xref: /linux/fs/f2fs/node.c (revision b6ebbac51bedf9e98e837688bc838f400196da5e)
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 		if (excess_cached_nats(sbi))
56 			res = false;
57 		if (nm_i->nat_cnt > DEF_NAT_CACHE_THRESHOLD)
58 			res = false;
59 	} else if (type == DIRTY_DENTS) {
60 		if (sbi->sb->s_bdi->wb.dirty_exceeded)
61 			return false;
62 		mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
63 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
64 	} else if (type == INO_ENTRIES) {
65 		int i;
66 
67 		for (i = 0; i <= UPDATE_INO; i++)
68 			mem_size += (sbi->im[i].ino_num *
69 				sizeof(struct ino_entry)) >> 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 {
78 		if (!sbi->sb->s_bdi->wb.dirty_exceeded)
79 			return true;
80 	}
81 	return res;
82 }
83 
84 static void clear_node_page_dirty(struct page *page)
85 {
86 	struct address_space *mapping = page->mapping;
87 	unsigned int long flags;
88 
89 	if (PageDirty(page)) {
90 		spin_lock_irqsave(&mapping->tree_lock, flags);
91 		radix_tree_tag_clear(&mapping->page_tree,
92 				page_index(page),
93 				PAGECACHE_TAG_DIRTY);
94 		spin_unlock_irqrestore(&mapping->tree_lock, flags);
95 
96 		clear_page_dirty_for_io(page);
97 		dec_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
98 	}
99 	ClearPageUptodate(page);
100 }
101 
102 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
103 {
104 	pgoff_t index = current_nat_addr(sbi, nid);
105 	return get_meta_page(sbi, index);
106 }
107 
108 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
109 {
110 	struct page *src_page;
111 	struct page *dst_page;
112 	pgoff_t src_off;
113 	pgoff_t dst_off;
114 	void *src_addr;
115 	void *dst_addr;
116 	struct f2fs_nm_info *nm_i = NM_I(sbi);
117 
118 	src_off = current_nat_addr(sbi, nid);
119 	dst_off = next_nat_addr(sbi, src_off);
120 
121 	/* get current nat block page with lock */
122 	src_page = get_meta_page(sbi, src_off);
123 	dst_page = grab_meta_page(sbi, dst_off);
124 	f2fs_bug_on(sbi, PageDirty(src_page));
125 
126 	src_addr = page_address(src_page);
127 	dst_addr = page_address(dst_page);
128 	memcpy(dst_addr, src_addr, PAGE_SIZE);
129 	set_page_dirty(dst_page);
130 	f2fs_put_page(src_page, 1);
131 
132 	set_to_next_nat(nm_i, nid);
133 
134 	return dst_page;
135 }
136 
137 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
138 {
139 	return radix_tree_lookup(&nm_i->nat_root, n);
140 }
141 
142 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
143 		nid_t start, unsigned int nr, struct nat_entry **ep)
144 {
145 	return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
146 }
147 
148 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
149 {
150 	list_del(&e->list);
151 	radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
152 	nm_i->nat_cnt--;
153 	kmem_cache_free(nat_entry_slab, e);
154 }
155 
156 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
157 						struct nat_entry *ne)
158 {
159 	nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
160 	struct nat_entry_set *head;
161 
162 	if (get_nat_flag(ne, IS_DIRTY))
163 		return;
164 
165 	head = radix_tree_lookup(&nm_i->nat_set_root, set);
166 	if (!head) {
167 		head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
168 
169 		INIT_LIST_HEAD(&head->entry_list);
170 		INIT_LIST_HEAD(&head->set_list);
171 		head->set = set;
172 		head->entry_cnt = 0;
173 		f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
174 	}
175 	list_move_tail(&ne->list, &head->entry_list);
176 	nm_i->dirty_nat_cnt++;
177 	head->entry_cnt++;
178 	set_nat_flag(ne, IS_DIRTY, true);
179 }
180 
181 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
182 						struct nat_entry *ne)
183 {
184 	nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
185 	struct nat_entry_set *head;
186 
187 	head = radix_tree_lookup(&nm_i->nat_set_root, set);
188 	if (head) {
189 		list_move_tail(&ne->list, &nm_i->nat_entries);
190 		set_nat_flag(ne, IS_DIRTY, false);
191 		head->entry_cnt--;
192 		nm_i->dirty_nat_cnt--;
193 	}
194 }
195 
196 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
197 		nid_t start, unsigned int nr, struct nat_entry_set **ep)
198 {
199 	return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
200 							start, nr);
201 }
202 
203 int need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
204 {
205 	struct f2fs_nm_info *nm_i = NM_I(sbi);
206 	struct nat_entry *e;
207 	bool need = false;
208 
209 	percpu_down_read(&nm_i->nat_tree_lock);
210 	e = __lookup_nat_cache(nm_i, nid);
211 	if (e) {
212 		if (!get_nat_flag(e, IS_CHECKPOINTED) &&
213 				!get_nat_flag(e, HAS_FSYNCED_INODE))
214 			need = true;
215 	}
216 	percpu_up_read(&nm_i->nat_tree_lock);
217 	return need;
218 }
219 
220 bool is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
221 {
222 	struct f2fs_nm_info *nm_i = NM_I(sbi);
223 	struct nat_entry *e;
224 	bool is_cp = true;
225 
226 	percpu_down_read(&nm_i->nat_tree_lock);
227 	e = __lookup_nat_cache(nm_i, nid);
228 	if (e && !get_nat_flag(e, IS_CHECKPOINTED))
229 		is_cp = false;
230 	percpu_up_read(&nm_i->nat_tree_lock);
231 	return is_cp;
232 }
233 
234 bool need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
235 {
236 	struct f2fs_nm_info *nm_i = NM_I(sbi);
237 	struct nat_entry *e;
238 	bool need_update = true;
239 
240 	percpu_down_read(&nm_i->nat_tree_lock);
241 	e = __lookup_nat_cache(nm_i, ino);
242 	if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
243 			(get_nat_flag(e, IS_CHECKPOINTED) ||
244 			 get_nat_flag(e, HAS_FSYNCED_INODE)))
245 		need_update = false;
246 	percpu_up_read(&nm_i->nat_tree_lock);
247 	return need_update;
248 }
249 
250 static struct nat_entry *grab_nat_entry(struct f2fs_nm_info *nm_i, nid_t nid)
251 {
252 	struct nat_entry *new;
253 
254 	new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_NOFS);
255 	f2fs_radix_tree_insert(&nm_i->nat_root, nid, new);
256 	memset(new, 0, sizeof(struct nat_entry));
257 	nat_set_nid(new, nid);
258 	nat_reset_flag(new);
259 	list_add_tail(&new->list, &nm_i->nat_entries);
260 	nm_i->nat_cnt++;
261 	return new;
262 }
263 
264 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
265 						struct f2fs_nat_entry *ne)
266 {
267 	struct f2fs_nm_info *nm_i = NM_I(sbi);
268 	struct nat_entry *e;
269 
270 	e = __lookup_nat_cache(nm_i, nid);
271 	if (!e) {
272 		e = grab_nat_entry(nm_i, nid);
273 		node_info_from_raw_nat(&e->ni, ne);
274 	} else {
275 		f2fs_bug_on(sbi, nat_get_ino(e) != ne->ino ||
276 				nat_get_blkaddr(e) != ne->block_addr ||
277 				nat_get_version(e) != ne->version);
278 	}
279 }
280 
281 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
282 			block_t new_blkaddr, bool fsync_done)
283 {
284 	struct f2fs_nm_info *nm_i = NM_I(sbi);
285 	struct nat_entry *e;
286 
287 	percpu_down_write(&nm_i->nat_tree_lock);
288 	e = __lookup_nat_cache(nm_i, ni->nid);
289 	if (!e) {
290 		e = grab_nat_entry(nm_i, ni->nid);
291 		copy_node_info(&e->ni, ni);
292 		f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
293 	} else if (new_blkaddr == NEW_ADDR) {
294 		/*
295 		 * when nid is reallocated,
296 		 * previous nat entry can be remained in nat cache.
297 		 * So, reinitialize it with new information.
