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/>.
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
36 /* Flags used in IP allocation algorithms. */
43 struct ctdb_iface *prev, *next;
49 static const char *ctdb_vnn_iface_string(const struct ctdb_vnn *vnn)
52 return vnn->iface->name;
58 static int ctdb_add_local_iface(struct ctdb_context *ctdb, const char *iface)
62 /* Verify that we dont have an entry for this ip yet */
63 for (i=ctdb->ifaces;i;i=i->next) {
64 if (strcmp(i->name, iface) == 0) {
69 /* create a new structure for this interface */
70 i = talloc_zero(ctdb, struct ctdb_iface);
71 CTDB_NO_MEMORY_FATAL(ctdb, i);
72 i->name = talloc_strdup(i, iface);
73 CTDB_NO_MEMORY(ctdb, i->name);
75 * If link_up defaults to true then IPs can be allocated to a
76 * node during the first recovery. However, then an interface
77 * could have its link marked down during the startup event,
78 * causing the IP to move almost immediately. If link_up
79 * defaults to false then, during normal operation, IPs added
80 * to a new interface can't be assigned until a monitor cycle
81 * has occurred and marked the new interfaces up. This makes
82 * IP allocation unpredictable. The following is a neat
83 * compromise: early in startup link_up defaults to false, so
84 * IPs can't be assigned, and after startup IPs can be
85 * assigned immediately.
87 i->link_up = (ctdb->runstate == CTDB_RUNSTATE_RUNNING);
89 DLIST_ADD(ctdb->ifaces, i);
94 static bool vnn_has_interface_with_name(struct ctdb_vnn *vnn,
99 for (n = 0; vnn->ifaces[n] != NULL; n++) {
100 if (strcmp(name, vnn->ifaces[n]) == 0) {
108 /* If any interfaces now have no possible IPs then delete them. This
109 * implementation is naive (i.e. simple) rather than clever
110 * (i.e. complex). Given that this is run on delip and that operation
111 * is rare, this doesn't need to be efficient - it needs to be
112 * foolproof. One alternative is reference counting, where the logic
113 * is distributed and can, therefore, be broken in multiple places.
114 * Another alternative is to build a red-black tree of interfaces that
115 * can have addresses (by walking ctdb->vnn and ctdb->single_ip_vnn
116 * once) and then walking ctdb->ifaces once and deleting those not in
117 * the tree. Let's go to one of those if the naive implementation
118 * causes problems... :-)
120 static void ctdb_remove_orphaned_ifaces(struct ctdb_context *ctdb,
121 struct ctdb_vnn *vnn,
124 struct ctdb_iface *i;
126 /* For each interface, check if there's an IP using it. */
127 for(i=ctdb->ifaces; i; i=i->next) {
131 /* Only consider interfaces named in the given VNN. */
132 if (!vnn_has_interface_with_name(vnn, i->name)) {
136 /* Is the "single IP" on this interface? */
137 if ((ctdb->single_ip_vnn != NULL) &&
138 (ctdb->single_ip_vnn->ifaces[0] != NULL) &&
139 (strcmp(i->name, ctdb->single_ip_vnn->ifaces[0]) == 0)) {
140 /* Found, next interface please... */
143 /* Search for a vnn with this interface. */
145 for (tv=ctdb->vnn; tv; tv=tv->next) {
146 if (vnn_has_interface_with_name(tv, i->name)) {
153 /* None of the VNNs are using this interface. */
154 DLIST_REMOVE(ctdb->ifaces, i);
155 /* Caller will free mem_ctx when convenient. */
156 talloc_steal(mem_ctx, i);
162 static struct ctdb_iface *ctdb_find_iface(struct ctdb_context *ctdb,
165 struct ctdb_iface *i;
167 for (i=ctdb->ifaces;i;i=i->next) {
168 if (strcmp(i->name, iface) == 0) {
176 static struct ctdb_iface *ctdb_vnn_best_iface(struct ctdb_context *ctdb,
177 struct ctdb_vnn *vnn)
180 struct ctdb_iface *cur = NULL;
181 struct ctdb_iface *best = NULL;
183 for (i=0; vnn->ifaces[i]; i++) {
185 cur = ctdb_find_iface(ctdb, vnn->ifaces[i]);
199 if (cur->references < best->references) {
208 static int32_t ctdb_vnn_assign_iface(struct ctdb_context *ctdb,
209 struct ctdb_vnn *vnn)
211 struct ctdb_iface *best = NULL;
214 DEBUG(DEBUG_INFO, (__location__ " public address '%s' "
215 "still assigned to iface '%s'\n",
216 ctdb_addr_to_str(&vnn->public_address),
217 ctdb_vnn_iface_string(vnn)));
221 best = ctdb_vnn_best_iface(ctdb, vnn);
223 DEBUG(DEBUG_ERR, (__location__ " public address '%s' "
224 "cannot assign to iface any iface\n",
225 ctdb_addr_to_str(&vnn->public_address)));
231 vnn->pnn = ctdb->pnn;
233 DEBUG(DEBUG_INFO, (__location__ " public address '%s' "
234 "now assigned to iface '%s' refs[%d]\n",
235 ctdb_addr_to_str(&vnn->public_address),
236 ctdb_vnn_iface_string(vnn),
241 static void ctdb_vnn_unassign_iface(struct ctdb_context *ctdb,
242 struct ctdb_vnn *vnn)
244 DEBUG(DEBUG_INFO, (__location__ " public address '%s' "
245 "now unassigned (old iface '%s' refs[%d])\n",
246 ctdb_addr_to_str(&vnn->public_address),
247 ctdb_vnn_iface_string(vnn),
248 vnn->iface?vnn->iface->references:0));
250 vnn->iface->references--;
253 if (vnn->pnn == ctdb->pnn) {
258 static bool ctdb_vnn_available(struct ctdb_context *ctdb,
259 struct ctdb_vnn *vnn)
263 if (vnn->iface && vnn->iface->link_up) {
267 for (i=0; vnn->ifaces[i]; i++) {
268 struct ctdb_iface *cur;
270 cur = ctdb_find_iface(ctdb, vnn->ifaces[i]);
283 struct ctdb_takeover_arp {
284 struct ctdb_context *ctdb;
287 struct ctdb_tcp_array *tcparray;
288 struct ctdb_vnn *vnn;
293 lists of tcp endpoints
295 struct ctdb_tcp_list {
296 struct ctdb_tcp_list *prev, *next;
297 struct ctdb_tcp_connection connection;
301 list of clients to kill on IP release
303 struct ctdb_client_ip {
304 struct ctdb_client_ip *prev, *next;
305 struct ctdb_context *ctdb;
312 send a gratuitous arp
314 static void ctdb_control_send_arp(struct event_context *ev, struct timed_event *te,
315 struct timeval t, void *private_data)
317 struct ctdb_takeover_arp *arp = talloc_get_type(private_data,
318 struct ctdb_takeover_arp);
320 struct ctdb_tcp_array *tcparray;
321 const char *iface = ctdb_vnn_iface_string(arp->vnn);
323 ret = ctdb_sys_send_arp(&arp->addr, iface);
325 DEBUG(DEBUG_CRIT,(__location__ " sending of arp failed on iface '%s' (%s)\n",
326 iface, strerror(errno)));
329 tcparray = arp->tcparray;
331 for (i=0;i<tcparray->num;i++) {
332 struct ctdb_tcp_connection *tcon;
334 tcon = &tcparray->connections[i];
335 DEBUG(DEBUG_INFO,("sending tcp tickle ack for %u->%s:%u\n",
336 (unsigned)ntohs(tcon->dst_addr.ip.sin_port),
337 ctdb_addr_to_str(&tcon->src_addr),
338 (unsigned)ntohs(tcon->src_addr.ip.sin_port)));
339 ret = ctdb_sys_send_tcp(
344 DEBUG(DEBUG_CRIT,(__location__ " Failed to send tcp tickle ack for %s\n",
345 ctdb_addr_to_str(&tcon->src_addr)));
352 if (arp->count == CTDB_ARP_REPEAT) {
357 event_add_timed(arp->ctdb->ev, arp->vnn->takeover_ctx,
358 timeval_current_ofs(CTDB_ARP_INTERVAL, 100000),
359 ctdb_control_send_arp, arp);
362 static int32_t ctdb_announce_vnn_iface(struct ctdb_context *ctdb,
363 struct ctdb_vnn *vnn)
365 struct ctdb_takeover_arp *arp;
366 struct ctdb_tcp_array *tcparray;
368 if (!vnn->takeover_ctx) {
369 vnn->takeover_ctx = talloc_new(vnn);
370 if (!vnn->takeover_ctx) {
375 arp = talloc_zero(vnn->takeover_ctx, struct ctdb_takeover_arp);
381 arp->addr = vnn->public_address;
384 tcparray = vnn->tcp_array;
386 /* add all of the known tcp connections for this IP to the
387 list of tcp connections to send tickle acks for */
388 arp->tcparray = talloc_steal(arp, tcparray);
390 vnn->tcp_array = NULL;
391 vnn->tcp_update_needed = true;
394 event_add_timed(arp->ctdb->ev, vnn->takeover_ctx,
395 timeval_zero(), ctdb_control_send_arp, arp);
400 struct takeover_callback_state {
401 struct ctdb_req_control *c;
402 ctdb_sock_addr *addr;
403 struct ctdb_vnn *vnn;
406 struct ctdb_do_takeip_state {
407 struct ctdb_req_control *c;
408 struct ctdb_vnn *vnn;
412 called when takeip event finishes
414 static void ctdb_do_takeip_callback(struct ctdb_context *ctdb, int status,
417 struct ctdb_do_takeip_state *state =
418 talloc_get_type(private_data, struct ctdb_do_takeip_state);
423 struct ctdb_node *node = ctdb->nodes[ctdb->pnn];
425 if (status == -ETIME) {
428 DEBUG(DEBUG_ERR,(__location__ " Failed to takeover IP %s on interface %s\n",
429 ctdb_addr_to_str(&state->vnn->public_address),
430 ctdb_vnn_iface_string(state->vnn)));
431 ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
433 node->flags |= NODE_FLAGS_UNHEALTHY;
438 if (ctdb->do_checkpublicip) {
440 ret = ctdb_announce_vnn_iface(ctdb, state->vnn);
442 ctdb_request_control_reply(ctdb, state->c, NULL, -1, NULL);
449 data.dptr = (uint8_t *)ctdb_addr_to_str(&state->vnn->public_address);
450 data.dsize = strlen((char *)data.dptr) + 1;
451 DEBUG(DEBUG_INFO,(__location__ " sending TAKE_IP for '%s'\n", data.dptr));
453 ctdb_daemon_send_message(ctdb, ctdb->pnn, CTDB_SRVID_TAKE_IP, data);
456 /* the control succeeded */
457 ctdb_request_control_reply(ctdb, state->c, NULL, 0, NULL);
462 static int ctdb_takeip_destructor(struct ctdb_do_takeip_state *state)
464 state->vnn->update_in_flight = false;
469 take over an ip address
471 static int32_t ctdb_do_takeip(struct ctdb_context *ctdb,
472 struct ctdb_req_control *c,
473 struct ctdb_vnn *vnn)
476 struct ctdb_do_takeip_state *state;
478 if (vnn->update_in_flight) {
479 DEBUG(DEBUG_NOTICE,("Takeover of IP %s/%u rejected "
480 "update for this IP already in flight\n",
481 ctdb_addr_to_str(&vnn->public_address),
482 vnn->public_netmask_bits));
486 ret = ctdb_vnn_assign_iface(ctdb, vnn);
488 DEBUG(DEBUG_ERR,("Takeover of IP %s/%u failed to "
489 "assign a usable interface\n",
490 ctdb_addr_to_str(&vnn->public_address),
491 vnn->public_netmask_bits));
495 state = talloc(vnn, struct ctdb_do_takeip_state);
496 CTDB_NO_MEMORY(ctdb, state);
498 state->c = talloc_steal(ctdb, c);
501 vnn->update_in_flight = true;
502 talloc_set_destructor(state, ctdb_takeip_destructor);
504 DEBUG(DEBUG_NOTICE,("Takeover of IP %s/%u on interface %s\n",
505 ctdb_addr_to_str(&vnn->public_address),
506 vnn->public_netmask_bits,
507 ctdb_vnn_iface_string(vnn)));
509 ret = ctdb_event_script_callback(ctdb,
511 ctdb_do_takeip_callback,
516 ctdb_vnn_iface_string(vnn),
517 ctdb_addr_to_str(&vnn->public_address),
518 vnn->public_netmask_bits);
521 DEBUG(DEBUG_ERR,(__location__ " Failed to takeover IP %s on interface %s\n",
522 ctdb_addr_to_str(&vnn->public_address),
523 ctdb_vnn_iface_string(vnn)));
531 struct ctdb_do_updateip_state {
532 struct ctdb_req_control *c;
533 struct ctdb_iface *old;
534 struct ctdb_vnn *vnn;
538 called when updateip event finishes
540 static void ctdb_do_updateip_callback(struct ctdb_context *ctdb, int status,
543 struct ctdb_do_updateip_state *state =
544 talloc_get_type(private_data, struct ctdb_do_updateip_state);
548 if (status == -ETIME) {
551 DEBUG(DEBUG_ERR,(__location__ " Failed to move IP %s from interface %s to %s\n",
552 ctdb_addr_to_str(&state->vnn->public_address),
554 ctdb_vnn_iface_string(state->vnn)));
557 * All we can do is reset the old interface
558 * and let the next run fix it
560 ctdb_vnn_unassign_iface(ctdb, state->vnn);
561 state->vnn->iface = state->old;
562 state->vnn->iface->references++;
564 ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
569 if (ctdb->do_checkpublicip) {
571 ret = ctdb_announce_vnn_iface(ctdb, state->vnn);
573 ctdb_request_control_reply(ctdb, state->c, NULL, -1, NULL);
580 /* the control succeeded */
581 ctdb_request_control_reply(ctdb, state->c, NULL, 0, NULL);
586 static int ctdb_updateip_destructor(struct ctdb_do_updateip_state *state)
588 state->vnn->update_in_flight = false;
593 update (move) an ip address
595 static int32_t ctdb_do_updateip(struct ctdb_context *ctdb,
596 struct ctdb_req_control *c,
597 struct ctdb_vnn *vnn)
600 struct ctdb_do_updateip_state *state;
601 struct ctdb_iface *old = vnn->iface;
602 const char *new_name;
604 if (vnn->update_in_flight) {
605 DEBUG(DEBUG_NOTICE,("Update of IP %s/%u rejected "
606 "update for this IP already in flight\n",
607 ctdb_addr_to_str(&vnn->public_address),
608 vnn->public_netmask_bits));
612 ctdb_vnn_unassign_iface(ctdb, vnn);
613 ret = ctdb_vnn_assign_iface(ctdb, vnn);
615 DEBUG(DEBUG_ERR,("update of IP %s/%u failed to "
616 "assin a usable interface (old iface '%s')\n",
617 ctdb_addr_to_str(&vnn->public_address),
618 vnn->public_netmask_bits,
623 new_name = ctdb_vnn_iface_string(vnn);
624 if (old->name != NULL && new_name != NULL && !strcmp(old->name, new_name)) {
625 /* A benign update from one interface onto itself.
