a core file. We need to release this process id immediately
so that someone else can take over without getting sharing
violations */
- _exit(0);
+ _exit(1);
}
/*
return status;
}
+/*
+ * Are we active (i.e. not banned or stopped?)
+ */
+static bool ctdbd_working(struct ctdbd_connection *conn, uint32_t vnn)
+{
+ int32_t cstatus=-1;
+ NTSTATUS status;
+ TDB_DATA outdata;
+ struct ctdb_node_map *m;
+ uint32_t failure_flags;
+ bool ret = false;
+ int i;
+
+ status = ctdbd_control(conn, CTDB_CURRENT_NODE,
+ CTDB_CONTROL_GET_NODEMAP, 0, 0,
+ tdb_null, talloc_tos(), &outdata, &cstatus);
+ if (!NT_STATUS_IS_OK(status)) {
+ cluster_fatal("ctdbd_control failed\n");
+ }
+ if ((cstatus != 0) || (outdata.dptr == NULL)) {
+ DEBUG(2, ("Received invalid ctdb data\n"));
+ return false;
+ }
+
+ m = (struct ctdb_node_map *)outdata.dptr;
+
+ for (i=0; i<m->num; i++) {
+ if (vnn == m->nodes[i].pnn) {
+ break;
+ }
+ }
+
+ if (i == m->num) {
+ DEBUG(2, ("Did not find ourselves (node %d) in nodemap\n",
+ (int)vnn));
+ goto fail;
+ }
+
+ failure_flags = NODE_FLAGS_BANNED | NODE_FLAGS_DISCONNECTED
+ | NODE_FLAGS_PERMANENTLY_DISABLED | NODE_FLAGS_STOPPED;
+
+ if ((m->nodes[i].flags & failure_flags) != 0) {
+ DEBUG(2, ("Node has status %x, not active\n",
+ (int)m->nodes[i].flags));
+ goto fail;
+ }
+
+ ret = true;
+fail:
+ TALLOC_FREE(outdata.dptr);
+ return ret;;
+}
+
uint32 ctdbd_vnn(const struct ctdbd_connection *conn)
{
return conn->our_vnn;
* Do we have a complete ctdb packet in the queue?
*/
-static bool ctdb_req_complete(const struct data_blob *data,
+static bool ctdb_req_complete(const uint8_t *buf, size_t available,
size_t *length,
void *private_data)
{
uint32 msglen;
- if (data->length < sizeof(msglen)) {
+ if (available < sizeof(msglen)) {
return False;
}
- msglen = *((uint32 *)data->data);
+ msglen = *((uint32 *)buf);
DEBUG(10, ("msglen = %d\n", msglen));
cluster_fatal("ctdbd protocol error\n");
}
- if (data->length >= msglen) {
- *length = msglen;
- return True;
+ if (available < msglen) {
+ return false;
}
- return False;
+ *length = msglen;
+ return true;
}
/*
static void deferred_message_dispatch(struct event_context *event_ctx,
struct timed_event *te,
- const struct timeval *now,
+ struct timeval now,
void *private_data)
{
struct deferred_msg_state *state = talloc_get_type_abort(
* Pull a ctdb request out of the incoming packet queue
*/
-static NTSTATUS ctdb_req_pull(const struct data_blob *data,
+static NTSTATUS ctdb_req_pull(uint8_t *buf, size_t length,
void *private_data)
{
struct req_pull_state *state = (struct req_pull_state *)private_data;
- state->req = data_blob_talloc(state->mem_ctx, data->data,
- data->length);
- if (state->req.data == NULL) {
- return NT_STATUS_NO_MEMORY;
- }
+ state->req.data = talloc_move(state->mem_ctx, &buf);
+ state->req.length = length;
return NT_STATUS_OK;
}
blob = data_blob_const(msg->data, msg->datalen);
ndr_err = ndr_pull_struct_blob(
- &blob, result, result,
+ &blob, result, NULL, result,
(ndr_pull_flags_fn_t)ndr_pull_messaging_rec);
if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
return result;
}
+static NTSTATUS ctdb_packet_fd_read_sync(struct packet_context *ctx)
+{
+ struct timeval timeout;
+ struct timeval *ptimeout;
+
+ timeout = timeval_set(lp_ctdb_timeout(), 0);
+ ptimeout = (timeout.tv_sec != 0) ? &timeout : NULL;
+
+ return packet_fd_read_sync(ctx, ptimeout);
+}
+
/*
* Read a full ctdbd request. If we have a messaging context, defer incoming
* messages that might come in between.
