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 NTSTATUS ctdbd_control(struct ctdbd_connection *conn,
67 uint32_t vnn, uint32_t opcode,
68 uint64_t srvid, uint32_t flags, TDB_DATA data,
69 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
73 * exit on fatal communications errors with the ctdbd daemon
75 static void cluster_fatal(const char *why)
77 DEBUG(0,("cluster fatal event: %s - exiting immediately\n", why));
78 /* we don't use smb_panic() as we don't want to delay to write
79 a core file. We need to release this process id immediately
80 so that someone else can take over without getting sharing
88 static void ctdb_packet_dump(struct ctdb_req_header *hdr)
90 if (DEBUGLEVEL < 11) {
93 DEBUGADD(11, ("len=%d, magic=%x, vers=%d, gen=%d, op=%d, reqid=%d\n",
94 (int)hdr->length, (int)hdr->ctdb_magic,
95 (int)hdr->ctdb_version, (int)hdr->generation,
96 (int)hdr->operation, (int)hdr->reqid));
100 * Register a srvid with ctdbd
102 NTSTATUS register_with_ctdbd(struct ctdbd_connection *conn, uint64_t srvid,
103 int (*cb)(uint32_t src_vnn, uint32_t dst_vnn,
105 const uint8_t *msg, size_t msglen,
112 size_t num_callbacks;
113 struct ctdbd_srvid_cb *tmp;
115 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
116 CTDB_CONTROL_REGISTER_SRVID, srvid, 0,
117 tdb_null, NULL, NULL, &cstatus);
118 if (!NT_STATUS_IS_OK(status)) {
122 num_callbacks = talloc_array_length(conn->callbacks);
124 tmp = talloc_realloc(conn, conn->callbacks, struct ctdbd_srvid_cb,
127 return NT_STATUS_NO_MEMORY;
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 NTSTATUS get_cluster_vnn(struct ctdbd_connection *conn, uint32_t *vnn)
184 status = ctdbd_control(conn,
185 CTDB_CURRENT_NODE, CTDB_CONTROL_GET_PNN, 0, 0,
186 tdb_null, NULL, NULL, &cstatus);
187 if (!NT_STATUS_IS_OK(status)) {
188 DEBUG(1, ("ctdbd_control failed: %s\n", nt_errstr(status)));
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)
203 struct ctdb_node_map *m;
204 uint32_t failure_flags;
208 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
209 CTDB_CONTROL_GET_NODEMAP, 0, 0,
210 tdb_null, talloc_tos(), &outdata, &cstatus);
211 if (!NT_STATUS_IS_OK(status)) {
212 DEBUG(1, ("ctdbd_control failed: %s\n", nt_errstr(status)));
215 if ((cstatus != 0) || (outdata.dptr == NULL)) {
216 DEBUG(2, ("Received invalid ctdb data\n"));
220 m = (struct ctdb_node_map *)outdata.dptr;
222 for (i=0; i<m->num; i++) {
223 if (vnn == m->nodes[i].pnn) {
229 DEBUG(2, ("Did not find ourselves (node %d) in nodemap\n",
234 failure_flags = NODE_FLAGS_BANNED | NODE_FLAGS_DISCONNECTED
235 | NODE_FLAGS_PERMANENTLY_DISABLED | NODE_FLAGS_STOPPED;
237 if ((m->nodes[i].flags & failure_flags) != 0) {
238 DEBUG(2, ("Node has status %x, not active\n",
239 (int)m->nodes[i].flags));
245 TALLOC_FREE(outdata.dptr);
249 uint32_t ctdbd_vnn(const struct ctdbd_connection *conn)
251 return conn->our_vnn;
255 * Get us a ctdb connection