626 * no need to run the eventscripts in this case, just return
629 ctdb_request_control_reply(ctdb, c, NULL, 0, NULL);
633 state = talloc(vnn, struct ctdb_do_updateip_state);
634 CTDB_NO_MEMORY(ctdb, state);
636 state->c = talloc_steal(ctdb, c);
640 vnn->update_in_flight = true;
641 talloc_set_destructor(state, ctdb_updateip_destructor);
643 DEBUG(DEBUG_NOTICE,("Update of IP %s/%u from "
644 "interface %s to %s\n",
645 ctdb_addr_to_str(&vnn->public_address),
646 vnn->public_netmask_bits,
650 ret = ctdb_event_script_callback(ctdb,
652 ctdb_do_updateip_callback,
655 CTDB_EVENT_UPDATE_IP,
659 ctdb_addr_to_str(&vnn->public_address),
660 vnn->public_netmask_bits);
662 DEBUG(DEBUG_ERR,(__location__ " Failed update IP %s from interface %s to %s\n",
663 ctdb_addr_to_str(&vnn->public_address),
664 old->name, new_name));
673 Find the vnn of the node that has a public ip address
674 returns -1 if the address is not known as a public address
676 static struct ctdb_vnn *find_public_ip_vnn(struct ctdb_context *ctdb, ctdb_sock_addr *addr)
678 struct ctdb_vnn *vnn;
680 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
681 if (ctdb_same_ip(&vnn->public_address, addr)) {
690 take over an ip address
692 int32_t ctdb_control_takeover_ip(struct ctdb_context *ctdb,
693 struct ctdb_req_control *c,
698 struct ctdb_public_ip *pip = (struct ctdb_public_ip *)indata.dptr;
699 struct ctdb_vnn *vnn;
700 bool have_ip = false;
701 bool do_updateip = false;
702 bool do_takeip = false;
703 struct ctdb_iface *best_iface = NULL;
705 if (pip->pnn != ctdb->pnn) {
706 DEBUG(DEBUG_ERR,(__location__" takeoverip called for an ip '%s' "
707 "with pnn %d, but we're node %d\n",
708 ctdb_addr_to_str(&pip->addr),
709 pip->pnn, ctdb->pnn));
713 /* update out vnn list */
714 vnn = find_public_ip_vnn(ctdb, &pip->addr);
716 DEBUG(DEBUG_INFO,("takeoverip called for an ip '%s' that is not a public address\n",
717 ctdb_addr_to_str(&pip->addr)));
721 if (ctdb->do_checkpublicip) {
722 have_ip = ctdb_sys_have_ip(&pip->addr);
724 best_iface = ctdb_vnn_best_iface(ctdb, vnn);
725 if (best_iface == NULL) {
726 DEBUG(DEBUG_ERR,("takeoverip of IP %s/%u failed to find"
727 "a usable interface (old %s, have_ip %d)\n",
728 ctdb_addr_to_str(&vnn->public_address),
729 vnn->public_netmask_bits,
730 ctdb_vnn_iface_string(vnn),
735 if (vnn->iface == NULL && vnn->pnn == -1 && have_ip && best_iface != NULL) {
736 DEBUG(DEBUG_ERR,("Taking over newly created ip\n"));
741 if (vnn->iface == NULL && have_ip) {
742 DEBUG(DEBUG_CRIT,(__location__ " takeoverip of IP %s is known to the kernel, "
743 "but we have no interface assigned, has someone manually configured it? Ignore for now.\n",
744 ctdb_addr_to_str(&vnn->public_address)));
748 if (vnn->pnn != ctdb->pnn && have_ip && vnn->pnn != -1) {
749 DEBUG(DEBUG_CRIT,(__location__ " takeoverip of IP %s is known to the kernel, "
750 "and we have it on iface[%s], but it was assigned to node %d"
751 "and we are node %d, banning ourself\n",
752 ctdb_addr_to_str(&vnn->public_address),
753 ctdb_vnn_iface_string(vnn), vnn->pnn, ctdb->pnn));
758 if (vnn->pnn == -1 && have_ip) {
759 vnn->pnn = ctdb->pnn;
760 DEBUG(DEBUG_CRIT,(__location__ " takeoverip of IP %s is known to the kernel, "
761 "and we already have it on iface[%s], update local daemon\n",
762 ctdb_addr_to_str(&vnn->public_address),
763 ctdb_vnn_iface_string(vnn)));
768 if (vnn->iface != best_iface) {
769 if (!vnn->iface->link_up) {
771 } else if (vnn->iface->references > (best_iface->references + 1)) {
772 /* only move when the rebalance gains something */
780 ctdb_vnn_unassign_iface(ctdb, vnn);
787 ret = ctdb_do_takeip(ctdb, c, vnn);
791 } else if (do_updateip) {
792 ret = ctdb_do_updateip(ctdb, c, vnn);
798 * The interface is up and the kernel known the ip
801 DEBUG(DEBUG_INFO,("Redundant takeover of IP %s/%u on interface %s (ip already held)\n",
802 ctdb_addr_to_str(&pip->addr),
803 vnn->public_netmask_bits,
804 ctdb_vnn_iface_string(vnn)));
808 /* tell ctdb_control.c that we will be replying asynchronously */
815 takeover an ip address old v4 style
817 int32_t ctdb_control_takeover_ipv4(struct ctdb_context *ctdb,
818 struct ctdb_req_control *c,
824 data.dsize = sizeof(struct ctdb_public_ip);
825 data.dptr = (uint8_t *)talloc_zero(c, struct ctdb_public_ip);
826 CTDB_NO_MEMORY(ctdb, data.dptr);
828 memcpy(data.dptr, indata.dptr, indata.dsize);
829 return ctdb_control_takeover_ip(ctdb, c, data, async_reply);
833 kill any clients that are registered with a IP that is being released
835 static void release_kill_clients(struct ctdb_context *ctdb, ctdb_sock_addr *addr)
837 struct ctdb_client_ip *ip;
839 DEBUG(DEBUG_INFO,("release_kill_clients for ip %s\n",
840 ctdb_addr_to_str(addr)));
842 for (ip=ctdb->client_ip_list; ip; ip=ip->next) {
843 ctdb_sock_addr tmp_addr;
846 DEBUG(DEBUG_INFO,("checking for client %u with IP %s\n",
848 ctdb_addr_to_str(&ip->addr)));
850 if (ctdb_same_ip(&tmp_addr, addr)) {
851 struct ctdb_client *client = ctdb_reqid_find(ctdb,
854 DEBUG(DEBUG_INFO,("matched client %u with IP %s and pid %u\n",
856 ctdb_addr_to_str(&ip->addr),
859 if (client->pid != 0) {
860 DEBUG(DEBUG_INFO,(__location__ " Killing client pid %u for IP %s on client_id %u\n",
861 (unsigned)client->pid,
862 ctdb_addr_to_str(addr),
864 kill(client->pid, SIGKILL);
871 called when releaseip event finishes
873 static void release_ip_callback(struct ctdb_context *ctdb, int status,
876 struct takeover_callback_state *state =
877 talloc_get_type(private_data, struct takeover_callback_state);
880 if (status == -ETIME) {
884 if (ctdb->do_checkpublicip && ctdb_sys_have_ip(state->addr)) {
885 DEBUG(DEBUG_ERR, ("IP %s still hosted during release IP callback, failing\n",
886 ctdb_addr_to_str(state->addr)));
887 ctdb_request_control_reply(ctdb, state->c, NULL, -1, NULL);
892 /* send a message to all clients of this node telling them
893 that the cluster has been reconfigured and they should
894 release any sockets on this IP */
895 data.dptr = (uint8_t *)talloc_strdup(state, ctdb_addr_to_str(state->addr));
896 CTDB_NO_MEMORY_VOID(ctdb, data.dptr);
897 data.dsize = strlen((char *)data.dptr)+1;
899 DEBUG(DEBUG_INFO,(__location__ " sending RELEASE_IP for '%s'\n", data.dptr));
901 ctdb_daemon_send_message(ctdb, ctdb->pnn, CTDB_SRVID_RELEASE_IP, data);
903 /* kill clients that have registered with this IP */
904 release_kill_clients(ctdb, state->addr);
906 ctdb_vnn_unassign_iface(ctdb, state->vnn);
908 /* the control succeeded */
909 ctdb_request_control_reply(ctdb, state->c, NULL, 0, NULL);
913 static int ctdb_releaseip_destructor(struct takeover_callback_state *state)
915 state->vnn->update_in_flight = false;
920 release an ip address
922 int32_t ctdb_control_release_ip(struct ctdb_context *ctdb,
923 struct ctdb_req_control *c,
928 struct takeover_callback_state *state;
929 struct ctdb_public_ip *pip = (struct ctdb_public_ip *)indata.dptr;
930 struct ctdb_vnn *vnn;
933 /* update our vnn list */
934 vnn = find_public_ip_vnn(ctdb, &pip->addr);
936 DEBUG(DEBUG_INFO,("releaseip called for an ip '%s' that is not a public address\n",
937 ctdb_addr_to_str(&pip->addr)));
942 /* stop any previous arps */
943 talloc_free(vnn->takeover_ctx);
944 vnn->takeover_ctx = NULL;
946 /* Some ctdb tool commands (e.g. moveip, rebalanceip) send
947 * lazy multicast to drop an IP from any node that isn't the
948 * intended new node. The following causes makes ctdbd ignore
949 * a release for any address it doesn't host.
951 if (ctdb->do_checkpublicip) {
952 if (!ctdb_sys_have_ip(&pip->addr)) {
953 DEBUG(DEBUG_DEBUG,("Redundant release of IP %s/%u on interface %s (ip not held)\n",
954 ctdb_addr_to_str(&pip->addr),
955 vnn->public_netmask_bits,
956 ctdb_vnn_iface_string(vnn)));
957 ctdb_vnn_unassign_iface(ctdb, vnn);
961 if (vnn->iface == NULL) {
962 DEBUG(DEBUG_DEBUG,("Redundant release of IP %s/%u (ip not held)\n",
963 ctdb_addr_to_str(&pip->addr),
964 vnn->public_netmask_bits));
969 /* There is a potential race between take_ip and us because we
970 * update the VNN via a callback that run when the
971 * eventscripts have been run. Avoid the race by allowing one
972 * update to be in flight at a time.
974 if (vnn->update_in_flight) {
975 DEBUG(DEBUG_NOTICE,("Release of IP %s/%u rejected "
976 "update for this IP already in flight\n",
977 ctdb_addr_to_str(&vnn->public_address),
978 vnn->public_netmask_bits));
982 if (ctdb->do_checkpublicip) {
983 iface = ctdb_sys_find_ifname(&pip->addr);
985 DEBUG(DEBUG_ERR, ("Could not find which interface the ip address is hosted on. can not release it\n"));
988 if (vnn->iface == NULL) {
990 ("Public IP %s is hosted on interface %s but we have no VNN\n",
991 ctdb_addr_to_str(&pip->addr),
993 } else if (strcmp(iface, ctdb_vnn_iface_string(vnn)) != 0) {
995 ("Public IP %s is hosted on inteterface %s but VNN says %s\n",
996 ctdb_addr_to_str(&pip->addr),
998 ctdb_vnn_iface_string(vnn)));
999 /* Should we fix vnn->iface? If we do, what
1000 * happens to reference counts?
1004 iface = strdup(ctdb_vnn_iface_string(vnn));
1007 DEBUG(DEBUG_NOTICE,("Release of IP %s/%u on interface %s node:%d\n",
1008 ctdb_addr_to_str(&pip->addr),
1009 vnn->public_netmask_bits,
1013 state = talloc(ctdb, struct takeover_callback_state);
1014 CTDB_NO_MEMORY(ctdb, state);
1016 state->c = talloc_steal(state, c);
1017 state->addr = talloc(state, ctdb_sock_addr);
1018 CTDB_NO_MEMORY(ctdb, state->addr);
1019 *state->addr = pip->addr;
1022 vnn->update_in_flight = true;
1023 talloc_set_destructor(state, ctdb_releaseip_destructor);
1025 ret = ctdb_event_script_callback(ctdb,
1026 state, release_ip_callback, state,
1028 CTDB_EVENT_RELEASE_IP,
1031 ctdb_addr_to_str(&pip->addr),
1032 vnn->public_netmask_bits);
1035 DEBUG(DEBUG_ERR,(__location__ " Failed to release IP %s on interface %s\n",
1036 ctdb_addr_to_str(&pip->addr),
1037 ctdb_vnn_iface_string(vnn)));
1042 /* tell the control that we will be reply asynchronously */
1043 *async_reply = true;
1048 release an ip address old v4 style
1050 int32_t ctdb_control_release_ipv4(struct ctdb_context *ctdb,
1051 struct ctdb_req_control *c,
1057 data.dsize = sizeof(struct ctdb_public_ip);
1058 data.dptr = (uint8_t *)talloc_zero(c, struct ctdb_public_ip);
1059 CTDB_NO_MEMORY(ctdb, data.dptr);
1061 memcpy(data.dptr, indata.dptr, indata.dsize);
1062 return ctdb_control_release_ip(ctdb, c, data, async_reply);
1066 static int ctdb_add_public_address(struct ctdb_context *ctdb,
1067 ctdb_sock_addr *addr,
1068 unsigned mask, const char *ifaces,
1071 struct ctdb_vnn *vnn;
1078 tmp = strdup(ifaces);
1079 for (iface = strtok(tmp, ","); iface; iface = strtok(NULL, ",")) {
1080 if (!ctdb_sys_check_iface_exists(iface)) {
1081 DEBUG(DEBUG_CRIT,("Interface %s does not exist. Can not add public-address : %s\n", iface, ctdb_addr_to_str(addr)));
1088 /* Verify that we dont have an entry for this ip yet */
1089 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
1090 if (ctdb_same_sockaddr(addr, &vnn->public_address)) {
1091 DEBUG(DEBUG_CRIT,("Same ip '%s' specified multiple times in the public address list \n",
1092 ctdb_addr_to_str(addr)));
1097 /* create a new vnn structure for this ip address */
1098 vnn = talloc_zero(ctdb, struct ctdb_vnn);
1099 CTDB_NO_MEMORY_FATAL(ctdb, vnn);
1100 vnn->ifaces = talloc_array(vnn, const char *, num + 2);
1101 tmp = talloc_strdup(vnn, ifaces);
1102 CTDB_NO_MEMORY_FATAL(ctdb, tmp);
1103 for (iface = strtok(tmp, ","); iface; iface = strtok(NULL, ",")) {
1104 vnn->ifaces = talloc_realloc(vnn, vnn->ifaces, const char *, num + 2);
1105 CTDB_NO_MEMORY_FATAL(ctdb, vnn->ifaces);
1106 vnn->ifaces[num] = talloc_strdup(vnn, iface);
1107 CTDB_NO_MEMORY_FATAL(ctdb, vnn->ifaces[num]);
1111 vnn->ifaces[num] = NULL;
1112 vnn->public_address = *addr;
1113 vnn->public_netmask_bits = mask;
1115 if (check_address) {
1116 if (ctdb_sys_have_ip(addr)) {
1117 DEBUG(DEBUG_ERR,("We are already hosting public address '%s'. setting PNN to ourself:%d\n", ctdb_addr_to_str(addr), ctdb->pnn));
1118 vnn->pnn = ctdb->pnn;
1122 for (i=0; vnn->ifaces[i]; i++) {
1123 ret = ctdb_add_local_iface(ctdb, vnn->ifaces[i]);
1125 DEBUG(DEBUG_CRIT, (__location__ " failed to add iface[%s] "
1126 "for public_address[%s]\n",
1127 vnn->ifaces[i], ctdb_addr_to_str(addr)));
1133 DLIST_ADD(ctdb->vnn, vnn);
1139 setup the event script directory
1141 int ctdb_set_event_script_dir(struct ctdb_context *ctdb, const char *script_dir)
1143 ctdb->event_script_dir = talloc_strdup(ctdb, script_dir);
1144 CTDB_NO_MEMORY(ctdb, ctdb->event_script_dir);
1148 static void ctdb_check_interfaces_event(struct event_context *ev, struct timed_event *te,
1149 struct timeval t, void *private_data)
1151 struct ctdb_context *ctdb = talloc_get_type(private_data,
1152 struct ctdb_context);
1153 struct ctdb_vnn *vnn;
1155 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
1158 for (i=0; vnn->ifaces[i] != NULL; i++) {
1159 if (!ctdb_sys_check_iface_exists(vnn->ifaces[i])) {
1160 DEBUG(DEBUG_CRIT,("Interface %s does not exist but is used by public ip %s\n",
1162 ctdb_addr_to_str(&vnn->public_address)));
1167 event_add_timed(ctdb->ev, ctdb->check_public_ifaces_ctx,
1168 timeval_current_ofs(30, 0),
1169 ctdb_check_interfaces_event, ctdb);
1173 int ctdb_start_monitoring_interfaces(struct ctdb_context *ctdb)
1175 if (ctdb->check_public_ifaces_ctx != NULL) {
1176 talloc_free(ctdb->check_public_ifaces_ctx);
1177 ctdb->check_public_ifaces_ctx = NULL;
1180 ctdb->check_public_ifaces_ctx = talloc_new(ctdb);
1181 if (ctdb->check_public_ifaces_ctx == NULL) {
1182 ctdb_fatal(ctdb, "failed to allocate context for checking interfaces");
1185 event_add_timed(ctdb->ev, ctdb->check_public_ifaces_ctx,
1186 timeval_current_ofs(30, 0),
1187 ctdb_check_interfaces_event, ctdb);
1194 setup the public address lists from a file
1196 int ctdb_set_public_addresses(struct ctdb_context *ctdb, bool check_addresses)
1202 lines = file_lines_load(ctdb->public_addresses_file, &nlines, ctdb);
1203 if (lines == NULL) {
1204 ctdb_set_error(ctdb, "Failed to load public address list '%s'\n", ctdb->public_addresses_file);
1207 while (nlines > 0 && strcmp(lines[nlines-1], "") == 0) {
1211 for (i=0;i<nlines;i++) {
1213 ctdb_sock_addr addr;
1214 const char *addrstr;
1219 while ((*line == ' ') || (*line == '\t')) {
1225 if (strcmp(line, "") == 0) {
1228 tok = strtok(line, " \t");
1230 tok = strtok(NULL, " \t");
1232 if (NULL == ctdb->default_public_interface) {
1233 DEBUG(DEBUG_CRIT,("No default public interface and no interface specified at line %u of public address list\n",
1238 ifaces = ctdb->default_public_interface;
1243 if (!addrstr || !parse_ip_mask(addrstr, ifaces, &addr, &mask)) {
1244 DEBUG(DEBUG_CRIT,("Badly formed line %u in public address list\n", i+1));
1248 if (ctdb_add_public_address(ctdb, &addr, mask, ifaces, check_addresses)) {
1249 DEBUG(DEBUG_CRIT,("Failed to add line %u to the public address list\n", i+1));
1260 int ctdb_set_single_public_ip(struct ctdb_context *ctdb,
1264 struct ctdb_vnn *svnn;
1265 struct ctdb_iface *cur = NULL;
1269 svnn = talloc_zero(ctdb, struct ctdb_vnn);
1270 CTDB_NO_MEMORY(ctdb, svnn);
1272 svnn->ifaces = talloc_array(svnn, const char *, 2);
1273 CTDB_NO_MEMORY(ctdb, svnn->ifaces);
1274 svnn->ifaces[0] = talloc_strdup(svnn->ifaces, iface);
1275 CTDB_NO_MEMORY(ctdb, svnn->ifaces[0]);
1276 svnn->ifaces[1] = NULL;
1278 ok = parse_ip(ip, iface, 0, &svnn->public_address);
1284 ret = ctdb_add_local_iface(ctdb, svnn->ifaces[0]);
1286 DEBUG(DEBUG_CRIT, (__location__ " failed to add iface[%s] "
1287 "for single_ip[%s]\n",
1289 ctdb_addr_to_str(&svnn->public_address)));
1294 /* assume the single public ip interface is initially "good" */
1295 cur = ctdb_find_iface(ctdb, iface);
1297 DEBUG(DEBUG_CRIT,("Can not find public interface %s used by --single-public-ip", iface));
1300 cur->link_up = true;
1302 ret = ctdb_vnn_assign_iface(ctdb, svnn);
1308 ctdb->single_ip_vnn = svnn;
1312 struct ctdb_public_ip_list {
1313 struct ctdb_public_ip_list *next;
1315 ctdb_sock_addr addr;
1318 /* Given a physical node, return the number of
1319 public addresses that is currently assigned to this node.
