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);
1138 static void ctdb_check_interfaces_event(struct event_context *ev, struct timed_event *te,
1139 struct timeval t, void *private_data)
1141 struct ctdb_context *ctdb = talloc_get_type(private_data,
1142 struct ctdb_context);
1143 struct ctdb_vnn *vnn;
1145 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
1148 for (i=0; vnn->ifaces[i] != NULL; i++) {
1149 if (!ctdb_sys_check_iface_exists(vnn->ifaces[i])) {
1150 DEBUG(DEBUG_CRIT,("Interface %s does not exist but is used by public ip %s\n",
1152 ctdb_addr_to_str(&vnn->public_address)));
1157 event_add_timed(ctdb->ev, ctdb->check_public_ifaces_ctx,
1158 timeval_current_ofs(30, 0),
1159 ctdb_check_interfaces_event, ctdb);
1163 int ctdb_start_monitoring_interfaces(struct ctdb_context *ctdb)
1165 if (ctdb->check_public_ifaces_ctx != NULL) {
1166 talloc_free(ctdb->check_public_ifaces_ctx);
1167 ctdb->check_public_ifaces_ctx = NULL;
1170 ctdb->check_public_ifaces_ctx = talloc_new(ctdb);
1171 if (ctdb->check_public_ifaces_ctx == NULL) {
1172 ctdb_fatal(ctdb, "failed to allocate context for checking interfaces");
1175 event_add_timed(ctdb->ev, ctdb->check_public_ifaces_ctx,
1176 timeval_current_ofs(30, 0),
1177 ctdb_check_interfaces_event, ctdb);
1184 setup the public address lists from a file
1186 int ctdb_set_public_addresses(struct ctdb_context *ctdb, bool check_addresses)
1192 lines = file_lines_load(ctdb->public_addresses_file, &nlines, ctdb);
1193 if (lines == NULL) {
1194 ctdb_set_error(ctdb, "Failed to load public address list '%s'\n", ctdb->public_addresses_file);
1197 while (nlines > 0 && strcmp(lines[nlines-1], "") == 0) {
1201 for (i=0;i<nlines;i++) {
1203 ctdb_sock_addr addr;
1204 const char *addrstr;
1209 while ((*line == ' ') || (*line == '\t')) {
1215 if (strcmp(line, "") == 0) {
1218 tok = strtok(line, " \t");
1220 tok = strtok(NULL, " \t");
1222 if (NULL == ctdb->default_public_interface) {
1223 DEBUG(DEBUG_CRIT,("No default public interface and no interface specified at line %u of public address list\n",
1228 ifaces = ctdb->default_public_interface;
1233 if (!addrstr || !parse_ip_mask(addrstr, ifaces, &addr, &mask)) {
1234 DEBUG(DEBUG_CRIT,("Badly formed line %u in public address list\n", i+1));
1238 if (ctdb_add_public_address(ctdb, &addr, mask, ifaces, check_addresses)) {
1239 DEBUG(DEBUG_CRIT,("Failed to add line %u to the public address list\n", i+1));
1250 int ctdb_set_single_public_ip(struct ctdb_context *ctdb,
1254 struct ctdb_vnn *svnn;
1255 struct ctdb_iface *cur = NULL;
1259 svnn = talloc_zero(ctdb, struct ctdb_vnn);
1260 CTDB_NO_MEMORY(ctdb, svnn);
1262 svnn->ifaces = talloc_array(svnn, const char *, 2);
1263 CTDB_NO_MEMORY(ctdb, svnn->ifaces);
1264 svnn->ifaces[0] = talloc_strdup(svnn->ifaces, iface);
1265 CTDB_NO_MEMORY(ctdb, svnn->ifaces[0]);
1266 svnn->ifaces[1] = NULL;
1268 ok = parse_ip(ip, iface, 0, &svnn->public_address);
1274 ret = ctdb_add_local_iface(ctdb, svnn->ifaces[0]);
1276 DEBUG(DEBUG_CRIT, (__location__ " failed to add iface[%s] "
1277 "for single_ip[%s]\n",
1279 ctdb_addr_to_str(&svnn->public_address)));
1284 /* assume the single public ip interface is initially "good" */
1285 cur = ctdb_find_iface(ctdb, iface);
1287 DEBUG(DEBUG_CRIT,("Can not find public interface %s used by --single-public-ip", iface));
1290 cur->link_up = true;
1292 ret = ctdb_vnn_assign_iface(ctdb, svnn);
1298 ctdb->single_ip_vnn = svnn;
1302 struct ctdb_public_ip_list {
1303 struct ctdb_public_ip_list *next;
1305 ctdb_sock_addr addr;
1308 /* Given a physical node, return the number of
1309 public addresses that is currently assigned to this node.
1311 static int node_ip_coverage(struct ctdb_context *ctdb,
1313 struct ctdb_public_ip_list *ips)
1317 for (;ips;ips=ips->next) {
1318 if (ips->pnn == pnn) {
1326 /* Can the given node host the given IP: is the public IP known to the
1327 * node and is NOIPHOST unset?
1329 static bool can_node_host_ip(struct ctdb_context *ctdb, int32_t pnn,
1330 struct ctdb_ipflags ipflags,
1331 struct ctdb_public_ip_list *ip)
1333 struct ctdb_all_public_ips *public_ips;
1336 if (ipflags.noiphost) {
1340 public_ips = ctdb->nodes[pnn]->available_public_ips;
1342 if (public_ips == NULL) {
1346 for (i=0; i<public_ips->num; i++) {
1347 if (ctdb_same_ip(&ip->addr, &public_ips->ips[i].addr)) {
1348 /* yes, this node can serve this public ip */
1356 static bool can_node_takeover_ip(struct ctdb_context *ctdb, int32_t pnn,
1357 struct ctdb_ipflags ipflags,
1358 struct ctdb_public_ip_list *ip)
1360 if (ipflags.noiptakeover) {
1364 return can_node_host_ip(ctdb, pnn, ipflags, ip);
1367 /* search the node lists list for a node to takeover this ip.
1368 pick the node that currently are serving the least number of ips
1369 so that the ips get spread out evenly.
1371 static int find_takeover_node(struct ctdb_context *ctdb,
1372 struct ctdb_ipflags *ipflags,
1373 struct ctdb_public_ip_list *ip,
1374 struct ctdb_public_ip_list *all_ips)
1376 int pnn, min=0, num;
1379 numnodes = talloc_array_length(ipflags);
1381 for (i=0; i<numnodes; i++) {
1382 /* verify that this node can serve this ip */
1383 if (!can_node_takeover_ip(ctdb, i, ipflags[i], ip)) {
1384 /* no it couldnt so skip to the next node */
1388 num = node_ip_coverage(ctdb, i, all_ips);
1389 /* was this the first node we checked ? */
1401 DEBUG(DEBUG_WARNING,(__location__ " Could not find node to take over public address '%s'\n",
1402 ctdb_addr_to_str(&ip->addr)));
1412 static uint32_t *ip_key(ctdb_sock_addr *ip)
1414 static uint32_t key[IP_KEYLEN];
1416 bzero(key, sizeof(key));
1418 switch (ip->sa.sa_family) {
1420 key[3] = htonl(ip->ip.sin_addr.s_addr);
1423 uint32_t *s6_a32 = (uint32_t *)&(ip->ip6.sin6_addr.s6_addr);
1424 key[0] = htonl(s6_a32[0]);
1425 key[1] = htonl(s6_a32[1]);
1426 key[2] = htonl(s6_a32[2]);
1427 key[3] = htonl(s6_a32[3]);
1431 DEBUG(DEBUG_ERR, (__location__ " ERROR, unknown family passed :%u\n", ip->sa.sa_family));
1438 static void *add_ip_callback(void *parm, void *data)
1440 struct ctdb_public_ip_list *this_ip = parm;
1441 struct ctdb_public_ip_list *prev_ip = data;
1443 if (prev_ip == NULL) {
1446 if (this_ip->pnn == -1) {
1447 this_ip->pnn = prev_ip->pnn;
1453 static int getips_count_callback(void *param, void *data)
1455 struct ctdb_public_ip_list **ip_list = (struct ctdb_public_ip_list **)param;
1456 struct ctdb_public_ip_list *new_ip = (struct ctdb_public_ip_list *)data;
1458 new_ip->next = *ip_list;
1463 static struct ctdb_public_ip_list *
1464 create_merged_ip_list(struct ctdb_context *ctdb)
1467 struct ctdb_public_ip_list *ip_list;
1468 struct ctdb_all_public_ips *public_ips;
1470 if (ctdb->ip_tree != NULL) {
1471 talloc_free(ctdb->ip_tree);
1472 ctdb->ip_tree = NULL;
1474 ctdb->ip_tree = trbt_create(ctdb, 0);
1476 for (i=0;i<ctdb->num_nodes;i++) {
1477 public_ips = ctdb->nodes[i]->known_public_ips;
1479 if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
1483 /* there were no public ips for this node */
1484 if (public_ips == NULL) {
1488 for (j=0;j<public_ips->num;j++) {
1489 struct ctdb_public_ip_list *tmp_ip;
1491 tmp_ip = talloc_zero(ctdb->ip_tree, struct ctdb_public_ip_list);
1492 CTDB_NO_MEMORY_NULL(ctdb, tmp_ip);
1493 /* Do not use information about IP addresses hosted
1494 * on other nodes, it may not be accurate */
1495 if (public_ips->ips[j].pnn == ctdb->nodes[i]->pnn) {
1496 tmp_ip->pnn = public_ips->ips[j].pnn;
1500 tmp_ip->addr = public_ips->ips[j].addr;
1501 tmp_ip->next = NULL;
1503 trbt_insertarray32_callback(ctdb->ip_tree,
1504 IP_KEYLEN, ip_key(&public_ips->ips[j].addr),
1511 trbt_traversearray32(ctdb->ip_tree, IP_KEYLEN, getips_count_callback, &ip_list);
1517 * This is the length of the longtest common prefix between the IPs.
1518 * It is calculated by XOR-ing the 2 IPs together and counting the
1519 * number of leading zeroes. The implementation means that all
1520 * addresses end up being 128 bits long.
1522 * FIXME? Should we consider IPv4 and IPv6 separately given that the
1523 * 12 bytes of 0 prefix padding will hurt the algorithm if there are
1524 * lots of nodes and IP addresses?
1526 static uint32_t ip_distance(ctdb_sock_addr *ip1, ctdb_sock_addr *ip2)
1528 uint32_t ip1_k[IP_KEYLEN];
1533 uint32_t distance = 0;
1535 memcpy(ip1_k, ip_key(ip1), sizeof(ip1_k));
1537 for (i=0; i<IP_KEYLEN; i++) {
1538 x = ip1_k[i] ^ t[i];
1542 /* Count number of leading zeroes.
1543 * FIXME? This could be optimised...
1545 while ((x & (1 << 31)) == 0) {
1555 /* Calculate the IP distance for the given IP relative to IPs on the
1556 given node. The ips argument is generally the all_ips variable
1557 used in the main part of the algorithm.
1559 static uint32_t ip_distance_2_sum(ctdb_sock_addr *ip,
1560 struct ctdb_public_ip_list *ips,
1563 struct ctdb_public_ip_list *t;
1568 for (t=ips; t != NULL; t=t->next) {
1569 if (t->pnn != pnn) {
1573 /* Optimisation: We never calculate the distance
1574 * between an address and itself. This allows us to
1575 * calculate the effect of removing an address from a
1576 * node by simply calculating the distance between
1577 * that address and all of the exitsing addresses.
1578 * Moreover, we assume that we're only ever dealing
1579 * with addresses from all_ips so we can identify an
1580 * address via a pointer rather than doing a more
1581 * expensive address comparison. */
1582 if (&(t->addr) == ip) {
1586 d = ip_distance(ip, &(t->addr));
1587 sum += d * d; /* Cheaper than pulling in math.h :-) */
1593 /* Return the LCP2 imbalance metric for addresses currently assigned
1596 static uint32_t lcp2_imbalance(struct ctdb_public_ip_list * all_ips, int pnn)
1598 struct ctdb_public_ip_list *t;
1600 uint32_t imbalance = 0;
1602 for (t=all_ips; t!=NULL; t=t->next) {
1603 if (t->pnn != pnn) {
1606 /* Pass the rest of the IPs rather than the whole
1609 imbalance += ip_distance_2_sum(&(t->addr), t->next, pnn);
1615 /* Allocate any unassigned IPs just by looping through the IPs and
1616 * finding the best node for each.
1618 static void basic_allocate_unassigned(struct ctdb_context *ctdb,
1619 struct ctdb_ipflags *ipflags,
1620 struct ctdb_public_ip_list *all_ips)
1622 struct ctdb_public_ip_list *tmp_ip;
1624 /* loop over all ip's and find a physical node to cover for
1627 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1628 if (tmp_ip->pnn == -1) {
1629 if (find_takeover_node(ctdb, ipflags, tmp_ip, all_ips)) {
1630 DEBUG(DEBUG_WARNING,("Failed to find node to cover ip %s\n",
1631 ctdb_addr_to_str(&tmp_ip->addr)));
1637 /* Basic non-deterministic rebalancing algorithm.
