xref: /linux/fs/f2fs/node.c (revision c1f3caff2450048ab6c053e5b23698b58f286159)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * fs/f2fs/node.c
4  *
5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6  *             http://www.samsung.com/
7  */
8 #include <linux/fs.h>
9 #include <linux/f2fs_fs.h>
10 #include <linux/mpage.h>
11 #include <linux/sched/mm.h>
12 #include <linux/blkdev.h>
13 #include <linux/pagevec.h>
14 #include <linux/swap.h>
15 
16 #include "f2fs.h"
17 #include "node.h"
18 #include "segment.h"
19 #include "xattr.h"
20 #include "iostat.h"
21 #include <trace/events/f2fs.h>
22 
23 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
24 
25 static struct kmem_cache *nat_entry_slab;
26 static struct kmem_cache *free_nid_slab;
27 static struct kmem_cache *nat_entry_set_slab;
28 static struct kmem_cache *fsync_node_entry_slab;
29 
30 /*
31  * Check whether the given nid is within node id range.
32  */
33 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
34 {
35 	if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
36 		set_sbi_flag(sbi, SBI_NEED_FSCK);
37 		f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
38 			  __func__, nid);
39 		return -EFSCORRUPTED;
40 	}
41 	return 0;
42 }
43 
44 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
45 {
46 	struct f2fs_nm_info *nm_i = NM_I(sbi);
47 	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
48 	struct sysinfo val;
49 	unsigned long avail_ram;
50 	unsigned long mem_size = 0;
51 	bool res = false;
52 
53 	if (!nm_i)
54 		return true;
55 
56 	si_meminfo(&val);
57 
58 	/* only uses low memory */
59 	avail_ram = val.totalram - val.totalhigh;
60 
61 	/*
62 	 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
63 	 */
64 	if (type == FREE_NIDS) {
65 		mem_size = (nm_i->nid_cnt[FREE_NID] *
66 				sizeof(struct free_nid)) >> PAGE_SHIFT;
67 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
68 	} else if (type == NAT_ENTRIES) {
69 		mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
70 				sizeof(struct nat_entry)) >> PAGE_SHIFT;
71 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
72 		if (excess_cached_nats(sbi))
73 			res = false;
74 	} else if (type == DIRTY_DENTS) {
75 		if (sbi->sb->s_bdi->wb.dirty_exceeded)
76 			return false;
77 		mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
78 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
79 	} else if (type == INO_ENTRIES) {
80 		int i;
81 
82 		for (i = 0; i < MAX_INO_ENTRY; i++)
83 			mem_size += sbi->im[i].ino_num *
84 						sizeof(struct ino_entry);
85 		mem_size >>= PAGE_SHIFT;
86 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
87 	} else if (type == EXTENT_CACHE) {
88 		mem_size = (atomic_read(&sbi->total_ext_tree) *
89 				sizeof(struct extent_tree) +
90 				atomic_read(&sbi->total_ext_node) *
91 				sizeof(struct extent_node)) >> PAGE_SHIFT;
92 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
93 	} else if (type == INMEM_PAGES) {
94 		/* it allows 20% / total_ram for inmemory pages */
95 		mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
96 		res = mem_size < (val.totalram / 5);
97 	} else if (type == DISCARD_CACHE) {
98 		mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
99 				sizeof(struct discard_cmd)) >> PAGE_SHIFT;
100 		res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
101 	} else if (type == COMPRESS_PAGE) {
102 #ifdef CONFIG_F2FS_FS_COMPRESSION
103 		unsigned long free_ram = val.freeram;
104 
105 		/*
106 		 * free memory is lower than watermark or cached page count
107 		 * exceed threshold, deny caching compress page.
108 		 */
109 		res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
110 			(COMPRESS_MAPPING(sbi)->nrpages <
111 			 free_ram * sbi->compress_percent / 100);
112 #else
113 		res = false;
114 #endif
115 	} else {
116 		if (!sbi->sb->s_bdi->wb.dirty_exceeded)
117 			return true;
118 	}
119 	return res;
120 }
121 
122 static void clear_node_page_dirty(struct page *page)
123 {
124 	if (PageDirty(page)) {
125 		f2fs_clear_page_cache_dirty_tag(page);
126 		clear_page_dirty_for_io(page);
127 		dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
128 	}
129 	ClearPageUptodate(page);
130 }
131 
132 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
133 {
134 	return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
135 }
136 
137 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
138 {
139 	struct page *src_page;
140 	struct page *dst_page;
141 	pgoff_t dst_off;
142 	void *src_addr;
143 	void *dst_addr;
144 	struct f2fs_nm_info *nm_i = NM_I(sbi);
145 
146 	dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
147 
148 	/* get current nat block page with lock */
149 	src_page = get_current_nat_page(sbi, nid);
150 	if (IS_ERR(src_page))
151 		return src_page;
152 	dst_page = f2fs_grab_meta_page(sbi, dst_off);
153 	f2fs_bug_on(sbi, PageDirty(src_page));
154 
155 	src_addr = page_address(src_page);
156 	dst_addr = page_address(dst_page);
157 	memcpy(dst_addr, src_addr, PAGE_SIZE);
158 	set_page_dirty(dst_page);
159 	f2fs_put_page(src_page, 1);
160 
161 	set_to_next_nat(nm_i, nid);
162 
163 	return dst_page;
164 }
165 
166 static struct nat_entry *__alloc_nat_entry(struct f2fs_sb_info *sbi,
167 						nid_t nid, bool no_fail)
168 {
169 	struct nat_entry *new;
170 
171 	new = f2fs_kmem_cache_alloc(nat_entry_slab,
172 					GFP_F2FS_ZERO, no_fail, sbi);
173 	if (new) {
174 		nat_set_nid(new, nid);
175 		nat_reset_flag(new);
176 	}
177 	return new;
178 }
179 
180 static void __free_nat_entry(struct nat_entry *e)
181 {
182 	kmem_cache_free(nat_entry_slab, e);
183 }
184 
185 /* must be locked by nat_tree_lock */
186 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
187 	struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
188 {
189 	if (no_fail)
190 		f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
191 	else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
192 		return NULL;
193 
194 	if (raw_ne)
195 		node_info_from_raw_nat(&ne->ni, raw_ne);
196 
197 	spin_lock(&nm_i->nat_list_lock);
198 	list_add_tail(&ne->list, &nm_i->nat_entries);
199 	spin_unlock(&nm_i->nat_list_lock);
200 
201 	nm_i->nat_cnt[TOTAL_NAT]++;
202 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
203 	return ne;
204 }
205 
206 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
207 {
208 	struct nat_entry *ne;
209 
210 	ne = radix_tree_lookup(&nm_i->nat_root, n);
211 
212 	/* for recent accessed nat entry, move it to tail of lru list */
213 	if (ne && !get_nat_flag(ne, IS_DIRTY)) {
214 		spin_lock(&nm_i->nat_list_lock);
215 		if (!list_empty(&ne->list))
216 			list_move_tail(&ne->list, &nm_i->nat_entries);
217 		spin_unlock(&nm_i->nat_list_lock);
218 	}
219 
220 	return ne;
221 }
222 
223 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
224 		nid_t start, unsigned int nr, struct nat_entry **ep)
225 {
226 	return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
227 }
228 
229 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
230 {
231 	radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
232 	nm_i->nat_cnt[TOTAL_NAT]--;
233 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
234 	__free_nat_entry(e);
235 }
236 
237 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
238 							struct nat_entry *ne)
239 {
240 	nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
241 	struct nat_entry_set *head;
242 
243 	head = radix_tree_lookup(&nm_i->nat_set_root, set);
244 	if (!head) {
245 		head = f2fs_kmem_cache_alloc(nat_entry_set_slab,
246 						GFP_NOFS, true, NULL);
247 
248 		INIT_LIST_HEAD(&head->entry_list);
249 		INIT_LIST_HEAD(&head->set_list);
250 		head->set = set;
251 		head->entry_cnt = 0;
252 		f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
253 	}
254 	return head;
255 }
256 
257 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
258 						struct nat_entry *ne)
259 {
260 	struct nat_entry_set *head;
261 	bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
262 
263 	if (!new_ne)
264 		head = __grab_nat_entry_set(nm_i, ne);
265 
266 	/*
267 	 * update entry_cnt in below condition:
268 	 * 1. update NEW_ADDR to valid block address;
269 	 * 2. update old block address to new one;
270 	 */
271 	if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
272 				!get_nat_flag(ne, IS_DIRTY)))
273 		head->entry_cnt++;
274 
275 	set_nat_flag(ne, IS_PREALLOC, new_ne);
276 
277 	if (get_nat_flag(ne, IS_DIRTY))
278 		goto refresh_list;
279 
280 	nm_i->nat_cnt[DIRTY_NAT]++;
281 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
282 	set_nat_flag(ne, IS_DIRTY, true);
283 refresh_list:
284 	spin_lock(&nm_i->nat_list_lock);
285 	if (new_ne)
286 		list_del_init(&ne->list);
287 	else
288 		list_move_tail(&ne->list, &head->entry_list);
289 	spin_unlock(&nm_i->nat_list_lock);
290 }
291 
292 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
293 		struct nat_entry_set *set, struct nat_entry *ne)
294 {
295 	spin_lock(&nm_i->nat_list_lock);
296 	list_move_tail(&ne->list, &nm_i->nat_entries);
297 	spin_unlock(&nm_i->nat_list_lock);
298 
299 	set_nat_flag(ne, IS_DIRTY, false);
300 	set->entry_cnt--;
301 	nm_i->nat_cnt[DIRTY_NAT]--;
302 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
303 }
304 
305 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
306 		nid_t start, unsigned int nr, struct nat_entry_set **ep)
307 {
308 	return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
309 							start, nr);
310 }
311 
312 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
313 {
314 	return NODE_MAPPING(sbi) == page->mapping &&
315 			IS_DNODE(page) && is_cold_node(page);
316 }
317 
318 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
319 {
320 	spin_lock_init(&sbi->fsync_node_lock);
321 	INIT_LIST_HEAD(&sbi->fsync_node_list);
322 	sbi->fsync_seg_id = 0;
323 	sbi->fsync_node_num = 0;
324 }
325 
326 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
327 							struct page *page)
328 {
329 	struct fsync_node_entry *fn;
330 	unsigned long flags;
331 	unsigned int seq_id;
332 
333 	fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab,
334 					GFP_NOFS, true, NULL);
335 
336 	get_page(page);
337 	fn->page = page;
338 	INIT_LIST_HEAD(&fn->list);
339 
340 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
341 	list_add_tail(&fn->list, &sbi->fsync_node_list);
342 	fn->seq_id = sbi->fsync_seg_id++;
343 	seq_id = fn->seq_id;
344 	sbi->fsync_node_num++;
345 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
346 
347 	return seq_id;
348 }
349 
350 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
351 {
352 	struct fsync_node_entry *fn;
353 	unsigned long flags;
354 
355 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
356 	list_for_each_entry(fn, &sbi->fsync_node_list, list) {
357 		if (fn->page == page) {
358 			list_del(&fn->list);
359 			sbi->fsync_node_num--;
360 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
361 			kmem_cache_free(fsync_node_entry_slab, fn);
362 			put_page(page);
363 			return;
364 		}
365 	}
366 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
367 	f2fs_bug_on(sbi, 1);
368 }
369 
370 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
371 {
372 	unsigned long flags;
373 
374 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
375 	sbi->fsync_seg_id = 0;
376 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
377 }
378 
379 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
380 {
381 	struct f2fs_nm_info *nm_i = NM_I(sbi);
382 	struct nat_entry *e;
383 	bool need = false;
384 
385 	down_read(&nm_i->nat_tree_lock);
386 	e = __lookup_nat_cache(nm_i, nid);
387 	if (e) {
388 		if (!get_nat_flag(e, IS_CHECKPOINTED) &&
389 				!get_nat_flag(e, HAS_FSYNCED_INODE))
390 			need = true;
391 	}
392 	up_read(&nm_i->nat_tree_lock);
393 	return need;
394 }
395 
396 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
397 {
398 	struct f2fs_nm_info *nm_i = NM_I(sbi);
399 	struct nat_entry *e;
400 	bool is_cp = true;
401 
402 	down_read(&nm_i->nat_tree_lock);
403 	e = __lookup_nat_cache(nm_i, nid);
404 	if (e && !get_nat_flag(e, IS_CHECKPOINTED))
405 		is_cp = false;
406 	up_read(&nm_i->nat_tree_lock);
407 	return is_cp;
408 }
409 
410 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
411 {
412 	struct f2fs_nm_info *nm_i = NM_I(sbi);
413 	struct nat_entry *e;
414 	bool need_update = true;
415 
416 	down_read(&nm_i->nat_tree_lock);
417 	e = __lookup_nat_cache(nm_i, ino);
418 	if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
419 			(get_nat_flag(e, IS_CHECKPOINTED) ||
420 			 get_nat_flag(e, HAS_FSYNCED_INODE)))
421 		need_update = false;
422 	up_read(&nm_i->nat_tree_lock);
423 	return need_update;
424 }
425 
426 /* must be locked by nat_tree_lock */
427 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
428 						struct f2fs_nat_entry *ne)
429 {
430 	struct f2fs_nm_info *nm_i = NM_I(sbi);
431 	struct nat_entry *new, *e;
432 
433 	/* Let's mitigate lock contention of nat_tree_lock during checkpoint */
434 	if (rwsem_is_locked(&sbi->cp_global_sem))
435 		return;
436 
437 	new = __alloc_nat_entry(sbi, nid, false);
438 	if (!new)
439 		return;
440 
441 	down_write(&nm_i->nat_tree_lock);
442 	e = __lookup_nat_cache(nm_i, nid);
443 	if (!e)
444 		e = __init_nat_entry(nm_i, new, ne, false);
445 	else
446 		f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
447 				nat_get_blkaddr(e) !=
448 					le32_to_cpu(ne->block_addr) ||
449 				nat_get_version(e) != ne->version);
450 	up_write(&nm_i->nat_tree_lock);
451 	if (e != new)
452 		__free_nat_entry(new);
453 }
454 
455 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
456 			block_t new_blkaddr, bool fsync_done)
457 {
458 	struct f2fs_nm_info *nm_i = NM_I(sbi);
459 	struct nat_entry *e;
460 	struct nat_entry *new = __alloc_nat_entry(sbi, ni->nid, true);
461 
462 	down_write(&nm_i->nat_tree_lock);
463 	e = __lookup_nat_cache(nm_i, ni->nid);
464 	if (!e) {
465 		e = __init_nat_entry(nm_i, new, NULL, true);
466 		copy_node_info(&e->ni, ni);
467 		f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
468 	} else if (new_blkaddr == NEW_ADDR) {
469 		/*
470 		 * when nid is reallocated,
471 		 * previous nat entry can be remained in nat cache.
