ctdb-scripts: Do not de-duplicate the interfaces list
[samba.git] / lib / tdb / common / open.c
index 66539c3f6c5dd9fece0010a3822355ec07c9af34..3fa7ce1389d690a43cd8605be0e3263b7da43d82 100644 (file)
@@ -1,4 +1,4 @@
- /* 
+ /*
    Unix SMB/CIFS implementation.
 
    trivial database library
@@ -51,7 +51,8 @@ void tdb_header_hash(struct tdb_context *tdb,
 }
 
 /* initialise a new database with a specified hash size */
-static int tdb_new_database(struct tdb_context *tdb, int hash_size)
+static int tdb_new_database(struct tdb_context *tdb, struct tdb_header *header,
+                           int hash_size)
 {
        struct tdb_header *newdb;
        size_t size;
@@ -75,10 +76,36 @@ static int tdb_new_database(struct tdb_context *tdb, int hash_size)
        if (tdb->flags & TDB_INCOMPATIBLE_HASH)
                newdb->rwlocks = TDB_HASH_RWLOCK_MAGIC;
 
+       /*
+        * We create a tdb with TDB_FEATURE_FLAG_MUTEX support,
+        * the flag combination and runtime feature checks
+        * are done by the caller already.
+        */
+       if (tdb->flags & TDB_MUTEX_LOCKING) {
+               newdb->feature_flags |= TDB_FEATURE_FLAG_MUTEX;
+       }
+
+       /*
+        * If we have any features we add the FEATURE_FLAG_MAGIC, overwriting the
+        * TDB_HASH_RWLOCK_MAGIC above.
+        */
+       if (newdb->feature_flags != 0) {
+               newdb->rwlocks = TDB_FEATURE_FLAG_MAGIC;
+       }
+
+       /*
+        * It's required for some following code paths
+        * to have the fields on 'tdb' up-to-date.
+        *
+        * E.g. tdb_mutex_size() requires it
+        */
+       tdb->feature_flags = newdb->feature_flags;
+       tdb->hash_size = newdb->hash_size;
+
        if (tdb->flags & TDB_INTERNAL) {
                tdb->map_size = size;
                tdb->map_ptr = (char *)newdb;
-               memcpy(&tdb->header, newdb, sizeof(tdb->header));
+               memcpy(header, newdb, sizeof(*header));
                /* Convert the `ondisk' version if asked. */
                CONVERT(*newdb);
                return 0;
@@ -89,15 +116,52 @@ static int tdb_new_database(struct tdb_context *tdb, int hash_size)
        if (ftruncate(tdb->fd, 0) == -1)
                goto fail;
 
+       if (newdb->feature_flags & TDB_FEATURE_FLAG_MUTEX) {
+               newdb->mutex_size = tdb_mutex_size(tdb);
+               tdb->hdr_ofs = newdb->mutex_size;
+       }
+
        /* This creates an endian-converted header, as if read from disk */
        CONVERT(*newdb);
-       memcpy(&tdb->header, newdb, sizeof(tdb->header));
+       memcpy(header, newdb, sizeof(*header));
        /* Don't endian-convert the magic food! */
        memcpy(newdb->magic_food, TDB_MAGIC_FOOD, strlen(TDB_MAGIC_FOOD)+1);
-       /* we still have "ret == -1" here */
-       if (tdb_write_all(tdb->fd, newdb, size))
-               ret = 0;
 
+       if (!tdb_write_all(tdb->fd, newdb, size))
+               goto fail;
+
+       if (newdb->feature_flags & TDB_FEATURE_FLAG_MUTEX) {
+
+               /*
+                * Now we init the mutex area
+                * followed by a second header.
+                */
+
+               ret = ftruncate(
+                       tdb->fd,
+                       newdb->mutex_size + sizeof(struct tdb_header));
+               if (ret == -1) {
+                       goto fail;
+               }
+               ret = tdb_mutex_init(tdb);
+               if (ret == -1) {
+                       goto fail;
+               }
+
+               /*
+                * Write a second header behind the mutexes. That's the area
+                * that will be mmapp'ed.
+                */
+               ret = lseek(tdb->fd, newdb->mutex_size, SEEK_SET);
+               if (ret == -1) {
+                       goto fail;
+               }
+               if (!tdb_write_all(tdb->fd, newdb, size)) {
+                       goto fail;
+               }
+       }
+
+       ret = 0;
   fail:
        SAFE_FREE(newdb);
        return ret;
@@ -119,17 +183,17 @@ static int tdb_already_open(dev_t device,
        return 0;
 }
 
