xref: /openbmc/linux/fs/f2fs/node.c (revision 70246286e94c335b5bea0cbc68a17a96dd620281)
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_SHIFT;
50 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
51 	} else if (type == NAT_ENTRIES) {
52 		mem_size = (nm_i->nat_cnt * sizeof(struct nat_entry)) >>
53 							PAGE_SHIFT;
54 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
55 	} 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_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_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_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_sb_info *sbi, nid_t nid,
261 						struct f2fs_nat_entry *ne)
262 {
263 	struct f2fs_nm_info *nm_i = NM_I(sbi);
264 	struct nat_entry *e;
265 
266 	e = __lookup_nat_cache(nm_i, nid);
267 	if (!e) {
268 		e = grab_nat_entry(nm_i, nid);
269 		node_info_from_raw_nat(&e->ni, ne);
270 	} else {
271 		f2fs_bug_on(sbi, nat_get_ino(e) != ne->ino ||
272 				nat_get_blkaddr(e) != ne->block_addr ||
273 				nat_get_version(e) != ne->version);
274 	}
275 }
276 
277 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
278 			block_t new_blkaddr, bool fsync_done)
279 {
280 	struct f2fs_nm_info *nm_i = NM_I(sbi);
281 	struct nat_entry *e;
282 
283 	down_write(&nm_i->nat_tree_lock);
284 	e = __lookup_nat_cache(nm_i, ni->nid);
285 	if (!e) {
286 		e = grab_nat_entry(nm_i, ni->nid);
287 		copy_node_info(&e->ni, ni);
288 		f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
289 	} else if (new_blkaddr == NEW_ADDR) {
290 		/*
291 		 * when nid is reallocated,
292 		 * previous nat entry can be remained in nat cache.
293 		 * So, reinitialize it with new information.
294 		 */
295 		copy_node_info(&e->ni, ni);
296 		f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
297 	}
298 
299 	/* sanity check */
300 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
301 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
302 			new_blkaddr == NULL_ADDR);
303 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
304 			new_blkaddr == NEW_ADDR);
305 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
306 			nat_get_blkaddr(e) != NULL_ADDR &&
307 			new_blkaddr == NEW_ADDR);
308 
309 	/* increment version no as node is removed */
310 	if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
311 		unsigned char version = nat_get_version(e);
312 		nat_set_version(e, inc_node_version(version));
313 
314 		/* in order to reuse the nid */
315 		if (nm_i->next_scan_nid > ni->nid)
316 			nm_i->next_scan_nid = ni->nid;
317 	}
318 
319 	/* change address */
320 	nat_set_blkaddr(e, new_blkaddr);
321 	if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
322 		set_nat_flag(e, IS_CHECKPOINTED, false);
323 	__set_nat_cache_dirty(nm_i, e);
324 
325 	/* update fsync_mark if its inode nat entry is still alive */
326 	if (ni->nid != ni->ino)
327 		e = __lookup_nat_cache(nm_i, ni->ino);
328 	if (e) {
329 		if (fsync_done && ni->nid == ni->ino)
330 			set_nat_flag(e, HAS_FSYNCED_INODE, true);
331 		set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
332 	}
333 	up_write(&nm_i->nat_tree_lock);
334 }
335 
336 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
337 {
338 	struct f2fs_nm_info *nm_i = NM_I(sbi);
339 	int nr = nr_shrink;
340 
341 	if (!down_write_trylock(&nm_i->nat_tree_lock))
342 		return 0;
343 
344 	while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
345 		struct nat_entry *ne;
346 		ne = list_first_entry(&nm_i->nat_entries,
347 					struct nat_entry, list);
348 		__del_from_nat_cache(nm_i, ne);
349 		nr_shrink--;
350 	}
351 	up_write(&nm_i->nat_tree_lock);
352 	return nr - nr_shrink;
353 }
354 
355 /*
356  * This function always returns success
357  */
358 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
359 {
360 	struct f2fs_nm_info *nm_i = NM_I(sbi);
361 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
362 	struct f2fs_journal *journal = curseg->journal;
363 	nid_t start_nid = START_NID(nid);
364 	struct f2fs_nat_block *nat_blk;
365 	struct page *page = NULL;
366 	struct f2fs_nat_entry ne;
367 	struct nat_entry *e;
368 	int i;
369 
370 	ni->nid = nid;
371 
372 	/* Check nat cache */
373 	down_read(&nm_i->nat_tree_lock);
374 	e = __lookup_nat_cache(nm_i, nid);
375 	if (e) {
376 		ni->ino = nat_get_ino(e);
377 		ni->blk_addr = nat_get_blkaddr(e);
378 		ni->version = nat_get_version(e);
379 		up_read(&nm_i->nat_tree_lock);
380 		return;
381 	}
382 
383 	memset(&ne, 0, sizeof(struct f2fs_nat_entry));
384 
385 	/* Check current segment summary */
386 	down_read(&curseg->journal_rwsem);
387 	i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
388 	if (i >= 0) {
389 		ne = nat_in_journal(journal, i);
390 		node_info_from_raw_nat(ni, &ne);
391 	}
392 	up_read(&curseg->journal_rwsem);
393 	if (i >= 0)
394 		goto cache;
395 
396 	/* Fill node_info from nat page */
397 	page = get_current_nat_page(sbi, start_nid);
398 	nat_blk = (struct f2fs_nat_block *)page_address(page);
399 	ne = nat_blk->entries[nid - start_nid];
400 	node_info_from_raw_nat(ni, &ne);
401 	f2fs_put_page(page, 1);
402 cache:
403 	up_read(&nm_i->nat_tree_lock);
404 	/* cache nat entry */
405 	down_write(&nm_i->nat_tree_lock);
406 	cache_nat_entry(sbi, nid, &ne);
407 	up_write(&nm_i->nat_tree_lock);
408 }
409 
410 /*
411  * readahead MAX_RA_NODE number of node pages.
412  */
413 static void ra_node_pages(struct page *parent, int start, int n)
414 {
415 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
416 	struct blk_plug plug;
417 	int i, end;
418 	nid_t nid;
419 
420 	blk_start_plug(&plug);
421 
422 	/* Then, try readahead for siblings of the desired node */
423 	end = start + n;
424 	end = min(end, NIDS_PER_BLOCK);
425 	for (i = start; i < end; i++) {
426 		nid = get_nid(parent, i, false);
427 		ra_node_page(sbi, nid);
428 	}
429 
430 	blk_finish_plug(&plug);
431 }
432 
433 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
434 {
435 	const long direct_index = ADDRS_PER_INODE(dn->inode);
436 	const long direct_blks = ADDRS_PER_BLOCK;
437 	const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
438 	unsigned int skipped_unit = ADDRS_PER_BLOCK;
439 	int cur_level = dn->cur_level;
440 	int max_level = dn->max_level;
441 	pgoff_t base = 0;
442 
443 	if (!dn->max_level)
444 		return pgofs + 1;
445 
446 	while (max_level-- > cur_level)
447 		skipped_unit *= NIDS_PER_BLOCK;
448 
449 	switch (dn->max_level) {
450 	case 3:
451 		base += 2 * indirect_blks;
452 	case 2:
453 		base += 2 * direct_blks;
454 	case 1:
455 		base += direct_index;
456 		break;
457 	default:
458 		f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
459 	}
460 
461 	return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
462 }
463 
464 /*
465  * The maximum depth is four.
466  * Offset[0] will have raw inode offset.
