xref: /openbmc/linux/fs/f2fs/node.c (revision 5085e036)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * fs/f2fs/node.c
4  *
5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6  *             http://www.samsung.com/
7  */
8 #include <linux/fs.h>
9 #include <linux/f2fs_fs.h>
10 #include <linux/mpage.h>
11 #include <linux/sched/mm.h>
12 #include <linux/blkdev.h>
13 #include <linux/pagevec.h>
14 #include <linux/swap.h>
15 
16 #include "f2fs.h"
17 #include "node.h"
18 #include "segment.h"
19 #include "xattr.h"
20 #include "iostat.h"
21 #include <trace/events/f2fs.h>
22 
23 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
24 
25 static struct kmem_cache *nat_entry_slab;
26 static struct kmem_cache *free_nid_slab;
27 static struct kmem_cache *nat_entry_set_slab;
28 static struct kmem_cache *fsync_node_entry_slab;
29 
30 /*
31  * Check whether the given nid is within node id range.
32  */
33 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
34 {
35 	if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
36 		set_sbi_flag(sbi, SBI_NEED_FSCK);
37 		f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
38 			  __func__, nid);
39 		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 discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
48 	struct sysinfo val;
49 	unsigned long avail_ram;
50 	unsigned long mem_size = 0;
51 	bool res = false;
52 
53 	if (!nm_i)
54 		return true;
55 
56 	si_meminfo(&val);
57 
58 	/* only uses low memory */
59 	avail_ram = val.totalram - val.totalhigh;
60 
61 	/*
62 	 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
63 	 */
64 	if (type == FREE_NIDS) {
65 		mem_size = (nm_i->nid_cnt[FREE_NID] *
66 				sizeof(struct free_nid)) >> PAGE_SHIFT;
67 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
68 	} else if (type == NAT_ENTRIES) {
69 		mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
70 				sizeof(struct nat_entry)) >> PAGE_SHIFT;
71 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
72 		if (excess_cached_nats(sbi))
73 			res = false;
74 	} else if (type == DIRTY_DENTS) {
75 		if (sbi->sb->s_bdi->wb.dirty_exceeded)
76 			return false;
77 		mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
78 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
79 	} else if (type == INO_ENTRIES) {
80 		int i;
81 
82 		for (i = 0; i < MAX_INO_ENTRY; i++)
83 			mem_size += sbi->im[i].ino_num *
84 						sizeof(struct ino_entry);
85 		mem_size >>= PAGE_SHIFT;
86 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
87 	} else if (type == EXTENT_CACHE) {
88 		mem_size = (atomic_read(&sbi->total_ext_tree) *
89 				sizeof(struct extent_tree) +
90 				atomic_read(&sbi->total_ext_node) *
91 				sizeof(struct extent_node)) >> PAGE_SHIFT;
92 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
93 	} else if (type == DISCARD_CACHE) {
94 		mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
95 				sizeof(struct discard_cmd)) >> PAGE_SHIFT;
96 		res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
97 	} else if (type == COMPRESS_PAGE) {
98 #ifdef CONFIG_F2FS_FS_COMPRESSION
99 		unsigned long free_ram = val.freeram;
100 
101 		/*
102 		 * free memory is lower than watermark or cached page count
103 		 * exceed threshold, deny caching compress page.
104 		 */
105 		res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
106 			(COMPRESS_MAPPING(sbi)->nrpages <
107 			 free_ram * sbi->compress_percent / 100);
108 #else
109 		res = false;
110 #endif
111 	} else {
112 		if (!sbi->sb->s_bdi->wb.dirty_exceeded)
113 			return true;
114 	}
115 	return res;
116 }
117 
118 static void clear_node_page_dirty(struct page *page)
119 {
120 	if (PageDirty(page)) {
121 		f2fs_clear_page_cache_dirty_tag(page);
122 		clear_page_dirty_for_io(page);
123 		dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
124 	}
125 	ClearPageUptodate(page);
126 }
127 
128 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
129 {
130 	return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
131 }
132 
133 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
134 {
135 	struct page *src_page;
136 	struct page *dst_page;
137 	pgoff_t dst_off;
138 	void *src_addr;
139 	void *dst_addr;
140 	struct f2fs_nm_info *nm_i = NM_I(sbi);
141 
142 	dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
143 
144 	/* get current nat block page with lock */
145 	src_page = get_current_nat_page(sbi, nid);
146 	if (IS_ERR(src_page))
147 		return src_page;
148 	dst_page = f2fs_grab_meta_page(sbi, dst_off);
149 	f2fs_bug_on(sbi, PageDirty(src_page));
150 
151 	src_addr = page_address(src_page);
152 	dst_addr = page_address(dst_page);
153 	memcpy(dst_addr, src_addr, PAGE_SIZE);
154 	set_page_dirty(dst_page);
155 	f2fs_put_page(src_page, 1);
156 
157 	set_to_next_nat(nm_i, nid);
158 
159 	return dst_page;
160 }
161 
162 static struct nat_entry *__alloc_nat_entry(struct f2fs_sb_info *sbi,
163 						nid_t nid, bool no_fail)
164 {
165 	struct nat_entry *new;
166 
167 	new = f2fs_kmem_cache_alloc(nat_entry_slab,
168 					GFP_F2FS_ZERO, no_fail, sbi);
169 	if (new) {
170 		nat_set_nid(new, nid);
171 		nat_reset_flag(new);
172 	}
173 	return new;
174 }
175 
176 static void __free_nat_entry(struct nat_entry *e)
177 {
178 	kmem_cache_free(nat_entry_slab, e);
179 }
180 
181 /* must be locked by nat_tree_lock */
182 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
183 	struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
184 {
185 	if (no_fail)
186 		f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
187 	else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
188 		return NULL;
189 
190 	if (raw_ne)
191 		node_info_from_raw_nat(&ne->ni, raw_ne);
192 
193 	spin_lock(&nm_i->nat_list_lock);
194 	list_add_tail(&ne->list, &nm_i->nat_entries);
195 	spin_unlock(&nm_i->nat_list_lock);
196 
197 	nm_i->nat_cnt[TOTAL_NAT]++;
198 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
199 	return ne;
200 }
201 
202 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
203 {
204 	struct nat_entry *ne;
205 
206 	ne = radix_tree_lookup(&nm_i->nat_root, n);
207 
208 	/* for recent accessed nat entry, move it to tail of lru list */
209 	if (ne && !get_nat_flag(ne, IS_DIRTY)) {
210 		spin_lock(&nm_i->nat_list_lock);
211 		if (!list_empty(&ne->list))
212 			list_move_tail(&ne->list, &nm_i->nat_entries);
213 		spin_unlock(&nm_i->nat_list_lock);
214 	}
215 
216 	return ne;
217 }
218 
219 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
220 		nid_t start, unsigned int nr, struct nat_entry **ep)
221 {
222 	return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
223 }
224 
225 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
226 {
227 	radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
228 	nm_i->nat_cnt[TOTAL_NAT]--;
229 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
230 	__free_nat_entry(e);
231 }
232 
233 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
234 							struct nat_entry *ne)
235 {
236 	nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
237 	struct nat_entry_set *head;
238 
239 	head = radix_tree_lookup(&nm_i->nat_set_root, set);
240 	if (!head) {
241 		head = f2fs_kmem_cache_alloc(nat_entry_set_slab,
242 						GFP_NOFS, true, NULL);
243 
244 		INIT_LIST_HEAD(&head->entry_list);
245 		INIT_LIST_HEAD(&head->set_list);
246 		head->set = set;
247 		head->entry_cnt = 0;
248 		f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
249 	}
250 	return head;
251 }
252 
253 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
254 						struct nat_entry *ne)
255 {
256 	struct nat_entry_set *head;
257 	bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
258 
259 	if (!new_ne)
260 		head = __grab_nat_entry_set(nm_i, ne);
261 
262 	/*
263 	 * update entry_cnt in below condition:
264 	 * 1. update NEW_ADDR to valid block address;
265 	 * 2. update old block address to new one;
266 	 */
267 	if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
268 				!get_nat_flag(ne, IS_DIRTY)))
269 		head->entry_cnt++;
270 
271 	set_nat_flag(ne, IS_PREALLOC, new_ne);
272 
273 	if (get_nat_flag(ne, IS_DIRTY))
274 		goto refresh_list;
275 
276 	nm_i->nat_cnt[DIRTY_NAT]++;
277 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
278 	set_nat_flag(ne, IS_DIRTY, true);
279 refresh_list:
280 	spin_lock(&nm_i->nat_list_lock);
281 	if (new_ne)
282 		list_del_init(&ne->list);
283 	else
284 		list_move_tail(&ne->list, &head->entry_list);
285 	spin_unlock(&nm_i->nat_list_lock);
286 }
287 
288 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
289 		struct nat_entry_set *set, struct nat_entry *ne)
290 {
291 	spin_lock(&nm_i->nat_list_lock);
292 	list_move_tail(&ne->list, &nm_i->nat_entries);
293 	spin_unlock(&nm_i->nat_list_lock);
294 
295 	set_nat_flag(ne, IS_DIRTY, false);
296 	set->entry_cnt--;
297 	nm_i->nat_cnt[DIRTY_NAT]--;
298 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
299 }
300 
301 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
302 		nid_t start, unsigned int nr, struct nat_entry_set **ep)
303 {
304 	return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
305 							start, nr);
306 }
307 
308 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
309 {
310 	return NODE_MAPPING(sbi) == page->mapping &&
311 			IS_DNODE(page) && is_cold_node(page);
312 }
313 
314 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
315 {
316 	spin_lock_init(&sbi->fsync_node_lock);
317 	INIT_LIST_HEAD(&sbi->fsync_node_list);
318 	sbi->fsync_seg_id = 0;
319 	sbi->fsync_node_num = 0;
320 }
321 
322 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
323 							struct page *page)
324 {
325 	struct fsync_node_entry *fn;
326 	unsigned long flags;
327 	unsigned int seq_id;
328 
329 	fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab,
330 					GFP_NOFS, true, NULL);
331 
332 	get_page(page);
333 	fn->page = page;
334 	INIT_LIST_HEAD(&fn->list);
335 
336 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
337 	list_add_tail(&fn->list, &sbi->fsync_node_list);
338 	fn->seq_id = sbi->fsync_seg_id++;
339 	seq_id = fn->seq_id;
340 	sbi->fsync_node_num++;
341 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
342 
343 	return seq_id;
344 }
345 
346 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
347 {
348 	struct fsync_node_entry *fn;
349 	unsigned long flags;
350 
351 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
352 	list_for_each_entry(fn, &sbi->fsync_node_list, list) {
353 		if (fn->page == page) {
354 			list_del(&fn->list);
355 			sbi->fsync_node_num--;
356 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
357 			kmem_cache_free(fsync_node_entry_slab, fn);
358 			put_page(page);
359 			return;
360 		}
361 	}
362 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
363 	f2fs_bug_on(sbi, 1);
364 }
365 
366 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
367 {
368 	unsigned long flags;
369 
370 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
371 	sbi->fsync_seg_id = 0;
372 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
373 }
374 
375 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
376 {
377 	struct f2fs_nm_info *nm_i = NM_I(sbi);
378 	struct nat_entry *e;
379 	bool need = false;
380 
381 	f2fs_down_read(&nm_i->nat_tree_lock);
382 	e = __lookup_nat_cache(nm_i, nid);
383 	if (e) {
384 		if (!get_nat_flag(e, IS_CHECKPOINTED) &&
385 				!get_nat_flag(e, HAS_FSYNCED_INODE))
386 			need = true;
387 	}
388 	f2fs_up_read(&nm_i->nat_tree_lock);
389 	return need;
390 }
391 
392 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
393 {
394 	struct f2fs_nm_info *nm_i = NM_I(sbi);
395 	struct nat_entry *e;
396 	bool is_cp = true;
397 
398 	f2fs_down_read(&nm_i->nat_tree_lock);
399 	e = __lookup_nat_cache(nm_i, nid);
400 	if (e && !get_nat_flag(e, IS_CHECKPOINTED))
401 		is_cp = false;
402 	f2fs_up_read(&nm_i->nat_tree_lock);
403 	return is_cp;
404 }
405 
406 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
407 {
408 	struct f2fs_nm_info *nm_i = NM_I(sbi);
409 	struct nat_entry *e;
410 	bool need_update = true;
411 
412 	f2fs_down_read(&nm_i->nat_tree_lock);
413 	e = __lookup_nat_cache(nm_i, ino);
414 	if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
415 			(get_nat_flag(e, IS_CHECKPOINTED) ||
416 			 get_nat_flag(e, HAS_FSYNCED_INODE)))
417 		need_update = false;
418 	f2fs_up_read(&nm_i->nat_tree_lock);
419 	return need_update;
420 }
421 
422 /* must be locked by nat_tree_lock */
423 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
424 						struct f2fs_nat_entry *ne)
425 {
426 	struct f2fs_nm_info *nm_i = NM_I(sbi);
427 	struct nat_entry *new, *e;
428 
429 	/* Let's mitigate lock contention of nat_tree_lock during checkpoint */
430 	if (f2fs_rwsem_is_locked(&sbi->cp_global_sem))
431 		return;
432 
433 	new = __alloc_nat_entry(sbi, nid, false);
434 	if (!new)
435 		return;
436 
437 	f2fs_down_write(&nm_i->nat_tree_lock);
438 	e = __lookup_nat_cache(nm_i, nid);
439 	if (!e)
440 		e = __init_nat_entry(nm_i, new, ne, false);
441 	else
442 		f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
443 				nat_get_blkaddr(e) !=
444 					le32_to_cpu(ne->block_addr) ||
445 				nat_get_version(e) != ne->version);
446 	f2fs_up_write(&nm_i->nat_tree_lock);
447 	if (e != new)
448 		__free_nat_entry(new);
449 }
450 
451 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
452 			block_t new_blkaddr, bool fsync_done)
453 {
454 	struct f2fs_nm_info *nm_i = NM_I(sbi);
455 	struct nat_entry *e;
456 	struct nat_entry *new = __alloc_nat_entry(sbi, ni->nid, true);
457 
458 	f2fs_down_write(&nm_i->nat_tree_lock);
459 	e = __lookup_nat_cache(nm_i, ni->nid);
460 	if (!e) {
461 		e = __init_nat_entry(nm_i, new, NULL, true);
462 		copy_node_info(&e->ni, ni);
463 		f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
464 	} else if (new_blkaddr == NEW_ADDR) {
465 		/*
466 		 * when nid is reallocated,
467 		 * previous nat entry can be remained in nat cache.
