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 "f2fs.h"
12 #include "node.h"
13 #include "segment.h"
14
15 /*
16 * Roll forward recovery scenarios.
17 *
18 * [Term] F: fsync_mark, D: dentry_mark
19 *
20 * 1. inode(x) | CP | inode(x) | dnode(F)
21 * -> Update the latest inode(x).
22 *
23 * 2. inode(x) | CP | inode(F) | dnode(F)
24 * -> No problem.
25 *
26 * 3. inode(x) | CP | dnode(F) | inode(x)
27 * -> Recover to the latest dnode(F), and drop the last inode(x)
28 *
29 * 4. inode(x) | CP | dnode(F) | inode(F)
30 * -> No problem.
31 *
32 * 5. CP | inode(x) | dnode(F)
33 * -> The inode(DF) was missing. Should drop this dnode(F).
34 *
35 * 6. CP | inode(DF) | dnode(F)
36 * -> No problem.
37 *
38 * 7. CP | dnode(F) | inode(DF)
39 * -> If f2fs_iget fails, then goto next to find inode(DF).
40 *
41 * 8. CP | dnode(F) | inode(x)
42 * -> If f2fs_iget fails, then goto next to find inode(DF).
43 * But it will fail due to no inode(DF).
44 */
45
46 static struct kmem_cache *fsync_entry_slab;
47
48 #ifdef CONFIG_UNICODE
49 extern struct kmem_cache *f2fs_cf_name_slab;
50 #endif
51
f2fs_space_for_roll_forward(struct f2fs_sb_info * sbi)52 bool f2fs_space_for_roll_forward(struct f2fs_sb_info *sbi)
53 {
54 s64 nalloc = percpu_counter_sum_positive(&sbi->alloc_valid_block_count);
55
56 if (sbi->last_valid_block_count + nalloc > sbi->user_block_count)
57 return false;
58 return true;
59 }
60
get_fsync_inode(struct list_head * head,nid_t ino)61 static struct fsync_inode_entry *get_fsync_inode(struct list_head *head,
62 nid_t ino)
63 {
64 struct fsync_inode_entry *entry;
65
66 list_for_each_entry(entry, head, list)
67 if (entry->inode->i_ino == ino)
68 return entry;
69
70 return NULL;
71 }
72
add_fsync_inode(struct f2fs_sb_info * sbi,struct list_head * head,nid_t ino,bool quota_inode)73 static struct fsync_inode_entry *add_fsync_inode(struct f2fs_sb_info *sbi,
74 struct list_head *head, nid_t ino, bool quota_inode)
75 {
76 struct inode *inode;
77 struct fsync_inode_entry *entry;
78 int err;
79
80 inode = f2fs_iget_retry(sbi->sb, ino);
81 if (IS_ERR(inode))
82 return ERR_CAST(inode);
83
84 err = dquot_initialize(inode);
85 if (err)
86 goto err_out;
87
88 if (quota_inode) {
89 err = dquot_alloc_inode(inode);
90 if (err)
91 goto err_out;
92 }
93
94 entry = f2fs_kmem_cache_alloc(fsync_entry_slab, GFP_F2FS_ZERO);
95 entry->inode = inode;
96 list_add_tail(&entry->list, head);
97
98 return entry;
99 err_out:
100 iput(inode);
101 return ERR_PTR(err);
102 }
103
del_fsync_inode(struct fsync_inode_entry * entry,int drop)104 static void del_fsync_inode(struct fsync_inode_entry *entry, int drop)
105 {
106 if (drop) {
107 /* inode should not be recovered, drop it */
108 f2fs_inode_synced(entry->inode);
109 }
110 iput(entry->inode);
111 list_del(&entry->list);
112 kmem_cache_free(fsync_entry_slab, entry);
113 }
114
init_recovered_filename(const struct inode * dir,struct f2fs_inode * raw_inode,struct f2fs_filename * fname,struct qstr * usr_fname)115 static int init_recovered_filename(const struct inode *dir,
116 struct f2fs_inode *raw_inode,
117 struct f2fs_filename *fname,
118 struct qstr *usr_fname)
119 {
120 int err;
121
122 memset(fname, 0, sizeof(*fname));
123 fname->disk_name.len = le32_to_cpu(raw_inode->i_namelen);
124 fname->disk_name.name = raw_inode->i_name;
125
126 if (WARN_ON(fname->disk_name.len > F2FS_NAME_LEN))
127 return -ENAMETOOLONG;
128
129 if (!IS_ENCRYPTED(dir)) {
130 usr_fname->name = fname->disk_name.name;
131 usr_fname->len = fname->disk_name.len;
132 fname->usr_fname = usr_fname;
133 }
134
135 /* Compute the hash of the filename */
136 if (IS_ENCRYPTED(dir) && IS_CASEFOLDED(dir)) {
137 /*
138 * In this case the hash isn't computable without the key, so it
139 * was saved on-disk.
