xref: /openbmc/linux/fs/f2fs/node.c (revision 6342a20e)
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 		f2fs_handle_error(sbi, ERROR_CORRUPTED_INODE);
40 		return -EFSCORRUPTED;
41 	}
42 	return 0;
43 }
44 
45 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
46 {
47 	struct f2fs_nm_info *nm_i = NM_I(sbi);
48 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
49 	struct sysinfo val;
50 	unsigned long avail_ram;
51 	unsigned long mem_size = 0;
52 	bool res = false;
53 
54 	if (!nm_i)
55 		return true;
56 
57 	si_meminfo(&val);
58 
59 	/* only uses low memory */
60 	avail_ram = val.totalram - val.totalhigh;
61 
62 	/*
63 	 * give 25%, 25%, 50%, 50%, 25%, 25% memory for each components respectively
64 	 */
65 	if (type == FREE_NIDS) {
66 		mem_size = (nm_i->nid_cnt[FREE_NID] *
67 				sizeof(struct free_nid)) >> PAGE_SHIFT;
68 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
69 	} else if (type == NAT_ENTRIES) {
70 		mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
71 				sizeof(struct nat_entry)) >> PAGE_SHIFT;
72 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
73 		if (excess_cached_nats(sbi))
74 			res = false;
75 	} else if (type == DIRTY_DENTS) {
76 		if (sbi->sb->s_bdi->wb.dirty_exceeded)
77 			return false;
78 		mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
79 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
80 	} else if (type == INO_ENTRIES) {
81 		int i;
82 
83 		for (i = 0; i < MAX_INO_ENTRY; i++)
84 			mem_size += sbi->im[i].ino_num *
85 						sizeof(struct ino_entry);
86 		mem_size >>= PAGE_SHIFT;
87 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
88 	} else if (type == READ_EXTENT_CACHE || type == AGE_EXTENT_CACHE) {
89 		enum extent_type etype = type == READ_EXTENT_CACHE ?
90 						EX_READ : EX_BLOCK_AGE;
91 		struct extent_tree_info *eti = &sbi->extent_tree[etype];
92 
93 		mem_size = (atomic_read(&eti->total_ext_tree) *
94 				sizeof(struct extent_tree) +
95 				atomic_read(&eti->total_ext_node) *
96 				sizeof(struct extent_node)) >> PAGE_SHIFT;
97 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
98 	} else if (type == DISCARD_CACHE) {
99 		mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
100 				sizeof(struct discard_cmd)) >> PAGE_SHIFT;
101 		res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
102 	} else if (type == COMPRESS_PAGE) {
103 #ifdef CONFIG_F2FS_FS_COMPRESSION
104 		unsigned long free_ram = val.freeram;
105 
106 		/*
107 		 * free memory is lower than watermark or cached page count
108 		 * exceed threshold, deny caching compress page.
109 		 */
110 		res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
111 			(COMPRESS_MAPPING(sbi)->nrpages <
112 			 free_ram * sbi->compress_percent / 100);
113 #else
114 		res = false;
115 #endif
116 	} else {
117 		if (!sbi->sb->s_bdi->wb.dirty_exceeded)
118 			return true;
119 	}
120 	return res;
121 }
122 
123 static void clear_node_page_dirty(struct page *page)
124 {
125 	if (PageDirty(page)) {
126 		f2fs_clear_page_cache_dirty_tag(page);
127 		clear_page_dirty_for_io(page);
128 		dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
129 	}
130 	ClearPageUptodate(page);
131 }
132 
133 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
134 {
135 	return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
136 }
137 
138 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
139 {
140 	struct page *src_page;
141 	struct page *dst_page;
142 	pgoff_t dst_off;
143 	void *src_addr;
144 	void *dst_addr;
145 	struct f2fs_nm_info *nm_i = NM_I(sbi);
146 
147 	dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
148 
149 	/* get current nat block page with lock */
150 	src_page = get_current_nat_page(sbi, nid);
151 	if (IS_ERR(src_page))
152 		return src_page;
153 	dst_page = f2fs_grab_meta_page(sbi, dst_off);
154 	f2fs_bug_on(sbi, PageDirty(src_page));
155 
156 	src_addr = page_address(src_page);
157 	dst_addr = page_address(dst_page);
158 	memcpy(dst_addr, src_addr, PAGE_SIZE);
159 	set_page_dirty(dst_page);
160 	f2fs_put_page(src_page, 1);
161 
162 	set_to_next_nat(nm_i, nid);
163 
164 	return dst_page;
165 }
166 
167 static struct nat_entry *__alloc_nat_entry(struct f2fs_sb_info *sbi,
168 						nid_t nid, bool no_fail)
169 {
170 	struct nat_entry *new;
171 
172 	new = f2fs_kmem_cache_alloc(nat_entry_slab,
173 					GFP_F2FS_ZERO, no_fail, sbi);
174 	if (new) {
175 		nat_set_nid(new, nid);
176 		nat_reset_flag(new);
177 	}
178 	return new;
179 }
180 
181 static void __free_nat_entry(struct nat_entry *e)
182 {
183 	kmem_cache_free(nat_entry_slab, e);
184 }
185 
186 /* must be locked by nat_tree_lock */
187 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
188 	struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
189 {
190 	if (no_fail)
191 		f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
192 	else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
193 		return NULL;
194 
195 	if (raw_ne)
196 		node_info_from_raw_nat(&ne->ni, raw_ne);
197 
198 	spin_lock(&nm_i->nat_list_lock);
199 	list_add_tail(&ne->list, &nm_i->nat_entries);
200 	spin_unlock(&nm_i->nat_list_lock);
201 
202 	nm_i->nat_cnt[TOTAL_NAT]++;
203 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
204 	return ne;
205 }
206 
207 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
208 {
209 	struct nat_entry *ne;
210 
211 	ne = radix_tree_lookup(&nm_i->nat_root, n);
212 
213 	/* for recent accessed nat entry, move it to tail of lru list */
214 	if (ne && !get_nat_flag(ne, IS_DIRTY)) {
215 		spin_lock(&nm_i->nat_list_lock);
216 		if (!list_empty(&ne->list))
217 			list_move_tail(&ne->list, &nm_i->nat_entries);
218 		spin_unlock(&nm_i->nat_list_lock);
219 	}
220 
221 	return ne;
222 }
223 
224 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
225 		nid_t start, unsigned int nr, struct nat_entry **ep)
226 {
227 	return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
228 }
229 
230 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
231 {
232 	radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
233 	nm_i->nat_cnt[TOTAL_NAT]--;
234 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
235 	__free_nat_entry(e);
236 }
237 
238 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
239 							struct nat_entry *ne)
240 {
241 	nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
242 	struct nat_entry_set *head;
243 
244 	head = radix_tree_lookup(&nm_i->nat_set_root, set);
245 	if (!head) {
246 		head = f2fs_kmem_cache_alloc(nat_entry_set_slab,
247 						GFP_NOFS, true, NULL);
248 
249 		INIT_LIST_HEAD(&head->entry_list);
250 		INIT_LIST_HEAD(&head->set_list);
251 		head->set = set;
252 		head->entry_cnt = 0;
253 		f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
254 	}
255 	return head;
256 }
257 
258 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
259 						struct nat_entry *ne)
260 {
261 	struct nat_entry_set *head;
262 	bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
263 
264 	if (!new_ne)
265 		head = __grab_nat_entry_set(nm_i, ne);
266 
267 	/*
268 	 * update entry_cnt in below condition:
269 	 * 1. update NEW_ADDR to valid block address;
270 	 * 2. update old block address to new one;
271 	 */
272 	if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
273 				!get_nat_flag(ne, IS_DIRTY)))
274 		head->entry_cnt++;
275 
276 	set_nat_flag(ne, IS_PREALLOC, new_ne);
277 
278 	if (get_nat_flag(ne, IS_DIRTY))
279 		goto refresh_list;
280 
281 	nm_i->nat_cnt[DIRTY_NAT]++;
282 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
283 	set_nat_flag(ne, IS_DIRTY, true);
284 refresh_list:
285 	spin_lock(&nm_i->nat_list_lock);
286 	if (new_ne)
287 		list_del_init(&ne->list);
288 	else
289 		list_move_tail(&ne->list, &head->entry_list);
290 	spin_unlock(&nm_i->nat_list_lock);
291 }
292 
293 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
294 		struct nat_entry_set *set, struct nat_entry *ne)
295 {
296 	spin_lock(&nm_i->nat_list_lock);
297 	list_move_tail(&ne->list, &nm_i->nat_entries);
298 	spin_unlock(&nm_i->nat_list_lock);
299 
300 	set_nat_flag(ne, IS_DIRTY, false);
301 	set->entry_cnt--;
302 	nm_i->nat_cnt[DIRTY_NAT]--;
303 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
304 }
305 
306 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
307 		nid_t start, unsigned int nr, struct nat_entry_set **ep)
308 {
309 	return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
310 							start, nr);
311 }
312 
313 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
314 {
315 	return NODE_MAPPING(sbi) == page->mapping &&
316 			IS_DNODE(page) && is_cold_node(page);
317 }
318 
319 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
320 {
321 	spin_lock_init(&sbi->fsync_node_lock);
322 	INIT_LIST_HEAD(&sbi->fsync_node_list);
323 	sbi->fsync_seg_id = 0;
324 	sbi->fsync_node_num = 0;
325 }
326 
327 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
328 							struct page *page)
329 {
330 	struct fsync_node_entry *fn;
331 	unsigned long flags;
332 	unsigned int seq_id;
333 
334 	fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab,
335 					GFP_NOFS, true, NULL);
336 
337 	get_page(page);
338 	fn->page = page;
339 	INIT_LIST_HEAD(&fn->list);
340 
341 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
342 	list_add_tail(&fn->list, &sbi->fsync_node_list);
343 	fn->seq_id = sbi->fsync_seg_id++;
344 	seq_id = fn->seq_id;
345 	sbi->fsync_node_num++;
346 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
347 
348 	return seq_id;
349 }
350 
351 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
352 {
353 	struct fsync_node_entry *fn;
354 	unsigned long flags;
355 
356 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
357 	list_for_each_entry(fn, &sbi->fsync_node_list, list) {
358 		if (fn->page == page) {
359 			list_del(&fn->list);
360 			sbi->fsync_node_num--;
361 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
362 			kmem_cache_free(fsync_node_entry_slab, fn);
363 			put_page(page);
364 			return;
365 		}
366 	}
367 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
368 	f2fs_bug_on(sbi, 1);
369 }
370 
371 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
372 {
373 	unsigned long flags;
374 
375 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
376 	sbi->fsync_seg_id = 0;
377 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
378 }
379 
380 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
381 {
382 	struct f2fs_nm_info *nm_i = NM_I(sbi);
383 	struct nat_entry *e;
384 	bool need = false;
385 
386 	f2fs_down_read(&nm_i->nat_tree_lock);
387 	e = __lookup_nat_cache(nm_i, nid);
388 	if (e) {
389 		if (!get_nat_flag(e, IS_CHECKPOINTED) &&
390 				!get_nat_flag(e, HAS_FSYNCED_INODE))
391 			need = true;
392 	}
393 	f2fs_up_read(&nm_i->nat_tree_lock);
394 	return need;
395 }
396 
397 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
398 {
399 	struct f2fs_nm_info *nm_i = NM_I(sbi);
400 	struct nat_entry *e;
401 	bool is_cp = true;
402 
403 	f2fs_down_read(&nm_i->nat_tree_lock);
404 	e = __lookup_nat_cache(nm_i, nid);
405 	if (e && !get_nat_flag(e, IS_CHECKPOINTED))
406 		is_cp = false;
407 	f2fs_up_read(&nm_i->nat_tree_lock);
408 	return is_cp;
409 }
410 
411 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
412 {
413 	struct f2fs_nm_info *nm_i = NM_I(sbi);
414 	struct nat_entry *e;
415 	bool need_update = true;
416 
417 	f2fs_down_read(&nm_i->nat_tree_lock);
418 	e = __lookup_nat_cache(nm_i, ino);
419 	if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
420 			(get_nat_flag(e, IS_CHECKPOINTED) ||
421 			 get_nat_flag(e, HAS_FSYNCED_INODE)))
422 		need_update = false;
423 	f2fs_up_read(&nm_i->nat_tree_lock);
424 	return need_update;
425 }
426 
427 /* must be locked by nat_tree_lock */
428 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
429 						struct f2fs_nat_entry *ne)
430 {
431 	struct f2fs_nm_info *nm_i = NM_I(sbi);
432 	struct nat_entry *new, *e;
433 
434 	/* Let's mitigate lock contention of nat_tree_lock during checkpoint */
435 	if (f2fs_rwsem_is_locked(&sbi->cp_global_sem))
436 		return;
437 
438 	new = __alloc_nat_entry(sbi, nid, false);
439 	if (!new)
440 		return;
441 
442 	f2fs_down_write(&nm_i->nat_tree_lock);
443 	e = __lookup_nat_cache(nm_i, nid);
444 	if (!e)
445 		e = __init_nat_entry(nm_i, new, ne, false);
446 	else
447 		f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
448 				nat_get_blkaddr(e) !=
449 					le32_to_cpu(ne->block_addr) ||
450 				nat_get_version(e) != ne->version);
451 	f2fs_up_write(&nm_i->nat_tree_lock);
452 	if (e != new)
453 		__free_nat_entry(new);
454 }
455 
456 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
457 			block_t new_blkaddr, bool fsync_done)
458 {
459 	struct f2fs_nm_info *nm_i = NM_I(sbi);
460 	struct nat_entry *e;
461 	struct nat_entry *new = __alloc_nat_entry(sbi, ni->nid, true);
462 
463 	f2fs_down_write(&nm_i->nat_tree_lock);
464 	e = __lookup_nat_cache(nm_i, ni->nid);
465 	if (!e) {
466 		e = __init_nat_entry(nm_i, new, NULL, true);
467 		copy_node_info(&e->ni, ni);
468 		f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
469 	} else if (new_blkaddr == NEW_ADDR) {
470 		/*
471 		 * when nid is reallocated,
472 		 * previous nat entry can be remained in nat cache.
