xref: /openbmc/linux/fs/f2fs/node.c (revision 51dd62493477923723c797c6da60121ed39900ed)
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/events/f2fs.h>
23 
24 static struct kmem_cache *nat_entry_slab;
25 static struct kmem_cache *free_nid_slab;
26 
27 static void clear_node_page_dirty(struct page *page)
28 {
29 	struct address_space *mapping = page->mapping;
30 	struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
31 	unsigned int long flags;
32 
33 	if (PageDirty(page)) {
34 		spin_lock_irqsave(&mapping->tree_lock, flags);
35 		radix_tree_tag_clear(&mapping->page_tree,
36 				page_index(page),
37 				PAGECACHE_TAG_DIRTY);
38 		spin_unlock_irqrestore(&mapping->tree_lock, flags);
39 
40 		clear_page_dirty_for_io(page);
41 		dec_page_count(sbi, F2FS_DIRTY_NODES);
42 	}
43 	ClearPageUptodate(page);
44 }
45 
46 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
47 {
48 	pgoff_t index = current_nat_addr(sbi, nid);
49 	return get_meta_page(sbi, index);
50 }
51 
52 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
53 {
54 	struct page *src_page;
55 	struct page *dst_page;
56 	pgoff_t src_off;
57 	pgoff_t dst_off;
58 	void *src_addr;
59 	void *dst_addr;
60 	struct f2fs_nm_info *nm_i = NM_I(sbi);
61 
62 	src_off = current_nat_addr(sbi, nid);
63 	dst_off = next_nat_addr(sbi, src_off);
64 
65 	/* get current nat block page with lock */
66 	src_page = get_meta_page(sbi, src_off);
67 
68 	/* Dirty src_page means that it is already the new target NAT page. */
69 	if (PageDirty(src_page))
70 		return src_page;
71 
72 	dst_page = grab_meta_page(sbi, dst_off);
73 
74 	src_addr = page_address(src_page);
75 	dst_addr = page_address(dst_page);
76 	memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
77 	set_page_dirty(dst_page);
78 	f2fs_put_page(src_page, 1);
79 
80 	set_to_next_nat(nm_i, nid);
81 
82 	return dst_page;
83 }
84 
85 /*
86  * Readahead NAT pages
87  */
88 static void ra_nat_pages(struct f2fs_sb_info *sbi, int nid)
89 {
90 	struct address_space *mapping = sbi->meta_inode->i_mapping;
91 	struct f2fs_nm_info *nm_i = NM_I(sbi);
92 	struct page *page;
93 	pgoff_t index;
94 	int i;
95 
96 	for (i = 0; i < FREE_NID_PAGES; i++, nid += NAT_ENTRY_PER_BLOCK) {
97 		if (nid >= nm_i->max_nid)
98 			nid = 0;
99 		index = current_nat_addr(sbi, nid);
100 
101 		page = grab_cache_page(mapping, index);
102 		if (!page)
103 			continue;
104 		if (PageUptodate(page)) {
105 			f2fs_put_page(page, 1);
106 			continue;
107 		}
108 		if (f2fs_readpage(sbi, page, index, READ))
109 			continue;
110 
111 		f2fs_put_page(page, 0);
112 	}
113 }
114 
115 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
116 {
117 	return radix_tree_lookup(&nm_i->nat_root, n);
118 }
119 
120 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
121 		nid_t start, unsigned int nr, struct nat_entry **ep)
122 {
123 	return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
124 }
125 
126 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
127 {
128 	list_del(&e->list);
129 	radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
130 	nm_i->nat_cnt--;
131 	kmem_cache_free(nat_entry_slab, e);
132 }
133 
134 int is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
135 {
136 	struct f2fs_nm_info *nm_i = NM_I(sbi);
137 	struct nat_entry *e;
138 	int is_cp = 1;
139 
140 	read_lock(&nm_i->nat_tree_lock);
141 	e = __lookup_nat_cache(nm_i, nid);
142 	if (e && !e->checkpointed)
143 		is_cp = 0;
144 	read_unlock(&nm_i->nat_tree_lock);
145 	return is_cp;
146 }
147 
148 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
149 {
150 	struct nat_entry *new;
151 
152 	new = kmem_cache_alloc(nat_entry_slab, GFP_ATOMIC);
153 	if (!new)
154 		return NULL;
155 	if (radix_tree_insert(&nm_i->nat_root, nid, new)) {
156 		kmem_cache_free(nat_entry_slab, new);
157 		return NULL;
158 	}
159 	memset(new, 0, sizeof(struct nat_entry));
160 	nat_set_nid(new, nid);
161 	list_add_tail(&new->list, &nm_i->nat_entries);
162 	nm_i->nat_cnt++;
163 	return new;
164 }
165 
166 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
167 						struct f2fs_nat_entry *ne)
168 {
169 	struct nat_entry *e;
170 retry:
171 	write_lock(&nm_i->nat_tree_lock);
172 	e = __lookup_nat_cache(nm_i, nid);
173 	if (!e) {
174 		e = grab_nat_entry(nm_i, nid);
175 		if (!e) {
176 			write_unlock(&nm_i->nat_tree_lock);
177 			goto retry;
178 		}
179 		nat_set_blkaddr(e, le32_to_cpu(ne->block_addr));
180 		nat_set_ino(e, le32_to_cpu(ne->ino));
181 		nat_set_version(e, ne->version);
182 		e->checkpointed = true;
183 	}
184 	write_unlock(&nm_i->nat_tree_lock);
185 }
186 
187 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
188 			block_t new_blkaddr)
189 {
190 	struct f2fs_nm_info *nm_i = NM_I(sbi);
191 	struct nat_entry *e;
192 retry:
193 	write_lock(&nm_i->nat_tree_lock);
194 	e = __lookup_nat_cache(nm_i, ni->nid);
195 	if (!e) {
196 		e = grab_nat_entry(nm_i, ni->nid);
197 		if (!e) {
198 			write_unlock(&nm_i->nat_tree_lock);
199 			goto retry;
200 		}
201 		e->ni = *ni;
202 		e->checkpointed = true;
203 		BUG_ON(ni->blk_addr == NEW_ADDR);
204 	} else if (new_blkaddr == NEW_ADDR) {
205 		/*
206 		 * when nid is reallocated,
207 		 * previous nat entry can be remained in nat cache.
208 		 * So, reinitialize it with new information.
