• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * fs/f2fs/node.c
4  *
5  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6  *             http://www.samsung.com/
7  */
8 #include <linux/fs.h>
9 #include <linux/f2fs_fs.h>
10 #include <linux/mpage.h>
11 #include <linux/backing-dev.h>
12 #include <linux/blkdev.h>
13 #include <linux/pagevec.h>
14 #include <linux/swap.h>
15 
16 #include "f2fs.h"
17 #include "node.h"
18 #include "segment.h"
19 #include "xattr.h"
20 #include "trace.h"
21 #include <trace/events/f2fs.h>
22 
23 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
24 
25 static struct kmem_cache *nat_entry_slab;
26 static struct kmem_cache *free_nid_slab;
27 static struct kmem_cache *nat_entry_set_slab;
28 static struct kmem_cache *fsync_node_entry_slab;
29 
30 /*
31  * Check whether the given nid is within node id range.
32  */
f2fs_check_nid_range(struct f2fs_sb_info * sbi,nid_t nid)33 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
34 {
35 	if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
36 		set_sbi_flag(sbi, SBI_NEED_FSCK);
37 		f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
38 			  __func__, nid);
39 		return -EFSCORRUPTED;
40 	}
41 	return 0;
42 }
43 
f2fs_available_free_memory(struct f2fs_sb_info * sbi,int type)44 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
45 {
46 	struct f2fs_nm_info *nm_i = NM_I(sbi);
47 	struct sysinfo val;
48 	unsigned long avail_ram;
49 	unsigned long mem_size = 0;
50 	bool res = false;
51 
52 	si_meminfo(&val);
53 
54 	/* only uses low memory */
55 	avail_ram = val.totalram - val.totalhigh;
56 
57 	/*
58 	 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
59 	 */
60 	if (type == FREE_NIDS) {
61 		mem_size = (nm_i->nid_cnt[FREE_NID] *
62 				sizeof(struct free_nid)) >> PAGE_SHIFT;
63 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
64 	} else if (type == NAT_ENTRIES) {
65 		mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
66 				sizeof(struct nat_entry)) >> PAGE_SHIFT;
67 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
68 		if (excess_cached_nats(sbi))
69 			res = false;
70 	} else if (type == DIRTY_DENTS) {
71 		if (sbi->sb->s_bdi->wb.dirty_exceeded)
72 			return false;
73 		mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
74 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
75 	} else if (type == INO_ENTRIES) {
76 		int i;
77 
78 		for (i = 0; i < MAX_INO_ENTRY; i++)
79 			mem_size += sbi->im[i].ino_num *
80 						sizeof(struct ino_entry);
81 		mem_size >>= PAGE_SHIFT;
82 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
83 	} else if (type == EXTENT_CACHE) {
84 		mem_size = (atomic_read(&sbi->total_ext_tree) *
85 				sizeof(struct extent_tree) +
86 				atomic_read(&sbi->total_ext_node) *
87 				sizeof(struct extent_node)) >> PAGE_SHIFT;
88 		res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
89 	} else if (type == INMEM_PAGES) {
90 		/* it allows 20% / total_ram for inmemory pages */
91 		mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
92 		res = mem_size < (val.totalram / 5);
93 	} else {
94 		if (!sbi->sb->s_bdi->wb.dirty_exceeded)
95 			return true;
96 	}
97 	return res;
98 }
99 
clear_node_page_dirty(struct page * page)100 static void clear_node_page_dirty(struct page *page)
101 {
102 	if (PageDirty(page)) {
103 		f2fs_clear_page_cache_dirty_tag(page);
104 		clear_page_dirty_for_io(page);
105 		dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
106 	}
107 	ClearPageUptodate(page);
108 }
109 
get_current_nat_page(struct f2fs_sb_info * sbi,nid_t nid)110 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
111 {
112 	return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
113 }
114 
get_next_nat_page(struct f2fs_sb_info * sbi,nid_t nid)115 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
116 {
117 	struct page *src_page;
118 	struct page *dst_page;
119 	pgoff_t dst_off;
120 	void *src_addr;
121 	void *dst_addr;
122 	struct f2fs_nm_info *nm_i = NM_I(sbi);
123 
124 	dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
125 
126 	/* get current nat block page with lock */
127 	src_page = get_current_nat_page(sbi, nid);
128 	if (IS_ERR(src_page))
129 		return src_page;
130 	dst_page = f2fs_grab_meta_page(sbi, dst_off);
131 	f2fs_bug_on(sbi, PageDirty(src_page));
132 
133 	src_addr = page_address(src_page);
134 	dst_addr = page_address(dst_page);
135 	memcpy(dst_addr, src_addr, PAGE_SIZE);
136 	set_page_dirty(dst_page);
137 	f2fs_put_page(src_page, 1);
138 
139 	set_to_next_nat(nm_i, nid);
140 
141 	return dst_page;
142 }
143 
__alloc_nat_entry(nid_t nid,bool no_fail)144 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
145 {
146 	struct nat_entry *new;
147 
148 	if (no_fail)
149 		new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
150 	else
151 		new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
152 	if (new) {
153 		nat_set_nid(new, nid);
154 		nat_reset_flag(new);
155 	}
156 	return new;
157 }
158 
__free_nat_entry(struct nat_entry * e)159 static void __free_nat_entry(struct nat_entry *e)
160 {
161 	kmem_cache_free(nat_entry_slab, e);
162 }
163 
164 /* must be locked by nat_tree_lock */
__init_nat_entry(struct f2fs_nm_info * nm_i,struct nat_entry * ne,struct f2fs_nat_entry * raw_ne,bool no_fail)165 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
166 	struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
167 {
168 	if (no_fail)
169 		f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
170 	else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
171 		return NULL;
172 
173 	if (raw_ne)
174 		node_info_from_raw_nat(&ne->ni, raw_ne);
175 
176 	spin_lock(&nm_i->nat_list_lock);
177 	list_add_tail(&ne->list, &nm_i->nat_entries);
178 	spin_unlock(&nm_i->nat_list_lock);
179 
180 	nm_i->nat_cnt[TOTAL_NAT]++;
181 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
182 	return ne;
183 }
184 
__lookup_nat_cache(struct f2fs_nm_info * nm_i,nid_t n)185 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
186 {
187 	struct nat_entry *ne;
188 
189 	ne = radix_tree_lookup(&nm_i->nat_root, n);
190 
191 	/* for recent accessed nat entry, move it to tail of lru list */
192 	if (ne && !get_nat_flag(ne, IS_DIRTY)) {
193 		spin_lock(&nm_i->nat_list_lock);
194 		if (!list_empty(&ne->list))
195 			list_move_tail(&ne->list, &nm_i->nat_entries);
196 		spin_unlock(&nm_i->nat_list_lock);
197 	}
198 
199 	return ne;
200 }
201 
__gang_lookup_nat_cache(struct f2fs_nm_info * nm_i,nid_t start,unsigned int nr,struct nat_entry ** ep)202 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
203 		nid_t start, unsigned int nr, struct nat_entry **ep)
204 {
205 	return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
206 }
207 
__del_from_nat_cache(struct f2fs_nm_info * nm_i,struct nat_entry * e)208 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
209 {
210 	radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
211 	nm_i->nat_cnt[TOTAL_NAT]--;
212 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
213 	__free_nat_entry(e);
214 }
215 
__grab_nat_entry_set(struct f2fs_nm_info * nm_i,struct nat_entry * ne)216 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
217 							struct nat_entry *ne)
218 {
219 	nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
220 	struct nat_entry_set *head;
221 
222 	head = radix_tree_lookup(&nm_i->nat_set_root, set);
223 	if (!head) {
224 		head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
225 
226 		INIT_LIST_HEAD(&head->entry_list);
227 		INIT_LIST_HEAD(&head->set_list);
228 		head->set = set;
229 		head->entry_cnt = 0;
230 		f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
231 	}
232 	return head;
233 }
234 
__set_nat_cache_dirty(struct f2fs_nm_info * nm_i,struct nat_entry * ne)235 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
236 						struct nat_entry *ne)
237 {
238 	struct nat_entry_set *head;
239 	bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
240 
241 	if (!new_ne)
242 		head = __grab_nat_entry_set(nm_i, ne);
243 
244 	/*
245 	 * update entry_cnt in below condition:
246 	 * 1. update NEW_ADDR to valid block address;
247 	 * 2. update old block address to new one;
248 	 */
249 	if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
250 				!get_nat_flag(ne, IS_DIRTY)))
251 		head->entry_cnt++;
252 
253 	set_nat_flag(ne, IS_PREALLOC, new_ne);
254 
255 	if (get_nat_flag(ne, IS_DIRTY))
256 		goto refresh_list;
257 
258 	nm_i->nat_cnt[DIRTY_NAT]++;
259 	nm_i->nat_cnt[RECLAIMABLE_NAT]--;
260 	set_nat_flag(ne, IS_DIRTY, true);
261 refresh_list:
262 	spin_lock(&nm_i->nat_list_lock);
263 	if (new_ne)
264 		list_del_init(&ne->list);
265 	else
266 		list_move_tail(&ne->list, &head->entry_list);
267 	spin_unlock(&nm_i->nat_list_lock);
268 }
269 
__clear_nat_cache_dirty(struct f2fs_nm_info * nm_i,struct nat_entry_set * set,struct nat_entry * ne)270 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
271 		struct nat_entry_set *set, struct nat_entry *ne)
272 {
273 	spin_lock(&nm_i->nat_list_lock);
274 	list_move_tail(&ne->list, &nm_i->nat_entries);
275 	spin_unlock(&nm_i->nat_list_lock);
276 
277 	set_nat_flag(ne, IS_DIRTY, false);
278 	set->entry_cnt--;
279 	nm_i->nat_cnt[DIRTY_NAT]--;
280 	nm_i->nat_cnt[RECLAIMABLE_NAT]++;
281 }
282 
__gang_lookup_nat_set(struct f2fs_nm_info * nm_i,nid_t start,unsigned int nr,struct nat_entry_set ** ep)283 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
284 		nid_t start, unsigned int nr, struct nat_entry_set **ep)
285 {
286 	return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
287 							start, nr);
288 }
289 
f2fs_in_warm_node_list(struct f2fs_sb_info * sbi,struct page * page)290 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
291 {
292 	return NODE_MAPPING(sbi) == page->mapping &&
293 			IS_DNODE(page) && is_cold_node(page);
294 }
295 
f2fs_init_fsync_node_info(struct f2fs_sb_info * sbi)296 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
297 {
298 	spin_lock_init(&sbi->fsync_node_lock);
299 	INIT_LIST_HEAD(&sbi->fsync_node_list);
300 	sbi->fsync_seg_id = 0;
301 	sbi->fsync_node_num = 0;
302 }
303 
f2fs_add_fsync_node_entry(struct f2fs_sb_info * sbi,struct page * page)304 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
305 							struct page *page)
306 {
307 	struct fsync_node_entry *fn;
308 	unsigned long flags;
309 	unsigned int seq_id;
310 
311 	fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS);
312 
313 	get_page(page);
314 	fn->page = page;
315 	INIT_LIST_HEAD(&fn->list);
316 
317 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
318 	list_add_tail(&fn->list, &sbi->fsync_node_list);
319 	fn->seq_id = sbi->fsync_seg_id++;
320 	seq_id = fn->seq_id;
321 	sbi->fsync_node_num++;
322 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
323 
324 	return seq_id;
325 }
326 
f2fs_del_fsync_node_entry(struct f2fs_sb_info * sbi,struct page * page)327 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
328 {
329 	struct fsync_node_entry *fn;
330 	unsigned long flags;
331 
332 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
333 	list_for_each_entry(fn, &sbi->fsync_node_list, list) {
334 		if (fn->page == page) {
335 			list_del(&fn->list);
336 			sbi->fsync_node_num--;
337 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
338 			kmem_cache_free(fsync_node_entry_slab, fn);
339 			put_page(page);
340 			return;
341 		}
342 	}
343 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
344 	f2fs_bug_on(sbi, 1);
345 }
346 
f2fs_reset_fsync_node_info(struct f2fs_sb_info * sbi)347 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
348 {
349 	unsigned long flags;
350 
351 	spin_lock_irqsave(&sbi->fsync_node_lock, flags);
352 	sbi->fsync_seg_id = 0;
353 	spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
354 }
355 
f2fs_need_dentry_mark(struct f2fs_sb_info * sbi,nid_t nid)356 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
357 {
358 	struct f2fs_nm_info *nm_i = NM_I(sbi);
359 	struct nat_entry *e;
360 	bool need = false;
361 
362 	down_read(&nm_i->nat_tree_lock);
363 	e = __lookup_nat_cache(nm_i, nid);
364 	if (e) {
365 		if (!get_nat_flag(e, IS_CHECKPOINTED) &&
366 				!get_nat_flag(e, HAS_FSYNCED_INODE))
367 			need = true;
368 	}
369 	up_read(&nm_i->nat_tree_lock);
370 	return need;
371 }
372 
f2fs_is_checkpointed_node(struct f2fs_sb_info * sbi,nid_t nid)373 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
374 {
375 	struct f2fs_nm_info *nm_i = NM_I(sbi);
376 	struct nat_entry *e;
377 	bool is_cp = true;
378 
379 	down_read(&nm_i->nat_tree_lock);
380 	e = __lookup_nat_cache(nm_i, nid);
381 	if (e && !get_nat_flag(e, IS_CHECKPOINTED))
382 		is_cp = false;
383 	up_read(&nm_i->nat_tree_lock);
384 	return is_cp;
385 }
386 
f2fs_need_inode_block_update(struct f2fs_sb_info * sbi,nid_t ino)387 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
388 {
389 	struct f2fs_nm_info *nm_i = NM_I(sbi);
390 	struct nat_entry *e;
391 	bool need_update = true;
392 
393 	down_read(&nm_i->nat_tree_lock);
394 	e = __lookup_nat_cache(nm_i, ino);
395 	if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
396 			(get_nat_flag(e, IS_CHECKPOINTED) ||
397 			 get_nat_flag(e, HAS_FSYNCED_INODE)))
398 		need_update = false;
399 	up_read(&nm_i->nat_tree_lock);
400 	return need_update;
401 }
402 
403 /* must be locked by nat_tree_lock */
cache_nat_entry(struct f2fs_sb_info * sbi,nid_t nid,struct f2fs_nat_entry * ne)404 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
405 						struct f2fs_nat_entry *ne)
406 {
407 	struct f2fs_nm_info *nm_i = NM_I(sbi);
408 	struct nat_entry *new, *e;
409 
410 	new = __alloc_nat_entry(nid, false);
411 	if (!new)
412 		return;
413 
414 	down_write(&nm_i->nat_tree_lock);
415 	e = __lookup_nat_cache(nm_i, nid);
416 	if (!e)
417 		e = __init_nat_entry(nm_i, new, ne, false);
418 	else
419 		f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
420 				nat_get_blkaddr(e) !=
421 					le32_to_cpu(ne->block_addr) ||
422 				nat_get_version(e) != ne->version);
423 	up_write(&nm_i->nat_tree_lock);
424 	if (e != new)
425 		__free_nat_entry(new);
426 }
427 
set_node_addr(struct f2fs_sb_info * sbi,struct node_info * ni,block_t new_blkaddr,bool fsync_done)428 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
429 			block_t new_blkaddr, bool fsync_done)
430 {
431 	struct f2fs_nm_info *nm_i = NM_I(sbi);
432 	struct nat_entry *e;
433 	struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
434 
435 	down_write(&nm_i->nat_tree_lock);
436 	e = __lookup_nat_cache(nm_i, ni->nid);
437 	if (!e) {
438 		e = __init_nat_entry(nm_i, new, NULL, true);
439 		copy_node_info(&e->ni, ni);
440 		f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
441 	} else if (new_blkaddr == NEW_ADDR) {
442 		/*
443 		 * when nid is reallocated,
444 		 * previous nat entry can be remained in nat cache.
