xref: /openbmc/linux/fs/f2fs/node.c (revision e0d07278)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * fs/f2fs/node.c
4  *
5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6  *             http://www.samsung.com/
7  */
8 #include <linux/fs.h>
9 #include <linux/f2fs_fs.h>
10 #include <linux/mpage.h>
11 #include <linux/backing-dev.h>
12 #include <linux/blkdev.h>
13 #include <linux/pagevec.h>
14 #include <linux/swap.h>
15 
16 #include "f2fs.h"
17 #include "node.h"
18 #include "segment.h"
19 #include "xattr.h"
20 #include "trace.h"
21 #include <trace/events/f2fs.h>
22 
23 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
24 
25 static struct kmem_cache *nat_entry_slab;
26 static struct kmem_cache *free_nid_slab;
27 static struct kmem_cache *nat_entry_set_slab;
28 static struct kmem_cache *fsync_node_entry_slab;
29 
30 /*
31  * Check whether the given nid is within node id range.
32  */
33 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
34 {
35 	if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
36 		set_sbi_flag(sbi, SBI_NEED_FSCK);
37 		f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
38 			  __func__, nid);
39 		return -EFSCORRUPTED;
40 	}
41 	return 0;
42 }
43 
44 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
45 {
46 	struct f2fs_nm_info *nm_i = NM_I(sbi);
47 	struct sysinfo val;
48 	unsigned long avail_ram;
49 	unsigned long mem_size = 0;
50 	bool res = false;
51 
52 	si_meminfo(&val);
53 
54 	/* only uses low memory */
55 	avail_ram = val.totalram - val.totalhigh;
56 
57 	/*
58 	 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
59 	 */
60 	if (type == FREE_NIDS) {
61 		mem_size = (nm_i->nid_cnt[FREE_NID] *
62 				sizeof(struct free_nid)) >> PAGE_SHIFT;
63 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
64 	} else if (type == NAT_ENTRIES) {
65 		mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
66 							PAGE_SHIFT;
67 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
68 		if (excess_cached_nats(sbi))
69 			res = false;
70 	} else if (type == DIRTY_DENTS) {
71 		if (sbi->sb->s_bdi->wb.dirty_exceeded)
72 			return false;
73 		mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
74 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
75 	} else if (type == INO_ENTRIES) {
76 		int i;
77 
78 		for (i = 0; i < MAX_INO_ENTRY; i++)
79 			mem_size += sbi->im[i].ino_num *
80 						sizeof(struct ino_entry);
81 		mem_size >>= PAGE_SHIFT;
82 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
83 	} else if (type == EXTENT_CACHE) {
84 		mem_size = (atomic_read(&sbi->total_ext_tree) *
85 				sizeof(struct extent_tree) +
86 				atomic_read(&sbi->total_ext_node) *
87 				sizeof(struct extent_node)) >> PAGE_SHIFT;
88 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
89 	} else if (type == INMEM_PAGES) {
90 		/* it allows 20% / total_ram for inmemory pages */
91 		mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
92 		res = mem_size < (val.totalram / 5);
93 	} else {
94 		if (!sbi->sb->s_bdi->wb.dirty_exceeded)
95 			return true;
96 	}
97 	return res;
98 }
99 
100 static void clear_node_page_dirty(struct page *page)
101 {
102 	if (PageDirty(page)) {
103 		f2fs_clear_page_cache_dirty_tag(page);
104 		clear_page_dirty_for_io(page);
105 		dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
106 	}
107 	ClearPageUptodate(page);
108 }
109 
110 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
111 {
112 	return f2fs_get_meta_page_nofail(sbi, current_nat_addr(sbi, nid));
113 }
114 
115 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
116 {
117 	struct page *src_page;
118 	struct page *dst_page;
119 	pgoff_t dst_off;
120 	void *src_addr;
121 	void *dst_addr;
122 	struct f2fs_nm_info *nm_i = NM_I(sbi);
123 
124 	dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
125 
126 	/* get current nat block page with lock */
127 	src_page = get_current_nat_page(sbi, nid);
128 	if (IS_ERR(src_page))
129 		return src_page;
130 	dst_page = f2fs_grab_meta_page(sbi, dst_off);
131 	f2fs_bug_on(sbi, PageDirty(src_page));
132 
133 	src_addr = page_address(src_page);
134 	dst_addr = page_address(dst_page);
135 	memcpy(dst_addr, src_addr, PAGE_SIZE);
136 	set_page_dirty(dst_page);
137 	f2fs_put_page(src_page, 1);
138 
139 	set_to_next_nat(nm_i, nid);
140 
141 	return dst_page;
142 }
143 
144 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
145 {
146 	struct nat_entry *new;
147 
148 	if (no_fail)
149 		new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
150 	else
151 		new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
152 	if (new) {
153 		nat_set_nid(new, nid);
154 		nat_reset_flag(new);
155 	}
156 	return new;
157 }
158 
159 static void __free_nat_entry(struct nat_entry *e)
160 {
161 	kmem_cache_free(nat_entry_slab, e);
162 }
163 
164 /* must be locked by nat_tree_lock */
165 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
166 	struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
167 {
168 	if (no_fail)
169 		f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
170 	else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
171 		return NULL;
172 
173 	if (raw_ne)
174 		node_info_from_raw_nat(&ne->ni, raw_ne);
175 
176 	spin_lock(&nm_i->nat_list_lock);
177 	list_add_tail(&ne->list, &nm_i->nat_entries);
178 	spin_unlock(&nm_i->nat_list_lock);
179 
180 	nm_i->nat_cnt++;
181 	return ne;
182 }
183 
184 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
185 {
186 	struct nat_entry *ne;
187 
188 	ne = radix_tree_lookup(&nm_i->nat_root, n);
189 
190 	/* for recent accessed nat entry, move it to tail of lru list */
191 	if (ne && !get_nat_flag(ne, IS_DIRTY)) {
192 		spin_lock(&nm_i->nat_list_lock);
193 		if (!list_empty(&ne->list))
194 			list_move_tail(&ne->list, &nm_i->nat_entries);
195 		spin_unlock(&nm_i->nat_list_lock);
196 	}
197 
198 	return ne;
199 }
200 
201 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
202 		nid_t start, unsigned int nr, struct nat_entry **ep)
203 {
204 	return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
205 }
206 
207 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
208 {
209 	radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
210 	nm_i->nat_cnt--;
211 	__free_nat_entry(e);
212 }
213 
214 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
215 							struct nat_entry *ne)
216 {
217 	nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
218 	struct nat_entry_set *head;
219 
220 	head = radix_tree_lookup(&nm_i->nat_set_root, set);
221 	if (!head) {
222 		head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
223 
224 		INIT_LIST_HEAD(&head->entry_list);
225 		INIT_LIST_HEAD(&head->set_list);
226 		head->set = set;
227 		head->entry_cnt = 0;
228 		f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
229 	}
230 	return head;
231 }
232 
233 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
234 						struct nat_entry *ne)
235 {
236 	struct nat_entry_set *head;
237 	bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
238 
239 	if (!new_ne)
240 		head = __grab_nat_entry_set(nm_i, ne);
241 
242 	/*
243 	 * update entry_cnt in below condition:
244 	 * 1. update NEW_ADDR to valid block address;
245 	 * 2. update old block address to new one;
246 	 */
247 	if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
248 				!get_nat_flag(ne, IS_DIRTY)))
249 		head->entry_cnt++;
250 
251 	set_nat_flag(ne, IS_PREALLOC, new_ne);
252 
253 	if (get_nat_flag(ne, IS_DIRTY))
254 		goto refresh_list;
255 
256 	nm_i->dirty_nat_cnt++;
257 	set_nat_flag(ne, IS_DIRTY, true);
258 refresh_list:
259 	spin_lock(&nm_i->nat_list_lock);
260 	if (new_ne)
261 		list_del_init(&ne->list);
262 	else
263 		list_move_tail(&ne->list, &head->entry_list);
264 	spin_unlock(&nm_i->nat_list_lock);
265 }
266 
267 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
268 		struct nat_entry_set *set, struct nat_entry *ne)
269 {
270 	spin_lock(&nm_i->nat_list_lock);
271 	list_move_tail(&ne->list, &nm_i->nat_entries);
272 	spin_unlock(&nm_i->nat_list_lock);
273 
274 	set_nat_flag(ne, IS_DIRTY, false);
275 	set->entry_cnt--;
276 	nm_i->dirty_nat_cnt--;
277 }
278 
279 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
280 		nid_t start, unsigned int nr, struct nat_entry_set **ep)
281 {
282 	return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
283 							start, nr);
284 }
285 
286 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
287 {
288 	return NODE_MAPPING(sbi) == page->mapping &&
289 			IS_DNODE(page) && is_cold_node(page);
290 }
291 
292 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
293 {
294 	spin_lock_init(&sbi->fsync_node_lock);
295 	INIT_LIST_HEAD(&sbi->fsync_node_list);
296 	sbi->fsync_seg_id = 0;
297 	sbi->fsync_node_num = 0;
298 }
299 
300 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
301 							struct page *page)
302 {
303 	struct fsync_node_entry *fn;
304 	unsigned long flags;
305 	unsigned int seq_id;
306 
307 	fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS);
308 
309 	get_page(page);
310 	fn->page = page;
311 	INIT_LIST_HEAD(&fn->list);
312 
313 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
314 	list_add_tail(&fn->list, &sbi->fsync_node_list);
315 	fn->seq_id = sbi->fsync_seg_id++;
316 	seq_id = fn->seq_id;
317 	sbi->fsync_node_num++;
318 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
319 
320 	return seq_id;
321 }
322 
323 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
324 {
325 	struct fsync_node_entry *fn;
326 	unsigned long flags;
327 
328 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
329 	list_for_each_entry(fn, &sbi->fsync_node_list, list) {
330 		if (fn->page == page) {
331 			list_del(&fn->list);
332 			sbi->fsync_node_num--;
333 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
334 			kmem_cache_free(fsync_node_entry_slab, fn);
335 			put_page(page);
336 			return;
337 		}
338 	}
339 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
340 	f2fs_bug_on(sbi, 1);
341 }
342 
343 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
344 {
345 	unsigned long flags;
346 
347 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
348 	sbi->fsync_seg_id = 0;
349 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
350 }
351 
352 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
353 {
354 	struct f2fs_nm_info *nm_i = NM_I(sbi);
355 	struct nat_entry *e;
356 	bool need = false;
357 
358 	down_read(&nm_i->nat_tree_lock);
359 	e = __lookup_nat_cache(nm_i, nid);
360 	if (e) {
361 		if (!get_nat_flag(e, IS_CHECKPOINTED) &&
362 				!get_nat_flag(e, HAS_FSYNCED_INODE))
363 			need = true;
364 	}
365 	up_read(&nm_i->nat_tree_lock);
366 	return need;
367 }
368 
369 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
370 {
371 	struct f2fs_nm_info *nm_i = NM_I(sbi);
372 	struct nat_entry *e;
373 	bool is_cp = true;
374 
375 	down_read(&nm_i->nat_tree_lock);
376 	e = __lookup_nat_cache(nm_i, nid);
377 	if (e && !get_nat_flag(e, IS_CHECKPOINTED))
378 		is_cp = false;
379 	up_read(&nm_i->nat_tree_lock);
380 	return is_cp;
381 }
382 
383 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
384 {
385 	struct f2fs_nm_info *nm_i = NM_I(sbi);
386 	struct nat_entry *e;
387 	bool need_update = true;
388 
389 	down_read(&nm_i->nat_tree_lock);
390 	e = __lookup_nat_cache(nm_i, ino);
391 	if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
392 			(get_nat_flag(e, IS_CHECKPOINTED) ||
393 			 get_nat_flag(e, HAS_FSYNCED_INODE)))
394 		need_update = false;
395 	up_read(&nm_i->nat_tree_lock);
396 	return need_update;
397 }
398 
399 /* must be locked by nat_tree_lock */
400 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
401 						struct f2fs_nat_entry *ne)
402 {
403 	struct f2fs_nm_info *nm_i = NM_I(sbi);
404 	struct nat_entry *new, *e;
405 
406 	new = __alloc_nat_entry(nid, false);
407 	if (!new)
408 		return;
409 
410 	down_write(&nm_i->nat_tree_lock);
411 	e = __lookup_nat_cache(nm_i, nid);
412 	if (!e)
413 		e = __init_nat_entry(nm_i, new, ne, false);
414 	else
415 		f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
416 				nat_get_blkaddr(e) !=
417 					le32_to_cpu(ne->block_addr) ||
418 				nat_get_version(e) != ne->version);
419 	up_write(&nm_i->nat_tree_lock);
420 	if (e != new)
421 		__free_nat_entry(new);
422 }
423 
424 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
425 			block_t new_blkaddr, bool fsync_done)
426 {
427 	struct f2fs_nm_info *nm_i = NM_I(sbi);
428 	struct nat_entry *e;
429 	struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
430 
431 	down_write(&nm_i->nat_tree_lock);
432 	e = __lookup_nat_cache(nm_i, ni->nid);
433 	if (!e) {
434 		e = __init_nat_entry(nm_i, new, NULL, true);
435 		copy_node_info(&e->ni, ni);
436 		f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
437 	} else if (new_blkaddr == NEW_ADDR) {
438 		/*
439 		 * when nid is reallocated,
440 		 * previous nat entry can be remained in nat cache.