468 		 * So, reinitialize it with new information.
469 		 */
470 		copy_node_info(&e->ni, ni);
471 		f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
472 	}
473 	/* let's free early to reduce memory consumption */
474 	if (e != new)
475 		__free_nat_entry(new);
476 
477 	/* sanity check */
478 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
479 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
480 			new_blkaddr == NULL_ADDR);
481 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
482 			new_blkaddr == NEW_ADDR);
483 	f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
484 			new_blkaddr == NEW_ADDR);
485 
486 	/* increment version no as node is removed */
487 	if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
488 		unsigned char version = nat_get_version(e);
489 
490 		nat_set_version(e, inc_node_version(version));
491 	}
492 
493 	/* change address */
494 	nat_set_blkaddr(e, new_blkaddr);
495 	if (!__is_valid_data_blkaddr(new_blkaddr))
496 		set_nat_flag(e, IS_CHECKPOINTED, false);
497 	__set_nat_cache_dirty(nm_i, e);
498 
499 	/* update fsync_mark if its inode nat entry is still alive */
500 	if (ni->nid != ni->ino)
501 		e = __lookup_nat_cache(nm_i, ni->ino);
502 	if (e) {
503 		if (fsync_done && ni->nid == ni->ino)
504 			set_nat_flag(e, HAS_FSYNCED_INODE, true);
505 		set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
506 	}
507 	f2fs_up_write(&nm_i->nat_tree_lock);
508 }
509 
510 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
511 {
512 	struct f2fs_nm_info *nm_i = NM_I(sbi);
513 	int nr = nr_shrink;
514 
515 	if (!f2fs_down_write_trylock(&nm_i->nat_tree_lock))
516 		return 0;
517 
518 	spin_lock(&nm_i->nat_list_lock);
519 	while (nr_shrink) {
520 		struct nat_entry *ne;
521 
522 		if (list_empty(&nm_i->nat_entries))
523 			break;
524 
525 		ne = list_first_entry(&nm_i->nat_entries,
526 					struct nat_entry, list);
527 		list_del(&ne->list);
528 		spin_unlock(&nm_i->nat_list_lock);
529 
530 		__del_from_nat_cache(nm_i, ne);
531 		nr_shrink--;
532 
533 		spin_lock(&nm_i->nat_list_lock);
534 	}
535 	spin_unlock(&nm_i->nat_list_lock);
536 
537 	f2fs_up_write(&nm_i->nat_tree_lock);
538 	return nr - nr_shrink;
539 }
540 
541 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
542 				struct node_info *ni, bool checkpoint_context)
543 {
544 	struct f2fs_nm_info *nm_i = NM_I(sbi);
545 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
546 	struct f2fs_journal *journal = curseg->journal;
547 	nid_t start_nid = START_NID(nid);
548 	struct f2fs_nat_block *nat_blk;
549 	struct page *page = NULL;
550 	struct f2fs_nat_entry ne;
551 	struct nat_entry *e;
552 	pgoff_t index;
553 	block_t blkaddr;
554 	int i;
555 
556 	ni->nid = nid;
557 retry:
558 	/* Check nat cache */
559 	f2fs_down_read(&nm_i->nat_tree_lock);
560 	e = __lookup_nat_cache(nm_i, nid);
561 	if (e) {
562 		ni->ino = nat_get_ino(e);
563 		ni->blk_addr = nat_get_blkaddr(e);
564 		ni->version = nat_get_version(e);
565 		f2fs_up_read(&nm_i->nat_tree_lock);
566 		return 0;
567 	}
568 
569 	/*
570 	 * Check current segment summary by trying to grab journal_rwsem first.
571 	 * This sem is on the critical path on the checkpoint requiring the above
572 	 * nat_tree_lock. Therefore, we should retry, if we failed to grab here
573 	 * while not bothering checkpoint.
574 	 */
575 	if (!f2fs_rwsem_is_locked(&sbi->cp_global_sem) || checkpoint_context) {
576 		down_read(&curseg->journal_rwsem);
577 	} else if (f2fs_rwsem_is_contended(&nm_i->nat_tree_lock) ||
578 				!down_read_trylock(&curseg->journal_rwsem)) {
579 		f2fs_up_read(&nm_i->nat_tree_lock);
580 		goto retry;
581 	}
582 
583 	i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
584 	if (i >= 0) {
585 		ne = nat_in_journal(journal, i);
586 		node_info_from_raw_nat(ni, &ne);
587 	}
588         up_read(&curseg->journal_rwsem);
589 	if (i >= 0) {
590 		f2fs_up_read(&nm_i->nat_tree_lock);
591 		goto cache;
592 	}
593 
594 	/* Fill node_info from nat page */
595 	index = current_nat_addr(sbi, nid);
596 	f2fs_up_read(&nm_i->nat_tree_lock);
597 
598 	page = f2fs_get_meta_page(sbi, index);
599 	if (IS_ERR(page))
600 		return PTR_ERR(page);
601 
602 	nat_blk = (struct f2fs_nat_block *)page_address(page);
603 	ne = nat_blk->entries[nid - start_nid];
604 	node_info_from_raw_nat(ni, &ne);
605 	f2fs_put_page(page, 1);
606 cache:
607 	blkaddr = le32_to_cpu(ne.block_addr);
608 	if (__is_valid_data_blkaddr(blkaddr) &&
609 		!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
610 		return -EFAULT;
611 
612 	/* cache nat entry */
613 	cache_nat_entry(sbi, nid, &ne);
614 	return 0;
615 }
616 
617 /*
618  * readahead MAX_RA_NODE number of node pages.
619  */
620 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
621 {
622 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
623 	struct blk_plug plug;
624 	int i, end;
625 	nid_t nid;
626 
627 	blk_start_plug(&plug);
628 
629 	/* Then, try readahead for siblings of the desired node */
630 	end = start + n;
631 	end = min(end, NIDS_PER_BLOCK);
632 	for (i = start; i < end; i++) {
633 		nid = get_nid(parent, i, false);
634 		f2fs_ra_node_page(sbi, nid);
635 	}
636 
637 	blk_finish_plug(&plug);
638 }
639 
640 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
641 {
642 	const long direct_index = ADDRS_PER_INODE(dn->inode);
643 	const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
644 	const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
645 	unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
646 	int cur_level = dn->cur_level;
647 	int max_level = dn->max_level;
648 	pgoff_t base = 0;
649 
650 	if (!dn->max_level)
651 		return pgofs + 1;
652 
653 	while (max_level-- > cur_level)
654 		skipped_unit *= NIDS_PER_BLOCK;
655 
656 	switch (dn->max_level) {
657 	case 3:
658 		base += 2 * indirect_blks;
659 		fallthrough;
660 	case 2:
661 		base += 2 * direct_blks;
662 		fallthrough;
663 	case 1:
664 		base += direct_index;
665 		break;
666 	default:
667 		f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
668 	}
669 
670 	return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
671 }
672 
673 /*
674  * The maximum depth is four.
675  * Offset[0] will have raw inode offset.
676  */
677 static int get_node_path(struct inode *inode, long block,
678 				int offset[4], unsigned int noffset[4])
679 {
680 	const long direct_index = ADDRS_PER_INODE(inode);
681 	const long direct_blks = ADDRS_PER_BLOCK(inode);
682 	const long dptrs_per_blk = NIDS_PER_BLOCK;
683 	const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
684 	const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
685 	int n = 0;
686 	int level = 0;
687 
688 	noffset[0] = 0;
689 
690 	if (block < direct_index) {
691 		offset[n] = block;
692 		goto got;
693 	}
694 	block -= direct_index;
695 	if (block < direct_blks) {
696 		offset[n++] = NODE_DIR1_BLOCK;
697 		noffset[n] = 1;
698 		offset[n] = block;
699 		level = 1;
700 		goto got;
701 	}
702 	block -= direct_blks;
703 	if (block < direct_blks) {
704 		offset[n++] = NODE_DIR2_BLOCK;
705 		noffset[n] = 2;
706 		offset[n] = block;
707 		level = 1;
708 		goto got;
709 	}
710 	block -= direct_blks;
711 	if (block < indirect_blks) {
712 		offset[n++] = NODE_IND1_BLOCK;
713 		noffset[n] = 3;
714 		offset[n++] = block / direct_blks;
715 		noffset[n] = 4 + offset[n - 1];
716 		offset[n] = block % direct_blks;
717 		level = 2;
718 		goto got;
719 	}
720 	block -= indirect_blks;
721 	if (block < indirect_blks) {
722 		offset[n++] = NODE_IND2_BLOCK;
723 		noffset[n] = 4 + dptrs_per_blk;
724 		offset[n++] = block / direct_blks;
725 		noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
726 		offset[n] = block % direct_blks;
727 		level = 2;
728 		goto got;
729 	}
730 	block -= indirect_blks;
731 	if (block < dindirect_blks) {
732 		offset[n++] = NODE_DIND_BLOCK;
733 		noffset[n] = 5 + (dptrs_per_blk * 2);
734 		offset[n++] = block / indirect_blks;
735 		noffset[n] = 6 + (dptrs_per_blk * 2) +
736 			      offset[n - 1] * (dptrs_per_blk + 1);
737 		offset[n++] = (block / direct_blks) % dptrs_per_blk;
738 		noffset[n] = 7 + (dptrs_per_blk * 2) +
739 			      offset[n - 2] * (dptrs_per_blk + 1) +
740 			      offset[n - 1];
741 		offset[n] = block % direct_blks;
742 		level = 3;
743 		goto got;
744 	} else {
745 		return -E2BIG;
746 	}
747 got:
748 	return level;
749 }
750 
751 /*
752  * Caller should call f2fs_put_dnode(dn).
