tdb.c

Linux下的一个关系数据库源代码

	TDB_DATA key, dbuf;
	struct list_struct rec;
	struct tdb_traverse_lock tl = { NULL, 0, 0 };
	int ret, count = 0;

	/* This was in the initializaton, above, but the IRIX compiler
	 * did not like it.  crh
	 */
	tl.next = tdb->travlocks.next;

	/* fcntl locks don't stack: beware traverse inside traverse */
	tdb->travlocks.next = &tl;

	/* tdb_next_lock places locks on the record returned, and its chain */
	while ((ret = tdb_next_lock(tdb, &tl, &rec)) > 0) {
		count++;
		/* now read the full record */
		key.dptr = tdb_alloc_read(tdb, tl.off + sizeof(rec), 
					  rec.key_len + rec.data_len);
		if (!key.dptr) {
			tdb_unlock(tdb, tl.hash, F_WRLCK);
			unlock_record(tdb, tl.off);
			tdb->travlocks.next = tl.next;
			return -1;
		}
		key.dsize = rec.key_len;
		dbuf.dptr = key.dptr + rec.key_len;
		dbuf.dsize = rec.data_len;

		/* Drop chain lock, call out */
		tdb_unlock(tdb, tl.hash, F_WRLCK);
		if (fn && fn(tdb, key, dbuf, state)) {
			/* They want us to terminate traversal */
			unlock_record(tdb, tl.off);
			tdb->travlocks.next = tl.next;
			free(key.dptr);
			return count;
		}
		free(key.dptr);
	}
	tdb->travlocks.next = tl.next;
	if (ret < 0)
		return -1;
	else
		return count;
}

/* find the first entry in the database and return its key */
TDB_DATA tdb_firstkey(TDB_CONTEXT *tdb)
{
	TDB_DATA key;
	struct list_struct rec;

	/* release any old lock */
	unlock_record(tdb, tdb->travlocks.off);
	tdb->travlocks.off = tdb->travlocks.hash = 0;

	if (tdb_next_lock(tdb, &tdb->travlocks, &rec) <= 0)
		return tdb_null;
	/* now read the key */
	key.dsize = rec.key_len;
	key.dptr =tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),key.dsize);
	tdb_unlock(tdb, BUCKET(tdb->travlocks.hash), F_WRLCK);
	return key;
}

/* find the next entry in the database, returning its key */
TDB_DATA tdb_nextkey(TDB_CONTEXT *tdb, TDB_DATA oldkey)
{
	u32 oldhash;
	TDB_DATA key = tdb_null;
	struct list_struct rec;
	char *k = NULL;

	/* Is locked key the old key?  If so, traverse will be reliable. */
	if (tdb->travlocks.off) {
		if (tdb_lock(tdb,tdb->travlocks.hash,F_WRLCK))
			return tdb_null;
		if (rec_read(tdb, tdb->travlocks.off, &rec) == -1
		    || !(k = tdb_alloc_read(tdb,tdb->travlocks.off+sizeof(rec),
					    rec.key_len))
		    || memcmp(k, oldkey.dptr, oldkey.dsize) != 0) {
			/* No, it wasn't: unlock it and start from scratch */
			unlock_record(tdb, tdb->travlocks.off);
			tdb_unlock(tdb, tdb->travlocks.hash, F_WRLCK);
			tdb->travlocks.off = 0;
		}

		if (k)
			free(k);
	}

	if (!tdb->travlocks.off) {
		/* No previous element: do normal find, and lock record */
		tdb->travlocks.off = tdb_find_lock(tdb, oldkey, F_WRLCK, &rec);
		if (!tdb->travlocks.off)
			return tdb_null;
		tdb->travlocks.hash = BUCKET(rec.full_hash);
		lock_record(tdb, tdb->travlocks.off);
	}
	oldhash = tdb->travlocks.hash;

