xref: /openbmc/linux/fs/f2fs/node.c (revision 8c749ce9)
1 /*
2  * fs/f2fs/node.c
3  *
4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
5  *             http://www.samsung.com/
6  *
7  * This program is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License version 2 as
9  * published by the Free Software Foundation.
10  */
11 #include <linux/fs.h>
12 #include <linux/f2fs_fs.h>
13 #include <linux/mpage.h>
14 #include <linux/backing-dev.h>
15 #include <linux/blkdev.h>
16 #include <linux/pagevec.h>
17 #include <linux/swap.h>
18 
19 #include "f2fs.h"
20 #include "node.h"
21 #include "segment.h"
22 #include "trace.h"
23 #include <trace/events/f2fs.h>
24 
25 #define on_build_free_nids(nmi) mutex_is_locked(&nm_i->build_lock)
26 
27 static struct kmem_cache *nat_entry_slab;
28 static struct kmem_cache *free_nid_slab;
29 static struct kmem_cache *nat_entry_set_slab;
30 
31 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
32 {
33 	struct f2fs_nm_info *nm_i = NM_I(sbi);
34 	struct sysinfo val;
35 	unsigned long avail_ram;
36 	unsigned long mem_size = 0;
37 	bool res = false;
38 
39 	si_meminfo(&val);
40 
41 	/* only uses low memory */
42 	avail_ram = val.totalram - val.totalhigh;
43 
44 	/*
45 	 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
46 	 */
47 	if (type == FREE_NIDS) {
48 		mem_size = (nm_i->fcnt * sizeof(struct free_nid)) >>
49 							PAGE_CACHE_SHIFT;
50 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
51 	} else if (type == NAT_ENTRIES) {
52 		mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
53 							PAGE_CACHE_SHIFT;
54 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
55 	} else if (type == DIRTY_DENTS) {
56 		if (sbi->sb->s_bdi->wb.dirty_exceeded)
57 			return false;
58 		mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
59 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
60 	} else if (type == INO_ENTRIES) {
61 		int i;
62 
63 		for (i = 0; i <= UPDATE_INO; i++)
64 			mem_size += (sbi->im[i].ino_num *
65 				sizeof(struct ino_entry)) >> PAGE_CACHE_SHIFT;
66 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
67 	} else if (type == EXTENT_CACHE) {
68 		mem_size = (atomic_read(&sbi->total_ext_tree) *
69 				sizeof(struct extent_tree) +
70 				atomic_read(&sbi->total_ext_node) *
71 				sizeof(struct extent_node)) >> PAGE_CACHE_SHIFT;
72 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
73 	} else {
74 		if (!sbi->sb->s_bdi->wb.dirty_exceeded)
75 			return true;
76 	}
77 	return res;
78 }
79 
80 static void clear_node_page_dirty(struct page *page)
81 {
82 	struct address_space *mapping = page->mapping;
83 	unsigned int long flags;
84 
85 	if (PageDirty(page)) {
86 		spin_lock_irqsave(&mapping->tree_lock, flags);
87 		radix_tree_tag_clear(&mapping->page_tree,
88 				page_index(page),
89 				PAGECACHE_TAG_DIRTY);
90 		spin_unlock_irqrestore(&mapping->tree_lock, flags);
91 
92 		clear_page_dirty_for_io(page);
93 		dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
94 	}
95 	ClearPageUptodate(page);
96 }
97 
98 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
99 {
100 	pgoff_t index = current_nat_addr(sbi, nid);
101 	return get_meta_page(sbi, index);
102 }
103 
104 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
105 {
106 	struct page *src_page;
107 	struct page *dst_page;
108 	pgoff_t src_off;
109 	pgoff_t dst_off;
110 	void *src_addr;
111 	void *dst_addr;
112 	struct f2fs_nm_info *nm_i = NM_I(sbi);
113 
114 	src_off = current_nat_addr(sbi, nid);
115 	dst_off = next_nat_addr(sbi, src_off);
116 
117 	/* get current nat block page with lock */
118 	src_page = get_meta_page(sbi, src_off);
119 	dst_page = grab_meta_page(sbi, dst_off);
120 	f2fs_bug_on(sbi, PageDirty(src_page));
121 
122 	src_addr = page_address(src_page);
123 	dst_addr = page_address(dst_page);
124 	memcpy(dst_addr, src_addr, PAGE_CACHE_SIZE);
125 	set_page_dirty(dst_page);
126 	f2fs_put_page(src_page, 1);
127 
128 	set_to_next_nat(nm_i, nid);
129 
130 	return dst_page;
131 }
132 
133 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
134 {
135 	return radix_tree_lookup(&nm_i->nat_root, n);
136 }
137 
138 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
139 		nid_t start, unsigned int nr, struct nat_entry **ep)
140 {
141 	return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
142 }
143 
144 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
145 {
146 	list_del(&e->list);
147 	radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
148 	nm_i->nat_cnt--;
149 	kmem_cache_free(nat_entry_slab, e);
150 }
151 
152 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
153 						struct nat_entry *ne)
154 {
155 	nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
156 	struct nat_entry_set *head;
157 
158 	if (get_nat_flag(ne, IS_DIRTY))
159 		return;
160 
161 	head = radix_tree_lookup(&nm_i->nat_set_root, set);
162 	if (!head) {
163 		head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
164 
165 		INIT_LIST_HEAD(&head->entry_list);
166 		INIT_LIST_HEAD(&head->set_list);
167 		head->set = set;
168 		head->entry_cnt = 0;
169 		f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
170 	}
171 	list_move_tail(&ne->list, &head->entry_list);
172 	nm_i->dirty_nat_cnt++;
173 	head->entry_cnt++;
174 	set_nat_flag(ne, IS_DIRTY, true);
175 }
176 
177 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
178 						struct nat_entry *ne)
179 {
180 	nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
181 	struct nat_entry_set *head;
182 
183 	head = radix_tree_lookup(&nm_i->nat_set_root, set);
184 	if (head) {
185 		list_move_tail(&ne->list, &nm_i->nat_entries);
186 		set_nat_flag(ne, IS_DIRTY, false);
187 		head->entry_cnt--;
188 		nm_i->dirty_nat_cnt--;
189 	}
190 }
191 
192 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
193 		nid_t start, unsigned int nr, struct nat_entry_set **ep)
194 {
195 	return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
196 							start, nr);
197 }
198 
199 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
200 {
201 	struct f2fs_nm_info *nm_i = NM_I(sbi);
202 	struct nat_entry *e;
203 	bool need = false;
204 
205 	down_read(&nm_i->nat_tree_lock);
206 	e = __lookup_nat_cache(nm_i, nid);
207 	if (e) {
208 		if (!get_nat_flag(e, IS_CHECKPOINTED) &&
209 				!get_nat_flag(e, HAS_FSYNCED_INODE))
210 			need = true;
211 	}
212 	up_read(&nm_i->nat_tree_lock);
213 	return need;
214 }
215 
216 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
217 {
218 	struct f2fs_nm_info *nm_i = NM_I(sbi);
219 	struct nat_entry *e;
220 	bool is_cp = true;
221 
222 	down_read(&nm_i->nat_tree_lock);
223 	e = __lookup_nat_cache(nm_i, nid);
224 	if (e && !get_nat_flag(e, IS_CHECKPOINTED))
225 		is_cp = false;
226 	up_read(&nm_i->nat_tree_lock);
227 	return is_cp;
228 }
229 
230 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
231 {
232 	struct f2fs_nm_info *nm_i = NM_I(sbi);
233 	struct nat_entry *e;
234 	bool need_update = true;
235 
236 	down_read(&nm_i->nat_tree_lock);
237 	e = __lookup_nat_cache(nm_i, ino);
238 	if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
239 			(get_nat_flag(e, IS_CHECKPOINTED) ||
240 			 get_nat_flag(e, HAS_FSYNCED_INODE)))
241 		need_update = false;
242 	up_read(&nm_i->nat_tree_lock);
243 	return need_update;
244 }
245 
246 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
247 {
248 	struct nat_entry *new;
249 
250 	new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
251 	f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
252 	memset(new, 0, sizeof(struct nat_entry));
253 	nat_set_nid(new, nid);
254 	nat_reset_flag(new);
255 	list_add_tail(&new->list, &nm_i->nat_entries);
256 	nm_i->nat_cnt++;
257 	return new;
258 }
259 
260 static void cache_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid,
261 						struct f2fs_nat_entry *ne)
262 {
263 	struct nat_entry *e;
264 
265 	e = __lookup_nat_cache(nm_i, nid);
266 	if (!e) {
267 		e = grab_nat_entry(nm_i, nid);
268 		node_info_from_raw_nat(&e->ni, ne);
269 	}
270 }
271 
272 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
273 			block_t new_blkaddr, bool fsync_done)
274 {
275 	struct f2fs_nm_info *nm_i = NM_I(sbi);
276 	struct nat_entry *e;
277 
278 	down_write(&nm_i->nat_tree_lock);
279 	e = __lookup_nat_cache(nm_i, ni->nid);
280 	if (!e) {
281 		e = grab_nat_entry(nm_i, ni->nid);
282 		copy_node_info(&e->ni, ni);
283 		f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
284 	} else if (new_blkaddr == NEW_ADDR) {
285 		/*
286 		 * when nid is reallocated,
287 		 * previous nat entry can be remained in nat cache.
