xref: /openbmc/linux/fs/f2fs/node.c (revision a72594ca)
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 "xattr.h"
23 #include "trace.h"
24 #include <trace/events/f2fs.h>
25 
26 #define on_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
27 
28 static struct kmem_cache *nat_entry_slab;
29 static struct kmem_cache *free_nid_slab;
30 static struct kmem_cache *nat_entry_set_slab;
31 
32 bool available_free_memory(struct f2fs_sb_info *sbi, int type)
33 {
34 	struct f2fs_nm_info *nm_i = NM_I(sbi);
35 	struct sysinfo val;
36 	unsigned long avail_ram;
37 	unsigned long mem_size = 0;
38 	bool res = false;
39 
40 	si_meminfo(&val);
41 
42 	/* only uses low memory */
43 	avail_ram = val.totalram - val.totalhigh;
44 
45 	/*
46 	 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
47 	 */
48 	if (type == FREE_NIDS) {
49 		mem_size = (nm_i->nid_cnt[FREE_NID] *
50 				sizeof(struct free_nid)) >> PAGE_SHIFT;
51 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
52 	} else if (type == NAT_ENTRIES) {
53 		mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
54 							PAGE_SHIFT;
55 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
56 		if (excess_cached_nats(sbi))
57 			res = false;
58 	} else if (type == DIRTY_DENTS) {
59 		if (sbi->sb->s_bdi->wb.dirty_exceeded)
60 			return false;
61 		mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
62 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
63 	} else if (type == INO_ENTRIES) {
64 		int i;
65 
66 		for (i = 0; i < MAX_INO_ENTRY; i++)
67 			mem_size += sbi->im[i].ino_num *
68 						sizeof(struct ino_entry);
69 		mem_size >>= PAGE_SHIFT;
70 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
71 	} else if (type == EXTENT_CACHE) {
72 		mem_size = (atomic_read(&sbi->total_ext_tree) *
73 				sizeof(struct extent_tree) +
74 				atomic_read(&sbi->total_ext_node) *
75 				sizeof(struct extent_node)) >> PAGE_SHIFT;
76 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
77 	} else if (type == INMEM_PAGES) {
78 		/* it allows 20% / total_ram for inmemory pages */
79 		mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
80 		res = mem_size < (val.totalram / 5);
81 	} else {
82 		if (!sbi->sb->s_bdi->wb.dirty_exceeded)
83 			return true;
84 	}
85 	return res;
86 }
87 
88 static void clear_node_page_dirty(struct page *page)
89 {
90 	struct address_space *mapping = page->mapping;
91 	unsigned int long flags;
92 
93 	if (PageDirty(page)) {
94 		spin_lock_irqsave(&mapping->tree_lock, flags);
95 		radix_tree_tag_clear(&mapping->page_tree,
96 				page_index(page),
97 				PAGECACHE_TAG_DIRTY);
98 		spin_unlock_irqrestore(&mapping->tree_lock, flags);
99 
100 		clear_page_dirty_for_io(page);
101 		dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
102 	}
103 	ClearPageUptodate(page);
104 }
105 
106 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
107 {
108 	pgoff_t index = current_nat_addr(sbi, nid);
109 	return get_meta_page(sbi, index);
110 }
111 
112 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
113 {
114 	struct page *src_page;
115 	struct page *dst_page;
116 	pgoff_t src_off;
117 	pgoff_t dst_off;
118 	void *src_addr;
119 	void *dst_addr;
120 	struct f2fs_nm_info *nm_i = NM_I(sbi);
121 
122 	src_off = current_nat_addr(sbi, nid);
123 	dst_off = next_nat_addr(sbi, src_off);
124 
125 	/* get current nat block page with lock */
126 	src_page = get_meta_page(sbi, src_off);
127 	dst_page = grab_meta_page(sbi, dst_off);
128 	f2fs_bug_on(sbi, PageDirty(src_page));
129 
130 	src_addr = page_address(src_page);
131 	dst_addr = page_address(dst_page);
132 	memcpy(dst_addr, src_addr, PAGE_SIZE);
133 	set_page_dirty(dst_page);
134 	f2fs_put_page(src_page, 1);
135 
136 	set_to_next_nat(nm_i, nid);
137 
138 	return dst_page;
139 }
140 
141 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
142 {
143 	struct nat_entry *new;
144 
145 	if (no_fail)
146 		new = f2fs_kmem_cache_alloc(nat_entry_slab,
147 						GFP_NOFS | __GFP_ZERO);
148 	else
149 		new = kmem_cache_alloc(nat_entry_slab,
150 						GFP_NOFS | __GFP_ZERO);
151 	if (new) {
152 		nat_set_nid(new, nid);
153 		nat_reset_flag(new);
154 	}
155 	return new;
156 }
157 
158 static void __free_nat_entry(struct nat_entry *e)
159 {
160 	kmem_cache_free(nat_entry_slab, e);
161 }
162 
163 /* must be locked by nat_tree_lock */
164 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
165 	struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
166 {
167 	if (no_fail)
168 		f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
169 	else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
170 		return NULL;
171 
172 	if (raw_ne)
173 		node_info_from_raw_nat(&ne->ni, raw_ne);
174 	list_add_tail(&ne->list, &nm_i->nat_entries);
175 	nm_i->nat_cnt++;
176 	return ne;
177 }
178 
179 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
180 {
181 	return radix_tree_lookup(&nm_i->nat_root, n);
182 }
183 
184 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
185 		nid_t start, unsigned int nr, struct nat_entry **ep)
186 {
187 	return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
188 }
189 
190 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
191 {
192 	list_del(&e->list);
193 	radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
194 	nm_i->nat_cnt--;
195 	__free_nat_entry(e);
196 }
197 
198 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
199 						struct nat_entry *ne)
200 {
201 	nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
202 	struct nat_entry_set *head;
203 
204 	head = radix_tree_lookup(&nm_i->nat_set_root, set);
205 	if (!head) {
206 		head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
207 
208 		INIT_LIST_HEAD(&head->entry_list);
209 		INIT_LIST_HEAD(&head->set_list);
210 		head->set = set;
211 		head->entry_cnt = 0;
212 		f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
213 	}
214 
215 	if (get_nat_flag(ne, IS_DIRTY))
216 		goto refresh_list;
217 
218 	nm_i->dirty_nat_cnt++;
219 	head->entry_cnt++;
220 	set_nat_flag(ne, IS_DIRTY, true);
221 refresh_list:
222 	if (nat_get_blkaddr(ne) == NEW_ADDR)
223 		list_del_init(&ne->list);
224 	else
225 		list_move_tail(&ne->list, &head->entry_list);
226 }
227 
228 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
229 		struct nat_entry_set *set, struct nat_entry *ne)
230 {
231 	list_move_tail(&ne->list, &nm_i->nat_entries);
232 	set_nat_flag(ne, IS_DIRTY, false);
233 	set->entry_cnt--;
234 	nm_i->dirty_nat_cnt--;
235 }
236 
237 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
238 		nid_t start, unsigned int nr, struct nat_entry_set **ep)
239 {
240 	return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
241 							start, nr);
242 }
243 
244 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
245 {
246 	struct f2fs_nm_info *nm_i = NM_I(sbi);
247 	struct nat_entry *e;
248 	bool need = false;
249 
250 	down_read(&nm_i->nat_tree_lock);
251 	e = __lookup_nat_cache(nm_i, nid);
252 	if (e) {
253 		if (!get_nat_flag(e, IS_CHECKPOINTED) &&
254 				!get_nat_flag(e, HAS_FSYNCED_INODE))
255 			need = true;
256 	}
257 	up_read(&nm_i->nat_tree_lock);
258 	return need;
259 }
260 
261 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
262 {
263 	struct f2fs_nm_info *nm_i = NM_I(sbi);
264 	struct nat_entry *e;
265 	bool is_cp = true;
266 
267 	down_read(&nm_i->nat_tree_lock);
268 	e = __lookup_nat_cache(nm_i, nid);
269 	if (e && !get_nat_flag(e, IS_CHECKPOINTED))
270 		is_cp = false;
271 	up_read(&nm_i->nat_tree_lock);
272 	return is_cp;
273 }
274 
275 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
276 {
277 	struct f2fs_nm_info *nm_i = NM_I(sbi);
278 	struct nat_entry *e;
279 	bool need_update = true;
280 
281 	down_read(&nm_i->nat_tree_lock);
282 	e = __lookup_nat_cache(nm_i, ino);
283 	if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
284 			(get_nat_flag(e, IS_CHECKPOINTED) ||
285 			 get_nat_flag(e, HAS_FSYNCED_INODE)))
286 		need_update = false;
287 	up_read(&nm_i->nat_tree_lock);
288 	return need_update;
289 }
290 
291 /* must be locked by nat_tree_lock */
292 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
293 						struct f2fs_nat_entry *ne)
294 {
295 	struct f2fs_nm_info *nm_i = NM_I(sbi);
296 	struct nat_entry *new, *e;
297 
298 	new = __alloc_nat_entry(nid, false);
299 	if (!new)
300 		return;
301 
302 	down_write(&nm_i->nat_tree_lock);
303 	e = __lookup_nat_cache(nm_i, nid);
304 	if (!e)
305 		e = __init_nat_entry(nm_i, new, ne, false);
306 	else
307 		f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
308 				nat_get_blkaddr(e) !=
309 					le32_to_cpu(ne->block_addr) ||
310 				nat_get_version(e) != ne->version);
311 	up_write(&nm_i->nat_tree_lock);
312 	if (e != new)
313 		__free_nat_entry(new);
314 }
315 
316 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
317 			block_t new_blkaddr, bool fsync_done)
318 {
319 	struct f2fs_nm_info *nm_i = NM_I(sbi);
320 	struct nat_entry *e;
321 	struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
322 
323 	down_write(&nm_i->nat_tree_lock);
324 	e = __lookup_nat_cache(nm_i, ni->nid);
325 	if (!e) {
326 		e = __init_nat_entry(nm_i, new, NULL, true);
327 		copy_node_info(&e->ni, ni);
328 		f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
329 	} else if (new_blkaddr == NEW_ADDR) {
330 		/*
331 		 * when nid is reallocated,
332 		 * previous nat entry can be remained in nat cache.
333 		 * So, reinitialize it with new information.