1321 static int node_ip_coverage(struct ctdb_context *ctdb,
1323 struct ctdb_public_ip_list *ips)
1327 for (;ips;ips=ips->next) {
1328 if (ips->pnn == pnn) {
1336 /* Can the given node host the given IP: is the public IP known to the
1337 * node and is NOIPHOST unset?
1339 static bool can_node_host_ip(struct ctdb_context *ctdb, int32_t pnn,
1340 struct ctdb_ipflags ipflags,
1341 struct ctdb_public_ip_list *ip)
1343 struct ctdb_all_public_ips *public_ips;
1346 if (ipflags.noiphost) {
1350 public_ips = ctdb->nodes[pnn]->available_public_ips;
1352 if (public_ips == NULL) {
1356 for (i=0; i<public_ips->num; i++) {
1357 if (ctdb_same_ip(&ip->addr, &public_ips->ips[i].addr)) {
1358 /* yes, this node can serve this public ip */
1366 static bool can_node_takeover_ip(struct ctdb_context *ctdb, int32_t pnn,
1367 struct ctdb_ipflags ipflags,
1368 struct ctdb_public_ip_list *ip)
1370 if (ipflags.noiptakeover) {
1374 return can_node_host_ip(ctdb, pnn, ipflags, ip);
1377 /* search the node lists list for a node to takeover this ip.
1378 pick the node that currently are serving the least number of ips
1379 so that the ips get spread out evenly.
1381 static int find_takeover_node(struct ctdb_context *ctdb,
1382 struct ctdb_ipflags *ipflags,
1383 struct ctdb_public_ip_list *ip,
1384 struct ctdb_public_ip_list *all_ips)
1386 int pnn, min=0, num;
1389 numnodes = talloc_array_length(ipflags);
1391 for (i=0; i<numnodes; i++) {
1392 /* verify that this node can serve this ip */
1393 if (!can_node_takeover_ip(ctdb, i, ipflags[i], ip)) {
1394 /* no it couldnt so skip to the next node */
1398 num = node_ip_coverage(ctdb, i, all_ips);
1399 /* was this the first node we checked ? */
1411 DEBUG(DEBUG_WARNING,(__location__ " Could not find node to take over public address '%s'\n",
1412 ctdb_addr_to_str(&ip->addr)));
1422 static uint32_t *ip_key(ctdb_sock_addr *ip)
1424 static uint32_t key[IP_KEYLEN];
1426 bzero(key, sizeof(key));
1428 switch (ip->sa.sa_family) {
1430 key[3] = htonl(ip->ip.sin_addr.s_addr);
1433 uint32_t *s6_a32 = (uint32_t *)&(ip->ip6.sin6_addr.s6_addr);
1434 key[0] = htonl(s6_a32[0]);
1435 key[1] = htonl(s6_a32[1]);
1436 key[2] = htonl(s6_a32[2]);
1437 key[3] = htonl(s6_a32[3]);
1441 DEBUG(DEBUG_ERR, (__location__ " ERROR, unknown family passed :%u\n", ip->sa.sa_family));
1448 static void *add_ip_callback(void *parm, void *data)
1450 struct ctdb_public_ip_list *this_ip = parm;
1451 struct ctdb_public_ip_list *prev_ip = data;
1453 if (prev_ip == NULL) {
1456 if (this_ip->pnn == -1) {
1457 this_ip->pnn = prev_ip->pnn;
1463 static int getips_count_callback(void *param, void *data)
1465 struct ctdb_public_ip_list **ip_list = (struct ctdb_public_ip_list **)param;
1466 struct ctdb_public_ip_list *new_ip = (struct ctdb_public_ip_list *)data;
1468 new_ip->next = *ip_list;
1473 static struct ctdb_public_ip_list *
1474 create_merged_ip_list(struct ctdb_context *ctdb)
1477 struct ctdb_public_ip_list *ip_list;
1478 struct ctdb_all_public_ips *public_ips;
1480 if (ctdb->ip_tree != NULL) {
1481 talloc_free(ctdb->ip_tree);
1482 ctdb->ip_tree = NULL;
1484 ctdb->ip_tree = trbt_create(ctdb, 0);
1486 for (i=0;i<ctdb->num_nodes;i++) {
1487 public_ips = ctdb->nodes[i]->known_public_ips;
1489 if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
1493 /* there were no public ips for this node */
1494 if (public_ips == NULL) {
1498 for (j=0;j<public_ips->num;j++) {
1499 struct ctdb_public_ip_list *tmp_ip;
1501 tmp_ip = talloc_zero(ctdb->ip_tree, struct ctdb_public_ip_list);
1502 CTDB_NO_MEMORY_NULL(ctdb, tmp_ip);
1503 /* Do not use information about IP addresses hosted
1504 * on other nodes, it may not be accurate */
1505 if (public_ips->ips[j].pnn == ctdb->nodes[i]->pnn) {
1506 tmp_ip->pnn = public_ips->ips[j].pnn;
1510 tmp_ip->addr = public_ips->ips[j].addr;
1511 tmp_ip->next = NULL;
1513 trbt_insertarray32_callback(ctdb->ip_tree,
1514 IP_KEYLEN, ip_key(&public_ips->ips[j].addr),
1521 trbt_traversearray32(ctdb->ip_tree, IP_KEYLEN, getips_count_callback, &ip_list);
1527 * This is the length of the longtest common prefix between the IPs.
1528 * It is calculated by XOR-ing the 2 IPs together and counting the
1529 * number of leading zeroes. The implementation means that all
1530 * addresses end up being 128 bits long.
1532 * FIXME? Should we consider IPv4 and IPv6 separately given that the
1533 * 12 bytes of 0 prefix padding will hurt the algorithm if there are
1534 * lots of nodes and IP addresses?
1536 static uint32_t ip_distance(ctdb_sock_addr *ip1, ctdb_sock_addr *ip2)
1538 uint32_t ip1_k[IP_KEYLEN];
1543 uint32_t distance = 0;
1545 memcpy(ip1_k, ip_key(ip1), sizeof(ip1_k));
1547 for (i=0; i<IP_KEYLEN; i++) {
1548 x = ip1_k[i] ^ t[i];
1552 /* Count number of leading zeroes.
1553 * FIXME? This could be optimised...
1555 while ((x & (1 << 31)) == 0) {
1565 /* Calculate the IP distance for the given IP relative to IPs on the
1566 given node. The ips argument is generally the all_ips variable
1567 used in the main part of the algorithm.
1569 static uint32_t ip_distance_2_sum(ctdb_sock_addr *ip,
1570 struct ctdb_public_ip_list *ips,
1573 struct ctdb_public_ip_list *t;
1578 for (t=ips; t != NULL; t=t->next) {
1579 if (t->pnn != pnn) {
1583 /* Optimisation: We never calculate the distance
1584 * between an address and itself. This allows us to
1585 * calculate the effect of removing an address from a
1586 * node by simply calculating the distance between
1587 * that address and all of the exitsing addresses.
1588 * Moreover, we assume that we're only ever dealing
1589 * with addresses from all_ips so we can identify an
1590 * address via a pointer rather than doing a more
1591 * expensive address comparison. */
1592 if (&(t->addr) == ip) {
1596 d = ip_distance(ip, &(t->addr));
1597 sum += d * d; /* Cheaper than pulling in math.h :-) */
1603 /* Return the LCP2 imbalance metric for addresses currently assigned
1606 static uint32_t lcp2_imbalance(struct ctdb_public_ip_list * all_ips, int pnn)
1608 struct ctdb_public_ip_list *t;
1610 uint32_t imbalance = 0;
1612 for (t=all_ips; t!=NULL; t=t->next) {
1613 if (t->pnn != pnn) {
1616 /* Pass the rest of the IPs rather than the whole
1619 imbalance += ip_distance_2_sum(&(t->addr), t->next, pnn);
1625 /* Allocate any unassigned IPs just by looping through the IPs and
1626 * finding the best node for each.
1628 static void basic_allocate_unassigned(struct ctdb_context *ctdb,
1629 struct ctdb_ipflags *ipflags,
1630 struct ctdb_public_ip_list *all_ips)
1632 struct ctdb_public_ip_list *tmp_ip;
1634 /* loop over all ip's and find a physical node to cover for
1637 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1638 if (tmp_ip->pnn == -1) {
1639 if (find_takeover_node(ctdb, ipflags, tmp_ip, all_ips)) {
1640 DEBUG(DEBUG_WARNING,("Failed to find node to cover ip %s\n",
1641 ctdb_addr_to_str(&tmp_ip->addr)));
1647 /* Basic non-deterministic rebalancing algorithm.
1649 static void basic_failback(struct ctdb_context *ctdb,
1650 struct ctdb_ipflags *ipflags,
1651 struct ctdb_public_ip_list *all_ips,
1655 int maxnode, maxnum, minnode, minnum, num, retries;
1656 struct ctdb_public_ip_list *tmp_ip;
1658 numnodes = talloc_array_length(ipflags);
1665 /* for each ip address, loop over all nodes that can serve
1666 this ip and make sure that the difference between the node
1667 serving the most and the node serving the least ip's are
1670 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1671 if (tmp_ip->pnn == -1) {
1675 /* Get the highest and lowest number of ips's served by any
1676 valid node which can serve this ip.
1680 for (i=0; i<numnodes; i++) {
1681 /* only check nodes that can actually serve this ip */
1682 if (!can_node_takeover_ip(ctdb, i, ipflags[i], tmp_ip)) {
1683 /* no it couldnt so skip to the next node */
1687 num = node_ip_coverage(ctdb, i, all_ips);
1688 if (maxnode == -1) {
1697 if (minnode == -1) {
1707 if (maxnode == -1) {
1708 DEBUG(DEBUG_WARNING,(__location__ " Could not find maxnode. May not be able to serve ip '%s'\n",
1709 ctdb_addr_to_str(&tmp_ip->addr)));
1714 /* if the spread between the smallest and largest coverage by
1715 a node is >=2 we steal one of the ips from the node with
1716 most coverage to even things out a bit.
1717 try to do this a limited number of times since we dont
1718 want to spend too much time balancing the ip coverage.
1720 if ( (maxnum > minnum+1)
1721 && (retries < (num_ips + 5)) ){
1722 struct ctdb_public_ip_list *tmp;
1724 /* Reassign one of maxnode's VNNs */
1725 for (tmp=all_ips;tmp;tmp=tmp->next) {
1726 if (tmp->pnn == maxnode) {
1727 (void)find_takeover_node(ctdb, ipflags, tmp, all_ips);
1736 static void lcp2_init(struct ctdb_context *tmp_ctx,
1737 struct ctdb_ipflags *ipflags,
1738 struct ctdb_public_ip_list *all_ips,
1739 uint32_t *force_rebalance_nodes,
1740 uint32_t **lcp2_imbalances,
1741 bool **rebalance_candidates)
1744 struct ctdb_public_ip_list *tmp_ip;
1746 numnodes = talloc_array_length(ipflags);
1748 *rebalance_candidates = talloc_array(tmp_ctx, bool, numnodes);
1749 CTDB_NO_MEMORY_FATAL(tmp_ctx, *rebalance_candidates);
1750 *lcp2_imbalances = talloc_array(tmp_ctx, uint32_t, numnodes);
1751 CTDB_NO_MEMORY_FATAL(tmp_ctx, *lcp2_imbalances);
1753 for (i=0; i<numnodes; i++) {
1754 (*lcp2_imbalances)[i] = lcp2_imbalance(all_ips, i);
1755 /* First step: assume all nodes are candidates */
1756 (*rebalance_candidates)[i] = true;
1759 /* 2nd step: if a node has IPs assigned then it must have been
1760 * healthy before, so we remove it from consideration. This
1761 * is overkill but is all we have because we don't maintain
1762 * state between takeover runs. An alternative would be to
1763 * keep state and invalidate it every time the recovery master
1766 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1767 if (tmp_ip->pnn != -1) {
1768 (*rebalance_candidates)[tmp_ip->pnn] = false;
1772 /* 3rd step: if a node is forced to re-balance then
1773 we allow failback onto the node */
1774 if (force_rebalance_nodes == NULL) {
1777 for (i = 0; i < talloc_array_length(force_rebalance_nodes); i++) {
1778 uint32_t pnn = force_rebalance_nodes[i];
1779 if (pnn >= numnodes) {
1781 (__location__ "unknown node %u\n", pnn));
1786 ("Forcing rebalancing of IPs to node %u\n", pnn));
1787 (*rebalance_candidates)[pnn] = true;
1791 /* Allocate any unassigned addresses using the LCP2 algorithm to find
1792 * the IP/node combination that will cost the least.
1794 static void lcp2_allocate_unassigned(struct ctdb_context *ctdb,
1795 struct ctdb_ipflags *ipflags,
1796 struct ctdb_public_ip_list *all_ips,
1797 uint32_t *lcp2_imbalances)
1799 struct ctdb_public_ip_list *tmp_ip;
1800 int dstnode, numnodes;
1803 uint32_t mindsum, dstdsum, dstimbl, minimbl;
1804 struct ctdb_public_ip_list *minip;
1806 bool should_loop = true;
1807 bool have_unassigned = true;
1809 numnodes = talloc_array_length(ipflags);
1811 while (have_unassigned && should_loop) {
1812 should_loop = false;
1814 DEBUG(DEBUG_DEBUG,(" ----------------------------------------\n"));
1815 DEBUG(DEBUG_DEBUG,(" CONSIDERING MOVES (UNASSIGNED)\n"));
1821 /* loop over each unassigned ip. */
1822 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1823 if (tmp_ip->pnn != -1) {
1827 for (dstnode=0; dstnode<numnodes; dstnode++) {
1828 /* only check nodes that can actually takeover this ip */
1829 if (!can_node_takeover_ip(ctdb, dstnode,
1832 /* no it couldnt so skip to the next node */
1836 dstdsum = ip_distance_2_sum(&(tmp_ip->addr), all_ips, dstnode);
1837 dstimbl = lcp2_imbalances[dstnode] + dstdsum;
1838 DEBUG(DEBUG_DEBUG,(" %s -> %d [+%d]\n",
1839 ctdb_addr_to_str(&(tmp_ip->addr)),
1841 dstimbl - lcp2_imbalances[dstnode]));
1844 if ((minnode == -1) || (dstdsum < mindsum)) {
1854 DEBUG(DEBUG_DEBUG,(" ----------------------------------------\n"));
1856 /* If we found one then assign it to the given node. */
1857 if (minnode != -1) {
1858 minip->pnn = minnode;
1859 lcp2_imbalances[minnode] = minimbl;
1860 DEBUG(DEBUG_INFO,(" %s -> %d [+%d]\n",
1861 ctdb_addr_to_str(&(minip->addr)),
1866 /* There might be a better way but at least this is clear. */
1867 have_unassigned = false;
1868 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1869 if (tmp_ip->pnn == -1) {
1870 have_unassigned = true;
1875 /* We know if we have an unassigned addresses so we might as
1878 if (have_unassigned) {
1879 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1880 if (tmp_ip->pnn == -1) {
1881 DEBUG(DEBUG_WARNING,("Failed to find node to cover ip %s\n",
1882 ctdb_addr_to_str(&tmp_ip->addr)));
1888 /* LCP2 algorithm for rebalancing the cluster. Given a candidate node
1889 * to move IPs from, determines the best IP/destination node
1890 * combination to move from the source node.