140 */
141 if (fname->disk_name.len + sizeof(f2fs_hash_t) > F2FS_NAME_LEN)
142 return -EINVAL;
143 fname->hash = get_unaligned((f2fs_hash_t *)
144 &raw_inode->i_name[fname->disk_name.len]);
145 } else if (IS_CASEFOLDED(dir)) {
146 err = f2fs_init_casefolded_name(dir, fname);
147 if (err)
148 return err;
149 f2fs_hash_filename(dir, fname);
150 #ifdef CONFIG_UNICODE
151 /* Case-sensitive match is fine for recovery */
152 kmem_cache_free(f2fs_cf_name_slab, fname->cf_name.name);
153 fname->cf_name.name = NULL;
154 #endif
155 } else {
156 f2fs_hash_filename(dir, fname);
157 }
158 return 0;
159 }
160
recover_dentry(struct inode * inode,struct page * ipage,struct list_head * dir_list)161 static int recover_dentry(struct inode *inode, struct page *ipage,
162 struct list_head *dir_list)
163 {
164 struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
165 nid_t pino = le32_to_cpu(raw_inode->i_pino);
166 struct f2fs_dir_entry *de;
167 struct f2fs_filename fname;
168 struct qstr usr_fname;
169 struct page *page;
170 struct inode *dir, *einode;
171 struct fsync_inode_entry *entry;
172 int err = 0;
173 char *name;
174
175 entry = get_fsync_inode(dir_list, pino);
176 if (!entry) {
177 entry = add_fsync_inode(F2FS_I_SB(inode), dir_list,
178 pino, false);
179 if (IS_ERR(entry)) {
180 dir = ERR_CAST(entry);
181 err = PTR_ERR(entry);
182 goto out;
183 }
184 }
185
186 dir = entry->inode;
187 err = init_recovered_filename(dir, raw_inode, &fname, &usr_fname);
188 if (err)
189 goto out;
190 retry:
191 de = __f2fs_find_entry(dir, &fname, &page);
192 if (de && inode->i_ino == le32_to_cpu(de->ino))
193 goto out_put;
194
195 if (de) {
196 einode = f2fs_iget_retry(inode->i_sb, le32_to_cpu(de->ino));
197 if (IS_ERR(einode)) {
198 WARN_ON(1);
199 err = PTR_ERR(einode);
200 if (err == -ENOENT)
201 err = -EEXIST;
202 goto out_put;
203 }
204
205 err = dquot_initialize(einode);
206 if (err) {
207 iput(einode);
208 goto out_put;
209 }
210
211 err = f2fs_acquire_orphan_inode(F2FS_I_SB(inode));
212 if (err) {
213 iput(einode);
214 goto out_put;
215 }
216 f2fs_delete_entry(de, page, dir, einode);
217 iput(einode);
218 goto retry;
219 } else if (IS_ERR(page)) {
220 err = PTR_ERR(page);
221 } else {
222 err = f2fs_add_dentry(dir, &fname, inode,
223 inode->i_ino, inode->i_mode);
224 }
225 if (err == -ENOMEM)
226 goto retry;
227 goto out;
228
229 out_put:
230 f2fs_put_page(page, 0);
231 out:
232 if (file_enc_name(inode))
233 name = "<encrypted>";
234 else
235 name = raw_inode->i_name;
236 f2fs_notice(F2FS_I_SB(inode), "%s: ino = %x, name = %s, dir = %lx, err = %d",
237 __func__, ino_of_node(ipage), name,
238 IS_ERR(dir) ? 0 : dir->i_ino, err);
239 return err;
240 }
241
recover_quota_data(struct inode * inode,struct page * page)242 static int recover_quota_data(struct inode *inode, struct page *page)
