xref: /openbmc/linux/fs/f2fs/node.c (revision 2891f2d5)
1 /*
2  * fs/f2fs/node.c
3  *
4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5  *             http://www.samsung.com/
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 as
9  * published by the Free Software Foundation.
10  */
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/mpage.h>
14 #include <linux/backing-dev.h>
15 #include <linux/blkdev.h>
16 #include <linux/pagevec.h>
17 #include <linux/swap.h>
18 
19 #include "f2fs.h"
20 #include "node.h"
21 #include "segment.h"
22 #include "trace.h"
23 #include <trace/events/f2fs.h>
24 
25 #define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
26 
27 static struct kmem_cache *nat_entry_slab;
28 static struct kmem_cache *free_nid_slab;
29 static struct kmem_cache *nat_entry_set_slab;
30 
31 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
32 {
33 	struct f2fs_nm_info *nm_i = NM_I(sbi);
34 	struct sysinfo val;
35 	unsigned long avail_ram;
36 	unsigned long mem_size = 0;
37 	bool res = false;
38 
39 	si_meminfo(&val);
40 
41 	/* only uses low memory */
42 	avail_ram = val.totalram - val.totalhigh;
43 
44 	/*
45 	 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
46 	 */
47 	if (type == FREE_NIDS) {
48 		mem_size = (nm_i->nid_cnt[FREE_NID_LIST] *
49 				sizeof(struct free_nid)) >> PAGE_SHIFT;
50 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
51 	} else if (type == NAT_ENTRIES) {
52 		mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
53 							PAGE_SHIFT;
54 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
55 		if (excess_cached_nats(sbi))
56 			res = false;
57 	} else if (type == DIRTY_DENTS) {
58 		if (sbi->sb->s_bdi->wb.dirty_exceeded)
59 			return false;
60 		mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
61 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
62 	} else if (type == INO_ENTRIES) {
63 		int i;
64 
65 		for (i = 0; i <= UPDATE_INO; i++)
66 			mem_size += (sbi->im[i].ino_num *
67 				sizeof(struct ino_entry)) >> PAGE_SHIFT;
68 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
69 	} else if (type == EXTENT_CACHE) {
70 		mem_size = (atomic_read(&sbi->total_ext_tree) *
71 				sizeof(struct extent_tree) +
72 				atomic_read(&sbi->total_ext_node) *
73 				sizeof(struct extent_node)) >> PAGE_SHIFT;
74 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
75 	} else {
76 		if (!sbi->sb->s_bdi->wb.dirty_exceeded)
77 			return true;
78 	}
79 	return res;
80 }
81 
82 static void clear_node_page_dirty(struct page *page)
83 {
84 	struct address_space *mapping = page->mapping;
85 	unsigned int long flags;
86 
87 	if (PageDirty(page)) {
88 		spin_lock_irqsave(&mapping->tree_lock, flags);
89 		radix_tree_tag_clear(&mapping->page_tree,
90 				page_index(page),
91 				PAGECACHE_TAG_DIRTY);
92 		spin_unlock_irqrestore(&mapping->tree_lock, flags);
93 
94 		clear_page_dirty_for_io(page);
95 		dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
96 	}
97 	ClearPageUptodate(page);
98 }
99 
100 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
101 {
102 	pgoff_t index = current_nat_addr(sbi, nid);
103 	return get_meta_page(sbi, index);
104 }
105 
106 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
107 {
108 	struct page *src_page;
109 	struct page *dst_page;
110 	pgoff_t src_off;
111 	pgoff_t dst_off;
112 	void *src_addr;
113 	void *dst_addr;
114 	struct f2fs_nm_info *nm_i = NM_I(sbi);
115 
116 	src_off = current_nat_addr(sbi, nid);
117 	dst_off = next_nat_addr(sbi, src_off);
118 
119 	/* get current nat block page with lock */
120 	src_page = get_meta_page(sbi, src_off);
121 	dst_page = grab_meta_page(sbi, dst_off);
122 	f2fs_bug_on(sbi, PageDirty(src_page));
123 
124 	src_addr = page_address(src_page);
125 	dst_addr = page_address(dst_page);
126 	memcpy(dst_addr, src_addr, PAGE_SIZE);
127 	set_page_dirty(dst_page);
128 	f2fs_put_page(src_page, 1);
129 
130 	set_to_next_nat(nm_i, nid);
131 
132 	return dst_page;
133 }
134 
135 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
136 {
137 	return radix_tree_lookup(&nm_i->nat_root, n);
138 }
139 
140 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
141 		nid_t start, unsigned int nr, struct nat_entry **ep)
142 {
143 	return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
144 }
145 
146 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
147 {
148 	list_del(&e->list);
149 	radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
150 	nm_i->nat_cnt--;
151 	kmem_cache_free(nat_entry_slab, e);
152 }
153 
154 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
155 						struct nat_entry *ne)
156 {
157 	nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
158 	struct nat_entry_set *head;
159 
160 	if (get_nat_flag(ne, IS_DIRTY))
161 		return;
162 
163 	head = radix_tree_lookup(&nm_i->nat_set_root, set);
164 	if (!head) {
165 		head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
166 
167 		INIT_LIST_HEAD(&head->entry_list);
168 		INIT_LIST_HEAD(&head->set_list);
169 		head->set = set;
170 		head->entry_cnt = 0;
171 		f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
172 	}
173 	list_move_tail(&ne->list, &head->entry_list);
174 	nm_i->dirty_nat_cnt++;
175 	head->entry_cnt++;
176 	set_nat_flag(ne, IS_DIRTY, true);
177 }
178 
179 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
180 						struct nat_entry *ne)
181 {
182 	nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
183 	struct nat_entry_set *head;
184 
185 	head = radix_tree_lookup(&nm_i->nat_set_root, set);
186 	if (head) {
187 		list_move_tail(&ne->list, &nm_i->nat_entries);
188 		set_nat_flag(ne, IS_DIRTY, false);
189 		head->entry_cnt--;
190 		nm_i->dirty_nat_cnt--;
191 	}
192 }
193 
194 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
195 		nid_t start, unsigned int nr, struct nat_entry_set **ep)
196 {
197 	return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
198 							start, nr);
199 }
200 
201 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
202 {
203 	struct f2fs_nm_info *nm_i = NM_I(sbi);
204 	struct nat_entry *e;
205 	bool need = false;
206 
207 	down_read(&nm_i->nat_tree_lock);
208 	e = __lookup_nat_cache(nm_i, nid);
209 	if (e) {
210 		if (!get_nat_flag(e, IS_CHECKPOINTED) &&
211 				!get_nat_flag(e, HAS_FSYNCED_INODE))
212 			need = true;
213 	}
214 	up_read(&nm_i->nat_tree_lock);
215 	return need;
216 }
217 
218 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
219 {
220 	struct f2fs_nm_info *nm_i = NM_I(sbi);
221 	struct nat_entry *e;
222 	bool is_cp = true;
223 
224 	down_read(&nm_i->nat_tree_lock);
225 	e = __lookup_nat_cache(nm_i, nid);
226 	if (e && !get_nat_flag(e, IS_CHECKPOINTED))
227 		is_cp = false;
228 	up_read(&nm_i->nat_tree_lock);
229 	return is_cp;
230 }
231 
232 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
233 {
234 	struct f2fs_nm_info *nm_i = NM_I(sbi);
235 	struct nat_entry *e;
236 	bool need_update = true;
237 
238 	down_read(&nm_i->nat_tree_lock);
239 	e = __lookup_nat_cache(nm_i, ino);
240 	if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
241 			(get_nat_flag(e, IS_CHECKPOINTED) ||
242 			 get_nat_flag(e, HAS_FSYNCED_INODE)))
243 		need_update = false;
244 	up_read(&nm_i->nat_tree_lock);
245 	return need_update;
246 }
247 
248 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
249 								bool no_fail)
250 {
251 	struct nat_entry *new;
252 
253 	if (no_fail) {
254 		new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
255 		f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
256 	} else {
257 		new = kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
258 		if (!new)
259 			return NULL;
260 		if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
261 			kmem_cache_free(nat_entry_slab, new);
262 			return NULL;
263 		}
264 	}
265 
266 	memset(new, 0, sizeof(struct nat_entry));
267 	nat_set_nid(new, nid);
268 	nat_reset_flag(new);
269 	list_add_tail(&new->list, &nm_i->nat_entries);
270 	nm_i->nat_cnt++;
271 	return new;
272 }
273 
274 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
275 						struct f2fs_nat_entry *ne)
276 {
277 	struct f2fs_nm_info *nm_i = NM_I(sbi);
278 	struct nat_entry *e;
279 
280 	e = __lookup_nat_cache(nm_i, nid);
281 	if (!e) {
282 		e = grab_nat_entry(nm_i, nid, false);
283 		if (e)
284 			node_info_from_raw_nat(&e->ni, ne);
285 	} else {
286 		f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
287 				nat_get_blkaddr(e) !=
288 					le32_to_cpu(ne->block_addr) ||
289 				nat_get_version(e) != ne->version);
290 	}
291 }
292 
293 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
294 			block_t new_blkaddr, bool fsync_done)
295 {
296 	struct f2fs_nm_info *nm_i = NM_I(sbi);
297 	struct nat_entry *e;
298 
299 	down_write(&nm_i->nat_tree_lock);
300 	e = __lookup_nat_cache(nm_i, ni->nid);
301 	if (!e) {
302 		e = grab_nat_entry(nm_i, ni->nid, true);
303 		copy_node_info(&e->ni, ni);
304 		f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
305 	} else if (new_blkaddr == NEW_ADDR) {
306 		/*
307 		 * when nid is reallocated,
308 		 * previous nat entry can be remained in nat cache.
309 		 * So, reinitialize it with new information.