1639 static void basic_failback(struct ctdb_context *ctdb,
1640 struct ctdb_ipflags *ipflags,
1641 struct ctdb_public_ip_list *all_ips,
1645 int maxnode, maxnum, minnode, minnum, num, retries;
1646 struct ctdb_public_ip_list *tmp_ip;
1648 numnodes = talloc_array_length(ipflags);
1655 /* for each ip address, loop over all nodes that can serve
1656 this ip and make sure that the difference between the node
1657 serving the most and the node serving the least ip's are
1660 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1661 if (tmp_ip->pnn == -1) {
1665 /* Get the highest and lowest number of ips's served by any
1666 valid node which can serve this ip.
1670 for (i=0; i<numnodes; i++) {
1671 /* only check nodes that can actually serve this ip */
1672 if (!can_node_takeover_ip(ctdb, i, ipflags[i], tmp_ip)) {
1673 /* no it couldnt so skip to the next node */
1677 num = node_ip_coverage(ctdb, i, all_ips);
1678 if (maxnode == -1) {
1687 if (minnode == -1) {
1697 if (maxnode == -1) {
1698 DEBUG(DEBUG_WARNING,(__location__ " Could not find maxnode. May not be able to serve ip '%s'\n",
1699 ctdb_addr_to_str(&tmp_ip->addr)));
1704 /* if the spread between the smallest and largest coverage by
1705 a node is >=2 we steal one of the ips from the node with
1706 most coverage to even things out a bit.
1707 try to do this a limited number of times since we dont
1708 want to spend too much time balancing the ip coverage.
1710 if ( (maxnum > minnum+1)
1711 && (retries < (num_ips + 5)) ){
1712 struct ctdb_public_ip_list *tmp;
1714 /* Reassign one of maxnode's VNNs */
1715 for (tmp=all_ips;tmp;tmp=tmp->next) {
1716 if (tmp->pnn == maxnode) {
1717 (void)find_takeover_node(ctdb, ipflags, tmp, all_ips);
1726 static void lcp2_init(struct ctdb_context *tmp_ctx,
1727 struct ctdb_ipflags *ipflags,
1728 struct ctdb_public_ip_list *all_ips,
1729 uint32_t *force_rebalance_nodes,
1730 uint32_t **lcp2_imbalances,
1731 bool **rebalance_candidates)
1734 struct ctdb_public_ip_list *tmp_ip;
1736 numnodes = talloc_array_length(ipflags);
1738 *rebalance_candidates = talloc_array(tmp_ctx, bool, numnodes);
1739 CTDB_NO_MEMORY_FATAL(tmp_ctx, *rebalance_candidates);
1740 *lcp2_imbalances = talloc_array(tmp_ctx, uint32_t, numnodes);
1741 CTDB_NO_MEMORY_FATAL(tmp_ctx, *lcp2_imbalances);
1743 for (i=0; i<numnodes; i++) {
1744 (*lcp2_imbalances)[i] = lcp2_imbalance(all_ips, i);
1745 /* First step: assume all nodes are candidates */
1746 (*rebalance_candidates)[i] = true;
1749 /* 2nd step: if a node has IPs assigned then it must have been
1750 * healthy before, so we remove it from consideration. This
1751 * is overkill but is all we have because we don't maintain
1752 * state between takeover runs. An alternative would be to
1753 * keep state and invalidate it every time the recovery master
1756 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1757 if (tmp_ip->pnn != -1) {
1758 (*rebalance_candidates)[tmp_ip->pnn] = false;
1762 /* 3rd step: if a node is forced to re-balance then
1763 we allow failback onto the node */
1764 if (force_rebalance_nodes == NULL) {
1767 for (i = 0; i < talloc_array_length(force_rebalance_nodes); i++) {
1768 uint32_t pnn = force_rebalance_nodes[i];
1769 if (pnn >= numnodes) {
1771 (__location__ "unknown node %u\n", pnn));
1776 ("Forcing rebalancing of IPs to node %u\n", pnn));
1777 (*rebalance_candidates)[pnn] = true;
1781 /* Allocate any unassigned addresses using the LCP2 algorithm to find
1782 * the IP/node combination that will cost the least.
1784 static void lcp2_allocate_unassigned(struct ctdb_context *ctdb,
1785 struct ctdb_ipflags *ipflags,
1786 struct ctdb_public_ip_list *all_ips,
1787 uint32_t *lcp2_imbalances)
1789 struct ctdb_public_ip_list *tmp_ip;
1790 int dstnode, numnodes;
1793 uint32_t mindsum, dstdsum, dstimbl, minimbl;
1794 struct ctdb_public_ip_list *minip;
1796 bool should_loop = true;
1797 bool have_unassigned = true;
1799 numnodes = talloc_array_length(ipflags);
1801 while (have_unassigned && should_loop) {
1802 should_loop = false;
1804 DEBUG(DEBUG_DEBUG,(" ----------------------------------------\n"));
1805 DEBUG(DEBUG_DEBUG,(" CONSIDERING MOVES (UNASSIGNED)\n"));
1811 /* loop over each unassigned ip. */
1812 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1813 if (tmp_ip->pnn != -1) {
1817 for (dstnode=0; dstnode<numnodes; dstnode++) {
1818 /* only check nodes that can actually takeover this ip */
1819 if (!can_node_takeover_ip(ctdb, dstnode,
1822 /* no it couldnt so skip to the next node */
1826 dstdsum = ip_distance_2_sum(&(tmp_ip->addr), all_ips, dstnode);
1827 dstimbl = lcp2_imbalances[dstnode] + dstdsum;
1828 DEBUG(DEBUG_DEBUG,(" %s -> %d [+%d]\n",
1829 ctdb_addr_to_str(&(tmp_ip->addr)),
1831 dstimbl - lcp2_imbalances[dstnode]));
1834 if ((minnode == -1) || (dstdsum < mindsum)) {
1844 DEBUG(DEBUG_DEBUG,(" ----------------------------------------\n"));
1846 /* If we found one then assign it to the given node. */
1847 if (minnode != -1) {
1848 minip->pnn = minnode;
1849 lcp2_imbalances[minnode] = minimbl;
1850 DEBUG(DEBUG_INFO,(" %s -> %d [+%d]\n",
1851 ctdb_addr_to_str(&(minip->addr)),
1856 /* There might be a better way but at least this is clear. */
1857 have_unassigned = false;
1858 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1859 if (tmp_ip->pnn == -1) {
1860 have_unassigned = true;
1865 /* We know if we have an unassigned addresses so we might as
1868 if (have_unassigned) {
1869 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
1870 if (tmp_ip->pnn == -1) {
1871 DEBUG(DEBUG_WARNING,("Failed to find node to cover ip %s\n",
1872 ctdb_addr_to_str(&tmp_ip->addr)));
1878 /* LCP2 algorithm for rebalancing the cluster. Given a candidate node
1879 * to move IPs from, determines the best IP/destination node
1880 * combination to move from the source node.
1882 static bool lcp2_failback_candidate(struct ctdb_context *ctdb,
1883 struct ctdb_ipflags *ipflags,
1884 struct ctdb_public_ip_list *all_ips,
1887 uint32_t *lcp2_imbalances,
1888 bool *rebalance_candidates)
1890 int dstnode, mindstnode, numnodes;
1891 uint32_t srcimbl, srcdsum, dstimbl, dstdsum;
1892 uint32_t minsrcimbl, mindstimbl;
1893 struct ctdb_public_ip_list *minip;
1894 struct ctdb_public_ip_list *tmp_ip;
1896 /* Find an IP and destination node that best reduces imbalance. */
1903 numnodes = talloc_array_length(ipflags);
1905 DEBUG(DEBUG_DEBUG,(" ----------------------------------------\n"));
1906 DEBUG(DEBUG_DEBUG,(" CONSIDERING MOVES FROM %d [%d]\n", srcnode, candimbl));
1908 for (tmp_ip=all_ips; tmp_ip; tmp_ip=tmp_ip->next) {
1909 /* Only consider addresses on srcnode. */
1910 if (tmp_ip->pnn != srcnode) {
1914 /* What is this IP address costing the source node? */
1915 srcdsum = ip_distance_2_sum(&(tmp_ip->addr), all_ips, srcnode);
1916 srcimbl = candimbl - srcdsum;
1918 /* Consider this IP address would cost each potential
1919 * destination node. Destination nodes are limited to
1920 * those that are newly healthy, since we don't want
1921 * to do gratuitous failover of IPs just to make minor
1922 * balance improvements.
1924 for (dstnode=0; dstnode<numnodes; dstnode++) {
1925 if (!rebalance_candidates[dstnode]) {
1929 /* only check nodes that can actually takeover this ip */
1930 if (!can_node_takeover_ip(ctdb, dstnode,
1931 ipflags[dstnode], tmp_ip)) {
1932 /* no it couldnt so skip to the next node */
1936 dstdsum = ip_distance_2_sum(&(tmp_ip->addr), all_ips, dstnode);
1937 dstimbl = lcp2_imbalances[dstnode] + dstdsum;
1938 DEBUG(DEBUG_DEBUG,(" %d [%d] -> %s -> %d [+%d]\n",
1939 srcnode, srcimbl - lcp2_imbalances[srcnode],
1940 ctdb_addr_to_str(&(tmp_ip->addr)),
1941 dstnode, dstimbl - lcp2_imbalances[dstnode]));
1943 if ((dstimbl < candimbl) && (dstdsum < srcdsum) && \
1944 ((mindstnode == -1) || \
1945 ((srcimbl + dstimbl) < (minsrcimbl + mindstimbl)))) {
1948 minsrcimbl = srcimbl;
1949 mindstnode = dstnode;
1950 mindstimbl = dstimbl;
1954 DEBUG(DEBUG_DEBUG,(" ----------------------------------------\n"));
1956 if (mindstnode != -1) {
1957 /* We found a move that makes things better... */
1958 DEBUG(DEBUG_INFO,("%d [%d] -> %s -> %d [+%d]\n",
1959 srcnode, minsrcimbl - lcp2_imbalances[srcnode],
1960 ctdb_addr_to_str(&(minip->addr)),
1961 mindstnode, mindstimbl - lcp2_imbalances[mindstnode]));
1964 lcp2_imbalances[srcnode] = srcimbl;
1965 lcp2_imbalances[mindstnode] = mindstimbl;
1966 minip->pnn = mindstnode;
1975 struct lcp2_imbalance_pnn {
1980 static int lcp2_cmp_imbalance_pnn(const void * a, const void * b)
1982 const struct lcp2_imbalance_pnn * lipa = (const struct lcp2_imbalance_pnn *) a;
1983 const struct lcp2_imbalance_pnn * lipb = (const struct lcp2_imbalance_pnn *) b;
1985 if (lipa->imbalance > lipb->imbalance) {
1987 } else if (lipa->imbalance == lipb->imbalance) {
1994 /* LCP2 algorithm for rebalancing the cluster. This finds the source
1995 * node with the highest LCP2 imbalance, and then determines the best
1996 * IP/destination node combination to move from the source node.
1998 static void lcp2_failback(struct ctdb_context *ctdb,
1999 struct ctdb_ipflags *ipflags,
2000 struct ctdb_public_ip_list *all_ips,
2001 uint32_t *lcp2_imbalances,
2002 bool *rebalance_candidates)
2004 int i, num_rebalance_candidates, numnodes;
2005 struct lcp2_imbalance_pnn * lips;
2008 numnodes = talloc_array_length(ipflags);
2012 /* It is only worth continuing if we have suitable target
2013 * nodes to transfer IPs to. This check is much cheaper than
2016 num_rebalance_candidates = 0;
2017 for (i=0; i<numnodes; i++) {
2018 if (rebalance_candidates[i]) {
2019 num_rebalance_candidates++;
2022 if (num_rebalance_candidates == 0) {
2026 /* Put the imbalances and nodes into an array, sort them and
2027 * iterate through candidates. Usually the 1st one will be
2028 * used, so this doesn't cost much...
2030 lips = talloc_array(ctdb, struct lcp2_imbalance_pnn, numnodes);
2031 for (i=0; i<numnodes; i++) {
2032 lips[i].imbalance = lcp2_imbalances[i];
2035 qsort(lips, numnodes, sizeof(struct lcp2_imbalance_pnn),
2036 lcp2_cmp_imbalance_pnn);
2039 for (i=0; i<numnodes; i++) {
2040 /* This means that all nodes had 0 or 1 addresses, so
2041 * can't be imbalanced.
2043 if (lips[i].imbalance == 0) {
2047 if (lcp2_failback_candidate(ctdb,
2053 rebalance_candidates)) {
2065 static void unassign_unsuitable_ips(struct ctdb_context *ctdb,
2066 struct ctdb_ipflags *ipflags,
2067 struct ctdb_public_ip_list *all_ips)
2069 struct ctdb_public_ip_list *tmp_ip;
2071 /* verify that the assigned nodes can serve that public ip
2072 and set it to -1 if not
2074 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
2075 if (tmp_ip->pnn == -1) {
2078 if (!can_node_host_ip(ctdb, tmp_ip->pnn,
2079 ipflags[tmp_ip->pnn], tmp_ip) != 0) {
2080 /* this node can not serve this ip. */
2081 DEBUG(DEBUG_DEBUG,("Unassign IP: %s from %d\n",
2082 ctdb_addr_to_str(&(tmp_ip->addr)),
2089 static void ip_alloc_deterministic_ips(struct ctdb_context *ctdb,
2090 struct ctdb_ipflags *ipflags,
2091 struct ctdb_public_ip_list *all_ips)
2093 struct ctdb_public_ip_list *tmp_ip;
2096 numnodes = talloc_array_length(ipflags);
2098 DEBUG(DEBUG_NOTICE,("Deterministic IPs enabled. Resetting all ip allocations\n"));
2099 /* Allocate IPs to nodes in a modulo fashion so that IPs will
2100 * always be allocated the same way for a specific set of
2101 * available/unavailable nodes.
