xref: /linux/fs/f2fs/node.c (revision 55d0969c451159cff86949b38c39171cab962069)
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(nm_i) mutex_is_locked(&(nm_i)->build_lock)
24 
25 static struct kmem_cache *nat_entry_slab;
26 static struct kmem_cache *free_nid_slab;
27 static struct kmem_cache *nat_entry_set_slab;
28 static struct kmem_cache *fsync_node_entry_slab;
29 
30 /*
31  * Check whether the given nid is within node id range.
32  */
33 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
34 {
35 	if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
36 		set_sbi_flag(sbi, SBI_NEED_FSCK);
37 		f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
38 			  __func__, nid);
39 		f2fs_handle_error(sbi, ERROR_CORRUPTED_INODE);
40 		return -EFSCORRUPTED;
41 	}
42 	return 0;
43 }
44 
45 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
46 {
47 	struct f2fs_nm_info *nm_i = NM_I(sbi);
48 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
49 	struct sysinfo val;
50 	unsigned long avail_ram;
51 	unsigned long mem_size = 0;
52 	bool res = false;
53 
54 	if (!nm_i)
55 		return true;
56 
57 	si_meminfo(&val);
58 
59 	/* only uses low memory */
60 	avail_ram = val.totalram - val.totalhigh;
61 
62 	/*
63 	 * give 25%, 25%, 50%, 50%, 25%, 25% memory for each components respectively
64 	 */
65 	if (type == FREE_NIDS) {
66 		mem_size = (nm_i->nid_cnt[FREE_NID] *
67 				sizeof(struct free_nid)) >> PAGE_SHIFT;
68 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
69 	} else if (type == NAT_ENTRIES) {
70 		mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
71 				sizeof(struct nat_entry)) >> PAGE_SHIFT;
72 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
73 		if (excess_cached_nats(sbi))
74 			res = false;
75 	} else if (type == DIRTY_DENTS) {
76 		if (sbi->sb->s_bdi->wb.dirty_exceeded)
77 			return false;
78 		mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
79 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
80 	} else if (type == INO_ENTRIES) {
81 		int i;
82 
83 		for (i = 0; i < MAX_INO_ENTRY; i++)
84 			mem_size += sbi->im[i].ino_num *
85 						sizeof(struct ino_entry);
86 		mem_size >>= PAGE_SHIFT;
87 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
88 	} else if (type == READ_EXTENT_CACHE || type == AGE_EXTENT_CACHE) {
89 		enum extent_type etype = type == READ_EXTENT_CACHE ?
90 						EX_READ : EX_BLOCK_AGE;
91 		struct extent_tree_info *eti = &sbi->extent_tree[etype];
92 
93 		mem_size = (atomic_read(&eti->total_ext_tree) *
94 				sizeof(struct extent_tree) +
95 				atomic_read(&eti->total_ext_node) *
96 				sizeof(struct extent_node)) >> PAGE_SHIFT;
97 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
98 	} else if (type == DISCARD_CACHE) {
99 		mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
100 				sizeof(struct discard_cmd)) >> PAGE_SHIFT;
101 		res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
102 	} else if (type == COMPRESS_PAGE) {
103 #ifdef CONFIG_F2FS_FS_COMPRESSION
104 		unsigned long free_ram = val.freeram;
105 
106 		/*
107 		 * free memory is lower than watermark or cached page count
108 		 * exceed threshold, deny caching compress page.
109 		 */
110 		res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
111 			(COMPRESS_MAPPING(sbi)->nrpages <
112 			 free_ram * sbi->compress_percent / 100);
113 #else
114 		res = false;
115 #endif
116 	} else {
117 		if (!sbi->sb->s_bdi->wb.dirty_exceeded)
118 			return true;
119 	}
120 	return res;
121 }
122 
123 static void clear_node_page_dirty(struct page *page)
124 {
125 	if (PageDirty(page)) {
126 		f2fs_clear_page_cache_dirty_tag(page_folio(page));
127 		clear_page_dirty_for_io(page);
128 		dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
129 	}
130 	ClearPageUptodate(page);
131 }
132 
133 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
134 {
135 	return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
136 }
137 
138 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
139 {
140 	struct page *src_page;
141 	struct page *dst_page;
142 	pgoff_t dst_off;
143 	void *src_addr;
144 	void *dst_addr;
145 	struct f2fs_nm_info *nm_i = NM_I(sbi);
146 
147 	dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
148 
149 	/* get current nat block page with lock */
150 	src_page = get_current_nat_page(sbi, nid);
151 	if (IS_ERR(src_page))
152 		return src_page;
153 	dst_page = f2fs_grab_meta_page(sbi, dst_off);
154 	f2fs_bug_on(sbi, PageDirty(src_page));
155 
156 	src_addr = page_address(src_page);
157 	dst_addr = page_address(dst_page);
158 	memcpy(dst_addr, src_addr, PAGE_SIZE);
159 	set_page_dirty(dst_page);
160 	f2fs_put_page(src_page, 1);
161 
162 	set_to_next_nat(nm_i, nid);
163 
164 	return dst_page;
165 }
166 
167 static struct nat_entry *__alloc_nat_entry(struct f2fs_sb_info *sbi,
168 						nid_t nid, bool no_fail)
169 {
170 	struct nat_entry *new;
171 
172 	new = f2fs_kmem_cache_alloc(nat_entry_slab,
173 					GFP_F2FS_ZERO, no_fail, sbi);
174 	if (new) {
175 		nat_set_nid(new, nid);
176 		nat_reset_flag(new);
177 	}
178 	return new;
179 }
180 
181 static void __free_nat_entry(struct nat_entry *e)
182 {
183 	kmem_cache_free(nat_entry_slab, e);
184 }
185 
186 /* must be locked by nat_tree_lock */
187 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
188 	struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
189 {
190 	if (no_fail)
191 		f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
192 	else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
193 		return NULL;
194 
195 	if (raw_ne)
196 		node_info_from_raw_nat(&ne->ni, raw_ne);
197 
198 	spin_lock(&nm_i->nat_list_lock);
199 	list_add_tail(&ne->list, &nm_i->nat_entries);
200 	spin_unlock(&nm_i->nat_list_lock);
201 
202 	nm_i->nat_cnt[TOTAL_NAT]++;
203 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
204 	return ne;
205 }
206 
207 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
208 {
209 	struct nat_entry *ne;
210 
211 	ne = radix_tree_lookup(&nm_i->nat_root, n);
212 
213 	/* for recent accessed nat entry, move it to tail of lru list */
214 	if (ne && !get_nat_flag(ne, IS_DIRTY)) {
215 		spin_lock(&nm_i->nat_list_lock);
216 		if (!list_empty(&ne->list))
217 			list_move_tail(&ne->list, &nm_i->nat_entries);
218 		spin_unlock(&nm_i->nat_list_lock);
219 	}
220 
221 	return ne;
222 }
223 
224 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
225 		nid_t start, unsigned int nr, struct nat_entry **ep)
226 {
227 	return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
228 }
229 
230 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
231 {
232 	radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
233 	nm_i->nat_cnt[TOTAL_NAT]--;
234 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
235 	__free_nat_entry(e);
236 }
237 
238 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
239 							struct nat_entry *ne)
240 {
241 	nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
242 	struct nat_entry_set *head;
243 
244 	head = radix_tree_lookup(&nm_i->nat_set_root, set);
245 	if (!head) {
246 		head = f2fs_kmem_cache_alloc(nat_entry_set_slab,
247 						GFP_NOFS, true, NULL);
248 
249 		INIT_LIST_HEAD(&head->entry_list);
250 		INIT_LIST_HEAD(&head->set_list);
251 		head->set = set;
252 		head->entry_cnt = 0;
253 		f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
254 	}
255 	return head;
256 }
257 
258 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
259 						struct nat_entry *ne)
260 {
261 	struct nat_entry_set *head;
262 	bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
263 
264 	if (!new_ne)
265 		head = __grab_nat_entry_set(nm_i, ne);
266 
267 	/*
268 	 * update entry_cnt in below condition:
269 	 * 1. update NEW_ADDR to valid block address;
270 	 * 2. update old block address to new one;
271 	 */
272 	if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
273 				!get_nat_flag(ne, IS_DIRTY)))
274 		head->entry_cnt++;
275 
276 	set_nat_flag(ne, IS_PREALLOC, new_ne);
277 
278 	if (get_nat_flag(ne, IS_DIRTY))
279 		goto refresh_list;
280 
281 	nm_i->nat_cnt[DIRTY_NAT]++;
282 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
283 	set_nat_flag(ne, IS_DIRTY, true);
284 refresh_list:
285 	spin_lock(&nm_i->nat_list_lock);
286 	if (new_ne)
287 		list_del_init(&ne->list);
288 	else
289 		list_move_tail(&ne->list, &head->entry_list);
290 	spin_unlock(&nm_i->nat_list_lock);
291 }
292 
293 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
294 		struct nat_entry_set *set, struct nat_entry *ne)
295 {
296 	spin_lock(&nm_i->nat_list_lock);
297 	list_move_tail(&ne->list, &nm_i->nat_entries);
298 	spin_unlock(&nm_i->nat_list_lock);
299 
300 	set_nat_flag(ne, IS_DIRTY, false);
301 	set->entry_cnt--;
302 	nm_i->nat_cnt[DIRTY_NAT]--;
303 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
304 }
305 
306 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
307 		nid_t start, unsigned int nr, struct nat_entry_set **ep)
308 {
309 	return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
310 							start, nr);
311 }
312 
313 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
314 {
315 	return NODE_MAPPING(sbi) == page->mapping &&
316 			IS_DNODE(page) && is_cold_node(page);
317 }
318 
319 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
320 {
321 	spin_lock_init(&sbi->fsync_node_lock);
322 	INIT_LIST_HEAD(&sbi->fsync_node_list);
323 	sbi->fsync_seg_id = 0;
324 	sbi->fsync_node_num = 0;
325 }
326 
327 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
328 							struct page *page)
329 {
330 	struct fsync_node_entry *fn;
331 	unsigned long flags;
332 	unsigned int seq_id;
333 
334 	fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab,
335 					GFP_NOFS, true, NULL);
336 
337 	get_page(page);
338 	fn->page = page;
339 	INIT_LIST_HEAD(&fn->list);
340 
341 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
342 	list_add_tail(&fn->list, &sbi->fsync_node_list);
343 	fn->seq_id = sbi->fsync_seg_id++;
344 	seq_id = fn->seq_id;
345 	sbi->fsync_node_num++;
346 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
347 
348 	return seq_id;
349 }
350 
351 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
352 {
353 	struct fsync_node_entry *fn;
354 	unsigned long flags;
355 
356 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
357 	list_for_each_entry(fn, &sbi->fsync_node_list, list) {
358 		if (fn->page == page) {
359 			list_del(&fn->list);
360 			sbi->fsync_node_num--;
361 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
362 			kmem_cache_free(fsync_node_entry_slab, fn);
363 			put_page(page);
364 			return;
365 		}
366 	}
367 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
368 	f2fs_bug_on(sbi, 1);
369 }
370 
371 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
372 {
373 	unsigned long flags;
374 
375 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
376 	sbi->fsync_seg_id = 0;
377 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
378 }
379 
380 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
381 {
382 	struct f2fs_nm_info *nm_i = NM_I(sbi);
383 	struct nat_entry *e;
384 	bool need = false;
385 
386 	f2fs_down_read(&nm_i->nat_tree_lock);
387 	e = __lookup_nat_cache(nm_i, nid);
388 	if (e) {
389 		if (!get_nat_flag(e, IS_CHECKPOINTED) &&
390 				!get_nat_flag(e, HAS_FSYNCED_INODE))
391 			need = true;
392 	}
393 	f2fs_up_read(&nm_i->nat_tree_lock);
394 	return need;
395 }
396 
397 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
398 {
399 	struct f2fs_nm_info *nm_i = NM_I(sbi);
400 	struct nat_entry *e;
401 	bool is_cp = true;
402 
403 	f2fs_down_read(&nm_i->nat_tree_lock);
404 	e = __lookup_nat_cache(nm_i, nid);
405 	if (e && !get_nat_flag(e, IS_CHECKPOINTED))
406 		is_cp = false;
407 	f2fs_up_read(&nm_i->nat_tree_lock);
408 	return is_cp;
409 }
410 
411 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
412 {
413 	struct f2fs_nm_info *nm_i = NM_I(sbi);
414 	struct nat_entry *e;
415 	bool need_update = true;
416 
417 	f2fs_down_read(&nm_i->nat_tree_lock);
418 	e = __lookup_nat_cache(nm_i, ino);
419 	if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
420 			(get_nat_flag(e, IS_CHECKPOINTED) ||
421 			 get_nat_flag(e, HAS_FSYNCED_INODE)))
422 		need_update = false;
423 	f2fs_up_read(&nm_i->nat_tree_lock);
424 	return need_update;
425 }
426 
427 /* must be locked by nat_tree_lock */
428 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
429 						struct f2fs_nat_entry *ne)
430 {
431 	struct f2fs_nm_info *nm_i = NM_I(sbi);
432 	struct nat_entry *new, *e;
433 
434 	/* Let's mitigate lock contention of nat_tree_lock during checkpoint */
435 	if (f2fs_rwsem_is_locked(&sbi->cp_global_sem))
436 		return;
437 
438 	new = __alloc_nat_entry(sbi, nid, false);
439 	if (!new)
440 		return;
441 
442 	f2fs_down_write(&nm_i->nat_tree_lock);
443 	e = __lookup_nat_cache(nm_i, nid);
444 	if (!e)
445 		e = __init_nat_entry(nm_i, new, ne, false);
446 	else
447 		f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
448 				nat_get_blkaddr(e) !=
449 					le32_to_cpu(ne->block_addr) ||
450 				nat_get_version(e) != ne->version);
451 	f2fs_up_write(&nm_i->nat_tree_lock);
452 	if (e != new)
453 		__free_nat_entry(new);
454 }
455 
456 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
457 			block_t new_blkaddr, bool fsync_done)
458 {
459 	struct f2fs_nm_info *nm_i = NM_I(sbi);
460 	struct nat_entry *e;
461 	struct nat_entry *new = __alloc_nat_entry(sbi, ni->nid, true);
462 
463 	f2fs_down_write(&nm_i->nat_tree_lock);
464 	e = __lookup_nat_cache(nm_i, ni->nid);
465 	if (!e) {
466 		e = __init_nat_entry(nm_i, new, NULL, true);
467 		copy_node_info(&e->ni, ni);
468 		f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
469 	} else if (new_blkaddr == NEW_ADDR) {
470 		/*
471 		 * when nid is reallocated,
472 		 * previous nat entry can be remained in nat cache.
