4 Copyright (C) Ronnie Sahlberg 2007
5 Copyright (C) Andrew Tridgell 2007
6 Copyright (C) Martin Schwenke 2011
8 This program is free software; you can redistribute it and/or modify
9 it under the terms of the GNU General Public License as published by
10 the Free Software Foundation; either version 3 of the License, or
11 (at your option) any later version.
13 This program is distributed in the hope that it will be useful,
14 but WITHOUT ANY WARRANTY; without even the implied warranty of
15 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 GNU General Public License for more details.
18 You should have received a copy of the GNU General Public License
19 along with this program; if not, see <http://www.gnu.org/licenses/>.
22 #include "system/network.h"
23 #include "system/filesys.h"
24 #include "system/time.h"
25 #include "system/wait.h"
30 #include "lib/util/dlinklist.h"
31 #include "lib/util/debug.h"
32 #include "lib/util/samba_util.h"
33 #include "lib/util/util_process.h"
35 #include "ctdb_private.h"
36 #include "ctdb_client.h"
38 #include "common/rb_tree.h"
39 #include "common/reqid.h"
40 #include "common/system.h"
41 #include "common/common.h"
42 #include "common/logging.h"
44 #include "server/ipalloc.h"
46 #define TAKEOVER_TIMEOUT() timeval_current_ofs(ctdb->tunable.takeover_timeout,0)
48 #define CTDB_ARP_INTERVAL 1
49 #define CTDB_ARP_REPEAT 3
51 struct ctdb_interface {
52 struct ctdb_interface *prev, *next;
58 static const char *ctdb_vnn_iface_string(const struct ctdb_vnn *vnn)
61 return vnn->iface->name;
67 static int ctdb_add_local_iface(struct ctdb_context *ctdb, const char *iface)
69 struct ctdb_interface *i;
71 if (strlen(iface) > CTDB_IFACE_SIZE) {
72 DEBUG(DEBUG_ERR, ("Interface name too long \"%s\"\n", iface));
76 /* Verify that we don't have an entry for this ip yet */
77 for (i=ctdb->ifaces;i;i=i->next) {
78 if (strcmp(i->name, iface) == 0) {
83 /* create a new structure for this interface */
84 i = talloc_zero(ctdb, struct ctdb_interface);
85 CTDB_NO_MEMORY_FATAL(ctdb, i);
86 i->name = talloc_strdup(i, iface);
87 CTDB_NO_MEMORY(ctdb, i->name);
91 DLIST_ADD(ctdb->ifaces, i);
96 static bool vnn_has_interface_with_name(struct ctdb_vnn *vnn,
101 for (n = 0; vnn->ifaces[n] != NULL; n++) {
102 if (strcmp(name, vnn->ifaces[n]) == 0) {
110 /* If any interfaces now have no possible IPs then delete them. This
111 * implementation is naive (i.e. simple) rather than clever
112 * (i.e. complex). Given that this is run on delip and that operation
113 * is rare, this doesn't need to be efficient - it needs to be
114 * foolproof. One alternative is reference counting, where the logic
115 * is distributed and can, therefore, be broken in multiple places.
116 * Another alternative is to build a red-black tree of interfaces that
117 * can have addresses (by walking ctdb->vnn once) and then walking
118 * ctdb->ifaces once and deleting those not in the tree. Let's go to
119 * one of those if the naive implementation causes problems... :-)
121 static void ctdb_remove_orphaned_ifaces(struct ctdb_context *ctdb,
122 struct ctdb_vnn *vnn)
124 struct ctdb_interface *i, *next;
126 /* For each interface, check if there's an IP using it. */
127 for (i = ctdb->ifaces; i != NULL; i = next) {
132 /* Only consider interfaces named in the given VNN. */
133 if (!vnn_has_interface_with_name(vnn, i->name)) {
137 /* Search for a vnn with this interface. */
139 for (tv=ctdb->vnn; tv; tv=tv->next) {
140 if (vnn_has_interface_with_name(tv, i->name)) {
147 /* None of the VNNs are using this interface. */
148 DLIST_REMOVE(ctdb->ifaces, i);
155 static struct ctdb_interface *ctdb_find_iface(struct ctdb_context *ctdb,
158 struct ctdb_interface *i;
160 for (i=ctdb->ifaces;i;i=i->next) {
161 if (strcmp(i->name, iface) == 0) {
169 static struct ctdb_interface *ctdb_vnn_best_iface(struct ctdb_context *ctdb,
170 struct ctdb_vnn *vnn)
173 struct ctdb_interface *cur = NULL;
174 struct ctdb_interface *best = NULL;
176 for (i=0; vnn->ifaces[i]; i++) {
178 cur = ctdb_find_iface(ctdb, vnn->ifaces[i]);
192 if (cur->references < best->references) {
201 static int32_t ctdb_vnn_assign_iface(struct ctdb_context *ctdb,
202 struct ctdb_vnn *vnn)
204 struct ctdb_interface *best = NULL;
207 DEBUG(DEBUG_INFO, (__location__ " public address '%s' "
208 "still assigned to iface '%s'\n",
209 ctdb_addr_to_str(&vnn->public_address),
210 ctdb_vnn_iface_string(vnn)));
214 best = ctdb_vnn_best_iface(ctdb, vnn);
216 DEBUG(DEBUG_ERR, (__location__ " public address '%s' "
217 "cannot assign to iface any iface\n",
218 ctdb_addr_to_str(&vnn->public_address)));
224 vnn->pnn = ctdb->pnn;
226 DEBUG(DEBUG_INFO, (__location__ " public address '%s' "
227 "now assigned to iface '%s' refs[%d]\n",
228 ctdb_addr_to_str(&vnn->public_address),
229 ctdb_vnn_iface_string(vnn),
234 static void ctdb_vnn_unassign_iface(struct ctdb_context *ctdb,
235 struct ctdb_vnn *vnn)
237 DEBUG(DEBUG_INFO, (__location__ " public address '%s' "
238 "now unassigned (old iface '%s' refs[%d])\n",
239 ctdb_addr_to_str(&vnn->public_address),
240 ctdb_vnn_iface_string(vnn),
241 vnn->iface?vnn->iface->references:0));
243 vnn->iface->references--;
246 if (vnn->pnn == ctdb->pnn) {
251 static bool ctdb_vnn_available(struct ctdb_context *ctdb,
252 struct ctdb_vnn *vnn)
256 /* Nodes that are not RUNNING can not host IPs */
257 if (ctdb->runstate != CTDB_RUNSTATE_RUNNING) {
261 if (vnn->delete_pending) {
265 if (vnn->iface && vnn->iface->link_up) {
269 for (i=0; vnn->ifaces[i]; i++) {
270 struct ctdb_interface *cur;
272 cur = ctdb_find_iface(ctdb, vnn->ifaces[i]);
285 struct ctdb_takeover_arp {
286 struct ctdb_context *ctdb;
289 struct ctdb_tcp_array *tcparray;
290 struct ctdb_vnn *vnn;
295 lists of tcp endpoints
297 struct ctdb_tcp_list {
298 struct ctdb_tcp_list *prev, *next;
299 struct ctdb_connection connection;
303 list of clients to kill on IP release
305 struct ctdb_client_ip {
306 struct ctdb_client_ip *prev, *next;
307 struct ctdb_context *ctdb;
314 send a gratuitous arp
316 static void ctdb_control_send_arp(struct tevent_context *ev,
317 struct tevent_timer *te,
318 struct timeval t, void *private_data)
320 struct ctdb_takeover_arp *arp = talloc_get_type(private_data,
321 struct ctdb_takeover_arp);
323 struct ctdb_tcp_array *tcparray;
324 const char *iface = ctdb_vnn_iface_string(arp->vnn);
326 ret = ctdb_sys_send_arp(&arp->addr, iface);
328 DEBUG(DEBUG_CRIT,(__location__ " sending of arp failed on iface '%s' (%s)\n",
329 iface, strerror(errno)));
332 tcparray = arp->tcparray;
334 for (i=0;i<tcparray->num;i++) {
335 struct ctdb_connection *tcon;
337 tcon = &tcparray->connections[i];
338 DEBUG(DEBUG_INFO,("sending tcp tickle ack for %u->%s:%u\n",
339 (unsigned)ntohs(tcon->dst.ip.sin_port),
340 ctdb_addr_to_str(&tcon->src),
341 (unsigned)ntohs(tcon->src.ip.sin_port)));
342 ret = ctdb_sys_send_tcp(
347 DEBUG(DEBUG_CRIT,(__location__ " Failed to send tcp tickle ack for %s\n",
348 ctdb_addr_to_str(&tcon->src)));
355 if (arp->count == CTDB_ARP_REPEAT) {
360 tevent_add_timer(arp->ctdb->ev, arp->vnn->takeover_ctx,
361 timeval_current_ofs(CTDB_ARP_INTERVAL, 100000),
362 ctdb_control_send_arp, arp);
365 static int32_t ctdb_announce_vnn_iface(struct ctdb_context *ctdb,
366 struct ctdb_vnn *vnn)
368 struct ctdb_takeover_arp *arp;
369 struct ctdb_tcp_array *tcparray;
371 if (!vnn->takeover_ctx) {
372 vnn->takeover_ctx = talloc_new(vnn);
373 if (!vnn->takeover_ctx) {
378 arp = talloc_zero(vnn->takeover_ctx, struct ctdb_takeover_arp);
384 arp->addr = vnn->public_address;
387 tcparray = vnn->tcp_array;
389 /* add all of the known tcp connections for this IP to the
390 list of tcp connections to send tickle acks for */
391 arp->tcparray = talloc_steal(arp, tcparray);
393 vnn->tcp_array = NULL;
394 vnn->tcp_update_needed = true;
397 tevent_add_timer(arp->ctdb->ev, vnn->takeover_ctx,
398 timeval_zero(), ctdb_control_send_arp, arp);
403 struct takeover_callback_state {
404 struct ctdb_req_control_old *c;
405 ctdb_sock_addr *addr;
406 struct ctdb_vnn *vnn;
409 struct ctdb_do_takeip_state {
410 struct ctdb_req_control_old *c;
411 struct ctdb_vnn *vnn;
415 called when takeip event finishes
417 static void ctdb_do_takeip_callback(struct ctdb_context *ctdb, int status,
420 struct ctdb_do_takeip_state *state =
421 talloc_get_type(private_data, struct ctdb_do_takeip_state);
426 if (status == -ETIME) {
429 DEBUG(DEBUG_ERR,(__location__ " Failed to takeover IP %s on interface %s\n",
430 ctdb_addr_to_str(&state->vnn->public_address),
431 ctdb_vnn_iface_string(state->vnn)));
432 ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
438 if (ctdb->do_checkpublicip) {
440 ret = ctdb_announce_vnn_iface(ctdb, state->vnn);
442 ctdb_request_control_reply(ctdb, state->c, NULL, -1, NULL);
449 data.dptr = (uint8_t *)ctdb_addr_to_str(&state->vnn->public_address);
450 data.dsize = strlen((char *)data.dptr) + 1;
451 DEBUG(DEBUG_INFO,(__location__ " sending TAKE_IP for '%s'\n", data.dptr));
453 ctdb_daemon_send_message(ctdb, ctdb->pnn, CTDB_SRVID_TAKE_IP, data);
456 /* the control succeeded */
457 ctdb_request_control_reply(ctdb, state->c, NULL, 0, NULL);
462 static int ctdb_takeip_destructor(struct ctdb_do_takeip_state *state)
464 state->vnn->update_in_flight = false;
469 take over an ip address
471 static int32_t ctdb_do_takeip(struct ctdb_context *ctdb,
472 struct ctdb_req_control_old *c,
473 struct ctdb_vnn *vnn)
476 struct ctdb_do_takeip_state *state;
478 if (vnn->update_in_flight) {
479 DEBUG(DEBUG_NOTICE,("Takeover of IP %s/%u rejected "
480 "update for this IP already in flight\n",
481 ctdb_addr_to_str(&vnn->public_address),
