2 Unix SMB/CIFS implementation.
3 Samba internal messaging functions
4 Copyright (C) 2007 by Volker Lendecke
5 Copyright (C) 2007 by Andrew Tridgell
7 This program is free software; you can redistribute it and/or modify
8 it under the terms of the GNU General Public License as published by
9 the Free Software Foundation; either version 3 of the License, or
10 (at your option) any later version.
12 This program is distributed in the hope that it will be useful,
13 but WITHOUT ANY WARRANTY; without even the implied warranty of
14 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15 GNU General Public License for more details.
17 You should have received a copy of the GNU General Public License
18 along with this program. If not, see <http://www.gnu.org/licenses/>.
24 #ifdef CLUSTER_SUPPORT
26 #include "ctdbd_conn.h"
27 #include "ctdb_packet.h"
31 * It is not possible to include ctdb.h and tdb_compat.h (included via
32 * some other include above) without warnings. This fixes those
40 #ifdef typesafe_cb_preargs
41 #undef typesafe_cb_preargs
44 #ifdef typesafe_cb_postargs
45 #undef typesafe_cb_postargs
48 /* paths to these include files come from --with-ctdb= in configure */
51 #include "ctdb_private.h"
53 struct ctdbd_connection {
54 struct messaging_context *msg_ctx;
58 struct ctdb_packet_context *pkt;
61 void (*release_ip_handler)(const char *ip_addr, void *private_data);
62 void *release_ip_priv;
65 static uint32_t ctdbd_next_reqid(struct ctdbd_connection *conn)
68 if (conn->reqid == 0) {
74 static NTSTATUS ctdbd_control(struct ctdbd_connection *conn,
75 uint32_t vnn, uint32 opcode,
76 uint64_t srvid, uint32_t flags, TDB_DATA data,
77 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
81 * exit on fatal communications errors with the ctdbd daemon
83 static void cluster_fatal(const char *why)
85 DEBUG(0,("cluster fatal event: %s - exiting immediately\n", why));
86 /* we don't use smb_panic() as we don't want to delay to write
87 a core file. We need to release this process id immediately
88 so that someone else can take over without getting sharing
96 static void ctdb_packet_dump(struct ctdb_req_header *hdr)
98 if (DEBUGLEVEL < 10) {
101 DEBUGADD(10, ("len=%d, magic=%x, vers=%d, gen=%d, op=%d, reqid=%d\n",
102 (int)hdr->length, (int)hdr->ctdb_magic,
103 (int)hdr->ctdb_version, (int)hdr->generation,
104 (int)hdr->operation, (int)hdr->reqid));
108 * Register a srvid with ctdbd
110 static NTSTATUS register_with_ctdbd(struct ctdbd_connection *conn,
115 return ctdbd_control(conn, CTDB_CURRENT_NODE,
116 CTDB_CONTROL_REGISTER_SRVID, srvid, 0,
117 tdb_null, NULL, NULL, &cstatus);
121 * get our vnn from the cluster
123 static NTSTATUS get_cluster_vnn(struct ctdbd_connection *conn, uint32 *vnn)
127 status = ctdbd_control(conn,
128 CTDB_CURRENT_NODE, CTDB_CONTROL_GET_PNN, 0, 0,
129 tdb_null, NULL, NULL, &cstatus);
130 if (!NT_STATUS_IS_OK(status)) {
131 cluster_fatal("ctdbd_control failed\n");
133 *vnn = (uint32_t)cstatus;
138 * Are we active (i.e. not banned or stopped?)
140 static bool ctdbd_working(struct ctdbd_connection *conn, uint32_t vnn)
145 struct ctdb_node_map *m;
146 uint32_t failure_flags;
150 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
151 CTDB_CONTROL_GET_NODEMAP, 0, 0,
152 tdb_null, talloc_tos(), &outdata, &cstatus);
153 if (!NT_STATUS_IS_OK(status)) {
154 cluster_fatal("ctdbd_control failed\n");
156 if ((cstatus != 0) || (outdata.dptr == NULL)) {
157 DEBUG(2, ("Received invalid ctdb data\n"));
161 m = (struct ctdb_node_map *)outdata.dptr;
163 for (i=0; i<m->num; i++) {
164 if (vnn == m->nodes[i].pnn) {
170 DEBUG(2, ("Did not find ourselves (node %d) in nodemap\n",
175 failure_flags = NODE_FLAGS_BANNED | NODE_FLAGS_DISCONNECTED
176 | NODE_FLAGS_PERMANENTLY_DISABLED | NODE_FLAGS_STOPPED;
178 if ((m->nodes[i].flags & failure_flags) != 0) {
179 DEBUG(2, ("Node has status %x, not active\n",
180 (int)m->nodes[i].flags));
186 TALLOC_FREE(outdata.dptr);
190 uint32 ctdbd_vnn(const struct ctdbd_connection *conn)
192 return conn->our_vnn;
196 * Get us a ctdb connection
199 static NTSTATUS ctdbd_connect(TALLOC_CTX *mem_ctx,
200 struct ctdb_packet_context **presult)
202 struct ctdb_packet_context *result;
203 const char *sockname = lp_ctdbd_socket();
204 struct sockaddr_un addr;
207 if (!sockname || !*sockname) {
208 sockname = CTDB_PATH;
211 fd = socket(AF_UNIX, SOCK_STREAM, 0);
213 DEBUG(3, ("Could not create socket: %s\n", strerror(errno)));
214 return map_nt_error_from_unix(errno);
218 addr.sun_family = AF_UNIX;
219 strncpy(addr.sun_path, sockname, sizeof(addr.sun_path));
221 if (sys_connect(fd, (struct sockaddr *)(void *)&addr) == -1) {
222 DEBUG(1, ("connect(%s) failed: %s\n", sockname,
225 return map_nt_error_from_unix(errno);
228 if (!(result = ctdb_packet_init(mem_ctx, fd))) {
230 return NT_STATUS_NO_MEMORY;
238 * Do we have a complete ctdb packet in the queue?
