1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * fs/f2fs/node.c
4 *
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
7 */
8 #include <linux/fs.h>
9 #include <linux/f2fs_fs.h>
10 #include <linux/mpage.h>
11 #include <linux/backing-dev.h>
12 #include <linux/blkdev.h>
13 #include <linux/pagevec.h>
14 #include <linux/swap.h>
15
16 #include "f2fs.h"
17 #include "node.h"
18 #include "segment.h"
19 #include "xattr.h"
20 #include "trace.h"
21 #include <trace/events/f2fs.h>
22
23 #define on_f2fs_build_free_nids(nmi) mutex_is_locked(&(nm_i)->build_lock)
24
25 static struct kmem_cache *nat_entry_slab;
26 static struct kmem_cache *free_nid_slab;
27 static struct kmem_cache *nat_entry_set_slab;
28 static struct kmem_cache *fsync_node_entry_slab;
29
30 /*
31 * Check whether the given nid is within node id range.
32 */
f2fs_check_nid_range(struct f2fs_sb_info * sbi,nid_t nid)33 int f2fs_check_nid_range(struct f2fs_sb_info *sbi, nid_t nid)
34 {
35 if (unlikely(nid < F2FS_ROOT_INO(sbi) || nid >= NM_I(sbi)->max_nid)) {
36 set_sbi_flag(sbi, SBI_NEED_FSCK);
37 f2fs_warn(sbi, "%s: out-of-range nid=%x, run fsck to fix.",
38 __func__, nid);
39 return -EFSCORRUPTED;
40 }
41 return 0;
42 }
43
f2fs_available_free_memory(struct f2fs_sb_info * sbi,int type)44 bool f2fs_available_free_memory(struct f2fs_sb_info *sbi, int type)
45 {
46 struct f2fs_nm_info *nm_i = NM_I(sbi);
47 struct sysinfo val;
48 unsigned long avail_ram;
49 unsigned long mem_size = 0;
50 bool res = false;
51
52 si_meminfo(&val);
53
54 /* only uses low memory */
55 avail_ram = val.totalram - val.totalhigh;
56
57 /*
58 * give 25%, 25%, 50%, 50%, 50% memory for each components respectively
59 */
60 if (type == FREE_NIDS) {
61 mem_size = (nm_i->nid_cnt[FREE_NID] *
62 sizeof(struct free_nid)) >> PAGE_SHIFT;
63 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
64 } else if (type == NAT_ENTRIES) {
65 mem_size = (nm_i->nat_cnt[TOTAL_NAT] *
66 sizeof(struct nat_entry)) >> PAGE_SHIFT;
67 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 2);
68 if (excess_cached_nats(sbi))
69 res = false;
70 } else if (type == DIRTY_DENTS) {
71 if (sbi->sb->s_bdi->wb.dirty_exceeded)
72 return false;
73 mem_size = get_pages(sbi, F2FS_DIRTY_DENTS);
74 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
75 } else if (type == INO_ENTRIES) {
76 int i;
77
78 for (i = 0; i < MAX_INO_ENTRY; i++)
79 mem_size += sbi->im[i].ino_num *
80 sizeof(struct ino_entry);
81 mem_size >>= PAGE_SHIFT;
82 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
83 } else if (type == EXTENT_CACHE) {
84 mem_size = (atomic_read(&sbi->total_ext_tree) *
85 sizeof(struct extent_tree) +
86 atomic_read(&sbi->total_ext_node) *
87 sizeof(struct extent_node)) >> PAGE_SHIFT;
88 res = mem_size < ((avail_ram * nm_i->ram_thresh / 100) >> 1);
89 } else if (type == INMEM_PAGES) {
90 /* it allows 20% / total_ram for inmemory pages */
91 mem_size = get_pages(sbi, F2FS_INMEM_PAGES);
92 res = mem_size < (val.totalram / 5);
93 } else {
94 if (!sbi->sb->s_bdi->wb.dirty_exceeded)
95 return true;
96 }
97 return res;
98 }
99
clear_node_page_dirty(struct page * page)100 static void clear_node_page_dirty(struct page *page)
101 {
102 if (PageDirty(page)) {
103 f2fs_clear_page_cache_dirty_tag(page);
104 clear_page_dirty_for_io(page);
105 dec_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
106 }
107 ClearPageUptodate(page);
108 }
109
get_current_nat_page(struct f2fs_sb_info * sbi,nid_t nid)110 static struct page *get_current_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
111 {
112 return f2fs_get_meta_page_retry(sbi, current_nat_addr(sbi, nid));
113 }
114
get_next_nat_page(struct f2fs_sb_info * sbi,nid_t nid)115 static struct page *get_next_nat_page(struct f2fs_sb_info *sbi, nid_t nid)
116 {
117 struct page *src_page;
118 struct page *dst_page;
119 pgoff_t dst_off;
120 void *src_addr;
121 void *dst_addr;
122 struct f2fs_nm_info *nm_i = NM_I(sbi);
123
124 dst_off = next_nat_addr(sbi, current_nat_addr(sbi, nid));
125
126 /* get current nat block page with lock */
127 src_page = get_current_nat_page(sbi, nid);
128 if (IS_ERR(src_page))
129 return src_page;
130 dst_page = f2fs_grab_meta_page(sbi, dst_off);
131 f2fs_bug_on(sbi, PageDirty(src_page));
132
133 src_addr = page_address(src_page);
134 dst_addr = page_address(dst_page);
135 memcpy(dst_addr, src_addr, PAGE_SIZE);
136 set_page_dirty(dst_page);
137 f2fs_put_page(src_page, 1);
138
139 set_to_next_nat(nm_i, nid);
140
141 return dst_page;
142 }
143
__alloc_nat_entry(nid_t nid,bool no_fail)144 static struct nat_entry *__alloc_nat_entry(nid_t nid, bool no_fail)
145 {
146 struct nat_entry *new;
147
148 if (no_fail)
149 new = f2fs_kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
150 else
151 new = kmem_cache_alloc(nat_entry_slab, GFP_F2FS_ZERO);
152 if (new) {
153 nat_set_nid(new, nid);
154 nat_reset_flag(new);
155 }
156 return new;
157 }
158
__free_nat_entry(struct nat_entry * e)159 static void __free_nat_entry(struct nat_entry *e)
160 {
161 kmem_cache_free(nat_entry_slab, e);
162 }
163
164 /* must be locked by nat_tree_lock */
__init_nat_entry(struct f2fs_nm_info * nm_i,struct nat_entry * ne,struct f2fs_nat_entry * raw_ne,bool no_fail)165 static struct nat_entry *__init_nat_entry(struct f2fs_nm_info *nm_i,
166 struct nat_entry *ne, struct f2fs_nat_entry *raw_ne, bool no_fail)
167 {
168 if (no_fail)
169 f2fs_radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne);
170 else if (radix_tree_insert(&nm_i->nat_root, nat_get_nid(ne), ne))
171 return NULL;
172
173 if (raw_ne)
174 node_info_from_raw_nat(&ne->ni, raw_ne);
175
176 spin_lock(&nm_i->nat_list_lock);
177 list_add_tail(&ne->list, &nm_i->nat_entries);
178 spin_unlock(&nm_i->nat_list_lock);
179
180 nm_i->nat_cnt[TOTAL_NAT]++;
181 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
182 return ne;
183 }
184
__lookup_nat_cache(struct f2fs_nm_info * nm_i,nid_t n)185 static struct nat_entry *__lookup_nat_cache(struct f2fs_nm_info *nm_i, nid_t n)
186 {
187 struct nat_entry *ne;
188
189 ne = radix_tree_lookup(&nm_i->nat_root, n);
190
191 /* for recent accessed nat entry, move it to tail of lru list */
192 if (ne && !get_nat_flag(ne, IS_DIRTY)) {
193 spin_lock(&nm_i->nat_list_lock);
194 if (!list_empty(&ne->list))
195 list_move_tail(&ne->list, &nm_i->nat_entries);
196 spin_unlock(&nm_i->nat_list_lock);
197 }
198
199 return ne;
200 }
201
__gang_lookup_nat_cache(struct f2fs_nm_info * nm_i,nid_t start,unsigned int nr,struct nat_entry ** ep)202 static unsigned int __gang_lookup_nat_cache(struct f2fs_nm_info *nm_i,
203 nid_t start, unsigned int nr, struct nat_entry **ep)
204 {
205 return radix_tree_gang_lookup(&nm_i->nat_root, (void **)ep, start, nr);
206 }
207
__del_from_nat_cache(struct f2fs_nm_info * nm_i,struct nat_entry * e)208 static void __del_from_nat_cache(struct f2fs_nm_info *nm_i, struct nat_entry *e)
209 {
210 radix_tree_delete(&nm_i->nat_root, nat_get_nid(e));
211 nm_i->nat_cnt[TOTAL_NAT]--;
212 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
213 __free_nat_entry(e);
214 }
215
__grab_nat_entry_set(struct f2fs_nm_info * nm_i,struct nat_entry * ne)216 static struct nat_entry_set *__grab_nat_entry_set(struct f2fs_nm_info *nm_i,
217 struct nat_entry *ne)
218 {
219 nid_t set = NAT_BLOCK_OFFSET(ne->ni.nid);
220 struct nat_entry_set *head;
221
222 head = radix_tree_lookup(&nm_i->nat_set_root, set);
223 if (!head) {
224 head = f2fs_kmem_cache_alloc(nat_entry_set_slab, GFP_NOFS);
225
226 INIT_LIST_HEAD(&head->entry_list);
227 INIT_LIST_HEAD(&head->set_list);
228 head->set = set;
229 head->entry_cnt = 0;
230 f2fs_radix_tree_insert(&nm_i->nat_set_root, set, head);
231 }
232 return head;
233 }
234
__set_nat_cache_dirty(struct f2fs_nm_info * nm_i,struct nat_entry * ne)235 static void __set_nat_cache_dirty(struct f2fs_nm_info *nm_i,
236 struct nat_entry *ne)
237 {
238 struct nat_entry_set *head;
239 bool new_ne = nat_get_blkaddr(ne) == NEW_ADDR;
240
241 if (!new_ne)
242 head = __grab_nat_entry_set(nm_i, ne);
243
244 /*
245 * update entry_cnt in below condition:
246 * 1. update NEW_ADDR to valid block address;
247 * 2. update old block address to new one;
248 */
249 if (!new_ne && (get_nat_flag(ne, IS_PREALLOC) ||
250 !get_nat_flag(ne, IS_DIRTY)))
251 head->entry_cnt++;
252
253 set_nat_flag(ne, IS_PREALLOC, new_ne);
254
255 if (get_nat_flag(ne, IS_DIRTY))
256 goto refresh_list;
257
258 nm_i->nat_cnt[DIRTY_NAT]++;
259 nm_i->nat_cnt[RECLAIMABLE_NAT]--;
260 set_nat_flag(ne, IS_DIRTY, true);
261 refresh_list:
262 spin_lock(&nm_i->nat_list_lock);
263 if (new_ne)
264 list_del_init(&ne->list);
265 else
266 list_move_tail(&ne->list, &head->entry_list);
267 spin_unlock(&nm_i->nat_list_lock);
268 }
269
__clear_nat_cache_dirty(struct f2fs_nm_info * nm_i,struct nat_entry_set * set,struct nat_entry * ne)270 static void __clear_nat_cache_dirty(struct f2fs_nm_info *nm_i,
271 struct nat_entry_set *set, struct nat_entry *ne)
272 {
273 spin_lock(&nm_i->nat_list_lock);
274 list_move_tail(&ne->list, &nm_i->nat_entries);
275 spin_unlock(&nm_i->nat_list_lock);
276
277 set_nat_flag(ne, IS_DIRTY, false);
278 set->entry_cnt--;
279 nm_i->nat_cnt[DIRTY_NAT]--;
280 nm_i->nat_cnt[RECLAIMABLE_NAT]++;
281 }
282
__gang_lookup_nat_set(struct f2fs_nm_info * nm_i,nid_t start,unsigned int nr,struct nat_entry_set ** ep)283 static unsigned int __gang_lookup_nat_set(struct f2fs_nm_info *nm_i,
284 nid_t start, unsigned int nr, struct nat_entry_set **ep)
285 {
286 return radix_tree_gang_lookup(&nm_i->nat_set_root, (void **)ep,
287 start, nr);
288 }
289
f2fs_in_warm_node_list(struct f2fs_sb_info * sbi,struct page * page)290 bool f2fs_in_warm_node_list(struct f2fs_sb_info *sbi, struct page *page)
291 {
292 return NODE_MAPPING(sbi) == page->mapping &&
293 IS_DNODE(page) && is_cold_node(page);
294 }
295
f2fs_init_fsync_node_info(struct f2fs_sb_info * sbi)296 void f2fs_init_fsync_node_info(struct f2fs_sb_info *sbi)
297 {
298 spin_lock_init(&sbi->fsync_node_lock);
299 INIT_LIST_HEAD(&sbi->fsync_node_list);
300 sbi->fsync_seg_id = 0;
301 sbi->fsync_node_num = 0;
302 }
303
f2fs_add_fsync_node_entry(struct f2fs_sb_info * sbi,struct page * page)304 static unsigned int f2fs_add_fsync_node_entry(struct f2fs_sb_info *sbi,
305 struct page *page)
306 {
307 struct fsync_node_entry *fn;
308 unsigned long flags;
309 unsigned int seq_id;
310
311 fn = f2fs_kmem_cache_alloc(fsync_node_entry_slab, GFP_NOFS);
312
313 get_page(page);
314 fn->page = page;
315 INIT_LIST_HEAD(&fn->list);
316
317 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
318 list_add_tail(&fn->list, &sbi->fsync_node_list);
319 fn->seq_id = sbi->fsync_seg_id++;
320 seq_id = fn->seq_id;
321 sbi->fsync_node_num++;
322 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
323
324 return seq_id;
325 }
326
f2fs_del_fsync_node_entry(struct f2fs_sb_info * sbi,struct page * page)327 void f2fs_del_fsync_node_entry(struct f2fs_sb_info *sbi, struct page *page)
328 {
329 struct fsync_node_entry *fn;
330 unsigned long flags;
331
332 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
333 list_for_each_entry(fn, &sbi->fsync_node_list, list) {
334 if (fn->page == page) {
335 list_del(&fn->list);
336 sbi->fsync_node_num--;
337 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
338 kmem_cache_free(fsync_node_entry_slab, fn);
339 put_page(page);
340 return;
341 }
342 }
343 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
344 f2fs_bug_on(sbi, 1);
345 }
346
f2fs_reset_fsync_node_info(struct f2fs_sb_info * sbi)347 void f2fs_reset_fsync_node_info(struct f2fs_sb_info *sbi)
348 {
349 unsigned long flags;
350
351 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
352 sbi->fsync_seg_id = 0;
353 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
354 }
355
f2fs_need_dentry_mark(struct f2fs_sb_info * sbi,nid_t nid)356 int f2fs_need_dentry_mark(struct f2fs_sb_info *sbi, nid_t nid)
357 {
358 struct f2fs_nm_info *nm_i = NM_I(sbi);
359 struct nat_entry *e;
360 bool need = false;
361
362 down_read(&nm_i->nat_tree_lock);
363 e = __lookup_nat_cache(nm_i, nid);
364 if (e) {
365 if (!get_nat_flag(e, IS_CHECKPOINTED) &&
366 !get_nat_flag(e, HAS_FSYNCED_INODE))
367 need = true;
368 }
369 up_read(&nm_i->nat_tree_lock);
370 return need;
371 }
372
f2fs_is_checkpointed_node(struct f2fs_sb_info * sbi,nid_t nid)373 bool f2fs_is_checkpointed_node(struct f2fs_sb_info *sbi, nid_t nid)
374 {
375 struct f2fs_nm_info *nm_i = NM_I(sbi);
376 struct nat_entry *e;
377 bool is_cp = true;
378
379 down_read(&nm_i->nat_tree_lock);
380 e = __lookup_nat_cache(nm_i, nid);
381 if (e && !get_nat_flag(e, IS_CHECKPOINTED))
382 is_cp = false;
383 up_read(&nm_i->nat_tree_lock);
384 return is_cp;
385 }
386
f2fs_need_inode_block_update(struct f2fs_sb_info * sbi,nid_t ino)387 bool f2fs_need_inode_block_update(struct f2fs_sb_info *sbi, nid_t ino)
388 {
389 struct f2fs_nm_info *nm_i = NM_I(sbi);
390 struct nat_entry *e;
391 bool need_update = true;
392
393 down_read(&nm_i->nat_tree_lock);
394 e = __lookup_nat_cache(nm_i, ino);
395 if (e && get_nat_flag(e, HAS_LAST_FSYNC) &&
396 (get_nat_flag(e, IS_CHECKPOINTED) ||
397 get_nat_flag(e, HAS_FSYNCED_INODE)))
398 need_update = false;
399 up_read(&nm_i->nat_tree_lock);
400 return need_update;
401 }
402
403 /* must be locked by nat_tree_lock */
cache_nat_entry(struct f2fs_sb_info * sbi,nid_t nid,struct f2fs_nat_entry * ne)404 static void cache_nat_entry(struct f2fs_sb_info *sbi, nid_t nid,
405 struct f2fs_nat_entry *ne)
406 {
407 struct f2fs_nm_info *nm_i = NM_I(sbi);
408 struct nat_entry *new, *e;
409
410 new = __alloc_nat_entry(nid, false);
411 if (!new)
412 return;
413
414 down_write(&nm_i->nat_tree_lock);
415 e = __lookup_nat_cache(nm_i, nid);
416 if (!e)
417 e = __init_nat_entry(nm_i, new, ne, false);
418 else
419 f2fs_bug_on(sbi, nat_get_ino(e) != le32_to_cpu(ne->ino) ||
420 nat_get_blkaddr(e) !=
421 le32_to_cpu(ne->block_addr) ||
422 nat_get_version(e) != ne->version);
423 up_write(&nm_i->nat_tree_lock);
424 if (e != new)
425 __free_nat_entry(new);
426 }
427
set_node_addr(struct f2fs_sb_info * sbi,struct node_info * ni,block_t new_blkaddr,bool fsync_done)428 static void set_node_addr(struct f2fs_sb_info *sbi, struct node_info *ni,
429 block_t new_blkaddr, bool fsync_done)
430 {
431 struct f2fs_nm_info *nm_i = NM_I(sbi);
432 struct nat_entry *e;
433 struct nat_entry *new = __alloc_nat_entry(ni->nid, true);
434
435 down_write(&nm_i->nat_tree_lock);
436 e = __lookup_nat_cache(nm_i, ni->nid);
437 if (!e) {
438 e = __init_nat_entry(nm_i, new, NULL, true);
439 copy_node_info(&e->ni, ni);
440 f2fs_bug_on(sbi, ni->blk_addr == NEW_ADDR);
441 } else if (new_blkaddr == NEW_ADDR) {
442 /*
443 * when nid is reallocated,
444 * previous nat entry can be remained in nat cache.
