xref: /linux/fs/f2fs/node.c (revision 2ba9268dd603d23e17643437b2246acb6844953b)
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 	/* give 25%, 25%, 50%, 50% memory for each components respectively */
45 	if (type == FREE_NIDS) {
46 		mem_size = (nm_i->fcnt * sizeof(struct free_nid)) >>
47 							PAGE_CACHE_SHIFT;
48 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
49 	} else if (type == NAT_ENTRIES) {
50 		mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
51 							PAGE_CACHE_SHIFT;
52 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
53 	} else if (type == DIRTY_DENTS) {
54 		if (sbi->sb->s_bdi->dirty_exceeded)
55 			return false;
56 		mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
57 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
58 	} else if (type == INO_ENTRIES) {
59 		int i;
60 
61 		for (i = 0; i <= UPDATE_INO; i++)
62 			mem_size += (sbi->im[i].ino_num *
63 				sizeof(struct ino_entry)) >> PAGE_CACHE_SHIFT;
64 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
65 	} else {
66 		if (sbi->sb->s_bdi->dirty_exceeded)
67 			return false;
68 	}
69 	return res;
70 }
71 
72 static void clear_node_page_dirty(struct page *page)
73 {
74 	struct address_space *mapping = page->mapping;
75 	unsigned int long flags;
76 
77 	if (PageDirty(page)) {
78 		spin_lock_irqsave(&mapping->tree_lock, flags);
79 		radix_tree_tag_clear(&mapping->page_tree,
80 				page_index(page),
81 				PAGECACHE_TAG_DIRTY);
82 		spin_unlock_irqrestore(&mapping->tree_lock, flags);
83 
84 		clear_page_dirty_for_io(page);
85 		dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
86 	}
87 	ClearPageUptodate(page);
88 }
89 
90 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
91 {
92 	pgoff_t index = current_nat_addr(sbi, nid);
93 	return get_meta_page(sbi, index);
94 }
95 
96 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
97 {
98 	struct page *src_page;
99 	struct page *dst_page;
100 	pgoff_t src_off;
101 	pgoff_t dst_off;
102 	void *src_addr;
103 	void *dst_addr;
104 	struct f2fs_nm_info *nm_i = NM_I(sbi);
105 
106 	src_off = current_nat_addr(sbi, nid);
107 	dst_off = next_nat_addr(sbi, src_off);
108 
109 	/* get current nat block page with lock */
110 	src_page = get_meta_page(sbi, src_off);
111 	dst_page = grab_meta_page(sbi, dst_off);
112 	f2fs_bug_on(sbi, PageDirty(src_page));
113 
114 	src_addr = page_address(src_page);
115 	dst_addr = page_address(dst_page);
116 	memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
117 	set_page_dirty(dst_page);
118 	f2fs_put_page(src_page, 1);
119 
120 	set_to_next_nat(nm_i, nid);
121 
122 	return dst_page;
123 }
124 
125 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
126 {
127 	return radix_tree_lookup(&nm_i->nat_root, n);
128 }
129 
130 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
131 		nid_t start, unsigned int nr, struct nat_entry **ep)
132 {
133 	return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
134 }
135 
136 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
137 {
138 	list_del(&e->list);
139 	radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
140 	nm_i->nat_cnt--;
141 	kmem_cache_free(nat_entry_slab, e);
142 }
143 
144 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
145 						struct nat_entry *ne)
146 {
147 	nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
148 	struct nat_entry_set *head;
149 
150 	if (get_nat_flag(ne, IS_DIRTY))
151 		return;
152 
153 	head = radix_tree_lookup(&nm_i->nat_set_root, set);
154 	if (!head) {
155 		head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_ATOMIC);
156 
157 		INIT_LIST_HEAD(&head->entry_list);
158 		INIT_LIST_HEAD(&head->set_list);
159 		head->set = set;
160 		head->entry_cnt = 0;
161 		f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
162 	}
163 	list_move_tail(&ne->list, &head->entry_list);
164 	nm_i->dirty_nat_cnt++;
165 	head->entry_cnt++;
166 	set_nat_flag(ne, IS_DIRTY, true);
167 }
168 
169 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
170 						struct nat_entry *ne)
171 {
172 	nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
173 	struct nat_entry_set *head;
174 
175 	head = radix_tree_lookup(&nm_i->nat_set_root, set);
176 	if (head) {
177 		list_move_tail(&ne->list, &nm_i->nat_entries);
178 		set_nat_flag(ne, IS_DIRTY, false);
179 		head->entry_cnt--;
180 		nm_i->dirty_nat_cnt--;
181 	}
182 }
183 
184 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
185 		nid_t start, unsigned int nr, struct nat_entry_set **ep)
186 {
187 	return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
188 							start, nr);
189 }
190 
191 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
192 {
193 	struct f2fs_nm_info *nm_i = NM_I(sbi);
194 	struct nat_entry *e;
195 	bool is_cp = true;
196 
197 	down_read(&nm_i->nat_tree_lock);
198 	e = __lookup_nat_cache(nm_i, nid);
199 	if (e && !get_nat_flag(e, IS_CHECKPOINTED))
200 		is_cp = false;
201 	up_read(&nm_i->nat_tree_lock);
202 	return is_cp;
203 }
204 
205 bool has_fsynced_inode(struct f2fs_sb_info *sbi, nid_t ino)
206 {
207 	struct f2fs_nm_info *nm_i = NM_I(sbi);
208 	struct nat_entry *e;
209 	bool fsynced = false;
210 
211 	down_read(&nm_i->nat_tree_lock);
212 	e = __lookup_nat_cache(nm_i, ino);
213 	if (e && get_nat_flag(e, HAS_FSYNCED_INODE))
214 		fsynced = true;
215 	up_read(&nm_i->nat_tree_lock);
216 	return fsynced;
217 }
218 
219 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
220 {
221 	struct f2fs_nm_info *nm_i = NM_I(sbi);
222 	struct nat_entry *e;
223 	bool need_update = true;
224 
225 	down_read(&nm_i->nat_tree_lock);
226 	e = __lookup_nat_cache(nm_i, ino);
227 	if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
228 			(get_nat_flag(e, IS_CHECKPOINTED) ||
229 			 get_nat_flag(e, HAS_FSYNCED_INODE)))
230 		need_update = false;
231 	up_read(&nm_i->nat_tree_lock);
232 	return need_update;
233 }
234 
235 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
236 {
237 	struct nat_entry *new;
238 
239 	new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
240 	f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
241 	memset(new, 0, sizeof(struct nat_entry));
242 	nat_set_nid(new, nid);
243 	nat_reset_flag(new);
244 	list_add_tail(&new->list, &nm_i->nat_entries);
245 	nm_i->nat_cnt++;
246 	return new;
247 }
248 
249 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
250 						struct f2fs_nat_entry *ne)
251 {
252 	struct nat_entry *e;
253 
254 	down_write(&nm_i->nat_tree_lock);
255 	e = __lookup_nat_cache(nm_i, nid);
256 	if (!e) {
257 		e = grab_nat_entry(nm_i, nid);
258 		node_info_from_raw_nat(&e->ni, ne);
259 	}
260 	up_write(&nm_i->nat_tree_lock);
261 }
262 
263 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
264 			block_t new_blkaddr, bool fsync_done)
265 {
266 	struct f2fs_nm_info *nm_i = NM_I(sbi);
267 	struct nat_entry *e;
268 
269 	down_write(&nm_i->nat_tree_lock);
270 	e = __lookup_nat_cache(nm_i, ni->nid);
271 	if (!e) {
272 		e = grab_nat_entry(nm_i, ni->nid);
273 		copy_node_info(&e->ni, ni);
274 		f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
275 	} else if (new_blkaddr == NEW_ADDR) {
276 		/*
277 		 * when nid is reallocated,
278 		 * previous nat entry can be remained in nat cache.
279 		 * So, reinitialize it with new information.