473 		 * So, reinitialize it with new information.
474 		 */
475 		copy_node_info(&e->ni, ni);
476 		f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
477 	}
478 	/* let's free early to reduce memory consumption */
479 	if (e != new)
480 		__free_nat_entry(new);
481 
482 	/* sanity check */
483 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
484 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
485 			new_blkaddr == NULL_ADDR);
486 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
487 			new_blkaddr == NEW_ADDR);
488 	f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
489 			new_blkaddr == NEW_ADDR);
490 
491 	/* increment version no as node is removed */
492 	if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
493 		unsigned char version = nat_get_version(e);
494 
495 		nat_set_version(e, inc_node_version(version));
496 	}
497 
498 	/* change address */
499 	nat_set_blkaddr(e, new_blkaddr);
500 	if (!__is_valid_data_blkaddr(new_blkaddr))
501 		set_nat_flag(e, IS_CHECKPOINTED, false);
502 	__set_nat_cache_dirty(nm_i, e);
503 
504 	/* update fsync_mark if its inode nat entry is still alive */
505 	if (ni->nid != ni->ino)
506 		e = __lookup_nat_cache(nm_i, ni->ino);
507 	if (e) {
508 		if (fsync_done && ni->nid == ni->ino)
509 			set_nat_flag(e, HAS_FSYNCED_INODE, true);
510 		set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
511 	}
512 	f2fs_up_write(&nm_i->nat_tree_lock);
513 }
514 
515 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
516 {
517 	struct f2fs_nm_info *nm_i = NM_I(sbi);
518 	int nr = nr_shrink;
519 
520 	if (!f2fs_down_write_trylock(&nm_i->nat_tree_lock))
521 		return 0;
522 
523 	spin_lock(&nm_i->nat_list_lock);
524 	while (nr_shrink) {
525 		struct nat_entry *ne;
526 
527 		if (list_empty(&nm_i->nat_entries))
528 			break;
529 
530 		ne = list_first_entry(&nm_i->nat_entries,
531 					struct nat_entry, list);
532 		list_del(&ne->list);
533 		spin_unlock(&nm_i->nat_list_lock);
534 
535 		__del_from_nat_cache(nm_i, ne);
536 		nr_shrink--;
537 
538 		spin_lock(&nm_i->nat_list_lock);
539 	}
540 	spin_unlock(&nm_i->nat_list_lock);
541 
542 	f2fs_up_write(&nm_i->nat_tree_lock);
543 	return nr - nr_shrink;
544 }
545 
546 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
547 				struct node_info *ni, bool checkpoint_context)
548 {
549 	struct f2fs_nm_info *nm_i = NM_I(sbi);
550 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
551 	struct f2fs_journal *journal = curseg->journal;
552 	nid_t start_nid = START_NID(nid);
553 	struct f2fs_nat_block *nat_blk;
554 	struct page *page = NULL;
555 	struct f2fs_nat_entry ne;
556 	struct nat_entry *e;
557 	pgoff_t index;
558 	block_t blkaddr;
559 	int i;
560 
561 	ni->nid = nid;
562 retry:
563 	/* Check nat cache */
564 	f2fs_down_read(&nm_i->nat_tree_lock);
565 	e = __lookup_nat_cache(nm_i, nid);
566 	if (e) {
567 		ni->ino = nat_get_ino(e);
568 		ni->blk_addr = nat_get_blkaddr(e);
569 		ni->version = nat_get_version(e);
570 		f2fs_up_read(&nm_i->nat_tree_lock);
571 		return 0;
572 	}
573 
574 	/*
575 	 * Check current segment summary by trying to grab journal_rwsem first.
576 	 * This sem is on the critical path on the checkpoint requiring the above
577 	 * nat_tree_lock. Therefore, we should retry, if we failed to grab here
578 	 * while not bothering checkpoint.
579 	 */
580 	if (!f2fs_rwsem_is_locked(&sbi->cp_global_sem) || checkpoint_context) {
581 		down_read(&curseg->journal_rwsem);
582 	} else if (f2fs_rwsem_is_contended(&nm_i->nat_tree_lock) ||
583 				!down_read_trylock(&curseg->journal_rwsem)) {
584 		f2fs_up_read(&nm_i->nat_tree_lock);
585 		goto retry;
586 	}
587 
588 	i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
589 	if (i >= 0) {
590 		ne = nat_in_journal(journal, i);
591 		node_info_from_raw_nat(ni, &ne);
592 	}
593 	up_read(&curseg->journal_rwsem);
594 	if (i >= 0) {
595 		f2fs_up_read(&nm_i->nat_tree_lock);
596 		goto cache;
597 	}
598 
599 	/* Fill node_info from nat page */
600 	index = current_nat_addr(sbi, nid);
601 	f2fs_up_read(&nm_i->nat_tree_lock);
602 
603 	page = f2fs_get_meta_page(sbi, index);
604 	if (IS_ERR(page))
605 		return PTR_ERR(page);
606 
607 	nat_blk = (struct f2fs_nat_block *)page_address(page);
608 	ne = nat_blk->entries[nid - start_nid];
609 	node_info_from_raw_nat(ni, &ne);
610 	f2fs_put_page(page, 1);
611 cache:
612 	blkaddr = le32_to_cpu(ne.block_addr);
613 	if (__is_valid_data_blkaddr(blkaddr) &&
614 		!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
615 		return -EFAULT;
616 
617 	/* cache nat entry */
618 	cache_nat_entry(sbi, nid, &ne);
619 	return 0;
620 }
621 
622 /*
623  * readahead MAX_RA_NODE number of node pages.
624  */
625 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
626 {
627 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
628 	struct blk_plug plug;
629 	int i, end;
630 	nid_t nid;
631 
632 	blk_start_plug(&plug);
633 
634 	/* Then, try readahead for siblings of the desired node */
635 	end = start + n;
636 	end = min(end, NIDS_PER_BLOCK);
637 	for (i = start; i < end; i++) {
638 		nid = get_nid(parent, i, false);
639 		f2fs_ra_node_page(sbi, nid);
640 	}
641 
642 	blk_finish_plug(&plug);
643 }
644 
645 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
646 {
647 	const long direct_index = ADDRS_PER_INODE(dn->inode);
648 	const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
649 	const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
650 	unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
651 	int cur_level = dn->cur_level;
652 	int max_level = dn->max_level;
653 	pgoff_t base = 0;
654 
655 	if (!dn->max_level)
656 		return pgofs + 1;
657 
658 	while (max_level-- > cur_level)
659 		skipped_unit *= NIDS_PER_BLOCK;
660 
661 	switch (dn->max_level) {
662 	case 3:
663 		base += 2 * indirect_blks;
664 		fallthrough;
665 	case 2:
666 		base += 2 * direct_blks;
667 		fallthrough;
668 	case 1:
669 		base += direct_index;
670 		break;
671 	default:
672 		f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
673 	}
674 
675 	return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
676 }
677 
678 /*
679  * The maximum depth is four.
680  * Offset[0] will have raw inode offset.
681  */
682 static int get_node_path(struct inode *inode, long block,
683 				int offset[4], unsigned int noffset[4])
684 {
685 	const long direct_index = ADDRS_PER_INODE(inode);
686 	const long direct_blks = ADDRS_PER_BLOCK(inode);
687 	const long dptrs_per_blk = NIDS_PER_BLOCK;
688 	const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
689 	const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
690 	int n = 0;
691 	int level = 0;
692 
693 	noffset[0] = 0;
694 
695 	if (block < direct_index) {
696 		offset[n] = block;
697 		goto got;
698 	}
699 	block -= direct_index;
700 	if (block < direct_blks) {
701 		offset[n++] = NODE_DIR1_BLOCK;
702 		noffset[n] = 1;
703 		offset[n] = block;
704 		level = 1;
705 		goto got;
706 	}
707 	block -= direct_blks;
708 	if (block < direct_blks) {
709 		offset[n++] = NODE_DIR2_BLOCK;
710 		noffset[n] = 2;
711 		offset[n] = block;
712 		level = 1;
713 		goto got;
714 	}
715 	block -= direct_blks;
716 	if (block < indirect_blks) {
717 		offset[n++] = NODE_IND1_BLOCK;
718 		noffset[n] = 3;
719 		offset[n++] = block / direct_blks;
720 		noffset[n] = 4 + offset[n - 1];
721 		offset[n] = block % direct_blks;
722 		level = 2;
723 		goto got;
724 	}
725 	block -= indirect_blks;
726 	if (block < indirect_blks) {
727 		offset[n++] = NODE_IND2_BLOCK;
728 		noffset[n] = 4 + dptrs_per_blk;
729 		offset[n++] = block / direct_blks;
730 		noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
731 		offset[n] = block % direct_blks;
732 		level = 2;
733 		goto got;
734 	}
735 	block -= indirect_blks;
736 	if (block < dindirect_blks) {
737 		offset[n++] = NODE_DIND_BLOCK;
738 		noffset[n] = 5 + (dptrs_per_blk * 2);
739 		offset[n++] = block / indirect_blks;
740 		noffset[n] = 6 + (dptrs_per_blk * 2) +
741 			      offset[n - 1] * (dptrs_per_blk + 1);
742 		offset[n++] = (block / direct_blks) % dptrs_per_blk;
743 		noffset[n] = 7 + (dptrs_per_blk * 2) +
744 			      offset[n - 2] * (dptrs_per_blk + 1) +
745 			      offset[n - 1];
746 		offset[n] = block % direct_blks;
747 		level = 3;
748 		goto got;
749 	} else {
750 		return -E2BIG;
751 	}
752 got:
753 	return level;
754 }
755 
756 /*
757  * Caller should call f2fs_put_dnode(dn).
