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/sched/mm.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 "iostat.h"
21 #include <trace/events/f2fs.h>
22
23 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
24
25 static struct kmem_cache *nat_entry_slab;
26 static struct kmem_cache *free_nid_slab;
27 static struct kmem_cache *nat_entry_set_slab;
28 static struct kmem_cache *fsync_node_entry_slab;
29
30 /*
31 * Check whether the given nid is within node id range.
32 */
f2fs_check_nid_range(struct f2fs_sb_info * sbi,nid_t nid)33 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
34 {
35 if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
36 set_sbi_flag(sbi, SBI_NEED_FSCK);
37 f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
38 __func__, nid);
39 f2fs_handle_error(sbi, ERROR_CORRUPTED_INODE);
40 return -EFSCORRUPTED;
41 }
42 return 0;
43 }
44
f2fs_available_free_memory(struct f2fs_sb_info * sbi,int type)45 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
46 {
47 struct f2fs_nm_info *nm_i = NM_I(sbi);
48 struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
49 struct sysinfo val;
50 unsigned long avail_ram;
51 unsigned long mem_size = 0;
52 bool res = false;
53
54 if (!nm_i)
55 return true;
56
57 si_meminfo(&val);
58
59 /* only uses low memory */
60 avail_ram = val.totalram - val.totalhigh;
61
62 /*
63 * give 25%, 25%, 50%, 50%, 25%, 25% memory for each components respectively
64 */
65 if (type == FREE_NIDS) {
66 mem_size = (nm_i->nid_cnt[FREE_NID] *
67 sizeof(struct free_nid)) >> PAGE_SHIFT;
68 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
69 } else if (type == NAT_ENTRIES) {
70 mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
71 sizeof(struct nat_entry)) >> PAGE_SHIFT;
72 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
73 if (excess_cached_nats(sbi))
74 res = false;
75 } else if (type == DIRTY_DENTS) {
76 if (sbi->sb->s_bdi->wb.dirty_exceeded)
77 return false;
78 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
79 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
80 } else if (type == INO_ENTRIES) {
81 int i;
82
83 for (i = 0; i < MAX_INO_ENTRY; i++)
84 mem_size += sbi->im[i].ino_num *
85 sizeof(struct ino_entry);
86 mem_size >>= PAGE_SHIFT;
87 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
88 } else if (type == READ_EXTENT_CACHE || type == AGE_EXTENT_CACHE) {
89 enum extent_type etype = type == READ_EXTENT_CACHE ?
90 EX_READ : EX_BLOCK_AGE;
91 struct extent_tree_info *eti = &sbi->extent_tree[etype];
92
93 mem_size = (atomic_read(&eti->total_ext_tree) *
94 sizeof(struct extent_tree) +
95 atomic_read(&eti->total_ext_node) *
96 sizeof(struct extent_node)) >> PAGE_SHIFT;
97 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
98 } else if (type == DISCARD_CACHE) {
99 mem_size = (atomic_read(&dcc->discard_cmd_cnt) *
100 sizeof(struct discard_cmd)) >> PAGE_SHIFT;
101 res = mem_size < (avail_ram * nm_i->ram_thresh / 100);
102 } else if (type == COMPRESS_PAGE) {
103 #ifdef CONFIG_F2FS_FS_COMPRESSION
104 unsigned long free_ram = val.freeram;
105
106 /*
107 * free memory is lower than watermark or cached page count
108 * exceed threshold, deny caching compress page.
109 */
110 res = (free_ram > avail_ram * sbi->compress_watermark / 100) &&
111 (COMPRESS_MAPPING(sbi)->nrpages <
112 free_ram * sbi->compress_percent / 100);
113 #else
114 res = false;
115 #endif
116 } else {
117 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
118 return true;
119 }
120 return res;
121 }
122
clear_node_page_dirty(struct page * page)123 static void clear_node_page_dirty(struct page *page)
124 {
125 if (PageDirty(page)) {
126 f2fs_clear_page_cache_dirty_tag(page);
127 clear_page_dirty_for_io(page);
128 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
129 }
130 ClearPageUptodate(page);
131 }
132
get_current_nat_page(struct f2fs_sb_info * sbi,nid_t nid)133 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
134 {
135 return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
136 }
137
get_next_nat_page(struct f2fs_sb_info * sbi,nid_t nid)138 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
139 {
140 struct page *src_page;
141 struct page *dst_page;
142 pgoff_t dst_off;
143 void *src_addr;
144 void *dst_addr;
145 struct f2fs_nm_info *nm_i = NM_I(sbi);
146
147 dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
148
149 /* get current nat block page with lock */
150 src_page = get_current_nat_page(sbi, nid);
151 if (IS_ERR(src_page))
152 return src_page;
153 dst_page = f2fs_grab_meta_page(sbi, dst_off);
154 f2fs_bug_on(sbi, PageDirty(src_page));
155
156 src_addr = page_address(src_page);
157 dst_addr = page_address(dst_page);
158 memcpy(dst_addr, src_addr, PAGE_SIZE);
159 set_page_dirty(dst_page);
160 f2fs_put_page(src_page, 1);
161
162 set_to_next_nat(nm_i, nid);
163
164 return dst_page;
165 }
166
__alloc_nat_entry(struct f2fs_sb_info * sbi,nid_t nid,bool no_fail)167 static struct nat_entry *__alloc_nat_entry(struct f2fs_sb_info *sbi,
168 nid_t nid, bool no_fail)
169 {
170 struct nat_entry *new;
171
172 new = f2fs_kmem_cache_alloc(nat_entry_slab,
173 GFP_F2FS_ZERO, no_fail, sbi);
174 if (new) {
175 nat_set_nid(new, nid);
176 nat_reset_flag(new);
177 }
178 return new;
179 }
180
__free_nat_entry(struct nat_entry * e)181 static void __free_nat_entry(struct nat_entry *e)
182 {
183 kmem_cache_free(nat_entry_slab, e);
184 }
185
186 /* 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)187 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
188 struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
189 {
190 if (no_fail)
191 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
192 else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
193 return NULL;
194
195 if (raw_ne)
196 node_info_from_raw_nat(&ne->ni, raw_ne);
197
198 spin_lock(&nm_i->nat_list_lock);
199 list_add_tail(&ne->list, &nm_i->nat_entries);
200 spin_unlock(&nm_i->nat_list_lock);
201
202 nm_i->nat_cnt[TOTAL_NAT]++;
203 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
204 return ne;
205 }
206
__lookup_nat_cache(struct f2fs_nm_info * nm_i,nid_t n)207 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
208 {
209 struct nat_entry *ne;
210
211 ne = radix_tree_lookup(&nm_i->nat_root, n);
212
213 /* for recent accessed nat entry, move it to tail of lru list */
214 if (ne && !get_nat_flag(ne, IS_DIRTY)) {
215 spin_lock(&nm_i->nat_list_lock);
216 if (!list_empty(&ne->list))
217 list_move_tail(&ne->list, &nm_i->nat_entries);
218 spin_unlock(&nm_i->nat_list_lock);
219 }
220
221 return ne;
222 }
223
__gang_lookup_nat_cache(struct f2fs_nm_info * nm_i,nid_t start,unsigned int nr,struct nat_entry ** ep)224 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
225 nid_t start, unsigned int nr, struct nat_entry **ep)
226 {
227 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
228 }
229
__del_from_nat_cache(struct f2fs_nm_info * nm_i,struct nat_entry * e)230 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
231 {
232 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
233 nm_i->nat_cnt[TOTAL_NAT]--;
234 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
235 __free_nat_entry(e);
236 }
237
__grab_nat_entry_set(struct f2fs_nm_info * nm_i,struct nat_entry * ne)238 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
239 struct nat_entry *ne)
240 {
241 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
242 struct nat_entry_set *head;
243
244 head = radix_tree_lookup(&nm_i->nat_set_root, set);
245 if (!head) {
246 head = f2fs_kmem_cache_alloc(nat_entry_set_slab,
247 GFP_NOFS, true, NULL);
248
249 INIT_LIST_HEAD(&head->entry_list);
250 INIT_LIST_HEAD(&head->set_list);
251 head->set = set;
252 head->entry_cnt = 0;
253 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
254 }
255 return head;
256 }
257
__set_nat_cache_dirty(struct f2fs_nm_info * nm_i,struct nat_entry * ne)258 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
259 struct nat_entry *ne)
260 {
261 struct nat_entry_set *head;
262 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
263
264 if (!new_ne)
265 head = __grab_nat_entry_set(nm_i, ne);
266
267 /*
268 * update entry_cnt in below condition:
269 * 1. update NEW_ADDR to valid block address;
270 * 2. update old block address to new one;
271 */
272 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
273 !get_nat_flag(ne, IS_DIRTY)))
274 head->entry_cnt++;
275
276 set_nat_flag(ne, IS_PREALLOC, new_ne);
277
278 if (get_nat_flag(ne, IS_DIRTY))
279 goto refresh_list;
280
281 nm_i->nat_cnt[DIRTY_NAT]++;
282 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
283 set_nat_flag(ne, IS_DIRTY, true);
284 refresh_list:
285 spin_lock(&nm_i->nat_list_lock);
286 if (new_ne)
287 list_del_init(&ne->list);
288 else
289 list_move_tail(&ne->list, &head->entry_list);
290 spin_unlock(&nm_i->nat_list_lock);
291 }
292
__clear_nat_cache_dirty(struct f2fs_nm_info * nm_i,struct nat_entry_set * set,struct nat_entry * ne)293 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
294 struct nat_entry_set *set, struct nat_entry *ne)
295 {
296 spin_lock(&nm_i->nat_list_lock);
297 list_move_tail(&ne->list, &nm_i->nat_entries);
298 spin_unlock(&nm_i->nat_list_lock);
299
300 set_nat_flag(ne, IS_DIRTY, false);
301 set->entry_cnt--;
302 nm_i->nat_cnt[DIRTY_NAT]--;
303 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
304 }
305
__gang_lookup_nat_set(struct f2fs_nm_info * nm_i,nid_t start,unsigned int nr,struct nat_entry_set ** ep)306 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
307 nid_t start, unsigned int nr, struct nat_entry_set **ep)
308 {
309 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
310 start, nr);
311 }
312
f2fs_in_warm_node_list(struct f2fs_sb_info * sbi,struct page * page)313 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
314 {
315 return NODE_MAPPING(sbi) == page->mapping &&
316 IS_DNODE(page) && is_cold_node(page);
317 }
318
f2fs_init_fsync_node_info(struct f2fs_sb_info * sbi)319 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
320 {
321 spin_lock_init(&sbi->fsync_node_lock);
322 INIT_LIST_HEAD(&sbi->fsync_node_list);
323 sbi->fsync_seg_id = 0;
324 sbi->fsync_node_num = 0;
325 }
326
f2fs_add_fsync_node_entry(struct f2fs_sb_info * sbi,struct page * page)327 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
328 struct page *page)
329 {
330 struct fsync_node_entry *fn;
331 unsigned long flags;
332 unsigned int seq_id;
333
334 fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab,
335 GFP_NOFS, true, NULL);
336
337 get_page(page);
338 fn->page = page;
339 INIT_LIST_HEAD(&fn->list);
340
341 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
342 list_add_tail(&fn->list, &sbi->fsync_node_list);
343 fn->seq_id = sbi->fsync_seg_id++;
344 seq_id = fn->seq_id;
345 sbi->fsync_node_num++;
346 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
347
348 return seq_id;
349 }
350
f2fs_del_fsync_node_entry(struct f2fs_sb_info * sbi,struct page * page)351 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
352 {
353 struct fsync_node_entry *fn;
354 unsigned long flags;
355
356 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
357 list_for_each_entry(fn, &sbi->fsync_node_list, list) {
358 if (fn->page == page) {
359 list_del(&fn->list);
360 sbi->fsync_node_num--;
361 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
362 kmem_cache_free(fsync_node_entry_slab, fn);
363 put_page(page);
364 return;
365 }
366 }
367 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
368 f2fs_bug_on(sbi, 1);
369 }
370
f2fs_reset_fsync_node_info(struct f2fs_sb_info * sbi)371 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
372 {
373 unsigned long flags;
374
375 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
376 sbi->fsync_seg_id = 0;
377 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
378 }
379
f2fs_need_dentry_mark(struct f2fs_sb_info * sbi,nid_t nid)380 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
381 {
382 struct f2fs_nm_info *nm_i = NM_I(sbi);
383 struct nat_entry *e;
384 bool need = false;
385
386 f2fs_down_read(&nm_i->nat_tree_lock);
387 e = __lookup_nat_cache(nm_i, nid);
388 if (e) {
389 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
390 !get_nat_flag(e, HAS_FSYNCED_INODE))
391 need = true;
392 }
393 f2fs_up_read(&nm_i->nat_tree_lock);
394 return need;
395 }
396
f2fs_is_checkpointed_node(struct f2fs_sb_info * sbi,nid_t nid)397 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
398 {
399 struct f2fs_nm_info *nm_i = NM_I(sbi);
400 struct nat_entry *e;
401 bool is_cp = true;
402
403 f2fs_down_read(&nm_i->nat_tree_lock);
404 e = __lookup_nat_cache(nm_i, nid);
405 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
406 is_cp = false;
407 f2fs_up_read(&nm_i->nat_tree_lock);
408 return is_cp;
409 }
410
f2fs_need_inode_block_update(struct f2fs_sb_info * sbi,nid_t ino)411 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
412 {
413 struct f2fs_nm_info *nm_i = NM_I(sbi);
414 struct nat_entry *e;
415 bool need_update = true;
416
417 f2fs_down_read(&nm_i->nat_tree_lock);
418 e = __lookup_nat_cache(nm_i, ino);
419 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
420 (get_nat_flag(e, IS_CHECKPOINTED) ||
421 get_nat_flag(e, HAS_FSYNCED_INODE)))
422 need_update = false;
423 f2fs_up_read(&nm_i->nat_tree_lock);
424 return need_update;
425 }
426
427 /* must be locked by nat_tree_lock */
cache_nat_entry(struct f2fs_sb_info * sbi,nid_t nid,struct f2fs_nat_entry * ne)428 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
429 struct f2fs_nat_entry *ne)
430 {
431 struct f2fs_nm_info *nm_i = NM_I(sbi);
432 struct nat_entry *new, *e;
433
434 /* Let's mitigate lock contention of nat_tree_lock during checkpoint */
435 if (f2fs_rwsem_is_locked(&sbi->cp_global_sem))
436 return;
437
438 new = __alloc_nat_entry(sbi, nid, false);
439 if (!new)
440 return;
441
442 f2fs_down_write(&nm_i->nat_tree_lock);
443 e = __lookup_nat_cache(nm_i, nid);
444 if (!e)
445 e = __init_nat_entry(nm_i, new, ne, false);
446 else
447 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
448 nat_get_blkaddr(e) !=
449 le32_to_cpu(ne->block_addr) ||
450 nat_get_version(e) != ne->version);
451 f2fs_up_write(&nm_i->nat_tree_lock);
452 if (e != new)
453 __free_nat_entry(new);
454 }
455
set_node_addr(struct f2fs_sb_info * sbi,struct node_info * ni,block_t new_blkaddr,bool fsync_done)456 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
457 block_t new_blkaddr, bool fsync_done)
458 {
459 struct f2fs_nm_info *nm_i = NM_I(sbi);
460 struct nat_entry *e;
461 struct nat_entry *new = __alloc_nat_entry(sbi, ni->nid, true);
462
463 f2fs_down_write(&nm_i->nat_tree_lock);
464 e = __lookup_nat_cache(nm_i, ni->nid);
465 if (!e) {
466 e = __init_nat_entry(nm_i, new, NULL, true);
467 copy_node_info(&e->ni, ni);
468 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
469 } else if (new_blkaddr == NEW_ADDR) {
470 /*
471 * when nid is reallocated,
472 * previous nat entry can be remained in nat cache.
