 	dst = F2FS_INODE(ipage);
2813 
2814 	memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2815 	dst->i_size = 0;
2816 	dst->i_blocks = cpu_to_le64(1);
2817 	dst->i_links = cpu_to_le32(1);
2818 	dst->i_xattr_nid = 0;
2819 	dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2820 	if (dst->i_inline & F2FS_EXTRA_ATTR) {
2821 		dst->i_extra_isize = src->i_extra_isize;
2822 
2823 		if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2824 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2825 							i_inline_xattr_size))
2826 			dst->i_inline_xattr_size = src->i_inline_xattr_size;
2827 
2828 		if (f2fs_sb_has_project_quota(sbi) &&
2829 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2830 								i_projid))
2831 			dst->i_projid = src->i_projid;
2832 
2833 		if (f2fs_sb_has_inode_crtime(sbi) &&
2834 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2835 							i_crtime_nsec)) {
2836 			dst->i_crtime = src->i_crtime;
2837 			dst->i_crtime_nsec = src->i_crtime_nsec;
2838 		}
2839 	}
2840 
2841 	new_ni = old_ni;
2842 	new_ni.ino = ino;
2843 
2844 	if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2845 		WARN_ON(1);
2846 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2847 	inc_valid_inode_count(sbi);
2848 	set_page_dirty(ipage);
2849 	f2fs_put_page(ipage, 1);
2850 	return 0;
2851 }
2852 
2853 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2854 			unsigned int segno, struct f2fs_summary_block *sum)
2855 {
2856 	struct f2fs_node *rn;
2857 	struct f2fs_summary *sum_entry;
2858 	block_t addr;
2859 	int i, idx, last_offset, nrpages;
2860 
2861 	/* scan the node segment */
2862 	last_offset = BLKS_PER_SEG(sbi);
2863 	addr = START_BLOCK(sbi, segno);
2864 	sum_entry = &sum->entries[0];
2865 
2866 	for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2867 		nrpages = bio_max_segs(last_offset - i);
2868 
2869 		/* readahead node pages */
2870 		f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2871 
2872 		for (idx = addr; idx < addr + nrpages; idx++) {
2873 			struct page *page = f2fs_get_tmp_page(sbi, idx);
2874 
2875 			if (IS_ERR(page))
2876 				return PTR_ERR(page);
2877 
2878 			rn = F2FS_NODE(page);
2879 			sum_entry->nid = rn->footer.nid;
2880 			sum_entry->version = 0;
2881 			sum_entry->ofs_in_node = 0;
2882 			sum_entry++;
2883 			f2fs_put_page(page, 1);
2884 		}
2885 
2886 		invalidate_mapping_pages(META_MAPPING(sbi), addr,
2887 							addr + nrpages);
2888 	}
2889 	return 0;
2890 }
2891 
2892 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2893 {
2894 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2895 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2896 	struct f2fs_journal *journal = curseg->journal;
2897 	int i;
2898 
2899 	down_write(&curseg->journal_rwsem);
2900 	for (i = 0; i < nats_in_cursum(journal); i++) {
2901 		struct nat_entry *ne;
2902 		struct f2fs_nat_entry raw_ne;
2903 		nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2904 
2905 		if (f2fs_check_nid_range(sbi, nid))
2906 			continue;
2907 
2908 		raw_ne = nat_in_journal(journal, i);
2909 
2910 		ne = __lookup_nat_cache(nm_i, nid);
2911 		if (!ne) {
2912 			ne = __alloc_nat_entry(sbi, nid, true);
2913 			__init_nat_entry(nm_i, ne, &raw_ne, true);
2914 		}
2915 
2916 		/*
2917 		 * if a free nat in journal has not been used after last
2918 		 * checkpoint, we should remove it from available nids,
2919 		 * since later we will add it again.