758  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
759  * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
760  */
761 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
762 {
763 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
764 	struct page *npage[4];
765 	struct page *parent = NULL;
766 	int offset[4];
767 	unsigned int noffset[4];
768 	nid_t nids[4];
769 	int level, i = 0;
770 	int err = 0;
771 
772 	level = get_node_path(dn->inode, index, offset, noffset);
773 	if (level < 0)
774 		return level;
775 
776 	nids[0] = dn->inode->i_ino;
777 	npage[0] = dn->inode_page;
778 
779 	if (!npage[0]) {
780 		npage[0] = f2fs_get_node_page(sbi, nids[0]);
781 		if (IS_ERR(npage[0]))
782 			return PTR_ERR(npage[0]);
783 	}
784 
785 	/* if inline_data is set, should not report any block indices */
786 	if (f2fs_has_inline_data(dn->inode) && index) {
787 		err = -ENOENT;
788 		f2fs_put_page(npage[0], 1);
789 		goto release_out;
790 	}
791 
792 	parent = npage[0];
793 	if (level != 0)
794 		nids[1] = get_nid(parent, offset[0], true);
795 	dn->inode_page = npage[0];
796 	dn->inode_page_locked = true;
797 
798 	/* get indirect or direct nodes */
799 	for (i = 1; i <= level; i++) {
800 		bool done = false;
801 
802 		if (!nids[i] && mode == ALLOC_NODE) {
803 			/* alloc new node */
804 			if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
805 				err = -ENOSPC;
806 				goto release_pages;
807 			}
808 
809 			dn->nid = nids[i];
810 			npage[i] = f2fs_new_node_page(dn, noffset[i]);
811 			if (IS_ERR(npage[i])) {
812 				f2fs_alloc_nid_failed(sbi, nids[i]);
813 				err = PTR_ERR(npage[i]);
814 				goto release_pages;
815 			}
816 
817 			set_nid(parent, offset[i - 1], nids[i], i == 1);
818 			f2fs_alloc_nid_done(sbi, nids[i]);
819 			done = true;
820 		} else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
821 			npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
822 			if (IS_ERR(npage[i])) {
823 				err = PTR_ERR(npage[i]);
824 				goto release_pages;
825 			}
826 			done = true;
827 		}
828 		if (i == 1) {
829 			dn->inode_page_locked = false;
830 			unlock_page(parent);
831 		} else {
832 			f2fs_put_page(parent, 1);
833 		}
834 
835 		if (!done) {
836 			npage[i] = f2fs_get_node_page(sbi, nids[i]);
837 			if (IS_ERR(npage[i])) {
838 				err = PTR_ERR(npage[i]);
839 				f2fs_put_page(npage[0], 0);
840 				goto release_out;
841 			}
842 		}
843 		if (i < level) {
844 			parent = npage[i];
845 			nids[i + 1] = get_nid(parent, offset[i], false);
846 		}
847 	}
848 	dn->nid = nids[level];
849 	dn->ofs_in_node = offset[level];
850 	dn->node_page = npage[level];
851 	dn->data_blkaddr = f2fs_data_blkaddr(dn);
852 
853 	if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) &&
854 					f2fs_sb_has_readonly(sbi)) {
855 		unsigned int c_len = f2fs_cluster_blocks_are_contiguous(dn);
856 		block_t blkaddr;
857 
858 		if (!c_len)
859 			goto out;
860 
861 		blkaddr = f2fs_data_blkaddr(dn);
862 		if (blkaddr == COMPRESS_ADDR)
863 			blkaddr = data_blkaddr(dn->inode, dn->node_page,
864 						dn->ofs_in_node + 1);
865 
866 		f2fs_update_read_extent_tree_range_compressed(dn->inode,
867 					index, blkaddr,
868 					F2FS_I(dn->inode)->i_cluster_size,
869 					c_len);
870 	}
871 out:
872 	return 0;
873 
874 release_pages:
875 	f2fs_put_page(parent, 1);
876 	if (i > 1)
877 		f2fs_put_page(npage[0], 0);
878 release_out:
879 	dn->inode_page = NULL;
880 	dn->node_page = NULL;
881 	if (err == -ENOENT) {
882 		dn->cur_level = i;
883 		dn->max_level = level;
884 		dn->ofs_in_node = offset[level];
885 	}
886 	return err;
887 }
888 
889 static int truncate_node(struct dnode_of_data *dn)
890 {
891 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
892 	struct node_info ni;
893 	int err;
894 	pgoff_t index;
895 
896 	err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
897 	if (err)
898 		return err;
899 
900 	/* Deallocate node address */
901 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
902 	dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
903 	set_node_addr(sbi, &ni, NULL_ADDR, false);
904 
905 	if (dn->nid == dn->inode->i_ino) {
906 		f2fs_remove_orphan_inode(sbi, dn->nid);
907 		dec_valid_inode_count(sbi);
908 		f2fs_inode_synced(dn->inode);
909 	}
910 
911 	clear_node_page_dirty(dn->node_page);
912 	set_sbi_flag(sbi, SBI_IS_DIRTY);
913 
914 	index = dn->node_page->index;
915 	f2fs_put_page(dn->node_page, 1);
916 
917 	invalidate_mapping_pages(NODE_MAPPING(sbi),
918 			index, index);
919 
920 	dn->node_page = NULL;
921 	trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
922 
923 	return 0;
924 }
925 
926 static int truncate_dnode(struct dnode_of_data *dn)
927 {
928 	struct page *page;
929 	int err;
930 
931 	if (dn->nid == 0)
932 		return 1;
933 
934 	/* get direct node */
935 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
936 	if (PTR_ERR(page) == -ENOENT)
937 		return 1;
938 	else if (IS_ERR(page))
939 		return PTR_ERR(page);
940 
941 	/* Make dnode_of_data for parameter */
942 	dn->node_page = page;
943 	dn->ofs_in_node = 0;
944 	f2fs_truncate_data_blocks(dn);
945 	err = truncate_node(dn);
946 	if (err)
947 		return err;
948 
949 	return 1;
950 }
951 
952 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
953 						int ofs, int depth)
954 {
955 	struct dnode_of_data rdn = *dn;
956 	struct page *page;
957 	struct f2fs_node *rn;
958 	nid_t child_nid;
959 	unsigned int child_nofs;
960 	int freed = 0;
961 	int i, ret;
962 
963 	if (dn->nid == 0)
964 		return NIDS_PER_BLOCK + 1;
965 
966 	trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
967 
968 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
969 	if (IS_ERR(page)) {
970 		trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
971 		return PTR_ERR(page);
972 	}
973 
974 	f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
975 
976 	rn = F2FS_NODE(page);
977 	if (depth < 3) {
978 		for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
979 			child_nid = le32_to_cpu(rn->in.nid[i]);
980 			if (child_nid == 0)
981 				continue;
982 			rdn.nid = child_nid;
983 			ret = truncate_dnode(&rdn);
984 			if (ret < 0)
985 				goto out_err;
986 			if (set_nid(page, i, 0, false))
987 				dn->node_changed = true;
988 		}
989 	} else {
990 		child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
991 		for (i = ofs; i < NIDS_PER_BLOCK; i++) {
992 			child_nid = le32_to_cpu(rn->in.nid[i]);
993 			if (child_nid == 0) {
994 				child_nofs += NIDS_PER_BLOCK + 1;
995 				continue;
996 			}
997 			rdn.nid = child_nid;
998 			ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
999 			if (ret == (NIDS_PER_BLOCK + 1)) {
1000 				if (set_nid(page, i, 0, false))
1001 					dn->node_changed = true;
1002 				child_nofs += ret;
1003 			} else if (ret < 0 && ret != -ENOENT) {
1004 				goto out_err;
1005 			}
1006 		}
1007 		freed = child_nofs;
1008 	}
1009 
1010 	if (!ofs) {
1011 		/* remove current indirect node */
1012 		dn->node_page = page;
1013 		ret = truncate_node(dn);
1014 		if (ret)
1015 			goto out_err;
1016 		freed++;
1017 	} else {
1018 		f2fs_put_page(page, 1);
1019 	}
1020 	trace_f2fs_truncate_nodes_exit(dn->inode, freed);
1021 	return freed;
1022 
1023 out_err:
1024 	f2fs_put_page(page, 1);
1025 	trace_f2fs_truncate_nodes_exit(dn->inode, ret);
1026 	return ret;
1027 }
1028 
1029 static int truncate_partial_nodes(struct dnode_of_data *dn,
1030 			struct f2fs_inode *ri, int *offset, int depth)
1031 {
1032 	struct page *pages[2];
1033 	nid_t nid[3];
1034 	nid_t child_nid;
1035 	int err = 0;
1036 	int i;
1037 	int idx = depth - 2;
1038 
1039 	nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1040 	if (!nid[0])
1041 		return 0;
1042 
1043 	/* get indirect nodes in the path */
1044 	for (i = 0; i < idx + 1; i++) {
1045 		/* reference count'll be increased */
1046 		pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
1047 		if (IS_ERR(pages[i])) {
1048 			err = PTR_ERR(pages[i]);
1049 			idx = i - 1;
1050 			goto fail;
1051 		}
1052 		nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1053 	}
1054 
1055 	f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1056 
1057 	/* free direct nodes linked to a partial indirect node */
1058 	for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1059 		child_nid = get_nid(pages[idx], i, false);
1060 		if (!child_nid)
1061 			continue;
1062 		dn->nid = child_nid;
1063 		err = truncate_dnode(dn);
1064 		if (err < 0)
1065 			goto fail;
1066 		if (set_nid(pages[idx], i, 0, false))
1067 			dn->node_changed = true;
1068 	}
1069 
1070 	if (offset[idx + 1] == 0) {
1071 		dn->node_page = pages[idx];
1072 		dn->nid = nid[idx];
1073 		err = truncate_node(dn);
1074 		if (err)
1075 			goto fail;
1076 	} else {
1077 		f2fs_put_page(pages[idx], 1);
1078 	}
1079 	offset[idx]++;
1080 	offset[idx + 1] = 0;
1081 	idx--;
1082 fail:
1083 	for (i = idx; i >= 0; i--)
1084 		f2fs_put_page(pages[i], 1);
1085 
1086 	trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1087 
1088 	return err;
1089 }
1090 
1091 /*
1092  * All the block addresses of data and nodes should be nullified.
1093  */
1094 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1095 {
1096 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1097 	int err = 0, cont = 1;
1098 	int level, offset[4], noffset[4];
1099 	unsigned int nofs = 0;
1100 	struct f2fs_inode *ri;
1101 	struct dnode_of_data dn;
1102 	struct page *page;
1103 
1104 	trace_f2fs_truncate_inode_blocks_enter(inode, from);
1105 
1106 	level = get_node_path(inode, from, offset, noffset);
1107 	if (level < 0) {
1108 		trace_f2fs_truncate_inode_blocks_exit(inode, level);
1109 		return level;
1110 	}
1111 
1112 	page = f2fs_get_node_page(sbi, inode->i_ino);
1113 	if (IS_ERR(page)) {
1114 		trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1115 		return PTR_ERR(page);
1116 	}
1117 
1118 	set_new_dnode(&dn, inode, page, NULL, 0);
1119 	unlock_page(page);
1120 
1121 	ri = F2FS_INODE(page);
1122 	switch (level) {
1123 	case 0:
1124 	case 1:
1125 		nofs = noffset[1];
1126 		break;
1127 	case 2:
1128 		nofs = noffset[1];
1129 		if (!offset[level - 1])
1130 			goto skip_partial;
1131 		err = truncate_partial_nodes(&dn, ri, offset, level);
1132 		if (err < 0 && err != -ENOENT)
1133 			goto fail;
1134 		nofs += 1 + NIDS_PER_BLOCK;
1135 		break;
1136 	case 3:
1137 		nofs = 5 + 2 * NIDS_PER_BLOCK;
1138 		if (!offset[level - 1])
1139 			goto skip_partial;
1140 		err = truncate_partial_nodes(&dn, ri, offset, level);
1141 		if (err < 0 && err != -ENOENT)
1142 			goto fail;
1143 		break;
1144 	default:
1145 		BUG();
1146 	}
1147 
1148 skip_partial:
1149 	while (cont) {
1150 		dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1151 		switch (offset[0]) {
1152 		case NODE_DIR1_BLOCK:
1153 		case NODE_DIR2_BLOCK:
1154 			err = truncate_dnode(&dn);
1155 			break;
1156 
1157 		case NODE_IND1_BLOCK:
1158 		case NODE_IND2_BLOCK:
1159 			err = truncate_nodes(&dn, nofs, offset[1], 2);
1160 			break;
1161 
1162 		case NODE_DIND_BLOCK:
1163 			err = truncate_nodes(&dn, nofs, offset[1], 3);
1164 			cont = 0;
1165 			break;
1166 
1167 		default:
1168 			BUG();
1169 		}
1170 		if (err < 0 && err != -ENOENT)
1171 			goto fail;
1172 		if (offset[1] == 0 &&
1173 				ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1174 			lock_page(page);
1175 			BUG_ON(page->mapping != NODE_MAPPING(sbi));
1176 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1177 			ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1178 			set_page_dirty(page);
1179 			unlock_page(page);
1180 		}
1181 		offset[1] = 0;
1182 		offset[0]++;
1183 		nofs += err;
1184 	}
1185 fail:
1186 	f2fs_put_page(page, 0);
1187 	trace_f2fs_truncate_inode_blocks_exit(inode, err);
1188 	return err > 0 ? 0 : err;
1189 }
1190 
1191 /* caller must lock inode page */
1192 int f2fs_truncate_xattr_node(struct inode *inode)
1193 {
1194 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1195 	nid_t nid = F2FS_I(inode)->i_xattr_nid;
1196 	struct dnode_of_data dn;
1197 	struct page *npage;
1198 	int err;
1199 
1200 	if (!nid)
1201 		return 0;
1202 
1203 	npage = f2fs_get_node_page(sbi, nid);
1204 	if (IS_ERR(npage))
1205 		return PTR_ERR(npage);
1206 
1207 	set_new_dnode(&dn, inode, NULL, npage, nid);
1208 	err = truncate_node(&dn);
1209 	if (err) {
1210 		f2fs_put_page(npage, 1);
1211 		return err;
1212 	}
1213 
1214 	f2fs_i_xnid_write(inode, 0);
1215 
1216 	return 0;
1217 }
1218 
1219 /*
1220  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1221  * f2fs_unlock_op().
1222  */
1223 int f2fs_remove_inode_page(struct inode *inode)
1224 {
1225 	struct dnode_of_data dn;
1226 	int err;
1227 
1228 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1229 	err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1230 	if (err)
1231 		return err;
1232 
1233 	err = f2fs_truncate_xattr_node(inode);
1234 	if (err) {
1235 		f2fs_put_dnode(&dn);
1236 		return err;
1237 	}
1238 
1239 	/* remove potential inline_data blocks */
1240 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1241 				S_ISLNK(inode->i_mode))
1242 		f2fs_truncate_data_blocks_range(&dn, 1);
1243 
1244 	/* 0 is possible, after f2fs_new_inode() has failed */
1245 	if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1246 		f2fs_put_dnode(&dn);
1247 		return -EIO;
1248 	}
1249 
1250 	if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1251 		f2fs_warn(F2FS_I_SB(inode),
1252 			"f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1253 			inode->i_ino, (unsigned long long)inode->i_blocks);
1254 		set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1255 	}
1256 
1257 	/* will put inode & node pages */
1258 	err = truncate_node(&dn);
1259 	if (err) {
1260 		f2fs_put_dnode(&dn);
1261 		return err;
1262 	}
1263 	return 0;
1264 }
1265 
1266 struct page *f2fs_new_inode_page(struct inode *inode)
1267 {
1268 	struct dnode_of_data dn;
1269 
1270 	/* allocate inode page for new inode */
1271 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1272 
1273 	/* caller should f2fs_put_page(page, 1); */
1274 	return f2fs_new_node_page(&dn, 0);
1275 }
1276 
1277 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1278 {
1279 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1280 	struct node_info new_ni;
1281 	struct page *page;
1282 	int err;
1283 
1284 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1285 		return ERR_PTR(-EPERM);
1286 
1287 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1288 	if (!page)
1289 		return ERR_PTR(-ENOMEM);
1290 
1291 	if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1292 		goto fail;
1293 
1294 #ifdef CONFIG_F2FS_CHECK_FS
1295 	err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false);
1296 	if (err) {
1297 		dec_valid_node_count(sbi, dn->inode, !ofs);
1298 		goto fail;
1299 	}
1300 	if (unlikely(new_ni.blk_addr != NULL_ADDR)) {
1301 		err = -EFSCORRUPTED;
1302 		set_sbi_flag(sbi, SBI_NEED_FSCK);
1303 		f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1304 		goto fail;
1305 	}
1306 #endif
1307 	new_ni.nid = dn->nid;
1308 	new_ni.ino = dn->inode->i_ino;
1309 	new_ni.blk_addr = NULL_ADDR;
1310 	new_ni.flag = 0;
1311 	new_ni.version = 0;
1312 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1313 
1314 	f2fs_wait_on_page_writeback(page, NODE, true, true);
1315 	fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1316 	set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1317 	if (!PageUptodate(page))
1318 		SetPageUptodate(page);
1319 	if (set_page_dirty(page))
1320 		dn->node_changed = true;
1321 
1322 	if (f2fs_has_xattr_block(ofs))
1323 		f2fs_i_xnid_write(dn->inode, dn->nid);
1324 
1325 	if (ofs == 0)
1326 		inc_valid_inode_count(sbi);
1327 	return page;
1328 
1329 fail:
1330 	clear_node_page_dirty(page);
1331 	f2fs_put_page(page, 1);
1332 	return ERR_PTR(err);
1333 }
1334 
1335 /*
1336  * Caller should do after getting the following values.