288 		 * So, reinitialize it with new information.
289 		 */
290 		copy_node_info(&e->ni, ni);
291 		f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
292 	}
293 
294 	/* sanity check */
295 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
296 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
297 			new_blkaddr == NULL_ADDR);
298 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
299 			new_blkaddr == NEW_ADDR);
300 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
301 			nat_get_blkaddr(e) != NULL_ADDR &&
302 			new_blkaddr == NEW_ADDR);
303 
304 	/* increment version no as node is removed */
305 	if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
306 		unsigned char version = nat_get_version(e);
307 		nat_set_version(e, inc_node_version(version));
308 
309 		/* in order to reuse the nid */
310 		if (nm_i->next_scan_nid > ni->nid)
311 			nm_i->next_scan_nid = ni->nid;
312 	}
313 
314 	/* change address */
315 	nat_set_blkaddr(e, new_blkaddr);
316 	if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
317 		set_nat_flag(e, IS_CHECKPOINTED, false);
318 	__set_nat_cache_dirty(nm_i, e);
319 
320 	/* update fsync_mark if its inode nat entry is still alive */
321 	if (ni->nid != ni->ino)
322 		e = __lookup_nat_cache(nm_i, ni->ino);
323 	if (e) {
324 		if (fsync_done && ni->nid == ni->ino)
325 			set_nat_flag(e, HAS_FSYNCED_INODE, true);
326 		set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
327 	}
328 	up_write(&nm_i->nat_tree_lock);
329 }
330 
331 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
332 {
333 	struct f2fs_nm_info *nm_i = NM_I(sbi);
334 	int nr = nr_shrink;
335 
336 	if (!down_write_trylock(&nm_i->nat_tree_lock))
337 		return 0;
338 
339 	while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
340 		struct nat_entry *ne;
341 		ne = list_first_entry(&nm_i->nat_entries,
342 					struct nat_entry, list);
343 		__del_from_nat_cache(nm_i, ne);
344 		nr_shrink--;
345 	}
346 	up_write(&nm_i->nat_tree_lock);
347 	return nr - nr_shrink;
348 }
349 
350 /*
351  * This function always returns success
352  */
353 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
354 {
355 	struct f2fs_nm_info *nm_i = NM_I(sbi);
356 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
357 	struct f2fs_summary_block *sum = curseg->sum_blk;
358 	nid_t start_nid = START_NID(nid);
359 	struct f2fs_nat_block *nat_blk;
360 	struct page *page = NULL;
361 	struct f2fs_nat_entry ne;
362 	struct nat_entry *e;
363 	int i;
364 
365 	ni->nid = nid;
366 
367 	/* Check nat cache */
368 	down_read(&nm_i->nat_tree_lock);
369 	e = __lookup_nat_cache(nm_i, nid);
370 	if (e) {
371 		ni->ino = nat_get_ino(e);
372 		ni->blk_addr = nat_get_blkaddr(e);
373 		ni->version = nat_get_version(e);
374 	}
375 	up_read(&nm_i->nat_tree_lock);
376 	if (e)
377 		return;
378 
379 	memset(&ne, 0, sizeof(struct f2fs_nat_entry));
380 
381 	down_write(&nm_i->nat_tree_lock);
382 
383 	/* Check current segment summary */
384 	mutex_lock(&curseg->curseg_mutex);
385 	i = lookup_journal_in_cursum(sum, NAT_JOURNAL, nid, 0);
386 	if (i >= 0) {
387 		ne = nat_in_journal(sum, i);
388 		node_info_from_raw_nat(ni, &ne);
389 	}
390 	mutex_unlock(&curseg->curseg_mutex);
391 	if (i >= 0)
392 		goto cache;
393 
394 	/* Fill node_info from nat page */
395 	page = get_current_nat_page(sbi, start_nid);
396 	nat_blk = (struct f2fs_nat_block *)page_address(page);
397 	ne = nat_blk->entries[nid - start_nid];
398 	node_info_from_raw_nat(ni, &ne);
399 	f2fs_put_page(page, 1);
400 cache:
401 	/* cache nat entry */
402 	cache_nat_entry(NM_I(sbi), nid, &ne);
403 	up_write(&nm_i->nat_tree_lock);
404 }
405 
406 /*
407  * The maximum depth is four.
408  * Offset[0] will have raw inode offset.
409  */
410 static int get_node_path(struct f2fs_inode_info *fi, long block,
411 				int offset[4], unsigned int noffset[4])
412 {
413 	const long direct_index = ADDRS_PER_INODE(fi);
414 	const long direct_blks = ADDRS_PER_BLOCK;
415 	const long dptrs_per_blk = NIDS_PER_BLOCK;
416 	const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
417 	const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
418 	int n = 0;
419 	int level = 0;
420 
421 	noffset[0] = 0;
422 
423 	if (block < direct_index) {
424 		offset[n] = block;
425 		goto got;
426 	}
427 	block -= direct_index;
428 	if (block < direct_blks) {
429 		offset[n++] = NODE_DIR1_BLOCK;
430 		noffset[n] = 1;
431 		offset[n] = block;
432 		level = 1;
433 		goto got;
434 	}
435 	block -= direct_blks;
436 	if (block < direct_blks) {
437 		offset[n++] = NODE_DIR2_BLOCK;
438 		noffset[n] = 2;
439 		offset[n] = block;
440 		level = 1;
441 		goto got;
442 	}
443 	block -= direct_blks;
444 	if (block < indirect_blks) {
445 		offset[n++] = NODE_IND1_BLOCK;
446 		noffset[n] = 3;
447 		offset[n++] = block / direct_blks;
448 		noffset[n] = 4 + offset[n - 1];
449 		offset[n] = block % direct_blks;
450 		level = 2;
451 		goto got;
452 	}
453 	block -= indirect_blks;
454 	if (block < indirect_blks) {
455 		offset[n++] = NODE_IND2_BLOCK;
456 		noffset[n] = 4 + dptrs_per_blk;
457 		offset[n++] = block / direct_blks;
458 		noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
459 		offset[n] = block % direct_blks;
460 		level = 2;
461 		goto got;
462 	}
463 	block -= indirect_blks;
464 	if (block < dindirect_blks) {
465 		offset[n++] = NODE_DIND_BLOCK;
466 		noffset[n] = 5 + (dptrs_per_blk * 2);
467 		offset[n++] = block / indirect_blks;
468 		noffset[n] = 6 + (dptrs_per_blk * 2) +
469 			      offset[n - 1] * (dptrs_per_blk + 1);
470 		offset[n++] = (block / direct_blks) % dptrs_per_blk;
471 		noffset[n] = 7 + (dptrs_per_blk * 2) +
472 			      offset[n - 2] * (dptrs_per_blk + 1) +
473 			      offset[n - 1];
474 		offset[n] = block % direct_blks;
475 		level = 3;
476 		goto got;
477 	} else {
478 		BUG();
479 	}
480 got:
481 	return level;
482 }
483 
484 /*
485  * Caller should call f2fs_put_dnode(dn).
486  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
487  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
488  * In the case of RDONLY_NODE, we don't need to care about mutex.
