xref: /openbmc/linux/fs/f2fs/node.c (revision 62eab49f)
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/backing-dev.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 <trace/events/f2fs.h>
21 
22 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
23 
24 static struct kmem_cache *nat_entry_slab;
25 static struct kmem_cache *free_nid_slab;
26 static struct kmem_cache *nat_entry_set_slab;
27 static struct kmem_cache *fsync_node_entry_slab;
28 
29 /*
30  * Check whether the given nid is within node id range.
31  */
32 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
33 {
34 	if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
35 		set_sbi_flag(sbi, SBI_NEED_FSCK);
36 		f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
37 			  __func__, nid);
38 		return -EFSCORRUPTED;
39 	}
40 	return 0;
41 }
42 
43 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
44 {
45 	struct f2fs_nm_info *nm_i = NM_I(sbi);
46 	struct sysinfo val;
47 	unsigned long avail_ram;
48 	unsigned long mem_size = 0;
49 	bool res = false;
50 
51 	si_meminfo(&val);
52 
53 	/* only uses low memory */
54 	avail_ram = val.totalram - val.totalhigh;
55 
56 	/*
57 	 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
58 	 */
59 	if (type == FREE_NIDS) {
60 		mem_size = (nm_i->nid_cnt[FREE_NID] *
61 				sizeof(struct free_nid)) >> PAGE_SHIFT;
62 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
63 	} else if (type == NAT_ENTRIES) {
64 		mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
65 				sizeof(struct nat_entry)) >> PAGE_SHIFT;
66 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
67 		if (excess_cached_nats(sbi))
68 			res = false;
69 	} else if (type == DIRTY_DENTS) {
70 		if (sbi->sb->s_bdi->wb.dirty_exceeded)
71 			return false;
72 		mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
73 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
74 	} else if (type == INO_ENTRIES) {
75 		int i;
76 
77 		for (i = 0; i < MAX_INO_ENTRY; i++)
78 			mem_size += sbi->im[i].ino_num *
79 						sizeof(struct ino_entry);
80 		mem_size >>= PAGE_SHIFT;
81 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
82 	} else if (type == EXTENT_CACHE) {
83 		mem_size = (atomic_read(&sbi->total_ext_tree) *
84 				sizeof(struct extent_tree) +
85 				atomic_read(&sbi->total_ext_node) *
86 				sizeof(struct extent_node)) >> PAGE_SHIFT;
87 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
88 	} else if (type == INMEM_PAGES) {
89 		/* it allows 20% / total_ram for inmemory pages */
90 		mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
91 		res = mem_size < (val.totalram / 5);
92 	} else {
93 		if (!sbi->sb->s_bdi->wb.dirty_exceeded)
94 			return true;
95 	}
96 	return res;
97 }
98 
99 static void clear_node_page_dirty(struct page *page)
100 {
101 	if (PageDirty(page)) {
102 		f2fs_clear_page_cache_dirty_tag(page);
103 		clear_page_dirty_for_io(page);
104 		dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
105 	}
106 	ClearPageUptodate(page);
107 }
108 
109 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
110 {
111 	return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
112 }
113 
114 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
115 {
116 	struct page *src_page;
117 	struct page *dst_page;
118 	pgoff_t dst_off;
119 	void *src_addr;
120 	void *dst_addr;
121 	struct f2fs_nm_info *nm_i = NM_I(sbi);
122 
123 	dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
124 
125 	/* get current nat block page with lock */
126 	src_page = get_current_nat_page(sbi, nid);
127 	if (IS_ERR(src_page))
128 		return src_page;
129 	dst_page = f2fs_grab_meta_page(sbi, dst_off);
130 	f2fs_bug_on(sbi, PageDirty(src_page));
131 
132 	src_addr = page_address(src_page);
133 	dst_addr = page_address(dst_page);
134 	memcpy(dst_addr, src_addr, PAGE_SIZE);
135 	set_page_dirty(dst_page);
136 	f2fs_put_page(src_page, 1);
137 
138 	set_to_next_nat(nm_i, nid);
139 
140 	return dst_page;
141 }
142 
143 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
144 {
145 	struct nat_entry *new;
146 
147 	if (no_fail)
148 		new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
149 	else
150 		new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
151 	if (new) {
152 		nat_set_nid(new, nid);
153 		nat_reset_flag(new);
154 	}
155 	return new;
156 }
157 
158 static void __free_nat_entry(struct nat_entry *e)
159 {
160 	kmem_cache_free(nat_entry_slab, e);
161 }
162 
163 /* must be locked by nat_tree_lock */
164 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
165 	struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
166 {
167 	if (no_fail)
168 		f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
169 	else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
170 		return NULL;
171 
172 	if (raw_ne)
173 		node_info_from_raw_nat(&ne->ni, raw_ne);
174 
175 	spin_lock(&nm_i->nat_list_lock);
176 	list_add_tail(&ne->list, &nm_i->nat_entries);
177 	spin_unlock(&nm_i->nat_list_lock);
178 
179 	nm_i->nat_cnt[TOTAL_NAT]++;
180 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
181 	return ne;
182 }
183 
184 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
185 {
186 	struct nat_entry *ne;
187 
188 	ne = radix_tree_lookup(&nm_i->nat_root, n);
189 
190 	/* for recent accessed nat entry, move it to tail of lru list */
191 	if (ne && !get_nat_flag(ne, IS_DIRTY)) {
192 		spin_lock(&nm_i->nat_list_lock);
193 		if (!list_empty(&ne->list))
194 			list_move_tail(&ne->list, &nm_i->nat_entries);
195 		spin_unlock(&nm_i->nat_list_lock);
196 	}
197 
198 	return ne;
199 }
200 
201 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
202 		nid_t start, unsigned int nr, struct nat_entry **ep)
203 {
204 	return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
205 }
206 
207 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
208 {
209 	radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
210 	nm_i->nat_cnt[TOTAL_NAT]--;
211 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
212 	__free_nat_entry(e);
213 }
214 
215 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
216 							struct nat_entry *ne)
217 {
218 	nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
219 	struct nat_entry_set *head;
220 
221 	head = radix_tree_lookup(&nm_i->nat_set_root, set);
222 	if (!head) {
223 		head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
224 
225 		INIT_LIST_HEAD(&head->entry_list);
226 		INIT_LIST_HEAD(&head->set_list);
227 		head->set = set;
228 		head->entry_cnt = 0;
229 		f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
230 	}
231 	return head;
232 }
233 
234 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
235 						struct nat_entry *ne)
236 {
237 	struct nat_entry_set *head;
238 	bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
239 
240 	if (!new_ne)
241 		head = __grab_nat_entry_set(nm_i, ne);
242 
243 	/*
244 	 * update entry_cnt in below condition:
245 	 * 1. update NEW_ADDR to valid block address;
246 	 * 2. update old block address to new one;
247 	 */
248 	if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
249 				!get_nat_flag(ne, IS_DIRTY)))
250 		head->entry_cnt++;
251 
252 	set_nat_flag(ne, IS_PREALLOC, new_ne);
253 
254 	if (get_nat_flag(ne, IS_DIRTY))
255 		goto refresh_list;
256 
257 	nm_i->nat_cnt[DIRTY_NAT]++;
258 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
259 	set_nat_flag(ne, IS_DIRTY, true);
260 refresh_list:
261 	spin_lock(&nm_i->nat_list_lock);
262 	if (new_ne)
263 		list_del_init(&ne->list);
264 	else
265 		list_move_tail(&ne->list, &head->entry_list);
266 	spin_unlock(&nm_i->nat_list_lock);
267 }
268 
269 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
270 		struct nat_entry_set *set, struct nat_entry *ne)
271 {
272 	spin_lock(&nm_i->nat_list_lock);
273 	list_move_tail(&ne->list, &nm_i->nat_entries);
274 	spin_unlock(&nm_i->nat_list_lock);
275 
276 	set_nat_flag(ne, IS_DIRTY, false);
277 	set->entry_cnt--;
278 	nm_i->nat_cnt[DIRTY_NAT]--;
279 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
280 }
281 
282 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
283 		nid_t start, unsigned int nr, struct nat_entry_set **ep)
284 {
285 	return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
286 							start, nr);
287 }
288 
289 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
290 {
291 	return NODE_MAPPING(sbi) == page->mapping &&
292 			IS_DNODE(page) && is_cold_node(page);
293 }
294 
295 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
296 {
297 	spin_lock_init(&sbi->fsync_node_lock);
298 	INIT_LIST_HEAD(&sbi->fsync_node_list);
299 	sbi->fsync_seg_id = 0;
300 	sbi->fsync_node_num = 0;
301 }
302 
303 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
304 							struct page *page)
305 {
306 	struct fsync_node_entry *fn;
307 	unsigned long flags;
308 	unsigned int seq_id;
309 
310 	fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS);
311 
312 	get_page(page);
313 	fn->page = page;
314 	INIT_LIST_HEAD(&fn->list);
315 
316 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
317 	list_add_tail(&fn->list, &sbi->fsync_node_list);
318 	fn->seq_id = sbi->fsync_seg_id++;
319 	seq_id = fn->seq_id;
320 	sbi->fsync_node_num++;
321 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
322 
323 	return seq_id;
324 }
325 
326 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
327 {
328 	struct fsync_node_entry *fn;
329 	unsigned long flags;
330 
331 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
332 	list_for_each_entry(fn, &sbi->fsync_node_list, list) {
333 		if (fn->page == page) {
334 			list_del(&fn->list);
335 			sbi->fsync_node_num--;
336 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
337 			kmem_cache_free(fsync_node_entry_slab, fn);
338 			put_page(page);
339 			return;
340 		}
341 	}
342 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
343 	f2fs_bug_on(sbi, 1);
344 }
345 
346 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
347 {
348 	unsigned long flags;
349 
350 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
351 	sbi->fsync_seg_id = 0;
352 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
353 }
354 
355 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
356 {
357 	struct f2fs_nm_info *nm_i = NM_I(sbi);
358 	struct nat_entry *e;
359 	bool need = false;
360 
361 	down_read(&nm_i->nat_tree_lock);
362 	e = __lookup_nat_cache(nm_i, nid);
363 	if (e) {
364 		if (!get_nat_flag(e, IS_CHECKPOINTED) &&
365 				!get_nat_flag(e, HAS_FSYNCED_INODE))
366 			need = true;
367 	}
368 	up_read(&nm_i->nat_tree_lock);
369 	return need;
370 }
371 
372 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
373 {
374 	struct f2fs_nm_info *nm_i = NM_I(sbi);
375 	struct nat_entry *e;
376 	bool is_cp = true;
377 
378 	down_read(&nm_i->nat_tree_lock);
379 	e = __lookup_nat_cache(nm_i, nid);
380 	if (e && !get_nat_flag(e, IS_CHECKPOINTED))
381 		is_cp = false;
382 	up_read(&nm_i->nat_tree_lock);
383 	return is_cp;
384 }
385 
386 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
387 {
388 	struct f2fs_nm_info *nm_i = NM_I(sbi);
389 	struct nat_entry *e;
390 	bool need_update = true;
391 
392 	down_read(&nm_i->nat_tree_lock);
393 	e = __lookup_nat_cache(nm_i, ino);
394 	if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
395 			(get_nat_flag(e, IS_CHECKPOINTED) ||
396 			 get_nat_flag(e, HAS_FSYNCED_INODE)))
397 		need_update = false;
398 	up_read(&nm_i->nat_tree_lock);
399 	return need_update;
400 }
401 
402 /* must be locked by nat_tree_lock */
403 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
404 						struct f2fs_nat_entry *ne)
405 {
406 	struct f2fs_nm_info *nm_i = NM_I(sbi);
407 	struct nat_entry *new, *e;
408 
409 	new = __alloc_nat_entry(nid, false);
410 	if (!new)
411 		return;
412 
413 	down_write(&nm_i->nat_tree_lock);
414 	e = __lookup_nat_cache(nm_i, nid);
415 	if (!e)
416 		e = __init_nat_entry(nm_i, new, ne, false);
417 	else
418 		f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
419 				nat_get_blkaddr(e) !=
420 					le32_to_cpu(ne->block_addr) ||
421 				nat_get_version(e) != ne->version);
422 	up_write(&nm_i->nat_tree_lock);
423 	if (e != new)
424 		__free_nat_entry(new);
425 }
426 
427 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
428 			block_t new_blkaddr, bool fsync_done)
429 {
430 	struct f2fs_nm_info *nm_i = NM_I(sbi);
431 	struct nat_entry *e;
432 	struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
433 
434 	down_write(&nm_i->nat_tree_lock);
435 	e = __lookup_nat_cache(nm_i, ni->nid);
436 	if (!e) {
437 		e = __init_nat_entry(nm_i, new, NULL, true);
438 		copy_node_info(&e->ni, ni);
439 		f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
440 	} else if (new_blkaddr == NEW_ADDR) {
441 		/*
442 		 * when nid is reallocated,
443 		 * previous nat entry can be remained in nat cache.