310 		 */
311 		copy_node_info(&e->ni, ni);
312 		f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
313 	}
314 
315 	/* sanity check */
316 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
317 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
318 			new_blkaddr == NULL_ADDR);
319 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
320 			new_blkaddr == NEW_ADDR);
321 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
322 			nat_get_blkaddr(e) != NULL_ADDR &&
323 			new_blkaddr == NEW_ADDR);
324 
325 	/* increment version no as node is removed */
326 	if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
327 		unsigned char version = nat_get_version(e);
328 		nat_set_version(e, inc_node_version(version));
329 
330 		/* in order to reuse the nid */
331 		if (nm_i->next_scan_nid > ni->nid)
332 			nm_i->next_scan_nid = ni->nid;
333 	}
334 
335 	/* change address */
336 	nat_set_blkaddr(e, new_blkaddr);
337 	if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
338 		set_nat_flag(e, IS_CHECKPOINTED, false);
339 	__set_nat_cache_dirty(nm_i, e);
340 
341 	/* update fsync_mark if its inode nat entry is still alive */
342 	if (ni->nid != ni->ino)
343 		e = __lookup_nat_cache(nm_i, ni->ino);
344 	if (e) {
345 		if (fsync_done && ni->nid == ni->ino)
346 			set_nat_flag(e, HAS_FSYNCED_INODE, true);
347 		set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
348 	}
349 	up_write(&nm_i->nat_tree_lock);
350 }
351 
352 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
353 {
354 	struct f2fs_nm_info *nm_i = NM_I(sbi);
355 	int nr = nr_shrink;
356 
357 	if (!down_write_trylock(&nm_i->nat_tree_lock))
358 		return 0;
359 
360 	while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
361 		struct nat_entry *ne;
362 		ne = list_first_entry(&nm_i->nat_entries,
363 					struct nat_entry, list);
364 		__del_from_nat_cache(nm_i, ne);
365 		nr_shrink--;
366 	}
367 	up_write(&nm_i->nat_tree_lock);
368 	return nr - nr_shrink;
369 }
370 
371 /*
372  * This function always returns success
373  */
374 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
375 {
376 	struct f2fs_nm_info *nm_i = NM_I(sbi);
377 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
378 	struct f2fs_journal *journal = curseg->journal;
379 	nid_t start_nid = START_NID(nid);
380 	struct f2fs_nat_block *nat_blk;
381 	struct page *page = NULL;
382 	struct f2fs_nat_entry ne;
383 	struct nat_entry *e;
384 	int i;
385 
386 	ni->nid = nid;
387 
388 	/* Check nat cache */
389 	down_read(&nm_i->nat_tree_lock);
390 	e = __lookup_nat_cache(nm_i, nid);
391 	if (e) {
392 		ni->ino = nat_get_ino(e);
393 		ni->blk_addr = nat_get_blkaddr(e);
394 		ni->version = nat_get_version(e);
395 		up_read(&nm_i->nat_tree_lock);
396 		return;
397 	}
398 
399 	memset(&ne, 0, sizeof(struct f2fs_nat_entry));
400 
401 	/* Check current segment summary */
402 	down_read(&curseg->journal_rwsem);
403 	i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
404 	if (i >= 0) {
405 		ne = nat_in_journal(journal, i);
406 		node_info_from_raw_nat(ni, &ne);
407 	}
408 	up_read(&curseg->journal_rwsem);
409 	if (i >= 0)
410 		goto cache;
411 
412 	/* Fill node_info from nat page */
413 	page = get_current_nat_page(sbi, start_nid);
414 	nat_blk = (struct f2fs_nat_block *)page_address(page);
415 	ne = nat_blk->entries[nid - start_nid];
416 	node_info_from_raw_nat(ni, &ne);
417 	f2fs_put_page(page, 1);
418 cache:
419 	up_read(&nm_i->nat_tree_lock);
420 	/* cache nat entry */
421 	down_write(&nm_i->nat_tree_lock);
422 	cache_nat_entry(sbi, nid, &ne);
423 	up_write(&nm_i->nat_tree_lock);
424 }
425 
426 /*
427  * readahead MAX_RA_NODE number of node pages.
428  */
429 static void ra_node_pages(struct page *parent, int start, int n)
430 {
431 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
432 	struct blk_plug plug;
433 	int i, end;
434 	nid_t nid;
435 
436 	blk_start_plug(&plug);
437 
438 	/* Then, try readahead for siblings of the desired node */
439 	end = start + n;
440 	end = min(end, NIDS_PER_BLOCK);
441 	for (i = start; i < end; i++) {
442 		nid = get_nid(parent, i, false);
443 		ra_node_page(sbi, nid);
444 	}
445 
446 	blk_finish_plug(&plug);
447 }
448 
449 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
450 {
451 	const long direct_index = ADDRS_PER_INODE(dn->inode);
452 	const long direct_blks = ADDRS_PER_BLOCK;
453 	const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
454 	unsigned int skipped_unit = ADDRS_PER_BLOCK;
455 	int cur_level = dn->cur_level;
456 	int max_level = dn->max_level;
457 	pgoff_t base = 0;
458 
459 	if (!dn->max_level)
460 		return pgofs + 1;
461 
462 	while (max_level-- > cur_level)
463 		skipped_unit *= NIDS_PER_BLOCK;
464 
465 	switch (dn->max_level) {
466 	case 3:
467 		base += 2 * indirect_blks;
468 	case 2:
469 		base += 2 * direct_blks;
470 	case 1:
471 		base += direct_index;
472 		break;
473 	default:
474 		f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
475 	}
476 
477 	return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
478 }
479 
480 /*
481  * The maximum depth is four.
482  * Offset[0] will have raw inode offset.
483  */
484 static int get_node_path(struct inode *inode, long block,
485 				int offset[4], unsigned int noffset[4])
486 {
487 	const long direct_index = ADDRS_PER_INODE(inode);
488 	const long direct_blks = ADDRS_PER_BLOCK;
489 	const long dptrs_per_blk = NIDS_PER_BLOCK;
490 	const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
491 	const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
492 	int n = 0;
493 	int level = 0;
494 
495 	noffset[0] = 0;
496 
497 	if (block < direct_index) {
498 		offset[n] = block;
499 		goto got;
500 	}
501 	block -= direct_index;
502 	if (block < direct_blks) {
503 		offset[n++] = NODE_DIR1_BLOCK;
504 		noffset[n] = 1;
505 		offset[n] = block;
506 		level = 1;
507 		goto got;
508 	}
509 	block -= direct_blks;
510 	if (block < direct_blks) {
511 		offset[n++] = NODE_DIR2_BLOCK;
512 		noffset[n] = 2;
513 		offset[n] = block;
514 		level = 1;
515 		goto got;
516 	}
517 	block -= direct_blks;
518 	if (block < indirect_blks) {
519 		offset[n++] = NODE_IND1_BLOCK;
520 		noffset[n] = 3;
521 		offset[n++] = block / direct_blks;
522 		noffset[n] = 4 + offset[n - 1];
523 		offset[n] = block % direct_blks;
524 		level = 2;
525 		goto got;
526 	}
527 	block -= indirect_blks;
528 	if (block < indirect_blks) {
529 		offset[n++] = NODE_IND2_BLOCK;
530 		noffset[n] = 4 + dptrs_per_blk;
531 		offset[n++] = block / direct_blks;
532 		noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
533 		offset[n] = block % direct_blks;
534 		level = 2;
535 		goto got;
536 	}
537 	block -= indirect_blks;
538 	if (block < dindirect_blks) {
539 		offset[n++] = NODE_DIND_BLOCK;
540 		noffset[n] = 5 + (dptrs_per_blk * 2);
541 		offset[n++] = block / indirect_blks;
542 		noffset[n] = 6 + (dptrs_per_blk * 2) +
543 			      offset[n - 1] * (dptrs_per_blk + 1);
544 		offset[n++] = (block / direct_blks) % dptrs_per_blk;
545 		noffset[n] = 7 + (dptrs_per_blk * 2) +
546 			      offset[n - 2] * (dptrs_per_blk + 1) +
547 			      offset[n - 1];
548 		offset[n] = block % direct_blks;
549 		level = 3;
550 		goto got;
551 	} else {
552 		BUG();
553 	}
554 got:
555 	return level;
556 }
557 
558 /*
559  * Caller should call f2fs_put_dnode(dn).
560  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
561  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
562  * In the case of RDONLY_NODE, we don't need to care about mutex.
563  */
564 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
565 {
566 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
567 	struct page *npage[4];
568 	struct page *parent = NULL;
569 	int offset[4];
570 	unsigned int noffset[4];
571 	nid_t nids[4];
572 	int level, i = 0;
573 	int err = 0;
574 
575 	level = get_node_path(dn->inode, index, offset, noffset);
576 
577 	nids[0] = dn->inode->i_ino;
578 	npage[0] = dn->inode_page;
579 
580 	if (!npage[0]) {
581 		npage[0] = get_node_page(sbi, nids[0]);
582 		if (IS_ERR(npage[0]))
583 			return PTR_ERR(npage[0]);
584 	}
585 
586 	/* if inline_data is set, should not report any block indices */
587 	if (f2fs_has_inline_data(dn->inode) && index) {
588 		err = -ENOENT;
589 		f2fs_put_page(npage[0], 1);
590 		goto release_out;
591 	}
592 
593 	parent = npage[0];
594 	if (level != 0)
595 		nids[1] = get_nid(parent, offset[0], true);
596 	dn->inode_page = npage[0];
597 	dn->inode_page_locked = true;
598 
599 	/* get indirect or direct nodes */
600 	for (i = 1; i <= level; i++) {
601 		bool done = false;
602 
603 		if (!nids[i] && mode == ALLOC_NODE) {
604 			/* alloc new node */
605 			if (!alloc_nid(sbi, &(nids[i]))) {
606 				err = -ENOSPC;
607 				goto release_pages;
608 			}
609 
610 			dn->nid = nids[i];
611 			npage[i] = new_node_page(dn, noffset[i], NULL);
612 			if (IS_ERR(npage[i])) {
613 				alloc_nid_failed(sbi, nids[i]);
614 				err = PTR_ERR(npage[i]);
615 				goto release_pages;
616 			}
617 
618 			set_nid(parent, offset[i - 1], nids[i], i == 1);
619 			alloc_nid_done(sbi, nids[i]);
620 			done = true;
621 		} else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
622 			npage[i] = get_node_page_ra(parent, offset[i - 1]);
623 			if (IS_ERR(npage[i])) {
624 				err = PTR_ERR(npage[i]);
625 				goto release_pages;
626 			}
627 			done = true;
628 		}
629 		if (i == 1) {
630 			dn->inode_page_locked = false;
631 			unlock_page(parent);
632 		} else {
633 			f2fs_put_page(parent, 1);
634 		}
635 
636 		if (!done) {
637 			npage[i] = get_node_page(sbi, nids[i]);
638 			if (IS_ERR(npage[i])) {
639 				err = PTR_ERR(npage[i]);
640 				f2fs_put_page(npage[0], 0);
641 				goto release_out;
642 			}
643 		}
644 		if (i < level) {
645 			parent = npage[i];
646 			nids[i + 1] = get_nid(parent, offset[i], false);
647 		}
648 	}
649 	dn->nid = nids[level];
650 	dn->ofs_in_node = offset[level];
651 	dn->node_page = npage[level];
652 	dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
653 	return 0;
654 
655 release_pages:
656 	f2fs_put_page(parent, 1);
657 	if (i > 1)
658 		f2fs_put_page(npage[0], 0);
659 release_out:
660 	dn->inode_page = NULL;
661 	dn->node_page = NULL;
662 	if (err == -ENOENT) {
663 		dn->cur_level = i;
664 		dn->max_level = level;
665 		dn->ofs_in_node = offset[level];
666 	}
667 	return err;
668 }
669 
670 static void truncate_node(struct dnode_of_data *dn)
671 {
672 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
673 	struct node_info ni;
674 
675 	get_node_info(sbi, dn->nid, &ni);
676 	if (dn->inode->i_blocks == 0) {
677 		f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
678 		goto invalidate;
679 	}
680 	f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
681 
682 	/* Deallocate node address */
