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 #include "ctdbd_conn.h"
25 #include "system/select.h"
26 #include "lib/sys_rw_data.h"
27 #include "lib/util/iov_buf.h"
31 /* paths to these include files come from --with-ctdb= in configure */
34 #include "ctdb_private.h"
36 struct ctdbd_srvid_cb {
38 int (*cb)(uint32_t src_vnn, uint32_t dst_vnn,
40 const uint8_t *msg, size_t msglen,
45 struct ctdbd_connection {
46 const char *sockname; /* Needed in ctdbd_traverse */
47 struct messaging_context *msg_ctx;
51 struct ctdbd_srvid_cb *callbacks;
53 struct tevent_fd *fde;
57 static uint32_t ctdbd_next_reqid(struct ctdbd_connection *conn)
60 if (conn->reqid == 0) {
66 static int ctdbd_control(struct ctdbd_connection *conn,
67 uint32_t vnn, uint32_t opcode,
68 uint64_t srvid, uint32_t flags,
70 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
74 * exit on fatal communications errors with the ctdbd daemon
76 static void cluster_fatal(const char *why)
78 DEBUG(0,("cluster fatal event: %s - exiting immediately\n", why));
79 /* we don't use smb_panic() as we don't want to delay to write
80 a core file. We need to release this process id immediately
81 so that someone else can take over without getting sharing
89 static void ctdb_packet_dump(struct ctdb_req_header *hdr)
91 if (DEBUGLEVEL < 11) {
94 DEBUGADD(11, ("len=%d, magic=%x, vers=%d, gen=%d, op=%d, reqid=%d\n",
95 (int)hdr->length, (int)hdr->ctdb_magic,
96 (int)hdr->ctdb_version, (int)hdr->generation,
97 (int)hdr->operation, (int)hdr->reqid));
101 * Register a srvid with ctdbd
103 int register_with_ctdbd(struct ctdbd_connection *conn, uint64_t srvid,
104 int (*cb)(uint32_t src_vnn, uint32_t dst_vnn,
106 const uint8_t *msg, size_t msglen,
112 size_t num_callbacks;
113 struct ctdbd_srvid_cb *tmp;
115 ret = ctdbd_control(conn, CTDB_CURRENT_NODE,
116 CTDB_CONTROL_REGISTER_SRVID, srvid, 0,
117 tdb_null, NULL, NULL, &cstatus);
122 num_callbacks = talloc_array_length(conn->callbacks);
124 tmp = talloc_realloc(conn, conn->callbacks, struct ctdbd_srvid_cb,
129 conn->callbacks = tmp;
131 conn->callbacks[num_callbacks] = (struct ctdbd_srvid_cb) {
132 .srvid = srvid, .cb = cb, .private_data = private_data
138 static int ctdbd_msg_call_back(struct ctdbd_connection *conn,
139 struct ctdb_req_message *msg)
142 size_t i, num_callbacks;
144 msg_len = msg->hdr.length;
145 if (msg_len < offsetof(struct ctdb_req_message, data)) {
146 DEBUG(10, ("%s: len %u too small\n", __func__,
150 msg_len -= offsetof(struct ctdb_req_message, data);
152 if (msg_len < msg->datalen) {
153 DEBUG(10, ("%s: msg_len=%u < msg->datalen=%u\n", __func__,
154 (unsigned)msg_len, (unsigned)msg->datalen));
158 num_callbacks = talloc_array_length(conn->callbacks);
160 for (i=0; i<num_callbacks; i++) {
161 struct ctdbd_srvid_cb *cb = &conn->callbacks[i];
163 if ((cb->srvid == msg->srvid) && (cb->cb != NULL)) {
166 ret = cb->cb(msg->hdr.srcnode, msg->hdr.destnode,
167 msg->srvid, msg->data, msg->datalen,
178 * get our vnn from the cluster
180 static int get_cluster_vnn(struct ctdbd_connection *conn, uint32_t *vnn)
184 ret = ctdbd_control(conn,
185 CTDB_CURRENT_NODE, CTDB_CONTROL_GET_PNN, 0, 0,
186 tdb_null, NULL, NULL, &cstatus);
188 DEBUG(1, ("ctdbd_control failed: %s\n", strerror(ret)));
191 *vnn = (uint32_t)cstatus;
196 * Are we active (i.e. not banned or stopped?)