334 		 */
335 		copy_node_info(&e->ni, ni);
336 		f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
337 	}
338 	/* let's free early to reduce memory consumption */
339 	if (e != new)
340 		__free_nat_entry(new);
341 
342 	/* sanity check */
343 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
344 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
345 			new_blkaddr == NULL_ADDR);
346 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
347 			new_blkaddr == NEW_ADDR);
348 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
349 			nat_get_blkaddr(e) != NULL_ADDR &&
350 			new_blkaddr == NEW_ADDR);
351 
352 	/* increment version no as node is removed */
353 	if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
354 		unsigned char version = nat_get_version(e);
355 		nat_set_version(e, inc_node_version(version));
356 	}
357 
358 	/* change address */
359 	nat_set_blkaddr(e, new_blkaddr);
360 	if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
361 		set_nat_flag(e, IS_CHECKPOINTED, false);
362 	__set_nat_cache_dirty(nm_i, e);
363 
364 	/* update fsync_mark if its inode nat entry is still alive */
365 	if (ni->nid != ni->ino)
366 		e = __lookup_nat_cache(nm_i, ni->ino);
367 	if (e) {
368 		if (fsync_done && ni->nid == ni->ino)
369 			set_nat_flag(e, HAS_FSYNCED_INODE, true);
370 		set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
371 	}
372 	up_write(&nm_i->nat_tree_lock);
373 }
374 
375 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
376 {
377 	struct f2fs_nm_info *nm_i = NM_I(sbi);
378 	int nr = nr_shrink;
379 
380 	if (!down_write_trylock(&nm_i->nat_tree_lock))
381 		return 0;
382 
383 	while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
384 		struct nat_entry *ne;
385 		ne = list_first_entry(&nm_i->nat_entries,
386 					struct nat_entry, list);
387 		__del_from_nat_cache(nm_i, ne);
388 		nr_shrink--;
389 	}
390 	up_write(&nm_i->nat_tree_lock);
391 	return nr - nr_shrink;
392 }
393 
394 /*
395  * This function always returns success
396  */
397 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
398 {
399 	struct f2fs_nm_info *nm_i = NM_I(sbi);
400 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
401 	struct f2fs_journal *journal = curseg->journal;
402 	nid_t start_nid = START_NID(nid);
403 	struct f2fs_nat_block *nat_blk;
404 	struct page *page = NULL;
405 	struct f2fs_nat_entry ne;
406 	struct nat_entry *e;
407 	pgoff_t index;
408 	int i;
409 
410 	ni->nid = nid;
411 
412 	/* Check nat cache */
413 	down_read(&nm_i->nat_tree_lock);
414 	e = __lookup_nat_cache(nm_i, nid);
415 	if (e) {
416 		ni->ino = nat_get_ino(e);
417 		ni->blk_addr = nat_get_blkaddr(e);
418 		ni->version = nat_get_version(e);
419 		up_read(&nm_i->nat_tree_lock);
420 		return;
421 	}
422 
423 	memset(&ne, 0, sizeof(struct f2fs_nat_entry));
424 
425 	/* Check current segment summary */
426 	down_read(&curseg->journal_rwsem);
427 	i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
428 	if (i >= 0) {
429 		ne = nat_in_journal(journal, i);
430 		node_info_from_raw_nat(ni, &ne);
431 	}
432 	up_read(&curseg->journal_rwsem);
433 	if (i >= 0) {
434 		up_read(&nm_i->nat_tree_lock);
435 		goto cache;
436 	}
437 
438 	/* Fill node_info from nat page */
439 	index = current_nat_addr(sbi, nid);
440 	up_read(&nm_i->nat_tree_lock);
441 
442 	page = get_meta_page(sbi, index);
443 	nat_blk = (struct f2fs_nat_block *)page_address(page);
444 	ne = nat_blk->entries[nid - start_nid];
445 	node_info_from_raw_nat(ni, &ne);
446 	f2fs_put_page(page, 1);
447 cache:
448 	/* cache nat entry */
449 	cache_nat_entry(sbi, nid, &ne);
450 }
451 
452 /*
453  * readahead MAX_RA_NODE number of node pages.
454  */
455 static void ra_node_pages(struct page *parent, int start, int n)
456 {
457 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
458 	struct blk_plug plug;
459 	int i, end;
460 	nid_t nid;
461 
462 	blk_start_plug(&plug);
463 
464 	/* Then, try readahead for siblings of the desired node */
465 	end = start + n;
466 	end = min(end, NIDS_PER_BLOCK);
467 	for (i = start; i < end; i++) {
468 		nid = get_nid(parent, i, false);
469 		ra_node_page(sbi, nid);
470 	}
471 
472 	blk_finish_plug(&plug);
473 }
474 
475 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
476 {
477 	const long direct_index = ADDRS_PER_INODE(dn->inode);
478 	const long direct_blks = ADDRS_PER_BLOCK;
479 	const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
480 	unsigned int skipped_unit = ADDRS_PER_BLOCK;
481 	int cur_level = dn->cur_level;
482 	int max_level = dn->max_level;
483 	pgoff_t base = 0;
484 
485 	if (!dn->max_level)
486 		return pgofs + 1;
487 
488 	while (max_level-- > cur_level)
489 		skipped_unit *= NIDS_PER_BLOCK;
490 
491 	switch (dn->max_level) {
492 	case 3:
493 		base += 2 * indirect_blks;
494 	case 2:
495 		base += 2 * direct_blks;
496 	case 1:
497 		base += direct_index;
498 		break;
499 	default:
500 		f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
501 	}
502 
503 	return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
504 }
505 
506 /*
507  * The maximum depth is four.
508  * Offset[0] will have raw inode offset.
509  */
510 static int get_node_path(struct inode *inode, long block,
511 				int offset[4], unsigned int noffset[4])
512 {
513 	const long direct_index = ADDRS_PER_INODE(inode);
514 	const long direct_blks = ADDRS_PER_BLOCK;
515 	const long dptrs_per_blk = NIDS_PER_BLOCK;
516 	const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
517 	const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
518 	int n = 0;
519 	int level = 0;
520 
521 	noffset[0] = 0;
522 
523 	if (block < direct_index) {
524 		offset[n] = block;
525 		goto got;
526 	}
527 	block -= direct_index;
528 	if (block < direct_blks) {
529 		offset[n++] = NODE_DIR1_BLOCK;
530 		noffset[n] = 1;
531 		offset[n] = block;
532 		level = 1;
533 		goto got;
534 	}
535 	block -= direct_blks;
536 	if (block < direct_blks) {
537 		offset[n++] = NODE_DIR2_BLOCK;
538 		noffset[n] = 2;
539 		offset[n] = block;
540 		level = 1;
541 		goto got;
542 	}
543 	block -= direct_blks;
544 	if (block < indirect_blks) {
545 		offset[n++] = NODE_IND1_BLOCK;
546 		noffset[n] = 3;
547 		offset[n++] = block / direct_blks;
548 		noffset[n] = 4 + offset[n - 1];
549 		offset[n] = block % direct_blks;
550 		level = 2;
551 		goto got;
552 	}
553 	block -= indirect_blks;
554 	if (block < indirect_blks) {
555 		offset[n++] = NODE_IND2_BLOCK;
556 		noffset[n] = 4 + dptrs_per_blk;
557 		offset[n++] = block / direct_blks;
558 		noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
559 		offset[n] = block % direct_blks;
560 		level = 2;
561 		goto got;
562 	}
563 	block -= indirect_blks;
564 	if (block < dindirect_blks) {
565 		offset[n++] = NODE_DIND_BLOCK;
566 		noffset[n] = 5 + (dptrs_per_blk * 2);
567 		offset[n++] = block / indirect_blks;
568 		noffset[n] = 6 + (dptrs_per_blk * 2) +
569 			      offset[n - 1] * (dptrs_per_blk + 1);
570 		offset[n++] = (block / direct_blks) % dptrs_per_blk;
571 		noffset[n] = 7 + (dptrs_per_blk * 2) +
572 			      offset[n - 2] * (dptrs_per_blk + 1) +
573 			      offset[n - 1];
574 		offset[n] = block % direct_blks;
575 		level = 3;
576 		goto got;
577 	} else {
578 		return -E2BIG;
579 	}
580 got:
581 	return level;
582 }
583 
584 /*
585  * Caller should call f2fs_put_dnode(dn).
586  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
587  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
588  * In the case of RDONLY_NODE, we don't need to care about mutex.
589  */
590 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
591 {
592 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
593 	struct page *npage[4];
594 	struct page *parent = NULL;
595 	int offset[4];
596 	unsigned int noffset[4];
597 	nid_t nids[4];
598 	int level, i = 0;
599 	int err = 0;
600 
601 	level = get_node_path(dn->inode, index, offset, noffset);
602 	if (level < 0)
603 		return level;
604 
605 	nids[0] = dn->inode->i_ino;
606 	npage[0] = dn->inode_page;
607 
608 	if (!npage[0]) {
609 		npage[0] = get_node_page(sbi, nids[0]);
610 		if (IS_ERR(npage[0]))
611 			return PTR_ERR(npage[0]);
612 	}
613 
614 	/* if inline_data is set, should not report any block indices */
615 	if (f2fs_has_inline_data(dn->inode) && index) {
616 		err = -ENOENT;
617 		f2fs_put_page(npage[0], 1);
618 		goto release_out;
619 	}
620 
621 	parent = npage[0];
622 	if (level != 0)
623 		nids[1] = get_nid(parent, offset[0], true);
624 	dn->inode_page = npage[0];
625 	dn->inode_page_locked = true;
626 
627 	/* get indirect or direct nodes */
628 	for (i = 1; i <= level; i++) {
629 		bool done = false;
630 
631 		if (!nids[i] && mode == ALLOC_NODE) {
632 			/* alloc new node */
633 			if (!alloc_nid(sbi, &(nids[i]))) {
634 				err = -ENOSPC;
635 				goto release_pages;
636 			}
637 
638 			dn->nid = nids[i];
639 			npage[i] = new_node_page(dn, noffset[i]);
640 			if (IS_ERR(npage[i])) {
641 				alloc_nid_failed(sbi, nids[i]);
642 				err = PTR_ERR(npage[i]);
643 				goto release_pages;
644 			}
645 
646 			set_nid(parent, offset[i - 1], nids[i], i == 1);
647 			alloc_nid_done(sbi, nids[i]);
648 			done = true;
649 		} else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
650 			npage[i] = get_node_page_ra(parent, offset[i - 1]);
651 			if (IS_ERR(npage[i])) {
652 				err = PTR_ERR(npage[i]);
653 				goto release_pages;
654 			}
655 			done = true;
656 		}
657 		if (i == 1) {
658 			dn->inode_page_locked = false;
659 			unlock_page(parent);
660 		} else {
661 			f2fs_put_page(parent, 1);
662 		}
663 
664 		if (!done) {
665 			npage[i] = get_node_page(sbi, nids[i]);
666 			if (IS_ERR(npage[i])) {
667 				err = PTR_ERR(npage[i]);
668 				f2fs_put_page(npage[0], 0);
669 				goto release_out;
670 			}
671 		}
672 		if (i < level) {
673 			parent = npage[i];
674 			nids[i + 1] = get_nid(parent, offset[i], false);
675 		}
676 	}
677 	dn->nid = nids[level];
678 	dn->ofs_in_node = offset[level];
679 	dn->node_page = npage[level];
680 	dn->data_blkaddr = datablock_addr(dn->inode,
681 				dn->node_page, dn->ofs_in_node);
682 	return 0;
683 
684 release_pages:
685 	f2fs_put_page(parent, 1);
686 	if (i > 1)
687 		f2fs_put_page(npage[0], 0);
688 release_out:
689 	dn->inode_page = NULL;
690 	dn->node_page = NULL;
691 	if (err == -ENOENT) {
692 		dn->cur_level = i;
693 		dn->max_level = level;
694 		dn->ofs_in_node = offset[level];
695 	}
696 	return err;
697 }
698 
699 static void truncate_node(struct dnode_of_data *dn)
700 {
701 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
702 	struct node_info ni;
703 
704 	get_node_info(sbi, dn->nid, &ni);
705 	f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
706 
707 	/* Deallocate node address */