473 * So, reinitialize it with new information.
474 */
475 copy_node_info(&e->ni, ni);
476 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
477 }
478 /* let's free early to reduce memory consumption */
479 if (e != new)
480 __free_nat_entry(new);
481
482 /* sanity check */
483 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
484 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
485 new_blkaddr == NULL_ADDR);
486 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
487 new_blkaddr == NEW_ADDR);
488 f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
489 new_blkaddr == NEW_ADDR);
490
491 /* increment version no as node is removed */
492 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
493 unsigned char version = nat_get_version(e);
494
495 nat_set_version(e, inc_node_version(version));
496 }
497
498 /* change address */
499 nat_set_blkaddr(e, new_blkaddr);
500 if (!__is_valid_data_blkaddr(new_blkaddr))
501 set_nat_flag(e, IS_CHECKPOINTED, false);
502 __set_nat_cache_dirty(nm_i, e);
503
504 /* update fsync_mark if its inode nat entry is still alive */
505 if (ni->nid != ni->ino)
506 e = __lookup_nat_cache(nm_i, ni->ino);
507 if (e) {
508 if (fsync_done && ni->nid == ni->ino)
509 set_nat_flag(e, HAS_FSYNCED_INODE, true);
510 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
511 }
512 f2fs_up_write(&nm_i->nat_tree_lock);
513 }
514
f2fs_try_to_free_nats(struct f2fs_sb_info * sbi,int nr_shrink)515 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
516 {
517 struct f2fs_nm_info *nm_i = NM_I(sbi);
518 int nr = nr_shrink;
519
520 if (!f2fs_down_write_trylock(&nm_i->nat_tree_lock))
521 return 0;
522
523 spin_lock(&nm_i->nat_list_lock);
524 while (nr_shrink) {
525 struct nat_entry *ne;
526
527 if (list_empty(&nm_i->nat_entries))
528 break;
529
530 ne = list_first_entry(&nm_i->nat_entries,
531 struct nat_entry, list);
532 list_del(&ne->list);
533 spin_unlock(&nm_i->nat_list_lock);
534
535 __del_from_nat_cache(nm_i, ne);
536 nr_shrink--;
537
538 spin_lock(&nm_i->nat_list_lock);
539 }
540 spin_unlock(&nm_i->nat_list_lock);
541
542 f2fs_up_write(&nm_i->nat_tree_lock);
543 return nr - nr_shrink;
544 }
545
f2fs_get_node_info(struct f2fs_sb_info * sbi,nid_t nid,struct node_info * ni,bool checkpoint_context)546 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
547 struct node_info *ni, bool checkpoint_context)
548 {
549 struct f2fs_nm_info *nm_i = NM_I(sbi);
550 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
551 struct f2fs_journal *journal = curseg->journal;
552 nid_t start_nid = START_NID(nid);
553 struct f2fs_nat_block *nat_blk;
554 struct page *page = NULL;
555 struct f2fs_nat_entry ne;
556 struct nat_entry *e;
557 pgoff_t index;
558 block_t blkaddr;
559 int i;
560
561 ni->nid = nid;
562 retry:
563 /* Check nat cache */
564 f2fs_down_read(&nm_i->nat_tree_lock);
565 e = __lookup_nat_cache(nm_i, nid);
566 if (e) {
567 ni->ino = nat_get_ino(e);
568 ni->blk_addr = nat_get_blkaddr(e);
569 ni->version = nat_get_version(e);
570 f2fs_up_read(&nm_i->nat_tree_lock);
571 return 0;
572 }
573
574 /*
575 * Check current segment summary by trying to grab journal_rwsem first.
576 * This sem is on the critical path on the checkpoint requiring the above
577 * nat_tree_lock. Therefore, we should retry, if we failed to grab here
578 * while not bothering checkpoint.
579 */
580 if (!f2fs_rwsem_is_locked(&sbi->cp_global_sem) || checkpoint_context) {
581 down_read(&curseg->journal_rwsem);
582 } else if (f2fs_rwsem_is_contended(&nm_i->nat_tree_lock) ||
583 !down_read_trylock(&curseg->journal_rwsem)) {
584 f2fs_up_read(&nm_i->nat_tree_lock);
585 goto retry;
586 }
587
588 i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
589 if (i >= 0) {
590 ne = nat_in_journal(journal, i);
591 node_info_from_raw_nat(ni, &ne);
592 }
593 up_read(&curseg->journal_rwsem);
594 if (i >= 0) {
595 f2fs_up_read(&nm_i->nat_tree_lock);
596 goto cache;
597 }
598
599 /* Fill node_info from nat page */
600 index = current_nat_addr(sbi, nid);
601 f2fs_up_read(&nm_i->nat_tree_lock);
602
603 page = f2fs_get_meta_page(sbi, index);
604 if (IS_ERR(page))
605 return PTR_ERR(page);
606
607 nat_blk = (struct f2fs_nat_block *)page_address(page);
608 ne = nat_blk->entries[nid - start_nid];
609 node_info_from_raw_nat(ni, &ne);
610 f2fs_put_page(page, 1);
611 cache:
612 blkaddr = le32_to_cpu(ne.block_addr);
613 if (__is_valid_data_blkaddr(blkaddr) &&
614 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
615 return -EFAULT;
616
617 /* cache nat entry */
618 cache_nat_entry(sbi, nid, &ne);
619 return 0;
620 }
621
622 /*
623 * readahead MAX_RA_NODE number of node pages.
624 */
f2fs_ra_node_pages(struct page * parent,int start,int n)625 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
626 {
627 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
628 struct blk_plug plug;
629 int i, end;
630 nid_t nid;
631
632 blk_start_plug(&plug);
633
634 /* Then, try readahead for siblings of the desired node */
635 end = start + n;
636 end = min(end, NIDS_PER_BLOCK);
637 for (i = start; i < end; i++) {
638 nid = get_nid(parent, i, false);
639 f2fs_ra_node_page(sbi, nid);
640 }
641
642 blk_finish_plug(&plug);
643 }
644
f2fs_get_next_page_offset(struct dnode_of_data * dn,pgoff_t pgofs)645 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
646 {
647 const long direct_index = ADDRS_PER_INODE(dn->inode);
648 const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
649 const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
650 unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
651 int cur_level = dn->cur_level;
652 int max_level = dn->max_level;
653 pgoff_t base = 0;
654
655 if (!dn->max_level)
656 return pgofs + 1;
657
658 while (max_level-- > cur_level)
659 skipped_unit *= NIDS_PER_BLOCK;
660
661 switch (dn->max_level) {
662 case 3:
663 base += 2 * indirect_blks;
664 fallthrough;
665 case 2:
666 base += 2 * direct_blks;
667 fallthrough;
668 case 1:
669 base += direct_index;
670 break;
671 default:
672 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
673 }
674
675 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
676 }
677
678 /*
679 * The maximum depth is four.
680 * Offset[0] will have raw inode offset.
681 */
get_node_path(struct inode * inode,long block,int offset[4],unsigned int noffset[4])682 static int get_node_path(struct inode *inode, long block,
683 int offset[4], unsigned int noffset[4])
684 {
685 const long direct_index = ADDRS_PER_INODE(inode);
686 const long direct_blks = ADDRS_PER_BLOCK(inode);
687 const long dptrs_per_blk = NIDS_PER_BLOCK;
688 const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
689 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
690 int n = 0;
691 int level = 0;
692
693 noffset[0] = 0;
694
695 if (block < direct_index) {
696 offset[n] = block;
697 goto got;
698 }
699 block -= direct_index;
700 if (block < direct_blks) {
701 offset[n++] = NODE_DIR1_BLOCK;
702 noffset[n] = 1;
703 offset[n] = block;
704 level = 1;
705 goto got;
706 }
707 block -= direct_blks;
708 if (block < direct_blks) {
709 offset[n++] = NODE_DIR2_BLOCK;
710 noffset[n] = 2;
711 offset[n] = block;
712 level = 1;
713 goto got;
714 }
715 block -= direct_blks;
716 if (block < indirect_blks) {
717 offset[n++] = NODE_IND1_BLOCK;
718 noffset[n] = 3;
719 offset[n++] = block / direct_blks;
720 noffset[n] = 4 + offset[n - 1];
721 offset[n] = block % direct_blks;
722 level = 2;
723 goto got;
724 }
725 block -= indirect_blks;
726 if (block < indirect_blks) {
727 offset[n++] = NODE_IND2_BLOCK;
728 noffset[n] = 4 + dptrs_per_blk;
729 offset[n++] = block / direct_blks;
730 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
731 offset[n] = block % direct_blks;
732 level = 2;
733 goto got;
734 }
735 block -= indirect_blks;
736 if (block < dindirect_blks) {
737 offset[n++] = NODE_DIND_BLOCK;
738 noffset[n] = 5 + (dptrs_per_blk * 2);
739 offset[n++] = block / indirect_blks;
740 noffset[n] = 6 + (dptrs_per_blk * 2) +
741 offset[n - 1] * (dptrs_per_blk + 1);
742 offset[n++] = (block / direct_blks) % dptrs_per_blk;
743 noffset[n] = 7 + (dptrs_per_blk * 2) +
744 offset[n - 2] * (dptrs_per_blk + 1) +
745 offset[n - 1];
746 offset[n] = block % direct_blks;
747 level = 3;
748 goto got;
749 } else {
750 return -E2BIG;
751 }
752 got:
753 return level;
754 }
755
756 /*
757 * Caller should call f2fs_put_dnode(dn).