445 		 * So, reinitialize it with new information.
446 		 */
447 		copy_node_info(&e->ni, ni);
448 		f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
449 	}
450 	/* let's free early to reduce memory consumption */
451 	if (e != new)
452 		__free_nat_entry(new);
453 
454 	/* sanity check */
455 	f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
456 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
457 			new_blkaddr == NULL_ADDR);
458 	f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
459 			new_blkaddr == NEW_ADDR);
460 	f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
461 			new_blkaddr == NEW_ADDR);
462 
463 	/* increment version no as node is removed */
464 	if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
465 		unsigned char version = nat_get_version(e);
466 		nat_set_version(e, inc_node_version(version));
467 	}
468 
469 	/* change address */
470 	nat_set_blkaddr(e, new_blkaddr);
471 	if (!__is_valid_data_blkaddr(new_blkaddr))
472 		set_nat_flag(e, IS_CHECKPOINTED, false);
473 	__set_nat_cache_dirty(nm_i, e);
474 
475 	/* update fsync_mark if its inode nat entry is still alive */
476 	if (ni->nid != ni->ino)
477 		e = __lookup_nat_cache(nm_i, ni->ino);
478 	if (e) {
479 		if (fsync_done && ni->nid == ni->ino)
480 			set_nat_flag(e, HAS_FSYNCED_INODE, true);
481 		set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
482 	}
483 	up_write(&nm_i->nat_tree_lock);
484 }
485 
f2fs_try_to_free_nats(struct f2fs_sb_info * sbi,int nr_shrink)486 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
487 {
488 	struct f2fs_nm_info *nm_i = NM_I(sbi);
489 	int nr = nr_shrink;
490 
491 	if (!down_write_trylock(&nm_i->nat_tree_lock))
492 		return 0;
493 
494 	spin_lock(&nm_i->nat_list_lock);
495 	while (nr_shrink) {
496 		struct nat_entry *ne;
497 
498 		if (list_empty(&nm_i->nat_entries))
499 			break;
500 
501 		ne = list_first_entry(&nm_i->nat_entries,
502 					struct nat_entry, list);
503 		list_del(&ne->list);
504 		spin_unlock(&nm_i->nat_list_lock);
505 
506 		__del_from_nat_cache(nm_i, ne);
507 		nr_shrink--;
508 
509 		spin_lock(&nm_i->nat_list_lock);
510 	}
511 	spin_unlock(&nm_i->nat_list_lock);
512 
513 	up_write(&nm_i->nat_tree_lock);
514 	return nr - nr_shrink;
515 }
516 
f2fs_get_node_info(struct f2fs_sb_info * sbi,nid_t nid,struct node_info * ni)517 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
518 						struct node_info *ni)
519 {
520 	struct f2fs_nm_info *nm_i = NM_I(sbi);
521 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
522 	struct f2fs_journal *journal = curseg->journal;
523 	nid_t start_nid = START_NID(nid);
524 	struct f2fs_nat_block *nat_blk;
525 	struct page *page = NULL;
526 	struct f2fs_nat_entry ne;
527 	struct nat_entry *e;
528 	pgoff_t index;
529 	block_t blkaddr;
530 	int i;
531 
532 	ni->nid = nid;
533 
534 	/* Check nat cache */
535 	down_read(&nm_i->nat_tree_lock);
536 	e = __lookup_nat_cache(nm_i, nid);
537 	if (e) {
538 		ni->ino = nat_get_ino(e);
539 		ni->blk_addr = nat_get_blkaddr(e);
540 		ni->version = nat_get_version(e);
541 		up_read(&nm_i->nat_tree_lock);
542 		return 0;
543 	}
544 
545 	memset(&ne, 0, sizeof(struct f2fs_nat_entry));
546 
547 	/* Check current segment summary */
548 	down_read(&curseg->journal_rwsem);
549 	i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
550 	if (i >= 0) {
551 		ne = nat_in_journal(journal, i);
552 		node_info_from_raw_nat(ni, &ne);
553 	}
554 	up_read(&curseg->journal_rwsem);
555 	if (i >= 0) {
556 		up_read(&nm_i->nat_tree_lock);
557 		goto cache;
558 	}
559 
560 	/* Fill node_info from nat page */
561 	index = current_nat_addr(sbi, nid);
562 	up_read(&nm_i->nat_tree_lock);
563 
564 	page = f2fs_get_meta_page(sbi, index);
565 	if (IS_ERR(page))
566 		return PTR_ERR(page);
567 
568 	nat_blk = (struct f2fs_nat_block *)page_address(page);
569 	ne = nat_blk->entries[nid - start_nid];
570 	node_info_from_raw_nat(ni, &ne);
571 	f2fs_put_page(page, 1);
572 cache:
573 	blkaddr = le32_to_cpu(ne.block_addr);
574 	if (__is_valid_data_blkaddr(blkaddr) &&
575 		!f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
576 		return -EFAULT;
577 
578 	/* cache nat entry */
579 	cache_nat_entry(sbi, nid, &ne);
580 	return 0;
581 }
582 
583 /*
584  * readahead MAX_RA_NODE number of node pages.
585  */
f2fs_ra_node_pages(struct page * parent,int start,int n)586 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
587 {
588 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
589 	struct blk_plug plug;
590 	int i, end;
591 	nid_t nid;
592 
593 	blk_start_plug(&plug);
594 
595 	/* Then, try readahead for siblings of the desired node */
596 	end = start + n;
597 	end = min(end, NIDS_PER_BLOCK);
598 	for (i = start; i < end; i++) {
599 		nid = get_nid(parent, i, false);
600 		f2fs_ra_node_page(sbi, nid);
601 	}
602 
603 	blk_finish_plug(&plug);
604 }
605 
f2fs_get_next_page_offset(struct dnode_of_data * dn,pgoff_t pgofs)606 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
607 {
608 	const long direct_index = ADDRS_PER_INODE(dn->inode);
609 	const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
610 	const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
611 	unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
612 	int cur_level = dn->cur_level;
613 	int max_level = dn->max_level;
614 	pgoff_t base = 0;
615 
616 	if (!dn->max_level)
617 		return pgofs + 1;
618 
619 	while (max_level-- > cur_level)
620 		skipped_unit *= NIDS_PER_BLOCK;
621 
622 	switch (dn->max_level) {
623 	case 3:
624 		base += 2 * indirect_blks;
625 		fallthrough;
626 	case 2:
627 		base += 2 * direct_blks;
628 		fallthrough;
629 	case 1:
630 		base += direct_index;
631 		break;
632 	default:
633 		f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
634 	}
635 
636 	return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
637 }
638 
639 /*
640  * The maximum depth is four.
641  * Offset[0] will have raw inode offset.
642  */
get_node_path(struct inode * inode,long block,int offset[4],unsigned int noffset[4])643 static int get_node_path(struct inode *inode, long block,
644 				int offset[4], unsigned int noffset[4])
645 {
646 	const long direct_index = ADDRS_PER_INODE(inode);
647 	const long direct_blks = ADDRS_PER_BLOCK(inode);
648 	const long dptrs_per_blk = NIDS_PER_BLOCK;
649 	const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
650 	const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
651 	int n = 0;
652 	int level = 0;
653 
654 	noffset[0] = 0;
655 
656 	if (block < direct_index) {
657 		offset[n] = block;
658 		goto got;
659 	}
660 	block -= direct_index;
661 	if (block < direct_blks) {
662 		offset[n++] = NODE_DIR1_BLOCK;
663 		noffset[n] = 1;
664 		offset[n] = block;
665 		level = 1;
666 		goto got;
667 	}
668 	block -= direct_blks;
669 	if (block < direct_blks) {
670 		offset[n++] = NODE_DIR2_BLOCK;
671 		noffset[n] = 2;
672 		offset[n] = block;
673 		level = 1;
674 		goto got;
675 	}
676 	block -= direct_blks;
677 	if (block < indirect_blks) {
678 		offset[n++] = NODE_IND1_BLOCK;
679 		noffset[n] = 3;
680 		offset[n++] = block / direct_blks;
681 		noffset[n] = 4 + offset[n - 1];
682 		offset[n] = block % direct_blks;
683 		level = 2;
684 		goto got;
685 	}
686 	block -= indirect_blks;
687 	if (block < indirect_blks) {
688 		offset[n++] = NODE_IND2_BLOCK;
689 		noffset[n] = 4 + dptrs_per_blk;
690 		offset[n++] = block / direct_blks;
691 		noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
692 		offset[n] = block % direct_blks;
693 		level = 2;
694 		goto got;
695 	}
696 	block -= indirect_blks;
697 	if (block < dindirect_blks) {
698 		offset[n++] = NODE_DIND_BLOCK;
699 		noffset[n] = 5 + (dptrs_per_blk * 2);
700 		offset[n++] = block / indirect_blks;
701 		noffset[n] = 6 + (dptrs_per_blk * 2) +
702 			      offset[n - 1] * (dptrs_per_blk + 1);
703 		offset[n++] = (block / direct_blks) % dptrs_per_blk;
704 		noffset[n] = 7 + (dptrs_per_blk * 2) +
705 			      offset[n - 2] * (dptrs_per_blk + 1) +
706 			      offset[n - 1];
707 		offset[n] = block % direct_blks;
708 		level = 3;
709 		goto got;
710 	} else {
711 		return -E2BIG;
712 	}
713 got:
714 	return level;
715 }
716 
717 /*
718  * Caller should call f2fs_put_dnode(dn).
