xref: /openbmc/linux/fs/f2fs/node.c (revision c4e78957)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * fs/f2fs/node.c
4  *
5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6  *             http://www.samsung.com/
7  */
8 #include <linux/fs.h>
9 #include <linux/f2fs_fs.h>
10 #include <linux/mpage.h>
11 #include <linux/sched/mm.h>
12 #include <linux/blkdev.h>
13 #include <linux/pagevec.h>
14 #include <linux/swap.h>
15 
16 #include "f2fs.h"
17 #include "node.h"
18 #include "segment.h"
19 #include "xattr.h"
20 #include "iostat.h"
21 #include <trace/events/f2fs.h>
22 
23 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
24 
25 static struct kmem_cache *nat_entry_slab;
26 static struct kmem_cache *free_nid_slab;
27 static struct kmem_cache *nat_entry_set_slab;
28 static struct kmem_cache *fsync_node_entry_slab;
29 
30 /*
31  * Check whether the given nid is within node id range.
32  */
33 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
34 {
35 	if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
36 		set_sbi_flag(sbi, SBI_NEED_FSCK);
37 		f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
38 			  __func__, nid);
39 		return -EFSCORRUPTED;
40 	}
41 	return 0;
42 }
43 
44 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
45 {
46 	struct f2fs_nm_info *nm_i = NM_I(sbi);
47 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
48 	struct sysinfo val;
49 	unsigned long avail_ram;
50 	unsigned long mem_size = 0;
51 	bool res = false;
52 
53 	if (!nm_i)
54 		return true;
55 
56 	si_meminfo(&val);
57 
58 	/* only uses low memory */
59 	avail_ram = val.totalram - val.totalhigh;
60 
61 	/*
62 	 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
63 	 */
64 	if (type == FREE_NIDS) {
65 		mem_size = (nm_i->nid_cnt[FREE_NID] *
66 				sizeof(struct free_nid)) >> PAGE_SHIFT;
67 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
68 	} else if (type == NAT_ENTRIES) {
69 		mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
70 				sizeof(struct nat_entry)) >> PAGE_SHIFT;
71 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
72 		if (excess_cached_nats(sbi))
73 			res = false;
74 	} else if (type == DIRTY_DENTS) {
75 		if (sbi->sb->s_bdi->wb.dirty_exceeded)
76 			return false;
77 		mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
78 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
79 	} else if (type == INO_ENTRIES) {
80 		int i;
81 
82 		for (i = 0; i < MAX_INO_ENTRY; i++)
83 			mem_size += sbi->im[i].ino_num *
84 						sizeof(struct ino_entry);
85 		mem_size >>= PAGE_SHIFT;
86 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
87 	} else if (type == EXTENT_CACHE) {
88 		mem_size = (atomic_read(&sbi->total_ext_tree) *
89 				sizeof(struct extent_tree) +
90 				atomic_read(&sbi->total_ext_node) *
91 				sizeof(struct extent_node)) >> PAGE_SHIFT;
92 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
93 	} else if (type == DISCARD_CACHE) {
94 		mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
95 				sizeof(struct discard_cmd)) >> PAGE_SHIFT;
96 		res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
97 	} else if (type == COMPRESS_PAGE) {
98 #ifdef CONFIG_F2FS_FS_COMPRESSION
99 		unsigned long free_ram = val.freeram;
100 
101 		/*
102 		 * free memory is lower than watermark or cached page count
103 		 * exceed threshold, deny caching compress page.
104 		 */
105 		res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
106 			(COMPRESS_MAPPING(sbi)->nrpages <
107 			 free_ram * sbi->compress_percent / 100);
108 #else
109 		res = false;
110 #endif
111 	} else {
112 		if (!sbi->sb->s_bdi->wb.dirty_exceeded)
113 			return true;
114 	}
115 	return res;
116 }
117 
118 static void clear_node_page_dirty(struct page *page)
119 {
120 	if (PageDirty(page)) {
121 		f2fs_clear_page_cache_dirty_tag(page);
122 		clear_page_dirty_for_io(page);
123 		dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
124 	}
125 	ClearPageUptodate(page);
126 }
127 
128 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
129 {
130 	return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
131 }
132 
133 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
134 {
135 	struct page *src_page;
136 	struct page *dst_page;
137 	pgoff_t dst_off;
138 	void *src_addr;
139 	void *dst_addr;
140 	struct f2fs_nm_info *nm_i = NM_I(sbi);
141 
142 	dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
143 
144 	/* get current nat block page with lock */
145 	src_page = get_current_nat_page(sbi, nid);
146 	if (IS_ERR(src_page))
147 		return src_page;
148 	dst_page = f2fs_grab_meta_page(sbi, dst_off);
149 	f2fs_bug_on(sbi, PageDirty(src_page));
150 
151 	src_addr = page_address(src_page);
152 	dst_addr = page_address(dst_page);
153 	memcpy(dst_addr, src_addr, PAGE_SIZE);
154 	set_page_dirty(dst_page);
155 	f2fs_put_page(src_page, 1);
156 
157 	set_to_next_nat(nm_i, nid);
158 
159 	return dst_page;
160 }
161 
162 static struct nat_entry *__alloc_nat_entry(struct f2fs_sb_info *sbi,
163 						nid_t nid, bool no_fail)
164 {
165 	struct nat_entry *new;
166 
167 	new = f2fs_kmem_cache_alloc(nat_entry_slab,
168 					GFP_F2FS_ZERO, no_fail, sbi);
169 	if (new) {
170 		nat_set_nid(new, nid);
171 		nat_reset_flag(new);
172 	}
173 	return new;
174 }
175 
176 static void __free_nat_entry(struct nat_entry *e)
177 {
178 	kmem_cache_free(nat_entry_slab, e);
179 }
180 
181 /* must be locked by nat_tree_lock */
182 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
183 	struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
184 {
185 	if (no_fail)
186 		f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
187 	else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
188 		return NULL;
189 
190 	if (raw_ne)
191 		node_info_from_raw_nat(&ne->ni, raw_ne);
192 
193 	spin_lock(&nm_i->nat_list_lock);
194 	list_add_tail(&ne->list, &nm_i->nat_entries);
195 	spin_unlock(&nm_i->nat_list_lock);
196 
197 	nm_i->nat_cnt[TOTAL_NAT]++;
198 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
199 	return ne;
200 }
201 
202 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
203 {
204 	struct nat_entry *ne;
205 
206 	ne = radix_tree_lookup(&nm_i->nat_root, n);
207 
208 	/* for recent accessed nat entry, move it to tail of lru list */
209 	if (ne && !get_nat_flag(ne, IS_DIRTY)) {
210 		spin_lock(&nm_i->nat_list_lock);
211 		if (!list_empty(&ne->list))
212 			list_move_tail(&ne->list, &nm_i->nat_entries);
213 		spin_unlock(&nm_i->nat_list_lock);
214 	}
215 
216 	return ne;
217 }
218 
219 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
220 		nid_t start, unsigned int nr, struct nat_entry **ep)
221 {
222 	return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
223 }
224 
225 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
226 {
227 	radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
228 	nm_i->nat_cnt[TOTAL_NAT]--;
229 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
230 	__free_nat_entry(e);
231 }
232 
233 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
234 							struct nat_entry *ne)
235 {
236 	nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
237 	struct nat_entry_set *head;
238 
239 	head = radix_tree_lookup(&nm_i->nat_set_root, set);
240 	if (!head) {
241 		head = f2fs_kmem_cache_alloc(nat_entry_set_slab,
242 						GFP_NOFS, true, NULL);
243 
244 		INIT_LIST_HEAD(&head->entry_list);
245 		INIT_LIST_HEAD(&head->set_list);
246 		head->set = set;
247 		head->entry_cnt = 0;
248 		f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
249 	}
250 	return head;
251 }
252 
253 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
254 						struct nat_entry *ne)
255 {
256 	struct nat_entry_set *head;
257 	bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
258 
259 	if (!new_ne)
260 		head = __grab_nat_entry_set(nm_i, ne);
261 
262 	/*
263 	 * update entry_cnt in below condition:
264 	 * 1. update NEW_ADDR to valid block address;
265 	 * 2. update old block address to new one;
266 	 */
267 	if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
268 				!get_nat_flag(ne, IS_DIRTY)))
269 		head->entry_cnt++;
270 
271 	set_nat_flag(ne, IS_PREALLOC, new_ne);
272 
273 	if (get_nat_flag(ne, IS_DIRTY))
274 		goto refresh_list;
275 
276 	nm_i->nat_cnt[DIRTY_NAT]++;
277 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
278 	set_nat_flag(ne, IS_DIRTY, true);
279 refresh_list:
280 	spin_lock(&nm_i->nat_list_lock);
281 	if (new_ne)
282 		list_del_init(&ne->list);
283 	else
284 		list_move_tail(&ne->list, &head->entry_list);
285 	spin_unlock(&nm_i->nat_list_lock);
286 }
287 
288 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
289 		struct nat_entry_set *set, struct nat_entry *ne)
290 {
291 	spin_lock(&nm_i->nat_list_lock);
292 	list_move_tail(&ne->list, &nm_i->nat_entries);
293 	spin_unlock(&nm_i->nat_list_lock);
294 
295 	set_nat_flag(ne, IS_DIRTY, false);
296 	set->entry_cnt--;
297 	nm_i->nat_cnt[DIRTY_NAT]--;
298 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
299 }
300 
301 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
302 		nid_t start, unsigned int nr, struct nat_entry_set **ep)
303 {
304 	return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
305 							start, nr);
306 }
307 
308 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
309 {
310 	return NODE_MAPPING(sbi) == page->mapping &&
311 			IS_DNODE(page) && is_cold_node(page);
312 }
313 
314 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
315 {
316 	spin_lock_init(&sbi->fsync_node_lock);
317 	INIT_LIST_HEAD(&sbi->fsync_node_list);
318 	sbi->fsync_seg_id = 0;
319 	sbi->fsync_node_num = 0;
320 }
321 
322 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
323 							struct page *page)
324 {
325 	struct fsync_node_entry *fn;
326 	unsigned long flags;
327 	unsigned int seq_id;
328 
329 	fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab,
330 					GFP_NOFS, true, NULL);
331 
332 	get_page(page);
333 	fn->page = page;
334 	INIT_LIST_HEAD(&fn->list);
335 
336 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
337 	list_add_tail(&fn->list, &sbi->fsync_node_list);
338 	fn->seq_id = sbi->fsync_seg_id++;
339 	seq_id = fn->seq_id;
340 	sbi->fsync_node_num++;
341 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
342 
343 	return seq_id;
344 }
345 
346 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
347 {
348 	struct fsync_node_entry *fn;
349 	unsigned long flags;
350 
351 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
352 	list_for_each_entry(fn, &sbi->fsync_node_list, list) {
353 		if (fn->page == page) {
354 			list_del(&fn->list);
355 			sbi->fsync_node_num--;
356 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
357 			kmem_cache_free(fsync_node_entry_slab, fn);
358 			put_page(page);
359 			return;
360 		}
361 	}
362 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
363 	f2fs_bug_on(sbi, 1);
364 }
365 
366 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
367 {
368 	unsigned long flags;
369 
370 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
371 	sbi->fsync_seg_id = 0;
372 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
373 }
374 
375 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
376 {
377 	struct f2fs_nm_info *nm_i = NM_I(sbi);
378 	struct nat_entry *e;
379 	bool need = false;
380 
381 	f2fs_down_read(&nm_i->nat_tree_lock);
382 	e = __lookup_nat_cache(nm_i, nid);
383 	if (e) {
384 		if (!get_nat_flag(e, IS_CHECKPOINTED) &&
385 				!get_nat_flag(e, HAS_FSYNCED_INODE))
386 			need = true;
387 	}
388 	f2fs_up_read(&nm_i->nat_tree_lock);
389 	return need;
390 }
391 
392 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
393 {
394 	struct f2fs_nm_info *nm_i = NM_I(sbi);
395 	struct nat_entry *e;
396 	bool is_cp = true;
397 
398 	f2fs_down_read(&nm_i->nat_tree_lock);
399 	e = __lookup_nat_cache(nm_i, nid);
400 	if (e && !get_nat_flag(e, IS_CHECKPOINTED))
401 		is_cp = false;
402 	f2fs_up_read(&nm_i->nat_tree_lock);
403 	return is_cp;
404 }
405 
406 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
407 {
408 	struct f2fs_nm_info *nm_i = NM_I(sbi);
409 	struct nat_entry *e;
410 	bool need_update = true;
411 
412 	f2fs_down_read(&nm_i->nat_tree_lock);
413 	e = __lookup_nat_cache(nm_i, ino);
414 	if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
415 			(get_nat_flag(e, IS_CHECKPOINTED) ||
416 			 get_nat_flag(e, HAS_FSYNCED_INODE)))
417 		need_update = false;
418 	f2fs_up_read(&nm_i->nat_tree_lock);
419 	return need_update;
420 }
421 
422 /* must be locked by nat_tree_lock */
423 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
424 						struct f2fs_nat_entry *ne)
425 {
426 	struct f2fs_nm_info *nm_i = NM_I(sbi);
427 	struct nat_entry *new, *e;
428 
429 	/* Let's mitigate lock contention of nat_tree_lock during checkpoint */
430 	if (f2fs_rwsem_is_locked(&sbi->cp_global_sem))
431 		return;
432 
433 	new = __alloc_nat_entry(sbi, nid, false);
434 	if (!new)
435 		return;
436 
437 	f2fs_down_write(&nm_i->nat_tree_lock);
438 	e = __lookup_nat_cache(nm_i, nid);
439 	if (!e)
440 		e = __init_nat_entry(nm_i, new, ne, false);
441 	else
442 		f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
443 				nat_get_blkaddr(e) !=
444 					le32_to_cpu(ne->block_addr) ||
445 				nat_get_version(e) != ne->version);
446 	f2fs_up_write(&nm_i->nat_tree_lock);
447 	if (e != new)
448 		__free_nat_entry(new);
449 }
450 
451 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
452 			block_t new_blkaddr, bool fsync_done)
453 {
454 	struct f2fs_nm_info *nm_i = NM_I(sbi);
455 	struct nat_entry *e;
456 	struct nat_entry *new = __alloc_nat_entry(sbi, ni->nid, true);
457 
458 	f2fs_down_write(&nm_i->nat_tree_lock);
459 	e = __lookup_nat_cache(nm_i, ni->nid);
460 	if (!e) {
461 		e = __init_nat_entry(nm_i, new, NULL, true);
462 		copy_node_info(&e->ni, ni);
463 		f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
464 	} else if (new_blkaddr == NEW_ADDR) {
465 		/*
466 		 * when nid is reallocated,
467 		 * previous nat entry can be remained in nat cache.
