xref: /openbmc/linux/fs/f2fs/node.c (revision 7e60e389)
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.h"
21 #include <trace/events/f2fs.h>
22 
23 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
24 
25 static struct kmem_cache *nat_entry_slab;
26 static struct kmem_cache *free_nid_slab;
27 static struct kmem_cache *nat_entry_set_slab;
28 static struct kmem_cache *fsync_node_entry_slab;
29 
30 /*
31  * Check whether the given nid is within node id range.
32  */
33 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
34 {
35 	if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
36 		set_sbi_flag(sbi, SBI_NEED_FSCK);
37 		f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
38 			  __func__, nid);
39 		return -EFSCORRUPTED;
40 	}
41 	return 0;
42 }
43 
44 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
45 {
46 	struct f2fs_nm_info *nm_i = NM_I(sbi);
47 	struct sysinfo val;
48 	unsigned long avail_ram;
49 	unsigned long mem_size = 0;
50 	bool res = false;
51 
52 	si_meminfo(&val);
53 
54 	/* only uses low memory */
55 	avail_ram = val.totalram - val.totalhigh;
56 
57 	/*
58 	 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
59 	 */
60 	if (type == FREE_NIDS) {
61 		mem_size = (nm_i->nid_cnt[FREE_NID] *
62 				sizeof(struct free_nid)) >> PAGE_SHIFT;
63 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
64 	} else if (type == NAT_ENTRIES) {
65 		mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
66 				sizeof(struct nat_entry)) >> PAGE_SHIFT;
67 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
68 		if (excess_cached_nats(sbi))
69 			res = false;
70 	} else if (type == DIRTY_DENTS) {
71 		if (sbi->sb->s_bdi->wb.dirty_exceeded)
72 			return false;
73 		mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
74 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
75 	} else if (type == INO_ENTRIES) {
76 		int i;
77 
78 		for (i = 0; i < MAX_INO_ENTRY; i++)
79 			mem_size += sbi->im[i].ino_num *
80 						sizeof(struct ino_entry);
81 		mem_size >>= PAGE_SHIFT;
82 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
83 	} else if (type == EXTENT_CACHE) {
84 		mem_size = (atomic_read(&sbi->total_ext_tree) *
85 				sizeof(struct extent_tree) +
86 				atomic_read(&sbi->total_ext_node) *
87 				sizeof(struct extent_node)) >> PAGE_SHIFT;
88 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
89 	} else if (type == INMEM_PAGES) {
90 		/* it allows 20% / total_ram for inmemory pages */
91 		mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
92 		res = mem_size < (val.totalram / 5);
93 	} else {
94 		if (!sbi->sb->s_bdi->wb.dirty_exceeded)
95 			return true;
96 	}
97 	return res;
98 }
99 
100 static void clear_node_page_dirty(struct page *page)
101 {
102 	if (PageDirty(page)) {
103 		f2fs_clear_page_cache_dirty_tag(page);
104 		clear_page_dirty_for_io(page);
105 		dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
106 	}
107 	ClearPageUptodate(page);
108 }
109 
110 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
111 {
112 	return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
113 }
114 
115 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
116 {
117 	struct page *src_page;
118 	struct page *dst_page;
119 	pgoff_t dst_off;
120 	void *src_addr;
121 	void *dst_addr;
122 	struct f2fs_nm_info *nm_i = NM_I(sbi);
123 
124 	dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
125 
126 	/* get current nat block page with lock */
127 	src_page = get_current_nat_page(sbi, nid);
128 	if (IS_ERR(src_page))
129 		return src_page;
130 	dst_page = f2fs_grab_meta_page(sbi, dst_off);
131 	f2fs_bug_on(sbi, PageDirty(src_page));
132 
133 	src_addr = page_address(src_page);
134 	dst_addr = page_address(dst_page);
135 	memcpy(dst_addr, src_addr, PAGE_SIZE);
136 	set_page_dirty(dst_page);
137 	f2fs_put_page(src_page, 1);
138 
139 	set_to_next_nat(nm_i, nid);
140 
141 	return dst_page;
142 }
143 
144 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
145 {
146 	struct nat_entry *new;
147 
148 	if (no_fail)
149 		new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
150 	else
151 		new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
152 	if (new) {
153 		nat_set_nid(new, nid);
154 		nat_reset_flag(new);
155 	}
156 	return new;
157 }
158 
159 static void __free_nat_entry(struct nat_entry *e)
160 {
161 	kmem_cache_free(nat_entry_slab, e);
162 }
163 
164 /* must be locked by nat_tree_lock */
165 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
166 	struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
167 {
168 	if (no_fail)
169 		f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
170 	else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
171 		return NULL;
172 
173 	if (raw_ne)
174 		node_info_from_raw_nat(&ne->ni, raw_ne);
175 
176 	spin_lock(&nm_i->nat_list_lock);
177 	list_add_tail(&ne->list, &nm_i->nat_entries);
178 	spin_unlock(&nm_i->nat_list_lock);
179 
180 	nm_i->nat_cnt[TOTAL_NAT]++;
181 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
182 	return ne;
183 }
184 
185 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
186 {
187 	struct nat_entry *ne;
188 
189 	ne = radix_tree_lookup(&nm_i->nat_root, n);
190 
191 	/* for recent accessed nat entry, move it to tail of lru list */
192 	if (ne && !get_nat_flag(ne, IS_DIRTY)) {
193 		spin_lock(&nm_i->nat_list_lock);
194 		if (!list_empty(&ne->list))
195 			list_move_tail(&ne->list, &nm_i->nat_entries);
196 		spin_unlock(&nm_i->nat_list_lock);
197 	}
198 
199 	return ne;
200 }
201 
202 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
203 		nid_t start, unsigned int nr, struct nat_entry **ep)
204 {
205 	return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
206 }
207 
208 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
209 {
210 	radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
211 	nm_i->nat_cnt[TOTAL_NAT]--;
212 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
213 	__free_nat_entry(e);
214 }
215 
216 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
217 							struct nat_entry *ne)
218 {
219 	nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
220 	struct nat_entry_set *head;
221 
222 	head = radix_tree_lookup(&nm_i->nat_set_root, set);
223 	if (!head) {
224 		head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
225 
226 		INIT_LIST_HEAD(&head->entry_list);
227 		INIT_LIST_HEAD(&head->set_list);
228 		head->set = set;
229 		head->entry_cnt = 0;
230 		f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
231 	}
232 	return head;
233 }
234 
235 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
236 						struct nat_entry *ne)
237 {
238 	struct nat_entry_set *head;
239 	bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
240 
241 	if (!new_ne)
242 		head = __grab_nat_entry_set(nm_i, ne);
243 
244 	/*
245 	 * update entry_cnt in below condition:
246 	 * 1. update NEW_ADDR to valid block address;
247 	 * 2. update old block address to new one;
248 	 */
249 	if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
250 				!get_nat_flag(ne, IS_DIRTY)))
251 		head->entry_cnt++;
252 
253 	set_nat_flag(ne, IS_PREALLOC, new_ne);
254 
255 	if (get_nat_flag(ne, IS_DIRTY))
256 		goto refresh_list;
257 
258 	nm_i->nat_cnt[DIRTY_NAT]++;
259 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
260 	set_nat_flag(ne, IS_DIRTY, true);
261 refresh_list:
262 	spin_lock(&nm_i->nat_list_lock);
263 	if (new_ne)
264 		list_del_init(&ne->list);
265 	else
266 		list_move_tail(&ne->list, &head->entry_list);
267 	spin_unlock(&nm_i->nat_list_lock);
268 }
269 
270 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
271 		struct nat_entry_set *set, struct nat_entry *ne)
272 {
273 	spin_lock(&nm_i->nat_list_lock);
274 	list_move_tail(&ne->list, &nm_i->nat_entries);
275 	spin_unlock(&nm_i->nat_list_lock);
276 
277 	set_nat_flag(ne, IS_DIRTY, false);
278 	set->entry_cnt--;
279 	nm_i->nat_cnt[DIRTY_NAT]--;
280 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
281 }
282 
283 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
284 		nid_t start, unsigned int nr, struct nat_entry_set **ep)
285 {
286 	return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
287 							start, nr);
288 }
289 
290 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
291 {
292 	return NODE_MAPPING(sbi) == page->mapping &&
293 			IS_DNODE(page) && is_cold_node(page);
294 }
295 
296 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
297 {
298 	spin_lock_init(&sbi->fsync_node_lock);
299 	INIT_LIST_HEAD(&sbi->fsync_node_list);
300 	sbi->fsync_seg_id = 0;
301 	sbi->fsync_node_num = 0;
302 }
303 
304 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
305 							struct page *page)
306 {
307 	struct fsync_node_entry *fn;
308 	unsigned long flags;
309 	unsigned int seq_id;
310 
311 	fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS);
312 
313 	get_page(page);
314 	fn->page = page;
315 	INIT_LIST_HEAD(&fn->list);
316 
317 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
318 	list_add_tail(&fn->list, &sbi->fsync_node_list);
319 	fn->seq_id = sbi->fsync_seg_id++;
320 	seq_id = fn->seq_id;
321 	sbi->fsync_node_num++;
322 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
323 
324 	return seq_id;
325 }
326 
327 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
328 {
329 	struct fsync_node_entry *fn;
330 	unsigned long flags;
331 
332 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
333 	list_for_each_entry(fn, &sbi->fsync_node_list, list) {
334 		if (fn->page == page) {
335 			list_del(&fn->list);
336 			sbi->fsync_node_num--;
337 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
338 			kmem_cache_free(fsync_node_entry_slab, fn);
339 			put_page(page);
340 			return;
341 		}
342 	}
343 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
344 	f2fs_bug_on(sbi, 1);
345 }
346 
347 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
348 {
349 	unsigned long flags;
350 
351 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
352 	sbi->fsync_seg_id = 0;
353 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
354 }
355 
356 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
357 {
358 	struct f2fs_nm_info *nm_i = NM_I(sbi);
359 	struct nat_entry *e;
360 	bool need = false;
361 
362 	down_read(&nm_i->nat_tree_lock);
363 	e = __lookup_nat_cache(nm_i, nid);
364 	if (e) {
365 		if (!get_nat_flag(e, IS_CHECKPOINTED) &&
366 				!get_nat_flag(e, HAS_FSYNCED_INODE))
367 			need = true;
368 	}
369 	up_read(&nm_i->nat_tree_lock);
370 	return need;
371 }
372 
373 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
374 {
375 	struct f2fs_nm_info *nm_i = NM_I(sbi);
376 	struct nat_entry *e;
377 	bool is_cp = true;
378 
379 	down_read(&nm_i->nat_tree_lock);
380 	e = __lookup_nat_cache(nm_i, nid);
381 	if (e && !get_nat_flag(e, IS_CHECKPOINTED))
382 		is_cp = false;
383 	up_read(&nm_i->nat_tree_lock);
384 	return is_cp;
385 }
386 
387 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
388 {
389 	struct f2fs_nm_info *nm_i = NM_I(sbi);
390 	struct nat_entry *e;
391 	bool need_update = true;
392 
393 	down_read(&nm_i->nat_tree_lock);
394 	e = __lookup_nat_cache(nm_i, ino);
395 	if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
396 			(get_nat_flag(e, IS_CHECKPOINTED) ||
397 			 get_nat_flag(e, HAS_FSYNCED_INODE)))
398 		need_update = false;
399 	up_read(&nm_i->nat_tree_lock);
400 	return need_update;
401 }
402 
403 /* must be locked by nat_tree_lock */
404 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
405 						struct f2fs_nat_entry *ne)
406 {
407 	struct f2fs_nm_info *nm_i = NM_I(sbi);
408 	struct nat_entry *new, *e;
409 
410 	new = __alloc_nat_entry(nid, false);
411 	if (!new)
412 		return;
413 
414 	down_write(&nm_i->nat_tree_lock);
415 	e = __lookup_nat_cache(nm_i, nid);
416 	if (!e)
417 		e = __init_nat_entry(nm_i, new, ne, false);
418 	else
419 		f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
420 				nat_get_blkaddr(e) !=
421 					le32_to_cpu(ne->block_addr) ||
422 				nat_get_version(e) != ne->version);
423 	up_write(&nm_i->nat_tree_lock);
424 	if (e != new)
425 		__free_nat_entry(new);
426 }
427 
428 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
429 			block_t new_blkaddr, bool fsync_done)
430 {
431 	struct f2fs_nm_info *nm_i = NM_I(sbi);
432 	struct nat_entry *e;
433 	struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
434 
435 	down_write(&nm_i->nat_tree_lock);
436 	e = __lookup_nat_cache(nm_i, ni->nid);
437 	if (!e) {
438 		e = __init_nat_entry(nm_i, new, NULL, true);
439 		copy_node_info(&e->ni, ni);
440 		f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
441 	} else if (new_blkaddr == NEW_ADDR) {
442 		/*
443 		 * when nid is reallocated,
444 		 * previous nat entry can be remained in nat cache.