445 * So, reinitialize it with new information.
446 */
447 copy_node_info(&e->ni, ni);
448 f2fs_bug_on(sbi, ni->blk_addr != NULL_ADDR);
449 }
450 /* let's free early to reduce memory consumption */
451 if (e != new)
452 __free_nat_entry(new);
453
454 /* sanity check */
455 f2fs_bug_on(sbi, nat_get_blkaddr(e) != ni->blk_addr);
456 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NULL_ADDR &&
457 new_blkaddr == NULL_ADDR);
458 f2fs_bug_on(sbi, nat_get_blkaddr(e) == NEW_ADDR &&
459 new_blkaddr == NEW_ADDR);
460 f2fs_bug_on(sbi, __is_valid_data_blkaddr(nat_get_blkaddr(e)) &&
461 new_blkaddr == NEW_ADDR);
462
463 /* increment version no as node is removed */
464 if (nat_get_blkaddr(e) != NEW_ADDR && new_blkaddr == NULL_ADDR) {
465 unsigned char version = nat_get_version(e);
466 nat_set_version(e, inc_node_version(version));
467 }
468
469 /* change address */
470 nat_set_blkaddr(e, new_blkaddr);
471 if (!__is_valid_data_blkaddr(new_blkaddr))
472 set_nat_flag(e, IS_CHECKPOINTED, false);
473 __set_nat_cache_dirty(nm_i, e);
474
475 /* update fsync_mark if its inode nat entry is still alive */
476 if (ni->nid != ni->ino)
477 e = __lookup_nat_cache(nm_i, ni->ino);
478 if (e) {
479 if (fsync_done && ni->nid == ni->ino)
480 set_nat_flag(e, HAS_FSYNCED_INODE, true);
481 set_nat_flag(e, HAS_LAST_FSYNC, fsync_done);
482 }
483 up_write(&nm_i->nat_tree_lock);
484 }
485
f2fs_try_to_free_nats(struct f2fs_sb_info * sbi,int nr_shrink)486 int f2fs_try_to_free_nats(struct f2fs_sb_info *sbi, int nr_shrink)
487 {
488 struct f2fs_nm_info *nm_i = NM_I(sbi);
489 int nr = nr_shrink;
490
491 if (!down_write_trylock(&nm_i->nat_tree_lock))
492 return 0;
493
494 spin_lock(&nm_i->nat_list_lock);
495 while (nr_shrink) {
496 struct nat_entry *ne;
497
498 if (list_empty(&nm_i->nat_entries))
499 break;
500
501 ne = list_first_entry(&nm_i->nat_entries,
502 struct nat_entry, list);
503 list_del(&ne->list);
504 spin_unlock(&nm_i->nat_list_lock);
505
506 __del_from_nat_cache(nm_i, ne);
507 nr_shrink--;
508
509 spin_lock(&nm_i->nat_list_lock);
510 }
511 spin_unlock(&nm_i->nat_list_lock);
512
513 up_write(&nm_i->nat_tree_lock);
514 return nr - nr_shrink;
515 }
516
f2fs_get_node_info(struct f2fs_sb_info * sbi,nid_t nid,struct node_info * ni)517 int f2fs_get_node_info(struct f2fs_sb_info *sbi, nid_t nid,
518 struct node_info *ni)
519 {
520 struct f2fs_nm_info *nm_i = NM_I(sbi);
521 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
522 struct f2fs_journal *journal = curseg->journal;
523 nid_t start_nid = START_NID(nid);
524 struct f2fs_nat_block *nat_blk;
525 struct page *page = NULL;
526 struct f2fs_nat_entry ne;
527 struct nat_entry *e;
528 pgoff_t index;
529 block_t blkaddr;
530 int i;
531
532 ni->nid = nid;
533
534 /* Check nat cache */
535 down_read(&nm_i->nat_tree_lock);
536 e = __lookup_nat_cache(nm_i, nid);
537 if (e) {
538 ni->ino = nat_get_ino(e);
539 ni->blk_addr = nat_get_blkaddr(e);
540 ni->version = nat_get_version(e);
541 up_read(&nm_i->nat_tree_lock);
542 return 0;
543 }
544
545 memset(&ne, 0, sizeof(struct f2fs_nat_entry));
546
547 /* Check current segment summary */
548 down_read(&curseg->journal_rwsem);
549 i = f2fs_lookup_journal_in_cursum(journal, NAT_JOURNAL, nid, 0);
550 if (i >= 0) {
551 ne = nat_in_journal(journal, i);
552 node_info_from_raw_nat(ni, &ne);
553 }
554 up_read(&curseg->journal_rwsem);
555 if (i >= 0) {
556 up_read(&nm_i->nat_tree_lock);
557 goto cache;
558 }
559
560 /* Fill node_info from nat page */
561 index = current_nat_addr(sbi, nid);
562 up_read(&nm_i->nat_tree_lock);
563
564 page = f2fs_get_meta_page(sbi, index);
565 if (IS_ERR(page))
566 return PTR_ERR(page);
567
568 nat_blk = (struct f2fs_nat_block *)page_address(page);
569 ne = nat_blk->entries[nid - start_nid];
570 node_info_from_raw_nat(ni, &ne);
571 f2fs_put_page(page, 1);
572 cache:
573 blkaddr = le32_to_cpu(ne.block_addr);
574 if (__is_valid_data_blkaddr(blkaddr) &&
575 !f2fs_is_valid_blkaddr(sbi, blkaddr, DATA_GENERIC_ENHANCE))
576 return -EFAULT;
577
578 /* cache nat entry */
579 cache_nat_entry(sbi, nid, &ne);
580 return 0;
581 }
582
583 /*
584 * readahead MAX_RA_NODE number of node pages.
585 */
f2fs_ra_node_pages(struct page * parent,int start,int n)586 static void f2fs_ra_node_pages(struct page *parent, int start, int n)
587 {
588 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
589 struct blk_plug plug;
590 int i, end;
591 nid_t nid;
592
593 blk_start_plug(&plug);
594
595 /* Then, try readahead for siblings of the desired node */
596 end = start + n;
597 end = min(end, NIDS_PER_BLOCK);
598 for (i = start; i < end; i++) {
599 nid = get_nid(parent, i, false);
600 f2fs_ra_node_page(sbi, nid);
601 }
602
603 blk_finish_plug(&plug);
604 }
605
f2fs_get_next_page_offset(struct dnode_of_data * dn,pgoff_t pgofs)606 pgoff_t f2fs_get_next_page_offset(struct dnode_of_data *dn, pgoff_t pgofs)
607 {
608 const long direct_index = ADDRS_PER_INODE(dn->inode);
609 const long direct_blks = ADDRS_PER_BLOCK(dn->inode);
610 const long indirect_blks = ADDRS_PER_BLOCK(dn->inode) * NIDS_PER_BLOCK;
611 unsigned int skipped_unit = ADDRS_PER_BLOCK(dn->inode);
612 int cur_level = dn->cur_level;
613 int max_level = dn->max_level;
614 pgoff_t base = 0;
615
616 if (!dn->max_level)
617 return pgofs + 1;
618
619 while (max_level-- > cur_level)
620 skipped_unit *= NIDS_PER_BLOCK;
621
622 switch (dn->max_level) {
623 case 3:
624 base += 2 * indirect_blks;
625 /* fall through */
626 case 2:
627 base += 2 * direct_blks;
628 /* fall through */
629 case 1:
630 base += direct_index;
631 break;
632 default:
633 f2fs_bug_on(F2FS_I_SB(dn->inode), 1);
634 }
635
636 return ((pgofs - base) / skipped_unit + 1) * skipped_unit + base;
637 }
638
639 /*
640 * The maximum depth is four.
641 * Offset[0] will have raw inode offset.
642 */
get_node_path(struct inode * inode,long block,int offset[4],unsigned int noffset[4])643 static int get_node_path(struct inode *inode, long block,
644 int offset[4], unsigned int noffset[4])
645 {
646 const long direct_index = ADDRS_PER_INODE(inode);
647 const long direct_blks = ADDRS_PER_BLOCK(inode);
648 const long dptrs_per_blk = NIDS_PER_BLOCK;
649 const long indirect_blks = ADDRS_PER_BLOCK(inode) * NIDS_PER_BLOCK;
650 const long dindirect_blks = indirect_blks * NIDS_PER_BLOCK;
651 int n = 0;
652 int level = 0;
653
654 noffset[0] = 0;
655
656 if (block < direct_index) {
657 offset[n] = block;
658 goto got;
659 }
660 block -= direct_index;
661 if (block < direct_blks) {
662 offset[n++] = NODE_DIR1_BLOCK;
663 noffset[n] = 1;
664 offset[n] = block;
665 level = 1;
666 goto got;
667 }
668 block -= direct_blks;
669 if (block < direct_blks) {
670 offset[n++] = NODE_DIR2_BLOCK;
671 noffset[n] = 2;
672 offset[n] = block;
673 level = 1;
674 goto got;
675 }
676 block -= direct_blks;
677 if (block < indirect_blks) {
678 offset[n++] = NODE_IND1_BLOCK;
679 noffset[n] = 3;
680 offset[n++] = block / direct_blks;
681 noffset[n] = 4 + offset[n - 1];
682 offset[n] = block % direct_blks;
683 level = 2;
684 goto got;
685 }
686 block -= indirect_blks;
687 if (block < indirect_blks) {
688 offset[n++] = NODE_IND2_BLOCK;
689 noffset[n] = 4 + dptrs_per_blk;
690 offset[n++] = block / direct_blks;
691 noffset[n] = 5 + dptrs_per_blk + offset[n - 1];
692 offset[n] = block % direct_blks;
693 level = 2;
694 goto got;
695 }
696 block -= indirect_blks;
697 if (block < dindirect_blks) {
698 offset[n++] = NODE_DIND_BLOCK;
699 noffset[n] = 5 + (dptrs_per_blk * 2);
700 offset[n++] = block / indirect_blks;
701 noffset[n] = 6 + (dptrs_per_blk * 2) +
702 offset[n - 1] * (dptrs_per_blk + 1);
703 offset[n++] = (block / direct_blks) % dptrs_per_blk;
704 noffset[n] = 7 + (dptrs_per_blk * 2) +
705 offset[n - 2] * (dptrs_per_blk + 1) +
706 offset[n - 1];
707 offset[n] = block % direct_blks;
708 level = 3;
709 goto got;
710 } else {
711 return -E2BIG;
712 }
713 got:
714 return level;
715 }
716
717 /*
718 * Caller should call f2fs_put_dnode(dn).
719 * Also, it should grab and release a rwsem by calling f2fs_lock_op() and
720 * f2fs_unlock_op() only if mode is set with ALLOC_NODE.
