1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * fs/f2fs/recovery.c
4 *
5 * Copyright (c) 2012 Samsung Electronics Co., Ltd.
6 * http://www.samsung.com/
7 */
8 #include <asm/unaligned.h>
9 #include <linux/fs.h>
10 #include <linux/f2fs_fs.h>
11 #include <linux/sched/mm.h>
12 #include "f2fs.h"
13 #include "node.h"
14 #include "segment.h"
15
16 /*
17 * Roll forward recovery scenarios.
18 *
19 * [Term] F: fsync_mark, D: dentry_mark
20 *
21 * 1. inode(x) | CP | inode(x) | dnode(F)
22 * -> Update the latest inode(x).
23 *
24 * 2. inode(x) | CP | inode(F) | dnode(F)
25 * -> No problem.
26 *
27 * 3. inode(x) | CP | dnode(F) | inode(x)
28 * -> Recover to the latest dnode(F), and drop the last inode(x)
29 *
30 * 4. inode(x) | CP | dnode(F) | inode(F)
31 * -> No problem.
32 *
33 * 5. CP | inode(x) | dnode(F)
34 * -> The inode(DF) was missing. Should drop this dnode(F).
35 *
36 * 6. CP | inode(DF) | dnode(F)
37 * -> No problem.
38 *
39 * 7. CP | dnode(F) | inode(DF)
40 * -> If f2fs_iget fails, then goto next to find inode(DF).
41 *
42 * 8. CP | dnode(F) | inode(x)
43 * -> If f2fs_iget fails, then goto next to find inode(DF).
44 * But it will fail due to no inode(DF).
45 */
46
47 static struct kmem_cache *fsync_entry_slab;
48
49 #if IS_ENABLED(CONFIG_UNICODE)
50 extern struct kmem_cache *f2fs_cf_name_slab;
51 #endif
52
f2fs_space_for_roll_forward(struct f2fs_sb_info * sbi)53 bool f2fs_space_for_roll_forward(struct f2fs_sb_info *sbi)
54 {
55 s64 nalloc = percpu_counter_sum_positive(&sbi->alloc_valid_block_count);
56
57 if (sbi->last_valid_block_count + nalloc > sbi->user_block_count)
58 return false;
59 if (NM_I(sbi)->max_rf_node_blocks &&
60 percpu_counter_sum_positive(&sbi->rf_node_block_count) >=
61 NM_I(sbi)->max_rf_node_blocks)
62 return false;
63 return true;
64 }
65
get_fsync_inode(struct list_head * head,nid_t ino)66 static struct fsync_inode_entry *get_fsync_inode(struct list_head *head,
67 nid_t ino)
68 {
69 struct fsync_inode_entry *entry;
70
71 list_for_each_entry(entry, head, list)
72 if (entry->inode->i_ino == ino)
73 return entry;
74
75 return NULL;
76 }
77
add_fsync_inode(struct f2fs_sb_info * sbi,struct list_head * head,nid_t ino,bool quota_inode)78 static struct fsync_inode_entry *add_fsync_inode(struct f2fs_sb_info *sbi,
79 struct list_head *head, nid_t ino, bool quota_inode)
80 {
81 struct inode *inode;
82 struct fsync_inode_entry *entry;
83 int err;
84
85 inode = f2fs_iget_retry(sbi->sb, ino);
86 if (IS_ERR(inode))
87 return ERR_CAST(inode);
88
89 err = f2fs_dquot_initialize(inode);
90 if (err)
91 goto err_out;
92
93 if (quota_inode) {
94 err = dquot_alloc_inode(inode);
95 if (err)
96 goto err_out;
97 }
98
99 entry = f2fs_kmem_cache_alloc(fsync_entry_slab,
100 GFP_F2FS_ZERO, true, NULL);
101 entry->inode = inode;
102 list_add_tail(&entry->list, head);
103
104 return entry;
105 err_out:
106 iput(inode);
107 return ERR_PTR(err);
108 }
109
del_fsync_inode(struct fsync_inode_entry * entry,int drop)110 static void del_fsync_inode(struct fsync_inode_entry *entry, int drop)
111 {
112 if (drop) {
113 /* inode should not be recovered, drop it */
114 f2fs_inode_synced(entry->inode);
115 }
116 iput(entry->inode);
117 list_del(&entry->list);
118 kmem_cache_free(fsync_entry_slab, entry);
119 }
120
init_recovered_filename(const struct inode * dir,struct f2fs_inode * raw_inode,struct f2fs_filename * fname,struct qstr * usr_fname)121 static int init_recovered_filename(const struct inode *dir,
122 struct f2fs_inode *raw_inode,
123 struct f2fs_filename *fname,
124 struct qstr *usr_fname)
125 {
126 int err;
127
128 memset(fname, 0, sizeof(*fname));
129 fname->disk_name.len = le32_to_cpu(raw_inode->i_namelen);
130 fname->disk_name.name = raw_inode->i_name;
131
132 if (WARN_ON(fname->disk_name.len > F2FS_NAME_LEN))
133 return -ENAMETOOLONG;
134
135 if (!IS_ENCRYPTED(dir)) {
136 usr_fname->name = fname->disk_name.name;
137 usr_fname->len = fname->disk_name.len;
138 fname->usr_fname = usr_fname;
139 }
140
141 /* Compute the hash of the filename */
142 if (IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir)) {
143 /*
144 * In this case the hash isn't computable without the key, so it
145 * was saved on-disk.