298 		 */
299 		copy_node_info(&e->ni, ni);
300 		f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
301 	}
302 
303 	/* sanity check */
304 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
305 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
306 			new_blkaddr == NULL_ADDR);
307 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
308 			new_blkaddr == NEW_ADDR);
309 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != NEW_ADDR &&
310 			nat_get_blkaddr(e) != NULL_ADDR &&
311 			new_blkaddr == NEW_ADDR);
312 
313 	/* increment version no as node is removed */
314 	if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
315 		unsigned char version = nat_get_version(e);
316 		nat_set_version(e, inc_node_version(version));
317 
318 		/* in order to reuse the nid */
319 		if (nm_i->next_scan_nid > ni->nid)
320 			nm_i->next_scan_nid = ni->nid;
321 	}
322 
323 	/* change address */
324 	nat_set_blkaddr(e, new_blkaddr);
325 	if (new_blkaddr == NEW_ADDR || new_blkaddr == NULL_ADDR)
326 		set_nat_flag(e, IS_CHECKPOINTED, false);
327 	__set_nat_cache_dirty(nm_i, e);
328 
329 	/* update fsync_mark if its inode nat entry is still alive */
330 	if (ni->nid != ni->ino)
331 		e = __lookup_nat_cache(nm_i, ni->ino);
332 	if (e) {
333 		if (fsync_done && ni->nid == ni->ino)
334 			set_nat_flag(e, HAS_FSYNCED_INODE, true);
335 		set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
336 	}
337 	percpu_up_write(&nm_i->nat_tree_lock);
338 }
339 
340 int try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
341 {
342 	struct f2fs_nm_info *nm_i = NM_I(sbi);
343 	int nr = nr_shrink;
344 
345 	percpu_down_write(&nm_i->nat_tree_lock);
346 
347 	while (nr_shrink && !list_empty(&nm_i->nat_entries)) {
348 		struct nat_entry *ne;
349 		ne = list_first_entry(&nm_i->nat_entries,
350 					struct nat_entry, list);
351 		__del_from_nat_cache(nm_i, ne);
352 		nr_shrink--;
353 	}
354 	percpu_up_write(&nm_i->nat_tree_lock);
355 	return nr - nr_shrink;
356 }
357 
358 /*
359  * This function always returns success
360  */
361 void get_node_info(struct f2fs_sb_info *sbi, nid_t nid, struct node_info *ni)
362 {
363 	struct f2fs_nm_info *nm_i = NM_I(sbi);
364 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
365 	struct f2fs_journal *journal = curseg->journal;
366 	nid_t start_nid = START_NID(nid);
367 	struct f2fs_nat_block *nat_blk;
368 	struct page *page = NULL;
369 	struct f2fs_nat_entry ne;
370 	struct nat_entry *e;
371 	int i;
372 
373 	ni->nid = nid;
374 
375 	/* Check nat cache */
376 	percpu_down_read(&nm_i->nat_tree_lock);
377 	e = __lookup_nat_cache(nm_i, nid);
378 	if (e) {
379 		ni->ino = nat_get_ino(e);
380 		ni->blk_addr = nat_get_blkaddr(e);
381 		ni->version = nat_get_version(e);
382 		percpu_up_read(&nm_i->nat_tree_lock);
383 		return;
384 	}
385 
386 	memset(&ne, 0, sizeof(struct f2fs_nat_entry));
387 
388 	/* Check current segment summary */
389 	down_read(&curseg->journal_rwsem);
390 	i = lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
391 	if (i >= 0) {
392 		ne = nat_in_journal(journal, i);
393 		node_info_from_raw_nat(ni, &ne);
394 	}
395 	up_read(&curseg->journal_rwsem);
396 	if (i >= 0)
397 		goto cache;
398 
399 	/* Fill node_info from nat page */
400 	page = get_current_nat_page(sbi, start_nid);
401 	nat_blk = (struct f2fs_nat_block *)page_address(page);
402 	ne = nat_blk->entries[nid - start_nid];
403 	node_info_from_raw_nat(ni, &ne);
404 	f2fs_put_page(page, 1);
405 cache:
406 	percpu_up_read(&nm_i->nat_tree_lock);
407 	/* cache nat entry */
408 	percpu_down_write(&nm_i->nat_tree_lock);
409 	cache_nat_entry(sbi, nid, &ne);
410 	percpu_up_write(&nm_i->nat_tree_lock);
411 }
412 
413 /*
414  * readahead MAX_RA_NODE number of node pages.
415  */
416 static void ra_node_pages(struct page *parent, int start, int n)
417 {
418 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
419 	struct blk_plug plug;
420 	int i, end;
421 	nid_t nid;
422 
423 	blk_start_plug(&plug);
424 
425 	/* Then, try readahead for siblings of the desired node */
426 	end = start + n;
427 	end = min(end, NIDS_PER_BLOCK);
428 	for (i = start; i < end; i++) {
429 		nid = get_nid(parent, i, false);
430 		ra_node_page(sbi, nid);
431 	}
432 
433 	blk_finish_plug(&plug);
434 }
435 
436 pgoff_t get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
437 {
438 	const long direct_index = ADDRS_PER_INODE(dn->inode);
439 	const long direct_blks = ADDRS_PER_BLOCK;
440 	const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
441 	unsigned int skipped_unit = ADDRS_PER_BLOCK;
442 	int cur_level = dn->cur_level;
443 	int max_level = dn->max_level;
444 	pgoff_t base = 0;
445 
446 	if (!dn->max_level)
447 		return pgofs + 1;
448 
449 	while (max_level-- > cur_level)
450 		skipped_unit *= NIDS_PER_BLOCK;
451 
452 	switch (dn->max_level) {
453 	case 3:
454 		base += 2 * indirect_blks;
455 	case 2:
456 		base += 2 * direct_blks;
457 	case 1:
458 		base += direct_index;
459 		break;
460 	default:
461 		f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
462 	}
463 
464 	return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
465 }
466 
467 /*
468  * The maximum depth is four.
469  * Offset[0] will have raw inode offset.
470  */
471 static int get_node_path(struct inode *inode, long block,
472 				int offset[4], unsigned int noffset[4])
473 {
474 	const long direct_index = ADDRS_PER_INODE(inode);
475 	const long direct_blks = ADDRS_PER_BLOCK;
476 	const long dptrs_per_blk = NIDS_PER_BLOCK;
477 	const long indirect_blks = ADDRS_PER_BLOCK * NIDS_PER_BLOCK;
478 	const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
479 	int n = 0;
480 	int level = 0;
481 
482 	noffset[0] = 0;
483 
484 	if (block < direct_index) {
485 		offset[n] = block;
486 		goto got;
487 	}
488 	block -= direct_index;
489 	if (block < direct_blks) {
490 		offset[n++] = NODE_DIR1_BLOCK;
491 		noffset[n] = 1;
492 		offset[n] = block;
493 		level = 1;
494 		goto got;
495 	}
496 	block -= direct_blks;
497 	if (block < direct_blks) {
498 		offset[n++] = NODE_DIR2_BLOCK;
499 		noffset[n] = 2;
500 		offset[n] = block;
501 		level = 1;
502 		goto got;
503 	}
504 	block -= direct_blks;
505 	if (block < indirect_blks) {
506 		offset[n++] = NODE_IND1_BLOCK;
507 		noffset[n] = 3;
508 		offset[n++] = block / direct_blks;
509 		noffset[n] = 4 + offset[n - 1];
510 		offset[n] = block % direct_blks;
511 		level = 2;
512 		goto got;
513 	}
514 	block -= indirect_blks;
515 	if (block < indirect_blks) {
516 		offset[n++] = NODE_IND2_BLOCK;
517 		noffset[n] = 4 + dptrs_per_blk;
518 		offset[n++] = block / direct_blks;
519 		noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
520 		offset[n] = block % direct_blks;
521 		level = 2;
522 		goto got;
523 	}
524 	block -= indirect_blks;
525 	if (block < dindirect_blks) {
526 		offset[n++] = NODE_DIND_BLOCK;
527 		noffset[n] = 5 + (dptrs_per_blk * 2);
528 		offset[n++] = block / indirect_blks;
529 		noffset[n] = 6 + (dptrs_per_blk * 2) +
530 			      offset[n - 1] * (dptrs_per_blk + 1);
531 		offset[n++] = (block / direct_blks) % dptrs_per_blk;
532 		noffset[n] = 7 + (dptrs_per_blk * 2) +
533 			      offset[n - 2] * (dptrs_per_blk + 1) +
534 			      offset[n - 1];
535 		offset[n] = block % direct_blks;
536 		level = 3;
537 		goto got;
538 	} else {
539 		BUG();
540 	}
541 got:
542 	return level;
543 }
544 
545 /*
546  * Caller should call f2fs_put_dnode(dn).
547  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
548  * f2fs_unlock_op() only if ro is not set RDONLY_NODE.
549  * In the case of RDONLY_NODE, we don't need to care about mutex.