243 {
244 struct f2fs_inode *raw = F2FS_INODE(page);
245 struct iattr attr;
246 uid_t i_uid = le32_to_cpu(raw->i_uid);
247 gid_t i_gid = le32_to_cpu(raw->i_gid);
248 int err;
249
250 memset(&attr, 0, sizeof(attr));
251
252 attr.ia_uid = make_kuid(inode->i_sb->s_user_ns, i_uid);
253 attr.ia_gid = make_kgid(inode->i_sb->s_user_ns, i_gid);
254
255 if (!uid_eq(attr.ia_uid, inode->i_uid))
256 attr.ia_valid |= ATTR_UID;
257 if (!gid_eq(attr.ia_gid, inode->i_gid))
258 attr.ia_valid |= ATTR_GID;
259
260 if (!attr.ia_valid)
261 return 0;
262
263 err = dquot_transfer(inode, &attr);
264 if (err)
265 set_sbi_flag(F2FS_I_SB(inode), SBI_QUOTA_NEED_REPAIR);
266 return err;
267 }
268
recover_inline_flags(struct inode * inode,struct f2fs_inode * ri)269 static void recover_inline_flags(struct inode *inode, struct f2fs_inode *ri)
270 {
271 if (ri->i_inline & F2FS_PIN_FILE)
272 set_inode_flag(inode, FI_PIN_FILE);
273 else
274 clear_inode_flag(inode, FI_PIN_FILE);
275 if (ri->i_inline & F2FS_DATA_EXIST)
276 set_inode_flag(inode, FI_DATA_EXIST);
277 else
278 clear_inode_flag(inode, FI_DATA_EXIST);
279 }
280
recover_inode(struct inode * inode,struct page * page)281 static int recover_inode(struct inode *inode, struct page *page)
282 {
283 struct f2fs_inode *raw = F2FS_INODE(page);
284 char *name;
285 int err;
286
287 inode->i_mode = le16_to_cpu(raw->i_mode);
288
289 err = recover_quota_data(inode, page);
290 if (err)
291 return err;
292
293 i_uid_write(inode, le32_to_cpu(raw->i_uid));
294 i_gid_write(inode, le32_to_cpu(raw->i_gid));
295
296 if (raw->i_inline & F2FS_EXTRA_ATTR) {
297 if (f2fs_sb_has_project_quota(F2FS_I_SB(inode)) &&
298 F2FS_FITS_IN_INODE(raw, le16_to_cpu(raw->i_extra_isize),
299 i_projid)) {
300 projid_t i_projid;
301 kprojid_t kprojid;
302
303 i_projid = (projid_t)le32_to_cpu(raw->i_projid);
304 kprojid = make_kprojid(&init_user_ns, i_projid);
305
306 if (!projid_eq(kprojid, F2FS_I(inode)->i_projid)) {
307 err = f2fs_transfer_project_quota(inode,
308 kprojid);
309 if (err)
310 return err;
311 F2FS_I(inode)->i_projid = kprojid;
312 }
313 }
314 }
315
316 f2fs_i_size_write(inode, le64_to_cpu(raw->i_size));
317 inode->i_atime.tv_sec = le64_to_cpu(raw->i_atime);
318 inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime);
319 inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime);
320 inode->i_atime.tv_nsec = le32_to_cpu(raw->i_atime_nsec);
321 inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec);
322 inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
323
324 F2FS_I(inode)->i_advise = raw->i_advise;
325 F2FS_I(inode)->i_flags = le32_to_cpu(raw->i_flags);
326 f2fs_set_inode_flags(inode);
327 F2FS_I(inode)->i_gc_failures[GC_FAILURE_PIN] =
328 le16_to_cpu(raw->i_gc_failures);
329
330 recover_inline_flags(inode, raw);