1337  * 0: f2fs_put_page(page, 0)
1338  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1339  */
1340 static int read_node_page(struct page *page, blk_opf_t op_flags)
1341 {
1342 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1343 	struct node_info ni;
1344 	struct f2fs_io_info fio = {
1345 		.sbi = sbi,
1346 		.type = NODE,
1347 		.op = REQ_OP_READ,
1348 		.op_flags = op_flags,
1349 		.page = page,
1350 		.encrypted_page = NULL,
1351 	};
1352 	int err;
1353 
1354 	if (PageUptodate(page)) {
1355 		if (!f2fs_inode_chksum_verify(sbi, page)) {
1356 			ClearPageUptodate(page);
1357 			return -EFSBADCRC;
1358 		}
1359 		return LOCKED_PAGE;
1360 	}
1361 
1362 	err = f2fs_get_node_info(sbi, page->index, &ni, false);
1363 	if (err)
1364 		return err;
1365 
1366 	/* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1367 	if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR)) {
1368 		ClearPageUptodate(page);
1369 		return -ENOENT;
1370 	}
1371 
1372 	fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1373 
1374 	err = f2fs_submit_page_bio(&fio);
1375 
1376 	if (!err)
1377 		f2fs_update_iostat(sbi, NULL, FS_NODE_READ_IO, F2FS_BLKSIZE);
1378 
1379 	return err;
1380 }
1381 
1382 /*
1383  * Readahead a node page
1384  */
1385 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1386 {
1387 	struct page *apage;
1388 	int err;
1389 
1390 	if (!nid)
1391 		return;
1392 	if (f2fs_check_nid_range(sbi, nid))
1393 		return;
1394 
1395 	apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1396 	if (apage)
1397 		return;
1398 
1399 	apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1400 	if (!apage)
1401 		return;
1402 
1403 	err = read_node_page(apage, REQ_RAHEAD);
1404 	f2fs_put_page(apage, err ? 1 : 0);
1405 }
1406 
1407 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1408 					struct page *parent, int start)
1409 {
1410 	struct page *page;
1411 	int err;
1412 
1413 	if (!nid)
1414 		return ERR_PTR(-ENOENT);
1415 	if (f2fs_check_nid_range(sbi, nid))
1416 		return ERR_PTR(-EINVAL);
1417 repeat:
1418 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1419 	if (!page)
1420 		return ERR_PTR(-ENOMEM);
1421 
1422 	err = read_node_page(page, 0);
1423 	if (err < 0) {
1424 		goto out_put_err;
1425 	} else if (err == LOCKED_PAGE) {
1426 		err = 0;
1427 		goto page_hit;
1428 	}
1429 
1430 	if (parent)
1431 		f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1432 
1433 	lock_page(page);
1434 
1435 	if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1436 		f2fs_put_page(page, 1);
1437 		goto repeat;
1438 	}
1439 
1440 	if (unlikely(!PageUptodate(page))) {
1441 		err = -EIO;
1442 		goto out_err;
1443 	}
1444 
1445 	if (!f2fs_inode_chksum_verify(sbi, page)) {
1446 		err = -EFSBADCRC;
1447 		goto out_err;
1448 	}
1449 page_hit:
1450 	if (likely(nid == nid_of_node(page)))
1451 		return page;
1452 
1453 	f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1454 			  nid, nid_of_node(page), ino_of_node(page),
1455 			  ofs_of_node(page), cpver_of_node(page),
1456 			  next_blkaddr_of_node(page));
1457 	set_sbi_flag(sbi, SBI_NEED_FSCK);
1458 	err = -EINVAL;
1459 out_err:
1460 	ClearPageUptodate(page);
1461 out_put_err:
1462 	/* ENOENT comes from read_node_page which is not an error. */
1463 	if (err != -ENOENT)
1464 		f2fs_handle_page_eio(sbi, page->index, NODE);
1465 	f2fs_put_page(page, 1);
1466 	return ERR_PTR(err);
1467 }
1468 
1469 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1470 {
1471 	return __get_node_page(sbi, nid, NULL, 0);
1472 }
1473 
1474 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1475 {
1476 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1477 	nid_t nid = get_nid(parent, start, false);
1478 
1479 	return __get_node_page(sbi, nid, parent, start);
1480 }
1481 
1482 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1483 {
1484 	struct inode *inode;
1485 	struct page *page;
1486 	int ret;
1487 
1488 	/* should flush inline_data before evict_inode */
1489 	inode = ilookup(sbi->sb, ino);
1490 	if (!inode)
1491 		return;
1492 
1493 	page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1494 					FGP_LOCK|FGP_NOWAIT, 0);
1495 	if (!page)
1496 		goto iput_out;
1497 
1498 	if (!PageUptodate(page))
1499 		goto page_out;
1500 
1501 	if (!PageDirty(page))
1502 		goto page_out;
1503 
1504 	if (!clear_page_dirty_for_io(page))
1505 		goto page_out;
1506 
1507 	ret = f2fs_write_inline_data(inode, page);
1508 	inode_dec_dirty_pages(inode);
1509 	f2fs_remove_dirty_inode(inode);
1510 	if (ret)
1511 		set_page_dirty(page);
1512 page_out:
1513 	f2fs_put_page(page, 1);
1514 iput_out:
1515 	iput(inode);
1516 }
1517 
1518 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1519 {
1520 	pgoff_t index;
1521 	struct folio_batch fbatch;
1522 	struct page *last_page = NULL;
1523 	int nr_folios;
1524 
1525 	folio_batch_init(&fbatch);
1526 	index = 0;
1527 
1528 	while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1529 					(pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1530 					&fbatch))) {
1531 		int i;
1532 
1533 		for (i = 0; i < nr_folios; i++) {
1534 			struct page *page = &fbatch.folios[i]->page;
1535 
1536 			if (unlikely(f2fs_cp_error(sbi))) {
1537 				f2fs_put_page(last_page, 0);
1538 				folio_batch_release(&fbatch);
1539 				return ERR_PTR(-EIO);
1540 			}
1541 
1542 			if (!IS_DNODE(page) || !is_cold_node(page))
1543 				continue;
1544 			if (ino_of_node(page) != ino)
1545 				continue;
1546 
1547 			lock_page(page);
1548 
1549 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1550 continue_unlock:
1551 				unlock_page(page);
1552 				continue;
1553 			}
1554 			if (ino_of_node(page) != ino)
1555 				goto continue_unlock;
1556 
1557 			if (!PageDirty(page)) {
1558 				/* someone wrote it for us */
1559 				goto continue_unlock;
1560 			}
1561 
1562 			if (last_page)
1563 				f2fs_put_page(last_page, 0);
1564 
1565 			get_page(page);
1566 			last_page = page;
1567 			unlock_page(page);
1568 		}
1569 		folio_batch_release(&fbatch);
1570 		cond_resched();
1571 	}
1572 	return last_page;
1573 }
1574 
1575 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1576 				struct writeback_control *wbc, bool do_balance,
1577 				enum iostat_type io_type, unsigned int *seq_id)
1578 {
1579 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1580 	nid_t nid;
1581 	struct node_info ni;
1582 	struct f2fs_io_info fio = {
1583 		.sbi = sbi,
1584 		.ino = ino_of_node(page),
1585 		.type = NODE,
1586 		.op = REQ_OP_WRITE,
1587 		.op_flags = wbc_to_write_flags(wbc),
1588 		.page = page,
1589 		.encrypted_page = NULL,
1590 		.submitted = 0,
1591 		.io_type = io_type,
1592 		.io_wbc = wbc,
1593 	};
1594 	unsigned int seq;
1595 
1596 	trace_f2fs_writepage(page, NODE);
1597 
1598 	if (unlikely(f2fs_cp_error(sbi))) {
1599 		ClearPageUptodate(page);
1600 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1601 		unlock_page(page);
1602 		return 0;
1603 	}
1604 
1605 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1606 		goto redirty_out;
1607 
1608 	if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1609 			wbc->sync_mode == WB_SYNC_NONE &&
1610 			IS_DNODE(page) && is_cold_node(page))
1611 		goto redirty_out;
1612 
1613 	/* get old block addr of this node page */
1614 	nid = nid_of_node(page);
1615 	f2fs_bug_on(sbi, page->index != nid);
1616 
1617 	if (f2fs_get_node_info(sbi, nid, &ni, !do_balance))
1618 		goto redirty_out;
1619 
1620 	if (wbc->for_reclaim) {
1621 		if (!f2fs_down_read_trylock(&sbi->node_write))
1622 			goto redirty_out;
1623 	} else {
1624 		f2fs_down_read(&sbi->node_write);
1625 	}
1626 
1627 	/* This page is already truncated */
1628 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1629 		ClearPageUptodate(page);
1630 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1631 		f2fs_up_read(&sbi->node_write);
1632 		unlock_page(page);
1633 		return 0;
1634 	}
1635 
1636 	if (__is_valid_data_blkaddr(ni.blk_addr) &&
1637 		!f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1638 					DATA_GENERIC_ENHANCE)) {
1639 		f2fs_up_read(&sbi->node_write);
1640 		goto redirty_out;
1641 	}
1642 
1643 	if (atomic && !test_opt(sbi, NOBARRIER) && !f2fs_sb_has_blkzoned(sbi))
1644 		fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1645 
1646 	/* should add to global list before clearing PAGECACHE status */
1647 	if (f2fs_in_warm_node_list(sbi, page)) {
1648 		seq = f2fs_add_fsync_node_entry(sbi, page);
1649 		if (seq_id)
1650 			*seq_id = seq;
1651 	}
1652 
1653 	set_page_writeback(page);
1654 
1655 	fio.old_blkaddr = ni.blk_addr;
1656 	f2fs_do_write_node_page(nid, &fio);
1657 	set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1658 	dec_page_count(sbi, F2FS_DIRTY_NODES);
1659 	f2fs_up_read(&sbi->node_write);
1660 
1661 	if (wbc->for_reclaim) {
1662 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1663 		submitted = NULL;
1664 	}
1665 
1666 	unlock_page(page);
1667 
1668 	if (unlikely(f2fs_cp_error(sbi))) {
1669 		f2fs_submit_merged_write(sbi, NODE);
1670 		submitted = NULL;
1671 	}
1672 	if (submitted)
1673 		*submitted = fio.submitted;
1674 
1675 	if (do_balance)
1676 		f2fs_balance_fs(sbi, false);
1677 	return 0;
1678 
1679 redirty_out:
1680 	redirty_page_for_writepage(wbc, page);
1681 	return AOP_WRITEPAGE_ACTIVATE;
1682 }
1683 
1684 int f2fs_move_node_page(struct page *node_page, int gc_type)
1685 {
1686 	int err = 0;
1687 
1688 	if (gc_type == FG_GC) {
1689 		struct writeback_control wbc = {
1690 			.sync_mode = WB_SYNC_ALL,
1691 			.nr_to_write = 1,
1692 			.for_reclaim = 0,
1693 		};
1694 
1695 		f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1696 
1697 		set_page_dirty(node_page);
1698 
1699 		if (!clear_page_dirty_for_io(node_page)) {
1700 			err = -EAGAIN;
1701 			goto out_page;
1702 		}
1703 
1704 		if (__write_node_page(node_page, false, NULL,
1705 					&wbc, false, FS_GC_NODE_IO, NULL)) {
1706 			err = -EAGAIN;
1707 			unlock_page(node_page);
1708 		}
1709 		goto release_page;
1710 	} else {
1711 		/* set page dirty and write it */
1712 		if (!PageWriteback(node_page))
1713 			set_page_dirty(node_page);
1714 	}
1715 out_page:
1716 	unlock_page(node_page);
1717 release_page:
1718 	f2fs_put_page(node_page, 0);
1719 	return err;
1720 }
1721 
1722 static int f2fs_write_node_page(struct page *page,
1723 				struct writeback_control *wbc)
1724 {
1725 	return __write_node_page(page, false, NULL, wbc, false,
1726 						FS_NODE_IO, NULL);
1727 }
1728 
1729 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1730 			struct writeback_control *wbc, bool atomic,
1731 			unsigned int *seq_id)
1732 {
1733 	pgoff_t index;
1734 	struct folio_batch fbatch;
1735 	int ret = 0;
1736 	struct page *last_page = NULL;
1737 	bool marked = false;
1738 	nid_t ino = inode->i_ino;
1739 	int nr_folios;
1740 	int nwritten = 0;
1741 
1742 	if (atomic) {
1743 		last_page = last_fsync_dnode(sbi, ino);
1744 		if (IS_ERR_OR_NULL(last_page))
1745 			return PTR_ERR_OR_ZERO(last_page);
1746 	}
1747 retry:
1748 	folio_batch_init(&fbatch);
1749 	index = 0;
1750 
1751 	while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1752 					(pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1753 					&fbatch))) {
1754 		int i;
1755 
1756 		for (i = 0; i < nr_folios; i++) {
1757 			struct page *page = &fbatch.folios[i]->page;
1758 			bool submitted = false;
1759 
1760 			if (unlikely(f2fs_cp_error(sbi))) {
1761 				f2fs_put_page(last_page, 0);
1762 				folio_batch_release(&fbatch);
1763 				ret = -EIO;
1764 				goto out;
1765 			}
1766 
1767 			if (!IS_DNODE(page) || !is_cold_node(page))
1768 				continue;
1769 			if (ino_of_node(page) != ino)
1770 				continue;
1771 
1772 			lock_page(page);
1773 
1774 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1775 continue_unlock:
1776 				unlock_page(page);
1777 				continue;
1778 			}
1779 			if (ino_of_node(page) != ino)
1780 				goto continue_unlock;
1781 
1782 			if (!PageDirty(page) && page != last_page) {
1783 				/* someone wrote it for us */
1784 				goto continue_unlock;
1785 			}
1786 
1787 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1788 
1789 			set_fsync_mark(page, 0);
1790 			set_dentry_mark(page, 0);
1791 
1792 			if (!atomic || page == last_page) {
1793 				set_fsync_mark(page, 1);
1794 				percpu_counter_inc(&sbi->rf_node_block_count);
1795 				if (IS_INODE(page)) {
1796 					if (is_inode_flag_set(inode,
1797 								FI_DIRTY_INODE))
1798 						f2fs_update_inode(inode, page);
1799 					set_dentry_mark(page,
1800 						f2fs_need_dentry_mark(sbi, ino));
1801 				}
1802 				/* may be written by other thread */
1803 				if (!PageDirty(page))
1804 					set_page_dirty(page);
1805 			}
1806 
1807 			if (!clear_page_dirty_for_io(page))
1808 				goto continue_unlock;
1809 
1810 			ret = __write_node_page(page, atomic &&
1811 						page == last_page,
1812 						&submitted, wbc, true,
1813 						FS_NODE_IO, seq_id);
1814 			if (ret) {
1815 				unlock_page(page);
1816 				f2fs_put_page(last_page, 0);
1817 				break;
1818 			} else if (submitted) {
1819 				nwritten++;
1820 			}
1821 
1822 			if (page == last_page) {
1823 				f2fs_put_page(page, 0);
1824 				marked = true;
1825 				break;
1826 			}
1827 		}
1828 		folio_batch_release(&fbatch);
1829 		cond_resched();
1830 
1831 		if (ret || marked)
1832 			break;
1833 	}
1834 	if (!ret && atomic && !marked) {
1835 		f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1836 			   ino, last_page->index);
1837 		lock_page(last_page);
1838 		f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1839 		set_page_dirty(last_page);
1840 		unlock_page(last_page);
1841 		goto retry;
1842 	}
1843 out:
1844 	if (nwritten)
1845 		f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1846 	return ret ? -EIO : 0;
1847 }
1848 
1849 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1850 {
1851 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1852 	bool clean;
1853 
1854 	if (inode->i_ino != ino)
1855 		return 0;
1856 
1857 	if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1858 		return 0;
1859 
1860 	spin_lock(&sbi->inode_lock[DIRTY_META]);
1861 	clean = list_empty(&F2FS_I(inode)->gdirty_list);
1862 	spin_unlock(&sbi->inode_lock[DIRTY_META]);
1863 
1864 	if (clean)
1865 		return 0;
1866 
1867 	inode = igrab(inode);
1868 	if (!inode)
1869 		return 0;
1870 	return 1;
1871 }
1872 
1873 static bool flush_dirty_inode(struct page *page)
1874 {
1875 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1876 	struct inode *inode;
1877 	nid_t ino = ino_of_node(page);
1878 
1879 	inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1880 	if (!inode)
1881 		return false;
1882 
1883 	f2fs_update_inode(inode, page);
1884 	unlock_page(page);
1885 
1886 	iput(inode);
1887 	return true;
1888 }
1889 
1890 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1891 {
1892 	pgoff_t index = 0;
1893 	struct folio_batch fbatch;
1894 	int nr_folios;
1895 
1896 	folio_batch_init(&fbatch);
1897 
1898 	while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1899 					(pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1900 					&fbatch))) {
1901 		int i;
1902 
1903 		for (i = 0; i < nr_folios; i++) {
1904 			struct page *page = &fbatch.folios[i]->page;
1905 
1906 			if (!IS_DNODE(page))
1907 				continue;
1908 
1909 			lock_page(page);
1910 
1911 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1912 continue_unlock:
1913 				unlock_page(page);
1914 				continue;
1915 			}
1916 
1917 			if (!PageDirty(page)) {
1918 				/* someone wrote it for us */
1919 				goto continue_unlock;
1920 			}
1921 
1922 			/* flush inline_data, if it's async context. */
1923 			if (page_private_inline(page)) {
1924 				clear_page_private_inline(page);
1925 				unlock_page(page);
1926 				flush_inline_data(sbi, ino_of_node(page));
1927 				continue;
1928 			}
1929 			unlock_page(page);
1930 		}
1931 		folio_batch_release(&fbatch);
1932 		cond_resched();
1933 	}
1934 }
1935 
1936 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1937 				struct writeback_control *wbc,
1938 				bool do_balance, enum iostat_type io_type)
1939 {
1940 	pgoff_t index;
1941 	struct folio_batch fbatch;
1942 	int step = 0;
1943 	int nwritten = 0;
1944 	int ret = 0;
1945 	int nr_folios, done = 0;
1946 
1947 	folio_batch_init(&fbatch);
1948 
1949 next_step:
1950 	index = 0;
1951 
1952 	while (!done && (nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi),
1953 				&index, (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1954 				&fbatch))) {
1955 		int i;
1956 
1957 		for (i = 0; i < nr_folios; i++) {
1958 			struct page *page = &fbatch.folios[i]->page;
1959 			bool submitted = false;
1960 
1961 			/* give a priority to WB_SYNC threads */
1962 			if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1963 					wbc->sync_mode == WB_SYNC_NONE) {
1964 				done = 1;
1965 				break;
1966 			}
1967 
1968 			/*
1969 			 * flushing sequence with step:
1970 			 * 0. indirect nodes
1971 			 * 1. dentry dnodes
1972 			 * 2. file dnodes
1973 			 */
1974 			if (step == 0 && IS_DNODE(page))
1975 				continue;
1976 			if (step == 1 && (!IS_DNODE(page) ||
1977 						is_cold_node(page)))
1978 				continue;
1979 			if (step == 2 && (!IS_DNODE(page) ||
1980 						!is_cold_node(page)))
1981 				continue;
1982 lock_node:
1983 			if (wbc->sync_mode == WB_SYNC_ALL)
1984 				lock_page(page);
1985 			else if (!trylock_page(page))
1986 				continue;
1987 
1988 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1989 continue_unlock:
1990 				unlock_page(page);
1991 				continue;
1992 			}
1993 
1994 			if (!PageDirty(page)) {
1995 				/* someone wrote it for us */
1996 				goto continue_unlock;
1997 			}
1998 
1999 			/* flush inline_data/inode, if it's async context. */
2000 			if (!do_balance)
2001 				goto write_node;
2002 
2003 			/* flush inline_data */
2004 			if (page_private_inline(page)) {
2005 				clear_page_private_inline(page);
2006 				unlock_page(page);
2007 				flush_inline_data(sbi, ino_of_node(page));
2008 				goto lock_node;
2009 			}
2010 
2011 			/* flush dirty inode */
2012 			if (IS_INODE(page) && flush_dirty_inode(page))
2013 				goto lock_node;
2014 write_node:
2015 			f2fs_wait_on_page_writeback(page, NODE, true, true);
2016 
2017 			if (!clear_page_dirty_for_io(page))
2018 				goto continue_unlock;
2019 
2020 			set_fsync_mark(page, 0);
2021 			set_dentry_mark(page, 0);
2022 
2023 			ret = __write_node_page(page, false, &submitted,
2024 						wbc, do_balance, io_type, NULL);
2025 			if (ret)
2026 				unlock_page(page);
2027 			else if (submitted)
2028 				nwritten++;
2029 
2030 			if (--wbc->nr_to_write == 0)
2031 				break;
2032 		}
2033 		folio_batch_release(&fbatch);
2034 		cond_resched();
2035 
2036 		if (wbc->nr_to_write == 0) {
2037 			step = 2;
2038 			break;
2039 		}
2040 	}
2041 
2042 	if (step < 2) {
2043 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2044 				wbc->sync_mode == WB_SYNC_NONE && step == 1)
2045 			goto out;
2046 		step++;
2047 		goto next_step;
2048 	}
2049 out:
2050 	if (nwritten)
2051 		f2fs_submit_merged_write(sbi, NODE);
2052 
2053 	if (unlikely(f2fs_cp_error(sbi)))
2054 		return -EIO;
2055 	return ret;
2056 }
2057 
2058 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2059 						unsigned int seq_id)
2060 {
2061 	struct fsync_node_entry *fn;
2062 	struct page *page;
2063 	struct list_head *head = &sbi->fsync_node_list;
2064 	unsigned long flags;
2065 	unsigned int cur_seq_id = 0;
2066 	int ret2, ret = 0;
2067 
2068 	while (seq_id && cur_seq_id < seq_id) {
2069 		spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2070 		if (list_empty(head)) {
2071 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2072 			break;
2073 		}
2074 		fn = list_first_entry(head, struct fsync_node_entry, list);
2075 		if (fn->seq_id > seq_id) {
2076 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2077 			break;
2078 		}
2079 		cur_seq_id = fn->seq_id;
2080 		page = fn->page;
2081 		get_page(page);
2082 		spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2083 
2084 		f2fs_wait_on_page_writeback(page, NODE, true, false);
2085 
2086 		put_page(page);
2087 
2088 		if (ret)
2089 			break;
2090 	}
2091 
2092 	ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2093 	if (!ret)
2094 		ret = ret2;
2095 
2096 	return ret;
2097 }
2098 
2099 static int f2fs_write_node_pages(struct address_space *mapping,
2100 			    struct writeback_control *wbc)
2101 {
2102 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2103 	struct blk_plug plug;
2104 	long diff;
2105 
2106 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2107 		goto skip_write;
2108 
2109 	/* balancing f2fs's metadata in background */
2110 	f2fs_balance_fs_bg(sbi, true);
2111 
2112 	/* collect a number of dirty node pages and write together */
2113 	if (wbc->sync_mode != WB_SYNC_ALL &&
2114 			get_pages(sbi, F2FS_DIRTY_NODES) <
2115 					nr_pages_to_skip(sbi, NODE))
2116 		goto skip_write;
2117 
2118 	if (wbc->sync_mode == WB_SYNC_ALL)
2119 		atomic_inc(&sbi->wb_sync_req[NODE]);
2120 	else if (atomic_read(&sbi->wb_sync_req[NODE])) {
2121 		/* to avoid potential deadlock */
2122 		if (current->plug)
2123 			blk_finish_plug(current->plug);
2124 		goto skip_write;
2125 	}
2126 
2127 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2128 
2129 	diff = nr_pages_to_write(sbi, NODE, wbc);
2130 	blk_start_plug(&plug);
2131 	f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2132 	blk_finish_plug(&plug);
2133 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2134 
2135 	if (wbc->sync_mode == WB_SYNC_ALL)
2136 		atomic_dec(&sbi->wb_sync_req[NODE]);
2137 	return 0;
2138 
2139 skip_write:
2140 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2141 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2142 	return 0;
2143 }
2144 
2145 static bool f2fs_dirty_node_folio(struct address_space *mapping,
2146 		struct folio *folio)
2147 {
2148 	trace_f2fs_set_page_dirty(&folio->page, NODE);
2149 
2150 	if (!folio_test_uptodate(folio))
2151 		folio_mark_uptodate(folio);
2152 #ifdef CONFIG_F2FS_CHECK_FS
2153 	if (IS_INODE(&folio->page))
2154 		f2fs_inode_chksum_set(F2FS_M_SB(mapping), &folio->page);
2155 #endif
2156 	if (filemap_dirty_folio(mapping, folio)) {
2157 		inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
2158 		set_page_private_reference(&folio->page);
2159 		return true;
2160 	}
2161 	return false;
2162 }
2163 
2164 /*
2165  * Structure of the f2fs node operations
2166  */
2167 const struct address_space_operations f2fs_node_aops = {
2168 	.writepage	= f2fs_write_node_page,
2169 	.writepages	= f2fs_write_node_pages,
2170 	.dirty_folio	= f2fs_dirty_node_folio,
2171 	.invalidate_folio = f2fs_invalidate_folio,
2172 	.release_folio	= f2fs_release_folio,
2173 	.migrate_folio	= filemap_migrate_folio,
2174 };
2175 
2176 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2177 						nid_t n)
2178 {
2179 	return radix_tree_lookup(&nm_i->free_nid_root, n);
2180 }
2181 
2182 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2183 				struct free_nid *i)
2184 {
2185 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2186 	int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2187 
2188 	if (err)
2189 		return err;
2190 
2191 	nm_i->nid_cnt[FREE_NID]++;
2192 	list_add_tail(&i->list, &nm_i->free_nid_list);
2193 	return 0;
2194 }
2195 
2196 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2197 			struct free_nid *i, enum nid_state state)
2198 {
2199 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2200 