719  * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
720  * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
721  */
f2fs_get_dnode_of_data(struct dnode_of_data * dn,pgoff_t index,int mode)722 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
723 {
724 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
725 	struct page *npage[4];
726 	struct page *parent = NULL;
727 	int offset[4];
728 	unsigned int noffset[4];
729 	nid_t nids[4];
730 	int level, i = 0;
731 	int err = 0;
732 
733 	level = get_node_path(dn->inode, index, offset, noffset);
734 	if (level < 0)
735 		return level;
736 
737 	nids[0] = dn->inode->i_ino;
738 	npage[0] = dn->inode_page;
739 
740 	if (!npage[0]) {
741 		npage[0] = f2fs_get_node_page(sbi, nids[0]);
742 		if (IS_ERR(npage[0]))
743 			return PTR_ERR(npage[0]);
744 	}
745 
746 	/* if inline_data is set, should not report any block indices */
747 	if (f2fs_has_inline_data(dn->inode) && index) {
748 		err = -ENOENT;
749 		f2fs_put_page(npage[0], 1);
750 		goto release_out;
751 	}
752 
753 	parent = npage[0];
754 	if (level != 0)
755 		nids[1] = get_nid(parent, offset[0], true);
756 	dn->inode_page = npage[0];
757 	dn->inode_page_locked = true;
758 
759 	/* get indirect or direct nodes */
760 	for (i = 1; i <= level; i++) {
761 		bool done = false;
762 
763 		if (!nids[i] && mode == ALLOC_NODE) {
764 			/* alloc new node */
765 			if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
766 				err = -ENOSPC;
767 				goto release_pages;
768 			}
769 
770 			dn->nid = nids[i];
771 			npage[i] = f2fs_new_node_page(dn, noffset[i]);
772 			if (IS_ERR(npage[i])) {
773 				f2fs_alloc_nid_failed(sbi, nids[i]);
774 				err = PTR_ERR(npage[i]);
775 				goto release_pages;
776 			}
777 
778 			set_nid(parent, offset[i - 1], nids[i], i == 1);
779 			f2fs_alloc_nid_done(sbi, nids[i]);
780 			done = true;
781 		} else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
782 			npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
783 			if (IS_ERR(npage[i])) {
784 				err = PTR_ERR(npage[i]);
785 				goto release_pages;
786 			}
787 			done = true;
788 		}
789 		if (i == 1) {
790 			dn->inode_page_locked = false;
791 			unlock_page(parent);
792 		} else {
793 			f2fs_put_page(parent, 1);
794 		}
795 
796 		if (!done) {
797 			npage[i] = f2fs_get_node_page(sbi, nids[i]);
798 			if (IS_ERR(npage[i])) {
799 				err = PTR_ERR(npage[i]);
800 				f2fs_put_page(npage[0], 0);
801 				goto release_out;
802 			}
803 		}
804 		if (i < level) {
805 			parent = npage[i];
806 			nids[i + 1] = get_nid(parent, offset[i], false);
807 		}
808 	}
809 	dn->nid = nids[level];
810 	dn->ofs_in_node = offset[level];
811 	dn->node_page = npage[level];
812 	dn->data_blkaddr = f2fs_data_blkaddr(dn);
813 	return 0;
814 
815 release_pages:
816 	f2fs_put_page(parent, 1);
817 	if (i > 1)
818 		f2fs_put_page(npage[0], 0);
819 release_out:
820 	dn->inode_page = NULL;
821 	dn->node_page = NULL;
822 	if (err == -ENOENT) {
823 		dn->cur_level = i;
824 		dn->max_level = level;
825 		dn->ofs_in_node = offset[level];
826 	}
827 	return err;
828 }
829 
truncate_node(struct dnode_of_data * dn)830 static int truncate_node(struct dnode_of_data *dn)
831 {
832 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
833 	struct node_info ni;
834 	int err;
835 	pgoff_t index;
836 
837 	err = f2fs_get_node_info(sbi, dn->nid, &ni);
838 	if (err)
839 		return err;
840 
841 	/* Deallocate node address */
842 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
843 	dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
844 	set_node_addr(sbi, &ni, NULL_ADDR, false);
845 
846 	if (dn->nid == dn->inode->i_ino) {
847 		f2fs_remove_orphan_inode(sbi, dn->nid);
848 		dec_valid_inode_count(sbi);
849 		f2fs_inode_synced(dn->inode);
850 	}
851 
852 	clear_node_page_dirty(dn->node_page);
853 	set_sbi_flag(sbi, SBI_IS_DIRTY);
854 
855 	index = dn->node_page->index;
856 	f2fs_put_page(dn->node_page, 1);
857 
858 	invalidate_mapping_pages(NODE_MAPPING(sbi),
859 			index, index);
860 
861 	dn->node_page = NULL;
862 	trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
863 
864 	return 0;
865 }
866 
truncate_dnode(struct dnode_of_data * dn)867 static int truncate_dnode(struct dnode_of_data *dn)
868 {
869 	struct page *page;
870 	int err;
871 
872 	if (dn->nid == 0)
873 		return 1;
874 
875 	/* get direct node */
876 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
877 	if (PTR_ERR(page) == -ENOENT)
878 		return 1;
879 	else if (IS_ERR(page))
880 		return PTR_ERR(page);
881 
882 	/* Make dnode_of_data for parameter */
883 	dn->node_page = page;
884 	dn->ofs_in_node = 0;
885 	f2fs_truncate_data_blocks(dn);
886 	err = truncate_node(dn);
887 	if (err)
888 		return err;
889 
890 	return 1;
891 }
892 
truncate_nodes(struct dnode_of_data * dn,unsigned int nofs,int ofs,int depth)893 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
894 						int ofs, int depth)
895 {
896 	struct dnode_of_data rdn = *dn;
897 	struct page *page;
898 	struct f2fs_node *rn;
899 	nid_t child_nid;
900 	unsigned int child_nofs;
901 	int freed = 0;
902 	int i, ret;
903 
904 	if (dn->nid == 0)
905 		return NIDS_PER_BLOCK + 1;
906 
907 	trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
908 
909 	page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
910 	if (IS_ERR(page)) {
911 		trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
912 		return PTR_ERR(page);
913 	}
914 
915 	f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
916 
917 	rn = F2FS_NODE(page);
918 	if (depth < 3) {
919 		for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
920 			child_nid = le32_to_cpu(rn->in.nid[i]);
921 			if (child_nid == 0)
922 				continue;
923 			rdn.nid = child_nid;
924 			ret = truncate_dnode(&rdn);
925 			if (ret < 0)
926 				goto out_err;
927 			if (set_nid(page, i, 0, false))
928 				dn->node_changed = true;
929 		}
930 	} else {
931 		child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
932 		for (i = ofs; i < NIDS_PER_BLOCK; i++) {
933 			child_nid = le32_to_cpu(rn->in.nid[i]);
934 			if (child_nid == 0) {
935 				child_nofs += NIDS_PER_BLOCK + 1;
936 				continue;
937 			}
938 			rdn.nid = child_nid;
939 			ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
940 			if (ret == (NIDS_PER_BLOCK + 1)) {
941 				if (set_nid(page, i, 0, false))
942 					dn->node_changed = true;
943 				child_nofs += ret;
944 			} else if (ret < 0 && ret != -ENOENT) {
945 				goto out_err;
946 			}
947 		}
948 		freed = child_nofs;
949 	}
950 
951 	if (!ofs) {
952 		/* remove current indirect node */
953 		dn->node_page = page;
954 		ret = truncate_node(dn);
955 		if (ret)
956 			goto out_err;
957 		freed++;
958 	} else {
959 		f2fs_put_page(page, 1);
960 	}
961 	trace_f2fs_truncate_nodes_exit(dn->inode, freed);
962 	return freed;
963 
964 out_err:
965 	f2fs_put_page(page, 1);
966 	trace_f2fs_truncate_nodes_exit(dn->inode, ret);
967 	return ret;
968 }
969 
truncate_partial_nodes(struct dnode_of_data * dn,struct f2fs_inode * ri,int * offset,int depth)970 static int truncate_partial_nodes(struct dnode_of_data *dn,
971 			struct f2fs_inode *ri, int *offset, int depth)
972 {
973 	struct page *pages[2];
974 	nid_t nid[3];
975 	nid_t child_nid;
976 	int err = 0;
977 	int i;
978 	int idx = depth - 2;
979 
980 	nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
981 	if (!nid[0])
982 		return 0;
983 
984 	/* get indirect nodes in the path */
985 	for (i = 0; i < idx + 1; i++) {
986 		/* reference count'll be increased */
987 		pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
988 		if (IS_ERR(pages[i])) {
989 			err = PTR_ERR(pages[i]);
990 			idx = i - 1;
991 			goto fail;
992 		}
993 		nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
994 	}
995 
996 	f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
997 
998 	/* free direct nodes linked to a partial indirect node */
999 	for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1000 		child_nid = get_nid(pages[idx], i, false);
1001 		if (!child_nid)
1002 			continue;
1003 		dn->nid = child_nid;
1004 		err = truncate_dnode(dn);
1005 		if (err < 0)
1006 			goto fail;
1007 		if (set_nid(pages[idx], i, 0, false))
1008 			dn->node_changed = true;
1009 	}
1010 
1011 	if (offset[idx + 1] == 0) {
1012 		dn->node_page = pages[idx];
1013 		dn->nid = nid[idx];
1014 		err = truncate_node(dn);
1015 		if (err)
1016 			goto fail;
1017 	} else {
1018 		f2fs_put_page(pages[idx], 1);
1019 	}
1020 	offset[idx]++;
1021 	offset[idx + 1] = 0;
1022 	idx--;
1023 fail:
1024 	for (i = idx; i >= 0; i--)
1025 		f2fs_put_page(pages[i], 1);
1026 
1027 	trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1028 
1029 	return err;
1030 }
1031 
1032 /*
1033  * All the block addresses of data and nodes should be nullified.
1034  */
f2fs_truncate_inode_blocks(struct inode * inode,pgoff_t from)1035 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1036 {
1037 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1038 	int err = 0, cont = 1;
1039 	int level, offset[4], noffset[4];
1040 	unsigned int nofs = 0;
1041 	struct f2fs_inode *ri;
1042 	struct dnode_of_data dn;
1043 	struct page *page;
1044 
1045 	trace_f2fs_truncate_inode_blocks_enter(inode, from);
1046 
1047 	level = get_node_path(inode, from, offset, noffset);
1048 	if (level < 0) {
1049 		trace_f2fs_truncate_inode_blocks_exit(inode, level);
1050 		return level;
1051 	}
1052 
1053 	page = f2fs_get_node_page(sbi, inode->i_ino);
1054 	if (IS_ERR(page)) {
1055 		trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1056 		return PTR_ERR(page);
1057 	}
1058 
1059 	set_new_dnode(&dn, inode, page, NULL, 0);
1060 	unlock_page(page);
1061 
1062 	ri = F2FS_INODE(page);
1063 	switch (level) {
1064 	case 0:
1065 	case 1:
1066 		nofs = noffset[1];
1067 		break;
1068 	case 2:
1069 		nofs = noffset[1];
1070 		if (!offset[level - 1])
1071 			goto skip_partial;
1072 		err = truncate_partial_nodes(&dn, ri, offset, level);
1073 		if (err < 0 && err != -ENOENT)
1074 			goto fail;
1075 		nofs += 1 + NIDS_PER_BLOCK;
1076 		break;
1077 	case 3:
1078 		nofs = 5 + 2 * NIDS_PER_BLOCK;
1079 		if (!offset[level - 1])
1080 			goto skip_partial;
1081 		err = truncate_partial_nodes(&dn, ri, offset, level);
1082 		if (err < 0 && err != -ENOENT)
1083 			goto fail;
1084 		break;
1085 	default:
1086 		BUG();
1087 	}
1088 
1089 skip_partial:
1090 	while (cont) {
1091 		dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1092 		switch (offset[0]) {
1093 		case NODE_DIR1_BLOCK:
1094 		case NODE_DIR2_BLOCK:
1095 			err = truncate_dnode(&dn);
1096 			break;
1097 
1098 		case NODE_IND1_BLOCK:
1099 		case NODE_IND2_BLOCK:
1100 			err = truncate_nodes(&dn, nofs, offset[1], 2);
1101 			break;
1102 
1103 		case NODE_DIND_BLOCK:
1104 			err = truncate_nodes(&dn, nofs, offset[1], 3);
1105 			cont = 0;
1106 			break;
1107 
1108 		default:
1109 			BUG();
1110 		}
1111 		if (err < 0 && err != -ENOENT)
1112 			goto fail;
1113 		if (offset[1] == 0 &&
1114 				ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1115 			lock_page(page);
1116 			BUG_ON(page->mapping != NODE_MAPPING(sbi));
1117 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1118 			ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1119 			set_page_dirty(page);
1120 			unlock_page(page);
1121 		}
1122 		offset[1] = 0;
1123 		offset[0]++;
1124 		nofs += err;
1125 	}
1126 fail:
1127 	f2fs_put_page(page, 0);
1128 	trace_f2fs_truncate_inode_blocks_exit(inode, err);
1129 	return err > 0 ? 0 : err;
1130 }
1131 
1132 /* caller must lock inode page */
f2fs_truncate_xattr_node(struct inode * inode)1133 int f2fs_truncate_xattr_node(struct inode *inode)
1134 {
1135 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1136 	nid_t nid = F2FS_I(inode)->i_xattr_nid;
1137 	struct dnode_of_data dn;
1138 	struct page *npage;
1139 	int err;
1140 
1141 	if (!nid)
1142 		return 0;
1143 
1144 	npage = f2fs_get_node_page(sbi, nid);
1145 	if (IS_ERR(npage))
1146 		return PTR_ERR(npage);
1147 
1148 	set_new_dnode(&dn, inode, NULL, npage, nid);
1149 	err = truncate_node(&dn);
1150 	if (err) {
1151 		f2fs_put_page(npage, 1);
1152 		return err;
1153 	}
1154 
1155 	f2fs_i_xnid_write(inode, 0);
1156 
1157 	return 0;
1158 }
1159 
1160 /*
1161  * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1162  * f2fs_unlock_op().
1163  */
f2fs_remove_inode_page(struct inode * inode)1164 int f2fs_remove_inode_page(struct inode *inode)
1165 {
1166 	struct dnode_of_data dn;
1167 	int err;
1168 
1169 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1170 	err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1171 	if (err)
1172 		return err;
1173 
1174 	err = f2fs_truncate_xattr_node(inode);
1175 	if (err) {
1176 		f2fs_put_dnode(&dn);
1177 		return err;
1178 	}
1179 
1180 	/* remove potential inline_data blocks */
1181 	if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1182 				S_ISLNK(inode->i_mode))
1183 		f2fs_truncate_data_blocks_range(&dn, 1);
1184 
1185 	/* 0 is possible, after f2fs_new_inode() has failed */
1186 	if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1187 		f2fs_put_dnode(&dn);
1188 		return -EIO;
1189 	}
1190 
1191 	if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1192 		f2fs_warn(F2FS_I_SB(inode),
1193 			"f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1194 			inode->i_ino, (unsigned long long)inode->i_blocks);
1195 		set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1196 	}
1197 
1198 	/* will put inode & node pages */
1199 	err = truncate_node(&dn);
1200 	if (err) {
1201 		f2fs_put_dnode(&dn);
1202 		return err;
1203 	}
1204 	return 0;
1205 }
1206 
f2fs_new_inode_page(struct inode * inode)1207 struct page *f2fs_new_inode_page(struct inode *inode)
1208 {
1209 	struct dnode_of_data dn;
1210 
1211 	/* allocate inode page for new inode */
1212 	set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1213 
1214 	/* caller should f2fs_put_page(page, 1); */
1215 	return f2fs_new_node_page(&dn, 0);
1216 }
1217 
f2fs_new_node_page(struct dnode_of_data * dn,unsigned int ofs)1218 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1219 {
1220 	struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1221 	struct node_info new_ni;
1222 	struct page *page;
1223 	int err;
1224 
1225 	if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1226 		return ERR_PTR(-EPERM);
1227 
1228 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1229 	if (!page)
1230 		return ERR_PTR(-ENOMEM);
1231 
1232 	if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1233 		goto fail;
1234 
1235 #ifdef CONFIG_F2FS_CHECK_FS
1236 	err = f2fs_get_node_info(sbi, dn->nid, &new_ni);
1237 	if (err) {
1238 		dec_valid_node_count(sbi, dn->inode, !ofs);
1239 		goto fail;
1240 	}
1241 	f2fs_bug_on(sbi, new_ni.blk_addr != NULL_ADDR);
1242 #endif
1243 	new_ni.nid = dn->nid;
1244 	new_ni.ino = dn->inode->i_ino;
1245 	new_ni.blk_addr = NULL_ADDR;
1246 	new_ni.flag = 0;
1247 	new_ni.version = 0;
1248 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1249 
1250 	f2fs_wait_on_page_writeback(page, NODE, true, true);
1251 	fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1252 	set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1253 	if (!PageUptodate(page))
1254 		SetPageUptodate(page);
1255 	if (set_page_dirty(page))
1256 		dn->node_changed = true;
1257 
1258 	if (f2fs_has_xattr_block(ofs))
1259 		f2fs_i_xnid_write(dn->inode, dn->nid);
1260 
1261 	if (ofs == 0)
1262 		inc_valid_inode_count(sbi);
1263 	return page;
1264 
1265 fail:
1266 	clear_node_page_dirty(page);
1267 	f2fs_put_page(page, 1);
1268 	return ERR_PTR(err);
1269 }
1270 
1271 /*
1272  * Caller should do after getting the following values.
