1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * This file is part of UBIFS.
4 *
5 * Copyright (C) 2006-2008 Nokia Corporation.
6 *
7 * Authors: Adrian Hunter
8 * Artem Bityutskiy (Битюцкий Артём)
9 */
10
11/*
12 * This file contains journal replay code. It runs when the file-system is being
13 * mounted and requires no locking.
14 *
15 * The larger is the journal, the longer it takes to scan it, so the longer it
16 * takes to mount UBIFS. This is why the journal has limited size which may be
17 * changed depending on the system requirements. But a larger journal gives
18 * faster I/O speed because it writes the index less frequently. So this is a
19 * trade-off. Also, the journal is indexed by the in-memory index (TNC), so the
20 * larger is the journal, the more memory its index may consume.
21 */
22
23#include "ubifs.h"
24#include <linux/list_sort.h>
25#include <crypto/hash.h>
26
27/**
28 * struct replay_entry - replay list entry.
29 * @lnum: logical eraseblock number of the node
30 * @offs: node offset
31 * @len: node length
32 * @deletion: non-zero if this entry corresponds to a node deletion
33 * @sqnum: node sequence number
34 * @list: links the replay list
35 * @key: node key
36 * @nm: directory entry name
37 * @old_size: truncation old size
38 * @new_size: truncation new size
39 *
40 * The replay process first scans all buds and builds the replay list, then
41 * sorts the replay list in nodes sequence number order, and then inserts all
42 * the replay entries to the TNC.
43 */
44struct replay_entry {
45 int lnum;
46 int offs;
47 int len;
48 u8 hash[UBIFS_HASH_ARR_SZ];
49 unsigned int deletion:1;
50 unsigned long long sqnum;
51 struct list_head list;
52 union ubifs_key key;
53 union {
54 struct fscrypt_name nm;
55 struct {
56 loff_t old_size;
57 loff_t new_size;
58 };
59 };
60};
61
62/**
63 * struct bud_entry - entry in the list of buds to replay.
64 * @list: next bud in the list
65 * @bud: bud description object
66 * @sqnum: reference node sequence number
67 * @free: free bytes in the bud
68 * @dirty: dirty bytes in the bud
69 */
70struct bud_entry {
71 struct list_head list;
72 struct ubifs_bud *bud;
73 unsigned long long sqnum;
74 int free;
75 int dirty;
76};
77
78/**
79 * set_bud_lprops - set free and dirty space used by a bud.
80 * @c: UBIFS file-system description object
81 * @b: bud entry which describes the bud
82 *
83 * This function makes sure the LEB properties of bud @b are set correctly
84 * after the replay. Returns zero in case of success and a negative error code
85 * in case of failure.
86 */
87static int set_bud_lprops(struct ubifs_info *c, struct bud_entry *b)
88{
89 const struct ubifs_lprops *lp;
90 int err = 0, dirty;
91
92 ubifs_get_lprops(c);
93
94 lp = ubifs_lpt_lookup_dirty(c, lnum: b->bud->lnum);
95 if (IS_ERR(ptr: lp)) {
96 err = PTR_ERR(ptr: lp);
97 goto out;
98 }
99
100 dirty = lp->dirty;
101 if (b->bud->start == 0 && (lp->free != c->leb_size || lp->dirty != 0)) {
102 /*
103 * The LEB was added to the journal with a starting offset of
104 * zero which means the LEB must have been empty. The LEB
105 * property values should be @lp->free == @c->leb_size and
106 * @lp->dirty == 0, but that is not the case. The reason is that
107 * the LEB had been garbage collected before it became the bud,
108 * and there was no commit in between. The garbage collector
109 * resets the free and dirty space without recording it
110 * anywhere except lprops, so if there was no commit then
111 * lprops does not have that information.
112 *
113 * We do not need to adjust free space because the scan has told
114 * us the exact value which is recorded in the replay entry as
115 * @b->free.
116 *
117 * However we do need to subtract from the dirty space the
118 * amount of space that the garbage collector reclaimed, which
119 * is the whole LEB minus the amount of space that was free.
120 */
121 dbg_mnt("bud LEB %d was GC'd (%d free, %d dirty)", b->bud->lnum,
122 lp->free, lp->dirty);
123 dbg_gc("bud LEB %d was GC'd (%d free, %d dirty)", b->bud->lnum,
124 lp->free, lp->dirty);
125 dirty -= c->leb_size - lp->free;
126 /*
127 * If the replay order was perfect the dirty space would now be
128 * zero. The order is not perfect because the journal heads
129 * race with each other. This is not a problem but is does mean
130 * that the dirty space may temporarily exceed c->leb_size
131 * during the replay.
132 */
133 if (dirty != 0)
134 dbg_mnt("LEB %d lp: %d free %d dirty replay: %d free %d dirty",
135 b->bud->lnum, lp->free, lp->dirty, b->free,
136 b->dirty);
137 }
138 lp = ubifs_change_lp(c, lp, free: b->free, dirty: dirty + b->dirty,
139 flags: lp->flags | LPROPS_TAKEN, idx_gc_cnt: 0);
140 if (IS_ERR(ptr: lp)) {
141 err = PTR_ERR(ptr: lp);
142 goto out;
143 }
144
145 /* Make sure the journal head points to the latest bud */
146 err = ubifs_wbuf_seek_nolock(wbuf: &c->jheads[b->bud->jhead].wbuf,
147 lnum: b->bud->lnum, offs: c->leb_size - b->free);
148
149out:
150 ubifs_release_lprops(c);
151 return err;
152}
153
154/**
155 * set_buds_lprops - set free and dirty space for all replayed buds.
156 * @c: UBIFS file-system description object
157 *
158 * This function sets LEB properties for all replayed buds. Returns zero in
159 * case of success and a negative error code in case of failure.
