*/
#include "tdb1_private.h"
-
-TDB_DATA tdb1_null;
+#include <assert.h>
/*
non-blocking increment of the tdb sequence number if the tdb has been opened using
the TDB_SEQNUM flag
*/
-void tdb1_increment_seqnum_nonblock(struct tdb1_context *tdb)
+void tdb1_increment_seqnum_nonblock(struct tdb_context *tdb)
{
tdb1_off_t seqnum=0;
increment the tdb sequence number if the tdb has been opened using
the TDB_SEQNUM flag
*/
-static void tdb1_increment_seqnum(struct tdb1_context *tdb)
+static void tdb1_increment_seqnum(struct tdb_context *tdb)
{
if (!(tdb->flags & TDB_SEQNUM)) {
return;
tdb1_nest_unlock(tdb, TDB1_SEQNUM_OFS, F_WRLCK);
}
-static int tdb1_key_compare(TDB_DATA key, TDB_DATA data, void *private_data)
+static enum TDB_ERROR tdb1_key_compare(TDB_DATA key, TDB_DATA data,
+ void *matches_)
{
- return memcmp(data.dptr, key.dptr, data.dsize);
+ bool *matches = matches_;
+ *matches = (memcmp(data.dptr, key.dptr, data.dsize) == 0);
+ return TDB_SUCCESS;
}
-/* Returns 0 on fail. On success, return offset of record, and fills
- in rec */
-static tdb1_off_t tdb1_find(struct tdb1_context *tdb, TDB_DATA key, uint32_t hash,
+/* Returns 0 on fail; last_error will be TDB_ERR_NOEXIST if it simply
+ * wasn't there, otherwise a real error.
+ * On success, return offset of record, and fills in rec */
+static tdb1_off_t tdb1_find(struct tdb_context *tdb, TDB_DATA key, uint32_t hash,
struct tdb1_record *r)
{
tdb1_off_t rec_ptr;
if (tdb1_rec_read(tdb, rec_ptr, r) == -1)
return 0;
- if (!TDB1_DEAD(r) && hash==r->full_hash
- && key.dsize==r->key_len
- && tdb1_parse_data(tdb, key, rec_ptr + sizeof(*r),
- r->key_len, tdb1_key_compare,
- NULL) == 0) {
- return rec_ptr;
+ tdb->stats.compares++;
+ if (TDB1_DEAD(r)) {
+ tdb->stats.compare_wrong_bucket++;
+ } else if (key.dsize != r->key_len) {
+ tdb->stats.compare_wrong_keylen++;
+ } else if (hash != r->full_hash) {
+ tdb->stats.compare_wrong_rechash++;
+ } else {
+ enum TDB_ERROR ecode;
+ bool matches;
+ ecode = tdb1_parse_data(tdb, key, rec_ptr + sizeof(*r),
+ r->key_len, tdb1_key_compare,
+ &matches);
+
+ if (ecode != TDB_SUCCESS) {
+ tdb->last_error = ecode;
+ return 0;
+ }
+
+ if (!matches) {
+ tdb->stats.compare_wrong_keycmp++;
+ } else {
+ return rec_ptr;
+ }
}
/* detect tight infinite loop */
if (rec_ptr == r->next) {
}
/* As tdb1_find, but if you succeed, keep the lock */
-tdb1_off_t tdb1_find_lock_hash(struct tdb1_context *tdb, TDB_DATA key, uint32_t hash, int locktype,
+tdb1_off_t tdb1_find_lock_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash, int locktype,
struct tdb1_record *rec)
{
uint32_t rec_ptr;
return rec_ptr;
}
-static TDB_DATA _tdb1_fetch(struct tdb1_context *tdb, TDB_DATA key);
+static TDB_DATA _tdb1_fetch(struct tdb_context *tdb, TDB_DATA key);
/* update an entry in place - this only works if the new data size
is <= the old data size and the key exists.
on failure return -1.