280 		 */
281 		copy_node_info(&e->ni, ni);
282 		f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
283 	}
284 
285 	/* sanity check */
286 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
287 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
288 			new_blkaddr == NULL_ADDR);
289 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
290 			new_blkaddr == NEW_ADDR);
291 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
292 			nat_get_blkaddr(e) != NULL_ADDR &&
293 			new_blkaddr == NEW_ADDR);
294 
295 	/* increment version no as node is removed */
296 	if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
297 		unsigned char version = nat_get_version(e);
298 		nat_set_version(e, inc_node_version(version));
299 	}
300 
301 	/* change address */
302 	nat_set_blkaddr(e, new_blkaddr);
303 	if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
304 		set_nat_flag(e, IS_CHECKPOINTED, false);
305 	__set_nat_cache_dirty(nm_i, e);
306 
307 	/* update fsync_mark if its inode nat entry is still alive */
308 	e = __lookup_nat_cache(nm_i, ni->ino);
309 	if (e) {
310 		if (fsync_done && ni->nid == ni->ino)
311 			set_nat_flag(e, HAS_FSYNCED_INODE, true);
312 		set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
313 	}
314 	up_write(&nm_i->nat_tree_lock);
315 }
316 
317 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
318 {
319 	struct f2fs_nm_info *nm_i = NM_I(sbi);
320 
321 	if (available_free_memory(sbi, NAT_ENTRIES))
322 		return 0;
323 
324 	down_write(&nm_i->nat_tree_lock);
325 	while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
326 		struct nat_entry *ne;
327 		ne = list_first_entry(&nm_i->nat_entries,
328 					struct nat_entry, list);
329 		__del_from_nat_cache(nm_i, ne);
330 		nr_shrink--;
331 	}
332 	up_write(&nm_i->nat_tree_lock);
333 	return nr_shrink;
334 }
335 
336 /*
337  * This function always returns success
338  */
339 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
340 {
341 	struct f2fs_nm_info *nm_i = NM_I(sbi);
342 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
343 	struct f2fs_summary_block *sum = curseg->sum_blk;
344 	nid_t start_nid = START_NID(nid);
345 	struct f2fs_nat_block *nat_blk;
346 	struct page *page = NULL;
347 	struct f2fs_nat_entry ne;
348 	struct nat_entry *e;
349 	int i;
350 
351 	ni->nid = nid;
352 
353 	/* Check nat cache */
354 	down_read(&nm_i->nat_tree_lock);
355 	e = __lookup_nat_cache(nm_i, nid);
356 	if (e) {
357 		ni->ino = nat_get_ino(e);
358 		ni->blk_addr = nat_get_blkaddr(e);
359 		ni->version = nat_get_version(e);
360 	}
361 	up_read(&nm_i->nat_tree_lock);
362 	if (e)
363 		return;
364 
365 	memset(&ne, 0, sizeof(struct f2fs_nat_entry));
366 
367 	/* Check current segment summary */
368 	mutex_lock(&curseg->curseg_mutex);
369 	i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
370 	if (i >= 0) {
371 		ne = nat_in_journal(sum, i);
372 		node_info_from_raw_nat(ni, &ne);
373 	}
374 	mutex_unlock(&curseg->curseg_mutex);
375 	if (i >= 0)
376 		goto cache;
377 
378 	/* Fill node_info from nat page */
379 	page = get_current_nat_page(sbi, start_nid);
380 	nat_blk = (struct f2fs_nat_block *)page_address(page);
381 	ne = nat_blk->entries[nid - start_nid];
382 	node_info_from_raw_nat(ni, &ne);
383 	f2fs_put_page(page, 1);
384 cache:
385 	/* cache nat entry */
386 	cache_nat_entry(NM_I(sbi), nid, &ne);
387 }
388 
389 /*
390  * The maximum depth is four.
391  * Offset[0] will have raw inode offset.
392  */
393 static int get_node_path(struct f2fs_inode_info *fi, long block,
394 				int offset[4], unsigned int noffset[4])
395 {
396 	const long direct_index = ADDRS_PER_INODE(fi);
397 	const long direct_blks = ADDRS_PER_BLOCK;
398 	const long dptrs_per_blk = NIDS_PER_BLOCK;
399 	const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
400 	const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
401 	int n = 0;
402 	int level = 0;
403 
404 	noffset[0] = 0;
405 
406 	if (block < direct_index) {
407 		offset[n] = block;
408 		goto got;
409 	}
410 	block -= direct_index;
411 	if (block < direct_blks) {
412 		offset[n++] = NODE_DIR1_BLOCK;
413 		noffset[n] = 1;
414 		offset[n] = block;
415 		level = 1;
416 		goto got;
417 	}
418 	block -= direct_blks;
419 	if (block < direct_blks) {
420 		offset[n++] = NODE_DIR2_BLOCK;
421 		noffset[n] = 2;
422 		offset[n] = block;
423 		level = 1;
424 		goto got;
425 	}
426 	block -= direct_blks;
427 	if (block < indirect_blks) {
428 		offset[n++] = NODE_IND1_BLOCK;
429 		noffset[n] = 3;
430 		offset[n++] = block / direct_blks;
431 		noffset[n] = 4 + offset[n - 1];
432 		offset[n] = block % direct_blks;
433 		level = 2;
434 		goto got;
435 	}
436 	block -= indirect_blks;
437 	if (block < indirect_blks) {
438 		offset[n++] = NODE_IND2_BLOCK;
439 		noffset[n] = 4 + dptrs_per_blk;
440 		offset[n++] = block / direct_blks;
441 		noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
442 		offset[n] = block % direct_blks;
443 		level = 2;
444 		goto got;
445 	}
446 	block -= indirect_blks;
447 	if (block < dindirect_blks) {
448 		offset[n++] = NODE_DIND_BLOCK;
449 		noffset[n] = 5 + (dptrs_per_blk * 2);
450 		offset[n++] = block / indirect_blks;
451 		noffset[n] = 6 + (dptrs_per_blk * 2) +
452 			      offset[n - 1] * (dptrs_per_blk + 1);
453 		offset[n++] = (block / direct_blks) % dptrs_per_blk;
454 		noffset[n] = 7 + (dptrs_per_blk * 2) +
455 			      offset[n - 2] * (dptrs_per_blk + 1) +
456 			      offset[n - 1];
457 		offset[n] = block % direct_blks;
458 		level = 3;
459 		goto got;
460 	} else {
461 		BUG();
462 	}
463 got:
464 	return level;
465 }
466 
467 /*
468  * Caller should call f2fs_put_dnode(dn).
469  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
470  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
471  * In the case of RDONLY_NODE, we don't need to care about mutex.