198 static bool ctdbd_working(struct ctdbd_connection *conn, uint32_t vnn)
202 struct ctdb_node_map *m;
203 uint32_t failure_flags;
207 ret = ctdbd_control(conn, CTDB_CURRENT_NODE,
208 CTDB_CONTROL_GET_NODEMAP, 0, 0,
209 tdb_null, talloc_tos(), &outdata, &cstatus);
211 DEBUG(1, ("ctdbd_control failed: %s\n", strerror(ret)));
214 if ((cstatus != 0) || (outdata.dptr == NULL)) {
215 DEBUG(2, ("Received invalid ctdb data\n"));
219 m = (struct ctdb_node_map *)outdata.dptr;
221 for (i=0; i<m->num; i++) {
222 if (vnn == m->nodes[i].pnn) {
228 DEBUG(2, ("Did not find ourselves (node %d) in nodemap\n",
233 failure_flags = NODE_FLAGS_BANNED | NODE_FLAGS_DISCONNECTED
234 | NODE_FLAGS_PERMANENTLY_DISABLED | NODE_FLAGS_STOPPED;
236 if ((m->nodes[i].flags & failure_flags) != 0) {
237 DEBUG(2, ("Node has status %x, not active\n",
238 (int)m->nodes[i].flags));
244 TALLOC_FREE(outdata.dptr);
248 uint32_t ctdbd_vnn(const struct ctdbd_connection *conn)
250 return conn->our_vnn;
254 * Get us a ctdb connection
257 static int ctdbd_connect(const char *sockname, int *pfd)
259 struct sockaddr_un addr = { 0, };
264 fd = socket(AF_UNIX, SOCK_STREAM, 0);
267 DEBUG(3, ("Could not create socket: %s\n", strerror(err)));
271 addr.sun_family = AF_UNIX;
273 namelen = strlcpy(addr.sun_path, sockname, sizeof(addr.sun_path));
274 if (namelen >= sizeof(addr.sun_path)) {
275 DEBUG(3, ("%s: Socket name too long: %s\n", __func__,
281 salen = sizeof(struct sockaddr_un);
283 if (connect(fd, (struct sockaddr *)(void *)&addr, salen) == -1) {
285 DEBUG(1, ("connect(%s) failed: %s\n", sockname,
295 static int ctdb_read_packet(int fd, int timeout, TALLOC_CTX *mem_ctx,
296 struct ctdb_req_header **result)
298 struct ctdb_req_header *req;
304 ret = poll_one_fd(fd, POLLIN, timeout, &revents);
316 nread = read_data(fd, &msglen, sizeof(msglen));
324 if (msglen < sizeof(struct ctdb_req_header)) {
328 req = talloc_size(mem_ctx, msglen);
332 talloc_set_name_const(req, "struct ctdb_req_header");
334 req->length = msglen;
336 nread = read_data(fd, ((char *)req) + sizeof(msglen),
337 msglen - sizeof(msglen));
352 * Read a full ctdbd request. If we have a messaging context, defer incoming
353 * messages that might come in between.