2104 for (i=0,tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next,i++) {
2105 tmp_ip->pnn = i % numnodes;
2108 /* IP failback doesn't make sense with deterministic
2109 * IPs, since the modulo step above implicitly fails
2110 * back IPs to their "home" node.
2112 if (1 == ctdb->tunable.no_ip_failback) {
2113 DEBUG(DEBUG_WARNING, ("WARNING: 'NoIPFailback' set but ignored - incompatible with 'DeterministicIPs\n"));
2116 unassign_unsuitable_ips(ctdb, ipflags, all_ips);
2118 basic_allocate_unassigned(ctdb, ipflags, all_ips);
2120 /* No failback here! */
2123 static void ip_alloc_nondeterministic_ips(struct ctdb_context *ctdb,
2124 struct ctdb_ipflags *ipflags,
2125 struct ctdb_public_ip_list *all_ips)
2127 /* This should be pushed down into basic_failback. */
2128 struct ctdb_public_ip_list *tmp_ip;
2130 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
2134 unassign_unsuitable_ips(ctdb, ipflags, all_ips);
2136 basic_allocate_unassigned(ctdb, ipflags, all_ips);
2138 /* If we don't want IPs to fail back then don't rebalance IPs. */
2139 if (1 == ctdb->tunable.no_ip_failback) {
2143 /* Now, try to make sure the ip adresses are evenly distributed
2146 basic_failback(ctdb, ipflags, all_ips, num_ips);
2149 static void ip_alloc_lcp2(struct ctdb_context *ctdb,
2150 struct ctdb_ipflags *ipflags,
2151 struct ctdb_public_ip_list *all_ips,
2152 uint32_t *force_rebalance_nodes)
2154 uint32_t *lcp2_imbalances;
2155 bool *rebalance_candidates;
2157 TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
2159 unassign_unsuitable_ips(ctdb, ipflags, all_ips);
2161 lcp2_init(tmp_ctx, ipflags, all_ips,force_rebalance_nodes,
2162 &lcp2_imbalances, &rebalance_candidates);
2164 lcp2_allocate_unassigned(ctdb, ipflags, all_ips, lcp2_imbalances);
2166 /* If we don't want IPs to fail back then don't rebalance IPs. */
2167 if (1 == ctdb->tunable.no_ip_failback) {
2171 /* Now, try to make sure the ip adresses are evenly distributed
2174 lcp2_failback(ctdb, ipflags, all_ips,
2175 lcp2_imbalances, rebalance_candidates);
2178 talloc_free(tmp_ctx);
2181 static bool all_nodes_are_disabled(struct ctdb_node_map *nodemap)
2185 /* Count how many completely healthy nodes we have */
2187 for (i=0;i<nodemap->num;i++) {
2188 if (!(nodemap->nodes[i].flags & (NODE_FLAGS_INACTIVE|NODE_FLAGS_DISABLED))) {
2193 return num_healthy == 0;
2196 /* The calculation part of the IP allocation algorithm. */
2197 static void ctdb_takeover_run_core(struct ctdb_context *ctdb,
2198 struct ctdb_ipflags *ipflags,
2199 struct ctdb_public_ip_list **all_ips_p,
2200 uint32_t *force_rebalance_nodes)
2202 /* since nodes only know about those public addresses that
2203 can be served by that particular node, no single node has
2204 a full list of all public addresses that exist in the cluster.
2205 Walk over all node structures and create a merged list of
2206 all public addresses that exist in the cluster.
2208 keep the tree of ips around as ctdb->ip_tree
2210 *all_ips_p = create_merged_ip_list(ctdb);
2212 if (1 == ctdb->tunable.lcp2_public_ip_assignment) {
2213 ip_alloc_lcp2(ctdb, ipflags, *all_ips_p, force_rebalance_nodes);
2214 } else if (1 == ctdb->tunable.deterministic_public_ips) {
2215 ip_alloc_deterministic_ips(ctdb, ipflags, *all_ips_p);
2217 ip_alloc_nondeterministic_ips(ctdb, ipflags, *all_ips_p);
2220 /* at this point ->pnn is the node which will own each IP
2221 or -1 if there is no node that can cover this ip
2227 struct get_tunable_callback_data {
2228 const char *tunable;
2233 static void get_tunable_callback(struct ctdb_context *ctdb, uint32_t pnn,
2234 int32_t res, TDB_DATA outdata,
2237 struct get_tunable_callback_data *cd =
2238 (struct get_tunable_callback_data *)callback;
2242 /* Already handled in fail callback */
2246 if (outdata.dsize != sizeof(uint32_t)) {
2247 DEBUG(DEBUG_ERR,("Wrong size of returned data when reading \"%s\" tunable from node %d. Expected %d bytes but received %d bytes\n",
2248 cd->tunable, pnn, (int)sizeof(uint32_t),
2249 (int)outdata.dsize));
2254 size = talloc_array_length(cd->out);
2256 DEBUG(DEBUG_ERR,("Got %s reply from node %d but nodemap only has %d entries\n",
2257 cd->tunable, pnn, size));
2262 cd->out[pnn] = *(uint32_t *)outdata.dptr;
2265 static void get_tunable_fail_callback(struct ctdb_context *ctdb, uint32_t pnn,
2266 int32_t res, TDB_DATA outdata,
2269 struct get_tunable_callback_data *cd =
2270 (struct get_tunable_callback_data *)callback;
2275 ("Timed out getting tunable \"%s\" from node %d\n",
2281 DEBUG(DEBUG_WARNING,
2282 ("Tunable \"%s\" not implemented on node %d\n",
2287 ("Unexpected error getting tunable \"%s\" from node %d\n",
2293 static uint32_t *get_tunable_from_nodes(struct ctdb_context *ctdb,
2294 TALLOC_CTX *tmp_ctx,
2295 struct ctdb_node_map *nodemap,
2296 const char *tunable,
2297 uint32_t default_value)
2300 struct ctdb_control_get_tunable *t;
2303 struct get_tunable_callback_data callback_data;
2306 tvals = talloc_array(tmp_ctx, uint32_t, nodemap->num);
2307 CTDB_NO_MEMORY_NULL(ctdb, tvals);
2308 for (i=0; i<nodemap->num; i++) {
2309 tvals[i] = default_value;
2312 callback_data.out = tvals;
2313 callback_data.tunable = tunable;
2314 callback_data.fatal = false;
2316 data.dsize = offsetof(struct ctdb_control_get_tunable, name) + strlen(tunable) + 1;
2317 data.dptr = talloc_size(tmp_ctx, data.dsize);
2318 t = (struct ctdb_control_get_tunable *)data.dptr;
2319 t->length = strlen(tunable)+1;
2320 memcpy(t->name, tunable, t->length);
2321 nodes = list_of_connected_nodes(ctdb, nodemap, tmp_ctx, true);
2322 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_GET_TUNABLE,
2323 nodes, 0, TAKEOVER_TIMEOUT(),
2325 get_tunable_callback,
2326 get_tunable_fail_callback,
2327 &callback_data) != 0) {
2328 if (callback_data.fatal) {
2334 talloc_free(data.dptr);
2339 struct get_runstate_callback_data {
2340 enum ctdb_runstate *out;
2344 static void get_runstate_callback(struct ctdb_context *ctdb, uint32_t pnn,
2345 int32_t res, TDB_DATA outdata,
2346 void *callback_data)
2348 struct get_runstate_callback_data *cd =
2349 (struct get_runstate_callback_data *)callback_data;
2353 /* Already handled in fail callback */
2357 if (outdata.dsize != sizeof(uint32_t)) {
2358 DEBUG(DEBUG_ERR,("Wrong size of returned data when getting runstate from node %d. Expected %d bytes but received %d bytes\n",
2359 pnn, (int)sizeof(uint32_t),
2360 (int)outdata.dsize));
2365 size = talloc_array_length(cd->out);
2367 DEBUG(DEBUG_ERR,("Got reply from node %d but nodemap only has %d entries\n",
2372 cd->out[pnn] = (enum ctdb_runstate)*(uint32_t *)outdata.dptr;
2375 static void get_runstate_fail_callback(struct ctdb_context *ctdb, uint32_t pnn,
2376 int32_t res, TDB_DATA outdata,
2379 struct get_runstate_callback_data *cd =
2380 (struct get_runstate_callback_data *)callback;
2385 ("Timed out getting runstate from node %d\n", pnn));
2389 DEBUG(DEBUG_WARNING,
2390 ("Error getting runstate from node %d - assuming runstates not supported\n",
2395 static enum ctdb_runstate * get_runstate_from_nodes(struct ctdb_context *ctdb,
2396 TALLOC_CTX *tmp_ctx,
2397 struct ctdb_node_map *nodemap,
2398 enum ctdb_runstate default_value)
2401 enum ctdb_runstate *rs;
2402 struct get_runstate_callback_data callback_data;
2405 rs = talloc_array(tmp_ctx, enum ctdb_runstate, nodemap->num);
2406 CTDB_NO_MEMORY_NULL(ctdb, rs);
2407 for (i=0; i<nodemap->num; i++) {
2408 rs[i] = default_value;
2411 callback_data.out = rs;
2412 callback_data.fatal = false;
2414 nodes = list_of_connected_nodes(ctdb, nodemap, tmp_ctx, true);
2415 if (ctdb_client_async_control(ctdb, CTDB_CONTROL_GET_RUNSTATE,
2416 nodes, 0, TAKEOVER_TIMEOUT(),
2418 get_runstate_callback,
2419 get_runstate_fail_callback,
2420 &callback_data) != 0) {
2421 if (callback_data.fatal) {
2431 /* Set internal flags for IP allocation:
2433 * Set NOIPTAKOVER ip flags from per-node NoIPTakeover tunable
2434 * Set NOIPHOST ip flag for each INACTIVE node
2435 * if all nodes are disabled:
2436 * Set NOIPHOST ip flags from per-node NoIPHostOnAllDisabled tunable
2438 * Set NOIPHOST ip flags for disabled nodes
2440 static struct ctdb_ipflags *
2441 set_ipflags_internal(struct ctdb_context *ctdb,
2442 TALLOC_CTX *tmp_ctx,
2443 struct ctdb_node_map *nodemap,
2444 uint32_t *tval_noiptakeover,
2445 uint32_t *tval_noiphostonalldisabled,
2446 enum ctdb_runstate *runstate)
2449 struct ctdb_ipflags *ipflags;
2451 /* Clear IP flags - implicit due to talloc_zero */
2452 ipflags = talloc_zero_array(tmp_ctx, struct ctdb_ipflags, nodemap->num);
2453 CTDB_NO_MEMORY_NULL(ctdb, ipflags);
2455 for (i=0;i<nodemap->num;i++) {
2456 /* Can not take IPs on node with NoIPTakeover set */
2457 if (tval_noiptakeover[i] != 0) {
2458 ipflags[i].noiptakeover = true;
2461 /* Can not host IPs on node not in RUNNING state */
2462 if (runstate[i] != CTDB_RUNSTATE_RUNNING) {
2463 ipflags[i].noiphost = true;
2466 /* Can not host IPs on INACTIVE node */
2467 if (nodemap->nodes[i].flags & NODE_FLAGS_INACTIVE) {
2468 ipflags[i].noiphost = true;
2472 if (all_nodes_are_disabled(nodemap)) {
2473 /* If all nodes are disabled, can not host IPs on node
2474 * with NoIPHostOnAllDisabled set
2476 for (i=0;i<nodemap->num;i++) {
2477 if (tval_noiphostonalldisabled[i] != 0) {
2478 ipflags[i].noiphost = true;
2482 /* If some nodes are not disabled, then can not host
2483 * IPs on DISABLED node
2485 for (i=0;i<nodemap->num;i++) {
2486 if (nodemap->nodes[i].flags & NODE_FLAGS_DISABLED) {
2487 ipflags[i].noiphost = true;
2495 static struct ctdb_ipflags *set_ipflags(struct ctdb_context *ctdb,
2496 TALLOC_CTX *tmp_ctx,
2497 struct ctdb_node_map *nodemap)
2499 uint32_t *tval_noiptakeover;
2500 uint32_t *tval_noiphostonalldisabled;
2501 struct ctdb_ipflags *ipflags;
2502 enum ctdb_runstate *runstate;
2505 tval_noiptakeover = get_tunable_from_nodes(ctdb, tmp_ctx, nodemap,
2507 if (tval_noiptakeover == NULL) {
2511 tval_noiphostonalldisabled =
2512 get_tunable_from_nodes(ctdb, tmp_ctx, nodemap,
2513 "NoIPHostOnAllDisabled", 0);
2514 if (tval_noiphostonalldisabled == NULL) {
2515 /* Caller frees tmp_ctx */
2519 /* Any nodes where CTDB_CONTROL_GET_RUNSTATE is not supported
2520 * will default to CTDB_RUNSTATE_RUNNING. This ensures
2521 * reasonable behaviour on a mixed cluster during upgrade.