	/* Grab next record: locks chain and returned record,
	   unlocks old record */
	if (tdb_next_lock(tdb, &tdb->travlocks, &rec) > 0) {
		key.dsize = rec.key_len;
		key.dptr = tdb_alloc_read(tdb, tdb->travlocks.off+sizeof(rec),
					  key.dsize);
		/* Unlock the chain of this new record */
		tdb_unlock(tdb, tdb->travlocks.hash, F_WRLCK);
	}
	/* Unlock the chain of old record */
	tdb_unlock(tdb, BUCKET(oldhash), F_WRLCK);
	return key;
}

/* delete an entry in the database given a key */
int tdb_delete(TDB_CONTEXT *tdb, TDB_DATA key)
{
	tdb_off rec_ptr;
	struct list_struct rec;
	int ret;

	if (!(rec_ptr = tdb_find_lock(tdb, key, F_WRLCK, &rec)))
		return -1;
	ret = do_delete(tdb, rec_ptr, &rec);
	tdb_unlock(tdb, BUCKET(rec.full_hash), F_WRLCK);
	return ret;
}

/* store an element in the database, replacing any existing element
   with the same key 

   return 0 on success, -1 on failure
*/
int tdb_store(TDB_CONTEXT *tdb, TDB_DATA key, TDB_DATA dbuf, int flag)
{
	struct list_struct rec;
	u32 hash;
	tdb_off rec_ptr;
	char *p = NULL;
	int ret = 0;

	/* find which hash bucket it is in */
	hash = tdb_hash(&key);
	if (!tdb_keylocked(tdb, hash))
		return -1;
	if (tdb_lock(tdb, BUCKET(hash), F_WRLCK) == -1)
		return -1;

	/* check for it existing, on insert. */
	if (flag == TDB_INSERT) {
		if (tdb_exists(tdb, key)) {
			tdb->ecode = TDB_ERR_EXISTS;
			goto fail;
		}
	} else {
		/* first try in-place update, on modify or replace. */
		if (tdb_update(tdb, key, dbuf) == 0)
			goto out;
		if (flag == TDB_MODIFY && tdb->ecode == TDB_ERR_NOEXIST)
			goto fail;
	}
	/* reset the error code potentially set by the tdb_update() */
	tdb->ecode = TDB_SUCCESS;

	/* delete any existing record - if it doesn't exist we don't
           care.  Doing this first reduces fragmentation, and avoids
           coalescing with `allocated' block before it's updated. */
	if (flag != TDB_INSERT)
		tdb_delete(tdb, key);

	/* Copy key+value *before* allocating free space in case malloc
	   fails and we are left with a dead spot in the tdb. */

	if (!(p = (char *)malloc(key.dsize + dbuf.dsize))) {
		tdb->ecode = TDB_ERR_OOM;
		goto fail;
	}

	memcpy(p, key.dptr, key.dsize);
	memcpy(p+key.dsize, dbuf.dptr, dbuf.dsize);

	/* now we're into insert / modify / replace of a record which
	 * we know could not be optimised by an in-place store (for
	 * various reasons).  */
	if (!(rec_ptr = tdb_allocate(tdb, key.dsize + dbuf.dsize, &rec)))
		goto fail;

	/* Read hash top into next ptr */
	if (ofs_read(tdb, TDB_HASH_TOP(hash), &rec.next) == -1)
		goto fail;

	rec.key_len = key.dsize;
	rec.data_len = dbuf.dsize;
	rec.full_hash = hash;
	rec.magic = TDB_MAGIC;

	/* write out and point the top of the hash chain at it */
	if (rec_write(tdb, rec_ptr, &rec) == -1
	    || tdb_write(tdb, rec_ptr+sizeof(rec), p, key.dsize+dbuf.dsize)==-1
	    || ofs_write(tdb, TDB_HASH_TOP(hash), &rec_ptr) == -1) {
	fail:
		/* Need to tdb_unallocate() here */
		ret = -1;
	}
 out:
	if (p)
		free(p); 
	tdb_unlock(tdb, BUCKET(hash), F_WRLCK);
	return ret;
}

static int tdb_already_open(dev_t device,
			    ino_t ino)
{
	TDB_CONTEXT *i;
	
	for (i = tdbs; i; i = i->next) {
		if (i->device == device && i->inode == ino) {
			return 1;
		}
	}

	return 0;
}

/* open the database, creating it if necessary 

   The open_flags and mode are passed straight to the open call on the
   database file. A flags value of O_WRONLY is invalid. The hash size
   is advisory, use zero for a default value.