550  */
551 int get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
552 {
553 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
554 	struct page *npage[4];
555 	struct page *parent = NULL;
556 	int offset[4];
557 	unsigned int noffset[4];
558 	nid_t nids[4];
559 	int level, i = 0;
560 	int err = 0;
561 
562 	level = get_node_path(dn->inode, index, offset, noffset);
563 
564 	nids[0] = dn->inode->i_ino;
565 	npage[0] = dn->inode_page;
566 
567 	if (!npage[0]) {
568 		npage[0] = get_node_page(sbi, nids[0]);
569 		if (IS_ERR(npage[0]))
570 			return PTR_ERR(npage[0]);
571 	}
572 
573 	/* if inline_data is set, should not report any block indices */
574 	if (f2fs_has_inline_data(dn->inode) && index) {
575 		err = -ENOENT;
576 		f2fs_put_page(npage[0], 1);
577 		goto release_out;
578 	}
579 
580 	parent = npage[0];
581 	if (level != 0)
582 		nids[1] = get_nid(parent, offset[0], true);
583 	dn->inode_page = npage[0];
584 	dn->inode_page_locked = true;
585 
586 	/* get indirect or direct nodes */
587 	for (i = 1; i <= level; i++) {
588 		bool done = false;
589 
590 		if (!nids[i] && mode == ALLOC_NODE) {
591 			/* alloc new node */
592 			if (!alloc_nid(sbi, &(nids[i]))) {
593 				err = -ENOSPC;
594 				goto release_pages;
595 			}
596 
597 			dn->nid = nids[i];
598 			npage[i] = new_node_page(dn, noffset[i], NULL);
599 			if (IS_ERR(npage[i])) {
600 				alloc_nid_failed(sbi, nids[i]);
601 				err = PTR_ERR(npage[i]);
602 				goto release_pages;
603 			}
604 
605 			set_nid(parent, offset[i - 1], nids[i], i == 1);
606 			alloc_nid_done(sbi, nids[i]);
607 			done = true;
608 		} else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
609 			npage[i] = get_node_page_ra(parent, offset[i - 1]);
610 			if (IS_ERR(npage[i])) {
611 				err = PTR_ERR(npage[i]);
612 				goto release_pages;
613 			}
614 			done = true;
615 		}
616 		if (i == 1) {
617 			dn->inode_page_locked = false;
618 			unlock_page(parent);
619 		} else {
620 			f2fs_put_page(parent, 1);
621 		}
622 
623 		if (!done) {
624 			npage[i] = get_node_page(sbi, nids[i]);
625 			if (IS_ERR(npage[i])) {
626 				err = PTR_ERR(npage[i]);
627 				f2fs_put_page(npage[0], 0);
628 				goto release_out;
629 			}
630 		}
631 		if (i < level) {
632 			parent = npage[i];
633 			nids[i + 1] = get_nid(parent, offset[i], false);
634 		}
635 	}
636 	dn->nid = nids[level];
637 	dn->ofs_in_node = offset[level];
638 	dn->node_page = npage[level];
639 	dn->data_blkaddr = datablock_addr(dn->node_page, dn->ofs_in_node);
640 	return 0;
641 
642 release_pages:
643 	f2fs_put_page(parent, 1);
644 	if (i > 1)
645 		f2fs_put_page(npage[0], 0);
646 release_out:
647 	dn->inode_page = NULL;
648 	dn->node_page = NULL;
649 	if (err == -ENOENT) {
650 		dn->cur_level = i;
651 		dn->max_level = level;
652 		dn->ofs_in_node = offset[level];
653 	}
654 	return err;
655 }
656 
657 static void truncate_node(struct dnode_of_data *dn)
658 {
659 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
660 	struct node_info ni;
661 
662 	get_node_info(sbi, dn->nid, &ni);
663 	if (dn->inode->i_blocks == 0) {
664 		f2fs_bug_on(sbi, ni.blk_addr != NULL_ADDR);
665 		goto invalidate;
666 	}
667 	f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
668 
669 	/* Deallocate node address */
670 	invalidate_blocks(sbi, ni.blk_addr);
671 	dec_valid_node_count(sbi, dn->inode);
672 	set_node_addr(sbi, &ni, NULL_ADDR, false);
673 
674 	if (dn->nid == dn->inode->i_ino) {
675 		remove_orphan_inode(sbi, dn->nid);
676 		dec_valid_inode_count(sbi);
677 		f2fs_inode_synced(dn->inode);
678 	}
679 invalidate:
680 	clear_node_page_dirty(dn->node_page);
681 	set_sbi_flag(sbi, SBI_IS_DIRTY);
682 
683 	f2fs_put_page(dn->node_page, 1);
684 
685 	invalidate_mapping_pages(NODE_MAPPING(sbi),
686 			dn->node_page->index, dn->node_page->index);
687 
688 	dn->node_page = NULL;
689 	trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
690 }
691 
692 static int truncate_dnode(struct dnode_of_data *dn)
693 {
694 	struct page *page;
695 
696 	if (dn->nid == 0)
697 		return 1;
698 
699 	/* get direct node */
700 	page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
701 	if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
702 		return 1;
703 	else if (IS_ERR(page))
704 		return PTR_ERR(page);
705 
706 	/* Make dnode_of_data for parameter */
707 	dn->node_page = page;
708 	dn->ofs_in_node = 0;
709 	truncate_data_blocks(dn);
710 	truncate_node(dn);
711 	return 1;
712 }
713 
714 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
715 						int ofs, int depth)
716 {
717 	struct dnode_of_data rdn = *dn;
718 	struct page *page;
719 	struct f2fs_node *rn;
720 	nid_t child_nid;
721 	unsigned int child_nofs;
722 	int freed = 0;
723 	int i, ret;
724 
725 	if (dn->nid == 0)
726 		return NIDS_PER_BLOCK + 1;
727 
728 	trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
729 
730 	page = get_node_page(F2FS_I_SB(dn->inode), dn->nid);
731 	if (IS_ERR(page)) {
732 		trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
733 		return PTR_ERR(page);
734 	}
735 
736 	ra_node_pages(page, ofs, NIDS_PER_BLOCK);
737 
738 	rn = F2FS_NODE(page);
739 	if (depth < 3) {
740 		for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
741 			child_nid = le32_to_cpu(rn->in.nid[i]);
742 			if (child_nid == 0)
743 				continue;
744 			rdn.nid = child_nid;
745 			ret = truncate_dnode(&rdn);
746 			if (ret < 0)
747 				goto out_err;
748 			if (set_nid(page, i, 0, false))
749 				dn->node_changed = true;
750 		}
751 	} else {
752 		child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
753 		for (i = ofs; i < NIDS_PER_BLOCK; i++) {
754 			child_nid = le32_to_cpu(rn->in.nid[i]);
755 			if (child_nid == 0) {
756 				child_nofs += NIDS_PER_BLOCK + 1;
757 				continue;
758 			}
759 			rdn.nid = child_nid;
760 			ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
761 			if (ret == (NIDS_PER_BLOCK + 1)) {
762 				if (set_nid(page, i, 0, false))
763 					dn->node_changed = true;
764 				child_nofs += ret;
765 			} else if (ret < 0 && ret != -ENOENT) {
766 				goto out_err;
767 			}
768 		}
769 		freed = child_nofs;
770 	}
771 
772 	if (!ofs) {
773 		/* remove current indirect node */
774 		dn->node_page = page;
775 		truncate_node(dn);
776 		freed++;
777 	} else {
778 		f2fs_put_page(page, 1);
779 	}
780 	trace_f2fs_truncate_nodes_exit(dn->inode, freed);
781 	return freed;
782 
783 out_err:
784 	f2fs_put_page(page, 1);
785 	trace_f2fs_truncate_nodes_exit(dn->inode, ret);
786 	return ret;
787 }
788 
789 static int truncate_partial_nodes(struct dnode_of_data *dn,
790 			struct f2fs_inode *ri, int *offset, int depth)
791 {
792 	struct page *pages[2];
793 	nid_t nid[3];
794 	nid_t child_nid;
795 	int err = 0;
796 	int i;
797 	int idx = depth - 2;
798 
799 	nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
800 	if (!nid[0])
801 		return 0;
802 
803 	/* get indirect nodes in the path */
804 	for (i = 0; i < idx + 1; i++) {
805 		/* reference count'll be increased */
806 		pages[i] = get_node_page(F2FS_I_SB(dn->inode), nid[i]);
807 		if (IS_ERR(pages[i])) {
808 			err = PTR_ERR(pages[i]);
809 			idx = i - 1;
810 			goto fail;
811 		}
812 		nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
813 	}
814 
815 	ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
816 
817 	/* free direct nodes linked to a partial indirect node */
818 	for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
819 		child_nid = get_nid(pages[idx], i, false);
820 		if (!child_nid)
821 			continue;
822 		dn->nid = child_nid;
823 		err = truncate_dnode(dn);
824 		if (err < 0)
825 			goto fail;
826 		if (set_nid(pages[idx], i, 0, false))
827 			dn->node_changed = true;
828 	}
829 
830 	if (offset[idx + 1] == 0) {
831 		dn->node_page = pages[idx];
832 		dn->nid = nid[idx];
833 		truncate_node(dn);
834 	} else {
835 		f2fs_put_page(pages[idx], 1);
836 	}
837 	offset[idx]++;
838 	offset[idx + 1] = 0;
839 	idx--;
840 fail:
841 	for (i = idx; i >= 0; i--)
842 		f2fs_put_page(pages[i], 1);
843 
844 	trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
845 
846 	return err;
847 }
848 
849 /*
850  * All the block addresses of data and nodes should be nullified.
851  */
852 int truncate_inode_blocks(struct inode *inode, pgoff_t from)
853 {
854 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
855 	int err = 0, cont = 1;
856 	int level, offset[4], noffset[4];
857 	unsigned int nofs = 0;
858 	struct f2fs_inode *ri;
859 	struct dnode_of_data dn;
860 	struct page *page;
861 
862 	trace_f2fs_truncate_inode_blocks_enter(inode, from);
863 
864 	level = get_node_path(inode, from, offset, noffset);
865 
866 	page = get_node_page(sbi, inode->i_ino);
867 	if (IS_ERR(page)) {
868 		trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
869 		return PTR_ERR(page);
870 	}
871 
872 	set_new_dnode(&dn, inode, page, NULL, 0);
873 	unlock_page(page);
874 
875 	ri = F2FS_INODE(page);
876 	switch (level) {
877 	case 0:
878 	case 1:
879 		nofs = noffset[1];
880 		break;
881 	case 2:
882 		nofs = noffset[1];
883 		if (!offset[level - 1])
884 			goto skip_partial;
885 		err = truncate_partial_nodes(&dn, ri, offset, level);
886 		if (err < 0 && err != -ENOENT)
887 			goto fail;
888 		nofs += 1 + NIDS_PER_BLOCK;
889 		break;
890 	case 3:
891 		nofs = 5 + 2 * NIDS_PER_BLOCK;
892 		if (!offset[level - 1])
893 			goto skip_partial;
894 		err = truncate_partial_nodes(&dn, ri, offset, level);
895 		if (err < 0 && err != -ENOENT)
896 			goto fail;
897 		break;
898 	default:
899 		BUG();
900 	}
901 
902 skip_partial:
903 	while (cont) {
904 		dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
905 		switch (offset[0]) {
906 		case NODE_DIR1_BLOCK:
907 		case NODE_DIR2_BLOCK:
908 			err = truncate_dnode(&dn);
909 			break;
910 
911 		case NODE_IND1_BLOCK:
912 		case NODE_IND2_BLOCK:
913 			err = truncate_nodes(&dn, nofs, offset[1], 2);
914 			break;
915 
916 		case NODE_DIND_BLOCK:
917 			err = truncate_nodes(&dn, nofs, offset[1], 3);
918 			cont = 0;
919 			break;
920 
921 		default:
922 			BUG();
923 		}
924 		if (err < 0 && err != -ENOENT)
925 			goto fail;
926 		if (offset[1] == 0 &&
927 				ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
928 			lock_page(page);
929 			BUG_ON(page->mapping != NODE_MAPPING(sbi));
930 			f2fs_wait_on_page_writeback(page, NODE, true);
931 			ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
932 			set_page_dirty(page);
933 			unlock_page(page);
934 		}
935 		offset[1] = 0;
936 		offset[0]++;
937 		nofs += err;
938 	}
939 fail:
940 	f2fs_put_page(page, 0);
941 	trace_f2fs_truncate_inode_blocks_exit(inode, err);
942 	return err > 0 ? 0 : err;
943 }
944 
945 int truncate_xattr_node(struct inode *inode, struct page *page)
946 {
947 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
948 	nid_t nid = F2FS_I(inode)->i_xattr_nid;
949 	struct dnode_of_data dn;
950 	struct page *npage;
951 
952 	if (!nid)
953 		return 0;
954 
955 	npage = get_node_page(sbi, nid);
956 	if (IS_ERR(npage))
957 		return PTR_ERR(npage);
958 
959 	f2fs_i_xnid_write(inode, 0);
960 
961 	/* need to do checkpoint during fsync */
962 	F2FS_I(inode)->xattr_ver = cur_cp_version(F2FS_CKPT(sbi));
963 
964 	set_new_dnode(&dn, inode, page, npage, nid);
965 
966 	if (page)
967 		dn.inode_page_locked = true;
968 	truncate_node(&dn);
969 	return 0;
970 }
971 
972 /*
973  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
974  * f2fs_unlock_op().