753  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
754  * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
755  */
756 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
757 {
758 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
759 	struct page *npage[4];
760 	struct page *parent = NULL;
761 	int offset[4];
762 	unsigned int noffset[4];
763 	nid_t nids[4];
764 	int level, i = 0;
765 	int err = 0;
766 
767 	level = get_node_path(dn->inode, index, offset, noffset);
768 	if (level < 0)
769 		return level;
770 
771 	nids[0] = dn->inode->i_ino;
772 	npage[0] = dn->inode_page;
773 
774 	if (!npage[0]) {
775 		npage[0] = f2fs_get_node_page(sbi, nids[0]);
776 		if (IS_ERR(npage[0]))
777 			return PTR_ERR(npage[0]);
778 	}
779 
780 	/* if inline_data is set, should not report any block indices */
781 	if (f2fs_has_inline_data(dn->inode) && index) {
782 		err = -ENOENT;
783 		f2fs_put_page(npage[0], 1);
784 		goto release_out;
785 	}
786 
787 	parent = npage[0];
788 	if (level != 0)
789 		nids[1] = get_nid(parent, offset[0], true);
790 	dn->inode_page = npage[0];
791 	dn->inode_page_locked = true;
792 
793 	/* get indirect or direct nodes */
794 	for (i = 1; i <= level; i++) {
795 		bool done = false;
796 
797 		if (!nids[i] && mode == ALLOC_NODE) {
798 			/* alloc new node */
799 			if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
800 				err = -ENOSPC;
801 				goto release_pages;
802 			}
803 
804 			dn->nid = nids[i];
805 			npage[i] = f2fs_new_node_page(dn, noffset[i]);
806 			if (IS_ERR(npage[i])) {
807 				f2fs_alloc_nid_failed(sbi, nids[i]);
808 				err = PTR_ERR(npage[i]);
809 				goto release_pages;
810 			}
811 
812 			set_nid(parent, offset[i - 1], nids[i], i == 1);
813 			f2fs_alloc_nid_done(sbi, nids[i]);
814 			done = true;
815 		} else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
816 			npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
817 			if (IS_ERR(npage[i])) {
818 				err = PTR_ERR(npage[i]);
819 				goto release_pages;
820 			}
821 			done = true;
822 		}
823 		if (i == 1) {
824 			dn->inode_page_locked = false;
825 			unlock_page(parent);
826 		} else {
827 			f2fs_put_page(parent, 1);
828 		}
829 
830 		if (!done) {
831 			npage[i] = f2fs_get_node_page(sbi, nids[i]);
832 			if (IS_ERR(npage[i])) {
833 				err = PTR_ERR(npage[i]);
834 				f2fs_put_page(npage[0], 0);
835 				goto release_out;
836 			}
837 		}
838 		if (i < level) {
839 			parent = npage[i];
840 			nids[i + 1] = get_nid(parent, offset[i], false);
841 		}
842 	}
843 	dn->nid = nids[level];
844 	dn->ofs_in_node = offset[level];
845 	dn->node_page = npage[level];
846 	dn->data_blkaddr = f2fs_data_blkaddr(dn);
847 
848 	if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) &&
849 					f2fs_sb_has_readonly(sbi)) {
850 		unsigned int c_len = f2fs_cluster_blocks_are_contiguous(dn);
851 		block_t blkaddr;
852 
853 		if (!c_len)
854 			goto out;
855 
856 		blkaddr = f2fs_data_blkaddr(dn);
857 		if (blkaddr == COMPRESS_ADDR)
858 			blkaddr = data_blkaddr(dn->inode, dn->node_page,
859 						dn->ofs_in_node + 1);
860 
861 		f2fs_update_extent_tree_range_compressed(dn->inode,
862 					index, blkaddr,
863 					F2FS_I(dn->inode)->i_cluster_size,
864 					c_len);
865 	}
866 out:
867 	return 0;
868 
869 release_pages:
870 	f2fs_put_page(parent, 1);
871 	if (i > 1)
872 		f2fs_put_page(npage[0], 0);
873 release_out:
874 	dn->inode_page = NULL;
875 	dn->node_page = NULL;
876 	if (err == -ENOENT) {
877 		dn->cur_level = i;
878 		dn->max_level = level;
879 		dn->ofs_in_node = offset[level];
880 	}
881 	return err;
882 }
883 
884 static int truncate_node(struct dnode_of_data *dn)
885 {
886 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
887 	struct node_info ni;
888 	int err;
889 	pgoff_t index;
890 
891 	err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
892 	if (err)
893 		return err;
894 
895 	/* Deallocate node address */
896 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
897 	dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
898 	set_node_addr(sbi, &ni, NULL_ADDR, false);
899 
900 	if (dn->nid == dn->inode->i_ino) {
901 		f2fs_remove_orphan_inode(sbi, dn->nid);
902 		dec_valid_inode_count(sbi);
903 		f2fs_inode_synced(dn->inode);
904 	}
905 
906 	clear_node_page_dirty(dn->node_page);
907 	set_sbi_flag(sbi, SBI_IS_DIRTY);
908 
909 	index = dn->node_page->index;
910 	f2fs_put_page(dn->node_page, 1);
911 
912 	invalidate_mapping_pages(NODE_MAPPING(sbi),
913 			index, index);
914 
915 	dn->node_page = NULL;
916 	trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
917 
918 	return 0;
919 }
920 
921 static int truncate_dnode(struct dnode_of_data *dn)
922 {
923 	struct page *page;
924 	int err;
925 
926 	if (dn->nid == 0)
927 		return 1;
928 
929 	/* get direct node */
930 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
931 	if (PTR_ERR(page) == -ENOENT)
932 		return 1;
933 	else if (IS_ERR(page))
934 		return PTR_ERR(page);
935 
936 	/* Make dnode_of_data for parameter */
937 	dn->node_page = page;
938 	dn->ofs_in_node = 0;
939 	f2fs_truncate_data_blocks(dn);
940 	err = truncate_node(dn);
941 	if (err)
942 		return err;
943 
944 	return 1;
945 }
946 
947 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
948 						int ofs, int depth)
949 {
950 	struct dnode_of_data rdn = *dn;
951 	struct page *page;
952 	struct f2fs_node *rn;
953 	nid_t child_nid;
954 	unsigned int child_nofs;
955 	int freed = 0;
956 	int i, ret;
957 
958 	if (dn->nid == 0)
959 		return NIDS_PER_BLOCK + 1;
960 
961 	trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
962 
963 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
964 	if (IS_ERR(page)) {
965 		trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
966 		return PTR_ERR(page);
967 	}
968 
969 	f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
970 
971 	rn = F2FS_NODE(page);
972 	if (depth < 3) {
973 		for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
974 			child_nid = le32_to_cpu(rn->in.nid[i]);
975 			if (child_nid == 0)
976 				continue;
977 			rdn.nid = child_nid;
978 			ret = truncate_dnode(&rdn);
979 			if (ret < 0)
980 				goto out_err;
981 			if (set_nid(page, i, 0, false))
982 				dn->node_changed = true;
983 		}
984 	} else {
985 		child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
986 		for (i = ofs; i < NIDS_PER_BLOCK; i++) {
987 			child_nid = le32_to_cpu(rn->in.nid[i]);
988 			if (child_nid == 0) {
989 				child_nofs += NIDS_PER_BLOCK + 1;
990 				continue;
991 			}
992 			rdn.nid = child_nid;
993 			ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
994 			if (ret == (NIDS_PER_BLOCK + 1)) {
995 				if (set_nid(page, i, 0, false))
996 					dn->node_changed = true;
997 				child_nofs += ret;
998 			} else if (ret < 0 && ret != -ENOENT) {
999 				goto out_err;
1000 			}
1001 		}
1002 		freed = child_nofs;
1003 	}
1004 
1005 	if (!ofs) {
1006 		/* remove current indirect node */
1007 		dn->node_page = page;
1008 		ret = truncate_node(dn);
1009 		if (ret)
1010 			goto out_err;
1011 		freed++;
1012 	} else {
1013 		f2fs_put_page(page, 1);
1014 	}
1015 	trace_f2fs_truncate_nodes_exit(dn->inode, freed);
1016 	return freed;
1017 
1018 out_err:
1019 	f2fs_put_page(page, 1);
1020 	trace_f2fs_truncate_nodes_exit(dn->inode, ret);
1021 	return ret;
1022 }
1023 
1024 static int truncate_partial_nodes(struct dnode_of_data *dn,
1025 			struct f2fs_inode *ri, int *offset, int depth)
1026 {
1027 	struct page *pages[2];
1028 	nid_t nid[3];
1029 	nid_t child_nid;
1030 	int err = 0;
1031 	int i;
1032 	int idx = depth - 2;
1033 
1034 	nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1035 	if (!nid[0])
1036 		return 0;
1037 
1038 	/* get indirect nodes in the path */
1039 	for (i = 0; i < idx + 1; i++) {
1040 		/* reference count'll be increased */
1041 		pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
1042 		if (IS_ERR(pages[i])) {
1043 			err = PTR_ERR(pages[i]);
1044 			idx = i - 1;
1045 			goto fail;
1046 		}
1047 		nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1048 	}
1049 
1050 	f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1051 
1052 	/* free direct nodes linked to a partial indirect node */
1053 	for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1054 		child_nid = get_nid(pages[idx], i, false);
1055 		if (!child_nid)
1056 			continue;
1057 		dn->nid = child_nid;
1058 		err = truncate_dnode(dn);
1059 		if (err < 0)
1060 			goto fail;
1061 		if (set_nid(pages[idx], i, 0, false))
1062 			dn->node_changed = true;
1063 	}
1064 
1065 	if (offset[idx + 1] == 0) {
1066 		dn->node_page = pages[idx];
1067 		dn->nid = nid[idx];
1068 		err = truncate_node(dn);
1069 		if (err)
1070 			goto fail;
1071 	} else {
1072 		f2fs_put_page(pages[idx], 1);
1073 	}
1074 	offset[idx]++;
1075 	offset[idx + 1] = 0;
1076 	idx--;
1077 fail:
1078 	for (i = idx; i >= 0; i--)
1079 		f2fs_put_page(pages[i], 1);
1080 
1081 	trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1082 
1083 	return err;
1084 }
1085 
1086 /*
1087  * All the block addresses of data and nodes should be nullified.
1088  */
1089 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1090 {
1091 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1092 	int err = 0, cont = 1;
1093 	int level, offset[4], noffset[4];
1094 	unsigned int nofs = 0;
1095 	struct f2fs_inode *ri;
1096 	struct dnode_of_data dn;
1097 	struct page *page;
1098 
1099 	trace_f2fs_truncate_inode_blocks_enter(inode, from);
1100 
1101 	level = get_node_path(inode, from, offset, noffset);
1102 	if (level < 0) {
1103 		trace_f2fs_truncate_inode_blocks_exit(inode, level);
1104 		return level;
1105 	}
1106 
1107 	page = f2fs_get_node_page(sbi, inode->i_ino);
1108 	if (IS_ERR(page)) {
1109 		trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1110 		return PTR_ERR(page);
1111 	}
1112 
1113 	set_new_dnode(&dn, inode, page, NULL, 0);
1114 	unlock_page(page);
1115 
1116 	ri = F2FS_INODE(page);
1117 	switch (level) {
1118 	case 0:
1119 	case 1:
1120 		nofs = noffset[1];
1121 		break;
1122 	case 2:
1123 		nofs = noffset[1];
1124 		if (!offset[level - 1])
1125 			goto skip_partial;
1126 		err = truncate_partial_nodes(&dn, ri, offset, level);
1127 		if (err < 0 && err != -ENOENT)
1128 			goto fail;
1129 		nofs += 1 + NIDS_PER_BLOCK;
1130 		break;
1131 	case 3:
1132 		nofs = 5 + 2 * NIDS_PER_BLOCK;
1133 		if (!offset[level - 1])
1134 			goto skip_partial;
1135 		err = truncate_partial_nodes(&dn, ri, offset, level);
1136 		if (err < 0 && err != -ENOENT)
1137 			goto fail;
1138 		break;
1139 	default:
1140 		BUG();
1141 	}
1142 
1143 skip_partial:
1144 	while (cont) {
1145 		dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1146 		switch (offset[0]) {
1147 		case NODE_DIR1_BLOCK:
1148 		case NODE_DIR2_BLOCK:
1149 			err = truncate_dnode(&dn);
1150 			break;
1151 
1152 		case NODE_IND1_BLOCK:
1153 		case NODE_IND2_BLOCK:
1154 			err = truncate_nodes(&dn, nofs, offset[1], 2);
1155 			break;
1156 
1157 		case NODE_DIND_BLOCK:
1158 			err = truncate_nodes(&dn, nofs, offset[1], 3);
1159 			cont = 0;
1160 			break;
1161 
1162 		default:
1163 			BUG();
1164 		}
1165 		if (err < 0 && err != -ENOENT)
1166 			goto fail;
1167 		if (offset[1] == 0 &&
1168 				ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1169 			lock_page(page);
1170 			BUG_ON(page->mapping != NODE_MAPPING(sbi));
1171 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1172 			ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1173 			set_page_dirty(page);
1174 			unlock_page(page);
1175 		}
1176 		offset[1] = 0;
1177 		offset[0]++;
1178 		nofs += err;
1179 	}
1180 fail:
1181 	f2fs_put_page(page, 0);
1182 	trace_f2fs_truncate_inode_blocks_exit(inode, err);
1183 	return err > 0 ? 0 : err;
1184 }
1185 
1186 /* caller must lock inode page */
1187 int f2fs_truncate_xattr_node(struct inode *inode)
1188 {
1189 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1190 	nid_t nid = F2FS_I(inode)->i_xattr_nid;
1191 	struct dnode_of_data dn;
1192 	struct page *npage;
1193 	int err;
1194 
1195 	if (!nid)
1196 		return 0;
1197 
1198 	npage = f2fs_get_node_page(sbi, nid);
1199 	if (IS_ERR(npage))
1200 		return PTR_ERR(npage);
1201 
1202 	set_new_dnode(&dn, inode, NULL, npage, nid);
1203 	err = truncate_node(&dn);
1204 	if (err) {
1205 		f2fs_put_page(npage, 1);
1206 		return err;
1207 	}
1208 
1209 	f2fs_i_xnid_write(inode, 0);
1210 
1211 	return 0;
1212 }
1213 
1214 /*
1215  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1216  * f2fs_unlock_op().
1217  */
1218 int f2fs_remove_inode_page(struct inode *inode)
1219 {
1220 	struct dnode_of_data dn;
1221 	int err;
1222 
1223 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1224 	err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1225 	if (err)
1226 		return err;
1227 
1228 	err = f2fs_truncate_xattr_node(inode);
1229 	if (err) {
1230 		f2fs_put_dnode(&dn);
1231 		return err;
1232 	}
1233 
1234 	/* remove potential inline_data blocks */
1235 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1236 				S_ISLNK(inode->i_mode))
1237 		f2fs_truncate_data_blocks_range(&dn, 1);
1238 
1239 	/* 0 is possible, after f2fs_new_inode() has failed */
1240 	if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1241 		f2fs_put_dnode(&dn);
1242 		return -EIO;
1243 	}
1244 
1245 	if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1246 		f2fs_warn(F2FS_I_SB(inode),
1247 			"f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1248 			inode->i_ino, (unsigned long long)inode->i_blocks);
1249 		set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1250 	}
1251 
1252 	/* will put inode & node pages */
1253 	err = truncate_node(&dn);
1254 	if (err) {
1255 		f2fs_put_dnode(&dn);
1256 		return err;
1257 	}
1258 	return 0;
1259 }
1260 
1261 struct page *f2fs_new_inode_page(struct inode *inode)
1262 {
1263 	struct dnode_of_data dn;
1264 
1265 	/* allocate inode page for new inode */
1266 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1267 
1268 	/* caller should f2fs_put_page(page, 1); */
1269 	return f2fs_new_node_page(&dn, 0);
1270 }
1271 
1272 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1273 {
1274 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1275 	struct node_info new_ni;
1276 	struct page *page;
1277 	int err;
1278 
1279 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1280 		return ERR_PTR(-EPERM);
1281 
1282 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1283 	if (!page)
1284 		return ERR_PTR(-ENOMEM);
1285 
1286 	if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1287 		goto fail;
1288 
1289 #ifdef CONFIG_F2FS_CHECK_FS
1290 	err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false);
1291 	if (err) {
1292 		dec_valid_node_count(sbi, dn->inode, !ofs);
1293 		goto fail;
1294 	}
1295 	f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1296 #endif
1297 	new_ni.nid = dn->nid;
1298 	new_ni.ino = dn->inode->i_ino;
1299 	new_ni.blk_addr = NULL_ADDR;
1300 	new_ni.flag = 0;
1301 	new_ni.version = 0;
1302 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1303 
1304 	f2fs_wait_on_page_writeback(page, NODE, true, true);
1305 	fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1306 	set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1307 	if (!PageUptodate(page))
1308 		SetPageUptodate(page);
1309 	if (set_page_dirty(page))
1310 		dn->node_changed = true;
1311 
1312 	if (f2fs_has_xattr_block(ofs))
1313 		f2fs_i_xnid_write(dn->inode, dn->nid);
1314 
1315 	if (ofs == 0)
1316 		inc_valid_inode_count(sbi);
1317 	return page;
1318 
1319 fail:
1320 	clear_node_page_dirty(page);
1321 	f2fs_put_page(page, 1);
1322 	return ERR_PTR(err);
1323 }
1324 
1325 /*
1326  * Caller should do after getting the following values.