441 		 * So, reinitialize it with new information.
442 		 */
443 		copy_node_info(&e->ni, ni);
444 		f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
445 	}
446 	/* let's free early to reduce memory consumption */
447 	if (e != new)
448 		__free_nat_entry(new);
449 
450 	/* sanity check */
451 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
452 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
453 			new_blkaddr == NULL_ADDR);
454 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
455 			new_blkaddr == NEW_ADDR);
456 	f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
457 			new_blkaddr == NEW_ADDR);
458 
459 	/* increment version no as node is removed */
460 	if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
461 		unsigned char version = nat_get_version(e);
462 		nat_set_version(e, inc_node_version(version));
463 	}
464 
465 	/* change address */
466 	nat_set_blkaddr(e, new_blkaddr);
467 	if (!__is_valid_data_blkaddr(new_blkaddr))
468 		set_nat_flag(e, IS_CHECKPOINTED, false);
469 	__set_nat_cache_dirty(nm_i, e);
470 
471 	/* update fsync_mark if its inode nat entry is still alive */
472 	if (ni->nid != ni->ino)
473 		e = __lookup_nat_cache(nm_i, ni->ino);
474 	if (e) {
475 		if (fsync_done && ni->nid == ni->ino)
476 			set_nat_flag(e, HAS_FSYNCED_INODE, true);
477 		set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
478 	}
479 	up_write(&nm_i->nat_tree_lock);
480 }
481 
482 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
483 {
484 	struct f2fs_nm_info *nm_i = NM_I(sbi);
485 	int nr = nr_shrink;
486 
487 	if (!down_write_trylock(&nm_i->nat_tree_lock))
488 		return 0;
489 
490 	spin_lock(&nm_i->nat_list_lock);
491 	while (nr_shrink) {
492 		struct nat_entry *ne;
493 
494 		if (list_empty(&nm_i->nat_entries))
495 			break;
496 
497 		ne = list_first_entry(&nm_i->nat_entries,
498 					struct nat_entry, list);
499 		list_del(&ne->list);
500 		spin_unlock(&nm_i->nat_list_lock);
501 
502 		__del_from_nat_cache(nm_i, ne);
503 		nr_shrink--;
504 
505 		spin_lock(&nm_i->nat_list_lock);
506 	}
507 	spin_unlock(&nm_i->nat_list_lock);
508 
509 	up_write(&nm_i->nat_tree_lock);
510 	return nr - nr_shrink;
511 }
512 
513 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
514 						struct node_info *ni)
515 {
516 	struct f2fs_nm_info *nm_i = NM_I(sbi);
517 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
518 	struct f2fs_journal *journal = curseg->journal;
519 	nid_t start_nid = START_NID(nid);
520 	struct f2fs_nat_block *nat_blk;
521 	struct page *page = NULL;
522 	struct f2fs_nat_entry ne;
523 	struct nat_entry *e;
524 	pgoff_t index;
525 	block_t blkaddr;
526 	int i;
527 
528 	ni->nid = nid;
529 
530 	/* Check nat cache */
531 	down_read(&nm_i->nat_tree_lock);
532 	e = __lookup_nat_cache(nm_i, nid);
533 	if (e) {
534 		ni->ino = nat_get_ino(e);
535 		ni->blk_addr = nat_get_blkaddr(e);
536 		ni->version = nat_get_version(e);
537 		up_read(&nm_i->nat_tree_lock);
538 		return 0;
539 	}
540 
541 	memset(&ne, 0, sizeof(struct f2fs_nat_entry));
542 
543 	/* Check current segment summary */
544 	down_read(&curseg->journal_rwsem);
545 	i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
546 	if (i >= 0) {
547 		ne = nat_in_journal(journal, i);
548 		node_info_from_raw_nat(ni, &ne);
549 	}
550 	up_read(&curseg->journal_rwsem);
551 	if (i >= 0) {
552 		up_read(&nm_i->nat_tree_lock);
553 		goto cache;
554 	}
555 
556 	/* Fill node_info from nat page */
557 	index = current_nat_addr(sbi, nid);
558 	up_read(&nm_i->nat_tree_lock);
559 
560 	page = f2fs_get_meta_page(sbi, index);
561 	if (IS_ERR(page))
562 		return PTR_ERR(page);
563 
564 	nat_blk = (struct f2fs_nat_block *)page_address(page);
565 	ne = nat_blk->entries[nid - start_nid];
566 	node_info_from_raw_nat(ni, &ne);
567 	f2fs_put_page(page, 1);
568 cache:
569 	blkaddr = le32_to_cpu(ne.block_addr);
570 	if (__is_valid_data_blkaddr(blkaddr) &&
571 		!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
572 		return -EFAULT;
573 
574 	/* cache nat entry */
575 	cache_nat_entry(sbi, nid, &ne);
576 	return 0;
577 }
578 
579 /*
580  * readahead MAX_RA_NODE number of node pages.
581  */
582 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
583 {
584 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
585 	struct blk_plug plug;
586 	int i, end;
587 	nid_t nid;
588 
589 	blk_start_plug(&plug);
590 
591 	/* Then, try readahead for siblings of the desired node */
592 	end = start + n;
593 	end = min(end, NIDS_PER_BLOCK);
594 	for (i = start; i < end; i++) {
595 		nid = get_nid(parent, i, false);
596 		f2fs_ra_node_page(sbi, nid);
597 	}
598 
599 	blk_finish_plug(&plug);
600 }
601 
602 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
603 {
604 	const long direct_index = ADDRS_PER_INODE(dn->inode);
605 	const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
606 	const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
607 	unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
608 	int cur_level = dn->cur_level;
609 	int max_level = dn->max_level;
610 	pgoff_t base = 0;
611 
612 	if (!dn->max_level)
613 		return pgofs + 1;
614 
615 	while (max_level-- > cur_level)
616 		skipped_unit *= NIDS_PER_BLOCK;
617 
618 	switch (dn->max_level) {
619 	case 3:
620 		base += 2 * indirect_blks;
621 		fallthrough;
622 	case 2:
623 		base += 2 * direct_blks;
624 		fallthrough;
625 	case 1:
626 		base += direct_index;
627 		break;
628 	default:
629 		f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
630 	}
631 
632 	return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
633 }
634 
635 /*
636  * The maximum depth is four.
637  * Offset[0] will have raw inode offset.
638  */
639 static int get_node_path(struct inode *inode, long block,
640 				int offset[4], unsigned int noffset[4])
641 {
642 	const long direct_index = ADDRS_PER_INODE(inode);
643 	const long direct_blks = ADDRS_PER_BLOCK(inode);
644 	const long dptrs_per_blk = NIDS_PER_BLOCK;
645 	const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
646 	const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
647 	int n = 0;
648 	int level = 0;
649 
650 	noffset[0] = 0;
651 
652 	if (block < direct_index) {
653 		offset[n] = block;
654 		goto got;
655 	}
656 	block -= direct_index;
657 	if (block < direct_blks) {
658 		offset[n++] = NODE_DIR1_BLOCK;
659 		noffset[n] = 1;
660 		offset[n] = block;
661 		level = 1;
662 		goto got;
663 	}
664 	block -= direct_blks;
665 	if (block < direct_blks) {
666 		offset[n++] = NODE_DIR2_BLOCK;
667 		noffset[n] = 2;
668 		offset[n] = block;
669 		level = 1;
670 		goto got;
671 	}
672 	block -= direct_blks;
673 	if (block < indirect_blks) {
674 		offset[n++] = NODE_IND1_BLOCK;
675 		noffset[n] = 3;
676 		offset[n++] = block / direct_blks;
677 		noffset[n] = 4 + offset[n - 1];
678 		offset[n] = block % direct_blks;
679 		level = 2;
680 		goto got;
681 	}
682 	block -= indirect_blks;
683 	if (block < indirect_blks) {
684 		offset[n++] = NODE_IND2_BLOCK;
685 		noffset[n] = 4 + dptrs_per_blk;
686 		offset[n++] = block / direct_blks;
687 		noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
688 		offset[n] = block % direct_blks;
689 		level = 2;
690 		goto got;
691 	}
692 	block -= indirect_blks;
693 	if (block < dindirect_blks) {
694 		offset[n++] = NODE_DIND_BLOCK;
695 		noffset[n] = 5 + (dptrs_per_blk * 2);
696 		offset[n++] = block / indirect_blks;
697 		noffset[n] = 6 + (dptrs_per_blk * 2) +
698 			      offset[n - 1] * (dptrs_per_blk + 1);
699 		offset[n++] = (block / direct_blks) % dptrs_per_blk;
700 		noffset[n] = 7 + (dptrs_per_blk * 2) +
701 			      offset[n - 2] * (dptrs_per_blk + 1) +
702 			      offset[n - 1];
703 		offset[n] = block % direct_blks;
704 		level = 3;
705 		goto got;
706 	} else {
707 		return -E2BIG;
708 	}
709 got:
710 	return level;
711 }
712 
713 /*
714  * Caller should call f2fs_put_dnode(dn).