2920 		 */
2921 		if (!get_nat_flag(ne, IS_DIRTY) &&
2922 				le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2923 			spin_lock(&nm_i->nid_list_lock);
2924 			nm_i->available_nids--;
2925 			spin_unlock(&nm_i->nid_list_lock);
2926 		}
2927 
2928 		__set_nat_cache_dirty(nm_i, ne);
2929 	}
2930 	update_nats_in_cursum(journal, -i);
2931 	up_write(&curseg->journal_rwsem);
2932 }
2933 
2934 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2935 						struct list_head *head, int max)
2936 {
2937 	struct nat_entry_set *cur;
2938 
2939 	if (nes->entry_cnt >= max)
2940 		goto add_out;
2941 
2942 	list_for_each_entry(cur, head, set_list) {
2943 		if (cur->entry_cnt >= nes->entry_cnt) {
2944 			list_add(&nes->set_list, cur->set_list.prev);
2945 			return;
2946 		}
2947 	}
2948 add_out:
2949 	list_add_tail(&nes->set_list, head);
2950 }
2951 
2952 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs,
2953 							unsigned int valid)
2954 {
2955 	if (valid == 0) {
2956 		__set_bit_le(nat_ofs, nm_i->empty_nat_bits);
2957 		__clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2958 		return;
2959 	}
2960 
2961 	__clear_bit_le(nat_ofs, nm_i->empty_nat_bits);
2962 	if (valid == NAT_ENTRY_PER_BLOCK)
2963 		__set_bit_le(nat_ofs, nm_i->full_nat_bits);
2964 	else
2965 		__clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2966 }
2967 
2968 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2969 						struct page *page)
2970 {
2971 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2972 	unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2973 	struct f2fs_nat_block *nat_blk = page_address(page);
2974 	int valid = 0;
2975 	int i = 0;
2976 
2977 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
2978 		return;
2979 
2980 	if (nat_index == 0) {
2981 		valid = 1;
2982 		i = 1;
2983 	}
2984 	for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2985 		if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2986 			valid++;
2987 	}
2988 
2989 	__update_nat_bits(nm_i, nat_index, valid);
2990 }
2991 
2992 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi)
2993 {
2994 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2995 	unsigned int nat_ofs;
2996 
2997 	f2fs_down_read(&nm_i->nat_tree_lock);
2998 
2999 	for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) {
3000 		unsigned int valid = 0, nid_ofs = 0;
3001 
3002 		/* handle nid zero due to it should never be used */
3003 		if (unlikely(nat_ofs == 0)) {
3004 			valid = 1;
3005 			nid_ofs = 1;
3006 		}
3007 
3008 		for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) {
3009 			if (!test_bit_le(nid_ofs,
3010 					nm_i->free_nid_bitmap[nat_ofs]))
3011 				valid++;
3012 		}
3013 
3014 		__update_nat_bits(nm_i, nat_ofs, valid);
3015 	}
3016 
3017 	f2fs_up_read(&nm_i->nat_tree_lock);
3018 }
3019 
3020 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
3021 		struct nat_entry_set *set, struct cp_control *cpc)
3022 {
3023 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3024 	struct f2fs_journal *journal = curseg->journal;
3025 	nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
3026 	bool to_journal = true;
3027 	struct f2fs_nat_block *nat_blk;
3028 	struct nat_entry *ne, *cur;
3029 	struct page *page = NULL;
3030 
3031 	/*
3032 	 * there are two steps to flush nat entries:
3033 	 * #1, flush nat entries to journal in current hot data summary block.
3034 	 * #2, flush nat entries to nat page.
3035 	 */
3036 	if ((cpc->reason & CP_UMOUNT) ||
3037 		!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
3038 		to_journal = false;
3039 
3040 	if (to_journal) {
3041 		down_write(&curseg->journal_rwsem);
3042 	} else {
3043 		page = get_next_nat_page(sbi, start_nid);
3044 		if (IS_ERR(page))
3045 			return PTR_ERR(page);
3046 
3047 		nat_blk = page_address(page);
3048 		f2fs_bug_on(sbi, !nat_blk);
3049 	}
3050 
3051 	/* flush dirty nats in nat entry set */
3052 	list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
3053 		struct f2fs_nat_entry *raw_ne;
3054 		nid_t nid = nat_get_nid(ne);
3055 		int offset;
3056 
3057 		f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
3058 
3059 		if (to_journal) {
3060 			offset = f2fs_lookup_journal_in_cursum(journal,
3061 							NAT_JOURNAL, nid, 1);
3062 			f2fs_bug_on(sbi, offset < 0);
3063 			raw_ne = &nat_in_journal(journal, offset);
3064 			nid_in_journal(journal, offset) = cpu_to_le32(nid);
3065 		} else {
3066 			raw_ne = &nat_blk->entries[nid - start_nid];
3067 		}
3068 		raw_nat_from_node_info(raw_ne, &ne->ni);
3069 		nat_reset_flag(ne);
3070 		__clear_nat_cache_dirty(NM_I(sbi), set, ne);
3071 		if (nat_get_blkaddr(ne) == NULL_ADDR) {
3072 			add_free_nid(sbi, nid, false, true);
3073 		} else {
3074 			spin_lock(&NM_I(sbi)->nid_list_lock);
3075 			update_free_nid_bitmap(sbi, nid, false, false);
3076 			spin_unlock(&NM_I(sbi)->nid_list_lock);
3077 		}
3078 	}
3079 
3080 	if (to_journal) {
3081 		up_write(&curseg->journal_rwsem);
3082 	} else {
3083 		update_nat_bits(sbi, start_nid, page);
3084 		f2fs_put_page(page, 1);
3085 	}
3086 
3087 	/* Allow dirty nats by node block allocation in write_begin */
3088 	if (!set->entry_cnt) {
3089 		radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
3090 		kmem_cache_free(nat_entry_set_slab, set);
3091 	}
3092 	return 0;
3093 }
3094 
3095 /*
3096  * This function is called during the checkpointing process.
