xref: /openbmc/linux/fs/f2fs/node.c (revision 2984f26a)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * fs/f2fs/node.c
4  *
5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6  *             http://www.samsung.com/
7  */
8 #include <linux/fs.h>
9 #include <linux/f2fs_fs.h>
10 #include <linux/mpage.h>
11 #include <linux/sched/mm.h>
12 #include <linux/blkdev.h>
13 #include <linux/pagevec.h>
14 #include <linux/swap.h>
15 
16 #include "f2fs.h"
17 #include "node.h"
18 #include "segment.h"
19 #include "xattr.h"
20 #include "iostat.h"
21 #include <trace/events/f2fs.h>
22 
23 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
24 
25 static struct kmem_cache *nat_entry_slab;
26 static struct kmem_cache *free_nid_slab;
27 static struct kmem_cache *nat_entry_set_slab;
28 static struct kmem_cache *fsync_node_entry_slab;
29 
30 /*
31  * Check whether the given nid is within node id range.
32  */
33 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
34 {
35 	if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
36 		set_sbi_flag(sbi, SBI_NEED_FSCK);
37 		f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
38 			  __func__, nid);
39 		f2fs_handle_error(sbi, ERROR_CORRUPTED_INODE);
40 		return -EFSCORRUPTED;
41 	}
42 	return 0;
43 }
44 
45 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
46 {
47 	struct f2fs_nm_info *nm_i = NM_I(sbi);
48 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
49 	struct sysinfo val;
50 	unsigned long avail_ram;
51 	unsigned long mem_size = 0;
52 	bool res = false;
53 
54 	if (!nm_i)
55 		return true;
56 
57 	si_meminfo(&val);
58 
59 	/* only uses low memory */
60 	avail_ram = val.totalram - val.totalhigh;
61 
62 	/*
63 	 * give 25%, 25%, 50%, 50%, 25%, 25% memory for each components respectively
64 	 */
65 	if (type == FREE_NIDS) {
66 		mem_size = (nm_i->nid_cnt[FREE_NID] *
67 				sizeof(struct free_nid)) >> PAGE_SHIFT;
68 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
69 	} else if (type == NAT_ENTRIES) {
70 		mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
71 				sizeof(struct nat_entry)) >> PAGE_SHIFT;
72 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
73 		if (excess_cached_nats(sbi))
74 			res = false;
75 	} else if (type == DIRTY_DENTS) {
76 		if (sbi->sb->s_bdi->wb.dirty_exceeded)
77 			return false;
78 		mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
79 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
80 	} else if (type == INO_ENTRIES) {
81 		int i;
82 
83 		for (i = 0; i < MAX_INO_ENTRY; i++)
84 			mem_size += sbi->im[i].ino_num *
85 						sizeof(struct ino_entry);
86 		mem_size >>= PAGE_SHIFT;
87 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
88 	} else if (type == READ_EXTENT_CACHE || type == AGE_EXTENT_CACHE) {
89 		enum extent_type etype = type == READ_EXTENT_CACHE ?
90 						EX_READ : EX_BLOCK_AGE;
91 		struct extent_tree_info *eti = &sbi->extent_tree[etype];
92 
93 		mem_size = (atomic_read(&eti->total_ext_tree) *
94 				sizeof(struct extent_tree) +
95 				atomic_read(&eti->total_ext_node) *
96 				sizeof(struct extent_node)) >> PAGE_SHIFT;
97 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
98 	} else if (type == DISCARD_CACHE) {
99 		mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
100 				sizeof(struct discard_cmd)) >> PAGE_SHIFT;
101 		res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
102 	} else if (type == COMPRESS_PAGE) {
103 #ifdef CONFIG_F2FS_FS_COMPRESSION
104 		unsigned long free_ram = val.freeram;
105 
106 		/*
107 		 * free memory is lower than watermark or cached page count
108 		 * exceed threshold, deny caching compress page.
109 		 */
110 		res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
111 			(COMPRESS_MAPPING(sbi)->nrpages <
112 			 free_ram * sbi->compress_percent / 100);
113 #else
114 		res = false;
115 #endif
116 	} else {
117 		if (!sbi->sb->s_bdi->wb.dirty_exceeded)
118 			return true;
119 	}
120 	return res;
121 }
122 
123 static void clear_node_page_dirty(struct page *page)
124 {
125 	if (PageDirty(page)) {
126 		f2fs_clear_page_cache_dirty_tag(page);
127 		clear_page_dirty_for_io(page);
128 		dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
129 	}
130 	ClearPageUptodate(page);
131 }
132 
133 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
134 {
135 	return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
136 }
137 
138 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
139 {
140 	struct page *src_page;
141 	struct page *dst_page;
142 	pgoff_t dst_off;
143 	void *src_addr;
144 	void *dst_addr;
145 	struct f2fs_nm_info *nm_i = NM_I(sbi);
146 
147 	dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
148 
149 	/* get current nat block page with lock */
150 	src_page = get_current_nat_page(sbi, nid);
151 	if (IS_ERR(src_page))
152 		return src_page;
153 	dst_page = f2fs_grab_meta_page(sbi, dst_off);
154 	f2fs_bug_on(sbi, PageDirty(src_page));
155 
156 	src_addr = page_address(src_page);
157 	dst_addr = page_address(dst_page);
158 	memcpy(dst_addr, src_addr, PAGE_SIZE);
159 	set_page_dirty(dst_page);
160 	f2fs_put_page(src_page, 1);
161 
162 	set_to_next_nat(nm_i, nid);
163 
164 	return dst_page;
165 }
166 
167 static struct nat_entry *__alloc_nat_entry(struct f2fs_sb_info *sbi,
168 						nid_t nid, bool no_fail)
169 {
170 	struct nat_entry *new;
171 
172 	new = f2fs_kmem_cache_alloc(nat_entry_slab,
173 					GFP_F2FS_ZERO, no_fail, sbi);
174 	if (new) {
175 		nat_set_nid(new, nid);
176 		nat_reset_flag(new);
177 	}
178 	return new;
179 }
180 
181 static void __free_nat_entry(struct nat_entry *e)
182 {
183 	kmem_cache_free(nat_entry_slab, e);
184 }
185 
186 /* must be locked by nat_tree_lock */
187 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
188 	struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
189 {
190 	if (no_fail)
191 		f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
192 	else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
193 		return NULL;
194 
195 	if (raw_ne)
196 		node_info_from_raw_nat(&ne->ni, raw_ne);
197 
198 	spin_lock(&nm_i->nat_list_lock);
199 	list_add_tail(&ne->list, &nm_i->nat_entries);
200 	spin_unlock(&nm_i->nat_list_lock);
201 
202 	nm_i->nat_cnt[TOTAL_NAT]++;
203 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
204 	return ne;
205 }
206 
207 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
208 {
209 	struct nat_entry *ne;
210 
211 	ne = radix_tree_lookup(&nm_i->nat_root, n);
212 
213 	/* for recent accessed nat entry, move it to tail of lru list */
214 	if (ne && !get_nat_flag(ne, IS_DIRTY)) {
215 		spin_lock(&nm_i->nat_list_lock);
216 		if (!list_empty(&ne->list))
217 			list_move_tail(&ne->list, &nm_i->nat_entries);
218 		spin_unlock(&nm_i->nat_list_lock);
219 	}
220 
221 	return ne;
222 }
223 
224 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
225 		nid_t start, unsigned int nr, struct nat_entry **ep)
226 {
227 	return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
228 }
229 
230 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
231 {
232 	radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
233 	nm_i->nat_cnt[TOTAL_NAT]--;
234 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
235 	__free_nat_entry(e);
236 }
237 
238 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
239 							struct nat_entry *ne)
240 {
241 	nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
242 	struct nat_entry_set *head;
243 
244 	head = radix_tree_lookup(&nm_i->nat_set_root, set);
245 	if (!head) {
246 		head = f2fs_kmem_cache_alloc(nat_entry_set_slab,
247 						GFP_NOFS, true, NULL);
248 
249 		INIT_LIST_HEAD(&head->entry_list);
250 		INIT_LIST_HEAD(&head->set_list);
251 		head->set = set;
252 		head->entry_cnt = 0;
253 		f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
254 	}
255 	return head;
256 }
257 
258 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
259 						struct nat_entry *ne)
260 {
261 	struct nat_entry_set *head;
262 	bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
263 
264 	if (!new_ne)
265 		head = __grab_nat_entry_set(nm_i, ne);
266 
267 	/*
268 	 * update entry_cnt in below condition:
269 	 * 1. update NEW_ADDR to valid block address;
270 	 * 2. update old block address to new one;
271 	 */
272 	if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
273 				!get_nat_flag(ne, IS_DIRTY)))
274 		head->entry_cnt++;
275 
276 	set_nat_flag(ne, IS_PREALLOC, new_ne);
277 
278 	if (get_nat_flag(ne, IS_DIRTY))
279 		goto refresh_list;
280 
281 	nm_i->nat_cnt[DIRTY_NAT]++;
282 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
283 	set_nat_flag(ne, IS_DIRTY, true);
284 refresh_list:
285 	spin_lock(&nm_i->nat_list_lock);
286 	if (new_ne)
287 		list_del_init(&ne->list);
288 	else
289 		list_move_tail(&ne->list, &head->entry_list);
290 	spin_unlock(&nm_i->nat_list_lock);
291 }
292 
293 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
294 		struct nat_entry_set *set, struct nat_entry *ne)
295 {
296 	spin_lock(&nm_i->nat_list_lock);
297 	list_move_tail(&ne->list, &nm_i->nat_entries);
298 	spin_unlock(&nm_i->nat_list_lock);
299 
300 	set_nat_flag(ne, IS_DIRTY, false);
301 	set->entry_cnt--;
302 	nm_i->nat_cnt[DIRTY_NAT]--;
303 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
304 }
305 
306 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
307 		nid_t start, unsigned int nr, struct nat_entry_set **ep)
308 {
309 	return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
310 							start, nr);
311 }
312 
313 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
314 {
315 	return NODE_MAPPING(sbi) == page->mapping &&
316 			IS_DNODE(page) && is_cold_node(page);
317 }
318 
319 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
320 {
321 	spin_lock_init(&sbi->fsync_node_lock);
322 	INIT_LIST_HEAD(&sbi->fsync_node_list);
323 	sbi->fsync_seg_id = 0;
324 	sbi->fsync_node_num = 0;
325 }
326 
327 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
328 							struct page *page)
329 {
330 	struct fsync_node_entry *fn;
331 	unsigned long flags;
332 	unsigned int seq_id;
333 
334 	fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab,
335 					GFP_NOFS, true, NULL);
336 
337 	get_page(page);
338 	fn->page = page;
339 	INIT_LIST_HEAD(&fn->list);
340 
341 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
342 	list_add_tail(&fn->list, &sbi->fsync_node_list);
343 	fn->seq_id = sbi->fsync_seg_id++;
344 	seq_id = fn->seq_id;
345 	sbi->fsync_node_num++;
346 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
347 
348 	return seq_id;
349 }
350 
351 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
352 {
353 	struct fsync_node_entry *fn;
354 	unsigned long flags;
355 
356 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
357 	list_for_each_entry(fn, &sbi->fsync_node_list, list) {
358 		if (fn->page == page) {
359 			list_del(&fn->list);
360 			sbi->fsync_node_num--;
361 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
362 			kmem_cache_free(fsync_node_entry_slab, fn);
363 			put_page(page);
364 			return;
365 		}
366 	}
367 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
368 	f2fs_bug_on(sbi, 1);
369 }
370 
371 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
372 {
373 	unsigned long flags;
374 
375 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
376 	sbi->fsync_seg_id = 0;
377 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
378 }
379 
380 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
381 {
382 	struct f2fs_nm_info *nm_i = NM_I(sbi);
383 	struct nat_entry *e;
384 	bool need = false;
385 
386 	f2fs_down_read(&nm_i->nat_tree_lock);
387 	e = __lookup_nat_cache(nm_i, nid);
388 	if (e) {
389 		if (!get_nat_flag(e, IS_CHECKPOINTED) &&
390 				!get_nat_flag(e, HAS_FSYNCED_INODE))
391 			need = true;
392 	}
393 	f2fs_up_read(&nm_i->nat_tree_lock);
394 	return need;
395 }
396 
397 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
398 {
399 	struct f2fs_nm_info *nm_i = NM_I(sbi);
400 	struct nat_entry *e;
401 	bool is_cp = true;
402 
403 	f2fs_down_read(&nm_i->nat_tree_lock);
404 	e = __lookup_nat_cache(nm_i, nid);
405 	if (e && !get_nat_flag(e, IS_CHECKPOINTED))
406 		is_cp = false;
407 	f2fs_up_read(&nm_i->nat_tree_lock);
408 	return is_cp;
409 }
410 
411 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
412 {
413 	struct f2fs_nm_info *nm_i = NM_I(sbi);
414 	struct nat_entry *e;
415 	bool need_update = true;
416 
417 	f2fs_down_read(&nm_i->nat_tree_lock);
418 	e = __lookup_nat_cache(nm_i, ino);
419 	if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
420 			(get_nat_flag(e, IS_CHECKPOINTED) ||
421 			 get_nat_flag(e, HAS_FSYNCED_INODE)))
422 		need_update = false;
423 	f2fs_up_read(&nm_i->nat_tree_lock);
424 	return need_update;
425 }
426 
427 /* must be locked by nat_tree_lock */
428 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
429 						struct f2fs_nat_entry *ne)
430 {
431 	struct f2fs_nm_info *nm_i = NM_I(sbi);
432 	struct nat_entry *new, *e;
433 
434 	/* Let's mitigate lock contention of nat_tree_lock during checkpoint */
435 	if (f2fs_rwsem_is_locked(&sbi->cp_global_sem))
436 		return;
437 
438 	new = __alloc_nat_entry(sbi, nid, false);
439 	if (!new)
440 		return;
441 
442 	f2fs_down_write(&nm_i->nat_tree_lock);
443 	e = __lookup_nat_cache(nm_i, nid);
444 	if (!e)
445 		e = __init_nat_entry(nm_i, new, ne, false);
446 	else
447 		f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
448 				nat_get_blkaddr(e) !=
449 					le32_to_cpu(ne->block_addr) ||
450 				nat_get_version(e) != ne->version);
451 	f2fs_up_write(&nm_i->nat_tree_lock);
452 	if (e != new)
453 		__free_nat_entry(new);
454 }
455 
456 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
457 			block_t new_blkaddr, bool fsync_done)
458 {
459 	struct f2fs_nm_info *nm_i = NM_I(sbi);
460 	struct nat_entry *e;
461 	struct nat_entry *new = __alloc_nat_entry(sbi, ni->nid, true);
462 
463 	f2fs_down_write(&nm_i->nat_tree_lock);
464 	e = __lookup_nat_cache(nm_i, ni->nid);
465 	if (!e) {
466 		e = __init_nat_entry(nm_i, new, NULL, true);
467 		copy_node_info(&e->ni, ni);
468 		f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
469 	} else if (new_blkaddr == NEW_ADDR) {
470 		/*
471 		 * when nid is reallocated,
472 		 * previous nat entry can be remained in nat cache.
