xref: /linux/fs/f2fs/node.c (revision 8137a49e1567726eb10fcf55ad141ac19804ca6b)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * fs/f2fs/node.c
4  *
5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6  *             http://www.samsung.com/
7  */
8 #include <linux/fs.h>
9 #include <linux/f2fs_fs.h>
10 #include <linux/mpage.h>
11 #include <linux/backing-dev.h>
12 #include <linux/blkdev.h>
13 #include <linux/pagevec.h>
14 #include <linux/swap.h>
15 
16 #include "f2fs.h"
17 #include "node.h"
18 #include "segment.h"
19 #include "xattr.h"
20 #include <trace/events/f2fs.h>
21 
22 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
23 
24 static struct kmem_cache *nat_entry_slab;
25 static struct kmem_cache *free_nid_slab;
26 static struct kmem_cache *nat_entry_set_slab;
27 static struct kmem_cache *fsync_node_entry_slab;
28 
29 /*
30  * Check whether the given nid is within node id range.
31  */
32 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
33 {
34 	if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
35 		set_sbi_flag(sbi, SBI_NEED_FSCK);
36 		f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
37 			  __func__, nid);
38 		return -EFSCORRUPTED;
39 	}
40 	return 0;
41 }
42 
43 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
44 {
45 	struct f2fs_nm_info *nm_i = NM_I(sbi);
46 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
47 	struct sysinfo val;
48 	unsigned long avail_ram;
49 	unsigned long mem_size = 0;
50 	bool res = false;
51 
52 	if (!nm_i)
53 		return true;
54 
55 	si_meminfo(&val);
56 
57 	/* only uses low memory */
58 	avail_ram = val.totalram - val.totalhigh;
59 
60 	/*
61 	 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
62 	 */
63 	if (type == FREE_NIDS) {
64 		mem_size = (nm_i->nid_cnt[FREE_NID] *
65 				sizeof(struct free_nid)) >> PAGE_SHIFT;
66 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
67 	} else if (type == NAT_ENTRIES) {
68 		mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
69 				sizeof(struct nat_entry)) >> PAGE_SHIFT;
70 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
71 		if (excess_cached_nats(sbi))
72 			res = false;
73 	} else if (type == DIRTY_DENTS) {
74 		if (sbi->sb->s_bdi->wb.dirty_exceeded)
75 			return false;
76 		mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
77 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
78 	} else if (type == INO_ENTRIES) {
79 		int i;
80 
81 		for (i = 0; i < MAX_INO_ENTRY; i++)
82 			mem_size += sbi->im[i].ino_num *
83 						sizeof(struct ino_entry);
84 		mem_size >>= PAGE_SHIFT;
85 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
86 	} else if (type == EXTENT_CACHE) {
87 		mem_size = (atomic_read(&sbi->total_ext_tree) *
88 				sizeof(struct extent_tree) +
89 				atomic_read(&sbi->total_ext_node) *
90 				sizeof(struct extent_node)) >> PAGE_SHIFT;
91 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
92 	} else if (type == INMEM_PAGES) {
93 		/* it allows 20% / total_ram for inmemory pages */
94 		mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
95 		res = mem_size < (val.totalram / 5);
96 	} else if (type == DISCARD_CACHE) {
97 		mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
98 				sizeof(struct discard_cmd)) >> PAGE_SHIFT;
99 		res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
100 	} else if (type == COMPRESS_PAGE) {
101 #ifdef CONFIG_F2FS_FS_COMPRESSION
102 		unsigned long free_ram = val.freeram;
103 
104 		/*
105 		 * free memory is lower than watermark or cached page count
106 		 * exceed threshold, deny caching compress page.
107 		 */
108 		res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
109 			(COMPRESS_MAPPING(sbi)->nrpages <
110 			 free_ram * sbi->compress_percent / 100);
111 #else
112 		res = false;
113 #endif
114 	} else {
115 		if (!sbi->sb->s_bdi->wb.dirty_exceeded)
116 			return true;
117 	}
118 	return res;
119 }
120 
121 static void clear_node_page_dirty(struct page *page)
122 {
123 	if (PageDirty(page)) {
124 		f2fs_clear_page_cache_dirty_tag(page);
125 		clear_page_dirty_for_io(page);
126 		dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
127 	}
128 	ClearPageUptodate(page);
129 }
130 
131 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
132 {
133 	return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
134 }
135 
136 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
137 {
138 	struct page *src_page;
139 	struct page *dst_page;
140 	pgoff_t dst_off;
141 	void *src_addr;
142 	void *dst_addr;
143 	struct f2fs_nm_info *nm_i = NM_I(sbi);
144 
145 	dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
146 
147 	/* get current nat block page with lock */
148 	src_page = get_current_nat_page(sbi, nid);
149 	if (IS_ERR(src_page))
150 		return src_page;
151 	dst_page = f2fs_grab_meta_page(sbi, dst_off);
152 	f2fs_bug_on(sbi, PageDirty(src_page));
153 
154 	src_addr = page_address(src_page);
155 	dst_addr = page_address(dst_page);
156 	memcpy(dst_addr, src_addr, PAGE_SIZE);
157 	set_page_dirty(dst_page);
158 	f2fs_put_page(src_page, 1);
159 
160 	set_to_next_nat(nm_i, nid);
161 
162 	return dst_page;
163 }
164 
165 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
166 {
167 	struct nat_entry *new;
168 
169 	if (no_fail)
170 		new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
171 	else
172 		new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
173 	if (new) {
174 		nat_set_nid(new, nid);
175 		nat_reset_flag(new);
176 	}
177 	return new;
178 }
179 
180 static void __free_nat_entry(struct nat_entry *e)
181 {
182 	kmem_cache_free(nat_entry_slab, e);
183 }
184 
185 /* must be locked by nat_tree_lock */
186 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
187 	struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
188 {
189 	if (no_fail)
190 		f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
191 	else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
192 		return NULL;
193 
194 	if (raw_ne)
195 		node_info_from_raw_nat(&ne->ni, raw_ne);
196 
197 	spin_lock(&nm_i->nat_list_lock);
198 	list_add_tail(&ne->list, &nm_i->nat_entries);
199 	spin_unlock(&nm_i->nat_list_lock);
200 
201 	nm_i->nat_cnt[TOTAL_NAT]++;
202 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
203 	return ne;
204 }
205 
206 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
207 {
208 	struct nat_entry *ne;
209 
210 	ne = radix_tree_lookup(&nm_i->nat_root, n);
211 
212 	/* for recent accessed nat entry, move it to tail of lru list */
213 	if (ne && !get_nat_flag(ne, IS_DIRTY)) {
214 		spin_lock(&nm_i->nat_list_lock);
215 		if (!list_empty(&ne->list))
216 			list_move_tail(&ne->list, &nm_i->nat_entries);
217 		spin_unlock(&nm_i->nat_list_lock);
218 	}
219 
220 	return ne;
221 }
222 
223 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
224 		nid_t start, unsigned int nr, struct nat_entry **ep)
225 {
226 	return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
227 }
228 
229 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
230 {
231 	radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
232 	nm_i->nat_cnt[TOTAL_NAT]--;
233 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
234 	__free_nat_entry(e);
235 }
236 
237 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
238 							struct nat_entry *ne)
239 {
240 	nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
241 	struct nat_entry_set *head;
242 
243 	head = radix_tree_lookup(&nm_i->nat_set_root, set);
244 	if (!head) {
245 		head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
246 
247 		INIT_LIST_HEAD(&head->entry_list);
248 		INIT_LIST_HEAD(&head->set_list);
249 		head->set = set;
250 		head->entry_cnt = 0;
251 		f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
252 	}
253 	return head;
254 }
255 
256 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
257 						struct nat_entry *ne)
258 {
259 	struct nat_entry_set *head;
260 	bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
261 
262 	if (!new_ne)
263 		head = __grab_nat_entry_set(nm_i, ne);
264 
265 	/*
266 	 * update entry_cnt in below condition:
267 	 * 1. update NEW_ADDR to valid block address;
268 	 * 2. update old block address to new one;
269 	 */
270 	if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
271 				!get_nat_flag(ne, IS_DIRTY)))
272 		head->entry_cnt++;
273 
274 	set_nat_flag(ne, IS_PREALLOC, new_ne);
275 
276 	if (get_nat_flag(ne, IS_DIRTY))
277 		goto refresh_list;
278 
279 	nm_i->nat_cnt[DIRTY_NAT]++;
280 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
281 	set_nat_flag(ne, IS_DIRTY, true);
282 refresh_list:
283 	spin_lock(&nm_i->nat_list_lock);
284 	if (new_ne)
285 		list_del_init(&ne->list);
286 	else
287 		list_move_tail(&ne->list, &head->entry_list);
288 	spin_unlock(&nm_i->nat_list_lock);
289 }
290 
291 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
292 		struct nat_entry_set *set, struct nat_entry *ne)
293 {
294 	spin_lock(&nm_i->nat_list_lock);
295 	list_move_tail(&ne->list, &nm_i->nat_entries);
296 	spin_unlock(&nm_i->nat_list_lock);
297 
298 	set_nat_flag(ne, IS_DIRTY, false);
299 	set->entry_cnt--;
300 	nm_i->nat_cnt[DIRTY_NAT]--;
301 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
302 }
303 
304 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
305 		nid_t start, unsigned int nr, struct nat_entry_set **ep)
306 {
307 	return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
308 							start, nr);
309 }
310 
311 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
312 {
313 	return NODE_MAPPING(sbi) == page->mapping &&
314 			IS_DNODE(page) && is_cold_node(page);
315 }
316 
317 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
318 {
319 	spin_lock_init(&sbi->fsync_node_lock);
320 	INIT_LIST_HEAD(&sbi->fsync_node_list);
321 	sbi->fsync_seg_id = 0;
322 	sbi->fsync_node_num = 0;
323 }
324 
325 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
326 							struct page *page)
327 {
328 	struct fsync_node_entry *fn;
329 	unsigned long flags;
330 	unsigned int seq_id;
331 
332 	fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS);
333 
334 	get_page(page);
335 	fn->page = page;
336 	INIT_LIST_HEAD(&fn->list);
337 
338 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
339 	list_add_tail(&fn->list, &sbi->fsync_node_list);
340 	fn->seq_id = sbi->fsync_seg_id++;
341 	seq_id = fn->seq_id;
342 	sbi->fsync_node_num++;
343 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
344 
345 	return seq_id;
346 }
347 
348 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
349 {
350 	struct fsync_node_entry *fn;
351 	unsigned long flags;
352 
353 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
354 	list_for_each_entry(fn, &sbi->fsync_node_list, list) {
355 		if (fn->page == page) {
356 			list_del(&fn->list);
357 			sbi->fsync_node_num--;
358 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
359 			kmem_cache_free(fsync_node_entry_slab, fn);
360 			put_page(page);
361 			return;
362 		}
363 	}
364 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
365 	f2fs_bug_on(sbi, 1);
366 }
367 
368 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
369 {
370 	unsigned long flags;
371 
372 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
373 	sbi->fsync_seg_id = 0;
374 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
375 }
376 
377 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
378 {
379 	struct f2fs_nm_info *nm_i = NM_I(sbi);
380 	struct nat_entry *e;
381 	bool need = false;
382 
383 	down_read(&nm_i->nat_tree_lock);
384 	e = __lookup_nat_cache(nm_i, nid);
385 	if (e) {
386 		if (!get_nat_flag(e, IS_CHECKPOINTED) &&
387 				!get_nat_flag(e, HAS_FSYNCED_INODE))
388 			need = true;
389 	}
390 	up_read(&nm_i->nat_tree_lock);
391 	return need;
392 }
393 
394 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
395 {
396 	struct f2fs_nm_info *nm_i = NM_I(sbi);
397 	struct nat_entry *e;
398 	bool is_cp = true;
399 
400 	down_read(&nm_i->nat_tree_lock);
401 	e = __lookup_nat_cache(nm_i, nid);
402 	if (e && !get_nat_flag(e, IS_CHECKPOINTED))
403 		is_cp = false;
404 	up_read(&nm_i->nat_tree_lock);
405 	return is_cp;
406 }
407 
408 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
409 {
410 	struct f2fs_nm_info *nm_i = NM_I(sbi);
411 	struct nat_entry *e;
412 	bool need_update = true;
413 
414 	down_read(&nm_i->nat_tree_lock);
415 	e = __lookup_nat_cache(nm_i, ino);
416 	if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
417 			(get_nat_flag(e, IS_CHECKPOINTED) ||
418 			 get_nat_flag(e, HAS_FSYNCED_INODE)))
419 		need_update = false;
420 	up_read(&nm_i->nat_tree_lock);
421 	return need_update;
422 }
423 
424 /* must be locked by nat_tree_lock */
425 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
426 						struct f2fs_nat_entry *ne)
427 {
428 	struct f2fs_nm_info *nm_i = NM_I(sbi);
429 	struct nat_entry *new, *e;
430 
431 	new = __alloc_nat_entry(nid, false);
432 	if (!new)
433 		return;
434 
435 	down_write(&nm_i->nat_tree_lock);
436 	e = __lookup_nat_cache(nm_i, nid);
437 	if (!e)
438 		e = __init_nat_entry(nm_i, new, ne, false);
439 	else
440 		f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
441 				nat_get_blkaddr(e) !=
442 					le32_to_cpu(ne->block_addr) ||
443 				nat_get_version(e) != ne->version);
444 	up_write(&nm_i->nat_tree_lock);
445 	if (e != new)
446 		__free_nat_entry(new);
447 }
448 
449 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
450 			block_t new_blkaddr, bool fsync_done)
451 {
452 	struct f2fs_nm_info *nm_i = NM_I(sbi);
453 	struct nat_entry *e;
454 	struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
455 
456 	down_write(&nm_i->nat_tree_lock);
457 	e = __lookup_nat_cache(nm_i, ni->nid);
458 	if (!e) {
459 		e = __init_nat_entry(nm_i, new, NULL, true);
460 		copy_node_info(&e->ni, ni);
461 		f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
462 	} else if (new_blkaddr == NEW_ADDR) {
463 		/*
464 		 * when nid is reallocated,
465 		 * previous nat entry can be remained in nat cache.