-/* open the database, creating it if necessary 
+/* 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 
+   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. */
-struct tdb_context *tdb_open(const char *name, int hash_size, int tdb_flags,
+_PUBLIC_ struct tdb_context *tdb_open(const 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, NULL);
@@ -142,11 +206,12 @@ static void null_log_fn(struct tdb_context *tdb, enum tdb_debug_level level, con
 }
 
 static bool check_header_hash(struct tdb_context *tdb,
+                             struct tdb_header *header,
                              bool default_hash, uint32_t *m1, uint32_t *m2)
 {
        tdb_header_hash(tdb, m1, m2);
-       if (tdb->header.magic1_hash == *m1 &&
-           tdb->header.magic2_hash == *m2) {
+       if (header->magic1_hash == *m1 &&
+           header->magic2_hash == *m2) {
                return true;
        }
 
@@ -159,22 +224,59 @@ static bool check_header_hash(struct tdb_context *tdb,
                tdb->hash_fn = tdb_jenkins_hash;
        else
                tdb->hash_fn = tdb_old_hash;
-       return check_header_hash(tdb, false, m1, m2);
+       return check_header_hash(tdb, header, false, m1, m2);
 }
 
-struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
+static bool tdb_mutex_open_ok(struct tdb_context *tdb,
+                             const struct tdb_header *header)
+{
+       if (tdb->flags & TDB_NOLOCK) {
+               /*
+                * We don't look at locks, so it does not matter to have a
+                * compatible mutex implementation. Allow the open.
+                */
+               return true;
+       }
+
+       if (!(tdb->flags & TDB_MUTEX_LOCKING)) {
+               TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_mutex_open_ok[%s]: "
+                        "Can use mutexes only with "
+                        "MUTEX_LOCKING or NOLOCK\n",
+                        tdb->name));
+               return false;
+       }
+
+       if (tdb_mutex_size(tdb) != header->mutex_size) {
+               TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_mutex_open_ok[%s]: "
+                        "Mutex size changed from %"PRIu32" to %zu\n.",
+                        tdb->name,
+                        header->mutex_size,
+                        tdb_mutex_size(tdb)));
+               return false;
+       }
+
+       return true;
+}
+
+_PUBLIC_ 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)
 {
+       int orig_errno = errno;
+       struct tdb_header header = {
+               .version = 0,
+       };
        struct tdb_context *tdb;
        struct stat st;
-       int rev = 0, locked = 0;
+       int rev = 0;
+       bool locked = false;
        unsigned char *vp;
        uint32_t vertest;
        unsigned v;
        const char *hash_alg;
        uint32_t magic1, magic2;
+       int ret;
 
        if (!(tdb = (struct tdb_context *)calloc(1, sizeof *tdb))) {
                /* Can't log this */
@@ -182,6 +284,14 @@ struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
                goto fail;
        }
        tdb_io_init(tdb);
+
+       if (tdb_flags & TDB_INTERNAL) {
+               tdb_flags |= TDB_INCOMPATIBLE_HASH;
+       }
+       if (tdb_flags & TDB_MUTEX_LOCKING) {
+               tdb_flags |= TDB_INCOMPATIBLE_HASH;
+       }
+
        tdb->fd = -1;
 #ifdef TDB_TRACE
        tdb->tracefd = -1;
@@ -197,6 +307,32 @@ struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
                tdb->log.log_private = NULL;
        }
 
+       if (name == NULL && (tdb_flags & TDB_INTERNAL)) {
+               name = "__TDB_INTERNAL__";
+       }
+
+       if (name == NULL) {
+               tdb->name = discard_const_p(char, "__NULL__");
+               TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: called with name == NULL\n"));
+               tdb->name = NULL;
+               errno = EINVAL;
+               goto fail;
+       }
+
+       /* now make a copy of the name, as the caller memory might go away */
+       if (!(tdb->name = (char *)strdup(name))) {
+               /*
+                * set the name as the given string, so that tdb_name() will
+                * work in case of an error.
+                */
+               tdb->name = discard_const_p(char, name);
+               TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: can't strdup(%s)\n",
+                        name));
+               tdb->name = NULL;
+               errno = ENOMEM;
+               goto fail;
+       }
+
        if (hash_fn) {
                tdb->hash_fn = hash_fn;
                hash_alg = "the user defined";
@@ -231,7 +367,7 @@ struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
                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;
+               tdb->flags &= ~(TDB_CLEAR_IF_FIRST|TDB_MUTEX_LOCKING);
        }
 
        if ((tdb->flags & TDB_ALLOW_NESTING) &&
@@ -243,6 +379,56 @@ struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
                goto fail;
        }
 