708 	invalidate_blocks(sbi, ni.blk_addr);
709 	dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
710 	set_node_addr(sbi, &ni, NULL_ADDR, false);
711 
712 	if (dn->nid == dn->inode->i_ino) {
713 		remove_orphan_inode(sbi, dn->nid);
714 		dec_valid_inode_count(sbi);
715 		f2fs_inode_synced(dn->inode);
716 	}
717 
718 	clear_node_page_dirty(dn->node_page);
719 	set_sbi_flag(sbi, SBI_IS_DIRTY);
720 
721 	f2fs_put_page(dn->node_page, 1);
722 
723 	invalidate_mapping_pages(NODE_MAPPING(sbi),
724 			dn->node_page->index, dn->node_page->index);
725 
726 	dn->node_page = NULL;
727 	trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
728 }
729 
730 static int truncate_dnode(struct dnode_of_data *dn)
731 {
732 	struct page *page;
733 
734 	if (dn->nid == 0)
735 		return 1;
736 
737 	/* get direct node */
738 	page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
739 	if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
740 		return 1;
741 	else if (IS_ERR(page))
742 		return PTR_ERR(page);
743 
744 	/* Make dnode_of_data for parameter */
745 	dn->node_page = page;
746 	dn->ofs_in_node = 0;
747 	truncate_data_blocks(dn);
748 	truncate_node(dn);
749 	return 1;
750 }
751 
752 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
753 						int ofs, int depth)
754 {
755 	struct dnode_of_data rdn = *dn;
756 	struct page *page;
757 	struct f2fs_node *rn;
758 	nid_t child_nid;
759 	unsigned int child_nofs;
760 	int freed = 0;
761 	int i, ret;
762 
763 	if (dn->nid == 0)
764 		return NIDS_PER_BLOCK + 1;
765 
766 	trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
767 
768 	page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
769 	if (IS_ERR(page)) {
770 		trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
771 		return PTR_ERR(page);
772 	}
773 
774 	ra_node_pages(page, ofs, NIDS_PER_BLOCK);
775 
776 	rn = F2FS_NODE(page);
777 	if (depth < 3) {
778 		for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
779 			child_nid = le32_to_cpu(rn->in.nid[i]);
780 			if (child_nid == 0)
781 				continue;
782 			rdn.nid = child_nid;
783 			ret = truncate_dnode(&rdn);
784 			if (ret < 0)
785 				goto out_err;
786 			if (set_nid(page, i, 0, false))
787 				dn->node_changed = true;
788 		}
789 	} else {
790 		child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
791 		for (i = ofs; i < NIDS_PER_BLOCK; i++) {
792 			child_nid = le32_to_cpu(rn->in.nid[i]);
793 			if (child_nid == 0) {
794 				child_nofs += NIDS_PER_BLOCK + 1;
795 				continue;
796 			}
797 			rdn.nid = child_nid;
798 			ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
799 			if (ret == (NIDS_PER_BLOCK + 1)) {
800 				if (set_nid(page, i, 0, false))
801 					dn->node_changed = true;
802 				child_nofs += ret;
803 			} else if (ret < 0 && ret != -ENOENT) {
804 				goto out_err;
805 			}
806 		}
807 		freed = child_nofs;
808 	}
809 
810 	if (!ofs) {
811 		/* remove current indirect node */
812 		dn->node_page = page;
813 		truncate_node(dn);
814 		freed++;
815 	} else {
816 		f2fs_put_page(page, 1);
817 	}
818 	trace_f2fs_truncate_nodes_exit(dn->inode, freed);
819 	return freed;
820 
821 out_err:
822 	f2fs_put_page(page, 1);
823 	trace_f2fs_truncate_nodes_exit(dn->inode, ret);
824 	return ret;
825 }
826 
827 static int truncate_partial_nodes(struct dnode_of_data *dn,
828 			struct f2fs_inode *ri, int *offset, int depth)
829 {
830 	struct page *pages[2];
831 	nid_t nid[3];
832 	nid_t child_nid;
833 	int err = 0;
834 	int i;
835 	int idx = depth - 2;
836 
837 	nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
838 	if (!nid[0])
839 		return 0;
840 
841 	/* get indirect nodes in the path */
842 	for (i = 0; i < idx + 1; i++) {
843 		/* reference count'll be increased */
844 		pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
845 		if (IS_ERR(pages[i])) {
846 			err = PTR_ERR(pages[i]);
847 			idx = i - 1;
848 			goto fail;
849 		}
850 		nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
851 	}
852 
853 	ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
854 
855 	/* free direct nodes linked to a partial indirect node */
856 	for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
857 		child_nid = get_nid(pages[idx], i, false);
858 		if (!child_nid)
859 			continue;
860 		dn->nid = child_nid;
861 		err = truncate_dnode(dn);
862 		if (err < 0)
863 			goto fail;
864 		if (set_nid(pages[idx], i, 0, false))
865 			dn->node_changed = true;
866 	}
867 
868 	if (offset[idx + 1] == 0) {
869 		dn->node_page = pages[idx];
870 		dn->nid = nid[idx];
871 		truncate_node(dn);
872 	} else {
873 		f2fs_put_page(pages[idx], 1);
874 	}
875 	offset[idx]++;
876 	offset[idx + 1] = 0;
877 	idx--;
878 fail:
879 	for (i = idx; i >= 0; i--)
880 		f2fs_put_page(pages[i], 1);
881 
882 	trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
883 
884 	return err;
885 }
886 
887 /*
888  * All the block addresses of data and nodes should be nullified.
889  */
890 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
891 {
892 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
893 	int err = 0, cont = 1;
894 	int level, offset[4], noffset[4];
895 	unsigned int nofs = 0;
896 	struct f2fs_inode *ri;
897 	struct dnode_of_data dn;
898 	struct page *page;
899 
900 	trace_f2fs_truncate_inode_blocks_enter(inode, from);
901 
902 	level = get_node_path(inode, from, offset, noffset);
903 	if (level < 0)
904 		return level;
905 
906 	page = get_node_page(sbi, inode->i_ino);
907 	if (IS_ERR(page)) {
908 		trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
909 		return PTR_ERR(page);
910 	}
911 
912 	set_new_dnode(&dn, inode, page, NULL, 0);
913 	unlock_page(page);
914 
915 	ri = F2FS_INODE(page);
916 	switch (level) {
917 	case 0:
918 	case 1:
919 		nofs = noffset[1];
920 		break;
921 	case 2:
922 		nofs = noffset[1];
923 		if (!offset[level - 1])
924 			goto skip_partial;
925 		err = truncate_partial_nodes(&dn, ri, offset, level);
926 		if (err < 0 && err != -ENOENT)
927 			goto fail;
928 		nofs += 1 + NIDS_PER_BLOCK;
929 		break;
930 	case 3:
931 		nofs = 5 + 2 * NIDS_PER_BLOCK;
932 		if (!offset[level - 1])
933 			goto skip_partial;
934 		err = truncate_partial_nodes(&dn, ri, offset, level);
935 		if (err < 0 && err != -ENOENT)
936 			goto fail;
937 		break;
938 	default:
939 		BUG();
940 	}
941 
942 skip_partial:
943 	while (cont) {
944 		dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
945 		switch (offset[0]) {
946 		case NODE_DIR1_BLOCK:
947 		case NODE_DIR2_BLOCK:
948 			err = truncate_dnode(&dn);
949 			break;
950 
951 		case NODE_IND1_BLOCK:
952 		case NODE_IND2_BLOCK:
953 			err = truncate_nodes(&dn, nofs, offset[1], 2);
954 			break;
955 
956 		case NODE_DIND_BLOCK:
957 			err = truncate_nodes(&dn, nofs, offset[1], 3);
958 			cont = 0;
959 			break;
960 
961 		default:
962 			BUG();
963 		}
964 		if (err < 0 && err != -ENOENT)
965 			goto fail;
966 		if (offset[1] == 0 &&
967 				ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
968 			lock_page(page);
969 			BUG_ON(page->mapping != NODE_MAPPING(sbi));
970 			f2fs_wait_on_page_writeback(page, NODE, true);
971 			ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
972 			set_page_dirty(page);
973 			unlock_page(page);
974 		}
975 		offset[1] = 0;
976 		offset[0]++;
977 		nofs += err;
978 	}
979 fail:
980 	f2fs_put_page(page, 0);
981 	trace_f2fs_truncate_inode_blocks_exit(inode, err);
982 	return err > 0 ? 0 : err;
983 }
984 
985 /* caller must lock inode page */
986 int truncate_xattr_node(struct inode *inode)
987 {
988 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
989 	nid_t nid = F2FS_I(inode)->i_xattr_nid;
990 	struct dnode_of_data dn;
991 	struct page *npage;
992 
993 	if (!nid)
994 		return 0;
995 
996 	npage = get_node_page(sbi, nid);
997 	if (IS_ERR(npage))
998 		return PTR_ERR(npage);
999 
1000 	f2fs_i_xnid_write(inode, 0);
1001 
1002 	set_new_dnode(&dn, inode, NULL, npage, nid);
1003 	truncate_node(&dn);
1004 	return 0;
1005 }
1006 
1007 /*
1008  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1009  * f2fs_unlock_op().
1010  */
1011 int remove_inode_page(struct inode *inode)
1012 {
1013 	struct dnode_of_data dn;
1014 	int err;
1015 
1016 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1017 	err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1018 	if (err)
1019 		return err;
1020 
1021 	err = truncate_xattr_node(inode);
1022 	if (err) {
1023 		f2fs_put_dnode(&dn);
1024 		return err;
1025 	}
1026 
1027 	/* remove potential inline_data blocks */
1028 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1029 				S_ISLNK(inode->i_mode))
1030 		truncate_data_blocks_range(&dn, 1);
1031 
1032 	/* 0 is possible, after f2fs_new_inode() has failed */
1033 	f2fs_bug_on(F2FS_I_SB(inode),
1034 			inode->i_blocks != 0 && inode->i_blocks != 8);
1035 
1036 	/* will put inode & node pages */
1037 	truncate_node(&dn);
1038 	return 0;
1039 }
1040 
1041 struct page *new_inode_page(struct inode *inode)
1042 {
1043 	struct dnode_of_data dn;
1044 
1045 	/* allocate inode page for new inode */
1046 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1047 
1048 	/* caller should f2fs_put_page(page, 1); */
1049 	return new_node_page(&dn, 0);
1050 }
1051 
1052 struct page *new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1053 {
1054 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1055 	struct node_info new_ni;
1056 	struct page *page;
1057 	int err;
1058 
1059 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1060 		return ERR_PTR(-EPERM);
1061 
1062 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1063 	if (!page)
1064 		return ERR_PTR(-ENOMEM);
1065 
1066 	if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1067 		goto fail;
1068 
1069 #ifdef CONFIG_F2FS_CHECK_FS
1070 	get_node_info(sbi, dn->nid, &new_ni);
1071 	f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1072 #endif
1073 	new_ni.nid = dn->nid;
1074 	new_ni.ino = dn->inode->i_ino;
1075 	new_ni.blk_addr = NULL_ADDR;
1076 	new_ni.flag = 0;
1077 	new_ni.version = 0;
1078 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1079 
1080 	f2fs_wait_on_page_writeback(page, NODE, true);
1081 	fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1082 	set_cold_node(dn->inode, page);
1083 	if (!PageUptodate(page))
1084 		SetPageUptodate(page);
1085 	if (set_page_dirty(page))
1086 		dn->node_changed = true;
1087 
1088 	if (f2fs_has_xattr_block(ofs))
1089 		f2fs_i_xnid_write(dn->inode, dn->nid);
1090 
1091 	if (ofs == 0)
1092 		inc_valid_inode_count(sbi);
1093 	return page;
1094 
1095 fail:
1096 	clear_node_page_dirty(page);
1097 	f2fs_put_page(page, 1);
1098 	return ERR_PTR(err);
1099 }
1100 
1101 /*
1102  * Caller should do after getting the following values.