2523 runstate = get_runstate_from_nodes(ctdb, tmp_ctx, nodemap,
2524 CTDB_RUNSTATE_RUNNING);
2525 if (runstate == NULL) {
2526 /* Caller frees tmp_ctx */
2530 ipflags = set_ipflags_internal(ctdb, tmp_ctx, nodemap,
2532 tval_noiphostonalldisabled,
2535 talloc_free(tval_noiptakeover);
2536 talloc_free(tval_noiphostonalldisabled);
2537 talloc_free(runstate);
2542 struct iprealloc_callback_data {
2545 client_async_callback fail_callback;
2546 void *fail_callback_data;
2547 struct ctdb_node_map *nodemap;
2550 static void iprealloc_fail_callback(struct ctdb_context *ctdb, uint32_t pnn,
2551 int32_t res, TDB_DATA outdata,
2555 struct iprealloc_callback_data *cd =
2556 (struct iprealloc_callback_data *)callback;
2560 /* If the control timed out then that's a real error,
2561 * so call the real fail callback
2563 cd->fail_callback(ctdb, pnn, res, outdata,
2564 cd->fail_callback_data);
2567 /* If not a timeout then either the ipreallocated
2568 * eventscript (or some setup) failed. This might
2569 * have failed because the IPREALLOCATED control isn't
2570 * implemented - right now there is no way of knowing
2571 * because the error codes are all folded down to -1.
2572 * Consider retrying using EVENTSCRIPT control...
2575 numnodes = talloc_array_length(cd->retry_nodes);
2576 if (pnn > numnodes) {
2578 ("ipreallocated failure from node %d, but only %d nodes in nodemap\n",
2583 /* Can't run the "ipreallocated" event on a INACTIVE node */
2584 if (cd->nodemap->nodes[pnn].flags & NODE_FLAGS_INACTIVE) {
2586 ("ipreallocated failure from node %d, but node is inactive - not flagging a retry\n",
2591 DEBUG(DEBUG_WARNING,
2592 ("ipreallocated failure from node %d, flagging retry\n",
2594 cd->retry_nodes[pnn] = true;
2599 struct takeover_callback_data {
2601 client_async_callback fail_callback;
2602 void *fail_callback_data;
2603 struct ctdb_node_map *nodemap;
2606 static void takeover_run_fail_callback(struct ctdb_context *ctdb,
2607 uint32_t node_pnn, int32_t res,
2608 TDB_DATA outdata, void *callback_data)
2610 struct takeover_callback_data *cd =
2611 talloc_get_type_abort(callback_data,
2612 struct takeover_callback_data);
2615 for (i = 0; i < cd->nodemap->num; i++) {
2616 if (node_pnn == cd->nodemap->nodes[i].pnn) {
2621 if (i == cd->nodemap->num) {
2622 DEBUG(DEBUG_ERR, (__location__ " invalid PNN %u\n", node_pnn));
2626 if (!cd->node_failed[i]) {
2627 cd->node_failed[i] = true;
2628 cd->fail_callback(ctdb, node_pnn, res, outdata,
2629 cd->fail_callback_data);
2634 make any IP alias changes for public addresses that are necessary
2636 int ctdb_takeover_run(struct ctdb_context *ctdb, struct ctdb_node_map *nodemap,
2637 uint32_t *force_rebalance_nodes,
2638 client_async_callback fail_callback, void *callback_data)
2641 struct ctdb_public_ip ip;
2642 struct ctdb_public_ipv4 ipv4;
2644 struct ctdb_public_ip_list *all_ips, *tmp_ip;
2646 struct timeval timeout;
2647 struct client_async_data *async_data;
2648 struct ctdb_client_control_state *state;
2649 TALLOC_CTX *tmp_ctx = talloc_new(ctdb);
2650 struct ctdb_ipflags *ipflags;
2651 struct takeover_callback_data *takeover_data;
2652 struct iprealloc_callback_data iprealloc_data;
2656 * ip failover is completely disabled, just send out the
2657 * ipreallocated event.
2659 if (ctdb->tunable.disable_ip_failover != 0) {
2663 ipflags = set_ipflags(ctdb, tmp_ctx, nodemap);
2664 if (ipflags == NULL) {
2665 DEBUG(DEBUG_ERR,("Failed to set IP flags - aborting takeover run\n"));
2666 talloc_free(tmp_ctx);
2672 /* Do the IP reassignment calculations */
2673 ctdb_takeover_run_core(ctdb, ipflags, &all_ips, force_rebalance_nodes);
2675 /* Now tell all nodes to release any public IPs should not
2676 * host. This will be a NOOP on nodes that don't currently
2677 * hold the given IP.
2679 takeover_data = talloc_zero(tmp_ctx, struct takeover_callback_data);
2680 CTDB_NO_MEMORY_FATAL(ctdb, takeover_data);
2682 takeover_data->node_failed = talloc_zero_array(tmp_ctx,
2683 bool, nodemap->num);
2684 CTDB_NO_MEMORY_FATAL(ctdb, takeover_data->node_failed);
2685 takeover_data->fail_callback = fail_callback;
2686 takeover_data->fail_callback_data = callback_data;
2687 takeover_data->nodemap = nodemap;
2689 async_data = talloc_zero(tmp_ctx, struct client_async_data);
2690 CTDB_NO_MEMORY_FATAL(ctdb, async_data);
2692 async_data->fail_callback = takeover_run_fail_callback;
2693 async_data->callback_data = takeover_data;
2695 for (i=0;i<nodemap->num;i++) {
2696 /* don't talk to unconnected nodes, but do talk to banned nodes */
2697 if (nodemap->nodes[i].flags & NODE_FLAGS_DISCONNECTED) {
2701 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
2702 if (tmp_ip->pnn == nodemap->nodes[i].pnn) {
2703 /* This node should be serving this
2704 vnn so dont tell it to release the ip
2708 if (tmp_ip->addr.sa.sa_family == AF_INET) {
2709 ipv4.pnn = tmp_ip->pnn;
2710 ipv4.sin = tmp_ip->addr.ip;
2712 timeout = TAKEOVER_TIMEOUT();
2713 data.dsize = sizeof(ipv4);
2714 data.dptr = (uint8_t *)&ipv4;
2715 state = ctdb_control_send(ctdb, nodemap->nodes[i].pnn,
2716 0, CTDB_CONTROL_RELEASE_IPv4, 0,
2720 ip.pnn = tmp_ip->pnn;
2721 ip.addr = tmp_ip->addr;
2723 timeout = TAKEOVER_TIMEOUT();
2724 data.dsize = sizeof(ip);
2725 data.dptr = (uint8_t *)&ip;
2726 state = ctdb_control_send(ctdb, nodemap->nodes[i].pnn,
2727 0, CTDB_CONTROL_RELEASE_IP, 0,
2732 if (state == NULL) {
2733 DEBUG(DEBUG_ERR,(__location__ " Failed to call async control CTDB_CONTROL_RELEASE_IP to node %u\n", nodemap->nodes[i].pnn));
2734 talloc_free(tmp_ctx);
2738 ctdb_client_async_add(async_data, state);
2741 if (ctdb_client_async_wait(ctdb, async_data) != 0) {
2742 DEBUG(DEBUG_ERR,(__location__ " Async control CTDB_CONTROL_RELEASE_IP failed\n"));
2743 talloc_free(tmp_ctx);
2746 talloc_free(async_data);
2749 /* tell all nodes to get their own IPs */
2750 async_data = talloc_zero(tmp_ctx, struct client_async_data);
2751 CTDB_NO_MEMORY_FATAL(ctdb, async_data);
2753 async_data->fail_callback = fail_callback;
2754 async_data->callback_data = callback_data;
2756 for (tmp_ip=all_ips;tmp_ip;tmp_ip=tmp_ip->next) {
2757 if (tmp_ip->pnn == -1) {
2758 /* this IP won't be taken over */
2762 if (tmp_ip->addr.sa.sa_family == AF_INET) {
2763 ipv4.pnn = tmp_ip->pnn;
2764 ipv4.sin = tmp_ip->addr.ip;
2766 timeout = TAKEOVER_TIMEOUT();
2767 data.dsize = sizeof(ipv4);
2768 data.dptr = (uint8_t *)&ipv4;
2769 state = ctdb_control_send(ctdb, tmp_ip->pnn,
2770 0, CTDB_CONTROL_TAKEOVER_IPv4, 0,
2774 ip.pnn = tmp_ip->pnn;
2775 ip.addr = tmp_ip->addr;
2777 timeout = TAKEOVER_TIMEOUT();
2778 data.dsize = sizeof(ip);
2779 data.dptr = (uint8_t *)&ip;
2780 state = ctdb_control_send(ctdb, tmp_ip->pnn,
2781 0, CTDB_CONTROL_TAKEOVER_IP, 0,
2785 if (state == NULL) {
2786 DEBUG(DEBUG_ERR,(__location__ " Failed to call async control CTDB_CONTROL_TAKEOVER_IP to node %u\n", tmp_ip->pnn));
2787 talloc_free(tmp_ctx);
2791 ctdb_client_async_add(async_data, state);
2793 if (ctdb_client_async_wait(ctdb, async_data) != 0) {
2794 DEBUG(DEBUG_ERR,(__location__ " Async control CTDB_CONTROL_TAKEOVER_IP failed\n"));
2795 talloc_free(tmp_ctx);
2801 * Tell all nodes to run eventscripts to process the
2802 * "ipreallocated" event. This can do a lot of things,
2803 * including restarting services to reconfigure them if public
2804 * IPs have moved. Once upon a time this event only used to
2807 retry_data = talloc_zero_array(tmp_ctx, bool, nodemap->num);
2808 CTDB_NO_MEMORY_FATAL(ctdb, retry_data);
2809 iprealloc_data.retry_nodes = retry_data;
2810 iprealloc_data.retry_count = 0;
2811 iprealloc_data.fail_callback = fail_callback;
2812 iprealloc_data.fail_callback_data = callback_data;
2813 iprealloc_data.nodemap = nodemap;
2815 nodes = list_of_connected_nodes(ctdb, nodemap, tmp_ctx, true);
2816 ret = ctdb_client_async_control(ctdb, CTDB_CONTROL_IPREALLOCATED,
2817 nodes, 0, TAKEOVER_TIMEOUT(),
2819 NULL, iprealloc_fail_callback,
2822 /* If the control failed then we should retry to any
2823 * nodes flagged by iprealloc_fail_callback using the
2824 * EVENTSCRIPT control. This is a best-effort at
2825 * backward compatiblity when running a mixed cluster
2826 * where some nodes have not yet been upgraded to
2827 * support the IPREALLOCATED control.