again:
- status = packet_fd_read_sync(conn->pkt);
+ status = ctdb_packet_fd_read_sync(conn->pkt);
if (NT_STATUS_EQUAL(status, NT_STATUS_NETWORK_BUSY)) {
/* EAGAIN */
goto next_pkt;
}
+ if ((msg->srvid == CTDB_SRVID_RECONFIGURE)
+ || (msg->srvid == CTDB_SRVID_SAMBA_NOTIFY)) {
+
+ DEBUG(1, ("ctdb_read_req: Got %s message\n",
+ (msg->srvid == CTDB_SRVID_RECONFIGURE)
+ ? "cluster reconfigure" : "SAMBA_NOTIFY"));
+
+ messaging_send(conn->msg_ctx, procid_self(),
+ MSG_SMB_BRL_VALIDATE, &data_blob_null);
+ messaging_send(conn->msg_ctx, procid_self(),
+ MSG_DBWRAP_G_LOCK_RETRY,
+ &data_blob_null);
+ TALLOC_FREE(hdr);
+ goto next_pkt;
+ }
+
if (!(msg_state = TALLOC_P(NULL, struct deferred_msg_state))) {
DEBUG(0, ("talloc failed\n"));
TALLOC_FREE(hdr);
evt = event_add_timed(conn->msg_ctx->event_ctx,
conn->msg_ctx->event_ctx,
timeval_zero(),
- "deferred_message_dispatch",
deferred_message_dispatch,
msg_state);
if (evt == NULL) {
goto fail;
}
+ if (!ctdbd_working(conn, conn->our_vnn)) {
+ DEBUG(2, ("Node is not working, can not connect\n"));
+ status = NT_STATUS_INTERNAL_DB_ERROR;
+ goto fail;
+ }
+
generate_random_buffer((unsigned char *)&conn->rand_srvid,
sizeof(conn->rand_srvid));
goto fail;
}
+ status = register_with_ctdbd(conn, CTDB_SRVID_SAMBA_NOTIFY);
+ if (!NT_STATUS_IS_OK(status)) {
+ goto fail;
+ }
+
*pconn = conn;
return NT_STATUS_OK;
return status;
}
+struct messaging_context *ctdb_conn_msg_ctx(struct ctdbd_connection *conn)
+{
+ return conn->msg_ctx;
+}
+
+int ctdbd_conn_get_fd(struct ctdbd_connection *conn)
+{
+ return packet_get_fd(conn->pkt);
+}
+
/*
* Packet handler to receive and handle a ctdb message
*/
-static NTSTATUS ctdb_handle_message(const struct data_blob *data,
+static NTSTATUS ctdb_handle_message(uint8_t *buf, size_t length,
void *private_data)
{
struct ctdbd_connection *conn = talloc_get_type_abort(
struct ctdb_req_message *msg;
struct messaging_rec *msg_rec;
- msg = (struct ctdb_req_message *)data->data;
+ msg = (struct ctdb_req_message *)buf;
if (msg->hdr.operation != CTDB_REQ_MESSAGE) {
DEBUG(0, ("Received async msg of type %u, discarding\n",
msg->hdr.operation));
+ TALLOC_FREE(buf);
return NT_STATUS_INVALID_PARAMETER;
}
DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
conn->release_ip_handler((const char *)msg->data,
conn->release_ip_priv);
+ TALLOC_FREE(buf);
return NT_STATUS_OK;
}
SMB_ASSERT(conn->msg_ctx != NULL);
- if (msg->srvid == CTDB_SRVID_RECONFIGURE) {
+ if ((msg->srvid == CTDB_SRVID_RECONFIGURE)
+ || (msg->srvid == CTDB_SRVID_SAMBA_NOTIFY)){
DEBUG(0,("Got cluster reconfigure message\n"));
/*
- * when the cluster is reconfigured, we need to clean the brl
- * database
+ * when the cluster is reconfigured or someone of the
+ * family has passed away (SAMBA_NOTIFY), we need to
+ * clean the brl database
*/
messaging_send(conn->msg_ctx, procid_self(),
MSG_SMB_BRL_VALIDATE, &data_blob_null);
- /*
- * it's possible that we have just rejoined the cluster after
- * an outage. In that case our pending locks could have been
- * removed from the lockdb, so retry them once more
- */
