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->i_atime.tv_sec = le64_to_cpu(raw->i_atime);
324 inode_set_ctime(inode, le64_to_cpu(raw->i_ctime),
325 le32_to_cpu(raw->i_ctime_nsec));
326 inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime);
327 inode->i_atime.tv_nsec = le32_to_cpu(raw->i_atime_nsec);
328 inode->i_mtime.tv_nsec = 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[GC_FAILURE_PIN] =
334 le16_to_cpu(raw->i_gc_failures);
335
336 recover_inline_flags(inode, raw);
337
338 f2fs_mark_inode_dirty_sync(inode, true);
339
340 if (file_enc_name(inode))
341 name = "<encrypted>";
342 else
343 name = F2FS_INODE(page)->i_name;
344
345 f2fs_notice(F2FS_I_SB(inode), "recover_inode: ino = %x, name = %s, inline = %x",
346 ino_of_node(page), name, raw->i_inline);
347 return 0;
348 }
349
adjust_por_ra_blocks(struct f2fs_sb_info * sbi,unsigned int ra_blocks,unsigned int blkaddr,unsigned int next_blkaddr)350 static unsigned int adjust_por_ra_blocks(struct f2fs_sb_info *sbi,
351 unsigned int ra_blocks, unsigned int blkaddr,
352 unsigned int next_blkaddr)
353 {
354 if (blkaddr + 1 == next_blkaddr)
355 ra_blocks = min_t(unsigned int, RECOVERY_MAX_RA_BLOCKS,
356 ra_blocks * 2);
357 else if (next_blkaddr % sbi->blocks_per_seg)
358 ra_blocks = max_t(unsigned int, RECOVERY_MIN_RA_BLOCKS,
359 ra_blocks / 2);
360 return ra_blocks;
361 }
362
363 /* 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)364 static int sanity_check_node_chain(struct f2fs_sb_info *sbi, block_t blkaddr,
365 block_t *blkaddr_fast, bool *is_detecting)
366 {
367 unsigned int ra_blocks = RECOVERY_MAX_RA_BLOCKS;
368 struct page *page = NULL;
369 int i;
370
371 if (!*is_detecting)
372 return 0;
373
374 for (i = 0; i < 2; i++) {
375 if (!f2fs_is_valid_blkaddr(sbi, *blkaddr_fast, META_POR)) {
376 *is_detecting = false;
377 return 0;
378 }
379
380 page = f2fs_get_tmp_page(sbi, *blkaddr_fast);
381 if (IS_ERR(page))
382 return PTR_ERR(page);
383
384 if (!is_recoverable_dnode(page)) {
385 f2fs_put_page(page, 1);
386 *is_detecting = false;
387 return 0;
388 }
389
390 ra_blocks = adjust_por_ra_blocks(sbi, ra_blocks, *blkaddr_fast,
391 next_blkaddr_of_node(page));
392
393 *blkaddr_fast = next_blkaddr_of_node(page);
394 f2fs_put_page(page, 1);
395
396 f2fs_ra_meta_pages_cond(sbi, *blkaddr_fast, ra_blocks);
397 }
398
399 if (*blkaddr_fast == blkaddr) {
400 f2fs_notice(sbi, "%s: Detect looped node chain on blkaddr:%u."
401 " Run fsck to fix it.", __func__, blkaddr);
402 return -EINVAL;
403 }
404 return 0;
405 }
406
find_fsync_dnodes(struct f2fs_sb_info * sbi,struct list_head * head,bool check_only)407 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head,
408 bool check_only)
409 {
410 struct curseg_info *curseg;
411 struct page *page = NULL;
412 block_t blkaddr, blkaddr_fast;
413 bool is_detecting = true;
414 int err = 0;
415
416 /* get node pages in the current segment */
417 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
418 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
419 blkaddr_fast = blkaddr;
420
421 while (1) {
422 struct fsync_inode_entry *entry;
423
424 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
425 return 0;
426
427 page = f2fs_get_tmp_page(sbi, blkaddr);
428 if (IS_ERR(page)) {
429 err = PTR_ERR(page);
430 break;
431 }
432
433 if (!is_recoverable_dnode(page)) {
434 f2fs_put_page(page, 1);
435 break;
436 }
437
438 if (!is_fsync_dnode(page))
439 goto next;
440
441 entry = get_fsync_inode(head, ino_of_node(page));
442 if (!entry) {
443 bool quota_inode = false;
444
445 if (!check_only &&
446 IS_INODE(page) && is_dent_dnode(page)) {
447 err = f2fs_recover_inode_page(sbi, page);
448 if (err) {
449 f2fs_put_page(page, 1);
450 break;
451 }
452 quota_inode = true;
453 }
454
455 /*
456 * CP | dnode(F) | inode(DF)
457 * For this case, we should not give up now.
