/**
* Close a database.
*
* @returns -1 for error; 0 for success.
**/
int tdb_close(struct tdb_context *tdb)
{
struct tdb_context **i;
int ret = 0;
if (tdb->transaction) {
tdb_transaction_cancel(tdb);
}
if (tdb->map_ptr) {
if (tdb->flags & TDB_INTERNAL)
SAFE_FREE(tdb->map_ptr);
else
tdb_munmap(tdb);
}
SAFE_FREE(tdb->name);
if (tdb->fd != -1)
ret = close(tdb->fd);
SAFE_FREE(tdb->locked);
/* Remove from contexts list */
for (i = &tdbs; *i; i = &(*i)->next) {
if (*i == tdb) {
*i = tdb->next;
break;
}
}
memset(tdb, 0, sizeof(*tdb));
SAFE_FREE(tdb);
return ret;
}
void tdb_add_flag(struct tdb_context *tdb, unsigned flag)
{
if (tdb->flags & TDB_INTERNAL) {
tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
TDB_LOG_USE_ERROR,
"tdb_add_flag: internal db");
return;
}
switch (flag) {
case TDB_NOLOCK:
tdb->flags |= TDB_NOLOCK;
break;
case TDB_NOMMAP:
tdb->flags |= TDB_NOMMAP;
tdb_munmap(tdb->file);
break;
case TDB_NOSYNC:
tdb->flags |= TDB_NOSYNC;
break;
case TDB_SEQNUM:
tdb->flags |= TDB_SEQNUM;
break;
case TDB_ALLOW_NESTING:
tdb->flags |= TDB_ALLOW_NESTING;
break;
default:
tdb->last_error = tdb_logerr(tdb, TDB_ERR_EINVAL,
TDB_LOG_USE_ERROR,
"tdb_add_flag: Unknown flag %u",
flag);
}
}
/* reopen a tdb - this can be used after a fork to ensure that we have an independent
seek pointer from our parent and to re-establish locks */
int tdb_reopen(struct tdb_context *tdb)
{
struct stat st;
if (tdb->flags & TDB_INTERNAL) {
return 0; /* Nothing to do. */
}
if (tdb->num_locks != 0 || tdb->global_lock.count) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed with locks held\n"));
goto fail;
}
if (tdb->transaction != 0) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_reopen: reopen not allowed inside a transaction\n"));
goto fail;
}
if (tdb_munmap(tdb) != 0) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
goto fail;
}
if (close(tdb->fd) != 0)
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: WARNING closing tdb->fd failed!\n"));
tdb->fd = open(tdb->name, tdb->open_flags & ~(O_CREAT|O_TRUNC), 0);
if (tdb->fd == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
goto fail;
}
if ((tdb->flags & TDB_CLEAR_IF_FIRST) &&
(tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
goto fail;
}
if (fstat(tdb->fd, &st) != 0) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
goto fail;
}
if (st.st_ino != tdb->inode || st.st_dev != tdb->device) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
goto fail;
}
tdb_mmap(tdb);
return 0;
fail:
tdb_close(tdb);
return -1;
}
/* check for an out of bounds access - if it is out of bounds then
see if the database has been expanded by someone else and expand
if necessary
note that "len" is the minimum length needed for the db
*/
static int tdb_oob(struct tdb_context *tdb, tdb_off_t len, int probe)
{
struct stat st;
if (len <= tdb->map_size)
return 0;
if (tdb->flags & TDB_INTERNAL) {
if (!probe) {
/* Ensure ecode is set for log fn. */
tdb->ecode = TDB_ERR_IO;
TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_oob len %d beyond internal malloc size %d\n",
(int)len, (int)tdb->map_size));
}
return -1;
}
if (fstat(tdb->fd, &st) == -1) {
tdb->ecode = TDB_ERR_IO;
return -1;
}
/* Unmap, update size, remap */
if (tdb_munmap(tdb) == -1) {
tdb->ecode = TDB_ERR_IO;
return -1;
}
tdb->map_size = st.st_size;
tdb_mmap(tdb);
if (st.st_size < (size_t)len) {
if (!probe) {
/* Ensure ecode is set for log fn. */
tdb->ecode = TDB_ERR_IO;
TDB_LOG((tdb, TDB_DEBUG_FATAL,"tdb_oob len %d beyond eof at %d\n",
(int)len, (int)st.st_size));
}
return -1;
}
return 0;
}
struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
int open_flags, mode_t mode,
const struct tdb_logging_context *log_ctx,
tdb_hash_func hash_fn)
{
struct tdb_context *tdb;
struct stat st;
int rev = 0, locked = 0;
unsigned char *vp;
u32 vertest;
if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
/* Can't log this */
errno = ENOMEM;
goto fail;
}
tdb_io_init(tdb);
tdb->fd = -1;
tdb->name = NULL;
tdb->map_ptr = NULL;
tdb->flags = tdb_flags;
tdb->open_flags = open_flags;
if (log_ctx) {
tdb->log = *log_ctx;
} else {
tdb->log.log_fn = null_log_fn;
tdb->log.log_private = NULL;
}
tdb->hash_fn = hash_fn ? hash_fn : default_tdb_hash;
/* cache the page size */
tdb->page_size = getpagesize();
if (tdb->page_size <= 0) {