445 		 * So, reinitialize it with new information.
446 		 */
447 		copy_node_info(&e->ni, ni);
448 		f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
449 	}
450 	/* let's free early to reduce memory consumption */
451 	if (e != new)
452 		__free_nat_entry(new);
453 
454 	/* sanity check */
455 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
456 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
457 			new_blkaddr == NULL_ADDR);
458 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
459 			new_blkaddr == NEW_ADDR);
460 	f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
461 			new_blkaddr == NEW_ADDR);
462 
463 	/* increment version no as node is removed */
464 	if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
465 		unsigned char version = nat_get_version(e);
466 		nat_set_version(e, inc_node_version(version));
467 	}
468 
469 	/* change address */
470 	nat_set_blkaddr(e, new_blkaddr);
471 	if (!__is_valid_data_blkaddr(new_blkaddr))
472 		set_nat_flag(e, IS_CHECKPOINTED, false);
473 	__set_nat_cache_dirty(nm_i, e);
474 
475 	/* update fsync_mark if its inode nat entry is still alive */
476 	if (ni->nid != ni->ino)
477 		e = __lookup_nat_cache(nm_i, ni->ino);
478 	if (e) {
479 		if (fsync_done && ni->nid == ni->ino)
480 			set_nat_flag(e, HAS_FSYNCED_INODE, true);
481 		set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
482 	}
483 	up_write(&nm_i->nat_tree_lock);
484 }
485 
486 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
487 {
488 	struct f2fs_nm_info *nm_i = NM_I(sbi);
489 	int nr = nr_shrink;
490 
491 	if (!down_write_trylock(&nm_i->nat_tree_lock))
492 		return 0;
493 
494 	spin_lock(&nm_i->nat_list_lock);
495 	while (nr_shrink) {
496 		struct nat_entry *ne;
497 
498 		if (list_empty(&nm_i->nat_entries))
499 			break;
500 
501 		ne = list_first_entry(&nm_i->nat_entries,
502 					struct nat_entry, list);
503 		list_del(&ne->list);
504 		spin_unlock(&nm_i->nat_list_lock);
505 
506 		__del_from_nat_cache(nm_i, ne);
507 		nr_shrink--;
508 
509 		spin_lock(&nm_i->nat_list_lock);
510 	}
511 	spin_unlock(&nm_i->nat_list_lock);
512 
513 	up_write(&nm_i->nat_tree_lock);
514 	return nr - nr_shrink;
515 }
516 
517 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
518 						struct node_info *ni)
519 {
520 	struct f2fs_nm_info *nm_i = NM_I(sbi);
521 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
522 	struct f2fs_journal *journal = curseg->journal;
523 	nid_t start_nid = START_NID(nid);
524 	struct f2fs_nat_block *nat_blk;
525 	struct page *page = NULL;
526 	struct f2fs_nat_entry ne;
527 	struct nat_entry *e;
528 	pgoff_t index;
529 	block_t blkaddr;
530 	int i;
531 
532 	ni->nid = nid;
533 
534 	/* Check nat cache */
535 	down_read(&nm_i->nat_tree_lock);
536 	e = __lookup_nat_cache(nm_i, nid);
537 	if (e) {
538 		ni->ino = nat_get_ino(e);
539 		ni->blk_addr = nat_get_blkaddr(e);
540 		ni->version = nat_get_version(e);
541 		up_read(&nm_i->nat_tree_lock);
542 		return 0;
543 	}
544 
545 	memset(&ne, 0, sizeof(struct f2fs_nat_entry));
546 
547 	/* Check current segment summary */
548 	down_read(&curseg->journal_rwsem);
549 	i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
550 	if (i >= 0) {
551 		ne = nat_in_journal(journal, i);
552 		node_info_from_raw_nat(ni, &ne);
553 	}
554 	up_read(&curseg->journal_rwsem);
555 	if (i >= 0) {
556 		up_read(&nm_i->nat_tree_lock);
557 		goto cache;
558 	}
559 
560 	/* Fill node_info from nat page */
561 	index = current_nat_addr(sbi, nid);
562 	up_read(&nm_i->nat_tree_lock);
563 
564 	page = f2fs_get_meta_page(sbi, index);
565 	if (IS_ERR(page))
566 		return PTR_ERR(page);
567 
568 	nat_blk = (struct f2fs_nat_block *)page_address(page);
569 	ne = nat_blk->entries[nid - start_nid];
570 	node_info_from_raw_nat(ni, &ne);
571 	f2fs_put_page(page, 1);
572 cache:
573 	blkaddr = le32_to_cpu(ne.block_addr);
574 	if (__is_valid_data_blkaddr(blkaddr) &&
575 		!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
576 		return -EFAULT;
577 
578 	/* cache nat entry */
579 	cache_nat_entry(sbi, nid, &ne);
580 	return 0;
581 }
582 
583 /*
584  * readahead MAX_RA_NODE number of node pages.
585  */
586 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
587 {
588 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
589 	struct blk_plug plug;
590 	int i, end;
591 	nid_t nid;
592 
593 	blk_start_plug(&plug);
594 
595 	/* Then, try readahead for siblings of the desired node */
596 	end = start + n;
597 	end = min(end, NIDS_PER_BLOCK);
598 	for (i = start; i < end; i++) {
599 		nid = get_nid(parent, i, false);
600 		f2fs_ra_node_page(sbi, nid);
601 	}
602 
603 	blk_finish_plug(&plug);
604 }
605 
606 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
607 {
608 	const long direct_index = ADDRS_PER_INODE(dn->inode);
609 	const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
610 	const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
611 	unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
612 	int cur_level = dn->cur_level;
613 	int max_level = dn->max_level;
614 	pgoff_t base = 0;
615 
616 	if (!dn->max_level)
617 		return pgofs + 1;
618 
619 	while (max_level-- > cur_level)
620 		skipped_unit *= NIDS_PER_BLOCK;
621 
622 	switch (dn->max_level) {
623 	case 3:
624 		base += 2 * indirect_blks;
625 		fallthrough;
626 	case 2:
627 		base += 2 * direct_blks;
628 		fallthrough;
629 	case 1:
630 		base += direct_index;
631 		break;
632 	default:
633 		f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
634 	}
635 
636 	return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
637 }
638 
639 /*
640  * The maximum depth is four.
641  * Offset[0] will have raw inode offset.
642  */
643 static int get_node_path(struct inode *inode, long block,
644 				int offset[4], unsigned int noffset[4])
645 {
646 	const long direct_index = ADDRS_PER_INODE(inode);
647 	const long direct_blks = ADDRS_PER_BLOCK(inode);
648 	const long dptrs_per_blk = NIDS_PER_BLOCK;
649 	const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
650 	const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
651 	int n = 0;
652 	int level = 0;
653 
654 	noffset[0] = 0;
655 
656 	if (block < direct_index) {
657 		offset[n] = block;
658 		goto got;
659 	}
660 	block -= direct_index;
661 	if (block < direct_blks) {
662 		offset[n++] = NODE_DIR1_BLOCK;
663 		noffset[n] = 1;
664 		offset[n] = block;
665 		level = 1;
666 		goto got;
667 	}
668 	block -= direct_blks;
669 	if (block < direct_blks) {
670 		offset[n++] = NODE_DIR2_BLOCK;
671 		noffset[n] = 2;
672 		offset[n] = block;
673 		level = 1;
674 		goto got;
675 	}
676 	block -= direct_blks;
677 	if (block < indirect_blks) {
678 		offset[n++] = NODE_IND1_BLOCK;
679 		noffset[n] = 3;
680 		offset[n++] = block / direct_blks;
681 		noffset[n] = 4 + offset[n - 1];
682 		offset[n] = block % direct_blks;
683 		level = 2;
684 		goto got;
685 	}
686 	block -= indirect_blks;
687 	if (block < indirect_blks) {
688 		offset[n++] = NODE_IND2_BLOCK;
689 		noffset[n] = 4 + dptrs_per_blk;
690 		offset[n++] = block / direct_blks;
691 		noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
692 		offset[n] = block % direct_blks;
693 		level = 2;
694 		goto got;
695 	}
696 	block -= indirect_blks;
697 	if (block < dindirect_blks) {
698 		offset[n++] = NODE_DIND_BLOCK;
699 		noffset[n] = 5 + (dptrs_per_blk * 2);
700 		offset[n++] = block / indirect_blks;
701 		noffset[n] = 6 + (dptrs_per_blk * 2) +
702 			      offset[n - 1] * (dptrs_per_blk + 1);
703 		offset[n++] = (block / direct_blks) % dptrs_per_blk;
704 		noffset[n] = 7 + (dptrs_per_blk * 2) +
705 			      offset[n - 2] * (dptrs_per_blk + 1) +
706 			      offset[n - 1];
707 		offset[n] = block % direct_blks;
708 		level = 3;
709 		goto got;
710 	} else {
711 		return -E2BIG;
712 	}
713 got:
714 	return level;
715 }
716 
717 /*
718  * Caller should call f2fs_put_dnode(dn).