146 */
147 if (fname->disk_name.len + sizeof(f2fs_hash_t) > F2FS_NAME_LEN)
148 return -EINVAL;
149 fname->hash = get_unaligned((f2fs_hash_t *)
150 &raw_inode->i_name[fname->disk_name.len]);
151 } else if (IS_CASEFOLDED(dir)) {
152 err = f2fs_init_casefolded_name(dir, fname);
153 if (err)
154 return err;
155 f2fs_hash_filename(dir, fname);
156 #if IS_ENABLED(CONFIG_UNICODE)
157 /* Case-sensitive match is fine for recovery */
158 kmem_cache_free(f2fs_cf_name_slab, fname->cf_name.name);
159 fname->cf_name.name = NULL;
160 #endif
161 } else {
162 f2fs_hash_filename(dir, fname);
163 }
164 return 0;
165 }
166
recover_dentry(struct inode * inode,struct page * ipage,struct list_head * dir_list)167 static int recover_dentry(struct inode *inode, struct page *ipage,
168 struct list_head *dir_list)
169 {
170 struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
171 nid_t pino = le32_to_cpu(raw_inode->i_pino);
172 struct f2fs_dir_entry *de;
173 struct f2fs_filename fname;
174 struct qstr usr_fname;
175 struct page *page;
176 struct inode *dir, *einode;
177 struct fsync_inode_entry *entry;
178 int err = 0;
179 char *name;
180
181 entry = get_fsync_inode(dir_list, pino);
182 if (!entry) {
183 entry = add_fsync_inode(F2FS_I_SB(inode), dir_list,
184 pino, false);
185 if (IS_ERR(entry)) {
186 dir = ERR_CAST(entry);
187 err = PTR_ERR(entry);
188 goto out;
189 }
190 }
191
192 dir = entry->inode;
193 err = init_recovered_filename(dir, raw_inode, &fname, &usr_fname);
194 if (err)
195 goto out;
196 retry:
197 de = __f2fs_find_entry(dir, &fname, &page);
198 if (de && inode->i_ino == le32_to_cpu(de->ino))
199 goto out_put;
200
201 if (de) {
202 einode = f2fs_iget_retry(inode->i_sb, le32_to_cpu(de->ino));
203 if (IS_ERR(einode)) {
204 WARN_ON(1);
205 err = PTR_ERR(einode);
206 if (err == -ENOENT)
207 err = -EEXIST;
208 goto out_put;
209 }
210
211 err = f2fs_dquot_initialize(einode);
212 if (err) {
213 iput(einode);
214 goto out_put;
215 }
216
217 err = f2fs_acquire_orphan_inode(F2FS_I_SB(inode));
218 if (err) {
219 iput(einode);
220 goto out_put;
221 }
222 f2fs_delete_entry(de, page, dir, einode);
223 iput(einode);
224 goto retry;
225 } else if (IS_ERR(page)) {
226 err = PTR_ERR(page);
227 } else {
228 err = f2fs_add_dentry(dir, &fname, inode,
229 inode->i_ino, inode->i_mode);
230 }
231 if (err == -ENOMEM)
232 goto retry;
233 goto out;
234
235 out_put:
236 f2fs_put_page(page, 0);
237 out:
238 if (file_enc_name(inode))
239 name = "<encrypted>";
240 else
241 name = raw_inode->i_name;
242 f2fs_notice(F2FS_I_SB(inode), "%s: ino = %x, name = %s, dir = %lx, err = %d",
243 __func__, ino_of_node(ipage), name,
244 IS_ERR(dir) ? 0 : dir->i_ino, err);
245 return err;
246 }
247
recover_quota_data(struct inode * inode,struct page * page)248 static int recover_quota_data(struct inode *inode, struct page *page)
249 {
250 struct f2fs_inode *raw = F2FS_INODE(page);
251 struct iattr attr;
252 uid_t i_uid = le32_to_cpu(raw->i_uid);
253 gid_t i_gid = le32_to_cpu(raw->i_gid);
254 int err;
255