1103  * 0: f2fs_put_page(page, 0)
1104  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1105  */
1106 static int read_node_page(struct page *page, int op_flags)
1107 {
1108 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1109 	struct node_info ni;
1110 	struct f2fs_io_info fio = {
1111 		.sbi = sbi,
1112 		.type = NODE,
1113 		.op = REQ_OP_READ,
1114 		.op_flags = op_flags,
1115 		.page = page,
1116 		.encrypted_page = NULL,
1117 	};
1118 
1119 	if (PageUptodate(page))
1120 		return LOCKED_PAGE;
1121 
1122 	get_node_info(sbi, page->index, &ni);
1123 
1124 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1125 		ClearPageUptodate(page);
1126 		return -ENOENT;
1127 	}
1128 
1129 	fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1130 	return f2fs_submit_page_bio(&fio);
1131 }
1132 
1133 /*
1134  * Readahead a node page
1135  */
1136 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1137 {
1138 	struct page *apage;
1139 	int err;
1140 
1141 	if (!nid)
1142 		return;
1143 	f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1144 
1145 	rcu_read_lock();
1146 	apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1147 	rcu_read_unlock();
1148 	if (apage)
1149 		return;
1150 
1151 	apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1152 	if (!apage)
1153 		return;
1154 
1155 	err = read_node_page(apage, REQ_RAHEAD);
1156 	f2fs_put_page(apage, err ? 1 : 0);
1157 }
1158 
1159 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1160 					struct page *parent, int start)
1161 {
1162 	struct page *page;
1163 	int err;
1164 
1165 	if (!nid)
1166 		return ERR_PTR(-ENOENT);
1167 	f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1168 repeat:
1169 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1170 	if (!page)
1171 		return ERR_PTR(-ENOMEM);
1172 
1173 	err = read_node_page(page, 0);
1174 	if (err < 0) {
1175 		f2fs_put_page(page, 1);
1176 		return ERR_PTR(err);
1177 	} else if (err == LOCKED_PAGE) {
1178 		err = 0;
1179 		goto page_hit;
1180 	}
1181 
1182 	if (parent)
1183 		ra_node_pages(parent, start + 1, MAX_RA_NODE);
1184 
1185 	lock_page(page);
1186 
1187 	if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1188 		f2fs_put_page(page, 1);
1189 		goto repeat;
1190 	}
1191 
1192 	if (unlikely(!PageUptodate(page))) {
1193 		err = -EIO;
1194 		goto out_err;
1195 	}
1196 
1197 	if (!f2fs_inode_chksum_verify(sbi, page)) {
1198 		err = -EBADMSG;
1199 		goto out_err;
1200 	}
1201 page_hit:
1202 	if(unlikely(nid != nid_of_node(page))) {
1203 		f2fs_msg(sbi->sb, KERN_WARNING, "inconsistent node block, "
1204 			"nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1205 			nid, nid_of_node(page), ino_of_node(page),
1206 			ofs_of_node(page), cpver_of_node(page),
1207 			next_blkaddr_of_node(page));
1208 		err = -EINVAL;
1209 out_err:
1210 		ClearPageUptodate(page);
1211 		f2fs_put_page(page, 1);
1212 		return ERR_PTR(err);
1213 	}
1214 	return page;
1215 }
1216 
1217 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1218 {
1219 	return __get_node_page(sbi, nid, NULL, 0);
1220 }
1221 
1222 struct page *get_node_page_ra(struct page *parent, int start)
1223 {
1224 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1225 	nid_t nid = get_nid(parent, start, false);
1226 
1227 	return __get_node_page(sbi, nid, parent, start);
1228 }
1229 
1230 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1231 {
1232 	struct inode *inode;
1233 	struct page *page;
1234 	int ret;
1235 
1236 	/* should flush inline_data before evict_inode */
1237 	inode = ilookup(sbi->sb, ino);
1238 	if (!inode)
1239 		return;
1240 
1241 	page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1242 					FGP_LOCK|FGP_NOWAIT, 0);
1243 	if (!page)
1244 		goto iput_out;
1245 
1246 	if (!PageUptodate(page))
1247 		goto page_out;
1248 
1249 	if (!PageDirty(page))
1250 		goto page_out;
1251 
1252 	if (!clear_page_dirty_for_io(page))
1253 		goto page_out;
1254 
1255 	ret = f2fs_write_inline_data(inode, page);
1256 	inode_dec_dirty_pages(inode);
1257 	remove_dirty_inode(inode);
1258 	if (ret)
1259 		set_page_dirty(page);
1260 page_out:
1261 	f2fs_put_page(page, 1);
1262 iput_out:
1263 	iput(inode);
1264 }
1265 
1266 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1267 {
1268 	pgoff_t index;
1269 	struct pagevec pvec;
1270 	struct page *last_page = NULL;
1271 	int nr_pages;
1272 
1273 	pagevec_init(&pvec);
1274 	index = 0;
1275 
1276 	while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1277 				PAGECACHE_TAG_DIRTY))) {
1278 		int i;
1279 
1280 		for (i = 0; i < nr_pages; i++) {
1281 			struct page *page = pvec.pages[i];
1282 
1283 			if (unlikely(f2fs_cp_error(sbi))) {
1284 				f2fs_put_page(last_page, 0);
1285 				pagevec_release(&pvec);
1286 				return ERR_PTR(-EIO);
1287 			}
1288 
1289 			if (!IS_DNODE(page) || !is_cold_node(page))
1290 				continue;
1291 			if (ino_of_node(page) != ino)
1292 				continue;
1293 
1294 			lock_page(page);
1295 
1296 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1297 continue_unlock:
1298 				unlock_page(page);
1299 				continue;
1300 			}
1301 			if (ino_of_node(page) != ino)
1302 				goto continue_unlock;
1303 
1304 			if (!PageDirty(page)) {
1305 				/* someone wrote it for us */
1306 				goto continue_unlock;
1307 			}
1308 
1309 			if (last_page)
1310 				f2fs_put_page(last_page, 0);
1311 
1312 			get_page(page);
1313 			last_page = page;
1314 			unlock_page(page);
1315 		}
1316 		pagevec_release(&pvec);
1317 		cond_resched();
1318 	}
1319 	return last_page;
1320 }
1321 
1322 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1323 				struct writeback_control *wbc, bool do_balance,
1324 				enum iostat_type io_type)
1325 {
1326 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1327 	nid_t nid;
1328 	struct node_info ni;
1329 	struct f2fs_io_info fio = {
1330 		.sbi = sbi,
1331 		.ino = ino_of_node(page),
1332 		.type = NODE,
1333 		.op = REQ_OP_WRITE,
1334 		.op_flags = wbc_to_write_flags(wbc),
1335 		.page = page,
1336 		.encrypted_page = NULL,
1337 		.submitted = false,
1338 		.io_type = io_type,
1339 	};
1340 
1341 	trace_f2fs_writepage(page, NODE);
1342 
1343 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1344 		goto redirty_out;
1345 	if (unlikely(f2fs_cp_error(sbi)))
1346 		goto redirty_out;
1347 
1348 	/* get old block addr of this node page */
1349 	nid = nid_of_node(page);
1350 	f2fs_bug_on(sbi, page->index != nid);
1351 
1352 	if (wbc->for_reclaim) {
1353 		if (!down_read_trylock(&sbi->node_write))
1354 			goto redirty_out;
1355 	} else {
1356 		down_read(&sbi->node_write);
1357 	}
1358 
1359 	get_node_info(sbi, nid, &ni);
1360 
1361 	/* This page is already truncated */
1362 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1363 		ClearPageUptodate(page);
1364 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1365 		up_read(&sbi->node_write);
1366 		unlock_page(page);
1367 		return 0;
1368 	}
1369 
1370 	if (atomic && !test_opt(sbi, NOBARRIER))
1371 		fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1372 
1373 	set_page_writeback(page);
1374 	fio.old_blkaddr = ni.blk_addr;
1375 	write_node_page(nid, &fio);
1376 	set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1377 	dec_page_count(sbi, F2FS_DIRTY_NODES);
1378 	up_read(&sbi->node_write);
1379 
1380 	if (wbc->for_reclaim) {
1381 		f2fs_submit_merged_write_cond(sbi, page->mapping->host, 0,
1382 						page->index, NODE);
1383 		submitted = NULL;
1384 	}
1385 
1386 	unlock_page(page);
1387 
1388 	if (unlikely(f2fs_cp_error(sbi))) {
1389 		f2fs_submit_merged_write(sbi, NODE);
1390 		submitted = NULL;
1391 	}
1392 	if (submitted)
1393 		*submitted = fio.submitted;
1394 
1395 	if (do_balance)
1396 		f2fs_balance_fs(sbi, false);
1397 	return 0;
1398 
1399 redirty_out:
1400 	redirty_page_for_writepage(wbc, page);
1401 	return AOP_WRITEPAGE_ACTIVATE;
1402 }
1403 
1404 void move_node_page(struct page *node_page, int gc_type)
1405 {
1406 	if (gc_type == FG_GC) {
1407 		struct writeback_control wbc = {
1408 			.sync_mode = WB_SYNC_ALL,
1409 			.nr_to_write = 1,
1410 			.for_reclaim = 0,
1411 		};
1412 
1413 		set_page_dirty(node_page);
1414 		f2fs_wait_on_page_writeback(node_page, NODE, true);
1415 
1416 		f2fs_bug_on(F2FS_P_SB(node_page), PageWriteback(node_page));
1417 		if (!clear_page_dirty_for_io(node_page))
1418 			goto out_page;
1419 
1420 		if (__write_node_page(node_page, false, NULL,
1421 					&wbc, false, FS_GC_NODE_IO))
1422 			unlock_page(node_page);
1423 		goto release_page;
1424 	} else {
1425 		/* set page dirty and write it */
1426 		if (!PageWriteback(node_page))
1427 			set_page_dirty(node_page);
1428 	}
1429 out_page:
1430 	unlock_page(node_page);
1431 release_page:
1432 	f2fs_put_page(node_page, 0);
1433 }
1434 
1435 static int f2fs_write_node_page(struct page *page,
1436 				struct writeback_control *wbc)
1437 {
1438 	return __write_node_page(page, false, NULL, wbc, false, FS_NODE_IO);
1439 }
1440 
1441 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1442 			struct writeback_control *wbc, bool atomic)
1443 {
1444 	pgoff_t index;
1445 	pgoff_t last_idx = ULONG_MAX;
1446 	struct pagevec pvec;
1447 	int ret = 0;
1448 	struct page *last_page = NULL;
1449 	bool marked = false;
1450 	nid_t ino = inode->i_ino;
1451 	int nr_pages;
1452 
1453 	if (atomic) {
1454 		last_page = last_fsync_dnode(sbi, ino);
1455 		if (IS_ERR_OR_NULL(last_page))
1456 			return PTR_ERR_OR_ZERO(last_page);
1457 	}
1458 retry:
1459 	pagevec_init(&pvec);
1460 	index = 0;
1461 
1462 	while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1463 				PAGECACHE_TAG_DIRTY))) {