466 		 * So, reinitialize it with new information.
467 		 */
468 		copy_node_info(&e->ni, ni);
469 		f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
470 	}
471 	/* let's free early to reduce memory consumption */
472 	if (e != new)
473 		__free_nat_entry(new);
474 
475 	/* sanity check */
476 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
477 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
478 			new_blkaddr == NULL_ADDR);
479 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
480 			new_blkaddr == NEW_ADDR);
481 	f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
482 			new_blkaddr == NEW_ADDR);
483 
484 	/* increment version no as node is removed */
485 	if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
486 		unsigned char version = nat_get_version(e);
487 
488 		nat_set_version(e, inc_node_version(version));
489 	}
490 
491 	/* change address */
492 	nat_set_blkaddr(e, new_blkaddr);
493 	if (!__is_valid_data_blkaddr(new_blkaddr))
494 		set_nat_flag(e, IS_CHECKPOINTED, false);
495 	__set_nat_cache_dirty(nm_i, e);
496 
497 	/* update fsync_mark if its inode nat entry is still alive */
498 	if (ni->nid != ni->ino)
499 		e = __lookup_nat_cache(nm_i, ni->ino);
500 	if (e) {
501 		if (fsync_done && ni->nid == ni->ino)
502 			set_nat_flag(e, HAS_FSYNCED_INODE, true);
503 		set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
504 	}
505 	up_write(&nm_i->nat_tree_lock);
506 }
507 
508 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
509 {
510 	struct f2fs_nm_info *nm_i = NM_I(sbi);
511 	int nr = nr_shrink;
512 
513 	if (!down_write_trylock(&nm_i->nat_tree_lock))
514 		return 0;
515 
516 	spin_lock(&nm_i->nat_list_lock);
517 	while (nr_shrink) {
518 		struct nat_entry *ne;
519 
520 		if (list_empty(&nm_i->nat_entries))
521 			break;
522 
523 		ne = list_first_entry(&nm_i->nat_entries,
524 					struct nat_entry, list);
525 		list_del(&ne->list);
526 		spin_unlock(&nm_i->nat_list_lock);
527 
528 		__del_from_nat_cache(nm_i, ne);
529 		nr_shrink--;
530 
531 		spin_lock(&nm_i->nat_list_lock);
532 	}
533 	spin_unlock(&nm_i->nat_list_lock);
534 
535 	up_write(&nm_i->nat_tree_lock);
536 	return nr - nr_shrink;
537 }
538 
539 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
540 						struct node_info *ni)
541 {
542 	struct f2fs_nm_info *nm_i = NM_I(sbi);
543 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
544 	struct f2fs_journal *journal = curseg->journal;
545 	nid_t start_nid = START_NID(nid);
546 	struct f2fs_nat_block *nat_blk;
547 	struct page *page = NULL;
548 	struct f2fs_nat_entry ne;
549 	struct nat_entry *e;
550 	pgoff_t index;
551 	block_t blkaddr;
552 	int i;
553 
554 	ni->nid = nid;
555 
556 	/* Check nat cache */
557 	down_read(&nm_i->nat_tree_lock);
558 	e = __lookup_nat_cache(nm_i, nid);
559 	if (e) {
560 		ni->ino = nat_get_ino(e);
561 		ni->blk_addr = nat_get_blkaddr(e);
562 		ni->version = nat_get_version(e);
563 		up_read(&nm_i->nat_tree_lock);
564 		return 0;
565 	}
566 
567 	memset(&ne, 0, sizeof(struct f2fs_nat_entry));
568 
569 	/* Check current segment summary */
570 	down_read(&curseg->journal_rwsem);
571 	i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
572 	if (i >= 0) {
573 		ne = nat_in_journal(journal, i);
574 		node_info_from_raw_nat(ni, &ne);
575 	}
576 	up_read(&curseg->journal_rwsem);
577 	if (i >= 0) {
578 		up_read(&nm_i->nat_tree_lock);
579 		goto cache;
580 	}
581 
582 	/* Fill node_info from nat page */
583 	index = current_nat_addr(sbi, nid);
584 	up_read(&nm_i->nat_tree_lock);
585 
586 	page = f2fs_get_meta_page(sbi, index);
587 	if (IS_ERR(page))
588 		return PTR_ERR(page);
589 
590 	nat_blk = (struct f2fs_nat_block *)page_address(page);
591 	ne = nat_blk->entries[nid - start_nid];
592 	node_info_from_raw_nat(ni, &ne);
593 	f2fs_put_page(page, 1);
594 cache:
595 	blkaddr = le32_to_cpu(ne.block_addr);
596 	if (__is_valid_data_blkaddr(blkaddr) &&
597 		!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
598 		return -EFAULT;
599 
600 	/* cache nat entry */
601 	cache_nat_entry(sbi, nid, &ne);
602 	return 0;
603 }
604 
605 /*
606  * readahead MAX_RA_NODE number of node pages.
607  */
608 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
609 {
610 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
611 	struct blk_plug plug;
612 	int i, end;
613 	nid_t nid;
614 
615 	blk_start_plug(&plug);
616 
617 	/* Then, try readahead for siblings of the desired node */
618 	end = start + n;
619 	end = min(end, NIDS_PER_BLOCK);
620 	for (i = start; i < end; i++) {
621 		nid = get_nid(parent, i, false);
622 		f2fs_ra_node_page(sbi, nid);
623 	}
624 
625 	blk_finish_plug(&plug);
626 }
627 
628 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
629 {
630 	const long direct_index = ADDRS_PER_INODE(dn->inode);
631 	const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
632 	const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
633 	unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
634 	int cur_level = dn->cur_level;
635 	int max_level = dn->max_level;
636 	pgoff_t base = 0;
637 
638 	if (!dn->max_level)
639 		return pgofs + 1;
640 
641 	while (max_level-- > cur_level)
642 		skipped_unit *= NIDS_PER_BLOCK;
643 
644 	switch (dn->max_level) {
645 	case 3:
646 		base += 2 * indirect_blks;
647 		fallthrough;
648 	case 2:
649 		base += 2 * direct_blks;
650 		fallthrough;
651 	case 1:
652 		base += direct_index;
653 		break;
654 	default:
655 		f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
656 	}
657 
658 	return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
659 }
660 
661 /*
662  * The maximum depth is four.
663  * Offset[0] will have raw inode offset.
664  */
665 static int get_node_path(struct inode *inode, long block,
666 				int offset[4], unsigned int noffset[4])
667 {
668 	const long direct_index = ADDRS_PER_INODE(inode);
669 	const long direct_blks = ADDRS_PER_BLOCK(inode);
670 	const long dptrs_per_blk = NIDS_PER_BLOCK;
671 	const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
672 	const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
673 	int n = 0;
674 	int level = 0;
675 
676 	noffset[0] = 0;
677 
678 	if (block < direct_index) {
679 		offset[n] = block;
680 		goto got;
681 	}
682 	block -= direct_index;
683 	if (block < direct_blks) {
684 		offset[n++] = NODE_DIR1_BLOCK;
685 		noffset[n] = 1;
686 		offset[n] = block;
687 		level = 1;
688 		goto got;
689 	}
690 	block -= direct_blks;
691 	if (block < direct_blks) {
692 		offset[n++] = NODE_DIR2_BLOCK;
693 		noffset[n] = 2;
694 		offset[n] = block;
695 		level = 1;
696 		goto got;
697 	}
698 	block -= direct_blks;
699 	if (block < indirect_blks) {
700 		offset[n++] = NODE_IND1_BLOCK;
701 		noffset[n] = 3;
702 		offset[n++] = block / direct_blks;
703 		noffset[n] = 4 + offset[n - 1];
704 		offset[n] = block % direct_blks;
705 		level = 2;
706 		goto got;
707 	}
708 	block -= indirect_blks;
709 	if (block < indirect_blks) {
710 		offset[n++] = NODE_IND2_BLOCK;
711 		noffset[n] = 4 + dptrs_per_blk;
712 		offset[n++] = block / direct_blks;
713 		noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
714 		offset[n] = block % direct_blks;
715 		level = 2;
716 		goto got;
717 	}
718 	block -= indirect_blks;
719 	if (block < dindirect_blks) {
720 		offset[n++] = NODE_DIND_BLOCK;
721 		noffset[n] = 5 + (dptrs_per_blk * 2);
722 		offset[n++] = block / indirect_blks;
723 		noffset[n] = 6 + (dptrs_per_blk * 2) +
724 			      offset[n - 1] * (dptrs_per_blk + 1);
725 		offset[n++] = (block / direct_blks) % dptrs_per_blk;
726 		noffset[n] = 7 + (dptrs_per_blk * 2) +
727 			      offset[n - 2] * (dptrs_per_blk + 1) +
728 			      offset[n - 1];
729 		offset[n] = block % direct_blks;
730 		level = 3;
731 		goto got;
732 	} else {
733 		return -E2BIG;
734 	}
735 got:
736 	return level;
737 }
738 
739 /*
740  * Caller should call f2fs_put_dnode(dn).