472 		 * So, reinitialize it with new information.
473 		 */
474 		copy_node_info(&e->ni, ni);
475 		f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
476 	}
477 	/* let's free early to reduce memory consumption */
478 	if (e != new)
479 		__free_nat_entry(new);
480 
481 	/* sanity check */
482 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
483 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
484 			new_blkaddr == NULL_ADDR);
485 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
486 			new_blkaddr == NEW_ADDR);
487 	f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
488 			new_blkaddr == NEW_ADDR);
489 
490 	/* increment version no as node is removed */
491 	if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
492 		unsigned char version = nat_get_version(e);
493 
494 		nat_set_version(e, inc_node_version(version));
495 	}
496 
497 	/* change address */
498 	nat_set_blkaddr(e, new_blkaddr);
499 	if (!__is_valid_data_blkaddr(new_blkaddr))
500 		set_nat_flag(e, IS_CHECKPOINTED, false);
501 	__set_nat_cache_dirty(nm_i, e);
502 
503 	/* update fsync_mark if its inode nat entry is still alive */
504 	if (ni->nid != ni->ino)
505 		e = __lookup_nat_cache(nm_i, ni->ino);
506 	if (e) {
507 		if (fsync_done && ni->nid == ni->ino)
508 			set_nat_flag(e, HAS_FSYNCED_INODE, true);
509 		set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
510 	}
511 	up_write(&nm_i->nat_tree_lock);
512 }
513 
514 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
515 {
516 	struct f2fs_nm_info *nm_i = NM_I(sbi);
517 	int nr = nr_shrink;
518 
519 	if (!down_write_trylock(&nm_i->nat_tree_lock))
520 		return 0;
521 
522 	spin_lock(&nm_i->nat_list_lock);
523 	while (nr_shrink) {
524 		struct nat_entry *ne;
525 
526 		if (list_empty(&nm_i->nat_entries))
527 			break;
528 
529 		ne = list_first_entry(&nm_i->nat_entries,
530 					struct nat_entry, list);
531 		list_del(&ne->list);
532 		spin_unlock(&nm_i->nat_list_lock);
533 
534 		__del_from_nat_cache(nm_i, ne);
535 		nr_shrink--;
536 
537 		spin_lock(&nm_i->nat_list_lock);
538 	}
539 	spin_unlock(&nm_i->nat_list_lock);
540 
541 	up_write(&nm_i->nat_tree_lock);
542 	return nr - nr_shrink;
543 }
544 
545 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
546 				struct node_info *ni, bool checkpoint_context)
547 {
548 	struct f2fs_nm_info *nm_i = NM_I(sbi);
549 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
550 	struct f2fs_journal *journal = curseg->journal;
551 	nid_t start_nid = START_NID(nid);
552 	struct f2fs_nat_block *nat_blk;
553 	struct page *page = NULL;
554 	struct f2fs_nat_entry ne;
555 	struct nat_entry *e;
556 	pgoff_t index;
557 	block_t blkaddr;
558 	int i;
559 
560 	ni->nid = nid;
561 retry:
562 	/* Check nat cache */
563 	down_read(&nm_i->nat_tree_lock);
564 	e = __lookup_nat_cache(nm_i, nid);
565 	if (e) {
566 		ni->ino = nat_get_ino(e);
567 		ni->blk_addr = nat_get_blkaddr(e);
568 		ni->version = nat_get_version(e);
569 		up_read(&nm_i->nat_tree_lock);
570 		return 0;
571 	}
572 
573 	/*
574 	 * Check current segment summary by trying to grab journal_rwsem first.
575 	 * This sem is on the critical path on the checkpoint requiring the above
576 	 * nat_tree_lock. Therefore, we should retry, if we failed to grab here
577 	 * while not bothering checkpoint.
578 	 */
579 	if (!rwsem_is_locked(&sbi->cp_global_sem) || checkpoint_context) {
580 		down_read(&curseg->journal_rwsem);
581 	} else if (rwsem_is_contended(&nm_i->nat_tree_lock) ||
582 				!down_read_trylock(&curseg->journal_rwsem)) {
583 		up_read(&nm_i->nat_tree_lock);
584 		goto retry;
585 	}
586 
587 	i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
588 	if (i >= 0) {
589 		ne = nat_in_journal(journal, i);
590 		node_info_from_raw_nat(ni, &ne);
591 	}
592 	up_read(&curseg->journal_rwsem);
593 	if (i >= 0) {
594 		up_read(&nm_i->nat_tree_lock);
595 		goto cache;
596 	}
597 
598 	/* Fill node_info from nat page */
599 	index = current_nat_addr(sbi, nid);
600 	up_read(&nm_i->nat_tree_lock);
601 
602 	page = f2fs_get_meta_page(sbi, index);
603 	if (IS_ERR(page))
604 		return PTR_ERR(page);
605 
606 	nat_blk = (struct f2fs_nat_block *)page_address(page);
607 	ne = nat_blk->entries[nid - start_nid];
608 	node_info_from_raw_nat(ni, &ne);
609 	f2fs_put_page(page, 1);
610 cache:
611 	blkaddr = le32_to_cpu(ne.block_addr);
612 	if (__is_valid_data_blkaddr(blkaddr) &&
613 		!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
614 		return -EFAULT;
615 
616 	/* cache nat entry */
617 	cache_nat_entry(sbi, nid, &ne);
618 	return 0;
619 }
620 
621 /*
622  * readahead MAX_RA_NODE number of node pages.
623  */
624 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
625 {
626 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
627 	struct blk_plug plug;
628 	int i, end;
629 	nid_t nid;
630 
631 	blk_start_plug(&plug);
632 
633 	/* Then, try readahead for siblings of the desired node */
634 	end = start + n;
635 	end = min(end, NIDS_PER_BLOCK);
636 	for (i = start; i < end; i++) {
637 		nid = get_nid(parent, i, false);
638 		f2fs_ra_node_page(sbi, nid);
639 	}
640 
641 	blk_finish_plug(&plug);
642 }
643 
644 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
645 {
646 	const long direct_index = ADDRS_PER_INODE(dn->inode);
647 	const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
648 	const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
649 	unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
650 	int cur_level = dn->cur_level;
651 	int max_level = dn->max_level;
652 	pgoff_t base = 0;
653 
654 	if (!dn->max_level)
655 		return pgofs + 1;
656 
657 	while (max_level-- > cur_level)
658 		skipped_unit *= NIDS_PER_BLOCK;
659 
660 	switch (dn->max_level) {
661 	case 3:
662 		base += 2 * indirect_blks;
663 		fallthrough;
664 	case 2:
665 		base += 2 * direct_blks;
666 		fallthrough;
667 	case 1:
668 		base += direct_index;
669 		break;
670 	default:
671 		f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
672 	}
673 
674 	return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
675 }
676 
677 /*
678  * The maximum depth is four.
679  * Offset[0] will have raw inode offset.
680  */
681 static int get_node_path(struct inode *inode, long block,
682 				int offset[4], unsigned int noffset[4])
683 {
684 	const long direct_index = ADDRS_PER_INODE(inode);
685 	const long direct_blks = ADDRS_PER_BLOCK(inode);
686 	const long dptrs_per_blk = NIDS_PER_BLOCK;
687 	const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
688 	const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
689 	int n = 0;
690 	int level = 0;
691 
692 	noffset[0] = 0;
693 
694 	if (block < direct_index) {
695 		offset[n] = block;
696 		goto got;
697 	}
698 	block -= direct_index;
699 	if (block < direct_blks) {
700 		offset[n++] = NODE_DIR1_BLOCK;
701 		noffset[n] = 1;
702 		offset[n] = block;
703 		level = 1;
704 		goto got;
705 	}
706 	block -= direct_blks;
707 	if (block < direct_blks) {
708 		offset[n++] = NODE_DIR2_BLOCK;
709 		noffset[n] = 2;
710 		offset[n] = block;
711 		level = 1;
712 		goto got;
713 	}
714 	block -= direct_blks;
715 	if (block < indirect_blks) {
716 		offset[n++] = NODE_IND1_BLOCK;
717 		noffset[n] = 3;
718 		offset[n++] = block / direct_blks;
719 		noffset[n] = 4 + offset[n - 1];
720 		offset[n] = block % direct_blks;
721 		level = 2;
722 		goto got;
723 	}
724 	block -= indirect_blks;
725 	if (block < indirect_blks) {
726 		offset[n++] = NODE_IND2_BLOCK;
727 		noffset[n] = 4 + dptrs_per_blk;
728 		offset[n++] = block / direct_blks;
729 		noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
730 		offset[n] = block % direct_blks;
731 		level = 2;
732 		goto got;
733 	}
734 	block -= indirect_blks;
735 	if (block < dindirect_blks) {
736 		offset[n++] = NODE_DIND_BLOCK;
737 		noffset[n] = 5 + (dptrs_per_blk * 2);
738 		offset[n++] = block / indirect_blks;
739 		noffset[n] = 6 + (dptrs_per_blk * 2) +
740 			      offset[n - 1] * (dptrs_per_blk + 1);
741 		offset[n++] = (block / direct_blks) % dptrs_per_blk;
742 		noffset[n] = 7 + (dptrs_per_blk * 2) +
743 			      offset[n - 2] * (dptrs_per_blk + 1) +
744 			      offset[n - 1];
745 		offset[n] = block % direct_blks;
746 		level = 3;
747 		goto got;
748 	} else {
749 		return -E2BIG;
750 	}
751 got:
752 	return level;
753 }
754 
755 /*
756  * Caller should call f2fs_put_dnode(dn).