1464 		int i;
1465 
1466 		for (i = 0; i < nr_pages; i++) {
1467 			struct page *page = pvec.pages[i];
1468 			bool submitted = false;
1469 
1470 			if (unlikely(f2fs_cp_error(sbi))) {
1471 				f2fs_put_page(last_page, 0);
1472 				pagevec_release(&pvec);
1473 				ret = -EIO;
1474 				goto out;
1475 			}
1476 
1477 			if (!IS_DNODE(page) || !is_cold_node(page))
1478 				continue;
1479 			if (ino_of_node(page) != ino)
1480 				continue;
1481 
1482 			lock_page(page);
1483 
1484 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1485 continue_unlock:
1486 				unlock_page(page);
1487 				continue;
1488 			}
1489 			if (ino_of_node(page) != ino)
1490 				goto continue_unlock;
1491 
1492 			if (!PageDirty(page) && page != last_page) {
1493 				/* someone wrote it for us */
1494 				goto continue_unlock;
1495 			}
1496 
1497 			f2fs_wait_on_page_writeback(page, NODE, true);
1498 			BUG_ON(PageWriteback(page));
1499 
1500 			set_fsync_mark(page, 0);
1501 			set_dentry_mark(page, 0);
1502 
1503 			if (!atomic || page == last_page) {
1504 				set_fsync_mark(page, 1);
1505 				if (IS_INODE(page)) {
1506 					if (is_inode_flag_set(inode,
1507 								FI_DIRTY_INODE))
1508 						update_inode(inode, page);
1509 					set_dentry_mark(page,
1510 						need_dentry_mark(sbi, ino));
1511 				}
1512 				/*  may be written by other thread */
1513 				if (!PageDirty(page))
1514 					set_page_dirty(page);
1515 			}
1516 
1517 			if (!clear_page_dirty_for_io(page))
1518 				goto continue_unlock;
1519 
1520 			ret = __write_node_page(page, atomic &&
1521 						page == last_page,
1522 						&submitted, wbc, true,
1523 						FS_NODE_IO);
1524 			if (ret) {
1525 				unlock_page(page);
1526 				f2fs_put_page(last_page, 0);
1527 				break;
1528 			} else if (submitted) {
1529 				last_idx = page->index;
1530 			}
1531 
1532 			if (page == last_page) {
1533 				f2fs_put_page(page, 0);
1534 				marked = true;
1535 				break;
1536 			}
1537 		}
1538 		pagevec_release(&pvec);
1539 		cond_resched();
1540 
1541 		if (ret || marked)
1542 			break;
1543 	}
1544 	if (!ret && atomic && !marked) {
1545 		f2fs_msg(sbi->sb, KERN_DEBUG,
1546 			"Retry to write fsync mark: ino=%u, idx=%lx",
1547 					ino, last_page->index);
1548 		lock_page(last_page);
1549 		f2fs_wait_on_page_writeback(last_page, NODE, true);
1550 		set_page_dirty(last_page);
1551 		unlock_page(last_page);
1552 		goto retry;
1553 	}
1554 out:
1555 	if (last_idx != ULONG_MAX)
1556 		f2fs_submit_merged_write_cond(sbi, NULL, ino, last_idx, NODE);
1557 	return ret ? -EIO: 0;
1558 }
1559 
1560 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc,
1561 				bool do_balance, enum iostat_type io_type)
1562 {
1563 	pgoff_t index;
1564 	struct pagevec pvec;
1565 	int step = 0;
1566 	int nwritten = 0;
1567 	int ret = 0;
1568 	int nr_pages;
1569 
1570 	pagevec_init(&pvec);
1571 
1572 next_step:
1573 	index = 0;
1574 
1575 	while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1576 				PAGECACHE_TAG_DIRTY))) {
1577 		int i;
1578 
1579 		for (i = 0; i < nr_pages; i++) {
1580 			struct page *page = pvec.pages[i];
1581 			bool submitted = false;
1582 
1583 			if (unlikely(f2fs_cp_error(sbi))) {
1584 				pagevec_release(&pvec);
1585 				ret = -EIO;
1586 				goto out;
1587 			}
1588 
1589 			/*
1590 			 * flushing sequence with step:
1591 			 * 0. indirect nodes
1592 			 * 1. dentry dnodes
1593 			 * 2. file dnodes
1594 			 */
1595 			if (step == 0 && IS_DNODE(page))
1596 				continue;
1597 			if (step == 1 && (!IS_DNODE(page) ||
1598 						is_cold_node(page)))
1599 				continue;
1600 			if (step == 2 && (!IS_DNODE(page) ||
1601 						!is_cold_node(page)))
1602 				continue;
1603 lock_node:
1604 			if (!trylock_page(page))
1605 				continue;
1606 
1607 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1608 continue_unlock:
1609 				unlock_page(page);
1610 				continue;
1611 			}
1612 
1613 			if (!PageDirty(page)) {
1614 				/* someone wrote it for us */
1615 				goto continue_unlock;
1616 			}
1617 
1618 			/* flush inline_data */
1619 			if (is_inline_node(page)) {
1620 				clear_inline_node(page);
1621 				unlock_page(page);
1622 				flush_inline_data(sbi, ino_of_node(page));
1623 				goto lock_node;
1624 			}
1625 
1626 			f2fs_wait_on_page_writeback(page, NODE, true);
1627 
1628 			BUG_ON(PageWriteback(page));
1629 			if (!clear_page_dirty_for_io(page))
1630 				goto continue_unlock;
1631 
1632 			set_fsync_mark(page, 0);
1633 			set_dentry_mark(page, 0);
1634 
1635 			ret = __write_node_page(page, false, &submitted,
1636 						wbc, do_balance, io_type);
1637 			if (ret)
1638 				unlock_page(page);
1639 			else if (submitted)
1640 				nwritten++;
1641 
1642 			if (--wbc->nr_to_write == 0)
1643 				break;
1644 		}
1645 		pagevec_release(&pvec);
1646 		cond_resched();
1647 
1648 		if (wbc->nr_to_write == 0) {
1649 			step = 2;
1650 			break;
1651 		}
1652 	}
1653 
1654 	if (step < 2) {
1655 		step++;
1656 		goto next_step;
1657 	}
1658 out:
1659 	if (nwritten)
1660 		f2fs_submit_merged_write(sbi, NODE);
1661 	return ret;
1662 }
1663 
1664 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1665 {
1666 	pgoff_t index = 0;
1667 	struct pagevec pvec;
1668 	int ret2, ret = 0;
1669 	int nr_pages;
1670 
1671 	pagevec_init(&pvec);
1672 
1673 	while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1674 				PAGECACHE_TAG_WRITEBACK))) {
1675 		int i;
1676 
1677 		for (i = 0; i < nr_pages; i++) {
1678 			struct page *page = pvec.pages[i];
1679 
1680 			if (ino && ino_of_node(page) == ino) {
1681 				f2fs_wait_on_page_writeback(page, NODE, true);
1682 				if (TestClearPageError(page))
1683 					ret = -EIO;
1684 			}
1685 		}
1686 		pagevec_release(&pvec);
1687 		cond_resched();
1688 	}
1689 
1690 	ret2 = filemap_check_errors(NODE_MAPPING(sbi));
1691 	if (!ret)
1692 		ret = ret2;
1693 	return ret;
1694 }
1695 
1696 static int f2fs_write_node_pages(struct address_space *mapping,
1697 			    struct writeback_control *wbc)
1698 {
1699 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1700 	struct blk_plug plug;
1701 	long diff;
1702 
1703 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1704 		goto skip_write;
1705 
1706 	/* balancing f2fs's metadata in background */
1707 	f2fs_balance_fs_bg(sbi);
1708 
1709 	/* collect a number of dirty node pages and write together */
1710 	if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1711 		goto skip_write;
1712 
1713 	trace_f2fs_writepages(mapping->host, wbc, NODE);
1714 
1715 	diff = nr_pages_to_write(sbi, NODE, wbc);
1716 	wbc->sync_mode = WB_SYNC_NONE;
1717 	blk_start_plug(&plug);
1718 	sync_node_pages(sbi, wbc, true, FS_NODE_IO);
1719 	blk_finish_plug(&plug);
1720 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1721 	return 0;
1722 
1723 skip_write:
1724 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1725 	trace_f2fs_writepages(mapping->host, wbc, NODE);
1726 	return 0;
1727 }
1728 
1729 static int f2fs_set_node_page_dirty(struct page *page)
1730 {
1731 	trace_f2fs_set_page_dirty(page, NODE);
1732 
1733 	if (!PageUptodate(page))
1734 		SetPageUptodate(page);
1735 	if (!PageDirty(page)) {
1736 		f2fs_set_page_dirty_nobuffers(page);
1737 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1738 		SetPagePrivate(page);
1739 		f2fs_trace_pid(page);
1740 		return 1;
1741 	}
1742 	return 0;
1743 }
1744 
1745 /*
1746  * Structure of the f2fs node operations
1747  */
1748 const struct address_space_operations f2fs_node_aops = {
1749 	.writepage	= f2fs_write_node_page,
1750 	.writepages	= f2fs_write_node_pages,
1751 	.set_page_dirty	= f2fs_set_node_page_dirty,
1752 	.invalidatepage	= f2fs_invalidate_page,
1753 	.releasepage	= f2fs_release_page,
1754 #ifdef CONFIG_MIGRATION
1755 	.migratepage    = f2fs_migrate_page,
1756 #endif
1757 };
1758 
1759 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1760 						nid_t n)
1761 {
1762 	return radix_tree_lookup(&nm_i->free_nid_root, n);
1763 }
1764 
1765 static int __insert_free_nid(struct f2fs_sb_info *sbi,
1766 			struct free_nid *i, enum nid_state state)
1767 {
1768 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1769 
1770 	int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
1771 	if (err)
1772 		return err;
1773 
1774 	f2fs_bug_on(sbi, state != i->state);
1775 	nm_i->nid_cnt[state]++;
1776 	if (state == FREE_NID)
1777 		list_add_tail(&i->list, &nm_i->free_nid_list);
1778 	return 0;
1779 }
1780 
1781 static void __remove_free_nid(struct f2fs_sb_info *sbi,
1782 			struct free_nid *i, enum nid_state state)
1783 {
1784 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1785 
1786 	f2fs_bug_on(sbi, state != i->state);
1787 	nm_i->nid_cnt[state]--;
1788 	if (state == FREE_NID)
1789 		list_del(&i->list);
1790 	radix_tree_delete(&nm_i->free_nid_root, i->nid);
1791 }
1792 
1793 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
1794 			enum nid_state org_state, enum nid_state dst_state)
1795 {
1796 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1797 
1798 	f2fs_bug_on(sbi, org_state != i->state);
1799 	i->state = dst_state;
1800 	nm_i->nid_cnt[org_state]--;
1801 	nm_i->nid_cnt[dst_state]++;
1802 
1803 	switch (dst_state) {
1804 	case PREALLOC_NID:
1805 		list_del(&i->list);
1806 		break;
1807 	case FREE_NID:
1808 		list_add_tail(&i->list, &nm_i->free_nid_list);
1809 		break;
1810 	default:
1811 		BUG_ON(1);
1812 	}
1813 }
1814 
1815 /* return if the nid is recognized as free */
1816 static bool add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1817 {
1818 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1819 	struct free_nid *i, *e;