*/
-static int tdb1_update_hash(struct tdb1_context *tdb, TDB_DATA key, uint32_t hash, TDB_DATA dbuf)
+static int tdb1_update_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash, TDB_DATA dbuf)
{
struct tdb1_record rec;
tdb1_off_t rec_ptr;
return -1;
}
- if (tdb->methods->tdb1_write(tdb, rec_ptr + sizeof(rec) + rec.key_len,
+ if (tdb->tdb1.io->tdb1_write(tdb, rec_ptr + sizeof(rec) + rec.key_len,
dbuf.dptr, dbuf.dsize) == -1)
return -1;
* then the TDB_DATA will have zero length but
* a non-zero pointer
*/
-static TDB_DATA _tdb1_fetch(struct tdb1_context *tdb, TDB_DATA key)
+static TDB_DATA _tdb1_fetch(struct tdb_context *tdb, TDB_DATA key)
{
tdb1_off_t rec_ptr;
struct tdb1_record rec;
uint32_t hash;
/* find which hash bucket it is in */
- hash = tdb->hash_fn(&key);
- if (!(rec_ptr = tdb1_find_lock_hash(tdb,key,hash,F_RDLCK,&rec)))
- return tdb1_null;
+ hash = tdb_hash(tdb, key.dptr, key.dsize);
+ if (!(rec_ptr = tdb1_find_lock_hash(tdb,key,hash,F_RDLCK,&rec))) {
+ ret.dptr = NULL;
+ ret.dsize = 0;
+ return ret;
+ }
ret.dptr = tdb1_alloc_read(tdb, rec_ptr + sizeof(rec) + rec.key_len,
rec.data_len);
return ret;
}
-TDB_DATA tdb1_fetch(struct tdb1_context *tdb, TDB_DATA key)
+enum TDB_ERROR tdb1_fetch(struct tdb_context *tdb, TDB_DATA key, TDB_DATA *data)
{
- TDB_DATA ret = _tdb1_fetch(tdb, key);
-
- return ret;
+ *data = _tdb1_fetch(tdb, key);
+ if (data->dptr == NULL)
+ return tdb->last_error;
+ return TDB_SUCCESS;
}
-/*
- * Find an entry in the database and hand the record's data to a parsing
- * function. The parsing function is executed under the chain read lock, so it
- * should be fast and should not block on other syscalls.
- *
- * DON'T CALL OTHER TDB CALLS FROM THE PARSER, THIS MIGHT LEAD TO SEGFAULTS.
- *
- * For mmapped tdb's that do not have a transaction open it points the parsing
- * function directly at the mmap area, it avoids the malloc/memcpy in this
- * case. If a transaction is open or no mmap is available, it has to do
- * malloc/read/parse/free.
- *
- * This is interesting for all readers of potentially large data structures in
- * the tdb records, ldb indexes being one example.
- *
- * Return -1 if the record was not found.
- */
-
-int tdb1_parse_record(struct tdb1_context *tdb, TDB_DATA key,
- int (*parser)(TDB_DATA key, TDB_DATA data,
- void *private_data),
- void *private_data)
+enum TDB_ERROR tdb1_parse_record(struct tdb_context *tdb, TDB_DATA key,
+ enum TDB_ERROR (*parser)(TDB_DATA key,
+ TDB_DATA data,
+ void *private_data),
+ void *private_data)
{
tdb1_off_t rec_ptr;
struct tdb1_record rec;
- int ret;
+ enum TDB_ERROR ret;
uint32_t hash;
/* find which hash bucket it is in */
- hash = tdb->hash_fn(&key);
+ hash = tdb_hash(tdb, key.dptr, key.dsize);
if (!(rec_ptr = tdb1_find_lock_hash(tdb,key,hash,F_RDLCK,&rec))) {
- /* record not found */
- tdb->last_error = TDB_ERR_NOEXIST;
- return -1;
+ return tdb->last_error;
}
ret = tdb1_parse_data(tdb, key, rec_ptr + sizeof(rec) + rec.key_len,
this doesn't match the conventions in the rest of this module, but is
compatible with gdbm
*/
-static int tdb1_exists_hash(struct tdb1_context *tdb, TDB_DATA key, uint32_t hash)