473 		 * So, reinitialize it with new information.
474 		 */
475 		copy_node_info(&e->ni, ni);
476 		f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
477 	}
478 	/* let's free early to reduce memory consumption */
479 	if (e != new)
480 		__free_nat_entry(new);
481 
482 	/* sanity check */
483 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
484 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
485 			new_blkaddr == NULL_ADDR);
486 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
487 			new_blkaddr == NEW_ADDR);
488 	f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
489 			new_blkaddr == NEW_ADDR);
490 
491 	/* increment version no as node is removed */
492 	if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
493 		unsigned char version = nat_get_version(e);
494 
495 		nat_set_version(e, inc_node_version(version));
496 	}
497 
498 	/* change address */
499 	nat_set_blkaddr(e, new_blkaddr);
500 	if (!__is_valid_data_blkaddr(new_blkaddr))
501 		set_nat_flag(e, IS_CHECKPOINTED, false);
502 	__set_nat_cache_dirty(nm_i, e);
503 
504 	/* update fsync_mark if its inode nat entry is still alive */
505 	if (ni->nid != ni->ino)
506 		e = __lookup_nat_cache(nm_i, ni->ino);
507 	if (e) {
508 		if (fsync_done && ni->nid == ni->ino)
509 			set_nat_flag(e, HAS_FSYNCED_INODE, true);
510 		set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
511 	}
512 	f2fs_up_write(&nm_i->nat_tree_lock);
513 }
514 
515 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
516 {
517 	struct f2fs_nm_info *nm_i = NM_I(sbi);
518 	int nr = nr_shrink;
519 
520 	if (!f2fs_down_write_trylock(&nm_i->nat_tree_lock))
521 		return 0;
522 
523 	spin_lock(&nm_i->nat_list_lock);
524 	while (nr_shrink) {
525 		struct nat_entry *ne;
526 
527 		if (list_empty(&nm_i->nat_entries))
528 			break;
529 
530 		ne = list_first_entry(&nm_i->nat_entries,
531 					struct nat_entry, list);
532 		list_del(&ne->list);
533 		spin_unlock(&nm_i->nat_list_lock);
534 
535 		__del_from_nat_cache(nm_i, ne);
536 		nr_shrink--;
537 
538 		spin_lock(&nm_i->nat_list_lock);
539 	}
540 	spin_unlock(&nm_i->nat_list_lock);
541 
542 	f2fs_up_write(&nm_i->nat_tree_lock);
543 	return nr - nr_shrink;
544 }
545 
546 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
547 				struct node_info *ni, bool checkpoint_context)
548 {
549 	struct f2fs_nm_info *nm_i = NM_I(sbi);
550 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
551 	struct f2fs_journal *journal = curseg->journal;
552 	nid_t start_nid = START_NID(nid);
553 	struct f2fs_nat_block *nat_blk;
554 	struct page *page = NULL;
555 	struct f2fs_nat_entry ne;
556 	struct nat_entry *e;
557 	pgoff_t index;
558 	block_t blkaddr;
559 	int i;
560 
561 	ni->nid = nid;
562 retry:
563 	/* Check nat cache */
564 	f2fs_down_read(&nm_i->nat_tree_lock);
565 	e = __lookup_nat_cache(nm_i, nid);
566 	if (e) {
567 		ni->ino = nat_get_ino(e);
568 		ni->blk_addr = nat_get_blkaddr(e);
569 		ni->version = nat_get_version(e);
570 		f2fs_up_read(&nm_i->nat_tree_lock);
571 		return 0;
572 	}
573 
574 	/*
575 	 * Check current segment summary by trying to grab journal_rwsem first.
576 	 * This sem is on the critical path on the checkpoint requiring the above
577 	 * nat_tree_lock. Therefore, we should retry, if we failed to grab here
578 	 * while not bothering checkpoint.
579 	 */
580 	if (!f2fs_rwsem_is_locked(&sbi->cp_global_sem) || checkpoint_context) {
581 		down_read(&curseg->journal_rwsem);
582 	} else if (f2fs_rwsem_is_contended(&nm_i->nat_tree_lock) ||
583 				!down_read_trylock(&curseg->journal_rwsem)) {
584 		f2fs_up_read(&nm_i->nat_tree_lock);
585 		goto retry;
586 	}
587 
588 	i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
589 	if (i >= 0) {
590 		ne = nat_in_journal(journal, i);
591 		node_info_from_raw_nat(ni, &ne);
592 	}
593 	up_read(&curseg->journal_rwsem);
594 	if (i >= 0) {
595 		f2fs_up_read(&nm_i->nat_tree_lock);
596 		goto cache;
597 	}
598 
599 	/* Fill node_info from nat page */
600 	index = current_nat_addr(sbi, nid);
601 	f2fs_up_read(&nm_i->nat_tree_lock);
602 
603 	page = f2fs_get_meta_page(sbi, index);
604 	if (IS_ERR(page))
605 		return PTR_ERR(page);
606 
607 	nat_blk = (struct f2fs_nat_block *)page_address(page);
608 	ne = nat_blk->entries[nid - start_nid];
609 	node_info_from_raw_nat(ni, &ne);
610 	f2fs_put_page(page, 1);
611 cache:
612 	blkaddr = le32_to_cpu(ne.block_addr);
613 	if (__is_valid_data_blkaddr(blkaddr) &&
614 		!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
615 		return -EFAULT;
616 
617 	/* cache nat entry */
618 	cache_nat_entry(sbi, nid, &ne);
619 	return 0;
620 }
621 
622 /*
623  * readahead MAX_RA_NODE number of node pages.
624  */
625 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
626 {
627 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
628 	struct blk_plug plug;
629 	int i, end;
630 	nid_t nid;
631 
632 	blk_start_plug(&plug);
633 
634 	/* Then, try readahead for siblings of the desired node */
635 	end = start + n;
636 	end = min(end, (int)NIDS_PER_BLOCK);
637 	for (i = start; i < end; i++) {
638 		nid = get_nid(parent, i, false);
639 		f2fs_ra_node_page(sbi, nid);
640 	}
641 
642 	blk_finish_plug(&plug);
643 }
644 
645 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
646 {
647 	const long direct_index = ADDRS_PER_INODE(dn->inode);
648 	const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
649 	const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
650 	unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
651 	int cur_level = dn->cur_level;
652 	int max_level = dn->max_level;
653 	pgoff_t base = 0;
654 
655 	if (!dn->max_level)
656 		return pgofs + 1;
657 
658 	while (max_level-- > cur_level)
659 		skipped_unit *= NIDS_PER_BLOCK;
660 
661 	switch (dn->max_level) {
662 	case 3:
663 		base += 2 * indirect_blks;
664 		fallthrough;
665 	case 2:
666 		base += 2 * direct_blks;
667 		fallthrough;
668 	case 1:
669 		base += direct_index;
670 		break;
671 	default:
672 		f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
673 	}
674 
675 	return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
676 }
677 
678 /*
679  * The maximum depth is four.
680  * Offset[0] will have raw inode offset.
681  */
682 static int get_node_path(struct inode *inode, long block,
683 				int offset[4], unsigned int noffset[4])
684 {
685 	const long direct_index = ADDRS_PER_INODE(inode);
686 	const long direct_blks = ADDRS_PER_BLOCK(inode);
687 	const long dptrs_per_blk = NIDS_PER_BLOCK;
688 	const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
689 	const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
690 	int n = 0;
691 	int level = 0;
692 
693 	noffset[0] = 0;
694 
695 	if (block < direct_index) {
696 		offset[n] = block;
697 		goto got;
698 	}
699 	block -= direct_index;
700 	if (block < direct_blks) {
701 		offset[n++] = NODE_DIR1_BLOCK;
702 		noffset[n] = 1;
703 		offset[n] = block;
704 		level = 1;
705 		goto got;
706 	}
707 	block -= direct_blks;
708 	if (block < direct_blks) {
709 		offset[n++] = NODE_DIR2_BLOCK;
710 		noffset[n] = 2;
711 		offset[n] = block;
712 		level = 1;
713 		goto got;
714 	}
715 	block -= direct_blks;
716 	if (block < indirect_blks) {
717 		offset[n++] = NODE_IND1_BLOCK;
718 		noffset[n] = 3;
719 		offset[n++] = block / direct_blks;
720 		noffset[n] = 4 + offset[n - 1];
721 		offset[n] = block % direct_blks;
722 		level = 2;
723 		goto got;
724 	}
725 	block -= indirect_blks;
726 	if (block < indirect_blks) {
727 		offset[n++] = NODE_IND2_BLOCK;
728 		noffset[n] = 4 + dptrs_per_blk;
729 		offset[n++] = block / direct_blks;
730 		noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
731 		offset[n] = block % direct_blks;
732 		level = 2;
733 		goto got;
734 	}
735 	block -= indirect_blks;
736 	if (block < dindirect_blks) {
737 		offset[n++] = NODE_DIND_BLOCK;
738 		noffset[n] = 5 + (dptrs_per_blk * 2);
739 		offset[n++] = block / indirect_blks;
740 		noffset[n] = 6 + (dptrs_per_blk * 2) +
741 			      offset[n - 1] * (dptrs_per_blk + 1);
742 		offset[n++] = (block / direct_blks) % dptrs_per_blk;
743 		noffset[n] = 7 + (dptrs_per_blk * 2) +
744 			      offset[n - 2] * (dptrs_per_blk + 1) +
745 			      offset[n - 1];
746 		offset[n] = block % direct_blks;
747 		level = 3;
748 		goto got;
749 	} else {
750 		return -E2BIG;
751 	}
752 got:
753 	return level;
754 }
755 
756 /*
757  * Caller should call f2fs_put_dnode(dn).