1327  * 0: f2fs_put_page(page, 0)
1328  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1329  */
1330 static int read_node_page(struct page *page, int op_flags)
1331 {
1332 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1333 	struct node_info ni;
1334 	struct f2fs_io_info fio = {
1335 		.sbi = sbi,
1336 		.type = NODE,
1337 		.op = REQ_OP_READ,
1338 		.op_flags = op_flags,
1339 		.page = page,
1340 		.encrypted_page = NULL,
1341 	};
1342 	int err;
1343 
1344 	if (PageUptodate(page)) {
1345 		if (!f2fs_inode_chksum_verify(sbi, page)) {
1346 			ClearPageUptodate(page);
1347 			return -EFSBADCRC;
1348 		}
1349 		return LOCKED_PAGE;
1350 	}
1351 
1352 	err = f2fs_get_node_info(sbi, page->index, &ni, false);
1353 	if (err)
1354 		return err;
1355 
1356 	/* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1357 	if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR) ||
1358 			is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1359 		ClearPageUptodate(page);
1360 		return -ENOENT;
1361 	}
1362 
1363 	fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1364 
1365 	err = f2fs_submit_page_bio(&fio);
1366 
1367 	if (!err)
1368 		f2fs_update_iostat(sbi, FS_NODE_READ_IO, F2FS_BLKSIZE);
1369 
1370 	return err;
1371 }
1372 
1373 /*
1374  * Readahead a node page
1375  */
1376 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1377 {
1378 	struct page *apage;
1379 	int err;
1380 
1381 	if (!nid)
1382 		return;
1383 	if (f2fs_check_nid_range(sbi, nid))
1384 		return;
1385 
1386 	apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1387 	if (apage)
1388 		return;
1389 
1390 	apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1391 	if (!apage)
1392 		return;
1393 
1394 	err = read_node_page(apage, REQ_RAHEAD);
1395 	f2fs_put_page(apage, err ? 1 : 0);
1396 }
1397 
1398 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1399 					struct page *parent, int start)
1400 {
1401 	struct page *page;
1402 	int err;
1403 
1404 	if (!nid)
1405 		return ERR_PTR(-ENOENT);
1406 	if (f2fs_check_nid_range(sbi, nid))
1407 		return ERR_PTR(-EINVAL);
1408 repeat:
1409 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1410 	if (!page)
1411 		return ERR_PTR(-ENOMEM);
1412 
1413 	err = read_node_page(page, 0);
1414 	if (err < 0) {
1415 		goto out_put_err;
1416 	} else if (err == LOCKED_PAGE) {
1417 		err = 0;
1418 		goto page_hit;
1419 	}
1420 
1421 	if (parent)
1422 		f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1423 
1424 	lock_page(page);
1425 
1426 	if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1427 		f2fs_put_page(page, 1);
1428 		goto repeat;
1429 	}
1430 
1431 	if (unlikely(!PageUptodate(page))) {
1432 		err = -EIO;
1433 		goto out_err;
1434 	}
1435 
1436 	if (!f2fs_inode_chksum_verify(sbi, page)) {
1437 		err = -EFSBADCRC;
1438 		goto out_err;
1439 	}
1440 page_hit:
1441 	if (likely(nid == nid_of_node(page)))
1442 		return page;
1443 
1444 	f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1445 			  nid, nid_of_node(page), ino_of_node(page),
1446 			  ofs_of_node(page), cpver_of_node(page),
1447 			  next_blkaddr_of_node(page));
1448 	set_sbi_flag(sbi, SBI_NEED_FSCK);
1449 	err = -EINVAL;
1450 out_err:
1451 	ClearPageUptodate(page);
1452 out_put_err:
1453 	/* ENOENT comes from read_node_page which is not an error. */
1454 	if (err != -ENOENT)
1455 		f2fs_handle_page_eio(sbi, page->index, NODE);
1456 	f2fs_put_page(page, 1);
1457 	return ERR_PTR(err);
1458 }
1459 
1460 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1461 {
1462 	return __get_node_page(sbi, nid, NULL, 0);
1463 }
1464 
1465 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1466 {
1467 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1468 	nid_t nid = get_nid(parent, start, false);
1469 
1470 	return __get_node_page(sbi, nid, parent, start);
1471 }
1472 
1473 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1474 {
1475 	struct inode *inode;
1476 	struct page *page;
1477 	int ret;
1478 
1479 	/* should flush inline_data before evict_inode */
1480 	inode = ilookup(sbi->sb, ino);
1481 	if (!inode)
1482 		return;
1483 
1484 	page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1485 					FGP_LOCK|FGP_NOWAIT, 0);
1486 	if (!page)
1487 		goto iput_out;
1488 
1489 	if (!PageUptodate(page))
1490 		goto page_out;
1491 
1492 	if (!PageDirty(page))
1493 		goto page_out;
1494 
1495 	if (!clear_page_dirty_for_io(page))
1496 		goto page_out;
1497 
1498 	ret = f2fs_write_inline_data(inode, page);
1499 	inode_dec_dirty_pages(inode);
1500 	f2fs_remove_dirty_inode(inode);
1501 	if (ret)
1502 		set_page_dirty(page);
1503 page_out:
1504 	f2fs_put_page(page, 1);
1505 iput_out:
1506 	iput(inode);
1507 }
1508 
1509 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1510 {
1511 	pgoff_t index;
1512 	struct pagevec pvec;
1513 	struct page *last_page = NULL;
1514 	int nr_pages;
1515 
1516 	pagevec_init(&pvec);
1517 	index = 0;
1518 
1519 	while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1520 				PAGECACHE_TAG_DIRTY))) {
1521 		int i;
1522 
1523 		for (i = 0; i < nr_pages; i++) {
1524 			struct page *page = pvec.pages[i];
1525 
1526 			if (unlikely(f2fs_cp_error(sbi))) {
1527 				f2fs_put_page(last_page, 0);
1528 				pagevec_release(&pvec);
1529 				return ERR_PTR(-EIO);
1530 			}
1531 
1532 			if (!IS_DNODE(page) || !is_cold_node(page))
1533 				continue;
1534 			if (ino_of_node(page) != ino)
1535 				continue;
1536 
1537 			lock_page(page);
1538 
1539 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1540 continue_unlock:
1541 				unlock_page(page);
1542 				continue;
1543 			}
1544 			if (ino_of_node(page) != ino)
1545 				goto continue_unlock;
1546 
1547 			if (!PageDirty(page)) {
1548 				/* someone wrote it for us */
1549 				goto continue_unlock;
1550 			}
1551 
1552 			if (last_page)
1553 				f2fs_put_page(last_page, 0);
1554 
1555 			get_page(page);
1556 			last_page = page;
1557 			unlock_page(page);
1558 		}
1559 		pagevec_release(&pvec);
1560 		cond_resched();
1561 	}
1562 	return last_page;
1563 }
1564 
1565 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1566 				struct writeback_control *wbc, bool do_balance,
1567 				enum iostat_type io_type, unsigned int *seq_id)
1568 {
1569 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1570 	nid_t nid;
1571 	struct node_info ni;
1572 	struct f2fs_io_info fio = {
1573 		.sbi = sbi,
1574 		.ino = ino_of_node(page),
1575 		.type = NODE,
1576 		.op = REQ_OP_WRITE,
1577 		.op_flags = wbc_to_write_flags(wbc),
1578 		.page = page,
1579 		.encrypted_page = NULL,
1580 		.submitted = false,
1581 		.io_type = io_type,
1582 		.io_wbc = wbc,
1583 	};
1584 	unsigned int seq;
1585 
1586 	trace_f2fs_writepage(page, NODE);
1587 
1588 	if (unlikely(f2fs_cp_error(sbi))) {
1589 		ClearPageUptodate(page);
1590 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1591 		unlock_page(page);
1592 		return 0;
1593 	}
1594 
1595 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1596 		goto redirty_out;
1597 
1598 	if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1599 			wbc->sync_mode == WB_SYNC_NONE &&
1600 			IS_DNODE(page) && is_cold_node(page))
1601 		goto redirty_out;
1602 
1603 	/* get old block addr of this node page */
1604 	nid = nid_of_node(page);
1605 	f2fs_bug_on(sbi, page->index != nid);
1606 
1607 	if (f2fs_get_node_info(sbi, nid, &ni, !do_balance))
1608 		goto redirty_out;
1609 
1610 	if (wbc->for_reclaim) {
1611 		if (!f2fs_down_read_trylock(&sbi->node_write))
1612 			goto redirty_out;
1613 	} else {
1614 		f2fs_down_read(&sbi->node_write);
1615 	}
1616 
1617 	/* This page is already truncated */
1618 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1619 		ClearPageUptodate(page);
1620 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1621 		f2fs_up_read(&sbi->node_write);
1622 		unlock_page(page);
1623 		return 0;
1624 	}
1625 
1626 	if (__is_valid_data_blkaddr(ni.blk_addr) &&
1627 		!f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1628 					DATA_GENERIC_ENHANCE)) {
1629 		f2fs_up_read(&sbi->node_write);
1630 		goto redirty_out;
1631 	}
1632 
1633 	if (atomic && !test_opt(sbi, NOBARRIER) && !f2fs_sb_has_blkzoned(sbi))
1634 		fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1635 
1636 	/* should add to global list before clearing PAGECACHE status */
1637 	if (f2fs_in_warm_node_list(sbi, page)) {
1638 		seq = f2fs_add_fsync_node_entry(sbi, page);
1639 		if (seq_id)
1640 			*seq_id = seq;
1641 	}
1642 
1643 	set_page_writeback(page);
1644 	ClearPageError(page);
1645 
1646 	fio.old_blkaddr = ni.blk_addr;
1647 	f2fs_do_write_node_page(nid, &fio);
1648 	set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1649 	dec_page_count(sbi, F2FS_DIRTY_NODES);
1650 	f2fs_up_read(&sbi->node_write);
1651 
1652 	if (wbc->for_reclaim) {
1653 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1654 		submitted = NULL;
1655 	}
1656 
1657 	unlock_page(page);
1658 
1659 	if (unlikely(f2fs_cp_error(sbi))) {
1660 		f2fs_submit_merged_write(sbi, NODE);
1661 		submitted = NULL;
1662 	}
1663 	if (submitted)
1664 		*submitted = fio.submitted;
1665 
1666 	if (do_balance)
1667 		f2fs_balance_fs(sbi, false);
1668 	return 0;
1669 
1670 redirty_out:
1671 	redirty_page_for_writepage(wbc, page);
1672 	return AOP_WRITEPAGE_ACTIVATE;
1673 }
1674 
1675 int f2fs_move_node_page(struct page *node_page, int gc_type)
1676 {
1677 	int err = 0;
1678 
1679 	if (gc_type == FG_GC) {
1680 		struct writeback_control wbc = {
1681 			.sync_mode = WB_SYNC_ALL,
1682 			.nr_to_write = 1,
1683 			.for_reclaim = 0,
1684 		};
1685 
1686 		f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1687 
1688 		set_page_dirty(node_page);
1689 
1690 		if (!clear_page_dirty_for_io(node_page)) {
1691 			err = -EAGAIN;
1692 			goto out_page;
1693 		}
1694 
1695 		if (__write_node_page(node_page, false, NULL,
1696 					&wbc, false, FS_GC_NODE_IO, NULL)) {
1697 			err = -EAGAIN;
1698 			unlock_page(node_page);
1699 		}
1700 		goto release_page;
1701 	} else {
1702 		/* set page dirty and write it */
1703 		if (!PageWriteback(node_page))
1704 			set_page_dirty(node_page);
1705 	}
1706 out_page:
1707 	unlock_page(node_page);
1708 release_page:
1709 	f2fs_put_page(node_page, 0);
1710 	return err;
1711 }
1712 
1713 static int f2fs_write_node_page(struct page *page,
1714 				struct writeback_control *wbc)
1715 {
1716 	return __write_node_page(page, false, NULL, wbc, false,
1717 						FS_NODE_IO, NULL);
1718 }
1719 
1720 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1721 			struct writeback_control *wbc, bool atomic,
1722 			unsigned int *seq_id)
1723 {
1724 	pgoff_t index;
1725 	struct pagevec pvec;
1726 	int ret = 0;
1727 	struct page *last_page = NULL;
1728 	bool marked = false;
1729 	nid_t ino = inode->i_ino;
1730 	int nr_pages;
1731 	int nwritten = 0;
1732 
1733 	if (atomic) {
1734 		last_page = last_fsync_dnode(sbi, ino);
1735 		if (IS_ERR_OR_NULL(last_page))
1736 			return PTR_ERR_OR_ZERO(last_page);
1737 	}
1738 retry:
1739 	pagevec_init(&pvec);
1740 	index = 0;
1741 
1742 	while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1743 				PAGECACHE_TAG_DIRTY))) {
1744 		int i;
1745 
1746 		for (i = 0; i < nr_pages; i++) {
1747 			struct page *page = pvec.pages[i];
1748 			bool submitted = false;
1749 
1750 			if (unlikely(f2fs_cp_error(sbi))) {
1751 				f2fs_put_page(last_page, 0);
1752 				pagevec_release(&pvec);
1753 				ret = -EIO;
1754 				goto out;
1755 			}
1756 
1757 			if (!IS_DNODE(page) || !is_cold_node(page))
1758 				continue;
1759 			if (ino_of_node(page) != ino)
1760 				continue;
1761 
1762 			lock_page(page);
1763 
1764 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1765 continue_unlock:
1766 				unlock_page(page);
1767 				continue;
1768 			}
1769 			if (ino_of_node(page) != ino)