975  */
976 int remove_inode_page(struct inode *inode)
977 {
978 	struct dnode_of_data dn;
979 	int err;
980 
981 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
982 	err = get_dnode_of_data(&dn, 0, LOOKUP_NODE);
983 	if (err)
984 		return err;
985 
986 	err = truncate_xattr_node(inode, dn.inode_page);
987 	if (err) {
988 		f2fs_put_dnode(&dn);
989 		return err;
990 	}
991 
992 	/* remove potential inline_data blocks */
993 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
994 				S_ISLNK(inode->i_mode))
995 		truncate_data_blocks_range(&dn, 1);
996 
997 	/* 0 is possible, after f2fs_new_inode() has failed */
998 	f2fs_bug_on(F2FS_I_SB(inode),
999 			inode->i_blocks != 0 && inode->i_blocks != 1);
1000 
1001 	/* will put inode & node pages */
1002 	truncate_node(&dn);
1003 	return 0;
1004 }
1005 
1006 struct page *new_inode_page(struct inode *inode)
1007 {
1008 	struct dnode_of_data dn;
1009 
1010 	/* allocate inode page for new inode */
1011 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1012 
1013 	/* caller should f2fs_put_page(page, 1); */
1014 	return new_node_page(&dn, 0, NULL);
1015 }
1016 
1017 struct page *new_node_page(struct dnode_of_data *dn,
1018 				unsigned int ofs, struct page *ipage)
1019 {
1020 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1021 	struct node_info old_ni, new_ni;
1022 	struct page *page;
1023 	int err;
1024 
1025 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1026 		return ERR_PTR(-EPERM);
1027 
1028 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1029 	if (!page)
1030 		return ERR_PTR(-ENOMEM);
1031 
1032 	if (unlikely(!inc_valid_node_count(sbi, dn->inode))) {
1033 		err = -ENOSPC;
1034 		goto fail;
1035 	}
1036 
1037 	get_node_info(sbi, dn->nid, &old_ni);
1038 
1039 	/* Reinitialize old_ni with new node page */
1040 	f2fs_bug_on(sbi, old_ni.blk_addr != NULL_ADDR);
1041 	new_ni = old_ni;
1042 	new_ni.ino = dn->inode->i_ino;
1043 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1044 
1045 	f2fs_wait_on_page_writeback(page, NODE, true);
1046 	fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1047 	set_cold_node(dn->inode, page);
1048 	if (!PageUptodate(page))
1049 		SetPageUptodate(page);
1050 	if (set_page_dirty(page))
1051 		dn->node_changed = true;
1052 
1053 	if (f2fs_has_xattr_block(ofs))
1054 		f2fs_i_xnid_write(dn->inode, dn->nid);
1055 
1056 	if (ofs == 0)
1057 		inc_valid_inode_count(sbi);
1058 	return page;
1059 
1060 fail:
1061 	clear_node_page_dirty(page);
1062 	f2fs_put_page(page, 1);
1063 	return ERR_PTR(err);
1064 }
1065 
1066 /*
1067  * Caller should do after getting the following values.
1068  * 0: f2fs_put_page(page, 0)
1069  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1070  */
1071 static int read_node_page(struct page *page, int op_flags)
1072 {
1073 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1074 	struct node_info ni;
1075 	struct f2fs_io_info fio = {
1076 		.sbi = sbi,
1077 		.type = NODE,
1078 		.op = REQ_OP_READ,
1079 		.op_flags = op_flags,
1080 		.page = page,
1081 		.encrypted_page = NULL,
1082 	};
1083 
1084 	if (PageUptodate(page))
1085 		return LOCKED_PAGE;
1086 
1087 	get_node_info(sbi, page->index, &ni);
1088 
1089 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1090 		ClearPageUptodate(page);
1091 		return -ENOENT;
1092 	}
1093 
1094 	fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1095 	return f2fs_submit_page_bio(&fio);
1096 }
1097 
1098 /*
1099  * Readahead a node page
1100  */
1101 void ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1102 {
1103 	struct page *apage;
1104 	int err;
1105 
1106 	if (!nid)
1107 		return;
1108 	f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1109 
1110 	rcu_read_lock();
1111 	apage = radix_tree_lookup(&NODE_MAPPING(sbi)->page_tree, nid);
1112 	rcu_read_unlock();
1113 	if (apage)
1114 		return;
1115 
1116 	apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1117 	if (!apage)
1118 		return;
1119 
1120 	err = read_node_page(apage, REQ_RAHEAD);
1121 	f2fs_put_page(apage, err ? 1 : 0);
1122 }
1123 
1124 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1125 					struct page *parent, int start)
1126 {
1127 	struct page *page;
1128 	int err;
1129 
1130 	if (!nid)
1131 		return ERR_PTR(-ENOENT);
1132 	f2fs_bug_on(sbi, check_nid_range(sbi, nid));
1133 repeat:
1134 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1135 	if (!page)
1136 		return ERR_PTR(-ENOMEM);
1137 
1138 	err = read_node_page(page, READ_SYNC);
1139 	if (err < 0) {
1140 		f2fs_put_page(page, 1);
1141 		return ERR_PTR(err);
1142 	} else if (err == LOCKED_PAGE) {
1143 		goto page_hit;
1144 	}
1145 
1146 	if (parent)
1147 		ra_node_pages(parent, start + 1, MAX_RA_NODE);
1148 
1149 	lock_page(page);
1150 
1151 	if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1152 		f2fs_put_page(page, 1);
1153 		goto repeat;
1154 	}
1155 
1156 	if (unlikely(!PageUptodate(page)))
1157 		goto out_err;
1158 page_hit:
1159 	if(unlikely(nid != nid_of_node(page))) {
1160 		f2fs_bug_on(sbi, 1);
1161 		ClearPageUptodate(page);
1162 out_err:
1163 		f2fs_put_page(page, 1);
1164 		return ERR_PTR(-EIO);
1165 	}
1166 	return page;
1167 }
1168 
1169 struct page *get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1170 {
1171 	return __get_node_page(sbi, nid, NULL, 0);
1172 }
1173 
1174 struct page *get_node_page_ra(struct page *parent, int start)
1175 {
1176 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1177 	nid_t nid = get_nid(parent, start, false);
1178 
1179 	return __get_node_page(sbi, nid, parent, start);
1180 }
1181 
1182 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1183 {
1184 	struct inode *inode;
1185 	struct page *page;
1186 	int ret;
1187 
1188 	/* should flush inline_data before evict_inode */
1189 	inode = ilookup(sbi->sb, ino);
1190 	if (!inode)
1191 		return;
1192 
1193 	page = pagecache_get_page(inode->i_mapping, 0, FGP_LOCK|FGP_NOWAIT, 0);
1194 	if (!page)
1195 		goto iput_out;
1196 
1197 	if (!PageUptodate(page))
1198 		goto page_out;
1199 
1200 	if (!PageDirty(page))
1201 		goto page_out;
1202 
1203 	if (!clear_page_dirty_for_io(page))
1204 		goto page_out;
1205 
1206 	ret = f2fs_write_inline_data(inode, page);
1207 	inode_dec_dirty_pages(inode);
1208 	if (ret)
1209 		set_page_dirty(page);
1210 page_out:
1211 	f2fs_put_page(page, 1);
1212 iput_out:
1213 	iput(inode);
1214 }
1215 
1216 void move_node_page(struct page *node_page, int gc_type)
1217 {
1218 	if (gc_type == FG_GC) {
1219 		struct f2fs_sb_info *sbi = F2FS_P_SB(node_page);
1220 		struct writeback_control wbc = {
1221 			.sync_mode = WB_SYNC_ALL,
1222 			.nr_to_write = 1,
1223 			.for_reclaim = 0,
1224 		};
1225 
1226 		set_page_dirty(node_page);
1227 		f2fs_wait_on_page_writeback(node_page, NODE, true);
1228 
1229 		f2fs_bug_on(sbi, PageWriteback(node_page));
1230 		if (!clear_page_dirty_for_io(node_page))
1231 			goto out_page;
1232 
1233 		if (NODE_MAPPING(sbi)->a_ops->writepage(node_page, &wbc))
1234 			unlock_page(node_page);
1235 		goto release_page;
1236 	} else {
1237 		/* set page dirty and write it */
1238 		if (!PageWriteback(node_page))
1239 			set_page_dirty(node_page);
1240 	}
1241 out_page:
1242 	unlock_page(node_page);
1243 release_page:
1244 	f2fs_put_page(node_page, 0);
1245 }
1246 
1247 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1248 {
1249 	pgoff_t index, end;
1250 	struct pagevec pvec;
1251 	struct page *last_page = NULL;
1252 
1253 	pagevec_init(&pvec, 0);
1254 	index = 0;
1255 	end = ULONG_MAX;
1256 
1257 	while (index <= end) {
1258 		int i, nr_pages;
1259 		nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1260 				PAGECACHE_TAG_DIRTY,
1261 				min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1262 		if (nr_pages == 0)
1263 			break;
1264 
1265 		for (i = 0; i < nr_pages; i++) {
1266 			struct page *page = pvec.pages[i];
1267 
1268 			if (unlikely(f2fs_cp_error(sbi))) {
1269 				f2fs_put_page(last_page, 0);
1270 				pagevec_release(&pvec);
1271 				return ERR_PTR(-EIO);
1272 			}
1273 
1274 			if (!IS_DNODE(page) || !is_cold_node(page))
1275 				continue;
1276 			if (ino_of_node(page) != ino)
1277 				continue;
1278 
1279 			lock_page(page);
1280 
1281 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1282 continue_unlock:
1283 				unlock_page(page);
