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/events/f2fs.h>
21
22 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
23
24 static struct kmem_cache *nat_entry_slab;
25 static struct kmem_cache *free_nid_slab;
26 static struct kmem_cache *nat_entry_set_slab;
27 static struct kmem_cache *fsync_node_entry_slab;
28
29 /*
30 * Check whether the given nid is within node id range.
31 */
f2fs_check_nid_range(struct f2fs_sb_info * sbi,nid_t nid)32 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
33 {
34 if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
35 set_sbi_flag(sbi, SBI_NEED_FSCK);
36 f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
37 __func__, nid);
38 return -EFSCORRUPTED;
39 }
40 return 0;
41 }
42
f2fs_available_free_memory(struct f2fs_sb_info * sbi,int type)43 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
44 {
45 struct f2fs_nm_info *nm_i = NM_I(sbi);
46 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
47 struct sysinfo val;
48 unsigned long avail_ram;
49 unsigned long mem_size = 0;
50 bool res = false;
51
52 if (!nm_i)
53 return true;
54
55 si_meminfo(&val);
56
57 /* only uses low memory */
58 avail_ram = val.totalram - val.totalhigh;
59
60 /*
61 * give 25%, 25%, 50%, 50%, 25%, 25% memory for each components respectively
62 */
63 if (type == FREE_NIDS) {
64 mem_size = (nm_i->nid_cnt[FREE_NID] *
65 sizeof(struct free_nid)) >> PAGE_SHIFT;
66 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
67 } else if (type == NAT_ENTRIES) {
68 mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
69 sizeof(struct nat_entry)) >> PAGE_SHIFT;
70 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
71 if (excess_cached_nats(sbi))
72 res = false;
73 } else if (type == DIRTY_DENTS) {
74 if (sbi->sb->s_bdi->wb.dirty_exceeded)
75 return false;
76 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
77 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
78 } else if (type == INO_ENTRIES) {
79 int i;
80
81 for (i = 0; i < MAX_INO_ENTRY; i++)
82 mem_size += sbi->im[i].ino_num *
83 sizeof(struct ino_entry);
84 mem_size >>= PAGE_SHIFT;
85 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
86 } else if (type == READ_EXTENT_CACHE || type == AGE_EXTENT_CACHE) {
87 enum extent_type etype = type == READ_EXTENT_CACHE ?
88 EX_READ : EX_BLOCK_AGE;
89 struct extent_tree_info *eti = &sbi->extent_tree[etype];
90
91 mem_size = (atomic_read(&eti->total_ext_tree) *
92 sizeof(struct extent_tree) +
93 atomic_read(&eti->total_ext_node) *
94 sizeof(struct extent_node)) >> PAGE_SHIFT;
95 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
96 } else if (type == INMEM_PAGES) {
97 /* it allows 20% / total_ram for inmemory pages */
98 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
99 res = mem_size < (val.totalram / 5);
100 } else if (type == DISCARD_CACHE) {
101 mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
102 sizeof(struct discard_cmd)) >> PAGE_SHIFT;
103 res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
104 } else if (type == COMPRESS_PAGE) {
105 #ifdef CONFIG_F2FS_FS_COMPRESSION
106 unsigned long free_ram = val.freeram;
107
108 /*
109 * free memory is lower than watermark or cached page count
110 * exceed threshold, deny caching compress page.
111 */
112 res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
113 (COMPRESS_MAPPING(sbi)->nrpages <
114 free_ram * sbi->compress_percent / 100);
115 #else
116 res = false;
117 #endif
118 } else {
119 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
120 return true;
121 }
122 return res;
123 }
124
clear_node_page_dirty(struct page * page)125 static void clear_node_page_dirty(struct page *page)
126 {
127 if (PageDirty(page)) {
128 f2fs_clear_page_cache_dirty_tag(page);
129 clear_page_dirty_for_io(page);
130 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
131 }
132 ClearPageUptodate(page);
133 }
134
get_current_nat_page(struct f2fs_sb_info * sbi,nid_t nid)135 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
136 {
137 return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
138 }
139
get_next_nat_page(struct f2fs_sb_info * sbi,nid_t nid)140 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
141 {
142 struct page *src_page;
143 struct page *dst_page;
144 pgoff_t dst_off;
145 void *src_addr;
146 void *dst_addr;
147 struct f2fs_nm_info *nm_i = NM_I(sbi);
148
149 dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
150
151 /* get current nat block page with lock */
152 src_page = get_current_nat_page(sbi, nid);
153 if (IS_ERR(src_page))
154 return src_page;
155 dst_page = f2fs_grab_meta_page(sbi, dst_off);
156 f2fs_bug_on(sbi, PageDirty(src_page));
157
158 src_addr = page_address(src_page);
159 dst_addr = page_address(dst_page);
160 memcpy(dst_addr, src_addr, PAGE_SIZE);
161 set_page_dirty(dst_page);
162 f2fs_put_page(src_page, 1);
163
164 set_to_next_nat(nm_i, nid);
165
166 return dst_page;
167 }
168
__alloc_nat_entry(nid_t nid,bool no_fail)169 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
170 {
171 struct nat_entry *new;
172
173 if (no_fail)
174 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
175 else
176 new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
177 if (new) {
178 nat_set_nid(new, nid);
179 nat_reset_flag(new);
180 }
181 return new;
182 }
183
__free_nat_entry(struct nat_entry * e)184 static void __free_nat_entry(struct nat_entry *e)
185 {
186 kmem_cache_free(nat_entry_slab, e);
187 }
188
189 /* 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)190 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
191 struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
192 {
193 if (no_fail)
194 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
195 else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
196 return NULL;
197
198 if (raw_ne)
199 node_info_from_raw_nat(&ne->ni, raw_ne);
200
201 spin_lock(&nm_i->nat_list_lock);
202 list_add_tail(&ne->list, &nm_i->nat_entries);
203 spin_unlock(&nm_i->nat_list_lock);
204
205 nm_i->nat_cnt[TOTAL_NAT]++;
206 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
207 return ne;
208 }
209
__lookup_nat_cache(struct f2fs_nm_info * nm_i,nid_t n)210 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
211 {
212 struct nat_entry *ne;
213
214 ne = radix_tree_lookup(&nm_i->nat_root, n);
215
216 /* for recent accessed nat entry, move it to tail of lru list */
217 if (ne && !get_nat_flag(ne, IS_DIRTY)) {
218 spin_lock(&nm_i->nat_list_lock);
219 if (!list_empty(&ne->list))
220 list_move_tail(&ne->list, &nm_i->nat_entries);
221 spin_unlock(&nm_i->nat_list_lock);
222 }
223
224 return ne;
225 }
226
__gang_lookup_nat_cache(struct f2fs_nm_info * nm_i,nid_t start,unsigned int nr,struct nat_entry ** ep)227 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
228 nid_t start, unsigned int nr, struct nat_entry **ep)
229 {
230 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
231 }
232
__del_from_nat_cache(struct f2fs_nm_info * nm_i,struct nat_entry * e)233 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
234 {
235 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
236 nm_i->nat_cnt[TOTAL_NAT]--;
237 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
238 __free_nat_entry(e);
239 }
240
__grab_nat_entry_set(struct f2fs_nm_info * nm_i,struct nat_entry * ne)241 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
242 struct nat_entry *ne)
243 {
244 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
245 struct nat_entry_set *head;
246
247 head = radix_tree_lookup(&nm_i->nat_set_root, set);
248 if (!head) {
249 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
250
251 INIT_LIST_HEAD(&head->entry_list);
252 INIT_LIST_HEAD(&head->set_list);
253 head->set = set;
254 head->entry_cnt = 0;
255 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
256 }
257 return head;
258 }
259
__set_nat_cache_dirty(struct f2fs_nm_info * nm_i,struct nat_entry * ne)260 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
261 struct nat_entry *ne)
262 {
263 struct nat_entry_set *head;
264 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
265
266 if (!new_ne)
267 head = __grab_nat_entry_set(nm_i, ne);
268
269 /*
270 * update entry_cnt in below condition:
271 * 1. update NEW_ADDR to valid block address;
272 * 2. update old block address to new one;
273 */
274 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
275 !get_nat_flag(ne, IS_DIRTY)))
276 head->entry_cnt++;
277
278 set_nat_flag(ne, IS_PREALLOC, new_ne);
279
280 if (get_nat_flag(ne, IS_DIRTY))
281 goto refresh_list;
282
283 nm_i->nat_cnt[DIRTY_NAT]++;
284 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
285 set_nat_flag(ne, IS_DIRTY, true);
286 refresh_list:
287 spin_lock(&nm_i->nat_list_lock);
288 if (new_ne)
289 list_del_init(&ne->list);
290 else
291 list_move_tail(&ne->list, &head->entry_list);
292 spin_unlock(&nm_i->nat_list_lock);
293 }
294
__clear_nat_cache_dirty(struct f2fs_nm_info * nm_i,struct nat_entry_set * set,struct nat_entry * ne)295 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
296 struct nat_entry_set *set, struct nat_entry *ne)
297 {
298 spin_lock(&nm_i->nat_list_lock);
299 list_move_tail(&ne->list, &nm_i->nat_entries);
300 spin_unlock(&nm_i->nat_list_lock);
301
302 set_nat_flag(ne, IS_DIRTY, false);
303 set->entry_cnt--;
304 nm_i->nat_cnt[DIRTY_NAT]--;
305 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
306 }
307
__gang_lookup_nat_set(struct f2fs_nm_info * nm_i,nid_t start,unsigned int nr,struct nat_entry_set ** ep)308 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
309 nid_t start, unsigned int nr, struct nat_entry_set **ep)
310 {
311 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
312 start, nr);
313 }
314
f2fs_in_warm_node_list(struct f2fs_sb_info * sbi,struct page * page)315 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
316 {
317 return NODE_MAPPING(sbi) == page->mapping &&
318 IS_DNODE(page) && is_cold_node(page);
319 }
320
f2fs_init_fsync_node_info(struct f2fs_sb_info * sbi)321 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
322 {
323 spin_lock_init(&sbi->fsync_node_lock);
324 INIT_LIST_HEAD(&sbi->fsync_node_list);
325 sbi->fsync_seg_id = 0;
326 sbi->fsync_node_num = 0;
327 }
328
f2fs_add_fsync_node_entry(struct f2fs_sb_info * sbi,struct page * page)329 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
330 struct page *page)
331 {
332 struct fsync_node_entry *fn;
333 unsigned long flags;
334 unsigned int seq_id;
335
336 fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS);
337
338 get_page(page);
339 fn->page = page;
340 INIT_LIST_HEAD(&fn->list);
341
342 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
343 list_add_tail(&fn->list, &sbi->fsync_node_list);
344 fn->seq_id = sbi->fsync_seg_id++;
345 seq_id = fn->seq_id;
346 sbi->fsync_node_num++;
347 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
348
349 return seq_id;
350 }
351
f2fs_del_fsync_node_entry(struct f2fs_sb_info * sbi,struct page * page)352 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
353 {
354 struct fsync_node_entry *fn;
355 unsigned long flags;
356
357 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
358 list_for_each_entry(fn, &sbi->fsync_node_list, list) {
359 if (fn->page == page) {
360 list_del(&fn->list);
361 sbi->fsync_node_num--;
362 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
363 kmem_cache_free(fsync_node_entry_slab, fn);
364 put_page(page);
365 return;
366 }
367 }
368 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
369 f2fs_bug_on(sbi, 1);
370 }
371
f2fs_reset_fsync_node_info(struct f2fs_sb_info * sbi)372 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
373 {
374 unsigned long flags;
375
376 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
377 sbi->fsync_seg_id = 0;
378 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
379 }
380
f2fs_need_dentry_mark(struct f2fs_sb_info * sbi,nid_t nid)381 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
382 {
383 struct f2fs_nm_info *nm_i = NM_I(sbi);
384 struct nat_entry *e;
385 bool need = false;
386
387 f2fs_down_read(&nm_i->nat_tree_lock);
388 e = __lookup_nat_cache(nm_i, nid);
389 if (e) {
390 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
391 !get_nat_flag(e, HAS_FSYNCED_INODE))
392 need = true;
393 }
394 f2fs_up_read(&nm_i->nat_tree_lock);
395 return need;
396 }
397
f2fs_is_checkpointed_node(struct f2fs_sb_info * sbi,nid_t nid)398 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
399 {
400 struct f2fs_nm_info *nm_i = NM_I(sbi);
401 struct nat_entry *e;
402 bool is_cp = true;
403
404 f2fs_down_read(&nm_i->nat_tree_lock);
405 e = __lookup_nat_cache(nm_i, nid);
406 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
407 is_cp = false;
408 f2fs_up_read(&nm_i->nat_tree_lock);
409 return is_cp;
410 }
411
f2fs_need_inode_block_update(struct f2fs_sb_info * sbi,nid_t ino)412 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
413 {
414 struct f2fs_nm_info *nm_i = NM_I(sbi);
415 struct nat_entry *e;
416 bool need_update = true;
417
418 f2fs_down_read(&nm_i->nat_tree_lock);
419 e = __lookup_nat_cache(nm_i, ino);
420 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
421 (get_nat_flag(e, IS_CHECKPOINTED) ||
422 get_nat_flag(e, HAS_FSYNCED_INODE)))
423 need_update = false;
424 f2fs_up_read(&nm_i->nat_tree_lock);
425 return need_update;
426 }
427
428 /* must be locked by nat_tree_lock */
cache_nat_entry(struct f2fs_sb_info * sbi,nid_t nid,struct f2fs_nat_entry * ne)429 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
430 struct f2fs_nat_entry *ne)
431 {
432 struct f2fs_nm_info *nm_i = NM_I(sbi);
433 struct nat_entry *new, *e;
434
435 /* Let's mitigate lock contention of nat_tree_lock during checkpoint */
436 if (f2fs_rwsem_is_locked(&sbi->cp_global_sem))
437 return;
438
439 new = __alloc_nat_entry(nid, false);
440 if (!new)
441 return;
442
443 f2fs_down_write(&nm_i->nat_tree_lock);
444 e = __lookup_nat_cache(nm_i, nid);
445 if (!e)
446 e = __init_nat_entry(nm_i, new, ne, false);
447 else
448 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
449 nat_get_blkaddr(e) !=
450 le32_to_cpu(ne->block_addr) ||
451 nat_get_version(e) != ne->version);
452 f2fs_up_write(&nm_i->nat_tree_lock);
453 if (e != new)
454 __free_nat_entry(new);
455 }
456
set_node_addr(struct f2fs_sb_info * sbi,struct node_info * ni,block_t new_blkaddr,bool fsync_done)457 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
458 block_t new_blkaddr, bool fsync_done)
459 {
460 struct f2fs_nm_info *nm_i = NM_I(sbi);
461 struct nat_entry *e;
462 struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
463
464 f2fs_down_write(&nm_i->nat_tree_lock);
465 e = __lookup_nat_cache(nm_i, ni->nid);
466 if (!e) {
467 e = __init_nat_entry(nm_i, new, NULL, true);
468 copy_node_info(&e->ni, ni);
469 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
470 } else if (new_blkaddr == NEW_ADDR) {
471 /*
472 * when nid is reallocated,
473 * previous nat entry can be remained in nat cache.