741  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
742  * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
743  */
744 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
745 {
746 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
747 	struct page *npage[4];
748 	struct page *parent = NULL;
749 	int offset[4];
750 	unsigned int noffset[4];
751 	nid_t nids[4];
752 	int level, i = 0;
753 	int err = 0;
754 
755 	level = get_node_path(dn->inode, index, offset, noffset);
756 	if (level < 0)
757 		return level;
758 
759 	nids[0] = dn->inode->i_ino;
760 	npage[0] = dn->inode_page;
761 
762 	if (!npage[0]) {
763 		npage[0] = f2fs_get_node_page(sbi, nids[0]);
764 		if (IS_ERR(npage[0]))
765 			return PTR_ERR(npage[0]);
766 	}
767 
768 	/* if inline_data is set, should not report any block indices */
769 	if (f2fs_has_inline_data(dn->inode) && index) {
770 		err = -ENOENT;
771 		f2fs_put_page(npage[0], 1);
772 		goto release_out;
773 	}
774 
775 	parent = npage[0];
776 	if (level != 0)
777 		nids[1] = get_nid(parent, offset[0], true);
778 	dn->inode_page = npage[0];
779 	dn->inode_page_locked = true;
780 
781 	/* get indirect or direct nodes */
782 	for (i = 1; i <= level; i++) {
783 		bool done = false;
784 
785 		if (!nids[i] && mode == ALLOC_NODE) {
786 			/* alloc new node */
787 			if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
788 				err = -ENOSPC;
789 				goto release_pages;
790 			}
791 
792 			dn->nid = nids[i];
793 			npage[i] = f2fs_new_node_page(dn, noffset[i]);
794 			if (IS_ERR(npage[i])) {
795 				f2fs_alloc_nid_failed(sbi, nids[i]);
796 				err = PTR_ERR(npage[i]);
797 				goto release_pages;
798 			}
799 
800 			set_nid(parent, offset[i - 1], nids[i], i == 1);
801 			f2fs_alloc_nid_done(sbi, nids[i]);
802 			done = true;
803 		} else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
804 			npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
805 			if (IS_ERR(npage[i])) {
806 				err = PTR_ERR(npage[i]);
807 				goto release_pages;
808 			}
809 			done = true;
810 		}
811 		if (i == 1) {
812 			dn->inode_page_locked = false;
813 			unlock_page(parent);
814 		} else {
815 			f2fs_put_page(parent, 1);
816 		}
817 
818 		if (!done) {
819 			npage[i] = f2fs_get_node_page(sbi, nids[i]);
820 			if (IS_ERR(npage[i])) {
821 				err = PTR_ERR(npage[i]);
822 				f2fs_put_page(npage[0], 0);
823 				goto release_out;
824 			}
825 		}
826 		if (i < level) {
827 			parent = npage[i];
828 			nids[i + 1] = get_nid(parent, offset[i], false);
829 		}
830 	}
831 	dn->nid = nids[level];
832 	dn->ofs_in_node = offset[level];
833 	dn->node_page = npage[level];
834 	dn->data_blkaddr = f2fs_data_blkaddr(dn);
835 	return 0;
836 
837 release_pages:
838 	f2fs_put_page(parent, 1);
839 	if (i > 1)
840 		f2fs_put_page(npage[0], 0);
841 release_out:
842 	dn->inode_page = NULL;
843 	dn->node_page = NULL;
844 	if (err == -ENOENT) {
845 		dn->cur_level = i;
846 		dn->max_level = level;
847 		dn->ofs_in_node = offset[level];
848 	}
849 	return err;
850 }
851 
852 static int truncate_node(struct dnode_of_data *dn)
853 {
854 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
855 	struct node_info ni;
856 	int err;
857 	pgoff_t index;
858 
859 	err = f2fs_get_node_info(sbi, dn->nid, &ni);
860 	if (err)
861 		return err;
862 
863 	/* Deallocate node address */
864 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
865 	dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
866 	set_node_addr(sbi, &ni, NULL_ADDR, false);
867 
868 	if (dn->nid == dn->inode->i_ino) {
869 		f2fs_remove_orphan_inode(sbi, dn->nid);
870 		dec_valid_inode_count(sbi);
871 		f2fs_inode_synced(dn->inode);
872 	}
873 
874 	clear_node_page_dirty(dn->node_page);
875 	set_sbi_flag(sbi, SBI_IS_DIRTY);
876 
877 	index = dn->node_page->index;
878 	f2fs_put_page(dn->node_page, 1);
879 
880 	invalidate_mapping_pages(NODE_MAPPING(sbi),
881 			index, index);
882 
883 	dn->node_page = NULL;
884 	trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
885 
886 	return 0;
887 }
888 
889 static int truncate_dnode(struct dnode_of_data *dn)
890 {
891 	struct page *page;
892 	int err;
893 
894 	if (dn->nid == 0)
895 		return 1;
896 
897 	/* get direct node */
898 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
899 	if (PTR_ERR(page) == -ENOENT)
900 		return 1;
901 	else if (IS_ERR(page))
902 		return PTR_ERR(page);
903 
904 	/* Make dnode_of_data for parameter */
905 	dn->node_page = page;
906 	dn->ofs_in_node = 0;
907 	f2fs_truncate_data_blocks(dn);
908 	err = truncate_node(dn);
909 	if (err)
910 		return err;
911 
912 	return 1;
913 }
914 
915 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
916 						int ofs, int depth)
917 {
918 	struct dnode_of_data rdn = *dn;
919 	struct page *page;
920 	struct f2fs_node *rn;
921 	nid_t child_nid;
922 	unsigned int child_nofs;
923 	int freed = 0;
924 	int i, ret;
925 
926 	if (dn->nid == 0)
927 		return NIDS_PER_BLOCK + 1;
928 
929 	trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
930 
931 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
932 	if (IS_ERR(page)) {
933 		trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
934 		return PTR_ERR(page);
935 	}
936 
937 	f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
938 
939 	rn = F2FS_NODE(page);
940 	if (depth < 3) {
941 		for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
942 			child_nid = le32_to_cpu(rn->in.nid[i]);
943 			if (child_nid == 0)
944 				continue;
945 			rdn.nid = child_nid;
946 			ret = truncate_dnode(&rdn);
947 			if (ret < 0)
948 				goto out_err;
949 			if (set_nid(page, i, 0, false))
950 				dn->node_changed = true;
951 		}
952 	} else {
953 		child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
954 		for (i = ofs; i < NIDS_PER_BLOCK; i++) {
955 			child_nid = le32_to_cpu(rn->in.nid[i]);
956 			if (child_nid == 0) {
957 				child_nofs += NIDS_PER_BLOCK + 1;
958 				continue;
959 			}
960 			rdn.nid = child_nid;
961 			ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
962 			if (ret == (NIDS_PER_BLOCK + 1)) {
963 				if (set_nid(page, i, 0, false))
964 					dn->node_changed = true;
965 				child_nofs += ret;
966 			} else if (ret < 0 && ret != -ENOENT) {
967 				goto out_err;
968 			}
969 		}
970 		freed = child_nofs;
971 	}
972 
973 	if (!ofs) {
974 		/* remove current indirect node */
975 		dn->node_page = page;
976 		ret = truncate_node(dn);
977 		if (ret)
978 			goto out_err;
979 		freed++;
980 	} else {
981 		f2fs_put_page(page, 1);
982 	}
983 	trace_f2fs_truncate_nodes_exit(dn->inode, freed);
984 	return freed;
985 
986 out_err:
987 	f2fs_put_page(page, 1);
988 	trace_f2fs_truncate_nodes_exit(dn->inode, ret);
989 	return ret;
990 }
991 
992 static int truncate_partial_nodes(struct dnode_of_data *dn,
993 			struct f2fs_inode *ri, int *offset, int depth)
994 {
995 	struct page *pages[2];
996 	nid_t nid[3];
997 	nid_t child_nid;
998 	int err = 0;
999 	int i;
1000 	int idx = depth - 2;
1001 
1002 	nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1003 	if (!nid[0])
1004 		return 0;
1005 
1006 	/* get indirect nodes in the path */
1007 	for (i = 0; i < idx + 1; i++) {
1008 		/* reference count'll be increased */
1009 		pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
1010 		if (IS_ERR(pages[i])) {
1011 			err = PTR_ERR(pages[i]);
1012 			idx = i - 1;
1013 			goto fail;
1014 		}
1015 		nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1016 	}
1017 
1018 	f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1019 
1020 	/* free direct nodes linked to a partial indirect node */
1021 	for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1022 		child_nid = get_nid(pages[idx], i, false);
1023 		if (!child_nid)
1024 			continue;
1025 		dn->nid = child_nid;
1026 		err = truncate_dnode(dn);
1027 		if (err < 0)
1028 			goto fail;
1029 		if (set_nid(pages[idx], i, 0, false))
1030 			dn->node_changed = true;
1031 	}
1032 
1033 	if (offset[idx + 1] == 0) {
1034 		dn->node_page = pages[idx];
1035 		dn->nid = nid[idx];
1036 		err = truncate_node(dn);
1037 		if (err)
1038 			goto fail;
1039 	} else {
1040 		f2fs_put_page(pages[idx], 1);
1041 	}
1042 	offset[idx]++;
1043 	offset[idx + 1] = 0;
1044 	idx--;
1045 fail:
1046 	for (i = idx; i >= 0; i--)
1047 		f2fs_put_page(pages[i], 1);
1048 
1049 	trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1050 
1051 	return err;
1052 }
1053 
1054 /*
1055  * All the block addresses of data and nodes should be nullified.
1056  */
1057 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1058 {
1059 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1060 	int err = 0, cont = 1;
1061 	int level, offset[4], noffset[4];
1062 	unsigned int nofs = 0;
1063 	struct f2fs_inode *ri;
1064 	struct dnode_of_data dn;
1065 	struct page *page;
1066 
1067 	trace_f2fs_truncate_inode_blocks_enter(inode, from);
1068 
1069 	level = get_node_path(inode, from, offset, noffset);
1070 	if (level < 0) {
1071 		trace_f2fs_truncate_inode_blocks_exit(inode, level);
1072 		return level;
1073 	}
1074 
1075 	page = f2fs_get_node_page(sbi, inode->i_ino);
1076 	if (IS_ERR(page)) {
1077 		trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1078 		return PTR_ERR(page);
1079 	}
1080 
1081 	set_new_dnode(&dn, inode, page, NULL, 0);
1082 	unlock_page(page);
1083 
1084 	ri = F2FS_INODE(page);
1085 	switch (level) {
1086 	case 0:
1087 	case 1:
1088 		nofs = noffset[1];
1089 		break;
1090 	case 2:
1091 		nofs = noffset[1];
1092 		if (!offset[level - 1])
1093 			goto skip_partial;
1094 		err = truncate_partial_nodes(&dn, ri, offset, level);
1095 		if (err < 0 && err != -ENOENT)
1096 			goto fail;
1097 		nofs += 1 + NIDS_PER_BLOCK;
1098 		break;
1099 	case 3:
1100 		nofs = 5 + 2 * NIDS_PER_BLOCK;
1101 		if (!offset[level - 1])
1102 			goto skip_partial;
1103 		err = truncate_partial_nodes(&dn, ri, offset, level);
1104 		if (err < 0 && err != -ENOENT)
1105 			goto fail;
1106 		break;
1107 	default:
1108 		BUG();
1109 	}
1110 
1111 skip_partial:
1112 	while (cont) {
1113 		dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1114 		switch (offset[0]) {
1115 		case NODE_DIR1_BLOCK:
1116 		case NODE_DIR2_BLOCK:
1117 			err = truncate_dnode(&dn);
1118 			break;
1119 
1120 		case NODE_IND1_BLOCK:
1121 		case NODE_IND2_BLOCK:
1122 			err = truncate_nodes(&dn, nofs, offset[1], 2);
1123 			break;
1124 
1125 		case NODE_DIND_BLOCK:
1126 			err = truncate_nodes(&dn, nofs, offset[1], 3);
1127 			cont = 0;
1128 			break;
1129 
1130 		default:
1131 			BUG();
1132 		}
1133 		if (err < 0 && err != -ENOENT)
1134 			goto fail;
1135 		if (offset[1] == 0 &&
1136 				ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1137 			lock_page(page);
1138 			BUG_ON(page->mapping != NODE_MAPPING(sbi));
1139 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1140 			ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1141 			set_page_dirty(page);
1142 			unlock_page(page);
1143 		}
1144 		offset[1] = 0;
1145 		offset[0]++;
1146 		nofs += err;
1147 	}
1148 fail:
1149 	f2fs_put_page(page, 0);
1150 	trace_f2fs_truncate_inode_blocks_exit(inode, err);
1151 	return err > 0 ? 0 : err;
1152 }
1153 
1154 /* caller must lock inode page */
1155 int f2fs_truncate_xattr_node(struct inode *inode)
1156 {
1157 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1158 	nid_t nid = F2FS_I(inode)->i_xattr_nid;
1159 	struct dnode_of_data dn;
1160 	struct page *npage;
1161 	int err;
1162 
1163 	if (!nid)
1164 		return 0;
1165 
1166 	npage = f2fs_get_node_page(sbi, nid);
1167 	if (IS_ERR(npage))
1168 		return PTR_ERR(npage);
1169 
1170 	set_new_dnode(&dn, inode, NULL, npage, nid);
1171 	err = truncate_node(&dn);
1172 	if (err) {
1173 		f2fs_put_page(npage, 1);
1174 		return err;
1175 	}
1176 
1177 	f2fs_i_xnid_write(inode, 0);
1178 
1179 	return 0;
1180 }
1181 
1182 /*
1183  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1184  * f2fs_unlock_op().
1185  */
1186 int f2fs_remove_inode_page(struct inode *inode)
1187 {
1188 	struct dnode_of_data dn;
1189 	int err;
1190 
1191 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1192 	err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1193 	if (err)
1194 		return err;
1195 
1196 	err = f2fs_truncate_xattr_node(inode);
1197 	if (err) {
1198 		f2fs_put_dnode(&dn);
1199 		return err;
1200 	}
1201 
1202 	/* remove potential inline_data blocks */
1203 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1204 				S_ISLNK(inode->i_mode))
1205 		f2fs_truncate_data_blocks_range(&dn, 1);
1206 
1207 	/* 0 is possible, after f2fs_new_inode() has failed */
1208 	if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1209 		f2fs_put_dnode(&dn);
1210 		return -EIO;
1211 	}
1212 
1213 	if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1214 		f2fs_warn(F2FS_I_SB(inode),
1215 			"f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1216 			inode->i_ino, (unsigned long long)inode->i_blocks);
1217 		set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1218 	}
1219 
1220 	/* will put inode & node pages */
1221 	err = truncate_node(&dn);
1222 	if (err) {
1223 		f2fs_put_dnode(&dn);
1224 		return err;
1225 	}
1226 	return 0;
1227 }
1228 
1229 struct page *f2fs_new_inode_page(struct inode *inode)
1230 {
1231 	struct dnode_of_data dn;
1232 
1233 	/* allocate inode page for new inode */
1234 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1235 
1236 	/* caller should f2fs_put_page(page, 1); */
1237 	return f2fs_new_node_page(&dn, 0);
1238 }
1239 
1240 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1241 {
1242 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1243 	struct node_info new_ni;
1244 	struct page *page;
1245 	int err;
1246 
1247 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1248 		return ERR_PTR(-EPERM);
1249 
1250 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1251 	if (!page)
1252 		return ERR_PTR(-ENOMEM);
1253 
1254 	if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1255 		goto fail;
1256 
1257 #ifdef CONFIG_F2FS_CHECK_FS
1258 	err = f2fs_get_node_info(sbi, dn->nid, &new_ni);
1259 	if (err) {
1260 		dec_valid_node_count(sbi, dn->inode, !ofs);
1261 		goto fail;
1262 	}
1263 	f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1264 #endif
1265 	new_ni.nid = dn->nid;
1266 	new_ni.ino = dn->inode->i_ino;
1267 	new_ni.blk_addr = NULL_ADDR;
1268 	new_ni.flag = 0;
1269 	new_ni.version = 0;
1270 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1271 
1272 	f2fs_wait_on_page_writeback(page, NODE, true, true);
1273 	fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1274 	set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1275 	if (!PageUptodate(page))
1276 		SetPageUptodate(page);
1277 	if (set_page_dirty(page))
1278 		dn->node_changed = true;
1279 
1280 	if (f2fs_has_xattr_block(ofs))
1281 		f2fs_i_xnid_write(dn->inode, dn->nid);
1282 
1283 	if (ofs == 0)
1284 		inc_valid_inode_count(sbi);
1285 	return page;
1286 
1287 fail:
1288 	clear_node_page_dirty(page);
1289 	f2fs_put_page(page, 1);
1290 	return ERR_PTR(err);
1291 }
1292 
1293 /*
1294  * Caller should do after getting the following values.