715  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
716  * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
717  */
718 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
719 {
720 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
721 	struct page *npage[4];
722 	struct page *parent = NULL;
723 	int offset[4];
724 	unsigned int noffset[4];
725 	nid_t nids[4];
726 	int level, i = 0;
727 	int err = 0;
728 
729 	level = get_node_path(dn->inode, index, offset, noffset);
730 	if (level < 0)
731 		return level;
732 
733 	nids[0] = dn->inode->i_ino;
734 	npage[0] = dn->inode_page;
735 
736 	if (!npage[0]) {
737 		npage[0] = f2fs_get_node_page(sbi, nids[0]);
738 		if (IS_ERR(npage[0]))
739 			return PTR_ERR(npage[0]);
740 	}
741 
742 	/* if inline_data is set, should not report any block indices */
743 	if (f2fs_has_inline_data(dn->inode) && index) {
744 		err = -ENOENT;
745 		f2fs_put_page(npage[0], 1);
746 		goto release_out;
747 	}
748 
749 	parent = npage[0];
750 	if (level != 0)
751 		nids[1] = get_nid(parent, offset[0], true);
752 	dn->inode_page = npage[0];
753 	dn->inode_page_locked = true;
754 
755 	/* get indirect or direct nodes */
756 	for (i = 1; i <= level; i++) {
757 		bool done = false;
758 
759 		if (!nids[i] && mode == ALLOC_NODE) {
760 			/* alloc new node */
761 			if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
762 				err = -ENOSPC;
763 				goto release_pages;
764 			}
765 
766 			dn->nid = nids[i];
767 			npage[i] = f2fs_new_node_page(dn, noffset[i]);
768 			if (IS_ERR(npage[i])) {
769 				f2fs_alloc_nid_failed(sbi, nids[i]);
770 				err = PTR_ERR(npage[i]);
771 				goto release_pages;
772 			}
773 
774 			set_nid(parent, offset[i - 1], nids[i], i == 1);
775 			f2fs_alloc_nid_done(sbi, nids[i]);
776 			done = true;
777 		} else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
778 			npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
779 			if (IS_ERR(npage[i])) {
780 				err = PTR_ERR(npage[i]);
781 				goto release_pages;
782 			}
783 			done = true;
784 		}
785 		if (i == 1) {
786 			dn->inode_page_locked = false;
787 			unlock_page(parent);
788 		} else {
789 			f2fs_put_page(parent, 1);
790 		}
791 
792 		if (!done) {
793 			npage[i] = f2fs_get_node_page(sbi, nids[i]);
794 			if (IS_ERR(npage[i])) {
795 				err = PTR_ERR(npage[i]);
796 				f2fs_put_page(npage[0], 0);
797 				goto release_out;
798 			}
799 		}
800 		if (i < level) {
801 			parent = npage[i];
802 			nids[i + 1] = get_nid(parent, offset[i], false);
803 		}
804 	}
805 	dn->nid = nids[level];
806 	dn->ofs_in_node = offset[level];
807 	dn->node_page = npage[level];
808 	dn->data_blkaddr = f2fs_data_blkaddr(dn);
809 	return 0;
810 
811 release_pages:
812 	f2fs_put_page(parent, 1);
813 	if (i > 1)
814 		f2fs_put_page(npage[0], 0);
815 release_out:
816 	dn->inode_page = NULL;
817 	dn->node_page = NULL;
818 	if (err == -ENOENT) {
819 		dn->cur_level = i;
820 		dn->max_level = level;
821 		dn->ofs_in_node = offset[level];
822 	}
823 	return err;
824 }
825 
826 static int truncate_node(struct dnode_of_data *dn)
827 {
828 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
829 	struct node_info ni;
830 	int err;
831 	pgoff_t index;
832 
833 	err = f2fs_get_node_info(sbi, dn->nid, &ni);
834 	if (err)
835 		return err;
836 
837 	/* Deallocate node address */
838 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
839 	dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
840 	set_node_addr(sbi, &ni, NULL_ADDR, false);
841 
842 	if (dn->nid == dn->inode->i_ino) {
843 		f2fs_remove_orphan_inode(sbi, dn->nid);
844 		dec_valid_inode_count(sbi);
845 		f2fs_inode_synced(dn->inode);
846 	}
847 
848 	clear_node_page_dirty(dn->node_page);
849 	set_sbi_flag(sbi, SBI_IS_DIRTY);
850 
851 	index = dn->node_page->index;
852 	f2fs_put_page(dn->node_page, 1);
853 
854 	invalidate_mapping_pages(NODE_MAPPING(sbi),
855 			index, index);
856 
857 	dn->node_page = NULL;
858 	trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
859 
860 	return 0;
861 }
862 
863 static int truncate_dnode(struct dnode_of_data *dn)
864 {
865 	struct page *page;
866 	int err;
867 
868 	if (dn->nid == 0)
869 		return 1;
870 
871 	/* get direct node */
872 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
873 	if (PTR_ERR(page) == -ENOENT)
874 		return 1;
875 	else if (IS_ERR(page))
876 		return PTR_ERR(page);
877 
878 	/* Make dnode_of_data for parameter */
879 	dn->node_page = page;
880 	dn->ofs_in_node = 0;
881 	f2fs_truncate_data_blocks(dn);
882 	err = truncate_node(dn);
883 	if (err)
884 		return err;
885 
886 	return 1;
887 }
888 
889 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
890 						int ofs, int depth)
891 {
892 	struct dnode_of_data rdn = *dn;
893 	struct page *page;
894 	struct f2fs_node *rn;
895 	nid_t child_nid;
896 	unsigned int child_nofs;
897 	int freed = 0;
898 	int i, ret;
899 
900 	if (dn->nid == 0)
901 		return NIDS_PER_BLOCK + 1;
902 
903 	trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
904 
905 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
906 	if (IS_ERR(page)) {
907 		trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
908 		return PTR_ERR(page);
909 	}
910 
911 	f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
912 
913 	rn = F2FS_NODE(page);
914 	if (depth < 3) {
915 		for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
916 			child_nid = le32_to_cpu(rn->in.nid[i]);
917 			if (child_nid == 0)
918 				continue;
919 			rdn.nid = child_nid;
920 			ret = truncate_dnode(&rdn);
921 			if (ret < 0)
922 				goto out_err;
923 			if (set_nid(page, i, 0, false))
924 				dn->node_changed = true;
925 		}
926 	} else {
927 		child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
928 		for (i = ofs; i < NIDS_PER_BLOCK; i++) {
929 			child_nid = le32_to_cpu(rn->in.nid[i]);
930 			if (child_nid == 0) {
931 				child_nofs += NIDS_PER_BLOCK + 1;
932 				continue;
933 			}
934 			rdn.nid = child_nid;
935 			ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
936 			if (ret == (NIDS_PER_BLOCK + 1)) {
937 				if (set_nid(page, i, 0, false))
938 					dn->node_changed = true;
939 				child_nofs += ret;
940 			} else if (ret < 0 && ret != -ENOENT) {
941 				goto out_err;
942 			}
943 		}
944 		freed = child_nofs;
945 	}
946 
947 	if (!ofs) {
948 		/* remove current indirect node */
949 		dn->node_page = page;
950 		ret = truncate_node(dn);
951 		if (ret)
952 			goto out_err;
953 		freed++;
954 	} else {
955 		f2fs_put_page(page, 1);
956 	}
957 	trace_f2fs_truncate_nodes_exit(dn->inode, freed);
958 	return freed;
959 
960 out_err:
961 	f2fs_put_page(page, 1);
962 	trace_f2fs_truncate_nodes_exit(dn->inode, ret);
963 	return ret;
964 }
965 
966 static int truncate_partial_nodes(struct dnode_of_data *dn,
967 			struct f2fs_inode *ri, int *offset, int depth)
968 {
969 	struct page *pages[2];
970 	nid_t nid[3];
971 	nid_t child_nid;
972 	int err = 0;
973 	int i;
974 	int idx = depth - 2;
975 
976 	nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
977 	if (!nid[0])
978 		return 0;
979 
980 	/* get indirect nodes in the path */
981 	for (i = 0; i < idx + 1; i++) {
982 		/* reference count'll be increased */
983 		pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
984 		if (IS_ERR(pages[i])) {
985 			err = PTR_ERR(pages[i]);
986 			idx = i - 1;
987 			goto fail;
988 		}
989 		nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
990 	}
991 
992 	f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
993 
994 	/* free direct nodes linked to a partial indirect node */
995 	for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
996 		child_nid = get_nid(pages[idx], i, false);
997 		if (!child_nid)
998 			continue;
999 		dn->nid = child_nid;
1000 		err = truncate_dnode(dn);
1001 		if (err < 0)
1002 			goto fail;
1003 		if (set_nid(pages[idx], i, 0, false))
1004 			dn->node_changed = true;
1005 	}
1006 
1007 	if (offset[idx + 1] == 0) {
1008 		dn->node_page = pages[idx];
1009 		dn->nid = nid[idx];
1010 		err = truncate_node(dn);
1011 		if (err)
1012 			goto fail;
1013 	} else {
1014 		f2fs_put_page(pages[idx], 1);
1015 	}
1016 	offset[idx]++;
1017 	offset[idx + 1] = 0;
1018 	idx--;
1019 fail:
1020 	for (i = idx; i >= 0; i--)
1021 		f2fs_put_page(pages[i], 1);
1022 
1023 	trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1024 
1025 	return err;
1026 }
1027 
1028 /*
1029  * All the block addresses of data and nodes should be nullified.
1030  */
1031 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1032 {
1033 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1034 	int err = 0, cont = 1;
1035 	int level, offset[4], noffset[4];
1036 	unsigned int nofs = 0;
1037 	struct f2fs_inode *ri;
1038 	struct dnode_of_data dn;
1039 	struct page *page;
1040 
1041 	trace_f2fs_truncate_inode_blocks_enter(inode, from);
1042 
1043 	level = get_node_path(inode, from, offset, noffset);
1044 	if (level < 0) {
1045 		trace_f2fs_truncate_inode_blocks_exit(inode, level);
1046 		return level;
1047 	}
1048 
1049 	page = f2fs_get_node_page(sbi, inode->i_ino);
1050 	if (IS_ERR(page)) {
1051 		trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1052 		return PTR_ERR(page);
1053 	}
1054 
1055 	set_new_dnode(&dn, inode, page, NULL, 0);
1056 	unlock_page(page);
1057 
1058 	ri = F2FS_INODE(page);
1059 	switch (level) {
1060 	case 0:
1061 	case 1:
1062 		nofs = noffset[1];
1063 		break;
1064 	case 2:
1065 		nofs = noffset[1];
1066 		if (!offset[level - 1])
1067 			goto skip_partial;
1068 		err = truncate_partial_nodes(&dn, ri, offset, level);
1069 		if (err < 0 && err != -ENOENT)
1070 			goto fail;
1071 		nofs += 1 + NIDS_PER_BLOCK;
1072 		break;
1073 	case 3:
1074 		nofs = 5 + 2 * NIDS_PER_BLOCK;
1075 		if (!offset[level - 1])
1076 			goto skip_partial;
1077 		err = truncate_partial_nodes(&dn, ri, offset, level);
1078 		if (err < 0 && err != -ENOENT)
1079 			goto fail;
1080 		break;
1081 	default:
1082 		BUG();
1083 	}
1084 
1085 skip_partial:
1086 	while (cont) {
1087 		dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1088 		switch (offset[0]) {
1089 		case NODE_DIR1_BLOCK:
1090 		case NODE_DIR2_BLOCK:
1091 			err = truncate_dnode(&dn);
1092 			break;
1093 
1094 		case NODE_IND1_BLOCK:
1095 		case NODE_IND2_BLOCK:
1096 			err = truncate_nodes(&dn, nofs, offset[1], 2);
1097 			break;
1098 
1099 		case NODE_DIND_BLOCK:
1100 			err = truncate_nodes(&dn, nofs, offset[1], 3);
1101 			cont = 0;
1102 			break;
1103 
1104 		default:
1105 			BUG();
1106 		}
1107 		if (err < 0 && err != -ENOENT)
1108 			goto fail;
1109 		if (offset[1] == 0 &&
1110 				ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1111 			lock_page(page);
1112 			BUG_ON(page->mapping != NODE_MAPPING(sbi));
1113 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1114 			ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1115 			set_page_dirty(page);
1116 			unlock_page(page);
1117 		}
1118 		offset[1] = 0;
1119 		offset[0]++;
1120 		nofs += err;
1121 	}
1122 fail:
1123 	f2fs_put_page(page, 0);
1124 	trace_f2fs_truncate_inode_blocks_exit(inode, err);
1125 	return err > 0 ? 0 : err;
1126 }
1127 
1128 /* caller must lock inode page */
1129 int f2fs_truncate_xattr_node(struct inode *inode)
1130 {
1131 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1132 	nid_t nid = F2FS_I(inode)->i_xattr_nid;
1133 	struct dnode_of_data dn;
1134 	struct page *npage;
1135 	int err;
1136 
1137 	if (!nid)
1138 		return 0;
1139 
1140 	npage = f2fs_get_node_page(sbi, nid);
1141 	if (IS_ERR(npage))
1142 		return PTR_ERR(npage);
1143 
1144 	set_new_dnode(&dn, inode, NULL, npage, nid);
1145 	err = truncate_node(&dn);
1146 	if (err) {
1147 		f2fs_put_page(npage, 1);
1148 		return err;
1149 	}
1150 
1151 	f2fs_i_xnid_write(inode, 0);
1152 
1153 	return 0;
1154 }
1155 
1156 /*
1157  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1158  * f2fs_unlock_op().
1159  */
1160 int f2fs_remove_inode_page(struct inode *inode)
1161 {
1162 	struct dnode_of_data dn;
1163 	int err;
1164 
1165 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1166 	err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1167 	if (err)
1168 		return err;
1169 
1170 	err = f2fs_truncate_xattr_node(inode);
1171 	if (err) {
1172 		f2fs_put_dnode(&dn);
1173 		return err;
1174 	}
1175 
1176 	/* remove potential inline_data blocks */
1177 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1178 				S_ISLNK(inode->i_mode))
1179 		f2fs_truncate_data_blocks_range(&dn, 1);
1180 
1181 	/* 0 is possible, after f2fs_new_inode() has failed */
1182 	if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1183 		f2fs_put_dnode(&dn);
1184 		return -EIO;
1185 	}
1186 
1187 	if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1188 		f2fs_warn(F2FS_I_SB(inode),
1189 			"f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1190 			inode->i_ino, (unsigned long long)inode->i_blocks);
1191 		set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1192 	}
1193 
1194 	/* will put inode & node pages */
1195 	err = truncate_node(&dn);
1196 	if (err) {
1197 		f2fs_put_dnode(&dn);
1198 		return err;
1199 	}
1200 	return 0;
1201 }
1202 
1203 struct page *f2fs_new_inode_page(struct inode *inode)
1204 {
1205 	struct dnode_of_data dn;
1206 
1207 	/* allocate inode page for new inode */
1208 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1209 
1210 	/* caller should f2fs_put_page(page, 1); */
1211 	return f2fs_new_node_page(&dn, 0);
1212 }
1213 
1214 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1215 {
1216 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1217 	struct node_info new_ni;
1218 	struct page *page;
1219 	int err;
1220 
1221 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1222 		return ERR_PTR(-EPERM);
1223 
1224 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1225 	if (!page)
1226 		return ERR_PTR(-ENOMEM);
1227 
1228 	if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1229 		goto fail;
1230 
1231 #ifdef CONFIG_F2FS_CHECK_FS
1232 	err = f2fs_get_node_info(sbi, dn->nid, &new_ni);
1233 	if (err) {
1234 		dec_valid_node_count(sbi, dn->inode, !ofs);
1235 		goto fail;
1236 	}
1237 	f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1238 #endif
1239 	new_ni.nid = dn->nid;
1240 	new_ni.ino = dn->inode->i_ino;
1241 	new_ni.blk_addr = NULL_ADDR;
1242 	new_ni.flag = 0;
1243 	new_ni.version = 0;
1244 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1245 
1246 	f2fs_wait_on_page_writeback(page, NODE, true, true);
1247 	fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1248 	set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1249 	if (!PageUptodate(page))
1250 		SetPageUptodate(page);
1251 	if (set_page_dirty(page))
1252 		dn->node_changed = true;
1253 
1254 	if (f2fs_has_xattr_block(ofs))
1255 		f2fs_i_xnid_write(dn->inode, dn->nid);
1256 
1257 	if (ofs == 0)
1258 		inc_valid_inode_count(sbi);
1259 	return page;
1260 
1261 fail:
1262 	clear_node_page_dirty(page);
1263 	f2fs_put_page(page, 1);
1264 	return ERR_PTR(err);
1265 }
1266 
1267 /*
1268  * Caller should do after getting the following values.