1273  * 0: f2fs_put_page(page, 0)
1274  * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1275  */
read_node_page(struct page * page,int op_flags)1276 static int read_node_page(struct page *page, int op_flags)
1277 {
1278 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1279 	struct node_info ni;
1280 	struct f2fs_io_info fio = {
1281 		.sbi = sbi,
1282 		.type = NODE,
1283 		.op = REQ_OP_READ,
1284 		.op_flags = op_flags,
1285 		.page = page,
1286 		.encrypted_page = NULL,
1287 	};
1288 	int err;
1289 
1290 	if (PageUptodate(page)) {
1291 		if (!f2fs_inode_chksum_verify(sbi, page)) {
1292 			ClearPageUptodate(page);
1293 			return -EFSBADCRC;
1294 		}
1295 		return LOCKED_PAGE;
1296 	}
1297 
1298 	err = f2fs_get_node_info(sbi, page->index, &ni);
1299 	if (err)
1300 		return err;
1301 
1302 	if (unlikely(ni.blk_addr == NULL_ADDR) ||
1303 			is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1304 		ClearPageUptodate(page);
1305 		return -ENOENT;
1306 	}
1307 
1308 	fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1309 
1310 	err = f2fs_submit_page_bio(&fio);
1311 
1312 	if (!err)
1313 		f2fs_update_iostat(sbi, FS_NODE_READ_IO, F2FS_BLKSIZE);
1314 
1315 	return err;
1316 }
1317 
1318 /*
1319  * Readahead a node page
1320  */
f2fs_ra_node_page(struct f2fs_sb_info * sbi,nid_t nid)1321 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1322 {
1323 	struct page *apage;
1324 	int err;
1325 
1326 	if (!nid)
1327 		return;
1328 	if (f2fs_check_nid_range(sbi, nid))
1329 		return;
1330 
1331 	apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1332 	if (apage)
1333 		return;
1334 
1335 	apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1336 	if (!apage)
1337 		return;
1338 
1339 	err = read_node_page(apage, REQ_RAHEAD);
1340 	f2fs_put_page(apage, err ? 1 : 0);
1341 }
1342 
__get_node_page(struct f2fs_sb_info * sbi,pgoff_t nid,struct page * parent,int start)1343 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1344 					struct page *parent, int start)
1345 {
1346 	struct page *page;
1347 	int err;
1348 
1349 	if (!nid)
1350 		return ERR_PTR(-ENOENT);
1351 	if (f2fs_check_nid_range(sbi, nid))
1352 		return ERR_PTR(-EINVAL);
1353 repeat:
1354 	page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1355 	if (!page)
1356 		return ERR_PTR(-ENOMEM);
1357 
1358 	err = read_node_page(page, 0);
1359 	if (err < 0) {
1360 		f2fs_put_page(page, 1);
1361 		return ERR_PTR(err);
1362 	} else if (err == LOCKED_PAGE) {
1363 		err = 0;
1364 		goto page_hit;
1365 	}
1366 
1367 	if (parent)
1368 		f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1369 
1370 	lock_page(page);
1371 
1372 	if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1373 		f2fs_put_page(page, 1);
1374 		goto repeat;
1375 	}
1376 
1377 	if (unlikely(!PageUptodate(page))) {
1378 		err = -EIO;
1379 		goto out_err;
1380 	}
1381 
1382 	if (!f2fs_inode_chksum_verify(sbi, page)) {
1383 		err = -EFSBADCRC;
1384 		goto out_err;
1385 	}
1386 page_hit:
1387 	if(unlikely(nid != nid_of_node(page))) {
1388 		f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1389 			  nid, nid_of_node(page), ino_of_node(page),
1390 			  ofs_of_node(page), cpver_of_node(page),
1391 			  next_blkaddr_of_node(page));
1392 		set_sbi_flag(sbi, SBI_NEED_FSCK);
1393 		err = -EINVAL;
1394 out_err:
1395 		ClearPageUptodate(page);
1396 		f2fs_put_page(page, 1);
1397 		return ERR_PTR(err);
1398 	}
1399 	return page;
1400 }
1401 
f2fs_get_node_page(struct f2fs_sb_info * sbi,pgoff_t nid)1402 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1403 {
1404 	return __get_node_page(sbi, nid, NULL, 0);
1405 }
1406 
f2fs_get_node_page_ra(struct page * parent,int start)1407 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1408 {
1409 	struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1410 	nid_t nid = get_nid(parent, start, false);
1411 
1412 	return __get_node_page(sbi, nid, parent, start);
1413 }
1414 
flush_inline_data(struct f2fs_sb_info * sbi,nid_t ino)1415 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1416 {
1417 	struct inode *inode;
1418 	struct page *page;
1419 	int ret;
1420 
1421 	/* should flush inline_data before evict_inode */
1422 	inode = ilookup(sbi->sb, ino);
1423 	if (!inode)
1424 		return;
1425 
1426 	page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1427 					FGP_LOCK|FGP_NOWAIT, 0);
1428 	if (!page)
1429 		goto iput_out;
1430 
1431 	if (!PageUptodate(page))
1432 		goto page_out;
1433 
1434 	if (!PageDirty(page))
1435 		goto page_out;
1436 
1437 	if (!clear_page_dirty_for_io(page))
1438 		goto page_out;
1439 
1440 	ret = f2fs_write_inline_data(inode, page);
1441 	inode_dec_dirty_pages(inode);
1442 	f2fs_remove_dirty_inode(inode);
1443 	if (ret)
1444 		set_page_dirty(page);
1445 page_out:
1446 	f2fs_put_page(page, 1);
1447 iput_out:
1448 	iput(inode);
1449 }
1450 
last_fsync_dnode(struct f2fs_sb_info * sbi,nid_t ino)1451 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1452 {
1453 	pgoff_t index;
1454 	struct pagevec pvec;
1455 	struct page *last_page = NULL;
1456 	int nr_pages;
1457 
1458 	pagevec_init(&pvec);
1459 	index = 0;
1460 
1461 	while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1462 				PAGECACHE_TAG_DIRTY))) {
1463 		int i;
1464 
1465 		for (i = 0; i < nr_pages; i++) {
1466 			struct page *page = pvec.pages[i];
1467 
1468 			if (unlikely(f2fs_cp_error(sbi))) {
1469 				f2fs_put_page(last_page, 0);
1470 				pagevec_release(&pvec);
1471 				return ERR_PTR(-EIO);
1472 			}
1473 
1474 			if (!IS_DNODE(page) || !is_cold_node(page))
1475 				continue;
1476 			if (ino_of_node(page) != ino)
1477 				continue;
1478 
1479 			lock_page(page);
1480 
1481 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1482 continue_unlock:
1483 				unlock_page(page);
1484 				continue;
1485 			}
1486 			if (ino_of_node(page) != ino)
1487 				goto continue_unlock;
1488 
1489 			if (!PageDirty(page)) {
1490 				/* someone wrote it for us */
1491 				goto continue_unlock;
1492 			}
1493 
1494 			if (last_page)
1495 				f2fs_put_page(last_page, 0);
1496 
1497 			get_page(page);
1498 			last_page = page;
1499 			unlock_page(page);
1500 		}
1501 		pagevec_release(&pvec);
1502 		cond_resched();
1503 	}
1504 	return last_page;
1505 }
1506 
__write_node_page(struct page * page,bool atomic,bool * submitted,struct writeback_control * wbc,bool do_balance,enum iostat_type io_type,unsigned int * seq_id)1507 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1508 				struct writeback_control *wbc, bool do_balance,
1509 				enum iostat_type io_type, unsigned int *seq_id)
1510 {
1511 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1512 	nid_t nid;
1513 	struct node_info ni;
1514 	struct f2fs_io_info fio = {
1515 		.sbi = sbi,
1516 		.ino = ino_of_node(page),
1517 		.type = NODE,
1518 		.op = REQ_OP_WRITE,
1519 		.op_flags = wbc_to_write_flags(wbc),
1520 		.page = page,
1521 		.encrypted_page = NULL,
1522 		.submitted = false,
1523 		.io_type = io_type,
1524 		.io_wbc = wbc,
1525 	};
1526 	unsigned int seq;
1527 
1528 	trace_f2fs_writepage(page, NODE);
1529 
1530 	if (unlikely(f2fs_cp_error(sbi))) {
1531 		if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
1532 			ClearPageUptodate(page);
1533 			dec_page_count(sbi, F2FS_DIRTY_NODES);
1534 			unlock_page(page);
1535 			return 0;
1536 		}
1537 		goto redirty_out;
1538 	}
1539 
1540 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1541 		goto redirty_out;
1542 
1543 	if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1544 			wbc->sync_mode == WB_SYNC_NONE &&
1545 			IS_DNODE(page) && is_cold_node(page))
1546 		goto redirty_out;
1547 
1548 	/* get old block addr of this node page */
1549 	nid = nid_of_node(page);
1550 	f2fs_bug_on(sbi, page->index != nid);
1551 
1552 	if (f2fs_get_node_info(sbi, nid, &ni))
1553 		goto redirty_out;
1554 
1555 	if (wbc->for_reclaim) {
1556 		if (!down_read_trylock(&sbi->node_write))
1557 			goto redirty_out;
1558 	} else {
1559 		down_read(&sbi->node_write);
1560 	}
1561 
1562 	/* This page is already truncated */
1563 	if (unlikely(ni.blk_addr == NULL_ADDR)) {
1564 		ClearPageUptodate(page);
1565 		dec_page_count(sbi, F2FS_DIRTY_NODES);
1566 		up_read(&sbi->node_write);
1567 		unlock_page(page);
1568 		return 0;
1569 	}
1570 
1571 	if (__is_valid_data_blkaddr(ni.blk_addr) &&
1572 		!f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1573 					DATA_GENERIC_ENHANCE)) {
1574 		up_read(&sbi->node_write);
1575 		goto redirty_out;
1576 	}
1577 
1578 	if (atomic && !test_opt(sbi, NOBARRIER))
1579 		fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1580 
1581 	/* should add to global list before clearing PAGECACHE status */
1582 	if (f2fs_in_warm_node_list(sbi, page)) {
1583 		seq = f2fs_add_fsync_node_entry(sbi, page);
1584 		if (seq_id)
1585 			*seq_id = seq;
1586 	}
1587 
1588 	set_page_writeback(page);
1589 	ClearPageError(page);
1590 
1591 	fio.old_blkaddr = ni.blk_addr;
1592 	f2fs_do_write_node_page(nid, &fio);
1593 	set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1594 	dec_page_count(sbi, F2FS_DIRTY_NODES);
1595 	up_read(&sbi->node_write);
1596 
1597 	if (wbc->for_reclaim) {
1598 		f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1599 		submitted = NULL;
1600 	}
1601 
1602 	unlock_page(page);
1603 
1604 	if (unlikely(f2fs_cp_error(sbi))) {
1605 		f2fs_submit_merged_write(sbi, NODE);
1606 		submitted = NULL;
1607 	}
1608 	if (submitted)
1609 		*submitted = fio.submitted;
1610 
1611 	if (do_balance)
1612 		f2fs_balance_fs(sbi, false);
1613 	return 0;
1614 
1615 redirty_out:
1616 	redirty_page_for_writepage(wbc, page);
1617 	return AOP_WRITEPAGE_ACTIVATE;
1618 }
1619 
f2fs_move_node_page(struct page * node_page,int gc_type)1620 int f2fs_move_node_page(struct page *node_page, int gc_type)
1621 {
1622 	int err = 0;
1623 
1624 	if (gc_type == FG_GC) {
1625 		struct writeback_control wbc = {
1626 			.sync_mode = WB_SYNC_ALL,
1627 			.nr_to_write = 1,
1628 			.for_reclaim = 0,
1629 		};
1630 
1631 		f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1632 
1633 		set_page_dirty(node_page);
1634 
1635 		if (!clear_page_dirty_for_io(node_page)) {
1636 			err = -EAGAIN;
1637 			goto out_page;
1638 		}
1639 
1640 		if (__write_node_page(node_page, false, NULL,
1641 					&wbc, false, FS_GC_NODE_IO, NULL)) {
1642 			err = -EAGAIN;
1643 			unlock_page(node_page);
1644 		}
1645 		goto release_page;
1646 	} else {
1647 		/* set page dirty and write it */
1648 		if (!PageWriteback(node_page))
1649 			set_page_dirty(node_page);
1650 	}
1651 out_page:
1652 	unlock_page(node_page);
1653 release_page:
1654 	f2fs_put_page(node_page, 0);
1655 	return err;
1656 }
1657 
f2fs_write_node_page(struct page * page,struct writeback_control * wbc)1658 static int f2fs_write_node_page(struct page *page,
1659 				struct writeback_control *wbc)
1660 {
1661 	return __write_node_page(page, false, NULL, wbc, false,
1662 						FS_NODE_IO, NULL);
1663 }
1664 
f2fs_fsync_node_pages(struct f2fs_sb_info * sbi,struct inode * inode,struct writeback_control * wbc,bool atomic,unsigned int * seq_id)1665 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1666 			struct writeback_control *wbc, bool atomic,
1667 			unsigned int *seq_id)
1668 {
1669 	pgoff_t index;
1670 	struct pagevec pvec;
1671 	int ret = 0;
1672 	struct page *last_page = NULL;
1673 	bool marked = false;
1674 	nid_t ino = inode->i_ino;
1675 	int nr_pages;
1676 	int nwritten = 0;
1677 
1678 	if (atomic) {
1679 		last_page = last_fsync_dnode(sbi, ino);
1680 		if (IS_ERR_OR_NULL(last_page))
1681 			return PTR_ERR_OR_ZERO(last_page);
1682 	}
1683 retry:
1684 	pagevec_init(&pvec);
1685 	index = 0;
1686 
1687 	while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1688 				PAGECACHE_TAG_DIRTY))) {
1689 		int i;
1690 
1691 		for (i = 0; i < nr_pages; i++) {
1692 			struct page *page = pvec.pages[i];
1693 			bool submitted = false;
1694 
1695 			if (unlikely(f2fs_cp_error(sbi))) {
1696 				f2fs_put_page(last_page, 0);
1697 				pagevec_release(&pvec);
1698 				ret = -EIO;
1699 				goto out;
1700 			}
1701 
1702 			if (!IS_DNODE(page) || !is_cold_node(page))
1703 				continue;
1704 			if (ino_of_node(page) != ino)
1705 				continue;
1706 
1707 			lock_page(page);
