2 Trivial Database 2: free list/block handling
3 Copyright (C) Rusty Russell 2010
5 This library is free software; you can redistribute it and/or
6 modify it under the terms of the GNU Lesser General Public
7 License as published by the Free Software Foundation; either
8 version 3 of the License, or (at your option) any later version.
10 This library is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 Lesser General Public License for more details.
15 You should have received a copy of the GNU Lesser General Public
16 License along with this library; if not, see <http://www.gnu.org/licenses/>.
19 #include <ccan/likely/likely.h>
20 #include <ccan/asearch/asearch.h>
22 /* We keep an ordered array of offsets. */
23 static bool append(tdb_off_t **arr, size_t *num, tdb_off_t off)
25 tdb_off_t *new = realloc(*arr, (*num + 1) * sizeof(tdb_off_t));
33 static bool check_header(struct tdb_context *tdb, tdb_off_t *recovery)
36 struct tdb_header hdr;
38 if (tdb_read_convert(tdb, 0, &hdr, sizeof(hdr)) == -1)
40 /* magic food should not be converted, so convert back. */
41 tdb_convert(tdb, hdr.magic_food, sizeof(hdr.magic_food));
43 hash_test = TDB_HASH_MAGIC;
44 hash_test = tdb_hash(tdb, &hash_test, sizeof(hash_test));
45 if (hdr.hash_test != hash_test) {
46 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
47 "check: hash test %llu should be %llu\n",
48 (long long)hdr.hash_test,
49 (long long)hash_test);
53 if (strcmp(hdr.magic_food, TDB_MAGIC_FOOD) != 0) {
54 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
55 "check: bad magic '%.*s'\n",
56 (unsigned)sizeof(hdr.magic_food), hdr.magic_food);
60 *recovery = hdr.recovery;
62 if (*recovery < sizeof(hdr) || *recovery > tdb->map_size) {
63 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
64 "tdb_check: invalid recovery offset %zu\n",
70 /* Don't check reserved: they *can* be used later. */
74 static bool check_hash_tree(struct tdb_context *tdb,
75 tdb_off_t off, unsigned int group_bits,
77 unsigned hprefix_bits,
81 int (*check)(TDB_DATA, TDB_DATA, void *),
84 static bool check_hash_record(struct tdb_context *tdb,
87 unsigned hprefix_bits,
91 int (*check)(TDB_DATA, TDB_DATA, void *),
94 struct tdb_used_record rec;
96 if (tdb_read_convert(tdb, off, &rec, sizeof(rec)) == -1)
99 if (rec_data_length(&rec)
100 != sizeof(tdb_off_t) << TDB_SUBLEVEL_HASH_BITS) {
101 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
102 "tdb_check: Bad hash table length %llu vs %llu\n",
103 (long long)rec_data_length(&rec),
104 (long long)sizeof(tdb_off_t)<<TDB_SUBLEVEL_HASH_BITS);
107 if (rec_key_length(&rec) != 0) {
108 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
109 "tdb_check: Bad hash table key length %llu\n",
110 (long long)rec_key_length(&rec));
113 if (rec_hash(&rec) != 0) {
114 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
115 "tdb_check: Bad hash table hash value %llu\n",
116 (long long)rec_hash(&rec));
121 return check_hash_tree(tdb, off,
122 TDB_SUBLEVEL_HASH_BITS-TDB_HASH_GROUP_BITS,
123 hprefix, hprefix_bits,
124 used, num_used, num_found, check, private_data);
127 static int off_cmp(const tdb_off_t *a, const tdb_off_t *b)
129 /* Can overflow an int. */
135 static uint64_t get_bits(uint64_t h, unsigned num, unsigned *used)
139 return (h >> (64 - *used)) & ((1U << num) - 1);
142 static bool check_hash_tree(struct tdb_context *tdb,
143 tdb_off_t off, unsigned int group_bits,
145 unsigned hprefix_bits,
149 int (*check)(TDB_DATA, TDB_DATA, void *),
153 const tdb_off_t *hash;
154 struct tdb_used_record rec;
156 hash = tdb_access_read(tdb, off,
158 << (group_bits + TDB_HASH_GROUP_BITS),
163 for (g = 0; g < (1 << group_bits); g++) {
164 const tdb_off_t *group = hash + (g << TDB_HASH_GROUP_BITS);
165 for (b = 0; b < (1 << TDB_HASH_GROUP_BITS); b++) {
166 unsigned int bucket, i, used_bits;
172 off = group[b] & TDB_OFF_MASK;
173 p = asearch(&off, used, num_used, off_cmp);
175 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
176 "tdb_check: Invalid offset %llu "
181 /* Mark it invalid. */
185 if (is_subhash(group[b])) {
188 << (group_bits + TDB_HASH_GROUP_BITS))
189 + g * (1 << TDB_HASH_GROUP_BITS) + b;
191 if (!check_hash_record(tdb,
192 group[b] & TDB_OFF_MASK,
196 + TDB_HASH_GROUP_BITS,
197 used, num_used, num_found,
198 check, private_data))
204 /* Does it belong here at all? */