758  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
759  * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
760  */
761 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
762 {
763 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
764 	struct page *npage[4];
765 	struct page *parent = NULL;
766 	int offset[4];
767 	unsigned int noffset[4];
768 	nid_t nids[4];
769 	int level, i = 0;
770 	int err = 0;
771 
772 	level = get_node_path(dn->inode, index, offset, noffset);
773 	if (level < 0)
774 		return level;
775 
776 	nids[0] = dn->inode->i_ino;
777 	npage[0] = dn->inode_page;
778 
779 	if (!npage[0]) {
780 		npage[0] = f2fs_get_node_page(sbi, nids[0]);
781 		if (IS_ERR(npage[0]))
782 			return PTR_ERR(npage[0]);
783 	}
784 
785 	/* if inline_data is set, should not report any block indices */
786 	if (f2fs_has_inline_data(dn->inode) && index) {
787 		err = -ENOENT;
788 		f2fs_put_page(npage[0], 1);
789 		goto release_out;
790 	}
791 
792 	parent = npage[0];
793 	if (level != 0)
794 		nids[1] = get_nid(parent, offset[0], true);
795 	dn->inode_page = npage[0];
796 	dn->inode_page_locked = true;
797 
798 	/* get indirect or direct nodes */
799 	for (i = 1; i <= level; i++) {
800 		bool done = false;
801 
802 		if (!nids[i] && mode == ALLOC_NODE) {
803 			/* alloc new node */
804 			if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
805 				err = -ENOSPC;
806 				goto release_pages;
807 			}
808 
809 			dn->nid = nids[i];
810 			npage[i] = f2fs_new_node_page(dn, noffset[i]);
811 			if (IS_ERR(npage[i])) {
812 				f2fs_alloc_nid_failed(sbi, nids[i]);
813 				err = PTR_ERR(npage[i]);
814 				goto release_pages;
815 			}
816 
817 			set_nid(parent, offset[i - 1], nids[i], i == 1);
818 			f2fs_alloc_nid_done(sbi, nids[i]);
819 			done = true;
820 		} else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
821 			npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
822 			if (IS_ERR(npage[i])) {
823 				err = PTR_ERR(npage[i]);
824 				goto release_pages;
825 			}
826 			done = true;
827 		}
828 		if (i == 1) {
829 			dn->inode_page_locked = false;
830 			unlock_page(parent);
831 		} else {
832 			f2fs_put_page(parent, 1);
833 		}
834 
835 		if (!done) {
836 			npage[i] = f2fs_get_node_page(sbi, nids[i]);
837 			if (IS_ERR(npage[i])) {
838 				err = PTR_ERR(npage[i]);
839 				f2fs_put_page(npage[0], 0);
840 				goto release_out;
841 			}
842 		}
843 		if (i < level) {
844 			parent = npage[i];
845 			nids[i + 1] = get_nid(parent, offset[i], false);
846 		}
847 	}
848 	dn->nid = nids[level];
849 	dn->ofs_in_node = offset[level];
850 	dn->node_page = npage[level];
851 	dn->data_blkaddr = f2fs_data_blkaddr(dn);
852 
853 	if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) &&
854 					f2fs_sb_has_readonly(sbi)) {
855 		unsigned int cluster_size = F2FS_I(dn->inode)->i_cluster_size;
856 		unsigned int ofs_in_node = dn->ofs_in_node;
857 		pgoff_t fofs = index;
858 		unsigned int c_len;
859 		block_t blkaddr;
860 
861 		/* should align fofs and ofs_in_node to cluster_size */
862 		if (fofs % cluster_size) {
863 			fofs = round_down(fofs, cluster_size);
864 			ofs_in_node = round_down(ofs_in_node, cluster_size);
865 		}
866 
867 		c_len = f2fs_cluster_blocks_are_contiguous(dn, ofs_in_node);
868 		if (!c_len)
869 			goto out;
870 
871 		blkaddr = data_blkaddr(dn->inode, dn->node_page, ofs_in_node);
872 		if (blkaddr == COMPRESS_ADDR)
873 			blkaddr = data_blkaddr(dn->inode, dn->node_page,
874 						ofs_in_node + 1);
875 
876 		f2fs_update_read_extent_tree_range_compressed(dn->inode,
877 					fofs, blkaddr, cluster_size, c_len);
878 	}
879 out:
880 	return 0;
881 
882 release_pages:
883 	f2fs_put_page(parent, 1);
884 	if (i > 1)
885 		f2fs_put_page(npage[0], 0);
886 release_out:
887 	dn->inode_page = NULL;
888 	dn->node_page = NULL;
889 	if (err == -ENOENT) {
890 		dn->cur_level = i;
891 		dn->max_level = level;
892 		dn->ofs_in_node = offset[level];
893 	}
894 	return err;
895 }
896 
897 static int truncate_node(struct dnode_of_data *dn)
898 {
899 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
900 	struct node_info ni;
901 	int err;
902 	pgoff_t index;
903 
904 	err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
905 	if (err)
906 		return err;
907 
908 	/* Deallocate node address */
909 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
910 	dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
911 	set_node_addr(sbi, &ni, NULL_ADDR, false);
912 
913 	if (dn->nid == dn->inode->i_ino) {
914 		f2fs_remove_orphan_inode(sbi, dn->nid);
915 		dec_valid_inode_count(sbi);
916 		f2fs_inode_synced(dn->inode);
917 	}
918 
919 	clear_node_page_dirty(dn->node_page);
920 	set_sbi_flag(sbi, SBI_IS_DIRTY);
921 
922 	index = page_folio(dn->node_page)->index;
923 	f2fs_put_page(dn->node_page, 1);
924 
925 	invalidate_mapping_pages(NODE_MAPPING(sbi),
926 			index, index);
927 
928 	dn->node_page = NULL;
929 	trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
930 
931 	return 0;
932 }
933 
934 static int truncate_dnode(struct dnode_of_data *dn)
935 {
936 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
937 	struct page *page;
938 	int err;
939 
940 	if (dn->nid == 0)
941 		return 1;
942 
943 	/* get direct node */
944 	page = f2fs_get_node_page(sbi, dn->nid);
945 	if (PTR_ERR(page) == -ENOENT)
946 		return 1;
947 	else if (IS_ERR(page))
948 		return PTR_ERR(page);
949 
950 	if (IS_INODE(page) || ino_of_node(page) != dn->inode->i_ino) {
951 		f2fs_err(sbi, "incorrect node reference, ino: %lu, nid: %u, ino_of_node: %u",
952 				dn->inode->i_ino, dn->nid, ino_of_node(page));
953 		set_sbi_flag(sbi, SBI_NEED_FSCK);
954 		f2fs_handle_error(sbi, ERROR_INVALID_NODE_REFERENCE);
955 		f2fs_put_page(page, 1);
956 		return -EFSCORRUPTED;
957 	}
958 
959 	/* Make dnode_of_data for parameter */
960 	dn->node_page = page;
961 	dn->ofs_in_node = 0;
962 	f2fs_truncate_data_blocks_range(dn, ADDRS_PER_BLOCK(dn->inode));
963 	err = truncate_node(dn);
964 	if (err) {
965 		f2fs_put_page(page, 1);
966 		return err;
967 	}
968 
969 	return 1;
970 }
971 
972 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
973 						int ofs, int depth)
974 {
975 	struct dnode_of_data rdn = *dn;
976 	struct page *page;
977 	struct f2fs_node *rn;
978 	nid_t child_nid;
979 	unsigned int child_nofs;
980 	int freed = 0;
981 	int i, ret;
982 
983 	if (dn->nid == 0)
984 		return NIDS_PER_BLOCK + 1;
985 
986 	trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
987 
988 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
989 	if (IS_ERR(page)) {
990 		trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
991 		return PTR_ERR(page);
992 	}
993 
994 	f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
995 
996 	rn = F2FS_NODE(page);
997 	if (depth < 3) {
998 		for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
999 			child_nid = le32_to_cpu(rn->in.nid[i]);
1000 			if (child_nid == 0)
1001 				continue;
1002 			rdn.nid = child_nid;
1003 			ret = truncate_dnode(&rdn);
1004 			if (ret < 0)
1005 				goto out_err;
1006 			if (set_nid(page, i, 0, false))
1007 				dn->node_changed = true;
1008 		}
1009 	} else {
1010 		child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
1011 		for (i = ofs; i < NIDS_PER_BLOCK; i++) {
1012 			child_nid = le32_to_cpu(rn->in.nid[i]);
1013 			if (child_nid == 0) {
1014 				child_nofs += NIDS_PER_BLOCK + 1;
1015 				continue;
1016 			}
1017 			rdn.nid = child_nid;
1018 			ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
1019 			if (ret == (NIDS_PER_BLOCK + 1)) {
1020 				if (set_nid(page, i, 0, false))
1021 					dn->node_changed = true;
1022 				child_nofs += ret;
1023 			} else if (ret < 0 && ret != -ENOENT) {
1024 				goto out_err;
1025 			}
1026 		}
1027 		freed = child_nofs;
1028 	}
1029 
1030 	if (!ofs) {
1031 		/* remove current indirect node */
1032 		dn->node_page = page;
1033 		ret = truncate_node(dn);
1034 		if (ret)
1035 			goto out_err;
1036 		freed++;
1037 	} else {
1038 		f2fs_put_page(page, 1);
1039 	}
1040 	trace_f2fs_truncate_nodes_exit(dn->inode, freed);
1041 	return freed;
1042 
1043 out_err:
1044 	f2fs_put_page(page, 1);
1045 	trace_f2fs_truncate_nodes_exit(dn->inode, ret);
1046 	return ret;
1047 }
1048 
1049 static int truncate_partial_nodes(struct dnode_of_data *dn,
1050 			struct f2fs_inode *ri, int *offset, int depth)
1051 {
1052 	struct page *pages[2];
1053 	nid_t nid[3];
1054 	nid_t child_nid;
1055 	int err = 0;
1056 	int i;
1057 	int idx = depth - 2;
1058 
1059 	nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1060 	if (!nid[0])
1061 		return 0;
1062 
1063 	/* get indirect nodes in the path */
1064 	for (i = 0; i < idx + 1; i++) {
1065 		/* reference count'll be increased */
1066 		pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
1067 		if (IS_ERR(pages[i])) {
1068 			err = PTR_ERR(pages[i]);
1069 			idx = i - 1;
1070 			goto fail;
1071 		}
1072 		nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1073 	}
1074 
1075 	f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1076 
1077 	/* free direct nodes linked to a partial indirect node */
1078 	for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1079 		child_nid = get_nid(pages[idx], i, false);
1080 		if (!child_nid)
1081 			continue;
1082 		dn->nid = child_nid;
1083 		err = truncate_dnode(dn);
1084 		if (err < 0)
1085 			goto fail;
1086 		if (set_nid(pages[idx], i, 0, false))
1087 			dn->node_changed = true;
1088 	}
1089 
1090 	if (offset[idx + 1] == 0) {
1091 		dn->node_page = pages[idx];
1092 		dn->nid = nid[idx];
1093 		err = truncate_node(dn);
1094 		if (err)
1095 			goto fail;
1096 	} else {
1097 		f2fs_put_page(pages[idx], 1);
1098 	}
1099 	offset[idx]++;
1100 	offset[idx + 1] = 0;
1101 	idx--;
1102 fail:
1103 	for (i = idx; i >= 0; i--)
1104 		f2fs_put_page(pages[i], 1);
1105 
1106 	trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1107 
1108 	return err;
1109 }
1110 
1111 /*
1112  * All the block addresses of data and nodes should be nullified.
1113  */
1114 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1115 {
1116 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1117 	int err = 0, cont = 1;
1118 	int level, offset[4], noffset[4];
1119 	unsigned int nofs = 0;
1120 	struct f2fs_inode *ri;
1121 	struct dnode_of_data dn;
1122 	struct page *page;
1123 
1124 	trace_f2fs_truncate_inode_blocks_enter(inode, from);
1125 
1126 	level = get_node_path(inode, from, offset, noffset);
1127 	if (level < 0) {
1128 		trace_f2fs_truncate_inode_blocks_exit(inode, level);
1129 		return level;
1130 	}
1131 
1132 	page = f2fs_get_node_page(sbi, inode->i_ino);
1133 	if (IS_ERR(page)) {
1134 		trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1135 		return PTR_ERR(page);
1136 	}
1137 
1138 	set_new_dnode(&dn, inode, page, NULL, 0);
1139 	unlock_page(page);
1140 
1141 	ri = F2FS_INODE(page);
1142 	switch (level) {
1143 	case 0:
1144 	case 1:
1145 		nofs = noffset[1];
1146 		break;
1147 	case 2:
1148 		nofs = noffset[1];
1149 		if (!offset[level - 1])
1150 			goto skip_partial;
1151 		err = truncate_partial_nodes(&dn, ri, offset, level);
1152 		if (err < 0 && err != -ENOENT)
1153 			goto fail;
1154 		nofs += 1 + NIDS_PER_BLOCK;
1155 		break;
1156 	case 3:
1157 		nofs = 5 + 2 * NIDS_PER_BLOCK;
1158 		if (!offset[level - 1])
1159 			goto skip_partial;
1160 		err = truncate_partial_nodes(&dn, ri, offset, level);
1161 		if (err < 0 && err != -ENOENT)
1162 			goto fail;
1163 		break;
1164 	default:
1165 		BUG();
1166 	}
1167 
1168 skip_partial:
1169 	while (cont) {
1170 		dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1171 		switch (offset[0]) {
1172 		case NODE_DIR1_BLOCK:
1173 		case NODE_DIR2_BLOCK:
1174 			err = truncate_dnode(&dn);
1175 			break;
1176 
1177 		case NODE_IND1_BLOCK:
1178 		case NODE_IND2_BLOCK:
1179 			err = truncate_nodes(&dn, nofs, offset[1], 2);
1180 			break;
1181 
1182 		case NODE_DIND_BLOCK:
1183 			err = truncate_nodes(&dn, nofs, offset[1], 3);
1184 			cont = 0;
1185 			break;
1186 
1187 		default:
1188 			BUG();
1189 		}
1190 		if (err == -ENOENT) {
1191 			set_sbi_flag(F2FS_P_SB(page), SBI_NEED_FSCK);
1192 			f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1193 			f2fs_err_ratelimited(sbi,
1194 				"truncate node fail, ino:%lu, nid:%u, "
1195 				"offset[0]:%d, offset[1]:%d, nofs:%d",
1196 				inode->i_ino, dn.nid, offset[0],
1197 				offset[1], nofs);
1198 			err = 0;
1199 		}
1200 		if (err < 0)
1201 			goto fail;
1202 		if (offset[1] == 0 &&
1203 				ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1204 			lock_page(page);
1205 			BUG_ON(page->mapping != NODE_MAPPING(sbi));
1206 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1207 			ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1208 			set_page_dirty(page);
1209 			unlock_page(page);
1210 		}
1211 		offset[1] = 0;
1212 		offset[0]++;
1213 		nofs += err;
1214 	}
1215 fail:
1216 	f2fs_put_page(page, 0);
1217 	trace_f2fs_truncate_inode_blocks_exit(inode, err);
1218 	return err > 0 ? 0 : err;
1219 }
1220 
1221 /* caller must lock inode page */
1222 int f2fs_truncate_xattr_node(struct inode *inode)
1223 {
1224 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1225 	nid_t nid = F2FS_I(inode)->i_xattr_nid;
1226 	struct dnode_of_data dn;
1227 	struct page *npage;
1228 	int err;
1229 
1230 	if (!nid)
1231 		return 0;
1232 
1233 	npage = f2fs_get_node_page(sbi, nid);
1234 	if (IS_ERR(npage))
1235 		return PTR_ERR(npage);
1236 
1237 	set_new_dnode(&dn, inode, NULL, npage, nid);
1238 	err = truncate_node(&dn);
1239 	if (err) {
1240 		f2fs_put_page(npage, 1);
1241 		return err;
1242 	}
1243 
1244 	f2fs_i_xnid_write(inode, 0);
1245 
1246 	return 0;
1247 }
1248 
1249 /*
1250  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1251  * f2fs_unlock_op().
