return true;
}
-static bool check_header(struct tdb_context *tdb)
+static bool check_header(struct tdb_context *tdb, tdb_off_t *recovery)
{
uint64_t hash_test;
struct tdb_header hdr;
return false;
}
+ *recovery = hdr.recovery;
+ if (*recovery) {
+ if (*recovery < sizeof(hdr) || *recovery > tdb->map_size) {
+ tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
+ "tdb_check: invalid recovery offset %zu\n",
+ (size_t)*recovery);
+ return false;
+ }
+ }
+
/* Don't check reserved: they *can* be used later. */
return true;
}
unsigned hprefix_bits,
tdb_off_t used[],
size_t num_used,
- size_t *num_found);
+ size_t *num_found,
+ int (*check)(TDB_DATA, TDB_DATA, void *),
+ void *private_data);
static bool check_hash_record(struct tdb_context *tdb,
tdb_off_t off,
unsigned hprefix_bits,
tdb_off_t used[],
size_t num_used,
- size_t *num_found)
+ size_t *num_found,
+ int (*check)(TDB_DATA, TDB_DATA, void *),
+ void *private_data)
{
struct tdb_used_record rec;
return check_hash_tree(tdb, off,
TDB_SUBLEVEL_HASH_BITS-TDB_HASH_GROUP_BITS,
hprefix, hprefix_bits,
- used, num_used, num_found);
+ used, num_used, num_found, check, private_data);
}
static int off_cmp(const tdb_off_t *a, const tdb_off_t *b)
unsigned hprefix_bits,
tdb_off_t used[],
size_t num_used,
- size_t *num_found)
+ size_t *num_found,
+ int (*check)(TDB_DATA, TDB_DATA, void *),
+ void *private_data)
{
unsigned int g, b;
const tdb_off_t *hash;
hprefix_bits
+ group_bits
+ TDB_HASH_GROUP_BITS,
- used, num_used, num_found))
+ used, num_used, num_found,
+ check, private_data))
goto fail;
continue;
}
goto fail;
/* Bottom bits must match header. */
- if ((h & ((1 << 5)-1)) != rec_hash(&rec)) {
+ if ((h & ((1 << 11)-1)) != rec_hash(&rec)) {
tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
"tdb_check: Bad hash magic at"
" offset %llu (0x%llx vs 0x%llx)\n",
(long long)rec_hash(&rec));
goto fail;
}
+
+ if (check) {
+ TDB_DATA key, data;
+ key.dsize = rec_key_length(&rec);
+ data.dsize = rec_data_length(&rec);
+ key.dptr = (void *)tdb_access_read(tdb,
+ off + sizeof(rec),
+ key.dsize + data.dsize,
+ false);
+ if (!key.dptr)
+ goto fail;
+ data.dptr = key.dptr + key.dsize;
+ if (check(key, data, private_data) != 0)
+ goto fail;
+ tdb_access_release(tdb, key.dptr);
+ }
}
}
tdb_access_release(tdb, hash);
static bool check_hash(struct tdb_context *tdb,
tdb_off_t used[],
- size_t num_used)
+ size_t num_used, size_t num_flists,
+ int (*check)(TDB_DATA, TDB_DATA, void *),
+ void *private_data)
{
- size_t num_found = 0;
+ /* Free lists also show up as used. */
+ size_t num_found = num_flists;
if (!check_hash_tree(tdb, offsetof(struct tdb_header, hashtable),
TDB_TOPLEVEL_HASH_BITS-TDB_HASH_GROUP_BITS,
- 0, 0, used, num_used, &num_found))
+ 0, 0, used, num_used, &num_found,
+ check, private_data))
return false;
if (num_found != num_used) {
static bool check_free(struct tdb_context *tdb,
tdb_off_t off,
const struct tdb_free_record *frec,
- tdb_off_t prev,
- tdb_off_t zone_off, unsigned int bucket)
+ tdb_off_t prev, tdb_off_t flist_off, unsigned int bucket)
{
if (frec_magic(frec) != TDB_FREE_MAGIC) {
tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
(long long)off, (long long)frec->magic_and_meta);
return false;
}
+ if (frec_flist(frec) != flist_off) {
+ tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
+ "tdb_check: offset %llu bad freelist 0x%llx\n",
+ (long long)off, (long long)frec_flist(frec));
+ return false;
+ }
+
if (tdb->methods->oob(tdb, off
+ frec->data_len+sizeof(struct tdb_used_record),
false))
return false;
- if (off < zone_off || off >= zone_off + (1ULL<<frec_zone_bits(frec))) {
- tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
- "tdb_check: offset %llu outside zone %llu-%llu\n",
- (long long)off,
- (long long)zone_off,
- (long long)zone_off + (1ULL<<frec_zone_bits(frec)));
- return false;
- }
- if (size_to_bucket(frec_zone_bits(frec), frec->data_len) != bucket) {
+ if (size_to_bucket(frec->data_len) != bucket) {
tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
"tdb_check: offset %llu in wrong bucket %u vs %u\n",
(long long)off,
- bucket,
- size_to_bucket(frec_zone_bits(frec), frec->data_len));
+ bucket, size_to_bucket(frec->data_len));
return false;
}
if (prev != frec->prev) {
return true;
}
-static tdb_len_t check_free_list(struct tdb_context *tdb,
- tdb_off_t zone_off,
- tdb_off_t free[],
- size_t num_free,
- size_t *num_found)
+static bool check_free_list(struct tdb_context *tdb,
+ tdb_off_t flist_off,
+ tdb_off_t free[],
+ size_t num_free,
+ size_t *num_found)
{
- struct free_zone_header zhdr;
+ struct tdb_freelist flist;
tdb_off_t h;
unsigned int i;
- if (tdb_read_convert(tdb, zone_off, &zhdr, sizeof(zhdr)) == -1)
- return TDB_OFF_ERR;
+ if (tdb_read_convert(tdb, flist_off, &flist, sizeof(flist)) == -1)
+ return false;
+
+ if (rec_magic(&flist.hdr) != TDB_MAGIC
+ || rec_key_length(&flist.hdr) != 0
+ || rec_data_length(&flist.hdr) != sizeof(flist) - sizeof(flist.hdr)
+ || rec_hash(&flist.hdr) != 1) {
+ tdb->log(tdb, TDB_DEBUG_ERROR,
+ tdb->log_priv,
+ "tdb_check: Invalid header on free list\n");
+ return false;
+ }
- for (i = 0; i <= BUCKETS_FOR_ZONE(zhdr.zone_bits); i++) {
+ for (i = 0; i < TDB_FREE_BUCKETS; i++) {
tdb_off_t off, prev = 0, *p;
struct tdb_free_record f;
- h = bucket_off(zone_off, i);
+ h = bucket_off(flist_off, i);
for (off = tdb_read_off(tdb, h); off; off = f.next) {
if (off == TDB_OFF_ERR)
- return TDB_OFF_ERR;
+ return false;
if (tdb_read_convert(tdb, off, &f, sizeof(f)))
- return TDB_OFF_ERR;
- if (!check_free(tdb, off, &f, prev, zone_off, i))
- return TDB_OFF_ERR;
+ return false;
+ if (!check_free(tdb, off, &f, prev, flist_off, i))
+ return false;
/* FIXME: Check hash bits */
p = asearch(&off, free, num_free, off_cmp);
"tdb_check: Invalid offset"
" %llu in free table\n",
(long long)off);
- return TDB_OFF_ERR;
+ return false;
}
/* Mark it invalid. */
*p ^= 1;
prev = off;
}
}
- return 1ULL << zhdr.zone_bits;
+ return true;
}
-static tdb_off_t check_zone(struct tdb_context *tdb, tdb_off_t zone_off,
- tdb_off_t **used, size_t *num_used,
- tdb_off_t **free, size_t *num_free,
- unsigned int *max_zone_bits)
+/* Slow, but should be very rare. */
+size_t dead_space(struct tdb_context *tdb, tdb_off_t off)
{
- struct free_zone_header zhdr;
- tdb_off_t off, hdrlen;
- tdb_len_t len;
-
- if (tdb_read_convert(tdb, zone_off, &zhdr, sizeof(zhdr)) == -1)
- return TDB_OFF_ERR;
-
- if (zhdr.zone_bits < INITIAL_ZONE_BITS) {
- tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