1820 	struct nat_entry *ne;
1821 	int err = -EINVAL;
1822 	bool ret = false;
1823 
1824 	/* 0 nid should not be used */
1825 	if (unlikely(nid == 0))
1826 		return false;
1827 
1828 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1829 	i->nid = nid;
1830 	i->state = FREE_NID;
1831 
1832 	if (radix_tree_preload(GFP_NOFS))
1833 		goto err;
1834 
1835 	spin_lock(&nm_i->nid_list_lock);
1836 
1837 	if (build) {
1838 		/*
1839 		 *   Thread A             Thread B
1840 		 *  - f2fs_create
1841 		 *   - f2fs_new_inode
1842 		 *    - alloc_nid
1843 		 *     - __insert_nid_to_list(PREALLOC_NID)
1844 		 *                     - f2fs_balance_fs_bg
1845 		 *                      - build_free_nids
1846 		 *                       - __build_free_nids
1847 		 *                        - scan_nat_page
1848 		 *                         - add_free_nid
1849 		 *                          - __lookup_nat_cache
1850 		 *  - f2fs_add_link
1851 		 *   - init_inode_metadata
1852 		 *    - new_inode_page
1853 		 *     - new_node_page
1854 		 *      - set_node_addr
1855 		 *  - alloc_nid_done
1856 		 *   - __remove_nid_from_list(PREALLOC_NID)
1857 		 *                         - __insert_nid_to_list(FREE_NID)
1858 		 */
1859 		ne = __lookup_nat_cache(nm_i, nid);
1860 		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1861 				nat_get_blkaddr(ne) != NULL_ADDR))
1862 			goto err_out;
1863 
1864 		e = __lookup_free_nid_list(nm_i, nid);
1865 		if (e) {
1866 			if (e->state == FREE_NID)
1867 				ret = true;
1868 			goto err_out;
1869 		}
1870 	}
1871 	ret = true;
1872 	err = __insert_free_nid(sbi, i, FREE_NID);
1873 err_out:
1874 	spin_unlock(&nm_i->nid_list_lock);
1875 	radix_tree_preload_end();
1876 err:
1877 	if (err)
1878 		kmem_cache_free(free_nid_slab, i);
1879 	return ret;
1880 }
1881 
1882 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
1883 {
1884 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1885 	struct free_nid *i;
1886 	bool need_free = false;
1887 
1888 	spin_lock(&nm_i->nid_list_lock);
1889 	i = __lookup_free_nid_list(nm_i, nid);
1890 	if (i && i->state == FREE_NID) {
1891 		__remove_free_nid(sbi, i, FREE_NID);
1892 		need_free = true;
1893 	}
1894 	spin_unlock(&nm_i->nid_list_lock);
1895 
1896 	if (need_free)
1897 		kmem_cache_free(free_nid_slab, i);
1898 }
1899 
1900 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
1901 							bool set, bool build)
1902 {
1903 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1904 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
1905 	unsigned int nid_ofs = nid - START_NID(nid);
1906 
1907 	if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1908 		return;
1909 
1910 	if (set) {
1911 		if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
1912 			return;
1913 		__set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1914 		nm_i->free_nid_count[nat_ofs]++;
1915 	} else {
1916 		if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
1917 			return;
1918 		__clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
1919 		if (!build)
1920 			nm_i->free_nid_count[nat_ofs]--;
1921 	}
1922 }
1923 
1924 static void scan_nat_page(struct f2fs_sb_info *sbi,
1925 			struct page *nat_page, nid_t start_nid)
1926 {
1927 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1928 	struct f2fs_nat_block *nat_blk = page_address(nat_page);
1929 	block_t blk_addr;
1930 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
1931 	int i;
1932 
1933 	if (test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
1934 		return;
1935 
1936 	__set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
1937 
1938 	i = start_nid % NAT_ENTRY_PER_BLOCK;
1939 
1940 	for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1941 		bool freed = false;
1942 
1943 		if (unlikely(start_nid >= nm_i->max_nid))
1944 			break;
1945 
1946 		blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1947 		f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1948 		if (blk_addr == NULL_ADDR)
1949 			freed = add_free_nid(sbi, start_nid, true);
1950 		spin_lock(&NM_I(sbi)->nid_list_lock);
1951 		update_free_nid_bitmap(sbi, start_nid, freed, true);
1952 		spin_unlock(&NM_I(sbi)->nid_list_lock);
1953 	}
1954 }
1955 
1956 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
1957 {
1958 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1959 	struct f2fs_journal *journal = curseg->journal;
1960 	int i;
1961 
1962 	down_read(&curseg->journal_rwsem);
1963 	for (i = 0; i < nats_in_cursum(journal); i++) {
1964 		block_t addr;
1965 		nid_t nid;
1966 
1967 		addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1968 		nid = le32_to_cpu(nid_in_journal(journal, i));
1969 		if (addr == NULL_ADDR)
1970 			add_free_nid(sbi, nid, true);
1971 		else
1972 			remove_free_nid(sbi, nid);
1973 	}
1974 	up_read(&curseg->journal_rwsem);
1975 }
1976 
1977 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
1978 {
1979 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1980 	unsigned int i, idx;
1981 	nid_t nid;
1982 
1983 	down_read(&nm_i->nat_tree_lock);
1984 
1985 	for (i = 0; i < nm_i->nat_blocks; i++) {
1986 		if (!test_bit_le(i, nm_i->nat_block_bitmap))
1987 			continue;
1988 		if (!nm_i->free_nid_count[i])
1989 			continue;
1990 		for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
1991 			idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
1992 						NAT_ENTRY_PER_BLOCK, idx);
1993 			if (idx >= NAT_ENTRY_PER_BLOCK)
1994 				break;
1995 
1996 			nid = i * NAT_ENTRY_PER_BLOCK + idx;
1997 			add_free_nid(sbi, nid, true);
1998 
1999 			if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2000 				goto out;
2001 		}
2002 	}
2003 out:
2004 	scan_curseg_cache(sbi);
2005 
2006 	up_read(&nm_i->nat_tree_lock);
2007 }
2008 
2009 static void __build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2010 {
2011 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2012 	int i = 0;
2013 	nid_t nid = nm_i->next_scan_nid;
2014 
2015 	if (unlikely(nid >= nm_i->max_nid))
2016 		nid = 0;
2017 
2018 	/* Enough entries */
2019 	if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2020 		return;
2021 
2022 	if (!sync && !available_free_memory(sbi, FREE_NIDS))
2023 		return;
2024 
2025 	if (!mount) {
2026 		/* try to find free nids in free_nid_bitmap */
2027 		scan_free_nid_bits(sbi);
2028 
2029 		if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2030 			return;
2031 	}
2032 
2033 	/* readahead nat pages to be scanned */
2034 	ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2035 							META_NAT, true);
2036 
2037 	down_read(&nm_i->nat_tree_lock);
2038 
2039 	while (1) {
2040 		struct page *page = get_current_nat_page(sbi, nid);
2041 
2042 		scan_nat_page(sbi, page, nid);
2043 		f2fs_put_page(page, 1);
2044 
2045 		nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2046 		if (unlikely(nid >= nm_i->max_nid))
2047 			nid = 0;
2048 
2049 		if (++i >= FREE_NID_PAGES)
2050 			break;
2051 	}
2052 
2053 	/* go to the next free nat pages to find free nids abundantly */
2054 	nm_i->next_scan_nid = nid;
2055 
2056 	/* find free nids from current sum_pages */
2057 	scan_curseg_cache(sbi);
2058 
2059 	up_read(&nm_i->nat_tree_lock);
2060 
2061 	ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2062 					nm_i->ra_nid_pages, META_NAT, false);
2063 }
2064 
2065 void build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2066 {
2067 	mutex_lock(&NM_I(sbi)->build_lock);
2068 	__build_free_nids(sbi, sync, mount);
2069 	mutex_unlock(&NM_I(sbi)->build_lock);
2070 }
2071 
2072 /*
2073  * If this function returns success, caller can obtain a new nid
2074  * from second parameter of this function.
2075  * The returned nid could be used ino as well as nid when inode is created.
2076  */
2077 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2078 {
2079 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2080 	struct free_nid *i = NULL;
2081 retry:
2082 #ifdef CONFIG_F2FS_FAULT_INJECTION
2083 	if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2084 		f2fs_show_injection_info(FAULT_ALLOC_NID);
2085 		return false;
2086 	}
2087 #endif
2088 	spin_lock(&nm_i->nid_list_lock);
2089 
2090 	if (unlikely(nm_i->available_nids == 0)) {
2091 		spin_unlock(&nm_i->nid_list_lock);
2092 		return false;
2093 	}
2094 
2095 	/* We should not use stale free nids created by build_free_nids */
2096 	if (nm_i->nid_cnt[FREE_NID] && !on_build_free_nids(nm_i)) {
2097 		f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2098 		i = list_first_entry(&nm_i->free_nid_list,
2099 					struct free_nid, list);
2100 		*nid = i->nid;
2101 
2102 		__move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2103 		nm_i->available_nids--;
2104 
2105 		update_free_nid_bitmap(sbi, *nid, false, false);
2106 
2107 		spin_unlock(&nm_i->nid_list_lock);
2108 		return true;
2109 	}
2110 	spin_unlock(&nm_i->nid_list_lock);
2111 
2112 	/* Let's scan nat pages and its caches to get free nids */
2113 	build_free_nids(sbi, true, false);
2114 	goto retry;
2115 }
2116 
2117 /*
2118  * alloc_nid() should be called prior to this function.
2119  */
2120 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2121 {
2122 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2123 	struct free_nid *i;
2124 
2125 	spin_lock(&nm_i->nid_list_lock);
2126 	i = __lookup_free_nid_list(nm_i, nid);
2127 	f2fs_bug_on(sbi, !i);
2128 	__remove_free_nid(sbi, i, PREALLOC_NID);
2129 	spin_unlock(&nm_i->nid_list_lock);
2130 
2131 	kmem_cache_free(free_nid_slab, i);
2132 }
2133 
2134 /*
2135  * alloc_nid() should be called prior to this function.