472  */
473 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
474 {
475 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
476 	struct page *npage[4];
477 	struct page *parent = NULL;
478 	int offset[4];
479 	unsigned int noffset[4];
480 	nid_t nids[4];
481 	int level, i;
482 	int err = 0;
483 
484 	level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
485 
486 	nids[0] = dn->inode->i_ino;
487 	npage[0] = dn->inode_page;
488 
489 	if (!npage[0]) {
490 		npage[0] = get_node_page(sbi, nids[0]);
491 		if (IS_ERR(npage[0]))
492 			return PTR_ERR(npage[0]);
493 	}
494 
495 	/* if inline_data is set, should not report any block indices */
496 	if (f2fs_has_inline_data(dn->inode) && index) {
497 		err = -EINVAL;
498 		f2fs_put_page(npage[0], 1);
499 		goto release_out;
500 	}
501 
502 	parent = npage[0];
503 	if (level != 0)
504 		nids[1] = get_nid(parent, offset[0], true);
505 	dn->inode_page = npage[0];
506 	dn->inode_page_locked = true;
507 
508 	/* get indirect or direct nodes */
509 	for (i = 1; i <= level; i++) {
510 		bool done = false;
511 
512 		if (!nids[i] && mode == ALLOC_NODE) {
513 			/* alloc new node */
514 			if (!alloc_nid(sbi, &(nids[i]))) {
515 				err = -ENOSPC;
516 				goto release_pages;
517 			}
518 
519 			dn->nid = nids[i];
520 			npage[i] = new_node_page(dn, noffset[i], NULL);
521 			if (IS_ERR(npage[i])) {
522 				alloc_nid_failed(sbi, nids[i]);
523 				err = PTR_ERR(npage[i]);
524 				goto release_pages;
525 			}
526 
527 			set_nid(parent, offset[i - 1], nids[i], i == 1);
528 			alloc_nid_done(sbi, nids[i]);
529 			done = true;
530 		} else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
531 			npage[i] = get_node_page_ra(parent, offset[i - 1]);
532 			if (IS_ERR(npage[i])) {
533 				err = PTR_ERR(npage[i]);
534 				goto release_pages;
535 			}
536 			done = true;
537 		}
538 		if (i == 1) {
539 			dn->inode_page_locked = false;
540 			unlock_page(parent);
541 		} else {
542 			f2fs_put_page(parent, 1);
543 		}
544 
545 		if (!done) {
546 			npage[i] = get_node_page(sbi, nids[i]);
547 			if (IS_ERR(npage[i])) {
548 				err = PTR_ERR(npage[i]);
549 				f2fs_put_page(npage[0], 0);
550 				goto release_out;
551 			}
552 		}
553 		if (i < level) {
554 			parent = npage[i];
555 			nids[i + 1] = get_nid(parent, offset[i], false);
556 		}
557 	}
558 	dn->nid = nids[level];
559 	dn->ofs_in_node = offset[level];
560 	dn->node_page = npage[level];
561 	dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
562 	return 0;
563 
564 release_pages:
565 	f2fs_put_page(parent, 1);
566 	if (i > 1)
567 		f2fs_put_page(npage[0], 0);
568 release_out:
569 	dn->inode_page = NULL;
570 	dn->node_page = NULL;
571 	return err;
572 }
573 
574 static void truncate_node(struct dnode_of_data *dn)
575 {
576 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
577 	struct node_info ni;
578 
579 	get_node_info(sbi, dn->nid, &ni);
580 	if (dn->inode->i_blocks == 0) {
581 		f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
582 		goto invalidate;
583 	}
584 	f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
585 
586 	/* Deallocate node address */
587 	invalidate_blocks(sbi, ni.blk_addr);
588 	dec_valid_node_count(sbi, dn->inode);
589 	set_node_addr(sbi, &ni, NULL_ADDR, false);
590 
591 	if (dn->nid == dn->inode->i_ino) {
592 		remove_orphan_inode(sbi, dn->nid);
593 		dec_valid_inode_count(sbi);
594 	} else {
595 		sync_inode_page(dn);
596 	}
597 invalidate:
598 	clear_node_page_dirty(dn->node_page);
599 	set_sbi_flag(sbi, SBI_IS_DIRTY);
600 
601 	f2fs_put_page(dn->node_page, 1);
602 
603 	invalidate_mapping_pages(NODE_MAPPING(sbi),
604 			dn->node_page->index, dn->node_page->index);
605 
606 	dn->node_page = NULL;
607 	trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
608 }
609 
610 static int truncate_dnode(struct dnode_of_data *dn)
611 {
612 	struct page *page;
613 
614 	if (dn->nid == 0)
615 		return 1;
616 
617 	/* get direct node */
618 	page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
619 	if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
620 		return 1;
621 	else if (IS_ERR(page))
622 		return PTR_ERR(page);
623 
624 	/* Make dnode_of_data for parameter */
625 	dn->node_page = page;
626 	dn->ofs_in_node = 0;
627 	truncate_data_blocks(dn);
628 	truncate_node(dn);
629 	return 1;
630 }
631 
632 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
633 						int ofs, int depth)
634 {
635 	struct dnode_of_data rdn = *dn;
636 	struct page *page;
637 	struct f2fs_node *rn;
638 	nid_t child_nid;
639 	unsigned int child_nofs;
640 	int freed = 0;
641 	int i, ret;
642 
643 	if (dn->nid == 0)
644 		return NIDS_PER_BLOCK + 1;
645 
646 	trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
647 
648 	page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
649 	if (IS_ERR(page)) {
650 		trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
651 		return PTR_ERR(page);
652 	}
653 
654 	rn = F2FS_NODE(page);
655 	if (depth < 3) {
656 		for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
657 			child_nid = le32_to_cpu(rn->in.nid[i]);
658 			if (child_nid == 0)
659 				continue;
660 			rdn.nid = child_nid;
661 			ret = truncate_dnode(&rdn);
662 			if (ret < 0)
663 				goto out_err;
664 			set_nid(page, i, 0, false);
665 		}
666 	} else {
667 		child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
668 		for (i = ofs; i < NIDS_PER_BLOCK; i++) {
669 			child_nid = le32_to_cpu(rn->in.nid[i]);
670 			if (child_nid == 0) {
671 				child_nofs += NIDS_PER_BLOCK + 1;
672 				continue;
673 			}
674 			rdn.nid = child_nid;
675 			ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
676 			if (ret == (NIDS_PER_BLOCK + 1)) {
677 				set_nid(page, i, 0, false);
678 				child_nofs += ret;
679 			} else if (ret < 0 && ret != -ENOENT) {
680 				goto out_err;
681 			}
682 		}
683 		freed = child_nofs;
684 	}
685 
686 	if (!ofs) {
687 		/* remove current indirect node */
688 		dn->node_page = page;
689 		truncate_node(dn);
690 		freed++;
691 	} else {
692 		f2fs_put_page(page, 1);
693 	}
694 	trace_f2fs_truncate_nodes_exit(dn->inode, freed);
695 	return freed;
696 
697 out_err:
698 	f2fs_put_page(page, 1);
699 	trace_f2fs_truncate_nodes_exit(dn->inode, ret);
700 	return ret;
701 }
702 
703 static int truncate_partial_nodes(struct dnode_of_data *dn,
704 			struct f2fs_inode *ri, int *offset, int depth)
705 {
706 	struct page *pages[2];
707 	nid_t nid[3];
708 	nid_t child_nid;
709 	int err = 0;
710 	int i;
711 	int idx = depth - 2;
712 
713 	nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
714 	if (!nid[0])
715 		return 0;
716 
717 	/* get indirect nodes in the path */
718 	for (i = 0; i < idx + 1; i++) {
719 		/* reference count'll be increased */
720 		pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
721 		if (IS_ERR(pages[i])) {
722 			err = PTR_ERR(pages[i]);
723 			idx = i - 1;
724 			goto fail;
725 		}
726 		nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
727 	}
728 
729 	/* free direct nodes linked to a partial indirect node */
730 	for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
731 		child_nid = get_nid(pages[idx], i, false);
732 		if (!child_nid)
733 			continue;
734 		dn->nid = child_nid;
735 		err = truncate_dnode(dn);
736 		if (err < 0)
737 			goto fail;
738 		set_nid(pages[idx], i, 0, false);
739 	}
740 
741 	if (offset[idx + 1] == 0) {
742 		dn->node_page = pages[idx];
743 		dn->nid = nid[idx];
744 		truncate_node(dn);
745 	} else {
746 		f2fs_put_page(pages[idx], 1);
747 	}
748 	offset[idx]++;
749 	offset[idx + 1] = 0;
750 	idx--;
751 fail:
752 	for (i = idx; i >= 0; i--)
753 		f2fs_put_page(pages[i], 1);
754 
755 	trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
756 
757 	return err;
758 }
759 
760 /*
761  * All the block addresses of data and nodes should be nullified.