1770 				goto continue_unlock;
1771 
1772 			if (!PageDirty(page) && page != last_page) {
1773 				/* someone wrote it for us */
1774 				goto continue_unlock;
1775 			}
1776 
1777 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1778 
1779 			set_fsync_mark(page, 0);
1780 			set_dentry_mark(page, 0);
1781 
1782 			if (!atomic || page == last_page) {
1783 				set_fsync_mark(page, 1);
1784 				percpu_counter_inc(&sbi->rf_node_block_count);
1785 				if (IS_INODE(page)) {
1786 					if (is_inode_flag_set(inode,
1787 								FI_DIRTY_INODE))
1788 						f2fs_update_inode(inode, page);
1789 					set_dentry_mark(page,
1790 						f2fs_need_dentry_mark(sbi, ino));
1791 				}
1792 				/* may be written by other thread */
1793 				if (!PageDirty(page))
1794 					set_page_dirty(page);
1795 			}
1796 
1797 			if (!clear_page_dirty_for_io(page))
1798 				goto continue_unlock;
1799 
1800 			ret = __write_node_page(page, atomic &&
1801 						page == last_page,
1802 						&submitted, wbc, true,
1803 						FS_NODE_IO, seq_id);
1804 			if (ret) {
1805 				unlock_page(page);
1806 				f2fs_put_page(last_page, 0);
1807 				break;
1808 			} else if (submitted) {
1809 				nwritten++;
1810 			}
1811 
1812 			if (page == last_page) {
1813 				f2fs_put_page(page, 0);
1814 				marked = true;
1815 				break;
1816 			}
1817 		}
1818 		pagevec_release(&pvec);
1819 		cond_resched();
1820 
1821 		if (ret || marked)
1822 			break;
1823 	}
1824 	if (!ret && atomic && !marked) {
1825 		f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1826 			   ino, last_page->index);
1827 		lock_page(last_page);
1828 		f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1829 		set_page_dirty(last_page);
1830 		unlock_page(last_page);
1831 		goto retry;
1832 	}
1833 out:
1834 	if (nwritten)
1835 		f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1836 	return ret ? -EIO : 0;
1837 }
1838 
1839 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1840 {
1841 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1842 	bool clean;
1843 
1844 	if (inode->i_ino != ino)
1845 		return 0;
1846 
1847 	if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1848 		return 0;
1849 
1850 	spin_lock(&sbi->inode_lock[DIRTY_META]);
1851 	clean = list_empty(&F2FS_I(inode)->gdirty_list);
1852 	spin_unlock(&sbi->inode_lock[DIRTY_META]);
1853 
1854 	if (clean)
1855 		return 0;
1856 
1857 	inode = igrab(inode);
1858 	if (!inode)
1859 		return 0;
1860 	return 1;
1861 }
1862 
1863 static bool flush_dirty_inode(struct page *page)
1864 {
1865 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1866 	struct inode *inode;
1867 	nid_t ino = ino_of_node(page);
1868 
1869 	inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1870 	if (!inode)
1871 		return false;
1872 
1873 	f2fs_update_inode(inode, page);
1874 	unlock_page(page);
1875 
1876 	iput(inode);
1877 	return true;
1878 }
1879 
1880 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1881 {
1882 	pgoff_t index = 0;
1883 	struct pagevec pvec;
1884 	int nr_pages;
1885 
1886 	pagevec_init(&pvec);
1887 
1888 	while ((nr_pages = pagevec_lookup_tag(&pvec,
1889 			NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1890 		int i;
1891 
1892 		for (i = 0; i < nr_pages; i++) {
1893 			struct page *page = pvec.pages[i];
1894 
1895 			if (!IS_DNODE(page))
1896 				continue;
1897 
1898 			lock_page(page);
1899 
1900 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1901 continue_unlock:
1902 				unlock_page(page);
1903 				continue;
1904 			}
1905 
1906 			if (!PageDirty(page)) {
1907 				/* someone wrote it for us */
1908 				goto continue_unlock;
1909 			}
1910 
1911 			/* flush inline_data, if it's async context. */
1912 			if (page_private_inline(page)) {
1913 				clear_page_private_inline(page);
1914 				unlock_page(page);
1915 				flush_inline_data(sbi, ino_of_node(page));
1916 				continue;
1917 			}
1918 			unlock_page(page);
1919 		}
1920 		pagevec_release(&pvec);
1921 		cond_resched();
1922 	}
1923 }
1924 
1925 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1926 				struct writeback_control *wbc,
1927 				bool do_balance, enum iostat_type io_type)
1928 {
1929 	pgoff_t index;
1930 	struct pagevec pvec;
1931 	int step = 0;
1932 	int nwritten = 0;
1933 	int ret = 0;
1934 	int nr_pages, done = 0;
1935 
1936 	pagevec_init(&pvec);
1937 
1938 next_step:
1939 	index = 0;
1940 
1941 	while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1942 			NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1943 		int i;
1944 
1945 		for (i = 0; i < nr_pages; i++) {
1946 			struct page *page = pvec.pages[i];
1947 			bool submitted = false;
1948 			bool may_dirty = true;
1949 
1950 			/* give a priority to WB_SYNC threads */
1951 			if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1952 					wbc->sync_mode == WB_SYNC_NONE) {
1953 				done = 1;
1954 				break;
1955 			}
1956 
1957 			/*
1958 			 * flushing sequence with step:
1959 			 * 0. indirect nodes
1960 			 * 1. dentry dnodes
1961 			 * 2. file dnodes
1962 			 */
1963 			if (step == 0 && IS_DNODE(page))
1964 				continue;
1965 			if (step == 1 && (!IS_DNODE(page) ||
1966 						is_cold_node(page)))
1967 				continue;
1968 			if (step == 2 && (!IS_DNODE(page) ||
1969 						!is_cold_node(page)))
1970 				continue;
1971 lock_node:
1972 			if (wbc->sync_mode == WB_SYNC_ALL)
1973 				lock_page(page);
1974 			else if (!trylock_page(page))
1975 				continue;
1976 
1977 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1978 continue_unlock:
1979 				unlock_page(page);
1980 				continue;
1981 			}
1982 
1983 			if (!PageDirty(page)) {
1984 				/* someone wrote it for us */
1985 				goto continue_unlock;
1986 			}
1987 
1988 			/* flush inline_data/inode, if it's async context. */
1989 			if (!do_balance)
1990 				goto write_node;
1991 
1992 			/* flush inline_data */
1993 			if (page_private_inline(page)) {
1994 				clear_page_private_inline(page);
1995 				unlock_page(page);
1996 				flush_inline_data(sbi, ino_of_node(page));
1997 				goto lock_node;
1998 			}
1999 
2000 			/* flush dirty inode */
2001 			if (IS_INODE(page) && may_dirty) {
2002 				may_dirty = false;
2003 				if (flush_dirty_inode(page))
2004 					goto lock_node;
2005 			}
2006 write_node:
2007 			f2fs_wait_on_page_writeback(page, NODE, true, true);
2008 
2009 			if (!clear_page_dirty_for_io(page))
2010 				goto continue_unlock;
2011 
2012 			set_fsync_mark(page, 0);
2013 			set_dentry_mark(page, 0);
2014 
2015 			ret = __write_node_page(page, false, &submitted,
2016 						wbc, do_balance, io_type, NULL);
2017 			if (ret)
2018 				unlock_page(page);
2019 			else if (submitted)
2020 				nwritten++;
2021 
2022 			if (--wbc->nr_to_write == 0)
2023 				break;
2024 		}
2025 		pagevec_release(&pvec);
2026 		cond_resched();
2027 
2028 		if (wbc->nr_to_write == 0) {
2029 			step = 2;
2030 			break;
2031 		}
2032 	}
2033 
2034 	if (step < 2) {
2035 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2036 				wbc->sync_mode == WB_SYNC_NONE && step == 1)
2037 			goto out;
2038 		step++;
2039 		goto next_step;
2040 	}
2041 out:
2042 	if (nwritten)
2043 		f2fs_submit_merged_write(sbi, NODE);
2044 
2045 	if (unlikely(f2fs_cp_error(sbi)))
2046 		return -EIO;
2047 	return ret;
2048 }
2049 
2050 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2051 						unsigned int seq_id)
2052 {
2053 	struct fsync_node_entry *fn;
2054 	struct page *page;
2055 	struct list_head *head = &sbi->fsync_node_list;
2056 	unsigned long flags;
2057 	unsigned int cur_seq_id = 0;
2058 	int ret2, ret = 0;
2059 
2060 	while (seq_id && cur_seq_id < seq_id) {
2061 		spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2062 		if (list_empty(head)) {
2063 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2064 			break;
2065 		}
2066 		fn = list_first_entry(head, struct fsync_node_entry, list);
2067 		if (fn->seq_id > seq_id) {
2068 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2069 			break;
2070 		}
2071 		cur_seq_id = fn->seq_id;
2072 		page = fn->page;
2073 		get_page(page);
2074 		spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2075 
2076 		f2fs_wait_on_page_writeback(page, NODE, true, false);
2077 		if (TestClearPageError(page))
2078 			ret = -EIO;
2079 
2080 		put_page(page);
2081 
2082 		if (ret)
2083 			break;
2084 	}
2085 
2086 	ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2087 	if (!ret)
2088 		ret = ret2;
2089 
2090 	return ret;
2091 }
2092 
2093 static int f2fs_write_node_pages(struct address_space *mapping,
2094 			    struct writeback_control *wbc)
2095 {
2096 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2097 	struct blk_plug plug;
2098 	long diff;
2099 
2100 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2101 		goto skip_write;
2102 
2103 	/* balancing f2fs's metadata in background */
2104 	f2fs_balance_fs_bg(sbi, true);
2105 
2106 	/* collect a number of dirty node pages and write together */
2107 	if (wbc->sync_mode != WB_SYNC_ALL &&
2108 			get_pages(sbi, F2FS_DIRTY_NODES) <
2109 					nr_pages_to_skip(sbi, NODE))
2110 		goto skip_write;
2111 
2112 	if (wbc->sync_mode == WB_SYNC_ALL)
2113 		atomic_inc(&sbi->wb_sync_req[NODE]);
2114 	else if (atomic_read(&sbi->wb_sync_req[NODE])) {
2115 		/* to avoid potential deadlock */
2116 		if (current->plug)
2117 			blk_finish_plug(current->plug);
2118 		goto skip_write;
2119 	}
2120 
2121 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2122 
2123 	diff = nr_pages_to_write(sbi, NODE, wbc);
2124 	blk_start_plug(&plug);
2125 	f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2126 	blk_finish_plug(&plug);
2127 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2128 
2129 	if (wbc->sync_mode == WB_SYNC_ALL)
2130 		atomic_dec(&sbi->wb_sync_req[NODE]);
2131 	return 0;
2132 
2133 skip_write:
2134 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2135 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2136 	return 0;
2137 }
2138 
2139 static bool f2fs_dirty_node_folio(struct address_space *mapping,
2140 		struct folio *folio)
2141 {
2142 	trace_f2fs_set_page_dirty(&folio->page, NODE);
2143 
2144 	if (!folio_test_uptodate(folio))
2145 		folio_mark_uptodate(folio);
2146 #ifdef CONFIG_F2FS_CHECK_FS
2147 	if (IS_INODE(&folio->page))
2148 		f2fs_inode_chksum_set(F2FS_M_SB(mapping), &folio->page);
2149 #endif
2150 	if (!folio_test_dirty(folio)) {
2151 		filemap_dirty_folio(mapping, folio);
2152 		inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
2153 		set_page_private_reference(&folio->page);
2154 		return true;
2155 	}
2156 	return false;
2157 }
2158 
2159 /*
2160  * Structure of the f2fs node operations
2161  */
2162 const struct address_space_operations f2fs_node_aops = {
2163 	.writepage	= f2fs_write_node_page,
2164 	.writepages	= f2fs_write_node_pages,
2165 	.dirty_folio	= f2fs_dirty_node_folio,
2166 	.invalidate_folio = f2fs_invalidate_folio,
2167 	.release_folio	= f2fs_release_folio,
2168 #ifdef CONFIG_MIGRATION
2169 	.migratepage	= f2fs_migrate_page,
2170 #endif
2171 };
2172 
2173 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2174 						nid_t n)
2175 {
2176 	return radix_tree_lookup(&nm_i->free_nid_root, n);
2177 }
2178 
2179 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2180 				struct free_nid *i)
2181 {
2182 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2183 	int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2184 
2185 	if (err)
2186 		return err;
2187 
2188 	nm_i->nid_cnt[FREE_NID]++;
2189 	list_add_tail(&i->list, &nm_i->free_nid_list);
2190 	return 0;
2191 }
2192 
2193 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2194 			struct free_nid *i, enum nid_state state)
2195 {
2196 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2197 
2198 	f2fs_bug_on(sbi, state != i->state);
2199 	nm_i->nid_cnt[state]--;
