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 "lib/tdb/include/tdb.h"
23 #include "lib/util/dlinklist.h"
24 #include "system/network.h"
25 #include "system/filesys.h"
26 #include "system/wait.h"
27 #include "../include/ctdb_private.h"
28 #include "../common/rb_tree.h"
31 #define TAKEOVER_TIMEOUT() timeval_current_ofs(ctdb->tunable.takeover_timeout,0)
33 #define CTDB_ARP_INTERVAL 1
34 #define CTDB_ARP_REPEAT 3
37 struct ctdb_iface *prev, *next;
43 static const char *ctdb_vnn_iface_string(const struct ctdb_vnn *vnn)
46 return vnn->iface->name;
52 static int ctdb_add_local_iface(struct ctdb_context *ctdb, const char *iface)
56 /* Verify that we dont have an entry for this ip yet */
57 for (i=ctdb->ifaces;i;i=i->next) {
58 if (strcmp(i->name, iface) == 0) {
63 /* create a new structure for this interface */
64 i = talloc_zero(ctdb, struct ctdb_iface);
65 CTDB_NO_MEMORY_FATAL(ctdb, i);
66 i->name = talloc_strdup(i, iface);
67 CTDB_NO_MEMORY(ctdb, i->name);
69 * If link_up defaults to true then IPs can be allocated to a
70 * node during the first recovery. However, then an interface
71 * could have its link marked down during the startup event,
72 * causing the IP to move almost immediately. If link_up
73 * defaults to false then, during normal operation, IPs added
74 * to a new interface can't be assigned until a monitor cycle
75 * has occurred and marked the new interfaces up. This makes
76 * IP allocation unpredictable. The following is a neat
77 * compromise: early in startup link_up defaults to false, so
78 * IPs can't be assigned, and after startup IPs can be
79 * assigned immediately.
81 i->link_up = ctdb->done_startup;
83 DLIST_ADD(ctdb->ifaces, i);
88 static bool vnn_has_interface_with_name(struct ctdb_vnn *vnn,
93 for (n = 0; vnn->ifaces[n] != NULL; n++) {
94 if (strcmp(name, vnn->ifaces[n]) == 0) {
102 /* If any interfaces now have no possible IPs then delete them. This
103 * implementation is naive (i.e. simple) rather than clever
104 * (i.e. complex). Given that this is run on delip and that operation
105 * is rare, this doesn't need to be efficient - it needs to be
106 * foolproof. One alternative is reference counting, where the logic
107 * is distributed and can, therefore, be broken in multiple places.
108 * Another alternative is to build a red-black tree of interfaces that
109 * can have addresses (by walking ctdb->vnn and ctdb->single_ip_vnn
110 * once) and then walking ctdb->ifaces once and deleting those not in
111 * the tree. Let's go to one of those if the naive implementation
112 * causes problems... :-)
114 static void ctdb_remove_orphaned_ifaces(struct ctdb_context *ctdb,
115 struct ctdb_vnn *vnn,
118 struct ctdb_iface *i;
120 /* For each interface, check if there's an IP using it. */
121 for(i=ctdb->ifaces; i; i=i->next) {
125 /* Only consider interfaces named in the given VNN. */
126 if (!vnn_has_interface_with_name(vnn, i->name)) {
130 /* Is the "single IP" on this interface? */
131 if ((ctdb->single_ip_vnn != NULL) &&
132 (ctdb->single_ip_vnn->ifaces[0] != NULL) &&
133 (strcmp(i->name, ctdb->single_ip_vnn->ifaces[0]) == 0)) {
134 /* Found, next interface please... */
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);
149 /* Caller will free mem_ctx when convenient. */
150 talloc_steal(mem_ctx, i);
156 static struct ctdb_iface *ctdb_find_iface(struct ctdb_context *ctdb,
159 struct ctdb_iface *i;
161 /* Verify that we dont have an entry for this ip yet */
162 for (i=ctdb->ifaces;i;i=i->next) {
163 if (strcmp(i->name, iface) == 0) {
171 static struct ctdb_iface *ctdb_vnn_best_iface(struct ctdb_context *ctdb,
172 struct ctdb_vnn *vnn)
175 struct ctdb_iface *cur = NULL;
176 struct ctdb_iface *best = NULL;
178 for (i=0; vnn->ifaces[i]; i++) {
180 cur = ctdb_find_iface(ctdb, vnn->ifaces[i]);
194 if (cur->references < best->references) {
203 static int32_t ctdb_vnn_assign_iface(struct ctdb_context *ctdb,
204 struct ctdb_vnn *vnn)
206 struct ctdb_iface *best = NULL;
209 DEBUG(DEBUG_INFO, (__location__ " public address '%s' "
210 "still assigned to iface '%s'\n",
211 ctdb_addr_to_str(&vnn->public_address),
212 ctdb_vnn_iface_string(vnn)));
216 best = ctdb_vnn_best_iface(ctdb, vnn);
218 DEBUG(DEBUG_ERR, (__location__ " public address '%s' "
219 "cannot assign to iface any iface\n",
220 ctdb_addr_to_str(&vnn->public_address)));
226 vnn->pnn = ctdb->pnn;
228 DEBUG(DEBUG_INFO, (__location__ " public address '%s' "
229 "now assigned to iface '%s' refs[%d]\n",
230 ctdb_addr_to_str(&vnn->public_address),
231 ctdb_vnn_iface_string(vnn),
236 static void ctdb_vnn_unassign_iface(struct ctdb_context *ctdb,
237 struct ctdb_vnn *vnn)
239 DEBUG(DEBUG_INFO, (__location__ " public address '%s' "
240 "now unassigned (old iface '%s' refs[%d])\n",
241 ctdb_addr_to_str(&vnn->public_address),
242 ctdb_vnn_iface_string(vnn),
243 vnn->iface?vnn->iface->references:0));
245 vnn->iface->references--;
248 if (vnn->pnn == ctdb->pnn) {
253 static bool ctdb_vnn_available(struct ctdb_context *ctdb,
254 struct ctdb_vnn *vnn)
258 if (vnn->iface && vnn->iface->link_up) {
262 for (i=0; vnn->ifaces[i]; i++) {
263 struct ctdb_iface *cur;
265 cur = ctdb_find_iface(ctdb, vnn->ifaces[i]);
278 struct ctdb_takeover_arp {
279 struct ctdb_context *ctdb;
282 struct ctdb_tcp_array *tcparray;
283 struct ctdb_vnn *vnn;
288 lists of tcp endpoints
290 struct ctdb_tcp_list {
291 struct ctdb_tcp_list *prev, *next;
292 struct ctdb_tcp_connection connection;
296 list of clients to kill on IP release
298 struct ctdb_client_ip {
299 struct ctdb_client_ip *prev, *next;
300 struct ctdb_context *ctdb;
307 send a gratuitous arp
309 static void ctdb_control_send_arp(struct event_context *ev, struct timed_event *te,
310 struct timeval t, void *private_data)
312 struct ctdb_takeover_arp *arp = talloc_get_type(private_data,
313 struct ctdb_takeover_arp);
315 struct ctdb_tcp_array *tcparray;
316 const char *iface = ctdb_vnn_iface_string(arp->vnn);
318 ret = ctdb_sys_send_arp(&arp->addr, iface);
320 DEBUG(DEBUG_CRIT,(__location__ " sending of arp failed on iface '%s' (%s)\n",
321 iface, strerror(errno)));
324 tcparray = arp->tcparray;
326 for (i=0;i<tcparray->num;i++) {
327 struct ctdb_tcp_connection *tcon;
329 tcon = &tcparray->connections[i];
330 DEBUG(DEBUG_INFO,("sending tcp tickle ack for %u->%s:%u\n",
331 (unsigned)ntohs(tcon->dst_addr.ip.sin_port),
332 ctdb_addr_to_str(&tcon->src_addr),
333 (unsigned)ntohs(tcon->src_addr.ip.sin_port)));
334 ret = ctdb_sys_send_tcp(
339 DEBUG(DEBUG_CRIT,(__location__ " Failed to send tcp tickle ack for %s\n",
340 ctdb_addr_to_str(&tcon->src_addr)));
347 if (arp->count == CTDB_ARP_REPEAT) {
352 event_add_timed(arp->ctdb->ev, arp->vnn->takeover_ctx,
353 timeval_current_ofs(CTDB_ARP_INTERVAL, 100000),
354 ctdb_control_send_arp, arp);
357 static int32_t ctdb_announce_vnn_iface(struct ctdb_context *ctdb,
358 struct ctdb_vnn *vnn)
360 struct ctdb_takeover_arp *arp;
361 struct ctdb_tcp_array *tcparray;
363 if (!vnn->takeover_ctx) {
364 vnn->takeover_ctx = talloc_new(vnn);
365 if (!vnn->takeover_ctx) {
370 arp = talloc_zero(vnn->takeover_ctx, struct ctdb_takeover_arp);
376 arp->addr = vnn->public_address;
379 tcparray = vnn->tcp_array;
381 /* add all of the known tcp connections for this IP to the
382 list of tcp connections to send tickle acks for */
383 arp->tcparray = talloc_steal(arp, tcparray);
385 vnn->tcp_array = NULL;
386 vnn->tcp_update_needed = true;
389 event_add_timed(arp->ctdb->ev, vnn->takeover_ctx,
390 timeval_zero(), ctdb_control_send_arp, arp);
395 struct takeover_callback_state {
396 struct ctdb_req_control *c;
397 ctdb_sock_addr *addr;
398 struct ctdb_vnn *vnn;
401 struct ctdb_do_takeip_state {
402 struct ctdb_req_control *c;
403 struct ctdb_vnn *vnn;
407 called when takeip event finishes
409 static void ctdb_do_takeip_callback(struct ctdb_context *ctdb, int status,
412 struct ctdb_do_takeip_state *state =
413 talloc_get_type(private_data, struct ctdb_do_takeip_state);
418 struct ctdb_node *node = ctdb->nodes[ctdb->pnn];
420 if (status == -ETIME) {
423 DEBUG(DEBUG_ERR,(__location__ " Failed to takeover IP %s on interface %s\n",
424 ctdb_addr_to_str(&state->vnn->public_address),
425 ctdb_vnn_iface_string(state->vnn)));
426 ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
428 node->flags |= NODE_FLAGS_UNHEALTHY;
433 if (ctdb->do_checkpublicip) {
435 ret = ctdb_announce_vnn_iface(ctdb, state->vnn);
437 ctdb_request_control_reply(ctdb, state->c, NULL, -1, NULL);
444 data.dptr = (uint8_t *)ctdb_addr_to_str(&state->vnn->public_address);
445 data.dsize = strlen((char *)data.dptr) + 1;
446 DEBUG(DEBUG_INFO,(__location__ " sending TAKE_IP for '%s'\n", data.dptr));
448 ctdb_daemon_send_message(ctdb, ctdb->pnn, CTDB_SRVID_TAKE_IP, data);
451 /* the control succeeded */
452 ctdb_request_control_reply(ctdb, state->c, NULL, 0, NULL);
457 static int ctdb_takeip_destructor(struct ctdb_do_takeip_state *state)
459 state->vnn->update_in_flight = false;
464 take over an ip address
466 static int32_t ctdb_do_takeip(struct ctdb_context *ctdb,
467 struct ctdb_req_control *c,
468 struct ctdb_vnn *vnn)
471 struct ctdb_do_takeip_state *state;
473 if (vnn->update_in_flight) {
474 DEBUG(DEBUG_NOTICE,("Takeover of IP %s/%u rejected "
475 "update for this IP already in flight\n",
476 ctdb_addr_to_str(&vnn->public_address),
477 vnn->public_netmask_bits));
481 ret = ctdb_vnn_assign_iface(ctdb, vnn);
483 DEBUG(DEBUG_ERR,("Takeover of IP %s/%u failed to "
484 "assign a usable interface\n",
485 ctdb_addr_to_str(&vnn->public_address),
486 vnn->public_netmask_bits));
490 state = talloc(vnn, struct ctdb_do_takeip_state);
491 CTDB_NO_MEMORY(ctdb, state);
493 state->c = talloc_steal(ctdb, c);
496 vnn->update_in_flight = true;
497 talloc_set_destructor(state, ctdb_takeip_destructor);
499 DEBUG(DEBUG_NOTICE,("Takeover of IP %s/%u on interface %s\n",
500 ctdb_addr_to_str(&vnn->public_address),
501 vnn->public_netmask_bits,
502 ctdb_vnn_iface_string(vnn)));
504 ret = ctdb_event_script_callback(ctdb,
506 ctdb_do_takeip_callback,
511 ctdb_vnn_iface_string(vnn),
512 ctdb_addr_to_str(&vnn->public_address),
513 vnn->public_netmask_bits);
516 DEBUG(DEBUG_ERR,(__location__ " Failed to takeover IP %s on interface %s\n",
517 ctdb_addr_to_str(&vnn->public_address),
518 ctdb_vnn_iface_string(vnn)));
526 struct ctdb_do_updateip_state {
527 struct ctdb_req_control *c;
528 struct ctdb_iface *old;
529 struct ctdb_vnn *vnn;
533 called when updateip event finishes
535 static void ctdb_do_updateip_callback(struct ctdb_context *ctdb, int status,
538 struct ctdb_do_updateip_state *state =
539 talloc_get_type(private_data, struct ctdb_do_updateip_state);
543 if (status == -ETIME) {
546 DEBUG(DEBUG_ERR,(__location__ " Failed to move IP %s from interface %s to %s\n",
547 ctdb_addr_to_str(&state->vnn->public_address),
549 ctdb_vnn_iface_string(state->vnn)));
552 * All we can do is reset the old interface
553 * and let the next run fix it
555 ctdb_vnn_unassign_iface(ctdb, state->vnn);
556 state->vnn->iface = state->old;
557 state->vnn->iface->references++;
559 ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
564 if (ctdb->do_checkpublicip) {
566 ret = ctdb_announce_vnn_iface(ctdb, state->vnn);
568 ctdb_request_control_reply(ctdb, state->c, NULL, -1, NULL);
575 /* the control succeeded */
576 ctdb_request_control_reply(ctdb, state->c, NULL, 0, NULL);
581 static int ctdb_updateip_destructor(struct ctdb_do_updateip_state *state)
583 state->vnn->update_in_flight = false;
588 update (move) an ip address
590 static int32_t ctdb_do_updateip(struct ctdb_context *ctdb,
591 struct ctdb_req_control *c,
592 struct ctdb_vnn *vnn)
595 struct ctdb_do_updateip_state *state;
596 struct ctdb_iface *old = vnn->iface;
597 const char *new_name;
599 if (vnn->update_in_flight) {
600 DEBUG(DEBUG_NOTICE,("Update of IP %s/%u rejected "
601 "update for this IP already in flight\n",
602 ctdb_addr_to_str(&vnn->public_address),
603 vnn->public_netmask_bits));
607 ctdb_vnn_unassign_iface(ctdb, vnn);
608 ret = ctdb_vnn_assign_iface(ctdb, vnn);
610 DEBUG(DEBUG_ERR,("update of IP %s/%u failed to "
611 "assin a usable interface (old iface '%s')\n",
612 ctdb_addr_to_str(&vnn->public_address),
613 vnn->public_netmask_bits,
618 new_name = ctdb_vnn_iface_string(vnn);
619 if (old->name != NULL && new_name != NULL && !strcmp(old->name, new_name)) {
620 /* A benign update from one interface onto itself.