356 static int ctdb_read_req(struct ctdbd_connection *conn, uint32_t reqid,
357 TALLOC_CTX *mem_ctx, struct ctdb_req_header **result)
359 struct ctdb_req_header *hdr;
364 ret = ctdb_read_packet(conn->fd, conn->timeout, mem_ctx, &hdr);
366 DEBUG(0, ("ctdb_read_packet failed: %s\n", strerror(ret)));
367 cluster_fatal("ctdbd died\n");
370 DEBUG(11, ("Received ctdb packet\n"));
371 ctdb_packet_dump(hdr);
373 if (hdr->operation == CTDB_REQ_MESSAGE) {
374 struct ctdb_req_message *msg = (struct ctdb_req_message *)hdr;
376 if (conn->msg_ctx == NULL) {
377 DEBUG(1, ("Got a message without having a msg ctx, "
378 "dropping msg %llu\n",
379 (long long unsigned)msg->srvid));
384 ret = ctdbd_msg_call_back(conn, msg);
394 if ((reqid != 0) && (hdr->reqid != reqid)) {
395 /* we got the wrong reply */
396 DEBUG(0,("Discarding mismatched ctdb reqid %u should have "
397 "been %u\n", hdr->reqid, reqid));
402 *result = talloc_move(mem_ctx, &hdr);
407 static int ctdbd_connection_destructor(struct ctdbd_connection *c)
417 * Get us a ctdbd connection
420 static NTSTATUS ctdbd_init_connection(TALLOC_CTX *mem_ctx,
421 const char *sockname, int timeout,
422 struct ctdbd_connection **pconn)
424 struct ctdbd_connection *conn;
428 if (!(conn = talloc_zero(mem_ctx, struct ctdbd_connection))) {
429 DEBUG(0, ("talloc failed\n"));
430 return NT_STATUS_NO_MEMORY;
433 conn->sockname = talloc_strdup(conn, sockname);
434 if (conn->sockname == NULL) {
435 DBG_ERR("%s: talloc failed\n", __func__);
436 status = NT_STATUS_NO_MEMORY;
440 conn->timeout = timeout;
442 if (conn->timeout == 0) {
446 ret = ctdbd_connect(conn->sockname, &conn->fd);
448 status = map_nt_error_from_unix(ret);
449 DEBUG(1, ("ctdbd_connect failed: %s\n", strerror(ret)));
452 talloc_set_destructor(conn, ctdbd_connection_destructor);
454 ret = get_cluster_vnn(conn, &conn->our_vnn);
457 DEBUG(10, ("get_cluster_vnn failed: %s\n", strerror(ret)));
458 status = map_nt_error_from_unix(ret);
462 if (!ctdbd_working(conn, conn->our_vnn)) {
463 DEBUG(2, ("Node is not working, can not connect\n"));
464 status = NT_STATUS_INTERNAL_DB_ERROR;
468 generate_random_buffer((unsigned char *)&conn->rand_srvid,
469 sizeof(conn->rand_srvid));
471 ret = register_with_ctdbd(conn, conn->rand_srvid, NULL, NULL);
474 DEBUG(5, ("Could not register random srvid: %s\n",
476 status = map_nt_error_from_unix(ret);
489 * Get us a ctdbd connection and register us as a process
492 NTSTATUS ctdbd_messaging_connection(TALLOC_CTX *mem_ctx,
493 const char *sockname, int timeout,
494 struct ctdbd_connection **pconn)
496 struct ctdbd_connection *conn;
500 status = ctdbd_init_connection(mem_ctx, sockname, timeout, &conn);
502 if (!NT_STATUS_IS_OK(status)) {
506 ret = register_with_ctdbd(conn, MSG_SRVID_SAMBA, NULL, NULL);
508 status = map_nt_error_from_unix(ret);
520 struct messaging_context *ctdb_conn_msg_ctx(struct ctdbd_connection *conn)
522 return conn->msg_ctx;
525 int ctdbd_conn_get_fd(struct ctdbd_connection *conn)
531 * Packet handler to receive and handle a ctdb message
533 static int ctdb_handle_message(struct ctdbd_connection *conn,
534 struct ctdb_req_header *hdr)
536 struct ctdb_req_message *msg;
538 if (hdr->operation != CTDB_REQ_MESSAGE) {
539 DEBUG(0, ("Received async msg of type %u, discarding\n",
544 msg = (struct ctdb_req_message *)hdr;
546 ctdbd_msg_call_back(conn, msg);
552 * The ctdbd socket is readable asynchronuously
555 static void ctdbd_socket_handler(struct tevent_context *event_ctx,
556 struct tevent_fd *event,
560 struct ctdbd_connection *conn = talloc_get_type_abort(
561 private_data, struct ctdbd_connection);
562 struct ctdb_req_header *hdr = NULL;
565 ret = ctdb_read_packet(conn->fd, conn->timeout, talloc_tos(), &hdr);
567 DEBUG(0, ("ctdb_read_packet failed: %s\n", strerror(ret)));