1892 static bool lcp2_failback_candidate(struct ctdb_context *ctdb,
1893 struct ctdb_ipflags *ipflags,
1894 struct ctdb_public_ip_list *all_ips,
1897 uint32_t *lcp2_imbalances,
1898 bool *rebalance_candidates)
1900 int dstnode, mindstnode, numnodes;
1901 uint32_t srcimbl, srcdsum, dstimbl, dstdsum;
1902 uint32_t minsrcimbl, mindstimbl;
1903 struct ctdb_public_ip_list *minip;
1904 struct ctdb_public_ip_list *tmp_ip;
1906 /* Find an IP and destination node that best reduces imbalance. */
1913 numnodes = talloc_array_length(ipflags);
1915 DEBUG(DEBUG_DEBUG,(" ----------------------------------------\n"));
1916 DEBUG(DEBUG_DEBUG,(" CONSIDERING MOVES FROM %d [%d]\n", srcnode, candimbl));
1918 for (tmp_ip=all_ips; tmp_ip; tmp_ip=tmp_ip->next) {
1919 /* Only consider addresses on srcnode. */
1920 if (tmp_ip->pnn != srcnode) {
1924 /* What is this IP address costing the source node? */
1925 srcdsum = ip_distance_2_sum(&(tmp_ip->addr), all_ips, srcnode);
1926 srcimbl = candimbl - srcdsum;
1928 /* Consider this IP address would cost each potential
1929 * destination node. Destination nodes are limited to
1930 * those that are newly healthy, since we don't want
1931 * to do gratuitous failover of IPs just to make minor
1932 * balance improvements.
1934 for (dstnode=0; dstnode<numnodes; dstnode++) {
1935 if (!rebalance_candidates[dstnode]) {
1939 /* only check nodes that can actually takeover this ip */
1940 if (!can_node_takeover_ip(ctdb, dstnode,
1941 ipflags[dstnode], tmp_ip)) {
1942 /* no it couldnt so skip to the next node */
1946 dstdsum = ip_distance_2_sum(&(tmp_ip->addr), all_ips, dstnode);
1947 dstimbl = lcp2_imbalances[dstnode] + dstdsum;
1948 DEBUG(DEBUG_DEBUG,(" %d [%d] -> %s -> %d [+%d]\n",
1949 srcnode, srcimbl - lcp2_imbalances[srcnode],
1950 ctdb_addr_to_str(&(tmp_ip->addr)),
1951 dstnode, dstimbl - lcp2_imbalances[dstnode]));
1953 if ((dstimbl < candimbl) && (dstdsum < srcdsum) && \
1954 ((mindstnode == -1) || \
1955 ((srcimbl + dstimbl) < (minsrcimbl + mindstimbl)))) {
1958 minsrcimbl = srcimbl;
1959 mindstnode = dstnode;
1960 mindstimbl = dstimbl;
1964 DEBUG(DEBUG_DEBUG,(" ----------------------------------------\n"));
1966 if (mindstnode != -1) {
1967 /* We found a move that makes things better... */
1968 DEBUG(DEBUG_INFO,("%d [%d] -> %s -> %d [+%d]\n",
1969 srcnode, minsrcimbl - lcp2_imbalances[srcnode],
1970 ctdb_addr_to_str(&(minip->addr)),
1971 mindstnode, mindstimbl - lcp2_imbalances[mindstnode]));
1974 lcp2_imbalances[srcnode] = srcimbl;
1975 lcp2_imbalances[mindstnode] = mindstimbl;
1976 minip->pnn = mindstnode;
1985 struct lcp2_imbalance_pnn {
1990 static int lcp2_cmp_imbalance_pnn(const void * a, const void * b)
1992 const struct lcp2_imbalance_pnn * lipa = (const struct lcp2_imbalance_pnn *) a;
1993 const struct lcp2_imbalance_pnn * lipb = (const struct lcp2_imbalance_pnn *) b;
1995 if (lipa->imbalance > lipb->imbalance) {
1997 } else if (lipa->imbalance == lipb->imbalance) {
2004 /* LCP2 algorithm for rebalancing the cluster. This finds the source
2005 * node with the highest LCP2 imbalance, and then determines the best
2006 * IP/destination node combination to move from the source node.
2008 static void lcp2_failback(struct ctdb_context *ctdb,
2009 struct ctdb_ipflags *ipflags,
2010 struct ctdb_public_ip_list *all_ips,
2011 uint32_t *lcp2_imbalances,
2012 bool *rebalance_candidates)
2014 int i, num_rebalance_candidates, numnodes;
2015 struct lcp2_imbalance_pnn * lips;
2018 numnodes = talloc_array_length(ipflags);
2022 /* It is only worth continuing if we have suitable target
2023 * nodes to transfer IPs to. This check is much cheaper than
2026 num_rebalance_candidates = 0;
2027 for (i=0; i<numnodes; i++) {
2028 if (rebalance_candidates[i]) {
2029 num_rebalance_candidates++;
2032 if (num_rebalance_candidates == 0) {
2036 /* Put the imbalances and nodes into an array, sort them and
2037 * iterate through candidates. Usually the 1st one will be
2038 * used, so this doesn't cost much...
2040 lips = talloc_array(ctdb, struct lcp2_imbalance_pnn, numnodes);
2041 for (i=0; i<numnodes; i++) {
2042 lips[i].imbalance = lcp2_imbalances[i];
2045 qsort(lips, numnodes, sizeof(struct lcp2_imbalance_pnn),
2046 lcp2_cmp_imbalance_pnn);
2049 for (i=0; i<numnodes; i++) {
2050 /* This means that all nodes had 0 or 1 addresses, so
2051 * can't be imbalanced.
2053 if (lips[i].imbalance == 0) {
2057 if (lcp2_failback_candidate(ctdb,
2063 rebalance_candidates)) {
2075 static void unassign_unsuitable_ips(struct ctdb_context *ctdb,
2076 struct ctdb_ipflags *ipflags,
2077 struct ctdb_public_ip_list *all_ips)
2079 struct ctdb_public_ip_list *tmp_ip;
2081 /* verify that the assigned nodes can serve that public ip
2082 and set it to -1 if not
2084 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
2085 if (tmp_ip->pnn == -1) {
2088 if (!can_node_host_ip(ctdb, tmp_ip->pnn,
2089 ipflags[tmp_ip->pnn], tmp_ip) != 0) {
2090 /* this node can not serve this ip. */
2091 DEBUG(DEBUG_DEBUG,("Unassign IP: %s from %d\n",
2092 ctdb_addr_to_str(&(tmp_ip->addr)),
2099 static void ip_alloc_deterministic_ips(struct ctdb_context *ctdb,
2100 struct ctdb_ipflags *ipflags,
2101 struct ctdb_public_ip_list *all_ips)
2103 struct ctdb_public_ip_list *tmp_ip;
2106 numnodes = talloc_array_length(ipflags);
2108 DEBUG(DEBUG_NOTICE,("Deterministic IPs enabled. Resetting all ip allocations\n"));
2109 /* Allocate IPs to nodes in a modulo fashion so that IPs will
2110 * always be allocated the same way for a specific set of
2111 * available/unavailable nodes.
2114 for (i=0,tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next,i++) {
2115 tmp_ip->pnn = i % numnodes;
2118 /* IP failback doesn't make sense with deterministic
2119 * IPs, since the modulo step above implicitly fails
2120 * back IPs to their "home" node.
2122 if (1 == ctdb->tunable.no_ip_failback) {
2123 DEBUG(DEBUG_WARNING, ("WARNING: 'NoIPFailback' set but ignored - incompatible with 'DeterministicIPs\n"));
2126 unassign_unsuitable_ips(ctdb, ipflags, all_ips);
2128 basic_allocate_unassigned(ctdb, ipflags, all_ips);
2130 /* No failback here! */
2133 static void ip_alloc_nondeterministic_ips(struct ctdb_context *ctdb,
2134 struct ctdb_ipflags *ipflags,
2135 struct ctdb_public_ip_list *all_ips)
2137 /* This should be pushed down into basic_failback. */
2138 struct ctdb_public_ip_list *tmp_ip;
2140 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
2144 unassign_unsuitable_ips(ctdb, ipflags, all_ips);
2146 basic_allocate_unassigned(ctdb, ipflags, all_ips);
2148 /* If we don't want IPs to fail back then don't rebalance IPs. */
2149 if (1 == ctdb->tunable.no_ip_failback) {
2153 /* Now, try to make sure the ip adresses are evenly distributed
2156 basic_failback(ctdb, ipflags, all_ips, num_ips);
2159 static void ip_alloc_lcp2(struct ctdb_context *ctdb,
2160 struct ctdb_ipflags *ipflags,
2161 struct ctdb_public_ip_list *all_ips,
2162 uint32_t *force_rebalance_nodes)
2164 uint32_t *lcp2_imbalances;
2165 bool *rebalance_candidates;
2167 TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
2169 unassign_unsuitable_ips(ctdb, ipflags, all_ips);
2171 lcp2_init(tmp_ctx, ipflags, all_ips,force_rebalance_nodes,
2172 &lcp2_imbalances, &rebalance_candidates);
2174 lcp2_allocate_unassigned(ctdb, ipflags, all_ips, lcp2_imbalances);
2176 /* If we don't want IPs to fail back then don't rebalance IPs. */
2177 if (1 == ctdb->tunable.no_ip_failback) {
2181 /* Now, try to make sure the ip adresses are evenly distributed
2184 lcp2_failback(ctdb, ipflags, all_ips,
2185 lcp2_imbalances, rebalance_candidates);
2188 talloc_free(tmp_ctx);
2191 static bool all_nodes_are_disabled(struct ctdb_node_map *nodemap)
2195 /* Count how many completely healthy nodes we have */
2197 for (i=0;i<nodemap->num;i++) {
2198 if (!(nodemap->nodes[i].flags & (NODE_FLAGS_INACTIVE|NODE_FLAGS_DISABLED))) {
2203 return num_healthy == 0;
2206 /* The calculation part of the IP allocation algorithm. */
2207 static void ctdb_takeover_run_core(struct ctdb_context *ctdb,
2208 struct ctdb_ipflags *ipflags,
2209 struct ctdb_public_ip_list **all_ips_p,
2210 uint32_t *force_rebalance_nodes)
2212 /* since nodes only know about those public addresses that
2213 can be served by that particular node, no single node has
2214 a full list of all public addresses that exist in the cluster.
2215 Walk over all node structures and create a merged list of
2216 all public addresses that exist in the cluster.
2218 keep the tree of ips around as ctdb->ip_tree
2220 *all_ips_p = create_merged_ip_list(ctdb);
2222 if (1 == ctdb->tunable.lcp2_public_ip_assignment) {
2223 ip_alloc_lcp2(ctdb, ipflags, *all_ips_p, force_rebalance_nodes);
2224 } else if (1 == ctdb->tunable.deterministic_public_ips) {
2225 ip_alloc_deterministic_ips(ctdb, ipflags, *all_ips_p);
2227 ip_alloc_nondeterministic_ips(ctdb, ipflags, *all_ips_p);
2230 /* at this point ->pnn is the node which will own each IP
2231 or -1 if there is no node that can cover this ip
2237 struct get_tunable_callback_data {
2238 const char *tunable;
2243 static void get_tunable_callback(struct ctdb_context *ctdb, uint32_t pnn,
2244 int32_t res, TDB_DATA outdata,
2247 struct get_tunable_callback_data *cd =
2248 (struct get_tunable_callback_data *)callback;
2252 /* Already handled in fail callback */
2256 if (outdata.dsize != sizeof(uint32_t)) {
2257 DEBUG(DEBUG_ERR,("Wrong size of returned data when reading \"%s\" tunable from node %d. Expected %d bytes but received %d bytes\n",
2258 cd->tunable, pnn, (int)sizeof(uint32_t),
2259 (int)outdata.dsize));
2264 size = talloc_array_length(cd->out);
2266 DEBUG(DEBUG_ERR,("Got %s reply from node %d but nodemap only has %d entries\n",
2267 cd->tunable, pnn, size));
2272 cd->out[pnn] = *(uint32_t *)outdata.dptr;
2275 static void get_tunable_fail_callback(struct ctdb_context *ctdb, uint32_t pnn,
2276 int32_t res, TDB_DATA outdata,
2279 struct get_tunable_callback_data *cd =
2280 (struct get_tunable_callback_data *)callback;
2285 ("Timed out getting tunable \"%s\" from node %d\n",
2291 DEBUG(DEBUG_WARNING,
2292 ("Tunable \"%s\" not implemented on node %d\n",
2297 ("Unexpected error getting tunable \"%s\" from node %d\n",
2303 static uint32_t *get_tunable_from_nodes(struct ctdb_context *ctdb,
2304 TALLOC_CTX *tmp_ctx,
2305 struct ctdb_node_map *nodemap,
2306 const char *tunable,
2307 uint32_t default_value)
2310 struct ctdb_control_get_tunable *t;
2313 struct get_tunable_callback_data callback_data;
2316 tvals = talloc_array(tmp_ctx, uint32_t, nodemap->num);
2317 CTDB_NO_MEMORY_NULL(ctdb, tvals);
2318 for (i=0; i<nodemap->num; i++) {
2319 tvals[i] = default_value;
2322 callback_data.out = tvals;
2323 callback_data.tunable = tunable;
2324 callback_data.fatal = false;
2326 data.dsize = offsetof(struct ctdb_control_get_tunable, name) + strlen(tunable) + 1;
2327 data.dptr = talloc_size(tmp_ctx, data.dsize);
2328 t = (struct ctdb_control_get_tunable *)data.dptr;
2329 t->length = strlen(tunable)+1;
2330 memcpy(t->name, tunable, t->length);
2331 nodes = list_of_connected_nodes(ctdb, nodemap, tmp_ctx, true);
2332 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_GET_TUNABLE,
2333 nodes, 0, TAKEOVER_TIMEOUT(),
2335 get_tunable_callback,
2336 get_tunable_fail_callback,
2337 &callback_data) != 0) {
2338 if (callback_data.fatal) {
2344 talloc_free(data.dptr);
2349 struct get_runstate_callback_data {
2350 enum ctdb_runstate *out;
2354 static void get_runstate_callback(struct ctdb_context *ctdb, uint32_t pnn,
2355 int32_t res, TDB_DATA outdata,
2356 void *callback_data)
2358 struct get_runstate_callback_data *cd =
2359 (struct get_runstate_callback_data *)callback_data;
2363 /* Already handled in fail callback */
2367 if (outdata.dsize != sizeof(uint32_t)) {
2368 DEBUG(DEBUG_ERR,("Wrong size of returned data when getting runstate from node %d. Expected %d bytes but received %d bytes\n",
2369 pnn, (int)sizeof(uint32_t),
2370 (int)outdata.dsize));
2375 size = talloc_array_length(cd->out);
2377 DEBUG(DEBUG_ERR,("Got reply from node %d but nodemap only has %d entries\n",
2382 cd->out[pnn] = (enum ctdb_runstate)*(uint32_t *)outdata.dptr;
2385 static void get_runstate_fail_callback(struct ctdb_context *ctdb, uint32_t pnn,
2386 int32_t res, TDB_DATA outdata,
2389 struct get_runstate_callback_data *cd =
2390 (struct get_runstate_callback_data *)callback;
2395 ("Timed out getting runstate from node %d\n", pnn));
2399 DEBUG(DEBUG_WARNING,
2400 ("Error getting runstate from node %d - assuming runstates not supported\n",
2405 static enum ctdb_runstate * get_runstate_from_nodes(struct ctdb_context *ctdb,
2406 TALLOC_CTX *tmp_ctx,
2407 struct ctdb_node_map *nodemap,
2408 enum ctdb_runstate default_value)
2411 enum ctdb_runstate *rs;
2412 struct get_runstate_callback_data callback_data;
2415 rs = talloc_array(tmp_ctx, enum ctdb_runstate, nodemap->num);
2416 CTDB_NO_MEMORY_NULL(ctdb, rs);
2417 for (i=0; i<nodemap->num; i++) {
2418 rs[i] = default_value;
2421 callback_data.out = rs;
2422 callback_data.fatal = false;
2424 nodes = list_of_connected_nodes(ctdb, nodemap, tmp_ctx, true);
2425 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_GET_RUNSTATE,
2426 nodes, 0, TAKEOVER_TIMEOUT(),
2428 get_runstate_callback,
2429 get_runstate_fail_callback,
2430 &callback_data) != 0) {
2431 if (callback_data.fatal) {
2441 /* Set internal flags for IP allocation:
2443 * Set NOIPTAKOVER ip flags from per-node NoIPTakeover tunable
2444 * Set NOIPHOST ip flag for each INACTIVE node
2445 * if all nodes are disabled:
2446 * Set NOIPHOST ip flags from per-node NoIPHostOnAllDisabled tunable
2448 * Set NOIPHOST ip flags for disabled nodes
2450 static struct ctdb_ipflags *
2451 set_ipflags_internal(struct ctdb_context *ctdb,
2452 TALLOC_CTX *tmp_ctx,
2453 struct ctdb_node_map *nodemap,
2454 uint32_t *tval_noiptakeover,
2455 uint32_t *tval_noiphostonalldisabled,
2456 enum ctdb_runstate *runstate)
2459 struct ctdb_ipflags *ipflags;
2461 /* Clear IP flags - implicit due to talloc_zero */
2462 ipflags = talloc_zero_array(tmp_ctx, struct ctdb_ipflags, nodemap->num);
2463 CTDB_NO_MEMORY_NULL(ctdb, ipflags);
2465 for (i=0;i<nodemap->num;i++) {
2466 /* Can not take IPs on node with NoIPTakeover set */
2467 if (tval_noiptakeover[i] != 0) {
2468 ipflags[i].noiptakeover = true;
2471 /* Can not host IPs on node not in RUNNING state */
2472 if (runstate[i] != CTDB_RUNSTATE_RUNNING) {
2473 ipflags[i].noiphost = true;
2476 /* Can not host IPs on INACTIVE node */
2477 if (nodemap->nodes[i].flags & NODE_FLAGS_INACTIVE) {
2478 ipflags[i].noiphost = true;
2482 if (all_nodes_are_disabled(nodemap)) {
2483 /* If all nodes are disabled, can not host IPs on node
2484 * with NoIPHostOnAllDisabled set
2486 for (i=0;i<nodemap->num;i++) {
2487 if (tval_noiphostonalldisabled[i] != 0) {
2488 ipflags[i].noiphost = true;
2492 /* If some nodes are not disabled, then can not host
2493 * IPs on DISABLED node
2495 for (i=0;i<nodemap->num;i++) {
2496 if (nodemap->nodes[i].flags & NODE_FLAGS_DISABLED) {
2497 ipflags[i].noiphost = true;
2505 static struct ctdb_ipflags *set_ipflags(struct ctdb_context *ctdb,
2506 TALLOC_CTX *tmp_ctx,
2507 struct ctdb_node_map *nodemap)
2509 uint32_t *tval_noiptakeover;
2510 uint32_t *tval_noiphostonalldisabled;
2511 struct ctdb_ipflags *ipflags;
2512 enum ctdb_runstate *runstate;
2515 tval_noiptakeover = get_tunable_from_nodes(ctdb, tmp_ctx, nodemap,
2517 if (tval_noiptakeover == NULL) {
2521 tval_noiphostonalldisabled =
2522 get_tunable_from_nodes(ctdb, tmp_ctx, nodemap,
2523 "NoIPHostOnAllDisabled", 0);
2524 if (tval_noiphostonalldisabled == NULL) {
2525 /* Caller frees tmp_ctx */
2529 /* Any nodes where CTDB_CONTROL_GET_RUNSTATE is not supported
2530 * will default to CTDB_RUNSTATE_RUNNING. This ensures
2531 * reasonable behaviour on a mixed cluster during upgrade.