489  */
490 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
491 {
492 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
493 	struct page *npage[4];
494 	struct page *parent = NULL;
495 	int offset[4];
496 	unsigned int noffset[4];
497 	nid_t nids[4];
498 	int level, i;
499 	int err = 0;
500 
501 	level = get_node_path(F2FS_I(dn->inode), index, offset, noffset);
502 
503 	nids[0] = dn->inode->i_ino;
504 	npage[0] = dn->inode_page;
505 
506 	if (!npage[0]) {
507 		npage[0] = get_node_page(sbi, nids[0]);
508 		if (IS_ERR(npage[0]))
509 			return PTR_ERR(npage[0]);
510 	}
511 
512 	/* if inline_data is set, should not report any block indices */
513 	if (f2fs_has_inline_data(dn->inode) && index) {
514 		err = -ENOENT;
515 		f2fs_put_page(npage[0], 1);
516 		goto release_out;
517 	}
518 
519 	parent = npage[0];
520 	if (level != 0)
521 		nids[1] = get_nid(parent, offset[0], true);
522 	dn->inode_page = npage[0];
523 	dn->inode_page_locked = true;
524 
525 	/* get indirect or direct nodes */
526 	for (i = 1; i <= level; i++) {
527 		bool done = false;
528 
529 		if (!nids[i] && mode == ALLOC_NODE) {
530 			/* alloc new node */
531 			if (!alloc_nid(sbi, &(nids[i]))) {
532 				err = -ENOSPC;
533 				goto release_pages;
534 			}
535 
536 			dn->nid = nids[i];
537 			npage[i] = new_node_page(dn, noffset[i], NULL);
538 			if (IS_ERR(npage[i])) {
539 				alloc_nid_failed(sbi, nids[i]);
540 				err = PTR_ERR(npage[i]);
541 				goto release_pages;
542 			}
543 
544 			set_nid(parent, offset[i - 1], nids[i], i == 1);
545 			alloc_nid_done(sbi, nids[i]);
546 			done = true;
547 		} else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
548 			npage[i] = get_node_page_ra(parent, offset[i - 1]);
549 			if (IS_ERR(npage[i])) {
550 				err = PTR_ERR(npage[i]);
551 				goto release_pages;
552 			}
553 			done = true;
554 		}
555 		if (i == 1) {
556 			dn->inode_page_locked = false;
557 			unlock_page(parent);
558 		} else {
559 			f2fs_put_page(parent, 1);
560 		}
561 
562 		if (!done) {
563 			npage[i] = get_node_page(sbi, nids[i]);
564 			if (IS_ERR(npage[i])) {
565 				err = PTR_ERR(npage[i]);
566 				f2fs_put_page(npage[0], 0);
567 				goto release_out;
568 			}
569 		}
570 		if (i < level) {
571 			parent = npage[i];
572 			nids[i + 1] = get_nid(parent, offset[i], false);
573 		}
574 	}
575 	dn->nid = nids[level];
576 	dn->ofs_in_node = offset[level];
577 	dn->node_page = npage[level];
578 	dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
579 	return 0;
580 
581 release_pages:
582 	f2fs_put_page(parent, 1);
583 	if (i > 1)
584 		f2fs_put_page(npage[0], 0);
585 release_out:
586 	dn->inode_page = NULL;
587 	dn->node_page = NULL;
588 	return err;
589 }
590 
591 static void truncate_node(struct dnode_of_data *dn)
592 {
593 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
594 	struct node_info ni;
595 
596 	get_node_info(sbi, dn->nid, &ni);
597 	if (dn->inode->i_blocks == 0) {
598 		f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
599 		goto invalidate;
600 	}
601 	f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
602 
603 	/* Deallocate node address */
604 	invalidate_blocks(sbi, ni.blk_addr);
605 	dec_valid_node_count(sbi, dn->inode);
606 	set_node_addr(sbi, &ni, NULL_ADDR, false);
607 
608 	if (dn->nid == dn->inode->i_ino) {
609 		remove_orphan_inode(sbi, dn->nid);
610 		dec_valid_inode_count(sbi);
611 	} else {
612 		sync_inode_page(dn);
613 	}
614 invalidate:
615 	clear_node_page_dirty(dn->node_page);
616 	set_sbi_flag(sbi, SBI_IS_DIRTY);
617 
618 	f2fs_put_page(dn->node_page, 1);
619 
620 	invalidate_mapping_pages(NODE_MAPPING(sbi),
621 			dn->node_page->index, dn->node_page->index);
622 
623 	dn->node_page = NULL;
624 	trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
625 }
626 
627 static int truncate_dnode(struct dnode_of_data *dn)
628 {
629 	struct page *page;
630 
631 	if (dn->nid == 0)
632 		return 1;
633 
634 	/* get direct node */
635 	page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
636 	if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
637 		return 1;
638 	else if (IS_ERR(page))
639 		return PTR_ERR(page);
640 
641 	/* Make dnode_of_data for parameter */
642 	dn->node_page = page;
643 	dn->ofs_in_node = 0;
644 	truncate_data_blocks(dn);
645 	truncate_node(dn);
646 	return 1;
647 }
648 
649 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
650 						int ofs, int depth)
651 {
652 	struct dnode_of_data rdn = *dn;
653 	struct page *page;
654 	struct f2fs_node *rn;
655 	nid_t child_nid;
656 	unsigned int child_nofs;
657 	int freed = 0;
658 	int i, ret;
659 
660 	if (dn->nid == 0)
661 		return NIDS_PER_BLOCK + 1;
662 
663 	trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
664 
665 	page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
666 	if (IS_ERR(page)) {
667 		trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
668 		return PTR_ERR(page);
669 	}
670 
671 	rn = F2FS_NODE(page);
672 	if (depth < 3) {
673 		for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
674 			child_nid = le32_to_cpu(rn->in.nid[i]);
675 			if (child_nid == 0)
676 				continue;
677 			rdn.nid = child_nid;
678 			ret = truncate_dnode(&rdn);
679 			if (ret < 0)
680 				goto out_err;
681 			if (set_nid(page, i, 0, false))
682 				dn->node_changed = true;
683 		}
684 	} else {
685 		child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
686 		for (i = ofs; i < NIDS_PER_BLOCK; i++) {
687 			child_nid = le32_to_cpu(rn->in.nid[i]);
688 			if (child_nid == 0) {
689 				child_nofs += NIDS_PER_BLOCK + 1;
690 				continue;
691 			}
692 			rdn.nid = child_nid;
693 			ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
694 			if (ret == (NIDS_PER_BLOCK + 1)) {
695 				if (set_nid(page, i, 0, false))
696 					dn->node_changed = true;
697 				child_nofs += ret;
698 			} else if (ret < 0 && ret != -ENOENT) {
699 				goto out_err;
700 			}
701 		}
702 		freed = child_nofs;
703 	}
704 
705 	if (!ofs) {
706 		/* remove current indirect node */
707 		dn->node_page = page;
708 		truncate_node(dn);
709 		freed++;
710 	} else {
711 		f2fs_put_page(page, 1);
712 	}
713 	trace_f2fs_truncate_nodes_exit(dn->inode, freed);
714 	return freed;
715 
716 out_err:
717 	f2fs_put_page(page, 1);
718 	trace_f2fs_truncate_nodes_exit(dn->inode, ret);
719 	return ret;
720 }
721 
722 static int truncate_partial_nodes(struct dnode_of_data *dn,
723 			struct f2fs_inode *ri, int *offset, int depth)
724 {
725 	struct page *pages[2];
726 	nid_t nid[3];
727 	nid_t child_nid;
728 	int err = 0;
729 	int i;
730 	int idx = depth - 2;
731 
732 	nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
733 	if (!nid[0])
734 		return 0;
735 
736 	/* get indirect nodes in the path */
737 	for (i = 0; i < idx + 1; i++) {
738 		/* reference count'll be increased */
739 		pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
740 		if (IS_ERR(pages[i])) {
741 			err = PTR_ERR(pages[i]);
742 			idx = i - 1;
743 			goto fail;
744 		}
745 		nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
746 	}
747 
748 	/* free direct nodes linked to a partial indirect node */
749 	for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
750 		child_nid = get_nid(pages[idx], i, false);
751 		if (!child_nid)
752 			continue;
753 		dn->nid = child_nid;
754 		err = truncate_dnode(dn);
755 		if (err < 0)
756 			goto fail;
757 		if (set_nid(pages[idx], i, 0, false))
758 			dn->node_changed = true;
759 	}
760 
761 	if (offset[idx + 1] == 0) {
762 		dn->node_page = pages[idx];
763 		dn->nid = nid[idx];
764 		truncate_node(dn);
765 	} else {
766 		f2fs_put_page(pages[idx], 1);
767 	}
768 	offset[idx]++;
769 	offset[idx + 1] = 0;
770 	idx--;
771 fail:
772 	for (i = idx; i >= 0; i--)
773 		f2fs_put_page(pages[i], 1);
774 
775 	trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
776 
777 	return err;
778 }
779 
780 /*
781  * All the block addresses of data and nodes should be nullified.