209 		 */
210 		e->ni = *ni;
211 		BUG_ON(ni->blk_addr != NULL_ADDR);
212 	}
213 
214 	if (new_blkaddr == NEW_ADDR)
215 		e->checkpointed = false;
216 
217 	/* sanity check */
218 	BUG_ON(nat_get_blkaddr(e) != ni->blk_addr);
219 	BUG_ON(nat_get_blkaddr(e) == NULL_ADDR &&
220 			new_blkaddr == NULL_ADDR);
221 	BUG_ON(nat_get_blkaddr(e) == NEW_ADDR &&
222 			new_blkaddr == NEW_ADDR);
223 	BUG_ON(nat_get_blkaddr(e) != NEW_ADDR &&
224 			nat_get_blkaddr(e) != NULL_ADDR &&
225 			new_blkaddr == NEW_ADDR);
226 
227 	/* increament version no as node is removed */
228 	if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
229 		unsigned char version = nat_get_version(e);
230 		nat_set_version(e, inc_node_version(version));
231 	}
232 
233 	/* change address */
234 	nat_set_blkaddr(e, new_blkaddr);
235 	__set_nat_cache_dirty(nm_i, e);
236 	write_unlock(&nm_i->nat_tree_lock);
237 }
238 
239 static int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
240 {
241 	struct f2fs_nm_info *nm_i = NM_I(sbi);
242 
243 	if (nm_i->nat_cnt < 2 * NM_WOUT_THRESHOLD)
244 		return 0;
245 
246 	write_lock(&nm_i->nat_tree_lock);
247 	while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
248 		struct nat_entry *ne;
249 		ne = list_first_entry(&nm_i->nat_entries,
250 					struct nat_entry, list);
251 		__del_from_nat_cache(nm_i, ne);
252 		nr_shrink--;
253 	}
254 	write_unlock(&nm_i->nat_tree_lock);
255 	return nr_shrink;
256 }
257 
258 /*
259  * This function returns always success
260  */
261 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
262 {
263 	struct f2fs_nm_info *nm_i = NM_I(sbi);
264 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
265 	struct f2fs_summary_block *sum = curseg->sum_blk;
266 	nid_t start_nid = START_NID(nid);
267 	struct f2fs_nat_block *nat_blk;
268 	struct page *page = NULL;
269 	struct f2fs_nat_entry ne;
270 	struct nat_entry *e;
271 	int i;
272 
273 	memset(&ne, 0, sizeof(struct f2fs_nat_entry));
274 	ni->nid = nid;
275 
276 	/* Check nat cache */
277 	read_lock(&nm_i->nat_tree_lock);
278 	e = __lookup_nat_cache(nm_i, nid);
279 	if (e) {
280 		ni->ino = nat_get_ino(e);
281 		ni->blk_addr = nat_get_blkaddr(e);
282 		ni->version = nat_get_version(e);
283 	}
284 	read_unlock(&nm_i->nat_tree_lock);
285 	if (e)
286 		return;
287 
288 	/* Check current segment summary */
289 	mutex_lock(&curseg->curseg_mutex);
290 	i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
291 	if (i >= 0) {
292 		ne = nat_in_journal(sum, i);
293 		node_info_from_raw_nat(ni, &ne);
294 	}
295 	mutex_unlock(&curseg->curseg_mutex);
296 	if (i >= 0)
297 		goto cache;
298 
299 	/* Fill node_info from nat page */
300 	page = get_current_nat_page(sbi, start_nid);
301 	nat_blk = (struct f2fs_nat_block *)page_address(page);
302 	ne = nat_blk->entries[nid - start_nid];
303 	node_info_from_raw_nat(ni, &ne);
304 	f2fs_put_page(page, 1);
305 cache:
306 	/* cache nat entry */
307 	cache_nat_entry(NM_I(sbi), nid, &ne);
308 }
309 
310 /*
311  * The maximum depth is four.
312  * Offset[0] will have raw inode offset.
313  */
314 static int get_node_path(long block, int offset[4], unsigned int noffset[4])
315 {
316 	const long direct_index = ADDRS_PER_INODE;
317 	const long direct_blks = ADDRS_PER_BLOCK;
318 	const long dptrs_per_blk = NIDS_PER_BLOCK;
319 	const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
320 	const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
321 	int n = 0;
322 	int level = 0;
323 
324 	noffset[0] = 0;
325 
326 	if (block < direct_index) {
327 		offset[n] = block;
328 		goto got;
329 	}
330 	block -= direct_index;
331 	if (block < direct_blks) {
332 		offset[n++] = NODE_DIR1_BLOCK;
333 		noffset[n] = 1;
334 		offset[n] = block;
335 		level = 1;
336 		goto got;
337 	}
338 	block -= direct_blks;
339 	if (block < direct_blks) {
340 		offset[n++] = NODE_DIR2_BLOCK;
341 		noffset[n] = 2;
342 		offset[n] = block;
343 		level = 1;
344 		goto got;
345 	}
346 	block -= direct_blks;
347 	if (block < indirect_blks) {
348 		offset[n++] = NODE_IND1_BLOCK;
349 		noffset[n] = 3;
350 		offset[n++] = block / direct_blks;
351 		noffset[n] = 4 + offset[n - 1];
352 		offset[n] = block % direct_blks;
353 		level = 2;
354 		goto got;
355 	}
356 	block -= indirect_blks;
357 	if (block < indirect_blks) {
358 		offset[n++] = NODE_IND2_BLOCK;
359 		noffset[n] = 4 + dptrs_per_blk;
360 		offset[n++] = block / direct_blks;
361 		noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
362 		offset[n] = block % direct_blks;
363 		level = 2;
364 		goto got;
365 	}
366 	block -= indirect_blks;
367 	if (block < dindirect_blks) {
368 		offset[n++] = NODE_DIND_BLOCK;
369 		noffset[n] = 5 + (dptrs_per_blk * 2);
370 		offset[n++] = block / indirect_blks;
371 		noffset[n] = 6 + (dptrs_per_blk * 2) +
372 			      offset[n - 1] * (dptrs_per_blk + 1);
373 		offset[n++] = (block / direct_blks) % dptrs_per_blk;
374 		noffset[n] = 7 + (dptrs_per_blk * 2) +
375 			      offset[n - 2] * (dptrs_per_blk + 1) +
376 			      offset[n - 1];
377 		offset[n] = block % direct_blks;
378 		level = 3;
379 		goto got;
380 	} else {
381 		BUG();
382 	}
383 got:
384 	return level;
385 }
386 
387 /*
388  * Caller should call f2fs_put_dnode(dn).
389  * Also, it should grab and release a mutex by calling mutex_lock_op() and
390  * mutex_unlock_op() only if ro is not set RDONLY_NODE.
391  * In the case of RDONLY_NODE, we don't need to care about mutex.