1252  */
1253 int f2fs_remove_inode_page(struct inode *inode)
1254 {
1255 	struct dnode_of_data dn;
1256 	int err;
1257 
1258 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1259 	err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1260 	if (err)
1261 		return err;
1262 
1263 	err = f2fs_truncate_xattr_node(inode);
1264 	if (err) {
1265 		f2fs_put_dnode(&dn);
1266 		return err;
1267 	}
1268 
1269 	/* remove potential inline_data blocks */
1270 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1271 				S_ISLNK(inode->i_mode))
1272 		f2fs_truncate_data_blocks_range(&dn, 1);
1273 
1274 	/* 0 is possible, after f2fs_new_inode() has failed */
1275 	if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1276 		f2fs_put_dnode(&dn);
1277 		return -EIO;
1278 	}
1279 
1280 	if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1281 		f2fs_warn(F2FS_I_SB(inode),
1282 			"f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1283 			inode->i_ino, (unsigned long long)inode->i_blocks);
1284 		set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1285 	}
1286 
1287 	/* will put inode & node pages */
1288 	err = truncate_node(&dn);
1289 	if (err) {
1290 		f2fs_put_dnode(&dn);
1291 		return err;
1292 	}
1293 	return 0;
1294 }
1295 
1296 struct page *f2fs_new_inode_page(struct inode *inode)
1297 {
1298 	struct dnode_of_data dn;
1299 
1300 	/* allocate inode page for new inode */
1301 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1302 
1303 	/* caller should f2fs_put_page(page, 1); */
1304 	return f2fs_new_node_page(&dn, 0);
1305 }
1306 
1307 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1308 {
1309 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1310 	struct node_info new_ni;
1311 	struct page *page;
1312 	int err;
1313 
1314 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1315 		return ERR_PTR(-EPERM);
1316 
1317 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1318 	if (!page)
1319 		return ERR_PTR(-ENOMEM);
1320 
1321 	if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1322 		goto fail;
1323 
1324 #ifdef CONFIG_F2FS_CHECK_FS
1325 	err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false);
1326 	if (err) {
1327 		dec_valid_node_count(sbi, dn->inode, !ofs);
1328 		goto fail;
1329 	}
1330 	if (unlikely(new_ni.blk_addr != NULL_ADDR)) {
1331 		err = -EFSCORRUPTED;
1332 		dec_valid_node_count(sbi, dn->inode, !ofs);
1333 		set_sbi_flag(sbi, SBI_NEED_FSCK);
1334 		f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
1335 		goto fail;
1336 	}
1337 #endif
1338 	new_ni.nid = dn->nid;
1339 	new_ni.ino = dn->inode->i_ino;
1340 	new_ni.blk_addr = NULL_ADDR;
1341 	new_ni.flag = 0;
1342 	new_ni.version = 0;
1343 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1344 
1345 	f2fs_wait_on_page_writeback(page, NODE, true, true);
1346 	fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1347 	set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1348 	if (!PageUptodate(page))
1349 		SetPageUptodate(page);
1350 	if (set_page_dirty(page))
1351 		dn->node_changed = true;
1352 
1353 	if (f2fs_has_xattr_block(ofs))
1354 		f2fs_i_xnid_write(dn->inode, dn->nid);
1355 
1356 	if (ofs == 0)
1357 		inc_valid_inode_count(sbi);
1358 	return page;
1359 fail:
1360 	clear_node_page_dirty(page);
1361 	f2fs_put_page(page, 1);
1362 	return ERR_PTR(err);
1363 }
1364 
1365 /*
1366  * Caller should do after getting the following values.
1367  * 0: f2fs_put_page(page, 0)
1368  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1369  */
1370 static int read_node_page(struct page *page, blk_opf_t op_flags)
1371 {
1372 	struct folio *folio = page_folio(page);
1373 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1374 	struct node_info ni;
1375 	struct f2fs_io_info fio = {
1376 		.sbi = sbi,
1377 		.type = NODE,
1378 		.op = REQ_OP_READ,
1379 		.op_flags = op_flags,
1380 		.page = page,
1381 		.encrypted_page = NULL,
1382 	};
1383 	int err;
1384 
1385 	if (folio_test_uptodate(folio)) {
1386 		if (!f2fs_inode_chksum_verify(sbi, page)) {
1387 			folio_clear_uptodate(folio);
1388 			return -EFSBADCRC;
1389 		}
1390 		return LOCKED_PAGE;
1391 	}
1392 
1393 	err = f2fs_get_node_info(sbi, folio->index, &ni, false);
1394 	if (err)
1395 		return err;
1396 
1397 	/* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1398 	if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR)) {
1399 		folio_clear_uptodate(folio);
1400 		return -ENOENT;
1401 	}
1402 
1403 	fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1404 
1405 	err = f2fs_submit_page_bio(&fio);
1406 
1407 	if (!err)
1408 		f2fs_update_iostat(sbi, NULL, FS_NODE_READ_IO, F2FS_BLKSIZE);
1409 
1410 	return err;
1411 }
1412 
1413 /*
1414  * Readahead a node page
1415  */
1416 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1417 {
1418 	struct page *apage;
1419 	int err;
1420 
1421 	if (!nid)
1422 		return;
1423 	if (f2fs_check_nid_range(sbi, nid))
1424 		return;
1425 
1426 	apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1427 	if (apage)
1428 		return;
1429 
1430 	apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1431 	if (!apage)
1432 		return;
1433 
1434 	err = read_node_page(apage, REQ_RAHEAD);
1435 	f2fs_put_page(apage, err ? 1 : 0);
1436 }
1437 
1438 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1439 					struct page *parent, int start)
1440 {
1441 	struct page *page;
1442 	int err;
1443 
1444 	if (!nid)
1445 		return ERR_PTR(-ENOENT);
1446 	if (f2fs_check_nid_range(sbi, nid))
1447 		return ERR_PTR(-EINVAL);
1448 repeat:
1449 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1450 	if (!page)
1451 		return ERR_PTR(-ENOMEM);
1452 
1453 	err = read_node_page(page, 0);
1454 	if (err < 0) {
1455 		goto out_put_err;
1456 	} else if (err == LOCKED_PAGE) {
1457 		err = 0;
1458 		goto page_hit;
1459 	}
1460 
1461 	if (parent)
1462 		f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1463 
1464 	lock_page(page);
1465 
1466 	if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1467 		f2fs_put_page(page, 1);
1468 		goto repeat;
1469 	}
1470 
1471 	if (unlikely(!PageUptodate(page))) {
1472 		err = -EIO;
1473 		goto out_err;
1474 	}
1475 
1476 	if (!f2fs_inode_chksum_verify(sbi, page)) {
1477 		err = -EFSBADCRC;
1478 		goto out_err;
1479 	}
1480 page_hit:
1481 	if (likely(nid == nid_of_node(page)))
1482 		return page;
1483 
1484 	f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1485 			  nid, nid_of_node(page), ino_of_node(page),
1486 			  ofs_of_node(page), cpver_of_node(page),
1487 			  next_blkaddr_of_node(page));
1488 	set_sbi_flag(sbi, SBI_NEED_FSCK);
1489 	f2fs_handle_error(sbi, ERROR_INCONSISTENT_FOOTER);
1490 	err = -EFSCORRUPTED;
1491 out_err:
1492 	ClearPageUptodate(page);
1493 out_put_err:
1494 	/* ENOENT comes from read_node_page which is not an error. */
1495 	if (err != -ENOENT)
1496 		f2fs_handle_page_eio(sbi, page_folio(page), NODE);
1497 	f2fs_put_page(page, 1);
1498 	return ERR_PTR(err);
1499 }
1500 
1501 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1502 {
1503 	return __get_node_page(sbi, nid, NULL, 0);
1504 }
1505 
1506 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1507 {
1508 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1509 	nid_t nid = get_nid(parent, start, false);
1510 
1511 	return __get_node_page(sbi, nid, parent, start);
1512 }
1513 
1514 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1515 {
1516 	struct inode *inode;
1517 	struct page *page;
1518 	int ret;
1519 
1520 	/* should flush inline_data before evict_inode */
1521 	inode = ilookup(sbi->sb, ino);
1522 	if (!inode)
1523 		return;
1524 
1525 	page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1526 					FGP_LOCK|FGP_NOWAIT, 0);
1527 	if (!page)
1528 		goto iput_out;
1529 
1530 	if (!PageUptodate(page))
1531 		goto page_out;
1532 
1533 	if (!PageDirty(page))
1534 		goto page_out;
1535 
1536 	if (!clear_page_dirty_for_io(page))
1537 		goto page_out;
1538 
1539 	ret = f2fs_write_inline_data(inode, page_folio(page));
1540 	inode_dec_dirty_pages(inode);
1541 	f2fs_remove_dirty_inode(inode);
1542 	if (ret)
1543 		set_page_dirty(page);
1544 page_out:
1545 	f2fs_put_page(page, 1);
1546 iput_out:
1547 	iput(inode);
1548 }
1549 
1550 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1551 {
1552 	pgoff_t index;
1553 	struct folio_batch fbatch;
1554 	struct page *last_page = NULL;
1555 	int nr_folios;
1556 
1557 	folio_batch_init(&fbatch);
1558 	index = 0;
1559 
1560 	while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1561 					(pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1562 					&fbatch))) {
1563 		int i;
1564 
1565 		for (i = 0; i < nr_folios; i++) {
1566 			struct page *page = &fbatch.folios[i]->page;
1567 
1568 			if (unlikely(f2fs_cp_error(sbi))) {
1569 				f2fs_put_page(last_page, 0);
1570 				folio_batch_release(&fbatch);
1571 				return ERR_PTR(-EIO);
1572 			}
1573 
1574 			if (!IS_DNODE(page) || !is_cold_node(page))
1575 				continue;
1576 			if (ino_of_node(page) != ino)
1577 				continue;
1578 
1579 			lock_page(page);
1580 
1581 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1582 continue_unlock:
1583 				unlock_page(page);
1584 				continue;
1585 			}
1586 			if (ino_of_node(page) != ino)
1587 				goto continue_unlock;
1588 
1589 			if (!PageDirty(page)) {
1590 				/* someone wrote it for us */
1591 				goto continue_unlock;
1592 			}
1593 
1594 			if (last_page)
1595 				f2fs_put_page(last_page, 0);
1596 
1597 			get_page(page);
1598 			last_page = page;
1599 			unlock_page(page);
1600 		}
1601 		folio_batch_release(&fbatch);
1602 		cond_resched();
1603 	}
1604 	return last_page;
1605 }
1606 
1607 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1608 				struct writeback_control *wbc, bool do_balance,
1609 				enum iostat_type io_type, unsigned int *seq_id)
1610 {
1611 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1612 	struct folio *folio = page_folio(page);
1613 	nid_t nid;
1614 	struct node_info ni;
1615 	struct f2fs_io_info fio = {
1616 		.sbi = sbi,
1617 		.ino = ino_of_node(page),
1618 		.type = NODE,
1619 		.op = REQ_OP_WRITE,
1620 		.op_flags = wbc_to_write_flags(wbc),
1621 		.page = page,
1622 		.encrypted_page = NULL,
1623 		.submitted = 0,
1624 		.io_type = io_type,
1625 		.io_wbc = wbc,
1626 	};
1627 	unsigned int seq;
1628 
1629 	trace_f2fs_writepage(folio, NODE);
1630 
1631 	if (unlikely(f2fs_cp_error(sbi))) {
1632 		/* keep node pages in remount-ro mode */
1633 		if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY)
1634 			goto redirty_out;
1635 		folio_clear_uptodate(folio);
1636 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1637 		folio_unlock(folio);
1638 		return 0;
1639 	}
1640 
1641 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1642 		goto redirty_out;
1643 
1644 	if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1645 			wbc->sync_mode == WB_SYNC_NONE &&
1646 			IS_DNODE(page) && is_cold_node(page))
1647 		goto redirty_out;
1648 
1649 	/* get old block addr of this node page */
1650 	nid = nid_of_node(page);
1651 	f2fs_bug_on(sbi, folio->index != nid);
1652 
1653 	if (f2fs_get_node_info(sbi, nid, &ni, !do_balance))
1654 		goto redirty_out;
1655 
1656 	if (wbc->for_reclaim) {
1657 		if (!f2fs_down_read_trylock(&sbi->node_write))
1658 			goto redirty_out;
1659 	} else {
1660 		f2fs_down_read(&sbi->node_write);
1661 	}
1662 
1663 	/* This page is already truncated */
1664 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1665 		folio_clear_uptodate(folio);
1666 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1667 		f2fs_up_read(&sbi->node_write);
1668 		folio_unlock(folio);
1669 		return 0;
1670 	}
1671 
1672 	if (__is_valid_data_blkaddr(ni.blk_addr) &&
1673 		!f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1674 					DATA_GENERIC_ENHANCE)) {
1675 		f2fs_up_read(&sbi->node_write);
1676 		goto redirty_out;
1677 	}
1678 
1679 	if (atomic && !test_opt(sbi, NOBARRIER))
1680 		fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1681 
1682 	/* should add to global list before clearing PAGECACHE status */
1683 	if (f2fs_in_warm_node_list(sbi, page)) {
1684 		seq = f2fs_add_fsync_node_entry(sbi, page);
1685 		if (seq_id)
1686 			*seq_id = seq;
1687 	}
1688 
1689 	folio_start_writeback(folio);
1690 
1691 	fio.old_blkaddr = ni.blk_addr;
1692 	f2fs_do_write_node_page(nid, &fio);
1693 	set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1694 	dec_page_count(sbi, F2FS_DIRTY_NODES);
1695 	f2fs_up_read(&sbi->node_write);
1696 
1697 	if (wbc->for_reclaim) {
1698 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1699 		submitted = NULL;
1700 	}
1701 
1702 	folio_unlock(folio);
1703 
1704 	if (unlikely(f2fs_cp_error(sbi))) {
1705 		f2fs_submit_merged_write(sbi, NODE);
1706 		submitted = NULL;
1707 	}
1708 	if (submitted)
1709 		*submitted = fio.submitted;
1710 
1711 	if (do_balance)
1712 		f2fs_balance_fs(sbi, false);
1713 	return 0;
1714 
1715 redirty_out:
1716 	folio_redirty_for_writepage(wbc, folio);
1717 	return AOP_WRITEPAGE_ACTIVATE;
1718 }
1719 