782  */
783 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
784 {
785 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
786 	int err = 0, cont = 1;
787 	int level, offset[4], noffset[4];
788 	unsigned int nofs = 0;
789 	struct f2fs_inode *ri;
790 	struct dnode_of_data dn;
791 	struct page *page;
792 
793 	trace_f2fs_truncate_inode_blocks_enter(inode, from);
794 
795 	level = get_node_path(F2FS_I(inode), from, offset, noffset);
796 restart:
797 	page = get_node_page(sbi, inode->i_ino);
798 	if (IS_ERR(page)) {
799 		trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
800 		return PTR_ERR(page);
801 	}
802 
803 	set_new_dnode(&dn, inode, page, NULL, 0);
804 	unlock_page(page);
805 
806 	ri = F2FS_INODE(page);
807 	switch (level) {
808 	case 0:
809 	case 1:
810 		nofs = noffset[1];
811 		break;
812 	case 2:
813 		nofs = noffset[1];
814 		if (!offset[level - 1])
815 			goto skip_partial;
816 		err = truncate_partial_nodes(&dn, ri, offset, level);
817 		if (err < 0 && err != -ENOENT)
818 			goto fail;
819 		nofs += 1 + NIDS_PER_BLOCK;
820 		break;
821 	case 3:
822 		nofs = 5 + 2 * NIDS_PER_BLOCK;
823 		if (!offset[level - 1])
824 			goto skip_partial;
825 		err = truncate_partial_nodes(&dn, ri, offset, level);
826 		if (err < 0 && err != -ENOENT)
827 			goto fail;
828 		break;
829 	default:
830 		BUG();
831 	}
832 
833 skip_partial:
834 	while (cont) {
835 		dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
836 		switch (offset[0]) {
837 		case NODE_DIR1_BLOCK:
838 		case NODE_DIR2_BLOCK:
839 			err = truncate_dnode(&dn);
840 			break;
841 
842 		case NODE_IND1_BLOCK:
843 		case NODE_IND2_BLOCK:
844 			err = truncate_nodes(&dn, nofs, offset[1], 2);
845 			break;
846 
847 		case NODE_DIND_BLOCK:
848 			err = truncate_nodes(&dn, nofs, offset[1], 3);
849 			cont = 0;
850 			break;
851 
852 		default:
853 			BUG();
854 		}
855 		if (err < 0 && err != -ENOENT)
856 			goto fail;
857 		if (offset[1] == 0 &&
858 				ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
859 			lock_page(page);
860 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
861 				f2fs_put_page(page, 1);
862 				goto restart;
863 			}
864 			f2fs_wait_on_page_writeback(page, NODE);
865 			ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
866 			set_page_dirty(page);
867 			unlock_page(page);
868 		}
869 		offset[1] = 0;
870 		offset[0]++;
871 		nofs += err;
872 	}
873 fail:
874 	f2fs_put_page(page, 0);
875 	trace_f2fs_truncate_inode_blocks_exit(inode, err);
876 	return err > 0 ? 0 : err;
877 }
878 
879 int truncate_xattr_node(struct inode *inode, struct page *page)
880 {
881 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
882 	nid_t nid = F2FS_I(inode)->i_xattr_nid;
883 	struct dnode_of_data dn;
884 	struct page *npage;
885 
886 	if (!nid)
887 		return 0;
888 
889 	npage = get_node_page(sbi, nid);
890 	if (IS_ERR(npage))
891 		return PTR_ERR(npage);
892 
893 	F2FS_I(inode)->i_xattr_nid = 0;
894 
895 	/* need to do checkpoint during fsync */
896 	F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
897 
898 	set_new_dnode(&dn, inode, page, npage, nid);
899 
900 	if (page)
901 		dn.inode_page_locked = true;
902 	truncate_node(&dn);
903 	return 0;
904 }
905 
906 /*
907  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
908  * f2fs_unlock_op().
909  */
910 int remove_inode_page(struct inode *inode)
911 {
912 	struct dnode_of_data dn;
913 	int err;
914 
915 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
916 	err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
917 	if (err)
918 		return err;
919 
920 	err = truncate_xattr_node(inode, dn.inode_page);
921 	if (err) {
922 		f2fs_put_dnode(&dn);
923 		return err;
924 	}
925 
926 	/* remove potential inline_data blocks */
927 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
928 				S_ISLNK(inode->i_mode))
929 		truncate_data_blocks_range(&dn, 1);
930 
931 	/* 0 is possible, after f2fs_new_inode() has failed */
932 	f2fs_bug_on(F2FS_I_SB(inode),
933 			inode->i_blocks != 0 && inode->i_blocks != 1);
934 
935 	/* will put inode & node pages */
936 	truncate_node(&dn);
937 	return 0;
938 }
939 
940 struct page *new_inode_page(struct inode *inode)
941 {
942 	struct dnode_of_data dn;
943 
944 	/* allocate inode page for new inode */
945 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
946 
947 	/* caller should f2fs_put_page(page, 1); */
948 	return new_node_page(&dn, 0, NULL);
949 }
950 
951 struct page *new_node_page(struct dnode_of_data *dn,
952 				unsigned int ofs, struct page *ipage)
953 {
954 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
955 	struct node_info old_ni, new_ni;
956 	struct page *page;
957 	int err;
958 
959 	if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
960 		return ERR_PTR(-EPERM);
961 
962 	page = grab_cache_page(NODE_MAPPING(sbi), dn->nid);
963 	if (!page)
964 		return ERR_PTR(-ENOMEM);
965 
966 	if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
967 		err = -ENOSPC;
968 		goto fail;
969 	}
970 
971 	get_node_info(sbi, dn->nid, &old_ni);
972 
973 	/* Reinitialize old_ni with new node page */
974 	f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
975 	new_ni = old_ni;
976 	new_ni.ino = dn->inode->i_ino;
977 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
978 
979 	f2fs_wait_on_page_writeback(page, NODE);
980 	fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
981 	set_cold_node(dn->inode, page);
982 	SetPageUptodate(page);
983 	if (set_page_dirty(page))
984 		dn->node_changed = true;
985 
986 	if (f2fs_has_xattr_block(ofs))
987 		F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
988 
989 	dn->node_page = page;
990 	if (ipage)
991 		update_inode(dn->inode, ipage);
992 	else
993 		sync_inode_page(dn);
994 	if (ofs == 0)
995 		inc_valid_inode_count(sbi);
996 
997 	return page;
998 
999 fail:
1000 	clear_node_page_dirty(page);
1001 	f2fs_put_page(page, 1);
1002 	return ERR_PTR(err);
1003 }
1004 
1005 /*
1006  * Caller should do after getting the following values.
1007  * 0: f2fs_put_page(page, 0)
1008  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1009  */
1010 static int read_node_page(struct page *page, int rw)
1011 {
1012 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1013 	struct node_info ni;
1014 	struct f2fs_io_info fio = {
1015 		.sbi = sbi,
1016 		.type = NODE,
1017 		.rw = rw,
1018 		.page = page,
1019 		.encrypted_page = NULL,
1020 	};
1021 
1022 	get_node_info(sbi, page->index, &ni);
1023 
1024 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1025 		ClearPageUptodate(page);
1026 		return -ENOENT;
1027 	}
1028 
1029 	if (PageUptodate(page))
1030 		return LOCKED_PAGE;
1031 
1032 	fio.blk_addr = ni.blk_addr;
1033 	return f2fs_submit_page_bio(&fio);
1034 }
1035 
1036 /*
1037  * Readahead a node page
1038  */
1039 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1040 {
1041 	struct page *apage;
1042 	int err;
1043 
1044 	if (!nid)
1045 		return;
1046 	f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1047 
1048 	apage = find_get_page(NODE_MAPPING(sbi), nid);
1049 	if (apage && PageUptodate(apage)) {
1050 		f2fs_put_page(apage, 0);
1051 		return;
1052 	}
1053 	f2fs_put_page(apage, 0);
1054 
1055 	apage = grab_cache_page(NODE_MAPPING(sbi), nid);
1056 	if (!apage)
1057 		return;
1058 
1059 	err = read_node_page(apage, READA);
1060 	f2fs_put_page(apage, err ? 1 : 0);
1061 }
1062 
1063 /*
1064  * readahead MAX_RA_NODE number of node pages.