3097  */
3098 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3099 {
3100 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3101 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3102 	struct f2fs_journal *journal = curseg->journal;
3103 	struct nat_entry_set *setvec[NAT_VEC_SIZE];
3104 	struct nat_entry_set *set, *tmp;
3105 	unsigned int found;
3106 	nid_t set_idx = 0;
3107 	LIST_HEAD(sets);
3108 	int err = 0;
3109 
3110 	/*
3111 	 * during unmount, let's flush nat_bits before checking
3112 	 * nat_cnt[DIRTY_NAT].
3113 	 */
3114 	if (cpc->reason & CP_UMOUNT) {
3115 		f2fs_down_write(&nm_i->nat_tree_lock);
3116 		remove_nats_in_journal(sbi);
3117 		f2fs_up_write(&nm_i->nat_tree_lock);
3118 	}
3119 
3120 	if (!nm_i->nat_cnt[DIRTY_NAT])
3121 		return 0;
3122 
3123 	f2fs_down_write(&nm_i->nat_tree_lock);
3124 
3125 	/*
3126 	 * if there are no enough space in journal to store dirty nat
3127 	 * entries, remove all entries from journal and merge them
3128 	 * into nat entry set.
3129 	 */
3130 	if (cpc->reason & CP_UMOUNT ||
3131 		!__has_cursum_space(journal,
3132 			nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3133 		remove_nats_in_journal(sbi);
3134 
3135 	while ((found = __gang_lookup_nat_set(nm_i,
3136 					set_idx, NAT_VEC_SIZE, setvec))) {
3137 		unsigned idx;
3138 
3139 		set_idx = setvec[found - 1]->set + 1;
3140 		for (idx = 0; idx < found; idx++)
3141 			__adjust_nat_entry_set(setvec[idx], &sets,
3142 						MAX_NAT_JENTRIES(journal));
3143 	}
3144 
3145 	/* flush dirty nats in nat entry set */
3146 	list_for_each_entry_safe(set, tmp, &sets, set_list) {
3147 		err = __flush_nat_entry_set(sbi, set, cpc);
3148 		if (err)
3149 			break;
3150 	}
3151 
3152 	f2fs_up_write(&nm_i->nat_tree_lock);
3153 	/* Allow dirty nats by node block allocation in write_begin */
3154 
3155 	return err;
3156 }
3157 
3158 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3159 {
3160 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3161 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3162 	unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3163 	unsigned int i;
3164 	__u64 cp_ver = cur_cp_version(ckpt);
3165 	block_t nat_bits_addr;
3166 
3167 	nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3168 	nm_i->nat_bits = f2fs_kvzalloc(sbi,
3169 			nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3170 	if (!nm_i->nat_bits)
3171 		return -ENOMEM;
3172 
3173 	nm_i->full_nat_bits = nm_i->nat_bits + 8;
3174 	nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3175 
3176 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3177 		return 0;
3178 
3179 	nat_bits_addr = __start_cp_addr(sbi) + BLKS_PER_SEG(sbi) -
3180 						nm_i->nat_bits_blocks;
3181 	for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3182 		struct page *page;
3183 
3184 		page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3185 		if (IS_ERR(page))
3186 			return PTR_ERR(page);
3187 
3188 		memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3189 					page_address(page), F2FS_BLKSIZE);
3190 		f2fs_put_page(page, 1);
3191 	}
3192 
3193 	cp_ver |= (cur_cp_crc(ckpt) << 32);
3194 	if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3195 		clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
3196 		f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)",
3197 			cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits));
3198 		return 0;
3199 	}
3200 
3201 	f2fs_notice(sbi, "Found nat_bits in checkpoint");
3202 	return 0;
3203 }
3204 
3205 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3206 {
3207 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3208 	unsigned int i = 0;
3209 	nid_t nid, last_nid;
3210 
3211 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3212 		return;
3213 
3214 	for (i = 0; i < nm_i->nat_blocks; i++) {
3215 		i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3216 		if (i >= nm_i->nat_blocks)
3217 			break;
3218 
3219 		__set_bit_le(i, nm_i->nat_block_bitmap);
3220 