160 */
161static int set_buds_lprops(struct ubifs_info *c)
162{
163 struct bud_entry *b;
164 int err;
165
166 list_for_each_entry(b, &c->replay_buds, list) {
167 err = set_bud_lprops(c, b);
168 if (err)
169 return err;
170 }
171
172 return 0;
173}
174
175/**
176 * trun_remove_range - apply a replay entry for a truncation to the TNC.
177 * @c: UBIFS file-system description object
178 * @r: replay entry of truncation
179 */
180static int trun_remove_range(struct ubifs_info *c, struct replay_entry *r)
181{
182 unsigned min_blk, max_blk;
183 union ubifs_key min_key, max_key;
184 ino_t ino;
185
186 min_blk = r->new_size / UBIFS_BLOCK_SIZE;
187 if (r->new_size & (UBIFS_BLOCK_SIZE - 1))
188 min_blk += 1;
189
190 max_blk = r->old_size / UBIFS_BLOCK_SIZE;
191 if ((r->old_size & (UBIFS_BLOCK_SIZE - 1)) == 0)
192 max_blk -= 1;
193
194 ino = key_inum(c, k: &r->key);
195
196 data_key_init(c, key: &min_key, inum: ino, block: min_blk);
197 data_key_init(c, key: &max_key, inum: ino, block: max_blk);
198
199 return ubifs_tnc_remove_range(c, from_key: &min_key, to_key: &max_key);
200}
201
202/**
203 * inode_still_linked - check whether inode in question will be re-linked.
204 * @c: UBIFS file-system description object
205 * @rino: replay entry to test
206 *
207 * O_TMPFILE files can be re-linked, this means link count goes from 0 to 1.
208 * This case needs special care, otherwise all references to the inode will
209 * be removed upon the first replay entry of an inode with link count 0
210 * is found.
211 */
212static bool inode_still_linked(struct ubifs_info *c, struct replay_entry *rino)
213{
214 struct replay_entry *r;
215
216 ubifs_assert(c, rino->deletion);
217 ubifs_assert(c, key_type(c, &rino->key) == UBIFS_INO_KEY);
218
219 /*
220 * Find the most recent entry for the inode behind @rino and check
221 * whether it is a deletion.
222 */
223 list_for_each_entry_reverse(r, &c->replay_list, list) {
224 ubifs_assert(c, r->sqnum >= rino->sqnum);
225 if (key_inum(c, k: &r->key) == key_inum(c, k: &rino->key) &&
226 key_type(c, key: &r->key) == UBIFS_INO_KEY)
227 return r->deletion == 0;
228
229 }
230
231 ubifs_assert(c, 0);
232 return false;
233}
234
235/**
236 * apply_replay_entry - apply a replay entry to the TNC.
237 * @c: UBIFS file-system description object
238 * @r: replay entry to apply
239 *
240 * Apply a replay entry to the TNC.
241 */
242static int apply_replay_entry(struct ubifs_info *c, struct replay_entry *r)
243{
244 int err;
245
246 dbg_mntk(&r->key, "LEB %d:%d len %d deletion %d sqnum %llu key ",
247 r->lnum, r->offs, r->len, r->deletion, r->sqnum);
248
249 if (is_hash_key(c, key: &r->key)) {
250 if (r->deletion)
251 err = ubifs_tnc_remove_nm(c, key: &r->key, nm: &r->nm);
252 else
253 err = ubifs_tnc_add_nm(c, key: &r->key, lnum: r->lnum, offs: r->offs,
254 len: r->len, hash: r->hash, nm: &r->nm);
255 } else {
256 if (r->deletion)
257 switch (key_type(c, key: &r->key)) {
258 case UBIFS_INO_KEY:
259 {
260 ino_t inum = key_inum(c, k: &r->key);
261
262 if (inode_still_linked(c, rino: r)) {
263 err = 0;
264 break;
265 }
266
267 err = ubifs_tnc_remove_ino(c, inum);
268 break;
269 }
270 case UBIFS_TRUN_KEY:
271 err = trun_remove_range(c, r);
272 break;
273 default:
274 err = ubifs_tnc_remove(c, key: &r->key);
275 break;
276 }
277 else
278 err = ubifs_tnc_add(c, key: &r->key, lnum: r->lnum, offs: r->offs,
279 len: r->len, hash: r->hash);
280 if (err)
281 return err;
282
283 if (c->need_recovery)
284 err = ubifs_recover_size_accum(c, key: &r->key, deletion: r->deletion,
285 new_size: r->new_size);
286 }
287
288 return err;
289}
290
291/**
292 * replay_entries_cmp - compare 2 replay entries.
293 * @priv: UBIFS file-system description object
294 * @a: first replay entry
295 * @b: second replay entry
296 *
297 * This is a comparios function for 'list_sort()' which compares 2 replay
298 * entries @a and @b by comparing their sequence number. Returns %1 if @a has
299 * greater sequence number and %-1 otherwise.
300 */
301static int replay_entries_cmp(void *priv, const struct list_head *a,
302 const struct list_head *b)
303{
304 struct ubifs_info *c = priv;
305 struct replay_entry *ra, *rb;
306
307 cond_resched();
308 if (a == b)
309 return 0;
310
311 ra = list_entry(a, struct replay_entry, list);
312 rb = list_entry(b, struct replay_entry, list);
313 ubifs_assert(c, ra->sqnum != rb->sqnum);
314 if (ra->sqnum > rb->sqnum)
315 return 1;
316 return -1;
317}
318
319/**
320 * apply_replay_list - apply the replay list to the TNC.
321 * @c: UBIFS file-system description object
322 *
323 * Apply all entries in the replay list to the TNC. Returns zero in case of
324 * success and a negative error code in case of failure.