473 		 * So, reinitialize it with new information.
474 		 */
475 		copy_node_info(&e->ni, ni);
476 		f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
477 	}
478 	/* let's free early to reduce memory consumption */
479 	if (e != new)
480 		__free_nat_entry(new);
481 
482 	/* sanity check */
483 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
484 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
485 			new_blkaddr == NULL_ADDR);
486 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
487 			new_blkaddr == NEW_ADDR);
488 	f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
489 			new_blkaddr == NEW_ADDR);
490 
491 	/* increment version no as node is removed */
492 	if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
493 		unsigned char version = nat_get_version(e);
494 
495 		nat_set_version(e, inc_node_version(version));
496 	}
497 
498 	/* change address */
499 	nat_set_blkaddr(e, new_blkaddr);
500 	if (!__is_valid_data_blkaddr(new_blkaddr))
501 		set_nat_flag(e, IS_CHECKPOINTED, false);
502 	__set_nat_cache_dirty(nm_i, e);
503 
504 	/* update fsync_mark if its inode nat entry is still alive */
505 	if (ni->nid != ni->ino)
506 		e = __lookup_nat_cache(nm_i, ni->ino);
507 	if (e) {
508 		if (fsync_done && ni->nid == ni->ino)
509 			set_nat_flag(e, HAS_FSYNCED_INODE, true);
510 		set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
511 	}
512 	f2fs_up_write(&nm_i->nat_tree_lock);
513 }
514 
515 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
516 {
517 	struct f2fs_nm_info *nm_i = NM_I(sbi);
518 	int nr = nr_shrink;
519 
520 	if (!f2fs_down_write_trylock(&nm_i->nat_tree_lock))
521 		return 0;
522 
523 	spin_lock(&nm_i->nat_list_lock);
524 	while (nr_shrink) {
525 		struct nat_entry *ne;
526 
527 		if (list_empty(&nm_i->nat_entries))
528 			break;
529 
530 		ne = list_first_entry(&nm_i->nat_entries,
531 					struct nat_entry, list);
532 		list_del(&ne->list);
533 		spin_unlock(&nm_i->nat_list_lock);
534 
535 		__del_from_nat_cache(nm_i, ne);
536 		nr_shrink--;
537 
538 		spin_lock(&nm_i->nat_list_lock);
539 	}
540 	spin_unlock(&nm_i->nat_list_lock);
541 
542 	f2fs_up_write(&nm_i->nat_tree_lock);
543 	return nr - nr_shrink;
544 }
545 
546 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
547 				struct node_info *ni, bool checkpoint_context)
548 {
549 	struct f2fs_nm_info *nm_i = NM_I(sbi);
550 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
551 	struct f2fs_journal *journal = curseg->journal;
552 	nid_t start_nid = START_NID(nid);
553 	struct f2fs_nat_block *nat_blk;
554 	struct page *page = NULL;
555 	struct f2fs_nat_entry ne;
556 	struct nat_entry *e;
557 	pgoff_t index;
558 	block_t blkaddr;
559 	int i;
560 
561 	ni->nid = nid;
562 retry:
563 	/* Check nat cache */
564 	f2fs_down_read(&nm_i->nat_tree_lock);
565 	e = __lookup_nat_cache(nm_i, nid);
566 	if (e) {
567 		ni->ino = nat_get_ino(e);
568 		ni->blk_addr = nat_get_blkaddr(e);
569 		ni->version = nat_get_version(e);
570 		f2fs_up_read(&nm_i->nat_tree_lock);
571 		return 0;
572 	}
573 
574 	/*
575 	 * Check current segment summary by trying to grab journal_rwsem first.
576 	 * This sem is on the critical path on the checkpoint requiring the above
577 	 * nat_tree_lock. Therefore, we should retry, if we failed to grab here
578 	 * while not bothering checkpoint.
579 	 */
580 	if (!f2fs_rwsem_is_locked(&sbi->cp_global_sem) || checkpoint_context) {
581 		down_read(&curseg->journal_rwsem);
582 	} else if (f2fs_rwsem_is_contended(&nm_i->nat_tree_lock) ||
583 				!down_read_trylock(&curseg->journal_rwsem)) {
584 		f2fs_up_read(&nm_i->nat_tree_lock);
585 		goto retry;
586 	}
587 
588 	i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
589 	if (i >= 0) {
590 		ne = nat_in_journal(journal, i);
591 		node_info_from_raw_nat(ni, &ne);
592 	}
593 	up_read(&curseg->journal_rwsem);
594 	if (i >= 0) {
595 		f2fs_up_read(&nm_i->nat_tree_lock);
596 		goto cache;
597 	}
598 
599 	/* Fill node_info from nat page */
600 	index = current_nat_addr(sbi, nid);
601 	f2fs_up_read(&nm_i->nat_tree_lock);
602 
603 	page = f2fs_get_meta_page(sbi, index);
604 	if (IS_ERR(page))
605 		return PTR_ERR(page);
606 
607 	nat_blk = (struct f2fs_nat_block *)page_address(page);
608 	ne = nat_blk->entries[nid - start_nid];
609 	node_info_from_raw_nat(ni, &ne);
610 	f2fs_put_page(page, 1);
611 cache:
612 	blkaddr = le32_to_cpu(ne.block_addr);
613 	if (__is_valid_data_blkaddr(blkaddr) &&
614 		!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
615 		return -EFAULT;
616 
617 	/* cache nat entry */
618 	cache_nat_entry(sbi, nid, &ne);
619 	return 0;
620 }
621 
622 /*
623  * readahead MAX_RA_NODE number of node pages.
624  */
625 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
626 {
627 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
628 	struct blk_plug plug;
629 	int i, end;
630 	nid_t nid;
631 
632 	blk_start_plug(&plug);
633 
634 	/* Then, try readahead for siblings of the desired node */
635 	end = start + n;
636 	end = min(end, NIDS_PER_BLOCK);
637 	for (i = start; i < end; i++) {
638 		nid = get_nid(parent, i, false);
639 		f2fs_ra_node_page(sbi, nid);
640 	}
641 
642 	blk_finish_plug(&plug);
643 }
644 
645 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
646 {
647 	const long direct_index = ADDRS_PER_INODE(dn->inode);
648 	const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
649 	const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
650 	unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
651 	int cur_level = dn->cur_level;
652 	int max_level = dn->max_level;
653 	pgoff_t base = 0;
654 
655 	if (!dn->max_level)
656 		return pgofs + 1;
657 
658 	while (max_level-- > cur_level)
659 		skipped_unit *= NIDS_PER_BLOCK;
660 
661 	switch (dn->max_level) {
662 	case 3:
663 		base += 2 * indirect_blks;
664 		fallthrough;
665 	case 2:
666 		base += 2 * direct_blks;
667 		fallthrough;
668 	case 1:
669 		base += direct_index;
670 		break;
671 	default:
672 		f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
673 	}
674 
675 	return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
676 }
677 
678 /*
679  * The maximum depth is four.
680  * Offset[0] will have raw inode offset.
681  */
682 static int get_node_path(struct inode *inode, long block,
683 				int offset[4], unsigned int noffset[4])
684 {
685 	const long direct_index = ADDRS_PER_INODE(inode);
686 	const long direct_blks = ADDRS_PER_BLOCK(inode);
687 	const long dptrs_per_blk = NIDS_PER_BLOCK;
688 	const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
689 	const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
690 	int n = 0;
691 	int level = 0;
692 
693 	noffset[0] = 0;
694 
695 	if (block < direct_index) {
696 		offset[n] = block;
697 		goto got;
698 	}
699 	block -= direct_index;
700 	if (block < direct_blks) {
701 		offset[n++] = NODE_DIR1_BLOCK;
702 		noffset[n] = 1;
703 		offset[n] = block;
704 		level = 1;
705 		goto got;
706 	}
707 	block -= direct_blks;
708 	if (block < direct_blks) {
709 		offset[n++] = NODE_DIR2_BLOCK;
710 		noffset[n] = 2;
711 		offset[n] = block;
712 		level = 1;
713 		goto got;
714 	}
715 	block -= direct_blks;
716 	if (block < indirect_blks) {
717 		offset[n++] = NODE_IND1_BLOCK;
718 		noffset[n] = 3;
719 		offset[n++] = block / direct_blks;
720 		noffset[n] = 4 + offset[n - 1];
721 		offset[n] = block % direct_blks;
722 		level = 2;
723 		goto got;
724 	}
725 	block -= indirect_blks;
726 	if (block < indirect_blks) {
727 		offset[n++] = NODE_IND2_BLOCK;
728 		noffset[n] = 4 + dptrs_per_blk;
729 		offset[n++] = block / direct_blks;
730 		noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
731 		offset[n] = block % direct_blks;
732 		level = 2;
733 		goto got;
734 	}
735 	block -= indirect_blks;
736 	if (block < dindirect_blks) {
737 		offset[n++] = NODE_DIND_BLOCK;
738 		noffset[n] = 5 + (dptrs_per_blk * 2);
739 		offset[n++] = block / indirect_blks;
740 		noffset[n] = 6 + (dptrs_per_blk * 2) +
741 			      offset[n - 1] * (dptrs_per_blk + 1);
742 		offset[n++] = (block / direct_blks) % dptrs_per_blk;
743 		noffset[n] = 7 + (dptrs_per_blk * 2) +
744 			      offset[n - 2] * (dptrs_per_blk + 1) +
745 			      offset[n - 1];
746 		offset[n] = block % direct_blks;
747 		level = 3;
748 		goto got;
749 	} else {
750 		return -E2BIG;
751 	}
752 got:
753 	return level;
754 }
755 
756 /*
757  * Caller should call f2fs_put_dnode(dn).