758 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
759 * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
760 */
f2fs_get_dnode_of_data(struct dnode_of_data * dn,pgoff_t index,int mode)761 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
762 {
763 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
764 struct page *npage[4];
765 struct page *parent = NULL;
766 int offset[4];
767 unsigned int noffset[4];
768 nid_t nids[4];
769 int level, i = 0;
770 int err = 0;
771
772 level = get_node_path(dn->inode, index, offset, noffset);
773 if (level < 0)
774 return level;
775
776 nids[0] = dn->inode->i_ino;
777 npage[0] = dn->inode_page;
778
779 if (!npage[0]) {
780 npage[0] = f2fs_get_node_page(sbi, nids[0]);
781 if (IS_ERR(npage[0]))
782 return PTR_ERR(npage[0]);
783 }
784
785 /* if inline_data is set, should not report any block indices */
786 if (f2fs_has_inline_data(dn->inode) && index) {
787 err = -ENOENT;
788 f2fs_put_page(npage[0], 1);
789 goto release_out;
790 }
791
792 parent = npage[0];
793 if (level != 0)
794 nids[1] = get_nid(parent, offset[0], true);
795 dn->inode_page = npage[0];
796 dn->inode_page_locked = true;
797
798 /* get indirect or direct nodes */
799 for (i = 1; i <= level; i++) {
800 bool done = false;
801
802 if (!nids[i] && mode == ALLOC_NODE) {
803 /* alloc new node */
804 if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
805 err = -ENOSPC;
806 goto release_pages;
807 }
808
809 dn->nid = nids[i];
810 npage[i] = f2fs_new_node_page(dn, noffset[i]);
811 if (IS_ERR(npage[i])) {
812 f2fs_alloc_nid_failed(sbi, nids[i]);
813 err = PTR_ERR(npage[i]);
814 goto release_pages;
815 }
816
817 set_nid(parent, offset[i - 1], nids[i], i == 1);
818 f2fs_alloc_nid_done(sbi, nids[i]);
819 done = true;
820 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
821 npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
822 if (IS_ERR(npage[i])) {
823 err = PTR_ERR(npage[i]);
824 goto release_pages;
825 }
826 done = true;
827 }
828 if (i == 1) {
829 dn->inode_page_locked = false;
830 unlock_page(parent);
831 } else {
832 f2fs_put_page(parent, 1);
833 }
834
835 if (!done) {
836 npage[i] = f2fs_get_node_page(sbi, nids[i]);
837 if (IS_ERR(npage[i])) {
838 err = PTR_ERR(npage[i]);
839 f2fs_put_page(npage[0], 0);
840 goto release_out;
841 }
842 }
843 if (i < level) {
844 parent = npage[i];
845 nids[i + 1] = get_nid(parent, offset[i], false);
846 }
847 }
848 dn->nid = nids[level];
849 dn->ofs_in_node = offset[level];
850 dn->node_page = npage[level];
851 dn->data_blkaddr = f2fs_data_blkaddr(dn);
852
853 if (is_inode_flag_set(dn->inode, FI_COMPRESSED_FILE) &&
854 f2fs_sb_has_readonly(sbi)) {
855 unsigned int cluster_size = F2FS_I(dn->inode)->i_cluster_size;
856 unsigned int ofs_in_node = dn->ofs_in_node;
857 pgoff_t fofs = index;
858 unsigned int c_len;
859 block_t blkaddr;
860
861 /* should align fofs and ofs_in_node to cluster_size */
862 if (fofs % cluster_size) {
863 fofs = round_down(fofs, cluster_size);
864 ofs_in_node = round_down(ofs_in_node, cluster_size);
865 }
866
867 c_len = f2fs_cluster_blocks_are_contiguous(dn, ofs_in_node);
868 if (!c_len)
869 goto out;
870
871 blkaddr = data_blkaddr(dn->inode, dn->node_page, ofs_in_node);
872 if (blkaddr == COMPRESS_ADDR)
873 blkaddr = data_blkaddr(dn->inode, dn->node_page,
874 ofs_in_node + 1);
875
876 f2fs_update_read_extent_tree_range_compressed(dn->inode,
877 fofs, blkaddr, cluster_size, c_len);
878 }
879 out:
880 return 0;
881
882 release_pages:
883 f2fs_put_page(parent, 1);
884 if (i > 1)
885 f2fs_put_page(npage[0], 0);
886 release_out:
887 dn->inode_page = NULL;
888 dn->node_page = NULL;
889 if (err == -ENOENT) {
890 dn->cur_level = i;
891 dn->max_level = level;
892 dn->ofs_in_node = offset[level];
893 }
894 return err;
895 }
896
truncate_node(struct dnode_of_data * dn)897 static int truncate_node(struct dnode_of_data *dn)
898 {
899 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
900 struct node_info ni;
901 int err;
902 pgoff_t index;
903
904 err = f2fs_get_node_info(sbi, dn->nid, &ni, false);
905 if (err)
906 return err;
907
908 if (ni.blk_addr != NEW_ADDR &&
909 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr, DATA_GENERIC_ENHANCE)) {
910 f2fs_err_ratelimited(sbi,
911 "nat entry is corrupted, run fsck to fix it, ino:%u, "
912 "nid:%u, blkaddr:%u", ni.ino, ni.nid, ni.blk_addr);
913 set_sbi_flag(sbi, SBI_NEED_FSCK);
914 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NAT);
915 return -EFSCORRUPTED;
916 }
917
918 /* Deallocate node address */
919 f2fs_invalidate_blocks(sbi, ni.blk_addr);
920 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
921 set_node_addr(sbi, &ni, NULL_ADDR, false);
922
923 if (dn->nid == dn->inode->i_ino) {
924 f2fs_remove_orphan_inode(sbi, dn->nid);
925 dec_valid_inode_count(sbi);
926 f2fs_inode_synced(dn->inode);
927 }
928
929 clear_node_page_dirty(dn->node_page);
930 set_sbi_flag(sbi, SBI_IS_DIRTY);
931
932 index = dn->node_page->index;
933 f2fs_put_page(dn->node_page, 1);
934
935 invalidate_mapping_pages(NODE_MAPPING(sbi),
936 index, index);
937
938 dn->node_page = NULL;
939 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
940
941 return 0;
942 }
943
truncate_dnode(struct dnode_of_data * dn)944 static int truncate_dnode(struct dnode_of_data *dn)
945 {
946 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
947 struct page *page;
948 int err;
949
950 if (dn->nid == 0)
951 return 1;
952
953 /* get direct node */
954 page = f2fs_get_node_page(sbi, dn->nid);
955 if (PTR_ERR(page) == -ENOENT)
956 return 1;
957 else if (IS_ERR(page))
958 return PTR_ERR(page);
959
960 if (IS_INODE(page) || ino_of_node(page) != dn->inode->i_ino) {
961 f2fs_err(sbi, "incorrect node reference, ino: %lu, nid: %u, ino_of_node: %u",
962 dn->inode->i_ino, dn->nid, ino_of_node(page));
963 set_sbi_flag(sbi, SBI_NEED_FSCK);
964 f2fs_handle_error(sbi, ERROR_INVALID_NODE_REFERENCE);
965 f2fs_put_page(page, 1);
966 return -EFSCORRUPTED;
967 }
968
969 /* Make dnode_of_data for parameter */
970 dn->node_page = page;
971 dn->ofs_in_node = 0;
972 f2fs_truncate_data_blocks_range(dn, ADDRS_PER_BLOCK(dn->inode));
973 err = truncate_node(dn);
974 if (err) {
975 f2fs_put_page(page, 1);
976 return err;
977 }
978
979 return 1;
980 }
981
truncate_nodes(struct dnode_of_data * dn,unsigned int nofs,int ofs,int depth)982 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
983 int ofs, int depth)
984 {
985 struct dnode_of_data rdn = *dn;
986 struct page *page;
987 struct f2fs_node *rn;
988 nid_t child_nid;
989 unsigned int child_nofs;
990 int freed = 0;
991 int i, ret;
992
993 if (dn->nid == 0)
994 return NIDS_PER_BLOCK + 1;
995
996 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
997
998 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
999 if (IS_ERR(page)) {
1000 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
1001 return PTR_ERR(page);
1002 }
1003
1004 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
1005
1006 rn = F2FS_NODE(page);
1007 if (depth < 3) {
1008 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
1009 child_nid = le32_to_cpu(rn->in.nid[i]);
1010 if (child_nid == 0)
1011 continue;
1012 rdn.nid = child_nid;
1013 ret = truncate_dnode(&rdn);
1014 if (ret < 0)
1015 goto out_err;
1016 if (set_nid(page, i, 0, false))
1017 dn->node_changed = true;
1018 }
1019 } else {
1020 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
1021 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
1022 child_nid = le32_to_cpu(rn->in.nid[i]);
1023 if (child_nid == 0) {
1024 child_nofs += NIDS_PER_BLOCK + 1;
1025 continue;
1026 }
1027 rdn.nid = child_nid;
1028 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
1029 if (ret == (NIDS_PER_BLOCK + 1)) {
1030 if (set_nid(page, i, 0, false))
1031 dn->node_changed = true;
1032 child_nofs += ret;
1033 } else if (ret < 0 && ret != -ENOENT) {
1034 goto out_err;
1035 }
1036 }
1037 freed = child_nofs;
1038 }
1039
1040 if (!ofs) {
1041 /* remove current indirect node */
1042 dn->node_page = page;
1043 ret = truncate_node(dn);
1044 if (ret)
1045 goto out_err;
1046 freed++;
1047 } else {
1048 f2fs_put_page(page, 1);
1049 }
1050 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
1051 return freed;
1052
1053 out_err:
1054 f2fs_put_page(page, 1);
1055 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
1056 return ret;
1057 }
1058
truncate_partial_nodes(struct dnode_of_data * dn,struct f2fs_inode * ri,int * offset,int depth)1059 static int truncate_partial_nodes(struct dnode_of_data *dn,
1060 struct f2fs_inode *ri, int *offset, int depth)
1061 {
1062 struct page *pages[2];
1063 nid_t nid[3];
1064 nid_t child_nid;
1065 int err = 0;
1066 int i;
1067 int idx = depth - 2;
1068
1069 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1070 if (!nid[0])
1071 return 0;
1072
1073 /* get indirect nodes in the path */
1074 for (i = 0; i < idx + 1; i++) {
1075 /* reference count'll be increased */
1076 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
1077 if (IS_ERR(pages[i])) {
1078 err = PTR_ERR(pages[i]);
1079 idx = i - 1;
1080 goto fail;
1081 }
1082 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
1083 }
1084
1085 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
1086
1087 /* free direct nodes linked to a partial indirect node */
1088 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1089 child_nid = get_nid(pages[idx], i, false);
1090 if (!child_nid)
1091 continue;
1092 dn->nid = child_nid;
1093 err = truncate_dnode(dn);
1094 if (err < 0)
1095 goto fail;
1096 if (set_nid(pages[idx], i, 0, false))
1097 dn->node_changed = true;
1098 }
1099
1100 if (offset[idx + 1] == 0) {
1101 dn->node_page = pages[idx];
1102 dn->nid = nid[idx];
1103 err = truncate_node(dn);
1104 if (err)
1105 goto fail;
1106 } else {
1107 f2fs_put_page(pages[idx], 1);
1108 }
1109 offset[idx]++;
1110 offset[idx + 1] = 0;
1111 idx--;
1112 fail:
1113 for (i = idx; i >= 0; i--)
1114 f2fs_put_page(pages[i], 1);
1115
1116 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1117
1118 return err;
1119 }
1120
1121 /*
1122 * All the block addresses of data and nodes should be nullified.
1123 */
f2fs_truncate_inode_blocks(struct inode * inode,pgoff_t from)1124 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1125 {
1126 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1127 int err = 0, cont = 1;
1128 int level, offset[4], noffset[4];
1129 unsigned int nofs = 0;
1130 struct f2fs_inode *ri;
1131 struct dnode_of_data dn;
1132 struct page *page;
1133
1134 trace_f2fs_truncate_inode_blocks_enter(inode, from);
1135
1136 level = get_node_path(inode, from, offset, noffset);
1137 if (level < 0) {
1138 trace_f2fs_truncate_inode_blocks_exit(inode, level);
1139 return level;
1140 }
1141
1142 page = f2fs_get_node_page(sbi, inode->i_ino);
1143 if (IS_ERR(page)) {
1144 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1145 return PTR_ERR(page);
1146 }
1147
1148 set_new_dnode(&dn, inode, page, NULL, 0);
1149 unlock_page(page);
1150
1151 ri = F2FS_INODE(page);
1152 switch (level) {
1153 case 0:
1154 case 1:
1155 nofs = noffset[1];
1156 break;
1157 case 2:
1158 nofs = noffset[1];
1159 if (!offset[level - 1])
1160 goto skip_partial;
1161 err = truncate_partial_nodes(&dn, ri, offset, level);
1162 if (err < 0 && err != -ENOENT)
1163 goto fail;
1164 nofs += 1 + NIDS_PER_BLOCK;
1165 break;
1166 case 3:
1167 nofs = 5 + 2 * NIDS_PER_BLOCK;
1168 if (!offset[level - 1])
1169 goto skip_partial;
1170 err = truncate_partial_nodes(&dn, ri, offset, level);
1171 if (err < 0 && err != -ENOENT)
1172 goto fail;
1173 break;
1174 default:
1175 BUG();
1176 }
1177
1178 skip_partial:
1179 while (cont) {
1180 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1181 switch (offset[0]) {
1182 case NODE_DIR1_BLOCK:
1183 case NODE_DIR2_BLOCK:
1184 err = truncate_dnode(&dn);
1185 break;
1186
1187 case NODE_IND1_BLOCK:
1188 case NODE_IND2_BLOCK:
1189 err = truncate_nodes(&dn, nofs, offset[1], 2);
1190 break;
1191
1192 case NODE_DIND_BLOCK:
1193 err = truncate_nodes(&dn, nofs, offset[1], 3);
1194 cont = 0;
1195 break;
1196
1197 default:
1198 BUG();
1199 }
1200 if (err < 0 && err != -ENOENT)
1201 goto fail;
1202 if (offset[1] == 0 &&
1203 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1204 lock_page(page);
1205 BUG_ON(page->mapping != NODE_MAPPING(sbi));
1206 f2fs_wait_on_page_writeback(page, NODE, true, true);
1207 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1208 set_page_dirty(page);
1209 unlock_page(page);
1210 }
1211 offset[1] = 0;
1212 offset[0]++;
1213 nofs += err;
1214 }
1215 fail:
1216 f2fs_put_page(page, 0);
1217 trace_f2fs_truncate_inode_blocks_exit(inode, err);
1218 return err > 0 ? 0 : err;
1219 }
1220
1221 /* caller must lock inode page */
f2fs_truncate_xattr_node(struct inode * inode)1222 int f2fs_truncate_xattr_node(struct inode *inode)
1223 {
1224 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1225 nid_t nid = F2FS_I(inode)->i_xattr_nid;
1226 struct dnode_of_data dn;
1227 struct page *npage;
1228 int err;
1229
1230 if (!nid)
1231 return 0;
1232
1233 npage = f2fs_get_node_page(sbi, nid);
1234 if (IS_ERR(npage))
1235 return PTR_ERR(npage);
1236
1237 set_new_dnode(&dn, inode, NULL, npage, nid);
1238 err = truncate_node(&dn);
1239 if (err) {
1240 f2fs_put_page(npage, 1);
1241 return err;
1242 }
1243
1244 f2fs_i_xnid_write(inode, 0);
1245
1246 return 0;
1247 }
1248
1249 /*
1250 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1251 * f2fs_unlock_op().