331
332 f2fs_mark_inode_dirty_sync(inode, true);
333
334 if (file_enc_name(inode))
335 name = "<encrypted>";
336 else
337 name = F2FS_INODE(page)->i_name;
338
339 f2fs_notice(F2FS_I_SB(inode), "recover_inode: ino = %x, name = %s, inline = %x",
340 ino_of_node(page), name, raw->i_inline);
341 return 0;
342 }
343
find_fsync_dnodes(struct f2fs_sb_info * sbi,struct list_head * head,bool check_only)344 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head,
345 bool check_only)
346 {
347 struct curseg_info *curseg;
348 struct page *page = NULL;
349 block_t blkaddr;
350 unsigned int loop_cnt = 0;
351 unsigned int free_blocks = MAIN_SEGS(sbi) * sbi->blocks_per_seg -
352 valid_user_blocks(sbi);
353 int err = 0;
354
355 /* get node pages in the current segment */
356 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
357 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
358
359 while (1) {
360 struct fsync_inode_entry *entry;
361
362 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
363 return 0;
364
365 page = f2fs_get_tmp_page(sbi, blkaddr);
366 if (IS_ERR(page)) {
367 err = PTR_ERR(page);
368 break;
369 }
370
371 if (!is_recoverable_dnode(page)) {
372 f2fs_put_page(page, 1);
373 break;
374 }
375
376 if (!is_fsync_dnode(page))
377 goto next;
378
379 entry = get_fsync_inode(head, ino_of_node(page));
380 if (!entry) {
381 bool quota_inode = false;
382
383 if (!check_only &&
384 IS_INODE(page) && is_dent_dnode(page)) {
385 err = f2fs_recover_inode_page(sbi, page);
386 if (err) {
387 f2fs_put_page(page, 1);
388 break;
389 }
390 quota_inode = true;
391 }
392
393 /*
394 * CP | dnode(F) | inode(DF)
395 * For this case, we should not give up now.
396 */
397 entry = add_fsync_inode(sbi, head, ino_of_node(page),
398 quota_inode);
399 if (IS_ERR(entry)) {
400 err = PTR_ERR(entry);
401 if (err == -ENOENT) {
402 err = 0;
403 goto next;
404 }
405 f2fs_put_page(page, 1);
406 break;
407 }
408 }
409 entry->blkaddr = blkaddr;
410
411 if (IS_INODE(page) && is_dent_dnode(page))
412 entry->last_dentry = blkaddr;
413 next:
414 /* sanity check in order to detect looped node chain */
415 if (++loop_cnt >= free_blocks ||
416 blkaddr == next_blkaddr_of_node(page)) {
417 f2fs_notice(sbi, "%s: detect looped node chain, blkaddr:%u, next:%u",
418 __func__, blkaddr,
419 next_blkaddr_of_node(page));
420 f2fs_put_page(page, 1);
421 err = -EINVAL;
422 break;
423 }
424
425 /* check next segment */
426 blkaddr = next_blkaddr_of_node(page);
427 f2fs_put_page(page, 1);
428
429 f2fs_ra_meta_pages_cond(sbi, blkaddr);
430 }
431 return err;
432 }
433
destroy_fsync_dnodes(struct list_head * head,int drop)434 static void destroy_fsync_dnodes(struct list_head *head, int drop)
435 {
436 struct fsync_inode_entry *entry, *tmp;
437
438 list_for_each_entry_safe(entry, tmp, head, list)