258 static int ctdbd_connect(const char *sockname, int *pfd)
260 struct sockaddr_un addr = { 0, };
265 fd = socket(AF_UNIX, SOCK_STREAM, 0);
268 DEBUG(3, ("Could not create socket: %s\n", strerror(err)));
272 addr.sun_family = AF_UNIX;
274 namelen = strlcpy(addr.sun_path, sockname, sizeof(addr.sun_path));
275 if (namelen >= sizeof(addr.sun_path)) {
276 DEBUG(3, ("%s: Socket name too long: %s\n", __func__,
282 salen = sizeof(struct sockaddr_un);
284 if (connect(fd, (struct sockaddr *)(void *)&addr, salen) == -1) {
286 DEBUG(1, ("connect(%s) failed: %s\n", sockname,
296 static int ctdb_read_packet(int fd, int timeout, TALLOC_CTX *mem_ctx,
297 struct ctdb_req_header **result)
299 struct ctdb_req_header *req;
305 ret = poll_one_fd(fd, POLLIN, timeout, &revents);
317 nread = read_data(fd, &msglen, sizeof(msglen));
325 if (msglen < sizeof(struct ctdb_req_header)) {
329 req = talloc_size(mem_ctx, msglen);
333 talloc_set_name_const(req, "struct ctdb_req_header");
335 req->length = msglen;
337 nread = read_data(fd, ((char *)req) + sizeof(msglen),
338 msglen - sizeof(msglen));
353 * Read a full ctdbd request. If we have a messaging context, defer incoming
354 * messages that might come in between.
357 static int ctdb_read_req(struct ctdbd_connection *conn, uint32_t reqid,
358 TALLOC_CTX *mem_ctx, struct ctdb_req_header **result)
360 struct ctdb_req_header *hdr;
365 ret = ctdb_read_packet(conn->fd, conn->timeout, mem_ctx, &hdr);
367 DEBUG(0, ("ctdb_read_packet failed: %s\n", strerror(ret)));
368 cluster_fatal("ctdbd died\n");
371 DEBUG(11, ("Received ctdb packet\n"));
372 ctdb_packet_dump(hdr);
374 if (hdr->operation == CTDB_REQ_MESSAGE) {
375 struct ctdb_req_message *msg = (struct ctdb_req_message *)hdr;
377 if (conn->msg_ctx == NULL) {
378 DEBUG(1, ("Got a message without having a msg ctx, "
379 "dropping msg %llu\n",
380 (long long unsigned)msg->srvid));
385 ret = ctdbd_msg_call_back(conn, msg);
395 if ((reqid != 0) && (hdr->reqid != reqid)) {
396 /* we got the wrong reply */
397 DEBUG(0,("Discarding mismatched ctdb reqid %u should have "
398 "been %u\n", hdr->reqid, reqid));
403 *result = talloc_move(mem_ctx, &hdr);
408 static int ctdbd_connection_destructor(struct ctdbd_connection *c)
418 * Get us a ctdbd connection
421 static NTSTATUS ctdbd_init_connection(TALLOC_CTX *mem_ctx,
422 const char *sockname, int timeout,
423 struct ctdbd_connection **pconn)
425 struct ctdbd_connection *conn;
429 if (!(conn = talloc_zero(mem_ctx, struct ctdbd_connection))) {
430 DEBUG(0, ("talloc failed\n"));
431 return NT_STATUS_NO_MEMORY;
434 conn->sockname = talloc_strdup(conn, sockname);
435 if (conn->sockname == NULL) {
436 DBG_ERR("%s: talloc failed\n", __func__);
437 status = NT_STATUS_NO_MEMORY;
441 conn->timeout = timeout;
443 if (conn->timeout == 0) {
447 ret = ctdbd_connect(conn->sockname, &conn->fd);
449 status = map_nt_error_from_unix(ret);
450 DEBUG(1, ("ctdbd_connect failed: %s\n", strerror(ret)));