+static int tdb1_exists_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash)
{
struct tdb1_record rec;
return 1;
}
-int tdb1_exists(struct tdb1_context *tdb, TDB_DATA key)
+int tdb1_exists(struct tdb_context *tdb, TDB_DATA key)
{
- uint32_t hash = tdb->hash_fn(&key);
+ uint32_t hash = tdb_hash(tdb, key.dptr, key.dsize);
int ret;
+ assert(tdb->flags & TDB_VERSION1);
ret = tdb1_exists_hash(tdb, key, hash);
return ret;
}
/* actually delete an entry in the database given the offset */
-int tdb1_do_delete(struct tdb1_context *tdb, tdb1_off_t rec_ptr, struct tdb1_record *rec)
+int tdb1_do_delete(struct tdb_context *tdb, tdb1_off_t rec_ptr, struct tdb1_record *rec)
{
tdb1_off_t last_ptr, i;
struct tdb1_record lastrec;
- if (tdb->read_only || tdb->traverse_read) return -1;
+ if ((tdb->flags & TDB_RDONLY) || tdb->tdb1.traverse_read) return -1;
- if (((tdb->traverse_write != 0) && (!TDB1_DEAD(rec))) ||
+ if (((tdb->tdb1.traverse_write != 0) && (!TDB1_DEAD(rec))) ||
tdb1_write_lock_record(tdb, rec_ptr) == -1) {
/* Someone traversing here: mark it as dead */
rec->magic = TDB1_DEAD_MAGIC;
return 0;
}
-static int tdb1_count_dead(struct tdb1_context *tdb, uint32_t hash)
+static int tdb1_count_dead(struct tdb_context *tdb, uint32_t hash)
{
int res = 0;
tdb1_off_t rec_ptr;
/*
* Purge all DEAD records from a hash chain
*/
-static int tdb1_purge_dead(struct tdb1_context *tdb, uint32_t hash)
+static int tdb1_purge_dead(struct tdb_context *tdb, uint32_t hash)
{
int res = -1;
struct tdb1_record rec;
}
/* delete an entry in the database given a key */
-static int tdb1_delete_hash(struct tdb1_context *tdb, TDB_DATA key, uint32_t hash)
+static int tdb1_delete_hash(struct tdb_context *tdb, TDB_DATA key, uint32_t hash)
{
tdb1_off_t rec_ptr;
struct tdb1_record rec;
int ret;
- if (tdb->max_dead_records != 0) {
+ if (tdb->tdb1.max_dead_records != 0) {
/*
* Allow for some dead records per hash chain, mainly for
if (tdb1_lock(tdb, TDB1_BUCKET(hash), F_WRLCK) == -1)
return -1;
- if (tdb1_count_dead(tdb, hash) >= tdb->max_dead_records) {
+ if (tdb1_count_dead(tdb, hash) >= tdb->tdb1.max_dead_records) {
/*
* Don't let the per-chain freelist grow too large,
* delete all existing dead records
return ret;
}
-int tdb1_delete(struct tdb1_context *tdb, TDB_DATA key)
+int tdb1_delete(struct tdb_context *tdb, TDB_DATA key)
{
- uint32_t hash = tdb->hash_fn(&key);
+ uint32_t hash = tdb_hash(tdb, key.dptr, key.dsize);
int ret;
+ assert(tdb->flags & TDB_VERSION1);
ret = tdb1_delete_hash(tdb, key, hash);
return ret;
}
/*
* See if we have a dead record around with enough space
*/
-static tdb1_off_t tdb1_find_dead(struct tdb1_context *tdb, uint32_t hash,
+static tdb1_off_t tdb1_find_dead(struct tdb_context *tdb, uint32_t hash,
struct tdb1_record *r, tdb1_len_t length)
{
tdb1_off_t rec_ptr;
return 0;
}
-static int _tdb1_store(struct tdb1_context *tdb, TDB_DATA key,
+static int _tdb1_store(struct tdb_context *tdb, TDB_DATA key,
TDB_DATA dbuf, int flag, uint32_t hash)
{
struct tdb1_record rec;
tdb->last_error = TDB_ERR_EXISTS;
goto fail;
}
+ if (tdb->last_error != TDB_ERR_NOEXIST) {
+ goto fail;
+ }
} else {
/* first try in-place update, on modify or replace. */