205 h = hash_record(tdb, off);
207 if (get_bits(h, hprefix_bits, &used_bits) != hprefix
209 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
210 "check: bad hash placement"
211 " 0x%llx vs 0x%llx\n",
212 (long long)h, (long long)hprefix);
216 /* Does it belong in this group? */
217 if (get_bits(h, group_bits, &used_bits) != g) {
218 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
219 "check: bad group %llu vs %u\n",
224 /* Are bucket bits correct? */
225 bucket = group[b] & TDB_OFF_HASH_GROUP_MASK;
226 if (get_bits(h, TDB_HASH_GROUP_BITS, &used_bits)
228 used_bits -= TDB_HASH_GROUP_BITS;
229 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
230 "check: bad bucket %u vs %u\n",
231 (unsigned)get_bits(h,
238 /* There must not be any zero entries between
239 * the bucket it belongs in and this one! */
242 i = (i + 1) % (1 << TDB_HASH_GROUP_BITS)) {
244 tdb->log(tdb, TDB_DEBUG_ERROR,
246 "check: bad group placement"
253 if (tdb_read_convert(tdb, off, &rec, sizeof(rec)) == -1)
256 /* Bottom bits must match header. */
257 if ((h & ((1 << 11)-1)) != rec_hash(&rec)) {
258 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
259 "tdb_check: Bad hash magic at"
260 " offset %llu (0x%llx vs 0x%llx)\n",
263 (long long)rec_hash(&rec));
269 key.dsize = rec_key_length(&rec);
270 data.dsize = rec_data_length(&rec);
271 key.dptr = (void *)tdb_access_read(tdb,
273 key.dsize + data.dsize,
277 data.dptr = key.dptr + key.dsize;
278 if (check(key, data, private_data) != 0)
280 tdb_access_release(tdb, key.dptr);
284 tdb_access_release(tdb, hash);
288 tdb_access_release(tdb, hash);
292 static bool check_hash(struct tdb_context *tdb,
294 size_t num_used, size_t num_flists,
295 int (*check)(TDB_DATA, TDB_DATA, void *),
298 /* Free lists also show up as used. */
299 size_t num_found = num_flists;
301 if (!check_hash_tree(tdb, offsetof(struct tdb_header, hashtable),
302 TDB_TOPLEVEL_HASH_BITS-TDB_HASH_GROUP_BITS,
303 0, 0, used, num_used, &num_found,
304 check, private_data))
307 if (num_found != num_used) {
308 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
309 "tdb_check: Not all entries are in hash\n");
315 static bool check_free(struct tdb_context *tdb,
317 const struct tdb_free_record *frec,
318 tdb_off_t prev, unsigned int flist, unsigned int bucket)
320 if (frec_magic(frec) != TDB_FREE_MAGIC) {
321 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
322 "tdb_check: offset %llu bad magic 0x%llx\n",
323 (long long)off, (long long)frec->magic_and_prev);
326 if (frec_flist(frec) != flist) {
327 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
328 "tdb_check: offset %llu bad freelist %u\n",
329 (long long)off, frec_flist(frec));
333 if (tdb->methods->oob(tdb, off
334 + frec_len(frec) + sizeof(struct tdb_used_record),
337 if (size_to_bucket(frec_len(frec)) != bucket) {
338 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
339 "tdb_check: offset %llu in wrong bucket %u vs %u\n",
341 bucket, size_to_bucket(frec_len(frec)));
344 if (prev != frec_prev(frec)) {
345 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
346 "tdb_check: offset %llu bad prev %llu vs %llu\n",
348 (long long)prev, (long long)frec_len(frec));
354 static bool check_free_list(struct tdb_context *tdb,
361 struct tdb_freelist flist;
365 if (tdb_read_convert(tdb, flist_off, &flist, sizeof(flist)) == -1)
368 if (rec_magic(&flist.hdr) != TDB_MAGIC
369 || rec_key_length(&flist.hdr) != 0
370 || rec_data_length(&flist.hdr) != sizeof(flist) - sizeof(flist.hdr)
371 || rec_hash(&flist.hdr) != 1) {
372 tdb->log(tdb, TDB_DEBUG_ERROR,
374 "tdb_check: Invalid header on free list\n");
378 for (i = 0; i < TDB_FREE_BUCKETS; i++) {
379 tdb_off_t off, prev = 0, *p;
380 struct tdb_free_record f;
382 h = bucket_off(flist_off, i);
383 for (off = tdb_read_off(tdb, h); off; off = f.next) {
384 if (off == TDB_OFF_ERR)
386 if (tdb_read_convert(tdb, off, &f, sizeof(f)))
388 if (!check_free(tdb, off, &f, prev, flist_num, i))
391 /* FIXME: Check hash bits */
392 p = asearch(&off, free, num_free, off_cmp);
394 tdb->log(tdb, TDB_DEBUG_ERROR,
396 "tdb_check: Invalid offset"
397 " %llu in free table\n",
401 /* Mark it invalid. */
410 /* Slow, but should be very rare. */
411 size_t dead_space(struct tdb_context *tdb, tdb_off_t off)
415 for (len = 0; off + len < tdb->map_size; len++) {