467  */
468 static int get_node_path(struct inode *inode, long block,
469 				int offset[4], unsigned int noffset[4])
470 {
471 	const long direct_index = ADDRS_PER_INODE(inode);
472 	const long direct_blks = ADDRS_PER_BLOCK;
473 	const long dptrs_per_blk = NIDS_PER_BLOCK;
474 	const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
475 	const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
476 	int n = 0;
477 	int level = 0;
478 
479 	noffset[0] = 0;
480 
481 	if (block < direct_index) {
482 		offset[n] = block;
483 		goto got;
484 	}
485 	block -= direct_index;
486 	if (block < direct_blks) {
487 		offset[n++] = NODE_DIR1_BLOCK;
488 		noffset[n] = 1;
489 		offset[n] = block;
490 		level = 1;
491 		goto got;
492 	}
493 	block -= direct_blks;
494 	if (block < direct_blks) {
495 		offset[n++] = NODE_DIR2_BLOCK;
496 		noffset[n] = 2;
497 		offset[n] = block;
498 		level = 1;
499 		goto got;
500 	}
501 	block -= direct_blks;
502 	if (block < indirect_blks) {
503 		offset[n++] = NODE_IND1_BLOCK;
504 		noffset[n] = 3;
505 		offset[n++] = block / direct_blks;
506 		noffset[n] = 4 + offset[n - 1];
507 		offset[n] = block % direct_blks;
508 		level = 2;
509 		goto got;
510 	}
511 	block -= indirect_blks;
512 	if (block < indirect_blks) {
513 		offset[n++] = NODE_IND2_BLOCK;
514 		noffset[n] = 4 + dptrs_per_blk;
515 		offset[n++] = block / direct_blks;
516 		noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
517 		offset[n] = block % direct_blks;
518 		level = 2;
519 		goto got;
520 	}
521 	block -= indirect_blks;
522 	if (block < dindirect_blks) {
523 		offset[n++] = NODE_DIND_BLOCK;
524 		noffset[n] = 5 + (dptrs_per_blk * 2);
525 		offset[n++] = block / indirect_blks;
526 		noffset[n] = 6 + (dptrs_per_blk * 2) +
527 			      offset[n - 1] * (dptrs_per_blk + 1);
528 		offset[n++] = (block / direct_blks) % dptrs_per_blk;
529 		noffset[n] = 7 + (dptrs_per_blk * 2) +
530 			      offset[n - 2] * (dptrs_per_blk + 1) +
531 			      offset[n - 1];
532 		offset[n] = block % direct_blks;
533 		level = 3;
534 		goto got;
535 	} else {
536 		BUG();
537 	}
538 got:
539 	return level;
540 }
541 
542 /*
543  * Caller should call f2fs_put_dnode(dn).
544  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
545  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
546  * In the case of RDONLY_NODE, we don't need to care about mutex.
547  */
548 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
549 {
550 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
551 	struct page *npage[4];
552 	struct page *parent = NULL;
553 	int offset[4];
554 	unsigned int noffset[4];
555 	nid_t nids[4];
556 	int level, i = 0;
557 	int err = 0;
558 
559 	level = get_node_path(dn->inode, index, offset, noffset);
560 
561 	nids[0] = dn->inode->i_ino;
562 	npage[0] = dn->inode_page;
563 
564 	if (!npage[0]) {
565 		npage[0] = get_node_page(sbi, nids[0]);
566 		if (IS_ERR(npage[0]))
567 			return PTR_ERR(npage[0]);
568 	}
569 
570 	/* if inline_data is set, should not report any block indices */
571 	if (f2fs_has_inline_data(dn->inode) && index) {
572 		err = -ENOENT;
573 		f2fs_put_page(npage[0], 1);
574 		goto release_out;
575 	}
576 
577 	parent = npage[0];
578 	if (level != 0)
579 		nids[1] = get_nid(parent, offset[0], true);
580 	dn->inode_page = npage[0];
581 	dn->inode_page_locked = true;
582 
583 	/* get indirect or direct nodes */
584 	for (i = 1; i <= level; i++) {
585 		bool done = false;
586 
587 		if (!nids[i] && mode == ALLOC_NODE) {
588 			/* alloc new node */
589 			if (!alloc_nid(sbi, &(nids[i]))) {
590 				err = -ENOSPC;
591 				goto release_pages;
592 			}
593 
594 			dn->nid = nids[i];
595 			npage[i] = new_node_page(dn, noffset[i], NULL);
596 			if (IS_ERR(npage[i])) {
597 				alloc_nid_failed(sbi, nids[i]);
598 				err = PTR_ERR(npage[i]);
599 				goto release_pages;
600 			}
601 
602 			set_nid(parent, offset[i - 1], nids[i], i == 1);
603 			alloc_nid_done(sbi, nids[i]);
604 			done = true;
605 		} else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
606 			npage[i] = get_node_page_ra(parent, offset[i - 1]);
607 			if (IS_ERR(npage[i])) {
608 				err = PTR_ERR(npage[i]);
609 				goto release_pages;
610 			}
611 			done = true;
612 		}
613 		if (i == 1) {
614 			dn->inode_page_locked = false;
615 			unlock_page(parent);
616 		} else {
617 			f2fs_put_page(parent, 1);
618 		}
619 
620 		if (!done) {
621 			npage[i] = get_node_page(sbi, nids[i]);
622 			if (IS_ERR(npage[i])) {
623 				err = PTR_ERR(npage[i]);
624 				f2fs_put_page(npage[0], 0);
625 				goto release_out;
626 			}
627 		}
628 		if (i < level) {
629 			parent = npage[i];
630 			nids[i + 1] = get_nid(parent, offset[i], false);
631 		}
632 	}
633 	dn->nid = nids[level];
634 	dn->ofs_in_node = offset[level];
635 	dn->node_page = npage[level];
636 	dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
637 	return 0;
638 
639 release_pages:
640 	f2fs_put_page(parent, 1);
641 	if (i > 1)
642 		f2fs_put_page(npage[0], 0);
643 release_out:
644 	dn->inode_page = NULL;
645 	dn->node_page = NULL;
646 	if (err == -ENOENT) {
647 		dn->cur_level = i;
648 		dn->max_level = level;
649 	}
650 	return err;
651 }
652 
653 static void truncate_node(struct dnode_of_data *dn)
654 {
655 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
656 	struct node_info ni;
657 
658 	get_node_info(sbi, dn->nid, &ni);
659 	if (dn->inode->i_blocks == 0) {
660 		f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
661 		goto invalidate;
662 	}
663 	f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
664 
665 	/* Deallocate node address */
666 	invalidate_blocks(sbi, ni.blk_addr);
667 	dec_valid_node_count(sbi, dn->inode);
668 	set_node_addr(sbi, &ni, NULL_ADDR, false);
669 
670 	if (dn->nid == dn->inode->i_ino) {
671 		remove_orphan_inode(sbi, dn->nid);
672 		dec_valid_inode_count(sbi);
673 	} else {
674 		sync_inode_page(dn);
675 	}
676 invalidate:
677 	clear_node_page_dirty(dn->node_page);
678 	set_sbi_flag(sbi, SBI_IS_DIRTY);
679 
680 	f2fs_put_page(dn->node_page, 1);
681 
682 	invalidate_mapping_pages(NODE_MAPPING(sbi),
683 			dn->node_page->index, dn->node_page->index);
684 
685 	dn->node_page = NULL;
686 	trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
687 }
688 
689 static int truncate_dnode(struct dnode_of_data *dn)
690 {
691 	struct page *page;
692 
693 	if (dn->nid == 0)
694 		return 1;
695 
696 	/* get direct node */
697 	page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
698 	if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
699 		return 1;
700 	else if (IS_ERR(page))
701 		return PTR_ERR(page);
702 
703 	/* Make dnode_of_data for parameter */
704 	dn->node_page = page;
705 	dn->ofs_in_node = 0;
706 	truncate_data_blocks(dn);
707 	truncate_node(dn);
708 	return 1;
709 }
710 
711 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
712 						int ofs, int depth)
713 {
714 	struct dnode_of_data rdn = *dn;
715 	struct page *page;
716 	struct f2fs_node *rn;
717 	nid_t child_nid;
718 	unsigned int child_nofs;
719 	int freed = 0;
720 	int i, ret;
721 
722 	if (dn->nid == 0)
723 		return NIDS_PER_BLOCK + 1;
724 
725 	trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
726 
727 	page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
728 	if (IS_ERR(page)) {
729 		trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
730 		return PTR_ERR(page);
731 	}
732 
733 	ra_node_pages(page, ofs, NIDS_PER_BLOCK);
734 
735 	rn = F2FS_NODE(page);
736 	if (depth < 3) {
737 		for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
738 			child_nid = le32_to_cpu(rn->in.nid[i]);
739 			if (child_nid == 0)
740 				continue;
741 			rdn.nid = child_nid;
742 			ret = truncate_dnode(&rdn);
743 			if (ret < 0)
744 				goto out_err;
745 			if (set_nid(page, i, 0, false))
746 				dn->node_changed = true;
747 		}
748 	} else {
749 		child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
750 		for (i = ofs; i < NIDS_PER_BLOCK; i++) {
751 			child_nid = le32_to_cpu(rn->in.nid[i]);
752 			if (child_nid == 0) {
753 				child_nofs += NIDS_PER_BLOCK + 1;
754 				continue;
755 			}
756 			rdn.nid = child_nid;
757 			ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
758 			if (ret == (NIDS_PER_BLOCK + 1)) {
759 				if (set_nid(page, i, 0, false))
760 					dn->node_changed = true;
761 				child_nofs += ret;
762 			} else if (ret < 0 && ret != -ENOENT) {
763 				goto out_err;
764 			}
765 		}
766 		freed = child_nofs;
767 	}
768 
769 	if (!ofs) {
770 		/* remove current indirect node */
771 		dn->node_page = page;
772 		truncate_node(dn);
773 		freed++;
774 	} else {
775 		f2fs_put_page(page, 1);
776 	}
777 	trace_f2fs_truncate_nodes_exit(dn->inode, freed);
778 	return freed;
779 
780 out_err:
781 	f2fs_put_page(page, 1);
782 	trace_f2fs_truncate_nodes_exit(dn->inode, ret);
783 	return ret;
784 }
785 
786 static int truncate_partial_nodes(struct dnode_of_data *dn,
787 			struct f2fs_inode *ri, int *offset, int depth)
788 {
789 	struct page *pages[2];
790 	nid_t nid[3];
791 	nid_t child_nid;
792 	int err = 0;
793 	int i;
794 	int idx = depth - 2;
795 
796 	nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
797 	if (!nid[0])
798 		return 0;
799 
800 	/* get indirect nodes in the path */
801 	for (i = 0; i < idx + 1; i++) {
802 		/* reference count'll be increased */
803 		pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
804 		if (IS_ERR(pages[i])) {
805 			err = PTR_ERR(pages[i]);
806 			idx = i - 1;
807 			goto fail;
808 		}
809 		nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
810 	}
811 
812 	ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
813 
814 	/* free direct nodes linked to a partial indirect node */
815 	for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
816 		child_nid = get_nid(pages[idx], i, false);
817 		if (!child_nid)
818 			continue;
819 		dn->nid = child_nid;
820 		err = truncate_dnode(dn);
821 		if (err < 0)
822 			goto fail;
823 		if (set_nid(pages[idx], i, 0, false))
824 			dn->node_changed = true;
825 	}
826 
827 	if (offset[idx + 1] == 0) {
828 		dn->node_page = pages[idx];
829 		dn->nid = nid[idx];
830 		truncate_node(dn);
831 	} else {
832 		f2fs_put_page(pages[idx], 1);
833 	}
834 	offset[idx]++;
835 	offset[idx + 1] = 0;
836 	idx--;
837 fail:
838 	for (i = idx; i >= 0; i--)
839 		f2fs_put_page(pages[i], 1);
840 
841 	trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
842 
843 	return err;
844 }
845 
846 /*
847  * All the block addresses of data and nodes should be nullified.