762  */
763 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
764 {
765 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
766 	int err = 0, cont = 1;
767 	int level, offset[4], noffset[4];
768 	unsigned int nofs = 0;
769 	struct f2fs_inode *ri;
770 	struct dnode_of_data dn;
771 	struct page *page;
772 
773 	trace_f2fs_truncate_inode_blocks_enter(inode, from);
774 
775 	level = get_node_path(F2FS_I(inode), from, offset, noffset);
776 restart:
777 	page = get_node_page(sbi, inode->i_ino);
778 	if (IS_ERR(page)) {
779 		trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
780 		return PTR_ERR(page);
781 	}
782 
783 	set_new_dnode(&dn, inode, page, NULL, 0);
784 	unlock_page(page);
785 
786 	ri = F2FS_INODE(page);
787 	switch (level) {
788 	case 0:
789 	case 1:
790 		nofs = noffset[1];
791 		break;
792 	case 2:
793 		nofs = noffset[1];
794 		if (!offset[level - 1])
795 			goto skip_partial;
796 		err = truncate_partial_nodes(&dn, ri, offset, level);
797 		if (err < 0 && err != -ENOENT)
798 			goto fail;
799 		nofs += 1 + NIDS_PER_BLOCK;
800 		break;
801 	case 3:
802 		nofs = 5 + 2 * NIDS_PER_BLOCK;
803 		if (!offset[level - 1])
804 			goto skip_partial;
805 		err = truncate_partial_nodes(&dn, ri, offset, level);
806 		if (err < 0 && err != -ENOENT)
807 			goto fail;
808 		break;
809 	default:
810 		BUG();
811 	}
812 
813 skip_partial:
814 	while (cont) {
815 		dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
816 		switch (offset[0]) {
817 		case NODE_DIR1_BLOCK:
818 		case NODE_DIR2_BLOCK:
819 			err = truncate_dnode(&dn);
820 			break;
821 
822 		case NODE_IND1_BLOCK:
823 		case NODE_IND2_BLOCK:
824 			err = truncate_nodes(&dn, nofs, offset[1], 2);
825 			break;
826 
827 		case NODE_DIND_BLOCK:
828 			err = truncate_nodes(&dn, nofs, offset[1], 3);
829 			cont = 0;
830 			break;
831 
832 		default:
833 			BUG();
834 		}
835 		if (err < 0 && err != -ENOENT)
836 			goto fail;
837 		if (offset[1] == 0 &&
838 				ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
839 			lock_page(page);
840 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
841 				f2fs_put_page(page, 1);
842 				goto restart;
843 			}
844 			f2fs_wait_on_page_writeback(page, NODE);
845 			ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
846 			set_page_dirty(page);
847 			unlock_page(page);
848 		}
849 		offset[1] = 0;
850 		offset[0]++;
851 		nofs += err;
852 	}
853 fail:
854 	f2fs_put_page(page, 0);
855 	trace_f2fs_truncate_inode_blocks_exit(inode, err);
856 	return err > 0 ? 0 : err;
857 }
858 
859 int truncate_xattr_node(struct inode *inode, struct page *page)
860 {
861 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
862 	nid_t nid = F2FS_I(inode)->i_xattr_nid;
863 	struct dnode_of_data dn;
864 	struct page *npage;
865 
866 	if (!nid)
867 		return 0;
868 
869 	npage = get_node_page(sbi, nid);
870 	if (IS_ERR(npage))
871 		return PTR_ERR(npage);
872 
873 	F2FS_I(inode)->i_xattr_nid = 0;
874 
875 	/* need to do checkpoint during fsync */
876 	F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
877 
878 	set_new_dnode(&dn, inode, page, npage, nid);
879 
880 	if (page)
881 		dn.inode_page_locked = true;
882 	truncate_node(&dn);
883 	return 0;
884 }
885 
886 /*
887  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
888  * f2fs_unlock_op().
889  */
890 void remove_inode_page(struct inode *inode)
891 {
892 	struct dnode_of_data dn;
893 
894 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
895 	if (get_dnode_of_data(&dn, 0, LOOKUP_NODE))
896 		return;
897 
898 	if (truncate_xattr_node(inode, dn.inode_page)) {
899 		f2fs_put_dnode(&dn);
900 		return;
901 	}
902 
903 	/* remove potential inline_data blocks */
904 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
905 				S_ISLNK(inode->i_mode))
906 		truncate_data_blocks_range(&dn, 1);
907 
908 	/* 0 is possible, after f2fs_new_inode() has failed */
909 	f2fs_bug_on(F2FS_I_SB(inode),
910 			inode->i_blocks != 0 && inode->i_blocks != 1);
911 
912 	/* will put inode & node pages */
913 	truncate_node(&dn);
914 }
915 
916 struct page *new_inode_page(struct inode *inode)
917 {
918 	struct dnode_of_data dn;
919 
920 	/* allocate inode page for new inode */
921 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
922 
923 	/* caller should f2fs_put_page(page, 1); */
924 	return new_node_page(&dn, 0, NULL);
925 }
926 
927 struct page *new_node_page(struct dnode_of_data *dn,
928 				unsigned int ofs, struct page *ipage)
929 {
930 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
931 	struct node_info old_ni, new_ni;
932 	struct page *page;
933 	int err;
934 
935 	if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
936 		return ERR_PTR(-EPERM);
937 
938 	page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
939 	if (!page)
940 		return ERR_PTR(-ENOMEM);
941 
942 	if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
943 		err = -ENOSPC;
944 		goto fail;
945 	}
946 
947 	get_node_info(sbi, dn->nid, &old_ni);
948 
949 	/* Reinitialize old_ni with new node page */
950 	f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
951 	new_ni = old_ni;
952 	new_ni.ino = dn->inode->i_ino;
953 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
954 
955 	f2fs_wait_on_page_writeback(page, NODE);
956 	fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
957 	set_cold_node(dn->inode, page);
958 	SetPageUptodate(page);
959 	set_page_dirty(page);
960 
961 	if (f2fs_has_xattr_block(ofs))
962 		F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
963 
964 	dn->node_page = page;
965 	if (ipage)
966 		update_inode(dn->inode, ipage);
967 	else
968 		sync_inode_page(dn);
969 	if (ofs == 0)
970 		inc_valid_inode_count(sbi);
971 
972 	return page;
973 
974 fail:
975 	clear_node_page_dirty(page);
976 	f2fs_put_page(page, 1);
977 	return ERR_PTR(err);
978 }
979 
980 /*
981  * Caller should do after getting the following values.
982  * 0: f2fs_put_page(page, 0)
983  * LOCKED_PAGE: f2fs_put_page(page, 1)
984  * error: nothing
985  */
986 static int read_node_page(struct page *page, int rw)
987 {
988 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
989 	struct node_info ni;
990 	struct f2fs_io_info fio = {
991 		.type = NODE,
992 		.rw = rw,
993 	};
994 
995 	get_node_info(sbi, page->index, &ni);
996 
997 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
998 		f2fs_put_page(page, 1);
999 		return -ENOENT;
1000 	}
1001 
1002 	if (PageUptodate(page))
1003 		return LOCKED_PAGE;
1004 
1005 	fio.blk_addr = ni.blk_addr;
1006 	return f2fs_submit_page_bio(sbi, page, &fio);
1007 }
1008 
1009 /*
1010  * Readahead a node page
1011  */
1012 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1013 {
1014 	struct page *apage;
1015 	int err;
1016 
1017 	apage = find_get_page(NODE_MAPPING(sbi), nid);
1018 	if (apage && PageUptodate(apage)) {
1019 		f2fs_put_page(apage, 0);
1020 		return;
1021 	}
1022 	f2fs_put_page(apage, 0);
1023 
1024 	apage = grab_cache_page(NODE_MAPPING(sbi), nid);
1025 	if (!apage)
1026 		return;
1027 
1028 	err = read_node_page(apage, READA);
1029 	if (err == 0)
1030 		f2fs_put_page(apage, 0);
1031 	else if (err == LOCKED_PAGE)
1032 		f2fs_put_page(apage, 1);
1033 }
1034 
1035 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1036 {
1037 	struct page *page;
1038 	int err;
1039 repeat:
1040 	page = grab_cache_page(NODE_MAPPING(sbi), nid);
1041 	if (!page)
1042 		return ERR_PTR(-ENOMEM);
1043 
1044 	err = read_node_page(page, READ_SYNC);
1045 	if (err < 0)
1046 		return ERR_PTR(err);
1047 	else if (err != LOCKED_PAGE)
1048 		lock_page(page);
1049 
1050 	if (unlikely(!PageUptodate(page) || nid != nid_of_node(page))) {
1051 		ClearPageUptodate(page);
1052 		f2fs_put_page(page, 1);
1053 		return ERR_PTR(-EIO);
1054 	}
1055 	if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1056 		f2fs_put_page(page, 1);
1057 		goto repeat;
1058 	}
1059 	return page;
1060 }
1061 
1062 /*
1063  * Return a locked page for the desired node page.
1064  * And, readahead MAX_RA_NODE number of node pages.