758  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
759  * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
760  */
761 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
762 {
763 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
764 	struct page *npage[4];
765 	struct page *parent = NULL;
766 	int offset[4];
767 	unsigned int noffset[4];
768 	nid_t nids[4];
769 	int level, i = 0;
770 	int err = 0;
771 
772 	level = get_node_path(dn->inode, index, offset, noffset);
773 	if (level < 0)
774 		return level;
775 
776 	nids[0] = dn->inode->i_ino;
777 	npage[0] = dn->inode_page;
778 
779 	if (!npage[0]) {
780 		npage[0] = f2fs_get_node_page(sbi, nids[0]);
781 		if (IS_ERR(npage[0]))
782 			return PTR_ERR(npage[0]);
783 	}
784 
785 	/* if inline_data is set, should not report any block indices */
786 	if (f2fs_has_inline_data(dn->inode) && index) {
787 		err = -ENOENT;
788 		f2fs_put_page(npage[0], 1);
789 		goto release_out;
790 	}
791 
792 	parent = npage[0];
793 	if (level != 0)
794 		nids[1] = get_nid(parent, offset[0], true);
795 	dn->inode_page = npage[0];
796 	dn->inode_page_locked = true;
797 
798 	/* get indirect or direct nodes */
799 	for (i = 1; i <= level; i++) {
800 		bool done = false;
801 
802 		if (!nids[i] && mode == ALLOC_NODE) {
803 			/* alloc new node */
804 			if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
805 				err = -ENOSPC;
806 				goto release_pages;
807 			}
808 
809 			dn->nid = nids[i];
810 			npage[i] = f2fs_new_node_page(dn, noffset[i]);
811 			if (IS_ERR(npage[i])) {
812 				f2fs_alloc_nid_failed(sbi, nids[i]);
813 				err = PTR_ERR(npage[i]);
814 				goto release_pages;
815 			}
816 
817 			set_nid(parent, offset[i - 1], nids[i], i == 1);
818 			f2fs_alloc_nid_done(sbi, nids[i]);
819 			done = true;
820 		} else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
821 			npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
822 			if (IS_ERR(npage[i])) {
823 				err = PTR_ERR(npage[i]);
824 				goto release_pages;
825 			}
826 			done = true;
827 		}
828 		if (i == 1) {
829 			dn->inode_page_locked = false;
830 			unlock_page(parent);
831 		} else {
832 			f2fs_put_page(parent, 1);
833 		}
834 
835 		if (!done) {
836 			npage[i] = f2fs_get_node_page(sbi, nids[i]);
837 			if (IS_ERR(npage[i])) {
838 				err = PTR_ERR(npage[i]);
839 				f2fs_put_page(npage[0], 0);
840 				goto release_out;
841 			}
842 		}
843 		if (i < level) {
844 			parent = npage[i];
845 			nids[i + 1] = get_nid(parent, offset[i], false);
846 		}
847 	}
848 	dn->nid = nids[level];
849 	dn->ofs_in_node = offset[level];
850 	dn->node_page = npage[level];
851 	dn->data_blkaddr = f2fs_data_blkaddr(dn);
852 
853 	if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) &&
854 					f2fs_sb_has_readonly(sbi)) {
855 		unsigned int c_len = f2fs_cluster_blocks_are_contiguous(dn);
856 		block_t blkaddr;
857 
858 		if (!c_len)
859 			goto out;
860 
861 		blkaddr = f2fs_data_blkaddr(dn);
862 		if (blkaddr == COMPRESS_ADDR)
863 			blkaddr = data_blkaddr(dn->inode, dn->node_page,
864 						dn->ofs_in_node + 1);
865 
866 		f2fs_update_read_extent_tree_range_compressed(dn->inode,
867 					index, blkaddr,
868 					F2FS_I(dn->inode)->i_cluster_size,
869 					c_len);
870 	}
871 out:
872 	return 0;
873 
874 release_pages:
875 	f2fs_put_page(parent, 1);
876 	if (i > 1)
877 		f2fs_put_page(npage[0], 0);
878 release_out:
879 	dn->inode_page = NULL;
880 	dn->node_page = NULL;
881 	if (err == -ENOENT) {
882 		dn->cur_level = i;
883 		dn->max_level = level;
884 		dn->ofs_in_node = offset[level];
885 	}
886 	return err;
887 }
888 
889 static int truncate_node(struct dnode_of_data *dn)
890 {
891 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
892 	struct node_info ni;
893 	int err;
894 	pgoff_t index;
895 
896 	err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
897 	if (err)
898 		return err;
899 
900 	/* Deallocate node address */
901 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
902 	dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
903 	set_node_addr(sbi, &ni, NULL_ADDR, false);
904 
905 	if (dn->nid == dn->inode->i_ino) {
906 		f2fs_remove_orphan_inode(sbi, dn->nid);
907 		dec_valid_inode_count(sbi);
908 		f2fs_inode_synced(dn->inode);
909 	}
910 
911 	clear_node_page_dirty(dn->node_page);
912 	set_sbi_flag(sbi, SBI_IS_DIRTY);
913 
914 	index = dn->node_page->index;
915 	f2fs_put_page(dn->node_page, 1);
916 
917 	invalidate_mapping_pages(NODE_MAPPING(sbi),
918 			index, index);
919 
920 	dn->node_page = NULL;
921 	trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
922 
923 	return 0;
924 }
925 
926 static int truncate_dnode(struct dnode_of_data *dn)
927 {
928 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
929 	struct page *page;
930 	int err;
931 
932 	if (dn->nid == 0)
933 		return 1;
934 
935 	/* get direct node */
936 	page = f2fs_get_node_page(sbi, dn->nid);
937 	if (PTR_ERR(page) == -ENOENT)
938 		return 1;
939 	else if (IS_ERR(page))
940 		return PTR_ERR(page);
941 
942 	if (IS_INODE(page) || ino_of_node(page) != dn->inode->i_ino) {
943 		f2fs_err(sbi, "incorrect node reference, ino: %lu, nid: %u, ino_of_node: %u",
944 				dn->inode->i_ino, dn->nid, ino_of_node(page));
945 		set_sbi_flag(sbi, SBI_NEED_FSCK);
946 		f2fs_handle_error(sbi, ERROR_INVALID_NODE_REFERENCE);
947 		f2fs_put_page(page, 1);
948 		return -EFSCORRUPTED;
949 	}
950 
951 	/* Make dnode_of_data for parameter */
952 	dn->node_page = page;
953 	dn->ofs_in_node = 0;
954 	f2fs_truncate_data_blocks_range(dn, ADDRS_PER_BLOCK(dn->inode));
955 	err = truncate_node(dn);
956 	if (err) {
957 		f2fs_put_page(page, 1);
958 		return err;
959 	}
960 
961 	return 1;
962 }
963 
964 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
965 						int ofs, int depth)
966 {
967 	struct dnode_of_data rdn = *dn;
968 	struct page *page;
969 	struct f2fs_node *rn;
970 	nid_t child_nid;
971 	unsigned int child_nofs;
972 	int freed = 0;
973 	int i, ret;
974 
975 	if (dn->nid == 0)
976 		return NIDS_PER_BLOCK + 1;
977 
978 	trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
979 
980 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
981 	if (IS_ERR(page)) {
982 		trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
983 		return PTR_ERR(page);
984 	}
985 
986 	f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
987 
988 	rn = F2FS_NODE(page);
989 	if (depth < 3) {
990 		for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
991 			child_nid = le32_to_cpu(rn->in.nid[i]);
992 			if (child_nid == 0)
993 				continue;
994 			rdn.nid = child_nid;
995 			ret = truncate_dnode(&rdn);
996 			if (ret < 0)
997 				goto out_err;
998 			if (set_nid(page, i, 0, false))
999 				dn->node_changed = true;
1000 		}
1001 	} else {
1002 		child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
1003 		for (i = ofs; i < NIDS_PER_BLOCK; i++) {
1004 			child_nid = le32_to_cpu(rn->in.nid[i]);
1005 			if (child_nid == 0) {
1006 				child_nofs += NIDS_PER_BLOCK + 1;
1007 				continue;
1008 			}
1009 			rdn.nid = child_nid;
1010 			ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
1011 			if (ret == (NIDS_PER_BLOCK + 1)) {
1012 				if (set_nid(page, i, 0, false))
1013 					dn->node_changed = true;
1014 				child_nofs += ret;
1015 			} else if (ret < 0 && ret != -ENOENT) {
1016 				goto out_err;
1017 			}
1018 		}
1019 		freed = child_nofs;
1020 	}
1021 
1022 	if (!ofs) {
1023 		/* remove current indirect node */
1024 		dn->node_page = page;
1025 		ret = truncate_node(dn);
1026 		if (ret)
1027 			goto out_err;
1028 		freed++;
1029 	} else {
1030 		f2fs_put_page(page, 1);
1031 	}
1032 	trace_f2fs_truncate_nodes_exit(dn->inode, freed);
1033 	return freed;
1034 
1035 out_err:
1036 	f2fs_put_page(page, 1);
1037 	trace_f2fs_truncate_nodes_exit(dn->inode, ret);
1038 	return ret;
1039 }
1040 
1041 static int truncate_partial_nodes(struct dnode_of_data *dn,
1042 			struct f2fs_inode *ri, int *offset, int depth)
1043 {
1044 	struct page *pages[2];
1045 	nid_t nid[3];
1046 	nid_t child_nid;
1047 	int err = 0;
1048 	int i;
1049 	int idx = depth - 2;
1050 
1051 	nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1052 	if (!nid[0])
1053 		return 0;
1054 
1055 	/* get indirect nodes in the path */
1056 	for (i = 0; i < idx + 1; i++) {
1057 		/* reference count'll be increased */
1058 		pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
1059 		if (IS_ERR(pages[i])) {
1060 			err = PTR_ERR(pages[i]);
1061 			idx = i - 1;
1062 			goto fail;
1063 		}
1064 		nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1065 	}
1066 
1067 	f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1068 
1069 	/* free direct nodes linked to a partial indirect node */
1070 	for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1071 		child_nid = get_nid(pages[idx], i, false);
1072 		if (!child_nid)
1073 			continue;
1074 		dn->nid = child_nid;
1075 		err = truncate_dnode(dn);
1076 		if (err < 0)
1077 			goto fail;
1078 		if (set_nid(pages[idx], i, 0, false))
1079 			dn->node_changed = true;
1080 	}
1081 
1082 	if (offset[idx + 1] == 0) {
1083 		dn->node_page = pages[idx];
1084 		dn->nid = nid[idx];
1085 		err = truncate_node(dn);
1086 		if (err)
1087 			goto fail;
1088 	} else {
1089 		f2fs_put_page(pages[idx], 1);
1090 	}
1091 	offset[idx]++;
1092 	offset[idx + 1] = 0;
1093 	idx--;
1094 fail:
1095 	for (i = idx; i >= 0; i--)
1096 		f2fs_put_page(pages[i], 1);
1097 
1098 	trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1099 
1100 	return err;
1101 }
1102 
1103 /*
1104  * All the block addresses of data and nodes should be nullified.
1105  */
1106 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1107 {
1108 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1109 	int err = 0, cont = 1;
1110 	int level, offset[4], noffset[4];
1111 	unsigned int nofs = 0;
1112 	struct f2fs_inode *ri;
1113 	struct dnode_of_data dn;
1114 	struct page *page;
1115 
1116 	trace_f2fs_truncate_inode_blocks_enter(inode, from);
1117 
1118 	level = get_node_path(inode, from, offset, noffset);
1119 	if (level < 0) {
1120 		trace_f2fs_truncate_inode_blocks_exit(inode, level);
1121 		return level;
1122 	}
1123 
1124 	page = f2fs_get_node_page(sbi, inode->i_ino);
1125 	if (IS_ERR(page)) {
1126 		trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1127 		return PTR_ERR(page);
1128 	}
1129 
1130 	set_new_dnode(&dn, inode, page, NULL, 0);
1131 	unlock_page(page);
1132 
1133 	ri = F2FS_INODE(page);
1134 	switch (level) {
1135 	case 0:
1136 	case 1:
1137 		nofs = noffset[1];
1138 		break;
1139 	case 2:
1140 		nofs = noffset[1];
1141 		if (!offset[level - 1])
1142 			goto skip_partial;
1143 		err = truncate_partial_nodes(&dn, ri, offset, level);
1144 		if (err < 0 && err != -ENOENT)
1145 			goto fail;
1146 		nofs += 1 + NIDS_PER_BLOCK;
1147 		break;
1148 	case 3:
1149 		nofs = 5 + 2 * NIDS_PER_BLOCK;
1150 		if (!offset[level - 1])
1151 			goto skip_partial;
1152 		err = truncate_partial_nodes(&dn, ri, offset, level);
1153 		if (err < 0 && err != -ENOENT)
1154 			goto fail;
1155 		break;
1156 	default:
1157 		BUG();
1158 	}
1159 
1160 skip_partial:
1161 	while (cont) {
1162 		dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1163 		switch (offset[0]) {
1164 		case NODE_DIR1_BLOCK:
1165 		case NODE_DIR2_BLOCK:
1166 			err = truncate_dnode(&dn);
1167 			break;
1168 
1169 		case NODE_IND1_BLOCK:
1170 		case NODE_IND2_BLOCK:
1171 			err = truncate_nodes(&dn, nofs, offset[1], 2);
1172 			break;
1173 
1174 		case NODE_DIND_BLOCK:
1175 			err = truncate_nodes(&dn, nofs, offset[1], 3);
1176 			cont = 0;
1177 			break;
1178 
1179 		default:
1180 			BUG();
1181 		}
1182 		if (err < 0 && err != -ENOENT)
1183 			goto fail;
1184 		if (offset[1] == 0 &&
1185 				ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1186 			lock_page(page);
1187 			BUG_ON(page->mapping != NODE_MAPPING(sbi));
1188 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1189 			ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1190 			set_page_dirty(page);
1191 			unlock_page(page);
1192 		}
1193 		offset[1] = 0;
1194 		offset[0]++;
1195 		nofs += err;
1196 	}
1197 fail:
1198 	f2fs_put_page(page, 0);
1199 	trace_f2fs_truncate_inode_blocks_exit(inode, err);
1200 	return err > 0 ? 0 : err;
1201 }
1202 
1203 /* caller must lock inode page */
1204 int f2fs_truncate_xattr_node(struct inode *inode)
1205 {
1206 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1207 	nid_t nid = F2FS_I(inode)->i_xattr_nid;
1208 	struct dnode_of_data dn;
1209 	struct page *npage;
1210 	int err;
1211 
1212 	if (!nid)
1213 		return 0;
1214 
1215 	npage = f2fs_get_node_page(sbi, nid);
1216 	if (IS_ERR(npage))
1217 		return PTR_ERR(npage);
1218 
1219 	set_new_dnode(&dn, inode, NULL, npage, nid);
1220 	err = truncate_node(&dn);
1221 	if (err) {
1222 		f2fs_put_page(npage, 1);
1223 		return err;
1224 	}
1225 
1226 	f2fs_i_xnid_write(inode, 0);
1227 
1228 	return 0;
1229 }
1230 
1231 /*
1232  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1233  * f2fs_unlock_op().