2200 	if (state == FREE_NID)
2201 		list_del(&i->list);
2202 	radix_tree_delete(&nm_i->free_nid_root, i->nid);
2203 }
2204 
2205 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2206 			enum nid_state org_state, enum nid_state dst_state)
2207 {
2208 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2209 
2210 	f2fs_bug_on(sbi, org_state != i->state);
2211 	i->state = dst_state;
2212 	nm_i->nid_cnt[org_state]--;
2213 	nm_i->nid_cnt[dst_state]++;
2214 
2215 	switch (dst_state) {
2216 	case PREALLOC_NID:
2217 		list_del(&i->list);
2218 		break;
2219 	case FREE_NID:
2220 		list_add_tail(&i->list, &nm_i->free_nid_list);
2221 		break;
2222 	default:
2223 		BUG_ON(1);
2224 	}
2225 }
2226 
2227 bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi)
2228 {
2229 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2230 	unsigned int i;
2231 	bool ret = true;
2232 
2233 	f2fs_down_read(&nm_i->nat_tree_lock);
2234 	for (i = 0; i < nm_i->nat_blocks; i++) {
2235 		if (!test_bit_le(i, nm_i->nat_block_bitmap)) {
2236 			ret = false;
2237 			break;
2238 		}
2239 	}
2240 	f2fs_up_read(&nm_i->nat_tree_lock);
2241 
2242 	return ret;
2243 }
2244 
2245 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2246 							bool set, bool build)
2247 {
2248 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2249 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2250 	unsigned int nid_ofs = nid - START_NID(nid);
2251 
2252 	if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2253 		return;
2254 
2255 	if (set) {
2256 		if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2257 			return;
2258 		__set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2259 		nm_i->free_nid_count[nat_ofs]++;
2260 	} else {
2261 		if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2262 			return;
2263 		__clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2264 		if (!build)
2265 			nm_i->free_nid_count[nat_ofs]--;
2266 	}
2267 }
2268 
2269 /* return if the nid is recognized as free */
2270 static bool add_free_nid(struct f2fs_sb_info *sbi,
2271 				nid_t nid, bool build, bool update)
2272 {
2273 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2274 	struct free_nid *i, *e;
2275 	struct nat_entry *ne;
2276 	int err = -EINVAL;
2277 	bool ret = false;
2278 
2279 	/* 0 nid should not be used */
2280 	if (unlikely(nid == 0))
2281 		return false;
2282 
2283 	if (unlikely(f2fs_check_nid_range(sbi, nid)))
2284 		return false;
2285 
2286 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL);
2287 	i->nid = nid;
2288 	i->state = FREE_NID;
2289 
2290 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2291 
2292 	spin_lock(&nm_i->nid_list_lock);
2293 
2294 	if (build) {
2295 		/*
2296 		 *   Thread A             Thread B
2297 		 *  - f2fs_create
2298 		 *   - f2fs_new_inode
2299 		 *    - f2fs_alloc_nid
2300 		 *     - __insert_nid_to_list(PREALLOC_NID)
2301 		 *                     - f2fs_balance_fs_bg
2302 		 *                      - f2fs_build_free_nids
2303 		 *                       - __f2fs_build_free_nids
2304 		 *                        - scan_nat_page
2305 		 *                         - add_free_nid
2306 		 *                          - __lookup_nat_cache
2307 		 *  - f2fs_add_link
2308 		 *   - f2fs_init_inode_metadata
2309 		 *    - f2fs_new_inode_page
2310 		 *     - f2fs_new_node_page
2311 		 *      - set_node_addr
2312 		 *  - f2fs_alloc_nid_done
2313 		 *   - __remove_nid_from_list(PREALLOC_NID)
2314 		 *                         - __insert_nid_to_list(FREE_NID)
2315 		 */
2316 		ne = __lookup_nat_cache(nm_i, nid);
2317 		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2318 				nat_get_blkaddr(ne) != NULL_ADDR))
2319 			goto err_out;
2320 
2321 		e = __lookup_free_nid_list(nm_i, nid);
2322 		if (e) {
2323 			if (e->state == FREE_NID)
2324 				ret = true;
2325 			goto err_out;
2326 		}
2327 	}
2328 	ret = true;
2329 	err = __insert_free_nid(sbi, i);
2330 err_out:
2331 	if (update) {
2332 		update_free_nid_bitmap(sbi, nid, ret, build);
2333 		if (!build)
2334 			nm_i->available_nids++;
2335 	}
2336 	spin_unlock(&nm_i->nid_list_lock);
2337 	radix_tree_preload_end();
2338 
2339 	if (err)
2340 		kmem_cache_free(free_nid_slab, i);
2341 	return ret;
2342 }
2343 
2344 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2345 {
2346 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2347 	struct free_nid *i;
2348 	bool need_free = false;
2349 
2350 	spin_lock(&nm_i->nid_list_lock);
2351 	i = __lookup_free_nid_list(nm_i, nid);
2352 	if (i && i->state == FREE_NID) {
2353 		__remove_free_nid(sbi, i, FREE_NID);
2354 		need_free = true;
2355 	}
2356 	spin_unlock(&nm_i->nid_list_lock);
2357 
2358 	if (need_free)
2359 		kmem_cache_free(free_nid_slab, i);
2360 }
2361 
2362 static int scan_nat_page(struct f2fs_sb_info *sbi,
2363 			struct page *nat_page, nid_t start_nid)
2364 {
2365 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2366 	struct f2fs_nat_block *nat_blk = page_address(nat_page);
2367 	block_t blk_addr;
2368 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2369 	int i;
2370 
2371 	__set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2372 
2373 	i = start_nid % NAT_ENTRY_PER_BLOCK;
2374 
2375 	for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2376 		if (unlikely(start_nid >= nm_i->max_nid))
2377 			break;
2378 
2379 		blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2380 
2381 		if (blk_addr == NEW_ADDR)
2382 			return -EINVAL;
2383 
2384 		if (blk_addr == NULL_ADDR) {
2385 			add_free_nid(sbi, start_nid, true, true);
2386 		} else {
2387 			spin_lock(&NM_I(sbi)->nid_list_lock);
2388 			update_free_nid_bitmap(sbi, start_nid, false, true);
2389 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2390 		}
2391 	}
2392 
2393 	return 0;
2394 }
2395 
2396 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2397 {
2398 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2399 	struct f2fs_journal *journal = curseg->journal;
2400 	int i;
2401 
2402 	down_read(&curseg->journal_rwsem);
2403 	for (i = 0; i < nats_in_cursum(journal); i++) {
2404 		block_t addr;
2405 		nid_t nid;
2406 
2407 		addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2408 		nid = le32_to_cpu(nid_in_journal(journal, i));
2409 		if (addr == NULL_ADDR)
2410 			add_free_nid(sbi, nid, true, false);
2411 		else
2412 			remove_free_nid(sbi, nid);
2413 	}
2414 	up_read(&curseg->journal_rwsem);
2415 }
2416 
2417 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2418 {
2419 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2420 	unsigned int i, idx;
2421 	nid_t nid;
2422 
2423 	f2fs_down_read(&nm_i->nat_tree_lock);
2424 
2425 	for (i = 0; i < nm_i->nat_blocks; i++) {
2426 		if (!test_bit_le(i, nm_i->nat_block_bitmap))
2427 			continue;
2428 		if (!nm_i->free_nid_count[i])
2429 			continue;
2430 		for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2431 			idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2432 						NAT_ENTRY_PER_BLOCK, idx);
2433 			if (idx >= NAT_ENTRY_PER_BLOCK)
2434 				break;
2435 
2436 			nid = i * NAT_ENTRY_PER_BLOCK + idx;
2437 			add_free_nid(sbi, nid, true, false);
2438 
2439 			if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2440 				goto out;
2441 		}
2442 	}
2443 out:
2444 	scan_curseg_cache(sbi);
2445 
2446 	f2fs_up_read(&nm_i->nat_tree_lock);
2447 }
2448 
2449 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2450 						bool sync, bool mount)
2451 {
2452 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2453 	int i = 0, ret;
2454 	nid_t nid = nm_i->next_scan_nid;
2455 
2456 	if (unlikely(nid >= nm_i->max_nid))
2457 		nid = 0;
2458 
2459 	if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2460 		nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2461 
2462 	/* Enough entries */
2463 	if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2464 		return 0;
2465 
2466 	if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2467 		return 0;
2468 
2469 	if (!mount) {
2470 		/* try to find free nids in free_nid_bitmap */
2471 		scan_free_nid_bits(sbi);
2472 
2473 		if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2474 			return 0;
2475 	}
2476 
2477 	/* readahead nat pages to be scanned */
2478 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2479 							META_NAT, true);
2480 
2481 	f2fs_down_read(&nm_i->nat_tree_lock);
2482 
2483 	while (1) {
2484 		if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2485 						nm_i->nat_block_bitmap)) {
2486 			struct page *page = get_current_nat_page(sbi, nid);
2487 
2488 			if (IS_ERR(page)) {
2489 				ret = PTR_ERR(page);
2490 			} else {
2491 				ret = scan_nat_page(sbi, page, nid);
2492 				f2fs_put_page(page, 1);
2493 			}
2494 
2495 			if (ret) {
2496 				f2fs_up_read(&nm_i->nat_tree_lock);
2497 				f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2498 				return ret;
2499 			}
2500 		}
2501 
2502 		nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2503 		if (unlikely(nid >= nm_i->max_nid))
2504 			nid = 0;
2505 
2506 		if (++i >= FREE_NID_PAGES)
2507 			break;
2508 	}
2509 
2510 	/* go to the next free nat pages to find free nids abundantly */
2511 	nm_i->next_scan_nid = nid;
2512 
2513 	/* find free nids from current sum_pages */
2514 	scan_curseg_cache(sbi);
2515 
2516 	f2fs_up_read(&nm_i->nat_tree_lock);
2517 
2518 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2519 					nm_i->ra_nid_pages, META_NAT, false);
2520 
2521 	return 0;
2522 }
2523 
2524 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2525 {
2526 	int ret;
2527 
2528 	mutex_lock(&NM_I(sbi)->build_lock);
2529 	ret = __f2fs_build_free_nids(sbi, sync, mount);
2530 	mutex_unlock(&NM_I(sbi)->build_lock);
2531 
2532 	return ret;
2533 }
2534 
2535 /*
2536  * If this function returns success, caller can obtain a new nid
2537  * from second parameter of this function.
2538  * The returned nid could be used ino as well as nid when inode is created.
2539  */
2540 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2541 {
2542 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2543 	struct free_nid *i = NULL;
2544 retry:
2545 	if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2546 		f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2547 		return false;
2548 	}
2549 
2550 	spin_lock(&nm_i->nid_list_lock);
2551 
2552 	if (unlikely(nm_i->available_nids == 0)) {
2553 		spin_unlock(&nm_i->nid_list_lock);
2554 		return false;
2555 	}
2556 
2557 	/* We should not use stale free nids created by f2fs_build_free_nids */
2558 	if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2559 		f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2560 		i = list_first_entry(&nm_i->free_nid_list,
2561 					struct free_nid, list);
2562 		*nid = i->nid;
2563 
2564 		__move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2565 		nm_i->available_nids--;
2566 
2567 		update_free_nid_bitmap(sbi, *nid, false, false);
2568 
2569 		spin_unlock(&nm_i->nid_list_lock);
2570 		return true;
2571 	}
2572 	spin_unlock(&nm_i->nid_list_lock);
2573 
2574 	/* Let's scan nat pages and its caches to get free nids */
2575 	if (!f2fs_build_free_nids(sbi, true, false))
2576 		goto retry;
2577 	return false;
2578 }
2579 
2580 /*
2581  * f2fs_alloc_nid() should be called prior to this function.
2582  */
2583 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2584 {
2585 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2586 	struct free_nid *i;
2587 
2588 	spin_lock(&nm_i->nid_list_lock);
2589 	i = __lookup_free_nid_list(nm_i, nid);
2590 	f2fs_bug_on(sbi, !i);
2591 	__remove_free_nid(sbi, i, PREALLOC_NID);
2592 	spin_unlock(&nm_i->nid_list_lock);
2593 
2594 	kmem_cache_free(free_nid_slab, i);
2595 }
2596 
2597 /*
2598  * f2fs_alloc_nid() should be called prior to this function.