482 vnn->public_netmask_bits));
486 ret = ctdb_vnn_assign_iface(ctdb, vnn);
488 DEBUG(DEBUG_ERR,("Takeover of IP %s/%u failed to "
489 "assign a usable interface\n",
490 ctdb_addr_to_str(&vnn->public_address),
491 vnn->public_netmask_bits));
495 state = talloc(vnn, struct ctdb_do_takeip_state);
496 CTDB_NO_MEMORY(ctdb, state);
498 state->c = talloc_steal(ctdb, c);
501 vnn->update_in_flight = true;
502 talloc_set_destructor(state, ctdb_takeip_destructor);
504 DEBUG(DEBUG_NOTICE,("Takeover of IP %s/%u on interface %s\n",
505 ctdb_addr_to_str(&vnn->public_address),
506 vnn->public_netmask_bits,
507 ctdb_vnn_iface_string(vnn)));
509 ret = ctdb_event_script_callback(ctdb,
511 ctdb_do_takeip_callback,
515 ctdb_vnn_iface_string(vnn),
516 ctdb_addr_to_str(&vnn->public_address),
517 vnn->public_netmask_bits);
520 DEBUG(DEBUG_ERR,(__location__ " Failed to takeover IP %s on interface %s\n",
521 ctdb_addr_to_str(&vnn->public_address),
522 ctdb_vnn_iface_string(vnn)));
530 struct ctdb_do_updateip_state {
531 struct ctdb_req_control_old *c;
532 struct ctdb_interface *old;
533 struct ctdb_vnn *vnn;
537 called when updateip event finishes
539 static void ctdb_do_updateip_callback(struct ctdb_context *ctdb, int status,
542 struct ctdb_do_updateip_state *state =
543 talloc_get_type(private_data, struct ctdb_do_updateip_state);
547 if (status == -ETIME) {
550 DEBUG(DEBUG_ERR,(__location__ " Failed to move IP %s from interface %s to %s\n",
551 ctdb_addr_to_str(&state->vnn->public_address),
553 ctdb_vnn_iface_string(state->vnn)));
556 * All we can do is reset the old interface
557 * and let the next run fix it
559 ctdb_vnn_unassign_iface(ctdb, state->vnn);
560 state->vnn->iface = state->old;
561 state->vnn->iface->references++;
563 ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
568 if (ctdb->do_checkpublicip) {
570 ret = ctdb_announce_vnn_iface(ctdb, state->vnn);
572 ctdb_request_control_reply(ctdb, state->c, NULL, -1, NULL);
579 /* the control succeeded */
580 ctdb_request_control_reply(ctdb, state->c, NULL, 0, NULL);
585 static int ctdb_updateip_destructor(struct ctdb_do_updateip_state *state)
587 state->vnn->update_in_flight = false;
592 update (move) an ip address
594 static int32_t ctdb_do_updateip(struct ctdb_context *ctdb,
595 struct ctdb_req_control_old *c,
596 struct ctdb_vnn *vnn)
599 struct ctdb_do_updateip_state *state;
600 struct ctdb_interface *old = vnn->iface;
601 const char *new_name;
603 if (vnn->update_in_flight) {
604 DEBUG(DEBUG_NOTICE,("Update of IP %s/%u rejected "
605 "update for this IP already in flight\n",
606 ctdb_addr_to_str(&vnn->public_address),
607 vnn->public_netmask_bits));
611 ctdb_vnn_unassign_iface(ctdb, vnn);
612 ret = ctdb_vnn_assign_iface(ctdb, vnn);
614 DEBUG(DEBUG_ERR,("update of IP %s/%u failed to "
615 "assin a usable interface (old iface '%s')\n",
616 ctdb_addr_to_str(&vnn->public_address),
617 vnn->public_netmask_bits,
622 new_name = ctdb_vnn_iface_string(vnn);
623 if (old->name != NULL && new_name != NULL && !strcmp(old->name, new_name)) {
624 /* A benign update from one interface onto itself.
625 * no need to run the eventscripts in this case, just return
628 ctdb_request_control_reply(ctdb, c, NULL, 0, NULL);
632 state = talloc(vnn, struct ctdb_do_updateip_state);
633 CTDB_NO_MEMORY(ctdb, state);
635 state->c = talloc_steal(ctdb, c);
639 vnn->update_in_flight = true;
640 talloc_set_destructor(state, ctdb_updateip_destructor);
642 DEBUG(DEBUG_NOTICE,("Update of IP %s/%u from "
643 "interface %s to %s\n",
644 ctdb_addr_to_str(&vnn->public_address),
645 vnn->public_netmask_bits,
649 ret = ctdb_event_script_callback(ctdb,
651 ctdb_do_updateip_callback,
653 CTDB_EVENT_UPDATE_IP,
657 ctdb_addr_to_str(&vnn->public_address),
658 vnn->public_netmask_bits);
660 DEBUG(DEBUG_ERR,(__location__ " Failed update IP %s from interface %s to %s\n",
661 ctdb_addr_to_str(&vnn->public_address),
662 old->name, new_name));
671 Find the vnn of the node that has a public ip address
672 returns -1 if the address is not known as a public address
674 static struct ctdb_vnn *find_public_ip_vnn(struct ctdb_context *ctdb, ctdb_sock_addr *addr)
676 struct ctdb_vnn *vnn;
678 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
679 if (ctdb_same_ip(&vnn->public_address, addr)) {
688 take over an ip address
690 int32_t ctdb_control_takeover_ip(struct ctdb_context *ctdb,
691 struct ctdb_req_control_old *c,
696 struct ctdb_public_ip *pip = (struct ctdb_public_ip *)indata.dptr;
697 struct ctdb_vnn *vnn;
698 bool have_ip = false;
699 bool do_updateip = false;
700 bool do_takeip = false;
701 struct ctdb_interface *best_iface = NULL;
703 if (pip->pnn != ctdb->pnn) {
704 DEBUG(DEBUG_ERR,(__location__" takeoverip called for an ip '%s' "
705 "with pnn %d, but we're node %d\n",
706 ctdb_addr_to_str(&pip->addr),
707 pip->pnn, ctdb->pnn));
711 /* update out vnn list */
712 vnn = find_public_ip_vnn(ctdb, &pip->addr);
714 DEBUG(DEBUG_INFO,("takeoverip called for an ip '%s' that is not a public address\n",
715 ctdb_addr_to_str(&pip->addr)));
719 if (ctdb->tunable.disable_ip_failover == 0 && ctdb->do_checkpublicip) {
720 have_ip = ctdb_sys_have_ip(&pip->addr);
722 best_iface = ctdb_vnn_best_iface(ctdb, vnn);
723 if (best_iface == NULL) {
724 DEBUG(DEBUG_ERR,("takeoverip of IP %s/%u failed to find"
725 "a usable interface (old %s, have_ip %d)\n",
726 ctdb_addr_to_str(&vnn->public_address),
727 vnn->public_netmask_bits,
728 ctdb_vnn_iface_string(vnn),
733 if (vnn->iface == NULL && vnn->pnn == -1 && have_ip && best_iface != NULL) {
734 DEBUG(DEBUG_ERR,("Taking over newly created ip\n"));
739 if (vnn->iface == NULL && have_ip) {
740 DEBUG(DEBUG_CRIT,(__location__ " takeoverip of IP %s is known to the kernel, "
741 "but we have no interface assigned, has someone manually configured it? Ignore for now.\n",
742 ctdb_addr_to_str(&vnn->public_address)));
746 if (vnn->pnn != ctdb->pnn && have_ip && vnn->pnn != -1) {
747 DEBUG(DEBUG_CRIT,(__location__ " takeoverip of IP %s is known to the kernel, "
748 "and we have it on iface[%s], but it was assigned to node %d"
749 "and we are node %d, banning ourself\n",
750 ctdb_addr_to_str(&vnn->public_address),
751 ctdb_vnn_iface_string(vnn), vnn->pnn, ctdb->pnn));
756 if (vnn->pnn == -1 && have_ip) {
757 vnn->pnn = ctdb->pnn;
758 DEBUG(DEBUG_CRIT,(__location__ " takeoverip of IP %s is known to the kernel, "
759 "and we already have it on iface[%s], update local daemon\n",
760 ctdb_addr_to_str(&vnn->public_address),
761 ctdb_vnn_iface_string(vnn)));
766 if (vnn->iface != best_iface) {
767 if (!vnn->iface->link_up) {
769 } else if (vnn->iface->references > (best_iface->references + 1)) {
770 /* only move when the rebalance gains something */
778 ctdb_vnn_unassign_iface(ctdb, vnn);
785 ret = ctdb_do_takeip(ctdb, c, vnn);
789 } else if (do_updateip) {
790 ret = ctdb_do_updateip(ctdb, c, vnn);
796 * The interface is up and the kernel known the ip
799 DEBUG(DEBUG_INFO,("Redundant takeover of IP %s/%u on interface %s (ip already held)\n",
800 ctdb_addr_to_str(&pip->addr),
801 vnn->public_netmask_bits,
802 ctdb_vnn_iface_string(vnn)));
806 /* tell ctdb_control.c that we will be replying asynchronously */
812 static void do_delete_ip(struct ctdb_context *ctdb, struct ctdb_vnn *vnn)
814 DLIST_REMOVE(ctdb->vnn, vnn);
815 ctdb_vnn_unassign_iface(ctdb, vnn);
816 ctdb_remove_orphaned_ifaces(ctdb, vnn);
821 called when releaseip event finishes
823 static void release_ip_callback(struct ctdb_context *ctdb, int status,
826 struct takeover_callback_state *state =
827 talloc_get_type(private_data, struct takeover_callback_state);
830 if (status == -ETIME) {
834 if (ctdb->tunable.disable_ip_failover == 0 && ctdb->do_checkpublicip) {
835 if (ctdb_sys_have_ip(state->addr)) {
837 ("IP %s still hosted during release IP callback, failing\n",
838 ctdb_addr_to_str(state->addr)));
839 ctdb_request_control_reply(ctdb, state->c,
846 /* send a message to all clients of this node telling them
847 that the cluster has been reconfigured and they should
848 release any sockets on this IP */
849 data.dptr = (uint8_t *)talloc_strdup(state, ctdb_addr_to_str(state->addr));
850 CTDB_NO_MEMORY_VOID(ctdb, data.dptr);
851 data.dsize = strlen((char *)data.dptr)+1;
853 DEBUG(DEBUG_INFO,(__location__ " sending RELEASE_IP for '%s'\n", data.dptr));
855 ctdb_daemon_send_message(ctdb, ctdb->pnn, CTDB_SRVID_RELEASE_IP, data);
857 ctdb_vnn_unassign_iface(ctdb, state->vnn);
859 /* Process the IP if it has been marked for deletion */
860 if (state->vnn->delete_pending) {
861 do_delete_ip(ctdb, state->vnn);
865 /* the control succeeded */
866 ctdb_request_control_reply(ctdb, state->c, NULL, 0, NULL);
870 static int ctdb_releaseip_destructor(struct takeover_callback_state *state)
872 if (state->vnn != NULL) {
873 state->vnn->update_in_flight = false;
879 release an ip address
881 int32_t ctdb_control_release_ip(struct ctdb_context *ctdb,
882 struct ctdb_req_control_old *c,
887 struct takeover_callback_state *state;
888 struct ctdb_public_ip *pip = (struct ctdb_public_ip *)indata.dptr;
889 struct ctdb_vnn *vnn;
892 /* update our vnn list */
893 vnn = find_public_ip_vnn(ctdb, &pip->addr);
895 DEBUG(DEBUG_INFO,("releaseip called for an ip '%s' that is not a public address\n",
896 ctdb_addr_to_str(&pip->addr)));
901 /* stop any previous arps */
902 talloc_free(vnn->takeover_ctx);
903 vnn->takeover_ctx = NULL;
905 /* Some ctdb tool commands (e.g. moveip) send
906 * lazy multicast to drop an IP from any node that isn't the
907 * intended new node. The following causes makes ctdbd ignore
908 * a release for any address it doesn't host.