848  */
849 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
850 {
851 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
852 	int err = 0, cont = 1;
853 	int level, offset[4], noffset[4];
854 	unsigned int nofs = 0;
855 	struct f2fs_inode *ri;
856 	struct dnode_of_data dn;
857 	struct page *page;
858 
859 	trace_f2fs_truncate_inode_blocks_enter(inode, from);
860 
861 	level = get_node_path(inode, from, offset, noffset);
862 
863 	page = get_node_page(sbi, inode->i_ino);
864 	if (IS_ERR(page)) {
865 		trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
866 		return PTR_ERR(page);
867 	}
868 
869 	set_new_dnode(&dn, inode, page, NULL, 0);
870 	unlock_page(page);
871 
872 	ri = F2FS_INODE(page);
873 	switch (level) {
874 	case 0:
875 	case 1:
876 		nofs = noffset[1];
877 		break;
878 	case 2:
879 		nofs = noffset[1];
880 		if (!offset[level - 1])
881 			goto skip_partial;
882 		err = truncate_partial_nodes(&dn, ri, offset, level);
883 		if (err < 0 && err != -ENOENT)
884 			goto fail;
885 		nofs += 1 + NIDS_PER_BLOCK;
886 		break;
887 	case 3:
888 		nofs = 5 + 2 * NIDS_PER_BLOCK;
889 		if (!offset[level - 1])
890 			goto skip_partial;
891 		err = truncate_partial_nodes(&dn, ri, offset, level);
892 		if (err < 0 && err != -ENOENT)
893 			goto fail;
894 		break;
895 	default:
896 		BUG();
897 	}
898 
899 skip_partial:
900 	while (cont) {
901 		dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
902 		switch (offset[0]) {
903 		case NODE_DIR1_BLOCK:
904 		case NODE_DIR2_BLOCK:
905 			err = truncate_dnode(&dn);
906 			break;
907 
908 		case NODE_IND1_BLOCK:
909 		case NODE_IND2_BLOCK:
910 			err = truncate_nodes(&dn, nofs, offset[1], 2);
911 			break;
912 
913 		case NODE_DIND_BLOCK:
914 			err = truncate_nodes(&dn, nofs, offset[1], 3);
915 			cont = 0;
916 			break;
917 
918 		default:
919 			BUG();
920 		}
921 		if (err < 0 && err != -ENOENT)
922 			goto fail;
923 		if (offset[1] == 0 &&
924 				ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
925 			lock_page(page);
926 			BUG_ON(page->mapping != NODE_MAPPING(sbi));
927 			f2fs_wait_on_page_writeback(page, NODE, true);
928 			ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
929 			set_page_dirty(page);
930 			unlock_page(page);
931 		}
932 		offset[1] = 0;
933 		offset[0]++;
934 		nofs += err;
935 	}
936 fail:
937 	f2fs_put_page(page, 0);
938 	trace_f2fs_truncate_inode_blocks_exit(inode, err);
939 	return err > 0 ? 0 : err;
940 }
941 
942 int truncate_xattr_node(struct inode *inode, struct page *page)
943 {
944 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
945 	nid_t nid = F2FS_I(inode)->i_xattr_nid;
946 	struct dnode_of_data dn;
947 	struct page *npage;
948 
949 	if (!nid)
950 		return 0;
951 
952 	npage = get_node_page(sbi, nid);
953 	if (IS_ERR(npage))
954 		return PTR_ERR(npage);
955 
956 	F2FS_I(inode)->i_xattr_nid = 0;
957 
958 	/* need to do checkpoint during fsync */
959 	F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
960 
961 	set_new_dnode(&dn, inode, page, npage, nid);
962 
963 	if (page)
964 		dn.inode_page_locked = true;
965 	truncate_node(&dn);
966 	return 0;
967 }
968 
969 /*
970  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
971  * f2fs_unlock_op().
972  */
973 int remove_inode_page(struct inode *inode)
974 {
975 	struct dnode_of_data dn;
976 	int err;
977 
978 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
979 	err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
980 	if (err)
981 		return err;
982 
983 	err = truncate_xattr_node(inode, dn.inode_page);
984 	if (err) {
985 		f2fs_put_dnode(&dn);
986 		return err;
987 	}
988 
989 	/* remove potential inline_data blocks */
990 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
991 				S_ISLNK(inode->i_mode))
992 		truncate_data_blocks_range(&dn, 1);
993 
994 	/* 0 is possible, after f2fs_new_inode() has failed */
995 	f2fs_bug_on(F2FS_I_SB(inode),
996 			inode->i_blocks != 0 && inode->i_blocks != 1);
997 
998 	/* will put inode & node pages */
999 	truncate_node(&dn);
1000 	return 0;
1001 }
1002 
1003 struct page *new_inode_page(struct inode *inode)
1004 {
1005 	struct dnode_of_data dn;
1006 
1007 	/* allocate inode page for new inode */
1008 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1009 
1010 	/* caller should f2fs_put_page(page, 1); */
1011 	return new_node_page(&dn, 0, NULL);
1012 }
1013 
1014 struct page *new_node_page(struct dnode_of_data *dn,
1015 				unsigned int ofs, struct page *ipage)
1016 {
1017 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1018 	struct node_info old_ni, new_ni;
1019 	struct page *page;
1020 	int err;
1021 
1022 	if (unlikely(is_inode_flag_set(F2FS_I(dn->inode), FI_NO_ALLOC)))
1023 		return ERR_PTR(-EPERM);
1024 
1025 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1026 	if (!page)
1027 		return ERR_PTR(-ENOMEM);
1028 
1029 	if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1030 		err = -ENOSPC;
1031 		goto fail;
1032 	}
1033 
1034 	get_node_info(sbi, dn->nid, &old_ni);
1035 
1036 	/* Reinitialize old_ni with new node page */
1037 	f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
1038 	new_ni = old_ni;
1039 	new_ni.ino = dn->inode->i_ino;
1040 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1041 
1042 	f2fs_wait_on_page_writeback(page, NODE, true);
1043 	fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1044 	set_cold_node(dn->inode, page);
1045 	SetPageUptodate(page);
1046 	if (set_page_dirty(page))
1047 		dn->node_changed = true;
1048 
1049 	if (f2fs_has_xattr_block(ofs))
1050 		F2FS_I(dn->inode)->i_xattr_nid = dn->nid;
1051 
1052 	dn->node_page = page;
1053 	if (ipage)
1054 		update_inode(dn->inode, ipage);
1055 	else
1056 		sync_inode_page(dn);
1057 	if (ofs == 0)
1058 		inc_valid_inode_count(sbi);
1059 
1060 	return page;
1061 
1062 fail:
1063 	clear_node_page_dirty(page);
1064 	f2fs_put_page(page, 1);
1065 	return ERR_PTR(err);
1066 }
1067 
1068 /*
1069  * Caller should do after getting the following values.