if (tdb1_update_hash(tdb, key, hash, dbuf) == 0) {
goto done;
}
- if (tdb->last_error == TDB_ERR_NOEXIST &&
- flag == TDB_MODIFY) {
- /* if the record doesn't exist and we are in TDB1_MODIFY mode then
- we should fail the store */
- goto fail;
+ if (tdb->last_error != TDB_SUCCESS) {
+ if (tdb->last_error != TDB_ERR_NOEXIST) {
+ goto fail;
+ }
+ if (flag == TDB_MODIFY) {
+ /* if the record doesn't exist and we are in TDB1_MODIFY mode then
+ we should fail the store */
+ goto fail;
+ }
}
}
/* reset the error code potentially set by the tdb1_update() */
fails and we are left with a dead spot in the tdb. */
if (!(p = (char *)malloc(key.dsize + dbuf.dsize))) {
- tdb->last_error = TDB_ERR_OOM;
+ tdb->last_error = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
+ "tdb1_store: out of memory"
+ " allocating copy");
goto fail;
}
if (dbuf.dsize)
memcpy(p+key.dsize, dbuf.dptr, dbuf.dsize);
- if (tdb->max_dead_records != 0) {
+ if (tdb->tdb1.max_dead_records != 0) {
/*
* Allow for some dead records per hash chain, look if we can
* find one that can hold the new record. We need enough space
rec.full_hash = hash;
rec.magic = TDB1_MAGIC;
if (tdb1_rec_write(tdb, rec_ptr, &rec) == -1
- || tdb->methods->tdb1_write(
+ || tdb->tdb1.io->tdb1_write(
tdb, rec_ptr + sizeof(rec),
p, key.dsize + dbuf.dsize) == -1) {
goto fail;
goto fail;
}
- if ((tdb->max_dead_records != 0)
+ if ((tdb->tdb1.max_dead_records != 0)
&& (tdb1_purge_dead(tdb, hash) == -1)) {
tdb1_unlock(tdb, -1, F_WRLCK);
goto fail;
/* write out and point the top of the hash chain at it */
if (tdb1_rec_write(tdb, rec_ptr, &rec) == -1
- || tdb->methods->tdb1_write(tdb, rec_ptr+sizeof(rec), p, key.dsize+dbuf.dsize)==-1
+ || tdb->tdb1.io->tdb1_write(tdb, rec_ptr+sizeof(rec), p, key.dsize+dbuf.dsize)==-1
|| tdb1_ofs_write(tdb, TDB1_HASH_TOP(hash), &rec_ptr) == -1) {
/* Need to tdb1_unallocate() here */
goto fail;
return 0 on success, -1 on failure
*/
-int tdb1_store(struct tdb1_context *tdb, TDB_DATA key, TDB_DATA dbuf, int flag)
+int tdb1_store(struct tdb_context *tdb, TDB_DATA key, TDB_DATA dbuf, int flag)
{
uint32_t hash;
int ret;
- if (tdb->read_only || tdb->traverse_read) {
- tdb->last_error = TDB_ERR_RDONLY;
+ assert(tdb->flags & TDB_VERSION1);
+
+ if ((tdb->flags & TDB_RDONLY) || tdb->tdb1.traverse_read) {
+ tdb->last_error = tdb_logerr(tdb, TDB_ERR_RDONLY,
+ TDB_LOG_USE_ERROR,
+ "tdb_store: read-only tdb");
return -1;
}
/* find which hash bucket it is in */
- hash = tdb->hash_fn(&key);
+ hash = tdb_hash(tdb, key.dptr, key.dsize);
if (tdb1_lock(tdb, TDB1_BUCKET(hash), F_WRLCK) == -1)
return -1;
}
/* Append to an entry. Create if not exist. */
-int tdb1_append(struct tdb1_context *tdb, TDB_DATA key, TDB_DATA new_dbuf)
+int tdb1_append(struct tdb_context *tdb, TDB_DATA key, TDB_DATA new_dbuf)
{
uint32_t hash;
TDB_DATA dbuf;
int ret = -1;
+ assert(tdb->flags & TDB_VERSION1);
+
/* find which hash bucket it is in */
- hash = tdb->hash_fn(&key);
+ hash = tdb_hash(tdb, key.dptr, key.dsize);
if (tdb1_lock(tdb, TDB1_BUCKET(hash), F_WRLCK) == -1)
return -1;
The aim of this sequence number is to allow for a very lightweight
test of a possible tdb change.