2201 	f2fs_bug_on(sbi, state != i->state);
2202 	nm_i->nid_cnt[state]--;
2203 	if (state == FREE_NID)
2204 		list_del(&i->list);
2205 	radix_tree_delete(&nm_i->free_nid_root, i->nid);
2206 }
2207 
2208 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2209 			enum nid_state org_state, enum nid_state dst_state)
2210 {
2211 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2212 
2213 	f2fs_bug_on(sbi, org_state != i->state);
2214 	i->state = dst_state;
2215 	nm_i->nid_cnt[org_state]--;
2216 	nm_i->nid_cnt[dst_state]++;
2217 
2218 	switch (dst_state) {
2219 	case PREALLOC_NID:
2220 		list_del(&i->list);
2221 		break;
2222 	case FREE_NID:
2223 		list_add_tail(&i->list, &nm_i->free_nid_list);
2224 		break;
2225 	default:
2226 		BUG_ON(1);
2227 	}
2228 }
2229 
2230 bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi)
2231 {
2232 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2233 	unsigned int i;
2234 	bool ret = true;
2235 
2236 	f2fs_down_read(&nm_i->nat_tree_lock);
2237 	for (i = 0; i < nm_i->nat_blocks; i++) {
2238 		if (!test_bit_le(i, nm_i->nat_block_bitmap)) {
2239 			ret = false;
2240 			break;
2241 		}
2242 	}
2243 	f2fs_up_read(&nm_i->nat_tree_lock);
2244 
2245 	return ret;
2246 }
2247 
2248 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2249 							bool set, bool build)
2250 {
2251 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2252 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2253 	unsigned int nid_ofs = nid - START_NID(nid);
2254 
2255 	if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2256 		return;
2257 
2258 	if (set) {
2259 		if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2260 			return;
2261 		__set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2262 		nm_i->free_nid_count[nat_ofs]++;
2263 	} else {
2264 		if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2265 			return;
2266 		__clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2267 		if (!build)
2268 			nm_i->free_nid_count[nat_ofs]--;
2269 	}
2270 }
2271 
2272 /* return if the nid is recognized as free */
2273 static bool add_free_nid(struct f2fs_sb_info *sbi,
2274 				nid_t nid, bool build, bool update)
2275 {
2276 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2277 	struct free_nid *i, *e;
2278 	struct nat_entry *ne;
2279 	int err = -EINVAL;
2280 	bool ret = false;
2281 
2282 	/* 0 nid should not be used */
2283 	if (unlikely(nid == 0))
2284 		return false;
2285 
2286 	if (unlikely(f2fs_check_nid_range(sbi, nid)))
2287 		return false;
2288 
2289 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL);
2290 	i->nid = nid;
2291 	i->state = FREE_NID;
2292 
2293 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2294 
2295 	spin_lock(&nm_i->nid_list_lock);
2296 
2297 	if (build) {
2298 		/*
2299 		 *   Thread A             Thread B
2300 		 *  - f2fs_create
2301 		 *   - f2fs_new_inode
2302 		 *    - f2fs_alloc_nid
2303 		 *     - __insert_nid_to_list(PREALLOC_NID)
2304 		 *                     - f2fs_balance_fs_bg
2305 		 *                      - f2fs_build_free_nids
2306 		 *                       - __f2fs_build_free_nids
2307 		 *                        - scan_nat_page
2308 		 *                         - add_free_nid
2309 		 *                          - __lookup_nat_cache
2310 		 *  - f2fs_add_link
2311 		 *   - f2fs_init_inode_metadata
2312 		 *    - f2fs_new_inode_page
2313 		 *     - f2fs_new_node_page
2314 		 *      - set_node_addr
2315 		 *  - f2fs_alloc_nid_done
2316 		 *   - __remove_nid_from_list(PREALLOC_NID)
2317 		 *                         - __insert_nid_to_list(FREE_NID)
2318 		 */
2319 		ne = __lookup_nat_cache(nm_i, nid);
2320 		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2321 				nat_get_blkaddr(ne) != NULL_ADDR))
2322 			goto err_out;
2323 
2324 		e = __lookup_free_nid_list(nm_i, nid);
2325 		if (e) {
2326 			if (e->state == FREE_NID)
2327 				ret = true;
2328 			goto err_out;
2329 		}
2330 	}
2331 	ret = true;
2332 	err = __insert_free_nid(sbi, i);
2333 err_out:
2334 	if (update) {
2335 		update_free_nid_bitmap(sbi, nid, ret, build);
2336 		if (!build)
2337 			nm_i->available_nids++;
2338 	}
2339 	spin_unlock(&nm_i->nid_list_lock);
2340 	radix_tree_preload_end();
2341 
2342 	if (err)
2343 		kmem_cache_free(free_nid_slab, i);
2344 	return ret;
2345 }
2346 
2347 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2348 {
2349 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2350 	struct free_nid *i;
2351 	bool need_free = false;
2352 
2353 	spin_lock(&nm_i->nid_list_lock);
2354 	i = __lookup_free_nid_list(nm_i, nid);
2355 	if (i && i->state == FREE_NID) {
2356 		__remove_free_nid(sbi, i, FREE_NID);
2357 		need_free = true;
2358 	}
2359 	spin_unlock(&nm_i->nid_list_lock);
2360 
2361 	if (need_free)
2362 		kmem_cache_free(free_nid_slab, i);
2363 }
2364 
2365 static int scan_nat_page(struct f2fs_sb_info *sbi,
2366 			struct page *nat_page, nid_t start_nid)
2367 {
2368 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2369 	struct f2fs_nat_block *nat_blk = page_address(nat_page);
2370 	block_t blk_addr;
2371 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2372 	int i;
2373 
2374 	__set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2375 
2376 	i = start_nid % NAT_ENTRY_PER_BLOCK;
2377 
2378 	for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2379 		if (unlikely(start_nid >= nm_i->max_nid))
2380 			break;
2381 
2382 		blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2383 
2384 		if (blk_addr == NEW_ADDR)
2385 			return -EINVAL;
2386 
2387 		if (blk_addr == NULL_ADDR) {
2388 			add_free_nid(sbi, start_nid, true, true);
2389 		} else {
2390 			spin_lock(&NM_I(sbi)->nid_list_lock);
2391 			update_free_nid_bitmap(sbi, start_nid, false, true);
2392 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2393 		}
2394 	}
2395 
2396 	return 0;
2397 }
2398 
2399 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2400 {
2401 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2402 	struct f2fs_journal *journal = curseg->journal;
2403 	int i;
2404 
2405 	down_read(&curseg->journal_rwsem);
2406 	for (i = 0; i < nats_in_cursum(journal); i++) {
2407 		block_t addr;
2408 		nid_t nid;
2409 
2410 		addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2411 		nid = le32_to_cpu(nid_in_journal(journal, i));
2412 		if (addr == NULL_ADDR)
2413 			add_free_nid(sbi, nid, true, false);
2414 		else
2415 			remove_free_nid(sbi, nid);
2416 	}
2417 	up_read(&curseg->journal_rwsem);
2418 }
2419 
2420 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2421 {
2422 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2423 	unsigned int i, idx;
2424 	nid_t nid;
2425 
2426 	f2fs_down_read(&nm_i->nat_tree_lock);
2427 
2428 	for (i = 0; i < nm_i->nat_blocks; i++) {
2429 		if (!test_bit_le(i, nm_i->nat_block_bitmap))
2430 			continue;
2431 		if (!nm_i->free_nid_count[i])
2432 			continue;
2433 		for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2434 			idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2435 						NAT_ENTRY_PER_BLOCK, idx);
2436 			if (idx >= NAT_ENTRY_PER_BLOCK)
2437 				break;
2438 
2439 			nid = i * NAT_ENTRY_PER_BLOCK + idx;
2440 			add_free_nid(sbi, nid, true, false);
2441 
2442 			if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2443 				goto out;
2444 		}
2445 	}
2446 out:
2447 	scan_curseg_cache(sbi);
2448 
2449 	f2fs_up_read(&nm_i->nat_tree_lock);
2450 }
2451 
2452 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2453 						bool sync, bool mount)
2454 {
2455 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2456 	int i = 0, ret;
2457 	nid_t nid = nm_i->next_scan_nid;
2458 
2459 	if (unlikely(nid >= nm_i->max_nid))
2460 		nid = 0;
2461 
2462 	if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2463 		nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2464 
2465 	/* Enough entries */
2466 	if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2467 		return 0;
2468 
2469 	if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2470 		return 0;
2471 
2472 	if (!mount) {
2473 		/* try to find free nids in free_nid_bitmap */
2474 		scan_free_nid_bits(sbi);
2475 
2476 		if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2477 			return 0;
2478 	}
2479 
2480 	/* readahead nat pages to be scanned */
2481 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2482 							META_NAT, true);
2483 
2484 	f2fs_down_read(&nm_i->nat_tree_lock);
2485 
2486 	while (1) {
2487 		if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2488 						nm_i->nat_block_bitmap)) {
2489 			struct page *page = get_current_nat_page(sbi, nid);
2490 
2491 			if (IS_ERR(page)) {
2492 				ret = PTR_ERR(page);
2493 			} else {
2494 				ret = scan_nat_page(sbi, page, nid);
2495 				f2fs_put_page(page, 1);
2496 			}
2497 
2498 			if (ret) {
2499 				f2fs_up_read(&nm_i->nat_tree_lock);
2500 				f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2501 				return ret;
2502 			}
2503 		}
2504 
2505 		nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2506 		if (unlikely(nid >= nm_i->max_nid))
2507 			nid = 0;
2508 
2509 		if (++i >= FREE_NID_PAGES)
2510 			break;
2511 	}
2512 
2513 	/* go to the next free nat pages to find free nids abundantly */
2514 	nm_i->next_scan_nid = nid;
2515 
2516 	/* find free nids from current sum_pages */
2517 	scan_curseg_cache(sbi);
2518 
2519 	f2fs_up_read(&nm_i->nat_tree_lock);
2520 
2521 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2522 					nm_i->ra_nid_pages, META_NAT, false);
2523 
2524 	return 0;
2525 }
2526 
2527 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2528 {
2529 	int ret;
2530 
2531 	mutex_lock(&NM_I(sbi)->build_lock);
2532 	ret = __f2fs_build_free_nids(sbi, sync, mount);
2533 	mutex_unlock(&NM_I(sbi)->build_lock);
2534 
2535 	return ret;
2536 }
2537 
2538 /*
2539  * If this function returns success, caller can obtain a new nid
2540  * from second parameter of this function.
2541  * The returned nid could be used ino as well as nid when inode is created.
2542  */
2543 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2544 {
2545 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2546 	struct free_nid *i = NULL;
2547 retry:
2548 	if (time_to_inject(sbi, FAULT_ALLOC_NID))
2549 		return false;
2550 
2551 	spin_lock(&nm_i->nid_list_lock);
2552 
2553 	if (unlikely(nm_i->available_nids == 0)) {
2554 		spin_unlock(&nm_i->nid_list_lock);
2555 		return false;
2556 	}
2557 
2558 	/* We should not use stale free nids created by f2fs_build_free_nids */
2559 	if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2560 		f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2561 		i = list_first_entry(&nm_i->free_nid_list,
2562 					struct free_nid, list);
2563 		*nid = i->nid;
2564 
2565 		__move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2566 		nm_i->available_nids--;
2567 
2568 		update_free_nid_bitmap(sbi, *nid, false, false);
2569 
2570 		spin_unlock(&nm_i->nid_list_lock);
2571 		return true;
2572 	}
2573 	spin_unlock(&nm_i->nid_list_lock);
2574 
2575 	/* Let's scan nat pages and its caches to get free nids */
2576 	if (!f2fs_build_free_nids(sbi, true, false))
2577 		goto retry;
2578 	return false;
2579 }
2580 
2581 /*
2582  * f2fs_alloc_nid() should be called prior to this function.
2583  */
2584 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2585 {
2586 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2587 	struct free_nid *i;
2588 
2589 	spin_lock(&nm_i->nid_list_lock);
2590 	i = __lookup_free_nid_list(nm_i, nid);
2591 	f2fs_bug_on(sbi, !i);
2592 	__remove_free_nid(sbi, i, PREALLOC_NID);
2593 	spin_unlock(&nm_i->nid_list_lock);
2594 
2595 	kmem_cache_free(free_nid_slab, i);
2596 }
2597 
2598 /*
2599  * f2fs_alloc_nid() should be called prior to this function.