721 */
f2fs_get_dnode_of_data(struct dnode_of_data * dn,pgoff_t index,int mode)722 int f2fs_get_dnode_of_data(struct dnode_of_data *dn, pgoff_t index, int mode)
723 {
724 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
725 struct page *npage[4];
726 struct page *parent = NULL;
727 int offset[4];
728 unsigned int noffset[4];
729 nid_t nids[4];
730 int level, i = 0;
731 int err = 0;
732
733 level = get_node_path(dn->inode, index, offset, noffset);
734 if (level < 0)
735 return level;
736
737 nids[0] = dn->inode->i_ino;
738 npage[0] = dn->inode_page;
739
740 if (!npage[0]) {
741 npage[0] = f2fs_get_node_page(sbi, nids[0]);
742 if (IS_ERR(npage[0]))
743 return PTR_ERR(npage[0]);
744 }
745
746 /* if inline_data is set, should not report any block indices */
747 if (f2fs_has_inline_data(dn->inode) && index) {
748 err = -ENOENT;
749 f2fs_put_page(npage[0], 1);
750 goto release_out;
751 }
752
753 parent = npage[0];
754 if (level != 0)
755 nids[1] = get_nid(parent, offset[0], true);
756 dn->inode_page = npage[0];
757 dn->inode_page_locked = true;
758
759 /* get indirect or direct nodes */
760 for (i = 1; i <= level; i++) {
761 bool done = false;
762
763 if (!nids[i] && mode == ALLOC_NODE) {
764 /* alloc new node */
765 if (!f2fs_alloc_nid(sbi, &(nids[i]))) {
766 err = -ENOSPC;
767 goto release_pages;
768 }
769
770 dn->nid = nids[i];
771 npage[i] = f2fs_new_node_page(dn, noffset[i]);
772 if (IS_ERR(npage[i])) {
773 f2fs_alloc_nid_failed(sbi, nids[i]);
774 err = PTR_ERR(npage[i]);
775 goto release_pages;
776 }
777
778 set_nid(parent, offset[i - 1], nids[i], i == 1);
779 f2fs_alloc_nid_done(sbi, nids[i]);
780 done = true;
781 } else if (mode == LOOKUP_NODE_RA && i == level && level > 1) {
782 npage[i] = f2fs_get_node_page_ra(parent, offset[i - 1]);
783 if (IS_ERR(npage[i])) {
784 err = PTR_ERR(npage[i]);
785 goto release_pages;
786 }
787 done = true;
788 }
789 if (i == 1) {
790 dn->inode_page_locked = false;
791 unlock_page(parent);
792 } else {
793 f2fs_put_page(parent, 1);
794 }
795
796 if (!done) {
797 npage[i] = f2fs_get_node_page(sbi, nids[i]);
798 if (IS_ERR(npage[i])) {
799 err = PTR_ERR(npage[i]);
800 f2fs_put_page(npage[0], 0);
801 goto release_out;
802 }
803 }
804 if (i < level) {
805 parent = npage[i];
806 nids[i + 1] = get_nid(parent, offset[i], false);
807 }
808 }
809 dn->nid = nids[level];
810 dn->ofs_in_node = offset[level];
811 dn->node_page = npage[level];
812 dn->data_blkaddr = f2fs_data_blkaddr(dn);
813 return 0;
814
815 release_pages:
816 f2fs_put_page(parent, 1);
817 if (i > 1)
818 f2fs_put_page(npage[0], 0);
819 release_out:
820 dn->inode_page = NULL;
821 dn->node_page = NULL;
822 if (err == -ENOENT) {
823 dn->cur_level = i;
824 dn->max_level = level;
825 dn->ofs_in_node = offset[level];
826 }
827 return err;
828 }
829
truncate_node(struct dnode_of_data * dn)830 static int truncate_node(struct dnode_of_data *dn)
831 {
832 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
833 struct node_info ni;
834 int err;
835 pgoff_t index;
836
837 err = f2fs_get_node_info(sbi, dn->nid, &ni);
838 if (err)
839 return err;
840
841 /* Deallocate node address */
842 f2fs_invalidate_blocks(sbi, ni.blk_addr);
843 dec_valid_node_count(sbi, dn->inode, dn->nid == dn->inode->i_ino);
844 set_node_addr(sbi, &ni, NULL_ADDR, false);
845
846 if (dn->nid == dn->inode->i_ino) {
847 f2fs_remove_orphan_inode(sbi, dn->nid);
848 dec_valid_inode_count(sbi);
849 f2fs_inode_synced(dn->inode);
850 }
851
852 clear_node_page_dirty(dn->node_page);
853 set_sbi_flag(sbi, SBI_IS_DIRTY);
854
855 index = dn->node_page->index;
856 f2fs_put_page(dn->node_page, 1);
857
858 invalidate_mapping_pages(NODE_MAPPING(sbi),
859 index, index);
860
861 dn->node_page = NULL;
862 trace_f2fs_truncate_node(dn->inode, dn->nid, ni.blk_addr);
863
864 return 0;
865 }
866
truncate_dnode(struct dnode_of_data * dn)867 static int truncate_dnode(struct dnode_of_data *dn)
868 {
869 struct page *page;
870 int err;
871
872 if (dn->nid == 0)
873 return 1;
874
875 /* get direct node */
876 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
877 if (IS_ERR(page) && PTR_ERR(page) == -ENOENT)
878 return 1;
879 else if (IS_ERR(page))
880 return PTR_ERR(page);
881
882 /* Make dnode_of_data for parameter */
883 dn->node_page = page;
884 dn->ofs_in_node = 0;
885 f2fs_truncate_data_blocks(dn);
886 err = truncate_node(dn);
887 if (err) {
888 f2fs_put_page(page, 1);
889 return err;
890 }
891
892 return 1;
893 }
894
truncate_nodes(struct dnode_of_data * dn,unsigned int nofs,int ofs,int depth)895 static int truncate_nodes(struct dnode_of_data *dn, unsigned int nofs,
896 int ofs, int depth)
897 {
898 struct dnode_of_data rdn = *dn;
899 struct page *page;
900 struct f2fs_node *rn;
901 nid_t child_nid;
902 unsigned int child_nofs;
903 int freed = 0;
904 int i, ret;
905
906 if (dn->nid == 0)
907 return NIDS_PER_BLOCK + 1;
908
909 trace_f2fs_truncate_nodes_enter(dn->inode, dn->nid, dn->data_blkaddr);
910
911 page = f2fs_get_node_page(F2FS_I_SB(dn->inode), dn->nid);
912 if (IS_ERR(page)) {
913 trace_f2fs_truncate_nodes_exit(dn->inode, PTR_ERR(page));
914 return PTR_ERR(page);
915 }
916
917 f2fs_ra_node_pages(page, ofs, NIDS_PER_BLOCK);
918
919 rn = F2FS_NODE(page);
920 if (depth < 3) {
921 for (i = ofs; i < NIDS_PER_BLOCK; i++, freed++) {
922 child_nid = le32_to_cpu(rn->in.nid[i]);
923 if (child_nid == 0)
924 continue;
925 rdn.nid = child_nid;
926 ret = truncate_dnode(&rdn);
927 if (ret < 0)
928 goto out_err;
929 if (set_nid(page, i, 0, false))
930 dn->node_changed = true;
931 }
932 } else {
933 child_nofs = nofs + ofs * (NIDS_PER_BLOCK + 1) + 1;
934 for (i = ofs; i < NIDS_PER_BLOCK; i++) {
935 child_nid = le32_to_cpu(rn->in.nid[i]);
936 if (child_nid == 0) {
937 child_nofs += NIDS_PER_BLOCK + 1;
938 continue;
939 }
940 rdn.nid = child_nid;
941 ret = truncate_nodes(&rdn, child_nofs, 0, depth - 1);
942 if (ret == (NIDS_PER_BLOCK + 1)) {
943 if (set_nid(page, i, 0, false))
944 dn->node_changed = true;
945 child_nofs += ret;
946 } else if (ret < 0 && ret != -ENOENT) {
947 goto out_err;
948 }
949 }
950 freed = child_nofs;
951 }
952
953 if (!ofs) {
954 /* remove current indirect node */
955 dn->node_page = page;
956 ret = truncate_node(dn);
957 if (ret)
958 goto out_err;
959 freed++;
960 } else {
961 f2fs_put_page(page, 1);
962 }
963 trace_f2fs_truncate_nodes_exit(dn->inode, freed);
964 return freed;
965
966 out_err:
967 f2fs_put_page(page, 1);
968 trace_f2fs_truncate_nodes_exit(dn->inode, ret);
969 return ret;
970 }
971
truncate_partial_nodes(struct dnode_of_data * dn,struct f2fs_inode * ri,int * offset,int depth)972 static int truncate_partial_nodes(struct dnode_of_data *dn,
973 struct f2fs_inode *ri, int *offset, int depth)
974 {
975 struct page *pages[2];
976 nid_t nid[3];
977 nid_t child_nid;
978 int err = 0;
979 int i;
980 int idx = depth - 2;
981
982 nid[0] = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
983 if (!nid[0])
984 return 0;
985
986 /* get indirect nodes in the path */
987 for (i = 0; i < idx + 1; i++) {
988 /* reference count'll be increased */
989 pages[i] = f2fs_get_node_page(F2FS_I_SB(dn->inode), nid[i]);
990 if (IS_ERR(pages[i])) {
991 err = PTR_ERR(pages[i]);
992 idx = i - 1;
993 goto fail;
994 }
995 nid[i + 1] = get_nid(pages[i], offset[i + 1], false);
996 }
997
998 f2fs_ra_node_pages(pages[idx], offset[idx + 1], NIDS_PER_BLOCK);
999
1000 /* free direct nodes linked to a partial indirect node */
1001 for (i = offset[idx + 1]; i < NIDS_PER_BLOCK; i++) {
1002 child_nid = get_nid(pages[idx], i, false);
1003 if (!child_nid)
1004 continue;
1005 dn->nid = child_nid;
1006 err = truncate_dnode(dn);
1007 if (err < 0)
1008 goto fail;
1009 if (set_nid(pages[idx], i, 0, false))
1010 dn->node_changed = true;
1011 }
1012
1013 if (offset[idx + 1] == 0) {
1014 dn->node_page = pages[idx];
1015 dn->nid = nid[idx];
1016 err = truncate_node(dn);
1017 if (err)
1018 goto fail;
1019 } else {
1020 f2fs_put_page(pages[idx], 1);
1021 }
1022 offset[idx]++;
1023 offset[idx + 1] = 0;
1024 idx--;
1025 fail:
1026 for (i = idx; i >= 0; i--)
1027 f2fs_put_page(pages[i], 1);
1028
1029 trace_f2fs_truncate_partial_nodes(dn->inode, nid, depth, err);
1030
1031 return err;
1032 }
1033
1034 /*
1035 * All the block addresses of data and nodes should be nullified.
1036 */
f2fs_truncate_inode_blocks(struct inode * inode,pgoff_t from)1037 int f2fs_truncate_inode_blocks(struct inode *inode, pgoff_t from)
1038 {
1039 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1040 int err = 0, cont = 1;
1041 int level, offset[4], noffset[4];
1042 unsigned int nofs = 0;
1043 struct f2fs_inode *ri;
1044 struct dnode_of_data dn;
1045 struct page *page;
1046
1047 trace_f2fs_truncate_inode_blocks_enter(inode, from);
1048
1049 level = get_node_path(inode, from, offset, noffset);
1050 if (level < 0)
1051 return level;
1052
1053 page = f2fs_get_node_page(sbi, inode->i_ino);
1054 if (IS_ERR(page)) {
1055 trace_f2fs_truncate_inode_blocks_exit(inode, PTR_ERR(page));
1056 return PTR_ERR(page);
1057 }
1058
1059 set_new_dnode(&dn, inode, page, NULL, 0);
1060 unlock_page(page);
1061
1062 ri = F2FS_INODE(page);
1063 switch (level) {
1064 case 0:
1065 case 1:
1066 nofs = noffset[1];
1067 break;
1068 case 2:
1069 nofs = noffset[1];
1070 if (!offset[level - 1])
1071 goto skip_partial;
1072 err = truncate_partial_nodes(&dn, ri, offset, level);
1073 if (err < 0 && err != -ENOENT)
1074 goto fail;
1075 nofs += 1 + NIDS_PER_BLOCK;
1076 break;
1077 case 3:
1078 nofs = 5 + 2 * NIDS_PER_BLOCK;
1079 if (!offset[level - 1])
1080 goto skip_partial;
1081 err = truncate_partial_nodes(&dn, ri, offset, level);
1082 if (err < 0 && err != -ENOENT)
1083 goto fail;
1084 break;
1085 default:
1086 BUG();
1087 }
1088
1089 skip_partial:
1090 while (cont) {
1091 dn.nid = le32_to_cpu(ri->i_nid[offset[0] - NODE_DIR1_BLOCK]);
1092 switch (offset[0]) {
1093 case NODE_DIR1_BLOCK:
1094 case NODE_DIR2_BLOCK:
1095 err = truncate_dnode(&dn);
1096 break;
1097
1098 case NODE_IND1_BLOCK:
1099 case NODE_IND2_BLOCK:
1100 err = truncate_nodes(&dn, nofs, offset[1], 2);
1101 break;
1102
1103 case NODE_DIND_BLOCK:
1104 err = truncate_nodes(&dn, nofs, offset[1], 3);
1105 cont = 0;
1106 break;
1107
1108 default:
1109 BUG();
1110 }
1111 if (err < 0 && err != -ENOENT)
1112 goto fail;
1113 if (offset[1] == 0 &&
1114 ri->i_nid[offset[0] - NODE_DIR1_BLOCK]) {
1115 lock_page(page);
1116 BUG_ON(page->mapping != NODE_MAPPING(sbi));
1117 f2fs_wait_on_page_writeback(page, NODE, true, true);
1118 ri->i_nid[offset[0] - NODE_DIR1_BLOCK] = 0;
1119 set_page_dirty(page);
1120 unlock_page(page);
1121 }
1122 offset[1] = 0;
1123 offset[0]++;
1124 nofs += err;
1125 }
1126 fail:
1127 f2fs_put_page(page, 0);
1128 trace_f2fs_truncate_inode_blocks_exit(inode, err);
1129 return err > 0 ? 0 : err;
1130 }
1131
1132 /* caller must lock inode page */
f2fs_truncate_xattr_node(struct inode * inode)1133 int f2fs_truncate_xattr_node(struct inode *inode)
1134 {
1135 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1136 nid_t nid = F2FS_I(inode)->i_xattr_nid;
1137 struct dnode_of_data dn;
1138 struct page *npage;
1139 int err;
1140
1141 if (!nid)
1142 return 0;
1143
1144 npage = f2fs_get_node_page(sbi, nid);
1145 if (IS_ERR(npage))
1146 return PTR_ERR(npage);
1147
1148 set_new_dnode(&dn, inode, NULL, npage, nid);
1149 err = truncate_node(&dn);
1150 if (err) {
1151 f2fs_put_page(npage, 1);
1152 return err;
1153 }
1154
1155 f2fs_i_xnid_write(inode, 0);
1156
1157 return 0;
1158 }
1159
1160 /*
1161 * Caller should grab and release a rwsem by calling f2fs_lock_op() and
1162 * f2fs_unlock_op().
1163 */
f2fs_remove_inode_page(struct inode * inode)1164 int f2fs_remove_inode_page(struct inode *inode)
1165 {
1166 struct dnode_of_data dn;
1167 int err;
1168
1169 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1170 err = f2fs_get_dnode_of_data(&dn, 0, LOOKUP_NODE);
1171 if (err)
1172 return err;
1173
1174 err = f2fs_truncate_xattr_node(inode);
1175 if (err) {
1176 f2fs_put_dnode(&dn);
1177 return err;
1178 }
1179
1180 /* remove potential inline_data blocks */
1181 if (S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
1182 S_ISLNK(inode->i_mode))
1183 f2fs_truncate_data_blocks_range(&dn, 1);
1184
1185 /* 0 is possible, after f2fs_new_inode() has failed */
1186 if (unlikely(f2fs_cp_error(F2FS_I_SB(inode)))) {
1187 f2fs_put_dnode(&dn);
1188 return -EIO;
1189 }
1190
1191 if (unlikely(inode->i_blocks != 0 && inode->i_blocks != 8)) {
1192 f2fs_warn(F2FS_I_SB(inode),
1193 "f2fs_remove_inode_page: inconsistent i_blocks, ino:%lu, iblocks:%llu",
1194 inode->i_ino, (unsigned long long)inode->i_blocks);
1195 set_sbi_flag(F2FS_I_SB(inode), SBI_NEED_FSCK);
1196 }
1197
1198 /* will put inode & node pages */
1199 err = truncate_node(&dn);
1200 if (err) {
1201 f2fs_put_dnode(&dn);
1202 return err;
1203 }
1204 return 0;
1205 }
1206
f2fs_new_inode_page(struct inode * inode)1207 struct page *f2fs_new_inode_page(struct inode *inode)
1208 {
1209 struct dnode_of_data dn;
1210
1211 /* allocate inode page for new inode */
1212 set_new_dnode(&dn, inode, NULL, NULL, inode->i_ino);
1213
1214 /* caller should f2fs_put_page(page, 1); */
1215 return f2fs_new_node_page(&dn, 0);
1216 }
1217
f2fs_new_node_page(struct dnode_of_data * dn,unsigned int ofs)1218 struct page *f2fs_new_node_page(struct dnode_of_data *dn, unsigned int ofs)
1219 {
1220 struct f2fs_sb_info *sbi = F2FS_I_SB(dn->inode);
1221 struct node_info new_ni;
1222 struct page *page;
1223 int err;
1224
1225 if (unlikely(is_inode_flag_set(dn->inode, FI_NO_ALLOC)))
1226 return ERR_PTR(-EPERM);
1227
1228 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), dn->nid, false);
1229 if (!page)
1230 return ERR_PTR(-ENOMEM);
1231
1232 if (unlikely((err = inc_valid_node_count(sbi, dn->inode, !ofs))))
1233 goto fail;
1234
1235 #ifdef CONFIG_F2FS_CHECK_FS
1236 err = f2fs_get_node_info(sbi, dn->nid, &new_ni);
1237 if (err) {
1238 dec_valid_node_count(sbi, dn->inode, !ofs);
1239 goto fail;
1240 }
1241 if (unlikely(new_ni.blk_addr != NULL_ADDR)) {
1242 err = -EFSCORRUPTED;
1243 set_sbi_flag(sbi, SBI_NEED_FSCK);
1244 goto fail;
1245 }
1246 #endif
1247 new_ni.nid = dn->nid;
1248 new_ni.ino = dn->inode->i_ino;
1249 new_ni.blk_addr = NULL_ADDR;
1250 new_ni.flag = 0;
1251 new_ni.version = 0;
1252 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
1253
1254 f2fs_wait_on_page_writeback(page, NODE, true, true);
1255 fill_node_footer(page, dn->nid, dn->inode->i_ino, ofs, true);
1256 set_cold_node(page, S_ISDIR(dn->inode->i_mode));
1257 if (!PageUptodate(page))
1258 SetPageUptodate(page);
1259 if (set_page_dirty(page))
1260 dn->node_changed = true;
1261
1262 if (f2fs_has_xattr_block(ofs))
1263 f2fs_i_xnid_write(dn->inode, dn->nid);
1264
1265 if (ofs == 0)
1266 inc_valid_inode_count(sbi);
1267 return page;
1268
1269 fail:
1270 clear_node_page_dirty(page);
1271 f2fs_put_page(page, 1);
1272 return ERR_PTR(err);
1273 }
1274
1275 /*
1276 * Caller should do after getting the following values.