453 talloc_set_destructor(conn, ctdbd_connection_destructor);
455 status = get_cluster_vnn(conn, &conn->our_vnn);
457 if (!NT_STATUS_IS_OK(status)) {
458 DEBUG(10, ("get_cluster_vnn failed: %s\n", nt_errstr(status)));
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 status = register_with_ctdbd(conn, conn->rand_srvid, NULL, NULL);
473 if (!NT_STATUS_IS_OK(status)) {
474 DEBUG(5, ("Could not register random srvid: %s\n",
488 * Get us a ctdbd connection and register us as a process
491 NTSTATUS ctdbd_messaging_connection(TALLOC_CTX *mem_ctx,
492 const char *sockname, int timeout,
493 struct ctdbd_connection **pconn)
495 struct ctdbd_connection *conn;
498 status = ctdbd_init_connection(mem_ctx, sockname, timeout, &conn);
500 if (!NT_STATUS_IS_OK(status)) {
504 status = register_with_ctdbd(conn, MSG_SRVID_SAMBA, NULL, NULL);
505 if (!NT_STATUS_IS_OK(status)) {
517 struct messaging_context *ctdb_conn_msg_ctx(struct ctdbd_connection *conn)
519 return conn->msg_ctx;
522 int ctdbd_conn_get_fd(struct ctdbd_connection *conn)
528 * Packet handler to receive and handle a ctdb message
530 static int ctdb_handle_message(struct ctdbd_connection *conn,
531 struct ctdb_req_header *hdr)
533 struct ctdb_req_message *msg;
535 if (hdr->operation != CTDB_REQ_MESSAGE) {
536 DEBUG(0, ("Received async msg of type %u, discarding\n",
541 msg = (struct ctdb_req_message *)hdr;
543 ctdbd_msg_call_back(conn, msg);
549 * The ctdbd socket is readable asynchronuously
552 static void ctdbd_socket_handler(struct tevent_context *event_ctx,
553 struct tevent_fd *event,
557 struct ctdbd_connection *conn = talloc_get_type_abort(
558 private_data, struct ctdbd_connection);
559 struct ctdb_req_header *hdr = NULL;
562 ret = ctdb_read_packet(conn->fd, conn->timeout, talloc_tos(), &hdr);
564 DEBUG(0, ("ctdb_read_packet failed: %s\n", strerror(ret)));
565 cluster_fatal("ctdbd died\n");
568 ret = ctdb_handle_message(conn, hdr);
573 DEBUG(10, ("could not handle incoming message: %s\n",
579 * Prepare a ctdbd connection to receive messages
582 NTSTATUS ctdbd_register_msg_ctx(struct ctdbd_connection *conn,
583 struct messaging_context *msg_ctx)
585 SMB_ASSERT(conn->msg_ctx == NULL);
586 SMB_ASSERT(conn->fde == NULL);
588 if (!(conn->fde = tevent_add_fd(messaging_tevent_context(msg_ctx),
592 ctdbd_socket_handler,
594 DEBUG(0, ("event_add_fd failed\n"));
595 return NT_STATUS_NO_MEMORY;
598 conn->msg_ctx = msg_ctx;
603 NTSTATUS ctdbd_messaging_send_iov(struct ctdbd_connection *conn,
604 uint32_t dst_vnn, uint64_t dst_srvid,
605 const struct iovec *iov, int iovlen)
607 struct ctdb_req_message r;
608 struct iovec iov2[iovlen+1];
609 size_t buflen = iov_buflen(iov, iovlen);
612 r.hdr.length = offsetof(struct ctdb_req_message, data) + buflen;
613 r.hdr.ctdb_magic = CTDB_MAGIC;
614 r.hdr.ctdb_version = CTDB_PROTOCOL;
615 r.hdr.generation = 1;
616 r.hdr.operation = CTDB_REQ_MESSAGE;