+       if (tdb->flags & TDB_MUTEX_LOCKING) {
+               /*
+                * Here we catch bugs in the callers,
+                * the runtime check for existing tdb's comes later.
+                */
+
+               if (tdb->flags & TDB_INTERNAL) {
+                       TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
+                               "invalid flags for %s - TDB_MUTEX_LOCKING and "
+                               "TDB_INTERNAL are not allowed together\n", name));
+                       errno = EINVAL;
+                       goto fail;
+               }
+
+               if (tdb->flags & TDB_NOMMAP) {
+                       TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
+                               "invalid flags for %s - TDB_MUTEX_LOCKING and "
+                               "TDB_NOMMAP are not allowed together\n", name));
+                       errno = EINVAL;
+                       goto fail;
+               }
+
+               if (tdb->read_only) {
+                       TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
+                               "invalid flags for %s - TDB_MUTEX_LOCKING "
+                               "not allowed read only\n", name));
+                       errno = EINVAL;
+                       goto fail;
+               }
+
+               /*
+                * The callers should have called
+                * tdb_runtime_check_for_robust_mutexes()
+                * before using TDB_MUTEX_LOCKING!
+                *
+                * This makes sure the caller understands
+                * that the locking may behave a bit differently
+                * than with pure fcntl locking. E.g. multiple
+                * read locks are not supported.
+                */
+               if (!tdb_runtime_check_for_robust_mutexes()) {
+                       TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
+                               "invalid flags for %s - TDB_MUTEX_LOCKING "
+                               "requires support for robust_mutexes\n",
+                               name));
+                       errno = ENOSYS;
+                       goto fail;
+               }
+       }
+
        if (getenv("TDB_NO_FSYNC")) {
                tdb->flags |= TDB_NOSYNC;
        }
@@ -259,10 +445,11 @@ struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
        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) {
+               if (tdb_new_database(tdb, &header, hash_size) != 0) {
                        TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: tdb_new_database failed!"));
                        goto fail;
                }
+               tdb->hash_size = hash_size;
                goto internal;
        }
 
@@ -285,34 +472,69 @@ struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
 
        /* we need to zero database if we are the only one with it open */
        if ((tdb_flags & TDB_CLEAR_IF_FIRST) &&
-           (!tdb->read_only) &&
-           (locked = (tdb_nest_lock(tdb, ACTIVE_LOCK, F_WRLCK, 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: "
-                                "failed to truncate %s: %s\n",
-                                name, strerror(errno)));
-                       goto fail; /* errno set by ftruncate */
+           (!tdb->read_only)) {
+               ret = tdb_nest_lock(tdb, ACTIVE_LOCK, F_WRLCK,
+                                   TDB_LOCK_NOWAIT|TDB_LOCK_PROBE);
+               locked = (ret == 0);
+
+               if (locked) {
+                       ret = tdb_brlock(tdb, F_WRLCK, FREELIST_TOP, 0,
+                                        TDB_LOCK_WAIT);
+                       if (ret == -1) {
+                               TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
+                                        "tdb_brlock failed for %s: %s\n",
+                                        name, strerror(errno)));
+                               goto fail;
+                       }
+                       ret = tdb_new_database(tdb, &header, hash_size);
+                       if (ret == -1) {
+                               TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
+                                        "tdb_new_database failed for "
+                                        "%s: %s\n", name, strerror(errno)));
+                               tdb_unlockall(tdb);
+                               goto fail;
+                       }
+                       ret = tdb_brunlock(tdb, F_WRLCK, FREELIST_TOP, 0);
+                       if (ret == -1) {
+                               TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
+                                        "tdb_unlockall failed for %s: %s\n",
+                                        name, strerror(errno)));
+                               goto fail;
+                       }
+                       ret = lseek(tdb->fd, 0, SEEK_SET);
+                       if (ret == -1) {
+                               TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
+                                        "lseek failed for %s: %s\n",
+                                        name, strerror(errno)));
+                               goto fail;
+                       }
                }
        }
 