444 		 * So, reinitialize it with new information.
445 		 */
446 		copy_node_info(&e->ni, ni);
447 		f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
448 	}
449 	/* let's free early to reduce memory consumption */
450 	if (e != new)
451 		__free_nat_entry(new);
452 
453 	/* sanity check */
454 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
455 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
456 			new_blkaddr == NULL_ADDR);
457 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
458 			new_blkaddr == NEW_ADDR);
459 	f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
460 			new_blkaddr == NEW_ADDR);
461 
462 	/* increment version no as node is removed */
463 	if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
464 		unsigned char version = nat_get_version(e);
465 		nat_set_version(e, inc_node_version(version));
466 	}
467 
468 	/* change address */
469 	nat_set_blkaddr(e, new_blkaddr);
470 	if (!__is_valid_data_blkaddr(new_blkaddr))
471 		set_nat_flag(e, IS_CHECKPOINTED, false);
472 	__set_nat_cache_dirty(nm_i, e);
473 
474 	/* update fsync_mark if its inode nat entry is still alive */
475 	if (ni->nid != ni->ino)
476 		e = __lookup_nat_cache(nm_i, ni->ino);
477 	if (e) {
478 		if (fsync_done && ni->nid == ni->ino)
479 			set_nat_flag(e, HAS_FSYNCED_INODE, true);
480 		set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
481 	}
482 	up_write(&nm_i->nat_tree_lock);
483 }
484 
485 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
486 {
487 	struct f2fs_nm_info *nm_i = NM_I(sbi);
488 	int nr = nr_shrink;
489 
490 	if (!down_write_trylock(&nm_i->nat_tree_lock))
491 		return 0;
492 
493 	spin_lock(&nm_i->nat_list_lock);
494 	while (nr_shrink) {
495 		struct nat_entry *ne;
496 
497 		if (list_empty(&nm_i->nat_entries))
498 			break;
499 
500 		ne = list_first_entry(&nm_i->nat_entries,
501 					struct nat_entry, list);
502 		list_del(&ne->list);
503 		spin_unlock(&nm_i->nat_list_lock);
504 
505 		__del_from_nat_cache(nm_i, ne);
506 		nr_shrink--;
507 
508 		spin_lock(&nm_i->nat_list_lock);
509 	}
510 	spin_unlock(&nm_i->nat_list_lock);
511 
512 	up_write(&nm_i->nat_tree_lock);
513 	return nr - nr_shrink;
514 }
515 
516 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
517 						struct node_info *ni)
518 {
519 	struct f2fs_nm_info *nm_i = NM_I(sbi);
520 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
521 	struct f2fs_journal *journal = curseg->journal;
522 	nid_t start_nid = START_NID(nid);
523 	struct f2fs_nat_block *nat_blk;
524 	struct page *page = NULL;
525 	struct f2fs_nat_entry ne;
526 	struct nat_entry *e;
527 	pgoff_t index;
528 	block_t blkaddr;
529 	int i;
530 
531 	ni->nid = nid;
532 
533 	/* Check nat cache */
534 	down_read(&nm_i->nat_tree_lock);
535 	e = __lookup_nat_cache(nm_i, nid);
536 	if (e) {
537 		ni->ino = nat_get_ino(e);
538 		ni->blk_addr = nat_get_blkaddr(e);
539 		ni->version = nat_get_version(e);
540 		up_read(&nm_i->nat_tree_lock);
541 		return 0;
542 	}
543 
544 	memset(&ne, 0, sizeof(struct f2fs_nat_entry));
545 
546 	/* Check current segment summary */
547 	down_read(&curseg->journal_rwsem);
548 	i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
549 	if (i >= 0) {
550 		ne = nat_in_journal(journal, i);
551 		node_info_from_raw_nat(ni, &ne);
552 	}
553 	up_read(&curseg->journal_rwsem);
554 	if (i >= 0) {
555 		up_read(&nm_i->nat_tree_lock);
556 		goto cache;
557 	}
558 
559 	/* Fill node_info from nat page */
560 	index = current_nat_addr(sbi, nid);
561 	up_read(&nm_i->nat_tree_lock);
562 
563 	page = f2fs_get_meta_page(sbi, index);
564 	if (IS_ERR(page))
565 		return PTR_ERR(page);
566 
567 	nat_blk = (struct f2fs_nat_block *)page_address(page);
568 	ne = nat_blk->entries[nid - start_nid];
569 	node_info_from_raw_nat(ni, &ne);
570 	f2fs_put_page(page, 1);
571 cache:
572 	blkaddr = le32_to_cpu(ne.block_addr);
573 	if (__is_valid_data_blkaddr(blkaddr) &&
574 		!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
575 		return -EFAULT;
576 
577 	/* cache nat entry */
578 	cache_nat_entry(sbi, nid, &ne);
579 	return 0;
580 }
581 
582 /*
583  * readahead MAX_RA_NODE number of node pages.
584  */
585 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
586 {
587 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
588 	struct blk_plug plug;
589 	int i, end;
590 	nid_t nid;
591 
592 	blk_start_plug(&plug);
593 
594 	/* Then, try readahead for siblings of the desired node */
595 	end = start + n;
596 	end = min(end, NIDS_PER_BLOCK);
597 	for (i = start; i < end; i++) {
598 		nid = get_nid(parent, i, false);
599 		f2fs_ra_node_page(sbi, nid);
600 	}
601 
602 	blk_finish_plug(&plug);
603 }
604 
605 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
606 {
607 	const long direct_index = ADDRS_PER_INODE(dn->inode);
608 	const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
609 	const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
610 	unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
611 	int cur_level = dn->cur_level;
612 	int max_level = dn->max_level;
613 	pgoff_t base = 0;
614 
615 	if (!dn->max_level)
616 		return pgofs + 1;
617 
618 	while (max_level-- > cur_level)
619 		skipped_unit *= NIDS_PER_BLOCK;
620 
621 	switch (dn->max_level) {
622 	case 3:
623 		base += 2 * indirect_blks;
624 		fallthrough;
625 	case 2:
626 		base += 2 * direct_blks;
627 		fallthrough;
628 	case 1:
629 		base += direct_index;
630 		break;
631 	default:
632 		f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
633 	}
634 
635 	return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
636 }
637 
638 /*
639  * The maximum depth is four.
640  * Offset[0] will have raw inode offset.
641  */
642 static int get_node_path(struct inode *inode, long block,
643 				int offset[4], unsigned int noffset[4])
644 {
645 	const long direct_index = ADDRS_PER_INODE(inode);
646 	const long direct_blks = ADDRS_PER_BLOCK(inode);
647 	const long dptrs_per_blk = NIDS_PER_BLOCK;
648 	const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
649 	const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
650 	int n = 0;
651 	int level = 0;
652 
653 	noffset[0] = 0;
654 
655 	if (block < direct_index) {
656 		offset[n] = block;
657 		goto got;
658 	}
659 	block -= direct_index;
660 	if (block < direct_blks) {
661 		offset[n++] = NODE_DIR1_BLOCK;
662 		noffset[n] = 1;
663 		offset[n] = block;
664 		level = 1;
665 		goto got;
666 	}
667 	block -= direct_blks;
668 	if (block < direct_blks) {
669 		offset[n++] = NODE_DIR2_BLOCK;
670 		noffset[n] = 2;
671 		offset[n] = block;
672 		level = 1;
673 		goto got;
674 	}
675 	block -= direct_blks;
676 	if (block < indirect_blks) {
677 		offset[n++] = NODE_IND1_BLOCK;
678 		noffset[n] = 3;
679 		offset[n++] = block / direct_blks;
680 		noffset[n] = 4 + offset[n - 1];
681 		offset[n] = block % direct_blks;
682 		level = 2;
683 		goto got;
684 	}
685 	block -= indirect_blks;
686 	if (block < indirect_blks) {
687 		offset[n++] = NODE_IND2_BLOCK;
688 		noffset[n] = 4 + dptrs_per_blk;
689 		offset[n++] = block / direct_blks;
690 		noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
691 		offset[n] = block % direct_blks;
692 		level = 2;
693 		goto got;
694 	}
695 	block -= indirect_blks;
696 	if (block < dindirect_blks) {
697 		offset[n++] = NODE_DIND_BLOCK;
698 		noffset[n] = 5 + (dptrs_per_blk * 2);
699 		offset[n++] = block / indirect_blks;
700 		noffset[n] = 6 + (dptrs_per_blk * 2) +
701 			      offset[n - 1] * (dptrs_per_blk + 1);
702 		offset[n++] = (block / direct_blks) % dptrs_per_blk;
703 		noffset[n] = 7 + (dptrs_per_blk * 2) +
704 			      offset[n - 2] * (dptrs_per_blk + 1) +
705 			      offset[n - 1];
706 		offset[n] = block % direct_blks;
707 		level = 3;
708 		goto got;
709 	} else {
710 		return -E2BIG;
711 	}
712 got:
713 	return level;
714 }
715 
716 /*
717  * Caller should call f2fs_put_dnode(dn).