474 * So, reinitialize it with new information.
475 */
476 copy_node_info(&e->ni, ni);
477 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
478 }
479 /* let's free early to reduce memory consumption */
480 if (e != new)
481 __free_nat_entry(new);
482
483 /* sanity check */
484 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
485 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
486 new_blkaddr == NULL_ADDR);
487 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
488 new_blkaddr == NEW_ADDR);
489 f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
490 new_blkaddr == NEW_ADDR);
491
492 /* increment version no as node is removed */
493 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
494 unsigned char version = nat_get_version(e);
495
496 nat_set_version(e, inc_node_version(version));
497 }
498
499 /* change address */
500 nat_set_blkaddr(e, new_blkaddr);
501 if (!__is_valid_data_blkaddr(new_blkaddr))
502 set_nat_flag(e, IS_CHECKPOINTED, false);
503 __set_nat_cache_dirty(nm_i, e);
504
505 /* update fsync_mark if its inode nat entry is still alive */
506 if (ni->nid != ni->ino)
507 e = __lookup_nat_cache(nm_i, ni->ino);
508 if (e) {
509 if (fsync_done && ni->nid == ni->ino)
510 set_nat_flag(e, HAS_FSYNCED_INODE, true);
511 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
512 }
513 f2fs_up_write(&nm_i->nat_tree_lock);
514 }
515
f2fs_try_to_free_nats(struct f2fs_sb_info * sbi,int nr_shrink)516 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
517 {
518 struct f2fs_nm_info *nm_i = NM_I(sbi);
519 int nr = nr_shrink;
520
521 if (!f2fs_down_write_trylock(&nm_i->nat_tree_lock))
522 return 0;
523
524 spin_lock(&nm_i->nat_list_lock);
525 while (nr_shrink) {
526 struct nat_entry *ne;
527
528 if (list_empty(&nm_i->nat_entries))
529 break;
530
531 ne = list_first_entry(&nm_i->nat_entries,
532 struct nat_entry, list);
533 list_del(&ne->list);
534 spin_unlock(&nm_i->nat_list_lock);
535
536 __del_from_nat_cache(nm_i, ne);
537 nr_shrink--;
538
539 spin_lock(&nm_i->nat_list_lock);
540 }
541 spin_unlock(&nm_i->nat_list_lock);
542
543 f2fs_up_write(&nm_i->nat_tree_lock);
544 return nr - nr_shrink;
545 }
546
f2fs_get_node_info(struct f2fs_sb_info * sbi,nid_t nid,struct node_info * ni,bool checkpoint_context)547 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
548 struct node_info *ni, bool checkpoint_context)
549 {
550 struct f2fs_nm_info *nm_i = NM_I(sbi);
551 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
552 struct f2fs_journal *journal = curseg->journal;
553 nid_t start_nid = START_NID(nid);
554 struct f2fs_nat_block *nat_blk;
555 struct page *page = NULL;
556 struct f2fs_nat_entry ne;
557 struct nat_entry *e;
558 pgoff_t index;
559 block_t blkaddr;
560 int i;
561
562 ni->nid = nid;
563 retry:
564 /* Check nat cache */
565 f2fs_down_read(&nm_i->nat_tree_lock);
566 e = __lookup_nat_cache(nm_i, nid);
567 if (e) {
568 ni->ino = nat_get_ino(e);
569 ni->blk_addr = nat_get_blkaddr(e);
570 ni->version = nat_get_version(e);
571 f2fs_up_read(&nm_i->nat_tree_lock);
572 return 0;
573 }
574
575 /*
576 * Check current segment summary by trying to grab journal_rwsem first.
577 * This sem is on the critical path on the checkpoint requiring the above
578 * nat_tree_lock. Therefore, we should retry, if we failed to grab here
579 * while not bothering checkpoint.
580 */
581 if (!f2fs_rwsem_is_locked(&sbi->cp_global_sem) || checkpoint_context) {
582 down_read(&curseg->journal_rwsem);
583 } else if (f2fs_rwsem_is_contended(&nm_i->nat_tree_lock) ||
584 !down_read_trylock(&curseg->journal_rwsem)) {
585 f2fs_up_read(&nm_i->nat_tree_lock);
586 goto retry;
587 }
588
589 i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
590 if (i >= 0) {
591 ne = nat_in_journal(journal, i);
592 node_info_from_raw_nat(ni, &ne);
593 }
594 up_read(&curseg->journal_rwsem);
595 if (i >= 0) {
596 f2fs_up_read(&nm_i->nat_tree_lock);
597 goto cache;
598 }
599
600 /* Fill node_info from nat page */
601 index = current_nat_addr(sbi, nid);
602 f2fs_up_read(&nm_i->nat_tree_lock);
603
604 page = f2fs_get_meta_page(sbi, index);
605 if (IS_ERR(page))
606 return PTR_ERR(page);
607
608 nat_blk = (struct f2fs_nat_block *)page_address(page);
609 ne = nat_blk->entries[nid - start_nid];
610 node_info_from_raw_nat(ni, &ne);
611 f2fs_put_page(page, 1);
612 cache:
613 blkaddr = le32_to_cpu(ne.block_addr);
614 if (__is_valid_data_blkaddr(blkaddr) &&
615 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
616 return -EFAULT;
617
618 /* cache nat entry */
619 cache_nat_entry(sbi, nid, &ne);
620 return 0;
621 }
622
623 /*
624 * readahead MAX_RA_NODE number of node pages.
625 */
f2fs_ra_node_pages(struct page * parent,int start,int n)626 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
627 {
628 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
629 struct blk_plug plug;
630 int i, end;
631 nid_t nid;
632
633 blk_start_plug(&plug);
634
635 /* Then, try readahead for siblings of the desired node */
636 end = start + n;
637 end = min(end, NIDS_PER_BLOCK);
638 for (i = start; i < end; i++) {
639 nid = get_nid(parent, i, false);
640 f2fs_ra_node_page(sbi, nid);
641 }
642
643 blk_finish_plug(&plug);
644 }
645
f2fs_get_next_page_offset(struct dnode_of_data * dn,pgoff_t pgofs)646 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
647 {
648 const long direct_index = ADDRS_PER_INODE(dn->inode);
649 const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
650 const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
651 unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
652 int cur_level = dn->cur_level;
653 int max_level = dn->max_level;
654 pgoff_t base = 0;
655
656 if (!dn->max_level)
657 return pgofs + 1;
658
659 while (max_level-- > cur_level)
660 skipped_unit *= NIDS_PER_BLOCK;
661
662 switch (dn->max_level) {
663 case 3:
664 base += 2 * indirect_blks;
665 fallthrough;
666 case 2:
667 base += 2 * direct_blks;
668 fallthrough;
669 case 1:
670 base += direct_index;
671 break;
672 default:
673 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
674 }
675
676 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
677 }
678
679 /*
680 * The maximum depth is four.
681 * Offset[0] will have raw inode offset.
682 */
get_node_path(struct inode * inode,long block,int offset[4],unsigned int noffset[4])683 static int get_node_path(struct inode *inode, long block,
684 int offset[4], unsigned int noffset[4])
685 {
686 const long direct_index = ADDRS_PER_INODE(inode);
687 const long direct_blks = ADDRS_PER_BLOCK(inode);
688 const long dptrs_per_blk = NIDS_PER_BLOCK;
689 const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
690 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
691 int n = 0;
692 int level = 0;
693
694 noffset[0] = 0;
695
696 if (block < direct_index) {
697 offset[n] = block;
698 goto got;
699 }
700 block -= direct_index;
701 if (block < direct_blks) {
702 offset[n++] = NODE_DIR1_BLOCK;
703 noffset[n] = 1;
704 offset[n] = block;
705 level = 1;
706 goto got;
707 }
708 block -= direct_blks;
709 if (block < direct_blks) {
710 offset[n++] = NODE_DIR2_BLOCK;
711 noffset[n] = 2;
712 offset[n] = block;
713 level = 1;
714 goto got;
715 }
716 block -= direct_blks;
717 if (block < indirect_blks) {
718 offset[n++] = NODE_IND1_BLOCK;
719 noffset[n] = 3;
720 offset[n++] = block / direct_blks;
721 noffset[n] = 4 + offset[n - 1];
722 offset[n] = block % direct_blks;
723 level = 2;
724 goto got;
725 }
726 block -= indirect_blks;
727 if (block < indirect_blks) {
728 offset[n++] = NODE_IND2_BLOCK;
729 noffset[n] = 4 + dptrs_per_blk;
730 offset[n++] = block / direct_blks;
731 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
732 offset[n] = block % direct_blks;
733 level = 2;
734 goto got;
735 }
736 block -= indirect_blks;
737 if (block < dindirect_blks) {
738 offset[n++] = NODE_DIND_BLOCK;
739 noffset[n] = 5 + (dptrs_per_blk * 2);
740 offset[n++] = block / indirect_blks;
741 noffset[n] = 6 + (dptrs_per_blk * 2) +
742 offset[n - 1] * (dptrs_per_blk + 1);
743 offset[n++] = (block / direct_blks) % dptrs_per_blk;
744 noffset[n] = 7 + (dptrs_per_blk * 2) +
745 offset[n - 2] * (dptrs_per_blk + 1) +
746 offset[n - 1];
747 offset[n] = block % direct_blks;
748 level = 3;
749 goto got;
750 } else {
751 return -E2BIG;
752 }
753 got:
754 return level;
755 }
756
757 /*
758 * Caller should call f2fs_put_dnode(dn).