617 r.hdr.destnode = dst_vnn;
618 r.hdr.srcnode = conn->our_vnn;
623 DEBUG(10, ("ctdbd_messaging_send: Sending ctdb packet\n"));
624 ctdb_packet_dump(&r.hdr);
626 iov2[0].iov_base = &r;
627 iov2[0].iov_len = offsetof(struct ctdb_req_message, data);
628 memcpy(&iov2[1], iov, iovlen * sizeof(struct iovec));
630 nwritten = write_data_iov(conn->fd, iov2, iovlen+1);
631 if (nwritten == -1) {
632 DEBUG(3, ("write_data_iov failed: %s\n", strerror(errno)));
633 cluster_fatal("cluster dispatch daemon msg write error\n");
640 * send/recv a generic ctdb control message
642 static NTSTATUS ctdbd_control(struct ctdbd_connection *conn,
643 uint32_t vnn, uint32_t opcode,
644 uint64_t srvid, uint32_t flags,
646 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
649 struct ctdb_req_control req;
650 struct ctdb_req_header *hdr;
651 struct ctdb_reply_control *reply = NULL;
658 req.hdr.length = offsetof(struct ctdb_req_control, data) + data.dsize;
659 req.hdr.ctdb_magic = CTDB_MAGIC;
660 req.hdr.ctdb_version = CTDB_PROTOCOL;
661 req.hdr.operation = CTDB_REQ_CONTROL;
662 req.hdr.reqid = ctdbd_next_reqid(conn);
663 req.hdr.destnode = vnn;
666 req.datalen = data.dsize;
669 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
670 ctdb_packet_dump(&req.hdr);
672 iov[0].iov_base = &req;
673 iov[0].iov_len = offsetof(struct ctdb_req_control, data);
674 iov[1].iov_base = data.dptr;
675 iov[1].iov_len = data.dsize;
677 nwritten = write_data_iov(conn->fd, iov, ARRAY_SIZE(iov));
678 if (nwritten == -1) {
679 DEBUG(3, ("write_data_iov failed: %s\n", strerror(errno)));
680 cluster_fatal("cluster dispatch daemon msg write error\n");
683 if (flags & CTDB_CTRL_FLAG_NOREPLY) {
690 ret = ctdb_read_req(conn, req.hdr.reqid, NULL, &hdr);
692 DEBUG(10, ("ctdb_read_req failed: %s\n", strerror(ret)));
693 status = map_nt_error_from_unix(ret);
697 if (hdr->operation != CTDB_REPLY_CONTROL) {
698 DEBUG(0, ("received invalid reply\n"));
699 status = NT_STATUS_INVALID_NETWORK_RESPONSE;
702 reply = (struct ctdb_reply_control *)hdr;
705 if (!(outdata->dptr = (uint8_t *)talloc_memdup(
706 mem_ctx, reply->data, reply->datalen))) {
708 return NT_STATUS_NO_MEMORY;
710 outdata->dsize = reply->datalen;
713 (*cstatus) = reply->status;
716 status = NT_STATUS_OK;
724 * see if a remote process exists
726 bool ctdbd_process_exists(struct ctdbd_connection *conn, uint32_t vnn, pid_t pid)
734 if (!ctdb_processes_exist(conn, &id, 1, &result)) {
735 DEBUG(10, ("ctdb_processes_exist failed\n"));
741 bool ctdb_processes_exist(struct ctdbd_connection *conn,
742 const struct server_id *pids, int num_pids,
745 TALLOC_CTX *frame = talloc_stackframe();
750 reqids = talloc_array(talloc_tos(), uint32_t, num_pids);
751 if (reqids == NULL) {
755 for (i=0; i<num_pids; i++) {
756 struct ctdb_req_control req;
762 reqids[i] = ctdbd_next_reqid(conn);
767 * pids[i].pid is uint64_t, scale down to pid_t which
768 * is the wire protocol towards ctdb.