439 del_fsync_inode(entry, drop);
440 }
441
check_index_in_prev_nodes(struct f2fs_sb_info * sbi,block_t blkaddr,struct dnode_of_data * dn)442 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
443 block_t blkaddr, struct dnode_of_data *dn)
444 {
445 struct seg_entry *sentry;
446 unsigned int segno = GET_SEGNO(sbi, blkaddr);
447 unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
448 struct f2fs_summary_block *sum_node;
449 struct f2fs_summary sum;
450 struct page *sum_page, *node_page;
451 struct dnode_of_data tdn = *dn;
452 nid_t ino, nid;
453 struct inode *inode;
454 unsigned int offset, ofs_in_node, max_addrs;
455 block_t bidx;
456 int i;
457
458 sentry = get_seg_entry(sbi, segno);
459 if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
460 return 0;
461
462 /* Get the previous summary */
463 for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
464 struct curseg_info *curseg = CURSEG_I(sbi, i);
465
466 if (curseg->segno == segno) {
467 sum = curseg->sum_blk->entries[blkoff];
468 goto got_it;
469 }
470 }
471
472 sum_page = f2fs_get_sum_page(sbi, segno);
473 if (IS_ERR(sum_page))
474 return PTR_ERR(sum_page);
475 sum_node = (struct f2fs_summary_block *)page_address(sum_page);
476 sum = sum_node->entries[blkoff];
477 f2fs_put_page(sum_page, 1);
478 got_it:
479 /* Use the locked dnode page and inode */
480 nid = le32_to_cpu(sum.nid);
481 ofs_in_node = le16_to_cpu(sum.ofs_in_node);
482
483 max_addrs = ADDRS_PER_PAGE(dn->node_page, dn->inode);
484 if (ofs_in_node >= max_addrs) {
485 f2fs_err(sbi, "Inconsistent ofs_in_node:%u in summary, ino:%lu, nid:%u, max:%u",
486 ofs_in_node, dn->inode->i_ino, nid, max_addrs);
487 return -EFSCORRUPTED;
488 }
489
490 if (dn->inode->i_ino == nid) {
491 tdn.nid = nid;
492 if (!dn->inode_page_locked)
493 lock_page(dn->inode_page);
494 tdn.node_page = dn->inode_page;
495 tdn.ofs_in_node = ofs_in_node;
496 goto truncate_out;
497 } else if (dn->nid == nid) {
498 tdn.ofs_in_node = ofs_in_node;
499 goto truncate_out;
500 }
501
502 /* Get the node page */
503 node_page = f2fs_get_node_page(sbi, nid);
504 if (IS_ERR(node_page))
505 return PTR_ERR(node_page);
506
507 offset = ofs_of_node(node_page);
508 ino = ino_of_node(node_page);
509 f2fs_put_page(node_page, 1);
510
511 if (ino != dn->inode->i_ino) {
512 int ret;
513
514 /* Deallocate previous index in the node page */
515 inode = f2fs_iget_retry(sbi->sb, ino);
516 if (IS_ERR(inode))
517 return PTR_ERR(inode);
518
519 ret = dquot_initialize(inode);
520 if (ret) {
521 iput(inode);
522 return ret;
523 }
524 } else {
525 inode = dn->inode;
526 }
527
528 bidx = f2fs_start_bidx_of_node(offset, inode) +
529 le16_to_cpu(sum.ofs_in_node);
530
531 /*
532 * if inode page is locked, unlock temporarily, but its reference
533 * count keeps alive.