568 cluster_fatal("ctdbd died\n");
571 ret = ctdb_handle_message(conn, hdr);
576 DEBUG(10, ("could not handle incoming message: %s\n",
582 * Prepare a ctdbd connection to receive messages
585 NTSTATUS ctdbd_register_msg_ctx(struct ctdbd_connection *conn,
586 struct messaging_context *msg_ctx)
588 SMB_ASSERT(conn->msg_ctx == NULL);
589 SMB_ASSERT(conn->fde == NULL);
591 if (!(conn->fde = tevent_add_fd(messaging_tevent_context(msg_ctx),
595 ctdbd_socket_handler,
597 DEBUG(0, ("event_add_fd failed\n"));
598 return NT_STATUS_NO_MEMORY;
601 conn->msg_ctx = msg_ctx;
606 NTSTATUS ctdbd_messaging_send_iov(struct ctdbd_connection *conn,
607 uint32_t dst_vnn, uint64_t dst_srvid,
608 const struct iovec *iov, int iovlen)
610 struct ctdb_req_message r;
611 struct iovec iov2[iovlen+1];
612 size_t buflen = iov_buflen(iov, iovlen);
615 r.hdr.length = offsetof(struct ctdb_req_message, data) + buflen;
616 r.hdr.ctdb_magic = CTDB_MAGIC;
617 r.hdr.ctdb_version = CTDB_PROTOCOL;
618 r.hdr.generation = 1;
619 r.hdr.operation = CTDB_REQ_MESSAGE;
620 r.hdr.destnode = dst_vnn;
621 r.hdr.srcnode = conn->our_vnn;
626 DEBUG(10, ("ctdbd_messaging_send: Sending ctdb packet\n"));
627 ctdb_packet_dump(&r.hdr);
629 iov2[0].iov_base = &r;
630 iov2[0].iov_len = offsetof(struct ctdb_req_message, data);
631 memcpy(&iov2[1], iov, iovlen * sizeof(struct iovec));
633 nwritten = write_data_iov(conn->fd, iov2, iovlen+1);
634 if (nwritten == -1) {
635 DEBUG(3, ("write_data_iov failed: %s\n", strerror(errno)));
636 cluster_fatal("cluster dispatch daemon msg write error\n");
643 * send/recv a generic ctdb control message
645 static int ctdbd_control(struct ctdbd_connection *conn,
646 uint32_t vnn, uint32_t opcode,
647 uint64_t srvid, uint32_t flags,
649 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
652 struct ctdb_req_control req;
653 struct ctdb_req_header *hdr;
654 struct ctdb_reply_control *reply = NULL;
660 req.hdr.length = offsetof(struct ctdb_req_control, data) + data.dsize;
661 req.hdr.ctdb_magic = CTDB_MAGIC;
662 req.hdr.ctdb_version = CTDB_PROTOCOL;
663 req.hdr.operation = CTDB_REQ_CONTROL;
664 req.hdr.reqid = ctdbd_next_reqid(conn);
665 req.hdr.destnode = vnn;
668 req.datalen = data.dsize;
671 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
672 ctdb_packet_dump(&req.hdr);
674 iov[0].iov_base = &req;
675 iov[0].iov_len = offsetof(struct ctdb_req_control, data);
676 iov[1].iov_base = data.dptr;
677 iov[1].iov_len = data.dsize;
679 nwritten = write_data_iov(conn->fd, iov, ARRAY_SIZE(iov));
680 if (nwritten == -1) {
681 DEBUG(3, ("write_data_iov failed: %s\n", strerror(errno)));
682 cluster_fatal("cluster dispatch daemon msg write error\n");
685 if (flags & CTDB_CTRL_FLAG_NOREPLY) {
692 ret = ctdb_read_req(conn, req.hdr.reqid, NULL, &hdr);
694 DEBUG(10, ("ctdb_read_req failed: %s\n", strerror(ret)));
698 if (hdr->operation != CTDB_REPLY_CONTROL) {
699 DEBUG(0, ("received invalid reply\n"));
703 reply = (struct ctdb_reply_control *)hdr;
706 if (!(outdata->dptr = (uint8_t *)talloc_memdup(
707 mem_ctx, reply->data, reply->datalen))) {
711 outdata->dsize = reply->datalen;
714 (*cstatus) = reply->status;
722 * see if a remote process exists
724 bool ctdbd_process_exists(struct ctdbd_connection *conn, uint32_t vnn, pid_t pid)
732 if (!ctdb_processes_exist(conn, &id, 1, &result)) {
733 DEBUG(10, ("ctdb_processes_exist failed\n"));
739 bool ctdb_processes_exist(struct ctdbd_connection *conn,
740 const struct server_id *pids, int num_pids,
743 TALLOC_CTX *frame = talloc_stackframe();
748 reqids = talloc_array(talloc_tos(), uint32_t, num_pids);
749 if (reqids == NULL) {
753 for (i=0; i<num_pids; i++) {
754 struct ctdb_req_control req;
760 reqids[i] = ctdbd_next_reqid(conn);
765 * pids[i].pid is uint64_t, scale down to pid_t which
766 * is the wire protocol towards ctdb.