256 memset(&attr, 0, sizeof(attr));
257
258 attr.ia_vfsuid = VFSUIDT_INIT(make_kuid(inode->i_sb->s_user_ns, i_uid));
259 attr.ia_vfsgid = VFSGIDT_INIT(make_kgid(inode->i_sb->s_user_ns, i_gid));
260
261 if (!vfsuid_eq(attr.ia_vfsuid, i_uid_into_vfsuid(&nop_mnt_idmap, inode)))
262 attr.ia_valid |= ATTR_UID;
263 if (!vfsgid_eq(attr.ia_vfsgid, i_gid_into_vfsgid(&nop_mnt_idmap, inode)))
264 attr.ia_valid |= ATTR_GID;
265
266 if (!attr.ia_valid)
267 return 0;
268
269 err = dquot_transfer(&nop_mnt_idmap, inode, &attr);
270 if (err)
271 set_sbi_flag(F2FS_I_SB(inode), SBI_QUOTA_NEED_REPAIR);
272 return err;
273 }
274
recover_inline_flags(struct inode * inode,struct f2fs_inode * ri)275 static void recover_inline_flags(struct inode *inode, struct f2fs_inode *ri)
276 {
277 if (ri->i_inline & F2FS_PIN_FILE)
278 set_inode_flag(inode, FI_PIN_FILE);
279 else
280 clear_inode_flag(inode, FI_PIN_FILE);
281 if (ri->i_inline & F2FS_DATA_EXIST)
282 set_inode_flag(inode, FI_DATA_EXIST);
283 else
284 clear_inode_flag(inode, FI_DATA_EXIST);
285 }
286
recover_inode(struct inode * inode,struct page * page)287 static int recover_inode(struct inode *inode, struct page *page)
288 {
289 struct f2fs_inode *raw = F2FS_INODE(page);
290 char *name;
291 int err;
292
293 inode->i_mode = le16_to_cpu(raw->i_mode);
294
295 err = recover_quota_data(inode, page);
296 if (err)
297 return err;
298
299 i_uid_write(inode, le32_to_cpu(raw->i_uid));
300 i_gid_write(inode, le32_to_cpu(raw->i_gid));
301
302 if (raw->i_inline & F2FS_EXTRA_ATTR) {
303 if (f2fs_sb_has_project_quota(F2FS_I_SB(inode)) &&
304 F2FS_FITS_IN_INODE(raw, le16_to_cpu(raw->i_extra_isize),
305 i_projid)) {
306 projid_t i_projid;
307 kprojid_t kprojid;
308
309 i_projid = (projid_t)le32_to_cpu(raw->i_projid);
310 kprojid = make_kprojid(&init_user_ns, i_projid);
311
312 if (!projid_eq(kprojid, F2FS_I(inode)->i_projid)) {
313 err = f2fs_transfer_project_quota(inode,
314 kprojid);
315 if (err)
316 return err;
317 F2FS_I(inode)->i_projid = kprojid;
318 }
319 }
320 }
321
322 f2fs_i_size_write(inode, le64_to_cpu(raw->i_size));
323 inode_set_atime(inode, le64_to_cpu(raw->i_atime),
324 le32_to_cpu(raw->i_atime_nsec));
325 inode_set_ctime(inode, le64_to_cpu(raw->i_ctime),
326 le32_to_cpu(raw->i_ctime_nsec));
327 inode_set_mtime(inode, le64_to_cpu(raw->i_mtime),
328 le32_to_cpu(raw->i_mtime_nsec));
329
330 F2FS_I(inode)->i_advise = raw->i_advise;
331 F2FS_I(inode)->i_flags = le32_to_cpu(raw->i_flags);
332 f2fs_set_inode_flags(inode);
333 F2FS_I(inode)->i_gc_failures = le16_to_cpu(raw->i_gc_failures);
334
335 recover_inline_flags(inode, raw);
336
337 f2fs_mark_inode_dirty_sync(inode, true);
338
339 if (file_enc_name(inode))
340 name = "<encrypted>";
341 else
342 name = F2FS_INODE(page)->i_name;
343
344 f2fs_notice(F2FS_I_SB(inode), "recover_inode: ino = %x, name = %s, inline = %x",
345 ino_of_node(page), name, raw->i_inline);
346 return 0;
347 }
348
adjust_por_ra_blocks(struct f2fs_sb_info * sbi,unsigned int ra_blocks,unsigned int blkaddr,unsigned int next_blkaddr)349 static unsigned int adjust_por_ra_blocks(struct f2fs_sb_info *sbi,