1234  */
1235 int f2fs_remove_inode_page(struct inode *inode)
1236 {
1237 	struct dnode_of_data dn;
1238 	int err;
1239 
1240 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1241 	err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1242 	if (err)
1243 		return err;
1244 
1245 	err = f2fs_truncate_xattr_node(inode);
1246 	if (err) {
1247 		f2fs_put_dnode(&dn);
1248 		return err;
1249 	}
1250 
1251 	/* remove potential inline_data blocks */
1252 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1253 				S_ISLNK(inode->i_mode))
1254 		f2fs_truncate_data_blocks_range(&dn, 1);
1255 
1256 	/* 0 is possible, after f2fs_new_inode() has failed */
1257 	if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1258 		f2fs_put_dnode(&dn);
1259 		return -EIO;
1260 	}
1261 
1262 	if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1263 		f2fs_warn(F2FS_I_SB(inode),
1264 			"f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1265 			inode->i_ino, (unsigned long long)inode->i_blocks);
1266 		set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1267 	}
1268 
1269 	/* will put inode & node pages */
1270 	err = truncate_node(&dn);
1271 	if (err) {
1272 		f2fs_put_dnode(&dn);
1273 		return err;
1274 	}
1275 	return 0;
1276 }
1277 
1278 struct page *f2fs_new_inode_page(struct inode *inode)
1279 {
1280 	struct dnode_of_data dn;
1281 
1282 	/* allocate inode page for new inode */
1283 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1284 
1285 	/* caller should f2fs_put_page(page, 1); */
1286 	return f2fs_new_node_page(&dn, 0);
1287 }
1288 
1289 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1290 {
1291 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1292 	struct node_info new_ni;
1293 	struct page *page;
1294 	int err;
1295 
1296 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1297 		return ERR_PTR(-EPERM);
1298 
1299 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1300 	if (!page)
1301 		return ERR_PTR(-ENOMEM);
1302 
1303 	if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1304 		goto fail;
1305 
1306 #ifdef CONFIG_F2FS_CHECK_FS
1307 	err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false);
1308 	if (err) {
1309 		dec_valid_node_count(sbi, dn->inode, !ofs);
1310 		goto fail;
1311 	}
1312 	if (unlikely(new_ni.blk_addr != NULL_ADDR)) {
1313 		err = -EFSCORRUPTED;
1314 		set_sbi_flag(sbi, SBI_NEED_FSCK);
1315 		f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1316 		goto fail;
1317 	}
1318 #endif
1319 	new_ni.nid = dn->nid;
1320 	new_ni.ino = dn->inode->i_ino;
1321 	new_ni.blk_addr = NULL_ADDR;
1322 	new_ni.flag = 0;
1323 	new_ni.version = 0;
1324 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1325 
1326 	f2fs_wait_on_page_writeback(page, NODE, true, true);
1327 	fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1328 	set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1329 	if (!PageUptodate(page))
1330 		SetPageUptodate(page);
1331 	if (set_page_dirty(page))
1332 		dn->node_changed = true;
1333 
1334 	if (f2fs_has_xattr_block(ofs))
1335 		f2fs_i_xnid_write(dn->inode, dn->nid);
1336 
1337 	if (ofs == 0)
1338 		inc_valid_inode_count(sbi);
1339 	return page;
1340 
1341 fail:
1342 	clear_node_page_dirty(page);
1343 	f2fs_put_page(page, 1);
1344 	return ERR_PTR(err);
1345 }
1346 
1347 /*
1348  * Caller should do after getting the following values.
1349  * 0: f2fs_put_page(page, 0)
1350  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1351  */
1352 static int read_node_page(struct page *page, blk_opf_t op_flags)
1353 {
1354 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1355 	struct node_info ni;
1356 	struct f2fs_io_info fio = {
1357 		.sbi = sbi,
1358 		.type = NODE,
1359 		.op = REQ_OP_READ,
1360 		.op_flags = op_flags,
1361 		.page = page,
1362 		.encrypted_page = NULL,
1363 	};
1364 	int err;
1365 
1366 	if (PageUptodate(page)) {
1367 		if (!f2fs_inode_chksum_verify(sbi, page)) {
1368 			ClearPageUptodate(page);
1369 			return -EFSBADCRC;
1370 		}
1371 		return LOCKED_PAGE;
1372 	}
1373 
1374 	err = f2fs_get_node_info(sbi, page->index, &ni, false);
1375 	if (err)
1376 		return err;
1377 
1378 	/* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1379 	if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR)) {
1380 		ClearPageUptodate(page);
1381 		return -ENOENT;
1382 	}
1383 
1384 	fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1385 
1386 	err = f2fs_submit_page_bio(&fio);
1387 
1388 	if (!err)
1389 		f2fs_update_iostat(sbi, NULL, FS_NODE_READ_IO, F2FS_BLKSIZE);
1390 
1391 	return err;
1392 }
1393 
1394 /*
1395  * Readahead a node page
1396  */
1397 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1398 {
1399 	struct page *apage;
1400 	int err;
1401 
1402 	if (!nid)
1403 		return;
1404 	if (f2fs_check_nid_range(sbi, nid))
1405 		return;
1406 
1407 	apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1408 	if (apage)
1409 		return;
1410 
1411 	apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1412 	if (!apage)
1413 		return;
1414 
1415 	err = read_node_page(apage, REQ_RAHEAD);
1416 	f2fs_put_page(apage, err ? 1 : 0);
1417 }
1418 
1419 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1420 					struct page *parent, int start)
1421 {
1422 	struct page *page;
1423 	int err;
1424 
1425 	if (!nid)
1426 		return ERR_PTR(-ENOENT);
1427 	if (f2fs_check_nid_range(sbi, nid))
1428 		return ERR_PTR(-EINVAL);
1429 repeat:
1430 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1431 	if (!page)
1432 		return ERR_PTR(-ENOMEM);
1433 
1434 	err = read_node_page(page, 0);
1435 	if (err < 0) {
1436 		goto out_put_err;
1437 	} else if (err == LOCKED_PAGE) {
1438 		err = 0;
1439 		goto page_hit;
1440 	}
1441 
1442 	if (parent)
1443 		f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1444 
1445 	lock_page(page);
1446 
1447 	if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1448 		f2fs_put_page(page, 1);
1449 		goto repeat;
1450 	}
1451 
1452 	if (unlikely(!PageUptodate(page))) {
1453 		err = -EIO;
1454 		goto out_err;
1455 	}
1456 
1457 	if (!f2fs_inode_chksum_verify(sbi, page)) {
1458 		err = -EFSBADCRC;
1459 		goto out_err;
1460 	}
1461 page_hit:
1462 	if (likely(nid == nid_of_node(page)))
1463 		return page;
1464 
1465 	f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1466 			  nid, nid_of_node(page), ino_of_node(page),
1467 			  ofs_of_node(page), cpver_of_node(page),
1468 			  next_blkaddr_of_node(page));
1469 	set_sbi_flag(sbi, SBI_NEED_FSCK);
1470 	f2fs_handle_error(sbi, ERROR_INCONSISTENT_FOOTER);
1471 	err = -EFSCORRUPTED;
1472 out_err:
1473 	ClearPageUptodate(page);
1474 out_put_err:
1475 	/* ENOENT comes from read_node_page which is not an error. */
1476 	if (err != -ENOENT)
1477 		f2fs_handle_page_eio(sbi, page->index, NODE);
1478 	f2fs_put_page(page, 1);
1479 	return ERR_PTR(err);
1480 }
1481 
1482 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1483 {
1484 	return __get_node_page(sbi, nid, NULL, 0);
1485 }
1486 
1487 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1488 {
1489 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1490 	nid_t nid = get_nid(parent, start, false);
1491 
1492 	return __get_node_page(sbi, nid, parent, start);
1493 }
1494 
1495 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1496 {
1497 	struct inode *inode;
1498 	struct page *page;
1499 	int ret;
1500 
1501 	/* should flush inline_data before evict_inode */
1502 	inode = ilookup(sbi->sb, ino);
1503 	if (!inode)
1504 		return;
1505 
1506 	page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1507 					FGP_LOCK|FGP_NOWAIT, 0);
1508 	if (!page)
1509 		goto iput_out;
1510 
1511 	if (!PageUptodate(page))
1512 		goto page_out;
1513 
1514 	if (!PageDirty(page))
1515 		goto page_out;
1516 
1517 	if (!clear_page_dirty_for_io(page))
1518 		goto page_out;
1519 
1520 	ret = f2fs_write_inline_data(inode, page);
1521 	inode_dec_dirty_pages(inode);
1522 	f2fs_remove_dirty_inode(inode);
1523 	if (ret)
1524 		set_page_dirty(page);
1525 page_out:
1526 	f2fs_put_page(page, 1);
1527 iput_out:
1528 	iput(inode);
1529 }
1530 
1531 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1532 {
1533 	pgoff_t index;
1534 	struct folio_batch fbatch;
1535 	struct page *last_page = NULL;
1536 	int nr_folios;
1537 
1538 	folio_batch_init(&fbatch);
1539 	index = 0;
1540 
1541 	while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1542 					(pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1543 					&fbatch))) {
1544 		int i;
1545 
1546 		for (i = 0; i < nr_folios; i++) {
1547 			struct page *page = &fbatch.folios[i]->page;
1548 
1549 			if (unlikely(f2fs_cp_error(sbi))) {
1550 				f2fs_put_page(last_page, 0);
1551 				folio_batch_release(&fbatch);
1552 				return ERR_PTR(-EIO);
1553 			}
1554 
1555 			if (!IS_DNODE(page) || !is_cold_node(page))
1556 				continue;
1557 			if (ino_of_node(page) != ino)
1558 				continue;
1559 
1560 			lock_page(page);
1561 
1562 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1563 continue_unlock:
1564 				unlock_page(page);
1565 				continue;
1566 			}
1567 			if (ino_of_node(page) != ino)
1568 				goto continue_unlock;
1569 
1570 			if (!PageDirty(page)) {
1571 				/* someone wrote it for us */
1572 				goto continue_unlock;
1573 			}
1574 
1575 			if (last_page)
1576 				f2fs_put_page(last_page, 0);
1577 
1578 			get_page(page);
1579 			last_page = page;
1580 			unlock_page(page);
1581 		}
1582 		folio_batch_release(&fbatch);
1583 		cond_resched();
1584 	}
1585 	return last_page;
1586 }
1587 
1588 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1589 				struct writeback_control *wbc, bool do_balance,
1590 				enum iostat_type io_type, unsigned int *seq_id)
1591 {
1592 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1593 	nid_t nid;
1594 	struct node_info ni;
1595 	struct f2fs_io_info fio = {
1596 		.sbi = sbi,
1597 		.ino = ino_of_node(page),
1598 		.type = NODE,
1599 		.op = REQ_OP_WRITE,
1600 		.op_flags = wbc_to_write_flags(wbc),
1601 		.page = page,
1602 		.encrypted_page = NULL,
1603 		.submitted = 0,
1604 		.io_type = io_type,
1605 		.io_wbc = wbc,
1606 	};
1607 	unsigned int seq;
1608 
1609 	trace_f2fs_writepage(page, NODE);
1610 
1611 	if (unlikely(f2fs_cp_error(sbi))) {
1612 		/* keep node pages in remount-ro mode */
1613 		if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY)
1614 			goto redirty_out;
1615 		ClearPageUptodate(page);
1616 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1617 		unlock_page(page);
1618 		return 0;
1619 	}
1620 
1621 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1622 		goto redirty_out;
1623 
1624 	if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1625 			wbc->sync_mode == WB_SYNC_NONE &&
1626 			IS_DNODE(page) && is_cold_node(page))
1627 		goto redirty_out;
1628 
1629 	/* get old block addr of this node page */
1630 	nid = nid_of_node(page);
1631 	f2fs_bug_on(sbi, page->index != nid);
1632 
1633 	if (f2fs_get_node_info(sbi, nid, &ni, !do_balance))
1634 		goto redirty_out;
1635 
1636 	if (wbc->for_reclaim) {
1637 		if (!f2fs_down_read_trylock(&sbi->node_write))
1638 			goto redirty_out;
1639 	} else {
1640 		f2fs_down_read(&sbi->node_write);
1641 	}
1642 
1643 	/* This page is already truncated */
1644 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1645 		ClearPageUptodate(page);
1646 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1647 		f2fs_up_read(&sbi->node_write);
1648 		unlock_page(page);
1649 		return 0;
1650 	}
1651 
1652 	if (__is_valid_data_blkaddr(ni.blk_addr) &&
1653 		!f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1654 					DATA_GENERIC_ENHANCE)) {
1655 		f2fs_up_read(&sbi->node_write);
1656 		goto redirty_out;
1657 	}
1658 
1659 	if (atomic && !test_opt(sbi, NOBARRIER) && !f2fs_sb_has_blkzoned(sbi))
1660 		fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1661 
1662 	/* should add to global list before clearing PAGECACHE status */
1663 	if (f2fs_in_warm_node_list(sbi, page)) {
1664 		seq = f2fs_add_fsync_node_entry(sbi, page);
1665 		if (seq_id)
1666 			*seq_id = seq;
1667 	}
1668 
1669 	set_page_writeback(page);
1670 
1671 	fio.old_blkaddr = ni.blk_addr;
1672 	f2fs_do_write_node_page(nid, &fio);
1673 	set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1674 	dec_page_count(sbi, F2FS_DIRTY_NODES);
1675 	f2fs_up_read(&sbi->node_write);
1676 
1677 	if (wbc->for_reclaim) {
1678 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1679 		submitted = NULL;
1680 	}
1681 
1682 	unlock_page(page);
1683 
1684 	if (unlikely(f2fs_cp_error(sbi))) {
1685 		f2fs_submit_merged_write(sbi, NODE);
1686 		submitted = NULL;
1687 	}
1688 	if (submitted)
1689 		*submitted = fio.submitted;
1690 
1691 	if (do_balance)
1692 		f2fs_balance_fs(sbi, false);
1693 	return 0;
1694 
1695 redirty_out:
1696 	redirty_page_for_writepage(wbc, page);
1697 	return AOP_WRITEPAGE_ACTIVATE;
1698 }
1699 
1700 int f2fs_move_node_page(struct page *node_page, int gc_type)
1701 {
1702 	int err = 0;
1703 