910 if (ctdb->tunable.disable_ip_failover == 0 && ctdb->do_checkpublicip) {
911 if (!ctdb_sys_have_ip(&pip->addr)) {
912 DEBUG(DEBUG_DEBUG,("Redundant release of IP %s/%u on interface %s (ip not held)\n",
913 ctdb_addr_to_str(&pip->addr),
914 vnn->public_netmask_bits,
915 ctdb_vnn_iface_string(vnn)));
916 ctdb_vnn_unassign_iface(ctdb, vnn);
920 if (vnn->iface == NULL) {
921 DEBUG(DEBUG_DEBUG,("Redundant release of IP %s/%u (ip not held)\n",
922 ctdb_addr_to_str(&pip->addr),
923 vnn->public_netmask_bits));
928 /* There is a potential race between take_ip and us because we
929 * update the VNN via a callback that run when the
930 * eventscripts have been run. Avoid the race by allowing one
931 * update to be in flight at a time.
933 if (vnn->update_in_flight) {
934 DEBUG(DEBUG_NOTICE,("Release of IP %s/%u rejected "
935 "update for this IP already in flight\n",
936 ctdb_addr_to_str(&vnn->public_address),
937 vnn->public_netmask_bits));
941 iface = strdup(ctdb_vnn_iface_string(vnn));
943 DEBUG(DEBUG_NOTICE,("Release of IP %s/%u on interface %s node:%d\n",
944 ctdb_addr_to_str(&pip->addr),
945 vnn->public_netmask_bits,
949 state = talloc(ctdb, struct takeover_callback_state);
951 ctdb_set_error(ctdb, "Out of memory at %s:%d",
957 state->c = talloc_steal(state, c);
958 state->addr = talloc(state, ctdb_sock_addr);
959 if (state->addr == NULL) {
960 ctdb_set_error(ctdb, "Out of memory at %s:%d",
966 *state->addr = pip->addr;
969 vnn->update_in_flight = true;
970 talloc_set_destructor(state, ctdb_releaseip_destructor);
972 ret = ctdb_event_script_callback(ctdb,
973 state, release_ip_callback, state,
974 CTDB_EVENT_RELEASE_IP,
977 ctdb_addr_to_str(&pip->addr),
978 vnn->public_netmask_bits);
981 DEBUG(DEBUG_ERR,(__location__ " Failed to release IP %s on interface %s\n",
982 ctdb_addr_to_str(&pip->addr),
983 ctdb_vnn_iface_string(vnn)));
988 /* tell the control that we will be reply asynchronously */
993 static int ctdb_add_public_address(struct ctdb_context *ctdb,
994 ctdb_sock_addr *addr,
995 unsigned mask, const char *ifaces,
998 struct ctdb_vnn *vnn;
1005 tmp = strdup(ifaces);
1006 for (iface = strtok(tmp, ","); iface; iface = strtok(NULL, ",")) {
1007 if (!ctdb_sys_check_iface_exists(iface)) {
1008 DEBUG(DEBUG_CRIT,("Interface %s does not exist. Can not add public-address : %s\n", iface, ctdb_addr_to_str(addr)));
1015 /* Verify that we don't have an entry for this ip yet */
1016 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
1017 if (ctdb_same_sockaddr(addr, &vnn->public_address)) {
1018 DEBUG(DEBUG_CRIT,("Same ip '%s' specified multiple times in the public address list \n",
1019 ctdb_addr_to_str(addr)));
1024 /* create a new vnn structure for this ip address */
1025 vnn = talloc_zero(ctdb, struct ctdb_vnn);
1026 CTDB_NO_MEMORY_FATAL(ctdb, vnn);
1027 vnn->ifaces = talloc_array(vnn, const char *, num + 2);
1028 tmp = talloc_strdup(vnn, ifaces);
1029 CTDB_NO_MEMORY_FATAL(ctdb, tmp);
1030 for (iface = strtok(tmp, ","); iface; iface = strtok(NULL, ",")) {
1031 vnn->ifaces = talloc_realloc(vnn, vnn->ifaces, const char *, num + 2);
1032 CTDB_NO_MEMORY_FATAL(ctdb, vnn->ifaces);
1033 vnn->ifaces[num] = talloc_strdup(vnn, iface);
1034 CTDB_NO_MEMORY_FATAL(ctdb, vnn->ifaces[num]);
1038 vnn->ifaces[num] = NULL;
1039 vnn->public_address = *addr;
1040 vnn->public_netmask_bits = mask;
1042 if (check_address) {
1043 if (ctdb_sys_have_ip(addr)) {
1044 DEBUG(DEBUG_ERR,("We are already hosting public address '%s'. setting PNN to ourself:%d\n", ctdb_addr_to_str(addr), ctdb->pnn));
1045 vnn->pnn = ctdb->pnn;
1049 for (i=0; vnn->ifaces[i]; i++) {
1050 ret = ctdb_add_local_iface(ctdb, vnn->ifaces[i]);
1052 DEBUG(DEBUG_CRIT, (__location__ " failed to add iface[%s] "
1053 "for public_address[%s]\n",
1054 vnn->ifaces[i], ctdb_addr_to_str(addr)));
1060 DLIST_ADD(ctdb->vnn, vnn);
1066 setup the public address lists from a file
1068 int ctdb_set_public_addresses(struct ctdb_context *ctdb, bool check_addresses)
1074 lines = file_lines_load(ctdb->public_addresses_file, &nlines, 0, ctdb);
1075 if (lines == NULL) {
1076 ctdb_set_error(ctdb, "Failed to load public address list '%s'\n", ctdb->public_addresses_file);
1079 while (nlines > 0 && strcmp(lines[nlines-1], "") == 0) {
1083 for (i=0;i<nlines;i++) {
1085 ctdb_sock_addr addr;
1086 const char *addrstr;
1091 while ((*line == ' ') || (*line == '\t')) {
1097 if (strcmp(line, "") == 0) {
1100 tok = strtok(line, " \t");
1102 tok = strtok(NULL, " \t");
1104 if (NULL == ctdb->default_public_interface) {
1105 DEBUG(DEBUG_CRIT,("No default public interface and no interface specified at line %u of public address list\n",
1110 ifaces = ctdb->default_public_interface;
1115 if (!addrstr || !parse_ip_mask(addrstr, ifaces, &addr, &mask)) {
1116 DEBUG(DEBUG_CRIT,("Badly formed line %u in public address list\n", i+1));
1120 if (ctdb_add_public_address(ctdb, &addr, mask, ifaces, check_addresses)) {
1121 DEBUG(DEBUG_CRIT,("Failed to add line %u to the public address list\n", i+1));
1132 static void *add_ip_callback(void *parm, void *data)
1134 struct public_ip_list *this_ip = parm;
1135 struct public_ip_list *prev_ip = data;
1137 if (prev_ip == NULL) {
1140 if (this_ip->pnn == -1) {
1141 this_ip->pnn = prev_ip->pnn;
1147 static int getips_count_callback(void *param, void *data)
1149 struct public_ip_list **ip_list = (struct public_ip_list **)param;
1150 struct public_ip_list *new_ip = (struct public_ip_list *)data;
1152 new_ip->next = *ip_list;
1157 static int ctdb_reload_remote_public_ips(struct ctdb_context *ctdb,
1158 struct ipalloc_state *ipalloc_state,
1159 struct ctdb_node_map_old *nodemap)
1163 struct ctdb_public_ip_list_old *ip_list;
1165 if (ipalloc_state->num != nodemap->num) {
1168 " ipalloc_state->num (%d) != nodemap->num (%d) invalid param\n",
1169 ipalloc_state->num, nodemap->num));
1173 for (j=0; j<nodemap->num; j++) {
1174 if (nodemap->nodes[j].flags & NODE_FLAGS_INACTIVE) {
1178 /* Retrieve the list of known public IPs from the node */
1179 ret = ctdb_ctrl_get_public_ips_flags(ctdb,
1182 ipalloc_state->known_public_ips,
1187 ("Failed to read known public IPs from node: %u\n",
1191 ipalloc_state->known_public_ips[j].num = ip_list->num;
1192 /* This could be copied and freed. However, ip_list
1193 * is allocated off ipalloc_state->known_public_ips,
1194 * so this is a safe hack. This will go away in a
1195 * while anyway... */
1196 ipalloc_state->known_public_ips[j].ip = &ip_list->ips[0];
1198 /* Retrieve the list of available public IPs from the node */
1199 ret = ctdb_ctrl_get_public_ips_flags(ctdb,
1202 ipalloc_state->available_public_ips,
1203 CTDB_PUBLIC_IP_FLAGS_ONLY_AVAILABLE,
1207 ("Failed to read available public IPs from node: %u\n",
1211 ipalloc_state->available_public_ips[j].num = ip_list->num;
1212 /* This could be copied and freed. However, ip_list
1213 * is allocated off ipalloc_state->available_public_ips,
1214 * so this is a safe hack. This will go away in a
1215 * while anyway... */
1216 ipalloc_state->available_public_ips[j].ip = &ip_list->ips[0];
1222 static struct public_ip_list *
1223 create_merged_ip_list(struct ipalloc_state *ipalloc_state)
1226 struct public_ip_list *ip_list;
1227 struct ctdb_public_ip_list *public_ips;
1228 struct trbt_tree *ip_tree;
1230 ip_tree = trbt_create(ipalloc_state, 0);
1232 if (ipalloc_state->known_public_ips == NULL) {
1233 DEBUG(DEBUG_ERR, ("Known public IPs not set\n"));
1237 for (i=0; i < ipalloc_state->num; i++) {
1239 public_ips = &ipalloc_state->known_public_ips[i];
1241 for (j=0; j < public_ips->num; j++) {
1242 struct public_ip_list *tmp_ip;
1244 /* This is returned as part of ip_list */
1245 tmp_ip = talloc_zero(ipalloc_state, struct public_ip_list);
1246 if (tmp_ip == NULL) {
1248 (__location__ " out of memory\n"));
1249 talloc_free(ip_tree);
1253 /* Do not use information about IP addresses hosted
1254 * on other nodes, it may not be accurate */
1255 if (public_ips->ip[j].pnn == i) {
1256 tmp_ip->pnn = public_ips->ip[j].pnn;
1260 tmp_ip->addr = public_ips->ip[j].addr;
1261 tmp_ip->next = NULL;
1263 trbt_insertarray32_callback(ip_tree,
1264 IP_KEYLEN, ip_key(&public_ips->ip[j].addr),
1271 trbt_traversearray32(ip_tree, IP_KEYLEN, getips_count_callback, &ip_list);
1272 talloc_free(ip_tree);
1277 static bool all_nodes_are_disabled(struct ctdb_node_map_old *nodemap)
1281 for (i=0;i<nodemap->num;i++) {
1282 if (!(nodemap->nodes[i].flags & (NODE_FLAGS_INACTIVE|NODE_FLAGS_DISABLED))) {
1283 /* Found one completely healthy node */
1291 struct get_tunable_callback_data {
1292 const char *tunable;
1297 static void get_tunable_callback(struct ctdb_context *ctdb, uint32_t pnn,
1298 int32_t res, TDB_DATA outdata,
1301 struct get_tunable_callback_data *cd =
1302 (struct get_tunable_callback_data *)callback;
1306 /* Already handled in fail callback */
1310 if (outdata.dsize != sizeof(uint32_t)) {
1311 DEBUG(DEBUG_ERR,("Wrong size of returned data when reading \"%s\" tunable from node %d. Expected %d bytes but received %d bytes\n",
1312 cd->tunable, pnn, (int)sizeof(uint32_t),
1313 (int)outdata.dsize));
1318 size = talloc_array_length(cd->out);
1320 DEBUG(DEBUG_ERR,("Got %s reply from node %d but nodemap only has %d entries\n",
1321 cd->tunable, pnn, size));
1326 cd->out[pnn] = *(uint32_t *)outdata.dptr;
1329 static void get_tunable_fail_callback(struct ctdb_context *ctdb, uint32_t pnn,
1330 int32_t res, TDB_DATA outdata,
1333 struct get_tunable_callback_data *cd =
1334 (struct get_tunable_callback_data *)callback;
1339 ("Timed out getting tunable \"%s\" from node %d\n",
1345 DEBUG(DEBUG_WARNING,
1346 ("Tunable \"%s\" not implemented on node %d\n",
1351 ("Unexpected error getting tunable \"%s\" from node %d\n",
1357 static uint32_t *get_tunable_from_nodes(struct ctdb_context *ctdb,
1358 TALLOC_CTX *tmp_ctx,
1359 struct ctdb_node_map_old *nodemap,
1360 const char *tunable,