325 */
326static int apply_replay_list(struct ubifs_info *c)
327{
328 struct replay_entry *r;
329 int err;
330
331 list_sort(priv: c, head: &c->replay_list, cmp: &replay_entries_cmp);
332
333 list_for_each_entry(r, &c->replay_list, list) {
334 cond_resched();
335
336 err = apply_replay_entry(c, r);
337 if (err)
338 return err;
339 }
340
341 return 0;
342}
343
344/**
345 * destroy_replay_list - destroy the replay.
346 * @c: UBIFS file-system description object
347 *
348 * Destroy the replay list.
349 */
350static void destroy_replay_list(struct ubifs_info *c)
351{
352 struct replay_entry *r, *tmp;
353
354 list_for_each_entry_safe(r, tmp, &c->replay_list, list) {
355 if (is_hash_key(c, key: &r->key))
356 kfree(fname_name(&r->nm));
357 list_del(entry: &r->list);
358 kfree(objp: r);
359 }
360}
361
362/**
363 * insert_node - insert a node to the replay list
364 * @c: UBIFS file-system description object
365 * @lnum: node logical eraseblock number
366 * @offs: node offset
367 * @len: node length
368 * @hash: node hash
369 * @key: node key
370 * @sqnum: sequence number
371 * @deletion: non-zero if this is a deletion
372 * @used: number of bytes in use in a LEB
373 * @old_size: truncation old size
374 * @new_size: truncation new size
375 *
376 * This function inserts a scanned non-direntry node to the replay list. The
377 * replay list contains @struct replay_entry elements, and we sort this list in
378 * sequence number order before applying it. The replay list is applied at the
379 * very end of the replay process. Since the list is sorted in sequence number
380 * order, the older modifications are applied first. This function returns zero
381 * in case of success and a negative error code in case of failure.
382 */
383static int insert_node(struct ubifs_info *c, int lnum, int offs, int len,
384 const u8 *hash, union ubifs_key *key,
385 unsigned long long sqnum, int deletion, int *used,
386 loff_t old_size, loff_t new_size)
387{
388 struct replay_entry *r;
389
390 dbg_mntk(key, "add LEB %d:%d, key ", lnum, offs);
391
392 if (key_inum(c, k: key) >= c->highest_inum)
393 c->highest_inum = key_inum(c, k: key);
394
395 r = kzalloc(size: sizeof(struct replay_entry), GFP_KERNEL);
396 if (!r)
397 return -ENOMEM;
398
399 if (!deletion)
400 *used += ALIGN(len, 8);
401 r->lnum = lnum;
402 r->offs = offs;
403 r->len = len;
404 ubifs_copy_hash(c, from: hash, to: r->hash);
405 r->deletion = !!deletion;
406 r->sqnum = sqnum;
407 key_copy(c, from: key, to: &r->key);
408 r->old_size = old_size;
409 r->new_size = new_size;
410
411 list_add_tail(new: &r->list, head: &c->replay_list);
412 return 0;
413}
414
415/**
416 * insert_dent - insert a directory entry node into the replay list.
417 * @c: UBIFS file-system description object
418 * @lnum: node logical eraseblock number
419 * @offs: node offset
420 * @len: node length
421 * @hash: node hash
422 * @key: node key
423 * @name: directory entry name
424 * @nlen: directory entry name length
425 * @sqnum: sequence number
426 * @deletion: non-zero if this is a deletion
427 * @used: number of bytes in use in a LEB
428 *
429 * This function inserts a scanned directory entry node or an extended
430 * attribute entry to the replay list. Returns zero in case of success and a
431 * negative error code in case of failure.
432 */
433static int insert_dent(struct ubifs_info *c, int lnum, int offs, int len,
434 const u8 *hash, union ubifs_key *key,
435 const char *name, int nlen, unsigned long long sqnum,
436 int deletion, int *used)
437{
438 struct replay_entry *r;
439 char *nbuf;
440
441 dbg_mntk(key, "add LEB %d:%d, key ", lnum, offs);
442 if (key_inum(c, k: key) >= c->highest_inum)
443 c->highest_inum = key_inum(c, k: key);
444
445 r = kzalloc(size: sizeof(struct replay_entry), GFP_KERNEL);
446 if (!r)
447 return -ENOMEM;
448
449 nbuf = kmalloc(size: nlen + 1, GFP_KERNEL);
450 if (!nbuf) {
451 kfree(objp: r);
452 return -ENOMEM;
453 }
454
455 if (!deletion)
456 *used += ALIGN(len, 8);
457 r->lnum = lnum;
458 r->offs = offs;
459 r->len = len;
460 ubifs_copy_hash(c, from: hash, to: r->hash);
461 r->deletion = !!deletion;
462 r->sqnum = sqnum;
463 key_copy(c, from: key, to: &r->key);
464 fname_len(&r->nm) = nlen;
465 memcpy(nbuf, name, nlen);
466 nbuf[nlen] = '\0';
467 fname_name(&r->nm) = nbuf;
468
469 list_add_tail(new: &r->list, head: &c->replay_list);
470 return 0;
471}
472
473/**
474 * ubifs_validate_entry - validate directory or extended attribute entry node.
475 * @c: UBIFS file-system description object
476 * @dent: the node to validate
477 *
478 * This function validates directory or extended attribute entry node @dent.
479 * Returns zero if the node is all right and a %-EINVAL if not.