2829 DEBUG(DEBUG_WARNING,
2830 ("Retry ipreallocated to some nodes using eventscript control\n"));
2832 nodes = talloc_array(tmp_ctx, uint32_t,
2833 iprealloc_data.retry_count);
2834 CTDB_NO_MEMORY_FATAL(ctdb, nodes);
2837 for (i=0; i<nodemap->num; i++) {
2838 if (iprealloc_data.retry_nodes[i]) {
2844 data.dptr = discard_const("ipreallocated");
2845 data.dsize = strlen((char *)data.dptr) + 1;
2846 ret = ctdb_client_async_control(ctdb,
2847 CTDB_CONTROL_RUN_EVENTSCRIPTS,
2848 nodes, 0, TAKEOVER_TIMEOUT(),
2850 NULL, fail_callback,
2853 DEBUG(DEBUG_ERR, (__location__ " failed to send control to run eventscripts with \"ipreallocated\"\n"));
2857 talloc_free(tmp_ctx);
2863 destroy a ctdb_client_ip structure
2865 static int ctdb_client_ip_destructor(struct ctdb_client_ip *ip)
2867 DEBUG(DEBUG_DEBUG,("destroying client tcp for %s:%u (client_id %u)\n",
2868 ctdb_addr_to_str(&ip->addr),
2869 ntohs(ip->addr.ip.sin_port),
2872 DLIST_REMOVE(ip->ctdb->client_ip_list, ip);
2877 called by a client to inform us of a TCP connection that it is managing
2878 that should tickled with an ACK when IP takeover is done
2879 we handle both the old ipv4 style of packets as well as the new ipv4/6
2882 int32_t ctdb_control_tcp_client(struct ctdb_context *ctdb, uint32_t client_id,
2885 struct ctdb_client *client = ctdb_reqid_find(ctdb, client_id, struct ctdb_client);
2886 struct ctdb_control_tcp *old_addr = NULL;
2887 struct ctdb_control_tcp_addr new_addr;
2888 struct ctdb_control_tcp_addr *tcp_sock = NULL;
2889 struct ctdb_tcp_list *tcp;
2890 struct ctdb_tcp_connection t;
2893 struct ctdb_client_ip *ip;
2894 struct ctdb_vnn *vnn;
2895 ctdb_sock_addr addr;
2897 switch (indata.dsize) {
2898 case sizeof(struct ctdb_control_tcp):
2899 old_addr = (struct ctdb_control_tcp *)indata.dptr;
2900 ZERO_STRUCT(new_addr);
2901 tcp_sock = &new_addr;
2902 tcp_sock->src.ip = old_addr->src;
2903 tcp_sock->dest.ip = old_addr->dest;
2905 case sizeof(struct ctdb_control_tcp_addr):
2906 tcp_sock = (struct ctdb_control_tcp_addr *)indata.dptr;
2909 DEBUG(DEBUG_ERR,(__location__ " Invalid data structure passed "
2910 "to ctdb_control_tcp_client. size was %d but "
2911 "only allowed sizes are %lu and %lu\n",
2913 (long unsigned)sizeof(struct ctdb_control_tcp),
2914 (long unsigned)sizeof(struct ctdb_control_tcp_addr)));
2918 addr = tcp_sock->src;
2919 ctdb_canonicalize_ip(&addr, &tcp_sock->src);
2920 addr = tcp_sock->dest;
2921 ctdb_canonicalize_ip(&addr, &tcp_sock->dest);
2924 memcpy(&addr, &tcp_sock->dest, sizeof(addr));
2925 vnn = find_public_ip_vnn(ctdb, &addr);
2927 switch (addr.sa.sa_family) {
2929 if (ntohl(addr.ip.sin_addr.s_addr) != INADDR_LOOPBACK) {
2930 DEBUG(DEBUG_ERR,("Could not add client IP %s. This is not a public address.\n",
2931 ctdb_addr_to_str(&addr)));
2935 DEBUG(DEBUG_ERR,("Could not add client IP %s. This is not a public ipv6 address.\n",
2936 ctdb_addr_to_str(&addr)));
2939 DEBUG(DEBUG_ERR,(__location__ " Unknown family type %d\n", addr.sa.sa_family));
2945 if (vnn->pnn != ctdb->pnn) {
2946 DEBUG(DEBUG_ERR,("Attempt to register tcp client for IP %s we don't hold - failing (client_id %u pid %u)\n",
2947 ctdb_addr_to_str(&addr),
2948 client_id, client->pid));
2949 /* failing this call will tell smbd to die */
2953 ip = talloc(client, struct ctdb_client_ip);
2954 CTDB_NO_MEMORY(ctdb, ip);
2958 ip->client_id = client_id;
2959 talloc_set_destructor(ip, ctdb_client_ip_destructor);
2960 DLIST_ADD(ctdb->client_ip_list, ip);
2962 tcp = talloc(client, struct ctdb_tcp_list);
2963 CTDB_NO_MEMORY(ctdb, tcp);
2965 tcp->connection.src_addr = tcp_sock->src;
2966 tcp->connection.dst_addr = tcp_sock->dest;
2968 DLIST_ADD(client->tcp_list, tcp);
2970 t.src_addr = tcp_sock->src;
2971 t.dst_addr = tcp_sock->dest;
2973 data.dptr = (uint8_t *)&t;
2974 data.dsize = sizeof(t);
2976 switch (addr.sa.sa_family) {
2978 DEBUG(DEBUG_INFO,("registered tcp client for %u->%s:%u (client_id %u pid %u)\n",
2979 (unsigned)ntohs(tcp_sock->dest.ip.sin_port),
2980 ctdb_addr_to_str(&tcp_sock->src),
2981 (unsigned)ntohs(tcp_sock->src.ip.sin_port), client_id, client->pid));
2984 DEBUG(DEBUG_INFO,("registered tcp client for %u->%s:%u (client_id %u pid %u)\n",
2985 (unsigned)ntohs(tcp_sock->dest.ip6.sin6_port),
2986 ctdb_addr_to_str(&tcp_sock->src),
2987 (unsigned)ntohs(tcp_sock->src.ip6.sin6_port), client_id, client->pid));
2990 DEBUG(DEBUG_ERR,(__location__ " Unknown family %d\n", addr.sa.sa_family));
2994 /* tell all nodes about this tcp connection */
2995 ret = ctdb_daemon_send_control(ctdb, CTDB_BROADCAST_CONNECTED, 0,
2996 CTDB_CONTROL_TCP_ADD,
2997 0, CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL);
2999 DEBUG(DEBUG_ERR,(__location__ " Failed to send CTDB_CONTROL_TCP_ADD\n"));
3007 find a tcp address on a list
3009 static struct ctdb_tcp_connection *ctdb_tcp_find(struct ctdb_tcp_array *array,
3010 struct ctdb_tcp_connection *tcp)
3014 if (array == NULL) {
3018 for (i=0;i<array->num;i++) {
3019 if (ctdb_same_sockaddr(&array->connections[i].src_addr, &tcp->src_addr) &&
3020 ctdb_same_sockaddr(&array->connections[i].dst_addr, &tcp->dst_addr)) {
3021 return &array->connections[i];
3030 called by a daemon to inform us of a TCP connection that one of its
3031 clients managing that should tickled with an ACK when IP takeover is
3034 int32_t ctdb_control_tcp_add(struct ctdb_context *ctdb, TDB_DATA indata, bool tcp_update_needed)
3036 struct ctdb_tcp_connection *p = (struct ctdb_tcp_connection *)indata.dptr;
3037 struct ctdb_tcp_array *tcparray;
3038 struct ctdb_tcp_connection tcp;
3039 struct ctdb_vnn *vnn;
3041 vnn = find_public_ip_vnn(ctdb, &p->dst_addr);
3043 DEBUG(DEBUG_INFO,(__location__ " got TCP_ADD control for an address which is not a public address '%s'\n",
3044 ctdb_addr_to_str(&p->dst_addr)));
3050 tcparray = vnn->tcp_array;
3052 /* If this is the first tickle */
3053 if (tcparray == NULL) {
3054 tcparray = talloc_size(ctdb->nodes,
3055 offsetof(struct ctdb_tcp_array, connections) +
3056 sizeof(struct ctdb_tcp_connection) * 1);
3057 CTDB_NO_MEMORY(ctdb, tcparray);
3058 vnn->tcp_array = tcparray;
3061 tcparray->connections = talloc_size(tcparray, sizeof(struct ctdb_tcp_connection));
3062 CTDB_NO_MEMORY(ctdb, tcparray->connections);
3064 tcparray->connections[tcparray->num].src_addr = p->src_addr;
3065 tcparray->connections[tcparray->num].dst_addr = p->dst_addr;
3068 if (tcp_update_needed) {
3069 vnn->tcp_update_needed = true;
3075 /* Do we already have this tickle ?*/
3076 tcp.src_addr = p->src_addr;
3077 tcp.dst_addr = p->dst_addr;
3078 if (ctdb_tcp_find(vnn->tcp_array, &tcp) != NULL) {
3079 DEBUG(DEBUG_DEBUG,("Already had tickle info for %s:%u for vnn:%u\n",
3080 ctdb_addr_to_str(&tcp.dst_addr),
3081 ntohs(tcp.dst_addr.ip.sin_port),
3086 /* A new tickle, we must add it to the array */
3087 tcparray->connections = talloc_realloc(tcparray, tcparray->connections,
3088 struct ctdb_tcp_connection,
3090 CTDB_NO_MEMORY(ctdb, tcparray->connections);
3092 vnn->tcp_array = tcparray;
3093 tcparray->connections[tcparray->num].src_addr = p->src_addr;
3094 tcparray->connections[tcparray->num].dst_addr = p->dst_addr;
3097 DEBUG(DEBUG_INFO,("Added tickle info for %s:%u from vnn %u\n",
3098 ctdb_addr_to_str(&tcp.dst_addr),
3099 ntohs(tcp.dst_addr.ip.sin_port),
3102 if (tcp_update_needed) {
3103 vnn->tcp_update_needed = true;
3111 called by a daemon to inform us of a TCP connection that one of its
3112 clients managing that should tickled with an ACK when IP takeover is
3115 static void ctdb_remove_tcp_connection(struct ctdb_context *ctdb, struct ctdb_tcp_connection *conn)
3117 struct ctdb_tcp_connection *tcpp;
3118 struct ctdb_vnn *vnn = find_public_ip_vnn(ctdb, &conn->dst_addr);
3121 DEBUG(DEBUG_ERR,(__location__ " unable to find public address %s\n",
3122 ctdb_addr_to_str(&conn->dst_addr)));
3126 /* if the array is empty we cant remove it
3127 and we dont need to do anything
3129 if (vnn->tcp_array == NULL) {
3130 DEBUG(DEBUG_INFO,("Trying to remove tickle that doesnt exist (array is empty) %s:%u\n",
3131 ctdb_addr_to_str(&conn->dst_addr),
3132 ntohs(conn->dst_addr.ip.sin_port)));
3137 /* See if we know this connection
3138 if we dont know this connection then we dont need to do anything
3140 tcpp = ctdb_tcp_find(vnn->tcp_array, conn);
3142 DEBUG(DEBUG_INFO,("Trying to remove tickle that doesnt exist %s:%u\n",
3143 ctdb_addr_to_str(&conn->dst_addr),
3144 ntohs(conn->dst_addr.ip.sin_port)));
3149 /* We need to remove this entry from the array.
3150 Instead of allocating a new array and copying data to it
3151 we cheat and just copy the last entry in the existing array
3152 to the entry that is to be removed and just shring the
3155 *tcpp = vnn->tcp_array->connections[vnn->tcp_array->num - 1];
3156 vnn->tcp_array->num--;
3158 /* If we deleted the last entry we also need to remove the entire array
3160 if (vnn->tcp_array->num == 0) {
3161 talloc_free(vnn->tcp_array);
3162 vnn->tcp_array = NULL;
3165 vnn->tcp_update_needed = true;
3167 DEBUG(DEBUG_INFO,("Removed tickle info for %s:%u\n",
3168 ctdb_addr_to_str(&conn->src_addr),
3169 ntohs(conn->src_addr.ip.sin_port)));
3174 called by a daemon to inform us of a TCP connection that one of its
3175 clients used are no longer needed in the tickle database
3177 int32_t ctdb_control_tcp_remove(struct ctdb_context *ctdb, TDB_DATA indata)
3179 struct ctdb_tcp_connection *conn = (struct ctdb_tcp_connection *)indata.dptr;
3181 ctdb_remove_tcp_connection(ctdb, conn);
3188 called when a daemon restarts - send all tickes for all public addresses
3189 we are serving immediately to the new node.