1070  * 0: f2fs_put_page(page, 0)
1071  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1072  */
1073 static int read_node_page(struct page *page, int op_flags)
1074 {
1075 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1076 	struct node_info ni;
1077 	struct f2fs_io_info fio = {
1078 		.sbi = sbi,
1079 		.type = NODE,
1080 		.op = REQ_OP_READ,
1081 		.op_flags = op_flags,
1082 		.page = page,
1083 		.encrypted_page = NULL,
1084 	};
1085 
1086 	get_node_info(sbi, page->index, &ni);
1087 
1088 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1089 		ClearPageUptodate(page);
1090 		return -ENOENT;
1091 	}
1092 
1093 	if (PageUptodate(page))
1094 		return LOCKED_PAGE;
1095 
1096 	fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1097 	return f2fs_submit_page_bio(&fio);
1098 }
1099 
1100 /*
1101  * Readahead a node page
1102  */
1103 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1104 {
1105 	struct page *apage;
1106 	int err;
1107 
1108 	if (!nid)
1109 		return;
1110 	f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1111 
1112 	rcu_read_lock();
1113 	apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1114 	rcu_read_unlock();
1115 	if (apage)
1116 		return;
1117 
1118 	apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1119 	if (!apage)
1120 		return;
1121 
1122 	err = read_node_page(apage, REQ_RAHEAD);
1123 	f2fs_put_page(apage, err ? 1 : 0);
1124 }
1125 
1126 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1127 					struct page *parent, int start)
1128 {
1129 	struct page *page;
1130 	int err;
1131 
1132 	if (!nid)
1133 		return ERR_PTR(-ENOENT);
1134 	f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1135 repeat:
1136 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1137 	if (!page)
1138 		return ERR_PTR(-ENOMEM);
1139 
1140 	err = read_node_page(page, READ_SYNC);
1141 	if (err < 0) {
1142 		f2fs_put_page(page, 1);
1143 		return ERR_PTR(err);
1144 	} else if (err == LOCKED_PAGE) {
1145 		goto page_hit;
1146 	}
1147 
1148 	if (parent)
1149 		ra_node_pages(parent, start + 1, MAX_RA_NODE);
1150 
1151 	lock_page(page);
1152 
1153 	if (unlikely(!PageUptodate(page))) {
1154 		f2fs_put_page(page, 1);
1155 		return ERR_PTR(-EIO);
1156 	}
1157 	if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1158 		f2fs_put_page(page, 1);
1159 		goto repeat;
1160 	}
1161 page_hit:
1162 	f2fs_bug_on(sbi, nid != nid_of_node(page));
1163 	return page;
1164 }
1165 
1166 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1167 {
1168 	return __get_node_page(sbi, nid, NULL, 0);
1169 }
1170 
1171 struct page *get_node_page_ra(struct page *parent, int start)
1172 {
1173 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1174 	nid_t nid = get_nid(parent, start, false);
1175 
1176 	return __get_node_page(sbi, nid, parent, start);
1177 }
1178 
1179 void sync_inode_page(struct dnode_of_data *dn)
1180 {
1181 	int ret = 0;
1182 
1183 	if (IS_INODE(dn->node_page) || dn->inode_page == dn->node_page) {
1184 		ret = update_inode(dn->inode, dn->node_page);
1185 	} else if (dn->inode_page) {
1186 		if (!dn->inode_page_locked)
1187 			lock_page(dn->inode_page);
1188 		ret = update_inode(dn->inode, dn->inode_page);
1189 		if (!dn->inode_page_locked)
1190 			unlock_page(dn->inode_page);
1191 	} else {
1192 		ret = update_inode_page(dn->inode);
1193 	}
1194 	dn->node_changed = ret ? true: false;
1195 }
1196 
1197 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1198 {
1199 	struct inode *inode;
1200 	struct page *page;
1201 	int ret;
1202 
1203 	/* should flush inline_data before evict_inode */
1204 	inode = ilookup(sbi->sb, ino);
1205 	if (!inode)
1206 		return;
1207 
1208 	page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1209 	if (!page)
1210 		goto iput_out;
1211 
1212 	if (!PageUptodate(page))
1213 		goto page_out;
1214 
1215 	if (!PageDirty(page))
1216 		goto page_out;
1217 
1218 	if (!clear_page_dirty_for_io(page))
1219 		goto page_out;
1220 
1221 	ret = f2fs_write_inline_data(inode, page);
1222 	inode_dec_dirty_pages(inode);
1223 	if (ret)
1224 		set_page_dirty(page);
1225 page_out:
1226 	f2fs_put_page(page, 1);
1227 iput_out:
1228 	iput(inode);
1229 }
1230 
1231 void move_node_page(struct page *node_page, int gc_type)
1232 {
1233 	if (gc_type == FG_GC) {
1234 		struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1235 		struct writeback_control wbc = {
1236 			.sync_mode = WB_SYNC_ALL,
1237 			.nr_to_write = 1,
1238 			.for_reclaim = 0,
1239 		};
1240 
1241 		set_page_dirty(node_page);
1242 		f2fs_wait_on_page_writeback(node_page, NODE, true);
1243 
1244 		f2fs_bug_on(sbi, PageWriteback(node_page));
1245 		if (!clear_page_dirty_for_io(node_page))
1246 			goto out_page;
1247 
1248 		if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1249 			unlock_page(node_page);
1250 		goto release_page;
1251 	} else {
1252 		/* set page dirty and write it */
1253 		if (!PageWriteback(node_page))
1254 			set_page_dirty(node_page);
1255 	}
1256 out_page:
1257 	unlock_page(node_page);
1258 release_page:
1259 	f2fs_put_page(node_page, 0);
1260 }
1261 
1262 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1263 {
1264 	pgoff_t index, end;
1265 	struct pagevec pvec;
1266 	struct page *last_page = NULL;
1267 
1268 	pagevec_init(&pvec, 0);
1269 	index = 0;
1270 	end = ULONG_MAX;
1271 
1272 	while (index <= end) {
1273 		int i, nr_pages;
1274 		nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1275 				PAGECACHE_TAG_DIRTY,
1276 				min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1277 		if (nr_pages == 0)
1278 			break;
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 int fsync_node_pages(struct f2fs_sb_info *sbi, nid_t ino,
1323 			struct writeback_control *wbc, bool atomic)
1324 {
1325 	pgoff_t index, end;
1326 	struct pagevec pvec;
1327 	int ret = 0;
1328 	struct page *last_page = NULL;
1329 	bool marked = false;
1330 
1331 	if (atomic) {
1332 		last_page = last_fsync_dnode(sbi, ino);
1333 		if (IS_ERR_OR_NULL(last_page))
1334 			return PTR_ERR_OR_ZERO(last_page);
1335 	}
1336 retry:
1337 	pagevec_init(&pvec, 0);
1338 	index = 0;
1339 	end = ULONG_MAX;
1340 
1341 	while (index <= end) {
1342 		int i, nr_pages;
1343 		nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1344 				PAGECACHE_TAG_DIRTY,
1345 				min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1346 		if (nr_pages == 0)
1347 			break;
1348 
1349 		for (i = 0; i < nr_pages; i++) {
1350 			struct page *page = pvec.pages[i];
1351 
1352 			if (unlikely(f2fs_cp_error(sbi))) {
1353 				f2fs_put_page(last_page, 0);
1354 				pagevec_release(&pvec);
1355 				return -EIO;
1356 			}
1357 
1358 			if (!IS_DNODE(page) || !is_cold_node(page))
1359 				continue;
1360 			if (ino_of_node(page) != ino)
1361 				continue;
1362 
1363 			lock_page(page);
1364 
1365 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1366 continue_unlock:
1367 				unlock_page(page);
1368 				continue;
1369 			}
1370 			if (ino_of_node(page) != ino)
1371 				goto continue_unlock;
1372 
1373 			if (!PageDirty(page) && page != last_page) {
1374 				/* someone wrote it for us */
1375 				goto continue_unlock;
1376 			}
1377 
1378 			f2fs_wait_on_page_writeback(page, NODE, true);
1379 			BUG_ON(PageWriteback(page));
1380 
1381 			if (!atomic || page == last_page) {