621 * no need to run the eventscripts in this case, just return
624 ctdb_request_control_reply(ctdb, c, NULL, 0, NULL);
628 state = talloc(vnn, struct ctdb_do_updateip_state);
629 CTDB_NO_MEMORY(ctdb, state);
631 state->c = talloc_steal(ctdb, c);
635 vnn->update_in_flight = true;
636 talloc_set_destructor(state, ctdb_updateip_destructor);
638 DEBUG(DEBUG_NOTICE,("Update of IP %s/%u from "
639 "interface %s to %s\n",
640 ctdb_addr_to_str(&vnn->public_address),
641 vnn->public_netmask_bits,
645 ret = ctdb_event_script_callback(ctdb,
647 ctdb_do_updateip_callback,
650 CTDB_EVENT_UPDATE_IP,
654 ctdb_addr_to_str(&vnn->public_address),
655 vnn->public_netmask_bits);
657 DEBUG(DEBUG_ERR,(__location__ " Failed update IP %s from interface %s to %s\n",
658 ctdb_addr_to_str(&vnn->public_address),
659 old->name, new_name));
668 Find the vnn of the node that has a public ip address
669 returns -1 if the address is not known as a public address
671 static struct ctdb_vnn *find_public_ip_vnn(struct ctdb_context *ctdb, ctdb_sock_addr *addr)
673 struct ctdb_vnn *vnn;
675 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
676 if (ctdb_same_ip(&vnn->public_address, addr)) {
685 take over an ip address
687 int32_t ctdb_control_takeover_ip(struct ctdb_context *ctdb,
688 struct ctdb_req_control *c,
693 struct ctdb_public_ip *pip = (struct ctdb_public_ip *)indata.dptr;
694 struct ctdb_vnn *vnn;
695 bool have_ip = false;
696 bool do_updateip = false;
697 bool do_takeip = false;
698 struct ctdb_iface *best_iface = NULL;
700 if (pip->pnn != ctdb->pnn) {
701 DEBUG(DEBUG_ERR,(__location__" takeoverip called for an ip '%s' "
702 "with pnn %d, but we're node %d\n",
703 ctdb_addr_to_str(&pip->addr),
704 pip->pnn, ctdb->pnn));
708 /* update out vnn list */
709 vnn = find_public_ip_vnn(ctdb, &pip->addr);
711 DEBUG(DEBUG_INFO,("takeoverip called for an ip '%s' that is not a public address\n",
712 ctdb_addr_to_str(&pip->addr)));
716 if (ctdb->do_checkpublicip) {
717 have_ip = ctdb_sys_have_ip(&pip->addr);
719 best_iface = ctdb_vnn_best_iface(ctdb, vnn);
720 if (best_iface == NULL) {
721 DEBUG(DEBUG_ERR,("takeoverip of IP %s/%u failed to find"
722 "a usable interface (old %s, have_ip %d)\n",
723 ctdb_addr_to_str(&vnn->public_address),
724 vnn->public_netmask_bits,
725 ctdb_vnn_iface_string(vnn),
730 if (vnn->iface == NULL && vnn->pnn == -1 && have_ip && best_iface != NULL) {
731 DEBUG(DEBUG_ERR,("Taking over newly created ip\n"));
736 if (vnn->iface == NULL && have_ip) {
737 DEBUG(DEBUG_CRIT,(__location__ " takeoverip of IP %s is known to the kernel, "
738 "but we have no interface assigned, has someone manually configured it? Ignore for now.\n",
739 ctdb_addr_to_str(&vnn->public_address)));
743 if (vnn->pnn != ctdb->pnn && have_ip && vnn->pnn != -1) {
744 DEBUG(DEBUG_CRIT,(__location__ " takeoverip of IP %s is known to the kernel, "
745 "and we have it on iface[%s], but it was assigned to node %d"
746 "and we are node %d, banning ourself\n",
747 ctdb_addr_to_str(&vnn->public_address),
748 ctdb_vnn_iface_string(vnn), vnn->pnn, ctdb->pnn));
753 if (vnn->pnn == -1 && have_ip) {
754 vnn->pnn = ctdb->pnn;
755 DEBUG(DEBUG_CRIT,(__location__ " takeoverip of IP %s is known to the kernel, "
756 "and we already have it on iface[%s], update local daemon\n",
757 ctdb_addr_to_str(&vnn->public_address),
758 ctdb_vnn_iface_string(vnn)));
763 if (vnn->iface != best_iface) {
764 if (!vnn->iface->link_up) {
766 } else if (vnn->iface->references > (best_iface->references + 1)) {
767 /* only move when the rebalance gains something */
775 ctdb_vnn_unassign_iface(ctdb, vnn);
782 ret = ctdb_do_takeip(ctdb, c, vnn);
786 } else if (do_updateip) {
787 ret = ctdb_do_updateip(ctdb, c, vnn);
793 * The interface is up and the kernel known the ip
796 DEBUG(DEBUG_INFO,("Redundant takeover of IP %s/%u on interface %s (ip already held)\n",
797 ctdb_addr_to_str(&pip->addr),
798 vnn->public_netmask_bits,
799 ctdb_vnn_iface_string(vnn)));
803 /* tell ctdb_control.c that we will be replying asynchronously */
810 takeover an ip address old v4 style
812 int32_t ctdb_control_takeover_ipv4(struct ctdb_context *ctdb,
813 struct ctdb_req_control *c,
819 data.dsize = sizeof(struct ctdb_public_ip);
820 data.dptr = (uint8_t *)talloc_zero(c, struct ctdb_public_ip);
821 CTDB_NO_MEMORY(ctdb, data.dptr);
823 memcpy(data.dptr, indata.dptr, indata.dsize);
824 return ctdb_control_takeover_ip(ctdb, c, data, async_reply);
828 kill any clients that are registered with a IP that is being released
830 static void release_kill_clients(struct ctdb_context *ctdb, ctdb_sock_addr *addr)
832 struct ctdb_client_ip *ip;
834 DEBUG(DEBUG_INFO,("release_kill_clients for ip %s\n",
835 ctdb_addr_to_str(addr)));
837 for (ip=ctdb->client_ip_list; ip; ip=ip->next) {
838 ctdb_sock_addr tmp_addr;
841 DEBUG(DEBUG_INFO,("checking for client %u with IP %s\n",
843 ctdb_addr_to_str(&ip->addr)));
845 if (ctdb_same_ip(&tmp_addr, addr)) {
846 struct ctdb_client *client = ctdb_reqid_find(ctdb,
849 DEBUG(DEBUG_INFO,("matched client %u with IP %s and pid %u\n",
851 ctdb_addr_to_str(&ip->addr),
854 if (client->pid != 0) {
855 DEBUG(DEBUG_INFO,(__location__ " Killing client pid %u for IP %s on client_id %u\n",
856 (unsigned)client->pid,
857 ctdb_addr_to_str(addr),
859 ctdb_kill(ctdb, client->pid, SIGKILL);
866 called when releaseip event finishes
868 static void release_ip_callback(struct ctdb_context *ctdb, int status,
871 struct takeover_callback_state *state =
872 talloc_get_type(private_data, struct takeover_callback_state);
875 if (status == -ETIME) {
879 /* send a message to all clients of this node telling them
880 that the cluster has been reconfigured and they should
881 release any sockets on this IP */
882 data.dptr = (uint8_t *)talloc_strdup(state, ctdb_addr_to_str(state->addr));
883 CTDB_NO_MEMORY_VOID(ctdb, data.dptr);
884 data.dsize = strlen((char *)data.dptr)+1;
886 DEBUG(DEBUG_INFO,(__location__ " sending RELEASE_IP for '%s'\n", data.dptr));
888 ctdb_daemon_send_message(ctdb, ctdb->pnn, CTDB_SRVID_RELEASE_IP, data);
890 /* kill clients that have registered with this IP */
891 release_kill_clients(ctdb, state->addr);
893 ctdb_vnn_unassign_iface(ctdb, state->vnn);
895 /* the control succeeded */
896 ctdb_request_control_reply(ctdb, state->c, NULL, 0, NULL);
900 static int ctdb_releaseip_destructor(struct takeover_callback_state *state)
902 state->vnn->update_in_flight = false;
907 release an ip address
909 int32_t ctdb_control_release_ip(struct ctdb_context *ctdb,
910 struct ctdb_req_control *c,
915 struct takeover_callback_state *state;
916 struct ctdb_public_ip *pip = (struct ctdb_public_ip *)indata.dptr;
917 struct ctdb_vnn *vnn;
920 /* update our vnn list */
921 vnn = find_public_ip_vnn(ctdb, &pip->addr);
923 DEBUG(DEBUG_INFO,("releaseip called for an ip '%s' that is not a public address\n",
924 ctdb_addr_to_str(&pip->addr)));
929 /* stop any previous arps */
930 talloc_free(vnn->takeover_ctx);
931 vnn->takeover_ctx = NULL;
933 /* Some ctdb tool commands (e.g. moveip, rebalanceip) send
934 * lazy multicast to drop an IP from any node that isn't the
935 * intended new node. The following causes makes ctdbd ignore
936 * a release for any address it doesn't host.
938 if (ctdb->do_checkpublicip) {
939 if (!ctdb_sys_have_ip(&pip->addr)) {
940 DEBUG(DEBUG_DEBUG,("Redundant release of IP %s/%u on interface %s (ip not held)\n",
941 ctdb_addr_to_str(&pip->addr),
942 vnn->public_netmask_bits,
943 ctdb_vnn_iface_string(vnn)));
944 ctdb_vnn_unassign_iface(ctdb, vnn);
948 if (vnn->iface == NULL) {
949 DEBUG(DEBUG_DEBUG,("Redundant release of IP %s/%u (ip not held)\n",
950 ctdb_addr_to_str(&pip->addr),
951 vnn->public_netmask_bits));
956 /* There is a potential race between take_ip and us because we
957 * update the VNN via a callback that run when the
958 * eventscripts have been run. Avoid the race by allowing one
959 * update to be in flight at a time.
961 if (vnn->update_in_flight) {
962 DEBUG(DEBUG_NOTICE,("Release of IP %s/%u rejected "
963 "update for this IP already in flight\n",
964 ctdb_addr_to_str(&vnn->public_address),
965 vnn->public_netmask_bits));
969 if (ctdb->do_checkpublicip) {
970 iface = ctdb_sys_find_ifname(&pip->addr);
972 DEBUG(DEBUG_ERR, ("Could not find which interface the ip address is hosted on. can not release it\n"));
976 iface = strdup(ctdb_vnn_iface_string(vnn));
979 DEBUG(DEBUG_NOTICE,("Release of IP %s/%u on interface %s node:%d\n",
980 ctdb_addr_to_str(&pip->addr),
981 vnn->public_netmask_bits,
985 state = talloc(ctdb, struct takeover_callback_state);
986 CTDB_NO_MEMORY(ctdb, state);
988 state->c = talloc_steal(state, c);
989 state->addr = talloc(state, ctdb_sock_addr);
990 CTDB_NO_MEMORY(ctdb, state->addr);
991 *state->addr = pip->addr;
994 vnn->update_in_flight = true;
995 talloc_set_destructor(state, ctdb_releaseip_destructor);
997 ret = ctdb_event_script_callback(ctdb,
998 state, release_ip_callback, state,
1000 CTDB_EVENT_RELEASE_IP,
1003 ctdb_addr_to_str(&pip->addr),
1004 vnn->public_netmask_bits);
1007 DEBUG(DEBUG_ERR,(__location__ " Failed to release IP %s on interface %s\n",
1008 ctdb_addr_to_str(&pip->addr),
1009 ctdb_vnn_iface_string(vnn)));
1014 /* tell the control that we will be reply asynchronously */
1015 *async_reply = true;
1020 release an ip address old v4 style
1022 int32_t ctdb_control_release_ipv4(struct ctdb_context *ctdb,
1023 struct ctdb_req_control *c,
1029 data.dsize = sizeof(struct ctdb_public_ip);
1030 data.dptr = (uint8_t *)talloc_zero(c, struct ctdb_public_ip);
1031 CTDB_NO_MEMORY(ctdb, data.dptr);
1033 memcpy(data.dptr, indata.dptr, indata.dsize);
1034 return ctdb_control_release_ip(ctdb, c, data, async_reply);
1038 static int ctdb_add_public_address(struct ctdb_context *ctdb,
1039 ctdb_sock_addr *addr,
1040 unsigned mask, const char *ifaces,
1043 struct ctdb_vnn *vnn;
1050 tmp = strdup(ifaces);
1051 for (iface = strtok(tmp, ","); iface; iface = strtok(NULL, ",")) {
1052 if (!ctdb_sys_check_iface_exists(iface)) {
1053 DEBUG(DEBUG_CRIT,("Interface %s does not exist. Can not add public-address : %s\n", iface, ctdb_addr_to_str(addr)));
1060 /* Verify that we dont have an entry for this ip yet */
1061 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
1062 if (ctdb_same_sockaddr(addr, &vnn->public_address)) {
1063 DEBUG(DEBUG_CRIT,("Same ip '%s' specified multiple times in the public address list \n",
1064 ctdb_addr_to_str(addr)));
1069 /* create a new vnn structure for this ip address */
1070 vnn = talloc_zero(ctdb, struct ctdb_vnn);
1071 CTDB_NO_MEMORY_FATAL(ctdb, vnn);
1072 vnn->ifaces = talloc_array(vnn, const char *, num + 2);
1073 tmp = talloc_strdup(vnn, ifaces);
1074 CTDB_NO_MEMORY_FATAL(ctdb, tmp);
1075 for (iface = strtok(tmp, ","); iface; iface = strtok(NULL, ",")) {
1076 vnn->ifaces = talloc_realloc(vnn, vnn->ifaces, const char *, num + 2);
1077 CTDB_NO_MEMORY_FATAL(ctdb, vnn->ifaces);
1078 vnn->ifaces[num] = talloc_strdup(vnn, iface);
1079 CTDB_NO_MEMORY_FATAL(ctdb, vnn->ifaces[num]);
1083 vnn->ifaces[num] = NULL;
1084 vnn->public_address = *addr;
1085 vnn->public_netmask_bits = mask;
1087 if (check_address) {
1088 if (ctdb_sys_have_ip(addr)) {
1089 DEBUG(DEBUG_ERR,("We are already hosting public address '%s'. setting PNN to ourself:%d\n", ctdb_addr_to_str(addr), ctdb->pnn));
1090 vnn->pnn = ctdb->pnn;
1094 for (i=0; vnn->ifaces[i]; i++) {
1095 ret = ctdb_add_local_iface(ctdb, vnn->ifaces[i]);
1097 DEBUG(DEBUG_CRIT, (__location__ " failed to add iface[%s] "
1098 "for public_address[%s]\n",
1099 vnn->ifaces[i], ctdb_addr_to_str(addr)));
1105 DLIST_ADD(ctdb->vnn, vnn);
1111 setup the event script directory
1113 int ctdb_set_event_script_dir(struct ctdb_context *ctdb, const char *script_dir)
1115 ctdb->event_script_dir = talloc_strdup(ctdb, script_dir);
1116 CTDB_NO_MEMORY(ctdb, ctdb->event_script_dir);
1120 static void ctdb_check_interfaces_event(struct event_context *ev, struct timed_event *te,
1121 struct timeval t, void *private_data)
1123 struct ctdb_context *ctdb = talloc_get_type(private_data,
1124 struct ctdb_context);
1125 struct ctdb_vnn *vnn;
1127 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
1130 for (i=0; vnn->ifaces[i] != NULL; i++) {
1131 if (!ctdb_sys_check_iface_exists(vnn->ifaces[i])) {
1132 DEBUG(DEBUG_CRIT,("Interface %s does not exist but is used by public ip %s\n",
1134 ctdb_addr_to_str(&vnn->public_address)));
1139 event_add_timed(ctdb->ev, ctdb->check_public_ifaces_ctx,
1140 timeval_current_ofs(30, 0),
1141 ctdb_check_interfaces_event, ctdb);
1145 int ctdb_start_monitoring_interfaces(struct ctdb_context *ctdb)
1147 if (ctdb->check_public_ifaces_ctx != NULL) {
1148 talloc_free(ctdb->check_public_ifaces_ctx);
1149 ctdb->check_public_ifaces_ctx = NULL;
1152 ctdb->check_public_ifaces_ctx = talloc_new(ctdb);
1153 if (ctdb->check_public_ifaces_ctx == NULL) {
1154 ctdb_fatal(ctdb, "failed to allocate context for checking interfaces");
1157 event_add_timed(ctdb->ev, ctdb->check_public_ifaces_ctx,
1158 timeval_current_ofs(30, 0),
1159 ctdb_check_interfaces_event, ctdb);
1166 setup the public address lists from a file
1168 int ctdb_set_public_addresses(struct ctdb_context *ctdb, bool check_addresses)
1174 lines = file_lines_load(ctdb->public_addresses_file, &nlines, ctdb);
1175 if (lines == NULL) {
1176 ctdb_set_error(ctdb, "Failed to load public address list '%s'\n", ctdb->public_addresses_file);
1179 while (nlines > 0 && strcmp(lines[nlines-1], "") == 0) {
1183 for (i=0;i<nlines;i++) {
1185 ctdb_sock_addr addr;
1186 const char *addrstr;
1191 while ((*line == ' ') || (*line == '\t')) {
1197 if (strcmp(line, "") == 0) {
1200 tok = strtok(line, " \t");
1202 tok = strtok(NULL, " \t");
1204 if (NULL == ctdb->default_public_interface) {
1205 DEBUG(DEBUG_CRIT,("No default public interface and no interface specified at line %u of public address list\n",
1210 ifaces = ctdb->default_public_interface;
1215 if (!addrstr || !parse_ip_mask(addrstr, ifaces, &addr, &mask)) {
1216 DEBUG(DEBUG_CRIT,("Badly formed line %u in public address list\n", i+1));
1220 if (ctdb_add_public_address(ctdb, &addr, mask, ifaces, check_addresses)) {
1221 DEBUG(DEBUG_CRIT,("Failed to add line %u to the public address list\n", i+1));
1232 int ctdb_set_single_public_ip(struct ctdb_context *ctdb,
1236 struct ctdb_vnn *svnn;
1237 struct ctdb_iface *cur = NULL;
1241 svnn = talloc_zero(ctdb, struct ctdb_vnn);
1242 CTDB_NO_MEMORY(ctdb, svnn);
1244 svnn->ifaces = talloc_array(svnn, const char *, 2);
1245 CTDB_NO_MEMORY(ctdb, svnn->ifaces);
1246 svnn->ifaces[0] = talloc_strdup(svnn->ifaces, iface);
1247 CTDB_NO_MEMORY(ctdb, svnn->ifaces[0]);
1248 svnn->ifaces[1] = NULL;
1250 ok = parse_ip(ip, iface, 0, &svnn->public_address);
1256 ret = ctdb_add_local_iface(ctdb, svnn->ifaces[0]);
1258 DEBUG(DEBUG_CRIT, (__location__ " failed to add iface[%s] "
1259 "for single_ip[%s]\n",
1261 ctdb_addr_to_str(&svnn->public_address)));
1266 /* assume the single public ip interface is initially "good" */
1267 cur = ctdb_find_iface(ctdb, iface);
1269 DEBUG(DEBUG_CRIT,("Can not find public interface %s used by --single-public-ip", iface));
1272 cur->link_up = true;
1274 ret = ctdb_vnn_assign_iface(ctdb, svnn);
1280 ctdb->single_ip_vnn = svnn;
1284 /* Given a physical node, return the number of
1285 public addresses that is currently assigned to this node.