2600  */
2601 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2602 {
2603 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2604 	struct free_nid *i;
2605 	bool need_free = false;
2606 
2607 	if (!nid)
2608 		return;
2609 
2610 	spin_lock(&nm_i->nid_list_lock);
2611 	i = __lookup_free_nid_list(nm_i, nid);
2612 	f2fs_bug_on(sbi, !i);
2613 
2614 	if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2615 		__remove_free_nid(sbi, i, PREALLOC_NID);
2616 		need_free = true;
2617 	} else {
2618 		__move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2619 	}
2620 
2621 	nm_i->available_nids++;
2622 
2623 	update_free_nid_bitmap(sbi, nid, true, false);
2624 
2625 	spin_unlock(&nm_i->nid_list_lock);
2626 
2627 	if (need_free)
2628 		kmem_cache_free(free_nid_slab, i);
2629 }
2630 
2631 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2632 {
2633 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2634 	int nr = nr_shrink;
2635 
2636 	if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2637 		return 0;
2638 
2639 	if (!mutex_trylock(&nm_i->build_lock))
2640 		return 0;
2641 
2642 	while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2643 		struct free_nid *i, *next;
2644 		unsigned int batch = SHRINK_NID_BATCH_SIZE;
2645 
2646 		spin_lock(&nm_i->nid_list_lock);
2647 		list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2648 			if (!nr_shrink || !batch ||
2649 				nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2650 				break;
2651 			__remove_free_nid(sbi, i, FREE_NID);
2652 			kmem_cache_free(free_nid_slab, i);
2653 			nr_shrink--;
2654 			batch--;
2655 		}
2656 		spin_unlock(&nm_i->nid_list_lock);
2657 	}
2658 
2659 	mutex_unlock(&nm_i->build_lock);
2660 
2661 	return nr - nr_shrink;
2662 }
2663 
2664 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2665 {
2666 	void *src_addr, *dst_addr;
2667 	size_t inline_size;
2668 	struct page *ipage;
2669 	struct f2fs_inode *ri;
2670 
2671 	ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2672 	if (IS_ERR(ipage))
2673 		return PTR_ERR(ipage);
2674 
2675 	ri = F2FS_INODE(page);
2676 	if (ri->i_inline & F2FS_INLINE_XATTR) {
2677 		if (!f2fs_has_inline_xattr(inode)) {
2678 			set_inode_flag(inode, FI_INLINE_XATTR);
2679 			stat_inc_inline_xattr(inode);
2680 		}
2681 	} else {
2682 		if (f2fs_has_inline_xattr(inode)) {
2683 			stat_dec_inline_xattr(inode);
2684 			clear_inode_flag(inode, FI_INLINE_XATTR);
2685 		}
2686 		goto update_inode;
2687 	}
2688 
2689 	dst_addr = inline_xattr_addr(inode, ipage);
2690 	src_addr = inline_xattr_addr(inode, page);
2691 	inline_size = inline_xattr_size(inode);
2692 
2693 	f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2694 	memcpy(dst_addr, src_addr, inline_size);
2695 update_inode:
2696 	f2fs_update_inode(inode, ipage);
2697 	f2fs_put_page(ipage, 1);
2698 	return 0;
2699 }
2700 
2701 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2702 {
2703 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2704 	nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2705 	nid_t new_xnid;
2706 	struct dnode_of_data dn;
2707 	struct node_info ni;
2708 	struct page *xpage;
2709 	int err;
2710 
2711 	if (!prev_xnid)
2712 		goto recover_xnid;
2713 
2714 	/* 1: invalidate the previous xattr nid */
2715 	err = f2fs_get_node_info(sbi, prev_xnid, &ni, false);
2716 	if (err)
2717 		return err;
2718 
2719 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
2720 	dec_valid_node_count(sbi, inode, false);
2721 	set_node_addr(sbi, &ni, NULL_ADDR, false);
2722 
2723 recover_xnid:
2724 	/* 2: update xattr nid in inode */
2725 	if (!f2fs_alloc_nid(sbi, &new_xnid))
2726 		return -ENOSPC;
2727 
2728 	set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2729 	xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2730 	if (IS_ERR(xpage)) {
2731 		f2fs_alloc_nid_failed(sbi, new_xnid);
2732 		return PTR_ERR(xpage);
2733 	}
2734 
2735 	f2fs_alloc_nid_done(sbi, new_xnid);
2736 	f2fs_update_inode_page(inode);
2737 
2738 	/* 3: update and set xattr node page dirty */
2739 	memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2740 
2741 	set_page_dirty(xpage);
2742 	f2fs_put_page(xpage, 1);
2743 
2744 	return 0;
2745 }
2746 
2747 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2748 {
2749 	struct f2fs_inode *src, *dst;
2750 	nid_t ino = ino_of_node(page);
2751 	struct node_info old_ni, new_ni;
2752 	struct page *ipage;
2753 	int err;
2754 
2755 	err = f2fs_get_node_info(sbi, ino, &old_ni, false);
2756 	if (err)
2757 		return err;
2758 
2759 	if (unlikely(old_ni.blk_addr != NULL_ADDR))
2760 		return -EINVAL;
2761 retry:
2762 	ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2763 	if (!ipage) {
2764 		memalloc_retry_wait(GFP_NOFS);
2765 		goto retry;
2766 	}
2767 
2768 	/* Should not use this inode from free nid list */
2769 	remove_free_nid(sbi, ino);
2770 
2771 	if (!PageUptodate(ipage))
2772 		SetPageUptodate(ipage);
2773 	fill_node_footer(ipage, ino, ino, 0, true);
2774 	set_cold_node(ipage, false);
2775 
2776 	src = F2FS_INODE(page);
2777 	dst = F2FS_INODE(ipage);
2778 
2779 	memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2780 	dst->i_size = 0;
2781 	dst->i_blocks = cpu_to_le64(1);
2782 	dst->i_links = cpu_to_le32(1);
2783 	dst->i_xattr_nid = 0;
2784 	dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2785 	if (dst->i_inline & F2FS_EXTRA_ATTR) {
2786 		dst->i_extra_isize = src->i_extra_isize;
2787 
2788 		if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2789 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2790 							i_inline_xattr_size))
2791 			dst->i_inline_xattr_size = src->i_inline_xattr_size;
2792 
2793 		if (f2fs_sb_has_project_quota(sbi) &&
2794 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2795 								i_projid))
2796 			dst->i_projid = src->i_projid;
2797 
2798 		if (f2fs_sb_has_inode_crtime(sbi) &&
2799 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2800 							i_crtime_nsec)) {
2801 			dst->i_crtime = src->i_crtime;
2802 			dst->i_crtime_nsec = src->i_crtime_nsec;
2803 		}
2804 	}
2805 
2806 	new_ni = old_ni;
2807 	new_ni.ino = ino;
2808 
2809 	if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2810 		WARN_ON(1);
2811 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2812 	inc_valid_inode_count(sbi);
2813 	set_page_dirty(ipage);
2814 	f2fs_put_page(ipage, 1);
2815 	return 0;
2816 }
2817 
2818 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2819 			unsigned int segno, struct f2fs_summary_block *sum)
2820 {
2821 	struct f2fs_node *rn;
2822 	struct f2fs_summary *sum_entry;
2823 	block_t addr;
2824 	int i, idx, last_offset, nrpages;
2825 
2826 	/* scan the node segment */
2827 	last_offset = sbi->blocks_per_seg;
2828 	addr = START_BLOCK(sbi, segno);
2829 	sum_entry = &sum->entries[0];
2830 
2831 	for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2832 		nrpages = bio_max_segs(last_offset - i);
2833 
2834 		/* readahead node pages */
2835 		f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2836 
2837 		for (idx = addr; idx < addr + nrpages; idx++) {
2838 			struct page *page = f2fs_get_tmp_page(sbi, idx);
2839 
2840 			if (IS_ERR(page))
2841 				return PTR_ERR(page);
2842 
2843 			rn = F2FS_NODE(page);
2844 			sum_entry->nid = rn->footer.nid;
2845 			sum_entry->version = 0;
2846 			sum_entry->ofs_in_node = 0;
2847 			sum_entry++;
2848 			f2fs_put_page(page, 1);
2849 		}
2850 
2851 		invalidate_mapping_pages(META_MAPPING(sbi), addr,
2852 							addr + nrpages);
2853 	}
2854 	return 0;
2855 }
2856 
2857 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2858 {
2859 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2860 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2861 	struct f2fs_journal *journal = curseg->journal;
2862 	int i;
2863 
2864 	down_write(&curseg->journal_rwsem);
2865 	for (i = 0; i < nats_in_cursum(journal); i++) {
2866 		struct nat_entry *ne;
2867 		struct f2fs_nat_entry raw_ne;
2868 		nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2869 
2870 		if (f2fs_check_nid_range(sbi, nid))
2871 			continue;
2872 
2873 		raw_ne = nat_in_journal(journal, i);
2874 
2875 		ne = __lookup_nat_cache(nm_i, nid);
2876 		if (!ne) {
2877 			ne = __alloc_nat_entry(sbi, nid, true);
2878 			__init_nat_entry(nm_i, ne, &raw_ne, true);
2879 		}
2880 
2881 		/*
2882 		 * if a free nat in journal has not been used after last
2883 		 * checkpoint, we should remove it from available nids,
2884 		 * since later we will add it again.
2885 		 */
2886 		if (!get_nat_flag(ne, IS_DIRTY) &&
2887 				le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2888 			spin_lock(&nm_i->nid_list_lock);
2889 			nm_i->available_nids--;
2890 			spin_unlock(&nm_i->nid_list_lock);
2891 		}
2892 
2893 		__set_nat_cache_dirty(nm_i, ne);
2894 	}
2895 	update_nats_in_cursum(journal, -i);
2896 	up_write(&curseg->journal_rwsem);
2897 }
2898 
2899 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2900 						struct list_head *head, int max)
2901 {
2902 	struct nat_entry_set *cur;
2903 
2904 	if (nes->entry_cnt >= max)
2905 		goto add_out;
2906 
2907 	list_for_each_entry(cur, head, set_list) {
2908 		if (cur->entry_cnt >= nes->entry_cnt) {
2909 			list_add(&nes->set_list, cur->set_list.prev);
2910 			return;
2911 		}
2912 	}
2913 add_out:
2914 	list_add_tail(&nes->set_list, head);
2915 }
2916 
2917 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs,
2918 							unsigned int valid)
2919 {
2920 	if (valid == 0) {
2921 		__set_bit_le(nat_ofs, nm_i->empty_nat_bits);
2922 		__clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2923 		return;
2924 	}
2925 
2926 	__clear_bit_le(nat_ofs, nm_i->empty_nat_bits);
2927 	if (valid == NAT_ENTRY_PER_BLOCK)
2928 		__set_bit_le(nat_ofs, nm_i->full_nat_bits);
2929 	else
2930 		__clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2931 }
2932 
2933 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2934 						struct page *page)
2935 {
2936 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2937 	unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2938 	struct f2fs_nat_block *nat_blk = page_address(page);
2939 	int valid = 0;
2940 	int i = 0;
2941 
2942 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
2943 		return;
2944 
2945 	if (nat_index == 0) {
2946 		valid = 1;
2947 		i = 1;
2948 	}
2949 	for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2950 		if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2951 			valid++;
2952 	}
2953 
2954 	__update_nat_bits(nm_i, nat_index, valid);
2955 }
2956 
2957 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi)
2958 {
2959 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2960 	unsigned int nat_ofs;
2961 
2962 	f2fs_down_read(&nm_i->nat_tree_lock);
2963 
2964 	for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) {
2965 		unsigned int valid = 0, nid_ofs = 0;
2966 
2967 		/* handle nid zero due to it should never be used */
2968 		if (unlikely(nat_ofs == 0)) {
2969 			valid = 1;
2970 			nid_ofs = 1;
2971 		}
2972 
2973 		for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) {
2974 			if (!test_bit_le(nid_ofs,
2975 					nm_i->free_nid_bitmap[nat_ofs]))
2976 				valid++;
2977 		}
2978 
2979 		__update_nat_bits(nm_i, nat_ofs, valid);
2980 	}
2981 
2982 	f2fs_up_read(&nm_i->nat_tree_lock);
2983 }
2984 
2985 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2986 		struct nat_entry_set *set, struct cp_control *cpc)
2987 {
2988 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2989 	struct f2fs_journal *journal = curseg->journal;
2990 	nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2991 	bool to_journal = true;
2992 	struct f2fs_nat_block *nat_blk;
2993 	struct nat_entry *ne, *cur;
2994 	struct page *page = NULL;
2995 
2996 	/*
2997 	 * there are two steps to flush nat entries:
2998 	 * #1, flush nat entries to journal in current hot data summary block.
2999 	 * #2, flush nat entries to nat page.
3000 	 */
3001 	if ((cpc->reason & CP_UMOUNT) ||
3002 		!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
3003 		to_journal = false;
3004 
3005 	if (to_journal) {
3006 		down_write(&curseg->journal_rwsem);
3007 	} else {
3008 		page = get_next_nat_page(sbi, start_nid);
3009 		if (IS_ERR(page))
3010 			return PTR_ERR(page);
3011 
3012 		nat_blk = page_address(page);
3013 		f2fs_bug_on(sbi, !nat_blk);
3014 	}
3015 
3016 	/* flush dirty nats in nat entry set */
3017 	list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
3018 		struct f2fs_nat_entry *raw_ne;
3019 		nid_t nid = nat_get_nid(ne);
3020 		int offset;
3021 
3022 		f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
3023 
3024 		if (to_journal) {
3025 			offset = f2fs_lookup_journal_in_cursum(journal,
3026 							NAT_JOURNAL, nid, 1);
3027 			f2fs_bug_on(sbi, offset < 0);
3028 			raw_ne = &nat_in_journal(journal, offset);
3029 			nid_in_journal(journal, offset) = cpu_to_le32(nid);
3030 		} else {
3031 			raw_ne = &nat_blk->entries[nid - start_nid];
3032 		}
3033 		raw_nat_from_node_info(raw_ne, &ne->ni);
3034 		nat_reset_flag(ne);
3035 		__clear_nat_cache_dirty(NM_I(sbi), set, ne);
3036 		if (nat_get_blkaddr(ne) == NULL_ADDR) {
3037 			add_free_nid(sbi, nid, false, true);
3038 		} else {
3039 			spin_lock(&NM_I(sbi)->nid_list_lock);
3040 			update_free_nid_bitmap(sbi, nid, false, false);
3041 			spin_unlock(&NM_I(sbi)->nid_list_lock);
3042 		}
3043 	}
3044 
3045 	if (to_journal) {
3046 		up_write(&curseg->journal_rwsem);
3047 	} else {
3048 		update_nat_bits(sbi, start_nid, page);
3049 		f2fs_put_page(page, 1);
3050 	}
3051 
3052 	/* Allow dirty nats by node block allocation in write_begin */
3053 	if (!set->entry_cnt) {
3054 		radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
3055 		kmem_cache_free(nat_entry_set_slab, set);
3056 	}
3057 	return 0;
3058 }
3059 
3060 /*
3061  * This function is called during the checkpointing process.