719  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
720  * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
721  */
722 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
723 {
724 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
725 	struct page *npage[4];
726 	struct page *parent = NULL;
727 	int offset[4];
728 	unsigned int noffset[4];
729 	nid_t nids[4];
730 	int level, i = 0;
731 	int err = 0;
732 
733 	level = get_node_path(dn->inode, index, offset, noffset);
734 	if (level < 0)
735 		return level;
736 
737 	nids[0] = dn->inode->i_ino;
738 	npage[0] = dn->inode_page;
739 
740 	if (!npage[0]) {
741 		npage[0] = f2fs_get_node_page(sbi, nids[0]);
742 		if (IS_ERR(npage[0]))
743 			return PTR_ERR(npage[0]);
744 	}
745 
746 	/* if inline_data is set, should not report any block indices */
747 	if (f2fs_has_inline_data(dn->inode) && index) {
748 		err = -ENOENT;
749 		f2fs_put_page(npage[0], 1);
750 		goto release_out;
751 	}
752 
753 	parent = npage[0];
754 	if (level != 0)
755 		nids[1] = get_nid(parent, offset[0], true);
756 	dn->inode_page = npage[0];
757 	dn->inode_page_locked = true;
758 
759 	/* get indirect or direct nodes */
760 	for (i = 1; i <= level; i++) {
761 		bool done = false;
762 
763 		if (!nids[i] && mode == ALLOC_NODE) {
764 			/* alloc new node */
765 			if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
766 				err = -ENOSPC;
767 				goto release_pages;
768 			}
769 
770 			dn->nid = nids[i];
771 			npage[i] = f2fs_new_node_page(dn, noffset[i]);
772 			if (IS_ERR(npage[i])) {
773 				f2fs_alloc_nid_failed(sbi, nids[i]);
774 				err = PTR_ERR(npage[i]);
775 				goto release_pages;
776 			}
777 
778 			set_nid(parent, offset[i - 1], nids[i], i == 1);
779 			f2fs_alloc_nid_done(sbi, nids[i]);
780 			done = true;
781 		} else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
782 			npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
783 			if (IS_ERR(npage[i])) {
784 				err = PTR_ERR(npage[i]);
785 				goto release_pages;
786 			}
787 			done = true;
788 		}
789 		if (i == 1) {
790 			dn->inode_page_locked = false;
791 			unlock_page(parent);
792 		} else {
793 			f2fs_put_page(parent, 1);
794 		}
795 
796 		if (!done) {
797 			npage[i] = f2fs_get_node_page(sbi, nids[i]);
798 			if (IS_ERR(npage[i])) {
799 				err = PTR_ERR(npage[i]);
800 				f2fs_put_page(npage[0], 0);
801 				goto release_out;
802 			}
803 		}
804 		if (i < level) {
805 			parent = npage[i];
806 			nids[i + 1] = get_nid(parent, offset[i], false);
807 		}
808 	}
809 	dn->nid = nids[level];
810 	dn->ofs_in_node = offset[level];
811 	dn->node_page = npage[level];
812 	dn->data_blkaddr = f2fs_data_blkaddr(dn);
813 	return 0;
814 
815 release_pages:
816 	f2fs_put_page(parent, 1);
817 	if (i > 1)
818 		f2fs_put_page(npage[0], 0);
819 release_out:
820 	dn->inode_page = NULL;
821 	dn->node_page = NULL;
822 	if (err == -ENOENT) {
823 		dn->cur_level = i;
824 		dn->max_level = level;
825 		dn->ofs_in_node = offset[level];
826 	}
827 	return err;
828 }
829 
830 static int truncate_node(struct dnode_of_data *dn)
831 {
832 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
833 	struct node_info ni;
834 	int err;
835 	pgoff_t index;
836 
837 	err = f2fs_get_node_info(sbi, dn->nid, &ni);
838 	if (err)
839 		return err;
840 
841 	/* Deallocate node address */
842 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
843 	dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
844 	set_node_addr(sbi, &ni, NULL_ADDR, false);
845 
846 	if (dn->nid == dn->inode->i_ino) {
847 		f2fs_remove_orphan_inode(sbi, dn->nid);
848 		dec_valid_inode_count(sbi);
849 		f2fs_inode_synced(dn->inode);
850 	}
851 
852 	clear_node_page_dirty(dn->node_page);
853 	set_sbi_flag(sbi, SBI_IS_DIRTY);
854 
855 	index = dn->node_page->index;
856 	f2fs_put_page(dn->node_page, 1);
857 
858 	invalidate_mapping_pages(NODE_MAPPING(sbi),
859 			index, index);
860 
861 	dn->node_page = NULL;
862 	trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
863 
864 	return 0;
865 }
866 
867 static int truncate_dnode(struct dnode_of_data *dn)
868 {
869 	struct page *page;
870 	int err;
871 
872 	if (dn->nid == 0)
873 		return 1;
874 
875 	/* get direct node */
876 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
877 	if (PTR_ERR(page) == -ENOENT)
878 		return 1;
879 	else if (IS_ERR(page))
880 		return PTR_ERR(page);
881 
882 	/* Make dnode_of_data for parameter */
883 	dn->node_page = page;
884 	dn->ofs_in_node = 0;
885 	f2fs_truncate_data_blocks(dn);
886 	err = truncate_node(dn);
887 	if (err)
888 		return err;
889 
890 	return 1;
891 }
892 
893 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
894 						int ofs, int depth)
895 {
896 	struct dnode_of_data rdn = *dn;
897 	struct page *page;
898 	struct f2fs_node *rn;
899 	nid_t child_nid;
900 	unsigned int child_nofs;
901 	int freed = 0;
902 	int i, ret;
903 
904 	if (dn->nid == 0)
905 		return NIDS_PER_BLOCK + 1;
906 
907 	trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
908 
909 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
910 	if (IS_ERR(page)) {
911 		trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
912 		return PTR_ERR(page);
913 	}
914 
915 	f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
916 
917 	rn = F2FS_NODE(page);
918 	if (depth < 3) {
919 		for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
920 			child_nid = le32_to_cpu(rn->in.nid[i]);
921 			if (child_nid == 0)
922 				continue;
923 			rdn.nid = child_nid;
924 			ret = truncate_dnode(&rdn);
925 			if (ret < 0)
926 				goto out_err;
927 			if (set_nid(page, i, 0, false))
928 				dn->node_changed = true;
929 		}
930 	} else {
931 		child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
932 		for (i = ofs; i < NIDS_PER_BLOCK; i++) {
933 			child_nid = le32_to_cpu(rn->in.nid[i]);
934 			if (child_nid == 0) {
935 				child_nofs += NIDS_PER_BLOCK + 1;
936 				continue;
937 			}
938 			rdn.nid = child_nid;
939 			ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
940 			if (ret == (NIDS_PER_BLOCK + 1)) {
941 				if (set_nid(page, i, 0, false))
942 					dn->node_changed = true;
943 				child_nofs += ret;
944 			} else if (ret < 0 && ret != -ENOENT) {
945 				goto out_err;
946 			}
947 		}
948 		freed = child_nofs;
949 	}
950 
951 	if (!ofs) {
952 		/* remove current indirect node */
953 		dn->node_page = page;
954 		ret = truncate_node(dn);
955 		if (ret)
956 			goto out_err;
957 		freed++;
958 	} else {
959 		f2fs_put_page(page, 1);
960 	}
961 	trace_f2fs_truncate_nodes_exit(dn->inode, freed);
962 	return freed;
963 
964 out_err:
965 	f2fs_put_page(page, 1);
966 	trace_f2fs_truncate_nodes_exit(dn->inode, ret);
967 	return ret;
968 }
969 
970 static int truncate_partial_nodes(struct dnode_of_data *dn,
971 			struct f2fs_inode *ri, int *offset, int depth)
972 {
973 	struct page *pages[2];
974 	nid_t nid[3];
975 	nid_t child_nid;
976 	int err = 0;
977 	int i;
978 	int idx = depth - 2;
979 
980 	nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
981 	if (!nid[0])
982 		return 0;
983 
984 	/* get indirect nodes in the path */
985 	for (i = 0; i < idx + 1; i++) {
986 		/* reference count'll be increased */
987 		pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
988 		if (IS_ERR(pages[i])) {
989 			err = PTR_ERR(pages[i]);
990 			idx = i - 1;
991 			goto fail;
992 		}
993 		nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
994 	}
995 
996 	f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
997 
998 	/* free direct nodes linked to a partial indirect node */
999 	for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1000 		child_nid = get_nid(pages[idx], i, false);
1001 		if (!child_nid)
1002 			continue;
1003 		dn->nid = child_nid;
1004 		err = truncate_dnode(dn);
1005 		if (err < 0)
1006 			goto fail;
1007 		if (set_nid(pages[idx], i, 0, false))
1008 			dn->node_changed = true;
1009 	}
1010 
1011 	if (offset[idx + 1] == 0) {
1012 		dn->node_page = pages[idx];
1013 		dn->nid = nid[idx];
1014 		err = truncate_node(dn);
1015 		if (err)
1016 			goto fail;
1017 	} else {
1018 		f2fs_put_page(pages[idx], 1);
1019 	}
1020 	offset[idx]++;
1021 	offset[idx + 1] = 0;
1022 	idx--;
1023 fail:
1024 	for (i = idx; i >= 0; i--)
1025 		f2fs_put_page(pages[i], 1);
1026 
1027 	trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1028 
1029 	return err;
1030 }
1031 
1032 /*
1033  * All the block addresses of data and nodes should be nullified.
1034  */
1035 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1036 {
1037 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1038 	int err = 0, cont = 1;
1039 	int level, offset[4], noffset[4];
1040 	unsigned int nofs = 0;
1041 	struct f2fs_inode *ri;
1042 	struct dnode_of_data dn;
1043 	struct page *page;
1044 
1045 	trace_f2fs_truncate_inode_blocks_enter(inode, from);
1046 
1047 	level = get_node_path(inode, from, offset, noffset);
1048 	if (level < 0) {
1049 		trace_f2fs_truncate_inode_blocks_exit(inode, level);
1050 		return level;
1051 	}
1052 
1053 	page = f2fs_get_node_page(sbi, inode->i_ino);
1054 	if (IS_ERR(page)) {
1055 		trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1056 		return PTR_ERR(page);
1057 	}
1058 
1059 	set_new_dnode(&dn, inode, page, NULL, 0);
1060 	unlock_page(page);
1061 
1062 	ri = F2FS_INODE(page);
1063 	switch (level) {
1064 	case 0:
1065 	case 1:
1066 		nofs = noffset[1];
1067 		break;
1068 	case 2:
1069 		nofs = noffset[1];
1070 		if (!offset[level - 1])
1071 			goto skip_partial;
1072 		err = truncate_partial_nodes(&dn, ri, offset, level);
1073 		if (err < 0 && err != -ENOENT)
1074 			goto fail;
1075 		nofs += 1 + NIDS_PER_BLOCK;
1076 		break;
1077 	case 3:
1078 		nofs = 5 + 2 * NIDS_PER_BLOCK;
1079 		if (!offset[level - 1])
1080 			goto skip_partial;
1081 		err = truncate_partial_nodes(&dn, ri, offset, level);
1082 		if (err < 0 && err != -ENOENT)
1083 			goto fail;
1084 		break;
1085 	default:
1086 		BUG();
1087 	}
1088 
1089 skip_partial:
1090 	while (cont) {
1091 		dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1092 		switch (offset[0]) {
1093 		case NODE_DIR1_BLOCK:
1094 		case NODE_DIR2_BLOCK:
1095 			err = truncate_dnode(&dn);
1096 			break;
1097 
1098 		case NODE_IND1_BLOCK:
1099 		case NODE_IND2_BLOCK:
1100 			err = truncate_nodes(&dn, nofs, offset[1], 2);
1101 			break;
1102 
1103 		case NODE_DIND_BLOCK:
1104 			err = truncate_nodes(&dn, nofs, offset[1], 3);
1105 			cont = 0;
1106 			break;
1107 
1108 		default:
1109 			BUG();
1110 		}
1111 		if (err < 0 && err != -ENOENT)
1112 			goto fail;
1113 		if (offset[1] == 0 &&
1114 				ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1115 			lock_page(page);
1116 			BUG_ON(page->mapping != NODE_MAPPING(sbi));
1117 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1118 			ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1119 			set_page_dirty(page);
1120 			unlock_page(page);
1121 		}
1122 		offset[1] = 0;
1123 		offset[0]++;
1124 		nofs += err;
1125 	}
1126 fail:
1127 	f2fs_put_page(page, 0);
1128 	trace_f2fs_truncate_inode_blocks_exit(inode, err);
1129 	return err > 0 ? 0 : err;
1130 }
1131 
1132 /* caller must lock inode page */
1133 int f2fs_truncate_xattr_node(struct inode *inode)
1134 {
1135 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1136 	nid_t nid = F2FS_I(inode)->i_xattr_nid;
1137 	struct dnode_of_data dn;
1138 	struct page *npage;
1139 	int err;
1140 
1141 	if (!nid)
1142 		return 0;
1143 
1144 	npage = f2fs_get_node_page(sbi, nid);
1145 	if (IS_ERR(npage))
1146 		return PTR_ERR(npage);
1147 
1148 	set_new_dnode(&dn, inode, NULL, npage, nid);
1149 	err = truncate_node(&dn);
1150 	if (err) {
1151 		f2fs_put_page(npage, 1);
1152 		return err;
1153 	}
1154 
1155 	f2fs_i_xnid_write(inode, 0);
1156 
1157 	return 0;
1158 }
1159 
1160 /*
1161  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1162  * f2fs_unlock_op().
1163  */
1164 int f2fs_remove_inode_page(struct inode *inode)
1165 {
1166 	struct dnode_of_data dn;
1167 	int err;
1168 
1169 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1170 	err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1171 	if (err)
1172 		return err;
1173 
1174 	err = f2fs_truncate_xattr_node(inode);
1175 	if (err) {
1176 		f2fs_put_dnode(&dn);
1177 		return err;
1178 	}
1179 
1180 	/* remove potential inline_data blocks */
1181 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1182 				S_ISLNK(inode->i_mode))
1183 		f2fs_truncate_data_blocks_range(&dn, 1);
1184 
1185 	/* 0 is possible, after f2fs_new_inode() has failed */
1186 	if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1187 		f2fs_put_dnode(&dn);
1188 		return -EIO;
1189 	}
1190 
1191 	if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1192 		f2fs_warn(F2FS_I_SB(inode),
1193 			"f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1194 			inode->i_ino, (unsigned long long)inode->i_blocks);
1195 		set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1196 	}
1197 
1198 	/* will put inode & node pages */
1199 	err = truncate_node(&dn);
1200 	if (err) {
1201 		f2fs_put_dnode(&dn);
1202 		return err;
1203 	}
1204 	return 0;
1205 }
1206 
1207 struct page *f2fs_new_inode_page(struct inode *inode)
1208 {
1209 	struct dnode_of_data dn;
1210 
1211 	/* allocate inode page for new inode */
1212 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1213 
1214 	/* caller should f2fs_put_page(page, 1); */
1215 	return f2fs_new_node_page(&dn, 0);
1216 }
1217 
1218 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1219 {
1220 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1221 	struct node_info new_ni;
1222 	struct page *page;
1223 	int err;
1224 
1225 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1226 		return ERR_PTR(-EPERM);
1227 
1228 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1229 	if (!page)
1230 		return ERR_PTR(-ENOMEM);
1231 
1232 	if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1233 		goto fail;
1234 
1235 #ifdef CONFIG_F2FS_CHECK_FS
1236 	err = f2fs_get_node_info(sbi, dn->nid, &new_ni);
1237 	if (err) {
1238 		dec_valid_node_count(sbi, dn->inode, !ofs);
1239 		goto fail;
1240 	}
1241 	f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1242 #endif
1243 	new_ni.nid = dn->nid;
1244 	new_ni.ino = dn->inode->i_ino;
1245 	new_ni.blk_addr = NULL_ADDR;
1246 	new_ni.flag = 0;
1247 	new_ni.version = 0;
1248 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1249 
1250 	f2fs_wait_on_page_writeback(page, NODE, true, true);
1251 	fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1252 	set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1253 	if (!PageUptodate(page))
1254 		SetPageUptodate(page);
1255 	if (set_page_dirty(page))
1256 		dn->node_changed = true;
1257 
1258 	if (f2fs_has_xattr_block(ofs))
1259 		f2fs_i_xnid_write(dn->inode, dn->nid);
1260 
1261 	if (ofs == 0)
1262 		inc_valid_inode_count(sbi);
1263 	return page;
1264 
1265 fail:
1266 	clear_node_page_dirty(page);
1267 	f2fs_put_page(page, 1);
1268 	return ERR_PTR(err);
1269 }
1270 
1271 /*
1272  * Caller should do after getting the following values.