   Return is NULL on error, in which case errno is also set.  Don't 
   try to call tdb_error or tdb_errname, just do strerror(errno).

   @param name may be NULL for internal databases. */
TDB_CONTEXT *tdb_open(char *name, int hash_size, int tdb_flags,
		      int open_flags, mode_t mode)
{
	return tdb_open_ex(name, hash_size, tdb_flags, open_flags, mode, NULL);
}


TDB_CONTEXT *tdb_open_ex(char *name, int hash_size, int tdb_flags,
			 int open_flags, mode_t mode,
			 tdb_log_func log_fn)
{
	TDB_CONTEXT *tdb;
	struct stat st;
	int rev = 0, locked;

	if (!(tdb = calloc(1, sizeof *tdb))) {
		/* Can't log this */
		errno = ENOMEM;
		goto fail;
	}
	tdb->fd = -1;
	tdb->name = NULL;
	tdb->map_ptr = NULL;
	tdb->lockedkeys = NULL;
	tdb->flags = tdb_flags;
	tdb->open_flags = open_flags;
	tdb->log_fn = log_fn;
	
	if ((open_flags & O_ACCMODE) == O_WRONLY) {
		TDB_LOG((tdb, 0, "tdb_open_ex: can't open tdb %s write-only\n",
			 name));
		errno = EINVAL;
		goto fail;
	}
	
	if (hash_size == 0)
		hash_size = DEFAULT_HASH_SIZE;
	if ((open_flags & O_ACCMODE) == O_RDONLY) {
		tdb->read_only = 1;
		/* read only databases don't do locking or clear if first */
		tdb->flags |= TDB_NOLOCK;
		tdb->flags &= ~TDB_CLEAR_IF_FIRST;
	}

	/* internal databases don't mmap or lock, and start off cleared */
	if (tdb->flags & TDB_INTERNAL) {
		tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
		tdb->flags &= ~TDB_CLEAR_IF_FIRST;
		tdb_new_database(tdb, hash_size);
		goto internal;
	}

	if ((tdb->fd = open(name, open_flags, mode)) == -1) {
		TDB_LOG((tdb, 0, "tdb_open_ex: could not open file %s: %s\n",
			 name, strerror(errno)));
		goto fail;	/* errno set by open(2) */
	}

	/* ensure there is only one process initialising at once */
	if (tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0) == -1) {
		TDB_LOG((tdb, 0, "tdb_open_ex: failed to get global lock on %s: %s\n",
			 name, strerror(errno)));
		goto fail;	/* errno set by tdb_brlock */
	}

	/* we need to zero database if we are the only one with it open */
	if ((locked = (tdb_brlock(tdb, ACTIVE_LOCK, F_WRLCK, F_SETLK, 0) == 0))
	    && (tdb_flags & TDB_CLEAR_IF_FIRST)) {
		open_flags |= O_CREAT;
		if (ftruncate(tdb->fd, 0) == -1) {
			TDB_LOG((tdb, 0, "tdb_open_ex: "
				 "failed to truncate %s: %s\n",
				 name, strerror(errno)));
			goto fail; /* errno set by ftruncate */
		}
	}

	if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
	    || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0
	    || (tdb->header.version != TDB_VERSION
		&& !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION))))) {
		/* its not a valid database - possibly initialise it */
		if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
			errno = EIO; /* ie bad format or something */
			goto fail;
		}
		rev = (tdb->flags & TDB_CONVERT);
	}
	if (!rev)
		tdb->flags &= ~TDB_CONVERT;
	else {
		tdb->flags |= TDB_CONVERT;
		convert(&tdb->header, sizeof(tdb->header));
	}
	if (fstat(tdb->fd, &st) == -1)
		goto fail;