1361 uint32_t default_value)
1364 struct ctdb_control_get_tunable *t;
1367 struct get_tunable_callback_data callback_data;
1370 tvals = talloc_array(tmp_ctx, uint32_t, nodemap->num);
1371 CTDB_NO_MEMORY_NULL(ctdb, tvals);
1372 for (i=0; i<nodemap->num; i++) {
1373 tvals[i] = default_value;
1376 callback_data.out = tvals;
1377 callback_data.tunable = tunable;
1378 callback_data.fatal = false;
1380 data.dsize = offsetof(struct ctdb_control_get_tunable, name) + strlen(tunable) + 1;
1381 data.dptr = talloc_size(tmp_ctx, data.dsize);
1382 t = (struct ctdb_control_get_tunable *)data.dptr;
1383 t->length = strlen(tunable)+1;
1384 memcpy(t->name, tunable, t->length);
1385 nodes = list_of_connected_nodes(ctdb, nodemap, tmp_ctx, true);
1386 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_GET_TUNABLE,
1387 nodes, 0, TAKEOVER_TIMEOUT(),
1389 get_tunable_callback,
1390 get_tunable_fail_callback,
1391 &callback_data) != 0) {
1392 if (callback_data.fatal) {
1398 talloc_free(data.dptr);
1403 /* Set internal flags for IP allocation:
1405 * Set NOIPTAKOVER ip flags from per-node NoIPTakeover tunable
1406 * Set NOIPHOST ip flag for each INACTIVE node
1407 * if all nodes are disabled:
1408 * Set NOIPHOST ip flags from per-node NoIPHostOnAllDisabled tunable
1410 * Set NOIPHOST ip flags for disabled nodes
1412 static void set_ipflags_internal(struct ipalloc_state *ipalloc_state,
1413 struct ctdb_node_map_old *nodemap,
1414 uint32_t *tval_noiptakeover,
1415 uint32_t *tval_noiphostonalldisabled)
1419 for (i=0;i<nodemap->num;i++) {
1420 /* Can not take IPs on node with NoIPTakeover set */
1421 if (tval_noiptakeover[i] != 0) {
1422 ipalloc_state->noiptakeover[i] = true;
1425 /* Can not host IPs on INACTIVE node */
1426 if (nodemap->nodes[i].flags & NODE_FLAGS_INACTIVE) {
1427 ipalloc_state->noiphost[i] = true;
1431 if (all_nodes_are_disabled(nodemap)) {
1432 /* If all nodes are disabled, can not host IPs on node
1433 * with NoIPHostOnAllDisabled set
1435 for (i=0;i<nodemap->num;i++) {
1436 if (tval_noiphostonalldisabled[i] != 0) {
1437 ipalloc_state->noiphost[i] = true;
1441 /* If some nodes are not disabled, then can not host
1442 * IPs on DISABLED node
1444 for (i=0;i<nodemap->num;i++) {
1445 if (nodemap->nodes[i].flags & NODE_FLAGS_DISABLED) {
1446 ipalloc_state->noiphost[i] = true;
1452 static bool set_ipflags(struct ctdb_context *ctdb,
1453 struct ipalloc_state *ipalloc_state,
1454 struct ctdb_node_map_old *nodemap)
1456 uint32_t *tval_noiptakeover;
1457 uint32_t *tval_noiphostonalldisabled;
1459 tval_noiptakeover = get_tunable_from_nodes(ctdb, ipalloc_state, nodemap,
1461 if (tval_noiptakeover == NULL) {
1465 tval_noiphostonalldisabled =
1466 get_tunable_from_nodes(ctdb, ipalloc_state, nodemap,
1467 "NoIPHostOnAllDisabled", 0);
1468 if (tval_noiphostonalldisabled == NULL) {
1469 /* Caller frees tmp_ctx */
1473 set_ipflags_internal(ipalloc_state, nodemap,
1475 tval_noiphostonalldisabled);
1477 talloc_free(tval_noiptakeover);
1478 talloc_free(tval_noiphostonalldisabled);
1483 static struct ipalloc_state * ipalloc_state_init(struct ctdb_context *ctdb,
1484 TALLOC_CTX *mem_ctx)
1486 struct ipalloc_state *ipalloc_state =
1487 talloc_zero(mem_ctx, struct ipalloc_state);
1488 if (ipalloc_state == NULL) {
1489 DEBUG(DEBUG_ERR, (__location__ " Out of memory\n"));
1493 ipalloc_state->num = ctdb->num_nodes;
1495 ipalloc_state->known_public_ips =
1496 talloc_zero_array(ipalloc_state,
1497 struct ctdb_public_ip_list,
1498 ipalloc_state->num);
1499 if (ipalloc_state->known_public_ips == NULL) {
1500 DEBUG(DEBUG_ERR, (__location__ " Out of memory\n"));
1504 ipalloc_state->available_public_ips =
1505 talloc_zero_array(ipalloc_state,
1506 struct ctdb_public_ip_list,
1507 ipalloc_state->num);
1508 if (ipalloc_state->available_public_ips == NULL) {
1509 DEBUG(DEBUG_ERR, (__location__ " Out of memory\n"));
1512 ipalloc_state->noiptakeover =
1513 talloc_zero_array(ipalloc_state,
1515 ipalloc_state->num);
1516 if (ipalloc_state->noiptakeover == NULL) {
1517 DEBUG(DEBUG_ERR, (__location__ " Out of memory\n"));
1520 ipalloc_state->noiphost =
1521 talloc_zero_array(ipalloc_state,
1523 ipalloc_state->num);
1524 if (ipalloc_state->noiphost == NULL) {
1525 DEBUG(DEBUG_ERR, (__location__ " Out of memory\n"));
1529 if (1 == ctdb->tunable.lcp2_public_ip_assignment) {
1530 ipalloc_state->algorithm = IPALLOC_LCP2;
1531 } else if (1 == ctdb->tunable.deterministic_public_ips) {
1532 ipalloc_state->algorithm = IPALLOC_DETERMINISTIC;
1534 ipalloc_state->algorithm = IPALLOC_NONDETERMINISTIC;
1537 ipalloc_state->no_ip_failback = ctdb->tunable.no_ip_failback;
1539 return ipalloc_state;
1541 talloc_free(ipalloc_state);
1545 struct takeover_callback_data {
1547 unsigned int *fail_count;
1550 static struct takeover_callback_data *
1551 takeover_callback_data_init(TALLOC_CTX *mem_ctx,
1554 static struct takeover_callback_data *takeover_data;
1556 takeover_data = talloc_zero(mem_ctx, struct takeover_callback_data);
1557 if (takeover_data == NULL) {
1558 DEBUG(DEBUG_ERR, (__location__ " out of memory\n"));
1562 takeover_data->fail_count = talloc_zero_array(takeover_data,
1563 unsigned int, num_nodes);
1564 if (takeover_data->fail_count == NULL) {
1565 DEBUG(DEBUG_ERR, (__location__ " out of memory\n"));
1566 talloc_free(takeover_data);
1570 takeover_data->num_nodes = num_nodes;
1572 return takeover_data;
1575 static void takeover_run_fail_callback(struct ctdb_context *ctdb,
1576 uint32_t node_pnn, int32_t res,
1577 TDB_DATA outdata, void *callback_data)
1579 struct takeover_callback_data *cd =
1580 talloc_get_type_abort(callback_data,
1581 struct takeover_callback_data);
1583 if (node_pnn >= cd->num_nodes) {
1584 DEBUG(DEBUG_ERR, (__location__ " invalid PNN %u\n", node_pnn));
1588 if (cd->fail_count[node_pnn] == 0) {
1590 ("Node %u failed the takeover run\n", node_pnn));
1593 cd->fail_count[node_pnn]++;
1596 static void takeover_run_process_failures(struct ctdb_context *ctdb,
1597 struct takeover_callback_data *tcd)
1599 unsigned int max_fails = 0;
1600 uint32_t max_pnn = -1;
1603 for (i = 0; i < tcd->num_nodes; i++) {
1604 if (tcd->fail_count[i] > max_fails) {
1606 max_fails = tcd->fail_count[i];
1610 if (max_fails > 0) {
1615 ("Sending banning credits to %u with fail count %u\n",
1616 max_pnn, max_fails));
1618 data.dptr = (uint8_t *)&max_pnn;
1619 data.dsize = sizeof(uint32_t);
1620 ret = ctdb_client_send_message(ctdb,
1621 CTDB_BROADCAST_CONNECTED,
1626 ("Failed to set banning credits for node %u\n",
1633 * Recalculate the allocation of public IPs to nodes and have the
1634 * nodes host their allocated addresses.
1636 * - Allocate memory for IP allocation state, including per node
1638 * - Populate IP allocation algorithm in IP allocation state
1639 * - Populate local value of tunable NoIPFailback in IP allocation
1640 state - this is really a cluster-wide configuration variable and
1641 only the value form the master node is used
1642 * - Retrieve tunables NoIPTakeover and NoIPHostOnAllDisabled from all
1643 * connected nodes - this is done separately so tunable values can
1644 * be faked in unit testing
1645 * - Populate NoIPTakover tunable in IP allocation state
1646 * - Populate NoIPHost in IP allocation state, derived from node flags
1647 * and NoIPHostOnAllDisabled tunable
1648 * - Retrieve and populate known and available IP lists in IP
1650 * - If no available IP addresses then early exit
1651 * - Build list of (known IPs, currently assigned node)
1652 * - Populate list of nodes to force rebalance - internal structure,
1653 * currently no way to fetch, only used by LCP2 for nodes that have
1654 * had new IP addresses added
1655 * - Run IP allocation algorithm
1656 * - Send RELEASE_IP to all nodes for IPs they should not host
1657 * - Send TAKE_IP to all nodes for IPs they should host
1658 * - Send IPREALLOCATED to all nodes (with backward compatibility hack)
1660 int ctdb_takeover_run(struct ctdb_context *ctdb, struct ctdb_node_map_old *nodemap,
1661 uint32_t *force_rebalance_nodes)
1664 struct ctdb_public_ip ip;
1666 struct public_ip_list *all_ips, *tmp_ip;
1668 struct timeval timeout;
1669 struct client_async_data *async_data;
1670 struct ctdb_client_control_state *state;
1671 TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
1672 struct ipalloc_state *ipalloc_state;
1673 struct takeover_callback_data *takeover_data;
1676 /* Initialise fail callback data to be used with
1677 * takeover_run_fail_callback(). A failure in any of the
1678 * following steps will cause an early return, so this can be
1679 * reused for each of those steps without re-initialising. */
1680 takeover_data = takeover_callback_data_init(tmp_ctx,
1682 if (takeover_data == NULL) {
1683 talloc_free(tmp_ctx);
1688 * ip failover is completely disabled, just send out the
1689 * ipreallocated event.
1691 if (ctdb->tunable.disable_ip_failover != 0) {
1695 ipalloc_state = ipalloc_state_init(ctdb, tmp_ctx);
1696 if (ipalloc_state == NULL) {
1697 talloc_free(tmp_ctx);
1701 if (!set_ipflags(ctdb, ipalloc_state, nodemap)) {
1702 DEBUG(DEBUG_ERR,("Failed to set IP flags - aborting takeover run\n"));
1703 talloc_free(tmp_ctx);
1707 /* Fetch known/available public IPs from each active node */
1708 ret = ctdb_reload_remote_public_ips(ctdb, ipalloc_state, nodemap);
1710 talloc_free(tmp_ctx);
1714 /* Short-circuit IP allocation if no node has available IPs */
1715 can_host_ips = false;
1716 for (i=0; i < ipalloc_state->num; i++) {
1717 if (ipalloc_state->available_public_ips[i].num != 0) {
1718 can_host_ips = true;
1721 if (!can_host_ips) {
1722 DEBUG(DEBUG_WARNING,("No nodes available to host public IPs yet\n"));
1726 /* since nodes only know about those public addresses that
1727 can be served by that particular node, no single node has
1728 a full list of all public addresses that exist in the cluster.