718  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
719  * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
720  */
721 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
722 {
723 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
724 	struct page *npage[4];
725 	struct page *parent = NULL;
726 	int offset[4];
727 	unsigned int noffset[4];
728 	nid_t nids[4];
729 	int level, i = 0;
730 	int err = 0;
731 
732 	level = get_node_path(dn->inode, index, offset, noffset);
733 	if (level < 0)
734 		return level;
735 
736 	nids[0] = dn->inode->i_ino;
737 	npage[0] = dn->inode_page;
738 
739 	if (!npage[0]) {
740 		npage[0] = f2fs_get_node_page(sbi, nids[0]);
741 		if (IS_ERR(npage[0]))
742 			return PTR_ERR(npage[0]);
743 	}
744 
745 	/* if inline_data is set, should not report any block indices */
746 	if (f2fs_has_inline_data(dn->inode) && index) {
747 		err = -ENOENT;
748 		f2fs_put_page(npage[0], 1);
749 		goto release_out;
750 	}
751 
752 	parent = npage[0];
753 	if (level != 0)
754 		nids[1] = get_nid(parent, offset[0], true);
755 	dn->inode_page = npage[0];
756 	dn->inode_page_locked = true;
757 
758 	/* get indirect or direct nodes */
759 	for (i = 1; i <= level; i++) {
760 		bool done = false;
761 
762 		if (!nids[i] && mode == ALLOC_NODE) {
763 			/* alloc new node */
764 			if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
765 				err = -ENOSPC;
766 				goto release_pages;
767 			}
768 
769 			dn->nid = nids[i];
770 			npage[i] = f2fs_new_node_page(dn, noffset[i]);
771 			if (IS_ERR(npage[i])) {
772 				f2fs_alloc_nid_failed(sbi, nids[i]);
773 				err = PTR_ERR(npage[i]);
774 				goto release_pages;
775 			}
776 
777 			set_nid(parent, offset[i - 1], nids[i], i == 1);
778 			f2fs_alloc_nid_done(sbi, nids[i]);
779 			done = true;
780 		} else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
781 			npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
782 			if (IS_ERR(npage[i])) {
783 				err = PTR_ERR(npage[i]);
784 				goto release_pages;
785 			}
786 			done = true;
787 		}
788 		if (i == 1) {
789 			dn->inode_page_locked = false;
790 			unlock_page(parent);
791 		} else {
792 			f2fs_put_page(parent, 1);
793 		}
794 
795 		if (!done) {
796 			npage[i] = f2fs_get_node_page(sbi, nids[i]);
797 			if (IS_ERR(npage[i])) {
798 				err = PTR_ERR(npage[i]);
799 				f2fs_put_page(npage[0], 0);
800 				goto release_out;
801 			}
802 		}
803 		if (i < level) {
804 			parent = npage[i];
805 			nids[i + 1] = get_nid(parent, offset[i], false);
806 		}
807 	}
808 	dn->nid = nids[level];
809 	dn->ofs_in_node = offset[level];
810 	dn->node_page = npage[level];
811 	dn->data_blkaddr = f2fs_data_blkaddr(dn);
812 	return 0;
813 
814 release_pages:
815 	f2fs_put_page(parent, 1);
816 	if (i > 1)
817 		f2fs_put_page(npage[0], 0);
818 release_out:
819 	dn->inode_page = NULL;
820 	dn->node_page = NULL;
821 	if (err == -ENOENT) {
822 		dn->cur_level = i;
823 		dn->max_level = level;
824 		dn->ofs_in_node = offset[level];
825 	}
826 	return err;
827 }
828 
829 static int truncate_node(struct dnode_of_data *dn)
830 {
831 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
832 	struct node_info ni;
833 	int err;
834 	pgoff_t index;
835 
836 	err = f2fs_get_node_info(sbi, dn->nid, &ni);
837 	if (err)
838 		return err;
839 
840 	/* Deallocate node address */
841 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
842 	dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
843 	set_node_addr(sbi, &ni, NULL_ADDR, false);
844 
845 	if (dn->nid == dn->inode->i_ino) {
846 		f2fs_remove_orphan_inode(sbi, dn->nid);
847 		dec_valid_inode_count(sbi);
848 		f2fs_inode_synced(dn->inode);
849 	}
850 
851 	clear_node_page_dirty(dn->node_page);
852 	set_sbi_flag(sbi, SBI_IS_DIRTY);
853 
854 	index = dn->node_page->index;
855 	f2fs_put_page(dn->node_page, 1);
856 
857 	invalidate_mapping_pages(NODE_MAPPING(sbi),
858 			index, index);
859 
860 	dn->node_page = NULL;
861 	trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
862 
863 	return 0;
864 }
865 
866 static int truncate_dnode(struct dnode_of_data *dn)
867 {
868 	struct page *page;
869 	int err;
870 
871 	if (dn->nid == 0)
872 		return 1;
873 
874 	/* get direct node */
875 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
876 	if (PTR_ERR(page) == -ENOENT)
877 		return 1;
878 	else if (IS_ERR(page))
879 		return PTR_ERR(page);
880 
881 	/* Make dnode_of_data for parameter */
882 	dn->node_page = page;
883 	dn->ofs_in_node = 0;
884 	f2fs_truncate_data_blocks(dn);
885 	err = truncate_node(dn);
886 	if (err)
887 		return err;
888 
889 	return 1;
890 }
891 
892 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
893 						int ofs, int depth)
894 {
895 	struct dnode_of_data rdn = *dn;
896 	struct page *page;
897 	struct f2fs_node *rn;
898 	nid_t child_nid;
899 	unsigned int child_nofs;
900 	int freed = 0;
901 	int i, ret;
902 
903 	if (dn->nid == 0)
904 		return NIDS_PER_BLOCK + 1;
905 
906 	trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
907 
908 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
909 	if (IS_ERR(page)) {
910 		trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
911 		return PTR_ERR(page);
912 	}
913 
914 	f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
915 
916 	rn = F2FS_NODE(page);
917 	if (depth < 3) {
918 		for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
919 			child_nid = le32_to_cpu(rn->in.nid[i]);
920 			if (child_nid == 0)
921 				continue;
922 			rdn.nid = child_nid;
923 			ret = truncate_dnode(&rdn);
924 			if (ret < 0)
925 				goto out_err;
926 			if (set_nid(page, i, 0, false))
927 				dn->node_changed = true;
928 		}
929 	} else {
930 		child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
931 		for (i = ofs; i < NIDS_PER_BLOCK; i++) {
932 			child_nid = le32_to_cpu(rn->in.nid[i]);
933 			if (child_nid == 0) {
934 				child_nofs += NIDS_PER_BLOCK + 1;
935 				continue;
936 			}
937 			rdn.nid = child_nid;
938 			ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
939 			if (ret == (NIDS_PER_BLOCK + 1)) {
940 				if (set_nid(page, i, 0, false))
941 					dn->node_changed = true;
942 				child_nofs += ret;
943 			} else if (ret < 0 && ret != -ENOENT) {
944 				goto out_err;
945 			}
946 		}
947 		freed = child_nofs;
948 	}
949 
950 	if (!ofs) {
951 		/* remove current indirect node */
952 		dn->node_page = page;
953 		ret = truncate_node(dn);
954 		if (ret)
955 			goto out_err;
956 		freed++;
957 	} else {
958 		f2fs_put_page(page, 1);
959 	}
960 	trace_f2fs_truncate_nodes_exit(dn->inode, freed);
961 	return freed;
962 
963 out_err:
964 	f2fs_put_page(page, 1);
965 	trace_f2fs_truncate_nodes_exit(dn->inode, ret);
966 	return ret;
967 }
968 
969 static int truncate_partial_nodes(struct dnode_of_data *dn,
970 			struct f2fs_inode *ri, int *offset, int depth)
971 {
972 	struct page *pages[2];
973 	nid_t nid[3];
974 	nid_t child_nid;
975 	int err = 0;
976 	int i;
977 	int idx = depth - 2;
978 
979 	nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
980 	if (!nid[0])
981 		return 0;
982 
983 	/* get indirect nodes in the path */
984 	for (i = 0; i < idx + 1; i++) {
985 		/* reference count'll be increased */
986 		pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
987 		if (IS_ERR(pages[i])) {
988 			err = PTR_ERR(pages[i]);
989 			idx = i - 1;
990 			goto fail;
991 		}
992 		nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
993 	}
994 
995 	f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
996 
997 	/* free direct nodes linked to a partial indirect node */
998 	for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
999 		child_nid = get_nid(pages[idx], i, false);
1000 		if (!child_nid)
1001 			continue;
1002 		dn->nid = child_nid;
1003 		err = truncate_dnode(dn);
1004 		if (err < 0)
1005 			goto fail;
1006 		if (set_nid(pages[idx], i, 0, false))
1007 			dn->node_changed = true;
1008 	}
1009 
1010 	if (offset[idx + 1] == 0) {
1011 		dn->node_page = pages[idx];
1012 		dn->nid = nid[idx];
1013 		err = truncate_node(dn);
1014 		if (err)
1015 			goto fail;
1016 	} else {
1017 		f2fs_put_page(pages[idx], 1);
1018 	}
1019 	offset[idx]++;
1020 	offset[idx + 1] = 0;
1021 	idx--;
1022 fail:
1023 	for (i = idx; i >= 0; i--)
1024 		f2fs_put_page(pages[i], 1);
1025 
1026 	trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1027 
1028 	return err;
1029 }
1030 
1031 /*
1032  * All the block addresses of data and nodes should be nullified.
1033  */
1034 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1035 {
1036 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1037 	int err = 0, cont = 1;
1038 	int level, offset[4], noffset[4];
1039 	unsigned int nofs = 0;
1040 	struct f2fs_inode *ri;
1041 	struct dnode_of_data dn;
1042 	struct page *page;
1043 
1044 	trace_f2fs_truncate_inode_blocks_enter(inode, from);
1045 
1046 	level = get_node_path(inode, from, offset, noffset);
1047 	if (level < 0) {
1048 		trace_f2fs_truncate_inode_blocks_exit(inode, level);
1049 		return level;
1050 	}
1051 
1052 	page = f2fs_get_node_page(sbi, inode->i_ino);
1053 	if (IS_ERR(page)) {
1054 		trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1055 		return PTR_ERR(page);
1056 	}
1057 
1058 	set_new_dnode(&dn, inode, page, NULL, 0);
1059 	unlock_page(page);
1060 
1061 	ri = F2FS_INODE(page);
1062 	switch (level) {
1063 	case 0:
1064 	case 1:
1065 		nofs = noffset[1];
1066 		break;
1067 	case 2:
1068 		nofs = noffset[1];
1069 		if (!offset[level - 1])
1070 			goto skip_partial;
1071 		err = truncate_partial_nodes(&dn, ri, offset, level);
1072 		if (err < 0 && err != -ENOENT)
1073 			goto fail;
1074 		nofs += 1 + NIDS_PER_BLOCK;
1075 		break;
1076 	case 3:
1077 		nofs = 5 + 2 * NIDS_PER_BLOCK;
1078 		if (!offset[level - 1])
1079 			goto skip_partial;
1080 		err = truncate_partial_nodes(&dn, ri, offset, level);
1081 		if (err < 0 && err != -ENOENT)
1082 			goto fail;
1083 		break;
1084 	default:
1085 		BUG();
1086 	}
1087 
1088 skip_partial:
1089 	while (cont) {
1090 		dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1091 		switch (offset[0]) {
1092 		case NODE_DIR1_BLOCK:
1093 		case NODE_DIR2_BLOCK:
1094 			err = truncate_dnode(&dn);
1095 			break;
1096 
1097 		case NODE_IND1_BLOCK:
1098 		case NODE_IND2_BLOCK:
1099 			err = truncate_nodes(&dn, nofs, offset[1], 2);
1100 			break;
1101 
1102 		case NODE_DIND_BLOCK:
1103 			err = truncate_nodes(&dn, nofs, offset[1], 3);
1104 			cont = 0;
1105 			break;
1106 
1107 		default:
1108 			BUG();
1109 		}
1110 		if (err < 0 && err != -ENOENT)
1111 			goto fail;
1112 		if (offset[1] == 0 &&
1113 				ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1114 			lock_page(page);
1115 			BUG_ON(page->mapping != NODE_MAPPING(sbi));
1116 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1117 			ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1118 			set_page_dirty(page);
1119 			unlock_page(page);
1120 		}
1121 		offset[1] = 0;
1122 		offset[0]++;
1123 		nofs += err;
1124 	}
1125 fail:
1126 	f2fs_put_page(page, 0);
1127 	trace_f2fs_truncate_inode_blocks_exit(inode, err);
1128 	return err > 0 ? 0 : err;
1129 }
1130 
1131 /* caller must lock inode page */
1132 int f2fs_truncate_xattr_node(struct inode *inode)
1133 {
1134 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1135 	nid_t nid = F2FS_I(inode)->i_xattr_nid;
1136 	struct dnode_of_data dn;
1137 	struct page *npage;
1138 	int err;
1139 
1140 	if (!nid)
1141 		return 0;
1142 
1143 	npage = f2fs_get_node_page(sbi, nid);
1144 	if (IS_ERR(npage))
1145 		return PTR_ERR(npage);
1146 
1147 	set_new_dnode(&dn, inode, NULL, npage, nid);
1148 	err = truncate_node(&dn);
1149 	if (err) {
1150 		f2fs_put_page(npage, 1);
1151 		return err;
1152 	}
1153 
1154 	f2fs_i_xnid_write(inode, 0);
1155 
1156 	return 0;
1157 }
1158 
1159 /*
1160  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1161  * f2fs_unlock_op().
1162  */
1163 int f2fs_remove_inode_page(struct inode *inode)
1164 {
1165 	struct dnode_of_data dn;
1166 	int err;
1167 
1168 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1169 	err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1170 	if (err)
1171 		return err;
1172 
1173 	err = f2fs_truncate_xattr_node(inode);
1174 	if (err) {
1175 		f2fs_put_dnode(&dn);
1176 		return err;
1177 	}
1178 
1179 	/* remove potential inline_data blocks */
1180 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1181 				S_ISLNK(inode->i_mode))
1182 		f2fs_truncate_data_blocks_range(&dn, 1);
1183 
1184 	/* 0 is possible, after f2fs_new_inode() has failed */
1185 	if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1186 		f2fs_put_dnode(&dn);
1187 		return -EIO;
1188 	}
1189 
1190 	if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1191 		f2fs_warn(F2FS_I_SB(inode),
1192 			"f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1193 			inode->i_ino, (unsigned long long)inode->i_blocks);
1194 		set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1195 	}
1196 
1197 	/* will put inode & node pages */
1198 	err = truncate_node(&dn);
1199 	if (err) {
1200 		f2fs_put_dnode(&dn);
1201 		return err;
1202 	}
1203 	return 0;
1204 }
1205 
1206 struct page *f2fs_new_inode_page(struct inode *inode)
1207 {
1208 	struct dnode_of_data dn;
1209 
1210 	/* allocate inode page for new inode */
1211 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1212 
1213 	/* caller should f2fs_put_page(page, 1); */
1214 	return f2fs_new_node_page(&dn, 0);
1215 }
1216 
1217 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1218 {
1219 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1220 	struct node_info new_ni;
1221 	struct page *page;
1222 	int err;
1223 
1224 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1225 		return ERR_PTR(-EPERM);
1226 
1227 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1228 	if (!page)
1229 		return ERR_PTR(-ENOMEM);
1230 
1231 	if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1232 		goto fail;
1233 
1234 #ifdef CONFIG_F2FS_CHECK_FS
1235 	err = f2fs_get_node_info(sbi, dn->nid, &new_ni);
1236 	if (err) {
1237 		dec_valid_node_count(sbi, dn->inode, !ofs);
1238 		goto fail;
1239 	}
1240 	f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1241 #endif
1242 	new_ni.nid = dn->nid;
1243 	new_ni.ino = dn->inode->i_ino;
1244 	new_ni.blk_addr = NULL_ADDR;
1245 	new_ni.flag = 0;
1246 	new_ni.version = 0;
1247 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1248 
1249 	f2fs_wait_on_page_writeback(page, NODE, true, true);
1250 	fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1251 	set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1252 	if (!PageUptodate(page))
1253 		SetPageUptodate(page);
1254 	if (set_page_dirty(page))
1255 		dn->node_changed = true;
1256 
1257 	if (f2fs_has_xattr_block(ofs))
1258 		f2fs_i_xnid_write(dn->inode, dn->nid);
1259 
1260 	if (ofs == 0)
1261 		inc_valid_inode_count(sbi);
1262 	return page;
1263 
1264 fail:
1265 	clear_node_page_dirty(page);
1266 	f2fs_put_page(page, 1);
1267 	return ERR_PTR(err);
1268 }
1269 
1270 /*
1271  * Caller should do after getting the following values.