1287 static int node_ip_coverage(struct ctdb_context *ctdb,
1289 struct ctdb_public_ip_list *ips)
1293 for (;ips;ips=ips->next) {
1294 if (ips->pnn == pnn) {
1302 /* Can the given node host the given IP: is the public IP known to the
1303 * node and is NOIPHOST unset?
1305 static bool can_node_host_ip(struct ctdb_context *ctdb, int32_t pnn,
1306 struct ctdb_node_map *nodemap,
1307 struct ctdb_public_ip_list *ip)
1309 struct ctdb_all_public_ips *public_ips;
1312 if (nodemap->nodes[pnn].flags & NODE_FLAGS_NOIPHOST) {
1316 public_ips = ctdb->nodes[pnn]->available_public_ips;
1318 if (public_ips == NULL) {
1322 for (i=0;i<public_ips->num;i++) {
1323 if (ctdb_same_ip(&ip->addr, &public_ips->ips[i].addr)) {
1324 /* yes, this node can serve this public ip */
1332 static bool can_node_takeover_ip(struct ctdb_context *ctdb, int32_t pnn,
1333 struct ctdb_node_map *nodemap,
1334 struct ctdb_public_ip_list *ip)
1336 if (nodemap->nodes[pnn].flags & NODE_FLAGS_NOIPTAKEOVER) {
1340 return can_node_host_ip(ctdb, pnn, nodemap, ip);
1343 /* search the node lists list for a node to takeover this ip.
1344 pick the node that currently are serving the least number of ips
1345 so that the ips get spread out evenly.
1347 static int find_takeover_node(struct ctdb_context *ctdb,
1348 struct ctdb_node_map *nodemap,
1349 struct ctdb_public_ip_list *ip,
1350 struct ctdb_public_ip_list *all_ips)
1352 int pnn, min=0, num;
1356 for (i=0;i<nodemap->num;i++) {
1357 /* verify that this node can serve this ip */
1358 if (!can_node_takeover_ip(ctdb, i, nodemap, ip)) {
1359 /* no it couldnt so skip to the next node */
1363 num = node_ip_coverage(ctdb, i, all_ips);
1364 /* was this the first node we checked ? */
1376 DEBUG(DEBUG_WARNING,(__location__ " Could not find node to take over public address '%s'\n",
1377 ctdb_addr_to_str(&ip->addr)));
1387 static uint32_t *ip_key(ctdb_sock_addr *ip)
1389 static uint32_t key[IP_KEYLEN];
1391 bzero(key, sizeof(key));
1393 switch (ip->sa.sa_family) {
1395 key[3] = htonl(ip->ip.sin_addr.s_addr);
1398 uint32_t *s6_a32 = (uint32_t *)&(ip->ip6.sin6_addr.s6_addr);
1399 key[0] = htonl(s6_a32[0]);
1400 key[1] = htonl(s6_a32[1]);
1401 key[2] = htonl(s6_a32[2]);
1402 key[3] = htonl(s6_a32[3]);
1406 DEBUG(DEBUG_ERR, (__location__ " ERROR, unknown family passed :%u\n", ip->sa.sa_family));
1413 static void *add_ip_callback(void *parm, void *data)
1415 struct ctdb_public_ip_list *this_ip = parm;
1416 struct ctdb_public_ip_list *prev_ip = data;
1418 if (prev_ip == NULL) {
1421 if (this_ip->pnn == -1) {
1422 this_ip->pnn = prev_ip->pnn;
1428 static int getips_count_callback(void *param, void *data)
1430 struct ctdb_public_ip_list **ip_list = (struct ctdb_public_ip_list **)param;
1431 struct ctdb_public_ip_list *new_ip = (struct ctdb_public_ip_list *)data;
1433 new_ip->next = *ip_list;
1438 static struct ctdb_public_ip_list *
1439 create_merged_ip_list(struct ctdb_context *ctdb)
1442 struct ctdb_public_ip_list *ip_list;
1443 struct ctdb_all_public_ips *public_ips;
1445 if (ctdb->ip_tree != NULL) {
1446 talloc_free(ctdb->ip_tree);
1447 ctdb->ip_tree = NULL;
1449 ctdb->ip_tree = trbt_create(ctdb, 0);
1451 for (i=0;i<ctdb->num_nodes;i++) {
1452 public_ips = ctdb->nodes[i]->known_public_ips;
1454 if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
1458 /* there were no public ips for this node */
1459 if (public_ips == NULL) {
1463 for (j=0;j<public_ips->num;j++) {
1464 struct ctdb_public_ip_list *tmp_ip;
1466 tmp_ip = talloc_zero(ctdb->ip_tree, struct ctdb_public_ip_list);
1467 CTDB_NO_MEMORY_NULL(ctdb, tmp_ip);
1468 /* Do not use information about IP addresses hosted
1469 * on other nodes, it may not be accurate */
1470 if (public_ips->ips[j].pnn == ctdb->nodes[i]->pnn) {
1471 tmp_ip->pnn = public_ips->ips[j].pnn;
1475 tmp_ip->addr = public_ips->ips[j].addr;
1476 tmp_ip->next = NULL;
1478 trbt_insertarray32_callback(ctdb->ip_tree,
1479 IP_KEYLEN, ip_key(&public_ips->ips[j].addr),
1486 trbt_traversearray32(ctdb->ip_tree, IP_KEYLEN, getips_count_callback, &ip_list);
1492 * This is the length of the longtest common prefix between the IPs.
1493 * It is calculated by XOR-ing the 2 IPs together and counting the
1494 * number of leading zeroes. The implementation means that all
1495 * addresses end up being 128 bits long.
1497 * FIXME? Should we consider IPv4 and IPv6 separately given that the
1498 * 12 bytes of 0 prefix padding will hurt the algorithm if there are
1499 * lots of nodes and IP addresses?
1501 static uint32_t ip_distance(ctdb_sock_addr *ip1, ctdb_sock_addr *ip2)
1503 uint32_t ip1_k[IP_KEYLEN];
1508 uint32_t distance = 0;
1510 memcpy(ip1_k, ip_key(ip1), sizeof(ip1_k));
1512 for (i=0; i<IP_KEYLEN; i++) {
1513 x = ip1_k[i] ^ t[i];
1517 /* Count number of leading zeroes.
1518 * FIXME? This could be optimised...
1520 while ((x & (1 << 31)) == 0) {
1530 /* Calculate the IP distance for the given IP relative to IPs on the
1531 given node. The ips argument is generally the all_ips variable
1532 used in the main part of the algorithm.
1534 static uint32_t ip_distance_2_sum(ctdb_sock_addr *ip,
1535 struct ctdb_public_ip_list *ips,
1538 struct ctdb_public_ip_list *t;
1543 for (t=ips; t != NULL; t=t->next) {
1544 if (t->pnn != pnn) {
1548 /* Optimisation: We never calculate the distance
1549 * between an address and itself. This allows us to
1550 * calculate the effect of removing an address from a
1551 * node by simply calculating the distance between
1552 * that address and all of the exitsing addresses.
1553 * Moreover, we assume that we're only ever dealing
1554 * with addresses from all_ips so we can identify an
1555 * address via a pointer rather than doing a more
1556 * expensive address comparison. */
1557 if (&(t->addr) == ip) {
1561 d = ip_distance(ip, &(t->addr));
1562 sum += d * d; /* Cheaper than pulling in math.h :-) */
1568 /* Return the LCP2 imbalance metric for addresses currently assigned
1571 static uint32_t lcp2_imbalance(struct ctdb_public_ip_list * all_ips, int pnn)
1573 struct ctdb_public_ip_list *t;
1575 uint32_t imbalance = 0;
1577 for (t=all_ips; t!=NULL; t=t->next) {
1578 if (t->pnn != pnn) {
1581 /* Pass the rest of the IPs rather than the whole
1584 imbalance += ip_distance_2_sum(&(t->addr), t->next, pnn);
1590 /* Allocate any unassigned IPs just by looping through the IPs and
1591 * finding the best node for each.
1593 static void basic_allocate_unassigned(struct ctdb_context *ctdb,
1594 struct ctdb_node_map *nodemap,
1595 struct ctdb_public_ip_list *all_ips)
1597 struct ctdb_public_ip_list *tmp_ip;
1599 /* loop over all ip's and find a physical node to cover for
1602 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1603 if (tmp_ip->pnn == -1) {
1604 if (find_takeover_node(ctdb, nodemap, tmp_ip, all_ips)) {
1605 DEBUG(DEBUG_WARNING,("Failed to find node to cover ip %s\n",
1606 ctdb_addr_to_str(&tmp_ip->addr)));
1612 /* Basic non-deterministic rebalancing algorithm.
1614 static void basic_failback(struct ctdb_context *ctdb,
1615 struct ctdb_node_map *nodemap,
1616 struct ctdb_public_ip_list *all_ips,
1620 int maxnode, maxnum, minnode, minnum, num, retries;
1621 struct ctdb_public_ip_list *tmp_ip;
1629 /* for each ip address, loop over all nodes that can serve
1630 this ip and make sure that the difference between the node
1631 serving the most and the node serving the least ip's are
1634 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1635 if (tmp_ip->pnn == -1) {
1639 /* Get the highest and lowest number of ips's served by any
1640 valid node which can serve this ip.
1644 for (i=0;i<nodemap->num;i++) {
1645 /* only check nodes that can actually serve this ip */
1646 if (!can_node_takeover_ip(ctdb, i, nodemap, tmp_ip)) {
1647 /* no it couldnt so skip to the next node */
1651 num = node_ip_coverage(ctdb, i, all_ips);
1652 if (maxnode == -1) {
1661 if (minnode == -1) {
1671 if (maxnode == -1) {
1672 DEBUG(DEBUG_WARNING,(__location__ " Could not find maxnode. May not be able to serve ip '%s'\n",
1673 ctdb_addr_to_str(&tmp_ip->addr)));
1678 /* if the spread between the smallest and largest coverage by
1679 a node is >=2 we steal one of the ips from the node with
1680 most coverage to even things out a bit.
1681 try to do this a limited number of times since we dont
1682 want to spend too much time balancing the ip coverage.
1684 if ( (maxnum > minnum+1)
1685 && (retries < (num_ips + 5)) ){
1686 struct ctdb_public_ip_list *tmp;
1688 /* Reassign one of maxnode's VNNs */
1689 for (tmp=all_ips;tmp;tmp=tmp->next) {
1690 if (tmp->pnn == maxnode) {
1691 (void)find_takeover_node(ctdb, nodemap, tmp, all_ips);
1700 struct ctdb_rebalancenodes {
1701 struct ctdb_rebalancenodes *next;
1704 static struct ctdb_rebalancenodes *force_rebalance_list = NULL;
1707 /* set this flag to force the node to be rebalanced even if it just didnt
1708 become healthy again.
1710 void lcp2_forcerebalance(struct ctdb_context *ctdb, uint32_t pnn)
1712 struct ctdb_rebalancenodes *rebalance;
1714 for (rebalance = force_rebalance_list; rebalance; rebalance = rebalance->next) {
1715 if (rebalance->pnn == pnn) {
1720 rebalance = talloc(ctdb, struct ctdb_rebalancenodes);
1721 rebalance->pnn = pnn;
1722 rebalance->next = force_rebalance_list;
1723 force_rebalance_list = rebalance;
1726 /* Do necessary LCP2 initialisation. Bury it in a function here so
1727 * that we can unit test it.