1708 
1709 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1710 continue_unlock:
1711 				unlock_page(page);
1712 				continue;
1713 			}
1714 			if (ino_of_node(page) != ino)
1715 				goto continue_unlock;
1716 
1717 			if (!PageDirty(page) && page != last_page) {
1718 				/* someone wrote it for us */
1719 				goto continue_unlock;
1720 			}
1721 
1722 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1723 
1724 			set_fsync_mark(page, 0);
1725 			set_dentry_mark(page, 0);
1726 
1727 			if (!atomic || page == last_page) {
1728 				set_fsync_mark(page, 1);
1729 				if (IS_INODE(page)) {
1730 					if (is_inode_flag_set(inode,
1731 								FI_DIRTY_INODE))
1732 						f2fs_update_inode(inode, page);
1733 					set_dentry_mark(page,
1734 						f2fs_need_dentry_mark(sbi, ino));
1735 				}
1736 				/* may be written by other thread */
1737 				if (!PageDirty(page))
1738 					set_page_dirty(page);
1739 			}
1740 
1741 			if (!clear_page_dirty_for_io(page))
1742 				goto continue_unlock;
1743 
1744 			ret = __write_node_page(page, atomic &&
1745 						page == last_page,
1746 						&submitted, wbc, true,
1747 						FS_NODE_IO, seq_id);
1748 			if (ret) {
1749 				unlock_page(page);
1750 				f2fs_put_page(last_page, 0);
1751 				break;
1752 			} else if (submitted) {
1753 				nwritten++;
1754 			}
1755 
1756 			if (page == last_page) {
1757 				f2fs_put_page(page, 0);
1758 				marked = true;
1759 				break;
1760 			}
1761 		}
1762 		pagevec_release(&pvec);
1763 		cond_resched();
1764 
1765 		if (ret || marked)
1766 			break;
1767 	}
1768 	if (!ret && atomic && !marked) {
1769 		f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1770 			   ino, last_page->index);
1771 		lock_page(last_page);
1772 		f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1773 		set_page_dirty(last_page);
1774 		unlock_page(last_page);
1775 		goto retry;
1776 	}
1777 out:
1778 	if (nwritten)
1779 		f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1780 	return ret ? -EIO: 0;
1781 }
1782 
f2fs_match_ino(struct inode * inode,unsigned long ino,void * data)1783 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1784 {
1785 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1786 	bool clean;
1787 
1788 	if (inode->i_ino != ino)
1789 		return 0;
1790 
1791 	if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1792 		return 0;
1793 
1794 	spin_lock(&sbi->inode_lock[DIRTY_META]);
1795 	clean = list_empty(&F2FS_I(inode)->gdirty_list);
1796 	spin_unlock(&sbi->inode_lock[DIRTY_META]);
1797 
1798 	if (clean)
1799 		return 0;
1800 
1801 	inode = igrab(inode);
1802 	if (!inode)
1803 		return 0;
1804 	return 1;
1805 }
1806 
flush_dirty_inode(struct page * page)1807 static bool flush_dirty_inode(struct page *page)
1808 {
1809 	struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1810 	struct inode *inode;
1811 	nid_t ino = ino_of_node(page);
1812 
1813 	inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1814 	if (!inode)
1815 		return false;
1816 
1817 	f2fs_update_inode(inode, page);
1818 	unlock_page(page);
1819 
1820 	iput(inode);
1821 	return true;
1822 }
1823 
f2fs_flush_inline_data(struct f2fs_sb_info * sbi)1824 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1825 {
1826 	pgoff_t index = 0;
1827 	struct pagevec pvec;
1828 	int nr_pages;
1829 
1830 	pagevec_init(&pvec);
1831 
1832 	while ((nr_pages = pagevec_lookup_tag(&pvec,
1833 			NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1834 		int i;
1835 
1836 		for (i = 0; i < nr_pages; i++) {
1837 			struct page *page = pvec.pages[i];
1838 
1839 			if (!IS_DNODE(page))
1840 				continue;
1841 
1842 			lock_page(page);
1843 
1844 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1845 continue_unlock:
1846 				unlock_page(page);
1847 				continue;
1848 			}
1849 
1850 			if (!PageDirty(page)) {
1851 				/* someone wrote it for us */
1852 				goto continue_unlock;
1853 			}
1854 
1855 			/* flush inline_data, if it's async context. */
1856 			if (is_inline_node(page)) {
1857 				clear_inline_node(page);
1858 				unlock_page(page);
1859 				flush_inline_data(sbi, ino_of_node(page));
1860 				continue;
1861 			}
1862 			unlock_page(page);
1863 		}
1864 		pagevec_release(&pvec);
1865 		cond_resched();
1866 	}
1867 }
1868 
f2fs_sync_node_pages(struct f2fs_sb_info * sbi,struct writeback_control * wbc,bool do_balance,enum iostat_type io_type)1869 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1870 				struct writeback_control *wbc,
1871 				bool do_balance, enum iostat_type io_type)
1872 {
1873 	pgoff_t index;
1874 	struct pagevec pvec;
1875 	int step = 0;
1876 	int nwritten = 0;
1877 	int ret = 0;
1878 	int nr_pages, done = 0;
1879 
1880 	pagevec_init(&pvec);
1881 
1882 next_step:
1883 	index = 0;
1884 
1885 	while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1886 			NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1887 		int i;
1888 
1889 		for (i = 0; i < nr_pages; i++) {
1890 			struct page *page = pvec.pages[i];
1891 			bool submitted = false;
1892 			bool may_dirty = true;
1893 
1894 			/* give a priority to WB_SYNC threads */
1895 			if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1896 					wbc->sync_mode == WB_SYNC_NONE) {
1897 				done = 1;
1898 				break;
1899 			}
1900 
1901 			/*
1902 			 * flushing sequence with step:
1903 			 * 0. indirect nodes
1904 			 * 1. dentry dnodes
1905 			 * 2. file dnodes
1906 			 */
1907 			if (step == 0 && IS_DNODE(page))
1908 				continue;
1909 			if (step == 1 && (!IS_DNODE(page) ||
1910 						is_cold_node(page)))
1911 				continue;
1912 			if (step == 2 && (!IS_DNODE(page) ||
1913 						!is_cold_node(page)))
1914 				continue;
1915 lock_node:
1916 			if (wbc->sync_mode == WB_SYNC_ALL)
1917 				lock_page(page);
1918 			else if (!trylock_page(page))
1919 				continue;
1920 
1921 			if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1922 continue_unlock:
1923 				unlock_page(page);
1924 				continue;
1925 			}
1926 
1927 			if (!PageDirty(page)) {
1928 				/* someone wrote it for us */
1929 				goto continue_unlock;
1930 			}
1931 
1932 			/* flush inline_data/inode, if it's async context. */
1933 			if (!do_balance)
1934 				goto write_node;
1935 
1936 			/* flush inline_data */
1937 			if (is_inline_node(page)) {
1938 				clear_inline_node(page);
1939 				unlock_page(page);
1940 				flush_inline_data(sbi, ino_of_node(page));
1941 				goto lock_node;
1942 			}
1943 
1944 			/* flush dirty inode */
1945 			if (IS_INODE(page) && may_dirty) {
1946 				may_dirty = false;
1947 				if (flush_dirty_inode(page))
1948 					goto lock_node;
1949 			}
1950 write_node:
1951 			f2fs_wait_on_page_writeback(page, NODE, true, true);
1952 
1953 			if (!clear_page_dirty_for_io(page))
1954 				goto continue_unlock;
1955 
1956 			set_fsync_mark(page, 0);
1957 			set_dentry_mark(page, 0);
1958 
1959 			ret = __write_node_page(page, false, &submitted,
1960 						wbc, do_balance, io_type, NULL);
1961 			if (ret)
1962 				unlock_page(page);
1963 			else if (submitted)
1964 				nwritten++;
1965 
1966 			if (--wbc->nr_to_write == 0)
1967 				break;
1968 		}
1969 		pagevec_release(&pvec);
1970 		cond_resched();
1971 
1972 		if (wbc->nr_to_write == 0) {
1973 			step = 2;
1974 			break;
1975 		}
1976 	}
1977 
1978 	if (step < 2) {
1979 		if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1980 				wbc->sync_mode == WB_SYNC_NONE && step == 1)
1981 			goto out;
1982 		step++;
1983 		goto next_step;
1984 	}
1985 out:
1986 	if (nwritten)
1987 		f2fs_submit_merged_write(sbi, NODE);
1988 
1989 	if (unlikely(f2fs_cp_error(sbi)))
1990 		return -EIO;
1991 	return ret;
1992 }
1993 
f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info * sbi,unsigned int seq_id)1994 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
1995 						unsigned int seq_id)
1996 {
1997 	struct fsync_node_entry *fn;
1998 	struct page *page;
1999 	struct list_head *head = &sbi->fsync_node_list;
2000 	unsigned long flags;
2001 	unsigned int cur_seq_id = 0;
2002 	int ret2, ret = 0;
2003 
2004 	while (seq_id && cur_seq_id < seq_id) {
2005 		spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2006 		if (list_empty(head)) {
2007 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2008 			break;
2009 		}
2010 		fn = list_first_entry(head, struct fsync_node_entry, list);
2011 		if (fn->seq_id > seq_id) {
2012 			spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2013 			break;
2014 		}
2015 		cur_seq_id = fn->seq_id;
2016 		page = fn->page;
2017 		get_page(page);
2018 		spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2019 
2020 		f2fs_wait_on_page_writeback(page, NODE, true, false);
2021 		if (TestClearPageError(page))
2022 			ret = -EIO;
2023 
2024 		put_page(page);
2025 
2026 		if (ret)
2027 			break;
2028 	}
2029 
2030 	ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2031 	if (!ret)
2032 		ret = ret2;
2033 
2034 	return ret;
2035 }
2036 
f2fs_write_node_pages(struct address_space * mapping,struct writeback_control * wbc)2037 static int f2fs_write_node_pages(struct address_space *mapping,
2038 			    struct writeback_control *wbc)
2039 {
2040 	struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2041 	struct blk_plug plug;
2042 	long diff;
2043 
2044 	if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2045 		goto skip_write;
2046 
2047 	/* balancing f2fs's metadata in background */
2048 	f2fs_balance_fs_bg(sbi, true);
2049 
2050 	/* collect a number of dirty node pages and write together */
2051 	if (wbc->sync_mode != WB_SYNC_ALL &&
2052 			get_pages(sbi, F2FS_DIRTY_NODES) <
2053 					nr_pages_to_skip(sbi, NODE))
2054 		goto skip_write;
2055 
2056 	if (wbc->sync_mode == WB_SYNC_ALL)
2057 		atomic_inc(&sbi->wb_sync_req[NODE]);
2058 	else if (atomic_read(&sbi->wb_sync_req[NODE]))
2059 		goto skip_write;
2060 
2061 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2062 
2063 	diff = nr_pages_to_write(sbi, NODE, wbc);
2064 	blk_start_plug(&plug);
2065 	f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2066 	blk_finish_plug(&plug);
2067 	wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2068 
2069 	if (wbc->sync_mode == WB_SYNC_ALL)
2070 		atomic_dec(&sbi->wb_sync_req[NODE]);
2071 	return 0;
2072 
2073 skip_write:
2074 	wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2075 	trace_f2fs_writepages(mapping->host, wbc, NODE);
2076 	return 0;
2077 }
2078 
f2fs_set_node_page_dirty(struct page * page)2079 static int f2fs_set_node_page_dirty(struct page *page)
2080 {
2081 	trace_f2fs_set_page_dirty(page, NODE);
2082 
2083 	if (!PageUptodate(page))
2084 		SetPageUptodate(page);
2085 #ifdef CONFIG_F2FS_CHECK_FS
2086 	if (IS_INODE(page))
2087 		f2fs_inode_chksum_set(F2FS_P_SB(page), page);
2088 #endif
2089 	if (!PageDirty(page)) {
2090 		__set_page_dirty_nobuffers(page);
2091 		inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
2092 		f2fs_set_page_private(page, 0);
2093 		f2fs_trace_pid(page);
2094 		return 1;
2095 	}
2096 	return 0;
2097 }
2098 
2099 /*
2100  * Structure of the f2fs node operations
2101  */
2102 const struct address_space_operations f2fs_node_aops = {
2103 	.writepage	= f2fs_write_node_page,
2104 	.writepages	= f2fs_write_node_pages,
2105 	.set_page_dirty	= f2fs_set_node_page_dirty,
2106 	.invalidatepage	= f2fs_invalidate_page,
2107 	.releasepage	= f2fs_release_page,
2108 #ifdef CONFIG_MIGRATION
2109 	.migratepage	= f2fs_migrate_page,
2110 #endif
2111 };
2112 
__lookup_free_nid_list(struct f2fs_nm_info * nm_i,nid_t n)2113 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2114 						nid_t n)
2115 {
2116 	return radix_tree_lookup(&nm_i->free_nid_root, n);
2117 }
2118 
__insert_free_nid(struct f2fs_sb_info * sbi,struct free_nid * i)2119 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2120 				struct free_nid *i)
2121 {
2122 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2123 
2124 	int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2125 	if (err)
2126 		return err;
2127 
2128 	nm_i->nid_cnt[FREE_NID]++;
2129 	list_add_tail(&i->list, &nm_i->free_nid_list);
2130 	return 0;
2131 }
2132 
__remove_free_nid(struct f2fs_sb_info * sbi,struct free_nid * i,enum nid_state state)2133 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2134 			struct free_nid *i, enum nid_state state)
2135 {