1720 int f2fs_move_node_page(struct page *node_page, int gc_type)
1721 {
1722 	int err = 0;
1723 
1724 	if (gc_type == FG_GC) {
1725 		struct writeback_control wbc = {
1726 			.sync_mode = WB_SYNC_ALL,
1727 			.nr_to_write = 1,
1728 			.for_reclaim = 0,
1729 		};
1730 
1731 		f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1732 
1733 		set_page_dirty(node_page);
1734 
1735 		if (!clear_page_dirty_for_io(node_page)) {
1736 			err = -EAGAIN;
1737 			goto out_page;
1738 		}
1739 
1740 		if (__write_node_page(node_page, false, NULL,
1741 					&wbc, false, FS_GC_NODE_IO, NULL)) {
1742 			err = -EAGAIN;
1743 			unlock_page(node_page);
1744 		}
1745 		goto release_page;
1746 	} else {
1747 		/* set page dirty and write it */
1748 		if (!folio_test_writeback(page_folio(node_page)))
1749 			set_page_dirty(node_page);
1750 	}
1751 out_page:
1752 	unlock_page(node_page);
1753 release_page:
1754 	f2fs_put_page(node_page, 0);
1755 	return err;
1756 }
1757 
1758 static int f2fs_write_node_page(struct page *page,
1759 				struct writeback_control *wbc)
1760 {
1761 	return __write_node_page(page, false, NULL, wbc, false,
1762 						FS_NODE_IO, NULL);
1763 }
1764 
1765 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1766 			struct writeback_control *wbc, bool atomic,
1767 			unsigned int *seq_id)
1768 {
1769 	pgoff_t index;
1770 	struct folio_batch fbatch;
1771 	int ret = 0;
1772 	struct page *last_page = NULL;
1773 	bool marked = false;
1774 	nid_t ino = inode->i_ino;
1775 	int nr_folios;
1776 	int nwritten = 0;
1777 
1778 	if (atomic) {
1779 		last_page = last_fsync_dnode(sbi, ino);
1780 		if (IS_ERR_OR_NULL(last_page))
1781 			return PTR_ERR_OR_ZERO(last_page);
1782 	}
1783 retry:
1784 	folio_batch_init(&fbatch);
1785 	index = 0;
1786 
1787 	while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1788 					(pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1789 					&fbatch))) {
1790 		int i;
1791 
1792 		for (i = 0; i < nr_folios; i++) {
1793 			struct page *page = &fbatch.folios[i]->page;
1794 			bool submitted = false;
1795 
1796 			if (unlikely(f2fs_cp_error(sbi))) {
1797 				f2fs_put_page(last_page, 0);
1798 				folio_batch_release(&fbatch);
1799 				ret = -EIO;
1800 				goto out;
1801 			}
1802 
1803 			if (!IS_DNODE(page) || !is_cold_node(page))
1804 				continue;
1805 			if (ino_of_node(page) != ino)
1806 				continue;
1807 
1808 			lock_page(page);
1809 
1810 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1811 continue_unlock:
1812 				unlock_page(page);
1813 				continue;
1814 			}
1815 			if (ino_of_node(page) != ino)
1816 				goto continue_unlock;
1817 
1818 			if (!PageDirty(page) && page != last_page) {
1819 				/* someone wrote it for us */
1820 				goto continue_unlock;
1821 			}
1822 
1823 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1824 
1825 			set_fsync_mark(page, 0);
1826 			set_dentry_mark(page, 0);
1827 
1828 			if (!atomic || page == last_page) {
1829 				set_fsync_mark(page, 1);
1830 				percpu_counter_inc(&sbi->rf_node_block_count);
1831 				if (IS_INODE(page)) {
1832 					if (is_inode_flag_set(inode,
1833 								FI_DIRTY_INODE))
1834 						f2fs_update_inode(inode, page);
1835 					set_dentry_mark(page,
1836 						f2fs_need_dentry_mark(sbi, ino));
1837 				}
1838 				/* may be written by other thread */
1839 				if (!PageDirty(page))
1840 					set_page_dirty(page);
1841 			}
1842 
1843 			if (!clear_page_dirty_for_io(page))
1844 				goto continue_unlock;
1845 
1846 			ret = __write_node_page(page, atomic &&
1847 						page == last_page,
1848 						&submitted, wbc, true,
1849 						FS_NODE_IO, seq_id);
1850 			if (ret) {
1851 				unlock_page(page);
1852 				f2fs_put_page(last_page, 0);
1853 				break;
1854 			} else if (submitted) {
1855 				nwritten++;
1856 			}
1857 
1858 			if (page == last_page) {
1859 				f2fs_put_page(page, 0);
1860 				marked = true;
1861 				break;
1862 			}
1863 		}
1864 		folio_batch_release(&fbatch);
1865 		cond_resched();
1866 
1867 		if (ret || marked)
1868 			break;
1869 	}
1870 	if (!ret && atomic && !marked) {
1871 		f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1872 			   ino, page_folio(last_page)->index);
1873 		lock_page(last_page);
1874 		f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1875 		set_page_dirty(last_page);
1876 		unlock_page(last_page);
1877 		goto retry;
1878 	}
1879 out:
1880 	if (nwritten)
1881 		f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1882 	return ret ? -EIO : 0;
1883 }
1884 
1885 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1886 {
1887 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1888 	bool clean;
1889 
1890 	if (inode->i_ino != ino)
1891 		return 0;
1892 
1893 	if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1894 		return 0;
1895 
1896 	spin_lock(&sbi->inode_lock[DIRTY_META]);
1897 	clean = list_empty(&F2FS_I(inode)->gdirty_list);
1898 	spin_unlock(&sbi->inode_lock[DIRTY_META]);
1899 
1900 	if (clean)
1901 		return 0;
1902 
1903 	inode = igrab(inode);
1904 	if (!inode)
1905 		return 0;
1906 	return 1;
1907 }
1908 
1909 static bool flush_dirty_inode(struct page *page)
1910 {
1911 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1912 	struct inode *inode;
1913 	nid_t ino = ino_of_node(page);
1914 
1915 	inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1916 	if (!inode)
1917 		return false;
1918 
1919 	f2fs_update_inode(inode, page);
1920 	unlock_page(page);
1921 
1922 	iput(inode);
1923 	return true;
1924 }
1925 
1926 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1927 {
1928 	pgoff_t index = 0;
1929 	struct folio_batch fbatch;
1930 	int nr_folios;
1931 
1932 	folio_batch_init(&fbatch);
1933 
1934 	while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1935 					(pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1936 					&fbatch))) {
1937 		int i;
1938 
1939 		for (i = 0; i < nr_folios; i++) {
1940 			struct page *page = &fbatch.folios[i]->page;
1941 
1942 			if (!IS_INODE(page))
1943 				continue;
1944 
1945 			lock_page(page);
1946 
1947 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1948 continue_unlock:
1949 				unlock_page(page);
1950 				continue;
1951 			}
1952 
1953 			if (!PageDirty(page)) {
1954 				/* someone wrote it for us */
1955 				goto continue_unlock;
1956 			}
1957 
1958 			/* flush inline_data, if it's async context. */
1959 			if (page_private_inline(page)) {
1960 				clear_page_private_inline(page);
1961 				unlock_page(page);
1962 				flush_inline_data(sbi, ino_of_node(page));
1963 				continue;
1964 			}
1965 			unlock_page(page);
1966 		}
1967 		folio_batch_release(&fbatch);
1968 		cond_resched();
1969 	}
1970 }
1971 
1972 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1973 				struct writeback_control *wbc,
1974 				bool do_balance, enum iostat_type io_type)
1975 {
1976 	pgoff_t index;
1977 	struct folio_batch fbatch;
1978 	int step = 0;
1979 	int nwritten = 0;
1980 	int ret = 0;
1981 	int nr_folios, done = 0;
1982 
1983 	folio_batch_init(&fbatch);
1984 
1985 next_step:
1986 	index = 0;
1987 
1988 	while (!done && (nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi),
1989 				&index, (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1990 				&fbatch))) {
1991 		int i;
1992 
1993 		for (i = 0; i < nr_folios; i++) {
1994 			struct page *page = &fbatch.folios[i]->page;
1995 			bool submitted = false;
1996 
1997 			/* give a priority to WB_SYNC threads */
1998 			if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1999 					wbc->sync_mode == WB_SYNC_NONE) {
2000 				done = 1;
2001 				break;
2002 			}
2003 
2004 			/*
2005 			 * flushing sequence with step:
2006 			 * 0. indirect nodes
2007 			 * 1. dentry dnodes
2008 			 * 2. file dnodes
2009 			 */
2010 			if (step == 0 && IS_DNODE(page))
2011 				continue;
2012 			if (step == 1 && (!IS_DNODE(page) ||
2013 						is_cold_node(page)))
2014 				continue;
2015 			if (step == 2 && (!IS_DNODE(page) ||
2016 						!is_cold_node(page)))
2017 				continue;
2018 lock_node:
2019 			if (wbc->sync_mode == WB_SYNC_ALL)
2020 				lock_page(page);
2021 			else if (!trylock_page(page))
2022 				continue;
2023 
2024 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
2025 continue_unlock:
2026 				unlock_page(page);
2027 				continue;
2028 			}
2029 
2030 			if (!PageDirty(page)) {
2031 				/* someone wrote it for us */
2032 				goto continue_unlock;
2033 			}
2034 
2035 			/* flush inline_data/inode, if it's async context. */
2036 			if (!do_balance)
2037 				goto write_node;
2038 
2039 			/* flush inline_data */
2040 			if (page_private_inline(page)) {
2041 				clear_page_private_inline(page);
2042 				unlock_page(page);
2043 				flush_inline_data(sbi, ino_of_node(page));
2044 				goto lock_node;
2045 			}
2046 
2047 			/* flush dirty inode */
2048 			if (IS_INODE(page) && flush_dirty_inode(page))
2049 				goto lock_node;
2050 write_node:
2051 			f2fs_wait_on_page_writeback(page, NODE, true, true);
2052 
2053 			if (!clear_page_dirty_for_io(page))
2054 				goto continue_unlock;
2055 
2056 			set_fsync_mark(page, 0);
2057 			set_dentry_mark(page, 0);
2058 
2059 			ret = __write_node_page(page, false, &submitted,
2060 						wbc, do_balance, io_type, NULL);
2061 			if (ret)
2062 				unlock_page(page);
2063 			else if (submitted)
2064 				nwritten++;
2065 
2066 			if (--wbc->nr_to_write == 0)
2067 				break;
2068 		}
2069 		folio_batch_release(&fbatch);
2070 		cond_resched();
2071 
2072 		if (wbc->nr_to_write == 0) {
2073 			step = 2;
2074 			break;
2075 		}
2076 	}
2077 
2078 	if (step < 2) {
2079 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2080 				wbc->sync_mode == WB_SYNC_NONE && step == 1)
2081 			goto out;
2082 		step++;
2083 		goto next_step;
2084 	}
2085 out:
2086 	if (nwritten)
2087 		f2fs_submit_merged_write(sbi, NODE);
2088 
2089 	if (unlikely(f2fs_cp_error(sbi)))
2090 		return -EIO;
2091 	return ret;
2092 }
2093 
2094 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2095 						unsigned int seq_id)
2096 {
2097 	struct fsync_node_entry *fn;
2098 	struct page *page;
2099 	struct list_head *head = &sbi->fsync_node_list;
2100 	unsigned long flags;
2101 	unsigned int cur_seq_id = 0;
2102 
2103 	while (seq_id && cur_seq_id < seq_id) {
2104 		spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2105 		if (list_empty(head)) {
2106 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2107 			break;
2108 		}
2109 		fn = list_first_entry(head, struct fsync_node_entry, list);
2110 		if (fn->seq_id > seq_id) {
2111 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2112 			break;
2113 		}
2114 		cur_seq_id = fn->seq_id;
2115 		page = fn->page;
2116 		get_page(page);
2117 		spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2118 
2119 		f2fs_wait_on_page_writeback(page, NODE, true, false);
2120 
2121 		put_page(page);
2122 	}
2123 
2124 	return filemap_check_errors(NODE_MAPPING(sbi));
2125 }
2126 
2127 static int f2fs_write_node_pages(struct address_space *mapping,
2128 			    struct writeback_control *wbc)
2129 {
2130 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2131 	struct blk_plug plug;
2132 	long diff;
2133 
2134 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2135 		goto skip_write;
2136 
2137 	/* balancing f2fs's metadata in background */
2138 	f2fs_balance_fs_bg(sbi, true);
2139 
2140 	/* collect a number of dirty node pages and write together */
2141 	if (wbc->sync_mode != WB_SYNC_ALL &&
2142 			get_pages(sbi, F2FS_DIRTY_NODES) <
2143 					nr_pages_to_skip(sbi, NODE))
2144 		goto skip_write;
2145 
2146 	if (wbc->sync_mode == WB_SYNC_ALL)
2147 		atomic_inc(&sbi->wb_sync_req[NODE]);
2148 	else if (atomic_read(&sbi->wb_sync_req[NODE])) {
2149 		/* to avoid potential deadlock */
2150 		if (current->plug)
2151 			blk_finish_plug(current->plug);
2152 		goto skip_write;
2153 	}
2154 
2155 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2156 
2157 	diff = nr_pages_to_write(sbi, NODE, wbc);
2158 	blk_start_plug(&plug);
2159 	f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2160 	blk_finish_plug(&plug);
2161 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2162 
2163 	if (wbc->sync_mode == WB_SYNC_ALL)
2164 		atomic_dec(&sbi->wb_sync_req[NODE]);
2165 	return 0;
2166 
2167 skip_write:
2168 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2169 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2170 	return 0;
2171 }
2172 
2173 static bool f2fs_dirty_node_folio(struct address_space *mapping,
2174 		struct folio *folio)
2175 {
2176 	trace_f2fs_set_page_dirty(folio, NODE);
2177 
2178 	if (!folio_test_uptodate(folio))
2179 		folio_mark_uptodate(folio);
2180 #ifdef CONFIG_F2FS_CHECK_FS
2181 	if (IS_INODE(&folio->page))
2182 		f2fs_inode_chksum_set(F2FS_M_SB(mapping), &folio->page);
2183 #endif
2184 	if (filemap_dirty_folio(mapping, folio)) {
2185 		inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
2186 		set_page_private_reference(&folio->page);
2187 		return true;
2188 	}
2189 	return false;
2190 }
2191 
2192 /*
2193  * Structure of the f2fs node operations