	/* Is it already in the open list?  If so, fail. */
	if (tdb_already_open(st.st_dev, st.st_ino)) {
		TDB_LOG((tdb, 2, "tdb_open_ex: "
			 "%s (%d,%d) is already open in this process\n",
			 name, st.st_dev, st.st_ino));
		errno = EBUSY;
		goto fail;
	}

	if (!(tdb->name = (char *)strdup(name))) {
		errno = ENOMEM;
		goto fail;
	}

	tdb->map_size = st.st_size;
	tdb->device = st.st_dev;
	tdb->inode = st.st_ino;
	tdb->locked = calloc(tdb->header.hash_size+1, sizeof(tdb->locked[0]));
	if (!tdb->locked) {
		TDB_LOG((tdb, 2, "tdb_open_ex: "
			 "failed to allocate lock structure for %s\n",
			 name));
		errno = ENOMEM;
		goto fail;
	}
	tdb_mmap(tdb);
	if (locked) {
		if (!tdb->read_only)
			tdb_clear_spinlocks(tdb);
		if (tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0) == -1) {
			TDB_LOG((tdb, 0, "tdb_open_ex: "
				 "failed to take ACTIVE_LOCK on %s: %s\n",
				 name, strerror(errno)));
			goto fail;
		}
	}
	/* leave this lock in place to indicate it's in use */
	if (tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0) == -1)
		goto fail;

 internal:
	/* Internal (memory-only) databases skip all the code above to
	 * do with disk files, and resume here by releasing their
	 * global lock and hooking into the active list. */
	if (tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0) == -1)
		goto fail;
	tdb->next = tdbs;
	tdbs = tdb;
	return tdb;

 fail:
	{ int save_errno = errno;

	if (!tdb)
		return NULL;
	
	if (tdb->map_ptr) {
		if (tdb->flags & TDB_INTERNAL)
			free(tdb->map_ptr);
		else
			tdb_munmap(tdb);
	}
	if (tdb->name)
		free(tdb->name);
	if (tdb->fd != -1)
		close(tdb->fd);
	if (tdb->locked)
		free(tdb->locked);
	errno = save_errno;
	return NULL;
	}
}

/* close a database */
int tdb_close(TDB_CONTEXT *tdb)
{
	TDB_CONTEXT **i;
	int ret = 0;

	if (tdb->map_ptr) {
		if (tdb->flags & TDB_INTERNAL)
			free(tdb->map_ptr);
		else
			tdb_munmap(tdb);
	}
	if (tdb->name)
		free(tdb->name);
	if (tdb->fd != -1)
		ret = close(tdb->fd);
	if (tdb->locked)
		free(tdb->locked);
	if (tdb->lockedkeys)
		free(tdb->lockedkeys);

	/* Remove from contexts list */
	for (i = &tdbs; *i; i = &(*i)->next) {
		if (*i == tdb) {
			*i = tdb->next;
			break;
		}
	}

	memset(tdb, 0, sizeof(*tdb));
	free(tdb);

	return ret;
}

/* lock/unlock entire database */
int tdb_lockall(TDB_CONTEXT *tdb)
{
	u32 i;

	/* There are no locks on read-only dbs */
	if (tdb->read_only)
		return TDB_ERRCODE(TDB_ERR_LOCK, -1);
	if (tdb->lockedkeys)
		return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
	for (i = 0; i < tdb->header.hash_size; i++) 
		if (tdb_lock(tdb, i, F_WRLCK))
			break;