1065  */
1066 void ra_node_pages(struct page *parent, int start)
1067 {
1068 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1069 	struct blk_plug plug;
1070 	int i, end;
1071 	nid_t nid;
1072 
1073 	blk_start_plug(&plug);
1074 
1075 	/* Then, try readahead for siblings of the desired node */
1076 	end = start + MAX_RA_NODE;
1077 	end = min(end, NIDS_PER_BLOCK);
1078 	for (i = start; i < end; i++) {
1079 		nid = get_nid(parent, i, false);
1080 		ra_node_page(sbi, nid);
1081 	}
1082 
1083 	blk_finish_plug(&plug);
1084 }
1085 
1086 struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1087 					struct page *parent, int start)
1088 {
1089 	struct page *page;
1090 	int err;
1091 
1092 	if (!nid)
1093 		return ERR_PTR(-ENOENT);
1094 	f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1095 repeat:
1096 	page = grab_cache_page(NODE_MAPPING(sbi), nid);
1097 	if (!page)
1098 		return ERR_PTR(-ENOMEM);
1099 
1100 	err = read_node_page(page, READ_SYNC);
1101 	if (err < 0) {
1102 		f2fs_put_page(page, 1);
1103 		return ERR_PTR(err);
1104 	} else if (err == LOCKED_PAGE) {
1105 		goto page_hit;
1106 	}
1107 
1108 	if (parent)
1109 		ra_node_pages(parent, start + 1);
1110 
1111 	lock_page(page);
1112 
1113 	if (unlikely(!PageUptodate(page))) {
1114 		f2fs_put_page(page, 1);
1115 		return ERR_PTR(-EIO);
1116 	}
1117 	if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1118 		f2fs_put_page(page, 1);
1119 		goto repeat;
1120 	}
1121 page_hit:
1122 	f2fs_bug_on(sbi, nid != nid_of_node(page));
1123 	return page;
1124 }
1125 
1126 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1127 {
1128 	return __get_node_page(sbi, nid, NULL, 0);
1129 }
1130 
1131 struct page *get_node_page_ra(struct page *parent, int start)
1132 {
1133 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1134 	nid_t nid = get_nid(parent, start, false);
1135 
1136 	return __get_node_page(sbi, nid, parent, start);
1137 }
1138 
1139 void sync_inode_page(struct dnode_of_data *dn)
1140 {
1141 	int ret = 0;
1142 
1143 	if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1144 		ret = update_inode(dn->inode, dn->node_page);
1145 	} else if (dn->inode_page) {
1146 		if (!dn->inode_page_locked)
1147 			lock_page(dn->inode_page);
1148 		ret = update_inode(dn->inode, dn->inode_page);
1149 		if (!dn->inode_page_locked)
1150 			unlock_page(dn->inode_page);
1151 	} else {
1152 		ret = update_inode_page(dn->inode);
1153 	}
1154 	dn->node_changed = ret ? true: false;
1155 }
1156 
1157 int sync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1158 					struct writeback_control *wbc)
1159 {
1160 	pgoff_t index, end;
1161 	struct pagevec pvec;
1162 	int step = ino ? 2 : 0;
1163 	int nwritten = 0, wrote = 0;
1164 
1165 	pagevec_init(&pvec, 0);
1166 
1167 next_step:
1168 	index = 0;
1169 	end = LONG_MAX;
1170 
1171 	while (index <= end) {
1172 		int i, nr_pages;
1173 		nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1174 				PAGECACHE_TAG_DIRTY,
1175 				min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1176 		if (nr_pages == 0)
1177 			break;
1178 
1179 		for (i = 0; i < nr_pages; i++) {
1180 			struct page *page = pvec.pages[i];
1181 
1182 			if (unlikely(f2fs_cp_error(sbi))) {
1183 				pagevec_release(&pvec);
1184 				return -EIO;
1185 			}
1186 
1187 			/*
1188 			 * flushing sequence with step:
1189 			 * 0. indirect nodes
1190 			 * 1. dentry dnodes
1191 			 * 2. file dnodes
1192 			 */
1193 			if (step == 0 && IS_DNODE(page))
1194 				continue;
1195 			if (step == 1 && (!IS_DNODE(page) ||
1196 						is_cold_node(page)))
1197 				continue;
1198 			if (step == 2 && (!IS_DNODE(page) ||
1199 						!is_cold_node(page)))
1200 				continue;
1201 
1202 			/*
1203 			 * If an fsync mode,
1204 			 * we should not skip writing node pages.
1205 			 */
1206 			if (ino && ino_of_node(page) == ino)
1207 				lock_page(page);
1208 			else if (!trylock_page(page))
1209 				continue;
1210 
1211 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1212 continue_unlock:
1213 				unlock_page(page);
1214 				continue;
1215 			}
1216 			if (ino && ino_of_node(page) != ino)
1217 				goto continue_unlock;
1218 
1219 			if (!PageDirty(page)) {
1220 				/* someone wrote it for us */
1221 				goto continue_unlock;
1222 			}
1223 
1224 			if (!clear_page_dirty_for_io(page))
1225 				goto continue_unlock;
1226 
1227 			/* called by fsync() */
1228 			if (ino && IS_DNODE(page)) {
1229 				set_fsync_mark(page, 1);
1230 				if (IS_INODE(page))
1231 					set_dentry_mark(page,
1232 						need_dentry_mark(sbi, ino));
1233 				nwritten++;
1234 			} else {
1235 				set_fsync_mark(page, 0);
1236 				set_dentry_mark(page, 0);
1237 			}
1238 
1239 			if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1240 				unlock_page(page);
1241 			else
1242 				wrote++;
1243 
1244 			if (--wbc->nr_to_write == 0)
1245 				break;
1246 		}
1247 		pagevec_release(&pvec);
1248 		cond_resched();
1249 
1250 		if (wbc->nr_to_write == 0) {
1251 			step = 2;
1252 			break;
1253 		}
1254 	}
1255 
1256 	if (step < 2) {
1257 		step++;
1258 		goto next_step;
1259 	}
1260 
1261 	if (wrote)
1262 		f2fs_submit_merged_bio(sbi, NODE, WRITE);
1263 	return nwritten;
1264 }
1265 
1266 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1267 {
1268 	pgoff_t index = 0, end = LONG_MAX;
1269 	struct pagevec pvec;
1270 	int ret2 = 0, ret = 0;
1271 
1272 	pagevec_init(&pvec, 0);
1273 
1274 	while (index <= end) {
1275 		int i, nr_pages;
1276 		nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1277 				PAGECACHE_TAG_WRITEBACK,
1278 				min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1279 		if (nr_pages == 0)
1280 			break;
1281 
1282 		for (i = 0; i < nr_pages; i++) {
1283 			struct page *page = pvec.pages[i];
1284 
1285 			/* until radix tree lookup accepts end_index */
1286 			if (unlikely(page->index > end))
1287 				continue;
1288 
1289 			if (ino && ino_of_node(page) == ino) {
1290 				f2fs_wait_on_page_writeback(page, NODE);
1291 				if (TestClearPageError(page))
1292 					ret = -EIO;
1293 			}
1294 		}
1295 		pagevec_release(&pvec);
1296 		cond_resched();
1297 	}
1298 
1299 	if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1300 		ret2 = -ENOSPC;
1301 	if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1302 		ret2 = -EIO;
1303 	if (!ret)
1304 		ret = ret2;
1305 	return ret;
1306 }
1307 
1308 static int f2fs_write_node_page(struct page *page,
1309 				struct writeback_control *wbc)
1310 {
1311 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1312 	nid_t nid;
1313 	struct node_info ni;
1314 	struct f2fs_io_info fio = {
1315 		.sbi = sbi,
1316 		.type = NODE,
1317 		.rw = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : WRITE,
1318 		.page = page,
1319 		.encrypted_page = NULL,
1320 	};
1321 
1322 	trace_f2fs_writepage(page, NODE);
1323 
1324 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1325 		goto redirty_out;
1326 	if (unlikely(f2fs_cp_error(sbi)))
1327 		goto redirty_out;
1328 
1329 	f2fs_wait_on_page_writeback(page, NODE);
1330 
1331 	/* get old block addr of this node page */
1332 	nid = nid_of_node(page);
1333 	f2fs_bug_on(sbi, page->index != nid);
1334 
1335 	if (wbc->for_reclaim) {
1336 		if (!down_read_trylock(&sbi->node_write))
1337 			goto redirty_out;
1338 	} else {
1339 		down_read(&sbi->node_write);
1340 	}
1341 
1342 	get_node_info(sbi, nid, &ni);
1343 
1344 	/* This page is already truncated */
1345 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1346 		ClearPageUptodate(page);
1347 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1348 		up_read(&sbi->node_write);
1349 		unlock_page(page);
1350 		return 0;
1351 	}
1352 
1353 	set_page_writeback(page);
1354 	fio.blk_addr = ni.blk_addr;
1355 	write_node_page(nid, &fio);
1356 	set_node_addr(sbi, &ni, fio.blk_addr, is_fsync_dnode(page));
1357 	dec_page_count(sbi, F2FS_DIRTY_NODES);
1358 	up_read(&sbi->node_write);
1359 	unlock_page(page);
1360 
1361 	if (wbc->for_reclaim || unlikely(f2fs_cp_error(sbi)))
1362 		f2fs_submit_merged_bio(sbi, NODE, WRITE);
1363 
1364 	return 0;
1365 