683 	invalidate_blocks(sbi, ni.blk_addr);
684 	dec_valid_node_count(sbi, dn->inode);
685 	set_node_addr(sbi, &ni, NULL_ADDR, false);
686 
687 	if (dn->nid == dn->inode->i_ino) {
688 		remove_orphan_inode(sbi, dn->nid);
689 		dec_valid_inode_count(sbi);
690 		f2fs_inode_synced(dn->inode);
691 	}
692 invalidate:
693 	clear_node_page_dirty(dn->node_page);
694 	set_sbi_flag(sbi, SBI_IS_DIRTY);
695 
696 	f2fs_put_page(dn->node_page, 1);
697 
698 	invalidate_mapping_pages(NODE_MAPPING(sbi),
699 			dn->node_page->index, dn->node_page->index);
700 
701 	dn->node_page = NULL;
702 	trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
703 }
704 
705 static int truncate_dnode(struct dnode_of_data *dn)
706 {
707 	struct page *page;
708 
709 	if (dn->nid == 0)
710 		return 1;
711 
712 	/* get direct node */
713 	page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
714 	if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
715 		return 1;
716 	else if (IS_ERR(page))
717 		return PTR_ERR(page);
718 
719 	/* Make dnode_of_data for parameter */
720 	dn->node_page = page;
721 	dn->ofs_in_node = 0;
722 	truncate_data_blocks(dn);
723 	truncate_node(dn);
724 	return 1;
725 }
726 
727 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
728 						int ofs, int depth)
729 {
730 	struct dnode_of_data rdn = *dn;
731 	struct page *page;
732 	struct f2fs_node *rn;
733 	nid_t child_nid;
734 	unsigned int child_nofs;
735 	int freed = 0;
736 	int i, ret;
737 
738 	if (dn->nid == 0)
739 		return NIDS_PER_BLOCK + 1;
740 
741 	trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
742 
743 	page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
744 	if (IS_ERR(page)) {
745 		trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
746 		return PTR_ERR(page);
747 	}
748 
749 	ra_node_pages(page, ofs, NIDS_PER_BLOCK);
750 
751 	rn = F2FS_NODE(page);
752 	if (depth < 3) {
753 		for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
754 			child_nid = le32_to_cpu(rn->in.nid[i]);
755 			if (child_nid == 0)
756 				continue;
757 			rdn.nid = child_nid;
758 			ret = truncate_dnode(&rdn);
759 			if (ret < 0)
760 				goto out_err;
761 			if (set_nid(page, i, 0, false))
762 				dn->node_changed = true;
763 		}
764 	} else {
765 		child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
766 		for (i = ofs; i < NIDS_PER_BLOCK; i++) {
767 			child_nid = le32_to_cpu(rn->in.nid[i]);
768 			if (child_nid == 0) {
769 				child_nofs += NIDS_PER_BLOCK + 1;
770 				continue;
771 			}
772 			rdn.nid = child_nid;
773 			ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
774 			if (ret == (NIDS_PER_BLOCK + 1)) {
775 				if (set_nid(page, i, 0, false))
776 					dn->node_changed = true;
777 				child_nofs += ret;
778 			} else if (ret < 0 && ret != -ENOENT) {
779 				goto out_err;
780 			}
781 		}
782 		freed = child_nofs;
783 	}
784 
785 	if (!ofs) {
786 		/* remove current indirect node */
787 		dn->node_page = page;
788 		truncate_node(dn);
789 		freed++;
790 	} else {
791 		f2fs_put_page(page, 1);
792 	}
793 	trace_f2fs_truncate_nodes_exit(dn->inode, freed);
794 	return freed;
795 
796 out_err:
797 	f2fs_put_page(page, 1);
798 	trace_f2fs_truncate_nodes_exit(dn->inode, ret);
799 	return ret;
800 }
801 
802 static int truncate_partial_nodes(struct dnode_of_data *dn,
803 			struct f2fs_inode *ri, int *offset, int depth)
804 {
805 	struct page *pages[2];
806 	nid_t nid[3];
807 	nid_t child_nid;
808 	int err = 0;
809 	int i;
810 	int idx = depth - 2;
811 
812 	nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
813 	if (!nid[0])
814 		return 0;
815 
816 	/* get indirect nodes in the path */
817 	for (i = 0; i < idx + 1; i++) {
818 		/* reference count'll be increased */
819 		pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
820 		if (IS_ERR(pages[i])) {
821 			err = PTR_ERR(pages[i]);
822 			idx = i - 1;
823 			goto fail;
824 		}
825 		nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
826 	}
827 
828 	ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
829 
830 	/* free direct nodes linked to a partial indirect node */
831 	for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
832 		child_nid = get_nid(pages[idx], i, false);
833 		if (!child_nid)
834 			continue;
835 		dn->nid = child_nid;
836 		err = truncate_dnode(dn);
837 		if (err < 0)
838 			goto fail;
839 		if (set_nid(pages[idx], i, 0, false))
840 			dn->node_changed = true;
841 	}
842 
843 	if (offset[idx + 1] == 0) {
844 		dn->node_page = pages[idx];
845 		dn->nid = nid[idx];
846 		truncate_node(dn);
847 	} else {
848 		f2fs_put_page(pages[idx], 1);
849 	}
850 	offset[idx]++;
851 	offset[idx + 1] = 0;
852 	idx--;
853 fail:
854 	for (i = idx; i >= 0; i--)
855 		f2fs_put_page(pages[i], 1);
856 
857 	trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
858 
859 	return err;
860 }
861 
862 /*
863  * All the block addresses of data and nodes should be nullified.
864  */
865 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
866 {
867 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
868 	int err = 0, cont = 1;
869 	int level, offset[4], noffset[4];
870 	unsigned int nofs = 0;
871 	struct f2fs_inode *ri;
872 	struct dnode_of_data dn;
873 	struct page *page;
874 
875 	trace_f2fs_truncate_inode_blocks_enter(inode, from);
876 
877 	level = get_node_path(inode, from, offset, noffset);
878 
879 	page = get_node_page(sbi, inode->i_ino);
880 	if (IS_ERR(page)) {
881 		trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
882 		return PTR_ERR(page);
883 	}
884 
885 	set_new_dnode(&dn, inode, page, NULL, 0);
886 	unlock_page(page);
887 
888 	ri = F2FS_INODE(page);
889 	switch (level) {
890 	case 0:
891 	case 1:
892 		nofs = noffset[1];
893 		break;
894 	case 2:
895 		nofs = noffset[1];
896 		if (!offset[level - 1])
897 			goto skip_partial;
898 		err = truncate_partial_nodes(&dn, ri, offset, level);
899 		if (err < 0 && err != -ENOENT)
900 			goto fail;
901 		nofs += 1 + NIDS_PER_BLOCK;
902 		break;
903 	case 3:
904 		nofs = 5 + 2 * NIDS_PER_BLOCK;
905 		if (!offset[level - 1])
906 			goto skip_partial;
907 		err = truncate_partial_nodes(&dn, ri, offset, level);
908 		if (err < 0 && err != -ENOENT)
909 			goto fail;
910 		break;
911 	default:
912 		BUG();
913 	}
914 
915 skip_partial:
916 	while (cont) {
917 		dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
918 		switch (offset[0]) {
919 		case NODE_DIR1_BLOCK:
920 		case NODE_DIR2_BLOCK:
921 			err = truncate_dnode(&dn);
922 			break;
923 
924 		case NODE_IND1_BLOCK:
925 		case NODE_IND2_BLOCK:
926 			err = truncate_nodes(&dn, nofs, offset[1], 2);
927 			break;
928 
929 		case NODE_DIND_BLOCK:
930 			err = truncate_nodes(&dn, nofs, offset[1], 3);
931 			cont = 0;
932 			break;
933 
934 		default:
935 			BUG();
936 		}
937 		if (err < 0 && err != -ENOENT)
938 			goto fail;
939 		if (offset[1] == 0 &&
940 				ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
941 			lock_page(page);
942 			BUG_ON(page->mapping != NODE_MAPPING(sbi));
943 			f2fs_wait_on_page_writeback(page, NODE, true);
944 			ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
945 			set_page_dirty(page);
946 			unlock_page(page);
947 		}
948 		offset[1] = 0;
949 		offset[0]++;
950 		nofs += err;
951 	}
952 fail:
953 	f2fs_put_page(page, 0);
954 	trace_f2fs_truncate_inode_blocks_exit(inode, err);
955 	return err > 0 ? 0 : err;
956 }
957 
958 int truncate_xattr_node(struct inode *inode, struct page *page)
959 {
960 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
961 	nid_t nid = F2FS_I(inode)->i_xattr_nid;
962 	struct dnode_of_data dn;
963 	struct page *npage;
964 
965 	if (!nid)
966 		return 0;
967 
968 	npage = get_node_page(sbi, nid);
969 	if (IS_ERR(npage))
970 		return PTR_ERR(npage);
971 
972 	f2fs_i_xnid_write(inode, 0);
973 
974 	set_new_dnode(&dn, inode, page, npage, nid);
975 
976 	if (page)
977 		dn.inode_page_locked = true;
978 	truncate_node(&dn);
979 	return 0;
980 }
981 
982 /*
983  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
984  * f2fs_unlock_op().
985  */
986 int remove_inode_page(struct inode *inode)
987 {
988 	struct dnode_of_data dn;
989 	int err;
990 
991 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
992 	err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
993 	if (err)
994 		return err;
995 
996 	err = truncate_xattr_node(inode, dn.inode_page);
997 	if (err) {
998 		f2fs_put_dnode(&dn);
999 		return err;
1000 	}
1001 
1002 	/* remove potential inline_data blocks */
1003 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1004 				S_ISLNK(inode->i_mode))
1005 		truncate_data_blocks_range(&dn, 1);
1006 
1007 	/* 0 is possible, after f2fs_new_inode() has failed */
1008 	f2fs_bug_on(F2FS_I_SB(inode),
1009 			inode->i_blocks != 0 && inode->i_blocks != 1);
1010 
1011 	/* will put inode & node pages */
1012 	truncate_node(&dn);
1013 	return 0;
1014 }
1015 
1016 struct page *new_inode_page(struct inode *inode)
1017 {
1018 	struct dnode_of_data dn;
1019 
1020 	/* allocate inode page for new inode */
1021 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1022 
1023 	/* caller should f2fs_put_page(page, 1); */
1024 	return new_node_page(&dn, 0, NULL);
1025 }
1026 
1027 struct page *new_node_page(struct dnode_of_data *dn,
1028 				unsigned int ofs, struct page *ipage)
1029 {
1030 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1031 	struct node_info new_ni;
1032 	struct page *page;
1033 	int err;
1034 
1035 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1036 		return ERR_PTR(-EPERM);
1037 
1038 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1039 	if (!page)
1040 		return ERR_PTR(-ENOMEM);
1041 
1042 	if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1043 		err = -ENOSPC;
1044 		goto fail;
1045 	}
1046 #ifdef CONFIG_F2FS_CHECK_FS
1047 	get_node_info(sbi, dn->nid, &new_ni);
1048 	f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1049 #endif
1050 	new_ni.nid = dn->nid;
1051 	new_ni.ino = dn->inode->i_ino;
1052 	new_ni.blk_addr = NULL_ADDR;
1053 	new_ni.flag = 0;
1054 	new_ni.version = 0;
1055 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1056 
1057 	f2fs_wait_on_page_writeback(page, NODE, true);
1058 	fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1059 	set_cold_node(dn->inode, page);
1060 	if (!PageUptodate(page))
1061 		SetPageUptodate(page);
1062 	if (set_page_dirty(page))
1063 		dn->node_changed = true;
1064 
1065 	if (f2fs_has_xattr_block(ofs))
1066 		f2fs_i_xnid_write(dn->inode, dn->nid);
1067 
1068 	if (ofs == 0)
1069 		inc_valid_inode_count(sbi);
1070 	return page;
1071 
1072 fail:
1073 	clear_node_page_dirty(page);
1074 	f2fs_put_page(page, 1);
1075 	return ERR_PTR(err);
1076 }
1077 
1078 /*
1079  * Caller should do after getting the following values.