1277 * 0: f2fs_put_page(page, 0)
1278 * LOCKED_PAGE or error: f2fs_put_page(page, 1)
1279 */
read_node_page(struct page * page,int op_flags)1280 static int read_node_page(struct page *page, int op_flags)
1281 {
1282 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1283 struct node_info ni;
1284 struct f2fs_io_info fio = {
1285 .sbi = sbi,
1286 .type = NODE,
1287 .op = REQ_OP_READ,
1288 .op_flags = op_flags,
1289 .page = page,
1290 .encrypted_page = NULL,
1291 };
1292 int err;
1293
1294 if (PageUptodate(page)) {
1295 if (!f2fs_inode_chksum_verify(sbi, page)) {
1296 ClearPageUptodate(page);
1297 return -EFSBADCRC;
1298 }
1299 return LOCKED_PAGE;
1300 }
1301
1302 err = f2fs_get_node_info(sbi, page->index, &ni);
1303 if (err)
1304 return err;
1305
1306 if (unlikely(ni.blk_addr == NULL_ADDR) ||
1307 is_sbi_flag_set(sbi, SBI_IS_SHUTDOWN)) {
1308 ClearPageUptodate(page);
1309 return -ENOENT;
1310 }
1311
1312 fio.new_blkaddr = fio.old_blkaddr = ni.blk_addr;
1313
1314 err = f2fs_submit_page_bio(&fio);
1315
1316 if (!err)
1317 f2fs_update_iostat(sbi, FS_NODE_READ_IO, F2FS_BLKSIZE);
1318
1319 return err;
1320 }
1321
1322 /*
1323 * Readahead a node page
1324 */
f2fs_ra_node_page(struct f2fs_sb_info * sbi,nid_t nid)1325 void f2fs_ra_node_page(struct f2fs_sb_info *sbi, nid_t nid)
1326 {
1327 struct page *apage;
1328 int err;
1329
1330 if (!nid)
1331 return;
1332 if (f2fs_check_nid_range(sbi, nid))
1333 return;
1334
1335 apage = xa_load(&NODE_MAPPING(sbi)->i_pages, nid);
1336 if (apage)
1337 return;
1338
1339 apage = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1340 if (!apage)
1341 return;
1342
1343 err = read_node_page(apage, REQ_RAHEAD);
1344 f2fs_put_page(apage, err ? 1 : 0);
1345 }
1346
__get_node_page(struct f2fs_sb_info * sbi,pgoff_t nid,struct page * parent,int start)1347 static struct page *__get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid,
1348 struct page *parent, int start)
1349 {
1350 struct page *page;
1351 int err;
1352
1353 if (!nid)
1354 return ERR_PTR(-ENOENT);
1355 if (f2fs_check_nid_range(sbi, nid))
1356 return ERR_PTR(-EINVAL);
1357 repeat:
1358 page = f2fs_grab_cache_page(NODE_MAPPING(sbi), nid, false);
1359 if (!page)
1360 return ERR_PTR(-ENOMEM);
1361
1362 err = read_node_page(page, 0);
1363 if (err < 0) {
1364 f2fs_put_page(page, 1);
1365 return ERR_PTR(err);
1366 } else if (err == LOCKED_PAGE) {
1367 err = 0;
1368 goto page_hit;
1369 }
1370
1371 if (parent)
1372 f2fs_ra_node_pages(parent, start + 1, MAX_RA_NODE);
1373
1374 lock_page(page);
1375
1376 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1377 f2fs_put_page(page, 1);
1378 goto repeat;
1379 }
1380
1381 if (unlikely(!PageUptodate(page))) {
1382 err = -EIO;
1383 goto out_err;
1384 }
1385
1386 if (!f2fs_inode_chksum_verify(sbi, page)) {
1387 err = -EFSBADCRC;
1388 goto out_err;
1389 }
1390 page_hit:
1391 if(unlikely(nid != nid_of_node(page))) {
1392 f2fs_warn(sbi, "inconsistent node block, nid:%lu, node_footer[nid:%u,ino:%u,ofs:%u,cpver:%llu,blkaddr:%u]",
1393 nid, nid_of_node(page), ino_of_node(page),
1394 ofs_of_node(page), cpver_of_node(page),
1395 next_blkaddr_of_node(page));
1396 set_sbi_flag(sbi, SBI_NEED_FSCK);
1397 err = -EINVAL;
1398 out_err:
1399 ClearPageUptodate(page);
1400 f2fs_put_page(page, 1);
1401 return ERR_PTR(err);
1402 }
1403 return page;
1404 }
1405
f2fs_get_node_page(struct f2fs_sb_info * sbi,pgoff_t nid)1406 struct page *f2fs_get_node_page(struct f2fs_sb_info *sbi, pgoff_t nid)
1407 {
1408 return __get_node_page(sbi, nid, NULL, 0);
1409 }
1410
f2fs_get_node_page_ra(struct page * parent,int start)1411 struct page *f2fs_get_node_page_ra(struct page *parent, int start)
1412 {
1413 struct f2fs_sb_info *sbi = F2FS_P_SB(parent);
1414 nid_t nid = get_nid(parent, start, false);
1415
1416 return __get_node_page(sbi, nid, parent, start);
1417 }
1418
flush_inline_data(struct f2fs_sb_info * sbi,nid_t ino)1419 static void flush_inline_data(struct f2fs_sb_info *sbi, nid_t ino)
1420 {
1421 struct inode *inode;
1422 struct page *page;
1423 int ret;
1424
1425 /* should flush inline_data before evict_inode */
1426 inode = ilookup(sbi->sb, ino);
1427 if (!inode)
1428 return;
1429
1430 page = f2fs_pagecache_get_page(inode->i_mapping, 0,
1431 FGP_LOCK|FGP_NOWAIT, 0);
1432 if (!page)
1433 goto iput_out;
1434
1435 if (!PageUptodate(page))
1436 goto page_out;
1437
1438 if (!PageDirty(page))
1439 goto page_out;
1440
1441 if (!clear_page_dirty_for_io(page))
1442 goto page_out;
1443
1444 ret = f2fs_write_inline_data(inode, page);
1445 inode_dec_dirty_pages(inode);
1446 f2fs_remove_dirty_inode(inode);
1447 if (ret)
1448 set_page_dirty(page);
1449 page_out:
1450 f2fs_put_page(page, 1);
1451 iput_out:
1452 iput(inode);
1453 }
1454
last_fsync_dnode(struct f2fs_sb_info * sbi,nid_t ino)1455 static struct page *last_fsync_dnode(struct f2fs_sb_info *sbi, nid_t ino)
1456 {
1457 pgoff_t index;
1458 struct pagevec pvec;
1459 struct page *last_page = NULL;
1460 int nr_pages;
1461
1462 pagevec_init(&pvec);
1463 index = 0;
1464
1465 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1466 PAGECACHE_TAG_DIRTY))) {
1467 int i;
1468
1469 for (i = 0; i < nr_pages; i++) {
1470 struct page *page = pvec.pages[i];
1471
1472 if (unlikely(f2fs_cp_error(sbi))) {
1473 f2fs_put_page(last_page, 0);
1474 pagevec_release(&pvec);
1475 return ERR_PTR(-EIO);
1476 }
1477
1478 if (!IS_DNODE(page) || !is_cold_node(page))
1479 continue;
1480 if (ino_of_node(page) != ino)
1481 continue;
1482
1483 lock_page(page);
1484
1485 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1486 continue_unlock:
1487 unlock_page(page);
1488 continue;
1489 }
1490 if (ino_of_node(page) != ino)
1491 goto continue_unlock;
1492
1493 if (!PageDirty(page)) {
1494 /* someone wrote it for us */
1495 goto continue_unlock;
1496 }
1497
1498 if (last_page)
1499 f2fs_put_page(last_page, 0);
1500
1501 get_page(page);
1502 last_page = page;
1503 unlock_page(page);
1504 }
1505 pagevec_release(&pvec);
1506 cond_resched();
1507 }
1508 return last_page;
1509 }
1510
__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)1511 static int __write_node_page(struct page *page, bool atomic, bool *submitted,
1512 struct writeback_control *wbc, bool do_balance,
1513 enum iostat_type io_type, unsigned int *seq_id)
1514 {
1515 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1516 nid_t nid;
1517 struct node_info ni;
1518 struct f2fs_io_info fio = {
1519 .sbi = sbi,
1520 .ino = ino_of_node(page),
1521 .type = NODE,
1522 .op = REQ_OP_WRITE,
1523 .op_flags = wbc_to_write_flags(wbc),
1524 .page = page,
1525 .encrypted_page = NULL,
1526 .submitted = false,
1527 .io_type = io_type,
1528 .io_wbc = wbc,
1529 };
1530 unsigned int seq;
1531
1532 trace_f2fs_writepage(page, NODE);
1533
1534 if (unlikely(f2fs_cp_error(sbi))) {
1535 if (is_sbi_flag_set(sbi, SBI_IS_CLOSE)) {
1536 ClearPageUptodate(page);
1537 dec_page_count(sbi, F2FS_DIRTY_NODES);
1538 unlock_page(page);
1539 return 0;
1540 }
1541 goto redirty_out;
1542 }
1543
1544 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
1545 goto redirty_out;
1546
1547 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1548 wbc->sync_mode == WB_SYNC_NONE &&
1549 IS_DNODE(page) && is_cold_node(page))
1550 goto redirty_out;
1551
1552 /* get old block addr of this node page */
1553 nid = nid_of_node(page);
1554 f2fs_bug_on(sbi, page->index != nid);
1555
1556 if (f2fs_get_node_info(sbi, nid, &ni))
1557 goto redirty_out;
1558
1559 if (wbc->for_reclaim) {
1560 if (!down_read_trylock(&sbi->node_write))
1561 goto redirty_out;
1562 } else {
1563 down_read(&sbi->node_write);
1564 }
1565
1566 /* This page is already truncated */
1567 if (unlikely(ni.blk_addr == NULL_ADDR)) {
1568 ClearPageUptodate(page);
1569 dec_page_count(sbi, F2FS_DIRTY_NODES);
1570 up_read(&sbi->node_write);
1571 unlock_page(page);
1572 return 0;
1573 }
1574
1575 if (__is_valid_data_blkaddr(ni.blk_addr) &&
1576 !f2fs_is_valid_blkaddr(sbi, ni.blk_addr,
1577 DATA_GENERIC_ENHANCE)) {
1578 up_read(&sbi->node_write);
1579 goto redirty_out;
1580 }
1581
1582 if (atomic && !test_opt(sbi, NOBARRIER))
1583 fio.op_flags |= REQ_PREFLUSH | REQ_FUA;
1584
1585 /* should add to global list before clearing PAGECACHE status */
1586 if (f2fs_in_warm_node_list(sbi, page)) {
1587 seq = f2fs_add_fsync_node_entry(sbi, page);
1588 if (seq_id)
1589 *seq_id = seq;
1590 }
1591
1592 set_page_writeback(page);
1593 ClearPageError(page);
1594
1595 fio.old_blkaddr = ni.blk_addr;
1596 f2fs_do_write_node_page(nid, &fio);
1597 set_node_addr(sbi, &ni, fio.new_blkaddr, is_fsync_dnode(page));
1598 dec_page_count(sbi, F2FS_DIRTY_NODES);
1599 up_read(&sbi->node_write);
1600
1601 if (wbc->for_reclaim) {
1602 f2fs_submit_merged_write_cond(sbi, NULL, page, 0, NODE);
1603 submitted = NULL;
1604 }
1605
1606 unlock_page(page);
1607
1608 if (unlikely(f2fs_cp_error(sbi))) {
1609 f2fs_submit_merged_write(sbi, NODE);
1610 submitted = NULL;
1611 }
1612 if (submitted)
1613 *submitted = fio.submitted;
1614
1615 if (do_balance)
1616 f2fs_balance_fs(sbi, false);
1617 return 0;
1618
1619 redirty_out:
1620 redirty_page_for_writepage(wbc, page);
1621 return AOP_WRITEPAGE_ACTIVATE;
1622 }
1623
f2fs_move_node_page(struct page * node_page,int gc_type)1624 int f2fs_move_node_page(struct page *node_page, int gc_type)
1625 {
1626 int err = 0;
1627
1628 if (gc_type == FG_GC) {
1629 struct writeback_control wbc = {
1630 .sync_mode = WB_SYNC_ALL,
1631 .nr_to_write = 1,
1632 .for_reclaim = 0,
1633 };
1634
1635 f2fs_wait_on_page_writeback(node_page, NODE, true, true);
1636
1637 set_page_dirty(node_page);
1638
1639 if (!clear_page_dirty_for_io(node_page)) {
1640 err = -EAGAIN;
1641 goto out_page;
1642 }
1643
1644 if (__write_node_page(node_page, false, NULL,
1645 &wbc, false, FS_GC_NODE_IO, NULL)) {
1646 err = -EAGAIN;
1647 unlock_page(node_page);
1648 }
1649 goto release_page;
1650 } else {
1651 /* set page dirty and write it */
1652 if (!PageWriteback(node_page))
1653 set_page_dirty(node_page);
1654 }
1655 out_page:
1656 unlock_page(node_page);
1657 release_page:
1658 f2fs_put_page(node_page, 0);
1659 return err;
1660 }
1661
f2fs_write_node_page(struct page * page,struct writeback_control * wbc)1662 static int f2fs_write_node_page(struct page *page,
1663 struct writeback_control *wbc)
1664 {
1665 return __write_node_page(page, false, NULL, wbc, false,
1666 FS_NODE_IO, NULL);
1667 }
1668
f2fs_fsync_node_pages(struct f2fs_sb_info * sbi,struct inode * inode,struct writeback_control * wbc,bool atomic,unsigned int * seq_id)1669 int f2fs_fsync_node_pages(struct f2fs_sb_info *sbi, struct inode *inode,
1670 struct writeback_control *wbc, bool atomic,
1671 unsigned int *seq_id)
1672 {
1673 pgoff_t index;
1674 struct pagevec pvec;
1675 int ret = 0;
1676 struct page *last_page = NULL;
1677 bool marked = false;
1678 nid_t ino = inode->i_ino;
1679 int nr_pages;
1680 int nwritten = 0;
1681
1682 if (atomic) {
1683 last_page = last_fsync_dnode(sbi, ino);
1684 if (IS_ERR_OR_NULL(last_page))
1685 return PTR_ERR_OR_ZERO(last_page);
1686 }
1687 retry:
1688 pagevec_init(&pvec);
1689 index = 0;
1690
1691 while ((nr_pages = pagevec_lookup_tag(&pvec, NODE_MAPPING(sbi), &index,
1692 PAGECACHE_TAG_DIRTY))) {
1693 int i;
1694
1695 for (i = 0; i < nr_pages; i++) {
1696 struct page *page = pvec.pages[i];
1697 bool submitted = false;
1698
1699 if (unlikely(f2fs_cp_error(sbi))) {
1700 f2fs_put_page(last_page, 0);
1701 pagevec_release(&pvec);
1702 ret = -EIO;
1703 goto out;
1704 }
1705
1706 if (!IS_DNODE(page) || !is_cold_node(page))