2533 runstate = get_runstate_from_nodes(ctdb, tmp_ctx, nodemap,
2534 CTDB_RUNSTATE_RUNNING);
2535 if (runstate == NULL) {
2536 /* Caller frees tmp_ctx */
2540 ipflags = set_ipflags_internal(ctdb, tmp_ctx, nodemap,
2542 tval_noiphostonalldisabled,
2545 talloc_free(tval_noiptakeover);
2546 talloc_free(tval_noiphostonalldisabled);
2547 talloc_free(runstate);
2552 struct iprealloc_callback_data {
2555 client_async_callback fail_callback;
2556 void *fail_callback_data;
2557 struct ctdb_node_map *nodemap;
2560 static void iprealloc_fail_callback(struct ctdb_context *ctdb, uint32_t pnn,
2561 int32_t res, TDB_DATA outdata,
2565 struct iprealloc_callback_data *cd =
2566 (struct iprealloc_callback_data *)callback;
2570 /* If the control timed out then that's a real error,
2571 * so call the real fail callback
2573 cd->fail_callback(ctdb, pnn, res, outdata,
2574 cd->fail_callback_data);
2577 /* If not a timeout then either the ipreallocated
2578 * eventscript (or some setup) failed. This might
2579 * have failed because the IPREALLOCATED control isn't
2580 * implemented - right now there is no way of knowing
2581 * because the error codes are all folded down to -1.
2582 * Consider retrying using EVENTSCRIPT control...
2585 numnodes = talloc_array_length(cd->retry_nodes);
2586 if (pnn > numnodes) {
2588 ("ipreallocated failure from node %d, but only %d nodes in nodemap\n",
2593 /* Can't run the "ipreallocated" event on a INACTIVE node */
2594 if (cd->nodemap->nodes[pnn].flags & NODE_FLAGS_INACTIVE) {
2596 ("ipreallocated failure from node %d, but node is inactive - not flagging a retry\n",
2601 DEBUG(DEBUG_WARNING,
2602 ("ipreallocated failure from node %d, flagging retry\n",
2604 cd->retry_nodes[pnn] = true;
2609 struct takeover_callback_data {
2611 client_async_callback fail_callback;
2612 void *fail_callback_data;
2613 struct ctdb_node_map *nodemap;
2616 static void takeover_run_fail_callback(struct ctdb_context *ctdb,
2617 uint32_t node_pnn, int32_t res,
2618 TDB_DATA outdata, void *callback_data)
2620 struct takeover_callback_data *cd =
2621 talloc_get_type_abort(callback_data,
2622 struct takeover_callback_data);
2625 for (i = 0; i < cd->nodemap->num; i++) {
2626 if (node_pnn == cd->nodemap->nodes[i].pnn) {
2631 if (i == cd->nodemap->num) {
2632 DEBUG(DEBUG_ERR, (__location__ " invalid PNN %u\n", node_pnn));
2636 if (!cd->node_failed[i]) {
2637 cd->node_failed[i] = true;
2638 cd->fail_callback(ctdb, node_pnn, res, outdata,
2639 cd->fail_callback_data);
2644 make any IP alias changes for public addresses that are necessary
2646 int ctdb_takeover_run(struct ctdb_context *ctdb, struct ctdb_node_map *nodemap,
2647 uint32_t *force_rebalance_nodes,
2648 client_async_callback fail_callback, void *callback_data)
2651 struct ctdb_public_ip ip;
2652 struct ctdb_public_ipv4 ipv4;
2654 struct ctdb_public_ip_list *all_ips, *tmp_ip;
2656 struct timeval timeout;
2657 struct client_async_data *async_data;
2658 struct ctdb_client_control_state *state;
2659 TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
2660 struct ctdb_ipflags *ipflags;
2661 struct takeover_callback_data *takeover_data;
2662 struct iprealloc_callback_data iprealloc_data;
2666 * ip failover is completely disabled, just send out the
2667 * ipreallocated event.
2669 if (ctdb->tunable.disable_ip_failover != 0) {
2673 ipflags = set_ipflags(ctdb, tmp_ctx, nodemap);
2674 if (ipflags == NULL) {
2675 DEBUG(DEBUG_ERR,("Failed to set IP flags - aborting takeover run\n"));
2676 talloc_free(tmp_ctx);
2682 /* Do the IP reassignment calculations */
2683 ctdb_takeover_run_core(ctdb, ipflags, &all_ips, force_rebalance_nodes);
2685 /* Now tell all nodes to release any public IPs should not
2686 * host. This will be a NOOP on nodes that don't currently
2687 * hold the given IP.
2689 takeover_data = talloc_zero(tmp_ctx, struct takeover_callback_data);
2690 CTDB_NO_MEMORY_FATAL(ctdb, takeover_data);
2692 takeover_data->node_failed = talloc_zero_array(tmp_ctx,
2693 bool, nodemap->num);
2694 CTDB_NO_MEMORY_FATAL(ctdb, takeover_data->node_failed);
2695 takeover_data->fail_callback = fail_callback;
2696 takeover_data->fail_callback_data = callback_data;
2697 takeover_data->nodemap = nodemap;
2699 async_data = talloc_zero(tmp_ctx, struct client_async_data);
2700 CTDB_NO_MEMORY_FATAL(ctdb, async_data);
2702 async_data->fail_callback = takeover_run_fail_callback;
2703 async_data->callback_data = takeover_data;
2705 for (i=0;i<nodemap->num;i++) {
2706 /* don't talk to unconnected nodes, but do talk to banned nodes */
2707 if (nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED) {
2711 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
2712 if (tmp_ip->pnn == nodemap->nodes[i].pnn) {
2713 /* This node should be serving this
2714 vnn so dont tell it to release the ip
2718 if (tmp_ip->addr.sa.sa_family == AF_INET) {
2719 ipv4.pnn = tmp_ip->pnn;
2720 ipv4.sin = tmp_ip->addr.ip;
2722 timeout = TAKEOVER_TIMEOUT();
2723 data.dsize = sizeof(ipv4);
2724 data.dptr = (uint8_t *)&ipv4;
2725 state = ctdb_control_send(ctdb, nodemap->nodes[i].pnn,
2726 0, CTDB_CONTROL_RELEASE_IPv4, 0,
2730 ip.pnn = tmp_ip->pnn;
2731 ip.addr = tmp_ip->addr;
2733 timeout = TAKEOVER_TIMEOUT();
2734 data.dsize = sizeof(ip);
2735 data.dptr = (uint8_t *)&ip;
2736 state = ctdb_control_send(ctdb, nodemap->nodes[i].pnn,
2737 0, CTDB_CONTROL_RELEASE_IP, 0,
2742 if (state == NULL) {
2743 DEBUG(DEBUG_ERR,(__location__ " Failed to call async control CTDB_CONTROL_RELEASE_IP to node %u\n", nodemap->nodes[i].pnn));
2744 talloc_free(tmp_ctx);
2748 ctdb_client_async_add(async_data, state);
2751 if (ctdb_client_async_wait(ctdb, async_data) != 0) {
2752 DEBUG(DEBUG_ERR,(__location__ " Async control CTDB_CONTROL_RELEASE_IP failed\n"));
2753 talloc_free(tmp_ctx);
2756 talloc_free(async_data);
2759 /* tell all nodes to get their own IPs */
2760 async_data = talloc_zero(tmp_ctx, struct client_async_data);
2761 CTDB_NO_MEMORY_FATAL(ctdb, async_data);
2763 async_data->fail_callback = fail_callback;
2764 async_data->callback_data = callback_data;
2766 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
2767 if (tmp_ip->pnn == -1) {
2768 /* this IP won't be taken over */
2772 if (tmp_ip->addr.sa.sa_family == AF_INET) {
2773 ipv4.pnn = tmp_ip->pnn;
2774 ipv4.sin = tmp_ip->addr.ip;
2776 timeout = TAKEOVER_TIMEOUT();
2777 data.dsize = sizeof(ipv4);
2778 data.dptr = (uint8_t *)&ipv4;
2779 state = ctdb_control_send(ctdb, tmp_ip->pnn,
2780 0, CTDB_CONTROL_TAKEOVER_IPv4, 0,
2784 ip.pnn = tmp_ip->pnn;
2785 ip.addr = tmp_ip->addr;
2787 timeout = TAKEOVER_TIMEOUT();
2788 data.dsize = sizeof(ip);
2789 data.dptr = (uint8_t *)&ip;
2790 state = ctdb_control_send(ctdb, tmp_ip->pnn,
2791 0, CTDB_CONTROL_TAKEOVER_IP, 0,
2795 if (state == NULL) {
2796 DEBUG(DEBUG_ERR,(__location__ " Failed to call async control CTDB_CONTROL_TAKEOVER_IP to node %u\n", tmp_ip->pnn));
2797 talloc_free(tmp_ctx);
2801 ctdb_client_async_add(async_data, state);
2803 if (ctdb_client_async_wait(ctdb, async_data) != 0) {
2804 DEBUG(DEBUG_ERR,(__location__ " Async control CTDB_CONTROL_TAKEOVER_IP failed\n"));
2805 talloc_free(tmp_ctx);
2811 * Tell all nodes to run eventscripts to process the
2812 * "ipreallocated" event. This can do a lot of things,
2813 * including restarting services to reconfigure them if public
2814 * IPs have moved. Once upon a time this event only used to
2817 retry_data = talloc_zero_array(tmp_ctx, bool, nodemap->num);
2818 CTDB_NO_MEMORY_FATAL(ctdb, retry_data);
2819 iprealloc_data.retry_nodes = retry_data;
2820 iprealloc_data.retry_count = 0;
2821 iprealloc_data.fail_callback = fail_callback;
2822 iprealloc_data.fail_callback_data = callback_data;
2823 iprealloc_data.nodemap = nodemap;
2825 nodes = list_of_connected_nodes(ctdb, nodemap, tmp_ctx, true);
2826 ret = ctdb_client_async_control(ctdb, CTDB_CONTROL_IPREALLOCATED,
2827 nodes, 0, TAKEOVER_TIMEOUT(),
2829 NULL, iprealloc_fail_callback,
2832 /* If the control failed then we should retry to any
2833 * nodes flagged by iprealloc_fail_callback using the
2834 * EVENTSCRIPT control. This is a best-effort at
2835 * backward compatiblity when running a mixed cluster
2836 * where some nodes have not yet been upgraded to
2837 * support the IPREALLOCATED control.