1729 Walk over all node structures and create a merged list of
1730 all public addresses that exist in the cluster.
1732 all_ips = create_merged_ip_list(ipalloc_state);
1733 if (all_ips == NULL) {
1734 talloc_free(tmp_ctx);
1737 ipalloc_state->all_ips = all_ips;
1739 ipalloc_state->force_rebalance_nodes = force_rebalance_nodes;
1741 /* Do the IP reassignment calculations */
1742 ipalloc(ipalloc_state);
1744 /* Now tell all nodes to release any public IPs should not
1745 * host. This will be a NOOP on nodes that don't currently
1746 * hold the given IP.
1748 async_data = talloc_zero(tmp_ctx, struct client_async_data);
1749 CTDB_NO_MEMORY_FATAL(ctdb, async_data);
1751 async_data->fail_callback = takeover_run_fail_callback;
1752 async_data->callback_data = takeover_data;
1754 ZERO_STRUCT(ip); /* Avoid valgrind warnings for union */
1756 /* Send a RELEASE_IP to all nodes that should not be hosting
1757 * each IP. For each IP, all but one of these will be
1758 * redundant. However, the redundant ones are used to tell
1759 * nodes which node should be hosting the IP so that commands
1760 * like "ctdb ip" can display a particular nodes idea of who
1761 * is hosting what. */
1762 for (i=0;i<nodemap->num;i++) {
1763 /* don't talk to unconnected nodes, but do talk to banned nodes */
1764 if (nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED) {
1768 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1769 if (tmp_ip->pnn == nodemap->nodes[i].pnn) {
1770 /* This node should be serving this
1771 vnn so don't tell it to release the ip
1775 ip.pnn = tmp_ip->pnn;
1776 ip.addr = tmp_ip->addr;
1778 timeout = TAKEOVER_TIMEOUT();
1779 data.dsize = sizeof(ip);
1780 data.dptr = (uint8_t *)&ip;
1781 state = ctdb_control_send(ctdb, nodemap->nodes[i].pnn,
1782 0, CTDB_CONTROL_RELEASE_IP, 0,
1785 if (state == NULL) {
1786 DEBUG(DEBUG_ERR,(__location__ " Failed to call async control CTDB_CONTROL_RELEASE_IP to node %u\n", nodemap->nodes[i].pnn));
1787 talloc_free(tmp_ctx);
1791 ctdb_client_async_add(async_data, state);
1794 if (ctdb_client_async_wait(ctdb, async_data) != 0) {
1796 ("Async control CTDB_CONTROL_RELEASE_IP failed\n"));
1799 talloc_free(async_data);
1802 /* For each IP, send a TAKOVER_IP to the node that should be
1803 * hosting it. Many of these will often be redundant (since
1804 * the allocation won't have changed) but they can be useful
1805 * to recover from inconsistencies. */
1806 async_data = talloc_zero(tmp_ctx, struct client_async_data);
1807 CTDB_NO_MEMORY_FATAL(ctdb, async_data);
1809 async_data->fail_callback = takeover_run_fail_callback;
1810 async_data->callback_data = takeover_data;
1812 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1813 if (tmp_ip->pnn == -1) {
1814 /* this IP won't be taken over */
1818 ip.pnn = tmp_ip->pnn;
1819 ip.addr = tmp_ip->addr;
1821 timeout = TAKEOVER_TIMEOUT();
1822 data.dsize = sizeof(ip);
1823 data.dptr = (uint8_t *)&ip;
1824 state = ctdb_control_send(ctdb, tmp_ip->pnn,
1825 0, CTDB_CONTROL_TAKEOVER_IP, 0,
1826 data, async_data, &timeout, NULL);
1827 if (state == NULL) {
1828 DEBUG(DEBUG_ERR,(__location__ " Failed to call async control CTDB_CONTROL_TAKEOVER_IP to node %u\n", tmp_ip->pnn));
1829 talloc_free(tmp_ctx);
1833 ctdb_client_async_add(async_data, state);
1835 if (ctdb_client_async_wait(ctdb, async_data) != 0) {
1837 ("Async control CTDB_CONTROL_TAKEOVER_IP failed\n"));
1843 * Tell all nodes to run eventscripts to process the
1844 * "ipreallocated" event. This can do a lot of things,
1845 * including restarting services to reconfigure them if public
1846 * IPs have moved. Once upon a time this event only used to
1849 nodes = list_of_connected_nodes(ctdb, nodemap, tmp_ctx, true);
1850 ret = ctdb_client_async_control(ctdb, CTDB_CONTROL_IPREALLOCATED,
1851 nodes, 0, TAKEOVER_TIMEOUT(),
1853 NULL, takeover_run_fail_callback,
1857 ("Async CTDB_CONTROL_IPREALLOCATED control failed\n"));
1861 talloc_free(tmp_ctx);
1865 takeover_run_process_failures(ctdb, takeover_data);
1866 talloc_free(tmp_ctx);
1872 destroy a ctdb_client_ip structure
1874 static int ctdb_client_ip_destructor(struct ctdb_client_ip *ip)
1876 DEBUG(DEBUG_DEBUG,("destroying client tcp for %s:%u (client_id %u)\n",
1877 ctdb_addr_to_str(&ip->addr),
1878 ntohs(ip->addr.ip.sin_port),
1881 DLIST_REMOVE(ip->ctdb->client_ip_list, ip);
1886 called by a client to inform us of a TCP connection that it is managing
1887 that should tickled with an ACK when IP takeover is done
1889 int32_t ctdb_control_tcp_client(struct ctdb_context *ctdb, uint32_t client_id,
1892 struct ctdb_client *client = reqid_find(ctdb->idr, client_id, struct ctdb_client);
1893 struct ctdb_connection *tcp_sock = NULL;
1894 struct ctdb_tcp_list *tcp;
1895 struct ctdb_connection t;
1898 struct ctdb_client_ip *ip;
1899 struct ctdb_vnn *vnn;
1900 ctdb_sock_addr addr;
1902 /* If we don't have public IPs, tickles are useless */
1903 if (ctdb->vnn == NULL) {
1907 tcp_sock = (struct ctdb_connection *)indata.dptr;
1909 addr = tcp_sock->src;
1910 ctdb_canonicalize_ip(&addr, &tcp_sock->src);
1911 addr = tcp_sock->dst;
1912 ctdb_canonicalize_ip(&addr, &tcp_sock->dst);
1915 memcpy(&addr, &tcp_sock->dst, sizeof(addr));
1916 vnn = find_public_ip_vnn(ctdb, &addr);
1918 switch (addr.sa.sa_family) {
1920 if (ntohl(addr.ip.sin_addr.s_addr) != INADDR_LOOPBACK) {
1921 DEBUG(DEBUG_ERR,("Could not add client IP %s. This is not a public address.\n",
1922 ctdb_addr_to_str(&addr)));
1926 DEBUG(DEBUG_ERR,("Could not add client IP %s. This is not a public ipv6 address.\n",
1927 ctdb_addr_to_str(&addr)));
1930 DEBUG(DEBUG_ERR,(__location__ " Unknown family type %d\n", addr.sa.sa_family));
1936 if (vnn->pnn != ctdb->pnn) {
1937 DEBUG(DEBUG_ERR,("Attempt to register tcp client for IP %s we don't hold - failing (client_id %u pid %u)\n",
1938 ctdb_addr_to_str(&addr),
1939 client_id, client->pid));
1940 /* failing this call will tell smbd to die */
1944 ip = talloc(client, struct ctdb_client_ip);
1945 CTDB_NO_MEMORY(ctdb, ip);
1949 ip->client_id = client_id;
1950 talloc_set_destructor(ip, ctdb_client_ip_destructor);
1951 DLIST_ADD(ctdb->client_ip_list, ip);
1953 tcp = talloc(client, struct ctdb_tcp_list);
1954 CTDB_NO_MEMORY(ctdb, tcp);
1956 tcp->connection.src = tcp_sock->src;
1957 tcp->connection.dst = tcp_sock->dst;
1959 DLIST_ADD(client->tcp_list, tcp);
1961 t.src = tcp_sock->src;
1962 t.dst = tcp_sock->dst;
1964 data.dptr = (uint8_t *)&t;
1965 data.dsize = sizeof(t);
1967 switch (addr.sa.sa_family) {
1969 DEBUG(DEBUG_INFO,("registered tcp client for %u->%s:%u (client_id %u pid %u)\n",
1970 (unsigned)ntohs(tcp_sock->dst.ip.sin_port),
1971 ctdb_addr_to_str(&tcp_sock->src),
1972 (unsigned)ntohs(tcp_sock->src.ip.sin_port), client_id, client->pid));
1975 DEBUG(DEBUG_INFO,("registered tcp client for %u->%s:%u (client_id %u pid %u)\n",
1976 (unsigned)ntohs(tcp_sock->dst.ip6.sin6_port),
1977 ctdb_addr_to_str(&tcp_sock->src),
1978 (unsigned)ntohs(tcp_sock->src.ip6.sin6_port), client_id, client->pid));
1981 DEBUG(DEBUG_ERR,(__location__ " Unknown family %d\n", addr.sa.sa_family));
1985 /* tell all nodes about this tcp connection */
1986 ret = ctdb_daemon_send_control(ctdb, CTDB_BROADCAST_CONNECTED, 0,
1987 CTDB_CONTROL_TCP_ADD,
1988 0, CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL);
1990 DEBUG(DEBUG_ERR,(__location__ " Failed to send CTDB_CONTROL_TCP_ADD\n"));
1998 find a tcp address on a list
2000 static struct ctdb_connection *ctdb_tcp_find(struct ctdb_tcp_array *array,
2001 struct ctdb_connection *tcp)
2005 if (array == NULL) {
2009 for (i=0;i<array->num;i++) {
2010 if (ctdb_same_sockaddr(&array->connections[i].src, &tcp->src) &&
2011 ctdb_same_sockaddr(&array->connections[i].dst, &tcp->dst)) {
2012 return &array->connections[i];
2021 called by a daemon to inform us of a TCP connection that one of its
2022 clients managing that should tickled with an ACK when IP takeover is
2025 int32_t ctdb_control_tcp_add(struct ctdb_context *ctdb, TDB_DATA indata, bool tcp_update_needed)
2027 struct ctdb_connection *p = (struct ctdb_connection *)indata.dptr;
2028 struct ctdb_tcp_array *tcparray;
2029 struct ctdb_connection tcp;
2030 struct ctdb_vnn *vnn;
2032 /* If we don't have public IPs, tickles are useless */
2033 if (ctdb->vnn == NULL) {
2037 vnn = find_public_ip_vnn(ctdb, &p->dst);
2039 DEBUG(DEBUG_INFO,(__location__ " got TCP_ADD control for an address which is not a public address '%s'\n",
2040 ctdb_addr_to_str(&p->dst)));
2046 tcparray = vnn->tcp_array;
2048 /* If this is the first tickle */
2049 if (tcparray == NULL) {
2050 tcparray = talloc(vnn, struct ctdb_tcp_array);
2051 CTDB_NO_MEMORY(ctdb, tcparray);
2052 vnn->tcp_array = tcparray;
2055 tcparray->connections = talloc_size(tcparray, sizeof(struct ctdb_connection));
2056 CTDB_NO_MEMORY(ctdb, tcparray->connections);
2058 tcparray->connections[tcparray->num].src = p->src;
2059 tcparray->connections[tcparray->num].dst = p->dst;
2062 if (tcp_update_needed) {
2063 vnn->tcp_update_needed = true;
2069 /* Do we already have this tickle ?*/
2072 if (ctdb_tcp_find(tcparray, &tcp) != NULL) {
2073 DEBUG(DEBUG_DEBUG,("Already had tickle info for %s:%u for vnn:%u\n",
2074 ctdb_addr_to_str(&tcp.dst),
2075 ntohs(tcp.dst.ip.sin_port),
2080 /* A new tickle, we must add it to the array */
2081 tcparray->connections = talloc_realloc(tcparray, tcparray->connections,
2082 struct ctdb_connection,
2084 CTDB_NO_MEMORY(ctdb, tcparray->connections);
2086 tcparray->connections[tcparray->num].src = p->src;
2087 tcparray->connections[tcparray->num].dst = p->dst;
2090 DEBUG(DEBUG_INFO,("Added tickle info for %s:%u from vnn %u\n",
2091 ctdb_addr_to_str(&tcp.dst),
2092 ntohs(tcp.dst.ip.sin_port),
2095 if (tcp_update_needed) {
2096 vnn->tcp_update_needed = true;
2103 static void ctdb_remove_connection(struct ctdb_vnn *vnn, struct ctdb_connection *conn)
2105 struct ctdb_connection *tcpp;
2111 /* if the array is empty we cant remove it
2112 and we don't need to do anything
2114 if (vnn->tcp_array == NULL) {
2115 DEBUG(DEBUG_INFO,("Trying to remove tickle that doesnt exist (array is empty) %s:%u\n",
2116 ctdb_addr_to_str(&conn->dst),
2117 ntohs(conn->dst.ip.sin_port)));
2122 /* See if we know this connection
2123 if we don't know this connection then we dont need to do anything
2125 tcpp = ctdb_tcp_find(vnn->tcp_array, conn);
2127 DEBUG(DEBUG_INFO,("Trying to remove tickle that doesnt exist %s:%u\n",
2128 ctdb_addr_to_str(&conn->dst),
2129 ntohs(conn->dst.ip.sin_port)));
2134 /* We need to remove this entry from the array.