534 */
535 if (ino == dn->inode->i_ino && dn->inode_page_locked)
536 unlock_page(dn->inode_page);
537
538 set_new_dnode(&tdn, inode, NULL, NULL, 0);
539 if (f2fs_get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
540 goto out;
541
542 if (tdn.data_blkaddr == blkaddr)
543 f2fs_truncate_data_blocks_range(&tdn, 1);
544
545 f2fs_put_dnode(&tdn);
546 out:
547 if (ino != dn->inode->i_ino)
548 iput(inode);
549 else if (dn->inode_page_locked)
550 lock_page(dn->inode_page);
551 return 0;
552
553 truncate_out:
554 if (f2fs_data_blkaddr(&tdn) == blkaddr)
555 f2fs_truncate_data_blocks_range(&tdn, 1);
556 if (dn->inode->i_ino == nid && !dn->inode_page_locked)
557 unlock_page(dn->inode_page);
558 return 0;
559 }
560
do_recover_data(struct f2fs_sb_info * sbi,struct inode * inode,struct page * page)561 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
562 struct page *page)
563 {
564 struct dnode_of_data dn;
565 struct node_info ni;
566 unsigned int start, end;
567 int err = 0, recovered = 0;
568
569 /* step 1: recover xattr */
570 if (IS_INODE(page)) {
571 err = f2fs_recover_inline_xattr(inode, page);
572 if (err)
573 goto out;
574 } else if (f2fs_has_xattr_block(ofs_of_node(page))) {
575 err = f2fs_recover_xattr_data(inode, page);
576 if (!err)
577 recovered++;
578 goto out;
579 }
580
581 /* step 2: recover inline data */
582 err = f2fs_recover_inline_data(inode, page);
583 if (err) {
584 if (err == 1)
585 err = 0;
586 goto out;
587 }
588
589 /* step 3: recover data indices */
590 start = f2fs_start_bidx_of_node(ofs_of_node(page), inode);
591 end = start + ADDRS_PER_PAGE(page, inode);
592
593 set_new_dnode(&dn, inode, NULL, NULL, 0);
594 retry_dn:
595 err = f2fs_get_dnode_of_data(&dn, start, ALLOC_NODE);
596 if (err) {
597 if (err == -ENOMEM) {
598 congestion_wait(BLK_RW_ASYNC, DEFAULT_IO_TIMEOUT);
599 goto retry_dn;
600 }
601 goto out;
602 }
603
604 f2fs_wait_on_page_writeback(dn.node_page, NODE, true, true);
605
606 err = f2fs_get_node_info(sbi, dn.nid, &ni, false);
607 if (err)
608 goto err;
609
610 f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
611
612 if (ofs_of_node(dn.node_page) != ofs_of_node(page)) {
613 f2fs_warn(sbi, "Inconsistent ofs_of_node, ino:%lu, ofs:%u, %u",
614 inode->i_ino, ofs_of_node(dn.node_page),
615 ofs_of_node(page));
616 err = -EFSCORRUPTED;
617 goto err;
618 }
619
620 for (; start < end; start++, dn.ofs_in_node++) {
621 block_t src, dest;
622
623 src = f2fs_data_blkaddr(&dn);
624 dest = data_blkaddr(dn.inode, page, dn.ofs_in_node);
625
626 if (__is_valid_data_blkaddr(src) &&
627 !f2fs_is_valid_blkaddr(sbi, src, META_POR)) {
628 err = -EFSCORRUPTED;
629 goto err;
630 }
631
632 if (__is_valid_data_blkaddr(dest) &&
633 !f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
634 err = -EFSCORRUPTED;
635 goto err;
636 }
637
638 /* skip recovering if dest is the same as src */
639 if (src == dest)
640 continue;
641
642 /* dest is invalid, just invalidate src block */
643 if (dest == NULL_ADDR) {
644 f2fs_truncate_data_blocks_range(&dn, 1);
645 continue;
646 }
647
648 if (!file_keep_isize(inode) &&
649 (i_size_read(inode) <= ((loff_t)start << PAGE_SHIFT)))
650 f2fs_i_size_write(inode,
651 (loff_t)(start + 1) << PAGE_SHIFT);
652
653 /*
654 * dest is reserved block, invalidate src block
655 * and then reserve one new block in dnode page.