759 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
760 * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
761 */
f2fs_get_dnode_of_data(struct dnode_of_data * dn,pgoff_t index,int mode)762 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
763 {
764 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
765 struct page *npage[4];
766 struct page *parent = NULL;
767 int offset[4];
768 unsigned int noffset[4];
769 nid_t nids[4];
770 int level, i = 0;
771 int err = 0;
772
773 level = get_node_path(dn->inode, index, offset, noffset);
774 if (level < 0)
775 return level;
776
777 nids[0] = dn->inode->i_ino;
778 npage[0] = dn->inode_page;
779
780 if (!npage[0]) {
781 npage[0] = f2fs_get_node_page(sbi, nids[0]);
782 if (IS_ERR(npage[0]))
783 return PTR_ERR(npage[0]);
784 }
785
786 /* if inline_data is set, should not report any block indices */
787 if (f2fs_has_inline_data(dn->inode) && index) {
788 err = -ENOENT;
789 f2fs_put_page(npage[0], 1);
790 goto release_out;
791 }
792
793 parent = npage[0];
794 if (level != 0)
795 nids[1] = get_nid(parent, offset[0], true);
796 dn->inode_page = npage[0];
797 dn->inode_page_locked = true;
798
799 /* get indirect or direct nodes */
800 for (i = 1; i <= level; i++) {
801 bool done = false;
802
803 if (!nids[i] && mode == ALLOC_NODE) {
804 /* alloc new node */
805 if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
806 err = -ENOSPC;
807 goto release_pages;
808 }
809
810 dn->nid = nids[i];
811 npage[i] = f2fs_new_node_page(dn, noffset[i]);
812 if (IS_ERR(npage[i])) {
813 f2fs_alloc_nid_failed(sbi, nids[i]);
814 err = PTR_ERR(npage[i]);
815 goto release_pages;
816 }
817
818 set_nid(parent, offset[i - 1], nids[i], i == 1);
819 f2fs_alloc_nid_done(sbi, nids[i]);
820 done = true;
821 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
822 npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
823 if (IS_ERR(npage[i])) {
824 err = PTR_ERR(npage[i]);
825 goto release_pages;
826 }
827 done = true;
828 }
829 if (i == 1) {
830 dn->inode_page_locked = false;
831 unlock_page(parent);
832 } else {
833 f2fs_put_page(parent, 1);
834 }
835
836 if (!done) {
837 npage[i] = f2fs_get_node_page(sbi, nids[i]);
838 if (IS_ERR(npage[i])) {
839 err = PTR_ERR(npage[i]);
840 f2fs_put_page(npage[0], 0);
841 goto release_out;
842 }
843 }
844 if (i < level) {
845 parent = npage[i];
846 nids[i + 1] = get_nid(parent, offset[i], false);
847 }
848 }
849 dn->nid = nids[level];
850 dn->ofs_in_node = offset[level];
851 dn->node_page = npage[level];
852 dn->data_blkaddr = f2fs_data_blkaddr(dn);
853
854 if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) &&
855 f2fs_sb_has_readonly(sbi)) {
856 unsigned int c_len = f2fs_cluster_blocks_are_contiguous(dn);
857 block_t blkaddr;
858
859 if (!c_len)
860 goto out;
861
862 blkaddr = f2fs_data_blkaddr(dn);
863 if (blkaddr == COMPRESS_ADDR)
864 blkaddr = data_blkaddr(dn->inode, dn->node_page,
865 dn->ofs_in_node + 1);
866
867 f2fs_update_read_extent_tree_range_compressed(dn->inode,
868 index, blkaddr,
869 F2FS_I(dn->inode)->i_cluster_size,
870 c_len);
871 }
872 out:
873 return 0;
874
875 release_pages:
876 f2fs_put_page(parent, 1);
877 if (i > 1)
878 f2fs_put_page(npage[0], 0);
879 release_out:
880 dn->inode_page = NULL;
881 dn->node_page = NULL;
882 if (err == -ENOENT) {
883 dn->cur_level = i;
884 dn->max_level = level;
885 dn->ofs_in_node = offset[level];
886 }
887 return err;
888 }
889
truncate_node(struct dnode_of_data * dn)890 static int truncate_node(struct dnode_of_data *dn)
891 {
892 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
893 struct node_info ni;
894 int err;
895 pgoff_t index;
896
897 err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
898 if (err)
899 return err;
900
901 /* Deallocate node address */
902 f2fs_invalidate_blocks(sbi, ni.blk_addr);
903 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
904 set_node_addr(sbi, &ni, NULL_ADDR, false);
905
906 if (dn->nid == dn->inode->i_ino) {
907 f2fs_remove_orphan_inode(sbi, dn->nid);
908 dec_valid_inode_count(sbi);
909 f2fs_inode_synced(dn->inode);
910 }
911
912 clear_node_page_dirty(dn->node_page);
913 set_sbi_flag(sbi, SBI_IS_DIRTY);
914
915 index = dn->node_page->index;
916 f2fs_put_page(dn->node_page, 1);
917
918 invalidate_mapping_pages(NODE_MAPPING(sbi),
919 index, index);
920
921 dn->node_page = NULL;
922 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
923
924 return 0;
925 }
926
truncate_dnode(struct dnode_of_data * dn)927 static int truncate_dnode(struct dnode_of_data *dn)
928 {
929 struct page *page;
930 int err;
931
932 if (dn->nid == 0)
933 return 1;
934
935 /* get direct node */
936 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
937 if (PTR_ERR(page) == -ENOENT)
938 return 1;
939 else if (IS_ERR(page))
940 return PTR_ERR(page);
941
942 /* Make dnode_of_data for parameter */
943 dn->node_page = page;
944 dn->ofs_in_node = 0;
945 f2fs_truncate_data_blocks(dn);
946 err = truncate_node(dn);
947 if (err)
948 return err;
949
950 return 1;
951 }
952
truncate_nodes(struct dnode_of_data * dn,unsigned int nofs,int ofs,int depth)953 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
954 int ofs, int depth)
955 {
956 struct dnode_of_data rdn = *dn;
957 struct page *page;
958 struct f2fs_node *rn;
959 nid_t child_nid;
960 unsigned int child_nofs;
961 int freed = 0;
962 int i, ret;
963
964 if (dn->nid == 0)
965 return NIDS_PER_BLOCK + 1;
966
967 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
968
969 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
970 if (IS_ERR(page)) {
971 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
972 return PTR_ERR(page);
973 }
974
975 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
976
977 rn = F2FS_NODE(page);
978 if (depth < 3) {
979 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
980 child_nid = le32_to_cpu(rn->in.nid[i]);
981 if (child_nid == 0)
982 continue;
983 rdn.nid = child_nid;
984 ret = truncate_dnode(&rdn);
985 if (ret < 0)
986 goto out_err;
987 if (set_nid(page, i, 0, false))
988 dn->node_changed = true;
989 }
990 } else {
991 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
992 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
993 child_nid = le32_to_cpu(rn->in.nid[i]);
994 if (child_nid == 0) {
995 child_nofs += NIDS_PER_BLOCK + 1;
996 continue;
997 }
998 rdn.nid = child_nid;
999 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
1000 if (ret == (NIDS_PER_BLOCK + 1)) {
1001 if (set_nid(page, i, 0, false))
1002 dn->node_changed = true;
1003 child_nofs += ret;
1004 } else if (ret < 0 && ret != -ENOENT) {
1005 goto out_err;
1006 }
1007 }
1008 freed = child_nofs;
1009 }
1010
1011 if (!ofs) {
1012 /* remove current indirect node */
1013 dn->node_page = page;
1014 ret = truncate_node(dn);
1015 if (ret)
1016 goto out_err;
1017 freed++;
1018 } else {
1019 f2fs_put_page(page, 1);
1020 }
1021 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
1022 return freed;
1023
1024 out_err:
1025 f2fs_put_page(page, 1);
1026 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
1027 return ret;
1028 }
1029
truncate_partial_nodes(struct dnode_of_data * dn,struct f2fs_inode * ri,int * offset,int depth)1030 static int truncate_partial_nodes(struct dnode_of_data *dn,
1031 struct f2fs_inode *ri, int *offset, int depth)
1032 {
1033 struct page *pages[2];
1034 nid_t nid[3];
1035 nid_t child_nid;
1036 int err = 0;
1037 int i;
1038 int idx = depth - 2;
1039
1040 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1041 if (!nid[0])
1042 return 0;
1043
1044 /* get indirect nodes in the path */
1045 for (i = 0; i < idx + 1; i++) {
1046 /* reference count'll be increased */
1047 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
1048 if (IS_ERR(pages[i])) {
1049 err = PTR_ERR(pages[i]);
1050 idx = i - 1;
1051 goto fail;
1052 }
1053 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1054 }
1055
1056 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1057
1058 /* free direct nodes linked to a partial indirect node */
1059 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1060 child_nid = get_nid(pages[idx], i, false);
1061 if (!child_nid)
1062 continue;
1063 dn->nid = child_nid;
1064 err = truncate_dnode(dn);
1065 if (err < 0)
1066 goto fail;
1067 if (set_nid(pages[idx], i, 0, false))
1068 dn->node_changed = true;
1069 }
1070
1071 if (offset[idx + 1] == 0) {
1072 dn->node_page = pages[idx];
1073 dn->nid = nid[idx];
1074 err = truncate_node(dn);
1075 if (err)
1076 goto fail;
1077 } else {
1078 f2fs_put_page(pages[idx], 1);
1079 }
1080 offset[idx]++;
1081 offset[idx + 1] = 0;
1082 idx--;
1083 fail:
1084 for (i = idx; i >= 0; i--)
1085 f2fs_put_page(pages[i], 1);
1086
1087 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1088
1089 return err;
1090 }
1091
1092 /*
1093 * All the block addresses of data and nodes should be nullified.
1094 */
f2fs_truncate_inode_blocks(struct inode * inode,pgoff_t from)1095 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1096 {
1097 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1098 int err = 0, cont = 1;
1099 int level, offset[4], noffset[4];
1100 unsigned int nofs = 0;
1101 struct f2fs_inode *ri;
1102 struct dnode_of_data dn;
1103 struct page *page;
1104
1105 trace_f2fs_truncate_inode_blocks_enter(inode, from);
1106
1107 level = get_node_path(inode, from, offset, noffset);
1108 if (level < 0) {
1109 trace_f2fs_truncate_inode_blocks_exit(inode, level);
1110 return level;
1111 }
1112
1113 page = f2fs_get_node_page(sbi, inode->i_ino);
1114 if (IS_ERR(page)) {
1115 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1116 return PTR_ERR(page);
1117 }
1118
1119 set_new_dnode(&dn, inode, page, NULL, 0);
1120 unlock_page(page);
1121
1122 ri = F2FS_INODE(page);
1123 switch (level) {
1124 case 0:
1125 case 1:
1126 nofs = noffset[1];
1127 break;
1128 case 2:
1129 nofs = noffset[1];
1130 if (!offset[level - 1])
1131 goto skip_partial;
1132 err = truncate_partial_nodes(&dn, ri, offset, level);
1133 if (err < 0 && err != -ENOENT)
1134 goto fail;
1135 nofs += 1 + NIDS_PER_BLOCK;
1136 break;
1137 case 3:
1138 nofs = 5 + 2 * NIDS_PER_BLOCK;
1139 if (!offset[level - 1])
1140 goto skip_partial;
1141 err = truncate_partial_nodes(&dn, ri, offset, level);
1142 if (err < 0 && err != -ENOENT)
1143 goto fail;
1144 break;
1145 default:
1146 BUG();
1147 }
1148
1149 skip_partial:
1150 while (cont) {
1151 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1152 switch (offset[0]) {
1153 case NODE_DIR1_BLOCK:
1154 case NODE_DIR2_BLOCK:
1155 err = truncate_dnode(&dn);
1156 break;
1157
1158 case NODE_IND1_BLOCK:
1159 case NODE_IND2_BLOCK:
1160 err = truncate_nodes(&dn, nofs, offset[1], 2);
1161 break;
1162
1163 case NODE_DIND_BLOCK:
1164 err = truncate_nodes(&dn, nofs, offset[1], 3);
1165 cont = 0;
1166 break;
1167
1168 default:
1169 BUG();
1170 }
1171 if (err < 0 && err != -ENOENT)
1172 goto fail;
1173 if (offset[1] == 0 &&
1174 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1175 lock_page(page);
1176 BUG_ON(page->mapping != NODE_MAPPING(sbi));
1177 f2fs_wait_on_page_writeback(page, NODE, true, true);
1178 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1179 set_page_dirty(page);
1180 unlock_page(page);
1181 }
1182 offset[1] = 0;
1183 offset[0]++;
1184 nofs += err;
1185 }
1186 fail:
1187 f2fs_put_page(page, 0);
1188 trace_f2fs_truncate_inode_blocks_exit(inode, err);
1189 return err > 0 ? 0 : err;
1190 }
1191
1192 /* caller must lock inode page */
f2fs_truncate_xattr_node(struct inode * inode)1193 int f2fs_truncate_xattr_node(struct inode *inode)
1194 {
1195 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1196 nid_t nid = F2FS_I(inode)->i_xattr_nid;
1197 struct dnode_of_data dn;
1198 struct page *npage;
1199 int err;
1200
1201 if (!nid)
1202 return 0;
1203
1204 npage = f2fs_get_node_page(sbi, nid);
1205 if (IS_ERR(npage))
1206 return PTR_ERR(npage);
1207
1208 set_new_dnode(&dn, inode, NULL, npage, nid);
1209 err = truncate_node(&dn);
1210 if (err) {
1211 f2fs_put_page(npage, 1);
1212 return err;
1213 }
1214
1215 f2fs_i_xnid_write(inode, 0);
1216
1217 return 0;
1218 }
1219
1220 /*
1221 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1222 * f2fs_unlock_op().