350 unsigned int ra_blocks, unsigned int blkaddr,
351 unsigned int next_blkaddr)
352 {
353 if (blkaddr + 1 == next_blkaddr)
354 ra_blocks = min_t(unsigned int, RECOVERY_MAX_RA_BLOCKS,
355 ra_blocks * 2);
356 else if (next_blkaddr % BLKS_PER_SEG(sbi))
357 ra_blocks = max_t(unsigned int, RECOVERY_MIN_RA_BLOCKS,
358 ra_blocks / 2);
359 return ra_blocks;
360 }
361
362 /* Detect looped node chain with Floyd's cycle detection algorithm. */
sanity_check_node_chain(struct f2fs_sb_info * sbi,block_t blkaddr,block_t * blkaddr_fast,bool * is_detecting)363 static int sanity_check_node_chain(struct f2fs_sb_info *sbi, block_t blkaddr,
364 block_t *blkaddr_fast, bool *is_detecting)
365 {
366 unsigned int ra_blocks = RECOVERY_MAX_RA_BLOCKS;
367 struct page *page = NULL;
368 int i;
369
370 if (!*is_detecting)
371 return 0;
372
373 for (i = 0; i < 2; i++) {
374 if (!f2fs_is_valid_blkaddr(sbi, *blkaddr_fast, META_POR)) {
375 *is_detecting = false;
376 return 0;
377 }
378
379 page = f2fs_get_tmp_page(sbi, *blkaddr_fast);
380 if (IS_ERR(page))
381 return PTR_ERR(page);
382
383 if (!is_recoverable_dnode(page)) {
384 f2fs_put_page(page, 1);
385 *is_detecting = false;
386 return 0;
387 }
388
389 ra_blocks = adjust_por_ra_blocks(sbi, ra_blocks, *blkaddr_fast,
390 next_blkaddr_of_node(page));
391
392 *blkaddr_fast = next_blkaddr_of_node(page);
393 f2fs_put_page(page, 1);
394
395 f2fs_ra_meta_pages_cond(sbi, *blkaddr_fast, ra_blocks);
396 }
397
398 if (*blkaddr_fast == blkaddr) {
399 f2fs_notice(sbi, "%s: Detect looped node chain on blkaddr:%u."
400 " Run fsck to fix it.", __func__, blkaddr);
401 return -EINVAL;
402 }
403 return 0;
404 }
405
find_fsync_dnodes(struct f2fs_sb_info * sbi,struct list_head * head,bool check_only)406 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head,
407 bool check_only)
408 {
409 struct curseg_info *curseg;
410 struct page *page = NULL;
411 block_t blkaddr, blkaddr_fast;
412 bool is_detecting = true;
413 int err = 0;
414
415 /* get node pages in the current segment */
416 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
417 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
418 blkaddr_fast = blkaddr;
419
420 while (1) {
421 struct fsync_inode_entry *entry;
422
423 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
424 return 0;
425
426 page = f2fs_get_tmp_page(sbi, blkaddr);
427 if (IS_ERR(page)) {
428 err = PTR_ERR(page);
429 break;
430 }
431
432 if (!is_recoverable_dnode(page)) {
433 f2fs_put_page(page, 1);
434 break;
435 }
436
437 if (!is_fsync_dnode(page))
438 goto next;
439
440 entry = get_fsync_inode(head, ino_of_node(page));
441 if (!entry) {
442 bool quota_inode = false;
443
444 if (!check_only &&
445 IS_INODE(page) && is_dent_dnode(page)) {
446 err = f2fs_recover_inode_page(sbi, page);
447 if (err) {
448 f2fs_put_page(page, 1);
449 break;
450 }
451 quota_inode = true;
452 }
453
454 /*
455 * CP | dnode(F) | inode(DF)
456 * For this case, we should not give up now.