3221 		nid = i * NAT_ENTRY_PER_BLOCK;
3222 		last_nid = nid + NAT_ENTRY_PER_BLOCK;
3223 
3224 		spin_lock(&NM_I(sbi)->nid_list_lock);
3225 		for (; nid < last_nid; nid++)
3226 			update_free_nid_bitmap(sbi, nid, true, true);
3227 		spin_unlock(&NM_I(sbi)->nid_list_lock);
3228 	}
3229 
3230 	for (i = 0; i < nm_i->nat_blocks; i++) {
3231 		i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3232 		if (i >= nm_i->nat_blocks)
3233 			break;
3234 
3235 		__set_bit_le(i, nm_i->nat_block_bitmap);
3236 	}
3237 }
3238 
3239 static int init_node_manager(struct f2fs_sb_info *sbi)
3240 {
3241 	struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3242 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3243 	unsigned char *version_bitmap;
3244 	unsigned int nat_segs;
3245 	int err;
3246 
3247 	nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3248 
3249 	/* segment_count_nat includes pair segment so divide to 2. */
3250 	nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3251 	nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3252 	nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3253 
3254 	/* not used nids: 0, node, meta, (and root counted as valid node) */
3255 	nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3256 						F2FS_RESERVED_NODE_NUM;
3257 	nm_i->nid_cnt[FREE_NID] = 0;
3258 	nm_i->nid_cnt[PREALLOC_NID] = 0;
3259 	nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3260 	nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3261 	nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3262 	nm_i->max_rf_node_blocks = DEF_RF_NODE_BLOCKS;
3263 
3264 	INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3265 	INIT_LIST_HEAD(&nm_i->free_nid_list);
3266 	INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3267 	INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3268 	INIT_LIST_HEAD(&nm_i->nat_entries);
3269 	spin_lock_init(&nm_i->nat_list_lock);
3270 
3271 	mutex_init(&nm_i->build_lock);
3272 	spin_lock_init(&nm_i->nid_list_lock);
3273 	init_f2fs_rwsem(&nm_i->nat_tree_lock);
3274 
3275 	nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3276 	nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3277 	version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3278 	nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3279 					GFP_KERNEL);
3280 	if (!nm_i->nat_bitmap)
3281 		return -ENOMEM;
3282 
3283 	err = __get_nat_bitmaps(sbi);
3284 	if (err)
3285 		return err;
3286 
3287 #ifdef CONFIG_F2FS_CHECK_FS
3288 	nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3289 					GFP_KERNEL);
3290 	if (!nm_i->nat_bitmap_mir)
3291 		return -ENOMEM;
3292 #endif
3293 
3294 	return 0;
3295 }
3296 
3297 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3298 {
3299 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3300 	int i;
3301 
3302 	nm_i->free_nid_bitmap =
3303 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3304 					      nm_i->nat_blocks),
3305 			      GFP_KERNEL);
3306 	if (!nm_i->free_nid_bitmap)
3307 		return -ENOMEM;
3308 
3309 	for (i = 0; i < nm_i->nat_blocks; i++) {
3310 		nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3311 			f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3312 		if (!nm_i->free_nid_bitmap[i])
3313 			return -ENOMEM;
3314 	}
3315 
3316 	nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3317 								GFP_KERNEL);
3318 	if (!nm_i->nat_block_bitmap)
3319 		return -ENOMEM;
3320 
3321 	nm_i->free_nid_count =
3322 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3323 					      nm_i->nat_blocks),
3324 			      GFP_KERNEL);
3325 	if (!nm_i->free_nid_count)
3326 		return -ENOMEM;
3327 	return 0;
3328 }
3329 
3330 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3331 {
3332 	int err;
3333 
3334 	sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3335 							GFP_KERNEL);
3336 	if (!sbi->nm_info)
3337 		return -ENOMEM;
3338 
3339 	err = init_node_manager(sbi);
3340 	if (err)
3341 		return err;
3342 
3343 	err = init_free_nid_cache(sbi);
3344 	if (err)