1366 redirty_out:
1367 	redirty_page_for_writepage(wbc, page);
1368 	return AOP_WRITEPAGE_ACTIVATE;
1369 }
1370 
1371 static int f2fs_write_node_pages(struct address_space *mapping,
1372 			    struct writeback_control *wbc)
1373 {
1374 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1375 	long diff;
1376 
1377 	trace_f2fs_writepages(mapping->host, wbc, NODE);
1378 
1379 	/* balancing f2fs's metadata in background */
1380 	f2fs_balance_fs_bg(sbi);
1381 
1382 	/* collect a number of dirty node pages and write together */
1383 	if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1384 		goto skip_write;
1385 
1386 	diff = nr_pages_to_write(sbi, NODE, wbc);
1387 	wbc->sync_mode = WB_SYNC_NONE;
1388 	sync_node_pages(sbi, 0, wbc);
1389 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1390 	return 0;
1391 
1392 skip_write:
1393 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1394 	return 0;
1395 }
1396 
1397 static int f2fs_set_node_page_dirty(struct page *page)
1398 {
1399 	trace_f2fs_set_page_dirty(page, NODE);
1400 
1401 	SetPageUptodate(page);
1402 	if (!PageDirty(page)) {
1403 		__set_page_dirty_nobuffers(page);
1404 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1405 		SetPagePrivate(page);
1406 		f2fs_trace_pid(page);
1407 		return 1;
1408 	}
1409 	return 0;
1410 }
1411 
1412 /*
1413  * Structure of the f2fs node operations
1414  */
1415 const struct address_space_operations f2fs_node_aops = {
1416 	.writepage	= f2fs_write_node_page,
1417 	.writepages	= f2fs_write_node_pages,
1418 	.set_page_dirty	= f2fs_set_node_page_dirty,
1419 	.invalidatepage	= f2fs_invalidate_page,
1420 	.releasepage	= f2fs_release_page,
1421 };
1422 
1423 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1424 						nid_t n)
1425 {
1426 	return radix_tree_lookup(&nm_i->free_nid_root, n);
1427 }
1428 
1429 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1430 						struct free_nid *i)
1431 {
1432 	list_del(&i->list);
1433 	radix_tree_delete(&nm_i->free_nid_root, i->nid);
1434 }
1435 
1436 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1437 {
1438 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1439 	struct free_nid *i;
1440 	struct nat_entry *ne;
1441 	bool allocated = false;
1442 
1443 	if (!available_free_memory(sbi, FREE_NIDS))
1444 		return -1;
1445 
1446 	/* 0 nid should not be used */
1447 	if (unlikely(nid == 0))
1448 		return 0;
1449 
1450 	if (build) {
1451 		/* do not add allocated nids */
1452 		ne = __lookup_nat_cache(nm_i, nid);
1453 		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1454 				nat_get_blkaddr(ne) != NULL_ADDR))
1455 			allocated = true;
1456 		if (allocated)
1457 			return 0;
1458 	}
1459 
1460 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1461 	i->nid = nid;
1462 	i->state = NID_NEW;
1463 
1464 	if (radix_tree_preload(GFP_NOFS)) {
1465 		kmem_cache_free(free_nid_slab, i);
1466 		return 0;
1467 	}
1468 
1469 	spin_lock(&nm_i->free_nid_list_lock);
1470 	if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1471 		spin_unlock(&nm_i->free_nid_list_lock);
1472 		radix_tree_preload_end();
1473 		kmem_cache_free(free_nid_slab, i);
1474 		return 0;
1475 	}
1476 	list_add_tail(&i->list, &nm_i->free_nid_list);
1477 	nm_i->fcnt++;
1478 	spin_unlock(&nm_i->free_nid_list_lock);
1479 	radix_tree_preload_end();
1480 	return 1;
1481 }
1482 
1483 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1484 {
1485 	struct free_nid *i;
1486 	bool need_free = false;
1487 
1488 	spin_lock(&nm_i->free_nid_list_lock);
1489 	i = __lookup_free_nid_list(nm_i, nid);
1490 	if (i && i->state == NID_NEW) {
1491 		__del_from_free_nid_list(nm_i, i);
1492 		nm_i->fcnt--;
1493 		need_free = true;
1494 	}
1495 	spin_unlock(&nm_i->free_nid_list_lock);
1496 
1497 	if (need_free)
1498 		kmem_cache_free(free_nid_slab, i);
1499 }
1500 
1501 static void scan_nat_page(struct f2fs_sb_info *sbi,
1502 			struct page *nat_page, nid_t start_nid)
1503 {
1504 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1505 	struct f2fs_nat_block *nat_blk = page_address(nat_page);
1506 	block_t blk_addr;
1507 	int i;
1508 
1509 	i = start_nid % NAT_ENTRY_PER_BLOCK;
1510 
1511 	for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1512 
1513 		if (unlikely(start_nid >= nm_i->max_nid))
1514 			break;
1515 
1516 		blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1517 		f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1518 		if (blk_addr == NULL_ADDR) {
1519 			if (add_free_nid(sbi, start_nid, true) < 0)
1520 				break;
1521 		}
1522 	}
1523 }
1524 
1525 static void build_free_nids(struct f2fs_sb_info *sbi)
1526 {
1527 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1528 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1529 	struct f2fs_summary_block *sum = curseg->sum_blk;
1530 	int i = 0;
1531 	nid_t nid = nm_i->next_scan_nid;
1532 
1533 	/* Enough entries */
1534 	if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1535 		return;
1536 
1537 	/* readahead nat pages to be scanned */
1538 	ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1539 							META_NAT, true);
1540 
1541 	down_read(&nm_i->nat_tree_lock);
1542 
1543 	while (1) {
1544 		struct page *page = get_current_nat_page(sbi, nid);
1545 
1546 		scan_nat_page(sbi, page, nid);
1547 		f2fs_put_page(page, 1);
1548 
1549 		nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1550 		if (unlikely(nid >= nm_i->max_nid))
1551 			nid = 0;
1552 
1553 		if (++i >= FREE_NID_PAGES)
1554 			break;
1555 	}
1556 
1557 	/* go to the next free nat pages to find free nids abundantly */
1558 	nm_i->next_scan_nid = nid;
1559 
1560 	/* find free nids from current sum_pages */
1561 	mutex_lock(&curseg->curseg_mutex);
1562 	for (i = 0; i < nats_in_cursum(sum); i++) {
1563 		block_t addr = le32_to_cpu(nat_in_journal(sum, i).block_addr);
1564 		nid = le32_to_cpu(nid_in_journal(sum, i));
1565 		if (addr == NULL_ADDR)
1566 			add_free_nid(sbi, nid, true);
1567 		else
1568 			remove_free_nid(nm_i, nid);
1569 	}
1570 	mutex_unlock(&curseg->curseg_mutex);
1571 	up_read(&nm_i->nat_tree_lock);
1572 
1573 	ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1574 					nm_i->ra_nid_pages, META_NAT, false);
1575 }
1576 
1577 /*
1578  * If this function returns success, caller can obtain a new nid
1579  * from second parameter of this function.
1580  * The returned nid could be used ino as well as nid when inode is created.
1581  */
1582 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1583 {
1584 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1585 	struct free_nid *i = NULL;
1586 retry:
1587 	if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1588 		return false;
1589 
1590 	spin_lock(&nm_i->free_nid_list_lock);
1591 
1592 	/* We should not use stale free nids created by build_free_nids */
1593 	if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1594 		f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1595 		list_for_each_entry(i, &nm_i->free_nid_list, list)
1596 			if (i->state == NID_NEW)
1597 				break;
1598 
1599 		f2fs_bug_on(sbi, i->state != NID_NEW);
1600 		*nid = i->nid;
1601 		i->state = NID_ALLOC;
1602 		nm_i->fcnt--;
1603 		spin_unlock(&nm_i->free_nid_list_lock);
1604 		return true;
1605 	}
1606 	spin_unlock(&nm_i->free_nid_list_lock);
1607 
1608 	/* Let's scan nat pages and its caches to get free nids */
1609 	mutex_lock(&nm_i->build_lock);
1610 	build_free_nids(sbi);
1611 	mutex_unlock(&nm_i->build_lock);
1612 	goto retry;
1613 }
1614 
1615 /*
1616  * alloc_nid() should be called prior to this function.
1617  */
1618 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1619 {
1620 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1621 	struct free_nid *i;
1622 
1623 	spin_lock(&nm_i->free_nid_list_lock);
1624 	i = __lookup_free_nid_list(nm_i, nid);
1625 	f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1626 	__del_from_free_nid_list(nm_i, i);
1627 	spin_unlock(&nm_i->free_nid_list_lock);
1628 
1629 	kmem_cache_free(free_nid_slab, i);
1630 }
1631 
1632 /*
1633  * alloc_nid() should be called prior to this function.