1284 				continue;
1285 			}
1286 			if (ino_of_node(page) != ino)
1287 				goto continue_unlock;
1288 
1289 			if (!PageDirty(page)) {
1290 				/* someone wrote it for us */
1291 				goto continue_unlock;
1292 			}
1293 
1294 			if (last_page)
1295 				f2fs_put_page(last_page, 0);
1296 
1297 			get_page(page);
1298 			last_page = page;
1299 			unlock_page(page);
1300 		}
1301 		pagevec_release(&pvec);
1302 		cond_resched();
1303 	}
1304 	return last_page;
1305 }
1306 
1307 int fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1308 			struct writeback_control *wbc, bool atomic)
1309 {
1310 	pgoff_t index, end;
1311 	struct pagevec pvec;
1312 	int ret = 0;
1313 	struct page *last_page = NULL;
1314 	bool marked = false;
1315 	nid_t ino = inode->i_ino;
1316 
1317 	if (atomic) {
1318 		last_page = last_fsync_dnode(sbi, ino);
1319 		if (IS_ERR_OR_NULL(last_page))
1320 			return PTR_ERR_OR_ZERO(last_page);
1321 	}
1322 retry:
1323 	pagevec_init(&pvec, 0);
1324 	index = 0;
1325 	end = ULONG_MAX;
1326 
1327 	while (index <= end) {
1328 		int i, nr_pages;
1329 		nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1330 				PAGECACHE_TAG_DIRTY,
1331 				min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1332 		if (nr_pages == 0)
1333 			break;
1334 
1335 		for (i = 0; i < nr_pages; i++) {
1336 			struct page *page = pvec.pages[i];
1337 
1338 			if (unlikely(f2fs_cp_error(sbi))) {
1339 				f2fs_put_page(last_page, 0);
1340 				pagevec_release(&pvec);
1341 				return -EIO;
1342 			}
1343 
1344 			if (!IS_DNODE(page) || !is_cold_node(page))
1345 				continue;
1346 			if (ino_of_node(page) != ino)
1347 				continue;
1348 
1349 			lock_page(page);
1350 
1351 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1352 continue_unlock:
1353 				unlock_page(page);
1354 				continue;
1355 			}
1356 			if (ino_of_node(page) != ino)
1357 				goto continue_unlock;
1358 
1359 			if (!PageDirty(page) && page != last_page) {
1360 				/* someone wrote it for us */
1361 				goto continue_unlock;
1362 			}
1363 
1364 			f2fs_wait_on_page_writeback(page, NODE, true);
1365 			BUG_ON(PageWriteback(page));
1366 
1367 			if (!atomic || page == last_page) {
1368 				set_fsync_mark(page, 1);
1369 				if (IS_INODE(page)) {
1370 					if (is_inode_flag_set(inode,
1371 								FI_DIRTY_INODE))
1372 						update_inode(inode, page);
1373 					set_dentry_mark(page,
1374 						need_dentry_mark(sbi, ino));
1375 				}
1376 				/*  may be written by other thread */
1377 				if (!PageDirty(page))
1378 					set_page_dirty(page);
1379 			}
1380 
1381 			if (!clear_page_dirty_for_io(page))
1382 				goto continue_unlock;
1383 
1384 			ret = NODE_MAPPING(sbi)->a_ops->writepage(page, wbc);
1385 			if (ret) {
1386 				unlock_page(page);
1387 				f2fs_put_page(last_page, 0);
1388 				break;
1389 			}
1390 			if (page == last_page) {
1391 				f2fs_put_page(page, 0);
1392 				marked = true;
1393 				break;
1394 			}
1395 		}
1396 		pagevec_release(&pvec);
1397 		cond_resched();
1398 
1399 		if (ret || marked)
1400 			break;
1401 	}
1402 	if (!ret && atomic && !marked) {
1403 		f2fs_msg(sbi->sb, KERN_DEBUG,
1404 			"Retry to write fsync mark: ino=%u, idx=%lx",
1405 					ino, last_page->index);
1406 		lock_page(last_page);
1407 		set_page_dirty(last_page);
1408 		unlock_page(last_page);
1409 		goto retry;
1410 	}
1411 	return ret ? -EIO: 0;
1412 }
1413 
1414 int sync_node_pages(struct f2fs_sb_info *sbi, struct writeback_control *wbc)
1415 {
1416 	pgoff_t index, end;
1417 	struct pagevec pvec;
1418 	int step = 0;
1419 	int nwritten = 0;
1420 
1421 	pagevec_init(&pvec, 0);
1422 
1423 next_step:
1424 	index = 0;
1425 	end = ULONG_MAX;
1426 
1427 	while (index <= end) {
1428 		int i, nr_pages;
1429 		nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1430 				PAGECACHE_TAG_DIRTY,
1431 				min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1432 		if (nr_pages == 0)
1433 			break;
1434 
1435 		for (i = 0; i < nr_pages; i++) {
1436 			struct page *page = pvec.pages[i];
1437 
1438 			if (unlikely(f2fs_cp_error(sbi))) {
1439 				pagevec_release(&pvec);
1440 				return -EIO;
1441 			}
1442 
1443 			/*
1444 			 * flushing sequence with step:
1445 			 * 0. indirect nodes
1446 			 * 1. dentry dnodes
1447 			 * 2. file dnodes
1448 			 */
1449 			if (step == 0 && IS_DNODE(page))
1450 				continue;
1451 			if (step == 1 && (!IS_DNODE(page) ||
1452 						is_cold_node(page)))
1453 				continue;
1454 			if (step == 2 && (!IS_DNODE(page) ||
1455 						!is_cold_node(page)))
1456 				continue;
1457 lock_node:
1458 			if (!trylock_page(page))
1459 				continue;
1460 
1461 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1462 continue_unlock:
1463 				unlock_page(page);
1464 				continue;
1465 			}
1466 
1467 			if (!PageDirty(page)) {
1468 				/* someone wrote it for us */
1469 				goto continue_unlock;
1470 			}
1471 
1472 			/* flush inline_data */
1473 			if (is_inline_node(page)) {
1474 				clear_inline_node(page);
1475 				unlock_page(page);
1476 				flush_inline_data(sbi, ino_of_node(page));
1477 				goto lock_node;
1478 			}
1479 
1480 			f2fs_wait_on_page_writeback(page, NODE, true);
1481 
1482 			BUG_ON(PageWriteback(page));
1483 			if (!clear_page_dirty_for_io(page))
1484 				goto continue_unlock;
1485 
1486 			set_fsync_mark(page, 0);
1487 			set_dentry_mark(page, 0);
1488 
1489 			if (NODE_MAPPING(sbi)->a_ops->writepage(page, wbc))
1490 				unlock_page(page);
1491 
1492 			if (--wbc->nr_to_write == 0)
1493 				break;
1494 		}
1495 		pagevec_release(&pvec);
1496 		cond_resched();
1497 
1498 		if (wbc->nr_to_write == 0) {
1499 			step = 2;
1500 			break;
1501 		}
1502 	}
1503 
1504 	if (step < 2) {
1505 		step++;
1506 		goto next_step;
1507 	}
1508 	return nwritten;
1509 }
1510 
1511 int wait_on_node_pages_writeback(struct f2fs_sb_info *sbi, nid_t ino)
1512 {
1513 	pgoff_t index = 0, end = ULONG_MAX;
1514 	struct pagevec pvec;
1515 	int ret2 = 0, ret = 0;
1516 
1517 	pagevec_init(&pvec, 0);
1518 
1519 	while (index <= end) {
1520 		int i, nr_pages;
1521 		nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1522 				PAGECACHE_TAG_WRITEBACK,
1523 				min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1);
1524 		if (nr_pages == 0)
1525 			break;
1526 
1527 		for (i = 0; i < nr_pages; i++) {
1528 			struct page *page = pvec.pages[i];
1529 
1530 			/* until radix tree lookup accepts end_index */
1531 			if (unlikely(page->index > end))
1532 				continue;
1533 
1534 			if (ino && ino_of_node(page) == ino) {
1535 				f2fs_wait_on_page_writeback(page, NODE, true);
1536 				if (TestClearPageError(page))
1537 					ret = -EIO;
1538 			}
1539 		}
1540 		pagevec_release(&pvec);
1541 		cond_resched();
1542 	}
1543 
1544 	if (unlikely(test_and_clear_bit(AS_ENOSPC, &NODE_MAPPING(sbi)->flags)))
1545 		ret2 = -ENOSPC;
1546 	if (unlikely(test_and_clear_bit(AS_EIO, &NODE_MAPPING(sbi)->flags)))
1547 		ret2 = -EIO;
1548 	if (!ret)
1549 		ret = ret2;
1550 	return ret;
1551 }
1552 
1553 static int f2fs_write_node_page(struct page *page,
1554 				struct writeback_control *wbc)
1555 {
1556 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1557 	nid_t nid;
1558 	struct node_info ni;
1559 	struct f2fs_io_info fio = {
1560 		.sbi = sbi,
1561 		.type = NODE,
1562 		.op = REQ_OP_WRITE,
1563 		.op_flags = (wbc->sync_mode == WB_SYNC_ALL) ? WRITE_SYNC : 0,
1564 		.page = page,
1565 		.encrypted_page = NULL,
1566 	};
1567 
1568 	trace_f2fs_writepage(page, NODE);
1569 
1570 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1571 		goto redirty_out;
1572 	if (unlikely(f2fs_cp_error(sbi)))
1573 		goto redirty_out;
1574 
1575 	/* get old block addr of this node page */
1576 	nid = nid_of_node(page);
1577 	f2fs_bug_on(sbi, page->index != nid);
1578 
1579 	if (wbc->for_reclaim) {
1580 		if (!down_read_trylock(&sbi->node_write))
1581 			goto redirty_out;
1582 	} else {
1583 		down_read(&sbi->node_write);
1584 	}
1585 
1586 	get_node_info(sbi, nid, &ni);
1587 
1588 	/* This page is already truncated */
1589 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1590 		ClearPageUptodate(page);
1591 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1592 		up_read(&sbi->node_write);
1593 		unlock_page(page);
1594 		return 0;
1595 	}
1596 
1597 	set_page_writeback(page);
1598 	fio.old_blkaddr = ni.blk_addr;