1065  */
1066 struct page *get_node_page_ra(struct page *parent, int start)
1067 {
1068 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1069 	struct blk_plug plug;
1070 	struct page *page;
1071 	int err, i, end;
1072 	nid_t nid;
1073 
1074 	/* First, try getting the desired direct node. */
1075 	nid = get_nid(parent, start, false);
1076 	if (!nid)
1077 		return ERR_PTR(-ENOENT);
1078 repeat:
1079 	page = grab_cache_page(NODE_MAPPING(sbi), nid);
1080 	if (!page)
1081 		return ERR_PTR(-ENOMEM);
1082 
1083 	err = read_node_page(page, READ_SYNC);
1084 	if (err < 0)
1085 		return ERR_PTR(err);
1086 	else if (err == LOCKED_PAGE)
1087 		goto page_hit;
1088 
1089 	blk_start_plug(&plug);
1090 
1091 	/* Then, try readahead for siblings of the desired node */
1092 	end = start + MAX_RA_NODE;
1093 	end = min(end, NIDS_PER_BLOCK);
1094 	for (i = start + 1; i < end; i++) {
1095 		nid = get_nid(parent, i, false);
1096 		if (!nid)
1097 			continue;
1098 		ra_node_page(sbi, nid);
1099 	}
1100 
1101 	blk_finish_plug(&plug);
1102 
1103 	lock_page(page);
1104 	if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1105 		f2fs_put_page(page, 1);
1106 		goto repeat;
1107 	}
1108 page_hit:
1109 	if (unlikely(!PageUptodate(page))) {
1110 		f2fs_put_page(page, 1);
1111 		return ERR_PTR(-EIO);
1112 	}
1113 	return page;
1114 }
1115 
1116 void sync_inode_page(struct dnode_of_data *dn)
1117 {
1118 	if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1119 		update_inode(dn->inode, dn->node_page);
1120 	} else if (dn->inode_page) {
1121 		if (!dn->inode_page_locked)
1122 			lock_page(dn->inode_page);
1123 		update_inode(dn->inode, dn->inode_page);
1124 		if (!dn->inode_page_locked)
1125 			unlock_page(dn->inode_page);
1126 	} else {
1127 		update_inode_page(dn->inode);
1128 	}
1129 }
1130 
1131 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1132 					struct writeback_control *wbc)
1133 {
1134 	pgoff_t index, end;
1135 	struct pagevec pvec;
1136 	int step = ino ? 2 : 0;
1137 	int nwritten = 0, wrote = 0;
1138 
1139 	pagevec_init(&pvec, 0);
1140 
1141 next_step:
1142 	index = 0;
1143 	end = LONG_MAX;
1144 
1145 	while (index <= end) {
1146 		int i, nr_pages;
1147 		nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1148 				PAGECACHE_TAG_DIRTY,
1149 				min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1150 		if (nr_pages == 0)
1151 			break;
1152 
1153 		for (i = 0; i < nr_pages; i++) {
1154 			struct page *page = pvec.pages[i];
1155 
1156 			/*
1157 			 * flushing sequence with step:
1158 			 * 0. indirect nodes
1159 			 * 1. dentry dnodes
1160 			 * 2. file dnodes
1161 			 */
1162 			if (step == 0 && IS_DNODE(page))
1163 				continue;
1164 			if (step == 1 && (!IS_DNODE(page) ||
1165 						is_cold_node(page)))
1166 				continue;
1167 			if (step == 2 && (!IS_DNODE(page) ||
1168 						!is_cold_node(page)))
1169 				continue;
1170 
1171 			/*
1172 			 * If an fsync mode,
1173 			 * we should not skip writing node pages.
1174 			 */
1175 			if (ino && ino_of_node(page) == ino)
1176 				lock_page(page);
1177 			else if (!trylock_page(page))
1178 				continue;
1179 
1180 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1181 continue_unlock:
1182 				unlock_page(page);
1183 				continue;
1184 			}
1185 			if (ino && ino_of_node(page) != ino)
1186 				goto continue_unlock;
1187 
1188 			if (!PageDirty(page)) {
1189 				/* someone wrote it for us */
1190 				goto continue_unlock;
1191 			}
1192 
1193 			if (!clear_page_dirty_for_io(page))
1194 				goto continue_unlock;
1195 
1196 			/* called by fsync() */
1197 			if (ino && IS_DNODE(page)) {
1198 				set_fsync_mark(page, 1);
1199 				if (IS_INODE(page)) {
1200 					if (!is_checkpointed_node(sbi, ino) &&
1201 						!has_fsynced_inode(sbi, ino))
1202 						set_dentry_mark(page, 1);
1203 					else
1204 						set_dentry_mark(page, 0);
1205 				}
1206 				nwritten++;
1207 			} else {
1208 				set_fsync_mark(page, 0);
1209 				set_dentry_mark(page, 0);
1210 			}
1211 
1212 			if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1213 				unlock_page(page);
1214 			else
1215 				wrote++;
1216 
1217 			if (--wbc->nr_to_write == 0)
1218 				break;
1219 		}
1220 		pagevec_release(&pvec);
1221 		cond_resched();
1222 
1223 		if (wbc->nr_to_write == 0) {
1224 			step = 2;
1225 			break;
1226 		}
1227 	}
1228 
1229 	if (step < 2) {
1230 		step++;
1231 		goto next_step;
1232 	}
1233 
1234 	if (wrote)
1235 		f2fs_submit_merged_bio(sbi, NODE, WRITE);
1236 	return nwritten;
1237 }
1238 
1239 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1240 {
1241 	pgoff_t index = 0, end = LONG_MAX;
1242 	struct pagevec pvec;
1243 	int ret2 = 0, ret = 0;
1244 
1245 	pagevec_init(&pvec, 0);
1246 
1247 	while (index <= end) {
1248 		int i, nr_pages;
1249 		nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1250 				PAGECACHE_TAG_WRITEBACK,
1251 				min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1252 		if (nr_pages == 0)
1253 			break;
1254 
1255 		for (i = 0; i < nr_pages; i++) {
1256 			struct page *page = pvec.pages[i];
1257 
1258 			/* until radix tree lookup accepts end_index */
1259 			if (unlikely(page->index > end))
1260 				continue;
1261 
1262 			if (ino && ino_of_node(page) == ino) {
1263 				f2fs_wait_on_page_writeback(page, NODE);
1264 				if (TestClearPageError(page))
1265 					ret = -EIO;
1266 			}
1267 		}
1268 		pagevec_release(&pvec);
1269 		cond_resched();
1270 	}
1271 
1272 	if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1273 		ret2 = -ENOSPC;
1274 	if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1275 		ret2 = -EIO;
1276 	if (!ret)
1277 		ret = ret2;
1278 	return ret;
1279 }
1280 
1281 static int f2fs_write_node_page(struct page *page,
1282 				struct writeback_control *wbc)
1283 {
1284 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1285 	nid_t nid;
1286 	struct node_info ni;
1287 	struct f2fs_io_info fio = {
1288 		.type = NODE,
1289 		.rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1290 	};
1291 
1292 	trace_f2fs_writepage(page, NODE);
1293 
1294 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1295 		goto redirty_out;
1296 	if (unlikely(f2fs_cp_error(sbi)))
1297 		goto redirty_out;
1298 
1299 	f2fs_wait_on_page_writeback(page, NODE);
1300 
1301 	/* get old block addr of this node page */
1302 	nid = nid_of_node(page);
1303 	f2fs_bug_on(sbi, page->index != nid);
1304 
1305 	get_node_info(sbi, nid, &ni);
1306 
1307 	/* This page is already truncated */
1308 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1309 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1310 		unlock_page(page);
1311 		return 0;
1312 	}
1313 
1314 	if (wbc->for_reclaim) {
1315 		if (!down_read_trylock(&sbi->node_write))
1316 			goto redirty_out;
1317 	} else {
1318 		down_read(&sbi->node_write);
1319 	}
1320 
1321 	set_page_writeback(page);
1322 	fio.blk_addr = ni.blk_addr;
1323 	write_node_page(sbi, page, nid, &fio);
1324 	set_node_addr(sbi, &ni, fio.blk_addr, is_fsync_dnode(page));
1325 	dec_page_count(sbi, F2FS_DIRTY_NODES);
1326 	up_read(&sbi->node_write);
1327 	unlock_page(page);
1328 
1329 	if (wbc->for_reclaim)
1330 		f2fs_submit_merged_bio(sbi, NODE, WRITE);
1331 
1332 	return 0;
1333 
1334 redirty_out:
1335 	redirty_page_for_writepage(wbc, page);
1336 	return AOP_WRITEPAGE_ACTIVATE;
1337 }
1338 
1339 static int f2fs_write_node_pages(struct address_space *mapping,
1340 			    struct writeback_control *wbc)
1341 {
1342 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1343 	long diff;
1344 
1345 	trace_f2fs_writepages(mapping->host, wbc, NODE);