        errno = 0;
-       if (read(tdb->fd, &tdb->header, sizeof(tdb->header)) != sizeof(tdb->header)
-           || strcmp(tdb->header.magic_food, TDB_MAGIC_FOOD) != 0) {
-               if (!(open_flags & O_CREAT) || tdb_new_database(tdb, hash_size) == -1) {
+       if (read(tdb->fd, &header, sizeof(header)) != sizeof(header)
+           /*
+            * Call strncmp() rather than strcmp() in case header.magic_food is
+            * not zero‐terminated. We’re still checking the full string for
+            * equality, as tdb_header::magic_food is larger than
+            * TDB_MAGIC_FOOD.
+            */
+           || strncmp(header.magic_food, TDB_MAGIC_FOOD, sizeof(header.magic_food)) != 0) {
+               if (!(open_flags & O_CREAT) ||
+                   tdb_new_database(tdb, &header, hash_size) == -1) {
                        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)))) {
+       } else if (header.version != TDB_VERSION
+                  && !(rev = (header.version==TDB_BYTEREV(TDB_VERSION)))) {
                /* wrong version */
                errno = EIO;
                goto fail;
        }
-       vp = (unsigned char *)&tdb->header.version;
+       vp = (unsigned char *)&header.version;
        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;
@@ -320,54 +542,155 @@ struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
                tdb->flags &= ~TDB_CONVERT;
        else {
                tdb->flags |= TDB_CONVERT;
-               tdb_convert(&tdb->header, sizeof(tdb->header));
+               tdb_convert(&header, sizeof(header));
        }
-       if (fstat(tdb->fd, &st) == -1)
+
+       /*
+        * We only use st.st_dev and st.st_ino from the raw fstat()
+        * call, everything else needs to use tdb_fstat() in order
+        * to skip tdb->hdr_ofs!
+        */
+       if (fstat(tdb->fd, &st) == -1) {
                goto fail;
+       }
+       tdb->device = st.st_dev;
+       tdb->inode = st.st_ino;
+       ZERO_STRUCT(st);
 
-       if (tdb->header.rwlocks != 0 &&
-           tdb->header.rwlocks != TDB_HASH_RWLOCK_MAGIC) {
+       if (header.rwlocks != 0 &&
+           header.rwlocks != TDB_FEATURE_FLAG_MAGIC &&
+           header.rwlocks != TDB_HASH_RWLOCK_MAGIC) {
                TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: spinlocks no longer supported\n"));
+               errno = ENOSYS;
+               goto fail;
+       }
+
+       if (header.hash_size == 0) {
+               TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: invalid database: 0 hash_size\n"));
+               errno = ENOSYS;
                goto fail;
        }
 