468 		 * So, reinitialize it with new information.
469 		 */
470 		copy_node_info(&e->ni, ni);
471 		f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
472 	}
473 	/* let's free early to reduce memory consumption */
474 	if (e != new)
475 		__free_nat_entry(new);
476 
477 	/* sanity check */
478 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
479 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
480 			new_blkaddr == NULL_ADDR);
481 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
482 			new_blkaddr == NEW_ADDR);
483 	f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
484 			new_blkaddr == NEW_ADDR);
485 
486 	/* increment version no as node is removed */
487 	if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
488 		unsigned char version = nat_get_version(e);
489 
490 		nat_set_version(e, inc_node_version(version));
491 	}
492 
493 	/* change address */
494 	nat_set_blkaddr(e, new_blkaddr);
495 	if (!__is_valid_data_blkaddr(new_blkaddr))
496 		set_nat_flag(e, IS_CHECKPOINTED, false);
497 	__set_nat_cache_dirty(nm_i, e);
498 
499 	/* update fsync_mark if its inode nat entry is still alive */
500 	if (ni->nid != ni->ino)
501 		e = __lookup_nat_cache(nm_i, ni->ino);
502 	if (e) {
503 		if (fsync_done && ni->nid == ni->ino)
504 			set_nat_flag(e, HAS_FSYNCED_INODE, true);
505 		set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
506 	}
507 	f2fs_up_write(&nm_i->nat_tree_lock);
508 }
509 
510 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
511 {
512 	struct f2fs_nm_info *nm_i = NM_I(sbi);
513 	int nr = nr_shrink;
514 
515 	if (!f2fs_down_write_trylock(&nm_i->nat_tree_lock))
516 		return 0;
517 
518 	spin_lock(&nm_i->nat_list_lock);
519 	while (nr_shrink) {
520 		struct nat_entry *ne;
521 
522 		if (list_empty(&nm_i->nat_entries))
523 			break;
524 
525 		ne = list_first_entry(&nm_i->nat_entries,
526 					struct nat_entry, list);
527 		list_del(&ne->list);
528 		spin_unlock(&nm_i->nat_list_lock);
529 
530 		__del_from_nat_cache(nm_i, ne);
531 		nr_shrink--;
532 
533 		spin_lock(&nm_i->nat_list_lock);
534 	}
535 	spin_unlock(&nm_i->nat_list_lock);
536 
537 	f2fs_up_write(&nm_i->nat_tree_lock);
538 	return nr - nr_shrink;
539 }
540 
541 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
542 				struct node_info *ni, bool checkpoint_context)
543 {
544 	struct f2fs_nm_info *nm_i = NM_I(sbi);
545 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
546 	struct f2fs_journal *journal = curseg->journal;
547 	nid_t start_nid = START_NID(nid);
548 	struct f2fs_nat_block *nat_blk;
549 	struct page *page = NULL;
550 	struct f2fs_nat_entry ne;
551 	struct nat_entry *e;
552 	pgoff_t index;
553 	block_t blkaddr;
554 	int i;
555 
556 	ni->nid = nid;
557 retry:
558 	/* Check nat cache */
559 	f2fs_down_read(&nm_i->nat_tree_lock);
560 	e = __lookup_nat_cache(nm_i, nid);
561 	if (e) {
562 		ni->ino = nat_get_ino(e);
563 		ni->blk_addr = nat_get_blkaddr(e);
564 		ni->version = nat_get_version(e);
565 		f2fs_up_read(&nm_i->nat_tree_lock);
566 		return 0;
567 	}
568 
569 	/*
570 	 * Check current segment summary by trying to grab journal_rwsem first.
571 	 * This sem is on the critical path on the checkpoint requiring the above
572 	 * nat_tree_lock. Therefore, we should retry, if we failed to grab here
573 	 * while not bothering checkpoint.
574 	 */
575 	if (!f2fs_rwsem_is_locked(&sbi->cp_global_sem) || checkpoint_context) {
576 		down_read(&curseg->journal_rwsem);
577 	} else if (f2fs_rwsem_is_contended(&nm_i->nat_tree_lock) ||
578 				!down_read_trylock(&curseg->journal_rwsem)) {
579 		f2fs_up_read(&nm_i->nat_tree_lock);
580 		goto retry;
581 	}
582 
583 	i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
584 	if (i >= 0) {
585 		ne = nat_in_journal(journal, i);
586 		node_info_from_raw_nat(ni, &ne);
587 	}
588         up_read(&curseg->journal_rwsem);
589 	if (i >= 0) {
590 		f2fs_up_read(&nm_i->nat_tree_lock);
591 		goto cache;
592 	}
593 
594 	/* Fill node_info from nat page */
595 	index = current_nat_addr(sbi, nid);
596 	f2fs_up_read(&nm_i->nat_tree_lock);
597 
598 	page = f2fs_get_meta_page(sbi, index);
599 	if (IS_ERR(page))
600 		return PTR_ERR(page);
601 
602 	nat_blk = (struct f2fs_nat_block *)page_address(page);
603 	ne = nat_blk->entries[nid - start_nid];
604 	node_info_from_raw_nat(ni, &ne);
605 	f2fs_put_page(page, 1);
606 cache:
607 	blkaddr = le32_to_cpu(ne.block_addr);
608 	if (__is_valid_data_blkaddr(blkaddr) &&
609 		!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
610 		return -EFAULT;
611 
612 	/* cache nat entry */
613 	cache_nat_entry(sbi, nid, &ne);
614 	return 0;
615 }
616 
617 /*
618  * readahead MAX_RA_NODE number of node pages.
619  */
620 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
621 {
622 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
623 	struct blk_plug plug;
624 	int i, end;
625 	nid_t nid;
626 
627 	blk_start_plug(&plug);
628 
629 	/* Then, try readahead for siblings of the desired node */
630 	end = start + n;
631 	end = min(end, NIDS_PER_BLOCK);
632 	for (i = start; i < end; i++) {
633 		nid = get_nid(parent, i, false);
634 		f2fs_ra_node_page(sbi, nid);
635 	}
636 
637 	blk_finish_plug(&plug);
638 }
639 
640 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
641 {
642 	const long direct_index = ADDRS_PER_INODE(dn->inode);
643 	const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
644 	const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
645 	unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
646 	int cur_level = dn->cur_level;
647 	int max_level = dn->max_level;
648 	pgoff_t base = 0;
649 
650 	if (!dn->max_level)
651 		return pgofs + 1;
652 
653 	while (max_level-- > cur_level)
654 		skipped_unit *= NIDS_PER_BLOCK;
655 
656 	switch (dn->max_level) {
657 	case 3:
658 		base += 2 * indirect_blks;
659 		fallthrough;
660 	case 2:
661 		base += 2 * direct_blks;
662 		fallthrough;
663 	case 1:
664 		base += direct_index;
665 		break;
666 	default:
667 		f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
668 	}
669 
670 	return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
671 }
672 
673 /*
674  * The maximum depth is four.
675  * Offset[0] will have raw inode offset.
676  */
677 static int get_node_path(struct inode *inode, long block,
678 				int offset[4], unsigned int noffset[4])
679 {
680 	const long direct_index = ADDRS_PER_INODE(inode);
681 	const long direct_blks = ADDRS_PER_BLOCK(inode);
682 	const long dptrs_per_blk = NIDS_PER_BLOCK;
683 	const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
684 	const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
685 	int n = 0;
686 	int level = 0;
687 
688 	noffset[0] = 0;
689 
690 	if (block < direct_index) {
691 		offset[n] = block;
692 		goto got;
693 	}
694 	block -= direct_index;
695 	if (block < direct_blks) {
696 		offset[n++] = NODE_DIR1_BLOCK;
697 		noffset[n] = 1;
698 		offset[n] = block;
699 		level = 1;
700 		goto got;
701 	}
702 	block -= direct_blks;
703 	if (block < direct_blks) {
704 		offset[n++] = NODE_DIR2_BLOCK;
705 		noffset[n] = 2;
706 		offset[n] = block;
707 		level = 1;
708 		goto got;
709 	}
710 	block -= direct_blks;
711 	if (block < indirect_blks) {
712 		offset[n++] = NODE_IND1_BLOCK;
713 		noffset[n] = 3;
714 		offset[n++] = block / direct_blks;
715 		noffset[n] = 4 + offset[n - 1];
716 		offset[n] = block % direct_blks;
717 		level = 2;
718 		goto got;
719 	}
720 	block -= indirect_blks;
721 	if (block < indirect_blks) {
722 		offset[n++] = NODE_IND2_BLOCK;
723 		noffset[n] = 4 + dptrs_per_blk;
724 		offset[n++] = block / direct_blks;
725 		noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
726 		offset[n] = block % direct_blks;
727 		level = 2;
728 		goto got;
729 	}
730 	block -= indirect_blks;
731 	if (block < dindirect_blks) {
732 		offset[n++] = NODE_DIND_BLOCK;
733 		noffset[n] = 5 + (dptrs_per_blk * 2);
734 		offset[n++] = block / indirect_blks;
735 		noffset[n] = 6 + (dptrs_per_blk * 2) +
736 			      offset[n - 1] * (dptrs_per_blk + 1);
737 		offset[n++] = (block / direct_blks) % dptrs_per_blk;
738 		noffset[n] = 7 + (dptrs_per_blk * 2) +
739 			      offset[n - 2] * (dptrs_per_blk + 1) +
740 			      offset[n - 1];
741 		offset[n] = block % direct_blks;
742 		level = 3;
743 		goto got;
744 	} else {
745 		return -E2BIG;
746 	}
747 got:
748 	return level;
749 }
750 
751 /*
752  * Caller should call f2fs_put_dnode(dn).