tdb->page_size = 0x2000;
}
if ((open_flags & O_ACCMODE) == O_WRONLY) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "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;
if (tdb_new_database(tdb, hash_size) != 0) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
goto fail;
}
goto internal;
}
if ((tdb->fd = open(name, open_flags, mode)) == -1) {
TDB_LOG((tdb, TDB_DEBUG_WARNING, "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->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_WRLCK, F_SETLKW, 0, 1) == -1) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "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 ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
(locked = (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_WRLCK, F_SETLK, 0, 1) == 0))) {
open_flags |= O_CREAT;
if (ftruncate(tdb->fd, 0) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "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);
}
vp = (unsigned char *)&tdb->header.version;
vertest = (((u32)vp[0]) << 24) | (((u32)vp[1]) << 16) |
(((u32)vp[2]) << 8) | (u32)vp[3];
tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
if (!rev)
tdb->flags &= ~TDB_CONVERT;
else {
tdb->flags |= TDB_CONVERT;
tdb_convert(&tdb->header, sizeof(tdb->header));
}
if (fstat(tdb->fd, &st) == -1)
goto fail;
if (tdb->header.rwlocks != 0) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
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, TDB_DEBUG_ERROR, "tdb_open_ex: "
"%s (%d,%d) is already open in this process\n",
name, (int)st.st_dev, (int)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 = (struct tdb_lock_type *)calloc(tdb->header.hash_size+1,
sizeof(tdb->locked[0]));
if (!tdb->locked) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
"failed to allocate lock structure for %s\n",
name));
errno = ENOMEM;
goto fail;
}
tdb_mmap(tdb);
if (locked) {
if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0, 1) == -1) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
"failed to take ACTIVE_LOCK on %s: %s\n",
name, strerror(errno)));
goto fail;
}
}
/* We always need to do this if the CLEAR_IF_FIRST flag is set, even if
we didn't get the initial exclusive lock as we need to let all other
users know we're using it. */
if (tdb_flags & TDB_CLEAR_IF_FIRST) {
/* leave this lock in place to indicate it's in use */
if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_RDLCK, F_SETLKW, 0, 1) == -1)
goto fail;
}
/* if needed, run recovery */
if (tdb_transaction_recover(tdb) == -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->methods->tdb_brlock(tdb, GLOBAL_LOCK, F_UNLCK, F_SETLKW, 0, 1) == -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)
SAFE_FREE(tdb->map_ptr);
else
tdb_munmap(tdb);
}
SAFE_FREE(tdb->name);
if (tdb->fd != -1)
if (close(tdb->fd) != 0)
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to close tdb->fd on error!\n"));
SAFE_FREE(tdb->locked);
SAFE_FREE(tdb);
errno = save_errno;
return NULL;
}
}
/* expand the database at least size bytes by expanding the underlying
file and doing the mmap again if necessary */
int tdb_expand(struct tdb_context *tdb, tdb_off_t size)
{
struct tdb_record rec;
tdb_off_t offset, new_size, top_size, map_size;
if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "lock failed in tdb_expand\n"));
return -1;
}
/* must know about any previous expansions by another process */
tdb->methods->tdb_oob(tdb, tdb->map_size + 1, 1);
/* limit size in order to avoid using up huge amounts of memory for
* in memory tdbs if an oddball huge record creeps in */
if (size > 100 * 1024) {
top_size = tdb->map_size + size * 2;
} else {
top_size = tdb->map_size + size * 100;
}
/* always make room for at least top_size more records, and at
least 25% more space. if the DB is smaller than 100MiB,
otherwise grow it by 10% only. */
if (tdb->map_size > 100 * 1024 * 1024) {
map_size = tdb->map_size * 1.10;
} else {
map_size = tdb->map_size * 1.25;
}
/* Round the database up to a multiple of the page size */
new_size = MAX(top_size, map_size);
size = TDB_ALIGN(new_size, tdb->page_size) - tdb->map_size;
if (!(tdb->flags & TDB_INTERNAL))
tdb_munmap(tdb);
/*
* We must ensure the file is unmapped before doing this
* to ensure consistency with systems like OpenBSD where
* writes and mmaps are not consistent.