772 DEBUG(10, ("Requesting PID %d/%d, reqid=%d\n",
773 (int)pids[i].vnn, (int)pid,
776 req.hdr.length = offsetof(struct ctdb_req_control, data);
777 req.hdr.length += sizeof(pid);
778 req.hdr.ctdb_magic = CTDB_MAGIC;
779 req.hdr.ctdb_version = CTDB_PROTOCOL;
780 req.hdr.operation = CTDB_REQ_CONTROL;
781 req.hdr.reqid = reqids[i];
782 req.hdr.destnode = pids[i].vnn;
783 req.opcode = CTDB_CONTROL_PROCESS_EXISTS;
785 req.datalen = sizeof(pid);
788 DEBUG(10, ("ctdbd_control: Sending ctdb packet\n"));
789 ctdb_packet_dump(&req.hdr);
791 iov[0].iov_base = &req;
792 iov[0].iov_len = offsetof(struct ctdb_req_control, data);
793 iov[1].iov_base = &pid;
794 iov[1].iov_len = sizeof(pid);
796 nwritten = write_data_iov(conn->fd, iov, ARRAY_SIZE(iov));
797 if (nwritten == -1) {
798 DEBUG(10, ("write_data_iov failed: %s\n",
806 while (num_received < num_pids) {
807 struct ctdb_req_header *hdr;
808 struct ctdb_reply_control *reply;
812 ret = ctdb_read_req(conn, 0, talloc_tos(), &hdr);
814 DEBUG(10, ("ctdb_read_req failed: %s\n",
819 if (hdr->operation != CTDB_REPLY_CONTROL) {
820 DEBUG(10, ("Received invalid reply\n"));
823 reply = (struct ctdb_reply_control *)hdr;
825 reqid = reply->hdr.reqid;
827 DEBUG(10, ("Received reqid %d\n", (int)reqid));
829 for (i=0; i<num_pids; i++) {
830 if (reqid == reqids[i]) {
835 DEBUG(10, ("Received unknown record number %u\n",
839 results[i] = ((reply->status) == 0);
853 char *ctdbd_dbpath(struct ctdbd_connection *conn,
854 TALLOC_CTX *mem_ctx, uint32_t db_id)
858 TDB_DATA rdata = {0};
861 data.dptr = (uint8_t*)&db_id;
862 data.dsize = sizeof(db_id);
864 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
865 CTDB_CONTROL_GETDBPATH, 0, 0, data,
866 mem_ctx, &rdata, &cstatus);
867 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
868 DEBUG(0,(__location__ " ctdb_control for getdbpath failed\n"));
872 return (char *)rdata.dptr;
876 * attach to a ctdb database
878 NTSTATUS ctdbd_db_attach(struct ctdbd_connection *conn,
879 const char *name, uint32_t *db_id, int tdb_flags)
884 bool persistent = (tdb_flags & TDB_CLEAR_IF_FIRST) == 0;
886 data = string_term_tdb_data(name);
888 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
890 ? CTDB_CONTROL_DB_ATTACH_PERSISTENT
891 : CTDB_CONTROL_DB_ATTACH,
892 tdb_flags, 0, data, NULL, &data, &cstatus);
893 if (!NT_STATUS_IS_OK(status)) {
894 DEBUG(0, (__location__ " ctdb_control for db_attach "
895 "failed: %s\n", nt_errstr(status)));
899 if (cstatus != 0 || data.dsize != sizeof(uint32_t)) {
900 DEBUG(0,(__location__ " ctdb_control for db_attach failed\n"));
901 return NT_STATUS_INTERNAL_ERROR;
904 *db_id = *(uint32_t *)data.dptr;
905 talloc_free(data.dptr);
907 if (!(tdb_flags & TDB_SEQNUM)) {