1272  * 0: f2fs_put_page(page, 0)
1273  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1274  */
1275 static int read_node_page(struct page *page, int op_flags)
1276 {
1277 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1278 	struct node_info ni;
1279 	struct f2fs_io_info fio = {
1280 		.sbi = sbi,
1281 		.type = NODE,
1282 		.op = REQ_OP_READ,
1283 		.op_flags = op_flags,
1284 		.page = page,
1285 		.encrypted_page = NULL,
1286 	};
1287 	int err;
1288 
1289 	if (PageUptodate(page)) {
1290 		if (!f2fs_inode_chksum_verify(sbi, page)) {
1291 			ClearPageUptodate(page);
1292 			return -EFSBADCRC;
1293 		}
1294 		return LOCKED_PAGE;
1295 	}
1296 
1297 	err = f2fs_get_node_info(sbi, page->index, &ni);
1298 	if (err)
1299 		return err;
1300 
1301 	if (unlikely(ni.blk_addr == NULL_ADDR) ||
1302 			is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1303 		ClearPageUptodate(page);
1304 		return -ENOENT;
1305 	}
1306 
1307 	fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1308 
1309 	err = f2fs_submit_page_bio(&fio);
1310 
1311 	if (!err)
1312 		f2fs_update_iostat(sbi, FS_NODE_READ_IO, F2FS_BLKSIZE);
1313 
1314 	return err;
1315 }
1316 
1317 /*
1318  * Readahead a node page
1319  */
1320 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1321 {
1322 	struct page *apage;
1323 	int err;
1324 
1325 	if (!nid)
1326 		return;
1327 	if (f2fs_check_nid_range(sbi, nid))
1328 		return;
1329 
1330 	apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1331 	if (apage)
1332 		return;
1333 
1334 	apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1335 	if (!apage)
1336 		return;
1337 
1338 	err = read_node_page(apage, REQ_RAHEAD);
1339 	f2fs_put_page(apage, err ? 1 : 0);
1340 }
1341 
1342 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1343 					struct page *parent, int start)
1344 {
1345 	struct page *page;
1346 	int err;
1347 
1348 	if (!nid)
1349 		return ERR_PTR(-ENOENT);
1350 	if (f2fs_check_nid_range(sbi, nid))
1351 		return ERR_PTR(-EINVAL);
1352 repeat:
1353 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1354 	if (!page)
1355 		return ERR_PTR(-ENOMEM);
1356 
1357 	err = read_node_page(page, 0);
1358 	if (err < 0) {
1359 		f2fs_put_page(page, 1);
1360 		return ERR_PTR(err);
1361 	} else if (err == LOCKED_PAGE) {
1362 		err = 0;
1363 		goto page_hit;
1364 	}
1365 
1366 	if (parent)
1367 		f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1368 
1369 	lock_page(page);
1370 
1371 	if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1372 		f2fs_put_page(page, 1);
1373 		goto repeat;
1374 	}
1375 
1376 	if (unlikely(!PageUptodate(page))) {
1377 		err = -EIO;
1378 		goto out_err;
1379 	}
1380 
1381 	if (!f2fs_inode_chksum_verify(sbi, page)) {
1382 		err = -EFSBADCRC;
1383 		goto out_err;
1384 	}
1385 page_hit:
1386 	if(unlikely(nid != nid_of_node(page))) {
1387 		f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1388 			  nid, nid_of_node(page), ino_of_node(page),
1389 			  ofs_of_node(page), cpver_of_node(page),
1390 			  next_blkaddr_of_node(page));
1391 		err = -EINVAL;
1392 out_err:
1393 		ClearPageUptodate(page);
1394 		f2fs_put_page(page, 1);
1395 		return ERR_PTR(err);
1396 	}
1397 	return page;
1398 }
1399 
1400 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1401 {
1402 	return __get_node_page(sbi, nid, NULL, 0);
1403 }
1404 
1405 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1406 {
1407 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1408 	nid_t nid = get_nid(parent, start, false);
1409 
1410 	return __get_node_page(sbi, nid, parent, start);
1411 }
1412 
1413 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1414 {
1415 	struct inode *inode;
1416 	struct page *page;
1417 	int ret;
1418 
1419 	/* should flush inline_data before evict_inode */
1420 	inode = ilookup(sbi->sb, ino);
1421 	if (!inode)
1422 		return;
1423 
1424 	page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1425 					FGP_LOCK|FGP_NOWAIT, 0);
1426 	if (!page)
1427 		goto iput_out;
1428 
1429 	if (!PageUptodate(page))
1430 		goto page_out;
1431 
1432 	if (!PageDirty(page))
1433 		goto page_out;
1434 
1435 	if (!clear_page_dirty_for_io(page))
1436 		goto page_out;
1437 
1438 	ret = f2fs_write_inline_data(inode, page);
1439 	inode_dec_dirty_pages(inode);
1440 	f2fs_remove_dirty_inode(inode);
1441 	if (ret)
1442 		set_page_dirty(page);
1443 page_out:
1444 	f2fs_put_page(page, 1);
1445 iput_out:
1446 	iput(inode);
1447 }
1448 
1449 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1450 {
1451 	pgoff_t index;
1452 	struct pagevec pvec;
1453 	struct page *last_page = NULL;
1454 	int nr_pages;
1455 
1456 	pagevec_init(&pvec);
1457 	index = 0;
1458 
1459 	while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1460 				PAGECACHE_TAG_DIRTY))) {
1461 		int i;
1462 
1463 		for (i = 0; i < nr_pages; i++) {
1464 			struct page *page = pvec.pages[i];
1465 
1466 			if (unlikely(f2fs_cp_error(sbi))) {
1467 				f2fs_put_page(last_page, 0);
1468 				pagevec_release(&pvec);
1469 				return ERR_PTR(-EIO);
1470 			}
1471 
1472 			if (!IS_DNODE(page) || !is_cold_node(page))
1473 				continue;
1474 			if (ino_of_node(page) != ino)
1475 				continue;
1476 
1477 			lock_page(page);
1478 
1479 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1480 continue_unlock:
1481 				unlock_page(page);
1482 				continue;
1483 			}
1484 			if (ino_of_node(page) != ino)
1485 				goto continue_unlock;
1486 
1487 			if (!PageDirty(page)) {
1488 				/* someone wrote it for us */
1489 				goto continue_unlock;
1490 			}
1491 
1492 			if (last_page)
1493 				f2fs_put_page(last_page, 0);
1494 
1495 			get_page(page);
1496 			last_page = page;
1497 			unlock_page(page);
1498 		}
1499 		pagevec_release(&pvec);
1500 		cond_resched();
1501 	}
1502 	return last_page;
1503 }
1504 
1505 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1506 				struct writeback_control *wbc, bool do_balance,
1507 				enum iostat_type io_type, unsigned int *seq_id)
1508 {
1509 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1510 	nid_t nid;
1511 	struct node_info ni;
1512 	struct f2fs_io_info fio = {
1513 		.sbi = sbi,
1514 		.ino = ino_of_node(page),
1515 		.type = NODE,
1516 		.op = REQ_OP_WRITE,
1517 		.op_flags = wbc_to_write_flags(wbc),
1518 		.page = page,
1519 		.encrypted_page = NULL,
1520 		.submitted = false,
1521 		.io_type = io_type,
1522 		.io_wbc = wbc,
1523 	};
1524 	unsigned int seq;
1525 
1526 	trace_f2fs_writepage(page, NODE);
1527 
1528 	if (unlikely(f2fs_cp_error(sbi))) {
1529 		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
1530 			ClearPageUptodate(page);
1531 			dec_page_count(sbi, F2FS_DIRTY_NODES);
1532 			unlock_page(page);
1533 			return 0;
1534 		}
1535 		goto redirty_out;
1536 	}
1537 
1538 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1539 		goto redirty_out;
1540 
1541 	if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1542 			wbc->sync_mode == WB_SYNC_NONE &&
1543 			IS_DNODE(page) && is_cold_node(page))
1544 		goto redirty_out;
1545 
1546 	/* get old block addr of this node page */
1547 	nid = nid_of_node(page);
1548 	f2fs_bug_on(sbi, page->index != nid);
1549 
1550 	if (f2fs_get_node_info(sbi, nid, &ni))
1551 		goto redirty_out;
1552 
1553 	if (wbc->for_reclaim) {
1554 		if (!down_read_trylock(&sbi->node_write))
1555 			goto redirty_out;
1556 	} else {
1557 		down_read(&sbi->node_write);
1558 	}
1559 
1560 	/* This page is already truncated */
1561 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1562 		ClearPageUptodate(page);
1563 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1564 		up_read(&sbi->node_write);
1565 		unlock_page(page);
1566 		return 0;
1567 	}
1568 
1569 	if (__is_valid_data_blkaddr(ni.blk_addr) &&
1570 		!f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1571 					DATA_GENERIC_ENHANCE)) {
1572 		up_read(&sbi->node_write);
1573 		goto redirty_out;
1574 	}
1575 
1576 	if (atomic && !test_opt(sbi, NOBARRIER))
1577 		fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1578 
1579 	/* should add to global list before clearing PAGECACHE status */
1580 	if (f2fs_in_warm_node_list(sbi, page)) {
1581 		seq = f2fs_add_fsync_node_entry(sbi, page);
1582 		if (seq_id)
1583 			*seq_id = seq;
1584 	}
1585 
1586 	set_page_writeback(page);
1587 	ClearPageError(page);
1588 
1589 	fio.old_blkaddr = ni.blk_addr;
1590 	f2fs_do_write_node_page(nid, &fio);
1591 	set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1592 	dec_page_count(sbi, F2FS_DIRTY_NODES);
1593 	up_read(&sbi->node_write);
1594 
1595 	if (wbc->for_reclaim) {
1596 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1597 		submitted = NULL;
1598 	}
1599 
1600 	unlock_page(page);
1601 
1602 	if (unlikely(f2fs_cp_error(sbi))) {
1603 		f2fs_submit_merged_write(sbi, NODE);
1604 		submitted = NULL;
1605 	}
1606 	if (submitted)
1607 		*submitted = fio.submitted;
1608 
1609 	if (do_balance)
1610 		f2fs_balance_fs(sbi, false);
1611 	return 0;
1612 
1613 redirty_out:
1614 	redirty_page_for_writepage(wbc, page);
1615 	return AOP_WRITEPAGE_ACTIVATE;
1616 }
1617 
1618 int f2fs_move_node_page(struct page *node_page, int gc_type)
1619 {
1620 	int err = 0;
1621 
1622 	if (gc_type == FG_GC) {
1623 		struct writeback_control wbc = {
1624 			.sync_mode = WB_SYNC_ALL,
1625 			.nr_to_write = 1,
1626 			.for_reclaim = 0,
1627 		};
1628 
1629 		f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1630 
1631 		set_page_dirty(node_page);
1632 
1633 		if (!clear_page_dirty_for_io(node_page)) {
1634 			err = -EAGAIN;
1635 			goto out_page;
1636 		}
1637 
1638 		if (__write_node_page(node_page, false, NULL,
1639 					&wbc, false, FS_GC_NODE_IO, NULL)) {
1640 			err = -EAGAIN;
1641 			unlock_page(node_page);
1642 		}
1643 		goto release_page;
1644 	} else {
1645 		/* set page dirty and write it */
1646 		if (!PageWriteback(node_page))
1647 			set_page_dirty(node_page);
1648 	}
1649 out_page:
1650 	unlock_page(node_page);
1651 release_page:
1652 	f2fs_put_page(node_page, 0);
1653 	return err;
1654 }
1655 
1656 static int f2fs_write_node_page(struct page *page,
1657 				struct writeback_control *wbc)
1658 {
1659 	return __write_node_page(page, false, NULL, wbc, false,
1660 						FS_NODE_IO, NULL);
1661 }
1662 
1663 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1664 			struct writeback_control *wbc, bool atomic,
1665 			unsigned int *seq_id)
1666 {
1667 	pgoff_t index;
1668 	struct pagevec pvec;
1669 	int ret = 0;
1670 	struct page *last_page = NULL;
1671 	bool marked = false;
1672 	nid_t ino = inode->i_ino;
1673 	int nr_pages;
1674 	int nwritten = 0;
1675 
1676 	if (atomic) {
1677 		last_page = last_fsync_dnode(sbi, ino);
1678 		if (IS_ERR_OR_NULL(last_page))
1679 			return PTR_ERR_OR_ZERO(last_page);
1680 	}
1681 retry:
1682 	pagevec_init(&pvec);
1683 	index = 0;
1684 
1685 	while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1686 				PAGECACHE_TAG_DIRTY))) {
1687 		int i;
1688 
1689 		for (i = 0; i < nr_pages; i++) {
1690 			struct page *page = pvec.pages[i];
1691 			bool submitted = false;
1692 
1693 			if (unlikely(f2fs_cp_error(sbi))) {
1694 				f2fs_put_page(last_page, 0);
1695 				pagevec_release(&pvec);
1696 				ret = -EIO;
1697 				goto out;
1698 			}
1699 
1700 			if (!IS_DNODE(page) || !is_cold_node(page))
1701 				continue;
1702 			if (ino_of_node(page) != ino)
1703 				continue;
1704 
1705 			lock_page(page);
1706 
1707 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1708 continue_unlock:
1709 				unlock_page(page);
1710 				continue;
1711 			}
1712 			if (ino_of_node(page) != ino)
1713 				goto continue_unlock;
1714 
1715 			if (!PageDirty(page) && page != last_page) {
1716 				/* someone wrote it for us */
1717 				goto continue_unlock;
1718 			}
1719 
1720 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1721 
1722 			set_fsync_mark(page, 0);
1723 			set_dentry_mark(page, 0);
1724 
1725 			if (!atomic || page == last_page) {
1726 				set_fsync_mark(page, 1);
1727 				if (IS_INODE(page)) {
1728 					if (is_inode_flag_set(inode,
1729 								FI_DIRTY_INODE))
1730 						f2fs_update_inode(inode, page);
1731 					set_dentry_mark(page,
1732 						f2fs_need_dentry_mark(sbi, ino));
1733 				}
1734 				/* may be written by other thread */
1735 				if (!PageDirty(page))
1736 					set_page_dirty(page);
1737 			}
1738 
1739 			if (!clear_page_dirty_for_io(page))
1740 				goto continue_unlock;
1741 
1742 			ret = __write_node_page(page, atomic &&
1743 						page == last_page,
1744 						&submitted, wbc, true,
1745 						FS_NODE_IO, seq_id);
1746 			if (ret) {
1747 				unlock_page(page);
1748 				f2fs_put_page(last_page, 0);
1749 				break;
1750 			} else if (submitted) {
1751 				nwritten++;
1752 			}
1753 
1754 			if (page == last_page) {
1755 				f2fs_put_page(page, 0);
1756 				marked = true;
1757 				break;
1758 			}
1759 		}
1760 		pagevec_release(&pvec);
1761 		cond_resched();
1762 
1763 		if (ret || marked)
1764 			break;
1765 	}
1766 	if (!ret && atomic && !marked) {
1767 		f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1768 			   ino, last_page->index);
1769 		lock_page(last_page);
1770 		f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1771 		set_page_dirty(last_page);
1772 		unlock_page(last_page);
1773 		goto retry;
1774 	}
1775 out:
1776 	if (nwritten)
1777 		f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1778 	return ret ? -EIO: 0;
1779 }
1780 
1781 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1782 {
1783 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1784 	bool clean;
1785 
1786 	if (inode->i_ino != ino)
1787 		return 0;
1788 
1789 	if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1790 		return 0;
1791 
1792 	spin_lock(&sbi->inode_lock[DIRTY_META]);
1793 	clean = list_empty(&F2FS_I(inode)->gdirty_list);
1794 	spin_unlock(&sbi->inode_lock[DIRTY_META]);
1795 
1796 	if (clean)
1797 		return 0;
1798 
1799 	inode = igrab(inode);
1800 	if (!inode)
1801 		return 0;
1802 	return 1;
1803 }
1804 
1805 static bool flush_dirty_inode(struct page *page)
1806 {
1807 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1808 	struct inode *inode;
1809 	nid_t ino = ino_of_node(page);
1810 
1811 	inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1812 	if (!inode)
1813 		return false;
1814 
1815 	f2fs_update_inode(inode, page);
1816 	unlock_page(page);
1817 
1818 	iput(inode);
1819 	return true;
1820 }
1821 
1822 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1823 {
1824 	pgoff_t index = 0;
1825 	struct pagevec pvec;
1826 	int nr_pages;
1827 
1828 	pagevec_init(&pvec);
1829 
1830 	while ((nr_pages = pagevec_lookup_tag(&pvec,
1831 			NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1832 		int i;
1833 
1834 		for (i = 0; i < nr_pages; i++) {
1835 			struct page *page = pvec.pages[i];
1836 
1837 			if (!IS_DNODE(page))
1838 				continue;
1839 
1840 			lock_page(page);
1841 
1842 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1843 continue_unlock:
1844 				unlock_page(page);
1845 				continue;
1846 			}
1847 
1848 			if (!PageDirty(page)) {
1849 				/* someone wrote it for us */
1850 				goto continue_unlock;
1851 			}
1852 
1853 			/* flush inline_data, if it's async context. */
1854 			if (is_inline_node(page)) {
1855 				clear_inline_node(page);
1856 				unlock_page(page);
1857 				flush_inline_data(sbi, ino_of_node(page));
1858 				continue;
1859 			}
1860 			unlock_page(page);
1861 		}
1862 		pagevec_release(&pvec);
1863 		cond_resched();
1864 	}
1865 }
1866 
1867 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1868 				struct writeback_control *wbc,
1869 				bool do_balance, enum iostat_type io_type)
1870 {
1871 	pgoff_t index;
1872 	struct pagevec pvec;
1873 	int step = 0;
1874 	int nwritten = 0;
1875 	int ret = 0;
1876 	int nr_pages, done = 0;
1877 
1878 	pagevec_init(&pvec);
1879 
1880 next_step:
1881 	index = 0;
1882 
1883 	while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1884 			NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1885 		int i;
1886 
1887 		for (i = 0; i < nr_pages; i++) {
1888 			struct page *page = pvec.pages[i];
1889 			bool submitted = false;
1890 			bool may_dirty = true;
1891 
1892 			/* give a priority to WB_SYNC threads */
1893 			if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1894 					wbc->sync_mode == WB_SYNC_NONE) {
1895 				done = 1;
1896 				break;
1897 			}
1898 
1899 			/*
1900 			 * flushing sequence with step:
1901 			 * 0. indirect nodes
1902 			 * 1. dentry dnodes
1903 			 * 2. file dnodes
1904 			 */
1905 			if (step == 0 && IS_DNODE(page))
1906 				continue;
1907 			if (step == 1 && (!IS_DNODE(page) ||
1908 						is_cold_node(page)))
1909 				continue;
1910 			if (step == 2 && (!IS_DNODE(page) ||
1911 						!is_cold_node(page)))
1912 				continue;
1913 lock_node:
1914 			if (wbc->sync_mode == WB_SYNC_ALL)
1915 				lock_page(page);
1916 			else if (!trylock_page(page))
1917 				continue;
1918 
1919 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1920 continue_unlock:
1921 				unlock_page(page);
1922 				continue;
1923 			}
1924 
1925 			if (!PageDirty(page)) {
1926 				/* someone wrote it for us */
1927 				goto continue_unlock;
1928 			}
1929 
1930 			/* flush inline_data/inode, if it's async context. */
1931 			if (!do_balance)
1932 				goto write_node;
1933 
1934 			/* flush inline_data */
1935 			if (is_inline_node(page)) {
1936 				clear_inline_node(page);
1937 				unlock_page(page);
1938 				flush_inline_data(sbi, ino_of_node(page));
1939 				goto lock_node;
1940 			}
1941 
1942 			/* flush dirty inode */
1943 			if (IS_INODE(page) && may_dirty) {
1944 				may_dirty = false;
1945 				if (flush_dirty_inode(page))
1946 					goto lock_node;
1947 			}
1948 write_node:
1949 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1950 
1951 			if (!clear_page_dirty_for_io(page))
1952 				goto continue_unlock;
1953 
1954 			set_fsync_mark(page, 0);
1955 			set_dentry_mark(page, 0);
1956 
1957 			ret = __write_node_page(page, false, &submitted,
1958 						wbc, do_balance, io_type, NULL);
1959 			if (ret)
1960 				unlock_page(page);
1961 			else if (submitted)
1962 				nwritten++;
1963 
1964 			if (--wbc->nr_to_write == 0)
1965 				break;
1966 		}
1967 		pagevec_release(&pvec);
1968 		cond_resched();
1969 
1970 		if (wbc->nr_to_write == 0) {
1971 			step = 2;
1972 			break;
1973 		}
1974 	}
1975 
1976 	if (step < 2) {
1977 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1978 				wbc->sync_mode == WB_SYNC_NONE && step == 1)
1979 			goto out;
1980 		step++;
1981 		goto next_step;
1982 	}
1983 out:
1984 	if (nwritten)
1985 		f2fs_submit_merged_write(sbi, NODE);
1986 
1987 	if (unlikely(f2fs_cp_error(sbi)))
1988 		return -EIO;
1989 	return ret;
1990 }
1991 
1992 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
1993 						unsigned int seq_id)
1994 {
1995 	struct fsync_node_entry *fn;
1996 	struct page *page;
1997 	struct list_head *head = &sbi->fsync_node_list;
1998 	unsigned long flags;
1999 	unsigned int cur_seq_id = 0;
2000 	int ret2, ret = 0;
2001 
2002 	while (seq_id && cur_seq_id < seq_id) {
2003 		spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2004 		if (list_empty(head)) {
2005 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2006 			break;
2007 		}
2008 		fn = list_first_entry(head, struct fsync_node_entry, list);
2009 		if (fn->seq_id > seq_id) {
2010 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2011 			break;
2012 		}
2013 		cur_seq_id = fn->seq_id;
2014 		page = fn->page;
2015 		get_page(page);
2016 		spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2017 
2018 		f2fs_wait_on_page_writeback(page, NODE, true, false);
2019 		if (TestClearPageError(page))
2020 			ret = -EIO;
2021 
2022 		put_page(page);
2023 
2024 		if (ret)
2025 			break;
2026 	}
2027 
2028 	ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2029 	if (!ret)
2030 		ret = ret2;
2031 
2032 	return ret;
2033 }
2034 
2035 static int f2fs_write_node_pages(struct address_space *mapping,
2036 			    struct writeback_control *wbc)
2037 {
2038 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2039 	struct blk_plug plug;
2040 	long diff;
2041 
2042 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2043 		goto skip_write;
2044 
2045 	/* balancing f2fs's metadata in background */
2046 	f2fs_balance_fs_bg(sbi, true);
2047 
2048 	/* collect a number of dirty node pages and write together */
2049 	if (wbc->sync_mode != WB_SYNC_ALL &&
2050 			get_pages(sbi, F2FS_DIRTY_NODES) <
2051 					nr_pages_to_skip(sbi, NODE))
2052 		goto skip_write;
2053 
2054 	if (wbc->sync_mode == WB_SYNC_ALL)
2055 		atomic_inc(&sbi->wb_sync_req[NODE]);
2056 	else if (atomic_read(&sbi->wb_sync_req[NODE]))
2057 		goto skip_write;
2058 
2059 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2060 
2061 	diff = nr_pages_to_write(sbi, NODE, wbc);
2062 	blk_start_plug(&plug);
2063 	f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2064 	blk_finish_plug(&plug);
2065 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2066 
2067 	if (wbc->sync_mode == WB_SYNC_ALL)
2068 		atomic_dec(&sbi->wb_sync_req[NODE]);
2069 	return 0;
2070 
2071 skip_write:
2072 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2073 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2074 	return 0;
2075 }
2076 
2077 static int f2fs_set_node_page_dirty(struct page *page)
2078 {
2079 	trace_f2fs_set_page_dirty(page, NODE);
2080 
2081 	if (!PageUptodate(page))
2082 		SetPageUptodate(page);
2083 #ifdef CONFIG_F2FS_CHECK_FS
2084 	if (IS_INODE(page))
2085 		f2fs_inode_chksum_set(F2FS_P_SB(page), page);
2086 #endif
2087 	if (!PageDirty(page)) {
2088 		__set_page_dirty_nobuffers(page);
2089 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
2090 		f2fs_set_page_private(page, 0);
2091 		return 1;
2092 	}
2093 	return 0;
2094 }
2095 
2096 /*
2097  * Structure of the f2fs node operations
2098  */
2099 const struct address_space_operations f2fs_node_aops = {
2100 	.writepage	= f2fs_write_node_page,
2101 	.writepages	= f2fs_write_node_pages,
2102 	.set_page_dirty	= f2fs_set_node_page_dirty,
2103 	.invalidatepage	= f2fs_invalidate_page,
2104 	.releasepage	= f2fs_release_page,
2105 #ifdef CONFIG_MIGRATION
2106 	.migratepage	= f2fs_migrate_page,
2107 #endif
2108 };
2109 
2110 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2111 						nid_t n)
2112 {
2113 	return radix_tree_lookup(&nm_i->free_nid_root, n);
2114 }
2115 
2116 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2117 				struct free_nid *i)
2118 {
2119 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2120 
2121 	int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2122 	if (err)
2123 		return err;
2124 
2125 	nm_i->nid_cnt[FREE_NID]++;
2126 	list_add_tail(&i->list, &nm_i->free_nid_list);
2127 	return 0;
2128 }
2129 
2130 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2131 			struct free_nid *i, enum nid_state state)