770 DEBUG(10, ("Requesting PID %d/%d, reqid=%d\n",
771 (int)pids[i].vnn, (int)pid,
774 req.hdr.length = offsetof(struct ctdb_req_control, data);
775 req.hdr.length += sizeof(pid);
776 req.hdr.ctdb_magic = CTDB_MAGIC;
777 req.hdr.ctdb_version = CTDB_PROTOCOL;
778 req.hdr.operation = CTDB_REQ_CONTROL;
779 req.hdr.reqid = reqids[i];
780 req.hdr.destnode = pids[i].vnn;
781 req.opcode = CTDB_CONTROL_PROCESS_EXISTS;
783 req.datalen = sizeof(pid);
786 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
787 ctdb_packet_dump(&req.hdr);
789 iov[0].iov_base = &req;
790 iov[0].iov_len = offsetof(struct ctdb_req_control, data);
791 iov[1].iov_base = &pid;
792 iov[1].iov_len = sizeof(pid);
794 nwritten = write_data_iov(conn->fd, iov, ARRAY_SIZE(iov));
795 if (nwritten == -1) {
796 DEBUG(10, ("write_data_iov failed: %s\n",
804 while (num_received < num_pids) {
805 struct ctdb_req_header *hdr;
806 struct ctdb_reply_control *reply;
810 ret = ctdb_read_req(conn, 0, talloc_tos(), &hdr);
812 DEBUG(10, ("ctdb_read_req failed: %s\n",
817 if (hdr->operation != CTDB_REPLY_CONTROL) {
818 DEBUG(10, ("Received invalid reply\n"));
821 reply = (struct ctdb_reply_control *)hdr;
823 reqid = reply->hdr.reqid;
825 DEBUG(10, ("Received reqid %d\n", (int)reqid));
827 for (i=0; i<num_pids; i++) {
828 if (reqid == reqids[i]) {
833 DEBUG(10, ("Received unknown record number %u\n",
837 results[i] = ((reply->status) == 0);
851 char *ctdbd_dbpath(struct ctdbd_connection *conn,
852 TALLOC_CTX *mem_ctx, uint32_t db_id)
856 TDB_DATA rdata = {0};
859 data.dptr = (uint8_t*)&db_id;
860 data.dsize = sizeof(db_id);
862 ret = ctdbd_control(conn, CTDB_CURRENT_NODE,
863 CTDB_CONTROL_GETDBPATH, 0, 0, data,
864 mem_ctx, &rdata, &cstatus);
865 if ((ret != 0) || cstatus != 0) {
866 DEBUG(0, (__location__ " ctdb_control for getdbpath failed: %s\n",
871 return (char *)rdata.dptr;
875 * attach to a ctdb database
877 NTSTATUS ctdbd_db_attach(struct ctdbd_connection *conn,
878 const char *name, uint32_t *db_id, int tdb_flags)
883 bool persistent = (tdb_flags & TDB_CLEAR_IF_FIRST) == 0;
885 data = string_term_tdb_data(name);
887 ret = ctdbd_control(conn, CTDB_CURRENT_NODE,
889 ? CTDB_CONTROL_DB_ATTACH_PERSISTENT
890 : CTDB_CONTROL_DB_ATTACH,
891 tdb_flags, 0, data, NULL, &data, &cstatus);
893 DEBUG(0, (__location__ " ctdb_control for db_attach "
894 "failed: %s\n", strerror(ret)));
895 return map_nt_error_from_unix(ret);
898 if (cstatus != 0 || data.dsize != sizeof(uint32_t)) {
899 DEBUG(0,(__location__ " ctdb_control for db_attach failed\n"));
900 return NT_STATUS_INTERNAL_ERROR;
903 *db_id = *(uint32_t *)data.dptr;
904 talloc_free(data.dptr);
906 if (!(tdb_flags & TDB_SEQNUM)) {
910 data.dptr = (uint8_t *)db_id;
911 data.dsize = sizeof(*db_id);
