This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
- version 2 of the License, or (at your option) any later version.
+ version 3 of the License, or (at your option) any later version.
This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
- License along with this library; if not, write to the Free Software
- Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
+ License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#include "tdb_private.h"
/* This is based on the hash algorithm from gdbm */
static unsigned int default_tdb_hash(TDB_DATA *key)
{
- u32 value; /* Used to compute the hash value. */
- u32 i; /* Used to cycle through random values. */
+ uint32_t value; /* Used to compute the hash value. */
+ uint32_t i; /* Used to cycle through random values. */
/* Set the initial value from the key size. */
for (value = 0x238F13AF * key->dsize, i=0; i < key->dsize; i++)
static int tdb_new_database(struct tdb_context *tdb, int hash_size)
{
struct tdb_header *newdb;
- int size, ret = -1;
+ size_t size;
+ int ret = -1;
+ ssize_t written;
/* We make it up in memory, then write it out if not internal */
size = sizeof(struct tdb_header) + (hash_size+1)*sizeof(tdb_off_t);
- if (!(newdb = (struct tdb_header *)calloc(size, 1)))
- return TDB_ERRCODE(TDB_ERR_OOM, -1);
+ if (!(newdb = (struct tdb_header *)calloc(size, 1))) {
+ tdb->ecode = TDB_ERR_OOM;
+ return -1;
+ }
/* Fill in the header */
newdb->version = TDB_VERSION;
memcpy(&tdb->header, newdb, sizeof(tdb->header));
/* Don't endian-convert the magic food! */
memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
- if (write(tdb->fd, newdb, size) != size) {
- ret = -1;
- } else {
+ /* we still have "ret == -1" here */
+ written = write(tdb->fd, newdb, size);
+ if (written == size) {
+ ret = 0;
+ } else if (written != -1) {
+ /* call write once again, this usually should return -1 and
+ * set errno appropriately */
+ size -= written;
+ written = write(tdb->fd, newdb+written, size);
+ if (written == size) {
ret = 0;
+ } else if (written >= 0) {
+ /* a second incomplete write - we give up.
+ * guessing the errno... */
+ errno = ENOSPC;
+ }
}
fail:
struct stat st;
int rev = 0, locked = 0;
unsigned char *vp;
- u32 vertest;
+ uint32_t vertest;
+ unsigned v;
if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
/* Can't log this */
}
tdb_io_init(tdb);
tdb->fd = -1;
+#ifdef TDB_TRACE
+ tdb->tracefd = -1;
+#endif
tdb->name = NULL;
tdb->map_ptr = NULL;
tdb->flags = tdb_flags;
tdb->page_size = 0x2000;
}
+ tdb->max_dead_records = (tdb_flags & TDB_VOLATILE) ? 5 : 0;
+
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));
tdb->flags &= ~TDB_CLEAR_IF_FIRST;
}
+ if ((tdb->flags & TDB_ALLOW_NESTING) &&
+ (tdb->flags & TDB_DISALLOW_NESTING)) {
+ tdb->ecode = TDB_ERR_NESTING;
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
+ "allow_nesting and disallow_nesting are not allowed together!"));
+ errno = EINVAL;
+ goto fail;
+ }
+
+ /*
+ * TDB_ALLOW_NESTING is the default behavior.
+ * Note: this may change in future versions!
+ */
+ if (!(tdb->flags & TDB_DISALLOW_NESTING)) {
+ tdb->flags |= TDB_ALLOW_NESTING;
+ }
+
/* internal databases don't mmap or lock, and start off cleared */
if (tdb->flags & TDB_INTERNAL) {
tdb->flags |= (TDB_NOLOCK | TDB_NOMMAP);
goto fail; /* errno set by open(2) */
}
+ /* on exec, don't inherit the fd */
+ v = fcntl(tdb->fd, F_GETFD, 0);
+ fcntl(tdb->fd, F_SETFD, v | FD_CLOEXEC);
+
/* 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",
+ if (tdb->methods->brlock(tdb, F_WRLCK, OPEN_LOCK, 1, TDB_LOCK_WAIT) == -1) {
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to get open 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))) {
+ (!tdb->read_only) &&
+ (locked = (tdb->methods->brlock(tdb, F_WRLCK, ACTIVE_LOCK, 1, TDB_LOCK_NOWAIT|TDB_LOCK_PROBE) == 0))) {
open_flags |= O_CREAT;
if (ftruncate(tdb->fd, 0) == -1) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
}
}
+ errno = 0;
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 */
+ || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0) {
if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