2132 {
2133 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2134 
2135 	f2fs_bug_on(sbi, state != i->state);
2136 	nm_i->nid_cnt[state]--;
2137 	if (state == FREE_NID)
2138 		list_del(&i->list);
2139 	radix_tree_delete(&nm_i->free_nid_root, i->nid);
2140 }
2141 
2142 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2143 			enum nid_state org_state, enum nid_state dst_state)
2144 {
2145 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2146 
2147 	f2fs_bug_on(sbi, org_state != i->state);
2148 	i->state = dst_state;
2149 	nm_i->nid_cnt[org_state]--;
2150 	nm_i->nid_cnt[dst_state]++;
2151 
2152 	switch (dst_state) {
2153 	case PREALLOC_NID:
2154 		list_del(&i->list);
2155 		break;
2156 	case FREE_NID:
2157 		list_add_tail(&i->list, &nm_i->free_nid_list);
2158 		break;
2159 	default:
2160 		BUG_ON(1);
2161 	}
2162 }
2163 
2164 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2165 							bool set, bool build)
2166 {
2167 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2168 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2169 	unsigned int nid_ofs = nid - START_NID(nid);
2170 
2171 	if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2172 		return;
2173 
2174 	if (set) {
2175 		if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2176 			return;
2177 		__set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2178 		nm_i->free_nid_count[nat_ofs]++;
2179 	} else {
2180 		if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2181 			return;
2182 		__clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2183 		if (!build)
2184 			nm_i->free_nid_count[nat_ofs]--;
2185 	}
2186 }
2187 
2188 /* return if the nid is recognized as free */
2189 static bool add_free_nid(struct f2fs_sb_info *sbi,
2190 				nid_t nid, bool build, bool update)
2191 {
2192 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2193 	struct free_nid *i, *e;
2194 	struct nat_entry *ne;
2195 	int err = -EINVAL;
2196 	bool ret = false;
2197 
2198 	/* 0 nid should not be used */
2199 	if (unlikely(nid == 0))
2200 		return false;
2201 
2202 	if (unlikely(f2fs_check_nid_range(sbi, nid)))
2203 		return false;
2204 
2205 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2206 	i->nid = nid;
2207 	i->state = FREE_NID;
2208 
2209 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2210 
2211 	spin_lock(&nm_i->nid_list_lock);
2212 
2213 	if (build) {
2214 		/*
2215 		 *   Thread A             Thread B
2216 		 *  - f2fs_create
2217 		 *   - f2fs_new_inode
2218 		 *    - f2fs_alloc_nid
2219 		 *     - __insert_nid_to_list(PREALLOC_NID)
2220 		 *                     - f2fs_balance_fs_bg
2221 		 *                      - f2fs_build_free_nids
2222 		 *                       - __f2fs_build_free_nids
2223 		 *                        - scan_nat_page
2224 		 *                         - add_free_nid
2225 		 *                          - __lookup_nat_cache
2226 		 *  - f2fs_add_link
2227 		 *   - f2fs_init_inode_metadata
2228 		 *    - f2fs_new_inode_page
2229 		 *     - f2fs_new_node_page
2230 		 *      - set_node_addr
2231 		 *  - f2fs_alloc_nid_done
2232 		 *   - __remove_nid_from_list(PREALLOC_NID)
2233 		 *                         - __insert_nid_to_list(FREE_NID)
2234 		 */
2235 		ne = __lookup_nat_cache(nm_i, nid);
2236 		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2237 				nat_get_blkaddr(ne) != NULL_ADDR))
2238 			goto err_out;
2239 
2240 		e = __lookup_free_nid_list(nm_i, nid);
2241 		if (e) {
2242 			if (e->state == FREE_NID)
2243 				ret = true;
2244 			goto err_out;
2245 		}
2246 	}
2247 	ret = true;
2248 	err = __insert_free_nid(sbi, i);
2249 err_out:
2250 	if (update) {
2251 		update_free_nid_bitmap(sbi, nid, ret, build);
2252 		if (!build)
2253 			nm_i->available_nids++;
2254 	}
2255 	spin_unlock(&nm_i->nid_list_lock);
2256 	radix_tree_preload_end();
2257 
2258 	if (err)
2259 		kmem_cache_free(free_nid_slab, i);
2260 	return ret;
2261 }
2262 
2263 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2264 {
2265 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2266 	struct free_nid *i;
2267 	bool need_free = false;
2268 
2269 	spin_lock(&nm_i->nid_list_lock);
2270 	i = __lookup_free_nid_list(nm_i, nid);
2271 	if (i && i->state == FREE_NID) {
2272 		__remove_free_nid(sbi, i, FREE_NID);
2273 		need_free = true;
2274 	}
2275 	spin_unlock(&nm_i->nid_list_lock);
2276 
2277 	if (need_free)
2278 		kmem_cache_free(free_nid_slab, i);
2279 }
2280 
2281 static int scan_nat_page(struct f2fs_sb_info *sbi,
2282 			struct page *nat_page, nid_t start_nid)
2283 {
2284 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2285 	struct f2fs_nat_block *nat_blk = page_address(nat_page);
2286 	block_t blk_addr;
2287 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2288 	int i;
2289 
2290 	__set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2291 
2292 	i = start_nid % NAT_ENTRY_PER_BLOCK;
2293 
2294 	for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2295 		if (unlikely(start_nid >= nm_i->max_nid))
2296 			break;
2297 
2298 		blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2299 
2300 		if (blk_addr == NEW_ADDR)
2301 			return -EINVAL;
2302 
2303 		if (blk_addr == NULL_ADDR) {
2304 			add_free_nid(sbi, start_nid, true, true);
2305 		} else {
2306 			spin_lock(&NM_I(sbi)->nid_list_lock);
2307 			update_free_nid_bitmap(sbi, start_nid, false, true);
2308 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2309 		}
2310 	}
2311 
2312 	return 0;
2313 }
2314 
2315 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2316 {
2317 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2318 	struct f2fs_journal *journal = curseg->journal;
2319 	int i;
2320 
2321 	down_read(&curseg->journal_rwsem);
2322 	for (i = 0; i < nats_in_cursum(journal); i++) {
2323 		block_t addr;
2324 		nid_t nid;
2325 
2326 		addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2327 		nid = le32_to_cpu(nid_in_journal(journal, i));
2328 		if (addr == NULL_ADDR)
2329 			add_free_nid(sbi, nid, true, false);
2330 		else
2331 			remove_free_nid(sbi, nid);
2332 	}
2333 	up_read(&curseg->journal_rwsem);
2334 }
2335 
2336 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2337 {
2338 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2339 	unsigned int i, idx;
2340 	nid_t nid;
2341 
2342 	down_read(&nm_i->nat_tree_lock);
2343 
2344 	for (i = 0; i < nm_i->nat_blocks; i++) {
2345 		if (!test_bit_le(i, nm_i->nat_block_bitmap))
2346 			continue;
2347 		if (!nm_i->free_nid_count[i])
2348 			continue;
2349 		for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2350 			idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2351 						NAT_ENTRY_PER_BLOCK, idx);
2352 			if (idx >= NAT_ENTRY_PER_BLOCK)
2353 				break;
2354 
2355 			nid = i * NAT_ENTRY_PER_BLOCK + idx;
2356 			add_free_nid(sbi, nid, true, false);
2357 
2358 			if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2359 				goto out;
2360 		}
2361 	}
2362 out:
2363 	scan_curseg_cache(sbi);
2364 
2365 	up_read(&nm_i->nat_tree_lock);
2366 }
2367 
2368 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2369 						bool sync, bool mount)
2370 {
2371 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2372 	int i = 0, ret;
2373 	nid_t nid = nm_i->next_scan_nid;
2374 
2375 	if (unlikely(nid >= nm_i->max_nid))
2376 		nid = 0;
2377 
2378 	if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2379 		nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2380 
2381 	/* Enough entries */
2382 	if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2383 		return 0;
2384 
2385 	if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2386 		return 0;
2387 
2388 	if (!mount) {
2389 		/* try to find free nids in free_nid_bitmap */
2390 		scan_free_nid_bits(sbi);
2391 
2392 		if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2393 			return 0;
2394 	}
2395 
2396 	/* readahead nat pages to be scanned */
2397 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2398 							META_NAT, true);
2399 
2400 	down_read(&nm_i->nat_tree_lock);
2401 
2402 	while (1) {
2403 		if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2404 						nm_i->nat_block_bitmap)) {
2405 			struct page *page = get_current_nat_page(sbi, nid);
2406 
2407 			if (IS_ERR(page)) {
2408 				ret = PTR_ERR(page);
2409 			} else {
2410 				ret = scan_nat_page(sbi, page, nid);
2411 				f2fs_put_page(page, 1);
2412 			}
2413 
2414 			if (ret) {
2415 				up_read(&nm_i->nat_tree_lock);
2416 				f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2417 				return ret;
2418 			}
2419 		}
2420 
2421 		nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2422 		if (unlikely(nid >= nm_i->max_nid))
2423 			nid = 0;
2424 
2425 		if (++i >= FREE_NID_PAGES)
2426 			break;
2427 	}
2428 
2429 	/* go to the next free nat pages to find free nids abundantly */
2430 	nm_i->next_scan_nid = nid;
2431 
2432 	/* find free nids from current sum_pages */
2433 	scan_curseg_cache(sbi);
2434 
2435 	up_read(&nm_i->nat_tree_lock);
2436 
2437 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2438 					nm_i->ra_nid_pages, META_NAT, false);
2439 
2440 	return 0;
2441 }
2442 
2443 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2444 {
2445 	int ret;
2446 
2447 	mutex_lock(&NM_I(sbi)->build_lock);
2448 	ret = __f2fs_build_free_nids(sbi, sync, mount);
2449 	mutex_unlock(&NM_I(sbi)->build_lock);
2450 
2451 	return ret;
2452 }
2453 
2454 /*
2455  * If this function returns success, caller can obtain a new nid
2456  * from second parameter of this function.
2457  * The returned nid could be used ino as well as nid when inode is created.
2458  */
2459 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2460 {
2461 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2462 	struct free_nid *i = NULL;
2463 retry:
2464 	if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2465 		f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2466 		return false;
2467 	}
2468 
2469 	spin_lock(&nm_i->nid_list_lock);
2470 
2471 	if (unlikely(nm_i->available_nids == 0)) {
2472 		spin_unlock(&nm_i->nid_list_lock);
2473 		return false;
2474 	}
2475 
2476 	/* We should not use stale free nids created by f2fs_build_free_nids */
2477 	if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2478 		f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2479 		i = list_first_entry(&nm_i->free_nid_list,
2480 					struct free_nid, list);
2481 		*nid = i->nid;
2482 
2483 		__move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2484 		nm_i->available_nids--;
2485 
2486 		update_free_nid_bitmap(sbi, *nid, false, false);
2487 
2488 		spin_unlock(&nm_i->nid_list_lock);
2489 		return true;
2490 	}
2491 	spin_unlock(&nm_i->nid_list_lock);
2492 
2493 	/* Let's scan nat pages and its caches to get free nids */
2494 	if (!f2fs_build_free_nids(sbi, true, false))
2495 		goto retry;
2496 	return false;
2497 }
2498 
2499 /*
2500  * f2fs_alloc_nid() should be called prior to this function.
2501  */
2502 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2503 {
2504 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2505 	struct free_nid *i;
2506 
2507 	spin_lock(&nm_i->nid_list_lock);
2508 	i = __lookup_free_nid_list(nm_i, nid);
2509 	f2fs_bug_on(sbi, !i);
2510 	__remove_free_nid(sbi, i, PREALLOC_NID);
2511 	spin_unlock(&nm_i->nid_list_lock);
2512 
2513 	kmem_cache_free(free_nid_slab, i);
2514 }
2515 
2516 /*
2517  * f2fs_alloc_nid() should be called prior to this function.