1729 static void lcp2_init(struct ctdb_context * tmp_ctx,
1730 struct ctdb_node_map * nodemap,
1732 struct ctdb_public_ip_list *all_ips,
1733 uint32_t **lcp2_imbalances,
1734 bool **rebalance_candidates)
1737 struct ctdb_public_ip_list *tmp_ip;
1739 *rebalance_candidates = talloc_array(tmp_ctx, bool, nodemap->num);
1740 CTDB_NO_MEMORY_FATAL(tmp_ctx, *rebalance_candidates);
1741 *lcp2_imbalances = talloc_array(tmp_ctx, uint32_t, nodemap->num);
1742 CTDB_NO_MEMORY_FATAL(tmp_ctx, *lcp2_imbalances);
1744 for (i=0;i<nodemap->num;i++) {
1745 (*lcp2_imbalances)[i] = lcp2_imbalance(all_ips, i);
1746 /* First step: assume all nodes are candidates */
1747 (*rebalance_candidates)[i] = true;
1750 /* 2nd step: if a node has IPs assigned then it must have been
1751 * healthy before, so we remove it from consideration. This
1752 * is overkill but is all we have because we don't maintain
1753 * state between takeover runs. An alternative would be to
1754 * keep state and invalidate it every time the recovery master
1757 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1758 if (tmp_ip->pnn != -1) {
1759 (*rebalance_candidates)[tmp_ip->pnn] = false;
1763 /* 3rd step: if a node is forced to re-balance then
1764 we allow failback onto the node */
1765 while (force_rebalance_list != NULL) {
1766 struct ctdb_rebalancenodes *next = force_rebalance_list->next;
1768 if (force_rebalance_list->pnn <= nodemap->num) {
1769 (*rebalance_candidates)[force_rebalance_list->pnn] = true;
1772 DEBUG(DEBUG_ERR,("During ipreallocation, forced rebalance of node %d\n", force_rebalance_list->pnn));
1773 talloc_free(force_rebalance_list);
1774 force_rebalance_list = next;
1778 /* Allocate any unassigned addresses using the LCP2 algorithm to find
1779 * the IP/node combination that will cost the least.
1781 static void lcp2_allocate_unassigned(struct ctdb_context *ctdb,
1782 struct ctdb_node_map *nodemap,
1783 struct ctdb_public_ip_list *all_ips,
1784 uint32_t *lcp2_imbalances)
1786 struct ctdb_public_ip_list *tmp_ip;
1790 uint32_t mindsum, dstdsum, dstimbl, minimbl;
1791 struct ctdb_public_ip_list *minip;
1793 bool should_loop = true;
1794 bool have_unassigned = true;
1796 while (have_unassigned && should_loop) {
1797 should_loop = false;
1799 DEBUG(DEBUG_DEBUG,(" ----------------------------------------\n"));
1800 DEBUG(DEBUG_DEBUG,(" CONSIDERING MOVES (UNASSIGNED)\n"));
1806 /* loop over each unassigned ip. */
1807 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1808 if (tmp_ip->pnn != -1) {
1812 for (dstnode=0; dstnode < nodemap->num; dstnode++) {
1813 /* only check nodes that can actually takeover this ip */
1814 if (!can_node_takeover_ip(ctdb, dstnode,
1816 /* no it couldnt so skip to the next node */
1820 dstdsum = ip_distance_2_sum(&(tmp_ip->addr), all_ips, dstnode);
1821 dstimbl = lcp2_imbalances[dstnode] + dstdsum;
1822 DEBUG(DEBUG_DEBUG,(" %s -> %d [+%d]\n",
1823 ctdb_addr_to_str(&(tmp_ip->addr)),
1825 dstimbl - lcp2_imbalances[dstnode]));
1828 if ((minnode == -1) || (dstdsum < mindsum)) {
1838 DEBUG(DEBUG_DEBUG,(" ----------------------------------------\n"));
1840 /* If we found one then assign it to the given node. */
1841 if (minnode != -1) {
1842 minip->pnn = minnode;
1843 lcp2_imbalances[minnode] = minimbl;
1844 DEBUG(DEBUG_INFO,(" %s -> %d [+%d]\n",
1845 ctdb_addr_to_str(&(minip->addr)),
1850 /* There might be a better way but at least this is clear. */
1851 have_unassigned = false;
1852 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1853 if (tmp_ip->pnn == -1) {
1854 have_unassigned = true;
1859 /* We know if we have an unassigned addresses so we might as
1862 if (have_unassigned) {
1863 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1864 if (tmp_ip->pnn == -1) {
1865 DEBUG(DEBUG_WARNING,("Failed to find node to cover ip %s\n",
1866 ctdb_addr_to_str(&tmp_ip->addr)));
1872 /* LCP2 algorithm for rebalancing the cluster. Given a candidate node
1873 * to move IPs from, determines the best IP/destination node
1874 * combination to move from the source node.
1876 static bool lcp2_failback_candidate(struct ctdb_context *ctdb,
1877 struct ctdb_node_map *nodemap,
1878 struct ctdb_public_ip_list *all_ips,
1881 uint32_t *lcp2_imbalances,
1882 bool *rebalance_candidates)
1884 int dstnode, mindstnode;
1885 uint32_t srcimbl, srcdsum, dstimbl, dstdsum;
1886 uint32_t minsrcimbl, mindstimbl;
1887 struct ctdb_public_ip_list *minip;
1888 struct ctdb_public_ip_list *tmp_ip;
1890 /* Find an IP and destination node that best reduces imbalance. */
1896 DEBUG(DEBUG_DEBUG,(" ----------------------------------------\n"));
1897 DEBUG(DEBUG_DEBUG,(" CONSIDERING MOVES FROM %d [%d]\n", srcnode, candimbl));
1899 for (tmp_ip=all_ips; tmp_ip; tmp_ip=tmp_ip->next) {
1900 /* Only consider addresses on srcnode. */
1901 if (tmp_ip->pnn != srcnode) {
1905 /* What is this IP address costing the source node? */
1906 srcdsum = ip_distance_2_sum(&(tmp_ip->addr), all_ips, srcnode);
1907 srcimbl = candimbl - srcdsum;
1909 /* Consider this IP address would cost each potential
1910 * destination node. Destination nodes are limited to
1911 * those that are newly healthy, since we don't want
1912 * to do gratuitous failover of IPs just to make minor
1913 * balance improvements.
1915 for (dstnode=0; dstnode < nodemap->num; dstnode++) {
1916 if (!rebalance_candidates[dstnode]) {
1920 /* only check nodes that can actually takeover this ip */
1921 if (!can_node_takeover_ip(ctdb, dstnode,
1923 /* no it couldnt so skip to the next node */
1927 dstdsum = ip_distance_2_sum(&(tmp_ip->addr), all_ips, dstnode);
1928 dstimbl = lcp2_imbalances[dstnode] + dstdsum;
1929 DEBUG(DEBUG_DEBUG,(" %d [%d] -> %s -> %d [+%d]\n",
1930 srcnode, srcimbl - lcp2_imbalances[srcnode],
1931 ctdb_addr_to_str(&(tmp_ip->addr)),
1932 dstnode, dstimbl - lcp2_imbalances[dstnode]));
1934 if ((dstimbl < candimbl) && (dstdsum < srcdsum) && \
1935 ((mindstnode == -1) || \
1936 ((srcimbl + dstimbl) < (minsrcimbl + mindstimbl)))) {
1939 minsrcimbl = srcimbl;
1940 mindstnode = dstnode;
1941 mindstimbl = dstimbl;
1945 DEBUG(DEBUG_DEBUG,(" ----------------------------------------\n"));
1947 if (mindstnode != -1) {
1948 /* We found a move that makes things better... */
1949 DEBUG(DEBUG_INFO,("%d [%d] -> %s -> %d [+%d]\n",
1950 srcnode, minsrcimbl - lcp2_imbalances[srcnode],
1951 ctdb_addr_to_str(&(minip->addr)),
1952 mindstnode, mindstimbl - lcp2_imbalances[mindstnode]));
1955 lcp2_imbalances[srcnode] = srcimbl;
1956 lcp2_imbalances[mindstnode] = mindstimbl;
1957 minip->pnn = mindstnode;
1966 struct lcp2_imbalance_pnn {
1971 static int lcp2_cmp_imbalance_pnn(const void * a, const void * b)
1973 const struct lcp2_imbalance_pnn * lipa = (const struct lcp2_imbalance_pnn *) a;
1974 const struct lcp2_imbalance_pnn * lipb = (const struct lcp2_imbalance_pnn *) b;
1976 if (lipa->imbalance > lipb->imbalance) {
1978 } else if (lipa->imbalance == lipb->imbalance) {
1985 /* LCP2 algorithm for rebalancing the cluster. This finds the source
1986 * node with the highest LCP2 imbalance, and then determines the best
1987 * IP/destination node combination to move from the source node.
1989 static void lcp2_failback(struct ctdb_context *ctdb,
1990 struct ctdb_node_map *nodemap,
1991 struct ctdb_public_ip_list *all_ips,
1992 uint32_t *lcp2_imbalances,
1993 bool *rebalance_candidates)
1995 int i, num_rebalance_candidates;
1996 struct lcp2_imbalance_pnn * lips;
2001 /* It is only worth continuing if we have suitable target
2002 * nodes to transfer IPs to. This check is much cheaper than
2005 num_rebalance_candidates = 0;
2006 for (i = 0; i < nodemap->num; i++) {
2007 if (rebalance_candidates[i]) {
2008 num_rebalance_candidates++;
2011 if (num_rebalance_candidates == 0) {
2015 /* Put the imbalances and nodes into an array, sort them and
2016 * iterate through candidates. Usually the 1st one will be
2017 * used, so this doesn't cost much...
2019 lips = talloc_array(ctdb, struct lcp2_imbalance_pnn, nodemap->num);
2020 for (i = 0; i < nodemap->num; i++) {
2021 lips[i].imbalance = lcp2_imbalances[i];
2024 qsort(lips, nodemap->num, sizeof(struct lcp2_imbalance_pnn),
2025 lcp2_cmp_imbalance_pnn);
2028 for (i = 0; i < nodemap->num; i++) {
2029 /* This means that all nodes had 0 or 1 addresses, so
2030 * can't be imbalanced.
2032 if (lips[i].imbalance == 0) {
2036 if (lcp2_failback_candidate(ctdb,
2042 rebalance_candidates)) {
2054 static void unassign_unsuitable_ips(struct ctdb_context *ctdb,
2055 struct ctdb_node_map *nodemap,
2056 struct ctdb_public_ip_list *all_ips)
2058 struct ctdb_public_ip_list *tmp_ip;
2060 /* verify that the assigned nodes can serve that public ip
2061 and set it to -1 if not
2063 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
2064 if (tmp_ip->pnn == -1) {
2067 if (!can_node_host_ip(ctdb, tmp_ip->pnn,
2068 nodemap, tmp_ip) != 0) {
2069 /* this node can not serve this ip. */
2070 DEBUG(DEBUG_DEBUG,("Unassign IP: %s from %d\n",
2071 ctdb_addr_to_str(&(tmp_ip->addr)),
2078 static void ip_alloc_deterministic_ips(struct ctdb_context *ctdb,
2079 struct ctdb_node_map *nodemap,
2080 struct ctdb_public_ip_list *all_ips)
2082 struct ctdb_public_ip_list *tmp_ip;
2085 DEBUG(DEBUG_NOTICE,("Deterministic IPs enabled. Resetting all ip allocations\n"));
2086 /* Allocate IPs to nodes in a modulo fashion so that IPs will
2087 * always be allocated the same way for a specific set of
2088 * available/unavailable nodes.
2091 for (i=0,tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next,i++) {
2092 tmp_ip->pnn = i%nodemap->num;
2095 /* IP failback doesn't make sense with deterministic
2096 * IPs, since the modulo step above implicitly fails
2097 * back IPs to their "home" node.
2099 if (1 == ctdb->tunable.no_ip_failback) {
2100 DEBUG(DEBUG_WARNING, ("WARNING: 'NoIPFailback' set but ignored - incompatible with 'DeterministicIPs\n"));
2103 unassign_unsuitable_ips(ctdb, nodemap, all_ips);
2105 basic_allocate_unassigned(ctdb, nodemap, all_ips);
2107 /* No failback here! */
2110 static void ip_alloc_nondeterministic_ips(struct ctdb_context *ctdb,
2111 struct ctdb_node_map *nodemap,
2112 struct ctdb_public_ip_list *all_ips)
2114 /* This should be pushed down into basic_failback. */
2115 struct ctdb_public_ip_list *tmp_ip;
2117 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
2121 unassign_unsuitable_ips(ctdb, nodemap, all_ips);
2123 basic_allocate_unassigned(ctdb, nodemap, all_ips);
2125 /* If we don't want IPs to fail back then don't rebalance IPs. */
2126 if (1 == ctdb->tunable.no_ip_failback) {
2130 /* Now, try to make sure the ip adresses are evenly distributed
2133 basic_failback(ctdb, nodemap, all_ips, num_ips);
2136 static void ip_alloc_lcp2(struct ctdb_context *ctdb,
2137 struct ctdb_node_map *nodemap,
2138 struct ctdb_public_ip_list *all_ips,
2141 uint32_t *lcp2_imbalances;
2142 bool *rebalance_candidates;
2144 TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
2146 unassign_unsuitable_ips(ctdb, nodemap, all_ips);
2148 lcp2_init(tmp_ctx, nodemap, mask, all_ips,
2149 &lcp2_imbalances, &rebalance_candidates);
2151 lcp2_allocate_unassigned(ctdb, nodemap, all_ips, lcp2_imbalances);
2153 /* If we don't want IPs to fail back then don't rebalance IPs. */
2154 if (1 == ctdb->tunable.no_ip_failback) {
2158 /* Now, try to make sure the ip adresses are evenly distributed
2161 lcp2_failback(ctdb, nodemap, all_ips,
2162 lcp2_imbalances, rebalance_candidates);
2165 talloc_free(tmp_ctx);
2168 static bool all_nodes_are_disabled(struct ctdb_node_map *nodemap)
2172 /* Count how many completely healthy nodes we have */
2174 for (i=0;i<nodemap->num;i++) {
2175 if (!(nodemap->nodes[i].flags & (NODE_FLAGS_INACTIVE|NODE_FLAGS_DISABLED))) {
2180 return num_healthy == 0;
2183 /* The calculation part of the IP allocation algorithm. */
2184 static void ctdb_takeover_run_core(struct ctdb_context *ctdb,
2185 struct ctdb_node_map *nodemap,
2186 struct ctdb_public_ip_list **all_ips_p)
2190 /* If we have healthy nodes then we will only consider them
2191 for serving public addresses
2193 mask = NODE_FLAGS_INACTIVE|NODE_FLAGS_DISABLED;
2194 if (all_nodes_are_disabled(nodemap) &&
2195 (ctdb->tunable.no_ip_host_on_all_disabled == 0)) {
2196 /* We didnt have any completely healthy nodes so
2197 use "disabled" nodes as a fallback
2199 mask = NODE_FLAGS_INACTIVE;
2202 /* since nodes only know about those public addresses that
2203 can be served by that particular node, no single node has
2204 a full list of all public addresses that exist in the cluster.
2205 Walk over all node structures and create a merged list of
2206 all public addresses that exist in the cluster.