241 static bool ctdb_req_complete(const uint8_t *buf, size_t available,
247 if (available < sizeof(msglen)) {
251 msglen = *((uint32 *)buf);
253 DEBUG(10, ("msglen = %d\n", msglen));
255 if (msglen < sizeof(struct ctdb_req_header)) {
256 DEBUG(0, ("Got invalid msglen: %d, expected at least %d for "
257 "the req_header\n", (int)msglen,
258 (int)sizeof(struct ctdb_req_header)));
259 cluster_fatal("ctdbd protocol error\n");
262 if (available < msglen) {
271 * State necessary to defer an incoming message while we are waiting for a
275 struct deferred_msg_state {
276 struct messaging_context *msg_ctx;
277 struct messaging_rec *rec;
281 * Timed event handler for the deferred message
284 static void deferred_message_dispatch(struct event_context *event_ctx,
285 struct timed_event *te,
289 struct deferred_msg_state *state = talloc_get_type_abort(
290 private_data, struct deferred_msg_state);
292 messaging_dispatch_rec(state->msg_ctx, state->rec);
297 struct req_pull_state {
303 * Pull a ctdb request out of the incoming ctdb_packet queue
306 static NTSTATUS ctdb_req_pull(uint8_t *buf, size_t length,
309 struct req_pull_state *state = (struct req_pull_state *)private_data;
311 state->req.data = talloc_move(state->mem_ctx, &buf);
312 state->req.length = length;
317 * Fetch a messaging_rec from an incoming ctdb style message
320 static struct messaging_rec *ctdb_pull_messaging_rec(TALLOC_CTX *mem_ctx,
321 size_t overall_length,
322 struct ctdb_req_message *msg)
324 struct messaging_rec *result;
326 enum ndr_err_code ndr_err;
328 if ((overall_length < offsetof(struct ctdb_req_message, data))
330 < offsetof(struct ctdb_req_message, data) + msg->datalen)) {
332 cluster_fatal("got invalid msg length");
335 if (!(result = talloc(mem_ctx, struct messaging_rec))) {
336 DEBUG(0, ("talloc failed\n"));
340 blob = data_blob_const(msg->data, msg->datalen);
342 ndr_err = ndr_pull_struct_blob(
343 &blob, result, result,
344 (ndr_pull_flags_fn_t)ndr_pull_messaging_rec);
346 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
347 DEBUG(0, ("ndr_pull_struct_blob failed: %s\n",
348 ndr_errstr(ndr_err)));
353 if (DEBUGLEVEL >= 10) {
354 DEBUG(10, ("ctdb_pull_messaging_rec:\n"));
355 NDR_PRINT_DEBUG(messaging_rec, result);
361 static NTSTATUS ctdb_packet_fd_read_sync(struct ctdb_packet_context *ctx)
363 int timeout = lp_ctdb_timeout();
368 return ctdb_packet_fd_read_sync_timeout(ctx, timeout);
372 * Read a full ctdbd request. If we have a messaging context, defer incoming
373 * messages that might come in between.