1269  * 0: f2fs_put_page(page, 0)
1270  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1271  */
1272 static int read_node_page(struct page *page, int op_flags)
1273 {
1274 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1275 	struct node_info ni;
1276 	struct f2fs_io_info fio = {
1277 		.sbi = sbi,
1278 		.type = NODE,
1279 		.op = REQ_OP_READ,
1280 		.op_flags = op_flags,
1281 		.page = page,
1282 		.encrypted_page = NULL,
1283 	};
1284 	int err;
1285 
1286 	if (PageUptodate(page)) {
1287 		if (!f2fs_inode_chksum_verify(sbi, page)) {
1288 			ClearPageUptodate(page);
1289 			return -EFSBADCRC;
1290 		}
1291 		return LOCKED_PAGE;
1292 	}
1293 
1294 	err = f2fs_get_node_info(sbi, page->index, &ni);
1295 	if (err)
1296 		return err;
1297 
1298 	if (unlikely(ni.blk_addr == NULL_ADDR) ||
1299 			is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1300 		ClearPageUptodate(page);
1301 		return -ENOENT;
1302 	}
1303 
1304 	fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1305 
1306 	err = f2fs_submit_page_bio(&fio);
1307 
1308 	if (!err)
1309 		f2fs_update_iostat(sbi, FS_NODE_READ_IO, F2FS_BLKSIZE);
1310 
1311 	return err;
1312 }
1313 
1314 /*
1315  * Readahead a node page
1316  */
1317 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1318 {
1319 	struct page *apage;
1320 	int err;
1321 
1322 	if (!nid)
1323 		return;
1324 	if (f2fs_check_nid_range(sbi, nid))
1325 		return;
1326 
1327 	apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1328 	if (apage)
1329 		return;
1330 
1331 	apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1332 	if (!apage)
1333 		return;
1334 
1335 	err = read_node_page(apage, REQ_RAHEAD);
1336 	f2fs_put_page(apage, err ? 1 : 0);
1337 }
1338 
1339 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1340 					struct page *parent, int start)
1341 {
1342 	struct page *page;
1343 	int err;
1344 
1345 	if (!nid)
1346 		return ERR_PTR(-ENOENT);
1347 	if (f2fs_check_nid_range(sbi, nid))
1348 		return ERR_PTR(-EINVAL);
1349 repeat:
1350 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1351 	if (!page)
1352 		return ERR_PTR(-ENOMEM);
1353 
1354 	err = read_node_page(page, 0);
1355 	if (err < 0) {
1356 		f2fs_put_page(page, 1);
1357 		return ERR_PTR(err);
1358 	} else if (err == LOCKED_PAGE) {
1359 		err = 0;
1360 		goto page_hit;
1361 	}
1362 
1363 	if (parent)
1364 		f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1365 
1366 	lock_page(page);
1367 
1368 	if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1369 		f2fs_put_page(page, 1);
1370 		goto repeat;
1371 	}
1372 
1373 	if (unlikely(!PageUptodate(page))) {
1374 		err = -EIO;
1375 		goto out_err;
1376 	}
1377 
1378 	if (!f2fs_inode_chksum_verify(sbi, page)) {
1379 		err = -EFSBADCRC;
1380 		goto out_err;
1381 	}
1382 page_hit:
1383 	if(unlikely(nid != nid_of_node(page))) {
1384 		f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1385 			  nid, nid_of_node(page), ino_of_node(page),
1386 			  ofs_of_node(page), cpver_of_node(page),
1387 			  next_blkaddr_of_node(page));
1388 		err = -EINVAL;
1389 out_err:
1390 		ClearPageUptodate(page);
1391 		f2fs_put_page(page, 1);
1392 		return ERR_PTR(err);
1393 	}
1394 	return page;
1395 }
1396 
1397 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1398 {
1399 	return __get_node_page(sbi, nid, NULL, 0);
1400 }
1401 
1402 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1403 {
1404 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1405 	nid_t nid = get_nid(parent, start, false);
1406 
1407 	return __get_node_page(sbi, nid, parent, start);
1408 }
1409 
1410 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1411 {
1412 	struct inode *inode;
1413 	struct page *page;
1414 	int ret;
1415 
1416 	/* should flush inline_data before evict_inode */
1417 	inode = ilookup(sbi->sb, ino);
1418 	if (!inode)
1419 		return;
1420 
1421 	page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1422 					FGP_LOCK|FGP_NOWAIT, 0);
1423 	if (!page)
1424 		goto iput_out;
1425 
1426 	if (!PageUptodate(page))
1427 		goto page_out;
1428 
1429 	if (!PageDirty(page))
1430 		goto page_out;
1431 
1432 	if (!clear_page_dirty_for_io(page))
1433 		goto page_out;
1434 
1435 	ret = f2fs_write_inline_data(inode, page);
1436 	inode_dec_dirty_pages(inode);
1437 	f2fs_remove_dirty_inode(inode);
1438 	if (ret)
1439 		set_page_dirty(page);
1440 page_out:
1441 	f2fs_put_page(page, 1);
1442 iput_out:
1443 	iput(inode);
1444 }
1445 
1446 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1447 {
1448 	pgoff_t index;
1449 	struct pagevec pvec;
1450 	struct page *last_page = NULL;
1451 	int nr_pages;
1452 
1453 	pagevec_init(&pvec);
1454 	index = 0;
1455 
1456 	while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1457 				PAGECACHE_TAG_DIRTY))) {
1458 		int i;
1459 
1460 		for (i = 0; i < nr_pages; i++) {
1461 			struct page *page = pvec.pages[i];
1462 
1463 			if (unlikely(f2fs_cp_error(sbi))) {
1464 				f2fs_put_page(last_page, 0);
1465 				pagevec_release(&pvec);
1466 				return ERR_PTR(-EIO);
1467 			}
1468 
1469 			if (!IS_DNODE(page) || !is_cold_node(page))
1470 				continue;
1471 			if (ino_of_node(page) != ino)
1472 				continue;
1473 
1474 			lock_page(page);
1475 
1476 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1477 continue_unlock:
1478 				unlock_page(page);
1479 				continue;
1480 			}
1481 			if (ino_of_node(page) != ino)
1482 				goto continue_unlock;
1483 
1484 			if (!PageDirty(page)) {
1485 				/* someone wrote it for us */
1486 				goto continue_unlock;
1487 			}
1488 
1489 			if (last_page)
1490 				f2fs_put_page(last_page, 0);
1491 
1492 			get_page(page);
1493 			last_page = page;
1494 			unlock_page(page);
1495 		}
1496 		pagevec_release(&pvec);
1497 		cond_resched();
1498 	}
1499 	return last_page;
1500 }
1501 
1502 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1503 				struct writeback_control *wbc, bool do_balance,
1504 				enum iostat_type io_type, unsigned int *seq_id)
1505 {
1506 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1507 	nid_t nid;
1508 	struct node_info ni;
1509 	struct f2fs_io_info fio = {
1510 		.sbi = sbi,
1511 		.ino = ino_of_node(page),
1512 		.type = NODE,
1513 		.op = REQ_OP_WRITE,
1514 		.op_flags = wbc_to_write_flags(wbc),
1515 		.page = page,
1516 		.encrypted_page = NULL,
1517 		.submitted = false,
1518 		.io_type = io_type,
1519 		.io_wbc = wbc,
1520 	};
1521 	unsigned int seq;
1522 
1523 	trace_f2fs_writepage(page, NODE);
1524 
1525 	if (unlikely(f2fs_cp_error(sbi))) {
1526 		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
1527 			ClearPageUptodate(page);
1528 			dec_page_count(sbi, F2FS_DIRTY_NODES);
1529 			unlock_page(page);
1530 			return 0;
1531 		}
1532 		goto redirty_out;
1533 	}
1534 
1535 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1536 		goto redirty_out;
1537 
1538 	if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1539 			wbc->sync_mode == WB_SYNC_NONE &&
1540 			IS_DNODE(page) && is_cold_node(page))
1541 		goto redirty_out;
1542 
1543 	/* get old block addr of this node page */
1544 	nid = nid_of_node(page);
1545 	f2fs_bug_on(sbi, page->index != nid);
1546 
1547 	if (f2fs_get_node_info(sbi, nid, &ni))
1548 		goto redirty_out;
1549 
1550 	if (wbc->for_reclaim) {
1551 		if (!down_read_trylock(&sbi->node_write))
1552 			goto redirty_out;
1553 	} else {
1554 		down_read(&sbi->node_write);
1555 	}
1556 
1557 	/* This page is already truncated */
1558 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1559 		ClearPageUptodate(page);
1560 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1561 		up_read(&sbi->node_write);
1562 		unlock_page(page);
1563 		return 0;
1564 	}
1565 
1566 	if (__is_valid_data_blkaddr(ni.blk_addr) &&
1567 		!f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1568 					DATA_GENERIC_ENHANCE)) {
1569 		up_read(&sbi->node_write);
1570 		goto redirty_out;
1571 	}
1572 
1573 	if (atomic && !test_opt(sbi, NOBARRIER))
1574 		fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1575 
1576 	/* should add to global list before clearing PAGECACHE status */
1577 	if (f2fs_in_warm_node_list(sbi, page)) {
1578 		seq = f2fs_add_fsync_node_entry(sbi, page);
1579 		if (seq_id)
1580 			*seq_id = seq;
1581 	}
1582 
1583 	set_page_writeback(page);
1584 	ClearPageError(page);
1585 
1586 	fio.old_blkaddr = ni.blk_addr;
1587 	f2fs_do_write_node_page(nid, &fio);
1588 	set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1589 	dec_page_count(sbi, F2FS_DIRTY_NODES);
1590 	up_read(&sbi->node_write);
1591 
1592 	if (wbc->for_reclaim) {
1593 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1594 		submitted = NULL;
1595 	}
1596 
1597 	unlock_page(page);
1598 
1599 	if (unlikely(f2fs_cp_error(sbi))) {
1600 		f2fs_submit_merged_write(sbi, NODE);
1601 		submitted = NULL;
1602 	}
1603 	if (submitted)
1604 		*submitted = fio.submitted;
1605 
1606 	if (do_balance)
1607 		f2fs_balance_fs(sbi, false);
1608 	return 0;
1609 
1610 redirty_out:
1611 	redirty_page_for_writepage(wbc, page);
1612 	return AOP_WRITEPAGE_ACTIVATE;
1613 }
1614 
1615 int f2fs_move_node_page(struct page *node_page, int gc_type)
1616 {
1617 	int err = 0;
1618 
1619 	if (gc_type == FG_GC) {
1620 		struct writeback_control wbc = {
1621 			.sync_mode = WB_SYNC_ALL,
1622 			.nr_to_write = 1,
1623 			.for_reclaim = 0,
1624 		};
1625 
1626 		f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1627 
1628 		set_page_dirty(node_page);
1629 
1630 		if (!clear_page_dirty_for_io(node_page)) {
1631 			err = -EAGAIN;
1632 			goto out_page;
1633 		}
1634 
1635 		if (__write_node_page(node_page, false, NULL,
1636 					&wbc, false, FS_GC_NODE_IO, NULL)) {
1637 			err = -EAGAIN;
1638 			unlock_page(node_page);
1639 		}
1640 		goto release_page;
1641 	} else {
1642 		/* set page dirty and write it */
1643 		if (!PageWriteback(node_page))
1644 			set_page_dirty(node_page);
1645 	}
1646 out_page:
1647 	unlock_page(node_page);
1648 release_page:
1649 	f2fs_put_page(node_page, 0);
1650 	return err;
1651 }
1652 
1653 static int f2fs_write_node_page(struct page *page,
1654 				struct writeback_control *wbc)
1655 {
1656 	return __write_node_page(page, false, NULL, wbc, false,
1657 						FS_NODE_IO, NULL);
1658 }
1659 
1660 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1661 			struct writeback_control *wbc, bool atomic,
1662 			unsigned int *seq_id)
1663 {
1664 	pgoff_t index;
1665 	struct pagevec pvec;
1666 	int ret = 0;
1667 	struct page *last_page = NULL;
1668 	bool marked = false;
1669 	nid_t ino = inode->i_ino;
1670 	int nr_pages;
1671 	int nwritten = 0;
1672 
1673 	if (atomic) {
1674 		last_page = last_fsync_dnode(sbi, ino);
1675 		if (IS_ERR_OR_NULL(last_page))
1676 			return PTR_ERR_OR_ZERO(last_page);
1677 	}
1678 retry:
1679 	pagevec_init(&pvec);
1680 	index = 0;
1681 
1682 	while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1683 				PAGECACHE_TAG_DIRTY))) {
1684 		int i;
1685 
1686 		for (i = 0; i < nr_pages; i++) {
1687 			struct page *page = pvec.pages[i];
1688 			bool submitted = false;
1689 
1690 			if (unlikely(f2fs_cp_error(sbi))) {
1691 				f2fs_put_page(last_page, 0);
1692 				pagevec_release(&pvec);
1693 				ret = -EIO;
1694 				goto out;
1695 			}
1696 
1697 			if (!IS_DNODE(page) || !is_cold_node(page))
1698 				continue;
1699 			if (ino_of_node(page) != ino)
1700 				continue;
1701 
1702 			lock_page(page);
1703 