753  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
754  * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
755  */
756 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
757 {
758 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
759 	struct page *npage[4];
760 	struct page *parent = NULL;
761 	int offset[4];
762 	unsigned int noffset[4];
763 	nid_t nids[4];
764 	int level, i = 0;
765 	int err = 0;
766 
767 	level = get_node_path(dn->inode, index, offset, noffset);
768 	if (level < 0)
769 		return level;
770 
771 	nids[0] = dn->inode->i_ino;
772 	npage[0] = dn->inode_page;
773 
774 	if (!npage[0]) {
775 		npage[0] = f2fs_get_node_page(sbi, nids[0]);
776 		if (IS_ERR(npage[0]))
777 			return PTR_ERR(npage[0]);
778 	}
779 
780 	/* if inline_data is set, should not report any block indices */
781 	if (f2fs_has_inline_data(dn->inode) && index) {
782 		err = -ENOENT;
783 		f2fs_put_page(npage[0], 1);
784 		goto release_out;
785 	}
786 
787 	parent = npage[0];
788 	if (level != 0)
789 		nids[1] = get_nid(parent, offset[0], true);
790 	dn->inode_page = npage[0];
791 	dn->inode_page_locked = true;
792 
793 	/* get indirect or direct nodes */
794 	for (i = 1; i <= level; i++) {
795 		bool done = false;
796 
797 		if (!nids[i] && mode == ALLOC_NODE) {
798 			/* alloc new node */
799 			if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
800 				err = -ENOSPC;
801 				goto release_pages;
802 			}
803 
804 			dn->nid = nids[i];
805 			npage[i] = f2fs_new_node_page(dn, noffset[i]);
806 			if (IS_ERR(npage[i])) {
807 				f2fs_alloc_nid_failed(sbi, nids[i]);
808 				err = PTR_ERR(npage[i]);
809 				goto release_pages;
810 			}
811 
812 			set_nid(parent, offset[i - 1], nids[i], i == 1);
813 			f2fs_alloc_nid_done(sbi, nids[i]);
814 			done = true;
815 		} else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
816 			npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
817 			if (IS_ERR(npage[i])) {
818 				err = PTR_ERR(npage[i]);
819 				goto release_pages;
820 			}
821 			done = true;
822 		}
823 		if (i == 1) {
824 			dn->inode_page_locked = false;
825 			unlock_page(parent);
826 		} else {
827 			f2fs_put_page(parent, 1);
828 		}
829 
830 		if (!done) {
831 			npage[i] = f2fs_get_node_page(sbi, nids[i]);
832 			if (IS_ERR(npage[i])) {
833 				err = PTR_ERR(npage[i]);
834 				f2fs_put_page(npage[0], 0);
835 				goto release_out;
836 			}
837 		}
838 		if (i < level) {
839 			parent = npage[i];
840 			nids[i + 1] = get_nid(parent, offset[i], false);
841 		}
842 	}
843 	dn->nid = nids[level];
844 	dn->ofs_in_node = offset[level];
845 	dn->node_page = npage[level];
846 	dn->data_blkaddr = f2fs_data_blkaddr(dn);
847 
848 	if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) &&
849 					f2fs_sb_has_readonly(sbi)) {
850 		unsigned int c_len = f2fs_cluster_blocks_are_contiguous(dn);
851 		block_t blkaddr;
852 
853 		if (!c_len)
854 			goto out;
855 
856 		blkaddr = f2fs_data_blkaddr(dn);
857 		if (blkaddr == COMPRESS_ADDR)
858 			blkaddr = data_blkaddr(dn->inode, dn->node_page,
859 						dn->ofs_in_node + 1);
860 
861 		f2fs_update_extent_tree_range_compressed(dn->inode,
862 					index, blkaddr,
863 					F2FS_I(dn->inode)->i_cluster_size,
864 					c_len);
865 	}
866 out:
867 	return 0;
868 
869 release_pages:
870 	f2fs_put_page(parent, 1);
871 	if (i > 1)
872 		f2fs_put_page(npage[0], 0);
873 release_out:
874 	dn->inode_page = NULL;
875 	dn->node_page = NULL;
876 	if (err == -ENOENT) {
877 		dn->cur_level = i;
878 		dn->max_level = level;
879 		dn->ofs_in_node = offset[level];
880 	}
881 	return err;
882 }
883 
884 static int truncate_node(struct dnode_of_data *dn)
885 {
886 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
887 	struct node_info ni;
888 	int err;
889 	pgoff_t index;
890 
891 	err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
892 	if (err)
893 		return err;
894 
895 	/* Deallocate node address */
896 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
897 	dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
898 	set_node_addr(sbi, &ni, NULL_ADDR, false);
899 
900 	if (dn->nid == dn->inode->i_ino) {
901 		f2fs_remove_orphan_inode(sbi, dn->nid);
902 		dec_valid_inode_count(sbi);
903 		f2fs_inode_synced(dn->inode);
904 	}
905 
906 	clear_node_page_dirty(dn->node_page);
907 	set_sbi_flag(sbi, SBI_IS_DIRTY);
908 
909 	index = dn->node_page->index;
910 	f2fs_put_page(dn->node_page, 1);
911 
912 	invalidate_mapping_pages(NODE_MAPPING(sbi),
913 			index, index);
914 
915 	dn->node_page = NULL;
916 	trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
917 
918 	return 0;
919 }
920 
921 static int truncate_dnode(struct dnode_of_data *dn)
922 {
923 	struct page *page;
924 	int err;
925 
926 	if (dn->nid == 0)
927 		return 1;
928 
929 	/* get direct node */
930 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
931 	if (PTR_ERR(page) == -ENOENT)
932 		return 1;
933 	else if (IS_ERR(page))
934 		return PTR_ERR(page);
935 
936 	/* Make dnode_of_data for parameter */
937 	dn->node_page = page;
938 	dn->ofs_in_node = 0;
939 	f2fs_truncate_data_blocks(dn);
940 	err = truncate_node(dn);
941 	if (err)
942 		return err;
943 
944 	return 1;
945 }
946 
947 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
948 						int ofs, int depth)
949 {
950 	struct dnode_of_data rdn = *dn;
951 	struct page *page;
952 	struct f2fs_node *rn;
953 	nid_t child_nid;
954 	unsigned int child_nofs;
955 	int freed = 0;
956 	int i, ret;
957 
958 	if (dn->nid == 0)
959 		return NIDS_PER_BLOCK + 1;
960 
961 	trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
962 
963 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
964 	if (IS_ERR(page)) {
965 		trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
966 		return PTR_ERR(page);
967 	}
968 
969 	f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
970 
971 	rn = F2FS_NODE(page);
972 	if (depth < 3) {
973 		for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
974 			child_nid = le32_to_cpu(rn->in.nid[i]);
975 			if (child_nid == 0)
976 				continue;
977 			rdn.nid = child_nid;
978 			ret = truncate_dnode(&rdn);
979 			if (ret < 0)
980 				goto out_err;
981 			if (set_nid(page, i, 0, false))
982 				dn->node_changed = true;
983 		}
984 	} else {
985 		child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
986 		for (i = ofs; i < NIDS_PER_BLOCK; i++) {
987 			child_nid = le32_to_cpu(rn->in.nid[i]);
988 			if (child_nid == 0) {
989 				child_nofs += NIDS_PER_BLOCK + 1;
990 				continue;
991 			}
992 			rdn.nid = child_nid;
993 			ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
994 			if (ret == (NIDS_PER_BLOCK + 1)) {
995 				if (set_nid(page, i, 0, false))
996 					dn->node_changed = true;
997 				child_nofs += ret;
998 			} else if (ret < 0 && ret != -ENOENT) {
999 				goto out_err;
1000 			}
1001 		}
1002 		freed = child_nofs;
1003 	}
1004 
1005 	if (!ofs) {
1006 		/* remove current indirect node */
1007 		dn->node_page = page;
1008 		ret = truncate_node(dn);
1009 		if (ret)
1010 			goto out_err;
1011 		freed++;
1012 	} else {
1013 		f2fs_put_page(page, 1);
1014 	}
1015 	trace_f2fs_truncate_nodes_exit(dn->inode, freed);
1016 	return freed;
1017 
1018 out_err:
1019 	f2fs_put_page(page, 1);
1020 	trace_f2fs_truncate_nodes_exit(dn->inode, ret);
1021 	return ret;
1022 }
1023 
1024 static int truncate_partial_nodes(struct dnode_of_data *dn,
1025 			struct f2fs_inode *ri, int *offset, int depth)
1026 {
1027 	struct page *pages[2];
1028 	nid_t nid[3];
1029 	nid_t child_nid;
1030 	int err = 0;
1031 	int i;
1032 	int idx = depth - 2;
1033 
1034 	nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1035 	if (!nid[0])
1036 		return 0;
1037 
1038 	/* get indirect nodes in the path */
1039 	for (i = 0; i < idx + 1; i++) {
1040 		/* reference count'll be increased */
1041 		pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
1042 		if (IS_ERR(pages[i])) {
1043 			err = PTR_ERR(pages[i]);
1044 			idx = i - 1;
1045 			goto fail;
1046 		}
1047 		nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1048 	}
1049 
1050 	f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1051 
1052 	/* free direct nodes linked to a partial indirect node */
1053 	for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1054 		child_nid = get_nid(pages[idx], i, false);
1055 		if (!child_nid)
1056 			continue;
1057 		dn->nid = child_nid;
1058 		err = truncate_dnode(dn);
1059 		if (err < 0)
1060 			goto fail;
1061 		if (set_nid(pages[idx], i, 0, false))
1062 			dn->node_changed = true;
1063 	}
1064 
1065 	if (offset[idx + 1] == 0) {
1066 		dn->node_page = pages[idx];
1067 		dn->nid = nid[idx];
1068 		err = truncate_node(dn);
1069 		if (err)
1070 			goto fail;
1071 	} else {
1072 		f2fs_put_page(pages[idx], 1);
1073 	}
1074 	offset[idx]++;
1075 	offset[idx + 1] = 0;
1076 	idx--;
1077 fail:
1078 	for (i = idx; i >= 0; i--)
1079 		f2fs_put_page(pages[i], 1);
1080 
1081 	trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1082 
1083 	return err;
1084 }
1085 
1086 /*
1087  * All the block addresses of data and nodes should be nullified.
1088  */
1089 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1090 {
1091 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1092 	int err = 0, cont = 1;
1093 	int level, offset[4], noffset[4];
1094 	unsigned int nofs = 0;
1095 	struct f2fs_inode *ri;
1096 	struct dnode_of_data dn;
1097 	struct page *page;
1098 
1099 	trace_f2fs_truncate_inode_blocks_enter(inode, from);
1100 
1101 	level = get_node_path(inode, from, offset, noffset);
1102 	if (level < 0) {
1103 		trace_f2fs_truncate_inode_blocks_exit(inode, level);
1104 		return level;
1105 	}
1106 
1107 	page = f2fs_get_node_page(sbi, inode->i_ino);
1108 	if (IS_ERR(page)) {
1109 		trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1110 		return PTR_ERR(page);
1111 	}
1112 
1113 	set_new_dnode(&dn, inode, page, NULL, 0);
1114 	unlock_page(page);
1115 
1116 	ri = F2FS_INODE(page);
1117 	switch (level) {
1118 	case 0:
1119 	case 1:
1120 		nofs = noffset[1];
1121 		break;
1122 	case 2:
1123 		nofs = noffset[1];
1124 		if (!offset[level - 1])
1125 			goto skip_partial;
1126 		err = truncate_partial_nodes(&dn, ri, offset, level);
1127 		if (err < 0 && err != -ENOENT)
1128 			goto fail;
1129 		nofs += 1 + NIDS_PER_BLOCK;
1130 		break;
1131 	case 3:
1132 		nofs = 5 + 2 * NIDS_PER_BLOCK;
1133 		if (!offset[level - 1])
1134 			goto skip_partial;
1135 		err = truncate_partial_nodes(&dn, ri, offset, level);
1136 		if (err < 0 && err != -ENOENT)
1137 			goto fail;
1138 		break;
1139 	default:
1140 		BUG();
1141 	}
1142 
1143 skip_partial:
1144 	while (cont) {
1145 		dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1146 		switch (offset[0]) {
1147 		case NODE_DIR1_BLOCK:
1148 		case NODE_DIR2_BLOCK:
1149 			err = truncate_dnode(&dn);
1150 			break;
1151 
1152 		case NODE_IND1_BLOCK:
1153 		case NODE_IND2_BLOCK:
1154 			err = truncate_nodes(&dn, nofs, offset[1], 2);
1155 			break;
1156 
1157 		case NODE_DIND_BLOCK:
1158 			err = truncate_nodes(&dn, nofs, offset[1], 3);
1159 			cont = 0;
1160 			break;
1161 
1162 		default:
1163 			BUG();
1164 		}
1165 		if (err < 0 && err != -ENOENT)
1166 			goto fail;
1167 		if (offset[1] == 0 &&
1168 				ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1169 			lock_page(page);
1170 			BUG_ON(page->mapping != NODE_MAPPING(sbi));
1171 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1172 			ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1173 			set_page_dirty(page);
1174 			unlock_page(page);
1175 		}
1176 		offset[1] = 0;
1177 		offset[0]++;
1178 		nofs += err;
1179 	}
1180 fail:
1181 	f2fs_put_page(page, 0);
1182 	trace_f2fs_truncate_inode_blocks_exit(inode, err);
1183 	return err > 0 ? 0 : err;
1184 }
1185 
1186 /* caller must lock inode page */
1187 int f2fs_truncate_xattr_node(struct inode *inode)
1188 {
1189 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1190 	nid_t nid = F2FS_I(inode)->i_xattr_nid;
1191 	struct dnode_of_data dn;
1192 	struct page *npage;
1193 	int err;
1194 
1195 	if (!nid)
1196 		return 0;
1197 
1198 	npage = f2fs_get_node_page(sbi, nid);
1199 	if (IS_ERR(npage))
1200 		return PTR_ERR(npage);
1201 
1202 	set_new_dnode(&dn, inode, NULL, npage, nid);
1203 	err = truncate_node(&dn);
1204 	if (err) {
1205 		f2fs_put_page(npage, 1);
1206 		return err;
1207 	}
1208 
1209 	f2fs_i_xnid_write(inode, 0);
1210 
1211 	return 0;
1212 }
1213 
1214 /*
1215  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1216  * f2fs_unlock_op().
1217  */
1218 int f2fs_remove_inode_page(struct inode *inode)
1219 {
1220 	struct dnode_of_data dn;
1221 	int err;
1222 
1223 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1224 	err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1225 	if (err)
1226 		return err;
1227 
1228 	err = f2fs_truncate_xattr_node(inode);
1229 	if (err) {
1230 		f2fs_put_dnode(&dn);
1231 		return err;
1232 	}
1233 
1234 	/* remove potential inline_data blocks */
1235 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1236 				S_ISLNK(inode->i_mode))
1237 		f2fs_truncate_data_blocks_range(&dn, 1);
1238 
1239 	/* 0 is possible, after f2fs_new_inode() has failed */
1240 	if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1241 		f2fs_put_dnode(&dn);
1242 		return -EIO;
1243 	}
1244 
1245 	if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1246 		f2fs_warn(F2FS_I_SB(inode),
1247 			"f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1248 			inode->i_ino, (unsigned long long)inode->i_blocks);
1249 		set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1250 	}
1251 
1252 	/* will put inode & node pages */
1253 	err = truncate_node(&dn);
1254 	if (err) {
1255 		f2fs_put_dnode(&dn);
1256 		return err;
1257 	}
1258 	return 0;
1259 }
1260 
1261 struct page *f2fs_new_inode_page(struct inode *inode)
1262 {
1263 	struct dnode_of_data dn;
1264 
1265 	/* allocate inode page for new inode */
1266 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1267 
1268 	/* caller should f2fs_put_page(page, 1); */
1269 	return f2fs_new_node_page(&dn, 0);
1270 }
1271 
1272 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1273 {
1274 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1275 	struct node_info new_ni;
1276 	struct page *page;
1277 	int err;
1278 
1279 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1280 		return ERR_PTR(-EPERM);
1281 
1282 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1283 	if (!page)
1284 		return ERR_PTR(-ENOMEM);
1285 
1286 	if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1287 		goto fail;
1288 
1289 #ifdef CONFIG_F2FS_CHECK_FS
1290 	err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false);
1291 	if (err) {
1292 		dec_valid_node_count(sbi, dn->inode, !ofs);
1293 		goto fail;
1294 	}
1295 	f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1296 #endif
1297 	new_ni.nid = dn->nid;
1298 	new_ni.ino = dn->inode->i_ino;
1299 	new_ni.blk_addr = NULL_ADDR;
1300 	new_ni.flag = 0;
1301 	new_ni.version = 0;
1302 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1303 
1304 	f2fs_wait_on_page_writeback(page, NODE, true, true);
1305 	fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1306 	set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1307 	if (!PageUptodate(page))
1308 		SetPageUptodate(page);
1309 	if (set_page_dirty(page))
1310 		dn->node_changed = true;
1311 
1312 	if (f2fs_has_xattr_block(ofs))
1313 		f2fs_i_xnid_write(dn->inode, dn->nid);
1314 
1315 	if (ofs == 0)
1316 		inc_valid_inode_count(sbi);
1317 	return page;
1318 
1319 fail:
1320 	clear_node_page_dirty(page);
1321 	f2fs_put_page(page, 1);
1322 	return ERR_PTR(err);
1323 }
1324 
1325 /*
1326  * Caller should do after getting the following values.