376 static NTSTATUS ctdb_read_req(struct ctdbd_connection *conn, uint32 reqid,
377 TALLOC_CTX *mem_ctx, void *result)
379 struct ctdb_req_header *hdr;
380 struct req_pull_state state;
385 status = ctdb_packet_fd_read_sync(conn->pkt);
387 if (NT_STATUS_EQUAL(status, NT_STATUS_NETWORK_BUSY)) {
390 } else if (NT_STATUS_EQUAL(status, NT_STATUS_RETRY)) {
395 if (!NT_STATUS_IS_OK(status)) {
396 DEBUG(0, ("ctdb_packet_fd_read failed: %s\n", nt_errstr(status)));
397 cluster_fatal("ctdbd died\n");
403 state.mem_ctx = mem_ctx;
405 if (!ctdb_packet_handler(conn->pkt, ctdb_req_complete, ctdb_req_pull,
410 DEBUG(10, ("not enough data from ctdb socket, retrying\n"));
414 if (!NT_STATUS_IS_OK(status)) {
415 DEBUG(0, ("Could not read ctdb_packet: %s\n", nt_errstr(status)));
416 cluster_fatal("ctdbd died\n");
419 hdr = (struct ctdb_req_header *)state.req.data;
421 DEBUG(10, ("Received ctdb packet\n"));
422 ctdb_packet_dump(hdr);
424 if (hdr->operation == CTDB_REQ_MESSAGE) {
425 struct timed_event *evt;
426 struct deferred_msg_state *msg_state;
427 struct ctdb_req_message *msg = (struct ctdb_req_message *)hdr;
429 if (conn->msg_ctx == NULL) {
430 DEBUG(1, ("Got a message without having a msg ctx, "
431 "dropping msg %llu\n",
432 (long long unsigned)msg->srvid));
436 if ((conn->release_ip_handler != NULL)
437 && (msg->srvid == CTDB_SRVID_RELEASE_IP)) {
438 /* must be dispatched immediately */
439 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
440 conn->release_ip_handler((const char *)msg->data,
441 conn->release_ip_priv);
446 if ((msg->srvid == CTDB_SRVID_RECONFIGURE)
447 || (msg->srvid == CTDB_SRVID_SAMBA_NOTIFY)) {
449 DEBUG(1, ("ctdb_read_req: Got %s message\n",
450 (msg->srvid == CTDB_SRVID_RECONFIGURE)
451 ? "cluster reconfigure" : "SAMBA_NOTIFY"));
453 messaging_send(conn->msg_ctx,
454 messaging_server_id(conn->msg_ctx),
455 MSG_SMB_BRL_VALIDATE, &data_blob_null);
456 messaging_send(conn->msg_ctx,
457 messaging_server_id(conn->msg_ctx),
458 MSG_DBWRAP_G_LOCK_RETRY,
464 msg_state = talloc(NULL, struct deferred_msg_state);
465 if (msg_state == NULL) {
466 DEBUG(0, ("talloc failed\n"));
471 if (!(msg_state->rec = ctdb_pull_messaging_rec(
472 msg_state, state.req.length, msg))) {
473 DEBUG(0, ("ctdbd_pull_messaging_rec failed\n"));
474 TALLOC_FREE(msg_state);
481 msg_state->msg_ctx = conn->msg_ctx;
484 * We're waiting for a call reply, but an async message has
485 * crossed. Defer dispatching to the toplevel event loop.
487 evt = event_add_timed(conn->msg_ctx->event_ctx,
488 conn->msg_ctx->event_ctx,
490 deferred_message_dispatch,
493 DEBUG(0, ("event_add_timed failed\n"));
494 TALLOC_FREE(msg_state);
502 if ((reqid != 0) && (hdr->reqid != reqid)) {
503 /* we got the wrong reply */
504 DEBUG(0,("Discarding mismatched ctdb reqid %u should have "
505 "been %u\n", hdr->reqid, reqid));
510 *((void **)result) = talloc_move(mem_ctx, &hdr);
516 * Get us a ctdbd connection
519 static NTSTATUS ctdbd_init_connection(TALLOC_CTX *mem_ctx,
520 struct ctdbd_connection **pconn)
522 struct ctdbd_connection *conn;
525 if (!(conn = talloc_zero(mem_ctx, struct ctdbd_connection))) {
526 DEBUG(0, ("talloc failed\n"));
527 return NT_STATUS_NO_MEMORY;
530 status = ctdbd_connect(conn, &conn->pkt);
532 if (!NT_STATUS_IS_OK(status)) {
533 DEBUG(10, ("ctdbd_connect failed: %s\n", nt_errstr(status)));
537 status = get_cluster_vnn(conn, &conn->our_vnn);
539 if (!NT_STATUS_IS_OK(status)) {
540 DEBUG(10, ("get_cluster_vnn failed: %s\n", nt_errstr(status)));
544 if (!ctdbd_working(conn, conn->our_vnn)) {
545 DEBUG(2, ("Node is not working, can not connect\n"));
546 status = NT_STATUS_INTERNAL_DB_ERROR;
550 generate_random_buffer((unsigned char *)&conn->rand_srvid,
551 sizeof(conn->rand_srvid));
553 status = register_with_ctdbd(conn, conn->rand_srvid);
555 if (!NT_STATUS_IS_OK(status)) {
556 DEBUG(5, ("Could not register random srvid: %s\n",
570 * Get us a ctdbd connection and register us as a process
573 NTSTATUS ctdbd_messaging_connection(TALLOC_CTX *mem_ctx,
574 struct ctdbd_connection **pconn)
576 struct ctdbd_connection *conn;
579 status = ctdbd_init_connection(mem_ctx, &conn);
581 if (!NT_STATUS_IS_OK(status)) {
585 status = register_with_ctdbd(conn, (uint64_t)sys_getpid());