1704 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1705 continue_unlock:
1706 				unlock_page(page);
1707 				continue;
1708 			}
1709 			if (ino_of_node(page) != ino)
1710 				goto continue_unlock;
1711 
1712 			if (!PageDirty(page) && page != last_page) {
1713 				/* someone wrote it for us */
1714 				goto continue_unlock;
1715 			}
1716 
1717 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1718 
1719 			set_fsync_mark(page, 0);
1720 			set_dentry_mark(page, 0);
1721 
1722 			if (!atomic || page == last_page) {
1723 				set_fsync_mark(page, 1);
1724 				if (IS_INODE(page)) {
1725 					if (is_inode_flag_set(inode,
1726 								FI_DIRTY_INODE))
1727 						f2fs_update_inode(inode, page);
1728 					set_dentry_mark(page,
1729 						f2fs_need_dentry_mark(sbi, ino));
1730 				}
1731 				/* may be written by other thread */
1732 				if (!PageDirty(page))
1733 					set_page_dirty(page);
1734 			}
1735 
1736 			if (!clear_page_dirty_for_io(page))
1737 				goto continue_unlock;
1738 
1739 			ret = __write_node_page(page, atomic &&
1740 						page == last_page,
1741 						&submitted, wbc, true,
1742 						FS_NODE_IO, seq_id);
1743 			if (ret) {
1744 				unlock_page(page);
1745 				f2fs_put_page(last_page, 0);
1746 				break;
1747 			} else if (submitted) {
1748 				nwritten++;
1749 			}
1750 
1751 			if (page == last_page) {
1752 				f2fs_put_page(page, 0);
1753 				marked = true;
1754 				break;
1755 			}
1756 		}
1757 		pagevec_release(&pvec);
1758 		cond_resched();
1759 
1760 		if (ret || marked)
1761 			break;
1762 	}
1763 	if (!ret && atomic && !marked) {
1764 		f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1765 			   ino, last_page->index);
1766 		lock_page(last_page);
1767 		f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1768 		set_page_dirty(last_page);
1769 		unlock_page(last_page);
1770 		goto retry;
1771 	}
1772 out:
1773 	if (nwritten)
1774 		f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1775 	return ret ? -EIO: 0;
1776 }
1777 
1778 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1779 {
1780 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1781 	bool clean;
1782 
1783 	if (inode->i_ino != ino)
1784 		return 0;
1785 
1786 	if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1787 		return 0;
1788 
1789 	spin_lock(&sbi->inode_lock[DIRTY_META]);
1790 	clean = list_empty(&F2FS_I(inode)->gdirty_list);
1791 	spin_unlock(&sbi->inode_lock[DIRTY_META]);
1792 
1793 	if (clean)
1794 		return 0;
1795 
1796 	inode = igrab(inode);
1797 	if (!inode)
1798 		return 0;
1799 	return 1;
1800 }
1801 
1802 static bool flush_dirty_inode(struct page *page)
1803 {
1804 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1805 	struct inode *inode;
1806 	nid_t ino = ino_of_node(page);
1807 
1808 	inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1809 	if (!inode)
1810 		return false;
1811 
1812 	f2fs_update_inode(inode, page);
1813 	unlock_page(page);
1814 
1815 	iput(inode);
1816 	return true;
1817 }
1818 
1819 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1820 {
1821 	pgoff_t index = 0;
1822 	struct pagevec pvec;
1823 	int nr_pages;
1824 
1825 	pagevec_init(&pvec);
1826 
1827 	while ((nr_pages = pagevec_lookup_tag(&pvec,
1828 			NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1829 		int i;
1830 
1831 		for (i = 0; i < nr_pages; i++) {
1832 			struct page *page = pvec.pages[i];
1833 
1834 			if (!IS_DNODE(page))
1835 				continue;
1836 
1837 			lock_page(page);
1838 
1839 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1840 continue_unlock:
1841 				unlock_page(page);
1842 				continue;
1843 			}
1844 
1845 			if (!PageDirty(page)) {
1846 				/* someone wrote it for us */
1847 				goto continue_unlock;
1848 			}
1849 
1850 			/* flush inline_data, if it's async context. */
1851 			if (is_inline_node(page)) {
1852 				clear_inline_node(page);
1853 				unlock_page(page);
1854 				flush_inline_data(sbi, ino_of_node(page));
1855 				continue;
1856 			}
1857 			unlock_page(page);
1858 		}
1859 		pagevec_release(&pvec);
1860 		cond_resched();
1861 	}
1862 }
1863 
1864 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1865 				struct writeback_control *wbc,
1866 				bool do_balance, enum iostat_type io_type)
1867 {
1868 	pgoff_t index;
1869 	struct pagevec pvec;
1870 	int step = 0;
1871 	int nwritten = 0;
1872 	int ret = 0;
1873 	int nr_pages, done = 0;
1874 
1875 	pagevec_init(&pvec);
1876 
1877 next_step:
1878 	index = 0;
1879 
1880 	while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1881 			NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1882 		int i;
1883 
1884 		for (i = 0; i < nr_pages; i++) {
1885 			struct page *page = pvec.pages[i];
1886 			bool submitted = false;
1887 			bool may_dirty = true;
1888 
1889 			/* give a priority to WB_SYNC threads */
1890 			if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1891 					wbc->sync_mode == WB_SYNC_NONE) {
1892 				done = 1;
1893 				break;
1894 			}
1895 
1896 			/*
1897 			 * flushing sequence with step:
1898 			 * 0. indirect nodes
1899 			 * 1. dentry dnodes
1900 			 * 2. file dnodes
1901 			 */
1902 			if (step == 0 && IS_DNODE(page))
1903 				continue;
1904 			if (step == 1 && (!IS_DNODE(page) ||
1905 						is_cold_node(page)))
1906 				continue;
1907 			if (step == 2 && (!IS_DNODE(page) ||
1908 						!is_cold_node(page)))
1909 				continue;
1910 lock_node:
1911 			if (wbc->sync_mode == WB_SYNC_ALL)
1912 				lock_page(page);
1913 			else if (!trylock_page(page))
1914 				continue;
1915 
1916 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1917 continue_unlock:
1918 				unlock_page(page);
1919 				continue;
1920 			}
1921 
1922 			if (!PageDirty(page)) {
1923 				/* someone wrote it for us */
1924 				goto continue_unlock;
1925 			}
1926 
1927 			/* flush inline_data/inode, if it's async context. */
1928 			if (!do_balance)
1929 				goto write_node;
1930 
1931 			/* flush inline_data */
1932 			if (is_inline_node(page)) {
1933 				clear_inline_node(page);
1934 				unlock_page(page);
1935 				flush_inline_data(sbi, ino_of_node(page));
1936 				goto lock_node;
1937 			}
1938 
1939 			/* flush dirty inode */
1940 			if (IS_INODE(page) && may_dirty) {
1941 				may_dirty = false;
1942 				if (flush_dirty_inode(page))
1943 					goto lock_node;
1944 			}
1945 write_node:
1946 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1947 
1948 			if (!clear_page_dirty_for_io(page))
1949 				goto continue_unlock;
1950 
1951 			set_fsync_mark(page, 0);
1952 			set_dentry_mark(page, 0);
1953 
1954 			ret = __write_node_page(page, false, &submitted,
1955 						wbc, do_balance, io_type, NULL);
1956 			if (ret)
1957 				unlock_page(page);
1958 			else if (submitted)
1959 				nwritten++;
1960 
1961 			if (--wbc->nr_to_write == 0)
1962 				break;
1963 		}
1964 		pagevec_release(&pvec);
1965 		cond_resched();
1966 
1967 		if (wbc->nr_to_write == 0) {
1968 			step = 2;
1969 			break;
1970 		}
1971 	}
1972 
1973 	if (step < 2) {
1974 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1975 				wbc->sync_mode == WB_SYNC_NONE && step == 1)
1976 			goto out;
1977 		step++;
1978 		goto next_step;
1979 	}
1980 out:
1981 	if (nwritten)
1982 		f2fs_submit_merged_write(sbi, NODE);
1983 
1984 	if (unlikely(f2fs_cp_error(sbi)))
1985 		return -EIO;
1986 	return ret;
1987 }
1988 
1989 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
1990 						unsigned int seq_id)
1991 {
1992 	struct fsync_node_entry *fn;
1993 	struct page *page;
1994 	struct list_head *head = &sbi->fsync_node_list;
1995 	unsigned long flags;
1996 	unsigned int cur_seq_id = 0;
1997 	int ret2, ret = 0;
1998 
1999 	while (seq_id && cur_seq_id < seq_id) {
2000 		spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2001 		if (list_empty(head)) {
2002 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2003 			break;
2004 		}
2005 		fn = list_first_entry(head, struct fsync_node_entry, list);
2006 		if (fn->seq_id > seq_id) {
2007 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2008 			break;
2009 		}
2010 		cur_seq_id = fn->seq_id;
2011 		page = fn->page;
2012 		get_page(page);
2013 		spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2014 
2015 		f2fs_wait_on_page_writeback(page, NODE, true, false);
2016 		if (TestClearPageError(page))
2017 			ret = -EIO;
2018 
2019 		put_page(page);
2020 
2021 		if (ret)
2022 			break;
2023 	}
2024 
2025 	ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2026 	if (!ret)
2027 		ret = ret2;
2028 
2029 	return ret;
2030 }
2031 
2032 static int f2fs_write_node_pages(struct address_space *mapping,
2033 			    struct writeback_control *wbc)
2034 {
2035 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2036 	struct blk_plug plug;
2037 	long diff;
2038 
2039 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2040 		goto skip_write;
2041 
2042 	/* balancing f2fs's metadata in background */
2043 	f2fs_balance_fs_bg(sbi, true);
2044 
2045 	/* collect a number of dirty node pages and write together */
2046 	if (wbc->sync_mode != WB_SYNC_ALL &&
2047 			get_pages(sbi, F2FS_DIRTY_NODES) <
2048 					nr_pages_to_skip(sbi, NODE))
2049 		goto skip_write;
2050 
2051 	if (wbc->sync_mode == WB_SYNC_ALL)
2052 		atomic_inc(&sbi->wb_sync_req[NODE]);
2053 	else if (atomic_read(&sbi->wb_sync_req[NODE]))
2054 		goto skip_write;
2055 
2056 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2057 
2058 	diff = nr_pages_to_write(sbi, NODE, wbc);
2059 	blk_start_plug(&plug);
2060 	f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2061 	blk_finish_plug(&plug);
2062 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2063 
2064 	if (wbc->sync_mode == WB_SYNC_ALL)
2065 		atomic_dec(&sbi->wb_sync_req[NODE]);
2066 	return 0;
2067 
2068 skip_write:
2069 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2070 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2071 	return 0;
2072 }
2073 
2074 static int f2fs_set_node_page_dirty(struct page *page)
2075 {
2076 	trace_f2fs_set_page_dirty(page, NODE);
2077 
2078 	if (!PageUptodate(page))
2079 		SetPageUptodate(page);
2080 #ifdef CONFIG_F2FS_CHECK_FS
2081 	if (IS_INODE(page))
2082 		f2fs_inode_chksum_set(F2FS_P_SB(page), page);
2083 #endif
2084 	if (!PageDirty(page)) {
2085 		__set_page_dirty_nobuffers(page);
2086 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
2087 		f2fs_set_page_private(page, 0);
2088 		f2fs_trace_pid(page);
2089 		return 1;
2090 	}
2091 	return 0;
2092 }
2093 
2094 /*
2095  * Structure of the f2fs node operations
2096  */
2097 const struct address_space_operations f2fs_node_aops = {
2098 	.writepage	= f2fs_write_node_page,
2099 	.writepages	= f2fs_write_node_pages,
2100 	.set_page_dirty	= f2fs_set_node_page_dirty,
2101 	.invalidatepage	= f2fs_invalidate_page,
2102 	.releasepage	= f2fs_release_page,
2103 #ifdef CONFIG_MIGRATION
2104 	.migratepage	= f2fs_migrate_page,
2105 #endif
2106 };
2107 
2108 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2109 						nid_t n)
2110 {
2111 	return radix_tree_lookup(&nm_i->free_nid_root, n);
2112 }
2113 
2114 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2115 				struct free_nid *i)
2116 {
2117 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2118 
2119 	int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2120 	if (err)
2121 		return err;
2122 
2123 	nm_i->nid_cnt[FREE_NID]++;
2124 	list_add_tail(&i->list, &nm_i->free_nid_list);
2125 	return 0;
2126 }
2127 