911 data.dptr = (uint8_t *)db_id;
912 data.dsize = sizeof(*db_id);
914 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
915 CTDB_CONTROL_ENABLE_SEQNUM, 0, 0, data,
916 NULL, NULL, &cstatus);
917 if (!NT_STATUS_IS_OK(status) || cstatus != 0) {
918 DEBUG(0,(__location__ " ctdb_control for enable seqnum "
920 return NT_STATUS_IS_OK(status) ? NT_STATUS_INTERNAL_ERROR :
928 * force the migration of a record to this node
930 NTSTATUS ctdbd_migrate(struct ctdbd_connection *conn, uint32_t db_id,
933 struct ctdb_req_call req;
934 struct ctdb_req_header *hdr;
942 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
943 req.hdr.ctdb_magic = CTDB_MAGIC;
944 req.hdr.ctdb_version = CTDB_PROTOCOL;
945 req.hdr.operation = CTDB_REQ_CALL;
946 req.hdr.reqid = ctdbd_next_reqid(conn);
947 req.flags = CTDB_IMMEDIATE_MIGRATION;
948 req.callid = CTDB_NULL_FUNC;
950 req.keylen = key.dsize;
952 DEBUG(10, ("ctdbd_migrate: Sending ctdb packet\n"));
953 ctdb_packet_dump(&req.hdr);
955 iov[0].iov_base = &req;
956 iov[0].iov_len = offsetof(struct ctdb_req_call, data);
957 iov[1].iov_base = key.dptr;
958 iov[1].iov_len = key.dsize;
960 nwritten = write_data_iov(conn->fd, iov, ARRAY_SIZE(iov));
961 if (nwritten == -1) {
962 DEBUG(3, ("write_data_iov failed: %s\n", strerror(errno)));
963 cluster_fatal("cluster dispatch daemon msg write error\n");
966 ret = ctdb_read_req(conn, req.hdr.reqid, NULL, &hdr);
968 DEBUG(10, ("ctdb_read_req failed: %s\n", strerror(ret)));
969 status = map_nt_error_from_unix(ret);
973 if (hdr->operation != CTDB_REPLY_CALL) {
974 DEBUG(0, ("received invalid reply\n"));
975 status = NT_STATUS_INTERNAL_ERROR;
979 status = NT_STATUS_OK;
987 * Fetch a record and parse it
989 NTSTATUS ctdbd_parse(struct ctdbd_connection *conn, uint32_t db_id,
990 TDB_DATA key, bool local_copy,
991 void (*parser)(TDB_DATA key, TDB_DATA data,
995 struct ctdb_req_call req;
996 struct ctdb_req_header *hdr = NULL;
997 struct ctdb_reply_call *reply;
1004 flags = local_copy ? CTDB_WANT_READONLY : 0;
1008 req.hdr.length = offsetof(struct ctdb_req_call, data) + key.dsize;
1009 req.hdr.ctdb_magic = CTDB_MAGIC;
1010 req.hdr.ctdb_version = CTDB_PROTOCOL;
1011 req.hdr.operation = CTDB_REQ_CALL;
1012 req.hdr.reqid = ctdbd_next_reqid(conn);
1014 req.callid = CTDB_FETCH_FUNC;
1016 req.keylen = key.dsize;
1018 iov[0].iov_base = &req;
1019 iov[0].iov_len = offsetof(struct ctdb_req_call, data);
1020 iov[1].iov_base = key.dptr;
1021 iov[1].iov_len = key.dsize;
1023 nwritten = write_data_iov(conn->fd, iov, ARRAY_SIZE(iov));
1024 if (nwritten == -1) {
1025 DEBUG(3, ("write_data_iov failed: %s\n", strerror(errno)));
1026 cluster_fatal("cluster dispatch daemon msg write error\n");