457 */
458 entry = add_fsync_inode(sbi, head, ino_of_node(page),
459 quota_inode);
460 if (IS_ERR(entry)) {
461 err = PTR_ERR(entry);
462 if (err == -ENOENT)
463 goto next;
464 f2fs_put_page(page, 1);
465 break;
466 }
467 }
468 entry->blkaddr = blkaddr;
469
470 if (IS_INODE(page) && is_dent_dnode(page))
471 entry->last_dentry = blkaddr;
472 next:
473 /* check next segment */
474 blkaddr = next_blkaddr_of_node(page);
475 f2fs_put_page(page, 1);
476
477 err = sanity_check_node_chain(sbi, blkaddr, &blkaddr_fast,
478 &is_detecting);
479 if (err)
480 break;
481 }
482 return err;
483 }
484
destroy_fsync_dnodes(struct list_head * head,int drop)485 static void destroy_fsync_dnodes(struct list_head *head, int drop)
486 {
487 struct fsync_inode_entry *entry, *tmp;
488
489 list_for_each_entry_safe(entry, tmp, head, list)
490 del_fsync_inode(entry, drop);
491 }
492
check_index_in_prev_nodes(struct f2fs_sb_info * sbi,block_t blkaddr,struct dnode_of_data * dn)493 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
494 block_t blkaddr, struct dnode_of_data *dn)
495 {
496 struct seg_entry *sentry;
497 unsigned int segno = GET_SEGNO(sbi, blkaddr);
498 unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
499 struct f2fs_summary_block *sum_node;
500 struct f2fs_summary sum;
501 struct page *sum_page, *node_page;
502 struct dnode_of_data tdn = *dn;
503 nid_t ino, nid;
504 struct inode *inode;
505 unsigned int offset, ofs_in_node, max_addrs;
506 block_t bidx;
507 int i;
508
509 sentry = get_seg_entry(sbi, segno);
510 if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
511 return 0;
512
513 /* Get the previous summary */
514 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
515 struct curseg_info *curseg = CURSEG_I(sbi, i);
516
517 if (curseg->segno == segno) {
518 sum = curseg->sum_blk->entries[blkoff];
519 goto got_it;
520 }
521 }
522
523 sum_page = f2fs_get_sum_page(sbi, segno);
524 if (IS_ERR(sum_page))
525 return PTR_ERR(sum_page);
526 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
527 sum = sum_node->entries[blkoff];
528 f2fs_put_page(sum_page, 1);
529 got_it:
530 /* Use the locked dnode page and inode */
531 nid = le32_to_cpu(sum.nid);
532 ofs_in_node = le16_to_cpu(sum.ofs_in_node);
533
534 max_addrs = ADDRS_PER_PAGE(dn->node_page, dn->inode);
535 if (ofs_in_node >= max_addrs) {
536 f2fs_err(sbi, "Inconsistent ofs_in_node:%u in summary, ino:%lu, nid:%u, max:%u",
537 ofs_in_node, dn->inode->i_ino, nid, max_addrs);
538 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUMMARY);
539 return -EFSCORRUPTED;
540 }
541
542 if (dn->inode->i_ino == nid) {
543 tdn.nid = nid;
544 if (!dn->inode_page_locked)
545 lock_page(dn->inode_page);
546 tdn.node_page = dn->inode_page;
547 tdn.ofs_in_node = ofs_in_node;
548 goto truncate_out;
549 } else if (dn->nid == nid) {
550 tdn.ofs_in_node = ofs_in_node;
551 goto truncate_out;
552 }
553
554 /* Get the node page */
555 node_page = f2fs_get_node_page(sbi, nid);
556 if (IS_ERR(node_page))
557 return PTR_ERR(node_page);
558
559 offset = ofs_of_node(node_page);
560 ino = ino_of_node(node_page);
561 f2fs_put_page(node_page, 1);
562
563 if (ino != dn->inode->i_ino) {
564 int ret;
565
566 /* Deallocate previous index in the node page */
567 inode = f2fs_iget_retry(sbi->sb, ino);
568 if (IS_ERR(inode))
569 return PTR_ERR(inode);
570
571 ret = f2fs_dquot_initialize(inode);
572 if (ret) {
573 iput(inode);
574 return ret;
575 }
576 } else {
577 inode = dn->inode;
578 }
579
580 bidx = f2fs_start_bidx_of_node(offset, inode) +
581 le16_to_cpu(sum.ofs_in_node);
582
583 /*
584 * if inode page is locked, unlock temporarily, but its reference
585 * count keeps alive.