480 */
481int ubifs_validate_entry(struct ubifs_info *c,
482 const struct ubifs_dent_node *dent)
483{
484 int key_type = key_type_flash(c, k: dent->key);
485 int nlen = le16_to_cpu(dent->nlen);
486
487 if (le32_to_cpu(dent->ch.len) != nlen + UBIFS_DENT_NODE_SZ + 1 ||
488 dent->type >= UBIFS_ITYPES_CNT ||
489 nlen > UBIFS_MAX_NLEN || dent->name[nlen] != 0 ||
490 (key_type == UBIFS_XENT_KEY && strnlen(p: dent->name, maxlen: nlen) != nlen) ||
491 le64_to_cpu(dent->inum) > MAX_INUM) {
492 ubifs_err(c, fmt: "bad %s node", key_type == UBIFS_DENT_KEY ?
493 "directory entry" : "extended attribute entry");
494 return -EINVAL;
495 }
496
497 if (key_type != UBIFS_DENT_KEY && key_type != UBIFS_XENT_KEY) {
498 ubifs_err(c, fmt: "bad key type %d", key_type);
499 return -EINVAL;
500 }
501
502 return 0;
503}
504
505/**
506 * is_last_bud - check if the bud is the last in the journal head.
507 * @c: UBIFS file-system description object
508 * @bud: bud description object
509 *
510 * This function checks if bud @bud is the last bud in its journal head. This
511 * information is then used by 'replay_bud()' to decide whether the bud can
512 * have corruptions or not. Indeed, only last buds can be corrupted by power
513 * cuts. Returns %1 if this is the last bud, and %0 if not.
514 */
515static int is_last_bud(struct ubifs_info *c, struct ubifs_bud *bud)
516{
517 struct ubifs_jhead *jh = &c->jheads[bud->jhead];
518 struct ubifs_bud *next;
519 uint32_t data;
520 int err;
521
522 if (list_is_last(list: &bud->list, head: &jh->buds_list))
523 return 1;
524
525 /*
526 * The following is a quirk to make sure we work correctly with UBIFS
527 * images used with older UBIFS.
528 *
529 * Normally, the last bud will be the last in the journal head's list
530 * of bud. However, there is one exception if the UBIFS image belongs
531 * to older UBIFS. This is fairly unlikely: one would need to use old
532 * UBIFS, then have a power cut exactly at the right point, and then
533 * try to mount this image with new UBIFS.
534 *
535 * The exception is: it is possible to have 2 buds A and B, A goes
536 * before B, and B is the last, bud B is contains no data, and bud A is
537 * corrupted at the end. The reason is that in older versions when the
538 * journal code switched the next bud (from A to B), it first added a
539 * log reference node for the new bud (B), and only after this it
540 * synchronized the write-buffer of current bud (A). But later this was
541 * changed and UBIFS started to always synchronize the write-buffer of
542 * the bud (A) before writing the log reference for the new bud (B).
543 *
544 * But because older UBIFS always synchronized A's write-buffer before
545 * writing to B, we can recognize this exceptional situation but
546 * checking the contents of bud B - if it is empty, then A can be
547 * treated as the last and we can recover it.
548 *
549 * TODO: remove this piece of code in a couple of years (today it is
550 * 16.05.2011).
551 */
552 next = list_entry(bud->list.next, struct ubifs_bud, list);
553 if (!list_is_last(list: &next->list, head: &jh->buds_list))
554 return 0;
555
556 err = ubifs_leb_read(c, lnum: next->lnum, buf: (char *)&data, offs: next->start, len: 4, even_ebadmsg: 1);
557 if (err)
558 return 0;
559
560 return data == 0xFFFFFFFF;
561}
562
563/* authenticate_sleb_hash is split out for stack usage */
564static int noinline_for_stack
565authenticate_sleb_hash(struct ubifs_info *c,
566 struct shash_desc *log_hash, u8 *hash)
567{
568 SHASH_DESC_ON_STACK(hash_desc, c->hash_tfm);
569
570 hash_desc->tfm = c->hash_tfm;
571
572 ubifs_shash_copy_state(c, src: log_hash, target: hash_desc);
573 return crypto_shash_final(desc: hash_desc, out: hash);
574}
575
576/**
577 * authenticate_sleb - authenticate one scan LEB
578 * @c: UBIFS file-system description object
579 * @sleb: the scan LEB to authenticate
580 * @log_hash:
581 * @is_last: if true, this is the last LEB
582 *
583 * This function iterates over the buds of a single LEB authenticating all buds
584 * with the authentication nodes on this LEB. Authentication nodes are written
585 * after some buds and contain a HMAC covering the authentication node itself
586 * and the buds between the last authentication node and the current
587 * authentication node. It can happen that the last buds cannot be authenticated
588 * because a powercut happened when some nodes were written but not the
589 * corresponding authentication node. This function returns the number of nodes
590 * that could be authenticated or a negative error code.
591 */
592static int authenticate_sleb(struct ubifs_info *c, struct ubifs_scan_leb *sleb,
593 struct shash_desc *log_hash, int is_last)
594{
595 int n_not_auth = 0;
596 struct ubifs_scan_node *snod;
597 int n_nodes = 0;
598 int err;
599 u8 hash[UBIFS_HASH_ARR_SZ];
600 u8 hmac[UBIFS_HMAC_ARR_SZ];
601
602 if (!ubifs_authenticated(c))
603 return sleb->nodes_cnt;
604
605 list_for_each_entry(snod, &sleb->nodes, list) {
606
607 n_nodes++;
608
609 if (snod->type == UBIFS_AUTH_NODE) {
610 struct ubifs_auth_node *auth = snod->node;
611
612 err = authenticate_sleb_hash(c, log_hash, hash);
613 if (err)
614 goto out;
615
616 err = crypto_shash_tfm_digest(tfm: c->hmac_tfm, data: hash,
617 len: c->hash_len, out: hmac);
618 if (err)
619 goto out;
620
621 err = ubifs_check_hmac(c, expected: auth->hmac, got: hmac);
622 if (err) {
623 err = -EPERM;
624 goto out;
625 }
626 n_not_auth = 0;
627 } else {
628 err = crypto_shash_update(desc: log_hash, data: snod->node,
629 len: snod->len);
630 if (err)
631 goto out;
632 n_not_auth++;
633 }
634 }
635
636 /*
637 * A powercut can happen when some nodes were written, but not yet
638 * the corresponding authentication node. This may only happen on
639 * the last bud though.