1382 				set_fsync_mark(page, 1);
1383 				if (IS_INODE(page))
1384 					set_dentry_mark(page,
1385 						need_dentry_mark(sbi, ino));
1386 				/*  may be written by other thread */
1387 				if (!PageDirty(page))
1388 					set_page_dirty(page);
1389 			}
1390 
1391 			if (!clear_page_dirty_for_io(page))
1392 				goto continue_unlock;
1393 
1394 			ret = NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1395 			if (ret) {
1396 				unlock_page(page);
1397 				f2fs_put_page(last_page, 0);
1398 				break;
1399 			}
1400 			if (page == last_page) {
1401 				f2fs_put_page(page, 0);
1402 				marked = true;
1403 				break;
1404 			}
1405 		}
1406 		pagevec_release(&pvec);
1407 		cond_resched();
1408 
1409 		if (ret || marked)
1410 			break;
1411 	}
1412 	if (!ret && atomic && !marked) {
1413 		f2fs_msg(sbi->sb, KERN_DEBUG,
1414 			"Retry to write fsync mark: ino=%u, idx=%lx",
1415 					ino, last_page->index);
1416 		lock_page(last_page);
1417 		set_page_dirty(last_page);
1418 		unlock_page(last_page);
1419 		goto retry;
1420 	}
1421 	return ret ? -EIO: 0;
1422 }
1423 
1424 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc)
1425 {
1426 	pgoff_t index, end;
1427 	struct pagevec pvec;
1428 	int step = 0;
1429 	int nwritten = 0;
1430 
1431 	pagevec_init(&pvec, 0);
1432 
1433 next_step:
1434 	index = 0;
1435 	end = ULONG_MAX;
1436 
1437 	while (index <= end) {
1438 		int i, nr_pages;
1439 		nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1440 				PAGECACHE_TAG_DIRTY,
1441 				min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1442 		if (nr_pages == 0)
1443 			break;
1444 
1445 		for (i = 0; i < nr_pages; i++) {
1446 			struct page *page = pvec.pages[i];
1447 
1448 			if (unlikely(f2fs_cp_error(sbi))) {
1449 				pagevec_release(&pvec);
1450 				return -EIO;
1451 			}
1452 
1453 			/*
1454 			 * flushing sequence with step:
1455 			 * 0. indirect nodes
1456 			 * 1. dentry dnodes
1457 			 * 2. file dnodes
1458 			 */
1459 			if (step == 0 && IS_DNODE(page))
1460 				continue;
1461 			if (step == 1 && (!IS_DNODE(page) ||
1462 						is_cold_node(page)))
1463 				continue;
1464 			if (step == 2 && (!IS_DNODE(page) ||
1465 						!is_cold_node(page)))
1466 				continue;
1467 lock_node:
1468 			if (!trylock_page(page))
1469 				continue;
1470 
1471 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1472 continue_unlock:
1473 				unlock_page(page);
1474 				continue;
1475 			}
1476 
1477 			if (!PageDirty(page)) {
1478 				/* someone wrote it for us */
1479 				goto continue_unlock;
1480 			}
1481 
1482 			/* flush inline_data */
1483 			if (is_inline_node(page)) {
1484 				clear_inline_node(page);
1485 				unlock_page(page);
1486 				flush_inline_data(sbi, ino_of_node(page));
1487 				goto lock_node;
1488 			}
1489 
1490 			f2fs_wait_on_page_writeback(page, NODE, true);
1491 
1492 			BUG_ON(PageWriteback(page));
1493 			if (!clear_page_dirty_for_io(page))
1494 				goto continue_unlock;
1495 
1496 			set_fsync_mark(page, 0);
1497 			set_dentry_mark(page, 0);
1498 
1499 			if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1500 				unlock_page(page);
1501 
1502 			if (--wbc->nr_to_write == 0)
1503 				break;
1504 		}
1505 		pagevec_release(&pvec);
1506 		cond_resched();
1507 
1508 		if (wbc->nr_to_write == 0) {
1509 			step = 2;
1510 			break;
1511 		}
1512 	}
1513 
1514 	if (step < 2) {
1515 		step++;
1516 		goto next_step;
1517 	}
1518 	return nwritten;
1519 }
1520 
1521 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1522 {
1523 	pgoff_t index = 0, end = ULONG_MAX;
1524 	struct pagevec pvec;
1525 	int ret2 = 0, ret = 0;
1526 
1527 	pagevec_init(&pvec, 0);
1528 
1529 	while (index <= end) {
1530 		int i, nr_pages;
1531 		nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1532 				PAGECACHE_TAG_WRITEBACK,
1533 				min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1534 		if (nr_pages == 0)
1535 			break;
1536 
1537 		for (i = 0; i < nr_pages; i++) {
1538 			struct page *page = pvec.pages[i];
1539 
1540 			/* until radix tree lookup accepts end_index */
1541 			if (unlikely(page->index > end))
1542 				continue;
1543 
1544 			if (ino && ino_of_node(page) == ino) {
1545 				f2fs_wait_on_page_writeback(page, NODE, true);
1546 				if (TestClearPageError(page))
1547 					ret = -EIO;
1548 			}
1549 		}
1550 		pagevec_release(&pvec);
1551 		cond_resched();
1552 	}
1553 
1554 	if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1555 		ret2 = -ENOSPC;
1556 	if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1557 		ret2 = -EIO;
1558 	if (!ret)
1559 		ret = ret2;
1560 	return ret;
1561 }
1562 
1563 static int f2fs_write_node_page(struct page *page,
1564 				struct writeback_control *wbc)
1565 {
1566 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1567 	nid_t nid;
1568 	struct node_info ni;
1569 	struct f2fs_io_info fio = {
1570 		.sbi = sbi,
1571 		.type = NODE,
1572 		.op = REQ_OP_WRITE,
1573 		.op_flags = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0,
1574 		.page = page,
1575 		.encrypted_page = NULL,
1576 	};
1577 
1578 	trace_f2fs_writepage(page, NODE);
1579 
1580 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1581 		goto redirty_out;
1582 	if (unlikely(f2fs_cp_error(sbi)))
1583 		goto redirty_out;
1584 
1585 	/* get old block addr of this node page */
1586 	nid = nid_of_node(page);
1587 	f2fs_bug_on(sbi, page->index != nid);
1588 
1589 	if (wbc->for_reclaim) {
1590 		if (!down_read_trylock(&sbi->node_write))
1591 			goto redirty_out;
1592 	} else {
1593 		down_read(&sbi->node_write);
1594 	}
1595 
1596 	get_node_info(sbi, nid, &ni);
1597 
1598 	/* This page is already truncated */
1599 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1600 		ClearPageUptodate(page);
1601 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1602 		up_read(&sbi->node_write);
1603 		unlock_page(page);
1604 		return 0;
1605 	}
1606 
1607 	set_page_writeback(page);
1608 	fio.old_blkaddr = ni.blk_addr;
1609 	write_node_page(nid, &fio);
1610 	set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1611 	dec_page_count(sbi, F2FS_DIRTY_NODES);
1612 	up_read(&sbi->node_write);
1613 
1614 	if (wbc->for_reclaim)
1615 		f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, NODE, WRITE);
1616 
1617 	unlock_page(page);
1618 
1619 	if (unlikely(f2fs_cp_error(sbi)))
1620 		f2fs_submit_merged_bio(sbi, NODE, WRITE);
1621 
1622 	return 0;
1623 
1624 redirty_out:
1625 	redirty_page_for_writepage(wbc, page);
1626 	return AOP_WRITEPAGE_ACTIVATE;
1627 }
1628 
1629 static int f2fs_write_node_pages(struct address_space *mapping,
1630 			    struct writeback_control *wbc)
1631 {
1632 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1633 	long diff;
1634 
1635 	/* balancing f2fs's metadata in background */
1636 	f2fs_balance_fs_bg(sbi);
1637 
1638 	/* collect a number of dirty node pages and write together */
1639 	if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1640 		goto skip_write;
1641 
1642 	trace_f2fs_writepages(mapping->host, wbc, NODE);
1643 
1644 	diff = nr_pages_to_write(sbi, NODE, wbc);
1645 	wbc->sync_mode = WB_SYNC_NONE;
1646 	sync_node_pages(sbi, wbc);
1647 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1648 	return 0;
1649 
1650 skip_write:
1651 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1652 	trace_f2fs_writepages(mapping->host, wbc, NODE);
1653 	return 0;
1654 }
1655 
1656 static int f2fs_set_node_page_dirty(struct page *page)
1657 {
1658 	trace_f2fs_set_page_dirty(page, NODE);
1659 
1660 	SetPageUptodate(page);
1661 	if (!PageDirty(page)) {
1662 		__set_page_dirty_nobuffers(page);
1663 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1664 		SetPagePrivate(page);
1665 		f2fs_trace_pid(page);
1666 		return 1;
1667 	}
1668 	return 0;
1669 }
1670 
1671 /*
1672  * Structure of the f2fs node operations
1673  */
1674 const struct address_space_operations f2fs_node_aops = {
1675 	.writepage	= f2fs_write_node_page,
1676 	.writepages	= f2fs_write_node_pages,
1677 	.set_page_dirty	= f2fs_set_node_page_dirty,
1678 	.invalidatepage	= f2fs_invalidate_page,
1679 	.releasepage	= f2fs_release_page,
1680 };
1681 
1682 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1683 						nid_t n)
1684 {
1685 	return radix_tree_lookup(&nm_i->free_nid_root, n);
1686 }
1687 
1688 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1689 						struct free_nid *i)
1690 {
1691 	list_del(&i->list);
1692 	radix_tree_delete(&nm_i->free_nid_root, i->nid);
1693 }
1694 
1695 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1696 {
1697 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1698 	struct free_nid *i;
1699 	struct nat_entry *ne;
1700 
1701 	if (!available_free_memory(sbi, FREE_NIDS))
1702 		return -1;
1703 
1704 	/* 0 nid should not be used */
1705 	if (unlikely(nid == 0))
1706 		return 0;
1707 
1708 	if (build) {
1709 		/* do not add allocated nids */
1710 		ne = __lookup_nat_cache(nm_i, nid);
1711 		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1712 				nat_get_blkaddr(ne) != NULL_ADDR))
1713 			return 0;
1714 	}
1715 
1716 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1717 	i->nid = nid;
1718 	i->state = NID_NEW;
1719 
1720 	if (radix_tree_preload(GFP_NOFS)) {
1721 		kmem_cache_free(free_nid_slab, i);
1722 		return 0;
1723 	}
1724 
1725 	spin_lock(&nm_i->free_nid_list_lock);
1726 	if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1727 		spin_unlock(&nm_i->free_nid_list_lock);
1728 		radix_tree_preload_end();
1729 		kmem_cache_free(free_nid_slab, i);
1730 		return 0;
1731 	}
1732 	list_add_tail(&i->list, &nm_i->free_nid_list);
1733 	nm_i->fcnt++;
1734 	spin_unlock(&nm_i->free_nid_list_lock);
1735 	radix_tree_preload_end();
1736 	return 1;
1737 }
1738 
1739 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1740 {
1741 	struct free_nid *i;
1742 	bool need_free = false;
1743 
1744 	spin_lock(&nm_i->free_nid_list_lock);
1745 	i = __lookup_free_nid_list(nm_i, nid);
1746 	if (i && i->state == NID_NEW) {
1747 		__del_from_free_nid_list(nm_i, i);
1748 		nm_i->fcnt--;
1749 		need_free = true;
1750 	}
1751 	spin_unlock(&nm_i->free_nid_list_lock);
1752 
1753 	if (need_free)
1754 		kmem_cache_free(free_nid_slab, i);
1755 }
1756 
1757 static void scan_nat_page(struct f2fs_sb_info *sbi,
1758 			struct page *nat_page, nid_t start_nid)
1759 {
1760 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1761 	struct f2fs_nat_block *nat_blk = page_address(nat_page);
1762 	block_t blk_addr;
1763 	int i;
1764 
1765 	i = start_nid % NAT_ENTRY_PER_BLOCK;
1766 
1767 	for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1768 
1769 		if (unlikely(start_nid >= nm_i->max_nid))
1770 			break;
1771 
1772 		blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1773 		f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1774 		if (blk_addr == NULL_ADDR) {
1775 			if (add_free_nid(sbi, start_nid, true) < 0)
1776 				break;
1777 		}
1778 	}
1779 }
1780 
1781 static void build_free_nids(struct f2fs_sb_info *sbi)
1782 {
1783 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1784 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1785 	struct f2fs_journal *journal = curseg->journal;
1786 	int i = 0;
1787 	nid_t nid = nm_i->next_scan_nid;
1788 
1789 	/* Enough entries */
1790 	if (nm_i->fcnt > NAT_ENTRY_PER_BLOCK)
1791 		return;
1792 
1793 	/* readahead nat pages to be scanned */
1794 	ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1795 							META_NAT, true);
1796 
1797 	down_read(&nm_i->nat_tree_lock);
1798 
1799 	while (1) {
1800 		struct page *page = get_current_nat_page(sbi, nid);
1801 
1802 		scan_nat_page(sbi, page, nid);
1803 		f2fs_put_page(page, 1);
1804 
1805 		nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1806 		if (unlikely(nid >= nm_i->max_nid))
1807 			nid = 0;
1808 
1809 		if (++i >= FREE_NID_PAGES)
1810 			break;
1811 	}
1812 
1813 	/* go to the next free nat pages to find free nids abundantly */
1814 	nm_i->next_scan_nid = nid;
1815 
1816 	/* find free nids from current sum_pages */
1817 	down_read(&curseg->journal_rwsem);
1818 	for (i = 0; i < nats_in_cursum(journal); i++) {
1819 		block_t addr;
1820 
1821 		addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1822 		nid = le32_to_cpu(nid_in_journal(journal, i));
1823 		if (addr == NULL_ADDR)
1824 			add_free_nid(sbi, nid, true);
1825 		else
1826 			remove_free_nid(nm_i, nid);
1827 	}
1828 	up_read(&curseg->journal_rwsem);
1829 	up_read(&nm_i->nat_tree_lock);
1830 
1831 	ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1832 					nm_i->ra_nid_pages, META_NAT, false);
1833 }
1834 
1835 /*
1836  * If this function returns success, caller can obtain a new nid
1837  * from second parameter of this function.
1838  * The returned nid could be used ino as well as nid when inode is created.
1839  */
1840 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1841 {
1842 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1843 	struct free_nid *i = NULL;
1844 retry:
1845 #ifdef CONFIG_F2FS_FAULT_INJECTION
1846 	if (time_to_inject(FAULT_ALLOC_NID))
1847 		return false;
1848 #endif
1849 	if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1850 		return false;
1851 
1852 	spin_lock(&nm_i->free_nid_list_lock);
1853 
1854 	/* We should not use stale free nids created by build_free_nids */
1855 	if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1856 		f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1857 		list_for_each_entry(i, &nm_i->free_nid_list, list)
1858 			if (i->state == NID_NEW)
1859 				break;
1860 
1861 		f2fs_bug_on(sbi, i->state != NID_NEW);
1862 		*nid = i->nid;
1863 		i->state = NID_ALLOC;
1864 		nm_i->fcnt--;
1865 		spin_unlock(&nm_i->free_nid_list_lock);
1866 		return true;
1867 	}
1868 	spin_unlock(&nm_i->free_nid_list_lock);
1869 
1870 	/* Let's scan nat pages and its caches to get free nids */
1871 	mutex_lock(&nm_i->build_lock);
1872 	build_free_nids(sbi);
1873 	mutex_unlock(&nm_i->build_lock);
1874 	goto retry;
1875 }
1876 
1877 /*
1878  * alloc_nid() should be called prior to this function.
1879  */
1880 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1881 {
1882 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1883 	struct free_nid *i;
1884 
1885 	spin_lock(&nm_i->free_nid_list_lock);
1886 	i = __lookup_free_nid_list(nm_i, nid);
1887 	f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1888 	__del_from_free_nid_list(nm_i, i);
1889 	spin_unlock(&nm_i->free_nid_list_lock);
1890 
1891 	kmem_cache_free(free_nid_slab, i);
1892 }
1893 
1894 /*
1895  * alloc_nid() should be called prior to this function.