1704 	if (gc_type == FG_GC) {
1705 		struct writeback_control wbc = {
1706 			.sync_mode = WB_SYNC_ALL,
1707 			.nr_to_write = 1,
1708 			.for_reclaim = 0,
1709 		};
1710 
1711 		f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1712 
1713 		set_page_dirty(node_page);
1714 
1715 		if (!clear_page_dirty_for_io(node_page)) {
1716 			err = -EAGAIN;
1717 			goto out_page;
1718 		}
1719 
1720 		if (__write_node_page(node_page, false, NULL,
1721 					&wbc, false, FS_GC_NODE_IO, NULL)) {
1722 			err = -EAGAIN;
1723 			unlock_page(node_page);
1724 		}
1725 		goto release_page;
1726 	} else {
1727 		/* set page dirty and write it */
1728 		if (!PageWriteback(node_page))
1729 			set_page_dirty(node_page);
1730 	}
1731 out_page:
1732 	unlock_page(node_page);
1733 release_page:
1734 	f2fs_put_page(node_page, 0);
1735 	return err;
1736 }
1737 
1738 static int f2fs_write_node_page(struct page *page,
1739 				struct writeback_control *wbc)
1740 {
1741 	return __write_node_page(page, false, NULL, wbc, false,
1742 						FS_NODE_IO, NULL);
1743 }
1744 
1745 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1746 			struct writeback_control *wbc, bool atomic,
1747 			unsigned int *seq_id)
1748 {
1749 	pgoff_t index;
1750 	struct folio_batch fbatch;
1751 	int ret = 0;
1752 	struct page *last_page = NULL;
1753 	bool marked = false;
1754 	nid_t ino = inode->i_ino;
1755 	int nr_folios;
1756 	int nwritten = 0;
1757 
1758 	if (atomic) {
1759 		last_page = last_fsync_dnode(sbi, ino);
1760 		if (IS_ERR_OR_NULL(last_page))
1761 			return PTR_ERR_OR_ZERO(last_page);
1762 	}
1763 retry:
1764 	folio_batch_init(&fbatch);
1765 	index = 0;
1766 
1767 	while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1768 					(pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1769 					&fbatch))) {
1770 		int i;
1771 
1772 		for (i = 0; i < nr_folios; i++) {
1773 			struct page *page = &fbatch.folios[i]->page;
1774 			bool submitted = false;
1775 
1776 			if (unlikely(f2fs_cp_error(sbi))) {
1777 				f2fs_put_page(last_page, 0);
1778 				folio_batch_release(&fbatch);
1779 				ret = -EIO;
1780 				goto out;
1781 			}
1782 
1783 			if (!IS_DNODE(page) || !is_cold_node(page))
1784 				continue;
1785 			if (ino_of_node(page) != ino)
1786 				continue;
1787 
1788 			lock_page(page);
1789 
1790 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1791 continue_unlock:
1792 				unlock_page(page);
1793 				continue;
1794 			}
1795 			if (ino_of_node(page) != ino)
1796 				goto continue_unlock;
1797 
1798 			if (!PageDirty(page) && page != last_page) {
1799 				/* someone wrote it for us */
1800 				goto continue_unlock;
1801 			}
1802 
1803 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1804 
1805 			set_fsync_mark(page, 0);
1806 			set_dentry_mark(page, 0);
1807 
1808 			if (!atomic || page == last_page) {
1809 				set_fsync_mark(page, 1);
1810 				percpu_counter_inc(&sbi->rf_node_block_count);
1811 				if (IS_INODE(page)) {
1812 					if (is_inode_flag_set(inode,
1813 								FI_DIRTY_INODE))
1814 						f2fs_update_inode(inode, page);
1815 					set_dentry_mark(page,
1816 						f2fs_need_dentry_mark(sbi, ino));
1817 				}
1818 				/* may be written by other thread */
1819 				if (!PageDirty(page))
1820 					set_page_dirty(page);
1821 			}
1822 
1823 			if (!clear_page_dirty_for_io(page))
1824 				goto continue_unlock;
1825 
1826 			ret = __write_node_page(page, atomic &&
1827 						page == last_page,
1828 						&submitted, wbc, true,
1829 						FS_NODE_IO, seq_id);
1830 			if (ret) {
1831 				unlock_page(page);
1832 				f2fs_put_page(last_page, 0);
1833 				break;
1834 			} else if (submitted) {
1835 				nwritten++;
1836 			}
1837 
1838 			if (page == last_page) {
1839 				f2fs_put_page(page, 0);
1840 				marked = true;
1841 				break;
1842 			}
1843 		}
1844 		folio_batch_release(&fbatch);
1845 		cond_resched();
1846 
1847 		if (ret || marked)
1848 			break;
1849 	}
1850 	if (!ret && atomic && !marked) {
1851 		f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1852 			   ino, last_page->index);
1853 		lock_page(last_page);
1854 		f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1855 		set_page_dirty(last_page);
1856 		unlock_page(last_page);
1857 		goto retry;
1858 	}
1859 out:
1860 	if (nwritten)
1861 		f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1862 	return ret ? -EIO : 0;
1863 }
1864 
1865 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1866 {
1867 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1868 	bool clean;
1869 
1870 	if (inode->i_ino != ino)
1871 		return 0;
1872 
1873 	if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1874 		return 0;
1875 
1876 	spin_lock(&sbi->inode_lock[DIRTY_META]);
1877 	clean = list_empty(&F2FS_I(inode)->gdirty_list);
1878 	spin_unlock(&sbi->inode_lock[DIRTY_META]);
1879 
1880 	if (clean)
1881 		return 0;
1882 
1883 	inode = igrab(inode);
1884 	if (!inode)
1885 		return 0;
1886 	return 1;
1887 }
1888 
1889 static bool flush_dirty_inode(struct page *page)
1890 {
1891 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1892 	struct inode *inode;
1893 	nid_t ino = ino_of_node(page);
1894 
1895 	inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1896 	if (!inode)
1897 		return false;
1898 
1899 	f2fs_update_inode(inode, page);
1900 	unlock_page(page);
1901 
1902 	iput(inode);
1903 	return true;
1904 }
1905 
1906 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1907 {
1908 	pgoff_t index = 0;
1909 	struct folio_batch fbatch;
1910 	int nr_folios;
1911 
1912 	folio_batch_init(&fbatch);
1913 
1914 	while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1915 					(pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1916 					&fbatch))) {
1917 		int i;
1918 
1919 		for (i = 0; i < nr_folios; i++) {
1920 			struct page *page = &fbatch.folios[i]->page;
1921 
1922 			if (!IS_DNODE(page))
1923 				continue;
1924 
1925 			lock_page(page);
1926 
1927 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1928 continue_unlock:
1929 				unlock_page(page);
1930 				continue;
1931 			}
1932 
1933 			if (!PageDirty(page)) {
1934 				/* someone wrote it for us */
1935 				goto continue_unlock;
1936 			}
1937 
1938 			/* flush inline_data, if it's async context. */
1939 			if (page_private_inline(page)) {
1940 				clear_page_private_inline(page);
1941 				unlock_page(page);
1942 				flush_inline_data(sbi, ino_of_node(page));
1943 				continue;
1944 			}
1945 			unlock_page(page);
1946 		}
1947 		folio_batch_release(&fbatch);
1948 		cond_resched();
1949 	}
1950 }
1951 
1952 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1953 				struct writeback_control *wbc,
1954 				bool do_balance, enum iostat_type io_type)
1955 {
1956 	pgoff_t index;
1957 	struct folio_batch fbatch;
1958 	int step = 0;
1959 	int nwritten = 0;
1960 	int ret = 0;
1961 	int nr_folios, done = 0;
1962 
1963 	folio_batch_init(&fbatch);
1964 
1965 next_step:
1966 	index = 0;
1967 
1968 	while (!done && (nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi),
1969 				&index, (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1970 				&fbatch))) {
1971 		int i;
1972 
1973 		for (i = 0; i < nr_folios; i++) {
1974 			struct page *page = &fbatch.folios[i]->page;
1975 			bool submitted = false;
1976 
1977 			/* give a priority to WB_SYNC threads */
1978 			if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1979 					wbc->sync_mode == WB_SYNC_NONE) {
1980 				done = 1;
1981 				break;
1982 			}
1983 
1984 			/*
1985 			 * flushing sequence with step:
1986 			 * 0. indirect nodes
1987 			 * 1. dentry dnodes
1988 			 * 2. file dnodes
1989 			 */
1990 			if (step == 0 && IS_DNODE(page))
1991 				continue;
1992 			if (step == 1 && (!IS_DNODE(page) ||
1993 						is_cold_node(page)))
1994 				continue;
1995 			if (step == 2 && (!IS_DNODE(page) ||
1996 						!is_cold_node(page)))
1997 				continue;
1998 lock_node:
1999 			if (wbc->sync_mode == WB_SYNC_ALL)
2000 				lock_page(page);
2001 			else if (!trylock_page(page))
2002 				continue;
2003 
2004 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
2005 continue_unlock:
2006 				unlock_page(page);
2007 				continue;
2008 			}
2009 
2010 			if (!PageDirty(page)) {
2011 				/* someone wrote it for us */
2012 				goto continue_unlock;
2013 			}
2014 
2015 			/* flush inline_data/inode, if it's async context. */
2016 			if (!do_balance)
2017 				goto write_node;
2018 
2019 			/* flush inline_data */
2020 			if (page_private_inline(page)) {
2021 				clear_page_private_inline(page);
2022 				unlock_page(page);
2023 				flush_inline_data(sbi, ino_of_node(page));
2024 				goto lock_node;
2025 			}
2026 
2027 			/* flush dirty inode */
2028 			if (IS_INODE(page) && flush_dirty_inode(page))
2029 				goto lock_node;
2030 write_node:
2031 			f2fs_wait_on_page_writeback(page, NODE, true, true);
2032 
2033 			if (!clear_page_dirty_for_io(page))
2034 				goto continue_unlock;
2035 
2036 			set_fsync_mark(page, 0);
2037 			set_dentry_mark(page, 0);
2038 
2039 			ret = __write_node_page(page, false, &submitted,
2040 						wbc, do_balance, io_type, NULL);
2041 			if (ret)
2042 				unlock_page(page);
2043 			else if (submitted)
2044 				nwritten++;
2045 
2046 			if (--wbc->nr_to_write == 0)
2047 				break;
2048 		}
2049 		folio_batch_release(&fbatch);
2050 		cond_resched();
2051 
2052 		if (wbc->nr_to_write == 0) {
2053 			step = 2;
2054 			break;
2055 		}
2056 	}
2057 
2058 	if (step < 2) {
2059 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2060 				wbc->sync_mode == WB_SYNC_NONE && step == 1)
2061 			goto out;
2062 		step++;
2063 		goto next_step;
2064 	}
2065 out:
2066 	if (nwritten)
2067 		f2fs_submit_merged_write(sbi, NODE);
2068 
2069 	if (unlikely(f2fs_cp_error(sbi)))
2070 		return -EIO;
2071 	return ret;
2072 }
2073 
2074 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2075 						unsigned int seq_id)
2076 {
2077 	struct fsync_node_entry *fn;
2078 	struct page *page;
2079 	struct list_head *head = &sbi->fsync_node_list;
2080 	unsigned long flags;
2081 	unsigned int cur_seq_id = 0;
2082 
2083 	while (seq_id && cur_seq_id < seq_id) {
2084 		spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2085 		if (list_empty(head)) {
2086 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2087 			break;
2088 		}
2089 		fn = list_first_entry(head, struct fsync_node_entry, list);
2090 		if (fn->seq_id > seq_id) {
2091 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2092 			break;
2093 		}
2094 		cur_seq_id = fn->seq_id;
2095 		page = fn->page;
2096 		get_page(page);
2097 		spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2098 
2099 		f2fs_wait_on_page_writeback(page, NODE, true, false);
2100 
2101 		put_page(page);
2102 	}
2103 
2104 	return filemap_check_errors(NODE_MAPPING(sbi));
2105 }
2106 
2107 static int f2fs_write_node_pages(struct address_space *mapping,
2108 			    struct writeback_control *wbc)
2109 {
2110 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2111 	struct blk_plug plug;
2112 	long diff;
2113 
2114 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2115 		goto skip_write;
2116 
2117 	/* balancing f2fs's metadata in background */
2118 	f2fs_balance_fs_bg(sbi, true);
2119 
2120 	/* collect a number of dirty node pages and write together */
2121 	if (wbc->sync_mode != WB_SYNC_ALL &&
2122 			get_pages(sbi, F2FS_DIRTY_NODES) <
2123 					nr_pages_to_skip(sbi, NODE))
2124 		goto skip_write;
2125 
2126 	if (wbc->sync_mode == WB_SYNC_ALL)
2127 		atomic_inc(&sbi->wb_sync_req[NODE]);
2128 	else if (atomic_read(&sbi->wb_sync_req[NODE])) {
2129 		/* to avoid potential deadlock */
2130 		if (current->plug)
2131 			blk_finish_plug(current->plug);
2132 		goto skip_write;
2133 	}
2134 
2135 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2136 
2137 	diff = nr_pages_to_write(sbi, NODE, wbc);
2138 	blk_start_plug(&plug);
2139 	f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2140 	blk_finish_plug(&plug);
2141 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2142 
2143 	if (wbc->sync_mode == WB_SYNC_ALL)
2144 		atomic_dec(&sbi->wb_sync_req[NODE]);
2145 	return 0;
2146 
2147 skip_write:
2148 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2149 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2150 	return 0;
2151 }
2152 
2153 static bool f2fs_dirty_node_folio(struct address_space *mapping,
2154 		struct folio *folio)
2155 {
2156 	trace_f2fs_set_page_dirty(&folio->page, NODE);
2157 
2158 	if (!folio_test_uptodate(folio))
2159 		folio_mark_uptodate(folio);
2160 #ifdef CONFIG_F2FS_CHECK_FS
2161 	if (IS_INODE(&folio->page))
2162 		f2fs_inode_chksum_set(F2FS_M_SB(mapping), &folio->page);
2163 #endif
2164 	if (filemap_dirty_folio(mapping, folio)) {
2165 		inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
2166 		set_page_private_reference(&folio->page);
2167 		return true;
2168 	}
2169 	return false;
2170 }
2171 
2172 /*
2173  * Structure of the f2fs node operations
2174  */
2175 const struct address_space_operations f2fs_node_aops = {