-       if ((tdb->header.magic1_hash == 0) && (tdb->header.magic2_hash == 0)) {
+       tdb->hash_size = header.hash_size;
+
+       if (header.rwlocks == TDB_FEATURE_FLAG_MAGIC) {
+               tdb->feature_flags = header.feature_flags;
+       }
+
+       if (tdb->feature_flags & ~TDB_SUPPORTED_FEATURE_FLAGS) {
+               TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: unsupported "
+                        "features in tdb %s: 0x%08x (supported: 0x%08x)\n",
+                        name, (unsigned)tdb->feature_flags,
+                        (unsigned)TDB_SUPPORTED_FEATURE_FLAGS));
+               errno = ENOSYS;
+               goto fail;
+       }
+
+       if (tdb->feature_flags & TDB_FEATURE_FLAG_MUTEX) {
+               if (!tdb_mutex_open_ok(tdb, &header)) {
+                       errno = EINVAL;
+                       goto fail;
+               }
+
+               /*
+                * We need to remember the hdr_ofs
+                * also for the TDB_NOLOCK case
+                * if the current library doesn't support
+                * mutex locking.
+                */
+               tdb->hdr_ofs = header.mutex_size;
+
+               if ((!(tdb_flags & TDB_CLEAR_IF_FIRST)) && (!tdb->read_only)) {
+                       /*
+                        * Open an existing mutexed tdb, but without
+                        * CLEAR_IF_FIRST. We need to initialize the
+                        * mutex array and keep the CLEAR_IF_FIRST
+                        * lock locked.
+                        */
+                       ret = tdb_nest_lock(tdb, ACTIVE_LOCK, F_WRLCK,
+                                           TDB_LOCK_NOWAIT|TDB_LOCK_PROBE);
+                       locked = (ret == 0);
+
+                       if (locked) {
+                               ret = tdb_mutex_init(tdb);
+                               if (ret == -1) {
+                                       TDB_LOG((tdb,
+                                                TDB_DEBUG_FATAL,
+                                                "tdb_open_ex: tdb_mutex_init "
+                                                "failed for ""%s: %s\n",
+                                                name, strerror(errno)));
+                                       goto fail;
+                               }
+                       }
+               }
+       }
+
+       if ((header.magic1_hash == 0) && (header.magic2_hash == 0)) {
                /* older TDB without magic hash references */
                tdb->hash_fn = tdb_old_hash;
-       } else if (!check_header_hash(tdb, !hash_fn, &magic1, &magic2)) {
+       } else if (!check_header_hash(tdb, &header, !hash_fn,
+                                     &magic1, &magic2)) {
                TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
                         "%s was not created with %s hash function we are using\n"
                         "magic1_hash[0x%08X %s 0x%08X] "
                         "magic2_hash[0x%08X %s 0x%08X]\n",
                         name, hash_alg,
-                        tdb->header.magic1_hash,
-                        (tdb->header.magic1_hash == magic1) ? "==" : "!=",
+                        header.magic1_hash,
+                        (header.magic1_hash == magic1) ? "==" : "!=",
                         magic1,
-                        tdb->header.magic2_hash,
-                        (tdb->header.magic2_hash == magic2) ? "==" : "!=",
+                        header.magic2_hash,
+                        (header.magic2_hash == magic2) ? "==" : "!=",
                         magic2));
                errno = EINVAL;
                goto fail;
        }
 
        /* Is it already in the open list?  If so, fail. */
-       if (tdb_already_open(st.st_dev, st.st_ino)) {
+       if (tdb_already_open(tdb->device, tdb->inode)) {
                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));
+                        name, (int)tdb->device, (int)tdb->inode));
                errno = EBUSY;
                goto fail;
        }
 
-       if (!(tdb->name = (char *)strdup(name))) {
-               errno = ENOMEM;
+       /*
+        * We had tdb_mmap(tdb) here before,
+        * but we need to use tdb_fstat(),
+        * which is triggered from tdb_oob() before calling tdb_mmap().
+        * As this skips tdb->hdr_ofs.
+        */
+       tdb->map_size = 0;
+       ret = tdb_oob(tdb, 0, 1, 0);
+       if (ret == -1) {
+               errno = EIO;
+               goto fail;
+       }
+
+       if (tdb->feature_flags & TDB_FEATURE_FLAG_MUTEX) {
+               if (!(tdb->flags & TDB_NOLOCK)) {
+                       ret = tdb_mutex_mmap(tdb);
+                       if (ret != 0) {
+                               goto fail;
+                       }
+               }
+       }
+
+       if (tdb->hash_size > UINT32_MAX/4) {
+               TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
+                        "hash size %"PRIu32" too large\n", tdb->hash_size));
+               errno = EINVAL;
+               goto fail;
+       }
+
+       ret = tdb_oob(tdb, FREELIST_TOP, 4*tdb->hash_size, 1);
+       if (ret == -1) {
+               TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_open_ex: "
+                        "hash size %"PRIu32" does not fit\n", tdb->hash_size));
+               errno = EINVAL;
                goto fail;
        }
 
-       tdb->map_size = st.st_size;
-       tdb->device = st.st_dev;
-       tdb->inode = st.st_ino;
-       tdb_mmap(tdb);
        if (locked) {
                if (tdb_nest_unlock(tdb, ACTIVE_LOCK, F_WRLCK, false) == -1) {
                        TDB_LOG((tdb, TDB_DEBUG_ERROR, "tdb_open_ex: "
@@ -376,15 +699,19 @@ struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
                        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 (locked || (tdb_flags & TDB_CLEAR_IF_FIRST)) {
+               /*
+                * 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_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT) == -1) {
+               ret = tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT);
+               if (ret == -1) {
                        goto fail;
                }
        }
@@ -419,6 +746,7 @@ struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
        }
        tdb->next = tdbs;
        tdbs = tdb;
+       errno = orig_errno;
        return tdb;
 
  fail:
@@ -436,11 +764,11 @@ struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
                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->lockrecs);
+       SAFE_FREE(tdb->name);
        SAFE_FREE(tdb);
        errno = save_errno;
        return NULL;
@@ -451,7 +779,7 @@ struct tdb_context *tdb_open_ex(const char *name, int hash_size, int tdb_flags,
  * Set the maximum number of dead records per hash chain
  */
 