757  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
758  * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
759  */
760 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
761 {
762 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
763 	struct page *npage[4];
764 	struct page *parent = NULL;
765 	int offset[4];
766 	unsigned int noffset[4];
767 	nid_t nids[4];
768 	int level, i = 0;
769 	int err = 0;
770 
771 	level = get_node_path(dn->inode, index, offset, noffset);
772 	if (level < 0)
773 		return level;
774 
775 	nids[0] = dn->inode->i_ino;
776 	npage[0] = dn->inode_page;
777 
778 	if (!npage[0]) {
779 		npage[0] = f2fs_get_node_page(sbi, nids[0]);
780 		if (IS_ERR(npage[0]))
781 			return PTR_ERR(npage[0]);
782 	}
783 
784 	/* if inline_data is set, should not report any block indices */
785 	if (f2fs_has_inline_data(dn->inode) && index) {
786 		err = -ENOENT;
787 		f2fs_put_page(npage[0], 1);
788 		goto release_out;
789 	}
790 
791 	parent = npage[0];
792 	if (level != 0)
793 		nids[1] = get_nid(parent, offset[0], true);
794 	dn->inode_page = npage[0];
795 	dn->inode_page_locked = true;
796 
797 	/* get indirect or direct nodes */
798 	for (i = 1; i <= level; i++) {
799 		bool done = false;
800 
801 		if (!nids[i] && mode == ALLOC_NODE) {
802 			/* alloc new node */
803 			if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
804 				err = -ENOSPC;
805 				goto release_pages;
806 			}
807 
808 			dn->nid = nids[i];
809 			npage[i] = f2fs_new_node_page(dn, noffset[i]);
810 			if (IS_ERR(npage[i])) {
811 				f2fs_alloc_nid_failed(sbi, nids[i]);
812 				err = PTR_ERR(npage[i]);
813 				goto release_pages;
814 			}
815 
816 			set_nid(parent, offset[i - 1], nids[i], i == 1);
817 			f2fs_alloc_nid_done(sbi, nids[i]);
818 			done = true;
819 		} else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
820 			npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
821 			if (IS_ERR(npage[i])) {
822 				err = PTR_ERR(npage[i]);
823 				goto release_pages;
824 			}
825 			done = true;
826 		}
827 		if (i == 1) {
828 			dn->inode_page_locked = false;
829 			unlock_page(parent);
830 		} else {
831 			f2fs_put_page(parent, 1);
832 		}
833 
834 		if (!done) {
835 			npage[i] = f2fs_get_node_page(sbi, nids[i]);
836 			if (IS_ERR(npage[i])) {
837 				err = PTR_ERR(npage[i]);
838 				f2fs_put_page(npage[0], 0);
839 				goto release_out;
840 			}
841 		}
842 		if (i < level) {
843 			parent = npage[i];
844 			nids[i + 1] = get_nid(parent, offset[i], false);
845 		}
846 	}
847 	dn->nid = nids[level];
848 	dn->ofs_in_node = offset[level];
849 	dn->node_page = npage[level];
850 	dn->data_blkaddr = f2fs_data_blkaddr(dn);
851 
852 	if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) &&
853 					f2fs_sb_has_readonly(sbi)) {
854 		unsigned int c_len = f2fs_cluster_blocks_are_contiguous(dn);
855 		block_t blkaddr;
856 
857 		if (!c_len)
858 			goto out;
859 
860 		blkaddr = f2fs_data_blkaddr(dn);
861 		if (blkaddr == COMPRESS_ADDR)
862 			blkaddr = data_blkaddr(dn->inode, dn->node_page,
863 						dn->ofs_in_node + 1);
864 
865 		f2fs_update_extent_tree_range_compressed(dn->inode,
866 					index, blkaddr,
867 					F2FS_I(dn->inode)->i_cluster_size,
868 					c_len);
869 	}
870 out:
871 	return 0;
872 
873 release_pages:
874 	f2fs_put_page(parent, 1);
875 	if (i > 1)
876 		f2fs_put_page(npage[0], 0);
877 release_out:
878 	dn->inode_page = NULL;
879 	dn->node_page = NULL;
880 	if (err == -ENOENT) {
881 		dn->cur_level = i;
882 		dn->max_level = level;
883 		dn->ofs_in_node = offset[level];
884 	}
885 	return err;
886 }
887 
888 static int truncate_node(struct dnode_of_data *dn)
889 {
890 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
891 	struct node_info ni;
892 	int err;
893 	pgoff_t index;
894 
895 	err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
896 	if (err)
897 		return err;
898 
899 	/* Deallocate node address */
900 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
901 	dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
902 	set_node_addr(sbi, &ni, NULL_ADDR, false);
903 
904 	if (dn->nid == dn->inode->i_ino) {
905 		f2fs_remove_orphan_inode(sbi, dn->nid);
906 		dec_valid_inode_count(sbi);
907 		f2fs_inode_synced(dn->inode);
908 	}
909 
910 	clear_node_page_dirty(dn->node_page);
911 	set_sbi_flag(sbi, SBI_IS_DIRTY);
912 
913 	index = dn->node_page->index;
914 	f2fs_put_page(dn->node_page, 1);
915 
916 	invalidate_mapping_pages(NODE_MAPPING(sbi),
917 			index, index);
918 
919 	dn->node_page = NULL;
920 	trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
921 
922 	return 0;
923 }
924 
925 static int truncate_dnode(struct dnode_of_data *dn)
926 {
927 	struct page *page;
928 	int err;
929 
930 	if (dn->nid == 0)
931 		return 1;
932 
933 	/* get direct node */
934 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
935 	if (PTR_ERR(page) == -ENOENT)
936 		return 1;
937 	else if (IS_ERR(page))
938 		return PTR_ERR(page);
939 
940 	/* Make dnode_of_data for parameter */
941 	dn->node_page = page;
942 	dn->ofs_in_node = 0;
943 	f2fs_truncate_data_blocks(dn);
944 	err = truncate_node(dn);
945 	if (err)
946 		return err;
947 
948 	return 1;
949 }
950 
951 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
952 						int ofs, int depth)
953 {
954 	struct dnode_of_data rdn = *dn;
955 	struct page *page;
956 	struct f2fs_node *rn;
957 	nid_t child_nid;
958 	unsigned int child_nofs;
959 	int freed = 0;
960 	int i, ret;
961 
962 	if (dn->nid == 0)
963 		return NIDS_PER_BLOCK + 1;
964 
965 	trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
966 
967 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
968 	if (IS_ERR(page)) {
969 		trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
970 		return PTR_ERR(page);
971 	}
972 
973 	f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
974 
975 	rn = F2FS_NODE(page);
976 	if (depth < 3) {
977 		for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
978 			child_nid = le32_to_cpu(rn->in.nid[i]);
979 			if (child_nid == 0)
980 				continue;
981 			rdn.nid = child_nid;
982 			ret = truncate_dnode(&rdn);
983 			if (ret < 0)
984 				goto out_err;
985 			if (set_nid(page, i, 0, false))
986 				dn->node_changed = true;
987 		}
988 	} else {
989 		child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
990 		for (i = ofs; i < NIDS_PER_BLOCK; i++) {
991 			child_nid = le32_to_cpu(rn->in.nid[i]);
992 			if (child_nid == 0) {
993 				child_nofs += NIDS_PER_BLOCK + 1;
994 				continue;
995 			}
996 			rdn.nid = child_nid;
997 			ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
998 			if (ret == (NIDS_PER_BLOCK + 1)) {
999 				if (set_nid(page, i, 0, false))
1000 					dn->node_changed = true;
1001 				child_nofs += ret;
1002 			} else if (ret < 0 && ret != -ENOENT) {
1003 				goto out_err;
1004 			}
1005 		}
1006 		freed = child_nofs;
1007 	}
1008 
1009 	if (!ofs) {
1010 		/* remove current indirect node */
1011 		dn->node_page = page;
1012 		ret = truncate_node(dn);
1013 		if (ret)
1014 			goto out_err;
1015 		freed++;
1016 	} else {
1017 		f2fs_put_page(page, 1);
1018 	}
1019 	trace_f2fs_truncate_nodes_exit(dn->inode, freed);
1020 	return freed;
1021 
1022 out_err:
1023 	f2fs_put_page(page, 1);
1024 	trace_f2fs_truncate_nodes_exit(dn->inode, ret);
1025 	return ret;
1026 }
1027 
1028 static int truncate_partial_nodes(struct dnode_of_data *dn,
1029 			struct f2fs_inode *ri, int *offset, int depth)
1030 {
1031 	struct page *pages[2];
1032 	nid_t nid[3];
1033 	nid_t child_nid;
1034 	int err = 0;
1035 	int i;
1036 	int idx = depth - 2;
1037 
1038 	nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1039 	if (!nid[0])
1040 		return 0;
1041 
1042 	/* get indirect nodes in the path */
1043 	for (i = 0; i < idx + 1; i++) {
1044 		/* reference count'll be increased */
1045 		pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
1046 		if (IS_ERR(pages[i])) {
1047 			err = PTR_ERR(pages[i]);
1048 			idx = i - 1;
1049 			goto fail;
1050 		}
1051 		nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1052 	}
1053 
1054 	f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1055 
1056 	/* free direct nodes linked to a partial indirect node */
1057 	for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1058 		child_nid = get_nid(pages[idx], i, false);
1059 		if (!child_nid)
1060 			continue;
1061 		dn->nid = child_nid;
1062 		err = truncate_dnode(dn);
1063 		if (err < 0)
1064 			goto fail;
1065 		if (set_nid(pages[idx], i, 0, false))
1066 			dn->node_changed = true;
1067 	}
1068 
1069 	if (offset[idx + 1] == 0) {
1070 		dn->node_page = pages[idx];
1071 		dn->nid = nid[idx];
1072 		err = truncate_node(dn);
1073 		if (err)
1074 			goto fail;
1075 	} else {
1076 		f2fs_put_page(pages[idx], 1);
1077 	}
1078 	offset[idx]++;
1079 	offset[idx + 1] = 0;
1080 	idx--;
1081 fail:
1082 	for (i = idx; i >= 0; i--)
1083 		f2fs_put_page(pages[i], 1);
1084 
1085 	trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1086 
1087 	return err;
1088 }
1089 
1090 /*
1091  * All the block addresses of data and nodes should be nullified.
1092  */
1093 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1094 {
1095 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1096 	int err = 0, cont = 1;
1097 	int level, offset[4], noffset[4];
1098 	unsigned int nofs = 0;
1099 	struct f2fs_inode *ri;
1100 	struct dnode_of_data dn;
1101 	struct page *page;
1102 
1103 	trace_f2fs_truncate_inode_blocks_enter(inode, from);
1104 
1105 	level = get_node_path(inode, from, offset, noffset);
1106 	if (level < 0) {
1107 		trace_f2fs_truncate_inode_blocks_exit(inode, level);
1108 		return level;
1109 	}
1110 
1111 	page = f2fs_get_node_page(sbi, inode->i_ino);
1112 	if (IS_ERR(page)) {
1113 		trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1114 		return PTR_ERR(page);
1115 	}
1116 
1117 	set_new_dnode(&dn, inode, page, NULL, 0);
1118 	unlock_page(page);
1119 
1120 	ri = F2FS_INODE(page);
1121 	switch (level) {
1122 	case 0:
1123 	case 1:
1124 		nofs = noffset[1];
1125 		break;
1126 	case 2:
1127 		nofs = noffset[1];
1128 		if (!offset[level - 1])
1129 			goto skip_partial;
1130 		err = truncate_partial_nodes(&dn, ri, offset, level);
1131 		if (err < 0 && err != -ENOENT)
1132 			goto fail;
1133 		nofs += 1 + NIDS_PER_BLOCK;
1134 		break;
1135 	case 3:
1136 		nofs = 5 + 2 * NIDS_PER_BLOCK;
1137 		if (!offset[level - 1])
1138 			goto skip_partial;
1139 		err = truncate_partial_nodes(&dn, ri, offset, level);
1140 		if (err < 0 && err != -ENOENT)
1141 			goto fail;
1142 		break;
1143 	default:
1144 		BUG();
1145 	}
1146 
1147 skip_partial:
1148 	while (cont) {
1149 		dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1150 		switch (offset[0]) {
1151 		case NODE_DIR1_BLOCK:
1152 		case NODE_DIR2_BLOCK:
1153 			err = truncate_dnode(&dn);
1154 			break;
1155 
1156 		case NODE_IND1_BLOCK:
1157 		case NODE_IND2_BLOCK:
1158 			err = truncate_nodes(&dn, nofs, offset[1], 2);
1159 			break;
1160 
1161 		case NODE_DIND_BLOCK:
1162 			err = truncate_nodes(&dn, nofs, offset[1], 3);
1163 			cont = 0;
1164 			break;
1165 
1166 		default:
1167 			BUG();
1168 		}
1169 		if (err < 0 && err != -ENOENT)
1170 			goto fail;
1171 		if (offset[1] == 0 &&
1172 				ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1173 			lock_page(page);
1174 			BUG_ON(page->mapping != NODE_MAPPING(sbi));
1175 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1176 			ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1177 			set_page_dirty(page);
1178 			unlock_page(page);
1179 		}
1180 		offset[1] = 0;
1181 		offset[0]++;
1182 		nofs += err;
1183 	}
1184 fail:
1185 	f2fs_put_page(page, 0);
1186 	trace_f2fs_truncate_inode_blocks_exit(inode, err);
1187 	return err > 0 ? 0 : err;
1188 }
1189 
1190 /* caller must lock inode page */
1191 int f2fs_truncate_xattr_node(struct inode *inode)
1192 {
1193 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1194 	nid_t nid = F2FS_I(inode)->i_xattr_nid;
1195 	struct dnode_of_data dn;
1196 	struct page *npage;
1197 	int err;
1198 
1199 	if (!nid)
1200 		return 0;
1201 
1202 	npage = f2fs_get_node_page(sbi, nid);
1203 	if (IS_ERR(npage))
1204 		return PTR_ERR(npage);
1205 
1206 	set_new_dnode(&dn, inode, NULL, npage, nid);
1207 	err = truncate_node(&dn);
1208 	if (err) {
1209 		f2fs_put_page(npage, 1);
1210 		return err;
1211 	}
1212 
1213 	f2fs_i_xnid_write(inode, 0);
1214 
1215 	return 0;
1216 }
1217 
1218 /*
1219  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1220  * f2fs_unlock_op().