392  */
393 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
394 {
395 	struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
396 	struct page *npage[4];
397 	struct page *parent;
398 	int offset[4];
399 	unsigned int noffset[4];
400 	nid_t nids[4];
401 	int level, i;
402 	int err = 0;
403 
404 	level = get_node_path(index, offset, noffset);
405 
406 	nids[0] = dn->inode->i_ino;
407 	npage[0] = get_node_page(sbi, nids[0]);
408 	if (IS_ERR(npage[0]))
409 		return PTR_ERR(npage[0]);
410 
411 	parent = npage[0];
412 	if (level != 0)
413 		nids[1] = get_nid(parent, offset[0], true);
414 	dn->inode_page = npage[0];
415 	dn->inode_page_locked = true;
416 
417 	/* get indirect or direct nodes */
418 	for (i = 1; i <= level; i++) {
419 		bool done = false;
420 
421 		if (!nids[i] && mode == ALLOC_NODE) {
422 			/* alloc new node */
423 			if (!alloc_nid(sbi, &(nids[i]))) {
424 				err = -ENOSPC;
425 				goto release_pages;
426 			}
427 
428 			dn->nid = nids[i];
429 			npage[i] = new_node_page(dn, noffset[i]);
430 			if (IS_ERR(npage[i])) {
431 				alloc_nid_failed(sbi, nids[i]);
432 				err = PTR_ERR(npage[i]);
433 				goto release_pages;
434 			}
435 
436 			set_nid(parent, offset[i - 1], nids[i], i == 1);
437 			alloc_nid_done(sbi, nids[i]);
438 			done = true;
439 		} else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
440 			npage[i] = get_node_page_ra(parent, offset[i - 1]);
441 			if (IS_ERR(npage[i])) {
442 				err = PTR_ERR(npage[i]);
443 				goto release_pages;
444 			}
445 			done = true;
446 		}
447 		if (i == 1) {
448 			dn->inode_page_locked = false;
449 			unlock_page(parent);
450 		} else {
451 			f2fs_put_page(parent, 1);
452 		}
453 
454 		if (!done) {
455 			npage[i] = get_node_page(sbi, nids[i]);
456 			if (IS_ERR(npage[i])) {
457 				err = PTR_ERR(npage[i]);
458 				f2fs_put_page(npage[0], 0);
459 				goto release_out;
460 			}
461 		}
462 		if (i < level) {
463 			parent = npage[i];
464 			nids[i + 1] = get_nid(parent, offset[i], false);
465 		}
466 	}
467 	dn->nid = nids[level];
468 	dn->ofs_in_node = offset[level];
469 	dn->node_page = npage[level];
470 	dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
471 	return 0;
472 
473 release_pages:
474 	f2fs_put_page(parent, 1);
475 	if (i > 1)
476 		f2fs_put_page(npage[0], 0);
477 release_out:
478 	dn->inode_page = NULL;
479 	dn->node_page = NULL;
480 	return err;
481 }
482 
483 static void truncate_node(struct dnode_of_data *dn)
484 {
485 	struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
486 	struct node_info ni;
487 
488 	get_node_info(sbi, dn->nid, &ni);
489 	if (dn->inode->i_blocks == 0) {
490 		BUG_ON(ni.blk_addr != NULL_ADDR);
491 		goto invalidate;
492 	}
493 	BUG_ON(ni.blk_addr == NULL_ADDR);
494 
495 	/* Deallocate node address */
496 	invalidate_blocks(sbi, ni.blk_addr);
497 	dec_valid_node_count(sbi, dn->inode, 1);
498 	set_node_addr(sbi, &ni, NULL_ADDR);
499 
500 	if (dn->nid == dn->inode->i_ino) {
501 		remove_orphan_inode(sbi, dn->nid);
502 		dec_valid_inode_count(sbi);
503 	} else {
504 		sync_inode_page(dn);
505 	}
506 invalidate:
507 	clear_node_page_dirty(dn->node_page);
508 	F2FS_SET_SB_DIRT(sbi);
509 
510 	f2fs_put_page(dn->node_page, 1);
511 	dn->node_page = NULL;
512 	trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
513 }
514 
515 static int truncate_dnode(struct dnode_of_data *dn)
516 {
517 	struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
518 	struct page *page;
519 
520 	if (dn->nid == 0)
521 		return 1;
522 
523 	/* get direct node */
524 	page = get_node_page(sbi, dn->nid);
525 	if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
526 		return 1;
527 	else if (IS_ERR(page))
528 		return PTR_ERR(page);
529 
530 	/* Make dnode_of_data for parameter */
531 	dn->node_page = page;
532 	dn->ofs_in_node = 0;
533 	truncate_data_blocks(dn);
534 	truncate_node(dn);
535 	return 1;
536 }
537 
538 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
539 						int ofs, int depth)
540 {
541 	struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
542 	struct dnode_of_data rdn = *dn;
543 	struct page *page;
544 	struct f2fs_node *rn;
545 	nid_t child_nid;
546 	unsigned int child_nofs;
547 	int freed = 0;
548 	int i, ret;
549 
550 	if (dn->nid == 0)
551 		return NIDS_PER_BLOCK + 1;
552 
553 	trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
554 
555 	page = get_node_page(sbi, dn->nid);
556 	if (IS_ERR(page)) {
557 		trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
558 		return PTR_ERR(page);
559 	}
560 
561 	rn = (struct f2fs_node *)page_address(page);
562 	if (depth < 3) {
563 		for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
564 			child_nid = le32_to_cpu(rn->in.nid[i]);
565 			if (child_nid == 0)
566 				continue;
567 			rdn.nid = child_nid;
568 			ret = truncate_dnode(&rdn);
569 			if (ret < 0)
570 				goto out_err;
571 			set_nid(page, i, 0, false);
572 		}
573 	} else {
574 		child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
575 		for (i = ofs; i < NIDS_PER_BLOCK; i++) {
576 			child_nid = le32_to_cpu(rn->in.nid[i]);
577 			if (child_nid == 0) {
578 				child_nofs += NIDS_PER_BLOCK + 1;
579 				continue;
580 			}
581 			rdn.nid = child_nid;
582 			ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
583 			if (ret == (NIDS_PER_BLOCK + 1)) {
584 				set_nid(page, i, 0, false);
585 				child_nofs += ret;
586 			} else if (ret < 0 && ret != -ENOENT) {
587 				goto out_err;
588 			}
589 		}
590 		freed = child_nofs;
591 	}
592 
593 	if (!ofs) {
594 		/* remove current indirect node */
595 		dn->node_page = page;
596 		truncate_node(dn);
597 		freed++;
598 	} else {
599 		f2fs_put_page(page, 1);
600 	}
601 	trace_f2fs_truncate_nodes_exit(dn->inode, freed);
602 	return freed;
603 
604 out_err:
605 	f2fs_put_page(page, 1);
606 	trace_f2fs_truncate_nodes_exit(dn->inode, ret);
607 	return ret;
608 }
609 
610 static int truncate_partial_nodes(struct dnode_of_data *dn,
611 			struct f2fs_inode *ri, int *offset, int depth)
612 {
613 	struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
614 	struct page *pages[2];
615 	nid_t nid[3];
616 	nid_t child_nid;
617 	int err = 0;
618 	int i;
619 	int idx = depth - 2;
620 
621 	nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
622 	if (!nid[0])
623 		return 0;
624 
625 	/* get indirect nodes in the path */
626 	for (i = 0; i < depth - 1; i++) {
627 		/* refernece count'll be increased */
628 		pages[i] = get_node_page(sbi, nid[i]);
629 		if (IS_ERR(pages[i])) {
630 			depth = i + 1;
631 			err = PTR_ERR(pages[i]);
632 			goto fail;
633 		}
634 		nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
635 	}
636 
637 	/* free direct nodes linked to a partial indirect node */
638 	for (i = offset[depth - 1]; i < NIDS_PER_BLOCK; i++) {
639 		child_nid = get_nid(pages[idx], i, false);
640 		if (!child_nid)
641 			continue;
642 		dn->nid = child_nid;
643 		err = truncate_dnode(dn);
644 		if (err < 0)
645 			goto fail;
646 		set_nid(pages[idx], i, 0, false);
647 	}
648 
649 	if (offset[depth - 1] == 0) {
650 		dn->node_page = pages[idx];
651 		dn->nid = nid[idx];
652 		truncate_node(dn);
653 	} else {
654 		f2fs_put_page(pages[idx], 1);
655 	}
656 	offset[idx]++;
657 	offset[depth - 1] = 0;
658 fail:
659 	for (i = depth - 3; i >= 0; i--)
660 		f2fs_put_page(pages[i], 1);
661 
662 	trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
663 
664 	return err;
665 }
666 
667 /*
668  * All the block addresses of data and nodes should be nullified.