2208 keep the tree of ips around as ctdb->ip_tree
2210 *all_ips_p = create_merged_ip_list(ctdb);
2212 if (1 == ctdb->tunable.lcp2_public_ip_assignment) {
2213 ip_alloc_lcp2(ctdb, nodemap, *all_ips_p, mask);
2214 } else if (1 == ctdb->tunable.deterministic_public_ips) {
2215 ip_alloc_deterministic_ips(ctdb, nodemap, *all_ips_p);
2217 ip_alloc_nondeterministic_ips(ctdb, nodemap, *all_ips_p);
2220 /* at this point ->pnn is the node which will own each IP
2221 or -1 if there is no node that can cover this ip
2227 struct get_tunable_callback_data {
2228 const char *tunable;
2232 static void get_tunable_callback(struct ctdb_context *ctdb, uint32_t pnn,
2233 int32_t res, TDB_DATA outdata,
2236 struct get_tunable_callback_data *cd =
2237 (struct get_tunable_callback_data *)callback;
2242 ("Failure to read \"%s\" tunable from remote node %d\n",
2247 if (outdata.dsize != sizeof(uint32_t)) {
2248 DEBUG(DEBUG_ERR,("Wrong size of returned data when reading \"%s\" tunable from node %d. Expected %d bytes but received %d bytes\n",
2249 cd->tunable, pnn, (int)sizeof(uint32_t),
2250 (int)outdata.dsize));
2254 size = talloc_get_size(cd->out) / sizeof(uint32_t);
2256 DEBUG(DEBUG_ERR,("Got %s reply from node %d but nodemap only has %d entries\n",
2257 cd->tunable, pnn, size));
2262 cd->out[pnn] = *(uint32_t *)outdata.dptr;
2265 static uint32_t *get_tunable_from_nodes(struct ctdb_context *ctdb,
2266 TALLOC_CTX *tmp_ctx,
2267 struct ctdb_node_map *nodemap,
2268 const char *tunable)
2271 struct ctdb_control_get_tunable *t;
2274 struct get_tunable_callback_data callback_data;
2276 tvals = talloc_zero_array(tmp_ctx, uint32_t, nodemap->num);
2277 CTDB_NO_MEMORY_NULL(ctdb, tvals);
2278 callback_data.out = tvals;
2279 callback_data.tunable = tunable;
2281 data.dsize = offsetof(struct ctdb_control_get_tunable, name) + strlen(tunable) + 1;
2282 data.dptr = talloc_size(tmp_ctx, data.dsize);
2283 t = (struct ctdb_control_get_tunable *)data.dptr;
2284 t->length = strlen(tunable)+1;
2285 memcpy(t->name, tunable, t->length);
2286 nodes = list_of_connected_nodes(ctdb, nodemap, tmp_ctx, true);
2287 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_GET_TUNABLE,
2288 nodes, 0, TAKEOVER_TIMEOUT(),
2290 get_tunable_callback, NULL,
2291 &callback_data) != 0) {
2292 DEBUG(DEBUG_ERR, (__location__ " ctdb_control to get %s tunable failed\n", tunable));
2295 talloc_free(data.dptr);
2300 /* Set internal flags for IP allocation:
2302 * Set NOIPTAKOVER ip flags from per-node NoIPTakeover tunable
2303 * Set NOIPHOST ip flag for each INACTIVE node
2304 * if all nodes are disabled:
2305 * Set NOIPHOST ip flags from per-node NoIPHostOnAllDisabled tunable
2307 * Set NOIPHOST ip flags for disabled nodes
2309 static void set_ipflags_internal(struct ctdb_node_map *nodemap,
2310 uint32_t *tval_noiptakeover,
2311 uint32_t *tval_noiphostonalldisabled)
2315 /* Clear IP flags */
2316 for (i=0;i<nodemap->num;i++) {
2317 nodemap->nodes[i].flags &=
2318 ~(NODE_FLAGS_NOIPTAKEOVER|NODE_FLAGS_NOIPHOST);
2321 for (i=0;i<nodemap->num;i++) {
2322 /* Can not take IPs on node with NoIPTakeover set */
2323 if (tval_noiptakeover[i] != 0) {
2324 nodemap->nodes[i].flags |= NODE_FLAGS_NOIPTAKEOVER;
2327 /* Can not host IPs on INACTIVE node */
2328 if (nodemap->nodes[i].flags & NODE_FLAGS_INACTIVE) {
2329 nodemap->nodes[i].flags |= NODE_FLAGS_NOIPHOST;
2333 if (all_nodes_are_disabled(nodemap)) {
2334 /* If all nodes are disabled, can not host IPs on node
2335 * with NoIPHostOnAllDisabled set
2337 for (i=0;i<nodemap->num;i++) {
2338 if (tval_noiphostonalldisabled[i] != 0) {
2339 nodemap->nodes[i].flags |= NODE_FLAGS_NOIPHOST;
2343 /* If some nodes are not disabled, then can not host
2344 * IPs on DISABLED node
2346 for (i=0;i<nodemap->num;i++) {
2347 if (nodemap->nodes[i].flags & NODE_FLAGS_DISABLED) {
2348 nodemap->nodes[i].flags |= NODE_FLAGS_NOIPHOST;
2354 static bool set_ipflags(struct ctdb_context *ctdb,
2355 TALLOC_CTX *tmp_ctx,
2356 struct ctdb_node_map *nodemap)
2358 uint32_t *tval_noiptakeover;
2359 uint32_t *tval_noiphostonalldisabled;
2361 tval_noiptakeover = get_tunable_from_nodes(ctdb, tmp_ctx, nodemap,
2363 if (tval_noiptakeover == NULL) {
2367 tval_noiphostonalldisabled =
2368 get_tunable_from_nodes(ctdb, tmp_ctx, nodemap,
2369 "NoIPHostOnAllDisabled");
2370 if (tval_noiphostonalldisabled == NULL) {
2374 set_ipflags_internal(nodemap,
2375 tval_noiptakeover, tval_noiphostonalldisabled);
2377 talloc_free(tval_noiptakeover);
2378 talloc_free(tval_noiphostonalldisabled);
2384 make any IP alias changes for public addresses that are necessary
2386 int ctdb_takeover_run(struct ctdb_context *ctdb, struct ctdb_node_map *nodemap,
2387 client_async_callback fail_callback, void *callback_data)
2390 struct ctdb_public_ip ip;
2391 struct ctdb_public_ipv4 ipv4;
2393 struct ctdb_public_ip_list *all_ips, *tmp_ip;
2395 struct timeval timeout;
2396 struct client_async_data *async_data;
2397 struct ctdb_client_control_state *state;
2398 TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
2399 uint32_t disable_timeout;
2402 * ip failover is completely disabled, just send out the
2403 * ipreallocated event.
2405 if (ctdb->tunable.disable_ip_failover != 0) {
2409 if (!set_ipflags(ctdb, tmp_ctx, nodemap)) {
2410 DEBUG(DEBUG_ERR,("Failed to set IP flags from tunables\n"));
2416 /* Do the IP reassignment calculations */
2417 ctdb_takeover_run_core(ctdb, nodemap, &all_ips);
2419 /* The recovery daemon does regular sanity checks of the IPs.
2420 * However, sometimes it is overzealous and thinks changes are
2421 * required when they're already underway. This stops the
2422 * checks for a while before we start moving IPs.
2424 disable_timeout = ctdb->tunable.takeover_timeout;
2425 data.dptr = (uint8_t*)&disable_timeout;
2426 data.dsize = sizeof(disable_timeout);
2427 if (ctdb_client_send_message(ctdb, CTDB_BROADCAST_CONNECTED,
2428 CTDB_SRVID_DISABLE_IP_CHECK, data) != 0) {
2429 DEBUG(DEBUG_INFO,("Failed to disable ip verification\n"));
2432 /* now tell all nodes to delete any alias that they should not
2433 have. This will be a NOOP on nodes that don't currently
2434 hold the given alias */
2435 async_data = talloc_zero(tmp_ctx, struct client_async_data);
2436 CTDB_NO_MEMORY_FATAL(ctdb, async_data);
2438 async_data->fail_callback = fail_callback;
2439 async_data->callback_data = callback_data;
2441 for (i=0;i<nodemap->num;i++) {
2442 /* don't talk to unconnected nodes, but do talk to banned nodes */
2443 if (nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED) {
2447 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
2448 if (tmp_ip->pnn == nodemap->nodes[i].pnn) {
2449 /* This node should be serving this
2450 vnn so dont tell it to release the ip
2454 if (tmp_ip->addr.sa.sa_family == AF_INET) {
2455 ipv4.pnn = tmp_ip->pnn;
2456 ipv4.sin = tmp_ip->addr.ip;
2458 timeout = TAKEOVER_TIMEOUT();
2459 data.dsize = sizeof(ipv4);
2460 data.dptr = (uint8_t *)&ipv4;
2461 state = ctdb_control_send(ctdb, nodemap->nodes[i].pnn,
2462 0, CTDB_CONTROL_RELEASE_IPv4, 0,
2466 ip.pnn = tmp_ip->pnn;
2467 ip.addr = tmp_ip->addr;
2469 timeout = TAKEOVER_TIMEOUT();
2470 data.dsize = sizeof(ip);
2471 data.dptr = (uint8_t *)&ip;
2472 state = ctdb_control_send(ctdb, nodemap->nodes[i].pnn,
2473 0, CTDB_CONTROL_RELEASE_IP, 0,
2478 if (state == NULL) {
2479 DEBUG(DEBUG_ERR,(__location__ " Failed to call async control CTDB_CONTROL_RELEASE_IP to node %u\n", nodemap->nodes[i].pnn));
2480 talloc_free(tmp_ctx);
2484 ctdb_client_async_add(async_data, state);
2487 if (ctdb_client_async_wait(ctdb, async_data) != 0) {
2488 DEBUG(DEBUG_ERR,(__location__ " Async control CTDB_CONTROL_RELEASE_IP failed\n"));
2489 talloc_free(tmp_ctx);
2492 talloc_free(async_data);
2495 /* tell all nodes to get their own IPs */
2496 async_data = talloc_zero(tmp_ctx, struct client_async_data);
2497 CTDB_NO_MEMORY_FATAL(ctdb, async_data);
2499 async_data->fail_callback = fail_callback;
2500 async_data->callback_data = callback_data;
2502 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
2503 if (tmp_ip->pnn == -1) {
2504 /* this IP won't be taken over */
2508 if (tmp_ip->addr.sa.sa_family == AF_INET) {
2509 ipv4.pnn = tmp_ip->pnn;
2510 ipv4.sin = tmp_ip->addr.ip;
2512 timeout = TAKEOVER_TIMEOUT();
2513 data.dsize = sizeof(ipv4);
2514 data.dptr = (uint8_t *)&ipv4;
2515 state = ctdb_control_send(ctdb, tmp_ip->pnn,
2516 0, CTDB_CONTROL_TAKEOVER_IPv4, 0,
2520 ip.pnn = tmp_ip->pnn;
2521 ip.addr = tmp_ip->addr;
2523 timeout = TAKEOVER_TIMEOUT();
2524 data.dsize = sizeof(ip);
2525 data.dptr = (uint8_t *)&ip;
2526 state = ctdb_control_send(ctdb, tmp_ip->pnn,
2527 0, CTDB_CONTROL_TAKEOVER_IP, 0,
2531 if (state == NULL) {
2532 DEBUG(DEBUG_ERR,(__location__ " Failed to call async control CTDB_CONTROL_TAKEOVER_IP to node %u\n", tmp_ip->pnn));
2533 talloc_free(tmp_ctx);
2537 ctdb_client_async_add(async_data, state);
2539 if (ctdb_client_async_wait(ctdb, async_data) != 0) {
2540 DEBUG(DEBUG_ERR,(__location__ " Async control CTDB_CONTROL_TAKEOVER_IP failed\n"));
2541 talloc_free(tmp_ctx);
2547 * Tell all nodes to run eventscripts to process the
2548 * "ipreallocated" event. This can do a lot of things,
2549 * including restarting services to reconfigure them if public
2550 * IPs have moved. Once upon a time this event only used to
2553 nodes = list_of_connected_nodes(ctdb, nodemap, tmp_ctx, true);
2554 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_IPREALLOCATED,
2555 nodes, 0, TAKEOVER_TIMEOUT(),
2557 NULL, fail_callback,
2558 callback_data) != 0) {
2559 DEBUG(DEBUG_ERR, (__location__ " failed to send control to run eventscripts with \"ipreallocated\"\n"));
2562 talloc_free(tmp_ctx);
2568 destroy a ctdb_client_ip structure
2570 static int ctdb_client_ip_destructor(struct ctdb_client_ip *ip)
2572 DEBUG(DEBUG_DEBUG,("destroying client tcp for %s:%u (client_id %u)\n",
2573 ctdb_addr_to_str(&ip->addr),
2574 ntohs(ip->addr.ip.sin_port),
2577 DLIST_REMOVE(ip->ctdb->client_ip_list, ip);
2582 called by a client to inform us of a TCP connection that it is managing
2583 that should tickled with an ACK when IP takeover is done
2584 we handle both the old ipv4 style of packets as well as the new ipv4/6
2587 int32_t ctdb_control_tcp_client(struct ctdb_context *ctdb, uint32_t client_id,
2590 struct ctdb_client *client = ctdb_reqid_find(ctdb, client_id, struct ctdb_client);
2591 struct ctdb_control_tcp *old_addr = NULL;
2592 struct ctdb_control_tcp_addr new_addr;
2593 struct ctdb_control_tcp_addr *tcp_sock = NULL;
2594 struct ctdb_tcp_list *tcp;
2595 struct ctdb_tcp_connection t;
2598 struct ctdb_client_ip *ip;
2599 struct ctdb_vnn *vnn;
2600 ctdb_sock_addr addr;
2602 switch (indata.dsize) {
2603 case sizeof(struct ctdb_control_tcp):
2604 old_addr = (struct ctdb_control_tcp *)indata.dptr;
2605 ZERO_STRUCT(new_addr);
2606 tcp_sock = &new_addr;
2607 tcp_sock->src.ip = old_addr->src;
2608 tcp_sock->dest.ip = old_addr->dest;
2610 case sizeof(struct ctdb_control_tcp_addr):
2611 tcp_sock = (struct ctdb_control_tcp_addr *)indata.dptr;
2614 DEBUG(DEBUG_ERR,(__location__ " Invalid data structure passed "
2615 "to ctdb_control_tcp_client. size was %d but "
2616 "only allowed sizes are %lu and %lu\n",
2618 (long unsigned)sizeof(struct ctdb_control_tcp),
2619 (long unsigned)sizeof(struct ctdb_control_tcp_addr)));
2623 addr = tcp_sock->src;
2624 ctdb_canonicalize_ip(&addr, &tcp_sock->src);
2625 addr = tcp_sock->dest;
2626 ctdb_canonicalize_ip(&addr, &tcp_sock->dest);
2629 memcpy(&addr, &tcp_sock->dest, sizeof(addr));
2630 vnn = find_public_ip_vnn(ctdb, &addr);
2632 switch (addr.sa.sa_family) {
2634 if (ntohl(addr.ip.sin_addr.s_addr) != INADDR_LOOPBACK) {
2635 DEBUG(DEBUG_ERR,("Could not add client IP %s. This is not a public address.\n",
2636 ctdb_addr_to_str(&addr)));
2640 DEBUG(DEBUG_ERR,("Could not add client IP %s. This is not a public ipv6 address.\n",
2641 ctdb_addr_to_str(&addr)));
2644 DEBUG(DEBUG_ERR,(__location__ " Unknown family type %d\n", addr.sa.sa_family));
2650 if (vnn->pnn != ctdb->pnn) {
2651 DEBUG(DEBUG_ERR,("Attempt to register tcp client for IP %s we don't hold - failing (client_id %u pid %u)\n",
2652 ctdb_addr_to_str(&addr),
2653 client_id, client->pid));
2654 /* failing this call will tell smbd to die */
2658 ip = talloc(client, struct ctdb_client_ip);
2659 CTDB_NO_MEMORY(ctdb, ip);
2663 ip->client_id = client_id;
2664 talloc_set_destructor(ip, ctdb_client_ip_destructor);
2665 DLIST_ADD(ctdb->client_ip_list, ip);
2667 tcp = talloc(client, struct ctdb_tcp_list);
2668 CTDB_NO_MEMORY(ctdb, tcp);
2670 tcp->connection.src_addr = tcp_sock->src;
2671 tcp->connection.dst_addr = tcp_sock->dest;
2673 DLIST_ADD(client->tcp_list, tcp);
2675 t.src_addr = tcp_sock->src;
2676 t.dst_addr = tcp_sock->dest;
2678 data.dptr = (uint8_t *)&t;