	/* If error, release locks we have... */
	if (i < tdb->header.hash_size) {
		u32 j;

		for ( j = 0; j < i; j++)
			tdb_unlock(tdb, j, F_WRLCK);
		return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
	}

	return 0;
}
void tdb_unlockall(TDB_CONTEXT *tdb)
{
	u32 i;
	for (i=0; i < tdb->header.hash_size; i++)
		tdb_unlock(tdb, i, F_WRLCK);
}

int tdb_lockkeys(TDB_CONTEXT *tdb, u32 number, TDB_DATA keys[])
{
	u32 i, j, hash;

	/* Can't lock more keys if already locked */
	if (tdb->lockedkeys)
		return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
	if (!(tdb->lockedkeys = malloc(sizeof(u32) * (number+1))))
		return TDB_ERRCODE(TDB_ERR_OOM, -1);
	/* First number in array is # keys */
	tdb->lockedkeys[0] = number;

	/* Insertion sort by bucket */
	for (i = 0; i < number; i++) {
		hash = tdb_hash(&keys[i]);
		for (j = 0; j < i && BUCKET(tdb->lockedkeys[j+1]) < BUCKET(hash); j++);
			memmove(&tdb->lockedkeys[j+2], &tdb->lockedkeys[j+1], sizeof(u32) * (i-j));
		tdb->lockedkeys[j+1] = hash;
	}
	/* Finally, lock in order */
	for (i = 0; i < number; i++)
		if (tdb_lock(tdb, i, F_WRLCK))
			break;

	/* If error, release locks we have... */
	if (i < number) {
		for ( j = 0; j < i; j++)
			tdb_unlock(tdb, j, F_WRLCK);
		free(tdb->lockedkeys);
		tdb->lockedkeys = NULL;
		return TDB_ERRCODE(TDB_ERR_NOLOCK, -1);
	}
	return 0;
}

/* Unlock the keys previously locked by tdb_lockkeys() */
void tdb_unlockkeys(TDB_CONTEXT *tdb)
{
	u32 i;
	for (i = 0; i < tdb->lockedkeys[0]; i++)
		tdb_unlock(tdb, tdb->lockedkeys[i+1], F_WRLCK);
	free(tdb->lockedkeys);
	tdb->lockedkeys = NULL;
}

/* lock/unlock one hash chain. This is meant to be used to reduce
   contention - it cannot guarantee how many records will be locked */
int tdb_chainlock(TDB_CONTEXT *tdb, TDB_DATA key)
{
	return tdb_lock(tdb, BUCKET(tdb_hash(&key)), F_WRLCK);
}
void tdb_chainunlock(TDB_CONTEXT *tdb, TDB_DATA key)
{
	tdb_unlock(tdb, BUCKET(tdb_hash(&key)), F_WRLCK);
}


/* register a loging function */
void tdb_logging_function(TDB_CONTEXT *tdb, void (*fn)(TDB_CONTEXT *, int , const char *, ...))
{
	tdb->log_fn = fn;
}


/* reopen a tdb - this is used after a fork to ensure that we have an independent
   seek pointer from our parent and to re-establish locks */
int tdb_reopen(TDB_CONTEXT *tdb)
{
	struct stat st;

	tdb_munmap(tdb);
	close(tdb->fd);
	tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
	if (tdb->fd == -1) {
		TDB_LOG((tdb, 0, "tdb_reopen: open failed (%s)\n", strerror(errno)));
		goto fail;
	}
	fstat(tdb->fd, &st);
	if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
		TDB_LOG((tdb, 0, "tdb_reopen: file dev/inode has changed!\n"));
		goto fail;
	}
	tdb_mmap(tdb);
	if (tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0) == -1) {
		TDB_LOG((tdb, 0, "tdb_reopen: failed to obtain active lock\n"));
		goto fail;
	}

	return 0;

fail:
	tdb_close(tdb);
	return -1;
}

/* reopen all tdb's */
int tdb_reopen_all(void)
{
	TDB_CONTEXT *tdb;

	for (tdb=tdbs; tdb; tdb = tdb->next) {
		if (tdb_reopen(tdb) != 0) return -1;
	}

	return 0;
}