1221  */
1222 int f2fs_remove_inode_page(struct inode *inode)
1223 {
1224 	struct dnode_of_data dn;
1225 	int err;
1226 
1227 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1228 	err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1229 	if (err)
1230 		return err;
1231 
1232 	err = f2fs_truncate_xattr_node(inode);
1233 	if (err) {
1234 		f2fs_put_dnode(&dn);
1235 		return err;
1236 	}
1237 
1238 	/* remove potential inline_data blocks */
1239 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1240 				S_ISLNK(inode->i_mode))
1241 		f2fs_truncate_data_blocks_range(&dn, 1);
1242 
1243 	/* 0 is possible, after f2fs_new_inode() has failed */
1244 	if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1245 		f2fs_put_dnode(&dn);
1246 		return -EIO;
1247 	}
1248 
1249 	if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1250 		f2fs_warn(F2FS_I_SB(inode),
1251 			"f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1252 			inode->i_ino, (unsigned long long)inode->i_blocks);
1253 		set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1254 	}
1255 
1256 	/* will put inode & node pages */
1257 	err = truncate_node(&dn);
1258 	if (err) {
1259 		f2fs_put_dnode(&dn);
1260 		return err;
1261 	}
1262 	return 0;
1263 }
1264 
1265 struct page *f2fs_new_inode_page(struct inode *inode)
1266 {
1267 	struct dnode_of_data dn;
1268 
1269 	/* allocate inode page for new inode */
1270 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1271 
1272 	/* caller should f2fs_put_page(page, 1); */
1273 	return f2fs_new_node_page(&dn, 0);
1274 }
1275 
1276 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1277 {
1278 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1279 	struct node_info new_ni;
1280 	struct page *page;
1281 	int err;
1282 
1283 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1284 		return ERR_PTR(-EPERM);
1285 
1286 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1287 	if (!page)
1288 		return ERR_PTR(-ENOMEM);
1289 
1290 	if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1291 		goto fail;
1292 
1293 #ifdef CONFIG_F2FS_CHECK_FS
1294 	err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false);
1295 	if (err) {
1296 		dec_valid_node_count(sbi, dn->inode, !ofs);
1297 		goto fail;
1298 	}
1299 	f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1300 #endif
1301 	new_ni.nid = dn->nid;
1302 	new_ni.ino = dn->inode->i_ino;
1303 	new_ni.blk_addr = NULL_ADDR;
1304 	new_ni.flag = 0;
1305 	new_ni.version = 0;
1306 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1307 
1308 	f2fs_wait_on_page_writeback(page, NODE, true, true);
1309 	fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1310 	set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1311 	if (!PageUptodate(page))
1312 		SetPageUptodate(page);
1313 	if (set_page_dirty(page))
1314 		dn->node_changed = true;
1315 
1316 	if (f2fs_has_xattr_block(ofs))
1317 		f2fs_i_xnid_write(dn->inode, dn->nid);
1318 
1319 	if (ofs == 0)
1320 		inc_valid_inode_count(sbi);
1321 	return page;
1322 
1323 fail:
1324 	clear_node_page_dirty(page);
1325 	f2fs_put_page(page, 1);
1326 	return ERR_PTR(err);
1327 }
1328 
1329 /*
1330  * Caller should do after getting the following values.
1331  * 0: f2fs_put_page(page, 0)
1332  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1333  */
1334 static int read_node_page(struct page *page, int op_flags)
1335 {
1336 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1337 	struct node_info ni;
1338 	struct f2fs_io_info fio = {
1339 		.sbi = sbi,
1340 		.type = NODE,
1341 		.op = REQ_OP_READ,
1342 		.op_flags = op_flags,
1343 		.page = page,
1344 		.encrypted_page = NULL,
1345 	};
1346 	int err;
1347 
1348 	if (PageUptodate(page)) {
1349 		if (!f2fs_inode_chksum_verify(sbi, page)) {
1350 			ClearPageUptodate(page);
1351 			return -EFSBADCRC;
1352 		}
1353 		return LOCKED_PAGE;
1354 	}
1355 
1356 	err = f2fs_get_node_info(sbi, page->index, &ni, false);
1357 	if (err)
1358 		return err;
1359 
1360 	/* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1361 	if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR) ||
1362 			is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1363 		ClearPageUptodate(page);
1364 		return -ENOENT;
1365 	}
1366 
1367 	fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1368 
1369 	err = f2fs_submit_page_bio(&fio);
1370 
1371 	if (!err)
1372 		f2fs_update_iostat(sbi, FS_NODE_READ_IO, F2FS_BLKSIZE);
1373 
1374 	return err;
1375 }
1376 
1377 /*
1378  * Readahead a node page
1379  */
1380 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1381 {
1382 	struct page *apage;
1383 	int err;
1384 
1385 	if (!nid)
1386 		return;
1387 	if (f2fs_check_nid_range(sbi, nid))
1388 		return;
1389 
1390 	apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1391 	if (apage)
1392 		return;
1393 
1394 	apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1395 	if (!apage)
1396 		return;
1397 
1398 	err = read_node_page(apage, REQ_RAHEAD);
1399 	f2fs_put_page(apage, err ? 1 : 0);
1400 }
1401 
1402 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1403 					struct page *parent, int start)
1404 {
1405 	struct page *page;
1406 	int err;
1407 
1408 	if (!nid)
1409 		return ERR_PTR(-ENOENT);
1410 	if (f2fs_check_nid_range(sbi, nid))
1411 		return ERR_PTR(-EINVAL);
1412 repeat:
1413 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1414 	if (!page)
1415 		return ERR_PTR(-ENOMEM);
1416 
1417 	err = read_node_page(page, 0);
1418 	if (err < 0) {
1419 		f2fs_put_page(page, 1);
1420 		return ERR_PTR(err);
1421 	} else if (err == LOCKED_PAGE) {
1422 		err = 0;
1423 		goto page_hit;
1424 	}
1425 
1426 	if (parent)
1427 		f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1428 
1429 	lock_page(page);
1430 
1431 	if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1432 		f2fs_put_page(page, 1);
1433 		goto repeat;
1434 	}
1435 
1436 	if (unlikely(!PageUptodate(page))) {
1437 		err = -EIO;
1438 		goto out_err;
1439 	}
1440 
1441 	if (!f2fs_inode_chksum_verify(sbi, page)) {
1442 		err = -EFSBADCRC;
1443 		goto out_err;
1444 	}
1445 page_hit:
1446 	if (unlikely(nid != nid_of_node(page))) {
1447 		f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1448 			  nid, nid_of_node(page), ino_of_node(page),
1449 			  ofs_of_node(page), cpver_of_node(page),
1450 			  next_blkaddr_of_node(page));
1451 		set_sbi_flag(sbi, SBI_NEED_FSCK);
1452 		err = -EINVAL;
1453 out_err:
1454 		ClearPageUptodate(page);
1455 		f2fs_put_page(page, 1);
1456 		return ERR_PTR(err);
1457 	}
1458 	return page;
1459 }
1460 
1461 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1462 {
1463 	return __get_node_page(sbi, nid, NULL, 0);
1464 }
1465 
1466 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1467 {
1468 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1469 	nid_t nid = get_nid(parent, start, false);
1470 
1471 	return __get_node_page(sbi, nid, parent, start);
1472 }
1473 
1474 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1475 {
1476 	struct inode *inode;
1477 	struct page *page;
1478 	int ret;
1479 
1480 	/* should flush inline_data before evict_inode */
1481 	inode = ilookup(sbi->sb, ino);
1482 	if (!inode)
1483 		return;
1484 
1485 	page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1486 					FGP_LOCK|FGP_NOWAIT, 0);
1487 	if (!page)
1488 		goto iput_out;
1489 
1490 	if (!PageUptodate(page))
1491 		goto page_out;
1492 
1493 	if (!PageDirty(page))
1494 		goto page_out;
1495 
1496 	if (!clear_page_dirty_for_io(page))
1497 		goto page_out;
1498 
1499 	ret = f2fs_write_inline_data(inode, page);
1500 	inode_dec_dirty_pages(inode);
1501 	f2fs_remove_dirty_inode(inode);
1502 	if (ret)
1503 		set_page_dirty(page);
1504 page_out:
1505 	f2fs_put_page(page, 1);
1506 iput_out:
1507 	iput(inode);
1508 }
1509 
1510 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1511 {
1512 	pgoff_t index;
1513 	struct pagevec pvec;
1514 	struct page *last_page = NULL;
1515 	int nr_pages;
1516 
1517 	pagevec_init(&pvec);
1518 	index = 0;
1519 
1520 	while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1521 				PAGECACHE_TAG_DIRTY))) {
1522 		int i;
1523 
1524 		for (i = 0; i < nr_pages; i++) {
1525 			struct page *page = pvec.pages[i];
1526 
1527 			if (unlikely(f2fs_cp_error(sbi))) {
1528 				f2fs_put_page(last_page, 0);
1529 				pagevec_release(&pvec);
1530 				return ERR_PTR(-EIO);
1531 			}
1532 
1533 			if (!IS_DNODE(page) || !is_cold_node(page))
1534 				continue;
1535 			if (ino_of_node(page) != ino)
1536 				continue;
1537 
1538 			lock_page(page);
1539 
1540 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1541 continue_unlock:
1542 				unlock_page(page);
1543 				continue;
1544 			}
1545 			if (ino_of_node(page) != ino)
1546 				goto continue_unlock;
1547 
1548 			if (!PageDirty(page)) {
1549 				/* someone wrote it for us */
1550 				goto continue_unlock;
1551 			}
1552 
1553 			if (last_page)
1554 				f2fs_put_page(last_page, 0);
1555 
1556 			get_page(page);
1557 			last_page = page;
1558 			unlock_page(page);
1559 		}
1560 		pagevec_release(&pvec);
1561 		cond_resched();
1562 	}
1563 	return last_page;
1564 }
1565 
1566 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1567 				struct writeback_control *wbc, bool do_balance,
1568 				enum iostat_type io_type, unsigned int *seq_id)
1569 {
1570 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1571 	nid_t nid;
1572 	struct node_info ni;
1573 	struct f2fs_io_info fio = {
1574 		.sbi = sbi,
1575 		.ino = ino_of_node(page),
1576 		.type = NODE,
1577 		.op = REQ_OP_WRITE,
1578 		.op_flags = wbc_to_write_flags(wbc),
1579 		.page = page,
1580 		.encrypted_page = NULL,
1581 		.submitted = false,
1582 		.io_type = io_type,
1583 		.io_wbc = wbc,
1584 	};
1585 	unsigned int seq;
1586 
1587 	trace_f2fs_writepage(page, NODE);
1588 
1589 	if (unlikely(f2fs_cp_error(sbi))) {
1590 		ClearPageUptodate(page);
1591 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1592 		unlock_page(page);
1593 		return 0;
1594 	}
1595 
1596 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1597 		goto redirty_out;
1598 
1599 	if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1600 			wbc->sync_mode == WB_SYNC_NONE &&
1601 			IS_DNODE(page) && is_cold_node(page))
1602 		goto redirty_out;
1603 
1604 	/* get old block addr of this node page */
1605 	nid = nid_of_node(page);
1606 	f2fs_bug_on(sbi, page->index != nid);
1607 
1608 	if (f2fs_get_node_info(sbi, nid, &ni, !do_balance))
1609 		goto redirty_out;
1610 
1611 	if (wbc->for_reclaim) {
1612 		if (!down_read_trylock(&sbi->node_write))
1613 			goto redirty_out;
1614 	} else {
1615 		down_read(&sbi->node_write);
1616 	}
1617 
1618 	/* This page is already truncated */
1619 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1620 		ClearPageUptodate(page);
1621 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1622 		up_read(&sbi->node_write);
1623 		unlock_page(page);
1624 		return 0;
1625 	}
1626 
1627 	if (__is_valid_data_blkaddr(ni.blk_addr) &&
1628 		!f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1629 					DATA_GENERIC_ENHANCE)) {
1630 		up_read(&sbi->node_write);
1631 		goto redirty_out;
1632 	}
1633 
1634 	if (atomic && !test_opt(sbi, NOBARRIER))
1635 		fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1636 
1637 	/* should add to global list before clearing PAGECACHE status */
1638 	if (f2fs_in_warm_node_list(sbi, page)) {
1639 		seq = f2fs_add_fsync_node_entry(sbi, page);
1640 		if (seq_id)
1641 			*seq_id = seq;
1642 	}
1643 
1644 	set_page_writeback(page);
1645 	ClearPageError(page);
1646 
1647 	fio.old_blkaddr = ni.blk_addr;
1648 	f2fs_do_write_node_page(nid, &fio);
1649 	set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1650 	dec_page_count(sbi, F2FS_DIRTY_NODES);
1651 	up_read(&sbi->node_write);
1652 
1653 	if (wbc->for_reclaim) {
1654 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1655 		submitted = NULL;
1656 	}
1657 
1658 	unlock_page(page);
1659 
1660 	if (unlikely(f2fs_cp_error(sbi))) {
1661 		f2fs_submit_merged_write(sbi, NODE);
1662 		submitted = NULL;
1663 	}
1664 	if (submitted)
1665 		*submitted = fio.submitted;
1666 
1667 	if (do_balance)
1668 		f2fs_balance_fs(sbi, false);
1669 	return 0;
1670 
1671 redirty_out:
1672 	redirty_page_for_writepage(wbc, page);
1673 	return AOP_WRITEPAGE_ACTIVATE;
1674 }
1675 
1676 int f2fs_move_node_page(struct page *node_page, int gc_type)
1677 {
1678 	int err = 0;
1679 
1680 	if (gc_type == FG_GC) {
1681 		struct writeback_control wbc = {
1682 			.sync_mode = WB_SYNC_ALL,
1683 			.nr_to_write = 1,
1684 			.for_reclaim = 0,
1685 		};
1686 
1687 		f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1688 