*/
/* expand the file itself */
if (!(tdb->flags & TDB_INTERNAL)) {
if (tdb->methods->tdb_expand_file(tdb, tdb->map_size, size) != 0)
goto fail;
}
tdb->map_size += size;
if (tdb->flags & TDB_INTERNAL) {
char *new_map_ptr = (char *)realloc(tdb->map_ptr,
tdb->map_size);
if (!new_map_ptr) {
tdb->map_size -= size;
goto fail;
}
tdb->map_ptr = new_map_ptr;
} else {
/*
* We must ensure the file is remapped before adding the space
* to ensure consistency with systems like OpenBSD where
* writes and mmaps are not consistent.
*/
/* We're ok if the mmap fails as we'll fallback to read/write */
tdb_mmap(tdb);
}
/* form a new freelist record */
memset(&rec,'\0',sizeof(rec));
rec.rec_len = size - sizeof(rec);
/* link it into the free list */
offset = tdb->map_size - size;
if (tdb_free(tdb, offset, &rec) == -1)
goto fail;
tdb_unlock(tdb, -1, F_WRLCK);
return 0;
fail:
tdb_unlock(tdb, -1, F_WRLCK);
return -1;
}
/* expand the database at least size bytes by expanding the underlying
file and doing the mmap again if necessary */
int tdb_expand(struct tdb_context *tdb, tdb_off_t size)
{
struct list_struct rec;
tdb_off_t offset;
if (tdb_lock(tdb, -1, F_WRLCK) == -1) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "lock failed in tdb_expand\n"));
return -1;
}
/* must know about any previous expansions by another process */
tdb->methods->tdb_oob(tdb, tdb->map_size + 1, 1);
/* always make room for at least 10 more records, and round
the database up to a multiple of the page size */
size = TDB_ALIGN(tdb->map_size + size*10, tdb->page_size) - tdb->map_size;
if (!(tdb->flags & TDB_INTERNAL))
tdb_munmap(tdb);
/*
* We must ensure the file is unmapped before doing this
* to ensure consistency with systems like OpenBSD where
* writes and mmaps are not consistent.
*/
/* expand the file itself */
if (!(tdb->flags & TDB_INTERNAL)) {
if (tdb->methods->tdb_expand_file(tdb, tdb->map_size, size) != 0)
goto fail;
}
tdb->map_size += size;
if (tdb->flags & TDB_INTERNAL) {
char *new_map_ptr = (char *)realloc(tdb->map_ptr,
tdb->map_size);
if (!new_map_ptr) {
tdb->map_size -= size;
goto fail;
}
tdb->map_ptr = new_map_ptr;
} else {
/*
* We must ensure the file is remapped before adding the space
* to ensure consistency with systems like OpenBSD where
* writes and mmaps are not consistent.
*/
/* We're ok if the mmap fails as we'll fallback to read/write */
tdb_mmap(tdb);
}
/* form a new freelist record */
memset(&rec,'\0',sizeof(rec));
rec.rec_len = size - sizeof(rec);
/* link it into the free list */
offset = tdb->map_size - size;
if (tdb_free(tdb, offset, &rec) == -1)
goto fail;
tdb_unlock(tdb, -1, F_WRLCK);
return 0;
fail:
tdb_unlock(tdb, -1, F_WRLCK);
return -1;
}