- errno = EIO; /* ie bad format or something */
+ if (errno == 0) {
+ errno = EIO; /* ie bad format or something */
+ }
goto fail;
}
rev = (tdb->flags & TDB_CONVERT);
+ } else if (tdb->header.version != TDB_VERSION
+ && !(rev = (tdb->header.version==TDB_BYTEREV(TDB_VERSION)))) {
+ /* wrong version */
+ errno = EIO;
+ goto fail;
}
vp = (unsigned char *)&tdb->header.version;
- vertest = (((u32)vp[0]) << 24) | (((u32)vp[1]) << 16) |
- (((u32)vp[2]) << 8) | (u32)vp[3];
+ vertest = (((uint32_t)vp[0]) << 24) | (((uint32_t)vp[1]) << 16) |
+ (((uint32_t)vp[2]) << 8) | (uint32_t)vp[3];
tdb->flags |= (vertest==TDB_VERSION) ? TDB_BIGENDIAN : 0;
if (!rev)
tdb->flags &= ~TDB_CONVERT;
tdb->map_size = st.st_size;
tdb->device = st.st_dev;
tdb->inode = st.st_ino;
- tdb->max_dead_records = 0;
tdb_mmap(tdb);
if (locked) {
- if (tdb->methods->tdb_brlock(tdb, ACTIVE_LOCK, F_UNLCK, F_SETLK, 0, 1) == -1) {
+ if (tdb->methods->brunlock(tdb, F_WRLCK, ACTIVE_LOCK, 1) == -1) {
TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
- "failed to take ACTIVE_LOCK on %s: %s\n",
+ "failed to release ACTIVE_LOCK on %s: %s\n",
name, strerror(errno)));
goto fail;
}
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)
+ if (tdb->methods->brlock(tdb, F_RDLCK, ACTIVE_LOCK, 1, TDB_LOCK_WAIT) == -1)
goto fail;
}
goto fail;
}
+#ifdef TDB_TRACE
+ {
+ char tracefile[strlen(name) + 32];
+
+ snprintf(tracefile, sizeof(tracefile),
+ "%s.trace.%li", name, (long)getpid());
+ tdb->tracefd = open(tracefile, O_WRONLY|O_CREAT|O_EXCL, 0600);
+ if (tdb->tracefd >= 0) {
+ tdb_enable_seqnum(tdb);
+ tdb_trace_open(tdb, "tdb_open", hash_size, tdb_flags,
+ open_flags);
+ } else
+ TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: failed to open trace file %s!\n", tracefile));
+ }
+#endif
+
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)
+ * open lock and hooking into the active list. */
+ if (tdb->methods->brunlock(tdb, F_WRLCK, OPEN_LOCK, 1) == -1)
goto fail;
tdb->next = tdbs;
tdbs = tdb;
if (!tdb)
return NULL;
-
+
+#ifdef TDB_TRACE
+ close(tdb->tracefd);
+#endif
if (tdb->map_ptr) {
if (tdb->flags & TDB_INTERNAL)
SAFE_FREE(tdb->map_ptr);
if (tdb->transaction) {
tdb_transaction_cancel(tdb);
}
+ tdb_trace(tdb, "tdb_close");
if (tdb->map_ptr) {
if (tdb->flags & TDB_INTERNAL)
tdb_munmap(tdb);
}
SAFE_FREE(tdb->name);
- if (tdb->fd != -1)
+ if (tdb->fd != -1) {
ret = close(tdb->fd);
+ tdb->fd = -1;
+ }
SAFE_FREE(tdb->lockrecs);
/* Remove from contexts list */
}
}
+#ifdef TDB_TRACE
+ close(tdb->tracefd);
+#endif
memset(tdb, 0, sizeof(*tdb));
SAFE_FREE(tdb);
return tdb->log.log_private;
}
-/* 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)
+static int tdb_reopen_internal(struct tdb_context *tdb, bool active_lock)
{
+#if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
+ !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
struct stat st;
+#endif
if (tdb->flags & TDB_INTERNAL) {
return 0; /* Nothing to do. */
goto fail;
}
+/* If we have real pread & pwrite, we can skip reopen. */
+#if !defined(LIBREPLACE_PREAD_NOT_REPLACED) || \
+ !defined(LIBREPLACE_PWRITE_NOT_REPLACED)
if (tdb_munmap(tdb) != 0) {
TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: munmap failed (%s)\n", strerror(errno)));
goto fail;
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;
goto fail;
}
tdb_mmap(tdb);
+#endif /* fake pread or pwrite */
+
+ if (active_lock &&
+ (tdb->methods->brlock(tdb, F_RDLCK, ACTIVE_LOCK, 1, TDB_LOCK_WAIT) == -1)) {
+ TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
+ goto fail;
+ }
return 0;
return -1;
}
+/* 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)
+{
+ return tdb_reopen_internal(tdb, tdb->flags & TDB_CLEAR_IF_FIRST);
+}
+
/* reopen all tdb's */
int tdb_reopen_all(int parent_longlived)
{
struct tdb_context *tdb;
for (tdb=tdbs; tdb; tdb = tdb->next) {
+ bool active_lock = (tdb->flags & TDB_CLEAR_IF_FIRST);
+
/*
* If the parent is longlived (ie. a
* parent daemon architecture), we know
*/
if (parent_longlived) {
/* Ensure no clear-if-first. */
- tdb->flags &= ~TDB_CLEAR_IF_FIRST;
+ active_lock = false;
}
- if (tdb_reopen(tdb) != 0)
+ if (tdb_reopen_internal(tdb, active_lock) != 0)
return -1;
}