1707 continue;
1708 if (ino_of_node(page) != ino)
1709 continue;
1710
1711 lock_page(page);
1712
1713 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1714 continue_unlock:
1715 unlock_page(page);
1716 continue;
1717 }
1718 if (ino_of_node(page) != ino)
1719 goto continue_unlock;
1720
1721 if (!PageDirty(page) && page != last_page) {
1722 /* someone wrote it for us */
1723 goto continue_unlock;
1724 }
1725
1726 f2fs_wait_on_page_writeback(page, NODE, true, true);
1727
1728 set_fsync_mark(page, 0);
1729 set_dentry_mark(page, 0);
1730
1731 if (!atomic || page == last_page) {
1732 set_fsync_mark(page, 1);
1733 if (IS_INODE(page)) {
1734 if (is_inode_flag_set(inode,
1735 FI_DIRTY_INODE))
1736 f2fs_update_inode(inode, page);
1737 set_dentry_mark(page,
1738 f2fs_need_dentry_mark(sbi, ino));
1739 }
1740 /* may be written by other thread */
1741 if (!PageDirty(page))
1742 set_page_dirty(page);
1743 }
1744
1745 if (!clear_page_dirty_for_io(page))
1746 goto continue_unlock;
1747
1748 ret = __write_node_page(page, atomic &&
1749 page == last_page,
1750 &submitted, wbc, true,
1751 FS_NODE_IO, seq_id);
1752 if (ret) {
1753 unlock_page(page);
1754 f2fs_put_page(last_page, 0);
1755 break;
1756 } else if (submitted) {
1757 nwritten++;
1758 }
1759
1760 if (page == last_page) {
1761 f2fs_put_page(page, 0);
1762 marked = true;
1763 break;
1764 }
1765 }
1766 pagevec_release(&pvec);
1767 cond_resched();
1768
1769 if (ret || marked)
1770 break;
1771 }
1772 if (!ret && atomic && !marked) {
1773 f2fs_debug(sbi, "Retry to write fsync mark: ino=%u, idx=%lx",
1774 ino, last_page->index);
1775 lock_page(last_page);
1776 f2fs_wait_on_page_writeback(last_page, NODE, true, true);
1777 set_page_dirty(last_page);
1778 unlock_page(last_page);
1779 goto retry;
1780 }
1781 out:
1782 if (nwritten)
1783 f2fs_submit_merged_write_cond(sbi, NULL, NULL, ino, NODE);
1784 return ret ? -EIO: 0;
1785 }
1786
f2fs_match_ino(struct inode * inode,unsigned long ino,void * data)1787 static int f2fs_match_ino(struct inode *inode, unsigned long ino, void *data)
1788 {
1789 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
1790 bool clean;
1791
1792 if (inode->i_ino != ino)
1793 return 0;
1794
1795 if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
1796 return 0;
1797
1798 spin_lock(&sbi->inode_lock[DIRTY_META]);
1799 clean = list_empty(&F2FS_I(inode)->gdirty_list);
1800 spin_unlock(&sbi->inode_lock[DIRTY_META]);
1801
1802 if (clean)
1803 return 0;
1804
1805 inode = igrab(inode);
1806 if (!inode)
1807 return 0;
1808 return 1;
1809 }
1810
flush_dirty_inode(struct page * page)1811 static bool flush_dirty_inode(struct page *page)
1812 {
1813 struct f2fs_sb_info *sbi = F2FS_P_SB(page);
1814 struct inode *inode;
1815 nid_t ino = ino_of_node(page);
1816
1817 inode = find_inode_nowait(sbi->sb, ino, f2fs_match_ino, NULL);
1818 if (!inode)
1819 return false;
1820
1821 f2fs_update_inode(inode, page);
1822 unlock_page(page);
1823
1824 iput(inode);
1825 return true;
1826 }
1827
f2fs_flush_inline_data(struct f2fs_sb_info * sbi)1828 int f2fs_flush_inline_data(struct f2fs_sb_info *sbi)
1829 {
1830 pgoff_t index = 0;
1831 struct pagevec pvec;
1832 int nr_pages;
1833 int ret = 0;
1834
1835 pagevec_init(&pvec);
1836
1837 while ((nr_pages = pagevec_lookup_tag(&pvec,
1838 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1839 int i;
1840
1841 for (i = 0; i < nr_pages; i++) {
1842 struct page *page = pvec.pages[i];
1843
1844 if (!IS_DNODE(page))
1845 continue;
1846
1847 lock_page(page);
1848
1849 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1850 continue_unlock:
1851 unlock_page(page);
1852 continue;
1853 }
1854
1855 if (!PageDirty(page)) {
1856 /* someone wrote it for us */
1857 goto continue_unlock;
1858 }
1859
1860 /* flush inline_data, if it's async context. */
1861 if (is_inline_node(page)) {
1862 clear_inline_node(page);
1863 unlock_page(page);
1864 flush_inline_data(sbi, ino_of_node(page));
1865 continue;
1866 }
1867 unlock_page(page);
1868 }
1869 pagevec_release(&pvec);
1870 cond_resched();
1871 }
1872 return ret;
1873 }
1874
f2fs_sync_node_pages(struct f2fs_sb_info * sbi,struct writeback_control * wbc,bool do_balance,enum iostat_type io_type)1875 int f2fs_sync_node_pages(struct f2fs_sb_info *sbi,
1876 struct writeback_control *wbc,
1877 bool do_balance, enum iostat_type io_type)
1878 {
1879 pgoff_t index;
1880 struct pagevec pvec;
1881 int step = 0;
1882 int nwritten = 0;
1883 int ret = 0;
1884 int nr_pages, done = 0;
1885
1886 pagevec_init(&pvec);
1887
1888 next_step:
1889 index = 0;
1890
1891 while (!done && (nr_pages = pagevec_lookup_tag(&pvec,
1892 NODE_MAPPING(sbi), &index, PAGECACHE_TAG_DIRTY))) {
1893 int i;
1894
1895 for (i = 0; i < nr_pages; i++) {
1896 struct page *page = pvec.pages[i];
1897 bool submitted = false;
1898 bool may_dirty = true;
1899
1900 /* give a priority to WB_SYNC threads */
1901 if (atomic_read(&sbi->wb_sync_req[NODE]) &&
1902 wbc->sync_mode == WB_SYNC_NONE) {
1903 done = 1;
1904 break;
1905 }
1906
1907 /*
1908 * flushing sequence with step:
1909 * 0. indirect nodes
1910 * 1. dentry dnodes
1911 * 2. file dnodes
1912 */
1913 if (step == 0 && IS_DNODE(page))
1914 continue;
1915 if (step == 1 && (!IS_DNODE(page) ||
1916 is_cold_node(page)))
1917 continue;
1918 if (step == 2 && (!IS_DNODE(page) ||
1919 !is_cold_node(page)))
1920 continue;
1921 lock_node:
1922 if (wbc->sync_mode == WB_SYNC_ALL)
1923 lock_page(page);
1924 else if (!trylock_page(page))
1925 continue;
1926
1927 if (unlikely(page->mapping != NODE_MAPPING(sbi))) {
1928 continue_unlock:
1929 unlock_page(page);
1930 continue;
1931 }
1932
1933 if (!PageDirty(page)) {
1934 /* someone wrote it for us */
1935 goto continue_unlock;
1936 }
1937
1938 /* flush inline_data, if it's async context. */
1939 if (do_balance && is_inline_node(page)) {
1940 clear_inline_node(page);
1941 unlock_page(page);
1942 flush_inline_data(sbi, ino_of_node(page));
1943 goto lock_node;
1944 }
1945
1946 /* flush dirty inode */
1947 if (IS_INODE(page) && may_dirty) {
1948 may_dirty = false;
1949 if (flush_dirty_inode(page))
1950 goto lock_node;
1951 }
1952
1953 f2fs_wait_on_page_writeback(page, NODE, true, true);
1954
1955 if (!clear_page_dirty_for_io(page))
1956 goto continue_unlock;
1957
1958 set_fsync_mark(page, 0);
1959 set_dentry_mark(page, 0);
1960
1961 ret = __write_node_page(page, false, &submitted,
1962 wbc, do_balance, io_type, NULL);
1963 if (ret)
1964 unlock_page(page);
1965 else if (submitted)
1966 nwritten++;
1967
1968 if (--wbc->nr_to_write == 0)
1969 break;
1970 }
1971 pagevec_release(&pvec);
1972 cond_resched();
1973
1974 if (wbc->nr_to_write == 0) {
1975 step = 2;
1976 break;
1977 }
1978 }
1979
1980 if (step < 2) {
1981 if (!is_sbi_flag_set(sbi, SBI_CP_DISABLED) &&
1982 wbc->sync_mode == WB_SYNC_NONE && step == 1)
1983 goto out;
1984 step++;
1985 goto next_step;
1986 }
1987 out:
1988 if (nwritten)
1989 f2fs_submit_merged_write(sbi, NODE);
1990
1991 if (unlikely(f2fs_cp_error(sbi)))
1992 return -EIO;
1993 return ret;
1994 }
1995
f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info * sbi,unsigned int seq_id)1996 int f2fs_wait_on_node_pages_writeback(struct f2fs_sb_info *sbi,
1997 unsigned int seq_id)
1998 {
1999 struct fsync_node_entry *fn;
2000 struct page *page;
2001 struct list_head *head = &sbi->fsync_node_list;
2002 unsigned long flags;
2003 unsigned int cur_seq_id = 0;
2004 int ret2, ret = 0;
2005
2006 while (seq_id && cur_seq_id < seq_id) {
2007 spin_lock_irqsave(&sbi->fsync_node_lock, flags);
2008 if (list_empty(head)) {
2009 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2010 break;
2011 }
2012 fn = list_first_entry(head, struct fsync_node_entry, list);
2013 if (fn->seq_id > seq_id) {
2014 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2015 break;
2016 }
2017 cur_seq_id = fn->seq_id;
2018 page = fn->page;
2019 get_page(page);
2020 spin_unlock_irqrestore(&sbi->fsync_node_lock, flags);
2021
2022 f2fs_wait_on_page_writeback(page, NODE, true, false);
2023 if (TestClearPageError(page))
2024 ret = -EIO;
2025
2026 put_page(page);
2027
2028 if (ret)
2029 break;
2030 }
2031
2032 ret2 = filemap_check_errors(NODE_MAPPING(sbi));
2033 if (!ret)
2034 ret = ret2;
2035
2036 return ret;
2037 }
2038
f2fs_write_node_pages(struct address_space * mapping,struct writeback_control * wbc)2039 static int f2fs_write_node_pages(struct address_space *mapping,
2040 struct writeback_control *wbc)
2041 {
2042 struct f2fs_sb_info *sbi = F2FS_M_SB(mapping);
2043 struct blk_plug plug;
2044 long diff;
2045
2046 if (unlikely(is_sbi_flag_set(sbi, SBI_POR_DOING)))
2047 goto skip_write;
2048
2049 /* balancing f2fs's metadata in background */
2050 f2fs_balance_fs_bg(sbi, true);
2051
2052 /* collect a number of dirty node pages and write together */
2053 if (wbc->sync_mode != WB_SYNC_ALL &&
2054 get_pages(sbi, F2FS_DIRTY_NODES) <
2055 nr_pages_to_skip(sbi, NODE))
2056 goto skip_write;
2057
2058 if (wbc->sync_mode == WB_SYNC_ALL)
2059 atomic_inc(&sbi->wb_sync_req[NODE]);
2060 else if (atomic_read(&sbi->wb_sync_req[NODE])) {
2061 /* to avoid potential deadlock */
2062 if (current->plug)
2063 blk_finish_plug(current->plug);
2064 goto skip_write;
2065 }
2066
2067 trace_f2fs_writepages(mapping->host, wbc, NODE);
2068
2069 diff = nr_pages_to_write(sbi, NODE, wbc);
2070 blk_start_plug(&plug);
2071 f2fs_sync_node_pages(sbi, wbc, true, FS_NODE_IO);
2072 blk_finish_plug(&plug);
2073 wbc->nr_to_write = max((long)0, wbc->nr_to_write - diff);
2074
2075 if (wbc->sync_mode == WB_SYNC_ALL)
2076 atomic_dec(&sbi->wb_sync_req[NODE]);
2077 return 0;
2078
2079 skip_write:
2080 wbc->pages_skipped += get_pages(sbi, F2FS_DIRTY_NODES);
2081 trace_f2fs_writepages(mapping->host, wbc, NODE);
2082 return 0;
2083 }
2084
f2fs_set_node_page_dirty(struct page * page)2085 static int f2fs_set_node_page_dirty(struct page *page)
2086 {
2087 trace_f2fs_set_page_dirty(page, NODE);
2088
2089 if (!PageUptodate(page))
2090 SetPageUptodate(page);
2091 #ifdef CONFIG_F2FS_CHECK_FS
2092 if (IS_INODE(page))
2093 f2fs_inode_chksum_set(F2FS_P_SB(page), page);
2094 #endif
2095 if (!PageDirty(page)) {
2096 __set_page_dirty_nobuffers(page);
2097 inc_page_count(F2FS_P_SB(page), F2FS_DIRTY_NODES);
2098 f2fs_set_page_private(page, 0);
2099 f2fs_trace_pid(page);
2100 return 1;
2101 }
2102 return 0;
2103 }
2104
2105 /*
2106 * Structure of the f2fs node operations
2107 */
2108 const struct address_space_operations f2fs_node_aops = {
2109 .writepage = f2fs_write_node_page,
2110 .writepages = f2fs_write_node_pages,
2111 .set_page_dirty = f2fs_set_node_page_dirty,
2112 .invalidatepage = f2fs_invalidate_page,
2113 .releasepage = f2fs_release_page,
2114 #ifdef CONFIG_MIGRATION
2115 .migratepage = f2fs_migrate_page,
2116 #endif
2117 };
2118
__lookup_free_nid_list(struct f2fs_nm_info * nm_i,nid_t n)2119 static struct free_nid *__lookup_free_nid_list(struct f2fs_nm_info *nm_i,
2120 nid_t n)
2121 {
2122 return radix_tree_lookup(&nm_i->free_nid_root, n);
2123 }
2124
__insert_free_nid(struct f2fs_sb_info * sbi,struct free_nid * i,enum nid_state state)2125 static int __insert_free_nid(struct f2fs_sb_info *sbi,
2126 struct free_nid *i, enum nid_state state)
2127 {
2128 struct f2fs_nm_info *nm_i = NM_I(sbi);
2129
2130 int err = radix_tree_insert(&nm_i->free_nid_root, i->nid, i);
2131 if (err)
2132 return err;
2133
2134 f2fs_bug_on(sbi, state != i->state);
2135 nm_i->nid_cnt[state]++;
2136 if (state == FREE_NID)
2137 list_add_tail(&i->list, &nm_i->free_nid_list);
2138 return 0;
2139 }
2140