1599 	write_node_page(nid, &fio);
1600 	set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1601 	dec_page_count(sbi, F2FS_DIRTY_NODES);
1602 	up_read(&sbi->node_write);
1603 
1604 	if (wbc->for_reclaim)
1605 		f2fs_submit_merged_bio_cond(sbi, NULL, page, 0, NODE, WRITE);
1606 
1607 	unlock_page(page);
1608 
1609 	if (unlikely(f2fs_cp_error(sbi)))
1610 		f2fs_submit_merged_bio(sbi, NODE, WRITE);
1611 
1612 	return 0;
1613 
1614 redirty_out:
1615 	redirty_page_for_writepage(wbc, page);
1616 	return AOP_WRITEPAGE_ACTIVATE;
1617 }
1618 
1619 static int f2fs_write_node_pages(struct address_space *mapping,
1620 			    struct writeback_control *wbc)
1621 {
1622 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
1623 	struct blk_plug plug;
1624 	long diff;
1625 
1626 	/* balancing f2fs's metadata in background */
1627 	f2fs_balance_fs_bg(sbi);
1628 
1629 	/* collect a number of dirty node pages and write together */
1630 	if (get_pages(sbi, F2FS_DIRTY_NODES) < nr_pages_to_skip(sbi, NODE))
1631 		goto skip_write;
1632 
1633 	trace_f2fs_writepages(mapping->host, wbc, NODE);
1634 
1635 	diff = nr_pages_to_write(sbi, NODE, wbc);
1636 	wbc->sync_mode = WB_SYNC_NONE;
1637 	blk_start_plug(&plug);
1638 	sync_node_pages(sbi, wbc);
1639 	blk_finish_plug(&plug);
1640 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
1641 	return 0;
1642 
1643 skip_write:
1644 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
1645 	trace_f2fs_writepages(mapping->host, wbc, NODE);
1646 	return 0;
1647 }
1648 
1649 static int f2fs_set_node_page_dirty(struct page *page)
1650 {
1651 	trace_f2fs_set_page_dirty(page, NODE);
1652 
1653 	if (!PageUptodate(page))
1654 		SetPageUptodate(page);
1655 	if (!PageDirty(page)) {
1656 		f2fs_set_page_dirty_nobuffers(page);
1657 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
1658 		SetPagePrivate(page);
1659 		f2fs_trace_pid(page);
1660 		return 1;
1661 	}
1662 	return 0;
1663 }
1664 
1665 /*
1666  * Structure of the f2fs node operations
1667  */
1668 const struct address_space_operations f2fs_node_aops = {
1669 	.writepage	= f2fs_write_node_page,
1670 	.writepages	= f2fs_write_node_pages,
1671 	.set_page_dirty	= f2fs_set_node_page_dirty,
1672 	.invalidatepage	= f2fs_invalidate_page,
1673 	.releasepage	= f2fs_release_page,
1674 };
1675 
1676 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
1677 						nid_t n)
1678 {
1679 	return radix_tree_lookup(&nm_i->free_nid_root, n);
1680 }
1681 
1682 static void __del_from_free_nid_list(struct f2fs_nm_info *nm_i,
1683 						struct free_nid *i)
1684 {
1685 	list_del(&i->list);
1686 	radix_tree_delete(&nm_i->free_nid_root, i->nid);
1687 }
1688 
1689 static int add_free_nid(struct f2fs_sb_info *sbi, nid_t nid, bool build)
1690 {
1691 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1692 	struct free_nid *i;
1693 	struct nat_entry *ne;
1694 
1695 	if (!available_free_memory(sbi, FREE_NIDS))
1696 		return -1;
1697 
1698 	/* 0 nid should not be used */
1699 	if (unlikely(nid == 0))
1700 		return 0;
1701 
1702 	if (build) {
1703 		/* do not add allocated nids */
1704 		ne = __lookup_nat_cache(nm_i, nid);
1705 		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
1706 				nat_get_blkaddr(ne) != NULL_ADDR))
1707 			return 0;
1708 	}
1709 
1710 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
1711 	i->nid = nid;
1712 	i->state = NID_NEW;
1713 
1714 	if (radix_tree_preload(GFP_NOFS)) {
1715 		kmem_cache_free(free_nid_slab, i);
1716 		return 0;
1717 	}
1718 
1719 	spin_lock(&nm_i->free_nid_list_lock);
1720 	if (radix_tree_insert(&nm_i->free_nid_root, i->nid, i)) {
1721 		spin_unlock(&nm_i->free_nid_list_lock);
1722 		radix_tree_preload_end();
1723 		kmem_cache_free(free_nid_slab, i);
1724 		return 0;
1725 	}
1726 	list_add_tail(&i->list, &nm_i->free_nid_list);
1727 	nm_i->fcnt++;
1728 	spin_unlock(&nm_i->free_nid_list_lock);
1729 	radix_tree_preload_end();
1730 	return 1;
1731 }
1732 
1733 static void remove_free_nid(struct f2fs_nm_info *nm_i, nid_t nid)
1734 {
1735 	struct free_nid *i;
1736 	bool need_free = false;
1737 
1738 	spin_lock(&nm_i->free_nid_list_lock);
1739 	i = __lookup_free_nid_list(nm_i, nid);
1740 	if (i && i->state == NID_NEW) {
1741 		__del_from_free_nid_list(nm_i, i);
1742 		nm_i->fcnt--;
1743 		need_free = true;
1744 	}
1745 	spin_unlock(&nm_i->free_nid_list_lock);
1746 
1747 	if (need_free)
1748 		kmem_cache_free(free_nid_slab, i);
1749 }
1750 
1751 static void scan_nat_page(struct f2fs_sb_info *sbi,
1752 			struct page *nat_page, nid_t start_nid)
1753 {
1754 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1755 	struct f2fs_nat_block *nat_blk = page_address(nat_page);
1756 	block_t blk_addr;
1757 	int i;
1758 
1759 	i = start_nid % NAT_ENTRY_PER_BLOCK;
1760 
1761 	for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
1762 
1763 		if (unlikely(start_nid >= nm_i->max_nid))
1764 			break;
1765 
1766 		blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
1767 		f2fs_bug_on(sbi, blk_addr == NEW_ADDR);
1768 		if (blk_addr == NULL_ADDR) {
1769 			if (add_free_nid(sbi, start_nid, true) < 0)
1770 				break;
1771 		}
1772 	}
1773 }
1774 
1775 void build_free_nids(struct f2fs_sb_info *sbi)
1776 {
1777 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1778 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
1779 	struct f2fs_journal *journal = curseg->journal;
1780 	int i = 0;
1781 	nid_t nid = nm_i->next_scan_nid;
1782 
1783 	/* Enough entries */
1784 	if (nm_i->fcnt >= NAT_ENTRY_PER_BLOCK)
1785 		return;
1786 
1787 	/* readahead nat pages to be scanned */
1788 	ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
1789 							META_NAT, true);
1790 
1791 	percpu_down_read(&nm_i->nat_tree_lock);
1792 
1793 	while (1) {
1794 		struct page *page = get_current_nat_page(sbi, nid);
1795 
1796 		scan_nat_page(sbi, page, nid);
1797 		f2fs_put_page(page, 1);
1798 
1799 		nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
1800 		if (unlikely(nid >= nm_i->max_nid))
1801 			nid = 0;
1802 
1803 		if (++i >= FREE_NID_PAGES)
1804 			break;
1805 	}
1806 
1807 	/* go to the next free nat pages to find free nids abundantly */
1808 	nm_i->next_scan_nid = nid;
1809 
1810 	/* find free nids from current sum_pages */
1811 	down_read(&curseg->journal_rwsem);
1812 	for (i = 0; i < nats_in_cursum(journal); i++) {
1813 		block_t addr;
1814 
1815 		addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
1816 		nid = le32_to_cpu(nid_in_journal(journal, i));
1817 		if (addr == NULL_ADDR)
1818 			add_free_nid(sbi, nid, true);
1819 		else
1820 			remove_free_nid(nm_i, nid);
1821 	}
1822 	up_read(&curseg->journal_rwsem);
1823 	percpu_up_read(&nm_i->nat_tree_lock);
1824 
1825 	ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
1826 					nm_i->ra_nid_pages, META_NAT, false);
1827 }
1828 
1829 /*
1830  * If this function returns success, caller can obtain a new nid
1831  * from second parameter of this function.
1832  * The returned nid could be used ino as well as nid when inode is created.
1833  */
1834 bool alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
1835 {
1836 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1837 	struct free_nid *i = NULL;
1838 retry:
1839 #ifdef CONFIG_F2FS_FAULT_INJECTION
1840 	if (time_to_inject(FAULT_ALLOC_NID))
1841 		return false;
1842 #endif
1843 	if (unlikely(sbi->total_valid_node_count + 1 > nm_i->available_nids))
1844 		return false;
1845 
1846 	spin_lock(&nm_i->free_nid_list_lock);
1847 
1848 	/* We should not use stale free nids created by build_free_nids */
1849 	if (nm_i->fcnt && !on_build_free_nids(nm_i)) {
1850 		f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
1851 		list_for_each_entry(i, &nm_i->free_nid_list, list)
1852 			if (i->state == NID_NEW)
1853 				break;
1854 
1855 		f2fs_bug_on(sbi, i->state != NID_NEW);
1856 		*nid = i->nid;
1857 		i->state = NID_ALLOC;
1858 		nm_i->fcnt--;
1859 		spin_unlock(&nm_i->free_nid_list_lock);
1860 		return true;
1861 	}
1862 	spin_unlock(&nm_i->free_nid_list_lock);
1863 
1864 	/* Let's scan nat pages and its caches to get free nids */
1865 	mutex_lock(&nm_i->build_lock);
1866 	build_free_nids(sbi);
1867 	mutex_unlock(&nm_i->build_lock);
1868 	goto retry;
1869 }
1870 
1871 /*
1872  * alloc_nid() should be called prior to this function.