669  */
670 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
671 {
672 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
673 	int err = 0, cont = 1;
674 	int level, offset[4], noffset[4];
675 	unsigned int nofs = 0;
676 	struct f2fs_node *rn;
677 	struct dnode_of_data dn;
678 	struct page *page;
679 
680 	trace_f2fs_truncate_inode_blocks_enter(inode, from);
681 
682 	level = get_node_path(from, offset, noffset);
683 
684 	page = get_node_page(sbi, inode->i_ino);
685 	if (IS_ERR(page)) {
686 		trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
687 		return PTR_ERR(page);
688 	}
689 
690 	set_new_dnode(&dn, inode, page, NULL, 0);
691 	unlock_page(page);
692 
693 	rn = page_address(page);
694 	switch (level) {
695 	case 0:
696 	case 1:
697 		nofs = noffset[1];
698 		break;
699 	case 2:
700 		nofs = noffset[1];
701 		if (!offset[level - 1])
702 			goto skip_partial;
703 		err = truncate_partial_nodes(&dn, &rn->i, offset, level);
704 		if (err < 0 && err != -ENOENT)
705 			goto fail;
706 		nofs += 1 + NIDS_PER_BLOCK;
707 		break;
708 	case 3:
709 		nofs = 5 + 2 * NIDS_PER_BLOCK;
710 		if (!offset[level - 1])
711 			goto skip_partial;
712 		err = truncate_partial_nodes(&dn, &rn->i, offset, level);
713 		if (err < 0 && err != -ENOENT)
714 			goto fail;
715 		break;
716 	default:
717 		BUG();
718 	}
719 
720 skip_partial:
721 	while (cont) {
722 		dn.nid = le32_to_cpu(rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]);
723 		switch (offset[0]) {
724 		case NODE_DIR1_BLOCK:
725 		case NODE_DIR2_BLOCK:
726 			err = truncate_dnode(&dn);
727 			break;
728 
729 		case NODE_IND1_BLOCK:
730 		case NODE_IND2_BLOCK:
731 			err = truncate_nodes(&dn, nofs, offset[1], 2);
732 			break;
733 
734 		case NODE_DIND_BLOCK:
735 			err = truncate_nodes(&dn, nofs, offset[1], 3);
736 			cont = 0;
737 			break;
738 
739 		default:
740 			BUG();
741 		}
742 		if (err < 0 && err != -ENOENT)
743 			goto fail;
744 		if (offset[1] == 0 &&
745 				rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK]) {
746 			lock_page(page);
747 			wait_on_page_writeback(page);
748 			rn->i.i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
749 			set_page_dirty(page);
750 			unlock_page(page);
751 		}
752 		offset[1] = 0;
753 		offset[0]++;
754 		nofs += err;
755 	}
756 fail:
757 	f2fs_put_page(page, 0);
758 	trace_f2fs_truncate_inode_blocks_exit(inode, err);
759 	return err > 0 ? 0 : err;
760 }
761 
762 /*
763  * Caller should grab and release a mutex by calling mutex_lock_op() and
764  * mutex_unlock_op().
765  */
766 int remove_inode_page(struct inode *inode)
767 {
768 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
769 	struct page *page;
770 	nid_t ino = inode->i_ino;
771 	struct dnode_of_data dn;
772 
773 	page = get_node_page(sbi, ino);
774 	if (IS_ERR(page))
775 		return PTR_ERR(page);
776 
777 	if (F2FS_I(inode)->i_xattr_nid) {
778 		nid_t nid = F2FS_I(inode)->i_xattr_nid;
779 		struct page *npage = get_node_page(sbi, nid);
780 
781 		if (IS_ERR(npage))
782 			return PTR_ERR(npage);
783 
784 		F2FS_I(inode)->i_xattr_nid = 0;
785 		set_new_dnode(&dn, inode, page, npage, nid);
786 		dn.inode_page_locked = 1;
787 		truncate_node(&dn);
788 	}
789 
790 	/* 0 is possible, after f2fs_new_inode() is failed */
791 	BUG_ON(inode->i_blocks != 0 && inode->i_blocks != 1);
792 	set_new_dnode(&dn, inode, page, page, ino);
793 	truncate_node(&dn);
794 	return 0;
795 }
796 
797 int new_inode_page(struct inode *inode, const struct qstr *name)
798 {
799 	struct page *page;
800 	struct dnode_of_data dn;
801 
802 	/* allocate inode page for new inode */
803 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
804 	page = new_node_page(&dn, 0);
805 	init_dent_inode(name, page);
806 	if (IS_ERR(page))
807 		return PTR_ERR(page);
808 	f2fs_put_page(page, 1);
809 	return 0;
810 }
811 
812 struct page *new_node_page(struct dnode_of_data *dn, unsigned int ofs)
813 {
814 	struct f2fs_sb_info *sbi = F2FS_SB(dn->inode->i_sb);
815 	struct address_space *mapping = sbi->node_inode->i_mapping;
816 	struct node_info old_ni, new_ni;
817 	struct page *page;
818 	int err;
819 
820 	if (is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC))
821 		return ERR_PTR(-EPERM);
822 
823 	page = grab_cache_page(mapping, dn->nid);
824 	if (!page)
825 		return ERR_PTR(-ENOMEM);
826 
827 	get_node_info(sbi, dn->nid, &old_ni);
828 
829 	SetPageUptodate(page);
830 	fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
831 
832 	/* Reinitialize old_ni with new node page */
833 	BUG_ON(old_ni.blk_addr != NULL_ADDR);
834 	new_ni = old_ni;
835 	new_ni.ino = dn->inode->i_ino;
836 
837 	if (!inc_valid_node_count(sbi, dn->inode, 1)) {
838 		err = -ENOSPC;
839 		goto fail;
840 	}
841 	set_node_addr(sbi, &new_ni, NEW_ADDR);
842 	set_cold_node(dn->inode, page);
843 
844 	dn->node_page = page;
845 	sync_inode_page(dn);
846 	set_page_dirty(page);
847 	if (ofs == 0)
848 		inc_valid_inode_count(sbi);
849 
850 	return page;
851 
852 fail:
853 	clear_node_page_dirty(page);
854 	f2fs_put_page(page, 1);
855 	return ERR_PTR(err);
856 }
857 
858 /*
859  * Caller should do after getting the following values.
860  * 0: f2fs_put_page(page, 0)
861  * LOCKED_PAGE: f2fs_put_page(page, 1)
862  * error: nothing
863  */
864 static int read_node_page(struct page *page, int type)
865 {
866 	struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
867 	struct node_info ni;
868 
869 	get_node_info(sbi, page->index, &ni);
870 
871 	if (ni.blk_addr == NULL_ADDR) {
872 		f2fs_put_page(page, 1);
873 		return -ENOENT;
874 	}
875 
876 	if (PageUptodate(page))
877 		return LOCKED_PAGE;
878 
879 	return f2fs_readpage(sbi, page, ni.blk_addr, type);
880 }
881 
882 /*
883  * Readahead a node page
884  */
885 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
886 {
887 	struct address_space *mapping = sbi->node_inode->i_mapping;
888 	struct page *apage;
889 	int err;
890 
891 	apage = find_get_page(mapping, nid);
892 	if (apage && PageUptodate(apage)) {
893 		f2fs_put_page(apage, 0);
894 		return;
895 	}
896 	f2fs_put_page(apage, 0);
897 
898 	apage = grab_cache_page(mapping, nid);
899 	if (!apage)
900 		return;
901 
902 	err = read_node_page(apage, READA);
903 	if (err == 0)
904 		f2fs_put_page(apage, 0);
905 	else if (err == LOCKED_PAGE)
906 		f2fs_put_page(apage, 1);
907 	return;
908 }
909 
910 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
911 {
912 	struct address_space *mapping = sbi->node_inode->i_mapping;
913 	struct page *page;
914 	int err;
915 
916 	page = grab_cache_page(mapping, nid);
917 	if (!page)
918 		return ERR_PTR(-ENOMEM);
919 
920 	err = read_node_page(page, READ_SYNC);
921 	if (err < 0)
922 		return ERR_PTR(err);
923 	else if (err == LOCKED_PAGE)
924 		goto got_it;
925 
926 	lock_page(page);
927 	if (!PageUptodate(page)) {
928 		f2fs_put_page(page, 1);
929 		return ERR_PTR(-EIO);
930 	}
931 got_it:
932 	BUG_ON(nid != nid_of_node(page));
933 	mark_page_accessed(page);
934 	return page;
935 }
936 
937 /*
938  * Return a locked page for the desired node page.
939  * And, readahead MAX_RA_NODE number of node pages.