417 if (tdb->methods->read(tdb, off, &c, 1))
419 if (c != 0 && c != 0x43)
425 static bool check_linear(struct tdb_context *tdb,
426 tdb_off_t **used, size_t *num_used,
427 tdb_off_t **free, size_t *num_free,
432 bool found_recovery = false;
434 for (off = sizeof(struct tdb_header); off < tdb->map_size; off += len) {
436 struct tdb_used_record u;
437 struct tdb_free_record f;
438 struct tdb_recovery_record r;
440 /* r is larger: only get that if we need to. */
441 p = tdb_get(tdb, off, &pad, sizeof(pad.f));
445 /* If we crash after ftruncate, we can get zeroes or fill. */
446 if (p->r.magic == TDB_RECOVERY_INVALID_MAGIC
447 || p->r.magic == 0x4343434343434343ULL) {
448 p = tdb_get(tdb, off, &pad, sizeof(pad.r));
451 if (recovery == off) {
452 found_recovery = true;
453 len = sizeof(p->r) + p->r.max_len;
455 len = dead_space(tdb, off);
456 if (len < sizeof(p->r)) {
457 tdb->log(tdb, TDB_DEBUG_ERROR,
459 "tdb_check: invalid dead space"
460 " at %zu\n", (size_t)off);
464 tdb->log(tdb, TDB_DEBUG_WARNING, tdb->log_priv,
465 "Dead space at %zu-%zu (of %zu)\n",
466 (size_t)off, (size_t)(off + len),
467 (size_t)tdb->map_size);
469 } else if (p->r.magic == TDB_RECOVERY_MAGIC) {
470 p = tdb_get(tdb, off, &pad, sizeof(pad.r));
473 if (recovery != off) {
474 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
475 "tdb_check: unexpected recovery"
476 " record at offset %zu\n",
480 if (p->r.len > p->r.max_len) {
481 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
482 "tdb_check: invalid recovery length"
483 " %zu\n", (size_t)p->r.len);
486 if (p->r.eof > tdb->map_size) {
487 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
488 "tdb_check: invalid old EOF"
489 " %zu\n", (size_t)p->r.eof);
492 found_recovery = true;
493 len = sizeof(p->r) + p->r.max_len;
494 } else if (frec_magic(&p->f) == TDB_FREE_MAGIC
495 || frec_magic(&p->f) == TDB_COALESCING_MAGIC) {
496 len = sizeof(p->u) + frec_len(&p->f);
497 if (off + len > tdb->map_size) {
498 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
499 "tdb_check: free overlength %llu"
501 (long long)len, (long long)off);
504 /* This record is free! */
505 if (frec_magic(&p->f) == TDB_FREE_MAGIC
506 && !append(free, num_free, off))
509 uint64_t klen, dlen, extra;
511 /* This record is used! */
512 if (rec_magic(&p->u) != TDB_MAGIC) {
513 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
514 "tdb_check: Bad magic 0x%llx"
516 (long long)rec_magic(&p->u),
521 if (!append(used, num_used, off))
524 klen = rec_key_length(&p->u);
525 dlen = rec_data_length(&p->u);
526 extra = rec_extra_padding(&p->u);
528 len = sizeof(p->u) + klen + dlen + extra;
529 if (off + len > tdb->map_size) {
530 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
531 "tdb_check: used overlength %llu"
533 (long long)len, (long long)off);
537 if (len < sizeof(p->f)) {
538 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
539 "tdb_check: too short record %llu at"
541 (long long)len, (long long)off);
547 /* We must have found recovery area if there was one. */
548 if (recovery != 0 && !found_recovery) {
549 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
550 "tdb_check: expected a recovery area at %zu\n",
558 int tdb_check(struct tdb_context *tdb,
559 int (*check)(TDB_DATA key, TDB_DATA data, void *private_data),
562 tdb_off_t *free = NULL, *used = NULL, flist, recovery;
563 size_t num_free = 0, num_used = 0, num_found = 0, num_flists = 0;
565 if (tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, false) != 0)
568 if (tdb_lock_expand(tdb, F_RDLCK) != 0) {
569 tdb_allrecord_unlock(tdb, F_RDLCK);
573 if (!check_header(tdb, &recovery))
576 /* First we do a linear scan, checking all records. */
577 if (!check_linear(tdb, &used, &num_used, &free, &num_free, recovery))
580 for (flist = first_flist(tdb); flist; flist = next_flist(tdb, flist)) {
581 if (flist == TDB_OFF_ERR)
583 if (!check_free_list(tdb, flist, num_flists, free, num_free,
589 /* FIXME: Check key uniqueness? */
590 if (!check_hash(tdb, used, num_used, num_flists, check, private_data))
593 if (num_found != num_free) {
594 tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
595 "tdb_check: Not all entries are in free table\n");
599 tdb_allrecord_unlock(tdb, F_RDLCK);
600 tdb_unlock_expand(tdb, F_RDLCK);
604 tdb_allrecord_unlock(tdb, F_RDLCK);
605 tdb_unlock_expand(tdb, F_RDLCK);