458 */
459 entry = add_fsync_inode(sbi, head, ino_of_node(page),
460 quota_inode);
461 if (IS_ERR(entry)) {
462 err = PTR_ERR(entry);
463 if (err == -ENOENT)
464 goto next;
465 f2fs_put_page(page, 1);
466 break;
467 }
468 }
469 entry->blkaddr = blkaddr;
470
471 if (IS_INODE(page) && is_dent_dnode(page))
472 entry->last_dentry = blkaddr;
473 next:
474 /* check next segment */
475 blkaddr = next_blkaddr_of_node(page);
476 f2fs_put_page(page, 1);
477
478 err = sanity_check_node_chain(sbi, blkaddr, &blkaddr_fast,
479 &is_detecting);
480 if (err)
481 break;
482 }
483 return err;
484 }
485
destroy_fsync_dnodes(struct list_head * head,int drop)486 static void destroy_fsync_dnodes(struct list_head *head, int drop)
487 {
488 struct fsync_inode_entry *entry, *tmp;
489
490 list_for_each_entry_safe(entry, tmp, head, list)
491 del_fsync_inode(entry, drop);
492 }
493
check_index_in_prev_nodes(struct f2fs_sb_info * sbi,block_t blkaddr,struct dnode_of_data * dn)494 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
495 block_t blkaddr, struct dnode_of_data *dn)
496 {
497 struct seg_entry *sentry;
498 unsigned int segno = GET_SEGNO(sbi, blkaddr);
499 unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
500 struct f2fs_summary_block *sum_node;
501 struct f2fs_summary sum;
502 struct page *sum_page, *node_page;
503 struct dnode_of_data tdn = *dn;
504 nid_t ino, nid;
505 struct inode *inode;
506 unsigned int offset, ofs_in_node, max_addrs;
507 block_t bidx;
508 int i;
509
510 sentry = get_seg_entry(sbi, segno);
511 if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
512 return 0;
513
514 /* Get the previous summary */
515 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
516 struct curseg_info *curseg = CURSEG_I(sbi, i);
517
518 if (curseg->segno == segno) {
519 sum = curseg->sum_blk->entries[blkoff];
520 goto got_it;
521 }
522 }
523
524 sum_page = f2fs_get_sum_page(sbi, segno);
525 if (IS_ERR(sum_page))
526 return PTR_ERR(sum_page);
527 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
528 sum = sum_node->entries[blkoff];
529 f2fs_put_page(sum_page, 1);
530 got_it:
531 /* Use the locked dnode page and inode */
532 nid = le32_to_cpu(sum.nid);
533 ofs_in_node = le16_to_cpu(sum.ofs_in_node);
534
535 max_addrs = ADDRS_PER_PAGE(dn->node_page, dn->inode);
536 if (ofs_in_node >= max_addrs) {
537 f2fs_err(sbi, "Inconsistent ofs_in_node:%u in summary, ino:%lu, nid:%u, max:%u",
538 ofs_in_node, dn->inode->i_ino, nid, max_addrs);
539 f2fs_handle_error(sbi, ERROR_INCONSISTENT_SUMMARY);
540 return -EFSCORRUPTED;
541 }
542
543 if (dn->inode->i_ino == nid) {
544 tdn.nid = nid;
545 if (!dn->inode_page_locked)
546 lock_page(dn->inode_page);
547 tdn.node_page = dn->inode_page;
548 tdn.ofs_in_node = ofs_in_node;
549 goto truncate_out;
550 } else if (dn->nid == nid) {
551 tdn.ofs_in_node = ofs_in_node;
552 goto truncate_out;
553 }
554
555 /* Get the node page */
556 node_page = f2fs_get_node_page(sbi, nid);
557 if (IS_ERR(node_page))
558 return PTR_ERR(node_page);
559
560 offset = ofs_of_node(node_page);
561 ino = ino_of_node(node_page);
562 f2fs_put_page(node_page, 1);
563
564 if (ino != dn->inode->i_ino) {
565 int ret;
566
567 /* Deallocate previous index in the node page */
568 inode = f2fs_iget_retry(sbi->sb, ino);
569 if (IS_ERR(inode))
570 return PTR_ERR(inode);
571
572 ret = f2fs_dquot_initialize(inode);
573 if (ret) {
574 iput(inode);
575 return ret;
576 }
577 } else {
578 inode = dn->inode;
579 }
580
581 bidx = f2fs_start_bidx_of_node(offset, inode) +
582 le16_to_cpu(sum.ofs_in_node);
583
584 /*
585 * if inode page is locked, unlock temporarily, but its reference
586 * count keeps alive.