2128 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2129 			struct free_nid *i, enum nid_state state)
2130 {
2131 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2132 
2133 	f2fs_bug_on(sbi, state != i->state);
2134 	nm_i->nid_cnt[state]--;
2135 	if (state == FREE_NID)
2136 		list_del(&i->list);
2137 	radix_tree_delete(&nm_i->free_nid_root, i->nid);
2138 }
2139 
2140 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2141 			enum nid_state org_state, enum nid_state dst_state)
2142 {
2143 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2144 
2145 	f2fs_bug_on(sbi, org_state != i->state);
2146 	i->state = dst_state;
2147 	nm_i->nid_cnt[org_state]--;
2148 	nm_i->nid_cnt[dst_state]++;
2149 
2150 	switch (dst_state) {
2151 	case PREALLOC_NID:
2152 		list_del(&i->list);
2153 		break;
2154 	case FREE_NID:
2155 		list_add_tail(&i->list, &nm_i->free_nid_list);
2156 		break;
2157 	default:
2158 		BUG_ON(1);
2159 	}
2160 }
2161 
2162 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2163 							bool set, bool build)
2164 {
2165 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2166 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2167 	unsigned int nid_ofs = nid - START_NID(nid);
2168 
2169 	if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2170 		return;
2171 
2172 	if (set) {
2173 		if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2174 			return;
2175 		__set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2176 		nm_i->free_nid_count[nat_ofs]++;
2177 	} else {
2178 		if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2179 			return;
2180 		__clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2181 		if (!build)
2182 			nm_i->free_nid_count[nat_ofs]--;
2183 	}
2184 }
2185 
2186 /* return if the nid is recognized as free */
2187 static bool add_free_nid(struct f2fs_sb_info *sbi,
2188 				nid_t nid, bool build, bool update)
2189 {
2190 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2191 	struct free_nid *i, *e;
2192 	struct nat_entry *ne;
2193 	int err = -EINVAL;
2194 	bool ret = false;
2195 
2196 	/* 0 nid should not be used */
2197 	if (unlikely(nid == 0))
2198 		return false;
2199 
2200 	if (unlikely(f2fs_check_nid_range(sbi, nid)))
2201 		return false;
2202 
2203 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2204 	i->nid = nid;
2205 	i->state = FREE_NID;
2206 
2207 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2208 
2209 	spin_lock(&nm_i->nid_list_lock);
2210 
2211 	if (build) {
2212 		/*
2213 		 *   Thread A             Thread B
2214 		 *  - f2fs_create
2215 		 *   - f2fs_new_inode
2216 		 *    - f2fs_alloc_nid
2217 		 *     - __insert_nid_to_list(PREALLOC_NID)
2218 		 *                     - f2fs_balance_fs_bg
2219 		 *                      - f2fs_build_free_nids
2220 		 *                       - __f2fs_build_free_nids
2221 		 *                        - scan_nat_page
2222 		 *                         - add_free_nid
2223 		 *                          - __lookup_nat_cache
2224 		 *  - f2fs_add_link
2225 		 *   - f2fs_init_inode_metadata
2226 		 *    - f2fs_new_inode_page
2227 		 *     - f2fs_new_node_page
2228 		 *      - set_node_addr
2229 		 *  - f2fs_alloc_nid_done
2230 		 *   - __remove_nid_from_list(PREALLOC_NID)
2231 		 *                         - __insert_nid_to_list(FREE_NID)
2232 		 */
2233 		ne = __lookup_nat_cache(nm_i, nid);
2234 		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2235 				nat_get_blkaddr(ne) != NULL_ADDR))
2236 			goto err_out;
2237 
2238 		e = __lookup_free_nid_list(nm_i, nid);
2239 		if (e) {
2240 			if (e->state == FREE_NID)
2241 				ret = true;
2242 			goto err_out;
2243 		}
2244 	}
2245 	ret = true;
2246 	err = __insert_free_nid(sbi, i);
2247 err_out:
2248 	if (update) {
2249 		update_free_nid_bitmap(sbi, nid, ret, build);
2250 		if (!build)
2251 			nm_i->available_nids++;
2252 	}
2253 	spin_unlock(&nm_i->nid_list_lock);
2254 	radix_tree_preload_end();
2255 
2256 	if (err)
2257 		kmem_cache_free(free_nid_slab, i);
2258 	return ret;
2259 }
2260 
2261 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2262 {
2263 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2264 	struct free_nid *i;
2265 	bool need_free = false;
2266 
2267 	spin_lock(&nm_i->nid_list_lock);
2268 	i = __lookup_free_nid_list(nm_i, nid);
2269 	if (i && i->state == FREE_NID) {
2270 		__remove_free_nid(sbi, i, FREE_NID);
2271 		need_free = true;
2272 	}
2273 	spin_unlock(&nm_i->nid_list_lock);
2274 
2275 	if (need_free)
2276 		kmem_cache_free(free_nid_slab, i);
2277 }
2278 
2279 static int scan_nat_page(struct f2fs_sb_info *sbi,
2280 			struct page *nat_page, nid_t start_nid)
2281 {
2282 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2283 	struct f2fs_nat_block *nat_blk = page_address(nat_page);
2284 	block_t blk_addr;
2285 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2286 	int i;
2287 
2288 	__set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2289 
2290 	i = start_nid % NAT_ENTRY_PER_BLOCK;
2291 
2292 	for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2293 		if (unlikely(start_nid >= nm_i->max_nid))
2294 			break;
2295 
2296 		blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2297 
2298 		if (blk_addr == NEW_ADDR)
2299 			return -EINVAL;
2300 
2301 		if (blk_addr == NULL_ADDR) {
2302 			add_free_nid(sbi, start_nid, true, true);
2303 		} else {
2304 			spin_lock(&NM_I(sbi)->nid_list_lock);
2305 			update_free_nid_bitmap(sbi, start_nid, false, true);
2306 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2307 		}
2308 	}
2309 
2310 	return 0;
2311 }
2312 
2313 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2314 {
2315 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2316 	struct f2fs_journal *journal = curseg->journal;
2317 	int i;
2318 
2319 	down_read(&curseg->journal_rwsem);
2320 	for (i = 0; i < nats_in_cursum(journal); i++) {
2321 		block_t addr;
2322 		nid_t nid;
2323 
2324 		addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2325 		nid = le32_to_cpu(nid_in_journal(journal, i));
2326 		if (addr == NULL_ADDR)
2327 			add_free_nid(sbi, nid, true, false);
2328 		else
2329 			remove_free_nid(sbi, nid);
2330 	}
2331 	up_read(&curseg->journal_rwsem);
2332 }
2333 
2334 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2335 {
2336 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2337 	unsigned int i, idx;
2338 	nid_t nid;
2339 
2340 	down_read(&nm_i->nat_tree_lock);
2341 
2342 	for (i = 0; i < nm_i->nat_blocks; i++) {
2343 		if (!test_bit_le(i, nm_i->nat_block_bitmap))
2344 			continue;
2345 		if (!nm_i->free_nid_count[i])
2346 			continue;
2347 		for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2348 			idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2349 						NAT_ENTRY_PER_BLOCK, idx);
2350 			if (idx >= NAT_ENTRY_PER_BLOCK)
2351 				break;
2352 
2353 			nid = i * NAT_ENTRY_PER_BLOCK + idx;
2354 			add_free_nid(sbi, nid, true, false);
2355 
2356 			if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2357 				goto out;
2358 		}
2359 	}
2360 out:
2361 	scan_curseg_cache(sbi);
2362 
2363 	up_read(&nm_i->nat_tree_lock);
2364 }
2365 
2366 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2367 						bool sync, bool mount)
2368 {
2369 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2370 	int i = 0, ret;
2371 	nid_t nid = nm_i->next_scan_nid;
2372 
2373 	if (unlikely(nid >= nm_i->max_nid))
2374 		nid = 0;
2375 
2376 	/* Enough entries */
2377 	if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2378 		return 0;
2379 
2380 	if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2381 		return 0;
2382 
2383 	if (!mount) {
2384 		/* try to find free nids in free_nid_bitmap */
2385 		scan_free_nid_bits(sbi);
2386 
2387 		if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2388 			return 0;
2389 	}
2390 
2391 	/* readahead nat pages to be scanned */
2392 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2393 							META_NAT, true);
2394 
2395 	down_read(&nm_i->nat_tree_lock);
2396 
2397 	while (1) {
2398 		if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2399 						nm_i->nat_block_bitmap)) {
2400 			struct page *page = get_current_nat_page(sbi, nid);
2401 
2402 			if (IS_ERR(page)) {
2403 				ret = PTR_ERR(page);
2404 			} else {
2405 				ret = scan_nat_page(sbi, page, nid);
2406 				f2fs_put_page(page, 1);
2407 			}
2408 
2409 			if (ret) {
2410 				up_read(&nm_i->nat_tree_lock);
2411 				f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2412 				return ret;
2413 			}
2414 		}
2415 
2416 		nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2417 		if (unlikely(nid >= nm_i->max_nid))
2418 			nid = 0;
2419 
2420 		if (++i >= FREE_NID_PAGES)
2421 			break;
2422 	}
2423 
2424 	/* go to the next free nat pages to find free nids abundantly */
2425 	nm_i->next_scan_nid = nid;
2426 
2427 	/* find free nids from current sum_pages */
2428 	scan_curseg_cache(sbi);
2429 
2430 	up_read(&nm_i->nat_tree_lock);
2431 
2432 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2433 					nm_i->ra_nid_pages, META_NAT, false);
2434 
2435 	return 0;
2436 }
2437 
2438 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2439 {
2440 	int ret;
2441 
2442 	mutex_lock(&NM_I(sbi)->build_lock);
2443 	ret = __f2fs_build_free_nids(sbi, sync, mount);
2444 	mutex_unlock(&NM_I(sbi)->build_lock);
2445 
2446 	return ret;
2447 }
2448 
2449 /*
2450  * If this function returns success, caller can obtain a new nid
2451  * from second parameter of this function.
2452  * The returned nid could be used ino as well as nid when inode is created.
2453  */
2454 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2455 {
2456 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2457 	struct free_nid *i = NULL;
2458 retry:
2459 	if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2460 		f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2461 		return false;
2462 	}
2463 
2464 	spin_lock(&nm_i->nid_list_lock);
2465 
2466 	if (unlikely(nm_i->available_nids == 0)) {
2467 		spin_unlock(&nm_i->nid_list_lock);
2468 		return false;
2469 	}
2470 
2471 	/* We should not use stale free nids created by f2fs_build_free_nids */
2472 	if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2473 		f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2474 		i = list_first_entry(&nm_i->free_nid_list,
2475 					struct free_nid, list);
2476 		*nid = i->nid;
2477 
2478 		__move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2479 		nm_i->available_nids--;
2480 
2481 		update_free_nid_bitmap(sbi, *nid, false, false);
2482 
2483 		spin_unlock(&nm_i->nid_list_lock);
2484 		return true;
2485 	}
2486 	spin_unlock(&nm_i->nid_list_lock);
2487 
2488 	/* Let's scan nat pages and its caches to get free nids */
2489 	if (!f2fs_build_free_nids(sbi, true, false))
2490 		goto retry;
2491 	return false;
2492 }
2493 
2494 /*
2495  * f2fs_alloc_nid() should be called prior to this function.
2496  */
2497 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2498 {
2499 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2500 	struct free_nid *i;
2501 
2502 	spin_lock(&nm_i->nid_list_lock);
2503 	i = __lookup_free_nid_list(nm_i, nid);
2504 	f2fs_bug_on(sbi, !i);
2505 	__remove_free_nid(sbi, i, PREALLOC_NID);
2506 	spin_unlock(&nm_i->nid_list_lock);
2507 
2508 	kmem_cache_free(free_nid_slab, i);
2509 }
2510 
2511 /*
2512  * f2fs_alloc_nid() should be called prior to this function.