940  */
941 struct page *get_node_page_ra(struct page *parent, int start)
942 {
943 	struct f2fs_sb_info *sbi = F2FS_SB(parent->mapping->host->i_sb);
944 	struct address_space *mapping = sbi->node_inode->i_mapping;
945 	struct page *page;
946 	int err, i, end;
947 	nid_t nid;
948 
949 	/* First, try getting the desired direct node. */
950 	nid = get_nid(parent, start, false);
951 	if (!nid)
952 		return ERR_PTR(-ENOENT);
953 
954 	page = grab_cache_page(mapping, nid);
955 	if (!page)
956 		return ERR_PTR(-ENOMEM);
957 
958 	err = read_node_page(page, READ_SYNC);
959 	if (err < 0)
960 		return ERR_PTR(err);
961 	else if (err == LOCKED_PAGE)
962 		goto page_hit;
963 
964 	/* Then, try readahead for siblings of the desired node */
965 	end = start + MAX_RA_NODE;
966 	end = min(end, NIDS_PER_BLOCK);
967 	for (i = start + 1; i < end; i++) {
968 		nid = get_nid(parent, i, false);
969 		if (!nid)
970 			continue;
971 		ra_node_page(sbi, nid);
972 	}
973 
974 	lock_page(page);
975 
976 page_hit:
977 	if (!PageUptodate(page)) {
978 		f2fs_put_page(page, 1);
979 		return ERR_PTR(-EIO);
980 	}
981 	mark_page_accessed(page);
982 	return page;
983 }
984 
985 void sync_inode_page(struct dnode_of_data *dn)
986 {
987 	if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
988 		update_inode(dn->inode, dn->node_page);
989 	} else if (dn->inode_page) {
990 		if (!dn->inode_page_locked)
991 			lock_page(dn->inode_page);
992 		update_inode(dn->inode, dn->inode_page);
993 		if (!dn->inode_page_locked)
994 			unlock_page(dn->inode_page);
995 	} else {
996 		update_inode_page(dn->inode);
997 	}
998 }
999 
1000 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1001 					struct writeback_control *wbc)
1002 {
1003 	struct address_space *mapping = sbi->node_inode->i_mapping;
1004 	pgoff_t index, end;
1005 	struct pagevec pvec;
1006 	int step = ino ? 2 : 0;
1007 	int nwritten = 0, wrote = 0;
1008 
1009 	pagevec_init(&pvec, 0);
1010 
1011 next_step:
1012 	index = 0;
1013 	end = LONG_MAX;
1014 
1015 	while (index <= end) {
1016 		int i, nr_pages;
1017 		nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
1018 				PAGECACHE_TAG_DIRTY,
1019 				min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1020 		if (nr_pages == 0)
1021 			break;
1022 
1023 		for (i = 0; i < nr_pages; i++) {
1024 			struct page *page = pvec.pages[i];
1025 
1026 			/*
1027 			 * flushing sequence with step:
1028 			 * 0. indirect nodes
1029 			 * 1. dentry dnodes
1030 			 * 2. file dnodes
1031 			 */
1032 			if (step == 0 && IS_DNODE(page))
1033 				continue;
1034 			if (step == 1 && (!IS_DNODE(page) ||
1035 						is_cold_node(page)))
1036 				continue;
1037 			if (step == 2 && (!IS_DNODE(page) ||
1038 						!is_cold_node(page)))
1039 				continue;
1040 
1041 			/*
1042 			 * If an fsync mode,
1043 			 * we should not skip writing node pages.
1044 			 */
1045 			if (ino && ino_of_node(page) == ino)
1046 				lock_page(page);
1047 			else if (!trylock_page(page))
1048 				continue;
1049 
1050 			if (unlikely(page->mapping != mapping)) {
1051 continue_unlock:
1052 				unlock_page(page);
1053 				continue;
1054 			}
1055 			if (ino && ino_of_node(page) != ino)
1056 				goto continue_unlock;
1057 
1058 			if (!PageDirty(page)) {
1059 				/* someone wrote it for us */
1060 				goto continue_unlock;
1061 			}
1062 
1063 			if (!clear_page_dirty_for_io(page))
1064 				goto continue_unlock;
1065 
1066 			/* called by fsync() */
1067 			if (ino && IS_DNODE(page)) {
1068 				int mark = !is_checkpointed_node(sbi, ino);
1069 				set_fsync_mark(page, 1);
1070 				if (IS_INODE(page))
1071 					set_dentry_mark(page, mark);
1072 				nwritten++;
1073 			} else {
1074 				set_fsync_mark(page, 0);
1075 				set_dentry_mark(page, 0);
1076 			}
1077 			mapping->a_ops->writepage(page, wbc);
1078 			wrote++;
1079 
1080 			if (--wbc->nr_to_write == 0)
1081 				break;
1082 		}
1083 		pagevec_release(&pvec);
1084 		cond_resched();
1085 
1086 		if (wbc->nr_to_write == 0) {
1087 			step = 2;
1088 			break;
1089 		}
1090 	}
1091 
1092 	if (step < 2) {
1093 		step++;
1094 		goto next_step;
1095 	}
1096 
1097 	if (wrote)
1098 		f2fs_submit_bio(sbi, NODE, wbc->sync_mode == WB_SYNC_ALL);
1099 
1100 	return nwritten;
1101 }
1102 
1103 static int f2fs_write_node_page(struct page *page,
1104 				struct writeback_control *wbc)
1105 {
1106 	struct f2fs_sb_info *sbi = F2FS_SB(page->mapping->host->i_sb);
1107 	nid_t nid;
1108 	block_t new_addr;
1109 	struct node_info ni;
1110 
1111 	wait_on_page_writeback(page);
1112 
1113 	/* get old block addr of this node page */
1114 	nid = nid_of_node(page);
1115 	BUG_ON(page->index != nid);
1116 
1117 	get_node_info(sbi, nid, &ni);
1118 
1119 	/* This page is already truncated */
1120 	if (ni.blk_addr == NULL_ADDR) {
1121 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1122 		unlock_page(page);
1123 		return 0;
1124 	}
1125 
1126 	if (wbc->for_reclaim) {
1127 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1128 		wbc->pages_skipped++;
1129 		set_page_dirty(page);
1130 		return AOP_WRITEPAGE_ACTIVATE;
1131 	}
1132 
1133 	mutex_lock(&sbi->node_write);
1134 	set_page_writeback(page);
1135 	write_node_page(sbi, page, nid, ni.blk_addr, &new_addr);
1136 	set_node_addr(sbi, &ni, new_addr);
1137 	dec_page_count(sbi, F2FS_DIRTY_NODES);
1138 	mutex_unlock(&sbi->node_write);
1139 	unlock_page(page);
1140 	return 0;
1141 }
1142 
1143 /*
1144  * It is very important to gather dirty pages and write at once, so that we can
1145  * submit a big bio without interfering other data writes.
1146  * Be default, 512 pages (2MB), a segment size, is quite reasonable.