1029 ret = ctdb_read_req(conn, req.hdr.reqid, NULL, &hdr);
1031 DEBUG(10, ("ctdb_read_req failed: %s\n", strerror(ret)));
1032 status = map_nt_error_from_unix(ret);
1036 if ((hdr == NULL) || (hdr->operation != CTDB_REPLY_CALL)) {
1037 DEBUG(0, ("received invalid reply\n"));
1038 status = NT_STATUS_INTERNAL_ERROR;
1041 reply = (struct ctdb_reply_call *)hdr;
1043 if (reply->datalen == 0) {
1045 * Treat an empty record as non-existing
1047 status = NT_STATUS_NOT_FOUND;
1051 parser(key, make_tdb_data(&reply->data[0], reply->datalen),
1054 status = NT_STATUS_OK;
1061 Traverse a ctdb database. This uses a kind-of hackish way to open a second
1062 connection to ctdbd to avoid the hairy recursive and async problems with
1066 NTSTATUS ctdbd_traverse(struct ctdbd_connection *master, uint32_t db_id,
1067 void (*fn)(TDB_DATA key, TDB_DATA data,
1068 void *private_data),
1071 struct ctdbd_connection *conn;
1075 struct ctdb_traverse_start t;
1079 status = ctdbd_init_connection(NULL, master->sockname, master->timeout,
1082 if (!NT_STATUS_IS_OK(status)) {
1083 DEBUG(0, ("ctdbd_init_connection failed: %s\n",
1084 nt_errstr(status)));
1089 t.srvid = conn->rand_srvid;
1090 t.reqid = ctdbd_next_reqid(conn);
1092 data.dptr = (uint8_t *)&t;
1093 data.dsize = sizeof(t);
1095 status = ctdbd_control(conn, CTDB_CURRENT_NODE,
1096 CTDB_CONTROL_TRAVERSE_START, conn->rand_srvid, 0,
1097 data, NULL, NULL, &cstatus);
1099 if (!NT_STATUS_IS_OK(status) || (cstatus != 0)) {
1101 DEBUG(0,("ctdbd_control failed: %s, %d\n", nt_errstr(status),
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;
1120 ret = ctdb_read_packet(conn->fd, conn->timeout, conn, &hdr);
1122 DEBUG(0, ("ctdb_read_packet failed: %s\n",
1124 cluster_fatal("ctdbd died\n");
1127 if (hdr->operation != CTDB_REQ_MESSAGE) {
1128 DEBUG(0, ("Got operation %u, expected a message\n",
1129 (unsigned)hdr->operation));
1131 return NT_STATUS_UNEXPECTED_IO_ERROR;
1134 m = (struct ctdb_req_message *)hdr;
1135 d = (struct ctdb_rec_data *)&m->data[0];
1136 if (m->datalen < sizeof(uint32_t) || m->datalen != d->length) {
1137 DEBUG(0, ("Got invalid traverse data of length %d\n",
1140 return NT_STATUS_UNEXPECTED_IO_ERROR;
1143 key.dsize = d->keylen;
1144 key.dptr = &d->data[0];
1145 data.dsize = d->datalen;
1146 data.dptr = &d->data[d->keylen];
1148 if (key.dsize == 0 && data.dsize == 0) {
1149 /* end of traverse */
1151 return NT_STATUS_OK;
1154 if (data.dsize < sizeof(struct ctdb_ltdb_header)) {
1155 DEBUG(0, ("Got invalid ltdb header length %d\n",
1158 return NT_STATUS_UNEXPECTED_IO_ERROR;
1160 data.dsize -= sizeof(struct ctdb_ltdb_header);
1161 data.dptr += sizeof(struct ctdb_ltdb_header);
1164 fn(key, data, private_data);
1167 return NT_STATUS_OK;