2839 DEBUG(DEBUG_WARNING,
2840 ("Retry ipreallocated to some nodes using eventscript control\n"));
2842 nodes = talloc_array(tmp_ctx, uint32_t,
2843 iprealloc_data.retry_count);
2844 CTDB_NO_MEMORY_FATAL(ctdb, nodes);
2847 for (i=0; i<nodemap->num; i++) {
2848 if (iprealloc_data.retry_nodes[i]) {
2854 data.dptr = discard_const("ipreallocated");
2855 data.dsize = strlen((char *)data.dptr) + 1;
2856 ret = ctdb_client_async_control(ctdb,
2857 CTDB_CONTROL_RUN_EVENTSCRIPTS,
2858 nodes, 0, TAKEOVER_TIMEOUT(),
2860 NULL, fail_callback,
2863 DEBUG(DEBUG_ERR, (__location__ " failed to send control to run eventscripts with \"ipreallocated\"\n"));
2867 talloc_free(tmp_ctx);
2873 destroy a ctdb_client_ip structure
2875 static int ctdb_client_ip_destructor(struct ctdb_client_ip *ip)
2877 DEBUG(DEBUG_DEBUG,("destroying client tcp for %s:%u (client_id %u)\n",
2878 ctdb_addr_to_str(&ip->addr),
2879 ntohs(ip->addr.ip.sin_port),
2882 DLIST_REMOVE(ip->ctdb->client_ip_list, ip);
2887 called by a client to inform us of a TCP connection that it is managing
2888 that should tickled with an ACK when IP takeover is done
2889 we handle both the old ipv4 style of packets as well as the new ipv4/6
2892 int32_t ctdb_control_tcp_client(struct ctdb_context *ctdb, uint32_t client_id,
2895 struct ctdb_client *client = ctdb_reqid_find(ctdb, client_id, struct ctdb_client);
2896 struct ctdb_control_tcp *old_addr = NULL;
2897 struct ctdb_control_tcp_addr new_addr;
2898 struct ctdb_control_tcp_addr *tcp_sock = NULL;
2899 struct ctdb_tcp_list *tcp;
2900 struct ctdb_tcp_connection t;
2903 struct ctdb_client_ip *ip;
2904 struct ctdb_vnn *vnn;
2905 ctdb_sock_addr addr;
2907 switch (indata.dsize) {
2908 case sizeof(struct ctdb_control_tcp):
2909 old_addr = (struct ctdb_control_tcp *)indata.dptr;
2910 ZERO_STRUCT(new_addr);
2911 tcp_sock = &new_addr;
2912 tcp_sock->src.ip = old_addr->src;
2913 tcp_sock->dest.ip = old_addr->dest;
2915 case sizeof(struct ctdb_control_tcp_addr):
2916 tcp_sock = (struct ctdb_control_tcp_addr *)indata.dptr;
2919 DEBUG(DEBUG_ERR,(__location__ " Invalid data structure passed "
2920 "to ctdb_control_tcp_client. size was %d but "
2921 "only allowed sizes are %lu and %lu\n",
2923 (long unsigned)sizeof(struct ctdb_control_tcp),
2924 (long unsigned)sizeof(struct ctdb_control_tcp_addr)));
2928 addr = tcp_sock->src;
2929 ctdb_canonicalize_ip(&addr, &tcp_sock->src);
2930 addr = tcp_sock->dest;
2931 ctdb_canonicalize_ip(&addr, &tcp_sock->dest);
2934 memcpy(&addr, &tcp_sock->dest, sizeof(addr));
2935 vnn = find_public_ip_vnn(ctdb, &addr);
2937 switch (addr.sa.sa_family) {
2939 if (ntohl(addr.ip.sin_addr.s_addr) != INADDR_LOOPBACK) {
2940 DEBUG(DEBUG_ERR,("Could not add client IP %s. This is not a public address.\n",
2941 ctdb_addr_to_str(&addr)));
2945 DEBUG(DEBUG_ERR,("Could not add client IP %s. This is not a public ipv6 address.\n",
2946 ctdb_addr_to_str(&addr)));
2949 DEBUG(DEBUG_ERR,(__location__ " Unknown family type %d\n", addr.sa.sa_family));
2955 if (vnn->pnn != ctdb->pnn) {
2956 DEBUG(DEBUG_ERR,("Attempt to register tcp client for IP %s we don't hold - failing (client_id %u pid %u)\n",
2957 ctdb_addr_to_str(&addr),
2958 client_id, client->pid));
2959 /* failing this call will tell smbd to die */
2963 ip = talloc(client, struct ctdb_client_ip);
2964 CTDB_NO_MEMORY(ctdb, ip);
2968 ip->client_id = client_id;
2969 talloc_set_destructor(ip, ctdb_client_ip_destructor);
2970 DLIST_ADD(ctdb->client_ip_list, ip);
2972 tcp = talloc(client, struct ctdb_tcp_list);
2973 CTDB_NO_MEMORY(ctdb, tcp);
2975 tcp->connection.src_addr = tcp_sock->src;
2976 tcp->connection.dst_addr = tcp_sock->dest;
2978 DLIST_ADD(client->tcp_list, tcp);
2980 t.src_addr = tcp_sock->src;
2981 t.dst_addr = tcp_sock->dest;
2983 data.dptr = (uint8_t *)&t;
2984 data.dsize = sizeof(t);
2986 switch (addr.sa.sa_family) {
2988 DEBUG(DEBUG_INFO,("registered tcp client for %u->%s:%u (client_id %u pid %u)\n",
2989 (unsigned)ntohs(tcp_sock->dest.ip.sin_port),
2990 ctdb_addr_to_str(&tcp_sock->src),
2991 (unsigned)ntohs(tcp_sock->src.ip.sin_port), client_id, client->pid));
2994 DEBUG(DEBUG_INFO,("registered tcp client for %u->%s:%u (client_id %u pid %u)\n",
2995 (unsigned)ntohs(tcp_sock->dest.ip6.sin6_port),
2996 ctdb_addr_to_str(&tcp_sock->src),
2997 (unsigned)ntohs(tcp_sock->src.ip6.sin6_port), client_id, client->pid));
3000 DEBUG(DEBUG_ERR,(__location__ " Unknown family %d\n", addr.sa.sa_family));
3004 /* tell all nodes about this tcp connection */
3005 ret = ctdb_daemon_send_control(ctdb, CTDB_BROADCAST_CONNECTED, 0,
3006 CTDB_CONTROL_TCP_ADD,
3007 0, CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL);
3009 DEBUG(DEBUG_ERR,(__location__ " Failed to send CTDB_CONTROL_TCP_ADD\n"));
3017 find a tcp address on a list
3019 static struct ctdb_tcp_connection *ctdb_tcp_find(struct ctdb_tcp_array *array,
3020 struct ctdb_tcp_connection *tcp)
3024 if (array == NULL) {
3028 for (i=0;i<array->num;i++) {
3029 if (ctdb_same_sockaddr(&array->connections[i].src_addr, &tcp->src_addr) &&
3030 ctdb_same_sockaddr(&array->connections[i].dst_addr, &tcp->dst_addr)) {
3031 return &array->connections[i];
3040 called by a daemon to inform us of a TCP connection that one of its
3041 clients managing that should tickled with an ACK when IP takeover is
3044 int32_t ctdb_control_tcp_add(struct ctdb_context *ctdb, TDB_DATA indata, bool tcp_update_needed)
3046 struct ctdb_tcp_connection *p = (struct ctdb_tcp_connection *)indata.dptr;
3047 struct ctdb_tcp_array *tcparray;
3048 struct ctdb_tcp_connection tcp;
3049 struct ctdb_vnn *vnn;
3051 vnn = find_public_ip_vnn(ctdb, &p->dst_addr);
3053 DEBUG(DEBUG_INFO,(__location__ " got TCP_ADD control for an address which is not a public address '%s'\n",
3054 ctdb_addr_to_str(&p->dst_addr)));
3060 tcparray = vnn->tcp_array;
3062 /* If this is the first tickle */
3063 if (tcparray == NULL) {
3064 tcparray = talloc_size(ctdb->nodes,
3065 offsetof(struct ctdb_tcp_array, connections) +
3066 sizeof(struct ctdb_tcp_connection) * 1);
3067 CTDB_NO_MEMORY(ctdb, tcparray);
3068 vnn->tcp_array = tcparray;
3071 tcparray->connections = talloc_size(tcparray, sizeof(struct ctdb_tcp_connection));
3072 CTDB_NO_MEMORY(ctdb, tcparray->connections);
3074 tcparray->connections[tcparray->num].src_addr = p->src_addr;
3075 tcparray->connections[tcparray->num].dst_addr = p->dst_addr;
3078 if (tcp_update_needed) {
3079 vnn->tcp_update_needed = true;
3085 /* Do we already have this tickle ?*/
3086 tcp.src_addr = p->src_addr;
3087 tcp.dst_addr = p->dst_addr;
3088 if (ctdb_tcp_find(vnn->tcp_array, &tcp) != NULL) {
3089 DEBUG(DEBUG_DEBUG,("Already had tickle info for %s:%u for vnn:%u\n",
3090 ctdb_addr_to_str(&tcp.dst_addr),
3091 ntohs(tcp.dst_addr.ip.sin_port),
3096 /* A new tickle, we must add it to the array */
3097 tcparray->connections = talloc_realloc(tcparray, tcparray->connections,
3098 struct ctdb_tcp_connection,
3100 CTDB_NO_MEMORY(ctdb, tcparray->connections);
3102 vnn->tcp_array = tcparray;
3103 tcparray->connections[tcparray->num].src_addr = p->src_addr;
3104 tcparray->connections[tcparray->num].dst_addr = p->dst_addr;
3107 DEBUG(DEBUG_INFO,("Added tickle info for %s:%u from vnn %u\n",
3108 ctdb_addr_to_str(&tcp.dst_addr),
3109 ntohs(tcp.dst_addr.ip.sin_port),
3112 if (tcp_update_needed) {
3113 vnn->tcp_update_needed = true;
3121 called by a daemon to inform us of a TCP connection that one of its
3122 clients managing that should tickled with an ACK when IP takeover is
3125 static void ctdb_remove_tcp_connection(struct ctdb_context *ctdb, struct ctdb_tcp_connection *conn)
3127 struct ctdb_tcp_connection *tcpp;
3128 struct ctdb_vnn *vnn = find_public_ip_vnn(ctdb, &conn->dst_addr);
3131 DEBUG(DEBUG_ERR,(__location__ " unable to find public address %s\n",
3132 ctdb_addr_to_str(&conn->dst_addr)));
3136 /* if the array is empty we cant remove it
3137 and we dont need to do anything
3139 if (vnn->tcp_array == NULL) {
3140 DEBUG(DEBUG_INFO,("Trying to remove tickle that doesnt exist (array is empty) %s:%u\n",
3141 ctdb_addr_to_str(&conn->dst_addr),
3142 ntohs(conn->dst_addr.ip.sin_port)));
3147 /* See if we know this connection
3148 if we dont know this connection then we dont need to do anything
3150 tcpp = ctdb_tcp_find(vnn->tcp_array, conn);
3152 DEBUG(DEBUG_INFO,("Trying to remove tickle that doesnt exist %s:%u\n",
3153 ctdb_addr_to_str(&conn->dst_addr),
3154 ntohs(conn->dst_addr.ip.sin_port)));
3159 /* We need to remove this entry from the array.
3160 Instead of allocating a new array and copying data to it
3161 we cheat and just copy the last entry in the existing array
3162 to the entry that is to be removed and just shring the
3165 *tcpp = vnn->tcp_array->connections[vnn->tcp_array->num - 1];
3166 vnn->tcp_array->num--;
3168 /* If we deleted the last entry we also need to remove the entire array
3170 if (vnn->tcp_array->num == 0) {
3171 talloc_free(vnn->tcp_array);
3172 vnn->tcp_array = NULL;
3175 vnn->tcp_update_needed = true;
3177 DEBUG(DEBUG_INFO,("Removed tickle info for %s:%u\n",
3178 ctdb_addr_to_str(&conn->src_addr),
3179 ntohs(conn->src_addr.ip.sin_port)));
3184 called by a daemon to inform us of a TCP connection that one of its
3185 clients used are no longer needed in the tickle database
3187 int32_t ctdb_control_tcp_remove(struct ctdb_context *ctdb, TDB_DATA indata)
3189 struct ctdb_tcp_connection *conn = (struct ctdb_tcp_connection *)indata.dptr;
3191 ctdb_remove_tcp_connection(ctdb, conn);
3198 called when a daemon restarts - send all tickes for all public addresses
3199 we are serving immediately to the new node.
3201 int32_t ctdb_control_startup(struct ctdb_context *ctdb, uint32_t vnn)
3203 /*XXX here we should send all tickes we are serving to the new node */
3209 called when a client structure goes away - hook to remove
3210 elements from the tcp_list in all daemons
3212 void ctdb_takeover_client_destructor_hook(struct ctdb_client *client)
3214 while (client->tcp_list) {
3215 struct ctdb_tcp_list *tcp = client->tcp_list;
3216 DLIST_REMOVE(client->tcp_list, tcp);
3217 ctdb_remove_tcp_connection(client->ctdb, &tcp->connection);
3223 release all IPs on shutdown
3225 void ctdb_release_all_ips(struct ctdb_context *ctdb)
3227 struct ctdb_vnn *vnn;
3230 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
3231 if (!ctdb_sys_have_ip(&vnn->public_address)) {
3232 ctdb_vnn_unassign_iface(ctdb, vnn);
3239 DEBUG(DEBUG_INFO,("Release of IP %s/%u on interface %s node:-1\n",
3240 ctdb_addr_to_str(&vnn->public_address),
3241 vnn->public_netmask_bits,
3242 ctdb_vnn_iface_string(vnn)));
3244 ctdb_event_script_args(ctdb, CTDB_EVENT_RELEASE_IP, "%s %s %u",
3245 ctdb_vnn_iface_string(vnn),
3246 ctdb_addr_to_str(&vnn->public_address),
3247 vnn->public_netmask_bits);
3248 release_kill_clients(ctdb, &vnn->public_address);
3249 ctdb_vnn_unassign_iface(ctdb, vnn);
3253 DEBUG(DEBUG_NOTICE,(__location__ " Released %d public IPs\n", count));
3258 get list of public IPs
3260 int32_t ctdb_control_get_public_ips(struct ctdb_context *ctdb,
3261 struct ctdb_req_control *c, TDB_DATA *outdata)
3264 struct ctdb_all_public_ips *ips;
3265 struct ctdb_vnn *vnn;
3266 bool only_available = false;
3268 if (c->flags & CTDB_PUBLIC_IP_FLAGS_ONLY_AVAILABLE) {
3269 only_available = true;
3272 /* count how many public ip structures we have */
3274 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
3278 len = offsetof(struct ctdb_all_public_ips, ips) +
3279 num*sizeof(struct ctdb_public_ip);
3280 ips = talloc_zero_size(outdata, len);
3281 CTDB_NO_MEMORY(ctdb, ips);
3284 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
3285 if (only_available && !ctdb_vnn_available(ctdb, vnn)) {
3288 ips->ips[i].pnn = vnn->pnn;
3289 ips->ips[i].addr = vnn->public_address;
3293 len = offsetof(struct ctdb_all_public_ips, ips) +
3294 i*sizeof(struct ctdb_public_ip);
3296 outdata->dsize = len;
3297 outdata->dptr = (uint8_t *)ips;
3304 get list of public IPs, old ipv4 style. only returns ipv4 addresses
3306 int32_t ctdb_control_get_public_ipsv4(struct ctdb_context *ctdb,
3307 struct ctdb_req_control *c, TDB_DATA *outdata)
3310 struct ctdb_all_public_ipsv4 *ips;
3311 struct ctdb_vnn *vnn;
3313 /* count how many public ip structures we have */
3315 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
3316 if (vnn->public_address.sa.sa_family != AF_INET) {
3322 len = offsetof(struct ctdb_all_public_ipsv4, ips) +
3323 num*sizeof(struct ctdb_public_ipv4);
3324 ips = talloc_zero_size(outdata, len);
3325 CTDB_NO_MEMORY(ctdb, ips);
3327 outdata->dsize = len;
3328 outdata->dptr = (uint8_t *)ips;
3332 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
3333 if (vnn->public_address.sa.sa_family != AF_INET) {
3336 ips->ips[i].pnn = vnn->pnn;
3337 ips->ips[i].sin = vnn->public_address.ip;
3344 int32_t ctdb_control_get_public_ip_info(struct ctdb_context *ctdb,
3345 struct ctdb_req_control *c,
3350 ctdb_sock_addr *addr;
3351 struct ctdb_control_public_ip_info *info;
3352 struct ctdb_vnn *vnn;
3354 addr = (ctdb_sock_addr *)indata.dptr;
3356 vnn = find_public_ip_vnn(ctdb, addr);
3358 /* if it is not a public ip it could be our 'single ip' */
3359 if (ctdb->single_ip_vnn) {
3360 if (ctdb_same_ip(&ctdb->single_ip_vnn->public_address, addr)) {
3361 vnn = ctdb->single_ip_vnn;
3366 DEBUG(DEBUG_ERR,(__location__ " Could not get public ip info, "
3367 "'%s'not a public address\n",
3368 ctdb_addr_to_str(addr)));
3372 /* count how many public ip structures we have */
3374 for (;vnn->ifaces[num];) {
3378 len = offsetof(struct ctdb_control_public_ip_info, ifaces) +
3379 num*sizeof(struct ctdb_control_iface_info);
3380 info = talloc_zero_size(outdata, len);
3381 CTDB_NO_MEMORY(ctdb, info);
3383 info->ip.addr = vnn->public_address;
3384 info->ip.pnn = vnn->pnn;
3385 info->active_idx = 0xFFFFFFFF;
3387 for (i=0; vnn->ifaces[i]; i++) {
3388 struct ctdb_iface *cur;
3390 cur = ctdb_find_iface(ctdb, vnn->ifaces[i]);
3392 DEBUG(DEBUG_CRIT, (__location__ " internal error iface[%s] unknown\n",
3396 if (vnn->iface == cur) {
3397 info->active_idx = i;
3399 strcpy(info->ifaces[i].name, cur->name);
3400 info->ifaces[i].link_state = cur->link_up;
3401 info->ifaces[i].references = cur->references;
3404 len = offsetof(struct ctdb_control_public_ip_info, ifaces) +
3405 i*sizeof(struct ctdb_control_iface_info);
3407 outdata->dsize = len;
3408 outdata->dptr = (uint8_t *)info;
3413 int32_t ctdb_control_get_ifaces(struct ctdb_context *ctdb,
3414 struct ctdb_req_control *c,
3418 struct ctdb_control_get_ifaces *ifaces;
3419 struct ctdb_iface *cur;
3421 /* count how many public ip structures we have */
3423 for (cur=ctdb->ifaces;cur;cur=cur->next) {
3427 len = offsetof(struct ctdb_control_get_ifaces, ifaces) +
3428 num*sizeof(struct ctdb_control_iface_info);
3429 ifaces = talloc_zero_size(outdata, len);
3430 CTDB_NO_MEMORY(ctdb, ifaces);
3433 for (cur=ctdb->ifaces;cur;cur=cur->next) {
3434 strcpy(ifaces->ifaces[i].name, cur->name);
3435 ifaces->ifaces[i].link_state = cur->link_up;
3436 ifaces->ifaces[i].references = cur->references;
3440 len = offsetof(struct ctdb_control_get_ifaces, ifaces) +
3441 i*sizeof(struct ctdb_control_iface_info);
3443 outdata->dsize = len;
3444 outdata->dptr = (uint8_t *)ifaces;
3449 int32_t ctdb_control_set_iface_link(struct ctdb_context *ctdb,
3450 struct ctdb_req_control *c,
3453 struct ctdb_control_iface_info *info;
3454 struct ctdb_iface *iface;
3455 bool link_up = false;
3457 info = (struct ctdb_control_iface_info *)indata.dptr;
3459 if (info->name[CTDB_IFACE_SIZE] != '\0') {
3460 int len = strnlen(info->name, CTDB_IFACE_SIZE);
3461 DEBUG(DEBUG_ERR, (__location__ " name[%*.*s] not terminated\n",
3462 len, len, info->name));
3466 switch (info->link_state) {
3474 DEBUG(DEBUG_ERR, (__location__ " link_state[%u] invalid\n",
3475 (unsigned int)info->link_state));
3479 if (info->references != 0) {
3480 DEBUG(DEBUG_ERR, (__location__ " references[%u] should be 0\n",
3481 (unsigned int)info->references));
3485 iface = ctdb_find_iface(ctdb, info->name);
3486 if (iface == NULL) {
3490 if (link_up == iface->link_up) {
3494 DEBUG(iface->link_up?DEBUG_ERR:DEBUG_NOTICE,
3495 ("iface[%s] has changed it's link status %s => %s\n",
3497 iface->link_up?"up":"down",
3498 link_up?"up":"down"));
3500 iface->link_up = link_up;
3506 structure containing the listening socket and the list of tcp connections
3507 that the ctdb daemon is to kill
3509 struct ctdb_kill_tcp {
3510 struct ctdb_vnn *vnn;
3511 struct ctdb_context *ctdb;
3513 struct fd_event *fde;
3514 trbt_tree_t *connections;
3519 a tcp connection that is to be killed
3521 struct ctdb_killtcp_con {
3522 ctdb_sock_addr src_addr;
3523 ctdb_sock_addr dst_addr;
3525 struct ctdb_kill_tcp *killtcp;
3528 /* this function is used to create a key to represent this socketpair
3529 in the killtcp tree.