1147  */
1148 #define COLLECT_DIRTY_NODES	512
1149 static int f2fs_write_node_pages(struct address_space *mapping,
1150 			    struct writeback_control *wbc)
1151 {
1152 	struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1153 	struct block_device *bdev = sbi->sb->s_bdev;
1154 	long nr_to_write = wbc->nr_to_write;
1155 
1156 	/* First check balancing cached NAT entries */
1157 	if (try_to_free_nats(sbi, NAT_ENTRY_PER_BLOCK)) {
1158 		f2fs_sync_fs(sbi->sb, true);
1159 		return 0;
1160 	}
1161 
1162 	/* collect a number of dirty node pages and write together */
1163 	if (get_pages(sbi, F2FS_DIRTY_NODES) < COLLECT_DIRTY_NODES)
1164 		return 0;
1165 
1166 	/* if mounting is failed, skip writing node pages */
1167 	wbc->nr_to_write = bio_get_nr_vecs(bdev);
1168 	sync_node_pages(sbi, 0, wbc);
1169 	wbc->nr_to_write = nr_to_write -
1170 		(bio_get_nr_vecs(bdev) - wbc->nr_to_write);
1171 	return 0;
1172 }
1173 
1174 static int f2fs_set_node_page_dirty(struct page *page)
1175 {
1176 	struct address_space *mapping = page->mapping;
1177 	struct f2fs_sb_info *sbi = F2FS_SB(mapping->host->i_sb);
1178 
1179 	SetPageUptodate(page);
1180 	if (!PageDirty(page)) {
1181 		__set_page_dirty_nobuffers(page);
1182 		inc_page_count(sbi, F2FS_DIRTY_NODES);
1183 		SetPagePrivate(page);
1184 		return 1;
1185 	}
1186 	return 0;
1187 }
1188 
1189 static void f2fs_invalidate_node_page(struct page *page, unsigned long offset)
1190 {
1191 	struct inode *inode = page->mapping->host;
1192 	struct f2fs_sb_info *sbi = F2FS_SB(inode->i_sb);
1193 	if (PageDirty(page))
1194 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1195 	ClearPagePrivate(page);
1196 }
1197 
1198 static int f2fs_release_node_page(struct page *page, gfp_t wait)
1199 {
1200 	ClearPagePrivate(page);
1201 	return 1;
1202 }
1203 
1204 /*
1205  * Structure of the f2fs node operations
1206  */
1207 const struct address_space_operations f2fs_node_aops = {
1208 	.writepage	= f2fs_write_node_page,
1209 	.writepages	= f2fs_write_node_pages,
1210 	.set_page_dirty	= f2fs_set_node_page_dirty,
1211 	.invalidatepage	= f2fs_invalidate_node_page,
1212 	.releasepage	= f2fs_release_node_page,
1213 };
1214 
1215 static struct free_nid *__lookup_free_nid_list(nid_t n, struct list_head *head)
1216 {
1217 	struct list_head *this;
1218 	struct free_nid *i;
1219 	list_for_each(this, head) {
1220 		i = list_entry(this, struct free_nid, list);
1221 		if (i->nid == n)
1222 			return i;
1223 	}
1224 	return NULL;
1225 }
1226 
1227 static void __del_from_free_nid_list(struct free_nid *i)
1228 {
1229 	list_del(&i->list);
1230 	kmem_cache_free(free_nid_slab, i);
1231 }
1232 
1233 static int add_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1234 {
1235 	struct free_nid *i;
1236 
1237 	if (nm_i->fcnt > 2 * MAX_FREE_NIDS)
1238 		return 0;
1239 retry:
1240 	i = kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1241 	if (!i) {
1242 		cond_resched();
1243 		goto retry;
1244 	}
1245 	i->nid = nid;
1246 	i->state = NID_NEW;
1247 
1248 	spin_lock(&nm_i->free_nid_list_lock);
1249 	if (__lookup_free_nid_list(nid, &nm_i->free_nid_list)) {
1250 		spin_unlock(&nm_i->free_nid_list_lock);
1251 		kmem_cache_free(free_nid_slab, i);
1252 		return 0;
1253 	}
1254 	list_add_tail(&i->list, &nm_i->free_nid_list);
1255 	nm_i->fcnt++;
1256 	spin_unlock(&nm_i->free_nid_list_lock);
1257 	return 1;
1258 }
1259 
1260 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1261 {
1262 	struct free_nid *i;
1263 	spin_lock(&nm_i->free_nid_list_lock);
1264 	i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1265 	if (i && i->state == NID_NEW) {
1266 		__del_from_free_nid_list(i);
1267 		nm_i->fcnt--;
1268 	}
1269 	spin_unlock(&nm_i->free_nid_list_lock);
1270 }
1271 
1272 static int scan_nat_page(struct f2fs_nm_info *nm_i,
1273 			struct page *nat_page, nid_t start_nid)
1274 {
1275 	struct f2fs_nat_block *nat_blk = page_address(nat_page);
1276 	block_t blk_addr;
1277 	int fcnt = 0;
1278 	int i;
1279 
1280 	/* 0 nid should not be used */
1281 	if (start_nid == 0)
1282 		++start_nid;
1283 
1284 	i = start_nid % NAT_ENTRY_PER_BLOCK;
1285 
1286 	for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1287 		if (start_nid >= nm_i->max_nid)
1288 			break;
1289 		blk_addr  = le32_to_cpu(nat_blk->entries[i].block_addr);
1290 		BUG_ON(blk_addr == NEW_ADDR);
1291 		if (blk_addr == NULL_ADDR)
1292 			fcnt += add_free_nid(nm_i, start_nid);
1293 	}
1294 	return fcnt;
1295 }
1296 
1297 static void build_free_nids(struct f2fs_sb_info *sbi)
1298 {
1299 	struct free_nid *fnid, *next_fnid;
1300 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1301 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1302 	struct f2fs_summary_block *sum = curseg->sum_blk;
1303 	nid_t nid = 0;
1304 	bool is_cycled = false;
1305 	int fcnt = 0;
1306 	int i;
1307 
1308 	nid = nm_i->next_scan_nid;
1309 	nm_i->init_scan_nid = nid;
1310 
1311 	ra_nat_pages(sbi, nid);
1312 
1313 	while (1) {
1314 		struct page *page = get_current_nat_page(sbi, nid);
1315 
1316 		fcnt += scan_nat_page(nm_i, page, nid);
1317 		f2fs_put_page(page, 1);
1318 
1319 		nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1320 
1321 		if (nid >= nm_i->max_nid) {
1322 			nid = 0;
1323 			is_cycled = true;
1324 		}
1325 		if (fcnt > MAX_FREE_NIDS)
1326 			break;
1327 		if (is_cycled && nm_i->init_scan_nid <= nid)
1328 			break;
1329 	}
1330 
1331 	/* go to the next nat page in order to reuse free nids first */
1332 	nm_i->next_scan_nid = nm_i->init_scan_nid + NAT_ENTRY_PER_BLOCK;
1333 
1334 	/* find free nids from current sum_pages */
1335 	mutex_lock(&curseg->curseg_mutex);
1336 	for (i = 0; i < nats_in_cursum(sum); i++) {
1337 		block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1338 		nid = le32_to_cpu(nid_in_journal(sum, i));
1339 		if (addr == NULL_ADDR)
1340 			add_free_nid(nm_i, nid);
1341 		else
1342 			remove_free_nid(nm_i, nid);
1343 	}
1344 	mutex_unlock(&curseg->curseg_mutex);
1345 
1346 	/* remove the free nids from current allocated nids */
1347 	list_for_each_entry_safe(fnid, next_fnid, &nm_i->free_nid_list, list) {
1348 		struct nat_entry *ne;
1349 
1350 		read_lock(&nm_i->nat_tree_lock);
1351 		ne = __lookup_nat_cache(nm_i, fnid->nid);
1352 		if (ne && nat_get_blkaddr(ne) != NULL_ADDR)
1353 			remove_free_nid(nm_i, fnid->nid);
1354 		read_unlock(&nm_i->nat_tree_lock);
1355 	}
1356 }
1357 
1358 /*
1359  * If this function returns success, caller can obtain a new nid
1360  * from second parameter of this function.
1361  * The returned nid could be used ino as well as nid when inode is created.
1362  */
1363 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1364 {
1365 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1366 	struct free_nid *i = NULL;
1367 	struct list_head *this;
1368 retry:
1369 	mutex_lock(&nm_i->build_lock);
1370 	if (!nm_i->fcnt) {
1371 		/* scan NAT in order to build free nid list */
1372 		build_free_nids(sbi);
1373 		if (!nm_i->fcnt) {
1374 			mutex_unlock(&nm_i->build_lock);
1375 			return false;
1376 		}
1377 	}
1378 	mutex_unlock(&nm_i->build_lock);
1379 
1380 	/*
1381 	 * We check fcnt again since previous check is racy as
1382 	 * we didn't hold free_nid_list_lock. So other thread
1383 	 * could consume all of free nids.