2679 data.dsize = sizeof(t);
2681 switch (addr.sa.sa_family) {
2683 DEBUG(DEBUG_INFO,("registered tcp client for %u->%s:%u (client_id %u pid %u)\n",
2684 (unsigned)ntohs(tcp_sock->dest.ip.sin_port),
2685 ctdb_addr_to_str(&tcp_sock->src),
2686 (unsigned)ntohs(tcp_sock->src.ip.sin_port), client_id, client->pid));
2689 DEBUG(DEBUG_INFO,("registered tcp client for %u->%s:%u (client_id %u pid %u)\n",
2690 (unsigned)ntohs(tcp_sock->dest.ip6.sin6_port),
2691 ctdb_addr_to_str(&tcp_sock->src),
2692 (unsigned)ntohs(tcp_sock->src.ip6.sin6_port), client_id, client->pid));
2695 DEBUG(DEBUG_ERR,(__location__ " Unknown family %d\n", addr.sa.sa_family));
2699 /* tell all nodes about this tcp connection */
2700 ret = ctdb_daemon_send_control(ctdb, CTDB_BROADCAST_CONNECTED, 0,
2701 CTDB_CONTROL_TCP_ADD,
2702 0, CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL);
2704 DEBUG(DEBUG_ERR,(__location__ " Failed to send CTDB_CONTROL_TCP_ADD\n"));
2712 find a tcp address on a list
2714 static struct ctdb_tcp_connection *ctdb_tcp_find(struct ctdb_tcp_array *array,
2715 struct ctdb_tcp_connection *tcp)
2719 if (array == NULL) {
2723 for (i=0;i<array->num;i++) {
2724 if (ctdb_same_sockaddr(&array->connections[i].src_addr, &tcp->src_addr) &&
2725 ctdb_same_sockaddr(&array->connections[i].dst_addr, &tcp->dst_addr)) {
2726 return &array->connections[i];
2735 called by a daemon to inform us of a TCP connection that one of its
2736 clients managing that should tickled with an ACK when IP takeover is
2739 int32_t ctdb_control_tcp_add(struct ctdb_context *ctdb, TDB_DATA indata, bool tcp_update_needed)
2741 struct ctdb_tcp_connection *p = (struct ctdb_tcp_connection *)indata.dptr;
2742 struct ctdb_tcp_array *tcparray;
2743 struct ctdb_tcp_connection tcp;
2744 struct ctdb_vnn *vnn;
2746 vnn = find_public_ip_vnn(ctdb, &p->dst_addr);
2748 DEBUG(DEBUG_INFO,(__location__ " got TCP_ADD control for an address which is not a public address '%s'\n",
2749 ctdb_addr_to_str(&p->dst_addr)));
2755 tcparray = vnn->tcp_array;
2757 /* If this is the first tickle */
2758 if (tcparray == NULL) {
2759 tcparray = talloc_size(ctdb->nodes,
2760 offsetof(struct ctdb_tcp_array, connections) +
2761 sizeof(struct ctdb_tcp_connection) * 1);
2762 CTDB_NO_MEMORY(ctdb, tcparray);
2763 vnn->tcp_array = tcparray;
2766 tcparray->connections = talloc_size(tcparray, sizeof(struct ctdb_tcp_connection));
2767 CTDB_NO_MEMORY(ctdb, tcparray->connections);
2769 tcparray->connections[tcparray->num].src_addr = p->src_addr;
2770 tcparray->connections[tcparray->num].dst_addr = p->dst_addr;
2773 if (tcp_update_needed) {
2774 vnn->tcp_update_needed = true;
2780 /* Do we already have this tickle ?*/
2781 tcp.src_addr = p->src_addr;
2782 tcp.dst_addr = p->dst_addr;
2783 if (ctdb_tcp_find(vnn->tcp_array, &tcp) != NULL) {
2784 DEBUG(DEBUG_DEBUG,("Already had tickle info for %s:%u for vnn:%u\n",
2785 ctdb_addr_to_str(&tcp.dst_addr),
2786 ntohs(tcp.dst_addr.ip.sin_port),
2791 /* A new tickle, we must add it to the array */
2792 tcparray->connections = talloc_realloc(tcparray, tcparray->connections,
2793 struct ctdb_tcp_connection,
2795 CTDB_NO_MEMORY(ctdb, tcparray->connections);
2797 vnn->tcp_array = tcparray;
2798 tcparray->connections[tcparray->num].src_addr = p->src_addr;
2799 tcparray->connections[tcparray->num].dst_addr = p->dst_addr;
2802 DEBUG(DEBUG_INFO,("Added tickle info for %s:%u from vnn %u\n",
2803 ctdb_addr_to_str(&tcp.dst_addr),
2804 ntohs(tcp.dst_addr.ip.sin_port),
2807 if (tcp_update_needed) {
2808 vnn->tcp_update_needed = true;
2816 called by a daemon to inform us of a TCP connection that one of its
2817 clients managing that should tickled with an ACK when IP takeover is
2820 static void ctdb_remove_tcp_connection(struct ctdb_context *ctdb, struct ctdb_tcp_connection *conn)
2822 struct ctdb_tcp_connection *tcpp;
2823 struct ctdb_vnn *vnn = find_public_ip_vnn(ctdb, &conn->dst_addr);
2826 DEBUG(DEBUG_ERR,(__location__ " unable to find public address %s\n",
2827 ctdb_addr_to_str(&conn->dst_addr)));
2831 /* if the array is empty we cant remove it
2832 and we dont need to do anything
2834 if (vnn->tcp_array == NULL) {
2835 DEBUG(DEBUG_INFO,("Trying to remove tickle that doesnt exist (array is empty) %s:%u\n",
2836 ctdb_addr_to_str(&conn->dst_addr),
2837 ntohs(conn->dst_addr.ip.sin_port)));
2842 /* See if we know this connection
2843 if we dont know this connection then we dont need to do anything
2845 tcpp = ctdb_tcp_find(vnn->tcp_array, conn);
2847 DEBUG(DEBUG_INFO,("Trying to remove tickle that doesnt exist %s:%u\n",
2848 ctdb_addr_to_str(&conn->dst_addr),
2849 ntohs(conn->dst_addr.ip.sin_port)));
2854 /* We need to remove this entry from the array.
2855 Instead of allocating a new array and copying data to it
2856 we cheat and just copy the last entry in the existing array
2857 to the entry that is to be removed and just shring the
2860 *tcpp = vnn->tcp_array->connections[vnn->tcp_array->num - 1];
2861 vnn->tcp_array->num--;
2863 /* If we deleted the last entry we also need to remove the entire array
2865 if (vnn->tcp_array->num == 0) {
2866 talloc_free(vnn->tcp_array);
2867 vnn->tcp_array = NULL;
2870 vnn->tcp_update_needed = true;
2872 DEBUG(DEBUG_INFO,("Removed tickle info for %s:%u\n",
2873 ctdb_addr_to_str(&conn->src_addr),
2874 ntohs(conn->src_addr.ip.sin_port)));
2879 called by a daemon to inform us of a TCP connection that one of its
2880 clients used are no longer needed in the tickle database
2882 int32_t ctdb_control_tcp_remove(struct ctdb_context *ctdb, TDB_DATA indata)
2884 struct ctdb_tcp_connection *conn = (struct ctdb_tcp_connection *)indata.dptr;
2886 ctdb_remove_tcp_connection(ctdb, conn);
2893 called when a daemon restarts - send all tickes for all public addresses
2894 we are serving immediately to the new node.
2896 int32_t ctdb_control_startup(struct ctdb_context *ctdb, uint32_t vnn)
2898 /*XXX here we should send all tickes we are serving to the new node */
2904 called when a client structure goes away - hook to remove
2905 elements from the tcp_list in all daemons
2907 void ctdb_takeover_client_destructor_hook(struct ctdb_client *client)
2909 while (client->tcp_list) {
2910 struct ctdb_tcp_list *tcp = client->tcp_list;
2911 DLIST_REMOVE(client->tcp_list, tcp);
2912 ctdb_remove_tcp_connection(client->ctdb, &tcp->connection);
2918 release all IPs on shutdown
2920 void ctdb_release_all_ips(struct ctdb_context *ctdb)
2922 struct ctdb_vnn *vnn;
2924 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
2925 if (!ctdb_sys_have_ip(&vnn->public_address)) {
2926 ctdb_vnn_unassign_iface(ctdb, vnn);
2932 ctdb_event_script_args(ctdb, CTDB_EVENT_RELEASE_IP, "%s %s %u",
2933 ctdb_vnn_iface_string(vnn),
2934 ctdb_addr_to_str(&vnn->public_address),
2935 vnn->public_netmask_bits);
2936 release_kill_clients(ctdb, &vnn->public_address);
2937 ctdb_vnn_unassign_iface(ctdb, vnn);
2943 get list of public IPs
2945 int32_t ctdb_control_get_public_ips(struct ctdb_context *ctdb,
2946 struct ctdb_req_control *c, TDB_DATA *outdata)
2949 struct ctdb_all_public_ips *ips;
2950 struct ctdb_vnn *vnn;
2951 bool only_available = false;
2953 if (c->flags & CTDB_PUBLIC_IP_FLAGS_ONLY_AVAILABLE) {
2954 only_available = true;
2957 /* count how many public ip structures we have */
2959 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
2963 len = offsetof(struct ctdb_all_public_ips, ips) +
2964 num*sizeof(struct ctdb_public_ip);
2965 ips = talloc_zero_size(outdata, len);
2966 CTDB_NO_MEMORY(ctdb, ips);
2969 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
2970 if (only_available && !ctdb_vnn_available(ctdb, vnn)) {
2973 ips->ips[i].pnn = vnn->pnn;
2974 ips->ips[i].addr = vnn->public_address;
2978 len = offsetof(struct ctdb_all_public_ips, ips) +
2979 i*sizeof(struct ctdb_public_ip);
2981 outdata->dsize = len;
2982 outdata->dptr = (uint8_t *)ips;
2989 get list of public IPs, old ipv4 style. only returns ipv4 addresses
2991 int32_t ctdb_control_get_public_ipsv4(struct ctdb_context *ctdb,
2992 struct ctdb_req_control *c, TDB_DATA *outdata)
2995 struct ctdb_all_public_ipsv4 *ips;
2996 struct ctdb_vnn *vnn;
2998 /* count how many public ip structures we have */
3000 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
3001 if (vnn->public_address.sa.sa_family != AF_INET) {
3007 len = offsetof(struct ctdb_all_public_ipsv4, ips) +
3008 num*sizeof(struct ctdb_public_ipv4);
3009 ips = talloc_zero_size(outdata, len);
3010 CTDB_NO_MEMORY(ctdb, ips);
3012 outdata->dsize = len;
3013 outdata->dptr = (uint8_t *)ips;
3017 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
3018 if (vnn->public_address.sa.sa_family != AF_INET) {
3021 ips->ips[i].pnn = vnn->pnn;
3022 ips->ips[i].sin = vnn->public_address.ip;
3029 int32_t ctdb_control_get_public_ip_info(struct ctdb_context *ctdb,
3030 struct ctdb_req_control *c,
3035 ctdb_sock_addr *addr;
3036 struct ctdb_control_public_ip_info *info;
3037 struct ctdb_vnn *vnn;
3039 addr = (ctdb_sock_addr *)indata.dptr;
3041 vnn = find_public_ip_vnn(ctdb, addr);
3043 /* if it is not a public ip it could be our 'single ip' */
3044 if (ctdb->single_ip_vnn) {
3045 if (ctdb_same_ip(&ctdb->single_ip_vnn->public_address, addr)) {
3046 vnn = ctdb->single_ip_vnn;
3051 DEBUG(DEBUG_ERR,(__location__ " Could not get public ip info, "
3052 "'%s'not a public address\n",
3053 ctdb_addr_to_str(addr)));
3057 /* count how many public ip structures we have */
3059 for (;vnn->ifaces[num];) {
3063 len = offsetof(struct ctdb_control_public_ip_info, ifaces) +
3064 num*sizeof(struct ctdb_control_iface_info);
3065 info = talloc_zero_size(outdata, len);
3066 CTDB_NO_MEMORY(ctdb, info);
3068 info->ip.addr = vnn->public_address;
3069 info->ip.pnn = vnn->pnn;
3070 info->active_idx = 0xFFFFFFFF;
3072 for (i=0; vnn->ifaces[i]; i++) {
3073 struct ctdb_iface *cur;
3075 cur = ctdb_find_iface(ctdb, vnn->ifaces[i]);
3077 DEBUG(DEBUG_CRIT, (__location__ " internal error iface[%s] unknown\n",
3081 if (vnn->iface == cur) {
3082 info->active_idx = i;
3084 strcpy(info->ifaces[i].name, cur->name);
3085 info->ifaces[i].link_state = cur->link_up;
3086 info->ifaces[i].references = cur->references;
3089 len = offsetof(struct ctdb_control_public_ip_info, ifaces) +
3090 i*sizeof(struct ctdb_control_iface_info);
3092 outdata->dsize = len;
3093 outdata->dptr = (uint8_t *)info;
3098 int32_t ctdb_control_get_ifaces(struct ctdb_context *ctdb,
3099 struct ctdb_req_control *c,
3103 struct ctdb_control_get_ifaces *ifaces;
3104 struct ctdb_iface *cur;
3106 /* count how many public ip structures we have */
3108 for (cur=ctdb->ifaces;cur;cur=cur->next) {
3112 len = offsetof(struct ctdb_control_get_ifaces, ifaces) +
3113 num*sizeof(struct ctdb_control_iface_info);
3114 ifaces = talloc_zero_size(outdata, len);
3115 CTDB_NO_MEMORY(ctdb, ifaces);
3118 for (cur=ctdb->ifaces;cur;cur=cur->next) {
3119 strcpy(ifaces->ifaces[i].name, cur->name);
3120 ifaces->ifaces[i].link_state = cur->link_up;
3121 ifaces->ifaces[i].references = cur->references;
3125 len = offsetof(struct ctdb_control_get_ifaces, ifaces) +
3126 i*sizeof(struct ctdb_control_iface_info);
3128 outdata->dsize = len;
3129 outdata->dptr = (uint8_t *)ifaces;
3134 int32_t ctdb_control_set_iface_link(struct ctdb_context *ctdb,
3135 struct ctdb_req_control *c,
3138 struct ctdb_control_iface_info *info;
3139 struct ctdb_iface *iface;
3140 bool link_up = false;
3142 info = (struct ctdb_control_iface_info *)indata.dptr;
3144 if (info->name[CTDB_IFACE_SIZE] != '\0') {
3145 int len = strnlen(info->name, CTDB_IFACE_SIZE);
3146 DEBUG(DEBUG_ERR, (__location__ " name[%*.*s] not terminated\n",
3147 len, len, info->name));
3151 switch (info->link_state) {
3159 DEBUG(DEBUG_ERR, (__location__ " link_state[%u] invalid\n",
3160 (unsigned int)info->link_state));
3164 if (info->references != 0) {
3165 DEBUG(DEBUG_ERR, (__location__ " references[%u] should be 0\n",
3166 (unsigned int)info->references));
3170 iface = ctdb_find_iface(ctdb, info->name);
3171 if (iface == NULL) {
3175 if (link_up == iface->link_up) {
3179 DEBUG(iface->link_up?DEBUG_ERR:DEBUG_NOTICE,
3180 ("iface[%s] has changed it's link status %s => %s\n",
3182 iface->link_up?"up":"down",
3183 link_up?"up":"down"));
3185 iface->link_up = link_up;
3191 structure containing the listening socket and the list of tcp connections
3192 that the ctdb daemon is to kill
3194 struct ctdb_kill_tcp {
3195 struct ctdb_vnn *vnn;
3196 struct ctdb_context *ctdb;
3198 struct fd_event *fde;
3199 trbt_tree_t *connections;
3204 a tcp connection that is to be killed
3206 struct ctdb_killtcp_con {
3207 ctdb_sock_addr src_addr;
3208 ctdb_sock_addr dst_addr;
3210 struct ctdb_kill_tcp *killtcp;
3213 /* this function is used to create a key to represent this socketpair
3214 in the killtcp tree.