2176 	.writepage	= f2fs_write_node_page,
2177 	.writepages	= f2fs_write_node_pages,
2178 	.dirty_folio	= f2fs_dirty_node_folio,
2179 	.invalidate_folio = f2fs_invalidate_folio,
2180 	.release_folio	= f2fs_release_folio,
2181 	.migrate_folio	= filemap_migrate_folio,
2182 };
2183 
2184 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2185 						nid_t n)
2186 {
2187 	return radix_tree_lookup(&nm_i->free_nid_root, n);
2188 }
2189 
2190 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2191 				struct free_nid *i)
2192 {
2193 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2194 	int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2195 
2196 	if (err)
2197 		return err;
2198 
2199 	nm_i->nid_cnt[FREE_NID]++;
2200 	list_add_tail(&i->list, &nm_i->free_nid_list);
2201 	return 0;
2202 }
2203 
2204 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2205 			struct free_nid *i, enum nid_state state)
2206 {
2207 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2208 
2209 	f2fs_bug_on(sbi, state != i->state);
2210 	nm_i->nid_cnt[state]--;
2211 	if (state == FREE_NID)
2212 		list_del(&i->list);
2213 	radix_tree_delete(&nm_i->free_nid_root, i->nid);
2214 }
2215 
2216 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2217 			enum nid_state org_state, enum nid_state dst_state)
2218 {
2219 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2220 
2221 	f2fs_bug_on(sbi, org_state != i->state);
2222 	i->state = dst_state;
2223 	nm_i->nid_cnt[org_state]--;
2224 	nm_i->nid_cnt[dst_state]++;
2225 
2226 	switch (dst_state) {
2227 	case PREALLOC_NID:
2228 		list_del(&i->list);
2229 		break;
2230 	case FREE_NID:
2231 		list_add_tail(&i->list, &nm_i->free_nid_list);
2232 		break;
2233 	default:
2234 		BUG_ON(1);
2235 	}
2236 }
2237 
2238 bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi)
2239 {
2240 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2241 	unsigned int i;
2242 	bool ret = true;
2243 
2244 	f2fs_down_read(&nm_i->nat_tree_lock);
2245 	for (i = 0; i < nm_i->nat_blocks; i++) {
2246 		if (!test_bit_le(i, nm_i->nat_block_bitmap)) {
2247 			ret = false;
2248 			break;
2249 		}
2250 	}
2251 	f2fs_up_read(&nm_i->nat_tree_lock);
2252 
2253 	return ret;
2254 }
2255 
2256 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2257 							bool set, bool build)
2258 {
2259 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2260 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2261 	unsigned int nid_ofs = nid - START_NID(nid);
2262 
2263 	if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2264 		return;
2265 
2266 	if (set) {
2267 		if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2268 			return;
2269 		__set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2270 		nm_i->free_nid_count[nat_ofs]++;
2271 	} else {
2272 		if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2273 			return;
2274 		__clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2275 		if (!build)
2276 			nm_i->free_nid_count[nat_ofs]--;
2277 	}
2278 }
2279 
2280 /* return if the nid is recognized as free */
2281 static bool add_free_nid(struct f2fs_sb_info *sbi,
2282 				nid_t nid, bool build, bool update)
2283 {
2284 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2285 	struct free_nid *i, *e;
2286 	struct nat_entry *ne;
2287 	int err = -EINVAL;
2288 	bool ret = false;
2289 
2290 	/* 0 nid should not be used */
2291 	if (unlikely(nid == 0))
2292 		return false;
2293 
2294 	if (unlikely(f2fs_check_nid_range(sbi, nid)))
2295 		return false;
2296 
2297 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL);
2298 	i->nid = nid;
2299 	i->state = FREE_NID;
2300 
2301 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2302 
2303 	spin_lock(&nm_i->nid_list_lock);
2304 
2305 	if (build) {
2306 		/*
2307 		 *   Thread A             Thread B
2308 		 *  - f2fs_create
2309 		 *   - f2fs_new_inode
2310 		 *    - f2fs_alloc_nid
2311 		 *     - __insert_nid_to_list(PREALLOC_NID)
2312 		 *                     - f2fs_balance_fs_bg
2313 		 *                      - f2fs_build_free_nids
2314 		 *                       - __f2fs_build_free_nids
2315 		 *                        - scan_nat_page
2316 		 *                         - add_free_nid
2317 		 *                          - __lookup_nat_cache
2318 		 *  - f2fs_add_link
2319 		 *   - f2fs_init_inode_metadata
2320 		 *    - f2fs_new_inode_page
2321 		 *     - f2fs_new_node_page
2322 		 *      - set_node_addr
2323 		 *  - f2fs_alloc_nid_done
2324 		 *   - __remove_nid_from_list(PREALLOC_NID)
2325 		 *                         - __insert_nid_to_list(FREE_NID)
2326 		 */
2327 		ne = __lookup_nat_cache(nm_i, nid);
2328 		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2329 				nat_get_blkaddr(ne) != NULL_ADDR))
2330 			goto err_out;
2331 
2332 		e = __lookup_free_nid_list(nm_i, nid);
2333 		if (e) {
2334 			if (e->state == FREE_NID)
2335 				ret = true;
2336 			goto err_out;
2337 		}
2338 	}
2339 	ret = true;
2340 	err = __insert_free_nid(sbi, i);
2341 err_out:
2342 	if (update) {
2343 		update_free_nid_bitmap(sbi, nid, ret, build);
2344 		if (!build)
2345 			nm_i->available_nids++;
2346 	}
2347 	spin_unlock(&nm_i->nid_list_lock);
2348 	radix_tree_preload_end();
2349 
2350 	if (err)
2351 		kmem_cache_free(free_nid_slab, i);
2352 	return ret;
2353 }
2354 
2355 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2356 {
2357 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2358 	struct free_nid *i;
2359 	bool need_free = false;
2360 
2361 	spin_lock(&nm_i->nid_list_lock);
2362 	i = __lookup_free_nid_list(nm_i, nid);
2363 	if (i && i->state == FREE_NID) {
2364 		__remove_free_nid(sbi, i, FREE_NID);
2365 		need_free = true;
2366 	}
2367 	spin_unlock(&nm_i->nid_list_lock);
2368 
2369 	if (need_free)
2370 		kmem_cache_free(free_nid_slab, i);
2371 }
2372 
2373 static int scan_nat_page(struct f2fs_sb_info *sbi,
2374 			struct page *nat_page, nid_t start_nid)
2375 {
2376 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2377 	struct f2fs_nat_block *nat_blk = page_address(nat_page);
2378 	block_t blk_addr;
2379 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2380 	int i;
2381 
2382 	__set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2383 
2384 	i = start_nid % NAT_ENTRY_PER_BLOCK;
2385 
2386 	for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2387 		if (unlikely(start_nid >= nm_i->max_nid))
2388 			break;
2389 
2390 		blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2391 
2392 		if (blk_addr == NEW_ADDR)
2393 			return -EFSCORRUPTED;
2394 
2395 		if (blk_addr == NULL_ADDR) {
2396 			add_free_nid(sbi, start_nid, true, true);
2397 		} else {
2398 			spin_lock(&NM_I(sbi)->nid_list_lock);
2399 			update_free_nid_bitmap(sbi, start_nid, false, true);
2400 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2401 		}
2402 	}
2403 
2404 	return 0;
2405 }
2406 
2407 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2408 {
2409 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2410 	struct f2fs_journal *journal = curseg->journal;
2411 	int i;
2412 
2413 	down_read(&curseg->journal_rwsem);
2414 	for (i = 0; i < nats_in_cursum(journal); i++) {
2415 		block_t addr;
2416 		nid_t nid;
2417 
2418 		addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2419 		nid = le32_to_cpu(nid_in_journal(journal, i));
2420 		if (addr == NULL_ADDR)
2421 			add_free_nid(sbi, nid, true, false);
2422 		else
2423 			remove_free_nid(sbi, nid);
2424 	}
2425 	up_read(&curseg->journal_rwsem);
2426 }
2427 
2428 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2429 {
2430 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2431 	unsigned int i, idx;
2432 	nid_t nid;
2433 
2434 	f2fs_down_read(&nm_i->nat_tree_lock);
2435 
2436 	for (i = 0; i < nm_i->nat_blocks; i++) {
2437 		if (!test_bit_le(i, nm_i->nat_block_bitmap))
2438 			continue;
2439 		if (!nm_i->free_nid_count[i])
2440 			continue;
2441 		for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2442 			idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2443 						NAT_ENTRY_PER_BLOCK, idx);
2444 			if (idx >= NAT_ENTRY_PER_BLOCK)
2445 				break;
2446 
2447 			nid = i * NAT_ENTRY_PER_BLOCK + idx;
2448 			add_free_nid(sbi, nid, true, false);
2449 
2450 			if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2451 				goto out;
2452 		}
2453 	}
2454 out:
2455 	scan_curseg_cache(sbi);
2456 
2457 	f2fs_up_read(&nm_i->nat_tree_lock);
2458 }
2459 
2460 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2461 						bool sync, bool mount)
2462 {
2463 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2464 	int i = 0, ret;
2465 	nid_t nid = nm_i->next_scan_nid;
2466 
2467 	if (unlikely(nid >= nm_i->max_nid))
2468 		nid = 0;
2469 
2470 	if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2471 		nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2472 
2473 	/* Enough entries */
2474 	if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2475 		return 0;
2476 
2477 	if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2478 		return 0;
2479 
2480 	if (!mount) {
2481 		/* try to find free nids in free_nid_bitmap */
2482 		scan_free_nid_bits(sbi);
2483 
2484 		if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2485 			return 0;
2486 	}
2487 
2488 	/* readahead nat pages to be scanned */
2489 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2490 							META_NAT, true);
2491 
2492 	f2fs_down_read(&nm_i->nat_tree_lock);
2493 
2494 	while (1) {
2495 		if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2496 						nm_i->nat_block_bitmap)) {
2497 			struct page *page = get_current_nat_page(sbi, nid);
2498 
2499 			if (IS_ERR(page)) {
2500 				ret = PTR_ERR(page);
2501 			} else {
2502 				ret = scan_nat_page(sbi, page, nid);
2503 				f2fs_put_page(page, 1);
2504 			}
2505 
2506 			if (ret) {
2507 				f2fs_up_read(&nm_i->nat_tree_lock);
2508 
2509 				if (ret == -EFSCORRUPTED) {
2510 					f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2511 					set_sbi_flag(sbi, SBI_NEED_FSCK);
2512 					f2fs_handle_error(sbi,
2513 						ERROR_INCONSISTENT_NAT);
2514 				}
2515 
2516 				return ret;
2517 			}
2518 		}
2519 
2520 		nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2521 		if (unlikely(nid >= nm_i->max_nid))
2522 			nid = 0;
2523 
2524 		if (++i >= FREE_NID_PAGES)
2525 			break;
2526 	}
2527 
2528 	/* go to the next free nat pages to find free nids abundantly */
2529 	nm_i->next_scan_nid = nid;
2530 
2531 	/* find free nids from current sum_pages */
2532 	scan_curseg_cache(sbi);
2533 
2534 	f2fs_up_read(&nm_i->nat_tree_lock);
2535 
2536 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2537 					nm_i->ra_nid_pages, META_NAT, false);
2538 
2539 	return 0;
2540 }
2541 
2542 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2543 {
2544 	int ret;
2545 
2546 	mutex_lock(&NM_I(sbi)->build_lock);
2547 	ret = __f2fs_build_free_nids(sbi, sync, mount);
2548 	mutex_unlock(&NM_I(sbi)->build_lock);
2549 
2550 	return ret;
2551 }
2552 
2553 /*
2554  * If this function returns success, caller can obtain a new nid
2555  * from second parameter of this function.
2556  * The returned nid could be used ino as well as nid when inode is created.
2557  */
2558 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2559 {
2560 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2561 	struct free_nid *i = NULL;
2562 retry:
2563 	if (time_to_inject(sbi, FAULT_ALLOC_NID))
2564 		return false;
2565 
2566 	spin_lock(&nm_i->nid_list_lock);
2567 
2568 	if (unlikely(nm_i->available_nids == 0)) {
2569 		spin_unlock(&nm_i->nid_list_lock);
2570 		return false;
2571 	}
2572 
2573 	/* We should not use stale free nids created by f2fs_build_free_nids */
2574 	if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2575 		f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2576 		i = list_first_entry(&nm_i->free_nid_list,
2577 					struct free_nid, list);
2578 		*nid = i->nid;
2579 
2580 		__move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2581 		nm_i->available_nids--;
2582 
2583 		update_free_nid_bitmap(sbi, *nid, false, false);
2584 
2585 		spin_unlock(&nm_i->nid_list_lock);
2586 		return true;
2587 	}
2588 	spin_unlock(&nm_i->nid_list_lock);
2589 
2590 	/* Let's scan nat pages and its caches to get free nids */
2591 	if (!f2fs_build_free_nids(sbi, true, false))
2592 		goto retry;
2593 	return false;
2594 }
2595 
2596 /*
2597  * f2fs_alloc_nid() should be called prior to this function.
2598  */
2599 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2600 {
2601 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2602 	struct free_nid *i;
2603 
2604 	spin_lock(&nm_i->nid_list_lock);
2605 	i = __lookup_free_nid_list(nm_i, nid);
2606 	f2fs_bug_on(sbi, !i);
2607 	__remove_free_nid(sbi, i, PREALLOC_NID);
2608 	spin_unlock(&nm_i->nid_list_lock);
2609 
2610 	kmem_cache_free(free_nid_slab, i);
2611 }
2612 
2613 /*
2614  * f2fs_alloc_nid() should be called prior to this function.