1273  * 0: f2fs_put_page(page, 0)
1274  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1275  */
1276 static int read_node_page(struct page *page, int op_flags)
1277 {
1278 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1279 	struct node_info ni;
1280 	struct f2fs_io_info fio = {
1281 		.sbi = sbi,
1282 		.type = NODE,
1283 		.op = REQ_OP_READ,
1284 		.op_flags = op_flags,
1285 		.page = page,
1286 		.encrypted_page = NULL,
1287 	};
1288 	int err;
1289 
1290 	if (PageUptodate(page)) {
1291 		if (!f2fs_inode_chksum_verify(sbi, page)) {
1292 			ClearPageUptodate(page);
1293 			return -EFSBADCRC;
1294 		}
1295 		return LOCKED_PAGE;
1296 	}
1297 
1298 	err = f2fs_get_node_info(sbi, page->index, &ni);
1299 	if (err)
1300 		return err;
1301 
1302 	if (unlikely(ni.blk_addr == NULL_ADDR) ||
1303 			is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1304 		ClearPageUptodate(page);
1305 		return -ENOENT;
1306 	}
1307 
1308 	fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1309 
1310 	err = f2fs_submit_page_bio(&fio);
1311 
1312 	if (!err)
1313 		f2fs_update_iostat(sbi, FS_NODE_READ_IO, F2FS_BLKSIZE);
1314 
1315 	return err;
1316 }
1317 
1318 /*
1319  * Readahead a node page
1320  */
1321 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1322 {
1323 	struct page *apage;
1324 	int err;
1325 
1326 	if (!nid)
1327 		return;
1328 	if (f2fs_check_nid_range(sbi, nid))
1329 		return;
1330 
1331 	apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1332 	if (apage)
1333 		return;
1334 
1335 	apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1336 	if (!apage)
1337 		return;
1338 
1339 	err = read_node_page(apage, REQ_RAHEAD);
1340 	f2fs_put_page(apage, err ? 1 : 0);
1341 }
1342 
1343 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1344 					struct page *parent, int start)
1345 {
1346 	struct page *page;
1347 	int err;
1348 
1349 	if (!nid)
1350 		return ERR_PTR(-ENOENT);
1351 	if (f2fs_check_nid_range(sbi, nid))
1352 		return ERR_PTR(-EINVAL);
1353 repeat:
1354 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1355 	if (!page)
1356 		return ERR_PTR(-ENOMEM);
1357 
1358 	err = read_node_page(page, 0);
1359 	if (err < 0) {
1360 		f2fs_put_page(page, 1);
1361 		return ERR_PTR(err);
1362 	} else if (err == LOCKED_PAGE) {
1363 		err = 0;
1364 		goto page_hit;
1365 	}
1366 
1367 	if (parent)
1368 		f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1369 
1370 	lock_page(page);
1371 
1372 	if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1373 		f2fs_put_page(page, 1);
1374 		goto repeat;
1375 	}
1376 
1377 	if (unlikely(!PageUptodate(page))) {
1378 		err = -EIO;
1379 		goto out_err;
1380 	}
1381 
1382 	if (!f2fs_inode_chksum_verify(sbi, page)) {
1383 		err = -EFSBADCRC;
1384 		goto out_err;
1385 	}
1386 page_hit:
1387 	if(unlikely(nid != nid_of_node(page))) {
1388 		f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1389 			  nid, nid_of_node(page), ino_of_node(page),
1390 			  ofs_of_node(page), cpver_of_node(page),
1391 			  next_blkaddr_of_node(page));
1392 		err = -EINVAL;
1393 out_err:
1394 		ClearPageUptodate(page);
1395 		f2fs_put_page(page, 1);
1396 		return ERR_PTR(err);
1397 	}
1398 	return page;
1399 }
1400 
1401 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1402 {
1403 	return __get_node_page(sbi, nid, NULL, 0);
1404 }
1405 
1406 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1407 {
1408 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1409 	nid_t nid = get_nid(parent, start, false);
1410 
1411 	return __get_node_page(sbi, nid, parent, start);
1412 }
1413 
1414 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1415 {
1416 	struct inode *inode;
1417 	struct page *page;
1418 	int ret;
1419 
1420 	/* should flush inline_data before evict_inode */
1421 	inode = ilookup(sbi->sb, ino);
1422 	if (!inode)
1423 		return;
1424 
1425 	page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1426 					FGP_LOCK|FGP_NOWAIT, 0);
1427 	if (!page)
1428 		goto iput_out;
1429 
1430 	if (!PageUptodate(page))
1431 		goto page_out;
1432 
1433 	if (!PageDirty(page))
1434 		goto page_out;
1435 
1436 	if (!clear_page_dirty_for_io(page))
1437 		goto page_out;
1438 
1439 	ret = f2fs_write_inline_data(inode, page);
1440 	inode_dec_dirty_pages(inode);
1441 	f2fs_remove_dirty_inode(inode);
1442 	if (ret)
1443 		set_page_dirty(page);
1444 page_out:
1445 	f2fs_put_page(page, 1);
1446 iput_out:
1447 	iput(inode);
1448 }
1449 
1450 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1451 {
1452 	pgoff_t index;
1453 	struct pagevec pvec;
1454 	struct page *last_page = NULL;
1455 	int nr_pages;
1456 
1457 	pagevec_init(&pvec);
1458 	index = 0;
1459 
1460 	while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1461 				PAGECACHE_TAG_DIRTY))) {
1462 		int i;
1463 
1464 		for (i = 0; i < nr_pages; i++) {
1465 			struct page *page = pvec.pages[i];
1466 
1467 			if (unlikely(f2fs_cp_error(sbi))) {
1468 				f2fs_put_page(last_page, 0);
1469 				pagevec_release(&pvec);
1470 				return ERR_PTR(-EIO);
1471 			}
1472 
1473 			if (!IS_DNODE(page) || !is_cold_node(page))
1474 				continue;
1475 			if (ino_of_node(page) != ino)
1476 				continue;
1477 
1478 			lock_page(page);
1479 
1480 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1481 continue_unlock:
1482 				unlock_page(page);
1483 				continue;
1484 			}
1485 			if (ino_of_node(page) != ino)
1486 				goto continue_unlock;
1487 
1488 			if (!PageDirty(page)) {
1489 				/* someone wrote it for us */
1490 				goto continue_unlock;
1491 			}
1492 
1493 			if (last_page)
1494 				f2fs_put_page(last_page, 0);
1495 
1496 			get_page(page);
1497 			last_page = page;
1498 			unlock_page(page);
1499 		}
1500 		pagevec_release(&pvec);
1501 		cond_resched();
1502 	}
1503 	return last_page;
1504 }
1505 
1506 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1507 				struct writeback_control *wbc, bool do_balance,
1508 				enum iostat_type io_type, unsigned int *seq_id)
1509 {
1510 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1511 	nid_t nid;
1512 	struct node_info ni;
1513 	struct f2fs_io_info fio = {
1514 		.sbi = sbi,
1515 		.ino = ino_of_node(page),
1516 		.type = NODE,
1517 		.op = REQ_OP_WRITE,
1518 		.op_flags = wbc_to_write_flags(wbc),
1519 		.page = page,
1520 		.encrypted_page = NULL,
1521 		.submitted = false,
1522 		.io_type = io_type,
1523 		.io_wbc = wbc,
1524 	};
1525 	unsigned int seq;
1526 
1527 	trace_f2fs_writepage(page, NODE);
1528 
1529 	if (unlikely(f2fs_cp_error(sbi))) {
1530 		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
1531 			ClearPageUptodate(page);
1532 			dec_page_count(sbi, F2FS_DIRTY_NODES);
1533 			unlock_page(page);
1534 			return 0;
1535 		}
1536 		goto redirty_out;
1537 	}
1538 
1539 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1540 		goto redirty_out;
1541 
1542 	if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1543 			wbc->sync_mode == WB_SYNC_NONE &&
1544 			IS_DNODE(page) && is_cold_node(page))
1545 		goto redirty_out;
1546 
1547 	/* get old block addr of this node page */
1548 	nid = nid_of_node(page);
1549 	f2fs_bug_on(sbi, page->index != nid);
1550 
1551 	if (f2fs_get_node_info(sbi, nid, &ni))
1552 		goto redirty_out;
1553 
1554 	if (wbc->for_reclaim) {
1555 		if (!down_read_trylock(&sbi->node_write))
1556 			goto redirty_out;
1557 	} else {
1558 		down_read(&sbi->node_write);
1559 	}
1560 
1561 	/* This page is already truncated */
1562 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1563 		ClearPageUptodate(page);
1564 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1565 		up_read(&sbi->node_write);
1566 		unlock_page(page);
1567 		return 0;
1568 	}
1569 
1570 	if (__is_valid_data_blkaddr(ni.blk_addr) &&
1571 		!f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1572 					DATA_GENERIC_ENHANCE)) {
1573 		up_read(&sbi->node_write);
1574 		goto redirty_out;
1575 	}
1576 
1577 	if (atomic && !test_opt(sbi, NOBARRIER))
1578 		fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1579 
1580 	/* should add to global list before clearing PAGECACHE status */
1581 	if (f2fs_in_warm_node_list(sbi, page)) {
1582 		seq = f2fs_add_fsync_node_entry(sbi, page);
1583 		if (seq_id)
1584 			*seq_id = seq;
1585 	}
1586 
1587 	set_page_writeback(page);
1588 	ClearPageError(page);
1589 
1590 	fio.old_blkaddr = ni.blk_addr;
1591 	f2fs_do_write_node_page(nid, &fio);
1592 	set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1593 	dec_page_count(sbi, F2FS_DIRTY_NODES);
1594 	up_read(&sbi->node_write);
1595 
1596 	if (wbc->for_reclaim) {
1597 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1598 		submitted = NULL;
1599 	}
1600 
1601 	unlock_page(page);
1602 
1603 	if (unlikely(f2fs_cp_error(sbi))) {
1604 		f2fs_submit_merged_write(sbi, NODE);
1605 		submitted = NULL;
1606 	}
1607 	if (submitted)
1608 		*submitted = fio.submitted;
1609 
1610 	if (do_balance)
1611 		f2fs_balance_fs(sbi, false);
1612 	return 0;
1613 
1614 redirty_out:
1615 	redirty_page_for_writepage(wbc, page);
1616 	return AOP_WRITEPAGE_ACTIVATE;
1617 }
1618 
1619 int f2fs_move_node_page(struct page *node_page, int gc_type)
1620 {
1621 	int err = 0;
1622 
1623 	if (gc_type == FG_GC) {
1624 		struct writeback_control wbc = {
1625 			.sync_mode = WB_SYNC_ALL,
1626 			.nr_to_write = 1,
1627 			.for_reclaim = 0,
1628 		};
1629 
1630 		f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1631 
1632 		set_page_dirty(node_page);
1633 
1634 		if (!clear_page_dirty_for_io(node_page)) {
1635 			err = -EAGAIN;
1636 			goto out_page;
1637 		}
1638 
1639 		if (__write_node_page(node_page, false, NULL,
1640 					&wbc, false, FS_GC_NODE_IO, NULL)) {
1641 			err = -EAGAIN;
1642 			unlock_page(node_page);
1643 		}
1644 		goto release_page;
1645 	} else {
1646 		/* set page dirty and write it */
1647 		if (!PageWriteback(node_page))
1648 			set_page_dirty(node_page);
1649 	}
1650 out_page:
1651 	unlock_page(node_page);
1652 release_page:
1653 	f2fs_put_page(node_page, 0);
1654 	return err;
1655 }
1656 
1657 static int f2fs_write_node_page(struct page *page,
1658 				struct writeback_control *wbc)
1659 {
1660 	return __write_node_page(page, false, NULL, wbc, false,
1661 						FS_NODE_IO, NULL);
1662 }
1663 
1664 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1665 			struct writeback_control *wbc, bool atomic,
1666 			unsigned int *seq_id)
1667 {
1668 	pgoff_t index;
1669 	struct pagevec pvec;
1670 	int ret = 0;
1671 	struct page *last_page = NULL;
1672 	bool marked = false;
1673 	nid_t ino = inode->i_ino;
1674 	int nr_pages;
1675 	int nwritten = 0;
1676 
1677 	if (atomic) {
1678 		last_page = last_fsync_dnode(sbi, ino);
1679 		if (IS_ERR_OR_NULL(last_page))
1680 			return PTR_ERR_OR_ZERO(last_page);
1681 	}
1682 retry:
1683 	pagevec_init(&pvec);
1684 	index = 0;
1685 
1686 	while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1687 				PAGECACHE_TAG_DIRTY))) {
1688 		int i;
1689 
1690 		for (i = 0; i < nr_pages; i++) {
1691 			struct page *page = pvec.pages[i];
1692 			bool submitted = false;
1693 
1694 			if (unlikely(f2fs_cp_error(sbi))) {
1695 				f2fs_put_page(last_page, 0);
1696 				pagevec_release(&pvec);
1697 				ret = -EIO;
1698 				goto out;
1699 			}
1700 
1701 			if (!IS_DNODE(page) || !is_cold_node(page))
1702 				continue;
1703 			if (ino_of_node(page) != ino)
1704 				continue;
1705 
1706 			lock_page(page);
1707 
1708 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1709 continue_unlock:
1710 				unlock_page(page);
1711 				continue;
1712 			}
1713 			if (ino_of_node(page) != ino)
1714 				goto continue_unlock;
1715 
1716 			if (!PageDirty(page) && page != last_page) {
1717 				/* someone wrote it for us */
1718 				goto continue_unlock;
1719 			}
1720 
1721 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1722 
1723 			set_fsync_mark(page, 0);
1724 			set_dentry_mark(page, 0);
1725 
1726 			if (!atomic || page == last_page) {
1727 				set_fsync_mark(page, 1);
1728 				if (IS_INODE(page)) {
1729 					if (is_inode_flag_set(inode,
1730 								FI_DIRTY_INODE))
1731 						f2fs_update_inode(inode, page);
1732 					set_dentry_mark(page,
1733 						f2fs_need_dentry_mark(sbi, ino));
1734 				}
1735 				/* may be written by other thread */
1736 				if (!PageDirty(page))
1737 					set_page_dirty(page);
1738 			}
1739 
1740 			if (!clear_page_dirty_for_io(page))
1741 				goto continue_unlock;
1742 
1743 			ret = __write_node_page(page, atomic &&
1744 						page == last_page,
1745 						&submitted, wbc, true,
1746 						FS_NODE_IO, seq_id);
1747 			if (ret) {
1748 				unlock_page(page);
1749 				f2fs_put_page(last_page, 0);
1750 				break;
1751 			} else if (submitted) {
1752 				nwritten++;
1753 			}
1754 
1755 			if (page == last_page) {
1756 				f2fs_put_page(page, 0);
1757 				marked = true;
1758 				break;
1759 			}
1760 		}
1761 		pagevec_release(&pvec);
1762 		cond_resched();
1763 
1764 		if (ret || marked)
1765 			break;
1766 	}
1767 	if (!ret && atomic && !marked) {
1768 		f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1769 			   ino, last_page->index);
1770 		lock_page(last_page);
1771 		f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1772 		set_page_dirty(last_page);
1773 		unlock_page(last_page);
1774 		goto retry;
1775 	}
1776 out:
1777 	if (nwritten)
1778 		f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1779 	return ret ? -EIO: 0;
1780 }
1781 
1782 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1783 {
1784 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1785 	bool clean;
1786 
1787 	if (inode->i_ino != ino)
1788 		return 0;
1789 
1790 	if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1791 		return 0;
1792 
1793 	spin_lock(&sbi->inode_lock[DIRTY_META]);
1794 	clean = list_empty(&F2FS_I(inode)->gdirty_list);
1795 	spin_unlock(&sbi->inode_lock[DIRTY_META]);
1796 
1797 	if (clean)
1798 		return 0;
1799 
1800 	inode = igrab(inode);
1801 	if (!inode)
1802 		return 0;
1803 	return 1;
1804 }
1805 
1806 static bool flush_dirty_inode(struct page *page)
1807 {
1808 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1809 	struct inode *inode;
1810 	nid_t ino = ino_of_node(page);
1811 
1812 	inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1813 	if (!inode)
1814 		return false;
1815 
1816 	f2fs_update_inode(inode, page);
1817 	unlock_page(page);
1818 
1819 	iput(inode);
1820 	return true;
1821 }
1822 
1823 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1824 {
1825 	pgoff_t index = 0;
1826 	struct pagevec pvec;
1827 	int nr_pages;
1828 
1829 	pagevec_init(&pvec);
1830 
1831 	while ((nr_pages = pagevec_lookup_tag(&pvec,
1832 			NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1833 		int i;
1834 
1835 		for (i = 0; i < nr_pages; i++) {
1836 			struct page *page = pvec.pages[i];
1837 
1838 			if (!IS_DNODE(page))
1839 				continue;
1840 
1841 			lock_page(page);
1842 
1843 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1844 continue_unlock:
1845 				unlock_page(page);
1846 				continue;
1847 			}
1848 
1849 			if (!PageDirty(page)) {
1850 				/* someone wrote it for us */
1851 				goto continue_unlock;
1852 			}
1853 
1854 			/* flush inline_data, if it's async context. */
1855 			if (is_inline_node(page)) {
1856 				clear_inline_node(page);
1857 				unlock_page(page);
1858 				flush_inline_data(sbi, ino_of_node(page));
1859 				continue;
1860 			}
1861 			unlock_page(page);
1862 		}
1863 		pagevec_release(&pvec);
1864 		cond_resched();
1865 	}
1866 }
1867 
1868 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1869 				struct writeback_control *wbc,
1870 				bool do_balance, enum iostat_type io_type)
1871 {
1872 	pgoff_t index;
1873 	struct pagevec pvec;
1874 	int step = 0;
1875 	int nwritten = 0;
1876 	int ret = 0;
1877 	int nr_pages, done = 0;
1878 
1879 	pagevec_init(&pvec);
1880 
1881 next_step:
1882 	index = 0;
1883 
1884 	while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1885 			NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1886 		int i;
1887 
1888 		for (i = 0; i < nr_pages; i++) {
1889 			struct page *page = pvec.pages[i];
1890 			bool submitted = false;
1891 			bool may_dirty = true;
1892 
1893 			/* give a priority to WB_SYNC threads */
1894 			if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1895 					wbc->sync_mode == WB_SYNC_NONE) {
1896 				done = 1;
1897 				break;
1898 			}
1899 
1900 			/*
1901 			 * flushing sequence with step:
1902 			 * 0. indirect nodes
1903 			 * 1. dentry dnodes
1904 			 * 2. file dnodes
1905 			 */
1906 			if (step == 0 && IS_DNODE(page))
1907 				continue;
1908 			if (step == 1 && (!IS_DNODE(page) ||
1909 						is_cold_node(page)))
1910 				continue;
1911 			if (step == 2 && (!IS_DNODE(page) ||
1912 						!is_cold_node(page)))
1913 				continue;
1914 lock_node:
1915 			if (wbc->sync_mode == WB_SYNC_ALL)
1916 				lock_page(page);
1917 			else if (!trylock_page(page))
1918 				continue;
1919 
1920 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1921 continue_unlock:
1922 				unlock_page(page);
1923 				continue;
1924 			}
1925 
1926 			if (!PageDirty(page)) {
1927 				/* someone wrote it for us */
1928 				goto continue_unlock;
1929 			}
1930 
1931 			/* flush inline_data/inode, if it's async context. */
1932 			if (!do_balance)
1933 				goto write_node;
1934 
1935 			/* flush inline_data */
1936 			if (is_inline_node(page)) {
1937 				clear_inline_node(page);
1938 				unlock_page(page);
1939 				flush_inline_data(sbi, ino_of_node(page));
1940 				goto lock_node;
1941 			}
1942 
1943 			/* flush dirty inode */
1944 			if (IS_INODE(page) && may_dirty) {
1945 				may_dirty = false;
1946 				if (flush_dirty_inode(page))
1947 					goto lock_node;
1948 			}
1949 write_node:
1950 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1951 
1952 			if (!clear_page_dirty_for_io(page))
1953 				goto continue_unlock;
1954 
1955 			set_fsync_mark(page, 0);
1956 			set_dentry_mark(page, 0);
1957 
1958 			ret = __write_node_page(page, false, &submitted,
1959 						wbc, do_balance, io_type, NULL);
1960 			if (ret)
1961 				unlock_page(page);
1962 			else if (submitted)
1963 				nwritten++;
1964 
1965 			if (--wbc->nr_to_write == 0)
1966 				break;
1967 		}
1968 		pagevec_release(&pvec);
1969 		cond_resched();
1970 
1971 		if (wbc->nr_to_write == 0) {
1972 			step = 2;
1973 			break;
1974 		}
1975 	}
1976 
1977 	if (step < 2) {
1978 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1979 				wbc->sync_mode == WB_SYNC_NONE && step == 1)
1980 			goto out;
1981 		step++;
1982 		goto next_step;
1983 	}
1984 out:
1985 	if (nwritten)
1986 		f2fs_submit_merged_write(sbi, NODE);
1987 
1988 	if (unlikely(f2fs_cp_error(sbi)))
1989 		return -EIO;
1990 	return ret;
1991 }
1992 
1993 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
1994 						unsigned int seq_id)
1995 {
1996 	struct fsync_node_entry *fn;
1997 	struct page *page;
1998 	struct list_head *head = &sbi->fsync_node_list;
1999 	unsigned long flags;
2000 	unsigned int cur_seq_id = 0;
2001 	int ret2, ret = 0;
2002 
2003 	while (seq_id && cur_seq_id < seq_id) {
2004 		spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2005 		if (list_empty(head)) {
2006 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2007 			break;
2008 		}
2009 		fn = list_first_entry(head, struct fsync_node_entry, list);
2010 		if (fn->seq_id > seq_id) {
2011 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2012 			break;
2013 		}
2014 		cur_seq_id = fn->seq_id;
2015 		page = fn->page;
2016 		get_page(page);
2017 		spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2018 
2019 		f2fs_wait_on_page_writeback(page, NODE, true, false);
2020 		if (TestClearPageError(page))
2021 			ret = -EIO;
2022 
2023 		put_page(page);
2024 
2025 		if (ret)
2026 			break;
2027 	}
2028 
2029 	ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2030 	if (!ret)
2031 		ret = ret2;
2032 
2033 	return ret;
2034 }
2035 
2036 static int f2fs_write_node_pages(struct address_space *mapping,
2037 			    struct writeback_control *wbc)
2038 {
2039 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2040 	struct blk_plug plug;
2041 	long diff;
2042 
2043 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2044 		goto skip_write;
2045 
2046 	/* balancing f2fs's metadata in background */
2047 	f2fs_balance_fs_bg(sbi, true);
2048 
2049 	/* collect a number of dirty node pages and write together */
2050 	if (wbc->sync_mode != WB_SYNC_ALL &&
2051 			get_pages(sbi, F2FS_DIRTY_NODES) <
2052 					nr_pages_to_skip(sbi, NODE))
2053 		goto skip_write;
2054 
2055 	if (wbc->sync_mode == WB_SYNC_ALL)
2056 		atomic_inc(&sbi->wb_sync_req[NODE]);
2057 	else if (atomic_read(&sbi->wb_sync_req[NODE]))
2058 		goto skip_write;
2059 
2060 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2061 
2062 	diff = nr_pages_to_write(sbi, NODE, wbc);
2063 	blk_start_plug(&plug);
2064 	f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2065 	blk_finish_plug(&plug);
2066 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2067 
2068 	if (wbc->sync_mode == WB_SYNC_ALL)
2069 		atomic_dec(&sbi->wb_sync_req[NODE]);
2070 	return 0;
2071 
2072 skip_write:
2073 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2074 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2075 	return 0;
2076 }
2077 
2078 static int f2fs_set_node_page_dirty(struct page *page)
2079 {
2080 	trace_f2fs_set_page_dirty(page, NODE);
2081 
2082 	if (!PageUptodate(page))
2083 		SetPageUptodate(page);
2084 #ifdef CONFIG_F2FS_CHECK_FS
2085 	if (IS_INODE(page))
2086 		f2fs_inode_chksum_set(F2FS_P_SB(page), page);
2087 #endif
2088 	if (!PageDirty(page)) {
2089 		__set_page_dirty_nobuffers(page);
2090 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
2091 		f2fs_set_page_private(page, 0);
2092 		f2fs_trace_pid(page);
2093 		return 1;
2094 	}
2095 	return 0;
2096 }
2097 
2098 /*
2099  * Structure of the f2fs node operations
2100  */
2101 const struct address_space_operations f2fs_node_aops = {
2102 	.writepage	= f2fs_write_node_page,
2103 	.writepages	= f2fs_write_node_pages,
2104 	.set_page_dirty	= f2fs_set_node_page_dirty,
2105 	.invalidatepage	= f2fs_invalidate_page,
2106 	.releasepage	= f2fs_release_page,
2107 #ifdef CONFIG_MIGRATION
2108 	.migratepage	= f2fs_migrate_page,
2109 #endif
2110 };
2111 
2112 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2113 						nid_t n)
2114 {
2115 	return radix_tree_lookup(&nm_i->free_nid_root, n);
2116 }
2117 
2118 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2119 				struct free_nid *i)
2120 {
2121 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2122 
2123 	int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2124 	if (err)
2125 		return err;
2126 
2127 	nm_i->nid_cnt[FREE_NID]++;
2128 	list_add_tail(&i->list, &nm_i->free_nid_list);
2129 	return 0;
2130 }
2131 
2132 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2133 			struct free_nid *i, enum nid_state state)