1223 */
f2fs_remove_inode_page(struct inode * inode)1224 int f2fs_remove_inode_page(struct inode *inode)
1225 {
1226 struct dnode_of_data dn;
1227 int err;
1228
1229 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1230 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1231 if (err)
1232 return err;
1233
1234 err = f2fs_truncate_xattr_node(inode);
1235 if (err) {
1236 f2fs_put_dnode(&dn);
1237 return err;
1238 }
1239
1240 /* remove potential inline_data blocks */
1241 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1242 S_ISLNK(inode->i_mode))
1243 f2fs_truncate_data_blocks_range(&dn, 1);
1244
1245 /* 0 is possible, after f2fs_new_inode() has failed */
1246 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1247 f2fs_put_dnode(&dn);
1248 return -EIO;
1249 }
1250
1251 if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1252 f2fs_warn(F2FS_I_SB(inode),
1253 "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1254 inode->i_ino, (unsigned long long)inode->i_blocks);
1255 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1256 }
1257
1258 /* will put inode & node pages */
1259 err = truncate_node(&dn);
1260 if (err) {
1261 f2fs_put_dnode(&dn);
1262 return err;
1263 }
1264 return 0;
1265 }
1266
f2fs_new_inode_page(struct inode * inode)1267 struct page *f2fs_new_inode_page(struct inode *inode)
1268 {
1269 struct dnode_of_data dn;
1270
1271 /* allocate inode page for new inode */
1272 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1273
1274 /* caller should f2fs_put_page(page, 1); */
1275 return f2fs_new_node_page(&dn, 0);
1276 }
1277
f2fs_new_node_page(struct dnode_of_data * dn,unsigned int ofs)1278 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1279 {
1280 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1281 struct node_info new_ni;
1282 struct page *page;
1283 int err;
1284
1285 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1286 return ERR_PTR(-EPERM);
1287
1288 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1289 if (!page)
1290 return ERR_PTR(-ENOMEM);
1291
1292 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1293 goto fail;
1294
1295 #ifdef CONFIG_F2FS_CHECK_FS
1296 err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false);
1297 if (err) {
1298 dec_valid_node_count(sbi, dn->inode, !ofs);
1299 goto fail;
1300 }
1301 if (unlikely(new_ni.blk_addr != NULL_ADDR)) {
1302 err = -EFSCORRUPTED;
1303 set_sbi_flag(sbi, SBI_NEED_FSCK);
1304 goto fail;
1305 }
1306 #endif
1307 new_ni.nid = dn->nid;
1308 new_ni.ino = dn->inode->i_ino;
1309 new_ni.blk_addr = NULL_ADDR;
1310 new_ni.flag = 0;
1311 new_ni.version = 0;
1312 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1313
1314 f2fs_wait_on_page_writeback(page, NODE, true, true);
1315 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1316 set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1317 if (!PageUptodate(page))
1318 SetPageUptodate(page);
1319 if (set_page_dirty(page))
1320 dn->node_changed = true;
1321
1322 if (f2fs_has_xattr_block(ofs))
1323 f2fs_i_xnid_write(dn->inode, dn->nid);
1324
1325 if (ofs == 0)
1326 inc_valid_inode_count(sbi);
1327 return page;
1328
1329 fail:
1330 clear_node_page_dirty(page);
1331 f2fs_put_page(page, 1);
1332 return ERR_PTR(err);
1333 }
1334
1335 /*
1336 * Caller should do after getting the following values.
1337 * 0: f2fs_put_page(page, 0)
1338 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1339 */
read_node_page(struct page * page,int op_flags)1340 static int read_node_page(struct page *page, int op_flags)
1341 {
1342 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1343 struct node_info ni;
1344 struct f2fs_io_info fio = {
1345 .sbi = sbi,
1346 .type = NODE,
1347 .op = REQ_OP_READ,
1348 .op_flags = op_flags,
1349 .page = page,
1350 .encrypted_page = NULL,
1351 };
1352 int err;
1353
1354 if (PageUptodate(page)) {
1355 if (!f2fs_inode_chksum_verify(sbi, page)) {
1356 ClearPageUptodate(page);
1357 return -EFSBADCRC;
1358 }
1359 return LOCKED_PAGE;
1360 }
1361
1362 err = f2fs_get_node_info(sbi, page->index, &ni, false);
1363 if (err)
1364 return err;
1365
1366 /* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1367 if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR)) {
1368 ClearPageUptodate(page);
1369 return -ENOENT;
1370 }
1371
1372 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1373
1374 err = f2fs_submit_page_bio(&fio);
1375
1376 if (!err)
1377 f2fs_update_iostat(sbi, FS_NODE_READ_IO, F2FS_BLKSIZE);
1378
1379 return err;
1380 }
1381
1382 /*
1383 * Readahead a node page
1384 */
f2fs_ra_node_page(struct f2fs_sb_info * sbi,nid_t nid)1385 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1386 {
1387 struct page *apage;
1388 int err;
1389
1390 if (!nid)
1391 return;
1392 if (f2fs_check_nid_range(sbi, nid))
1393 return;
1394
1395 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1396 if (apage)
1397 return;
1398
1399 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1400 if (!apage)
1401 return;
1402
1403 err = read_node_page(apage, REQ_RAHEAD);
1404 f2fs_put_page(apage, err ? 1 : 0);
1405 }
1406
__get_node_page(struct f2fs_sb_info * sbi,pgoff_t nid,struct page * parent,int start)1407 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1408 struct page *parent, int start)
1409 {
1410 struct page *page;
1411 int err;
1412
1413 if (!nid)
1414 return ERR_PTR(-ENOENT);
1415 if (f2fs_check_nid_range(sbi, nid))
1416 return ERR_PTR(-EINVAL);
1417 repeat:
1418 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1419 if (!page)
1420 return ERR_PTR(-ENOMEM);
1421
1422 err = read_node_page(page, 0);
1423 if (err < 0) {
1424 f2fs_put_page(page, 1);
1425 return ERR_PTR(err);
1426 } else if (err == LOCKED_PAGE) {
1427 err = 0;
1428 goto page_hit;
1429 }
1430
1431 if (parent)
1432 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1433
1434 lock_page(page);
1435
1436 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1437 f2fs_put_page(page, 1);
1438 goto repeat;
1439 }
1440
1441 if (unlikely(!PageUptodate(page))) {
1442 err = -EIO;
1443 goto out_err;
1444 }
1445
1446 if (!f2fs_inode_chksum_verify(sbi, page)) {
1447 err = -EFSBADCRC;
1448 goto out_err;
1449 }
1450 page_hit:
1451 if (unlikely(nid != nid_of_node(page))) {
1452 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1453 nid, nid_of_node(page), ino_of_node(page),
1454 ofs_of_node(page), cpver_of_node(page),
1455 next_blkaddr_of_node(page));
1456 set_sbi_flag(sbi, SBI_NEED_FSCK);
1457 err = -EINVAL;
1458 out_err:
1459 ClearPageUptodate(page);
1460 f2fs_put_page(page, 1);
1461 return ERR_PTR(err);
1462 }
1463 return page;
1464 }
1465
f2fs_get_node_page(struct f2fs_sb_info * sbi,pgoff_t nid)1466 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1467 {
1468 return __get_node_page(sbi, nid, NULL, 0);
1469 }
1470
f2fs_get_node_page_ra(struct page * parent,int start)1471 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1472 {
1473 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1474 nid_t nid = get_nid(parent, start, false);
1475
1476 return __get_node_page(sbi, nid, parent, start);
1477 }
1478
flush_inline_data(struct f2fs_sb_info * sbi,nid_t ino)1479 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1480 {
1481 struct inode *inode;
1482 struct page *page;
1483 int ret;
1484
1485 /* should flush inline_data before evict_inode */
1486 inode = ilookup(sbi->sb, ino);
1487 if (!inode)
1488 return;
1489
1490 page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1491 FGP_LOCK|FGP_NOWAIT, 0);
1492 if (!page)
1493 goto iput_out;
1494
1495 if (!PageUptodate(page))
1496 goto page_out;
1497
1498 if (!PageDirty(page))
1499 goto page_out;
1500
1501 if (!clear_page_dirty_for_io(page))
1502 goto page_out;
1503
1504 ret = f2fs_write_inline_data(inode, page);
1505 inode_dec_dirty_pages(inode);
1506 f2fs_remove_dirty_inode(inode);
1507 if (ret)
1508 set_page_dirty(page);
1509 page_out:
1510 f2fs_put_page(page, 1);
1511 iput_out:
1512 iput(inode);
1513 }
1514
last_fsync_dnode(struct f2fs_sb_info * sbi,nid_t ino)1515 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1516 {
1517 pgoff_t index;
1518 struct pagevec pvec;
1519 struct page *last_page = NULL;
1520 int nr_pages;
1521
1522 pagevec_init(&pvec);
1523 index = 0;
1524
1525 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1526 PAGECACHE_TAG_DIRTY))) {
1527 int i;
1528
1529 for (i = 0; i < nr_pages; i++) {
1530 struct page *page = pvec.pages[i];
1531
1532 if (unlikely(f2fs_cp_error(sbi))) {
1533 f2fs_put_page(last_page, 0);
1534 pagevec_release(&pvec);
1535 return ERR_PTR(-EIO);
1536 }
1537
1538 if (!IS_DNODE(page) || !is_cold_node(page))
1539 continue;
1540 if (ino_of_node(page) != ino)
1541 continue;
1542
1543 lock_page(page);
1544
1545 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1546 continue_unlock:
1547 unlock_page(page);
1548 continue;
1549 }
1550 if (ino_of_node(page) != ino)
1551 goto continue_unlock;
1552
1553 if (!PageDirty(page)) {
1554 /* someone wrote it for us */
1555 goto continue_unlock;
1556 }
1557
1558 if (last_page)
1559 f2fs_put_page(last_page, 0);
1560
1561 get_page(page);
1562 last_page = page;
1563 unlock_page(page);
1564 }
1565 pagevec_release(&pvec);
1566 cond_resched();
1567 }
1568 return last_page;
1569 }
1570
__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)1571 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1572 struct writeback_control *wbc, bool do_balance,
1573 enum iostat_type io_type, unsigned int *seq_id)
1574 {
1575 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1576 nid_t nid;
1577 struct node_info ni;
1578 struct f2fs_io_info fio = {
1579 .sbi = sbi,
1580 .ino = ino_of_node(page),
1581 .type = NODE,
1582 .op = REQ_OP_WRITE,
1583 .op_flags = wbc_to_write_flags(wbc),
1584 .page = page,
1585 .encrypted_page = NULL,
1586 .submitted = false,
1587 .io_type = io_type,
1588 .io_wbc = wbc,
1589 };
1590 unsigned int seq;
1591
1592 trace_f2fs_writepage(page, NODE);
1593
1594 if (unlikely(f2fs_cp_error(sbi))) {
1595 if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
1596 ClearPageUptodate(page);
1597 dec_page_count(sbi, F2FS_DIRTY_NODES);
1598 unlock_page(page);
1599 return 0;
1600 }
1601 goto redirty_out;
1602 }
1603
1604 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1605 goto redirty_out;
1606
1607 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1608 wbc->sync_mode == WB_SYNC_NONE &&
1609 IS_DNODE(page) && is_cold_node(page))
1610 goto redirty_out;
1611
1612 /* get old block addr of this node page */
1613 nid = nid_of_node(page);
1614 f2fs_bug_on(sbi, page->index != nid);
1615
1616 if (f2fs_get_node_info(sbi, nid, &ni, !do_balance))
1617 goto redirty_out;
1618
1619 if (wbc->for_reclaim) {
1620 if (!f2fs_down_read_trylock(&sbi->node_write))
1621 goto redirty_out;
1622 } else {
1623 f2fs_down_read(&sbi->node_write);
1624 }
1625
1626 /* This page is already truncated */
1627 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1628 ClearPageUptodate(page);
1629 dec_page_count(sbi, F2FS_DIRTY_NODES);
1630 f2fs_up_read(&sbi->node_write);
1631 unlock_page(page);
1632 return 0;
1633 }
1634
1635 if (__is_valid_data_blkaddr(ni.blk_addr) &&
1636 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1637 DATA_GENERIC_ENHANCE)) {
1638 f2fs_up_read(&sbi->node_write);
1639 goto redirty_out;
1640 }
1641
1642 if (atomic && !test_opt(sbi, NOBARRIER))
1643 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1644
1645 /* should add to global list before clearing PAGECACHE status */
1646 if (f2fs_in_warm_node_list(sbi, page)) {
1647 seq = f2fs_add_fsync_node_entry(sbi, page);
1648 if (seq_id)
1649 *seq_id = seq;
1650 }
1651
1652 set_page_writeback(page);
1653 ClearPageError(page);
1654
1655 fio.old_blkaddr = ni.blk_addr;
1656 f2fs_do_write_node_page(nid, &fio);
1657 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1658 dec_page_count(sbi, F2FS_DIRTY_NODES);
1659 f2fs_up_read(&sbi->node_write);
1660
1661 if (wbc->for_reclaim) {
1662 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1663 submitted = NULL;
1664 }
1665
1666 unlock_page(page);
1667
1668 if (unlikely(f2fs_cp_error(sbi))) {
1669 f2fs_submit_merged_write(sbi, NODE);
1670 submitted = NULL;
1671 }
1672 if (submitted)
1673 *submitted = fio.submitted;
1674
1675 if (do_balance)
1676 f2fs_balance_fs(sbi, false);
1677 return 0;
1678
1679 redirty_out:
1680 redirty_page_for_writepage(wbc, page);
1681 return AOP_WRITEPAGE_ACTIVATE;
1682 }
1683
f2fs_move_node_page(struct page * node_page,int gc_type)1684 int f2fs_move_node_page(struct page *node_page, int gc_type)
1685 {
1686 int err = 0;
1687
1688 if (gc_type == FG_GC) {
1689 struct writeback_control wbc = {
1690 .sync_mode = WB_SYNC_ALL,
1691 .nr_to_write = 1,
1692 .for_reclaim = 0,
1693 };
1694
1695 f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1696
1697 set_page_dirty(node_page);
1698
1699 if (!clear_page_dirty_for_io(node_page)) {
1700 err = -EAGAIN;
1701 goto out_page;
1702 }
1703
1704 if (__write_node_page(node_page, false, NULL,
1705 &wbc, false, FS_GC_NODE_IO, NULL)) {
1706 err = -EAGAIN;
1707 unlock_page(node_page);
1708 }
1709 goto release_page;
1710 } else {
1711 /* set page dirty and write it */
1712 if (!PageWriteback(node_page))
1713 set_page_dirty(node_page);
1714 }
1715 out_page:
1716 unlock_page(node_page);
1717 release_page:
1718 f2fs_put_page(node_page, 0);
1719 return err;
1720 }
1721
f2fs_write_node_page(struct page * page,struct writeback_control * wbc)1722 static int f2fs_write_node_page(struct page *page,
1723 struct writeback_control *wbc)
1724 {
1725 return __write_node_page(page, false, NULL, wbc, false,
1726 FS_NODE_IO, NULL);
1727 }
1728
f2fs_fsync_node_pages(struct f2fs_sb_info * sbi,struct inode * inode,struct writeback_control * wbc,bool atomic,unsigned int * seq_id)1729 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1730 struct writeback_control *wbc, bool atomic,
1731 unsigned int *seq_id)
1732 {
1733 pgoff_t index;
1734 struct pagevec pvec;
1735 int ret = 0;
1736 struct page *last_page = NULL;
1737 bool marked = false;
1738 nid_t ino = inode->i_ino;
1739 int nr_pages;
1740 int nwritten = 0;
1741
1742 if (atomic) {
1743 last_page = last_fsync_dnode(sbi, ino);
1744 if (IS_ERR_OR_NULL(last_page))
1745 return PTR_ERR_OR_ZERO(last_page);
1746 }
1747 retry:
1748 pagevec_init(&pvec);
1749 index = 0;
1750
1751 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1752 PAGECACHE_TAG_DIRTY))) {
1753 int i;
1754
1755 for (i = 0; i < nr_pages; i++) {
1756 struct page *page = pvec.pages[i];
1757 bool submitted = false;
1758
1759 if (unlikely(f2fs_cp_error(sbi))) {
1760 f2fs_put_page(last_page, 0);
1761 pagevec_release(&pvec);
1762 ret = -EIO;
1763 goto out;
1764 }
1765
1766 if (!IS_DNODE(page) || !is_cold_node(page))
1767 continue;
1768 if (ino_of_node(page) != ino)
1769 continue;
1770
1771 lock_page(page);
1772
1773 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1774 continue_unlock:
1775 unlock_page(page);
1776 continue;
1777 }
1778 if (ino_of_node(page) != ino)
1779 goto continue_unlock;
1780
1781 if (!PageDirty(page) && page != last_page) {
1782 /* someone wrote it for us */
1783 goto continue_unlock;
1784 }
1785
1786 f2fs_wait_on_page_writeback(page, NODE, true, true);
1787
1788 set_fsync_mark(page, 0);
1789 set_dentry_mark(page, 0);
1790
1791 if (!atomic || page == last_page) {
1792 set_fsync_mark(page, 1);
1793 if (IS_INODE(page)) {
1794 if (is_inode_flag_set(inode,
1795 FI_DIRTY_INODE))
1796 f2fs_update_inode(inode, page);
1797 set_dentry_mark(page,
1798 f2fs_need_dentry_mark(sbi, ino));
1799 }
1800 /* may be written by other thread */
1801 if (!PageDirty(page))
1802 set_page_dirty(page);
1803 }
1804
1805 if (!clear_page_dirty_for_io(page))
1806 goto continue_unlock;
1807
1808 ret = __write_node_page(page, atomic &&
1809 page == last_page,
1810 &submitted, wbc, true,
1811 FS_NODE_IO, seq_id);
1812 if (ret) {
1813 unlock_page(page);
1814 f2fs_put_page(last_page, 0);
1815 break;
1816 } else if (submitted) {
1817 nwritten++;
1818 }
1819
1820 if (page == last_page) {
1821 f2fs_put_page(page, 0);
1822 marked = true;
1823 break;
1824 }
1825 }
1826 pagevec_release(&pvec);
1827 cond_resched();
1828
1829 if (ret || marked)
1830 break;
1831 }
1832 if (!ret && atomic && !marked) {
1833 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1834 ino, last_page->index);
1835 lock_page(last_page);
1836 f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1837 set_page_dirty(last_page);
1838 unlock_page(last_page);
1839 goto retry;
1840 }
1841 out:
1842 if (nwritten)
1843 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1844 return ret ? -EIO : 0;
1845 }
1846
f2fs_match_ino(struct inode * inode,unsigned long ino,void * data)1847 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1848 {
1849 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1850 bool clean;
1851
1852 if (inode->i_ino != ino)
1853 return 0;
1854
1855 if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1856 return 0;
1857
1858 spin_lock(&sbi->inode_lock[DIRTY_META]);
1859 clean = list_empty(&F2FS_I(inode)->gdirty_list);
1860 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1861
1862 if (clean)
1863 return 0;
1864
1865 inode = igrab(inode);
1866 if (!inode)
1867 return 0;
1868 return 1;
1869 }
1870
flush_dirty_inode(struct page * page)1871 static bool flush_dirty_inode(struct page *page)
1872 {
1873 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1874 struct inode *inode;
1875 nid_t ino = ino_of_node(page);
1876
1877 inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1878 if (!inode)
1879 return false;
1880
1881 f2fs_update_inode(inode, page);
1882 unlock_page(page);
1883
1884 iput(inode);
1885 return true;
1886 }
1887
f2fs_flush_inline_data(struct f2fs_sb_info * sbi)1888 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1889 {
1890 pgoff_t index = 0;
1891 struct pagevec pvec;
1892 int nr_pages;
1893
1894 pagevec_init(&pvec);
1895
1896 while ((nr_pages = pagevec_lookup_tag(&pvec,
1897 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1898 int i;
1899
1900 for (i = 0; i < nr_pages; i++) {
1901 struct page *page = pvec.pages[i];
1902
1903 if (!IS_DNODE(page))
1904 continue;
1905
1906 lock_page(page);
1907
1908 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1909 continue_unlock:
1910 unlock_page(page);
1911 continue;
1912 }
1913
1914 if (!PageDirty(page)) {
1915 /* someone wrote it for us */
1916 goto continue_unlock;
1917 }
1918
1919 /* flush inline_data, if it's async context. */
1920 if (page_private_inline(page)) {
1921 clear_page_private_inline(page);
1922 unlock_page(page);
1923 flush_inline_data(sbi, ino_of_node(page));
1924 continue;
1925 }
1926 unlock_page(page);
1927 }
1928 pagevec_release(&pvec);
1929 cond_resched();
1930 }
1931 }
1932
f2fs_sync_node_pages(struct f2fs_sb_info * sbi,struct writeback_control * wbc,bool do_balance,enum iostat_type io_type)1933 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1934 struct writeback_control *wbc,
1935 bool do_balance, enum iostat_type io_type)
1936 {
1937 pgoff_t index;
1938 struct pagevec pvec;
1939 int step = 0;
1940 int nwritten = 0;
1941 int ret = 0;
1942 int nr_pages, done = 0;
1943
1944 pagevec_init(&pvec);
1945
1946 next_step:
1947 index = 0;
1948
1949 while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1950 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1951 int i;
1952
1953 for (i = 0; i < nr_pages; i++) {
1954 struct page *page = pvec.pages[i];
1955 bool submitted = false;
1956 bool may_dirty = true;
1957
1958 /* give a priority to WB_SYNC threads */
1959 if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1960 wbc->sync_mode == WB_SYNC_NONE) {
1961 done = 1;
1962 break;
1963 }
1964
1965 /*
1966 * flushing sequence with step:
1967 * 0. indirect nodes
1968 * 1. dentry dnodes
1969 * 2. file dnodes
1970 */
1971 if (step == 0 && IS_DNODE(page))
1972 continue;
1973 if (step == 1 && (!IS_DNODE(page) ||
1974 is_cold_node(page)))
1975 continue;
1976 if (step == 2 && (!IS_DNODE(page) ||
1977 !is_cold_node(page)))
1978 continue;
1979 lock_node:
1980 if (wbc->sync_mode == WB_SYNC_ALL)
1981 lock_page(page);
1982 else if (!trylock_page(page))
1983 continue;
1984
1985 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1986 continue_unlock:
1987 unlock_page(page);
1988 continue;
1989 }
1990
1991 if (!PageDirty(page)) {
1992 /* someone wrote it for us */
1993 goto continue_unlock;
1994 }
1995
1996 /* flush inline_data/inode, if it's async context. */
1997 if (!do_balance)
1998 goto write_node;
1999
2000 /* flush inline_data */
2001 if (page_private_inline(page)) {
2002 clear_page_private_inline(page);
2003 unlock_page(page);
2004 flush_inline_data(sbi, ino_of_node(page));
2005 goto lock_node;
2006 }
2007
2008 /* flush dirty inode */
2009 if (IS_INODE(page) && may_dirty) {
2010 may_dirty = false;
2011 if (flush_dirty_inode(page))
2012 goto lock_node;
2013 }
2014 write_node:
2015 f2fs_wait_on_page_writeback(page, NODE, true, true);
2016
2017 if (!clear_page_dirty_for_io(page))
2018 goto continue_unlock;
2019
2020 set_fsync_mark(page, 0);
2021 set_dentry_mark(page, 0);
2022
2023 ret = __write_node_page(page, false, &submitted,
2024 wbc, do_balance, io_type, NULL);
2025 if (ret)
2026 unlock_page(page);
2027 else if (submitted)
2028 nwritten++;
2029
2030 if (--wbc->nr_to_write == 0)
2031 break;
2032 }
2033 pagevec_release(&pvec);
2034 cond_resched();
2035
2036 if (wbc->nr_to_write == 0) {
2037 step = 2;
2038 break;
2039 }
2040 }
2041
2042 if (step < 2) {
2043 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2044 wbc->sync_mode == WB_SYNC_NONE && step == 1)
2045 goto out;
2046 step++;
2047 goto next_step;
2048 }
2049 out:
2050 if (nwritten)
2051 f2fs_submit_merged_write(sbi, NODE);
2052
2053 if (unlikely(f2fs_cp_error(sbi)))
2054 return -EIO;
2055 return ret;
2056 }
2057
f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info * sbi,unsigned int seq_id)2058 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2059 unsigned int seq_id)
2060 {
2061 struct fsync_node_entry *fn;
2062 struct page *page;
2063 struct list_head *head = &sbi->fsync_node_list;
2064 unsigned long flags;
2065 unsigned int cur_seq_id = 0;
2066 int ret2, ret = 0;
2067
2068 while (seq_id && cur_seq_id < seq_id) {
2069 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2070 if (list_empty(head)) {
2071 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2072 break;
2073 }
2074 fn = list_first_entry(head, struct fsync_node_entry, list);
2075 if (fn->seq_id > seq_id) {
2076 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2077 break;
2078 }
2079 cur_seq_id = fn->seq_id;
2080 page = fn->page;
2081 get_page(page);
2082 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2083
2084 f2fs_wait_on_page_writeback(page, NODE, true, false);
2085 if (TestClearPageError(page))
2086 ret = -EIO;
2087
2088 put_page(page);
2089
2090 if (ret)
2091 break;
2092 }
2093
2094 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2095 if (!ret)
2096 ret = ret2;
2097
2098 return ret;
2099 }
2100
f2fs_write_node_pages(struct address_space * mapping,struct writeback_control * wbc)2101 static int f2fs_write_node_pages(struct address_space *mapping,
2102 struct writeback_control *wbc)
2103 {
2104 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2105 struct blk_plug plug;
2106 long diff;
2107
2108 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2109 goto skip_write;
2110
2111 /* balancing f2fs's metadata in background */
2112 f2fs_balance_fs_bg(sbi, true);
2113
2114 /* collect a number of dirty node pages and write together */
2115 if (wbc->sync_mode != WB_SYNC_ALL &&
2116 get_pages(sbi, F2FS_DIRTY_NODES) <
2117 nr_pages_to_skip(sbi, NODE))
2118 goto skip_write;
2119
2120 if (wbc->sync_mode == WB_SYNC_ALL)
2121 atomic_inc(&sbi->wb_sync_req[NODE]);
2122 else if (atomic_read(&sbi->wb_sync_req[NODE])) {
2123 /* to avoid potential deadlock */
2124 if (current->plug)
2125 blk_finish_plug(current->plug);
2126 goto skip_write;
2127 }
2128
2129 trace_f2fs_writepages(mapping->host, wbc, NODE);
2130
2131 diff = nr_pages_to_write(sbi, NODE, wbc);
2132 blk_start_plug(&plug);
2133 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2134 blk_finish_plug(&plug);
2135 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2136
2137 if (wbc->sync_mode == WB_SYNC_ALL)
2138 atomic_dec(&sbi->wb_sync_req[NODE]);
2139 return 0;
2140
2141 skip_write:
2142 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2143 trace_f2fs_writepages(mapping->host, wbc, NODE);
2144 return 0;
2145 }
2146
f2fs_set_node_page_dirty(struct page * page)2147 static int f2fs_set_node_page_dirty(struct page *page)
2148 {
2149 trace_f2fs_set_page_dirty(page, NODE);