1689 		set_page_dirty(node_page);
1690 
1691 		if (!clear_page_dirty_for_io(node_page)) {
1692 			err = -EAGAIN;
1693 			goto out_page;
1694 		}
1695 
1696 		if (__write_node_page(node_page, false, NULL,
1697 					&wbc, false, FS_GC_NODE_IO, NULL)) {
1698 			err = -EAGAIN;
1699 			unlock_page(node_page);
1700 		}
1701 		goto release_page;
1702 	} else {
1703 		/* set page dirty and write it */
1704 		if (!PageWriteback(node_page))
1705 			set_page_dirty(node_page);
1706 	}
1707 out_page:
1708 	unlock_page(node_page);
1709 release_page:
1710 	f2fs_put_page(node_page, 0);
1711 	return err;
1712 }
1713 
1714 static int f2fs_write_node_page(struct page *page,
1715 				struct writeback_control *wbc)
1716 {
1717 	return __write_node_page(page, false, NULL, wbc, false,
1718 						FS_NODE_IO, NULL);
1719 }
1720 
1721 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1722 			struct writeback_control *wbc, bool atomic,
1723 			unsigned int *seq_id)
1724 {
1725 	pgoff_t index;
1726 	struct pagevec pvec;
1727 	int ret = 0;
1728 	struct page *last_page = NULL;
1729 	bool marked = false;
1730 	nid_t ino = inode->i_ino;
1731 	int nr_pages;
1732 	int nwritten = 0;
1733 
1734 	if (atomic) {
1735 		last_page = last_fsync_dnode(sbi, ino);
1736 		if (IS_ERR_OR_NULL(last_page))
1737 			return PTR_ERR_OR_ZERO(last_page);
1738 	}
1739 retry:
1740 	pagevec_init(&pvec);
1741 	index = 0;
1742 
1743 	while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1744 				PAGECACHE_TAG_DIRTY))) {
1745 		int i;
1746 
1747 		for (i = 0; i < nr_pages; i++) {
1748 			struct page *page = pvec.pages[i];
1749 			bool submitted = false;
1750 
1751 			if (unlikely(f2fs_cp_error(sbi))) {
1752 				f2fs_put_page(last_page, 0);
1753 				pagevec_release(&pvec);
1754 				ret = -EIO;
1755 				goto out;
1756 			}
1757 
1758 			if (!IS_DNODE(page) || !is_cold_node(page))
1759 				continue;
1760 			if (ino_of_node(page) != ino)
1761 				continue;
1762 
1763 			lock_page(page);
1764 
1765 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1766 continue_unlock:
1767 				unlock_page(page);
1768 				continue;
1769 			}
1770 			if (ino_of_node(page) != ino)
1771 				goto continue_unlock;
1772 
1773 			if (!PageDirty(page) && page != last_page) {
1774 				/* someone wrote it for us */
1775 				goto continue_unlock;
1776 			}
1777 
1778 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1779 
1780 			set_fsync_mark(page, 0);
1781 			set_dentry_mark(page, 0);
1782 
1783 			if (!atomic || page == last_page) {
1784 				set_fsync_mark(page, 1);
1785 				if (IS_INODE(page)) {
1786 					if (is_inode_flag_set(inode,
1787 								FI_DIRTY_INODE))
1788 						f2fs_update_inode(inode, page);
1789 					set_dentry_mark(page,
1790 						f2fs_need_dentry_mark(sbi, ino));
1791 				}
1792 				/* may be written by other thread */
1793 				if (!PageDirty(page))
1794 					set_page_dirty(page);
1795 			}
1796 
1797 			if (!clear_page_dirty_for_io(page))
1798 				goto continue_unlock;
1799 
1800 			ret = __write_node_page(page, atomic &&
1801 						page == last_page,
1802 						&submitted, wbc, true,
1803 						FS_NODE_IO, seq_id);
1804 			if (ret) {
1805 				unlock_page(page);
1806 				f2fs_put_page(last_page, 0);
1807 				break;
1808 			} else if (submitted) {
1809 				nwritten++;
1810 			}
1811 
1812 			if (page == last_page) {
1813 				f2fs_put_page(page, 0);
1814 				marked = true;
1815 				break;
1816 			}
1817 		}
1818 		pagevec_release(&pvec);
1819 		cond_resched();
1820 
1821 		if (ret || marked)
1822 			break;
1823 	}
1824 	if (!ret && atomic && !marked) {
1825 		f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1826 			   ino, last_page->index);
1827 		lock_page(last_page);
1828 		f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1829 		set_page_dirty(last_page);
1830 		unlock_page(last_page);
1831 		goto retry;
1832 	}
1833 out:
1834 	if (nwritten)
1835 		f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1836 	return ret ? -EIO : 0;
1837 }
1838 
1839 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1840 {
1841 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1842 	bool clean;
1843 
1844 	if (inode->i_ino != ino)
1845 		return 0;
1846 
1847 	if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1848 		return 0;
1849 
1850 	spin_lock(&sbi->inode_lock[DIRTY_META]);
1851 	clean = list_empty(&F2FS_I(inode)->gdirty_list);
1852 	spin_unlock(&sbi->inode_lock[DIRTY_META]);
1853 
1854 	if (clean)
1855 		return 0;
1856 
1857 	inode = igrab(inode);
1858 	if (!inode)
1859 		return 0;
1860 	return 1;
1861 }
1862 
1863 static bool flush_dirty_inode(struct page *page)
1864 {
1865 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1866 	struct inode *inode;
1867 	nid_t ino = ino_of_node(page);
1868 
1869 	inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1870 	if (!inode)
1871 		return false;
1872 
1873 	f2fs_update_inode(inode, page);
1874 	unlock_page(page);
1875 
1876 	iput(inode);
1877 	return true;
1878 }
1879 
1880 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1881 {
1882 	pgoff_t index = 0;
1883 	struct pagevec pvec;
1884 	int nr_pages;
1885 
1886 	pagevec_init(&pvec);
1887 
1888 	while ((nr_pages = pagevec_lookup_tag(&pvec,
1889 			NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1890 		int i;
1891 
1892 		for (i = 0; i < nr_pages; i++) {
1893 			struct page *page = pvec.pages[i];
1894 
1895 			if (!IS_DNODE(page))
1896 				continue;
1897 
1898 			lock_page(page);
1899 
1900 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1901 continue_unlock:
1902 				unlock_page(page);
1903 				continue;
1904 			}
1905 
1906 			if (!PageDirty(page)) {
1907 				/* someone wrote it for us */
1908 				goto continue_unlock;
1909 			}
1910 
1911 			/* flush inline_data, if it's async context. */
1912 			if (page_private_inline(page)) {
1913 				clear_page_private_inline(page);
1914 				unlock_page(page);
1915 				flush_inline_data(sbi, ino_of_node(page));
1916 				continue;
1917 			}
1918 			unlock_page(page);
1919 		}
1920 		pagevec_release(&pvec);
1921 		cond_resched();
1922 	}
1923 }
1924 
1925 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1926 				struct writeback_control *wbc,
1927 				bool do_balance, enum iostat_type io_type)
1928 {
1929 	pgoff_t index;
1930 	struct pagevec pvec;
1931 	int step = 0;
1932 	int nwritten = 0;
1933 	int ret = 0;
1934 	int nr_pages, done = 0;
1935 
1936 	pagevec_init(&pvec);
1937 
1938 next_step:
1939 	index = 0;
1940 
1941 	while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1942 			NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1943 		int i;
1944 
1945 		for (i = 0; i < nr_pages; i++) {
1946 			struct page *page = pvec.pages[i];
1947 			bool submitted = false;
1948 			bool may_dirty = true;
1949 
1950 			/* give a priority to WB_SYNC threads */
1951 			if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1952 					wbc->sync_mode == WB_SYNC_NONE) {
1953 				done = 1;
1954 				break;
1955 			}
1956 
1957 			/*
1958 			 * flushing sequence with step:
1959 			 * 0. indirect nodes
1960 			 * 1. dentry dnodes
1961 			 * 2. file dnodes
1962 			 */
1963 			if (step == 0 && IS_DNODE(page))
1964 				continue;
1965 			if (step == 1 && (!IS_DNODE(page) ||
1966 						is_cold_node(page)))
1967 				continue;
1968 			if (step == 2 && (!IS_DNODE(page) ||
1969 						!is_cold_node(page)))
1970 				continue;
1971 lock_node:
1972 			if (wbc->sync_mode == WB_SYNC_ALL)
1973 				lock_page(page);
1974 			else if (!trylock_page(page))
1975 				continue;
1976 
1977 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1978 continue_unlock:
1979 				unlock_page(page);
1980 				continue;
1981 			}
1982 
1983 			if (!PageDirty(page)) {
1984 				/* someone wrote it for us */
1985 				goto continue_unlock;
1986 			}
1987 
1988 			/* flush inline_data/inode, if it's async context. */
1989 			if (!do_balance)
1990 				goto write_node;
1991 
1992 			/* flush inline_data */
1993 			if (page_private_inline(page)) {
1994 				clear_page_private_inline(page);
1995 				unlock_page(page);
1996 				flush_inline_data(sbi, ino_of_node(page));
1997 				goto lock_node;
1998 			}
1999 
2000 			/* flush dirty inode */
2001 			if (IS_INODE(page) && may_dirty) {
2002 				may_dirty = false;
2003 				if (flush_dirty_inode(page))
2004 					goto lock_node;
2005 			}
2006 write_node:
2007 			f2fs_wait_on_page_writeback(page, NODE, true, true);
2008 
2009 			if (!clear_page_dirty_for_io(page))
2010 				goto continue_unlock;
2011 
2012 			set_fsync_mark(page, 0);
2013 			set_dentry_mark(page, 0);
2014 
2015 			ret = __write_node_page(page, false, &submitted,
2016 						wbc, do_balance, io_type, NULL);
2017 			if (ret)
2018 				unlock_page(page);
2019 			else if (submitted)
2020 				nwritten++;
2021 
2022 			if (--wbc->nr_to_write == 0)
2023 				break;
2024 		}
2025 		pagevec_release(&pvec);
2026 		cond_resched();
2027 
2028 		if (wbc->nr_to_write == 0) {
2029 			step = 2;
2030 			break;
2031 		}
2032 	}
2033 
2034 	if (step < 2) {
2035 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2036 				wbc->sync_mode == WB_SYNC_NONE && step == 1)
2037 			goto out;
2038 		step++;
2039 		goto next_step;
2040 	}
2041 out:
2042 	if (nwritten)
2043 		f2fs_submit_merged_write(sbi, NODE);
2044 
2045 	if (unlikely(f2fs_cp_error(sbi)))
2046 		return -EIO;
2047 	return ret;
2048 }
2049 
2050 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2051 						unsigned int seq_id)
2052 {
2053 	struct fsync_node_entry *fn;
2054 	struct page *page;
2055 	struct list_head *head = &sbi->fsync_node_list;
2056 	unsigned long flags;
2057 	unsigned int cur_seq_id = 0;
2058 	int ret2, ret = 0;
2059 
2060 	while (seq_id && cur_seq_id < seq_id) {
2061 		spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2062 		if (list_empty(head)) {
2063 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2064 			break;
2065 		}
2066 		fn = list_first_entry(head, struct fsync_node_entry, list);
2067 		if (fn->seq_id > seq_id) {
2068 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2069 			break;
2070 		}
2071 		cur_seq_id = fn->seq_id;
2072 		page = fn->page;
2073 		get_page(page);
2074 		spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2075 
2076 		f2fs_wait_on_page_writeback(page, NODE, true, false);
2077 		if (TestClearPageError(page))
2078 			ret = -EIO;
2079 
2080 		put_page(page);
2081 
2082 		if (ret)
2083 			break;
2084 	}
2085 
2086 	ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2087 	if (!ret)
2088 		ret = ret2;
2089 
2090 	return ret;
2091 }
2092 
2093 static int f2fs_write_node_pages(struct address_space *mapping,
2094 			    struct writeback_control *wbc)
2095 {
2096 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2097 	struct blk_plug plug;
2098 	long diff;
2099 
2100 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2101 		goto skip_write;
2102 
2103 	/* balancing f2fs's metadata in background */
2104 	f2fs_balance_fs_bg(sbi, true);
2105 
2106 	/* collect a number of dirty node pages and write together */
2107 	if (wbc->sync_mode != WB_SYNC_ALL &&
2108 			get_pages(sbi, F2FS_DIRTY_NODES) <
2109 					nr_pages_to_skip(sbi, NODE))
2110 		goto skip_write;
2111 
2112 	if (wbc->sync_mode == WB_SYNC_ALL)
2113 		atomic_inc(&sbi->wb_sync_req[NODE]);
2114 	else if (atomic_read(&sbi->wb_sync_req[NODE]))
2115 		goto skip_write;
2116 
2117 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2118 
2119 	diff = nr_pages_to_write(sbi, NODE, wbc);
2120 	blk_start_plug(&plug);
2121 	f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2122 	blk_finish_plug(&plug);
2123 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2124 
2125 	if (wbc->sync_mode == WB_SYNC_ALL)
2126 		atomic_dec(&sbi->wb_sync_req[NODE]);
2127 	return 0;
2128 
2129 skip_write:
2130 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2131 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2132 	return 0;
2133 }
2134 
2135 static int f2fs_set_node_page_dirty(struct page *page)
2136 {
2137 	trace_f2fs_set_page_dirty(page, NODE);
2138 
2139 	if (!PageUptodate(page))
2140 		SetPageUptodate(page);
2141 #ifdef CONFIG_F2FS_CHECK_FS
2142 	if (IS_INODE(page))
2143 		f2fs_inode_chksum_set(F2FS_P_SB(page), page);
2144 #endif
2145 	if (!PageDirty(page)) {
2146 		__set_page_dirty_nobuffers(page);
2147 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
2148 		set_page_private_reference(page);
2149 		return 1;
2150 	}
2151 	return 0;
2152 }
2153 
2154 /*
2155  * Structure of the f2fs node operations
2156  */
2157 const struct address_space_operations f2fs_node_aops = {
2158 	.writepage	= f2fs_write_node_page,
2159 	.writepages	= f2fs_write_node_pages,
2160 	.set_page_dirty	= f2fs_set_node_page_dirty,