- "check: bad zone_bits %llu at zone %llu\n",
- (long long)zhdr.zone_bits, (long long)zone_off);
- return TDB_OFF_ERR;
- }
-
- /* Zone bits can only increase... */
- if (zhdr.zone_bits > *max_zone_bits)
- *max_zone_bits = zhdr.zone_bits;
- else if (zhdr.zone_bits < *max_zone_bits) {
- tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
- "check: small zone_bits %llu at zone %llu\n",
- (long long)zhdr.zone_bits, (long long)zone_off);
- return TDB_OFF_ERR;
+ size_t len;
+
+ for (len = 0; off + len < tdb->map_size; len++) {
+ char c;
+ if (tdb->methods->read(tdb, off, &c, 1))
+ return 0;
+ if (c != 0 && c != 0x43)
+ break;
}
+ return len;
+}
- /* Zone must be within file! */
- if (tdb->methods->oob(tdb, zone_off + (1ULL << zhdr.zone_bits), false))
- return TDB_OFF_ERR;
+static bool check_linear(struct tdb_context *tdb,
+ tdb_off_t **used, size_t *num_used,
+ tdb_off_t **free, size_t *num_free,
+ tdb_off_t recovery)
+{
+ tdb_off_t off;
+ tdb_len_t len;
+ bool found_recovery = false;
- hdrlen = sizeof(zhdr)
- + (BUCKETS_FOR_ZONE(zhdr.zone_bits) + 1) * sizeof(tdb_off_t);
- for (off = zone_off + hdrlen;
- off < zone_off + (1ULL << zhdr.zone_bits);
- off += len) {
+ for (off = sizeof(struct tdb_header); off < tdb->map_size; off += len) {
union {
struct tdb_used_record u;
struct tdb_free_record f;
+ struct tdb_recovery_record r;
} pad, *p;
p = tdb_get(tdb, off, &pad, sizeof(pad));
if (!p)
- return TDB_OFF_ERR;
- if (frec_magic(&p->f) == TDB_FREE_MAGIC
- || frec_magic(&p->f) == TDB_COALESCING_MAGIC) {
- if (frec_zone_bits(&p->f) != zhdr.zone_bits) {
+ return false;
+
+ /* If we crash after ftruncate, we can get zeroes or fill. */
+ if (p->r.magic == TDB_RECOVERY_INVALID_MAGIC
+ || p->r.magic == 0x4343434343434343ULL) {
+ if (recovery == off) {
+ found_recovery = true;
+ len = sizeof(p->r) + p->r.max_len;
+ } else {
+ len = dead_space(tdb, off);
+ if (len < sizeof(p->r)) {
+ tdb->log(tdb, TDB_DEBUG_ERROR,
+ tdb->log_priv,
+ "tdb_check: invalid dead space"
+ " at %zu\n", (size_t)off);
+ return false;
+ }
+
+ tdb->log(tdb, TDB_DEBUG_WARNING, tdb->log_priv,
+ "Dead space at %zu-%zu (of %zu)\n",
+ (size_t)off, (size_t)(off + len),
+ (size_t)tdb->map_size);
+ }
+ } else if (p->r.magic == TDB_RECOVERY_MAGIC) {
+ if (recovery != off) {
tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
- "tdb_check: Bad free zone bits %u"
- " at offset %llu\n",
- frec_zone_bits(&p->f),
- (long long)off);
- return TDB_OFF_ERR;
+ "tdb_check: unexpected recovery"
+ " record at offset %zu\n",
+ (size_t)off);
+ return false;
}
+ found_recovery = true;
+ len = sizeof(p->r) + p->r.max_len;
+ } else if (frec_magic(&p->f) == TDB_FREE_MAGIC
+ || frec_magic(&p->f) == TDB_COALESCING_MAGIC) {
len = sizeof(p->u) + p->f.data_len;
- if (off + len > zone_off + (1ULL << zhdr.zone_bits)) {
+ if (off + len > tdb->map_size) {
tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
"tdb_check: free overlength %llu"
" at offset %llu\n",
(long long)len, (long long)off);
- return TDB_OFF_ERR;
+ return false;
}
/* This record is free! */
if (frec_magic(&p->f) == TDB_FREE_MAGIC
&& !append(free, num_free, off))
- return TDB_OFF_ERR;
+ return false;
} else {
uint64_t klen, dlen, extra;
" at offset %llu\n",
(long long)rec_magic(&p->u),
(long long)off);