-void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
+_PUBLIC_ void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
 {
        tdb->max_dead_records = max_dead;
 }
@@ -461,7 +789,7 @@ void tdb_set_max_dead(struct tdb_context *tdb, int max_dead)
  *
  * @returns -1 for error; 0 for success.
  **/
-int tdb_close(struct tdb_context *tdb)
+_PUBLIC_ int tdb_close(struct tdb_context *tdb)
 {
        struct tdb_context **i;
        int ret = 0;
@@ -477,6 +805,9 @@ int tdb_close(struct tdb_context *tdb)
                else
                        tdb_munmap(tdb);
        }
+
+       tdb_mutex_munmap(tdb);
+
        SAFE_FREE(tdb->name);
        if (tdb->fd != -1) {
                ret = close(tdb->fd);
@@ -501,14 +832,14 @@ int tdb_close(struct tdb_context *tdb)
        return ret;
 }
 
-/* register a loging function */
-void tdb_set_logging_function(struct tdb_context *tdb,
-                              const struct tdb_logging_context *log_ctx)
+/* register a logging function */
+_PUBLIC_ void tdb_set_logging_function(struct tdb_context *tdb,
+                                       const struct tdb_logging_context *log_ctx)
 {
         tdb->log = *log_ctx;
 }
 
-void *tdb_get_logging_private(struct tdb_context *tdb)
+_PUBLIC_ void *tdb_get_logging_private(struct tdb_context *tdb)
 {
        return tdb->log.log_private;
 }
@@ -548,6 +879,11 @@ static int tdb_reopen_internal(struct tdb_context *tdb, bool active_lock)
                TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: open failed (%s)\n", strerror(errno)));
                goto fail;
        }
+       /*
+        * We only use st.st_dev and st.st_ino from the raw fstat()
+        * call, everything else needs to use tdb_fstat() in order
+        * to skip tdb->hdr_ofs!
+        */
        if (fstat(tdb->fd, &st) != 0) {
                TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: fstat failed (%s)\n", strerror(errno)));
                goto fail;
@@ -556,12 +892,24 @@ static int tdb_reopen_internal(struct tdb_context *tdb, bool active_lock)
                TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: file dev/inode has changed!\n"));
                goto fail;
        }
-       tdb_mmap(tdb);
+       ZERO_STRUCT(st);
+
+       /*
+        * We had tdb_mmap(tdb) here before,
+        * but we need to use tdb_fstat(),
+        * which is triggered from tdb_oob() before calling tdb_mmap().
+        * As this skips tdb->hdr_ofs.
+        */
+       tdb->map_size = 0;
+       if (tdb_oob(tdb, 0, 1, 0) != 0) {
+               goto fail;
+       }
 #endif /* fake pread or pwrite */
 
        /* We may still think we hold the active lock. */
        tdb->num_lockrecs = 0;
        SAFE_FREE(tdb->lockrecs);
+       tdb->lockrecs_array_length = 0;
 
        if (active_lock && tdb_nest_lock(tdb, ACTIVE_LOCK, F_RDLCK, TDB_LOCK_WAIT) == -1) {
                TDB_LOG((tdb, TDB_DEBUG_FATAL, "tdb_reopen: failed to obtain active lock\n"));
@@ -577,18 +925,24 @@ fail:
 
 /* 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)
+_PUBLIC_ int tdb_reopen(struct tdb_context *tdb)
 {
-       return tdb_reopen_internal(tdb, tdb->flags & TDB_CLEAR_IF_FIRST);
+       bool active_lock;
+       active_lock = (tdb->flags & (TDB_CLEAR_IF_FIRST|TDB_MUTEX_LOCKING));
+
+       return tdb_reopen_internal(tdb, active_lock);
 }
 
 /* reopen all tdb's */
-int tdb_reopen_all(int parent_longlived)
+_PUBLIC_ 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);
+               bool active_lock;
+
+               active_lock =
+                       (tdb->flags & (TDB_CLEAR_IF_FIRST|TDB_MUTEX_LOCKING));
 
                /*
                 * If the parent is longlived (ie. a