1295  * 0: f2fs_put_page(page, 0)
1296  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1297  */
1298 static int read_node_page(struct page *page, int op_flags)
1299 {
1300 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1301 	struct node_info ni;
1302 	struct f2fs_io_info fio = {
1303 		.sbi = sbi,
1304 		.type = NODE,
1305 		.op = REQ_OP_READ,
1306 		.op_flags = op_flags,
1307 		.page = page,
1308 		.encrypted_page = NULL,
1309 	};
1310 	int err;
1311 
1312 	if (PageUptodate(page)) {
1313 		if (!f2fs_inode_chksum_verify(sbi, page)) {
1314 			ClearPageUptodate(page);
1315 			return -EFSBADCRC;
1316 		}
1317 		return LOCKED_PAGE;
1318 	}
1319 
1320 	err = f2fs_get_node_info(sbi, page->index, &ni);
1321 	if (err)
1322 		return err;
1323 
1324 	if (unlikely(ni.blk_addr == NULL_ADDR) ||
1325 			is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1326 		ClearPageUptodate(page);
1327 		return -ENOENT;
1328 	}
1329 
1330 	fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1331 
1332 	err = f2fs_submit_page_bio(&fio);
1333 
1334 	if (!err)
1335 		f2fs_update_iostat(sbi, FS_NODE_READ_IO, F2FS_BLKSIZE);
1336 
1337 	return err;
1338 }
1339 
1340 /*
1341  * Readahead a node page
1342  */
1343 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1344 {
1345 	struct page *apage;
1346 	int err;
1347 
1348 	if (!nid)
1349 		return;
1350 	if (f2fs_check_nid_range(sbi, nid))
1351 		return;
1352 
1353 	apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1354 	if (apage)
1355 		return;
1356 
1357 	apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1358 	if (!apage)
1359 		return;
1360 
1361 	err = read_node_page(apage, REQ_RAHEAD);
1362 	f2fs_put_page(apage, err ? 1 : 0);
1363 }
1364 
1365 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1366 					struct page *parent, int start)
1367 {
1368 	struct page *page;
1369 	int err;
1370 
1371 	if (!nid)
1372 		return ERR_PTR(-ENOENT);
1373 	if (f2fs_check_nid_range(sbi, nid))
1374 		return ERR_PTR(-EINVAL);
1375 repeat:
1376 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1377 	if (!page)
1378 		return ERR_PTR(-ENOMEM);
1379 
1380 	err = read_node_page(page, 0);
1381 	if (err < 0) {
1382 		f2fs_put_page(page, 1);
1383 		return ERR_PTR(err);
1384 	} else if (err == LOCKED_PAGE) {
1385 		err = 0;
1386 		goto page_hit;
1387 	}
1388 
1389 	if (parent)
1390 		f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1391 
1392 	lock_page(page);
1393 
1394 	if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1395 		f2fs_put_page(page, 1);
1396 		goto repeat;
1397 	}
1398 
1399 	if (unlikely(!PageUptodate(page))) {
1400 		err = -EIO;
1401 		goto out_err;
1402 	}
1403 
1404 	if (!f2fs_inode_chksum_verify(sbi, page)) {
1405 		err = -EFSBADCRC;
1406 		goto out_err;
1407 	}
1408 page_hit:
1409 	if (unlikely(nid != nid_of_node(page))) {
1410 		f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1411 			  nid, nid_of_node(page), ino_of_node(page),
1412 			  ofs_of_node(page), cpver_of_node(page),
1413 			  next_blkaddr_of_node(page));
1414 		err = -EINVAL;
1415 out_err:
1416 		ClearPageUptodate(page);
1417 		f2fs_put_page(page, 1);
1418 		return ERR_PTR(err);
1419 	}
1420 	return page;
1421 }
1422 
1423 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1424 {
1425 	return __get_node_page(sbi, nid, NULL, 0);
1426 }
1427 
1428 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1429 {
1430 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1431 	nid_t nid = get_nid(parent, start, false);
1432 
1433 	return __get_node_page(sbi, nid, parent, start);
1434 }
1435 
1436 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1437 {
1438 	struct inode *inode;
1439 	struct page *page;
1440 	int ret;
1441 
1442 	/* should flush inline_data before evict_inode */
1443 	inode = ilookup(sbi->sb, ino);
1444 	if (!inode)
1445 		return;
1446 
1447 	page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1448 					FGP_LOCK|FGP_NOWAIT, 0);
1449 	if (!page)
1450 		goto iput_out;
1451 
1452 	if (!PageUptodate(page))
1453 		goto page_out;
1454 
1455 	if (!PageDirty(page))
1456 		goto page_out;
1457 
1458 	if (!clear_page_dirty_for_io(page))
1459 		goto page_out;
1460 
1461 	ret = f2fs_write_inline_data(inode, page);
1462 	inode_dec_dirty_pages(inode);
1463 	f2fs_remove_dirty_inode(inode);
1464 	if (ret)
1465 		set_page_dirty(page);
1466 page_out:
1467 	f2fs_put_page(page, 1);
1468 iput_out:
1469 	iput(inode);
1470 }
1471 
1472 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1473 {
1474 	pgoff_t index;
1475 	struct pagevec pvec;
1476 	struct page *last_page = NULL;
1477 	int nr_pages;
1478 
1479 	pagevec_init(&pvec);
1480 	index = 0;
1481 
1482 	while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1483 				PAGECACHE_TAG_DIRTY))) {
1484 		int i;
1485 
1486 		for (i = 0; i < nr_pages; i++) {
1487 			struct page *page = pvec.pages[i];
1488 
1489 			if (unlikely(f2fs_cp_error(sbi))) {
1490 				f2fs_put_page(last_page, 0);
1491 				pagevec_release(&pvec);
1492 				return ERR_PTR(-EIO);
1493 			}
1494 
1495 			if (!IS_DNODE(page) || !is_cold_node(page))
1496 				continue;
1497 			if (ino_of_node(page) != ino)
1498 				continue;
1499 
1500 			lock_page(page);
1501 
1502 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1503 continue_unlock:
1504 				unlock_page(page);
1505 				continue;
1506 			}
1507 			if (ino_of_node(page) != ino)
1508 				goto continue_unlock;
1509 
1510 			if (!PageDirty(page)) {
1511 				/* someone wrote it for us */
1512 				goto continue_unlock;
1513 			}
1514 
1515 			if (last_page)
1516 				f2fs_put_page(last_page, 0);
1517 
1518 			get_page(page);
1519 			last_page = page;
1520 			unlock_page(page);
1521 		}
1522 		pagevec_release(&pvec);
1523 		cond_resched();
1524 	}
1525 	return last_page;
1526 }
1527 
1528 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1529 				struct writeback_control *wbc, bool do_balance,
1530 				enum iostat_type io_type, unsigned int *seq_id)
1531 {
1532 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1533 	nid_t nid;
1534 	struct node_info ni;
1535 	struct f2fs_io_info fio = {
1536 		.sbi = sbi,
1537 		.ino = ino_of_node(page),
1538 		.type = NODE,
1539 		.op = REQ_OP_WRITE,
1540 		.op_flags = wbc_to_write_flags(wbc),
1541 		.page = page,
1542 		.encrypted_page = NULL,
1543 		.submitted = false,
1544 		.io_type = io_type,
1545 		.io_wbc = wbc,
1546 	};
1547 	unsigned int seq;
1548 
1549 	trace_f2fs_writepage(page, NODE);
1550 
1551 	if (unlikely(f2fs_cp_error(sbi))) {
1552 		ClearPageUptodate(page);
1553 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1554 		unlock_page(page);
1555 		return 0;
1556 	}
1557 
1558 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1559 		goto redirty_out;
1560 
1561 	if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1562 			wbc->sync_mode == WB_SYNC_NONE &&
1563 			IS_DNODE(page) && is_cold_node(page))
1564 		goto redirty_out;
1565 
1566 	/* get old block addr of this node page */
1567 	nid = nid_of_node(page);
1568 	f2fs_bug_on(sbi, page->index != nid);
1569 
1570 	if (f2fs_get_node_info(sbi, nid, &ni))
1571 		goto redirty_out;
1572 
1573 	if (wbc->for_reclaim) {
1574 		if (!down_read_trylock(&sbi->node_write))
1575 			goto redirty_out;
1576 	} else {
1577 		down_read(&sbi->node_write);
1578 	}
1579 
1580 	/* This page is already truncated */
1581 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1582 		ClearPageUptodate(page);
1583 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1584 		up_read(&sbi->node_write);
1585 		unlock_page(page);
1586 		return 0;
1587 	}
1588 
1589 	if (__is_valid_data_blkaddr(ni.blk_addr) &&
1590 		!f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1591 					DATA_GENERIC_ENHANCE)) {
1592 		up_read(&sbi->node_write);
1593 		goto redirty_out;
1594 	}
1595 
1596 	if (atomic && !test_opt(sbi, NOBARRIER))
1597 		fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1598 
1599 	/* should add to global list before clearing PAGECACHE status */
1600 	if (f2fs_in_warm_node_list(sbi, page)) {
1601 		seq = f2fs_add_fsync_node_entry(sbi, page);
1602 		if (seq_id)
1603 			*seq_id = seq;
1604 	}
1605 
1606 	set_page_writeback(page);
1607 	ClearPageError(page);
1608 
1609 	fio.old_blkaddr = ni.blk_addr;
1610 	f2fs_do_write_node_page(nid, &fio);
1611 	set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1612 	dec_page_count(sbi, F2FS_DIRTY_NODES);
1613 	up_read(&sbi->node_write);
1614 
1615 	if (wbc->for_reclaim) {
1616 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1617 		submitted = NULL;
1618 	}
1619 
1620 	unlock_page(page);
1621 
1622 	if (unlikely(f2fs_cp_error(sbi))) {
1623 		f2fs_submit_merged_write(sbi, NODE);
1624 		submitted = NULL;
1625 	}
1626 	if (submitted)
1627 		*submitted = fio.submitted;
1628 
1629 	if (do_balance)
1630 		f2fs_balance_fs(sbi, false);
1631 	return 0;
1632 
1633 redirty_out:
1634 	redirty_page_for_writepage(wbc, page);
1635 	return AOP_WRITEPAGE_ACTIVATE;
1636 }
1637 
1638 int f2fs_move_node_page(struct page *node_page, int gc_type)
1639 {
1640 	int err = 0;
1641 
1642 	if (gc_type == FG_GC) {
1643 		struct writeback_control wbc = {
1644 			.sync_mode = WB_SYNC_ALL,
1645 			.nr_to_write = 1,
1646 			.for_reclaim = 0,
1647 		};
1648 
1649 		f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1650 
1651 		set_page_dirty(node_page);
1652 
1653 		if (!clear_page_dirty_for_io(node_page)) {
1654 			err = -EAGAIN;
1655 			goto out_page;
1656 		}
1657 
1658 		if (__write_node_page(node_page, false, NULL,
1659 					&wbc, false, FS_GC_NODE_IO, NULL)) {
1660 			err = -EAGAIN;
1661 			unlock_page(node_page);
1662 		}
1663 		goto release_page;
1664 	} else {
1665 		/* set page dirty and write it */
1666 		if (!PageWriteback(node_page))
1667 			set_page_dirty(node_page);
1668 	}
1669 out_page:
1670 	unlock_page(node_page);
1671 release_page:
1672 	f2fs_put_page(node_page, 0);
1673 	return err;
1674 }
1675 
1676 static int f2fs_write_node_page(struct page *page,
1677 				struct writeback_control *wbc)
1678 {
1679 	return __write_node_page(page, false, NULL, wbc, false,
1680 						FS_NODE_IO, NULL);
1681 }
1682 
1683 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1684 			struct writeback_control *wbc, bool atomic,
1685 			unsigned int *seq_id)
1686 {
1687 	pgoff_t index;
1688 	struct pagevec pvec;
1689 	int ret = 0;
1690 	struct page *last_page = NULL;
1691 	bool marked = false;
1692 	nid_t ino = inode->i_ino;
1693 	int nr_pages;
1694 	int nwritten = 0;
1695 
1696 	if (atomic) {
1697 		last_page = last_fsync_dnode(sbi, ino);
1698 		if (IS_ERR_OR_NULL(last_page))
1699 			return PTR_ERR_OR_ZERO(last_page);
1700 	}
1701 retry:
1702 	pagevec_init(&pvec);
1703 	index = 0;
1704 
1705 	while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1706 				PAGECACHE_TAG_DIRTY))) {
1707 		int i;
1708 
1709 		for (i = 0; i < nr_pages; i++) {
1710 			struct page *page = pvec.pages[i];
1711 			bool submitted = false;
1712 
1713 			if (unlikely(f2fs_cp_error(sbi))) {
1714 				f2fs_put_page(last_page, 0);
1715 				pagevec_release(&pvec);
1716 				ret = -EIO;
1717 				goto out;
1718 			}
1719 
1720 			if (!IS_DNODE(page) || !is_cold_node(page))
1721 				continue;
1722 			if (ino_of_node(page) != ino)
1723 				continue;
1724 
1725 			lock_page(page);
1726 
1727 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1728 continue_unlock:
1729 				unlock_page(page);
1730 				continue;