1252 */
f2fs_remove_inode_page(struct inode * inode)1253 int f2fs_remove_inode_page(struct inode *inode)
1254 {
1255 struct dnode_of_data dn;
1256 int err;
1257
1258 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1259 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1260 if (err)
1261 return err;
1262
1263 err = f2fs_truncate_xattr_node(inode);
1264 if (err) {
1265 f2fs_put_dnode(&dn);
1266 return err;
1267 }
1268
1269 /* remove potential inline_data blocks */
1270 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1271 S_ISLNK(inode->i_mode))
1272 f2fs_truncate_data_blocks_range(&dn, 1);
1273
1274 /* 0 is possible, after f2fs_new_inode() has failed */
1275 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1276 f2fs_put_dnode(&dn);
1277 return -EIO;
1278 }
1279
1280 if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1281 f2fs_warn(F2FS_I_SB(inode),
1282 "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1283 inode->i_ino, (unsigned long long)inode->i_blocks);
1284 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1285 }
1286
1287 /* will put inode & node pages */
1288 err = truncate_node(&dn);
1289 if (err) {
1290 f2fs_put_dnode(&dn);
1291 return err;
1292 }
1293 return 0;
1294 }
1295
f2fs_new_inode_page(struct inode * inode)1296 struct page *f2fs_new_inode_page(struct inode *inode)
1297 {
1298 struct dnode_of_data dn;
1299
1300 /* allocate inode page for new inode */
1301 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1302
1303 /* caller should f2fs_put_page(page, 1); */
1304 return f2fs_new_node_page(&dn, 0);
1305 }
1306
f2fs_new_node_page(struct dnode_of_data * dn,unsigned int ofs)1307 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1308 {
1309 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1310 struct node_info new_ni;
1311 struct page *page;
1312 int err;
1313
1314 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1315 return ERR_PTR(-EPERM);
1316
1317 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1318 if (!page)
1319 return ERR_PTR(-ENOMEM);
1320
1321 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1322 goto fail;
1323
1324 #ifdef CONFIG_F2FS_CHECK_FS
1325 err = f2fs_get_node_info(sbi, dn->nid, &new_ni, false);
1326 if (err) {
1327 dec_valid_node_count(sbi, dn->inode, !ofs);
1328 goto fail;
1329 }
1330 if (unlikely(new_ni.blk_addr != NULL_ADDR)) {
1331 err = -EFSCORRUPTED;
1332 dec_valid_node_count(sbi, dn->inode, !ofs);
1333 set_sbi_flag(sbi, SBI_NEED_FSCK);
1334 f2fs_warn_ratelimited(sbi,
1335 "f2fs_new_node_page: inconsistent nat entry, "
1336 "ino:%u, nid:%u, blkaddr:%u, ver:%u, flag:%u",
1337 new_ni.ino, new_ni.nid, new_ni.blk_addr,
1338 new_ni.version, new_ni.flag);
1339 f2fs_handle_error(sbi, ERROR_INCONSISTENT_NAT);
1340 goto fail;
1341 }
1342 #endif
1343 new_ni.nid = dn->nid;
1344 new_ni.ino = dn->inode->i_ino;
1345 new_ni.blk_addr = NULL_ADDR;
1346 new_ni.flag = 0;
1347 new_ni.version = 0;
1348 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1349
1350 f2fs_wait_on_page_writeback(page, NODE, true, true);
1351 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1352 set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1353 if (!PageUptodate(page))
1354 SetPageUptodate(page);
1355 if (set_page_dirty(page))
1356 dn->node_changed = true;
1357
1358 if (f2fs_has_xattr_block(ofs))
1359 f2fs_i_xnid_write(dn->inode, dn->nid);
1360
1361 if (ofs == 0)
1362 inc_valid_inode_count(sbi);
1363 return page;
1364 fail:
1365 clear_node_page_dirty(page);
1366 f2fs_put_page(page, 1);
1367 return ERR_PTR(err);
1368 }
1369
1370 /*
1371 * Caller should do after getting the following values.
1372 * 0: f2fs_put_page(page, 0)
1373 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1374 */
read_node_page(struct page * page,blk_opf_t op_flags)1375 static int read_node_page(struct page *page, blk_opf_t op_flags)
1376 {
1377 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1378 struct node_info ni;
1379 struct f2fs_io_info fio = {
1380 .sbi = sbi,
1381 .type = NODE,
1382 .op = REQ_OP_READ,
1383 .op_flags = op_flags,
1384 .page = page,
1385 .encrypted_page = NULL,
1386 };
1387 int err;
1388
1389 if (PageUptodate(page)) {
1390 if (!f2fs_inode_chksum_verify(sbi, page)) {
1391 ClearPageUptodate(page);
1392 return -EFSBADCRC;
1393 }
1394 return LOCKED_PAGE;
1395 }
1396
1397 err = f2fs_get_node_info(sbi, page->index, &ni, false);
1398 if (err)
1399 return err;
1400
1401 /* NEW_ADDR can be seen, after cp_error drops some dirty node pages */
1402 if (unlikely(ni.blk_addr == NULL_ADDR || ni.blk_addr == NEW_ADDR)) {
1403 ClearPageUptodate(page);
1404 return -ENOENT;
1405 }
1406
1407 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1408
1409 err = f2fs_submit_page_bio(&fio);
1410
1411 if (!err)
1412 f2fs_update_iostat(sbi, NULL, FS_NODE_READ_IO, F2FS_BLKSIZE);
1413
1414 return err;
1415 }
1416
1417 /*
1418 * Readahead a node page
1419 */
f2fs_ra_node_page(struct f2fs_sb_info * sbi,nid_t nid)1420 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1421 {
1422 struct page *apage;
1423 int err;
1424
1425 if (!nid)
1426 return;
1427 if (f2fs_check_nid_range(sbi, nid))
1428 return;
1429
1430 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1431 if (apage)
1432 return;
1433
1434 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1435 if (!apage)
1436 return;
1437
1438 err = read_node_page(apage, REQ_RAHEAD);
1439 f2fs_put_page(apage, err ? 1 : 0);
1440 }
1441
__get_node_page(struct f2fs_sb_info * sbi,pgoff_t nid,struct page * parent,int start)1442 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1443 struct page *parent, int start)
1444 {
1445 struct page *page;
1446 int err;
1447
1448 if (!nid)
1449 return ERR_PTR(-ENOENT);
1450 if (f2fs_check_nid_range(sbi, nid))
1451 return ERR_PTR(-EINVAL);
1452 repeat:
1453 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1454 if (!page)
1455 return ERR_PTR(-ENOMEM);
1456
1457 err = read_node_page(page, 0);
1458 if (err < 0) {
1459 goto out_put_err;
1460 } else if (err == LOCKED_PAGE) {
1461 err = 0;
1462 goto page_hit;
1463 }
1464
1465 if (parent)
1466 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1467
1468 lock_page(page);
1469
1470 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1471 f2fs_put_page(page, 1);
1472 goto repeat;
1473 }
1474
1475 if (unlikely(!PageUptodate(page))) {
1476 err = -EIO;
1477 goto out_err;
1478 }
1479
1480 if (!f2fs_inode_chksum_verify(sbi, page)) {
1481 err = -EFSBADCRC;
1482 goto out_err;
1483 }
1484 page_hit:
1485 if (likely(nid == nid_of_node(page)))
1486 return page;
1487
1488 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1489 nid, nid_of_node(page), ino_of_node(page),
1490 ofs_of_node(page), cpver_of_node(page),
1491 next_blkaddr_of_node(page));
1492 set_sbi_flag(sbi, SBI_NEED_FSCK);
1493 f2fs_handle_error(sbi, ERROR_INCONSISTENT_FOOTER);
1494 err = -EFSCORRUPTED;
1495 out_err:
1496 ClearPageUptodate(page);
1497 out_put_err:
1498 /* ENOENT comes from read_node_page which is not an error. */
1499 if (err != -ENOENT)
1500 f2fs_handle_page_eio(sbi, page->index, NODE);
1501 f2fs_put_page(page, 1);
1502 return ERR_PTR(err);
1503 }
1504
f2fs_get_node_page(struct f2fs_sb_info * sbi,pgoff_t nid)1505 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1506 {
1507 return __get_node_page(sbi, nid, NULL, 0);
1508 }
1509
f2fs_get_node_page_ra(struct page * parent,int start)1510 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1511 {
1512 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1513 nid_t nid = get_nid(parent, start, false);
1514
1515 return __get_node_page(sbi, nid, parent, start);
1516 }
1517
flush_inline_data(struct f2fs_sb_info * sbi,nid_t ino)1518 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1519 {
1520 struct inode *inode;
1521 struct page *page;
1522 int ret;
1523
1524 /* should flush inline_data before evict_inode */
1525 inode = ilookup(sbi->sb, ino);
1526 if (!inode)
1527 return;
1528
1529 page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1530 FGP_LOCK|FGP_NOWAIT, 0);
1531 if (!page)
1532 goto iput_out;
1533
1534 if (!PageUptodate(page))
1535 goto page_out;
1536
1537 if (!PageDirty(page))
1538 goto page_out;
1539
1540 if (!clear_page_dirty_for_io(page))
1541 goto page_out;
1542
1543 ret = f2fs_write_inline_data(inode, page);
1544 inode_dec_dirty_pages(inode);
1545 f2fs_remove_dirty_inode(inode);
1546 if (ret)
1547 set_page_dirty(page);
1548 page_out:
1549 f2fs_put_page(page, 1);
1550 iput_out:
1551 iput(inode);
1552 }
1553
last_fsync_dnode(struct f2fs_sb_info * sbi,nid_t ino)1554 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1555 {
1556 pgoff_t index;
1557 struct folio_batch fbatch;
1558 struct page *last_page = NULL;
1559 int nr_folios;
1560
1561 folio_batch_init(&fbatch);
1562 index = 0;
1563
1564 while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1565 (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1566 &fbatch))) {
1567 int i;
1568
1569 for (i = 0; i < nr_folios; i++) {
1570 struct page *page = &fbatch.folios[i]->page;
1571
1572 if (unlikely(f2fs_cp_error(sbi))) {
1573 f2fs_put_page(last_page, 0);
1574 folio_batch_release(&fbatch);
1575 return ERR_PTR(-EIO);
1576 }
1577
1578 if (!IS_DNODE(page) || !is_cold_node(page))
1579 continue;
1580 if (ino_of_node(page) != ino)
1581 continue;
1582
1583 lock_page(page);
1584
1585 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1586 continue_unlock:
1587 unlock_page(page);
1588 continue;
1589 }
1590 if (ino_of_node(page) != ino)
1591 goto continue_unlock;
1592
1593 if (!PageDirty(page)) {
1594 /* someone wrote it for us */
1595 goto continue_unlock;
1596 }
1597
1598 if (last_page)
1599 f2fs_put_page(last_page, 0);
1600
1601 get_page(page);
1602 last_page = page;
1603 unlock_page(page);
1604 }
1605 folio_batch_release(&fbatch);
1606 cond_resched();
1607 }
1608 return last_page;
1609 }
1610
__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)1611 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1612 struct writeback_control *wbc, bool do_balance,
1613 enum iostat_type io_type, unsigned int *seq_id)
1614 {
1615 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1616 nid_t nid;
1617 struct node_info ni;
1618 struct f2fs_io_info fio = {
1619 .sbi = sbi,
1620 .ino = ino_of_node(page),
1621 .type = NODE,
1622 .op = REQ_OP_WRITE,
1623 .op_flags = wbc_to_write_flags(wbc),
1624 .page = page,
1625 .encrypted_page = NULL,
1626 .submitted = 0,
1627 .io_type = io_type,
1628 .io_wbc = wbc,
1629 };
1630 unsigned int seq;
1631
1632 trace_f2fs_writepage(page, NODE);
1633
1634 if (unlikely(f2fs_cp_error(sbi))) {
1635 /* keep node pages in remount-ro mode */
1636 if (F2FS_OPTION(sbi).errors == MOUNT_ERRORS_READONLY)
1637 goto redirty_out;
1638 ClearPageUptodate(page);
1639 dec_page_count(sbi, F2FS_DIRTY_NODES);
1640 unlock_page(page);
1641 return 0;
1642 }
1643
1644 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1645 goto redirty_out;
1646
1647 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1648 wbc->sync_mode == WB_SYNC_NONE &&
1649 IS_DNODE(page) && is_cold_node(page))
1650 goto redirty_out;
1651
1652 /* get old block addr of this node page */
1653 nid = nid_of_node(page);
1654 f2fs_bug_on(sbi, page->index != nid);
1655
1656 if (f2fs_get_node_info(sbi, nid, &ni, !do_balance))
1657 goto redirty_out;
1658
1659 if (wbc->for_reclaim) {
1660 if (!f2fs_down_read_trylock(&sbi->node_write))
1661 goto redirty_out;
1662 } else {
1663 f2fs_down_read(&sbi->node_write);
1664 }
1665
1666 /* This page is already truncated */
1667 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1668 ClearPageUptodate(page);
1669 dec_page_count(sbi, F2FS_DIRTY_NODES);
1670 f2fs_up_read(&sbi->node_write);
1671 unlock_page(page);
1672 return 0;
1673 }
1674
1675 if (__is_valid_data_blkaddr(ni.blk_addr) &&
1676 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1677 DATA_GENERIC_ENHANCE)) {
1678 f2fs_up_read(&sbi->node_write);
1679 goto redirty_out;
1680 }
1681
1682 if (atomic && !test_opt(sbi, NOBARRIER) && !f2fs_sb_has_blkzoned(sbi))
1683 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1684
1685 /* should add to global list before clearing PAGECACHE status */
1686 if (f2fs_in_warm_node_list(sbi, page)) {
1687 seq = f2fs_add_fsync_node_entry(sbi, page);
1688 if (seq_id)
1689 *seq_id = seq;
1690 }
1691
1692 set_page_writeback(page);
1693
1694 fio.old_blkaddr = ni.blk_addr;
1695 f2fs_do_write_node_page(nid, &fio);
1696 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1697 dec_page_count(sbi, F2FS_DIRTY_NODES);
1698 f2fs_up_read(&sbi->node_write);
1699
1700 if (wbc->for_reclaim) {
1701 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1702 submitted = NULL;
1703 }
1704
1705 unlock_page(page);
1706
1707 if (unlikely(f2fs_cp_error(sbi))) {
1708 f2fs_submit_merged_write(sbi, NODE);
1709 submitted = NULL;
1710 }
1711 if (submitted)
1712 *submitted = fio.submitted;
1713
1714 if (do_balance)
1715 f2fs_balance_fs(sbi, false);