586 */
587 if (ino == dn->inode->i_ino && dn->inode_page_locked)
588 unlock_page(dn->inode_page);
589
590 set_new_dnode(&tdn, inode, NULL, NULL, 0);
591 if (f2fs_get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
592 goto out;
593
594 if (tdn.data_blkaddr == blkaddr)
595 f2fs_truncate_data_blocks_range(&tdn, 1);
596
597 f2fs_put_dnode(&tdn);
598 out:
599 if (ino != dn->inode->i_ino)
600 iput(inode);
601 else if (dn->inode_page_locked)
602 lock_page(dn->inode_page);
603 return 0;
604
605 truncate_out:
606 if (f2fs_data_blkaddr(&tdn) == blkaddr)
607 f2fs_truncate_data_blocks_range(&tdn, 1);
608 if (dn->inode->i_ino == nid && !dn->inode_page_locked)
609 unlock_page(dn->inode_page);
610 return 0;
611 }
612
f2fs_reserve_new_block_retry(struct dnode_of_data * dn)613 static int f2fs_reserve_new_block_retry(struct dnode_of_data *dn)
614 {
615 int i, err = 0;
616
617 for (i = DEFAULT_FAILURE_RETRY_COUNT; i > 0; i--) {
618 err = f2fs_reserve_new_block(dn);
619 if (!err)
620 break;
621 }
622
623 return err;
624 }
625
do_recover_data(struct f2fs_sb_info * sbi,struct inode * inode,struct page * page)626 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
627 struct page *page)
628 {
629 struct dnode_of_data dn;
630 struct node_info ni;
631 unsigned int start, end;
632 int err = 0, recovered = 0;
633
634 /* step 1: recover xattr */
635 if (IS_INODE(page)) {
636 err = f2fs_recover_inline_xattr(inode, page);
637 if (err)
638 goto out;
639 } else if (f2fs_has_xattr_block(ofs_of_node(page))) {
640 err = f2fs_recover_xattr_data(inode, page);
641 if (!err)
642 recovered++;
643 goto out;
644 }
645
646 /* step 2: recover inline data */
647 err = f2fs_recover_inline_data(inode, page);
648 if (err) {
649 if (err == 1)
650 err = 0;
651 goto out;
652 }
653
654 /* step 3: recover data indices */
655 start = f2fs_start_bidx_of_node(ofs_of_node(page), inode);
656 end = start + ADDRS_PER_PAGE(page, inode);
657
658 set_new_dnode(&dn, inode, NULL, NULL, 0);
659 retry_dn:
660 err = f2fs_get_dnode_of_data(&dn, start, ALLOC_NODE);
661 if (err) {
662 if (err == -ENOMEM) {
663 memalloc_retry_wait(GFP_NOFS);
664 goto retry_dn;
665 }
666 goto out;
667 }
668
669 f2fs_wait_on_page_writeback(dn.node_page, NODE, true, true);
670
671 err = f2fs_get_node_info(sbi, dn.nid, &ni, false);
672 if (err)
673 goto err;
674
675 f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
676
677 if (ofs_of_node(dn.node_page) != ofs_of_node(page)) {
678 f2fs_warn(sbi, "Inconsistent ofs_of_node, ino:%lu, ofs:%u, %u",
679 inode->i_ino, ofs_of_node(dn.node_page),
680 ofs_of_node(page));
681 err = -EFSCORRUPTED;
682 f2fs_handle_error(sbi, ERROR_INCONSISTENT_FOOTER);
683 goto err;
684 }
685
686 for (; start < end; start++, dn.ofs_in_node++) {
687 block_t src, dest;
688
689 src = f2fs_data_blkaddr(&dn);
690 dest = data_blkaddr(dn.inode, page, dn.ofs_in_node);
691
692 if (__is_valid_data_blkaddr(src) &&
693 !f2fs_is_valid_blkaddr(sbi, src, META_POR)) {
694 err = -EFSCORRUPTED;
695 goto err;
696 }
697
698 if (__is_valid_data_blkaddr(dest) &&
699 !f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
700 err = -EFSCORRUPTED;
701 goto err;
702 }
703
704 /* skip recovering if dest is the same as src */
705 if (src == dest)
706 continue;
707
708 /* dest is invalid, just invalidate src block */
709 if (dest == NULL_ADDR) {
710 f2fs_truncate_data_blocks_range(&dn, 1);
711 continue;
712 }
713
714 if (!file_keep_isize(inode) &&
715 (i_size_read(inode) <= ((loff_t)start << PAGE_SHIFT)))
716 f2fs_i_size_write(inode,
717 (loff_t)(start + 1) << PAGE_SHIFT);
718
719 /*
720 * dest is reserved block, invalidate src block
721 * and then reserve one new block in dnode page.