1873  */
1874 void alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
1875 {
1876 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1877 	struct free_nid *i;
1878 
1879 	spin_lock(&nm_i->free_nid_list_lock);
1880 	i = __lookup_free_nid_list(nm_i, nid);
1881 	f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1882 	__del_from_free_nid_list(nm_i, i);
1883 	spin_unlock(&nm_i->free_nid_list_lock);
1884 
1885 	kmem_cache_free(free_nid_slab, i);
1886 }
1887 
1888 /*
1889  * alloc_nid() should be called prior to this function.
1890  */
1891 void alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
1892 {
1893 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1894 	struct free_nid *i;
1895 	bool need_free = false;
1896 
1897 	if (!nid)
1898 		return;
1899 
1900 	spin_lock(&nm_i->free_nid_list_lock);
1901 	i = __lookup_free_nid_list(nm_i, nid);
1902 	f2fs_bug_on(sbi, !i || i->state != NID_ALLOC);
1903 	if (!available_free_memory(sbi, FREE_NIDS)) {
1904 		__del_from_free_nid_list(nm_i, i);
1905 		need_free = true;
1906 	} else {
1907 		i->state = NID_NEW;
1908 		nm_i->fcnt++;
1909 	}
1910 	spin_unlock(&nm_i->free_nid_list_lock);
1911 
1912 	if (need_free)
1913 		kmem_cache_free(free_nid_slab, i);
1914 }
1915 
1916 int try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
1917 {
1918 	struct f2fs_nm_info *nm_i = NM_I(sbi);
1919 	struct free_nid *i, *next;
1920 	int nr = nr_shrink;
1921 
1922 	if (nm_i->fcnt <= MAX_FREE_NIDS)
1923 		return 0;
1924 
1925 	if (!mutex_trylock(&nm_i->build_lock))
1926 		return 0;
1927 
1928 	spin_lock(&nm_i->free_nid_list_lock);
1929 	list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
1930 		if (nr_shrink <= 0 || nm_i->fcnt <= MAX_FREE_NIDS)
1931 			break;
1932 		if (i->state == NID_ALLOC)
1933 			continue;
1934 		__del_from_free_nid_list(nm_i, i);
1935 		kmem_cache_free(free_nid_slab, i);
1936 		nm_i->fcnt--;
1937 		nr_shrink--;
1938 	}
1939 	spin_unlock(&nm_i->free_nid_list_lock);
1940 	mutex_unlock(&nm_i->build_lock);
1941 
1942 	return nr - nr_shrink;
1943 }
1944 
1945 void recover_inline_xattr(struct inode *inode, struct page *page)
1946 {
1947 	void *src_addr, *dst_addr;
1948 	size_t inline_size;
1949 	struct page *ipage;
1950 	struct f2fs_inode *ri;
1951 
1952 	ipage = get_node_page(F2FS_I_SB(inode), inode->i_ino);
1953 	f2fs_bug_on(F2FS_I_SB(inode), IS_ERR(ipage));
1954 
1955 	ri = F2FS_INODE(page);
1956 	if (!(ri->i_inline & F2FS_INLINE_XATTR)) {
1957 		clear_inode_flag(inode, FI_INLINE_XATTR);
1958 		goto update_inode;
1959 	}
1960 
1961 	dst_addr = inline_xattr_addr(ipage);
1962 	src_addr = inline_xattr_addr(page);
1963 	inline_size = inline_xattr_size(inode);
1964 
1965 	f2fs_wait_on_page_writeback(ipage, NODE, true);
1966 	memcpy(dst_addr, src_addr, inline_size);
1967 update_inode:
1968 	update_inode(inode, ipage);
1969 	f2fs_put_page(ipage, 1);
1970 }
1971 
1972 void recover_xattr_data(struct inode *inode, struct page *page, block_t blkaddr)
1973 {
1974 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1975 	nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
1976 	nid_t new_xnid = nid_of_node(page);
1977 	struct node_info ni;
1978 
1979 	/* 1: invalidate the previous xattr nid */
1980 	if (!prev_xnid)
1981 		goto recover_xnid;
1982 
1983 	/* Deallocate node address */
1984 	get_node_info(sbi, prev_xnid, &ni);
1985 	f2fs_bug_on(sbi, ni.blk_addr == NULL_ADDR);
1986 	invalidate_blocks(sbi, ni.blk_addr);
1987 	dec_valid_node_count(sbi, inode);
1988 	set_node_addr(sbi, &ni, NULL_ADDR, false);
1989 
1990 recover_xnid:
1991 	/* 2: allocate new xattr nid */
1992 	if (unlikely(!inc_valid_node_count(sbi, inode)))
1993 		f2fs_bug_on(sbi, 1);
1994 
1995 	remove_free_nid(NM_I(sbi), new_xnid);
1996 	get_node_info(sbi, new_xnid, &ni);
1997 	ni.ino = inode->i_ino;
1998 	set_node_addr(sbi, &ni, NEW_ADDR, false);
1999 	f2fs_i_xnid_write(inode, new_xnid);
2000 
2001 	/* 3: update xattr blkaddr */
2002 	refresh_sit_entry(sbi, NEW_ADDR, blkaddr);
2003 	set_node_addr(sbi, &ni, blkaddr, false);
2004 }
2005 
2006 int recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2007 {
2008 	struct f2fs_inode *src, *dst;
2009 	nid_t ino = ino_of_node(page);
2010 	struct node_info old_ni, new_ni;
2011 	struct page *ipage;
2012 
2013 	get_node_info(sbi, ino, &old_ni);
2014 
2015 	if (unlikely(old_ni.blk_addr != NULL_ADDR))
2016 		return -EINVAL;
2017 
2018 	ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2019 	if (!ipage)
2020 		return -ENOMEM;
2021 
2022 	/* Should not use this inode from free nid list */
2023 	remove_free_nid(NM_I(sbi), ino);
2024 
2025 	if (!PageUptodate(ipage))
2026 		SetPageUptodate(ipage);
2027 	fill_node_footer(ipage, ino, ino, 0, true);
2028 
2029 	src = F2FS_INODE(page);
2030 	dst = F2FS_INODE(ipage);
2031 
2032 	memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2033 	dst->i_size = 0;
2034 	dst->i_blocks = cpu_to_le64(1);
2035 	dst->i_links = cpu_to_le32(1);
2036 	dst->i_xattr_nid = 0;
2037 	dst->i_inline = src->i_inline & F2FS_INLINE_XATTR;
2038 
2039 	new_ni = old_ni;
2040 	new_ni.ino = ino;
2041 
2042 	if (unlikely(!inc_valid_node_count(sbi, NULL)))
2043 		WARN_ON(1);
2044 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2045 	inc_valid_inode_count(sbi);
2046 	set_page_dirty(ipage);
2047 	f2fs_put_page(ipage, 1);
2048 	return 0;
2049 }
2050 
2051 int restore_node_summary(struct f2fs_sb_info *sbi,
2052 			unsigned int segno, struct f2fs_summary_block *sum)
2053 {
2054 	struct f2fs_node *rn;
2055 	struct f2fs_summary *sum_entry;
2056 	block_t addr;
2057 	int bio_blocks = MAX_BIO_BLOCKS(sbi);
2058 	int i, idx, last_offset, nrpages;
2059 
2060 	/* scan the node segment */
2061 	last_offset = sbi->blocks_per_seg;
2062 	addr = START_BLOCK(sbi, segno);
2063 	sum_entry = &sum->entries[0];
2064 
2065 	for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2066 		nrpages = min(last_offset - i, bio_blocks);
2067 
2068 		/* readahead node pages */
2069 		ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2070 
2071 		for (idx = addr; idx < addr + nrpages; idx++) {
2072 			struct page *page = get_tmp_page(sbi, idx);
2073 
2074 			rn = F2FS_NODE(page);
2075 			sum_entry->nid = rn->footer.nid;
2076 			sum_entry->version = 0;
2077 			sum_entry->ofs_in_node = 0;
2078 			sum_entry++;
2079 			f2fs_put_page(page, 1);
2080 		}
2081 
2082 		invalidate_mapping_pages(META_MAPPING(sbi), addr,
2083 							addr + nrpages);
2084 	}
2085 	return 0;
2086 }
2087 
2088 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2089 {
2090 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2091 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2092 	struct f2fs_journal *journal = curseg->journal;
2093 	int i;
2094 
2095 	down_write(&curseg->journal_rwsem);
2096 	for (i = 0; i < nats_in_cursum(journal); i++) {
2097 		struct nat_entry *ne;
2098 		struct f2fs_nat_entry raw_ne;
2099 		nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2100 
2101 		raw_ne = nat_in_journal(journal, i);
2102 
2103 		ne = __lookup_nat_cache(nm_i, nid);
2104 		if (!ne) {
2105 			ne = grab_nat_entry(nm_i, nid);
2106 			node_info_from_raw_nat(&ne->ni, &raw_ne);
2107 		}
2108 		__set_nat_cache_dirty(nm_i, ne);
2109 	}
2110 	update_nats_in_cursum(journal, -i);
2111 	up_write(&curseg->journal_rwsem);
2112 }
2113 
2114 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2115 						struct list_head *head, int max)
2116 {
2117 	struct nat_entry_set *cur;
2118 
2119 	if (nes->entry_cnt >= max)
2120 		goto add_out;
2121 
2122 	list_for_each_entry(cur, head, set_list) {
2123 		if (cur->entry_cnt >= nes->entry_cnt) {
2124 			list_add(&nes->set_list, cur->set_list.prev);
2125 			return;
2126 		}
2127 	}
2128 add_out:
2129 	list_add_tail(&nes->set_list, head);
2130 }
2131 
2132 static void __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2133 					struct nat_entry_set *set)
2134 {
2135 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2136 	struct f2fs_journal *journal = curseg->journal;
2137 	nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2138 	bool to_journal = true;
2139 	struct f2fs_nat_block *nat_blk;
2140 	struct nat_entry *ne, *cur;
2141 	struct page *page = NULL;
2142 
2143 	/*
2144 	 * there are two steps to flush nat entries:
2145 	 * #1, flush nat entries to journal in current hot data summary block.