- message_send_all(conn->msg_ctx, MSG_SMB_UNLOCK, NULL, 0, NULL);
+ messaging_send(conn->msg_ctx, procid_self(),
+ MSG_DBWRAP_G_LOCK_RETRY,
+ &data_blob_null);
+
+ TALLOC_FREE(buf);
return NT_STATUS_OK;
-
}
/* only messages to our pid or the broadcast are valid here */
if (msg->srvid != sys_getpid() && msg->srvid != MSG_SRVID_SAMBA) {
DEBUG(0,("Got unexpected message with srvid=%llu\n",
(unsigned long long)msg->srvid));
+ TALLOC_FREE(buf);
return NT_STATUS_OK;
}
- if (!(msg_rec = ctdb_pull_messaging_rec(NULL, data->length, msg))) {
+ if (!(msg_rec = ctdb_pull_messaging_rec(NULL, length, msg))) {
DEBUG(10, ("ctdb_pull_messaging_rec failed\n"));
+ TALLOC_FREE(buf);
return NT_STATUS_NO_MEMORY;
}
messaging_dispatch_rec(conn->msg_ctx, msg_rec);
TALLOC_FREE(msg_rec);
+ TALLOC_FREE(buf);
return NT_STATUS_OK;
}
}
ndr_err = ndr_push_struct_blob(
- &blob, mem_ctx, msg,
+ &blob, mem_ctx, NULL, msg,
(ndr_push_flags_fn_t)ndr_push_messaging_rec);
if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
req.opcode = opcode;
req.srvid = srvid;
req.datalen = data.dsize;
+ req.flags = flags;
DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
ctdb_packet_dump(&req.hdr);
if (flags & CTDB_CTRL_FLAG_NOREPLY) {
TALLOC_FREE(new_conn);
+ if (cstatus) {
+ *cstatus = 0;
+ }
return NT_STATUS_OK;
}
* Handle a traverse record coming in on the ctdbd connection
*/
-static NTSTATUS ctdb_traverse_handler(const struct data_blob *blob,
+static NTSTATUS ctdb_traverse_handler(uint8_t *buf, size_t length,
void *private_data)
{
struct ctdbd_traverse_state *state =
struct ctdb_rec_data *d;
TDB_DATA key, data;
- m = (struct ctdb_req_message *)blob->data;
+ m = (struct ctdb_req_message *)buf;
- if (blob->length < sizeof(*m) || m->hdr.length != blob->length) {
- DEBUG(0, ("Got invalid message of length %d\n",
- (int)blob->length));
+ if (length < sizeof(*m) || m->hdr.length != length) {
+ DEBUG(0, ("Got invalid message of length %d\n", (int)length));
+ TALLOC_FREE(buf);
return NT_STATUS_UNEXPECTED_IO_ERROR;
}
if (m->datalen < sizeof(uint32_t) || m->datalen != d->length) {
DEBUG(0, ("Got invalid traverse data of length %d\n",
(int)m->datalen));
+ TALLOC_FREE(buf);
return NT_STATUS_UNEXPECTED_IO_ERROR;
}
if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
DEBUG(0, ("Got invalid ltdb header length %d\n",
(int)data.dsize));
+ TALLOC_FREE(buf);
return NT_STATUS_UNEXPECTED_IO_ERROR;
}
data.dsize -= sizeof(struct ctdb_ltdb_header);
state->fn(key, data, state->private_data);
}
+ TALLOC_FREE(buf);
return NT_STATUS_OK;
}
struct ctdbd_traverse_state state;
status = ctdbd_init_connection(NULL, &conn);
+ if (!NT_STATUS_IS_OK(status)) {
+ DEBUG(0, ("ctdbd_init_connection failed: %s\n",
+ nt_errstr(status)));
+ return status;
+ }
t.db_id = db_id;
t.srvid = conn->rand_srvid;
break;
}
- status = packet_fd_read_sync(conn->pkt);
+ status = ctdb_packet_fd_read_sync(conn->pkt);
if (NT_STATUS_EQUAL(status, NT_STATUS_RETRY)) {
/*
if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
status = NT_STATUS_OK;
+ break;
}
if (!NT_STATUS_IS_OK(status)) {
return status;
}
+/*
+ This is used to canonicalize a ctdb_sock_addr structure.