640 */
641 if (n_not_auth) {
642 if (is_last) {
643 dbg_mnt("%d unauthenticated nodes found on LEB %d, Ignoring them",
644 n_not_auth, sleb->lnum);
645 err = 0;
646 } else {
647 dbg_mnt("%d unauthenticated nodes found on non-last LEB %d",
648 n_not_auth, sleb->lnum);
649 err = -EPERM;
650 }
651 } else {
652 err = 0;
653 }
654out:
655 return err ? err : n_nodes - n_not_auth;
656}
657
658/**
659 * replay_bud - replay a bud logical eraseblock.
660 * @c: UBIFS file-system description object
661 * @b: bud entry which describes the bud
662 *
663 * This function replays bud @bud, recovers it if needed, and adds all nodes
664 * from this bud to the replay list. Returns zero in case of success and a
665 * negative error code in case of failure.
666 */
667static int replay_bud(struct ubifs_info *c, struct bud_entry *b)
668{
669 int is_last = is_last_bud(c, bud: b->bud);
670 int err = 0, used = 0, lnum = b->bud->lnum, offs = b->bud->start;
671 int n_nodes, n = 0;
672 struct ubifs_scan_leb *sleb;
673 struct ubifs_scan_node *snod;
674
675 dbg_mnt("replay bud LEB %d, head %d, offs %d, is_last %d",
676 lnum, b->bud->jhead, offs, is_last);
677
678 if (c->need_recovery && is_last)
679 /*
680 * Recover only last LEBs in the journal heads, because power
681 * cuts may cause corruptions only in these LEBs, because only
682 * these LEBs could possibly be written to at the power cut
683 * time.
684 */
685 sleb = ubifs_recover_leb(c, lnum, offs, sbuf: c->sbuf, jhead: b->bud->jhead);
686 else
687 sleb = ubifs_scan(c, lnum, offs, sbuf: c->sbuf, quiet: 0);
688 if (IS_ERR(ptr: sleb))
689 return PTR_ERR(ptr: sleb);
690
691 n_nodes = authenticate_sleb(c, sleb, log_hash: b->bud->log_hash, is_last);
692 if (n_nodes < 0) {
693 err = n_nodes;
694 goto out;
695 }
696
697 ubifs_shash_copy_state(c, src: b->bud->log_hash,
698 target: c->jheads[b->bud->jhead].log_hash);
699
700 /*
701 * The bud does not have to start from offset zero - the beginning of
702 * the 'lnum' LEB may contain previously committed data. One of the
703 * things we have to do in replay is to correctly update lprops with
704 * newer information about this LEB.
705 *
706 * At this point lprops thinks that this LEB has 'c->leb_size - offs'
707 * bytes of free space because it only contain information about
708 * committed data.
709 *
710 * But we know that real amount of free space is 'c->leb_size -
711 * sleb->endpt', and the space in the 'lnum' LEB between 'offs' and
712 * 'sleb->endpt' is used by bud data. We have to correctly calculate
713 * how much of these data are dirty and update lprops with this
714 * information.
715 *
716 * The dirt in that LEB region is comprised of padding nodes, deletion
717 * nodes, truncation nodes and nodes which are obsoleted by subsequent
718 * nodes in this LEB. So instead of calculating clean space, we
719 * calculate used space ('used' variable).
720 */
721
722 list_for_each_entry(snod, &sleb->nodes, list) {
723 u8 hash[UBIFS_HASH_ARR_SZ];
724 int deletion = 0;
725
726 cond_resched();
727
728 if (snod->sqnum >= SQNUM_WATERMARK) {
729 ubifs_err(c, fmt: "file system's life ended");
730 goto out_dump;
731 }
732
733 ubifs_node_calc_hash(c, buf: snod->node, hash);
734
735 if (snod->sqnum > c->max_sqnum)
736 c->max_sqnum = snod->sqnum;
737
738 switch (snod->type) {
739 case UBIFS_INO_NODE:
740 {
741 struct ubifs_ino_node *ino = snod->node;
742 loff_t new_size = le64_to_cpu(ino->size);
743
744 if (le32_to_cpu(ino->nlink) == 0)
745 deletion = 1;
746 err = insert_node(c, lnum, offs: snod->offs, len: snod->len, hash,
747 key: &snod->key, sqnum: snod->sqnum, deletion,
748 used: &used, old_size: 0, new_size);
749 break;
750 }
751 case UBIFS_DATA_NODE:
752 {
753 struct ubifs_data_node *dn = snod->node;
754 loff_t new_size = le32_to_cpu(dn->size) +
755 key_block(c, key: &snod->key) *
756 UBIFS_BLOCK_SIZE;
757
758 err = insert_node(c, lnum, offs: snod->offs, len: snod->len, hash,
759 key: &snod->key, sqnum: snod->sqnum, deletion,
760 used: &used, old_size: 0, new_size);
761 break;
762 }
763 case UBIFS_DENT_NODE:
764 case UBIFS_XENT_NODE:
765 {
766 struct ubifs_dent_node *dent = snod->node;
767
768 err = ubifs_validate_entry(c, dent);
769 if (err)
770 goto out_dump;
771
772 err = insert_dent(c, lnum, offs: snod->offs, len: snod->len, hash,
773 key: &snod->key, name: dent->name,
774 le16_to_cpu(dent->nlen), sqnum: snod->sqnum,
775 deletion: !le64_to_cpu(dent->inum), used: &used);
776 break;
777 }
778 case UBIFS_TRUN_NODE:
779 {
780 struct ubifs_trun_node *trun = snod->node;
781 loff_t old_size = le64_to_cpu(trun->old_size);
782 loff_t new_size = le64_to_cpu(trun->new_size);
783 union ubifs_key key;
784
785 /* Validate truncation node */
786 if (old_size < 0 || old_size > c->max_inode_sz ||
787 new_size < 0 || new_size > c->max_inode_sz ||
788 old_size <= new_size) {
789 ubifs_err(c, fmt: "bad truncation node");
790 goto out_dump;
791 }
792
793 /*
794 * Create a fake truncation key just to use the same
795 * functions which expect nodes to have keys.