1327  * 0: f2fs_put_page(page, 0)
1328  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1329  */
1330 static int read_node_page(struct page *page, int op_flags)
1331 {
1332 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1333 	struct node_info ni;
1334 	struct f2fs_io_info fio = {
1335 		.sbi = sbi,
1336 		.type = NODE,
1337 		.op = REQ_OP_READ,
1338 		.op_flags = op_flags,
1339 		.page = page,
1340 		.encrypted_page = NULL,
1341 	};
1342 	int err;
1343 
1344 	if (PageUptodate(page)) {
1345 		if (!f2fs_inode_chksum_verify(sbi, page)) {
1346 			ClearPageUptodate(page);
1347 			return -EFSBADCRC;
1348 		}
1349 		return LOCKED_PAGE;
1350 	}
1351 
1352 	err = f2fs_get_node_info(sbi, page->index, &ni, false);
1353 	if (err)
1354 		return err;
1355 
1356 	/* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1357 	if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR) ||
1358 			is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1359 		ClearPageUptodate(page);
1360 		return -ENOENT;
1361 	}
1362 
1363 	fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1364 
1365 	err = f2fs_submit_page_bio(&fio);
1366 
1367 	if (!err)
1368 		f2fs_update_iostat(sbi, FS_NODE_READ_IO, F2FS_BLKSIZE);
1369 
1370 	return err;
1371 }
1372 
1373 /*
1374  * Readahead a node page
1375  */
1376 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1377 {
1378 	struct page *apage;
1379 	int err;
1380 
1381 	if (!nid)
1382 		return;
1383 	if (f2fs_check_nid_range(sbi, nid))
1384 		return;
1385 
1386 	apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1387 	if (apage)
1388 		return;
1389 
1390 	apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1391 	if (!apage)
1392 		return;
1393 
1394 	err = read_node_page(apage, REQ_RAHEAD);
1395 	f2fs_put_page(apage, err ? 1 : 0);
1396 }
1397 
1398 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1399 					struct page *parent, int start)
1400 {
1401 	struct page *page;
1402 	int err;
1403 
1404 	if (!nid)
1405 		return ERR_PTR(-ENOENT);
1406 	if (f2fs_check_nid_range(sbi, nid))
1407 		return ERR_PTR(-EINVAL);
1408 repeat:
1409 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1410 	if (!page)
1411 		return ERR_PTR(-ENOMEM);
1412 
1413 	err = read_node_page(page, 0);
1414 	if (err < 0) {
1415 		goto out_put_err;
1416 	} else if (err == LOCKED_PAGE) {
1417 		err = 0;
1418 		goto page_hit;
1419 	}
1420 
1421 	if (parent)
1422 		f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1423 
1424 	lock_page(page);
1425 
1426 	if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1427 		f2fs_put_page(page, 1);
1428 		goto repeat;
1429 	}
1430 
1431 	if (unlikely(!PageUptodate(page))) {
1432 		err = -EIO;
1433 		goto out_err;
1434 	}
1435 
1436 	if (!f2fs_inode_chksum_verify(sbi, page)) {
1437 		err = -EFSBADCRC;
1438 		goto out_err;
1439 	}
1440 page_hit:
1441 	if (likely(nid == nid_of_node(page)))
1442 		return page;
1443 
1444 	f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1445 			  nid, nid_of_node(page), ino_of_node(page),
1446 			  ofs_of_node(page), cpver_of_node(page),
1447 			  next_blkaddr_of_node(page));
1448 	set_sbi_flag(sbi, SBI_NEED_FSCK);
1449 	err = -EINVAL;
1450 out_err:
1451 	ClearPageUptodate(page);
1452 out_put_err:
1453 	f2fs_handle_page_eio(sbi, page->index, NODE);
1454 	f2fs_put_page(page, 1);
1455 	return ERR_PTR(err);
1456 }
1457 
1458 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1459 {
1460 	return __get_node_page(sbi, nid, NULL, 0);
1461 }
1462 
1463 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1464 {
1465 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1466 	nid_t nid = get_nid(parent, start, false);
1467 
1468 	return __get_node_page(sbi, nid, parent, start);
1469 }
1470 
1471 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1472 {
1473 	struct inode *inode;
1474 	struct page *page;
1475 	int ret;
1476 
1477 	/* should flush inline_data before evict_inode */
1478 	inode = ilookup(sbi->sb, ino);
1479 	if (!inode)
1480 		return;
1481 
1482 	page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1483 					FGP_LOCK|FGP_NOWAIT, 0);
1484 	if (!page)
1485 		goto iput_out;
1486 
1487 	if (!PageUptodate(page))
1488 		goto page_out;
1489 
1490 	if (!PageDirty(page))
1491 		goto page_out;
1492 
1493 	if (!clear_page_dirty_for_io(page))
1494 		goto page_out;
1495 
1496 	ret = f2fs_write_inline_data(inode, page);
1497 	inode_dec_dirty_pages(inode);
1498 	f2fs_remove_dirty_inode(inode);
1499 	if (ret)
1500 		set_page_dirty(page);
1501 page_out:
1502 	f2fs_put_page(page, 1);
1503 iput_out:
1504 	iput(inode);
1505 }
1506 
1507 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1508 {
1509 	pgoff_t index;
1510 	struct pagevec pvec;
1511 	struct page *last_page = NULL;
1512 	int nr_pages;
1513 
1514 	pagevec_init(&pvec);
1515 	index = 0;
1516 
1517 	while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1518 				PAGECACHE_TAG_DIRTY))) {
1519 		int i;
1520 
1521 		for (i = 0; i < nr_pages; i++) {
1522 			struct page *page = pvec.pages[i];
1523 
1524 			if (unlikely(f2fs_cp_error(sbi))) {
1525 				f2fs_put_page(last_page, 0);
1526 				pagevec_release(&pvec);
1527 				return ERR_PTR(-EIO);
1528 			}
1529 
1530 			if (!IS_DNODE(page) || !is_cold_node(page))
1531 				continue;
1532 			if (ino_of_node(page) != ino)
1533 				continue;
1534 
1535 			lock_page(page);
1536 
1537 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1538 continue_unlock:
1539 				unlock_page(page);
1540 				continue;
1541 			}
1542 			if (ino_of_node(page) != ino)
1543 				goto continue_unlock;
1544 
1545 			if (!PageDirty(page)) {
1546 				/* someone wrote it for us */
1547 				goto continue_unlock;
1548 			}
1549 
1550 			if (last_page)
1551 				f2fs_put_page(last_page, 0);
1552 
1553 			get_page(page);
1554 			last_page = page;
1555 			unlock_page(page);
1556 		}
1557 		pagevec_release(&pvec);
1558 		cond_resched();
1559 	}
1560 	return last_page;
1561 }
1562 
1563 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1564 				struct writeback_control *wbc, bool do_balance,
1565 				enum iostat_type io_type, unsigned int *seq_id)
1566 {
1567 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1568 	nid_t nid;
1569 	struct node_info ni;
1570 	struct f2fs_io_info fio = {
1571 		.sbi = sbi,
1572 		.ino = ino_of_node(page),
1573 		.type = NODE,
1574 		.op = REQ_OP_WRITE,
1575 		.op_flags = wbc_to_write_flags(wbc),
1576 		.page = page,
1577 		.encrypted_page = NULL,
1578 		.submitted = false,
1579 		.io_type = io_type,
1580 		.io_wbc = wbc,
1581 	};
1582 	unsigned int seq;
1583 
1584 	trace_f2fs_writepage(page, NODE);
1585 
1586 	if (unlikely(f2fs_cp_error(sbi))) {
1587 		ClearPageUptodate(page);
1588 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1589 		unlock_page(page);
1590 		return 0;
1591 	}
1592 
1593 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1594 		goto redirty_out;
1595 
1596 	if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1597 			wbc->sync_mode == WB_SYNC_NONE &&
1598 			IS_DNODE(page) && is_cold_node(page))
1599 		goto redirty_out;
1600 
1601 	/* get old block addr of this node page */
1602 	nid = nid_of_node(page);
1603 	f2fs_bug_on(sbi, page->index != nid);
1604 
1605 	if (f2fs_get_node_info(sbi, nid, &ni, !do_balance))
1606 		goto redirty_out;
1607 
1608 	if (wbc->for_reclaim) {
1609 		if (!f2fs_down_read_trylock(&sbi->node_write))
1610 			goto redirty_out;
1611 	} else {
1612 		f2fs_down_read(&sbi->node_write);
1613 	}
1614 
1615 	/* This page is already truncated */
1616 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1617 		ClearPageUptodate(page);
1618 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1619 		f2fs_up_read(&sbi->node_write);
1620 		unlock_page(page);
1621 		return 0;
1622 	}
1623 
1624 	if (__is_valid_data_blkaddr(ni.blk_addr) &&
1625 		!f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1626 					DATA_GENERIC_ENHANCE)) {
1627 		f2fs_up_read(&sbi->node_write);
1628 		goto redirty_out;
1629 	}
1630 
1631 	if (atomic && !test_opt(sbi, NOBARRIER) && !f2fs_sb_has_blkzoned(sbi))
1632 		fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1633 
1634 	/* should add to global list before clearing PAGECACHE status */
1635 	if (f2fs_in_warm_node_list(sbi, page)) {
1636 		seq = f2fs_add_fsync_node_entry(sbi, page);
1637 		if (seq_id)
1638 			*seq_id = seq;
1639 	}
1640 
1641 	set_page_writeback(page);
1642 	ClearPageError(page);
1643 
1644 	fio.old_blkaddr = ni.blk_addr;
1645 	f2fs_do_write_node_page(nid, &fio);
1646 	set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1647 	dec_page_count(sbi, F2FS_DIRTY_NODES);
1648 	f2fs_up_read(&sbi->node_write);
1649 
1650 	if (wbc->for_reclaim) {
1651 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1652 		submitted = NULL;
1653 	}
1654 
1655 	unlock_page(page);
1656 
1657 	if (unlikely(f2fs_cp_error(sbi))) {
1658 		f2fs_submit_merged_write(sbi, NODE);
1659 		submitted = NULL;
1660 	}
1661 	if (submitted)
1662 		*submitted = fio.submitted;
1663 
1664 	if (do_balance)
1665 		f2fs_balance_fs(sbi, false);
1666 	return 0;
1667 
1668 redirty_out:
1669 	redirty_page_for_writepage(wbc, page);
1670 	return AOP_WRITEPAGE_ACTIVATE;
1671 }
1672 
1673 int f2fs_move_node_page(struct page *node_page, int gc_type)
1674 {
1675 	int err = 0;
1676 
1677 	if (gc_type == FG_GC) {
1678 		struct writeback_control wbc = {
1679 			.sync_mode = WB_SYNC_ALL,
1680 			.nr_to_write = 1,
1681 			.for_reclaim = 0,
1682 		};
1683 
1684 		f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1685 
1686 		set_page_dirty(node_page);
1687 
1688 		if (!clear_page_dirty_for_io(node_page)) {
1689 			err = -EAGAIN;
1690 			goto out_page;
1691 		}
1692 
1693 		if (__write_node_page(node_page, false, NULL,
1694 					&wbc, false, FS_GC_NODE_IO, NULL)) {
1695 			err = -EAGAIN;
1696 			unlock_page(node_page);
1697 		}
1698 		goto release_page;
1699 	} else {
1700 		/* set page dirty and write it */
1701 		if (!PageWriteback(node_page))
1702 			set_page_dirty(node_page);
1703 	}
1704 out_page:
1705 	unlock_page(node_page);
1706 release_page:
1707 	f2fs_put_page(node_page, 0);
1708 	return err;
1709 }
1710 
1711 static int f2fs_write_node_page(struct page *page,
1712 				struct writeback_control *wbc)
1713 {
1714 	return __write_node_page(page, false, NULL, wbc, false,
1715 						FS_NODE_IO, NULL);
1716 }
1717 
1718 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1719 			struct writeback_control *wbc, bool atomic,
1720 			unsigned int *seq_id)
1721 {
1722 	pgoff_t index;
1723 	struct pagevec pvec;
1724 	int ret = 0;
1725 	struct page *last_page = NULL;
1726 	bool marked = false;
1727 	nid_t ino = inode->i_ino;
1728 	int nr_pages;
1729 	int nwritten = 0;
1730 
1731 	if (atomic) {
1732 		last_page = last_fsync_dnode(sbi, ino);
1733 		if (IS_ERR_OR_NULL(last_page))
1734 			return PTR_ERR_OR_ZERO(last_page);
1735 	}
1736 retry:
1737 	pagevec_init(&pvec);
1738 	index = 0;
1739 
1740 	while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1741 				PAGECACHE_TAG_DIRTY))) {
1742 		int i;
1743 
1744 		for (i = 0; i < nr_pages; i++) {
1745 			struct page *page = pvec.pages[i];
1746 			bool submitted = false;
1747 
1748 			if (unlikely(f2fs_cp_error(sbi))) {
1749 				f2fs_put_page(last_page, 0);
1750 				pagevec_release(&pvec);
1751 				ret = -EIO;
1752 				goto out;
1753 			}
1754 
1755 			if (!IS_DNODE(page) || !is_cold_node(page))
1756 				continue;
1757 			if (ino_of_node(page) != ino)
1758 				continue;
1759 
1760 			lock_page(page);
1761 
1762 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1763 continue_unlock:
1764 				unlock_page(page);
1765 				continue;
1766 			}
1767 			if (ino_of_node(page) != ino)
1768 				goto continue_unlock;
1769 