1080  * 0: f2fs_put_page(page, 0)
1081  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1082  */
1083 static int read_node_page(struct page *page, int op_flags)
1084 {
1085 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1086 	struct node_info ni;
1087 	struct f2fs_io_info fio = {
1088 		.sbi = sbi,
1089 		.type = NODE,
1090 		.op = REQ_OP_READ,
1091 		.op_flags = op_flags,
1092 		.page = page,
1093 		.encrypted_page = NULL,
1094 	};
1095 
1096 	if (PageUptodate(page))
1097 		return LOCKED_PAGE;
1098 
1099 	get_node_info(sbi, page->index, &ni);
1100 
1101 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1102 		ClearPageUptodate(page);
1103 		return -ENOENT;
1104 	}
1105 
1106 	fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1107 	return f2fs_submit_page_bio(&fio);
1108 }
1109 
1110 /*
1111  * Readahead a node page
1112  */
1113 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1114 {
1115 	struct page *apage;
1116 	int err;
1117 
1118 	if (!nid)
1119 		return;
1120 	f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1121 
1122 	rcu_read_lock();
1123 	apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1124 	rcu_read_unlock();
1125 	if (apage)
1126 		return;
1127 
1128 	apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1129 	if (!apage)
1130 		return;
1131 
1132 	err = read_node_page(apage, REQ_RAHEAD);
1133 	f2fs_put_page(apage, err ? 1 : 0);
1134 }
1135 
1136 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1137 					struct page *parent, int start)
1138 {
1139 	struct page *page;
1140 	int err;
1141 
1142 	if (!nid)
1143 		return ERR_PTR(-ENOENT);
1144 	f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1145 repeat:
1146 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1147 	if (!page)
1148 		return ERR_PTR(-ENOMEM);
1149 
1150 	err = read_node_page(page, 0);
1151 	if (err < 0) {
1152 		f2fs_put_page(page, 1);
1153 		return ERR_PTR(err);
1154 	} else if (err == LOCKED_PAGE) {
1155 		goto page_hit;
1156 	}
1157 
1158 	if (parent)
1159 		ra_node_pages(parent, start + 1, MAX_RA_NODE);
1160 
1161 	lock_page(page);
1162 
1163 	if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1164 		f2fs_put_page(page, 1);
1165 		goto repeat;
1166 	}
1167 
1168 	if (unlikely(!PageUptodate(page)))
1169 		goto out_err;
1170 page_hit:
1171 	if(unlikely(nid != nid_of_node(page))) {
1172 		f2fs_bug_on(sbi, 1);
1173 		ClearPageUptodate(page);
1174 out_err:
1175 		f2fs_put_page(page, 1);
1176 		return ERR_PTR(-EIO);
1177 	}
1178 	return page;
1179 }
1180 
1181 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1182 {
1183 	return __get_node_page(sbi, nid, NULL, 0);
1184 }
1185 
1186 struct page *get_node_page_ra(struct page *parent, int start)
1187 {
1188 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1189 	nid_t nid = get_nid(parent, start, false);
1190 
1191 	return __get_node_page(sbi, nid, parent, start);
1192 }
1193 
1194 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1195 {
1196 	struct inode *inode;
1197 	struct page *page;
1198 	int ret;
1199 
1200 	/* should flush inline_data before evict_inode */
1201 	inode = ilookup(sbi->sb, ino);
1202 	if (!inode)
1203 		return;
1204 
1205 	page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1206 	if (!page)
1207 		goto iput_out;
1208 
1209 	if (!PageUptodate(page))
1210 		goto page_out;
1211 
1212 	if (!PageDirty(page))
1213 		goto page_out;
1214 
1215 	if (!clear_page_dirty_for_io(page))
1216 		goto page_out;
1217 
1218 	ret = f2fs_write_inline_data(inode, page);
1219 	inode_dec_dirty_pages(inode);
1220 	remove_dirty_inode(inode);
1221 	if (ret)
1222 		set_page_dirty(page);
1223 page_out:
1224 	f2fs_put_page(page, 1);
1225 iput_out:
1226 	iput(inode);
1227 }
1228 
1229 void move_node_page(struct page *node_page, int gc_type)
1230 {
1231 	if (gc_type == FG_GC) {
1232 		struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1233 		struct writeback_control wbc = {
1234 			.sync_mode = WB_SYNC_ALL,
1235 			.nr_to_write = 1,
1236 			.for_reclaim = 0,
1237 		};
1238 
1239 		set_page_dirty(node_page);
1240 		f2fs_wait_on_page_writeback(node_page, NODE, true);
1241 
1242 		f2fs_bug_on(sbi, PageWriteback(node_page));
1243 		if (!clear_page_dirty_for_io(node_page))
1244 			goto out_page;
1245 
1246 		if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1247 			unlock_page(node_page);
1248 		goto release_page;
1249 	} else {
1250 		/* set page dirty and write it */
1251 		if (!PageWriteback(node_page))
1252 			set_page_dirty(node_page);
1253 	}
1254 out_page:
1255 	unlock_page(node_page);
1256 release_page:
1257 	f2fs_put_page(node_page, 0);
1258 }
1259 
1260 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1261 {
1262 	pgoff_t index, end;
1263 	struct pagevec pvec;
1264 	struct page *last_page = NULL;
1265 
1266 	pagevec_init(&pvec, 0);
1267 	index = 0;
1268 	end = ULONG_MAX;
1269 
1270 	while (index <= end) {
1271 		int i, nr_pages;
1272 		nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1273 				PAGECACHE_TAG_DIRTY,
1274 				min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1275 		if (nr_pages == 0)
1276 			break;
1277 
1278 		for (i = 0; i < nr_pages; i++) {
1279 			struct page *page = pvec.pages[i];
1280 
1281 			if (unlikely(f2fs_cp_error(sbi))) {
1282 				f2fs_put_page(last_page, 0);
1283 				pagevec_release(&pvec);
1284 				return ERR_PTR(-EIO);
1285 			}
1286 
1287 			if (!IS_DNODE(page) || !is_cold_node(page))
1288 				continue;
1289 			if (ino_of_node(page) != ino)
1290 				continue;
1291 
1292 			lock_page(page);
1293 
1294 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1295 continue_unlock:
1296 				unlock_page(page);
1297 				continue;
1298 			}
1299 			if (ino_of_node(page) != ino)
1300 				goto continue_unlock;
1301 
1302 			if (!PageDirty(page)) {
1303 				/* someone wrote it for us */
1304 				goto continue_unlock;
1305 			}
1306 
1307 			if (last_page)
1308 				f2fs_put_page(last_page, 0);
1309 
1310 			get_page(page);
1311 			last_page = page;
1312 			unlock_page(page);
1313 		}
1314 		pagevec_release(&pvec);
1315 		cond_resched();
1316 	}
1317 	return last_page;
1318 }
1319 
1320 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1321 				struct writeback_control *wbc)
1322 {
1323 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1324 	nid_t nid;
1325 	struct node_info ni;
1326 	struct f2fs_io_info fio = {
1327 		.sbi = sbi,
1328 		.type = NODE,
1329 		.op = REQ_OP_WRITE,
1330 		.op_flags = wbc_to_write_flags(wbc),
1331 		.page = page,
1332 		.encrypted_page = NULL,
1333 		.submitted = false,
1334 	};
1335 
1336 	trace_f2fs_writepage(page, NODE);
1337 
1338 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1339 		goto redirty_out;
1340 	if (unlikely(f2fs_cp_error(sbi)))
1341 		goto redirty_out;
1342 
1343 	/* get old block addr of this node page */
1344 	nid = nid_of_node(page);
1345 	f2fs_bug_on(sbi, page->index != nid);
1346 
1347 	if (wbc->for_reclaim) {
1348 		if (!down_read_trylock(&sbi->node_write))
1349 			goto redirty_out;
1350 	} else {
1351 		down_read(&sbi->node_write);
1352 	}
1353 
1354 	get_node_info(sbi, nid, &ni);
1355 
1356 	/* This page is already truncated */
1357 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1358 		ClearPageUptodate(page);
1359 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1360 		up_read(&sbi->node_write);
1361 		unlock_page(page);
1362 		return 0;
1363 	}
1364 
1365 	if (atomic && !test_opt(sbi, NOBARRIER))
1366 		fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1367 
1368 	set_page_writeback(page);
1369 	fio.old_blkaddr = ni.blk_addr;
1370 	write_node_page(nid, &fio);
1371 	set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1372 	dec_page_count(sbi, F2FS_DIRTY_NODES);
1373 	up_read(&sbi->node_write);
1374 
1375 	if (wbc->for_reclaim) {
1376 		f2fs_submit_merged_bio_cond(sbi, page->mapping->host, 0,
1377 						page->index, NODE, WRITE);
1378 		submitted = NULL;
1379 	}
1380 
1381 	unlock_page(page);
1382 
1383 	if (unlikely(f2fs_cp_error(sbi))) {
1384 		f2fs_submit_merged_bio(sbi, NODE, WRITE);
1385 		submitted = NULL;
1386 	}
1387 	if (submitted)
1388 		*submitted = fio.submitted;
1389 
1390 	return 0;
1391 
1392 redirty_out:
1393 	redirty_page_for_writepage(wbc, page);
1394 	return AOP_WRITEPAGE_ACTIVATE;
1395 }
1396 
1397 static int f2fs_write_node_page(struct page *page,
1398 				struct writeback_control *wbc)
1399 {
1400 	return __write_node_page(page, false, NULL, wbc);
1401 }
1402 
1403 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1404 			struct writeback_control *wbc, bool atomic)
1405 {
1406 	pgoff_t index, end;
1407 	pgoff_t last_idx = ULONG_MAX;
1408 	struct pagevec pvec;
1409 	int ret = 0;
1410 	struct page *last_page = NULL;
1411 	bool marked = false;
1412 	nid_t ino = inode->i_ino;
1413 
1414 	if (atomic) {
1415 		last_page = last_fsync_dnode(sbi, ino);
1416 		if (IS_ERR_OR_NULL(last_page))
1417 			return PTR_ERR_OR_ZERO(last_page);
1418 	}
1419 retry:
1420 	pagevec_init(&pvec, 0);
1421 	index = 0;
1422 	end = ULONG_MAX;
1423 
1424 	while (index <= end) {
1425 		int i, nr_pages;
1426 		nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1427 				PAGECACHE_TAG_DIRTY,