587 */
588 if (ino == dn->inode->i_ino && dn->inode_page_locked)
589 unlock_page(dn->inode_page);
590
591 set_new_dnode(&tdn, inode, NULL, NULL, 0);
592 if (f2fs_get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
593 goto out;
594
595 if (tdn.data_blkaddr == blkaddr)
596 f2fs_truncate_data_blocks_range(&tdn, 1);
597
598 f2fs_put_dnode(&tdn);
599 out:
600 if (ino != dn->inode->i_ino)
601 iput(inode);
602 else if (dn->inode_page_locked)
603 lock_page(dn->inode_page);
604 return 0;
605
606 truncate_out:
607 if (f2fs_data_blkaddr(&tdn) == blkaddr)
608 f2fs_truncate_data_blocks_range(&tdn, 1);
609 if (dn->inode->i_ino == nid && !dn->inode_page_locked)
610 unlock_page(dn->inode_page);
611 return 0;
612 }
613
do_recover_data(struct f2fs_sb_info * sbi,struct inode * inode,struct page * page)614 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
615 struct page *page)
616 {
617 struct dnode_of_data dn;
618 struct node_info ni;
619 unsigned int start, end;
620 int err = 0, recovered = 0;
621
622 /* step 1: recover xattr */
623 if (IS_INODE(page)) {
624 err = f2fs_recover_inline_xattr(inode, page);
625 if (err)
626 goto out;
627 } else if (f2fs_has_xattr_block(ofs_of_node(page))) {
628 err = f2fs_recover_xattr_data(inode, page);
629 if (!err)
630 recovered++;
631 goto out;
632 }
633
634 /* step 2: recover inline data */
635 err = f2fs_recover_inline_data(inode, page);
636 if (err) {
637 if (err == 1)
638 err = 0;
639 goto out;
640 }
641
642 /* step 3: recover data indices */
643 start = f2fs_start_bidx_of_node(ofs_of_node(page), inode);
644 end = start + ADDRS_PER_PAGE(page, inode);
645
646 set_new_dnode(&dn, inode, NULL, NULL, 0);
647 retry_dn:
648 err = f2fs_get_dnode_of_data(&dn, start, ALLOC_NODE);
649 if (err) {
650 if (err == -ENOMEM) {
651 memalloc_retry_wait(GFP_NOFS);
652 goto retry_dn;
653 }
654 goto out;
655 }
656
657 f2fs_wait_on_page_writeback(dn.node_page, NODE, true, true);
658
659 err = f2fs_get_node_info(sbi, dn.nid, &ni, false);
660 if (err)
661 goto err;
662
663 f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
664
665 if (ofs_of_node(dn.node_page) != ofs_of_node(page)) {
666 f2fs_warn(sbi, "Inconsistent ofs_of_node, ino:%lu, ofs:%u, %u",
667 inode->i_ino, ofs_of_node(dn.node_page),
668 ofs_of_node(page));
669 err = -EFSCORRUPTED;
670 f2fs_handle_error(sbi, ERROR_INCONSISTENT_FOOTER);
671 goto err;
672 }
673
674 for (; start < end; start++, dn.ofs_in_node++) {
675 block_t src, dest;
676
677 src = f2fs_data_blkaddr(&dn);
678 dest = data_blkaddr(dn.inode, page, dn.ofs_in_node);
679
680 if (__is_valid_data_blkaddr(src) &&
681 !f2fs_is_valid_blkaddr(sbi, src, META_POR)) {
682 err = -EFSCORRUPTED;
683 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
684 goto err;
685 }
686
687 if (__is_valid_data_blkaddr(dest) &&
688 !f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
689 err = -EFSCORRUPTED;
690 f2fs_handle_error(sbi, ERROR_INVALID_BLKADDR);
691 goto err;
692 }
693
694 /* skip recovering if dest is the same as src */
695 if (src == dest)
696 continue;
697
698 /* dest is invalid, just invalidate src block */
699 if (dest == NULL_ADDR) {
700 f2fs_truncate_data_blocks_range(&dn, 1);
701 continue;
702 }
703
704 if (!file_keep_isize(inode) &&
705 (i_size_read(inode) <= ((loff_t)start << PAGE_SHIFT)))
706 f2fs_i_size_write(inode,
707 (loff_t)(start + 1) << PAGE_SHIFT);
708
709 /*
710 * dest is reserved block, invalidate src block
711 * and then reserve one new block in dnode page.