1384 	 */
1385 	spin_lock(&nm_i->free_nid_list_lock);
1386 	if (!nm_i->fcnt) {
1387 		spin_unlock(&nm_i->free_nid_list_lock);
1388 		goto retry;
1389 	}
1390 
1391 	BUG_ON(list_empty(&nm_i->free_nid_list));
1392 	list_for_each(this, &nm_i->free_nid_list) {
1393 		i = list_entry(this, struct free_nid, list);
1394 		if (i->state == NID_NEW)
1395 			break;
1396 	}
1397 
1398 	BUG_ON(i->state != NID_NEW);
1399 	*nid = i->nid;
1400 	i->state = NID_ALLOC;
1401 	nm_i->fcnt--;
1402 	spin_unlock(&nm_i->free_nid_list_lock);
1403 	return true;
1404 }
1405 
1406 /*
1407  * alloc_nid() should be called prior to this function.
1408  */
1409 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1410 {
1411 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1412 	struct free_nid *i;
1413 
1414 	spin_lock(&nm_i->free_nid_list_lock);
1415 	i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1416 	BUG_ON(!i || i->state != NID_ALLOC);
1417 	__del_from_free_nid_list(i);
1418 	spin_unlock(&nm_i->free_nid_list_lock);
1419 }
1420 
1421 /*
1422  * alloc_nid() should be called prior to this function.
1423  */
1424 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1425 {
1426 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1427 	struct free_nid *i;
1428 
1429 	spin_lock(&nm_i->free_nid_list_lock);
1430 	i = __lookup_free_nid_list(nid, &nm_i->free_nid_list);
1431 	BUG_ON(!i || i->state != NID_ALLOC);
1432 	i->state = NID_NEW;
1433 	nm_i->fcnt++;
1434 	spin_unlock(&nm_i->free_nid_list_lock);
1435 }
1436 
1437 void recover_node_page(struct f2fs_sb_info *sbi, struct page *page,
1438 		struct f2fs_summary *sum, struct node_info *ni,
1439 		block_t new_blkaddr)
1440 {
1441 	rewrite_node_page(sbi, page, sum, ni->blk_addr, new_blkaddr);
1442 	set_node_addr(sbi, ni, new_blkaddr);
1443 	clear_node_page_dirty(page);
1444 }
1445 
1446 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1447 {
1448 	struct address_space *mapping = sbi->node_inode->i_mapping;
1449 	struct f2fs_node *src, *dst;
1450 	nid_t ino = ino_of_node(page);
1451 	struct node_info old_ni, new_ni;
1452 	struct page *ipage;
1453 
1454 	ipage = grab_cache_page(mapping, ino);
1455 	if (!ipage)
1456 		return -ENOMEM;
1457 
1458 	/* Should not use this inode  from free nid list */
1459 	remove_free_nid(NM_I(sbi), ino);
1460 
1461 	get_node_info(sbi, ino, &old_ni);
1462 	SetPageUptodate(ipage);
1463 	fill_node_footer(ipage, ino, ino, 0, true);
1464 
1465 	src = (struct f2fs_node *)page_address(page);
1466 	dst = (struct f2fs_node *)page_address(ipage);
1467 
1468 	memcpy(dst, src, (unsigned long)&src->i.i_ext - (unsigned long)&src->i);
1469 	dst->i.i_size = 0;
1470 	dst->i.i_blocks = cpu_to_le64(1);
1471 	dst->i.i_links = cpu_to_le32(1);
1472 	dst->i.i_xattr_nid = 0;
1473 
1474 	new_ni = old_ni;
1475 	new_ni.ino = ino;
1476 
1477 	set_node_addr(sbi, &new_ni, NEW_ADDR);
1478 	inc_valid_inode_count(sbi);
1479 
1480 	f2fs_put_page(ipage, 1);
1481 	return 0;
1482 }
1483 
1484 int restore_node_summary(struct f2fs_sb_info *sbi,
1485 			unsigned int segno, struct f2fs_summary_block *sum)
1486 {
1487 	struct f2fs_node *rn;
1488 	struct f2fs_summary *sum_entry;
1489 	struct page *page;
1490 	block_t addr;
1491 	int i, last_offset;
1492 
1493 	/* alloc temporal page for read node */
1494 	page = alloc_page(GFP_NOFS | __GFP_ZERO);
1495 	if (IS_ERR(page))
1496 		return PTR_ERR(page);
1497 	lock_page(page);
1498 
1499 	/* scan the node segment */
1500 	last_offset = sbi->blocks_per_seg;
1501 	addr = START_BLOCK(sbi, segno);
1502 	sum_entry = &sum->entries[0];
1503 
1504 	for (i = 0; i < last_offset; i++, sum_entry++) {
1505 		/*
1506 		 * In order to read next node page,
1507 		 * we must clear PageUptodate flag.
1508 		 */
1509 		ClearPageUptodate(page);
1510 
1511 		if (f2fs_readpage(sbi, page, addr, READ_SYNC))
1512 			goto out;
1513 
1514 		lock_page(page);
1515 		rn = (struct f2fs_node *)page_address(page);
1516 		sum_entry->nid = rn->footer.nid;
1517 		sum_entry->version = 0;
1518 		sum_entry->ofs_in_node = 0;
1519 		addr++;
1520 	}
1521 	unlock_page(page);
1522 out:
1523 	__free_pages(page, 0);
1524 	return 0;
1525 }
1526 
1527 static bool flush_nats_in_journal(struct f2fs_sb_info *sbi)
1528 {
1529 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1530 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1531 	struct f2fs_summary_block *sum = curseg->sum_blk;
1532 	int i;
1533 
1534 	mutex_lock(&curseg->curseg_mutex);
1535 
1536 	if (nats_in_cursum(sum) < NAT_JOURNAL_ENTRIES) {
1537 		mutex_unlock(&curseg->curseg_mutex);
1538 		return false;
1539 	}
1540 
1541 	for (i = 0; i < nats_in_cursum(sum); i++) {
1542 		struct nat_entry *ne;
1543 		struct f2fs_nat_entry raw_ne;
1544 		nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1545 
1546 		raw_ne = nat_in_journal(sum, i);
1547 retry:
1548 		write_lock(&nm_i->nat_tree_lock);
1549 		ne = __lookup_nat_cache(nm_i, nid);
1550 		if (ne) {
1551 			__set_nat_cache_dirty(nm_i, ne);
1552 			write_unlock(&nm_i->nat_tree_lock);
1553 			continue;
1554 		}
1555 		ne = grab_nat_entry(nm_i, nid);
1556 		if (!ne) {
1557 			write_unlock(&nm_i->nat_tree_lock);
1558 			goto retry;
1559 		}
1560 		nat_set_blkaddr(ne, le32_to_cpu(raw_ne.block_addr));
1561 		nat_set_ino(ne, le32_to_cpu(raw_ne.ino));
1562 		nat_set_version(ne, raw_ne.version);
1563 		__set_nat_cache_dirty(nm_i, ne);
1564 		write_unlock(&nm_i->nat_tree_lock);
1565 	}
1566 	update_nats_in_cursum(sum, -i);
1567 	mutex_unlock(&curseg->curseg_mutex);
1568 	return true;
1569 }
1570 
1571 /*
1572  * This function is called during the checkpointing process.