2150
2151 if (!PageUptodate(page))
2152 SetPageUptodate(page);
2153 #ifdef CONFIG_F2FS_CHECK_FS
2154 if (IS_INODE(page))
2155 f2fs_inode_chksum_set(F2FS_P_SB(page), page);
2156 #endif
2157 if (!PageDirty(page)) {
2158 __set_page_dirty_nobuffers(page);
2159 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
2160 set_page_private_reference(page);
2161 return 1;
2162 }
2163 return 0;
2164 }
2165
2166 /*
2167 * Structure of the f2fs node operations
2168 */
2169 const struct address_space_operations f2fs_node_aops = {
2170 .writepage = f2fs_write_node_page,
2171 .writepages = f2fs_write_node_pages,
2172 .set_page_dirty = f2fs_set_node_page_dirty,
2173 .invalidatepage = f2fs_invalidate_page,
2174 .releasepage = f2fs_release_page,
2175 #ifdef CONFIG_MIGRATION
2176 .migratepage = f2fs_migrate_page,
2177 #endif
2178 };
2179
__lookup_free_nid_list(struct f2fs_nm_info * nm_i,nid_t n)2180 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2181 nid_t n)
2182 {
2183 return radix_tree_lookup(&nm_i->free_nid_root, n);
2184 }
2185
__insert_free_nid(struct f2fs_sb_info * sbi,struct free_nid * i)2186 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2187 struct free_nid *i)
2188 {
2189 struct f2fs_nm_info *nm_i = NM_I(sbi);
2190 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2191
2192 if (err)
2193 return err;
2194
2195 nm_i->nid_cnt[FREE_NID]++;
2196 list_add_tail(&i->list, &nm_i->free_nid_list);
2197 return 0;
2198 }
2199
__remove_free_nid(struct f2fs_sb_info * sbi,struct free_nid * i,enum nid_state state)2200 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2201 struct free_nid *i, enum nid_state state)
2202 {
2203 struct f2fs_nm_info *nm_i = NM_I(sbi);
2204
2205 f2fs_bug_on(sbi, state != i->state);
2206 nm_i->nid_cnt[state]--;
2207 if (state == FREE_NID)
2208 list_del(&i->list);
2209 radix_tree_delete(&nm_i->free_nid_root, i->nid);
2210 }
2211
__move_free_nid(struct f2fs_sb_info * sbi,struct free_nid * i,enum nid_state org_state,enum nid_state dst_state)2212 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2213 enum nid_state org_state, enum nid_state dst_state)
2214 {
2215 struct f2fs_nm_info *nm_i = NM_I(sbi);
2216
2217 f2fs_bug_on(sbi, org_state != i->state);
2218 i->state = dst_state;
2219 nm_i->nid_cnt[org_state]--;
2220 nm_i->nid_cnt[dst_state]++;
2221
2222 switch (dst_state) {
2223 case PREALLOC_NID:
2224 list_del(&i->list);
2225 break;
2226 case FREE_NID:
2227 list_add_tail(&i->list, &nm_i->free_nid_list);
2228 break;
2229 default:
2230 BUG_ON(1);
2231 }
2232 }
2233
update_free_nid_bitmap(struct f2fs_sb_info * sbi,nid_t nid,bool set,bool build)2234 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2235 bool set, bool build)
2236 {
2237 struct f2fs_nm_info *nm_i = NM_I(sbi);
2238 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2239 unsigned int nid_ofs = nid - START_NID(nid);
2240
2241 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2242 return;
2243
2244 if (set) {
2245 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2246 return;
2247 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2248 nm_i->free_nid_count[nat_ofs]++;
2249 } else {
2250 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2251 return;
2252 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2253 if (!build)
2254 nm_i->free_nid_count[nat_ofs]--;
2255 }
2256 }
2257
2258 /* return if the nid is recognized as free */
add_free_nid(struct f2fs_sb_info * sbi,nid_t nid,bool build,bool update)2259 static bool add_free_nid(struct f2fs_sb_info *sbi,
2260 nid_t nid, bool build, bool update)
2261 {
2262 struct f2fs_nm_info *nm_i = NM_I(sbi);
2263 struct free_nid *i, *e;
2264 struct nat_entry *ne;
2265 int err = -EINVAL;
2266 bool ret = false;
2267
2268 /* 0 nid should not be used */
2269 if (unlikely(nid == 0))
2270 return false;
2271
2272 if (unlikely(f2fs_check_nid_range(sbi, nid)))
2273 return false;
2274
2275 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2276 i->nid = nid;
2277 i->state = FREE_NID;
2278
2279 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2280
2281 spin_lock(&nm_i->nid_list_lock);
2282
2283 if (build) {
2284 /*
2285 * Thread A Thread B
2286 * - f2fs_create
2287 * - f2fs_new_inode
2288 * - f2fs_alloc_nid
2289 * - __insert_nid_to_list(PREALLOC_NID)
2290 * - f2fs_balance_fs_bg
2291 * - f2fs_build_free_nids
2292 * - __f2fs_build_free_nids
2293 * - scan_nat_page
2294 * - add_free_nid
2295 * - __lookup_nat_cache
2296 * - f2fs_add_link
2297 * - f2fs_init_inode_metadata
2298 * - f2fs_new_inode_page
2299 * - f2fs_new_node_page
2300 * - set_node_addr
2301 * - f2fs_alloc_nid_done
2302 * - __remove_nid_from_list(PREALLOC_NID)
2303 * - __insert_nid_to_list(FREE_NID)
2304 */
2305 ne = __lookup_nat_cache(nm_i, nid);
2306 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2307 nat_get_blkaddr(ne) != NULL_ADDR))
2308 goto err_out;
2309
2310 e = __lookup_free_nid_list(nm_i, nid);
2311 if (e) {
2312 if (e->state == FREE_NID)
2313 ret = true;
2314 goto err_out;
2315 }
2316 }
2317 ret = true;
2318 err = __insert_free_nid(sbi, i);
2319 err_out:
2320 if (update) {
2321 update_free_nid_bitmap(sbi, nid, ret, build);
2322 if (!build)
2323 nm_i->available_nids++;
2324 }
2325 spin_unlock(&nm_i->nid_list_lock);
2326 radix_tree_preload_end();
2327
2328 if (err)
2329 kmem_cache_free(free_nid_slab, i);
2330 return ret;
2331 }
2332
remove_free_nid(struct f2fs_sb_info * sbi,nid_t nid)2333 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2334 {
2335 struct f2fs_nm_info *nm_i = NM_I(sbi);
2336 struct free_nid *i;
2337 bool need_free = false;
2338
2339 spin_lock(&nm_i->nid_list_lock);
2340 i = __lookup_free_nid_list(nm_i, nid);
2341 if (i && i->state == FREE_NID) {
2342 __remove_free_nid(sbi, i, FREE_NID);
2343 need_free = true;
2344 }
2345 spin_unlock(&nm_i->nid_list_lock);
2346
2347 if (need_free)
2348 kmem_cache_free(free_nid_slab, i);
2349 }
2350
scan_nat_page(struct f2fs_sb_info * sbi,struct page * nat_page,nid_t start_nid)2351 static int scan_nat_page(struct f2fs_sb_info *sbi,
2352 struct page *nat_page, nid_t start_nid)
2353 {
2354 struct f2fs_nm_info *nm_i = NM_I(sbi);
2355 struct f2fs_nat_block *nat_blk = page_address(nat_page);
2356 block_t blk_addr;
2357 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2358 int i;
2359
2360 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2361
2362 i = start_nid % NAT_ENTRY_PER_BLOCK;
2363
2364 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2365 if (unlikely(start_nid >= nm_i->max_nid))
2366 break;
2367
2368 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2369
2370 if (blk_addr == NEW_ADDR)
2371 return -EINVAL;
2372
2373 if (blk_addr == NULL_ADDR) {
2374 add_free_nid(sbi, start_nid, true, true);
2375 } else {
2376 spin_lock(&NM_I(sbi)->nid_list_lock);
2377 update_free_nid_bitmap(sbi, start_nid, false, true);
2378 spin_unlock(&NM_I(sbi)->nid_list_lock);
2379 }
2380 }
2381
2382 return 0;
2383 }
2384
scan_curseg_cache(struct f2fs_sb_info * sbi)2385 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2386 {
2387 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2388 struct f2fs_journal *journal = curseg->journal;
2389 int i;
2390
2391 down_read(&curseg->journal_rwsem);
2392 for (i = 0; i < nats_in_cursum(journal); i++) {
2393 block_t addr;
2394 nid_t nid;
2395
2396 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2397 nid = le32_to_cpu(nid_in_journal(journal, i));
2398 if (addr == NULL_ADDR)
2399 add_free_nid(sbi, nid, true, false);
2400 else
2401 remove_free_nid(sbi, nid);
2402 }
2403 up_read(&curseg->journal_rwsem);
2404 }
2405
scan_free_nid_bits(struct f2fs_sb_info * sbi)2406 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2407 {
2408 struct f2fs_nm_info *nm_i = NM_I(sbi);
2409 unsigned int i, idx;
2410 nid_t nid;
2411
2412 f2fs_down_read(&nm_i->nat_tree_lock);
2413
2414 for (i = 0; i < nm_i->nat_blocks; i++) {
2415 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2416 continue;
2417 if (!nm_i->free_nid_count[i])
2418 continue;
2419 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2420 idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2421 NAT_ENTRY_PER_BLOCK, idx);
2422 if (idx >= NAT_ENTRY_PER_BLOCK)
2423 break;
2424
2425 nid = i * NAT_ENTRY_PER_BLOCK + idx;
2426 add_free_nid(sbi, nid, true, false);
2427
2428 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2429 goto out;
2430 }
2431 }
2432 out:
2433 scan_curseg_cache(sbi);
2434
2435 f2fs_up_read(&nm_i->nat_tree_lock);
2436 }
2437
__f2fs_build_free_nids(struct f2fs_sb_info * sbi,bool sync,bool mount)2438 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2439 bool sync, bool mount)
2440 {
2441 struct f2fs_nm_info *nm_i = NM_I(sbi);
2442 int i = 0, ret;
2443 nid_t nid = nm_i->next_scan_nid;
2444
2445 if (unlikely(nid >= nm_i->max_nid))
2446 nid = 0;
2447
2448 if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2449 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2450
2451 /* Enough entries */
2452 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2453 return 0;
2454
2455 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2456 return 0;
2457
2458 if (!mount) {
2459 /* try to find free nids in free_nid_bitmap */
2460 scan_free_nid_bits(sbi);
2461
2462 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2463 return 0;
2464 }
2465
2466 /* readahead nat pages to be scanned */
2467 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2468 META_NAT, true);
2469
2470 f2fs_down_read(&nm_i->nat_tree_lock);
2471
2472 while (1) {
2473 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2474 nm_i->nat_block_bitmap)) {
2475 struct page *page = get_current_nat_page(sbi, nid);
2476
2477 if (IS_ERR(page)) {
2478 ret = PTR_ERR(page);
2479 } else {
2480 ret = scan_nat_page(sbi, page, nid);
2481 f2fs_put_page(page, 1);
2482 }
2483
2484 if (ret) {
2485 f2fs_up_read(&nm_i->nat_tree_lock);
2486 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2487 return ret;
2488 }
2489 }
2490
2491 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2492 if (unlikely(nid >= nm_i->max_nid))
2493 nid = 0;
2494
2495 if (++i >= FREE_NID_PAGES)
2496 break;
2497 }
2498
2499 /* go to the next free nat pages to find free nids abundantly */
2500 nm_i->next_scan_nid = nid;
2501
2502 /* find free nids from current sum_pages */
2503 scan_curseg_cache(sbi);
2504
2505 f2fs_up_read(&nm_i->nat_tree_lock);
2506
2507 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2508 nm_i->ra_nid_pages, META_NAT, false);
2509
2510 return 0;
2511 }
2512
f2fs_build_free_nids(struct f2fs_sb_info * sbi,bool sync,bool mount)2513 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2514 {
2515 int ret;
2516
2517 mutex_lock(&NM_I(sbi)->build_lock);
2518 ret = __f2fs_build_free_nids(sbi, sync, mount);
2519 mutex_unlock(&NM_I(sbi)->build_lock);
2520
2521 return ret;
2522 }
2523
2524 /*
2525 * If this function returns success, caller can obtain a new nid
2526 * from second parameter of this function.
2527 * The returned nid could be used ino as well as nid when inode is created.
2528 */
f2fs_alloc_nid(struct f2fs_sb_info * sbi,nid_t * nid)2529 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2530 {
2531 struct f2fs_nm_info *nm_i = NM_I(sbi);
2532 struct free_nid *i = NULL;
2533 retry:
2534 if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2535 f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2536 return false;
2537 }
2538
2539 spin_lock(&nm_i->nid_list_lock);
2540
2541 if (unlikely(nm_i->available_nids == 0)) {
2542 spin_unlock(&nm_i->nid_list_lock);
2543 return false;
2544 }
2545
2546 /* We should not use stale free nids created by f2fs_build_free_nids */
2547 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2548 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2549 i = list_first_entry(&nm_i->free_nid_list,
2550 struct free_nid, list);
2551 *nid = i->nid;
2552
2553 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2554 nm_i->available_nids--;
2555
2556 update_free_nid_bitmap(sbi, *nid, false, false);
2557
2558 spin_unlock(&nm_i->nid_list_lock);
2559 return true;
2560 }
2561 spin_unlock(&nm_i->nid_list_lock);
2562
2563 /* Let's scan nat pages and its caches to get free nids */
2564 if (!f2fs_build_free_nids(sbi, true, false))
2565 goto retry;
2566 return false;
2567 }
2568
2569 /*
2570 * f2fs_alloc_nid() should be called prior to this function.