*/
-int tdb1_get_seqnum(struct tdb1_context *tdb)
+int tdb1_get_seqnum(struct tdb_context *tdb)
{
tdb1_off_t seqnum=0;
return seqnum;
}
-int tdb1_hash_size(struct tdb1_context *tdb)
-{
- return tdb->header.hash_size;
-}
-
/*
add a region of the file to the freelist. Length is the size of the region in bytes,
which includes the free list header that needs to be added
*/
-static int tdb1_free_region(struct tdb1_context *tdb, tdb1_off_t offset, ssize_t length)
+static int tdb1_free_region(struct tdb_context *tdb, tdb1_off_t offset, ssize_t length)
{
struct tdb1_record rec;
if (length <= sizeof(rec)) {
This code carefully steps around the recovery area, leaving it alone
*/
-int tdb1_wipe_all(struct tdb1_context *tdb)
+int tdb1_wipe_all(struct tdb_context *tdb)
{
int i;
tdb1_off_t offset = 0;
tdb1_off_t recovery_head;
tdb1_len_t recovery_size = 0;
- if (tdb1_lockall(tdb) != 0) {
+ if (tdb_lockall(tdb) != TDB_SUCCESS) {
return -1;
}
if (recovery_head != 0) {
struct tdb1_record rec;
- if (tdb->methods->tdb1_read(tdb, recovery_head, &rec, sizeof(rec), TDB1_DOCONV()) == -1) {
+ if (tdb->tdb1.io->tdb1_read(tdb, recovery_head, &rec, sizeof(rec), TDB1_DOCONV()) == -1) {
tdb_logerr(tdb, tdb->last_error, TDB_LOG_ERROR,
"tdb1_wipe_all: failed to read recovery record");
return -1;
}
/* wipe the hashes */
- for (i=0;i<tdb->header.hash_size;i++) {
+ for (i=0;i<tdb->tdb1.header.hash_size;i++) {
if (tdb1_ofs_write(tdb, TDB1_HASH_TOP(i), &offset) == -1) {
tdb_logerr(tdb, tdb->last_error, TDB_LOG_ERROR,
"tdb1_wipe_all: failed to write hash %d", i);
for the recovery area */
if (recovery_size == 0) {
/* the simple case - the whole file can be used as a freelist */
- data_len = (tdb->file->map_size - TDB1_DATA_START(tdb->header.hash_size));
- if (tdb1_free_region(tdb, TDB1_DATA_START(tdb->header.hash_size), data_len) != 0) {
+ data_len = (tdb->file->map_size - TDB1_DATA_START(tdb->tdb1.header.hash_size));
+ if (tdb1_free_region(tdb, TDB1_DATA_START(tdb->tdb1.header.hash_size), data_len) != 0) {
goto failed;
}
} else {
move the recovery area or we risk subtle data
corruption
*/
- data_len = (recovery_head - TDB1_DATA_START(tdb->header.hash_size));
- if (tdb1_free_region(tdb, TDB1_DATA_START(tdb->header.hash_size), data_len) != 0) {
+ data_len = (recovery_head - TDB1_DATA_START(tdb->tdb1.header.hash_size));
+ if (tdb1_free_region(tdb, TDB1_DATA_START(tdb->tdb1.header.hash_size), data_len) != 0) {
goto failed;
}
/* and the 2nd free list entry after the recovery area - if any */
}
}
- if (tdb1_unlockall(tdb) != 0) {
- tdb_logerr(tdb, tdb->last_error, TDB_LOG_ERROR,
- "tdb1_wipe_all: failed to unlock");
- goto failed;
- }
-