1731 			}
1732 			if (ino_of_node(page) != ino)
1733 				goto continue_unlock;
1734 
1735 			if (!PageDirty(page) && page != last_page) {
1736 				/* someone wrote it for us */
1737 				goto continue_unlock;
1738 			}
1739 
1740 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1741 
1742 			set_fsync_mark(page, 0);
1743 			set_dentry_mark(page, 0);
1744 
1745 			if (!atomic || page == last_page) {
1746 				set_fsync_mark(page, 1);
1747 				if (IS_INODE(page)) {
1748 					if (is_inode_flag_set(inode,
1749 								FI_DIRTY_INODE))
1750 						f2fs_update_inode(inode, page);
1751 					set_dentry_mark(page,
1752 						f2fs_need_dentry_mark(sbi, ino));
1753 				}
1754 				/* may be written by other thread */
1755 				if (!PageDirty(page))
1756 					set_page_dirty(page);
1757 			}
1758 
1759 			if (!clear_page_dirty_for_io(page))
1760 				goto continue_unlock;
1761 
1762 			ret = __write_node_page(page, atomic &&
1763 						page == last_page,
1764 						&submitted, wbc, true,
1765 						FS_NODE_IO, seq_id);
1766 			if (ret) {
1767 				unlock_page(page);
1768 				f2fs_put_page(last_page, 0);
1769 				break;
1770 			} else if (submitted) {
1771 				nwritten++;
1772 			}
1773 
1774 			if (page == last_page) {
1775 				f2fs_put_page(page, 0);
1776 				marked = true;
1777 				break;
1778 			}
1779 		}
1780 		pagevec_release(&pvec);
1781 		cond_resched();
1782 
1783 		if (ret || marked)
1784 			break;
1785 	}
1786 	if (!ret && atomic && !marked) {
1787 		f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1788 			   ino, last_page->index);
1789 		lock_page(last_page);
1790 		f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1791 		set_page_dirty(last_page);
1792 		unlock_page(last_page);
1793 		goto retry;
1794 	}
1795 out:
1796 	if (nwritten)
1797 		f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1798 	return ret ? -EIO : 0;
1799 }
1800 
1801 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1802 {
1803 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1804 	bool clean;
1805 
1806 	if (inode->i_ino != ino)
1807 		return 0;
1808 
1809 	if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1810 		return 0;
1811 
1812 	spin_lock(&sbi->inode_lock[DIRTY_META]);
1813 	clean = list_empty(&F2FS_I(inode)->gdirty_list);
1814 	spin_unlock(&sbi->inode_lock[DIRTY_META]);
1815 
1816 	if (clean)
1817 		return 0;
1818 
1819 	inode = igrab(inode);
1820 	if (!inode)
1821 		return 0;
1822 	return 1;
1823 }
1824 
1825 static bool flush_dirty_inode(struct page *page)
1826 {
1827 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1828 	struct inode *inode;
1829 	nid_t ino = ino_of_node(page);
1830 
1831 	inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1832 	if (!inode)
1833 		return false;
1834 
1835 	f2fs_update_inode(inode, page);
1836 	unlock_page(page);
1837 
1838 	iput(inode);
1839 	return true;
1840 }
1841 
1842 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1843 {
1844 	pgoff_t index = 0;
1845 	struct pagevec pvec;
1846 	int nr_pages;
1847 
1848 	pagevec_init(&pvec);
1849 
1850 	while ((nr_pages = pagevec_lookup_tag(&pvec,
1851 			NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1852 		int i;
1853 
1854 		for (i = 0; i < nr_pages; i++) {
1855 			struct page *page = pvec.pages[i];
1856 
1857 			if (!IS_DNODE(page))
1858 				continue;
1859 
1860 			lock_page(page);
1861 
1862 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1863 continue_unlock:
1864 				unlock_page(page);
1865 				continue;
1866 			}
1867 
1868 			if (!PageDirty(page)) {
1869 				/* someone wrote it for us */
1870 				goto continue_unlock;
1871 			}
1872 
1873 			/* flush inline_data, if it's async context. */
1874 			if (page_private_inline(page)) {
1875 				clear_page_private_inline(page);
1876 				unlock_page(page);
1877 				flush_inline_data(sbi, ino_of_node(page));
1878 				continue;
1879 			}
1880 			unlock_page(page);
1881 		}
1882 		pagevec_release(&pvec);
1883 		cond_resched();
1884 	}
1885 }
1886 
1887 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1888 				struct writeback_control *wbc,
1889 				bool do_balance, enum iostat_type io_type)
1890 {
1891 	pgoff_t index;
1892 	struct pagevec pvec;
1893 	int step = 0;
1894 	int nwritten = 0;
1895 	int ret = 0;
1896 	int nr_pages, done = 0;
1897 
1898 	pagevec_init(&pvec);
1899 
1900 next_step:
1901 	index = 0;
1902 
1903 	while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1904 			NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1905 		int i;
1906 
1907 		for (i = 0; i < nr_pages; i++) {
1908 			struct page *page = pvec.pages[i];
1909 			bool submitted = false;
1910 			bool may_dirty = true;
1911 
1912 			/* give a priority to WB_SYNC threads */
1913 			if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1914 					wbc->sync_mode == WB_SYNC_NONE) {
1915 				done = 1;
1916 				break;
1917 			}
1918 
1919 			/*
1920 			 * flushing sequence with step:
1921 			 * 0. indirect nodes
1922 			 * 1. dentry dnodes
1923 			 * 2. file dnodes
1924 			 */
1925 			if (step == 0 && IS_DNODE(page))
1926 				continue;
1927 			if (step == 1 && (!IS_DNODE(page) ||
1928 						is_cold_node(page)))
1929 				continue;
1930 			if (step == 2 && (!IS_DNODE(page) ||
1931 						!is_cold_node(page)))
1932 				continue;
1933 lock_node:
1934 			if (wbc->sync_mode == WB_SYNC_ALL)
1935 				lock_page(page);
1936 			else if (!trylock_page(page))
1937 				continue;
1938 
1939 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1940 continue_unlock:
1941 				unlock_page(page);
1942 				continue;
1943 			}
1944 
1945 			if (!PageDirty(page)) {
1946 				/* someone wrote it for us */
1947 				goto continue_unlock;
1948 			}
1949 
1950 			/* flush inline_data/inode, if it's async context. */
1951 			if (!do_balance)
1952 				goto write_node;
1953 
1954 			/* flush inline_data */
1955 			if (page_private_inline(page)) {
1956 				clear_page_private_inline(page);
1957 				unlock_page(page);
1958 				flush_inline_data(sbi, ino_of_node(page));
1959 				goto lock_node;
1960 			}
1961 
1962 			/* flush dirty inode */
1963 			if (IS_INODE(page) && may_dirty) {
1964 				may_dirty = false;
1965 				if (flush_dirty_inode(page))
1966 					goto lock_node;
1967 			}
1968 write_node:
1969 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1970 
1971 			if (!clear_page_dirty_for_io(page))
1972 				goto continue_unlock;
1973 
1974 			set_fsync_mark(page, 0);
1975 			set_dentry_mark(page, 0);
1976 
1977 			ret = __write_node_page(page, false, &submitted,
1978 						wbc, do_balance, io_type, NULL);
1979 			if (ret)
1980 				unlock_page(page);
1981 			else if (submitted)
1982 				nwritten++;
1983 
1984 			if (--wbc->nr_to_write == 0)
1985 				break;
1986 		}
1987 		pagevec_release(&pvec);
1988 		cond_resched();
1989 
1990 		if (wbc->nr_to_write == 0) {
1991 			step = 2;
1992 			break;
1993 		}
1994 	}
1995 
1996 	if (step < 2) {
1997 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1998 				wbc->sync_mode == WB_SYNC_NONE && step == 1)
1999 			goto out;
2000 		step++;
2001 		goto next_step;
2002 	}
2003 out:
2004 	if (nwritten)
2005 		f2fs_submit_merged_write(sbi, NODE);
2006 
2007 	if (unlikely(f2fs_cp_error(sbi)))
2008 		return -EIO;
2009 	return ret;
2010 }
2011 
2012 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2013 						unsigned int seq_id)
2014 {
2015 	struct fsync_node_entry *fn;
2016 	struct page *page;
2017 	struct list_head *head = &sbi->fsync_node_list;
2018 	unsigned long flags;
2019 	unsigned int cur_seq_id = 0;
2020 	int ret2, ret = 0;
2021 
2022 	while (seq_id && cur_seq_id < seq_id) {
2023 		spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2024 		if (list_empty(head)) {
2025 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2026 			break;
2027 		}
2028 		fn = list_first_entry(head, struct fsync_node_entry, list);
2029 		if (fn->seq_id > seq_id) {
2030 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2031 			break;
2032 		}
2033 		cur_seq_id = fn->seq_id;
2034 		page = fn->page;
2035 		get_page(page);
2036 		spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2037 
2038 		f2fs_wait_on_page_writeback(page, NODE, true, false);
2039 		if (TestClearPageError(page))
2040 			ret = -EIO;
2041 
2042 		put_page(page);
2043 
2044 		if (ret)
2045 			break;
2046 	}
2047 
2048 	ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2049 	if (!ret)
2050 		ret = ret2;
2051 
2052 	return ret;
2053 }
2054 
2055 static int f2fs_write_node_pages(struct address_space *mapping,
2056 			    struct writeback_control *wbc)
2057 {
2058 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2059 	struct blk_plug plug;
2060 	long diff;
2061 
2062 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2063 		goto skip_write;
2064 
2065 	/* balancing f2fs's metadata in background */
2066 	f2fs_balance_fs_bg(sbi, true);
2067 
2068 	/* collect a number of dirty node pages and write together */
2069 	if (wbc->sync_mode != WB_SYNC_ALL &&
2070 			get_pages(sbi, F2FS_DIRTY_NODES) <
2071 					nr_pages_to_skip(sbi, NODE))
2072 		goto skip_write;
2073 
2074 	if (wbc->sync_mode == WB_SYNC_ALL)
2075 		atomic_inc(&sbi->wb_sync_req[NODE]);
2076 	else if (atomic_read(&sbi->wb_sync_req[NODE]))
2077 		goto skip_write;
2078 
2079 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2080 
2081 	diff = nr_pages_to_write(sbi, NODE, wbc);
2082 	blk_start_plug(&plug);
2083 	f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2084 	blk_finish_plug(&plug);
2085 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2086 
2087 	if (wbc->sync_mode == WB_SYNC_ALL)
2088 		atomic_dec(&sbi->wb_sync_req[NODE]);
2089 	return 0;
2090 
2091 skip_write:
2092 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2093 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2094 	return 0;
2095 }
2096 
2097 static int f2fs_set_node_page_dirty(struct page *page)
2098 {
2099 	trace_f2fs_set_page_dirty(page, NODE);
2100 
2101 	if (!PageUptodate(page))
2102 		SetPageUptodate(page);
2103 #ifdef CONFIG_F2FS_CHECK_FS
2104 	if (IS_INODE(page))
2105 		f2fs_inode_chksum_set(F2FS_P_SB(page), page);
2106 #endif
2107 	if (!PageDirty(page)) {
2108 		__set_page_dirty_nobuffers(page);
2109 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
2110 		set_page_private_reference(page);
2111 		return 1;
2112 	}
2113 	return 0;
2114 }
2115 
2116 /*
2117  * Structure of the f2fs node operations
2118  */
2119 const struct address_space_operations f2fs_node_aops = {
2120 	.writepage	= f2fs_write_node_page,
2121 	.writepages	= f2fs_write_node_pages,
2122 	.set_page_dirty	= f2fs_set_node_page_dirty,
2123 	.invalidatepage	= f2fs_invalidate_page,
2124 	.releasepage	= f2fs_release_page,
2125 #ifdef CONFIG_MIGRATION
2126 	.migratepage	= f2fs_migrate_page,
2127 #endif
2128 };
2129 
2130 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2131 						nid_t n)
2132 {
2133 	return radix_tree_lookup(&nm_i->free_nid_root, n);
2134 }
2135 
2136 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2137 				struct free_nid *i)
2138 {
2139 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2140 	int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2141 
2142 	if (err)
2143 		return err;
2144 
2145 	nm_i->nid_cnt[FREE_NID]++;
2146 	list_add_tail(&i->list, &nm_i->free_nid_list);
2147 	return 0;
2148 }
2149 
2150 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2151 			struct free_nid *i, enum nid_state state)
2152 {