586 if (!NT_STATUS_IS_OK(status)) {
590 status = register_with_ctdbd(conn, MSG_SRVID_SAMBA);
591 if (!NT_STATUS_IS_OK(status)) {
595 status = register_with_ctdbd(conn, CTDB_SRVID_SAMBA_NOTIFY);
596 if (!NT_STATUS_IS_OK(status)) {
608 struct messaging_context *ctdb_conn_msg_ctx(struct ctdbd_connection *conn)
610 return conn->msg_ctx;
613 int ctdbd_conn_get_fd(struct ctdbd_connection *conn)
615 return ctdb_packet_get_fd(conn->pkt);
619 * Packet handler to receive and handle a ctdb message
621 static NTSTATUS ctdb_handle_message(uint8_t *buf, size_t length,
624 struct ctdbd_connection *conn = talloc_get_type_abort(
625 private_data, struct ctdbd_connection);
626 struct ctdb_req_message *msg;
627 struct messaging_rec *msg_rec;
629 msg = (struct ctdb_req_message *)buf;
631 if (msg->hdr.operation != CTDB_REQ_MESSAGE) {
632 DEBUG(0, ("Received async msg of type %u, discarding\n",
633 msg->hdr.operation));
635 return NT_STATUS_INVALID_PARAMETER;
638 if ((conn->release_ip_handler != NULL)
639 && (msg->srvid == CTDB_SRVID_RELEASE_IP)) {
640 /* must be dispatched immediately */
641 DEBUG(10, ("received CTDB_SRVID_RELEASE_IP\n"));
642 conn->release_ip_handler((const char *)msg->data,
643 conn->release_ip_priv);
648 SMB_ASSERT(conn->msg_ctx != NULL);
650 if ((msg->srvid == CTDB_SRVID_RECONFIGURE)
651 || (msg->srvid == CTDB_SRVID_SAMBA_NOTIFY)){
652 DEBUG(0,("Got cluster reconfigure message\n"));
654 * when the cluster is reconfigured or someone of the
655 * family has passed away (SAMBA_NOTIFY), we need to
656 * clean the brl database
658 messaging_send(conn->msg_ctx,
659 messaging_server_id(conn->msg_ctx),
660 MSG_SMB_BRL_VALIDATE, &data_blob_null);
662 messaging_send(conn->msg_ctx,
663 messaging_server_id(conn->msg_ctx),
664 MSG_DBWRAP_G_LOCK_RETRY,
671 /* only messages to our pid or the broadcast are valid here */
672 if (msg->srvid != sys_getpid() && msg->srvid != MSG_SRVID_SAMBA) {
673 DEBUG(0,("Got unexpected message with srvid=%llu\n",
674 (unsigned long long)msg->srvid));
679 if (!(msg_rec = ctdb_pull_messaging_rec(NULL, length, msg))) {
680 DEBUG(10, ("ctdb_pull_messaging_rec failed\n"));
682 return NT_STATUS_NO_MEMORY;
685 messaging_dispatch_rec(conn->msg_ctx, msg_rec);
687 TALLOC_FREE(msg_rec);
693 * The ctdbd socket is readable asynchronuously
696 static void ctdbd_socket_handler(struct event_context *event_ctx,
697 struct fd_event *event,
701 struct ctdbd_connection *conn = talloc_get_type_abort(
702 private_data, struct ctdbd_connection);
706 status = ctdb_packet_fd_read(conn->pkt);
708 if (!NT_STATUS_IS_OK(status)) {
709 DEBUG(0, ("packet_fd_read failed: %s\n", nt_errstr(status)));
710 cluster_fatal("ctdbd died\n");
713 while (ctdb_packet_handler(conn->pkt, ctdb_req_complete,
714 ctdb_handle_message, conn, &status)) {
715 if (!NT_STATUS_IS_OK(status)) {
716 DEBUG(10, ("could not handle incoming message: %s\n",
723 * Prepare a ctdbd connection to receive messages
726 NTSTATUS ctdbd_register_msg_ctx(struct ctdbd_connection *conn,
727 struct messaging_context *msg_ctx)
729 SMB_ASSERT(conn->msg_ctx == NULL);
730 SMB_ASSERT(conn->fde == NULL);
732 if (!(conn->fde = event_add_fd(msg_ctx->event_ctx, conn,
733 ctdb_packet_get_fd(conn->pkt),
735 ctdbd_socket_handler,
737 DEBUG(0, ("event_add_fd failed\n"));
738 return NT_STATUS_NO_MEMORY;
741 conn->msg_ctx = msg_ctx;
747 * Send a messaging message across a ctdbd
750 NTSTATUS ctdbd_messaging_send(struct ctdbd_connection *conn,
751 uint32 dst_vnn, uint64 dst_srvid,
752 struct messaging_rec *msg)
754 struct ctdb_req_message r;
758 enum ndr_err_code ndr_err;
760 if (!(mem_ctx = talloc_init("ctdbd_messaging_send"))) {
761 DEBUG(0, ("talloc failed\n"));
762 return NT_STATUS_NO_MEMORY;
765 ndr_err = ndr_push_struct_blob(
767 (ndr_push_flags_fn_t)ndr_push_messaging_rec);
769 if (!NDR_ERR_CODE_IS_SUCCESS(ndr_err)) {
770 DEBUG(0, ("ndr_push_struct_blob failed: %s\n",
771 ndr_errstr(ndr_err)));
772 status = ndr_map_error2ntstatus(ndr_err);
776 r.hdr.length = offsetof(struct ctdb_req_message, data) + blob.length;
777 r.hdr.ctdb_magic = CTDB_MAGIC;
778 r.hdr.ctdb_version = CTDB_VERSION;
779 r.hdr.generation = 1;
780 r.hdr.operation = CTDB_REQ_MESSAGE;