2146 	 * #2, flush nat entries to nat page.
2147 	 */
2148 	if (!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2149 		to_journal = false;
2150 
2151 	if (to_journal) {
2152 		down_write(&curseg->journal_rwsem);
2153 	} else {
2154 		page = get_next_nat_page(sbi, start_nid);
2155 		nat_blk = page_address(page);
2156 		f2fs_bug_on(sbi, !nat_blk);
2157 	}
2158 
2159 	/* flush dirty nats in nat entry set */
2160 	list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2161 		struct f2fs_nat_entry *raw_ne;
2162 		nid_t nid = nat_get_nid(ne);
2163 		int offset;
2164 
2165 		if (nat_get_blkaddr(ne) == NEW_ADDR)
2166 			continue;
2167 
2168 		if (to_journal) {
2169 			offset = lookup_journal_in_cursum(journal,
2170 							NAT_JOURNAL, nid, 1);
2171 			f2fs_bug_on(sbi, offset < 0);
2172 			raw_ne = &nat_in_journal(journal, offset);
2173 			nid_in_journal(journal, offset) = cpu_to_le32(nid);
2174 		} else {
2175 			raw_ne = &nat_blk->entries[nid - start_nid];
2176 		}
2177 		raw_nat_from_node_info(raw_ne, &ne->ni);
2178 		nat_reset_flag(ne);
2179 		__clear_nat_cache_dirty(NM_I(sbi), ne);
2180 		if (nat_get_blkaddr(ne) == NULL_ADDR)
2181 			add_free_nid(sbi, nid, false);
2182 	}
2183 
2184 	if (to_journal)
2185 		up_write(&curseg->journal_rwsem);
2186 	else
2187 		f2fs_put_page(page, 1);
2188 
2189 	f2fs_bug_on(sbi, set->entry_cnt);
2190 
2191 	radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2192 	kmem_cache_free(nat_entry_set_slab, set);
2193 }
2194 
2195 /*
2196  * This function is called during the checkpointing process.
2197  */
2198 void flush_nat_entries(struct f2fs_sb_info *sbi)
2199 {
2200 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2201 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2202 	struct f2fs_journal *journal = curseg->journal;
2203 	struct nat_entry_set *setvec[SETVEC_SIZE];
2204 	struct nat_entry_set *set, *tmp;
2205 	unsigned int found;
2206 	nid_t set_idx = 0;
2207 	LIST_HEAD(sets);
2208 
2209 	if (!nm_i->dirty_nat_cnt)
2210 		return;
2211 
2212 	percpu_down_write(&nm_i->nat_tree_lock);
2213 
2214 	/*
2215 	 * if there are no enough space in journal to store dirty nat
2216 	 * entries, remove all entries from journal and merge them
2217 	 * into nat entry set.
2218 	 */
2219 	if (!__has_cursum_space(journal, nm_i->dirty_nat_cnt, NAT_JOURNAL))
2220 		remove_nats_in_journal(sbi);
2221 
2222 	while ((found = __gang_lookup_nat_set(nm_i,
2223 					set_idx, SETVEC_SIZE, setvec))) {
2224 		unsigned idx;
2225 		set_idx = setvec[found - 1]->set + 1;
2226 		for (idx = 0; idx < found; idx++)
2227 			__adjust_nat_entry_set(setvec[idx], &sets,
2228 						MAX_NAT_JENTRIES(journal));
2229 	}
2230 
2231 	/* flush dirty nats in nat entry set */
2232 	list_for_each_entry_safe(set, tmp, &sets, set_list)
2233 		__flush_nat_entry_set(sbi, set);
2234 
2235 	percpu_up_write(&nm_i->nat_tree_lock);
2236 
2237 	f2fs_bug_on(sbi, nm_i->dirty_nat_cnt);
2238 }
2239 
2240 static int init_node_manager(struct f2fs_sb_info *sbi)
2241 {
2242 	struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
2243 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2244 	unsigned char *version_bitmap;
2245 	unsigned int nat_segs, nat_blocks;
2246 
2247 	nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
2248 
2249 	/* segment_count_nat includes pair segment so divide to 2. */
2250 	nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
2251 	nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
2252 
2253 	nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nat_blocks;
2254 
2255 	/* not used nids: 0, node, meta, (and root counted as valid node) */
2256 	nm_i->available_nids = nm_i->max_nid - F2FS_RESERVED_NODE_NUM;
2257 	nm_i->fcnt = 0;
2258 	nm_i->nat_cnt = 0;
2259 	nm_i->ram_thresh = DEF_RAM_THRESHOLD;
2260 	nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
2261 	nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
2262 
2263 	INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
2264 	INIT_LIST_HEAD(&nm_i->free_nid_list);
2265 	INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
2266 	INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
2267 	INIT_LIST_HEAD(&nm_i->nat_entries);
2268 
2269 	mutex_init(&nm_i->build_lock);
2270 	spin_lock_init(&nm_i->free_nid_list_lock);
2271 	if (percpu_init_rwsem(&nm_i->nat_tree_lock))
2272 		return -ENOMEM;
2273 
2274 	nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
2275 	nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
2276 	version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
2277 	if (!version_bitmap)
2278 		return -EFAULT;
2279 
2280 	nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
2281 					GFP_KERNEL);
2282 	if (!nm_i->nat_bitmap)
2283 		return -ENOMEM;
2284 	return 0;
2285 }
2286 
2287 int build_node_manager(struct f2fs_sb_info *sbi)
2288 {
2289 	int err;
2290 
2291 	sbi->nm_info = kzalloc(sizeof(struct f2fs_nm_info), GFP_KERNEL);
2292 	if (!sbi->nm_info)
2293 		return -ENOMEM;
2294 
2295 	err = init_node_manager(sbi);
2296 	if (err)
2297 		return err;
2298 
2299 	build_free_nids(sbi);
2300 	return 0;
2301 }
2302 
2303 void destroy_node_manager(struct f2fs_sb_info *sbi)
2304 {
2305 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2306 	struct free_nid *i, *next_i;
2307 	struct nat_entry *natvec[NATVEC_SIZE];
2308 	struct nat_entry_set *setvec[SETVEC_SIZE];
2309 	nid_t nid = 0;
2310 	unsigned int found;
2311 
2312 	if (!nm_i)
2313 		return;
2314 
2315 	/* destroy free nid list */
2316 	spin_lock(&nm_i->free_nid_list_lock);
2317 	list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
2318 		f2fs_bug_on(sbi, i->state == NID_ALLOC);
2319 		__del_from_free_nid_list(nm_i, i);
2320 		nm_i->fcnt--;
2321 		spin_unlock(&nm_i->free_nid_list_lock);
2322 		kmem_cache_free(free_nid_slab, i);
2323 		spin_lock(&nm_i->free_nid_list_lock);
2324 	}
2325 	f2fs_bug_on(sbi, nm_i->fcnt);
2326 	spin_unlock(&nm_i->free_nid_list_lock);
2327 
2328 	/* destroy nat cache */
2329 	percpu_down_write(&nm_i->nat_tree_lock);
2330 	while ((found = __gang_lookup_nat_cache(nm_i,
2331 					nid, NATVEC_SIZE, natvec))) {
2332 		unsigned idx;
2333 
2334 		nid = nat_get_nid(natvec[found - 1]) + 1;
2335 		for (idx = 0; idx < found; idx++)
2336 			__del_from_nat_cache(nm_i, natvec[idx]);
2337 	}
2338 	f2fs_bug_on(sbi, nm_i->nat_cnt);
2339 
2340 	/* destroy nat set cache */
2341 	nid = 0;
2342 	while ((found = __gang_lookup_nat_set(nm_i,
2343 					nid, SETVEC_SIZE, setvec))) {
2344 		unsigned idx;
2345 
2346 		nid = setvec[found - 1]->set + 1;
2347 		for (idx = 0; idx < found; idx++) {
2348 			/* entry_cnt is not zero, when cp_error was occurred */
2349 			f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
2350 			radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
2351 			kmem_cache_free(nat_entry_set_slab, setvec[idx]);
2352 		}
2353 	}
2354 	percpu_up_write(&nm_i->nat_tree_lock);
2355 
2356 	percpu_free_rwsem(&nm_i->nat_tree_lock);
2357 	kfree(nm_i->nat_bitmap);
2358 	sbi->nm_info = NULL;
2359 	kfree(nm_i);
2360 }
2361 
2362 int __init create_node_manager_caches(void)
2363 {
2364 	nat_entry_slab = f2fs_kmem_cache_create("nat_entry",
2365 			sizeof(struct nat_entry));
2366 	if (!nat_entry_slab)
2367 		goto fail;
2368 
2369 	free_nid_slab = f2fs_kmem_cache_create("free_nid",
2370 			sizeof(struct free_nid));
2371 	if (!free_nid_slab)
2372 		goto destroy_nat_entry;
2373 
2374 	nat_entry_set_slab = f2fs_kmem_cache_create("nat_entry_set",
2375 			sizeof(struct nat_entry_set));
2376 	if (!nat_entry_set_slab)
2377 		goto destroy_free_nid;
2378 	return 0;
2379 
2380 destroy_free_nid:
2381 	kmem_cache_destroy(free_nid_slab);
2382 destroy_nat_entry:
2383 	kmem_cache_destroy(nat_entry_slab);
2384 fail:
2385 	return -ENOMEM;
2386 }
2387 
2388 void destroy_node_manager_caches(void)
2389 {
2390 	kmem_cache_destroy(nat_entry_set_slab);
2391 	kmem_cache_destroy(free_nid_slab);
2392 	kmem_cache_destroy(nat_entry_slab);
2393 }
2394