2599  */
2600 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2601 {
2602 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2603 	struct free_nid *i;
2604 	bool need_free = false;
2605 
2606 	if (!nid)
2607 		return;
2608 
2609 	spin_lock(&nm_i->nid_list_lock);
2610 	i = __lookup_free_nid_list(nm_i, nid);
2611 	f2fs_bug_on(sbi, !i);
2612 
2613 	if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2614 		__remove_free_nid(sbi, i, PREALLOC_NID);
2615 		need_free = true;
2616 	} else {
2617 		__move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2618 	}
2619 
2620 	nm_i->available_nids++;
2621 
2622 	update_free_nid_bitmap(sbi, nid, true, false);
2623 
2624 	spin_unlock(&nm_i->nid_list_lock);
2625 
2626 	if (need_free)
2627 		kmem_cache_free(free_nid_slab, i);
2628 }
2629 
2630 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2631 {
2632 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2633 	int nr = nr_shrink;
2634 
2635 	if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2636 		return 0;
2637 
2638 	if (!mutex_trylock(&nm_i->build_lock))
2639 		return 0;
2640 
2641 	while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2642 		struct free_nid *i, *next;
2643 		unsigned int batch = SHRINK_NID_BATCH_SIZE;
2644 
2645 		spin_lock(&nm_i->nid_list_lock);
2646 		list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2647 			if (!nr_shrink || !batch ||
2648 				nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2649 				break;
2650 			__remove_free_nid(sbi, i, FREE_NID);
2651 			kmem_cache_free(free_nid_slab, i);
2652 			nr_shrink--;
2653 			batch--;
2654 		}
2655 		spin_unlock(&nm_i->nid_list_lock);
2656 	}
2657 
2658 	mutex_unlock(&nm_i->build_lock);
2659 
2660 	return nr - nr_shrink;
2661 }
2662 
2663 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2664 {
2665 	void *src_addr, *dst_addr;
2666 	size_t inline_size;
2667 	struct page *ipage;
2668 	struct f2fs_inode *ri;
2669 
2670 	ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2671 	if (IS_ERR(ipage))
2672 		return PTR_ERR(ipage);
2673 
2674 	ri = F2FS_INODE(page);
2675 	if (ri->i_inline & F2FS_INLINE_XATTR) {
2676 		if (!f2fs_has_inline_xattr(inode)) {
2677 			set_inode_flag(inode, FI_INLINE_XATTR);
2678 			stat_inc_inline_xattr(inode);
2679 		}
2680 	} else {
2681 		if (f2fs_has_inline_xattr(inode)) {
2682 			stat_dec_inline_xattr(inode);
2683 			clear_inode_flag(inode, FI_INLINE_XATTR);
2684 		}
2685 		goto update_inode;
2686 	}
2687 
2688 	dst_addr = inline_xattr_addr(inode, ipage);
2689 	src_addr = inline_xattr_addr(inode, page);
2690 	inline_size = inline_xattr_size(inode);
2691 
2692 	f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2693 	memcpy(dst_addr, src_addr, inline_size);
2694 update_inode:
2695 	f2fs_update_inode(inode, ipage);
2696 	f2fs_put_page(ipage, 1);
2697 	return 0;
2698 }
2699 
2700 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2701 {
2702 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2703 	nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2704 	nid_t new_xnid;
2705 	struct dnode_of_data dn;
2706 	struct node_info ni;
2707 	struct page *xpage;
2708 	int err;
2709 
2710 	if (!prev_xnid)
2711 		goto recover_xnid;
2712 
2713 	/* 1: invalidate the previous xattr nid */
2714 	err = f2fs_get_node_info(sbi, prev_xnid, &ni, false);
2715 	if (err)
2716 		return err;
2717 
2718 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
2719 	dec_valid_node_count(sbi, inode, false);
2720 	set_node_addr(sbi, &ni, NULL_ADDR, false);
2721 
2722 recover_xnid:
2723 	/* 2: update xattr nid in inode */
2724 	if (!f2fs_alloc_nid(sbi, &new_xnid))
2725 		return -ENOSPC;
2726 
2727 	set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2728 	xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2729 	if (IS_ERR(xpage)) {
2730 		f2fs_alloc_nid_failed(sbi, new_xnid);
2731 		return PTR_ERR(xpage);
2732 	}
2733 
2734 	f2fs_alloc_nid_done(sbi, new_xnid);
2735 	f2fs_update_inode_page(inode);
2736 
2737 	/* 3: update and set xattr node page dirty */
2738 	memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2739 
2740 	set_page_dirty(xpage);
2741 	f2fs_put_page(xpage, 1);
2742 
2743 	return 0;
2744 }
2745 
2746 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2747 {
2748 	struct f2fs_inode *src, *dst;
2749 	nid_t ino = ino_of_node(page);
2750 	struct node_info old_ni, new_ni;
2751 	struct page *ipage;
2752 	int err;
2753 
2754 	err = f2fs_get_node_info(sbi, ino, &old_ni, false);
2755 	if (err)
2756 		return err;
2757 
2758 	if (unlikely(old_ni.blk_addr != NULL_ADDR))
2759 		return -EINVAL;
2760 retry:
2761 	ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2762 	if (!ipage) {
2763 		memalloc_retry_wait(GFP_NOFS);
2764 		goto retry;
2765 	}
2766 
2767 	/* Should not use this inode from free nid list */
2768 	remove_free_nid(sbi, ino);
2769 
2770 	if (!PageUptodate(ipage))
2771 		SetPageUptodate(ipage);
2772 	fill_node_footer(ipage, ino, ino, 0, true);
2773 	set_cold_node(ipage, false);
2774 
2775 	src = F2FS_INODE(page);
2776 	dst = F2FS_INODE(ipage);
2777 
2778 	memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2779 	dst->i_size = 0;
2780 	dst->i_blocks = cpu_to_le64(1);
2781 	dst->i_links = cpu_to_le32(1);
2782 	dst->i_xattr_nid = 0;
2783 	dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2784 	if (dst->i_inline & F2FS_EXTRA_ATTR) {
2785 		dst->i_extra_isize = src->i_extra_isize;
2786 
2787 		if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2788 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2789 							i_inline_xattr_size))
2790 			dst->i_inline_xattr_size = src->i_inline_xattr_size;
2791 
2792 		if (f2fs_sb_has_project_quota(sbi) &&
2793 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2794 								i_projid))
2795 			dst->i_projid = src->i_projid;
2796 
2797 		if (f2fs_sb_has_inode_crtime(sbi) &&
2798 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2799 							i_crtime_nsec)) {
2800 			dst->i_crtime = src->i_crtime;
2801 			dst->i_crtime_nsec = src->i_crtime_nsec;
2802 		}
2803 	}
2804 
2805 	new_ni = old_ni;
2806 	new_ni.ino = ino;
2807 
2808 	if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2809 		WARN_ON(1);
2810 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2811 	inc_valid_inode_count(sbi);
2812 	set_page_dirty(ipage);
2813 	f2fs_put_page(ipage, 1);
2814 	return 0;
2815 }
2816 
2817 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2818 			unsigned int segno, struct f2fs_summary_block *sum)
2819 {
2820 	struct f2fs_node *rn;
2821 	struct f2fs_summary *sum_entry;
2822 	block_t addr;
2823 	int i, idx, last_offset, nrpages;
2824 
2825 	/* scan the node segment */
2826 	last_offset = sbi->blocks_per_seg;
2827 	addr = START_BLOCK(sbi, segno);
2828 	sum_entry = &sum->entries[0];
2829 
2830 	for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2831 		nrpages = bio_max_segs(last_offset - i);
2832 
2833 		/* readahead node pages */
2834 		f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2835 
2836 		for (idx = addr; idx < addr + nrpages; idx++) {
2837 			struct page *page = f2fs_get_tmp_page(sbi, idx);
2838 
2839 			if (IS_ERR(page))
2840 				return PTR_ERR(page);
2841 
2842 			rn = F2FS_NODE(page);
2843 			sum_entry->nid = rn->footer.nid;
2844 			sum_entry->version = 0;
2845 			sum_entry->ofs_in_node = 0;
2846 			sum_entry++;
2847 			f2fs_put_page(page, 1);
2848 		}
2849 
2850 		invalidate_mapping_pages(META_MAPPING(sbi), addr,
2851 							addr + nrpages);
2852 	}
2853 	return 0;
2854 }
2855 
2856 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2857 {
2858 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2859 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2860 	struct f2fs_journal *journal = curseg->journal;
2861 	int i;
2862 
2863 	down_write(&curseg->journal_rwsem);
2864 	for (i = 0; i < nats_in_cursum(journal); i++) {
2865 		struct nat_entry *ne;
2866 		struct f2fs_nat_entry raw_ne;
2867 		nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2868 
2869 		if (f2fs_check_nid_range(sbi, nid))
2870 			continue;
2871 
2872 		raw_ne = nat_in_journal(journal, i);
2873 
2874 		ne = __lookup_nat_cache(nm_i, nid);
2875 		if (!ne) {
2876 			ne = __alloc_nat_entry(sbi, nid, true);
2877 			__init_nat_entry(nm_i, ne, &raw_ne, true);
2878 		}
2879 
2880 		/*
2881 		 * if a free nat in journal has not been used after last
2882 		 * checkpoint, we should remove it from available nids,
2883 		 * since later we will add it again.
2884 		 */
2885 		if (!get_nat_flag(ne, IS_DIRTY) &&
2886 				le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2887 			spin_lock(&nm_i->nid_list_lock);
2888 			nm_i->available_nids--;
2889 			spin_unlock(&nm_i->nid_list_lock);
2890 		}
2891 
2892 		__set_nat_cache_dirty(nm_i, ne);
2893 	}
2894 	update_nats_in_cursum(journal, -i);
2895 	up_write(&curseg->journal_rwsem);
2896 }
2897 
2898 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2899 						struct list_head *head, int max)
2900 {
2901 	struct nat_entry_set *cur;
2902 
2903 	if (nes->entry_cnt >= max)
2904 		goto add_out;
2905 
2906 	list_for_each_entry(cur, head, set_list) {
2907 		if (cur->entry_cnt >= nes->entry_cnt) {
2908 			list_add(&nes->set_list, cur->set_list.prev);
2909 			return;
2910 		}
2911 	}
2912 add_out:
2913 	list_add_tail(&nes->set_list, head);
2914 }
2915 
2916 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs,
2917 							unsigned int valid)
2918 {
2919 	if (valid == 0) {
2920 		__set_bit_le(nat_ofs, nm_i->empty_nat_bits);
2921 		__clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2922 		return;
2923 	}
2924 
2925 	__clear_bit_le(nat_ofs, nm_i->empty_nat_bits);
2926 	if (valid == NAT_ENTRY_PER_BLOCK)
2927 		__set_bit_le(nat_ofs, nm_i->full_nat_bits);
2928 	else
2929 		__clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2930 }
2931 
2932 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2933 						struct page *page)
2934 {
2935 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2936 	unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2937 	struct f2fs_nat_block *nat_blk = page_address(page);
2938 	int valid = 0;
2939 	int i = 0;
2940 
2941 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
2942 		return;
2943 
2944 	if (nat_index == 0) {
2945 		valid = 1;
2946 		i = 1;
2947 	}
2948 	for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2949 		if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2950 			valid++;
2951 	}
2952 
2953 	__update_nat_bits(nm_i, nat_index, valid);
2954 }
2955 
2956 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi)
2957 {
2958 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2959 	unsigned int nat_ofs;
2960 
2961 	f2fs_down_read(&nm_i->nat_tree_lock);
2962 
2963 	for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) {
2964 		unsigned int valid = 0, nid_ofs = 0;
2965 
2966 		/* handle nid zero due to it should never be used */
2967 		if (unlikely(nat_ofs == 0)) {
2968 			valid = 1;
2969 			nid_ofs = 1;
2970 		}
2971 
2972 		for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) {
2973 			if (!test_bit_le(nid_ofs,
2974 					nm_i->free_nid_bitmap[nat_ofs]))
2975 				valid++;
2976 		}
2977 
2978 		__update_nat_bits(nm_i, nat_ofs, valid);
2979 	}
2980 
2981 	f2fs_up_read(&nm_i->nat_tree_lock);
2982 }
2983 
2984 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2985 		struct nat_entry_set *set, struct cp_control *cpc)
2986 {
2987 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2988 	struct f2fs_journal *journal = curseg->journal;
2989 	nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2990 	bool to_journal = true;
2991 	struct f2fs_nat_block *nat_blk;
2992 	struct nat_entry *ne, *cur;
2993 	struct page *page = NULL;
2994 
2995 	/*
2996 	 * there are two steps to flush nat entries:
2997 	 * #1, flush nat entries to journal in current hot data summary block.
2998 	 * #2, flush nat entries to nat page.
2999 	 */
3000 	if ((cpc->reason & CP_UMOUNT) ||
3001 		!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
3002 		to_journal = false;
3003 
3004 	if (to_journal) {
3005 		down_write(&curseg->journal_rwsem);
3006 	} else {
3007 		page = get_next_nat_page(sbi, start_nid);
3008 		if (IS_ERR(page))
3009 			return PTR_ERR(page);
3010 
3011 		nat_blk = page_address(page);
3012 		f2fs_bug_on(sbi, !nat_blk);
3013 	}
3014 
3015 	/* flush dirty nats in nat entry set */
3016 	list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
3017 		struct f2fs_nat_entry *raw_ne;
3018 		nid_t nid = nat_get_nid(ne);
3019 		int offset;
3020 
3021 		f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
3022 
3023 		if (to_journal) {
3024 			offset = f2fs_lookup_journal_in_cursum(journal,
3025 							NAT_JOURNAL, nid, 1);
3026 			f2fs_bug_on(sbi, offset < 0);
3027 			raw_ne = &nat_in_journal(journal, offset);
3028 			nid_in_journal(journal, offset) = cpu_to_le32(nid);
3029 		} else {
3030 			raw_ne = &nat_blk->entries[nid - start_nid];
3031 		}
3032 		raw_nat_from_node_info(raw_ne, &ne->ni);
3033 		nat_reset_flag(ne);
3034 		__clear_nat_cache_dirty(NM_I(sbi), set, ne);
3035 		if (nat_get_blkaddr(ne) == NULL_ADDR) {
3036 			add_free_nid(sbi, nid, false, true);
3037 		} else {
3038 			spin_lock(&NM_I(sbi)->nid_list_lock);
3039 			update_free_nid_bitmap(sbi, nid, false, false);
3040 			spin_unlock(&NM_I(sbi)->nid_list_lock);
3041 		}
3042 	}
3043 
3044 	if (to_journal) {
3045 		up_write(&curseg->journal_rwsem);
3046 	} else {
3047 		update_nat_bits(sbi, start_nid, page);
3048 		f2fs_put_page(page, 1);
3049 	}
3050 
3051 	/* Allow dirty nats by node block allocation in write_begin */
3052 	if (!set->entry_cnt) {
3053 		radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
3054 		kmem_cache_free(nat_entry_set_slab, set);
3055 	}
3056 	return 0;
3057 }
3058 
3059 /*
3060  * This function is called during the checkpointing process.