781 r.hdr.destnode = dst_vnn;
782 r.hdr.srcnode = conn->our_vnn;
785 r.datalen = blob.length;
787 DEBUG(10, ("ctdbd_messaging_send: Sending ctdb packet\n"));
788 ctdb_packet_dump(&r.hdr);
790 status = ctdb_packet_send(
792 data_blob_const(&r, offsetof(struct ctdb_req_message, data)),
795 if (!NT_STATUS_IS_OK(status)) {
796 DEBUG(0, ("ctdb_packet_send failed: %s\n", nt_errstr(status)));
800 status = ctdb_packet_flush(conn->pkt);
802 if (!NT_STATUS_IS_OK(status)) {
803 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
804 cluster_fatal("cluster dispatch daemon msg write error\n");
807 status = NT_STATUS_OK;
809 TALLOC_FREE(mem_ctx);
814 * send/recv a generic ctdb control message
816 static NTSTATUS ctdbd_control(struct ctdbd_connection *conn,
817 uint32_t vnn, uint32 opcode,
818 uint64_t srvid, uint32_t flags,
820 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
823 struct ctdb_req_control req;
824 struct ctdb_reply_control *reply = NULL;
825 struct ctdbd_connection *new_conn = NULL;
829 status = ctdbd_init_connection(NULL, &new_conn);
831 if (!NT_STATUS_IS_OK(status)) {
832 DEBUG(10, ("Could not init temp connection: %s\n",
841 req.hdr.length = offsetof(struct ctdb_req_control, data) + data.dsize;
842 req.hdr.ctdb_magic = CTDB_MAGIC;
843 req.hdr.ctdb_version = CTDB_VERSION;
844 req.hdr.operation = CTDB_REQ_CONTROL;
845 req.hdr.reqid = ctdbd_next_reqid(conn);
846 req.hdr.destnode = vnn;
849 req.datalen = data.dsize;
852 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
853 ctdb_packet_dump(&req.hdr);
855 status = ctdb_packet_send(
857 data_blob_const(&req, offsetof(struct ctdb_req_control, data)),
858 data_blob_const(data.dptr, data.dsize));
860 if (!NT_STATUS_IS_OK(status)) {
861 DEBUG(3, ("ctdb_packet_send failed: %s\n", nt_errstr(status)));
865 status = ctdb_packet_flush(conn->pkt);
867 if (!NT_STATUS_IS_OK(status)) {
868 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
869 cluster_fatal("cluster dispatch daemon control write error\n");
872 if (flags & CTDB_CTRL_FLAG_NOREPLY) {
873 TALLOC_FREE(new_conn);
880 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
882 if (!NT_STATUS_IS_OK(status)) {
883 DEBUG(10, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
887 if (reply->hdr.operation != CTDB_REPLY_CONTROL) {
888 DEBUG(0, ("received invalid reply\n"));
893 if (!(outdata->dptr = (uint8 *)talloc_memdup(
894 mem_ctx, reply->data, reply->datalen))) {
896 return NT_STATUS_NO_MEMORY;
898 outdata->dsize = reply->datalen;
901 (*cstatus) = reply->status;
904 status = NT_STATUS_OK;
907 TALLOC_FREE(new_conn);
913 * see if a remote process exists
915 bool ctdbd_process_exists(struct ctdbd_connection *conn, uint32 vnn, pid_t pid)
923 if (!ctdb_processes_exist(conn, &id, 1, &result)) {
924 DEBUG(10, ("ctdb_processes_exist failed\n"));
930 bool ctdb_processes_exist(struct ctdbd_connection *conn,
931 const struct server_id *pids, int num_pids,
934 TALLOC_CTX *frame = talloc_stackframe();
940 reqids = talloc_array(talloc_tos(), uint32_t, num_pids);
941 if (reqids == NULL) {
945 for (i=0; i<num_pids; i++) {
946 struct ctdb_req_control req;
949 reqids[i] = ctdbd_next_reqid(conn);
953 DEBUG(10, ("Requesting PID %d/%d, reqid=%d\n",
954 (int)pids[i].vnn, (int)pids[i].pid,
957 req.hdr.length = offsetof(struct ctdb_req_control, data);
958 req.hdr.length += sizeof(pid_t);
959 req.hdr.ctdb_magic = CTDB_MAGIC;
960 req.hdr.ctdb_version = CTDB_VERSION;
961 req.hdr.operation = CTDB_REQ_CONTROL;
962 req.hdr.reqid = reqids[i];
963 req.hdr.destnode = pids[i].vnn;
964 req.opcode = CTDB_CONTROL_PROCESS_EXISTS;
966 req.datalen = sizeof(pids[i].pid);
969 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
970 ctdb_packet_dump(&req.hdr);
972 status = ctdb_packet_send(
975 &req, offsetof(struct ctdb_req_control, data)),
976 data_blob_const(&pids[i].pid, sizeof(pids[i].pid)));
977 if (!NT_STATUS_IS_OK(status)) {
978 DEBUG(10, ("ctdb_packet_send failed: %s\n",
984 status = ctdb_packet_flush(conn->pkt);
985 if (!NT_STATUS_IS_OK(status)) {
986 DEBUG(10, ("ctdb_packet_flush failed: %s\n",
993 while (num_received < num_pids) {
994 struct ctdb_reply_control *reply = NULL;
997 status = ctdb_read_req(conn, 0, talloc_tos(), (void *)&reply);