2136 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2137 
2138 	f2fs_bug_on(sbi, state != i->state);
2139 	nm_i->nid_cnt[state]--;
2140 	if (state == FREE_NID)
2141 		list_del(&i->list);
2142 	radix_tree_delete(&nm_i->free_nid_root, i->nid);
2143 }
2144 
__move_free_nid(struct f2fs_sb_info * sbi,struct free_nid * i,enum nid_state org_state,enum nid_state dst_state)2145 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2146 			enum nid_state org_state, enum nid_state dst_state)
2147 {
2148 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2149 
2150 	f2fs_bug_on(sbi, org_state != i->state);
2151 	i->state = dst_state;
2152 	nm_i->nid_cnt[org_state]--;
2153 	nm_i->nid_cnt[dst_state]++;
2154 
2155 	switch (dst_state) {
2156 	case PREALLOC_NID:
2157 		list_del(&i->list);
2158 		break;
2159 	case FREE_NID:
2160 		list_add_tail(&i->list, &nm_i->free_nid_list);
2161 		break;
2162 	default:
2163 		BUG_ON(1);
2164 	}
2165 }
2166 
update_free_nid_bitmap(struct f2fs_sb_info * sbi,nid_t nid,bool set,bool build)2167 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2168 							bool set, bool build)
2169 {
2170 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2171 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2172 	unsigned int nid_ofs = nid - START_NID(nid);
2173 
2174 	if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2175 		return;
2176 
2177 	if (set) {
2178 		if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2179 			return;
2180 		__set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2181 		nm_i->free_nid_count[nat_ofs]++;
2182 	} else {
2183 		if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2184 			return;
2185 		__clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2186 		if (!build)
2187 			nm_i->free_nid_count[nat_ofs]--;
2188 	}
2189 }
2190 
2191 /* return if the nid is recognized as free */
add_free_nid(struct f2fs_sb_info * sbi,nid_t nid,bool build,bool update)2192 static bool add_free_nid(struct f2fs_sb_info *sbi,
2193 				nid_t nid, bool build, bool update)
2194 {
2195 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2196 	struct free_nid *i, *e;
2197 	struct nat_entry *ne;
2198 	int err = -EINVAL;
2199 	bool ret = false;
2200 
2201 	/* 0 nid should not be used */
2202 	if (unlikely(nid == 0))
2203 		return false;
2204 
2205 	if (unlikely(f2fs_check_nid_range(sbi, nid)))
2206 		return false;
2207 
2208 	i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2209 	i->nid = nid;
2210 	i->state = FREE_NID;
2211 
2212 	radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2213 
2214 	spin_lock(&nm_i->nid_list_lock);
2215 
2216 	if (build) {
2217 		/*
2218 		 *   Thread A             Thread B
2219 		 *  - f2fs_create
2220 		 *   - f2fs_new_inode
2221 		 *    - f2fs_alloc_nid
2222 		 *     - __insert_nid_to_list(PREALLOC_NID)
2223 		 *                     - f2fs_balance_fs_bg
2224 		 *                      - f2fs_build_free_nids
2225 		 *                       - __f2fs_build_free_nids
2226 		 *                        - scan_nat_page
2227 		 *                         - add_free_nid
2228 		 *                          - __lookup_nat_cache
2229 		 *  - f2fs_add_link
2230 		 *   - f2fs_init_inode_metadata
2231 		 *    - f2fs_new_inode_page
2232 		 *     - f2fs_new_node_page
2233 		 *      - set_node_addr
2234 		 *  - f2fs_alloc_nid_done
2235 		 *   - __remove_nid_from_list(PREALLOC_NID)
2236 		 *                         - __insert_nid_to_list(FREE_NID)
2237 		 */
2238 		ne = __lookup_nat_cache(nm_i, nid);
2239 		if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2240 				nat_get_blkaddr(ne) != NULL_ADDR))
2241 			goto err_out;
2242 
2243 		e = __lookup_free_nid_list(nm_i, nid);
2244 		if (e) {
2245 			if (e->state == FREE_NID)
2246 				ret = true;
2247 			goto err_out;
2248 		}
2249 	}
2250 	ret = true;
2251 	err = __insert_free_nid(sbi, i);
2252 err_out:
2253 	if (update) {
2254 		update_free_nid_bitmap(sbi, nid, ret, build);
2255 		if (!build)
2256 			nm_i->available_nids++;
2257 	}
2258 	spin_unlock(&nm_i->nid_list_lock);
2259 	radix_tree_preload_end();
2260 
2261 	if (err)
2262 		kmem_cache_free(free_nid_slab, i);
2263 	return ret;
2264 }
2265 
remove_free_nid(struct f2fs_sb_info * sbi,nid_t nid)2266 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2267 {
2268 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2269 	struct free_nid *i;
2270 	bool need_free = false;
2271 
2272 	spin_lock(&nm_i->nid_list_lock);
2273 	i = __lookup_free_nid_list(nm_i, nid);
2274 	if (i && i->state == FREE_NID) {
2275 		__remove_free_nid(sbi, i, FREE_NID);
2276 		need_free = true;
2277 	}
2278 	spin_unlock(&nm_i->nid_list_lock);
2279 
2280 	if (need_free)
2281 		kmem_cache_free(free_nid_slab, i);
2282 }
2283 
scan_nat_page(struct f2fs_sb_info * sbi,struct page * nat_page,nid_t start_nid)2284 static int scan_nat_page(struct f2fs_sb_info *sbi,
2285 			struct page *nat_page, nid_t start_nid)
2286 {
2287 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2288 	struct f2fs_nat_block *nat_blk = page_address(nat_page);
2289 	block_t blk_addr;
2290 	unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2291 	int i;
2292 
2293 	__set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2294 
2295 	i = start_nid % NAT_ENTRY_PER_BLOCK;
2296 
2297 	for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2298 		if (unlikely(start_nid >= nm_i->max_nid))
2299 			break;
2300 
2301 		blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2302 
2303 		if (blk_addr == NEW_ADDR)
2304 			return -EINVAL;
2305 
2306 		if (blk_addr == NULL_ADDR) {
2307 			add_free_nid(sbi, start_nid, true, true);
2308 		} else {
2309 			spin_lock(&NM_I(sbi)->nid_list_lock);
2310 			update_free_nid_bitmap(sbi, start_nid, false, true);
2311 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2312 		}
2313 	}
2314 
2315 	return 0;
2316 }
2317 
scan_curseg_cache(struct f2fs_sb_info * sbi)2318 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2319 {
2320 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2321 	struct f2fs_journal *journal = curseg->journal;
2322 	int i;
2323 
2324 	down_read(&curseg->journal_rwsem);
2325 	for (i = 0; i < nats_in_cursum(journal); i++) {
2326 		block_t addr;
2327 		nid_t nid;
2328 
2329 		addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2330 		nid = le32_to_cpu(nid_in_journal(journal, i));
2331 		if (addr == NULL_ADDR)
2332 			add_free_nid(sbi, nid, true, false);
2333 		else
2334 			remove_free_nid(sbi, nid);
2335 	}
2336 	up_read(&curseg->journal_rwsem);
2337 }
2338 
scan_free_nid_bits(struct f2fs_sb_info * sbi)2339 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2340 {
2341 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2342 	unsigned int i, idx;
2343 	nid_t nid;
2344 
2345 	down_read(&nm_i->nat_tree_lock);
2346 
2347 	for (i = 0; i < nm_i->nat_blocks; i++) {
2348 		if (!test_bit_le(i, nm_i->nat_block_bitmap))
2349 			continue;
2350 		if (!nm_i->free_nid_count[i])
2351 			continue;
2352 		for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2353 			idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2354 						NAT_ENTRY_PER_BLOCK, idx);
2355 			if (idx >= NAT_ENTRY_PER_BLOCK)
2356 				break;
2357 
2358 			nid = i * NAT_ENTRY_PER_BLOCK + idx;
2359 			add_free_nid(sbi, nid, true, false);
2360 
2361 			if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2362 				goto out;
2363 		}
2364 	}
2365 out:
2366 	scan_curseg_cache(sbi);
2367 
2368 	up_read(&nm_i->nat_tree_lock);
2369 }
2370 
__f2fs_build_free_nids(struct f2fs_sb_info * sbi,bool sync,bool mount)2371 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2372 						bool sync, bool mount)
2373 {
2374 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2375 	int i = 0, ret;
2376 	nid_t nid = nm_i->next_scan_nid;
2377 
2378 	if (unlikely(nid >= nm_i->max_nid))
2379 		nid = 0;
2380 
2381 	if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2382 		nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2383 
2384 	/* Enough entries */
2385 	if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2386 		return 0;
2387 
2388 	if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2389 		return 0;
2390 
2391 	if (!mount) {
2392 		/* try to find free nids in free_nid_bitmap */
2393 		scan_free_nid_bits(sbi);
2394 
2395 		if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2396 			return 0;
2397 	}
2398 
2399 	/* readahead nat pages to be scanned */
2400 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2401 							META_NAT, true);
2402 
2403 	down_read(&nm_i->nat_tree_lock);
2404 
2405 	while (1) {
2406 		if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2407 						nm_i->nat_block_bitmap)) {
2408 			struct page *page = get_current_nat_page(sbi, nid);
2409 
2410 			if (IS_ERR(page)) {
2411 				ret = PTR_ERR(page);
2412 			} else {
2413 				ret = scan_nat_page(sbi, page, nid);
2414 				f2fs_put_page(page, 1);
2415 			}
2416 
2417 			if (ret) {
2418 				up_read(&nm_i->nat_tree_lock);
2419 				f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2420 				return ret;
2421 			}
2422 		}
2423 
2424 		nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2425 		if (unlikely(nid >= nm_i->max_nid))
2426 			nid = 0;
2427 
2428 		if (++i >= FREE_NID_PAGES)
2429 			break;
2430 	}
2431 
2432 	/* go to the next free nat pages to find free nids abundantly */
2433 	nm_i->next_scan_nid = nid;
2434 
2435 	/* find free nids from current sum_pages */
2436 	scan_curseg_cache(sbi);
2437 
2438 	up_read(&nm_i->nat_tree_lock);
2439 
2440 	f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2441 					nm_i->ra_nid_pages, META_NAT, false);
2442 
2443 	return 0;
2444 }
2445 
f2fs_build_free_nids(struct f2fs_sb_info * sbi,bool sync,bool mount)2446 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2447 {
2448 	int ret;
2449 
2450 	mutex_lock(&NM_I(sbi)->build_lock);
2451 	ret = __f2fs_build_free_nids(sbi, sync, mount);
2452 	mutex_unlock(&NM_I(sbi)->build_lock);
2453 
2454 	return ret;
2455 }
2456 
2457 /*
2458  * If this function returns success, caller can obtain a new nid
2459  * from second parameter of this function.
2460  * The returned nid could be used ino as well as nid when inode is created.
2461  */
f2fs_alloc_nid(struct f2fs_sb_info * sbi,nid_t * nid)2462 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2463 {
2464 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2465 	struct free_nid *i = NULL;
2466 retry:
2467 	if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2468 		f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2469 		return false;
2470 	}
2471 
2472 	spin_lock(&nm_i->nid_list_lock);
2473 
2474 	if (unlikely(nm_i->available_nids == 0)) {
2475 		spin_unlock(&nm_i->nid_list_lock);
2476 		return false;
2477 	}
2478 
2479 	/* We should not use stale free nids created by f2fs_build_free_nids */
2480 	if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2481 		f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2482 		i = list_first_entry(&nm_i->free_nid_list,
2483 					struct free_nid, list);
2484 		*nid = i->nid;
2485 
2486 		__move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2487 		nm_i->available_nids--;
2488 
2489 		update_free_nid_bitmap(sbi, *nid, false, false);
2490 
2491 		spin_unlock(&nm_i->nid_list_lock);
2492 		return true;
2493 	}
2494 	spin_unlock(&nm_i->nid_list_lock);
2495 
2496 	/* Let's scan nat pages and its caches to get free nids */
2497 	if (!f2fs_build_free_nids(sbi, true, false))
2498 		goto retry;
2499 	return false;
2500 }
2501 
2502 /*
2503  * f2fs_alloc_nid() should be called prior to this function.
2504  */
f2fs_alloc_nid_done(struct f2fs_sb_info * sbi,nid_t nid)2505 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2506 {
2507 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2508 	struct free_nid *i;
2509 
2510 	spin_lock(&nm_i->nid_list_lock);
2511 	i = __lookup_free_nid_list(nm_i, nid);
2512 	f2fs_bug_on(sbi, !i);
2513 	__remove_free_nid(sbi, i, PREALLOC_NID);
2514 	spin_unlock(&nm_i->nid_list_lock);
2515 
2516 	kmem_cache_free(free_nid_slab, i);
2517 }
2518 
2519 /*
2520  * f2fs_alloc_nid() should be called prior to this function.