1171 This is used to canonicalize a ctdb_sock_addr structure.
1173 static void smbd_ctdb_canonicalize_ip(const struct sockaddr_storage *in,
1174 struct sockaddr_storage *out)
1176 memcpy(out, in, sizeof (*out));
1179 if (in->ss_family == AF_INET6) {
1180 const char prefix[12] = { 0,0,0,0,0,0,0,0,0,0,0xff,0xff };
1181 const struct sockaddr_in6 *in6 =
1182 (const struct sockaddr_in6 *)in;
1183 struct sockaddr_in *out4 = (struct sockaddr_in *)out;
1184 if (memcmp(&in6->sin6_addr, prefix, 12) == 0) {
1185 memset(out, 0, sizeof(*out));
1186 #ifdef HAVE_SOCK_SIN_LEN
1187 out4->sin_len = sizeof(*out);
1189 out4->sin_family = AF_INET;
1190 out4->sin_port = in6->sin6_port;
1191 memcpy(&out4->sin_addr, &in6->sin6_addr.s6_addr[12], 4);
1198 * Register us as a server for a particular tcp connection
1201 NTSTATUS ctdbd_register_ips(struct ctdbd_connection *conn,
1202 const struct sockaddr_storage *_server,
1203 const struct sockaddr_storage *_client,
1204 int (*cb)(uint32_t src_vnn, uint32_t dst_vnn,
1206 const uint8_t *msg, size_t msglen,
1207 void *private_data),
1210 struct ctdb_control_tcp_addr p;
1211 TDB_DATA data = { .dptr = (uint8_t *)&p, .dsize = sizeof(p) };
1213 struct sockaddr_storage client;
1214 struct sockaddr_storage server;
1217 * Only one connection so far
1220 smbd_ctdb_canonicalize_ip(_client, &client);
1221 smbd_ctdb_canonicalize_ip(_server, &server);
1223 switch (client.ss_family) {
1225 memcpy(&p.dest.ip, &server, sizeof(p.dest.ip));
1226 memcpy(&p.src.ip, &client, sizeof(p.src.ip));
1229 memcpy(&p.dest.ip6, &server, sizeof(p.dest.ip6));
1230 memcpy(&p.src.ip6, &client, sizeof(p.src.ip6));
1233 return NT_STATUS_INTERNAL_ERROR;
1237 * We want to be told about IP releases
1240 status = register_with_ctdbd(conn, CTDB_SRVID_RELEASE_IP,
1242 if (!NT_STATUS_IS_OK(status)) {
1247 * inform ctdb of our tcp connection, so if IP takeover happens ctdb
1248 * can send an extra ack to trigger a reset for our client, so it
1249 * immediately reconnects
1251 return ctdbd_control(conn, CTDB_CURRENT_NODE,
1252 CTDB_CONTROL_TCP_CLIENT, 0,
1253 CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL, NULL);
1257 call a control on the local node
1259 NTSTATUS ctdbd_control_local(struct ctdbd_connection *conn, uint32_t opcode,
1260 uint64_t srvid, uint32_t flags, TDB_DATA data,
1261 TALLOC_CTX *mem_ctx, TDB_DATA *outdata,
1264 return ctdbd_control(conn, CTDB_CURRENT_NODE, opcode, srvid, flags, data, mem_ctx, outdata, cstatus);
1267 NTSTATUS ctdb_watch_us(struct ctdbd_connection *conn)
1269 struct ctdb_client_notify_register reg_data;
1274 reg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
1276 reg_data.notify_data[0] = 0;
1278 struct_len = offsetof(struct ctdb_client_notify_register,
1279 notify_data) + reg_data.len;
1281 status = ctdbd_control_local(
1282 conn, CTDB_CONTROL_REGISTER_NOTIFY, conn->rand_srvid, 0,
1283 make_tdb_data((uint8_t *)®_data, struct_len),
1284 NULL, NULL, &cstatus);
1285 if (!NT_STATUS_IS_OK(status)) {
1286 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1287 nt_errstr(status)));
1292 NTSTATUS ctdb_unwatch(struct ctdbd_connection *conn)
1294 struct ctdb_client_notify_deregister dereg_data;
1298 dereg_data.srvid = CTDB_SRVID_SAMBA_NOTIFY;
1300 status = ctdbd_control_local(
1301 conn, CTDB_CONTROL_DEREGISTER_NOTIFY, conn->rand_srvid, 0,
1302 make_tdb_data((uint8_t *)&dereg_data, sizeof(dereg_data)),
1303 NULL, NULL, &cstatus);
1304 if (!NT_STATUS_IS_OK(status)) {
1305 DEBUG(1, ("ctdbd_control_local failed: %s\n",
1306 nt_errstr(status)));
1311 NTSTATUS ctdbd_probe(const char *sockname, int timeout)
1314 * Do a very early check if ctdbd is around to avoid an abort and core
1317 struct ctdbd_connection *conn = NULL;
1320 status = ctdbd_messaging_connection(talloc_tos(), sockname, timeout,
1324 * We only care if we can connect.