__remove_free_nid(struct f2fs_sb_info * sbi,struct free_nid * i,enum nid_state state)2141 static void __remove_free_nid(struct f2fs_sb_info *sbi,
2142 struct free_nid *i, enum nid_state state)
2143 {
2144 struct f2fs_nm_info *nm_i = NM_I(sbi);
2145
2146 f2fs_bug_on(sbi, state != i->state);
2147 nm_i->nid_cnt[state]--;
2148 if (state == FREE_NID)
2149 list_del(&i->list);
2150 radix_tree_delete(&nm_i->free_nid_root, i->nid);
2151 }
2152
__move_free_nid(struct f2fs_sb_info * sbi,struct free_nid * i,enum nid_state org_state,enum nid_state dst_state)2153 static void __move_free_nid(struct f2fs_sb_info *sbi, struct free_nid *i,
2154 enum nid_state org_state, enum nid_state dst_state)
2155 {
2156 struct f2fs_nm_info *nm_i = NM_I(sbi);
2157
2158 f2fs_bug_on(sbi, org_state != i->state);
2159 i->state = dst_state;
2160 nm_i->nid_cnt[org_state]--;
2161 nm_i->nid_cnt[dst_state]++;
2162
2163 switch (dst_state) {
2164 case PREALLOC_NID:
2165 list_del(&i->list);
2166 break;
2167 case FREE_NID:
2168 list_add_tail(&i->list, &nm_i->free_nid_list);
2169 break;
2170 default:
2171 BUG_ON(1);
2172 }
2173 }
2174
update_free_nid_bitmap(struct f2fs_sb_info * sbi,nid_t nid,bool set,bool build)2175 static void update_free_nid_bitmap(struct f2fs_sb_info *sbi, nid_t nid,
2176 bool set, bool build)
2177 {
2178 struct f2fs_nm_info *nm_i = NM_I(sbi);
2179 unsigned int nat_ofs = NAT_BLOCK_OFFSET(nid);
2180 unsigned int nid_ofs = nid - START_NID(nid);
2181
2182 if (!test_bit_le(nat_ofs, nm_i->nat_block_bitmap))
2183 return;
2184
2185 if (set) {
2186 if (test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2187 return;
2188 __set_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2189 nm_i->free_nid_count[nat_ofs]++;
2190 } else {
2191 if (!test_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]))
2192 return;
2193 __clear_bit_le(nid_ofs, nm_i->free_nid_bitmap[nat_ofs]);
2194 if (!build)
2195 nm_i->free_nid_count[nat_ofs]--;
2196 }
2197 }
2198
2199 /* return if the nid is recognized as free */
add_free_nid(struct f2fs_sb_info * sbi,nid_t nid,bool build,bool update)2200 static bool add_free_nid(struct f2fs_sb_info *sbi,
2201 nid_t nid, bool build, bool update)
2202 {
2203 struct f2fs_nm_info *nm_i = NM_I(sbi);
2204 struct free_nid *i, *e;
2205 struct nat_entry *ne;
2206 int err = -EINVAL;
2207 bool ret = false;
2208
2209 /* 0 nid should not be used */
2210 if (unlikely(nid == 0))
2211 return false;
2212
2213 if (unlikely(f2fs_check_nid_range(sbi, nid)))
2214 return false;
2215
2216 i = f2fs_kmem_cache_alloc(free_nid_slab, GFP_NOFS);
2217 i->nid = nid;
2218 i->state = FREE_NID;
2219
2220 radix_tree_preload(GFP_NOFS | __GFP_NOFAIL);
2221
2222 spin_lock(&nm_i->nid_list_lock);
2223
2224 if (build) {
2225 /*
2226 * Thread A Thread B
2227 * - f2fs_create
2228 * - f2fs_new_inode
2229 * - f2fs_alloc_nid
2230 * - __insert_nid_to_list(PREALLOC_NID)
2231 * - f2fs_balance_fs_bg
2232 * - f2fs_build_free_nids
2233 * - __f2fs_build_free_nids
2234 * - scan_nat_page
2235 * - add_free_nid
2236 * - __lookup_nat_cache
2237 * - f2fs_add_link
2238 * - f2fs_init_inode_metadata
2239 * - f2fs_new_inode_page
2240 * - f2fs_new_node_page
2241 * - set_node_addr
2242 * - f2fs_alloc_nid_done
2243 * - __remove_nid_from_list(PREALLOC_NID)
2244 * - __insert_nid_to_list(FREE_NID)
2245 */
2246 ne = __lookup_nat_cache(nm_i, nid);
2247 if (ne && (!get_nat_flag(ne, IS_CHECKPOINTED) ||
2248 nat_get_blkaddr(ne) != NULL_ADDR))
2249 goto err_out;
2250
2251 e = __lookup_free_nid_list(nm_i, nid);
2252 if (e) {
2253 if (e->state == FREE_NID)
2254 ret = true;
2255 goto err_out;
2256 }
2257 }
2258 ret = true;
2259 err = __insert_free_nid(sbi, i, FREE_NID);
2260 err_out:
2261 if (update) {
2262 update_free_nid_bitmap(sbi, nid, ret, build);
2263 if (!build)
2264 nm_i->available_nids++;
2265 }
2266 spin_unlock(&nm_i->nid_list_lock);
2267 radix_tree_preload_end();
2268
2269 if (err)
2270 kmem_cache_free(free_nid_slab, i);
2271 return ret;
2272 }
2273
remove_free_nid(struct f2fs_sb_info * sbi,nid_t nid)2274 static void remove_free_nid(struct f2fs_sb_info *sbi, nid_t nid)
2275 {
2276 struct f2fs_nm_info *nm_i = NM_I(sbi);
2277 struct free_nid *i;
2278 bool need_free = false;
2279
2280 spin_lock(&nm_i->nid_list_lock);
2281 i = __lookup_free_nid_list(nm_i, nid);
2282 if (i && i->state == FREE_NID) {
2283 __remove_free_nid(sbi, i, FREE_NID);
2284 need_free = true;
2285 }
2286 spin_unlock(&nm_i->nid_list_lock);
2287
2288 if (need_free)
2289 kmem_cache_free(free_nid_slab, i);
2290 }
2291
scan_nat_page(struct f2fs_sb_info * sbi,struct page * nat_page,nid_t start_nid)2292 static int scan_nat_page(struct f2fs_sb_info *sbi,
2293 struct page *nat_page, nid_t start_nid)
2294 {
2295 struct f2fs_nm_info *nm_i = NM_I(sbi);
2296 struct f2fs_nat_block *nat_blk = page_address(nat_page);
2297 block_t blk_addr;
2298 unsigned int nat_ofs = NAT_BLOCK_OFFSET(start_nid);
2299 int i;
2300
2301 __set_bit_le(nat_ofs, nm_i->nat_block_bitmap);
2302
2303 i = start_nid % NAT_ENTRY_PER_BLOCK;
2304
2305 for (; i < NAT_ENTRY_PER_BLOCK; i++, start_nid++) {
2306 if (unlikely(start_nid >= nm_i->max_nid))
2307 break;
2308
2309 blk_addr = le32_to_cpu(nat_blk->entries[i].block_addr);
2310
2311 if (blk_addr == NEW_ADDR)
2312 return -EINVAL;
2313
2314 if (blk_addr == NULL_ADDR) {
2315 add_free_nid(sbi, start_nid, true, true);
2316 } else {
2317 spin_lock(&NM_I(sbi)->nid_list_lock);
2318 update_free_nid_bitmap(sbi, start_nid, false, true);
2319 spin_unlock(&NM_I(sbi)->nid_list_lock);
2320 }
2321 }
2322
2323 return 0;
2324 }
2325
scan_curseg_cache(struct f2fs_sb_info * sbi)2326 static void scan_curseg_cache(struct f2fs_sb_info *sbi)
2327 {
2328 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2329 struct f2fs_journal *journal = curseg->journal;
2330 int i;
2331
2332 down_read(&curseg->journal_rwsem);
2333 for (i = 0; i < nats_in_cursum(journal); i++) {
2334 block_t addr;
2335 nid_t nid;
2336
2337 addr = le32_to_cpu(nat_in_journal(journal, i).block_addr);
2338 nid = le32_to_cpu(nid_in_journal(journal, i));
2339 if (addr == NULL_ADDR)
2340 add_free_nid(sbi, nid, true, false);
2341 else
2342 remove_free_nid(sbi, nid);
2343 }
2344 up_read(&curseg->journal_rwsem);
2345 }
2346
scan_free_nid_bits(struct f2fs_sb_info * sbi)2347 static void scan_free_nid_bits(struct f2fs_sb_info *sbi)
2348 {
2349 struct f2fs_nm_info *nm_i = NM_I(sbi);
2350 unsigned int i, idx;
2351 nid_t nid;
2352
2353 down_read(&nm_i->nat_tree_lock);
2354
2355 for (i = 0; i < nm_i->nat_blocks; i++) {
2356 if (!test_bit_le(i, nm_i->nat_block_bitmap))
2357 continue;
2358 if (!nm_i->free_nid_count[i])
2359 continue;
2360 for (idx = 0; idx < NAT_ENTRY_PER_BLOCK; idx++) {
2361 idx = find_next_bit_le(nm_i->free_nid_bitmap[i],
2362 NAT_ENTRY_PER_BLOCK, idx);
2363 if (idx >= NAT_ENTRY_PER_BLOCK)
2364 break;
2365
2366 nid = i * NAT_ENTRY_PER_BLOCK + idx;
2367 add_free_nid(sbi, nid, true, false);
2368
2369 if (nm_i->nid_cnt[FREE_NID] >= MAX_FREE_NIDS)
2370 goto out;
2371 }
2372 }
2373 out:
2374 scan_curseg_cache(sbi);
2375
2376 up_read(&nm_i->nat_tree_lock);
2377 }
2378
__f2fs_build_free_nids(struct f2fs_sb_info * sbi,bool sync,bool mount)2379 static int __f2fs_build_free_nids(struct f2fs_sb_info *sbi,
2380 bool sync, bool mount)
2381 {
2382 struct f2fs_nm_info *nm_i = NM_I(sbi);
2383 int i = 0, ret;
2384 nid_t nid = nm_i->next_scan_nid;
2385
2386 if (unlikely(nid >= nm_i->max_nid))
2387 nid = 0;
2388
2389 if (unlikely(nid % NAT_ENTRY_PER_BLOCK))
2390 nid = NAT_BLOCK_OFFSET(nid) * NAT_ENTRY_PER_BLOCK;
2391
2392 /* Enough entries */
2393 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2394 return 0;
2395
2396 if (!sync && !f2fs_available_free_memory(sbi, FREE_NIDS))
2397 return 0;
2398
2399 if (!mount) {
2400 /* try to find free nids in free_nid_bitmap */
2401 scan_free_nid_bits(sbi);
2402
2403 if (nm_i->nid_cnt[FREE_NID] >= NAT_ENTRY_PER_BLOCK)
2404 return 0;
2405 }
2406
2407 /* readahead nat pages to be scanned */
2408 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nid), FREE_NID_PAGES,
2409 META_NAT, true);
2410
2411 down_read(&nm_i->nat_tree_lock);
2412
2413 while (1) {
2414 if (!test_bit_le(NAT_BLOCK_OFFSET(nid),
2415 nm_i->nat_block_bitmap)) {
2416 struct page *page = get_current_nat_page(sbi, nid);
2417
2418 if (IS_ERR(page)) {
2419 ret = PTR_ERR(page);
2420 } else {
2421 ret = scan_nat_page(sbi, page, nid);
2422 f2fs_put_page(page, 1);
2423 }
2424
2425 if (ret) {
2426 up_read(&nm_i->nat_tree_lock);
2427 f2fs_err(sbi, "NAT is corrupt, run fsck to fix it");
2428 return ret;
2429 }
2430 }
2431
2432 nid += (NAT_ENTRY_PER_BLOCK - (nid % NAT_ENTRY_PER_BLOCK));
2433 if (unlikely(nid >= nm_i->max_nid))
2434 nid = 0;
2435
2436 if (++i >= FREE_NID_PAGES)
2437 break;
2438 }
2439
2440 /* go to the next free nat pages to find free nids abundantly */
2441 nm_i->next_scan_nid = nid;
2442
2443 /* find free nids from current sum_pages */
2444 scan_curseg_cache(sbi);
2445
2446 up_read(&nm_i->nat_tree_lock);
2447
2448 f2fs_ra_meta_pages(sbi, NAT_BLOCK_OFFSET(nm_i->next_scan_nid),
2449 nm_i->ra_nid_pages, META_NAT, false);
2450
2451 return 0;
2452 }
2453
f2fs_build_free_nids(struct f2fs_sb_info * sbi,bool sync,bool mount)2454 int f2fs_build_free_nids(struct f2fs_sb_info *sbi, bool sync, bool mount)
2455 {
2456 int ret;
2457
2458 mutex_lock(&NM_I(sbi)->build_lock);
2459 ret = __f2fs_build_free_nids(sbi, sync, mount);
2460 mutex_unlock(&NM_I(sbi)->build_lock);
2461
2462 return ret;
2463 }
2464
2465 /*
2466 * If this function returns success, caller can obtain a new nid
2467 * from second parameter of this function.
2468 * The returned nid could be used ino as well as nid when inode is created.
2469 */
f2fs_alloc_nid(struct f2fs_sb_info * sbi,nid_t * nid)2470 bool f2fs_alloc_nid(struct f2fs_sb_info *sbi, nid_t *nid)
2471 {
2472 struct f2fs_nm_info *nm_i = NM_I(sbi);
2473 struct free_nid *i = NULL;
2474 retry:
2475 if (time_to_inject(sbi, FAULT_ALLOC_NID)) {
2476 f2fs_show_injection_info(sbi, FAULT_ALLOC_NID);
2477 return false;
2478 }
2479
2480 spin_lock(&nm_i->nid_list_lock);
2481
2482 if (unlikely(nm_i->available_nids == 0)) {
2483 spin_unlock(&nm_i->nid_list_lock);
2484 return false;
2485 }
2486
2487 /* We should not use stale free nids created by f2fs_build_free_nids */
2488 if (nm_i->nid_cnt[FREE_NID] && !on_f2fs_build_free_nids(nm_i)) {
2489 f2fs_bug_on(sbi, list_empty(&nm_i->free_nid_list));
2490 i = list_first_entry(&nm_i->free_nid_list,
2491 struct free_nid, list);
2492 *nid = i->nid;
2493
2494 __move_free_nid(sbi, i, FREE_NID, PREALLOC_NID);
2495 nm_i->available_nids--;
2496
2497 update_free_nid_bitmap(sbi, *nid, false, false);
2498
2499 spin_unlock(&nm_i->nid_list_lock);
2500 return true;
2501 }
2502 spin_unlock(&nm_i->nid_list_lock);
2503
2504 /* Let's scan nat pages and its caches to get free nids */
2505 if (!f2fs_build_free_nids(sbi, true, false))
2506 goto retry;
2507 return false;
2508 }
2509
2510 /*
2511 * f2fs_alloc_nid() should be called prior to this function.
2512 */
f2fs_alloc_nid_done(struct f2fs_sb_info * sbi,nid_t nid)2513 void f2fs_alloc_nid_done(struct f2fs_sb_info *sbi, nid_t nid)
2514 {
2515 struct f2fs_nm_info *nm_i = NM_I(sbi);
2516 struct free_nid *i;
2517
2518 spin_lock(&nm_i->nid_list_lock);
2519 i = __lookup_free_nid_list(nm_i, nid);
2520 f2fs_bug_on(sbi, !i);
2521 __remove_free_nid(sbi, i, PREALLOC_NID);
2522 spin_unlock(&nm_i->nid_list_lock);
2523
2524 kmem_cache_free(free_nid_slab, i);
2525 }
2526
2527 /*
2528 * f2fs_alloc_nid() should be called prior to this function.