3215 this key is used to insert and lookup matching socketpairs that are
3216 to be tickled and RST
3218 #define KILLTCP_KEYLEN 10
3219 static uint32_t *killtcp_key(ctdb_sock_addr *src, ctdb_sock_addr *dst)
3221 static uint32_t key[KILLTCP_KEYLEN];
3223 bzero(key, sizeof(key));
3225 if (src->sa.sa_family != dst->sa.sa_family) {
3226 DEBUG(DEBUG_ERR, (__location__ " ERROR, different families passed :%u vs %u\n", src->sa.sa_family, dst->sa.sa_family));
3230 switch (src->sa.sa_family) {
3232 key[0] = dst->ip.sin_addr.s_addr;
3233 key[1] = src->ip.sin_addr.s_addr;
3234 key[2] = dst->ip.sin_port;
3235 key[3] = src->ip.sin_port;
3238 uint32_t *dst6_addr32 =
3239 (uint32_t *)&(dst->ip6.sin6_addr.s6_addr);
3240 uint32_t *src6_addr32 =
3241 (uint32_t *)&(src->ip6.sin6_addr.s6_addr);
3242 key[0] = dst6_addr32[3];
3243 key[1] = src6_addr32[3];
3244 key[2] = dst6_addr32[2];
3245 key[3] = src6_addr32[2];
3246 key[4] = dst6_addr32[1];
3247 key[5] = src6_addr32[1];
3248 key[6] = dst6_addr32[0];
3249 key[7] = src6_addr32[0];
3250 key[8] = dst->ip6.sin6_port;
3251 key[9] = src->ip6.sin6_port;
3255 DEBUG(DEBUG_ERR, (__location__ " ERROR, unknown family passed :%u\n", src->sa.sa_family));
3263 called when we get a read event on the raw socket
3265 static void capture_tcp_handler(struct event_context *ev, struct fd_event *fde,
3266 uint16_t flags, void *private_data)
3268 struct ctdb_kill_tcp *killtcp = talloc_get_type(private_data, struct ctdb_kill_tcp);
3269 struct ctdb_killtcp_con *con;
3270 ctdb_sock_addr src, dst;
3271 uint32_t ack_seq, seq;
3273 if (!(flags & EVENT_FD_READ)) {
3277 if (ctdb_sys_read_tcp_packet(killtcp->capture_fd,
3278 killtcp->private_data,
3280 &ack_seq, &seq) != 0) {
3281 /* probably a non-tcp ACK packet */
3285 /* check if we have this guy in our list of connections
3288 con = trbt_lookuparray32(killtcp->connections,
3289 KILLTCP_KEYLEN, killtcp_key(&src, &dst));
3291 /* no this was some other packet we can just ignore */
3295 /* This one has been tickled !
3296 now reset him and remove him from the list.
3298 DEBUG(DEBUG_INFO, ("sending a tcp reset to kill connection :%d -> %s:%d\n",
3299 ntohs(con->dst_addr.ip.sin_port),
3300 ctdb_addr_to_str(&con->src_addr),
3301 ntohs(con->src_addr.ip.sin_port)));
3303 ctdb_sys_send_tcp(&con->dst_addr, &con->src_addr, ack_seq, seq, 1);
3308 /* when traversing the list of all tcp connections to send tickle acks to
3309 (so that we can capture the ack coming back and kill the connection
3311 this callback is called for each connection we are currently trying to kill
3313 static int tickle_connection_traverse(void *param, void *data)
3315 struct ctdb_killtcp_con *con = talloc_get_type(data, struct ctdb_killtcp_con);
3317 /* have tried too many times, just give up */
3318 if (con->count >= 5) {
3319 /* can't delete in traverse: reparent to delete_cons */
3320 talloc_steal(param, con);
3324 /* othervise, try tickling it again */
3327 (ctdb_sock_addr *)&con->dst_addr,
3328 (ctdb_sock_addr *)&con->src_addr,
3335 called every second until all sentenced connections have been reset
3337 static void ctdb_tickle_sentenced_connections(struct event_context *ev, struct timed_event *te,
3338 struct timeval t, void *private_data)
3340 struct ctdb_kill_tcp *killtcp = talloc_get_type(private_data, struct ctdb_kill_tcp);
3341 void *delete_cons = talloc_new(NULL);
3343 /* loop over all connections sending tickle ACKs */
3344 trbt_traversearray32(killtcp->connections, KILLTCP_KEYLEN, tickle_connection_traverse, delete_cons);
3346 /* now we've finished traverse, it's safe to do deletion. */
3347 talloc_free(delete_cons);
3349 /* If there are no more connections to kill we can remove the
3350 entire killtcp structure
3352 if ( (killtcp->connections == NULL) ||
3353 (killtcp->connections->root == NULL) ) {
3354 talloc_free(killtcp);
3358 /* try tickling them again in a seconds time
3360 event_add_timed(killtcp->ctdb->ev, killtcp, timeval_current_ofs(1, 0),
3361 ctdb_tickle_sentenced_connections, killtcp);
3365 destroy the killtcp structure
3367 static int ctdb_killtcp_destructor(struct ctdb_kill_tcp *killtcp)
3369 struct ctdb_vnn *tmpvnn;
3371 /* verify that this vnn is still active */
3372 for (tmpvnn = killtcp->ctdb->vnn; tmpvnn; tmpvnn = tmpvnn->next) {
3373 if (tmpvnn == killtcp->vnn) {
3378 if (tmpvnn == NULL) {
3382 if (killtcp->vnn->killtcp != killtcp) {
3386 killtcp->vnn->killtcp = NULL;
3392 /* nothing fancy here, just unconditionally replace any existing
3393 connection structure with the new one.
3395 dont even free the old one if it did exist, that one is talloc_stolen
3396 by the same node in the tree anyway and will be deleted when the new data
3399 static void *add_killtcp_callback(void *parm, void *data)
3405 add a tcp socket to the list of connections we want to RST
3407 static int ctdb_killtcp_add_connection(struct ctdb_context *ctdb,
3411 ctdb_sock_addr src, dst;
3412 struct ctdb_kill_tcp *killtcp;
3413 struct ctdb_killtcp_con *con;
3414 struct ctdb_vnn *vnn;
3416 ctdb_canonicalize_ip(s, &src);
3417 ctdb_canonicalize_ip(d, &dst);
3419 vnn = find_public_ip_vnn(ctdb, &dst);
3421 vnn = find_public_ip_vnn(ctdb, &src);
3424 /* if it is not a public ip it could be our 'single ip' */
3425 if (ctdb->single_ip_vnn) {
3426 if (ctdb_same_ip(&ctdb->single_ip_vnn->public_address, &dst)) {
3427 vnn = ctdb->single_ip_vnn;
3432 DEBUG(DEBUG_ERR,(__location__ " Could not killtcp, not a public address\n"));
3436 killtcp = vnn->killtcp;
3438 /* If this is the first connection to kill we must allocate
3441 if (killtcp == NULL) {
3442 killtcp = talloc_zero(vnn, struct ctdb_kill_tcp);
3443 CTDB_NO_MEMORY(ctdb, killtcp);
3446 killtcp->ctdb = ctdb;
3447 killtcp->capture_fd = -1;
3448 killtcp->connections = trbt_create(killtcp, 0);
3450 vnn->killtcp = killtcp;
3451 talloc_set_destructor(killtcp, ctdb_killtcp_destructor);
3456 /* create a structure that describes this connection we want to
3457 RST and store it in killtcp->connections
3459 con = talloc(killtcp, struct ctdb_killtcp_con);
3460 CTDB_NO_MEMORY(ctdb, con);
3461 con->src_addr = src;
3462 con->dst_addr = dst;
3464 con->killtcp = killtcp;
3467 trbt_insertarray32_callback(killtcp->connections,
3468 KILLTCP_KEYLEN, killtcp_key(&con->dst_addr, &con->src_addr),
3469 add_killtcp_callback, con);
3472 If we dont have a socket to listen on yet we must create it
3474 if (killtcp->capture_fd == -1) {
3475 const char *iface = ctdb_vnn_iface_string(vnn);
3476 killtcp->capture_fd = ctdb_sys_open_capture_socket(iface, &killtcp->private_data);
3477 if (killtcp->capture_fd == -1) {
3478 DEBUG(DEBUG_CRIT,(__location__ " Failed to open capturing "
3479 "socket on iface '%s' for killtcp (%s)\n",
3480 iface, strerror(errno)));
3486 if (killtcp->fde == NULL) {
3487 killtcp->fde = event_add_fd(ctdb->ev, killtcp, killtcp->capture_fd,
3489 capture_tcp_handler, killtcp);
3490 tevent_fd_set_auto_close(killtcp->fde);
3492 /* We also need to set up some events to tickle all these connections
3493 until they are all reset
3495 event_add_timed(ctdb->ev, killtcp, timeval_current_ofs(1, 0),
3496 ctdb_tickle_sentenced_connections, killtcp);
3499 /* tickle him once now */
3508 talloc_free(vnn->killtcp);
3509 vnn->killtcp = NULL;
3514 kill a TCP connection.
3516 int32_t ctdb_control_kill_tcp(struct ctdb_context *ctdb, TDB_DATA indata)
3518 struct ctdb_control_killtcp *killtcp = (struct ctdb_control_killtcp *)indata.dptr;
3520 return ctdb_killtcp_add_connection(ctdb, &killtcp->src_addr, &killtcp->dst_addr);
3524 called by a daemon to inform us of the entire list of TCP tickles for
3525 a particular public address.
3526 this control should only be sent by the node that is currently serving
3527 that public address.
3529 int32_t ctdb_control_set_tcp_tickle_list(struct ctdb_context *ctdb, TDB_DATA indata)
3531 struct ctdb_control_tcp_tickle_list *list = (struct ctdb_control_tcp_tickle_list *)indata.dptr;
3532 struct ctdb_tcp_array *tcparray;
3533 struct ctdb_vnn *vnn;
3535 /* We must at least have tickles.num or else we cant verify the size
3536 of the received data blob
3538 if (indata.dsize < offsetof(struct ctdb_control_tcp_tickle_list,
3539 tickles.connections)) {
3540 DEBUG(DEBUG_ERR,("Bad indata in ctdb_control_set_tcp_tickle_list. Not enough data for the tickle.num field\n"));
3544 /* verify that the size of data matches what we expect */
3545 if (indata.dsize < offsetof(struct ctdb_control_tcp_tickle_list,
3546 tickles.connections)
3547 + sizeof(struct ctdb_tcp_connection)
3548 * list->tickles.num) {
3549 DEBUG(DEBUG_ERR,("Bad indata in ctdb_control_set_tcp_tickle_list\n"));
3553 vnn = find_public_ip_vnn(ctdb, &list->addr);
3555 DEBUG(DEBUG_INFO,(__location__ " Could not set tcp tickle list, '%s' is not a public address\n",
3556 ctdb_addr_to_str(&list->addr)));
3561 /* remove any old ticklelist we might have */
3562 talloc_free(vnn->tcp_array);
3563 vnn->tcp_array = NULL;
3565 tcparray = talloc(ctdb->nodes, struct ctdb_tcp_array);
3566 CTDB_NO_MEMORY(ctdb, tcparray);
3568 tcparray->num = list->tickles.num;
3570 tcparray->connections = talloc_array(tcparray, struct ctdb_tcp_connection, tcparray->num);
3571 CTDB_NO_MEMORY(ctdb, tcparray->connections);
3573 memcpy(tcparray->connections, &list->tickles.connections[0],
3574 sizeof(struct ctdb_tcp_connection)*tcparray->num);
3576 /* We now have a new fresh tickle list array for this vnn */
3577 vnn->tcp_array = talloc_steal(vnn, tcparray);
3583 called to return the full list of tickles for the puclic address associated
3584 with the provided vnn
3586 int32_t ctdb_control_get_tcp_tickle_list(struct ctdb_context *ctdb, TDB_DATA indata, TDB_DATA *outdata)
3588 ctdb_sock_addr *addr = (ctdb_sock_addr *)indata.dptr;
3589 struct ctdb_control_tcp_tickle_list *list;
3590 struct ctdb_tcp_array *tcparray;
3592 struct ctdb_vnn *vnn;
3594 vnn = find_public_ip_vnn(ctdb, addr);
3596 DEBUG(DEBUG_ERR,(__location__ " Could not get tcp tickle list, '%s' is not a public address\n",
3597 ctdb_addr_to_str(addr)));
3602 tcparray = vnn->tcp_array;
3604 num = tcparray->num;
3609 outdata->dsize = offsetof(struct ctdb_control_tcp_tickle_list,
3610 tickles.connections)
3611 + sizeof(struct ctdb_tcp_connection) * num;
3613 outdata->dptr = talloc_size(outdata, outdata->dsize);
3614 CTDB_NO_MEMORY(ctdb, outdata->dptr);
3615 list = (struct ctdb_control_tcp_tickle_list *)outdata->dptr;
3618 list->tickles.num = num;
3620 memcpy(&list->tickles.connections[0], tcparray->connections,
3621 sizeof(struct ctdb_tcp_connection) * num);
3629 set the list of all tcp tickles for a public address
3631 static int ctdb_ctrl_set_tcp_tickles(struct ctdb_context *ctdb,
3632 struct timeval timeout, uint32_t destnode,
3633 ctdb_sock_addr *addr,
3634 struct ctdb_tcp_array *tcparray)
3638 struct ctdb_control_tcp_tickle_list *list;
3641 num = tcparray->num;
3646 data.dsize = offsetof(struct ctdb_control_tcp_tickle_list,
3647 tickles.connections) +
3648 sizeof(struct ctdb_tcp_connection) * num;
3649 data.dptr = talloc_size(ctdb, data.dsize);
3650 CTDB_NO_MEMORY(ctdb, data.dptr);
3652 list = (struct ctdb_control_tcp_tickle_list *)data.dptr;
3654 list->tickles.num = num;
3656 memcpy(&list->tickles.connections[0], tcparray->connections, sizeof(struct ctdb_tcp_connection) * num);
3659 ret = ctdb_daemon_send_control(ctdb, CTDB_BROADCAST_CONNECTED, 0,
3660 CTDB_CONTROL_SET_TCP_TICKLE_LIST,
3661 0, CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL);
3663 DEBUG(DEBUG_ERR,(__location__ " ctdb_control for set tcp tickles failed\n"));
3667 talloc_free(data.dptr);
3674 perform tickle updates if required
3676 static void ctdb_update_tcp_tickles(struct event_context *ev,
3677 struct timed_event *te,
3678 struct timeval t, void *private_data)
3680 struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
3682 struct ctdb_vnn *vnn;
3684 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
3685 /* we only send out updates for public addresses that
3688 if (ctdb->pnn != vnn->pnn) {
3691 /* We only send out the updates if we need to */
3692 if (!vnn->tcp_update_needed) {
3695 ret = ctdb_ctrl_set_tcp_tickles(ctdb,
3697 CTDB_BROADCAST_CONNECTED,
3698 &vnn->public_address,
3701 DEBUG(DEBUG_ERR,("Failed to send the tickle update for public address %s\n",
3702 ctdb_addr_to_str(&vnn->public_address)));
3706 event_add_timed(ctdb->ev, ctdb->tickle_update_context,
3707 timeval_current_ofs(ctdb->tunable.tickle_update_interval, 0),
3708 ctdb_update_tcp_tickles, ctdb);
3713 start periodic update of tcp tickles