3530 this key is used to insert and lookup matching socketpairs that are
3531 to be tickled and RST
3533 #define KILLTCP_KEYLEN 10
3534 static uint32_t *killtcp_key(ctdb_sock_addr *src, ctdb_sock_addr *dst)
3536 static uint32_t key[KILLTCP_KEYLEN];
3538 bzero(key, sizeof(key));
3540 if (src->sa.sa_family != dst->sa.sa_family) {
3541 DEBUG(DEBUG_ERR, (__location__ " ERROR, different families passed :%u vs %u\n", src->sa.sa_family, dst->sa.sa_family));
3545 switch (src->sa.sa_family) {
3547 key[0] = dst->ip.sin_addr.s_addr;
3548 key[1] = src->ip.sin_addr.s_addr;
3549 key[2] = dst->ip.sin_port;
3550 key[3] = src->ip.sin_port;
3553 uint32_t *dst6_addr32 =
3554 (uint32_t *)&(dst->ip6.sin6_addr.s6_addr);
3555 uint32_t *src6_addr32 =
3556 (uint32_t *)&(src->ip6.sin6_addr.s6_addr);
3557 key[0] = dst6_addr32[3];
3558 key[1] = src6_addr32[3];
3559 key[2] = dst6_addr32[2];
3560 key[3] = src6_addr32[2];
3561 key[4] = dst6_addr32[1];
3562 key[5] = src6_addr32[1];
3563 key[6] = dst6_addr32[0];
3564 key[7] = src6_addr32[0];
3565 key[8] = dst->ip6.sin6_port;
3566 key[9] = src->ip6.sin6_port;
3570 DEBUG(DEBUG_ERR, (__location__ " ERROR, unknown family passed :%u\n", src->sa.sa_family));
3578 called when we get a read event on the raw socket
3580 static void capture_tcp_handler(struct event_context *ev, struct fd_event *fde,
3581 uint16_t flags, void *private_data)
3583 struct ctdb_kill_tcp *killtcp = talloc_get_type(private_data, struct ctdb_kill_tcp);
3584 struct ctdb_killtcp_con *con;
3585 ctdb_sock_addr src, dst;
3586 uint32_t ack_seq, seq;
3588 if (!(flags & EVENT_FD_READ)) {
3592 if (ctdb_sys_read_tcp_packet(killtcp->capture_fd,
3593 killtcp->private_data,
3595 &ack_seq, &seq) != 0) {
3596 /* probably a non-tcp ACK packet */
3600 /* check if we have this guy in our list of connections
3603 con = trbt_lookuparray32(killtcp->connections,
3604 KILLTCP_KEYLEN, killtcp_key(&src, &dst));
3606 /* no this was some other packet we can just ignore */
3610 /* This one has been tickled !
3611 now reset him and remove him from the list.
3613 DEBUG(DEBUG_INFO, ("sending a tcp reset to kill connection :%d -> %s:%d\n",
3614 ntohs(con->dst_addr.ip.sin_port),
3615 ctdb_addr_to_str(&con->src_addr),
3616 ntohs(con->src_addr.ip.sin_port)));
3618 ctdb_sys_send_tcp(&con->dst_addr, &con->src_addr, ack_seq, seq, 1);
3623 /* when traversing the list of all tcp connections to send tickle acks to
3624 (so that we can capture the ack coming back and kill the connection
3626 this callback is called for each connection we are currently trying to kill
3628 static int tickle_connection_traverse(void *param, void *data)
3630 struct ctdb_killtcp_con *con = talloc_get_type(data, struct ctdb_killtcp_con);
3632 /* have tried too many times, just give up */
3633 if (con->count >= 5) {
3634 /* can't delete in traverse: reparent to delete_cons */
3635 talloc_steal(param, con);
3639 /* othervise, try tickling it again */
3642 (ctdb_sock_addr *)&con->dst_addr,
3643 (ctdb_sock_addr *)&con->src_addr,
3650 called every second until all sentenced connections have been reset
3652 static void ctdb_tickle_sentenced_connections(struct event_context *ev, struct timed_event *te,
3653 struct timeval t, void *private_data)
3655 struct ctdb_kill_tcp *killtcp = talloc_get_type(private_data, struct ctdb_kill_tcp);
3656 void *delete_cons = talloc_new(NULL);
3658 /* loop over all connections sending tickle ACKs */
3659 trbt_traversearray32(killtcp->connections, KILLTCP_KEYLEN, tickle_connection_traverse, delete_cons);
3661 /* now we've finished traverse, it's safe to do deletion. */
3662 talloc_free(delete_cons);
3664 /* If there are no more connections to kill we can remove the
3665 entire killtcp structure
3667 if ( (killtcp->connections == NULL) ||
3668 (killtcp->connections->root == NULL) ) {
3669 talloc_free(killtcp);
3673 /* try tickling them again in a seconds time
3675 event_add_timed(killtcp->ctdb->ev, killtcp, timeval_current_ofs(1, 0),
3676 ctdb_tickle_sentenced_connections, killtcp);
3680 destroy the killtcp structure
3682 static int ctdb_killtcp_destructor(struct ctdb_kill_tcp *killtcp)
3684 struct ctdb_vnn *tmpvnn;
3686 /* verify that this vnn is still active */
3687 for (tmpvnn = killtcp->ctdb->vnn; tmpvnn; tmpvnn = tmpvnn->next) {
3688 if (tmpvnn == killtcp->vnn) {
3693 if (tmpvnn == NULL) {
3697 if (killtcp->vnn->killtcp != killtcp) {
3701 killtcp->vnn->killtcp = NULL;
3707 /* nothing fancy here, just unconditionally replace any existing
3708 connection structure with the new one.
3710 dont even free the old one if it did exist, that one is talloc_stolen
3711 by the same node in the tree anyway and will be deleted when the new data
3714 static void *add_killtcp_callback(void *parm, void *data)
3720 add a tcp socket to the list of connections we want to RST
3722 static int ctdb_killtcp_add_connection(struct ctdb_context *ctdb,
3726 ctdb_sock_addr src, dst;
3727 struct ctdb_kill_tcp *killtcp;
3728 struct ctdb_killtcp_con *con;
3729 struct ctdb_vnn *vnn;
3731 ctdb_canonicalize_ip(s, &src);
3732 ctdb_canonicalize_ip(d, &dst);
3734 vnn = find_public_ip_vnn(ctdb, &dst);
3736 vnn = find_public_ip_vnn(ctdb, &src);
3739 /* if it is not a public ip it could be our 'single ip' */
3740 if (ctdb->single_ip_vnn) {
3741 if (ctdb_same_ip(&ctdb->single_ip_vnn->public_address, &dst)) {
3742 vnn = ctdb->single_ip_vnn;
3747 DEBUG(DEBUG_ERR,(__location__ " Could not killtcp, not a public address\n"));
3751 killtcp = vnn->killtcp;
3753 /* If this is the first connection to kill we must allocate
3756 if (killtcp == NULL) {
3757 killtcp = talloc_zero(vnn, struct ctdb_kill_tcp);
3758 CTDB_NO_MEMORY(ctdb, killtcp);
3761 killtcp->ctdb = ctdb;
3762 killtcp->capture_fd = -1;
3763 killtcp->connections = trbt_create(killtcp, 0);
3765 vnn->killtcp = killtcp;
3766 talloc_set_destructor(killtcp, ctdb_killtcp_destructor);
3771 /* create a structure that describes this connection we want to
3772 RST and store it in killtcp->connections
3774 con = talloc(killtcp, struct ctdb_killtcp_con);
3775 CTDB_NO_MEMORY(ctdb, con);
3776 con->src_addr = src;
3777 con->dst_addr = dst;
3779 con->killtcp = killtcp;
3782 trbt_insertarray32_callback(killtcp->connections,
3783 KILLTCP_KEYLEN, killtcp_key(&con->dst_addr, &con->src_addr),
3784 add_killtcp_callback, con);
3787 If we dont have a socket to listen on yet we must create it
3789 if (killtcp->capture_fd == -1) {
3790 const char *iface = ctdb_vnn_iface_string(vnn);
3791 killtcp->capture_fd = ctdb_sys_open_capture_socket(iface, &killtcp->private_data);
3792 if (killtcp->capture_fd == -1) {
3793 DEBUG(DEBUG_CRIT,(__location__ " Failed to open capturing "
3794 "socket on iface '%s' for killtcp (%s)\n",
3795 iface, strerror(errno)));
3801 if (killtcp->fde == NULL) {
3802 killtcp->fde = event_add_fd(ctdb->ev, killtcp, killtcp->capture_fd,
3804 capture_tcp_handler, killtcp);
3805 tevent_fd_set_auto_close(killtcp->fde);
3807 /* We also need to set up some events to tickle all these connections
3808 until they are all reset
3810 event_add_timed(ctdb->ev, killtcp, timeval_current_ofs(1, 0),
3811 ctdb_tickle_sentenced_connections, killtcp);
3814 /* tickle him once now */
3823 talloc_free(vnn->killtcp);
3824 vnn->killtcp = NULL;
3829 kill a TCP connection.
3831 int32_t ctdb_control_kill_tcp(struct ctdb_context *ctdb, TDB_DATA indata)
3833 struct ctdb_control_killtcp *killtcp = (struct ctdb_control_killtcp *)indata.dptr;
3835 return ctdb_killtcp_add_connection(ctdb, &killtcp->src_addr, &killtcp->dst_addr);
3839 called by a daemon to inform us of the entire list of TCP tickles for
3840 a particular public address.
3841 this control should only be sent by the node that is currently serving
3842 that public address.
3844 int32_t ctdb_control_set_tcp_tickle_list(struct ctdb_context *ctdb, TDB_DATA indata)
3846 struct ctdb_control_tcp_tickle_list *list = (struct ctdb_control_tcp_tickle_list *)indata.dptr;
3847 struct ctdb_tcp_array *tcparray;
3848 struct ctdb_vnn *vnn;
3850 /* We must at least have tickles.num or else we cant verify the size
3851 of the received data blob
3853 if (indata.dsize < offsetof(struct ctdb_control_tcp_tickle_list,
3854 tickles.connections)) {
3855 DEBUG(DEBUG_ERR,("Bad indata in ctdb_control_set_tcp_tickle_list. Not enough data for the tickle.num field\n"));
3859 /* verify that the size of data matches what we expect */
3860 if (indata.dsize < offsetof(struct ctdb_control_tcp_tickle_list,
3861 tickles.connections)
3862 + sizeof(struct ctdb_tcp_connection)
3863 * list->tickles.num) {
3864 DEBUG(DEBUG_ERR,("Bad indata in ctdb_control_set_tcp_tickle_list\n"));
3868 vnn = find_public_ip_vnn(ctdb, &list->addr);
3870 DEBUG(DEBUG_INFO,(__location__ " Could not set tcp tickle list, '%s' is not a public address\n",
3871 ctdb_addr_to_str(&list->addr)));
3876 /* remove any old ticklelist we might have */
3877 talloc_free(vnn->tcp_array);
3878 vnn->tcp_array = NULL;
3880 tcparray = talloc(ctdb->nodes, struct ctdb_tcp_array);
3881 CTDB_NO_MEMORY(ctdb, tcparray);
3883 tcparray->num = list->tickles.num;
3885 tcparray->connections = talloc_array(tcparray, struct ctdb_tcp_connection, tcparray->num);
3886 CTDB_NO_MEMORY(ctdb, tcparray->connections);
3888 memcpy(tcparray->connections, &list->tickles.connections[0],
3889 sizeof(struct ctdb_tcp_connection)*tcparray->num);
3891 /* We now have a new fresh tickle list array for this vnn */
3892 vnn->tcp_array = talloc_steal(vnn, tcparray);
3898 called to return the full list of tickles for the puclic address associated
3899 with the provided vnn
3901 int32_t ctdb_control_get_tcp_tickle_list(struct ctdb_context *ctdb, TDB_DATA indata, TDB_DATA *outdata)
3903 ctdb_sock_addr *addr = (ctdb_sock_addr *)indata.dptr;
3904 struct ctdb_control_tcp_tickle_list *list;
3905 struct ctdb_tcp_array *tcparray;
3907 struct ctdb_vnn *vnn;
3909 vnn = find_public_ip_vnn(ctdb, addr);
3911 DEBUG(DEBUG_ERR,(__location__ " Could not get tcp tickle list, '%s' is not a public address\n",
3912 ctdb_addr_to_str(addr)));
3917 tcparray = vnn->tcp_array;
3919 num = tcparray->num;
3924 outdata->dsize = offsetof(struct ctdb_control_tcp_tickle_list,
3925 tickles.connections)
3926 + sizeof(struct ctdb_tcp_connection) * num;
3928 outdata->dptr = talloc_size(outdata, outdata->dsize);
3929 CTDB_NO_MEMORY(ctdb, outdata->dptr);
3930 list = (struct ctdb_control_tcp_tickle_list *)outdata->dptr;
3933 list->tickles.num = num;
3935 memcpy(&list->tickles.connections[0], tcparray->connections,
3936 sizeof(struct ctdb_tcp_connection) * num);
3944 set the list of all tcp tickles for a public address
3946 static int ctdb_ctrl_set_tcp_tickles(struct ctdb_context *ctdb,
3947 struct timeval timeout, uint32_t destnode,
3948 ctdb_sock_addr *addr,
3949 struct ctdb_tcp_array *tcparray)
3953 struct ctdb_control_tcp_tickle_list *list;
3956 num = tcparray->num;
3961 data.dsize = offsetof(struct ctdb_control_tcp_tickle_list,
3962 tickles.connections) +
3963 sizeof(struct ctdb_tcp_connection) * num;
3964 data.dptr = talloc_size(ctdb, data.dsize);
3965 CTDB_NO_MEMORY(ctdb, data.dptr);
3967 list = (struct ctdb_control_tcp_tickle_list *)data.dptr;
3969 list->tickles.num = num;
3971 memcpy(&list->tickles.connections[0], tcparray->connections, sizeof(struct ctdb_tcp_connection) * num);
3974 ret = ctdb_daemon_send_control(ctdb, CTDB_BROADCAST_CONNECTED, 0,
3975 CTDB_CONTROL_SET_TCP_TICKLE_LIST,
3976 0, CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL);
3978 DEBUG(DEBUG_ERR,(__location__ " ctdb_control for set tcp tickles failed\n"));
3982 talloc_free(data.dptr);
3989 perform tickle updates if required
3991 static void ctdb_update_tcp_tickles(struct event_context *ev,
3992 struct timed_event *te,
3993 struct timeval t, void *private_data)
3995 struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
3997 struct ctdb_vnn *vnn;
3999 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
4000 /* we only send out updates for public addresses that
4003 if (ctdb->pnn != vnn->pnn) {
4006 /* We only send out the updates if we need to */
4007 if (!vnn->tcp_update_needed) {
4010 ret = ctdb_ctrl_set_tcp_tickles(ctdb,
4012 CTDB_BROADCAST_CONNECTED,
4013 &vnn->public_address,
4016 DEBUG(DEBUG_ERR,("Failed to send the tickle update for public address %s\n",
4017 ctdb_addr_to_str(&vnn->public_address)));
4021 event_add_timed(ctdb->ev, ctdb->tickle_update_context,
4022 timeval_current_ofs(ctdb->tunable.tickle_update_interval, 0),
4023 ctdb_update_tcp_tickles, ctdb);
4028 start periodic update of tcp tickles
4030 void ctdb_start_tcp_tickle_update(struct ctdb_context *ctdb)
4032 ctdb->tickle_update_context = talloc_new(ctdb);
4034 event_add_timed(ctdb->ev, ctdb->tickle_update_context,
4035 timeval_current_ofs(ctdb->tunable.tickle_update_interval, 0),
4036 ctdb_update_tcp_tickles, ctdb);
4042 struct control_gratious_arp {
4043 struct ctdb_context *ctdb;
4044 ctdb_sock_addr addr;
4050 send a control_gratuitous arp
4052 static void send_gratious_arp(struct event_context *ev, struct timed_event *te,
4053 struct timeval t, void *private_data)
4056 struct control_gratious_arp *arp = talloc_get_type(private_data,
4057 struct control_gratious_arp);