2513  */
2514 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2515 {
2516 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2517 	struct free_nid *i;
2518 	bool need_free = false;
2519 
2520 	if (!nid)
2521 		return;
2522 
2523 	spin_lock(&nm_i->nid_list_lock);
2524 	i = __lookup_free_nid_list(nm_i, nid);
2525 	f2fs_bug_on(sbi, !i);
2526 
2527 	if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2528 		__remove_free_nid(sbi, i, PREALLOC_NID);
2529 		need_free = true;
2530 	} else {
2531 		__move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2532 	}
2533 
2534 	nm_i->available_nids++;
2535 
2536 	update_free_nid_bitmap(sbi, nid, true, false);
2537 
2538 	spin_unlock(&nm_i->nid_list_lock);
2539 
2540 	if (need_free)
2541 		kmem_cache_free(free_nid_slab, i);
2542 }
2543 
2544 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2545 {
2546 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2547 	int nr = nr_shrink;
2548 
2549 	if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2550 		return 0;
2551 
2552 	if (!mutex_trylock(&nm_i->build_lock))
2553 		return 0;
2554 
2555 	while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2556 		struct free_nid *i, *next;
2557 		unsigned int batch = SHRINK_NID_BATCH_SIZE;
2558 
2559 		spin_lock(&nm_i->nid_list_lock);
2560 		list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2561 			if (!nr_shrink || !batch ||
2562 				nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2563 				break;
2564 			__remove_free_nid(sbi, i, FREE_NID);
2565 			kmem_cache_free(free_nid_slab, i);
2566 			nr_shrink--;
2567 			batch--;
2568 		}
2569 		spin_unlock(&nm_i->nid_list_lock);
2570 	}
2571 
2572 	mutex_unlock(&nm_i->build_lock);
2573 
2574 	return nr - nr_shrink;
2575 }
2576 
2577 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2578 {
2579 	void *src_addr, *dst_addr;
2580 	size_t inline_size;
2581 	struct page *ipage;
2582 	struct f2fs_inode *ri;
2583 
2584 	ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2585 	if (IS_ERR(ipage))
2586 		return PTR_ERR(ipage);
2587 
2588 	ri = F2FS_INODE(page);
2589 	if (ri->i_inline & F2FS_INLINE_XATTR) {
2590 		set_inode_flag(inode, FI_INLINE_XATTR);
2591 	} else {
2592 		clear_inode_flag(inode, FI_INLINE_XATTR);
2593 		goto update_inode;
2594 	}
2595 
2596 	dst_addr = inline_xattr_addr(inode, ipage);
2597 	src_addr = inline_xattr_addr(inode, page);
2598 	inline_size = inline_xattr_size(inode);
2599 
2600 	f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2601 	memcpy(dst_addr, src_addr, inline_size);
2602 update_inode:
2603 	f2fs_update_inode(inode, ipage);
2604 	f2fs_put_page(ipage, 1);
2605 	return 0;
2606 }
2607 
2608 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2609 {
2610 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2611 	nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2612 	nid_t new_xnid;
2613 	struct dnode_of_data dn;
2614 	struct node_info ni;
2615 	struct page *xpage;
2616 	int err;
2617 
2618 	if (!prev_xnid)
2619 		goto recover_xnid;
2620 
2621 	/* 1: invalidate the previous xattr nid */
2622 	err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2623 	if (err)
2624 		return err;
2625 
2626 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
2627 	dec_valid_node_count(sbi, inode, false);
2628 	set_node_addr(sbi, &ni, NULL_ADDR, false);
2629 
2630 recover_xnid:
2631 	/* 2: update xattr nid in inode */
2632 	if (!f2fs_alloc_nid(sbi, &new_xnid))
2633 		return -ENOSPC;
2634 
2635 	set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2636 	xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2637 	if (IS_ERR(xpage)) {
2638 		f2fs_alloc_nid_failed(sbi, new_xnid);
2639 		return PTR_ERR(xpage);
2640 	}
2641 
2642 	f2fs_alloc_nid_done(sbi, new_xnid);
2643 	f2fs_update_inode_page(inode);
2644 
2645 	/* 3: update and set xattr node page dirty */
2646 	memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2647 
2648 	set_page_dirty(xpage);
2649 	f2fs_put_page(xpage, 1);
2650 
2651 	return 0;
2652 }
2653 
2654 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2655 {
2656 	struct f2fs_inode *src, *dst;
2657 	nid_t ino = ino_of_node(page);
2658 	struct node_info old_ni, new_ni;
2659 	struct page *ipage;
2660 	int err;
2661 
2662 	err = f2fs_get_node_info(sbi, ino, &old_ni);
2663 	if (err)
2664 		return err;
2665 
2666 	if (unlikely(old_ni.blk_addr != NULL_ADDR))
2667 		return -EINVAL;
2668 retry:
2669 	ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2670 	if (!ipage) {
2671 		congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2672 		goto retry;
2673 	}
2674 
2675 	/* Should not use this inode from free nid list */
2676 	remove_free_nid(sbi, ino);
2677 
2678 	if (!PageUptodate(ipage))
2679 		SetPageUptodate(ipage);
2680 	fill_node_footer(ipage, ino, ino, 0, true);
2681 	set_cold_node(ipage, false);
2682 
2683 	src = F2FS_INODE(page);
2684 	dst = F2FS_INODE(ipage);
2685 
2686 	memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2687 	dst->i_size = 0;
2688 	dst->i_blocks = cpu_to_le64(1);
2689 	dst->i_links = cpu_to_le32(1);
2690 	dst->i_xattr_nid = 0;
2691 	dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2692 	if (dst->i_inline & F2FS_EXTRA_ATTR) {
2693 		dst->i_extra_isize = src->i_extra_isize;
2694 
2695 		if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2696 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2697 							i_inline_xattr_size))
2698 			dst->i_inline_xattr_size = src->i_inline_xattr_size;
2699 
2700 		if (f2fs_sb_has_project_quota(sbi) &&
2701 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2702 								i_projid))
2703 			dst->i_projid = src->i_projid;
2704 
2705 		if (f2fs_sb_has_inode_crtime(sbi) &&
2706 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2707 							i_crtime_nsec)) {
2708 			dst->i_crtime = src->i_crtime;
2709 			dst->i_crtime_nsec = src->i_crtime_nsec;
2710 		}
2711 	}
2712 
2713 	new_ni = old_ni;
2714 	new_ni.ino = ino;
2715 
2716 	if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2717 		WARN_ON(1);
2718 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2719 	inc_valid_inode_count(sbi);
2720 	set_page_dirty(ipage);
2721 	f2fs_put_page(ipage, 1);
2722 	return 0;
2723 }
2724 
2725 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2726 			unsigned int segno, struct f2fs_summary_block *sum)
2727 {
2728 	struct f2fs_node *rn;
2729 	struct f2fs_summary *sum_entry;
2730 	block_t addr;
2731 	int i, idx, last_offset, nrpages;
2732 
2733 	/* scan the node segment */
2734 	last_offset = sbi->blocks_per_seg;
2735 	addr = START_BLOCK(sbi, segno);
2736 	sum_entry = &sum->entries[0];
2737 
2738 	for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2739 		nrpages = min(last_offset - i, BIO_MAX_PAGES);
2740 
2741 		/* readahead node pages */
2742 		f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2743 
2744 		for (idx = addr; idx < addr + nrpages; idx++) {
2745 			struct page *page = f2fs_get_tmp_page(sbi, idx);
2746 
2747 			if (IS_ERR(page))
2748 				return PTR_ERR(page);
2749 
2750 			rn = F2FS_NODE(page);
2751 			sum_entry->nid = rn->footer.nid;
2752 			sum_entry->version = 0;
2753 			sum_entry->ofs_in_node = 0;
2754 			sum_entry++;
2755 			f2fs_put_page(page, 1);
2756 		}
2757 
2758 		invalidate_mapping_pages(META_MAPPING(sbi), addr,
2759 							addr + nrpages);
2760 	}
2761 	return 0;
2762 }
2763 
2764 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2765 {
2766 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2767 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2768 	struct f2fs_journal *journal = curseg->journal;
2769 	int i;
2770 
2771 	down_write(&curseg->journal_rwsem);
2772 	for (i = 0; i < nats_in_cursum(journal); i++) {
2773 		struct nat_entry *ne;
2774 		struct f2fs_nat_entry raw_ne;
2775 		nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2776 
2777 		raw_ne = nat_in_journal(journal, i);
2778 
2779 		ne = __lookup_nat_cache(nm_i, nid);
2780 		if (!ne) {
2781 			ne = __alloc_nat_entry(nid, true);
2782 			__init_nat_entry(nm_i, ne, &raw_ne, true);
2783 		}
2784 
2785 		/*
2786 		 * if a free nat in journal has not been used after last
2787 		 * checkpoint, we should remove it from available nids,
2788 		 * since later we will add it again.
2789 		 */
2790 		if (!get_nat_flag(ne, IS_DIRTY) &&
2791 				le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2792 			spin_lock(&nm_i->nid_list_lock);
2793 			nm_i->available_nids--;
2794 			spin_unlock(&nm_i->nid_list_lock);
2795 		}
2796 
2797 		__set_nat_cache_dirty(nm_i, ne);
2798 	}
2799 	update_nats_in_cursum(journal, -i);
2800 	up_write(&curseg->journal_rwsem);
2801 }
2802 
2803 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2804 						struct list_head *head, int max)
2805 {
2806 	struct nat_entry_set *cur;
2807 
2808 	if (nes->entry_cnt >= max)
2809 		goto add_out;
2810 
2811 	list_for_each_entry(cur, head, set_list) {
2812 		if (cur->entry_cnt >= nes->entry_cnt) {
2813 			list_add(&nes->set_list, cur->set_list.prev);
2814 			return;
2815 		}
2816 	}
2817 add_out:
2818 	list_add_tail(&nes->set_list, head);
2819 }
2820 
2821 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2822 						struct page *page)
2823 {
2824 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2825 	unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2826 	struct f2fs_nat_block *nat_blk = page_address(page);
2827 	int valid = 0;
2828 	int i = 0;
2829 
2830 	if (!enabled_nat_bits(sbi, NULL))
2831 		return;
2832 
2833 	if (nat_index == 0) {
2834 		valid = 1;
2835 		i = 1;
2836 	}
2837 	for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2838 		if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2839 			valid++;
2840 	}
2841 	if (valid == 0) {
2842 		__set_bit_le(nat_index, nm_i->empty_nat_bits);
2843 		__clear_bit_le(nat_index, nm_i->full_nat_bits);
2844 		return;
2845 	}
2846 
2847 	__clear_bit_le(nat_index, nm_i->empty_nat_bits);
2848 	if (valid == NAT_ENTRY_PER_BLOCK)
2849 		__set_bit_le(nat_index, nm_i->full_nat_bits);
2850 	else
2851 		__clear_bit_le(nat_index, nm_i->full_nat_bits);
2852 }
2853 
2854 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2855 		struct nat_entry_set *set, struct cp_control *cpc)
2856 {
2857 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2858 	struct f2fs_journal *journal = curseg->journal;
2859 	nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2860 	bool to_journal = true;
2861 	struct f2fs_nat_block *nat_blk;
2862 	struct nat_entry *ne, *cur;
2863 	struct page *page = NULL;
2864 
2865 	/*
2866 	 * there are two steps to flush nat entries:
2867 	 * #1, flush nat entries to journal in current hot data summary block.
2868 	 * #2, flush nat entries to nat page.
2869 	 */
2870 	if (enabled_nat_bits(sbi, cpc) ||
2871 		!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2872 		to_journal = false;
2873 
2874 	if (to_journal) {
2875 		down_write(&curseg->journal_rwsem);
2876 	} else {
2877 		page = get_next_nat_page(sbi, start_nid);
2878 		if (IS_ERR(page))
2879 			return PTR_ERR(page);
2880 
2881 		nat_blk = page_address(page);
2882 		f2fs_bug_on(sbi, !nat_blk);
2883 	}
2884 
2885 	/* flush dirty nats in nat entry set */
2886 	list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2887 		struct f2fs_nat_entry *raw_ne;
2888 		nid_t nid = nat_get_nid(ne);
2889 		int offset;
2890 
2891 		f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2892 
2893 		if (to_journal) {
2894 			offset = f2fs_lookup_journal_in_cursum(journal,
2895 							NAT_JOURNAL, nid, 1);
2896 			f2fs_bug_on(sbi, offset < 0);
2897 			raw_ne = &nat_in_journal(journal, offset);
2898 			nid_in_journal(journal, offset) = cpu_to_le32(nid);
2899 		} else {
2900 			raw_ne = &nat_blk->entries[nid - start_nid];
2901 		}
2902 		raw_nat_from_node_info(raw_ne, &ne->ni);
2903 		nat_reset_flag(ne);
2904 		__clear_nat_cache_dirty(NM_I(sbi), set, ne);
2905 		if (nat_get_blkaddr(ne) == NULL_ADDR) {
2906 			add_free_nid(sbi, nid, false, true);
2907 		} else {
2908 			spin_lock(&NM_I(sbi)->nid_list_lock);
2909 			update_free_nid_bitmap(sbi, nid, false, false);
2910 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2911 		}
2912 	}
2913 
2914 	if (to_journal) {
2915 		up_write(&curseg->journal_rwsem);
2916 	} else {
2917 		__update_nat_bits(sbi, start_nid, page);
2918 		f2fs_put_page(page, 1);
2919 	}
2920 
2921 	/* Allow dirty nats by node block allocation in write_begin */
2922 	if (!set->entry_cnt) {
2923 		radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2924 		kmem_cache_free(nat_entry_set_slab, set);
2925 	}
2926 	return 0;
2927 }
2928 
2929 /*
2930  * This function is called during the checkpointing process.