2571 */
f2fs_alloc_nid_done(struct f2fs_sb_info * sbi,nid_t nid)2572 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2573 {
2574 struct f2fs_nm_info *nm_i = NM_I(sbi);
2575 struct free_nid *i;
2576
2577 spin_lock(&nm_i->nid_list_lock);
2578 i = __lookup_free_nid_list(nm_i, nid);
2579 f2fs_bug_on(sbi, !i);
2580 __remove_free_nid(sbi, i, PREALLOC_NID);
2581 spin_unlock(&nm_i->nid_list_lock);
2582
2583 kmem_cache_free(free_nid_slab, i);
2584 }
2585
2586 /*
2587 * f2fs_alloc_nid() should be called prior to this function.
2588 */
f2fs_alloc_nid_failed(struct f2fs_sb_info * sbi,nid_t nid)2589 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2590 {
2591 struct f2fs_nm_info *nm_i = NM_I(sbi);
2592 struct free_nid *i;
2593 bool need_free = false;
2594
2595 if (!nid)
2596 return;
2597
2598 spin_lock(&nm_i->nid_list_lock);
2599 i = __lookup_free_nid_list(nm_i, nid);
2600 f2fs_bug_on(sbi, !i);
2601
2602 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2603 __remove_free_nid(sbi, i, PREALLOC_NID);
2604 need_free = true;
2605 } else {
2606 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2607 }
2608
2609 nm_i->available_nids++;
2610
2611 update_free_nid_bitmap(sbi, nid, true, false);
2612
2613 spin_unlock(&nm_i->nid_list_lock);
2614
2615 if (need_free)
2616 kmem_cache_free(free_nid_slab, i);
2617 }
2618
f2fs_try_to_free_nids(struct f2fs_sb_info * sbi,int nr_shrink)2619 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2620 {
2621 struct f2fs_nm_info *nm_i = NM_I(sbi);
2622 int nr = nr_shrink;
2623
2624 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2625 return 0;
2626
2627 if (!mutex_trylock(&nm_i->build_lock))
2628 return 0;
2629
2630 while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2631 struct free_nid *i, *next;
2632 unsigned int batch = SHRINK_NID_BATCH_SIZE;
2633
2634 spin_lock(&nm_i->nid_list_lock);
2635 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2636 if (!nr_shrink || !batch ||
2637 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2638 break;
2639 __remove_free_nid(sbi, i, FREE_NID);
2640 kmem_cache_free(free_nid_slab, i);
2641 nr_shrink--;
2642 batch--;
2643 }
2644 spin_unlock(&nm_i->nid_list_lock);
2645 }
2646
2647 mutex_unlock(&nm_i->build_lock);
2648
2649 return nr - nr_shrink;
2650 }
2651
f2fs_recover_inline_xattr(struct inode * inode,struct page * page)2652 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2653 {
2654 void *src_addr, *dst_addr;
2655 size_t inline_size;
2656 struct page *ipage;
2657 struct f2fs_inode *ri;
2658
2659 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2660 if (IS_ERR(ipage))
2661 return PTR_ERR(ipage);
2662
2663 ri = F2FS_INODE(page);
2664 if (ri->i_inline & F2FS_INLINE_XATTR) {
2665 if (!f2fs_has_inline_xattr(inode)) {
2666 set_inode_flag(inode, FI_INLINE_XATTR);
2667 stat_inc_inline_xattr(inode);
2668 }
2669 } else {
2670 if (f2fs_has_inline_xattr(inode)) {
2671 stat_dec_inline_xattr(inode);
2672 clear_inode_flag(inode, FI_INLINE_XATTR);
2673 }
2674 goto update_inode;
2675 }
2676
2677 dst_addr = inline_xattr_addr(inode, ipage);
2678 src_addr = inline_xattr_addr(inode, page);
2679 inline_size = inline_xattr_size(inode);
2680
2681 f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2682 memcpy(dst_addr, src_addr, inline_size);
2683 update_inode:
2684 f2fs_update_inode(inode, ipage);
2685 f2fs_put_page(ipage, 1);
2686 return 0;
2687 }
2688
f2fs_recover_xattr_data(struct inode * inode,struct page * page)2689 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2690 {
2691 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2692 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2693 nid_t new_xnid;
2694 struct dnode_of_data dn;
2695 struct node_info ni;
2696 struct page *xpage;
2697 int err;
2698
2699 if (!prev_xnid)
2700 goto recover_xnid;
2701
2702 /* 1: invalidate the previous xattr nid */
2703 err = f2fs_get_node_info(sbi, prev_xnid, &ni, false);
2704 if (err)
2705 return err;
2706
2707 f2fs_invalidate_blocks(sbi, ni.blk_addr);
2708 dec_valid_node_count(sbi, inode, false);
2709 set_node_addr(sbi, &ni, NULL_ADDR, false);
2710
2711 recover_xnid:
2712 /* 2: update xattr nid in inode */
2713 if (!f2fs_alloc_nid(sbi, &new_xnid))
2714 return -ENOSPC;
2715
2716 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2717 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2718 if (IS_ERR(xpage)) {
2719 f2fs_alloc_nid_failed(sbi, new_xnid);
2720 return PTR_ERR(xpage);
2721 }
2722
2723 f2fs_alloc_nid_done(sbi, new_xnid);
2724 f2fs_update_inode_page(inode);
2725
2726 /* 3: update and set xattr node page dirty */
2727 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2728
2729 set_page_dirty(xpage);
2730 f2fs_put_page(xpage, 1);
2731
2732 return 0;
2733 }
2734
f2fs_recover_inode_page(struct f2fs_sb_info * sbi,struct page * page)2735 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2736 {
2737 struct f2fs_inode *src, *dst;
2738 nid_t ino = ino_of_node(page);
2739 struct node_info old_ni, new_ni;
2740 struct page *ipage;
2741 int err;
2742
2743 err = f2fs_get_node_info(sbi, ino, &old_ni, false);
2744 if (err)
2745 return err;
2746
2747 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2748 return -EINVAL;
2749 retry:
2750 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2751 if (!ipage) {
2752 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2753 goto retry;
2754 }
2755
2756 /* Should not use this inode from free nid list */
2757 remove_free_nid(sbi, ino);
2758
2759 if (!PageUptodate(ipage))
2760 SetPageUptodate(ipage);
2761 fill_node_footer(ipage, ino, ino, 0, true);
2762 set_cold_node(ipage, false);
2763
2764 src = F2FS_INODE(page);
2765 dst = F2FS_INODE(ipage);
2766
2767 memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2768 dst->i_size = 0;
2769 dst->i_blocks = cpu_to_le64(1);
2770 dst->i_links = cpu_to_le32(1);
2771 dst->i_xattr_nid = 0;
2772 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2773 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2774 dst->i_extra_isize = src->i_extra_isize;
2775
2776 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2777 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2778 i_inline_xattr_size))
2779 dst->i_inline_xattr_size = src->i_inline_xattr_size;
2780
2781 if (f2fs_sb_has_project_quota(sbi) &&
2782 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2783 i_projid))
2784 dst->i_projid = src->i_projid;
2785
2786 if (f2fs_sb_has_inode_crtime(sbi) &&
2787 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2788 i_crtime_nsec)) {
2789 dst->i_crtime = src->i_crtime;
2790 dst->i_crtime_nsec = src->i_crtime_nsec;
2791 }
2792 }
2793
2794 new_ni = old_ni;
2795 new_ni.ino = ino;
2796
2797 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2798 WARN_ON(1);
2799 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2800 inc_valid_inode_count(sbi);
2801 set_page_dirty(ipage);
2802 f2fs_put_page(ipage, 1);
2803 return 0;
2804 }
2805
f2fs_restore_node_summary(struct f2fs_sb_info * sbi,unsigned int segno,struct f2fs_summary_block * sum)2806 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2807 unsigned int segno, struct f2fs_summary_block *sum)
2808 {
2809 struct f2fs_node *rn;
2810 struct f2fs_summary *sum_entry;
2811 block_t addr;
2812 int i, idx, last_offset, nrpages;
2813
2814 /* scan the node segment */
2815 last_offset = sbi->blocks_per_seg;
2816 addr = START_BLOCK(sbi, segno);
2817 sum_entry = &sum->entries[0];
2818
2819 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2820 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2821
2822 /* readahead node pages */
2823 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2824
2825 for (idx = addr; idx < addr + nrpages; idx++) {
2826 struct page *page = f2fs_get_tmp_page(sbi, idx);
2827
2828 if (IS_ERR(page))
2829 return PTR_ERR(page);
2830
2831 rn = F2FS_NODE(page);
2832 sum_entry->nid = rn->footer.nid;
2833 sum_entry->version = 0;
2834 sum_entry->ofs_in_node = 0;
2835 sum_entry++;
2836 f2fs_put_page(page, 1);
2837 }
2838
2839 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2840 addr + nrpages);
2841 }
2842 return 0;
2843 }
2844
remove_nats_in_journal(struct f2fs_sb_info * sbi)2845 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2846 {
2847 struct f2fs_nm_info *nm_i = NM_I(sbi);
2848 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2849 struct f2fs_journal *journal = curseg->journal;
2850 int i;
2851
2852 down_write(&curseg->journal_rwsem);
2853 for (i = 0; i < nats_in_cursum(journal); i++) {
2854 struct nat_entry *ne;
2855 struct f2fs_nat_entry raw_ne;
2856 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2857
2858 if (f2fs_check_nid_range(sbi, nid))
2859 continue;
2860
2861 raw_ne = nat_in_journal(journal, i);
2862
2863 ne = __lookup_nat_cache(nm_i, nid);
2864 if (!ne) {
2865 ne = __alloc_nat_entry(nid, true);
2866 __init_nat_entry(nm_i, ne, &raw_ne, true);
2867 }
2868
2869 /*
2870 * if a free nat in journal has not been used after last
2871 * checkpoint, we should remove it from available nids,
2872 * since later we will add it again.
2873 */
2874 if (!get_nat_flag(ne, IS_DIRTY) &&
2875 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2876 spin_lock(&nm_i->nid_list_lock);
2877 nm_i->available_nids--;
2878 spin_unlock(&nm_i->nid_list_lock);
2879 }
2880
2881 __set_nat_cache_dirty(nm_i, ne);
2882 }
2883 update_nats_in_cursum(journal, -i);
2884 up_write(&curseg->journal_rwsem);
2885 }
2886
__adjust_nat_entry_set(struct nat_entry_set * nes,struct list_head * head,int max)2887 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2888 struct list_head *head, int max)
2889 {
2890 struct nat_entry_set *cur;
2891
2892 if (nes->entry_cnt >= max)
2893 goto add_out;
2894
2895 list_for_each_entry(cur, head, set_list) {
2896 if (cur->entry_cnt >= nes->entry_cnt) {
2897 list_add(&nes->set_list, cur->set_list.prev);
2898 return;
2899 }
2900 }
2901 add_out:
2902 list_add_tail(&nes->set_list, head);
2903 }
2904
__update_nat_bits(struct f2fs_sb_info * sbi,nid_t start_nid,struct page * page)2905 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2906 struct page *page)
2907 {
2908 struct f2fs_nm_info *nm_i = NM_I(sbi);
2909 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2910 struct f2fs_nat_block *nat_blk = page_address(page);
2911 int valid = 0;
2912 int i = 0;
2913
2914 if (!enabled_nat_bits(sbi, NULL))
2915 return;
2916
2917 if (nat_index == 0) {
2918 valid = 1;
2919 i = 1;
2920 }
2921 for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2922 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2923 valid++;
2924 }
2925 if (valid == 0) {
2926 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2927 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2928 return;
2929 }
2930
2931 __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2932 if (valid == NAT_ENTRY_PER_BLOCK)
2933 __set_bit_le(nat_index, nm_i->full_nat_bits);
2934 else
2935 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2936 }
2937
__flush_nat_entry_set(struct f2fs_sb_info * sbi,struct nat_entry_set * set,struct cp_control * cpc)2938 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2939 struct nat_entry_set *set, struct cp_control *cpc)
2940 {
2941 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2942 struct f2fs_journal *journal = curseg->journal;
2943 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2944 bool to_journal = true;
2945 struct f2fs_nat_block *nat_blk;
2946 struct nat_entry *ne, *cur;
2947 struct page *page = NULL;
2948
2949 /*
2950 * there are two steps to flush nat entries:
2951 * #1, flush nat entries to journal in current hot data summary block.
2952 * #2, flush nat entries to nat page.
2953 */
2954 if (enabled_nat_bits(sbi, cpc) ||
2955 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2956 to_journal = false;
2957
2958 if (to_journal) {
2959 down_write(&curseg->journal_rwsem);
2960 } else {
2961 page = get_next_nat_page(sbi, start_nid);
2962 if (IS_ERR(page))
2963 return PTR_ERR(page);
2964
2965 nat_blk = page_address(page);
2966 f2fs_bug_on(sbi, !nat_blk);
2967 }
2968
2969 /* flush dirty nats in nat entry set */
2970 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2971 struct f2fs_nat_entry *raw_ne;
2972 nid_t nid = nat_get_nid(ne);
2973 int offset;
2974
2975 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2976
2977 if (to_journal) {
2978 offset = f2fs_lookup_journal_in_cursum(journal,
2979 NAT_JOURNAL, nid, 1);
2980 f2fs_bug_on(sbi, offset < 0);
2981 raw_ne = &nat_in_journal(journal, offset);
2982 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2983 } else {
2984 raw_ne = &nat_blk->entries[nid - start_nid];
2985 }
2986 raw_nat_from_node_info(raw_ne, &ne->ni);
2987 nat_reset_flag(ne);
2988 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2989 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2990 add_free_nid(sbi, nid, false, true);
2991 } else {
2992 spin_lock(&NM_I(sbi)->nid_list_lock);
2993 update_free_nid_bitmap(sbi, nid, false, false);
2994 spin_unlock(&NM_I(sbi)->nid_list_lock);
2995 }
2996 }
2997
2998 if (to_journal) {
2999 up_write(&curseg->journal_rwsem);
3000 } else {
3001 __update_nat_bits(sbi, start_nid, page);
3002 f2fs_put_page(page, 1);
3003 }
3004
3005 /* Allow dirty nats by node block allocation in write_begin */
3006 if (!set->entry_cnt) {
3007 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
3008 kmem_cache_free(nat_entry_set_slab, set);
3009 }
3010 return 0;
3011 }
3012
3013 /*
3014 * This function is called during the checkpointing process.