1634  */
1635 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1636 {
1637 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1638 	struct free_nid *i;
1639 	bool need_free = false;
1640 
1641 	if (!nid)
1642 		return;
1643 
1644 	spin_lock(&nm_i->free_nid_list_lock);
1645 	i = __lookup_free_nid_list(nm_i, nid);
1646 	f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1647 	if (!available_free_memory(sbi, FREE_NIDS)) {
1648 		__del_from_free_nid_list(nm_i, i);
1649 		need_free = true;
1650 	} else {
1651 		i->state = NID_NEW;
1652 		nm_i->fcnt++;
1653 	}
1654 	spin_unlock(&nm_i->free_nid_list_lock);
1655 
1656 	if (need_free)
1657 		kmem_cache_free(free_nid_slab, i);
1658 }
1659 
1660 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1661 {
1662 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1663 	struct free_nid *i, *next;
1664 	int nr = nr_shrink;
1665 
1666 	if (!mutex_trylock(&nm_i->build_lock))
1667 		return 0;
1668 
1669 	spin_lock(&nm_i->free_nid_list_lock);
1670 	list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1671 		if (nr_shrink <= 0 || nm_i->fcnt <= NAT_ENTRY_PER_BLOCK)
1672 			break;
1673 		if (i->state == NID_ALLOC)
1674 			continue;
1675 		__del_from_free_nid_list(nm_i, i);
1676 		kmem_cache_free(free_nid_slab, i);
1677 		nm_i->fcnt--;
1678 		nr_shrink--;
1679 	}
1680 	spin_unlock(&nm_i->free_nid_list_lock);
1681 	mutex_unlock(&nm_i->build_lock);
1682 
1683 	return nr - nr_shrink;
1684 }
1685 
1686 void recover_inline_xattr(struct inode *inode, struct page *page)
1687 {
1688 	void *src_addr, *dst_addr;
1689 	size_t inline_size;
1690 	struct page *ipage;
1691 	struct f2fs_inode *ri;
1692 
1693 	ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1694 	f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1695 
1696 	ri = F2FS_INODE(page);
1697 	if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1698 		clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1699 		goto update_inode;
1700 	}
1701 
1702 	dst_addr = inline_xattr_addr(ipage);
1703 	src_addr = inline_xattr_addr(page);
1704 	inline_size = inline_xattr_size(inode);
1705 
1706 	f2fs_wait_on_page_writeback(ipage, NODE);
1707 	memcpy(dst_addr, src_addr, inline_size);
1708 update_inode:
1709 	update_inode(inode, ipage);
1710 	f2fs_put_page(ipage, 1);
1711 }
1712 
1713 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1714 {
1715 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1716 	nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1717 	nid_t new_xnid = nid_of_node(page);
1718 	struct node_info ni;
1719 
1720 	/* 1: invalidate the previous xattr nid */
1721 	if (!prev_xnid)
1722 		goto recover_xnid;
1723 
1724 	/* Deallocate node address */
1725 	get_node_info(sbi, prev_xnid, &ni);
1726 	f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1727 	invalidate_blocks(sbi, ni.blk_addr);
1728 	dec_valid_node_count(sbi, inode);
1729 	set_node_addr(sbi, &ni, NULL_ADDR, false);
1730 
1731 recover_xnid:
1732 	/* 2: allocate new xattr nid */
1733 	if (unlikely(!inc_valid_node_count(sbi, inode)))
1734 		f2fs_bug_on(sbi, 1);
1735 
1736 	remove_free_nid(NM_I(sbi), new_xnid);
1737 	get_node_info(sbi, new_xnid, &ni);
1738 	ni.ino = inode->i_ino;
1739 	set_node_addr(sbi, &ni, NEW_ADDR, false);
1740 	F2FS_I(inode)->i_xattr_nid = new_xnid;
1741 
1742 	/* 3: update xattr blkaddr */
1743 	refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
1744 	set_node_addr(sbi, &ni, blkaddr, false);
1745 
1746 	update_inode_page(inode);
1747 }
1748 
1749 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
1750 {
1751 	struct f2fs_inode *src, *dst;
1752 	nid_t ino = ino_of_node(page);
1753 	struct node_info old_ni, new_ni;
1754 	struct page *ipage;
1755 
1756 	get_node_info(sbi, ino, &old_ni);
1757 
1758 	if (unlikely(old_ni.blk_addr != NULL_ADDR))
1759 		return -EINVAL;
1760 
1761 	ipage = grab_cache_page(NODE_MAPPING(sbi), ino);
1762 	if (!ipage)
1763 		return -ENOMEM;
1764 
1765 	/* Should not use this inode from free nid list */
1766 	remove_free_nid(NM_I(sbi), ino);
1767 
1768 	SetPageUptodate(ipage);
1769 	fill_node_footer(ipage, ino, ino, 0, true);
1770 
1771 	src = F2FS_INODE(page);
1772 	dst = F2FS_INODE(ipage);
1773 
1774 	memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
1775 	dst->i_size = 0;
1776 	dst->i_blocks = cpu_to_le64(1);
1777 	dst->i_links = cpu_to_le32(1);
1778 	dst->i_xattr_nid = 0;
1779 	dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
1780 
1781 	new_ni = old_ni;
1782 	new_ni.ino = ino;
1783 
1784 	if (unlikely(!inc_valid_node_count(sbi, NULL)))
1785 		WARN_ON(1);
1786 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1787 	inc_valid_inode_count(sbi);
1788 	set_page_dirty(ipage);
1789 	f2fs_put_page(ipage, 1);
1790 	return 0;
1791 }
1792 
1793 int restore_node_summary(struct f2fs_sb_info *sbi,
1794 			unsigned int segno, struct f2fs_summary_block *sum)
1795 {
1796 	struct f2fs_node *rn;
1797 	struct f2fs_summary *sum_entry;
1798 	block_t addr;
1799 	int bio_blocks = MAX_BIO_BLOCKS(sbi);
1800 	int i, idx, last_offset, nrpages;
1801 
1802 	/* scan the node segment */
1803 	last_offset = sbi->blocks_per_seg;
1804 	addr = START_BLOCK(sbi, segno);
1805 	sum_entry = &sum->entries[0];
1806 
1807 	for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
1808 		nrpages = min(last_offset - i, bio_blocks);
1809 
1810 		/* readahead node pages */
1811 		ra_meta_pages(sbi, addr, nrpages, META_POR, true);
1812 
1813 		for (idx = addr; idx < addr + nrpages; idx++) {
1814 			struct page *page = get_tmp_page(sbi, idx);
1815 
1816 			rn = F2FS_NODE(page);
1817 			sum_entry->nid = rn->footer.nid;
1818 			sum_entry->version = 0;
1819 			sum_entry->ofs_in_node = 0;
1820 			sum_entry++;
1821 			f2fs_put_page(page, 1);
1822 		}
1823 
1824 		invalidate_mapping_pages(META_MAPPING(sbi), addr,
1825 							addr + nrpages);
1826 	}
1827 	return 0;
1828 }
1829 
1830 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
1831 {
1832 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1833 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1834 	struct f2fs_summary_block *sum = curseg->sum_blk;
1835 	int i;
1836 
1837 	mutex_lock(&curseg->curseg_mutex);
1838 	for (i = 0; i < nats_in_cursum(sum); i++) {
1839 		struct nat_entry *ne;
1840 		struct f2fs_nat_entry raw_ne;
1841 		nid_t nid = le32_to_cpu(nid_in_journal(sum, i));
1842 
1843 		raw_ne = nat_in_journal(sum, i);
1844 
1845 		ne = __lookup_nat_cache(nm_i, nid);
1846 		if (!ne) {
1847 			ne = grab_nat_entry(nm_i, nid);
1848 			node_info_from_raw_nat(&ne->ni, &raw_ne);
1849 		}
1850 		__set_nat_cache_dirty(nm_i, ne);
1851 	}
1852 	update_nats_in_cursum(sum, -i);
1853 	mutex_unlock(&curseg->curseg_mutex);
1854 }
1855 
1856 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
1857 						struct list_head *head, int max)
1858 {
1859 	struct nat_entry_set *cur;
1860 
1861 	if (nes->entry_cnt >= max)
1862 		goto add_out;
1863 
1864 	list_for_each_entry(cur, head, set_list) {
1865 		if (cur->entry_cnt >= nes->entry_cnt) {
1866 			list_add(&nes->set_list, cur->set_list.prev);
1867 			return;
1868 		}
1869 	}
1870 add_out:
1871 	list_add_tail(&nes->set_list, head);
1872 }
1873 
1874 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
1875 					struct nat_entry_set *set)
1876 {
1877 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1878 	struct f2fs_summary_block *sum = curseg->sum_blk;
1879 	nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
1880 	bool to_journal = true;
1881 	struct f2fs_nat_block *nat_blk;
1882 	struct nat_entry *ne, *cur;
1883 	struct page *page = NULL;
1884 
1885 	/*
1886 	 * there are two steps to flush nat entries:
1887 	 * #1, flush nat entries to journal in current hot data summary block.
1888 	 * #2, flush nat entries to nat page.