+*/
+static void smbd_ctdb_canonicalize_ip(const struct sockaddr_storage *in,
+ struct sockaddr_storage *out)
+{
+ memcpy(out, in, sizeof (*out));
+
+#ifdef HAVE_IPV6
+ if (in->ss_family == AF_INET6) {
+ const char prefix[12] = { 0,0,0,0,0,0,0,0,0,0,0xff,0xff };
+ const struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)in;
+ struct sockaddr_in *out4 = (struct sockaddr_in *)out;
+ if (memcmp(&in6->sin6_addr, prefix, 12) == 0) {
+ memset(out, 0, sizeof(*out));
+#ifdef HAVE_SOCK_SIN_LEN
+ out4->sin_len = sizeof(*out);
+#endif
+ out4->sin_family = AF_INET;
+ out4->sin_port = in6->sin6_port;
+ memcpy(&out4->sin_addr, &in6->sin6_addr.s6_addr32[3], 4);
+ }
+ }
+#endif
+}
+
/*
* Register us as a server for a particular tcp connection
*/
NTSTATUS ctdbd_register_ips(struct ctdbd_connection *conn,
- const struct sockaddr_in *server,
- const struct sockaddr_in *client,
+ const struct sockaddr_storage *_server,
+ const struct sockaddr_storage *_client,
void (*release_ip_handler)(const char *ip_addr,
void *private_data),
void *private_data)
{
- struct ctdb_control_tcp p;
+ /*
+ * we still use ctdb_control_tcp for ipv4
+ * because we want to work against older ctdb
+ * versions at runtime
+ */
+ struct ctdb_control_tcp p4;
+#ifdef HAVE_STRUCT_CTDB_CONTROL_TCP_ADDR
+ struct ctdb_control_tcp_addr p;
+#endif
TDB_DATA data;
NTSTATUS status;
+ struct sockaddr_storage client;
+ struct sockaddr_storage server;
/*
* Only one connection so far
*/
SMB_ASSERT(conn->release_ip_handler == NULL);
+ smbd_ctdb_canonicalize_ip(_client, &client);
+ smbd_ctdb_canonicalize_ip(_server, &server);
+
+ switch (client.ss_family) {
+ case AF_INET:
+ p4.dest = *(struct sockaddr_in *)&server;
+ p4.src = *(struct sockaddr_in *)&client;
+ data.dptr = (uint8_t *)&p4;
+ data.dsize = sizeof(p4);
+ break;
+#ifdef HAVE_STRUCT_CTDB_CONTROL_TCP_ADDR
+ case AF_INET6:
+ p.dest.ip6 = *(struct sockaddr_in6 *)&server;
+ p.src.ip6 = *(struct sockaddr_in6 *)&client;
+ data.dptr = (uint8_t *)&p;
+ data.dsize = sizeof(p);
+ break;
+#endif
+ default:
+ return NT_STATUS_INTERNAL_ERROR;
+ }
+
+ conn->release_ip_handler = release_ip_handler;
+
/*
* We want to be told about IP releases
*/
return status;
}
- p.dest = *server;
- p.src = *client;
-
/*
* inform ctdb of our tcp connection, so if IP takeover happens ctdb
* can send an extra ack to trigger a reset for our client, so it
* immediately reconnects
*/
- data.dptr = (uint8_t *)&p;
- data.dsize = sizeof(p);
-
return ctdbd_control(conn, CTDB_CURRENT_NODE,
CTDB_CONTROL_TCP_CLIENT, 0,
CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL, NULL);
return register_with_ctdbd(conn, CTDB_SRVID_RECONFIGURE);
}
-/*
- persistent call. Used to start, store or cancel persistent updates.