1770 			if (!PageDirty(page) && page != last_page) {
1771 				/* someone wrote it for us */
1772 				goto continue_unlock;
1773 			}
1774 
1775 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1776 
1777 			set_fsync_mark(page, 0);
1778 			set_dentry_mark(page, 0);
1779 
1780 			if (!atomic || page == last_page) {
1781 				set_fsync_mark(page, 1);
1782 				percpu_counter_inc(&sbi->rf_node_block_count);
1783 				if (IS_INODE(page)) {
1784 					if (is_inode_flag_set(inode,
1785 								FI_DIRTY_INODE))
1786 						f2fs_update_inode(inode, page);
1787 					set_dentry_mark(page,
1788 						f2fs_need_dentry_mark(sbi, ino));
1789 				}
1790 				/* may be written by other thread */
1791 				if (!PageDirty(page))
1792 					set_page_dirty(page);
1793 			}
1794 
1795 			if (!clear_page_dirty_for_io(page))
1796 				goto continue_unlock;
1797 
1798 			ret = __write_node_page(page, atomic &&
1799 						page == last_page,
1800 						&submitted, wbc, true,
1801 						FS_NODE_IO, seq_id);
1802 			if (ret) {
1803 				unlock_page(page);
1804 				f2fs_put_page(last_page, 0);
1805 				break;
1806 			} else if (submitted) {
1807 				nwritten++;
1808 			}
1809 
1810 			if (page == last_page) {
1811 				f2fs_put_page(page, 0);
1812 				marked = true;
1813 				break;
1814 			}
1815 		}
1816 		pagevec_release(&pvec);
1817 		cond_resched();
1818 
1819 		if (ret || marked)
1820 			break;
1821 	}
1822 	if (!ret && atomic && !marked) {
1823 		f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1824 			   ino, last_page->index);
1825 		lock_page(last_page);
1826 		f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1827 		set_page_dirty(last_page);
1828 		unlock_page(last_page);
1829 		goto retry;
1830 	}
1831 out:
1832 	if (nwritten)
1833 		f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1834 	return ret ? -EIO : 0;
1835 }
1836 
1837 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1838 {
1839 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1840 	bool clean;
1841 
1842 	if (inode->i_ino != ino)
1843 		return 0;
1844 
1845 	if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1846 		return 0;
1847 
1848 	spin_lock(&sbi->inode_lock[DIRTY_META]);
1849 	clean = list_empty(&F2FS_I(inode)->gdirty_list);
1850 	spin_unlock(&sbi->inode_lock[DIRTY_META]);
1851 
1852 	if (clean)
1853 		return 0;
1854 
1855 	inode = igrab(inode);
1856 	if (!inode)
1857 		return 0;
1858 	return 1;
1859 }
1860 
1861 static bool flush_dirty_inode(struct page *page)
1862 {
1863 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1864 	struct inode *inode;
1865 	nid_t ino = ino_of_node(page);
1866 
1867 	inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1868 	if (!inode)
1869 		return false;
1870 
1871 	f2fs_update_inode(inode, page);
1872 	unlock_page(page);
1873 
1874 	iput(inode);
1875 	return true;
1876 }
1877 
1878 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1879 {
1880 	pgoff_t index = 0;
1881 	struct pagevec pvec;
1882 	int nr_pages;
1883 
1884 	pagevec_init(&pvec);
1885 
1886 	while ((nr_pages = pagevec_lookup_tag(&pvec,
1887 			NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1888 		int i;
1889 
1890 		for (i = 0; i < nr_pages; i++) {
1891 			struct page *page = pvec.pages[i];
1892 
1893 			if (!IS_DNODE(page))
1894 				continue;
1895 
1896 			lock_page(page);
1897 
1898 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1899 continue_unlock:
1900 				unlock_page(page);
1901 				continue;
1902 			}
1903 
1904 			if (!PageDirty(page)) {
1905 				/* someone wrote it for us */
1906 				goto continue_unlock;
1907 			}
1908 
1909 			/* flush inline_data, if it's async context. */
1910 			if (page_private_inline(page)) {
1911 				clear_page_private_inline(page);
1912 				unlock_page(page);
1913 				flush_inline_data(sbi, ino_of_node(page));
1914 				continue;
1915 			}
1916 			unlock_page(page);
1917 		}
1918 		pagevec_release(&pvec);
1919 		cond_resched();
1920 	}
1921 }
1922 
1923 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1924 				struct writeback_control *wbc,
1925 				bool do_balance, enum iostat_type io_type)
1926 {
1927 	pgoff_t index;
1928 	struct pagevec pvec;
1929 	int step = 0;
1930 	int nwritten = 0;
1931 	int ret = 0;
1932 	int nr_pages, done = 0;
1933 
1934 	pagevec_init(&pvec);
1935 
1936 next_step:
1937 	index = 0;
1938 
1939 	while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1940 			NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1941 		int i;
1942 
1943 		for (i = 0; i < nr_pages; i++) {
1944 			struct page *page = pvec.pages[i];
1945 			bool submitted = false;
1946 			bool may_dirty = true;
1947 
1948 			/* give a priority to WB_SYNC threads */
1949 			if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1950 					wbc->sync_mode == WB_SYNC_NONE) {
1951 				done = 1;
1952 				break;
1953 			}
1954 
1955 			/*
1956 			 * flushing sequence with step:
1957 			 * 0. indirect nodes
1958 			 * 1. dentry dnodes
1959 			 * 2. file dnodes
1960 			 */
1961 			if (step == 0 && IS_DNODE(page))
1962 				continue;
1963 			if (step == 1 && (!IS_DNODE(page) ||
1964 						is_cold_node(page)))
1965 				continue;
1966 			if (step == 2 && (!IS_DNODE(page) ||
1967 						!is_cold_node(page)))
1968 				continue;
1969 lock_node:
1970 			if (wbc->sync_mode == WB_SYNC_ALL)
1971 				lock_page(page);
1972 			else if (!trylock_page(page))
1973 				continue;
1974 
1975 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1976 continue_unlock:
1977 				unlock_page(page);
1978 				continue;
1979 			}
1980 
1981 			if (!PageDirty(page)) {
1982 				/* someone wrote it for us */
1983 				goto continue_unlock;
1984 			}
1985 
1986 			/* flush inline_data/inode, if it's async context. */
1987 			if (!do_balance)
1988 				goto write_node;
1989 
1990 			/* flush inline_data */
1991 			if (page_private_inline(page)) {
1992 				clear_page_private_inline(page);
1993 				unlock_page(page);
1994 				flush_inline_data(sbi, ino_of_node(page));
1995 				goto lock_node;
1996 			}
1997 
1998 			/* flush dirty inode */
1999 			if (IS_INODE(page) && may_dirty) {
2000 				may_dirty = false;
2001 				if (flush_dirty_inode(page))
2002 					goto lock_node;
2003 			}
2004 write_node:
2005 			f2fs_wait_on_page_writeback(page, NODE, true, true);
2006 
2007 			if (!clear_page_dirty_for_io(page))
2008 				goto continue_unlock;
2009 
2010 			set_fsync_mark(page, 0);
2011 			set_dentry_mark(page, 0);
2012 
2013 			ret = __write_node_page(page, false, &submitted,
2014 						wbc, do_balance, io_type, NULL);
2015 			if (ret)
2016 				unlock_page(page);
2017 			else if (submitted)
2018 				nwritten++;
2019 
2020 			if (--wbc->nr_to_write == 0)
2021 				break;
2022 		}
2023 		pagevec_release(&pvec);
2024 		cond_resched();
2025 
2026 		if (wbc->nr_to_write == 0) {
2027 			step = 2;
2028 			break;
2029 		}
2030 	}
2031 
2032 	if (step < 2) {
2033 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2034 				wbc->sync_mode == WB_SYNC_NONE && step == 1)
2035 			goto out;
2036 		step++;
2037 		goto next_step;
2038 	}
2039 out:
2040 	if (nwritten)
2041 		f2fs_submit_merged_write(sbi, NODE);
2042 
2043 	if (unlikely(f2fs_cp_error(sbi)))
2044 		return -EIO;
2045 	return ret;
2046 }
2047 
2048 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2049 						unsigned int seq_id)
2050 {
2051 	struct fsync_node_entry *fn;
2052 	struct page *page;
2053 	struct list_head *head = &sbi->fsync_node_list;
2054 	unsigned long flags;
2055 	unsigned int cur_seq_id = 0;
2056 	int ret2, ret = 0;
2057 
2058 	while (seq_id && cur_seq_id < seq_id) {
2059 		spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2060 		if (list_empty(head)) {
2061 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2062 			break;
2063 		}
2064 		fn = list_first_entry(head, struct fsync_node_entry, list);
2065 		if (fn->seq_id > seq_id) {
2066 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2067 			break;
2068 		}
2069 		cur_seq_id = fn->seq_id;
2070 		page = fn->page;
2071 		get_page(page);
2072 		spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2073 
2074 		f2fs_wait_on_page_writeback(page, NODE, true, false);
2075 		if (TestClearPageError(page))
2076 			ret = -EIO;
2077 
2078 		put_page(page);
2079 
2080 		if (ret)
2081 			break;
2082 	}
2083 
2084 	ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2085 	if (!ret)
2086 		ret = ret2;
2087 
2088 	return ret;
2089 }
2090 
2091 static int f2fs_write_node_pages(struct address_space *mapping,
2092 			    struct writeback_control *wbc)
2093 {
2094 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2095 	struct blk_plug plug;
2096 	long diff;
2097 
2098 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2099 		goto skip_write;
2100 
2101 	/* balancing f2fs's metadata in background */
2102 	f2fs_balance_fs_bg(sbi, true);
2103 
2104 	/* collect a number of dirty node pages and write together */
2105 	if (wbc->sync_mode != WB_SYNC_ALL &&
2106 			get_pages(sbi, F2FS_DIRTY_NODES) <
2107 					nr_pages_to_skip(sbi, NODE))
2108 		goto skip_write;
2109 
2110 	if (wbc->sync_mode == WB_SYNC_ALL)
2111 		atomic_inc(&sbi->wb_sync_req[NODE]);
2112 	else if (atomic_read(&sbi->wb_sync_req[NODE])) {
2113 		/* to avoid potential deadlock */
2114 		if (current->plug)
2115 			blk_finish_plug(current->plug);
2116 		goto skip_write;
2117 	}
2118 
2119 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2120 
2121 	diff = nr_pages_to_write(sbi, NODE, wbc);
2122 	blk_start_plug(&plug);
2123 	f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2124 	blk_finish_plug(&plug);
2125 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2126 
2127 	if (wbc->sync_mode == WB_SYNC_ALL)
2128 		atomic_dec(&sbi->wb_sync_req[NODE]);
2129 	return 0;
2130 
2131 skip_write:
2132 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2133 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2134 	return 0;
2135 }
2136 
2137 static bool f2fs_dirty_node_folio(struct address_space *mapping,
2138 		struct folio *folio)
2139 {
2140 	trace_f2fs_set_page_dirty(&folio->page, NODE);
2141 
2142 	if (!folio_test_uptodate(folio))
2143 		folio_mark_uptodate(folio);
2144 #ifdef CONFIG_F2FS_CHECK_FS
2145 	if (IS_INODE(&folio->page))
2146 		f2fs_inode_chksum_set(F2FS_M_SB(mapping), &folio->page);
2147 #endif
2148 	if (!folio_test_dirty(folio)) {
2149 		filemap_dirty_folio(mapping, folio);
2150 		inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
2151 		set_page_private_reference(&folio->page);
2152 		return true;
2153 	}
2154 	return false;
2155 }
2156 
2157 /*
2158  * Structure of the f2fs node operations
2159  */
2160 const struct address_space_operations f2fs_node_aops = {
2161 	.writepage	= f2fs_write_node_page,
2162 	.writepages	= f2fs_write_node_pages,
2163 	.dirty_folio	= f2fs_dirty_node_folio,
2164 	.invalidate_folio = f2fs_invalidate_folio,
2165 	.release_folio	= f2fs_release_folio,
2166 #ifdef CONFIG_MIGRATION
2167 	.migratepage	= f2fs_migrate_page,
2168 #endif
2169 };
2170 
2171 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2172 						nid_t n)
2173 {
2174 	return radix_tree_lookup(&nm_i->free_nid_root, n);
2175 }
2176 
2177 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2178 				struct free_nid *i)
2179 {
2180 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2181 	int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2182 
2183 	if (err)
2184 		return err;
2185 
2186 	nm_i->nid_cnt[FREE_NID]++;
2187 	list_add_tail(&i->list, &nm_i->free_nid_list);
2188 	return 0;
2189 }
2190 
2191 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2192 			struct free_nid *i, enum nid_state state)
2193 {
2194 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2195 
2196 	f2fs_bug_on(sbi, state != i->state);
2197 	nm_i->nid_cnt[state]--;
2198 	if (state == FREE_NID)