2529 */
f2fs_alloc_nid_failed(struct f2fs_sb_info * sbi,nid_t nid)2530 void f2fs_alloc_nid_failed(struct f2fs_sb_info *sbi, nid_t nid)
2531 {
2532 struct f2fs_nm_info *nm_i = NM_I(sbi);
2533 struct free_nid *i;
2534 bool need_free = false;
2535
2536 if (!nid)
2537 return;
2538
2539 spin_lock(&nm_i->nid_list_lock);
2540 i = __lookup_free_nid_list(nm_i, nid);
2541 f2fs_bug_on(sbi, !i);
2542
2543 if (!f2fs_available_free_memory(sbi, FREE_NIDS)) {
2544 __remove_free_nid(sbi, i, PREALLOC_NID);
2545 need_free = true;
2546 } else {
2547 __move_free_nid(sbi, i, PREALLOC_NID, FREE_NID);
2548 }
2549
2550 nm_i->available_nids++;
2551
2552 update_free_nid_bitmap(sbi, nid, true, false);
2553
2554 spin_unlock(&nm_i->nid_list_lock);
2555
2556 if (need_free)
2557 kmem_cache_free(free_nid_slab, i);
2558 }
2559
f2fs_try_to_free_nids(struct f2fs_sb_info * sbi,int nr_shrink)2560 int f2fs_try_to_free_nids(struct f2fs_sb_info *sbi, int nr_shrink)
2561 {
2562 struct f2fs_nm_info *nm_i = NM_I(sbi);
2563 int nr = nr_shrink;
2564
2565 if (nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2566 return 0;
2567
2568 if (!mutex_trylock(&nm_i->build_lock))
2569 return 0;
2570
2571 while (nr_shrink && nm_i->nid_cnt[FREE_NID] > MAX_FREE_NIDS) {
2572 struct free_nid *i, *next;
2573 unsigned int batch = SHRINK_NID_BATCH_SIZE;
2574
2575 spin_lock(&nm_i->nid_list_lock);
2576 list_for_each_entry_safe(i, next, &nm_i->free_nid_list, list) {
2577 if (!nr_shrink || !batch ||
2578 nm_i->nid_cnt[FREE_NID] <= MAX_FREE_NIDS)
2579 break;
2580 __remove_free_nid(sbi, i, FREE_NID);
2581 kmem_cache_free(free_nid_slab, i);
2582 nr_shrink--;
2583 batch--;
2584 }
2585 spin_unlock(&nm_i->nid_list_lock);
2586 }
2587
2588 mutex_unlock(&nm_i->build_lock);
2589
2590 return nr - nr_shrink;
2591 }
2592
f2fs_recover_inline_xattr(struct inode * inode,struct page * page)2593 int f2fs_recover_inline_xattr(struct inode *inode, struct page *page)
2594 {
2595 void *src_addr, *dst_addr;
2596 size_t inline_size;
2597 struct page *ipage;
2598 struct f2fs_inode *ri;
2599
2600 ipage = f2fs_get_node_page(F2FS_I_SB(inode), inode->i_ino);
2601 if (IS_ERR(ipage))
2602 return PTR_ERR(ipage);
2603
2604 ri = F2FS_INODE(page);
2605 if (ri->i_inline & F2FS_INLINE_XATTR) {
2606 set_inode_flag(inode, FI_INLINE_XATTR);
2607 } else {
2608 clear_inode_flag(inode, FI_INLINE_XATTR);
2609 goto update_inode;
2610 }
2611
2612 dst_addr = inline_xattr_addr(inode, ipage);
2613 src_addr = inline_xattr_addr(inode, page);
2614 inline_size = inline_xattr_size(inode);
2615
2616 f2fs_wait_on_page_writeback(ipage, NODE, true, true);
2617 memcpy(dst_addr, src_addr, inline_size);
2618 update_inode:
2619 f2fs_update_inode(inode, ipage);
2620 f2fs_put_page(ipage, 1);
2621 return 0;
2622 }
2623
f2fs_recover_xattr_data(struct inode * inode,struct page * page)2624 int f2fs_recover_xattr_data(struct inode *inode, struct page *page)
2625 {
2626 struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
2627 nid_t prev_xnid = F2FS_I(inode)->i_xattr_nid;
2628 nid_t new_xnid;
2629 struct dnode_of_data dn;
2630 struct node_info ni;
2631 struct page *xpage;
2632 int err;
2633
2634 if (!prev_xnid)
2635 goto recover_xnid;
2636
2637 /* 1: invalidate the previous xattr nid */
2638 err = f2fs_get_node_info(sbi, prev_xnid, &ni);
2639 if (err)
2640 return err;
2641
2642 f2fs_invalidate_blocks(sbi, ni.blk_addr);
2643 dec_valid_node_count(sbi, inode, false);
2644 set_node_addr(sbi, &ni, NULL_ADDR, false);
2645
2646 recover_xnid:
2647 /* 2: update xattr nid in inode */
2648 if (!f2fs_alloc_nid(sbi, &new_xnid))
2649 return -ENOSPC;
2650
2651 set_new_dnode(&dn, inode, NULL, NULL, new_xnid);
2652 xpage = f2fs_new_node_page(&dn, XATTR_NODE_OFFSET);
2653 if (IS_ERR(xpage)) {
2654 f2fs_alloc_nid_failed(sbi, new_xnid);
2655 return PTR_ERR(xpage);
2656 }
2657
2658 f2fs_alloc_nid_done(sbi, new_xnid);
2659 f2fs_update_inode_page(inode);
2660
2661 /* 3: update and set xattr node page dirty */
2662 memcpy(F2FS_NODE(xpage), F2FS_NODE(page), VALID_XATTR_BLOCK_SIZE);
2663
2664 set_page_dirty(xpage);
2665 f2fs_put_page(xpage, 1);
2666
2667 return 0;
2668 }
2669
f2fs_recover_inode_page(struct f2fs_sb_info * sbi,struct page * page)2670 int f2fs_recover_inode_page(struct f2fs_sb_info *sbi, struct page *page)
2671 {
2672 struct f2fs_inode *src, *dst;
2673 nid_t ino = ino_of_node(page);
2674 struct node_info old_ni, new_ni;
2675 struct page *ipage;
2676 int err;
2677
2678 err = f2fs_get_node_info(sbi, ino, &old_ni);
2679 if (err)
2680 return err;
2681
2682 if (unlikely(old_ni.blk_addr != NULL_ADDR))
2683 return -EINVAL;
2684 retry:
2685 ipage = f2fs_grab_cache_page(NODE_MAPPING(sbi), ino, false);
2686 if (!ipage) {
2687 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
2688 goto retry;
2689 }
2690
2691 /* Should not use this inode from free nid list */
2692 remove_free_nid(sbi, ino);
2693
2694 if (!PageUptodate(ipage))
2695 SetPageUptodate(ipage);
2696 fill_node_footer(ipage, ino, ino, 0, true);
2697 set_cold_node(ipage, false);
2698
2699 src = F2FS_INODE(page);
2700 dst = F2FS_INODE(ipage);
2701
2702 memcpy(dst, src, (unsigned long)&src->i_ext - (unsigned long)src);
2703 dst->i_size = 0;
2704 dst->i_blocks = cpu_to_le64(1);
2705 dst->i_links = cpu_to_le32(1);
2706 dst->i_xattr_nid = 0;
2707 dst->i_inline = src->i_inline & (F2FS_INLINE_XATTR | F2FS_EXTRA_ATTR);
2708 if (dst->i_inline & F2FS_EXTRA_ATTR) {
2709 dst->i_extra_isize = src->i_extra_isize;
2710
2711 if (f2fs_sb_has_flexible_inline_xattr(sbi) &&
2712 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2713 i_inline_xattr_size))
2714 dst->i_inline_xattr_size = src->i_inline_xattr_size;
2715
2716 if (f2fs_sb_has_project_quota(sbi) &&
2717 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2718 i_projid))
2719 dst->i_projid = src->i_projid;
2720
2721 if (f2fs_sb_has_inode_crtime(sbi) &&
2722 F2FS_FITS_IN_INODE(src, le16_to_cpu(src->i_extra_isize),
2723 i_crtime_nsec)) {
2724 dst->i_crtime = src->i_crtime;
2725 dst->i_crtime_nsec = src->i_crtime_nsec;
2726 }
2727 }
2728
2729 new_ni = old_ni;
2730 new_ni.ino = ino;
2731
2732 if (unlikely(inc_valid_node_count(sbi, NULL, true)))
2733 WARN_ON(1);
2734 set_node_addr(sbi, &new_ni, NEW_ADDR, false);
2735 inc_valid_inode_count(sbi);
2736 set_page_dirty(ipage);
2737 f2fs_put_page(ipage, 1);
2738 return 0;
2739 }
2740
f2fs_restore_node_summary(struct f2fs_sb_info * sbi,unsigned int segno,struct f2fs_summary_block * sum)2741 int f2fs_restore_node_summary(struct f2fs_sb_info *sbi,
2742 unsigned int segno, struct f2fs_summary_block *sum)
2743 {
2744 struct f2fs_node *rn;
2745 struct f2fs_summary *sum_entry;
2746 block_t addr;
2747 int i, idx, last_offset, nrpages;
2748
2749 /* scan the node segment */
2750 last_offset = sbi->blocks_per_seg;
2751 addr = START_BLOCK(sbi, segno);
2752 sum_entry = &sum->entries[0];
2753
2754 for (i = 0; i < last_offset; i += nrpages, addr += nrpages) {
2755 nrpages = min(last_offset - i, BIO_MAX_PAGES);
2756
2757 /* readahead node pages */
2758 f2fs_ra_meta_pages(sbi, addr, nrpages, META_POR, true);
2759
2760 for (idx = addr; idx < addr + nrpages; idx++) {
2761 struct page *page = f2fs_get_tmp_page(sbi, idx);
2762
2763 if (IS_ERR(page))
2764 return PTR_ERR(page);
2765
2766 rn = F2FS_NODE(page);
2767 sum_entry->nid = rn->footer.nid;
2768 sum_entry->version = 0;
2769 sum_entry->ofs_in_node = 0;
2770 sum_entry++;
2771 f2fs_put_page(page, 1);
2772 }
2773
2774 invalidate_mapping_pages(META_MAPPING(sbi), addr,
2775 addr + nrpages);
2776 }
2777 return 0;
2778 }
2779
remove_nats_in_journal(struct f2fs_sb_info * sbi)2780 static void remove_nats_in_journal(struct f2fs_sb_info *sbi)
2781 {
2782 struct f2fs_nm_info *nm_i = NM_I(sbi);
2783 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2784 struct f2fs_journal *journal = curseg->journal;
2785 int i;
2786
2787 down_write(&curseg->journal_rwsem);
2788 for (i = 0; i < nats_in_cursum(journal); i++) {
2789 struct nat_entry *ne;
2790 struct f2fs_nat_entry raw_ne;
2791 nid_t nid = le32_to_cpu(nid_in_journal(journal, i));
2792
2793 if (f2fs_check_nid_range(sbi, nid))
2794 continue;
2795
2796 raw_ne = nat_in_journal(journal, i);
2797
2798 ne = __lookup_nat_cache(nm_i, nid);
2799 if (!ne) {
2800 ne = __alloc_nat_entry(nid, true);
2801 __init_nat_entry(nm_i, ne, &raw_ne, true);
2802 }
2803
2804 /*
2805 * if a free nat in journal has not been used after last
2806 * checkpoint, we should remove it from available nids,
2807 * since later we will add it again.
2808 */
2809 if (!get_nat_flag(ne, IS_DIRTY) &&
2810 le32_to_cpu(raw_ne.block_addr) == NULL_ADDR) {
2811 spin_lock(&nm_i->nid_list_lock);
2812 nm_i->available_nids--;
2813 spin_unlock(&nm_i->nid_list_lock);
2814 }
2815
2816 __set_nat_cache_dirty(nm_i, ne);
2817 }
2818 update_nats_in_cursum(journal, -i);
2819 up_write(&curseg->journal_rwsem);
2820 }
2821
__adjust_nat_entry_set(struct nat_entry_set * nes,struct list_head * head,int max)2822 static void __adjust_nat_entry_set(struct nat_entry_set *nes,
2823 struct list_head *head, int max)
2824 {
2825 struct nat_entry_set *cur;
2826
2827 if (nes->entry_cnt >= max)
2828 goto add_out;
2829
2830 list_for_each_entry(cur, head, set_list) {
2831 if (cur->entry_cnt >= nes->entry_cnt) {
2832 list_add(&nes->set_list, cur->set_list.prev);
2833 return;
2834 }
2835 }
2836 add_out:
2837 list_add_tail(&nes->set_list, head);
2838 }
2839
__update_nat_bits(struct f2fs_sb_info * sbi,nid_t start_nid,struct page * page)2840 static void __update_nat_bits(struct f2fs_sb_info *sbi, nid_t start_nid,
2841 struct page *page)
2842 {
2843 struct f2fs_nm_info *nm_i = NM_I(sbi);
2844 unsigned int nat_index = start_nid / NAT_ENTRY_PER_BLOCK;
2845 struct f2fs_nat_block *nat_blk = page_address(page);
2846 int valid = 0;
2847 int i = 0;
2848
2849 if (!enabled_nat_bits(sbi, NULL))
2850 return;
2851
2852 if (nat_index == 0) {
2853 valid = 1;
2854 i = 1;
2855 }
2856 for (; i < NAT_ENTRY_PER_BLOCK; i++) {
2857 if (le32_to_cpu(nat_blk->entries[i].block_addr) != NULL_ADDR)
2858 valid++;
2859 }
2860 if (valid == 0) {
2861 __set_bit_le(nat_index, nm_i->empty_nat_bits);
2862 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2863 return;
2864 }
2865
2866 __clear_bit_le(nat_index, nm_i->empty_nat_bits);
2867 if (valid == NAT_ENTRY_PER_BLOCK)
2868 __set_bit_le(nat_index, nm_i->full_nat_bits);
2869 else
2870 __clear_bit_le(nat_index, nm_i->full_nat_bits);
2871 }
2872
__flush_nat_entry_set(struct f2fs_sb_info * sbi,struct nat_entry_set * set,struct cp_control * cpc)2873 static int __flush_nat_entry_set(struct f2fs_sb_info *sbi,
2874 struct nat_entry_set *set, struct cp_control *cpc)
2875 {
2876 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2877 struct f2fs_journal *journal = curseg->journal;
2878 nid_t start_nid = set->set * NAT_ENTRY_PER_BLOCK;
2879 bool to_journal = true;
2880 struct f2fs_nat_block *nat_blk;
2881 struct nat_entry *ne, *cur;
2882 struct page *page = NULL;
2883
2884 /*
2885 * there are two steps to flush nat entries:
2886 * #1, flush nat entries to journal in current hot data summary block.
2887 * #2, flush nat entries to nat page.
2888 */
2889 if (enabled_nat_bits(sbi, cpc) ||
2890 !__has_cursum_space(journal, set->entry_cnt, NAT_JOURNAL))
2891 to_journal = false;
2892
2893 if (to_journal) {
2894 down_write(&curseg->journal_rwsem);
2895 } else {
2896 page = get_next_nat_page(sbi, start_nid);
2897 if (IS_ERR(page))
2898 return PTR_ERR(page);
2899
2900 nat_blk = page_address(page);
2901 f2fs_bug_on(sbi, !nat_blk);
2902 }
2903
2904 /* flush dirty nats in nat entry set */
2905 list_for_each_entry_safe(ne, cur, &set->entry_list, list) {
2906 struct f2fs_nat_entry *raw_ne;
2907 nid_t nid = nat_get_nid(ne);
2908 int offset;
2909
2910 f2fs_bug_on(sbi, nat_get_blkaddr(ne) == NEW_ADDR);
2911
2912 if (to_journal) {
2913 offset = f2fs_lookup_journal_in_cursum(journal,
2914 NAT_JOURNAL, nid, 1);
2915 f2fs_bug_on(sbi, offset < 0);
2916 raw_ne = &nat_in_journal(journal, offset);
2917 nid_in_journal(journal, offset) = cpu_to_le32(nid);
2918 } else {
2919 raw_ne = &nat_blk->entries[nid - start_nid];
2920 }
2921 raw_nat_from_node_info(raw_ne, &ne->ni);
2922 nat_reset_flag(ne);
2923 __clear_nat_cache_dirty(NM_I(sbi), set, ne);
2924 if (nat_get_blkaddr(ne) == NULL_ADDR) {
2925 add_free_nid(sbi, nid, false, true);
2926 } else {
2927 spin_lock(&NM_I(sbi)->nid_list_lock);
2928 update_free_nid_bitmap(sbi, nid, false, false);
2929 spin_unlock(&NM_I(sbi)->nid_list_lock);
2930 }
2931 }
2932
2933 if (to_journal) {
2934 up_write(&curseg->journal_rwsem);
2935 } else {
2936 __update_nat_bits(sbi, start_nid, page);
2937 f2fs_put_page(page, 1);
2938 }
2939
2940 /* Allow dirty nats by node block allocation in write_begin */
2941 if (!set->entry_cnt) {
2942 radix_tree_delete(&NM_I(sbi)->nat_set_root, set->set);
2943 kmem_cache_free(nat_entry_set_slab, set);
2944 }
2945 return 0;
2946 }
2947
2948 /*
2949 * This function is called during the checkpointing process.
2950 */
f2fs_flush_nat_entries(struct f2fs_sb_info * sbi,struct cp_control * cpc)2951 int f2fs_flush_nat_entries(struct f2fs_sb_info *sbi, struct cp_control *cpc)
2952 {
2953 struct f2fs_nm_info *nm_i = NM_I(sbi);
2954 struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_HOT_DATA);
2955 struct f2fs_journal *journal = curseg->journal;
2956 struct nat_entry_set *setvec[SETVEC_SIZE];
2957 struct nat_entry_set *set, *tmp;
2958 unsigned int found;
2959 nid_t set_idx = 0;
2960 LIST_HEAD(sets);
2961 int err = 0;
2962
2963 /*
2964 * during unmount, let's flush nat_bits before checking
2965 * nat_cnt[DIRTY_NAT].