3191 int32_t ctdb_control_startup(struct ctdb_context *ctdb, uint32_t vnn)
3193 /*XXX here we should send all tickes we are serving to the new node */
3199 called when a client structure goes away - hook to remove
3200 elements from the tcp_list in all daemons
3202 void ctdb_takeover_client_destructor_hook(struct ctdb_client *client)
3204 while (client->tcp_list) {
3205 struct ctdb_tcp_list *tcp = client->tcp_list;
3206 DLIST_REMOVE(client->tcp_list, tcp);
3207 ctdb_remove_tcp_connection(client->ctdb, &tcp->connection);
3213 release all IPs on shutdown
3215 void ctdb_release_all_ips(struct ctdb_context *ctdb)
3217 struct ctdb_vnn *vnn;
3220 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
3221 if (!ctdb_sys_have_ip(&vnn->public_address)) {
3222 ctdb_vnn_unassign_iface(ctdb, vnn);
3229 DEBUG(DEBUG_INFO,("Release of IP %s/%u on interface %s node:-1\n",
3230 ctdb_addr_to_str(&vnn->public_address),
3231 vnn->public_netmask_bits,
3232 ctdb_vnn_iface_string(vnn)));
3234 ctdb_event_script_args(ctdb, CTDB_EVENT_RELEASE_IP, "%s %s %u",
3235 ctdb_vnn_iface_string(vnn),
3236 ctdb_addr_to_str(&vnn->public_address),
3237 vnn->public_netmask_bits);
3238 release_kill_clients(ctdb, &vnn->public_address);
3239 ctdb_vnn_unassign_iface(ctdb, vnn);
3243 DEBUG(DEBUG_NOTICE,(__location__ " Released %d public IPs\n", count));
3248 get list of public IPs
3250 int32_t ctdb_control_get_public_ips(struct ctdb_context *ctdb,
3251 struct ctdb_req_control *c, TDB_DATA *outdata)
3254 struct ctdb_all_public_ips *ips;
3255 struct ctdb_vnn *vnn;
3256 bool only_available = false;
3258 if (c->flags & CTDB_PUBLIC_IP_FLAGS_ONLY_AVAILABLE) {
3259 only_available = true;
3262 /* count how many public ip structures we have */
3264 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
3268 len = offsetof(struct ctdb_all_public_ips, ips) +
3269 num*sizeof(struct ctdb_public_ip);
3270 ips = talloc_zero_size(outdata, len);
3271 CTDB_NO_MEMORY(ctdb, ips);
3274 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
3275 if (only_available && !ctdb_vnn_available(ctdb, vnn)) {
3278 ips->ips[i].pnn = vnn->pnn;
3279 ips->ips[i].addr = vnn->public_address;
3283 len = offsetof(struct ctdb_all_public_ips, ips) +
3284 i*sizeof(struct ctdb_public_ip);
3286 outdata->dsize = len;
3287 outdata->dptr = (uint8_t *)ips;
3294 get list of public IPs, old ipv4 style. only returns ipv4 addresses
3296 int32_t ctdb_control_get_public_ipsv4(struct ctdb_context *ctdb,
3297 struct ctdb_req_control *c, TDB_DATA *outdata)
3300 struct ctdb_all_public_ipsv4 *ips;
3301 struct ctdb_vnn *vnn;
3303 /* count how many public ip structures we have */
3305 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
3306 if (vnn->public_address.sa.sa_family != AF_INET) {
3312 len = offsetof(struct ctdb_all_public_ipsv4, ips) +
3313 num*sizeof(struct ctdb_public_ipv4);
3314 ips = talloc_zero_size(outdata, len);
3315 CTDB_NO_MEMORY(ctdb, ips);
3317 outdata->dsize = len;
3318 outdata->dptr = (uint8_t *)ips;
3322 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
3323 if (vnn->public_address.sa.sa_family != AF_INET) {
3326 ips->ips[i].pnn = vnn->pnn;
3327 ips->ips[i].sin = vnn->public_address.ip;
3334 int32_t ctdb_control_get_public_ip_info(struct ctdb_context *ctdb,
3335 struct ctdb_req_control *c,
3340 ctdb_sock_addr *addr;
3341 struct ctdb_control_public_ip_info *info;
3342 struct ctdb_vnn *vnn;
3344 addr = (ctdb_sock_addr *)indata.dptr;
3346 vnn = find_public_ip_vnn(ctdb, addr);
3348 /* if it is not a public ip it could be our 'single ip' */
3349 if (ctdb->single_ip_vnn) {
3350 if (ctdb_same_ip(&ctdb->single_ip_vnn->public_address, addr)) {
3351 vnn = ctdb->single_ip_vnn;
3356 DEBUG(DEBUG_ERR,(__location__ " Could not get public ip info, "
3357 "'%s'not a public address\n",
3358 ctdb_addr_to_str(addr)));
3362 /* count how many public ip structures we have */
3364 for (;vnn->ifaces[num];) {
3368 len = offsetof(struct ctdb_control_public_ip_info, ifaces) +
3369 num*sizeof(struct ctdb_control_iface_info);
3370 info = talloc_zero_size(outdata, len);
3371 CTDB_NO_MEMORY(ctdb, info);
3373 info->ip.addr = vnn->public_address;
3374 info->ip.pnn = vnn->pnn;
3375 info->active_idx = 0xFFFFFFFF;
3377 for (i=0; vnn->ifaces[i]; i++) {
3378 struct ctdb_iface *cur;
3380 cur = ctdb_find_iface(ctdb, vnn->ifaces[i]);
3382 DEBUG(DEBUG_CRIT, (__location__ " internal error iface[%s] unknown\n",
3386 if (vnn->iface == cur) {
3387 info->active_idx = i;
3389 strcpy(info->ifaces[i].name, cur->name);
3390 info->ifaces[i].link_state = cur->link_up;
3391 info->ifaces[i].references = cur->references;
3394 len = offsetof(struct ctdb_control_public_ip_info, ifaces) +
3395 i*sizeof(struct ctdb_control_iface_info);
3397 outdata->dsize = len;
3398 outdata->dptr = (uint8_t *)info;
3403 int32_t ctdb_control_get_ifaces(struct ctdb_context *ctdb,
3404 struct ctdb_req_control *c,
3408 struct ctdb_control_get_ifaces *ifaces;
3409 struct ctdb_iface *cur;
3411 /* count how many public ip structures we have */
3413 for (cur=ctdb->ifaces;cur;cur=cur->next) {
3417 len = offsetof(struct ctdb_control_get_ifaces, ifaces) +
3418 num*sizeof(struct ctdb_control_iface_info);
3419 ifaces = talloc_zero_size(outdata, len);
3420 CTDB_NO_MEMORY(ctdb, ifaces);
3423 for (cur=ctdb->ifaces;cur;cur=cur->next) {
3424 strcpy(ifaces->ifaces[i].name, cur->name);
3425 ifaces->ifaces[i].link_state = cur->link_up;
3426 ifaces->ifaces[i].references = cur->references;
3430 len = offsetof(struct ctdb_control_get_ifaces, ifaces) +
3431 i*sizeof(struct ctdb_control_iface_info);
3433 outdata->dsize = len;
3434 outdata->dptr = (uint8_t *)ifaces;
3439 int32_t ctdb_control_set_iface_link(struct ctdb_context *ctdb,
3440 struct ctdb_req_control *c,
3443 struct ctdb_control_iface_info *info;
3444 struct ctdb_iface *iface;
3445 bool link_up = false;
3447 info = (struct ctdb_control_iface_info *)indata.dptr;
3449 if (info->name[CTDB_IFACE_SIZE] != '\0') {
3450 int len = strnlen(info->name, CTDB_IFACE_SIZE);
3451 DEBUG(DEBUG_ERR, (__location__ " name[%*.*s] not terminated\n",
3452 len, len, info->name));
3456 switch (info->link_state) {
3464 DEBUG(DEBUG_ERR, (__location__ " link_state[%u] invalid\n",
3465 (unsigned int)info->link_state));
3469 if (info->references != 0) {
3470 DEBUG(DEBUG_ERR, (__location__ " references[%u] should be 0\n",
3471 (unsigned int)info->references));
3475 iface = ctdb_find_iface(ctdb, info->name);
3476 if (iface == NULL) {
3480 if (link_up == iface->link_up) {
3484 DEBUG(iface->link_up?DEBUG_ERR:DEBUG_NOTICE,
3485 ("iface[%s] has changed it's link status %s => %s\n",
3487 iface->link_up?"up":"down",
3488 link_up?"up":"down"));
3490 iface->link_up = link_up;
3496 structure containing the listening socket and the list of tcp connections
3497 that the ctdb daemon is to kill
3499 struct ctdb_kill_tcp {
3500 struct ctdb_vnn *vnn;
3501 struct ctdb_context *ctdb;
3503 struct fd_event *fde;
3504 trbt_tree_t *connections;
3509 a tcp connection that is to be killed
3511 struct ctdb_killtcp_con {
3512 ctdb_sock_addr src_addr;
3513 ctdb_sock_addr dst_addr;
3515 struct ctdb_kill_tcp *killtcp;
3518 /* this function is used to create a key to represent this socketpair
3519 in the killtcp tree.
3520 this key is used to insert and lookup matching socketpairs that are
3521 to be tickled and RST
3523 #define KILLTCP_KEYLEN 10
3524 static uint32_t *killtcp_key(ctdb_sock_addr *src, ctdb_sock_addr *dst)
3526 static uint32_t key[KILLTCP_KEYLEN];
3528 bzero(key, sizeof(key));
3530 if (src->sa.sa_family != dst->sa.sa_family) {
3531 DEBUG(DEBUG_ERR, (__location__ " ERROR, different families passed :%u vs %u\n", src->sa.sa_family, dst->sa.sa_family));
3535 switch (src->sa.sa_family) {
3537 key[0] = dst->ip.sin_addr.s_addr;
3538 key[1] = src->ip.sin_addr.s_addr;
3539 key[2] = dst->ip.sin_port;
3540 key[3] = src->ip.sin_port;
3543 uint32_t *dst6_addr32 =
3544 (uint32_t *)&(dst->ip6.sin6_addr.s6_addr);
3545 uint32_t *src6_addr32 =
3546 (uint32_t *)&(src->ip6.sin6_addr.s6_addr);
3547 key[0] = dst6_addr32[3];
3548 key[1] = src6_addr32[3];
3549 key[2] = dst6_addr32[2];
3550 key[3] = src6_addr32[2];
3551 key[4] = dst6_addr32[1];
3552 key[5] = src6_addr32[1];
3553 key[6] = dst6_addr32[0];
3554 key[7] = src6_addr32[0];
3555 key[8] = dst->ip6.sin6_port;
3556 key[9] = src->ip6.sin6_port;
3560 DEBUG(DEBUG_ERR, (__location__ " ERROR, unknown family passed :%u\n", src->sa.sa_family));
3568 called when we get a read event on the raw socket
3570 static void capture_tcp_handler(struct event_context *ev, struct fd_event *fde,
3571 uint16_t flags, void *private_data)
3573 struct ctdb_kill_tcp *killtcp = talloc_get_type(private_data, struct ctdb_kill_tcp);
3574 struct ctdb_killtcp_con *con;
3575 ctdb_sock_addr src, dst;
3576 uint32_t ack_seq, seq;
3578 if (!(flags & EVENT_FD_READ)) {
3582 if (ctdb_sys_read_tcp_packet(killtcp->capture_fd,
3583 killtcp->private_data,
3585 &ack_seq, &seq) != 0) {
3586 /* probably a non-tcp ACK packet */
3590 /* check if we have this guy in our list of connections
3593 con = trbt_lookuparray32(killtcp->connections,
3594 KILLTCP_KEYLEN, killtcp_key(&src, &dst));
3596 /* no this was some other packet we can just ignore */
3600 /* This one has been tickled !
3601 now reset him and remove him from the list.
3603 DEBUG(DEBUG_INFO, ("sending a tcp reset to kill connection :%d -> %s:%d\n",
3604 ntohs(con->dst_addr.ip.sin_port),
3605 ctdb_addr_to_str(&con->src_addr),
3606 ntohs(con->src_addr.ip.sin_port)));
3608 ctdb_sys_send_tcp(&con->dst_addr, &con->src_addr, ack_seq, seq, 1);
3613 /* when traversing the list of all tcp connections to send tickle acks to
3614 (so that we can capture the ack coming back and kill the connection
3616 this callback is called for each connection we are currently trying to kill
3618 static int tickle_connection_traverse(void *param, void *data)
3620 struct ctdb_killtcp_con *con = talloc_get_type(data, struct ctdb_killtcp_con);
3622 /* have tried too many times, just give up */
3623 if (con->count >= 5) {
3624 /* can't delete in traverse: reparent to delete_cons */
3625 talloc_steal(param, con);
3629 /* othervise, try tickling it again */
3632 (ctdb_sock_addr *)&con->dst_addr,
3633 (ctdb_sock_addr *)&con->src_addr,
3640 called every second until all sentenced connections have been reset
3642 static void ctdb_tickle_sentenced_connections(struct event_context *ev, struct timed_event *te,
3643 struct timeval t, void *private_data)
3645 struct ctdb_kill_tcp *killtcp = talloc_get_type(private_data, struct ctdb_kill_tcp);
3646 void *delete_cons = talloc_new(NULL);
3648 /* loop over all connections sending tickle ACKs */
3649 trbt_traversearray32(killtcp->connections, KILLTCP_KEYLEN, tickle_connection_traverse, delete_cons);
3651 /* now we've finished traverse, it's safe to do deletion. */
3652 talloc_free(delete_cons);
3654 /* If there are no more connections to kill we can remove the
3655 entire killtcp structure
3657 if ( (killtcp->connections == NULL) ||
3658 (killtcp->connections->root == NULL) ) {
3659 talloc_free(killtcp);
3663 /* try tickling them again in a seconds time
3665 event_add_timed(killtcp->ctdb->ev, killtcp, timeval_current_ofs(1, 0),
3666 ctdb_tickle_sentenced_connections, killtcp);
3670 destroy the killtcp structure
3672 static int ctdb_killtcp_destructor(struct ctdb_kill_tcp *killtcp)
3674 struct ctdb_vnn *tmpvnn;
3676 /* verify that this vnn is still active */
3677 for (tmpvnn = killtcp->ctdb->vnn; tmpvnn; tmpvnn = tmpvnn->next) {
3678 if (tmpvnn == killtcp->vnn) {
3683 if (tmpvnn == NULL) {
3687 if (killtcp->vnn->killtcp != killtcp) {
3691 killtcp->vnn->killtcp = NULL;
3697 /* nothing fancy here, just unconditionally replace any existing
3698 connection structure with the new one.
3700 dont even free the old one if it did exist, that one is talloc_stolen
3701 by the same node in the tree anyway and will be deleted when the new data
3704 static void *add_killtcp_callback(void *parm, void *data)
3710 add a tcp socket to the list of connections we want to RST
3712 static int ctdb_killtcp_add_connection(struct ctdb_context *ctdb,
3716 ctdb_sock_addr src, dst;
3717 struct ctdb_kill_tcp *killtcp;
3718 struct ctdb_killtcp_con *con;
3719 struct ctdb_vnn *vnn;
3721 ctdb_canonicalize_ip(s, &src);
3722 ctdb_canonicalize_ip(d, &dst);
3724 vnn = find_public_ip_vnn(ctdb, &dst);
3726 vnn = find_public_ip_vnn(ctdb, &src);
3729 /* if it is not a public ip it could be our 'single ip' */
3730 if (ctdb->single_ip_vnn) {
3731 if (ctdb_same_ip(&ctdb->single_ip_vnn->public_address, &dst)) {
3732 vnn = ctdb->single_ip_vnn;
3737 DEBUG(DEBUG_ERR,(__location__ " Could not killtcp, not a public address\n"));
3741 killtcp = vnn->killtcp;
3743 /* If this is the first connection to kill we must allocate
3746 if (killtcp == NULL) {
3747 killtcp = talloc_zero(vnn, struct ctdb_kill_tcp);
3748 CTDB_NO_MEMORY(ctdb, killtcp);
3751 killtcp->ctdb = ctdb;
3752 killtcp->capture_fd = -1;
3753 killtcp->connections = trbt_create(killtcp, 0);
3755 vnn->killtcp = killtcp;
3756 talloc_set_destructor(killtcp, ctdb_killtcp_destructor);
3761 /* create a structure that describes this connection we want to
3762 RST and store it in killtcp->connections
3764 con = talloc(killtcp, struct ctdb_killtcp_con);
3765 CTDB_NO_MEMORY(ctdb, con);
3766 con->src_addr = src;
3767 con->dst_addr = dst;
3769 con->killtcp = killtcp;
3772 trbt_insertarray32_callback(killtcp->connections,
3773 KILLTCP_KEYLEN, killtcp_key(&con->dst_addr, &con->src_addr),
3774 add_killtcp_callback, con);
3777 If we dont have a socket to listen on yet we must create it
3779 if (killtcp->capture_fd == -1) {
3780 const char *iface = ctdb_vnn_iface_string(vnn);
3781 killtcp->capture_fd = ctdb_sys_open_capture_socket(iface, &killtcp->private_data);
3782 if (killtcp->capture_fd == -1) {
3783 DEBUG(DEBUG_CRIT,(__location__ " Failed to open capturing "
3784 "socket on iface '%s' for killtcp (%s)\n",
3785 iface, strerror(errno)));
3791 if (killtcp->fde == NULL) {
3792 killtcp->fde = event_add_fd(ctdb->ev, killtcp, killtcp->capture_fd,
3794 capture_tcp_handler, killtcp);
3795 tevent_fd_set_auto_close(killtcp->fde);
3797 /* We also need to set up some events to tickle all these connections
3798 until they are all reset
3800 event_add_timed(ctdb->ev, killtcp, timeval_current_ofs(1, 0),
3801 ctdb_tickle_sentenced_connections, killtcp);
3804 /* tickle him once now */
3813 talloc_free(vnn->killtcp);
3814 vnn->killtcp = NULL;
3819 kill a TCP connection.