2194  */
2195 const struct address_space_operations f2fs_node_aops = {
2196 	.writepage	= f2fs_write_node_page,
2197 	.writepages	= f2fs_write_node_pages,
2198 	.dirty_folio	= f2fs_dirty_node_folio,
2199 	.invalidate_folio = f2fs_invalidate_folio,
2200 	.release_folio	= f2fs_release_folio,
2201 	.migrate_folio	= filemap_migrate_folio,
2202 };
2203 
2204 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2205 						nid_t n)
2206 {
2207 	return radix_tree_lookup(&nm_i->free_nid_root, n);
2208 }
2209 
2210 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2211 				struct free_nid *i)
2212 {
2213 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2214 	int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2215 
2216 	if (err)
2217 		return err;
2218 
2219 	nm_i->nid_cnt[FREE_NID]++;
2220 	list_add_tail(&i->list, &nm_i->free_nid_list);
2221 	return 0;
2222 }
2223 
2224 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2225 			struct free_nid *i, enum nid_state state)
2226 {
2227 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2228 
2229 	f2fs_bug_on(sbi, state != i->state);
2230 	nm_i->nid_cnt[state]--;
2231 	if (state == FREE_NID)
2232 		list_del(&i->list);
2233 	radix_tree_delete(&nm_i->free_nid_root, i->nid);
2234 }
2235 
2236 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2237 			enum nid_state org_state, enum nid_state dst_state)
2238 {
2239 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2240 
2241 	f2fs_bug_on(sbi, org_state != i->state);
2242 	i->state = dst_state;
2243 	nm_i->nid_cnt[org_state]--;
2244 	nm_i->nid_cnt[dst_state]++;
2245 
2246 	switch (dst_state) {
2247 	case PREALLOC_NID:
2248 		list_del(&i->list);
2249 		break;
2250 	case FREE_NID:
2251 		list_add_tail(&i->list, &nm_i->free_nid_list);
2252 		break;
2253 	default:
2254 		BUG_ON(1);
2255 	}
2256 }
2257 
2258 bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi)
2259 {
2260 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2261 	unsigned int i;
2262 	bool ret = true;
2263 
2264 	f2fs_down_read(&nm_i->nat_tree_lock);
2265 	for (i = 0; i < nm_i->nat_blocks; i++) {
2266 		if (!test_bit_le(i, nm_i->nat_block_bitmap)) {
2267 			ret = false;
2268 			break;
2269 		}
2270 	}
2271 	f2fs_up_read(&nm_i->nat_tree_lock);
2272 
2273 	return ret;
2274 }
2275 
2276 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2277 							bool set, bool build)
2278 {
2279 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2280 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2281 	unsigned int nid_ofs = nid - START_NID(nid);
2282 
2283 	if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2284 		return;
2285 
2286 	if (set) {
2287 		if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2288 			return;
2289 		__set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2290 		nm_i->free_nid_count[nat_ofs]++;
2291 	} else {
2292 		if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2293 			return;
2294 		__clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2295 		if (!build)
2296 			nm_i->free_nid_count[nat_ofs]--;
2297 	}
2298 }
2299 
2300 /* return if the nid is recognized as free */
2301 static bool add_free_nid(struct f2fs_sb_info *sbi,
2302 				nid_t nid, bool build, bool update)
2303 {
2304 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2305 	struct free_nid *i, *e;
2306 	struct nat_entry *ne;
2307 	int err = -EINVAL;
2308 	bool ret = false;
2309 
2310 	/* 0 nid should not be used */
2311 	if (unlikely(nid == 0))
2312 		return false;
2313 
2314 	if (unlikely(f2fs_check_nid_range(sbi, nid)))
2315 		return false;
2316 
2317 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL);
2318 	i->nid = nid;
2319 	i->state = FREE_NID;
2320 
2321 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2322 
2323 	spin_lock(&nm_i->nid_list_lock);
2324 
2325 	if (build) {
2326 		/*
2327 		 *   Thread A             Thread B
2328 		 *  - f2fs_create
2329 		 *   - f2fs_new_inode
2330 		 *    - f2fs_alloc_nid
2331 		 *     - __insert_nid_to_list(PREALLOC_NID)
2332 		 *                     - f2fs_balance_fs_bg
2333 		 *                      - f2fs_build_free_nids
2334 		 *                       - __f2fs_build_free_nids
2335 		 *                        - scan_nat_page
2336 		 *                         - add_free_nid
2337 		 *                          - __lookup_nat_cache
2338 		 *  - f2fs_add_link
2339 		 *   - f2fs_init_inode_metadata
2340 		 *    - f2fs_new_inode_page
2341 		 *     - f2fs_new_node_page
2342 		 *      - set_node_addr
2343 		 *  - f2fs_alloc_nid_done
2344 		 *   - __remove_nid_from_list(PREALLOC_NID)
2345 		 *                         - __insert_nid_to_list(FREE_NID)
2346 		 */
2347 		ne = __lookup_nat_cache(nm_i, nid);
2348 		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2349 				nat_get_blkaddr(ne) != NULL_ADDR))
2350 			goto err_out;
2351 
2352 		e = __lookup_free_nid_list(nm_i, nid);
2353 		if (e) {
2354 			if (e->state == FREE_NID)
2355 				ret = true;
2356 			goto err_out;
2357 		}
2358 	}
2359 	ret = true;
2360 	err = __insert_free_nid(sbi, i);
2361 err_out:
2362 	if (update) {
2363 		update_free_nid_bitmap(sbi, nid, ret, build);
2364 		if (!build)
2365 			nm_i->available_nids++;
2366 	}
2367 	spin_unlock(&nm_i->nid_list_lock);
2368 	radix_tree_preload_end();
2369 
2370 	if (err)
2371 		kmem_cache_free(free_nid_slab, i);
2372 	return ret;
2373 }
2374 
2375 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2376 {
2377 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2378 	struct free_nid *i;
2379 	bool need_free = false;
2380 
2381 	spin_lock(&nm_i->nid_list_lock);
2382 	i = __lookup_free_nid_list(nm_i, nid);
2383 	if (i && i->state == FREE_NID) {
2384 		__remove_free_nid(sbi, i, FREE_NID);
2385 		need_free = true;
2386 	}
2387 	spin_unlock(&nm_i->nid_list_lock);
2388 
2389 	if (need_free)
2390 		kmem_cache_free(free_nid_slab, i);
2391 }
2392 
2393 static int scan_nat_page(struct f2fs_sb_info *sbi,
2394 			struct page *nat_page, nid_t start_nid)
2395 {
2396 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2397 	struct f2fs_nat_block *nat_blk = page_address(nat_page);
2398 	block_t blk_addr;
2399 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2400 	int i;
2401 
2402 	__set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2403 
2404 	i = start_nid % NAT_ENTRY_PER_BLOCK;
2405 
2406 	for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2407 		if (unlikely(start_nid >= nm_i->max_nid))
2408 			break;
2409 
2410 		blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2411 
2412 		if (blk_addr == NEW_ADDR)
2413 			return -EFSCORRUPTED;
2414 
2415 		if (blk_addr == NULL_ADDR) {
2416 			add_free_nid(sbi, start_nid, true, true);
2417 		} else {
2418 			spin_lock(&NM_I(sbi)->nid_list_lock);
2419 			update_free_nid_bitmap(sbi, start_nid, false, true);
2420 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2421 		}
2422 	}
2423 
2424 	return 0;
2425 }
2426 
2427 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2428 {
2429 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2430 	struct f2fs_journal *journal = curseg->journal;
2431 	int i;
2432 
2433 	down_read(&curseg->journal_rwsem);
2434 	for (i = 0; i < nats_in_cursum(journal); i++) {
2435 		block_t addr;
2436 		nid_t nid;
2437 
2438 		addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2439 		nid = le32_to_cpu(nid_in_journal(journal, i));
2440 		if (addr == NULL_ADDR)
2441 			add_free_nid(sbi, nid, true, false);
2442 		else
2443 			remove_free_nid(sbi, nid);
2444 	}
2445 	up_read(&curseg->journal_rwsem);
2446 }
2447 
2448 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2449 {
2450 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2451 	unsigned int i, idx;
2452 	nid_t nid;
2453 
2454 	f2fs_down_read(&nm_i->nat_tree_lock);
2455 
2456 	for (i = 0; i < nm_i->nat_blocks; i++) {
2457 		if (!test_bit_le(i, nm_i->nat_block_bitmap))
2458 			continue;
2459 		if (!nm_i->free_nid_count[i])
2460 			continue;
2461 		for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2462 			idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2463 						NAT_ENTRY_PER_BLOCK, idx);
2464 			if (idx >= NAT_ENTRY_PER_BLOCK)
2465 				break;
2466 
2467 			nid = i * NAT_ENTRY_PER_BLOCK + idx;
2468 			add_free_nid(sbi, nid, true, false);
2469 
2470 			if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2471 				goto out;
2472 		}
2473 	}
2474 out:
2475 	scan_curseg_cache(sbi);
2476 
2477 	f2fs_up_read(&nm_i->nat_tree_lock);
2478 }
2479 
2480 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2481 						bool sync, bool mount)
2482 {
2483 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2484 	int i = 0, ret;
2485 	nid_t nid = nm_i->next_scan_nid;
2486 
2487 	if (unlikely(nid >= nm_i->max_nid))
2488 		nid = 0;
2489 
2490 	if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2491 		nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2492 
2493 	/* Enough entries */
2494 	if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2495 		return 0;
2496 
2497 	if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2498 		return 0;
2499 
2500 	if (!mount) {
2501 		/* try to find free nids in free_nid_bitmap */
2502 		scan_free_nid_bits(sbi);
2503 
2504 		if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2505 			return 0;
2506 	}
2507 
2508 	/* readahead nat pages to be scanned */
2509 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2510 							META_NAT, true);
2511 
2512 	f2fs_down_read(&nm_i->nat_tree_lock);
2513 
2514 	while (1) {
2515 		if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2516 						nm_i->nat_block_bitmap)) {
2517 			struct page *page = get_current_nat_page(sbi, nid);
2518 
2519 			if (IS_ERR(page)) {
2520 				ret = PTR_ERR(page);
2521 			} else {
2522 				ret = scan_nat_page(sbi, page, nid);
2523 				f2fs_put_page(page, 1);
2524 			}
2525 
2526 			if (ret) {
2527 				f2fs_up_read(&nm_i->nat_tree_lock);
2528 
2529 				if (ret == -EFSCORRUPTED) {
2530 					f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2531 					set_sbi_flag(sbi, SBI_NEED_FSCK);
2532 					f2fs_handle_error(sbi,
2533 						ERROR_INCONSISTENT_NAT);
2534 				}
2535 
2536 				return ret;
2537 			}
2538 		}
2539 
2540 		nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2541 		if (unlikely(nid >= nm_i->max_nid))
2542 			nid = 0;
2543 
2544 		if (++i >= FREE_NID_PAGES)
2545 			break;
2546 	}
2547 
2548 	/* go to the next free nat pages to find free nids abundantly */
2549 	nm_i->next_scan_nid = nid;
2550 
2551 	/* find free nids from current sum_pages */
2552 	scan_curseg_cache(sbi);
2553 
2554 	f2fs_up_read(&nm_i->nat_tree_lock);
2555 
2556 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2557 					nm_i->ra_nid_pages, META_NAT, false);
2558 
2559 	return 0;
2560 }
2561 
2562 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2563 {
2564 	int ret;
2565 
2566 	mutex_lock(&NM_I(sbi)->build_lock);
2567 	ret = __f2fs_build_free_nids(sbi, sync, mount);
2568 	mutex_unlock(&NM_I(sbi)->build_lock);
2569 
2570 	return ret;
2571 }
2572 
2573 /*
2574  * If this function returns success, caller can obtain a new nid
2575  * from second parameter of this function.
2576  * The returned nid could be used ino as well as nid when inode is created.
2577  */
2578 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2579 {
2580 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2581 	struct free_nid *i = NULL;
2582 retry:
2583 	if (time_to_inject(sbi, FAULT_ALLOC_NID))
2584 		return false;
2585 
2586 	spin_lock(&nm_i->nid_list_lock);
2587 
2588 	if (unlikely(nm_i->available_nids == 0)) {
2589 		spin_unlock(&nm_i->nid_list_lock);
2590 		return false;
2591 	}
2592 
2593 	/* We should not use stale free nids created by f2fs_build_free_nids */
2594 	if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2595 		f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2596 		i = list_first_entry(&nm_i->free_nid_list,
2597 					struct free_nid, list);
2598 		*nid = i->nid;
2599 
2600 		__move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2601 		nm_i->available_nids--;
2602 
2603 		update_free_nid_bitmap(sbi, *nid, false, false);
2604 
2605 		spin_unlock(&nm_i->nid_list_lock);
2606 		return true;
2607 	}
2608 	spin_unlock(&nm_i->nid_list_lock);
2609 
2610 	/* Let's scan nat pages and its caches to get free nids */
2611 	if (!f2fs_build_free_nids(sbi, true, false))
2612 		goto retry;
2613 	return false;
2614 }
2615 
2616 /*
2617  * f2fs_alloc_nid() should be called prior to this function.
2618  */
2619 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2620 {
2621 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2622 	struct free_nid *i;
2623 
2624 	spin_lock(&nm_i->nid_list_lock);
2625 	i = __lookup_free_nid_list(nm_i, nid);
2626 	f2fs_bug_on(sbi, !i);
2627 	__remove_free_nid(sbi, i, PREALLOC_NID);
2628 	spin_unlock(&nm_i->nid_list_lock);
2629 
2630 	kmem_cache_free(free_nid_slab, i);
2631 }
2632 
2633 /*
2634  * f2fs_alloc_nid() should be called prior to this function.