2134 {
2135 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2136 
2137 	f2fs_bug_on(sbi, state != i->state);
2138 	nm_i->nid_cnt[state]--;
2139 	if (state == FREE_NID)
2140 		list_del(&i->list);
2141 	radix_tree_delete(&nm_i->free_nid_root, i->nid);
2142 }
2143 
2144 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2145 			enum nid_state org_state, enum nid_state dst_state)
2146 {
2147 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2148 
2149 	f2fs_bug_on(sbi, org_state != i->state);
2150 	i->state = dst_state;
2151 	nm_i->nid_cnt[org_state]--;
2152 	nm_i->nid_cnt[dst_state]++;
2153 
2154 	switch (dst_state) {
2155 	case PREALLOC_NID:
2156 		list_del(&i->list);
2157 		break;
2158 	case FREE_NID:
2159 		list_add_tail(&i->list, &nm_i->free_nid_list);
2160 		break;
2161 	default:
2162 		BUG_ON(1);
2163 	}
2164 }
2165 
2166 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2167 							bool set, bool build)
2168 {
2169 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2170 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2171 	unsigned int nid_ofs = nid - START_NID(nid);
2172 
2173 	if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2174 		return;
2175 
2176 	if (set) {
2177 		if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2178 			return;
2179 		__set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2180 		nm_i->free_nid_count[nat_ofs]++;
2181 	} else {
2182 		if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2183 			return;
2184 		__clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2185 		if (!build)
2186 			nm_i->free_nid_count[nat_ofs]--;
2187 	}
2188 }
2189 
2190 /* return if the nid is recognized as free */
2191 static bool add_free_nid(struct f2fs_sb_info *sbi,
2192 				nid_t nid, bool build, bool update)
2193 {
2194 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2195 	struct free_nid *i, *e;
2196 	struct nat_entry *ne;
2197 	int err = -EINVAL;
2198 	bool ret = false;
2199 
2200 	/* 0 nid should not be used */
2201 	if (unlikely(nid == 0))
2202 		return false;
2203 
2204 	if (unlikely(f2fs_check_nid_range(sbi, nid)))
2205 		return false;
2206 
2207 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2208 	i->nid = nid;
2209 	i->state = FREE_NID;
2210 
2211 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2212 
2213 	spin_lock(&nm_i->nid_list_lock);
2214 
2215 	if (build) {
2216 		/*
2217 		 *   Thread A             Thread B
2218 		 *  - f2fs_create
2219 		 *   - f2fs_new_inode
2220 		 *    - f2fs_alloc_nid
2221 		 *     - __insert_nid_to_list(PREALLOC_NID)
2222 		 *                     - f2fs_balance_fs_bg
2223 		 *                      - f2fs_build_free_nids
2224 		 *                       - __f2fs_build_free_nids
2225 		 *                        - scan_nat_page
2226 		 *                         - add_free_nid
2227 		 *                          - __lookup_nat_cache
2228 		 *  - f2fs_add_link
2229 		 *   - f2fs_init_inode_metadata
2230 		 *    - f2fs_new_inode_page
2231 		 *     - f2fs_new_node_page
2232 		 *      - set_node_addr
2233 		 *  - f2fs_alloc_nid_done
2234 		 *   - __remove_nid_from_list(PREALLOC_NID)
2235 		 *                         - __insert_nid_to_list(FREE_NID)
2236 		 */
2237 		ne = __lookup_nat_cache(nm_i, nid);
2238 		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2239 				nat_get_blkaddr(ne) != NULL_ADDR))
2240 			goto err_out;
2241 
2242 		e = __lookup_free_nid_list(nm_i, nid);
2243 		if (e) {
2244 			if (e->state == FREE_NID)
2245 				ret = true;
2246 			goto err_out;
2247 		}
2248 	}
2249 	ret = true;
2250 	err = __insert_free_nid(sbi, i);
2251 err_out:
2252 	if (update) {
2253 		update_free_nid_bitmap(sbi, nid, ret, build);
2254 		if (!build)
2255 			nm_i->available_nids++;
2256 	}
2257 	spin_unlock(&nm_i->nid_list_lock);
2258 	radix_tree_preload_end();
2259 
2260 	if (err)
2261 		kmem_cache_free(free_nid_slab, i);
2262 	return ret;
2263 }
2264 
2265 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2266 {
2267 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2268 	struct free_nid *i;
2269 	bool need_free = false;
2270 
2271 	spin_lock(&nm_i->nid_list_lock);
2272 	i = __lookup_free_nid_list(nm_i, nid);
2273 	if (i && i->state == FREE_NID) {
2274 		__remove_free_nid(sbi, i, FREE_NID);
2275 		need_free = true;
2276 	}
2277 	spin_unlock(&nm_i->nid_list_lock);
2278 
2279 	if (need_free)
2280 		kmem_cache_free(free_nid_slab, i);
2281 }
2282 
2283 static int scan_nat_page(struct f2fs_sb_info *sbi,
2284 			struct page *nat_page, nid_t start_nid)
2285 {
2286 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2287 	struct f2fs_nat_block *nat_blk = page_address(nat_page);
2288 	block_t blk_addr;
2289 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2290 	int i;
2291 
2292 	__set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2293 
2294 	i = start_nid % NAT_ENTRY_PER_BLOCK;
2295 
2296 	for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2297 		if (unlikely(start_nid >= nm_i->max_nid))
2298 			break;
2299 
2300 		blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2301 
2302 		if (blk_addr == NEW_ADDR)
2303 			return -EINVAL;
2304 
2305 		if (blk_addr == NULL_ADDR) {
2306 			add_free_nid(sbi, start_nid, true, true);
2307 		} else {
2308 			spin_lock(&NM_I(sbi)->nid_list_lock);
2309 			update_free_nid_bitmap(sbi, start_nid, false, true);
2310 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2311 		}
2312 	}
2313 
2314 	return 0;
2315 }
2316 
2317 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2318 {
2319 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2320 	struct f2fs_journal *journal = curseg->journal;
2321 	int i;
2322 
2323 	down_read(&curseg->journal_rwsem);
2324 	for (i = 0; i < nats_in_cursum(journal); i++) {
2325 		block_t addr;
2326 		nid_t nid;
2327 
2328 		addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2329 		nid = le32_to_cpu(nid_in_journal(journal, i));
2330 		if (addr == NULL_ADDR)
2331 			add_free_nid(sbi, nid, true, false);
2332 		else
2333 			remove_free_nid(sbi, nid);
2334 	}
2335 	up_read(&curseg->journal_rwsem);
2336 }
2337 
2338 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2339 {
2340 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2341 	unsigned int i, idx;
2342 	nid_t nid;
2343 
2344 	down_read(&nm_i->nat_tree_lock);
2345 
2346 	for (i = 0; i < nm_i->nat_blocks; i++) {
2347 		if (!test_bit_le(i, nm_i->nat_block_bitmap))
2348 			continue;
2349 		if (!nm_i->free_nid_count[i])
2350 			continue;
2351 		for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2352 			idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2353 						NAT_ENTRY_PER_BLOCK, idx);
2354 			if (idx >= NAT_ENTRY_PER_BLOCK)
2355 				break;
2356 
2357 			nid = i * NAT_ENTRY_PER_BLOCK + idx;
2358 			add_free_nid(sbi, nid, true, false);
2359 
2360 			if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2361 				goto out;
2362 		}
2363 	}
2364 out:
2365 	scan_curseg_cache(sbi);
2366 
2367 	up_read(&nm_i->nat_tree_lock);
2368 }
2369 
2370 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2371 						bool sync, bool mount)
2372 {
2373 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2374 	int i = 0, ret;
2375 	nid_t nid = nm_i->next_scan_nid;
2376 
2377 	if (unlikely(nid >= nm_i->max_nid))
2378 		nid = 0;
2379 
2380 	if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2381 		nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2382 
2383 	/* Enough entries */
2384 	if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2385 		return 0;
2386 
2387 	if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2388 		return 0;
2389 
2390 	if (!mount) {
2391 		/* try to find free nids in free_nid_bitmap */
2392 		scan_free_nid_bits(sbi);
2393 
2394 		if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2395 			return 0;
2396 	}
2397 
2398 	/* readahead nat pages to be scanned */
2399 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2400 							META_NAT, true);
2401 
2402 	down_read(&nm_i->nat_tree_lock);
2403 
2404 	while (1) {
2405 		if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2406 						nm_i->nat_block_bitmap)) {
2407 			struct page *page = get_current_nat_page(sbi, nid);
2408 
2409 			if (IS_ERR(page)) {
2410 				ret = PTR_ERR(page);
2411 			} else {
2412 				ret = scan_nat_page(sbi, page, nid);
2413 				f2fs_put_page(page, 1);
2414 			}
2415 
2416 			if (ret) {
2417 				up_read(&nm_i->nat_tree_lock);
2418 				f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2419 				return ret;
2420 			}
2421 		}
2422 
2423 		nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2424 		if (unlikely(nid >= nm_i->max_nid))
2425 			nid = 0;
2426 
2427 		if (++i >= FREE_NID_PAGES)
2428 			break;
2429 	}
2430 
2431 	/* go to the next free nat pages to find free nids abundantly */
2432 	nm_i->next_scan_nid = nid;
2433 
2434 	/* find free nids from current sum_pages */
2435 	scan_curseg_cache(sbi);
2436 
2437 	up_read(&nm_i->nat_tree_lock);
2438 
2439 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2440 					nm_i->ra_nid_pages, META_NAT, false);
2441 
2442 	return 0;
2443 }
2444 
2445 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2446 {
2447 	int ret;
2448 
2449 	mutex_lock(&NM_I(sbi)->build_lock);
2450 	ret = __f2fs_build_free_nids(sbi, sync, mount);
2451 	mutex_unlock(&NM_I(sbi)->build_lock);
2452 
2453 	return ret;
2454 }
2455 
2456 /*
2457  * If this function returns success, caller can obtain a new nid
2458  * from second parameter of this function.
2459  * The returned nid could be used ino as well as nid when inode is created.
2460  */
2461 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2462 {
2463 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2464 	struct free_nid *i = NULL;
2465 retry:
2466 	if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2467 		f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2468 		return false;
2469 	}
2470 
2471 	spin_lock(&nm_i->nid_list_lock);
2472 
2473 	if (unlikely(nm_i->available_nids == 0)) {
2474 		spin_unlock(&nm_i->nid_list_lock);
2475 		return false;
2476 	}
2477 
2478 	/* We should not use stale free nids created by f2fs_build_free_nids */
2479 	if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2480 		f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2481 		i = list_first_entry(&nm_i->free_nid_list,
2482 					struct free_nid, list);
2483 		*nid = i->nid;
2484 
2485 		__move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2486 		nm_i->available_nids--;
2487 
2488 		update_free_nid_bitmap(sbi, *nid, false, false);
2489 
2490 		spin_unlock(&nm_i->nid_list_lock);
2491 		return true;
2492 	}
2493 	spin_unlock(&nm_i->nid_list_lock);
2494 
2495 	/* Let's scan nat pages and its caches to get free nids */
2496 	if (!f2fs_build_free_nids(sbi, true, false))
2497 		goto retry;
2498 	return false;
2499 }
2500 
2501 /*
2502  * f2fs_alloc_nid() should be called prior to this function.
2503  */
2504 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2505 {
2506 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2507 	struct free_nid *i;
2508 
2509 	spin_lock(&nm_i->nid_list_lock);
2510 	i = __lookup_free_nid_list(nm_i, nid);
2511 	f2fs_bug_on(sbi, !i);
2512 	__remove_free_nid(sbi, i, PREALLOC_NID);
2513 	spin_unlock(&nm_i->nid_list_lock);
2514 
2515 	kmem_cache_free(free_nid_slab, i);
2516 }
2517 
2518 /*
2519  * f2fs_alloc_nid() should be called prior to this function.