2966 */
2967 if (enabled_nat_bits(sbi, cpc)) {
2968 down_write(&nm_i->nat_tree_lock);
2969 remove_nats_in_journal(sbi);
2970 up_write(&nm_i->nat_tree_lock);
2971 }
2972
2973 if (!nm_i->nat_cnt[DIRTY_NAT])
2974 return 0;
2975
2976 down_write(&nm_i->nat_tree_lock);
2977
2978 /*
2979 * if there are no enough space in journal to store dirty nat
2980 * entries, remove all entries from journal and merge them
2981 * into nat entry set.
2982 */
2983 if (enabled_nat_bits(sbi, cpc) ||
2984 !__has_cursum_space(journal,
2985 nm_i->nat_cnt[DIRTY_NAT], NAT_JOURNAL))
2986 remove_nats_in_journal(sbi);
2987
2988 while ((found = __gang_lookup_nat_set(nm_i,
2989 set_idx, SETVEC_SIZE, setvec))) {
2990 unsigned idx;
2991 set_idx = setvec[found - 1]->set + 1;
2992 for (idx = 0; idx < found; idx++)
2993 __adjust_nat_entry_set(setvec[idx], &sets,
2994 MAX_NAT_JENTRIES(journal));
2995 }
2996
2997 /* flush dirty nats in nat entry set */
2998 list_for_each_entry_safe(set, tmp, &sets, set_list) {
2999 err = __flush_nat_entry_set(sbi, set, cpc);
3000 if (err)
3001 break;
3002 }
3003
3004 up_write(&nm_i->nat_tree_lock);
3005 /* Allow dirty nats by node block allocation in write_begin */
3006
3007 return err;
3008 }
3009
__get_nat_bitmaps(struct f2fs_sb_info * sbi)3010 static int __get_nat_bitmaps(struct f2fs_sb_info *sbi)
3011 {
3012 struct f2fs_checkpoint *ckpt = F2FS_CKPT(sbi);
3013 struct f2fs_nm_info *nm_i = NM_I(sbi);
3014 unsigned int nat_bits_bytes = nm_i->nat_blocks / BITS_PER_BYTE;
3015 unsigned int i;
3016 __u64 cp_ver = cur_cp_version(ckpt);
3017 block_t nat_bits_addr;
3018
3019 if (!enabled_nat_bits(sbi, NULL))
3020 return 0;
3021
3022 nm_i->nat_bits_blocks = F2FS_BLK_ALIGN((nat_bits_bytes << 1) + 8);
3023 nm_i->nat_bits = f2fs_kvzalloc(sbi,
3024 nm_i->nat_bits_blocks << F2FS_BLKSIZE_BITS, GFP_KERNEL);
3025 if (!nm_i->nat_bits)
3026 return -ENOMEM;
3027
3028 nat_bits_addr = __start_cp_addr(sbi) + sbi->blocks_per_seg -
3029 nm_i->nat_bits_blocks;
3030 for (i = 0; i < nm_i->nat_bits_blocks; i++) {
3031 struct page *page;
3032
3033 page = f2fs_get_meta_page(sbi, nat_bits_addr++);
3034 if (IS_ERR(page))
3035 return PTR_ERR(page);
3036
3037 memcpy(nm_i->nat_bits + (i << F2FS_BLKSIZE_BITS),
3038 page_address(page), F2FS_BLKSIZE);
3039 f2fs_put_page(page, 1);
3040 }
3041
3042 cp_ver |= (cur_cp_crc(ckpt) << 32);
3043 if (cpu_to_le64(cp_ver) != *(__le64 *)nm_i->nat_bits) {
3044 disable_nat_bits(sbi, true);
3045 return 0;
3046 }
3047
3048 nm_i->full_nat_bits = nm_i->nat_bits + 8;
3049 nm_i->empty_nat_bits = nm_i->full_nat_bits + nat_bits_bytes;
3050
3051 f2fs_notice(sbi, "Found nat_bits in checkpoint");
3052 return 0;
3053 }
3054
load_free_nid_bitmap(struct f2fs_sb_info * sbi)3055 static inline void load_free_nid_bitmap(struct f2fs_sb_info *sbi)
3056 {
3057 struct f2fs_nm_info *nm_i = NM_I(sbi);
3058 unsigned int i = 0;
3059 nid_t nid, last_nid;
3060
3061 if (!enabled_nat_bits(sbi, NULL))
3062 return;
3063
3064 for (i = 0; i < nm_i->nat_blocks; i++) {
3065 i = find_next_bit_le(nm_i->empty_nat_bits, nm_i->nat_blocks, i);
3066 if (i >= nm_i->nat_blocks)
3067 break;
3068
3069 __set_bit_le(i, nm_i->nat_block_bitmap);
3070
3071 nid = i * NAT_ENTRY_PER_BLOCK;
3072 last_nid = nid + NAT_ENTRY_PER_BLOCK;
3073
3074 spin_lock(&NM_I(sbi)->nid_list_lock);
3075 for (; nid < last_nid; nid++)
3076 update_free_nid_bitmap(sbi, nid, true, true);
3077 spin_unlock(&NM_I(sbi)->nid_list_lock);
3078 }
3079
3080 for (i = 0; i < nm_i->nat_blocks; i++) {
3081 i = find_next_bit_le(nm_i->full_nat_bits, nm_i->nat_blocks, i);
3082 if (i >= nm_i->nat_blocks)
3083 break;
3084
3085 __set_bit_le(i, nm_i->nat_block_bitmap);
3086 }
3087 }
3088
init_node_manager(struct f2fs_sb_info * sbi)3089 static int init_node_manager(struct f2fs_sb_info *sbi)
3090 {
3091 struct f2fs_super_block *sb_raw = F2FS_RAW_SUPER(sbi);
3092 struct f2fs_nm_info *nm_i = NM_I(sbi);
3093 unsigned char *version_bitmap;
3094 unsigned int nat_segs;
3095 int err;
3096
3097 nm_i->nat_blkaddr = le32_to_cpu(sb_raw->nat_blkaddr);
3098
3099 /* segment_count_nat includes pair segment so divide to 2. */
3100 nat_segs = le32_to_cpu(sb_raw->segment_count_nat) >> 1;
3101 nm_i->nat_blocks = nat_segs << le32_to_cpu(sb_raw->log_blocks_per_seg);
3102 nm_i->max_nid = NAT_ENTRY_PER_BLOCK * nm_i->nat_blocks;
3103
3104 /* not used nids: 0, node, meta, (and root counted as valid node) */
3105 nm_i->available_nids = nm_i->max_nid - sbi->total_valid_node_count -
3106 F2FS_RESERVED_NODE_NUM;
3107 nm_i->nid_cnt[FREE_NID] = 0;
3108 nm_i->nid_cnt[PREALLOC_NID] = 0;
3109 nm_i->ram_thresh = DEF_RAM_THRESHOLD;
3110 nm_i->ra_nid_pages = DEF_RA_NID_PAGES;
3111 nm_i->dirty_nats_ratio = DEF_DIRTY_NAT_RATIO_THRESHOLD;
3112
3113 INIT_RADIX_TREE(&nm_i->free_nid_root, GFP_ATOMIC);
3114 INIT_LIST_HEAD(&nm_i->free_nid_list);
3115 INIT_RADIX_TREE(&nm_i->nat_root, GFP_NOIO);
3116 INIT_RADIX_TREE(&nm_i->nat_set_root, GFP_NOIO);
3117 INIT_LIST_HEAD(&nm_i->nat_entries);
3118 spin_lock_init(&nm_i->nat_list_lock);
3119
3120 mutex_init(&nm_i->build_lock);
3121 spin_lock_init(&nm_i->nid_list_lock);
3122 init_rwsem(&nm_i->nat_tree_lock);
3123
3124 nm_i->next_scan_nid = le32_to_cpu(sbi->ckpt->next_free_nid);
3125 nm_i->bitmap_size = __bitmap_size(sbi, NAT_BITMAP);
3126 version_bitmap = __bitmap_ptr(sbi, NAT_BITMAP);
3127 if (!version_bitmap)
3128 return -EFAULT;
3129
3130 nm_i->nat_bitmap = kmemdup(version_bitmap, nm_i->bitmap_size,
3131 GFP_KERNEL);
3132 if (!nm_i->nat_bitmap)
3133 return -ENOMEM;
3134
3135 err = __get_nat_bitmaps(sbi);
3136 if (err)
3137 return err;
3138
3139 #ifdef CONFIG_F2FS_CHECK_FS
3140 nm_i->nat_bitmap_mir = kmemdup(version_bitmap, nm_i->bitmap_size,
3141 GFP_KERNEL);
3142 if (!nm_i->nat_bitmap_mir)
3143 return -ENOMEM;
3144 #endif
3145
3146 return 0;
3147 }
3148
init_free_nid_cache(struct f2fs_sb_info * sbi)3149 static int init_free_nid_cache(struct f2fs_sb_info *sbi)
3150 {
3151 struct f2fs_nm_info *nm_i = NM_I(sbi);
3152 int i;
3153
3154 nm_i->free_nid_bitmap =
3155 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned char *),
3156 nm_i->nat_blocks),
3157 GFP_KERNEL);
3158 if (!nm_i->free_nid_bitmap)
3159 return -ENOMEM;
3160
3161 for (i = 0; i < nm_i->nat_blocks; i++) {
3162 nm_i->free_nid_bitmap[i] = f2fs_kvzalloc(sbi,
3163 f2fs_bitmap_size(NAT_ENTRY_PER_BLOCK), GFP_KERNEL);
3164 if (!nm_i->free_nid_bitmap[i])
3165 return -ENOMEM;
3166 }
3167
3168 nm_i->nat_block_bitmap = f2fs_kvzalloc(sbi, nm_i->nat_blocks / 8,
3169 GFP_KERNEL);
3170 if (!nm_i->nat_block_bitmap)
3171 return -ENOMEM;
3172
3173 nm_i->free_nid_count =
3174 f2fs_kvzalloc(sbi, array_size(sizeof(unsigned short),
3175 nm_i->nat_blocks),
3176 GFP_KERNEL);
3177 if (!nm_i->free_nid_count)
3178 return -ENOMEM;
3179 return 0;
3180 }
3181
f2fs_build_node_manager(struct f2fs_sb_info * sbi)3182 int f2fs_build_node_manager(struct f2fs_sb_info *sbi)
3183 {
3184 int err;
3185
3186 sbi->nm_info = f2fs_kzalloc(sbi, sizeof(struct f2fs_nm_info),
3187 GFP_KERNEL);
3188 if (!sbi->nm_info)
3189 return -ENOMEM;
3190
3191 err = init_node_manager(sbi);
3192 if (err)
3193 return err;
3194
3195 err = init_free_nid_cache(sbi);
3196 if (err)
3197 return err;
3198
3199 /* load free nid status from nat_bits table */
3200 load_free_nid_bitmap(sbi);
3201
3202 return f2fs_build_free_nids(sbi, true, true);
3203 }
3204
f2fs_destroy_node_manager(struct f2fs_sb_info * sbi)3205 void f2fs_destroy_node_manager(struct f2fs_sb_info *sbi)
3206 {
3207 struct f2fs_nm_info *nm_i = NM_I(sbi);
3208 struct free_nid *i, *next_i;
3209 struct nat_entry *natvec[NATVEC_SIZE];
3210 struct nat_entry_set *setvec[SETVEC_SIZE];
3211 nid_t nid = 0;
3212 unsigned int found;
3213
3214 if (!nm_i)
3215 return;
3216
3217 /* destroy free nid list */
3218 spin_lock(&nm_i->nid_list_lock);
3219 list_for_each_entry_safe(i, next_i, &nm_i->free_nid_list, list) {
3220 __remove_free_nid(sbi, i, FREE_NID);
3221 spin_unlock(&nm_i->nid_list_lock);
3222 kmem_cache_free(free_nid_slab, i);
3223 spin_lock(&nm_i->nid_list_lock);
3224 }
3225 f2fs_bug_on(sbi, nm_i->nid_cnt[FREE_NID]);
3226 f2fs_bug_on(sbi, nm_i->nid_cnt[PREALLOC_NID]);
3227 f2fs_bug_on(sbi, !list_empty(&nm_i->free_nid_list));
3228 spin_unlock(&nm_i->nid_list_lock);
3229
3230 /* destroy nat cache */
3231 down_write(&nm_i->nat_tree_lock);
3232 while ((found = __gang_lookup_nat_cache(nm_i,
3233 nid, NATVEC_SIZE, natvec))) {
3234 unsigned idx;
3235
3236 nid = nat_get_nid(natvec[found - 1]) + 1;
3237 for (idx = 0; idx < found; idx++) {
3238 spin_lock(&nm_i->nat_list_lock);
3239 list_del(&natvec[idx]->list);
3240 spin_unlock(&nm_i->nat_list_lock);
3241
3242 __del_from_nat_cache(nm_i, natvec[idx]);
3243 }
3244 }
3245 f2fs_bug_on(sbi, nm_i->nat_cnt[TOTAL_NAT]);
3246
3247 /* destroy nat set cache */
3248 nid = 0;
3249 while ((found = __gang_lookup_nat_set(nm_i,
3250 nid, SETVEC_SIZE, setvec))) {
3251 unsigned idx;
3252
3253 nid = setvec[found - 1]->set + 1;
3254 for (idx = 0; idx < found; idx++) {
3255 /* entry_cnt is not zero, when cp_error was occurred */
3256 f2fs_bug_on(sbi, !list_empty(&setvec[idx]->entry_list));
3257 radix_tree_delete(&nm_i->nat_set_root, setvec[idx]->set);
3258 kmem_cache_free(nat_entry_set_slab, setvec[idx]);
3259 }
3260 }
3261 up_write(&nm_i->nat_tree_lock);
3262
3263 kvfree(nm_i->nat_block_bitmap);
3264 if (nm_i->free_nid_bitmap) {
3265 int i;
3266
3267 for (i = 0; i < nm_i->nat_blocks; i++)
3268 kvfree(nm_i->free_nid_bitmap[i]);
3269 kvfree(nm_i->free_nid_bitmap);
3270 }
3271 kvfree(nm_i->free_nid_count);
3272
3273 kvfree(nm_i->nat_bitmap);
3274 kvfree(nm_i->nat_bits);
3275 #ifdef CONFIG_F2FS_CHECK_FS
3276 kvfree(nm_i->nat_bitmap_mir);
3277 #endif
3278 sbi->nm_info = NULL;
3279 kvfree(nm_i);
3280 }
3281
f2fs_create_node_manager_caches(void)3282 int __init f2fs_create_node_manager_caches(void)
3283 {
3284 nat_entry_slab = f2fs_kmem_cache_create("f2fs_nat_entry",
3285 sizeof(struct nat_entry));
3286 if (!nat_entry_slab)
3287 goto fail;
3288
3289 free_nid_slab = f2fs_kmem_cache_create("f2fs_free_nid",
3290 sizeof(struct free_nid));
3291 if (!free_nid_slab)
3292 goto destroy_nat_entry;
3293
3294 nat_entry_set_slab = f2fs_kmem_cache_create("f2fs_nat_entry_set",
3295 sizeof(struct nat_entry_set));
3296 if (!nat_entry_set_slab)
3297 goto destroy_free_nid;
3298
3299 fsync_node_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_node_entry",
3300 sizeof(struct fsync_node_entry));
3301 if (!fsync_node_entry_slab)
3302 goto destroy_nat_entry_set;
3303 return 0;
3304
3305 destroy_nat_entry_set:
3306 kmem_cache_destroy(nat_entry_set_slab);
3307 destroy_free_nid:
3308 kmem_cache_destroy(free_nid_slab);
3309 destroy_nat_entry:
3310 kmem_cache_destroy(nat_entry_slab);
3311 fail:
3312 return -ENOMEM;
3313 }
3314
f2fs_destroy_node_manager_caches(void)3315 void f2fs_destroy_node_manager_caches(void)
3316 {
3317 kmem_cache_destroy(fsync_node_entry_slab);
3318 kmem_cache_destroy(nat_entry_set_slab);
3319 kmem_cache_destroy(free_nid_slab);
3320 kmem_cache_destroy(nat_entry_slab);
3321 }
3322