712 */
713 if (dest == NEW_ADDR) {
714 f2fs_truncate_data_blocks_range(&dn, 1);
715 do {
716 err = f2fs_reserve_new_block(&dn);
717 if (err == -ENOSPC) {
718 f2fs_bug_on(sbi, 1);
719 break;
720 }
721 } while (err &&
722 IS_ENABLED(CONFIG_F2FS_FAULT_INJECTION));
723 if (err)
724 goto err;
725 continue;
726 }
727
728 /* dest is valid block, try to recover from src to dest */
729 if (f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
730
731 if (src == NULL_ADDR) {
732 do {
733 err = f2fs_reserve_new_block(&dn);
734 if (err == -ENOSPC) {
735 f2fs_bug_on(sbi, 1);
736 break;
737 }
738 } while (err &&
739 IS_ENABLED(CONFIG_F2FS_FAULT_INJECTION));
740 if (err)
741 goto err;
742 }
743 retry_prev:
744 /* Check the previous node page having this index */
745 err = check_index_in_prev_nodes(sbi, dest, &dn);
746 if (err) {
747 if (err == -ENOMEM) {
748 memalloc_retry_wait(GFP_NOFS);
749 goto retry_prev;
750 }
751 goto err;
752 }
753
754 if (f2fs_is_valid_blkaddr(sbi, dest,
755 DATA_GENERIC_ENHANCE_UPDATE)) {
756 f2fs_err(sbi, "Inconsistent dest blkaddr:%u, ino:%lu, ofs:%u",
757 dest, inode->i_ino, dn.ofs_in_node);
758 err = -EFSCORRUPTED;
759 f2fs_handle_error(sbi,
760 ERROR_INVALID_BLKADDR);
761 goto err;
762 }
763
764 /* write dummy data page */
765 f2fs_replace_block(sbi, &dn, src, dest,
766 ni.version, false, false);
767 recovered++;
768 }
769 }
770
771 copy_node_footer(dn.node_page, page);
772 fill_node_footer(dn.node_page, dn.nid, ni.ino,
773 ofs_of_node(page), false);
774 set_page_dirty(dn.node_page);
775 err:
776 f2fs_put_dnode(&dn);
777 out:
778 f2fs_notice(sbi, "recover_data: ino = %lx (i_size: %s) recovered = %d, err = %d",
779 inode->i_ino, file_keep_isize(inode) ? "keep" : "recover",
780 recovered, err);
781 return err;
782 }
783
recover_data(struct f2fs_sb_info * sbi,struct list_head * inode_list,struct list_head * tmp_inode_list,struct list_head * dir_list)784 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list,
785 struct list_head *tmp_inode_list, struct list_head *dir_list)
786 {
787 struct curseg_info *curseg;
788 struct page *page = NULL;
789 int err = 0;
790 block_t blkaddr;
791 unsigned int ra_blocks = RECOVERY_MAX_RA_BLOCKS;
792
793 /* get node pages in the current segment */
794 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
795 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
796
797 while (1) {
798 struct fsync_inode_entry *entry;
799
800 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
801 break;
802
803 page = f2fs_get_tmp_page(sbi, blkaddr);
804 if (IS_ERR(page)) {
805 err = PTR_ERR(page);
806 break;
807 }
808
809 if (!is_recoverable_dnode(page)) {
810 f2fs_put_page(page, 1);
811 break;
812 }
813
814 entry = get_fsync_inode(inode_list, ino_of_node(page));
815 if (!entry)
816 goto next;
817 /*
818 * inode(x) | CP | inode(x) | dnode(F)
819 * In this case, we can lose the latest inode(x).
820 * So, call recover_inode for the inode update.