913 ret = ctdbd_control(conn, CTDB_CURRENT_NODE,
914 CTDB_CONTROL_ENABLE_SEQNUM, 0, 0, data,
915 NULL, NULL, &cstatus);
916 if ((ret != 0) || cstatus != 0) {
917 DEBUG(0, (__location__ " ctdb_control for enable seqnum "
918 "failed: %s\n", strerror(ret)));
919 return (ret == 0) ? NT_STATUS_INTERNAL_ERROR :
920 map_nt_error_from_unix(ret);
927 * force the migration of a record to this node
929 NTSTATUS ctdbd_migrate(struct ctdbd_connection *conn, uint32_t db_id,
932 struct ctdb_req_call req;
933 struct ctdb_req_header *hdr;
941 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
942 req.hdr.ctdb_magic = CTDB_MAGIC;
943 req.hdr.ctdb_version = CTDB_PROTOCOL;
944 req.hdr.operation = CTDB_REQ_CALL;
945 req.hdr.reqid = ctdbd_next_reqid(conn);
946 req.flags = CTDB_IMMEDIATE_MIGRATION;
947 req.callid = CTDB_NULL_FUNC;
949 req.keylen = key.dsize;
951 DEBUG(10, ("ctdbd_migrate: Sending ctdb packet\n"));
952 ctdb_packet_dump(&req.hdr);
954 iov[0].iov_base = &req;
955 iov[0].iov_len = offsetof(struct ctdb_req_call, data);
956 iov[1].iov_base = key.dptr;
957 iov[1].iov_len = key.dsize;
959 nwritten = write_data_iov(conn->fd, iov, ARRAY_SIZE(iov));
960 if (nwritten == -1) {
961 DEBUG(3, ("write_data_iov failed: %s\n", strerror(errno)));
962 cluster_fatal("cluster dispatch daemon msg write error\n");
965 ret = ctdb_read_req(conn, req.hdr.reqid, NULL, &hdr);
967 DEBUG(10, ("ctdb_read_req failed: %s\n", strerror(ret)));
968 status = map_nt_error_from_unix(ret);
972 if (hdr->operation != CTDB_REPLY_CALL) {
973 DEBUG(0, ("received invalid reply\n"));
974 status = NT_STATUS_INTERNAL_ERROR;
978 status = NT_STATUS_OK;
986 * Fetch a record and parse it
988 NTSTATUS ctdbd_parse(struct ctdbd_connection *conn, uint32_t db_id,
989 TDB_DATA key, bool local_copy,
990 void (*parser)(TDB_DATA key, TDB_DATA data,
994 struct ctdb_req_call req;
995 struct ctdb_req_header *hdr = NULL;
996 struct ctdb_reply_call *reply;
1003 flags = local_copy ? CTDB_WANT_READONLY : 0;
1007 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
1008 req.hdr.ctdb_magic = CTDB_MAGIC;
1009 req.hdr.ctdb_version = CTDB_PROTOCOL;
1010 req.hdr.operation = CTDB_REQ_CALL;
1011 req.hdr.reqid = ctdbd_next_reqid(conn);
1013 req.callid = CTDB_FETCH_FUNC;
1015 req.keylen = key.dsize;
1017 iov[0].iov_base = &req;
1018 iov[0].iov_len = offsetof(struct ctdb_req_call, data);
1019 iov[1].iov_base = key.dptr;
1020 iov[1].iov_len = key.dsize;
1022 nwritten = write_data_iov(conn->fd, iov, ARRAY_SIZE(iov));
1023 if (nwritten == -1) {
1024 DEBUG(3, ("write_data_iov failed: %s\n", strerror(errno)));
1025 cluster_fatal("cluster dispatch daemon msg write error\n");
1028 ret = ctdb_read_req(conn, req.hdr.reqid, NULL, &hdr);
1030 DEBUG(10, ("ctdb_read_req failed: %s\n", strerror(ret)));