2135 Instead of allocating a new array and copying data to it
2136 we cheat and just copy the last entry in the existing array
2137 to the entry that is to be removed and just shring the
2140 *tcpp = vnn->tcp_array->connections[vnn->tcp_array->num - 1];
2141 vnn->tcp_array->num--;
2143 /* If we deleted the last entry we also need to remove the entire array
2145 if (vnn->tcp_array->num == 0) {
2146 talloc_free(vnn->tcp_array);
2147 vnn->tcp_array = NULL;
2150 vnn->tcp_update_needed = true;
2152 DEBUG(DEBUG_INFO,("Removed tickle info for %s:%u\n",
2153 ctdb_addr_to_str(&conn->src),
2154 ntohs(conn->src.ip.sin_port)));
2159 called by a daemon to inform us of a TCP connection that one of its
2160 clients used are no longer needed in the tickle database
2162 int32_t ctdb_control_tcp_remove(struct ctdb_context *ctdb, TDB_DATA indata)
2164 struct ctdb_vnn *vnn;
2165 struct ctdb_connection *conn = (struct ctdb_connection *)indata.dptr;
2167 /* If we don't have public IPs, tickles are useless */
2168 if (ctdb->vnn == NULL) {
2172 vnn = find_public_ip_vnn(ctdb, &conn->dst);
2175 (__location__ " unable to find public address %s\n",
2176 ctdb_addr_to_str(&conn->dst)));
2180 ctdb_remove_connection(vnn, conn);
2187 Called when another daemon starts - causes all tickles for all
2188 public addresses we are serving to be sent to the new node on the
2189 next check. This actually causes the next scheduled call to
2190 tdb_update_tcp_tickles() to update all nodes. This is simple and
2191 doesn't require careful error handling.
2193 int32_t ctdb_control_startup(struct ctdb_context *ctdb, uint32_t pnn)
2195 struct ctdb_vnn *vnn;
2197 DEBUG(DEBUG_INFO, ("Received startup control from node %lu\n",
2198 (unsigned long) pnn));
2200 for (vnn = ctdb->vnn; vnn != NULL; vnn = vnn->next) {
2201 vnn->tcp_update_needed = true;
2209 called when a client structure goes away - hook to remove
2210 elements from the tcp_list in all daemons
2212 void ctdb_takeover_client_destructor_hook(struct ctdb_client *client)
2214 while (client->tcp_list) {
2215 struct ctdb_vnn *vnn;
2216 struct ctdb_tcp_list *tcp = client->tcp_list;
2217 struct ctdb_connection *conn = &tcp->connection;
2219 DLIST_REMOVE(client->tcp_list, tcp);
2221 vnn = find_public_ip_vnn(client->ctdb,
2225 (__location__ " unable to find public address %s\n",
2226 ctdb_addr_to_str(&conn->dst)));
2230 /* If the IP address is hosted on this node then
2231 * remove the connection. */
2232 if (vnn->pnn == client->ctdb->pnn) {
2233 ctdb_remove_connection(vnn, conn);
2236 /* Otherwise this function has been called because the
2237 * server IP address has been released to another node
2238 * and the client has exited. This means that we
2239 * should not delete the connection information. The
2240 * takeover node processes connections too. */
2245 void ctdb_release_all_ips(struct ctdb_context *ctdb)
2247 struct ctdb_vnn *vnn;
2251 if (ctdb->tunable.disable_ip_failover == 1) {
2255 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
2256 if (!ctdb_sys_have_ip(&vnn->public_address)) {
2257 ctdb_vnn_unassign_iface(ctdb, vnn);
2264 /* Don't allow multiple releases at once. Some code,
2265 * particularly ctdb_tickle_sentenced_connections() is
2267 if (vnn->update_in_flight) {
2268 DEBUG(DEBUG_WARNING,
2270 " Not releasing IP %s/%u on interface %s, an update is already in progess\n",
2271 ctdb_addr_to_str(&vnn->public_address),
2272 vnn->public_netmask_bits,
2273 ctdb_vnn_iface_string(vnn)));
2276 vnn->update_in_flight = true;
2278 DEBUG(DEBUG_INFO,("Release of IP %s/%u on interface %s node:-1\n",
2279 ctdb_addr_to_str(&vnn->public_address),
2280 vnn->public_netmask_bits,
2281 ctdb_vnn_iface_string(vnn)));
2283 ctdb_event_script_args(ctdb, CTDB_EVENT_RELEASE_IP, "%s %s %u",
2284 ctdb_vnn_iface_string(vnn),
2285 ctdb_addr_to_str(&vnn->public_address),
2286 vnn->public_netmask_bits);
2288 data.dptr = (uint8_t *)talloc_strdup(
2289 vnn, ctdb_addr_to_str(&vnn->public_address));
2290 if (data.dptr != NULL) {
2291 data.dsize = strlen((char *)data.dptr) + 1;
2292 ctdb_daemon_send_message(ctdb, ctdb->pnn,
2293 CTDB_SRVID_RELEASE_IP, data);
2294 talloc_free(data.dptr);
2297 ctdb_vnn_unassign_iface(ctdb, vnn);
2298 vnn->update_in_flight = false;
2302 DEBUG(DEBUG_NOTICE,(__location__ " Released %d public IPs\n", count));
2307 get list of public IPs
2309 int32_t ctdb_control_get_public_ips(struct ctdb_context *ctdb,
2310 struct ctdb_req_control_old *c, TDB_DATA *outdata)
2313 struct ctdb_public_ip_list_old *ips;
2314 struct ctdb_vnn *vnn;
2315 bool only_available = false;
2317 if (c->flags & CTDB_PUBLIC_IP_FLAGS_ONLY_AVAILABLE) {
2318 only_available = true;
2321 /* count how many public ip structures we have */
2323 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
2327 len = offsetof(struct ctdb_public_ip_list_old, ips) +
2328 num*sizeof(struct ctdb_public_ip);
2329 ips = talloc_zero_size(outdata, len);
2330 CTDB_NO_MEMORY(ctdb, ips);
2333 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
2334 if (only_available && !ctdb_vnn_available(ctdb, vnn)) {
2337 ips->ips[i].pnn = vnn->pnn;
2338 ips->ips[i].addr = vnn->public_address;
2342 len = offsetof(struct ctdb_public_ip_list_old, ips) +
2343 i*sizeof(struct ctdb_public_ip);
2345 outdata->dsize = len;
2346 outdata->dptr = (uint8_t *)ips;
2352 int32_t ctdb_control_get_public_ip_info(struct ctdb_context *ctdb,
2353 struct ctdb_req_control_old *c,
2358 ctdb_sock_addr *addr;
2359 struct ctdb_public_ip_info_old *info;
2360 struct ctdb_vnn *vnn;
2362 addr = (ctdb_sock_addr *)indata.dptr;
2364 vnn = find_public_ip_vnn(ctdb, addr);
2366 DEBUG(DEBUG_ERR,(__location__ " Could not get public ip info, "
2367 "'%s'not a public address\n",
2368 ctdb_addr_to_str(addr)));
2372 /* count how many public ip structures we have */
2374 for (;vnn->ifaces[num];) {
2378 len = offsetof(struct ctdb_public_ip_info_old, ifaces) +
2379 num*sizeof(struct ctdb_iface);
2380 info = talloc_zero_size(outdata, len);
2381 CTDB_NO_MEMORY(ctdb, info);
2383 info->ip.addr = vnn->public_address;
2384 info->ip.pnn = vnn->pnn;
2385 info->active_idx = 0xFFFFFFFF;
2387 for (i=0; vnn->ifaces[i]; i++) {
2388 struct ctdb_interface *cur;
2390 cur = ctdb_find_iface(ctdb, vnn->ifaces[i]);
2392 DEBUG(DEBUG_CRIT, (__location__ " internal error iface[%s] unknown\n",
2396 if (vnn->iface == cur) {
2397 info->active_idx = i;
2399 strncpy(info->ifaces[i].name, cur->name,
2400 sizeof(info->ifaces[i].name));
2401 info->ifaces[i].name[sizeof(info->ifaces[i].name)-1] = '\0';
2402 info->ifaces[i].link_state = cur->link_up;
2403 info->ifaces[i].references = cur->references;
2406 len = offsetof(struct ctdb_public_ip_info_old, ifaces) +
2407 i*sizeof(struct ctdb_iface);
2409 outdata->dsize = len;
2410 outdata->dptr = (uint8_t *)info;
2415 int32_t ctdb_control_get_ifaces(struct ctdb_context *ctdb,
2416 struct ctdb_req_control_old *c,
2420 struct ctdb_iface_list_old *ifaces;
2421 struct ctdb_interface *cur;
2423 /* count how many public ip structures we have */
2425 for (cur=ctdb->ifaces;cur;cur=cur->next) {
2429 len = offsetof(struct ctdb_iface_list_old, ifaces) +
2430 num*sizeof(struct ctdb_iface);
2431 ifaces = talloc_zero_size(outdata, len);
2432 CTDB_NO_MEMORY(ctdb, ifaces);
2435 for (cur=ctdb->ifaces;cur;cur=cur->next) {
2436 strncpy(ifaces->ifaces[i].name, cur->name,
2437 sizeof(ifaces->ifaces[i].name));
2438 ifaces->ifaces[i].name[sizeof(ifaces->ifaces[i].name)-1] = '\0';
2439 ifaces->ifaces[i].link_state = cur->link_up;
2440 ifaces->ifaces[i].references = cur->references;
2444 len = offsetof(struct ctdb_iface_list_old, ifaces) +
2445 i*sizeof(struct ctdb_iface);
2447 outdata->dsize = len;
2448 outdata->dptr = (uint8_t *)ifaces;
2453 int32_t ctdb_control_set_iface_link(struct ctdb_context *ctdb,
2454 struct ctdb_req_control_old *c,
2457 struct ctdb_iface *info;
2458 struct ctdb_interface *iface;
2459 bool link_up = false;
2461 info = (struct ctdb_iface *)indata.dptr;
2463 if (info->name[CTDB_IFACE_SIZE] != '\0') {
2464 int len = strnlen(info->name, CTDB_IFACE_SIZE);
2465 DEBUG(DEBUG_ERR, (__location__ " name[%*.*s] not terminated\n",
2466 len, len, info->name));
2470 switch (info->link_state) {
2478 DEBUG(DEBUG_ERR, (__location__ " link_state[%u] invalid\n",
2479 (unsigned int)info->link_state));
2483 if (info->references != 0) {
2484 DEBUG(DEBUG_ERR, (__location__ " references[%u] should be 0\n",
2485 (unsigned int)info->references));
2489 iface = ctdb_find_iface(ctdb, info->name);
2490 if (iface == NULL) {
2494 if (link_up == iface->link_up) {
2498 DEBUG(iface->link_up?DEBUG_ERR:DEBUG_NOTICE,
2499 ("iface[%s] has changed it's link status %s => %s\n",
2501 iface->link_up?"up":"down",
2502 link_up?"up":"down"));
2504 iface->link_up = link_up;
2510 called by a daemon to inform us of the entire list of TCP tickles for
2511 a particular public address.