998 if (!NT_STATUS_IS_OK(status)) {
999 DEBUG(10, ("ctdb_read_req failed: %s\n",
1000 nt_errstr(status)));
1004 if (reply->hdr.operation != CTDB_REPLY_CONTROL) {
1005 DEBUG(10, ("Received invalid reply\n"));
1009 reqid = reply->hdr.reqid;
1011 DEBUG(10, ("Received reqid %d\n", (int)reqid));
1013 for (i=0; i<num_pids; i++) {
1014 if (reqid == reqids[i]) {
1018 if (i == num_pids) {
1019 DEBUG(10, ("Received unknown record number %u\n",
1023 results[i] = ((reply->status) == 0);
1037 char *ctdbd_dbpath(struct ctdbd_connection *conn,
1038 TALLOC_CTX *mem_ctx, uint32_t db_id)
1044 data.dptr = (uint8_t*)&db_id;
1045 data.dsize = sizeof(db_id);
1047 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
1048 CTDB_CONTROL_GETDBPATH, 0, 0, data,
1049 mem_ctx, &data, &cstatus);
1050 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
1051 DEBUG(0,(__location__ " ctdb_control for getdbpath failed\n"));
1055 return (char *)data.dptr;
1059 * attach to a ctdb database
1061 NTSTATUS ctdbd_db_attach(struct ctdbd_connection *conn,
1062 const char *name, uint32_t *db_id, int tdb_flags)
1067 bool persistent = (tdb_flags & TDB_CLEAR_IF_FIRST) == 0;
1069 data.dptr = (uint8_t*)name;
1070 data.dsize = strlen(name)+1;
1072 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
1074 ? CTDB_CONTROL_DB_ATTACH_PERSISTENT
1075 : CTDB_CONTROL_DB_ATTACH,
1076 tdb_flags, 0, data, NULL, &data, &cstatus);
1077 if (!NT_STATUS_IS_OK(status)) {
1078 DEBUG(0, (__location__ " ctdb_control for db_attach "
1079 "failed: %s\n", nt_errstr(status)));
1083 if (cstatus != 0 || data.dsize != sizeof(uint32_t)) {
1084 DEBUG(0,(__location__ " ctdb_control for db_attach failed\n"));
1085 return NT_STATUS_INTERNAL_ERROR;
1088 *db_id = *(uint32_t *)data.dptr;
1089 talloc_free(data.dptr);
1091 if (!(tdb_flags & TDB_SEQNUM)) {
1092 return NT_STATUS_OK;
1095 data.dptr = (uint8_t *)db_id;
1096 data.dsize = sizeof(*db_id);
1098 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
1099 CTDB_CONTROL_ENABLE_SEQNUM, 0, 0, data,
1100 NULL, NULL, &cstatus);
1101 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
1102 DEBUG(0,(__location__ " ctdb_control for enable seqnum "
1104 return NT_STATUS_IS_OK(status) ? NT_STATUS_INTERNAL_ERROR :
1108 return NT_STATUS_OK;
1112 * force the migration of a record to this node
1114 NTSTATUS ctdbd_migrate(struct ctdbd_connection *conn, uint32 db_id,
1117 struct ctdb_req_call req;
1118 struct ctdb_reply_call *reply;
1123 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
1124 req.hdr.ctdb_magic = CTDB_MAGIC;
1125 req.hdr.ctdb_version = CTDB_VERSION;
1126 req.hdr.operation = CTDB_REQ_CALL;
1127 req.hdr.reqid = ctdbd_next_reqid(conn);
1128 req.flags = CTDB_IMMEDIATE_MIGRATION;
1129 req.callid = CTDB_NULL_FUNC;
1131 req.keylen = key.dsize;
1133 DEBUG(10, ("ctdbd_migrate: Sending ctdb packet\n"));
1134 ctdb_packet_dump(&req.hdr);
1136 status = ctdb_packet_send(
1138 data_blob_const(&req, offsetof(struct ctdb_req_call, data)),
1139 data_blob_const(key.dptr, key.dsize));
1141 if (!NT_STATUS_IS_OK(status)) {
1142 DEBUG(3, ("ctdb_packet_send failed: %s\n", nt_errstr(status)));
1146 status = ctdb_packet_flush(conn->pkt);
1148 if (!NT_STATUS_IS_OK(status)) {
1149 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
1150 cluster_fatal("cluster dispatch daemon control write error\n");
1153 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
1155 if (!NT_STATUS_IS_OK(status)) {
1156 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
1160 if (reply->hdr.operation != CTDB_REPLY_CALL) {
1161 DEBUG(0, ("received invalid reply\n"));
1162 status = NT_STATUS_INTERNAL_ERROR;
1166 status = NT_STATUS_OK;
1174 * remotely fetch a record without locking it or forcing a migration
1176 NTSTATUS ctdbd_fetch(struct ctdbd_connection *conn, uint32 db_id,
1177 TDB_DATA key, TALLOC_CTX *mem_ctx, TDB_DATA *data)
1179 struct ctdb_req_call req;
1180 struct ctdb_reply_call *reply;
1185 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
1186 req.hdr.ctdb_magic = CTDB_MAGIC;
1187 req.hdr.ctdb_version = CTDB_VERSION;
1188 req.hdr.operation = CTDB_REQ_CALL;
1189 req.hdr.reqid = ctdbd_next_reqid(conn);
1191 req.callid = CTDB_FETCH_FUNC;
1193 req.keylen = key.dsize;