1716 return 0;
1717
1718 redirty_out:
1719 redirty_page_for_writepage(wbc, page);
1720 return AOP_WRITEPAGE_ACTIVATE;
1721 }
1722
f2fs_move_node_page(struct page * node_page,int gc_type)1723 int f2fs_move_node_page(struct page *node_page, int gc_type)
1724 {
1725 int err = 0;
1726
1727 if (gc_type == FG_GC) {
1728 struct writeback_control wbc = {
1729 .sync_mode = WB_SYNC_ALL,
1730 .nr_to_write = 1,
1731 .for_reclaim = 0,
1732 };
1733
1734 f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1735
1736 set_page_dirty(node_page);
1737
1738 if (!clear_page_dirty_for_io(node_page)) {
1739 err = -EAGAIN;
1740 goto out_page;
1741 }
1742
1743 if (__write_node_page(node_page, false, NULL,
1744 &wbc, false, FS_GC_NODE_IO, NULL)) {
1745 err = -EAGAIN;
1746 unlock_page(node_page);
1747 }
1748 goto release_page;
1749 } else {
1750 /* set page dirty and write it */
1751 if (!PageWriteback(node_page))
1752 set_page_dirty(node_page);
1753 }
1754 out_page:
1755 unlock_page(node_page);
1756 release_page:
1757 f2fs_put_page(node_page, 0);
1758 return err;
1759 }
1760
f2fs_write_node_page(struct page * page,struct writeback_control * wbc)1761 static int f2fs_write_node_page(struct page *page,
1762 struct writeback_control *wbc)
1763 {
1764 return __write_node_page(page, false, NULL, wbc, false,
1765 FS_NODE_IO, NULL);
1766 }
1767
f2fs_fsync_node_pages(struct f2fs_sb_info * sbi,struct inode * inode,struct writeback_control * wbc,bool atomic,unsigned int * seq_id)1768 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1769 struct writeback_control *wbc, bool atomic,
1770 unsigned int *seq_id)
1771 {
1772 pgoff_t index;
1773 struct folio_batch fbatch;
1774 int ret = 0;
1775 struct page *last_page = NULL;
1776 bool marked = false;
1777 nid_t ino = inode->i_ino;
1778 int nr_folios;
1779 int nwritten = 0;
1780
1781 if (atomic) {
1782 last_page = last_fsync_dnode(sbi, ino);
1783 if (IS_ERR_OR_NULL(last_page))
1784 return PTR_ERR_OR_ZERO(last_page);
1785 }
1786 retry:
1787 folio_batch_init(&fbatch);
1788 index = 0;
1789
1790 while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1791 (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1792 &fbatch))) {
1793 int i;
1794
1795 for (i = 0; i < nr_folios; i++) {
1796 struct page *page = &fbatch.folios[i]->page;
1797 bool submitted = false;
1798
1799 if (unlikely(f2fs_cp_error(sbi))) {
1800 f2fs_put_page(last_page, 0);
1801 folio_batch_release(&fbatch);
1802 ret = -EIO;
1803 goto out;
1804 }
1805
1806 if (!IS_DNODE(page) || !is_cold_node(page))
1807 continue;
1808 if (ino_of_node(page) != ino)
1809 continue;
1810
1811 lock_page(page);
1812
1813 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1814 continue_unlock:
1815 unlock_page(page);
1816 continue;
1817 }
1818 if (ino_of_node(page) != ino)
1819 goto continue_unlock;
1820
1821 if (!PageDirty(page) && page != last_page) {
1822 /* someone wrote it for us */
1823 goto continue_unlock;
1824 }
1825
1826 f2fs_wait_on_page_writeback(page, NODE, true, true);
1827
1828 set_fsync_mark(page, 0);
1829 set_dentry_mark(page, 0);
1830
1831 if (!atomic || page == last_page) {
1832 set_fsync_mark(page, 1);
1833 percpu_counter_inc(&sbi->rf_node_block_count);
1834 if (IS_INODE(page)) {
1835 if (is_inode_flag_set(inode,
1836 FI_DIRTY_INODE))
1837 f2fs_update_inode(inode, page);
1838 set_dentry_mark(page,
1839 f2fs_need_dentry_mark(sbi, ino));
1840 }
1841 /* may be written by other thread */
1842 if (!PageDirty(page))
1843 set_page_dirty(page);
1844 }
1845
1846 if (!clear_page_dirty_for_io(page))
1847 goto continue_unlock;
1848
1849 ret = __write_node_page(page, atomic &&
1850 page == last_page,
1851 &submitted, wbc, true,
1852 FS_NODE_IO, seq_id);
1853 if (ret) {
1854 unlock_page(page);
1855 f2fs_put_page(last_page, 0);
1856 break;
1857 } else if (submitted) {
1858 nwritten++;
1859 }
1860
1861 if (page == last_page) {
1862 f2fs_put_page(page, 0);
1863 marked = true;
1864 break;
1865 }
1866 }
1867 folio_batch_release(&fbatch);
1868 cond_resched();
1869
1870 if (ret || marked)
1871 break;
1872 }
1873 if (!ret && atomic && !marked) {
1874 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1875 ino, last_page->index);
1876 lock_page(last_page);
1877 f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1878 set_page_dirty(last_page);
1879 unlock_page(last_page);
1880 goto retry;
1881 }
1882 out:
1883 if (nwritten)
1884 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1885 return ret ? -EIO : 0;
1886 }
1887
f2fs_match_ino(struct inode * inode,unsigned long ino,void * data)1888 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1889 {
1890 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1891 bool clean;
1892
1893 if (inode->i_ino != ino)
1894 return 0;
1895
1896 if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1897 return 0;
1898
1899 spin_lock(&sbi->inode_lock[DIRTY_META]);
1900 clean = list_empty(&F2FS_I(inode)->gdirty_list);
1901 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1902
1903 if (clean)
1904 return 0;
1905
1906 inode = igrab(inode);
1907 if (!inode)
1908 return 0;
1909 return 1;
1910 }
1911
flush_dirty_inode(struct page * page)1912 static bool flush_dirty_inode(struct page *page)
1913 {
1914 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1915 struct inode *inode;
1916 nid_t ino = ino_of_node(page);
1917
1918 inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1919 if (!inode)
1920 return false;
1921
1922 f2fs_update_inode(inode, page);
1923 unlock_page(page);
1924
1925 iput(inode);
1926 return true;
1927 }
1928
f2fs_flush_inline_data(struct f2fs_sb_info * sbi)1929 void f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1930 {
1931 pgoff_t index = 0;
1932 struct folio_batch fbatch;
1933 int nr_folios;
1934
1935 folio_batch_init(&fbatch);
1936
1937 while ((nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi), &index,
1938 (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1939 &fbatch))) {
1940 int i;
1941
1942 for (i = 0; i < nr_folios; i++) {
1943 struct page *page = &fbatch.folios[i]->page;
1944
1945 if (!IS_DNODE(page))
1946 continue;
1947
1948 lock_page(page);
1949
1950 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1951 continue_unlock:
1952 unlock_page(page);
1953 continue;
1954 }
1955
1956 if (!PageDirty(page)) {
1957 /* someone wrote it for us */
1958 goto continue_unlock;
1959 }
1960
1961 /* flush inline_data, if it's async context. */
1962 if (page_private_inline(page)) {
1963 clear_page_private_inline(page);
1964 unlock_page(page);
1965 flush_inline_data(sbi, ino_of_node(page));
1966 continue;
1967 }
1968 unlock_page(page);
1969 }
1970 folio_batch_release(&fbatch);
1971 cond_resched();
1972 }
1973 }
1974
f2fs_sync_node_pages(struct f2fs_sb_info * sbi,struct writeback_control * wbc,bool do_balance,enum iostat_type io_type)1975 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1976 struct writeback_control *wbc,
1977 bool do_balance, enum iostat_type io_type)
1978 {
1979 pgoff_t index;
1980 struct folio_batch fbatch;
1981 int step = 0;
1982 int nwritten = 0;
1983 int ret = 0;
1984 int nr_folios, done = 0;
1985
1986 folio_batch_init(&fbatch);
1987
1988 next_step:
1989 index = 0;
1990
1991 while (!done && (nr_folios = filemap_get_folios_tag(NODE_MAPPING(sbi),
1992 &index, (pgoff_t)-1, PAGECACHE_TAG_DIRTY,
1993 &fbatch))) {
1994 int i;
1995
1996 for (i = 0; i < nr_folios; i++) {
1997 struct page *page = &fbatch.folios[i]->page;
1998 bool submitted = false;
1999
2000 /* give a priority to WB_SYNC threads */
2001 if (atomic_read(&sbi->wb_sync_req[NODE]) &&
2002 wbc->sync_mode == WB_SYNC_NONE) {
2003 done = 1;
2004 break;
2005 }
2006
2007 /*
2008 * flushing sequence with step:
2009 * 0. indirect nodes
2010 * 1. dentry dnodes
2011 * 2. file dnodes
2012 */
2013 if (step == 0 && IS_DNODE(page))
2014 continue;
2015 if (step == 1 && (!IS_DNODE(page) ||
2016 is_cold_node(page)))
2017 continue;
2018 if (step == 2 && (!IS_DNODE(page) ||
2019 !is_cold_node(page)))
2020 continue;
2021 lock_node:
2022 if (wbc->sync_mode == WB_SYNC_ALL)
2023 lock_page(page);
2024 else if (!trylock_page(page))
2025 continue;
2026
2027 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
2028 continue_unlock:
2029 unlock_page(page);
2030 continue;
2031 }
2032
2033 if (!PageDirty(page)) {
2034 /* someone wrote it for us */
2035 goto continue_unlock;
2036 }
2037
2038 /* flush inline_data/inode, if it's async context. */
2039 if (!do_balance)
2040 goto write_node;
2041
2042 /* flush inline_data */
2043 if (page_private_inline(page)) {
2044 clear_page_private_inline(page);
2045 unlock_page(page);
2046 flush_inline_data(sbi, ino_of_node(page));
2047 goto lock_node;
2048 }
2049
2050 /* flush dirty inode */
2051 if (IS_INODE(page) && flush_dirty_inode(page))
2052 goto lock_node;
2053 write_node:
2054 f2fs_wait_on_page_writeback(page, NODE, true, true);
2055
2056 if (!clear_page_dirty_for_io(page))
2057 goto continue_unlock;
2058
2059 set_fsync_mark(page, 0);
2060 set_dentry_mark(page, 0);
2061
2062 ret = __write_node_page(page, false, &submitted,
2063 wbc, do_balance, io_type, NULL);
2064 if (ret)
2065 unlock_page(page);
2066 else if (submitted)
2067 nwritten++;
2068
2069 if (--wbc->nr_to_write == 0)
2070 break;
2071 }
2072 folio_batch_release(&fbatch);
2073 cond_resched();
2074
2075 if (wbc->nr_to_write == 0) {
2076 step = 2;
2077 break;
2078 }
2079 }
2080
2081 if (step < 2) {
2082 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
2083 wbc->sync_mode == WB_SYNC_NONE && step == 1)
2084 goto out;
2085 step++;
2086 goto next_step;
2087 }
2088 out:
2089 if (nwritten)
2090 f2fs_submit_merged_write(sbi, NODE);
2091
2092 if (unlikely(f2fs_cp_error(sbi)))
2093 return -EIO;
2094 return ret;
2095 }
2096
f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info * sbi,unsigned int seq_id)2097 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
2098 unsigned int seq_id)
2099 {
2100 struct fsync_node_entry *fn;
2101 struct page *page;
2102 struct list_head *head = &sbi->fsync_node_list;
2103 unsigned long flags;
2104 unsigned int cur_seq_id = 0;
2105
2106 while (seq_id && cur_seq_id < seq_id) {
2107 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2108 if (list_empty(head)) {
2109 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2110 break;
2111 }
2112 fn = list_first_entry(head, struct fsync_node_entry, list);
2113 if (fn->seq_id > seq_id) {
2114 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2115 break;
2116 }
2117 cur_seq_id = fn->seq_id;
2118 page = fn->page;
2119 get_page(page);
2120 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2121
2122 f2fs_wait_on_page_writeback(page, NODE, true, false);
2123
2124 put_page(page);
2125 }
2126
2127 return filemap_check_errors(NODE_MAPPING(sbi));
2128 }
2129
f2fs_write_node_pages(struct address_space * mapping,struct writeback_control * wbc)2130 static int f2fs_write_node_pages(struct address_space *mapping,
2131 struct writeback_control *wbc)
2132 {
2133 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2134 struct blk_plug plug;
2135 long diff;
2136
2137 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2138 goto skip_write;
2139
2140 /* balancing f2fs's metadata in background */
2141 f2fs_balance_fs_bg(sbi, true);
2142
2143 /* collect a number of dirty node pages and write together */
2144 if (wbc->sync_mode != WB_SYNC_ALL &&
2145 get_pages(sbi, F2FS_DIRTY_NODES) <
2146 nr_pages_to_skip(sbi, NODE))
2147 goto skip_write;
2148
2149 if (wbc->sync_mode == WB_SYNC_ALL)
2150 atomic_inc(&sbi->wb_sync_req[NODE]);
2151 else if (atomic_read(&sbi->wb_sync_req[NODE])) {
2152 /* to avoid potential deadlock */
2153 if (current->plug)
2154 blk_finish_plug(current->plug);
2155 goto skip_write;
2156 }
2157
2158 trace_f2fs_writepages(mapping->host, wbc, NODE);
2159
2160 diff = nr_pages_to_write(sbi, NODE, wbc);
2161 blk_start_plug(&plug);
2162 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2163 blk_finish_plug(&plug);
2164 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2165
2166 if (wbc->sync_mode == WB_SYNC_ALL)
2167 atomic_dec(&sbi->wb_sync_req[NODE]);
2168 return 0;
2169
2170 skip_write:
2171 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2172 trace_f2fs_writepages(mapping->host, wbc, NODE);
2173 return 0;
2174 }
2175
f2fs_dirty_node_folio(struct address_space * mapping,struct folio * folio)2176 static bool f2fs_dirty_node_folio(struct address_space *mapping,
2177 struct folio *folio)
2178 {
2179 trace_f2fs_set_page_dirty(&folio->page, NODE);
2180
2181 if (!folio_test_uptodate(folio))
2182 folio_mark_uptodate(folio);
2183 #ifdef CONFIG_F2FS_CHECK_FS
2184 if (IS_INODE(&folio->page))
2185 f2fs_inode_chksum_set(F2FS_M_SB(mapping), &folio->page);
2186 #endif
2187 if (filemap_dirty_folio(mapping, folio)) {
2188 inc_page_count(F2FS_M_SB(mapping), F2FS_DIRTY_NODES);
2189 set_page_private_reference(&folio->page);
2190 return true;
2191 }
2192 return false;
2193 }
2194
2195 /*
2196 * Structure of the f2fs node operations