2199 		list_del(&i->list);
2200 	radix_tree_delete(&nm_i->free_nid_root, i->nid);
2201 }
2202 
2203 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2204 			enum nid_state org_state, enum nid_state dst_state)
2205 {
2206 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2207 
2208 	f2fs_bug_on(sbi, org_state != i->state);
2209 	i->state = dst_state;
2210 	nm_i->nid_cnt[org_state]--;
2211 	nm_i->nid_cnt[dst_state]++;
2212 
2213 	switch (dst_state) {
2214 	case PREALLOC_NID:
2215 		list_del(&i->list);
2216 		break;
2217 	case FREE_NID:
2218 		list_add_tail(&i->list, &nm_i->free_nid_list);
2219 		break;
2220 	default:
2221 		BUG_ON(1);
2222 	}
2223 }
2224 
2225 bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi)
2226 {
2227 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2228 	unsigned int i;
2229 	bool ret = true;
2230 
2231 	f2fs_down_read(&nm_i->nat_tree_lock);
2232 	for (i = 0; i < nm_i->nat_blocks; i++) {
2233 		if (!test_bit_le(i, nm_i->nat_block_bitmap)) {
2234 			ret = false;
2235 			break;
2236 		}
2237 	}
2238 	f2fs_up_read(&nm_i->nat_tree_lock);
2239 
2240 	return ret;
2241 }
2242 
2243 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2244 							bool set, bool build)
2245 {
2246 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2247 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2248 	unsigned int nid_ofs = nid - START_NID(nid);
2249 
2250 	if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2251 		return;
2252 
2253 	if (set) {
2254 		if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2255 			return;
2256 		__set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2257 		nm_i->free_nid_count[nat_ofs]++;
2258 	} else {
2259 		if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2260 			return;
2261 		__clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2262 		if (!build)
2263 			nm_i->free_nid_count[nat_ofs]--;
2264 	}
2265 }
2266 
2267 /* return if the nid is recognized as free */
2268 static bool add_free_nid(struct f2fs_sb_info *sbi,
2269 				nid_t nid, bool build, bool update)
2270 {
2271 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2272 	struct free_nid *i, *e;
2273 	struct nat_entry *ne;
2274 	int err = -EINVAL;
2275 	bool ret = false;
2276 
2277 	/* 0 nid should not be used */
2278 	if (unlikely(nid == 0))
2279 		return false;
2280 
2281 	if (unlikely(f2fs_check_nid_range(sbi, nid)))
2282 		return false;
2283 
2284 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL);
2285 	i->nid = nid;
2286 	i->state = FREE_NID;
2287 
2288 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2289 
2290 	spin_lock(&nm_i->nid_list_lock);
2291 
2292 	if (build) {
2293 		/*
2294 		 *   Thread A             Thread B
2295 		 *  - f2fs_create
2296 		 *   - f2fs_new_inode
2297 		 *    - f2fs_alloc_nid
2298 		 *     - __insert_nid_to_list(PREALLOC_NID)
2299 		 *                     - f2fs_balance_fs_bg
2300 		 *                      - f2fs_build_free_nids
2301 		 *                       - __f2fs_build_free_nids
2302 		 *                        - scan_nat_page
2303 		 *                         - add_free_nid
2304 		 *                          - __lookup_nat_cache
2305 		 *  - f2fs_add_link
2306 		 *   - f2fs_init_inode_metadata
2307 		 *    - f2fs_new_inode_page
2308 		 *     - f2fs_new_node_page
2309 		 *      - set_node_addr
2310 		 *  - f2fs_alloc_nid_done
2311 		 *   - __remove_nid_from_list(PREALLOC_NID)
2312 		 *                         - __insert_nid_to_list(FREE_NID)
2313 		 */
2314 		ne = __lookup_nat_cache(nm_i, nid);
2315 		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2316 				nat_get_blkaddr(ne) != NULL_ADDR))
2317 			goto err_out;
2318 
2319 		e = __lookup_free_nid_list(nm_i, nid);
2320 		if (e) {
2321 			if (e->state == FREE_NID)
2322 				ret = true;
2323 			goto err_out;
2324 		}
2325 	}
2326 	ret = true;
2327 	err = __insert_free_nid(sbi, i);
2328 err_out:
2329 	if (update) {
2330 		update_free_nid_bitmap(sbi, nid, ret, build);
2331 		if (!build)
2332 			nm_i->available_nids++;
2333 	}
2334 	spin_unlock(&nm_i->nid_list_lock);
2335 	radix_tree_preload_end();
2336 
2337 	if (err)
2338 		kmem_cache_free(free_nid_slab, i);
2339 	return ret;
2340 }
2341 
2342 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2343 {
2344 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2345 	struct free_nid *i;
2346 	bool need_free = false;
2347 
2348 	spin_lock(&nm_i->nid_list_lock);
2349 	i = __lookup_free_nid_list(nm_i, nid);
2350 	if (i && i->state == FREE_NID) {
2351 		__remove_free_nid(sbi, i, FREE_NID);
2352 		need_free = true;
2353 	}
2354 	spin_unlock(&nm_i->nid_list_lock);
2355 
2356 	if (need_free)
2357 		kmem_cache_free(free_nid_slab, i);
2358 }
2359 
2360 static int scan_nat_page(struct f2fs_sb_info *sbi,
2361 			struct page *nat_page, nid_t start_nid)
2362 {
2363 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2364 	struct f2fs_nat_block *nat_blk = page_address(nat_page);
2365 	block_t blk_addr;
2366 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2367 	int i;
2368 
2369 	__set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2370 
2371 	i = start_nid % NAT_ENTRY_PER_BLOCK;
2372 
2373 	for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2374 		if (unlikely(start_nid >= nm_i->max_nid))
2375 			break;
2376 
2377 		blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2378 
2379 		if (blk_addr == NEW_ADDR)
2380 			return -EINVAL;
2381 
2382 		if (blk_addr == NULL_ADDR) {
2383 			add_free_nid(sbi, start_nid, true, true);
2384 		} else {
2385 			spin_lock(&NM_I(sbi)->nid_list_lock);
2386 			update_free_nid_bitmap(sbi, start_nid, false, true);
2387 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2388 		}
2389 	}
2390 
2391 	return 0;
2392 }
2393 
2394 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2395 {
2396 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2397 	struct f2fs_journal *journal = curseg->journal;
2398 	int i;
2399 
2400 	down_read(&curseg->journal_rwsem);
2401 	for (i = 0; i < nats_in_cursum(journal); i++) {
2402 		block_t addr;
2403 		nid_t nid;
2404 
2405 		addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2406 		nid = le32_to_cpu(nid_in_journal(journal, i));
2407 		if (addr == NULL_ADDR)
2408 			add_free_nid(sbi, nid, true, false);
2409 		else
2410 			remove_free_nid(sbi, nid);
2411 	}
2412 	up_read(&curseg->journal_rwsem);
2413 }
2414 
2415 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2416 {
2417 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2418 	unsigned int i, idx;
2419 	nid_t nid;
2420 
2421 	f2fs_down_read(&nm_i->nat_tree_lock);
2422 
2423 	for (i = 0; i < nm_i->nat_blocks; i++) {
2424 		if (!test_bit_le(i, nm_i->nat_block_bitmap))
2425 			continue;
2426 		if (!nm_i->free_nid_count[i])
2427 			continue;
2428 		for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2429 			idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2430 						NAT_ENTRY_PER_BLOCK, idx);
2431 			if (idx >= NAT_ENTRY_PER_BLOCK)
2432 				break;
2433 
2434 			nid = i * NAT_ENTRY_PER_BLOCK + idx;
2435 			add_free_nid(sbi, nid, true, false);
2436 
2437 			if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2438 				goto out;
2439 		}
2440 	}
2441 out:
2442 	scan_curseg_cache(sbi);
2443 
2444 	f2fs_up_read(&nm_i->nat_tree_lock);
2445 }
2446 
2447 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2448 						bool sync, bool mount)
2449 {
2450 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2451 	int i = 0, ret;
2452 	nid_t nid = nm_i->next_scan_nid;
2453 
2454 	if (unlikely(nid >= nm_i->max_nid))
2455 		nid = 0;
2456 
2457 	if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2458 		nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2459 
2460 	/* Enough entries */
2461 	if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2462 		return 0;
2463 
2464 	if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2465 		return 0;
2466 
2467 	if (!mount) {
2468 		/* try to find free nids in free_nid_bitmap */
2469 		scan_free_nid_bits(sbi);
2470 
2471 		if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2472 			return 0;
2473 	}
2474 
2475 	/* readahead nat pages to be scanned */
2476 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2477 							META_NAT, true);
2478 
2479 	f2fs_down_read(&nm_i->nat_tree_lock);
2480 
2481 	while (1) {
2482 		if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2483 						nm_i->nat_block_bitmap)) {
2484 			struct page *page = get_current_nat_page(sbi, nid);
2485 
2486 			if (IS_ERR(page)) {
2487 				ret = PTR_ERR(page);
2488 			} else {
2489 				ret = scan_nat_page(sbi, page, nid);
2490 				f2fs_put_page(page, 1);
2491 			}
2492 
2493 			if (ret) {
2494 				f2fs_up_read(&nm_i->nat_tree_lock);
2495 				f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2496 				return ret;
2497 			}
2498 		}
2499 
2500 		nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2501 		if (unlikely(nid >= nm_i->max_nid))
2502 			nid = 0;
2503 
2504 		if (++i >= FREE_NID_PAGES)
2505 			break;
2506 	}
2507 
2508 	/* go to the next free nat pages to find free nids abundantly */
2509 	nm_i->next_scan_nid = nid;
2510 
2511 	/* find free nids from current sum_pages */
2512 	scan_curseg_cache(sbi);
2513 
2514 	f2fs_up_read(&nm_i->nat_tree_lock);
2515 
2516 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2517 					nm_i->ra_nid_pages, META_NAT, false);
2518 
2519 	return 0;
2520 }
2521 
2522 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2523 {
2524 	int ret;
2525 
2526 	mutex_lock(&NM_I(sbi)->build_lock);
2527 	ret = __f2fs_build_free_nids(sbi, sync, mount);
2528 	mutex_unlock(&NM_I(sbi)->build_lock);
2529 
2530 	return ret;
2531 }
2532 
2533 /*
2534  * If this function returns success, caller can obtain a new nid
2535  * from second parameter of this function.
2536  * The returned nid could be used ino as well as nid when inode is created.
2537  */
2538 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2539 {
2540 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2541 	struct free_nid *i = NULL;
2542 retry:
2543 	if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2544 		f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2545 		return false;
2546 	}
2547 
2548 	spin_lock(&nm_i->nid_list_lock);
2549 
2550 	if (unlikely(nm_i->available_nids == 0)) {
2551 		spin_unlock(&nm_i->nid_list_lock);
2552 		return false;
2553 	}
2554 
2555 	/* We should not use stale free nids created by f2fs_build_free_nids */
2556 	if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2557 		f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2558 		i = list_first_entry(&nm_i->free_nid_list,
2559 					struct free_nid, list);
2560 		*nid = i->nid;
2561 
2562 		__move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2563 		nm_i->available_nids--;
2564 
2565 		update_free_nid_bitmap(sbi, *nid, false, false);
2566 
2567 		spin_unlock(&nm_i->nid_list_lock);
2568 		return true;
2569 	}
2570 	spin_unlock(&nm_i->nid_list_lock);
2571 
2572 	/* Let's scan nat pages and its caches to get free nids */
2573 	if (!f2fs_build_free_nids(sbi, true, false))
2574 		goto retry;
2575 	return false;
2576 }
2577 
2578 /*
2579  * f2fs_alloc_nid() should be called prior to this function.
2580  */
2581 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2582 {
2583 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2584 	struct free_nid *i;
2585 
2586 	spin_lock(&nm_i->nid_list_lock);
2587 	i = __lookup_free_nid_list(nm_i, nid);
2588 	f2fs_bug_on(sbi, !i);
2589 	__remove_free_nid(sbi, i, PREALLOC_NID);
2590 	spin_unlock(&nm_i->nid_list_lock);
2591 
2592 	kmem_cache_free(free_nid_slab, i);
2593 }
2594 
2595 /*
2596  * f2fs_alloc_nid() should be called prior to this function.