2635  */
2636 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2637 {
2638 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2639 	struct free_nid *i;
2640 	bool need_free = false;
2641 
2642 	if (!nid)
2643 		return;
2644 
2645 	spin_lock(&nm_i->nid_list_lock);
2646 	i = __lookup_free_nid_list(nm_i, nid);
2647 	f2fs_bug_on(sbi, !i);
2648 
2649 	if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2650 		__remove_free_nid(sbi, i, PREALLOC_NID);
2651 		need_free = true;
2652 	} else {
2653 		__move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2654 	}
2655 
2656 	nm_i->available_nids++;
2657 
2658 	update_free_nid_bitmap(sbi, nid, true, false);
2659 
2660 	spin_unlock(&nm_i->nid_list_lock);
2661 
2662 	if (need_free)
2663 		kmem_cache_free(free_nid_slab, i);
2664 }
2665 
2666 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2667 {
2668 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2669 	int nr = nr_shrink;
2670 
2671 	if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2672 		return 0;
2673 
2674 	if (!mutex_trylock(&nm_i->build_lock))
2675 		return 0;
2676 
2677 	while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2678 		struct free_nid *i, *next;
2679 		unsigned int batch = SHRINK_NID_BATCH_SIZE;
2680 
2681 		spin_lock(&nm_i->nid_list_lock);
2682 		list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2683 			if (!nr_shrink || !batch ||
2684 				nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2685 				break;
2686 			__remove_free_nid(sbi, i, FREE_NID);
2687 			kmem_cache_free(free_nid_slab, i);
2688 			nr_shrink--;
2689 			batch--;
2690 		}
2691 		spin_unlock(&nm_i->nid_list_lock);
2692 	}
2693 
2694 	mutex_unlock(&nm_i->build_lock);
2695 
2696 	return nr - nr_shrink;
2697 }
2698 
2699 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2700 {
2701 	void *src_addr, *dst_addr;
2702 	size_t inline_size;
2703 	struct page *ipage;
2704 	struct f2fs_inode *ri;
2705 
2706 	ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2707 	if (IS_ERR(ipage))
2708 		return PTR_ERR(ipage);
2709 
2710 	ri = F2FS_INODE(page);
2711 	if (ri->i_inline & F2FS_INLINE_XATTR) {
2712 		if (!f2fs_has_inline_xattr(inode)) {
2713 			set_inode_flag(inode, FI_INLINE_XATTR);
2714 			stat_inc_inline_xattr(inode);
2715 		}
2716 	} else {
2717 		if (f2fs_has_inline_xattr(inode)) {
2718 			stat_dec_inline_xattr(inode);
2719 			clear_inode_flag(inode, FI_INLINE_XATTR);
2720 		}
2721 		goto update_inode;
2722 	}
2723 
2724 	dst_addr = inline_xattr_addr(inode, ipage);
2725 	src_addr = inline_xattr_addr(inode, page);
2726 	inline_size = inline_xattr_size(inode);
2727 
2728 	f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2729 	memcpy(dst_addr, src_addr, inline_size);
2730 update_inode:
2731 	f2fs_update_inode(inode, ipage);
2732 	f2fs_put_page(ipage, 1);
2733 	return 0;
2734 }
2735 
2736 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2737 {
2738 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2739 	nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2740 	nid_t new_xnid;
2741 	struct dnode_of_data dn;
2742 	struct node_info ni;
2743 	struct page *xpage;
2744 	int err;
2745 
2746 	if (!prev_xnid)
2747 		goto recover_xnid;
2748 
2749 	/* 1: invalidate the previous xattr nid */
2750 	err = f2fs_get_node_info(sbi, prev_xnid, &ni, false);
2751 	if (err)
2752 		return err;
2753 
2754 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
2755 	dec_valid_node_count(sbi, inode, false);
2756 	set_node_addr(sbi, &ni, NULL_ADDR, false);
2757 
2758 recover_xnid:
2759 	/* 2: update xattr nid in inode */
2760 	if (!f2fs_alloc_nid(sbi, &new_xnid))
2761 		return -ENOSPC;
2762 
2763 	set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2764 	xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2765 	if (IS_ERR(xpage)) {
2766 		f2fs_alloc_nid_failed(sbi, new_xnid);
2767 		return PTR_ERR(xpage);
2768 	}
2769 
2770 	f2fs_alloc_nid_done(sbi, new_xnid);
2771 	f2fs_update_inode_page(inode);
2772 
2773 	/* 3: update and set xattr node page dirty */
2774 	if (page) {
2775 		memcpy(F2FS_NODE(xpage), F2FS_NODE(page),
2776 				VALID_XATTR_BLOCK_SIZE);
2777 		set_page_dirty(xpage);
2778 	}
2779 	f2fs_put_page(xpage, 1);
2780 
2781 	return 0;
2782 }
2783 
2784 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2785 {
2786 	struct f2fs_inode *src, *dst;
2787 	nid_t ino = ino_of_node(page);
2788 	struct node_info old_ni, new_ni;
2789 	struct page *ipage;
2790 	int err;
2791 
2792 	err = f2fs_get_node_info(sbi, ino, &old_ni, false);
2793 	if (err)
2794 		return err;
2795 
2796 	if (unlikely(old_ni.blk_addr != NULL_ADDR))
2797 		return -EINVAL;
2798 retry:
2799 	ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2800 	if (!ipage) {
2801 		memalloc_retry_wait(GFP_NOFS);
2802 		goto retry;
2803 	}
2804 
2805 	/* Should not use this inode from free nid list */
2806 	remove_free_nid(sbi, ino);
2807 
2808 	if (!PageUptodate(ipage))
2809 		SetPageUptodate(ipage);
2810 	fill_node_footer(ipage, ino, ino, 0, true);
2811 	set_cold_node(ipage, false);
2812 
2813 	src = F2FS_INODE(page);
2814 	dst = F2FS_INODE(ipage);
2815 
2816 	memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2817 	dst->i_size = 0;
2818 	dst->i_blocks = cpu_to_le64(1);
2819 	dst->i_links = cpu_to_le32(1);
2820 	dst->i_xattr_nid = 0;
2821 	dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2822 	if (dst->i_inline & F2FS_EXTRA_ATTR) {
2823 		dst->i_extra_isize = src->i_extra_isize;
2824 
2825 		if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2826 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2827 							i_inline_xattr_size))
2828 			dst->i_inline_xattr_size = src->i_inline_xattr_size;
2829 
2830 		if (f2fs_sb_has_project_quota(sbi) &&
2831 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2832 								i_projid))
2833 			dst->i_projid = src->i_projid;
2834 
2835 		if (f2fs_sb_has_inode_crtime(sbi) &&
2836 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2837 							i_crtime_nsec)) {
2838 			dst->i_crtime = src->i_crtime;
2839 			dst->i_crtime_nsec = src->i_crtime_nsec;
2840 		}
2841 	}
2842 
2843 	new_ni = old_ni;
2844 	new_ni.ino = ino;
2845 
2846 	if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2847 		WARN_ON(1);
2848 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2849 	inc_valid_inode_count(sbi);
2850 	set_page_dirty(ipage);
2851 	f2fs_put_page(ipage, 1);
2852 	return 0;
2853 }
2854 
2855 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2856 			unsigned int segno, struct f2fs_summary_block *sum)
2857 {
2858 	struct f2fs_node *rn;
2859 	struct f2fs_summary *sum_entry;
2860 	block_t addr;
2861 	int i, idx, last_offset, nrpages;
2862 
2863 	/* scan the node segment */
2864 	last_offset = BLKS_PER_SEG(sbi);
2865 	addr = START_BLOCK(sbi, segno);
2866 	sum_entry = &sum->entries[0];
2867 
2868 	for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2869 		nrpages = bio_max_segs(last_offset - i);
2870 
2871 		/* readahead node pages */
2872 		f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2873 
2874 		for (idx = addr; idx < addr + nrpages; idx++) {
2875 			struct page *page = f2fs_get_tmp_page(sbi, idx);
2876 
2877 			if (IS_ERR(page))
2878 				return PTR_ERR(page);
2879 
2880 			rn = F2FS_NODE(page);
2881 			sum_entry->nid = rn->footer.nid;
2882 			sum_entry->version = 0;
2883 			sum_entry->ofs_in_node = 0;
2884 			sum_entry++;
2885 			f2fs_put_page(page, 1);
2886 		}
2887 
2888 		invalidate_mapping_pages(META_MAPPING(sbi), addr,
2889 							addr + nrpages);
2890 	}
2891 	return 0;
2892 }
2893 
2894 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2895 {
2896 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2897 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2898 	struct f2fs_journal *journal = curseg->journal;
2899 	int i;
2900 
2901 	down_write(&curseg->journal_rwsem);
2902 	for (i = 0; i < nats_in_cursum(journal); i++) {
2903 		struct nat_entry *ne;
2904 		struct f2fs_nat_entry raw_ne;
2905 		nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2906 
2907 		if (f2fs_check_nid_range(sbi, nid))
2908 			continue;
2909 
2910 		raw_ne = nat_in_journal(journal, i);
2911 
2912 		ne = __lookup_nat_cache(nm_i, nid);
2913 		if (!ne) {
2914 			ne = __alloc_nat_entry(sbi, nid, true);
2915 			__init_nat_entry(nm_i, ne, &raw_ne, true);
2916 		}
2917 
2918 		/*
2919 		 * if a free nat in journal has not been used after last
2920 		 * checkpoint, we should remove it from available nids,
2921 		 * since later we will add it again.
2922 		 */
2923 		if (!get_nat_flag(ne, IS_DIRTY) &&
2924 				le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2925 			spin_lock(&nm_i->nid_list_lock);
2926 			nm_i->available_nids--;
2927 			spin_unlock(&nm_i->nid_list_lock);
2928 		}
2929 
2930 		__set_nat_cache_dirty(nm_i, ne);
2931 	}
2932 	update_nats_in_cursum(journal, -i);
2933 	up_write(&curseg->journal_rwsem);
2934 }
2935 
2936 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2937 						struct list_head *head, int max)
2938 {
2939 	struct nat_entry_set *cur;
2940 
2941 	if (nes->entry_cnt >= max)
2942 		goto add_out;
2943 
2944 	list_for_each_entry(cur, head, set_list) {
2945 		if (cur->entry_cnt >= nes->entry_cnt) {
2946 			list_add(&nes->set_list, cur->set_list.prev);
2947 			return;
2948 		}
2949 	}
2950 add_out:
2951 	list_add_tail(&nes->set_list, head);
2952 }
2953 
2954 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs,
2955 							unsigned int valid)
2956 {
2957 	if (valid == 0) {
2958 		__set_bit_le(nat_ofs, nm_i->empty_nat_bits);
2959 		__clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2960 		return;
2961 	}
2962 
2963 	__clear_bit_le(nat_ofs, nm_i->empty_nat_bits);
2964 	if (valid == NAT_ENTRY_PER_BLOCK)
2965 		__set_bit_le(nat_ofs, nm_i->full_nat_bits);
2966 	else
2967 		__clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2968 }
2969 
2970 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2971 						struct page *page)
2972 {
2973 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2974 	unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2975 	struct f2fs_nat_block *nat_blk = page_address(page);
2976 	int valid = 0;
2977 	int i = 0;
2978 
2979 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
2980 		return;
2981 
2982 	if (nat_index == 0) {
2983 		valid = 1;
2984 		i = 1;
2985 	}
2986 	for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2987 		if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2988 			valid++;
2989 	}
2990 
2991 	__update_nat_bits(nm_i, nat_index, valid);
2992 }
2993 
2994 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi)
2995 {
2996 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2997 	unsigned int nat_ofs;
2998 
2999 	f2fs_down_read(&nm_i->nat_tree_lock);
3000 
3001 	for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) {
3002 		unsigned int valid = 0, nid_ofs = 0;
3003 
3004 		/* handle nid zero due to it should never be used */
3005 		if (unlikely(nat_ofs == 0)) {
3006 			valid = 1;
3007 			nid_ofs = 1;
3008 		}
3009 
3010 		for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) {
3011 			if (!test_bit_le(nid_ofs,
3012 					nm_i->free_nid_bitmap[nat_ofs]))
3013 				valid++;
3014 		}
3015 
3016 		__update_nat_bits(nm_i, nat_ofs, valid);
3017 	}
3018 
3019 	f2fs_up_read(&nm_i->nat_tree_lock);
3020 }
3021 
3022 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
3023 		struct nat_entry_set *set, struct cp_control *cpc)
3024 {
3025 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3026 	struct f2fs_journal *journal = curseg->journal;
3027 	nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
3028 	bool to_journal = true;
3029 	struct f2fs_nat_block *nat_blk;
3030 	struct nat_entry *ne, *cur;
3031 	struct page *page = NULL;
3032 
3033 	/*
3034 	 * there are two steps to flush nat entries:
3035 	 * #1, flush nat entries to journal in current hot data summary block.
3036 	 * #2, flush nat entries to nat page.
3037 	 */
3038 	if ((cpc->reason & CP_UMOUNT) ||
3039 		!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
3040 		to_journal = false;
3041 
3042 	if (to_journal) {
3043 		down_write(&curseg->journal_rwsem);
3044 	} else {
3045 		page = get_next_nat_page(sbi, start_nid);
3046 		if (IS_ERR(page))
3047 			return PTR_ERR(page);
3048 
3049 		nat_blk = page_address(page);
3050 		f2fs_bug_on(sbi, !nat_blk);
3051 	}
3052 
3053 	/* flush dirty nats in nat entry set */
3054 	list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
3055 		struct f2fs_nat_entry *raw_ne;
3056 		nid_t nid = nat_get_nid(ne);
3057 		int offset;
3058 
3059 		f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
3060 
3061 		if (to_journal) {
3062 			offset = f2fs_lookup_journal_in_cursum(journal,
3063 							NAT_JOURNAL, nid, 1);
3064 			f2fs_bug_on(sbi, offset < 0);
3065 			raw_ne = &nat_in_journal(journal, offset);
3066 			nid_in_journal(journal, offset) = cpu_to_le32(nid);
3067 		} else {
3068 			raw_ne = &nat_blk->entries[nid - start_nid];
3069 		}
3070 		raw_nat_from_node_info(raw_ne, &ne->ni);
3071 		nat_reset_flag(ne);
3072 		__clear_nat_cache_dirty(NM_I(sbi), set, ne);
3073 		if (nat_get_blkaddr(ne) == NULL_ADDR) {
3074 			add_free_nid(sbi, nid, false, true);
3075 		} else {
3076 			spin_lock(&NM_I(sbi)->nid_list_lock);
3077 			update_free_nid_bitmap(sbi, nid, false, false);
3078 			spin_unlock(&NM_I(sbi)->nid_list_lock);
3079 		}
3080 	}
3081 
3082 	if (to_journal) {
3083 		up_write(&curseg->journal_rwsem);
3084 	} else {
3085 		update_nat_bits(sbi, start_nid, page);
3086 		f2fs_put_page(page, 1);
3087 	}
3088 
3089 	/* Allow dirty nats by node block allocation in write_begin */
3090 	if (!set->entry_cnt) {
3091 		radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
3092 		kmem_cache_free(nat_entry_set_slab, set);
3093 	}
3094 	return 0;
3095 }
3096 
3097 /*
3098  * This function is called during the checkpointing process.