3821 int32_t ctdb_control_kill_tcp(struct ctdb_context *ctdb, TDB_DATA indata)
3823 struct ctdb_control_killtcp *killtcp = (struct ctdb_control_killtcp *)indata.dptr;
3825 return ctdb_killtcp_add_connection(ctdb, &killtcp->src_addr, &killtcp->dst_addr);
3829 called by a daemon to inform us of the entire list of TCP tickles for
3830 a particular public address.
3831 this control should only be sent by the node that is currently serving
3832 that public address.
3834 int32_t ctdb_control_set_tcp_tickle_list(struct ctdb_context *ctdb, TDB_DATA indata)
3836 struct ctdb_control_tcp_tickle_list *list = (struct ctdb_control_tcp_tickle_list *)indata.dptr;
3837 struct ctdb_tcp_array *tcparray;
3838 struct ctdb_vnn *vnn;
3840 /* We must at least have tickles.num or else we cant verify the size
3841 of the received data blob
3843 if (indata.dsize < offsetof(struct ctdb_control_tcp_tickle_list,
3844 tickles.connections)) {
3845 DEBUG(DEBUG_ERR,("Bad indata in ctdb_control_set_tcp_tickle_list. Not enough data for the tickle.num field\n"));
3849 /* verify that the size of data matches what we expect */
3850 if (indata.dsize < offsetof(struct ctdb_control_tcp_tickle_list,
3851 tickles.connections)
3852 + sizeof(struct ctdb_tcp_connection)
3853 * list->tickles.num) {
3854 DEBUG(DEBUG_ERR,("Bad indata in ctdb_control_set_tcp_tickle_list\n"));
3858 vnn = find_public_ip_vnn(ctdb, &list->addr);
3860 DEBUG(DEBUG_INFO,(__location__ " Could not set tcp tickle list, '%s' is not a public address\n",
3861 ctdb_addr_to_str(&list->addr)));
3866 /* remove any old ticklelist we might have */
3867 talloc_free(vnn->tcp_array);
3868 vnn->tcp_array = NULL;
3870 tcparray = talloc(ctdb->nodes, struct ctdb_tcp_array);
3871 CTDB_NO_MEMORY(ctdb, tcparray);
3873 tcparray->num = list->tickles.num;
3875 tcparray->connections = talloc_array(tcparray, struct ctdb_tcp_connection, tcparray->num);
3876 CTDB_NO_MEMORY(ctdb, tcparray->connections);
3878 memcpy(tcparray->connections, &list->tickles.connections[0],
3879 sizeof(struct ctdb_tcp_connection)*tcparray->num);
3881 /* We now have a new fresh tickle list array for this vnn */
3882 vnn->tcp_array = talloc_steal(vnn, tcparray);
3888 called to return the full list of tickles for the puclic address associated
3889 with the provided vnn
3891 int32_t ctdb_control_get_tcp_tickle_list(struct ctdb_context *ctdb, TDB_DATA indata, TDB_DATA *outdata)
3893 ctdb_sock_addr *addr = (ctdb_sock_addr *)indata.dptr;
3894 struct ctdb_control_tcp_tickle_list *list;
3895 struct ctdb_tcp_array *tcparray;
3897 struct ctdb_vnn *vnn;
3899 vnn = find_public_ip_vnn(ctdb, addr);
3901 DEBUG(DEBUG_ERR,(__location__ " Could not get tcp tickle list, '%s' is not a public address\n",
3902 ctdb_addr_to_str(addr)));
3907 tcparray = vnn->tcp_array;
3909 num = tcparray->num;
3914 outdata->dsize = offsetof(struct ctdb_control_tcp_tickle_list,
3915 tickles.connections)
3916 + sizeof(struct ctdb_tcp_connection) * num;
3918 outdata->dptr = talloc_size(outdata, outdata->dsize);
3919 CTDB_NO_MEMORY(ctdb, outdata->dptr);
3920 list = (struct ctdb_control_tcp_tickle_list *)outdata->dptr;
3923 list->tickles.num = num;
3925 memcpy(&list->tickles.connections[0], tcparray->connections,
3926 sizeof(struct ctdb_tcp_connection) * num);
3934 set the list of all tcp tickles for a public address
3936 static int ctdb_ctrl_set_tcp_tickles(struct ctdb_context *ctdb,
3937 struct timeval timeout, uint32_t destnode,
3938 ctdb_sock_addr *addr,
3939 struct ctdb_tcp_array *tcparray)
3943 struct ctdb_control_tcp_tickle_list *list;
3946 num = tcparray->num;
3951 data.dsize = offsetof(struct ctdb_control_tcp_tickle_list,
3952 tickles.connections) +
3953 sizeof(struct ctdb_tcp_connection) * num;
3954 data.dptr = talloc_size(ctdb, data.dsize);
3955 CTDB_NO_MEMORY(ctdb, data.dptr);
3957 list = (struct ctdb_control_tcp_tickle_list *)data.dptr;
3959 list->tickles.num = num;
3961 memcpy(&list->tickles.connections[0], tcparray->connections, sizeof(struct ctdb_tcp_connection) * num);
3964 ret = ctdb_daemon_send_control(ctdb, CTDB_BROADCAST_CONNECTED, 0,
3965 CTDB_CONTROL_SET_TCP_TICKLE_LIST,
3966 0, CTDB_CTRL_FLAG_NOREPLY, data, NULL, NULL);
3968 DEBUG(DEBUG_ERR,(__location__ " ctdb_control for set tcp tickles failed\n"));
3972 talloc_free(data.dptr);
3979 perform tickle updates if required
3981 static void ctdb_update_tcp_tickles(struct event_context *ev,
3982 struct timed_event *te,
3983 struct timeval t, void *private_data)
3985 struct ctdb_context *ctdb = talloc_get_type(private_data, struct ctdb_context);
3987 struct ctdb_vnn *vnn;
3989 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
3990 /* we only send out updates for public addresses that
3993 if (ctdb->pnn != vnn->pnn) {
3996 /* We only send out the updates if we need to */
3997 if (!vnn->tcp_update_needed) {
4000 ret = ctdb_ctrl_set_tcp_tickles(ctdb,
4002 CTDB_BROADCAST_CONNECTED,
4003 &vnn->public_address,
4006 DEBUG(DEBUG_ERR,("Failed to send the tickle update for public address %s\n",
4007 ctdb_addr_to_str(&vnn->public_address)));
4011 event_add_timed(ctdb->ev, ctdb->tickle_update_context,
4012 timeval_current_ofs(ctdb->tunable.tickle_update_interval, 0),
4013 ctdb_update_tcp_tickles, ctdb);
4018 start periodic update of tcp tickles
4020 void ctdb_start_tcp_tickle_update(struct ctdb_context *ctdb)
4022 ctdb->tickle_update_context = talloc_new(ctdb);
4024 event_add_timed(ctdb->ev, ctdb->tickle_update_context,
4025 timeval_current_ofs(ctdb->tunable.tickle_update_interval, 0),
4026 ctdb_update_tcp_tickles, ctdb);
4032 struct control_gratious_arp {
4033 struct ctdb_context *ctdb;
4034 ctdb_sock_addr addr;
4040 send a control_gratuitous arp
4042 static void send_gratious_arp(struct event_context *ev, struct timed_event *te,
4043 struct timeval t, void *private_data)
4046 struct control_gratious_arp *arp = talloc_get_type(private_data,
4047 struct control_gratious_arp);
4049 ret = ctdb_sys_send_arp(&arp->addr, arp->iface);
4051 DEBUG(DEBUG_ERR,(__location__ " sending of gratious arp on iface '%s' failed (%s)\n",
4052 arp->iface, strerror(errno)));
4057 if (arp->count == CTDB_ARP_REPEAT) {
4062 event_add_timed(arp->ctdb->ev, arp,
4063 timeval_current_ofs(CTDB_ARP_INTERVAL, 0),
4064 send_gratious_arp, arp);
4071 int32_t ctdb_control_send_gratious_arp(struct ctdb_context *ctdb, TDB_DATA indata)
4073 struct ctdb_control_gratious_arp *gratious_arp = (struct ctdb_control_gratious_arp *)indata.dptr;
4074 struct control_gratious_arp *arp;
4076 /* verify the size of indata */
4077 if (indata.dsize < offsetof(struct ctdb_control_gratious_arp, iface)) {
4078 DEBUG(DEBUG_ERR,(__location__ " Too small indata to hold a ctdb_control_gratious_arp structure. Got %u require %u bytes\n",
4079 (unsigned)indata.dsize,
4080 (unsigned)offsetof(struct ctdb_control_gratious_arp, iface)));
4084 ( offsetof(struct ctdb_control_gratious_arp, iface)
4085 + gratious_arp->len ) ){
4087 DEBUG(DEBUG_ERR,(__location__ " Wrong size of indata. Was %u bytes "
4088 "but should be %u bytes\n",
4089 (unsigned)indata.dsize,
4090 (unsigned)(offsetof(struct ctdb_control_gratious_arp, iface)+gratious_arp->len)));
4095 arp = talloc(ctdb, struct control_gratious_arp);
4096 CTDB_NO_MEMORY(ctdb, arp);
4099 arp->addr = gratious_arp->addr;
4100 arp->iface = talloc_strdup(arp, gratious_arp->iface);
4101 CTDB_NO_MEMORY(ctdb, arp->iface);
4104 event_add_timed(arp->ctdb->ev, arp,
4105 timeval_zero(), send_gratious_arp, arp);
4110 int32_t ctdb_control_add_public_address(struct ctdb_context *ctdb, TDB_DATA indata)
4112 struct ctdb_control_ip_iface *pub = (struct ctdb_control_ip_iface *)indata.dptr;
4115 /* verify the size of indata */
4116 if (indata.dsize < offsetof(struct ctdb_control_ip_iface, iface)) {
4117 DEBUG(DEBUG_ERR,(__location__ " Too small indata to hold a ctdb_control_ip_iface structure\n"));
4121 ( offsetof(struct ctdb_control_ip_iface, iface)
4124 DEBUG(DEBUG_ERR,(__location__ " Wrong size of indata. Was %u bytes "
4125 "but should be %u bytes\n",
4126 (unsigned)indata.dsize,
4127 (unsigned)(offsetof(struct ctdb_control_ip_iface, iface)+pub->len)));
4131 DEBUG(DEBUG_NOTICE,("Add IP %s\n", ctdb_addr_to_str(&pub->addr)));
4133 ret = ctdb_add_public_address(ctdb, &pub->addr, pub->mask, &pub->iface[0], true);
4136 DEBUG(DEBUG_ERR,(__location__ " Failed to add public address\n"));
4144 called when releaseip event finishes for del_public_address
4146 static void delete_ip_callback(struct ctdb_context *ctdb, int status,
4149 talloc_free(private_data);
4152 int32_t ctdb_control_del_public_address(struct ctdb_context *ctdb, TDB_DATA indata)
4154 struct ctdb_control_ip_iface *pub = (struct ctdb_control_ip_iface *)indata.dptr;
4155 struct ctdb_vnn *vnn;
4158 /* verify the size of indata */
4159 if (indata.dsize < offsetof(struct ctdb_control_ip_iface, iface)) {
4160 DEBUG(DEBUG_ERR,(__location__ " Too small indata to hold a ctdb_control_ip_iface structure\n"));
4164 ( offsetof(struct ctdb_control_ip_iface, iface)
4167 DEBUG(DEBUG_ERR,(__location__ " Wrong size of indata. Was %u bytes "
4168 "but should be %u bytes\n",
4169 (unsigned)indata.dsize,
4170 (unsigned)(offsetof(struct ctdb_control_ip_iface, iface)+pub->len)));
4174 DEBUG(DEBUG_NOTICE,("Delete IP %s\n", ctdb_addr_to_str(&pub->addr)));
4176 /* walk over all public addresses until we find a match */
4177 for (vnn=ctdb->vnn;vnn;vnn=vnn->next) {
4178 if (ctdb_same_ip(&vnn->public_address, &pub->addr)) {
4179 TALLOC_CTX *mem_ctx = talloc_new(ctdb);
4181 DLIST_REMOVE(ctdb->vnn, vnn);
4182 talloc_steal(mem_ctx, vnn);
4183 ctdb_remove_orphaned_ifaces(ctdb, vnn, mem_ctx);
4184 if (vnn->pnn != ctdb->pnn) {
4185 if (vnn->iface != NULL) {
4186 ctdb_vnn_unassign_iface(ctdb, vnn);
4188 talloc_free(mem_ctx);
4193 ret = ctdb_event_script_callback(ctdb,
4194 mem_ctx, delete_ip_callback, mem_ctx,
4196 CTDB_EVENT_RELEASE_IP,
4198 ctdb_vnn_iface_string(vnn),
4199 ctdb_addr_to_str(&vnn->public_address),
4200 vnn->public_netmask_bits);
4201 if (vnn->iface != NULL) {
4202 ctdb_vnn_unassign_iface(ctdb, vnn);
4215 struct ipreallocated_callback_state {
4216 struct ctdb_req_control *c;
4219 static void ctdb_ipreallocated_callback(struct ctdb_context *ctdb,
4220 int status, void *p)
4222 struct ipreallocated_callback_state *state =
4223 talloc_get_type(p, struct ipreallocated_callback_state);
4227 (" \"ipreallocated\" event script failed (status %d)\n",
4229 if (status == -ETIME) {
4230 ctdb_ban_self(ctdb);
4234 ctdb_request_control_reply(ctdb, state->c, NULL, status, NULL);
4238 /* A control to run the ipreallocated event */
4239 int32_t ctdb_control_ipreallocated(struct ctdb_context *ctdb,
4240 struct ctdb_req_control *c,
4244 struct ipreallocated_callback_state *state;
4246 state = talloc(ctdb, struct ipreallocated_callback_state);
4247 CTDB_NO_MEMORY(ctdb, state);
4249 DEBUG(DEBUG_INFO,(__location__ " Running \"ipreallocated\" event\n"));
4251 ret = ctdb_event_script_callback(ctdb, state,
4252 ctdb_ipreallocated_callback, state,
4253 false, CTDB_EVENT_IPREALLOCATED,
4257 DEBUG(DEBUG_ERR,("Failed to run \"ipreallocated\" event \n"));
4262 /* tell the control that we will be reply asynchronously */
4263 state->c = talloc_steal(state, c);
4264 *async_reply = true;
4270 /* This function is called from the recovery daemon to verify that a remote
4271 node has the expected ip allocation.