2520  */
2521 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2522 {
2523 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2524 	struct free_nid *i;
2525 	bool need_free = false;
2526 
2527 	if (!nid)
2528 		return;
2529 
2530 	spin_lock(&nm_i->nid_list_lock);
2531 	i = __lookup_free_nid_list(nm_i, nid);
2532 	f2fs_bug_on(sbi, !i);
2533 
2534 	if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2535 		__remove_free_nid(sbi, i, PREALLOC_NID);
2536 		need_free = true;
2537 	} else {
2538 		__move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2539 	}
2540 
2541 	nm_i->available_nids++;
2542 
2543 	update_free_nid_bitmap(sbi, nid, true, false);
2544 
2545 	spin_unlock(&nm_i->nid_list_lock);
2546 
2547 	if (need_free)
2548 		kmem_cache_free(free_nid_slab, i);
2549 }
2550 
2551 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2552 {
2553 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2554 	int nr = nr_shrink;
2555 
2556 	if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2557 		return 0;
2558 
2559 	if (!mutex_trylock(&nm_i->build_lock))
2560 		return 0;
2561 
2562 	while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2563 		struct free_nid *i, *next;
2564 		unsigned int batch = SHRINK_NID_BATCH_SIZE;
2565 
2566 		spin_lock(&nm_i->nid_list_lock);
2567 		list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2568 			if (!nr_shrink || !batch ||
2569 				nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2570 				break;
2571 			__remove_free_nid(sbi, i, FREE_NID);
2572 			kmem_cache_free(free_nid_slab, i);
2573 			nr_shrink--;
2574 			batch--;
2575 		}
2576 		spin_unlock(&nm_i->nid_list_lock);
2577 	}
2578 
2579 	mutex_unlock(&nm_i->build_lock);
2580 
2581 	return nr - nr_shrink;
2582 }
2583 
2584 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2585 {
2586 	void *src_addr, *dst_addr;
2587 	size_t inline_size;
2588 	struct page *ipage;
2589 	struct f2fs_inode *ri;
2590 
2591 	ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2592 	if (IS_ERR(ipage))
2593 		return PTR_ERR(ipage);
2594 
2595 	ri = F2FS_INODE(page);
2596 	if (ri->i_inline & F2FS_INLINE_XATTR) {
2597 		if (!f2fs_has_inline_xattr(inode)) {
2598 			set_inode_flag(inode, FI_INLINE_XATTR);
2599 			stat_inc_inline_xattr(inode);
2600 		}
2601 	} else {
2602 		if (f2fs_has_inline_xattr(inode)) {
2603 			stat_dec_inline_xattr(inode);
2604 			clear_inode_flag(inode, FI_INLINE_XATTR);
2605 		}
2606 		goto update_inode;
2607 	}
2608 
2609 	dst_addr = inline_xattr_addr(inode, ipage);
2610 	src_addr = inline_xattr_addr(inode, page);
2611 	inline_size = inline_xattr_size(inode);
2612 
2613 	f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2614 	memcpy(dst_addr, src_addr, inline_size);
2615 update_inode:
2616 	f2fs_update_inode(inode, ipage);
2617 	f2fs_put_page(ipage, 1);
2618 	return 0;
2619 }
2620 
2621 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2622 {
2623 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2624 	nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2625 	nid_t new_xnid;
2626 	struct dnode_of_data dn;
2627 	struct node_info ni;
2628 	struct page *xpage;
2629 	int err;
2630 
2631 	if (!prev_xnid)
2632 		goto recover_xnid;
2633 
2634 	/* 1: invalidate the previous xattr nid */
2635 	err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2636 	if (err)
2637 		return err;
2638 
2639 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
2640 	dec_valid_node_count(sbi, inode, false);
2641 	set_node_addr(sbi, &ni, NULL_ADDR, false);
2642 
2643 recover_xnid:
2644 	/* 2: update xattr nid in inode */
2645 	if (!f2fs_alloc_nid(sbi, &new_xnid))
2646 		return -ENOSPC;
2647 
2648 	set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2649 	xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2650 	if (IS_ERR(xpage)) {
2651 		f2fs_alloc_nid_failed(sbi, new_xnid);
2652 		return PTR_ERR(xpage);
2653 	}
2654 
2655 	f2fs_alloc_nid_done(sbi, new_xnid);
2656 	f2fs_update_inode_page(inode);
2657 
2658 	/* 3: update and set xattr node page dirty */
2659 	memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2660 
2661 	set_page_dirty(xpage);
2662 	f2fs_put_page(xpage, 1);
2663 
2664 	return 0;
2665 }
2666 
2667 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2668 {
2669 	struct f2fs_inode *src, *dst;
2670 	nid_t ino = ino_of_node(page);
2671 	struct node_info old_ni, new_ni;
2672 	struct page *ipage;
2673 	int err;
2674 
2675 	err = f2fs_get_node_info(sbi, ino, &old_ni);
2676 	if (err)
2677 		return err;
2678 
2679 	if (unlikely(old_ni.blk_addr != NULL_ADDR))
2680 		return -EINVAL;
2681 retry:
2682 	ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2683 	if (!ipage) {
2684 		congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2685 		goto retry;
2686 	}
2687 
2688 	/* Should not use this inode from free nid list */
2689 	remove_free_nid(sbi, ino);
2690 
2691 	if (!PageUptodate(ipage))
2692 		SetPageUptodate(ipage);
2693 	fill_node_footer(ipage, ino, ino, 0, true);
2694 	set_cold_node(ipage, false);
2695 
2696 	src = F2FS_INODE(page);
2697 	dst = F2FS_INODE(ipage);
2698 
2699 	memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2700 	dst->i_size = 0;
2701 	dst->i_blocks = cpu_to_le64(1);
2702 	dst->i_links = cpu_to_le32(1);
2703 	dst->i_xattr_nid = 0;
2704 	dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2705 	if (dst->i_inline & F2FS_EXTRA_ATTR) {
2706 		dst->i_extra_isize = src->i_extra_isize;
2707 
2708 		if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2709 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2710 							i_inline_xattr_size))
2711 			dst->i_inline_xattr_size = src->i_inline_xattr_size;
2712 
2713 		if (f2fs_sb_has_project_quota(sbi) &&
2714 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2715 								i_projid))
2716 			dst->i_projid = src->i_projid;
2717 
2718 		if (f2fs_sb_has_inode_crtime(sbi) &&
2719 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2720 							i_crtime_nsec)) {
2721 			dst->i_crtime = src->i_crtime;
2722 			dst->i_crtime_nsec = src->i_crtime_nsec;
2723 		}
2724 	}
2725 
2726 	new_ni = old_ni;
2727 	new_ni.ino = ino;
2728 
2729 	if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2730 		WARN_ON(1);
2731 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2732 	inc_valid_inode_count(sbi);
2733 	set_page_dirty(ipage);
2734 	f2fs_put_page(ipage, 1);
2735 	return 0;
2736 }
2737 
2738 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2739 			unsigned int segno, struct f2fs_summary_block *sum)
2740 {
2741 	struct f2fs_node *rn;
2742 	struct f2fs_summary *sum_entry;
2743 	block_t addr;
2744 	int i, idx, last_offset, nrpages;
2745 
2746 	/* scan the node segment */
2747 	last_offset = sbi->blocks_per_seg;
2748 	addr = START_BLOCK(sbi, segno);
2749 	sum_entry = &sum->entries[0];
2750 
2751 	for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2752 		nrpages = min(last_offset - i, BIO_MAX_PAGES);
2753 
2754 		/* readahead node pages */
2755 		f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2756 
2757 		for (idx = addr; idx < addr + nrpages; idx++) {
2758 			struct page *page = f2fs_get_tmp_page(sbi, idx);
2759 
2760 			if (IS_ERR(page))
2761 				return PTR_ERR(page);
2762 
2763 			rn = F2FS_NODE(page);
2764 			sum_entry->nid = rn->footer.nid;
2765 			sum_entry->version = 0;
2766 			sum_entry->ofs_in_node = 0;
2767 			sum_entry++;
2768 			f2fs_put_page(page, 1);
2769 		}
2770 
2771 		invalidate_mapping_pages(META_MAPPING(sbi), addr,
2772 							addr + nrpages);
2773 	}
2774 	return 0;
2775 }
2776 
2777 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2778 {
2779 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2780 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2781 	struct f2fs_journal *journal = curseg->journal;
2782 	int i;
2783 
2784 	down_write(&curseg->journal_rwsem);
2785 	for (i = 0; i < nats_in_cursum(journal); i++) {
2786 		struct nat_entry *ne;
2787 		struct f2fs_nat_entry raw_ne;
2788 		nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2789 
2790 		raw_ne = nat_in_journal(journal, i);
2791 
2792 		ne = __lookup_nat_cache(nm_i, nid);
2793 		if (!ne) {
2794 			ne = __alloc_nat_entry(nid, true);
2795 			__init_nat_entry(nm_i, ne, &raw_ne, true);
2796 		}
2797 
2798 		/*
2799 		 * if a free nat in journal has not been used after last
2800 		 * checkpoint, we should remove it from available nids,
2801 		 * since later we will add it again.
2802 		 */
2803 		if (!get_nat_flag(ne, IS_DIRTY) &&
2804 				le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2805 			spin_lock(&nm_i->nid_list_lock);
2806 			nm_i->available_nids--;
2807 			spin_unlock(&nm_i->nid_list_lock);
2808 		}
2809 
2810 		__set_nat_cache_dirty(nm_i, ne);
2811 	}
2812 	update_nats_in_cursum(journal, -i);
2813 	up_write(&curseg->journal_rwsem);
2814 }
2815 
2816 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2817 						struct list_head *head, int max)
2818 {
2819 	struct nat_entry_set *cur;
2820 
2821 	if (nes->entry_cnt >= max)
2822 		goto add_out;
2823 
2824 	list_for_each_entry(cur, head, set_list) {
2825 		if (cur->entry_cnt >= nes->entry_cnt) {
2826 			list_add(&nes->set_list, cur->set_list.prev);
2827 			return;
2828 		}
2829 	}
2830 add_out:
2831 	list_add_tail(&nes->set_list, head);
2832 }
2833 
2834 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2835 						struct page *page)
2836 {
2837 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2838 	unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2839 	struct f2fs_nat_block *nat_blk = page_address(page);
2840 	int valid = 0;
2841 	int i = 0;
2842 
2843 	if (!enabled_nat_bits(sbi, NULL))
2844 		return;
2845 
2846 	if (nat_index == 0) {
2847 		valid = 1;
2848 		i = 1;
2849 	}
2850 	for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2851 		if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2852 			valid++;
2853 	}
2854 	if (valid == 0) {
2855 		__set_bit_le(nat_index, nm_i->empty_nat_bits);
2856 		__clear_bit_le(nat_index, nm_i->full_nat_bits);
2857 		return;
2858 	}
2859 
2860 	__clear_bit_le(nat_index, nm_i->empty_nat_bits);
2861 	if (valid == NAT_ENTRY_PER_BLOCK)
2862 		__set_bit_le(nat_index, nm_i->full_nat_bits);
2863 	else
2864 		__clear_bit_le(nat_index, nm_i->full_nat_bits);
2865 }
2866 
2867 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2868 		struct nat_entry_set *set, struct cp_control *cpc)
2869 {
2870 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2871 	struct f2fs_journal *journal = curseg->journal;
2872 	nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2873 	bool to_journal = true;
2874 	struct f2fs_nat_block *nat_blk;
2875 	struct nat_entry *ne, *cur;
2876 	struct page *page = NULL;
2877 
2878 	/*
2879 	 * there are two steps to flush nat entries:
2880 	 * #1, flush nat entries to journal in current hot data summary block.
2881 	 * #2, flush nat entries to nat page.
2882 	 */
2883 	if (enabled_nat_bits(sbi, cpc) ||
2884 		!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2885 		to_journal = false;
2886 
2887 	if (to_journal) {
2888 		down_write(&curseg->journal_rwsem);
2889 	} else {
2890 		page = get_next_nat_page(sbi, start_nid);
2891 		if (IS_ERR(page))
2892 			return PTR_ERR(page);
2893 
2894 		nat_blk = page_address(page);
2895 		f2fs_bug_on(sbi, !nat_blk);
2896 	}
2897 
2898 	/* flush dirty nats in nat entry set */
2899 	list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2900 		struct f2fs_nat_entry *raw_ne;
2901 		nid_t nid = nat_get_nid(ne);
2902 		int offset;
2903 
2904 		f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2905 
2906 		if (to_journal) {
2907 			offset = f2fs_lookup_journal_in_cursum(journal,
2908 							NAT_JOURNAL, nid, 1);
2909 			f2fs_bug_on(sbi, offset < 0);
2910 			raw_ne = &nat_in_journal(journal, offset);
2911 			nid_in_journal(journal, offset) = cpu_to_le32(nid);
2912 		} else {
2913 			raw_ne = &nat_blk->entries[nid - start_nid];
2914 		}
2915 		raw_nat_from_node_info(raw_ne, &ne->ni);
2916 		nat_reset_flag(ne);
2917 		__clear_nat_cache_dirty(NM_I(sbi), set, ne);
2918 		if (nat_get_blkaddr(ne) == NULL_ADDR) {
2919 			add_free_nid(sbi, nid, false, true);
2920 		} else {
2921 			spin_lock(&NM_I(sbi)->nid_list_lock);
2922 			update_free_nid_bitmap(sbi, nid, false, false);
2923 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2924 		}
2925 	}
2926 
2927 	if (to_journal) {
2928 		up_write(&curseg->journal_rwsem);
2929 	} else {
2930 		__update_nat_bits(sbi, start_nid, page);
2931 		f2fs_put_page(page, 1);
2932 	}
2933 
2934 	/* Allow dirty nats by node block allocation in write_begin */
2935 	if (!set->entry_cnt) {
2936 		radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2937 		kmem_cache_free(nat_entry_set_slab, set);
2938 	}
2939 	return 0;
2940 }
2941 
2942 /*
2943  * This function is called during the checkpointing process.