2136  */
2137 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2138 {
2139 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2140 	struct free_nid *i;
2141 	bool need_free = false;
2142 
2143 	if (!nid)
2144 		return;
2145 
2146 	spin_lock(&nm_i->nid_list_lock);
2147 	i = __lookup_free_nid_list(nm_i, nid);
2148 	f2fs_bug_on(sbi, !i);
2149 
2150 	if (!available_free_memory(sbi, FREE_NIDS)) {
2151 		__remove_free_nid(sbi, i, PREALLOC_NID);
2152 		need_free = true;
2153 	} else {
2154 		__move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2155 	}
2156 
2157 	nm_i->available_nids++;
2158 
2159 	update_free_nid_bitmap(sbi, nid, true, false);
2160 
2161 	spin_unlock(&nm_i->nid_list_lock);
2162 
2163 	if (need_free)
2164 		kmem_cache_free(free_nid_slab, i);
2165 }
2166 
2167 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2168 {
2169 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2170 	struct free_nid *i, *next;
2171 	int nr = nr_shrink;
2172 
2173 	if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2174 		return 0;
2175 
2176 	if (!mutex_trylock(&nm_i->build_lock))
2177 		return 0;
2178 
2179 	spin_lock(&nm_i->nid_list_lock);
2180 	list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2181 		if (nr_shrink <= 0 ||
2182 				nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2183 			break;
2184 
2185 		__remove_free_nid(sbi, i, FREE_NID);
2186 		kmem_cache_free(free_nid_slab, i);
2187 		nr_shrink--;
2188 	}
2189 	spin_unlock(&nm_i->nid_list_lock);
2190 	mutex_unlock(&nm_i->build_lock);
2191 
2192 	return nr - nr_shrink;
2193 }
2194 
2195 void recover_inline_xattr(struct inode *inode, struct page *page)
2196 {
2197 	void *src_addr, *dst_addr;
2198 	size_t inline_size;
2199 	struct page *ipage;
2200 	struct f2fs_inode *ri;
2201 
2202 	ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
2203 	f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
2204 
2205 	ri = F2FS_INODE(page);
2206 	if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
2207 		clear_inode_flag(inode, FI_INLINE_XATTR);
2208 		goto update_inode;
2209 	}
2210 
2211 	dst_addr = inline_xattr_addr(inode, ipage);
2212 	src_addr = inline_xattr_addr(inode, page);
2213 	inline_size = inline_xattr_size(inode);
2214 
2215 	f2fs_wait_on_page_writeback(ipage, NODE, true);
2216 	memcpy(dst_addr, src_addr, inline_size);
2217 update_inode:
2218 	update_inode(inode, ipage);
2219 	f2fs_put_page(ipage, 1);
2220 }
2221 
2222 int recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
2223 {
2224 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2225 	nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2226 	nid_t new_xnid;
2227 	struct dnode_of_data dn;
2228 	struct node_info ni;
2229 	struct page *xpage;
2230 
2231 	if (!prev_xnid)
2232 		goto recover_xnid;
2233 
2234 	/* 1: invalidate the previous xattr nid */
2235 	get_node_info(sbi, prev_xnid, &ni);
2236 	f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
2237 	invalidate_blocks(sbi, ni.blk_addr);
2238 	dec_valid_node_count(sbi, inode, false);
2239 	set_node_addr(sbi, &ni, NULL_ADDR, false);
2240 
2241 recover_xnid:
2242 	/* 2: update xattr nid in inode */
2243 	if (!alloc_nid(sbi, &new_xnid))
2244 		return -ENOSPC;
2245 
2246 	set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2247 	xpage = new_node_page(&dn, XATTR_NODE_OFFSET);
2248 	if (IS_ERR(xpage)) {
2249 		alloc_nid_failed(sbi, new_xnid);
2250 		return PTR_ERR(xpage);
2251 	}
2252 
2253 	alloc_nid_done(sbi, new_xnid);
2254 	update_inode_page(inode);
2255 
2256 	/* 3: update and set xattr node page dirty */
2257 	memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2258 
2259 	set_page_dirty(xpage);
2260 	f2fs_put_page(xpage, 1);
2261 
2262 	return 0;
2263 }
2264 
2265 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2266 {
2267 	struct f2fs_inode *src, *dst;
2268 	nid_t ino = ino_of_node(page);
2269 	struct node_info old_ni, new_ni;
2270 	struct page *ipage;
2271 
2272 	get_node_info(sbi, ino, &old_ni);
2273 
2274 	if (unlikely(old_ni.blk_addr != NULL_ADDR))
2275 		return -EINVAL;
2276 retry:
2277 	ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2278 	if (!ipage) {
2279 		congestion_wait(BLK_RW_ASYNC, HZ/50);
2280 		goto retry;
2281 	}
2282 
2283 	/* Should not use this inode from free nid list */
2284 	remove_free_nid(sbi, ino);
2285 
2286 	if (!PageUptodate(ipage))
2287 		SetPageUptodate(ipage);
2288 	fill_node_footer(ipage, ino, ino, 0, true);
2289 
2290 	src = F2FS_INODE(page);
2291 	dst = F2FS_INODE(ipage);
2292 
2293 	memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2294 	dst->i_size = 0;
2295 	dst->i_blocks = cpu_to_le64(1);
2296 	dst->i_links = cpu_to_le32(1);
2297 	dst->i_xattr_nid = 0;
2298 	dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2299 	if (dst->i_inline & F2FS_EXTRA_ATTR) {
2300 		dst->i_extra_isize = src->i_extra_isize;
2301 
2302 		if (f2fs_sb_has_flexible_inline_xattr(sbi->sb) &&
2303 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2304 							i_inline_xattr_size))
2305 			dst->i_inline_xattr_size = src->i_inline_xattr_size;
2306 
2307 		if (f2fs_sb_has_project_quota(sbi->sb) &&
2308 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2309 								i_projid))
2310 			dst->i_projid = src->i_projid;
2311 	}
2312 
2313 	new_ni = old_ni;
2314 	new_ni.ino = ino;
2315 
2316 	if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2317 		WARN_ON(1);
2318 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2319 	inc_valid_inode_count(sbi);
2320 	set_page_dirty(ipage);
2321 	f2fs_put_page(ipage, 1);
2322 	return 0;
2323 }
2324 
2325 int restore_node_summary(struct f2fs_sb_info *sbi,
2326 			unsigned int segno, struct f2fs_summary_block *sum)
2327 {
2328 	struct f2fs_node *rn;
2329 	struct f2fs_summary *sum_entry;
2330 	block_t addr;
2331 	int i, idx, last_offset, nrpages;
2332 
2333 	/* scan the node segment */
2334 	last_offset = sbi->blocks_per_seg;
2335 	addr = START_BLOCK(sbi, segno);
2336 	sum_entry = &sum->entries[0];
2337 
2338 	for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2339 		nrpages = min(last_offset - i, BIO_MAX_PAGES);
2340 
2341 		/* readahead node pages */
2342 		ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2343 
2344 		for (idx = addr; idx < addr + nrpages; idx++) {
2345 			struct page *page = get_tmp_page(sbi, idx);
2346 
2347 			rn = F2FS_NODE(page);
2348 			sum_entry->nid = rn->footer.nid;
2349 			sum_entry->version = 0;
2350 			sum_entry->ofs_in_node = 0;
2351 			sum_entry++;
2352 			f2fs_put_page(page, 1);
2353 		}
2354 
2355 		invalidate_mapping_pages(META_MAPPING(sbi), addr,
2356 							addr + nrpages);
2357 	}
2358 	return 0;
2359 }
2360 
2361 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2362 {
2363 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2364 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2365 	struct f2fs_journal *journal = curseg->journal;
2366 	int i;
2367 
2368 	down_write(&curseg->journal_rwsem);
2369 	for (i = 0; i < nats_in_cursum(journal); i++) {
2370 		struct nat_entry *ne;
2371 		struct f2fs_nat_entry raw_ne;
2372 		nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2373 
2374 		raw_ne = nat_in_journal(journal, i);
2375 
2376 		ne = __lookup_nat_cache(nm_i, nid);
2377 		if (!ne) {
2378 			ne = __alloc_nat_entry(nid, true);
2379 			__init_nat_entry(nm_i, ne, &raw_ne, true);
2380 		}
2381 
2382 		/*
2383 		 * if a free nat in journal has not been used after last
2384 		 * checkpoint, we should remove it from available nids,
2385 		 * since later we will add it again.
2386 		 */
2387 		if (!get_nat_flag(ne, IS_DIRTY) &&
2388 				le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2389 			spin_lock(&nm_i->nid_list_lock);
2390 			nm_i->available_nids--;
2391 			spin_unlock(&nm_i->nid_list_lock);
2392 		}
2393 
2394 		__set_nat_cache_dirty(nm_i, ne);
2395 	}
2396 	update_nats_in_cursum(journal, -i);
2397 	up_write(&curseg->journal_rwsem);
2398 }
2399 
2400 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2401 						struct list_head *head, int max)
2402 {
2403 	struct nat_entry_set *cur;
2404 
2405 	if (nes->entry_cnt >= max)
2406 		goto add_out;
2407 
2408 	list_for_each_entry(cur, head, set_list) {
2409 		if (cur->entry_cnt >= nes->entry_cnt) {
2410 			list_add(&nes->set_list, cur->set_list.prev);
2411 			return;
2412 		}
2413 	}
2414 add_out:
2415 	list_add_tail(&nes->set_list, head);
2416 }
2417 
2418 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2419 						struct page *page)
2420 {
2421 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2422 	unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2423 	struct f2fs_nat_block *nat_blk = page_address(page);
2424 	int valid = 0;
2425 	int i = 0;
2426 
2427 	if (!enabled_nat_bits(sbi, NULL))
2428 		return;
2429 
2430 	if (nat_index == 0) {
2431 		valid = 1;
2432 		i = 1;
2433 	}
2434 	for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2435 		if (nat_blk->entries[i].block_addr != NULL_ADDR)
2436 			valid++;
2437 	}
2438 	if (valid == 0) {
2439 		__set_bit_le(nat_index, nm_i->empty_nat_bits);
2440 		__clear_bit_le(nat_index, nm_i->full_nat_bits);
2441 		return;
2442 	}
2443 
2444 	__clear_bit_le(nat_index, nm_i->empty_nat_bits);
2445 	if (valid == NAT_ENTRY_PER_BLOCK)
2446 		__set_bit_le(nat_index, nm_i->full_nat_bits);
2447 	else
2448 		__clear_bit_le(nat_index, nm_i->full_nat_bits);
2449 }
2450 
2451 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2452 		struct nat_entry_set *set, struct cp_control *cpc)
2453 {
2454 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2455 	struct f2fs_journal *journal = curseg->journal;
2456 	nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2457 	bool to_journal = true;
2458 	struct f2fs_nat_block *nat_blk;
2459 	struct nat_entry *ne, *cur;
2460 	struct page *page = NULL;
2461 
2462 	/*
2463 	 * there are two steps to flush nat entries:
2464 	 * #1, flush nat entries to journal in current hot data summary block.
2465 	 * #2, flush nat entries to nat page.
2466 	 */
2467 	if (enabled_nat_bits(sbi, cpc) ||
2468 		!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2469 		to_journal = false;
2470 
2471 	if (to_journal) {
2472 		down_write(&curseg->journal_rwsem);
2473 	} else {
2474 		page = get_next_nat_page(sbi, start_nid);
2475 		nat_blk = page_address(page);
2476 		f2fs_bug_on(sbi, !nat_blk);
2477 	}
2478 
2479 	/* flush dirty nats in nat entry set */
2480 	list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2481 		struct f2fs_nat_entry *raw_ne;
2482 		nid_t nid = nat_get_nid(ne);
2483 		int offset;
2484 
2485 		f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2486 
2487 		if (to_journal) {
2488 			offset = lookup_journal_in_cursum(journal,
2489 							NAT_JOURNAL, nid, 1);
2490 			f2fs_bug_on(sbi, offset < 0);
2491 			raw_ne = &nat_in_journal(journal, offset);
2492 			nid_in_journal(journal, offset) = cpu_to_le32(nid);
2493 		} else {
2494 			raw_ne = &nat_blk->entries[nid - start_nid];
2495 		}
2496 		raw_nat_from_node_info(raw_ne, &ne->ni);
2497 		nat_reset_flag(ne);
2498 		__clear_nat_cache_dirty(NM_I(sbi), set, ne);
2499 		if (nat_get_blkaddr(ne) == NULL_ADDR) {
2500 			add_free_nid(sbi, nid, false);
2501 			spin_lock(&NM_I(sbi)->nid_list_lock);
2502 			NM_I(sbi)->available_nids++;
2503 			update_free_nid_bitmap(sbi, nid, true, false);
2504 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2505 		} else {
2506 			spin_lock(&NM_I(sbi)->nid_list_lock);
2507 			update_free_nid_bitmap(sbi, nid, false, false);
2508 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2509 		}
2510 	}
2511 
2512 	if (to_journal) {
2513 		up_write(&curseg->journal_rwsem);
2514 	} else {
2515 		__update_nat_bits(sbi, start_nid, page);
2516 		f2fs_put_page(page, 1);
2517 	}
2518 
2519 	/* Allow dirty nats by node block allocation in write_begin */
2520 	if (!set->entry_cnt) {
2521 		radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2522 		kmem_cache_free(nat_entry_set_slab, set);
2523 	}
2524 }
2525 
2526 /*
2527  * This function is called during the checkpointing process.