4272 This is verified against ctdb->ip_tree
4274 int verify_remote_ip_allocation(struct ctdb_context *ctdb,
4275 struct ctdb_all_public_ips *ips,
4278 struct ctdb_public_ip_list *tmp_ip;
4281 if (ctdb->ip_tree == NULL) {
4282 /* dont know the expected allocation yet, assume remote node
4291 for (i=0; i<ips->num; i++) {
4292 tmp_ip = trbt_lookuparray32(ctdb->ip_tree, IP_KEYLEN, ip_key(&ips->ips[i].addr));
4293 if (tmp_ip == NULL) {
4294 DEBUG(DEBUG_ERR,("Node %u has new or unknown public IP %s\n", pnn, ctdb_addr_to_str(&ips->ips[i].addr)));
4298 if (tmp_ip->pnn == -1 || ips->ips[i].pnn == -1) {
4302 if (tmp_ip->pnn != ips->ips[i].pnn) {
4304 ("Inconsistent IP allocation - node %u thinks %s is held by node %u while it is assigned to node %u\n",
4306 ctdb_addr_to_str(&ips->ips[i].addr),
4307 ips->ips[i].pnn, tmp_ip->pnn));
4315 int update_ip_assignment_tree(struct ctdb_context *ctdb, struct ctdb_public_ip *ip)
4317 struct ctdb_public_ip_list *tmp_ip;
4319 if (ctdb->ip_tree == NULL) {
4320 DEBUG(DEBUG_ERR,("No ctdb->ip_tree yet. Failed to update ip assignment\n"));
4324 tmp_ip = trbt_lookuparray32(ctdb->ip_tree, IP_KEYLEN, ip_key(&ip->addr));
4325 if (tmp_ip == NULL) {
4326 DEBUG(DEBUG_ERR,(__location__ " Could not find record for address %s, update ip\n", ctdb_addr_to_str(&ip->addr)));
4330 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));
4331 tmp_ip->pnn = ip->pnn;
4337 struct ctdb_reloadips_handle {
4338 struct ctdb_context *ctdb;
4339 struct ctdb_req_control *c;
4343 struct fd_event *fde;
4346 static int ctdb_reloadips_destructor(struct ctdb_reloadips_handle *h)
4348 if (h == h->ctdb->reload_ips) {
4349 h->ctdb->reload_ips = NULL;
4352 ctdb_request_control_reply(h->ctdb, h->c, NULL, h->status, NULL);
4355 ctdb_kill(h->ctdb, h->child, SIGKILL);
4359 static void ctdb_reloadips_timeout_event(struct event_context *ev,
4360 struct timed_event *te,
4361 struct timeval t, void *private_data)
4363 struct ctdb_reloadips_handle *h = talloc_get_type(private_data, struct ctdb_reloadips_handle);
4368 static void ctdb_reloadips_child_handler(struct event_context *ev, struct fd_event *fde,
4369 uint16_t flags, void *private_data)
4371 struct ctdb_reloadips_handle *h = talloc_get_type(private_data, struct ctdb_reloadips_handle);
4376 ret = read(h->fd[0], &res, 1);
4377 if (ret < 1 || res != 0) {
4378 DEBUG(DEBUG_ERR, (__location__ " Reloadips child process returned error\n"));
4386 static int ctdb_reloadips_child(struct ctdb_context *ctdb)
4388 TALLOC_CTX *mem_ctx = talloc_new(NULL);
4389 struct ctdb_all_public_ips *ips;
4390 struct ctdb_vnn *vnn;
4391 struct client_async_data *async_data;
4392 struct timeval timeout;
4394 struct ctdb_client_control_state *state;
4398 CTDB_NO_MEMORY(ctdb, mem_ctx);
4400 /* Read IPs from local node */
4401 ret = ctdb_ctrl_get_public_ips(ctdb, TAKEOVER_TIMEOUT(),
4402 CTDB_CURRENT_NODE, mem_ctx, &ips);
4405 ("Unable to fetch public IPs from local node\n"));
4406 talloc_free(mem_ctx);
4410 /* Read IPs file - this is safe since this is a child process */
4412 if (ctdb_set_public_addresses(ctdb, false) != 0) {
4413 DEBUG(DEBUG_ERR,("Failed to re-read public addresses file\n"));
4414 talloc_free(mem_ctx);
4418 async_data = talloc_zero(mem_ctx, struct client_async_data);
4419 CTDB_NO_MEMORY(ctdb, async_data);
4421 /* Compare IPs between node and file for IPs to be deleted */
4422 for (i = 0; i < ips->num; i++) {
4424 for (vnn = ctdb->vnn; vnn; vnn = vnn->next) {
4425 if (ctdb_same_ip(&vnn->public_address,
4426 &ips->ips[i].addr)) {
4427 /* IP is still in file */
4433 /* Delete IP ips->ips[i] */
4434 struct ctdb_control_ip_iface *pub;
4437 ("IP %s no longer configured, deleting it\n",
4438 ctdb_addr_to_str(&ips->ips[i].addr)));
4440 pub = talloc_zero(mem_ctx,
4441 struct ctdb_control_ip_iface);
4442 CTDB_NO_MEMORY(ctdb, pub);
4444 pub->addr = ips->ips[i].addr;
4448 timeout = TAKEOVER_TIMEOUT();
4450 data.dsize = offsetof(struct ctdb_control_ip_iface,
4452 data.dptr = (uint8_t *)pub;
4454 state = ctdb_control_send(ctdb, CTDB_CURRENT_NODE, 0,
4455 CTDB_CONTROL_DEL_PUBLIC_IP,
4456 0, data, async_data,
4458 if (state == NULL) {
4461 " failed sending CTDB_CONTROL_DEL_PUBLIC_IP\n"));
4468 /* Compare IPs between node and file for IPs to be added */
4470 for (vnn = ctdb->vnn; vnn; vnn = vnn->next) {
4471 for (i = 0; i < ips->num; i++) {
4472 if (ctdb_same_ip(&vnn->public_address,
4473 &ips->ips[i].addr)) {
4474 /* IP already on node */
4478 if (i == ips->num) {
4479 /* Add IP ips->ips[i] */
4480 struct ctdb_control_ip_iface *pub;
4481 const char *ifaces = NULL;
4486 ("New IP %s configured, adding it\n",
4487 ctdb_addr_to_str(&vnn->public_address)));
4489 uint32_t pnn = ctdb_get_pnn(ctdb);
4491 data.dsize = sizeof(pnn);
4492 data.dptr = (uint8_t *)&pnn;
4494 ret = ctdb_client_send_message(
4496 CTDB_BROADCAST_CONNECTED,
4497 CTDB_SRVID_REBALANCE_NODE,
4500 DEBUG(DEBUG_WARNING,
4501 ("Failed to send message to force node reallocation - IPs may be unbalanced\n"));
4507 ifaces = vnn->ifaces[0];
4509 while (vnn->ifaces[iface] != NULL) {
4510 ifaces = talloc_asprintf(vnn, "%s,%s", ifaces,
4511 vnn->ifaces[iface]);
4515 len = strlen(ifaces) + 1;
4516 pub = talloc_zero_size(mem_ctx,
4517 offsetof(struct ctdb_control_ip_iface, iface) + len);
4518 CTDB_NO_MEMORY(ctdb, pub);
4520 pub->addr = vnn->public_address;
4521 pub->mask = vnn->public_netmask_bits;
4523 memcpy(&pub->iface[0], ifaces, pub->len);
4525 timeout = TAKEOVER_TIMEOUT();
4527 data.dsize = offsetof(struct ctdb_control_ip_iface,
4529 data.dptr = (uint8_t *)pub;
4531 state = ctdb_control_send(ctdb, CTDB_CURRENT_NODE, 0,
4532 CTDB_CONTROL_ADD_PUBLIC_IP,
4533 0, data, async_data,
4535 if (state == NULL) {
4538 " failed sending CTDB_CONTROL_ADD_PUBLIC_IP\n"));
4544 if (ctdb_client_async_wait(ctdb, async_data) != 0) {
4545 DEBUG(DEBUG_ERR,(__location__ " Add/delete IPs failed\n"));
4549 talloc_free(mem_ctx);
4553 talloc_free(mem_ctx);
4557 /* This control is sent to force the node to re-read the public addresses file
4558 and drop any addresses we should nnot longer host, and add new addresses
4559 that we are now able to host
4561 int32_t ctdb_control_reload_public_ips(struct ctdb_context *ctdb, struct ctdb_req_control *c, bool *async_reply)
4563 struct ctdb_reloadips_handle *h;
4564 pid_t parent = getpid();
4566 if (ctdb->reload_ips != NULL) {
4567 talloc_free(ctdb->reload_ips);
4568 ctdb->reload_ips = NULL;
4571 h = talloc(ctdb, struct ctdb_reloadips_handle);
4572 CTDB_NO_MEMORY(ctdb, h);
4577 if (pipe(h->fd) == -1) {
4578 DEBUG(DEBUG_ERR,("Failed to create pipe for ctdb_freeze_lock\n"));
4583 h->child = ctdb_fork(ctdb);
4584 if (h->child == (pid_t)-1) {
4585 DEBUG(DEBUG_ERR, ("Failed to fork a child for reloadips\n"));
4593 if (h->child == 0) {
4594 signed char res = 0;
4597 debug_extra = talloc_asprintf(NULL, "reloadips:");
4599 ctdb_set_process_name("ctdb_reloadips");
4600 if (switch_from_server_to_client(ctdb, "reloadips-child") != 0) {
4601 DEBUG(DEBUG_CRIT,("ERROR: Failed to switch reloadips child into client mode\n"));
4604 res = ctdb_reloadips_child(ctdb);
4606 DEBUG(DEBUG_ERR,("Failed to reload ips on local node\n"));
4610 write(h->fd[1], &res, 1);
4611 /* make sure we die when our parent dies */
4612 while (ctdb_kill(ctdb, parent, 0) == 0 || errno != ESRCH) {
4618 h->c = talloc_steal(h, c);
4621 set_close_on_exec(h->fd[0]);
4623 talloc_set_destructor(h, ctdb_reloadips_destructor);
4626 h->fde = event_add_fd(ctdb->ev, h, h->fd[0],
4627 EVENT_FD_READ, ctdb_reloadips_child_handler,
4629 tevent_fd_set_auto_close(h->fde);
4631 event_add_timed(ctdb->ev, h,
4632 timeval_current_ofs(120, 0),
4633 ctdb_reloadips_timeout_event, h);
4635 /* we reply later */
4636 *async_reply = true;