3061  */
3062 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3063 {
3064 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3065 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3066 	struct f2fs_journal *journal = curseg->journal;
3067 	struct nat_entry_set *setvec[SETVEC_SIZE];
3068 	struct nat_entry_set *set, *tmp;
3069 	unsigned int found;
3070 	nid_t set_idx = 0;
3071 	LIST_HEAD(sets);
3072 	int err = 0;
3073 
3074 	/*
3075 	 * during unmount, let's flush nat_bits before checking
3076 	 * nat_cnt[DIRTY_NAT].
3077 	 */
3078 	if (cpc->reason & CP_UMOUNT) {
3079 		f2fs_down_write(&nm_i->nat_tree_lock);
3080 		remove_nats_in_journal(sbi);
3081 		f2fs_up_write(&nm_i->nat_tree_lock);
3082 	}
3083 
3084 	if (!nm_i->nat_cnt[DIRTY_NAT])
3085 		return 0;
3086 
3087 	f2fs_down_write(&nm_i->nat_tree_lock);
3088 
3089 	/*
3090 	 * if there are no enough space in journal to store dirty nat
3091 	 * entries, remove all entries from journal and merge them
3092 	 * into nat entry set.
3093 	 */
3094 	if (cpc->reason & CP_UMOUNT ||
3095 		!__has_cursum_space(journal,
3096 			nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3097 		remove_nats_in_journal(sbi);
3098 
3099 	while ((found = __gang_lookup_nat_set(nm_i,
3100 					set_idx, SETVEC_SIZE, setvec))) {
3101 		unsigned idx;
3102 
3103 		set_idx = setvec[found - 1]->set + 1;
3104 		for (idx = 0; idx < found; idx++)
3105 			__adjust_nat_entry_set(setvec[idx], &sets,
3106 						MAX_NAT_JENTRIES(journal));
3107 	}
3108 
3109 	/* flush dirty nats in nat entry set */
3110 	list_for_each_entry_safe(set, tmp, &sets, set_list) {
3111 		err = __flush_nat_entry_set(sbi, set, cpc);
3112 		if (err)
3113 			break;
3114 	}
3115 
3116 	f2fs_up_write(&nm_i->nat_tree_lock);
3117 	/* Allow dirty nats by node block allocation in write_begin */
3118 
3119 	return err;
3120 }
3121 
3122 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3123 {
3124 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3125 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3126 	unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3127 	unsigned int i;
3128 	__u64 cp_ver = cur_cp_version(ckpt);
3129 	block_t nat_bits_addr;
3130 
3131 	nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3132 	nm_i->nat_bits = f2fs_kvzalloc(sbi,
3133 			nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3134 	if (!nm_i->nat_bits)
3135 		return -ENOMEM;
3136 
3137 	nm_i->full_nat_bits = nm_i->nat_bits + 8;
3138 	nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3139 
3140 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3141 		return 0;
3142 
3143 	nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3144 						nm_i->nat_bits_blocks;
3145 	for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3146 		struct page *page;
3147 
3148 		page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3149 		if (IS_ERR(page))
3150 			return PTR_ERR(page);
3151 
3152 		memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3153 					page_address(page), F2FS_BLKSIZE);
3154 		f2fs_put_page(page, 1);
3155 	}
3156 
3157 	cp_ver |= (cur_cp_crc(ckpt) << 32);
3158 	if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3159 		clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
3160 		f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)",
3161 			cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits));
3162 		return 0;
3163 	}
3164 
3165 	f2fs_notice(sbi, "Found nat_bits in checkpoint");
3166 	return 0;
3167 }
3168 
3169 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3170 {
3171 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3172 	unsigned int i = 0;
3173 	nid_t nid, last_nid;
3174 
3175 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3176 		return;
3177 
3178 	for (i = 0; i < nm_i->nat_blocks; i++) {
3179 		i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3180 		if (i >= nm_i->nat_blocks)
3181 			break;
3182 
3183 		__set_bit_le(i, nm_i->nat_block_bitmap);
3184 
3185 		nid = i * NAT_ENTRY_PER_BLOCK;
3186 		last_nid = nid + NAT_ENTRY_PER_BLOCK;
3187 
3188 		spin_lock(&NM_I(sbi)->nid_list_lock);
3189 		for (; nid < last_nid; nid++)
3190 			update_free_nid_bitmap(sbi, nid, true, true);
3191 		spin_unlock(&NM_I(sbi)->nid_list_lock);
3192 	}
3193 
3194 	for (i = 0; i < nm_i->nat_blocks; i++) {
3195 		i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3196 		if (i >= nm_i->nat_blocks)
3197 			break;
3198 
3199 		__set_bit_le(i, nm_i->nat_block_bitmap);
3200 	}
3201 }
3202 
3203 static int init_node_manager(struct f2fs_sb_info *sbi)
3204 {
3205 	struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3206 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3207 	unsigned char *version_bitmap;
3208 	unsigned int nat_segs;
3209 	int err;
3210 
3211 	nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3212 
3213 	/* segment_count_nat includes pair segment so divide to 2. */
3214 	nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3215 	nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3216 	nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3217 
3218 	/* not used nids: 0, node, meta, (and root counted as valid node) */
3219 	nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3220 						F2FS_RESERVED_NODE_NUM;
3221 	nm_i->nid_cnt[FREE_NID] = 0;
3222 	nm_i->nid_cnt[PREALLOC_NID] = 0;
3223 	nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3224 	nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3225 	nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3226 	nm_i->max_rf_node_blocks = DEF_RF_NODE_BLOCKS;
3227 
3228 	INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3229 	INIT_LIST_HEAD(&nm_i->free_nid_list);
3230 	INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3231 	INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3232 	INIT_LIST_HEAD(&nm_i->nat_entries);
3233 	spin_lock_init(&nm_i->nat_list_lock);
3234 
3235 	mutex_init(&nm_i->build_lock);
3236 	spin_lock_init(&nm_i->nid_list_lock);
3237 	init_f2fs_rwsem(&nm_i->nat_tree_lock);
3238 
3239 	nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3240 	nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3241 	version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3242 	nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3243 					GFP_KERNEL);
3244 	if (!nm_i->nat_bitmap)
3245 		return -ENOMEM;
3246 
3247 	err = __get_nat_bitmaps(sbi);
3248 	if (err)
3249 		return err;
3250 
3251 #ifdef CONFIG_F2FS_CHECK_FS
3252 	nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3253 					GFP_KERNEL);
3254 	if (!nm_i->nat_bitmap_mir)
3255 		return -ENOMEM;
3256 #endif
3257 
3258 	return 0;
3259 }
3260 
3261 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3262 {
3263 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3264 	int i;
3265 
3266 	nm_i->free_nid_bitmap =
3267 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3268 					      nm_i->nat_blocks),
3269 			      GFP_KERNEL);
3270 	if (!nm_i->free_nid_bitmap)
3271 		return -ENOMEM;
3272 
3273 	for (i = 0; i < nm_i->nat_blocks; i++) {
3274 		nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3275 			f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3276 		if (!nm_i->free_nid_bitmap[i])
3277 			return -ENOMEM;
3278 	}
3279 
3280 	nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3281 								GFP_KERNEL);
3282 	if (!nm_i->nat_block_bitmap)
3283 		return -ENOMEM;
3284 
3285 	nm_i->free_nid_count =
3286 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3287 					      nm_i->nat_blocks),
3288 			      GFP_KERNEL);
3289 	if (!nm_i->free_nid_count)
3290 		return -ENOMEM;
3291 	return 0;
3292 }
3293 
3294 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3295 {
3296 	int err;
3297 
3298 	sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3299 							GFP_KERNEL);
3300 	if (!sbi->nm_info)
3301 		return -ENOMEM;
3302 
3303 	err = init_node_manager(sbi);
3304 	if (err)
3305 		return err;
3306 
3307 	err = init_free_nid_cache(sbi);
3308 	if (err)
3309 		return err;
3310 
3311 	/* load free nid status from nat_bits table */
3312 	load_free_nid_bitmap(sbi);
3313 
3314 	return f2fs_build_free_nids(sbi, true, true);
3315 }
3316 
3317 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3318 {
3319 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3320 	struct free_nid *i, *next_i;
3321 	struct nat_entry *natvec[NATVEC_SIZE];
3322 	struct nat_entry_set *setvec[SETVEC_SIZE];
3323 	nid_t nid = 0;
3324 	unsigned int found;
3325 
3326 	if (!nm_i)
3327 		return;
3328 
3329 	/* destroy free nid list */
3330 	spin_lock(&nm_i->nid_list_lock);
3331 	list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3332 		__remove_free_nid(sbi, i, FREE_NID);
3333 		spin_unlock(&nm_i->nid_list_lock);
3334 		kmem_cache_free(free_nid_slab, i);
3335 		spin_lock(&nm_i->nid_list_lock);
3336 	}
3337 	f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3338 	f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3339 	f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3340 	spin_unlock(&nm_i->nid_list_lock);
3341 
3342 	/* destroy nat cache */
3343 	f2fs_down_write(&nm_i->nat_tree_lock);
3344 	while ((found = __gang_lookup_nat_cache(nm_i,
3345 					nid, NATVEC_SIZE, natvec))) {
3346 		unsigned idx;
3347 
3348 		nid = nat_get_nid(natvec[found - 1]) + 1;
3349 		for (idx = 0; idx < found; idx++) {
3350 			spin_lock(&nm_i->nat_list_lock);
3351 			list_del(&natvec[idx]->list);
3352 			spin_unlock(&nm_i->nat_list_lock);
3353 
3354 			__del_from_nat_cache(nm_i, natvec[idx]);
3355 		}
3356 	}
3357 	f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3358 
3359 	/* destroy nat set cache */
3360 	nid = 0;
3361 	while ((found = __gang_lookup_nat_set(nm_i,
3362 					nid, SETVEC_SIZE, setvec))) {
3363 		unsigned idx;
3364 
3365 		nid = setvec[found - 1]->set + 1;
3366 		for (idx = 0; idx < found; idx++) {
3367 			/* entry_cnt is not zero, when cp_error was occurred */
3368 			f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3369 			radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3370 			kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3371 		}
3372 	}
3373 	f2fs_up_write(&nm_i->nat_tree_lock);
3374 
3375 	kvfree(nm_i->nat_block_bitmap);
3376 	if (nm_i->free_nid_bitmap) {
3377 		int i;
3378 
3379 		for (i = 0; i < nm_i->nat_blocks; i++)
3380 			kvfree(nm_i->free_nid_bitmap[i]);
3381 		kvfree(nm_i->free_nid_bitmap);
3382 	}
3383 	kvfree(nm_i->free_nid_count);
3384 
3385 	kvfree(nm_i->nat_bitmap);
3386 	kvfree(nm_i->nat_bits);
3387 #ifdef CONFIG_F2FS_CHECK_FS
3388 	kvfree(nm_i->nat_bitmap_mir);
3389 #endif
3390 	sbi->nm_info = NULL;
3391 	kfree(nm_i);
3392 }
3393 
3394 int __init f2fs_create_node_manager_caches(void)
3395 {
3396 	nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3397 			sizeof(struct nat_entry));
3398 	if (!nat_entry_slab)
3399 		goto fail;
3400 
3401 	free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3402 			sizeof(struct free_nid));
3403 	if (!free_nid_slab)
3404 		goto destroy_nat_entry;
3405 
3406 	nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3407 			sizeof(struct nat_entry_set));
3408 	if (!nat_entry_set_slab)
3409 		goto destroy_free_nid;
3410 
3411 	fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3412 			sizeof(struct fsync_node_entry));
3413 	if (!fsync_node_entry_slab)
3414 		goto destroy_nat_entry_set;
3415 	return 0;
3416 
3417 destroy_nat_entry_set:
3418 	kmem_cache_destroy(nat_entry_set_slab);
3419 destroy_free_nid:
3420 	kmem_cache_destroy(free_nid_slab);
3421 destroy_nat_entry:
3422 	kmem_cache_destroy(nat_entry_slab);
3423 fail:
3424 	return -ENOMEM;
3425 }
3426 
3427 void f2fs_destroy_node_manager_caches(void)
3428 {
3429 	kmem_cache_destroy(fsync_node_entry_slab);
3430 	kmem_cache_destroy(nat_entry_set_slab);
3431 	kmem_cache_destroy(free_nid_slab);
3432 	kmem_cache_destroy(nat_entry_slab);
3433 }
3434