1428 				min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1429 		if (nr_pages == 0)
1430 			break;
1431 
1432 		for (i = 0; i < nr_pages; i++) {
1433 			struct page *page = pvec.pages[i];
1434 			bool submitted = false;
1435 
1436 			if (unlikely(f2fs_cp_error(sbi))) {
1437 				f2fs_put_page(last_page, 0);
1438 				pagevec_release(&pvec);
1439 				ret = -EIO;
1440 				goto out;
1441 			}
1442 
1443 			if (!IS_DNODE(page) || !is_cold_node(page))
1444 				continue;
1445 			if (ino_of_node(page) != ino)
1446 				continue;
1447 
1448 			lock_page(page);
1449 
1450 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1451 continue_unlock:
1452 				unlock_page(page);
1453 				continue;
1454 			}
1455 			if (ino_of_node(page) != ino)
1456 				goto continue_unlock;
1457 
1458 			if (!PageDirty(page) && page != last_page) {
1459 				/* someone wrote it for us */
1460 				goto continue_unlock;
1461 			}
1462 
1463 			f2fs_wait_on_page_writeback(page, NODE, true);
1464 			BUG_ON(PageWriteback(page));
1465 
1466 			if (!atomic || page == last_page) {
1467 				set_fsync_mark(page, 1);
1468 				if (IS_INODE(page)) {
1469 					if (is_inode_flag_set(inode,
1470 								FI_DIRTY_INODE))
1471 						update_inode(inode, page);
1472 					set_dentry_mark(page,
1473 						need_dentry_mark(sbi, ino));
1474 				}
1475 				/*  may be written by other thread */
1476 				if (!PageDirty(page))
1477 					set_page_dirty(page);
1478 			}
1479 
1480 			if (!clear_page_dirty_for_io(page))
1481 				goto continue_unlock;
1482 
1483 			ret = __write_node_page(page, atomic &&
1484 						page == last_page,
1485 						&submitted, wbc);
1486 			if (ret) {
1487 				unlock_page(page);
1488 				f2fs_put_page(last_page, 0);
1489 				break;
1490 			} else if (submitted) {
1491 				last_idx = page->index;
1492 			}
1493 
1494 			if (page == last_page) {
1495 				f2fs_put_page(page, 0);
1496 				marked = true;
1497 				break;
1498 			}
1499 		}
1500 		pagevec_release(&pvec);
1501 		cond_resched();
1502 
1503 		if (ret || marked)
1504 			break;
1505 	}
1506 	if (!ret && atomic && !marked) {
1507 		f2fs_msg(sbi->sb, KERN_DEBUG,
1508 			"Retry to write fsync mark: ino=%u, idx=%lx",
1509 					ino, last_page->index);
1510 		lock_page(last_page);
1511 		f2fs_wait_on_page_writeback(last_page, NODE, true);
1512 		set_page_dirty(last_page);
1513 		unlock_page(last_page);
1514 		goto retry;
1515 	}
1516 out:
1517 	if (last_idx != ULONG_MAX)
1518 		f2fs_submit_merged_bio_cond(sbi, NULL, ino, last_idx,
1519 							NODE, WRITE);
1520 	return ret ? -EIO: 0;
1521 }
1522 
1523 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc)
1524 {
1525 	pgoff_t index, end;
1526 	struct pagevec pvec;
1527 	int step = 0;
1528 	int nwritten = 0;
1529 	int ret = 0;
1530 
1531 	pagevec_init(&pvec, 0);
1532 
1533 next_step:
1534 	index = 0;
1535 	end = ULONG_MAX;
1536 
1537 	while (index <= end) {
1538 		int i, nr_pages;
1539 		nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1540 				PAGECACHE_TAG_DIRTY,
1541 				min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1542 		if (nr_pages == 0)
1543 			break;
1544 
1545 		for (i = 0; i < nr_pages; i++) {
1546 			struct page *page = pvec.pages[i];
1547 			bool submitted = false;
1548 
1549 			if (unlikely(f2fs_cp_error(sbi))) {
1550 				pagevec_release(&pvec);
1551 				ret = -EIO;
1552 				goto out;
1553 			}
1554 
1555 			/*
1556 			 * flushing sequence with step:
1557 			 * 0. indirect nodes
1558 			 * 1. dentry dnodes
1559 			 * 2. file dnodes
1560 			 */
1561 			if (step == 0 && IS_DNODE(page))
1562 				continue;
1563 			if (step == 1 && (!IS_DNODE(page) ||
1564 						is_cold_node(page)))
1565 				continue;
1566 			if (step == 2 && (!IS_DNODE(page) ||
1567 						!is_cold_node(page)))
1568 				continue;
1569 lock_node:
1570 			if (!trylock_page(page))
1571 				continue;
1572 
1573 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1574 continue_unlock:
1575 				unlock_page(page);
1576 				continue;
1577 			}
1578 
1579 			if (!PageDirty(page)) {
1580 				/* someone wrote it for us */
1581 				goto continue_unlock;
1582 			}
1583 
1584 			/* flush inline_data */
1585 			if (is_inline_node(page)) {
1586 				clear_inline_node(page);
1587 				unlock_page(page);
1588 				flush_inline_data(sbi, ino_of_node(page));
1589 				goto lock_node;
1590 			}
1591 
1592 			f2fs_wait_on_page_writeback(page, NODE, true);
1593 
1594 			BUG_ON(PageWriteback(page));
1595 			if (!clear_page_dirty_for_io(page))
1596 				goto continue_unlock;
1597 
1598 			set_fsync_mark(page, 0);
1599 			set_dentry_mark(page, 0);
1600 
1601 			ret = __write_node_page(page, false, &submitted, wbc);
1602 			if (ret)
1603 				unlock_page(page);
1604 			else if (submitted)
1605 				nwritten++;
1606 
1607 			if (--wbc->nr_to_write == 0)
1608 				break;
1609 		}
1610 		pagevec_release(&pvec);
1611 		cond_resched();
1612 
1613 		if (wbc->nr_to_write == 0) {
1614 			step = 2;
1615 			break;
1616 		}
1617 	}
1618 
1619 	if (step < 2) {
1620 		step++;
1621 		goto next_step;
1622 	}
1623 out:
1624 	if (nwritten)
1625 		f2fs_submit_merged_bio(sbi, NODE, WRITE);
1626 	return ret;
1627 }
1628 
1629 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1630 {
1631 	pgoff_t index = 0, end = ULONG_MAX;
1632 	struct pagevec pvec;
1633 	int ret2, ret = 0;
1634 
1635 	pagevec_init(&pvec, 0);
1636 
1637 	while (index <= end) {
1638 		int i, nr_pages;
1639 		nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1640 				PAGECACHE_TAG_WRITEBACK,
1641 				min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1642 		if (nr_pages == 0)
1643 			break;
1644 
1645 		for (i = 0; i < nr_pages; i++) {
1646 			struct page *page = pvec.pages[i];
1647 
1648 			/* until radix tree lookup accepts end_index */
1649 			if (unlikely(page->index > end))
1650 				continue;
1651 
1652 			if (ino && ino_of_node(page) == ino) {
1653 				f2fs_wait_on_page_writeback(page, NODE, true);
1654 				if (TestClearPageError(page))
1655 					ret = -EIO;
1656 			}
1657 		}
1658 		pagevec_release(&pvec);
1659 		cond_resched();
1660 	}
1661 
1662 	ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1663 	if (!ret)
1664 		ret = ret2;
1665 	return ret;
1666 }
1667 
1668 static int f2fs_write_node_pages(struct address_space *mapping,
1669 			    struct writeback_control *wbc)
1670 {
1671 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1672 	struct blk_plug plug;
1673 	long diff;
1674 
1675 	/* balancing f2fs's metadata in background */
1676 	f2fs_balance_fs_bg(sbi);
1677 
1678 	/* collect a number of dirty node pages and write together */
1679 	if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1680 		goto skip_write;
1681 
1682 	trace_f2fs_writepages(mapping->host, wbc, NODE);
1683 
1684 	diff = nr_pages_to_write(sbi, NODE, wbc);
1685 	wbc->sync_mode = WB_SYNC_NONE;
1686 	blk_start_plug(&plug);
1687 	sync_node_pages(sbi, wbc);
1688 	blk_finish_plug(&plug);
1689 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1690 	return 0;
1691 
1692 skip_write:
1693 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1694 	trace_f2fs_writepages(mapping->host, wbc, NODE);
1695 	return 0;
1696 }
1697 
1698 static int f2fs_set_node_page_dirty(struct page *page)
1699 {
1700 	trace_f2fs_set_page_dirty(page, NODE);
1701 
1702 	if (!PageUptodate(page))
1703 		SetPageUptodate(page);
1704 	if (!PageDirty(page)) {
1705 		f2fs_set_page_dirty_nobuffers(page);
1706 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1707 		SetPagePrivate(page);
1708 		f2fs_trace_pid(page);
1709 		return 1;
1710 	}
1711 	return 0;
1712 }
1713 
1714 /*
1715  * Structure of the f2fs node operations
1716  */
1717 const struct address_space_operations f2fs_node_aops = {
1718 	.writepage	= f2fs_write_node_page,
1719 	.writepages	= f2fs_write_node_pages,
1720 	.set_page_dirty	= f2fs_set_node_page_dirty,
1721 	.invalidatepage	= f2fs_invalidate_page,
1722 	.releasepage	= f2fs_release_page,
1723 #ifdef CONFIG_MIGRATION
1724 	.migratepage    = f2fs_migrate_page,
1725 #endif
1726 };
1727 
1728 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1729 						nid_t n)
1730 {
1731 	return radix_tree_lookup(&nm_i->free_nid_root, n);
1732 }
1733 
1734 static int __insert_nid_to_list(struct f2fs_sb_info *sbi,
1735 			struct free_nid *i, enum nid_list list, bool new)
1736 {
1737 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1738 
1739 	if (new) {
1740 		int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
1741 		if (err)
1742 			return err;
1743 	}
1744 
1745 	f2fs_bug_on(sbi, list == FREE_NID_LIST ? i->state != NID_NEW :
1746 						i->state != NID_ALLOC);
1747 	nm_i->nid_cnt[list]++;
1748 	list_add_tail(&i->list, &nm_i->nid_list[list]);
1749 	return 0;
1750 }
1751 
1752 static void __remove_nid_from_list(struct f2fs_sb_info *sbi,
1753 			struct free_nid *i, enum nid_list list, bool reuse)
1754 {
1755 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1756 
1757 	f2fs_bug_on(sbi, list == FREE_NID_LIST ? i->state != NID_NEW :
1758 						i->state != NID_ALLOC);
1759 	nm_i->nid_cnt[list]--;
1760 	list_del(&i->list);
1761 	if (!reuse)
1762 		radix_tree_delete(&nm_i->free_nid_root, i->nid);