2615  */
2616 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2617 {
2618 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2619 	struct free_nid *i;
2620 	bool need_free = false;
2621 
2622 	if (!nid)
2623 		return;
2624 
2625 	spin_lock(&nm_i->nid_list_lock);
2626 	i = __lookup_free_nid_list(nm_i, nid);
2627 	f2fs_bug_on(sbi, !i);
2628 
2629 	if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2630 		__remove_free_nid(sbi, i, PREALLOC_NID);
2631 		need_free = true;
2632 	} else {
2633 		__move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2634 	}
2635 
2636 	nm_i->available_nids++;
2637 
2638 	update_free_nid_bitmap(sbi, nid, true, false);
2639 
2640 	spin_unlock(&nm_i->nid_list_lock);
2641 
2642 	if (need_free)
2643 		kmem_cache_free(free_nid_slab, i);
2644 }
2645 
2646 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2647 {
2648 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2649 	int nr = nr_shrink;
2650 
2651 	if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2652 		return 0;
2653 
2654 	if (!mutex_trylock(&nm_i->build_lock))
2655 		return 0;
2656 
2657 	while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2658 		struct free_nid *i, *next;
2659 		unsigned int batch = SHRINK_NID_BATCH_SIZE;
2660 
2661 		spin_lock(&nm_i->nid_list_lock);
2662 		list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2663 			if (!nr_shrink || !batch ||
2664 				nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2665 				break;
2666 			__remove_free_nid(sbi, i, FREE_NID);
2667 			kmem_cache_free(free_nid_slab, i);
2668 			nr_shrink--;
2669 			batch--;
2670 		}
2671 		spin_unlock(&nm_i->nid_list_lock);
2672 	}
2673 
2674 	mutex_unlock(&nm_i->build_lock);
2675 
2676 	return nr - nr_shrink;
2677 }
2678 
2679 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2680 {
2681 	void *src_addr, *dst_addr;
2682 	size_t inline_size;
2683 	struct page *ipage;
2684 	struct f2fs_inode *ri;
2685 
2686 	ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2687 	if (IS_ERR(ipage))
2688 		return PTR_ERR(ipage);
2689 
2690 	ri = F2FS_INODE(page);
2691 	if (ri->i_inline & F2FS_INLINE_XATTR) {
2692 		if (!f2fs_has_inline_xattr(inode)) {
2693 			set_inode_flag(inode, FI_INLINE_XATTR);
2694 			stat_inc_inline_xattr(inode);
2695 		}
2696 	} else {
2697 		if (f2fs_has_inline_xattr(inode)) {
2698 			stat_dec_inline_xattr(inode);
2699 			clear_inode_flag(inode, FI_INLINE_XATTR);
2700 		}
2701 		goto update_inode;
2702 	}
2703 
2704 	dst_addr = inline_xattr_addr(inode, ipage);
2705 	src_addr = inline_xattr_addr(inode, page);
2706 	inline_size = inline_xattr_size(inode);
2707 
2708 	f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2709 	memcpy(dst_addr, src_addr, inline_size);
2710 update_inode:
2711 	f2fs_update_inode(inode, ipage);
2712 	f2fs_put_page(ipage, 1);
2713 	return 0;
2714 }
2715 
2716 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2717 {
2718 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2719 	nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2720 	nid_t new_xnid;
2721 	struct dnode_of_data dn;
2722 	struct node_info ni;
2723 	struct page *xpage;
2724 	int err;
2725 
2726 	if (!prev_xnid)
2727 		goto recover_xnid;
2728 
2729 	/* 1: invalidate the previous xattr nid */
2730 	err = f2fs_get_node_info(sbi, prev_xnid, &ni, false);
2731 	if (err)
2732 		return err;
2733 
2734 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
2735 	dec_valid_node_count(sbi, inode, false);
2736 	set_node_addr(sbi, &ni, NULL_ADDR, false);
2737 
2738 recover_xnid:
2739 	/* 2: update xattr nid in inode */
2740 	if (!f2fs_alloc_nid(sbi, &new_xnid))
2741 		return -ENOSPC;
2742 
2743 	set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2744 	xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2745 	if (IS_ERR(xpage)) {
2746 		f2fs_alloc_nid_failed(sbi, new_xnid);
2747 		return PTR_ERR(xpage);
2748 	}
2749 
2750 	f2fs_alloc_nid_done(sbi, new_xnid);
2751 	f2fs_update_inode_page(inode);
2752 
2753 	/* 3: update and set xattr node page dirty */
2754 	if (page) {
2755 		memcpy(F2FS_NODE(xpage), F2FS_NODE(page),
2756 				VALID_XATTR_BLOCK_SIZE);
2757 		set_page_dirty(xpage);
2758 	}
2759 	f2fs_put_page(xpage, 1);
2760 
2761 	return 0;
2762 }
2763 
2764 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2765 {
2766 	struct f2fs_inode *src, *dst;
2767 	nid_t ino = ino_of_node(page);
2768 	struct node_info old_ni, new_ni;
2769 	struct page *ipage;
2770 	int err;
2771 
2772 	err = f2fs_get_node_info(sbi, ino, &old_ni, false);
2773 	if (err)
2774 		return err;
2775 
2776 	if (unlikely(old_ni.blk_addr != NULL_ADDR))
2777 		return -EINVAL;
2778 retry:
2779 	ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2780 	if (!ipage) {
2781 		memalloc_retry_wait(GFP_NOFS);
2782 		goto retry;
2783 	}
2784 
2785 	/* Should not use this inode from free nid list */
2786 	remove_free_nid(sbi, ino);
2787 
2788 	if (!PageUptodate(ipage))
2789 		SetPageUptodate(ipage);
2790 	fill_node_footer(ipage, ino, ino, 0, true);
2791 	set_cold_node(ipage, false);
2792 
2793 	src = F2FS_INODE(page);
2794 	dst = F2FS_INODE(ipage);
2795 
2796 	memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2797 	dst->i_size = 0;
2798 	dst->i_blocks = cpu_to_le64(1);
2799 	dst->i_links = cpu_to_le32(1);
2800 	dst->i_xattr_nid = 0;
2801 	dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2802 	if (dst->i_inline & F2FS_EXTRA_ATTR) {
2803 		dst->i_extra_isize = src->i_extra_isize;
2804 
2805 		if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2806 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2807 							i_inline_xattr_size))
2808 			dst->i_inline_xattr_size = src->i_inline_xattr_size;
2809 
2810 		if (f2fs_sb_has_project_quota(sbi) &&
2811 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2812 								i_projid))
2813 			dst->i_projid = src->i_projid;
2814 
2815 		if (f2fs_sb_has_inode_crtime(sbi) &&
2816 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2817 							i_crtime_nsec)) {
2818 			dst->i_crtime = src->i_crtime;
2819 			dst->i_crtime_nsec = src->i_crtime_nsec;
2820 		}
2821 	}
2822 
2823 	new_ni = old_ni;
2824 	new_ni.ino = ino;
2825 
2826 	if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2827 		WARN_ON(1);
2828 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2829 	inc_valid_inode_count(sbi);
2830 	set_page_dirty(ipage);
2831 	f2fs_put_page(ipage, 1);
2832 	return 0;
2833 }
2834 
2835 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2836 			unsigned int segno, struct f2fs_summary_block *sum)
2837 {
2838 	struct f2fs_node *rn;
2839 	struct f2fs_summary *sum_entry;
2840 	block_t addr;
2841 	int i, idx, last_offset, nrpages;
2842 
2843 	/* scan the node segment */
2844 	last_offset = sbi->blocks_per_seg;
2845 	addr = START_BLOCK(sbi, segno);
2846 	sum_entry = &sum->entries[0];
2847 
2848 	for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2849 		nrpages = bio_max_segs(last_offset - i);
2850 
2851 		/* readahead node pages */
2852 		f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2853 
2854 		for (idx = addr; idx < addr + nrpages; idx++) {
2855 			struct page *page = f2fs_get_tmp_page(sbi, idx);
2856 
2857 			if (IS_ERR(page))
2858 				return PTR_ERR(page);
2859 
2860 			rn = F2FS_NODE(page);
2861 			sum_entry->nid = rn->footer.nid;
2862 			sum_entry->version = 0;
2863 			sum_entry->ofs_in_node = 0;
2864 			sum_entry++;
2865 			f2fs_put_page(page, 1);
2866 		}
2867 
2868 		invalidate_mapping_pages(META_MAPPING(sbi), addr,
2869 							addr + nrpages);
2870 	}
2871 	return 0;
2872 }
2873 
2874 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2875 {
2876 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2877 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2878 	struct f2fs_journal *journal = curseg->journal;
2879 	int i;
2880 
2881 	down_write(&curseg->journal_rwsem);
2882 	for (i = 0; i < nats_in_cursum(journal); i++) {
2883 		struct nat_entry *ne;
2884 		struct f2fs_nat_entry raw_ne;
2885 		nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2886 
2887 		if (f2fs_check_nid_range(sbi, nid))
2888 			continue;
2889 
2890 		raw_ne = nat_in_journal(journal, i);
2891 
2892 		ne = __lookup_nat_cache(nm_i, nid);
2893 		if (!ne) {
2894 			ne = __alloc_nat_entry(sbi, nid, true);
2895 			__init_nat_entry(nm_i, ne, &raw_ne, true);
2896 		}
2897 
2898 		/*
2899 		 * if a free nat in journal has not been used after last
2900 		 * checkpoint, we should remove it from available nids,
2901 		 * since later we will add it again.
2902 		 */
2903 		if (!get_nat_flag(ne, IS_DIRTY) &&
2904 				le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2905 			spin_lock(&nm_i->nid_list_lock);
2906 			nm_i->available_nids--;
2907 			spin_unlock(&nm_i->nid_list_lock);
2908 		}
2909 
2910 		__set_nat_cache_dirty(nm_i, ne);
2911 	}
2912 	update_nats_in_cursum(journal, -i);
2913 	up_write(&curseg->journal_rwsem);
2914 }
2915 
2916 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2917 						struct list_head *head, int max)
2918 {
2919 	struct nat_entry_set *cur;
2920 
2921 	if (nes->entry_cnt >= max)
2922 		goto add_out;
2923 
2924 	list_for_each_entry(cur, head, set_list) {
2925 		if (cur->entry_cnt >= nes->entry_cnt) {
2926 			list_add(&nes->set_list, cur->set_list.prev);
2927 			return;
2928 		}
2929 	}
2930 add_out:
2931 	list_add_tail(&nes->set_list, head);
2932 }
2933 
2934 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs,
2935 							unsigned int valid)
2936 {
2937 	if (valid == 0) {
2938 		__set_bit_le(nat_ofs, nm_i->empty_nat_bits);
2939 		__clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2940 		return;
2941 	}
2942 
2943 	__clear_bit_le(nat_ofs, nm_i->empty_nat_bits);
2944 	if (valid == NAT_ENTRY_PER_BLOCK)
2945 		__set_bit_le(nat_ofs, nm_i->full_nat_bits);
2946 	else
2947 		__clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2948 }
2949 
2950 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2951 						struct page *page)
2952 {
2953 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2954 	unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2955 	struct f2fs_nat_block *nat_blk = page_address(page);
2956 	int valid = 0;
2957 	int i = 0;
2958 
2959 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
2960 		return;
2961 
2962 	if (nat_index == 0) {
2963 		valid = 1;
2964 		i = 1;
2965 	}
2966 	for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2967 		if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2968 			valid++;
2969 	}
2970 
2971 	__update_nat_bits(nm_i, nat_index, valid);
2972 }
2973 
2974 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi)
2975 {
2976 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2977 	unsigned int nat_ofs;
2978 
2979 	f2fs_down_read(&nm_i->nat_tree_lock);
2980 
2981 	for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) {
2982 		unsigned int valid = 0, nid_ofs = 0;
2983 
2984 		/* handle nid zero due to it should never be used */
2985 		if (unlikely(nat_ofs == 0)) {
2986 			valid = 1;
2987 			nid_ofs = 1;
2988 		}
2989 
2990 		for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) {
2991 			if (!test_bit_le(nid_ofs,
2992 					nm_i->free_nid_bitmap[nat_ofs]))
2993 				valid++;
2994 		}
2995 
2996 		__update_nat_bits(nm_i, nat_ofs, valid);
2997 	}
2998 
2999 	f2fs_up_read(&nm_i->nat_tree_lock);
3000 }
3001 
3002 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
3003 		struct nat_entry_set *set, struct cp_control *cpc)
3004 {
3005 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3006 	struct f2fs_journal *journal = curseg->journal;
3007 	nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
3008 	bool to_journal = true;
3009 	struct f2fs_nat_block *nat_blk;
3010 	struct nat_entry *ne, *cur;
3011 	struct page *page = NULL;
3012 
3013 	/*
3014 	 * there are two steps to flush nat entries:
3015 	 * #1, flush nat entries to journal in current hot data summary block.
3016 	 * #2, flush nat entries to nat page.