656 */
657 if (dest == NEW_ADDR) {
658 f2fs_truncate_data_blocks_range(&dn, 1);
659 do {
660 err = f2fs_reserve_new_block(&dn);
661 if (err == -ENOSPC) {
662 f2fs_bug_on(sbi, 1);
663 break;
664 }
665 } while (err &&
666 IS_ENABLED(CONFIG_F2FS_FAULT_INJECTION));
667 if (err)
668 goto err;
669 continue;
670 }
671
672 /* dest is valid block, try to recover from src to dest */
673 if (f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
674
675 if (src == NULL_ADDR) {
676 do {
677 err = f2fs_reserve_new_block(&dn);
678 if (err == -ENOSPC) {
679 f2fs_bug_on(sbi, 1);
680 break;
681 }
682 } while (err &&
683 IS_ENABLED(CONFIG_F2FS_FAULT_INJECTION));
684 if (err)
685 goto err;
686 }
687 retry_prev:
688 /* Check the previous node page having this index */
689 err = check_index_in_prev_nodes(sbi, dest, &dn);
690 if (err) {
691 if (err == -ENOMEM) {
692 congestion_wait(BLK_RW_ASYNC,
693 DEFAULT_IO_TIMEOUT);
694 goto retry_prev;
695 }
696 goto err;
697 }
698
699 if (f2fs_is_valid_blkaddr(sbi, dest,
700 DATA_GENERIC_ENHANCE_UPDATE)) {
701 f2fs_err(sbi, "Inconsistent dest blkaddr:%u, ino:%lu, ofs:%u",
702 dest, inode->i_ino, dn.ofs_in_node);
703 err = -EFSCORRUPTED;
704 goto err;
705 }
706
707 /* write dummy data page */
708 f2fs_replace_block(sbi, &dn, src, dest,
709 ni.version, false, false);
710 recovered++;
711 }
712 }
713
714 copy_node_footer(dn.node_page, page);
715 fill_node_footer(dn.node_page, dn.nid, ni.ino,
716 ofs_of_node(page), false);
717 set_page_dirty(dn.node_page);
718 err:
719 f2fs_put_dnode(&dn);
720 out:
721 f2fs_notice(sbi, "recover_data: ino = %lx (i_size: %s) recovered = %d, err = %d",
722 inode->i_ino, file_keep_isize(inode) ? "keep" : "recover",
723 recovered, err);
724 return err;
725 }
726
recover_data(struct f2fs_sb_info * sbi,struct list_head * inode_list,struct list_head * tmp_inode_list,struct list_head * dir_list)727 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list,
728 struct list_head *tmp_inode_list, struct list_head *dir_list)
729 {
730 struct curseg_info *curseg;
731 struct page *page = NULL;
732 int err = 0;
733 block_t blkaddr;
734
735 /* get node pages in the current segment */
736 curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
737 blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
738
739 while (1) {
740 struct fsync_inode_entry *entry;
741
742 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
743 break;
744
745 f2fs_ra_meta_pages_cond(sbi, blkaddr);
746
747 page = f2fs_get_tmp_page(sbi, blkaddr);
748 if (IS_ERR(page)) {
749 err = PTR_ERR(page);
750 break;
751 }
752
753 if (!is_recoverable_dnode(page)) {
754 f2fs_put_page(page, 1);
755 break;
756 }
757
758 entry = get_fsync_inode(inode_list, ino_of_node(page));
759 if (!entry)
760 goto next;
761 /*
762 * inode(x) | CP | inode(x) | dnode(F)
763 * In this case, we can lose the latest inode(x).
764 * So, call recover_inode for the inode update.
765 */
766 if (IS_INODE(page)) {
767 err = recover_inode(entry->inode, page);
768 if (err) {
769 f2fs_put_page(page, 1);
770 break;
771 }
772 }
773 if (entry->last_dentry == blkaddr) {
774 err = recover_dentry(entry->inode, page, dir_list);
775 if (err) {
776 f2fs_put_page(page, 1);
777 break;
778 }
779 }
780 err = do_recover_data(sbi, entry->inode, page);
781 if (err) {
782 f2fs_put_page(page, 1);
783 break;
784 }
785
786 if (entry->blkaddr == blkaddr)
787 list_move_tail(&entry->list, tmp_inode_list);
788 next:
789 /* check next segment */
790 blkaddr = next_blkaddr_of_node(page);
791 f2fs_put_page(page, 1);
792 }
793 if (!err)