3345 		return err;
3346 
3347 	/* load free nid status from nat_bits table */
3348 	load_free_nid_bitmap(sbi);
3349 
3350 	return f2fs_build_free_nids(sbi, true, true);
3351 }
3352 
3353 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3354 {
3355 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3356 	struct free_nid *i, *next_i;
3357 	void *vec[NAT_VEC_SIZE];
3358 	struct nat_entry **natvec = (struct nat_entry **)vec;
3359 	struct nat_entry_set **setvec = (struct nat_entry_set **)vec;
3360 	nid_t nid = 0;
3361 	unsigned int found;
3362 
3363 	if (!nm_i)
3364 		return;
3365 
3366 	/* destroy free nid list */
3367 	spin_lock(&nm_i->nid_list_lock);
3368 	list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3369 		__remove_free_nid(sbi, i, FREE_NID);
3370 		spin_unlock(&nm_i->nid_list_lock);
3371 		kmem_cache_free(free_nid_slab, i);
3372 		spin_lock(&nm_i->nid_list_lock);
3373 	}
3374 	f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3375 	f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3376 	f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3377 	spin_unlock(&nm_i->nid_list_lock);
3378 
3379 	/* destroy nat cache */
3380 	f2fs_down_write(&nm_i->nat_tree_lock);
3381 	while ((found = __gang_lookup_nat_cache(nm_i,
3382 					nid, NAT_VEC_SIZE, natvec))) {
3383 		unsigned idx;
3384 
3385 		nid = nat_get_nid(natvec[found - 1]) + 1;
3386 		for (idx = 0; idx < found; idx++) {
3387 			spin_lock(&nm_i->nat_list_lock);
3388 			list_del(&natvec[idx]->list);
3389 			spin_unlock(&nm_i->nat_list_lock);
3390 
3391 			__del_from_nat_cache(nm_i, natvec[idx]);
3392 		}
3393 	}
3394 	f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3395 
3396 	/* destroy nat set cache */
3397 	nid = 0;
3398 	memset(vec, 0, sizeof(void *) * NAT_VEC_SIZE);
3399 	while ((found = __gang_lookup_nat_set(nm_i,
3400 					nid, NAT_VEC_SIZE, setvec))) {
3401 		unsigned idx;
3402 
3403 		nid = setvec[found - 1]->set + 1;
3404 		for (idx = 0; idx < found; idx++) {
3405 			/* entry_cnt is not zero, when cp_error was occurred */
3406 			f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3407 			radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3408 			kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3409 		}
3410 	}
3411 	f2fs_up_write(&nm_i->nat_tree_lock);
3412 
3413 	kvfree(nm_i->nat_block_bitmap);
3414 	if (nm_i->free_nid_bitmap) {
3415 		int i;
3416 
3417 		for (i = 0; i < nm_i->nat_blocks; i++)
3418 			kvfree(nm_i->free_nid_bitmap[i]);
3419 		kvfree(nm_i->free_nid_bitmap);
3420 	}
3421 	kvfree(nm_i->free_nid_count);
3422 
3423 	kvfree(nm_i->nat_bitmap);
3424 	kvfree(nm_i->nat_bits);
3425 #ifdef CONFIG_F2FS_CHECK_FS
3426 	kvfree(nm_i->nat_bitmap_mir);
3427 #endif
3428 	sbi->nm_info = NULL;
3429 	kfree(nm_i);
3430 }
3431 
3432 int __init f2fs_create_node_manager_caches(void)
3433 {
3434 	nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3435 			sizeof(struct nat_entry));
3436 	if (!nat_entry_slab)
3437 		goto fail;
3438 
3439 	free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3440 			sizeof(struct free_nid));
3441 	if (!free_nid_slab)
3442 		goto destroy_nat_entry;
3443 
3444 	nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3445 			sizeof(struct nat_entry_set));
3446 	if (!nat_entry_set_slab)
3447 		goto destroy_free_nid;
3448 
3449 	fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3450 			sizeof(struct fsync_node_entry));
3451 	if (!fsync_node_entry_slab)
3452 		goto destroy_nat_entry_set;
3453 	return 0;
3454 
3455 destroy_nat_entry_set:
3456 	kmem_cache_destroy(nat_entry_set_slab);
3457 destroy_free_nid:
3458 	kmem_cache_destroy(free_nid_slab);
3459 destroy_nat_entry:
3460 	kmem_cache_destroy(nat_entry_slab);
3461 fail:
3462 	return -ENOMEM;
3463 }
3464 
3465 void f2fs_destroy_node_manager_caches(void)
3466 {
3467 	kmem_cache_destroy(fsync_node_entry_slab);
3468 	kmem_cache_destroy(nat_entry_set_slab);
3469 	kmem_cache_destroy(free_nid_slab);
3470 	kmem_cache_destroy(nat_entry_slab);
3471 }
3472