3017 	 */
3018 	if ((cpc->reason & CP_UMOUNT) ||
3019 		!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
3020 		to_journal = false;
3021 
3022 	if (to_journal) {
3023 		down_write(&curseg->journal_rwsem);
3024 	} else {
3025 		page = get_next_nat_page(sbi, start_nid);
3026 		if (IS_ERR(page))
3027 			return PTR_ERR(page);
3028 
3029 		nat_blk = page_address(page);
3030 		f2fs_bug_on(sbi, !nat_blk);
3031 	}
3032 
3033 	/* flush dirty nats in nat entry set */
3034 	list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
3035 		struct f2fs_nat_entry *raw_ne;
3036 		nid_t nid = nat_get_nid(ne);
3037 		int offset;
3038 
3039 		f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
3040 
3041 		if (to_journal) {
3042 			offset = f2fs_lookup_journal_in_cursum(journal,
3043 							NAT_JOURNAL, nid, 1);
3044 			f2fs_bug_on(sbi, offset < 0);
3045 			raw_ne = &nat_in_journal(journal, offset);
3046 			nid_in_journal(journal, offset) = cpu_to_le32(nid);
3047 		} else {
3048 			raw_ne = &nat_blk->entries[nid - start_nid];
3049 		}
3050 		raw_nat_from_node_info(raw_ne, &ne->ni);
3051 		nat_reset_flag(ne);
3052 		__clear_nat_cache_dirty(NM_I(sbi), set, ne);
3053 		if (nat_get_blkaddr(ne) == NULL_ADDR) {
3054 			add_free_nid(sbi, nid, false, true);
3055 		} else {
3056 			spin_lock(&NM_I(sbi)->nid_list_lock);
3057 			update_free_nid_bitmap(sbi, nid, false, false);
3058 			spin_unlock(&NM_I(sbi)->nid_list_lock);
3059 		}
3060 	}
3061 
3062 	if (to_journal) {
3063 		up_write(&curseg->journal_rwsem);
3064 	} else {
3065 		update_nat_bits(sbi, start_nid, page);
3066 		f2fs_put_page(page, 1);
3067 	}
3068 
3069 	/* Allow dirty nats by node block allocation in write_begin */
3070 	if (!set->entry_cnt) {
3071 		radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
3072 		kmem_cache_free(nat_entry_set_slab, set);
3073 	}
3074 	return 0;
3075 }
3076 
3077 /*
3078  * This function is called during the checkpointing process.
3079  */
3080 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3081 {
3082 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3083 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3084 	struct f2fs_journal *journal = curseg->journal;
3085 	struct nat_entry_set *setvec[NAT_VEC_SIZE];
3086 	struct nat_entry_set *set, *tmp;
3087 	unsigned int found;
3088 	nid_t set_idx = 0;
3089 	LIST_HEAD(sets);
3090 	int err = 0;
3091 
3092 	/*
3093 	 * during unmount, let's flush nat_bits before checking
3094 	 * nat_cnt[DIRTY_NAT].
3095 	 */
3096 	if (cpc->reason & CP_UMOUNT) {
3097 		f2fs_down_write(&nm_i->nat_tree_lock);
3098 		remove_nats_in_journal(sbi);
3099 		f2fs_up_write(&nm_i->nat_tree_lock);
3100 	}
3101 
3102 	if (!nm_i->nat_cnt[DIRTY_NAT])
3103 		return 0;
3104 
3105 	f2fs_down_write(&nm_i->nat_tree_lock);
3106 
3107 	/*
3108 	 * if there are no enough space in journal to store dirty nat
3109 	 * entries, remove all entries from journal and merge them
3110 	 * into nat entry set.
3111 	 */
3112 	if (cpc->reason & CP_UMOUNT ||
3113 		!__has_cursum_space(journal,
3114 			nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3115 		remove_nats_in_journal(sbi);
3116 
3117 	while ((found = __gang_lookup_nat_set(nm_i,
3118 					set_idx, NAT_VEC_SIZE, setvec))) {
3119 		unsigned idx;
3120 
3121 		set_idx = setvec[found - 1]->set + 1;
3122 		for (idx = 0; idx < found; idx++)
3123 			__adjust_nat_entry_set(setvec[idx], &sets,
3124 						MAX_NAT_JENTRIES(journal));
3125 	}
3126 
3127 	/* flush dirty nats in nat entry set */
3128 	list_for_each_entry_safe(set, tmp, &sets, set_list) {
3129 		err = __flush_nat_entry_set(sbi, set, cpc);
3130 		if (err)
3131 			break;
3132 	}
3133 
3134 	f2fs_up_write(&nm_i->nat_tree_lock);
3135 	/* Allow dirty nats by node block allocation in write_begin */
3136 
3137 	return err;
3138 }
3139 
3140 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3141 {
3142 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3143 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3144 	unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3145 	unsigned int i;
3146 	__u64 cp_ver = cur_cp_version(ckpt);
3147 	block_t nat_bits_addr;
3148 
3149 	nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3150 	nm_i->nat_bits = f2fs_kvzalloc(sbi,
3151 			nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3152 	if (!nm_i->nat_bits)
3153 		return -ENOMEM;
3154 
3155 	nm_i->full_nat_bits = nm_i->nat_bits + 8;
3156 	nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3157 
3158 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3159 		return 0;
3160 
3161 	nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3162 						nm_i->nat_bits_blocks;
3163 	for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3164 		struct page *page;
3165 
3166 		page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3167 		if (IS_ERR(page))
3168 			return PTR_ERR(page);
3169 
3170 		memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3171 					page_address(page), F2FS_BLKSIZE);
3172 		f2fs_put_page(page, 1);
3173 	}
3174 
3175 	cp_ver |= (cur_cp_crc(ckpt) << 32);
3176 	if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3177 		clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
3178 		f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)",
3179 			cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits));
3180 		return 0;
3181 	}
3182 
3183 	f2fs_notice(sbi, "Found nat_bits in checkpoint");
3184 	return 0;
3185 }
3186 
3187 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3188 {
3189 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3190 	unsigned int i = 0;
3191 	nid_t nid, last_nid;
3192 
3193 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3194 		return;
3195 
3196 	for (i = 0; i < nm_i->nat_blocks; i++) {
3197 		i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3198 		if (i >= nm_i->nat_blocks)
3199 			break;
3200 
3201 		__set_bit_le(i, nm_i->nat_block_bitmap);
3202 
3203 		nid = i * NAT_ENTRY_PER_BLOCK;
3204 		last_nid = nid + NAT_ENTRY_PER_BLOCK;
3205 
3206 		spin_lock(&NM_I(sbi)->nid_list_lock);
3207 		for (; nid < last_nid; nid++)
3208 			update_free_nid_bitmap(sbi, nid, true, true);
3209 		spin_unlock(&NM_I(sbi)->nid_list_lock);
3210 	}
3211 
3212 	for (i = 0; i < nm_i->nat_blocks; i++) {
3213 		i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3214 		if (i >= nm_i->nat_blocks)
3215 			break;
3216 
3217 		__set_bit_le(i, nm_i->nat_block_bitmap);
3218 	}
3219 }
3220 
3221 static int init_node_manager(struct f2fs_sb_info *sbi)
3222 {
3223 	struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3224 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3225 	unsigned char *version_bitmap;
3226 	unsigned int nat_segs;
3227 	int err;
3228 
3229 	nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3230 
3231 	/* segment_count_nat includes pair segment so divide to 2. */
3232 	nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3233 	nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3234 	nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3235 
3236 	/* not used nids: 0, node, meta, (and root counted as valid node) */
3237 	nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3238 						F2FS_RESERVED_NODE_NUM;
3239 	nm_i->nid_cnt[FREE_NID] = 0;
3240 	nm_i->nid_cnt[PREALLOC_NID] = 0;
3241 	nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3242 	nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3243 	nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3244 	nm_i->max_rf_node_blocks = DEF_RF_NODE_BLOCKS;
3245 
3246 	INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3247 	INIT_LIST_HEAD(&nm_i->free_nid_list);
3248 	INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3249 	INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3250 	INIT_LIST_HEAD(&nm_i->nat_entries);
3251 	spin_lock_init(&nm_i->nat_list_lock);
3252 
3253 	mutex_init(&nm_i->build_lock);
3254 	spin_lock_init(&nm_i->nid_list_lock);
3255 	init_f2fs_rwsem(&nm_i->nat_tree_lock);
3256 
3257 	nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3258 	nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3259 	version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3260 	nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3261 					GFP_KERNEL);
3262 	if (!nm_i->nat_bitmap)
3263 		return -ENOMEM;
3264 
3265 	err = __get_nat_bitmaps(sbi);
3266 	if (err)
3267 		return err;
3268 
3269 #ifdef CONFIG_F2FS_CHECK_FS
3270 	nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3271 					GFP_KERNEL);
3272 	if (!nm_i->nat_bitmap_mir)
3273 		return -ENOMEM;
3274 #endif
3275 
3276 	return 0;
3277 }
3278 
3279 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3280 {
3281 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3282 	int i;
3283 
3284 	nm_i->free_nid_bitmap =
3285 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3286 					      nm_i->nat_blocks),
3287 			      GFP_KERNEL);
3288 	if (!nm_i->free_nid_bitmap)
3289 		return -ENOMEM;
3290 
3291 	for (i = 0; i < nm_i->nat_blocks; i++) {
3292 		nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3293 			f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3294 		if (!nm_i->free_nid_bitmap[i])
3295 			return -ENOMEM;
3296 	}
3297 
3298 	nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3299 								GFP_KERNEL);
3300 	if (!nm_i->nat_block_bitmap)
3301 		return -ENOMEM;
3302 
3303 	nm_i->free_nid_count =
3304 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3305 					      nm_i->nat_blocks),
3306 			      GFP_KERNEL);
3307 	if (!nm_i->free_nid_count)
3308 		return -ENOMEM;
3309 	return 0;
3310 }
3311 
3312 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3313 {
3314 	int err;
3315 
3316 	sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3317 							GFP_KERNEL);
3318 	if (!sbi->nm_info)
3319 		return -ENOMEM;
3320 
3321 	err = init_node_manager(sbi);
3322 	if (err)
3323 		return err;
3324 
3325 	err = init_free_nid_cache(sbi);
3326 	if (err)
3327 		return err;
3328 
3329 	/* load free nid status from nat_bits table */
3330 	load_free_nid_bitmap(sbi);
3331 
3332 	return f2fs_build_free_nids(sbi, true, true);
3333 }
3334 
3335 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3336 {
3337 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3338 	struct free_nid *i, *next_i;
3339 	void *vec[NAT_VEC_SIZE];
3340 	struct nat_entry **natvec = (struct nat_entry **)vec;
3341 	struct nat_entry_set **setvec = (struct nat_entry_set **)vec;
3342 	nid_t nid = 0;
3343 	unsigned int found;
3344 
3345 	if (!nm_i)
3346 		return;
3347 
3348 	/* destroy free nid list */
3349 	spin_lock(&nm_i->nid_list_lock);
3350 	list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3351 		__remove_free_nid(sbi, i, FREE_NID);
3352 		spin_unlock(&nm_i->nid_list_lock);
3353 		kmem_cache_free(free_nid_slab, i);
3354 		spin_lock(&nm_i->nid_list_lock);
3355 	}
3356 	f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3357 	f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3358 	f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3359 	spin_unlock(&nm_i->nid_list_lock);
3360 
3361 	/* destroy nat cache */
3362 	f2fs_down_write(&nm_i->nat_tree_lock);
3363 	while ((found = __gang_lookup_nat_cache(nm_i,
3364 					nid, NAT_VEC_SIZE, natvec))) {
3365 		unsigned idx;
3366 
3367 		nid = nat_get_nid(natvec[found - 1]) + 1;
3368 		for (idx = 0; idx < found; idx++) {
3369 			spin_lock(&nm_i->nat_list_lock);
3370 			list_del(&natvec[idx]->list);
3371 			spin_unlock(&nm_i->nat_list_lock);
3372 
3373 			__del_from_nat_cache(nm_i, natvec[idx]);
3374 		}
3375 	}
3376 	f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3377 
3378 	/* destroy nat set cache */
3379 	nid = 0;
3380 	memset(vec, 0, sizeof(void *) * NAT_VEC_SIZE);
3381 	while ((found = __gang_lookup_nat_set(nm_i,
3382 					nid, NAT_VEC_SIZE, setvec))) {
3383 		unsigned idx;
3384 
3385 		nid = setvec[found - 1]->set + 1;
3386 		for (idx = 0; idx < found; idx++) {
3387 			/* entry_cnt is not zero, when cp_error was occurred */
3388 			f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3389 			radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3390 			kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3391 		}
3392 	}
3393 	f2fs_up_write(&nm_i->nat_tree_lock);
3394 
3395 	kvfree(nm_i->nat_block_bitmap);
3396 	if (nm_i->free_nid_bitmap) {
3397 		int i;
3398 
3399 		for (i = 0; i < nm_i->nat_blocks; i++)
3400 			kvfree(nm_i->free_nid_bitmap[i]);
3401 		kvfree(nm_i->free_nid_bitmap);
3402 	}
3403 	kvfree(nm_i->free_nid_count);
3404 
3405 	kvfree(nm_i->nat_bitmap);
3406 	kvfree(nm_i->nat_bits);
3407 #ifdef CONFIG_F2FS_CHECK_FS
3408 	kvfree(nm_i->nat_bitmap_mir);
3409 #endif
3410 	sbi->nm_info = NULL;
3411 	kfree(nm_i);
3412 }
3413 
3414 int __init f2fs_create_node_manager_caches(void)
3415 {
3416 	nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3417 			sizeof(struct nat_entry));
3418 	if (!nat_entry_slab)
3419 		goto fail;
3420 
3421 	free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3422 			sizeof(struct free_nid));
3423 	if (!free_nid_slab)
3424 		goto destroy_nat_entry;
3425 
3426 	nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3427 			sizeof(struct nat_entry_set));
3428 	if (!nat_entry_set_slab)
3429 		goto destroy_free_nid;
3430 
3431 	fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3432 			sizeof(struct fsync_node_entry));
3433 	if (!fsync_node_entry_slab)
3434 		goto destroy_nat_entry_set;
3435 	return 0;
3436 
3437 destroy_nat_entry_set:
3438 	kmem_cache_destroy(nat_entry_set_slab);
3439 destroy_free_nid:
3440 	kmem_cache_destroy(free_nid_slab);
3441 destroy_nat_entry:
3442 	kmem_cache_destroy(nat_entry_slab);
3443 fail:
3444 	return -ENOMEM;
3445 }
3446 
3447 void f2fs_destroy_node_manager_caches(void)
3448 {
3449 	kmem_cache_destroy(fsync_node_entry_slab);
3450 	kmem_cache_destroy(nat_entry_set_slab);
3451 	kmem_cache_destroy(free_nid_slab);
3452 	kmem_cache_destroy(nat_entry_slab);
3453 }
3454