2521  */
f2fs_alloc_nid_failed(struct f2fs_sb_info * sbi,nid_t nid)2522 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2523 {
2524 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2525 	struct free_nid *i;
2526 	bool need_free = false;
2527 
2528 	if (!nid)
2529 		return;
2530 
2531 	spin_lock(&nm_i->nid_list_lock);
2532 	i = __lookup_free_nid_list(nm_i, nid);
2533 	f2fs_bug_on(sbi, !i);
2534 
2535 	if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2536 		__remove_free_nid(sbi, i, PREALLOC_NID);
2537 		need_free = true;
2538 	} else {
2539 		__move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2540 	}
2541 
2542 	nm_i->available_nids++;
2543 
2544 	update_free_nid_bitmap(sbi, nid, true, false);
2545 
2546 	spin_unlock(&nm_i->nid_list_lock);
2547 
2548 	if (need_free)
2549 		kmem_cache_free(free_nid_slab, i);
2550 }
2551 
f2fs_try_to_free_nids(struct f2fs_sb_info * sbi,int nr_shrink)2552 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2553 {
2554 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2555 	int nr = nr_shrink;
2556 
2557 	if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2558 		return 0;
2559 
2560 	if (!mutex_trylock(&nm_i->build_lock))
2561 		return 0;
2562 
2563 	while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2564 		struct free_nid *i, *next;
2565 		unsigned int batch = SHRINK_NID_BATCH_SIZE;
2566 
2567 		spin_lock(&nm_i->nid_list_lock);
2568 		list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2569 			if (!nr_shrink || !batch ||
2570 				nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2571 				break;
2572 			__remove_free_nid(sbi, i, FREE_NID);
2573 			kmem_cache_free(free_nid_slab, i);
2574 			nr_shrink--;
2575 			batch--;
2576 		}
2577 		spin_unlock(&nm_i->nid_list_lock);
2578 	}
2579 
2580 	mutex_unlock(&nm_i->build_lock);
2581 
2582 	return nr - nr_shrink;
2583 }
2584 
f2fs_recover_inline_xattr(struct inode * inode,struct page * page)2585 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2586 {
2587 	void *src_addr, *dst_addr;
2588 	size_t inline_size;
2589 	struct page *ipage;
2590 	struct f2fs_inode *ri;
2591 
2592 	ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2593 	if (IS_ERR(ipage))
2594 		return PTR_ERR(ipage);
2595 
2596 	ri = F2FS_INODE(page);
2597 	if (ri->i_inline & F2FS_INLINE_XATTR) {
2598 		set_inode_flag(inode, FI_INLINE_XATTR);
2599 	} else {
2600 		clear_inode_flag(inode, FI_INLINE_XATTR);
2601 		goto update_inode;
2602 	}
2603 
2604 	dst_addr = inline_xattr_addr(inode, ipage);
2605 	src_addr = inline_xattr_addr(inode, page);
2606 	inline_size = inline_xattr_size(inode);
2607 
2608 	f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2609 	memcpy(dst_addr, src_addr, inline_size);
2610 update_inode:
2611 	f2fs_update_inode(inode, ipage);
2612 	f2fs_put_page(ipage, 1);
2613 	return 0;
2614 }
2615 
f2fs_recover_xattr_data(struct inode * inode,struct page * page)2616 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2617 {
2618 	struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2619 	nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2620 	nid_t new_xnid;
2621 	struct dnode_of_data dn;
2622 	struct node_info ni;
2623 	struct page *xpage;
2624 	int err;
2625 
2626 	if (!prev_xnid)
2627 		goto recover_xnid;
2628 
2629 	/* 1: invalidate the previous xattr nid */
2630 	err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2631 	if (err)
2632 		return err;
2633 
2634 	f2fs_invalidate_blocks(sbi, ni.blk_addr);
2635 	dec_valid_node_count(sbi, inode, false);
2636 	set_node_addr(sbi, &ni, NULL_ADDR, false);
2637 
2638 recover_xnid:
2639 	/* 2: update xattr nid in inode */
2640 	if (!f2fs_alloc_nid(sbi, &new_xnid))
2641 		return -ENOSPC;
2642 
2643 	set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2644 	xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2645 	if (IS_ERR(xpage)) {
2646 		f2fs_alloc_nid_failed(sbi, new_xnid);
2647 		return PTR_ERR(xpage);
2648 	}
2649 
2650 	f2fs_alloc_nid_done(sbi, new_xnid);
2651 	f2fs_update_inode_page(inode);
2652 
2653 	/* 3: update and set xattr node page dirty */
2654 	memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2655 
2656 	set_page_dirty(xpage);
2657 	f2fs_put_page(xpage, 1);
2658 
2659 	return 0;
2660 }
2661 
f2fs_recover_inode_page(struct f2fs_sb_info * sbi,struct page * page)2662 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2663 {
2664 	struct f2fs_inode *src, *dst;
2665 	nid_t ino = ino_of_node(page);
2666 	struct node_info old_ni, new_ni;
2667 	struct page *ipage;
2668 	int err;
2669 
2670 	err = f2fs_get_node_info(sbi, ino, &old_ni);
2671 	if (err)
2672 		return err;
2673 
2674 	if (unlikely(old_ni.blk_addr != NULL_ADDR))
2675 		return -EINVAL;
2676 retry:
2677 	ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2678 	if (!ipage) {
2679 		congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2680 		goto retry;
2681 	}
2682 
2683 	/* Should not use this inode from free nid list */
2684 	remove_free_nid(sbi, ino);
2685 
2686 	if (!PageUptodate(ipage))
2687 		SetPageUptodate(ipage);
2688 	fill_node_footer(ipage, ino, ino, 0, true);
2689 	set_cold_node(ipage, false);
2690 
2691 	src = F2FS_INODE(page);
2692 	dst = F2FS_INODE(ipage);
2693 
2694 	memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2695 	dst->i_size = 0;
2696 	dst->i_blocks = cpu_to_le64(1);
2697 	dst->i_links = cpu_to_le32(1);
2698 	dst->i_xattr_nid = 0;
2699 	dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2700 	if (dst->i_inline & F2FS_EXTRA_ATTR) {
2701 		dst->i_extra_isize = src->i_extra_isize;
2702 
2703 		if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2704 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2705 							i_inline_xattr_size))
2706 			dst->i_inline_xattr_size = src->i_inline_xattr_size;
2707 
2708 		if (f2fs_sb_has_project_quota(sbi) &&
2709 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2710 								i_projid))
2711 			dst->i_projid = src->i_projid;
2712 
2713 		if (f2fs_sb_has_inode_crtime(sbi) &&
2714 			F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2715 							i_crtime_nsec)) {
2716 			dst->i_crtime = src->i_crtime;
2717 			dst->i_crtime_nsec = src->i_crtime_nsec;
2718 		}
2719 	}
2720 
2721 	new_ni = old_ni;
2722 	new_ni.ino = ino;
2723 
2724 	if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2725 		WARN_ON(1);
2726 	set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2727 	inc_valid_inode_count(sbi);
2728 	set_page_dirty(ipage);
2729 	f2fs_put_page(ipage, 1);
2730 	return 0;
2731 }
2732 
f2fs_restore_node_summary(struct f2fs_sb_info * sbi,unsigned int segno,struct f2fs_summary_block * sum)2733 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2734 			unsigned int segno, struct f2fs_summary_block *sum)
2735 {
2736 	struct f2fs_node *rn;
2737 	struct f2fs_summary *sum_entry;
2738 	block_t addr;
2739 	int i, idx, last_offset, nrpages;
2740 
2741 	/* scan the node segment */
2742 	last_offset = sbi->blocks_per_seg;
2743 	addr = START_BLOCK(sbi, segno);
2744 	sum_entry = &sum->entries[0];
2745 
2746 	for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2747 		nrpages = min(last_offset - i, BIO_MAX_PAGES);
2748 
2749 		/* readahead node pages */
2750 		f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2751 
2752 		for (idx = addr; idx < addr + nrpages; idx++) {
2753 			struct page *page = f2fs_get_tmp_page(sbi, idx);
2754 
2755 			if (IS_ERR(page))
2756 				return PTR_ERR(page);
2757 
2758 			rn = F2FS_NODE(page);
2759 			sum_entry->nid = rn->footer.nid;
2760 			sum_entry->version = 0;
2761 			sum_entry->ofs_in_node = 0;
2762 			sum_entry++;
2763 			f2fs_put_page(page, 1);
2764 		}
2765 
2766 		invalidate_mapping_pages(META_MAPPING(sbi), addr,
2767 							addr + nrpages);
2768 	}
2769 	return 0;
2770 }
2771 
remove_nats_in_journal(struct f2fs_sb_info * sbi)2772 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2773 {
2774 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2775 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2776 	struct f2fs_journal *journal = curseg->journal;
2777 	int i;
2778 
2779 	down_write(&curseg->journal_rwsem);
2780 	for (i = 0; i < nats_in_cursum(journal); i++) {
2781 		struct nat_entry *ne;
2782 		struct f2fs_nat_entry raw_ne;
2783 		nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2784 
2785 		if (f2fs_check_nid_range(sbi, nid))
2786 			continue;
2787 
2788 		raw_ne = nat_in_journal(journal, i);
2789 
2790 		ne = __lookup_nat_cache(nm_i, nid);
2791 		if (!ne) {
2792 			ne = __alloc_nat_entry(nid, true);
2793 			__init_nat_entry(nm_i, ne, &raw_ne, true);
2794 		}
2795 
2796 		/*
2797 		 * if a free nat in journal has not been used after last
2798 		 * checkpoint, we should remove it from available nids,
2799 		 * since later we will add it again.
2800 		 */
2801 		if (!get_nat_flag(ne, IS_DIRTY) &&
2802 				le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2803 			spin_lock(&nm_i->nid_list_lock);
2804 			nm_i->available_nids--;
2805 			spin_unlock(&nm_i->nid_list_lock);
2806 		}
2807 
2808 		__set_nat_cache_dirty(nm_i, ne);
2809 	}
2810 	update_nats_in_cursum(journal, -i);
2811 	up_write(&curseg->journal_rwsem);
2812 }
2813 
__adjust_nat_entry_set(struct nat_entry_set * nes,struct list_head * head,int max)2814 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2815 						struct list_head *head, int max)
2816 {
2817 	struct nat_entry_set *cur;
2818 
2819 	if (nes->entry_cnt >= max)
2820 		goto add_out;
2821 
2822 	list_for_each_entry(cur, head, set_list) {
2823 		if (cur->entry_cnt >= nes->entry_cnt) {
2824 			list_add(&nes->set_list, cur->set_list.prev);
2825 			return;
2826 		}
2827 	}
2828 add_out:
2829 	list_add_tail(&nes->set_list, head);
2830 }
2831 
__update_nat_bits(struct f2fs_sb_info * sbi,nid_t start_nid,struct page * page)2832 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2833 						struct page *page)
2834 {
2835 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2836 	unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2837 	struct f2fs_nat_block *nat_blk = page_address(page);
2838 	int valid = 0;
2839 	int i = 0;
2840 
2841 	if (!enabled_nat_bits(sbi, NULL))
2842 		return;
2843 
2844 	if (nat_index == 0) {
2845 		valid = 1;
2846 		i = 1;
2847 	}
2848 	for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2849 		if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2850 			valid++;
2851 	}
2852 	if (valid == 0) {
2853 		__set_bit_le(nat_index, nm_i->empty_nat_bits);
2854 		__clear_bit_le(nat_index, nm_i->full_nat_bits);
2855 		return;
2856 	}
2857 
2858 	__clear_bit_le(nat_index, nm_i->empty_nat_bits);
2859 	if (valid == NAT_ENTRY_PER_BLOCK)
2860 		__set_bit_le(nat_index, nm_i->full_nat_bits);
2861 	else
2862 		__clear_bit_le(nat_index, nm_i->full_nat_bits);
2863 }
2864 
__flush_nat_entry_set(struct f2fs_sb_info * sbi,struct nat_entry_set * set,struct cp_control * cpc)2865 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2866 		struct nat_entry_set *set, struct cp_control *cpc)
2867 {
2868 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2869 	struct f2fs_journal *journal = curseg->journal;
2870 	nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2871 	bool to_journal = true;
2872 	struct f2fs_nat_block *nat_blk;
2873 	struct nat_entry *ne, *cur;
2874 	struct page *page = NULL;
2875 
2876 	/*
2877 	 * there are two steps to flush nat entries:
2878 	 * #1, flush nat entries to journal in current hot data summary block.
2879 	 * #2, flush nat entries to nat page.
2880 	 */
2881 	if (enabled_nat_bits(sbi, cpc) ||
2882 		!__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2883 		to_journal = false;
2884 
2885 	if (to_journal) {
2886 		down_write(&curseg->journal_rwsem);
2887 	} else {
2888 		page = get_next_nat_page(sbi, start_nid);
2889 		if (IS_ERR(page))
2890 			return PTR_ERR(page);
2891 
2892 		nat_blk = page_address(page);
2893 		f2fs_bug_on(sbi, !nat_blk);
2894 	}
2895 
2896 	/* flush dirty nats in nat entry set */
2897 	list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2898 		struct f2fs_nat_entry *raw_ne;
2899 		nid_t nid = nat_get_nid(ne);
2900 		int offset;
2901 
2902 		f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2903 
2904 		if (to_journal) {
2905 			offset = f2fs_lookup_journal_in_cursum(journal,
2906 							NAT_JOURNAL, nid, 1);
2907 			f2fs_bug_on(sbi, offset < 0);
2908 			raw_ne = &nat_in_journal(journal, offset);
2909 			nid_in_journal(journal, offset) = cpu_to_le32(nid);
2910 		} else {
2911 			raw_ne = &nat_blk->entries[nid - start_nid];
2912 		}
2913 		raw_nat_from_node_info(raw_ne, &ne->ni);
2914 		nat_reset_flag(ne);
2915 		__clear_nat_cache_dirty(NM_I(sbi), set, ne);
2916 		if (nat_get_blkaddr(ne) == NULL_ADDR) {
2917 			add_free_nid(sbi, nid, false, true);
2918 		} else {
2919 			spin_lock(&NM_I(sbi)->nid_list_lock);
2920 			update_free_nid_bitmap(sbi, nid, false, false);
2921 			spin_unlock(&NM_I(sbi)->nid_list_lock);
2922 		}
2923 	}
2924 
2925 	if (to_journal) {
2926 		up_write(&curseg->journal_rwsem);
2927 	} else {
2928 		__update_nat_bits(sbi, start_nid, page);
2929 		f2fs_put_page(page, 1);
2930 	}
2931 
2932 	/* Allow dirty nats by node block allocation in write_begin */
2933 	if (!set->entry_cnt) {
2934 		radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2935 		kmem_cache_free(nat_entry_set_slab, set);
2936 	}
2937 	return 0;
2938 }
2939 
2940 /*
2941  * This function is called during the checkpointing process.