2197 */
2198 const struct address_space_operations f2fs_node_aops = {
2199 .writepage = f2fs_write_node_page,
2200 .writepages = f2fs_write_node_pages,
2201 .dirty_folio = f2fs_dirty_node_folio,
2202 .invalidate_folio = f2fs_invalidate_folio,
2203 .release_folio = f2fs_release_folio,
2204 .migrate_folio = filemap_migrate_folio,
2205 };
2206
__lookup_free_nid_list(struct f2fs_nm_info * nm_i,nid_t n)2207 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2208 nid_t n)
2209 {
2210 return radix_tree_lookup(&nm_i->free_nid_root, n);
2211 }
2212
__insert_free_nid(struct f2fs_sb_info * sbi,struct free_nid * i)2213 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2214 struct free_nid *i)
2215 {
2216 struct f2fs_nm_info *nm_i = NM_I(sbi);
2217 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2218
2219 if (err)
2220 return err;
2221
2222 nm_i->nid_cnt[FREE_NID]++;
2223 list_add_tail(&i->list, &nm_i->free_nid_list);
2224 return 0;
2225 }
2226
__remove_free_nid(struct f2fs_sb_info * sbi,struct free_nid * i,enum nid_state state)2227 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2228 struct free_nid *i, enum nid_state state)
2229 {
2230 struct f2fs_nm_info *nm_i = NM_I(sbi);
2231
2232 f2fs_bug_on(sbi, state != i->state);
2233 nm_i->nid_cnt[state]--;
2234 if (state == FREE_NID)
2235 list_del(&i->list);
2236 radix_tree_delete(&nm_i->free_nid_root, i->nid);
2237 }
2238
__move_free_nid(struct f2fs_sb_info * sbi,struct free_nid * i,enum nid_state org_state,enum nid_state dst_state)2239 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2240 enum nid_state org_state, enum nid_state dst_state)
2241 {
2242 struct f2fs_nm_info *nm_i = NM_I(sbi);
2243
2244 f2fs_bug_on(sbi, org_state != i->state);
2245 i->state = dst_state;
2246 nm_i->nid_cnt[org_state]--;
2247 nm_i->nid_cnt[dst_state]++;
2248
2249 switch (dst_state) {
2250 case PREALLOC_NID:
2251 list_del(&i->list);
2252 break;
2253 case FREE_NID:
2254 list_add_tail(&i->list, &nm_i->free_nid_list);
2255 break;
2256 default:
2257 BUG_ON(1);
2258 }
2259 }
2260
f2fs_nat_bitmap_enabled(struct f2fs_sb_info * sbi)2261 bool f2fs_nat_bitmap_enabled(struct f2fs_sb_info *sbi)
2262 {
2263 struct f2fs_nm_info *nm_i = NM_I(sbi);
2264 unsigned int i;
2265 bool ret = true;
2266
2267 f2fs_down_read(&nm_i->nat_tree_lock);
2268 for (i = 0; i < nm_i->nat_blocks; i++) {
2269 if (!test_bit_le(i, nm_i->nat_block_bitmap)) {
2270 ret = false;
2271 break;
2272 }
2273 }
2274 f2fs_up_read(&nm_i->nat_tree_lock);
2275
2276 return ret;
2277 }
2278
update_free_nid_bitmap(struct f2fs_sb_info * sbi,nid_t nid,bool set,bool build)2279 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2280 bool set, bool build)
2281 {
2282 struct f2fs_nm_info *nm_i = NM_I(sbi);
2283 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2284 unsigned int nid_ofs = nid - START_NID(nid);
2285
2286 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2287 return;
2288
2289 if (set) {
2290 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2291 return;
2292 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2293 nm_i->free_nid_count[nat_ofs]++;
2294 } else {
2295 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2296 return;
2297 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2298 if (!build)
2299 nm_i->free_nid_count[nat_ofs]--;
2300 }
2301 }
2302
2303 /* return if the nid is recognized as free */
add_free_nid(struct f2fs_sb_info * sbi,nid_t nid,bool build,bool update)2304 static bool add_free_nid(struct f2fs_sb_info *sbi,
2305 nid_t nid, bool build, bool update)
2306 {
2307 struct f2fs_nm_info *nm_i = NM_I(sbi);
2308 struct free_nid *i, *e;
2309 struct nat_entry *ne;
2310 int err = -EINVAL;
2311 bool ret = false;
2312
2313 /* 0 nid should not be used */
2314 if (unlikely(nid == 0))
2315 return false;
2316
2317 if (unlikely(f2fs_check_nid_range(sbi, nid)))
2318 return false;
2319
2320 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS, true, NULL);
2321 i->nid = nid;
2322 i->state = FREE_NID;
2323
2324 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2325
2326 spin_lock(&nm_i->nid_list_lock);
2327
2328 if (build) {
2329 /*
2330 * Thread A Thread B
2331 * - f2fs_create
2332 * - f2fs_new_inode
2333 * - f2fs_alloc_nid
2334 * - __insert_nid_to_list(PREALLOC_NID)
2335 * - f2fs_balance_fs_bg
2336 * - f2fs_build_free_nids
2337 * - __f2fs_build_free_nids
2338 * - scan_nat_page
2339 * - add_free_nid
2340 * - __lookup_nat_cache
2341 * - f2fs_add_link
2342 * - f2fs_init_inode_metadata
2343 * - f2fs_new_inode_page
2344 * - f2fs_new_node_page
2345 * - set_node_addr
2346 * - f2fs_alloc_nid_done
2347 * - __remove_nid_from_list(PREALLOC_NID)
2348 * - __insert_nid_to_list(FREE_NID)
2349 */
2350 ne = __lookup_nat_cache(nm_i, nid);
2351 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2352 nat_get_blkaddr(ne) != NULL_ADDR))
2353 goto err_out;
2354
2355 e = __lookup_free_nid_list(nm_i, nid);
2356 if (e) {
2357 if (e->state == FREE_NID)
2358 ret = true;
2359 goto err_out;
2360 }
2361 }
2362 ret = true;
2363 err = __insert_free_nid(sbi, i);
2364 err_out:
2365 if (update) {
2366 update_free_nid_bitmap(sbi, nid, ret, build);
2367 if (!build)
2368 nm_i->available_nids++;
2369 }
2370 spin_unlock(&nm_i->nid_list_lock);
2371 radix_tree_preload_end();
2372
2373 if (err)
2374 kmem_cache_free(free_nid_slab, i);
2375 return ret;
2376 }
2377
remove_free_nid(struct f2fs_sb_info * sbi,nid_t nid)2378 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2379 {
2380 struct f2fs_nm_info *nm_i = NM_I(sbi);
2381 struct free_nid *i;
2382 bool need_free = false;
2383
2384 spin_lock(&nm_i->nid_list_lock);
2385 i = __lookup_free_nid_list(nm_i, nid);
2386 if (i && i->state == FREE_NID) {
2387 __remove_free_nid(sbi, i, FREE_NID);
2388 need_free = true;
2389 }
2390 spin_unlock(&nm_i->nid_list_lock);
2391
2392 if (need_free)
2393 kmem_cache_free(free_nid_slab, i);
2394 }
2395
scan_nat_page(struct f2fs_sb_info * sbi,struct page * nat_page,nid_t start_nid)2396 static int scan_nat_page(struct f2fs_sb_info *sbi,
2397 struct page *nat_page, nid_t start_nid)
2398 {
2399 struct f2fs_nm_info *nm_i = NM_I(sbi);
2400 struct f2fs_nat_block *nat_blk = page_address(nat_page);
2401 block_t blk_addr;
2402 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2403 int i;
2404
2405 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2406
2407 i = start_nid % NAT_ENTRY_PER_BLOCK;
2408
2409 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2410 if (unlikely(start_nid >= nm_i->max_nid))
2411 break;
2412
2413 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2414
2415 if (blk_addr == NEW_ADDR)
2416 return -EFSCORRUPTED;
2417
2418 if (blk_addr == NULL_ADDR) {
2419 add_free_nid(sbi, start_nid, true, true);
2420 } else {
2421 spin_lock(&NM_I(sbi)->nid_list_lock);
2422 update_free_nid_bitmap(sbi, start_nid, false, true);
2423 spin_unlock(&NM_I(sbi)->nid_list_lock);
2424 }
2425 }
2426
2427 return 0;
2428 }
2429
scan_curseg_cache(struct f2fs_sb_info * sbi)2430 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2431 {
2432 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2433 struct f2fs_journal *journal = curseg->journal;
2434 int i;
2435
2436 down_read(&curseg->journal_rwsem);
2437 for (i = 0; i < nats_in_cursum(journal); i++) {
2438 block_t addr;
2439 nid_t nid;
2440
2441 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2442 nid = le32_to_cpu(nid_in_journal(journal, i));
2443 if (addr == NULL_ADDR)
2444 add_free_nid(sbi, nid, true, false);
2445 else
2446 remove_free_nid(sbi, nid);
2447 }
2448 up_read(&curseg->journal_rwsem);
2449 }
2450
scan_free_nid_bits(struct f2fs_sb_info * sbi)2451 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2452 {
2453 struct f2fs_nm_info *nm_i = NM_I(sbi);
2454 unsigned int i, idx;
2455 nid_t nid;
2456
2457 f2fs_down_read(&nm_i->nat_tree_lock);
2458
2459 for (i = 0; i < nm_i->nat_blocks; i++) {
2460 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2461 continue;
2462 if (!nm_i->free_nid_count[i])
2463 continue;
2464 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2465 idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2466 NAT_ENTRY_PER_BLOCK, idx);
2467 if (idx >= NAT_ENTRY_PER_BLOCK)
2468 break;
2469
2470 nid = i * NAT_ENTRY_PER_BLOCK + idx;
2471 add_free_nid(sbi, nid, true, false);
2472
2473 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2474 goto out;
2475 }
2476 }
2477 out:
2478 scan_curseg_cache(sbi);
2479
2480 f2fs_up_read(&nm_i->nat_tree_lock);
2481 }
2482
__f2fs_build_free_nids(struct f2fs_sb_info * sbi,bool sync,bool mount)2483 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2484 bool sync, bool mount)
2485 {
2486 struct f2fs_nm_info *nm_i = NM_I(sbi);
2487 int i = 0, ret;
2488 nid_t nid = nm_i->next_scan_nid;
2489
2490 if (unlikely(nid >= nm_i->max_nid))
2491 nid = 0;
2492
2493 if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2494 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2495
2496 /* Enough entries */
2497 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2498 return 0;
2499
2500 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2501 return 0;
2502
2503 if (!mount) {
2504 /* try to find free nids in free_nid_bitmap */
2505 scan_free_nid_bits(sbi);
2506
2507 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2508 return 0;
2509 }
2510
2511 /* readahead nat pages to be scanned */
2512 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2513 META_NAT, true);
2514
2515 f2fs_down_read(&nm_i->nat_tree_lock);
2516
2517 while (1) {
2518 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2519 nm_i->nat_block_bitmap)) {
2520 struct page *page = get_current_nat_page(sbi, nid);
2521
2522 if (IS_ERR(page)) {
2523 ret = PTR_ERR(page);
2524 } else {
2525 ret = scan_nat_page(sbi, page, nid);
2526 f2fs_put_page(page, 1);
2527 }
2528
2529 if (ret) {
2530 f2fs_up_read(&nm_i->nat_tree_lock);
2531
2532 if (ret == -EFSCORRUPTED) {
2533 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2534 set_sbi_flag(sbi, SBI_NEED_FSCK);
2535 f2fs_handle_error(sbi,
2536 ERROR_INCONSISTENT_NAT);
2537 }
2538
2539 return ret;
2540 }
2541 }
2542
2543 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2544 if (unlikely(nid >= nm_i->max_nid))
2545 nid = 0;
2546
2547 if (++i >= FREE_NID_PAGES)
2548 break;
2549 }
2550
2551 /* go to the next free nat pages to find free nids abundantly */
2552 nm_i->next_scan_nid = nid;
2553
2554 /* find free nids from current sum_pages */
2555 scan_curseg_cache(sbi);
2556
2557 f2fs_up_read(&nm_i->nat_tree_lock);
2558
2559 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2560 nm_i->ra_nid_pages, META_NAT, false);
2561
2562 return 0;
2563 }
2564
f2fs_build_free_nids(struct f2fs_sb_info * sbi,bool sync,bool mount)2565 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2566 {
2567 int ret;
2568
2569 mutex_lock(&NM_I(sbi)->build_lock);
2570 ret = __f2fs_build_free_nids(sbi, sync, mount);
2571 mutex_unlock(&NM_I(sbi)->build_lock);
2572
2573 return ret;
2574 }
2575
2576 /*
2577 * If this function returns success, caller can obtain a new nid
2578 * from second parameter of this function.
2579 * The returned nid could be used ino as well as nid when inode is created.
2580 */
f2fs_alloc_nid(struct f2fs_sb_info * sbi,nid_t * nid)2581 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2582 {
2583 struct f2fs_nm_info *nm_i = NM_I(sbi);
2584 struct free_nid *i = NULL;
2585 retry:
2586 if (time_to_inject(sbi, FAULT_ALLOC_NID))
2587 return false;
2588
2589 spin_lock(&nm_i->nid_list_lock);
2590
2591 if (unlikely(nm_i->available_nids == 0)) {
2592 spin_unlock(&nm_i->nid_list_lock);
2593 return false;
2594 }
2595
2596 /* We should not use stale free nids created by f2fs_build_free_nids */
2597 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2598 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2599 i = list_first_entry(&nm_i->free_nid_list,
2600 struct free_nid, list);
2601 *nid = i->nid;
2602
2603 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2604 nm_i->available_nids--;
2605
2606 update_free_nid_bitmap(sbi, *nid, false, false);
2607
2608 spin_unlock(&nm_i->nid_list_lock);
2609 return true;
2610 }
2611 spin_unlock(&nm_i->nid_list_lock);
2612
2613 /* Let's scan nat pages and its caches to get free nids */
2614 if (!f2fs_build_free_nids(sbi, true, false))
2615 goto retry;
2616 return false;
2617 }
2618
2619 /*
2620 * f2fs_alloc_nid() should be called prior to this function.
2621 */
f2fs_alloc_nid_done(struct f2fs_sb_info * sbi,nid_t nid)2622 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2623 {
2624 struct f2fs_nm_info *nm_i = NM_I(sbi);
2625 struct free_nid *i;
2626
2627 spin_lock(&nm_i->nid_list_lock);
2628 i = __lookup_free_nid_list(nm_i, nid);
2629 f2fs_bug_on(sbi, !i);
2630 __remove_free_nid(sbi, i, PREALLOC_NID);
2631 spin_unlock(&nm_i->nid_list_lock);
2632
2633 kmem_cache_free(free_nid_slab, i);
2634 }
2635
2636 /*
2637 * f2fs_alloc_nid() should be called prior to this function.