794 f2fs_allocate_new_segments(sbi);
795 return err;
796 }
797
f2fs_recover_fsync_data(struct f2fs_sb_info * sbi,bool check_only)798 int f2fs_recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
799 {
800 struct list_head inode_list, tmp_inode_list;
801 struct list_head dir_list;
802 int err;
803 int ret = 0;
804 unsigned long s_flags = sbi->sb->s_flags;
805 bool need_writecp = false;
806 bool fix_curseg_write_pointer = false;
807 #ifdef CONFIG_QUOTA
808 int quota_enabled;
809 #endif
810
811 if (s_flags & SB_RDONLY) {
812 f2fs_info(sbi, "recover fsync data on readonly fs");
813 sbi->sb->s_flags &= ~SB_RDONLY;
814 }
815
816 #ifdef CONFIG_QUOTA
817 /* Needed for iput() to work correctly and not trash data */
818 sbi->sb->s_flags |= SB_ACTIVE;
819 /* Turn on quotas so that they are updated correctly */
820 quota_enabled = f2fs_enable_quota_files(sbi, s_flags & SB_RDONLY);
821 #endif
822
823 INIT_LIST_HEAD(&inode_list);
824 INIT_LIST_HEAD(&tmp_inode_list);
825 INIT_LIST_HEAD(&dir_list);
826
827 /* prevent checkpoint */
828 f2fs_down_write(&sbi->cp_global_sem);
829
830 /* step #1: find fsynced inode numbers */
831 err = find_fsync_dnodes(sbi, &inode_list, check_only);
832 if (err || list_empty(&inode_list))
833 goto skip;
834
835 if (check_only) {
836 ret = 1;
837 goto skip;
838 }
839
840 need_writecp = true;
841
842 /* step #2: recover data */
843 err = recover_data(sbi, &inode_list, &tmp_inode_list, &dir_list);
844 if (!err)
845 f2fs_bug_on(sbi, !list_empty(&inode_list));
846 else {
847 /* restore s_flags to let iput() trash data */
848 sbi->sb->s_flags = s_flags;
849 }
850 skip:
851 fix_curseg_write_pointer = !check_only || list_empty(&inode_list);
852
853 destroy_fsync_dnodes(&inode_list, err);
854 destroy_fsync_dnodes(&tmp_inode_list, err);
855
856 /* truncate meta pages to be used by the recovery */
857 truncate_inode_pages_range(META_MAPPING(sbi),
858 (loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1);
859
860 if (err) {
861 truncate_inode_pages_final(NODE_MAPPING(sbi));
862 truncate_inode_pages_final(META_MAPPING(sbi));
863 }
864
865 /*
866 * If fsync data succeeds or there is no fsync data to recover,
867 * and the f2fs is not read only, check and fix zoned block devices'
868 * write pointer consistency.
869 */
870 if (!err && fix_curseg_write_pointer && !f2fs_readonly(sbi->sb) &&
871 f2fs_sb_has_blkzoned(sbi)) {
872 err = f2fs_fix_curseg_write_pointer(sbi);
873 if (!err)
874 err = f2fs_check_write_pointer(sbi);
875 ret = err;
876 }
877
878 if (!err)
879 clear_sbi_flag(sbi, SBI_POR_DOING);
880
881 f2fs_up_write(&sbi->cp_global_sem);
882
883 /* let's drop all the directory inodes for clean checkpoint */
884 destroy_fsync_dnodes(&dir_list, err);
885
886 if (need_writecp) {
887 set_sbi_flag(sbi, SBI_IS_RECOVERED);
888
889 if (!err) {
890 struct cp_control cpc = {
891 .reason = CP_RECOVERY,
892 };
893 err = f2fs_write_checkpoint(sbi, &cpc);
894 }
895 }
896
897 #ifdef CONFIG_QUOTA
898 /* Turn quotas off */
899 if (quota_enabled)
900 f2fs_quota_off_umount(sbi->sb);
901 #endif
902 sbi->sb->s_flags = s_flags; /* Restore SB_RDONLY status */
903
904 return ret ? ret : err;
905 }
906
f2fs_create_recovery_cache(void)907 int __init f2fs_create_recovery_cache(void)
908 {
909 fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
910 sizeof(struct fsync_inode_entry));
911 if (!fsync_entry_slab)
912 return -ENOMEM;
913 return 0;
914 }
915
f2fs_destroy_recovery_cache(void)916 void f2fs_destroy_recovery_cache(void)
917 {
918 kmem_cache_destroy(fsync_entry_slab);
919 }
920