1763 }
1764 
1765 /* return if the nid is recognized as free */
1766 static bool add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1767 {
1768 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1769 	struct free_nid *i;
1770 	struct nat_entry *ne;
1771 	int err;
1772 
1773 	/* 0 nid should not be used */
1774 	if (unlikely(nid == 0))
1775 		return false;
1776 
1777 	if (build) {
1778 		/* do not add allocated nids */
1779 		ne = __lookup_nat_cache(nm_i, nid);
1780 		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1781 				nat_get_blkaddr(ne) != NULL_ADDR))
1782 			return false;
1783 	}
1784 
1785 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1786 	i->nid = nid;
1787 	i->state = NID_NEW;
1788 
1789 	if (radix_tree_preload(GFP_NOFS)) {
1790 		kmem_cache_free(free_nid_slab, i);
1791 		return true;
1792 	}
1793 
1794 	spin_lock(&nm_i->nid_list_lock);
1795 	err = __insert_nid_to_list(sbi, i, FREE_NID_LIST, true);
1796 	spin_unlock(&nm_i->nid_list_lock);
1797 	radix_tree_preload_end();
1798 	if (err) {
1799 		kmem_cache_free(free_nid_slab, i);
1800 		return true;
1801 	}
1802 	return true;
1803 }
1804 
1805 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
1806 {
1807 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1808 	struct free_nid *i;
1809 	bool need_free = false;
1810 
1811 	spin_lock(&nm_i->nid_list_lock);
1812 	i = __lookup_free_nid_list(nm_i, nid);
1813 	if (i && i->state == NID_NEW) {
1814 		__remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
1815 		need_free = true;
1816 	}
1817 	spin_unlock(&nm_i->nid_list_lock);
1818 
1819 	if (need_free)
1820 		kmem_cache_free(free_nid_slab, i);
1821 }
1822 
1823 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
1824 			bool set, bool build, bool locked)
1825 {
1826 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1827 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
1828 	unsigned int nid_ofs = nid - START_NID(nid);
1829 
1830 	if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1831 		return;
1832 
1833 	if (set)
1834 		__set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1835 	else
1836 		__clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1837 
1838 	if (!locked)
1839 		spin_lock(&nm_i->free_nid_lock);
1840 	if (set)
1841 		nm_i->free_nid_count[nat_ofs]++;
1842 	else if (!build)
1843 		nm_i->free_nid_count[nat_ofs]--;
1844 	if (!locked)
1845 		spin_unlock(&nm_i->free_nid_lock);
1846 }
1847 
1848 static void scan_nat_page(struct f2fs_sb_info *sbi,
1849 			struct page *nat_page, nid_t start_nid)
1850 {
1851 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1852 	struct f2fs_nat_block *nat_blk = page_address(nat_page);
1853 	block_t blk_addr;
1854 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
1855 	int i;
1856 
1857 	if (test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1858 		return;
1859 
1860 	__set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
1861 
1862 	i = start_nid % NAT_ENTRY_PER_BLOCK;
1863 
1864 	for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1865 		bool freed = false;
1866 
1867 		if (unlikely(start_nid >= nm_i->max_nid))
1868 			break;
1869 
1870 		blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1871 		f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1872 		if (blk_addr == NULL_ADDR)
1873 			freed = add_free_nid(sbi, start_nid, true);
1874 		update_free_nid_bitmap(sbi, start_nid, freed, true, false);
1875 	}
1876 }
1877 
1878 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
1879 {
1880 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1881 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1882 	struct f2fs_journal *journal = curseg->journal;
1883 	unsigned int i, idx;
1884 
1885 	down_read(&nm_i->nat_tree_lock);
1886 
1887 	for (i = 0; i < nm_i->nat_blocks; i++) {
1888 		if (!test_bit_le(i, nm_i->nat_block_bitmap))
1889 			continue;
1890 		if (!nm_i->free_nid_count[i])
1891 			continue;
1892 		for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
1893 			nid_t nid;
1894 
1895 			if (!test_bit_le(idx, nm_i->free_nid_bitmap[i]))
1896 				continue;
1897 
1898 			nid = i * NAT_ENTRY_PER_BLOCK + idx;
1899 			add_free_nid(sbi, nid, true);
1900 
1901 			if (nm_i->nid_cnt[FREE_NID_LIST] >= MAX_FREE_NIDS)
1902 				goto out;
1903 		}
1904 	}
1905 out:
1906 	down_read(&curseg->journal_rwsem);
1907 	for (i = 0; i < nats_in_cursum(journal); i++) {
1908 		block_t addr;
1909 		nid_t nid;
1910 
1911 		addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1912 		nid = le32_to_cpu(nid_in_journal(journal, i));
1913 		if (addr == NULL_ADDR)
1914 			add_free_nid(sbi, nid, true);
1915 		else
1916 			remove_free_nid(sbi, nid);
1917 	}
1918 	up_read(&curseg->journal_rwsem);
1919 	up_read(&nm_i->nat_tree_lock);
1920 }
1921 
1922 static void __build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
1923 {
1924 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1925 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1926 	struct f2fs_journal *journal = curseg->journal;
1927 	int i = 0;
1928 	nid_t nid = nm_i->next_scan_nid;
1929 
1930 	/* Enough entries */
1931 	if (nm_i->nid_cnt[FREE_NID_LIST] >= NAT_ENTRY_PER_BLOCK)
1932 		return;
1933 
1934 	if (!sync && !available_free_memory(sbi, FREE_NIDS))
1935 		return;
1936 
1937 	if (!mount) {
1938 		/* try to find free nids in free_nid_bitmap */
1939 		scan_free_nid_bits(sbi);
1940 
1941 		if (nm_i->nid_cnt[FREE_NID_LIST])
1942 			return;
1943 	}
1944 
1945 	/* readahead nat pages to be scanned */
1946 	ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1947 							META_NAT, true);
1948 
1949 	down_read(&nm_i->nat_tree_lock);
1950 
1951 	while (1) {
1952 		struct page *page = get_current_nat_page(sbi, nid);
1953 
1954 		scan_nat_page(sbi, page, nid);
1955 		f2fs_put_page(page, 1);
1956 
1957 		nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1958 		if (unlikely(nid >= nm_i->max_nid))
1959 			nid = 0;
1960 
1961 		if (++i >= FREE_NID_PAGES)
1962 			break;
1963 	}
1964 
1965 	/* go to the next free nat pages to find free nids abundantly */
1966 	nm_i->next_scan_nid = nid;
1967 
1968 	/* find free nids from current sum_pages */
1969 	down_read(&curseg->journal_rwsem);
1970 	for (i = 0; i < nats_in_cursum(journal); i++) {
1971 		block_t addr;
1972 
1973 		addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1974 		nid = le32_to_cpu(nid_in_journal(journal, i));
1975 		if (addr == NULL_ADDR)
1976 			add_free_nid(sbi, nid, true);
1977 		else
1978 			remove_free_nid(sbi, nid);
1979 	}
1980 	up_read(&curseg->journal_rwsem);
1981 	up_read(&nm_i->nat_tree_lock);
1982 
1983 	ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1984 					nm_i->ra_nid_pages, META_NAT, false);
1985 }
1986 
1987 void build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
1988 {
1989 	mutex_lock(&NM_I(sbi)->build_lock);
1990 	__build_free_nids(sbi, sync, mount);
1991 	mutex_unlock(&NM_I(sbi)->build_lock);
1992 }
1993 
1994 /*
1995  * If this function returns success, caller can obtain a new nid
1996  * from second parameter of this function.
1997  * The returned nid could be used ino as well as nid when inode is created.
1998  */
1999 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2000 {
2001 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2002 	struct free_nid *i = NULL;
2003 retry:
2004 #ifdef CONFIG_F2FS_FAULT_INJECTION
2005 	if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2006 		f2fs_show_injection_info(FAULT_ALLOC_NID);
2007 		return false;
2008 	}
2009 #endif
2010 	spin_lock(&nm_i->nid_list_lock);
2011 
2012 	if (unlikely(nm_i->available_nids == 0)) {
2013 		spin_unlock(&nm_i->nid_list_lock);
2014 		return false;
2015 	}
2016 
2017 	/* We should not use stale free nids created by build_free_nids */
2018 	if (nm_i->nid_cnt[FREE_NID_LIST] && !on_build_free_nids(nm_i)) {
2019 		f2fs_bug_on(sbi, list_empty(&nm_i->nid_list[FREE_NID_LIST]));
2020 		i = list_first_entry(&nm_i->nid_list[FREE_NID_LIST],
2021 					struct free_nid, list);
2022 		*nid = i->nid;
2023 
2024 		__remove_nid_from_list(sbi, i, FREE_NID_LIST, true);
2025 		i->state = NID_ALLOC;
2026 		__insert_nid_to_list(sbi, i, ALLOC_NID_LIST, false);
2027 		nm_i->available_nids--;
2028 
2029 		update_free_nid_bitmap(sbi, *nid, false, false, false);
2030 
2031 		spin_unlock(&nm_i->nid_list_lock);
2032 		return true;
2033 	}
2034 	spin_unlock(&nm_i->nid_list_lock);
2035 
2036 	/* Let's scan nat pages and its caches to get free nids */
2037 	build_free_nids(sbi, true, false);
2038 	goto retry;
2039 }
2040 
2041 /*
2042  * alloc_nid() should be called prior to this function.
2043  */
2044 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2045 {
2046 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2047 	struct free_nid *i;
2048 
2049 	spin_lock(&nm_i->nid_list_lock);
2050 	i = __lookup_free_nid_list(nm_i, nid);
2051 	f2fs_bug_on(sbi, !i);
2052 	__remove_nid_from_list(sbi, i, ALLOC_NID_LIST, false);
2053 	spin_unlock(&nm_i->nid_list_lock);
2054 
2055 	kmem_cache_free(free_nid_slab, i);
2056 }
2057 
2058 /*
2059  * alloc_nid() should be called prior to this function.