1573  */
1574 void flush_nat_entries(struct f2fs_sb_info *sbi)
1575 {
1576 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1577 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1578 	struct f2fs_summary_block *sum = curseg->sum_blk;
1579 	struct list_head *cur, *n;
1580 	struct page *page = NULL;
1581 	struct f2fs_nat_block *nat_blk = NULL;
1582 	nid_t start_nid = 0, end_nid = 0;
1583 	bool flushed;
1584 
1585 	flushed = flush_nats_in_journal(sbi);
1586 
1587 	if (!flushed)
1588 		mutex_lock(&curseg->curseg_mutex);
1589 
1590 	/* 1) flush dirty nat caches */
1591 	list_for_each_safe(cur, n, &nm_i->dirty_nat_entries) {
1592 		struct nat_entry *ne;
1593 		nid_t nid;
1594 		struct f2fs_nat_entry raw_ne;
1595 		int offset = -1;
1596 		block_t new_blkaddr;
1597 
1598 		ne = list_entry(cur, struct nat_entry, list);
1599 		nid = nat_get_nid(ne);
1600 
1601 		if (nat_get_blkaddr(ne) == NEW_ADDR)
1602 			continue;
1603 		if (flushed)
1604 			goto to_nat_page;
1605 
1606 		/* if there is room for nat enries in curseg->sumpage */
1607 		offset = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 1);
1608 		if (offset >= 0) {
1609 			raw_ne = nat_in_journal(sum, offset);
1610 			goto flush_now;
1611 		}
1612 to_nat_page:
1613 		if (!page || (start_nid > nid || nid > end_nid)) {
1614 			if (page) {
1615 				f2fs_put_page(page, 1);
1616 				page = NULL;
1617 			}
1618 			start_nid = START_NID(nid);
1619 			end_nid = start_nid + NAT_ENTRY_PER_BLOCK - 1;
1620 
1621 			/*
1622 			 * get nat block with dirty flag, increased reference
1623 			 * count, mapped and lock
1624 			 */
1625 			page = get_next_nat_page(sbi, start_nid);
1626 			nat_blk = page_address(page);
1627 		}
1628 
1629 		BUG_ON(!nat_blk);
1630 		raw_ne = nat_blk->entries[nid - start_nid];
1631 flush_now:
1632 		new_blkaddr = nat_get_blkaddr(ne);
1633 
1634 		raw_ne.ino = cpu_to_le32(nat_get_ino(ne));
1635 		raw_ne.block_addr = cpu_to_le32(new_blkaddr);
1636 		raw_ne.version = nat_get_version(ne);
1637 
1638 		if (offset < 0) {
1639 			nat_blk->entries[nid - start_nid] = raw_ne;
1640 		} else {
1641 			nat_in_journal(sum, offset) = raw_ne;
1642 			nid_in_journal(sum, offset) = cpu_to_le32(nid);
1643 		}
1644 
1645 		if (nat_get_blkaddr(ne) == NULL_ADDR &&
1646 					!add_free_nid(NM_I(sbi), nid)) {
1647 			write_lock(&nm_i->nat_tree_lock);
1648 			__del_from_nat_cache(nm_i, ne);
1649 			write_unlock(&nm_i->nat_tree_lock);
1650 		} else {
1651 			write_lock(&nm_i->nat_tree_lock);
1652 			__clear_nat_cache_dirty(nm_i, ne);
1653 			ne->checkpointed = true;
1654 			write_unlock(&nm_i->nat_tree_lock);
1655 		}
1656 	}
1657 	if (!flushed)
1658 		mutex_unlock(&curseg->curseg_mutex);
1659 	f2fs_put_page(page, 1);
1660 
1661 	/* 2) shrink nat caches if necessary */
1662 	try_to_free_nats(sbi, nm_i->nat_cnt - NM_WOUT_THRESHOLD);
1663 }
1664 
1665 static int init_node_manager(struct f2fs_sb_info *sbi)
1666 {
1667 	struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1668 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1669 	unsigned char *version_bitmap;
1670 	unsigned int nat_segs, nat_blocks;
1671 
1672 	nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1673 
1674 	/* segment_count_nat includes pair segment so divide to 2. */
1675 	nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1676 	nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1677 	nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1678 	nm_i->fcnt = 0;
1679 	nm_i->nat_cnt = 0;
1680 
1681 	INIT_LIST_HEAD(&nm_i->free_nid_list);
1682 	INIT_RADIX_TREE(&nm_i->nat_root, GFP_ATOMIC);
1683 	INIT_LIST_HEAD(&nm_i->nat_entries);
1684 	INIT_LIST_HEAD(&nm_i->dirty_nat_entries);
1685 
1686 	mutex_init(&nm_i->build_lock);
1687 	spin_lock_init(&nm_i->free_nid_list_lock);
1688 	rwlock_init(&nm_i->nat_tree_lock);
1689 
1690 	nm_i->init_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1691 	nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
1692 	nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
1693 	version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
1694 	if (!version_bitmap)
1695 		return -EFAULT;
1696 
1697 	nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
1698 					GFP_KERNEL);
1699 	if (!nm_i->nat_bitmap)
1700 		return -ENOMEM;
1701 	return 0;
1702 }
1703 
1704 int build_node_manager(struct f2fs_sb_info *sbi)
1705 {
1706 	int err;
1707 
1708 	sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
1709 	if (!sbi->nm_info)
1710 		return -ENOMEM;
1711 
1712 	err = init_node_manager(sbi);
1713 	if (err)
1714 		return err;
1715 
1716 	build_free_nids(sbi);
1717 	return 0;
1718 }
1719 
1720 void destroy_node_manager(struct f2fs_sb_info *sbi)
1721 {
1722 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1723 	struct free_nid *i, *next_i;
1724 	struct nat_entry *natvec[NATVEC_SIZE];
1725 	nid_t nid = 0;
1726 	unsigned int found;
1727 
1728 	if (!nm_i)
1729 		return;
1730 
1731 	/* destroy free nid list */
1732 	spin_lock(&nm_i->free_nid_list_lock);
1733 	list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
1734 		BUG_ON(i->state == NID_ALLOC);
1735 		__del_from_free_nid_list(i);
1736 		nm_i->fcnt--;
1737 	}
1738 	BUG_ON(nm_i->fcnt);
1739 	spin_unlock(&nm_i->free_nid_list_lock);
1740 
1741 	/* destroy nat cache */
1742 	write_lock(&nm_i->nat_tree_lock);
1743 	while ((found = __gang_lookup_nat_cache(nm_i,
1744 					nid, NATVEC_SIZE, natvec))) {
1745 		unsigned idx;
1746 		for (idx = 0; idx < found; idx++) {
1747 			struct nat_entry *e = natvec[idx];
1748 			nid = nat_get_nid(e) + 1;
1749 			__del_from_nat_cache(nm_i, e);
1750 		}
1751 	}
1752 	BUG_ON(nm_i->nat_cnt);
1753 	write_unlock(&nm_i->nat_tree_lock);
1754 
1755 	kfree(nm_i->nat_bitmap);
1756 	sbi->nm_info = NULL;
1757 	kfree(nm_i);
1758 }
1759 
1760 int __init create_node_manager_caches(void)
1761 {
1762 	nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
1763 			sizeof(struct nat_entry), NULL);
1764 	if (!nat_entry_slab)
1765 		return -ENOMEM;
1766 
1767 	free_nid_slab = f2fs_kmem_cache_create("free_nid",
1768 			sizeof(struct free_nid), NULL);
1769 	if (!free_nid_slab) {
1770 		kmem_cache_destroy(nat_entry_slab);
1771 		return -ENOMEM;
1772 	}
1773 	return 0;
1774 }
1775 
1776 void destroy_node_manager_caches(void)
1777 {
1778 	kmem_cache_destroy(free_nid_slab);
1779 	kmem_cache_destroy(nat_entry_slab);
1780 }
1781