2161 	.invalidatepage	= f2fs_invalidate_page,
2162 	.releasepage	= f2fs_release_page,
2163 #ifdef CONFIG_MIGRATION
2164 	.migratepage	= f2fs_migrate_page,
2165 #endif
2166 };
2167 
2168 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2169 						nid_t n)
2170 {
2171 	return radix_tree_lookup(&nm_i->free_nid_root, n);
2172 }
2173 
2174 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2175 				struct free_nid *i)
2176 {
2177 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2178 	int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2179 
2180 	if (err)
2181 		return err;
2182 
2183 	nm_i->nid_cnt[FREE_NID]++;
2184 	list_add_tail(&i->list, &nm_i->free_nid_list);
2185 	return 0;
2186 }
2187 
2188 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2189 			struct free_nid *i, enum nid_state state)
2190 {
2191 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2192 
2193 	f2fs_bug_on(sbi, state != i->state);
2194 	nm_i->nid_cnt[state]--;
2195 	if (state == FREE_NID)
2196 		list_del(&i->list);
2197 	radix_tree_delete(&nm_i->free_nid_root, i->nid);
2198 }
2199 
2200 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2201 			enum nid_state org_state, enum nid_state dst_state)
2202 {
2203 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2204 
2205 	f2fs_bug_on(sbi, org_state != i->state);
2206 	i->state = dst_state;
2207 	nm_i->nid_cnt[org_state]--;
2208 	nm_i->nid_cnt[dst_state]++;
2209 
2210 	switch (dst_state) {
2211 	case PREALLOC_NID:
2212 		list_del(&i->list);
2213 		break;
2214 	case FREE_NID:
2215 		list_add_tail(&i->list, &nm_i->free_nid_list);
2216 		break;
2217 	default:
2218 		BUG_ON(1);
2219 	}
2220 }
2221 
2222 bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi)
2223 {
2224 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2225 	unsigned int i;
2226 	bool ret = true;
2227 
2228 	down_read(&nm_i->nat_tree_lock);
2229 	for (i = 0; i < nm_i->nat_blocks; i++) {
2230 		if (!test_bit_le(i, nm_i->nat_block_bitmap)) {
2231 			ret = false;
2232 			break;
2233 		}
2234 	}
2235 	up_read(&nm_i->nat_tree_lock);
2236 
2237 	return ret;
2238 }
2239 
2240 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2241 							bool set, bool build)
2242 {
2243 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2244 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2245 	unsigned int nid_ofs = nid - START_NID(nid);
2246 
2247 	if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2248 		return;
2249 
2250 	if (set) {
2251 		if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2252 			return;
2253 		__set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2254 		nm_i->free_nid_count[nat_ofs]++;
2255 	} else {
2256 		if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2257 			return;
2258 		__clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2259 		if (!build)
2260 			nm_i->free_nid_count[nat_ofs]--;
2261 	}
2262 }
2263 
2264 /* return if the nid is recognized as free */
2265 static bool add_free_nid(struct f2fs_sb_info *sbi,
2266 				nid_t nid, bool build, bool update)
2267 {
2268 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2269 	struct free_nid *i, *e;
2270 	struct nat_entry *ne;
2271 	int err = -EINVAL;
2272 	bool ret = false;
2273 
2274 	/* 0 nid should not be used */
2275 	if (unlikely(nid == 0))
2276 		return false;
2277 
2278 	if (unlikely(f2fs_check_nid_range(sbi, nid)))
2279 		return false;
2280 
2281 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL);
2282 	i->nid = nid;
2283 	i->state = FREE_NID;
2284 
2285 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2286 
2287 	spin_lock(&nm_i->nid_list_lock);
2288 
2289 	if (build) {
2290 		/*
2291 		 *   Thread A             Thread B
2292 		 *  - f2fs_create
2293 		 *   - f2fs_new_inode
2294 		 *    - f2fs_alloc_nid
2295 		 *     - __insert_nid_to_list(PREALLOC_NID)
2296 		 *                     - f2fs_balance_fs_bg
2297 		 *                      - f2fs_build_free_nids
2298 		 *                       - __f2fs_build_free_nids
2299 		 *                        - scan_nat_page
2300 		 *                         - add_free_nid
2301 		 *                          - __lookup_nat_cache
2302 		 *  - f2fs_add_link
2303 		 *   - f2fs_init_inode_metadata
2304 		 *    - f2fs_new_inode_page
2305 		 *     - f2fs_new_node_page
2306 		 *      - set_node_addr
2307 		 *  - f2fs_alloc_nid_done
2308 		 *   - __remove_nid_from_list(PREALLOC_NID)
2309 		 *                         - __insert_nid_to_list(FREE_NID)
2310 		 */
2311 		ne = __lookup_nat_cache(nm_i, nid);
2312 		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2313 				nat_get_blkaddr(ne) != NULL_ADDR))
2314 			goto err_out;
2315 
2316 		e = __lookup_free_nid_list(nm_i, nid);
2317 		if (e) {
2318 			if (e->state == FREE_NID)
2319 				ret = true;
2320 			goto err_out;
2321 		}
2322 	}
2323 	ret = true;
2324 	err = __insert_free_nid(sbi, i);
2325 err_out:
2326 	if (update) {
2327 		update_free_nid_bitmap(sbi, nid, ret, build);
2328 		if (!build)
2329 			nm_i->available_nids++;
2330 	}
2331 	spin_unlock(&nm_i->nid_list_lock);
2332 	radix_tree_preload_end();
2333 
2334 	if (err)
2335 		kmem_cache_free(free_nid_slab, i);
2336 	return ret;
2337 }
2338 
2339 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2340 {
2341 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2342 	struct free_nid *i;
2343 	bool need_free = false;
2344 
2345 	spin_lock(&nm_i->nid_list_lock);
2346 	i = __lookup_free_nid_list(nm_i, nid);
2347 	if (i && i->state == FREE_NID) {
2348 		__remove_free_nid(sbi, i, FREE_NID);
2349 		need_free = true;
2350 	}
2351 	spin_unlock(&nm_i->nid_list_lock);
2352 
2353 	if (need_free)
2354 		kmem_cache_free(free_nid_slab, i);
2355 }
2356 
2357 static int scan_nat_page(struct f2fs_sb_info *sbi,
2358 			struct page *nat_page, nid_t start_nid)
2359 {
2360 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2361 	struct f2fs_nat_block *nat_blk = page_address(nat_page);
2362 	block_t blk_addr;
2363 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2364 	int i;
2365 
2366 	__set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2367 
2368 	i = start_nid % NAT_ENTRY_PER_BLOCK;
2369 
2370 	for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2371 		if (unlikely(start_nid >= nm_i->max_nid))
2372 			break;
2373 
2374 		blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2375 
2376 		if (blk_addr == NEW_ADDR)
2377 			return -EINVAL;
2378 
2379 		if (blk_addr == NULL_ADDR) {
2380 			add_free_nid(sbi, start_nid, true, true);
2381 		} else {
2382 			spin_lock(&NM_I(sbi)->nid_list_lock);
2383 			update_free_nid_bitmap(sbi, start_nid, false, true);
2384 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2385 		}
2386 	}
2387 
2388 	return 0;
2389 }
2390 
2391 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2392 {
2393 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2394 	struct f2fs_journal *journal = curseg->journal;
2395 	int i;
2396 
2397 	down_read(&curseg->journal_rwsem);
2398 	for (i = 0; i < nats_in_cursum(journal); i++) {
2399 		block_t addr;
2400 		nid_t nid;
2401 
2402 		addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2403 		nid = le32_to_cpu(nid_in_journal(journal, i));
2404 		if (addr == NULL_ADDR)
2405 			add_free_nid(sbi, nid, true, false);
2406 		else
2407 			remove_free_nid(sbi, nid);
2408 	}
2409 	up_read(&curseg->journal_rwsem);
2410 }
2411 
2412 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2413 {
2414 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2415 	unsigned int i, idx;
2416 	nid_t nid;
2417 
2418 	down_read(&nm_i->nat_tree_lock);
2419 
2420 	for (i = 0; i < nm_i->nat_blocks; i++) {
2421 		if (!test_bit_le(i, nm_i->nat_block_bitmap))
2422 			continue;
2423 		if (!nm_i->free_nid_count[i])
2424 			continue;
2425 		for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2426 			idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2427 						NAT_ENTRY_PER_BLOCK, idx);
2428 			if (idx >= NAT_ENTRY_PER_BLOCK)
2429 				break;
2430 
2431 			nid = i * NAT_ENTRY_PER_BLOCK + idx;
2432 			add_free_nid(sbi, nid, true, false);
2433 
2434 			if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2435 				goto out;
2436 		}
2437 	}
2438 out:
2439 	scan_curseg_cache(sbi);
2440 
2441 	up_read(&nm_i->nat_tree_lock);
2442 }
2443 
2444 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2445 						bool sync, bool mount)
2446 {
2447 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2448 	int i = 0, ret;
2449 	nid_t nid = nm_i->next_scan_nid;
2450 
2451 	if (unlikely(nid >= nm_i->max_nid))
2452 		nid = 0;
2453 
2454 	if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2455 		nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2456 
2457 	/* Enough entries */
2458 	if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2459 		return 0;
2460 
2461 	if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2462 		return 0;
2463 
2464 	if (!mount) {
2465 		/* try to find free nids in free_nid_bitmap */
2466 		scan_free_nid_bits(sbi);
2467 
2468 		if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2469 			return 0;
2470 	}
2471 
2472 	/* readahead nat pages to be scanned */
2473 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2474 							META_NAT, true);
2475 
2476 	down_read(&nm_i->nat_tree_lock);
2477 
2478 	while (1) {
2479 		if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2480 						nm_i->nat_block_bitmap)) {
2481 			struct page *page = get_current_nat_page(sbi, nid);
2482 
2483 			if (IS_ERR(page)) {
2484 				ret = PTR_ERR(page);
2485 			} else {
2486 				ret = scan_nat_page(sbi, page, nid);
2487 				f2fs_put_page(page, 1);
2488 			}
2489 
2490 			if (ret) {
2491 				up_read(&nm_i->nat_tree_lock);
2492 				f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2493 				return ret;
2494 			}
2495 		}
2496 
2497 		nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2498 		if (unlikely(nid >= nm_i->max_nid))
2499 			nid = 0;
2500 
2501 		if (++i >= FREE_NID_PAGES)
2502 			break;
2503 	}
2504 
2505 	/* go to the next free nat pages to find free nids abundantly */
2506 	nm_i->next_scan_nid = nid;
2507 
2508 	/* find free nids from current sum_pages */
2509 	scan_curseg_cache(sbi);
2510 
2511 	up_read(&nm_i->nat_tree_lock);
2512 
2513 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2514 					nm_i->ra_nid_pages, META_NAT, false);
2515 
2516 	return 0;
2517 }
2518 
2519 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2520 {
2521 	int ret;
2522 
2523 	mutex_lock(&NM_I(sbi)->build_lock);
2524 	ret = __f2fs_build_free_nids(sbi, sync, mount);
2525 	mutex_unlock(&NM_I(sbi)->build_lock);
2526 
2527 	return ret;
2528 }
2529 
2530 /*
2531  * If this function returns success, caller can obtain a new nid
2532  * from second parameter of this function.
2533  * The returned nid could be used ino as well as nid when inode is created.
2534  */
2535 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2536 {
2537 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2538 	struct free_nid *i = NULL;
2539 retry:
2540 	if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2541 		f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2542 		return false;
2543 	}
2544 
2545 	spin_lock(&nm_i->nid_list_lock);
2546 
2547 	if (unlikely(nm_i->available_nids == 0)) {
2548 		spin_unlock(&nm_i->nid_list_lock);
2549 		return false;
2550 	}
2551 
2552 	/* We should not use stale free nids created by f2fs_build_free_nids */
2553 	if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2554 		f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2555 		i = list_first_entry(&nm_i->free_nid_list,
2556 					struct free_nid, list);
2557 		*nid = i->nid;
2558 
2559 		__move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2560 		nm_i->available_nids--;
2561 
2562 		update_free_nid_bitmap(sbi, *nid, false, false);
2563 
2564 		spin_unlock(&nm_i->nid_list_lock);
2565 		return true;
2566 	}
2567 	spin_unlock(&nm_i->nid_list_lock);
2568 
2569 	/* Let's scan nat pages and its caches to get free nids */
2570 	if (!f2fs_build_free_nids(sbi, true, false))
2571 		goto retry;
2572 	return false;
2573 }
2574 
2575 /*
2576  * f2fs_alloc_nid() should be called prior to this function.
2577  */
2578 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2579 {
2580 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2581 	struct free_nid *i;
2582 
2583 	spin_lock(&nm_i->nid_list_lock);
2584 	i = __lookup_free_nid_list(nm_i, nid);
2585 	f2fs_bug_on(sbi, !i);
2586 	__remove_free_nid(sbi, i, PREALLOC_NID);
2587 	spin_unlock(&nm_i->nid_list_lock);
2588 
2589 	kmem_cache_free(free_nid_slab, i);
2590 }
2591 
2592 /*
2593  * f2fs_alloc_nid() should be called prior to this function.