- return TDB_OFF_ERR;
+ return false;
}
- if (rec_zone_bits(&p->u) != zhdr.zone_bits) {
- tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
- "tdb_check: Bad zone bits %u"
- " at offset %llu\n",
- rec_zone_bits(&p->u),
- (long long)off);
- return TDB_OFF_ERR;
- }
-
if (!append(used, num_used, off))
- return TDB_OFF_ERR;
+ return false;
klen = rec_key_length(&p->u);
dlen = rec_data_length(&p->u);
extra = rec_extra_padding(&p->u);
len = sizeof(p->u) + klen + dlen + extra;
- if (off + len > zone_off + (1ULL << zhdr.zone_bits)) {
+ if (off + len > tdb->map_size) {
tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
"tdb_check: used overlength %llu"
" at offset %llu\n",
(long long)len, (long long)off);
- return TDB_OFF_ERR;
+ return false;
}
if (len < sizeof(p->f)) {
"tdb_check: too short record %llu at"
" %llu\n",
(long long)len, (long long)off);
- return TDB_OFF_ERR;
+ return false;
}
}
}
- return 1ULL << zhdr.zone_bits;
+
+ /* We must have found recovery area if there was one. */
+ if (recovery != 0 && !found_recovery) {
+ tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
+ "tdb_check: expected a recovery area at %zu\n",
+ (size_t)recovery);
+ return false;
+ }
+
+ return true;
}
-/* FIXME: call check() function. */
int tdb_check(struct tdb_context *tdb,
int (*check)(TDB_DATA key, TDB_DATA data, void *private_data),
void *private_data)
{
- tdb_off_t *free = NULL, *used = NULL, off;
- tdb_len_t len;
- size_t num_free = 0, num_used = 0, num_found = 0;
- unsigned max_zone_bits = INITIAL_ZONE_BITS;
- uint8_t tailer;
+ tdb_off_t *free = NULL, *used = NULL, flist, recovery;
+ size_t num_free = 0, num_used = 0, num_found = 0, num_flists = 0;
if (tdb_allrecord_lock(tdb, F_RDLCK, TDB_LOCK_WAIT, false) != 0)
return -1;
return -1;
}
- if (!check_header(tdb))
+ if (!check_header(tdb, &recovery))
goto fail;
/* First we do a linear scan, checking all records. */
- for (off = sizeof(struct tdb_header);
- off < tdb->map_size - 1;
- off += len) {
- len = check_zone(tdb, off, &used, &num_used, &free, &num_free,
- &max_zone_bits);
- if (len == TDB_OFF_ERR)
- goto fail;
- }
-
- /* Check tailer. */
- if (tdb->methods->read(tdb, tdb->map_size - 1, &tailer, 1) == -1)
- goto fail;
- if (tailer != max_zone_bits) {
- tdb->ecode = TDB_ERR_CORRUPT;
- tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
- "tdb_check: Bad tailer value %u vs %u\n", tailer,
- max_zone_bits);
+ if (!check_linear(tdb, &used, &num_used, &free, &num_free, recovery))
goto fail;
+
+ for (flist = first_flist(tdb); flist; flist = next_flist(tdb, flist)) {
+ if (flist == TDB_OFF_ERR)
+ goto fail;
+ if (!check_free_list(tdb, flist, free, num_free, &num_found))
+ goto fail;
+ num_flists++;
}
/* FIXME: Check key uniqueness? */
- if (!check_hash(tdb, used, num_used))
+ if (!check_hash(tdb, used, num_used, num_flists, check, private_data))
goto fail;
- for (off = sizeof(struct tdb_header);
- off < tdb->map_size - 1;
- off += len) {
- len = check_free_list(tdb, off, free, num_free, &num_found);
- if (len == TDB_OFF_ERR)
- goto fail;
- }
if (num_found != num_free) {
tdb->log(tdb, TDB_DEBUG_ERROR, tdb->log_priv,
"tdb_check: Not all entries are in free table\n");
- return false;
+ return -1;
}
tdb_allrecord_unlock(tdb, F_RDLCK);