1031 status = map_nt_error_from_unix(ret);
1035 if ((hdr == NULL) || (hdr->operation != CTDB_REPLY_CALL)) {
1036 DEBUG(0, ("received invalid reply\n"));
1037 status = NT_STATUS_INTERNAL_ERROR;
1040 reply = (struct ctdb_reply_call *)hdr;
1042 if (reply->datalen == 0) {
1044 * Treat an empty record as non-existing
1046 status = NT_STATUS_NOT_FOUND;
1050 parser(key, make_tdb_data(&reply->data[0], reply->datalen),
1053 status = NT_STATUS_OK;
1060 Traverse a ctdb database. This uses a kind-of hackish way to open a second
1061 connection to ctdbd to avoid the hairy recursive and async problems with
1065 NTSTATUS ctdbd_traverse(struct ctdbd_connection *master, uint32_t db_id,
1066 void (*fn)(TDB_DATA key, TDB_DATA data,
1067 void *private_data),
1070 struct ctdbd_connection *conn;
1074 struct ctdb_traverse_start t;
1078 status = ctdbd_init_connection(NULL, master->sockname, master->timeout,
1081 if (!NT_STATUS_IS_OK(status)) {
1082 DEBUG(0, ("ctdbd_init_connection failed: %s\n",
1083 nt_errstr(status)));
1088 t.srvid = conn->rand_srvid;
1089 t.reqid = ctdbd_next_reqid(conn);
1091 data.dptr = (uint8_t *)&t;
1092 data.dsize = sizeof(t);
1094 ret = ctdbd_control(conn, CTDB_CURRENT_NODE,
1095 CTDB_CONTROL_TRAVERSE_START, conn->rand_srvid,
1096 0, data, NULL, NULL, &cstatus);
1098 if ((ret != 0) || (cstatus != 0)) {
1099 status = map_nt_error_from_unix(ret);
1101 DEBUG(0,("ctdbd_control failed: %s, %d\n", strerror(ret),
1104 if (NT_STATUS_IS_OK(status)) {
1106 * We need a mapping here
1108 status = NT_STATUS_UNSUCCESSFUL;
1115 struct ctdb_req_header *hdr = NULL;
1116 struct ctdb_req_message *m;
1117 struct ctdb_rec_data *d;
1119 ret = ctdb_read_packet(conn->fd, conn->timeout, conn, &hdr);
1121 DEBUG(0, ("ctdb_read_packet failed: %s\n",
1123 cluster_fatal("ctdbd died\n");
1126 if (hdr->operation != CTDB_REQ_MESSAGE) {
1127 DEBUG(0, ("Got operation %u, expected a message\n",
1128 (unsigned)hdr->operation));
1130 return NT_STATUS_UNEXPECTED_IO_ERROR;
1133 m = (struct ctdb_req_message *)hdr;
1134 d = (struct ctdb_rec_data *)&m->data[0];
1135 if (m->datalen < sizeof(uint32_t) || m->datalen != d->length) {
1136 DEBUG(0, ("Got invalid traverse data of length %d\n",
1139 return NT_STATUS_UNEXPECTED_IO_ERROR;
1142 key.dsize = d->keylen;
1143 key.dptr = &d->data[0];
1144 data.dsize = d->datalen;
1145 data.dptr = &d->data[d->keylen];
1147 if (key.dsize == 0 && data.dsize == 0) {
1148 /* end of traverse */
1150 return NT_STATUS_OK;
1153 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1154 DEBUG(0, ("Got invalid ltdb header length %d\n",
1157 return NT_STATUS_UNEXPECTED_IO_ERROR;
1159 data.dsize -= sizeof(struct ctdb_ltdb_header);
1160 data.dptr += sizeof(struct ctdb_ltdb_header);
1163 fn(key, data, private_data);
1166 return NT_STATUS_OK;