2060  */
2061 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2062 {
2063 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2064 	struct free_nid *i;
2065 	bool need_free = false;
2066 
2067 	if (!nid)
2068 		return;
2069 
2070 	spin_lock(&nm_i->nid_list_lock);
2071 	i = __lookup_free_nid_list(nm_i, nid);
2072 	f2fs_bug_on(sbi, !i);
2073 
2074 	if (!available_free_memory(sbi, FREE_NIDS)) {
2075 		__remove_nid_from_list(sbi, i, ALLOC_NID_LIST, false);
2076 		need_free = true;
2077 	} else {
2078 		__remove_nid_from_list(sbi, i, ALLOC_NID_LIST, true);
2079 		i->state = NID_NEW;
2080 		__insert_nid_to_list(sbi, i, FREE_NID_LIST, false);
2081 	}
2082 
2083 	nm_i->available_nids++;
2084 
2085 	update_free_nid_bitmap(sbi, nid, true, false, false);
2086 
2087 	spin_unlock(&nm_i->nid_list_lock);
2088 
2089 	if (need_free)
2090 		kmem_cache_free(free_nid_slab, i);
2091 }
2092 
2093 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2094 {
2095 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2096 	struct free_nid *i, *next;
2097 	int nr = nr_shrink;
2098 
2099 	if (nm_i->nid_cnt[FREE_NID_LIST] <= MAX_FREE_NIDS)
2100 		return 0;
2101 
2102 	if (!mutex_trylock(&nm_i->build_lock))
2103 		return 0;
2104 
2105 	spin_lock(&nm_i->nid_list_lock);
2106 	list_for_each_entry_safe(i, next, &nm_i->nid_list[FREE_NID_LIST],
2107 									list) {
2108 		if (nr_shrink <= 0 ||
2109 				nm_i->nid_cnt[FREE_NID_LIST] <= MAX_FREE_NIDS)
2110 			break;
2111 
2112 		__remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
2113 		kmem_cache_free(free_nid_slab, i);
2114 		nr_shrink--;
2115 	}
2116 	spin_unlock(&nm_i->nid_list_lock);
2117 	mutex_unlock(&nm_i->build_lock);
2118 
2119 	return nr - nr_shrink;
2120 }
2121 
2122 void recover_inline_xattr(struct inode *inode, struct page *page)
2123 {
2124 	void *src_addr, *dst_addr;
2125 	size_t inline_size;
2126 	struct page *ipage;
2127 	struct f2fs_inode *ri;
2128 
2129 	ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
2130 	f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2131 
2132 	ri = F2FS_INODE(page);
2133 	if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
2134 		clear_inode_flag(inode, FI_INLINE_XATTR);
2135 		goto update_inode;
2136 	}
2137 
2138 	dst_addr = inline_xattr_addr(ipage);
2139 	src_addr = inline_xattr_addr(page);
2140 	inline_size = inline_xattr_size(inode);
2141 
2142 	f2fs_wait_on_page_writeback(ipage, NODE, true);
2143 	memcpy(dst_addr, src_addr, inline_size);
2144 update_inode:
2145 	update_inode(inode, ipage);
2146 	f2fs_put_page(ipage, 1);
2147 }
2148 
2149 int recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
2150 {
2151 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2152 	nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2153 	nid_t new_xnid = nid_of_node(page);
2154 	struct node_info ni;
2155 	struct page *xpage;
2156 
2157 	if (!prev_xnid)
2158 		goto recover_xnid;
2159 
2160 	/* 1: invalidate the previous xattr nid */
2161 	get_node_info(sbi, prev_xnid, &ni);
2162 	f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
2163 	invalidate_blocks(sbi, ni.blk_addr);
2164 	dec_valid_node_count(sbi, inode);
2165 	set_node_addr(sbi, &ni, NULL_ADDR, false);
2166 
2167 recover_xnid:
2168 	/* 2: update xattr nid in inode */
2169 	remove_free_nid(sbi, new_xnid);
2170 	f2fs_i_xnid_write(inode, new_xnid);
2171 	if (unlikely(!inc_valid_node_count(sbi, inode)))
2172 		f2fs_bug_on(sbi, 1);
2173 	update_inode_page(inode);
2174 
2175 	/* 3: update and set xattr node page dirty */
2176 	xpage = grab_cache_page(NODE_MAPPING(sbi), new_xnid);
2177 	if (!xpage)
2178 		return -ENOMEM;
2179 
2180 	memcpy(F2FS_NODE(xpage), F2FS_NODE(page), PAGE_SIZE);
2181 
2182 	get_node_info(sbi, new_xnid, &ni);
2183 	ni.ino = inode->i_ino;
2184 	set_node_addr(sbi, &ni, NEW_ADDR, false);
2185 	set_page_dirty(xpage);
2186 	f2fs_put_page(xpage, 1);
2187 
2188 	return 0;
2189 }
2190 
2191 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2192 {
2193 	struct f2fs_inode *src, *dst;
2194 	nid_t ino = ino_of_node(page);
2195 	struct node_info old_ni, new_ni;
2196 	struct page *ipage;
2197 
2198 	get_node_info(sbi, ino, &old_ni);
2199 
2200 	if (unlikely(old_ni.blk_addr != NULL_ADDR))
2201 		return -EINVAL;
2202 retry:
2203 	ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2204 	if (!ipage) {
2205 		congestion_wait(BLK_RW_ASYNC, HZ/50);
2206 		goto retry;
2207 	}
2208 
2209 	/* Should not use this inode from free nid list */
2210 	remove_free_nid(sbi, ino);
2211 
2212 	if (!PageUptodate(ipage))
2213 		SetPageUptodate(ipage);
2214 	fill_node_footer(ipage, ino, ino, 0, true);
2215 
2216 	src = F2FS_INODE(page);
2217 	dst = F2FS_INODE(ipage);
2218 
2219 	memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2220 	dst->i_size = 0;
2221 	dst->i_blocks = cpu_to_le64(1);
2222 	dst->i_links = cpu_to_le32(1);
2223 	dst->i_xattr_nid = 0;
2224 	dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
2225 
2226 	new_ni = old_ni;
2227 	new_ni.ino = ino;
2228 
2229 	if (unlikely(!inc_valid_node_count(sbi, NULL)))
2230 		WARN_ON(1);
2231 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2232 	inc_valid_inode_count(sbi);
2233 	set_page_dirty(ipage);
2234 	f2fs_put_page(ipage, 1);
2235 	return 0;
2236 }
2237 
2238 int restore_node_summary(struct f2fs_sb_info *sbi,
2239 			unsigned int segno, struct f2fs_summary_block *sum)
2240 {
2241 	struct f2fs_node *rn;
2242 	struct f2fs_summary *sum_entry;
2243 	block_t addr;
2244 	int i, idx, last_offset, nrpages;
2245 
2246 	/* scan the node segment */
2247 	last_offset = sbi->blocks_per_seg;
2248 	addr = START_BLOCK(sbi, segno);
2249 	sum_entry = &sum->entries[0];
2250 
2251 	for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2252 		nrpages = min(last_offset - i, BIO_MAX_PAGES);
2253 
2254 		/* readahead node pages */
2255 		ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2256 
2257 		for (idx = addr; idx < addr + nrpages; idx++) {
2258 			struct page *page = get_tmp_page(sbi, idx);
2259 
2260 			rn = F2FS_NODE(page);
2261 			sum_entry->nid = rn->footer.nid;
2262 			sum_entry->version = 0;
2263 			sum_entry->ofs_in_node = 0;
2264 			sum_entry++;
2265 			f2fs_put_page(page, 1);
2266 		}
2267 
2268 		invalidate_mapping_pages(META_MAPPING(sbi), addr,
2269 							addr + nrpages);
2270 	}
2271 	return 0;
2272 }
2273 
2274 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2275 {
2276 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2277 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2278 	struct f2fs_journal *journal = curseg->journal;
2279 	int i;
2280 
2281 	down_write(&curseg->journal_rwsem);
2282 	for (i = 0; i < nats_in_cursum(journal); i++) {
2283 		struct nat_entry *ne;
2284 		struct f2fs_nat_entry raw_ne;
2285 		nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2286 
2287 		raw_ne = nat_in_journal(journal, i);
2288 
2289 		ne = __lookup_nat_cache(nm_i, nid);
2290 		if (!ne) {
2291 			ne = grab_nat_entry(nm_i, nid, true);
2292 			node_info_from_raw_nat(&ne->ni, &raw_ne);
2293 		}
2294 
2295 		/*
2296 		 * if a free nat in journal has not been used after last
2297 		 * checkpoint, we should remove it from available nids,
2298 		 * since later we will add it again.
2299 		 */
2300 		if (!get_nat_flag(ne, IS_DIRTY) &&
2301 				le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2302 			spin_lock(&nm_i->nid_list_lock);
2303 			nm_i->available_nids--;
2304 			spin_unlock(&nm_i->nid_list_lock);
2305 		}
2306 
2307 		__set_nat_cache_dirty(nm_i, ne);
2308 	}
2309 	update_nats_in_cursum(journal, -i);
2310 	up_write(&curseg->journal_rwsem);
2311 }
2312 
2313 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2314 						struct list_head *head, int max)
2315 {
2316 	struct nat_entry_set *cur;
2317 
2318 	if (nes->entry_cnt >= max)
2319 		goto add_out;
2320 
2321 	list_for_each_entry(cur, head, set_list) {
2322 		if (cur->entry_cnt >= nes->entry_cnt) {
2323 			list_add(&nes->set_list, cur->set_list.prev);
2324 			return;
2325 		}
2326 	}
2327 add_out:
2328 	list_add_tail(&nes->set_list, head);
2329 }
2330 
2331 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2332 						struct page *page)
2333 {
2334 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2335 	unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2336 	struct f2fs_nat_block *nat_blk = page_address(page);
2337 	int valid = 0;
2338 	int i;
2339 
2340 	if (!enabled_nat_bits(sbi, NULL))
2341 		return;
2342 
2343 	for (i = 0; i < NAT_ENTRY_PER_BLOCK; i++) {
2344 		if (start_nid == 0 && i == 0)
2345 			valid++;
2346 		if (nat_blk->entries[i].block_addr)
2347 			valid++;
2348 	}
2349 	if (valid == 0) {
2350 		__set_bit_le(nat_index, nm_i->empty_nat_bits);
2351 		__clear_bit_le(nat_index, nm_i->full_nat_bits);
2352 		return;
2353 	}
2354 
2355 	__clear_bit_le(nat_index, nm_i->empty_nat_bits);
2356 	if (valid == NAT_ENTRY_PER_BLOCK)
2357 		__set_bit_le(nat_index, nm_i->full_nat_bits);
2358 	else
2359 		__clear_bit_le(nat_index, nm_i->full_nat_bits);
2360 }
2361 
2362 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2363 		struct nat_entry_set *set, struct cp_control *cpc)
2364 {
2365 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2366 	struct f2fs_journal *journal = curseg->journal;
2367 	nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2368 	bool to_journal = true;
2369 	struct f2fs_nat_block *nat_blk;
2370 	struct nat_entry *ne, *cur;
2371 	struct page *page = NULL;
2372 
2373 	/*
2374 	 * there are two steps to flush nat entries:
2375 	 * #1, flush nat entries to journal in current hot data summary block.
2376 	 * #2, flush nat entries to nat page.
2377 	 */
2378 	if (enabled_nat_bits(sbi, cpc) ||
2379 		!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2380 		to_journal = false;
2381 
2382 	if (to_journal) {
2383 		down_write(&curseg->journal_rwsem);
2384 	} else {
2385 		page = get_next_nat_page(sbi, start_nid);
2386 		nat_blk = page_address(page);
2387 		f2fs_bug_on(sbi, !nat_blk);
2388 	}
2389 
2390 	/* flush dirty nats in nat entry set */
2391 	list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2392 		struct f2fs_nat_entry *raw_ne;
2393 		nid_t nid = nat_get_nid(ne);
2394 		int offset;
2395 
2396 		if (nat_get_blkaddr(ne) == NEW_ADDR)
2397 			continue;
2398 
2399 		if (to_journal) {
2400 			offset = lookup_journal_in_cursum(journal,
2401 							NAT_JOURNAL, nid, 1);
2402 			f2fs_bug_on(sbi, offset < 0);
2403 			raw_ne = &nat_in_journal(journal, offset);
2404 			nid_in_journal(journal, offset) = cpu_to_le32(nid);
2405 		} else {
2406 			raw_ne = &nat_blk->entries[nid - start_nid];
2407 		}
2408 		raw_nat_from_node_info(raw_ne, &ne->ni);
2409 		nat_reset_flag(ne);
2410 		__clear_nat_cache_dirty(NM_I(sbi), ne);
2411 		if (nat_get_blkaddr(ne) == NULL_ADDR) {
2412 			add_free_nid(sbi, nid, false);
2413 			spin_lock(&NM_I(sbi)->nid_list_lock);
2414 			NM_I(sbi)->available_nids++;
2415 			update_free_nid_bitmap(sbi, nid, true, false, false);
2416 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2417 		} else {
2418 			spin_lock(&NM_I(sbi)->nid_list_lock);
2419 			update_free_nid_bitmap(sbi, nid, false, false, false);
2420 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2421 		}
2422 	}
2423 
2424 	if (to_journal) {
2425 		up_write(&curseg->journal_rwsem);
2426 	} else {
2427 		__update_nat_bits(sbi, start_nid, page);
2428 		f2fs_put_page(page, 1);
2429 	}
2430 
2431 	f2fs_bug_on(sbi, set->entry_cnt);
2432 
2433 	radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2434 	kmem_cache_free(nat_entry_set_slab, set);
2435 }
2436 
2437 /*
2438  * This function is called during the checkpointing process.
2439  */
2440 void flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2441 {
2442 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2443 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2444 	struct f2fs_journal *journal = curseg->journal;
2445 	struct nat_entry_set *setvec[SETVEC_SIZE];
2446 	struct nat_entry_set *set, *tmp;
2447 	unsigned int found;
2448 	nid_t set_idx = 0;
2449 	LIST_HEAD(sets);
2450 
2451 	if (!nm_i->dirty_nat_cnt)
2452 		return;
2453 
2454 	down_write(&nm_i->nat_tree_lock);
2455 
2456 	/*
2457 	 * if there are no enough space in journal to store dirty nat
2458 	 * entries, remove all entries from journal and merge them
2459 	 * into nat entry set.