2512 this control should only be sent by the node that is currently serving
2513 that public address.
2515 int32_t ctdb_control_set_tcp_tickle_list(struct ctdb_context *ctdb, TDB_DATA indata)
2517 struct ctdb_tickle_list_old *list = (struct ctdb_tickle_list_old *)indata.dptr;
2518 struct ctdb_tcp_array *tcparray;
2519 struct ctdb_vnn *vnn;
2521 /* We must at least have tickles.num or else we cant verify the size
2522 of the received data blob
2524 if (indata.dsize < offsetof(struct ctdb_tickle_list_old, connections)) {
2525 DEBUG(DEBUG_ERR,("Bad indata in ctdb_tickle_list. Not enough data for the tickle.num field\n"));
2529 /* verify that the size of data matches what we expect */
2530 if (indata.dsize < offsetof(struct ctdb_tickle_list_old, connections)
2531 + sizeof(struct ctdb_connection) * list->num) {
2532 DEBUG(DEBUG_ERR,("Bad indata in ctdb_tickle_list\n"));
2536 DEBUG(DEBUG_INFO, ("Received tickle update for public address %s\n",
2537 ctdb_addr_to_str(&list->addr)));
2539 vnn = find_public_ip_vnn(ctdb, &list->addr);
2541 DEBUG(DEBUG_INFO,(__location__ " Could not set tcp tickle list, '%s' is not a public address\n",
2542 ctdb_addr_to_str(&list->addr)));
2547 if (vnn->pnn == ctdb->pnn) {
2549 ("Ignoring redundant set tcp tickle list, this node hosts '%s'\n",
2550 ctdb_addr_to_str(&list->addr)));
2554 /* remove any old ticklelist we might have */
2555 talloc_free(vnn->tcp_array);
2556 vnn->tcp_array = NULL;
2558 tcparray = talloc(vnn, struct ctdb_tcp_array);
2559 CTDB_NO_MEMORY(ctdb, tcparray);
2561 tcparray->num = list->num;
2563 tcparray->connections = talloc_array(tcparray, struct ctdb_connection, tcparray->num);
2564 CTDB_NO_MEMORY(ctdb, tcparray->connections);
2566 memcpy(tcparray->connections, &list->connections[0],
2567 sizeof(struct ctdb_connection)*tcparray->num);
2569 /* We now have a new fresh tickle list array for this vnn */
2570 vnn->tcp_array = tcparray;
2576 called to return the full list of tickles for the puclic address associated
2577 with the provided vnn
2579 int32_t ctdb_control_get_tcp_tickle_list(struct ctdb_context *ctdb, TDB_DATA indata, TDB_DATA *outdata)
2581 ctdb_sock_addr *addr = (ctdb_sock_addr *)indata.dptr;
2582 struct ctdb_tickle_list_old *list;
2583 struct ctdb_tcp_array *tcparray;
2585 struct ctdb_vnn *vnn;
2588 vnn = find_public_ip_vnn(ctdb, addr);
2590 DEBUG(DEBUG_ERR,(__location__ " Could not get tcp tickle list, '%s' is not a public address\n",
2591 ctdb_addr_to_str(addr)));
2596 port = ctdb_addr_to_port(addr);
2598 tcparray = vnn->tcp_array;
2600 if (tcparray != NULL) {
2602 /* All connections */
2603 num = tcparray->num;
2605 /* Count connections for port */
2606 for (i = 0; i < tcparray->num; i++) {
2607 if (port == ctdb_addr_to_port(&tcparray->connections[i].dst)) {
2614 outdata->dsize = offsetof(struct ctdb_tickle_list_old, connections)
2615 + sizeof(struct ctdb_connection) * num;
2617 outdata->dptr = talloc_size(outdata, outdata->dsize);
2618 CTDB_NO_MEMORY(ctdb, outdata->dptr);
2619 list = (struct ctdb_tickle_list_old *)outdata->dptr;
2629 for (i = 0; i < tcparray->num; i++) {
2631 port == ctdb_addr_to_port(&tcparray->connections[i].dst)) {
2632 list->connections[num] = tcparray->connections[i];
2642 set the list of all tcp tickles for a public address
2644 static int ctdb_send_set_tcp_tickles_for_ip(struct ctdb_context *ctdb,
2645 ctdb_sock_addr *addr,
2646 struct ctdb_tcp_array *tcparray)
2650 struct ctdb_tickle_list_old *list;
2653 num = tcparray->num;
2658 data.dsize = offsetof(struct ctdb_tickle_list_old, connections) +
2659 sizeof(struct ctdb_connection) * num;
2660 data.dptr = talloc_size(ctdb, data.dsize);
2661 CTDB_NO_MEMORY(ctdb, data.dptr);
2663 list = (struct ctdb_tickle_list_old *)data.dptr;
2667 memcpy(&list->connections[0], tcparray->connections, sizeof(struct ctdb_connection) * num);
2670 ret = ctdb_daemon_send_control(ctdb, CTDB_BROADCAST_ALL, 0,
2671 CTDB_CONTROL_SET_TCP_TICKLE_LIST,
2672 0, CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL);
2674 DEBUG(DEBUG_ERR,(__location__ " ctdb_control for set tcp tickles failed\n"));
2678 talloc_free(data.dptr);
2685 perform tickle updates if required
2687 static void ctdb_update_tcp_tickles(struct tevent_context *ev,
2688 struct tevent_timer *te,
2689 struct timeval t, void *private_data)
2691 struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
2693 struct ctdb_vnn *vnn;
2695 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
2696 /* we only send out updates for public addresses that
2699 if (ctdb->pnn != vnn->pnn) {
2702 /* We only send out the updates if we need to */
2703 if (!vnn->tcp_update_needed) {
2706 ret = ctdb_send_set_tcp_tickles_for_ip(ctdb,
2707 &vnn->public_address,
2710 DEBUG(DEBUG_ERR,("Failed to send the tickle update for public address %s\n",
2711 ctdb_addr_to_str(&vnn->public_address)));
2714 ("Sent tickle update for public address %s\n",
2715 ctdb_addr_to_str(&vnn->public_address)));
2716 vnn->tcp_update_needed = false;
2720 tevent_add_timer(ctdb->ev, ctdb->tickle_update_context,
2721 timeval_current_ofs(ctdb->tunable.tickle_update_interval, 0),
2722 ctdb_update_tcp_tickles, ctdb);
2726 start periodic update of tcp tickles
2728 void ctdb_start_tcp_tickle_update(struct ctdb_context *ctdb)
2730 ctdb->tickle_update_context = talloc_new(ctdb);
2732 tevent_add_timer(ctdb->ev, ctdb->tickle_update_context,
2733 timeval_current_ofs(ctdb->tunable.tickle_update_interval, 0),
2734 ctdb_update_tcp_tickles, ctdb);
2740 struct control_gratious_arp {
2741 struct ctdb_context *ctdb;
2742 ctdb_sock_addr addr;
2748 send a control_gratuitous arp
2750 static void send_gratious_arp(struct tevent_context *ev,
2751 struct tevent_timer *te,
2752 struct timeval t, void *private_data)
2755 struct control_gratious_arp *arp = talloc_get_type(private_data,
2756 struct control_gratious_arp);
2758 ret = ctdb_sys_send_arp(&arp->addr, arp->iface);
2760 DEBUG(DEBUG_ERR,(__location__ " sending of gratious arp on iface '%s' failed (%s)\n",
2761 arp->iface, strerror(errno)));
2766 if (arp->count == CTDB_ARP_REPEAT) {
2771 tevent_add_timer(arp->ctdb->ev, arp,
2772 timeval_current_ofs(CTDB_ARP_INTERVAL, 0),
2773 send_gratious_arp, arp);
2780 int32_t ctdb_control_send_gratious_arp(struct ctdb_context *ctdb, TDB_DATA indata)
2782 struct ctdb_addr_info_old *gratious_arp = (struct ctdb_addr_info_old *)indata.dptr;
2783 struct control_gratious_arp *arp;
2785 /* verify the size of indata */
2786 if (indata.dsize < offsetof(struct ctdb_addr_info_old, iface)) {
2787 DEBUG(DEBUG_ERR,(__location__ " Too small indata to hold a ctdb_control_gratious_arp structure. Got %u require %u bytes\n",
2788 (unsigned)indata.dsize,
2789 (unsigned)offsetof(struct ctdb_addr_info_old, iface)));
2793 ( offsetof(struct ctdb_addr_info_old, iface)
2794 + gratious_arp->len ) ){
2796 DEBUG(DEBUG_ERR,(__location__ " Wrong size of indata. Was %u bytes "
2797 "but should be %u bytes\n",
2798 (unsigned)indata.dsize,
2799 (unsigned)(offsetof(struct ctdb_addr_info_old, iface)+gratious_arp->len)));
2804 arp = talloc(ctdb, struct control_gratious_arp);
2805 CTDB_NO_MEMORY(ctdb, arp);
2808 arp->addr = gratious_arp->addr;
2809 arp->iface = talloc_strdup(arp, gratious_arp->iface);
2810 CTDB_NO_MEMORY(ctdb, arp->iface);
2813 tevent_add_timer(arp->ctdb->ev, arp,
2814 timeval_zero(), send_gratious_arp, arp);
2819 int32_t ctdb_control_add_public_address(struct ctdb_context *ctdb, TDB_DATA indata)
2821 struct ctdb_addr_info_old *pub = (struct ctdb_addr_info_old *)indata.dptr;
2824 /* verify the size of indata */
2825 if (indata.dsize < offsetof(struct ctdb_addr_info_old, iface)) {
2826 DEBUG(DEBUG_ERR,(__location__ " Too small indata to hold a ctdb_addr_info structure\n"));
2830 ( offsetof(struct ctdb_addr_info_old, iface)
2833 DEBUG(DEBUG_ERR,(__location__ " Wrong size of indata. Was %u bytes "
2834 "but should be %u bytes\n",
2835 (unsigned)indata.dsize,
2836 (unsigned)(offsetof(struct ctdb_addr_info_old, iface)+pub->len)));
2840 DEBUG(DEBUG_NOTICE,("Add IP %s\n", ctdb_addr_to_str(&pub->addr)));
2842 ret = ctdb_add_public_address(ctdb, &pub->addr, pub->mask, &pub->iface[0], true);
2845 DEBUG(DEBUG_ERR,(__location__ " Failed to add public address\n"));
2852 struct delete_ip_callback_state {
2853 struct ctdb_req_control_old *c;
2857 called when releaseip event finishes for del_public_address
2859 static void delete_ip_callback(struct ctdb_context *ctdb,
2860 int32_t status, TDB_DATA data,
2861 const char *errormsg,
2864 struct delete_ip_callback_state *state =
2865 talloc_get_type(private_data, struct delete_ip_callback_state);
2867 /* If release failed then fail. */
2868 ctdb_request_control_reply(ctdb, state->c, NULL, status, errormsg);
2869 talloc_free(private_data);
2872 int32_t ctdb_control_del_public_address(struct ctdb_context *ctdb,
2873 struct ctdb_req_control_old *c,
2874 TDB_DATA indata, bool *async_reply)
2876 struct ctdb_addr_info_old *pub = (struct ctdb_addr_info_old *)indata.dptr;
2877 struct ctdb_vnn *vnn;
2879 /* verify the size of indata */