3062  */
3063 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3064 {
3065 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3066 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3067 	struct f2fs_journal *journal = curseg->journal;
3068 	struct nat_entry_set *setvec[SETVEC_SIZE];
3069 	struct nat_entry_set *set, *tmp;
3070 	unsigned int found;
3071 	nid_t set_idx = 0;
3072 	LIST_HEAD(sets);
3073 	int err = 0;
3074 
3075 	/*
3076 	 * during unmount, let's flush nat_bits before checking
3077 	 * nat_cnt[DIRTY_NAT].
3078 	 */
3079 	if (cpc->reason & CP_UMOUNT) {
3080 		f2fs_down_write(&nm_i->nat_tree_lock);
3081 		remove_nats_in_journal(sbi);
3082 		f2fs_up_write(&nm_i->nat_tree_lock);
3083 	}
3084 
3085 	if (!nm_i->nat_cnt[DIRTY_NAT])
3086 		return 0;
3087 
3088 	f2fs_down_write(&nm_i->nat_tree_lock);
3089 
3090 	/*
3091 	 * if there are no enough space in journal to store dirty nat
3092 	 * entries, remove all entries from journal and merge them
3093 	 * into nat entry set.
3094 	 */
3095 	if (cpc->reason & CP_UMOUNT ||
3096 		!__has_cursum_space(journal,
3097 			nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3098 		remove_nats_in_journal(sbi);
3099 
3100 	while ((found = __gang_lookup_nat_set(nm_i,
3101 					set_idx, SETVEC_SIZE, setvec))) {
3102 		unsigned idx;
3103 
3104 		set_idx = setvec[found - 1]->set + 1;
3105 		for (idx = 0; idx < found; idx++)
3106 			__adjust_nat_entry_set(setvec[idx], &sets,
3107 						MAX_NAT_JENTRIES(journal));
3108 	}
3109 
3110 	/* flush dirty nats in nat entry set */
3111 	list_for_each_entry_safe(set, tmp, &sets, set_list) {
3112 		err = __flush_nat_entry_set(sbi, set, cpc);
3113 		if (err)
3114 			break;
3115 	}
3116 
3117 	f2fs_up_write(&nm_i->nat_tree_lock);
3118 	/* Allow dirty nats by node block allocation in write_begin */
3119 
3120 	return err;
3121 }
3122 
3123 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3124 {
3125 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3126 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3127 	unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3128 	unsigned int i;
3129 	__u64 cp_ver = cur_cp_version(ckpt);
3130 	block_t nat_bits_addr;
3131 
3132 	nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3133 	nm_i->nat_bits = f2fs_kvzalloc(sbi,
3134 			nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3135 	if (!nm_i->nat_bits)
3136 		return -ENOMEM;
3137 
3138 	nm_i->full_nat_bits = nm_i->nat_bits + 8;
3139 	nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3140 
3141 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3142 		return 0;
3143 
3144 	nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3145 						nm_i->nat_bits_blocks;
3146 	for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3147 		struct page *page;
3148 
3149 		page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3150 		if (IS_ERR(page))
3151 			return PTR_ERR(page);
3152 
3153 		memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3154 					page_address(page), F2FS_BLKSIZE);
3155 		f2fs_put_page(page, 1);
3156 	}
3157 
3158 	cp_ver |= (cur_cp_crc(ckpt) << 32);
3159 	if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3160 		clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
3161 		f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)",
3162 			cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits));
3163 		return 0;
3164 	}
3165 
3166 	f2fs_notice(sbi, "Found nat_bits in checkpoint");
3167 	return 0;
3168 }
3169 
3170 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3171 {
3172 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3173 	unsigned int i = 0;
3174 	nid_t nid, last_nid;
3175 
3176 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3177 		return;
3178 
3179 	for (i = 0; i < nm_i->nat_blocks; i++) {
3180 		i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3181 		if (i >= nm_i->nat_blocks)
3182 			break;
3183 
3184 		__set_bit_le(i, nm_i->nat_block_bitmap);
3185 
3186 		nid = i * NAT_ENTRY_PER_BLOCK;
3187 		last_nid = nid + NAT_ENTRY_PER_BLOCK;
3188 
3189 		spin_lock(&NM_I(sbi)->nid_list_lock);
3190 		for (; nid < last_nid; nid++)
3191 			update_free_nid_bitmap(sbi, nid, true, true);
3192 		spin_unlock(&NM_I(sbi)->nid_list_lock);
3193 	}
3194 
3195 	for (i = 0; i < nm_i->nat_blocks; i++) {
3196 		i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3197 		if (i >= nm_i->nat_blocks)
3198 			break;
3199 
3200 		__set_bit_le(i, nm_i->nat_block_bitmap);
3201 	}
3202 }
3203 
3204 static int init_node_manager(struct f2fs_sb_info *sbi)
3205 {
3206 	struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3207 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3208 	unsigned char *version_bitmap;
3209 	unsigned int nat_segs;
3210 	int err;
3211 
3212 	nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3213 
3214 	/* segment_count_nat includes pair segment so divide to 2. */
3215 	nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3216 	nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3217 	nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3218 
3219 	/* not used nids: 0, node, meta, (and root counted as valid node) */
3220 	nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3221 						F2FS_RESERVED_NODE_NUM;
3222 	nm_i->nid_cnt[FREE_NID] = 0;
3223 	nm_i->nid_cnt[PREALLOC_NID] = 0;
3224 	nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3225 	nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3226 	nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3227 	nm_i->max_rf_node_blocks = DEF_RF_NODE_BLOCKS;
3228 
3229 	INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3230 	INIT_LIST_HEAD(&nm_i->free_nid_list);
3231 	INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3232 	INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3233 	INIT_LIST_HEAD(&nm_i->nat_entries);
3234 	spin_lock_init(&nm_i->nat_list_lock);
3235 
3236 	mutex_init(&nm_i->build_lock);
3237 	spin_lock_init(&nm_i->nid_list_lock);
3238 	init_f2fs_rwsem(&nm_i->nat_tree_lock);
3239 
3240 	nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3241 	nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3242 	version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3243 	nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3244 					GFP_KERNEL);
3245 	if (!nm_i->nat_bitmap)
3246 		return -ENOMEM;
3247 
3248 	err = __get_nat_bitmaps(sbi);
3249 	if (err)
3250 		return err;
3251 
3252 #ifdef CONFIG_F2FS_CHECK_FS
3253 	nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3254 					GFP_KERNEL);
3255 	if (!nm_i->nat_bitmap_mir)
3256 		return -ENOMEM;
3257 #endif
3258 
3259 	return 0;
3260 }
3261 
3262 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3263 {
3264 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3265 	int i;
3266 
3267 	nm_i->free_nid_bitmap =
3268 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3269 					      nm_i->nat_blocks),
3270 			      GFP_KERNEL);
3271 	if (!nm_i->free_nid_bitmap)
3272 		return -ENOMEM;
3273 
3274 	for (i = 0; i < nm_i->nat_blocks; i++) {
3275 		nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3276 			f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3277 		if (!nm_i->free_nid_bitmap[i])
3278 			return -ENOMEM;
3279 	}
3280 
3281 	nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3282 								GFP_KERNEL);
3283 	if (!nm_i->nat_block_bitmap)
3284 		return -ENOMEM;
3285 
3286 	nm_i->free_nid_count =
3287 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3288 					      nm_i->nat_blocks),
3289 			      GFP_KERNEL);
3290 	if (!nm_i->free_nid_count)
3291 		return -ENOMEM;
3292 	return 0;
3293 }
3294 
3295 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3296 {
3297 	int err;
3298 
3299 	sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3300 							GFP_KERNEL);
3301 	if (!sbi->nm_info)
3302 		return -ENOMEM;
3303 
3304 	err = init_node_manager(sbi);
3305 	if (err)
3306 		return err;
3307 
3308 	err = init_free_nid_cache(sbi);
3309 	if (err)
3310 		return err;
3311 
3312 	/* load free nid status from nat_bits table */
3313 	load_free_nid_bitmap(sbi);
3314 
3315 	return f2fs_build_free_nids(sbi, true, true);
3316 }
3317 
3318 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3319 {
3320 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3321 	struct free_nid *i, *next_i;
3322 	struct nat_entry *natvec[NATVEC_SIZE];
3323 	struct nat_entry_set *setvec[SETVEC_SIZE];
3324 	nid_t nid = 0;
3325 	unsigned int found;
3326 
3327 	if (!nm_i)
3328 		return;
3329 
3330 	/* destroy free nid list */
3331 	spin_lock(&nm_i->nid_list_lock);
3332 	list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3333 		__remove_free_nid(sbi, i, FREE_NID);
3334 		spin_unlock(&nm_i->nid_list_lock);
3335 		kmem_cache_free(free_nid_slab, i);
3336 		spin_lock(&nm_i->nid_list_lock);
3337 	}
3338 	f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3339 	f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3340 	f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3341 	spin_unlock(&nm_i->nid_list_lock);
3342 
3343 	/* destroy nat cache */
3344 	f2fs_down_write(&nm_i->nat_tree_lock);
3345 	while ((found = __gang_lookup_nat_cache(nm_i,
3346 					nid, NATVEC_SIZE, natvec))) {
3347 		unsigned idx;
3348 
3349 		nid = nat_get_nid(natvec[found - 1]) + 1;
3350 		for (idx = 0; idx < found; idx++) {
3351 			spin_lock(&nm_i->nat_list_lock);
3352 			list_del(&natvec[idx]->list);
3353 			spin_unlock(&nm_i->nat_list_lock);
3354 
3355 			__del_from_nat_cache(nm_i, natvec[idx]);
3356 		}
3357 	}
3358 	f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3359 
3360 	/* destroy nat set cache */
3361 	nid = 0;
3362 	while ((found = __gang_lookup_nat_set(nm_i,
3363 					nid, SETVEC_SIZE, setvec))) {
3364 		unsigned idx;
3365 
3366 		nid = setvec[found - 1]->set + 1;
3367 		for (idx = 0; idx < found; idx++) {
3368 			/* entry_cnt is not zero, when cp_error was occurred */
3369 			f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3370 			radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3371 			kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3372 		}
3373 	}
3374 	f2fs_up_write(&nm_i->nat_tree_lock);
3375 
3376 	kvfree(nm_i->nat_block_bitmap);
3377 	if (nm_i->free_nid_bitmap) {
3378 		int i;
3379 
3380 		for (i = 0; i < nm_i->nat_blocks; i++)
3381 			kvfree(nm_i->free_nid_bitmap[i]);
3382 		kvfree(nm_i->free_nid_bitmap);
3383 	}
3384 	kvfree(nm_i->free_nid_count);
3385 
3386 	kvfree(nm_i->nat_bitmap);
3387 	kvfree(nm_i->nat_bits);
3388 #ifdef CONFIG_F2FS_CHECK_FS
3389 	kvfree(nm_i->nat_bitmap_mir);
3390 #endif
3391 	sbi->nm_info = NULL;
3392 	kfree(nm_i);
3393 }
3394 
3395 int __init f2fs_create_node_manager_caches(void)
3396 {
3397 	nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3398 			sizeof(struct nat_entry));
3399 	if (!nat_entry_slab)
3400 		goto fail;
3401 
3402 	free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3403 			sizeof(struct free_nid));
3404 	if (!free_nid_slab)
3405 		goto destroy_nat_entry;
3406 
3407 	nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3408 			sizeof(struct nat_entry_set));
3409 	if (!nat_entry_set_slab)
3410 		goto destroy_free_nid;
3411 
3412 	fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3413 			sizeof(struct fsync_node_entry));
3414 	if (!fsync_node_entry_slab)
3415 		goto destroy_nat_entry_set;
3416 	return 0;
3417 
3418 destroy_nat_entry_set:
3419 	kmem_cache_destroy(nat_entry_set_slab);
3420 destroy_free_nid:
3421 	kmem_cache_destroy(free_nid_slab);
3422 destroy_nat_entry:
3423 	kmem_cache_destroy(nat_entry_slab);
3424 fail:
3425 	return -ENOMEM;
3426 }
3427 
3428 void f2fs_destroy_node_manager_caches(void)
3429 {
3430 	kmem_cache_destroy(fsync_node_entry_slab);
3431 	kmem_cache_destroy(nat_entry_set_slab);
3432 	kmem_cache_destroy(free_nid_slab);
3433 	kmem_cache_destroy(nat_entry_slab);
3434 }
3435