2518  */
2519 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2520 {
2521 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2522 	struct free_nid *i;
2523 	bool need_free = false;
2524 
2525 	if (!nid)
2526 		return;
2527 
2528 	spin_lock(&nm_i->nid_list_lock);
2529 	i = __lookup_free_nid_list(nm_i, nid);
2530 	f2fs_bug_on(sbi, !i);
2531 
2532 	if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2533 		__remove_free_nid(sbi, i, PREALLOC_NID);
2534 		need_free = true;
2535 	} else {
2536 		__move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2537 	}
2538 
2539 	nm_i->available_nids++;
2540 
2541 	update_free_nid_bitmap(sbi, nid, true, false);
2542 
2543 	spin_unlock(&nm_i->nid_list_lock);
2544 
2545 	if (need_free)
2546 		kmem_cache_free(free_nid_slab, i);
2547 }
2548 
2549 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2550 {
2551 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2552 	int nr = nr_shrink;
2553 
2554 	if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2555 		return 0;
2556 
2557 	if (!mutex_trylock(&nm_i->build_lock))
2558 		return 0;
2559 
2560 	while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2561 		struct free_nid *i, *next;
2562 		unsigned int batch = SHRINK_NID_BATCH_SIZE;
2563 
2564 		spin_lock(&nm_i->nid_list_lock);
2565 		list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2566 			if (!nr_shrink || !batch ||
2567 				nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2568 				break;
2569 			__remove_free_nid(sbi, i, FREE_NID);
2570 			kmem_cache_free(free_nid_slab, i);
2571 			nr_shrink--;
2572 			batch--;
2573 		}
2574 		spin_unlock(&nm_i->nid_list_lock);
2575 	}
2576 
2577 	mutex_unlock(&nm_i->build_lock);
2578 
2579 	return nr - nr_shrink;
2580 }
2581 
2582 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2583 {
2584 	void *src_addr, *dst_addr;
2585 	size_t inline_size;
2586 	struct page *ipage;
2587 	struct f2fs_inode *ri;
2588 
2589 	ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2590 	if (IS_ERR(ipage))
2591 		return PTR_ERR(ipage);
2592 
2593 	ri = F2FS_INODE(page);
2594 	if (ri->i_inline & F2FS_INLINE_XATTR) {
2595 		if (!f2fs_has_inline_xattr(inode)) {
2596 			set_inode_flag(inode, FI_INLINE_XATTR);
2597 			stat_inc_inline_xattr(inode);
2598 		}
2599 	} else {
2600 		if (f2fs_has_inline_xattr(inode)) {
2601 			stat_dec_inline_xattr(inode);
2602 			clear_inode_flag(inode, FI_INLINE_XATTR);
2603 		}
2604 		goto update_inode;
2605 	}
2606 
2607 	dst_addr = inline_xattr_addr(inode, ipage);
2608 	src_addr = inline_xattr_addr(inode, page);
2609 	inline_size = inline_xattr_size(inode);
2610 
2611 	f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2612 	memcpy(dst_addr, src_addr, inline_size);
2613 update_inode:
2614 	f2fs_update_inode(inode, ipage);
2615 	f2fs_put_page(ipage, 1);
2616 	return 0;
2617 }
2618 
2619 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2620 {
2621 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2622 	nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2623 	nid_t new_xnid;
2624 	struct dnode_of_data dn;
2625 	struct node_info ni;
2626 	struct page *xpage;
2627 	int err;
2628 
2629 	if (!prev_xnid)
2630 		goto recover_xnid;
2631 
2632 	/* 1: invalidate the previous xattr nid */
2633 	err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2634 	if (err)
2635 		return err;
2636 
2637 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
2638 	dec_valid_node_count(sbi, inode, false);
2639 	set_node_addr(sbi, &ni, NULL_ADDR, false);
2640 
2641 recover_xnid:
2642 	/* 2: update xattr nid in inode */
2643 	if (!f2fs_alloc_nid(sbi, &new_xnid))
2644 		return -ENOSPC;
2645 
2646 	set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2647 	xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2648 	if (IS_ERR(xpage)) {
2649 		f2fs_alloc_nid_failed(sbi, new_xnid);
2650 		return PTR_ERR(xpage);
2651 	}
2652 
2653 	f2fs_alloc_nid_done(sbi, new_xnid);
2654 	f2fs_update_inode_page(inode);
2655 
2656 	/* 3: update and set xattr node page dirty */
2657 	memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2658 
2659 	set_page_dirty(xpage);
2660 	f2fs_put_page(xpage, 1);
2661 
2662 	return 0;
2663 }
2664 
2665 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2666 {
2667 	struct f2fs_inode *src, *dst;
2668 	nid_t ino = ino_of_node(page);
2669 	struct node_info old_ni, new_ni;
2670 	struct page *ipage;
2671 	int err;
2672 
2673 	err = f2fs_get_node_info(sbi, ino, &old_ni);
2674 	if (err)
2675 		return err;
2676 
2677 	if (unlikely(old_ni.blk_addr != NULL_ADDR))
2678 		return -EINVAL;
2679 retry:
2680 	ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2681 	if (!ipage) {
2682 		congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2683 		goto retry;
2684 	}
2685 
2686 	/* Should not use this inode from free nid list */
2687 	remove_free_nid(sbi, ino);
2688 
2689 	if (!PageUptodate(ipage))
2690 		SetPageUptodate(ipage);
2691 	fill_node_footer(ipage, ino, ino, 0, true);
2692 	set_cold_node(ipage, false);
2693 
2694 	src = F2FS_INODE(page);
2695 	dst = F2FS_INODE(ipage);
2696 
2697 	memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2698 	dst->i_size = 0;
2699 	dst->i_blocks = cpu_to_le64(1);
2700 	dst->i_links = cpu_to_le32(1);
2701 	dst->i_xattr_nid = 0;
2702 	dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2703 	if (dst->i_inline & F2FS_EXTRA_ATTR) {
2704 		dst->i_extra_isize = src->i_extra_isize;
2705 
2706 		if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2707 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2708 							i_inline_xattr_size))
2709 			dst->i_inline_xattr_size = src->i_inline_xattr_size;
2710 
2711 		if (f2fs_sb_has_project_quota(sbi) &&
2712 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2713 								i_projid))
2714 			dst->i_projid = src->i_projid;
2715 
2716 		if (f2fs_sb_has_inode_crtime(sbi) &&
2717 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2718 							i_crtime_nsec)) {
2719 			dst->i_crtime = src->i_crtime;
2720 			dst->i_crtime_nsec = src->i_crtime_nsec;
2721 		}
2722 	}
2723 
2724 	new_ni = old_ni;
2725 	new_ni.ino = ino;
2726 
2727 	if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2728 		WARN_ON(1);
2729 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2730 	inc_valid_inode_count(sbi);
2731 	set_page_dirty(ipage);
2732 	f2fs_put_page(ipage, 1);
2733 	return 0;
2734 }
2735 
2736 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2737 			unsigned int segno, struct f2fs_summary_block *sum)
2738 {
2739 	struct f2fs_node *rn;
2740 	struct f2fs_summary *sum_entry;
2741 	block_t addr;
2742 	int i, idx, last_offset, nrpages;
2743 
2744 	/* scan the node segment */
2745 	last_offset = sbi->blocks_per_seg;
2746 	addr = START_BLOCK(sbi, segno);
2747 	sum_entry = &sum->entries[0];
2748 
2749 	for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2750 		nrpages = bio_max_segs(last_offset - i);
2751 
2752 		/* readahead node pages */
2753 		f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2754 
2755 		for (idx = addr; idx < addr + nrpages; idx++) {
2756 			struct page *page = f2fs_get_tmp_page(sbi, idx);
2757 
2758 			if (IS_ERR(page))
2759 				return PTR_ERR(page);
2760 
2761 			rn = F2FS_NODE(page);
2762 			sum_entry->nid = rn->footer.nid;
2763 			sum_entry->version = 0;
2764 			sum_entry->ofs_in_node = 0;
2765 			sum_entry++;
2766 			f2fs_put_page(page, 1);
2767 		}
2768 
2769 		invalidate_mapping_pages(META_MAPPING(sbi), addr,
2770 							addr + nrpages);
2771 	}
2772 	return 0;
2773 }
2774 
2775 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2776 {
2777 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2778 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2779 	struct f2fs_journal *journal = curseg->journal;
2780 	int i;
2781 
2782 	down_write(&curseg->journal_rwsem);
2783 	for (i = 0; i < nats_in_cursum(journal); i++) {
2784 		struct nat_entry *ne;
2785 		struct f2fs_nat_entry raw_ne;
2786 		nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2787 
2788 		raw_ne = nat_in_journal(journal, i);
2789 
2790 		ne = __lookup_nat_cache(nm_i, nid);
2791 		if (!ne) {
2792 			ne = __alloc_nat_entry(nid, true);
2793 			__init_nat_entry(nm_i, ne, &raw_ne, true);
2794 		}
2795 
2796 		/*
2797 		 * if a free nat in journal has not been used after last
2798 		 * checkpoint, we should remove it from available nids,
2799 		 * since later we will add it again.
2800 		 */
2801 		if (!get_nat_flag(ne, IS_DIRTY) &&
2802 				le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2803 			spin_lock(&nm_i->nid_list_lock);
2804 			nm_i->available_nids--;
2805 			spin_unlock(&nm_i->nid_list_lock);
2806 		}
2807 
2808 		__set_nat_cache_dirty(nm_i, ne);
2809 	}
2810 	update_nats_in_cursum(journal, -i);
2811 	up_write(&curseg->journal_rwsem);
2812 }
2813 
2814 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2815 						struct list_head *head, int max)
2816 {
2817 	struct nat_entry_set *cur;
2818 
2819 	if (nes->entry_cnt >= max)
2820 		goto add_out;
2821 
2822 	list_for_each_entry(cur, head, set_list) {
2823 		if (cur->entry_cnt >= nes->entry_cnt) {
2824 			list_add(&nes->set_list, cur->set_list.prev);
2825 			return;
2826 		}
2827 	}
2828 add_out:
2829 	list_add_tail(&nes->set_list, head);
2830 }
2831 
2832 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2833 						struct page *page)
2834 {
2835 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2836 	unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2837 	struct f2fs_nat_block *nat_blk = page_address(page);
2838 	int valid = 0;
2839 	int i = 0;
2840 
2841 	if (!enabled_nat_bits(sbi, NULL))
2842 		return;
2843 
2844 	if (nat_index == 0) {
2845 		valid = 1;
2846 		i = 1;
2847 	}
2848 	for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2849 		if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2850 			valid++;
2851 	}
2852 	if (valid == 0) {
2853 		__set_bit_le(nat_index, nm_i->empty_nat_bits);
2854 		__clear_bit_le(nat_index, nm_i->full_nat_bits);
2855 		return;
2856 	}
2857 
2858 	__clear_bit_le(nat_index, nm_i->empty_nat_bits);
2859 	if (valid == NAT_ENTRY_PER_BLOCK)
2860 		__set_bit_le(nat_index, nm_i->full_nat_bits);
2861 	else
2862 		__clear_bit_le(nat_index, nm_i->full_nat_bits);
2863 }
2864 
2865 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2866 		struct nat_entry_set *set, struct cp_control *cpc)
2867 {
2868 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2869 	struct f2fs_journal *journal = curseg->journal;
2870 	nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2871 	bool to_journal = true;
2872 	struct f2fs_nat_block *nat_blk;
2873 	struct nat_entry *ne, *cur;
2874 	struct page *page = NULL;
2875 
2876 	/*
2877 	 * there are two steps to flush nat entries:
2878 	 * #1, flush nat entries to journal in current hot data summary block.
2879 	 * #2, flush nat entries to nat page.
2880 	 */
2881 	if (enabled_nat_bits(sbi, cpc) ||
2882 		!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2883 		to_journal = false;
2884 
2885 	if (to_journal) {
2886 		down_write(&curseg->journal_rwsem);
2887 	} else {
2888 		page = get_next_nat_page(sbi, start_nid);
2889 		if (IS_ERR(page))
2890 			return PTR_ERR(page);
2891 
2892 		nat_blk = page_address(page);
2893 		f2fs_bug_on(sbi, !nat_blk);
2894 	}
2895 
2896 	/* flush dirty nats in nat entry set */
2897 	list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2898 		struct f2fs_nat_entry *raw_ne;
2899 		nid_t nid = nat_get_nid(ne);
2900 		int offset;
2901 
2902 		f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2903 
2904 		if (to_journal) {
2905 			offset = f2fs_lookup_journal_in_cursum(journal,
2906 							NAT_JOURNAL, nid, 1);
2907 			f2fs_bug_on(sbi, offset < 0);
2908 			raw_ne = &nat_in_journal(journal, offset);
2909 			nid_in_journal(journal, offset) = cpu_to_le32(nid);
2910 		} else {
2911 			raw_ne = &nat_blk->entries[nid - start_nid];
2912 		}
2913 		raw_nat_from_node_info(raw_ne, &ne->ni);
2914 		nat_reset_flag(ne);
2915 		__clear_nat_cache_dirty(NM_I(sbi), set, ne);
2916 		if (nat_get_blkaddr(ne) == NULL_ADDR) {
2917 			add_free_nid(sbi, nid, false, true);
2918 		} else {
2919 			spin_lock(&NM_I(sbi)->nid_list_lock);
2920 			update_free_nid_bitmap(sbi, nid, false, false);
2921 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2922 		}
2923 	}
2924 
2925 	if (to_journal) {
2926 		up_write(&curseg->journal_rwsem);
2927 	} else {
2928 		__update_nat_bits(sbi, start_nid, page);
2929 		f2fs_put_page(page, 1);
2930 	}
2931 
2932 	/* Allow dirty nats by node block allocation in write_begin */
2933 	if (!set->entry_cnt) {
2934 		radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2935 		kmem_cache_free(nat_entry_set_slab, set);
2936 	}
2937 	return 0;
2938 }
2939 
2940 /*
2941  * This function is called during the checkpointing process.