2153 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2154 
2155 	f2fs_bug_on(sbi, state != i->state);
2156 	nm_i->nid_cnt[state]--;
2157 	if (state == FREE_NID)
2158 		list_del(&i->list);
2159 	radix_tree_delete(&nm_i->free_nid_root, i->nid);
2160 }
2161 
2162 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2163 			enum nid_state org_state, enum nid_state dst_state)
2164 {
2165 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2166 
2167 	f2fs_bug_on(sbi, org_state != i->state);
2168 	i->state = dst_state;
2169 	nm_i->nid_cnt[org_state]--;
2170 	nm_i->nid_cnt[dst_state]++;
2171 
2172 	switch (dst_state) {
2173 	case PREALLOC_NID:
2174 		list_del(&i->list);
2175 		break;
2176 	case FREE_NID:
2177 		list_add_tail(&i->list, &nm_i->free_nid_list);
2178 		break;
2179 	default:
2180 		BUG_ON(1);
2181 	}
2182 }
2183 
2184 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2185 							bool set, bool build)
2186 {
2187 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2188 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2189 	unsigned int nid_ofs = nid - START_NID(nid);
2190 
2191 	if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2192 		return;
2193 
2194 	if (set) {
2195 		if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2196 			return;
2197 		__set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2198 		nm_i->free_nid_count[nat_ofs]++;
2199 	} else {
2200 		if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2201 			return;
2202 		__clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2203 		if (!build)
2204 			nm_i->free_nid_count[nat_ofs]--;
2205 	}
2206 }
2207 
2208 /* return if the nid is recognized as free */
2209 static bool add_free_nid(struct f2fs_sb_info *sbi,
2210 				nid_t nid, bool build, bool update)
2211 {
2212 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2213 	struct free_nid *i, *e;
2214 	struct nat_entry *ne;
2215 	int err = -EINVAL;
2216 	bool ret = false;
2217 
2218 	/* 0 nid should not be used */
2219 	if (unlikely(nid == 0))
2220 		return false;
2221 
2222 	if (unlikely(f2fs_check_nid_range(sbi, nid)))
2223 		return false;
2224 
2225 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2226 	i->nid = nid;
2227 	i->state = FREE_NID;
2228 
2229 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2230 
2231 	spin_lock(&nm_i->nid_list_lock);
2232 
2233 	if (build) {
2234 		/*
2235 		 *   Thread A             Thread B
2236 		 *  - f2fs_create
2237 		 *   - f2fs_new_inode
2238 		 *    - f2fs_alloc_nid
2239 		 *     - __insert_nid_to_list(PREALLOC_NID)
2240 		 *                     - f2fs_balance_fs_bg
2241 		 *                      - f2fs_build_free_nids
2242 		 *                       - __f2fs_build_free_nids
2243 		 *                        - scan_nat_page
2244 		 *                         - add_free_nid
2245 		 *                          - __lookup_nat_cache
2246 		 *  - f2fs_add_link
2247 		 *   - f2fs_init_inode_metadata
2248 		 *    - f2fs_new_inode_page
2249 		 *     - f2fs_new_node_page
2250 		 *      - set_node_addr
2251 		 *  - f2fs_alloc_nid_done
2252 		 *   - __remove_nid_from_list(PREALLOC_NID)
2253 		 *                         - __insert_nid_to_list(FREE_NID)
2254 		 */
2255 		ne = __lookup_nat_cache(nm_i, nid);
2256 		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2257 				nat_get_blkaddr(ne) != NULL_ADDR))
2258 			goto err_out;
2259 
2260 		e = __lookup_free_nid_list(nm_i, nid);
2261 		if (e) {
2262 			if (e->state == FREE_NID)
2263 				ret = true;
2264 			goto err_out;
2265 		}
2266 	}
2267 	ret = true;
2268 	err = __insert_free_nid(sbi, i);
2269 err_out:
2270 	if (update) {
2271 		update_free_nid_bitmap(sbi, nid, ret, build);
2272 		if (!build)
2273 			nm_i->available_nids++;
2274 	}
2275 	spin_unlock(&nm_i->nid_list_lock);
2276 	radix_tree_preload_end();
2277 
2278 	if (err)
2279 		kmem_cache_free(free_nid_slab, i);
2280 	return ret;
2281 }
2282 
2283 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2284 {
2285 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2286 	struct free_nid *i;
2287 	bool need_free = false;
2288 
2289 	spin_lock(&nm_i->nid_list_lock);
2290 	i = __lookup_free_nid_list(nm_i, nid);
2291 	if (i && i->state == FREE_NID) {
2292 		__remove_free_nid(sbi, i, FREE_NID);
2293 		need_free = true;
2294 	}
2295 	spin_unlock(&nm_i->nid_list_lock);
2296 
2297 	if (need_free)
2298 		kmem_cache_free(free_nid_slab, i);
2299 }
2300 
2301 static int scan_nat_page(struct f2fs_sb_info *sbi,
2302 			struct page *nat_page, nid_t start_nid)
2303 {
2304 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2305 	struct f2fs_nat_block *nat_blk = page_address(nat_page);
2306 	block_t blk_addr;
2307 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2308 	int i;
2309 
2310 	__set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2311 
2312 	i = start_nid % NAT_ENTRY_PER_BLOCK;
2313 
2314 	for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2315 		if (unlikely(start_nid >= nm_i->max_nid))
2316 			break;
2317 
2318 		blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2319 
2320 		if (blk_addr == NEW_ADDR)
2321 			return -EINVAL;
2322 
2323 		if (blk_addr == NULL_ADDR) {
2324 			add_free_nid(sbi, start_nid, true, true);
2325 		} else {
2326 			spin_lock(&NM_I(sbi)->nid_list_lock);
2327 			update_free_nid_bitmap(sbi, start_nid, false, true);
2328 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2329 		}
2330 	}
2331 
2332 	return 0;
2333 }
2334 
2335 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2336 {
2337 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2338 	struct f2fs_journal *journal = curseg->journal;
2339 	int i;
2340 
2341 	down_read(&curseg->journal_rwsem);
2342 	for (i = 0; i < nats_in_cursum(journal); i++) {
2343 		block_t addr;
2344 		nid_t nid;
2345 
2346 		addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2347 		nid = le32_to_cpu(nid_in_journal(journal, i));
2348 		if (addr == NULL_ADDR)
2349 			add_free_nid(sbi, nid, true, false);
2350 		else
2351 			remove_free_nid(sbi, nid);
2352 	}
2353 	up_read(&curseg->journal_rwsem);
2354 }
2355 
2356 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2357 {
2358 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2359 	unsigned int i, idx;
2360 	nid_t nid;
2361 
2362 	down_read(&nm_i->nat_tree_lock);
2363 
2364 	for (i = 0; i < nm_i->nat_blocks; i++) {
2365 		if (!test_bit_le(i, nm_i->nat_block_bitmap))
2366 			continue;
2367 		if (!nm_i->free_nid_count[i])
2368 			continue;
2369 		for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2370 			idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2371 						NAT_ENTRY_PER_BLOCK, idx);
2372 			if (idx >= NAT_ENTRY_PER_BLOCK)
2373 				break;
2374 
2375 			nid = i * NAT_ENTRY_PER_BLOCK + idx;
2376 			add_free_nid(sbi, nid, true, false);
2377 
2378 			if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2379 				goto out;
2380 		}
2381 	}
2382 out:
2383 	scan_curseg_cache(sbi);
2384 
2385 	up_read(&nm_i->nat_tree_lock);
2386 }
2387 
2388 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2389 						bool sync, bool mount)
2390 {
2391 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2392 	int i = 0, ret;
2393 	nid_t nid = nm_i->next_scan_nid;
2394 
2395 	if (unlikely(nid >= nm_i->max_nid))
2396 		nid = 0;
2397 
2398 	if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2399 		nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2400 
2401 	/* Enough entries */
2402 	if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2403 		return 0;
2404 
2405 	if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2406 		return 0;
2407 
2408 	if (!mount) {
2409 		/* try to find free nids in free_nid_bitmap */
2410 		scan_free_nid_bits(sbi);
2411 
2412 		if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2413 			return 0;
2414 	}
2415 
2416 	/* readahead nat pages to be scanned */
2417 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2418 							META_NAT, true);
2419 
2420 	down_read(&nm_i->nat_tree_lock);
2421 
2422 	while (1) {
2423 		if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2424 						nm_i->nat_block_bitmap)) {
2425 			struct page *page = get_current_nat_page(sbi, nid);
2426 
2427 			if (IS_ERR(page)) {
2428 				ret = PTR_ERR(page);
2429 			} else {
2430 				ret = scan_nat_page(sbi, page, nid);
2431 				f2fs_put_page(page, 1);
2432 			}
2433 
2434 			if (ret) {
2435 				up_read(&nm_i->nat_tree_lock);
2436 				f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2437 				return ret;
2438 			}
2439 		}
2440 
2441 		nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2442 		if (unlikely(nid >= nm_i->max_nid))
2443 			nid = 0;
2444 
2445 		if (++i >= FREE_NID_PAGES)
2446 			break;
2447 	}
2448 
2449 	/* go to the next free nat pages to find free nids abundantly */
2450 	nm_i->next_scan_nid = nid;
2451 
2452 	/* find free nids from current sum_pages */
2453 	scan_curseg_cache(sbi);
2454 
2455 	up_read(&nm_i->nat_tree_lock);
2456 
2457 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2458 					nm_i->ra_nid_pages, META_NAT, false);
2459 
2460 	return 0;
2461 }
2462 
2463 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2464 {
2465 	int ret;
2466 
2467 	mutex_lock(&NM_I(sbi)->build_lock);
2468 	ret = __f2fs_build_free_nids(sbi, sync, mount);
2469 	mutex_unlock(&NM_I(sbi)->build_lock);
2470 
2471 	return ret;
2472 }
2473 
2474 /*
2475  * If this function returns success, caller can obtain a new nid
2476  * from second parameter of this function.
2477  * The returned nid could be used ino as well as nid when inode is created.
2478  */
2479 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2480 {
2481 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2482 	struct free_nid *i = NULL;
2483 retry:
2484 	if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2485 		f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2486 		return false;
2487 	}
2488 
2489 	spin_lock(&nm_i->nid_list_lock);
2490 
2491 	if (unlikely(nm_i->available_nids == 0)) {
2492 		spin_unlock(&nm_i->nid_list_lock);
2493 		return false;
2494 	}
2495 
2496 	/* We should not use stale free nids created by f2fs_build_free_nids */
2497 	if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2498 		f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2499 		i = list_first_entry(&nm_i->free_nid_list,
2500 					struct free_nid, list);
2501 		*nid = i->nid;
2502 
2503 		__move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2504 		nm_i->available_nids--;
2505 
2506 		update_free_nid_bitmap(sbi, *nid, false, false);
2507 
2508 		spin_unlock(&nm_i->nid_list_lock);
2509 		return true;
2510 	}
2511 	spin_unlock(&nm_i->nid_list_lock);
2512 
2513 	/* Let's scan nat pages and its caches to get free nids */
2514 	if (!f2fs_build_free_nids(sbi, true, false))
2515 		goto retry;
2516 	return false;
2517 }
2518 
2519 /*
2520  * f2fs_alloc_nid() should be called prior to this function.
2521  */
2522 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2523 {
2524 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2525 	struct free_nid *i;
2526 
2527 	spin_lock(&nm_i->nid_list_lock);
2528 	i = __lookup_free_nid_list(nm_i, nid);
2529 	f2fs_bug_on(sbi, !i);
2530 	__remove_free_nid(sbi, i, PREALLOC_NID);
2531 	spin_unlock(&nm_i->nid_list_lock);
2532 
2533 	kmem_cache_free(free_nid_slab, i);
2534 }
2535 
2536 /*
2537  * f2fs_alloc_nid() should be called prior to this function.