1346 
1347 	/* balancing f2fs's metadata in background */
1348 	f2fs_balance_fs_bg(sbi);
1349 
1350 	/* collect a number of dirty node pages and write together */
1351 	if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1352 		goto skip_write;
1353 
1354 	diff = nr_pages_to_write(sbi, NODE, wbc);
1355 	wbc->sync_mode = WB_SYNC_NONE;
1356 	sync_node_pages(sbi, 0, wbc);
1357 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1358 	return 0;
1359 
1360 skip_write:
1361 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1362 	return 0;
1363 }
1364 
1365 static int f2fs_set_node_page_dirty(struct page *page)
1366 {
1367 	trace_f2fs_set_page_dirty(page, NODE);
1368 
1369 	SetPageUptodate(page);
1370 	if (!PageDirty(page)) {
1371 		__set_page_dirty_nobuffers(page);
1372 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1373 		SetPagePrivate(page);
1374 		f2fs_trace_pid(page);
1375 		return 1;
1376 	}
1377 	return 0;
1378 }
1379 
1380 /*
1381  * Structure of the f2fs node operations
1382  */
1383 const struct address_space_operations f2fs_node_aops = {
1384 	.writepage	= f2fs_write_node_page,
1385 	.writepages	= f2fs_write_node_pages,
1386 	.set_page_dirty	= f2fs_set_node_page_dirty,
1387 	.invalidatepage	= f2fs_invalidate_page,
1388 	.releasepage	= f2fs_release_page,
1389 };
1390 
1391 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1392 						nid_t n)
1393 {
1394 	return radix_tree_lookup(&nm_i->free_nid_root, n);
1395 }
1396 
1397 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1398 						struct free_nid *i)
1399 {
1400 	list_del(&i->list);
1401 	radix_tree_delete(&nm_i->free_nid_root, i->nid);
1402 }
1403 
1404 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1405 {
1406 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1407 	struct free_nid *i;
1408 	struct nat_entry *ne;
1409 	bool allocated = false;
1410 
1411 	if (!available_free_memory(sbi, FREE_NIDS))
1412 		return -1;
1413 
1414 	/* 0 nid should not be used */
1415 	if (unlikely(nid == 0))
1416 		return 0;
1417 
1418 	if (build) {
1419 		/* do not add allocated nids */
1420 		down_read(&nm_i->nat_tree_lock);
1421 		ne = __lookup_nat_cache(nm_i, nid);
1422 		if (ne &&
1423 			(!get_nat_flag(ne, IS_CHECKPOINTED) ||
1424 				nat_get_blkaddr(ne) != NULL_ADDR))
1425 			allocated = true;
1426 		up_read(&nm_i->nat_tree_lock);
1427 		if (allocated)
1428 			return 0;
1429 	}
1430 
1431 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1432 	i->nid = nid;
1433 	i->state = NID_NEW;
1434 
1435 	if (radix_tree_preload(GFP_NOFS)) {
1436 		kmem_cache_free(free_nid_slab, i);
1437 		return 0;
1438 	}
1439 
1440 	spin_lock(&nm_i->free_nid_list_lock);
1441 	if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1442 		spin_unlock(&nm_i->free_nid_list_lock);
1443 		radix_tree_preload_end();
1444 		kmem_cache_free(free_nid_slab, i);
1445 		return 0;
1446 	}
1447 	list_add_tail(&i->list, &nm_i->free_nid_list);
1448 	nm_i->fcnt++;
1449 	spin_unlock(&nm_i->free_nid_list_lock);
1450 	radix_tree_preload_end();
1451 	return 1;
1452 }
1453 
1454 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1455 {
1456 	struct free_nid *i;
1457 	bool need_free = false;
1458 
1459 	spin_lock(&nm_i->free_nid_list_lock);
1460 	i = __lookup_free_nid_list(nm_i, nid);
1461 	if (i && i->state == NID_NEW) {
1462 		__del_from_free_nid_list(nm_i, i);
1463 		nm_i->fcnt--;
1464 		need_free = true;
1465 	}
1466 	spin_unlock(&nm_i->free_nid_list_lock);
1467 
1468 	if (need_free)
1469 		kmem_cache_free(free_nid_slab, i);
1470 }
1471 
1472 static void scan_nat_page(struct f2fs_sb_info *sbi,
1473 			struct page *nat_page, nid_t start_nid)
1474 {
1475 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1476 	struct f2fs_nat_block *nat_blk = page_address(nat_page);
1477 	block_t blk_addr;
1478 	int i;
1479 
1480 	i = start_nid % NAT_ENTRY_PER_BLOCK;
1481 
1482 	for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1483 
1484 		if (unlikely(start_nid >= nm_i->max_nid))
1485 			break;
1486 
1487 		blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1488 		f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1489 		if (blk_addr == NULL_ADDR) {
1490 			if (add_free_nid(sbi, start_nid, true) < 0)
1491 				break;
1492 		}
1493 	}
1494 }
1495 
1496 static void build_free_nids(struct f2fs_sb_info *sbi)
1497 {
1498 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1499 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1500 	struct f2fs_summary_block *sum = curseg->sum_blk;
1501 	int i = 0;
1502 	nid_t nid = nm_i->next_scan_nid;
1503 
1504 	/* Enough entries */
1505 	if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1506 		return;
1507 
1508 	/* readahead nat pages to be scanned */
1509 	ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES, META_NAT);
1510 
1511 	while (1) {
1512 		struct page *page = get_current_nat_page(sbi, nid);
1513 
1514 		scan_nat_page(sbi, page, nid);
1515 		f2fs_put_page(page, 1);
1516 
1517 		nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1518 		if (unlikely(nid >= nm_i->max_nid))
1519 			nid = 0;
1520 
1521 		if (i++ == FREE_NID_PAGES)
1522 			break;
1523 	}
1524 
1525 	/* go to the next free nat pages to find free nids abundantly */
1526 	nm_i->next_scan_nid = nid;
1527 
1528 	/* find free nids from current sum_pages */
1529 	mutex_lock(&curseg->curseg_mutex);
1530 	for (i = 0; i < nats_in_cursum(sum); i++) {
1531 		block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1532 		nid = le32_to_cpu(nid_in_journal(sum, i));
1533 		if (addr == NULL_ADDR)
1534 			add_free_nid(sbi, nid, true);
1535 		else
1536 			remove_free_nid(nm_i, nid);
1537 	}
1538 	mutex_unlock(&curseg->curseg_mutex);
1539 }
1540 
1541 /*
1542  * If this function returns success, caller can obtain a new nid
1543  * from second parameter of this function.
1544  * The returned nid could be used ino as well as nid when inode is created.
1545  */
1546 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1547 {
1548 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1549 	struct free_nid *i = NULL;
1550 retry:
1551 	if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1552 		return false;
1553 
1554 	spin_lock(&nm_i->free_nid_list_lock);
1555 
1556 	/* We should not use stale free nids created by build_free_nids */
1557 	if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1558 		f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1559 		list_for_each_entry(i, &nm_i->free_nid_list, list)
1560 			if (i->state == NID_NEW)
1561 				break;
1562 
1563 		f2fs_bug_on(sbi, i->state != NID_NEW);
1564 		*nid = i->nid;
1565 		i->state = NID_ALLOC;
1566 		nm_i->fcnt--;
1567 		spin_unlock(&nm_i->free_nid_list_lock);
1568 		return true;
1569 	}
1570 	spin_unlock(&nm_i->free_nid_list_lock);
1571 
1572 	/* Let's scan nat pages and its caches to get free nids */
1573 	mutex_lock(&nm_i->build_lock);
1574 	build_free_nids(sbi);
1575 	mutex_unlock(&nm_i->build_lock);
1576 	goto retry;
1577 }
1578 
1579 /*
1580  * alloc_nid() should be called prior to this function.
1581  */
1582 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1583 {
1584 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1585 	struct free_nid *i;
1586 
1587 	spin_lock(&nm_i->free_nid_list_lock);
1588 	i = __lookup_free_nid_list(nm_i, nid);
1589 	f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1590 	__del_from_free_nid_list(nm_i, i);
1591 	spin_unlock(&nm_i->free_nid_list_lock);
1592 
1593 	kmem_cache_free(free_nid_slab, i);
1594 }
1595 
1596 /*
1597  * alloc_nid() should be called prior to this function.