3715 void ctdb_start_tcp_tickle_update(struct ctdb_context *ctdb)
3717 ctdb->tickle_update_context = talloc_new(ctdb);
3719 event_add_timed(ctdb->ev, ctdb->tickle_update_context,
3720 timeval_current_ofs(ctdb->tunable.tickle_update_interval, 0),
3721 ctdb_update_tcp_tickles, ctdb);
3727 struct control_gratious_arp {
3728 struct ctdb_context *ctdb;
3729 ctdb_sock_addr addr;
3735 send a control_gratuitous arp
3737 static void send_gratious_arp(struct event_context *ev, struct timed_event *te,
3738 struct timeval t, void *private_data)
3741 struct control_gratious_arp *arp = talloc_get_type(private_data,
3742 struct control_gratious_arp);
3744 ret = ctdb_sys_send_arp(&arp->addr, arp->iface);
3746 DEBUG(DEBUG_ERR,(__location__ " sending of gratious arp on iface '%s' failed (%s)\n",
3747 arp->iface, strerror(errno)));
3752 if (arp->count == CTDB_ARP_REPEAT) {
3757 event_add_timed(arp->ctdb->ev, arp,
3758 timeval_current_ofs(CTDB_ARP_INTERVAL, 0),
3759 send_gratious_arp, arp);
3766 int32_t ctdb_control_send_gratious_arp(struct ctdb_context *ctdb, TDB_DATA indata)
3768 struct ctdb_control_gratious_arp *gratious_arp = (struct ctdb_control_gratious_arp *)indata.dptr;
3769 struct control_gratious_arp *arp;
3771 /* verify the size of indata */
3772 if (indata.dsize < offsetof(struct ctdb_control_gratious_arp, iface)) {
3773 DEBUG(DEBUG_ERR,(__location__ " Too small indata to hold a ctdb_control_gratious_arp structure. Got %u require %u bytes\n",
3774 (unsigned)indata.dsize,
3775 (unsigned)offsetof(struct ctdb_control_gratious_arp, iface)));
3779 ( offsetof(struct ctdb_control_gratious_arp, iface)
3780 + gratious_arp->len ) ){
3782 DEBUG(DEBUG_ERR,(__location__ " Wrong size of indata. Was %u bytes "
3783 "but should be %u bytes\n",
3784 (unsigned)indata.dsize,
3785 (unsigned)(offsetof(struct ctdb_control_gratious_arp, iface)+gratious_arp->len)));
3790 arp = talloc(ctdb, struct control_gratious_arp);
3791 CTDB_NO_MEMORY(ctdb, arp);
3794 arp->addr = gratious_arp->addr;
3795 arp->iface = talloc_strdup(arp, gratious_arp->iface);
3796 CTDB_NO_MEMORY(ctdb, arp->iface);
3799 event_add_timed(arp->ctdb->ev, arp,
3800 timeval_zero(), send_gratious_arp, arp);
3805 int32_t ctdb_control_add_public_address(struct ctdb_context *ctdb, TDB_DATA indata)
3807 struct ctdb_control_ip_iface *pub = (struct ctdb_control_ip_iface *)indata.dptr;
3810 /* verify the size of indata */
3811 if (indata.dsize < offsetof(struct ctdb_control_ip_iface, iface)) {
3812 DEBUG(DEBUG_ERR,(__location__ " Too small indata to hold a ctdb_control_ip_iface structure\n"));
3816 ( offsetof(struct ctdb_control_ip_iface, iface)
3819 DEBUG(DEBUG_ERR,(__location__ " Wrong size of indata. Was %u bytes "
3820 "but should be %u bytes\n",
3821 (unsigned)indata.dsize,
3822 (unsigned)(offsetof(struct ctdb_control_ip_iface, iface)+pub->len)));
3826 ret = ctdb_add_public_address(ctdb, &pub->addr, pub->mask, &pub->iface[0], true);
3829 DEBUG(DEBUG_ERR,(__location__ " Failed to add public address\n"));
3837 called when releaseip event finishes for del_public_address
3839 static void delete_ip_callback(struct ctdb_context *ctdb, int status,
3842 talloc_free(private_data);
3845 int32_t ctdb_control_del_public_address(struct ctdb_context *ctdb, TDB_DATA indata)
3847 struct ctdb_control_ip_iface *pub = (struct ctdb_control_ip_iface *)indata.dptr;
3848 struct ctdb_vnn *vnn;
3851 /* verify the size of indata */
3852 if (indata.dsize < offsetof(struct ctdb_control_ip_iface, iface)) {
3853 DEBUG(DEBUG_ERR,(__location__ " Too small indata to hold a ctdb_control_ip_iface structure\n"));
3857 ( offsetof(struct ctdb_control_ip_iface, iface)
3860 DEBUG(DEBUG_ERR,(__location__ " Wrong size of indata. Was %u bytes "
3861 "but should be %u bytes\n",
3862 (unsigned)indata.dsize,
3863 (unsigned)(offsetof(struct ctdb_control_ip_iface, iface)+pub->len)));
3867 /* walk over all public addresses until we find a match */
3868 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
3869 if (ctdb_same_ip(&vnn->public_address, &pub->addr)) {
3870 TALLOC_CTX *mem_ctx = talloc_new(ctdb);
3872 DLIST_REMOVE(ctdb->vnn, vnn);
3873 talloc_steal(mem_ctx, vnn);
3874 ctdb_remove_orphaned_ifaces(ctdb, vnn, mem_ctx);
3875 if (vnn->pnn != ctdb->pnn) {
3876 if (vnn->iface != NULL) {
3877 ctdb_vnn_unassign_iface(ctdb, vnn);
3879 talloc_free(mem_ctx);
3884 ret = ctdb_event_script_callback(ctdb,
3885 mem_ctx, delete_ip_callback, mem_ctx,
3887 CTDB_EVENT_RELEASE_IP,
3889 ctdb_vnn_iface_string(vnn),
3890 ctdb_addr_to_str(&vnn->public_address),
3891 vnn->public_netmask_bits);
3892 if (vnn->iface != NULL) {
3893 ctdb_vnn_unassign_iface(ctdb, vnn);
3906 struct ipreallocated_callback_state {
3907 struct ctdb_req_control *c;
3910 static void ctdb_ipreallocated_callback(struct ctdb_context *ctdb,
3911 int status, void *p)
3913 struct ipreallocated_callback_state *state =
3914 talloc_get_type(p, struct ipreallocated_callback_state);
3918 (" \"ipreallocated\" event script failed (status %d)\n",
3920 if (status == -ETIME) {
3921 ctdb_ban_self(ctdb);
3925 ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
3929 /* A control to run the ipreallocated event */
3930 int32_t ctdb_control_ipreallocated(struct ctdb_context *ctdb,
3931 struct ctdb_req_control *c,
3935 struct ipreallocated_callback_state *state;
3937 state = talloc(ctdb, struct ipreallocated_callback_state);
3938 CTDB_NO_MEMORY(ctdb, state);
3940 DEBUG(DEBUG_INFO,(__location__ " Running \"ipreallocated\" event\n"));
3942 ret = ctdb_event_script_callback(ctdb, state,
3943 ctdb_ipreallocated_callback, state,
3944 false, CTDB_EVENT_IPREALLOCATED,
3948 DEBUG(DEBUG_ERR,("Failed to run \"ipreallocated\" event \n"));
3953 /* tell the control that we will be reply asynchronously */
3954 state->c = talloc_steal(state, c);
3955 *async_reply = true;
3961 /* This function is called from the recovery daemon to verify that a remote
3962 node has the expected ip allocation.
3963 This is verified against ctdb->ip_tree
3965 int verify_remote_ip_allocation(struct ctdb_context *ctdb, struct ctdb_all_public_ips *ips)
3967 struct ctdb_public_ip_list *tmp_ip;
3970 if (ctdb->ip_tree == NULL) {
3971 /* dont know the expected allocation yet, assume remote node
3980 for (i=0; i<ips->num; i++) {
3981 tmp_ip = trbt_lookuparray32(ctdb->ip_tree, IP_KEYLEN, ip_key(&ips->ips[i].addr));
3982 if (tmp_ip == NULL) {
3983 DEBUG(DEBUG_ERR,(__location__ " Could not find host for address %s, reassign ips\n", ctdb_addr_to_str(&ips->ips[i].addr)));
3987 if (tmp_ip->pnn == -1 || ips->ips[i].pnn == -1) {
3991 if (tmp_ip->pnn != ips->ips[i].pnn) {
3992 DEBUG(DEBUG_ERR,("Inconsistent ip allocation. Trigger reallocation. Thinks %s is held by node %u while it is held by node %u\n", ctdb_addr_to_str(&ips->ips[i].addr), ips->ips[i].pnn, tmp_ip->pnn));
4000 int update_ip_assignment_tree(struct ctdb_context *ctdb, struct ctdb_public_ip *ip)
4002 struct ctdb_public_ip_list *tmp_ip;
4004 if (ctdb->ip_tree == NULL) {
4005 DEBUG(DEBUG_ERR,("No ctdb->ip_tree yet. Failed to update ip assignment\n"));
4009 tmp_ip = trbt_lookuparray32(ctdb->ip_tree, IP_KEYLEN, ip_key(&ip->addr));
4010 if (tmp_ip == NULL) {
4011 DEBUG(DEBUG_ERR,(__location__ " Could not find record for address %s, update ip\n", ctdb_addr_to_str(&ip->addr)));
4015 DEBUG(DEBUG_NOTICE,("Updated ip assignment tree for ip : %s from node %u to node %u\n", ctdb_addr_to_str(&ip->addr), tmp_ip->pnn, ip->pnn));
4016 tmp_ip->pnn = ip->pnn;
4022 struct ctdb_reloadips_handle {
4023 struct ctdb_context *ctdb;
4024 struct ctdb_req_control *c;
4028 struct fd_event *fde;
4031 static int ctdb_reloadips_destructor(struct ctdb_reloadips_handle *h)
4033 if (h == h->ctdb->reload_ips) {
4034 h->ctdb->reload_ips = NULL;
4037 ctdb_request_control_reply(h->ctdb, h->c, NULL, h->status, NULL);
4040 ctdb_kill(h->ctdb, h->child, SIGKILL);
4044 static void ctdb_reloadips_timeout_event(struct event_context *ev,
4045 struct timed_event *te,
4046 struct timeval t, void *private_data)
4048 struct ctdb_reloadips_handle *h = talloc_get_type(private_data, struct ctdb_reloadips_handle);
4053 static void ctdb_reloadips_child_handler(struct event_context *ev, struct fd_event *fde,
4054 uint16_t flags, void *private_data)
4056 struct ctdb_reloadips_handle *h = talloc_get_type(private_data, struct ctdb_reloadips_handle);
4061 ret = read(h->fd[0], &res, 1);
4062 if (ret < 1 || res != 0) {
4063 DEBUG(DEBUG_ERR, (__location__ " Reloadips child process returned error\n"));
4071 static int ctdb_reloadips_child(struct ctdb_context *ctdb)
4073 TALLOC_CTX *mem_ctx = talloc_new(NULL);
4074 struct ctdb_all_public_ips *ips;
4075 struct ctdb_vnn *vnn;
4078 /* read the ip allocation from the local node */
4079 ret = ctdb_ctrl_get_public_ips(ctdb, TAKEOVER_TIMEOUT(), CTDB_CURRENT_NODE, mem_ctx, &ips);
4081 DEBUG(DEBUG_ERR, ("Unable to get public ips from local node\n"));
4082 talloc_free(mem_ctx);
4086 /* re-read the public ips file */
4088 if (ctdb_set_public_addresses(ctdb, false) != 0) {
4089 DEBUG(DEBUG_ERR,("Failed to re-read public addresses file\n"));
4090 talloc_free(mem_ctx);
4095 /* check the previous list of ips and scan for ips that have been
4098 for (i = 0; i < ips->num; i++) {
4099 for (vnn = ctdb->vnn; vnn; vnn = vnn->next) {
4100 if (ctdb_same_ip(&vnn->public_address, &ips->ips[i].addr)) {
4105 /* we need to delete this ip, no longer available on this node */
4107 struct ctdb_control_ip_iface pub;
4109 DEBUG(DEBUG_NOTICE,("RELOADIPS: IP%s is no longer available on this node. Deleting it.\n", ctdb_addr_to_str(&ips->ips[i].addr)));
4110 pub.addr = ips->ips[i].addr;
4114 ret = ctdb_ctrl_del_public_ip(ctdb, TAKEOVER_TIMEOUT(), CTDB_CURRENT_NODE, &pub);
4116 DEBUG(DEBUG_ERR, ("RELOADIPS: Unable to del public ip:%s from local node\n", ctdb_addr_to_str(&ips->ips[i].addr)));
4123 /* loop over all new ones and check the ones we need to add */
4124 for (vnn = ctdb->vnn; vnn; vnn = vnn->next) {
4125 for (i = 0; i < ips->num; i++) {
4126 if (ctdb_same_ip(&vnn->public_address, &ips->ips[i].addr)) {
4130 if (i == ips->num) {
4131 struct ctdb_control_ip_iface pub;
4132 const char *ifaces = NULL;
4135 DEBUG(DEBUG_NOTICE,("RELOADIPS: New ip:%s found, adding it.\n", ctdb_addr_to_str(&vnn->public_address)));
4137 pub.addr = vnn->public_address;
4138 pub.mask = vnn->public_netmask_bits;
4141 ifaces = vnn->ifaces[0];
4143 while (vnn->ifaces[iface] != NULL) {
4144 ifaces = talloc_asprintf(vnn, "%s,%s", ifaces, vnn->ifaces[iface]);
4147 pub.len = strlen(ifaces)+1;
4148 memcpy(&pub.iface[0], ifaces, strlen(ifaces)+1);
4150 ret = ctdb_ctrl_add_public_ip(ctdb, TAKEOVER_TIMEOUT(), CTDB_CURRENT_NODE, &pub);
4152 DEBUG(DEBUG_ERR, ("RELOADIPS: Unable to add public ip:%s to local node\n", ctdb_addr_to_str(&vnn->public_address)));
4161 /* This control is sent to force the node to re-read the public addresses file
4162 and drop any addresses we should nnot longer host, and add new addresses
4163 that we are now able to host
4165 int32_t ctdb_control_reload_public_ips(struct ctdb_context *ctdb, struct ctdb_req_control *c, bool *async_reply)
4167 struct ctdb_reloadips_handle *h;
4168 pid_t parent = getpid();
4170 if (ctdb->reload_ips != NULL) {
4171 talloc_free(ctdb->reload_ips);
4172 ctdb->reload_ips = NULL;
4175 h = talloc(ctdb, struct ctdb_reloadips_handle);
4176 CTDB_NO_MEMORY(ctdb, h);
4181 if (pipe(h->fd) == -1) {
4182 DEBUG(DEBUG_ERR,("Failed to create pipe for ctdb_freeze_lock\n"));
4187 h->child = ctdb_fork(ctdb);
4188 if (h->child == (pid_t)-1) {
4189 DEBUG(DEBUG_ERR, ("Failed to fork a child for reloadips\n"));
4197 if (h->child == 0) {
4198 signed char res = 0;
4201 debug_extra = talloc_asprintf(NULL, "reloadips:");
4203 if (switch_from_server_to_client(ctdb, "reloadips-child") != 0) {
4204 DEBUG(DEBUG_CRIT,("ERROR: Failed to switch reloadips child into client mode\n"));
4207 res = ctdb_reloadips_child(ctdb);
4209 DEBUG(DEBUG_ERR,("Failed to reload ips on local node\n"));
4213 write(h->fd[1], &res, 1);
4214 /* make sure we die when our parent dies */
4215 while (ctdb_kill(ctdb, parent, 0) == 0 || errno != ESRCH) {
4221 h->c = talloc_steal(h, c);
4224 set_close_on_exec(h->fd[0]);
4226 talloc_set_destructor(h, ctdb_reloadips_destructor);
4229 h->fde = event_add_fd(ctdb->ev, h, h->fd[0],
4230 EVENT_FD_READ, ctdb_reloadips_child_handler,
4232 tevent_fd_set_auto_close(h->fde);
4234 event_add_timed(ctdb->ev, h,
4235 timeval_current_ofs(120, 0),
4236 ctdb_reloadips_timeout_event, h);
4238 /* we reply later */
4239 *async_reply = true;