2942  */
f2fs_flush_nat_entries(struct f2fs_sb_info * sbi,struct cp_control * cpc)2943 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2944 {
2945 	struct f2fs_nm_info *nm_i = NM_I(sbi);
2946 	struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2947 	struct f2fs_journal *journal = curseg->journal;
2948 	struct nat_entry_set *setvec[SETVEC_SIZE];
2949 	struct nat_entry_set *set, *tmp;
2950 	unsigned int found;
2951 	nid_t set_idx = 0;
2952 	LIST_HEAD(sets);
2953 	int err = 0;
2954 
2955 	/*
2956 	 * during unmount, let's flush nat_bits before checking
2957 	 * nat_cnt[DIRTY_NAT].
2958 	 */
2959 	if (enabled_nat_bits(sbi, cpc)) {
2960 		down_write(&nm_i->nat_tree_lock);
2961 		remove_nats_in_journal(sbi);
2962 		up_write(&nm_i->nat_tree_lock);
2963 	}
2964 
2965 	if (!nm_i->nat_cnt[DIRTY_NAT])
2966 		return 0;
2967 
2968 	down_write(&nm_i->nat_tree_lock);
2969 
2970 	/*
2971 	 * if there are no enough space in journal to store dirty nat
2972 	 * entries, remove all entries from journal and merge them
2973 	 * into nat entry set.
2974 	 */
2975 	if (enabled_nat_bits(sbi, cpc) ||
2976 		!__has_cursum_space(journal,
2977 			nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
2978 		remove_nats_in_journal(sbi);
2979 
2980 	while ((found = __gang_lookup_nat_set(nm_i,
2981 					set_idx, SETVEC_SIZE, setvec))) {
2982 		unsigned idx;
2983 		set_idx = setvec[found - 1]->set + 1;
2984 		for (idx = 0; idx < found; idx++)
2985 			__adjust_nat_entry_set(setvec[idx], &sets,
2986 						MAX_NAT_JENTRIES(journal));
2987 	}
2988 
2989 	/* flush dirty nats in nat entry set */
2990 	list_for_each_entry_safe(set, tmp, &sets, set_list) {
2991 		err = __flush_nat_entry_set(sbi, set, cpc);
2992 		if (err)
2993 			break;
2994 	}
2995 
2996 	up_write(&nm_i->nat_tree_lock);
2997 	/* Allow dirty nats by node block allocation in write_begin */
2998 
2999 	return err;
3000 }
3001 
__get_nat_bitmaps(struct f2fs_sb_info * sbi)3002 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3003 {
3004 	struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3005 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3006 	unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3007 	unsigned int i;
3008 	__u64 cp_ver = cur_cp_version(ckpt);
3009 	block_t nat_bits_addr;
3010 
3011 	if (!enabled_nat_bits(sbi, NULL))
3012 		return 0;
3013 
3014 	nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3015 	nm_i->nat_bits = f2fs_kvzalloc(sbi,
3016 			nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3017 	if (!nm_i->nat_bits)
3018 		return -ENOMEM;
3019 
3020 	nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3021 						nm_i->nat_bits_blocks;
3022 	for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3023 		struct page *page;
3024 
3025 		page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3026 		if (IS_ERR(page))
3027 			return PTR_ERR(page);
3028 
3029 		memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3030 					page_address(page), F2FS_BLKSIZE);
3031 		f2fs_put_page(page, 1);
3032 	}
3033 
3034 	cp_ver |= (cur_cp_crc(ckpt) << 32);
3035 	if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3036 		disable_nat_bits(sbi, true);
3037 		return 0;
3038 	}
3039 
3040 	nm_i->full_nat_bits = nm_i->nat_bits + 8;
3041 	nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3042 
3043 	f2fs_notice(sbi, "Found nat_bits in checkpoint");
3044 	return 0;
3045 }
3046 
load_free_nid_bitmap(struct f2fs_sb_info * sbi)3047 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3048 {
3049 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3050 	unsigned int i = 0;
3051 	nid_t nid, last_nid;
3052 
3053 	if (!enabled_nat_bits(sbi, NULL))
3054 		return;
3055 
3056 	for (i = 0; i < nm_i->nat_blocks; i++) {
3057 		i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3058 		if (i >= nm_i->nat_blocks)
3059 			break;
3060 
3061 		__set_bit_le(i, nm_i->nat_block_bitmap);
3062 
3063 		nid = i * NAT_ENTRY_PER_BLOCK;
3064 		last_nid = nid + NAT_ENTRY_PER_BLOCK;
3065 
3066 		spin_lock(&NM_I(sbi)->nid_list_lock);
3067 		for (; nid < last_nid; nid++)
3068 			update_free_nid_bitmap(sbi, nid, true, true);
3069 		spin_unlock(&NM_I(sbi)->nid_list_lock);
3070 	}
3071 
3072 	for (i = 0; i < nm_i->nat_blocks; i++) {
3073 		i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3074 		if (i >= nm_i->nat_blocks)
3075 			break;
3076 
3077 		__set_bit_le(i, nm_i->nat_block_bitmap);
3078 	}
3079 }
3080 
init_node_manager(struct f2fs_sb_info * sbi)3081 static int init_node_manager(struct f2fs_sb_info *sbi)
3082 {
3083 	struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3084 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3085 	unsigned char *version_bitmap;
3086 	unsigned int nat_segs;
3087 	int err;
3088 
3089 	nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3090 
3091 	/* segment_count_nat includes pair segment so divide to 2. */
3092 	nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3093 	nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3094 	nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3095 
3096 	/* not used nids: 0, node, meta, (and root counted as valid node) */
3097 	nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3098 						F2FS_RESERVED_NODE_NUM;
3099 	nm_i->nid_cnt[FREE_NID] = 0;
3100 	nm_i->nid_cnt[PREALLOC_NID] = 0;
3101 	nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3102 	nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3103 	nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3104 
3105 	INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3106 	INIT_LIST_HEAD(&nm_i->free_nid_list);
3107 	INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3108 	INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3109 	INIT_LIST_HEAD(&nm_i->nat_entries);
3110 	spin_lock_init(&nm_i->nat_list_lock);
3111 
3112 	mutex_init(&nm_i->build_lock);
3113 	spin_lock_init(&nm_i->nid_list_lock);
3114 	init_rwsem(&nm_i->nat_tree_lock);
3115 
3116 	nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3117 	nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3118 	version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3119 	nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3120 					GFP_KERNEL);
3121 	if (!nm_i->nat_bitmap)
3122 		return -ENOMEM;
3123 
3124 	err = __get_nat_bitmaps(sbi);
3125 	if (err)
3126 		return err;
3127 
3128 #ifdef CONFIG_F2FS_CHECK_FS
3129 	nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3130 					GFP_KERNEL);
3131 	if (!nm_i->nat_bitmap_mir)
3132 		return -ENOMEM;
3133 #endif
3134 
3135 	return 0;
3136 }
3137 
init_free_nid_cache(struct f2fs_sb_info * sbi)3138 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3139 {
3140 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3141 	int i;
3142 
3143 	nm_i->free_nid_bitmap =
3144 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3145 					      nm_i->nat_blocks),
3146 			      GFP_KERNEL);
3147 	if (!nm_i->free_nid_bitmap)
3148 		return -ENOMEM;
3149 
3150 	for (i = 0; i < nm_i->nat_blocks; i++) {
3151 		nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3152 			f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3153 		if (!nm_i->free_nid_bitmap[i])
3154 			return -ENOMEM;
3155 	}
3156 
3157 	nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3158 								GFP_KERNEL);
3159 	if (!nm_i->nat_block_bitmap)
3160 		return -ENOMEM;
3161 
3162 	nm_i->free_nid_count =
3163 		f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3164 					      nm_i->nat_blocks),
3165 			      GFP_KERNEL);
3166 	if (!nm_i->free_nid_count)
3167 		return -ENOMEM;
3168 	return 0;
3169 }
3170 
f2fs_build_node_manager(struct f2fs_sb_info * sbi)3171 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3172 {
3173 	int err;
3174 
3175 	sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3176 							GFP_KERNEL);
3177 	if (!sbi->nm_info)
3178 		return -ENOMEM;
3179 
3180 	err = init_node_manager(sbi);
3181 	if (err)
3182 		return err;
3183 
3184 	err = init_free_nid_cache(sbi);
3185 	if (err)
3186 		return err;
3187 
3188 	/* load free nid status from nat_bits table */
3189 	load_free_nid_bitmap(sbi);
3190 
3191 	return f2fs_build_free_nids(sbi, true, true);
3192 }
3193 
f2fs_destroy_node_manager(struct f2fs_sb_info * sbi)3194 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3195 {
3196 	struct f2fs_nm_info *nm_i = NM_I(sbi);
3197 	struct free_nid *i, *next_i;
3198 	struct nat_entry *natvec[NATVEC_SIZE];
3199 	struct nat_entry_set *setvec[SETVEC_SIZE];
3200 	nid_t nid = 0;
3201 	unsigned int found;
3202 
3203 	if (!nm_i)
3204 		return;
3205 
3206 	/* destroy free nid list */
3207 	spin_lock(&nm_i->nid_list_lock);
3208 	list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3209 		__remove_free_nid(sbi, i, FREE_NID);
3210 		spin_unlock(&nm_i->nid_list_lock);
3211 		kmem_cache_free(free_nid_slab, i);
3212 		spin_lock(&nm_i->nid_list_lock);
3213 	}
3214 	f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3215 	f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3216 	f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3217 	spin_unlock(&nm_i->nid_list_lock);
3218 
3219 	/* destroy nat cache */
3220 	down_write(&nm_i->nat_tree_lock);
3221 	while ((found = __gang_lookup_nat_cache(nm_i,
3222 					nid, NATVEC_SIZE, natvec))) {
3223 		unsigned idx;
3224 
3225 		nid = nat_get_nid(natvec[found - 1]) + 1;
3226 		for (idx = 0; idx < found; idx++) {
3227 			spin_lock(&nm_i->nat_list_lock);
3228 			list_del(&natvec[idx]->list);
3229 			spin_unlock(&nm_i->nat_list_lock);
3230 
3231 			__del_from_nat_cache(nm_i, natvec[idx]);
3232 		}
3233 	}
3234 	f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3235 
3236 	/* destroy nat set cache */
3237 	nid = 0;
3238 	while ((found = __gang_lookup_nat_set(nm_i,
3239 					nid, SETVEC_SIZE, setvec))) {
3240 		unsigned idx;
3241 
3242 		nid = setvec[found - 1]->set + 1;
3243 		for (idx = 0; idx < found; idx++) {
3244 			/* entry_cnt is not zero, when cp_error was occurred */
3245 			f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3246 			radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3247 			kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3248 		}
3249 	}
3250 	up_write(&nm_i->nat_tree_lock);
3251 
3252 	kvfree(nm_i->nat_block_bitmap);
3253 	if (nm_i->free_nid_bitmap) {
3254 		int i;
3255 
3256 		for (i = 0; i < nm_i->nat_blocks; i++)
3257 			kvfree(nm_i->free_nid_bitmap[i]);
3258 		kvfree(nm_i->free_nid_bitmap);
3259 	}
3260 	kvfree(nm_i->free_nid_count);
3261 
3262 	kvfree(nm_i->nat_bitmap);
3263 	kvfree(nm_i->nat_bits);
3264 #ifdef CONFIG_F2FS_CHECK_FS
3265 	kvfree(nm_i->nat_bitmap_mir);
3266 #endif
3267 	sbi->nm_info = NULL;
3268 	kfree(nm_i);
3269 }
3270 
f2fs_create_node_manager_caches(void)3271 int __init f2fs_create_node_manager_caches(void)
3272 {
3273 	nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3274 			sizeof(struct nat_entry));
3275 	if (!nat_entry_slab)
3276 		goto fail;
3277 
3278 	free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3279 			sizeof(struct free_nid));
3280 	if (!free_nid_slab)
3281 		goto destroy_nat_entry;
3282 
3283 	nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3284 			sizeof(struct nat_entry_set));
3285 	if (!nat_entry_set_slab)
3286 		goto destroy_free_nid;
3287 
3288 	fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3289 			sizeof(struct fsync_node_entry));
3290 	if (!fsync_node_entry_slab)
3291 		goto destroy_nat_entry_set;
3292 	return 0;
3293 
3294 destroy_nat_entry_set:
3295 	kmem_cache_destroy(nat_entry_set_slab);
3296 destroy_free_nid:
3297 	kmem_cache_destroy(free_nid_slab);
3298 destroy_nat_entry:
3299 	kmem_cache_destroy(nat_entry_slab);
3300 fail:
3301 	return -ENOMEM;
3302 }
3303 
f2fs_destroy_node_manager_caches(void)3304 void f2fs_destroy_node_manager_caches(void)
3305 {
3306 	kmem_cache_destroy(fsync_node_entry_slab);
3307 	kmem_cache_destroy(nat_entry_set_slab);
3308 	kmem_cache_destroy(free_nid_slab);
3309 	kmem_cache_destroy(nat_entry_slab);
3310 }
3311