796 */
797 trun_key_init(c, key: &key, le32_to_cpu(trun->inum));
798 err = insert_node(c, lnum, offs: snod->offs, len: snod->len, hash,
799 key: &key, sqnum: snod->sqnum, deletion: 1, used: &used,
800 old_size, new_size);
801 break;
802 }
803 case UBIFS_AUTH_NODE:
804 break;
805 default:
806 ubifs_err(c, fmt: "unexpected node type %d in bud LEB %d:%d",
807 snod->type, lnum, snod->offs);
808 err = -EINVAL;
809 goto out_dump;
810 }
811 if (err)
812 goto out;
813
814 n++;
815 if (n == n_nodes)
816 break;
817 }
818
819 ubifs_assert(c, ubifs_search_bud(c, lnum));
820 ubifs_assert(c, sleb->endpt - offs >= used);
821 ubifs_assert(c, sleb->endpt % c->min_io_size == 0);
822
823 b->dirty = sleb->endpt - offs - used;
824 b->free = c->leb_size - sleb->endpt;
825 dbg_mnt("bud LEB %d replied: dirty %d, free %d",
826 lnum, b->dirty, b->free);
827
828out:
829 ubifs_scan_destroy(sleb);
830 return err;
831
832out_dump:
833 ubifs_err(c, fmt: "bad node is at LEB %d:%d", lnum, snod->offs);
834 ubifs_dump_node(c, node: snod->node, node_len: c->leb_size - snod->offs);
835 ubifs_scan_destroy(sleb);
836 return -EINVAL;
837}
838
839/**
840 * replay_buds - replay all buds.
841 * @c: UBIFS file-system description object
842 *
843 * This function returns zero in case of success and a negative error code in
844 * case of failure.
845 */
846static int replay_buds(struct ubifs_info *c)
847{
848 struct bud_entry *b;
849 int err;
850 unsigned long long prev_sqnum = 0;
851
852 list_for_each_entry(b, &c->replay_buds, list) {
853 err = replay_bud(c, b);
854 if (err)
855 return err;
856
857 ubifs_assert(c, b->sqnum > prev_sqnum);
858 prev_sqnum = b->sqnum;
859 }
860
861 return 0;
862}
863
864/**
865 * destroy_bud_list - destroy the list of buds to replay.
866 * @c: UBIFS file-system description object
867 */
868static void destroy_bud_list(struct ubifs_info *c)
869{
870 struct bud_entry *b;
871
872 while (!list_empty(head: &c->replay_buds)) {
873 b = list_entry(c->replay_buds.next, struct bud_entry, list);
874 list_del(entry: &b->list);
875 kfree(objp: b);
876 }
877}
878
879/**
880 * add_replay_bud - add a bud to the list of buds to replay.
881 * @c: UBIFS file-system description object
882 * @lnum: bud logical eraseblock number to replay
883 * @offs: bud start offset
884 * @jhead: journal head to which this bud belongs
885 * @sqnum: reference node sequence number
886 *
887 * This function returns zero in case of success and a negative error code in
888 * case of failure.
889 */
890static int add_replay_bud(struct ubifs_info *c, int lnum, int offs, int jhead,
891 unsigned long long sqnum)
892{
893 struct ubifs_bud *bud;
894 struct bud_entry *b;
895 int err;
896
897 dbg_mnt("add replay bud LEB %d:%d, head %d", lnum, offs, jhead);
898
899 bud = kmalloc(size: sizeof(struct ubifs_bud), GFP_KERNEL);
900 if (!bud)
901 return -ENOMEM;
902
903 b = kmalloc(size: sizeof(struct bud_entry), GFP_KERNEL);
904 if (!b) {
905 err = -ENOMEM;
906 goto out;
907 }
908
909 bud->lnum = lnum;
910 bud->start = offs;
911 bud->jhead = jhead;
912 bud->log_hash = ubifs_hash_get_desc(c);
913 if (IS_ERR(ptr: bud->log_hash)) {
914 err = PTR_ERR(ptr: bud->log_hash);
915 goto out;
916 }
917
918 ubifs_shash_copy_state(c, src: c->log_hash, target: bud->log_hash);
919
920 ubifs_add_bud(c, bud);
921
922 b->bud = bud;
923 b->sqnum = sqnum;
924 list_add_tail(new: &b->list, head: &c->replay_buds);
925
926 return 0;
927out:
928 kfree(objp: bud);
929 kfree(objp: b);
930
931 return err;
932}
933
934/**
935 * validate_ref - validate a reference node.
936 * @c: UBIFS file-system description object
937 * @ref: the reference node to validate
938 *
939 * This function returns %1 if a bud reference already exists for the LEB. %0 is
940 * returned if the reference node is new, otherwise %-EINVAL is returned if
941 * validation failed.
942 */
943static int validate_ref(struct ubifs_info *c, const struct ubifs_ref_node *ref)
944{
945 struct ubifs_bud *bud;
946 int lnum = le32_to_cpu(ref->lnum);
947 unsigned int offs = le32_to_cpu(ref->offs);
948 unsigned int jhead = le32_to_cpu(ref->jhead);
949
950 /*
951 * ref->offs may point to the end of LEB when the journal head points
952 * to the end of LEB and we write reference node for it during commit.