2538  */
2539 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2540 {
2541 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2542 	struct free_nid *i;
2543 	bool need_free = false;
2544 
2545 	if (!nid)
2546 		return;
2547 
2548 	spin_lock(&nm_i->nid_list_lock);
2549 	i = __lookup_free_nid_list(nm_i, nid);
2550 	f2fs_bug_on(sbi, !i);
2551 
2552 	if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2553 		__remove_free_nid(sbi, i, PREALLOC_NID);
2554 		need_free = true;
2555 	} else {
2556 		__move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2557 	}
2558 
2559 	nm_i->available_nids++;
2560 
2561 	update_free_nid_bitmap(sbi, nid, true, false);
2562 
2563 	spin_unlock(&nm_i->nid_list_lock);
2564 
2565 	if (need_free)
2566 		kmem_cache_free(free_nid_slab, i);
2567 }
2568 
2569 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2570 {
2571 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2572 	int nr = nr_shrink;
2573 
2574 	if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2575 		return 0;
2576 
2577 	if (!mutex_trylock(&nm_i->build_lock))
2578 		return 0;
2579 
2580 	while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2581 		struct free_nid *i, *next;
2582 		unsigned int batch = SHRINK_NID_BATCH_SIZE;
2583 
2584 		spin_lock(&nm_i->nid_list_lock);
2585 		list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2586 			if (!nr_shrink || !batch ||
2587 				nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2588 				break;
2589 			__remove_free_nid(sbi, i, FREE_NID);
2590 			kmem_cache_free(free_nid_slab, i);
2591 			nr_shrink--;
2592 			batch--;
2593 		}
2594 		spin_unlock(&nm_i->nid_list_lock);
2595 	}
2596 
2597 	mutex_unlock(&nm_i->build_lock);
2598 
2599 	return nr - nr_shrink;
2600 }
2601 
2602 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2603 {
2604 	void *src_addr, *dst_addr;
2605 	size_t inline_size;
2606 	struct page *ipage;
2607 	struct f2fs_inode *ri;
2608 
2609 	ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2610 	if (IS_ERR(ipage))
2611 		return PTR_ERR(ipage);
2612 
2613 	ri = F2FS_INODE(page);
2614 	if (ri->i_inline & F2FS_INLINE_XATTR) {
2615 		if (!f2fs_has_inline_xattr(inode)) {
2616 			set_inode_flag(inode, FI_INLINE_XATTR);
2617 			stat_inc_inline_xattr(inode);
2618 		}
2619 	} else {
2620 		if (f2fs_has_inline_xattr(inode)) {
2621 			stat_dec_inline_xattr(inode);
2622 			clear_inode_flag(inode, FI_INLINE_XATTR);
2623 		}
2624 		goto update_inode;
2625 	}
2626 
2627 	dst_addr = inline_xattr_addr(inode, ipage);
2628 	src_addr = inline_xattr_addr(inode, page);
2629 	inline_size = inline_xattr_size(inode);
2630 
2631 	f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2632 	memcpy(dst_addr, src_addr, inline_size);
2633 update_inode:
2634 	f2fs_update_inode(inode, ipage);
2635 	f2fs_put_page(ipage, 1);
2636 	return 0;
2637 }
2638 
2639 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2640 {
2641 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2642 	nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2643 	nid_t new_xnid;
2644 	struct dnode_of_data dn;
2645 	struct node_info ni;
2646 	struct page *xpage;
2647 	int err;
2648 
2649 	if (!prev_xnid)
2650 		goto recover_xnid;
2651 
2652 	/* 1: invalidate the previous xattr nid */
2653 	err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2654 	if (err)
2655 		return err;
2656 
2657 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
2658 	dec_valid_node_count(sbi, inode, false);
2659 	set_node_addr(sbi, &ni, NULL_ADDR, false);
2660 
2661 recover_xnid:
2662 	/* 2: update xattr nid in inode */
2663 	if (!f2fs_alloc_nid(sbi, &new_xnid))
2664 		return -ENOSPC;
2665 
2666 	set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2667 	xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2668 	if (IS_ERR(xpage)) {
2669 		f2fs_alloc_nid_failed(sbi, new_xnid);
2670 		return PTR_ERR(xpage);
2671 	}
2672 
2673 	f2fs_alloc_nid_done(sbi, new_xnid);
2674 	f2fs_update_inode_page(inode);
2675 
2676 	/* 3: update and set xattr node page dirty */
2677 	memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2678 
2679 	set_page_dirty(xpage);
2680 	f2fs_put_page(xpage, 1);
2681 
2682 	return 0;
2683 }
2684 
2685 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2686 {
2687 	struct f2fs_inode *src, *dst;
2688 	nid_t ino = ino_of_node(page);
2689 	struct node_info old_ni, new_ni;
2690 	struct page *ipage;
2691 	int err;
2692 
2693 	err = f2fs_get_node_info(sbi, ino, &old_ni);
2694 	if (err)
2695 		return err;
2696 
2697 	if (unlikely(old_ni.blk_addr != NULL_ADDR))
2698 		return -EINVAL;
2699 retry:
2700 	ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2701 	if (!ipage) {
2702 		congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2703 		goto retry;
2704 	}
2705 
2706 	/* Should not use this inode from free nid list */
2707 	remove_free_nid(sbi, ino);
2708 
2709 	if (!PageUptodate(ipage))
2710 		SetPageUptodate(ipage);
2711 	fill_node_footer(ipage, ino, ino, 0, true);
2712 	set_cold_node(ipage, false);
2713 
2714 	src = F2FS_INODE(page);
2715 	dst = F2FS_INODE(ipage);
2716 
2717 	memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2718 	dst->i_size = 0;
2719 	dst->i_blocks = cpu_to_le64(1);
2720 	dst->i_links = cpu_to_le32(1);
2721 	dst->i_xattr_nid = 0;
2722 	dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2723 	if (dst->i_inline & F2FS_EXTRA_ATTR) {
2724 		dst->i_extra_isize = src->i_extra_isize;
2725 
2726 		if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2727 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2728 							i_inline_xattr_size))
2729 			dst->i_inline_xattr_size = src->i_inline_xattr_size;
2730 
2731 		if (f2fs_sb_has_project_quota(sbi) &&
2732 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2733 								i_projid))
2734 			dst->i_projid = src->i_projid;
2735 
2736 		if (f2fs_sb_has_inode_crtime(sbi) &&
2737 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2738 							i_crtime_nsec)) {
2739 			dst->i_crtime = src->i_crtime;
2740 			dst->i_crtime_nsec = src->i_crtime_nsec;
2741 		}
2742 	}
2743 
2744 	new_ni = old_ni;
2745 	new_ni.ino = ino;
2746 
2747 	if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2748 		WARN_ON(1);
2749 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2750 	inc_valid_inode_count(sbi);
2751 	set_page_dirty(ipage);
2752 	f2fs_put_page(ipage, 1);
2753 	return 0;
2754 }
2755 
2756 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2757 			unsigned int segno, struct f2fs_summary_block *sum)
2758 {
2759 	struct f2fs_node *rn;
2760 	struct f2fs_summary *sum_entry;
2761 	block_t addr;
2762 	int i, idx, last_offset, nrpages;
2763 
2764 	/* scan the node segment */
2765 	last_offset = sbi->blocks_per_seg;
2766 	addr = START_BLOCK(sbi, segno);
2767 	sum_entry = &sum->entries[0];
2768 
2769 	for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2770 		nrpages = bio_max_segs(last_offset - i);
2771 
2772 		/* readahead node pages */
2773 		f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2774 
2775 		for (idx = addr; idx < addr + nrpages; idx++) {
2776 			struct page *page = f2fs_get_tmp_page(sbi, idx);
2777 
2778 			if (IS_ERR(page))
2779 				return PTR_ERR(page);
2780 
2781 			rn = F2FS_NODE(page);
2782 			sum_entry->nid = rn->footer.nid;
2783 			sum_entry->version = 0;
2784 			sum_entry->ofs_in_node = 0;
2785 			sum_entry++;
2786 			f2fs_put_page(page, 1);
2787 		}
2788 
2789 		invalidate_mapping_pages(META_MAPPING(sbi), addr,
2790 							addr + nrpages);
2791 	}
2792 	return 0;
2793 }
2794 
2795 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2796 {
2797 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2798 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2799 	struct f2fs_journal *journal = curseg->journal;
2800 	int i;
2801 
2802 	down_write(&curseg->journal_rwsem);
2803 	for (i = 0; i < nats_in_cursum(journal); i++) {
2804 		struct nat_entry *ne;
2805 		struct f2fs_nat_entry raw_ne;
2806 		nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2807 
2808 		if (f2fs_check_nid_range(sbi, nid))
2809 			continue;
2810 
2811 		raw_ne = nat_in_journal(journal, i);
2812 
2813 		ne = __lookup_nat_cache(nm_i, nid);
2814 		if (!ne) {
2815 			ne = __alloc_nat_entry(nid, true);
2816 			__init_nat_entry(nm_i, ne, &raw_ne, true);
2817 		}
2818 
2819 		/*
2820 		 * if a free nat in journal has not been used after last
2821 		 * checkpoint, we should remove it from available nids,
2822 		 * since later we will add it again.
2823 		 */
2824 		if (!get_nat_flag(ne, IS_DIRTY) &&
2825 				le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2826 			spin_lock(&nm_i->nid_list_lock);
2827 			nm_i->available_nids--;
2828 			spin_unlock(&nm_i->nid_list_lock);
2829 		}
2830 
2831 		__set_nat_cache_dirty(nm_i, ne);
2832 	}
2833 	update_nats_in_cursum(journal, -i);
2834 	up_write(&curseg->journal_rwsem);
2835 }
2836 
2837 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2838 						struct list_head *head, int max)
2839 {
2840 	struct nat_entry_set *cur;
2841 
2842 	if (nes->entry_cnt >= max)
2843 		goto add_out;
2844 
2845 	list_for_each_entry(cur, head, set_list) {
2846 		if (cur->entry_cnt >= nes->entry_cnt) {
2847 			list_add(&nes->set_list, cur->set_list.prev);
2848 			return;
2849 		}
2850 	}
2851 add_out:
2852 	list_add_tail(&nes->set_list, head);
2853 }
2854 
2855 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2856 						struct page *page)
2857 {
2858 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2859 	unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2860 	struct f2fs_nat_block *nat_blk = page_address(page);
2861 	int valid = 0;
2862 	int i = 0;
2863 
2864 	if (!enabled_nat_bits(sbi, NULL))
2865 		return;
2866 
2867 	if (nat_index == 0) {
2868 		valid = 1;
2869 		i = 1;
2870 	}
2871 	for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2872 		if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2873 			valid++;
2874 	}
2875 	if (valid == 0) {
2876 		__set_bit_le(nat_index, nm_i->empty_nat_bits);
2877 		__clear_bit_le(nat_index, nm_i->full_nat_bits);
2878 		return;
2879 	}
2880 
2881 	__clear_bit_le(nat_index, nm_i->empty_nat_bits);
2882 	if (valid == NAT_ENTRY_PER_BLOCK)
2883 		__set_bit_le(nat_index, nm_i->full_nat_bits);
2884 	else
2885 		__clear_bit_le(nat_index, nm_i->full_nat_bits);
2886 }
2887 
2888 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2889 		struct nat_entry_set *set, struct cp_control *cpc)
2890 {
2891 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2892 	struct f2fs_journal *journal = curseg->journal;
2893 	nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2894 	bool to_journal = true;
2895 	struct f2fs_nat_block *nat_blk;
2896 	struct nat_entry *ne, *cur;
2897 	struct page *page = NULL;
2898 
2899 	/*
2900 	 * there are two steps to flush nat entries:
2901 	 * #1, flush nat entries to journal in current hot data summary block.
2902 	 * #2, flush nat entries to nat page.
2903 	 */
2904 	if (enabled_nat_bits(sbi, cpc) ||
2905 		!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2906 		to_journal = false;
2907 
2908 	if (to_journal) {
2909 		down_write(&curseg->journal_rwsem);
2910 	} else {
2911 		page = get_next_nat_page(sbi, start_nid);
2912 		if (IS_ERR(page))
2913 			return PTR_ERR(page);
2914 
2915 		nat_blk = page_address(page);
2916 		f2fs_bug_on(sbi, !nat_blk);
2917 	}
2918 
2919 	/* flush dirty nats in nat entry set */
2920 	list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2921 		struct f2fs_nat_entry *raw_ne;
2922 		nid_t nid = nat_get_nid(ne);
2923 		int offset;
2924 
2925 		f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2926 
2927 		if (to_journal) {
2928 			offset = f2fs_lookup_journal_in_cursum(journal,
2929 							NAT_JOURNAL, nid, 1);
2930 			f2fs_bug_on(sbi, offset < 0);
2931 			raw_ne = &nat_in_journal(journal, offset);
2932 			nid_in_journal(journal, offset) = cpu_to_le32(nid);
2933 		} else {
2934 			raw_ne = &nat_blk->entries[nid - start_nid];
2935 		}
2936 		raw_nat_from_node_info(raw_ne, &ne->ni);
2937 		nat_reset_flag(ne);
2938 		__clear_nat_cache_dirty(NM_I(sbi), set, ne);
2939 		if (nat_get_blkaddr(ne) == NULL_ADDR) {
2940 			add_free_nid(sbi, nid, false, true);
2941 		} else {
2942 			spin_lock(&NM_I(sbi)->nid_list_lock);
2943 			update_free_nid_bitmap(sbi, nid, false, false);
2944 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2945 		}
2946 	}
2947 
2948 	if (to_journal) {
2949 		up_write(&curseg->journal_rwsem);
2950 	} else {
2951 		__update_nat_bits(sbi, start_nid, page);
2952 		f2fs_put_page(page, 1);
2953 	}
2954 
2955 	/* Allow dirty nats by node block allocation in write_begin */
2956 	if (!set->entry_cnt) {
2957 		radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2958 		kmem_cache_free(nat_entry_set_slab, set);
2959 	}
2960 	return 0;
2961 }
2962 
2963 /*
2964  * This function is called during the checkpointing process.