722 */
723 if (dest == NEW_ADDR) {
724 f2fs_truncate_data_blocks_range(&dn, 1);
725
726 err = f2fs_reserve_new_block_retry(&dn);
727 if (err)
728 goto err;
729 continue;
730 }
731
732 /* dest is valid block, try to recover from src to dest */
733 if (f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
734 if (src == NULL_ADDR) {
735 err = f2fs_reserve_new_block_retry(&dn);
736 if (err)
737 goto err;
738 }
739 retry_prev:
740 /* Check the previous node page having this index */
741 err = check_index_in_prev_nodes(sbi, dest, &dn);
742 if (err) {
743 if (err == -ENOMEM) {
744 memalloc_retry_wait(GFP_NOFS);
745 goto retry_prev;
746 }
747 goto err;
748 }
749
750 if (f2fs_is_valid_blkaddr(sbi, dest,
751 DATA_GENERIC_ENHANCE_UPDATE)) {
752 f2fs_err(sbi, "Inconsistent dest blkaddr:%u, ino:%lu, ofs:%u",
753 dest, inode->i_ino, dn.ofs_in_node);
754 err = -EFSCORRUPTED;
755 goto err;
756 }
757
758 /* write dummy data page */
759 f2fs_replace_block(sbi, &dn, src, dest,
760 ni.version, false, false);
761 recovered++;
762 }
763 }
764
765 copy_node_footer(dn.node_page, page);
766 fill_node_footer(dn.node_page, dn.nid, ni.ino,
767 ofs_of_node(page), false);
768 set_page_dirty(dn.node_page);
769 err:
770 f2fs_put_dnode(&dn);
771 out:
772 f2fs_notice(sbi, "recover_data: ino = %lx (i_size: %s) recovered = %d, err = %d",
773 inode->i_ino, file_keep_isize(inode) ? "keep" : "recover",
774 recovered, err);
775 return err;
776 }
777
recover_data(struct f2fs_sb_info * sbi,struct list_head * inode_list,struct list_head * tmp_inode_list,struct list_head * dir_list)778 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list,
779 struct list_head *tmp_inode_list, struct list_head *dir_list)
780 {
781 struct curseg_info *curseg;
782 struct page *page = NULL;
783 int err = 0;
784 block_t blkaddr;
785 unsigned int ra_blocks = RECOVERY_MAX_RA_BLOCKS;
786
787 /* get node pages in the current segment */
788 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
789 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
790
791 while (1) {
792 struct fsync_inode_entry *entry;
793
794 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
795 break;
796
797 page = f2fs_get_tmp_page(sbi, blkaddr);
798 if (IS_ERR(page)) {
799 err = PTR_ERR(page);
800 break;
801 }
802
803 if (!is_recoverable_dnode(page)) {
804 f2fs_put_page(page, 1);
805 break;
806 }
807
808 entry = get_fsync_inode(inode_list, ino_of_node(page));
809 if (!entry)
810 goto next;
811 /*
812 * inode(x) | CP | inode(x) | dnode(F)
813 * In this case, we can lose the latest inode(x).
814 * So, call recover_inode for the inode update.