1195 status = ctdb_packet_send(
1197 data_blob_const(&req, offsetof(struct ctdb_req_call, data)),
1198 data_blob_const(key.dptr, key.dsize));
1200 if (!NT_STATUS_IS_OK(status)) {
1201 DEBUG(3, ("ctdb_packet_send failed: %s\n", nt_errstr(status)));
1205 status = ctdb_packet_flush(conn->pkt);
1207 if (!NT_STATUS_IS_OK(status)) {
1208 DEBUG(3, ("write to ctdbd failed: %s\n", nt_errstr(status)));
1209 cluster_fatal("cluster dispatch daemon control write error\n");
1212 status = ctdb_read_req(conn, req.hdr.reqid, NULL, (void *)&reply);
1214 if (!NT_STATUS_IS_OK(status)) {
1215 DEBUG(0, ("ctdb_read_req failed: %s\n", nt_errstr(status)));
1219 if (reply->hdr.operation != CTDB_REPLY_CALL) {
1220 DEBUG(0, ("received invalid reply\n"));
1221 status = NT_STATUS_INTERNAL_ERROR;
1225 data->dsize = reply->datalen;
1226 if (data->dsize == 0) {
1231 data->dptr = (uint8 *)talloc_memdup(mem_ctx, &reply->data[0],
1233 if (data->dptr == NULL) {
1234 DEBUG(0, ("talloc failed\n"));
1235 status = NT_STATUS_NO_MEMORY;
1240 status = NT_STATUS_OK;
1246 struct ctdbd_traverse_state {
1247 void (*fn)(TDB_DATA key, TDB_DATA data, void *private_data);
1252 * Handle a traverse record coming in on the ctdbd connection
1255 static NTSTATUS ctdb_traverse_handler(uint8_t *buf, size_t length,
1258 struct ctdbd_traverse_state *state =
1259 (struct ctdbd_traverse_state *)private_data;
1261 struct ctdb_req_message *m;
1262 struct ctdb_rec_data *d;
1265 m = (struct ctdb_req_message *)buf;
1267 if (length < sizeof(*m) || m->hdr.length != length) {
1268 DEBUG(0, ("Got invalid message of length %d\n", (int)length));
1270 return NT_STATUS_UNEXPECTED_IO_ERROR;
1273 d = (struct ctdb_rec_data *)&m->data[0];
1274 if (m->datalen < sizeof(uint32_t) || m->datalen != d->length) {
1275 DEBUG(0, ("Got invalid traverse data of length %d\n",
1278 return NT_STATUS_UNEXPECTED_IO_ERROR;
1281 key.dsize = d->keylen;
1282 key.dptr = &d->data[0];
1283 data.dsize = d->datalen;
1284 data.dptr = &d->data[d->keylen];
1286 if (key.dsize == 0 && data.dsize == 0) {
1287 /* end of traverse */
1288 return NT_STATUS_END_OF_FILE;
1291 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1292 DEBUG(0, ("Got invalid ltdb header length %d\n",
1295 return NT_STATUS_UNEXPECTED_IO_ERROR;
1297 data.dsize -= sizeof(struct ctdb_ltdb_header);
1298 data.dptr += sizeof(struct ctdb_ltdb_header);
1301 state->fn(key, data, state->private_data);
1305 return NT_STATUS_OK;
1309 Traverse a ctdb database. This uses a kind-of hackish way to open a second
1310 connection to ctdbd to avoid the hairy recursive and async problems with
1314 NTSTATUS ctdbd_traverse(uint32 db_id,
1315 void (*fn)(TDB_DATA key, TDB_DATA data,
1316 void *private_data),
1319 struct ctdbd_connection *conn;
1323 struct ctdb_traverse_start t;
1325 struct ctdbd_traverse_state state;
1327 status = ctdbd_init_connection(NULL, &conn);
1328 if (!NT_STATUS_IS_OK(status)) {
1329 DEBUG(0, ("ctdbd_init_connection failed: %s\n",
1330 nt_errstr(status)));
1335 t.srvid = conn->rand_srvid;
1336 t.reqid = ctdbd_next_reqid(conn);
1338 data.dptr = (uint8_t *)&t;
1339 data.dsize = sizeof(t);
1341 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
1342 CTDB_CONTROL_TRAVERSE_START, conn->rand_srvid, 0,
1343 data, NULL, NULL, &cstatus);
1345 if (!NT_STATUS_IS_OK(status) || (cstatus != 0)) {
1347 DEBUG(0,("ctdbd_control failed: %s, %d\n", nt_errstr(status),
1350 if (NT_STATUS_IS_OK(status)) {
1352 * We need a mapping here
1354 status = NT_STATUS_UNSUCCESSFUL;
1360 state.private_data = private_data;
1364 status = NT_STATUS_OK;
1366 if (ctdb_packet_handler(conn->pkt, ctdb_req_complete,
1367 ctdb_traverse_handler, &state, &status)) {
1369 if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
1370 status = NT_STATUS_OK;
1375 * There might be more in the queue
1380 if (!NT_STATUS_IS_OK(status)) {
1384 status = ctdb_packet_fd_read_sync(conn->pkt);
1386 if (NT_STATUS_EQUAL(status, NT_STATUS_RETRY)) {
1388 * There might be more in the queue
1393 if (NT_STATUS_EQUAL(status, NT_STATUS_END_OF_FILE)) {
1394 status = NT_STATUS_OK;
1398 if (!NT_STATUS_IS_OK(status)) {
1399 DEBUG(0, ("ctdb_packet_fd_read_sync failed: %s\n", nt_errstr(status)));
1400 cluster_fatal("ctdbd died\n");