3015 */
f2fs_flush_nat_entries(struct f2fs_sb_info * sbi,struct cp_control * cpc)3016 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3017 {
3018 struct f2fs_nm_info *nm_i = NM_I(sbi);
3019 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3020 struct f2fs_journal *journal = curseg->journal;
3021 struct nat_entry_set *setvec[SETVEC_SIZE];
3022 struct nat_entry_set *set, *tmp;
3023 unsigned int found;
3024 nid_t set_idx = 0;
3025 LIST_HEAD(sets);
3026 int err = 0;
3027
3028 /*
3029 * during unmount, let's flush nat_bits before checking
3030 * nat_cnt[DIRTY_NAT].
3031 */
3032 if (enabled_nat_bits(sbi, cpc)) {
3033 f2fs_down_write(&nm_i->nat_tree_lock);
3034 remove_nats_in_journal(sbi);
3035 f2fs_up_write(&nm_i->nat_tree_lock);
3036 }
3037
3038 if (!nm_i->nat_cnt[DIRTY_NAT])
3039 return 0;
3040
3041 f2fs_down_write(&nm_i->nat_tree_lock);
3042
3043 /*
3044 * if there are no enough space in journal to store dirty nat
3045 * entries, remove all entries from journal and merge them
3046 * into nat entry set.
3047 */
3048 if (enabled_nat_bits(sbi, cpc) ||
3049 !__has_cursum_space(journal,
3050 nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3051 remove_nats_in_journal(sbi);
3052
3053 while ((found = __gang_lookup_nat_set(nm_i,
3054 set_idx, SETVEC_SIZE, setvec))) {
3055 unsigned idx;
3056
3057 set_idx = setvec[found - 1]->set + 1;
3058 for (idx = 0; idx < found; idx++)
3059 __adjust_nat_entry_set(setvec[idx], &sets,
3060 MAX_NAT_JENTRIES(journal));
3061 }
3062
3063 /* flush dirty nats in nat entry set */
3064 list_for_each_entry_safe(set, tmp, &sets, set_list) {
3065 err = __flush_nat_entry_set(sbi, set, cpc);
3066 if (err)
3067 break;
3068 }
3069
3070 f2fs_up_write(&nm_i->nat_tree_lock);
3071 /* Allow dirty nats by node block allocation in write_begin */
3072
3073 return err;
3074 }
3075
__get_nat_bitmaps(struct f2fs_sb_info * sbi)3076 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3077 {
3078 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3079 struct f2fs_nm_info *nm_i = NM_I(sbi);
3080 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3081 unsigned int i;
3082 __u64 cp_ver = cur_cp_version(ckpt);
3083 block_t nat_bits_addr;
3084
3085 if (!enabled_nat_bits(sbi, NULL))
3086 return 0;
3087
3088 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3089 nm_i->nat_bits = f2fs_kvzalloc(sbi,
3090 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3091 if (!nm_i->nat_bits)
3092 return -ENOMEM;
3093
3094 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3095 nm_i->nat_bits_blocks;
3096 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3097 struct page *page;
3098
3099 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3100 if (IS_ERR(page))
3101 return PTR_ERR(page);
3102
3103 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3104 page_address(page), F2FS_BLKSIZE);
3105 f2fs_put_page(page, 1);
3106 }
3107
3108 cp_ver |= (cur_cp_crc(ckpt) << 32);
3109 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3110 disable_nat_bits(sbi, true);
3111 return 0;
3112 }
3113
3114 nm_i->full_nat_bits = nm_i->nat_bits + 8;
3115 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3116
3117 f2fs_notice(sbi, "Found nat_bits in checkpoint");
3118 return 0;
3119 }
3120
load_free_nid_bitmap(struct f2fs_sb_info * sbi)3121 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3122 {
3123 struct f2fs_nm_info *nm_i = NM_I(sbi);
3124 unsigned int i = 0;
3125 nid_t nid, last_nid;
3126
3127 if (!enabled_nat_bits(sbi, NULL))
3128 return;
3129
3130 for (i = 0; i < nm_i->nat_blocks; i++) {
3131 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3132 if (i >= nm_i->nat_blocks)
3133 break;
3134
3135 __set_bit_le(i, nm_i->nat_block_bitmap);
3136
3137 nid = i * NAT_ENTRY_PER_BLOCK;
3138 last_nid = nid + NAT_ENTRY_PER_BLOCK;
3139
3140 spin_lock(&NM_I(sbi)->nid_list_lock);
3141 for (; nid < last_nid; nid++)
3142 update_free_nid_bitmap(sbi, nid, true, true);
3143 spin_unlock(&NM_I(sbi)->nid_list_lock);
3144 }
3145
3146 for (i = 0; i < nm_i->nat_blocks; i++) {
3147 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3148 if (i >= nm_i->nat_blocks)
3149 break;
3150
3151 __set_bit_le(i, nm_i->nat_block_bitmap);
3152 }
3153 }
3154
init_node_manager(struct f2fs_sb_info * sbi)3155 static int init_node_manager(struct f2fs_sb_info *sbi)
3156 {
3157 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3158 struct f2fs_nm_info *nm_i = NM_I(sbi);
3159 unsigned char *version_bitmap;
3160 unsigned int nat_segs;
3161 int err;
3162
3163 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3164
3165 /* segment_count_nat includes pair segment so divide to 2. */
3166 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3167 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3168 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3169
3170 /* not used nids: 0, node, meta, (and root counted as valid node) */
3171 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3172 F2FS_RESERVED_NODE_NUM;
3173 nm_i->nid_cnt[FREE_NID] = 0;
3174 nm_i->nid_cnt[PREALLOC_NID] = 0;
3175 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3176 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3177 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3178
3179 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3180 INIT_LIST_HEAD(&nm_i->free_nid_list);
3181 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3182 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3183 INIT_LIST_HEAD(&nm_i->nat_entries);
3184 spin_lock_init(&nm_i->nat_list_lock);
3185
3186 mutex_init(&nm_i->build_lock);
3187 spin_lock_init(&nm_i->nid_list_lock);
3188 init_f2fs_rwsem(&nm_i->nat_tree_lock);
3189
3190 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3191 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3192 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3193 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3194 GFP_KERNEL);
3195 if (!nm_i->nat_bitmap)
3196 return -ENOMEM;
3197
3198 err = __get_nat_bitmaps(sbi);
3199 if (err)
3200 return err;
3201
3202 #ifdef CONFIG_F2FS_CHECK_FS
3203 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3204 GFP_KERNEL);
3205 if (!nm_i->nat_bitmap_mir)
3206 return -ENOMEM;
3207 #endif
3208
3209 return 0;
3210 }
3211
init_free_nid_cache(struct f2fs_sb_info * sbi)3212 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3213 {
3214 struct f2fs_nm_info *nm_i = NM_I(sbi);
3215 int i;
3216
3217 nm_i->free_nid_bitmap =
3218 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3219 nm_i->nat_blocks),
3220 GFP_KERNEL);
3221 if (!nm_i->free_nid_bitmap)
3222 return -ENOMEM;
3223
3224 for (i = 0; i < nm_i->nat_blocks; i++) {
3225 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3226 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3227 if (!nm_i->free_nid_bitmap[i])
3228 return -ENOMEM;
3229 }
3230
3231 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3232 GFP_KERNEL);
3233 if (!nm_i->nat_block_bitmap)
3234 return -ENOMEM;
3235
3236 nm_i->free_nid_count =
3237 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3238 nm_i->nat_blocks),
3239 GFP_KERNEL);
3240 if (!nm_i->free_nid_count)
3241 return -ENOMEM;
3242 return 0;
3243 }
3244
f2fs_build_node_manager(struct f2fs_sb_info * sbi)3245 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3246 {
3247 int err;
3248
3249 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3250 GFP_KERNEL);
3251 if (!sbi->nm_info)
3252 return -ENOMEM;
3253
3254 err = init_node_manager(sbi);
3255 if (err)
3256 return err;
3257
3258 err = init_free_nid_cache(sbi);
3259 if (err)
3260 return err;
3261
3262 /* load free nid status from nat_bits table */
3263 load_free_nid_bitmap(sbi);
3264
3265 return f2fs_build_free_nids(sbi, true, true);
3266 }
3267
f2fs_destroy_node_manager(struct f2fs_sb_info * sbi)3268 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3269 {
3270 struct f2fs_nm_info *nm_i = NM_I(sbi);
3271 struct free_nid *i, *next_i;
3272 struct nat_entry *natvec[NATVEC_SIZE];
3273 struct nat_entry_set *setvec[SETVEC_SIZE];
3274 nid_t nid = 0;
3275 unsigned int found;
3276
3277 if (!nm_i)
3278 return;
3279
3280 /* destroy free nid list */
3281 spin_lock(&nm_i->nid_list_lock);
3282 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3283 __remove_free_nid(sbi, i, FREE_NID);
3284 spin_unlock(&nm_i->nid_list_lock);
3285 kmem_cache_free(free_nid_slab, i);
3286 spin_lock(&nm_i->nid_list_lock);
3287 }
3288 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3289 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3290 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3291 spin_unlock(&nm_i->nid_list_lock);
3292
3293 /* destroy nat cache */
3294 f2fs_down_write(&nm_i->nat_tree_lock);
3295 while ((found = __gang_lookup_nat_cache(nm_i,
3296 nid, NATVEC_SIZE, natvec))) {
3297 unsigned idx;
3298
3299 nid = nat_get_nid(natvec[found - 1]) + 1;
3300 for (idx = 0; idx < found; idx++) {
3301 spin_lock(&nm_i->nat_list_lock);
3302 list_del(&natvec[idx]->list);
3303 spin_unlock(&nm_i->nat_list_lock);
3304
3305 __del_from_nat_cache(nm_i, natvec[idx]);
3306 }
3307 }
3308 f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3309
3310 /* destroy nat set cache */
3311 nid = 0;
3312 while ((found = __gang_lookup_nat_set(nm_i,
3313 nid, SETVEC_SIZE, setvec))) {
3314 unsigned idx;
3315
3316 nid = setvec[found - 1]->set + 1;
3317 for (idx = 0; idx < found; idx++) {
3318 /* entry_cnt is not zero, when cp_error was occurred */
3319 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3320 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3321 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3322 }
3323 }
3324 f2fs_up_write(&nm_i->nat_tree_lock);
3325
3326 kvfree(nm_i->nat_block_bitmap);
3327 if (nm_i->free_nid_bitmap) {
3328 int i;
3329
3330 for (i = 0; i < nm_i->nat_blocks; i++)
3331 kvfree(nm_i->free_nid_bitmap[i]);
3332 kvfree(nm_i->free_nid_bitmap);
3333 }
3334 kvfree(nm_i->free_nid_count);
3335
3336 kvfree(nm_i->nat_bitmap);
3337 kvfree(nm_i->nat_bits);
3338 #ifdef CONFIG_F2FS_CHECK_FS
3339 kvfree(nm_i->nat_bitmap_mir);
3340 #endif
3341 sbi->nm_info = NULL;
3342 kfree(nm_i);
3343 }
3344
f2fs_create_node_manager_caches(void)3345 int __init f2fs_create_node_manager_caches(void)
3346 {
3347 nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3348 sizeof(struct nat_entry));
3349 if (!nat_entry_slab)
3350 goto fail;
3351
3352 free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3353 sizeof(struct free_nid));
3354 if (!free_nid_slab)
3355 goto destroy_nat_entry;
3356
3357 nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3358 sizeof(struct nat_entry_set));
3359 if (!nat_entry_set_slab)
3360 goto destroy_free_nid;
3361
3362 fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3363 sizeof(struct fsync_node_entry));
3364 if (!fsync_node_entry_slab)
3365 goto destroy_nat_entry_set;
3366 return 0;
3367
3368 destroy_nat_entry_set:
3369 kmem_cache_destroy(nat_entry_set_slab);
3370 destroy_free_nid:
3371 kmem_cache_destroy(free_nid_slab);
3372 destroy_nat_entry:
3373 kmem_cache_destroy(nat_entry_slab);
3374 fail:
3375 return -ENOMEM;
3376 }
3377
f2fs_destroy_node_manager_caches(void)3378 void f2fs_destroy_node_manager_caches(void)
3379 {
3380 kmem_cache_destroy(fsync_node_entry_slab);
3381 kmem_cache_destroy(nat_entry_set_slab);
3382 kmem_cache_destroy(free_nid_slab);
3383 kmem_cache_destroy(nat_entry_slab);
3384 }
3385