2944  */
2945 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2946 {
2947 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2948 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2949 	struct f2fs_journal *journal = curseg->journal;
2950 	struct nat_entry_set *setvec[SETVEC_SIZE];
2951 	struct nat_entry_set *set, *tmp;
2952 	unsigned int found;
2953 	nid_t set_idx = 0;
2954 	LIST_HEAD(sets);
2955 	int err = 0;
2956 
2957 	/*
2958 	 * during unmount, let's flush nat_bits before checking
2959 	 * nat_cnt[DIRTY_NAT].
2960 	 */
2961 	if (enabled_nat_bits(sbi, cpc)) {
2962 		down_write(&nm_i->nat_tree_lock);
2963 		remove_nats_in_journal(sbi);
2964 		up_write(&nm_i->nat_tree_lock);
2965 	}
2966 
2967 	if (!nm_i->nat_cnt[DIRTY_NAT])
2968 		return 0;
2969 
2970 	down_write(&nm_i->nat_tree_lock);
2971 
2972 	/*
2973 	 * if there are no enough space in journal to store dirty nat
2974 	 * entries, remove all entries from journal and merge them
2975 	 * into nat entry set.
2976 	 */
2977 	if (enabled_nat_bits(sbi, cpc) ||
2978 		!__has_cursum_space(journal,
2979 			nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
2980 		remove_nats_in_journal(sbi);
2981 
2982 	while ((found = __gang_lookup_nat_set(nm_i,
2983 					set_idx, SETVEC_SIZE, setvec))) {
2984 		unsigned idx;
2985 		set_idx = setvec[found - 1]->set + 1;
2986 		for (idx = 0; idx < found; idx++)
2987 			__adjust_nat_entry_set(setvec[idx], &sets,
2988 						MAX_NAT_JENTRIES(journal));
2989 	}
2990 
2991 	/* flush dirty nats in nat entry set */
2992 	list_for_each_entry_safe(set, tmp, &sets, set_list) {
2993 		err = __flush_nat_entry_set(sbi, set, cpc);
2994 		if (err)
2995 			break;
2996 	}
2997 
2998 	up_write(&nm_i->nat_tree_lock);
2999 	/* Allow dirty nats by node block allocation in write_begin */
3000 
3001 	return err;
3002 }
3003 
3004 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3005 {
3006 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3007 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3008 	unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3009 	unsigned int i;
3010 	__u64 cp_ver = cur_cp_version(ckpt);
3011 	block_t nat_bits_addr;
3012 
3013 	if (!enabled_nat_bits(sbi, NULL))
3014 		return 0;
3015 
3016 	nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3017 	nm_i->nat_bits = f2fs_kvzalloc(sbi,
3018 			nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3019 	if (!nm_i->nat_bits)
3020 		return -ENOMEM;
3021 
3022 	nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3023 						nm_i->nat_bits_blocks;
3024 	for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3025 		struct page *page;
3026 
3027 		page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3028 		if (IS_ERR(page))
3029 			return PTR_ERR(page);
3030 
3031 		memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3032 					page_address(page), F2FS_BLKSIZE);
3033 		f2fs_put_page(page, 1);
3034 	}
3035 
3036 	cp_ver |= (cur_cp_crc(ckpt) << 32);
3037 	if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3038 		disable_nat_bits(sbi, true);
3039 		return 0;
3040 	}
3041 
3042 	nm_i->full_nat_bits = nm_i->nat_bits + 8;
3043 	nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3044 
3045 	f2fs_notice(sbi, "Found nat_bits in checkpoint");
3046 	return 0;
3047 }
3048 
3049 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3050 {
3051 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3052 	unsigned int i = 0;
3053 	nid_t nid, last_nid;
3054 
3055 	if (!enabled_nat_bits(sbi, NULL))
3056 		return;
3057 
3058 	for (i = 0; i < nm_i->nat_blocks; i++) {
3059 		i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3060 		if (i >= nm_i->nat_blocks)
3061 			break;
3062 
3063 		__set_bit_le(i, nm_i->nat_block_bitmap);
3064 
3065 		nid = i * NAT_ENTRY_PER_BLOCK;
3066 		last_nid = nid + NAT_ENTRY_PER_BLOCK;
3067 
3068 		spin_lock(&NM_I(sbi)->nid_list_lock);
3069 		for (; nid < last_nid; nid++)
3070 			update_free_nid_bitmap(sbi, nid, true, true);
3071 		spin_unlock(&NM_I(sbi)->nid_list_lock);
3072 	}
3073 
3074 	for (i = 0; i < nm_i->nat_blocks; i++) {
3075 		i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3076 		if (i >= nm_i->nat_blocks)
3077 			break;
3078 
3079 		__set_bit_le(i, nm_i->nat_block_bitmap);
3080 	}
3081 }
3082 
3083 static int init_node_manager(struct f2fs_sb_info *sbi)
3084 {
3085 	struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3086 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3087 	unsigned char *version_bitmap;
3088 	unsigned int nat_segs;
3089 	int err;
3090 
3091 	nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3092 
3093 	/* segment_count_nat includes pair segment so divide to 2. */
3094 	nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3095 	nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3096 	nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3097 
3098 	/* not used nids: 0, node, meta, (and root counted as valid node) */
3099 	nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3100 						F2FS_RESERVED_NODE_NUM;
3101 	nm_i->nid_cnt[FREE_NID] = 0;
3102 	nm_i->nid_cnt[PREALLOC_NID] = 0;
3103 	nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3104 	nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3105 	nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3106 
3107 	INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3108 	INIT_LIST_HEAD(&nm_i->free_nid_list);
3109 	INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3110 	INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3111 	INIT_LIST_HEAD(&nm_i->nat_entries);
3112 	spin_lock_init(&nm_i->nat_list_lock);
3113 
3114 	mutex_init(&nm_i->build_lock);
3115 	spin_lock_init(&nm_i->nid_list_lock);
3116 	init_rwsem(&nm_i->nat_tree_lock);
3117 
3118 	nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3119 	nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3120 	version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3121 	nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3122 					GFP_KERNEL);
3123 	if (!nm_i->nat_bitmap)
3124 		return -ENOMEM;
3125 
3126 	err = __get_nat_bitmaps(sbi);
3127 	if (err)
3128 		return err;
3129 
3130 #ifdef CONFIG_F2FS_CHECK_FS
3131 	nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3132 					GFP_KERNEL);
3133 	if (!nm_i->nat_bitmap_mir)
3134 		return -ENOMEM;
3135 #endif
3136 
3137 	return 0;
3138 }
3139 
3140 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3141 {
3142 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3143 	int i;
3144 
3145 	nm_i->free_nid_bitmap =
3146 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3147 					      nm_i->nat_blocks),
3148 			      GFP_KERNEL);
3149 	if (!nm_i->free_nid_bitmap)
3150 		return -ENOMEM;
3151 
3152 	for (i = 0; i < nm_i->nat_blocks; i++) {
3153 		nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3154 			f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3155 		if (!nm_i->free_nid_bitmap[i])
3156 			return -ENOMEM;
3157 	}
3158 
3159 	nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3160 								GFP_KERNEL);
3161 	if (!nm_i->nat_block_bitmap)
3162 		return -ENOMEM;
3163 
3164 	nm_i->free_nid_count =
3165 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3166 					      nm_i->nat_blocks),
3167 			      GFP_KERNEL);
3168 	if (!nm_i->free_nid_count)
3169 		return -ENOMEM;
3170 	return 0;
3171 }
3172 
3173 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3174 {
3175 	int err;
3176 
3177 	sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3178 							GFP_KERNEL);
3179 	if (!sbi->nm_info)
3180 		return -ENOMEM;
3181 
3182 	err = init_node_manager(sbi);
3183 	if (err)
3184 		return err;
3185 
3186 	err = init_free_nid_cache(sbi);
3187 	if (err)
3188 		return err;
3189 
3190 	/* load free nid status from nat_bits table */
3191 	load_free_nid_bitmap(sbi);
3192 
3193 	return f2fs_build_free_nids(sbi, true, true);
3194 }
3195 
3196 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3197 {
3198 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3199 	struct free_nid *i, *next_i;
3200 	struct nat_entry *natvec[NATVEC_SIZE];
3201 	struct nat_entry_set *setvec[SETVEC_SIZE];
3202 	nid_t nid = 0;
3203 	unsigned int found;
3204 
3205 	if (!nm_i)
3206 		return;
3207 
3208 	/* destroy free nid list */
3209 	spin_lock(&nm_i->nid_list_lock);
3210 	list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3211 		__remove_free_nid(sbi, i, FREE_NID);
3212 		spin_unlock(&nm_i->nid_list_lock);
3213 		kmem_cache_free(free_nid_slab, i);
3214 		spin_lock(&nm_i->nid_list_lock);
3215 	}
3216 	f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3217 	f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3218 	f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3219 	spin_unlock(&nm_i->nid_list_lock);
3220 
3221 	/* destroy nat cache */
3222 	down_write(&nm_i->nat_tree_lock);
3223 	while ((found = __gang_lookup_nat_cache(nm_i,
3224 					nid, NATVEC_SIZE, natvec))) {
3225 		unsigned idx;
3226 
3227 		nid = nat_get_nid(natvec[found - 1]) + 1;
3228 		for (idx = 0; idx < found; idx++) {
3229 			spin_lock(&nm_i->nat_list_lock);
3230 			list_del(&natvec[idx]->list);
3231 			spin_unlock(&nm_i->nat_list_lock);
3232 
3233 			__del_from_nat_cache(nm_i, natvec[idx]);
3234 		}
3235 	}
3236 	f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3237 
3238 	/* destroy nat set cache */
3239 	nid = 0;
3240 	while ((found = __gang_lookup_nat_set(nm_i,
3241 					nid, SETVEC_SIZE, setvec))) {
3242 		unsigned idx;
3243 
3244 		nid = setvec[found - 1]->set + 1;
3245 		for (idx = 0; idx < found; idx++) {
3246 			/* entry_cnt is not zero, when cp_error was occurred */
3247 			f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3248 			radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3249 			kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3250 		}
3251 	}
3252 	up_write(&nm_i->nat_tree_lock);
3253 
3254 	kvfree(nm_i->nat_block_bitmap);
3255 	if (nm_i->free_nid_bitmap) {
3256 		int i;
3257 
3258 		for (i = 0; i < nm_i->nat_blocks; i++)
3259 			kvfree(nm_i->free_nid_bitmap[i]);
3260 		kvfree(nm_i->free_nid_bitmap);
3261 	}
3262 	kvfree(nm_i->free_nid_count);
3263 
3264 	kvfree(nm_i->nat_bitmap);
3265 	kvfree(nm_i->nat_bits);
3266 #ifdef CONFIG_F2FS_CHECK_FS
3267 	kvfree(nm_i->nat_bitmap_mir);
3268 #endif
3269 	sbi->nm_info = NULL;
3270 	kfree(nm_i);
3271 }
3272 
3273 int __init f2fs_create_node_manager_caches(void)
3274 {
3275 	nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3276 			sizeof(struct nat_entry));
3277 	if (!nat_entry_slab)
3278 		goto fail;
3279 
3280 	free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3281 			sizeof(struct free_nid));
3282 	if (!free_nid_slab)
3283 		goto destroy_nat_entry;
3284 
3285 	nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3286 			sizeof(struct nat_entry_set));
3287 	if (!nat_entry_set_slab)
3288 		goto destroy_free_nid;
3289 
3290 	fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3291 			sizeof(struct fsync_node_entry));
3292 	if (!fsync_node_entry_slab)
3293 		goto destroy_nat_entry_set;
3294 	return 0;
3295 
3296 destroy_nat_entry_set:
3297 	kmem_cache_destroy(nat_entry_set_slab);
3298 destroy_free_nid:
3299 	kmem_cache_destroy(free_nid_slab);
3300 destroy_nat_entry:
3301 	kmem_cache_destroy(nat_entry_slab);
3302 fail:
3303 	return -ENOMEM;
3304 }
3305 
3306 void f2fs_destroy_node_manager_caches(void)
3307 {
3308 	kmem_cache_destroy(fsync_node_entry_slab);
3309 	kmem_cache_destroy(nat_entry_set_slab);
3310 	kmem_cache_destroy(free_nid_slab);
3311 	kmem_cache_destroy(nat_entry_slab);
3312 }
3313