2597  */
2598 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2599 {
2600 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2601 	struct free_nid *i;
2602 	bool need_free = false;
2603 
2604 	if (!nid)
2605 		return;
2606 
2607 	spin_lock(&nm_i->nid_list_lock);
2608 	i = __lookup_free_nid_list(nm_i, nid);
2609 	f2fs_bug_on(sbi, !i);
2610 
2611 	if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2612 		__remove_free_nid(sbi, i, PREALLOC_NID);
2613 		need_free = true;
2614 	} else {
2615 		__move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2616 	}
2617 
2618 	nm_i->available_nids++;
2619 
2620 	update_free_nid_bitmap(sbi, nid, true, false);
2621 
2622 	spin_unlock(&nm_i->nid_list_lock);
2623 
2624 	if (need_free)
2625 		kmem_cache_free(free_nid_slab, i);
2626 }
2627 
2628 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2629 {
2630 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2631 	int nr = nr_shrink;
2632 
2633 	if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2634 		return 0;
2635 
2636 	if (!mutex_trylock(&nm_i->build_lock))
2637 		return 0;
2638 
2639 	while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2640 		struct free_nid *i, *next;
2641 		unsigned int batch = SHRINK_NID_BATCH_SIZE;
2642 
2643 		spin_lock(&nm_i->nid_list_lock);
2644 		list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2645 			if (!nr_shrink || !batch ||
2646 				nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2647 				break;
2648 			__remove_free_nid(sbi, i, FREE_NID);
2649 			kmem_cache_free(free_nid_slab, i);
2650 			nr_shrink--;
2651 			batch--;
2652 		}
2653 		spin_unlock(&nm_i->nid_list_lock);
2654 	}
2655 
2656 	mutex_unlock(&nm_i->build_lock);
2657 
2658 	return nr - nr_shrink;
2659 }
2660 
2661 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2662 {
2663 	void *src_addr, *dst_addr;
2664 	size_t inline_size;
2665 	struct page *ipage;
2666 	struct f2fs_inode *ri;
2667 
2668 	ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2669 	if (IS_ERR(ipage))
2670 		return PTR_ERR(ipage);
2671 
2672 	ri = F2FS_INODE(page);
2673 	if (ri->i_inline & F2FS_INLINE_XATTR) {
2674 		if (!f2fs_has_inline_xattr(inode)) {
2675 			set_inode_flag(inode, FI_INLINE_XATTR);
2676 			stat_inc_inline_xattr(inode);
2677 		}
2678 	} else {
2679 		if (f2fs_has_inline_xattr(inode)) {
2680 			stat_dec_inline_xattr(inode);
2681 			clear_inode_flag(inode, FI_INLINE_XATTR);
2682 		}
2683 		goto update_inode;
2684 	}
2685 
2686 	dst_addr = inline_xattr_addr(inode, ipage);
2687 	src_addr = inline_xattr_addr(inode, page);
2688 	inline_size = inline_xattr_size(inode);
2689 
2690 	f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2691 	memcpy(dst_addr, src_addr, inline_size);
2692 update_inode:
2693 	f2fs_update_inode(inode, ipage);
2694 	f2fs_put_page(ipage, 1);
2695 	return 0;
2696 }
2697 
2698 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2699 {
2700 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2701 	nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2702 	nid_t new_xnid;
2703 	struct dnode_of_data dn;
2704 	struct node_info ni;
2705 	struct page *xpage;
2706 	int err;
2707 
2708 	if (!prev_xnid)
2709 		goto recover_xnid;
2710 
2711 	/* 1: invalidate the previous xattr nid */
2712 	err = f2fs_get_node_info(sbi, prev_xnid, &ni, false);
2713 	if (err)
2714 		return err;
2715 
2716 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
2717 	dec_valid_node_count(sbi, inode, false);
2718 	set_node_addr(sbi, &ni, NULL_ADDR, false);
2719 
2720 recover_xnid:
2721 	/* 2: update xattr nid in inode */
2722 	if (!f2fs_alloc_nid(sbi, &new_xnid))
2723 		return -ENOSPC;
2724 
2725 	set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2726 	xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2727 	if (IS_ERR(xpage)) {
2728 		f2fs_alloc_nid_failed(sbi, new_xnid);
2729 		return PTR_ERR(xpage);
2730 	}
2731 
2732 	f2fs_alloc_nid_done(sbi, new_xnid);
2733 	f2fs_update_inode_page(inode);
2734 
2735 	/* 3: update and set xattr node page dirty */
2736 	memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2737 
2738 	set_page_dirty(xpage);
2739 	f2fs_put_page(xpage, 1);
2740 
2741 	return 0;
2742 }
2743 
2744 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2745 {
2746 	struct f2fs_inode *src, *dst;
2747 	nid_t ino = ino_of_node(page);
2748 	struct node_info old_ni, new_ni;
2749 	struct page *ipage;
2750 	int err;
2751 
2752 	err = f2fs_get_node_info(sbi, ino, &old_ni, false);
2753 	if (err)
2754 		return err;
2755 
2756 	if (unlikely(old_ni.blk_addr != NULL_ADDR))
2757 		return -EINVAL;
2758 retry:
2759 	ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2760 	if (!ipage) {
2761 		memalloc_retry_wait(GFP_NOFS);
2762 		goto retry;
2763 	}
2764 
2765 	/* Should not use this inode from free nid list */
2766 	remove_free_nid(sbi, ino);
2767 
2768 	if (!PageUptodate(ipage))
2769 		SetPageUptodate(ipage);
2770 	fill_node_footer(ipage, ino, ino, 0, true);
2771 	set_cold_node(ipage, false);
2772 
2773 	src = F2FS_INODE(page);
2774 	dst = F2FS_INODE(ipage);
2775 
2776 	memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2777 	dst->i_size = 0;
2778 	dst->i_blocks = cpu_to_le64(1);
2779 	dst->i_links = cpu_to_le32(1);
2780 	dst->i_xattr_nid = 0;
2781 	dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2782 	if (dst->i_inline & F2FS_EXTRA_ATTR) {
2783 		dst->i_extra_isize = src->i_extra_isize;
2784 
2785 		if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2786 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2787 							i_inline_xattr_size))
2788 			dst->i_inline_xattr_size = src->i_inline_xattr_size;
2789 
2790 		if (f2fs_sb_has_project_quota(sbi) &&
2791 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2792 								i_projid))
2793 			dst->i_projid = src->i_projid;
2794 
2795 		if (f2fs_sb_has_inode_crtime(sbi) &&
2796 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2797 							i_crtime_nsec)) {
2798 			dst->i_crtime = src->i_crtime;
2799 			dst->i_crtime_nsec = src->i_crtime_nsec;
2800 		}
2801 	}
2802 
2803 	new_ni = old_ni;
2804 	new_ni.ino = ino;
2805 
2806 	if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2807 		WARN_ON(1);
2808 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2809 	inc_valid_inode_count(sbi);
2810 	set_page_dirty(ipage);
2811 	f2fs_put_page(ipage, 1);
2812 	return 0;
2813 }
2814 
2815 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2816 			unsigned int segno, struct f2fs_summary_block *sum)
2817 {
2818 	struct f2fs_node *rn;
2819 	struct f2fs_summary *sum_entry;
2820 	block_t addr;
2821 	int i, idx, last_offset, nrpages;
2822 
2823 	/* scan the node segment */
2824 	last_offset = sbi->blocks_per_seg;
2825 	addr = START_BLOCK(sbi, segno);
2826 	sum_entry = &sum->entries[0];
2827 
2828 	for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2829 		nrpages = bio_max_segs(last_offset - i);
2830 
2831 		/* readahead node pages */
2832 		f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2833 
2834 		for (idx = addr; idx < addr + nrpages; idx++) {
2835 			struct page *page = f2fs_get_tmp_page(sbi, idx);
2836 
2837 			if (IS_ERR(page))
2838 				return PTR_ERR(page);
2839 
2840 			rn = F2FS_NODE(page);
2841 			sum_entry->nid = rn->footer.nid;
2842 			sum_entry->version = 0;
2843 			sum_entry->ofs_in_node = 0;
2844 			sum_entry++;
2845 			f2fs_put_page(page, 1);
2846 		}
2847 
2848 		invalidate_mapping_pages(META_MAPPING(sbi), addr,
2849 							addr + nrpages);
2850 	}
2851 	return 0;
2852 }
2853 
2854 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2855 {
2856 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2857 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2858 	struct f2fs_journal *journal = curseg->journal;
2859 	int i;
2860 
2861 	down_write(&curseg->journal_rwsem);
2862 	for (i = 0; i < nats_in_cursum(journal); i++) {
2863 		struct nat_entry *ne;
2864 		struct f2fs_nat_entry raw_ne;
2865 		nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2866 
2867 		if (f2fs_check_nid_range(sbi, nid))
2868 			continue;
2869 
2870 		raw_ne = nat_in_journal(journal, i);
2871 
2872 		ne = __lookup_nat_cache(nm_i, nid);
2873 		if (!ne) {
2874 			ne = __alloc_nat_entry(sbi, nid, true);
2875 			__init_nat_entry(nm_i, ne, &raw_ne, true);
2876 		}
2877 
2878 		/*
2879 		 * if a free nat in journal has not been used after last
2880 		 * checkpoint, we should remove it from available nids,
2881 		 * since later we will add it again.
2882 		 */
2883 		if (!get_nat_flag(ne, IS_DIRTY) &&
2884 				le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2885 			spin_lock(&nm_i->nid_list_lock);
2886 			nm_i->available_nids--;
2887 			spin_unlock(&nm_i->nid_list_lock);
2888 		}
2889 
2890 		__set_nat_cache_dirty(nm_i, ne);
2891 	}
2892 	update_nats_in_cursum(journal, -i);
2893 	up_write(&curseg->journal_rwsem);
2894 }
2895 
2896 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2897 						struct list_head *head, int max)
2898 {
2899 	struct nat_entry_set *cur;
2900 
2901 	if (nes->entry_cnt >= max)
2902 		goto add_out;
2903 
2904 	list_for_each_entry(cur, head, set_list) {
2905 		if (cur->entry_cnt >= nes->entry_cnt) {
2906 			list_add(&nes->set_list, cur->set_list.prev);
2907 			return;
2908 		}
2909 	}
2910 add_out:
2911 	list_add_tail(&nes->set_list, head);
2912 }
2913 
2914 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs,
2915 							unsigned int valid)
2916 {
2917 	if (valid == 0) {
2918 		__set_bit_le(nat_ofs, nm_i->empty_nat_bits);
2919 		__clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2920 		return;
2921 	}
2922 
2923 	__clear_bit_le(nat_ofs, nm_i->empty_nat_bits);
2924 	if (valid == NAT_ENTRY_PER_BLOCK)
2925 		__set_bit_le(nat_ofs, nm_i->full_nat_bits);
2926 	else
2927 		__clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2928 }
2929 
2930 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2931 						struct page *page)
2932 {
2933 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2934 	unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2935 	struct f2fs_nat_block *nat_blk = page_address(page);
2936 	int valid = 0;
2937 	int i = 0;
2938 
2939 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
2940 		return;
2941 
2942 	if (nat_index == 0) {
2943 		valid = 1;
2944 		i = 1;
2945 	}
2946 	for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2947 		if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2948 			valid++;
2949 	}
2950 
2951 	__update_nat_bits(nm_i, nat_index, valid);
2952 }
2953 
2954 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi)
2955 {
2956 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2957 	unsigned int nat_ofs;
2958 
2959 	f2fs_down_read(&nm_i->nat_tree_lock);
2960 
2961 	for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) {
2962 		unsigned int valid = 0, nid_ofs = 0;
2963 
2964 		/* handle nid zero due to it should never be used */
2965 		if (unlikely(nat_ofs == 0)) {
2966 			valid = 1;
2967 			nid_ofs = 1;
2968 		}
2969 
2970 		for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) {
2971 			if (!test_bit_le(nid_ofs,
2972 					nm_i->free_nid_bitmap[nat_ofs]))
2973 				valid++;
2974 		}
2975 
2976 		__update_nat_bits(nm_i, nat_ofs, valid);
2977 	}
2978 
2979 	f2fs_up_read(&nm_i->nat_tree_lock);
2980 }
2981 
2982 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2983 		struct nat_entry_set *set, struct cp_control *cpc)
2984 {
2985 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2986 	struct f2fs_journal *journal = curseg->journal;
2987 	nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2988 	bool to_journal = true;
2989 	struct f2fs_nat_block *nat_blk;
2990 	struct nat_entry *ne, *cur;
2991 	struct page *page = NULL;
2992 
2993 	/*
2994 	 * there are two steps to flush nat entries:
2995 	 * #1, flush nat entries to journal in current hot data summary block.
2996 	 * #2, flush nat entries to nat page.
2997 	 */
2998 	if ((cpc->reason & CP_UMOUNT) ||
2999 		!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
3000 		to_journal = false;
3001 
3002 	if (to_journal) {
3003 		down_write(&curseg->journal_rwsem);
3004 	} else {
3005 		page = get_next_nat_page(sbi, start_nid);
3006 		if (IS_ERR(page))
3007 			return PTR_ERR(page);
3008 
3009 		nat_blk = page_address(page);
3010 		f2fs_bug_on(sbi, !nat_blk);
3011 	}
3012 
3013 	/* flush dirty nats in nat entry set */
3014 	list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
3015 		struct f2fs_nat_entry *raw_ne;
3016 		nid_t nid = nat_get_nid(ne);
3017 		int offset;
3018 
3019 		f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
3020 
3021 		if (to_journal) {
3022 			offset = f2fs_lookup_journal_in_cursum(journal,
3023 							NAT_JOURNAL, nid, 1);
3024 			f2fs_bug_on(sbi, offset < 0);
3025 			raw_ne = &nat_in_journal(journal, offset);
3026 			nid_in_journal(journal, offset) = cpu_to_le32(nid);
3027 		} else {
3028 			raw_ne = &nat_blk->entries[nid - start_nid];
3029 		}
3030 		raw_nat_from_node_info(raw_ne, &ne->ni);
3031 		nat_reset_flag(ne);
3032 		__clear_nat_cache_dirty(NM_I(sbi), set, ne);
3033 		if (nat_get_blkaddr(ne) == NULL_ADDR) {
3034 			add_free_nid(sbi, nid, false, true);
3035 		} else {
3036 			spin_lock(&NM_I(sbi)->nid_list_lock);
3037 			update_free_nid_bitmap(sbi, nid, false, false);
3038 			spin_unlock(&NM_I(sbi)->nid_list_lock);
3039 		}
3040 	}
3041 
3042 	if (to_journal) {
3043 		up_write(&curseg->journal_rwsem);
3044 	} else {
3045 		update_nat_bits(sbi, start_nid, page);
3046 		f2fs_put_page(page, 1);
3047 	}
3048 
3049 	/* Allow dirty nats by node block allocation in write_begin */
3050 	if (!set->entry_cnt) {
3051 		radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
3052 		kmem_cache_free(nat_entry_set_slab, set);
3053 	}
3054 	return 0;
3055 }
3056 
3057 /*
3058  * This function is called during the checkpointing process.