815 */
816 if (IS_INODE(page)) {
817 err = recover_inode(entry->inode, page);
818 if (err) {
819 f2fs_put_page(page, 1);
820 break;
821 }
822 }
823 if (entry->last_dentry == blkaddr) {
824 err = recover_dentry(entry->inode, page, dir_list);
825 if (err) {
826 f2fs_put_page(page, 1);
827 break;
828 }
829 }
830 err = do_recover_data(sbi, entry->inode, page);
831 if (err) {
832 f2fs_put_page(page, 1);
833 break;
834 }
835
836 if (entry->blkaddr == blkaddr)
837 list_move_tail(&entry->list, tmp_inode_list);
838 next:
839 ra_blocks = adjust_por_ra_blocks(sbi, ra_blocks, blkaddr,
840 next_blkaddr_of_node(page));
841
842 /* check next segment */
843 blkaddr = next_blkaddr_of_node(page);
844 f2fs_put_page(page, 1);
845
846 f2fs_ra_meta_pages_cond(sbi, blkaddr, ra_blocks);
847 }
848 if (!err)
849 err = f2fs_allocate_new_segments(sbi);
850 return err;
851 }
852
f2fs_recover_fsync_data(struct f2fs_sb_info * sbi,bool check_only)853 int f2fs_recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
854 {
855 struct list_head inode_list, tmp_inode_list;
856 struct list_head dir_list;
857 int err;
858 int ret = 0;
859 unsigned long s_flags = sbi->sb->s_flags;
860 bool need_writecp = false;
861
862 if (is_sbi_flag_set(sbi, SBI_IS_WRITABLE))
863 f2fs_info(sbi, "recover fsync data on readonly fs");
864
865 INIT_LIST_HEAD(&inode_list);
866 INIT_LIST_HEAD(&tmp_inode_list);
867 INIT_LIST_HEAD(&dir_list);
868
869 /* prevent checkpoint */
870 f2fs_down_write(&sbi->cp_global_sem);
871
872 /* step #1: find fsynced inode numbers */
873 err = find_fsync_dnodes(sbi, &inode_list, check_only);
874 if (err || list_empty(&inode_list))
875 goto skip;
876
877 if (check_only) {
878 ret = 1;
879 goto skip;
880 }
881
882 need_writecp = true;
883
884 /* step #2: recover data */
885 err = recover_data(sbi, &inode_list, &tmp_inode_list, &dir_list);
886 if (!err)
887 f2fs_bug_on(sbi, !list_empty(&inode_list));
888 else
889 f2fs_bug_on(sbi, sbi->sb->s_flags & SB_ACTIVE);
890 skip:
891 destroy_fsync_dnodes(&inode_list, err);
892 destroy_fsync_dnodes(&tmp_inode_list, err);
893
894 /* truncate meta pages to be used by the recovery */
895 truncate_inode_pages_range(META_MAPPING(sbi),
896 (loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1);
897
898 if (err) {
899 truncate_inode_pages_final(NODE_MAPPING(sbi));
900 truncate_inode_pages_final(META_MAPPING(sbi));
901 }
902
903 /*
904 * If fsync data succeeds or there is no fsync data to recover,
905 * and the f2fs is not read only, check and fix zoned block devices'
906 * write pointer consistency.
907 */
908 if (f2fs_sb_has_blkzoned(sbi) && !f2fs_readonly(sbi->sb)) {
909 int err2 = f2fs_fix_curseg_write_pointer(sbi);
910
911 if (!err2)
912 err2 = f2fs_check_write_pointer(sbi);
913 if (err2)
914 err = err2;
915 ret = err;
916 }
917
918 if (!err)
919 clear_sbi_flag(sbi, SBI_POR_DOING);
920
921 f2fs_up_write(&sbi->cp_global_sem);
922
923 /* let's drop all the directory inodes for clean checkpoint */
924 destroy_fsync_dnodes(&dir_list, err);
925
926 if (need_writecp) {
927 set_sbi_flag(sbi, SBI_IS_RECOVERED);
928
929 if (!err) {
930 struct cp_control cpc = {
931 .reason = CP_RECOVERY,
932 };
933 stat_inc_cp_call_count(sbi, TOTAL_CALL);
934 err = f2fs_write_checkpoint(sbi, &cpc);
935 }
936 }
937
938 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
939
940 return ret ? ret : err;
941 }
942
f2fs_create_recovery_cache(void)943 int __init f2fs_create_recovery_cache(void)
944 {
945 fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
946 sizeof(struct fsync_inode_entry));
947 return fsync_entry_slab ? 0 : -ENOMEM;
948 }
949
f2fs_destroy_recovery_cache(void)950 void f2fs_destroy_recovery_cache(void)
951 {
952 kmem_cache_destroy(fsync_entry_slab);
953 }
954