2965  */
2966 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2967 {
2968 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2969 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2970 	struct f2fs_journal *journal = curseg->journal;
2971 	struct nat_entry_set *setvec[SETVEC_SIZE];
2972 	struct nat_entry_set *set, *tmp;
2973 	unsigned int found;
2974 	nid_t set_idx = 0;
2975 	LIST_HEAD(sets);
2976 	int err = 0;
2977 
2978 	/*
2979 	 * during unmount, let's flush nat_bits before checking
2980 	 * nat_cnt[DIRTY_NAT].
2981 	 */
2982 	if (enabled_nat_bits(sbi, cpc)) {
2983 		down_write(&nm_i->nat_tree_lock);
2984 		remove_nats_in_journal(sbi);
2985 		up_write(&nm_i->nat_tree_lock);
2986 	}
2987 
2988 	if (!nm_i->nat_cnt[DIRTY_NAT])
2989 		return 0;
2990 
2991 	down_write(&nm_i->nat_tree_lock);
2992 
2993 	/*
2994 	 * if there are no enough space in journal to store dirty nat
2995 	 * entries, remove all entries from journal and merge them
2996 	 * into nat entry set.
2997 	 */
2998 	if (enabled_nat_bits(sbi, cpc) ||
2999 		!__has_cursum_space(journal,
3000 			nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3001 		remove_nats_in_journal(sbi);
3002 
3003 	while ((found = __gang_lookup_nat_set(nm_i,
3004 					set_idx, SETVEC_SIZE, setvec))) {
3005 		unsigned idx;
3006 
3007 		set_idx = setvec[found - 1]->set + 1;
3008 		for (idx = 0; idx < found; idx++)
3009 			__adjust_nat_entry_set(setvec[idx], &sets,
3010 						MAX_NAT_JENTRIES(journal));
3011 	}
3012 
3013 	/* flush dirty nats in nat entry set */
3014 	list_for_each_entry_safe(set, tmp, &sets, set_list) {
3015 		err = __flush_nat_entry_set(sbi, set, cpc);
3016 		if (err)
3017 			break;
3018 	}
3019 
3020 	up_write(&nm_i->nat_tree_lock);
3021 	/* Allow dirty nats by node block allocation in write_begin */
3022 
3023 	return err;
3024 }
3025 
3026 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3027 {
3028 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3029 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3030 	unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3031 	unsigned int i;
3032 	__u64 cp_ver = cur_cp_version(ckpt);
3033 	block_t nat_bits_addr;
3034 
3035 	if (!enabled_nat_bits(sbi, NULL))
3036 		return 0;
3037 
3038 	nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3039 	nm_i->nat_bits = f2fs_kvzalloc(sbi,
3040 			nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3041 	if (!nm_i->nat_bits)
3042 		return -ENOMEM;
3043 
3044 	nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3045 						nm_i->nat_bits_blocks;
3046 	for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3047 		struct page *page;
3048 
3049 		page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3050 		if (IS_ERR(page))
3051 			return PTR_ERR(page);
3052 
3053 		memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3054 					page_address(page), F2FS_BLKSIZE);
3055 		f2fs_put_page(page, 1);
3056 	}
3057 
3058 	cp_ver |= (cur_cp_crc(ckpt) << 32);
3059 	if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3060 		disable_nat_bits(sbi, true);
3061 		return 0;
3062 	}
3063 
3064 	nm_i->full_nat_bits = nm_i->nat_bits + 8;
3065 	nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3066 
3067 	f2fs_notice(sbi, "Found nat_bits in checkpoint");
3068 	return 0;
3069 }
3070 
3071 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3072 {
3073 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3074 	unsigned int i = 0;
3075 	nid_t nid, last_nid;
3076 
3077 	if (!enabled_nat_bits(sbi, NULL))
3078 		return;
3079 
3080 	for (i = 0; i < nm_i->nat_blocks; i++) {
3081 		i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3082 		if (i >= nm_i->nat_blocks)
3083 			break;
3084 
3085 		__set_bit_le(i, nm_i->nat_block_bitmap);
3086 
3087 		nid = i * NAT_ENTRY_PER_BLOCK;
3088 		last_nid = nid + NAT_ENTRY_PER_BLOCK;
3089 
3090 		spin_lock(&NM_I(sbi)->nid_list_lock);
3091 		for (; nid < last_nid; nid++)
3092 			update_free_nid_bitmap(sbi, nid, true, true);
3093 		spin_unlock(&NM_I(sbi)->nid_list_lock);
3094 	}
3095 
3096 	for (i = 0; i < nm_i->nat_blocks; i++) {
3097 		i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3098 		if (i >= nm_i->nat_blocks)
3099 			break;
3100 
3101 		__set_bit_le(i, nm_i->nat_block_bitmap);
3102 	}
3103 }
3104 
3105 static int init_node_manager(struct f2fs_sb_info *sbi)
3106 {
3107 	struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3108 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3109 	unsigned char *version_bitmap;
3110 	unsigned int nat_segs;
3111 	int err;
3112 
3113 	nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3114 
3115 	/* segment_count_nat includes pair segment so divide to 2. */
3116 	nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3117 	nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3118 	nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3119 
3120 	/* not used nids: 0, node, meta, (and root counted as valid node) */
3121 	nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3122 						F2FS_RESERVED_NODE_NUM;
3123 	nm_i->nid_cnt[FREE_NID] = 0;
3124 	nm_i->nid_cnt[PREALLOC_NID] = 0;
3125 	nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3126 	nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3127 	nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3128 
3129 	INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3130 	INIT_LIST_HEAD(&nm_i->free_nid_list);
3131 	INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3132 	INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3133 	INIT_LIST_HEAD(&nm_i->nat_entries);
3134 	spin_lock_init(&nm_i->nat_list_lock);
3135 
3136 	mutex_init(&nm_i->build_lock);
3137 	spin_lock_init(&nm_i->nid_list_lock);
3138 	init_rwsem(&nm_i->nat_tree_lock);
3139 
3140 	nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3141 	nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3142 	version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3143 	nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3144 					GFP_KERNEL);
3145 	if (!nm_i->nat_bitmap)
3146 		return -ENOMEM;
3147 
3148 	err = __get_nat_bitmaps(sbi);
3149 	if (err)
3150 		return err;
3151 
3152 #ifdef CONFIG_F2FS_CHECK_FS
3153 	nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3154 					GFP_KERNEL);
3155 	if (!nm_i->nat_bitmap_mir)
3156 		return -ENOMEM;
3157 #endif
3158 
3159 	return 0;
3160 }
3161 
3162 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3163 {
3164 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3165 	int i;
3166 
3167 	nm_i->free_nid_bitmap =
3168 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3169 					      nm_i->nat_blocks),
3170 			      GFP_KERNEL);
3171 	if (!nm_i->free_nid_bitmap)
3172 		return -ENOMEM;
3173 
3174 	for (i = 0; i < nm_i->nat_blocks; i++) {
3175 		nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3176 			f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3177 		if (!nm_i->free_nid_bitmap[i])
3178 			return -ENOMEM;
3179 	}
3180 
3181 	nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3182 								GFP_KERNEL);
3183 	if (!nm_i->nat_block_bitmap)
3184 		return -ENOMEM;
3185 
3186 	nm_i->free_nid_count =
3187 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3188 					      nm_i->nat_blocks),
3189 			      GFP_KERNEL);
3190 	if (!nm_i->free_nid_count)
3191 		return -ENOMEM;
3192 	return 0;
3193 }
3194 
3195 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3196 {
3197 	int err;
3198 
3199 	sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3200 							GFP_KERNEL);
3201 	if (!sbi->nm_info)
3202 		return -ENOMEM;
3203 
3204 	err = init_node_manager(sbi);
3205 	if (err)
3206 		return err;
3207 
3208 	err = init_free_nid_cache(sbi);
3209 	if (err)
3210 		return err;
3211 
3212 	/* load free nid status from nat_bits table */
3213 	load_free_nid_bitmap(sbi);
3214 
3215 	return f2fs_build_free_nids(sbi, true, true);
3216 }
3217 
3218 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3219 {
3220 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3221 	struct free_nid *i, *next_i;
3222 	struct nat_entry *natvec[NATVEC_SIZE];
3223 	struct nat_entry_set *setvec[SETVEC_SIZE];
3224 	nid_t nid = 0;
3225 	unsigned int found;
3226 
3227 	if (!nm_i)
3228 		return;
3229 
3230 	/* destroy free nid list */
3231 	spin_lock(&nm_i->nid_list_lock);
3232 	list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3233 		__remove_free_nid(sbi, i, FREE_NID);
3234 		spin_unlock(&nm_i->nid_list_lock);
3235 		kmem_cache_free(free_nid_slab, i);
3236 		spin_lock(&nm_i->nid_list_lock);
3237 	}
3238 	f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3239 	f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3240 	f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3241 	spin_unlock(&nm_i->nid_list_lock);
3242 
3243 	/* destroy nat cache */
3244 	down_write(&nm_i->nat_tree_lock);
3245 	while ((found = __gang_lookup_nat_cache(nm_i,
3246 					nid, NATVEC_SIZE, natvec))) {
3247 		unsigned idx;
3248 
3249 		nid = nat_get_nid(natvec[found - 1]) + 1;
3250 		for (idx = 0; idx < found; idx++) {
3251 			spin_lock(&nm_i->nat_list_lock);
3252 			list_del(&natvec[idx]->list);
3253 			spin_unlock(&nm_i->nat_list_lock);
3254 
3255 			__del_from_nat_cache(nm_i, natvec[idx]);
3256 		}
3257 	}
3258 	f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3259 
3260 	/* destroy nat set cache */
3261 	nid = 0;
3262 	while ((found = __gang_lookup_nat_set(nm_i,
3263 					nid, SETVEC_SIZE, setvec))) {
3264 		unsigned idx;
3265 
3266 		nid = setvec[found - 1]->set + 1;
3267 		for (idx = 0; idx < found; idx++) {
3268 			/* entry_cnt is not zero, when cp_error was occurred */
3269 			f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3270 			radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3271 			kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3272 		}
3273 	}
3274 	up_write(&nm_i->nat_tree_lock);
3275 
3276 	kvfree(nm_i->nat_block_bitmap);
3277 	if (nm_i->free_nid_bitmap) {
3278 		int i;
3279 
3280 		for (i = 0; i < nm_i->nat_blocks; i++)
3281 			kvfree(nm_i->free_nid_bitmap[i]);
3282 		kvfree(nm_i->free_nid_bitmap);
3283 	}
3284 	kvfree(nm_i->free_nid_count);
3285 
3286 	kvfree(nm_i->nat_bitmap);
3287 	kvfree(nm_i->nat_bits);
3288 #ifdef CONFIG_F2FS_CHECK_FS
3289 	kvfree(nm_i->nat_bitmap_mir);
3290 #endif
3291 	sbi->nm_info = NULL;
3292 	kfree(nm_i);
3293 }
3294 
3295 int __init f2fs_create_node_manager_caches(void)
3296 {
3297 	nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3298 			sizeof(struct nat_entry));
3299 	if (!nat_entry_slab)
3300 		goto fail;
3301 
3302 	free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3303 			sizeof(struct free_nid));
3304 	if (!free_nid_slab)
3305 		goto destroy_nat_entry;
3306 
3307 	nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3308 			sizeof(struct nat_entry_set));
3309 	if (!nat_entry_set_slab)
3310 		goto destroy_free_nid;
3311 
3312 	fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3313 			sizeof(struct fsync_node_entry));
3314 	if (!fsync_node_entry_slab)
3315 		goto destroy_nat_entry_set;
3316 	return 0;
3317 
3318 destroy_nat_entry_set:
3319 	kmem_cache_destroy(nat_entry_set_slab);
3320 destroy_free_nid:
3321 	kmem_cache_destroy(free_nid_slab);
3322 destroy_nat_entry:
3323 	kmem_cache_destroy(nat_entry_slab);
3324 fail:
3325 	return -ENOMEM;
3326 }
3327 
3328 void f2fs_destroy_node_manager_caches(void)
3329 {
3330 	kmem_cache_destroy(fsync_node_entry_slab);
3331 	kmem_cache_destroy(nat_entry_set_slab);
3332 	kmem_cache_destroy(free_nid_slab);
3333 	kmem_cache_destroy(nat_entry_slab);
3334 }
3335