1889 	 */
1890 	if (!__has_cursum_space(sum, set->entry_cnt, NAT_JOURNAL))
1891 		to_journal = false;
1892 
1893 	if (to_journal) {
1894 		mutex_lock(&curseg->curseg_mutex);
1895 	} else {
1896 		page = get_next_nat_page(sbi, start_nid);
1897 		nat_blk = page_address(page);
1898 		f2fs_bug_on(sbi, !nat_blk);
1899 	}
1900 
1901 	/* flush dirty nats in nat entry set */
1902 	list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
1903 		struct f2fs_nat_entry *raw_ne;
1904 		nid_t nid = nat_get_nid(ne);
1905 		int offset;
1906 
1907 		if (nat_get_blkaddr(ne) == NEW_ADDR)
1908 			continue;
1909 
1910 		if (to_journal) {
1911 			offset = lookup_journal_in_cursum(sum,
1912 							NAT_JOURNAL, nid, 1);
1913 			f2fs_bug_on(sbi, offset < 0);
1914 			raw_ne = &nat_in_journal(sum, offset);
1915 			nid_in_journal(sum, offset) = cpu_to_le32(nid);
1916 		} else {
1917 			raw_ne = &nat_blk->entries[nid - start_nid];
1918 		}
1919 		raw_nat_from_node_info(raw_ne, &ne->ni);
1920 		nat_reset_flag(ne);
1921 		__clear_nat_cache_dirty(NM_I(sbi), ne);
1922 		if (nat_get_blkaddr(ne) == NULL_ADDR)
1923 			add_free_nid(sbi, nid, false);
1924 	}
1925 
1926 	if (to_journal)
1927 		mutex_unlock(&curseg->curseg_mutex);
1928 	else
1929 		f2fs_put_page(page, 1);
1930 
1931 	f2fs_bug_on(sbi, set->entry_cnt);
1932 
1933 	radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
1934 	kmem_cache_free(nat_entry_set_slab, set);
1935 }
1936 
1937 /*
1938  * This function is called during the checkpointing process.
1939  */
1940 void flush_nat_entries(struct f2fs_sb_info *sbi)
1941 {
1942 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1943 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1944 	struct f2fs_summary_block *sum = curseg->sum_blk;
1945 	struct nat_entry_set *setvec[SETVEC_SIZE];
1946 	struct nat_entry_set *set, *tmp;
1947 	unsigned int found;
1948 	nid_t set_idx = 0;
1949 	LIST_HEAD(sets);
1950 
1951 	if (!nm_i->dirty_nat_cnt)
1952 		return;
1953 
1954 	down_write(&nm_i->nat_tree_lock);
1955 
1956 	/*
1957 	 * if there are no enough space in journal to store dirty nat
1958 	 * entries, remove all entries from journal and merge them
1959 	 * into nat entry set.
1960 	 */
1961 	if (!__has_cursum_space(sum, nm_i->dirty_nat_cnt, NAT_JOURNAL))
1962 		remove_nats_in_journal(sbi);
1963 
1964 	while ((found = __gang_lookup_nat_set(nm_i,
1965 					set_idx, SETVEC_SIZE, setvec))) {
1966 		unsigned idx;
1967 		set_idx = setvec[found - 1]->set + 1;
1968 		for (idx = 0; idx < found; idx++)
1969 			__adjust_nat_entry_set(setvec[idx], &sets,
1970 							MAX_NAT_JENTRIES(sum));
1971 	}
1972 
1973 	/* flush dirty nats in nat entry set */
1974 	list_for_each_entry_safe(set, tmp, &sets, set_list)
1975 		__flush_nat_entry_set(sbi, set);
1976 
1977 	up_write(&nm_i->nat_tree_lock);
1978 
1979 	f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
1980 }
1981 
1982 static int init_node_manager(struct f2fs_sb_info *sbi)
1983 {
1984 	struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
1985 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1986 	unsigned char *version_bitmap;
1987 	unsigned int nat_segs, nat_blocks;
1988 
1989 	nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
1990 
1991 	/* segment_count_nat includes pair segment so divide to 2. */
1992 	nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
1993 	nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
1994 
1995 	nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
1996 
1997 	/* not used nids: 0, node, meta, (and root counted as valid node) */
1998 	nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
1999 	nm_i->fcnt = 0;
2000 	nm_i->nat_cnt = 0;
2001 	nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2002 	nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2003 
2004 	INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2005 	INIT_LIST_HEAD(&nm_i->free_nid_list);
2006 	INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2007 	INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2008 	INIT_LIST_HEAD(&nm_i->nat_entries);
2009 
2010 	mutex_init(&nm_i->build_lock);
2011 	spin_lock_init(&nm_i->free_nid_list_lock);
2012 	init_rwsem(&nm_i->nat_tree_lock);
2013 
2014 	nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2015 	nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2016 	version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2017 	if (!version_bitmap)
2018 		return -EFAULT;
2019 
2020 	nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2021 					GFP_KERNEL);
2022 	if (!nm_i->nat_bitmap)
2023 		return -ENOMEM;
2024 	return 0;
2025 }
2026 
2027 int build_node_manager(struct f2fs_sb_info *sbi)
2028 {
2029 	int err;
2030 
2031 	sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2032 	if (!sbi->nm_info)
2033 		return -ENOMEM;
2034 
2035 	err = init_node_manager(sbi);
2036 	if (err)
2037 		return err;
2038 
2039 	build_free_nids(sbi);
2040 	return 0;
2041 }
2042 
2043 void destroy_node_manager(struct f2fs_sb_info *sbi)
2044 {
2045 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2046 	struct free_nid *i, *next_i;
2047 	struct nat_entry *natvec[NATVEC_SIZE];
2048 	struct nat_entry_set *setvec[SETVEC_SIZE];
2049 	nid_t nid = 0;
2050 	unsigned int found;
2051 
2052 	if (!nm_i)
2053 		return;
2054 
2055 	/* destroy free nid list */
2056 	spin_lock(&nm_i->free_nid_list_lock);
2057 	list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2058 		f2fs_bug_on(sbi, i->state == NID_ALLOC);
2059 		__del_from_free_nid_list(nm_i, i);
2060 		nm_i->fcnt--;
2061 		spin_unlock(&nm_i->free_nid_list_lock);
2062 		kmem_cache_free(free_nid_slab, i);
2063 		spin_lock(&nm_i->free_nid_list_lock);
2064 	}
2065 	f2fs_bug_on(sbi, nm_i->fcnt);
2066 	spin_unlock(&nm_i->free_nid_list_lock);
2067 
2068 	/* destroy nat cache */
2069 	down_write(&nm_i->nat_tree_lock);
2070 	while ((found = __gang_lookup_nat_cache(nm_i,
2071 					nid, NATVEC_SIZE, natvec))) {
2072 		unsigned idx;
2073 
2074 		nid = nat_get_nid(natvec[found - 1]) + 1;
2075 		for (idx = 0; idx < found; idx++)
2076 			__del_from_nat_cache(nm_i, natvec[idx]);
2077 	}
2078 	f2fs_bug_on(sbi, nm_i->nat_cnt);
2079 
2080 	/* destroy nat set cache */
2081 	nid = 0;
2082 	while ((found = __gang_lookup_nat_set(nm_i,
2083 					nid, SETVEC_SIZE, setvec))) {
2084 		unsigned idx;
2085 
2086 		nid = setvec[found - 1]->set + 1;
2087 		for (idx = 0; idx < found; idx++) {
2088 			/* entry_cnt is not zero, when cp_error was occurred */
2089 			f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2090 			radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2091 			kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2092 		}
2093 	}
2094 	up_write(&nm_i->nat_tree_lock);
2095 
2096 	kfree(nm_i->nat_bitmap);
2097 	sbi->nm_info = NULL;
2098 	kfree(nm_i);
2099 }
2100 
2101 int __init create_node_manager_caches(void)
2102 {
2103 	nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2104 			sizeof(struct nat_entry));
2105 	if (!nat_entry_slab)
2106 		goto fail;
2107 
2108 	free_nid_slab = f2fs_kmem_cache_create("free_nid",
2109 			sizeof(struct free_nid));
2110 	if (!free_nid_slab)
2111 		goto destroy_nat_entry;
2112 
2113 	nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2114 			sizeof(struct nat_entry_set));
2115 	if (!nat_entry_set_slab)
2116 		goto destroy_free_nid;
2117 	return 0;
2118 
2119 destroy_free_nid:
2120 	kmem_cache_destroy(free_nid_slab);
2121 destroy_nat_entry:
2122 	kmem_cache_destroy(nat_entry_slab);
2123 fail:
2124 	return -ENOMEM;
2125 }
2126 
2127 void destroy_node_manager_caches(void)
2128 {
2129 	kmem_cache_destroy(nat_entry_set_slab);
2130 	kmem_cache_destroy(free_nid_slab);
2131 	kmem_cache_destroy(nat_entry_slab);
2132 }
2133