2528  */
2529 void flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2530 {
2531 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2532 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2533 	struct f2fs_journal *journal = curseg->journal;
2534 	struct nat_entry_set *setvec[SETVEC_SIZE];
2535 	struct nat_entry_set *set, *tmp;
2536 	unsigned int found;
2537 	nid_t set_idx = 0;
2538 	LIST_HEAD(sets);
2539 
2540 	if (!nm_i->dirty_nat_cnt)
2541 		return;
2542 
2543 	down_write(&nm_i->nat_tree_lock);
2544 
2545 	/*
2546 	 * if there are no enough space in journal to store dirty nat
2547 	 * entries, remove all entries from journal and merge them
2548 	 * into nat entry set.
2549 	 */
2550 	if (enabled_nat_bits(sbi, cpc) ||
2551 		!__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2552 		remove_nats_in_journal(sbi);
2553 
2554 	while ((found = __gang_lookup_nat_set(nm_i,
2555 					set_idx, SETVEC_SIZE, setvec))) {
2556 		unsigned idx;
2557 		set_idx = setvec[found - 1]->set + 1;
2558 		for (idx = 0; idx < found; idx++)
2559 			__adjust_nat_entry_set(setvec[idx], &sets,
2560 						MAX_NAT_JENTRIES(journal));
2561 	}
2562 
2563 	/* flush dirty nats in nat entry set */
2564 	list_for_each_entry_safe(set, tmp, &sets, set_list)
2565 		__flush_nat_entry_set(sbi, set, cpc);
2566 
2567 	up_write(&nm_i->nat_tree_lock);
2568 	/* Allow dirty nats by node block allocation in write_begin */
2569 }
2570 
2571 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
2572 {
2573 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
2574 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2575 	unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
2576 	unsigned int i;
2577 	__u64 cp_ver = cur_cp_version(ckpt);
2578 	block_t nat_bits_addr;
2579 
2580 	if (!enabled_nat_bits(sbi, NULL))
2581 		return 0;
2582 
2583 	nm_i->nat_bits_blocks = F2FS_BYTES_TO_BLK((nat_bits_bytes << 1) + 8 +
2584 						F2FS_BLKSIZE - 1);
2585 	nm_i->nat_bits = kzalloc(nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS,
2586 						GFP_KERNEL);
2587 	if (!nm_i->nat_bits)
2588 		return -ENOMEM;
2589 
2590 	nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
2591 						nm_i->nat_bits_blocks;
2592 	for (i = 0; i < nm_i->nat_bits_blocks; i++) {
2593 		struct page *page = get_meta_page(sbi, nat_bits_addr++);
2594 
2595 		memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
2596 					page_address(page), F2FS_BLKSIZE);
2597 		f2fs_put_page(page, 1);
2598 	}
2599 
2600 	cp_ver |= (cur_cp_crc(ckpt) << 32);
2601 	if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
2602 		disable_nat_bits(sbi, true);
2603 		return 0;
2604 	}
2605 
2606 	nm_i->full_nat_bits = nm_i->nat_bits + 8;
2607 	nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
2608 
2609 	f2fs_msg(sbi->sb, KERN_NOTICE, "Found nat_bits in checkpoint");
2610 	return 0;
2611 }
2612 
2613 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
2614 {
2615 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2616 	unsigned int i = 0;
2617 	nid_t nid, last_nid;
2618 
2619 	if (!enabled_nat_bits(sbi, NULL))
2620 		return;
2621 
2622 	for (i = 0; i < nm_i->nat_blocks; i++) {
2623 		i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
2624 		if (i >= nm_i->nat_blocks)
2625 			break;
2626 
2627 		__set_bit_le(i, nm_i->nat_block_bitmap);
2628 
2629 		nid = i * NAT_ENTRY_PER_BLOCK;
2630 		last_nid = nid + NAT_ENTRY_PER_BLOCK;
2631 
2632 		spin_lock(&NM_I(sbi)->nid_list_lock);
2633 		for (; nid < last_nid; nid++)
2634 			update_free_nid_bitmap(sbi, nid, true, true);
2635 		spin_unlock(&NM_I(sbi)->nid_list_lock);
2636 	}
2637 
2638 	for (i = 0; i < nm_i->nat_blocks; i++) {
2639 		i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
2640 		if (i >= nm_i->nat_blocks)
2641 			break;
2642 
2643 		__set_bit_le(i, nm_i->nat_block_bitmap);
2644 	}
2645 }
2646 
2647 static int init_node_manager(struct f2fs_sb_info *sbi)
2648 {
2649 	struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2650 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2651 	unsigned char *version_bitmap;
2652 	unsigned int nat_segs;
2653 	int err;
2654 
2655 	nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2656 
2657 	/* segment_count_nat includes pair segment so divide to 2. */
2658 	nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2659 	nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2660 	nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
2661 
2662 	/* not used nids: 0, node, meta, (and root counted as valid node) */
2663 	nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
2664 							F2FS_RESERVED_NODE_NUM;
2665 	nm_i->nid_cnt[FREE_NID] = 0;
2666 	nm_i->nid_cnt[PREALLOC_NID] = 0;
2667 	nm_i->nat_cnt = 0;
2668 	nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2669 	nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2670 	nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2671 
2672 	INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2673 	INIT_LIST_HEAD(&nm_i->free_nid_list);
2674 	INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2675 	INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2676 	INIT_LIST_HEAD(&nm_i->nat_entries);
2677 
2678 	mutex_init(&nm_i->build_lock);
2679 	spin_lock_init(&nm_i->nid_list_lock);
2680 	init_rwsem(&nm_i->nat_tree_lock);
2681 
2682 	nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2683 	nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2684 	version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2685 	if (!version_bitmap)
2686 		return -EFAULT;
2687 
2688 	nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2689 					GFP_KERNEL);
2690 	if (!nm_i->nat_bitmap)
2691 		return -ENOMEM;
2692 
2693 	err = __get_nat_bitmaps(sbi);
2694 	if (err)
2695 		return err;
2696 
2697 #ifdef CONFIG_F2FS_CHECK_FS
2698 	nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
2699 					GFP_KERNEL);
2700 	if (!nm_i->nat_bitmap_mir)
2701 		return -ENOMEM;
2702 #endif
2703 
2704 	return 0;
2705 }
2706 
2707 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
2708 {
2709 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2710 
2711 	nm_i->free_nid_bitmap = kvzalloc(nm_i->nat_blocks *
2712 					NAT_ENTRY_BITMAP_SIZE, GFP_KERNEL);
2713 	if (!nm_i->free_nid_bitmap)
2714 		return -ENOMEM;
2715 
2716 	nm_i->nat_block_bitmap = kvzalloc(nm_i->nat_blocks / 8,
2717 								GFP_KERNEL);
2718 	if (!nm_i->nat_block_bitmap)
2719 		return -ENOMEM;
2720 
2721 	nm_i->free_nid_count = kvzalloc(nm_i->nat_blocks *
2722 					sizeof(unsigned short), GFP_KERNEL);
2723 	if (!nm_i->free_nid_count)
2724 		return -ENOMEM;
2725 	return 0;
2726 }
2727 
2728 int build_node_manager(struct f2fs_sb_info *sbi)
2729 {
2730 	int err;
2731 
2732 	sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2733 	if (!sbi->nm_info)
2734 		return -ENOMEM;
2735 
2736 	err = init_node_manager(sbi);
2737 	if (err)
2738 		return err;
2739 
2740 	err = init_free_nid_cache(sbi);
2741 	if (err)
2742 		return err;
2743 
2744 	/* load free nid status from nat_bits table */
2745 	load_free_nid_bitmap(sbi);
2746 
2747 	build_free_nids(sbi, true, true);
2748 	return 0;
2749 }
2750 
2751 void destroy_node_manager(struct f2fs_sb_info *sbi)
2752 {
2753 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2754 	struct free_nid *i, *next_i;
2755 	struct nat_entry *natvec[NATVEC_SIZE];
2756 	struct nat_entry_set *setvec[SETVEC_SIZE];
2757 	nid_t nid = 0;
2758 	unsigned int found;
2759 
2760 	if (!nm_i)
2761 		return;
2762 
2763 	/* destroy free nid list */
2764 	spin_lock(&nm_i->nid_list_lock);
2765 	list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2766 		__remove_free_nid(sbi, i, FREE_NID);
2767 		spin_unlock(&nm_i->nid_list_lock);
2768 		kmem_cache_free(free_nid_slab, i);
2769 		spin_lock(&nm_i->nid_list_lock);
2770 	}
2771 	f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
2772 	f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
2773 	f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
2774 	spin_unlock(&nm_i->nid_list_lock);
2775 
2776 	/* destroy nat cache */
2777 	down_write(&nm_i->nat_tree_lock);
2778 	while ((found = __gang_lookup_nat_cache(nm_i,
2779 					nid, NATVEC_SIZE, natvec))) {
2780 		unsigned idx;
2781 
2782 		nid = nat_get_nid(natvec[found - 1]) + 1;
2783 		for (idx = 0; idx < found; idx++)
2784 			__del_from_nat_cache(nm_i, natvec[idx]);
2785 	}
2786 	f2fs_bug_on(sbi, nm_i->nat_cnt);
2787 
2788 	/* destroy nat set cache */
2789 	nid = 0;
2790 	while ((found = __gang_lookup_nat_set(nm_i,
2791 					nid, SETVEC_SIZE, setvec))) {
2792 		unsigned idx;
2793 
2794 		nid = setvec[found - 1]->set + 1;
2795 		for (idx = 0; idx < found; idx++) {
2796 			/* entry_cnt is not zero, when cp_error was occurred */
2797 			f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2798 			radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2799 			kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2800 		}
2801 	}
2802 	up_write(&nm_i->nat_tree_lock);
2803 
2804 	kvfree(nm_i->nat_block_bitmap);
2805 	kvfree(nm_i->free_nid_bitmap);
2806 	kvfree(nm_i->free_nid_count);
2807 
2808 	kfree(nm_i->nat_bitmap);
2809 	kfree(nm_i->nat_bits);
2810 #ifdef CONFIG_F2FS_CHECK_FS
2811 	kfree(nm_i->nat_bitmap_mir);
2812 #endif
2813 	sbi->nm_info = NULL;
2814 	kfree(nm_i);
2815 }
2816 
2817 int __init create_node_manager_caches(void)
2818 {
2819 	nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2820 			sizeof(struct nat_entry));
2821 	if (!nat_entry_slab)
2822 		goto fail;
2823 
2824 	free_nid_slab = f2fs_kmem_cache_create("free_nid",
2825 			sizeof(struct free_nid));
2826 	if (!free_nid_slab)
2827 		goto destroy_nat_entry;
2828 
2829 	nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2830 			sizeof(struct nat_entry_set));
2831 	if (!nat_entry_set_slab)
2832 		goto destroy_free_nid;
2833 	return 0;
2834 
2835 destroy_free_nid:
2836 	kmem_cache_destroy(free_nid_slab);
2837 destroy_nat_entry:
2838 	kmem_cache_destroy(nat_entry_slab);
2839 fail:
2840 	return -ENOMEM;
2841 }
2842 
2843 void destroy_node_manager_caches(void)
2844 {
2845 	kmem_cache_destroy(nat_entry_set_slab);
2846 	kmem_cache_destroy(free_nid_slab);
2847 	kmem_cache_destroy(nat_entry_slab);
2848 }
2849