821 */
822 if (IS_INODE(page)) {
823 err = recover_inode(entry->inode, page);
824 if (err) {
825 f2fs_put_page(page, 1);
826 break;
827 }
828 }
829 if (entry->last_dentry == blkaddr) {
830 err = recover_dentry(entry->inode, page, dir_list);
831 if (err) {
832 f2fs_put_page(page, 1);
833 break;
834 }
835 }
836 err = do_recover_data(sbi, entry->inode, page);
837 if (err) {
838 f2fs_put_page(page, 1);
839 break;
840 }
841
842 if (entry->blkaddr == blkaddr)
843 list_move_tail(&entry->list, tmp_inode_list);
844 next:
845 ra_blocks = adjust_por_ra_blocks(sbi, ra_blocks, blkaddr,
846 next_blkaddr_of_node(page));
847
848 /* check next segment */
849 blkaddr = next_blkaddr_of_node(page);
850 f2fs_put_page(page, 1);
851
852 f2fs_ra_meta_pages_cond(sbi, blkaddr, ra_blocks);
853 }
854 if (!err)
855 f2fs_allocate_new_segments(sbi);
856 return err;
857 }
858
f2fs_recover_fsync_data(struct f2fs_sb_info * sbi,bool check_only)859 int f2fs_recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
860 {
861 struct list_head inode_list, tmp_inode_list;
862 struct list_head dir_list;
863 int err;
864 int ret = 0;
865 unsigned long s_flags = sbi->sb->s_flags;
866 bool need_writecp = false;
867 bool fix_curseg_write_pointer = false;
868
869 if (is_sbi_flag_set(sbi, SBI_IS_WRITABLE))
870 f2fs_info(sbi, "recover fsync data on readonly fs");
871
872 INIT_LIST_HEAD(&inode_list);
873 INIT_LIST_HEAD(&tmp_inode_list);
874 INIT_LIST_HEAD(&dir_list);
875
876 /* prevent checkpoint */
877 f2fs_down_write(&sbi->cp_global_sem);
878
879 /* step #1: find fsynced inode numbers */
880 err = find_fsync_dnodes(sbi, &inode_list, check_only);
881 if (err || list_empty(&inode_list))
882 goto skip;
883
884 if (check_only) {
885 ret = 1;
886 goto skip;
887 }
888
889 need_writecp = true;
890
891 /* step #2: recover data */
892 err = recover_data(sbi, &inode_list, &tmp_inode_list, &dir_list);
893 if (!err)
894 f2fs_bug_on(sbi, !list_empty(&inode_list));
895 else
896 f2fs_bug_on(sbi, sbi->sb->s_flags & SB_ACTIVE);
897 skip:
898 fix_curseg_write_pointer = !check_only || list_empty(&inode_list);
899
900 destroy_fsync_dnodes(&inode_list, err);
901 destroy_fsync_dnodes(&tmp_inode_list, err);
902
903 /* truncate meta pages to be used by the recovery */
904 truncate_inode_pages_range(META_MAPPING(sbi),
905 (loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1);
906
907 if (err) {
908 truncate_inode_pages_final(NODE_MAPPING(sbi));
909 truncate_inode_pages_final(META_MAPPING(sbi));
910 }
911
912 /*
913 * If fsync data succeeds or there is no fsync data to recover,
914 * and the f2fs is not read only, check and fix zoned block devices'
915 * write pointer consistency.
916 */
917 if (!err && fix_curseg_write_pointer && !f2fs_readonly(sbi->sb) &&
918 f2fs_sb_has_blkzoned(sbi)) {
919 err = f2fs_fix_curseg_write_pointer(sbi);
920 if (!err)
921 err = f2fs_check_write_pointer(sbi);
922 ret = err;
923 }
924
925 if (!err)
926 clear_sbi_flag(sbi, SBI_POR_DOING);
927
928 f2fs_up_write(&sbi->cp_global_sem);
929
930 /* let's drop all the directory inodes for clean checkpoint */
931 destroy_fsync_dnodes(&dir_list, err);
932
933 if (need_writecp) {
934 set_sbi_flag(sbi, SBI_IS_RECOVERED);
935
936 if (!err) {
937 struct cp_control cpc = {
938 .reason = CP_RECOVERY,
939 };
940 stat_inc_cp_call_count(sbi, TOTAL_CALL);
941 err = f2fs_write_checkpoint(sbi, &cpc);
942 }
943 }
944
945 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
946
947 return ret ? ret : err;
948 }
949
f2fs_create_recovery_cache(void)950 int __init f2fs_create_recovery_cache(void)
951 {
952 fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
953 sizeof(struct fsync_inode_entry));
954 return fsync_entry_slab ? 0 : -ENOMEM;
955 }
956
f2fs_destroy_recovery_cache(void)957 void f2fs_destroy_recovery_cache(void)
958 {
959 kmem_cache_destroy(fsync_entry_slab);
960 }
961