3099  */
3100 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3101 {
3102 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3103 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3104 	struct f2fs_journal *journal = curseg->journal;
3105 	struct nat_entry_set *setvec[NAT_VEC_SIZE];
3106 	struct nat_entry_set *set, *tmp;
3107 	unsigned int found;
3108 	nid_t set_idx = 0;
3109 	LIST_HEAD(sets);
3110 	int err = 0;
3111 
3112 	/*
3113 	 * during unmount, let's flush nat_bits before checking
3114 	 * nat_cnt[DIRTY_NAT].
3115 	 */
3116 	if (cpc->reason & CP_UMOUNT) {
3117 		f2fs_down_write(&nm_i->nat_tree_lock);
3118 		remove_nats_in_journal(sbi);
3119 		f2fs_up_write(&nm_i->nat_tree_lock);
3120 	}
3121 
3122 	if (!nm_i->nat_cnt[DIRTY_NAT])
3123 		return 0;
3124 
3125 	f2fs_down_write(&nm_i->nat_tree_lock);
3126 
3127 	/*
3128 	 * if there are no enough space in journal to store dirty nat
3129 	 * entries, remove all entries from journal and merge them
3130 	 * into nat entry set.
3131 	 */
3132 	if (cpc->reason & CP_UMOUNT ||
3133 		!__has_cursum_space(journal,
3134 			nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3135 		remove_nats_in_journal(sbi);
3136 
3137 	while ((found = __gang_lookup_nat_set(nm_i,
3138 					set_idx, NAT_VEC_SIZE, setvec))) {
3139 		unsigned idx;
3140 
3141 		set_idx = setvec[found - 1]->set + 1;
3142 		for (idx = 0; idx < found; idx++)
3143 			__adjust_nat_entry_set(setvec[idx], &sets,
3144 						MAX_NAT_JENTRIES(journal));
3145 	}
3146 
3147 	/* flush dirty nats in nat entry set */
3148 	list_for_each_entry_safe(set, tmp, &sets, set_list) {
3149 		err = __flush_nat_entry_set(sbi, set, cpc);
3150 		if (err)
3151 			break;
3152 	}
3153 
3154 	f2fs_up_write(&nm_i->nat_tree_lock);
3155 	/* Allow dirty nats by node block allocation in write_begin */
3156 
3157 	return err;
3158 }
3159 
3160 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3161 {
3162 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3163 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3164 	unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3165 	unsigned int i;
3166 	__u64 cp_ver = cur_cp_version(ckpt);
3167 	block_t nat_bits_addr;
3168 
3169 	nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3170 	nm_i->nat_bits = f2fs_kvzalloc(sbi,
3171 			F2FS_BLK_TO_BYTES(nm_i->nat_bits_blocks), GFP_KERNEL);
3172 	if (!nm_i->nat_bits)
3173 		return -ENOMEM;
3174 
3175 	nm_i->full_nat_bits = nm_i->nat_bits + 8;
3176 	nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3177 
3178 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3179 		return 0;
3180 
3181 	nat_bits_addr = __start_cp_addr(sbi) + BLKS_PER_SEG(sbi) -
3182 						nm_i->nat_bits_blocks;
3183 	for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3184 		struct page *page;
3185 
3186 		page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3187 		if (IS_ERR(page))
3188 			return PTR_ERR(page);
3189 
3190 		memcpy(nm_i->nat_bits + F2FS_BLK_TO_BYTES(i),
3191 					page_address(page), F2FS_BLKSIZE);
3192 		f2fs_put_page(page, 1);
3193 	}
3194 
3195 	cp_ver |= (cur_cp_crc(ckpt) << 32);
3196 	if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3197 		clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
3198 		f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)",
3199 			cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits));
3200 		return 0;
3201 	}
3202 
3203 	f2fs_notice(sbi, "Found nat_bits in checkpoint");
3204 	return 0;
3205 }
3206 
3207 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3208 {
3209 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3210 	unsigned int i = 0;
3211 	nid_t nid, last_nid;
3212 
3213 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3214 		return;
3215 
3216 	for (i = 0; i < nm_i->nat_blocks; i++) {
3217 		i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3218 		if (i >= nm_i->nat_blocks)
3219 			break;
3220 
3221 		__set_bit_le(i, nm_i->nat_block_bitmap);
3222 
3223 		nid = i * NAT_ENTRY_PER_BLOCK;
3224 		last_nid = nid + NAT_ENTRY_PER_BLOCK;
3225 
3226 		spin_lock(&NM_I(sbi)->nid_list_lock);
3227 		for (; nid < last_nid; nid++)
3228 			update_free_nid_bitmap(sbi, nid, true, true);
3229 		spin_unlock(&NM_I(sbi)->nid_list_lock);
3230 	}
3231 
3232 	for (i = 0; i < nm_i->nat_blocks; i++) {
3233 		i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3234 		if (i >= nm_i->nat_blocks)
3235 			break;
3236 
3237 		__set_bit_le(i, nm_i->nat_block_bitmap);
3238 	}
3239 }
3240 
3241 static int init_node_manager(struct f2fs_sb_info *sbi)
3242 {
3243 	struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3244 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3245 	unsigned char *version_bitmap;
3246 	unsigned int nat_segs;
3247 	int err;
3248 
3249 	nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3250 
3251 	/* segment_count_nat includes pair segment so divide to 2. */
3252 	nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3253 	nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3254 	nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3255 
3256 	/* not used nids: 0, node, meta, (and root counted as valid node) */
3257 	nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3258 						F2FS_RESERVED_NODE_NUM;
3259 	nm_i->nid_cnt[FREE_NID] = 0;
3260 	nm_i->nid_cnt[PREALLOC_NID] = 0;
3261 	nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3262 	nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3263 	nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3264 	nm_i->max_rf_node_blocks = DEF_RF_NODE_BLOCKS;
3265 
3266 	INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3267 	INIT_LIST_HEAD(&nm_i->free_nid_list);
3268 	INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3269 	INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3270 	INIT_LIST_HEAD(&nm_i->nat_entries);
3271 	spin_lock_init(&nm_i->nat_list_lock);
3272 
3273 	mutex_init(&nm_i->build_lock);
3274 	spin_lock_init(&nm_i->nid_list_lock);
3275 	init_f2fs_rwsem(&nm_i->nat_tree_lock);
3276 
3277 	nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3278 	nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3279 	version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3280 	nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3281 					GFP_KERNEL);
3282 	if (!nm_i->nat_bitmap)
3283 		return -ENOMEM;
3284 
3285 	err = __get_nat_bitmaps(sbi);
3286 	if (err)
3287 		return err;
3288 
3289 #ifdef CONFIG_F2FS_CHECK_FS
3290 	nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3291 					GFP_KERNEL);
3292 	if (!nm_i->nat_bitmap_mir)
3293 		return -ENOMEM;
3294 #endif
3295 
3296 	return 0;
3297 }
3298 
3299 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3300 {
3301 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3302 	int i;
3303 
3304 	nm_i->free_nid_bitmap =
3305 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3306 					      nm_i->nat_blocks),
3307 			      GFP_KERNEL);
3308 	if (!nm_i->free_nid_bitmap)
3309 		return -ENOMEM;
3310 
3311 	for (i = 0; i < nm_i->nat_blocks; i++) {
3312 		nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3313 			f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3314 		if (!nm_i->free_nid_bitmap[i])
3315 			return -ENOMEM;
3316 	}
3317 
3318 	nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3319 								GFP_KERNEL);
3320 	if (!nm_i->nat_block_bitmap)
3321 		return -ENOMEM;
3322 
3323 	nm_i->free_nid_count =
3324 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3325 					      nm_i->nat_blocks),
3326 			      GFP_KERNEL);
3327 	if (!nm_i->free_nid_count)
3328 		return -ENOMEM;
3329 	return 0;
3330 }
3331 
3332 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3333 {
3334 	int err;
3335 
3336 	sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3337 							GFP_KERNEL);
3338 	if (!sbi->nm_info)
3339 		return -ENOMEM;
3340 
3341 	err = init_node_manager(sbi);
3342 	if (err)
3343 		return err;
3344 
3345 	err = init_free_nid_cache(sbi);
3346 	if (err)
3347 		return err;
3348 
3349 	/* load free nid status from nat_bits table */
3350 	load_free_nid_bitmap(sbi);
3351 
3352 	return f2fs_build_free_nids(sbi, true, true);
3353 }
3354 
3355 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3356 {
3357 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3358 	struct free_nid *i, *next_i;
3359 	void *vec[NAT_VEC_SIZE];
3360 	struct nat_entry **natvec = (struct nat_entry **)vec;
3361 	struct nat_entry_set **setvec = (struct nat_entry_set **)vec;
3362 	nid_t nid = 0;
3363 	unsigned int found;
3364 
3365 	if (!nm_i)
3366 		return;
3367 
3368 	/* destroy free nid list */
3369 	spin_lock(&nm_i->nid_list_lock);
3370 	list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3371 		__remove_free_nid(sbi, i, FREE_NID);
3372 		spin_unlock(&nm_i->nid_list_lock);
3373 		kmem_cache_free(free_nid_slab, i);
3374 		spin_lock(&nm_i->nid_list_lock);
3375 	}
3376 	f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3377 	f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3378 	f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3379 	spin_unlock(&nm_i->nid_list_lock);
3380 
3381 	/* destroy nat cache */
3382 	f2fs_down_write(&nm_i->nat_tree_lock);
3383 	while ((found = __gang_lookup_nat_cache(nm_i,
3384 					nid, NAT_VEC_SIZE, natvec))) {
3385 		unsigned idx;
3386 
3387 		nid = nat_get_nid(natvec[found - 1]) + 1;
3388 		for (idx = 0; idx < found; idx++) {
3389 			spin_lock(&nm_i->nat_list_lock);
3390 			list_del(&natvec[idx]->list);
3391 			spin_unlock(&nm_i->nat_list_lock);
3392 
3393 			__del_from_nat_cache(nm_i, natvec[idx]);
3394 		}
3395 	}
3396 	f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3397 
3398 	/* destroy nat set cache */
3399 	nid = 0;
3400 	memset(vec, 0, sizeof(void *) * NAT_VEC_SIZE);
3401 	while ((found = __gang_lookup_nat_set(nm_i,
3402 					nid, NAT_VEC_SIZE, setvec))) {
3403 		unsigned idx;
3404 
3405 		nid = setvec[found - 1]->set + 1;
3406 		for (idx = 0; idx < found; idx++) {
3407 			/* entry_cnt is not zero, when cp_error was occurred */
3408 			f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3409 			radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3410 			kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3411 		}
3412 	}
3413 	f2fs_up_write(&nm_i->nat_tree_lock);
3414 
3415 	kvfree(nm_i->nat_block_bitmap);
3416 	if (nm_i->free_nid_bitmap) {
3417 		int i;
3418 
3419 		for (i = 0; i < nm_i->nat_blocks; i++)
3420 			kvfree(nm_i->free_nid_bitmap[i]);
3421 		kvfree(nm_i->free_nid_bitmap);
3422 	}
3423 	kvfree(nm_i->free_nid_count);
3424 
3425 	kvfree(nm_i->nat_bitmap);
3426 	kvfree(nm_i->nat_bits);
3427 #ifdef CONFIG_F2FS_CHECK_FS
3428 	kvfree(nm_i->nat_bitmap_mir);
3429 #endif
3430 	sbi->nm_info = NULL;
3431 	kfree(nm_i);
3432 }
3433 
3434 int __init f2fs_create_node_manager_caches(void)
3435 {
3436 	nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3437 			sizeof(struct nat_entry));
3438 	if (!nat_entry_slab)
3439 		goto fail;
3440 
3441 	free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3442 			sizeof(struct free_nid));
3443 	if (!free_nid_slab)
3444 		goto destroy_nat_entry;
3445 
3446 	nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3447 			sizeof(struct nat_entry_set));
3448 	if (!nat_entry_set_slab)
3449 		goto destroy_free_nid;
3450 
3451 	fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3452 			sizeof(struct fsync_node_entry));
3453 	if (!fsync_node_entry_slab)
3454 		goto destroy_nat_entry_set;
3455 	return 0;
3456 
3457 destroy_nat_entry_set:
3458 	kmem_cache_destroy(nat_entry_set_slab);
3459 destroy_free_nid:
3460 	kmem_cache_destroy(free_nid_slab);
3461 destroy_nat_entry:
3462 	kmem_cache_destroy(nat_entry_slab);
3463 fail:
3464 	return -ENOMEM;
3465 }
3466 
3467 void f2fs_destroy_node_manager_caches(void)
3468 {
3469 	kmem_cache_destroy(fsync_node_entry_slab);
3470 	kmem_cache_destroy(nat_entry_set_slab);
3471 	kmem_cache_destroy(free_nid_slab);
3472 	kmem_cache_destroy(nat_entry_slab);
3473 }
3474