1410 This is used to canonicalize a ctdb_sock_addr structure.
1412 static void smbd_ctdb_canonicalize_ip(const struct sockaddr_storage *in,
1413 struct sockaddr_storage *out)
1415 memcpy(out, in, sizeof (*out));
1418 if (in->ss_family == AF_INET6) {
1419 const char prefix[12] = { 0,0,0,0,0,0,0,0,0,0,0xff,0xff };
1420 const struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)in;
1421 struct sockaddr_in *out4 = (struct sockaddr_in *)out;
1422 if (memcmp(&in6->sin6_addr, prefix, 12) == 0) {
1423 memset(out, 0, sizeof(*out));
1424 #ifdef HAVE_SOCK_SIN_LEN
1425 out4->sin_len = sizeof(*out);
1427 out4->sin_family = AF_INET;
1428 out4->sin_port = in6->sin6_port;
1429 memcpy(&out4->sin_addr, &in6->sin6_addr.s6_addr32[3], 4);
1436 * Register us as a server for a particular tcp connection
1439 NTSTATUS ctdbd_register_ips(struct ctdbd_connection *conn,
1440 const struct sockaddr_storage *_server,
1441 const struct sockaddr_storage *_client,
1442 void (*release_ip_handler)(const char *ip_addr,
1443 void *private_data),
1447 * we still use ctdb_control_tcp for ipv4
1448 * because we want to work against older ctdb
1449 * versions at runtime
1451 struct ctdb_control_tcp p4;
1452 #ifdef HAVE_STRUCT_CTDB_CONTROL_TCP_ADDR
1453 struct ctdb_control_tcp_addr p;
1457 struct sockaddr_storage client;
1458 struct sockaddr_storage server;
1461 * Only one connection so far
1463 SMB_ASSERT(conn->release_ip_handler == NULL);
1465 smbd_ctdb_canonicalize_ip(_client, &client);
1466 smbd_ctdb_canonicalize_ip(_server, &server);
1468 switch (client.ss_family) {
1470 memcpy(&p4.dest, &server, sizeof(p4.dest));
1471 memcpy(&p4.src, &client, sizeof(p4.src));
1472 data.dptr = (uint8_t *)&p4;
1473 data.dsize = sizeof(p4);
1475 #ifdef HAVE_STRUCT_CTDB_CONTROL_TCP_ADDR
1477 memcpy(&p.dest.ip6, &server, sizeof(p.dest.ip6));
1478 memcpy(&p.src.ip6, &client, sizeof(p.src.ip6));
1479 data.dptr = (uint8_t *)&p;
1480 data.dsize = sizeof(p);
1484 return NT_STATUS_INTERNAL_ERROR;
1487 conn->release_ip_handler = release_ip_handler;
1488 conn->release_ip_priv = private_data;
1491 * We want to be told about IP releases
1494 status = register_with_ctdbd(conn, CTDB_SRVID_RELEASE_IP);
1495 if (!NT_STATUS_IS_OK(status)) {
1500 * inform ctdb of our tcp connection, so if IP takeover happens ctdb
1501 * can send an extra ack to trigger a reset for our client, so it
1502 * immediately reconnects
1504 return ctdbd_control(conn, CTDB_CURRENT_NODE,
1505 CTDB_CONTROL_TCP_CLIENT, 0,
1506 CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL, NULL);
1510 * We want to handle reconfigure events
1512 NTSTATUS ctdbd_register_reconfigure(struct ctdbd_connection *conn)
1514 return register_with_ctdbd(conn, CTDB_SRVID_RECONFIGURE);
1518 call a control on the local node
1520 NTSTATUS ctdbd_control_local(struct ctdbd_connection *conn, uint32 opcode,
1521 uint64_t srvid, uint32_t flags, TDB_DATA data,
1522 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
1525 return ctdbd_control(conn, CTDB_CURRENT_NODE, opcode, srvid, flags, data, mem_ctx, outdata, cstatus);
1528 NTSTATUS ctdb_watch_us(struct ctdbd_connection *conn)
1530 struct ctdb_client_notify_register reg_data;
1535 reg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
1537 reg_data.notify_data[0] = 0;
1539 struct_len = offsetof(struct ctdb_client_notify_register,
1540 notify_data) + reg_data.len;
1542 status = ctdbd_control_local(
1543 conn, CTDB_CONTROL_REGISTER_NOTIFY, conn->rand_srvid, 0,
1544 make_tdb_data((uint8_t *)®_data, struct_len),
1545 NULL, NULL, &cstatus);
1546 if (!NT_STATUS_IS_OK(status)) {
1547 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1548 nt_errstr(status)));
1553 NTSTATUS ctdb_unwatch(struct ctdbd_connection *conn)
1555 struct ctdb_client_notify_deregister dereg_data;
1559 dereg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
1561 status = ctdbd_control_local(
1562 conn, CTDB_CONTROL_DEREGISTER_NOTIFY, conn->rand_srvid, 0,
1563 make_tdb_data((uint8_t *)&dereg_data, sizeof(dereg_data)),
1564 NULL, NULL, &cstatus);
1565 if (!NT_STATUS_IS_OK(status)) {
1566 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1567 nt_errstr(status)));