953 * So this is why we require 'offs > c->leb_size'.
954 */
955 if (jhead >= c->jhead_cnt || lnum >= c->leb_cnt ||
956 lnum < c->main_first || offs > c->leb_size ||
957 offs & (c->min_io_size - 1))
958 return -EINVAL;
959
960 /* Make sure we have not already looked at this bud */
961 bud = ubifs_search_bud(c, lnum);
962 if (bud) {
963 if (bud->jhead == jhead && bud->start <= offs)
964 return 1;
965 ubifs_err(c, fmt: "bud at LEB %d:%d was already referred", lnum, offs);
966 return -EINVAL;
967 }
968
969 return 0;
970}
971
972/**
973 * replay_log_leb - replay a log logical eraseblock.
974 * @c: UBIFS file-system description object
975 * @lnum: log logical eraseblock to replay
976 * @offs: offset to start replaying from
977 * @sbuf: scan buffer
978 *
979 * This function replays a log LEB and returns zero in case of success, %1 if
980 * this is the last LEB in the log, and a negative error code in case of
981 * failure.
982 */
983static int replay_log_leb(struct ubifs_info *c, int lnum, int offs, void *sbuf)
984{
985 int err;
986 struct ubifs_scan_leb *sleb;
987 struct ubifs_scan_node *snod;
988 const struct ubifs_cs_node *node;
989
990 dbg_mnt("replay log LEB %d:%d", lnum, offs);
991 sleb = ubifs_scan(c, lnum, offs, sbuf, quiet: c->need_recovery);
992 if (IS_ERR(ptr: sleb)) {
993 if (PTR_ERR(ptr: sleb) != -EUCLEAN || !c->need_recovery)
994 return PTR_ERR(ptr: sleb);
995 /*
996 * Note, the below function will recover this log LEB only if
997 * it is the last, because unclean reboots can possibly corrupt
998 * only the tail of the log.
999 */
1000 sleb = ubifs_recover_log_leb(c, lnum, offs, sbuf);
1001 if (IS_ERR(ptr: sleb))
1002 return PTR_ERR(ptr: sleb);
1003 }
1004
1005 if (sleb->nodes_cnt == 0) {
1006 err = 1;
1007 goto out;
1008 }
1009
1010 node = sleb->buf;
1011 snod = list_entry(sleb->nodes.next, struct ubifs_scan_node, list);
1012 if (c->cs_sqnum == 0) {
1013 /*
1014 * This is the first log LEB we are looking at, make sure that
1015 * the first node is a commit start node. Also record its
1016 * sequence number so that UBIFS can determine where the log
1017 * ends, because all nodes which were have higher sequence
1018 * numbers.
1019 */
1020 if (snod->type != UBIFS_CS_NODE) {
1021 ubifs_err(c, fmt: "first log node at LEB %d:%d is not CS node",
1022 lnum, offs);
1023 goto out_dump;
1024 }
1025 if (le64_to_cpu(node->cmt_no) != c->cmt_no) {
1026 ubifs_err(c, fmt: "first CS node at LEB %d:%d has wrong commit number %llu expected %llu",
1027 lnum, offs,
1028 (unsigned long long)le64_to_cpu(node->cmt_no),
1029 c->cmt_no);
1030 goto out_dump;
1031 }
1032
1033 c->cs_sqnum = le64_to_cpu(node->ch.sqnum);
1034 dbg_mnt("commit start sqnum %llu", c->cs_sqnum);
1035
1036 err = ubifs_shash_init(c, desc: c->log_hash);
1037 if (err)
1038 goto out;
1039
1040 err = ubifs_shash_update(c, desc: c->log_hash, buf: node, UBIFS_CS_NODE_SZ);
1041 if (err < 0)
1042 goto out;
1043 }
1044
1045 if (snod->sqnum < c->cs_sqnum) {
1046 /*
1047 * This means that we reached end of log and now
1048 * look to the older log data, which was already
1049 * committed but the eraseblock was not erased (UBIFS
1050 * only un-maps it). So this basically means we have to
1051 * exit with "end of log" code.
1052 */
1053 err = 1;
1054 goto out;
1055 }
1056
1057 /* Make sure the first node sits at offset zero of the LEB */
1058 if (snod->offs != 0) {
1059 ubifs_err(c, fmt: "first node is not at zero offset");
1060 goto out_dump;
1061 }
1062
1063 list_for_each_entry(snod, &sleb->nodes, list) {
1064 cond_resched();
1065
1066 if (snod->sqnum >= SQNUM_WATERMARK) {
1067 ubifs_err(c, fmt: "file system's life ended");
1068 goto out_dump;
1069 }
1070
1071 if (snod->sqnum < c->cs_sqnum) {
1072 ubifs_err(c, fmt: "bad sqnum %llu, commit sqnum %llu",
1073 snod->sqnum, c->cs_sqnum);
1074 goto out_dump;
1075 }
1076
1077 if (snod->sqnum > c->max_sqnum)
1078 c->max_sqnum = snod->sqnum;
1079
1080 switch (snod->type) {
1081 case UBIFS_REF_NODE: {
1082 const struct ubifs_ref_node *ref = snod->node;
1083
1084 err = validate_ref(c, ref);
1085 if (err == 1)
1086 break; /* Already have this bud */
1087 if (err)
1088 goto out_dump;
1089
1090 err = ubifs_shash_update(c, desc: c->log_hash, buf: ref,
1091 UBIFS_REF_NODE_SZ);
1092 if (err)
1093 goto out;
1094
1095 err = add_replay_bud(c, le32_to_cpu(ref->lnum),
1096 le32_to_cpu(ref->offs),
1097 le32_to_cpu(ref->jhead),
1098 sqnum: snod->sqnum);
1099 if (err)
1100 goto out;
1101
1102 break;
1103 }
1104 case UBIFS_CS_NODE:
1105 /* Make sure it sits at the beginning of LEB */
1106 if (snod->offs != 0) {
1107 ubifs_err(c, fmt: "unexpected node in log");
1108 goto out_dump;
1109 }
1110 break;
1111 default:
1112 ubifs_err(c, fmt: "unexpected node in log");
1113 goto out_dump;
1114 }
1115 }
1116
1117 if (sleb->endpt || c->lhead_offs >= c->leb_size) {
1118 c->lhead_lnum = lnum;
1119 c->lhead_offs = sleb->endpt;
1120 }
1121
1122 err = !sleb->endpt;
1123out:
1124 ubifs_scan_destroy(sleb);
1125 return err;
1126
1127out_dump:
1128 ubifs_err(c, fmt: "log error detected while replaying the log at LEB %d:%d",
1129 lnum, offs + snod->offs);
1130 ubifs_dump_node(c, node: snod->node, node_len: c->leb_size - snod->offs);
1131 ubifs_scan_destroy(sleb);
1132 return -EINVAL;
1133}
1134
1135/**
1136 * take_ihead - update the status of the index head in lprops to 'taken'.