1170 This is used to canonicalize a ctdb_sock_addr structure.
1172 static void smbd_ctdb_canonicalize_ip(const struct sockaddr_storage *in,
1173 struct sockaddr_storage *out)
1175 memcpy(out, in, sizeof (*out));
1178 if (in->ss_family == AF_INET6) {
1179 const char prefix[12] = { 0,0,0,0,0,0,0,0,0,0,0xff,0xff };
1180 const struct sockaddr_in6 *in6 =
1181 (const struct sockaddr_in6 *)in;
1182 struct sockaddr_in *out4 = (struct sockaddr_in *)out;
1183 if (memcmp(&in6->sin6_addr, prefix, 12) == 0) {
1184 memset(out, 0, sizeof(*out));
1185 #ifdef HAVE_SOCK_SIN_LEN
1186 out4->sin_len = sizeof(*out);
1188 out4->sin_family = AF_INET;
1189 out4->sin_port = in6->sin6_port;
1190 memcpy(&out4->sin_addr, &in6->sin6_addr.s6_addr[12], 4);
1197 * Register us as a server for a particular tcp connection
1200 NTSTATUS ctdbd_register_ips(struct ctdbd_connection *conn,
1201 const struct sockaddr_storage *_server,
1202 const struct sockaddr_storage *_client,
1203 int (*cb)(uint32_t src_vnn, uint32_t dst_vnn,
1205 const uint8_t *msg, size_t msglen,
1206 void *private_data),
1209 struct ctdb_control_tcp_addr p;
1210 TDB_DATA data = { .dptr = (uint8_t *)&p, .dsize = sizeof(p) };
1212 struct sockaddr_storage client;
1213 struct sockaddr_storage server;
1216 * Only one connection so far
1219 smbd_ctdb_canonicalize_ip(_client, &client);
1220 smbd_ctdb_canonicalize_ip(_server, &server);
1222 switch (client.ss_family) {
1224 memcpy(&p.dest.ip, &server, sizeof(p.dest.ip));
1225 memcpy(&p.src.ip, &client, sizeof(p.src.ip));
1228 memcpy(&p.dest.ip6, &server, sizeof(p.dest.ip6));
1229 memcpy(&p.src.ip6, &client, sizeof(p.src.ip6));
1232 return NT_STATUS_INTERNAL_ERROR;
1236 * We want to be told about IP releases
1239 ret = register_with_ctdbd(conn, CTDB_SRVID_RELEASE_IP,
1242 return map_nt_error_from_unix(ret);
1246 * inform ctdb of our tcp connection, so if IP takeover happens ctdb
1247 * can send an extra ack to trigger a reset for our client, so it
1248 * immediately reconnects
1250 ret = ctdbd_control(conn, CTDB_CURRENT_NODE,
1251 CTDB_CONTROL_TCP_CLIENT, 0,
1252 CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL,
1255 return map_nt_error_from_unix(ret);
1257 return NT_STATUS_OK;
1261 call a control on the local node
1263 NTSTATUS ctdbd_control_local(struct ctdbd_connection *conn, uint32_t opcode,
1264 uint64_t srvid, uint32_t flags, TDB_DATA data,
1265 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
1270 ret = ctdbd_control(conn, CTDB_CURRENT_NODE, opcode, srvid, flags, data,
1271 mem_ctx, outdata, cstatus);
1273 return map_nt_error_from_unix(ret);
1275 return NT_STATUS_OK;
1278 NTSTATUS ctdb_watch_us(struct ctdbd_connection *conn)
1280 struct ctdb_client_notify_register reg_data;
1285 reg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
1287 reg_data.notify_data[0] = 0;
1289 struct_len = offsetof(struct ctdb_client_notify_register,
1290 notify_data) + reg_data.len;
1292 status = ctdbd_control_local(
1293 conn, CTDB_CONTROL_REGISTER_NOTIFY, conn->rand_srvid, 0,
1294 make_tdb_data((uint8_t *)®_data, struct_len),
1295 NULL, NULL, &cstatus);
1296 if (!NT_STATUS_IS_OK(status)) {
1297 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1298 nt_errstr(status)));
1303 NTSTATUS ctdb_unwatch(struct ctdbd_connection *conn)
1305 struct ctdb_client_notify_deregister dereg_data;
1309 dereg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
1311 status = ctdbd_control_local(
1312 conn, CTDB_CONTROL_DEREGISTER_NOTIFY, conn->rand_srvid, 0,
1313 make_tdb_data((uint8_t *)&dereg_data, sizeof(dereg_data)),
1314 NULL, NULL, &cstatus);
1315 if (!NT_STATUS_IS_OK(status)) {
1316 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1317 nt_errstr(status)));
1322 NTSTATUS ctdbd_probe(const char *sockname, int timeout)
1325 * Do a very early check if ctdbd is around to avoid an abort and core
1328 struct ctdbd_connection *conn = NULL;
1331 status = ctdbd_messaging_connection(talloc_tos(), sockname, timeout,
1335 * We only care if we can connect.