2880 if (indata.dsize < offsetof(struct ctdb_addr_info_old, iface)) {
2881 DEBUG(DEBUG_ERR,(__location__ " Too small indata to hold a ctdb_addr_info structure\n"));
2885 ( offsetof(struct ctdb_addr_info_old, iface)
2888 DEBUG(DEBUG_ERR,(__location__ " Wrong size of indata. Was %u bytes "
2889 "but should be %u bytes\n",
2890 (unsigned)indata.dsize,
2891 (unsigned)(offsetof(struct ctdb_addr_info_old, iface)+pub->len)));
2895 DEBUG(DEBUG_NOTICE,("Delete IP %s\n", ctdb_addr_to_str(&pub->addr)));
2897 /* walk over all public addresses until we find a match */
2898 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
2899 if (ctdb_same_ip(&vnn->public_address, &pub->addr)) {
2900 if (vnn->pnn == ctdb->pnn) {
2901 struct delete_ip_callback_state *state;
2902 struct ctdb_public_ip *ip;
2906 vnn->delete_pending = true;
2908 state = talloc(ctdb,
2909 struct delete_ip_callback_state);
2910 CTDB_NO_MEMORY(ctdb, state);
2913 ip = talloc(state, struct ctdb_public_ip);
2916 (__location__ " Out of memory\n"));
2921 ip->addr = pub->addr;
2923 data.dsize = sizeof(struct ctdb_public_ip);
2924 data.dptr = (unsigned char *)ip;
2926 ret = ctdb_daemon_send_control(ctdb,
2929 CTDB_CONTROL_RELEASE_IP,
2936 (__location__ "Unable to send "
2937 "CTDB_CONTROL_RELEASE_IP\n"));
2942 state->c = talloc_steal(state, c);
2943 *async_reply = true;
2945 /* This IP is not hosted on the
2946 * current node so just delete it
2948 do_delete_ip(ctdb, vnn);
2955 DEBUG(DEBUG_ERR,("Delete IP of unknown public IP address %s\n",
2956 ctdb_addr_to_str(&pub->addr)));
2961 struct ipreallocated_callback_state {
2962 struct ctdb_req_control_old *c;
2965 static void ctdb_ipreallocated_callback(struct ctdb_context *ctdb,
2966 int status, void *p)
2968 struct ipreallocated_callback_state *state =
2969 talloc_get_type(p, struct ipreallocated_callback_state);
2973 (" \"ipreallocated\" event script failed (status %d)\n",
2975 if (status == -ETIME) {
2976 ctdb_ban_self(ctdb);
2980 ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
2984 /* A control to run the ipreallocated event */
2985 int32_t ctdb_control_ipreallocated(struct ctdb_context *ctdb,
2986 struct ctdb_req_control_old *c,
2990 struct ipreallocated_callback_state *state;
2992 state = talloc(ctdb, struct ipreallocated_callback_state);
2993 CTDB_NO_MEMORY(ctdb, state);
2995 DEBUG(DEBUG_INFO,(__location__ " Running \"ipreallocated\" event\n"));
2997 ret = ctdb_event_script_callback(ctdb, state,
2998 ctdb_ipreallocated_callback, state,
2999 CTDB_EVENT_IPREALLOCATED,
3003 DEBUG(DEBUG_ERR,("Failed to run \"ipreallocated\" event \n"));
3008 /* tell the control that we will be reply asynchronously */
3009 state->c = talloc_steal(state, c);
3010 *async_reply = true;
3016 struct ctdb_reloadips_handle {
3017 struct ctdb_context *ctdb;
3018 struct ctdb_req_control_old *c;
3022 struct tevent_fd *fde;
3025 static int ctdb_reloadips_destructor(struct ctdb_reloadips_handle *h)
3027 if (h == h->ctdb->reload_ips) {
3028 h->ctdb->reload_ips = NULL;
3031 ctdb_request_control_reply(h->ctdb, h->c, NULL, h->status, NULL);
3034 ctdb_kill(h->ctdb, h->child, SIGKILL);
3038 static void ctdb_reloadips_timeout_event(struct tevent_context *ev,
3039 struct tevent_timer *te,
3040 struct timeval t, void *private_data)
3042 struct ctdb_reloadips_handle *h = talloc_get_type(private_data, struct ctdb_reloadips_handle);
3047 static void ctdb_reloadips_child_handler(struct tevent_context *ev,
3048 struct tevent_fd *fde,
3049 uint16_t flags, void *private_data)
3051 struct ctdb_reloadips_handle *h = talloc_get_type(private_data, struct ctdb_reloadips_handle);
3056 ret = sys_read(h->fd[0], &res, 1);
3057 if (ret < 1 || res != 0) {
3058 DEBUG(DEBUG_ERR, (__location__ " Reloadips child process returned error\n"));
3066 static int ctdb_reloadips_child(struct ctdb_context *ctdb)
3068 TALLOC_CTX *mem_ctx = talloc_new(NULL);
3069 struct ctdb_public_ip_list_old *ips;
3070 struct ctdb_vnn *vnn;
3071 struct client_async_data *async_data;
3072 struct timeval timeout;
3074 struct ctdb_client_control_state *state;
3078 CTDB_NO_MEMORY(ctdb, mem_ctx);
3080 /* Read IPs from local node */
3081 ret = ctdb_ctrl_get_public_ips(ctdb, TAKEOVER_TIMEOUT(),
3082 CTDB_CURRENT_NODE, mem_ctx, &ips);
3085 ("Unable to fetch public IPs from local node\n"));
3086 talloc_free(mem_ctx);
3090 /* Read IPs file - this is safe since this is a child process */
3092 if (ctdb_set_public_addresses(ctdb, false) != 0) {
3093 DEBUG(DEBUG_ERR,("Failed to re-read public addresses file\n"));
3094 talloc_free(mem_ctx);
3098 async_data = talloc_zero(mem_ctx, struct client_async_data);
3099 CTDB_NO_MEMORY(ctdb, async_data);
3101 /* Compare IPs between node and file for IPs to be deleted */
3102 for (i = 0; i < ips->num; i++) {
3104 for (vnn = ctdb->vnn; vnn; vnn = vnn->next) {
3105 if (ctdb_same_ip(&vnn->public_address,
3106 &ips->ips[i].addr)) {
3107 /* IP is still in file */
3113 /* Delete IP ips->ips[i] */
3114 struct ctdb_addr_info_old *pub;
3117 ("IP %s no longer configured, deleting it\n",
3118 ctdb_addr_to_str(&ips->ips[i].addr)));
3120 pub = talloc_zero(mem_ctx, struct ctdb_addr_info_old);
3121 CTDB_NO_MEMORY(ctdb, pub);
3123 pub->addr = ips->ips[i].addr;
3127 timeout = TAKEOVER_TIMEOUT();
3129 data.dsize = offsetof(struct ctdb_addr_info_old,
3131 data.dptr = (uint8_t *)pub;
3133 state = ctdb_control_send(ctdb, CTDB_CURRENT_NODE, 0,
3134 CTDB_CONTROL_DEL_PUBLIC_IP,
3135 0, data, async_data,
3137 if (state == NULL) {
3140 " failed sending CTDB_CONTROL_DEL_PUBLIC_IP\n"));
3144 ctdb_client_async_add(async_data, state);
3148 /* Compare IPs between node and file for IPs to be added */
3150 for (vnn = ctdb->vnn; vnn; vnn = vnn->next) {
3151 for (i = 0; i < ips->num; i++) {
3152 if (ctdb_same_ip(&vnn->public_address,
3153 &ips->ips[i].addr)) {
3154 /* IP already on node */
3158 if (i == ips->num) {
3159 /* Add IP ips->ips[i] */
3160 struct ctdb_addr_info_old *pub;
3161 const char *ifaces = NULL;
3166 ("New IP %s configured, adding it\n",
3167 ctdb_addr_to_str(&vnn->public_address)));
3169 uint32_t pnn = ctdb_get_pnn(ctdb);
3171 data.dsize = sizeof(pnn);
3172 data.dptr = (uint8_t *)&pnn;
3174 ret = ctdb_client_send_message(
3176 CTDB_BROADCAST_CONNECTED,
3177 CTDB_SRVID_REBALANCE_NODE,
3180 DEBUG(DEBUG_WARNING,
3181 ("Failed to send message to force node reallocation - IPs may be unbalanced\n"));
3187 ifaces = vnn->ifaces[0];
3189 while (vnn->ifaces[iface] != NULL) {
3190 ifaces = talloc_asprintf(vnn, "%s,%s", ifaces,
3191 vnn->ifaces[iface]);
3195 len = strlen(ifaces) + 1;
3196 pub = talloc_zero_size(mem_ctx,
3197 offsetof(struct ctdb_addr_info_old, iface) + len);
3198 CTDB_NO_MEMORY(ctdb, pub);
3200 pub->addr = vnn->public_address;
3201 pub->mask = vnn->public_netmask_bits;
3203 memcpy(&pub->iface[0], ifaces, pub->len);
3205 timeout = TAKEOVER_TIMEOUT();
3207 data.dsize = offsetof(struct ctdb_addr_info_old,
3209 data.dptr = (uint8_t *)pub;
3211 state = ctdb_control_send(ctdb, CTDB_CURRENT_NODE, 0,
3212 CTDB_CONTROL_ADD_PUBLIC_IP,
3213 0, data, async_data,
3215 if (state == NULL) {
3218 " failed sending CTDB_CONTROL_ADD_PUBLIC_IP\n"));
3222 ctdb_client_async_add(async_data, state);
3226 if (ctdb_client_async_wait(ctdb, async_data) != 0) {
3227 DEBUG(DEBUG_ERR,(__location__ " Add/delete IPs failed\n"));
3231 talloc_free(mem_ctx);
3235 talloc_free(mem_ctx);
3239 /* This control is sent to force the node to re-read the public addresses file
3240 and drop any addresses we should nnot longer host, and add new addresses
3241 that we are now able to host
3243 int32_t ctdb_control_reload_public_ips(struct ctdb_context *ctdb, struct ctdb_req_control_old *c, bool *async_reply)
3245 struct ctdb_reloadips_handle *h;
3246 pid_t parent = getpid();
3248 if (ctdb->reload_ips != NULL) {
3249 talloc_free(ctdb->reload_ips);
3250 ctdb->reload_ips = NULL;
3253 h = talloc(ctdb, struct ctdb_reloadips_handle);
3254 CTDB_NO_MEMORY(ctdb, h);
3259 if (pipe(h->fd) == -1) {
3260 DEBUG(DEBUG_ERR,("Failed to create pipe for ctdb_freeze_lock\n"));
3265 h->child = ctdb_fork(ctdb);
3266 if (h->child == (pid_t)-1) {
3267 DEBUG(DEBUG_ERR, ("Failed to fork a child for reloadips\n"));
3275 if (h->child == 0) {
3276 signed char res = 0;
3279 debug_extra = talloc_asprintf(NULL, "reloadips:");
3281 prctl_set_comment("ctdb_reloadips");
3282 if (switch_from_server_to_client(ctdb, "reloadips-child") != 0) {
3283 DEBUG(DEBUG_CRIT,("ERROR: Failed to switch reloadips child into client mode\n"));
3286 res = ctdb_reloadips_child(ctdb);
3288 DEBUG(DEBUG_ERR,("Failed to reload ips on local node\n"));
3292 sys_write(h->fd[1], &res, 1);
3293 ctdb_wait_for_process_to_exit(parent);
3297 h->c = talloc_steal(h, c);
3300 set_close_on_exec(h->fd[0]);
3302 talloc_set_destructor(h, ctdb_reloadips_destructor);
3305 h->fde = tevent_add_fd(ctdb->ev, h, h->fd[0], TEVENT_FD_READ,
3306 ctdb_reloadips_child_handler, (void *)h);
3307 tevent_fd_set_auto_close(h->fde);
3309 tevent_add_timer(ctdb->ev, h, timeval_current_ofs(120, 0),
3310 ctdb_reloadips_timeout_event, h);
3312 /* we reply later */
3313 *async_reply = true;