2594  */
2595 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2596 {
2597 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2598 	struct free_nid *i;
2599 	bool need_free = false;
2600 
2601 	if (!nid)
2602 		return;
2603 
2604 	spin_lock(&nm_i->nid_list_lock);
2605 	i = __lookup_free_nid_list(nm_i, nid);
2606 	f2fs_bug_on(sbi, !i);
2607 
2608 	if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2609 		__remove_free_nid(sbi, i, PREALLOC_NID);
2610 		need_free = true;
2611 	} else {
2612 		__move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2613 	}
2614 
2615 	nm_i->available_nids++;
2616 
2617 	update_free_nid_bitmap(sbi, nid, true, false);
2618 
2619 	spin_unlock(&nm_i->nid_list_lock);
2620 
2621 	if (need_free)
2622 		kmem_cache_free(free_nid_slab, i);
2623 }
2624 
2625 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2626 {
2627 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2628 	int nr = nr_shrink;
2629 
2630 	if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2631 		return 0;
2632 
2633 	if (!mutex_trylock(&nm_i->build_lock))
2634 		return 0;
2635 
2636 	while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2637 		struct free_nid *i, *next;
2638 		unsigned int batch = SHRINK_NID_BATCH_SIZE;
2639 
2640 		spin_lock(&nm_i->nid_list_lock);
2641 		list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2642 			if (!nr_shrink || !batch ||
2643 				nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2644 				break;
2645 			__remove_free_nid(sbi, i, FREE_NID);
2646 			kmem_cache_free(free_nid_slab, i);
2647 			nr_shrink--;
2648 			batch--;
2649 		}
2650 		spin_unlock(&nm_i->nid_list_lock);
2651 	}
2652 
2653 	mutex_unlock(&nm_i->build_lock);
2654 
2655 	return nr - nr_shrink;
2656 }
2657 
2658 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2659 {
2660 	void *src_addr, *dst_addr;
2661 	size_t inline_size;
2662 	struct page *ipage;
2663 	struct f2fs_inode *ri;
2664 
2665 	ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2666 	if (IS_ERR(ipage))
2667 		return PTR_ERR(ipage);
2668 
2669 	ri = F2FS_INODE(page);
2670 	if (ri->i_inline & F2FS_INLINE_XATTR) {
2671 		if (!f2fs_has_inline_xattr(inode)) {
2672 			set_inode_flag(inode, FI_INLINE_XATTR);
2673 			stat_inc_inline_xattr(inode);
2674 		}
2675 	} else {
2676 		if (f2fs_has_inline_xattr(inode)) {
2677 			stat_dec_inline_xattr(inode);
2678 			clear_inode_flag(inode, FI_INLINE_XATTR);
2679 		}
2680 		goto update_inode;
2681 	}
2682 
2683 	dst_addr = inline_xattr_addr(inode, ipage);
2684 	src_addr = inline_xattr_addr(inode, page);
2685 	inline_size = inline_xattr_size(inode);
2686 
2687 	f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2688 	memcpy(dst_addr, src_addr, inline_size);
2689 update_inode:
2690 	f2fs_update_inode(inode, ipage);
2691 	f2fs_put_page(ipage, 1);
2692 	return 0;
2693 }
2694 
2695 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2696 {
2697 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2698 	nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2699 	nid_t new_xnid;
2700 	struct dnode_of_data dn;
2701 	struct node_info ni;
2702 	struct page *xpage;
2703 	int err;
2704 
2705 	if (!prev_xnid)
2706 		goto recover_xnid;
2707 
2708 	/* 1: invalidate the previous xattr nid */
2709 	err = f2fs_get_node_info(sbi, prev_xnid, &ni, false);
2710 	if (err)
2711 		return err;
2712 
2713 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
2714 	dec_valid_node_count(sbi, inode, false);
2715 	set_node_addr(sbi, &ni, NULL_ADDR, false);
2716 
2717 recover_xnid:
2718 	/* 2: update xattr nid in inode */
2719 	if (!f2fs_alloc_nid(sbi, &new_xnid))
2720 		return -ENOSPC;
2721 
2722 	set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2723 	xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2724 	if (IS_ERR(xpage)) {
2725 		f2fs_alloc_nid_failed(sbi, new_xnid);
2726 		return PTR_ERR(xpage);
2727 	}
2728 
2729 	f2fs_alloc_nid_done(sbi, new_xnid);
2730 	f2fs_update_inode_page(inode);
2731 
2732 	/* 3: update and set xattr node page dirty */
2733 	memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2734 
2735 	set_page_dirty(xpage);
2736 	f2fs_put_page(xpage, 1);
2737 
2738 	return 0;
2739 }
2740 
2741 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2742 {
2743 	struct f2fs_inode *src, *dst;
2744 	nid_t ino = ino_of_node(page);
2745 	struct node_info old_ni, new_ni;
2746 	struct page *ipage;
2747 	int err;
2748 
2749 	err = f2fs_get_node_info(sbi, ino, &old_ni, false);
2750 	if (err)
2751 		return err;
2752 
2753 	if (unlikely(old_ni.blk_addr != NULL_ADDR))
2754 		return -EINVAL;
2755 retry:
2756 	ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2757 	if (!ipage) {
2758 		memalloc_retry_wait(GFP_NOFS);
2759 		goto retry;
2760 	}
2761 
2762 	/* Should not use this inode from free nid list */
2763 	remove_free_nid(sbi, ino);
2764 
2765 	if (!PageUptodate(ipage))
2766 		SetPageUptodate(ipage);
2767 	fill_node_footer(ipage, ino, ino, 0, true);
2768 	set_cold_node(ipage, false);
2769 
2770 	src = F2FS_INODE(page);
2771 	dst = F2FS_INODE(ipage);
2772 
2773 	memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2774 	dst->i_size = 0;
2775 	dst->i_blocks = cpu_to_le64(1);
2776 	dst->i_links = cpu_to_le32(1);
2777 	dst->i_xattr_nid = 0;
2778 	dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2779 	if (dst->i_inline & F2FS_EXTRA_ATTR) {
2780 		dst->i_extra_isize = src->i_extra_isize;
2781 
2782 		if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2783 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2784 							i_inline_xattr_size))
2785 			dst->i_inline_xattr_size = src->i_inline_xattr_size;
2786 
2787 		if (f2fs_sb_has_project_quota(sbi) &&
2788 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2789 								i_projid))
2790 			dst->i_projid = src->i_projid;
2791 
2792 		if (f2fs_sb_has_inode_crtime(sbi) &&
2793 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2794 							i_crtime_nsec)) {
2795 			dst->i_crtime = src->i_crtime;
2796 			dst->i_crtime_nsec = src->i_crtime_nsec;
2797 		}
2798 	}
2799 
2800 	new_ni = old_ni;
2801 	new_ni.ino = ino;
2802 
2803 	if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2804 		WARN_ON(1);
2805 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2806 	inc_valid_inode_count(sbi);
2807 	set_page_dirty(ipage);
2808 	f2fs_put_page(ipage, 1);
2809 	return 0;
2810 }
2811 
2812 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2813 			unsigned int segno, struct f2fs_summary_block *sum)
2814 {
2815 	struct f2fs_node *rn;
2816 	struct f2fs_summary *sum_entry;
2817 	block_t addr;
2818 	int i, idx, last_offset, nrpages;
2819 
2820 	/* scan the node segment */
2821 	last_offset = sbi->blocks_per_seg;
2822 	addr = START_BLOCK(sbi, segno);
2823 	sum_entry = &sum->entries[0];
2824 
2825 	for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2826 		nrpages = bio_max_segs(last_offset - i);
2827 
2828 		/* readahead node pages */
2829 		f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2830 
2831 		for (idx = addr; idx < addr + nrpages; idx++) {
2832 			struct page *page = f2fs_get_tmp_page(sbi, idx);
2833 
2834 			if (IS_ERR(page))
2835 				return PTR_ERR(page);
2836 
2837 			rn = F2FS_NODE(page);
2838 			sum_entry->nid = rn->footer.nid;
2839 			sum_entry->version = 0;
2840 			sum_entry->ofs_in_node = 0;
2841 			sum_entry++;
2842 			f2fs_put_page(page, 1);
2843 		}
2844 
2845 		invalidate_mapping_pages(META_MAPPING(sbi), addr,
2846 							addr + nrpages);
2847 	}
2848 	return 0;
2849 }
2850 
2851 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2852 {
2853 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2854 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2855 	struct f2fs_journal *journal = curseg->journal;
2856 	int i;
2857 
2858 	down_write(&curseg->journal_rwsem);
2859 	for (i = 0; i < nats_in_cursum(journal); i++) {
2860 		struct nat_entry *ne;
2861 		struct f2fs_nat_entry raw_ne;
2862 		nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2863 
2864 		if (f2fs_check_nid_range(sbi, nid))
2865 			continue;
2866 
2867 		raw_ne = nat_in_journal(journal, i);
2868 
2869 		ne = __lookup_nat_cache(nm_i, nid);
2870 		if (!ne) {
2871 			ne = __alloc_nat_entry(sbi, nid, true);
2872 			__init_nat_entry(nm_i, ne, &raw_ne, true);
2873 		}
2874 
2875 		/*
2876 		 * if a free nat in journal has not been used after last
2877 		 * checkpoint, we should remove it from available nids,
2878 		 * since later we will add it again.
2879 		 */
2880 		if (!get_nat_flag(ne, IS_DIRTY) &&
2881 				le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2882 			spin_lock(&nm_i->nid_list_lock);
2883 			nm_i->available_nids--;
2884 			spin_unlock(&nm_i->nid_list_lock);
2885 		}
2886 
2887 		__set_nat_cache_dirty(nm_i, ne);
2888 	}
2889 	update_nats_in_cursum(journal, -i);
2890 	up_write(&curseg->journal_rwsem);
2891 }
2892 
2893 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2894 						struct list_head *head, int max)
2895 {
2896 	struct nat_entry_set *cur;
2897 
2898 	if (nes->entry_cnt >= max)
2899 		goto add_out;
2900 
2901 	list_for_each_entry(cur, head, set_list) {
2902 		if (cur->entry_cnt >= nes->entry_cnt) {
2903 			list_add(&nes->set_list, cur->set_list.prev);
2904 			return;
2905 		}
2906 	}
2907 add_out:
2908 	list_add_tail(&nes->set_list, head);
2909 }
2910 
2911 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs,
2912 							unsigned int valid)
2913 {
2914 	if (valid == 0) {
2915 		__set_bit_le(nat_ofs, nm_i->empty_nat_bits);
2916 		__clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2917 		return;
2918 	}
2919 
2920 	__clear_bit_le(nat_ofs, nm_i->empty_nat_bits);
2921 	if (valid == NAT_ENTRY_PER_BLOCK)
2922 		__set_bit_le(nat_ofs, nm_i->full_nat_bits);
2923 	else
2924 		__clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2925 }
2926 
2927 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2928 						struct page *page)
2929 {
2930 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2931 	unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2932 	struct f2fs_nat_block *nat_blk = page_address(page);
2933 	int valid = 0;
2934 	int i = 0;
2935 
2936 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
2937 		return;
2938 
2939 	if (nat_index == 0) {
2940 		valid = 1;
2941 		i = 1;
2942 	}
2943 	for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2944 		if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2945 			valid++;
2946 	}
2947 
2948 	__update_nat_bits(nm_i, nat_index, valid);
2949 }
2950 
2951 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi)
2952 {
2953 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2954 	unsigned int nat_ofs;
2955 
2956 	down_read(&nm_i->nat_tree_lock);
2957 
2958 	for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) {
2959 		unsigned int valid = 0, nid_ofs = 0;
2960 
2961 		/* handle nid zero due to it should never be used */
2962 		if (unlikely(nat_ofs == 0)) {
2963 			valid = 1;
2964 			nid_ofs = 1;
2965 		}
2966 
2967 		for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) {
2968 			if (!test_bit_le(nid_ofs,
2969 					nm_i->free_nid_bitmap[nat_ofs]))
2970 				valid++;
2971 		}
2972 
2973 		__update_nat_bits(nm_i, nat_ofs, valid);
2974 	}
2975 
2976 	up_read(&nm_i->nat_tree_lock);
2977 }
2978 
2979 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2980 		struct nat_entry_set *set, struct cp_control *cpc)
2981 {
2982 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2983 	struct f2fs_journal *journal = curseg->journal;
2984 	nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2985 	bool to_journal = true;
2986 	struct f2fs_nat_block *nat_blk;
2987 	struct nat_entry *ne, *cur;
2988 	struct page *page = NULL;
2989 
2990 	/*
2991 	 * there are two steps to flush nat entries:
2992 	 * #1, flush nat entries to journal in current hot data summary block.
2993 	 * #2, flush nat entries to nat page.
2994 	 */
2995 	if ((cpc->reason & CP_UMOUNT) ||
2996 		!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2997 		to_journal = false;
2998 
2999 	if (to_journal) {
3000 		down_write(&curseg->journal_rwsem);
3001 	} else {
3002 		page = get_next_nat_page(sbi, start_nid);
3003 		if (IS_ERR(page))
3004 			return PTR_ERR(page);
3005 
3006 		nat_blk = page_address(page);
3007 		f2fs_bug_on(sbi, !nat_blk);
3008 	}
3009 
3010 	/* flush dirty nats in nat entry set */
3011 	list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
3012 		struct f2fs_nat_entry *raw_ne;
3013 		nid_t nid = nat_get_nid(ne);
3014 		int offset;
3015 
3016 		f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
3017 
3018 		if (to_journal) {
3019 			offset = f2fs_lookup_journal_in_cursum(journal,
3020 							NAT_JOURNAL, nid, 1);
3021 			f2fs_bug_on(sbi, offset < 0);
3022 			raw_ne = &nat_in_journal(journal, offset);
3023 			nid_in_journal(journal, offset) = cpu_to_le32(nid);
3024 		} else {
3025 			raw_ne = &nat_blk->entries[nid - start_nid];
3026 		}
3027 		raw_nat_from_node_info(raw_ne, &ne->ni);
3028 		nat_reset_flag(ne);
3029 		__clear_nat_cache_dirty(NM_I(sbi), set, ne);
3030 		if (nat_get_blkaddr(ne) == NULL_ADDR) {
3031 			add_free_nid(sbi, nid, false, true);
3032 		} else {
3033 			spin_lock(&NM_I(sbi)->nid_list_lock);
3034 			update_free_nid_bitmap(sbi, nid, false, false);
3035 			spin_unlock(&NM_I(sbi)->nid_list_lock);
3036 		}
3037 	}
3038 
3039 	if (to_journal) {
3040 		up_write(&curseg->journal_rwsem);
3041 	} else {
3042 		update_nat_bits(sbi, start_nid, page);
3043 		f2fs_put_page(page, 1);
3044 	}
3045 
3046 	/* Allow dirty nats by node block allocation in write_begin */
3047 	if (!set->entry_cnt) {
3048 		radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
3049 		kmem_cache_free(nat_entry_set_slab, set);
3050 	}
3051 	return 0;
3052 }
3053 
3054 /*
3055  * This function is called during the checkpointing process.