2931  */
2932 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2933 {
2934 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2935 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2936 	struct f2fs_journal *journal = curseg->journal;
2937 	struct nat_entry_set *setvec[SETVEC_SIZE];
2938 	struct nat_entry_set *set, *tmp;
2939 	unsigned int found;
2940 	nid_t set_idx = 0;
2941 	LIST_HEAD(sets);
2942 	int err = 0;
2943 
2944 	/* during unmount, let's flush nat_bits before checking dirty_nat_cnt */
2945 	if (enabled_nat_bits(sbi, cpc)) {
2946 		down_write(&nm_i->nat_tree_lock);
2947 		remove_nats_in_journal(sbi);
2948 		up_write(&nm_i->nat_tree_lock);
2949 	}
2950 
2951 	if (!nm_i->dirty_nat_cnt)
2952 		return 0;
2953 
2954 	down_write(&nm_i->nat_tree_lock);
2955 
2956 	/*
2957 	 * if there are no enough space in journal to store dirty nat
2958 	 * entries, remove all entries from journal and merge them
2959 	 * into nat entry set.
2960 	 */
2961 	if (enabled_nat_bits(sbi, cpc) ||
2962 		!__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2963 		remove_nats_in_journal(sbi);
2964 
2965 	while ((found = __gang_lookup_nat_set(nm_i,
2966 					set_idx, SETVEC_SIZE, setvec))) {
2967 		unsigned idx;
2968 		set_idx = setvec[found - 1]->set + 1;
2969 		for (idx = 0; idx < found; idx++)
2970 			__adjust_nat_entry_set(setvec[idx], &sets,
2971 						MAX_NAT_JENTRIES(journal));
2972 	}
2973 
2974 	/* flush dirty nats in nat entry set */
2975 	list_for_each_entry_safe(set, tmp, &sets, set_list) {
2976 		err = __flush_nat_entry_set(sbi, set, cpc);
2977 		if (err)
2978 			break;
2979 	}
2980 
2981 	up_write(&nm_i->nat_tree_lock);
2982 	/* Allow dirty nats by node block allocation in write_begin */
2983 
2984 	return err;
2985 }
2986 
2987 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2988 {
2989 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2990 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2991 	unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2992 	unsigned int i;
2993 	__u64 cp_ver = cur_cp_version(ckpt);
2994 	block_t nat_bits_addr;
2995 
2996 	if (!enabled_nat_bits(sbi, NULL))
2997 		return 0;
2998 
2999 	nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3000 	nm_i->nat_bits = f2fs_kvzalloc(sbi,
3001 			nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3002 	if (!nm_i->nat_bits)
3003 		return -ENOMEM;
3004 
3005 	nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3006 						nm_i->nat_bits_blocks;
3007 	for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3008 		struct page *page;
3009 
3010 		page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3011 		if (IS_ERR(page))
3012 			return PTR_ERR(page);
3013 
3014 		memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3015 					page_address(page), F2FS_BLKSIZE);
3016 		f2fs_put_page(page, 1);
3017 	}
3018 
3019 	cp_ver |= (cur_cp_crc(ckpt) << 32);
3020 	if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3021 		disable_nat_bits(sbi, true);
3022 		return 0;
3023 	}
3024 
3025 	nm_i->full_nat_bits = nm_i->nat_bits + 8;
3026 	nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3027 
3028 	f2fs_notice(sbi, "Found nat_bits in checkpoint");
3029 	return 0;
3030 }
3031 
3032 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3033 {
3034 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3035 	unsigned int i = 0;
3036 	nid_t nid, last_nid;
3037 
3038 	if (!enabled_nat_bits(sbi, NULL))
3039 		return;
3040 
3041 	for (i = 0; i < nm_i->nat_blocks; i++) {
3042 		i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3043 		if (i >= nm_i->nat_blocks)
3044 			break;
3045 
3046 		__set_bit_le(i, nm_i->nat_block_bitmap);
3047 
3048 		nid = i * NAT_ENTRY_PER_BLOCK;
3049 		last_nid = nid + NAT_ENTRY_PER_BLOCK;
3050 
3051 		spin_lock(&NM_I(sbi)->nid_list_lock);
3052 		for (; nid < last_nid; nid++)
3053 			update_free_nid_bitmap(sbi, nid, true, true);
3054 		spin_unlock(&NM_I(sbi)->nid_list_lock);
3055 	}
3056 
3057 	for (i = 0; i < nm_i->nat_blocks; i++) {
3058 		i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3059 		if (i >= nm_i->nat_blocks)
3060 			break;
3061 
3062 		__set_bit_le(i, nm_i->nat_block_bitmap);
3063 	}
3064 }
3065 
3066 static int init_node_manager(struct f2fs_sb_info *sbi)
3067 {
3068 	struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3069 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3070 	unsigned char *version_bitmap;
3071 	unsigned int nat_segs;
3072 	int err;
3073 
3074 	nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3075 
3076 	/* segment_count_nat includes pair segment so divide to 2. */
3077 	nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3078 	nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3079 	nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3080 
3081 	/* not used nids: 0, node, meta, (and root counted as valid node) */
3082 	nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3083 						F2FS_RESERVED_NODE_NUM;
3084 	nm_i->nid_cnt[FREE_NID] = 0;
3085 	nm_i->nid_cnt[PREALLOC_NID] = 0;
3086 	nm_i->nat_cnt = 0;
3087 	nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3088 	nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3089 	nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3090 
3091 	INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3092 	INIT_LIST_HEAD(&nm_i->free_nid_list);
3093 	INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3094 	INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3095 	INIT_LIST_HEAD(&nm_i->nat_entries);
3096 	spin_lock_init(&nm_i->nat_list_lock);
3097 
3098 	mutex_init(&nm_i->build_lock);
3099 	spin_lock_init(&nm_i->nid_list_lock);
3100 	init_rwsem(&nm_i->nat_tree_lock);
3101 
3102 	nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3103 	nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3104 	version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3105 	if (!version_bitmap)
3106 		return -EFAULT;
3107 
3108 	nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3109 					GFP_KERNEL);
3110 	if (!nm_i->nat_bitmap)
3111 		return -ENOMEM;
3112 
3113 	err = __get_nat_bitmaps(sbi);
3114 	if (err)
3115 		return err;
3116 
3117 #ifdef CONFIG_F2FS_CHECK_FS
3118 	nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3119 					GFP_KERNEL);
3120 	if (!nm_i->nat_bitmap_mir)
3121 		return -ENOMEM;
3122 #endif
3123 
3124 	return 0;
3125 }
3126 
3127 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3128 {
3129 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3130 	int i;
3131 
3132 	nm_i->free_nid_bitmap =
3133 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3134 					      nm_i->nat_blocks),
3135 			      GFP_KERNEL);
3136 	if (!nm_i->free_nid_bitmap)
3137 		return -ENOMEM;
3138 
3139 	for (i = 0; i < nm_i->nat_blocks; i++) {
3140 		nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3141 			f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3142 		if (!nm_i->free_nid_bitmap[i])
3143 			return -ENOMEM;
3144 	}
3145 
3146 	nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3147 								GFP_KERNEL);
3148 	if (!nm_i->nat_block_bitmap)
3149 		return -ENOMEM;
3150 
3151 	nm_i->free_nid_count =
3152 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3153 					      nm_i->nat_blocks),
3154 			      GFP_KERNEL);
3155 	if (!nm_i->free_nid_count)
3156 		return -ENOMEM;
3157 	return 0;
3158 }
3159 
3160 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3161 {
3162 	int err;
3163 
3164 	sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3165 							GFP_KERNEL);
3166 	if (!sbi->nm_info)
3167 		return -ENOMEM;
3168 
3169 	err = init_node_manager(sbi);
3170 	if (err)
3171 		return err;
3172 
3173 	err = init_free_nid_cache(sbi);
3174 	if (err)
3175 		return err;
3176 
3177 	/* load free nid status from nat_bits table */
3178 	load_free_nid_bitmap(sbi);
3179 
3180 	return f2fs_build_free_nids(sbi, true, true);
3181 }
3182 
3183 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3184 {
3185 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3186 	struct free_nid *i, *next_i;
3187 	struct nat_entry *natvec[NATVEC_SIZE];
3188 	struct nat_entry_set *setvec[SETVEC_SIZE];
3189 	nid_t nid = 0;
3190 	unsigned int found;
3191 
3192 	if (!nm_i)
3193 		return;
3194 
3195 	/* destroy free nid list */
3196 	spin_lock(&nm_i->nid_list_lock);
3197 	list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3198 		__remove_free_nid(sbi, i, FREE_NID);
3199 		spin_unlock(&nm_i->nid_list_lock);
3200 		kmem_cache_free(free_nid_slab, i);
3201 		spin_lock(&nm_i->nid_list_lock);
3202 	}
3203 	f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3204 	f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3205 	f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3206 	spin_unlock(&nm_i->nid_list_lock);
3207 
3208 	/* destroy nat cache */
3209 	down_write(&nm_i->nat_tree_lock);
3210 	while ((found = __gang_lookup_nat_cache(nm_i,
3211 					nid, NATVEC_SIZE, natvec))) {
3212 		unsigned idx;
3213 
3214 		nid = nat_get_nid(natvec[found - 1]) + 1;
3215 		for (idx = 0; idx < found; idx++) {
3216 			spin_lock(&nm_i->nat_list_lock);
3217 			list_del(&natvec[idx]->list);
3218 			spin_unlock(&nm_i->nat_list_lock);
3219 
3220 			__del_from_nat_cache(nm_i, natvec[idx]);
3221 		}
3222 	}
3223 	f2fs_bug_on(sbi, nm_i->nat_cnt);
3224 
3225 	/* destroy nat set cache */
3226 	nid = 0;
3227 	while ((found = __gang_lookup_nat_set(nm_i,
3228 					nid, SETVEC_SIZE, setvec))) {
3229 		unsigned idx;
3230 
3231 		nid = setvec[found - 1]->set + 1;
3232 		for (idx = 0; idx < found; idx++) {
3233 			/* entry_cnt is not zero, when cp_error was occurred */
3234 			f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3235 			radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3236 			kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3237 		}
3238 	}
3239 	up_write(&nm_i->nat_tree_lock);
3240 
3241 	kvfree(nm_i->nat_block_bitmap);
3242 	if (nm_i->free_nid_bitmap) {
3243 		int i;
3244 
3245 		for (i = 0; i < nm_i->nat_blocks; i++)
3246 			kvfree(nm_i->free_nid_bitmap[i]);
3247 		kvfree(nm_i->free_nid_bitmap);
3248 	}
3249 	kvfree(nm_i->free_nid_count);
3250 
3251 	kvfree(nm_i->nat_bitmap);
3252 	kvfree(nm_i->nat_bits);
3253 #ifdef CONFIG_F2FS_CHECK_FS
3254 	kvfree(nm_i->nat_bitmap_mir);
3255 #endif
3256 	sbi->nm_info = NULL;
3257 	kvfree(nm_i);
3258 }
3259 
3260 int __init f2fs_create_node_manager_caches(void)
3261 {
3262 	nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3263 			sizeof(struct nat_entry));
3264 	if (!nat_entry_slab)
3265 		goto fail;
3266 
3267 	free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3268 			sizeof(struct free_nid));
3269 	if (!free_nid_slab)
3270 		goto destroy_nat_entry;
3271 
3272 	nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3273 			sizeof(struct nat_entry_set));
3274 	if (!nat_entry_set_slab)
3275 		goto destroy_free_nid;
3276 
3277 	fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3278 			sizeof(struct fsync_node_entry));
3279 	if (!fsync_node_entry_slab)
3280 		goto destroy_nat_entry_set;
3281 	return 0;
3282 
3283 destroy_nat_entry_set:
3284 	kmem_cache_destroy(nat_entry_set_slab);
3285 destroy_free_nid:
3286 	kmem_cache_destroy(free_nid_slab);
3287 destroy_nat_entry:
3288 	kmem_cache_destroy(nat_entry_slab);
3289 fail:
3290 	return -ENOMEM;
3291 }
3292 
3293 void f2fs_destroy_node_manager_caches(void)
3294 {
3295 	kmem_cache_destroy(fsync_node_entry_slab);
3296 	kmem_cache_destroy(nat_entry_set_slab);
3297 	kmem_cache_destroy(free_nid_slab);
3298 	kmem_cache_destroy(nat_entry_slab);
3299 }
3300