2638 */
f2fs_alloc_nid_failed(struct f2fs_sb_info * sbi,nid_t nid)2639 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2640 {
2641 struct f2fs_nm_info *nm_i = NM_I(sbi);
2642 struct free_nid *i;
2643 bool need_free = false;
2644
2645 if (!nid)
2646 return;
2647
2648 spin_lock(&nm_i->nid_list_lock);
2649 i = __lookup_free_nid_list(nm_i, nid);
2650 f2fs_bug_on(sbi, !i);
2651
2652 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2653 __remove_free_nid(sbi, i, PREALLOC_NID);
2654 need_free = true;
2655 } else {
2656 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2657 }
2658
2659 nm_i->available_nids++;
2660
2661 update_free_nid_bitmap(sbi, nid, true, false);
2662
2663 spin_unlock(&nm_i->nid_list_lock);
2664
2665 if (need_free)
2666 kmem_cache_free(free_nid_slab, i);
2667 }
2668
f2fs_try_to_free_nids(struct f2fs_sb_info * sbi,int nr_shrink)2669 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2670 {
2671 struct f2fs_nm_info *nm_i = NM_I(sbi);
2672 int nr = nr_shrink;
2673
2674 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2675 return 0;
2676
2677 if (!mutex_trylock(&nm_i->build_lock))
2678 return 0;
2679
2680 while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2681 struct free_nid *i, *next;
2682 unsigned int batch = SHRINK_NID_BATCH_SIZE;
2683
2684 spin_lock(&nm_i->nid_list_lock);
2685 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2686 if (!nr_shrink || !batch ||
2687 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2688 break;
2689 __remove_free_nid(sbi, i, FREE_NID);
2690 kmem_cache_free(free_nid_slab, i);
2691 nr_shrink--;
2692 batch--;
2693 }
2694 spin_unlock(&nm_i->nid_list_lock);
2695 }
2696
2697 mutex_unlock(&nm_i->build_lock);
2698
2699 return nr - nr_shrink;
2700 }
2701
f2fs_recover_inline_xattr(struct inode * inode,struct page * page)2702 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2703 {
2704 void *src_addr, *dst_addr;
2705 size_t inline_size;
2706 struct page *ipage;
2707 struct f2fs_inode *ri;
2708
2709 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2710 if (IS_ERR(ipage))
2711 return PTR_ERR(ipage);
2712
2713 ri = F2FS_INODE(page);
2714 if (ri->i_inline & F2FS_INLINE_XATTR) {
2715 if (!f2fs_has_inline_xattr(inode)) {
2716 set_inode_flag(inode, FI_INLINE_XATTR);
2717 stat_inc_inline_xattr(inode);
2718 }
2719 } else {
2720 if (f2fs_has_inline_xattr(inode)) {
2721 stat_dec_inline_xattr(inode);
2722 clear_inode_flag(inode, FI_INLINE_XATTR);
2723 }
2724 goto update_inode;
2725 }
2726
2727 dst_addr = inline_xattr_addr(inode, ipage);
2728 src_addr = inline_xattr_addr(inode, page);
2729 inline_size = inline_xattr_size(inode);
2730
2731 f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2732 memcpy(dst_addr, src_addr, inline_size);
2733 update_inode:
2734 f2fs_update_inode(inode, ipage);
2735 f2fs_put_page(ipage, 1);
2736 return 0;
2737 }
2738
f2fs_recover_xattr_data(struct inode * inode,struct page * page)2739 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2740 {
2741 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2742 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2743 nid_t new_xnid;
2744 struct dnode_of_data dn;
2745 struct node_info ni;
2746 struct page *xpage;
2747 int err;
2748
2749 if (!prev_xnid)
2750 goto recover_xnid;
2751
2752 /* 1: invalidate the previous xattr nid */
2753 err = f2fs_get_node_info(sbi, prev_xnid, &ni, false);
2754 if (err)
2755 return err;
2756
2757 f2fs_invalidate_blocks(sbi, ni.blk_addr);
2758 dec_valid_node_count(sbi, inode, false);
2759 set_node_addr(sbi, &ni, NULL_ADDR, false);
2760
2761 recover_xnid:
2762 /* 2: update xattr nid in inode */
2763 if (!f2fs_alloc_nid(sbi, &new_xnid))
2764 return -ENOSPC;
2765
2766 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2767 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2768 if (IS_ERR(xpage)) {
2769 f2fs_alloc_nid_failed(sbi, new_xnid);
2770 return PTR_ERR(xpage);
2771 }
2772
2773 f2fs_alloc_nid_done(sbi, new_xnid);
2774 f2fs_update_inode_page(inode);
2775
2776 /* 3: update and set xattr node page dirty */
2777 if (page) {
2778 memcpy(F2FS_NODE(xpage), F2FS_NODE(page),
2779 VALID_XATTR_BLOCK_SIZE);
2780 set_page_dirty(xpage);
2781 }
2782 f2fs_put_page(xpage, 1);
2783
2784 return 0;
2785 }
2786
f2fs_recover_inode_page(struct f2fs_sb_info * sbi,struct page * page)2787 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2788 {
2789 struct f2fs_inode *src, *dst;
2790 nid_t ino = ino_of_node(page);
2791 struct node_info old_ni, new_ni;
2792 struct page *ipage;
2793 int err;
2794
2795 err = f2fs_get_node_info(sbi, ino, &old_ni, false);
2796 if (err)
2797 return err;
2798
2799 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2800 return -EINVAL;
2801 retry:
2802 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2803 if (!ipage) {
2804 memalloc_retry_wait(GFP_NOFS);
2805 goto retry;
2806 }
2807
2808 /* Should not use this inode from free nid list */
2809 remove_free_nid(sbi, ino);
2810
2811 if (!PageUptodate(ipage))
2812 SetPageUptodate(ipage);
2813 fill_node_footer(ipage, ino, ino, 0, true);
2814 set_cold_node(ipage, false);
2815
2816 src = F2FS_INODE(page);
2817 dst = F2FS_INODE(ipage);
2818
2819 memcpy(dst, src, offsetof(struct f2fs_inode, i_ext));
2820 dst->i_size = 0;
2821 dst->i_blocks = cpu_to_le64(1);
2822 dst->i_links = cpu_to_le32(1);
2823 dst->i_xattr_nid = 0;
2824 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2825 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2826 dst->i_extra_isize = src->i_extra_isize;
2827
2828 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2829 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2830 i_inline_xattr_size))
2831 dst->i_inline_xattr_size = src->i_inline_xattr_size;
2832
2833 if (f2fs_sb_has_project_quota(sbi) &&
2834 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2835 i_projid))
2836 dst->i_projid = src->i_projid;
2837
2838 if (f2fs_sb_has_inode_crtime(sbi) &&
2839 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2840 i_crtime_nsec)) {
2841 dst->i_crtime = src->i_crtime;
2842 dst->i_crtime_nsec = src->i_crtime_nsec;
2843 }
2844 }
2845
2846 new_ni = old_ni;
2847 new_ni.ino = ino;
2848
2849 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2850 WARN_ON(1);
2851 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2852 inc_valid_inode_count(sbi);
2853 set_page_dirty(ipage);
2854 f2fs_put_page(ipage, 1);
2855 return 0;
2856 }
2857
f2fs_restore_node_summary(struct f2fs_sb_info * sbi,unsigned int segno,struct f2fs_summary_block * sum)2858 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2859 unsigned int segno, struct f2fs_summary_block *sum)
2860 {
2861 struct f2fs_node *rn;
2862 struct f2fs_summary *sum_entry;
2863 block_t addr;
2864 int i, idx, last_offset, nrpages;
2865
2866 /* scan the node segment */
2867 last_offset = BLKS_PER_SEG(sbi);
2868 addr = START_BLOCK(sbi, segno);
2869 sum_entry = &sum->entries[0];
2870
2871 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2872 nrpages = bio_max_segs(last_offset - i);
2873
2874 /* readahead node pages */
2875 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2876
2877 for (idx = addr; idx < addr + nrpages; idx++) {
2878 struct page *page = f2fs_get_tmp_page(sbi, idx);
2879
2880 if (IS_ERR(page))
2881 return PTR_ERR(page);
2882
2883 rn = F2FS_NODE(page);
2884 sum_entry->nid = rn->footer.nid;
2885 sum_entry->version = 0;
2886 sum_entry->ofs_in_node = 0;
2887 sum_entry++;
2888 f2fs_put_page(page, 1);
2889 }
2890
2891 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2892 addr + nrpages);
2893 }
2894 return 0;
2895 }
2896
remove_nats_in_journal(struct f2fs_sb_info * sbi)2897 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2898 {
2899 struct f2fs_nm_info *nm_i = NM_I(sbi);
2900 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2901 struct f2fs_journal *journal = curseg->journal;
2902 int i;
2903
2904 down_write(&curseg->journal_rwsem);
2905 for (i = 0; i < nats_in_cursum(journal); i++) {
2906 struct nat_entry *ne;
2907 struct f2fs_nat_entry raw_ne;
2908 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2909
2910 if (f2fs_check_nid_range(sbi, nid))
2911 continue;
2912
2913 raw_ne = nat_in_journal(journal, i);
2914
2915 ne = __lookup_nat_cache(nm_i, nid);
2916 if (!ne) {
2917 ne = __alloc_nat_entry(sbi, nid, true);
2918 __init_nat_entry(nm_i, ne, &raw_ne, true);
2919 }
2920
2921 /*
2922 * if a free nat in journal has not been used after last
2923 * checkpoint, we should remove it from available nids,
2924 * since later we will add it again.
2925 */
2926 if (!get_nat_flag(ne, IS_DIRTY) &&
2927 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2928 spin_lock(&nm_i->nid_list_lock);
2929 nm_i->available_nids--;
2930 spin_unlock(&nm_i->nid_list_lock);
2931 }
2932
2933 __set_nat_cache_dirty(nm_i, ne);
2934 }
2935 update_nats_in_cursum(journal, -i);
2936 up_write(&curseg->journal_rwsem);
2937 }
2938
__adjust_nat_entry_set(struct nat_entry_set * nes,struct list_head * head,int max)2939 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2940 struct list_head *head, int max)
2941 {
2942 struct nat_entry_set *cur;
2943
2944 if (nes->entry_cnt >= max)
2945 goto add_out;
2946
2947 list_for_each_entry(cur, head, set_list) {
2948 if (cur->entry_cnt >= nes->entry_cnt) {
2949 list_add(&nes->set_list, cur->set_list.prev);
2950 return;
2951 }
2952 }
2953 add_out:
2954 list_add_tail(&nes->set_list, head);
2955 }
2956
__update_nat_bits(struct f2fs_nm_info * nm_i,unsigned int nat_ofs,unsigned int valid)2957 static void __update_nat_bits(struct f2fs_nm_info *nm_i, unsigned int nat_ofs,
2958 unsigned int valid)
2959 {
2960 if (valid == 0) {
2961 __set_bit_le(nat_ofs, nm_i->empty_nat_bits);
2962 __clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2963 return;
2964 }
2965
2966 __clear_bit_le(nat_ofs, nm_i->empty_nat_bits);
2967 if (valid == NAT_ENTRY_PER_BLOCK)
2968 __set_bit_le(nat_ofs, nm_i->full_nat_bits);
2969 else
2970 __clear_bit_le(nat_ofs, nm_i->full_nat_bits);
2971 }
2972
update_nat_bits(struct f2fs_sb_info * sbi,nid_t start_nid,struct page * page)2973 static void update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2974 struct page *page)
2975 {
2976 struct f2fs_nm_info *nm_i = NM_I(sbi);
2977 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2978 struct f2fs_nat_block *nat_blk = page_address(page);
2979 int valid = 0;
2980 int i = 0;
2981
2982 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
2983 return;
2984
2985 if (nat_index == 0) {
2986 valid = 1;
2987 i = 1;
2988 }
2989 for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2990 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2991 valid++;
2992 }
2993
2994 __update_nat_bits(nm_i, nat_index, valid);
2995 }
2996
f2fs_enable_nat_bits(struct f2fs_sb_info * sbi)2997 void f2fs_enable_nat_bits(struct f2fs_sb_info *sbi)
2998 {
2999 struct f2fs_nm_info *nm_i = NM_I(sbi);
3000 unsigned int nat_ofs;
3001
3002 f2fs_down_read(&nm_i->nat_tree_lock);
3003
3004 for (nat_ofs = 0; nat_ofs < nm_i->nat_blocks; nat_ofs++) {
3005 unsigned int valid = 0, nid_ofs = 0;
3006
3007 /* handle nid zero due to it should never be used */
3008 if (unlikely(nat_ofs == 0)) {
3009 valid = 1;
3010 nid_ofs = 1;
3011 }
3012
3013 for (; nid_ofs < NAT_ENTRY_PER_BLOCK; nid_ofs++) {
3014 if (!test_bit_le(nid_ofs,
3015 nm_i->free_nid_bitmap[nat_ofs]))
3016 valid++;
3017 }
3018
3019 __update_nat_bits(nm_i, nat_ofs, valid);
3020 }
3021
3022 f2fs_up_read(&nm_i->nat_tree_lock);
3023 }
3024
__flush_nat_entry_set(struct f2fs_sb_info * sbi,struct nat_entry_set * set,struct cp_control * cpc)3025 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
3026 struct nat_entry_set *set, struct cp_control *cpc)
3027 {
3028 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3029 struct f2fs_journal *journal = curseg->journal;
3030 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
3031 bool to_journal = true;
3032 struct f2fs_nat_block *nat_blk;
3033 struct nat_entry *ne, *cur;
3034 struct page *page = NULL;
3035
3036 /*
3037 * there are two steps to flush nat entries:
3038 * #1, flush nat entries to journal in current hot data summary block.
3039 * #2, flush nat entries to nat page.