3056  */
3057 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3058 {
3059 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3060 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3061 	struct f2fs_journal *journal = curseg->journal;
3062 	struct nat_entry_set *setvec[SETVEC_SIZE];
3063 	struct nat_entry_set *set, *tmp;
3064 	unsigned int found;
3065 	nid_t set_idx = 0;
3066 	LIST_HEAD(sets);
3067 	int err = 0;
3068 
3069 	/*
3070 	 * during unmount, let's flush nat_bits before checking
3071 	 * nat_cnt[DIRTY_NAT].
3072 	 */
3073 	if (cpc->reason & CP_UMOUNT) {
3074 		down_write(&nm_i->nat_tree_lock);
3075 		remove_nats_in_journal(sbi);
3076 		up_write(&nm_i->nat_tree_lock);
3077 	}
3078 
3079 	if (!nm_i->nat_cnt[DIRTY_NAT])
3080 		return 0;
3081 
3082 	down_write(&nm_i->nat_tree_lock);
3083 
3084 	/*
3085 	 * if there are no enough space in journal to store dirty nat
3086 	 * entries, remove all entries from journal and merge them
3087 	 * into nat entry set.
3088 	 */
3089 	if (cpc->reason & CP_UMOUNT ||
3090 		!__has_cursum_space(journal,
3091 			nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3092 		remove_nats_in_journal(sbi);
3093 
3094 	while ((found = __gang_lookup_nat_set(nm_i,
3095 					set_idx, SETVEC_SIZE, setvec))) {
3096 		unsigned idx;
3097 
3098 		set_idx = setvec[found - 1]->set + 1;
3099 		for (idx = 0; idx < found; idx++)
3100 			__adjust_nat_entry_set(setvec[idx], &sets,
3101 						MAX_NAT_JENTRIES(journal));
3102 	}
3103 
3104 	/* flush dirty nats in nat entry set */
3105 	list_for_each_entry_safe(set, tmp, &sets, set_list) {
3106 		err = __flush_nat_entry_set(sbi, set, cpc);
3107 		if (err)
3108 			break;
3109 	}
3110 
3111 	up_write(&nm_i->nat_tree_lock);
3112 	/* Allow dirty nats by node block allocation in write_begin */
3113 
3114 	return err;
3115 }
3116 
3117 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3118 {
3119 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3120 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3121 	unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3122 	unsigned int i;
3123 	__u64 cp_ver = cur_cp_version(ckpt);
3124 	block_t nat_bits_addr;
3125 
3126 	nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3127 	nm_i->nat_bits = f2fs_kvzalloc(sbi,
3128 			nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3129 	if (!nm_i->nat_bits)
3130 		return -ENOMEM;
3131 
3132 	nm_i->full_nat_bits = nm_i->nat_bits + 8;
3133 	nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3134 
3135 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3136 		return 0;
3137 
3138 	nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3139 						nm_i->nat_bits_blocks;
3140 	for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3141 		struct page *page;
3142 
3143 		page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3144 		if (IS_ERR(page))
3145 			return PTR_ERR(page);
3146 
3147 		memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3148 					page_address(page), F2FS_BLKSIZE);
3149 		f2fs_put_page(page, 1);
3150 	}
3151 
3152 	cp_ver |= (cur_cp_crc(ckpt) << 32);
3153 	if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3154 		clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
3155 		f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)",
3156 			cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits));
3157 		return 0;
3158 	}
3159 
3160 	f2fs_notice(sbi, "Found nat_bits in checkpoint");
3161 	return 0;
3162 }
3163 
3164 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3165 {
3166 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3167 	unsigned int i = 0;
3168 	nid_t nid, last_nid;
3169 
3170 	if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3171 		return;
3172 
3173 	for (i = 0; i < nm_i->nat_blocks; i++) {
3174 		i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3175 		if (i >= nm_i->nat_blocks)
3176 			break;
3177 
3178 		__set_bit_le(i, nm_i->nat_block_bitmap);
3179 
3180 		nid = i * NAT_ENTRY_PER_BLOCK;
3181 		last_nid = nid + NAT_ENTRY_PER_BLOCK;
3182 
3183 		spin_lock(&NM_I(sbi)->nid_list_lock);
3184 		for (; nid < last_nid; nid++)
3185 			update_free_nid_bitmap(sbi, nid, true, true);
3186 		spin_unlock(&NM_I(sbi)->nid_list_lock);
3187 	}
3188 
3189 	for (i = 0; i < nm_i->nat_blocks; i++) {
3190 		i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3191 		if (i >= nm_i->nat_blocks)
3192 			break;
3193 
3194 		__set_bit_le(i, nm_i->nat_block_bitmap);
3195 	}
3196 }
3197 
3198 static int init_node_manager(struct f2fs_sb_info *sbi)
3199 {
3200 	struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3201 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3202 	unsigned char *version_bitmap;
3203 	unsigned int nat_segs;
3204 	int err;
3205 
3206 	nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3207 
3208 	/* segment_count_nat includes pair segment so divide to 2. */
3209 	nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3210 	nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3211 	nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3212 
3213 	/* not used nids: 0, node, meta, (and root counted as valid node) */
3214 	nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3215 						F2FS_RESERVED_NODE_NUM;
3216 	nm_i->nid_cnt[FREE_NID] = 0;
3217 	nm_i->nid_cnt[PREALLOC_NID] = 0;
3218 	nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3219 	nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3220 	nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3221 
3222 	INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3223 	INIT_LIST_HEAD(&nm_i->free_nid_list);
3224 	INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3225 	INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3226 	INIT_LIST_HEAD(&nm_i->nat_entries);
3227 	spin_lock_init(&nm_i->nat_list_lock);
3228 
3229 	mutex_init(&nm_i->build_lock);
3230 	spin_lock_init(&nm_i->nid_list_lock);
3231 	init_rwsem(&nm_i->nat_tree_lock);
3232 
3233 	nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3234 	nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3235 	version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3236 	nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3237 					GFP_KERNEL);
3238 	if (!nm_i->nat_bitmap)
3239 		return -ENOMEM;
3240 
3241 	err = __get_nat_bitmaps(sbi);
3242 	if (err)
3243 		return err;
3244 
3245 #ifdef CONFIG_F2FS_CHECK_FS
3246 	nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3247 					GFP_KERNEL);
3248 	if (!nm_i->nat_bitmap_mir)
3249 		return -ENOMEM;
3250 #endif
3251 
3252 	return 0;
3253 }
3254 
3255 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3256 {
3257 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3258 	int i;
3259 
3260 	nm_i->free_nid_bitmap =
3261 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3262 					      nm_i->nat_blocks),
3263 			      GFP_KERNEL);
3264 	if (!nm_i->free_nid_bitmap)
3265 		return -ENOMEM;
3266 
3267 	for (i = 0; i < nm_i->nat_blocks; i++) {
3268 		nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3269 			f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3270 		if (!nm_i->free_nid_bitmap[i])
3271 			return -ENOMEM;
3272 	}
3273 
3274 	nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3275 								GFP_KERNEL);
3276 	if (!nm_i->nat_block_bitmap)
3277 		return -ENOMEM;
3278 
3279 	nm_i->free_nid_count =
3280 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3281 					      nm_i->nat_blocks),
3282 			      GFP_KERNEL);
3283 	if (!nm_i->free_nid_count)
3284 		return -ENOMEM;
3285 	return 0;
3286 }
3287 
3288 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3289 {
3290 	int err;
3291 
3292 	sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3293 							GFP_KERNEL);
3294 	if (!sbi->nm_info)
3295 		return -ENOMEM;
3296 
3297 	err = init_node_manager(sbi);
3298 	if (err)
3299 		return err;
3300 
3301 	err = init_free_nid_cache(sbi);
3302 	if (err)
3303 		return err;
3304 
3305 	/* load free nid status from nat_bits table */
3306 	load_free_nid_bitmap(sbi);
3307 
3308 	return f2fs_build_free_nids(sbi, true, true);
3309 }
3310 
3311 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3312 {
3313 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3314 	struct free_nid *i, *next_i;
3315 	struct nat_entry *natvec[NATVEC_SIZE];
3316 	struct nat_entry_set *setvec[SETVEC_SIZE];
3317 	nid_t nid = 0;
3318 	unsigned int found;
3319 
3320 	if (!nm_i)
3321 		return;
3322 
3323 	/* destroy free nid list */
3324 	spin_lock(&nm_i->nid_list_lock);
3325 	list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3326 		__remove_free_nid(sbi, i, FREE_NID);
3327 		spin_unlock(&nm_i->nid_list_lock);
3328 		kmem_cache_free(free_nid_slab, i);
3329 		spin_lock(&nm_i->nid_list_lock);
3330 	}
3331 	f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3332 	f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3333 	f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3334 	spin_unlock(&nm_i->nid_list_lock);
3335 
3336 	/* destroy nat cache */
3337 	down_write(&nm_i->nat_tree_lock);
3338 	while ((found = __gang_lookup_nat_cache(nm_i,
3339 					nid, NATVEC_SIZE, natvec))) {
3340 		unsigned idx;
3341 
3342 		nid = nat_get_nid(natvec[found - 1]) + 1;
3343 		for (idx = 0; idx < found; idx++) {
3344 			spin_lock(&nm_i->nat_list_lock);
3345 			list_del(&natvec[idx]->list);
3346 			spin_unlock(&nm_i->nat_list_lock);
3347 
3348 			__del_from_nat_cache(nm_i, natvec[idx]);
3349 		}
3350 	}
3351 	f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3352 
3353 	/* destroy nat set cache */
3354 	nid = 0;
3355 	while ((found = __gang_lookup_nat_set(nm_i,
3356 					nid, SETVEC_SIZE, setvec))) {
3357 		unsigned idx;
3358 
3359 		nid = setvec[found - 1]->set + 1;
3360 		for (idx = 0; idx < found; idx++) {
3361 			/* entry_cnt is not zero, when cp_error was occurred */
3362 			f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3363 			radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3364 			kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3365 		}
3366 	}
3367 	up_write(&nm_i->nat_tree_lock);
3368 
3369 	kvfree(nm_i->nat_block_bitmap);
3370 	if (nm_i->free_nid_bitmap) {
3371 		int i;
3372 
3373 		for (i = 0; i < nm_i->nat_blocks; i++)
3374 			kvfree(nm_i->free_nid_bitmap[i]);
3375 		kvfree(nm_i->free_nid_bitmap);
3376 	}
3377 	kvfree(nm_i->free_nid_count);
3378 
3379 	kvfree(nm_i->nat_bitmap);
3380 	kvfree(nm_i->nat_bits);
3381 #ifdef CONFIG_F2FS_CHECK_FS
3382 	kvfree(nm_i->nat_bitmap_mir);
3383 #endif
3384 	sbi->nm_info = NULL;
3385 	kfree(nm_i);
3386 }
3387 
3388 int __init f2fs_create_node_manager_caches(void)
3389 {
3390 	nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3391 			sizeof(struct nat_entry));
3392 	if (!nat_entry_slab)
3393 		goto fail;
3394 
3395 	free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3396 			sizeof(struct free_nid));
3397 	if (!free_nid_slab)
3398 		goto destroy_nat_entry;
3399 
3400 	nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3401 			sizeof(struct nat_entry_set));
3402 	if (!nat_entry_set_slab)
3403 		goto destroy_free_nid;
3404 
3405 	fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3406 			sizeof(struct fsync_node_entry));
3407 	if (!fsync_node_entry_slab)
3408 		goto destroy_nat_entry_set;
3409 	return 0;
3410 
3411 destroy_nat_entry_set:
3412 	kmem_cache_destroy(nat_entry_set_slab);
3413 destroy_free_nid:
3414 	kmem_cache_destroy(free_nid_slab);
3415 destroy_nat_entry:
3416 	kmem_cache_destroy(nat_entry_slab);
3417 fail:
3418 	return -ENOMEM;
3419 }
3420 
3421 void f2fs_destroy_node_manager_caches(void)
3422 {
3423 	kmem_cache_destroy(fsync_node_entry_slab);
3424 	kmem_cache_destroy(nat_entry_set_slab);
3425 	kmem_cache_destroy(free_nid_slab);
3426 	kmem_cache_destroy(nat_entry_slab);
3427 }
3428