1137 * @c: UBIFS file-system description object
1138 *
1139 * This function returns the amount of free space in the index head LEB or a
1140 * negative error code.
1141 */
1142static int take_ihead(struct ubifs_info *c)
1143{
1144 const struct ubifs_lprops *lp;
1145 int err, free;
1146
1147 ubifs_get_lprops(c);
1148
1149 lp = ubifs_lpt_lookup_dirty(c, lnum: c->ihead_lnum);
1150 if (IS_ERR(ptr: lp)) {
1151 err = PTR_ERR(ptr: lp);
1152 goto out;
1153 }
1154
1155 free = lp->free;
1156
1157 lp = ubifs_change_lp(c, lp, LPROPS_NC, LPROPS_NC,
1158 flags: lp->flags | LPROPS_TAKEN, idx_gc_cnt: 0);
1159 if (IS_ERR(ptr: lp)) {
1160 err = PTR_ERR(ptr: lp);
1161 goto out;
1162 }
1163
1164 err = free;
1165out:
1166 ubifs_release_lprops(c);
1167 return err;
1168}
1169
1170/**
1171 * ubifs_replay_journal - replay journal.
1172 * @c: UBIFS file-system description object
1173 *
1174 * This function scans the journal, replays and cleans it up. It makes sure all
1175 * memory data structures related to uncommitted journal are built (dirty TNC
1176 * tree, tree of buds, modified lprops, etc).
1177 */
1178int ubifs_replay_journal(struct ubifs_info *c)
1179{
1180 int err, lnum, free;
1181
1182 BUILD_BUG_ON(UBIFS_TRUN_KEY > 5);
1183
1184 /* Update the status of the index head in lprops to 'taken' */
1185 free = take_ihead(c);
1186 if (free < 0)
1187 return free; /* Error code */
1188
1189 if (c->ihead_offs != c->leb_size - free) {
1190 ubifs_err(c, fmt: "bad index head LEB %d:%d", c->ihead_lnum,
1191 c->ihead_offs);
1192 return -EINVAL;
1193 }
1194
1195 dbg_mnt("start replaying the journal");
1196 c->replaying = 1;
1197 lnum = c->ltail_lnum = c->lhead_lnum;
1198
1199 do {
1200 err = replay_log_leb(c, lnum, offs: 0, sbuf: c->sbuf);
1201 if (err == 1) {
1202 if (lnum != c->lhead_lnum)
1203 /* We hit the end of the log */
1204 break;
1205
1206 /*
1207 * The head of the log must always start with the
1208 * "commit start" node on a properly formatted UBIFS.
1209 * But we found no nodes at all, which means that
1210 * something went wrong and we cannot proceed mounting
1211 * the file-system.
1212 */
1213 ubifs_err(c, fmt: "no UBIFS nodes found at the log head LEB %d:%d, possibly corrupted",
1214 lnum, 0);
1215 err = -EINVAL;
1216 }
1217 if (err)
1218 goto out;
1219 lnum = ubifs_next_log_lnum(c, lnum);
1220 } while (lnum != c->ltail_lnum);
1221
1222 err = replay_buds(c);
1223 if (err)
1224 goto out;
1225
1226 err = apply_replay_list(c);
1227 if (err)
1228 goto out;
1229
1230 err = set_buds_lprops(c);
1231 if (err)
1232 goto out;
1233
1234 /*
1235 * UBIFS budgeting calculations use @c->bi.uncommitted_idx variable
1236 * to roughly estimate index growth. Things like @c->bi.min_idx_lebs
1237 * depend on it. This means we have to initialize it to make sure
1238 * budgeting works properly.
1239 */
1240 c->bi.uncommitted_idx = atomic_long_read(v: &c->dirty_zn_cnt);
1241 c->bi.uncommitted_idx *= c->max_idx_node_sz;
1242
1243 ubifs_assert(c, c->bud_bytes <= c->max_bud_bytes || c->need_recovery);
1244 dbg_mnt("finished, log head LEB %d:%d, max_sqnum %llu, highest_inum %lu",
1245 c->lhead_lnum, c->lhead_offs, c->max_sqnum,
1246 (unsigned long)c->highest_inum);
1247out:
1248 destroy_replay_list(c);
1249 destroy_bud_list(c);
1250 c->replaying = 0;
1251 return err;
1252}
1253

source code of linux/fs/ubifs/replay.c