1598  */
1599 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1600 {
1601 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1602 	struct free_nid *i;
1603 	bool need_free = false;
1604 
1605 	if (!nid)
1606 		return;
1607 
1608 	spin_lock(&nm_i->free_nid_list_lock);
1609 	i = __lookup_free_nid_list(nm_i, nid);
1610 	f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1611 	if (!available_free_memory(sbi, FREE_NIDS)) {
1612 		__del_from_free_nid_list(nm_i, i);
1613 		need_free = true;
1614 	} else {
1615 		i->state = NID_NEW;
1616 		nm_i->fcnt++;
1617 	}
1618 	spin_unlock(&nm_i->free_nid_list_lock);
1619 
1620 	if (need_free)
1621 		kmem_cache_free(free_nid_slab, i);
1622 }
1623 
1624 void recover_inline_xattr(struct inode *inode, struct page *page)
1625 {
1626 	void *src_addr, *dst_addr;
1627 	size_t inline_size;
1628 	struct page *ipage;
1629 	struct f2fs_inode *ri;
1630 
1631 	ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1632 	f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1633 
1634 	ri = F2FS_INODE(page);
1635 	if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1636 		clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1637 		goto update_inode;
1638 	}
1639 
1640 	dst_addr = inline_xattr_addr(ipage);
1641 	src_addr = inline_xattr_addr(page);
1642 	inline_size = inline_xattr_size(inode);
1643 
1644 	f2fs_wait_on_page_writeback(ipage, NODE);
1645 	memcpy(dst_addr, src_addr, inline_size);
1646 update_inode:
1647 	update_inode(inode, ipage);
1648 	f2fs_put_page(ipage, 1);
1649 }
1650 
1651 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1652 {
1653 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1654 	nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1655 	nid_t new_xnid = nid_of_node(page);
1656 	struct node_info ni;
1657 
1658 	/* 1: invalidate the previous xattr nid */
1659 	if (!prev_xnid)
1660 		goto recover_xnid;
1661 
1662 	/* Deallocate node address */
1663 	get_node_info(sbi, prev_xnid, &ni);
1664 	f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1665 	invalidate_blocks(sbi, ni.blk_addr);
1666 	dec_valid_node_count(sbi, inode);
1667 	set_node_addr(sbi, &ni, NULL_ADDR, false);
1668 
1669 recover_xnid:
1670 	/* 2: allocate new xattr nid */
1671 	if (unlikely(!inc_valid_node_count(sbi, inode)))
1672 		f2fs_bug_on(sbi, 1);
1673 
1674 	remove_free_nid(NM_I(sbi), new_xnid);
1675 	get_node_info(sbi, new_xnid, &ni);
1676 	ni.ino = inode->i_ino;
1677 	set_node_addr(sbi, &ni, NEW_ADDR, false);
1678 	F2FS_I(inode)->i_xattr_nid = new_xnid;
1679 
1680 	/* 3: update xattr blkaddr */
1681 	refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1682 	set_node_addr(sbi, &ni, blkaddr, false);
1683 
1684 	update_inode_page(inode);
1685 }
1686 
1687 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1688 {
1689 	struct f2fs_inode *src, *dst;
1690 	nid_t ino = ino_of_node(page);
1691 	struct node_info old_ni, new_ni;
1692 	struct page *ipage;
1693 
1694 	get_node_info(sbi, ino, &old_ni);
1695 
1696 	if (unlikely(old_ni.blk_addr != NULL_ADDR))
1697 		return -EINVAL;
1698 
1699 	ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1700 	if (!ipage)
1701 		return -ENOMEM;
1702 
1703 	/* Should not use this inode from free nid list */
1704 	remove_free_nid(NM_I(sbi), ino);
1705 
1706 	SetPageUptodate(ipage);
1707 	fill_node_footer(ipage, ino, ino, 0, true);
1708 
1709 	src = F2FS_INODE(page);
1710 	dst = F2FS_INODE(ipage);
1711 
1712 	memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1713 	dst->i_size = 0;
1714 	dst->i_blocks = cpu_to_le64(1);
1715 	dst->i_links = cpu_to_le32(1);
1716 	dst->i_xattr_nid = 0;
1717 	dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
1718 
1719 	new_ni = old_ni;
1720 	new_ni.ino = ino;
1721 
1722 	if (unlikely(!inc_valid_node_count(sbi, NULL)))
1723 		WARN_ON(1);
1724 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1725 	inc_valid_inode_count(sbi);
1726 	set_page_dirty(ipage);
1727 	f2fs_put_page(ipage, 1);
1728 	return 0;
1729 }
1730 
1731 int restore_node_summary(struct f2fs_sb_info *sbi,
1732 			unsigned int segno, struct f2fs_summary_block *sum)
1733 {
1734 	struct f2fs_node *rn;
1735 	struct f2fs_summary *sum_entry;
1736 	block_t addr;
1737 	int bio_blocks = MAX_BIO_BLOCKS(sbi);
1738 	int i, idx, last_offset, nrpages;
1739 
1740 	/* scan the node segment */
1741 	last_offset = sbi->blocks_per_seg;
1742 	addr = START_BLOCK(sbi, segno);
1743 	sum_entry = &sum->entries[0];
1744 
1745 	for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
1746 		nrpages = min(last_offset - i, bio_blocks);
1747 
1748 		/* readahead node pages */
1749 		ra_meta_pages(sbi, addr, nrpages, META_POR);
1750 
1751 		for (idx = addr; idx < addr + nrpages; idx++) {
1752 			struct page *page = get_meta_page(sbi, idx);
1753 
1754 			rn = F2FS_NODE(page);
1755 			sum_entry->nid = rn->footer.nid;
1756 			sum_entry->version = 0;
1757 			sum_entry->ofs_in_node = 0;
1758 			sum_entry++;
1759 			f2fs_put_page(page, 1);
1760 		}
1761 
1762 		invalidate_mapping_pages(META_MAPPING(sbi), addr,
1763 							addr + nrpages);
1764 	}
1765 	return 0;
1766 }
1767 
1768 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
1769 {
1770 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1771 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1772 	struct f2fs_summary_block *sum = curseg->sum_blk;
1773 	int i;
1774 
1775 	mutex_lock(&curseg->curseg_mutex);
1776 	for (i = 0; i < nats_in_cursum(sum); i++) {
1777 		struct nat_entry *ne;
1778 		struct f2fs_nat_entry raw_ne;
1779 		nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1780 
1781 		raw_ne = nat_in_journal(sum, i);
1782 
1783 		down_write(&nm_i->nat_tree_lock);
1784 		ne = __lookup_nat_cache(nm_i, nid);
1785 		if (!ne) {
1786 			ne = grab_nat_entry(nm_i, nid);
1787 			node_info_from_raw_nat(&ne->ni, &raw_ne);
1788 		}
1789 		__set_nat_cache_dirty(nm_i, ne);
1790 		up_write(&nm_i->nat_tree_lock);
1791 	}
1792 	update_nats_in_cursum(sum, -i);
1793 	mutex_unlock(&curseg->curseg_mutex);
1794 }
1795 
1796 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
1797 						struct list_head *head, int max)
1798 {
1799 	struct nat_entry_set *cur;
1800 
1801 	if (nes->entry_cnt >= max)
1802 		goto add_out;
1803 
1804 	list_for_each_entry(cur, head, set_list) {
1805 		if (cur->entry_cnt >= nes->entry_cnt) {
1806 			list_add(&nes->set_list, cur->set_list.prev);
1807 			return;
1808 		}
1809 	}
1810 add_out:
1811 	list_add_tail(&nes->set_list, head);
1812 }
1813 
1814 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
1815 					struct nat_entry_set *set)
1816 {
1817 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1818 	struct f2fs_summary_block *sum = curseg->sum_blk;
1819 	nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
1820 	bool to_journal = true;
1821 	struct f2fs_nat_block *nat_blk;
1822 	struct nat_entry *ne, *cur;
1823 	struct page *page = NULL;
1824 
1825 	/*
1826 	 * there are two steps to flush nat entries:
1827 	 * #1, flush nat entries to journal in current hot data summary block.
1828 	 * #2, flush nat entries to nat page.