3059  */
3060 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3061 {
3062 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3063 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3064 	struct f2fs_journal *journal = curseg->journal;
3065 	struct nat_entry_set *setvec[SETVEC_SIZE];
3066 	struct nat_entry_set *set, *tmp;
3067 	unsigned int found;
3068 	nid_t set_idx = 0;
3069 	LIST_HEAD(sets);
3070 	int err = 0;
3071 
3072 	/*
3073 	 * during unmount, let's flush nat_bits before checking
3074 	 * nat_cnt[DIRTY_NAT].
3075 	 */
3076 	if (cpc->reason & CP_UMOUNT) {
3077 		f2fs_down_write(&nm_i->nat_tree_lock);
3078 		remove_nats_in_journal(sbi);
3079 		f2fs_up_write(&nm_i->nat_tree_lock);
3080 	}
3081 
3082 	if (!nm_i->nat_cnt[DIRTY_NAT])
3083 		return 0;
3084 
3085 	f2fs_down_write(&nm_i->nat_tree_lock);
3086 
3087 	/*
3088 	 * if there are no enough space in journal to store dirty nat
3089 	 * entries, remove all entries from journal and merge them
3090 	 * into nat entry set.
3091 	 */
3092 	if (cpc->reason & CP_UMOUNT ||
3093 		!__has_cursum_space(journal,
3094 			nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3095 		remove_nats_in_journal(sbi);
3096 
3097 	while ((found = __gang_lookup_nat_set(nm_i,
3098 					set_idx, SETVEC_SIZE, setvec))) {
3099 		unsigned idx;
3100 
3101 		set_idx = setvec[found - 1]->set + 1;
3102 		for (idx = 0; idx < found; idx++)
3103 			__adjust_nat_entry_set(setvec[idx], &sets,
3104 						MAX_NAT_JENTRIES(journal));
3105 	}
3106 
3107 	/* flush dirty nats in nat entry set */
3108 	list_for_each_entry_safe(set, tmp, &sets, set_list) {
3109 		err = __flush_nat_entry_set(sbi, set, cpc);
3110 		if (err)
3111 			break;
3112 	}
3113 
3114 	f2fs_up_write(&nm_i->nat_tree_lock);
3115 	/* Allow dirty nats by node block allocation in write_begin */
3116 
3117 	return err;
3118 }
3119 
3120 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3121 {
3122 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3123 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3124 	unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3125 	unsigned int i;
3126 	__u64 cp_ver = cur_cp_version(ckpt);
3127 	block_t nat_bits_addr;
3128 
3129 	nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3130 	nm_i->nat_bits = f2fs_kvzalloc(sbi,
3131 			nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3132 	if (!nm_i->nat_bits)
3133 		return -ENOMEM;
3134 
3135 	nm_i->full_nat_bits = nm_i->nat_bits + 8;
3136 	nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3137 
3138 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3139 		return 0;
3140 
3141 	nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3142 						nm_i->nat_bits_blocks;
3143 	for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3144 		struct page *page;
3145 
3146 		page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3147 		if (IS_ERR(page))
3148 			return PTR_ERR(page);
3149 
3150 		memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3151 					page_address(page), F2FS_BLKSIZE);
3152 		f2fs_put_page(page, 1);
3153 	}
3154 
3155 	cp_ver |= (cur_cp_crc(ckpt) << 32);
3156 	if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3157 		clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
3158 		f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)",
3159 			cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits));
3160 		return 0;
3161 	}
3162 
3163 	f2fs_notice(sbi, "Found nat_bits in checkpoint");
3164 	return 0;
3165 }
3166 
3167 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3168 {
3169 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3170 	unsigned int i = 0;
3171 	nid_t nid, last_nid;
3172 
3173 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3174 		return;
3175 
3176 	for (i = 0; i < nm_i->nat_blocks; i++) {
3177 		i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3178 		if (i >= nm_i->nat_blocks)
3179 			break;
3180 
3181 		__set_bit_le(i, nm_i->nat_block_bitmap);
3182 
3183 		nid = i * NAT_ENTRY_PER_BLOCK;
3184 		last_nid = nid + NAT_ENTRY_PER_BLOCK;
3185 
3186 		spin_lock(&NM_I(sbi)->nid_list_lock);
3187 		for (; nid < last_nid; nid++)
3188 			update_free_nid_bitmap(sbi, nid, true, true);
3189 		spin_unlock(&NM_I(sbi)->nid_list_lock);
3190 	}
3191 
3192 	for (i = 0; i < nm_i->nat_blocks; i++) {
3193 		i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3194 		if (i >= nm_i->nat_blocks)
3195 			break;
3196 
3197 		__set_bit_le(i, nm_i->nat_block_bitmap);
3198 	}
3199 }
3200 
3201 static int init_node_manager(struct f2fs_sb_info *sbi)
3202 {
3203 	struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3204 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3205 	unsigned char *version_bitmap;
3206 	unsigned int nat_segs;
3207 	int err;
3208 
3209 	nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3210 
3211 	/* segment_count_nat includes pair segment so divide to 2. */
3212 	nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3213 	nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3214 	nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3215 
3216 	/* not used nids: 0, node, meta, (and root counted as valid node) */
3217 	nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3218 						F2FS_RESERVED_NODE_NUM;
3219 	nm_i->nid_cnt[FREE_NID] = 0;
3220 	nm_i->nid_cnt[PREALLOC_NID] = 0;
3221 	nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3222 	nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3223 	nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3224 	nm_i->max_rf_node_blocks = DEF_RF_NODE_BLOCKS;
3225 
3226 	INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3227 	INIT_LIST_HEAD(&nm_i->free_nid_list);
3228 	INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3229 	INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3230 	INIT_LIST_HEAD(&nm_i->nat_entries);
3231 	spin_lock_init(&nm_i->nat_list_lock);
3232 
3233 	mutex_init(&nm_i->build_lock);
3234 	spin_lock_init(&nm_i->nid_list_lock);
3235 	init_f2fs_rwsem(&nm_i->nat_tree_lock);
3236 
3237 	nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3238 	nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3239 	version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3240 	nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3241 					GFP_KERNEL);
3242 	if (!nm_i->nat_bitmap)
3243 		return -ENOMEM;
3244 
3245 	err = __get_nat_bitmaps(sbi);
3246 	if (err)
3247 		return err;
3248 
3249 #ifdef CONFIG_F2FS_CHECK_FS
3250 	nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3251 					GFP_KERNEL);
3252 	if (!nm_i->nat_bitmap_mir)
3253 		return -ENOMEM;
3254 #endif
3255 
3256 	return 0;
3257 }
3258 
3259 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3260 {
3261 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3262 	int i;
3263 
3264 	nm_i->free_nid_bitmap =
3265 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3266 					      nm_i->nat_blocks),
3267 			      GFP_KERNEL);
3268 	if (!nm_i->free_nid_bitmap)
3269 		return -ENOMEM;
3270 
3271 	for (i = 0; i < nm_i->nat_blocks; i++) {
3272 		nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3273 			f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3274 		if (!nm_i->free_nid_bitmap[i])
3275 			return -ENOMEM;
3276 	}
3277 
3278 	nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3279 								GFP_KERNEL);
3280 	if (!nm_i->nat_block_bitmap)
3281 		return -ENOMEM;
3282 
3283 	nm_i->free_nid_count =
3284 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3285 					      nm_i->nat_blocks),
3286 			      GFP_KERNEL);
3287 	if (!nm_i->free_nid_count)
3288 		return -ENOMEM;
3289 	return 0;
3290 }
3291 
3292 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3293 {
3294 	int err;
3295 
3296 	sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3297 							GFP_KERNEL);
3298 	if (!sbi->nm_info)
3299 		return -ENOMEM;
3300 
3301 	err = init_node_manager(sbi);
3302 	if (err)
3303 		return err;
3304 
3305 	err = init_free_nid_cache(sbi);
3306 	if (err)
3307 		return err;
3308 
3309 	/* load free nid status from nat_bits table */
3310 	load_free_nid_bitmap(sbi);
3311 
3312 	return f2fs_build_free_nids(sbi, true, true);
3313 }
3314 
3315 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3316 {
3317 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3318 	struct free_nid *i, *next_i;
3319 	struct nat_entry *natvec[NATVEC_SIZE];
3320 	struct nat_entry_set *setvec[SETVEC_SIZE];
3321 	nid_t nid = 0;
3322 	unsigned int found;
3323 
3324 	if (!nm_i)
3325 		return;
3326 
3327 	/* destroy free nid list */
3328 	spin_lock(&nm_i->nid_list_lock);
3329 	list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3330 		__remove_free_nid(sbi, i, FREE_NID);
3331 		spin_unlock(&nm_i->nid_list_lock);
3332 		kmem_cache_free(free_nid_slab, i);
3333 		spin_lock(&nm_i->nid_list_lock);
3334 	}
3335 	f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3336 	f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3337 	f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3338 	spin_unlock(&nm_i->nid_list_lock);
3339 
3340 	/* destroy nat cache */
3341 	f2fs_down_write(&nm_i->nat_tree_lock);
3342 	while ((found = __gang_lookup_nat_cache(nm_i,
3343 					nid, NATVEC_SIZE, natvec))) {
3344 		unsigned idx;
3345 
3346 		nid = nat_get_nid(natvec[found - 1]) + 1;
3347 		for (idx = 0; idx < found; idx++) {
3348 			spin_lock(&nm_i->nat_list_lock);
3349 			list_del(&natvec[idx]->list);
3350 			spin_unlock(&nm_i->nat_list_lock);
3351 
3352 			__del_from_nat_cache(nm_i, natvec[idx]);
3353 		}
3354 	}
3355 	f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3356 
3357 	/* destroy nat set cache */
3358 	nid = 0;
3359 	while ((found = __gang_lookup_nat_set(nm_i,
3360 					nid, SETVEC_SIZE, setvec))) {
3361 		unsigned idx;
3362 
3363 		nid = setvec[found - 1]->set + 1;
3364 		for (idx = 0; idx < found; idx++) {
3365 			/* entry_cnt is not zero, when cp_error was occurred */
3366 			f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3367 			radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3368 			kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3369 		}
3370 	}
3371 	f2fs_up_write(&nm_i->nat_tree_lock);
3372 
3373 	kvfree(nm_i->nat_block_bitmap);
3374 	if (nm_i->free_nid_bitmap) {
3375 		int i;
3376 
3377 		for (i = 0; i < nm_i->nat_blocks; i++)
3378 			kvfree(nm_i->free_nid_bitmap[i]);
3379 		kvfree(nm_i->free_nid_bitmap);
3380 	}
3381 	kvfree(nm_i->free_nid_count);
3382 
3383 	kvfree(nm_i->nat_bitmap);
3384 	kvfree(nm_i->nat_bits);
3385 #ifdef CONFIG_F2FS_CHECK_FS
3386 	kvfree(nm_i->nat_bitmap_mir);
3387 #endif
3388 	sbi->nm_info = NULL;
3389 	kfree(nm_i);
3390 }
3391 
3392 int __init f2fs_create_node_manager_caches(void)
3393 {
3394 	nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3395 			sizeof(struct nat_entry));
3396 	if (!nat_entry_slab)
3397 		goto fail;
3398 
3399 	free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3400 			sizeof(struct free_nid));
3401 	if (!free_nid_slab)
3402 		goto destroy_nat_entry;
3403 
3404 	nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3405 			sizeof(struct nat_entry_set));
3406 	if (!nat_entry_set_slab)
3407 		goto destroy_free_nid;
3408 
3409 	fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3410 			sizeof(struct fsync_node_entry));
3411 	if (!fsync_node_entry_slab)
3412 		goto destroy_nat_entry_set;
3413 	return 0;
3414 
3415 destroy_nat_entry_set:
3416 	kmem_cache_destroy(nat_entry_set_slab);
3417 destroy_free_nid:
3418 	kmem_cache_destroy(free_nid_slab);
3419 destroy_nat_entry:
3420 	kmem_cache_destroy(nat_entry_slab);
3421 fail:
3422 	return -ENOMEM;
3423 }
3424 
3425 void f2fs_destroy_node_manager_caches(void)
3426 {
3427 	kmem_cache_destroy(fsync_node_entry_slab);
3428 	kmem_cache_destroy(nat_entry_set_slab);
3429 	kmem_cache_destroy(free_nid_slab);
3430 	kmem_cache_destroy(nat_entry_slab);
3431 }
3432