+ tdb1_increment_seqnum_nonblock(tdb);
+ tdb_unlockall(tdb);
return 0;
failed:
- tdb1_unlockall(tdb);
+ tdb_unlockall(tdb);
return -1;
}
-struct traverse_state {
- enum TDB_ERROR error;
- struct tdb1_context *dest_db;
-};
-
-/*
- traverse function for repacking
- */
-static int repack_traverse(struct tdb1_context *tdb, TDB_DATA key, TDB_DATA data, void *private_data)
-{
- struct traverse_state *state = (struct traverse_state *)private_data;
- if (tdb1_store(state->dest_db, key, data, TDB_INSERT) != 0) {
- state->error = state->dest_db->last_error;
- return -1;
- }
- return 0;
-}
-
-/*
- repack a tdb
- */
-int tdb1_repack(struct tdb1_context *tdb)
-{
- struct tdb1_context *tmp_db;
- struct traverse_state state;
-
- if (tdb1_transaction_start(tdb) != 0) {
- tdb_logerr(tdb, tdb->last_error, TDB_LOG_ERROR,
- __location__ " Failed to start transaction");
- return -1;
- }
-
- tmp_db = tdb1_open("tmpdb", tdb1_hash_size(tdb), TDB_INTERNAL, O_RDWR|O_CREAT, 0);
- if (tmp_db == NULL) {
- tdb->last_error = tdb_logerr(tdb, TDB_ERR_OOM, TDB_LOG_ERROR,
- __location__ " Failed to create tmp_db");
- tdb1_transaction_cancel(tdb);
- return -1;
- }
-
- state.error = TDB_SUCCESS;
- state.dest_db = tmp_db;
-
- if (tdb1_traverse_read(tdb, repack_traverse, &state) == -1) {
- tdb_logerr(tdb, tdb->last_error, TDB_LOG_ERROR,
- __location__ " Failed to traverse copying out");
- tdb1_transaction_cancel(tdb);
- tdb1_close(tmp_db);
- return -1;
- }
-
- if (state.error != TDB_SUCCESS) {
- tdb->last_error = tdb_logerr(tdb, state.error, TDB_LOG_ERROR,
- __location__ " Error during traversal");
- tdb1_transaction_cancel(tdb);
- tdb1_close(tmp_db);
- return -1;
- }
-
- if (tdb1_wipe_all(tdb) != 0) {
- tdb_logerr(tdb, tdb->last_error, TDB_LOG_ERROR,
- __location__ " Failed to wipe database\n");
- tdb1_transaction_cancel(tdb);
- tdb1_close(tmp_db);
- return -1;
- }
-
- state.error = TDB_SUCCESS;
- state.dest_db = tdb;
-
- if (tdb1_traverse_read(tmp_db, repack_traverse, &state) == -1) {
- tdb_logerr(tdb, tdb->last_error, TDB_LOG_ERROR,
- __location__ " Failed to traverse copying back");
- tdb1_transaction_cancel(tdb);
- tdb1_close(tmp_db);
- return -1;
- }
-
- if (state.error) {
- tdb->last_error = tdb_logerr(tdb, state.error, TDB_LOG_ERROR,
- __location__ " Error during second traversal");
- tdb1_transaction_cancel(tdb);
- tdb1_close(tmp_db);
- return -1;
- }
-
- tdb1_close(tmp_db);
-
- if (tdb1_transaction_commit(tdb) != 0) {
- tdb_logerr(tdb, tdb->last_error, TDB_LOG_ERROR,
- __location__ " Failed to commit");
- return -1;
- }
-
- return 0;
-}
-
/* Even on files, we can get partial writes due to signals. */
bool tdb1_write_all(int fd, const void *buf, size_t count)
{