1829 	 */
1830 	if (!__has_cursum_space(sum, set->entry_cnt, NAT_JOURNAL))
1831 		to_journal = false;
1832 
1833 	if (to_journal) {
1834 		mutex_lock(&curseg->curseg_mutex);
1835 	} else {
1836 		page = get_next_nat_page(sbi, start_nid);
1837 		nat_blk = page_address(page);
1838 		f2fs_bug_on(sbi, !nat_blk);
1839 	}
1840 
1841 	/* flush dirty nats in nat entry set */
1842 	list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
1843 		struct f2fs_nat_entry *raw_ne;
1844 		nid_t nid = nat_get_nid(ne);
1845 		int offset;
1846 
1847 		if (nat_get_blkaddr(ne) == NEW_ADDR)
1848 			continue;
1849 
1850 		if (to_journal) {
1851 			offset = lookup_journal_in_cursum(sum,
1852 							NAT_JOURNAL, nid, 1);
1853 			f2fs_bug_on(sbi, offset < 0);
1854 			raw_ne = &nat_in_journal(sum, offset);
1855 			nid_in_journal(sum, offset) = cpu_to_le32(nid);
1856 		} else {
1857 			raw_ne = &nat_blk->entries[nid - start_nid];
1858 		}
1859 		raw_nat_from_node_info(raw_ne, &ne->ni);
1860 
1861 		down_write(&NM_I(sbi)->nat_tree_lock);
1862 		nat_reset_flag(ne);
1863 		__clear_nat_cache_dirty(NM_I(sbi), ne);
1864 		up_write(&NM_I(sbi)->nat_tree_lock);
1865 
1866 		if (nat_get_blkaddr(ne) == NULL_ADDR)
1867 			add_free_nid(sbi, nid, false);
1868 	}
1869 
1870 	if (to_journal)
1871 		mutex_unlock(&curseg->curseg_mutex);
1872 	else
1873 		f2fs_put_page(page, 1);
1874 
1875 	f2fs_bug_on(sbi, set->entry_cnt);
1876 
1877 	radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
1878 	kmem_cache_free(nat_entry_set_slab, set);
1879 }
1880 
1881 /*
1882  * This function is called during the checkpointing process.
1883  */
1884 void flush_nat_entries(struct f2fs_sb_info *sbi)
1885 {
1886 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1887 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1888 	struct f2fs_summary_block *sum = curseg->sum_blk;
1889 	struct nat_entry_set *setvec[SETVEC_SIZE];
1890 	struct nat_entry_set *set, *tmp;
1891 	unsigned int found;
1892 	nid_t set_idx = 0;
1893 	LIST_HEAD(sets);
1894 
1895 	if (!nm_i->dirty_nat_cnt)
1896 		return;
1897 	/*
1898 	 * if there are no enough space in journal to store dirty nat
1899 	 * entries, remove all entries from journal and merge them
1900 	 * into nat entry set.
1901 	 */
1902 	if (!__has_cursum_space(sum, nm_i->dirty_nat_cnt, NAT_JOURNAL))
1903 		remove_nats_in_journal(sbi);
1904 
1905 	while ((found = __gang_lookup_nat_set(nm_i,
1906 					set_idx, SETVEC_SIZE, setvec))) {
1907 		unsigned idx;
1908 		set_idx = setvec[found - 1]->set + 1;
1909 		for (idx = 0; idx < found; idx++)
1910 			__adjust_nat_entry_set(setvec[idx], &sets,
1911 							MAX_NAT_JENTRIES(sum));
1912 	}
1913 
1914 	/* flush dirty nats in nat entry set */
1915 	list_for_each_entry_safe(set, tmp, &sets, set_list)
1916 		__flush_nat_entry_set(sbi, set);
1917 
1918 	f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
1919 }
1920 
1921 static int init_node_manager(struct f2fs_sb_info *sbi)
1922 {
1923 	struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1924 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1925 	unsigned char *version_bitmap;
1926 	unsigned int nat_segs, nat_blocks;
1927 
1928 	nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1929 
1930 	/* segment_count_nat includes pair segment so divide to 2. */
1931 	nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1932 	nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1933 
1934 	nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1935 
1936 	/* not used nids: 0, node, meta, (and root counted as valid node) */
1937 	nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
1938 	nm_i->fcnt = 0;
1939 	nm_i->nat_cnt = 0;
1940 	nm_i->ram_thresh = DEF_RAM_THRESHOLD;
1941 
1942 	INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
1943 	INIT_LIST_HEAD(&nm_i->free_nid_list);
1944 	INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
1945 	INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
1946 	INIT_LIST_HEAD(&nm_i->nat_entries);
1947 
1948 	mutex_init(&nm_i->build_lock);
1949 	spin_lock_init(&nm_i->free_nid_list_lock);
1950 	init_rwsem(&nm_i->nat_tree_lock);
1951 
1952 	nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1953 	nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1954 	version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1955 	if (!version_bitmap)
1956 		return -EFAULT;
1957 
1958 	nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1959 					GFP_KERNEL);
1960 	if (!nm_i->nat_bitmap)
1961 		return -ENOMEM;
1962 	return 0;
1963 }
1964 
1965 int build_node_manager(struct f2fs_sb_info *sbi)
1966 {
1967 	int err;
1968 
1969 	sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1970 	if (!sbi->nm_info)
1971 		return -ENOMEM;
1972 
1973 	err = init_node_manager(sbi);
1974 	if (err)
1975 		return err;
1976 
1977 	build_free_nids(sbi);
1978 	return 0;
1979 }
1980 
1981 void destroy_node_manager(struct f2fs_sb_info *sbi)
1982 {
1983 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1984 	struct free_nid *i, *next_i;
1985 	struct nat_entry *natvec[NATVEC_SIZE];
1986 	struct nat_entry_set *setvec[SETVEC_SIZE];
1987 	nid_t nid = 0;
1988 	unsigned int found;
1989 
1990 	if (!nm_i)
1991 		return;
1992 
1993 	/* destroy free nid list */
1994 	spin_lock(&nm_i->free_nid_list_lock);
1995 	list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1996 		f2fs_bug_on(sbi, i->state == NID_ALLOC);
1997 		__del_from_free_nid_list(nm_i, i);
1998 		nm_i->fcnt--;
1999 		spin_unlock(&nm_i->free_nid_list_lock);
2000 		kmem_cache_free(free_nid_slab, i);
2001 		spin_lock(&nm_i->free_nid_list_lock);
2002 	}
2003 	f2fs_bug_on(sbi, nm_i->fcnt);
2004 	spin_unlock(&nm_i->free_nid_list_lock);
2005 
2006 	/* destroy nat cache */
2007 	down_write(&nm_i->nat_tree_lock);
2008 	while ((found = __gang_lookup_nat_cache(nm_i,
2009 					nid, NATVEC_SIZE, natvec))) {
2010 		unsigned idx;
2011 
2012 		nid = nat_get_nid(natvec[found - 1]) + 1;
2013 		for (idx = 0; idx < found; idx++)
2014 			__del_from_nat_cache(nm_i, natvec[idx]);
2015 	}
2016 	f2fs_bug_on(sbi, nm_i->nat_cnt);
2017 
2018 	/* destroy nat set cache */
2019 	nid = 0;
2020 	while ((found = __gang_lookup_nat_set(nm_i,
2021 					nid, SETVEC_SIZE, setvec))) {
2022 		unsigned idx;
2023 
2024 		nid = setvec[found - 1]->set + 1;
2025 		for (idx = 0; idx < found; idx++) {
2026 			/* entry_cnt is not zero, when cp_error was occurred */
2027 			f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2028 			radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2029 			kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2030 		}
2031 	}
2032 	up_write(&nm_i->nat_tree_lock);
2033 
2034 	kfree(nm_i->nat_bitmap);
2035 	sbi->nm_info = NULL;
2036 	kfree(nm_i);
2037 }
2038 
2039 int __init create_node_manager_caches(void)
2040 {
2041 	nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2042 			sizeof(struct nat_entry));
2043 	if (!nat_entry_slab)
2044 		goto fail;
2045 
2046 	free_nid_slab = f2fs_kmem_cache_create("free_nid",
2047 			sizeof(struct free_nid));
2048 	if (!free_nid_slab)
2049 		goto destroy_nat_entry;
2050 
2051 	nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2052 			sizeof(struct nat_entry_set));
2053 	if (!nat_entry_set_slab)
2054 		goto destroy_free_nid;
2055 	return 0;
2056 
2057 destroy_free_nid:
2058 	kmem_cache_destroy(free_nid_slab);
2059 destroy_nat_entry:
2060 	kmem_cache_destroy(nat_entry_slab);
2061 fail:
2062 	return -ENOMEM;
2063 }
2064 
2065 void destroy_node_manager_caches(void)
2066 {
2067 	kmem_cache_destroy(nat_entry_set_slab);
2068 	kmem_cache_destroy(free_nid_slab);
2069 	kmem_cache_destroy(nat_entry_slab);
2070 }
2071