2942  */
2943 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2944 {
2945 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2946 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2947 	struct f2fs_journal *journal = curseg->journal;
2948 	struct nat_entry_set *setvec[SETVEC_SIZE];
2949 	struct nat_entry_set *set, *tmp;
2950 	unsigned int found;
2951 	nid_t set_idx = 0;
2952 	LIST_HEAD(sets);
2953 	int err = 0;
2954 
2955 	/*
2956 	 * during unmount, let's flush nat_bits before checking
2957 	 * nat_cnt[DIRTY_NAT].
2958 	 */
2959 	if (enabled_nat_bits(sbi, cpc)) {
2960 		down_write(&nm_i->nat_tree_lock);
2961 		remove_nats_in_journal(sbi);
2962 		up_write(&nm_i->nat_tree_lock);
2963 	}
2964 
2965 	if (!nm_i->nat_cnt[DIRTY_NAT])
2966 		return 0;
2967 
2968 	down_write(&nm_i->nat_tree_lock);
2969 
2970 	/*
2971 	 * if there are no enough space in journal to store dirty nat
2972 	 * entries, remove all entries from journal and merge them
2973 	 * into nat entry set.
2974 	 */
2975 	if (enabled_nat_bits(sbi, cpc) ||
2976 		!__has_cursum_space(journal,
2977 			nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
2978 		remove_nats_in_journal(sbi);
2979 
2980 	while ((found = __gang_lookup_nat_set(nm_i,
2981 					set_idx, SETVEC_SIZE, setvec))) {
2982 		unsigned idx;
2983 		set_idx = setvec[found - 1]->set + 1;
2984 		for (idx = 0; idx < found; idx++)
2985 			__adjust_nat_entry_set(setvec[idx], &sets,
2986 						MAX_NAT_JENTRIES(journal));
2987 	}
2988 
2989 	/* flush dirty nats in nat entry set */
2990 	list_for_each_entry_safe(set, tmp, &sets, set_list) {
2991 		err = __flush_nat_entry_set(sbi, set, cpc);
2992 		if (err)
2993 			break;
2994 	}
2995 
2996 	up_write(&nm_i->nat_tree_lock);
2997 	/* Allow dirty nats by node block allocation in write_begin */
2998 
2999 	return err;
3000 }
3001 
3002 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3003 {
3004 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3005 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3006 	unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3007 	unsigned int i;
3008 	__u64 cp_ver = cur_cp_version(ckpt);
3009 	block_t nat_bits_addr;
3010 
3011 	if (!enabled_nat_bits(sbi, NULL))
3012 		return 0;
3013 
3014 	nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3015 	nm_i->nat_bits = f2fs_kvzalloc(sbi,
3016 			nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3017 	if (!nm_i->nat_bits)
3018 		return -ENOMEM;
3019 
3020 	nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3021 						nm_i->nat_bits_blocks;
3022 	for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3023 		struct page *page;
3024 
3025 		page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3026 		if (IS_ERR(page))
3027 			return PTR_ERR(page);
3028 
3029 		memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3030 					page_address(page), F2FS_BLKSIZE);
3031 		f2fs_put_page(page, 1);
3032 	}
3033 
3034 	cp_ver |= (cur_cp_crc(ckpt) << 32);
3035 	if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3036 		disable_nat_bits(sbi, true);
3037 		return 0;
3038 	}
3039 
3040 	nm_i->full_nat_bits = nm_i->nat_bits + 8;
3041 	nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3042 
3043 	f2fs_notice(sbi, "Found nat_bits in checkpoint");
3044 	return 0;
3045 }
3046 
3047 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3048 {
3049 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3050 	unsigned int i = 0;
3051 	nid_t nid, last_nid;
3052 
3053 	if (!enabled_nat_bits(sbi, NULL))
3054 		return;
3055 
3056 	for (i = 0; i < nm_i->nat_blocks; i++) {
3057 		i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3058 		if (i >= nm_i->nat_blocks)
3059 			break;
3060 
3061 		__set_bit_le(i, nm_i->nat_block_bitmap);
3062 
3063 		nid = i * NAT_ENTRY_PER_BLOCK;
3064 		last_nid = nid + NAT_ENTRY_PER_BLOCK;
3065 
3066 		spin_lock(&NM_I(sbi)->nid_list_lock);
3067 		for (; nid < last_nid; nid++)
3068 			update_free_nid_bitmap(sbi, nid, true, true);
3069 		spin_unlock(&NM_I(sbi)->nid_list_lock);
3070 	}
3071 
3072 	for (i = 0; i < nm_i->nat_blocks; i++) {
3073 		i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3074 		if (i >= nm_i->nat_blocks)
3075 			break;
3076 
3077 		__set_bit_le(i, nm_i->nat_block_bitmap);
3078 	}
3079 }
3080 
3081 static int init_node_manager(struct f2fs_sb_info *sbi)
3082 {
3083 	struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3084 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3085 	unsigned char *version_bitmap;
3086 	unsigned int nat_segs;
3087 	int err;
3088 
3089 	nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3090 
3091 	/* segment_count_nat includes pair segment so divide to 2. */
3092 	nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3093 	nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3094 	nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3095 
3096 	/* not used nids: 0, node, meta, (and root counted as valid node) */
3097 	nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3098 						F2FS_RESERVED_NODE_NUM;
3099 	nm_i->nid_cnt[FREE_NID] = 0;
3100 	nm_i->nid_cnt[PREALLOC_NID] = 0;
3101 	nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3102 	nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3103 	nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3104 
3105 	INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3106 	INIT_LIST_HEAD(&nm_i->free_nid_list);
3107 	INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3108 	INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3109 	INIT_LIST_HEAD(&nm_i->nat_entries);
3110 	spin_lock_init(&nm_i->nat_list_lock);
3111 
3112 	mutex_init(&nm_i->build_lock);
3113 	spin_lock_init(&nm_i->nid_list_lock);
3114 	init_rwsem(&nm_i->nat_tree_lock);
3115 
3116 	nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3117 	nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3118 	version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3119 	nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3120 					GFP_KERNEL);
3121 	if (!nm_i->nat_bitmap)
3122 		return -ENOMEM;
3123 
3124 	err = __get_nat_bitmaps(sbi);
3125 	if (err)
3126 		return err;
3127 
3128 #ifdef CONFIG_F2FS_CHECK_FS
3129 	nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3130 					GFP_KERNEL);
3131 	if (!nm_i->nat_bitmap_mir)
3132 		return -ENOMEM;
3133 #endif
3134 
3135 	return 0;
3136 }
3137 
3138 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3139 {
3140 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3141 	int i;
3142 
3143 	nm_i->free_nid_bitmap =
3144 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3145 					      nm_i->nat_blocks),
3146 			      GFP_KERNEL);
3147 	if (!nm_i->free_nid_bitmap)
3148 		return -ENOMEM;
3149 
3150 	for (i = 0; i < nm_i->nat_blocks; i++) {
3151 		nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3152 			f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3153 		if (!nm_i->free_nid_bitmap[i])
3154 			return -ENOMEM;
3155 	}
3156 
3157 	nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3158 								GFP_KERNEL);
3159 	if (!nm_i->nat_block_bitmap)
3160 		return -ENOMEM;
3161 
3162 	nm_i->free_nid_count =
3163 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3164 					      nm_i->nat_blocks),
3165 			      GFP_KERNEL);
3166 	if (!nm_i->free_nid_count)
3167 		return -ENOMEM;
3168 	return 0;
3169 }
3170 
3171 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3172 {
3173 	int err;
3174 
3175 	sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3176 							GFP_KERNEL);
3177 	if (!sbi->nm_info)
3178 		return -ENOMEM;
3179 
3180 	err = init_node_manager(sbi);
3181 	if (err)
3182 		return err;
3183 
3184 	err = init_free_nid_cache(sbi);
3185 	if (err)
3186 		return err;
3187 
3188 	/* load free nid status from nat_bits table */
3189 	load_free_nid_bitmap(sbi);
3190 
3191 	return f2fs_build_free_nids(sbi, true, true);
3192 }
3193 
3194 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3195 {
3196 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3197 	struct free_nid *i, *next_i;
3198 	struct nat_entry *natvec[NATVEC_SIZE];
3199 	struct nat_entry_set *setvec[SETVEC_SIZE];
3200 	nid_t nid = 0;
3201 	unsigned int found;
3202 
3203 	if (!nm_i)
3204 		return;
3205 
3206 	/* destroy free nid list */
3207 	spin_lock(&nm_i->nid_list_lock);
3208 	list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3209 		__remove_free_nid(sbi, i, FREE_NID);
3210 		spin_unlock(&nm_i->nid_list_lock);
3211 		kmem_cache_free(free_nid_slab, i);
3212 		spin_lock(&nm_i->nid_list_lock);
3213 	}
3214 	f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3215 	f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3216 	f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3217 	spin_unlock(&nm_i->nid_list_lock);
3218 
3219 	/* destroy nat cache */
3220 	down_write(&nm_i->nat_tree_lock);
3221 	while ((found = __gang_lookup_nat_cache(nm_i,
3222 					nid, NATVEC_SIZE, natvec))) {
3223 		unsigned idx;
3224 
3225 		nid = nat_get_nid(natvec[found - 1]) + 1;
3226 		for (idx = 0; idx < found; idx++) {
3227 			spin_lock(&nm_i->nat_list_lock);
3228 			list_del(&natvec[idx]->list);
3229 			spin_unlock(&nm_i->nat_list_lock);
3230 
3231 			__del_from_nat_cache(nm_i, natvec[idx]);
3232 		}
3233 	}
3234 	f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3235 
3236 	/* destroy nat set cache */
3237 	nid = 0;
3238 	while ((found = __gang_lookup_nat_set(nm_i,
3239 					nid, SETVEC_SIZE, setvec))) {
3240 		unsigned idx;
3241 
3242 		nid = setvec[found - 1]->set + 1;
3243 		for (idx = 0; idx < found; idx++) {
3244 			/* entry_cnt is not zero, when cp_error was occurred */
3245 			f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3246 			radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3247 			kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3248 		}
3249 	}
3250 	up_write(&nm_i->nat_tree_lock);
3251 
3252 	kvfree(nm_i->nat_block_bitmap);
3253 	if (nm_i->free_nid_bitmap) {
3254 		int i;
3255 
3256 		for (i = 0; i < nm_i->nat_blocks; i++)
3257 			kvfree(nm_i->free_nid_bitmap[i]);
3258 		kvfree(nm_i->free_nid_bitmap);
3259 	}
3260 	kvfree(nm_i->free_nid_count);
3261 
3262 	kvfree(nm_i->nat_bitmap);
3263 	kvfree(nm_i->nat_bits);
3264 #ifdef CONFIG_F2FS_CHECK_FS
3265 	kvfree(nm_i->nat_bitmap_mir);
3266 #endif
3267 	sbi->nm_info = NULL;
3268 	kfree(nm_i);
3269 }
3270 
3271 int __init f2fs_create_node_manager_caches(void)
3272 {
3273 	nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3274 			sizeof(struct nat_entry));
3275 	if (!nat_entry_slab)
3276 		goto fail;
3277 
3278 	free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3279 			sizeof(struct free_nid));
3280 	if (!free_nid_slab)
3281 		goto destroy_nat_entry;
3282 
3283 	nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3284 			sizeof(struct nat_entry_set));
3285 	if (!nat_entry_set_slab)
3286 		goto destroy_free_nid;
3287 
3288 	fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3289 			sizeof(struct fsync_node_entry));
3290 	if (!fsync_node_entry_slab)
3291 		goto destroy_nat_entry_set;
3292 	return 0;
3293 
3294 destroy_nat_entry_set:
3295 	kmem_cache_destroy(nat_entry_set_slab);
3296 destroy_free_nid:
3297 	kmem_cache_destroy(free_nid_slab);
3298 destroy_nat_entry:
3299 	kmem_cache_destroy(nat_entry_slab);
3300 fail:
3301 	return -ENOMEM;
3302 }
3303 
3304 void f2fs_destroy_node_manager_caches(void)
3305 {
3306 	kmem_cache_destroy(fsync_node_entry_slab);
3307 	kmem_cache_destroy(nat_entry_set_slab);
3308 	kmem_cache_destroy(free_nid_slab);
3309 	kmem_cache_destroy(nat_entry_slab);
3310 }
3311