- */
-static NTSTATUS ctdbd_persistent_call(struct ctdbd_connection *conn, uint32_t opcode, uint32_t db_id, TDB_DATA key, TDB_DATA data)
-{
- int cstatus=0;
- struct ctdb_rec_data *rec;
- TDB_DATA recdata;
- size_t length;
- NTSTATUS status;
-
- length = offsetof(struct ctdb_rec_data, data) + key.dsize + data.dsize;
-
- rec = (struct ctdb_rec_data *)talloc_size(conn, length);
- NT_STATUS_HAVE_NO_MEMORY(rec);
-
- rec->length = length;
- rec->reqid = db_id;
- rec->keylen = key.dsize;
- rec->datalen= data.dsize;
- memcpy(&rec->data[0], key.dptr, key.dsize);
- memcpy(&rec->data[key.dsize], data.dptr, data.dsize);
-
- recdata.dptr = (uint8_t *)rec;
- recdata.dsize = length;
-
- status = ctdbd_control(conn, CTDB_CURRENT_NODE, opcode,
- 0, 0, recdata, NULL, NULL, &cstatus);
- if (cstatus != 0) {
- return NT_STATUS_INTERNAL_DB_CORRUPTION;
- }
- return status;
-}
-
-/*
- persistent store. Used when we update a record in a persistent database
- */
-NTSTATUS ctdbd_persistent_store(struct ctdbd_connection *conn, uint32_t db_id, TDB_DATA key, TDB_DATA data)
-{
- return ctdbd_persistent_call(conn,
- CTDB_CONTROL_PERSISTENT_STORE,
- db_id, key, data);
-}
-
-/*
- tell the ctdb daemon that we are starting a persistent update operation.
- If we terminate/disconnect from the daemon without first performing
- either a persistent_store or a cancel ctdbd will perform recovery.
- */
-NTSTATUS ctdbd_start_persistent_update(struct ctdbd_connection *conn, uint32_t db_id, TDB_DATA key, TDB_DATA data)
-{
- return ctdbd_persistent_call(conn,
- CTDB_CONTROL_START_PERSISTENT_UPDATE,
- db_id, key, data);
-
-}
-
-/*
- Cancel a persistent update operation. This is used if we have started a
- persistent update but we want to abort it before we have made changes to
- the tdb database.
- */
-NTSTATUS ctdbd_cancel_persistent_update(struct ctdbd_connection *conn, uint32_t db_id, TDB_DATA key, TDB_DATA data)
-{
- return ctdbd_persistent_call(conn,
- CTDB_CONTROL_CANCEL_PERSISTENT_UPDATE,
- db_id, key, data);
-}
-
/*
call a control on the local node
*/
return ctdbd_control(conn, CTDB_CURRENT_NODE, opcode, srvid, flags, data, mem_ctx, outdata, cstatus);
}
+NTSTATUS ctdb_watch_us(struct ctdbd_connection *conn)
+{
+ struct ctdb_client_notify_register reg_data;
+ size_t struct_len;
+ NTSTATUS status;
+ int cstatus;
+
+ reg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
+ reg_data.len = 1;
+ reg_data.notify_data[0] = 0;
+
+ struct_len = offsetof(struct ctdb_client_notify_register,
+ notify_data) + reg_data.len;
+
+ status = ctdbd_control_local(
+ conn, CTDB_CONTROL_REGISTER_NOTIFY, conn->rand_srvid, 0,
+ make_tdb_data((uint8_t *)®_data, struct_len),
+ NULL, NULL, &cstatus);
+ if (!NT_STATUS_IS_OK(status)) {
+ DEBUG(1, ("ctdbd_control_local failed: %s\n",
+ nt_errstr(status)));
+ }
+ return status;
+}
+
+NTSTATUS ctdb_unwatch(struct ctdbd_connection *conn)
+{
+ struct ctdb_client_notify_deregister dereg_data;
+ NTSTATUS status;
+ int cstatus;
+ dereg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
+ status = ctdbd_control_local(
+ conn, CTDB_CONTROL_DEREGISTER_NOTIFY, conn->rand_srvid, 0,
+ make_tdb_data((uint8_t *)&dereg_data, sizeof(dereg_data)),
+ NULL, NULL, &cstatus);
+ if (!NT_STATUS_IS_OK(status)) {
+ DEBUG(1, ("ctdbd_control_local failed: %s\n",
+ nt_errstr(status)));
+ }
+ return status;
+}
#else