1896  */
1897 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1898 {
1899 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1900 	struct free_nid *i;
1901 	bool need_free = false;
1902 
1903 	if (!nid)
1904 		return;
1905 
1906 	spin_lock(&nm_i->free_nid_list_lock);
1907 	i = __lookup_free_nid_list(nm_i, nid);
1908 	f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1909 	if (!available_free_memory(sbi, FREE_NIDS)) {
1910 		__del_from_free_nid_list(nm_i, i);
1911 		need_free = true;
1912 	} else {
1913 		i->state = NID_NEW;
1914 		nm_i->fcnt++;
1915 	}
1916 	spin_unlock(&nm_i->free_nid_list_lock);
1917 
1918 	if (need_free)
1919 		kmem_cache_free(free_nid_slab, i);
1920 }
1921 
1922 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1923 {
1924 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1925 	struct free_nid *i, *next;
1926 	int nr = nr_shrink;
1927 
1928 	if (!mutex_trylock(&nm_i->build_lock))
1929 		return 0;
1930 
1931 	spin_lock(&nm_i->free_nid_list_lock);
1932 	list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1933 		if (nr_shrink <= 0 || nm_i->fcnt <= NAT_ENTRY_PER_BLOCK)
1934 			break;
1935 		if (i->state == NID_ALLOC)
1936 			continue;
1937 		__del_from_free_nid_list(nm_i, i);
1938 		kmem_cache_free(free_nid_slab, i);
1939 		nm_i->fcnt--;
1940 		nr_shrink--;
1941 	}
1942 	spin_unlock(&nm_i->free_nid_list_lock);
1943 	mutex_unlock(&nm_i->build_lock);
1944 
1945 	return nr - nr_shrink;
1946 }
1947 
1948 void recover_inline_xattr(struct inode *inode, struct page *page)
1949 {
1950 	void *src_addr, *dst_addr;
1951 	size_t inline_size;
1952 	struct page *ipage;
1953 	struct f2fs_inode *ri;
1954 
1955 	ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1956 	f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1957 
1958 	ri = F2FS_INODE(page);
1959 	if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1960 		clear_inode_flag(F2FS_I(inode), FI_INLINE_XATTR);
1961 		goto update_inode;
1962 	}
1963 
1964 	dst_addr = inline_xattr_addr(ipage);
1965 	src_addr = inline_xattr_addr(page);
1966 	inline_size = inline_xattr_size(inode);
1967 
1968 	f2fs_wait_on_page_writeback(ipage, NODE, true);
1969 	memcpy(dst_addr, src_addr, inline_size);
1970 update_inode:
1971 	update_inode(inode, ipage);
1972 	f2fs_put_page(ipage, 1);
1973 }
1974 
1975 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1976 {
1977 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1978 	nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1979 	nid_t new_xnid = nid_of_node(page);
1980 	struct node_info ni;
1981 
1982 	/* 1: invalidate the previous xattr nid */
1983 	if (!prev_xnid)
1984 		goto recover_xnid;
1985 
1986 	/* Deallocate node address */
1987 	get_node_info(sbi, prev_xnid, &ni);
1988 	f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1989 	invalidate_blocks(sbi, ni.blk_addr);
1990 	dec_valid_node_count(sbi, inode);
1991 	set_node_addr(sbi, &ni, NULL_ADDR, false);
1992 
1993 recover_xnid:
1994 	/* 2: allocate new xattr nid */
1995 	if (unlikely(!inc_valid_node_count(sbi, inode)))
1996 		f2fs_bug_on(sbi, 1);
1997 
1998 	remove_free_nid(NM_I(sbi), new_xnid);
1999 	get_node_info(sbi, new_xnid, &ni);
2000 	ni.ino = inode->i_ino;
2001 	set_node_addr(sbi, &ni, NEW_ADDR, false);
2002 	F2FS_I(inode)->i_xattr_nid = new_xnid;
2003 
2004 	/* 3: update xattr blkaddr */
2005 	refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
2006 	set_node_addr(sbi, &ni, blkaddr, false);
2007 
2008 	update_inode_page(inode);
2009 }
2010 
2011 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2012 {
2013 	struct f2fs_inode *src, *dst;
2014 	nid_t ino = ino_of_node(page);
2015 	struct node_info old_ni, new_ni;
2016 	struct page *ipage;
2017 
2018 	get_node_info(sbi, ino, &old_ni);
2019 
2020 	if (unlikely(old_ni.blk_addr != NULL_ADDR))
2021 		return -EINVAL;
2022 
2023 	ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2024 	if (!ipage)
2025 		return -ENOMEM;
2026 
2027 	/* Should not use this inode from free nid list */
2028 	remove_free_nid(NM_I(sbi), ino);
2029 
2030 	SetPageUptodate(ipage);
2031 	fill_node_footer(ipage, ino, ino, 0, true);
2032 
2033 	src = F2FS_INODE(page);
2034 	dst = F2FS_INODE(ipage);
2035 
2036 	memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2037 	dst->i_size = 0;
2038 	dst->i_blocks = cpu_to_le64(1);
2039 	dst->i_links = cpu_to_le32(1);
2040 	dst->i_xattr_nid = 0;
2041 	dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
2042 
2043 	new_ni = old_ni;
2044 	new_ni.ino = ino;
2045 
2046 	if (unlikely(!inc_valid_node_count(sbi, NULL)))
2047 		WARN_ON(1);
2048 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2049 	inc_valid_inode_count(sbi);
2050 	set_page_dirty(ipage);
2051 	f2fs_put_page(ipage, 1);
2052 	return 0;
2053 }
2054 
2055 int restore_node_summary(struct f2fs_sb_info *sbi,
2056 			unsigned int segno, struct f2fs_summary_block *sum)
2057 {
2058 	struct f2fs_node *rn;
2059 	struct f2fs_summary *sum_entry;
2060 	block_t addr;
2061 	int bio_blocks = MAX_BIO_BLOCKS(sbi);
2062 	int i, idx, last_offset, nrpages;
2063 
2064 	/* scan the node segment */
2065 	last_offset = sbi->blocks_per_seg;
2066 	addr = START_BLOCK(sbi, segno);
2067 	sum_entry = &sum->entries[0];
2068 
2069 	for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2070 		nrpages = min(last_offset - i, bio_blocks);
2071 
2072 		/* readahead node pages */
2073 		ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2074 
2075 		for (idx = addr; idx < addr + nrpages; idx++) {
2076 			struct page *page = get_tmp_page(sbi, idx);
2077 
2078 			rn = F2FS_NODE(page);
2079 			sum_entry->nid = rn->footer.nid;
2080 			sum_entry->version = 0;
2081 			sum_entry->ofs_in_node = 0;
2082 			sum_entry++;
2083 			f2fs_put_page(page, 1);
2084 		}
2085 
2086 		invalidate_mapping_pages(META_MAPPING(sbi), addr,
2087 							addr + nrpages);
2088 	}
2089 	return 0;
2090 }
2091 
2092 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2093 {
2094 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2095 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2096 	struct f2fs_journal *journal = curseg->journal;
2097 	int i;
2098 
2099 	down_write(&curseg->journal_rwsem);
2100 	for (i = 0; i < nats_in_cursum(journal); i++) {
2101 		struct nat_entry *ne;
2102 		struct f2fs_nat_entry raw_ne;
2103 		nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2104 
2105 		raw_ne = nat_in_journal(journal, i);
2106 
2107 		ne = __lookup_nat_cache(nm_i, nid);
2108 		if (!ne) {
2109 			ne = grab_nat_entry(nm_i, nid);
2110 			node_info_from_raw_nat(&ne->ni, &raw_ne);
2111 		}
2112 		__set_nat_cache_dirty(nm_i, ne);
2113 	}
2114 	update_nats_in_cursum(journal, -i);
2115 	up_write(&curseg->journal_rwsem);
2116 }
2117 
2118 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2119 						struct list_head *head, int max)
2120 {
2121 	struct nat_entry_set *cur;
2122 
2123 	if (nes->entry_cnt >= max)
2124 		goto add_out;
2125 
2126 	list_for_each_entry(cur, head, set_list) {
2127 		if (cur->entry_cnt >= nes->entry_cnt) {
2128 			list_add(&nes->set_list, cur->set_list.prev);
2129 			return;
2130 		}
2131 	}
2132 add_out:
2133 	list_add_tail(&nes->set_list, head);
2134 }
2135 
2136 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2137 					struct nat_entry_set *set)
2138 {
2139 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2140 	struct f2fs_journal *journal = curseg->journal;
2141 	nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2142 	bool to_journal = true;
2143 	struct f2fs_nat_block *nat_blk;
2144 	struct nat_entry *ne, *cur;
2145 	struct page *page = NULL;
2146 
2147 	/*
2148 	 * there are two steps to flush nat entries:
2149 	 * #1, flush nat entries to journal in current hot data summary block.
2150 	 * #2, flush nat entries to nat page.