3040 */
3041 if ((cpc->reason & CP_UMOUNT) ||
3042 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
3043 to_journal = false;
3044
3045 if (to_journal) {
3046 down_write(&curseg->journal_rwsem);
3047 } else {
3048 page = get_next_nat_page(sbi, start_nid);
3049 if (IS_ERR(page))
3050 return PTR_ERR(page);
3051
3052 nat_blk = page_address(page);
3053 f2fs_bug_on(sbi, !nat_blk);
3054 }
3055
3056 /* flush dirty nats in nat entry set */
3057 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
3058 struct f2fs_nat_entry *raw_ne;
3059 nid_t nid = nat_get_nid(ne);
3060 int offset;
3061
3062 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
3063
3064 if (to_journal) {
3065 offset = f2fs_lookup_journal_in_cursum(journal,
3066 NAT_JOURNAL, nid, 1);
3067 f2fs_bug_on(sbi, offset < 0);
3068 raw_ne = &nat_in_journal(journal, offset);
3069 nid_in_journal(journal, offset) = cpu_to_le32(nid);
3070 } else {
3071 raw_ne = &nat_blk->entries[nid - start_nid];
3072 }
3073 raw_nat_from_node_info(raw_ne, &ne->ni);
3074 nat_reset_flag(ne);
3075 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
3076 if (nat_get_blkaddr(ne) == NULL_ADDR) {
3077 add_free_nid(sbi, nid, false, true);
3078 } else {
3079 spin_lock(&NM_I(sbi)->nid_list_lock);
3080 update_free_nid_bitmap(sbi, nid, false, false);
3081 spin_unlock(&NM_I(sbi)->nid_list_lock);
3082 }
3083 }
3084
3085 if (to_journal) {
3086 up_write(&curseg->journal_rwsem);
3087 } else {
3088 update_nat_bits(sbi, start_nid, page);
3089 f2fs_put_page(page, 1);
3090 }
3091
3092 /* Allow dirty nats by node block allocation in write_begin */
3093 if (!set->entry_cnt) {
3094 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
3095 kmem_cache_free(nat_entry_set_slab, set);
3096 }
3097 return 0;
3098 }
3099
3100 /*
3101 * This function is called during the checkpointing process.
3102 */
f2fs_flush_nat_entries(struct f2fs_sb_info * sbi,struct cp_control * cpc)3103 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
3104 {
3105 struct f2fs_nm_info *nm_i = NM_I(sbi);
3106 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
3107 struct f2fs_journal *journal = curseg->journal;
3108 struct nat_entry_set *setvec[NAT_VEC_SIZE];
3109 struct nat_entry_set *set, *tmp;
3110 unsigned int found;
3111 nid_t set_idx = 0;
3112 LIST_HEAD(sets);
3113 int err = 0;
3114
3115 /*
3116 * during unmount, let's flush nat_bits before checking
3117 * nat_cnt[DIRTY_NAT].
3118 */
3119 if (cpc->reason & CP_UMOUNT) {
3120 f2fs_down_write(&nm_i->nat_tree_lock);
3121 remove_nats_in_journal(sbi);
3122 f2fs_up_write(&nm_i->nat_tree_lock);
3123 }
3124
3125 if (!nm_i->nat_cnt[DIRTY_NAT])
3126 return 0;
3127
3128 f2fs_down_write(&nm_i->nat_tree_lock);
3129
3130 /*
3131 * if there are no enough space in journal to store dirty nat
3132 * entries, remove all entries from journal and merge them
3133 * into nat entry set.
3134 */
3135 if (cpc->reason & CP_UMOUNT ||
3136 !__has_cursum_space(journal,
3137 nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
3138 remove_nats_in_journal(sbi);
3139
3140 while ((found = __gang_lookup_nat_set(nm_i,
3141 set_idx, NAT_VEC_SIZE, setvec))) {
3142 unsigned idx;
3143
3144 set_idx = setvec[found - 1]->set + 1;
3145 for (idx = 0; idx < found; idx++)
3146 __adjust_nat_entry_set(setvec[idx], &sets,
3147 MAX_NAT_JENTRIES(journal));
3148 }
3149
3150 /* flush dirty nats in nat entry set */
3151 list_for_each_entry_safe(set, tmp, &sets, set_list) {
3152 err = __flush_nat_entry_set(sbi, set, cpc);
3153 if (err)
3154 break;
3155 }
3156
3157 f2fs_up_write(&nm_i->nat_tree_lock);
3158 /* Allow dirty nats by node block allocation in write_begin */
3159
3160 return err;
3161 }
3162
__get_nat_bitmaps(struct f2fs_sb_info * sbi)3163 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3164 {
3165 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3166 struct f2fs_nm_info *nm_i = NM_I(sbi);
3167 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3168 unsigned int i;
3169 __u64 cp_ver = cur_cp_version(ckpt);
3170 block_t nat_bits_addr;
3171
3172 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3173 nm_i->nat_bits = f2fs_kvzalloc(sbi,
3174 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3175 if (!nm_i->nat_bits)
3176 return -ENOMEM;
3177
3178 nm_i->full_nat_bits = nm_i->nat_bits + 8;
3179 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3180
3181 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3182 return 0;
3183
3184 nat_bits_addr = __start_cp_addr(sbi) + BLKS_PER_SEG(sbi) -
3185 nm_i->nat_bits_blocks;
3186 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3187 struct page *page;
3188
3189 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3190 if (IS_ERR(page))
3191 return PTR_ERR(page);
3192
3193 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3194 page_address(page), F2FS_BLKSIZE);
3195 f2fs_put_page(page, 1);
3196 }
3197
3198 cp_ver |= (cur_cp_crc(ckpt) << 32);
3199 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3200 clear_ckpt_flags(sbi, CP_NAT_BITS_FLAG);
3201 f2fs_notice(sbi, "Disable nat_bits due to incorrect cp_ver (%llu, %llu)",
3202 cp_ver, le64_to_cpu(*(__le64 *)nm_i->nat_bits));
3203 return 0;
3204 }
3205
3206 f2fs_notice(sbi, "Found nat_bits in checkpoint");
3207 return 0;
3208 }
3209
load_free_nid_bitmap(struct f2fs_sb_info * sbi)3210 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3211 {
3212 struct f2fs_nm_info *nm_i = NM_I(sbi);
3213 unsigned int i = 0;
3214 nid_t nid, last_nid;
3215
3216 if (!is_set_ckpt_flags(sbi, CP_NAT_BITS_FLAG))
3217 return;
3218
3219 for (i = 0; i < nm_i->nat_blocks; i++) {
3220 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3221 if (i >= nm_i->nat_blocks)
3222 break;
3223
3224 __set_bit_le(i, nm_i->nat_block_bitmap);
3225
3226 nid = i * NAT_ENTRY_PER_BLOCK;
3227 last_nid = nid + NAT_ENTRY_PER_BLOCK;
3228
3229 spin_lock(&NM_I(sbi)->nid_list_lock);
3230 for (; nid < last_nid; nid++)
3231 update_free_nid_bitmap(sbi, nid, true, true);
3232 spin_unlock(&NM_I(sbi)->nid_list_lock);
3233 }
3234
3235 for (i = 0; i < nm_i->nat_blocks; i++) {
3236 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3237 if (i >= nm_i->nat_blocks)
3238 break;
3239
3240 __set_bit_le(i, nm_i->nat_block_bitmap);
3241 }
3242 }
3243
init_node_manager(struct f2fs_sb_info * sbi)3244 static int init_node_manager(struct f2fs_sb_info *sbi)
3245 {
3246 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3247 struct f2fs_nm_info *nm_i = NM_I(sbi);
3248 unsigned char *version_bitmap;
3249 unsigned int nat_segs;
3250 int err;
3251
3252 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3253
3254 /* segment_count_nat includes pair segment so divide to 2. */
3255 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3256 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3257 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3258
3259 /* not used nids: 0, node, meta, (and root counted as valid node) */
3260 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3261 F2FS_RESERVED_NODE_NUM;
3262 nm_i->nid_cnt[FREE_NID] = 0;
3263 nm_i->nid_cnt[PREALLOC_NID] = 0;
3264 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3265 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3266 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3267 nm_i->max_rf_node_blocks = DEF_RF_NODE_BLOCKS;
3268
3269 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3270 INIT_LIST_HEAD(&nm_i->free_nid_list);
3271 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3272 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3273 INIT_LIST_HEAD(&nm_i->nat_entries);
3274 spin_lock_init(&nm_i->nat_list_lock);
3275
3276 mutex_init(&nm_i->build_lock);
3277 spin_lock_init(&nm_i->nid_list_lock);
3278 init_f2fs_rwsem(&nm_i->nat_tree_lock);
3279
3280 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3281 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3282 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3283 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3284 GFP_KERNEL);
3285 if (!nm_i->nat_bitmap)
3286 return -ENOMEM;
3287
3288 err = __get_nat_bitmaps(sbi);
3289 if (err)
3290 return err;
3291
3292 #ifdef CONFIG_F2FS_CHECK_FS
3293 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3294 GFP_KERNEL);
3295 if (!nm_i->nat_bitmap_mir)
3296 return -ENOMEM;
3297 #endif
3298
3299 return 0;
3300 }
3301
init_free_nid_cache(struct f2fs_sb_info * sbi)3302 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3303 {
3304 struct f2fs_nm_info *nm_i = NM_I(sbi);
3305 int i;
3306
3307 nm_i->free_nid_bitmap =
3308 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3309 nm_i->nat_blocks),
3310 GFP_KERNEL);
3311 if (!nm_i->free_nid_bitmap)
3312 return -ENOMEM;
3313
3314 for (i = 0; i < nm_i->nat_blocks; i++) {
3315 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3316 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3317 if (!nm_i->free_nid_bitmap[i])
3318 return -ENOMEM;
3319 }
3320
3321 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3322 GFP_KERNEL);
3323 if (!nm_i->nat_block_bitmap)
3324 return -ENOMEM;
3325
3326 nm_i->free_nid_count =
3327 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3328 nm_i->nat_blocks),
3329 GFP_KERNEL);
3330 if (!nm_i->free_nid_count)
3331 return -ENOMEM;
3332 return 0;
3333 }
3334
f2fs_build_node_manager(struct f2fs_sb_info * sbi)3335 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3336 {
3337 int err;
3338
3339 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3340 GFP_KERNEL);
3341 if (!sbi->nm_info)
3342 return -ENOMEM;
3343
3344 err = init_node_manager(sbi);
3345 if (err)
3346 return err;
3347
3348 err = init_free_nid_cache(sbi);
3349 if (err)
3350 return err;
3351
3352 /* load free nid status from nat_bits table */
3353 load_free_nid_bitmap(sbi);
3354
3355 return f2fs_build_free_nids(sbi, true, true);
3356 }
3357
f2fs_destroy_node_manager(struct f2fs_sb_info * sbi)3358 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3359 {
3360 struct f2fs_nm_info *nm_i = NM_I(sbi);
3361 struct free_nid *i, *next_i;
3362 void *vec[NAT_VEC_SIZE];
3363 struct nat_entry **natvec = (struct nat_entry **)vec;
3364 struct nat_entry_set **setvec = (struct nat_entry_set **)vec;
3365 nid_t nid = 0;
3366 unsigned int found;
3367
3368 if (!nm_i)
3369 return;
3370
3371 /* destroy free nid list */
3372 spin_lock(&nm_i->nid_list_lock);
3373 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3374 __remove_free_nid(sbi, i, FREE_NID);
3375 spin_unlock(&nm_i->nid_list_lock);
3376 kmem_cache_free(free_nid_slab, i);
3377 spin_lock(&nm_i->nid_list_lock);
3378 }
3379 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3380 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3381 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3382 spin_unlock(&nm_i->nid_list_lock);
3383
3384 /* destroy nat cache */
3385 f2fs_down_write(&nm_i->nat_tree_lock);
3386 while ((found = __gang_lookup_nat_cache(nm_i,
3387 nid, NAT_VEC_SIZE, natvec))) {
3388 unsigned idx;
3389
3390 nid = nat_get_nid(natvec[found - 1]) + 1;
3391 for (idx = 0; idx < found; idx++) {
3392 spin_lock(&nm_i->nat_list_lock);
3393 list_del(&natvec[idx]->list);
3394 spin_unlock(&nm_i->nat_list_lock);
3395
3396 __del_from_nat_cache(nm_i, natvec[idx]);
3397 }
3398 }
3399 f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3400
3401 /* destroy nat set cache */
3402 nid = 0;
3403 memset(vec, 0, sizeof(void *) * NAT_VEC_SIZE);
3404 while ((found = __gang_lookup_nat_set(nm_i,
3405 nid, NAT_VEC_SIZE, setvec))) {
3406 unsigned idx;
3407
3408 nid = setvec[found - 1]->set + 1;
3409 for (idx = 0; idx < found; idx++) {
3410 /* entry_cnt is not zero, when cp_error was occurred */
3411 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3412 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3413 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3414 }
3415 }
3416 f2fs_up_write(&nm_i->nat_tree_lock);
3417
3418 kvfree(nm_i->nat_block_bitmap);
3419 if (nm_i->free_nid_bitmap) {
3420 int i;
3421
3422 for (i = 0; i < nm_i->nat_blocks; i++)
3423 kvfree(nm_i->free_nid_bitmap[i]);
3424 kvfree(nm_i->free_nid_bitmap);
3425 }
3426 kvfree(nm_i->free_nid_count);
3427
3428 kvfree(nm_i->nat_bitmap);
3429 kvfree(nm_i->nat_bits);
3430 #ifdef CONFIG_F2FS_CHECK_FS
3431 kvfree(nm_i->nat_bitmap_mir);
3432 #endif
3433 sbi->nm_info = NULL;
3434 kfree(nm_i);
3435 }
3436
f2fs_create_node_manager_caches(void)3437 int __init f2fs_create_node_manager_caches(void)
3438 {
3439 nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3440 sizeof(struct nat_entry));
3441 if (!nat_entry_slab)
3442 goto fail;
3443
3444 free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3445 sizeof(struct free_nid));
3446 if (!free_nid_slab)
3447 goto destroy_nat_entry;
3448
3449 nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3450 sizeof(struct nat_entry_set));
3451 if (!nat_entry_set_slab)
3452 goto destroy_free_nid;
3453
3454 fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3455 sizeof(struct fsync_node_entry));
3456 if (!fsync_node_entry_slab)
3457 goto destroy_nat_entry_set;
3458 return 0;
3459
3460 destroy_nat_entry_set:
3461 kmem_cache_destroy(nat_entry_set_slab);
3462 destroy_free_nid:
3463 kmem_cache_destroy(free_nid_slab);
3464 destroy_nat_entry:
3465 kmem_cache_destroy(nat_entry_slab);
3466 fail:
3467 return -ENOMEM;
3468 }
3469
f2fs_destroy_node_manager_caches(void)3470 void f2fs_destroy_node_manager_caches(void)
3471 {
3472 kmem_cache_destroy(fsync_node_entry_slab);
3473 kmem_cache_destroy(nat_entry_set_slab);
3474 kmem_cache_destroy(free_nid_slab);
3475 kmem_cache_destroy(nat_entry_slab);
3476 }
3477