4059 ret = ctdb_sys_send_arp(&arp->addr, arp->iface);
4061 DEBUG(DEBUG_ERR,(__location__ " sending of gratious arp on iface '%s' failed (%s)\n",
4062 arp->iface, strerror(errno)));
4067 if (arp->count == CTDB_ARP_REPEAT) {
4072 event_add_timed(arp->ctdb->ev, arp,
4073 timeval_current_ofs(CTDB_ARP_INTERVAL, 0),
4074 send_gratious_arp, arp);
4081 int32_t ctdb_control_send_gratious_arp(struct ctdb_context *ctdb, TDB_DATA indata)
4083 struct ctdb_control_gratious_arp *gratious_arp = (struct ctdb_control_gratious_arp *)indata.dptr;
4084 struct control_gratious_arp *arp;
4086 /* verify the size of indata */
4087 if (indata.dsize < offsetof(struct ctdb_control_gratious_arp, iface)) {
4088 DEBUG(DEBUG_ERR,(__location__ " Too small indata to hold a ctdb_control_gratious_arp structure. Got %u require %u bytes\n",
4089 (unsigned)indata.dsize,
4090 (unsigned)offsetof(struct ctdb_control_gratious_arp, iface)));
4094 ( offsetof(struct ctdb_control_gratious_arp, iface)
4095 + gratious_arp->len ) ){
4097 DEBUG(DEBUG_ERR,(__location__ " Wrong size of indata. Was %u bytes "
4098 "but should be %u bytes\n",
4099 (unsigned)indata.dsize,
4100 (unsigned)(offsetof(struct ctdb_control_gratious_arp, iface)+gratious_arp->len)));
4105 arp = talloc(ctdb, struct control_gratious_arp);
4106 CTDB_NO_MEMORY(ctdb, arp);
4109 arp->addr = gratious_arp->addr;
4110 arp->iface = talloc_strdup(arp, gratious_arp->iface);
4111 CTDB_NO_MEMORY(ctdb, arp->iface);
4114 event_add_timed(arp->ctdb->ev, arp,
4115 timeval_zero(), send_gratious_arp, arp);
4120 int32_t ctdb_control_add_public_address(struct ctdb_context *ctdb, TDB_DATA indata)
4122 struct ctdb_control_ip_iface *pub = (struct ctdb_control_ip_iface *)indata.dptr;
4125 /* verify the size of indata */
4126 if (indata.dsize < offsetof(struct ctdb_control_ip_iface, iface)) {
4127 DEBUG(DEBUG_ERR,(__location__ " Too small indata to hold a ctdb_control_ip_iface structure\n"));
4131 ( offsetof(struct ctdb_control_ip_iface, iface)
4134 DEBUG(DEBUG_ERR,(__location__ " Wrong size of indata. Was %u bytes "
4135 "but should be %u bytes\n",
4136 (unsigned)indata.dsize,
4137 (unsigned)(offsetof(struct ctdb_control_ip_iface, iface)+pub->len)));
4141 DEBUG(DEBUG_NOTICE,("Add IP %s\n", ctdb_addr_to_str(&pub->addr)));
4143 ret = ctdb_add_public_address(ctdb, &pub->addr, pub->mask, &pub->iface[0], true);
4146 DEBUG(DEBUG_ERR,(__location__ " Failed to add public address\n"));
4154 called when releaseip event finishes for del_public_address
4156 static void delete_ip_callback(struct ctdb_context *ctdb, int status,
4159 talloc_free(private_data);
4162 int32_t ctdb_control_del_public_address(struct ctdb_context *ctdb, TDB_DATA indata)
4164 struct ctdb_control_ip_iface *pub = (struct ctdb_control_ip_iface *)indata.dptr;
4165 struct ctdb_vnn *vnn;
4168 /* verify the size of indata */
4169 if (indata.dsize < offsetof(struct ctdb_control_ip_iface, iface)) {
4170 DEBUG(DEBUG_ERR,(__location__ " Too small indata to hold a ctdb_control_ip_iface structure\n"));
4174 ( offsetof(struct ctdb_control_ip_iface, iface)
4177 DEBUG(DEBUG_ERR,(__location__ " Wrong size of indata. Was %u bytes "
4178 "but should be %u bytes\n",
4179 (unsigned)indata.dsize,
4180 (unsigned)(offsetof(struct ctdb_control_ip_iface, iface)+pub->len)));
4184 DEBUG(DEBUG_NOTICE,("Delete IP %s\n", ctdb_addr_to_str(&pub->addr)));
4186 /* walk over all public addresses until we find a match */
4187 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
4188 if (ctdb_same_ip(&vnn->public_address, &pub->addr)) {
4189 TALLOC_CTX *mem_ctx = talloc_new(ctdb);
4191 DLIST_REMOVE(ctdb->vnn, vnn);
4192 talloc_steal(mem_ctx, vnn);
4193 ctdb_remove_orphaned_ifaces(ctdb, vnn, mem_ctx);
4194 if (vnn->pnn != ctdb->pnn) {
4195 if (vnn->iface != NULL) {
4196 ctdb_vnn_unassign_iface(ctdb, vnn);
4198 talloc_free(mem_ctx);
4203 ret = ctdb_event_script_callback(ctdb,
4204 mem_ctx, delete_ip_callback, mem_ctx,
4206 CTDB_EVENT_RELEASE_IP,
4208 ctdb_vnn_iface_string(vnn),
4209 ctdb_addr_to_str(&vnn->public_address),
4210 vnn->public_netmask_bits);
4211 if (vnn->iface != NULL) {
4212 ctdb_vnn_unassign_iface(ctdb, vnn);
4225 struct ipreallocated_callback_state {
4226 struct ctdb_req_control *c;
4229 static void ctdb_ipreallocated_callback(struct ctdb_context *ctdb,
4230 int status, void *p)
4232 struct ipreallocated_callback_state *state =
4233 talloc_get_type(p, struct ipreallocated_callback_state);
4237 (" \"ipreallocated\" event script failed (status %d)\n",
4239 if (status == -ETIME) {
4240 ctdb_ban_self(ctdb);
4244 ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
4248 /* A control to run the ipreallocated event */
4249 int32_t ctdb_control_ipreallocated(struct ctdb_context *ctdb,
4250 struct ctdb_req_control *c,
4254 struct ipreallocated_callback_state *state;
4256 state = talloc(ctdb, struct ipreallocated_callback_state);
4257 CTDB_NO_MEMORY(ctdb, state);
4259 DEBUG(DEBUG_INFO,(__location__ " Running \"ipreallocated\" event\n"));
4261 ret = ctdb_event_script_callback(ctdb, state,
4262 ctdb_ipreallocated_callback, state,
4263 false, CTDB_EVENT_IPREALLOCATED,
4267 DEBUG(DEBUG_ERR,("Failed to run \"ipreallocated\" event \n"));
4272 /* tell the control that we will be reply asynchronously */
4273 state->c = talloc_steal(state, c);
4274 *async_reply = true;
4280 /* This function is called from the recovery daemon to verify that a remote
4281 node has the expected ip allocation.
4282 This is verified against ctdb->ip_tree
4284 int verify_remote_ip_allocation(struct ctdb_context *ctdb,
4285 struct ctdb_all_public_ips *ips,
4288 struct ctdb_public_ip_list *tmp_ip;
4291 if (ctdb->ip_tree == NULL) {
4292 /* dont know the expected allocation yet, assume remote node
4301 for (i=0; i<ips->num; i++) {
4302 tmp_ip = trbt_lookuparray32(ctdb->ip_tree, IP_KEYLEN, ip_key(&ips->ips[i].addr));
4303 if (tmp_ip == NULL) {
4304 DEBUG(DEBUG_ERR,("Node %u has new or unknown public IP %s\n", pnn, ctdb_addr_to_str(&ips->ips[i].addr)));
4308 if (tmp_ip->pnn == -1 || ips->ips[i].pnn == -1) {
4312 if (tmp_ip->pnn != ips->ips[i].pnn) {
4314 ("Inconsistent IP allocation - node %u thinks %s is held by node %u while it is assigned to node %u\n",
4316 ctdb_addr_to_str(&ips->ips[i].addr),
4317 ips->ips[i].pnn, tmp_ip->pnn));
4325 int update_ip_assignment_tree(struct ctdb_context *ctdb, struct ctdb_public_ip *ip)
4327 struct ctdb_public_ip_list *tmp_ip;
4329 if (ctdb->ip_tree == NULL) {
4330 DEBUG(DEBUG_ERR,("No ctdb->ip_tree yet. Failed to update ip assignment\n"));
4334 tmp_ip = trbt_lookuparray32(ctdb->ip_tree, IP_KEYLEN, ip_key(&ip->addr));
4335 if (tmp_ip == NULL) {
4336 DEBUG(DEBUG_ERR,(__location__ " Could not find record for address %s, update ip\n", ctdb_addr_to_str(&ip->addr)));
4340 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));
4341 tmp_ip->pnn = ip->pnn;
4347 struct ctdb_reloadips_handle {
4348 struct ctdb_context *ctdb;
4349 struct ctdb_req_control *c;
4353 struct fd_event *fde;
4356 static int ctdb_reloadips_destructor(struct ctdb_reloadips_handle *h)
4358 if (h == h->ctdb->reload_ips) {
4359 h->ctdb->reload_ips = NULL;
4362 ctdb_request_control_reply(h->ctdb, h->c, NULL, h->status, NULL);
4365 ctdb_kill(h->ctdb, h->child, SIGKILL);
4369 static void ctdb_reloadips_timeout_event(struct event_context *ev,
4370 struct timed_event *te,
4371 struct timeval t, void *private_data)
4373 struct ctdb_reloadips_handle *h = talloc_get_type(private_data, struct ctdb_reloadips_handle);
4378 static void ctdb_reloadips_child_handler(struct event_context *ev, struct fd_event *fde,
4379 uint16_t flags, void *private_data)
4381 struct ctdb_reloadips_handle *h = talloc_get_type(private_data, struct ctdb_reloadips_handle);
4386 ret = read(h->fd[0], &res, 1);
4387 if (ret < 1 || res != 0) {
4388 DEBUG(DEBUG_ERR, (__location__ " Reloadips child process returned error\n"));
4396 static int ctdb_reloadips_child(struct ctdb_context *ctdb)
4398 TALLOC_CTX *mem_ctx = talloc_new(NULL);
4399 struct ctdb_all_public_ips *ips;
4400 struct ctdb_vnn *vnn;
4401 struct client_async_data *async_data;
4402 struct timeval timeout;
4404 struct ctdb_client_control_state *state;
4408 CTDB_NO_MEMORY(ctdb, mem_ctx);
4410 /* Read IPs from local node */
4411 ret = ctdb_ctrl_get_public_ips(ctdb, TAKEOVER_TIMEOUT(),
4412 CTDB_CURRENT_NODE, mem_ctx, &ips);
4415 ("Unable to fetch public IPs from local node\n"));
4416 talloc_free(mem_ctx);
4420 /* Read IPs file - this is safe since this is a child process */
4422 if (ctdb_set_public_addresses(ctdb, false) != 0) {
4423 DEBUG(DEBUG_ERR,("Failed to re-read public addresses file\n"));
4424 talloc_free(mem_ctx);
4428 async_data = talloc_zero(mem_ctx, struct client_async_data);
4429 CTDB_NO_MEMORY(ctdb, async_data);
4431 /* Compare IPs between node and file for IPs to be deleted */
4432 for (i = 0; i < ips->num; i++) {
4434 for (vnn = ctdb->vnn; vnn; vnn = vnn->next) {
4435 if (ctdb_same_ip(&vnn->public_address,
4436 &ips->ips[i].addr)) {
4437 /* IP is still in file */
4443 /* Delete IP ips->ips[i] */
4444 struct ctdb_control_ip_iface *pub;
4447 ("IP %s no longer configured, deleting it\n",
4448 ctdb_addr_to_str(&ips->ips[i].addr)));
4450 pub = talloc_zero(mem_ctx,
4451 struct ctdb_control_ip_iface);
4452 CTDB_NO_MEMORY(ctdb, pub);
4454 pub->addr = ips->ips[i].addr;
4458 timeout = TAKEOVER_TIMEOUT();
4460 data.dsize = offsetof(struct ctdb_control_ip_iface,
4462 data.dptr = (uint8_t *)pub;
4464 state = ctdb_control_send(ctdb, CTDB_CURRENT_NODE, 0,
4465 CTDB_CONTROL_DEL_PUBLIC_IP,
4466 0, data, async_data,
4468 if (state == NULL) {
4471 " failed sending CTDB_CONTROL_DEL_PUBLIC_IP\n"));
4478 /* Compare IPs between node and file for IPs to be added */
4480 for (vnn = ctdb->vnn; vnn; vnn = vnn->next) {
4481 for (i = 0; i < ips->num; i++) {
4482 if (ctdb_same_ip(&vnn->public_address,
4483 &ips->ips[i].addr)) {
4484 /* IP already on node */
4488 if (i == ips->num) {
4489 /* Add IP ips->ips[i] */
4490 struct ctdb_control_ip_iface *pub;
4491 const char *ifaces = NULL;
4496 ("New IP %s configured, adding it\n",
4497 ctdb_addr_to_str(&vnn->public_address)));
4499 uint32_t pnn = ctdb_get_pnn(ctdb);
4501 data.dsize = sizeof(pnn);
4502 data.dptr = (uint8_t *)&pnn;
4504 ret = ctdb_client_send_message(
4506 CTDB_BROADCAST_CONNECTED,
4507 CTDB_SRVID_REBALANCE_NODE,
4510 DEBUG(DEBUG_WARNING,
4511 ("Failed to send message to force node reallocation - IPs may be unbalanced\n"));
4517 ifaces = vnn->ifaces[0];
4519 while (vnn->ifaces[iface] != NULL) {
4520 ifaces = talloc_asprintf(vnn, "%s,%s", ifaces,
4521 vnn->ifaces[iface]);
4525 len = strlen(ifaces) + 1;
4526 pub = talloc_zero_size(mem_ctx,
4527 offsetof(struct ctdb_control_ip_iface, iface) + len);
4528 CTDB_NO_MEMORY(ctdb, pub);
4530 pub->addr = vnn->public_address;
4531 pub->mask = vnn->public_netmask_bits;
4533 memcpy(&pub->iface[0], ifaces, pub->len);
4535 timeout = TAKEOVER_TIMEOUT();
4537 data.dsize = offsetof(struct ctdb_control_ip_iface,
4539 data.dptr = (uint8_t *)pub;
4541 state = ctdb_control_send(ctdb, CTDB_CURRENT_NODE, 0,
4542 CTDB_CONTROL_ADD_PUBLIC_IP,
4543 0, data, async_data,
4545 if (state == NULL) {
4548 " failed sending CTDB_CONTROL_ADD_PUBLIC_IP\n"));
4554 if (ctdb_client_async_wait(ctdb, async_data) != 0) {
4555 DEBUG(DEBUG_ERR,(__location__ " Add/delete IPs failed\n"));
4559 talloc_free(mem_ctx);
4563 talloc_free(mem_ctx);
4567 /* This control is sent to force the node to re-read the public addresses file
4568 and drop any addresses we should nnot longer host, and add new addresses
4569 that we are now able to host
4571 int32_t ctdb_control_reload_public_ips(struct ctdb_context *ctdb, struct ctdb_req_control *c, bool *async_reply)
4573 struct ctdb_reloadips_handle *h;
4574 pid_t parent = getpid();
4576 if (ctdb->reload_ips != NULL) {
4577 talloc_free(ctdb->reload_ips);
4578 ctdb->reload_ips = NULL;
4581 h = talloc(ctdb, struct ctdb_reloadips_handle);
4582 CTDB_NO_MEMORY(ctdb, h);
4587 if (pipe(h->fd) == -1) {
4588 DEBUG(DEBUG_ERR,("Failed to create pipe for ctdb_freeze_lock\n"));
4593 h->child = ctdb_fork(ctdb);
4594 if (h->child == (pid_t)-1) {
4595 DEBUG(DEBUG_ERR, ("Failed to fork a child for reloadips\n"));
4603 if (h->child == 0) {
4604 signed char res = 0;
4607 debug_extra = talloc_asprintf(NULL, "reloadips:");
4609 ctdb_set_process_name("ctdb_reloadips");
4610 if (switch_from_server_to_client(ctdb, "reloadips-child") != 0) {
4611 DEBUG(DEBUG_CRIT,("ERROR: Failed to switch reloadips child into client mode\n"));
4614 res = ctdb_reloadips_child(ctdb);
4616 DEBUG(DEBUG_ERR,("Failed to reload ips on local node\n"));
4620 write(h->fd[1], &res, 1);
4621 /* make sure we die when our parent dies */
4622 while (ctdb_kill(ctdb, parent, 0) == 0 || errno != ESRCH) {
4628 h->c = talloc_steal(h, c);
4631 set_close_on_exec(h->fd[0]);
4633 talloc_set_destructor(h, ctdb_reloadips_destructor);
4636 h->fde = event_add_fd(ctdb->ev, h, h->fd[0],
4637 EVENT_FD_READ, ctdb_reloadips_child_handler,
4639 tevent_fd_set_auto_close(h->fde);
4641 event_add_timed(ctdb->ev, h,
4642 timeval_current_ofs(120, 0),
4643 ctdb_reloadips_timeout_event, h);
4645 /* we reply later */
4646 *async_reply = true;