2151 	 */
2152 	if (!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2153 		to_journal = false;
2154 
2155 	if (to_journal) {
2156 		down_write(&curseg->journal_rwsem);
2157 	} else {
2158 		page = get_next_nat_page(sbi, start_nid);
2159 		nat_blk = page_address(page);
2160 		f2fs_bug_on(sbi, !nat_blk);
2161 	}
2162 
2163 	/* flush dirty nats in nat entry set */
2164 	list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2165 		struct f2fs_nat_entry *raw_ne;
2166 		nid_t nid = nat_get_nid(ne);
2167 		int offset;
2168 
2169 		if (nat_get_blkaddr(ne) == NEW_ADDR)
2170 			continue;
2171 
2172 		if (to_journal) {
2173 			offset = lookup_journal_in_cursum(journal,
2174 							NAT_JOURNAL, nid, 1);
2175 			f2fs_bug_on(sbi, offset < 0);
2176 			raw_ne = &nat_in_journal(journal, offset);
2177 			nid_in_journal(journal, offset) = cpu_to_le32(nid);
2178 		} else {
2179 			raw_ne = &nat_blk->entries[nid - start_nid];
2180 		}
2181 		raw_nat_from_node_info(raw_ne, &ne->ni);
2182 		nat_reset_flag(ne);
2183 		__clear_nat_cache_dirty(NM_I(sbi), ne);
2184 		if (nat_get_blkaddr(ne) == NULL_ADDR)
2185 			add_free_nid(sbi, nid, false);
2186 	}
2187 
2188 	if (to_journal)
2189 		up_write(&curseg->journal_rwsem);
2190 	else
2191 		f2fs_put_page(page, 1);
2192 
2193 	f2fs_bug_on(sbi, set->entry_cnt);
2194 
2195 	radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2196 	kmem_cache_free(nat_entry_set_slab, set);
2197 }
2198 
2199 /*
2200  * This function is called during the checkpointing process.
2201  */
2202 void flush_nat_entries(struct f2fs_sb_info *sbi)
2203 {
2204 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2205 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2206 	struct f2fs_journal *journal = curseg->journal;
2207 	struct nat_entry_set *setvec[SETVEC_SIZE];
2208 	struct nat_entry_set *set, *tmp;
2209 	unsigned int found;
2210 	nid_t set_idx = 0;
2211 	LIST_HEAD(sets);
2212 
2213 	if (!nm_i->dirty_nat_cnt)
2214 		return;
2215 
2216 	down_write(&nm_i->nat_tree_lock);
2217 
2218 	/*
2219 	 * if there are no enough space in journal to store dirty nat
2220 	 * entries, remove all entries from journal and merge them
2221 	 * into nat entry set.
2222 	 */
2223 	if (!__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2224 		remove_nats_in_journal(sbi);
2225 
2226 	while ((found = __gang_lookup_nat_set(nm_i,
2227 					set_idx, SETVEC_SIZE, setvec))) {
2228 		unsigned idx;
2229 		set_idx = setvec[found - 1]->set + 1;
2230 		for (idx = 0; idx < found; idx++)
2231 			__adjust_nat_entry_set(setvec[idx], &sets,
2232 						MAX_NAT_JENTRIES(journal));
2233 	}
2234 
2235 	/* flush dirty nats in nat entry set */
2236 	list_for_each_entry_safe(set, tmp, &sets, set_list)
2237 		__flush_nat_entry_set(sbi, set);
2238 
2239 	up_write(&nm_i->nat_tree_lock);
2240 
2241 	f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2242 }
2243 
2244 static int init_node_manager(struct f2fs_sb_info *sbi)
2245 {
2246 	struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2247 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2248 	unsigned char *version_bitmap;
2249 	unsigned int nat_segs, nat_blocks;
2250 
2251 	nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2252 
2253 	/* segment_count_nat includes pair segment so divide to 2. */
2254 	nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2255 	nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2256 
2257 	nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2258 
2259 	/* not used nids: 0, node, meta, (and root counted as valid node) */
2260 	nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
2261 	nm_i->fcnt = 0;
2262 	nm_i->nat_cnt = 0;
2263 	nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2264 	nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2265 	nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2266 
2267 	INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2268 	INIT_LIST_HEAD(&nm_i->free_nid_list);
2269 	INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2270 	INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2271 	INIT_LIST_HEAD(&nm_i->nat_entries);
2272 
2273 	mutex_init(&nm_i->build_lock);
2274 	spin_lock_init(&nm_i->free_nid_list_lock);
2275 	init_rwsem(&nm_i->nat_tree_lock);
2276 
2277 	nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2278 	nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2279 	version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2280 	if (!version_bitmap)
2281 		return -EFAULT;
2282 
2283 	nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2284 					GFP_KERNEL);
2285 	if (!nm_i->nat_bitmap)
2286 		return -ENOMEM;
2287 	return 0;
2288 }
2289 
2290 int build_node_manager(struct f2fs_sb_info *sbi)
2291 {
2292 	int err;
2293 
2294 	sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2295 	if (!sbi->nm_info)
2296 		return -ENOMEM;
2297 
2298 	err = init_node_manager(sbi);
2299 	if (err)
2300 		return err;
2301 
2302 	build_free_nids(sbi);
2303 	return 0;
2304 }
2305 
2306 void destroy_node_manager(struct f2fs_sb_info *sbi)
2307 {
2308 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2309 	struct free_nid *i, *next_i;
2310 	struct nat_entry *natvec[NATVEC_SIZE];
2311 	struct nat_entry_set *setvec[SETVEC_SIZE];
2312 	nid_t nid = 0;
2313 	unsigned int found;
2314 
2315 	if (!nm_i)
2316 		return;
2317 
2318 	/* destroy free nid list */
2319 	spin_lock(&nm_i->free_nid_list_lock);
2320 	list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2321 		f2fs_bug_on(sbi, i->state == NID_ALLOC);
2322 		__del_from_free_nid_list(nm_i, i);
2323 		nm_i->fcnt--;
2324 		spin_unlock(&nm_i->free_nid_list_lock);
2325 		kmem_cache_free(free_nid_slab, i);
2326 		spin_lock(&nm_i->free_nid_list_lock);
2327 	}
2328 	f2fs_bug_on(sbi, nm_i->fcnt);
2329 	spin_unlock(&nm_i->free_nid_list_lock);
2330 
2331 	/* destroy nat cache */
2332 	down_write(&nm_i->nat_tree_lock);
2333 	while ((found = __gang_lookup_nat_cache(nm_i,
2334 					nid, NATVEC_SIZE, natvec))) {
2335 		unsigned idx;
2336 
2337 		nid = nat_get_nid(natvec[found - 1]) + 1;
2338 		for (idx = 0; idx < found; idx++)
2339 			__del_from_nat_cache(nm_i, natvec[idx]);
2340 	}
2341 	f2fs_bug_on(sbi, nm_i->nat_cnt);
2342 
2343 	/* destroy nat set cache */
2344 	nid = 0;
2345 	while ((found = __gang_lookup_nat_set(nm_i,
2346 					nid, SETVEC_SIZE, setvec))) {
2347 		unsigned idx;
2348 
2349 		nid = setvec[found - 1]->set + 1;
2350 		for (idx = 0; idx < found; idx++) {
2351 			/* entry_cnt is not zero, when cp_error was occurred */
2352 			f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2353 			radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2354 			kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2355 		}
2356 	}
2357 	up_write(&nm_i->nat_tree_lock);
2358 
2359 	kfree(nm_i->nat_bitmap);
2360 	sbi->nm_info = NULL;
2361 	kfree(nm_i);
2362 }
2363 
2364 int __init create_node_manager_caches(void)
2365 {
2366 	nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2367 			sizeof(struct nat_entry));
2368 	if (!nat_entry_slab)
2369 		goto fail;
2370 
2371 	free_nid_slab = f2fs_kmem_cache_create("free_nid",
2372 			sizeof(struct free_nid));
2373 	if (!free_nid_slab)
2374 		goto destroy_nat_entry;
2375 
2376 	nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2377 			sizeof(struct nat_entry_set));
2378 	if (!nat_entry_set_slab)
2379 		goto destroy_free_nid;
2380 	return 0;
2381 
2382 destroy_free_nid:
2383 	kmem_cache_destroy(free_nid_slab);
2384 destroy_nat_entry:
2385 	kmem_cache_destroy(nat_entry_slab);
2386 fail:
2387 	return -ENOMEM;
2388 }
2389 
2390 void destroy_node_manager_caches(void)
2391 {
2392 	kmem_cache_destroy(nat_entry_set_slab);
2393 	kmem_cache_destroy(free_nid_slab);
2394 	kmem_cache_destroy(nat_entry_slab);
2395 }
2396