2460 	 */
2461 	if (enabled_nat_bits(sbi, cpc) ||
2462 		!__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2463 		remove_nats_in_journal(sbi);
2464 
2465 	while ((found = __gang_lookup_nat_set(nm_i,
2466 					set_idx, SETVEC_SIZE, setvec))) {
2467 		unsigned idx;
2468 		set_idx = setvec[found - 1]->set + 1;
2469 		for (idx = 0; idx < found; idx++)
2470 			__adjust_nat_entry_set(setvec[idx], &sets,
2471 						MAX_NAT_JENTRIES(journal));
2472 	}
2473 
2474 	/* flush dirty nats in nat entry set */
2475 	list_for_each_entry_safe(set, tmp, &sets, set_list)
2476 		__flush_nat_entry_set(sbi, set, cpc);
2477 
2478 	up_write(&nm_i->nat_tree_lock);
2479 
2480 	f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2481 }
2482 
2483 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2484 {
2485 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2486 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2487 	unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2488 	unsigned int i;
2489 	__u64 cp_ver = cur_cp_version(ckpt);
2490 	block_t nat_bits_addr;
2491 
2492 	if (!enabled_nat_bits(sbi, NULL))
2493 		return 0;
2494 
2495 	nm_i->nat_bits_blocks = F2FS_BYTES_TO_BLK((nat_bits_bytes << 1) + 8 +
2496 						F2FS_BLKSIZE - 1);
2497 	nm_i->nat_bits = kzalloc(nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS,
2498 						GFP_KERNEL);
2499 	if (!nm_i->nat_bits)
2500 		return -ENOMEM;
2501 
2502 	nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2503 						nm_i->nat_bits_blocks;
2504 	for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2505 		struct page *page = get_meta_page(sbi, nat_bits_addr++);
2506 
2507 		memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2508 					page_address(page), F2FS_BLKSIZE);
2509 		f2fs_put_page(page, 1);
2510 	}
2511 
2512 	cp_ver |= (cur_cp_crc(ckpt) << 32);
2513 	if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2514 		disable_nat_bits(sbi, true);
2515 		return 0;
2516 	}
2517 
2518 	nm_i->full_nat_bits = nm_i->nat_bits + 8;
2519 	nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2520 
2521 	f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint");
2522 	return 0;
2523 }
2524 
2525 inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2526 {
2527 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2528 	unsigned int i = 0;
2529 	nid_t nid, last_nid;
2530 
2531 	if (!enabled_nat_bits(sbi, NULL))
2532 		return;
2533 
2534 	for (i = 0; i < nm_i->nat_blocks; i++) {
2535 		i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2536 		if (i >= nm_i->nat_blocks)
2537 			break;
2538 
2539 		__set_bit_le(i, nm_i->nat_block_bitmap);
2540 
2541 		nid = i * NAT_ENTRY_PER_BLOCK;
2542 		last_nid = (i + 1) * NAT_ENTRY_PER_BLOCK;
2543 
2544 		spin_lock(&nm_i->free_nid_lock);
2545 		for (; nid < last_nid; nid++)
2546 			update_free_nid_bitmap(sbi, nid, true, true, true);
2547 		spin_unlock(&nm_i->free_nid_lock);
2548 	}
2549 
2550 	for (i = 0; i < nm_i->nat_blocks; i++) {
2551 		i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2552 		if (i >= nm_i->nat_blocks)
2553 			break;
2554 
2555 		__set_bit_le(i, nm_i->nat_block_bitmap);
2556 	}
2557 }
2558 
2559 static int init_node_manager(struct f2fs_sb_info *sbi)
2560 {
2561 	struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2562 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2563 	unsigned char *version_bitmap;
2564 	unsigned int nat_segs;
2565 	int err;
2566 
2567 	nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2568 
2569 	/* segment_count_nat includes pair segment so divide to 2. */
2570 	nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2571 	nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2572 	nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
2573 
2574 	/* not used nids: 0, node, meta, (and root counted as valid node) */
2575 	nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2576 							F2FS_RESERVED_NODE_NUM;
2577 	nm_i->nid_cnt[FREE_NID_LIST] = 0;
2578 	nm_i->nid_cnt[ALLOC_NID_LIST] = 0;
2579 	nm_i->nat_cnt = 0;
2580 	nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2581 	nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2582 	nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2583 
2584 	INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2585 	INIT_LIST_HEAD(&nm_i->nid_list[FREE_NID_LIST]);
2586 	INIT_LIST_HEAD(&nm_i->nid_list[ALLOC_NID_LIST]);
2587 	INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2588 	INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2589 	INIT_LIST_HEAD(&nm_i->nat_entries);
2590 
2591 	mutex_init(&nm_i->build_lock);
2592 	spin_lock_init(&nm_i->nid_list_lock);
2593 	init_rwsem(&nm_i->nat_tree_lock);
2594 
2595 	nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2596 	nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2597 	version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2598 	if (!version_bitmap)
2599 		return -EFAULT;
2600 
2601 	nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2602 					GFP_KERNEL);
2603 	if (!nm_i->nat_bitmap)
2604 		return -ENOMEM;
2605 
2606 	err = __get_nat_bitmaps(sbi);
2607 	if (err)
2608 		return err;
2609 
2610 #ifdef CONFIG_F2FS_CHECK_FS
2611 	nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
2612 					GFP_KERNEL);
2613 	if (!nm_i->nat_bitmap_mir)
2614 		return -ENOMEM;
2615 #endif
2616 
2617 	return 0;
2618 }
2619 
2620 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
2621 {
2622 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2623 
2624 	nm_i->free_nid_bitmap = f2fs_kvzalloc(nm_i->nat_blocks *
2625 					NAT_ENTRY_BITMAP_SIZE, GFP_KERNEL);
2626 	if (!nm_i->free_nid_bitmap)
2627 		return -ENOMEM;
2628 
2629 	nm_i->nat_block_bitmap = f2fs_kvzalloc(nm_i->nat_blocks / 8,
2630 								GFP_KERNEL);
2631 	if (!nm_i->nat_block_bitmap)
2632 		return -ENOMEM;
2633 
2634 	nm_i->free_nid_count = f2fs_kvzalloc(nm_i->nat_blocks *
2635 					sizeof(unsigned short), GFP_KERNEL);
2636 	if (!nm_i->free_nid_count)
2637 		return -ENOMEM;
2638 
2639 	spin_lock_init(&nm_i->free_nid_lock);
2640 
2641 	return 0;
2642 }
2643 
2644 int build_node_manager(struct f2fs_sb_info *sbi)
2645 {
2646 	int err;
2647 
2648 	sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2649 	if (!sbi->nm_info)
2650 		return -ENOMEM;
2651 
2652 	err = init_node_manager(sbi);
2653 	if (err)
2654 		return err;
2655 
2656 	err = init_free_nid_cache(sbi);
2657 	if (err)
2658 		return err;
2659 
2660 	/* load free nid status from nat_bits table */
2661 	load_free_nid_bitmap(sbi);
2662 
2663 	build_free_nids(sbi, true, true);
2664 	return 0;
2665 }
2666 
2667 void destroy_node_manager(struct f2fs_sb_info *sbi)
2668 {
2669 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2670 	struct free_nid *i, *next_i;
2671 	struct nat_entry *natvec[NATVEC_SIZE];
2672 	struct nat_entry_set *setvec[SETVEC_SIZE];
2673 	nid_t nid = 0;
2674 	unsigned int found;
2675 
2676 	if (!nm_i)
2677 		return;
2678 
2679 	/* destroy free nid list */
2680 	spin_lock(&nm_i->nid_list_lock);
2681 	list_for_each_entry_safe(i, next_i, &nm_i->nid_list[FREE_NID_LIST],
2682 									list) {
2683 		__remove_nid_from_list(sbi, i, FREE_NID_LIST, false);
2684 		spin_unlock(&nm_i->nid_list_lock);
2685 		kmem_cache_free(free_nid_slab, i);
2686 		spin_lock(&nm_i->nid_list_lock);
2687 	}
2688 	f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID_LIST]);
2689 	f2fs_bug_on(sbi, nm_i->nid_cnt[ALLOC_NID_LIST]);
2690 	f2fs_bug_on(sbi, !list_empty(&nm_i->nid_list[ALLOC_NID_LIST]));
2691 	spin_unlock(&nm_i->nid_list_lock);
2692 
2693 	/* destroy nat cache */
2694 	down_write(&nm_i->nat_tree_lock);
2695 	while ((found = __gang_lookup_nat_cache(nm_i,
2696 					nid, NATVEC_SIZE, natvec))) {
2697 		unsigned idx;
2698 
2699 		nid = nat_get_nid(natvec[found - 1]) + 1;
2700 		for (idx = 0; idx < found; idx++)
2701 			__del_from_nat_cache(nm_i, natvec[idx]);
2702 	}
2703 	f2fs_bug_on(sbi, nm_i->nat_cnt);
2704 
2705 	/* destroy nat set cache */
2706 	nid = 0;
2707 	while ((found = __gang_lookup_nat_set(nm_i,
2708 					nid, SETVEC_SIZE, setvec))) {
2709 		unsigned idx;
2710 
2711 		nid = setvec[found - 1]->set + 1;
2712 		for (idx = 0; idx < found; idx++) {
2713 			/* entry_cnt is not zero, when cp_error was occurred */
2714 			f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2715 			radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2716 			kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2717 		}
2718 	}
2719 	up_write(&nm_i->nat_tree_lock);
2720 
2721 	kvfree(nm_i->nat_block_bitmap);
2722 	kvfree(nm_i->free_nid_bitmap);
2723 	kvfree(nm_i->free_nid_count);
2724 
2725 	kfree(nm_i->nat_bitmap);
2726 	kfree(nm_i->nat_bits);
2727 #ifdef CONFIG_F2FS_CHECK_FS
2728 	kfree(nm_i->nat_bitmap_mir);
2729 #endif
2730 	sbi->nm_info = NULL;
2731 	kfree(nm_i);
2732 }
2733 
2734 int __init create_node_manager_caches(void)
2735 {
2736 	nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2737 			sizeof(struct nat_entry));
2738 	if (!nat_entry_slab)
2739 		goto fail;
2740 
2741 	free_nid_slab = f2fs_kmem_cache_create("free_nid",
2742 			sizeof(struct free_nid));
2743 	if (!free_nid_slab)
2744 		goto destroy_nat_entry;
2745 
2746 	nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2747 			sizeof(struct nat_entry_set));
2748 	if (!nat_entry_set_slab)
2749 		goto destroy_free_nid;
2750 	return 0;
2751 
2752 destroy_free_nid:
2753 	kmem_cache_destroy(free_nid_slab);
2754 destroy_nat_entry:
2755 	kmem_cache_destroy(nat_entry_slab);
2756 fail:
2757 	return -ENOMEM;
2758 }
2759 
2760 void destroy_node_manager_caches(void)
2761 {
2762 	kmem_cache_destroy(nat_entry_set_slab);
2763 	kmem_cache_destroy(free_nid_slab);
2764 	kmem_cache_destroy(nat_entry_slab);
2765 }
2766